Nucl - Phys.B v.784 PDF

Nuclear Physics B 784 (2007) 121
Chiral symmetry breaking from intersecting D-branes

E. Antonyan, J.A. Harvey , D. Kutasov
EFI and Department of Physics, University of Chicago, 5640 South Ellis Avenue, Chicago, IL 60637, USA
Received 22 January 2007; accepted 25 May 2007
Available online 6 June 2007
Abstract
We study a class of intersecting D-brane models in which fermions localized at different intersections
interact via exchange of bulk fields. In some cases these interactions lead to dynamical symmetry breaking
and generate a mass for the fermions. We analyze the conditions under which this happens as one varies the
dimensions of the branes and of the intersections.
2007 Elsevier B.V. All rights reserved.
1. Introduction
The system of Nc D4-branes wrapped around a circle with anti-periodic boundary conditions
for the fermions provides an interesting example of gauge-gravity duality [1]. In a certain region of the parameter space of the brane configuration, corresponding to small four-dimensional
t Hooft coupling, the low-energy theory on the D-branes is (3 + 1)-dimensional SU(Nc ) Yang
Mills (YM) theory without matter. Unfortunately, in that limit the theory on the branes is hard to
solve, even in the large Nc limit.
For large t Hooft coupling (and large Nc ), the dynamics reduces to supergravity in the nearhorizon geometry of the D4-branes and can be analyzed in some detail. The theory exhibits
confinement and has a spectrum of glueballs that can be calculated using supergravity. In this
regime the theory on the branes is not pure YM, since the adjoint scalars and fermions living
on the D4-branes are not decoupled, and their dynamics is not well described by the effective
Lagrangian obtained by dimensional reduction of N = 1 SYM in 9 + 1 dimensions. The dynamics can be described in terms of the (2, 0) superconformal field theory in 5 + 1 dimensions
* Corresponding author.
E-mail address: harvey@theory.uchicago.edu (J.A. Harvey).

0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.05.023
E. Antonyan et al. / Nuclear Physics B 784 (2007) 121
compactified on a two-torus with twisted boundary conditions around one of the cycles, but this
description is not easy to use.
While the theory with a good supergravity description is not YM, the two are related by
a continuous deformation, and one may hope that they are in the same phase. The reason for
this is that the (2, 0) theory is believed to be a standard QFT; compactifying it on a torus leads
to an RG flow, which is expected to be smooth. Changing the parameters associated with the
compactification probes different parts of this flow. Since physical properties change smoothly
along RG trajectories, it is natural to expect that the supergravity and YM regimes are in the same
phase. This means that some qualitative and perhaps even quantitative features of YM theory can
be addressed in supergravity. This is of course a general theme in gauge-gravity duality.
A natural way to add dynamical fermions in the fundamental representation of the gauge
group to the setup of [1] was proposed in [2] (see, e.g., [36], for other work on incorporating
dynamical quarks into gauge-gravity duality). It involves adding to the Nc color D4-branes Nf
flavor D8- and D8-branes, which intersect the color branes along an R3,1 . The flavor branes
and anti-branes are separated by a distance L in the remaining (compact) direction along the
D4-branes. Left-handed quarks live at the intersection of the D4- and D8-branes, while righthanded quarks live at the D4D8 intersection. A nice feature of this brane construction is that
quarks of different chiralities are physically separated in the extra dimensions, and the chiral
U (Nf )L U (Nf )R flavor symmetry is manifest.
QCD with massless quarks is obtained in a certain region of the parameter space of the brane
configuration where the dynamics is again difficult to analyze. In a different region of parameter
space one can analyze the theory by studying the DBI action for the D8-branes in the nearhorizon geometry of the D4-branes. As in the case without quarks, in this regime the theory is
not QCD but it is expected to be in the same phase. In particular, it exhibits confinement and
chiral symmetry breaking. The latter has a nice geometric realization.
It was pointed out in [7] that the brane construction of [2] has the interesting property that it
decouples the scales of confinement and chiral symmetry breaking.1 By varying the parameters
of the brane configuration one can make the energy scale of chiral symmetry breaking arbitrarily
higher than that of confinement. In fact, sending the size of the circle that the D4-branes wrap to
infinity one arrives at a theory which breaks chiral symmetry but does not confine. This theory
can again be studied at weak coupling using field theoretic techniques and at strong coupling
using supergravity.
The intersecting brane construction of [2,7] is a special case of a much more general class of
constructions, obtained by varying the dimensions of the color and flavor branes and that of the
intersection. Some other examples were considered recently in [9,10]. The purpose of this paper
is to present a more uniform treatment of these and other intersecting brane systems, using field
theory and supergravity, as appropriate, to analyze them.
The general setup we will consider is the following. We start with Nc color Dq-branes, which
generalize the color D4-branes of [2,7]. We will take these branes to be non-compact in all directions since we are mainly interested in chiral symmetry breaking. The low-energy theory on the
Dq-branes is (q + 1)-dimensional SYM with sixteen supercharges. For q < 3 it is strongly coupled in the infrared and weakly coupled in the UV, while for q > 3 it has the opposite behavior.
For q = 3 this theory is N = 4 SYM in 3 + 1 dimensions, which is conformally invariant. The
t Hooft coupling of the theory on the color branes, q+1 , has units of (length)q3 .
1 A similar phenomenon was observed in a different context in [8].
For q 4 the low-energy theory of the color branes is a local QFT, which can be decoupled
from gravity by taking a certain scaling limit. For q = 5 it is a non-local theory known as Little
String Theory (LST). This theory has a Hagedorn density of states; the Hagedorn temperature
sets the scale of non-locality (see, e.g., [11,12] for discussions). For q = 6, the theory on the
branes cannot be decoupled from bulk gravity [13].
In addition to the color branes we have Nf flavor Dp- and Dp-branes, which are analogs of
the flavor D8-branes and anti-branes in the construction of [2,7]. The color and flavor branes
intersect along an (r + 1)-dimensional spacetime, which we will denote by Dq Dp = Ir . The
flavor branes and anti-branes are separated by a distance L in a direction along the color branes
but transverse to the intersection Ir . In the cases of interest, the light degrees of freedom localized
at the intersections are fermions, which interact via exchange of fields living on the color branes.
One can think of the flavor branes, and in particular the dynamics of the fermions, as probing
the theory of the color branes at the scale L. By changing L we probe the color theory at different
points along its RG trajectory. For q 4 we expect the dependence on L to be smooth. For
q = 5 we expect it to be smooth for distances much larger than the scale of non-locality of
the underlying LST. For q = 6 there is no decoupling limit and one has to take into account
gravitational effects.
We will study the dynamics of these configurations in the limit Nc , gs 0 with gs Nc
and Nf held fixed. This dynamics depends non-trivially on the values of p, q and r, and on the
dimensionless parameter
(eff)
q+1 (L) = q+1 L3q ,
(1.1)
which can be thought of as the effective coupling of the color degrees of freedom at distance
scale L, or energy E 1/L. We will explore the dependence of the low-energy behavior on
these parameters, focusing on the question of dynamical symmetry breaking at weak and strong
coupling.
We have organized the paper as follows. In Section 2 we present a general analysis of the
intersecting brane systems we will consider. We review the spectrum of fields localized at a
particular intersection, and describe the leading interactions at weak coupling between fields at
different intersections. There are two qualitatively different cases that need to be considered,
depending on whether or not this interaction has finite range.
The supergravity description of the above intersecting brane systems is obtained by replacing
the color branes by their near-horizon geometry and studying the DBI action of the flavor branes
in this background. Holography suggests that in some cases (depending on the dimension of the
color branes) supergravity provides a useful description of the dynamics at strong coupling.
In the following sections we illustrate the general considerations of Section 2 with a number of
specific models. In Section 3 we discuss configurations with color D4-branes. These include the
models of [2,7,10] as well as a model with color and flavor D4-branes and a (1 + 1)-dimensional
intersection, which we discuss in some detail. In this model there are left- and right-moving
fermions with the same quantum numbers under U (Nc ) U (Nf ) at each intersection, and one
can give them a mass by separating the color and flavor branes in the two dimensions transverse
to both. For zero mass we show that the model still exhibits the dynamical breaking of a certain
chiral symmetry, both at weak coupling where it is described by a generalization of the Gross
Neveu model, and at strong coupling where it is described by probe brane dynamics in gravity.
For non-zero mass this symmetry is explicitly broken.
In Section 4 we discuss models with color Dq-branes for q > 4. These include a IIB model
with D5-branes intersecting in 1 + 1 dimensions. The weakly coupled theory is the GrossNeveu
model, with a slightly different UV cutoff than that of [10]. As the coupling increases (i.e., as
L decreases), one probes shorter and shorter distance physics in the LST of the fivebranes. As
mentioned above, this theory is non-local. One expects to encounter non-field theoretic behavior
when L reaches the scale of non-locality. Indeed, we find that the supergravity analysis gives
in this case a continuous set of solutions which exists for some critical value L = L . These
solutions appear to describe the interactions of the fermions localized at the intersection with
the continuum of closed string modes propagating in the fivebrane throat. We study a number
of additional models with color fivebranes, and find that all of them have similar solutions, with
slightly different values of L .
We also discuss models with color D6-branes. At weak coupling they can be studied using
field theoretic means. At finite coupling it is not clear that they makes sense due to the absence
of a good UV completion (which does not involve gravity). The supergravity analysis predicts
the existence of an unstable state with broken symmetry at weak coupling, and is inapplicable
for strong coupling.
In Section 5 we analyze several models with color D2- and D3-branes, and find qualitatively
similar behavior to the color D4-brane case. We conclude in Section 6 with a discussion.
2. General results
The brane configurations that we will consider consist of two intersections, Dq Dp = Ir
and Dq Dp = Ir , separated by a distance L in a direction transverse to the Dp- and Dp-branes
and along the Dq-branes. In this section we develop some tools for analyzing these systems. We
start by reviewing some standard facts about such intersections, following [14]. We then go on
to a discussion of systems with two intersections in a regime where the coupling between them
is weak and the dynamics can be studied using field theoretic techniques. In the last subsection
we describe these systems in supergravity, and discuss the implications of holography for them.
2.1. Classification
Consider an intersection of Nc Dq-branes and Nf Dp-branes2 along an Rr,1 (which we refer
to as Dq Dp = Ir ). We would like to determine the spectrum of massless states living at the
intersection, and in particular its chirality with respect to the U (Nc ) U (Nf ) symmetry on the
branes.
Imagine that all the spatial directions transverse to the intersection are compactified on circles,
so we can apply T -duality in these directions. The spectrum of massless states in r + 1 dimensions is invariant under these operations. Hence, we can map all intersections to a small class
of basic ones, and analyze those. For example, we can T -dualize the Dp-branes to D9-branes.
The Dq-branes turn in the process to Dr -branes, with r r, and the intersection becomes
Dr D9 = Ir .
There are four possibilities, r = 1, 3, 5, 7. For r = 7, the spectrum of strings stretched between the color and flavor branes (or 79 strings) contains a NS sector tachyon localized at the
intersection. We will not discuss this case further here. For r = 5, the intersection preserves eight
2 All the branes here and below are BPS (although the full brane configurations we will study break all supersymmetry).
We will not discuss intersecting brane systems that contain non-BPS branes, since the latter have tachyon instabilities
that are not of interest here.
supercharges, and the spectrum of 59 strings contains a massless hypermultiplet in the representation (Nc , N f ) of the gauge group. For r = 3 the massless spectrum contains a Weyl fermion
in the same representation, coming from the Ramond sector of 39 strings. All the states in the
NS sector are massive. Finally, for r = 1 the system again preserves eight supercharges, which
have a particular chirality in the 1 + 1 dimensions along the intersection. The massless spectrum
consists of chiral (Weyl) fermions, with the opposite chirality to that of the supercharges.
Note that in all the above cases, the massless spectrum at the intersection coming from r 9
strings is chiral. In cases where r = r, this means that the original intersection, Dq Dp = Ir ,
also has a chiral spectrum. When r > r, the spectrum at the original intersection Ir is obtained
by dimensionless reduction from r + 1 to r + 1 dimensions. As is well known, dimensionally
reducing chiral fermions gives non-chiral ones, so the resulting (r + 1)-dimensional spectrum is
non-chiral.
A closely related fact is that a transverse intersection, i.e., one with q + p r = 9 or equivalently r = r, has the property that there are no directions of space transverse to both kinds of
branes, so they always intersect. This is the geometric counterpart of the fact that one cannot give
a gauge invariant mass to chiral fermions.
On the other hand, intersections with r > r are not transverse, so the color and flavor branes
can be separated in directions transverse to both. Doing so gives a mass to the fermions at the
intersection which correspond to strings stretching from one brane to the other. The fact that it is
possible to give a gauge invariant mass to the fermions implies that the latter are not chiral.
It should be noted that the discussion above addressed the question of chirality with respect to
the U (Nc ) U (Nf ) symmetry on the branes. Even for non-transverse intersections, the fermions
may be chiral with respect to geometric symmetries from the normal bundle to the intersection.
The dynamical breaking of such geometric chiral symmetries has been studied in the context of
AdS/CFT, see, e.g., [1518].
2.2. Weak coupling
In the previous subsection we discussed the spectrum of states localized at a given intersection
of the form Dq Dp = Ir . Our main interest in this paper is in the systems with two such intersections described above. The important new feature of such systems is the interaction between
modes localized at the two intersections.
The leading interaction between the two intersections is due to exchange of a single color
gluon (and, for non-transverse intersections, scalars as well). This gives rise to a quartic interaction proportional to the gauge coupling of the color Dq-branes,
2
gq+1
(2.1)
= (2)q2 gs s q3 ,
where s = is the string length and gs is the string coupling. The (q + 1)-dimensional
t Hooft coupling q+1 can be defined in terms of gq+1 as
q+1 =
2 N
gq+1
c
(2)q2
The quartic interaction due to single gluon exchange is proportional to

q+1
d r+1 x d r+1 y Gq+1 (x y, L) L (x) R (y) R (y) L (x) ,
Nc
(2.2)
(2.3)
where L and R are fermion fields localized at the two intersections, respectively, and

1 (q1)
Gq+1 (x, L) = x 2 + L2 2
(2.4)
is proportional to the (q + 1)-dimensional massless propagator over a distance L in the directions

along which the flavor branes are separated and distance x along the intersection.3 Each term in
squared brackets in (2.3) is a singlet of global U (Nc ), and we suppress the flavor labels. The
expression (2.3) is schematic. For any given intersection one can write it more precisely, as was
done for some cases in [7,10] and will be done for some others below.
The non-local interaction (2.3) is non-singular in the UV. One can think of L as a UV cutoff.
The long distance behavior of the theory depends in an important way on whether this interaction
has finite range or not. For q r > 2, the integral

d r+1 x Gq+1 (x, L)
(2.5)
converges, and one can think of Gq+1 as an (r + 1)-dimensional -function smeared over a
distance of order L. Thus, at distances much larger than L one can replace (2.3) by the local
interaction

q+1
1
(2.6)
q3 Lr1 d r+1 x L (x) R (x) R (x) L (x) .
Nc L
Each of the factors in front of the integral in (2.6) has a simple interpretation. The first is nec(eff)
essary to get a smooth large Nc limit; the second is the effective coupling q+1 (L) (1.1). The
third is a power of the UV cutoff, that is needed to account for the scaling dimension of the local
quartic coupling. For example, for r = 3 (i.e., (3 + 1)-dimensional intersection) the four-Fermi
coupling has dimension six, which means that one needs a factor of L2 to reach the dimension
required of a Lagrangian. For r = 1 the third factor in (2.6) is absent, in agreement with the fact
that the four-Fermi coupling is in this case marginal (more precisely marginally relevant).
The theory with a local interaction (2.6) is solvable in the large Nc limit. For (1 + 1)dimensional intersections, such models exhibit dynamical symmetry breaking for arbitrarily
weak coupling (an example is the GrossNeveu model [19], which appears in the example studied
in [10], and some other brane configurations that will be mentioned below). For higher dimension (r > 1) they typically do not break chiral symmetry at weak coupling. An example is the
original NJL model [20], which as we will see appears in string theory as a low-energy model
corresponding to a certain brane configuration.
For q r 2 the integral (2.5) diverges and the range of the quartic interaction (2.3) is infinite.
This makes the analysis above more subtle and we will leave it to future work.
2.3. Supergravity analysis
The discussion of the previous subsection is valid when the effective coupling (1.1) is small.
For large (eff)
q+1 (L) the interactions between color and flavor degrees of freedom are strong and
one needs to use other tools to analyze them.
The problem without the flavor branes was studied in [13], where a qualitatively different
behavior was found for Dq-branes with q 4, and for those with q = 5, 6. In the former case the
3 We have written the interaction in Euclidean spacetime, which is convenient for studying the vacuum structure.
theory on the branes can be decoupled from gravity. As one changes the effective coupling (1.1),
the useful description changes from field theory, to ten-dimensional gravity, and sometimes to
eleven-dimensional gravity. The important fact for our purposes is that there is a wide range of
values of the effective coupling in which the field theoretic description is strongly coupled and
one has to use type II supergravity to study the dynamics.
For D5-branes the situation is more complex. The low-energy field theory degrees of freedom
do not decouple from a continuum of states that live in the throat of the fivebranes (see, e.g.,
[11,12,21]). Gravity in the near-horizon geometry of the fivebranes includes these states. For
D6-branes, the low-energy theory on the branes cannot be decoupled from gravity at all.
In this subsection we will analyze what happens when one adds to the system the flavor
branes and anti-branes discussed above. We will replace the color branes by their near-horizon
geometry and will study the flavor branes and anti-branes using their DiracBornInfeld (DBI)
action. We will see that the results are compatible with the above picture, and in particular exhibit
qualitatively different behavior for color branes with q 4, q = 5 and q = 6.
We will take the color Dq-branes to span the directions (0, 1, . . . , q), while the flavor Dpand Dp-branes are stretched in (0, 1, . . . , r, q + 1, q + 2, . . . , q + p r). The two intersections
lie along the (r + 1)-dimensional space with coordinates (0, 1, . . . , r) and are separated by a
distance L in the x q direction.
The near-horizon geometry of the Dq-branes is described by the metric and dilaton

(7q)/2

Rq+1 (7q)/2 2
U
2
2
dU + U 2 d8q
,
dx
ds =
Rq+1
U

Rq+1 (7q)(3q)/4
e = gs
,
U
2
(2.7)
where
7q
Rq+1
= (2 )
5q

7q
7q 2
72q
93q
( )
gs Nc = 2
gq+1 Nc .
2
2
(2.8)
There is also a RR flux through the (8 q)-sphere (2.7).

In the supergravity approximation, the dynamics of the flavor Dp-branes in the background
(2.7) is described by a DBI action, whose form (suppressing the gauge field on the branes) is
given by

SDp = Tp d p+1 x e det gDp
(2.9)
where gDp is the induced metric on the Dp-brane. There are also ChernSimons couplings in the
full action which are important for the analysis of anomalies, but play no role in what follows.
As in the examples studied in [7,10], in solving for the shape of the flavor branes we have
to allow for the possibility that the parallel separated Dp- and Dp-brane configuration that we
specified at weak coupling is deformed due to the effects of interactions with the color branes.
The brane configuration should still approach a brane and anti-brane at a distance |x q | = L
as U , and preserve the same symmetry as the intersecting brane system at weak coupling.
This implies that the Dp-brane wraps Rr,1 , a spherical subspace S pr1 of the (8 q)-sphere
transverse to the color branes, and a curve U (x q ) in the (U, x q ) plane, which approaches U
as x q = L/2. The induced metric is then given in terms of U = dU/dx q by

(7q)/2
(7q)/2

Rq+1 7q 2 q 2
U
U
dx dx
(U ) dx
1+
Rq+1
Rq+1
U

(7q)/2
Rq+1
2
U 2 dpr1
.
(2.10)
U

dsp2 =
Using (2.10) and (2.7) in (2.9) leads to the action

2
U
q
(U )2 ,
SDp = C(p, q, r) dx U 2 1 +
Rq+1
(2.11)
where we have defined

14 (q7)(2rpq+6)

Tp
Vol Rr,1 Vol S pr1 Rq+1
,
gs
7q
= (2r q p + 6)
+ 2(p r 1),
2
q 7
=
(2.12)
.
2
For the special case of a transverse intersection these expressions can be simplified by using the
relation p + q r = 9.
Since the Lagrangian (2.11) does not depend explicitly on x q , there is a first integral given by
C(p, q, r) =
U2
U 2
1 + ( Rq+1
) (U )2
= U02 ,
(2.13)
where U0 is the value of U where U = 0. Solving (2.13) for U and integrating gives
x (U ) =
q
U
Rq+1
U0
u du
.
(u/U0 ) 1
(2.14)
The integral can be evaluated in terms of complete and incomplete Beta functions,

U0 U 0
+1 1 1
U0
+1 1 1
;
+ ,
B
+ ,
.
B
x q (U ) =
Rq+1
2 2
U
2 2
(2.15)
The boundary conditions x q (U ) L/2 imply that

2U0 U0
+1 1 1
L = 2x q () =
B
+ ,
.
Rq+1
2 2
(2.16)
Using (2.2), (2.8), (2.12) and dropping constants this can be rewritten as
q5
L 2 U0
q+1 .
(2.17)
This is the holographic energydistance relation of [22], with the field theory energy scale
E 1/L. The solution (2.15) describes a curved, connected Dp-brane, which looks like Dpand Dp-branes connected by a wormhole whose width is determined by U0 (2.16).
The case of color fivebranes (q = 5) is special: the U0 dependence in (2.16) cancels and one
finds
2R6
.
L=
(2.18)
p1
Thus, in this case a solution exists only for a particular L of order R6 and any width U0 . The
scale (2.18) is of order the non-locality scale of the LST on the fivebranes. Thus, it is natural to
suspect that it is associated with interactions between the fermions at the intersections and high
energy, non-field theoretic, excitations in the fivebrane theory.
For a given value of L, the equations of motion of the DBI action (2.11) have two solutions.
One corresponds to U0 = 0 in (2.13) and describes a disconnected Dp- and Dp-brane pair,
running along the U axis at x q = L/2. The other is the curved, connected solution (2.15).
To determine the ground state of the system, one needs to compare their energies.
The energy difference between the two solutions, E Estraight Ecurved is

u/2+
C(p, q, r)
/2+
du u
du
.
E =
(2.19)
1 (U0 /u)
R
q+1
U0
Each integral in (2.19) is separately divergent at large u. The divergence can be regulated as in [7]
by regrouping terms, or equivalently by writing the integrals in terms of Beta functions which
are then defined by analytic continuation. This procedure should be equivalent to the holographic
renormalization reviewed in [23]. One finds

/2++1
1 +1 1
1 C(p, q, r)U0
B
,
.
E =
(2.20)
2
Rq+1
One can check that the sign of E depends only on q. For q 4, E > 0, so that the curved
solution has lower energy and is the ground state of the system. For q = 5, E = 0, so the energy
is independent of U0 . For q = 6, E < 0; hence, the curved solution is unstable4 and the straight
one is the ground state.
In order to trust the above supergravity analysis the curvature of the metric must be small at
the values of U which govern the dynamics, that is at U U0 . It was shown in [13] that this is
the case provided that the effective coupling (1.1) at the energy U0 is large,
q3
q+1 U0
1.
(2.21)
For q 4, eliminating U0 using (2.17) we find that the validity of supergravity requires
q+1 Lq3 ,
(2.22)
or, equivalently, large effective coupling at the scale 1/L. We see that for supergravity to be
valid,5 the effective coupling should be large both at the scale U0 , and at the scale 1/L. These
two scales are dynamically important; the former sets the dynamically generated mass of the
fermions at the intersection, while the latter governs the mass of the low-lying mesons, which
can be studied by expanding the DBI action around the background solution (2.15).
4 For the D6 D2 = I intersection, plugging into (2.12) one finds = 0, so the present discussion does not apply to
1
this case. The statements in the text apply to all the other intersections with color D6-branes.
5 There is also a bound on U which follows from the requirement that the string coupling remains small [13].
10
The relation between the two scales (2.17) can be rewritten as

U0

1
1 (eff)
1
(eff)
(U0 ).
q+1 (L) 5q
L
L q+1
(2.23)
For large effective coupling (2.21), (2.22), U0 is a higher energy scale than 1/L. Depending on
the value of q, one of the conditions (2.21), (2.22) can be more restrictive. For q = 2 the effective
coupling (1.1) decreases as the energy increases. Thus, the condition that the coupling be large
at energy U0 (2.21) is more restrictive than that at 1/L (2.22). For q = 3, the coupling does not
run and (2.21) and (2.22) are equivalent. For q = 4 the coupling increases with energy so (2.22)
is the more restrictive condition.
The cases q = 5 and q = 6 have to be treated separately. For q = 5, we see from (2.18) that the
curved solution exists only for a particular L of order R6 . Moreover, since the energy difference
(2.20) vanishes for this case, we have a continuum of solutions with the same energy, labeled by
the width of the throat connecting the flavor Dp-branes, U0 . As mentioned above, these solutions
are associated with the non-locality of the theory of the fivebranes.
For q = 6, combining (2.17) and (2.21) leads to
7 L3 .
(2.24)
In this case, validity of the supergravity analysis requires that the effective coupling at the scale
1/L be small, while that at the scale U0 should still be large (2.21). Note that the two requirements are consistent since U0 1/L and the coupling on the D6-branes increases with
energy.
At first sight it is surprising that the supergravity approximation should be valid at weak
coupling, where we already have a good description of the dynamics in terms of a low-energy
field theory. This is a manifestation of the non-decoupling of the D6-branes from bulk gravity.
The curved brane solution we found in this case describes the interactions of the fermions with
gravity, and it is not surprising that it is unstable. The interactions between the fermions and the
field theoretic degrees of freedom living on the color branes take place at much smaller values
of U , and cannot be described using supergravity (see [22] for related comments).
To summarize, we are led to a natural generalization of the picture in [13] to the system
with probe D-branes. For q 4, the supergravity analysis describes the strong coupling behavior
of the intersecting brane system. For q = 5, 6 it instead describes the interactions of the fields
associated with the intersection with non-field theoretic degrees of freedom, the LST modes
living in the throat of the fivebranes for q = 5, and gravity modes for q = 6.
3. D4D4 system
Our main purpose in the rest of this paper is to illustrate the general considerations of Section 2
in some examples. We have arranged the discussion by the dimension of the color branes. In this
section we focus on color D4-branes; in the next two we discuss systems with higher and lowerdimensional color branes respectively.
Two models with color D4-branes were discussed in detail in [2,7,10]. In both the intersections
were transverse. A model with a non-transverse intersection can be constructed with color D4branes intersecting flavor D6- and D6-branes on an I3 . The supergravity analysis of this model
was performed in [5].
11
In this section we will study one additional example with a non-transverse intersection. The
color and flavor branes are in this case both D4-branes, and are oriented as follows:
D4c
D4f , D4f
0
x
x
1
x
x
2
x
3
x
4
x
9
(3.1)
The color D4-branes are stretched in (01234) and located at the origin in (56789). The flavor
D4- and D4-branes are stretched in (01567) and separated by a distance L in (234). We will take
the separation to be in the x 4 direction and study the dynamics in the directions common to the
different branes, (01).
The subgroup of the Lorentz group preserved by this brane configuration is
SO(1, 1)01 SO(2)23 SO(3)567 SO(2)89 .
(3.2)
Further global symmetry arises from the gauge symmetry on the flavor D4-branes,
U (Nf )D4 U (Nf )D4 .
(3.3)
Comparing to the discussion of Section 2.1, we see that since the number of DN directions for
strings stretched between the color and flavor branes is equal to six, and this number is invariant
under T -duality, this system is T -dual to D3D9. Therefore r = 3, while the dimension of the
intersection is r = 1. Indeed, the intersection is not transverse as the color and flavor branes
can be separated in the directions (89), which are orthogonal to both. This deformation gives
mass to the fermions at each intersection and breaks the SO(2)89 symmetry (3.2). Since we are
interested in dynamical rather than explicit breaking of chiral symmetry, in what follows we set
the separation of the branes in the (89) direction to zero.
The low-energy degrees of freedom in this case are open strings stretched between color
branes, which give rise to (4+1)-dimensional SYM theory with sixteen supercharges, and strings
stretched between color and flavor branes, which give spacetime fermions. To see how these
fermions transform under the global symmetries, one can proceed as follows. If all the branes in
(3.1) were extended in the (89) directions instead of being localized in them, the SO(1, 1)01
SO(2)89 symmetry in (3.2) would have been extended to SO(1, 3)0189 . The spectrum at each
intersection would then be the same as in [2,7], i.e., a left-handed Weyl fermion, qL , at one
intersection, and a right-handed one, qR , at the other.
The configuration (3.1) can be obtained by compactifying (x 8 , x 9 ) on a torus, applying
T -duality in both directions and decompactifying back, in the process breaking the SO(1, 3)
symmetry back to SO(1, 1) SO(2). The massless states at each intersection are invariant under
this operation, so all we have to do is decompose the left- and right-handed SO(1, 3)0189 spinors
qL , qR under SO(1, 1)01 SO(2)89 :

L+
R+
qL =
(3.4)
;
qR =
.
R
L
and denote fermions localized at the two intersections. The subscripts (L, R) and (+, )
on the right-hand sides keep track of chirality in (01) and (89), respectively. For example, L+
in (3.4) is a complex, left-moving (one component) spinor field in 1 + 1 dimensions, with charge
, is a left-moving fermion with
+1/2 (i.e., half that of a vector) under SO(2)89 . Its adjoint, L+
the opposite SO(2)89 charge. Note that unlike the (3 + 1)-dimensional system discussed in [2,7]
and the (1 + 1)-dimensional one of [10], here there are left- and right-handed fermions at each
12
intersection. Thus, the two U (Nf ) factors in (3.3) no longer act purely on left- and right-handed
fermions.
In addition to their Lorentz charges, the fermions (3.4) transform in the fundamental (Nc )
representation of the color gauge group U (Nc ). Under the global symmetry (3.3), the fermions
L+ , R transform as (N f , 1), while R+ , L transform as (1, N f ).
Much of the discussion of the D4D6 system in [10] carries through with little change. In
particular, for L the color degrees of freedom are weakly coupled, and the dynamics of the
fermions is described by a Lagrangian of the form

Seff = i d 2 x qL qL + qR qR
+
g52
4 2

d 2 x d 2 y G5 (x y, L) qL (x) qR (y) qR (y) qL (x) ,
(3.5)
which is obtained by integrating out the color gauge field in the single gluon approximation.
G5 (x, L) is the five-dimensional massless propagator, (2.4).
The SU(Nc ) singlet fermion bilinear that enters the four Fermi interaction (3.5) can be expressed in terms of two-dimensional fermions as follows:
(x) R+ (y) + R
(x) L (y).
qL (x) qR (y) = L+
(3.6)
The resulting four-Fermi interaction is not equivalent to a Thirring model for U (Nc ). This is a
direct consequence of the fact that the color gluons that are exchanged by the fermions at the
two intersections and give rise to (3.5) include vectors and scalars under the (1 + 1)-dimensional
Lorentz group associated with the intersection.
Nevertheless, the theory can be solved at large Nc using standard methods, as in [10]. The
fermion bilinear (3.6) develops dynamically a non-zero vacuum expectation value. This breaks
U (Nf )D4 U (Nf )D4 U (Nf )diag .
(3.7)
Despite appearances, this symmetry breaking is chiral. Indeed, defining Q1 and Q2 to be the
U (1) generators in U (Nf )D4 and U (Nf )D4 , respectively, and R to be the generator of SO(2)89 ,
the combination6
Q5 = Q1 Q2 + 2R
(3.8)
acts chirally on the fermions. The left-moving fermions L+ , L have charge +2 and 2,
respectively, while the right-moving fermions are neutral. The symmetry (3.8) is preserved by
the action (3.5), and (if it is a symmetry of the vacuum) prevents the generation of a mass for the
quarks (3.4). The quark bilinear (3.6) has charge 2 under it. Thus, if it develops an expectation
value, the symmetry is broken and a quark mass can be generated.
A very similar analysis to that of [10] shows that the expectation value (3.6) takes again the
form

d 2k
eikx
qL (x) qR (0) = Nc mf
(3.9)
,
2
2
(2) k + m2f
|k|<
6 Note that Q is a linear combination of a symmetry that is broken by (3.7), Q Q , and one that is preserved, R.
5
1
2
13
with the dynamically generated mass mf given by

mf e
(3.10)
is the UV cutoff of the theory, 1/L. In this case, one can also analyze the system in the
presence of mass terms in the Lagrangian which explicitly break the chiral symmetry (3.8),
Leff = m1 L+
R + m2 R+
L + c.c.,
(3.11)
corresponding to separating the color and flavor branes in the (89) plane. This leads to a straightforward generalization of the analysis in [10].
As discussed in Section 2, at strong coupling the dynamics of the fermions and color degrees
of freedom can be described by studying the DBI action of the flavor D4-branes in the nearhorizon geometry of the color branes, (2.11).
The U -shaped solution in which the flavor branes and anti-branes are connected by a wormhole is given by (2.15),
x 4 (U ) =

4

3/2

U0 5 1
1 R5
5 1
;
,
B
,
.
B
4 U 1/2
8 2
U4 8 2
(3.12)
The energy difference between the straight brane and anti-brane configuration, and the U -shaped
one (3.12) is given by (2.20),

3
3
1
3 1
E = C(4, 4, 1)U02 B ,
(3.13)
0.225C(4, 4, 1)U02 .
4
8 2
Thus, the vacuum of the theory breaks chiral symmetry both for weak coupling and for strong
coupling, in agreement with the general analysis of Section 2. One can also analyze the system
for finite temperature, as was done for the D4D8 case in [24,25] and for the D4D6 case in [10].
Overall, we conclude that the D4D4 system behaves in a very similar way to the D4D6 one
analyzed in [10]. At weak coupling it reduces to a GN-type model which can be analyzed using
field theoretic techniques, and at strong coupling it can be studied using the DBI action for the
flavor branes in the near-horizon geometry of the color ones. One advantage of this system is
that one can turn on current masses to the fermions and study the dynamics as a function of these
masses. Another advantage is that the D4D4 brane configuration is simple to lift to M-theory,
where the color D4-brane background goes over to AdS7 S 4 , and the flavor D4-branes and
anti-branes become M5-branes in this background. These and other issues deserve further study.
4. Higher-dimensional color branes
In Section 2 we saw that when the color branes are higher than four-dimensional, the supergravity analysis exhibits some qualitative differences from the case when their dimension is four
or less. In this section we will examine some examples with color D5- and D6-branes to study
these phenomena in more detail.
4.1. Color fivebranes
In this subsection we discuss a few intersecting brane systems in which the color branes are
(5 + 1)-dimensional. They can be further subdivided by the type of flavor branes, the dimension
14
Table 1
Different systems with color D5-branes that are discussed in this section
Flavor branes
Dimension of intersection
Range of interaction
D5
D3
D5
D7
1+1
1+1
3+1
3+1
L
L
of the intersection and the range of the non-local four-Fermi interaction at weak coupling. In
Table 1 we list the four systems that will be discussed below.
Our first example is obtained by T -dualizing the D4D6 system discussed in [10]. It contains color D5-branes stretched in (012345), transversally intersecting flavor D5- and D5-branes
stretched in (016789). A single intersection of this sort was studied in [26,27]. The main new
phenomenon here is the attractive interactions between the fermions at the two intersections.
As before, we can try to analyze this system using QFT techniques at weak coupling, and
supergravity at strong coupling. The weak coupling analysis is quite analogous to that of [10].
The fermions at the two intersections, qL , qR , are chiral. Their dynamics is governed by the
effective action (3.5), with the coupling g5 replaced by g6 (2.1) and the Green function G5 (x, L)
replaced by G6 (x, L), (2.4). It is integrable,
d 2 x G6 (x, L) =
(4.1)
.
2L2
Thus, at distances much larger than L the system reduces to the GN model with action

gn
Sgn = d 2 x iqL qL + iqR qR +
(4.2)
qL (x) qR (x) qR (x) qL (x)
Nc
and coupling
6
(4.3)
.
2L2
In particular, it dynamically breaks chiral symmetry and generates a fermion condensate (3.9),
which leads to the fermion mass
1
mf e2/gn .
(4.4)
L
As before, this analysis is reliable for 6 L2 and breaks down when this condition is violated.
The DBI analysis does give a solution in which the flavor D5- and D5-branes are connected by a
wormhole whose shape is given by (2.15),

4

U0 1 1
1
1 1
;
B
,
.
x 5 (U ) = R6 B ,
(4.5)
4
2 2
U4 2 2
gn =
However, the asymptotic separation of the D5- and D5-branes (2.18) is fixed,
1
L = R6 .
(4.6)
2
For this value of L there are solutions with arbitrary width U0 , whose energy is independent
of U0 . This is different from the situation in systems with color D4-branes, where there is a solution for generic L, and the width of the wormhole U0 is a function of L, growing when L
decreases.
15
What does this mean for the dynamics of the fermions living at the two intersections? For
L R6 the field theoretic GN analysis is valid. Chiral symmetry is dynamically broken, and
the fermions get a mass (4.4). As L decreases, the coupling (4.3) grows, and the dynamically
generated mass (4.4) does as well. The system probes higher and higher energies in the fivebrane
theory. As L approaches the value (2.18) the dynamics becomes dominated by high energy LST
states.
The resulting physics is not field theoretic and its analysis is beyond the scope of this paper. One expects non-smooth behavior at the non-locality scale (4.6), and it is not clear whether
the system exists for smaller L. The problem is analogous to the analysis of fivebrane thermodynamics, with the inverse temperature being the analog of L and the energy density on the
fivebranes an analog of the fermion mass mf . The curved solution (4.5) is an analog of the
Euclidean continuation of the non-extremal fivebrane solution. The latter has the property that
the circumference of Euclidean time at infinity, , is independent of the energy density, just like
in (4.5) the asymptotic separation between the two arms of the U -shape, L, is independent of the
fermion mass (or U0 ).
In the case of fivebrane thermodynamics it is believed that the Euclidean black hole solution is
not continuously connected to the low temperature thermodynamics (see [12] for a discussion). It
would be interesting to understand whether in our case the solution (4.5) is continuously related
to the large L regime, and what happens for L smaller than (4.6).
We next move on to the brane system on the second line in Table 1, which is obtained by
T -duality from the one discussed in Section 3. It contains color D5-branes stretched in (012345)
and flavor D3- and D3-branes stretched in (0167).
At weak coupling this system reduces to a GN-type model of the sort discussed in Section 3.
In particular, it exhibits dynamical symmetry breaking. At strong coupling one needs to analyze
the DBI action (2.11), which leads to the brane profile

2

U0 1 1
1
1 1
;
B
,
.
x 5 (U ) = R6 B ,
(4.7)
2
2 2
U2 2 2
This looks very similar to the D5D5 solution (4.5). As there, the asymptotic separation (2.18)
is fixed, L = R6 . Here too we expect chiral symmetry breaking for sufficiently small values
of the coupling 6 /L2 , and non-smooth behavior when the coupling approaches the critical
value (2.18).
The last two lines in Table 1 correspond to systems with (3 + 1)-dimensional intersections.
In one, the flavor branes are D5- and D5-branes stretched in (012367). In the other, they are D7and D7-branes stretched in (01236789).
In the model with flavor D5-branes, each intersection preserves eight supercharges. Hence the
fermions living at a given intersection belong to a hypermultiplet. SUSY is completely broken in
the full system and exchange of fields living on the color branes leads to an attractive interaction
between the hypermultiplets localized at the two intersections. At weak coupling, this attractive
interaction has a structure similar to (3.5), however unlike the system analyzed in [10] and those
discussed earlier in this paper, the Green function G6 (x) is not integrable in this case:

d 4x
d 4 x G6 (x) =
(4.8)
2
(x + L2 )2
is logarithmically divergent, so the attractive interaction has infinite range. Such systems are in
general more subtle than those with a short-range interaction. We hope to return to their study in
a separate publication.
16
In the supergravity approximation, the solution of the equations of motion of the DBI action (2.11) is again given by (4.5), and we find that there is a curved brane solution in the
supergravity regime for L given by (4.6). The interpretation is the same as there.
The situation is similar for the last brane configuration in Table 1, which has flavor D7-branes
and a (3 + 1)-dimensional intersection with the color D5-branes. This model is T -dual to the
D4 D8 = I3 system studied in [7]. Thus the spectrum is the same as there: a left-handed fermion
qL in the (Nc , N f , 1) of U (Nc ) U (Nf )L U (Nf )R at the D5D7 intersection, and a righthanded fermion qR in the (Nc , 1, N f ) at the D5D7 one. The leading single gluon exchange
interaction between the left- and right-handed fermions takes a form similar to (3.5) (or, more
precisely, Eq. (3.5) in [7]). As in the previous example, this leads to a long-range interaction.
In the supergravity approximation, the solution to the DBI equations of motion is

6

U0 1 1
1
1 1
5
; ,
B
.
x (U ) = R6 B ,
(4.9)
6
2 2
U6 2 2
It again has a fixed value of L (2.18) at which the dynamics becomes dominated by highly excited
LST states.
Before leaving the case of color fivebranes we would like to point out that the curved brane
solutions we found for this case are closely related7 to the hairpin brane of [28,29] which plays
a role in analyzing the dynamics of D-branes propagating in the vicinity of NS5-branes [30,31].
They can be thought of as continuations to Euclidean space of accelerating brane solutions. This
also makes it plausible that, as mentioned above, they owe their existence to the interactions of
the fields at the intersections with the continuum of modes living in the fivebrane throat, i.e., to
Little String Theory dynamics [11,12].
4.2. Color sixbranes
As mentioned above, due to lack of decoupling, it is not clear whether intersecting brane
systems involving color D6-branes make sense beyond the field theory approximation (i.e., for
finite L). Nevertheless, in this subsection we will discuss two examples of such systems using
the tools outlined in Section 2.
The first system is obtained by reversing the roles of the D4- and D6-branes in the configuration studied in [10]. Thus, we have Nc color D6-branes stretched in (0156789) and Nf flavor D4and D4-branes stretched in (01234) and separated by a distance L in (56789). At weak coupling
the low-energy dynamics is again governed by a GN model which breaks chiral symmetry and
generates a mass for the fermions. The GN coupling gn (4.2) is proportional to 7 /L3 as can be
verified by integrating the (6 + 1)-dimensional propagator (2.4).
From our experience with the D5-brane case, we expect the supergravity analysis to be more
subtle. The DBI action leads in this case to the solution

3

U0 1 1
1
1 1
R7 U 0 B ,
;
B
,
.
x 6 (U ) =
(4.10)
3
3 2
U3 3 2
The asymptotic separation between the flavor branes and anti-branes is

1 1
2
R7 U 0 B ,
.
L=
3
3 2
7 In fact for the D5 D3 = I system they are S-dual.
1
(4.11)
17
Thus, unlike the fivebrane case, here there is a solution in which the flavor branes are connected by a wormhole for generic L.
As discussed in Section 2, the supergravity analysis is valid at large L, (2.24). In that region
the dynamics of the fermions is described by the GN model. The curved brane solution (4.10)
is not a consequence of that dynamics. Instead, its existence is due to gravitational interactions
of the fermions in the vicinity of the D6-branes. This dynamics should be unimportant at low
energies. Indeed, the solution (4.10) is unstable. The energy difference (2.20) is given by

1
2 1
E U02 B ,
(4.12)
0.351U02 .
3
3 2
One can show that the curved brane configuration is unstable to perturbations of the form
U (x 6 ) U (x 6 ) + U with U independent of x 6 . Thus, in the supergravity approximation
the vacuum corresponds to straight branes. As mentioned above, the GN analysis implies that
the flavor branes do curve towards each other and connect, but this happens at a much smaller
value of U than (4.11) and is well outside the regime of validity of supergravity.
We do not have a good description of chiral symmetry breaking in the strongly coupled regime
L3 . As mentioned above, it is likely that this is because the system does not exist in that
regime, or more generally for any finite /L3 .
Our second example consists of D6- and D6-branes with a (3 + 1)-dimensional intersection
(i.e., D6D6 = I3 ). The color branes can be taken to lie in (0123456), while the flavor branes and
anti-branes are stretched in (0123789) and separated by a distance L in x 6 . This configuration
is T -dual to the D4D8D8 one studied in [2,7]. The spectrum contains left-handed fermions at
one intersection, and right-handed ones at the other.
At weak coupling these fermions interact via the four-Fermi interaction given in Eq. (3.5)
in [7]. However, here this interaction is local, since

d 4x
d 4 x G7 (x, L) =
(4.13)
5
(x 2 + L2 ) 2
is finite. Therefore, the model reduces at long distances x L to a local NJL model [20], with
coupling 7 /L which has dimension length squared. This model does not break chiral symmetry
at arbitrarily weak coupling. Hence the same should be true for the intersecting D-brane system
in the limit L3 7 .
The supergravity analysis leads to results that are qualitatively similar to the previous case.
The DBI action (2.11) leads to a solution corresponding to flavor branes and anti-branes connected by a wormhole

4

U0 3 1
1
3 1
x 6 (U ) =
(4.14)
R7 U 0 B ,
;
B
,
.
4
8 2
U4 8 2
This solution is again valid for large L, (2.24), and is unstable,

5
1 52
5 1
E U0 B ,
0.193U02 .
4
8 2
(4.15)
5. Lower-dimensional color branes

In this section we will discuss intersecting brane configurations with D2- and D3-color branes.
The low-energy theories on these branes are renormalizable gauge theories which can be decoupled from gravity. For D3-branes this theory is N = 4 SYM. It is conformal, and its effective
18
coupling (1.1) is independent of the separation L. For D2-branes the theory is weakly coupled in
the UV and strongly coupled in the IR. Thus, the effective coupling (1.1) grows as L increases.
Due to the low dimension of the color branes, all our examples involve (1 + 1)-dimensional
intersections. Thus, the flavor branes are codimension one or two defects in the gauge theory. The
single gluon exchange interaction (2.3) relevant for the weak coupling analysis is a long range
one, since the integral (2.5) diverges. We will not discuss the weakly coupled theory in detail
here.
5.1. Color threebranes
Our first example contains Nc color D3-branes stretched in (0123) and Nf flavor D7- and
D7-branes stretched in (01456789) separated by the distance L in x 3 . This configuration is
T -dual to the D4D6 one considered in [10]. Therefore, it has the same massless spectrum
of fermions at the intersectionsa left-handed fermion qL transforming in the (Nc , N f , 1) of
U (Nc ) U (Nf )L U (Nf )R at the D3D7 intersection and a right-handed fermion qR in the
(Nc , 1, N f ) at the D3D7 intersection.
At weak coupling the leading interaction of the fermions is a long range GN coupling due to
single gluon exchange. At strong coupling, the D3-branes are replaced by their near-horizon
geometry, AdS5 S 5 , and the D7-branes are described by the DBI action (2.11). The solution (2.15) takes in this case the form

6

U0 2 1
1 R42
2 1
3
x (U ) =
(5.1)
; ,
B
.
B ,
6 U0
3 2
U6 3 2
The distance between the D7- and D7-branes is

R2
2 1
L=
B ,
.
3U0
3 2
(5.2)
One can check using (2.20) that this solution has lower energy than the one in which the flavor
branes are stretched in U at fixed values of x 3 . Thus, at strong coupling
(4 1) the chiral
symmetry is dynamically broken, and the fermions get a mass of order 4 /L.
Replacing the flavor branes by D5-branes extended in (014567) leads to a T -dual of the
D4D4 system studied in Section 3. Therefore, the spectrum is the same as there. It consists of
fermions L+ , R transforming in (Nc , N f , 1) and R+ , L transforming in (Nc , 1, N f ) of
U (Nc ) U (Nf )D5 U (Nf )D5 . At weak coupling there are long range interactions between the
fermions, the leading of which is given by (3.5), with G5 replaced by G4 . At strong coupling,
the DBI action of the fivebranes is proportional to that of the D7-branes discussed above. Thus,
the solution is given again by (5.1), (5.2) and chiral symmetry is broken at strong coupling.
5.2. Color twobranes
A system with color D2-branes and flavor D8- and D8-branes was considered recently in [9].
The D2-branes can be taken to lie in the directions (012), while the D8-branes span the directions
(013456789). This system is T -dual to the D3D7 one described above, so the massless spectrum
at the intersection is the same as there. At weak coupling (which in this case means small L) it
reduces to a non-local GN model, while at strong coupling (large L) it breaks chiral symmetry,
as follows from the DBI analysis of Section 2.3.
19
Replacing the D8-branes by D6-branes stretched in the directions (0134567) one finds
a D2D6 system T -dual to the D4D4 one of Section 3. At strong coupling, the DBI analysis of Section 2 leads to the solution
5

8

2

R
U0 11 1
1
11 1
3
2
x (U ) =
; ,
,
B
.
B
8 32
16 2
U 8 16 2
U0
(5.3)
This solution breaks chiral symmetry and has lower energy than the symmetry preserving one,
as in all other cases with Dq color branes with q 4.
6. Conclusions
In this paper we presented an analysis of dynamical symmetry breaking in a class of intersecting D-brane systems which generalize those investigated earlier in [2,7,9,10]. At weak coupling
these models are usefully classified by the range of the effective interaction between fermions
localized at the intersections. The models with short range interactions are straightforward to analyze. When the intersection is (1 + 1)-dimensional, they exhibit symmetry breaking of the type
studied by Gross and Neveu [19]. In 3 + 1 dimensions they reduce to variants of the Nambu
JonaLasinio model [20] and do not lead to symmetry breaking at weak coupling.
We also described the dynamics of these systems in the approximation where we replace
the color branes by their near-horizon geometry and study the DiracBornInfeld action of the
flavor branes in this geometry. As expected from [13,22], the results of this analysis depend on
the dimensionality of the color branes. For Dq color branes with q 4 it provides a holographic
description of the corresponding field theory at strong coupling. For all such systems the ground
state exhibits dynamical symmetry breaking.
For q > 4, the DBI analysis does not describe the strong coupling behavior of the field theory
at the intersection, but rather the interaction of modes associated with the intersection with other,
non-field theoretic degrees of freedom. For q = 5 these are Little String Theory modes that
propagate in the throat of the fivebranes. Their interactions with the modes at the intersections
lead to the existence of flavor brane configurations in the near-horizon geometry, all of the same
energy, which are labeled by the width of the throat, U0 . The asymptotic separation of the branes
has a particular value, (2.18), for these solutions. We interpreted this value as the non-locality
scale of LST for these probes. For q = 6 the modes in question are gravity modes of the full
theory. Their interaction with the fermions at the intersection leads to the existence of unstable
configurations of the flavor branes which exhibit dynamical symmetry breaking.
The main gap in our discussion is the weak-coupling analysis of intersecting brane systems
with long-range fermion interactions. These systems are subtle, but also potentially important as
they arise in embeddings of QCD in string theory in the limit where the scale of chiral symmetry
breaking is much larger than that of confinement [7]. We hope to return to them elsewhere.
One of the main motivations for this work was to see whether intersecting brane configurations
of the sort described in [2,7,9,10] and in this paper, are always in the same phase as far as
dynamical symmetry breaking is concerned (in the limit where the color branes are non-compact,
so there is no confinement). In all cases where we were able to analyze the dynamics for both
weak and strong coupling we found that there was no phase transition between the weak and
strong coupling regimes for Dq color branes with q 4. It is natural to conjecture that this is
always the case. A better understanding of weakly coupled systems with long range interactions
would allow us to test this conjecture in a wider class of models.
20
For q = 5 one does not expect such smoothness to extend beyond the non-locality scale of the
underlying LST, and indeed we found signs of this in the existence of curved branes solutions for
a particular value of L, (2.18). It would be interesting to understand the physics of these solutions
in LST better. For q = 6 it is not clear that the brane configurations in question exist for finite L,
so the question of smoothness in L does not arise.
Acknowledgements
We thank O. Aharony, N. Itzhaki and K. Skenderis for discussions. J.H. and D.K. thank the
Aspen Center for Physics for providing a supportive atmosphere during the completion of this
work. The work of E.A. and J.H. was supported in part by NSF Grant No. PHY-00506630.
D.K. was supported in part by DOE grant DE-FG02-90ER40560. This work was also supported
in part by the National Science Foundation under Grant 0529954. Any opinions, findings, and
conclusions or recommendations expressed in this material are those of the authors and do not
necessarily reflect the views of the National Science Foundation.
References
[1] E. Witten, Anti-de Sitter space, thermal phase transition, and confinement in gauge theories, Adv. Theor. Math.
Phys. 2 (1998) 505, hep-th/9803131.
[2] T. Sakai, S. Sugimoto, Low energy hadron physics in holographic QCD, Prog. Theor. Phys. 113 (2005) 843, hepth/0412141.
[3] A. Karch, E. Katz, Adding flavor to AdS/CFT, JHEP 0206 (2002) 043, hep-th/0205236.
[4] D.T. Son, M.A. Stephanov, Phys. Rev. D 69 (2004) 065020, hep-ph/0304182.
[5] M. Kruczenski, D. Mateos, R.C. Myers, D.J. Winters, Towards a holographic dual of large-N (c) QCD, JHEP 0405
(2004) 041, hep-th/0311270.
[6] J. Babington, J. Erdmenger, N.J. Evans, Z. Guralnik, I. Kirsch, Chiral symmetry breaking and pions in nonsupersymmetric gauge/gravity duals, Phys. Rev. D 69 (2004) 066007, hep-th/0306018.
[7] E. Antonyan, J.A. Harvey, S. Jensen, D. Kutasov, NJL and QCD from string theory, hep-th/0604017.
[8] D. Bak, H.U. Yee, Separation of spontaneous chiral symmetry breaking and confinement via AdS/CFT correspondence, Phys. Rev. D 71 (2005) 046003, hep-th/0412170.
[9] Y.H. Gao, W.S. Xu, D.F. Zeng, NGN, QCD(2) and chiral phase transition from string theory, hep-th/0605138.
[10] E. Antonyan, J.A. Harvey, D. Kutasov, The GrossNeveu model from string theory, hep-th/0608149.
[11] O. Aharony, M. Berkooz, D. Kutasov, N. Seiberg, Linear dilatons, NS5-branes and holography, JHEP 9810 (1998)
004, hep-th/9808149.
[12] O. Aharony, A. Giveon, D. Kutasov, LSZ in LST, Nucl. Phys. B 691 (2004) 3, hep-th/0404016.
[13] N. Itzhaki, J.M. Maldacena, J. Sonnenschein, S. Yankielowicz, Supergravity and the large N limit of theories with
sixteen supercharges, Phys. Rev. D 58 (1998) 046004, hep-th/9802042.
[14] J. Polchinski, Superstring Theory and Beyond, String Theory, vol. 2, Cambridge Univ. Press, Cambridge, UK, 1998.
[15] J. Distler, F. Zamora, Chiral symmetry breaking in the AdS/CFT correspondence, JHEP 0005 (2000) 005, hep-th/
9911040.
[16] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: Duality cascades and chiSB-resolution of
naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[17] N.J. Evans, J.P. Shock, Chiral dynamics from AdS space, Phys. Rev. D 70 (2004) 046002, hep-th/0403279.
[18] B.A. Burrington, J.T. Liu, M. Mahato, L.A. Pando Zayas, Towards supergravity duals of chiral symmetry breaking
in SasakiEinstein cascading quiver theories, JHEP 0507 (2005) 019, hep-th/0504155.
[19] D.J. Gross, A. Neveu, Dynamical symmetry breaking in asymptotically free field theories, Phys. Rev. D 10 (1974)
3235.
[20] Y. Nambu, G. Jona-Lasinio, Dynamical model of elementary particles based on an analogy with superconductivity. I,
Phys. Rev. 122 (1961) 345.
[21] J.M. Maldacena, A. Strominger, Semiclassical decay of near-extremal fivebranes, JHEP 9712 (1997) 008, hep-th/
9710014.
[22] A.W. Peet, J. Polchinski, UV/IR relations in AdS dynamics, Phys. Rev. D 59 (1999) 065011, hep-th/9809022.
21
[23] K. Skenderis, Lecture notes on holographic renormalization, Class. Quantum Grav. 19 (2002) 5849, hep-th/0209067.
[24] O. Aharony, J. Sonnenschein, S. Yankielowicz, A holographic model of deconfinement and chiral symmetry restoration, hep-th/0604161.
[25] A. Parnachev, D.A. Sahakyan, Chiral phase transition from string theory, hep-th/0604173.
[26] M.B. Green, J.A. Harvey, G.W. Moore, I-brane inflow and anomalous couplings on D-branes, Class. Quantum
Grav. 14 (1997) 47, hep-th/9605033.
[27] N. Itzhaki, D. Kutasov, N. Seiberg, I-brane dynamics, JHEP 0601 (2006) 119, hep-th/0508025.
[28] S.L. Lukyanov, E.S. Vitchev, A.B. Zamolodchikov, Integrable model of boundary interaction: The paperclip, Nucl.
Phys. B 683 (2004) 423, hep-th/0312168.
[29] S.L. Lukyanov, A.B. Zamolodchikov, Dual form of the paperclip model, Nucl. Phys. B 744 (2006) 295, hep-th/
0510145.
[30] D. Kutasov, D-brane dynamics near NS5-branes, hep-th/0405058.
[31] D. Kutasov, Accelerating branes and the string/black hole transition, hep-th/0509170.
Linear relations and their breakdown in high energy

massive string scatterings in compact spaces
Jen-Chi Lee a,b,,1 , Yi Yang a,c,2
a Institute of Physics, University of Tokyo, Komaba, Meguro-ku, Tokyo 153-8902, Japan
b Department of Electrophysics, National Chiao-Tung University, Hsinchu, Taiwan, ROC
c Physics Division, National Center for Theoretical Sciences, Hsinchu, Taiwan, ROC
Received 21 May 2007; accepted 1 June 2007

Abstract
We calculate high energy massive scattering amplitudes of closed bosonic string compactified on the
1
torus. For each fixed mass level with given quantized and winding momenta ( m
R , 2 nR), we obtain infinite
linear relations among high energy scattering amplitudes of different string states. For some kinematic
regimes, we discover that linear relations with NR = NL break down and, simultaneously, the amplitudes
enhance to power-law behavior instead of the usual exponential fall-off behavior at high energies. It is the
spacetime T-duality symmetry that plays a role here. This result is consistent with the coexistence of the
linear relations and the softer exponential fall-off behavior of high energy string scattering amplitudes as
we pointed out previously. It is also reminiscent of our previous work on the power-law behavior of high
energy string/domain-wall scatterings.
1. Introduction
It is well known that there are two fundamental characteristics of high energy string scattering
amplitudes, which make them very different from field theory scatterings. These are the softer
exponential fall-off behavior (in contrast to the power-law behavior of field theory scatterings)
* Corresponding author at: Department of Electrophysics, National Chiao-Tung University, Hsinchu, Taiwan, ROC.
E-mail addresses: jcclee@cc.nctu.edu.tw (J.-C. Lee), yyang@phys.cts.nthu.edu.tw (Y. Yang).

1 On leave of absence from National Chiao-Tung University.
2 On leave of absence from National Center for Theoretical Sciences.
doi:10.1016/j.nuclphysb.2007.06.005
J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235
23
and the existence of infinite Regge-pole structure in the form factor of the high energy string
scattering amplitudes.
Recently high-energy, fixed angle behavior of string scattering amplitudes [13] was intensively reinvestigated for massive string states at arbitrary mass levels [412]. An infinite number
of linear relations, or stringy symmetries, among string scattering amplitudes of different string
states were obtained. An important new ingredient of these calculations is the zero-norm states
(ZNS) [1315] in the old covariant first quantized (OCFQ) string spectrum. The existence of
these infinite linear relations constitutes the third fundamental characteristics of high energy
string scatterings, which is not shared by the usual point-particle field theory scatterings.
These linear relations persist for string scattered from generic Dp-brane [16] except
D-instanton and Domain-wall. For the scattering of D-instanton, the form factor exhibits the
well-known power-law behavior without Regge-pole structure, and thus resembles a field theory
amplitude. For the special case of Domain-wall (D24-brane) scattering, it was discovered [17]
recently that, in contrast to the common wisdom of exponential fall off behavior [18,19], its form
factor behaves as power-law with Regge-pole structure. This discovery makes Domain-wall scatterings a hybrid of string and field theory scatterings. Moreover, it was shown [17] that the linear
relations break down for the Domain-wall scattering due to this unusual power-law behavior.
This result gives a strong evidence that the existence of the infinite linear relations, or stringy
symmetries, of high-energy string scattering amplitudes is responsible for the softer (exponential
fall-off) high-energy string scatterings than the (power-law) field theory scatterings.
To further convince ourselves with the coexistence of the infinite linear relations and the softer
exponential fall-off behavior of string scatterings at high energies, it is important to find more
examples of high energy string scatterings, which show the unusual power-law behavior and,
simultaneously, give the breakdown of the infinite linear relations. With this in mind, in this paper we calculate high energy massive scattering amplitudes of closed bosonic string with some
coordinates compactified on the torus [20,21]. For each fixed mass level with given quantized
1
and winding momenta ( m
R , 2 nR), we obtain infinite linear relations among high energy scattering amplitudes of different string states. This result is reminiscent of the existence of an infinite
number of massive soliton ZNS in the compactified string constructed in [22]. We then discover
that, for some kinematic regime, so-called Mende regime (MR), infinite linear relations with
NR = NL break down and, simultaneously, the amplitudes enhance to power-law behavior instead of the usual exponential fall-off behavior at high energies. It is the spacetime T-duality
symmetry that plays a role here.
The power-law behavior of high energy string scatterings in a compact space was first suggested by Mende [20]. Here we give an explicit calculation of the conjecture. Moreover, in
addition to the high energy string/domain-wall scatterings mentioned above [17], our result in
this paper gives another evidence to support the coexistence of the infinite linear relations and
the softer exponential fall-off behavior of high energy string scattering amplitudes as we pointed
out previously [17,23]. The result also suggests that the infinite linear relations are closely related
to the full 26D spacetime symmetry of closed bosonic string theory. This paper is organized as
following. In Section 2, we set up the kinematic for the compactified string and calculate the
four-tachyon (NR = NL = 0) scattering amplitudes with arbitrary winding. In Section 3, we derive the infinite linear relations among high energy scattering amplitudes of different string states
1
with given ( m
R , 2 nR) for each fixed mass level. We then discuss the power-law behavior of the
amplitudes and breakdown of the infinite linear relations in the Mende regime. A brief conclusion
is given in Section 4.
24
2. String compactified on torus

2.1. Winding string and kinematic setup
We consider 26D closed bosonic string with one coordinate compactified on S 1 with radius R.
As we will see later, it is straightforward to generalize our calculation to more compactified
coordinates. The closed string boundary condition for the compactified coordinate is (we use the
notation in [24])
X 25 ( + 2, ) = X 25 (, ) + 2Rn,
(2.1)
where n is the winding number. The momentum in the X25 direction is then quantized to be
m
K= ,
(2.2)
R
where m is an integer. The mode expansion of the compactified coordinate is
25
( ) + XL25 ( + ),
X 25 (, ) = XR
(2.3)
where
1
1
25
( ) = x + KR ( ) + i
25 e2ik( ) ,
XR
2
k k
k=0
1
1
25 e2ik( + ) .
XL25 ( + ) = x + KL ( + ) + i
2
k k
(2.4)
(2.5)
k=0
The left and right momenta are defined to be

m 1
1
nR K = (KL + KR ),
R 2
2
and the mass spectrum can be calculated to be

2
M2 = ( m
+ 14 n2 R 2 ) + NR + NL 2 KL2 + ML2 KR2 + MR2 ,
R2
NR NL = mn,
KL,R = K L =
(2.6)
(2.7)
where NR and NL are the number operators for the right and left movers, which include the
counting of the compactified coordinate. We have also introduced the left and the right level
masses as
2
2(NL,R 1).
ML,R
(2.8)
Note that for the compactified closed string NR and NL are correlated through the winding
modes.
In the center of momentum frame, the kinematic can be set up to be

k1L,R = + p 2 + M12 , p, 0, K1L,R ,
(2.9)

k2L,R = + p 2 + M22 , +p, 0, +K2L,R ,
(2.10)

k3L,R = q 2 + M32 , q cos , q sin , K3L,R ,
(2.11)

k4L,R = q 2 + M42 , +q cos , +q sin , +K4L,R
(2.12)
25
where p |p| and q |q| and

1
ki (kiR + kiL ),
2
ki2 = Ki2 Mi2 ,
2
kiL,R
2
= KiL,R
Mi2
(2.13)
(2.14)
2
MiL,R
.
With this setup, the center of mass energy E is

1

1 2
2
2
2
2
2
p + M 1 + p + M2 =
q + M3 + q 2 + M42 .
E=
2
2
The conservation of momentum on the compactified direction gives
m1 m2 + m3 m4 = 0,
(2.15)
(2.16)
(2.17)
and T-duality symmetry implies conservation of winding number

n1 n2 + n3 n4 = 0.
(2.18)
One can easily calculate the following kinematic relations

k1L,R k2L,R = p 2 + M12 p 2 + M22 + p 2 + K 1L,R K 2L,R
1

1
1 2
2
2
2
sL,R + k1L,R
= sL,R M1L,R
,
+ k2L,R
+ M2L,R
2
2
2

k2L,R k3L,R = p 2 + M22 q 2 + M32 + pq cos + K 2L,R K 3L,R
=
1

1
1 2
2
2
2
tL,R + k2L,R
= tL,R M2L,R
,
+ k3L,R
+ M3L,R
2
2
2

k1L,R k3L,R = p 2 + M12 q 2 + M32 pq cos K 1L,R K 3L,R
=
1

1
1 2
2
2
2
uL,R + k1L,R
+ k3L,R
+ M3L,R
= uL,R M1L,R
,
2
2
2
where the left and the right Mandelstam variables are defined to be
=
(2.19)
(2.20)
(2.21)
(2.22)
(2.23)
(2.24)
sL,R (k1L,R + k2L,R )2 ,
(2.25)
tL,R (k2L,R + k3L,R ) ,
(2.26)
uL,R (k1L,R + k3L,R )2 ,
(2.27)
with
sL,R + tL,R + uL,R =
2
MiL,R
.
(2.28)
2.2. Four-tachyon scatterings with NR = NL = 0

We are now ready to calculate the string scattering amplitudes. Let us first calculate the case
with NR + NL = 0 (or NR = NL = 0),
26

(N +N =0)
R
L
Aclosed
(s, t, u)

= d 2 z exp k1L k2L ln z + k1R k2R ln z

+ k2L k3L ln(1 z) + k2R k3R ln(1 z )

= d 2 z exp 2k1R k2R ln |z| + 2k2R k3R ln |1 z|

+ (k1L k2L k1R k2R ) ln z + (k2L k3L k2R k3R ) ln(1 z) ,
(2.29)
where we have used = 2 for closed string propagators

X(z)X(z ) = ln(z z ),
(2.30)
2

z)X(
z ) = ln(z z ).
X(
(2.31)
2
Note that for this simple case, Eq. (2.7) implies either m = 0 or n = 0. However, we will keep
track of the general values of (m, n) here for the reference of future calculations. By using the
formula [25]
2
d z
I=
|z| |1 z| zn (1 z)m
(1 12 12 )(1 + n + 12 )(1 + m + 12 )

=
(2.32)
,
( 12 )( 12 )(2 + n + m + 12 + 12 )
we obtain
(N +NL =0)
R
Aclosed
(s, t, u)
(1 k1R k2R k2R k3R )(1 + k1L k2L )(1 + k2L k3L )
=
(k1R k2R )(k2R k3R )(2 + k1L k2L + k2L k3L )
sin(k1R k2R ) sin(k2R k3R )
=
sin( k1R k2R k2R k3R )
(1 + k1R k2R )(1 + k2R k3R ) (1 + k1L k2L )(1 + k2L k3L )
(2 + k1R k2R + k2R k3R )

(2 + k1L k2L + k2L k3L )
tR
u
t
u
sin(sL /2) sin(tR /2) (1 2 )(1 2R ) (1 2L )(1 2L )
,

sR
sL
sin(uL /2)
(2 + 2 )
(2 + 2 )
(2.33)
2
where we have used MiL,R
= 2 for i = 1, 2, 3, 4. In the above calculation, we have used the
following well-known formula for gamma function
(x) =
(2.34)
.
sin(x)(1 x)
3. High energy behaviors

3.1. High energy massive scatterings for general NR + NL
We now proceed to calculate the high energy scattering amplitudes for general higher mass
levels with fixed NR + NL . With one compactified coordinate, the mass spectrum of the second
27
Fig. 1. Different regimes of high energy limit. The high energy regime defined by E 2 M22 NR + NL will be called
Mende regime (MR). The high energy regime defined by E 2 M22 , E 2 NR + NL will be called Gross regime (GR).
vertex of the amplitude is

2

m2 1 2 2
2
M2 =
+ n R + NR + NL 2.
R2 4 2
(3.1)
We now have more mass parameters to define the high energy limit. So let us first clear and
redefine the concept of high energy limit in our following calculations. We are going to use
three quantities E 2 , M22 and NR + NL to define different regimes of high energy limit. See
Fig. 1. The high energy regime defined by E 2 M22 NR + NL will be called Mende regime
(MR). The high energy regime defined by E 2 M22 , E 2 NR + NL will be called Gross
region (GR). In the high energy limit, the polarizations on the scattering plane for the second
vertex operator are defined to be

1 2
eP =
(3.2)
p + M22 , p, 0, 0 ,
M2

1
p, p 2 + M22 , 0, 0 ,
eL =
(3.3)
M2
eT = (0, 0, 1, 0)
(3.4)
where the fourth component refers to the compactified direction. One can calculate the following
formulas in the high energy limit

1 2
eP k1L = eP k1R =
p + M12 p 2 + M22 + p 2
M2

M22
M12
p2
=
(3.5)
2 + 2 + 2 + O p 2 ,
M2
2p
2p

1 2
eP k3L = eP k3R =
q + M32 p 2 + M22 pq cos
M2

M32
M22
pq
=
(3.6)
1 cos + 2 + 2 + O p 2 ,
M2
2p
2q
28

p 2
p + M12 + p 2 + M22
M2

M22
M12
p2
=
2 + 2 + 2 + O p 2 ,
M2
2p
2p

1
p q 2 + M32 q p 2 + M22 cos
eL k3L = eL k3R =
M2

M32
M2
pq
=
1 + 2 1 + 22 cos + O p2 ,
M2
2q
2p
eL k1L = eL k1R =
eT k1L = eT k1R = 0,
e k3L = e k3R = q sin ,
T
(3.7)
(3.8)
(3.9)
(3.10)
which will be useful in the calculations of high energy string scattering amplitudes.
2 =
For the noncompactified open string, it was shown that [7,8], at each fixed mass level Mop
2(N 1), a four-point function is at the leading order in high energy (GR) only for states of the
following form
T N2l2q L 2l L q
|N, 2l, q 1
(3.11)
1
2 |0, k ,
where N 2l + 2q, l, q 0. To avoid the complicated subleading order calculation due to the
L operator, we will choose the simple case l = 0. We made a similar choice when dealing
1
with the high energy string/D-brane scatterings [16,17,23]. There is still one complication in the
25 and
25 in the vertex operator.
case of compactified string due to the possible choices of n
m
However, it can be easily shown that for each fixed mass level with given quantized and winding
1
25
25 are
momenta ( m
m
R , 2 nR), and thus fixed NR + NL level, vertex operators containing n or
T (
T )
subleading order in energy in the high energy expansion compared to other choices 1
1
L
L
and 2 ( 2 ) on the noncompact directions. In conclusion, in the calculation of compactified
closed string in the GR, we are going to consider tensor state of the form
T NL 2qL L qL T NR 2qR L qR
|NL,R , qL,R 1
(3.12)
2
2 |0 ,
1
at general NR + NL level scattered from three other tachyon states with NR + NL = 0.
Note that, in the GR, one can identify eP with eL as usual [4,5]. However, in the MR, one
cannot identify eP with eL . This can be seen from Eq. (3.5) to Eq. (3.8). In the MR, instead of
using the tensor vertex in Eq. (3.12), we will use
T NL 2qL P qL T NR 2qR P qR
|NL,R , qL,R 1
(3.13)
2
2 |0 ,
1
as the second vertex operator in the calculation of high energy scattering amplitudes. Note also
that, in the MR, states in Eq. (3.13) may not be the only states which contribute to the high
energy scattering amplitudes as in the GR. However, we will just choose these states to calculate
the scattering amplitudes in order to compare with the corresponding high energy scattering
amplitudes in the GR.
The high energy scattering amplitudes in the MR can be calculated to be

A = TNL 2qL PqL ,TNR 2qR PqR d 2 z1 d 2 z2 d 2 z3 d 2 z4

NL 2qL PqL ,TNR 2qR PqR
V1 (z1 , z 1 )V2T
(z2 , z 2 )V3 (z3 , z 3 )V4 (z4 , z 4 )

= TNL 2qL PqL ,TNR 2qR PqR
29
d 2 z1 d 2 z2 d 2 z3 d 2 z4 eik1L X (z1 )eik1R X (z1 )

N 2q
q
N 2q
q
X T L L i 2 X P L eik2L X (z2 ) X T R R i 2 X P R eik2R X (z2 )

eik3L X (z3 )eik3R X (z3 )eik4L X (z4 )eik4R X (z4 )

2
2
2
2
kiL kj L
kiR kj R
(zi zj )
(zi z j )
= d z1 d z2 d z3 d z4

i<j
k1L
k3L ieT k4L
+
+
z1 z2
z3 z2
z4 z2
ieT
ieT
NL 2qL

eP k1L
eP k3L
eP k4L qL
+
+
(z1 z2 )2 (z3 z2 )2 (z4 z2 )2
T

ie k1R ieT k3R ieT k4R NR 2qR
+
+
z 1 z 2
z 3 z 2
z 4 z 2
P

P
e k1R
e k3R
eP k4R qR
+
+
.
(z1 z 2 )2 (z3 z 2 )2 (z4 z 2 )2
(3.14)
After the standard SL(2, C) gauge fixing, we get

A (1)k1L k2L +k1R k2R +k1L k3L +k1R k3R +k2L k3L +k2R k3R

d 2 z zk1L k2L z k1R k2R (1 z)k2L k3L (1 z )k2R k3R

ieT k1L ieT k3L NL 2qL ieT k1R ieT k3R NR 2qR
z
1z
z
1 z
P

q
e k1L
eP k3L L eP k1R
eP k3R qR
+
.
z2
(1 z)2
z 2
(1 z )2
(3.15)
By using Eqs. (3.5) to (3.10), the amplitude can be written as

A (1)n+q+q +k1L k2L +k1R k2R +k1L k3L +k1R k3R +k2L k3L +k2R k3R (q sin )NL +NR 2qL 2qR


1 NL 2qL
1 NR 2qR
1z
1 z

1 2
M2 ( p + M12 p 2 + M22 + p 2 )
z2

1
2 + M 2 p 2 + M 2 pq cos ) qL
(
q
3
2
M2
+
(1 z)2

1 2
M2 ( p + M12 p 2 + M22 + p 2 )
z 2
30
1
M2 (

q 2 + M32 p 2 + M22 pq cos ) qR
(1 z )2
= (1)k1L k2L +k1R k2R +k1L k3L +k1R k3R +k2L k3L +k2R k3R

qL +qR
1
NL +NR
(q sin )
M2 q 2 sin2


1 NR 2qR
1 z

q
q
p 2 + M12 p 2 + M22 + p 2 i

q q
j
i
z2
i=0 j =0

p 2 + M 2 p 2 + M 2 + p 2 j
1
2
z 2
1
1z
NL 2qL
= (1)k1L k2L +k1R k2R +k1L k3L +k1R k3R +k2L k3L +k2R k3R (q sin )NL +NR

q 2 + M32 p 2 + M22 pq cos qL +qR

M2 q 2 sin2

i+j
qL
qR
p 2 + M12 p 2 + M22 + p 2

q L qR

i
j
q 2 + M32 p 2 + M22 + pq cos
i=0 j =0
sin[(k1R k2R 2j )] sin[(k2R k3R NR + 2j )]
sin[(1 + k1R k2R + k2R k3R NR )]
(1 + k1R k2R 2j )(1 + k2R k3R NR + 2j )
(2 + k1R k2R + k2R k3R NR )

(1 + k1L k2L 2i)(1 + k2L k3L + 2i NL )
(3.16)
,
(2 + k1L k2L + k2L k3L NL )
where, as in the calculation of Section 2 for the GR, we have used Eq. (2.32) to do the integration.
It is easy to do the following approximations for the gamma functions
A (1)k1L k2L +k1R k2R +k1L k3L +k1R k3R +k2L k3L +k2R k3R (q sin )NL +NR

q 2 + M32 p 2 + M22 pq cos qL +qR

M2 q 2 sin2

i+j
qL
qR
2 + M 2 p2 + M 2 + p2
p

q L qR
1
2

i
j
q 2 + M32 p 2 + M22 + pq cos
i=0 j =0
sin[k1R k2R ] sin[k2R k3R ]
sin[(1 + k1R k2R + k2R k3R )]
(1 + k1R k2R )(1 + k2R k3R )(1 + k1L k2L )(1 + k2L k3L )
(2 + k1R k2R + k2R k3R )(2 + k1L k2L + k2L k3L )
(k1R k2R )2j (k2R k3R )NR +2j (k1L k2L )2i (k2L k3L )NL +2i
.
(k1R k2R + k2R k3R )NR
(k1L k2L + k2L k3L )NL
31
(3.17)
One can now do the double summation and drop out the MiL,R terms to get

qL +qR
q sin (sL + tL ) NL
q sin (sR + tR ) NR
1
A
tL
tR
2M2 q 2 sin2

q
t 2 (sR 2K 1R K 2R ) R
(tR 2K 2R K 3R ) + R
sR2

t 2 (sL 2K 1L K 2L ) qL
(tL 2K 2L K 3L ) + L
sL2

sin(sL /2) sin(tR /2)
tR
uR
tL
uL
B 1 , 1
B 1 , 1
. (3.18)
sin(uL /2)
2
2
2
2
Eq. (3.18) is valid for E 2 NR + NL , M22 NR + NL .
3.2. The infinite linear relations in the GR
For the special case of GR with E 2 M22 , one can identify q with p, and the amplitude in
Eq. (3.18) further reduces to

2p cos3 2 NL +NR
1 qL +qR sin(sL /2) sin(tR /2)
lim A =
2M2
sin(uL /2)
E 2 M22
sin 2

tR
uR
tL
uL
B 1 , 1
(3.19)
B 1 , 1
.
2
2
2
2
It is crucial to note that the high energy limit of the beta function with s + t + u = 2n 8 is [4]

( 2t 1)( u2 1)
t
u
B 1 , 1
=
2
2
( 2s + 2)

3
52n
12n
sin
cos
E
2
2

t ln t + u ln u (t + u) ln(t + u)
exp
(3.20)
,
2
where we have calculated the approximation up to the next leading order in energy E. Note that
the appearance of the prepower factors in front of the exponential fall-off factor. For our purpose
here, with Eq. (2.28), we have

2
MiL,R
= 2NL,R 8,
sL,R + tL,R + uL,R =
(3.21)
i
and the high energy limit of the beta functions in Eq. (3.19) can be further calculated to be

tR
uR
tL
uL
B 1 , 1
B 1 , 1
2
2
2
2
3
52(NL +NR )

E 12(NL +NR ) sin

cos
2
2
32

tL ln tL + uL ln uL (tL + uL ) ln(tL + uL )
exp
2

tR ln tR + uR ln uR (tR + uR ) ln(tR + uR )
exp
2
3

52(NL +NR )
E 12(NL +NR ) sin
cos
2
2

exp
,
4
(3.22)

= 4open
= 2. By combining Eq. (3.19) and
where we have implicitly used the relation closed
Eq. (3.22), we end up with

2 cot 2 NL +NR
1 qL +qR 1
3
5
E
lim A
sin
cos
E
2M2
2
2
E 2 M22

exp
. (3.23)
sin(uL /2)
4
1
We see that there is a ( m
R , 2 nR) dependence in the

sin(uL /2)
factor in our final result.
1
This is physically consistent as one expects a ( m
R , 2 nR) dependent Regge-pole and zero structures in the high energy string scattering amplitudes. In conclusion, in the GR, for each fixed
1
mass level with given quantized and winding momenta ( m
R , 2 nR) (thus fixed NL and NR by
Eq. (2.7)), we have obtained infinite linear relations among high energy scattering amplitudes
of different string states with various (qL , qR ). Note also that this result reproduces the correct
1 qL +qR
)
obtained in the previous works [16,17,23]. However, the mass parameter M2
ratios ( 2M
2
1
m 1
here depends on ( m
R , 2 nR). It is also interesting to see that, if not for the ( R , 2 nR) dependence
in the
factor in the high energy scattering amplitudes in the GR, we would
sin(uL /2)
have had a linear relation among scattering amplitudes of different string states in different mass
levels with fixed (NR + NL ).
Presumably, the infinite linear relations obtained above can be reproduced by using the method
of high energy ZNS, or high energy Ward identities, adopted in the previous works [411]. The
existence of soliton ZNS at arbitrary mass levels was constructed in [22]. A closer look in this
direction seems worthwhile. In the paper, however, we are more interested in understanding the
power-law behavior of the high energy string scattering amplitudes and breakdown of the infinite
linear relations as we will discuss in the next section.
3.3. Power-law and breakdown of the infinite linear relations in the MR

In this section we discuss the power-law behavior of high energy string scattering amplitudes
in a compact space. We will see that, in the MR, the infinite linear relations derived in Section 3.2 break down and, simultaneously, the UV exponential fall-off behavior of high energy
string scattering amplitudes enhances to power-law behavior. The power-law behavior of high
energy string scatterings in a compact space was first suggested by Mende [20]. Here we give a
mathematically more concrete description. It is easy to see that the power law condition, i.e.
Eq. (3.7) in Mendes paper [20],
k1L k2L + k1R k2R = const,
(3.24)
33
turns out to be
(k1L k2L + k1R k2R )

= p 2 + M12 p 2 + M22 + p 2 + (K 1L K 2L + K 1R K 2R )

1 L
2)
= p 2 + M12 p 2 + M22 + p 2 + 2(K 1 K 2 + L
= const.
(3.25)
As p , due to the existence of winding modes in the compactified closed string, it is possible
1, L
2 ) such that
to choose (K 1 , K 2 ; L
1 L
2 < 0,
K 1 K 2 + L
(3.26)
1 L
2 ) to make
and let (K 1 K 2 + L
k1L k2L + k1R k2R const
(3.27)
sL + sR const.
(3.28)
In our calculation, this condition implies the beta functions in Eq. (3.18) reduce to

tR
uR
tL
uL
B 1 , 1
B 1 , 1
2
2
2
2
=
( t2R 1)( u2R 1) ( t2L 1)( u2L 1)

( s2R + 2)
( s2L + 2)
sin(sR /2)( t2R 1)( u2R 1)( t2L 1)( u2L 1)

,
s2R (1 + s2R )(1 + s2R )
(3.29)
which behaves as power-law in the high energy limit! On the other hand, it is obvious that the
(qL , qR ) dependent power factors of the amplitude in Eq. (3.18)

qL +qR
1
AqL ,qR
2M2 q 2 sin2

t 2 (sR 2K 1R K 2R ) qR
(tR 2K 2R K 3R ) + R
sR2

t 2 (sL 2K 1L K 2L ) qL
(tL 2K 2L K 3L ) + L
(3.30)
sL2
show no linear relations in the MR. This is very different from the case of high energy scattering
amplitude in Eq. (3.23) in the GR, which shows nice linear relations

1 qL +qR
AqL ,qR
(3.31)
.
2M2
Note that the mechanism to break the linear relations and the mechanism to enhance the amplitude to power-law are all due to E M2 in the MR. In our notation, Eq. (3.24) is equivalent to
the following condition

p 2 + M12 p 2 + M22 + p 2
E2
lim
(3.32)
m m
O(1).
1
1 2
2
p
1 L
2
K 1 K 2 + L
2 + n1 n2 R
R
34
For our purpose here, as we will see soon, it is good enough to choose only one compactified
coordinate to realize Eq. (3.32). First of all, in addition to Eqs. (2.17) and (2.18), Eq. (2.7) implies
mi ni = 0,
i = 1, 2, 3, 4 (no sum on i).
(3.33)
This is because three of the four vertex are tachyons. Also, since we are going to take n2 to
infinity with fixed NR + NL in order to satisfy Eq. (3.32), we are forced to take m2 = 0. In sum,
we can take, say, mi = 0 for i = 1, 2, 3, 4, and n1 = n2 = n, n3 = 2n, n4 = 0, and then
let n to realize Eq. (3.32). Note that it is crucial to choose different sign for n1 and n2 in
order to achieve the minus sign in Eq. (3.32). We stress that there are other choices to realize the
condition. One notes that all choices implies
NR = NL .
(3.34)
It is obvious that one can also compactify more than one coordinate to realize the Mende condition. We conclude that the high energy scatterings of the highly winding string states of the
compactified closed string in the MR behave as the unusual UV power-law, and the usual linear
relations among scattering amplitudes break down due to the unusual power-law behavior.
4. Conclusion
In this paper we calculate high energy scattering amplitudes of closed bosonic string compactified on torus. We define two regimes of high energy limit, the Gross regime (GR) and the
Mende regime (MR). In the GR, for each fixed mass level with given quantized and winding
1
momenta ( m
R , 2 nR), we obtain infinite linear relations among high energy scattering amplitudes
of different string states. In the MR, we discover that linear relations with NR = NL break down
and, simultaneously, the amplitudes enhance to power-law behavior instead of the usual exponential fall-off behavior at high energies. The result of this work gives a concrete example to
justify the coexistence of the linear relations and the softer exponential fall-off behavior of high
energy string scattering amplitudes as we pointed out previously [17,23]. It is also reminiscent of
our previous work on the power-law behavior of high energy string/domain-wall scatterings [17].
However, in the case of string/domain-wall scatterings, the high energy scattering amplitudes behaves as power-law for the whole UV kinematic regime, and one cannot see the transition from
UV power-law behavior to the UV exponential fall-off behavior, neither can one see the transition from nonlinear relations to the linear relations among high energy scattering amplitudes of
different string states. On the contrary, for the high energy scatterings of the compactified closed
1
string considered in this paper, one gets more kinematic variables, namely ( m
R , 2 nR), to cover
both GR and MR, so that one can see this interesting transition. Through the observation of this
transition, one further confirms that the infinite linear relations obtained in [411] are responsible
for the UV softer string scattering amplitudes than the field theory scatterings.
Acknowledgements
We would like to thank the hospitality of University of Tokyo at Komaba, where this work
is finalized. We are indebted to Chuan-Tsung Chan, Koji Hashimoto, Yoichi Kazama, Akitsugs
Miwa, Yu Nakayama, Naoto Yokoi and Tamiaki Yoneya for many of their enlightening discussions. This work is supported in part by the National Science Council, 50 billions project of
Ministry of Education and National Center for Theoretical Science, Taiwan, ROC.
35
References
[1] D.J. Gross, P.F. Mende, Phys. Lett. B 197 (1987) 129;
D.J. Gross, P.F. Mende, Nucl. Phys. B 303 (1988) 407.
[2] D.J. Gross, Phys. Rev. Lett. 60 (1988) 1229;
D.J. Gross, Philos. Trans. R. Soc. London A 329 (1989) 401.
[3] D.J. Gross, J.L. Manes, Nucl. Phys. B 326 (1989) 73, see Section 6 for details.
[4] C.T. Chan, J.C. Lee, Phys. Lett. B 611 (2005) 193;
J.C. Lee, hep-th/0303012.
[5] C.T. Chan, J.C. Lee, Nucl. Phys. B 690 (2004) 3.
[6] C.T. Chan, P.M. Ho, J.C. Lee, Nucl. Phys. B 708 (2005) 99.
[7] C.T. Chan, P.M. Ho, J.C. Lee, S. Teraguchi, Y. Yang, Nucl. Phys. B 725 (2005) 352.
[8] C.T. Chan, P.M. Ho, J.C. Lee, S. Teraguchi, Y. Yang, Phys. Rev. Lett. 96 (2006) 171601.
[9] C.T. Chan, P.M. Ho, J.C. Lee, S. Teraguchi, Y. Yang, Nucl. Phys. B 749 (2006) 266.
[10] C.T. Chan, J.C. Lee, Y. Yang, Nucl. Phys. B 738 (2006) 93.
[11] C.T. Chan, J.C. Lee, Y. Yang, Nucl. Phys. B 749 (2006) 280.
[12] P.-M. Ho, X.-Y. Lin, Phys. Rev. D 73 (2006) 126007.
[13] J.C. Lee, Phys. Lett. B 241 (1990) 336;
J.C. Lee, Phys. Rev. Lett. 64 (1990) 1636;
J.C. Lee, B. Ovrut, Nucl. Phys. B 336 (1990) 222;
J.C. Lee, Phys. Lett. B 326 (1994) 79.
[14] T.D. Chung, J.C. Lee, Phys. Lett. B 350 (1995) 22;
T.D. Chung, J.C. Lee, Z. Phys. C 75 (1997) 555;
J.C. Lee, Eur. Phys. J. C 1 (1998) 739.
[15] H.C. Kao, J.C. Lee, Phys. Rev. D 67 (2003) 086003;
C.T. Chan, J.C. Lee, Y. Yang, Phys. Rev. D 71 (2005) 086005.
[16] C.T. Chan, J.C. Lee, Y. Yang, Scatterings of massive string states from D-brane and their linear relations at high
energies, Nucl. Phys. B 764 (2007) 1.
[17] C.T. Chan, J.C. Lee, Y. Yang, Power-law behavior of strings scattered from domain-wall and breakdown of their
high energy linear relations, hep-th/0610219.
[18] For a review, see: A. Hashimoto, I.R. Klebanov, Scattering of strings from D-branes, hep-th/9611214.
[19] M.R. Garousi, R.C. Myers, Superstring scattering from D-branes, Nucl. Phys. B 475 (1996) 193, hep-th/9603194.
[20] P.F. Mende, High energy string collisions in a compact space, Phys. Lett. B 326 (1994) 216, hep-th/9401126.
[21] R.R. Khuri, Veneziano amplitude for winding strings, Phys. Rev. D 48 (1993) 2823, hep-th/9303074.
[22] J.-C. Lee, Soliton gauge states and T-duality of closed bosonic string compactified on torus, Eur. Phys. C 7 (1999)
669.
[23] J.C. Lee, Y. Yang, Linear relations of high energy absorption/emission amplitudes of D-brane, Phys. Lett. B 646
(2007) 120, hep-th/0612059.
[24] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, vol. 1, Cambridge Univ. Press, 1987.
[25] D. Gross, J. Harvey, E. Martinec, R. Rohm, Heterotic string theory II, Nucl. Phys. B 267 (1986) 75.
Single production of the doubly charged scalar

in the littlest Higgs model
Chong-Xing Yue , Shuang Zhao, Wei Ma
Department of Physics, Liaoning Normal University, Dalian 116029, China
Received 3 April 2007; received in revised form 27 May 2007; accepted 8 June 2007
Abstract
Single production of the doubly charged scalars via e , ep and pp collisions is studied in the
context of the little Higgs (LH) model. Our numerical results show that the new particles can be
abundantly produced and their possible signatures should be detected in future high energy linear e+ e
collider (ILC). The cross section for single production of at the LHC is much smaller than that at the
ILC or the THERA.
1. Introduction
It is widely believed that the mechanism of electroweak symmetry breaking (EWSB) and the
origin of the particle mass remain prominent mystery in current particle physics in spite of the
success of the Standard Model (SM) tested by high energy experimental data. There has been
no experimental evidence of the SM Higgs boson existing. Furthermore, the neutrino oscillation
experiments have made one believe that neutrinos are massive, oscillate in flavor, which presently
provides the only experimental hints of new physics (NP) [1]. Thus, the SM can only be an
effective theory below some high energy scales. Other EWSB mechanisms and extended Higgs
sectors have not been excluded in the theoretical point of view.
Doubly charged scalars arise in many new physics models with extended Higgs sectors, such
as leftright symmetric models [2], Higgs triplet models [3], 331 models [4], and little Higgs
models [5]. In these new physics models, doubly charged scalars appear typically as components
E-mail address: cxyue@lnnu.edu.cn (C.-X. Yue).

doi:10.1016/j.nuclphysb.2007.06.003
C.-X. Yue et al. / Nuclear Physics B 784 (2007) 3648
37
of the SU(2) triplet representations, which do not couple to quarks and their couplings to leptons
break the lepton number by two units. As a result, these new scalar particles have a distinct
experimental signature, namely a same sign pair of leptons. Thus, discovery of a doubly charged
scalar particle in future high energy colliders would be a definite signal of NP beyond the SM,
which would help us to understand the Higgs sector and more importantly what lies beyond
the SM.
Since charge conservation prevents doubly charged scalars from decaying to a pair of quarks
and their Yukawa couplings violate lepton number conservation, this kind of new particles would
have a distinct experimental signature, which has lower backgrounds. In addition, the presence
of doubly charged scalars provides a simple explanation to the lightness of neutrinos via the
see-saw mechanism [6], contents with recent data on neutrino oscillations [1]. This fact has lead
to many studies involving production and decay of doubly charged scalars at present or future
high energy collider experiments within specific popular models beyond the SM. For example,
studies of production for doubly charged scalars have been given in the context of some specific
popular models at hadron colliders [7,8] and different modes of operation of lepton colliders such
as e+ e mode [9], e e mode [10], and e or mode [11].
Although there are lot of works on production and decays of doubly charged scalars in the literature, it is need to be further studied in the context of the little Higgs models. There are several
motivations to perform this study. First, little Higgs theory can be seen as one of the important
candidates of the NP beyond the SM, most of the little Higgs models predict the existence of the
doubly charged scalars. However, in previous works on studying the phenomenology of the little
Higgs models, studies about doubly charged scalars are very little. Second, three types of colliders related to particle physics research seem to be promising in the next decade. Namely, they are
the CERN Large Hadron Collider (LHC), the high energy linear e+ e collider (ILC), and the
linear-ring type ep collider (LC LHC) called THERA. The phenomenology of the new physics
models should be analyzed taking into account all three types of colliders. So far, a complete
study on single production of the doubly charged scalars has not been presented in the context
of the little Higgs models. Third, studying the possible signals of the doubly charged scalars in
future high energy colliders can help the collider experiments to test little Higgs models, distinguish different NP models, and further to probe the production mechanism of the neutrino
mass. Thus, in this paper, we will consider single production of the doubly charged scalars
predicted by the littlest Higgs (LH) model [12] and see whether their signals can be detected in
future three types of high energy colliders (ILC, THERA, and LHC).
There are several variations of the little Higgs models, which differ in the assumed higher
symmetry and in the representations of the scalar multiples. In matter content, the LH model is
the most economical little Higgs model discussed in the literature, which has almost all of the
essential feature of the little Higgs models. The severe constraints on its free parameters coming
from the electroweak precision data can be solved by introducing the T-parity [13]. Using of
the fact that the LH model contains a complex triplet scalar , Refs. [14,15] have discussed
the possibility to introduce lepton number interactions and generation of neutrino mass in the
framework of the LH model. It has been shown [16] that the neutrino masses can be given by
the term Yij , in which is the vacuum expectation value (VEV) of the complex triplet scalar
and Yij is its Yukawa coupling constant. As long as the VEV is restricted to be extremely
small, the value of Yij is of naturel order one. So, the doubly charged scalars predicted
by the LH model might produce large contributions to some of lepton flavor violating (LFV)
processes [17,18].
38
Using the current experimental upper limits on the branching ratios Br(li lj ) and
Br(li lj lk lk ), Ref. [18] obtains constraints on the relevant free parameters of the LH model.
Taking into account these constraints, we further consider the contributions of to the LFV
processes e e li lj and e+ e li lj (li or lj = e) in Ref. [18]. Based on Ref. [18], in
this paper, we will discuss all possible processes for single production of the doubly charged
scalars in the future high energy collider experiments. The rest of this paper is organized as
follows. The relevant couplings of the doubly charged scalars are summarized in Section 2.
Sections 3, 4, and 5 are devoted to the computation of the cross sections for single production of
the doubly charged scalars at the ILC, THERA, and LHC, respectively. Some phenomenological analysis are also included in the three sections. Our conclusions and simple discussions
are given in Section 6.
2. The relevant couplings of the doubly charged scalars
The LH model [3] consists of a non-linear model with a global SU(5) symmetry and a
locally gauged symmetry [SU(2) U (1)]2 . The global SU(5) symmetry is broken down to its
subgroup SO(5) at a scale f TeV, which results in 14 Goldstone bosons (GBs). Four of
these GBs are eaten by the new gauge bosons (WH , ZH , BH ), resulting from the breaking of
[SU(2) U (1)]2 , giving them masses. The Higgs boson remains as a light pseudo-Goldstone
boson, and other GBs give masses to the SM gauge bosons and form the complex triplet scalar .
Thus, the remaining ten GBs can be parameterized as:
+
0
+
+
++

2
H

2
H
with =
0
=
(1)
.
2
2
+
0 +i p
T
H
Where H is the SM Higgs doublet, ++ , + , 0 , and p are the components of the complex
triplet scalar , which are degenerate at lowest order with a common mass M . The couplings
of the doubly charged scalar to other particles, which are related to our calculation, can be
written as [15,19]:
e2
g ,
2
SW
W+ W+ : 2i
WH+ WH+ : 2i
(2)
e2 (c4 + s 4 )
g ,
2
2s 2 c2
SW
(3)
e2 (c2 s 2 )
g ,
2
2sc
SW
(4)
W+ WH+ : 2i
++ : 2ie(p1 p2 ) ,
(5)
li+ lj+
(6)
: 2iYij CPR ,
e
(p1 p2 ) ,
SW
(7)
e (c2 s 2 )
(p1 p2 ) .
SW
2sc
(8)
W+ + : i
+ WH+ : i
39
(a)
(b)
(c)
Fig. 1. Feynman diagrams for the process e l + .
Where SW = sin W , W is the Weinberg angle. The free parameter c (s = 1 c2 ) is the mixing parameter between SU(2)1 and SU(2)2 gauge bosons. PR = (1 + 5 )/2 is the right-handed
projection operator and C is the charge conjugation operator. The indices i and j represent
three generation leptons e, , or . Eq. (6) gives the flavor-diagonal (FD) couplings and the
flavor-mixing (FX) couplings of the doubly charged scalar to leptons for i = j and i = j ,
respectively.
In the LH model, the neutrino mass matrix can be written as Mij = Yij [15,16]. Considering
the current bounds on the neutrino mass [20], we should have Yij 1010 GeV. If we assume
that the value of the triplet scalar VEV is smaller than 1 105 GeV, then there is the Yukawa
coupling constant Yij > 1 105 , which does not conflict with the most severe constraints on the
LH model from the LFV process eee [18]. In this case, the decay channel W W
can be neglected. As a result, the total decay width of the doubly charged scalar can be
approximately written as [15,18]:

3M Yii2
.
8
(9)
In the following sections, we will assume Yij > 1 105 GeV, which might produce large
contributions to some LFV processes, and take M and Yij as free parameters to calculate the
single production cross sections of the doubly charged scalar in future three types of high
energy collider experiments.
3. Single production of at the ILC
It is widely believed that the hadron colliders, such as Tevatron and LHC, can directly probe
possible NP beyond the SM up to a few TeV, while the TeV energy linear e+ e collider (ILC)
is also need to complement the probe of the new particles with detailed measurement [21]. An
unique feature of the ILC is that it can be transformed to or e collision with the photon
beam generated by laser-scatting method. The effective luminosity and energy of the and e
collisions are expected to be comparable to those of the ILC. In some scenarios, they are the best
instrument for discovery of the NP signatures. The e collision can produce particles, which
are kinematically not accessible in the e+ e collision at the same collider. For example, for the
process e AB with light particle A and new particle B, the discovery limits can be much
higher than in other reactions.
From discussions given in Section 2, we can see that the doubly charged scalar can be
produced via e collision associated with a lepton. The relevant Feynman diagrams are shown
in Fig. 1, in which l + is the lepton e+ , + , or + .
For the process e (P1 ) + (k1 ) l + (P2 ) + (k2 ), the renormalization amplitude can be
written as:
40
eYel
v T (P2 )C 1 PR (P
/ 1 + k/1 ) u(P1 ) (k1 )
(P1 + k1 )2
2eYel
v T (P2 )C 1 PR u(P1 )(2k2 k1 ) (k1 )

2
(k1 k2 )2 M
eYel
v T (P2 ) (P
/ 1 k/2 )C 1 PR u(P1 ) (k1 ).
(P1 k2 )2
M =
(10)
+
After calculating the cross section (s ) for the
subprocess e l , the effective cross
section 1 (s) at the ILC with the center-of-mass s = 2 TeV can be obtained by folding the cross
section 1 (s ) with the photon distribution function f /e [22]:
0.83
1 (s) =
dx 1 (s )f /e (x),
2 /s
M
(11)
where x = s /s, in which s is the center-of-mass energy of the subprocess e l + .

The effective cross section 1 (s) for the processe e+ is plotted in Fig. 2 as a
function of the FD coupling constant Y = Yee for s = 2 TeV and three values of the
can be abundantly
mass M . From this figure, one can see that the doubly
charged scalar
produced via e collision at the ILC experiment with s = 2 TeV. For example, for 0.1 Y
0.9 and 800 GeV M 1600 GeV, the value
of 1 (s) is in the range of 42.83474 fb. If we
assume that the future ILC experiment with s = 2 TeV has a yearly integrated luminosity of
= 500 fb1 , then there will be ten thousands up to several millions of the doubly charged scalar
associated with a positive electron e+ to be generated per year.
In general, the doubly charged scalar can decay to the modes , W , W W
and li lj . However, degeneracy among the components of the triplet scalar forbids appearance
of the first two modes. Furthermore, the decay width ( W W ) is controlled by the
Fig. 2. The cross section 1 (s) as a function of the FD coupling constant Y for three values of the mass M .
41
triplet scalar VEV . It has been shown that, for < 1 105 GeV and M < 2000 GeV,
mainly decays to li lj [10]. In this case, the signatures of is the same-sign lepton pair,
including lepton-number violating final states, which is the SM background free and has been
investigated in Ref. [11]. Thus, as long as the doubly charged scalar is not too heavy, its
possible signals should be detected in future ILC experiments.
The doubly charged scalar can also be singly produced via the LFV processes e
+
l (l = e) induced by the lepton number violating interactions. However, the experimental
upper limits on the LFV processes eee and eee can give severe constraints on the
4 (i = j ) [18], which make that the production cross sections of the
combination |Yii Yij |2 /M
LFV processes e l + (l = e) are very small. For example, for Y = 0.1 and M =
1000 GeV, the effective cross section for the process e + is smaller than 6.9
104 fb. Thus, it is very difficult to detect the signals of via the LFV processes e
l + (l = e) in future ILC experiments.
4. Single production of the doubly charged scalar at the THERA
Although the linear-ring-type ep collider (LC LHC) with the center-of-mass energy s =
3.7 TeV and with the integral luminosity = 100 pb1 has a lower luminosity, it can provide
better condition for studying a lot of phenomena, compared to the ILC due to the high center-ofmass energy and compared to the LHC due to clearer environment [23]. Thus, it can be used to
detect the possible signals of some new particles. In this section, we consider single production
of the doubly charged scalar predicted by the LH model via ep collision at the THERA.
The doubly charged scalar can be singly produced at the THERA via the processes:
e p e+ X ,
e p + X ,
e p + X ,
(12)
which were first studied by Ref. [24]. In terms of observability, the second and third processes
are induced by the lepton-number violating interactions. The corresponding subprocesses can be
unitive written as e l + (l = e, or ). The relevant Feynman diagrams are plotted in
Fig. 3. Since the production cross section for the process e + or e +
is much smaller than that of the process e e+ , we will use the equivalent photon
approximation (EPA) approach [25,26] to only consider the process e p e+ X in this
section.
Using the EPA method, the effective cross section 2 (s) at the THERA with s = 3.7 TeV
1
and = 100 pb can be folding the cross section 2 (s ) for the subprocess e e+ with
(a)
(b)
Fig. 3. Feynman diagrams for single production of at the THERA.
(c)
42
Fig. 4. The cross section 2 (s) as a function of the FD coupling constant Y for three values of the mass M .
the photon distribution function f /p (x, s ):

2
(1m/
s)
2 (s) =
dx (xs)f /p (x, xs),
(13)
2 /s
M
with x = s /s, m is the proton mass and

f /p (x, xs) = fel/p (x) + finel
/p (x, xs).
(14)
Where fel/p (x) and finel

/p (x, xs) are the elastic and inelastic components of the equivalent photon
distribution of the proton, which has been extensively studied in Refs. [2527].
+
is plotted in Fig. 4 as a funcThe effective cross section 2 (s) of the process
e p e X
tion of the FD Yukawa coupling constant Y for s = 3.7 TeV and three values of the mass M .
One can see from Fig. 4 that the cross section for single production of the doubly charged scalar
at the THERA is generally smaller than that at the ILC in most of the parameter space of
the LH model. For 0.1 Y 0.9 and 800 GeV M 1600 GeV, the value of the production
cross section 2 (s) is in the range of 1.11 102 5.9 102 fb. There will be several tens of
the doubly charged scalar to be generated per year at the THERA with s = 3.7 TeV and
= 100 pb1 .
Similar to single production of at the ILC, the process e p e+ X gives rise
to number of signal events with same-sign lepton pair and an isolated positive electron in the
l l + e+ signature, which is almost free of the SM backgrounds. Thus, possible signatures of
the doubly charged scalar might be detected in future THERA experiments.
43
5. Single production of the doubly charged scalar at the LHC
In this year, the LHC with s = 14 TeV and = 100 fb1 will begin operation, which has
an increase of a factor of seven in energy and a factor of 100 in luminosity over the Fermilab
Tevatron. The LHC is expected to directly probe possible NP beyond the SM up to a few TeV,
which might provide some striking evidence of NP.
From Eqs. (2)(8) given in Section 2, we can see that the doubly charged scalar can be
singly produced at the LHC via the partonic processes:
qq
W , WH W + ;
(15)
qq W
(16)

q q
qq
WH , W
qq
W , WH

qq;
WH+ ;
+
(17)
(q, q = u, d, c, or s).
(18)
However, the production cross sections of the partonic processes (15), (16), and (17) are
strongly suppressed by the factor 2 ( 1 105 GeV). All of their values are smaller than
W , WH
1 108 fb. So, in this section, we only consider the partonic processes qq
+
, as shown in Fig. 5.
Using Eqs. (7) and (8), and other relevant Feynman rules, we can write the scattering amplitude for the partonic process q(P
1 ) + q (P2 ) + (P3 ) + (P4 ):
g
e2 SM
1 )PL u(P2 )
M = 2 Vqq
(P3 P4 )
v(P
2
2
(P
+
P
)
M
2SW
1
2
W
2
2
2

g
ie (c s ) SM
Vqq
1 )PL u(P2 )
2
(P3 P4 ) .
v(P
2
(P1 + P2 )2 MW
2 2SW s 2
H
(19)
associated with
The cross section 3 (s) for single production of
the doubly charged scalar
+
a singly charged scalar at the LHC with s = 14 TeV can be obtained by convoluting the
production cross section 3 (s ) of the partonic process qq
+ with the quark distribution
functions (PDFs):
3 (s) =
ij
dx1
dx2 fi/p (x1 , )fj/p (x2 , ) (ij + ).
(20)
/x1
2 /s and s = x x s (we
Where i and j stand for the partons (light quarks u, d, c, or s), = 4M
1 2
have assumed M = M = M + ). In our numerical calculation, we will use CTEQ6L PDFs
[28] for the quark distribution functions and assume that the factorization scale is of order M /2.
From Eqs. (19) and (20), we can see that the production cross section 3 (s) is dependent on
the free parameters M , MWH , and c. It is well known that global fits to the electroweak precision
(a)
(b)
Fig. 5. Feynman diagrams for the partonic process qq

+ .
44
Fig. 6. The cross section 3 (s) as a function of the scalar mass M for three values of the mass MWH .
data impose rather severe constraints on the free parameters of the LH model. However, if the
SM fermions are charged under U (1)1 U (1)2 , the constraints can become relaxed [29]. As
numerical estimation, we will assume that the free parameters M , MWH , and c are in the ranges
of 8001600 GeV, 12 TeV, and 0.10.5, respectively.
Our numerical results are summarized in Fig. 6, in which we plot the production cross
section 3 (s) as a function of the mass M for different values of the free parameter MWH .
Since the contributions of the LH model to the process pp + X mainly come from
W exchange, the production cross section 3 (s) is insensitive to the free parameter c. So, we
have taken c = 0.3 in Fig. 6. One can see from Fig. 6 that the cross section for single production of the doubly charged scalar at the LHC is much smaller than that at the ILC
or the THERA. This is because, at tree level, the doubly charged scalar is singly produced at the LHC via the s-channel processes with highly virtual gauge boson propagators. For
800 GeV M 1600 GeV, 1 TeV MWH 2 TeV, and c = 0.3, the value of 3 (s) is in the
tens + events to be generrange of 7.4 101 1.3 104
fb. Then, there will be several
1
ated one year at the LHC with s = 14 TeV and = 100 fb .
T b,
W + Z, and W + H
The possible decay modes of the singly charged scalar + are l + l , t b,
+
+
+
+
[15]. The decay widths ( W Z) and ( W H ) are proportional to the factor 2 ,
which can be neglected. If we assume that the scalar + mass M is smaller than the vector-like
top quark mass MT (M MT ), then the decay channel + T b is kinematically forbidden.
If the singly charged
In this case, the singly charged scalar + mainly decays into l + l and t b.
+
+
scalar decays into l l , then the production of the doubly charged scalar associated
with a singly charged scalar + at the LHC gives rise to a number of signal events with like/ + jet, which has a relatively
sign di-leptons, with one jet and large missing energy, l l + E
high detection efficiency and enjoys essentially negligible background from SM processes. The
/ + jet is shown in Fig. 7 as a function of the FD
production rate of the signal event l l + E
Yukawa coupling constant Y for f = 1 TeV and M = 1 TeV. One can see from Fig. 7 that, for
45
Fig. 7. Production rate of the signal event l l + E

/ + jet as a function of the FD coupling constant Y for three values of
the WH mass MWH .
0.1 Y 0.9 and 1 TeV MWH 2 TeV, the production rate of the signal event is in the range
of 6.7 102 fb 1.4 101 fb. With the high luminosity option of the LHC, around 300 fb1 ,
there will be several tens signal events to be generated one year.
then the partonic process
If we assume that the singly charged scalar + decays into t b,

+
will generate the l l + E

/ + jets signature, which is similar to that coming
q q
from production of the first or second generation T-odd quark pair in the context of the LH
model with T-parity [30]. However, the production rate of this kind of signal events is too small
to be detected in future LHC experiments.
6. Conclusions and discussions
Doubly charged scalars appear in some popular NP models beyond the SM, motivated by their
usefulness in generating neutrino masses. This kind of new particles have distinct experimental
signals through their decay to same-sign lepton pairs. Their observation in future high energy
collider experiments would be a clear evidence of NP beyond the SM. Thus, searching for doubly
charged scalars is one of the main goals of future high energy collider experiments.
Little Higgs models have generated much interest as possible alternative to the EWSB mechanism, which can be regarded as one of the important candidates of NP beyond the SM. As the
most economical little Higgs model, the LH model can explain the observed neutrino mass by
introducing the lepton-number violation interaction of the triplet scalar to leptons. The neutrino
mass is proportional to the triplet scalar VEV multiplied by the Yukawa coupling constant
Yij without invoking a right-handed neutrino. This scenario predicts the existence of the doubly
charged scalars . In this paper, we consider single production of this kind of new particles
in future three types of high energy collider experiments (ILC, THERA, and LHC).
In future ILC experiments, the doubly charged scalar can be singly produced via e
collision. Our numerical results show that it can be abundantly produced in future ILC exper-
46
iments. In most of the parameter space of the LH model, there will be a large number of the
signal events with same-sign
lepton pair and an isolated positive electron to be generated in the
ILC experiment with s = 2 TeV and = 100 fb1 . Thus, the possible signals of the doubly
charged scalar should be detected in future ILC experiments.
Using the EPA method, we calculate the
cross section for single production of the doubly
charged scalar at the THERA with s = 3.7 TeV and = 100 pb1 via the subprocess
e e+ . In our numerical estimation, we have included both contributions of the elastic
photon and of the inelastic photon components from proton. We find that, for 0.1 Y 0.9
and 800 GeV M 1000 GeV, the value of the production cross section 2 is in the range of
1.11 102 5.9 102 fb. As long as the doubly charged scalar is not too heavy, its possible
signatures might be detected via the process e p e+ X in future THERA experiments.
At the LHC, the doubly charged scalar can be singly produced via several patronic
processes. However, considering that the triplet scalar VEV is very small ( 1 105 GeV),
the production channels involving the gauge bosons W or WH in the final states can be neglected.
Thus, in this paper, we only consider the production of the doubly charged scalar associated
with a single charged scalar + via the patronic processes qq
W , WH + at the

+
LHC. Since the partonic process qq

proceeds by the s-channel W exchange and
WH exchange, which have highly virtual gauge boson propagators, its production cross section is
much smaller than that at the ILC or the THERA. However, the partonic process qq
+
+
+
with l l and
l l can produce distinct experimental signature, which is almost
free of the SM background. If the mass of is smaller than 1 TeV, its possible signals might
be detected at the high luminosity option of the LHC experiments.
At the ILC, the doubly charged scalar can also be produced associated with a photon
via e e collision, i.e., the subprocess e e . The cleanliness of the final state photon
detection in future e e collider can be very helpful in identifying the doubly charged scalar.
However, this production process suffers from a large SM background process e + e +
e + e , which has to be carefully disposed by suitable technique [10].
Certainly, the doubly charged scalar ++ can also be singly produced at the ILC, the THERA,
and the LHC via the charge-conjugation processes of the corresponding processes for the doubly
charged scalar . Similar to above calculation, we can give the values of the production cross
sections for these processes. Thus, the conclusions for the doubly charged scalar also apply
to the doubly charged scalar ++ .
The LH model with T-parity also predicts the existence of the doubly charged scalars .
While, they have T-odd parity and their VEV is equal to zero, which can not generate the neutrino
masses via introducing the lepton-number violating interaction of the T-odd triplet scalar to
leptons. However, they can also be produced associated a T-odd new particle in future high energy
collider experiments, which might be need to be further studied.
Acknowledgements
This work was supported in part by Program for New Century Excellent Talents in University (NCET-04-0290), the National Natural Science Foundation of China under the Grants
Nos. 10475037 and 10675057.
References
[1] M.C. Gonzalez-Garcia, Y. Nir, Rev. Mod. Phys. 75 (2003) 345;
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
47
V. Barger, D. Marfatia, K. Whisnant, Int. J. Mod. Phys. E 12 (2003) 569;

A.Y. Smimov, hep-ph/402264.
J.C. Pati, A. Salam, Phys. Rev. D 10 (1975) 275;
R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 566;
G. Senjanovic, R.N. Mohapatra, Phys. Rev. D 12 (1975) 1502;
T.G. Rizzo, Phys. Rev. D 25 (1982) 1355.
G.B. Gelmini, M. Roncadelli, Phys. Lett. B 99 (1981) 411;
H. Georgi, M. Machacek, Nucl. Phys. B 262 (1985) 463;
J.F. Gunion, R. Vega, J. Wudka, Phys. Rev. D 42 (1990) 1673.
F. Pisano, V. Pleitez, Phys. Rev. D 46 (1992) 410;
P.H. Frampton, Phys. Rev. Lett. 69 (1992) 2889;
R. Foot, O.F. Hernandez, F. Pisano, V. Pleitez, Phys. Rev. D 47 (1993) 4158.
For recent review see: M. Schmaltz, D. Tucker-Smith, Annu. Rev. Nucl. Parti. Sci. 55 (2005) 229;
T. Han, H.E. Logan, L.T. Wang, JHEP 0601 (2006) 099.
H. Fritzsch, M. Gell-Mann, P. Minkowski, Phys. Lett. B 59 (1975) 256;
T.P. Cheng, Phys. Rev. D 14 (1976) 1367;
M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, D. Freedman (Eds.), Supergravity, Proceedings
of the Workshop, Stony Brook, New York, 1979, North-Holland, Amsterdam, 1979, p. 315;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on Unified Theories and Baryon
Number in the Universe, Tsukuba, Japan, KEK Report No. 79-18, Tsukuba, 1979, p. 95;
S. Weinberg, Phys. Rev. Lett. 43 (1979) 1566;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
K. Huitu, J. Maalampi, A. Pietila, M. Raidal, Nucl. Phys. B 487 (1997) 27;
A. Datta, A. Raychaudhuri, Phys. Rev. D 62 (2000) 055002;
J. Maalampi, N. Romanenko, Phys. Lett. B 532 (2002) 202;
M. Muhlleitner, M. Spira, Phys. Rev. D 68 (2003) 117701.
A.G. Akeroyd, M. Aoki, Phys. Rev. D 72 (2005) 035011;
G. Azuelos, K. Benslama, J. Ferland, J. Phys. G 32 (2006) 73;
J.E. Cieza Montalvo, N.V. Cortez Jr., J. S Borges, M.D. Tonasse, Nucl. Phys. B 756 (2006) 1.
S. Godfrey, P. Kalyniak, N. Romanenko, Phys. Lett. B 545 (2002) 361.
F. Cuypers, M. Raidal, Nucl. Phys. B 501 (1997) 3;
G. Barenboim, K. Huitu, J. Maalampi, M. Raidal, Phys. Lett. B 394 (1997) 132;
M. Raidal, Phys. Rev. D 57 (1998) 2013;
J. Maalampi, N. Romanenko, Phys. Lett. B 474 (2000) 347;
B. Mukhopadhyaya, S.K. Rai, Phys. Lett. B 633 (2006) 519;
O. Cakir, New J. Phys. 8 (2006) 145.
S. Chakrabarti, D. Choudhury, R.M. Godbole, B. Mukhopadhyaya, Phys. Lett. B 434 (1998) 347;
E.M. Gregores, A. Gusso, S.F. Novaes, Phys. Rev. D 64 (2001) 015004;
S. Godfrey, P. Kalyniak, N. Romanenko, Phys. Rev. D 65 (2002) 033009.
N. Arkani-Hamed, A.G. Cohen, E. Katz, A.E. Nelson, JHEP 0207 (2002) 034.
H.C. Cheng, I. Low, JHEP 0309 (2003) 051;
H.C. Cheng, I. Low, JHEP 0408 (2004) 061;
I. Low, JHEP 0410 (2004) 067.
W. Kilian, J. Reuter, Phys. Rev. D 70 (2004) 015004;
J.Y. Lee, JHEP 0506 (2005) 060.
T. Han, H.E. Logan, B. Mukhopadhyaya, R. Srikanth, Phys. Rev. D 72 (2005) 053007.
J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227;
G.B. Gelmini, M. Roncadelli, Phys. Lett. B 99 (1981) 411.
S.R. Choudhury, N. Gaur, A. Goyal, Phys. Rev. D 72 (2005) 097702;
J. Goyal, Mod. Phys. Lett. A 21 (2006) 1931.
C.-X. Yue, S. Zhao, Eur. Phys. J. C 50 (2007) 897.
T. Han, H.E. Logan, B.M. Elrath, L.-T. Wang, Phys. Rev. D 67 (2003) 095004.
G.L. Fogli, et al., Phys. Rev. D 70 (2004) 113003;
M. Tegmark, et al., SDSS Collaboration, Phys. Rev. D 69 (2004) 103501;
O. Elgaroy, O. Lahav, New J. Phys. 7 (2005) 61.
T. Abe, et al., American Linear Collider Group, hep-ex/0106057;
48
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
J.A. Aguilar-Saavedra, et al., ECFA/DESY LC Physics Working Group, hep-ph/0106315;

K. Abe, et al., ACFA Linear Collider Working Group, hep-ph/0109166;
G. Laow, et al., ILC Techinical Review Committee, second report, 2003 SLAC-R-606.
I.F. Ginzbury, et al., Nucl. Instrum. Methods Phys. Res. Sec. A 21 (1984) 5;
V.I. Telnov, Nucl. Instrum. Methods Phys. Res. Sec. A 294 (1990) 72.
S. Sultansoy, Eur. Phys. J. C 33 (2004) s1064, and reference therein.
E. Accomando, S. Petrarca, Phys. Lett. B 323 (1994) 212.
C.F. Von Weizscker, Z. Phys. 88 (1934) 612;
E.J. Williams, Phys. Rev. 45 (1934) 729(L).
M. Glck, C. Pisano, E. Reya, Phys. Lett. B 540 (2002) 75;
C. Pisano, Eur. Phys. J. C 38 (2004) 79, and reference therein.
G. Altarelli, G. Martinelli, B. Mele, R. Rckl, Nucl. Phys. B 262 (1985) 204;
B.A. Kniehl, Phys. Lett. B 254 (1991) 267;
A. Mukherjee, C. Pisano, Eur. Phys. J. C 30 (2003) 477.
J. Pumplin, et al., JHEP 0207 (2002) 012;
D. Stump, et al., JHEP 0310 (2003) 046.
J.L. Hewett, F.J. Petriello, T.G. Rizzo, JHEP 0310 (2003) 062;
C. Csaki, J. Hubisz, G.D. Kribs, P. Meade, J. Terning, Phys. Rev. D 67 (2003) 115002;
C. Csaki, et al., Phys. Rev. D 68 (2003) 035009;
R. Casalbuoni, A. Deandrea, M. Oertel, JHEP 0402 (2004) 032;
M.C. Chen, S. Dawson, Phys. Rev. D 70 (2004) 015003;
C.-X. Yue, W. Wang, Nucl. Phys. B 683 (2004) 48;
W. Kilian, J. Reuter, Phys. Rev. D 70 (2004) 015004;
T. Gregoire, D.R. Smith, J.G. Wacker, Phys. Rev. D 69 (2004) 115.
A. Belyaev, C.-R. Chen, K. Tobe, C.-P. Yuan, Phys. Rev. D 74 (2006) 11502.
The dipole form of the gluon part of the BFKL kernel

V.S. Fadin a,b , R. Fiore c , A.V. Grabovsky a,b , A. Papa c,
a Budker Institute of Nuclear Physics, 630090 Novosibirsk, Russia
b Novosibirsk State University, 630090 Novosibirsk, Russia
c Dipartimento di Fisica, Universit della Calabria, Istituto Nazionale di Fisica Nucleare,
Gruppo collegato di Cosenza, Arcavacata di Rende, I-87036 Cosenza, Italy

Abstract
The dipole form of the gluon part of the color singlet BFKL kernel in the next-to-leading order (NLO)
is obtained in the coordinate representation by direct transfer from the momentum representation, where
the kernel was calculated before. With this paper the transformation of the NLO BFKL kernel to the dipole
form, started a few months ago with the quark part of the kernel, is completed.
1. Introduction
The BFKL equation [1] allows to find the Greens function for the scattering of two reggeized
gluons, which determines the high-energy behavior of QCD amplitudes. It is an integral equation
in the space of momenta transverse to the momenta of colliding particles. The kernel of the BFKL
equation is known now in the next-to-leading order (NLO) both for the forward scattering [2],
i.e. for t = 0 and color singlet (pomeron) in the t-channel, and for any fixed (not growing with
energy) momentum transfer t and any possible color state in the t -channel [35].
The pomeron channel is the most important for phenomenological applications. However,
from the theoretical point of view the color octet case seems to be even more important because
Work supported in part by the RFBR grant 07-02-00953, in part by the RFBR-MSTI grant 06-02-72041 and in part
by Ministero Italiano dellIstruzione, dellUniversit e della Ricerca.
E-mail addresses: fadin@inp.nsk.su (V.S. Fadin), fiore@cs.infn.it (R. Fiore), a.v.grabovsky@inp.nsk.su
(A.V. Grabovsky), papa@cs.infn.it (A. Papa).
doi:10.1016/j.nuclphysb.2007.06.002
50
V.S. Fadin et al. / Nuclear Physics B 784 (2007) 4971
of the gluon reggeization. Indeed, high-energy QCD can be reformulated in terms of a gaugeinvariant effective field theory for reggeized gluon interactions [6], so that the primary reggeon
in QCD is not the pomeron, but the reggeized gluon. The main property of the octet BFKL kernel
which is required by the reggeization is that the gluon trajectory turns out to be its eigenfunction.
This property can be verified in the momentum representation, i.e. the representation in which the
BFKL approach was originally developed. On the contrary, a remarkable property of the color
singlet BFKL kernel in the leading approximation [7] is entirely concerned with the coordinate
representation. It is the famous conformal invariance, which is extremely important for finding
solutions of the equation.
In the NLO the coordinate representation of the color singlet BFKL kernel is also very interesting. First, it reveals conformal properties of the kernel. Evidently, the conformal invariance is
violated by renormalization. One may wonder, however, whether the renormalization is the only
source of the violation. If so, one can expect the conformal invariance of the NLO BFKL kernel
in supersymmetric extensions of QCD.
Another important reason for transferring the color singlet NLO BFKL kernel in the coordinate representation is to make possible the comparison with the color dipole approach for
high-energy processes [8]. This approach gives a clear physical picture of high-energy processes
and can be naturally extended from the regime of low parton densities to the saturation regime [9],
where the evolution equations of parton densities with energy become nonlinear. In general, there
is an infinite hierarchy of coupled equations [10,11]. In the simplest case of a large nucleus as a
target, this set of equations is reduced to the BK (BalitskyKovchegov) equation [10].
It is affirmed [8,10] that in the linear regime the color dipole framework gives the same results
as the BFKL one for the color singlet channel. Before the advent of the dipole approach, the
leading order color singlet BFKL kernel was investigated in the coordinate representation in
detail in Ref. [7]. More recently, the relation between BFKL and color dipole approaches was
analyzed in the leading order in Ref. [12]. The extension of the analysis to the NLO was started a
few months ago. In Ref. [13] the quark contribution to the color dipole approach at large number
of colors Nc has been transferred from the coordinate to the momentum representation and it has
been verified that the resulting contribution to the NLO pomeron intercept agrees with the wellknown result of Ref. [2]. In Refs. [14,15] the non-Abelian (leading in Nc ) and Abelian parts
of the quark contribution to the non-forward BFKL kernel in the momentum representation [3]
have been transformed to the coordinate one and it has been found that the dipole form of the
quark contribution agrees with the result obtained in Ref. [16] by the direct calculation of the
quark contribution to the dipole kernel in the coordinate representation.
Evidently, the main and the most important part of the BFKL kernel is given by the gluon
contribution. The aim of this work is to consider this part of the kernel in the NLO and to find its
dipole form.
The paper is organized as follows: in Section 2 we give basic definitions, fix our notations
and recall the main results of the papers in Refs. [14,15]; in Section 3 we describe the decomposition of the kernel into planar and symmetric parts; in Section 4 we present the NLO
gluon trajectory in a convenient form; in Section 5 we show a convenient representation for the
real part of the color octet kernel entering into the planar part; in Section 6 we discuss the
cancellation of the infrared singularities and give the resulting form of the color octet kernel;
in Section 7 we give the symmetric part of the kernel in a form suitable for the subsequent
transformation; in Section 8 we present the general structure of the dipole form of the kernel; in
Section 9 we describe the procedure to transfer the planar piece of the gluon part of the kernel
from the momentum to the coordinate representation; in Section 10 we outline this procedure
51
for the symmetric piece; in Section 11 we present our final result; in Section 12 we draw our
conclusions. Appendices contain all necessary integrals.
2. Basic definitions and notation
We use the same notation as in Ref. [14]: qi and qi , i = 1, 2, represent the transverse momenta
of reggeons in initial and final t -channel states, while ri and ri are the corresponding conjugate
coordinates. The state normalization is
q q ),

q |
q = (
r |r = (r r ),
(1)
so that
r |
q =
ei q r
,
(2)1+
(2)
where = (D 4)/2; D 2 is the dimension of the transverse space and it is taken different
from 2 for the regularization of divergences. We will also use the notation q = q1 + q2 , q =
q1 + q2 ; k = q1 q1 = q2 q2 and for brevity we will usually write pij = pi pj .
The BFKL kernel in the operator form is written as
K = 1 + 2 + K r ,
(3)
where
2
qi = (
qii )
qi ,

qi | i |
(4)
with (t) the gluon Regge trajectory, and K r represents real particle production in reggeon
collisions. The s-channel discontinuities of scattering amplitudes for the processes A + B
A + B have the form
B
YK
qA qB )discs AA
4i(2)D2 (
AB = A A|e
1
|B B.
q 2 q 2
(5)
1 2
In this equation Y = ln(s/s0 ), s0 is an appropriate energy scale, qA = pA pA , qB = pB pB ,

and
1
q , q = (

q1 , q2 |K|
q q ) 2 2 K(
q1 , q1 ; q),
1 2
q1 q2

q1 , q2 |B B = 4p (
qB q1 q2 )B B (
q1 , q2 ),
(6)
(7)
B
+
qA q1 q2 )A A (

q1 , q2 ),
A A|
(8)
q1 , q2 = 4pA (
where p = (p0 pz )/ 2; the kernel K(

q1 , q1 ; q) and the impact factors are expressed
through the reggeon vertices according to Ref. [17]. Note that the appearance of the factors
(q 21 q 22 )1 in (5) and (
q12 q22 )1 in (6) cannot be explained by a change of the normalization (1).
We have used a freedom in the definition of the kernel. Indeed, one can change the form of the
kernel (in any representation) performing the operator transformation
K O 1 K O,
A A|
O,
A A|
1
1
|B B O 1
|B B,
q2 q 2
q2 q 2
1 2
1 2
(9)
52
which does not change the discontinuity (5). In (9) O is an arbitrary non-singular operator. The
kernel K in (6) is related with the one defined in Ref. [17] by such transformation with O =
(q 21 q 22 )1/2 . The reason for this choice is that in the leading order the kernel which is conformal
invariant and simply related to the dipole kernel is not the kernel defined in Ref. [17], but just
the kernel K in (6) [7,12]. Note that after the choice of the operator O in the leading order,
where O g 2 , are still possible. At the NLO after
additional transformations with O = 1 O,
such transformation we get

K K K (B) , O ,
(10)
where K (B) is the leading order kernel.
In the NLO the BFKL kernel contains both gluon and quark contributions. Refs. [14,15] were
devoted to the quark contribution. In Ref. [14] the non-Abelian part of the quark contribution [3] was transformed to the coordinate representation. The transformation was carried out
in the most general way: at arbitrary D and for the case of arbitrary impact factors. Generally
speaking, the BFKL and dipole kernels are not equivalent. But in case of scattering of colorless
objects, besides the freedom in the definition of the kernel discussed above, there is an additional
freedom related to the gauge invariance (vanishing at zero reggeized gluon momenta) of the
r r can be written in the dipole form
impact factors [7,12]. In this case the kernel r1 r2 |K|
1 2
(see below). Owing to the possibility of the operator transformations (10), the dipole form is not
unique. It was shown in Ref. [14] that after the transformation (10) with a suitable operator O
the non-Abelian part of the quark contribution [3] to the BFKL kernel, transferred to the dipole form, agrees with the result obtained recently in Ref. [16] by direct calculation of the quark
contribution to the BK kernel in the coordinate representation.
In Ref. [15] the Abelian part of the quark contribution [3] was considered and its dipole form
was found. Since the Abelian part is known only in the limit 0, the approach adopted in
Ref. [15] was analogous to the one used in Section 5 of Ref. [14], i.e. the starting point was the
renormalized BFKL kernel, which was then simplified by the cancellation of the infrared singularities between virtual and real contributions and presented at D = 4. After that its dipole form
was found, which turned out to coincide with the corresponding part of the quark contribution to
the BK kernel calculated in Ref. [16] and to be evidently conformal invariant.
Here we consider the gluon part of the kernel (in the following sometimes for brevity we will
call this part simply kernel) and find its dipole form. Evidently this is the main and the most
important part. Since the gluon part is known only in the limit 0, we adopt the same strategy
as in Section 5 of Ref. [14] and in Ref. [15].
3. Decomposition of the gluon contribution
The real part of the gluon contribution to the BFKL kernel in the color singlet channel is
written [5] as
1 (s)
K r = 2K r(8) + K GG
,
2
(11)
where K r(8) is the real part of the color octet kernel, concerned with the gluon reggeization, and
(s)
K GG is the so-called symmetric part of the two-gluon contribution to the non-forward BFKL
kernel. The first of them was calculated in Ref. [4], and the second in Ref. [5]. It is convenient
to consider them separately. The use of the decomposition (11) is due to the fact that all infrared
53
singularities turn out to be located in K r , whereas the symmetric part K GG is infrared finite.
(8)
(s)
Moreover, in the momentum representation K r looks much simpler than K GG . The origin of
(8)
the relative simplicity of K r is that only planar diagrams do contribute to it. The symmetric
part includes contributions from both planar and non-planar diagrams, the latter being the most
complicated.
At fixed non-zero k2 , when the term (k2 /2 ) can be expanded in powers of , the piece
q1 , q2 is finite at = 0. But this part is singular at k2 = 0, therefore the region of

q1 , q2 |K r(8) |
k2 so small that | ln(k2 /2 )| 1 and the expansion of (k2 /2 ) cannot be done, is important.
Moreover, terms of order must be taken into account in the coefficient of the expression divergent at k2 = 0, in order to take all contributions non-vanishing in the limit 0 after the

integration over k.
The total BFKL kernel in the singlet case must be free from singularities. Therefore, the
(8)
infrared singularities of 2K r must cancel the singularities of the virtual contribution 1 +
2 in (3). In the following we join these parts. The experience of the transformation of the
non-Abelian part of the quark contribution [14] teaches us that the kernel in the coordinate
Here
representation can be simplified by the operator transformation (10) with an appropriate O.
we will use the transformed kernel with
(8)
(s)
11s ()Nc 2 2
ln q1 q2 ,
O =
24
(12)
and call
11s ()Nc (B) 2 2
K , ln q1 q2
K p = 1 + 2 + 2K r(8) +
24
(13)
(s)
planar part. We put also K GG
= 2K s , call K s symmetric part of the kernel and write the
kernel as
K = K p + K s .
(14)
The symmetric part

q1 , q2 |K s |
q1 , q2 is finite in the limit = 0. Moreover, it does not give
terms divergent in = 0 by action of the kernel, since it has no non-integrable singularities in the
limit = 0. Hence, we can consider the symmetric part in the physical spacetime dimension
D = 4 from the beginning.
4. The NLO gluon trajectory
We use the representation of the gluon trajectory in the form of integral in the transverse
momentum plane [18]. Taking into account that the bare coupling g in pure gluodynamics is
connected with the renormalized coupling g in the MS scheme through the relation

11 g 2
g = g 1 +
,
3 2
g 2 =
g2 Nc (1 )
(4)2+
where (x) is the Euler gamma-function, we have
(15)
54
2

qi =
g 2 qi2
1+ (1 )

2+2 2
k
d
11
0) + f (0, k qi ) .
k qi ) + f (k,
f (k,
1 + g 2
3
k2 (k qi )2
(16)
Here (x) = (x)/ (x) and

f (k1 , k2 ) =
2

2
k12
k12
d 2+2 q 2
ln
+ aD ,
1+ (1 )
q2
(k1 q)2 (k2 q)2
(17)
where

D
D
aD = 2(D 3) 3
+ 2
2 + (1)
2
2
+
D2
2
+
.
(D 3)(D 4) 4(D 1)(D 3)
(18)
The expressions (17) and (18) are exact in . We need to keep the terms of zero order in in
the trajectory (16), that means the terms of order in f (k1 , k2 ) and of order 2 in aD . With the
required accuracy

1 11
67
202
aD =
+
(2) 2
7 (3) ,

6
18
27
(19)
0) = f (0, k)
is known at
where (n) is the Riemann zeta-function. The integral entering f (k,
arbitrary D (see for instance Eq. (B.16) in the second of Ref. [4]):
2+2 2 2
d
q
k
k2
ln 2
1+
2
2

(1 )
q
q (k q)
2

2
1
() k
(1)
+
(1
)
(1
+
)
+
(1
+
2)
.
=
(2) 2
2
(20)
The integral for f (k1 , k2 ) was calculated with the required accuracy in Ref. [20] (see Eq. (A.13)):
2

2
k
k12
d 2+2 q 2
ln 12
1+
2
2
(1 )
q 2
(k1 q) (k2 q)
2
2
2

2
2
k1
k2
k
k12
k
2 ()
=
2 122
+
ln
ln 12
8 (3).
2
2
2
2(2)
k1
k22
(21)
We get then
2

qi =
g 2 qi2
1+ (1 )
where, with the required accuracy,

d 2+2 k2
k qi ) ,
1 + g 2 f (k,
2
2
k (k qi )
(22)
55
11
f (k1 , k2 ) =
f (k1 , k2 ) + f (k1 , 0) + f (0, k2 )
3

11
11 67
404 11
=
+
+ 2 (2) +
(2) 6 (3)
3
3
9
27
3
2
2
2

2
2
k1
k2
k12
k12
k
ln
ln 12
.
2
2
2
k2
k2
1
(23)
The representation (22)(23) is extremely convenient, since it permits to get easily the known
expression for the trajectory in the limit 0 [19]. But its main advantage is that it gives the
possibility to perform explicitly the cancellation of the infrared singularities and to write the
kernel at the physical spacetime dimension D = 4, as it will be shown in Section 6.
5. The real part of the color octet kernel
The real part of the color octet kernel (see Eq. (68) of Ref. [4]) expressed through the renormalized coupling constant g has the form

q1 , q2 |K r(8) |
q1 , q2
q12 q2 2 + q1 2 q22

2
q

2 1+ (1 )
q12 q22
k2

2

404
11
k
11 67
2 11
1 + g
+
2 (2) +
+ 14 (3) + (2)
+
3
3
9
27
3
2
2 2
2
2
2
2

q q
q
q
q
q
1
1
11
+ g 2 q 2
ln 1 1 + ln 12 ln 22 + ln 12 ln 22
3
2
2
q
q
q
q
q 2 k2

2
q12
q
q2 q 2 + q22 q1 2 2 q12
q12 q2 2 q22 q1 2
1
ln
ln
1 2
+
+ ln2 12
2
q1
q1 2
q1 2
k2
k2

2 2

q q
11 1
q12 q1 2 q12 q2 2 q22 q1 2

ln 1 1
+ q2 k2 q12 q1 2 + 2
4

3
2
k

q12 q2 2 q22 q1 2 2
q1 q1 2 I q12 , q1 2 , k2
+ (
q1 q2 , q1 q2 ),
+
(24)
k2
g 2 2
= (
q q )
where
I
q12 , q1 2 , k2
1
=
0
q1 (1 x) + q1 2 x
ln
.
q12 (1 x) + q1 2 x k2 x(1 x)
k2 x(1 x)
dx
(25)
Note that I (a, b, c) is a totally symmetric function of the variables a, b and c [21], as it is obvious
from the representation
1 1 1
I (a, b, c) =
0
dx1 dx2 dx3 (1 x1 x2 x3 )

.
(ax1 + bx2 + cx3 )(x1 x2 + x1 x3 + x2 x3 )
(26)
56
We will use also the representation

1
I (a, b, c) =
1
dx
dz
0
1
.
cx(1 x)z + (b(1 x) + ax)(1 z)
(27)
The leading singularity in (24) is 1/. It turns again into 1/ 2 after subsequent integrations
of the kernel because of the singular behavior at k2 = 0. The additional singularity arises from
the region where | ln k2 /2 | 1. For this reason, we have not expanded in the term (k2 /2 ) .
The terms are taken into account in the coefficient of the expression divergent at k2 = 0 in
order to save all non-vanishing contributions in the limit 0 after the integrations.
6. Cancellation of the infrared singularities
Let us introduce the cut-off 0, making it tending to zero after taking the limit 0,
and divide the integration region in the integral representation of the trajectory (22) into three
domains. In two of them either k2 2 , or (k qi )2 2 , and in the third one both k2 > 2 and
(k qi )2 > 2 . Then in the third domain we can take the limit = 0 in (23) and put f (k1 , k2 ) =
(0)
f (k1 , k2 ), where
2 2
2
2
k1 k2
k12
k
67
11
f(0) (k1 , k2 ) =
(28)
2 (2)
ln
ln
ln 12
.
2
2
9
3
k
k2
2 k
12
In the first domain we have
2

11
k
404 11
11 67

+ 2 (2) +
(2) 6 (3) ,
f (k, k qi ) =
3
3
9
27
3
2
(29)
and in the second one we have the same expression with the substitution k2 (k qi )2 . Writing
the real part of the color octet kernel as

q1 , q2 |K r(8) |
q1 , q2 =
q1 , q2 |K r(8) |
q1 , q2 2 k2 +
q1 , q2 k2 2 ,

q1 , q2 |K r(8) |
(30)
and comparing (24) with (29) we see that in the planar kernel (13) the first term in the r.h.s.
of (30) cancels almost completely the contributions of the regions k2 2 and (k qi )2 2
in the trajectories (
qi2 ). The only piece which remains uncancelled in each of the trajectories
for 0 is
2

2+2 2
2
2
g 4

2
d
k
k
2 = s ()Nc (3).
16
(3)
k
(31)
1+ (1 )
2
2 2
k2
On account of this cancellation and using the equality
2

2
k
q 2
(k q)2
d k
ln 2 ln
= (3),
4 k2 (k q)2
q
q 2
we can put
q1 , q2

q1 , q2 | 1 + 2 + 2K r(8) |

q1 k)
s ()Nc
2
k(
2
q22 )
k
+
2
= (
q11 )(
d
2
4 2
q1 k)
k2
k2 (
(32)
57

s ()Nc

+
V (k) + V (k, k q1 ) 3s ()Nc (3)
2 2
q1 q2 + q1 2 q22

s ()Nc
2
q

+ (
q q )
4 2 q12 q22
k2

2
s ()Nc
11
k
67
1+
ln 2 +
2 (2)
4
3
9

2 2
2
2
q1 q1
q
q
s ()Nc 2 11
1
q
ln
+
+ ln 12 ln 22
2
2

4
3
2
q

q
q k
2
2
q
q
q
q 2 q 2 + q22 q1 2 2 q12
1
1
+ ln 12 ln 22 + ln2 12
ln
1 2
2
2
q
q
q1
q1 2
k2
2

2 2

q2 q 2 q22 q1 2
q
q q
11 1
+ 1 2
q12 q1 2
ln 12
ln 1 1
+ q 2 k2 q12 q1 2 + 2
3
2
q1
k2
k4

q2 q 2 q22 q1 2 2
q1 q1 2 I q12 , q1 2 , k2
q12 q2 2 q22 q1 2 + 1 2
k2
+ (
q1 q2 , q1 q2 ),
where
=
V (k)
2

1 67
11
k
2 (2)
ln 2
,
3
2k2 9
q) =
V (k,

2 2
q 2
11
k q
67
11
kq
ln
ln
+ 2 (2)
9
2k2 q 2 3
2 (k q)2
12k2
(k q)2
2
k
11
ln
.
2

12
q
(k q)2
(33)
(34)
(35)
Thus we have obtained this piece of the kernel at D = 4. Of course, the infrared singularities
in (33) must be regularized either by limitations on the integration regions as discussed above or
in an equivalent way.
7. The symmetric part of the kernel
The symmetric part of the kernel was found in the momentum representation in Ref. [5].
It contains neither ultraviolet nor infrared singularities and therefore it does not require regularization and renormalization. For this reason one can use from the beginning physical spacetime
dimension D = 4 and renormalized coupling constant s (). Nevertheless, in the momentum
representation the symmetric part is the most complicated piece of the gluon contribution to
the kernel. In this respect the symmetric part is analogous to the Abelian part of the quark
contribution which was considered in Ref. [15] and turned out to be surprisingly simple in the
coordinate representation. But contrary to the Abelian part, the symmetric part contains the
subtraction term which makes its transformation more complicated and its form in the coordinate
representation considerably intricate.
As well as for the Abelian part it is better to start from the expressions for the symmetric part before integration over the momenta of the produced particles. The starting point is
58
Eq. (3.43) of Ref. [5], where a symmetrization operator appears. The use of this symmetrization
operator is inconvenient, because our definition of the kernel (6) differs from the definition of
Ref. [5]. Instead, we reconstruct the explicit symmetry by restoring the term represented by the
last line in Eq. (3.40) of Ref. [5], which was omitted in Eq. (3.43). As a result, we decompose
the symmetric part into two pieces:
K s = K s1 + K s2 ,
(36)
where
1 2 ()N 2
q1 , q2 = (
q q ) 2 2 s 3 c

q1 , q2 |K s1 |
q1 q2 2
1

dx
d 2 k1 Fs (k1 , k2 )
2
x(1 x) +
(37)
and
q1 , q2

q1 , q2 |K s2 |
2

k
q1 , q1 k1 ; q)KrB (
q1 k1 , q1 ; q)
d 2 k1 KrB (

ln 22 .
= (
q q )
2
2
2
2

4
q1 q2
k1
(
q1 k1 ) (
q 2 + k1 )
Here k1 + k2 = k = q11 , the subscript + means

f (x)
1
1
f (x) f (0) +
f (x) f (1) ,
x(1 x) + x
(1 x)
(38)
(39)
the function Fs (k1 , k2 ) is defined in Eqs. (3.44), (3.45) and (4.1) of Ref. [5], and KrB is the real
part of the leading order kernel,

s ()Nc q12 q2 2 + q1 2 q22

B

2
Kr (
(40)
q1 , q1 ; q) =
q .
2 2
k2
8. Dipole form of the kernel
With our normalizations, the kernel K in the coordinate representation is given by
2
2
d 2 q1 d 2 q2

r r = d q1 d q2
q1 , q2 ei[q1 r1 +q2 r2 q1 r1 q2 r2 ] .

q1 , q2 |K p + K s |
r1 r2 |K|
1 2
2 2 2 2
(41)
q1 , q2 vanishes when any of the qi s or qi s tends to zero.
The real part of the kernel
q1 , q2 |K r |
This property, together with the gauge invariance property of the impact factors of colorless
projectiles A A (0, q) = A A (
q , 0) = 0, permits us to change the target impact factors in (5)
so that they acquire the dipole property

1
r , r| q 21 q 22 |B Bd = 0.
(42)
Further, adding the terms independent either of r1 or of r2 (that is possible due to the gauge
r r so that it conserves the dipole property.
invariance) one can change the kernel r1 r2 |K|
1 2

r r can be omitted (see Ref. [14] for deThereafter, terms proportional to (r1 r2 ) in r1 r2 |K|
1 2
tails). We call the remaining part K d the dipole form of the BFKL kernel. In the LO, discussed
59
in detail in Ref. [14], it coincides with the dipole kernel:

r1 r2 |K dLO |r1 r2 =
s ()Nc
2 2

d
2

r12
2 r2
r1
2

(r11 )(r2 ) + (r1 )(r22 ) (r11 )(r22 ) .
(43)
Here ri = ri and rij = ri rj .

Note that the integrand in (43) contains ultraviolet singularities at = r1 and = r2 which
cancel in the sum of the contributions with account of the dipole property of the target
impact factors. The coefficient of (r11 )(r22 ) is written in the integral form in order to make
the cancellation evident. The singularities do not permit us to perform the integration in this
coefficient.
In the NLO the dipole form can be written as

s2 ()Nc2
NLO

(r11 )(r22 ) d g 0 (r1 , r2 ; )
r1 r2 |Kd |r1 r2 =
4 3

1
+ (r11 )g(r1 , r2 ; r2 ) + (r22 )g(r2 , r1 ; r1 ) + g(r1 , r2 ; r1 , r2 ) ,
(44)
with the functions g turning into zero when their first two arguments coincide.
The first three terms in the r.h.s. of (44) contain ultraviolet singularities which cancel in their
sum, as well as in the LO, with account of the dipole property of the target impact factors.
The coefficient of (r11 )(r22 ) is written in the integral form in order to make the cancellation
evident.
In the next sections we will find the functions g.
9. Transformation of the planar part
Using (13) and (33) and omitting terms with (r1 2 ) in the coordinate space, we reduce the
NLO piece of the planar part to the form:
q1 , q2

q1 , q2 |K pNLO |

2 ()N 2
+ V (k,
k q1 ) 3 (3)
d k V (k)
q22 )
q11 )(
s 3 c (
4

2
2
q
q
(
q1 q2 )

+ (
q q ) V (k) + 2V (k, q1 ) + 2 2 ln 12 ln 22
q

q
4
q1 q2
2

2
q
q
1
+ ln2 12
ln2 12
2

q1
q

2k
1

2
q2 (kq2 ) (
(
q1 + k)
q1 q2 ) (
q1 k) 2 2 2
(
q1 k)
+
+
I q1 , q1 , k
+
2
2
q2
q1
q22
q12 k2
k2
2
2 2
2
q2
2
q2 q2
1
(kq1 )
1
2 q
2 q
+
ln
ln
+
ln
+
ln
2
q2 2
q2 2
q1 2
2k2 q12
k4
60
2
2 2

2
2 2
2
2 2

q
q q
q
q q
q
q q
1
1
ln 12 ln 1 1
+ 2 ln 1 ln 12 2 2 + ln 22 ln 1 4 2
2
q
q
q1
4
q1
q2 q1
k4
k2

+ (
q1 q2 , q1 q2 ) .
(45)
Contributions to the coefficient of (r11 )(r22 ) in the square brackets of (44) come only from
and (3) in (45). The terms with V (k)
contribute to this coefficient as the
the terms with V (k)
integral

r

12 1 ,
ei k
V (r12 ) = 2 d k V (k)
(46)
with ultraviolet divergence. The divergence appears as a result of the separation of the ultraviolet
+ V (k,
k q1 ) into two pieces. We have to represent the contribution of
non-singular sum V (k)
in the integral form in order to make evident the cancellation of ultraviolet
the terms with V (k)
singularities of separate terms in (44), as it was already mentioned in Section 8. We use the same
trick as in Section 5 of Ref. [14] and obtain with the help of the integrals (A.2) and (A.3) of
Appendix A
2 2
67
q k)
11
k q
d k d q d (
2 (2)
ln
6
2(2)2 q2 k2 9
4
i[k r +q r ]

q )r1

2 e i(k+
2e 1
ei(k+q )r2
2 2

2
2

r1 r2
r2
r12
1
11
1
d 2 2 ln
=
+ 2 2 ln 2
,
4
12
r
r1 r2
r2
r1
r1
V (r12 ) =
(47)
where
ln r2
2
3 67
= 2(1) ln
2 (2) .
4
11 9
(48)
For uniformity we write also the contribution of the term with (3) in integral form via the
representation
1
(3) =
4

d
2
r12
2 r2
r1
2
2
2
r1
r2
ln 2 ln 2 .
r12
r12
(49)
Using the subscript p to denote the contributions of the planar part we have
=
gp0 (r1 , r2 ; )
2
2
2 2
2
2
r1
r1 r2
r2
3 r12
11 r12
ln
ln
ln
2
2
2
2
2
2
2 r1 r2
12 r1 r2
r4
r12
r12

2

r2
1
1
+ 2 2 ln 2
.
r2
r1
r1
(50)
The contribution to g(r1 , r2 ; r2 ) comes from all terms which do not depend on q1 but depend
on k and q2 . The transformation of these terms into the coordinate representation by means of
the integrals (A.2)(A.12) and (A.14) of Appendix A gives
61
g p (r1 , r2 ; r2 )
2
r12
r
r12
11 1
1
11 r12
11 1
ln
ln
+
+
ln 22
2
2
2
2
2
2
2
2
6 r12
6 r22 r12
6 r22
r
r
r12
r12

2
2
r
r
r
2
1
1 (r22 r12 )
1 2(r22 r12 )
2 ln2 212 +
2 ln 212 ln 212
+
2
2
2
2
4 r22 r12
4
r22
r12
r22 r12
r12
r22
r12

2

2
(r22 r12 )
2
2
.
1 I r22
12
12
+
(51)
, r
, r
2
2
r22 r12
The tilde is used since the expression obtained does not turn into zero when its first two arguments coincide, as it is required. One can see that the only non-vanishing term for r1 r2
in this expression is the first one. Since this term does not depend on r2 , we omit it to obtain
gp (r1 , r2 ; r2 ):
2
r12
11 1
gp (r1 , r2 ; r2 ) = g p (r1 , r2 ; r2 ) +
(52)
ln
.
2
6 r12
r2
Note that gp (r1 , r2 ; r2 ) has non-integrable ultraviolet singularities at r2 = r2 and r2 = r1 . The
former singularity cancels the corresponding singularity in gp0 (r1 , r2 ; ) at = r2 , while the
latter is unessential because of the dipole property of target impact factors.
To calculate the contribution of the remaining terms in (45) we need the integrals (A.13) and
(A.10)(A.11) of Appendix A and (B.18) of Appendix B, in addition to the integrals used before.
We get
g p (r1 , r2 ; r1 , r2 )
2 2
2 2

r21 r12
r r22
1
(r21 r12 )
(r11 r22 )
= 2
ln 2 2 + 2 2 ln 11
2
2
2
4r1 2 r11 r22
r1 2 r12
r21 r12
r12 2 r12

2 2
r11
(r12 r1 2 )
(r11 r12 )
r22 r1 2
2
+
+
1 2 , r12
2
I r11
, r

2
2
2
2
2r22 r1 2
r11
r1 2 r1
r11 r2
4
2 4

r
r r
r22
r12
r1 2
+ 2 2
ln 2 1 22
+ 2 ln 116 12
2
2r22 r11 r12
r11 r12
r1 2
r1 2
2
2

r
r
r22
r21
r12
+ 2 2
ln 212 + 2 ln 212
+ (1 2).
2
2r22 r1 2 r1 2
r11
r12
r1 2
(53)
Here and hereafter (1 2) means both 1 2 and 1 2 substitutions. The derivatives in this
equality can be calculated using the identity (B.8) of Appendix B. We find
g p (r1 , r2 ; r1 , r2 )

2
2
r
r12
(r22 r1 2 )
(r22 r12 )(r11 r12 )
(r22 r12 )
= 2 2 2 + 2 2 2 2
ln 2
2 2 2 ln 211
r11 r22 r1 2
r11 r22 r1 2 r12
r1 2
2r22 r1 2 r12
r1 2
2 2
2 2

r r12
r r22
1
(r11 r22 )
(r21 r12 )
+ 2
ln 21
+ 2 2 ln 11
2 r2
2 r2
2
4r1 2 r11
r

r

r

r
12 2 r12
22
1 2 12
21 12
2
r
(r22 r21 )
+ 2 2 2 ln 212 + (1 2).
2r22 r1 2 r21
r11
(54)
62
We could obtain gp (r1 , r2 ; r1 , r2 ) subtracting from g p the half-sum of its values with r2 changed
into r1 and vice versa. Since gp is not unique, we prefer to construct a shorter form for it. We
have
gp (r1 , r2 ; r1 , r2 )
= g p (r1 , r2 ; r1 , r2 ) +
r1 2
r11
(r22 r21 )
1
ln
ln
+
2
2
2
2
2
2
2r1 2 r22 r21
r22
2r1 2 r12
r12 2

+ (1 2)
2
r
1
2 2 ln 212
r11 r1 2
r1 2

2 2
r r22
1
(r11 r22 ) (r21 r12 ) 2(r22 r21 )
= 2
+ 2 2 2 2
ln 11
2
2
2
4r1 2 r11 r22
r21 r12
r22 r21
r12 2 r12
2 2
r r12
r11
(r11 r22 )
(r12 r22 )
+ 2 2 2 ln 21
ln
2
2
2
2
2
4r11 r22 r1 2
r11 r22
2r1 2 r22 r12
r12 2

2
r
(r22 r12 )
(r22 r11 )
(r22 r12 )(r11 r12 )
+ 2 2 2 2 2 2 + 2 2 2 2
ln 212 + (1 2).
r11 r22 r1 2
r11 r22 r1 2
r11 r22 r1 2 r12
r1 2
(55)
The dipole property of this term is explicit.

10. Transformation of the symmetric part
In the coordinate representation the piece K s1 , see Eq. (37), is given by the integral
2 ()N 2
r1 r2 |K s1 |r1 r2 = s 4 c
4

1
d q1 d q2 d k1 d k2 ei[q1 r11 +q2 r22 +kr1 2 ]

2 2 2 2
q12 q22
dx
0
Fs (k1 , k2 )
x(1 x)
(56)
.
+
Here Fs (k1 , k2 ) is given by Eqs. (3.44), (3.45) and (4.1) of Ref. [5] and k = k1 + k2 . We restrict
ourselves to the dipole form of the kernel. Hence, we omit those terms in Fs (k1 , k2 ), which lead
to (r1 2 ) in the coordinate representation. Decomposing the remaining terms and taking into
account the symmetry of the integration measure in (56) with regard to the substitution k1 k2 ,
we can make the replacement
j
k1i k2 x2 ij
x1 ij k1i k2
x1 ij x2 ij
Fs (k1 , k2 )
2
a
2
a1 2 2 + 2
a
a ,
2
2
2
2
2

11 k1 k2
11 1 22 2
q1 q2
k1 k2 22
where
x1 = x,
x2 = 1 x,
11 = (k1 x1 q1 )2 + x1 x2 q12 ,
22 = (k2 + x2 q2 )2 + x1 x2 q22 ,

and
(57)
63
ij
(
q1 k1 )
=
x2 1 2 2
+
2
q1

ij
(
q2 k2 )
ij
a2 =
x1 1 + 2 2
2
q2
ij
a1
q1i (q1 k1 )j
k1i (q1 k1 )j
x2 k1i q1
+
,
x1 q12
q12
k12
j
(q2 + k2 )i q2
(q2 + k2 )i k2
x1 q2i k2
.
x2 q22
q22
k22
(58)
Since the first term in (57) does not depend on q1 , the second is independent of q2 and the third
one depends on all momenta, they contribute to g(r1 , r2 ; r2 ), g(r2 , r1 ; r1 ) and g(r1 , r2 ; r1 ; r2 )
in (44), respectively. We denote these contributions with the s1 subscript. For the first term we
have
j
ij
ki k
k i (k2 + q2 )i 1
x2 a2
1
1
1
2 12 22
= 12
x1
22 (k2 + q2 )2
q22 k22
k1 k2 22 x(1 x) +
k1

qi
1
1
22
(59)
2 .
x2 q2 22 k
2
Note, that infrared singularities vanish here, as well as in other terms in the r.h.s. of (57), in a
rather tricky way, namely by means of the + prescription. Indeed, each term in (57) contains
non-integrable infrared singularities. But these terms do not depend on x so that the singularities
vanish after the subtraction.
For the term in question it seems more convenient to perform first the integration over x. We
obtain
1
2
0
j
ij
k1i k2
x2 a2
dx 2 2
k1 k2 22 x(1 x) +

k1i (k2 + q2 )i 1
q2i
1
(k2 + q2 )2
+
ln
.
k12 (k2 + q2 )2 q22 k22
k22
q22 k22
(60)
Then we use the integrals (B.1) and (B.2) of Appendix B to find the contribution of this term to
gs1 (r1 , r2 ; r2 )

2
2

r12
r12
(r22 r12 )
(r22 r12 )2 2 2
1
2
+
ln
.
gs1 (r1 , r2 ; r2 ) = 1 2 2
I r12 , r12 , r22
ln

2 r2
2
2
2
r22 r12
2r22
2r12
r22
r12
12

(61)
One can see that the second term in (57) equals the first one after the substitution q1
q2 ,
k1 k2 , x1 x2 . Hence, we can construct its integrated form replacing r22 r11 , r1 2
r12 , r12 r1 2 in (61). The third term in (57) is easier to integrate with respect to momenta
before the convolution. We introduce
l1 = k1 x q1 ,
l2 = k2 + (1 x)

q2 ,
p1 = q1 k1 ,
p2 = q2 + k2 .
In this notation

ij
ij
2x1 a1 x2 a2
x(1 x)11 22 +

j
j
j
l1 q1i
k1i p1
1 ij
(
q1 l1 )
x2 l1i q1
=2
+
+
x2 (1 2x1 ) 2 2
+
11 2
x1 q12
q1
q12
k12
64

j
j
j
l2i q2
k2 p2i
1 ij
(
q2 l2 )
x1 l2 q2i
2 2
x1 (1 2x2 ) + 2 2
22 2
x2 q22
q2
q2
k2

j j
j
j
k (k2 + q2 )i
2k2 q2i q1i
k1i (k1 q1 )j
q
2k i q
+
.
+ 12 12 22 + 22
x1 k1 q1 q2
k2 (k2 + q2 )2 x2 k22 q22 q12 k12 (k1 q1 )2
(62)
Then we use the integrals (B.3)(B.5) of Appendix B to find

ij
ij
2x1 a1 x2 a2
d 2 q 1 d 2 q 2 d 2 k1 d 2 k2

ei[q1 r11 +q2 r22 +kr1 2 ]
2 2 2 2 x(1 x)11 22 +

j
i rj
i rj
r12
r12
2 ij
x2 r11 r1i 2
(r11 r1 2 )
1 2
11
=
x2 1 + 2
2
d1 d2 2
x1 r12 2
r12 2
r12 2
r11
ij
j i
j
j i
i
r r
r r
x1 r22 r1 2
(r22 r1 2 )
212 22
x1 1 2
+ 212 1 2 +
2
2
2
x2 r1 2
r1 2
r1 2
r22
+
2 (r22 r12 ) (r11 r12 )

2 (r11 r21 ) (r22 r21 )
+
,
2 r2
2 r2
2 2
2 2
x1 r11
x
r

2 r22 r12 r11 r1 2
21
22 1 2
(63)
2 + x r2 and d = x r2 + x r2 . The subsequent integration over x is straightwhere d1 = x1 r12

2 11
2
1 22
2 21
forward. We obtain
g s1 (r1 , r2 ; r1 , r2 )

= gs1 (r1 , r2 ; r1 , r2 )
+
2
r11
2 r2 )
r14 2 (r12

11
2 + (
r11
r11 r12 )

2 r2 )
2
2
(r12

11 r1 2 r11
ln
2
r12

2
r11
ln
2
r12

2
r11
2
r22

2 r2 )
(r21

22
2 + (
r22 r21 )
r22

2 r2 )
2
2
(r21

22 r1 2 r22
ln
2
r21

2
r22
ln
2
r21

2
r22

2
(64)
Here
gs1 (r1 , r2 ; r1 , r2 )
2 2
2 2

2
2
r12 r21
22
(r1 2 r21 )r12
(r1 2 r12 )r12
1 r11
1
r
= 4
ln 2 2
1 + 2
2
2
d
r1 2
r11 r22
d r1 2
r11
r22

2(r22 r21 )(r12 r21 ) 2(r11 r12 )(r12 r12 )
2
2
r21
r12

2 2
r r21
2(r22 r12 )(r11 r21 ) 2
+
r1 2 2r12 2 ln 12
2 r2
2 r2
r11
r
11
22
22

r2
1
(r12 r12 ) (r22 r12 ) (r22 r12 )
+ 2 2 2 + 2 2 2 2
ln 21
2
r22
r1 2 r12
r1 2 r11
r11 r12
r22

2
r
1 (r12 r21 ) (r11 r12 ) (r11 r21 )
+ 2
2 2 2 2
,
ln 12
2
2
2
r11 r1 2 r21
r1 2 r22
r22 r21
r11

+
(65)
2 r2 r2 r2 . One can see that g vanishes when r equals r while the remaining
where d = r12
21
s1
1
2
11 22
terms in (64) are independent of either r1 or r2 , therefore we dropped them.
65
Now we turn to K s2 . We can rewrite this expression in a form convenient for the integration
r1 , r2 |K s2 |r1 , r2
2

k
s2 ()Nc2
d q1 d q2 d k1 d k2

=
ln 22 ei[q1 r11 +q2 r22 +kr1 2 ]
4

2
2
2
2
4
k1

1
1 k2
q2
q1 k1
q1
q2 + k1
2 2+ 2 2
+
q2 + k1 )2 q22 (q1 k1 )2 q12
k1 k2 k1 k2 (

(
q1 k1 )(
q2 + k1 )
(
q1 q2 )
+ 2 2
q1 q2
(
q1 k1 )2 (
q2 + k1 )2

q2 + k1 )
q1
q1 k1
(
q1 k1 )(
k2
q2 + k1
k1 q2
+ 2
q2 + k1 )2 (
q1 k1 )2
(
q1 k1 )2 (
q2 + k1 )2
k1 q22 q12
k22 (

(
q1 q2 ) k2
q2 + k1 )
q1 k1
(
q1 k1 )(
q2 + k1
2 2
(66)
.
q1 q2 k22 (
q2 + k1 )2 (
q1 k1 )2
(
q1 k1 )2 (
q2 + k1 )2
The first term in the square brackets vanishes in the integration. Again, we omit terms proportional to (r1 2 ) in the coordinate representation. The remaining terms which do not depend
on q1 contribute to gs2 (r1 , r2 ; r2 ), the ones independent of q2 contribute to gs2 (r2 , r1 ; r1 ) and the
others to gs2 (r1 , r2 ; r1 , r2 ). During the calculation we used the integrals (B.2) and (B.6)(B.18)
of Appendix B. Finally, after the integration we obtain
2
2
r12
r12
1
(r12 r22 ) 2 r12

2 2 ln
,
gs2 (r1 , r2 ; r2 ) = 2 ln 2 ln 2
(67)
2
4r12
r22
r12
4r12 r22
r12

g s2 (r1 , r2 ; r1 , r2 ) = g p (r1 , r2 ; r1 , r2 )
(r22 r12 )
1
+ 2 2
2
2
2
r11 r22 r1 2
r11 r1 2

r12
ln 2
+ (1 2) .
r1 2
(68)
Hence,
gs2 (r1 , r2 ; r1 , r2 ) = gp (r1 , r2 ; r1 , r2 )

r12
(r22 r12 )
ln 2
+ (1 2) .
2 r2 r2
r11
r1 2
22 1 2
(69)
Here we used the dipole form of gp and dropped the terms independent of r1 or of r2 .
11. Final result
In this section we gather the contributions from the planar and symmetric parts of the
kernel to the functions g defined in (44). We have
2
2
2
r1
r2
3 r12
g (r1 , r2 ; ) =
ln 2 ln 2
2
2
2 r1 r2
r12
r12
2 2

2
2

r1 r2
r2
1
11 r12
1
ln
+
ln
,
2 r2
2
2
2
12 r1
r4
r2
r1
r1
2
0
(70)
66
g(r1 , r2 ; r2 ) =
2
r
r12
11 r12
1
11 1
ln
+
ln 22
2 r2
2
2
2
2
6 r22
6
r
r

r

r
12
12
22
12
2
2
2
2
2
r
r
r
r
r
1
+ 2 ln 12
ln 212 2 12 2 ln 212 ln 212 ,
2
2r22
r22
r12
2r22 r12
r22
r12
where
ln r2 = 2(1) ln
2
3 67
2 (2) .
4
11 9
(71)
(72)
and g(r1 , r2 ; )

vanish at r1 = r2 . Then, these functions turn
One can see that both g 0 (r1 , r2 ; )
into zero for 2 faster than (2 )1 to provide the infrared safety. The ultraviolet singularities of these functions at = r2 and = r1 cancel in the sum of the first three contributions in
the r.h.s. of (44) on account of the dipole property of the target impact factors.
The last term in the r.h.s. of (44) is the most complicated one:
g(r1 , r2 ; r1 , r2 )

2
r
2(r22 r11 )
2(r22 r12 )(r11 r12 )
(r22 r12 )
= 2 2 2 2 2 2 +
ln 212
2
2
2
2
r11 r22 r1 2
r11 r22 r1 2
r11 r22 r1 2 r12
r1 2

2 2
r r22
1
(r11 r22 ) (r21 r12 ) 2(r22 r21 )
+ 2
+ 2 2 2 2
ln 11
2 r2
2
2r1 2 r11
r

r

r

r

r
12 2 r12
22
21 12
22 21
2 2

2
r r12
(r1 2 r12 )r12
(r11 r22 )
2(r22 r21 )(r12 r21 )
1
+ 2 2 2 ln 21
+
+
2
2
2
2
2
2r11 r22 r1 2
r11 r22
d r1 2
r11
r21

2 2
2 2
2 2
r12 r21
r12 r21
r11 r22
(r22 r12 )(r11 r21 ) 2
1
2
+
r1 2 r1 2 ln 2 2
ln 2 2
+ 4
1
2 r2
d
r11
r11 r22
2r1 2
r11 r22
22
2
r
r12
1 (r12 r21 ) (r11 r12 ) (r11 r21 )
(r12 r22 )
+ 2
2 2 2 2
ln 2
2 2 2 ln 211
2
2
r11 r1 2 r21
r1 2 r22
r22 r21
r11
r1 2 r22 r12
r1 2
+ (1 2),
2 r2
= r12
21
(73)
2 r2 .
r11
22
where d
This term also vanishes at r1 = r2 , so that it possesses the dipole property. It has ultraviolet
singularity only at r1 2 = 0 and tends to zero at large r1 2 and r2 2 sufficiently quickly in order to
provide the infrared safety.
12. Conclusion
The coordinate representation of the BFKL kernel is extremely interesting, because it gives
the possibility to understand its conformal properties and the relation between the BFKL and the
color dipole approaches. Generally speaking, the BFKL kernel is not equivalent to the dipole one.
Actually the first one is more general than the second. This is clear, because the BFKL kernel can
be applied not only in the case of scattering of colorless objects. However, when applied to the
latter case, we can use the dipole and gauge invariance properties of targets and projectiles
and omit the terms in the kernel proportional to (r1 2 ), as well as change the terms independent
either of r1 or of r2 in such a way that the resulting kernel becomes conserving the dipole
67
property, i.e. the property which provides the vanishing of cross-sections for scattering of zerosize dipoles. The coordinate representation of the kernel obtained in such a way is what we call
the dipole form of the BFKL kernel. We have found the dipole form of the gluon contribution
to the BFKL kernel in the NLO by the transfer of the kernel from the momentum representation
where it was calculated before. This paper completes the transformation of the NLO BFKL kernel
to the dipole form, started a few months ago with the quark part of the kernel [14,15].
The striking result of [14,15] was the simplicity of the dipole form of the quark contribution
to the kernel. Moreover, it was shown that the dipole form agrees with the quark contribution to
the BK kernel obtained in [16].
As it can be seen from the results of this paper, the dipole form of the gluon contribution to
the kernel is also extremely simple in the coordinate representation. This holds especially for the
symmetric part of the kernel, the momentum representation of which has a very complicated
form [5].
We do not have the possibility to compare the results obtained for the NLO gluon contribution
with the BK kernel, since the latter has not been obtained yet. Instead, our results can be used for
finding the NLO BK kernel. In this respect, one has to bear in mind the ambiguity of the NLO
kernel related to the operator transformation (10) with an appropriate O.

Note that in contrast to the LO, as well as to the quark contribution, the functions g 0 (r1 , r2 ; )

and g(r1 , r2 ; )
in the dipole form (44) turned out to be unequal. Although the function
g 0 (r1 , r2 ; )
can be changed by adding any function with zero integral over ,
the inequality
cannot be removed. On the other hand, according to [16] in the color dipole approach these
functions should be equal.
We have to say that our consideration was not completely rigorous. In particular, we did
not regularize the ultraviolet singularities arising as a result of separation of the ultraviolet non + V (k,
k q) in (45) into two pieces. Instead of the regularization we used the
singular sum V (k)
trick of representing the coefficient of (r11 )(r22 ) in (44) in integral form. Note, however, that
the trick is the same which was used in Ref. [14], where its validity was checked by completely
rigorous calculations. This permits to rely on the results obtained here.
Acknowledgement
We would like to thank R.E. Gerasimov for calculating the integral (B.18).
Appendix A
Here we present a list of the integrals necessary to perform the Fourier transform of the planar
part of the kernel. Most of them are calculated straightforwardly via the exponential representation:

1
j
a =
(A.1)
d j 1 ea .
(j )
0
We have

d k i k
ir
d k i k
2
r k

= 2,
e
e r ln k2 = 2 ,
2

2
2
r

r

k

2 2

d k i k
k2
ir
r
r k
ln 2 = 2 2(1) ln
e
,
2

2
4
r
k
(A.2)
(A.3)
68

d q
2
d q
2
d q
2
d q
2
d q
2
d q
2

q k)
(k + q)2
(r )

( r)2
d k i[q r+k]
(
ln
=
ln
(A.4)
e
,
2
q 2
r2 2
2
q2 k2

2 (k + q)2
d k i[q r+k]
q k)
(r )
2 ( r)2
(
ln
= 2 2 ln
e
,
(A.5)
2
q 2
r
2
q2 k2
2
2 k2
d k i[q r+k]
q k)
(r )

(
ln 2 = 2 2 ln2 2 ,
e
(A.6)
2
2
2
q
r
r
q k

2
q k)
1 2 (r + )
d k i[q r+k]
2
1
2 (
ln
=
ln
e
,
(A.7)
2
q2
2
r2
k2

2
d k i[q r+k]
(
q k)
1
(r + )
2
1
(A.8)
ln
= 2 ln
e
,
2
q2

r2
k2

2 (
2
(
q k)
q k)
1
(r + )
2
d k i[q r+k]
(r + )
2
1
ln
ln
= 2 ln
e
ln
.
2
q2
q2

r2
2
k2
(A.9)
The integral (A.9) is calculated most easily after the decomposition ln a ln b = 12 (ln2 a + ln2 b
ln2 ab ).

q1 q2 ) 2
d k i[q1 r1 +q2 r2 +k]
(
ln
e
2
q12 q22

r12 r22
4(r1 r2 )
= 2 2 ln
,
(r1 r2 )2 2
r1 r2 2

d q2
d k i[q1 r1 +q2 r2 +k]
q1 q2 ) 2
d q1
(
ln
e
2
2
2
q12 q22
d q1
2
d q2
2
(
q1 + q2 )2
k2
(A.10)
2
r
q12 4(r1 r2 )
= 2 2 ln 12 .
2
2

r

r

k
1 2
(A.11)
The following integrals contain I defined in (25):

d q
2
d q
2
d q
2

d k i[q r+k]

2 , k2 = I 2 , ( + r)2 , r2 ,
I q 2 , (
q k)
e
2
(A.12)
2

d k i[q r+k]
q k)
2
(
2 , k2 = (r )
I q , (
q k)
I , ( + r)2 , r2 ,
e
2
2

2
r

k
(A.13)
2 2

d k i[q r+k]
q k)
2 2
(
2 , k2 = (r )
I q , (
q k)
I , ( + r)2 , r2 .
e
2
2
2
2

2
r

q k
(A.14)
This integral can be expressed through (A.13) and simpler integrals with the help of the identity
(kq ) =
(1 x(1 z))k2 (z + x(1 z))

q2
+
2(1 x)(1 z)
2x(1 z)
2 x(1 x)(1 z) + (
(
q k)
q 2 (1 x) + k2 x)z
.
2(1 x)x(1 z)
(A.15)
69
Appendix B
Here we present a list of the integrals necessary to perform the Fourier transform of the symmetric part of the kernel:

d q
2

p2
d p i[p
p
e r +q ]
ln
2
2
2
q p
(
q + p)
2

2 2

r
i
2
2 , r2 , (r )
, r , (r )
= I
+ 2 (r )I
2
2
r

2

1

ln 2
,
+ ln
2
2
( r)
r
2
4

r

d q

q2
i
d k i q i k

q
e
e r1 ei kr2
ln
= 2 ln 12 ln 2 2 ,
2
2
2
2
2
4
r2
r1 r2
q k
k

d q d l
ei[q r+l ]
1
= 2
,
2
2
2 2 l + x(1 x)
r + x(1 x)2
q

ri j
d q d l qi lj
ei[q r+l ]
,
= 2 2
2
2
2

2 2 q l + x(1 x)
(r + x(1 x)2 )
q

ri j
d q d k qi kj
ei[q r+k ]
,
=
2 2 k2 (1 x)k2 + x q2 r2 ((1 x)r 2 + x 2 )
2

r
d k i k
1
r1
r2 1
i k
e
e
= ln 22 ,
2
2
2
r1
k

(

d q
q k)
q + k)
i r
2

r2
d k i[q r+k]
(
.
e
=
ln
+
ln
2
2
2
4 r2 (r )
2 2 (r )
2
q2 k2 (
q + k)
(B.1)
(B.2)
(B.3)
(B.4)
(B.5)
(B.6)
(B.7)
For calculating the following integrals we use the identities

1
a (c a b) b
I (a, b, c)
=
ln
,
(c + b a)I (a, b, c) + 2 ln +
a
c
a
c
(c a b)2 4ab
(B.8)

1
ax + b(1 x)
2cx dx
ln
ax + b(1 x) cx(1 x)
cx(1 x)
0
b
a
1 a ab
= I (a, b, c)(c + b a) + Li2 1
Li2 1
+ ln ln 2 ,
a
b
2 b c
(B.9)
where
1
Li2 (x) =
0
dt ln(1 xt)
,
t
(B.10)
70
(

q k)
q + k)
d k i[q r+k]
(
ln k2
e
2
2
q2 k2 (
q + k)

2 2
i r 1 2
2
2
2
=

ln
+ r I r , , (r )
4 r2 2
(r )
2

r2
2
r
(r )
2

ln
ln
+
2(1)
+
ln
+
4
2
(r )
2
r2
(r )
2
2
2

2 2
1

r
2
+ 2
(B.11)
,

ln
ln
(
r )I r , , (r )
2
(r )
2
(r )
2
2 2
i
j
d k i[q (r )
d q
k]
q (q + k)
ln k2
e
2
2
2
q2 (
q + k)

2 2

(r )
4
1
j r i
r
= 2 2 ij (r )
+ j r i i r j ln
+
4(1)
ln
16
4r
r2 2
2r 2 2
i j
i
j
2
2

1

r r
(r ) ij
+
(r )
2 +
j r i i r j ln 2 .
2(r )
2 2
r
2r 2 2
r
(B.12)
We simplified the tensor structure of this integral via the identity
d q
2
ij =
j r i + i r j )
i j r2 + r i r j 2 (r )(
.
r2 2 (r )
2
The integral

2
2
d q1
q1 q2 )
(
q1 k)
(
q2 + k)
d q2
d k i[q1 r +q2 r +k
r1 2 ] (
11
22
J=
ln
ln
e
2
2
2
(
q1 + q2 )2 (
q1 + q2 )2
q12 q22
appearing in the planar part can be written as

8
d (r1 r2 ) (r1 r2 )
J= 2
2 r2
2 2
2 r1
r1 2
2 r1 r2
putting

d
d q1 d q2 (
q1 k q1 )eir2 [q1 +q2 q1 q2 ] = 1
2
(2)
(B.13)
(B.14)
(B.15)
(B.16)
into the integrand in (B.14) and then integrating over momenta with the help of (A.2). Using the
same trick we obtain

(

r12 2
d q2
d k i[q1 r +q2 r +k
q1 q2 ) (
q1 k)
q2 + k)
d q1
r1 2 ] (
11
22
(B.17)
=
e
J
2 (
2
2
2
2
8
q12 q22 (
q1 k)
q2 + k)
for the integral appearing in the symmetric part. One can calculate J by means of complex
variables
a + c + a c +
a = ax iay ,
(
a c) =
,
a 2 = a+ a .
a+ = ax + iay ,
2
introducing z = ei , using
Shifting r1 ,
+ = z,
,
z
d =
dz
iz
71
and integrating over z via residues, one gets only trivial integrals over to perform. They yield

2 2
2 2
r21 r12
r r22
(r21 r12 )
2 (r11 r22 )
.
ln 2 2 + 2 2 ln 11
J= 2
(B.18)
2
2
2
r1 2 r11 r22
r1 2 r12
r21 r12
r12 2 r12
References
[1] V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Phys. Lett. B 60 (1975) 50;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Zh. Eksp. Teor. Fiz. 71 (1976) 840, Sov. Phys. JETP 44 (1976) 443;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 72 (1977) 377;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 (1977) 199;
Ya.Ya. Balitskii, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
[2] V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 127;
M. Ciafaloni, G. Camici, Phys. Lett. B 430 (1998) 349.
[3] V.S. Fadin, R. Fiore, A. Papa, Phys. Rev. D 60 (1999) 074025.
[4] V.S. Fadin, D.A. Gorbachev, Pisma Zh. Eksp. Teor. Fiz. 71 (2000) 322, JETP Lett. 71 (2000) 222;
V.S. Fadin, D.A. Gorbachev, Phys. At. Nucl. 63 (2000) 2157, Yad. Fiz. 63 (2000) 2253.
[5] V.S. Fadin, R. Fiore, Phys. Lett. B 610 (2005) 61;
V.S. Fadin, R. Fiore, Phys. Lett. B 621 (2005) 61, Erratum;
V.S. Fadin, R. Fiore, Phys. Rev. D 72 (2005) 014018.
[6] L.N. Lipatov, Nucl. Phys. B 452 (1995) 369;
L.N. Lipatov, Phys. Rep. 286 (1997) 131.
[7] L.N. Lipatov, Sov. Phys. JETP 63 (1986) 904, Zh. Eksp. Teor. Fiz. 90 (1986) 1536.
[8] N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 64 (1994) 631;
N.N. Nikolaev, B.G. Zakharov, V.R. Zoller, JETP Lett. 59 (1994) 6;
A.H. Mueller, Nucl. Phys. B 415 (1994) 373;
A.H. Mueller, B. Patel, Nucl. Phys. B 425 (1994) 471.
[9] L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 170 (1983) 1.
[10] I. Balitsky, Nucl. Phys. B 463 (1996) 99;
Yu. Kovchegov, Phys. Rev. D 60 (1999) 034008.
[11] L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233;
L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 3352;
L. McLerran, R. Venugopalan, Phys. Rev. D 50 (1994) 2225;
E. Iancu, A. Leonidov, L. McLerran, Phys. Lett. B 510 (2001) 133;
E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 692 (2001) 583;
E. Ferreiro, E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 703 (2002) 489;
J. Jalilian-Marian, A. Kovner, A. Leonidov, H. Weigert, Nucl. Phys. B 504 (1997) 415;
J. Jalilian-Marian, A. Kovner, A. Leonidov, H. Weigert, Phys. Rev. D 59 (1999) 014014.
[12] J. Bartels, L.N. Lipatov, G.P. Vacca, Nucl. Phys. B 706 (2005) 391;
J. Bartels, L.N. Lipatov, M. Salvadore, G.P. Vacca, Nucl. Phys. B 726 (2005) 53.
[13] Yu.V. Kovchegov, H. Weigert, hep-ph/0612071.
[14] V.S. Fadin, R. Fiore, A. Papa, Nucl. Phys. B 769 (2007) 108, hep-ph/0612284.
[15] V.S. Fadin, R. Fiore, A. Papa, Phys. Lett. B 647 (2007) 179, hep-ph/0701075.
[16] I. Balitsky, Phys. Rev. D 75 (2007) 014001, hep-ph/0609105.
[17] V.S. Fadin, R. Fiore, Phys. Lett. B 440 (1998) 359.
[18] V.S. Fadin, M.I. Kotsky, R. Fiore, Phys. Lett. B 359 (1995) 181.
[19] V.S. Fadin, R. Fiore, M.I. Kotsky, Phys. Lett. B 387 (1996) 593;
J. Blumlein, V. Ravindran, W.L. van Neerven, Phys. Rev. D 58 (1998) 091502;
V. Del Duca, E.W.N. Glover, JHEP 0110 (2001) 015.
[20] V.S. Fadin, R. Fiore, M.I. Kotsky, A. Papa, Phys. Rev. D 61 (2000) 094005, hep-ph/9908264.
[21] V.S. Fadin, A. Papa, Nucl. Phys. B 640 (2002) 309, hep-ph/0206079.
Thermodynamics of the N = 2 strongly coupled

plasma
Alex Buchel a,b, , Stan Deakin a , Patrick Kerner a , James T. Liu c
a Department of Applied Mathematics, University of Western Ontario, London, Ontario N6A 5B7, Canada
b Perimeter Institute for Theoretical Physics, Waterloo, Ontario N2J 2W9, Canada
c Michigan Center for Theoretical Physics, Randall Laboratory of Physics, The University of Michigan,
Ann Arbor, MI 48109-1040, USA

Received 29 January 2007; accepted 13 June 2007
Abstract
Gauge/string duality is a potentially important framework for addressing the properties of the strongly
coupled quark gluon plasma produced at RHIC. However, constructing an actual string theory dual to QCD
has so far proven elusive. In this paper, we take a partial step towards exploring the QCD plasma by investigating the thermodynamics of a non-conformal system, namely the N = 2 theory, which is obtained as
a mass deformation of the conformal N = 4 gauge theory. We find that at temperatures of order the mass
scale, the thermodynamics of the mass deformed plasma is surprisingly close to that of the conformal gauge
theory plasma. This suggests that many properties of the quark gluon plasma at RHIC may in fact be well
described by even relatively simple models such as that of the conformal N = 4 plasma.
1. Introduction
The purpose of this paper is to explore properties of strongly coupled non-conformal gauge
theory plasma within the framework of the gauge theory/string theory correspondence of Maldacena [1,2]. We are primarily motivated by recent attempts (see [3] and references therein)
to describe the quark gluon plasma (QGP) produced at RHIC [47] from the dual holographic
perspective. Unfortunately, we do not as yet have a string theory dual to QCD. In fact, most
applications of the Maldacena correspondence to RHIC physics are discussed in the context of
E-mail address: abuchel@uwo.ca (A. Buchel).

doi:10.1016/j.nuclphysb.2007.06.019
A. Buchel et al. / Nuclear Physics B 784 (2007) 72102
73
conformal N = 4 supersymmetric SU(N ) YangMills (SYM) theory in the t Hooft (planar)

limit and for large t Hooft coupling. Remarkably, N = 4 SYM plasma [816] as a model for the
RHIC QGP appears to be reasonably good [17]. We would like to stress, however, that such an
agreement is rather paradoxical. One requires a strong t Hooft coupling in order to have a controllable string theory dual, and the QGP does indeed appear to be strongly coupled. However,
it is strongly coupled because it is produced at temperatures of order the QCD strong coupling
scale, where, naively, conformal invariance is badly broken. So, why then does the conformal
gauge theory plasma serve as the model for RHIC physics?
As a first step towards answering the latter question, we study the thermodynamics of the N =
2 (i.e. mass deformed N = 4) SYM plasma over a wide range of temperatures and for different
mass deformations.1 The expectation here is that the deformation mass scale would provide a
model for the QCD strong coupling scale, and the thermodynamics of the mass deformed plasma
at temperatures of order the mass scale would then serve as a more realistic model for the RHIC
QGP.
The paper is organized as follows. In Section 2 we review the non-extremal N = 2 geometry constructed in [21], the holographic renormalization of this theory [24] and also recall the
explicit expressions for the free energy density, energy density and the entropy density [24,25].
In Section 3 we discuss the map between finite temperature N = 2 gauge theory parameters
and the parameters of the dual non-extremal geometry [21,26]. In Section 4 we review the hightemperature thermodynamics of the finite temperature PW flow [21,24,25]. After this, we turn
towards a numerical investigation of the flow away from the high-temperature regime. We describe our numerical procedure in Section 5, and present the results of the analysis in Section 6.
Finally, we conclude and also outline open problems in Section 7.
2. Non-extremal N = 2 geometry
The supergravity background dual to finite temperature N = 2 gauge theory [21] is a deformation of the original AdS5 S 5 geometry induced by a pair of scalars and of the
five-dimensional gauge supergravity. (At zero temperature, such a deformation was constructed
by Pilch and Warner [27].) According to the general scenario of a holographic RG flow, the asymptotic boundary behavior of the supergravity scalars is related to the bosonic and fermionic
mass parameters of the relevant operators inducing the RG flow in the boundary gauge theory.
Based on such a relation, and the fact that and have conformal dimensions two and one,
respectively, we call the supergravity scalar a bosonic deformation, and the supergravity scalar
a fermionic deformation of the D3-brane geometry.
The action of the five-dimensional gauged supergravity including the scalars and is
given by

S=
M5
d 5
gL5 =
1
4G5

d 5
M5

1
g R 3()2 ()2 P ,
4
(2.1)
1 Although there have been previous attempts to study the thermodynamics of strongly coupled non-conformal four
dimensional gauge theory plasmas, they have been limited to the high temperature regime, where the theory is almost
conformal [1822]. A notable difference is a recent study of thermodynamics of strongly coupled N = 2 gauge theory
plasma with massive fundamental hypermultiplets [23].
74
where the potential2

1 1 W 2
W 2
1
P=
+
W2
16 3
(2.2)
is a function of and , and is determined by the superpotential

1
W = e2 e4 cosh(2).
2
In our conventions, the five-dimensional Newtons constant is
G5
G10
5
2 volS 5
4
.
N2
(2.3)
(2.4)
The action (2.1) yields the Einstein equations

4
R = 12 + 4 + g P,
3
as well as the equations for the scalars
=
1 P
,
6
=
1 P
.
2
(2.5)
(2.6)
To construct a finite-temperature version of the PilchWarner flow, we choose an ansatz for

the metric respecting rotational but not the Lorentzian invariance

ds52 = c12 (r) dt 2 + c22 (r) dx12 + dx22 + dx32 + dr 2 .
(2.7)
With this ansatz, the equations of motion for the background become

1 P
= 0,
+ ln c1 c23
6

1 P
= 0,
+ ln c1 c23
2

4
c1 + c1 ln c23 + c1 P = 0,
3

4

2
c2 + c2 ln c1 c2 + c2 P = 0,
(2.8)
3
where the prime denotes a derivative with respect to the radial coordinate r. In addition, there is
a first-order constraint
1
1
1
( )2 + ( )2 P (ln c2 ) (ln c1 c2 ) = 0.
(2.9)
3
3
2
It was shown in [21] that any solution to (2.8) and (2.9) can be lifted to a full ten-dimensional
solution of type IIb supergravity. This includes the metric, the three- and five-form fluxes,
the dilaton and the axion. In particular, the ten-dimensional Einstein frame metric is given by
Eq. (4.12) in [21].
2 We set the five-dimensional gauged supergravity coupling to one. This corresponds to setting the radius L of the
five-dimensional sphere in the undeformed metric to 2.
75
For finite temperature flows, we find it convenient to introduce a new radial coordinate x
which spans the range from the horizon to the boundary in finite coordinate distance:
1 x(r) =
c1
,
c2
x [0, 1].
(2.10)
With this new coordinate, the black branes horizon is at x = 1, while the boundary of the asymptotically AdS5 spacetime is at x = 0.3 The background equations of motion (2.8) become
1
5
4
c2 (c2 )2 + c2 ( )2 = 0,
x1
c2
3

P

1
1
+

(x 1) 6( )2 + 2( )2 c22 3c2 c2 6(c2 )2 (x 1)
2
x 1
12Pc2 (x 1)
= 0,

P

1
1
+

(x 1) 6( )2 + 2( )2 c22 3c2 c2 6(c2 )2 (x 1)
2
x 1
4Pc2 (x 1)
= 0,
(2.11)
c2 + 4c2 ( )2
where the prime now denotes a derivative with respect to x. We demand that a physical RG flow
should correspond to a background geometry with a regular horizon. To ensure regularity, it is
necessary to impose the following boundary conditions at the horizon:

x 1 : (x), (x), c2 (x) {1 , 2 , 3 },
(2.12)
where the i are constants.
The boundary conditions at x = 0 are determined from the requirement that the solution
should approach the AdS5 geometry as x 0+ :

x 0+ : (x), (x), c2 (x) 0, 0, x 1/4 .
(2.13)
The three supergravity parameters i uniquely determine a non-singular RG flow in the dual
gauge theory. As we review in Section 3, they are unambiguously related to the three physical
parameters in the gauge theory: the temperature T , and the bosonic and fermionic masses mb
and mf of the N = 2 hypermultiplet components.
A general analytical solution of the system (2.11) with the boundary conditions (2.12), (2.13)
is unknown. However, it is possible to find an analytical solution in the regime of high temperatures4 (see Section 4). Notice that given (2.2) and (2.3), one can consistently truncate the
supergravity system (2.11) to bosonic deformation only, i.e., by setting = 0 identically. On the
other hand, it is inconsistent (beyond the linear approximation) to set bosonic deformation to
zero, i.e., to set = 0 identically while keeping the fermionic deformation = 0.
In the rest of this section we discuss the asymptotic singularity-free solution of (2.11) near the
boundary x 0+ and near the horizon x 1 , constraint by the boundary conditions (2.13)
and (2.12), respectively. We justify referring to the and scalars as bosonic and fermionic
3 This x coordinate is inappropriate for the extremal PW flow, since in that case x is always vanishing. In this sense,
backgrounds with and without horizons (or equivalently physics at finite temperature and zero temperature) may have
distinctly different characteristics.
4 Non-extremal AdS geometry is obtained as a trivial solution for vanishing bosonic and fermionic deformations:
5
= = 0 identically.
76
deformations. We also explain the advantages of using the radial coordinate (2.10) in numerical
integration. Finally, we recall holographic renormalization of the thermal PW flows.
In what follows we find it convenient to introduce
c2 eA ,
A ln 3

1
ln 2x x 2 + a(x),
4
e .
(2.14)
The form of A(x) is chosen to extract the leading asymptotic behavior c2 x 1/4 from the
metric function c2 (x). In addition, the introduction of = e is natural, as the bosonic scalar
enters exponentially in the superpotential (2.3). Imposing the boundary conditions (2.12) and
(2.13), we see that the new warp factor a(x) and scalar (x) satisfy the boundary conditions
x 0+ : a(x) 0,
(x) 1,
3
x 1 : a(x) ln ,
3
(x) e1 ,
(2.15)
where 3 is a new constant.

2.1. Asymptotics of the thermal PW flow near the boundary
The most general solution of (2.11) subject to the boundary conditions (2.13) and (2.15) may
be expanded as a series around x = 0:

= 1 + x 1/2 (10 + 11 ln x) + x 20 + 21 ln x + 22 ln2 x +

k

k/2
i
+x
(2.16)
ki ln x + ,
i=1

= 0 x 1/4 1 + x 1/2 (10 + 11 ln x) + x 20 + 21 ln x + 22 ln2 x +

+x
k/2
k

ki ln x + ,
i
(2.17)
i=1

a = x 1/2 (10 + 11 ln x) + x 20 + 21 ln x + 22 ln2 x +

k

k/2
i
+x
ki ln x + .
(2.18)
i=1
This solution is characterized by five independent parameters

{ln 3 ; 11 , 10 ; 0 , 10 },
(2.19)
where we have included 3 from (2.14). The remaining series coefficients are completely determined from the above parameters.
For the holographic renormalization (see Section 2.3) we will need the coefficients of the first
two subleading terms in (2.16)(2.18). Explicitly, these are given by
77
3 2
22 = 11
,
2
26 2
11 ,
9 0
3 2
26
104 2
1
2
20 = 2411
(2.20)
+ 10
810 11 + 02 10
0 11 + 04 ,
2
9
9
3
1
11 = 02 ,
3
9 2
22 = 11 ,
(2.21)
2
13
51 2
,
21 = 202 11 + 910 11 + 04 11
36
2
1
9 2
243 2
51
13
131 4
20 = 602 11 + 10
+
11 10 + 202 10 + 02 10
,
8
2
4 11
2
12
120 0
(2.22)
a11 = 0,
1
a10 = 02 ,
(2.23)
9
1 2
,
a22 = 11
2
1 2
1
04 ,
a21 = 10 11 11
2
12
1
1 2
3 2
13 4 1 2
+
.
a20 = 10 11 0 10 11
(2.24)
2
4
4
648 0 2 10
As indicated above, the non-extremal AdS5 geometry is obtained by setting 0, 0, which
in asymptotic expansions (2.20)(2.24) corresponds to taking 11 = 10 = 0 = 0. This leads to

1/2

ds52 = (2T )2 2x x 2
(1 x)2 dt 2 + dx12 + dx22 + dx32
2
+
21 = 310 11 811
dx 2
,
(2x x 2 )2
(2.25)
where the temperature T is given by

2T = 3 .
(2.26)
Matching with asymptotic extremal AdS5 geometry with a standard Poincar patch radial coordinate R (and with our conventional choice for the radius of curvature L = 2)

dR 2
R2
dt 2 + dx12 + dx22 + dx32 + 4 2 ,
4
R
we identify near the boundary
ds52 =
x R 4 ,
x 0+ .
(2.27)
(2.28)
Given (2.28) and the asymptotic expansions (2.16)(2.18), we identify the conformal weight two
supergravity scalar (x) as dual to turning on mass for the bosonic components of the N = 2
hypermultiplet. Parameters 11 and 10 are coefficients of its non-normalizable and normalizable modes, correspondingly. Similarly, the conformal weight one supergravity scalar (x) can
78
be identified as dual to turning on mass for the fermionic components of the N = 2 hypermultiplet. Parameters 0 and 10 are coefficients of its non-normalizable and normalizable modes,
correspondingly. We discuss the precise relation of 11 and 0 to the N = 2 gauge theory
bosonic and fermionic masses in Section 3. Here, we would simply like to emphasize the fol 11 , 0 }, the coefficients of the normalizable modes of
lowing well-known fact [29]: given {,
the supergravity scalars (x) and (x), namely {10 , 10 }, are uniquely fixed by requiring that
the resulting supergravity RG flow is nonsingular in the bulk. This statement is simply the supergravity dual to a gauge theory lore: for a fixed temperature and mass parameters, the bosonic
and fermionic condensates are determined uniquely.5 Thus, for nonsingular finite temperature
PilchWarner flows, we must have
10 = 10 (3 , 11 , 0 ),
10 = 10 (3 , 11 , 0 ).
(2.29)
2.2. Asymptotics of the thermal PW flow near the horizon

We now consider the behavior of the solution near the horizon, x = 1. Notice that Eq. (2.11)
are invariant under the transformation
(1 x) (1 x).
(2.30)
It is the straightforward to verify that boundary conditions (2.12), i.e. the finiteness of
{a(x), (x), (x)} near the horizon, implies that {a(x), (x), (x)} are even functions of (1 x)
for |1 x| 1. It is this fact that justifies our choice of the radial coordinate x in (2.10). Indeed,
keeping fixed {3 , 11 , 0 }, for generic values of {10 , 10 }, the functions {a(x), (x), (x)}
would diverge at the horizon. Thus, fixing coefficients of the normalizable modes {10 , 10 }
from the requirement of non-singularity of the holographic RG flow is equivalent to imposing
Neumann boundary conditions on {a(x), (x), (x)} at the horizon, i.e. as x 1
lim = lim = lim a = 0.
x1
x1
x1
(2.31)
The boundary condition (2.31) would determine bosonic and fermionic condensate dependence
on the hypermultiplet masses and the temperature (2.29).
2.3. Holographic thermodynamics of the finite temperature PW flow
We use the method of holographic renormalization in order to examine the thermodynamics
of the N = 2 RG flow. Holographic renormalization of this flow was explained in [24], where
it was investigated using the original radial coordinate r of the metric ansatz (2.7). In particular,
parameterizing
c1 (r) = eA(r)+B(r) ,
c2 (r) = eA(r) ,
(2.32)
the results of [24] indicate that the entropy density s, the energy density E, and the free energy
density F are given by
s=
1
4G5
lim
rrhorizon
e3A ,
5 Strictly speaking this is true in the absence of moduli. However, we do not expect moduli at finite temperature, and
thus unbroken supersymmetry, on the gauge theory side.

1
4A+B
4A+B
E=
lim 3e
A + 2e
1 + 3 + 4 2 + 5 2
8G5 r

2
+ 6 2 + 8
+ ln
10 4 + 11 4 ,
ln
1
lim e4A+B B ,
F = E sT = E
8G5 r
79
(2.33)
where
gtt = eA+B ,
(2.34)
and the boundary counterterm coefficients i take the values

3
1 = ,
4
1
1
4 = ,
6 = 0,
2 = ,
3 = 0,
5 = 3,
4
2
3
1
2
1
8 = ,
9 = ,
10 = ,
11 = .
7 = 0,
(2.35)
2
3
3
6
Using the asymptotic expansions (2.16)(2.18), and changing the radial coordinate in (2.33)
following (2.10), we find

34
2
(24 96 ln 3 + 24 ln 2) 2410 11 + 202 10
1 + 11
F =
32G5

2
8
4 4
+ 0
(2.36)
ln 2 + ln 3 ,
9 3
3
1
1
sT =
E =F
(2.37)
4 ,
4 .
8G5 3
8G5 3
Finally, the entropy density is given by
s=
33 e3ah
,
4G5
ah = lim a(x),
x1
(2.38)
and thus we extract the temperature from (2.37)

3 3ah
.
e
(2.39)
2
Of course, the same value of the Hawking temperature (2.39) can be extracted from the nearhorizon geometry, provided one recalls that sT is a renormalization group flow invariant in the
supergravity black brane geometries without a chemical potential [30]. Indeed, from the relation
between components of the Ricci tensor
T=
Rx1 x1 = Rt t ,
(2.40)
constraint6
we have a

c1
c24
= const.
c2
6 Eq. (2.41) can also be directly derived from (2.8).
(2.41)
80
Evaluating the LHS of (2.41) near the horizon, we have

4 c1
= 8G5 sT .
lim c
rrhorizon 2 c2
(2.42)
Changing to the radial coordinate x in (2.10) and using the asymptotic expansions (2.16)(2.18),
the LHS of (2.41) near the boundary takes form7

c1
dx 4 4
lim c24
= lim
(2.43)
c = 3 .
r
x0+
c2
dr 2
Comparing (2.42) with (2.43) leads to the value of sT presented in (2.37). It is precisely the holographic RG invariance of the combination sT that guarantees the basic thermodynamic relation
F = E sT .
(2.44)
= 2
Relation (2.44) was verified explicitly for the N

thermal RG flow in [21]; the same relation
holds in other thermal gauge/gravity duals, for example in the thermal cascading RG flow [22].
3. Gauge theory versus supergravity parameters
The relation between N = 2 gauge theory and the supergravity parameters of the (thermal)
PilchWarner flow was established in [21,26]. For completeness, we review the main points here.
We begin with the gauge theory, then move to the supersymmetric PW flow [26], and finally
discuss the gauge/gravity parameter identification at finite temperature.
3.1. N = 2 gauge theory
In the language of four-dimensional N = 1 supersymmetry, the mass deformed N = 4 SU(N )
YangMills theory (N = 2 ) in R3,1 consists of a vector multiplet V , an adjoint chiral superfield
related by N = 2 supersymmetry to the gauge field, and two additional adjoint chiral mul which form an N = 2 hypermultiplet. In addition to the usual gauge-invariant
tiplets Q and Q
kinetic terms for these fields,8 the theory has additional interactions and a hypermultiplet mass
term given by the superpotential

2 2
m
W = 2 Tr [Q, Q]
(3.1)
+ 2 Tr Q2 + Tr Q 2 .
gYM
gYM
When m = 0, the gauge theory is superconformal with gYM characterizing an exactly marginal
deformation. The theory has a classical 3(N 1) complex dimensional moduli space, which is
protected by supersymmetry against (non)-perturbative quantum corrections.
When m = 0, the N = 4 supersymmetry is softly broken to N = 2. This mass deformation
hypermultiplet moduli directions, leaving the (N 1) complex dimensional
lifts the {Q, Q}
Coulomb branch of the N = 2 SU(N ) YangMills theory, parameterized by expectation values
of the adjoint scalar

= diag(a1 , a2 , . . . , aN ),
(3.2)
ai = 0,
i
7 In principle, one does not have to take a limit in (2.43) since this expression is a constant.
Q].
8 The classical Khler potential is normalized according to (2/g 2 ) Tr[
+Q
+ QQ
YM
81
in the Cartan subalgebra of the gauge group. For generic values of the moduli ai , the gauge
symmetry is broken to that of the Cartan subalgebra U (1)N1 , up to the permutation of individual
U (1) factors. Additionally, the superpotential (3.1) induces the RG flow of the gauge coupling.
While from the gauge theory perspective it is straightforward to study this N = 2 theory at any
point on the Coulomb branch [28], the PW supergravity flow [27] corresponds to a particular
Coulomb branch vacuum. More specifically, matching the probe computation in gauge theory
and the dual PW supergravity flow, it was argued in [26] that the appropriate Coulomb branch
vacuum corresponds to a linear distribution of the vevs (3.2) as
2 N
m2 gYM
,
with (continuous in the large N limit) linear number density
ai [a0 , a0 ],
a02 =

2
(a) =
a02 a 2 ,
2
m2 gYM
(3.3)
a0
da (a) = N.
(3.4)
a0
Unfortunately, the extension of the N = 2 gauge/gravity correspondence of [26,27,31] for vacua

other than (3.4) is not known.
In [26,31] the dynamics of the gauge theory on the D3-brane probe in the PW background
was studied in detail. It was shown in [26] that the probe has a one complex dimensional moduli
space, with bulk induced metric precisely equal to the metric on the appropriate one complex
dimensional submanifold of the SU(N + 1) N = 2 DonagiWitten theory Coulomb branch.
This one-dimensional submanifold is parameterized by the expectation value u of the U (1) complex scalar on the Coulomb branch of the theory where SU(N + 1) U (1) SU(N )PW . Here
the PW subscript denotes that the SU(N ) factor is in the PilchWarner vacuum (3.4). Whenever
u coincides with any of the ai of the PW vacuum, the moduli space metric diverges, signaling
the appearance of additional massless states. An identical divergence is observed [26,31] for the
probe D3-brane at the enhancon singularity of the PW background. Away from the singularity
locus, u = a [a0 , a0 ], the gauge theory computation of the probe moduli space metric is oneloop exact. This is due to the suppression of instanton corrections in the large N limit [26,32] of
N = 2 gauge theories.
Consider now N = 2 gauge theory at finite temperature T . Of course, finite temperature
completely breaks supersymmetry. Thus, we can generalize the thermal N = 2 gauge theory by
allowing for different (non-equal) masses mb and mf for the bosonic and fermionic components
correspondingly. It is only when mb = mf = m, and at
of the N = 2 hypermultiplet {Q, Q}
zero temperature T = 0, that we have N = 2 supersymmetry. Since turning on mass terms for
the bosonic or fermionic components of the hypermultiplet corresponds to deforming the N = 4
supersymmetric YangMills Lagrangian by relevant operators of different classical dimension (a
dimension two operator for bosonic and a dimension three operator for fermionic mass terms),
such deformations will be encoded in different supergravity modes. As indicated in Section 2,
turning on bosonic/fermionic masses corresponds to turning on the five-dimensional supergravity
scalars / correspondingly.
We would like to conclude this section with a simple observation. Turning on bosonic/fermionic masses for the hypermultiplet components sets a strong coupling scale max{mb , mf }. In
this case, we expect to find two qualitatively different phases of this gauge theory, depending on
whether T or T . In the former case we expect the thermodynamics to be qualitatively
(and quantitatively in the limit T / ) similar to that of the N = 4 gauge theory plasma [33].
82
On the other hand, when T , and with mf = 0, we actually expect an instability in the system.
Indeed,9 turning on only the supergravity scalar , i.e., setting mb = 0 and mf = 0, corresponds
to giving positive mass-squared to four out of six N = 4 scalars (these are the bosonic components of the N = 2 hypermultiplet). The remaining two N = 4 scalars at the same time obtain a
negative mass-squared [27]they are tachyons at zero temperature. At high enough temperature
the thermal corrections would stabilize these tachyons. However, as we lower the temperature,
we expect the re-emergence of these tachyons. As argued in [34], dynamical instabilities of
gravitational backgrounds dual to thermal systems are reflected in thermodynamic instabilities.
Furthermore, it was argued in general (and demonstrated explicitly with a concrete example)
[35] that thermodynamic instabilities are reflected to developing cs2 < 0, where cs is the speed
of sound waves in the thermal gauge theory plasma. We discuss the supergravity realization of
these phenomena in Section 6.
3.2. Supersymmetric PW flow
It is clear that the following conditions must hold in order to preserve N = 2 supersymmetry
for the mass deformation of N = 4 SYM:
the temperature must be zero: T = 0;
the masses for the bosonic and fermionic components of the N = 2 hypermultiplet {Q, Q}
must be the same: mb = mf = m.
The former condition corresponds to a restriction The former condition corresponds to a restriction
c1 (r) = c2 (r),
(3.5)
ensuring Lorentz invariance of the metric. In this case, the supergravity RG flow cannot be parameterized by the radial coordinate x introduced in (2.10). Instead, the supersymmetric flow is
easiest to parameterize in terms of the fermionic scalar [0, +), where = 0 corresponds
to the asymptotic AdS5 boundary, and + to the enhancon location in the bulk [26,31].
The supersymmetric PilchWarner solution is then given by [27]
eA =
k 2
,
sinh(2)
6 = cosh(2) + sinh2 (2) ln
sinh()
,
cosh()
(3.6)
where the single integration constant k is related to the hypermultiplet mass m according to [26]
k = mL = 2m.
(3.7)
In order to identify the thermal RG parameters {11 , 0 } with the gauge theory masses, we need
the boundary asymptotics of the PW flow. This was computed in [21]. Introducing
x er/2 ,
we have
9 We would like to thank Ofer Aharony for clarifying this.
(3.8)
83

10
20 2
1 4
7
4 4
= k x 1 + k x
+ ln(k x)
+ k x +
ln(k x)
+
ln (k x)
3 3
90
3
9

+ O k 6 x 6 ln3 (k x)
,

2
2 2
2 2 1
4 4 1
+ O k 6 x 6 ln3 (k x)
,
+ ln(k x)
+ k x
+ 2 ln(k x)
+ ln (k x)
= 1 + k x
3 3
18
3

10
4
2
1 2 2
4 4
2
+ O k 6 x 6 ln3 (k x)
. (3.9)
+
ln(k x)
+ ln (k x)
A = ln(2x)
k x k x
3
9
9
9

2 2
3.3. Thermal PW flow

We now consider the thermal PW flow with mb = mf = m. Here we would like to identify the
parameters 11 and 0 for such a flow in terms of the PW quantity k (and thus, following (3.7),
the hypermultiplet mass m) by comparing the boundary asymptotics (2.16)(2.18) with (3.9). In
order to do so, we first establish the relation between the radial coordinate x in (2.16)(2.18) and
the radial coordinate x in (3.9). Matching the asymptotic warp factors eA for the zero temperature
PW flow and the thermal PW flow as x 0+ , we find
x 834 x 4 ,
x 0+ .
(3.10)
x 1/2 ln x
for (x) and

Next, using (3.10) and matching the leading nontrivial asymptotics [order
order x 1/4 for (x)] for both the thermal PW solution and the supersymmetric PW flow, we find

2 6ah m 2
21/2 k 2
21/4 k
1
3ah m
=
e
,
0 =
= 3/4 e
,
11 =
(3.11)
T
T
24 2
2
2432
23
where we used (2.39). We emphasize that the above matching procedure is well-definedit was
shown in [21] that finite temperature effects modify the asymptotic supersymmetric PW geometry at order O(x) O(x 4 ), which is subleading compared to the order at which the matching
(3.11) is done.
Motivated by (3.11), we now propose that in the general case, i.e. for mb = mf , we may
independently extract bosonic and fermionic masses according to

2 6ah mb 2
1
3ah mf
e
,
0 = 3/4 e
.
11 =
(3.12)
T
T
24 2
2
In Section 4 we recall that such an identification leads to a consistent thermodynamics of the
thermal PW flows [24,25] at high temperatures. Furthermore, we later show in Section 6 that this
identification in fact leads to a consistent thermodynamics at any temperature.
In the rest of this subsection we derive some useful expressions for the free energy and the
speed of sound waves. Specifically, we would like to express the free energy and the speed of
sound in terms of the coefficients of the non-normalizable modes of the supergravity scalars
(x), (x). We also present the differential constraint on coefficients of the normalizable modes
of these supergravity scalars following from the first law of thermodynamics.
Since we will constantly refer to the parameters 11 and 0 related to the bosonic and fermionic masses, we distinguish them from derived quantities (such as the subleading coefficients)
by introducing the notation
11 ,
0 .
(3.13)
84
Notice that because of (3.12), and are not independent, but are related through the ratio of
bosonic and fermionic masses according to
2 = 6
m2f
m2b
(3.14)
The non-singularity of the thermal PW flow at the horizon (2.31) will determine the coefficients {10 , 10 } of the normalizable modes of the supergravity scalars {(x), (x)} and also the
value ah of the warp factor a(x) at the horizon in terms of (3.13) and (3.14). When mb = 0, we
take
10 = 10 ( ),
10 = 10 ( ),
ah = ah ( ).
(3.15)
On the other hand, for mb = 0, we have instead

10 = 10 (),
10 = 10 (),
ah = ah ().
(3.16)
Notice that a given pair {, } can be unambiguously related to bosonic and fermionic masses,
measured with respect to temperature. Indeed, following (3.12)

2
mf
mb
= 12 2 2 e6ah ,
= 23/4 e3ah .
(3.17)
T
T
Solving for 3 from (2.39) and expressing G5 in gauge theory variables (2.4), we find for the free
energy

m4f
m2f

1
F = 2 N 2 T 4 e12ah ( ) 1 + 24 2 ln 2 1 4 + 12 2 10 ( ) 2410 ( )
8
mb
mb
+ F0 ,
(3.18)
assuming mb = 0, and

1
2
F = 2 N 2 T 4 e12ah () 1 4 ln 4 + 22 10 () + F0 ,
8
3
(3.19)
if = 0. In the above expressions, F0 does not depend on temperature. Such a constant arises
because holographic renormalization generically10 leads to ln T dependence in the free energy,
which requires the introduction of an arbitrary regularization scale
T
.
(3.20)
Changing this scale modifies F0 , but otherwise has no effect on the thermodynamics. This
phenomena has been encountered previously; it was discussed in perturbative gauge theory thermodynamics in [36], and in the context of gauge/string theory correspondence in [37].11 Notice
that in (3.18) we wrote ln( 2 ) instead of 2 ln( ). The reason is that the former expression allows us to study thermodynamics for < 0, which according to (3.12) we interpret as m2b < 0.
While introducing tachyonic masses for gauge theory scalars at zero temperature leads to instability, this is not the case at finite temperature. Here, the effective mass squared receives thermal
corrections of order T 2 , which might cure the zero temperature instability.
ln T ln
10 Notice that this is not the case when m = m .

b
f
11 It appears even in the high temperature thermodynamics of the thermal PW flow [25] (see also Section 4).
85
Besides the free energy, we can also evaluate the entropy density and the energy density
1
s = 2 N 2 T 3 e12ah ,
(3.21)
E = F + sT .
2
In addition, from the defining equations (3.17), we can evaluate the quantities d/dT and
d/dT :
1

m2
d
dah ( )
e6ah ( ) 6
, mb = 0;
= b
+1
dT
d
6 2 2 T 3
1

mf
dah ()
d
3ah ()
, mb = 0.
= 3/4 2 e
+1
3
(3.22)
dT
d
2 T
Using (3.18) or (3.19), (3.21) and (3.22) the first law of thermodynamics,
dF = s dT ,
(3.23)
quantities12
reduces to a differential constraint on the

, 10 ( ), 10 ( ), ah ( ) , for mb = 0;

, 10 (), 10 (), ah () , for mb = 0.
(3.24)
In particular, the first law in differential form reads

m2f
m4f
d10 ( )
d10 ( )
2 10 ( )
0 = 4 1 4 + 2 10 ( )
d
d
mb
mb
dah ( )
,
+
d
(3.25)
for mb = 0, and
4
d10 ()
dah ()
0 = 3 + 210 () 2
6
,
3
d
d
(3.26)
for mb = 0. Using either (3.25) or (3.26), we can evaluate the speed of sound entirely as a function
of or . Straightforward algebraic manipulations result in
cs2 =
P
F
1+
=
=
,
E
E
3
(3.27)
where

d10 ( )
1 + 12
10 ( )
( ) = 24
d
m4b

m2f
d10 ( )
+ 6 2 10 ( )
, mb = 0;
d
mb

m4f
1 d10 ()
2
,
() = 4 + 2 10 () 3
3
2
d
mb = 0,
(3.28)
12 Notice that even though, for = 0, the free energy density (3.19) does not depend on , one generically expects
10
that 10 = 10 ().
86
and where P = F is the pressure.

In general, one may explore the PW thermodynamics in the entire phase plane of mb mf .
However, here we focus on two special cases:
when mf = 0 (we call this the bosonic case)13 we have

2 N 2 T 4 12ah ( )
1 + 24 2 ln 2 2410 ( ) + F0 ,
e
8
d10 ( )
bosonic ( ) = 24 2 1210 ( ) + 12 2
,
d
Fbosonic =
and the first law of thermodynamics takes form

d10 ( )
dah ( )
0 = 4 + 2 10 ( )
+
;
d
d
(3.29)
(3.30)
(3.31)
when mf = mb m (we call this the supersymmetric case), or correspondingly from (3.14)
2 = 6,
(3.32)
we have

2 N 2 T 4 12ah ( )
1 2410 ( ) + 12 10 ( ) + F0 ,
e
8

d10 ( )
d10 ( )
susy ( ) = 12
10 ( ) + 6 10 ( )
,
d
d
Fsusy =
and the first law of thermodynamics takes form

d10 ( )
d10 ( )
dah ( )
0 = 2 10 ( )
10 ( )
+
.
d
d
d
(3.33)
(3.34)
(3.35)
Notice that in the supersymmetric case there is no ln( 2 ) dependence in the holographic free
energy (3.18). As a result one can unambiguously determine F0 directly from (2.33).14 We find
susy
F0 F0
2 N 2 T 4 12ah ( )
5N 2 4
40 2 =
m .
e
8
288 2
(3.36)
4. Thermodynamics of the high temperature PilchWarner flow

We now recall that the differential equations (2.11) describing the finite temperature PW
renormalization group flow admit a perturbative analytical solution at high temperature [21,24,
25]. In this regime, the appropriate expansion parameters are
2
mf
mb
1,
2
1.
1
(4.1)
T
T
13 Recall that this corresponds to a consistent truncation of the supergravity.
14 The latter holographic prescription was formulated in such a way that the holographic energy of the supersymmetric
PW flow vanishes [24].
87
In this case, the metric function A(x) is given to first nontrivial order in 1 , 2 by

1
A(x) = ln 3 ln 2x x 2 + 12 A1 (x) + 22 A2 (x),
4
(x) = 1 1 (x),
(x) = 2 2 (x),
(4.2)
where
1/2
1 1

2
1 = 2x x 2
2 F1 2 , 2 ; 1; (1 x) ,
3/4
3 3

2
2 = 2x x 2
2 F1 4 , 4 ; 1; (1 x) ,
1
A1 = 4
x
4
A2 =
3
1
x
(4.3)
(4.4)

1
(z 1) dz
1 2 (2y y 2 )2
,
1 dy
y
y 1
(2z z2 )2

(z 1) dz
2
(2z z2 )2
1
z
2
dy
y
2

(2y y 2 )2
.
y1
(4.5)
The constants i are fine-tuned to satisfy the boundary conditions, and are given by
8 2
8 3
,
2 =
.
8
2 2
Comparing (4.3), (4.4) and (2.16), (2.17) we find the relations
2
3 2 ln 2
10 =
1 ,
1 ,
11 =
21/2 ( 34 )4
21/4
0 =
.
2
10
2
( 34 )2
1 =
Also notice that
(4.6)
(4.7)

d 2 A2
d 2 A1
2
2 1
+ 2
2T = 3 1 + 1 2
2
dx 2 x=1
dx 2 x=1

4 2
16
= 3 1 + 2 12 +
(4.8)
.
3 2

The parameters {1 , 2 } can be related to the mb /T , mf /T parameters of the dual gauge theory
via
2
[( 34 )]2 mf
1
mb
1 =
(4.9)
,
2 =
.
24 T
2 3/2 T
To leading order in mass deformation, the free energy (3.18) takes form [24,25]

8
192
1
F = 2 N 2 T 4 1 2 ln(T ) 12 22 .
(4.10)
8
One can further evaluate the entropy density of the non-extremal PW geometry,

4
48
1
s = 2 N 2 T 3 1 2 12 22 ,
(4.11)
2
and verify that the first law of thermodynamics (3.23) is satisfied.
88
5. Numerical procedure and results

We are mainly interested in the thermodynamics of the PW flow away from the high temperature limit. In this case, since analytic solutions to the equations of motion (2.11) are unavailable,
we resort to numerical integration. In this section we describe the numerical procedure and some
of the consistency checks we have performed on the numerical results. Extracting the actual
thermodynamics from these solutions will be taken up in the following section.
The numerical procedure is conceptually the same for both the bosonic and supersymmetric
mass deformations of the non-extremal AdS5 geometry. In practice, however, it is much easier to
implement for the bosonic deformation
mf = 0,
mb = 0,
(5.1)
and so we will describe the details of this first. We then outline the modifications required for
handling the supersymmetric mass deformation
mf = mb m,
(5.2)
and comment on thermal flows corresponding to generic mass deformations. Finally, we discuss
the validity of the supergravity approximation.
5.1. Bosonic deformation
Simplicity of treating the bosonic mass deformation stems from the fact that thermal PW
flows allow for a consistent truncation with the supergravity scalar (x) identically set to zero.
The relevant equations can thus be obtained from (2.11) with
(x) 0,
ln ,
x 0+ ,
a(x) 0,

1
ln c2 (x) A(x) ln 3 ln 2x x 2 + a(x).
(5.3)
4
The boundary conditions (which guarantee that the flow is singularity-free) are given by (2.15)
and (2.31)
x 1 ,
a (x) 0,
(x) 1,
(x) 0.
(5.4)
Asymptotic expansions for {(x), a(x)} as x 0+ are given by (2.16), (2.18). The first couple
terms are presented in (2.20)(2.24), where one has to set 0 = 0 corresponding to (x) 0, see
(2.17). Notice that numerical integration of (x) and a(x) from x = 0+ is uniquely determined
specifying the non-normalizable 11 and normalizable 10 mode coefficients. However, a generic
choice of a pair {11 , 10 } would produce a solution that does not satisfy the horizon boundary
condition (the x 1 boundary condition in (5.4)). Solving SturmLiouville problem with
boundary conditions (5.4) would determine
10 10 (11 ) 10 10 ( ),
(5.5)
where on the RHS we used conventions of Section 3.3. For a flow satisfying (5.5) we can extract
ah ( ) = lim a(x).
x1
(5.6)
89
To summarize, each non-singular bosonic deformation flow would generate a triplet of numbers

, 10 ( ), ah ( ) ,
(5.7)
2
(3.27)
which can be used to evaluate the free energy Fbosonic (3.29), the speed of sound cs,bosonic
and (3.30), and verify (numerically) the first law of thermodynamics (3.31).
We use the following algorithm to generate triplets (5.7). First, we choose a mass deformation
parameter 11 = which we keep fixed during the iteration procedure. The iteration starts by
choosing a trial value of 10 . Given {11 , 10 } we integrate numerically15 (2.11) (with (5.3))
from xinitial 1 to xfinal (such that (1 xfinal ) 1) using the power series (2.16), (2.17) to
specify16
(xinitial ),
(xinitial ),
a(xinitial ),
a (xinitial ).
(5.8)
For a generic value of 10 ,

lim (x) ,
x1
lim (x) .
x1
(5.9)
We find that for a given value 11 = there is (generically) a unique 10 = 10 ( ), such that
(x) and a(x) are finite as we go to the horizon, x 1 , i.e.,
lim

d
x; 11 = , 10 = 10 ( ) = 0,
dx
(5.10)
lim

d
x; 11 = , 10 = 10 ( ) = .
dx
(5.11)
x1
while
x1
Now, the iterative algorithm is clear: we use Newtons shooting method of determining 10 ( )
for a fixed 11 = .
Results of the numerical analysis are presented in Figs. 13. From Fig. 1 notice that for 11 >
crit 0.03582 there is no solution to SturmLiouville problem with boundary condition (5.4).
Using (3.17) we find that corresponding to crit ,

mb
2.29(9).
(5.12)
T crit
We find that for each value of (0, crit ) there are two sets {10 ( ), ah ( )} which are represented by the plots in Fig. 1. We use a superscript + or to denote the larger or the smaller value
of 10 in this interval:
+
( ) > 10
( ),
10
(0, crit ).
(5.13)
For each < 0 we find a single set {10 ( ), ah ( )}, see Fig. 1. In Section 6 we show that for a
given value of , different values of the bosonic condensate 10 ( ) correspond to different phases
of the N = 2 plasma. We also show that there is = < crit such that for all > , N = 2
plasma becomes unstable both thermodynamically and dynamically, in agreement with general
arguments of [35].
15 We used Mathematica for numerical integration as well as our own C code based on fifth-order RungeKutta. Both
procedures (up to controllable numerical errors) produced equivalent results.
16 In practice we truncated the power series to k = 3 terms (inclusive), and used x
9
6
initial = 10 , xfinal = 110 .
90
Fig. 1. Altogether we obtained 3227 triplets {, 10 ( ), ah ( )}. The left plot represents 10 ( ), and the right
plot represents ah ( ). The critical value of is at crit 0.03582(2). For (0, crit ) there are two values of
+
( ), 10
( )}, where 10
( ) > 10
( ).
10 ( ): {10
( ), a ( ), and a numerical verification of the first law of thermodyFigs. 23 represent 10
h
namics (3.31). Since our numerical data is rather densely spaced17 in , we use
f ( ) =
df ( ) f ( + ) f ( )
=
,
d
2
(5.14)
as a numerical approximation for a derivative of a function f ( ). Notice that

d10 ( )
= ,
crit 0
d
lim
dah ( )
= ,
crit 0 d
lim
(5.15)
/10
correspondingly. Also, from Fig. 3
with minus/plus signs for 10
d10 ( )
= +,
+0
d
lim
dah ( )
= +.
+0 d
lim
(5.16)
We find that

( ) =+2104 2.10(6),
10
(5.17)
which is rather close to a predicted value from the high temperature analysis of Section 4 (see
(4.7))
high temperature

10
(5.18)
( ) =0
= 3 ln 2 2.07944.
+
A better agreement is achieved if we fit the first 200 points of 10
( ) on the left plot in Fig. 1
with a 10th order polynomial in . We find in this case
fit

( ) =0 2.07972.
10
(5.19)
The right plots in Figs. 23 demonstrate the cancellation of ah ( ) with a combination

d10 ( )
4 + 2 10 ( )
,
d
17 Mostly we have = 104 .
91
( ) for < 0 and ( + ( )) for (0, ). The right plot represents corresponding
Fig. 2. The left plot represents 10
crit
10
values of ah ( ). The right plot also includes numerical verification of the first law of the thermodynamics, where ah ( )
must be canceled with an appropriate combination of {, 10 ( ), ( )}, see (3.31).
Fig. 3. The left plot represents (10

( )) . The right plot represents corresponding values of ah ( ). The right plot also
includes numerical verification of the first law of the thermodynamics, where ah ( ) must be canceled with an appropriate
combination of {, 10 ( ), ( )}, see (3.31).
as required by the first law of thermodynamics, see (3.31). We find that such a cancellation is
achieved with an accuracy of better than 104 . Clearly, this provides a rather impressive check
on both the validity of the holographic renormalization of the thermal PW flows explained in
[24], and the numerical procedures developed in this paper.
5.2. Supersymmetric deformation
Albeit more complicated, the SturmLiouville problem for the supersymmetric mass deformation is conceptually similar to the one for the bosonic mass deformation studied in previous
subsection. Here we have to use all equations in (2.11). As for the bosonic deformation, we
rewrite these equations using {(x), a(x)} such that

1
(5.20)
ln 2x x 2 + a(x).
4
The boundary conditions (which guarantee that the flow is singularity-free) are given by (2.15)
and (2.31)
ln ,
ln c2 (x) A(x) ln 3
x 0+ ,
a(x) 0,
(x) 1,
(x) 0,
92
Fig. 4. Altogether we obtained 2712 sets {, 10 ( ), ( ), ah ( )}. The left plot represents 10 ( ), and the right plot
represents 10 ( ).
x 1 ,
a (x) 0,
(x) 0,
(x) 0.
(5.21)
For the bosonic mass deformation we have a one-dimensional SturmLiouville problem: given
11 , the boundary condition at the horizon determines 10 ; the pair {11 , 10 ( )} further
determines the value of the a(x) at the horizon, ah ( ). For the supersymmetric mass deformation
the SturmLiouville problem is two-dimensional: given 11 , the boundary condition (5.20)
would determine
10 10 ( ),
10 10 ( ).
(5.22)
For a flow satisfying (5.22) we can extract

ah ( ) = lim a(x).
x1
(5.23)
Thus, each non-singular supersymmetric mass deformation flow would generate a set of four
numbers

, 10 ( ), 10 ( ), ah ( ) ,
(5.24)
2
which can be used to evaluate the free energy Fsusy (3.33), the speed of sound cs,susy
(3.27) and
(3.34), and verify (numerically) the first law of thermodynamics (3.35).
Since the SturmLiouville problem is technically more complex here, we restrict our discussion to 0, or following (3.12) m2 0. Results of the numerical analysis are presented in
Figs. 45. Fitting the first 200 points of 10 ( ) and 10 ( ) with a 10th order polynomial in we
find
fit
fit

10
(5.25)
( ) =0 = 2.069(2),
10 =0 = 0.3219(7),
which should be compared with the high temperature predictions18 (4.7)

high temperature

10
( ) =0
= 3 ln 2 2.07944,
18 We reduce the accuracy compare to the bosonic mass deformation case in order to speed-up computations. For a
selected set of 41 sets (5.24) we improve accuracy by roughly a factor of 102 . More accurate data does not provide
noticeable quantitative difference.
93
Fig. 5. The left plot represents ah ( ) for the supersymmetric mass deformation. The right plot represents h ( ),
( )) and ( ( ) ( )) sets of points along with numerical verification of the first law of the
2(10 10
10
10
thermodynamics, see (3.35).
high temperature
21/2 ( 34 )4
10 ( ) =0
=
0.3231(1).
(5.26)
2
We do not find a signature for the existence of crit it appears for supersymmetric deformation
the boundary value problem (5.21) always have a solution.19
Fig. 5 represent ah ( ) (the left plot) and a numerical verification of the first law of thermodynamics (3.35) (the right plot). We find that (3.35) holds with an accuracy of 103 .
5.3. Generic deformation
Though in this paper we study two special cases of the thermal PW flowsone corresponding to a bosonic mass deformation where the supergravity scalar (x) 0, and the other one
corresponding to a supersymmetric mass deformation where the ratio of the coefficients of the
non-normalizable modes of the supergravity scalars {(x), (x)} is the same as for the supersymmetric PW flowthe numerical methods developed here can be applied for a generic mass
deformation as well. In fact, the SturmLiouville problem one has to solve in a generic case
is exactly the same as in the supersymmetric case where the boundary conditions are given by
(5.21). The only difference from the supersymmetric mass deformation case is the (fixed) relation
between the coefficients of the non-normalizable modes of the supergravity scalars {(x), (x)}
given by (3.14)
2 = 6
m2f
m2b
Recall (see Fig. 1) that when mf = 0 there is a critical value crit , such that for > crit there
is no nonsingular thermal PW flow. On the other hand (see Fig. 4), we do not find the evidence
for crit for the supersymmetric mass deformation mf = mb . Thus we expect that

m2f
crit crit
(5.27)
2 ,
mb
such that crit as
m2f
m2b
1. As we show in Section 6.1 the existence of crit for
mf = 0 implies destabilization of the thermal plasma at low temperatures. What we appear to

19 Recall for a bosonic deformation
crit 0.0358(2) which is well inside the set of considered in Fig. 4.
94
be finding is that this particular instability is cured by turning on mf = 0. Verifying the latter
conjecture is beyond the scope of this paper. However, we would like to point out a different type
of instability in a system closely related to the one studies here which is eliminated by turning on
sufficiently large mf . Consider thermal PW flow with the flat R 3 in (2.7) compactified on a three
torus20 T 3 . This represents holographic dual to thermal N = 2 plasma on finite volume T 3 .
Analysis identical to that in [38] shows that this system is generically non-perturbatively unstable
with respect to a spontaneous creation of the D3D3 brane pairs.21 This instability is suppressed
provided22
02 611 0 2 6 0
m2f
m2b
1.
(5.28)
So, for the non-perturbative instability due to D3D3 spontaneous brane pair creation for the
thermal PW flow compactified on T 3 we find that = 1.
5.4. Validity of the supergravity approximation
Our analysis of the holographic dual to thermal N = 2 gauge theory plasmas are done in the
supergravity approximation to type IIb string theory. In the planar limit considered here the string
loop corrections are suppressed and the supergravity approximation is valid provided curvature
invariants of the thermal PW flow geometry (2.7) are small. We find that along the flow from the
boundary to the horizon the absolute value of the Ricci scalar of the five-dimensional thermal
PW geometry (2.7) increases. As a quantitative criteria for such an increase we propose the ratio
of the scalar curvature at the horizon to the scalar curvature near the boundary
P horizon
4
R horizon
=
= P horizon ,
(5.29)
3
R boundary P boundary
where we used (2.5) and the fact that P boundary = 34 . Values23 of ( ) for both bosonic and
phase
supersymmetric mass deformations are presented in Fig. 6. Notice that for the 10
( ) = +,
lim bosonic
0+ 10
(5.30)
phase of the bosonic mass deformation for small

and thus supergravity approximation for the 10
values of is not valid.
6. N = 2 plasma at finite temperature

In this section we discuss thermodynamics of the N = 2 gauge theory plasma for two special
cases:
bosonic deformation, when mf = 0 and mb = 0;
supersymmetric deformation, when mf = mb = m.
20 Since a three torus is locally the same as R 3 , such a compactification does not change any of the above computations.
21 Similar instability was first identified in [39].
22 See Eq. (4.16) of [38] where one has to remove (2) from the inequality (coming from the curvature of the 3-space)
and replace 11 of [38] with 411 due to a different choice of a radial coordinate.
23 In order to evaluate P horizon besides a ( ) we retained horizon values of the supergravity scalars {(x), (x)} while
h
solving the SturmLiouville problem (5.4) or (5.21).
95
Fig. 6. The left plot represents ratio of thermal PW geometry horizon curvature to its curvature at the boundary for
bosonic mass deformation. The right plot represents this ratio for the supersymmetric mass deformation.
6.1. Thermal PW flows with mf = 0, mb = 0

As explained in Section 5.1, for the bosonic deformation we observe two phases of the thermal
gauge theory plasma for mass deformation parameter (0, crit ) characterized by two different
+
values for the bosonic condensate 10 ( ): the 10

phase and the 10
phase. For large values
of the bosonic deformation parameter ( > crit in Fig. 1) thermal N = 2 gauge theory plasma
with bosonic deformation discussed does not exist. The set of points in Fig. 1 for < 0, following (3.17), corresponds to m2b < 0. As we argued in Section 5.4, we can trust supergravity
+
( ) points (everywhere); for < 0 points we can trust the supergravapproximation for in the 10
ity approximation over the range we studied (corresponding to m2b /T 2 < 15). The supergravity
( ) points is much less reliable, especially for small values of .

approximation for the 10
We use (3.17), (3.27), (3.29) and (3.30) to evaluate the free energy density and the speed of
sound for the bosonic mass deformation. For the free energy density we set an overall (arbitrary)
constant part of the free energy24
F0bosonic = 0.
(6.1)
susy
F0
Notice that while in the supersymmetric case one can determine

unambiguously (see
(3.36)) by requiring the Fsusy (T = 0) = 0, we can not do so for the bosonic deformation: as
we describe shortly, the bosonic deformation becomes unstable both thermodynamically and dynamically at low temperatures. Results of the analysis are presented in Figs. 78.
The left plot in Fig. 7 describes the free energy density for the bosonic deformation Fbosonic
relative (up to a sign) to the free energy density of the conformal gauge theory plasma
1
FCFT = 2 N 2 T 4 ,
8
as a function of
m2b
.
T2
(6.2)
(6.3)
24 Such ambiguity is present both in thermal quantum field theory and in the framework of the holographic renormalization of the string theory duals to thermal gauge theories.
96
Fig. 7. The left plot represents bosonic mass deformation free energy Fbosonic as a function of m2b /T 2 . The critical
2
for bosonic mass deformation as a
mass is at crit 5.28(5). The right plot represents speed of sound squared cs,bosonic
2
= 1/3.
function of . Recall that the speed of sound in the conformal plasma cs,CFT
2
Fig. 8. More detailed plot of the speed of sound squared cs,bosonic
for bosonic mass deformation as a function of . The
speed of sound vanishes at 5.40(9). Notice that though > crit , corresponding to value of 0.035(1) is
2
= 1/3.
less than crit 0.03582(2), see Fig. 1. Recall that the speed of sound in the conformal plasma cs,CFT
+
Notice that the 10
phase has always a larger (or equal) free energy than the 10
phase
Fbosonic, + () Fbosonic, (),

10
10
(6.4)
with equality achieved only at = crit corresponding to crit according to (3.17), (6.3). For the
+
phase
10
Fbosonic, + ( = 0)
10
1 2 2 4
8 N T
= 1,
(6.5)
in agreement with the fact that for mb = 0 (or at very high temperatures T ) the free energy
density must be that of the conformal gauge theory.
The right plot in Fig. 7 and a more detailed plot in Fig. 8 present the speed of sound squared
for the thermal N = 2 gauge theory plasma, relative to the speed of sound in the conformal
gauge theory plasma
1
2
= ,
cs,CFT
3
(6.6)
97
for bosonic deformation as a function of . Although bosonic deformation phase discussed here
does not exist for > crit , bosonic phase actually extends to values of crit < (see
Fig. 8). The reason for this is that relation between and as given by (3.17) involves a factor
e6ah ( ) and is not monotonic around crit . A value of corresponding to is actually less
than crit :
5.40(9) > crit 5.28(5) 0.035(1) < crit 0.03582(2).
(6.7)
10
phase is superluminal, which is yet another indication

We find that the speed of sound for the
that this phase is unphysical. The speed of sound diverges for the 10
phase as one approach crit
+
either from above or below. The speed of sound for the 10 phase at = 0 is that as in the CFT
+
plasma; it vanishes at = and becomes purely imaginary for < < crit (still for the 10
phase and well within supergravity approximation)
1
2
2
,
cs,bosonic,
cs,bosonic,
+ ( = 0) =
+ ( = ) = 0,
3
10
10
2 6a ( )

2
h
< 0, < < crit .
cs,bosonic,
+ 12 2 e
(6.8)
10
We would like to stress that conclusions (6.8) are very robust and in fact, immediately follow
from the existence of crit for the bosonic condensate 10 ( ). Indeed, existence of crit , i.e.,
impossibility to determine 10 ( ) for > crit , along with the assumption for the smoothness of
10 ( ), implies that 10 ( ) is not a single valued function of around crit ; moreover,
lim
crit 0

10
( ) = ,
(6.9)
with the minus sign in the 10

phase and the plus sign in the 10
phase (precisely what is presented in Fig. 1). But then from (3.30)
lim
crit 0
bosonic ( ) = ,
(6.10)
again with the minus sign in the 10

phase and the plus sign in the 10
phase. Thus using (3.27)
+
we see that in the 10 phase

2
lim 3 cs,bosonic
(6.11)
( ) = 3
crit 0
(in agreement with Figs. 78), and for the 10

phase

2
lim 3 cs,bosonic
( ) = ,
bosonic, ( ) = 3,
0
10
(6.12)
+
where 0 < < crit . Since in the 10
phase at = 0 the speed of sound squared is 13 > 0 and
given (6.11) there is a guaranteed value of < crit at which the speed of sound vanishes, and for
< < crit the speed of sound is purely imaginary. As emphasized in [35], a thermal system
with cs2 < 0 is unstable simultaneously thermodynamically (it has a negative specific heat) and
dynamically (amplitude of small pressure/energy density fluctuations exponentially grows).
Finally, we do not see thermodynamic instability in thermal N = 2 plasma with bosonic
mass deformation for m2b < 0, see Fig. 7. In fact, we find that for negative values of the speed
of sound slowly (and monotonically) decreasesit reaches about 90% of the conformal plasma
98
speed of sound (6.6) at25 35. Of course, this does not exclude the fact that thermal N = 2
plasma with m2b < 0 might have dynamical instability without accompanying thermodynamics
instability [35].
6.2. Thermal PW flows with mf = mb m
Bosonic and fermionic condensates {10 ( ), 10 ( )} for thermal N = 2 gauge theory plasma
with supersymmetric mass deformation are presented in Fig. 4. As we argued in Section 5.4,
supergravity approximation here is valid of the whole range of deformation parameter considered. We use (3.17), (3.27), (3.33), (3.34) and (3.36) to evaluate the free energy density and
the speed of sound for the supersymmetric mass deformations. Results of the analysis are presented in Figs. 910. We do not see any signature of the phase transition,26 or the thermodynamic
instability.
Fig. 9. The left plot represents supersymmetric mass deformation free energy Fsusy as a function of m
T . The right plot
represents ln(fsusy ) as a function of m
T and a straight line fit (dashed curve) for the last 2500 points of the distribution.
2
Fig. 10. The speed of sound squared cs,susy
for supersymmetric mass deformation as a function of m
T .
25 This is outside the range of the right plot in Fig. 7.

m
m
26 This would show up as the free energy F
susy ( T ) changing sign for a particular value of T .
99
Probably the most unexpected result of the analysis is that strongly coupled thermal N = 2
plasma for supersymmetric mass deformation behaves much closer to a conformal plasma than
one would naively expect. The right plot on Fig. 9 presents the logarithm of the ratio of the
N = 2 plasma free energy density to the free energy density of the conformal plasma (6.2)
ln(fsusy ) ln
Fsusy
,
FCFT
(6.13)
as a function of m
T . We did a straight line fit (dashed line on the right plot on Fig. 9) for the low
temperature27
FIT
ln(fsusy ) (x) = 0.118(8) 0.138(1)x.
(6.14)
Thus, to a good approximation the free energy of the thermal N = 2 plasma can be approximated by
approx
Fsusy
1.1(3) FCFT e 7.2(4)T T 4 e 7T ,

m
(6.15)
so that noticeable deviation from conformality occurs at T 17 m. Corresponding to (6.15) we

find that the speed of sound vanishes28 as m
T 1
T
(6.16)
.
m
It is instructing to compare (6.15) with the zero coupling result for the N = 2 plasma. At
zero coupling we have29 N 2 massless N = 2 vector multiples and N 2 massive N = 2 hypermultiplets. The free energy density of the massless states is just 23 FCFT in (6.2), while the free
energy density of a mass m spin s particle is given by [40,41]

(2s + 1)m2 T 2 +1

m
s
K2
,
F =
(6.17)
T
2 2
2
2,approx
csusy
7.2(4)
=1
where = 1 for bosons/fermions correspondingly. As low temperatures

asymptotic expansion for the Bessel function K2 (x)

x
K2 (x)
e , x 1,
2x
m
T
1 we can use
(6.18)
to conclude that
Fs
(2s + 1)m3/2 T 5/2 m

e T,
(2)3/2
m
1.
T
(6.19)
Altogether we find that the free energy of the thermal N = 2 plasma for supersymmetric mass
deformation at zero coupling is

2N 2T 4
24 2 m 3/2 m
m
T
e
1 + 7/2
1.
,
FN =2 =
(6.20)
12
T
T
27 In practice, we define low temperatures as the subset of the mass deformation parameters m > 1.
T
28 This is true in the supergravity approximation only. At vanishingly low temperatures the supergravity approximation
is not valid, and thus we can not determine whether the speed of sound is actually zero at T = 0.
29 We neglect (1) in the large N limit.
100
From (6.20), we find that the free energy density is roughly 60% of the high temperature result
at T m, while the corresponding number for the strongly coupled plasma (6.14) is about 98%.
The hydrodynamic properties and the jet quenching of strongly coupled non-conformal gauge
theory plasma are typically parameterized in terms of [25,42]

1
cs2 .
(6.21)
3
For the strongly coupled N = 2 thermal plasma at30 T 1.5m we find (see Fig. 10)

N =2 T 1.5m 2 103 ,
(6.22)
which further emphasizes the striking similarity between strongly coupled N = 2 plasma and
conformal N = 4 gauge theory plasma.
7. Conclusion and future directions
In this paper we presented detailed analysis of the thermodynamics of the strongly coupled
N = 2 gauge theory plasma. We considered two special cases of the mass deformations: bosonic
mass deformation (where only the bosonic components of the N = 2 hypermultiplet get nonzero
mass) and supersymmetric mass deformation (with bosonic and fermionic components of the
hypermultiplet getting the same mass). We argued that both our theoretical framework and the
numerical procedures are under control, given a highly nontrivial consistency check of the first
law of thermodynamics for the extracted data. We also argued that the supergravity approximation is under control, as the curvature invariants of the dual geometry remain small.
The most important result of the analysis is the striking similarity between thermodynamics
of the mass deformed N = 2 gauge theory plasma at temperatures of order the mass scale and
the thermodynamics of the conformal N = 4 SYM plasma. We would like to reemphasize the
latter fact by presenting the plots for the energy for various mass deformations, see Fig. 11.
We could like to concluded with some open problems.
m
Fig. 11. The left plot represents bosonic mass deformation energy density Ebosonic as a function of Tb . The value
mb
(see Fig. 8) denotes the ratio T for which the speed of sound vanishes. The right plot represents supersymmetric mass
deformation energy density Esusy as a function of m
T .
30 The temperature of the quarkgluon plasma at RHIC is of order the QCD deconfinement temperature [43].
101
In is important to generalize our analysis to other examples of the strongly coupled

non-conformal gauge theory plasma. Most notably the cascading gauge theory plasma
[1820,22].
It is interesting to evaluate hydrodynamic properties and jet quenching in N = 2 gauge
theory plasma in the low temperature regime.
The speed of sound waves in N = 2 gauge theory plasma for the bosonic mass deformation
was found to vanish for certain value of mTb . It is interesting to study hydrodynamics of the
gauge theory plasma in the regime of the vanishing speed of sound and determine the critical
exponent of the bulk viscosity.
Acknowledgements
We would like to thank Jaume Gomis, Chris Herzog, Pavel Kovtun, Gerry McKeon, Chris
Pagnutti and Andrei Starinets for interesting discussions. We would like to thank Ofer Aharony
and Rob Myers for valuable comments on the manuscript. A.B.s research at Perimeter Institute is supported in part by the Government of Canada through NSERC and by the Province of
Ontario through MEDT. A.B. gratefully acknowledges further support by an NSERC Discovery grant. J.T.L.s research was supported in part by the US Department of Energy under grant
DE-FG02-95ER40899 and the National Science Foundation under grant PHY99-07949. J.T.L.
acknowledges the hospitality of the KITP, where this work was initiated.
References
[1] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
[2] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity, Phys.
Rep. 323 (2000) 183, hep-th/9905111.
[3] H. Liu, K. Rajagopal, U.A. Wiedemann, Wilson loops in heavy ion collisions and their calculation in AdS/CFT,
hep-ph/0612168.
[4] K. Adcox, et al., PHENIX Collaboration, Formation of dense partonic matter in relativistic nucleus collisions at
RHIC: Experimental evaluation by the PHENIX Collaboration, Nucl. Phys. A 757 (2005) 184, nucl-ex/0410003.
[5] B.B. Back, et al., The PHOBOS perspective on discoveries at RHIC, Nucl. Phys. A 757 (2005) 28, nucl-ex/0410022.
[6] I. Arsene, et al., BRAHMS Collaboration, Quarkgluon plasma and color glass condensate at RHIC? The perspective from the BRAHMS experiment, Nucl. Phys. A 757 (2005) 1, nucl-ex/0410020.
[7] J. Adams, et al., STAR Collaboration, Experimental and theoretical challenges in the search for the quarkgluon
plasma: The STAR Collaborations critical assessment of the evidence from RHIC collisions, Nucl. Phys. A 757
(2005) 102, nucl-ex/0501009.
[8] S.S. Gubser, I.R. Klebanov, A.W. Peet, Phys. Rev. D 54 (1996) 3915, hep-th/9602135.
[9] 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.
[10] G. Policastro, D.T. Son, A.O. Starinets, The shear viscosity of strongly coupled N = 4 supersymmetric YangMills
plasma, Phys. Rev. Lett. 87 (2001) 081601, hep-th/0104066.
[11] A. Buchel, J.T. Liu, A.O. Starinets, Coupling constant dependence of the shear viscosity in N = 4 supersymmetric
YangMills theory, Nucl. Phys. B 707 (2005) 56, hep-th/0406264.
[12] P. Benincasa, A. Buchel, Transport properties of N = 4 supersymmetric YangMills theory at finite coupling,
JHEP 0601 (2006) 103, hep-th/0510041.
[13] H. Liu, K. Rajagopal, U.A. Wiedemann, Calculating the jet quenching parameter from AdS/CFT, Phys. Rev. Lett. 97
(2006) 182301, hep-ph/0605178.
[14] C.P. Herzog, A. Karch, P. Kovtun, C. Kozcaz, L.G. Yaffe, JHEP 0607 (2006) 013, hep-th/0605158.
[15] H. Liu, K. Rajagopal, U.A. Wiedemann, An AdS/CFT calculation of screening in a hot wind, hep-ph/0607062.
[16] S. Caron-Huot, P. Kovtun, G.D. Moore, A. Starinets, L.G. Yaffe, Photon and dilepton production in supersymmetric
YangMills plasma, hep-th/0607237.
102
[17] E.V. Shuryak, Strongly coupled quarkgluon plasma: The status report, hep-ph/0608177.
[18] A. Buchel, Finite temperature resolution of the KlebanovTseytlin singularity, Nucl. Phys. B 600 (2001) 219, hepth/0011146.
[19] A. Buchel, C.P. Herzog, I.R. Klebanov, L.A. Pando Zayas, A.A. Tseytlin, Non-extremal gravity duals for fractional
D3-branes on the conifold, JHEP 0104 (2001) 033, hep-th/0102105.
[20] S.S. Gubser, C.P. Herzog, I.R. Klebanov, A.A. Tseytlin, Restoration of chiral symmetry: A supergravity perspective,
JHEP 0105 (2001) 028, hep-th/0102172.
[21] A. Buchel, J.T. Liu, Thermodynamics of the N = 2 flow, JHEP 0311 (2003) 031, hep-th/0305064.
[22] O. Aharony, A. Buchel, A. Yarom, Holographic renormalization of cascading gauge theories, Phys. Rev. D 72
(2005) 066003, hep-th/0506002.
[23] D. Mateos, R.C. Myers, R.M. Thomson, Thermodynamics of the brane, hep-th/0701132.
[24] A. Buchel, N = 2 hydrodynamics, Nucl. Phys. B 708 (2005) 451, hep-th/0406200.
[25] P. Benincasa, A. Buchel, A.O. Starinets, Sound waves in strongly coupled non-conformal gauge theory plasma,
Nucl. Phys. B 733 (2006) 160, hep-th/0507026.
[26] A. Buchel, A.W. Peet, J. Polchinski, Gauge dual and noncommutative extension of an N = 2 supergravity solution,
Phys. Rev. D 63 (2001) 044009, hep-th/0008076.
[27] K. Pilch, N.P. Warner, N = 2 supersymmetric RG flows and the IIB dilaton, Nucl. Phys. B 594 (2001) 209, hepth/0004063.
[28] R. Donagi, E. Witten, Supersymmetric YangMills theory and integrable systems, Nucl. Phys. B 460 (1996) 299,
hep-th/9510101.
[29] J. Polchinski, M.J. Strassler, The string dual of a confining four-dimensional gauge theory, hep-th/0003136.
[30] A. Buchel, J.T. Liu, Universality of the shear viscosity in supergravity, Phys. Rev. Lett. 93 (2004) 090602, hepth/0311175.
[31] N.J. Evans, C.V. Johnson, M. Petrini, The enhancon and N = 2 gauge theory/gravity RG flows, JHEP 0010 (2000)
022, hep-th/0008081.
[32] A. Buchel, Comments on fractional instantons in N = 2 gauge theories, Phys. Lett. B 514 (2001) 417, hepth/0101056.
[33] 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.
[34] S.S. Gubser, I. Mitra, Instability of charged black holes in anti-de Sitter space, hep-th/0009126.
[35] A. Buchel, A holographic perspective on GubserMitra conjecture, Nucl. Phys. B 731 (2005) 109, hep-th/0507275.
[36] J.O. Andersen, E. Petitgirard, M. Strickland, Two-loop HTL thermodynamics with quarks, Phys. Rev. D 70 (2004)
045001, hep-ph/0302069.
[37] A. Buchel, Higher derivative corrections to near-extremal black holes in type IIB supergravity, Nucl. Phys. B 750
(2006) 45, hep-th/0604167.
[38] A. Buchel, Gauge theories on hyperbolic spaces and dual wormhole instabilities, Phys. Rev. D 70 (2004) 066004,
hep-th/0402174.
[39] J.M. Maldacena, L. Maoz, Wormholes in AdS, JHEP 0402 (2004) 053, hep-th/0401024.
[40] J.I. Kapusta, C. Gale, Finite-Temperature Field Theory: Principles and Applications, Cambridge Monographs on
Mathematical Physics, Cambridge Univ. Press, 2006.
[41] A.O. Starinets, Effective potentials in quantum field theory at finite temperature, Cand. Sci. Dissertation, 1994
(in Russian).
[42] A. Buchel, On jet quenching parameters in strongly coupled non-conformal gauge theories, Phys. Rev. D 74 (2006)
046006, hep-th/0605178.
[43] E.V. Shuryak, What RHIC experiments and theory tell us about properties of quarkgluon plasma?, Nucl. Phys.
A 750 (2005) 64, hep-ph/0405066.
Phenomenological implications of a class

of neutrino mass matrices
S. Dev , Sanjeev Kumar, Surender Verma, Shivani Gupta
Department of Physics, Himachal Pradesh University, Shimla 171005, India
Available online 27 July 2007
Abstract
The generic predictions of two-texture zero neutrino mass matrices of class A in the flavor basis have
been reexamined especially in relation to the degeneracy between mass matrices of types A1 and A2 and
interesting constraints on the neutrino parameters have been obtained. It is shown that the octant of 23 and
the quadrant of the Dirac-type CP-violating phase can be used to lift this degeneracy.
Mass matrices provide important tools for the investigation of the underlying symmetries and
the resulting dynamics. The first step in this direction is the reconstruction of the neutrino mass
matrix in the flavor basis. However, the reconstruction results in a large variety of possible structures of mass matrices depending strongly on the mass scale, mass hierarchy and the Majorana
phases. However, the relatively weak dependence on some oscillation parameters (23 and )
results in the degeneracy of possible neutrino mass matrices since all the parameters are not
known at present. The mass matrix for Majorana neutrinos contains nine physical parameters
including the three mass eigenvalues, three mixing angles and the three CP-violating phases. The
two squared-mass differences (m221 and m232 ) and the two mixing angles (12 and 23 ) have
been measured in solar, atmospheric and reactor experiments. The third mixing angle 13 and
the Dirac-type CP-violating phase are expected to be measured in the forthcoming neutrino oscillation experiments. The possible measurement of the effective Majorana mass in neutrinoless
double decay searches will provide an additional constraint on the remaining three neutrino pa* Corresponding author.
E-mail addresses: dev5703@yahoo.com (S. Dev), sanjeev3kumar@yahoo.co.in (S. Kumar),

s_7verma@yahoo.co.in (S. Verma), shiroberts_1980@yahoo.co.in ( Gupta).
doi:10.1016/j.nuclphysb.2007.06.030
104
S. Dev et al. / Nuclear Physics B 784 (2007) 103117
rameters viz. the neutrino mass scale and two Majorana CP-violating phases. While the neutrino
mass scale will be determined by the direct beta decay searches and cosmological observations,
the two Majorana phases will not be uniquely determined even if the absolute neutrino mass
scale is known. Under the circumstances, it is natural to employ other theoretical inputs for the
reconstruction of the neutrino mass matrix. Several proposals have been made in the literature
to restrict the form of the neutrino mass matrix and to reduce the number of free parameters
which include presence of texture zeros [14], requirement of zero determinant [5,6] and the
zero trace condition [7] to name a few. However, the current neutrino oscillation data are consistent only with a limited number of texture schemes [14]. In particular, the current neutrino
oscillation data disallow all neutrino mass matrices with three or more texture zeros in the flavor
basis. Out of the fifteen possible neutrino mass matrices with two texture zeros, only seven are
compatible with the current neutrino oscillation data. The seven allowed two texture zero mass
matrices have been classified into three categories. The two class A matrices of the types A1 and
A2 give hierarchical neutrino masses. The class B matrices of types B1 , B2 , B3 and B4 yield a
quasi-degenerate spectrum of neutrino masses. The single class C matrix corresponds to inverted
hierarchy of neutrino masses. Furthermore, only the mass matrices of class A and C can accommodate maximal 23 mixing and are, hence, favored by the data. Thus, only matrices belonging
to class A can accommodate maximal 23 mixing and normal hierarchy of neutrino masses.
In the absence of a significant breakthrough in the theoretical understanding of the fermion
flavors, the phenomenological approaches are bound to play a crucial role in interpreting new
experimental data on quark and lepton mixing. These approaches are expected to provide useful
hints towards unraveling the dynamics of fermion mass generation, CP-violation and identification of possible underlying symmetries of the lepton flavors from which realistic models of
lepton mass generation and flavor mixing could be, hopefully, constructed.
Even though the grand unification on its own does not shed any light on the flavor problem, the
Grand Unified Theories (GUTs) provide the optimal framework in which possible solutions to
the flavor problem could be embedded. This is because the GUTs predict definite group theoretical relations between the fermion mass matrices. For this purpose, it is useful to find out possible
leading order forms of the neutrino mass matrix in a basis in which the charged lepton mass
matrix is diagonal. Such forms of neutrino mass matrix provide useful hints for model building
which will eventually shed important light on the dynamics of lepton mass generation and flavor
mixing. For example, a phenomenologically favored texture of quark mass matrix has been presented earlier [8]. In the spirit of quarklepton similarity, the same texture has been prescribed
for the charged lepton and Dirac neutrino mass matrices. The same texture for the right-handed
neutrino mass matrix in the see-saw mechanism might follow from universal flavor symmetry
hidden in a more fundamental theory of mass generation. Thus, the texture zeros in different
positions of the neutrino mass matrix, in particular and fermion mass matrices, in general could
be consequence of an underlying symmetry. Such universal textures of fermion mass matrices
can, theoretically, be obtained in the context of GUTs based on SO(10). The four texture zero
mass matrices [8] are basically of type A2 in the terminology followed in the present work. Thus,
the selection of A2 type neutrino mass matrices will have important implications for GUTs. On
the other hand, there exist models based on minimal supersymmetric SO(10) [9], for example,
which predict neutrino mass matrices of type A1 . An interesting way of realizing the textures
zeros is the triplet Higgs model based on an Abelian discrete symmetry [10] under which the
transformation properties of 2 and 3 get interchanged in neutrino mass matrices of type A1
and A2 . In another extension of the Standard Model (SM) with three SU(2) scalar triplets with
horizontal symmetry Z6 [11], mass matrices of types A1 and A2 have different transformation
105
properties for SU(2) triplets. Thus, the identification of phenomenologically admissible forms of
neutrino mass matrices will have important implications for the possible unification schemes at
the GUT scale. Furthermore, neutrino mass matrices of types A1 and A2 will have, significantly,
different implications for leptonic CP-violation which is a crucial ingredient of leptogenesis to
explain the cosmological baryon asymmetry. The sign of correlation between the Jarlskog parameter J and the baryon asymmetry is positive for A1 type mass matrices while the sign of
this correlation could be both positive or negative for A2 type mass matrices leading to vastly
different leptogenesis scenarios [12] for the two categories of mass matrices.
In the present work, we examine the neutrino mass matrices of class A for which mee = 0 and
the neutrino masses are hierarchical. There are two types of mass matrices in class A which are
consistent with the neutrino oscillation experiments. For neutrino mass matrices of type A1 ,
mee = me = 0
(1)
whereas
mee = me = 0
(2)
for neutrino mass matrices of type A2 . The texture zeros within a class were thought to have
identical phenomenological consequences [13] leading to the degeneracy of mass matrices of
types A1 and A2 , for example. In the present work, we discuss the ways to lift this degeneracy.
We find that the knowledge of the deviation of atmospheric mixing from maximality and of the
quadrant of the Dirac-type CP-violating phase can be used to distinguish the mass matrices
of types A1 and A2 . It is, also, found, that the prospects for the measurement of 13 in neutrino
mass matrices of class A are quite optimistic since a definite lower bound on 13 is obtained for
this class.
The neutrino mass matrix, in the flavor basis, is given by

(m )ij = U md U T ij ; i, j = e, , ,
(3)
for Majorana neutrinos where md = Diag{m1 , m2 , m3 } is the diagonal neutrino mass matrix and
U is the neutrino mixing matrix which is given by
c12 c13
s12 c13
s13 ei
1 0
0
U = s12 c23 c12 s23 s13 ei c12 c23 s12 s23 s13 ei
s23 c13 0 ei
0
0 0 ei(+)
s12 s23 c12 c23 s13 ei c12 s23 s12 c23 s13 ei c23 c13
(4)
in PDG representation [13]. Here, cij = cos ij and sij = sin ij . For the above parameterization,
the ee, e and e elements of the neutrino mass matrix are given by
2 2
2 2
2
mee = c13
(5)
c12 m1 + c13
s12 m2 e2i + s13
m3 e2i ,

2 2i
2
e m2 c12 c23 s12 m1 e2i m2 c12
s13 s23 ei m1
me = c13 s13 s23 ei e2i m3 s12
(6)
and

2 2i
2
e m2 + c12 s23 s12 m1 e2i m2 c12
s13 c23 ei m1 .
me = c13 s13 c23 ei e2i m3 s12
(7)
106
It will be helpful to note that me can be obtained from me by exchanging s23 with c23 and c23
with s23 . In other words, the transformation 23 23 + 2 transforms me to me . However,
the transformation 23 2 23 followed by the transformation + , also, transforms
me to me . Therefore, if me vanishes for 23 and , then me vanishes for 2 23 and + .
This symmetry argument clearly shows that the predictions of neutrino mass matrices for types
A1 and A2 will be identical for all neutrino parameters except 23 and . The predictions for 23
and can be parameterized as
A1 : 23 = + x, = y,
4
A2 : 23 = x, = + y.
(8)
4
Since, the two squared-mass differences m212 and m223 are known experimentally from the
solar, atmospheric and reactor neutrino experiments, we treat m1 as the free parameter and obtain
m2 and m3 from m1 using the following relations:
m2 = m21 + m212
(9)
and
m3 =
m22 + m223 .
(10)
Our present knowledge [14] of the oscillation parameters has been summarized below:
5
eV2 ,
m212 = 7.9+0.3,1.0
0.3,0.8 10
+0.02,0.09
2
s12
= 0.310.03,0.07
,
+0.37,1.1
m223 = 2.20.27,0.8
103 eV2 ,
+0.06,0.18
2
s23
= 0.500.05,0.16
,
2
s13
< 0.012(0.046).
(11)
Since, there is only an upper bound on 13 , we treat it as an unknown parameter in the beginning
and use the CHOOZ bound given above to constrain the allowed parameter space only at the end.
Thus, the element mee is a function of four unknown parameters viz. m1 , , and 13 while the
elements me and me are the functions of five unknown parameters viz. m1 , , , and 13 .
The condition mee = 0 has,
already, been examined in detail [15]. For 13 = 0, mee vanishes
2
for = 90 and m1 = s12
m212
cos 212 .
For non-zero 13 , the element mee can vanish only in the
normal hierarchy. The real and imaginary parts of mee are separately zero if
2
=
s13
2 m + s 2 m cos 2
c12
1
12 2
1
(12)
2
s13
=
2 m sin 2
s12
2
2
(13)
and
where 1 and 2 are given by

2
2
1 = m1 c12
+ m2 s12
cos 2 m3 cos 2,
(14)

2
2 = m2 s12
sin 2 m3 sin 2.
107
(15)
2 given in Eqs. (12) and (13) are equal if

The two values of s13
sin 2 =
2 m sin 2
s12
2
M
(16)
2 m + s 2 m cos 2
c12
1
12 2
M
(17)
and
cos 2 =
2 is given by
and the value of s13
2
=
s13
where
M=
M
M + m3
(18)
4 + m2 s 4 + 2m m c2 s 2 cos 2.
m21 c12
1 2 12 12
2 12
(19)
Using Eqs. (16) and (17), the parameters 1 and 2 defined in Eqs. (14) and (15) can be written
in terms of a single Majorana phase or . We express them as functions of :
1 = (M + m3 ) cos 2,
2 = (M + m3 ) sin 2.
(20)
Next, we examine the conditions me = 0 and me = 0. For vanishing 13 , me or me can

only be zero if m1 = m2 and = 90 which implies a lower bound on 13 [4]. Therefore, the
angle 13 will be bounded from below for neutrino mass matrices of class A.
The element me vanishes if Re(me ) = 0 and Im(me ) = 0 which yields the following two
relationships for s13 :
s13 =
c12 c23 s12 m1 m2 cos 2

s23
1 cos 2 sin
(21)
and
s13 =
m2 sin 2
c12 c23 s12
s23
1 sin + 2 cos
(22)
where the parameters 1 and 2 are given by Eqs. (14) and (15). The two values of s13 will be
equal if
sin =
1 m2 sin 2 2 (m2 cos 2 m1 )
,
3 21 + 22
cos =
2 m2 sin 2 + 1 (m2 cos 2 m1 )
3 21 + 22
and s13 is given by

3
c12 c23 s12
s13 =
s23
2
+ 2
1
(23)
(24)
108
where
3 =
m21 + m22 2m1 m2 cos 2.
(25)
Therefore, and s13 become functions of m1 , and for vanishing me . Similar expressions
for and s13 can be obtained for vanishing me from Eqs. (23) and (24) by applying the transformations 23 2 23 and + . Therefore, me vanishes if
sin =
1 m2 sin 2 2 (m2 cos 2 m1 )
,
3 21 + 22
cos =
2 m2 sin 2 + 1 (m2 cos 2 m1 )
3 21 + 22
and s13 is, now, given by

3
c12 s23 s12
s13 =
.
c23
2
+ 2
1
(26)
(27)
The transformation + keeps tan unchanged. Eqs. (26) and (27) can, also, be derived
directly in the same way as we derived Eqs. (23) and (24).
Now, we examine the implications of vanishing mee and me for the neutrino mass matrices
of type A1 . Since, 13 must vanish for = 90 if mee = 0, and a vanishing 13 is not allowed
by the condition me = 0, the points = 90 and 13 = 0 are not allowed for the neutrino mass
matrices of type A1 .
Substituting the values of 1 and 2 from Eqs. (20) in Eqs. (23), we obtain
m2 sin 2( ) + m1 sin 2
,
3
m2 cos 2( ) m1 cos 2
cos =
.
3
sin =
(28)
Eliminating from Eqs. (28) by using Eqs. (16) and (17), we obtain
sin =
m1 m2 sin 2
M3
(29)
cos =
2 m2 c2 m2 + m m (c2 s 2 ) cos 2
s12
1 2 12
2
12 1
12
.
M3
(30)
and
The Dirac-type CP-violating phase can, also, be written in terms of as

sin =
m1 sin 2
.
2
s12
3
(31)
The value = 90 corresponds to = 0 and = 0 . Since, = 90 is not allowed, = 0 and

= 0 are, also, not allowed. Therefore, two texture zero mass matrices of class A are necessarily
CP-violating. From Eq. (24), we obtain
c12 c23 s12 3
.
s13 =
(32)
s23
M + 3
109
For the simultaneous existence of texture zeros at the ee and entries, the two relations
for s13 calculated from the conditions mee = 0 and me = 0 should be consistent with each
other. Eqs. (18) and (32) give the same value of s13 for
2 2
M(M + m3 ) tan2 23 = 23 s12
c12
(33)
and the value of s13 is given by

s13 =
2
s23
M
.
2 c2 s 2
c12
23 12 3
(34)
Eq. (33) can be solved to obtain as a function of m1 for the neutrino mass matrices of type A1 .
With this value of , one can calculate and 13 from Eqs. (29), (30) and (34). Eq. (34) can be
used to obtain a lower bound on 13 :
s13 >
2 m c2 |
|m2 s12
1 12
.
2
2
2
m1 + m2
c12 c23 s12
2
s23
(35)
Similarly, one can show that for neutrino mass matrices of type A2 , and s13 are given by
sin =
m1 sin 2
2
s12
3
(36)
and
s13 =
2
c23
2
2 s2
c12 s23
12
M
3
(37)
and the condition for the simultaneous existence of two texture zeros is
2 2
c12 tan2 23 .
M(M + m3 ) = 23 s12
(38)
We note the reciprocity between the values of tan2 23 in Eqs. (33) and (38) for mass matrices
of types A1 and A2 , respectively. This reciprocity can be, gainfully, used to examine deviations
from maximality for the atmospheric mixing angle for matrices of types A1 and A2 and, hence,
to distinguish between them.
m
2
At = 90 , we have m1 = s12
cos 212 and + 2 = n where n is even for A1 and odd for
A2 type neutrino mass matrices. Although, = 90 is not allowed, the relation between and
remains approximately valid even for small deviations about = 90 .
The numerical analysis for the neutrino mass matrices of type A1 is done in the following
2 and s 2 are varied within their experimental
manner. The oscillation parameters m212 , m223 , s12
23
ranges given in Eqs. (11) and is calculated as a function of m1 by using Eq. (33). For these
(m1 , ) values, is calculated from Eqs. (16) and (17), from Eqs. (29) and (30) and 13 from
Eq. (34). The CHOOZ bound on 13 is used to constrain m1 and, hence, all the other neutrino
2
Table 1
The predictions for neutrino mass matrices of class A
Confidence level
m1 (103 eV)
(deg)
(deg)
13 (deg)
1 C.L.
2 C.L.
3 C.L.
2.84.2
2.010.0
1.616.2
83.496.6
76.4103.6
71.2108.8
59.459.4
9090
9090
5.66.3
4.69.8
3.512.4
0.0110.011
0.0210.021
0.0460.046
110
Table 2
The predictions for and 23
1 predictions
23
A1
A2
110.8 110.8
69.2 290.8
44.0 48.2
41.8 46.0
parameters. The neutrino mass matrices of type A2 can be studied in an analogous manner. The
results have been summarized in Table 1 at various confidence levels for m1 , , , 13 and
Jarlskog rephasing invariant quantity [16]
2
J = s12 s23 s13 c12 c23 c13
sin .
(39)
These quantities are the same for mass matrices of types A1 and A2 . It can be seen from Table 1
that the 3 lower bound on 13 is 3.5 . The range for the Majorana-type CP-violating phase at
1 C.L. is found to be 59.4 59.4 . However, if the neutrino oscillation parameters are allowed
to vary beyond their present 1.2 C.L. ranges, the full range for (90 90 ) is allowed. As
noted earlier, the neutrino mass matrices of type A1 and A2 differ in their predictions for
and 23 . At one standard deviation, the allowed range of is (110.8 110.8 ) for type A1
and (69.2 290.8 ) for type A2 and the allowed range of 23 is (44.0 48.2 ) for type A1 and
(41.8 46.0 ) for type A2 [Table 2]. Just like , no constraint on is obtained above 1.2 C.L.
The neutrino mass matrices of types A1 and A2 have some overlap in their predictions regarding
and 23 at one standard deviation. Therefore, a precise measurement of the oscillation parameters
is necessary to distinguish the neutrino mass matrices of types A1 and A2 .
In Fig. 1, we depict the allowed values of m1 , and as the correlation plots at one standard
deviation. The upper panel shows as a function of m1 and the lower panel shows as a function
of . The large spread in the (m1 , ) plot is due to the errors in the neutrino oscillation parameters.
However, there is a very strong correlation between and as indicated by Eqs. (16) and (17).
As seen earlier, the points = 90 and = 0 are excluded in Fig. 1. The two Majorana phases
are strongly correlated with each other. Such a correlation between the Majorana phases was
noted earlier [3]. However, full ranges (90 90 ) are allowed for the two Majorana phases in
that analysis [3]. In contrast, we obtain a very narrow range for the Majorana phase around 90
[Table 1]. Similarly, the Majorana phase is, also, constrained if the oscillation parameters are
limited to their 1 ranges. In Fig. 2, we depict the correlation plots of and with for matrices
of types A1 (left panel) and A2 (right panel). We, also, show the correlation plots of and 23
with one another as well as with 13 in Fig. 2. The Dirac-type CP-violating phase is strongly
correlated with the Majorana-type CP-violating phases and [cf. Eqs. (29) and (31)]. It can
be seen from (, ) and (, ) correlation plots that there are small deviations in the values of
around 90 and the correlation between and is almost linear. The fact that + 2 0 for
A1 type mass matrices and + 2 180 for A2 type mass matrices is apparent from the (, )
correlation plots given in the left and right panels in Fig. 2, respectively. The (, ) correlation
was noted earlier by Xing [3] in its approximate form in a different parameterization. The (, 23 )
plots, clearly, illustrate the point that the neutrino mass matrices of types A1 and A2 have different
predictions for these variables [Table 2] and only a limited region is allowed on the (, 23 ) plane.
This is contrary to the analysis done by Xing [3] where no constraints on and 23 have been
obtained. In fact, this feature is crucial for distinguishing mass matrices of types A1 and A2
which were found to be degenerate in the earlier analyses [13]. The constraints on and 23
are very sensitive to the values of 13 . For the values of 13 smaller than 1 CHOOZ bound,
111
Fig. 1. Correlation plots for neutrino mass matrices of class A at one standard deviation.
the constraints on and 23 become stronger which can be seen from the (13 , 23 ) and (13 , )
plots. For example, if 13 < 6 , then 23 > 46 (above maximal) for type A1 and 23 < 44
(below maximal) for type A2 . It can, also, be seen from the (13 , 23 ) correlation plot in Fig. 2
that the deviation of 23 from maximality is larger for smaller values of 13 . Therefore, if future
experiments measure 13 below its present 1 bound, the neutrino mass matrices of types A1 and
A2 will have different predictions for and 23 with no overlap. However, it would be difficult
to differentiate between matrices of types A1 and A2 if 13 is found to be above its present 1
112
range. As noted earlier, different quadrants for are selected for neutrino mass matrices of types
A1 and A2 . It can, also, be seen in Fig. 2 that the points = 0, are ruled out implying that
neutrino mass matrices of class A are necessarily CP-violating. This feature is, also, apparent
in Fig. 3 where we have plotted J as a function of m1 and in Fig. 4 (identical for A1 and A2 )
which depicts J as a function of for types A1 (top panel) and A2 (bottom panel). The symmetry
argument summarized in Eq. (8) is consistent with the numerical results presented in Fig. 2.
Fig. 2. Correlation plots for neutrino mass matrices of types A1 (left panel) and A2 (right panel).
Fig. 2. (continued)
Fig. 3. J as a function of m1 for the neutrino mass matrices of class A.
113
114
Fig. 4. J as a function of for the neutrino mass matrices of types A1 (top panel) and A2 (bottom panel).
As mentioned earlier, the accuracies of the oscillation parameters required to distinguish between the theoretical predictions of neutrino mass matrices of types A1 and A2 depend on the
upper bound on 13 . With the current precision of the oscillation parameters, the neutrino mass
matrices of types A1 and A2 can be distinguished at 1 , 2 and 3 C.L. for 13 < 6.2 , 5.6
and 5.0 respectively. For 13 > 6.2 the predictions of neutrino mass matrices of types A1 and
A2 cannot be differentiated with the presently available accuracies of the oscillation parameters.
Even with the reduction of the errors in the measurement of m223 to 6% expected at MINOS
[17] and to 2% expected at K2K [18], no significant improvement in the prospects for distin-
115
Table 3
The upper bounds on 13 at which the predictions of neutrino mass matrices of A1 and A2 differ at various confidence
levels for the present and future neutrino experiments
C.L.
Present data
MINOS
K2K
13 < 6.2
13 < 6.5
13 < 6.6
13 < 5.6
13 < 6.0
13 < 6.2
13 < 5.0
13 < 5.5
13 < 5.7
guishing between neutrino mass matrices of types A1 and A2 results [cf. Table 3]. We have
assumed the central value of atmospheric mixing angle 23 to be 45 for the calculations presented in Table 3. If the future experiments measure 23 to be above or below maximality at
a good statistical significance, the mass matrices of types A1 and A2 will be distinguishable
easily.
If neutrinoless double beta decay searches give positive results and mee is measured experimentally or the atmospheric/reactor neutrino oscillation experiments confirm inverted hierarchy,
the neutrino mass matrices of class A will be ruled out. A generic prediction of this class of
models is a lower bound on 13 [Table 1]. If the forthcoming neutrino oscillation experiments
measure 13 below 3.5 , the neutrino mass matrices of class A will, again, be ruled out. The
forthcoming neutrino experiments will definitely probe this range of 13 and aim at measuring the Dirac type CP-violating phase and the deviations of the atmospheric mixing angle from
maximality [19]. The results of these experiments will fall in one of the following four categories
of phenomenological interest:
1.
2.
3.
4.
23 > 45
23 < 45
23 < 45
23 > 45
and 90 < < 90 ,

and 90 < < 270 ,
and 90 < < 90 ,
and 90 < < 270 .
If the experiments confirm the first possibility, the neutrino mass matrices of type A1 may explain
the data. If the experiments confirm the second possibility, the mass matrices of type A2 may be
allowed. In case, the experiments select the third or fourth possibility, the neutrino mass matrices
of class A will be ruled out.
The four cases given above are degenerate for the present neutrino oscillation data because of
the octant degeneracy of 23 (i.e. if 23 < 45 or 23 > 45 ) and the intrinsic degeneracy in the
sign of cos . The above two two-fold degeneracies combined with the two-fold degeneracy in
the sign of m223 gives rise to the eightfold degeneracy in the neutrino parameter space which has
been studied extensively in the literature [2023]. A specific project named Tokai-to-KamiokaKorea (T2KK) two detector complex which will receive neutrino superbeams from J-PARC [23]
has been proposed to resolve these degeneracies. The intrinsic degeneracy in the sign of cos
will be resolved by the spectrum information at T2KK while the 23 octant degeneracy will be
resolved by observing the difference in m212 oscillation effects between the intermediate and
far detectors at T2KK. However, we find that the degeneracies in the neutrino parameter space
are inextricably linked to the degeneracies in the neutrino mass matrix and by finding the octant
of 23 and sign of cos , the future experiments will be able to settle the fate of neutrino mass
matrices of types A1 and A2 .
116
If the neutrino mass matrices of class A are confirmed experimentally, this would correlate
the CP-violation induced by the Dirac phase (which can be observed in the lepton number
conserving processes) with the CP-violation induced by Majorana phases (which can be observed
in the lepton number violating processes) in a definite way.
In conclusion, the degeneracy between A1 and A2 type neutrino mass matrices has been
examined in detail. In the earlier analyses [13], textures within a class were found to be experimentally indistinguishable. However, our analysis shows that the knowledge of the deviations of
the atmospheric mixing from the maximality and of the quadrant of the Dirac-type CP-violating
phase can be used to lift this degeneracy between mass matrices of types A1 and A2 . Contrary
to the results of earlier analyses [3], which leave the Dirac phase completely unconstrained, we
have obtained interesting constraints not only for the Dirac phase but, also, for the Majoranatype CP-violating phases. Our analysis, also, shows that the prospects for the measurement of
13 for neutrino mass matrices of class A are quite optimistic. Moreover, the neutrino mass matrices of class A connect the deviation of 23 from maximality with the quadrant of . Similarly,
the CP-violation observed in the neutrino oscillation experiments is linked with the CP-violation
induced by Majorana phases.
Acknowledgements
The research work of S.D. and S.V. is supported by the Board of Research in Nuclear Sciences (BRNS), Department of Atomic Energy, Government of India vide Grant No.
2004/37/23/BRNS/399. S.K. acknowledges the financial support provided by Council for Scientific and Industrial Research (CSIR), Government of India. We would like to thank Manmohan
Gupta for critical reading of the manuscript and helpful suggestions.
References
[1] P.H. Frampton, S.L. Glashow, D. Marfatia, Phys. Lett. B 536 (2002) 79.
[2] B.R. Desai, D.P. Roy, A.R. Vaucher, Mod. Phys. Lett. A 18 (2003) 1355.
[3] Z.-z. Xing, Phys. Lett. B 530 (2002) 159;
W. Guo, Z.-z. Xing, Phys. Rev. D 67 (2003) 053002.
[4] A. Merle, W. Rodejohann, Phys. Rev. D 73 (2006) 073012.
[5] G.C. Branco, R.G. Felipe, F.R. Joaquim, T. Yanagida, Phys. Lett. B 562 (2003) 265.
[6] B.C. Chauhan, J. Pulido, M. Picariello, Phys. Rev. D 73 (2006) 053003.
[7] X.-G. He, A. Zee, Phys. Rev. D 68 (2003) 037302.
[8] S.N. Gupta, S. Rajpoot, Quark Mass Matrices and the Top Quark Mass, Wayne State University Preprint, September
1990;
S.N. Gupta, J.M. Johnson, Phys. Rev. D 44 (1991) 2110;
S. Rajpoot, Mod. Phys. Lett. A 7 (1992) 309;
H. Fritzsch, Z.Z. Xing, Phys. Lett. B 353 (1995) 114.
[9] H.C. Goh, R.N. Mohapatra, S.-P. Ng, Phys. Rev. D 68 (2003) 115008.
[10] P.H. Frampton, M.C. Oh, T. Yoshikawa, Phys. Rev. D 66 (2002) 033007.
[11] W. Grimus, L. Lavoura, J. Phys. G 31 (2005) 693702.
[12] S. Kaneko, M. Katsumata, M. Tanimoto, JHEP 0307 (2003) 025.
[13] G.L. Fogli, et al., Prog. Part. Nucl. Phys. 57 (2006) 742795.
[14] M. Maltony, T. Schwetz, M.A. Tortola, J.W.F. Valle, New J. Phys. 6 (2004) 122.
[15] S. Dev, S. Kumar, Mod. Phys. Lett. A 22 (2007) 1401, hep-ph/0607048.
[16] C. Jarlskog, Phys. Rev. Lett. 55 (1985) 1039.
[17] D.G. Michael, et al., MINOS Collaboration, Phys. Rev. Lett. 97 (2006) 191801.
[18] M.H. Ahn, et al., K2K Collaboration, Phys. Rev. D 74 (2006) 072003, hep-ex/0606032.
[19]
[20]
[21]
[22]
[23]
E. Abouzaid et al., Report of the APS Neutrino Study Reactor Working Group, 2004.
V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. D 65 (2002) 073023.
H. Minakata, H. Nunokawa, S. Parke, Phys. Rev. D 66 (2002) 093012.
T. Kajita, H. Minakata, S. Nakayama, H. Nunokawa, Phys. Rev. D 75 (2007) 013006.
H. Minakata, hep-ph/0701070.
117
Radiative lifting of flat directions of the MSSM

in de Sitter background
Bjrn Garbrecht
School of Physics & Astronomy, The University of Manchester, Oxford Road, Manchester M13 9PL, United Kingdom
Received 10 December 2006; received in revised form 4 June 2007; accepted 15 June 2007
Abstract
The one-loop effective potential for a charged scalar field in de Sitter background is studied. We derive
an approximate form for the gauge boson propagator with a general covariant gauge-fixing parameter .
This expression is used to derive an effective potential, which agrees with earlier results obtained in Landau
gauge, and the independence of explicitly indicates the gauge-invariance of the calculation. Adding the
scalar- and fermion-loop contributions, that arise in a supersymmetric Higgs model, we find that the effective potential is vanishing to first order in the Hubble rate H 2 in the absence of additional chiral matter.
The contribution due to a chiral multiplet running in the loop is however non-vanishing in the presence of
Yukawa couplings to the Higgs field. When applied to a flat direction of the MSSM with VEV w, a logarithmically running correction of order h2 H 2 w2 arises, where h is a Yukawa coupling. This is of particular
importance for D-term inflation or models endowed with a Heisenberg symmetry, where Hubble scale mass
terms for flat directions from supergravity corrections are absent.
1. Introduction
The scalar potential landscape of the Minimal Supersymmetric Standard Model (MSSM)
is endowed with a large number of flat directions [1]. These lead to a variety of possible
cosmological scenarios [2], the most prominent of which is the AffleckDine mechanism for
baryogenesis [3]. Recently, also possible effects on the thermal history of the Universe have
gained attention [4,5]. Of course, flat directions are not utterly flat, and for both, qualitative and
quantitative predictions in cosmology, a detailed knowledge of their lifting is necessary. In the
E-mail address: bjorn.garbrecht@manchester.ac.uk.
doi:10.1016/j.nuclphysb.2007.06.013
B. Garbrecht / Nuclear Physics B 784 (2007) 118131
119
present day Universe, the flatness is lifted by renormalizable soft supersymmetry (SUSY) breaking operators of TeV scale, which are constrained by particle physics phenomenology, as well as
by non-renormalizable operators, which become important when the vacuum expectation value
(VEV) along the flat direction approaches the superstring scale. The non-renormalizable terms
are in general unknown, although some restraints can be gained from the supergravity scalar potential [68]. In addition, one can also derive the minimum dimension of a lifting operator for a
given flat direction [1].
During inflation, where the initial conditions for the subsequent evolution of the VEVs are set,
there are further possibilities for uplifting. We mention here three important ways. The first is
spontaneous breakdown of SUSY by the vacuum energy density driving inflation [9]. In F -term
models [6,10], the MSSM particle content generically couples to the non-zero F -term, which
drives inflation, via a Grand Unified symmetry breaking Higgs field or even through the Electroweak two-Higgs multiplet [11]. These couplings cause the flat directions to be radiatively
lifted. For the case of mediation by the Electroweak Higgs fields, an explicit expression for the
effective potential of the flat direction has been derived [9]. For the case of large Yukawa couplings, in especial for directions which involve MSSM fields of the third generation, the radiative
effect dominates over the non-renormalizable and the soft SUSY breaking terms, such that the
VEV along the flat direction during inflation can be predicted. This type of SUSY breaking does
not occur in D-term [12,13] or other types of inflation, where no direct renormalizable couplings
between the inflaton-waterfall and the MSSM sector occur. Note also, that this effect is not due
to the background curvature, and indeed, calculating this type of lifting in flat background is a
very good approximation.
The second contribution arises also within F -term inflation, but is absent in D-term models.
When taking supergravity into account, scalar fields generically have Planck-scale suppressed
couplings to the non-vanishing F -terms, which drive inflation [7,8]. Within minimal supergravity, a mass square of 3H 2 , where H is the Hubble rate, arises during inflation this way. For
non-minimal Khler potentials, the mass square can take different, in particular also negative,
values, but is expected to be of order H 2 . This type of mass term can however be forbidden at
tree level by imposing a Heisenberg symmetry [14,15].
The third possibility, the lifting by the background curvature, is the subject of the present
paper. Since various degrees of freedom of different spin, albeit being of equal mass, couple to
curvature in a different way, the mass term in the effective potential is renormalized. For the
effective potential in de Sitter space with a chiral multiplet running in the loop, this is pointed
out in [16].
This observation is however yet not sufficient for the application to flat directions of the
MSSM. Flat directions also couple to gauge sector fields, gauge bosons, gauginohiggsino
mixing states and scalars, which acquire masses through the D-terms. In flat space, the loop
corrections to the effective potential of the flat directions, which arise due to these degrees of
freedom, cancel to all orders in perturbation theory, but in a curved background, this does not
need to be the case. Therefore, we extend the analysis of the effective potential in de Sitter space
to a vector multiplet, which to our knowledge has not been reported in the existing literature so
far. We then use this result to determine the lifting of MSSM flat directions during inflation. We
find that the order H 2 contributions to the effective potential arising from the degrees of freedom
which acquire masses by the super Higgs mechanism cancel. This is a negative result, however,
besides the explicit calculation that is reported here, we are not aware of an alternative way
to predict it. Therefore, the only curvature-induced lifting terms are mediated by the Yukawacouplings and come from chiral supermultiplets in the loop. In combination with the work [16],
120
this paper contains a complete determination of the lifting of flat directions of the MSSM due to
the background curvature during inflation to order H 2 .
The single terms in the effective potential in de Sitter background are known and are constituted by contributions from the scalar and spin- 12 fermion loop contributions [1618] as well
as the vector boson loop [19,20]. While all authors agree about the same generic form of the
contributions, differences due to renormalization conditions [18] or due to minor numerical mistakes in the coefficients occur. While we believe to know the numerically correct answers for the
fermionic and the scalar case [16,18], a rederivation of the gauge-boson loop contribution is in
order. As a novel feature of the calculation presented here, we allow for a general gauge fixing
parameter . The final result is gauge-independent, which is an important consistency check for
our calculation. To our knowledge, this is the first one-loop calculation in a curved background,
which explicitly indicates gauge-invariance. This way, we reproduce Allens Ishikawas result
[20,21], obtained in Landau gauge, = 0.
2. Gauge-invariant effective potential for a charged scalar during inflation
In this section, we derive the one-loop effective potential due to massive gauge-boson-loops
Vgauge in de Sitter background for a charged scalar, which gives masses to the gauge bosons via
the Higgs-mechanism. We denote the mass-square matrix for the gauge-bosons by M 2 and express our result as Vgauge (M 2 ). Effective potentials in de Sitter backgrounds were first calculated
for fermion and scalar loops by Candelas and Raine [17], with numerical inaccuracies within
the finally quoted results however, which motivated rederivations [16,18]. The effective potential
for a gauge boson loop was first calculated by Shore [19] and later on by Allen and Ishikawa
[20,21], who agree with Shores overall form but find different numerical coefficients. For the
construction of renormalization group improved effective potentials in expanding cosmological
backgrounds, see [22].
In order to show that the curved space contributions to the effective potential do not cancel
within SUSY, as the flat space contributions do, we need reliable results with accurately determined coefficients. Here, we use techniques for deriving the gauge boson propagator pioneered
by Allen and Jacobson [23] and further elaborated by Kahya and Woodard [24,25]. Employing in
addition DeWitts method of WKB or adiabatic expansion [26], these allow us to find an approximate expression for the propagator with a general FaddeevPopov gauge-fixing parameter , and
as a novel feature of our derivation, we do not impose a particular choice on . We note that an
exact solution for the propagator in LandauLorenz gauge, = 0, has recently been obtained by
Tsamis and Woodard [27], and results in other specific gauges are derived in [24]. The somewhat
complementary topic, the effect of the scalar loops on photons during inflation, is investigated in
[2832].
In order to derive the effective potential, we proceed in several steps, as partitioned into the
following subsections.
We specify our choice of coordinates to describe the de Sitter background and introduce
coordinate-invariant distance functions, in terms of which the Green functions for the gauge,
ghost and Goldstone fields are expressed.
Then, we perform the covariant gauge-fixing of the Lagrangian, which is a generalisation of
the familiar procedure in flat-space.
Since the effective potential is a function of the masses of the particles, which run in the
loop, we discuss the mass matrices which arise due to the Higgs mechanism.
121
We enumerate the propagators for the fields in the loop. As a particular novel result, we
derive an expression for the gauge-boson propagator for a general gauge-fixing parameter .
We finally assemble the above intermediate results to calculate the effective potential. The
non-trivial cancellation of all contributions, which depend on the gauge-fixing is an important consistency-check of the calculation.
De Sitter background
In our calculations, we use the conventions of, e.g., [24,27]. For de Sitter space expanding at
the Hubble rate H , we use conformal coordinates, such that the scale factor is a = 1/(H ),
where ], 0[ is conformal time. The metric tensor reads
g = a 2 ,
(2.1)
where
= diag(1, 1, . . . , 1 )

D1 times
is the D-dimensional Minkowski metric. We seek to express two-point functions in terms of

one de Sitter-invariant distance variable. Therefore, we introduce the separation between two
coordinate points ( = x 0 )
D
i

2
x x i 2 ,
x 2 (x; x ) = | | i +
(2.2)
i=1
where the i-term is appropriately chosen to construct a Feynman-propagator. The function

x 2 (x; x ) is of course not de Sitter invariant, but it can be used to construct the geodesic distance
(x; x ) as

1
y = 4 sin2 H = a()a( )H 2 x 2 .
(2.3)
2
It has proved useful in many calculations to use the function y(x; x ) rather than the more familiar geodesic distance in order to express invariant lengths, and we adapt this practice in the
following. We refer especially to [27] for an appraisal of the y-basis for the construction of
bitensors.
Gauge fixed Lagrangian
In this setup, we consider the gauge invariant Lagrangian
1
1 a a 1
F g g [D ]T g D V (),
LGI = F
g
4
2
(2.4)
D = 1 + igT a Aa ,
(2.5)
where
and iT a , normalized as tr(iT a )2 = 12 , are the real antisymmetric generators of gauge transformations of the field , which is expressed as a real column vector. The field is decomposed into
122
a VEV v = and into dynamical degrees of freedom as

= + v.
(2.6)
In order to quantize, we apply the FaddeevPopov procedure. For that purpose, the covariant
generalization of the flat-space gauge fixing Lagrangian is
2
1
1 a
(2.7)
A + g T iT a v ,
LGF =
g
2
and the according ghost Lagrangian

ab

1
LFP = a a + gf abc b Ac a M 2 b
g
g 2 a b T T b T a v,
where the
rules from
(2.8)
are the anticommuting C-number ghost fields. We can then derive the Feynman
L = LGI + LGF + LFP .
(2.9)
For the calculation of the one-loop effective potential, it is useful to write down the expansion
of L to quadratic order in the dynamical fields,

1 a
1
1
D
a
1
Aa Aa
Lqu. = d x g A A
2
2
1
1 ab
Aa R Aa Aa M 2 Ab
2
2
u 1 p 2 pq q 1 r 2 rs s
1
u
MG MH
2
2
2

a

2 ab b
a
a
M

.
(2.10)
We note that the term proportional to the Ricci tensor arises due to the fact that the covariant
derivatives do not commute.
Mass matrices
The mass matrix for the gauge bosons is given by
1 2 ab
(2.11)
= g 2 v T T a T b v.
M
2
By an appropriate choice of the generators T a , it naturally decomposes into a zero submatrix for
the components a, b corresponding to the unbroken generators satisfying T a,b v = 0 and a nonzero submatrix for the broken generators with T a,b v
= 0. We denote the part of M 2 restricted to
the non-zero block as M 2 ,
2
M 0
M2 =
(2.12)
.
0 0
The scalar mass matrix MH2 occurring in (2.10) is given by

2
MH
rs
V
,
r s
(2.13)
123
this matrix is non-zero in general. According to the Higgs-mechanism, it gives masses to the
physical Higgs fields consisting of the components of which are perpendicular to T a v. How2 is of course a zero-matrix.
ever, when identifying v with the VEV along a flat direction, MH
The remaining degrees of freedom are the Goldstone fields spanned by the vectors T a v, with
mass terms
1 2 pq
a a
= g 2 vl Tpl
Tqk vk .
MG
(2.14)
2
We also recall the fact that while M 2 and MG2 have in general a different number of zero
eigenvalues, their spectrum of non-zero eigenvalues is identical. In particular, this implies that
tr MG2 = tr M 2 .
Since we want to obtain the effective potential as a function of the modulus of a flat direction,
which we denote by w, it is convenient to express the mass matrices as
1 2 ab 1 2 2 ab
= w
,
M
(2.15)
2
2
1 2 pq 1 2 2 pq
= w G ,
MG
(2.16)
2
2
where w = v i v i . Hence, when finding an expression of the form Vgauge (M 2 ) for the effective
potential, we can immediately read off the result as a function of the single parameter w.
We have defined Vgauge to be the one-loop effective potential due to massive gauge-bosons.
The full effective potential also gets corrections from the massless modes, which are however
independent of the VEV v and consequently of the parameter w. They are therefore irrelevant
for the lifting of the scalar potential and we do not consider them here. This means that we
include the following three contributions to Vgauge :
The potential VA due to the vector fields A , which get a non-zero mass from the matrix
(2.11). These are the modes associated with the non-zero block M 2 in (2.12).
The ghost contribution V , arising from massive ghosts with the mass matrix M 2 . Just as
for the vector fields, the full effective potential also includes corrections from the massless
modes, which however do not lift the scalar potential and are not considered here.
The Goldstone-loop potential VG , due to the mass-matrix MG2 .
These are all contributions, which arise from the gauge-bosons and depend on w.
2 depends on w, but unlike the matrices (2.15), (2.16), it is not
In general, also the matrix MH
2
simply proportional to w , unless one assumes a renormalizable potential and the cancellation
of linear terms in the Higgs field as a renormalization condition. The physical Higgs bosons
contribute to the full effective potential as calculated by Candelas and Raine [16,17]. For the
2 0, which means that it is independent of w
application to exactly flat directions however, MH
and does not contribute to the lifting of the flat direction. There may be of course corrections
2 = 0, as mentioned in the introduction, but these are at most of order H 2 , as in the case
to MH
of F -term inflation. To order H 2 , it is therefore possible to add these and the curvature-induced
corrections for an exactly flat direction.
Propagators
The propagators for the Goldstone and ghost fields are of the same form, since they both
correspond to minimally coupled scalar fields. Let m2i be any of the eigenvalues of MG2 or
124
M 2 . In the adiabatic limit, m2i H 2 for all these eigenvalues, and the scalar propagator in de
Sitter background can be expanded as [16,2831]
2

M
H2 1
1 2
2
iS (x, x ) =
(2.17)
M
log
2H
y + ,
+
2
2
4 y 16
H2
where M2 stands for either M 2 or MG2 .
Let us now derive an according expression for the gauge boson propagator in a general
gauge . Being a bitensor, it can be decomposed into the following components [23,24]:

i ab (y) = B ab (y)
2y
y y
+ C ab (y) .
x x
x x
The solution has to fulfill the Feynman propagator equation

2 ac c b

ac
g 1
i (x, x )
R M
= ig ab D (x x ).
(2.18)
(2.19)
Using the efficient relations by Kahya and Woodard [24], who choose the Feynman gauge = 1,
we can easily generalize their equations for B(y) and C(y) as

M2
4y y 2 B + (2 y)(D 1)B (2 y)(D 1)C 4y y 2 C 2 B
H

1

2
+ (2 y)B DB + 4y y C + (2 y)(D + 1)C = 0,
(2 y)C (D 1)C (2 y)B + (D 1)B
(2.20)
M2
C
H2

1
(2 y)B (D + 1)B + 4y y 2 C + (2 y)(D + 3)C (D + 1)C = 0,
(2.21)
where the prime denotes a derivative with respect to y.
Next, we set D = 4 and make the ansatz for the components corresponding to massive gauge
bosons
2
1
M
B(y) = 1 + 2 log
(2.22)
y + ,
y
H2
2
M
1
1
1
C(y) = 1 2 + 2 + 3 log
(2.23)
y + ,
y
y
y
H2
+
implementing the DeWitt [26] adiabatic expansion of the propagator, which is applicable when
the condition m2i H 2 holds, where m2i stands again for the eigenvalues of M 2 . This series
ansatz, which is similarly also applicable for the scalar and fermion propagators, is to be compared with the alternative possibility of finding exact solutions for the propagators. In de Sitter
background, the exact solutions are typically given in terms of hypergeometric functions, and for
the gauge boson propagator, the solution has only been constructed in particular gauges so far
[24,27]. The expansion ansatz in contrast appears simpler and works in all kind of backgrounds.
Its main shortcoming is however, that in the case where the adiabatic condition m2i H 2 does
not hold, there may be a constant contribution to the propagator of order H 4 /m2i , as it is the
125
case for a minimally coupled scalar field [31], which is not reliably captured in adiabatic expansion. Besides that exact solutions are desirable as a matter of principle, this justifies the efforts to
obtain them.
Plugging the series (2.22) and (2.23) into Eqs. (2.20) and (2.21) and likewise expanding up to
order y 2 and y 2 log y, we find a system of equations, that closes for those coefficients which
are explicitly given in (2.22) and (2.23). The result is

1
1
M 2 2
1
M 2
(1
+
+
2
+
3
y 128 2
16 2
H2
H2
2
M
log
y + ,
H2
B(y) =
C(y) =
1
1
( 1) 2 + .
2
16
y
(2.24)
(2.25)
The normalization has been found by expressing the -function in (2.19) as given in [16,33].
In order to calculate the effective potential, which is a function of one spacetime point only,
we need to take the coincidence limit. We therefore note
2y
y0
8H 2 a 2 ,
x x
1 y y y0
4H 2 a 2 .
y x x
(2.26)
(2.27)
We furthermore introduce a constant physical-length , which serves as a cutoff scale and is used
to regulate the divergent terms as y 0 [16]. The Minkowski trace over the coincident massive
gauge boson propagator then gives

b
a (x, x) =

1 1
1 2
3M + 6H 2 + 2 M 2 2 H 2 log 2 M 2
(3 + ) +
2 4 2
16 2
ab
,
(2.28)
while for the scalar propagators, we find the expressions [16]

2
ab
1 1
1 2 2
2
2
log

+
M
2
H
M
,
G
G
2 4 2
16 2

2
ab
1 1
1 2 2
2
2
a b (x, x) = 2 2 +
log

2
H
M
.
M
4
16 2

a bG (x, x) =
(2.29)
(2.30)
Effective potential
The effective Lagrangian Leff is given by the functional relation

i

d D x Leff = log N 1 D[A, , , ]eiS[A,,, ] ,
from which we can derive
(2.31)
126
w 2
d D x Leff + iN 1

N
w 2
pq

ab
D[A, , , ] d D x{ 2 Aa Ab ab a b + 2G p q }eiS[A,,, ]

=
D[A, , , ]eiS[A,,, ]

ba
1 ba
= d D x g 2 i a b (x, x) + 2 iab
(x, x)
2

1 qp pq
2G iG (x, x) .
(2.32)
2

Here, the functions [a b (x, x )] , ab
(x, x ) and G (x, x ) denote the Feynman propagators
for the gauge, ghost and Goldstone fields, respectively. The derivative of the effective action with
respect to w 2 does not involve any kinetic terms. We can therefore write

D
d xVgauge = d D x Leff ,
(2.33)
pq
for the effective potential arising from the massive gauge-boson loop. Here we have dropped
the terms involving the normalization N as an irrelevant integration constant. Decomposing
Vgauge = VA + VG + V into its individual contributions, we arrive at

VA 1 2 ba ab
= g
i (x, x) ,
(2.34)
2
2
w
VG 1 2 qp pq
= g G iG (x, x),
(2.35)
2
w 2
ba
V
= g 2 iab
(2.36)
(x, x).
2
w
We are now ready to calculate the individual contributions to the effective potential from
(2.34), (2.35) and (2.36)

M2
1 4
VA = tr
3M + 12H 2 M 2 + 2 M 4
(3
+
)
+
8 2 2
64 2

4H 2 M 2 log 2 M 2 ,
(2.37)

2 2
MG2
1 2 4
2
2
MG 4H MG log MG ,
+
VG = tr
8 2 2
64 2

2 2
M2
1 2 4
2
2
V = tr 2 2
2
log

,
M
8H
M
M
4
64 2
(2.38)
(2.39)
where we have dropped the finite, analytical, renormalization scheme dependent terms which are
quoted for the scalar and fermion loops in [16]. Using tr MG2 = tr M 2 = tr M 2 , we finally find the
net result

2 2
3M 2
1 4
2 2
3M + 12H M log M ,
+
Vgauge = VA + VG + V = tr
(2.40)
8 2 2 64 2
which agrees with the one by Allen and Ishikawa [20,21], which is derived in Landau gauge,
= 0. The dependences on the gauge fixing parameter all cancel, as they should. This is an
explicit demonstration of the gauge-independence of the FaddeevPopov method in a curved
127
background and also an important consistency check of the result. We note that to this end, our
observation of gauge invariance is limited to the parametric range, where the adiabatic condition
m2i H 2 , m2i H 2 holds. In the opposite limit, the situation may be different, and we have
also not given a proof that the gauge-invariance persists in the limit of LandauLorenz gauge,
0. We do not consider these questions here, because we do not have an exact expression
or an appropriate approximation for the gauge boson propagator in a general gauge . Since,
unlike the gauge boson, the scalar field allows for an adjustable curvature coupling, which is
inherited by the Goldstone boson, whereas the ghost field is always minimally coupled, it will
be interesting also to investigate the non-adiabatic limit in the future. Gauge invariance should
not be expected in the case of a non-vanishing curvature coupling, similar to the breakdown of
gauge invariance when the Higgs field in flat space is not settled at its minimum, see, e.g., [34]
for a discussion.
3. Lifting of flat directions
We now apply the above result to the possible curvature-induced lifting of flat directions in
a spontaneously broken SUSY-gauge theory during inflation. In order to complete the spectrum
of particles contributing to the one-loop effective potential, we need to enumerate the additional
scalar and fermionic degrees of freedom which attain masses due to the VEV along the flat
direction.
As we have worked with real representations so far, it is convenient to rewrite the complex
scalar fields occurring in SUSY in terms of the standard SO(2) representation of complex numbers,

cos sin
i
, a, R.
ae a
(3.1)
sin
cos
Then, there is for the scalars an additional D-term contribution to the Lagrangian,
2

L g 2 T T a .
(3.2)
When comparing with the Goldstone boson mass term arising from the gauge-fixing Lagrangian
(2.7), there is an additional factor i multiplying T a . This implies that the degrees of freedom,
which pick up masses through the D-terms are linearly independent from the Goldstone bosons.
Up to a factor of 1/ , the mass spectrum is however identical to the one of the Goldstone Bosons,
and we find the mass matrix (cf. (2.14))

1 2 pq
0 1
0 1
= g 2 vl
vk ,
MD
(3.3)
Ta
Ta
1 0
1
0
2
pl
qk
which implies as contribution to the effective potential [16]

2 2
M2
1 4
2 2
VD = tr
M 4H MG log M .
+
8 2 2 64 2
Finally, fermions acquire masses via the terms

L g 2 t a a + a t a ,
(3.4)
(3.5)
where we have denoted the fermion multiplet associated with by and the gauginos by a .
Supersymmetry or more specific the super-Higgs mechanism ensures that the square of the result2 , in particular
ing fermion mass matrix M2 has the same spectrum of eigenvalues as M 2 or MD
128
implying tr M2 = tr M 2 . The fermionic contribution to the effective potential then is [1618]

2 2
M4 1
1
4
2 2
V = tr 2 2 +
M 2H M log M .
2
16 2
(3.6)
Summing over the contributions from the gauge boson (2.40), scalar (3.4) and fermion (3.6)
loops gives for the curvature-induced effective potential along the flat direction
VHiggs = Vgauge + VD + V = 0,
(3.7)
up to possible corrections of order H 4 . This cancellation for the supersymmetric Higgs mechanism urges the question whether it was to be expected. In contrast, the effective potential for two
chiral supermultiplets, consisting together of one massive Dirac fermion and four real scalars of
the mass m, is given by [16]

3
(3.8)
H 2 m2 log 2 m2 ,
2
8
which implies that SUSY is broken during inflation and that the effective potential is indeed
lifted. We leave aside here the interesting question whether the disappointing cancellation (3.7)
could have been predicted in a way which is more than adding the individual contributions to the
effective potential. This leaves also open whether the cancellation is just accidental or whether it
persists beyond the order H 2 .
To summarize, we have found that to order H 2 , there is no curvature-induced lifting mediated
by the gauge coupling g. However, the degrees of freedom which attain masses via Yukawa couplings lift the flat direction as described by the potential Vchiral . In general, m is matrix-valued,
and the trace of (3.8) is to be taken. In order to obtain good estimates however, it is in most of the
cases sufficient to replace m by hw, where h is the largest Yukawa coupling of the fields composing the flat direction, because of the hierarchical nature of the couplings. We further note that
in principle the parameter needs to be fixed by a renormalization condition. There is however
no experimental information available which would fix the coefficient of the operator m2 H 2 at
a certain scale. Therefore, we can only make the statement that a logarithmically running mass
square of order 9/(4 2 )h2 H 2 is inevitably induced for flat directions during inflation. We can
now relate these observations to some aspects of flat direction cosmology.
First, we note that 9/16H 2 corresponds to a critical value for a scalar mass square term,
since for lower masses, the motion of the scalar field is overdamped, whereas for larger masses
the field performs damped oscillations around its minimum [35]. Depending on the value of
, a mass term of this size is likely to be induced by (3.8) for those directions involving topquark Yukawa couplings, such that h is of order one. These are directions containing the Higgs
field Hu and those which contain left- or right-handed stops. For directions which leave the
flavour symmetry partially intact at tree level, this indicates that depending on the sign of
the induced mass square, the fields tend to align either parallel or perpendicular to the stopdirection. If the direction does not involve the top-quark Yukawa coupling, we can estimate h
from the second generation couplings to be of order 103 , such that a much smaller lifting results.
For comparison, within F -term inflation and minimal supergravity, a tree level mass square
of 3H 2 is induced [7,8], which itself is subject to renormalization group running [36]. When
assuming a non-minimal Khler potential, the mass term is cH 2 , where c is a constant of order
one and can be either negative or positive. In that case, unless the form of the Khler potential is
Vchiral =
129
predicted by an underlying theory, the curvature induced lifting is just a correction, which can be
absorbed within the unknown parameter c.
The case is different when a tree-level mass term of Hubble scale is absent. Examples are
D-term inflation [12,13] or F -term inflation, when endowed with a Heisenberg symmetry which
protects the flatness against supergravity corrections [14], provided that also effects from spontaneous SUSY breaking are absent [9]. For these models, it assumed that the main contribution
to the lifting during inflation comes from non-renormalizable operators, which only become important at the Planck- or superstring scale [15,37,38]. In the light of the present work, which
suggests that a logarithmically running mass of order of the Hubble rate nonetheless arises
due to curvature effects, provided the top-quark Yukawa coupling is involved, this view needs
change.
4. Summary and conclusions
In this work, we rederive Allens and Ishikawas expression [20,21] for the one-loop, gaugeboson induced effective potential in de Sitter space. As a novel feature, the calculation explicitly
shows the independence of the gauge-fixing parameter , which is an important consistency
check. The gauge invariance holds in the adiabatic limit, that is when the gauge boson mass is
larger than the Hubble rate. The general case is left for future investigation.
The gauge boson induced potential can be assembled with the scalar and fermionic contributions to an effective potential for a SUSY-theory in de Sitter background. The cancellation of the
flat-space contributions by virtue of SUSY is another check of the results. Additionally, we also
find a cancellation for the order H 2 corrections. Since such a cancellation is absent for chiral
multiplets, it is not necessarily to be expected. A more detailed understanding of this observation
is desirable.
When applied to MSSM flat directions, our results imply that gauge-coupling mediated lifting in the inflationary background is absent. This is a novel result, and to our knowledge has
not appeared in the literature so far. There is however a lifting mediated by Yukawa-couplings.
For directions interacting via top-quark couplings, we find that these generically acquire a
logarithmically running mass term, which is of order of the Hubble rate. For scenarios with
non-vanishing F -term, this is an additional and equally important contribution to supergravity
corrections of the same form. For D-term models or F -term variants where the flat directions
are protected from supergravity-corrections by additional symmetries, this means that Hubblescale mass terms, which are absent at tree-level, may nonetheless be induced at one-loop in de
Sitter background. If it turns out that MSSM flat directions indeed acquire VEVs in the early
Universe and that this leads to observable consequences, the curvature-induced lifting is therefore of importance. Either, observations can be used to determine the unknown renormalization
parameter . Or, if fundamental theories are developed, which make predictions for the renormalization conditions, the lifting in inflationary background needs to be taken into account to
test their correctness.
Acknowledgement
The author is grateful to Apostolos Pilaftsis for interesting comments on gauge invariance of
effective potentials.
130
References
[1] T. Gherghetta, C.F. Kolda, S.P. Martin, Flat directions in the scalar potential of the supersymmetric standard model,
Nucl. Phys. B 468 (1996) 37, hep-ph/9510370.
[2] For a review, see K. Enqvist, A. Mazumdar, Cosmological consequences of MSSM flat directions, Phys. Rep. 380
(2003) 99, hep-ph/0209244.
[3] I. Affleck, M. Dine, A new mechanism for baryogenesis, Nucl. Phys. B 249 (1985) 361.
[4] R. Allahverdi, A. Mazumdar, Quasi-thermal universe: From cosmology to colliders, hep-ph/0505050.
[5] K.A. Olive, M. Peloso, The fate of SUSY flat directions and their role in reheating, hep-ph/0608096.
[6] E.J. Copeland, A.R. Liddle, D.H. Lyth, E.D. Stewart, D. Wands, False vacuum inflation with Einstein gravity, Phys.
Rev. D 49 (1994) 6410, astro-ph/9401011.
[7] M. Dine, L. Randall, S.D. Thomas, Baryogenesis from flat directions of the supersymmetric Standard Model, Nucl.
Phys. B 458 (1996) 291, hep-ph/9507453.
[8] M. Dine, L. Randall, S.D. Thomas, Phys. Rev. Lett. 75 (1995) 398, hep-ph/9503303.
[9] B. Garbrecht, C. Pallis, A. Pilaftsis, Anatomy of F(D)-term hybrid inflation, hep-ph/0605264.
[10] G.R. Dvali, Q. Shafi, R.K. Schaefer, Large scale structure and supersymmetric inflation without fine tuning, Phys.
Rev. Lett. 73 (1994) 1886, hep-ph/9406319.
[11] G.R. Dvali, G. Lazarides, Q. Shafi, Mu problem and hybrid inflation in supersymmetric SU(2)L SU(2)R
U(1)BL , Phys. Lett. B 424 (1998) 259, hep-ph/9710314.
[12] P. Binetruy, G.R. Dvali, D-term inflation, Phys. Lett. B 388 (1996) 241, hep-ph/9606342.
[13] E. Halyo, Hybrid inflation from supergravity D-terms, Phys. Lett. B 387 (1996) 43, hep-ph/9606423.
[14] P. Binetruy, M.K. Gaillard, Non-compact symmetries and scalar masses Ii superstring-inspired models, Phys. Lett.
B 195 (1987) 382.
[15] M.K. Gaillard, H. Murayama, K.A. Olive, Preserving flat directions during inflation, Phys. Lett. B 355 (1995) 71,
hep-ph/9504307.
[16] B. Garbrecht, Ultraviolet regularisation in de Sitter space, Phys. Rev. D 74 (2006) 043507, hep-th/0604166.
[17] P. Candelas, D.J. Raine, General relativistic quantum field theoryan exactly soluble model, Phys. Rev. D 12 (1975)
965.
[18] S.P. Miao, R.P. Woodard, Leading log solution for inflationary Yukawa, Phys. Rev. D 74 (2006) 044019, grqc/0602110.
[19] G.M. Shore, Radiatively induced spontaneous symmetry breaking and phase transitions in curved spacetime, Ann.
Phys. 128 (1980) 376.
[20] B. Allen, Phase transitions in de Sitter space, Nucl. Phys. B 226 (1983) 228.
[21] K. Ishikawa, Gravitational effect on effective potential, Phys. Rev. D 28 (1983) 2445.
[22] E. Elizalde, S.D. Odintsov, Renormalization group improved effective potential for gauge theories in curved space
time, Phys. Lett. B 303 (1993) 240, Russ. Phys. J. 37 (1994) 25, hep-th/9302074.
[23] B. Allen, T. Jacobson, Vector two point functions in maximally symmetric spaces, Commun. Math. Phys. 103 (1986)
669.
[24] E.O. Kahya, R.P. Woodard, Charged scalar self-mass during inflation, Phys. Rev. D 72 (2005) 104001, grqc/0508015.
[25] E.O. Kahya, R.P. Woodard, One loop corrected mode functions for SQED during inflation, Phys. Rev. D 74 (2006)
084012, gr-qc/0608049.
[26] B.S. DeWitt, Quantum field theory in curved spacetime, Phys. Rep. 19 (1975) 295.
[27] N.C. Tsamis, R.P. Woodard, A maximally symmetric vector propagator, gr-qc/0608069.
[28] T. Prokopec, O. Tornkvist, R.P. Woodard, Photon mass from inflation, Phys. Rev. Lett. 89 (2002) 101301, astroph/0205331.
[29] T. Prokopec, O. Tornkvist, R.P. Woodard, One loop vacuum polarization in a locally de Sitter background, Ann.
Phys. 303 (2003) 251, gr-qc/0205130.
[30] T. Prokopec, R.P. Woodard, Vacuum polarization and photon mass in inflation, Am. J. Phys. 72 (2004) 60, astroph/0303358.
[31] T. Prokopec, E. Puchwein, Photon mass generation during inflation: De Sitter invariant case, JCAP 0404 (2004)
007, astro-ph/0312274.
[32] T. Prokopec, N.C. Tsamis, R.P. Woodard, Two loop scalar bilinears for inflationary SQED, gr-qc/0607094.
[33] V.K. Onemli, R.P. Woodard, Super-acceleration from massless, minimally coupled 4 , Class. Quantum Grav. 19
(2002) 4607, gr-qc/0204065.
131
[34] D. Binosi, J. Papavassiliou, A. Pilaftsis, Displacement operator formalism for renormalization and gauge dependence to all orders, Phys. Rev. D 71 (2005) 085007, hep-ph/0501259.
[35] J. McDonald, Positive order H 2 mass corrections and AffleckDine baryogenesis, Phys. Lett. B 456 (1999) 118,
hep-ph/9902416.
[36] R. Allahverdi, M. Drees, A. Mazumdar, Hubble-induced radiative corrections and AffleckDine baryogenesis, Phys.
Rev. D 65 (2002) 065010, hep-ph/0110136.
[37] C.F. Kolda, J. March-Russell, Supersymmetric D-term inflation, reheating and AffleckDine baryogenesis, Phys.
Rev. D 60 (1999) 023504, hep-ph/9802358.
[38] K. Enqvist, J. McDonald, D-term inflation and B-ball baryogenesis, Phys. Rev. Lett. 81 (1998) 3071, hepph/9806213.
Antimatter in the Milky Way

C. Bambi a,b, , A.D. Dolgov a,b,c
a Istituto Nazionale di Fisica Nucleare, Sezione di Ferrara, I-44100 Ferrara, Italy
b Dipartimento di Fisica, Universit degli Studi di Ferrara, I-44100 Ferrara, Italy
c Institute of Theoretical and Experimental Physics, 113259 Moscow, Russia

Abstract
Observational signatures of existence of antimatter objects in the Galaxy are discussed. We focus on
point-like sources of gamma radiation, diffuse galactic gamma ray background and anti-nuclei in cosmic
rays.
1. Introduction
One can conclude from simple considerations that there is much more matter than antimatter
around us [1,2]. The Earth and the Solar System are evidently made of matter and the very small
antiproton-to-proton ratio in cosmic rays, 104 , suggests a secondary origin of antiprotons by
cosmic ray collisions in the interstellar medium and an absence of a large amount of antimatter in
the Galaxy. Still an excess of cosmic antiprotons at low energies [3] might point to non-standard
sources of their production and, in particular, to some antimatter objects in our neighborhood.
A similar conclusion may be obtained from observations of gamma rays originating from e+ e annihilation [4,5]. Though a conventional mechanism of positron production is the most probable
one, light dark matter might be a possible source of positrons, especially because of the observation of the 511-keV line from the halo emission, see the above quoted papers. This explanation
suffers from a necessity to fine-tune the mass of the dark matter particles, so that they would
decay or annihilate into non-relativistic positrons. Another possible source of positrons could
be primordial antimatter in the Galaxy. For other galaxies the sensitivity is not high enough to
* Corresponding author at: Dipartimento di Fisica, Universit degli Studi di Ferrara, I-44100 Ferrara, Italy.
E-mail addresses: bambi@fe.infn.it (C. Bambi), dolgov@fe.infn.it (A.D. Dolgov).

doi:10.1016/j.nuclphysb.2007.06.010
C. Bambi, A.D. Dolgov / Nuclear Physics B 784 (2007) 132150
133
exclude or observe significant pp and e+ e -annihilation indicating possible cosmic antimatter objects, but galaxies dominated by antimatter are excluded in our galactic cluster, i.e. up to
distances of about 10 Mpc [1], by non-detection of rays which would be produced by the
annihilation of the galactic antimatter with the matter from the infalling intergalactic gas.
On the other hand, a priori we could expect an approximately charge symmetric universe, or
at least a universe with some considerable amount of antimatter, since matter and antimatter have
(almost) the same properties.
Cosmological matterantimatter asymmetry can be expressed in terms of the baryon-tophoton ratio, . The Big Bang Nucleosynthesis (BBN) [6] and the Cosmic Microwave Background Radiation (CMBR) [7] provide two independent measurements which are consistent with
each other
=
nB nB
6 1010 ,
n
(1)
where nB nB , whereas the freeze-out abundances in a homogeneous baryo-symmetric universe would be nB /n = nB /n 1018 [8].
There can be three possible types of cosmological matterantimatter asymmetry:
1. is constant and the universe is 100% matter dominated.
2. The universe is globally baryo-symmetric. It consists of equal amount of similar domains of
matter and antimatter.
3. The universe has a non-vanishing average baryonic charge, but is not spatially constant
and can even be negative in some space regions. In other words there could be lumps of
antimatter in matter dominated universe.
The type of the asymmetry depends upon the mechanism of baryogenesis which took place in
the early universe. The baryogenesis scenarios mostly focus on the first possibility, for review
see Refs. [912], but at present there is neither experimental nor observational evidence in favor
of one model over another, since the involved physics operated at such high energies that it is
difficult or impossible to test it in laboratories on the Earth. On the other hand, a globally baryosymmetric universe is certainly theoretically appealing, but it seems observationally excluded or,
to be more precise, the size of the domain where we live should be (at least) comparable to the
present day cosmological horizon [13].
Most interesting phenomenologically is the third case because it allows for existence of antiobjects in our neighborhood and hence for peculiar features which may be observed in a near
future, thanks to the advent of antimatter research projects such as AMS and PAMELA. A small
amount of antimatter is not excluded even nearby in the Galaxy. The aim of our paper is to consider phenomenological manifestation of that. The effects from antimatter objects in the Galaxy
were analyzed also in Ref. [14] for a different mechanism of antimatter creation and because of
that for restricted types of such objects.
The content of the paper is the following. In Section 2 we briefly describe a mechanism for
generation of lumps of antimatter in baryon dominated universe. Their cosmological evolution is
considered in Section 3 and matterantimatter annihilation in contemporary universe is discussed
in Section 4. In Section 5 we focus on point-like sources of rays, in Section 6 on the diffuse
galactic gamma ray background, and in Section 7 on the possibility of anti-nuclei in cosmic rays.
In Section 8 we speculate on more violent events, where large amounts of matter and antimatter
come into contact. The results are summarized in Section 9.
134
2. Antimatter in baryon asymmetric universe

Baryon asymmetric universe with high density regions of matter and antimatter could be created if there existed two different sources of CP violation [15]: the background baryon asymmetry
could be provided by an explicit violation of CP in the Lagrangian, whereas small bubbles with
very high baryon asymmetry could be produced by the presence of a stochastic or dynamical
violation of CP. The concrete models are considered e.g. in Refs. [15,16].
In what follows we will use as a reference the scenario proposed in paper [17]. The starting
point is the AffleckDine mechanism [18], where scalar fields with non-zero baryonic and/or leptonic charges (predicted in supersymmetric extensions of the Standard Model) have the potential
with flat directions, that is the directions along which the potential energy does not change. As a
toy-model, we can consider the potential of the form:
U () = ||4 (1 cos 4 ),
(2)
is the scalar field with baryonic charge B = 0. Potential (2) has four flat diwhere =
rections along cos 4 = 1. It is not invariant under the transformation ei , i.e. B is not
conserved. Due to the infrared instability of massless (or light, m H ) fields in de Sitter space
time [19] during inflation, can condense along the flat directions of the potential, acquiring a
large expectation value. In the course of the cosmological expansion the Hubble parameter drops
down and when the mass of the field, m , exceeds the universe expansion rate H , evolves to
the equilibrium point = 0 and the baryonic charge stored in the condensate is transformed into
quarks by B-conserving processes. Since here CP is violated stochastically by a chaotic phase
of the field , then during the motion to the equilibrium state the matter and antimatter domains
in a globally symmetric universe would be created. An interesting feature of the model is that
regions with a very high , even close to one, could be formed.
If the scalar field is coupled to the inflaton with an interaction term of the kind
||ei
V (, ) = ||2 ( 1 )2 ,
(3)
the gates to the flat directions might be open only for a short time when the inflaton field was
close to 1 . In this case, the probability of the penetration to the flat directions is small and
could acquire a large expectation value only in a tiny fraction of space. The universe would have
a homogeneous background of baryon asymmetry 6 1010 generated by the same field
which did not penetrate to larger distance through the narrow gate or by another mechanism of
baryogenesis, while the high density matter, > 0, and antimatter, < 0, regions would be rare,
while their contribution to the cosmological mass density might be significant or even dominant.
Let us call these bubbles with high baryonic number density B-balls.
Since the conditions for B-ball creation had been prepared during inflation, their initial radius,
RB , might quite naturally exceed the cosmological horizon.
In a simple model, the mass spectrum of B-balls has a log-normal form [17]

dN
M
(4)
= C exp ln2
,
dM
M0
where C, and M0 are unknown parameters of the underlying theory. Depending on their values
and on , which are stochastic quantities, such bubbles could form clouds of matter or antimatter with high (anti)baryon number density, more compact object like (anti-)stars, and primordial
black holes. If all antimatter is hidden inside anti-black holes it would be unobservable. However,
it is natural to expect that there could be abundant anti-stars (either normal or compact ones, as
135
e.g. white dwarfs) or clouds of antimatter with higher than normal baryon density. Such antimatter objects could survive in the early universe due to their higher density, invalidating the bounds
of Ref. [14].
The compact matter/antimatter objects created by the described mechanism might make a part
of, or even the whole, cosmological dark matter.1 An interesting feature of such dark matter is
that it consists of (stellar size) particles with dispersed masses. In particular, even if M0 is close
to the solar mass, there is a non-zero probability that on the tail of distribution (4) very heavy
black holes of millions solar masses might be created. In particular, if there is one heavy black
hole per galaxy, they could serve as seeds for galaxy formation. On the other hand, in the noncollapsed regions with high baryonic number density primordial nucleosynthesis proceeded with
large , producing nuclei heavier than those formed in the standard BBN [21]. If these regions
are in our neighborhood, an observation of heavy anti-nuclei in cosmic rays would be plausible.
A different model of creation of (much smaller size) compact anti matter objects was suggested in Ref. [22]. Their observational signatures in cosmic gamma ray radiation are analyzed
in Ref. [23].
3. Cosmological evolution of B-balls
When baryogenesis was over, the regions with high values of || had the same energy density
as the background. Only the chemical content was different. These are the so called isocurvature
density perturbations. On the boundary of the bubbles the density contrast should be non-zero
due to the gradient term, ||2 , but it is relatively insignificant. Moreover, this term disappeared
at later stage when relaxed down to the equilibrium point, = 0. After the electroweak phase
transition, when quarks acquired masses due to the Higgs field condensate, the density contrast
became nonzero, especially because of the large mass of t -quark, mt 150 GeV. A short lifetime of t and other heavy quarks and a large number of different quark species in the primeval
plasma make the density contrast relatively small and we ignore it in what follows.
More essential is the QCD phase transition when quarks became confined forming nonrelativistic nucleons. The density contrast of B-bubbles would be equal to:

n mN
0.7mN
10.75
mN
rB =
(5)
=
= 2
0.068
,
g T
g
T
g T 4 /30
where mN 940 MeV is the nucleon mass, T is the temperature of the cosmic plasma, and
g 10 is the number of particle species living in the plasma. The non-relativistic baryonic
matter starts to dominate inside the bubble at
T = Tin 65 MeV.
The mass inside a baryon-rich bubble at the radiation dominated stage is

4 3
RB 3
RB 3 t
2
MB =
t
2 105 M (1 + rB )
RB = (1 + rB )MPl
,
3
2t
2t
s
(6)
(7)
2 (1 + r )/32t 2 . If at the moment of

where M is the solar mass and we assumed = 3MPl
B
the horizon crossing, i.e. at RB = 2t , the density contrast would be large, rB 1, the bubble
1 According to Ref. [20], the present observational data allow (at the 95% C.L.) macroscopic compact objects with
masses greater than 102 M to constitute 88% or less of the cosmological dark matter, so 100% is not highly probable
but still allowed.
136
gravitational radius Rg would be larger than RB and the bubble should form a primordial black
hole (BH). The BH masses are determined by the plasma temperature at the moment of horizon
crossing. Assuming the approximate relation t/s (MeV/T )2 , we find MBH 4 105 M
(t/s). The cosmological QCD phase transition took place at t = 105 104 s and the masses of
the first formed BHs should be near 10 solar masses. The baryon-rich bubbles with larger radius
could make much heavier BHs.
Smaller bubbles with RB < 2t and rB 1 could form compact stellar type objects with masses
given by Eq. (7) and with the initial (that is at T = Tin ) mass density of non-relativistic matter:
2
s
B 4.4 105 rB
(8)
g/cm3 1013 4 g/cm3 ,
tin
where Eq. (6) and t/s (MeV/T )2 were used.
The subsequent evolution of B-balls depends upon the ratio of their mass to the Jeans mass.
The Jeans wave length is:

2

MPl
32 2 T
J = cs
(9)
10t T /mN ,
=t
3 mN
where the speed of sound is cs T /mN and = 3m2Pl /32t 2 . It is evident from this expression
that J exceeds horizon, lh = 2t , if T < 40 MeV. If we take the time and temperature in Eq. (9)
at the beginning of matter dominance in the B-bubble given by Eq. (6), the initial value of J
would be
(in)
6.2 104 3/2 s.
(10)
Correspondingly the initial value of the Jeans mass is

2
t 100 M 1/2 .
MJ 135(T /mN )3/2 MPl
(11)
For relativistic expansion the temperature drops as T 1/a, where a(t) is the cosmological
scale factor. In such a case the Jeans mass would rise with time. However, the temperature of
matter dominated B-bubbles drops much faster, T 1/a 2 , and the Jeans mass would decrease
with time in agreement with intuitive expectations. According to Eq. (11) Jeans mass scales as
MJ T 3/2 t . For non-relativistic expansion law, a t 2/3 and consequently
MJ 100 M (ain /a)3/2 ,
(12)
where ain is cosmological scale factor at the onset of matter dominance inside B-ball.
MJ became of the order of the solar mass for a/ain 20. For an estimate let us take Tin
10 MeV. According to Eq. (6), it is realized for 0.15. Correspondingly the temperature inside
B-bubble when MJ dropped down to M , would be
Tin (ain /a)2 0.025 MeV.
(13)
The mass density evolved as 1/a 3 and thus it became

(in)
B = B (ain /a)3 6 105 g/cm3
(14)
to the moment when MJ = M . Correspondingly the radius of such B-ball (i.e. the Jeans wave
length) would be RB 109 cm.
137
It is noteworthy that B-bubbles would be supported against expansion by the pressure of the
hotter surrounding relativistic plasma. Since the photon mean free path inside B-bubbles is quite
small, see below, the heat exchange takes place only on the surface and is not efficient. The
external pressure slows down the expansion of B-balls and their adiabatic cooling.
If J < RB , the baryon-rich bubbles with such radius would decouple from the cosmological
expansion and for RB < 2t they would not be black holes but compact star-like objects. According to Eq. (14) their mass density could be much larger than the density of the normal stars and
similar to that of compact stars. They might survive to the present time or, if nuclear reactions
inside would be significant, they would explode enriching outer space with heavy elements or
anti-elements. Since their temperature at the moment of formation is very high, the pressure is
also high, higher than the usual pressure in normal stars powered by nuclear reactions. So we
assume that such objects mostly survive to the present time. After a while, their temperature
would become larger than the cosmological temperature because they stopped expanding and
the temperature is not red-shifted. They would cool down because of radiation from the surface
4 R 2 , where T
into external colder space. Their luminosity can be estimated as L Tsurf
surf is the
B
surface temperature. With infinitely large thermal conductivity, the life-time with respect to the
cooling would be quite short T 4 RB3 /L RB . Of course this result is an underestimate and
the real cooling time must be much longer. Without additional sources of energy, the cooling
time should be of the order of the photon diffusion time, tdiff , inside these compact objects. If the
photon mean free path lfree is smaller than the radius RB the diffusion time from the center to the
surface can be estimated as
tdiff RB2 /free .
(15)
The mean free time of photons is equal to

free 1/(e ne ),
(16)
where ne is the number density of electrons and e is the photon-electron cross-section. For low
energy photons, E < me , it is equal to the Thomson cross-section, Th = 8 2 /3m2e . For large
energies it drops as inverse center-of-mass energy squared.
The photon diffusion is quite slow and the cooling time of baryon-rich compact objects is
determined by tdiff . Assuming that e = Th we obtain:

e
MB
s
11
tdiff 2 10 s
(17)
.
M
RB
Th
The mass and size of the baryonic bubbles depends upon the parameters of the model of their
creation. We assume, for the sake of estimate, that they predominantly have mass close to the solar one and use the results presented after Eq. (12). The thermal energy of a solar mass B-bubble
taken at the moment when the Jeans mass dropped down to M is determined by the thermal
energy of nucleons and electrons, Eth = 3T . Taking T = 25 keV, see Eq. (13), we find for the
total energy stored inside the B-ball
(tot)
Etherm
= 3T MB /mN 1.5 1029 g 1.5 1050 erg.
(18)
With the cooling time given by Eq. (17) the luminosity of B-bubble would be L 1039 erg/s,
i.e. 106 of the solar luminosity. Such B-bubbles would be quite bright sources of keV radiation.
However, they would stop radiating their thermal before hydrogen recombination. The temperature of CMBR at t = tdiff 1011 s, would be about 3 eV. If the B-bubbles make all dark energy,
138
i.e. BB = 0.25, their mass density at tdiff would be equal to that of CMBR. Correspondingly
the energy density of the emitted keV radiation would be at most 104 105 of CMBR. This
radiation would red-shift today by 104 and move to eV region, i.e. to background light.
More efficient source of power could be nuclear reactions that might ignite inside B-balls.
In the considered example with 106 g/cm3 B-ball has the properties similar to those of the
core of red-giant at the initial stage of its evolution. The main source of energy under these conditions would be helium-4 burning, 34 He 12 C. However, the temperature would be somewhat
larger, T 25 KeV, instead of 10 keV. Since the probability of the above reaction is proportional to T 40 , the life-time of such helium flash would be extremely short. Naively taken these
numbers, we obtain life-time about 10 s. However, this simple estimate can be wrong by several
orders of magnitude because the efficiency of the process is very much different from that in
normal giant star. Still helium would be burnt very quickly and later nuclear reactions would be
presumably insignificant. More accurate estimates would demand development of astrophysics
for such unusual objects as B-balls and this is not a subject of this work. In the following sections
we consider observational manifestations of different types of B-balls allowing for existence of
any type of such objects. Here we note only that if the nuclear physics of B-balls with the chosen
above parameters is similar to that of the giant star core, the result of helium flash would be
a compact star with mass density of about 104 g/cm3 . The flash should take place at relatively
early stage of cosmological expansion to be safe from the CMBR restrictions. Indeed that relative
energy density of B-balls would be BB / < 104 /z. Since nuclear reactions result in emission
of approximately 103 of the total mas of a star, and we do not allow than more 104 distortion
of CMBR, we conclude that such B-balls should consume their helium at cosmological red-shift
z > 105 , i.e. at T 20 eV and t < 3 109 s, which does not look unreasonable in view of the
presented above estimate of the life-time 10 s.
KeV photons coming from the cooling of B-balls would heat up cosmic electrons and they
in turn might distort the spectrum of CMBR. However, since the total energy influx is about
104 105 of the total energy density of CMBR, the spectrum distortion should be not more
than at the same level. More detailed calculations are in order.
If B-bubbles consist of antimatter, the annihilation of the background matter on their surface
would create an additional radiation. However, due to a strong coupling of protons and electrons
to photons before hydrogen recombination, the proton diffusion to an anti-B-bubble is quite
slow and it is easy to see that the process is not efficient. After recombination the number of
annihilations on the surface of one B-ball per unit time would be
3

2
T
RB
2
31
N = 4RB n Vp 10 Vp
(19)
,
0.1 eV
109 cm
where = nB /n = 6 1010 and Vp 105 is the hydrogen velocity. Assuming that the
surface annihilation lasted approximately for one the Hubble time at T 0.1 eV, i.e. t = tH
1013 s and that the mass density of B-bubbles is not larger than 10CMBR , we obtain that the
energy density of gamma radiation from the annihilation on the surface is smaller than 1016 of
that of CMBR.
A more detailed analysis is of course necessary but it does not fit the frames of this letter.
The dynamics and astrophysics of B-bubbles in the early universe will be considered elsewhere.
There could be quite interesting signatures of B-bubbles as e.g. gamma ray background and
background light. They might also have an observable impact on CMBR. We will not go into
these interesting phenomena here. We only want to demonstrate that the model of Ref. [17]
allows an early production of dark stellar-mass objects consisting of matter and antimatter which
139
may survive to the present time. Such objects would behave as the usual cold dark matter and
so they would populate galactic halos in contrast to the normal stars which live in the smaller
luminous parts of galaxies.
4. Antimatter in contemporary universe
In what follows we will not dwell on possible scenarios of antimatter creation but simply
consider phenomenological consequences of their existence in the present day universe, in particular in the Galaxy, allowing for any types of such objects. According to the discussion in
the previous section, astronomically interesting domains of antimatter probably populate galactic halos but we will not confine ourselves to this case only, but consider phenomenology of
more general situation. If high density regions of antimatter have survived up to the present time,
we assume that astronomical anti-objects can be either inside and/or in the halo of our galaxy.
In particular there could be anti-clouds, anti-stars, anti-stellar clusters and anti-black holes (i.e.
black holes generated by the gravitational collapse of antimatter; they may be distinguishable
from standard black holes if they were surrounded by an anti-atmosphere). Of course, the
presence of anti-objects in the Galaxy today should lead to the production of the gamma radiation from matterantimatter annihilation. Hence we would expect 100 MeV from the
decay of 0 -mesons produced in p p-annihilation,
with an average of four per annihilation,

and 2 from e+ e -annihilation with E = 0.511 MeV if e+ e annihilate at rest. In addition to
the slow background positrons there should be also energetic secondary positrons produced by
pion decays from pp-annihilation.
Astronomical observations are seemingly more sensitive to

pp-annihilation
because the total energy release in pp-annihilation
is 3 orders of magnitude
larger than that in e+ e -annihilation and the galactic gamma ray background at 100 MeV is several orders of magnitude lower than the one at 0.5 MeV. On the other hand, e+ e -annihilation
gives the well-defined line which is easy to identify.
In order to study the observational constraints on the number of such anti-objects and to
speculate on possible future investigations, it is helpful to distinguish two different regimes of
matterantimatter annihilation, depending on the ratio R/free , where R is the size of the antiobject and
free = 1/(ann np )
(20)
is the proton or electron mean free path inside an anti-object, with ann and np being respectively
the annihilation cross-section for pp
or e+ e (they have similar order of magnitude) and the
antiproton number density in the anti-object.
4.1. Volume annihilation
Volume annihilation takes place if free is larger than the radius RB , that is:

np
RB
2
< 1.
RB ann np 3 10
0.1 pc
104 cm3
1023 cm2
(21)
One should remember that ann 1/v where v is the velocity of the annihilating particles in
their center-of-mass frame and thus the mean-free path is different for protons with different
velocities. We assume that the typical velocities are of the order of 103 c. The effective time of
annihilation, = free /v does not depend on velocity of the annihilating particles.
140
The annihilation rate per unit time and volume is equal to

n p = ann vnp np ,
(22)
where np 1/cm3 is the average galactic number density of protons. Hence the life-time of
volume-annihilated objects would be
3 15

cm3 /s
cm
10
.
vol = (ann vnp )1 1015 s
(23)
np
ann v
The result does not depend upon the size and density of the anti-object.
Thus the low density anti-objects which are annihilated by surrounding protons inside their
whole volume could not survive in galaxies to the present time, tU 3 1017 s. However, as we
argued in Section 3, B-balls could naturally populate galactic halos, where the proton number
density is lower, about 104 /cm3 , and the life-time of the anti-object would be at the level of
vol 1019 s.
The luminosity in gamma rays of the volume-annihilated objects can be estimated as follows.
(vol)
The total energy release per unit of time is given by Ltot = 2mp N ann , where mp is the proton
mass. The luminosity in gamma rays, produced by the annihilation, would be
L(vol)
0.3L(vol)
tot
1035 erg/s
ann v
1015 cm3 /s
RB
0.1 pc
3
np
104 cm3

np
,
104 cm3
(24)
if the mean free path of photons inside the anti-object is larger than its size, RB . If such an object
is at the distance of 10 kpc from the Earth, the expected flux would be about 1011 erg s1 cm2 ,
which corresponds to 107 photons s1 cm2 . It is below the gamma ray background with energy
100 MeV and in the sensitivity range of the present and future experiments.
We would expect approximately of the same photon flux from e+ e -annihilation because
but number of photons per annihilation is twice
v for e+ e is twice bigger than that for pp,
smaller. Thus the flux of 0.511 MeV line from an anti-cloud at 10 kpc would be about 107 photons cm2 s1 and it should be compared with the observed flux of the same line equal to
104 photons cm2 s1 [5].
Such anti-clouds could also be observed in visible light, created by the bremsstrahlung of
the energetic electrons originated from pp-annihilation.
The probability of such processes is

two orders of magnitude smaller than of the main one. The energy release is roughly 8 order of
magnitude smaller. So the visible luminosity would be 8 orders of magnitude weaker than the
solar one.
The influx of protons could be diminished by the radiation pressure, allowing the anti-objects
to survive up to the present time in more dense regions as well. This effect is considered in
Section 4.3, where we see, however, that the radiation pressure is usually not essential.
4.2. Surface annihilation
If the proton mean free path is much smaller than the size of the anti-object, free R, the
annihilation takes place on the surface. This is typical situation for stellar types anti-objects and,
even more, for compact anti-stars, as e.g. white dwarfs, (anti-)neutron stars, etc. In this case all
the protons that hit the surface of the anti-object annihilate. The annihilation cross-section is
given by the geometrical area of the anti-object, = 4R 2 , and the gamma ray luminosity of
141
such a compact anti-object is equal to:

RB 2 n p
v
2
27
4R
0.6m
F
10
erg/s
,
L(sur)
p p
B
R
cm3
103
(25)
where F np v is the proton flux and R 7 1010 cm is the Solar radius. With this luminosity
a solar mass anti-star would have the life-time of the order of 1027 s, if all the factors in Eq. (25)
are of order unity.
If such an anti-star lives in the galactic center, where np 1/cm3 , its luminosity may be
quite large. However, the pressure of emitted gamma radiation could reduce the proton flux and
(sur)
diminish L . This effect is considered in the following subsection and we see that quite high
luminosities are permitted.
4.3. Eddington limit
If the luminosity of an object is created by the influx of particles, there exists an upper bound
on its luminosity, which follows from the fact that the pressure of the emitted radiation diminishes
the incoming flux.
The force acting on proton, or time-derivative of the proton momentum, P , created by the
gamma ray pressure at distance r from the radiating objects is equal to:
P p n (r),
(26)
where p 1031 cm2 is the cross-section of the Compton scattering on proton, is the gamma
ray energy and n is the number density of emitted photons. The total luminosity of the object is
L = 4R 2 n (R). Here n is taken at the surface of the object, but it drops as (R/r)2 with the
increasing distance r.
The force of the gravitational attraction acting on the protons at distance r is equal to
GN mp M
(27)
.
r2
Demanding that the gravitational attraction should be stronger than the radiation pressure, we
obtain:

31

cm2
MB
10
45
L < 10 erg/s
(28)
.
M
p
Fgrav =
One may argue that the total flux of the particles to the anti-object must be electrically
neutral and hence the much larger force exerted by photons on electrons should be substituted into Eq. (26). For the low energy photons, < me , the Thomson cross-section, Th =
0.66 1024 cm2 , should be used. In the case of photons originating from the pp-annihilation,
their energies are much larger and the cross-section is about e 2 /(me ). This would diminish bound (28) by roughly 4 orders of magnitude. In any case, the Eddington limit is well
above prediction (25).
On the other hand, the excessive charge created by a larger influx of protons with respect to
electrons could be compensated by the out-flux of positrons from the antimatter object. If this is
the case, the limit on the gamma ray luminosity would be given by Eq. (28). Energetic gamma
rays from pp-annihilation
would be accompanied by the emission of low energy positrons, which

would be a source of 0.511 MeV line.
142
5. Point-like sources of gamma radiation

First, we consider the possibility of presence of anti-clouds in our galaxy. If condition (21)
is satisfied, the proton mean free path inside the anti-cloud is larger than the anti-cloud size and
matterantimatter annihilation proceeds in the whole volume. The life-time of such clouds is
given by Eq. (23), if the proton flux is sufficiently large. Such clouds would not survive to the
present time. However, if the proton flux is not big enough, the life-time of the cloud may exceed
the age of the Galaxy.
More favorable condition for survival of anti-clouds are in the galactic halo, where the proton
density is np 104 cm3 . The gamma ray luminosity of such a cloud is given by Eq. (24).
Such a source might be observed on the Earth as a ray source with the flux:

np
np
RB 3 10 kpc 2
7
Earth 10
(29)
cm2 s1 ,
0.1 pc
d
104 cm3
104 cm3
where d is the distance of the anti-cloud from the Earth. Eq. (29) should be compared with
the point source sensitivity of EGRET [24], at the level of 107 photons cm2 s1 for E >
100 MeV and a full two weeks exposure, and of the near-future GLAST [25], which is about two
order of magnitude better, 109 photons cm2 s1 .
As for possible anti-stars, emitting gamma rays from p p-annihilation
on their surface, they

should be quite close to us in order to be detectable point-like sources. For an anti-star in the
galactic disk, its flux would be

2
1 pc 2
7 RB
Earth 10
(30)
cm2 s1 .
R
d
To be observable in a near future such an anti-star should be really in solar neighborhood and if it
is a normal star powered by thermonuclear energy it could even be seen with a naked eye. On the
other hand, if anti-stars are similar to the Sun, the consideration of a possible anti-stellar wind
requires that their number in the Galaxy is smaller than 105 (see below Eq. (38)) and, assuming
they are uniformly distributed in the galactic disk, we would expect a mean anti-star number
density of 1/(100 pc)3 . In this case d 100 pc implies Earth 1011 cm2 s1 , a photon flux
too weak to be detectable from a point-like source.
The -flux from an anti-star may be strongly enhanced if it happens to be in a high density hydrogen cloud. This possibility is quite realistic: even if the mean proton density is np 1 cm3 ,
about one half of the interstellar medium is tied up in gas clouds with average proton density
np 103 cm3 . In other words, gas clouds occupy a fraction of about 103 of the Galaxy volume, so that, assuming 105 anti-stars in the Galaxy, about 100 of them are expected to live in
some gas cloud. Their gamma flux would increase by three orders of magnitude, or even more if
we consider that some clouds are much denser, up to np 106 109 cm3 .
Last, it could be interesting to consider observational signatures of anti-black holes, i.e. black
holes generated by the gravitational collapse of antimatter. They may be distinguished from the
ordinary black holes if they were surrounded by an atmosphere of antimatter (it is not difficult to imagine that it is possible, even if detailed calculations would be necessary for a precise
assertion). Their anti-atmosphere could be considered as an anti-cloud around a very compact
anti-object and, following the considerations of Section 4.1, such an anti-atmosphere could survive up to the present day only if such an anti-black hole was in the galactic halo. In this case this
strange anti-object could be detectable looking for high energy radiation (see Eq. (29)) from
143
a stellar mass object creating gravitational micro-lensing (MACHO) [26]. The same would be
true also for compact anti-stars in galactic halo.
The spectrum of photons from matterantimatter annihilations is well known and consists of
three different parts. First, most energetic are photons from pp
0 2 . If 0 were at rest
we would observe the single 67.5-MeV line. However, the life-time of 0 is very short and they
decay being relativistic. Thus the spectrum spreads both ways up and down and shifts from zero
to higher energies above 200 MeV. The second less energetic and also continuous part comes
from the chain of reactions pp
+ + e+ and subsequent e+ e -annihilation. The
probability of such double annihilation and the shape of the spectrum depends upon the object
where these processes take place. This process seems to be least efficient. The annihilation of
the slow positrons inside antimatter objects with the background electrons leads to the famous
0.511 MeV line, which is easy to identify. This anomalously bright line is observed recently in
the Galactic center [27], Galactic bulge [4] and possibly even in the halo [5]. Though an excess
of slow positrons is explained in a conventional way as a result of their creation by light dark
matter particles, such a suggestion is rather unnatural because it requires a fine-tuning of the
mass of the dark matter particle and the electron mass. More natural explanation is the origin of
these positrons from primordial antimatter objects.
6. Diffuse galactic gamma ray background
In the standard theory, the galactic production of rays is due to inelastic collisions of high
energy cosmic rays with the interstellar medium (the dominant processes are p + p, p + and
+ ), to bremsstrahlung radiation from cosmic ray electrons and to inverse Compton scattering
of electrons with low energy photons. Many authors have calculated the galactic production
rate of per hydrogen atom (see e.g. Ref. [28]) and the estimated rate in the energy range
E > 100 MeV is
2 1025 s1 atom1 ,
(31)
in good agreement with observational data [29]. From Eq. (31) we can deduce the total production rate of high energy rays in our galaxy
tot 1043 s1 .
(32)
The presence of NS anti-stars in the galactic disk, where the average proton number density
is np 1 cm3 , would create a new source of high energy rays, with the contribution
N N (sur) NS ,
(33)
(sur)
(sur)
(sur)
L /(100 MeV) is the photon flux coming from each anti-star and L
is
where N
given by Eq. (25). If we assume that N (33) cannot exceed 10% of the standard galactic production rate of high energy , given by Eq. (32), we obtain a bound on the present number of
anti-stars (for simplicity, we assume that all the anti-stars have the same radius RB )

2
12 R
NS 10
(34)
.
RB
A stronger constraint can be obtained from the consideration of the annihilation of antimatter
from the anti-stellar wind with protons in the interstellar medium. The rate of p emission by the
144
anti-stellar wind per anti-star is

N pwind
1036 W s1 ,
(35)
M is the ratio of an anti-star mass loss rate to the Solar one. The total number
where W = M/
of antiprotons in the Galaxy can be determined by an equation:
N ptot
= S T,
(36)
where S = N pwind
NS is the source term and T the sink term due to annihilation of p.
The life
time of antiprotons in the Galaxy with respect to annihilation is given by Eq. (23), i.e. it is much
smaller than the age of the Galaxy. Thus the number of antiprotons reached a stationary value,
i.e. N ptot
= 0 and the production rate of 100 MeV in the Galaxy is N 4S, because in each
act of pp-annihilation
4 photons are produced on the average. The flux of 100 MeV photons on
the Earth would be

NS
4N pwind
NS
(gal)
pp
(37)
100W
cm2 s1 ,
2
NS
4Rgal
where we took NS = 3 1011 and Rgal = 10 kpc. The luminosity of the Galaxy in 100 MeV
gamma rays from anti-stellar wind would be LS 1044 W NS /NS erg/s. Since from Eq. (32) we
39
find that the total Galaxy luminosity in 100 MeV is Ltot
10 erg/s, the related bound on the
number of anti-stars is:
NS /NS 106 W 1 .
(38)
Here, as always we assume that the contribution from new physics cannot exceed 10% of Ltot
.
A similar restriction can also be obtained from the 0.511 MeV line created by e+ e annihilation with positrons from the anti-stellar wind. Since the number of antiprotons in the
stellar wind is approximately the same as the number of positrons, the flux of 0.511 MeV photons would be close to that given by Eq. (37). Taking that the latter is smaller than 104 cm2 s1
we find
NS /NS 106 W 1 .
(39)
If anti-stars have been formed in the very early universe in the regions with a high antibaryonic
density [17], such primordial stars would most probably be compact ones, white or brown dwarfs,
neutron stars, etc. The stellar wind in this case would be much smaller that the solar one, W 1.
Their luminosity from the annihilation on the surface should be very low, because of their small
radius R, and their number in the Galaxy may be as large as the number of the usual stars. This
possibility is not excluded by the bounds (34) and (38). Such compact dark stars could make a
noticeable part of the cosmological dark matter. As we have argued in Section 3, the early created
compact stellar like objects behave as the usual collisionless cold dark matter (CDM). In this case
it is natural to expect that they would be distributed in and around galaxies as the standard CDM,
having a large number density in the galactic center and decreasing as 1/r 2 in the halo.
These compact objects would generate the diffuse gamma ray background not only now but
during all cosmological history. In particular, we expect a strong constraint on the number of
anti-stars from their emitted radiation during the so called dark age, after recombination but
before the advent of the early standard stars. We thank Francesco Villante for having pointed out
this possibility. We plan to evaluate the intensity of such cosmological -background in another
work.
145
7. Anti-nuclei in cosmic rays

Stable charged particles in cosmic rays consist of 86% of protons, 11% helium nuclei,
1% heavier nuclei, and 2% electrons. It is common belief that the abundances of the elements
in the cosmic rays reflect relative abundances in the Galaxy (even if the low energy cosmic rays
should be a mirror of the relative abundances in the Solar System). Hence, we can reasonably
expect that the antimatter-matter ratio in cosmic rays is more or less equal to the anti-starstar
ratio NS /NS , if the anti-stars are of the same kind as the stars in the Galaxy. As for antiprotons
and positrons, they are not direct indicators for the existence of primordial antimatter, because
they can be produced in many astrophysical processes. For example, the observed p/p
ratio is
at the level of 104 and is compatible with theoretical predictions for p production by the high
energy cosmic ray collisions with the interstellar medium. A possible contribution of p from
exotic sources (ES) is not more than about 10% of the total observed p flux, so the number of
anti-stars NS have to be

NS
p
(40)

105 NS 106 ,
NS
p ES
since the number of ordinary stars in the Galaxy is NS 1011 .
On the other hand, the possibility of producing heavier anti-nuclei (such as anti-helium) in
cosmic ray collisions is completely negligible and a possible future detection of the latter would
be a clear signature of antimatter objects. At present there exists an upper limit on the anti-helium
to helium ratio in cosmic rays, at the level of 106 [30], leading to the constraint

NS
He

106 NS 105 ,
(41)
NS
He ES
which is essentially equal to that from anti-stellar wind in Eq. (38). However, the sensitivity of
AMS [31] and PAMELA [32] space missions is expected to be two orders of magnitude better,
at about 108 . In this case the non-observation of anti-helium nuclei would lead to the much
stronger constraint
NS 103 .
(42)
Of course, the bound (42) cannot be applied if anti-stars are compact ones from the very beginning (i.e. from the moment of their formation in the early universe). In this case the stellar wind
from them and the shortage of anti-supernova events would spread much less (anti)helium than
the normal stars, but heavier anti-elements might be not so strongly suppressed. According to the
scenario described in Section 3, the compact (anti-)stars might form in (anti)baryon-rich regions,
where the primordial nuclear abundances would be quite different from the standard ones with
an enhanced amount of heavier (anti-)nuclei, as e.g. oxygen, nitrogen, carbon, and maybe even
heavier one up to calcium or iron [21]. Spectroscopy observations may detect stellar atmospheres
anomalously rich in heavy elements. Such chemical anomaly is a good signature that such object
are made of antimatter and a search for gamma rays from them looks promising. On the other
hand, the mechanisms capable of ejecting antimatter from such stars and of spreading it out in
the Galaxy are favorable for enrichment of galactic cosmic rays with heavier anti-nuclei. Unfortunately the amount of anti-nuclei depends upon many unknowns and it is impossible to make
a reliable estimate of their flux.
146
8. More exotic events

The presence of anti-stars in the Galaxy could lead to extraordinary events of staranti-star
annihilation. As a matter of fact, the radiation pressure produced in the collision prevents their
total destruction. Still the released energy can be huge.
The most spectacular phenomenon is a collision between a star and an anti-star with similar
masses M. The energy released in such a collision can be estimated as follows. The relative momentum of the colliding stars is approximately P Mv, where the typical value of the relative
velocity is about v 103 . This momentum would be canceled by the pressure of radiation
created by baryonantibaryon annihilation, p. The total pressure force is F r 2 p, where r is
the radius of the circle where the colliding stars penetrated into each other. The bounce would
take place when F tcoll Mv, where tcoll d/v is the collision time and d is the penetration
depth. Since the annihilation products are relativistic, their pressure density is of the same order
of magnitude as the energy density and hence the amount of the annihilated matter during the
staranti-star collision would be
mann Mv 2 ,
i.e. the total energy release would be
2

E1 1048 erg(M/M ) v/103 .
(43)
(44)
Most probably the radiation would be emitted in a narrow disk along the boundary of the colliding
stars.
The collision time can be estimated as follows. Let us introduce the angle at which the
radius, r, of the collision disk is seen from the star center, r = R , assuming that < 1. Here
R is the star radius. The penetration depth is d = R 2 . The volume where matter and antimatter
mix and annihilate is R 3 4 . This volume should be about v 2 of the total stellar volume. Thus the
penetration depth is d vR and the collision time is tcoll R. For the solar type star this time is
about 3 s. For colliding compact staranti-star the collision time would be much smaller but the
energy release could be much larger, because the velocity might approach relativistic values.
We expect that in the process of a star collision with the
similar anti-star the energy would
be emitted inside a narrow disk with the opening angle v. The characteristic time of the
emission is of the order of a second. The energy of the radiation should be noticeably smaller
than 100 MeV because the radiation should degrade in the process of forcing the star bounce.
This makes this collision similar to gamma bursts but unfortunately
some other features do not fit
so well: the released energy should be much larger, about 1053 v erg and it is difficult to explain
the features of the afterglow. A reasonable amount of the energy could be released in the case
of compact star annihilation. It may be possible to explain two or more bursts by the oscillating
motion with interchangingly dominated attraction by gravity and repulsion by the pressure of the
products of annihilation. The process is surely more complicated than our naive picture and we
cannot exclude that gamma bursts are the results of staranti-star annihilation.
The collision of a compact star, e.g. an anti-neutron star, with a usual one or with a red-giant
would look completely different. In this case, the mass densities of the two objects are so much
diverse that the anti-neutron star would go through the red-giant without stopping and would
annihilate all what it meets on the way. The released energy is about

Rns 2 D
2
,
E2 Rns Drg = Mrg
(45)
Rrg
Rrg
147
where Rns is the radius of the anti-neutron star, D is the crossed distance inside the red-giant,
Rrg is the radius of the red-giant, rg is the red-giant mass density, and Mrg is its mass. For
Mrg M and Rrg 1014 cm a reasonable estimate of this energy release is 1038 erg. The
crossing time of the red-giant is about Rrg /v 3 106 seconds. Hence the additional luminosity
during anti-neutron star propagation inside the red-giant would be an order of magnitude smaller
than the solar luminosity and, probably, most of the energy would reheat the interior of the star
and could not reach the free space. These estimates are valid for collision of a compact anti-star
with the envelope of red-giant. A collision of anti-star with the core of red-giant would look
similar to the compact staranti-star collision considered above. Most spectacular would be a
collision of red-giant with anti-red-giant, where a fantastic amount of energy would be released.
The probability of the collision of two stars can be estimated as follows. The collision rate is
= nS v, where R 2 is the geometrical area of the larger star and nS 1/pc3 the mean
star number density in the Galaxy. The total number of collisions per unit time would be:

2
NS
R
N S = NS = vR 2 NS nS 1013 yr1
(46)
.
105
1011 cm
If anti-stars are similar to the Sun, from the bound (38) we find NS 105 , which implies essentially no collisions during the whole history of the universe. On the other hand, if the anti-stars
are white dwarfs or anti-neutron stars, the bounds (34) and (38) are very weak or inapplicable
and the anti-star number can be as large as that of the ordinary stars, i.e. NS 1011 . In this case
we find one collision per 107 years.
The collision with a red-giant would have a larger cross-section because of much larger radius
of the red-giant, Rrg > 1013 cm. Since red-giants are about 1% of the stars in the Galaxy, the
number of collisions could increase by 2 orders of magnitude. Another factor which might also
enhance the probability is a larger gravitational attraction of heavy stars. We know, the majority
of the stars in the Galaxy are in multiple stellar systems. However, it is surely true for the normal
stars which were formed from the primeval hydrogen cloud in galactic disk, while B-balls were
created in the early universe by a completely different mechanism. However, the gravitational
capture of a normal star and B-ball has the same probability as the gravitational capture of the
normal stars, see discussion in the next paragraph.
Another interesting possibility is the transfer of material in a binary system. In the case of
the ordinary stars made of matter, it can happen that a binary system is formed by a red-giant
with a close compact star companion such as a white dwarf. If the former overflows the Roche
lobe of the latter, the compact star captures gases from the red-giant outer atmosphere. Then, on
the white dwarf surface, a large amount of hydrogen is rapidly converted into heavier elements
via the CNO cycle, producing an extremely bright outburst of light. The event taking place can
be a nova, where the star luminosity at the brightness peak is about 1038 erg/s, or the much
more spectacular phenomenon of supernova Ia, with a maximum luminosity of about 1052 erg/s
(however, the latter case is not of interest for us, because the emitted energy comes from star
gravitational collapse and does not from the CNO cycle). We can reasonably expect that something similar can happen in a binary system where one of the stars is made of matter and the other
one of antimatter. In this case hydrogen is not burnt via the CNO cycle, but the much more violent process of matterantimatter annihilation takes place and the white dwarf should be brighter
than an ordinary nova. Of course, the two stars cannot be born at the same time from the same
cloud, but they must have a different origin. For example, one star can be gravitationally captured
by another which is already in a multiple star system (energy conservation forbids the formation
of bound systems from two stars) and, even if today it is ruled out as the dominant formation
148
mechanism, it is expected to be not rare (see e.g. Ref. [33]). The event could be observed as
a long outburst of 100 MeV radiation and be easily distinguished from the standard novae:
the star luminosity could not exceed the Eddington limit (28), but would be probably close to
it. Moreover, even if the rate of the ordinary novae in the Galaxy is estimated to be only about
50 events per year, thanks to the large amount of energy release, we can monitor many galaxies at the same time, even quite distant from us, and increase our possibility of observing this
kind of phenomena. Even in this case, clear predictions are impossible, because observational
probabilities depend on many unknowns.
9. Conclusion
The conclusion to this paper, unfortunately, is not conclusivepractically anything is allowed. Gamma rays from pp-annihilation
may be observable with future or even with existing -telescopes. Quite promising for discovery of cosmic antimatter are point-like sources of
gamma radiation. The problem is to identify a source which is suspicious to consist of antimatter.
A possible manifestation of such a source is an anomalous abundance of chemical (anti-)elements
around it, which can be measured by spectroscopy.
The 100-MeV gamma ray background does not have pronounced features which would unambiguously tell that the photons came from the annihilation of antimatter. The photons produced as
a result of pp-annihilation
would have a well-known spectrum but it may be difficult to establish

a small variation of the conventional spectrum due to such photons.
In contrast, the 0.511 MeV line must originate from e+ e -annihilation and it is tempting to
conclude that the observed excessive signal from the Galaxy and, especially, from the galactic
bulge come from astronomical antimatter objects.
An important feature of the scenario of early formed compact antimatter objects [17] is that
such stellar type ones would behave as normal cold dark matter and they would concentrate in or
around normal galaxies.
Interesting candidates for being antimatter stars are the observed MACHO events [26]. According to the scenario of Ref. [17], these stellar mass gravitational lenses should consist of
practically equal number of matter and antimatter objects. The latter should emit 100 MeV and
0.511 MeV gamma rays and, though the luminosity might be rather weak, see Eq. (25), they still
may be observed with high angular resolution telescopes. The gamma ray luminosity would be
much stronger for MACHOs in the galactic disk, because the number density of protons there is
at least 4 orders of magnitude higher than in the galactic halo.
If an anti-star happens to be in the Galactic Center, its luminosity from the surface annihilation
of the background matter should be strongly enhanced due to the much larger density of the
interstellar matter there. So the search of the antimatter signatures in the direction of the Center
is quite promising.
There is a non-negligible chance to detect cosmic anti-nuclei and not only light anti-helium
but also much heavier ones, especially if anti-stars became early supernovae.
A possible discovery of cosmic antimatter would shed light on the mechanism of baryogenesis
and CP violation in cosmology. In the standard scenarios the baryon asymmetry is constant, just
one number, and it is impossible to distinguish between different mechanisms of baryogenesis
measuring this single number. More exotic models predicting a noticeable amount of antimatter
in our neighborhood are much more interesting because = (x) is a function of space points
and contains much more information about physics in the early universe.
149
Macroscopically large pieces of antimatter not far form the Earth may be interesting energy
sources, but this is not in foreseeable future, maybe only in science fiction.
Acknowledgements
We thank Filippo Frontera and Francesco Villante for useful comments and suggestions.
References
[1] G. Steigman, Annu. Rev. Astron. Astrophys. 14 (1976) 339.
[2] F.W. Stecker, Nucl. Phys. B 252 (1985) 25.
[3] M. Boezio, et al., WIZARD Collaboration, Astrophys. J. 487 (1997) 415;
S. Orito, et al., BESS Collaboration, Phys. Rev. Lett. 84 (2000) 1078, astro-ph/9906426.
[4] G. Weidenspointner, et al., Astron. Astrophys. 450 (2006) 1013, astro-ph/0601673.
[5] G. Weidenspointner, et al., astro-ph/0702621.
[6] W.-M. Yao, et al., J. Phys. G 33 (2006) 1, http://pdg.lbl.gov/.
[7] D.N. Spergel, et al., Astrophys. J. Suppl. 148 (2003) 175, astro-ph/0302209.
[8] Ya.B. Zeldovich, Zh. Eksp. Teor. Fiz. 48 (1965) 986;
A.D. Dolgov, Ya.B. Zeldovich, Rev. Mod. Phys. 53 (1981) 1;
E.W. Kolb, M.S. Turner, The Early Universe, AddisonWesley, Redwood City, USA, 1990.
[9] A.D. Dolgov, Phys. Rep. 222 (1992) 309.
[10] A.D. Dolgov, hep-ph/9707419, lectures at 25th ITEP Winter School of Physics, Moscow, Russia, 1827 February
1997.
[11] A. Riotto, M. Trodden, Annu. Rev. Nucl. Part. Sci. 49 (1999) 35, hep-ph/9901362.
[12] V.A. Rubakov, M.E. Shaposhnikov, Phys. Usp. 39 (1996) 461;
V.A. Rubakov, M.E. Shaposhnikov, Usp. Fiz. Nauk 166 (1996) 493, hep-ph/9603208.
[13] A.G. Cohen, A. De Rjula, S.L. Glashow, Astrophys. J. 495 (1998) 539, astro-ph/9707087;
A.G. Cohen, A. De Rjula, astro-ph/9709132.
[14] K.M. Belotsky, Yu.A. Golubkov, M.Yu. Khlopov, R.V. Konoplich, A.S. Sakharov, Phys. At. Nucl. 63 (2000) 233,
Yad. Fiz. 63 (2000) 290, astro-ph/9807027;
Yu.A. Golubkov, M.Yu. Khlopov, Phys. At. Nucl. 64 (2001) 1821, Yad. Fiz. 64 (2001) 1904, astro-ph/0005419;
D. Fargion, M. Khlopov, Astropart. Phys. 19 (2003) 441, hep-ph/0109133.
[15] A.D. Dolgov, hep-ph/0511213, lectures presented at Varenna School CP Violation: From Quarks to Leptons,
Varenna, Italy, July 2005.
[16] A.D. Dolgov, Antihydrogen Workshop, Munich, Germany, July 1992, hep-ph/9211241;
A.D. Dolgov, Nucl. Phys. B (Proc. Suppl.) 113 (2002) 40;
A.D. Dolgov, in: Proceedings of 14th Rencontres de Blois: MatterAnti-matter Asymmetry, Chateau de Blois,
France, 1722 June 2002, hep-ph/0211260.
[17] A. Dolgov, J. Silk, Phys. Rev. D 47 (1993) 4244.
[18] I. Affleck, M. Dine, Nucl. Phys. B 249 (1985) 361.
[19] A. Vilenkin, L. Ford, Phys. Rev. D 26 (1982) 1231;
A.D. Linde, Phys. Lett. B 116 (1982) 335;
A.A. Starobinsky, Phys. Lett. B 117 (1982) 175;
A.D. Dolgov, D.N. Pelliccia, Nucl. Phys. B 734 (2006) 208.
[20] R.B. Metcalf, J. Silk, astro-ph/0612253.
[21] S. Matsuura, A.D. Dolgov, S. Nagataki, K. Sato, Prog. Theor. Phys. 112 (2004) 971;
S. Matsuura, S.I. Fujimoto, S. Nishimura, M.A. Hashimoto, K. Sato, Phys. Rev. D 72 (2005) 123505, astro-ph/
0507439.
[22] D.H. Oaknin, A.R. Zhitnitsky, Phys. Rev. D 71 (2005) 023519, hep-ph/0309086;
A.R. Zhitnitsky, JCAP 0310 (2003) 010, hep-ph/0202161.
[23] D.H. Oaknin, A.R. Zhitnitsky, Phys. Rev. Lett. 94 (2005) 101301, hep-ph/0406146;
K. Lawson, A.R. Zhitnitsky, arXiv:0704.3064.
[24] D.J. Thompson, et al., Astrophys. J. Suppl. 86 (1993) 629;
See also http://heasarc.gsfc.nasa.gov/docs/cgro/egret/.
150
[25] N. Gehrels, P. Michelson, Astropart. Phys. 11 (1999) 277;

J.E. Carson, J. Phys. Conf. Ser. 60 (2007) 115, astro-ph/0610960;
See also http://glast.stanford.edu/.
[26] C. Alcock, et al., Astrophys. J. 486 (1997) 697, astro-ph/9606165.
[27] P. Jean, et al., Astron. Astrophys. 407 (2003) L55, astro-ph/0309484;
J. Kndlseder, et al., Astron. Astrophys. 411 (2003) L457, astro-ph/0309442;
J. Kndlseder, et al., Astron. Astrophys. 441 (2005) 513, astro-ph/0506026.
[28] M. Mori, Astrophys. J. 478 (1997) 225, astro-ph/9611233.
[29] S.W. Digel, S.D. Hunter, R. Mukherjee, Astrophys. J. 441 (1995) 270.
[30] M. Sasaki, et al., Nucl. Phys. B (Proc. Suppl.) 113 (2002) 202;
J. Alcaraz, et al., AMS Collaboration, Phys. Lett. B 461 (1999) 387, hep-ex/0002048.
[31] F. Barao, AMS Collaboration, Nucl. Instrum. Methods A 535 (2004) 134;
R. Battiston, AMS Collaboration, Nucl. Phys. B (Proc. Suppl.) 166 (2007) 19;
See also http://ams.cern.ch/.
[32] P. Picozza, A. Morselli, J. Phys. G 29 (2003) 903, astro-ph/0211286;
P. Picozza, et al., Astropart. Phys. 27 (2007) 296, astro-ph/0608697;
See also http://wizard.roma2.infn.it/pamela/.
[33] C. Clarke, in: H. Zinnecker, R.D. Mathieu (Eds.), The Formation of Binary Stars, IAU Symposium, vol. 200, 2001.
Nuclear Physics B 784 [FS] (2007) 151201
Notes on parafermionic QFTs with boundary interaction

Sergei L. Lukyanov a,b
a NHETC, Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855-0849, USA
b L.D. Landau Institute for Theoretical Physics, Chernogolovka 142432, Russia
Received 30 March 2007; accepted 18 April 2007

Available online 5 May 2007
Abstract
The main result of these notes is an analytical expression for the partition function of the circular brane
model [S.L. Lukyanov, A.B. Zamolodchikov, J. Stat. Mech.: Theor. Exp. (2004) P05003] for arbitrary values
of the topological angle. The model has important applications in condensed matter physics. It is related to
the dissipative rotator (AmbegaokarEckernSchn) model [V. Ambegaokar, U. Eckern, G. Schn, Phys.
Rev. Lett. 48 (1982) 1745] and describes a weakly blocked quantum dot with an infinite number of
tunneling channels under a finite gate voltage bias. A numerical check of the analytical solution by means
of Monte Carlo simulations has been performed recently in [S.L. Lukyanov, P. Werner, J. Stat. Mech.: Theor.
Exp. (2006) P11002]. To derive the main result we study the so-called boundary parafermionic sine-Gordon
model. The latter is of certain interest to condensed matter applications, namely as a toy model for a point
junction in the multichannel quantum wire [P. Fendley, H. Saleur, Phys. Rev. B 60 (1999) 11432].
Published by Elsevier B.V.
1. Introduction
Since the seminal work by Fateev and Zamolodchikov [1] CFT models with parafermionic
symmetry were extensively studied. In the original formulation the parafermionic current algebra
appears as an extended conformal symmetry of self-dual multicritical points of Zk symmetric
statistical systems [25]. Later this algebra has been employed in string theory [6]. Much effort
has been devoted to the parafermionic models with a boundary [711].
In conformal field theory there is a special class of Conformally invariant Boundary Conditions (CBCs) [7]. In the case of parafermionic models some of CBCs are easily visualized in
terms of the original Zk symmetric statistical systems [79]. The fluctuating variables (spins)
E-mail address: sergei@physics.rutgers.edu.
0550-3213/$ see front matter Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.04.034
152
S.L. Lukyanov / Nuclear Physics B 784 [FS] (2007) 151201
Fig. 1. Zk symmetric statistical system with fixed boundary condition. The spins (1.1) are located at the vertices of the
half-infinite square lattice. For an explicit form of self-dual Boltzmann weights see Ref. [5].
in such systems can be thought of as a set of k special points on the unit circle,
k1
2i
n n=0 with = e k .
(1.1)
The simplest microscopic boundary conditions are those for which all spin boundary values are
the same, say B = n . The scaling limit of the self-dual multicritical Zk system depicted in
Fig. 1, is described in terms of the minimal parafermionic model [1] on the half line x < 0 with
the so-called fixed CBC imposed at x = 0. With some abuse of notation we refer to these CBCs,
as well as the corresponding RG fixed points, as Bn,n (n = 0, 1, . . . , k 1).
Yet another simple type of CBC is the free one. In this case, the microscopic boundary spins
are free to take any value from the set (1.1). In the present work, we denote such CBC and the
associated RG fixed point by Bfree . The universal ratio [811],
gfree
= k,
(1.2)
gfixed
of the corresponding boundary degeneracies [12] is greater than one. Thus a unitary boundary
RG flow R(k) from Bfree to Bn,n seems to exist, or at least does not contradict the g-theorem
[12,13]. Indeed, R(2) (the Ising model case) is a textbook example of boundary flow [14]. For
an arbitrary k the existence of R(k) can be advocated within the general approach [15]. This
boundary flow has been also argued in Refs. [10,16].
In this paper we study the QFT model underlying the boundary flow R(k) . In Section 2.1 we
(k)
(k)
introduce its Hamiltonian H . It is crucial that H depends on an additional dimensionless
(k)
angular parameter, + 2k. The Hamiltonian H suffers from a specific ultraviolet divergence, which is very similar to the small instanton divergence in the 2D O(3) non-linear
(k)
-model [17,18]. In Section 2.2 we consider a certain non-perturbative regularization of H
involving an additional Bose field which does not suffer from this problem. The regularized
model is referred to as the Boundary Parafermionic (BP) sinh-Gordon model. The sinh-Gordon
parameter b plays the role of a regularization parameter and the divergence shows up when
b 0. In Section 2.3 we propose an analytical expression (2.36) for the partition function of
the BP sinh-Gordon model in terms of solutions of a certain second order linear differential
equation. Our motivation behind Eq. (2.36) follows closely along the lines of Al.B. Zamolodchikovs unpublished notes [19]. Namely, in a view of Appendices A and B, the proposed
153
formula is a straightforward generalization of Zamolodchikovs result for the conventional (without parafermions) boundary sinh-Gordon model. Unfortunately, even though Eq. (2.36) can be
easily guessed, its derivation from first principles is still lacking. For this reason we perform
some consistency checks for (2.36) in Appendix C.
(k)
The main quantity we are interested in, is the partition function Z corresponding to the
(k)
Hamiltonian H . It can be extracted from the limiting b2 +0 behavior of the BP sinh(k)
Gordon partition function. An analytical expression for Z is given in Section 2.4. Using this
(k)
result, we describe a qualitative picture of the boundary flow governed by H in Section 2.5.
It turns out that for each from the open segment (2n 1) < < (2n + 1) , the boundary
flow is terminated at the fixed point Bn,n . At the same time the RG trajectories associated with
= (2n + 1) with n = 1, . . . , k 1 posses non-trivial infrared fixed points whose boundary
degeneracies are all the same and equal to 2 cos( k+2

)gfixed . Hence, the infrared physics of the
boundary flow R(k) essentially depends on the parameter . For this reason we shall include an
(k)
additional symbol in the boundary flow notation: R .
In the rest of the main body of the paper we focus on the large-k limit. In Section 3 we identify
(k)
Z = limk Z with the partition function of the circular brane model [20], and the parameter is understood as the topological angle. In Section 4 a particular ultraviolet regularization
of the circular brane model is considered. The regularized theory is known as the dissipative
quantum rotator model [21] and commonly used in describing the Coulomb charging in quantum
dots. The applicability of our results to the quantum dot in weakly blockaded regime is briefly
discussed in Conclusion.
Numerous technical details of the work are relegated to appendices. In Appendices A and B
we study the model which, from a very formal point of view, is the BP sinh-Gordon model
considered at purely imaginary values of the sinh-Gordon parameter b. In essence, these appendices constitute work done in 1999 and which was never published before. It follows closely the
approach developed in the series of papers [2226]. In particular, some of the key statements
involved have been already proven. In those cases we omit the proofs and refer the reader for
details to [2426]. Among results obtained in Appendices A and B is Eq. (A.80), which is interesting in its own right. It generalizes a conjecture of P. Fendley and H. Saleur [27] for the
DC conductance in the multichannel quantum wire to the case of finite temperature. It should be
emphasized that, although the style of our presentation is somewhat sketchy, it seems possible to
transform the derivation of (A.80) into a rigorous mathematical proof.
In Appendix C we perform some consistency checks of the proposed exact analytical expression for the partition function of the BP sinh-Gordon model. Finally, for reference purposes, we
collect some formulas concerning the low-temperature expansion of the partition function Z(k)
in Appendix D.
2. Boundary flow R(k)
(k)
2.1. Hamiltonian H
Let us consider the minimal parafermionic model on a half lane, x 0, constrained by the free
CBC. By employing the magic of modular transformation [9] to the parafermionic characters [6,
154
28], one can observe operators of the scaling dimensions

n(k n)
(n = 0, . . . , k 1)
k
among boundary fields of the theory. Below they are denoted as
k1

n ( ) n=0 .
n =
(2.1)
(2.2)
The variable labels points along the boundary x = 0 (see Fig. 1). Notice that 0 coincides with
the unit operator I . We also reserve special notations for the two relevant boundary fields of the
lowest non-trivial scaling dimension 1 = k1 = 1 k1 , namely
+ 1,
k1 .
(2.3)
The boundary fields n ( ) can be thought of as FateevZamolodchikov chiral parafermionic

currents brought to the boundary point .1 Thus, their boundary Operator Product Expansions
(OPEs) have a form similar to those of the chiral parafermionic algebra:
j (1 ) m (2 ) =
Cj m
j +m (2 ) + .
|1 2 |j +m j m
(2.4)
Here the sum j + m should be understood modulo k and the structure constants are the same as
in [1]. With a properly normalized j , the structure constants read explicitly as

(j + m)!(k j )!(k m)!
Cj m =
(2.5)
.
j !m!k!(k m j )!
(k)
(k)
Let Hfree and Hn be the spaces of states of minimal parafermionic models on the half line
constrained, respectively, by the free (Bfree ) and fixed (Bn,n ) CBCs. Obviously, all the spaces
(k)
Hn are formally isomorphic for different integers n:
(k)
(k)
(k)
H0 H1 Hk1 .
(2.6)
By virtue of the simple microscopic nature of the fixed and free CBCs (see Introduction), it is
expected that
(k)
Hfree
k1

Hn(k) .
(2.7)
n=0
(k)
Then, the fields ( ), being considered as operators acting in Hfree , intertwine the linear
(k)
(k)
(k)
(k)
subspaces Hn Hfree and Hn1 (mod k) Hfree . Heuristically, one can think about ( ) as operators creating discontinuities in the fixed boundary condition Bn,n Bn1,n1 at the boundary
point (see heuristic Fig. 2).
In this work we study the model described by the Hamiltonian

i
i
(k)
(k)
H = H free e k + + e k ,
1 A precise meaning of this statement is discussed in Appendix B.2.
(2.8)
155
Fig. 2. Discontinuity diagram in the fixed boundary conditions corresponding to the insertion of the boundary parafermions.
(k)
where H free is the Hamiltonian of the minimal parafermionic model subject to the free CBC,
(0), and, hence,
(k)
(k)
(k)
H : Hfree Hfree .
(2.9)
1
The parameter in (2.8) carries the dimension [length] k , so that k sets the physical energy
scale (the Kondo temperature) in the theory. It is useful to keep in mind that the OPEs (2.4), (2.5)
specify the normalization of the boundary fields (2.2) modulo global Zk transformations,
j aj j
(a = 1, . . . , k 1).
(2.10)
For this reason the angular parameter + 2k in Eq. (2.8) is not completely unambiguous;
it is defined modulo transformations
2a
(a = 1, . . . , k 1).
The partition function

Z(k) = TrH(k) exp H (k)
/T
(2.11)
(2.12)
free
is an even 2 -periodic function of . Therefore the uncertainty in does not affect the thermodynamics of the model, and we may consider in the domain 0 only.
2.2. The BP sinh-Gordon model
The key idea behind our calculation of (2.12) is that H (k)
appears in a certain limit of a more
general Hamiltonian, involving additional bosonic degrees of freedom. The generalized model is
referred to below as the Boundary Parafermionic (BP) sinh-Gordon model.
Let (x) and (x) be Bose field operators obeying canonical commutation relations

(x), (x
) = 2i(x x
).
(2.13)
The Hamiltonian of the BP sinh-Gordon model is described as follows:

b
b
(k)
(k)
()
H bshg = H free + H free + h B + e k B + e k B ,
(2.14)
156
where
()
H free
1
=
4

dx 2 + (x )2 .
(2.15)
We do not impose any constraint on the boundary values of the canonical fields (2.13). Hence
the Hamiltonian (2.14) acts in the space
(k)
()
H = Hfree
Hfree ,
(2.16)
(k)
()
is the same as in Eq. (2.7) and Hfree
is the space of states of the one component
where Hfree
boson (x) with no constraint at the boundary x = 0. The parameter h in (2.14) plays the role
of an external field coupled to the boundary values B (x)|x=0 . To give an unambiguous
definition of the coupling constant , one should specify the normalization of the boundary
fields
b B
V = e
(2.17)
In what follows these composite fields are assumed to be canonically normalized with respect
their short-distance behavior, i.e.
V (1 )V (2 ) = |1 2 |2d I + ,
(2.18)
and
V (1 )V (2 ) V (k ) =
k!
k
k2
|i j |2(1d) eb
k B
+ ,
(2.19)
i<j
while
eb
k B
(1 )eb
k B
(2 ) = |1 2 |2kb I + .
(2.20)
Here
d =1
1 b2
k
k
(2.21)
k/2
is the scaling dimension
of the boundary fields V (2.17). Also notice that the factor k!/k in
(2.19) is a product k1
C
of
the
structure
constants
(2.5).
j =1 1j
Formally, when the dimensionless parameter b2 in (2.14) is equal to zero, the Hamiltonian
(k)
(k)
H bshg becomes the sum of a free Bose Hamiltonian and H . In fact, because of the particular
divergence generated by the interaction term in (2.14) at b2 0, the limiting procedure is not a
trivial matter. The divergence appears due to the leading term in the OPE (2.19). Keeping this in
mind, we rewrite the Hamiltonian (2.14) as a sum of two operators
H (k)
bshg = H 1 [] + H 2 [ , B ],
(2.22)
with
H 1 [] = H ()
free + h B +

2
A cosh(b k B ) BI
2
kb
(2.23)
157
and

b
b
(k)
H 2 [ , B ] = H free + e k B + e k B

2
(2.24)
2 A cosh(b k B ) BI .
kb
Our basic assumption is that the coefficients A and B for the local boundary counterterms in
(2.24) can be adjusted in such a way that the limit of the operator H 2 does exist when b2 +0.
In other words, we expect that the singular behavior at b2 +0 is fully controlled by the operator H 1 (2.23), which is just a Hamiltonian of the conventional boundary sinh-Gordon model.
The b2 +0 limit is the classical limit for H 1 . Indeed, after the field redefinition = b k
this becomes particularly striking. Since quantum fluctuations of are suppressed as b2 0 we
may apply the saddle point approximation to account for their contribution. The bulk Hamiltonian
for is free and massless, and, consequently, the saddle point is achieved for some constant
classical field configuration which we denote as i . Now, since all matrix elements of H 2 are exi
pected to be finite at b2 = 0, we can safely replace B in (2.24) by the c-number
. Obviously,
b k
the limit

(k)
i
H reg = lim H 2 ,
(2.25)
b2 +0
b k
(k)
should be treated as the regularized version of the Hamiltonian H .

i
The saddle point configuration =
corresponds to a minimum of the boundary potential
b k
U [ B ], which is a sum of the last two terms in the boundary sinh-Gordon Hamiltonian (2.23).
The boundary potential is minimized at
h
sin( ) = ib k
(2.26)
.
2A
We should also take into account the effect of Gaussian fluctuations around the classical saddle
+ (x). Then, expanding the boundary potential
point configuration. Let us write (x) = i
b k
near its minima, one gets
U [ B ] =

2
AS( ) B I + A cos( )( B )2 + O ( B )3 ,
2
kb
(2.27)
with
S( ) = sin( ) + cos( ).
(2.28)
Let ZGauss be the partition function of the Gaussian theory with the quadratic boundary poten(k)
tial, Z () be the partition function corresponding to the regularized Hamiltonian (2.25), and
(k)
Zbshg (h) be the partition function of the BP sinh-Gordon model:

(k)
(k)
Zbshg (h) = TrH exp H bshg /T ,
(2.29)
where the trace is taken over the space of states (2.16). Then, the above consideration implies the
following relation between these three partition functions as b2 +0:

2A
2 ( A S() B
(k)
T )Z
(k) .
e kb2 T
Zbshg
(2.30)
sin
Gauss Z
2
ib k
b +0
158
Finally, according to Ref. [29],

2 A cos()
T
(1 + 2 A
T cos ) C
.
ZGauss = gD
T
4 A
T cos( )
(2.31)
Here gD = 2 4 is the boundary degeneracy associated with the Dirichlet CBC of the uncompactified Bose field and C is some non-universal dimensionful constant.
(k)
2.3. Exact expression for the partition function Zbshg
(h)
Appendixes AC are meant to demonstrate that the partition function (2.29) can be expressed
in terms of the Wronskian W [+ , ] = + u u + of two particular solutions of
the Schrdinger equation
2
bu

u k
u + 2 e Q + e bQ 2 (u) = 0,
(2.32)
where
Q = b + b1 .
(2.33)
More precisely, let + and be regular solutions of (2.32) as u + and u , respectively. They are specified unambiguously via their own WKB asymptotics:

1
as u ,
(u) (2) 2 exp F b1 u
(2.34)
where F is expressed in terms of the conventional hypergeometric function,

ku
k
2bQ 2bQ
k
k u
k
F (b|u) =
u
e 2 F1
, ;1
.
e
4bQ
k
2bQ 2
2bQ
(2.35)
Then, the partition function (2.29) is given by2

(k)
Zbshg (h) = gD gfixed W [+ , ].

The parameters of the Hamiltonian (2.14) are related to and in (2.32) as follows:
kh
=i
,
2Q T
k

bQ
1 k
2
.
=
bQ
2T bQ
k(1 k )
(2.36)
(2.37)
(2.38)
Notice that the Wronskian W [+ , ] can be viewed as a spectral determinant of the

Schrdinger operator (2.32). Indeed the spectrum of (2.32) is simple, discrete and positive:

Spect = n2 n=1 n2 > m2 > 0 if n > m ,
(2.39)
2 Eq. (2.36) is a generalization of the original proposal from [19] to the case k > 1. For k = 1 Al.B. Zamolodchikov
presented a set of convincing arguments in support of (2.36). In particular, the Wronskian W [+ , ] was expressed in
terms of the solution of the massless thermodynamic Bethe ansatz equation associated with the boundary sinh-Gordon
model (2.23).
and, as follows from the BohrSommerfeld quantization condition,

k n
log(log n)
n = 2
1+O
as n .
log n
Q log n
159
(2.40)
Therefore the spectral determinant can be defined through the equation

D(, ) = D0 ()

2
1 2 ,
n
(2.41)
n=1
where D0 () is some overall -independent factor. At the same time, W [+ , ] is an entire

function of the variable 2 which shares the same set of zeros as (2.41).3 Using the standard
WKB methods [30], one can show that log W [+ , ] = o( 2 ) as 2 . Hence, as follows
from the Liouville theorem, the ratio D/W does not depend on , and we can always adjust the
overall normalization factor D0 in (2.41) to make it equal to one, i.e.,

D(, ) = W + (u), (u) .
(2.42)
(k)
2.4. Exact expression for the partition function Z
With the exact expression for Zbshg (h), we can now check the semiclassical behavior (2.30).
Consider the spectral determinant D(, sin ) in the limit b2 +0. There is no problem
with the limit for the differential equation (2.32) and for the solution + (2.34). Namely,

2
2 (k 1)
+ (u) exp
(2.43)
+
c
+ (u) as b2 +0,
sin
+
2
kb2
where + is a solution of

k

2
u + 2 1 + eu 2 sin2 ( ) (u) = 0
(2.44)
defined by the asymptotic condition

e(u+c+ )
+ (u)
exp
k
2(1 + eu ) 4
eu

k
dz
(1 + z) 2 1
z
as u +.
(2.45)
The constant c+ in (2.43) reads explicitly

k
+ E ,
c+ = 1 +
2
(2.46)
d
where (z) = dz
log (z) and E is Eulers constant. Notice that the argument of the exponential
function in (2.43) vanishes for any odd k.
3 Let 2 be a zero of W [ , ]. In this case the solution (u) and (u) are linearly dependent. It just means
+
that the Schrdinger equation possesses a solution which is both regular at x + and x , hence 2 Spect. It
is also obvious that W [+ , ]| 2 = 2 = 0.
n
160
The asymptotic condition (2.34) for is singular as b2 +0. For this reason the limiting
behavior of this solution is a slightly delicate issue. It can be analyzed using the WKB approximation for :

u

bk
bk
2Q 2Q
1
(wkb)
2Q
v
u
exp
dv P(v) e
e
,
(u) =
(2.47)
bk
2P(u)
where P(x) is given by

bu
u k
P(u) = e Q + e bQ sin2 ( )
with sin( ) = /.
(2.48)
bk
Notice that the subtraction term e 2Q v in the integrand comes from the asymptotic conditions
(2.34) and ensures convergence of the integral in (2.47). Consider now the argument {. . .} of the
exponential in (2.47) as b2 +0. It is easy to see that {. . .} F (u) kb22 S(), where S( ) is
given by (2.28), while F (u) u cos( ) as u . Then
ec cos 22 S()
e kb
(u) (u)
2 cos( )
as b2 +0,
(2.49)
where is a solution of the differential equation (2.44) such that

(u) e(u+c ) cos
as u ,
and c is an arbitrary constant. It will be convenient to choose

k
c =
+ E + 2 log 2.
2
(2.50)
(2.51)
We emphasize that can be defined by means of the asymptotic condition (2.50) for 0 < 2
only. For 2 < , the solution grows at large negative u and the asymptotics (2.45) does
not define unambiguously. Fortunately, the function / (1+2 cos ) is an entire function
of the complex variable = cos( ) for u < 0. So the solution (u) can be introduced within
2 through the analytic continuation with respect to the variable from the domain
0 < 2 .
In a view of relations (2.38), the singular behavior of D(, sin ) as b2 +0 matches
exactly the structure (2.30) provided

2 (k 1)
.
A = T ,
B = T sin
(2.52)
2
Furthermore, comparing the pre-exponent in (2.30) with a similar factor for D(, sin ), we
finally obtain the partition function for the regularized Hamiltonian (2.25)
2W [+ , ]
(k)
2
cos
.
Z () = gfixed
(2.53)
(1 + 2 cos )
Two important comments are in order here. First, Z (k) is the function of the dimensionless
variable , which is in fact the inverse temperature measured in the units of the RG invariant
scale E :
=
E
.
T
(2.54)
161
As follows from Eq. (2.38) considered at b2 = 0,

k
k
k 1 2
2
E =
.
2 (1 k1 )
(2.55)
The parameter should be understood now as the coupling of the Hamiltonian (2.25).
Second, the partition function for the regularized Hamiltonian (2.25) is defined modulo an
overall factor ec cos , where c is an arbitrary constant. By fixing the value of c via Eq. (2.51)
we have enclosed a particular normalization condition for the regularized ground state energy

(k)
(k)
E = lim T log Z .
(2.56)
T 0
Namely, if c is chosen as in (2.51), then E satisfies the condition

2 (k)
= 0.
E
2 =0
(k)
(2.57)
In the case of odd k the regularized ground state energy satisfying (2.57) is unambiguously
defined. In particular, the dimensionful constant

1
(k)
E
E =0 = 2 1
(2.58)
k
can be viewed as a universal physical energy scale in the model (2.8). Since the constant B (2.52)
does not vanish for even values of k, the regularized Hamiltonian (2.25) contains an additional
(k)
(k)
counterterm of unit operator. In this case only the difference E E =0 turns out to be a
universal quantity. For even k we choose Eq. (2.58) as an extra normalization condition for the
regularized ground state energy.
(k)
Whereas E satisfying (2.57) and (2.58) is a universal scaling function, the ground state
(k)
energy E for the original Hamiltonian H (k)
(2.8) does depend on details of the regularization
procedure. More precisely, the above consideration suggests (see Eqs. (2.24) and (2.25)) the
following general form for E(k)

(k 1)
(k)
(k)
(k)
(k)
E = E L1 cos( ) L2 sin2
(2.59)
+ E ,
2
(k)
where L1,2 are some dimensionless non-universal constants which are expected to absorb all
divergences in the theory (2.8).4 Hence, the partition function (2.12) has the form

(k 1)
(k)
(k)
(k)
(k)
Z = exp L1 cos( ) L2 sin2
(2.60)
Z (),
2
where Z (k) () is given by Eq. (2.53). For the regularization described above L(k)
1,2 =
(k)
with b2 +0. In general, the divergent constants L1,2
1
kb2
+ O(1)
are expressed in terms of some UV cutoff

equipped with the theory.
(k)
One should note that some particular cases of Z have been already discussed in the literature. E.g., the case k = 2 was studied in Ref. [14]. The differential equation (2.44) for k = 2 can
4 A very similar phenomenon was discussed in [31] for the minimal parafermionic models perturbed by the parafermionic currents in the bulk.
162
be transformed to Kummers equation, so that Z has the especially simple form,
2
g
2
fixed
(2)
Z () =
(2)2 cos ( 2 ) e .
1
2
( 2 + 2 cos ( 2 ))
(2)
(2.61)
Eq. (2.61) is in complete agreement with the result of [14]. It should be also mentioned that
for sin( ) = 0 the partition function is described in terms of the differential equation originally
introduced in [32] in the context of the so-called paperclip boundary flow. The difference be(=0)
(=)
and
are
tween = 0 and = appears in a choice of -solution. Namely,
the two Bloch-wave solutions of (2.44) with sin( ) = 0 [33]. We refer the reader to those papers
for a wealth of data about the differential equation (2.44) with sin( ) = 0. In particular, the ther(k)
modynamic Bethe ansatz equations describing the partition function Z (2.12) for = 0 and
= can be found in Ref. [33].
(k)
2.5. Qualitative pattern of the boundary flow R
(k)
Our prime interest here is in the high- and low-temperature behaviors of Z . With (2.53) it is
(k)
(k)
straightforward to check that the boundary entropy S = (1 + T ddT ) log Z has the following
-independent limit as T :
(k)
lim S (T ) = log(gfree ).
T
At the same time,

lim S(k) (T ) =
T 0
(2.62)
log(gfixed ) as 0 < ,
log(g2 )
(2.63)
as = .
Here the boundary degeneracy g2 is given by the formula

gs = gfixed
s
sin( k+2
)
sin( k+2
)
(2.64)
,
(k)
with s = 2. Whereas (2.62) is just a simple consequence of the fact that R starts from Bfree ,
Eq. (2.63) is much less trivial. It shows that within 0 < the infrared behavior of the theory
is defined by the fixed CBC, but for = the RG flow is terminated at some non-trivial fixed
point.
The CBC associated with the boundary degeneracy g2 can be easily identified. Recall that
the minimal parafermionic model possesses the Cardy states [9] with the additional conformal
symmetry, namely Wk symmetry introduced in Refs. [3436].5 They are in one-to-one correspondence with the k(k + 1)/2 number of Wk primary fields of the model. Let us denote the
CBCs corresponding to the Wk invariant Cardy states by Bl,n with 0 n l k 1. Their
boundary degeneracies are given by Eq. (2.64) with s = l n + 1.6 In particular the boundary
degeneracies associated with B1,0 , . . . , Bk1,k2 and Bk1,0 are all the same and given by the
)gfixed .
constant g2 = 2 cos( k+2
5 See Section B.2 for the brief review of W symmetry and W invariant CBCs in the minimal parafermionic models.
k
k
6 In string language [11], B
l,n are the so-called A-branes. The fixed CBCs Bn,n are D0-branes coinciding with the
points n on the unit circle. The CBCs Bl,n with l > n are interpreted as D1-branes, the chords stretched between the
points n and l . The lightest B-brane from [11] is identified with the free CBC Bfree .
163
(k=4)
Fig. 3. The boundary flow R

in the 2D plane of coupling constants. Polar coordinates (, ) in this plane
are some properly chosen running couplings, = (T , ), = (T , ), such that limT (T , ) = 0 and
max = limT 0 (T , ). The parameter is constant along each trajectory at T = 0.
Now, in view of the global Zk invariance (2.10), (2.11), it is easy to imagine the whole qual(k)
itative picture of the boundary RG flow R for 0 < 2k. In the case k = 4 it is depicted
in Fig. 3. Among the flow lines shown in Fig. 3 there are trajectories adjacent to the big circle
associated with zero temperature. In Appendix D we present an exact formula (D.3) for the nor(k)
malized ground state energy (2.56), (2.57). It turns out that E is a singular function at =
(see formula (D.2)):
1
(k)
E | | 1d ,
(2.65)
where
2
(2.66)
.
k+2
It is worth noting that d is the scaling dimension of the so-called first energy boundary operator, [1]. Hence, an effective Hamiltonian, describing the system in a vicinity of the non-trivial
infrared fixed points Bn+1,n , contains this relevant boundary field along with irrelevant conformal and Wk descendents of and I . Notice that the coupling constant for vanishes at
= , 3, . . . , so the leading low-temperature corrections come from the first W 3 descendent of
of the scaling dimension 1 + d . The low-temperature expansion of (2.53) for the radial trajectories = , 3, . . . was studied in Ref. [33]. More details about the low-temperature expansions
(k)
of Z for = are presented in Appendix D.
d =
3. Circular brane partition function

(k)
3.1. Z in the limit k

The regularized partition function (2.53) possesses a finite limit as k provided the
physical energy scale E is kept fixed. Indeed, let us shift the variable u in the differential equa-
164
tion (2.44):
u = y log(k).
(3.1)
Then, the large-k limit brings (2.44) to the form

2
y + 2 exp ey 2 sin2 ( ) (y) = 0.
(3.2)
The solutions + and remain finite for any fixed y when k . Thus
(k)
Z () = lim Z ()
(3.3)
can be calculated using the same Eq. (2.53) with W [+ , ] being the Wronskian of two solutions of the differential equation (3.2) which are fixed by the asymptotic conditions
cos y cos

e
as y ,
(y) 2eE
(3.4)
and

1
1
1
as y +,
+ (y) (2) 2 exp ey Ei ey
(3.5)
4
2
dx x
with Ei(z) =
z x e . Again, the asymptotic condition (3.4) is applicable for 0 < 2 only.
Within the domain 2 < the solution should be defined through the analytic continuation with respect to the variable .
Let us assume now that k is odd and consider the partition function (2.60) at k . One has
lim
k
k=3,5,...
Z = e cos()L Z (),
(k)
(3.6)
(k)
where L is the limiting value of the non-universal constant L1 . As follows from (2.60), the
large-k limit taken among the even k may cause the result to differ from (3.6) by an extra multiplicative factor e const . Obviously, this subtlety does not affect the physical content of the limiting
theory and we shall ignore it below.
3.2. Circular brane model
Here we identify the QFT describing the large-k limit of the model (2.8).
For finite k the minimal parafermionic model defined on the semi-infinite cylinder
+ 1/T , x 0, contains two holomorphic (, x) = ( ix) and two antiholomorphic
(, x) = ( + ix) currents of conformal dimensions 1 = 1 = 1 k1 . When k ,
the unlocal parafermionic currents turn into chiral spin-1 currents. The latter can be bosonized in
terms of a pair of fields satisfying free massless equations of motion in the bulk x < 0:
= Y
,
X
(3.7)
with = 12 ( + ix ) and = 12 ( ix ). Namely,

lim =
X iY
,
iY
X
lim =
.
k
2
(3.8)
165
Hence, the large-k limit of (2.8) is described by the theory with the bulk Euclidean action
1
Abulk =
1/T
0
d
+ Y Y
).
dx(X X
(3.9)
In view of the heuristic picture of the boundary interaction in (2.8) (see Fig. 2) one may expect the
boundary values X|x=0 XB and Y |x=0 YB satisfy some O(2)-invariant boundary condition.
This boundary condition has been already identified for = 0, in Refs. [20,33]. In both cases
the limit (3.6) coincides with the partition function of the circular brane model. The bulk action
of this two-dimensional model is indeed given by (3.9), while the boundary values of the Bose
fields are subjected to a non-linear constraint
1
(3.10)
.
g0
Let us recall some facts about the circular brane model [20]. Due to the non-linear boundary
condition (3.10), this theory needs renormalization. It has to be equipped with the ultraviolet
(UV) cut-off, and consistent removal of the UV divergences requires that the bare coupling g0 be
given a dependence on the cut-off energy scale , according to the renormalization group (RG)
flow equation
XB2 + YB2 =
dg0
= 2g02 4g03 + .
(3.11)
d
The leading two terms of the -function written down in (3.11) were computed in [37] and [38],
and indeed agree with the more general calculations in [39] and [40]. Eq. (3.11) implies that

1
1 2g0
m
E g0 e
(3.12)
cm g0 ,
1+
m=1
where the symbol stands for asymptotic equality. The coefficients cm in the asymptotic series
(3.12) are not universal, i.e., they depend on the details of the regularization.
It is important that the general definition of the circular brane model involves an additional
parameter, the topological angle . The configuration space for the two component Bose field
X = (X, Y ) consists of sectors characterized by an integer w, the number of times the boundary
value (XB , YB ) winds around the circle (3.10) when one goes around the boundary at x = 0. The
contributions from the topological sectors can be weighted with the factors eiw . Thus, in general
Z =
eiw Z (w) ,
(3.13)
w=
where Z (w) is the path integral
Z (w) = DX DY eAbulk [X,Y ] ,
(3.14)
(w)
evaluated over the fields from the sector w only. As it was pointed out in [41],7 the functional
integral (3.14) has a particular non-perturbative divergence due to the small-size instantons [44,
7 The circular brane model shows many similarities with the O(3) non-linear -model (or n-field) [42,43], where
a very similar effect of the small instanton divergence was known for a while [17,18].
166
45], which cannot be absorbed into the renormalization of the coupling constant g0 . Due to this
effect, the partition function (3.13) is expected to have the form (3.6), where

L = 2 log
(3.15)
,
E
with some ultraviolet cut-off regularizing the small instanton divergence, and the inverse
temperature measured in units of the physical scale (3.12).8 This suggests to extend the results of
Refs. [20,33] and identify (3.6) with the partition function of the circular brane model Z where
plays the role of the topological angle:

Z = exp 2 cos( ) log

(3.16)
Z ().
E
3.3. High-temperature expansion
Here we discuss the high-temperature behavior of the partition function Z .
First and foremost we recall a simple qualitative picture described in Refs. [20,32]. As follows from Eq. (3.11), the circular brane model is a asymptotically free theoryat short distances the effective size of the circle (3.10) becomes large. For small but non-zero values of the
bare coupling constant g0 we can locally ignore the curvature effects and write the Bose field
X (, x) = X0 + X (, x). It makes sense to split the fluctuational part, X , into the compo
nents normal and tangent to the circle (3.10) at the point X0 . The components must satisfy the
Dirichlet and the Neumann BC, respectively. Therefore the microscopic boundary degeneracy
for g0 1 does not depend on and coincides with the product of the boundary degeneracies
C
associated with the Dirichlet (gD ) and the Neumann CBC (gN = gD 2
, where C = 2/ g0
is the full length of the circle (3.10)). The above consideration supplemented by the common
wisdom of the renormalization group suggests that as E T , the partition function must
develop the following leading high-temperature behavior:

gfixed
1 + O g(T ) .
Z =
g(T )
(3.17)
In writing the above equation, we replace the bare coupling g0 in the microscopic boundary
degeneracy g2D / g0 by the running coupling constant g(T ). As follows from the two-loop
-function (3.11), g = g(T ) can be defined by the equation:
E
1
= g 1 e 2g .
T
Also, the factor gfixed in (3.17) should be understood as
(3.18)
gfixed = g2D .
(3.19)
The exact partition function (3.16) indeed possesses the leading short distance behavior (3.17).
The systematic high-temperature expansion for = 0 was obtained in Ref. [20]. That result can
8 Notice that the instanton cut-off is different from the perturbative cut-off in the RG flow equation (3.11), i.e.
their ratio /
is some non-universal function of the bare coupling g0 . In Section 4 we find this ratio for some particular
UV regularization of the circular brane model.
be easily generalized to = 0, yielding an expansion of the form

gfixed 2 cos cos n

Z
g
zn (g, ),
g T
167
(3.20)
n=0
where the coefficients zn (g, ) are power series in the running coupling constant (3.18). To describe some coefficients in (3.20) explicitly, it is convenient to represent Z as
gfixed
Z = e T cos Z .
g
(3.21)

= T 2 log(Z )=0 ,
(3.22)
Here
and therefore the factor Z is subject to the constraint

2 Z
= 0.
2 =0
It is possible to show that the topological susceptibility expands explicitly as

E 2 log
log g + 3E + log 2 2E g + O g 2
T

2

2
1 + O(g) + O 2 .
3
(3.23)
(3.24)
The expansion coefficients of Z (3.21) are somewhat simpler than zn in (3.20). For example,

Z 0 = 1 (1 + E )g + O g 2 , 2 ,
(3.25)
and

4 2
2
Z /Z 0 = 1
1 + O g 2 sin4
+ O 3 .
3
2
2
(3.26)
Notice that the coefficients 23 and 43 in Eqs. (3.24), (3.26) are in agreement with the result of
two-instanton calculation from Ref. [46].
3.4. Low-temperature expansion
For k = , the boundary degeneracy gfixed in (2.63) should be understood as in Eq. (3.19).
It implies that the circular brane boundary flow is terminated at the fixed point corresponding to
the Dirichlet CBC XB = 0 for any within the domain 0 < . At the same time, for =
the flow possesses a non-trivial IR fixed point whose boundary degeneracy is given by
g2 = 2g2D
(3.27)
(see Eq. (2.64) with s = 2 and k = ). The last equation suggests that the boundary values of X
are constrained to two points. We refer the reader to Ref. [33] for a comprehensive description
of this non-trivial RG fixed point.
168
Here we describe the low-temperature expansion for = . In this case the free energy, F =
T log Z , admits an asymptotic expansion in terms of T 2 :

T 2l
Fl ( ) 2l1 .
F 2E cos( ) log
(3.28)
+ E T log g2D
E
E
l=1
The regularized ground state energy E is a large-k limit of the function (D.3). It explicitly reads
E /E = C0 (E + log 2) cos

(x + 12 )
3
1 1
2
2
+ dx cos ( )
3 F2 x + , , 1; 2, cos ( )
(x)
2 2
2
0

5
2
cos3 ( )x 3 F2 x + 1, 1, 1; 2, cos2 ( ) .
3
2
(3.29)
Here 3 F2 is the generalized hypergeometric function and
(x 12 )
= 1.44142 . . . .
C0 = dx
(x)
(3.30)
The function (3.29) is plotted in Fig. 4. Notice that it develops a non-analytical behavior at = :

( )

+O
E /E = 2 +
(3.31)
.
log( )
log2 ( )
The first two coefficients Fl in the expansion (3.28) explicitly read as follows:
F1 ( ) =
,
12 sin
F2 ( ) =
(3.32)
and
720 sin3

sin
sin
dt 6 log2
7 log
5 .
sin t
sin t
(3.33)
4. The dissipative quantum rotator model

The circular brane model has useful interpretation in terms of Brownian dynamics of a quantum rotator. It was noticed a while ago in Ref. [47] that the free massless bulk dynamics (3.9) is
equivalent to the CaldeiraLeggett model of a quantum thermostat [48]. Upon fixing the bound
ary values XB ( ) and integrating out the bulk part of the field X (, x), Eqs. (3.13), (3.14)
reduce to
Z = gfixed D eAdiss []A [] ,

(4.1)
where
1
T2
Adiss [] =
2g0
T
d
d
0
)
sin2 ( ( )(
)
2
sin2 (T (
))
(4.2)
169
Fig. 4. The regularized ground state energy (3.29) in the circular brane model.
and
1
A [] =
i
2
d ( ).
(4.3)
The variable ( ) is the angular field defined through

1
(XB , YB ) = (cos , sin ).
g0
(4.4)
The path integral (4.1) requires some UV regularization. Among a large variety of regularizations
there is one of special interest for applications [21,49]:
ADQR [] = Adiss [] + A [] + AC [],
(4.5)
where
1
1
AC [] =
4EC
d 2 .
(4.6)
The functional (4.5) amounts to the CaldeiraLeggett action for the dissipative quantum rotator.
In the weak coupling regime (g0 1) the additional term (4.6) just provides an explicit UV
cutoff of the dissipative action (4.2) with the cutoff energy EC /g0 . Now, once the regularization scheme is chosen, all coefficients cl in the asymptotic expansion (3.12) are determined
unambiguously within the standard perturbation theory, in particular [20]
3
c1 = 2 .
(4.7)
4
The additional term (4.6) not only makes the circular brane model perturbatively well-defined,
but also regularizes the small instanton divergence. For example, the one-instanton contribution
to the topological susceptibility (3.22) in the Gaussian approximation reads as follows [41,50]:
(1)

da da eAC
EC 2g1
(pert)
0 2
=
e
+
O(g
)
.
(4.8)
0
2i 1 aa
2 2 g02
aa <1
170

(1)
Here AC is the functional (4.6) evaluated for the one instanton solution. The exponential factor
(1)
AC
makes the integral over the instanton moduli |a| < 1 finite. Indeed, using the explicit form
e
of the one instanton solution (1) [44,45],

za
exp i(1) ( ) =
(4.9)
, where z = e2T i ,
1 az
one finds
(1)
AC =
2 T 1 + aa
.
EC 1 aa
Now we can integrate over the instanton moduli explicitly:

EC 2g1
EC eE
(pert
)=
e 0 2 log
+ O(g0 , T /EC ) .
2 2 T
2 2 g02
(4.10)
(4.11)
It is instructive to compare (4.11) with the high-temperature expansion (3.22). In essence,

Eq. (3.22) requires that the bare coupling constant expansion should be of the form

(pert) = E 2 log
(4.12)
log g0 + 3E + log 2 + O(g0 , T /) .
T
Combining Eqs. (4.11) and (4.12), (3.12) yields the following relations between the perturbative
and the energy scale EC in (4.6):
cut-off (), the small instanton cut-off (),
=
EC
,
2 2 g0
(4.13)
and
5E
e 2
g0 .
= EC
8 2
(4.14)
5. Conclusion: Charge fluctuations in quantum dot

The dissipative rotator model (4.5) was introduced in Ref. [21] as an effective field theory
describing tunneling of quasiparticles between superconductors. Later it has been involved in
a surge of activity in the study of a low-capacitance metallic island (quantum dot), connected
to an outside lead by a tunnel junction [38,41,45,46,5052]. A remarkable feature of such a
device (see Fig. 5, borrowed from Ref. [52]) is that the essentially many-body phenomena may
be described via a single variable (conjugate to the number of excess charges on the island).
As was shown in Refs. [21,45], the fermionic degrees of freedom can be formally integrated out
yielding the Matsubara effective action Aeff for . The Coulomb part of this action coincides
with A + AC (4.3), (4.6) provided the energy scale EC is understood now as the single electron
charging energy, i.e.,
e2
,
2C
where C is the island capacitance. The parameter
EC =
ng =
(5.1)
(5.2)
171
Fig. 5. Schematic view of a quantum dot connected to a bulk 2D electrode. The dot is formed by applying a negative
voltage to the gates. Electrostatic conditions in the dot are controlled by the gate voltage Vg . Voltage Va applied to the
auxiliary gates controls the transmission through the constriction.
has also the simple physical meaning of a dimensionless applied gate voltage Vg : ng = Cg Vg /e,
where Cg is the gate capacitance (see Fig. 5).
In addition to the Coulomb terms, the effective action contains a highly non-trivial dissipative
piece which essentially depends on details of the islandlead interaction. In fact, the transmission
coefficients {Ta }N
a=1 in the islandlead interface are coupling constants of Aeff (N is a number
of spinless electron channels). The most general effective action appears to be too complicated
for practical purposes. It simplifies drastically in the special large-N limit when each individual
transmission coefficient Ta vanishes, but the dimensionless (measured in units e2 / h) tunneling
conductance
=
N
1
Ta ,
2 2
(5.3)
a=1
remains finite. In this limit Aeff takes the form of the dissipative rotator action (4.5) with
1
,
(5.4)
2 2
so the basic thermodynamic quantities can be extracted from the corresponding partition function, ZDQR . The renormalized charging energy E C ,
g0 =
E C = 2 2 T 2 log(ZDQR )
(5.5)
is of special interest. As it has been argued above, the quantum dissipative rotator model possesses the universal scaling behavior provided g0 1 and EC T . This corresponds to the
regime of almost perfect transmission for the low-capacitance metallic island. In the scaling
regime ZDQR turns into the circular brane partition function Z . Thus the exact results obtained
in this work give a complete description of the renormalized charging energy in the scaling limit.
In particular at T = 0,

5
EC e 2 E
2 E
E C
2
cos(
)
log
(, EC ).
+
(5.6)
2 2 E T =0
2 E
4 3 E
Recall that the function E /E is given by Eq. (3.29) and plotted in Fig. 4, while

2 2
3
2
2
e
.
+O
E = 2 EC 1
8
(5.7)
172
It is worth noting that the quoted analytical results for E C with = 0 have been numerically
checked in Ref. [53] via Monte Carlo simulations of ZDQR .
The quantitative description of the quantum dot with finite number of channels remains a
challenging problem. For finite N , one would still expect some sort of universal scaling behavior
characterized by the energy scale [45]
E E C
N

1 Ta .
(5.8)
a=1
In the case of N = 2 this was explicitly demonstrated in Ref. [52]. Matveev argued that the uni(k)
versal behavior of the two-channel quantum dot is described by the Hamiltonian H (2.8) with
k = 2. Thus the thermodynamic properties of the system can be extracted from the Chatergee
Zamolodchikov partition function (2.61). With regard to N = 2 and N = integrable cases,
it would be interesting to explore the possibility of using H (k)
(2.8) as a QFT Hamiltonian underlying the universal scaling behavior of the N -channel quantum dot provided N = k.
Acknowledgements
I am deeply indebted to Alexander B. Zamolodchikov for teaching me QFT, previous collaboration and interest in this work. I also would like to thank Alexei B. Zamolodchikov for
sharing insights from the unpublished work [19], and V.V. Bazhanov, G.Y. Chitov, V.A. Fateev,
A. Leclair, A.M. Tsvelik and P. Werner for interesting discussions and important comments.
This research is supported in part by DOE grant #DE-FG02-96 ER 40949.
Appendix A. The BP sine-Gordon model
In this appendix we shall study the BP sine-Gordon model. It provides an essential step toward
the main result (2.53).
A.1. Definition of the model
Let H free be a Hamiltonian of the boundary CFT model which contains two non-interacting
sectors, namely the minimal parafermionic CFT and the Gaussian theory of the free uncompactified Bose field:
()
H free = H (k)
free + H free .
(A.1)
Suppose the parameter b in Eq. (2.21) is purely imaginary,

b = i,
(A.2)
so that the boundary fields (2.17) take the form:

i B
V = e
(A.3)
Their scaling dimension is given by

d =1
1 2
+ .
k
k
(A.4)
173
For 0 < 2 < 1, V are relevant boundary fields normalized as in (2.18), and in what follows we
will focus on the model described by the Hamiltonian

(k)
H bsg = H free eiV t V + + eiV t V ,
(A.5)
where and V are parameters. The parameter carries the dimension of [length]d1 . The
boundary interaction contains an explicit time dependence through the factors eiV t . Below we
always assume that
V > 0. Notice that, although by means of the formal change of the field
variable, k t , the Hamiltonian (A.5) can be brought to the form (2.14), effects of the
external fields V in (A.5) and h in (2.14) are very different. Whereas the system (2.14) possesses
the thermal equilibrium state even for non-zero values of h, the system (A.5), at a non-zero V
and a temperature T , develops a non-equilibrium steady state which can be thought of as the
result of an infinite time evolution of the equilibrium state of the corresponding free system,
(k)
with the interaction term adiabatically switched on. We denote by Obsg the expectation value
of an observable O over this non-equilibrium steady state.
The model (A.5) was already discussed in the literature. In particular in the absence of the
parafermionic sector (k = 1) it coincides with the massless boundary sine-Gordon model [54,
55]. For this reason we shall call (A.5) the BP sine-Gordon model. Besides being an interesting
model of Quantum Field Theory on its own, the theory finds important applications in condensed
matter physics. In Ref. [27] the model (A.5) was suggested to describe a point-like impurity in the
multichannel quantum wire.9 In this context the integer k is the number of spinless electron
channels, 0 = k 2 is the conductance (in units e2 / h) of the wire without impurity, while V
is proportional to the voltage drop across the impurity. The quantities of special interest are
correlation functions of the Heisenberg boundary operators
V (t) = eiV t U 1 (t)V U (t),
(A.6)
where U (t) is the time evolution operator corresponding to the Hamiltonian (A.5). For instance,
the energy dissipation rate E for the steady state, produced by the time dependent boundary per(k)
turbation (A.5), can be expressed in terms of the expectation values V bsg . Indeed, as follows
from (A.5):

(k)
(k)
E = JD V with JD = i V + bsg V bsg .
(A.7)
Notice that the quantity

2
(V , T ) = 0 1 + 2 JD /V
k
(A.8)
is interpreted as a DC conductance of the multichannel quantum wire [27].

(k)
A.2. Perturbative expansions for V bsg

If V > 0 the system (A.5) evolves towards a steady state which is characterized by the ex(k)
pectation values V bsg . This is not a thermodynamic equilibrium state and no simple explicit
expression for its density matrix is known. Nevertheless, one can describe the definition of this
9 The cases k = 1 (without the parafermionic sector) and k = 2 were previously studied in Refs. [5658] and [59,60],
respectively.
174
state in terms of the real-time perturbation theory [61]. It is useful to represent the perturbative
(k)
expansions for V bsg in the form
T
log A ,
2
where log A are formal power series expansions,
(k)
V bsg =
log A =
(A.9)
with s = 2 (2T )2d2 .
an() s n
(A.10)
n=1
()
For k = 1 the compact formula for the perturbative coefficients an was obtained in Refs.[25,62].
In essence the derivation from [25] can be applied without change to the BP sine-Gordon model.
As a result, the following formula emerges:
an() = (1)n+1 22n
2n1

sin(dj )J,1 ,...,2n1 ,
(A.11)
1 ,...,2n1 j =1
where the sum

is taken over all arrangements of the charges 1 , . . . , 2n1 = 1 with zero total
charge + 2n1
s=1 s = 0, and
j = +
j 1

(A.12)
s .
s=1
The coefficients J,1 ,...,2n1 in (A.11) are expressed through the (2n 1)-fold integrals over the
real-time correlation functions:
0
J,1 ,...,2j =
t1
dt1
2n2
dt2n1 eiV
dt2
2n1
j =1
j tj
(2T )2n(1d) eid
2n1
j =1
j j

V (int) (0)V (int)
(t1 ) V (int)
(t
) .
1
2n1 2n1 0
(A.13)
Here V (t) = eitH free V eitH free are the boundary operators in the interaction representation
and
H free

H free

0 TrH e T / TrH e T
(A.14)
(int)
denotes an expectation value over the equilibrium thermal state of the free system.
As follows from the real-time OPE

2d
(int)
(int)
, 0 < t t
0,
V (t)V (t
) i(t t
)
(A.15)
the integrals (A.13) can be taken literally only if the scaling dimension d < 12 . For 12 < d < 1 we
assume the analytic regularization common in conformal perturbation theory, i.e. the integrals
should be understood as analytically continued from the convergence domain e d < 12 . For the
()
sake of illustration we consider explicitly the first coefficient a1 . In this case the correlators
appearing in the integrand of (A.13) are given by

2d
(int)

T
(int)
(t1 < 0).
eid V (0)V (t1 ) 0 =
(A.16)
sinh(T t1 )
175
Calculating the elementary integral over t1 , one obtains

()
a1
= 4 sin(d)
(1 2d)(d 2p)
(1 d 2p)
with p = i
V
.
4T
(A.17)
A.3. The system with q-oscillator

Remarkably, the formal power series (A.10) can be interpreted as high-temperature expansions of the equilibrium free energy for certain systems which differ from (A.5) in that they
involve additional boundary degrees of freedom. Originally this was demonstrated in Ref. [25]
for the boundary sine-Gordon model (k = 1), but the same reasoning can be applied to the generalized model (A.5).
Let us define the Hamiltonian
H + = H free V h (a V + + a + V ).
(A.18)
Here we use the same notations as in (A.5). Formula (A.18) includes a new ingredient, namely,
the operators h, a + , a which commute with all the parafermionic and bosonic degrees of freedom. They form among themselves the so-called q-oscillator algebra,
qa + a q 1 a a + = q q 1 ,
[h, a ] = a ,
(A.19)
with
q = eid .
(A.20)
Recall that the parameter 1 k1 < d < 1 is the scaling dimension of the boundary fields V (A.4).
Let + be some representation of (A.19) such that the spectrum of + (h) is real and bounded
from above. The Hamiltonian (A.18) acts in the space
H+ = H + ,
(A.21)
with H given by (2.16). For V > 0 this Hamiltonian is bounded from below. Then, the system (A.18) possesses a thermal equilibrium state described by the standard density matrix,
H+
1
(, V )e T .
Z+
Strictly speaking the partition function
H+
Z+ (, V ) = TrH+ e T
(A.22)
is ill-defined even for the non-interacting system. Indeed, since the space of states H+ has a structure of tensor product (see Eqs. (2.16), (A.21)), the partition function Z+ (, V )|=0 factorizes
into a product of three terms. The first one coincides with the boundary degeneracy gfree (1.2).
Vh
The component + in (A.21) gives rise to the factor Tr+ [e T ] which is well-defined if the spectrum of + (h) is bounded from above. The problem comes from a divergent contribution of the
zero-mode of the uncompactified Bose field subjected by the free CBC. To cancel this formal
divergence we shall consider below the ratio Z+ (, V )/Z+ (0, V ).
Carrying out the finite temperature Matsubara procedure, one obtains the weak coupling expansion of the form

Z+ (, V )
n
A(+)
=1+
n s
Z+ (0, V )
n=1
with s = 2 (2T )2d2 ,
(A.23)
176
where the expansion coefficients are given by

T1 2n G1 2n .
A(+)
n =
(A.24)
k =
1 ++2n =0
Whereas T1 2n are expressed in terms of traces over the representation + ,
T1 2n =
Tr+ [e
Vh
T
a 1 a 2n ]
Tr+ [e
Vh
T
(A.25)
the coefficients G1 2n are multiple integrals:

1
T
G1 ...2n =
d2n eV
d2
d1
0
2n1
2n
j =1 j j

(2T )2n(1d) V (Mats)
(1 ) V (Mats)
(2n ) 0 .
1
2n
(A.26)
( ) = e H free V e H free are unperturbed Matsubara operators associated with the

Here V
boundary operators (A.3), and 0 is defined in (A.14). As well as in formula (A.13), we
should assume the analytic regularization of G1 2n in the case 12 < d < 1. One might also note
that all the coefficients T1 2j in (A.24) are determined unambiguously via the commutation
relations (A.19) and the cyclic property of trace. Thus, they do not depend on a particular choice
Vh
of the representation + provided Tr+ [e T ] exist.
Now, following along the lines of Ref. [25], it is possible to prove that the formal power series
A+ defined in Eq. (A.10) coincides with (A.23), i.e.,
(Mats)
A+ =
Z+ (, V )
.
Z+ (0, V )
(A.27)
Of course, A (A.10) is also related to a certain equilibrium-state partition function. Namely,

let be any representation of (A.19) such that the spectrum of (h) is bounded from below.
Consider the Hamiltonian
H = H free + V h (a + V + + a V )
(A.28)
acting in H = H and the associated partition function

H
Z (, V ) = TrH e T .
(A.29)
Similarly to (A.27), one can prove the relation

A =
Z (, V )
.
Z (0, V )
(A.30)
A.4. Characteristic properties of A

Eq. (A.10) shows that log A are formal power series of the dimensionless variable
s = 2 (2T )2d2 ,
(A.31)
177
()
while the expansion coefficients an depend on the dimensionless ratio V /T . In what follows,
()
we treat an as functions of the complex variable
V
.
4T
Here are the main properties of A (s, p).
p = i
(A.32)
log A (s, p) are formal power series of the form (A.10), or, equivalently,
A (s, p) = 1 +
n
A()
n (p)s .
(A.33)
n=1
()
()
The coefficients an (p) (A.10) and An (p) (A.33) are meromorphic functions in the whole
complex plane of p and
(+)
A()
n (p) = An (p),
Both
(+)
an (p)
and
e p > pn
()
An (p)
an() (p) = an(+) (p).
(A.34)
are analytic in the right half-plane,
with 2pn = (1 d)n 1.
For n = 1, 2, . . . , k, they have a simple pole located at p = pn of residue

Resp=pn An (p) = Resp=pn an (p)

2n
k!(1 n(1 d))
2
.
=
2n!(k n)!(n(1 d))
k(d)
(A.35)
(A.36)
There is no singularity at p = pn for n > k, therefore the domain of analyticity (A.35) can
be extended to the half-plane e p > pn n with some finite n > 0.
The following leading power-low asymptotic holds in the domain of analyticity,
an(+) (p) Cn p 12n(1d)
as p (e p > pn ),
(A.37)
where Cn is some constant.

The formal power series A (s, p) obey the so-called quantum Wronskian relation,

e2ip A+ (qs, p)A q 1 s, p e2ip A+ q 1 s, p A (qs, p) = 2i sin(2p),
(A.38)
with q
= eid .
The proof of functional relation (A.38) is based on results of the work [24]. It is outlined
in Appendix B. All other properties readily follow from formulas (A.11) and (A.23). For example the integrals (A.26) are entire functions of the variable p (A.32). Thus, singularities of
the coefficients A(+)
n (p) are due to the traces over the representation + . A brief inspection of
(+)
(A.25) shows that An (p) possesses only simple poles. They may be located only at points
(+)
2p = (d(n + l) + m) where l = 0, 1, . . . , n 1 and m = 0, 1, 2, . . . . Hence, An are meromorphic functions in the whole complex plane of p. The same, of course, holds true for the
(+)
coefficients an (p). Furthermore, since (A.34) holds trivially for pure imaginary p, the claim is
also true for any complex p.
178
(+)
Fig. 6. The amplitude defining the boundary pn of the domain of analyticity e p > pn for an (p) (n k). The
(int)
(int)
incoming and outgoing legs correspond to the insertions of boundary fields V (t) and V + (t), respectively. For
n k the intermediate state carries the scaling dimension (A.41) and pn = d2nn .
(+)
The analyticity of an (p) in the right half-plane e p > pn for some pn and the asymptotic
formula (A.37) follow immediately from Eqs. (A.11), (A.13). To determine pn and calculate
the residue one needs to examine these equations in more details. It is possible to show that
the boundary of the domain of analyticity is defined by the integral J+,1 2n1 (A.13) whose
cumulative charges j (A.12) have maximum admissible values,10 i.e.
1 = = n1 = +1,
n = = 2n1 = 1.
(A.39)
Moreover, as p pn , the dominant contribution to this multiple integral comes from the re(int)
gion of integration where the operators V j with the same charges j combine into two
well-separated clusters: |t1 |T |tn1 |T 1 and |tn |T |t2n1 |T 1. Therefore,
to determine the singular behavior as p pn we invoke the cluster property and replace the
correlation function in the integrand (A.13) by
(int)

(int)
(int)
(int)
eind V + (t0 ) V + (tn1 )V (tn ) V (t2n1 ) 0
F (t0 , . . . , tn1 )e2T dn (tn t0 ) F (tn , . . . , t2n1 ),
where t0 = 0

dn = n n(d 1) + 1
(A.40)
and11
(A.41)
n
i
k B
is the scaling dimension of the boundary operator n e

(see Fig. 6). The latter is a product of
the bosonic exponential field and the boundary field n (2.2). The correlation function appearing
in (A.40),
n
(int)

(n) i k B
(t0 ) V (int)
() 0 ,
F (t0 , . . . , tn1 ) = (2T )dn V
(A.42)
(tn1 ) B e
has a very simple form in the case n k. Namely, for t1 > > tn
F (t1 , . . . , tn ) = (2T )nd
n1

m=1
C1m

2d2
2i sinh T (tm tj )
.
(A.43)
m<j
10 Note that because of the factor 2n1 sin(d ) the sum (A.11) enjoys a remarkable property, namely, it contains
j
j =1
only the terms where none of the cumulative charges j (j = 1, . . . , 2n 1) vanish.
11 Here it is assumed that n k.
179
Here C1m = (k m)(m + 1)/k are the structure constants (2.5) with j = 1. Combining (A.40)
with (A.11) one finds that the coefficients an (p) (n = 1, . . . , k) possess a simple pole singularity
dn
at p = pn = 2n
and
n1

22n1 k!n!
sin(nd)S 2
Resp=pn an(+) (p) = (1)n+1 2 n
sin2 (j d),
n k (k n)!
(A.44)
j =1
where S is the Selberg integral [63,64],
0
S=
u1
du2
du1
= (1)1+

ul uj 2d2
du2n1
2 sinh
2
u
2n2
n(n+1)
2
n
(d)
l<j
(1 n(1 d))
.

(n 1)! n1
j =1 sin(j d)
(A.45)
Plugging (A.45) into (A.44) one arrives at Eq. (A.36).

(+)
Finally it is easy to see that for n > k there is no pole at p = pn and an (p) is analytic in
some half-plane e p > pn n where n is a finite positive constant.
A.5. Uniqueness of the solution of the RiemannHilbert problem
The above-described properties of the formal power series expansions A (s, p) constitute a
certain RiemannHilbert problem defining these series uniquely. For k = 1 this was shown in
Appendix of the work [25]. Here we generalize the result of [25] to the case k > 1.
Let us substitute the formal power series (A.10) into the quantum Wronskian relation (A.38).
This yields an infinite set of relations for the meromorphic functions an(+) (p) of the following
form
sin(nd + 2p)an(+) (p) sin(nd 2p)an(+) (p) = Rn (p),
(+)
where Rn (p) are expressed through al
(A.46)
(p) with l = 1, . . . , n 1 only. For example,
R1 (p) = 0,
(+)
2
(+)
R2 (p) = qa1 (p) + q 1 a1 (p) e4ip sin(2p)/2.
(A.47)
Introduce the function rn (p) such that

rn (p) =
(n 1 nd + 2p) (+)
a (p),
(2 n + nd + 2p) n
(A.48)
then (A.46) takes the form

(1)n1
(A.49)
(n 1 nd + 2p)(n 1 nd 2p)Rn (p).
In the case n = 1, the RHS in (A.49) vanishes and this equation supplemented by the analyticity condition (A.35) implies that r1 (p) is an even entire function of the variable p. Moreover,
in view of the asymptotic condition (A.37), one should conclude that r1 = const. The constant is
defined by the residue condition (A.36). This yields formula (A.17).
rn (p) rn (p) =
180
We shall prove the uniqueness of the solution of (A.46) for n > 1 by using induction. Let us
(+)
assume all the functions am (p) for m = 1, . . . , n 1 are already determined, so the LHSs of
Eqs. (A.46) and (A.49) are given. Suppose also for the moment that the complex parameter d
n1
n1
belongs to the strip n1
n < e d < n , where > 0 is some small number. If e( n d)
is sufficiently small, then rn (p) is analytic in the half-plane e p 0 except for a simple pole
at p = pn 12 (n(1 d) 1). The residue of rn (p) at this pole is fixed by the condition (A.36).
Also, it is easy to see that rn (p) 0 as p within e p > 0. Hence, relation (A.49) supplemented by the above quoted requirements for rn (p) constitutes a simple factorization problem.
The uniqueness of this RiemannHilbert problem follows immediately from the Liouville theorem. To complete the induction step we note that the function rn (p) within e d < 1 can be
n1
obtained via the analytic continuation in the variable d from the strip n1
n < e d < n .
(+)
Thus we see that (A.46) provides a recursion for the evaluation of an (p) which allows one
(+)
(+)
(+)
to express an (p) in terms of an (n 1)-fold integral. Explicit formulas for a2 and a3 in
the case k = 1 can be found in Appendix of [25]. Those formulas can be easily generalized to
k > 1. It is worth noting that such integral representations prove to be convenient for numerical
calculations because the integrals converge very fast at infinity.
A.6. Exact expressions for A
It was observed some years ago in the work [65], that A (s, p) for k = 1 and some particular
values of p can be related exactly to the eigenvalue problem of a certain ordinary differential
operator. This remarkable observation was generalized and proven for any p in Ref. [26]. Here
we extend the result of [26] to all integers k > 1.
Let us consider the Schrdinger equation

2

k

u
u
2
2
2
u + exp
(A.50)
(u) = 0.
+ exp
1 2
1 2
For 0 < 2 < 1 the potential term decays at u and therefore for real > 0 there is a Jost
solution which is asymptotically plane wave as u :
(u, ) ei u .
(A.51)
As a matter of fact, this particular solution is a meromorphic function in the whole complex
plane of the parameter , so it solves (A.50) for all values except points where (u, ) has
simple poles (see, e.g., [66]). Notice that (u, ) is another, linearly independent (for = 0)
solution of (A.50).
Since the potential in (A.50) grows rapidly as u +, the Schrdinger equation admits
also a solution which decays at large positive u. We denote this solution by + (u, ) and fix its
normalization by the condition
1
+ (u, ) (2) 2 eF (i|u)
as u +,
(A.52)
2
where F (b|u) is defined in Eq. (2.35). + (u, ) is an entire function of and solves differential
equation (A.50) for all values of this complex parameter.
The main objects of our interest are properly normalized Wronskians of the above-introduced
solutions, namely,

2i(1d)
2(1 d)
W + (u, ), (u, ) .
(1 d)
D (, ) =
(A.53)
(1 2i(1 d))
181
Here d is given by (A.4) and the overall factor provides the normalization condition
lim D (, ) = 1.
(A.54)
In what follows we will show that for 0 < 2 < 1 and all values of and ,
A (s, p) = D (, ),
(A.55)
provided the variables are identified according to the relations:

1

ip
1
2 s 1d
=
(A.56)
.
,
=
1d
1d
k(d)
More specifically, we will check that D considered as functions of the variables s and p, obey
the same set of conditions (33)(37) as A (s, p). Thus (A.55) will follow from the uniqueness
of the solution of the RiemannHilbert problem.
In order to examine properties of D it is useful to make a change of the variable in (A.50),

u
+ log 2(1 d) ,
z=
(A.57)
2 2d
and rewrite the differential equation using the notations s, p (A.56) and d (A.4):
2

z + e2z + 4p 2 + U (z) (z)

(A.58)
= 0,
with
2d 2m
k!
2
s m e2z(1(1d)m) .
U (z) =
m!(k m)!
k(d)
m=1
k

(A.59)
The solutions (A.51), (A.52) should be considered now as functions of the complex
parameters s, p provided 1 k1 < d < 1 and z is real. In fact, it is convenient for the present
discussion to change slightly their normalizations:

2i(1d)
(z, p) = 2(1 d)
(u, ),
1
+ (z, p) =
(A.60)
+ (u, ).
1d
Then formula (A.53) takes the form:

22p
D (s, p) =
(A.61)
W + (z, p), (z, p) .
2 (1 2p)
Both + and are entire functions of s. In addition, is a meromorphic function of
p, while + is an even entire function of this complex parameter. Hence D (s, p) can be
()
expanded in a power series of s similar to (A.33), and the expansion coefficients, Dn (p):
Dn() (p) = Dn(+) (p), are meromorphic functions of p.
To proceed further we will need the following conventional formula for D+ (s, p), valid over
the domain e p 0,
1
=1
D+ (s, p)

dz I2p ez U (z)(z).
(A.62)
182
Here Ip (x) is the modified Bessel function and (z) solves the LipmanSchwinger equation,

(z) = K2p ez
dz
G(z, z
)U (z
)(z
),
(A.63)
G(z, z
)
is the Green function of (A.58) with U = 0 subject of the asymptotic conwhere

dition limz G(z, z
) = 0. Notice that (z) differs from the solution + (z, p) by some
(+)
overall z-independent factor only. The coefficients Dn (p) can be calculated within stan(+)
dard perturbation theory developed with respect to U . In fact, to determine Dn (p) for
given n, one needs to perform (n 1)-perturbative iterations in the LipmanSchwinger equation. In particular,
D+ = 1 +

dz I2p ez K2p ez U (z) + O (U )2 .
(A.64)
Making use of (A.59) and the table integral
dx x 21 I2p (x)K2p (x) =

0
( 12 )()( + 2p)
,
4 (1 + 2p)
one finds for n k:

2d 2n

2
1
(+)
Dn (p) =
+ n(1 d)
2
k(d)
k!
(1 n(1 d))(1 n(1 d) + 2p)
+ .
n!(k n)!
4 (n(1 d) + 2p)
(A.65)
(A.66)
Here the dots mean the terms corresponding to the higher-order perturbative in V contributions. Further, one may argue that the first-order contribution (A.66) is analytic in the
half-plane m(p) > pn = 12 (n(1 d) 1), whereas the omitted terms have a wider domain
(+)
of analyticity. Hence, Dn (p) with n k is an analytic function for m(p) > pn and has
a simple pole located at p = pn with the same residue as in (A.36). Similarly, it is possible
(+)
to show that Dn (p) with n > k is analytic in the half-plane m(p) > pn n with some
n > 0.
The large p behavior of the expansion coefficients dn(+) (p) for the series log Dn(+) (s, p)
can be explored by means of standard semiclassical methods. A simple WKB analysis
(+)
of the Schrdinger equation (A.58) shows that dn (p) Cn p 12n(1d) as p and
m p > pn , with the constants Cn given by

2n
2
Cn = (1)n+1 k
k(d)
(n (1 d)kn)(1 (1 d)n)( 12 + (1 d)n)

.
2 n!(1 (1 d)kn)
(A.67)
Finally, let us prove that D (s, p) (A.61) obey the quantum Wronskian relation. We start
with an observation that the following transformations of the variable z and the parameters
183
s and p,
z z,
z z + i,
s s,
p p,
s q 2 s,
(A.68)
p p,
leave the differential equation (A.58) unchanged while acting non-trivially on its solutions.
It has been mentioned already, that the transformation applied to the solution (z, p)
(A.60) yields another solution, and the pair
= (z, p),
= (z, p),
(A.69)
forms a basis in the space of solutions of (A.58). Indeed,

W [ , ] = 4p,
(A.70)
therefore the solutions (A.69) are linearly independent provided p = 0. The solution +
(A.60) can be always expanded in the basis (A.69),
+ = c1 + c2 .
(A.71)
Making use of (A.70) we have

c2 =
1
W (s, p),
4p
(A.72)
with W (s, p) = W [ + , ]. The solution + is an even function of p, therefore

c1 =
1
W (s, p).
4p
(A.73)
Let us apply now the transformation (A.68) to both sides of (A.71). Since
(z, p) = e2ip (z, p), so
+ =

1 2ip 2
W sq , p e2ip W sq 2 , p .
e
4p
(A.74)
Now, it is not difficult to check that

Wz [ + , + ] = 2i.
(A.75)
Combining formulae (A.71), (A.74) and (A.75) one obtains the relation

e2ip W (s, p)W q 2 s, p e2ip W q 2 s, p W (s, p) = 8ip.
(A.76)
Hence D (s, p) (A.61) satisfy the quantum Wronskian relation (A.38) with A (s, p) replaced by D (s, p).
Notice that once the relation (A.55) is established we may assert that series (A.23), (A.33)
have an infinite radius of convergence.
184
A.7. Formula for the DC conductance

Since the parafermionic boundary sine-Gordon model (A.5) has been of a certain interest in
the quantum impurity problem [27], we present here an exact formula for the DC conductance
(A.8) which follows immediately from the above consideration.
Using Eq. (A.9), the conductance can be written in the form

s
A+
s log
.
(V , T )/0 = 1 + 2
(A.77)
2pk
A
This formula has a nice interpretation in terms of the Schrdinger problem (A.50). Indeed, as
v , the solution + (v, ) (A.52) develops the asymptotic behavior:

+ (u, ) const ei u + S(, )ei u ,
(A.78)
where
S(, ) =
W [+ (u, ), (u, )]
W [+ (u, ), (u, )]
(A.79)
is the reflection scattering amplitude for the Schrdinger equation (A.50). Therefore, in view of
(A.53) and (A.55), the DC conductance is expressed in terms of S(, ) as follows,
(V , T )/0 = 1 +
k
4i( 1
)2
log S(, ),
V /T =

4
1 2 .
k
(A.80)
Recall that 2 = 0 /k is the dimensionless conductance per channel of the wire without impurity,
while in (A.80) is the dimensionless inverse temperature measured in unit of the Kondo scale:
k0
= E /T , E k2 .
Formula (A.80) extends the result of [26] to the case k > 1. We note also that as T 0, the
(+)
dimensionless variable p = iV /(4T ) . In this limit, the coefficients an (p) develop the
simple asymptotic behavior (A.37) where the Cn are given by (A.67). This immediately yields
the large-V expansion of the conductance at T = 0 proposed in Ref. [27]:

2n

(V , 0)
n (n + 1 kn )(1 n ) 4E
=1+
(1)
,
(A.81)
0
2
V
n!(1 kn )( 32 n )
n=1
where =
k0
.
k2
Appendix B. Quantum Wronskian relation

The purpose of this appendix is twofold. First, it is intended to sketch how the quantum Wronskian relation (A.38) can be proven for the perturbative series A (A.27), (A.30). Second, it
provides a short review of some well-known issues of CFT which are useful for understanding
the models under consideration.
Our previous discussion has always dealt with the so-called open string picture (channel).
In this Hamiltonian picture the variable x 0 plays the role of the space coordinate, while the
coordinate is treated as the compact Euclidean time. We now want to switch over to the closed
string channel. In this channel the world-sheet coordinate x is viewed as the Euclidean time, so
that the CFT lives in the finite volume 1/T . The closed string channel Hilbert space is embedded
in a tensor product of the chiral spaces of states of right and left movers. Below the following
185
notational convention is applied: If the open string Hilbert space was denoted by H, then the
right and left movers spaces of states in the closed string channel will be denoted (with some
abuse of notations) as H and H respectively. Of course, the spaces H and H are isomorphic to
each other. We will focus mainly on the right component H.

In the closed string channel an effect of the boundary at x = 0 can be described in terms of
the boundary state which incorporates all information about boundary conditions [9,54,67,68].
In the case of the free CFT (A.1) the corresponding conformal boundary state |Bfree takes the
form
()
(k)
|Bfree = |Bfree |Bfree ,
(B.1)
where
()
|Bfree H () H ()
(B.2)
is the boundary state of the Gaussian CFT subject to the von Neumann boundary condition, while
(k)
|Bfree H (k) H (k)
(B.3)
is the boundary state of the minimal parafermionic model with free boundary condition.
As an important step toward the derivation of (A.38), we will show that the power series
A (s, p) are vacuum eigenvalues of certain operators acting in the chiral Hilbert space H ()
H (k) . With this aim in mind we need to describe explicitly H () and H (k) , as well as the boundary
states (B.2) and (B.3).
B.1. Chiral space H () and boundary state |Bfree
()
A general solution to the bulk equation of motion = 0 has the form:

+ ix),
(, x) = ( ix) + (
(B.4)
where (v) and (v)

are right and left chiral fields respectively. In the case of the uncompactified
Bose field : ( + 1/T , x) = (, x), the chiral fields admit the Fourier mode expansions of
the form:
m
1
e2imvT (v = ix),
(v) = 0 + vT 0 + i
(B.5)
2
m
m=0
and
m
1
0 + i
e2imvT
(v)
= 0 vT
2
m
(v = + ix).
(B.6)
m=0
The canonical quantization procedure leads to the following non-trivial commutation relations:
n
[ 0 , 0 ] = i;
(B.7)
[ n , m ] = [ n , m ] = n+m,0 .
2
Let FP (F P ) be the Fock space, i.e. the space generated by a free action of the operators n ( n )
with n < 0 on the vacuum vector |P which satisfies
n |P = n |P = 0,
0 |P = P |P ,
for n > 0;
0 |P = P |P .
(B.8)
186
Then the closed string space of states coincides with the direct integral
corresponding chiral Hilbert space is given by
H () = FP .
p FP
F P , and the
(B.9)
The boundary state (B.2) associated with the von Neumann boundary condition obeys the
equation,

()
x (, x) x=0 |Bfree = 0.
(B.10)
It reads explicitly as follows [67,68]

2
()
n
|Bfree = gD exp
(1) n n |vac() ,
n
(B.11)
n=1
where |vac() = |0 |0 is the ground state of the Gaussian theory in the closed string channel.
Now, in view of (B.11) it is easy to see that for 1/T > 1 > > n > 0 the following relation,
involving the correlators of the boundary exponential fields, holds:
()
ia1 B
(1 ) eian B (n )|vac()
free B|e
n
= e2iaP j =1 j P |e2ia1 (1 ) e2ian (n )|P .
Here
2ia

m 2im T
( ) = (2iT ) exp 2a
e
exp(ia 0 + 2ia T 0 )
m
m>0

m
2im T
exp 2a
e
m
(B.12)
a2
2
(B.13)
m>0
is an intertwining operator for the Fock spaces:

e2ia ( ) : FP FP +a .
(B.14)
(k)
B.2. Chiral space H (k) and boundary state |Bfree
B.2.1. Parafermionic and Wk current algebras

Here we review shortly some well-known facts about the chiral Hilbert space of the minimal
parafermionic models. The details can be found in Refs. [1,35,36].
Again, let the variable v be a complex coordinate, v = ix, on the 2D cylinder. The chiral
parafermionic algebra contains a set of currents { n (v)}k1
n=1 of conformal dimensions (2.1). This
algebra possesses a global symmetry with respect to the Zk transformations n na n with
a = 0, . . . , k 1, and its Zk invariant component can be generated by the currents with the lowest
conformal dimension, i.e., + 1 and k1 , through the OPE:

2
k+2
+ (u) (v) = (u v)2 k I +
(u v)2 W 2 (u) + W 2 (v)
2k

1
+ 3 (u v)3 W 3 (u) + W 3 (v) + .
(B.15)
2k 2
187
The spin-2 current W 2 (B.15), in turn, generates the Virasoro algebra with the central charge
ck =
2(k 1)
.
k+2
(B.16)
The higher-order terms involve higher-spin currents, W 3 , W 4 , . . . forming the Wk algebra introduced in [35,36] with the special value (B.16) of the central charge. Notice that the higher
currents can also be produced recursively from singular parts of OPEs of the lower currents. In
this sense the chiral Wk algebra is generated by two basic currents W 2 and W 3 .
The parafermionic algebra has a set of irreducible representations (irreps),

1
k

,
MJ J = 0, , . . . ,
(B.17)
2
2
obtained through the action of the + and modes on the highest vectors (see Ref. [1] for
details). The irrep MJ decomposes into a direct sum of subspaces specified by the Zk charge
(number of + modes minus number of modes modulo k):
MJ =
k1

M()
J .
(B.18)
=0
()
Each component MJ in (B.18) is an irrep of the Wk algebra.

Recall that an irrep of the Wk algebra is obtained by factorizing a highest weight modules
over submodules of null-vectors [35,36]. In the case under consideration, the highest weight
can be thought of as a pair of eigenvalues (, w) of mutually commuting operators
ck
+ (2T )1
24
1/T
dv
W 2 (v),
2
(2T )
1/T
dv
W 3 (v),
2
(B.19)
corresponding to the highest weight vector. In fact, it is most convenient to label the highest
weight vectors by means of another pair of numbers (J, M) such that
J (J + 1) M 2
,
k+2
k

M 2(3k + 4) 2
M 6J (J + 1) + k .
w(J,M) =
k
3 k
(J,M) =
(B.20)
If W(J,M) denotes the irrep of highest weight ((J,M) , w(J,M) ) (B.20), then the Wk structure of
the parafermionic irreps (B.18) is described as follows:
(0)
M0 W(0,0) ,
()
MJ W(J,J )
(j )
M0 W( k , k j )
2 2
(j = 1, . . . , k 1),
for < 2J (J > 0),
(B.21)
for
= 2J (mod k).
(B.22)
and
()
(
)
2 J
MJ M k
188
Let be a set containing k(k+1)

number of pairs (J, M) of the form:
2

1
k

= (J, M) J = 0, , . . . , ; M = J, J + 1, . . . , J ;
2
2

k
k
J, J .
(J, J )
2
2
(B.23)
As implied above, the chiral parafermionic algebra can be thought of as an algebra of intertwiners
in the set
{W(J,M) }(J,M)
(B.24)
of integrable irreps of Wk . In particular,

(v) : W(J,M) W(J,M1) .
(B.25)
Notice that the pairs (J, J ) and ( k2 J, k2 J ) are subject to the equivalence relation in (B.23) since the corresponding highest weights (, w) (B.20) are the same, i.e.,
W(J,J ) W( k J, k J ) . Finally, the closed string Hilbert space in the case of minimal
2
2
(J,M) , while the corparafermionic models is given by the direct sum (J,M) W(J,M) W
responding chiral Hilbert space can be described as follows:

H (k) =
(B.26)
W(J,M) .
(J,M)
B.2.2. Wk invariant boundary states

The Wk algebra posses an automorphism of the form
C : W s (1)s W s ,
s = 2, 3, . . . .
(B.27)
Thus [9,68], there are two types of Wk invariant CBCs. In the string theory they are referred to
as A- and B-branes [11]. The corresponding boundary states obey the equations

s ( )
A-brane:
W s ( ) (1)s W
(B.28)
|B = 0,
x=0
and
B-brane:

s ( )
|B = 0.
W s ( ) W
x=0
(B.29)
Let |I(J,M) A be the A-type Ishibashi state [9,11,68] corresponding to the integrable irreps
W(J,M) from the set (B.24). In other words, it is a solution of (B.28) in the space W(J,M)
(J,M) , that is unique, provided
W

(J, M) (J, M) |I(J,M) A = 1,
(B.30)
where |(J, M) is the highest weight vector of W(J,M) . Now, following the formalism of Ref. [9],
we can construct a set of states satisfying both the Cardy consistency condition and (B.28):
|B(j,m) =

(J,M)
(J,M)
S(j,m)

|I(J,M) A ,
(J,M)
S(0,0)
(j, m) .
(B.31)
189
(J,M)
Here S(j,m) is a modular matrix of the characters of the integrable irreps.12 The common wisdom
amount of Cardy states (B.31) are the boundary states corresponding to
then is that the k(k+1)
2
some local, Wk invariant boundary conditions. In Section 2.5 these CBCs have been referred
to as Bl,n (0 n l k 1), where the integers l, n are related to the pair (j, m) in (B.31)
as follows: l = k j m and n = j m. In particular, the k amount of the boundary states
k1
{|B( k , k s) }k1
s=0 correspond to the k possible types of the fixed CBCs, while {|B( k1 , k1 s) }s=0
2 2
(k)
are the boundary states associated with the non-trivial fixed points of the boundary flow R .
Let us consider now Eq. (B.29). Again, to construct non-trivial boundary states of the B-type
one must draw on the B-type Ishibashi states |I(J,M) B , i.e., the unique normalized solutions of
(J,M) . Obviously |I(J,M) B exists only if J and M satisfy the
(B.29) in the space W(J,M) W
condition
w(J,M) = 0,
(B.32)
where the polynomial w(J,M) is given by (B.20). The following solutions of the Cardy consistency condition [9] can be easily verified [811]13 :
[k]
(J,0)
2
S(j,0)

|Bj = k
|I(J,0) B
(J,0)
J =0 S(0,0)

k
.
j = 0, 1, 2, . . . ,
2
(B.33)
There are strong reasons to believe [810] that the solution |Bj with j = 0 corresponds to
the free CBC:
(k)
|Bfree = |B0 .
(B.34)
By virtue of the above described

structure of this boundary state, one can show that the following
relations are satisfied when ns=1 s = 0 (mod k):
(k)
(k)
free B| 1 (v1 ) n (vn )|vac
= gfree (0, 0)| 1 (v1 ) n (vn )|(0, 0),
(B.35)
(k)
(k)
free B| 1 (v 1 ) n (v n )|vac
= gfree (0, 0)| 1 (v1 ) n (vn )|(0, 0).
(B.36)
and
Here we use the notation |vac(k) |(0, 0) |(0, 0) for the closed string ground state of the
minimal parafermionic model.
It is essential to point out that correlators (B.35) and (B.36) are not single-valued functions of
their variables. To resolve the phase ambiguity in (B.35) we require that, (a) when the arguments
are real, vs = s , then (B.35) is a real function within the domain 1/T > 1 > > n > 0,
and (b) the short distance expansions of (B.35) must be consistent with the OPEs (2.4) provided

n ( ) n ( ix)|x=0 . Making use of the relation between parafermions and SU
k (2) WZW
currents [1], it is possible to show that both conditions can be indeed satisfied. Similarly, one
12 In the case under consideration the modular matrix is given by [6,28],

(2j + 1)(2J + 1)
22Mm
(J,M)
sin
S(j,m) =
.
k+2
k(k + 2)
13 Eq. (B.33) requires special consideration for even k and j = k (see Ref. [11] for details).
4
190
can resolve the phase ambiguity of (B.36). As a result, the boundary values of the correlation
functions (B.35) and (B.36) at real vs = vs = s coincide within the domain 1/T > 1 > >
n > 0.
B.3. Baxters operators
In the boundary state formalism, thermal expectation values (A.14) are expressed in terms of
correlators of the form:

0
free B| |vac

free B|vac
|vac = |vac() |vac(k) ,
(B.37)
with |Bfree defined in (B.1). Our previous analysis implies, therefore, that the integrand in (A.26)
can be written as an expectation value over the highest vectors |P , (J, M) = |P |(J, M) with
J = M = 0:
2n

(1 ) V (Mats)
(2n ) 0
eV j =1 j j V (Mats)
1
2n

= (0, 0), P U (1 ) U (n )P , (0, 0) .
(B.38)
n
Here the parameter V is related to the zero-mode momentum P :

P
V = 2iT ,
k
(B.39)
and
U ( ) = e
2
i
(B.40)
( ).
From the mathematical point of view, U (v) are intertwining operators (chiral vertex operators)
of the conformal dimension d (A.4). They act in the chiral Hilbert space

VP ,(J,M) with VP ,(J,M) FP W(J,M)
H free =
(B.41)
P (J,M)
in accordance with the formula

U ( ) : VP ,(J,M) VP
,(J,M1)
(B.42)
Eq. (B.38), in turn, implies that the power series A (s,p) (A.24), (A.30) can be thought of as
expectation values over the state |P , (0, 0) with P = 2 kp/ of the operators

2i 0 h 1
k
A (s) = Tr 2Mh e
!
Tr 2Mh e
2i 0 h
k
(B.43)

1/T
T exp
"
d K ( )
Here K ( ) = a U+ ( ) + a U ( ) and s = 2 (2T )2d2 .

Some explanations of formula (B.43) are in order at this point. First, it contains the ordered
exponential (the symbol T denotes the path ordering) which is defined in terms of the power
191
series in as follows,

1/T
T exp
d K ( )
0
=1+
d1 dn K (1 )K (2 ) K (n ).
n=1
(B.44)
1/T >1 >2 >>n >0
In fact, only even powers of survive in the definition (B.43), so A are formal series in s 2
whose expansion coefficients are operators acting in the chiral Hilbert space (B.41). Second, the
factor 2Mh appearing in (B.43) was undefined yet. As follows from the commutation relations
(A.19), only integer powers of the operator 2M are involved in (B.43). The action of 2M in the
chiral Hilbert space (B.41) is defined by the formula
2M VP ,(J,M) = 2M VP ,(J,M) .
(B.45)
Notice that, because of the isomorphism W(J,J ) W( k J, k J ) , the operator M itself is ill2
2
defined.
In spite of the fact that the factor 2Mh does not contribute to the matrix element
P , (0, 0)|A |P , (0, 0), it plays an important role in the definition of A . Due to this phase
factor the operators A commute with the operator
1/T
I1 =

dv
()2 + W 2 .
2
(B.46)
A proof of the commutation relation

I1 , A (s) = 0,
(B.47)
is based on the quasiperiodic properties of the intertwining operators (u):

(u + 1/T ) = (u)12M .
(B.48)
It follows along the line of Appendix C of Ref. [24]. Notice that the operators I1 + I1 and
I1 I1 coincide, respectively, with Hamiltonian and the momentum operator in the closed string
channel. Hence, commutation relation (B.47) manifests the invariance of A (s) with respect to
translations along the -direction.
It is worth to note that the operators A (s) (B.43) act invariantly in VP ,(J,M) . Furthermore,
since each space VP ,(J,M) naturally splits into the direct sum of finite dimensional level subspaces
VP ,(J,M) =
()
VP ,(J,M) ;
()
()
LVP ,(J,M) = VP ,(J,M) ,
(B.49)
=0
and the grading operator L essentially coincides with I1 , then A (s) act invariantly in all the level
()
subspaces VP ,(J,M) . In particular, the highest vectors |P , (J, M) are eigenvectors for A (s) and

A (s)P , (0, 0) = A (s, p)P , (0, 0)
P
with p = .
2 k
(B.50)
192
Following the route of the work [24] it is possible to prove that A (s) commute among themselves for any values of s,

A (s), A (s
) = A+ (s), A (s
) = 0,
(B.51)
and satisfy the operator-valued identity

i 2
i 2
e k 0 A+ (qs)A q 1 s 2M A+ q 1 s A (qs) = e k 0 2M ,
(B.52)
Again, formulae (B.51) and (B.52) should be understood as an infinite set of

with q =
formal operator-valued relations for the expansion coefficients of A without any reference to
convergence issues. The quantum Wronskian relation (A.38) follows immediately from (B.50)
and (B.52).
A key idea of the proof of (B.51) and (B.52) is based on the observation that the formal
integrals
eid .
1
x0 =
q q 1
1/T
du U (u),
0
1
x1 =
q q 1
1/T
du U + (u)
(B.53)
and can be identified with the

satisfy the Serre relations for the quantum affine algebra Uq (sl(2))
generators of its Borel subalgebra. With this observation the construction elaborated in [24] can
be applied without any changes. Finally, we note that the operators A , acting in the CFT chiral
Hilbert space, play a role similar to that of the Q-operators of Baxters lattice theory [69].
Appendix C. Partition function of the BP sinh-Gordon model
If the parameter 2 = b2 is negative the Schrdinger equation (A.50) takes the form (2.32).
It is quite natural to expect that this differential equation within the parametric domain b2 > 0
is somehow related to the BP sinh-Gordon model (2.14). The precise relation is proposed in
Eq. (2.36). Unfortunately, a rigorous derivation of this formula does not currently exist. Here
we briefly argue in favour of (2.36) based on the high- and low-temperature expansions of the
(k)
partition function Zbshg (h).
C.1. High-temperature expansion
C.1.1. The BP Liouville model
First of all, let us discuss the leading high-temperature behavior of (2.29) based on properties of the Hamiltonian (2.14). Qualitatively, one may expect that if e h is not too small,
the limit T of (2.29) is controlled by either the boundary operator V + + e
b B
k
or
b B
V e k
in (2.14) depending on the sign e h, with some crossover at small e h.
More precisely, let us assume that e h < 0. Then we can treat the term V in the Hamiltonian (2.14) as a perturbation, and in leading approximation one has

(k)
(k)
Zbshg (h)T Zbl (h) (e h < 0),
(C.1)
(k)
where Zbl (h) is the partition function corresponding to the Hamiltonian of the BP Liouville
model:
(k)
H bl = H free + h B + e
b B
k
(C.2)
193
An important feature of the QFT (C.2) is the conformal invariance. To be precise, the theory can
be made scale and conformally invariant by an appropriate redefinition of the RG transformation,
namely by supplementing it by a formal field redefinition + Q tRG , where the RG
k
time tRG log(1/T ) and Q = b + b1 . In the nomenclature of string theory this corresponds
to introducing a linear dilaton which modifies slightly the stress-energy tensor of the model:
Q
T (2) = ()2 + 2 + W 2 ,
k
Q
(2)
2 + 2 + W
2.
T = ()
k
(C.3)
The first two terms in each equation (C.3) constitute the corresponding chiral components of the
2 are chiral components of
modified stress-energy tensor of the Gaussian theory, while W 2 and W
the stress-energy tensor of the minimal parafermionic model (see the OPE (B.15)). The scaling
b
dimension of the boundary operator + e k B with respect to the modified stress-energy tensor
(k)
(C.3) equals one and, hence, the temperature and -dependences of the partition function Zbl (h)
readily follow from the dimensional analysis:
(k)
Zbl (h) = (2T )
i PQ
k
P k
b
G(k) (P |b),
(C.4)
with
h
P =i .
T
(C.5)
Here G(k) is some function of the dimensionless parameters P , b and k.

In a view of the standard heuristic arguments of Ref. [70], it is expected that G(k) (P |b) is an
analytical function of the complex variable P in the lower half-plane m P < 0. Furthermore,
(k)
following [70] in the perturbative calculation of the partition function Zbl , one can first integrate out the constant mode of the
Bose field . This integration produces simple poles in the
variable P at the points P = inb k (n = 0, 1, . . .), and the corresponding residues
(k)

G(k)
n = ResP =ib kn G (P |b)
(n = 0, 1, 2, . . .)
(C.6)
are expressed through the integrals of the (k n)-points Matsubara correlation functions
G(k)
n =
gD gfree
2
(2T )n(kn1)nb
2i
1
kn1
(Mats)

(Mats)
dkn V +
(1 ) V +
(kn ) 0 .
d1
0
(C.7)
Here gfree is the boundary degeneracy of the free CBC in the minimal parafermionic model (see
Eq.(1.2)), and the prime means that the constant mode contribution is excluded from the thermal
averaging 0 (A.14). The free-field correlators in (C.7) have an especially simple form for
n = 1. In this case the integral (C.7) can be brought to the form of the Selberg integral [63,64].
194
This yields the explicit formula

(k)
G1
gD gfree k!
=
2i k k2
=
v
k1
dv1
0
dvk
0
2
gD gfree 2
b
2i
k(1

i<j
1+b2
k )
2(1+b2 )

vi v j k
2 sin
2
k
(C.8)
Another highly non-trivial property of the BP Liouville partition function can be guessed
by virtue of the form of stress-energy tensor (C.3). Namely, it is invariant with respect to the
transformation b b1 . In the case of the conventional Liouville model a similar phenomenon
manifests a remarkable non-perturbative duality of the theory. If one admits that the same sym(k)
metry occurs in the BP Liouville model, then,
the function G (P |b) should also possess an
1
additional series of dual poles at P = ib
kn.
It should be mentioned that the model (C.2) has been already studied in the works [71] and
[72] for the cases k = 1 and k = 2, respectively. In both cases the function G(k) (C.4) looks as
follows:

iP k

2
b
gD gfree
ibP
iP
2b k
(k)
G (P |b) =
.
1+ 1+
2
2iP
k
b k
k(1 1+b
k )
(C.9)
Notice that (C.9) satisfies the all above-mentioned conditions even thought k > 2.
C.1.2. General structure of the high-temperature expansion
(k)
Above we have discussed the leading high-temperature behavior of Zbshg in the case e h < 0.
(k)
(k)
Obviously for e h > 0, Zbl (h) in asymptotic formula (C.1) should be replaced by Zbl (h).
(k)
Since Zbl (h) vanishes in the limit T if h is taken in the wrong half-plane (note the
i PQ
factor T k in (C.4)), this in turn suggests that the overall T asymptotics of the partition
function is correctly expressed by the sum
(k)
(k)
(k)
Zbshg T Zbl (h) + Zbl (h).
(C.10)
What can be said about corrections to the leading asymptotic? Again we assume that e h < 0
and consider the perturbative effect of the term V in (2.14) to the partition function (C.4). In
the unperturbed BP Liouville model theory the parameter is a dimensionless constant. Let us
eliminate from the Hamiltonian (C.2) by shifting the field . Then the coupling in front of
the V in Eq.(2.14) is replaced by 2 . Since the anomalous dimension of the boundary operator V with respect to the BP Liouville stress-energy tensor (C.3) is given by 1 2bQ
k , then
2bQ
2 E k , where E is the physical energy scale of the BP sinh-Gordon model. Hence we

deduce that the perturbative corrections to the leading asymptotics (C.1) should be in a form of
2bQ
power series expansion of the dimensionless parameter k , with = E /T . The case e h > 0
can be analyzed similarly and one comes to the same conclusion about the form of perturbative
corrections to (C.10). As was mentioned above, it is quite natural to expect that the leading asymptotics (C.10) is invariant with respect to the duality transformation b b1 . If one assumes
that the BP sinh-Gordon partition function possesses this property of self-duality as well, then
195
(k)
the high-temperature expansion of Zbshg should be of the form

(k)
(k)
(k)
Zbshg = Zbl (h)M(, h) + Zbl (h)M(, h).

Here M(, h) is a formal double power series in integer powers of

2Q
1
Mn,m (h) k (nb+mb ) .
M(, h) 1 +
(C.11)
2bQ
k
2Q
and bk :
(C.12)
n,m
C.1.3. Small expansion of the spectral determinant

(k)
In order to justify the proposed formula for Zbshg (2.36), we should observe the structure
(C.11) in the small- expansion of the spectral determinant D(, ) (2.42). When goes to zero
ku
and u 1 the potential in the Schrdinger problem (2.32) can be approximated by 2 e bQ . To

understand the quality of this approximation one can make a change the variable (A.57) with
d = 1 bQ/k and bring the differential equation (2.32) to the form (A.58). Therefore for u 1

ku
2Qb
2Qb 2bQ
+ (u)
(C.13)
K bP
e
.
k i k
k
Here and below, in order to write formulae in the most instructive form, we use the notation
rather than (see relations (C.5) and (2.38)). The overall normalization in (C.13) is
P = 2Q
k
chosen to ensure the asymptotic condition (2.34). Within the domain,

Q
Qb
1
1
log( 2 ) u
(C.14)
log 2 ,
kb
k
the potential in (A.58) develops a wide plateau and the solution + (C.13) becomes a combination of the two plane waves,

bP

kP
Qb
bP
Qb i k
A(, P |b)ei 2Q u
i
+ (u) =
2k
k
k

bP

kP
Qb
bP
Qb i k
+
A(, P |b)ei 2Q u ,
i
(C.15)
2k
k
k
with
2Qb
A(, P |b) = 1 + O k .
(C.16)
When is small but finite, corrections to (C.16) can be obtained using the perturbation theory.
In view of Eq. (A.59), these corrections have the form of a power series in
A(, P |b) 1 +
An (P |b)
2Qbn
k
2Qb
k
:
(C.17)
n=1
The formal power series A(, P |b) amounts to A+ (s, p) (A.55) taken at 2 = b2 < 0 with the
. We have no reason to expect the convergence
variable s related to as in (A.56) and p = i bP
2 k
of the power series (C.17) for b2 > 0, i.e., the series should be understood as a formal asymptotic
expansion.
To find the form of the solution in the domain (C.14), we note that the transformation
u u and b b1 leaves the Schrdinger equation (2.32) unchanged while interchanging
196
the solutions + and (2.34).14 Hence, can be obtained by means of this transformation
from (C.15) within the domain (C.14). Now the Wronskian (2.42) can be calculated at any point
from (C.14), where both are combinations of plane waves. It yields the following form of
the small- expansion of the spectral determinant:

D(, ) R(P )A(, P |b)A , P |b1 + R(P )A(, P |b)A , P |b1 ,
with
i QP

P
k
iP
ibP
iP
i (b1 b) Q
R(P ) =
,
b k
k
2 k
k
b k
(C.18)
.
and P = 2Q
k
The high-temperature expansion (C.18) is in agreement with (C.11). Furthermore, it implies
that formula (C.9) holds true for any k 1.
C.2. Low-temperature expansion

Here we elucidate the low-temperature behavior of Zbshg (h) using the boundary state formalism described in Appendix B.
In the case of the BP sinh-Gordon model the boundary state |Bbshg is some particular vector
in the closed string space of state

VP ,(J,M) V P ,(J,M) ,
|Bbshg
(C.19)
P (J,M)
where we use the same notations as in formula (B.41). Unlike the conformal boundary states
discussed in Appendix B, |Bbshg essentially depends on the temperature via the dimensionless parameter = E /T . In the presence of the external field h (2.14), the partition function
Zbshg (h) can be expressed in terms of the vacuum overlap of the boundary state (C.19) analytically continued to pure imaginary values of the variable P (C.5):

Zbshg (h) =bshg B| P , (0, 0) P , (0, 0) P =i h .
(C.20)
T
The motivation behind (C.20) can be found in Ref.[20] where a similar relation was exploited in
the context of the circular brane model.
The simplest idea about the infrared fixed point of the BP sinh-Gordon boundary flow is that
it corresponds to the fixed CBC for both bosonic and parafermionic sectors of the theory. This
scenario implies that
|Bbshg e
Ebshg
T
()
|BD |B(0,m)
as .
(C.21)
Here Ebshg E is the ground state energy, |B(0,m) is the boundary state corresponding to the
fixed CBC of the minimal parafermionic model (B.31),15 and

2
()

|BD = gD exp
(C.22)
|P |P
n n n
P
n=1
14 For this reason the partition function (2.36) is invariant with respect to the duality transformation b b1 .
15 Recall, in order to fix the value of integer m = 0, 1, . . . , k 1 in (C.21), one needs to specify the normalization of the
boundary fields unambiguously (see Section 2.1).
197
is the boundary state associated with the Dirichlet CBC [67,68]. The assumption (C.21) leads, in
turn, to the leading low-temperature asymptotics for the BP sinh-Gordon partition function:
Ebshg
+ log(gD gfixed ) + O(T ).
(C.23)
T
To make more definite predictions about the low-temperature expansion of Zbshg (h), one
should take into account the integrability of the model. In a forthcoming paper [73] we intend
to present a comprehensive discussion of a relevant set of local Integral of Motions (IMs) in the
context of more general QFT model with boundary interaction. Skipping details of the computational complexity we mention here that the boundary state (C.19) is expected to satisfy an infinite
set of conditions
log Zbshg (h) =
(I2l1 I2l1 )|Bbshg = 0
(l = 1, 2, . . .)
(C.24)
for some operators
1/T
I2l1 (x) =
d
T 2l ( ix),
2
I2l1 (x) =
1/T
d
T 2l ( + ix),
2
(C.25)
where T 2l and T 2l are chiral currents of the Lorentz spin 2l and (2l), respectively. As has been
explained in Appendix C, the local chiral currents of the parafermionic minimal model constitute
Wk algebra. Therefore the local densities T 2l are, in fact, appropriately regularized polynomial
and
in and Wk currents as well as their derivatives. Similarly, T 2l are constructed using
Wk currents. Notice that any operators in the form (C.25) are conserving charges in the sense
that
d
d
(C.26)
I2l1 (x) =
I2l1 (x) = 0,
dx
dx
and the meaning of (C.24) is that the boundary neither emits nor absorbs any amount of the
combined charges I2l1 I2l1 [54]. For this reason I2l1 and I2l1 are referred to as local IMs.
The local IMs I1 and I3 are known in the closed form. Namely, I1 is given by (B.46) and an
explicit bosonized form of the first nontrivial local density T 4 is presented in Ref. [74]. There
are strong indications that the commutativity conditions
[I2l1 , I1 ] = [I2l1 , I2 ] = 0
(C.27)
(a) fix all the operators I2l1 for l > 2 uniquely up to the normalization, and (b) such defined
operators I2l1 mutually commute:
[I2l1 , I2m1 ] = 0.
(C.28)
In order to specify I2l1 unambiguously, it is convenient to normalize the corresponding local

density in such a way that the monomial ()2l appears in T 2l with the coefficient equal one, i.e.
1/T
I2l1 =

d
()2l + .
2
(C.29)
We can now take advantage of the general prediction from Introduction of Ref. [32] about the
form of the low-temperature expansion in the case of an RG boundary flow which possesses an
198
infinite set of mutually commuting local IMs, and terminated at the trivial infrared fixed point
(the one which in Cardys classification [9] corresponds to the identity primary state). In the case
under consideration the prediction implies that the low-temperature expansion of log Zbshg (h)
looks as follows
2l1

E
T
h
log Zbshg (h) f0
fl
I2l1 i
+ log(gD gfixed )
,
T
E
T
(C.30)
l=1
where fl are some coefficients and I2l1 (P ) are vacuum eigenvalues of the operators I2l1 :

I2l1 P , (0, 0) = (2T )2l1 I2l1 (P )P , (0, 0) .
(C.31)
Let us emphasize that the I2l1 (P ) are unique polynomials of order l of the variable P 2 , and, as
follows from the normalization condition (C.29)

I2l1 (P ) =
P
2
2l
+ ,
(C.32)
where the omitted terms contain lower powers of P 2 . In particular, in view of the explicit formulae for I1 (B.46) and I3 [74] it is possible to show that

I1 (P ) =

I3 (P ) =
P
2
P
2
2
4
k
,
8(2 + k)
(C.33)
2
5k
k(9k + 4Q2 )
P
+
,
4(2 + 3k) 2
64(2 + k)(2 + 3k)
with Q = b + b1 .
Applying the standard WKB iterational scheme [30] to the spectral determinant (2.42), one
can reproduce exactly the structure of the low-temperature expansion (C.30) as well as the eigenvalues (C.33). En passant, the WKB calculation provides an explicit form of the expansion
coefficients:
(2l1)k
(2l1)k
(l 12 ) ( 2Q b)( 2Q b1 )
fl =
2 l!
(kl k2 )
2l
k
.
Q
(C.34)
Appendix D. Low-temperature expansion of Z (k)

At low temperature the regularized partition function (2.53) can be studied using the WKB
expansion. The leading WKB approximation yields an explicit expression for the regularized
ground state energy (2.56):

(m 12 )
E (k)
1
1
k
m
= 2 1
+
mk
E
k
2
2
2 m!
m=2

k
1
+
sin2m ( ).
2
2
(D.1)
199
It is useful to bear in mind that (D.1) is applicable for 0 < 2 only. It can be analytically
continued within the domain 2 < by means of the relation:
(k)
(k)
E /E = E /E +

k1
1+ 2l
( kl 12 )
3 l 2
l
k
sin( )
, 1; + sin ( )
2 F1
k
k
2 k
(1 kl )
l=1
l= k2

4 2 (k 1)
+ sin
1 cos( ) log sin
k
2
2

+ 1 + cos( ) log cos

.
2

The following formula for the regularized ground state energy holds within 0 <

k
(k)
(k)
log k cos( )
E /E = C0 E + 2 log 2 +
2

k1
( l + 1 )
1 1
3
l
k
2
2
2
cos ( )
+ , , 1; 2, cos ( )
+
3 F2
k
k 2 2
2
( kl )
l=1

k1

2
l
5
l
3
2
cos ( )
+ 1, 1, 1; 2, cos ( ) ,
3 F2
3k
k
k
2
(D.2)
(D.3)
l=1
where
(k)
C0 =

k1
( kl 12 ) 2 log 2 2 (k 1)
sin
.
k
k
2
( kl )
l=1
(D.4)
l= k2
The systematic WKB expansion of (2.53) leads to an asymptotic series of the form
2l1
(k)

E
T
(k)
(k)
log Z
Fl ( )
,
+ log(gfixed ) +
T
E
(D.5)
l=1
(k)
where the coefficient F1
reads explicitly as

k1
1 1 1 3 1 2
(k)
, + ; + sin ( )
F1 ( ) =
2 F1
12(k + 2)
2 2 k 2 k
1 2
k( 12 + k1 )
k
=
sin( )
1
24( k 1)

k1
3
1
2
cos( )2 F1 1 + , 1; cos ( ) .
12k
k
2
References
[1] A.B. Zamolodchikov, V.A. Fateev, Sov. Phys. JETP 62 (1985) 215.
[2] F.Y. Wu, Y.K. Wang, J. Math. Phys. 17 (1976) 439.
[3] S. Elitsur, R. Pearson, J. Shigemitau, Phys. Rev. 19D (1979) 3698.
(D.6)
200
[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]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
E. Fradkin, L. Kadanoff, Nucl. Phys. B 170 (FS1) (1980) 1.

V.A. Fateev, A.B. Zamolodchikov, Phys. Lett. A 92 (1982) 37.
D. Gepner, Z. Qiu, Nucl. Phys. B 285 (1987) 423.
J.L. Cardy, Nucl. Phys. B 275 (FS12) (1986) 200.
H. Saleur, M. Bauer, Nucl. Phys. B 320 (1989) 591.
J.L. Cardy, Nucl. Phys. B 324 (1989) 581.
I. Affleck, M. Oshikawa, H. Saleur, Nucl. Phys. B 594 (2001) 535.
J. Maldacena, G. Moore, N. Seiberg, JHEP 0107 (2001) 046.
I. Affleck, A.W.W. Ludwig, Phys. Rev. Lett. 67 (1991) 161.
D. Friedan, A. Konechny, Phys. Rev. Lett. 93 (2004) 030402.
R. Chatterjee, A. Zamolodchikov, Nucl. Phys. B 432 (1994) 427.
S. Fredenhagen, V. Schomerus, JHEP 0202 (2002) 005.
I. Affleck, M. Oshikawa, H. Saleur, J. Phys. A 31 (1998) 5827.
V.A. Fateev, I.V. Frolov, A.S. Shwarz, Nucl. Phys. B 154 (1979) 1.
M. Lscher, Nucl. Phys. B 200 (FS4) (1982) 61.
Al.B. Zamolodchikov, unpublished notes, 2001.
S.L. Lukyanov, A.B. Zamolodchikov, J. Stat. Mech.: Theor. Exp. (2004) P05003.
V. Ambegaokar, U. Eckern, G. Schn, Phys. Rev. Lett. 48 (1982) 1745.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Commun. Math. Phys. 177 (1996) 381.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Nucl. Phys. B 549 (1999) 529.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, J. Stat. Phys. 102 (2001) 567.
P. Fendley, H. Saleur, Phys. Rev. B 60 (1999) 11432.
V.G. Kac, Infinite Dimensional Lie Algebras, Birkhuser, Boston, 1983.
E. Witten, Phys. Rev. D 47 (1993) 3405.
L.D. Landau, E.M. Lifshitz, Quantum Mechanics: Non-Relativistic Theory, vol. 3, third ed., Pergamon Press, Oxford, 1997.
V.A. Fateev, Al.B. Zamolodchikov, Phys. Lett. B 271 (1991) 91.
S.L. Lukyanov, E.S. Vitchev, A.B. Zamolodchikov, Nucl. Phys. B 683 (2004) 423.
S.L. Lukyanov, A.M. Tsvelik, A.B. Zamolodchikov, Nucl. Phys. B 719 (2005) 103.
A.B. Zamolodchikov, Teor. Mat. Fiz. 65 (1985) 347.
V.A. Fateev, A.B. Zamolodchikov, Nucl. Phys. B 280 (1987) 644.
V. Fateev, S. Lukyanov, Int. J. Mod. Phys. A 3 (1988) 507.
J.M. Kosterlitz, Phys. Rev. Lett. 37 (1976) 1577.
W. Hofstetter, W. Zwerger, Phys. Rev. Lett. 78 (1997) 3737.
R.G. Leigh, Mod. Phys. Lett. A 4 (1989) 2767.
V.S. Rychkov, JHEP 0212 (2002) 68.
X. Wang, H. Grabert, Phys. Rev. B 53 (1996) 12621.
A.M. Polyakov, Phys. Lett. B 59 (1975) 79.
A.A. Belavin, A.M. Polyakov, Pisma Zh. Eksp. Teor. Fiz. 22 (1975) 503, JETP Lett. 22 (1975) 245.
S.E. Korshunov, Pisma Zh. Eksp. Theor. Phys. 45 (1987) 342, JETP Lett. 45 (1987) 434.
Y.V. Nazarov, Phys. Rev. Lett. 82 (1999) 1245.
M.V. Feigelman, A. Kamenev, A.I. Larkin, M.A. Skvortsov, Phys. Rev. B 66 (2002) 054502.
C.G. Callan, L. Thorlacius, Nucl. Phys. B 329 (1990) 117.
A.O. Caldeira, A.J. Leggett, Phys. Rev. Lett. 46 (1981) 211.
G. Schn, A.D. Zaikin, Phys. Rep. 198 (1990) 237.
I.S. Beloborodov, A.V. Andreev, A.I. Larkin, Phys. Rev. B 68 (2003) 024204.
D.V. Averin, K.K. Likharev, in: B.L. Altshuler, et al. (Eds.), Mesoscopic Phenomena in Solids, Elsevier, Amsterdam,
1991, p. 173.
K.A. Matveev, Phys. Rev. B 51 (1995) 1743.
S.L. Lukyanov, P. Werner, J. Stat. Mech.: Theor. Exp. (2006) P11002.
S. Ghoshal, A. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841.
P. Fendley, H. Saleur, N.P. Warner, Nucl. Phys. B 430 (1994) 577.
K. Moon, H. Yi, C.L. Kane, S.M. Girvin, M.P.A. Fisher, Phys. Rev. Lett. 27 (1993) 4381.
P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. Lett. 74 (1995) 3005.
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
201
P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. B 52 (1995) 8934.

C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 15233.
F. Lesage, H. Saleur, P. Simonetti, Phys. Rev. B 56 (1997) 7598.
L.V. Keldysh, Sov. Phys. JETP 20 (1965) 1018.
U. Weiss, Solid State Commun. 100 (1996) 281.
A. Selberg, Norsk. Mat. Tidsskr. 26 (1944) 71.
V. Dotsenko, V. Fateev, Nucl. Phys. B 251 (1985) 691.
P. Dorey, R. Tateo, J. Phys. A 32 (1999) L419.
F. Calogero, A. Degasperis, Spectral Transform and Solitons, North-Holland, Amsterdam, 1982.
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 293 (1987) 83.
N. Ishibashi, Mod. Phys. Lett. A 4 (1989) 251.
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, 1982.
M. Goulian, M. Li, Phys. Rev. Lett. 66 (1991) 2051.
V. Fateev, A. Zamolodchikov, Al. Zamolodchikov, Boundary Liouville field theory. 1. Boundary state and boundary
two point function, Preprint RUNHETC-2000-01, hep-th/0001012.
[72] C. Ahn, C. Rim, M. Stanishkov, Nucl. Phys. B 636 (2002) 497.
[73] S.L. Lukyanov, A.B. Zamolodchikov, in preparation.
[74] V.A. Fateev, Nucl. Phys. B 473 (1996) 509.
Bootstrap in supersymmetric Liouville field theory.

I. NS sector
A. Belavin a, , V. Belavin b,c,1 , A. Neveu d , Al. Zamolodchikov d,2
a L.D. Landau Institute for Theoretical Physics RAS, 142432 Chernogolovka, Russia
b International School for Advanced Studies (SISSA), Via Beirut 2-4, 34014 Trieste, Italy
c INFN, Sezione di Trieste, Italy
d Laboratoire de Physique Thorique et Astroparticules, UMR-5207 CNRS-UM2, Universit Montpellier II,
Pl.E. Bataillon, 34095 Montpellier, France

Received 20 March 2007; accepted 18 April 2007
Available online 4 May 2007
In memory of Yasha Belavin
Abstract
A four point function of basic NeveuSchwarz exponential fields is constructed in the N = 1 supersymmetric Liouville field theory. Although the basic NS structure constants were known previously, we present
a new derivation, based on a singular vector decoupling in the NS sector. This allows to stay completely
inside the NS sector of the space of states, without referencing to the Ramond fields. The four-point construction involves also the NS blocks, for which we suggest a new recursion representation, the so-called
elliptic one. The bootstrap conditions for this four point correlation function are verified numerically for
different values of the parameters.
2007 Published by Elsevier B.V.
E-mail address: belavin@itp.ac.ru (A. Belavin).

1 Present address: Institute for Theoretical and Experimental Physics (ITEP), B. Cheremushkinskaya 25, 117259
Moscow, Russia.
2 On leave of absence from: Service de Physique Thorique, CNRS-URA 2306, C.E.A.-Saclay, F-91191, Gif-surYvette, France.
0550-3213/$ see front matter 2007 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.04.018
A. Belavin et al. / Nuclear Physics B 784 [FS] (2007) 202233
203
1. The N = 1 super-Liouville
Construction of the super-Liouville field theory (SLFT) is motivated by the following action,
which appeared in the non-critical superstring theory in 1981 [1]
1
1
+
)
+ 2ib2 e
b + 2b2 2 e2b ,
(1)
(a )2 +
(
8
2
where b is the standard quantum parameter related through the background charge Q =
b1 + b to the central charge
LSLFT =
c = 1 + 2Q2
(2)
z) and the stress tensor

of the superconformal algebra generated by the supercurrent S(z), S(
T (z), T (z). Traditionally the scale parameter is called the (super) cosmological constant.
Let us recall some details about SLFT necessary for the forthcoming discussion (see e.g., [2
6]). The space of fields splits into the so-called NeveuSchwarz [7] (NS) and Ramond [8] (R)
being respectively single or double valued near the
sectors, the supercurrent components (S, S)
point where the corresponding operator is located. Apparently the first, NS sector, is closed under
the operator product expansions (OPE). It is completely consistent to consider it separately. This
is what we are going to do in the present publication, meaning to include the R sector in the
future.
Respectively, the NS fields belong to highest weight representations of the NS superconformal
algebra

c 2 1
{Gk , Gl } = 2Lk+l +
k
k+l ,
2
4

n
k Gn+k ,
[Ln , Gk ] =
2

c
[Lm , Ln ] = (m n)Lm+n + m3 m m+n ,
(3)
8
where the subscripts m, n are integer and k, l are half-integer. In fact there are two copies of algebra (3), the right one SVir generated by S(z) and T (z), and the left SVir constructed from
z) and T (z). The space is classified in the highest weight reprethe left-moving components S(
sentations of SVir SVir. The basic NS fields are the scalar primary fields Va (x) corresponding
to the highest weight vectors
Ln Va = 0,
L n Va = 0, for n > 0,
k Va = 0, for k > 0,
G
Gk Va = 0,
L0 Va = L 0 Va = a Va ,
(4)
where
a(Q a)
(5)
2
and a is a (complex) continuous parameter. It is sometimes instructive to think of these basic
operators as of the properly regularized exponentials Va = exp(a) of the fundamental bosonic
field entering the Lagrangian (1). This is particularly useful in the region of the configuration
space where . Here one can neglect the interaction terms in the action (1), the fields
a =
204
behave as a free boson and a free Majorana fermion and the exponential expression
and (, )
can be given an exact sense.
All other NS fields are the SVir SVir descendents of these basic ones. It will prove convenient to distinguish the descendents of integer and half-integer level, for which we reserve
(somewhat loosely) the terms even descendents, often marked by the index e, and the odd
ones, referred to as o. It will be also useful to introduce special notations for the components of
the multiplets under the standard super Poincar algebra, a subalgebra of SVir SVir generated
1/2 and L0 L 0 . From this point of view Va is the bottom component of the
by G1/2 , G
supermultiplet, which includes also
a = G1/2 Va = iaea ,
1/2 Va = ia e
a ,
a = G
1/2 Va = a 2 e
a 2iabe(a+b) .
Wa = G1/2 G
(6)
Here we partially borrow apt notations from Ref. [9] and also give a free field interpretations
of the corresponding components. The basic Ward identities are
a
1
Va (0) + Va (0) + reg,
2
z
z
a + 1/2
1
T (z)a (0) =
a (0) + a (0) + reg,
z
z2
1
S(z)Va (0) = a (0) + reg,
z
2a
1
S(z)a (0) = 2 Va (0) + Va (0) + reg.
z
z
T (z)Va (0) =
(7)
We explicitly present the holomorphic relations and quote them for Va and a only. The right
superconformal properties of the doublet a , Wa are the same as of Va , a and the left ones of
Va , a and a , Wa are similar to (7) with obvious modifications caused by the anticommutativity
k.
of the right and left fermionic generators Gk and G
Local properties of SLFT in the NS sector are encoded in the basic operator product expansion
(here and below for the sake of brevity we denote = Q/2+iP and i = ai , wherever it
cannot cause any misunderstanding)

Q/2+iP

dP
VQ/2+iP (0) ee
(x x)
1 2 Ca1 ,a2
Va1 (x)Va2 (0) =
4

Q/2+iP
VQ/2+iP (0)
.
+C
(8)
a ,a
1
oo
This OPE is continuous and involves integration over the momentum P . Precisely as in the
bosonic Liouville field theory [10] the integration contour is basically along the real axis, but
should be deformed sometimes under analytic continuation in the parameters a1 and a2 . It is a
good idea to make such deformations explicit collecting the result in the form of the so-called
discrete terms [10]. In (8) [Vp ] denotes the contribution of the primary field Vp and its superconformal descendents to the operator product expansion. Unlike standard conformal symmetry,
not all these contributions are prescribed unambiguously by the superconformal invariance, the
even and the odd ones entering independently. From here come two different structure constants
p
pa a in the OPE (8), while [Va ]ee and [Va ]oo denote respectively the collections of
Ca1 a2 and C
1 2
205
eveneven3 and oddodd descendents. As usual [11] both towers of descendents enjoy the
factorization in the product of holomorphic and antiholomorphic chains

Va (0) ee = Ce1 ,2 (a , x)Ce1 ,2 (a , x)V
a (0),

1 ,2
,
(a , x)Co 1 2 (a , x)V
a (0),
Va (0) oo = Co
(9)
where each of the chain operators

+ 1 2
L1 + O x 2 ,
2

x 1/2
Co1 ,2 (, x) =
(10)
G1/2 + O x 3/2
2
(and the same for Ce and Co with the right SVir operators Gk and Ln replaced by the left
k and L n ) is determined uniquely by superconformal symmetry once the normalization
ones G
of the first term is fixed, e.g., as in (10), essentially in the same way as it occurs in the usual CFT
case.
Q/2iP
Q/2iP
and C
in (8) have been evaluated through
The basic NS structure constants Ca1 a2
a1 ,a2
the bootstrap technique quite a while ago in Refs. [6,12]. Here we quote their result in terms of
the three-point functions
Ce1 ,2 (, x) = 1 + x
Va1 (x1 )Va2 (x2 )Va3 (x3 ) =
(x12 x12
Wa1 (x1 )Wa2 (x2 )Wa3 (x3 ) =
Ca1 ,a2 ,a3

,
23 x 23 ) 2+31 (x31 x 31 ) 3+12

(1/2 1 2 3 )2 C a1 ,a2 ,a3
)1+23 (x
.
(x12 x12 )1+23 +1/2 (x23 x23 )2+31 +1/2 (x31 x31 )3+12 +1/2
(11)
Here and henceforth we denote xij = xi xj and also use the abbreviations like 1+23 =
1 + 2 3 , etc. All the other three point functions of different supermultiplet components
are expressed through these via the superprojective invariance SL(2|1) SL(2|1). For example

Wa1 (x1 )Wa2 (x2 )Va3 (x3 ) =
(1 + 2 3 )2 Ca1 ,a2 ,a3

,
(x12 x12 ) 1+23 +1 (x23 x23 )2+31 (x31 x31 )3+12
C a1 ,a2 ,a3
.
(x12 x12 )1+23 +1/2 (x23 x23 )2+31 1/2 (x31 x31 )3+12 +1/2
(12)
Wa1 (x1 )Va2 (x2 )Va3 (x3 ) =
The superconformal symmetry allows to express all three point functions with different components (6) of the supermultiplets through the two constants Ca1 a2 a3 and C a1 a2 a3 .4 Both constants
Ca1 a2 a3 and C a1 a2 a3 are symmetric in all their arguments and related to the structure constants as
(normalizations chosen in (10) are important in these relations)
Qp
Ca1 a2 = Ca1 a2 p ,
Qp
a1 a2 p .
C
a1 a2 = C
(13)
Following [6,12] they have the following explicit form (a stands here for a1 + a2 + a3 )
3 This means even in the left and even in the right sector. Terms oddodd, evenodd, etc., have similar sense.
4 A derivation based on the superprojective Ward identities, which does not exploit a superfield formalism, is presented
in Appendix A.
206

Qb 1b2 (Qa)/b
Ca1 a2 a3 =
b
2
(0) (2a ) (2a ) (2a )
NS
NS
1 NS
2 NS
3
,
NS (a Q)NS (a1+23 )NS (a2+31 )NS (a3+12 )

Qb 1b2 (Qa)/b
C a1 a2 a3 =
b
2
(0) (2a ) (2a ) (2a )
2iNS
NS
1 NS
2 NS
3
,
R (a Q)R (a1+23 )R (a2+31 )R (a3+12 )
where we make use of convenient notations from Ref. [9] for the special functions

x
x +Q
NS (x) = b
b
,
2
2

x+b
x + b1
b
R (x) = b
2
2
(14)
(15)
expressed in terms of the standard in the Liouville field theory upsilon function b (see [10,
13]). For us the following functional relations are important

1 bx
bx
NS (x + b) = b
+
R (x),
2
2

bx
R (x + b) = b1bx
(16)
NS (x)
2
and the same with b replaced by b1 . Finally

1
b
1

NS
(0) = b
b
.
2
2
2b
(17)
All these expressions correspond to the natural normalization of the exponential fields,
where the two-point function reads

DNS (a)
Va (x)Va (0) =
(x x)
2a
(18)
with
bQ
DNS (a) =
2
(Q2a)b1

2
b2 ba 12 b2
.

2
12 + b 2 ab1
(19)
As usual, in the natural normalization the fields Va and VQa are identified through the following
reflection relations
Va = DNS (a)VQa .
(20)
In Refs. [6,12] (see also [9]) expressions (14) were derived on the basis of the singular vector
decoupling in the singular Ramond representation. This approach has many advantages, being
the simplest known and giving simultaneously the OPE structure constants in the whole space of
fields, NeveuSchwarz and Ramond. In this note we find it instructive, however, to rederive (14)
with the help of pure NS bootstrap, requiring the NS null-vector decoupling. Albeit technically
207
more complicated, this derivation allows to stay completely inside the NS sector of SLFT and
provides a good exercise in classical analysis. This program is described in the following section.
Once the structure constants are determined, expression (8) gives directly the integral representation for the four point function of four bottom NS primary fields

Va1 (x1 )Va2 (x2 )Va3 (x3 )Va4 (x4 )

21
= (x41 x41 )
(x24 x24 )
1+324

(x34 x34 )
1+234
(x23 x23 )
4123
a1
a2

a3
x, x ,
a4
(21)
where
x=
x12 x34
x23 x41
(22)
and the reduced four point function admits the following s-channel representation

G
=
a1
a2

a3
x, x
a4

dP Q/2iP Q/2+iP
1 3
1 3
||x Fe
||x
Ca1 a2 Ca3 a4 Fe
2 4
2 4
4

dP Q/2iP Q/2+iP
1 3
1 3
||x Fo
||x .
Ca3 a4 Fo
Ca a
2 4
2 4
4 1 2
(23)
For simplicity we omit possible discrete terms. Superconformal NS blocks Fe and Fo effectively sum up respectively even and odd descendants of the primary NS field of dimension
= Q2 /8 + P 2 /2. Notice the minus sign in front of the second term in (23). It is due to the anticommutativity of the right and left odd chain operators in (9).5 The blocks are constructed
unambiguously on the basis of superconformal invariance. The problem of their evaluation has
been addressed recently in Refs. [14,15] and will be reconsidered below in Sections 3, 4 and 5.
First we recapitulate the so-called recursive c-representation [16], which has been developed
in [14,15] and allows to effectively reconstruct the blocks iteratively as a series in the parameter x. Better convergent is the so-called delta, or elliptic representation, whose recursions
give a much faster convergent series in the elliptic parameter q. Such representation has been
constructed in [17] for the ordinary conformal algebra. Presently in Section 5 we suggest a generalization of the elliptic representation for the N = 1 superconformal case. In this paper we
restrict ourselves only to the blocks with all four external fields bottom supermultiplet components (and those related to this case by the superprojective symmetry).
An important property of a consistent Euclidean quantum field theory is the associativity of
the algebra of operator product expansions. This property ensures the correlation functions to be
single-valued over the Euclidean slice of the complex spacetime. It is commonly believed to
give a Euclidean interpretation of the standard Minkowskian locality. In particular, the four-point
function is single-valued (or, enjoys the crossing symmetry) if the following two identities hold
5 This convention is consistent with the minus sign in the second relation in Eq. (13). The pure imaginary structure
p
constant Ca1 a2 in (14) assures positive sign before the net contribution of the odd levels (see below).
208

G

G
a1
a2
a1
a2

21
a1
a3
G
x, x = (1 x)(1 x)
a4
a2

a1 a2
a3
x, x = G
1 x, 1 x .
a4
a3 a4

x
a4 x
,
,
a3 x 1 x 1
(24)
In the form (23) the first relation is a trivial consequence of the transformation properties of
the blocks. The second, however, becomes a non-trivial relation for the structure constants. Its
closed formulation in terms of the structure constants requires explicit knowledge of the fusion
(or crossing) matrix for a general superconformal block, which is not currently available (see
[18] for an explicit construction in the bosonic case). Under these circumstances, we find it of
value to verify this relation numerically, taking the explicit form of the structure constants and
using the fast convergent elliptic representation for the superconformal blocks. This we perform
in Section 6 and find a reasonable numerical support for the relations (24) for certain randomly
chosen values of the parameters.
Section 7 is devoted to the discussion and outlook.
2. Singular vector bootstrap in NS sector
At certain values of the parameter a the SLFT operator Va is a highest vector of a singular
representation of SVir. In the NS sector, which we only consider in this paper, the simplest singular representation has dimension 1,3 = 1/2 b2 with the primary field Vb as the highest
weight vector. The singular vector appearing at the level 3/2 reads

3
G1/2 + b2 G3/2 Vb .
(25)
There is also a singular vector of the same form in the left SVir sector. Vanishing of all singular
vectors in the singular representations can be taken as the basic dynamical principle of SLFT,
precisely like in the ordinary bosonic LFT. In particular, setting (25) and its left counterpart to
zero leads to non-trivial equations for the correlation functions. It is easy to see, e.g., considering
the three point function with the field Vb , that such decoupling equation restricts the form of
the operator product expansion of a product of this degenerate field and arbitrary primary Va to
the following discrete form
2
2
ab C (a)[Vab ]ee + (x x)
1/2+b C 0 (a)[Va ]oo + C+ (a)(x x)
1ba+b [Va+b ]ee ,
Vb Va = (x x)
(26)
where C (a), C0 (a) and C+ (a) are special (unlike the generic ones in the OPE (8)), or
discrete structure constants. It is instructive to understand how the general OPE (8) with the
structure constants (14) results in the discrete expansion (26) if one of the parameters a1 or a2 is
set to the singular value b. This calculation is performed in Appendix B.
Below in this section we will use the bootstrap conditions together with the decoupling principle to derive unambiguously the generic NS structure constants (14). In particular, the values
of the special structure constants in (26) are recovered uniquely up to an overall scale. It is
instructive, however, to give a perturbative derivation, similar to that presented e.g., in [19],
because firstly it is simpler and more transparent, and secondly it gives a heuristic link with the
SLFT Lagrangian (1), in particular relates the scale parameter to the cosmological constant .
The idea is simply a version of the screening calculus, invented by B. Feigin and D. Fuchs (and
further developed by V. Fateev and V. Dotsenko [20]) where the role of the screening operator is
played by the interaction term 2iWb in (1). E.g., neglecting formally this term, one considers
209
and as free fields, so that the exponentials Vb and Va enjoy the free field fusion to Vab with
C (a) = 1.
(27)
2 1/2 W + requires one insertion of the perturbation

The next
a
termbwith2 [Va ]oo = (2a ) (x x)
2
2ib e d x. Thus

b (y)Va (0)Vb (1)e

(Qa) ()
C 0 (a) = 2ib2 (Q a)2 e
2i (ab b2 )
.
(b2 ) (ab)
(28)
Finally, we need to make two perturbative insertions in order to create the last term [Va+b ]ee =
Va+b + in the OPE (26). This results in

(2ib2 )2
b (y2 )Va (0)Vb (1)VQab ()
b (y1 )e
C+ (a) =
e
2

2

b2 1
b 2
(y1 y2 )(y1 y2 )
(yi yi )ab (1 yi )(1 yi ) d 2 yi
= 22 b4

= 2 2 b 4 2

b2
1
+
2
2
i=1

1 b2
1 b2

+ ab
ab .
2
2
2
2
(29)
In the last calculation we have used the following integration formula

n

1 2
2g
1
(zi zj )(zi z j )
(zi z i )1 (1 zi )(1 z i )
d zi
n!
i>j
i=1

n
= (1 g)
n1

(g + kg) ( + kg) ( + kg) ( + kg),
(30)
k=0
where + + = 1 2(n 1)g. This formula belongs to Dotsenko and Fateev [21].
It is a simple exercise in operator product expansion (see e.g. [19]) to show that the two-point
function (18) satisfies the functional relation
DNS (a)
= C+ (a).
DNS (a + b)
(31)
This is indeed the case for (19) together with (29). The dual functional relation

b2
1 b2
1 b2
DNS (a)
2 2 4 2 1
1
1
=
+
ab
ab
2
2
2
2
2
2
DNS (a + b1 )
(32)
is also satisfied if the dual cosmological constant is related to as

b2
1 b2
1 b2

(33)
.
+
=
+
2
2
2
2
Now, let us turn to the four point function with one singular primary field Vb and three
arbitrary ones. It is natural to renumber the operators, setting in (21) Va1 to be Vb and denoting
Va2 , Va3 and Va4 as V1 , V2 and V3 , their dimensions being 1 , 2 and 3 respectively. Thus, in
210
the notations of (21)

b
g(x, x)
=G
a1

a2
x,
x
.
a3
(34)
It has been shown in [14] that this function satisfies the following third order linear differential
equation
2

1 1 2b2 1 2x
b + 21 b2 + 22 2 3b2 + 21+23
g +
+
+
g +
g
x(1 x)
b2
b2 x(1 x)
x2
(1 x)2

22 (1 + b2 ) 21 (1 + b2 )
+
(1 x)3
x3

21 + (1 2x)(b4 + b2 (1/2 1+23 ) 1+2 )
+
(35)
g = 0.
x 2 (1 x)2
The same equation holds with respect to x.
Near x 0 it has three exponents ba1 , 1 + b2 and
2
1 ba1 + b , which correspond, respectively, to the dimensions a1 b , a1 + 1/2 and a1 +b .
They give rise to three s-channel blocks
Fe() (x) = x a1 b (1 + ),

1
2
Fo(0) (x) = x 1+b
+ ,
2a1
Fe(+) (x) = x 1ba1 +b (1 + ),
2
(36)
where in the brackets stand regular series in x and the normalization correspond to the general
convention of (10). The correlation function is combined as
+ C (a1 )Ca1 b,a2 ,a3 Fe() (x)Fe() (x)
g(x, x)
= C 0 (a1 )C a1 ,a2 ,a3 Fo(0) (x)Fo(0) (x)
+ C+ (a1 )Ca1 +b,a2 ,a3 Fe(+) (x)Fe(+) (x).
(37)
It turns out that (35) is of the type considered by Dotsenko and Fateev in [20] and can be
solved in terms of two-fold contour integrals. Explicitly
Fe() (x)

2
2
x a1 b (1 x)a2 b 12 b2 12 b2 + ba1 ba1 b2 F (A, B, C, g; x)
=
ba1+32 b2 1 ba1+23 ba1+23 b2 ,
(1 b2 ) 12 + ba1+32
2 2+ 2
2
2
2
2
Fo(0) (x)
2
x 1+b (1 x)a2 b b2 (1 ba1 + b2 ) (ba1 )F0 (A, B, C, g; x)

=
1 ba1+32 3 ba1+2+3

2+ 2
2 2 + b2 12 + ba1+23
12 + ba2+31
2
2
(38)
and
Fe(+) (x)

2
2
2
x 1ba1 +b (1 x)a2 b 12 b2 32 ba1 + b2 (1 ba1 )F+ (A, B, C, g; x)
.
=
ba2+31 b2 3 ba1+2+3

2
(1 b2 ) 12 + ba2+31
2 2 2 + b2 1 ba1+2+3
+ b2
2
2
2
(39)
211
Functions F (A, B, C, g; x) and F0 (A, B, C, g; x) are regular series in x and admit the following integral representations
1
F (x) =
t1
dt1
1
t1
F+ (x) =
dt1
0
1
F0 (x) =

B
C
dt2 (t1 t2 )2g (t1 t2 )ABC2 (1 t1 )(1 t2 ) (1 xt1 )(1 xt2 ) ,

C
B
dt2 (t1 t2 )2g (t1 t2 )A (1 t1 )(1 t2 ) (1 xt1 )(1 xt2 ) ,
0
2ABC2g
t1
(1 t1 )B (1 xt1 )C dt1
1

dt2 t2A (1 xt2 )B (1 t2 )C (1 xt1 t2 )2g .
(40)
Parameters A, B, C and g are related to the super-Liouville ones a1 , a2 , a3 and b as

1 ba2+31
1 ba12+3
A= +
,
B = +
,
2
2
2
2
1 b2
1 ba1+2+3
g= .
+ b2 ,
C=
2
2
2
2
(41)
For the sake of completeness in Appendix C we recapitulate from [20] the third order differential
equation of DotsenkoFateev type, relevant integral representations of the solutions, as well as
their monodromy properties.
In particular, the combination (37) is a single-valued function of the two-dimensional coordinate (x, x)
if
C+ (a1 )Ca1 +b,a2 ,a3
C (a1 )Ca1 b,a2 ,a3

2
(ba1 ) (ba1 b2 ) 2 12 b2 + ba1
=
1

b2 2
2 ba1 + 2
ba2+31 b2 3 ba1+2+3

2 2 2 + b2 1 ba1+2+3
+
12 + ba2+31
2
2
2
1 ba1+32 ba1+32 b2 1 ba1+23 ba1+23 b2
2+ 2
2 2+ 2
2
2
2
b2
2

(42)
and

2
2
12 b2 12 b2 + ba1 2 (ba1 b2 )
C 0 (a1 )C a1 ,a2 ,a3
=
C (a1 )Ca1 b,a2 ,a3
b4 (1 b2 ) (ba1 )
3 ba1+2+3

2 2 + b2
12 + ba2+31
2
.
ba

2
3+12
12 b2 ba1+23
b2
2
2
(43)
212
These formulas, together with the explicit expressions (27)(29) for the special structure constants, result in the following functional relations for the three-point functions Ca1 ,a2 ,a3 and
C a1 ,a2 ,a3
Ca1 +b,a2 ,a3
Ca1 b,a2 ,a3

2
(ba1 ) (ba1 b2 ) 12 b2 + ba1
=

2
2
2 2 b4 2 12 + b2 12 b2 a1 b

+ b2 12 + 12 ba2+31 12 ba2+31 12 b2
1 12 ba1+2+3 + 12 b2 32 ba1+2+3
2

12 + 12 ba1+32 12 ba1+32 12 b2 12 + 12 ba1+23 12 ba1+23 12 b2
(44)
and
2i
C a1 ,a2 ,a3
=
Ca1 b,a2 ,a3
1
2

3 ba1+2+3

2 2 + b2
+ ba1 (ba1 b2 ) 12 + ba2+31
2
.

2
2
b4 12 + b2 ba3+12
12 b2 ba1+23
b2
2
2
b2
2
(45)
It is verified directly through the shift relations (16) that the structure constants (14) satisfy
these functional relations. A standard consideration [22] with the dual functional relation (i.e.,
the relation with b b1 and )
now applies and allows to argue that, at least at real
irrational values of b2 , expressions (14) give the unique solution to the singular vector decoupling
equations.
3. Analytic structure of NS block
Here we analyze in more details the properties of the superconformal blocks Fe and Fo which
enter the integral representation (23) of the four-point function. The chain operators introduced
in (9) are mostly important in this analysis. The rightleft factorization allows to concentrate
on the holomorphic part only and then combine it with the (mostly identical) antiholomorphic
one. As it has been found in [15] and [14], apart from the OPE (8) one has to consider the similar
OPE a1 (x)Va2 (0). Commutation relations

Gk , Va (x) = x k+1/2 a (x),

k+1/2 2a (k + 1/2)
Va (x) +
Va (x)
Gk , a (x) = x
x
x

(46)
follow directly from (7). Acting by any generator Gk with k > 0 on the left-hand side of (8) one
finds
a1 (x)Va2 (0)

Q/2+iP
dP
1/2
=x
Q/2+iP (0)
(x x)
1 2 Ca1 ,a2 Co (a , x)Ce (a , x)V
4

Q/2+iP
+C
Ce (a , x)Co (a , x)V
Q/2+iP (0) ,
a1 ,a2
(47)
213
where Ce (a , x) and Co (a , x) are new chain operators.6 It is convenient to unify the even and
odd ones as the joint series in integer and half-integer powers of x
C(, x) = Ce (, x) + Co (, x),
C(,
x) = Ce (, x) + Co (, x).
(48)
From (46) one finds
C(,
x)V = x k Gk C(, x)V ,

x)V .
Gk C(,
C(, x)V = x + 21 k 2 k + x
x
For the coefficients in the level expansion

x N CN (),
C(,
x) =
x N CN (),
C(, x) =
N
(49)
(50)
where N runs over non-negative integer and half-integer numbers (levels), these relations read
(k N )
Gk CN ()V = CNk ()V ,
Gk CN ()V = ( + 21 k 2 + N k)CNk ()V .
(51)
This system turns out to be enough to reconstruct the chain operators up to two overall constants.
The latter can be fixed by the conditions C0 = 1 and C0 = 1. Explicitly one finds, level by level
1
G1/2 ,
2
+ 1 2
L1 ,
C1 () =
2
+ 1 2 + 1/2 3
2(1 2 )
O3/2 ,
G1/2 +
C3/2 () =
2
2(2 + 1)
4 + 2c
6 + c
( + 1 2 )( + 1 2 + 1) 2
C2 () =
L1
4(2 + 1)
2(1 2 )2 + (2 + 1)(1 + 2 )
G1/2 O3/2
+
(2 + 3)(42 + 2c
6 + c)
3(1 2 )2 2(1 + 2 ) 2
+
O2
2(2 + 3)(16 + 3c 3)
C1/2 () =
(52)
and
+ 1 2
C1/2 () =
G1/2 ,
2
+ 1 2 + 1/2
L1 ,
C1 () =
2
( + 1 2 )( + 1 2 + 1) 3
G1/2
C3/2 () =
2(2 + 1)
6 These operators depend also on the dimensions = and = . For not to overload the notations, this
a1
a2
1
2
dependence is not indicated explicitly.
214
2(1 2 )2 + (2 + 1)(1 + 2 )
O3/2 ,
42 + 2c
6 + c
( + 1 2 + 1/2)( + 1 2 + 3/2) 2
L1
C2 () =
4(2 + 1)
2(1 2 )( + 1 2 + 3/2)
G1/2 O3/2
+
(2 + 3)(42 + 2c
6 + c)
12(1 2 )2 (2 + 3)(2 1 + 4(1 + 2 ))

O2 ,
8(2 + 3)(16 + 3c 3)
(53)
where we have used the abbreviations

2
L1 G1/2 G3/2 ,
2 + 1
O2 = 3L21 4L2 3G3/2 G1/2 .
O3/2 =
(54)
Further coefficients are systematically evaluated level by level as a solution of (49). At each
level (51) is a finite dimensional linear problem, the coefficients being polynomials in c,
and
the external dimensions 1 and 2 . The determinant of the system is the Kac determinant [23]
of the corresponding level in the representation of the superconformal algebra. Therefore, the
singularities of C() are (simple in general) poles located at the singular dimensions = m,n ,
where (m, n) is a pair of positive integers,
Q2 2m,n
8
2
and we introduced a notation
m,n =
(55)
mb1 + nb
(56)
.
2
In the NS sector, which we only consider here, m and n are either both even or both odd. At this
values in the SVir module over V appears a singular vector Dm,n Vm,n , a primary field of dimension m,n + mn/2 = m,n . Here, as in Ref. [24], we denoted Vm,n = Vm,n and introduced
a set of singular vector creation operators Dm,n , which are graded polynomials in the generators Gk and Ln (with the coefficients depending in c (or b) only) such that Gk Dm,n Vm,n = 0
for every half integer k > 0. These operators are supposed to be normalized as in [24]
m,n =
Dm,n = Gmn
1/2 + .
(57)
Below we will need also the conjugate operator Dm,n

defined, as in Ref. [24] through the
following conjugation rules Ln = Ln and Gk = Gk . Apparently

Dm,n
(58)
Dm,n V (0) = rm,n
( m,n )V (0) + O ( m,n )2 V (0).

(the logarithmic norm of the singular vector) has been explicitly evaluated
The coefficient rm,n
in [24]

= 2mn1
rm,n
m,n,k+l2Z

k=1m,l=1n
(k,l)=(0,0),(m,n)
k,l .
(59)
215
The singular vector Dm,n Vm,n , once appeared in the chain vectors (48), gives rise to its own
m,n , x)Dm,n Vm,n which by themselves satisfy the chain
chains C(m,n , x)Dm,n Vm,n and C(
equations (51) with = m,n . Apparently
res C(, x) = x mn/2 Xm,n C(m,n , x)Dm,n ,
=m,n
(60)
where Xm,n are certain coefficients. Since Ce (, x) with Co (, x) and Co (, x) with Ce (, x)

form two independent systems, we need to treat separately the integer and half-integer chains.
Denote

(e)
mn/2 Xm,n Ce (m,n , x)Dm,n , m, n even,
res Ce (, x) = x
(e)
=m,n
Xm,n Co (m,n , x)Dm,n , m, n odd,

(o)
Xm,n Co (m,n , x)Dm,n , m, n even,
res Co (, x) = x mn/2
(61)
(o)
=m,n
Xm,n Ce (m,n , x)Dm,n , m, n odd,
(e)
(o)
and Xm,n
are then
and let C(, x) with C(,
x) be normalized as above. The coefficients Xm,n
uniquely defined. By construction they are polynomials in the external dimensions 1 and 2 .
To describe these coefficients it is convenient to define the following fusion polynomials [14,
15]
(e)
(x) =
Pm,n
m+nkl
mod 4=0
(x k,l ),
k{1m,2,m1}
l{1n,2,n1}
(o)
Pm,n
(x) =
m+nkl
mod 4=2
(x k,l ).
(62)
k{1m,2,m1}
l{1n,2,n1}
Here e.g. {1m, 2, m1} means from 1m to m1 with step 2, i.e., 1m, 3m, . . . , m1.
The degree of these polynomials
(e,o)
pe,o (m, n) = deg Pm,n
(x)
(63)
coincides with the number of multipliers in the products (62)

pe,o (m, n) = mn/2
pe (m, n) = mn/2 1/2
po (m, n) = mn/2 + 1/2
at m, n even,
at m, n odd.
(64)
(e,o)
In particular, the parity of Pm,n (x) is that of the integer pe,o (m, n). In the current context it turns
out convenient to parameterize the external dimensions in terms of new variables i as
i =
Q2 2i
.
8
2
(65)
We will need some more notations. Consider a three point function of formal chiral fields,
V1 (1)V2 (0)
and Vm,n ()Dm,n
V1 (1)V2 (0)
. Application of the Ward
like Vm,n ()Dm,n
identities, which read in this context simply as
216

Gk , V1 (1) = 1 (1),

Gk , 1 (1) = (2k1 2 + m,n + N )V1 (1)
(66)
(N is some integer or half integer, corresponding to the level of the descendent of Vm,n ()) is a
purely algebraic procedure and leads to the relations

Vm,n ()Dm,n
V1 (1)V2 (0)

Vm,n ()V1 (1)V2 (0)
at m, n even,
(e)
= Ym,n (1 , 2 )
Vm,n ()1 (1)V2 (0)
at m, n odd,

Vm,n ()Dm,n
1 (1)V2 (0)

Vm,n ()1 (1)V2 (0)
at m, n even,
(o)
= Ym,n (1 , 2 )
(67)
Vm,n ()V1 (1)V2 (0)
at m, n odd.
(e,o)
Apparently the Ym,n (1 , 2 ) are polynomials in 1 and 2 , the leading order term being generated by the term in Dm,n with maximum number of generators Gk , i.e., the one quoted in (57).
(e,o)
Thus the degree of Ym,n (1 , 2 ) is pe,o (m, n) with leading term
(e,o)
Ym,n
(1 , 2 ) = (1 2 )pe,o (m,n) + lower order terms.
(68)
On the other hand, the standard decoupling consideration requires this polynomials to be propor(e,o)
(e,o)
tional to the product Pm,n (1 + 2 )Pm,n (2 1 ). The last is apparently a polynomial in 1
and 2 with leading order term (21 22 )pe,o (m,n) and therefore
(e,o)
(e,o)
(e,o)
Ym,n
(1 , 2 ) = 2pe,o (m,n) Pm,n
(1 + 2 )Pm,n
(2 1 ).
(69)
Consider first the case m, n even and the following products

V ()Dm,n
Ce (, x)V (0)
(e)
x mn/2 Xm,n
V ()Dm,n
Ce (m,n , x)Dm,n V (0) + O( m,n ),
m,n
G1/2 Co (, x)V (0)

V ()Dm,n
(o)
x mn/2 Xm,n
V ()Dm,n
G1/2 Co (m,n , x)Dm,n V (0) + O( m,n ).
m,n
(70)
The estimates O( m,n ) follow from the observation, that all non-polar terms in Ce (, x) or
at = m,n . Similarly, at m, n odd we have

G1/2 Co (, x) are orthogonal to V ()Dm,n

V ()Dm,n
Co (, x)V (0)
(o)
x mn/2 Xm,n
V ()Dm,n
Ce (m,n , x)Dm,n V (0) + O( m,n ),
m,n
G1/2 Ce (, x)V (0)

V ()Dm,n
(e)
x mn/2 Xm,n
V ()Dm,n
G1/2 Co (m,n , x)Dm,n V (0) + O( m,n ).
m,n
(71)
217
Then the right-hand sides are evaluated using (58) and compared with the left-hand sides coming
from (67). This results in [14,15]
(e,o)
(e,o)
Xm,n
=
Ym,n (1 , 2 )
.

rm,n
(72)
Notice, that a simple change of the notations turns the above consideration to a proof of the
residue formula for the complementary chain C(,

x)

(o)
Ce (m,n , x)Dm,n , m, n even,
Xm,n
res Ce (, x) = x mn/2
(o)
=m,n
Xm,n Co (m,n , x)Dm,n , m, n odd,

(e)
Xm,n Co (m,n , x)Dm,n , m, n even,
res Co (, x) = x mn/2
(73)
(e)
=m,n
Xm,n Ce (m,n , x)Dm,n , m, n odd.
These simple analytic properties are inherited by the corresponding superconformal blocks.
Define the odd and even blocks as

1 3
1 ,2
Fe,o
(74)
||x = x 1 2 V4 ()V3 (1)Ce,o
(, x)V (0) ,
2 4
where again V are formal (chiral) primary fields and we have restored the implicit dependence
of the chain operator on the external dimensions. Normalization is fixed by

V4 ()V3 (1)V (0) = 1,

V4 ()V3 (1)G1/2 V (0) = 1.
(75)
1 ,2
() turn to the poles of the blocks, the residues being evaluated similarly (to
The poles of Ce,o
be more compact we suppress the external dimensions in the arguments of the blocks)

Fe (m,n , x) at m, n even,
1 3
(e)
res Fe (, x) = Bm,n
F
=m,n
e (m,n , x) at m, n odd,
2
4

Fo (m,n , x) at m, n even,
1 3
(o)
res Fo (, x) = Bm,n
(76)
Fe (m,n , x) at m, n odd.
2 4
=m,n
Matrix elements V4 ()V3 (1)Dm,n Vm,n (0)

are evaluated by the same algebraic procedure

V4 ()V3 (1)Dm,n Vm,n (0)

V4 ()V3 (1)Vm,n (0)
at m, n even,
(e)
(3 , 4 )
= Ym,n
V4 ()V3 (1)Vm,n (0)
at m, n odd,

V4 ()V3 (1)D m,n Vm,n (0)

V4 ()V3 (1)Vm,n (0)
at m, n even,
(o)
= Ym,n (3 , 4 )
(77)
V4 ()V3 (1)Vm,n (0)
at m, n odd,
where D m,n = G1/2 Dm,n and Vm,n (0) = G1/2 Vm,n . Combining all together we find [14,15]

(e,o)
Bm,n
1
2
3
4
(e,o)
(e,o)
Ym,n (1 , 2 )Ym,n (3 , 4 )
.

rm,n
(78)
218
Important symmetries

(e,o) 1 3
(e,o) 3 1
(e,o) 2
= Bm,n
= Bm,n
Bm,n
2 4
4 2
1

(e,o) 1 3
(e,o) 2 3
Bm,n
= ()pe,o (m,n) Bm,n
2 4
1 4
4
3

,
(79)
follow directly from the symmetry properties of the fusion polynomials.

4. c-recursion
Analytic properties observed in the previous section give rise to convenient relations for the
superconformal blocks Fe (, x) and Fo (, x) (we suppress sometimes the explicit dependence
on the external dimensions) which allow their simple recursive evaluation, e.g. as a power series
in x. The first way is to consider analytic properties in the central charge c of the superconformal
algebra [14,15]. In this case the Kac dimensions (55) appear as (again in general simple) poles
in c at
1
c = cm,n () = 5 + 2Tm,n () + 2Tm,n
(),
(80)
where again (m, n) is a pair of natural numbers, both even or both odd, while Tm,n () is a root
of the quadric (55) in b2

1 4 mn + 162 + 8(mn 1) + (m n)2
,
Tm,n () =
n2 1

1 4 mn 162 + 8(mn 1) + (m n)2
1
.
Tm,n () =
(81)
m2 1
For this particular root all singularities corresponding to m = 1 are sent to infinity and only
the pairs with m > 1 count. Notice that the root chosen is non-singular at n = 1 so that (81) is
understood as
Tm,1 () =
m2 1
.
2(1 4 m)
(82)
Corresponding residues are read off from those in (78) when being expressed in terms of . In
the present context it is convenient to use the symmetry of the polynomials (62) to make the
multipliers in the residues (69) explicit polynomials in 1 and 2 . At m, n even the multipli(e,o)
(x) always enter in pairs (x k,l )(x k,l ) and one can fold the product
ers in Pm,n
(e,o)
(e,o)
p
(m,n)
2 e,o
Pm,n (1 2 )Pm,n (1 + 2 ) as follows
(e,o)
(1 , 2 ) =
Ym,n
m+nkl
mod 4=0,2

(m,n)
Yk,l
(1 , 2 , ),
(83)
k{1,2,m1}
l{1n,2,n1}
where

1 (m,n)
2
(m,n)
(m,n)
(1 , 2 , ) = (1 2 )2 + (m,n)
k,l ()
Yk,l
k,l () 1 + 2 1,1 () +
4
(84)
219
and
1 () + 2kl + l 2 T
k 2 Tm,n
m,n ()
(85)
.
4
Similar folding is possible also at m, n odd if the degree of Pm,n is even. If it is odd, the term
with (k, l) = (0, 0) leads to an extra multiplier 1 2 .
Once the multipliers in (78) are expressed in terms of , the corresponding residues in c read

Fe (cm,n , + mn/2, x) at m, n even,
(e)
res Fe (c,
, x) = B m,n
()
Fo (cm,n , + mn/2, x) at m, n odd,
c=
cm,n

Fo (cm,n , + mn/2, x) at m, n even,
(o)
res Fo (c,
(86)
, x) = Bm,n ()
F
e (cm,n , + mn/2, x) at m, n odd,
c=
cm,n
(m,n)
k,l () =
where
(e,o)
(e.o)
() = Bm,n
()
B m,n
1 ())
16(Tm,n () Tm,n
1
(n2 1)Tm,n () (m2 1)Tm,n
()
(87)
the last fraction corresponding to c/
along the (m, n) Kac quadric (55).

The asymptotic of Fe,o (c,
, x) at c has been recovered in [14,15]. Analytic properties
in c sum up to the following relations
Fe,o (c,
, x) = fe,o (, x) +
+

m,n odd
m>1
(e,o)
B m,n ()
Fe,o (cm,n , + mn/2, x)
c cm,n ()
m,n even
(e,o)
B m,n ()
Fo,e (cm,n , + mn/2, x),
c cm,n ()
(88)
where
fe (, x) = x 1 2 2 F 1 ( + 1 2 , + 3 4 , 2, x),

1 1 2 +1/2
1
1
x
fo (, x) =
2 F 1 + 1 2 + , + 3 4 + , 2 + 1, x .
2
2
2
(89)
Eqs. (88) apparently can be used for recursive evaluation of the coefficients in the series
expansions

(N )
Fe,o (c,
(90)
, x) = x 1 2
x N Fe,o
(c,
),
k
where the sum is over non-negative integer (for Fe ) or half-integer (for Fo ) numbers
(N )
(N )
Fe,o
(c,
) = fe,o
() +
mn/2N

m,n even
mn/2N

m,n odd
m>1
(e,o)
() (Nmn/2)
B m,n
(cm,n , + mn/2)
Fe,o
c cm,n ()
(e,o)
B m,n () (Nmn/2)
(cm,n , + mn/2).
Fo,e
c cm,n ()
(91)
220
In (91)
( + 1 2 )N ( + 3 4 )N
,
N!(2)N
( + 1 2 + 1/2)N1/2 ( + 3 4 + 1/2)N1/2
.
fo(N ) () =
(N 1/2)!(2)N+1/2
fe(N ) () =
(92)
5. Elliptic recursion
For practical calculations another relation turns out to be much more convenient. This relation
is called the elliptic recursion (since it requires a parametrization in terms of the elliptic functions)
or, sometimes, the -recursion, because it is based on the analytic properties of the blocks in
instead of c.
As in Ref. [17] we understand the variable x as the modulus of the elliptic curve
y 2 = t (1 t)(1 xt) and introduce the ratio of its periods
=i
K(1 x)
,
K(x)
(93)
where
1
K(x) =
dt
2y(t)
(94)
is the complete elliptic integral of the first kind. Let also q = exp(i ) and denote in the standard
way
3 (q) =

n=
n2
q ,
2 (q) =
q (n+1/2)
(95)
n=
so that
x=
24 (q)
34 (q)
(96)
inverts (93). This elliptic parametrization has important advantages. Eq. (96) maps the half plane
Im > 0 to the universal covering of the x-plane with punctures at 0, 1 and . The power expansions (90) of the blocks in x converge inside the disk |x| < 1. Once reexpanded in q it converges
inside |q| < 1, i.e., on the whole covering and therefore gives there a uniform approximation.
Naturally, even in the region |x| < 1 it is expected to converge faster. The elliptic recursion,
which we describe now, gives the blocks directly in terms of the elliptic variable and allows to
generate the q-series in a simple way.
Define the elliptic blocks He,o (, q) through the relations

1 3
||x
Fe
2 4

2
2
x Q /81 2 (1 x)Q /81 3
2
1 3
H
||q
,
= (16q)Q /8

e
2 4
4 4 3Q2 /2
3 i=1 i
(q)

1 3
||x
2Fo
2 4

Q2 /8 x
= (16q)
Q2 /81 2 (1 x)Q2 /81 3

4
4
i=1 i 3Q
2 /2

Ho
(q)
1
2

3
||q ,
4
221
(97)
where the parametrization (65) of the external dimensions i in terms of i is implied. In order,
the elliptic blocks satisfy the following relations (the elliptic recursion)

He (, q) = 3 q 2 +
(e)
q mn/2 Rm,n
He (m,n , q)
m,n
m,n even
(e)
q mn/2 Rm,n
Ho (m,n , q),
m,n
m,n odd
Ho (, q) =
(o)
(o)
q mn/2 Rm,n
q mn/2 Rm,n
Ho (m,n , q) +
He (m,n , q)
m,n
m,n
m,n even
(98)
m,n odd
the residues reading simply

(e,o)
(e,o)
=
Rm,n
(e,o)
(e,o)
(e,o)
Pm,n (1 + 2 )Pm,n (1 2 )Pm,n (3 + 4 )Pm,n (3 4 )

.

rm,n
(99)
These relations take into account the analytic properties of the superconformal blocks in described in Section 3. In addition they imply that in the limit

He (, q) = 3 q 2 + O 1 ,

Ho (, q) = 1 2 q 2 + O 2 .
(100)
The first asymptotic is plugged explicitly into the relations, while the second is automatically
generated by the recursion. To derive the relations (98) we need to justify the asymptotic
of the blocks, summed up in (97) and (100). The arguments will be reported elsewhere.
Like the c-recursion
of Section 4, relations (98) allow to evaluate recursively the series expansions of the elliptic blocks in powers of q

)
He.o (, q) =
(101)
q N h(N
e,o ().
N
Again He (, q) expands in non-negative integer powers of q while Ho (, q) is a series in positive half-integer ones. Relations (98) give
)
(N )
h(N
e,o () = e,o
mn/2N

m,n even
(N )
(N )
(e,o) (Nmn/2)
Rm,n he,o
(m,n )
+
m,n
mn/2N

m,n odd
(e,o) (Nmn/2)
Rm,n ho,e
m,n )
,
m,n
(102)
where o = 0 and e are coefficients in the q-expansion of 3 (q 2 ). In practice this relation

allows a much better algorithm as compared to the c-recursion
(91), mostly because c is fixed

(e,o)
and the values Rm,n , m,n remain the same at all iteration steps (unlike (91), where at each level
they have to be recomputed for the shifted values of ). Once a necessary number of residues
and dimensions is available, the remaining recursive procedure runs very fast.
222
6. Superconformal bootstrap
With the structure constants (14) and superconformal blocks known one is in the position to
evaluate the four-point function (23) of basic NS fields in SLFT. The goal of this section is to
verify numerically the crossing symmetry relations (24).
For the reasons discussed above, we will use the elliptic representation (97) of the blocks. The
four-point function (23) acquires the form

2
2
Q /81 3
(x x)
Q /81 2 [(1 x)(1 x)]
a1 a3
a1 a3

x,
x
=
,
,
g
G
2
a2 a4
a2 a4
[3 (q)3 (q)]
4 i i 3Q /2
(103)
where

dP
2 Q/2+iP Q/2iP
a1 a3
g
,
|16q|P Ca1 ,a2 Ca3 ,a4 He (, q)He (, q)

a2 a4
4

Q/2+iP
Q/2iP
Ho (, q)Ho (, q)
C
(104)
C
a1 ,a2
a3 ,a4
and = Q2 /8 + P 2 /2. The first of the relations (24) is verified analytically. Indeed, the identities

1 3
1 4
He
|| q = He
||q ,
2 4
2 3

1 3
1 3
i
i/2
||e q = e
Ho
||q
Ho
(105)
2 4
2 4
for the elliptic blocks (the latter are easily derived from (98) and the symmetries (79) of the
residues) directly result in

a1 a3
a1 a4
g
(106)
, = g
+ 1, + 1 .
a2 a4
a2 a3
The second relation in Eq. (24) in terms of the function g reads

1
a1 a2 1
a1 a3
3Q2 /42 i i
, = ( )
,
g
.
g
a2 a4
a3 a4
(107)
It is a difficult mathematical problem to recover this property from the representation (34). However, with the fast algorithms based on the elliptic representation of the blocks, it is an affordable
problem for numerical analysis.
As a simplest numerical test we have chosen the external parameters a1 = a2 = a3 = a4 =
Q/2. The structure constants (14) vanish for these values of the external parameters, so that we
take a first derivative in all four of them. This means that we consider the four point function of

= exp(Q/2). Denoting
the primary fields VQ/2

4
a1 a3
4

(108)
,
g
= 4NS
(0)R2 (b)f (, )

a2 a4
a1 a2 a3 a4
a =Q/2
i
we find
f (, ) =

2
2

dP
0 0
0 0
P2

|16q|
re (P )He
||q + ro (P )Ho
||q .
0
0
0
0
4
(109)
223
Fig. 1. The results of comparison for different values of b.
The auxiliary functions re (P ) and ro (P ) read in terms of standard upsilon functions b (x)
re (P ) =
ro (P ) =
b (iP )b (iP )b2 (iP + Q/2)

b8 (Q/4 + iP /2)b8 (Q/4 iP /2)
b (iP )b (iP )b2 (Q/2 + iP )
(110)
b8 (Q/4 + b/2 + iP /2) 8 (Q/4 + b1 /2 + iP /2)
and allow the following integral representations

dt
(1 + 6b2 + b4 )et
re (P ) = P 2 exp
t
2b2
0
8 cos(P t/2) cosh[(b + 1/b)t/4] 2 cos(P t) cosh2 [(b 1/b)t/4] 6

+
sinh(t/2b) sinh(bt/2)

ro (P ) = P 2 exp
dt

,
(111)
(1 2b2
+ b4 )et
2b2
8 cos(P t/2) cosh[(b 1/b)t/4] 2 cos(P t) cosh2 [(b 1/b)t/4] 6

+
sinh(t/2b) sinh(bt/2)

.
(112)
Q2 /4
In Fig. 1 the values of f (, ) (solid curves) are compared with that of ( )

f (1/, 1/ ) (symbols) for pure imaginary = it and certain values of the parameter b.
To give an idea about the accuracy, some numbers are shown in Table 1. These numbers correspond to the approximation of the elliptic blocks He (q) and q 1/2 Ho (q) as power series up to the
order q 6 . It seems like the main source of the discrepancy is in this approximation. For example
at t = 0.5 (q 2 0.0432) the numbers differ in the fifth decimal digit (e.g., the numbers 7.12534
224
Table 1
Numerical data for f (t) for b = 0.8 and b = i/4
t
f (t) for b = 0.8
2
t Q /2 f (1/t) for b = 0.8
f (t) for b = i/4
2
t Q /2 f (1/t) for b = i/4
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
25.900821
15.372272
11.123984
8.7059097
7.1251188
6.0067008
5.1733043
4.528805
4.0160968
25.456246
15.365205
11.123971
8.7060123
7.125178
6.0067341
5.1733228
4.5288147
4.0161008
1.5305811
1.5884637
1.5267458
1.4398383
1.3522805
1.2708572
1.1971168
1.1308767
1.0713992
1.5232342
1.5881483
1.5264759
1.4396855
1.3521968
1.2708109
1.1970912
1.1308632
1.0713936
and 7.12512 in the table), while the approximation improved up to the order q 10 gives seven
correct decimal digits (respectively 7.12511575 and 7.12511599).
We have performed a similar comparison for some different values of a1 , a2 , a3 , a4 (chosen
at random, but close enough to Q/2 to preserve the convergence of the representations (111) and
(112)). E.g., at
9
2Q
Q,
a3 =
,
14
3
we have for b = 2.8 and t = 0.3
a1 =
a2 =
13Q
,
30
a4 =
7Q
10
(113)
fa1 ,a2 ,a3 ,a4 (t) = 2141.5325,

t 3Q
2 /24
fa1 ,a3 ,a2 ,a4 (1/t) = 2141.5101.
(114)
7. Discussion
In the present paper a preliminary analysis of the bootstrap properties has been performed
for the four point functions in the N = 1 supersymmetric Liouville field theory. Some, mostly
numerical, arguments are presented that the NS operator algebra, based on the structure constants (14), satisfies the locality property. Of course, a separate analysis is needed to include the
Ramond sector. This problem remains for future work.
Even in the purely NS sector the program is not yet completely finished. We considered
only the four point functions of four bottom components V of any supermultiplet. Correlation
functions involving other components, like V V V W
or V V
, etc., remain to be studied.
Obviously they can be expressed in terms of the structure constants (14), like in Eq. (23) for
V V V V
, with different superconformal blocks. We arrive at the problem to describe completely
the set of 32 different blocks like (in obvious notations)

1 3
||x = V4 ()3 (1), V1 (x)V2 (0) ,
Fe,o
2 4

1 3
||x = V4 ()3 (1), 1 (x)2 (0) , etc.,
Fe,o
(115)
2 4
where in the present context V and are formal chiral components of the right supermultiplet,
and, by definition

1 ,2
(, x)V (0)
V4 ()3 (1), V1 (x)V2 (0) = V4 ()3 (1)Ce,o
(116)
225
(similarly for other components).

The problem is slightly less involved than it seems at first glance. A study of the superprojective Ward identities, similar to that in Appendix A, allows to reduce the set to 8 independent
functions. Moreover, their analytic properties follow directly form those of the basic chains
Ce,o (, x) and Ce,o (, x). In particular, the residues at the singular dimensions = m,n are
expressed similarly to (78) in terms of the fusion polynomials. Construction of the recursive representations, analogous to (88) or (98), presents a separate problem, which we hope to analyze
in the future.
Another topic of a future report is the justifications of the asymptotic (97) and (100) of the
block at , which has been simply conjectured in Section 5 in order to write down the
elliptic recursion relations. This analysis requires several essential steps and presently we prefer
to skip it.
The superconformal blocks, considered above, are expected to satisfy certain crossing, or
fusion relations, the set with different values of the intermediate dimension forming an infinite
dimensional representation of the monodromy group. In the present case the group is equivalent
to the modular group generated by the maps x x/(x 1) and x 1x. In terms of the elliptic
parameter these are + 1 and 1/ . The representation is completely determined
by the relations

xei
1 3
1 4
i(1 2 )
21
Fe
||x = e
(1 x)
Fe
||
,
2 4
2 3
1x

xei
1 3
1 4
Fo
(117)
||x = ei(1 2 +1/2) (1 x)21 Fo
||
2 4
2 3
1x
and

F
1
2

dP
3
1
|P |x =
K
4
4
2

=e,o

1
3
|P , P F
4
3

2
|P |x .
4
(118)
Here we introduced an index = e, o to unify the notations, and K (P , P ) is the kernel of a

certain integral operator, called the crossing matrix (or, sometimes, the fusion matrix). In our
construction the first relation follow directly from the symmetries (105) of the elliptic blocks.

The second one is very non-trivial, the crossing matrix K (P , P ) remaining to be found. In
the case of the ordinary (non-supersymmetric) conformal block an explicit expression for the
crossing matrix has been conjectured in Ref. [18] on the basis of an appropriate quantum group
analysis. Similar analysis seems to be feasible in the present superconformal situation.
Finally, we hope that the construction of the SLFT four point function, started in the present
paper, will turn useful for the applications in the Liouville supergravity as well as in non-critical
superstring theory.
Acknowledgements
V. Belavin thanks sincerely G. Mussardo and G. Delfino for their hospitality at SISSA and
valuable discussions. His work was partially supported by the MIUR programme Quantum field
theory and statistical mechanics in low dimensions and also by the grant RFBR 05-01-01007.
A. Belavin acknowledges the hospitality of MPIM during his visit in FebruaryMarch 2007 and
a support by the grants RFBR 07-02-00799, RFBR-JSPS 05-01-02934, SS-2044-2003 and by
226
the program RAS Elementary particles and Fundamental nuclear physics. Also he is grateful
to R. Poghossian and L. Hadasz for useful discussions. Both A. Belavin and V. Belavin would
like to express their sincere gratitude to G. Delfino, G. Mussardo, G. von Gehlen, R. Flume,
A. Klumper and T. Miwa, whose help made their collaboration possible at different stages of the
work. Work of Al.Z. was sponsored by the European Committee under contract EUCLID HRPNCT-2002-00325. An important part has been made during his visit at the Department of Physics
and Astronomy of the Rutgers University. Hospitality and stimulating scientific atmosphere of
the Theory Group are greatly appreciated, as well as many discussions with A. Zamolodchikov.
Appendix A. Superprojective invariance. Three point function
In this appendix we consider only Ward identities related to the right superconformal algebra, formed by the holomorphic components S(z) and T (z). Respectively, supermultiplets consist
of the highest weight vectors Vi or i defined as in (6) (in this appendix we omit the parameter a
near the primary field to give place for the identification number 1, 2, etc.) and the top components i or Wi . To be definite, we will talk about the multiplet (Vi , i ). It seems to require
no comments how to combine the results below to the complete holomorphicantiholomorphic
combinations.
To describe all 23 possible three point functions denote
C123 = V1 V2 V3
,
C123
= 1 V2 V3
,
etc.
(A.1)
From the operator product expansions (7) we have the following supercurrent Ward identities

21
1
1
1
S(z)1 V2 V3 =
+
C1 23
C ,
C123

2
z x1 x1
z x2
z x3 123
(z x1 )

22
1
1
1
S(z)V1 2 V3 =
+
C1 23
C ,
C123 +

2
z x2 x2
z x1
z x3 123
(z x2 )

23
1
1
1
S(z)V1 V2 3 =
(A.2)
+
C12
C ,
C123 +
3 +
2
z x3 x3
z x1
z x2 123
(z x3 )
where 1 , 2 and 3 are respectively the dimensions of V1 , V2 and V3 . As S(z) = O(z3 ) at
z , the following super projective identities hold
C123 = C1 23
+ C12
3 ,
x1
C123 = C1 23
+ C12 3 ,
x2
C123 = C12
3 C12 3 ,
x

3
+ 21 C123 = x2 C1 23
x1
+ x3 C12
3 ,
x1

x2
+ 22 C123 = x1 C1 23
+ x3 C12 3 ,
x2

x3
+ 23 C123 = x1 C12
3 x2 C12 3 .
x3
(A.3)
227
These identities involve only the correlation functions with even number of fermions i . Eliminating the derivatives one finds
21 C123 = x12 C1 23
+ x31 C12
3 ,
22 C123 = x12 C1 23
x23 C12 3 ,
23 C123 = x31 C12
3 x23 C12 3 ,
(A.4)
and thus
2 + 3 1
C123 ,
x23
1 + 3 2
C123 ,
C12
3 =
x31
1 + 2 3
C1 23
C123 .
=
x12
C12 3 =
(A.5)
Being substituted to differential equations this sums up to
1 + 2 3
1 + 3 2
C123 =
C123
C123 ,
x1
x12
x13
1 + 2 3
2 + 3 1
C123 =
C123
C123 ,
x2
x21
x23
1 + 3 2
2 + 3 1
C123 =
C123
C123
x3
x31
x32
(A.6)
and gives finally

C123 =
C
+ + +
x121 2 3 x232 3 1 x311 3 2
(A.7)
where C is an integration constant, independent on x1 , x2 and x3 .

Another Ward identity, relevant for the even in fermions functions, is

21
1
22
1
S(z)1 2 3 =
+
+
C
C12
3
12 3
(z x1 )2 z x1 x1
(z x2 )2 z x2 x2

23
1
+
+
(A.8)
C1 23
.
2
z x3 x3
(z x3 )
In the same manner it gives
C12 3
C12
C = 0,
3 +
x
x2
x3 123
1

21 + x1
C12 3 22 + x2
C12
C1 23
3 + 23 + x3
= 0.
x1
x2
x3
(A.9)
It is straightforward to verify that these relations are satisfied identically with the explicit expressions (A.5).
For the odd fermion number functions consider the Ward identity

S(z)V1 V2 V3 =
1
1
1
C +
C +
C .
z x1 123 z x2 123 z x3 123
(A.10)
228
It follows that
C123
+ C123 = 0,
+ C123
x1 C123
+ x3 C123 = 0.
+ x2 C123
(A.11)
This system is solved in terms of a single function C 123
C123
= x23 C123 ,
C123
= x31 C123 ,
C123 = x12 C 123 .
Next, we need the identities

1
22
1
C123
C +
+
S(z)V1 2 3 =
z x1 123
(z x2 )2 z x2 x2

23
1
C123
+
,
2
z x3 x3
(z x3 )

1
21
1
C123
S(z)1 V2 3 =
C +
+
z x2 123
(z x1 )2 z x1 x1

23
1
C123
+
,
2
z x3 x3
(z x3 )

21
1
1
S(z)1 2 V3 =
C1 2 3 +
+
C123
z x3
(z x1 )2 z x1 x1

22
1
C123
+
.
2
z x2 x2
(z x2 )
(A.12)
They result in the relations
C
C + C1 2 3 = 0,
x2 123 x3 123
C123
C C1 2 3 = 0,
x1
x3 123
C
C + C1 2 3 = 0,
x1 123 x2 123

22 + x2
C123 23 + x3
C123
+ x1 C1 2 3 = 0,
x2
x3

C123 23 + x3
C123
21 + x1
x2 C1 2 3 = 0,
x1
x3

C123
C123
21 + x1
+ x3 C1 2 3 = 0.
22 + x2
x1
x2
(A.13)
All of them are satisfied by

C 123 =
C
+2 3 +1/2 2 +3 1 +1/2 1 +3 2 +1/2
x23
x31
x121
(A.14)
229
and
with a new integration constant C,
C1 2 3 = (1/2 1 2 3 )C 123 .
(A.15)
Appendix B. General OPE and special structure constants

It is instructive to show how the general continuous OPE (8) turns to the discrete one (26) if
one of the parameters a1 or a2 is set to the degenerate value b. Let us take a2 = b + and
p
consider the first term in (8) with Ca1 ,a2 given by (14)

Qb 1b2 (pa1 a2 )/b
p
b
Ca1 ,a2 =
2
(0) (2a ) (2a ) (2Q 2p)

NS
NS
1 NS
2 NS
.
NS (Q + p a1 a2 )NS (a2 + p a1 )NS (a1 + p a2 )NS (a1 + a2 + p Q)
(B.1)
At a2 b this expression vanishes due to the zero of the multiplier NS (2a2 ). This means that
the integral term in (8) disappears and only the discrete terms contribute. The latter are due to
the singularities of the integral, which come from the pole structure of the integrand. Expression
(B.1) has poles in p at (the four lines of singularities here correspond respectively to the four
multipliers in the denominator of (B.1))
a1 + a2 Q mb1 nb
and a1 + a2 + mb1 + nb,
a1 a2 mb1 nb
and Q + a1 a2 + mb1 + nb,
a2 a1 mb1 nb
and Q + a2 a1 + mb1 + nb,
Q a1 a2 mb1 nb
and 2Q a1 a2 + mb1 + nb,
(B.2)
where (m, n)any pair of non-negative integers of the same parity. At a2 = b + the poles
at p = a1 b + and p = a1 + b + of the first multiplier in the denominator of (B.1) come
across the poles at p = a1 b and p = a1 + b of the second multiplier producing two
singular terms. The same singularity appears from the two reflection symmetric pinches at
p = Q a1 + b and p = Q a1 b. Due to the symmetry properties of the integrand and the
reflection relation (20) the reflected terms give the same contributions and do not need separate
consideration.
First, let us pick up the pole at p = a1 b
res
p=a1 b
Ca1 ,a2 =
(0) (2a ) (2b + 2) (2Q 2a + 2b)

NS
NS
1 NS
NS
1
(2b) (2a ) (2a 2b Q) = 1.
NS (Q 2)NS
NS
1 NS
1
Similar calculation for the residue at p = a1 + b results in

Qb 1b2 2 NS (2a1 )NS (2a1 + 2b Q)
p
res Ca1 ,a2 =
b
p=a1 +b
2
NS (2a1 + 2b)NS (2a1 Q)
2

2
4
() b (a1 b 1/2 b2 /2)
Qb
,
=
2
(1/2 + b2 /2 + a1 b)
where we systematically use the shift relations (16).
(B.3)
(B.4)
230
Second, let us work out the contribution of the second term with
pa ,a
C
1 2

Qb 1b2 (pa1 a2 )/b
=
b
2
(0) (2a ) (2a ) (2Q 2p)

2iNS
NS
1 NS
2 NS
.
R (Q + p a1 a2 )R (a2 + p a1 )R (a1 + p a2 )R (a1 + a2 + p Q)
(B.5)
The pole structure is given by the same formula (B.2) where now (m, n) is a pair of non-negative
integers of opposite parity. At a2 = b + we have to pick up a singular term at p = a1
p
res C
a1 ,b+

(0) (2a ) (2b + 2) (2Q 2p)
2iNS
Qb
NS
1 NS
NS
=
2
R (Q + b 2)R (b)R (2a1 + b)R (2a1 b Q)
2i
.
=
2
(b ) (ba1 ) (1 + b2 ba1 )
p=a1
(B.6)
Residues (B.3), (B.4) and (B.6) can be compared with the special stricture constants, derived in
Section 2 in terms of the screening integrals.
Appendix C. DotsenkoFateev type equation
Substitution
g = x a1 b (1 x)a2 b F
(C.1)
renders Eq. (35) to the form

x 2 (1 x)2 F x(1 x) K1 x K2 (1 x) F

+ L1 x 2 + L2 (1 x)2 L3 x(1 x) F + M1 x M2 (1 x) F = 0,
(C.2)
where
K1 = 2g 3B 3C,
K2 = 2g 3A 3C,
L1 = (B + C)(2B + 2C + 2g + 1),
L2 = (A + C)(2A + 2C + 2g + 1),
L3 = 4AB + 4(2A + 2B + 2C + 1)C + 4(A + B + 3C)g + 4g 2 + 2g,

M1 = 2C(A + B + C + g + 1)(2B + 2C + 2g + 1),
M2 = 2C(A + B + C + g + 1)(2A + 2C + 2g + 1),
while the parameters A, B, C and g are related to a1 , a2 , a3 and b as in Eq. (41).
Consider the two-fold contour integrals

B
C

I (x) =
dt1 dt2 |t1 t2 |A (1 t1 )(1 t2 ) (x t1 )(x t2 ) |t1 t2 |2g
C C
(C.3)
(C.4)
at = and
1
I (x) =
2

B
C

dt1 dt2 |t1 t2 |A (1 t1 )(1 t2 ) (x t1 )(x t2 ) |t1 t2 |2g ,
231
(C.5)
C Ca
where the contours C , = 1, 2, 3, 4, are numbered as follows

C1 = (, 0],
C2 = [0, x],
C3 = [x, 1],
C4 = [1, ).
(C.6)
It is verified directly7 that all these integrals are solutions to Eq. (C.2).
Of all the nine integrals only three are independent. As a base one can choose the set with a
diagonal monodromy around the point x = 0
I1 (x) = I44 (x) I1(0) (1 + ),
(0)
I2 (x) = I24 (x) x 1+A+C I2 (1 + ),

(0)
I3 (x) = I22 (x) x 2+2A+2C+2g I3 (1 + ),
(C.7)
where stands for a regular series in x. The constants

I1(0)
(2g) (1 + B) (1 + B + g) (1 2g A B C) (1 g A B C)
,
(g) (g A C) (A C)
(1 + A) (1 + B) (1 + C) (1 2g A B C)
(0)
I2 =
,
(2 + A + C) (2g A C)
(2g) (1 + A) (1 + A + g) (1 + C) (1 + C + g)
(0)
I3 =
(C.8)
(g) (2 + A + C + g) (2 + A + C + 2g)
=
are calculated using the Selberg integral [25]

1
n!
1
n
n1

(g + kg) ( + kg) ( + kg)
1
dti ti (1 ti )1
|ti tj |2g =
.
(g) ( + + (n 1 + k)g)
i>j
0 i=1
(C.9)
k=0
Another base
(0)
J1 (x) = I11 (x) J1 (1 + ),

(0)
J2 (x) = I13 (x) (1 x)1+B+C J2 (1 + ),

(0)
J3 (x) = I33 (x) (1 x)2+2B+2C+2g J3 (1 + ),
(C.10)
where the dots now replace a regular series in 1 x, enjoys diagonal monodromy around x = 1.
Apparently

I =
(C.11)
M J ,
7 It is implied that the parameters are chosen in a way to ensure convergence of all these integrals. Otherwise, standard
regularization is in order.
232
where the fusion matrix M is evaluated by manipulating the contours of integration. It has
the following entries [21]
sin A sin (g + A)
,
sin (C + B) sin (g + C + B)
sin A sin C
M12 =
,
sin (C + B) sin (2g + C + B)
sin C sin (g + C)
,
M13 =
sin (g + C + B) sin (2g + C + B)
2 cos g sin (g + A) sin (g + A + B + C)
,
M21 =
sin C sin (g + A + B + C)
sin (g + B) sin A
,
M22 =
sin (C + B) sin (g + C + B) sin (g + C + B) sin (2g + C + B)
2 cos g sin (g + C) sin (g + B)
M23 =
,
sin (g + C + B) sin (2g + C + B)
sin (g + A + B + C) sin (2g + A + B + C)
,
M31 =
sin B sin (2g + A + B + C)
,
M32 =
sin (C + B) sin (2g + C + B)
sin B sin (g + B)
.
M33 =
(C.12)
sin (g + C + B) sin (2g + C + B)
M11 =
Next, it is verified [21] that the combination

+ X2 I2 (x)I2 (x)
+ X3 I3 (x)I3 (x)
X1 I1 (x)I1 (x)
(C.13)
is a single-valued function of (x, x)

if
sin A sin C sin (A + C) sin (A + g) sin (C + g)
X3
=
,
X1
sin B sin (B + g) sin (A + B + C + g) sin (A + C + 2g) sin (A + B + C + 2g)
sin (A + C + g) sin A sin C

X2
=
.
X1 2 cos g sin (B + g) sin (A + B + C + g) sin (A + C + 2g)
(C.14)
References
[1] A. Polyakov, Quantum geometry of fermionic strings, Phys. Lett. B 103 (1981) 211213.
[2] A. Zamolodchikov, R. Pogossian, Operator algebra in two-dimensional superconformal field theory, Sov. J. Nucl.
Phys. 47 (1988) 929936.
[3] G. Mussardo, G. Sotkov, M. Stanishkov, Fine structure of the supersymmetric operator product expansion algebras,
Nucl. Phys. B 305 (1988) 69108.
[4] J. Distler, Z. Hloused, H. Kawai, Super-Liouville theory as a two-dimensional superconformal supergravity theory,
Int. J. Mod. Phys. A 5 (1990) 391.
[5] L. Alvarez-Gaum, P. Zaugg, Structure constants in the N = 1 superoperator algebra, Ann. Phys. 215 (1992) 171
230, hep-th/9109050.
[6] R. Poghossian, Structure constants in the N = 1 super-Liouville field theory, Nucl. Phys. B 496 (1997) 451.
[7] A. Neveu, J. Schwarz, Factorizable dual model of pions, Nucl. Phys. B 31 (1971) 86112.
[8] P. Ramond, Dual theory of free fermions, Phys. Rev. D 3 (1971) 24152418.
[9] T. Fukuda, K. Hosomichi, Super-Liouville theory with boundary, Nucl. Phys. B 635 (2002) 215254, hepth/0202032.
233
[10] A. Zamolodchikov, Al. Zamolodchikov, Structure constants and conformal bootstrap in Liouville field theory, Nucl.
Phys. B 477 (1996) 577605, hep-th/9506136.
[11] A. Belavin, A. Polyakov, A. Zamolodchikov, Nucl. Phys. B 241 (1984) 333380.
[12] R. Rashkov, M. Stanishkov, Three point correlation functions in N = 1 super-Liouville theory, Phys. Lett. B 380
(1996) 49.
[13] H. Dorn, H.-J. Otto, On correlation functions for noncritical strings with c 1, d 1, Phys. Lett. B 291 (1992)
3943;
H. Dorn, H.-J. Otto, Two three point functions in Liouville theory, Nucl. Phys. B 429 (1994) 375.
[14] V. Belavin, N = 1 SUSY conformal block recursive relations, SISSA 72/2006/EP hep-th/0611295.
[15] L. Hadasz, Z. Jaskolski, P. Suchanek, Recursion representation of the NeveuSchwarz superconformal block, hepth/0611266.
[16] Al. Zamolodchikov, Conformal symmetry in two dimensions: An explicit recurrence formula for the conformal
partial wave amplitude, Commun. Math. Phys. 96 (1984) 419422.
[17] Al. Zamolodchikov, Conformal symmetry in two-dimensional space: Recursion representation of conformal block,
Theor. Math. Phys. 73 (1987) 1088.
[18] B. Ponsot, J. Teschner, Liouville bootstrap via harmonic analysis on a noncompact quantum group, hep-th/9911110.
[19] V. Fateev, A. Zamolodchikov, Al. Zamolodchikov, Boundary Liouville field theory. 1. Boundary state and boundary
two point function, hep-th/0001012.
[20] V. Dotsenko, V. Fateev, Conformal algebra and multipoint correlation functions in two-dimensional statistical models, Nucl. Phys. B 240 (1984) 312.
[21] V. Dotsenko, V. Fateev, Four point correlation functions and the operator algebra in the two-dimensional conformal
invariant theories with the central charge c < 1, Nucl. Phys. B 251 (1985) 691.
[22] J. Teschner, On the Liouville three point function, Phys. Lett. B 363 (1995) 6570.
[23] V. Kac, Infinite-Dimensional Lie Algebras, Progress in Mathematics, vol. 44, Birkhuser, Boston, 1984.
[24] A. Belavin, Al. Zamolodchikov, Higher equations of motion in N = 1 SUSY Liouville field theory, JETP Lett. 84
(2006) 418424, hep-th/0610316.
[25] A. Selberg, Remarks on a multiple integral, Norsk Mat. Tidsskr. 26 (1944) 7178 (in Norwegian).
FaddeevVolkov solution of the YangBaxter equation

and discrete conformal symmetry
Vladimir V. Bazhanov , Vladimir V. Mangazeev, Sergey M. Sergeev
Department of Theoretical Physics, Research School of Physical Sciences and Engineering,
Australian National University, Canberra, ACT 0200, Australia
Received 30 March 2007; accepted 7 May 2007
Abstract
The FaddeevVolkov solution of the star-triangle relation is connected with the modular double of the
quantum group Uq (sl2 ). It defines an Ising-type lattice model with positive Boltzmann weights where the
spin variables take continuous values on the real line. The free energy of the model is exactly calculated
in the thermodynamic limit. The model describes quantum fluctuations of circle patterns and the associated discrete conformal transformations connected with the Thurstons discrete analogue of the Riemann
mappings theorem. In particular, in the quasi-classical limit the model precisely describe the geometry of
integrable circle patterns with prescribed intersection angles.
2007 Published by Elsevier B.V.
1. Introduction
The FaddeevVolkov solution [13] of the YangBaxter equation [4] possesses many remarkable properties. From algebraic point of view it is distinguished by its modular duality. The
solution is related with non-compact representations of the modular double [5] of the quantum
2
2
group Uq (sl2 ) Uq (sl2 ), where q = eib and q = ei/b . It was studied in connection with the
lattice Liouville and sinh-Gorgon models [69]. Note, that the parameter b above is related to the
Liouville central charge cL = 1 + 6 (b + b1 )2 .
E-mail addresses: vladimir.bazhanov@anu.edu.au (V.V. Bazhanov), vladimir.mangazeev@anu.edu.au

(V.V. Mangazeev), sergey.sergeev@anu.edu.au (S.M. Sergeev).
0550-3213/$ see front matter 2007 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.05.013
V.V. Bazhanov et al. / Nuclear Physics B 784 [FS] (2007) 234258
235
It is well known that every solution of the YangBaxter equation defines an integrable model
of statistical mechanics on a two-dimensional lattice. The FaddeevVolkov model, defined by
their solution, has strictly positive Boltzmann weights. It is an Ising-type model with continuous
spin variables i R, taking arbitrary real values. Its partition function is given by the integral

Z=
eE[ ]
(1)
di ,
where the energy E[] is a convex and bounded from below function of the spin variables =
{1 , 2 , . . .}. In this paper we study this model and reveal its remarkable connections to discrete
geometry, namely, to circle patterns and discrete conformal transformations (for an introduction
see a popular review [10] and a monograph [11]). Recently these topics attracted much attention
due to Thurstons circle packing approach to the discrete Riemann mapping theorem [1215] and
the discrete uniformization theorem [1619].
In this approach circle packings and, more generally, circle patterns (see Section 5) serve as
discrete analogues of conformal transformations, where the circle radii can be thought as the local
dilatation factors. They satisfy certain integrable non-linear difference equations which play the
role of discrete CauchyRiemann conditions. These equations are called the cross-ratio equations [2022]. In the simplest case they reduce to the Hirota form of the discrete sine-Gordon
equation [23,24]. (Curiously enough, the continuous Riemann mapping problem for domains
with a piece-wise smooth boundary is also connected to the Hirota equations [25], but of the
different type [26].) An appropriate variational approach for the circle patterns was developed
by Bobenko and Springborn [27]. Their action A[] is a positive definite function of the logarithmic radii of the circles = {1 , 2 , . . .}, i = log ri . We show that the leading quasi-classical
asymptotics of the energy functional in (1)
E[ ]
1
A[2b ],
2b2
b 0,
cL +,
(2)
precisely coincide with this action provided logarithmic radii of the circles are identified with
the (rescaled) spin variables in (1), namely i = 2bi . Thus at cL + stationary points of
the integral (1) correspond to circle patterns. At finite values of cL the model describes quantum
fluctuations of these patterns and the associated discrete conformal transformation. Given that
the circle radii define local dilatation factors the FaddeevVolkov model describes a quantum
discrete dilaton.
The continuous quantum field theory with the conformal symmetry [28] has remarkable applications in physics and mathematics [29]. It would be interesting to understand which aspects
of the continuous conformal field theory can be transferred to the discrete case.
The organization of the paper is as follows. In Section 2 we present definitions and discuss
properties of the FaddeevVolkov model. In Section 3 we consider Z-invariant lattices and rhombic tilings which are closely related to the combinatorics of the circle patterns. The quasi-classical
limit is considered in Section 4. An introduction to circle patterns is given in Section 5. We review the derivation of the cross-ratio equations and show that they are exactly coincide with
the quasi-classical equation of motion in the FaddeevVolkov model. Some open problem are
discussed in the conclusion. In Appendix A we discuss some implications of the star-triangle
equation and Z-invariance to the circle patterns and volumes of polyhedra in the Lobachevskii
3-space. Appendix B contains useful information on the special functions used in this paper.
236
Fig. 1. The square lattice (solid lines) and its medial lattice (dashed lines).
Fig. 2. Typical horizontal and vertical edges and the corresponding Boltzmann weights.
2. The FaddeevVolkov model

Consider an Ising-type model on the square lattice shown in Fig. 1. Each site i of the lattice
is assigned with a spin variable i R, taking continuous real values. Two spins a and b interact
only if they are adjacent (connected with an edge of the lattice). Typical horizontal and vertical
edges are shown in Fig. 2. The corresponding Boltzmann weights are parametrized through the
so-called rapidity variables p and q associated with the oriented dashed lines in Fig. 1. In our
case the weights depends only on the spin and rapidity differences a b and p q, where a and
pq (a b)
b are the spins at the ends of the edge. We will denote them as Wpq (a b) and W
for the horizontal and vertical edges, respectively.
The partition function is defined as

pq (k l )
Z=
(3)
Wpq (i j )
di ,
W
(i,j )
(k,l)
237
where the first product is over all horizontal edges (i, j ), the second is over all vertical edges
(k, l). The integral is taken over all the interior spins; the boundary spins are kept fixed.
Explicit expressions for the Boltzmann weights contain two special functions. The first one

e2izw
1
dw
def
,
(z) = exp
(4)
4
sinh(wb) sinh(w/b) w
R+i0
is the Barnes double-sine function [30]. It was introduced to this field in [3] as the non-compact
version of the quantum dilogarithm [31]. The second function has no special name,

e2izw
1
dw
def
(z) = exp
(5)
,
8
sinh(wb) sinh(wb1 ) cosh(w(b + b1 )) w
R+i0
but previously appeared [32] in the context of the quantum sinh-Gordon model. The properties
of these functions are briefly summarized in Appendix B.
(s) are defined as
The Boltzmann weights W (s) and W
1 2s (s + i/)
(s) def
(6)
e
= W (s),
,
W
F
(s i/)
where and s stand for the rapidity and spin differences, respectively, and the parameter reads

= b + b1 /2.
(7)
def
W (s) =
The normalization factor F has the form

def
F = ei
2 2 /+i(182 )/24
(2i/).
(8)
The positivity requirement for the Boltzmann weights (point (b) below) singles out two different
regimes

(i) b > 0, or (ii) |b| = 1, Im b2 > 0.
(9)
The following consideration apply to either of these cases. Note that due to the symmetry b
b1 , for the regime (i) it is enough to consider the range 0 < b 1. The weights (6) possess the
following key properties:
(s) are even functions of s. The definition (6) implies
(a) Symmetry: W (s) and W
(s) = W (s).
W
(10)
(b) Positivity: when the parameter b belongs to either of the regimes (9) the functions W (s) and
(s) are real and positive for 0 < < and real s. They have no zeros and poles on the
W
real axis of s.
(c) Asymptotics:

s ,
W (s) F1 e2|s| 1 + O e2|s| ,
(11)
where = min(Re b, Re b1 ).
(d) Star-triangle relation:

qr (a )Wpr (c )W
pq ( b)
d W
R
pr (a b)Wqr (c b).

= Wpq (c a)W
(12)
238
(e) Inversion relations:

W (a b)W (a b) = 1,

it+ (a c)W
it+ (c b) = (a b),
dc W
lim
0+
(13)
where t is real.
(f) Self-duality:
(s) =
W

dx e2 ixs W (x).
(14)
(g) Initial conditions:

W (s) = 1 + O()
(s) = (s) + O()

and W
as 0,
(15)
where (s) is the -function.

The normalization factor F is the minimal solution of the inversion and crossing symmetry
relations [3335],

F
2i
2i2 ()/+i(1+42 )/12
=e
i ,
F F = 1,
(16)
F
in the sense that it does not have poles and zeroes in the strip < Re < ; it is real and
positive for real < < , and has a simple pole at = and a simple zero at = . Note
also that F0 = 1. Eq. (14) is a consequence of (B.8) and the second relation in (16).
Let N be the total number of edges on the lattice. In the thermodynamic limit, when N is
large, the partition function grows as
log Z = fbulk N fboundary N + O(1), N ,

(17)
where the first and second terms describe the bulk and boundary contributions to the free energy.
The inversion relation arguments of [3335] imply that with our normalization of Boltzmann
weights (6) the bulk free energy for the system (3) vanishes exactly
fbulk = lim
1
log Z = 0.
N
(18)
Thus, log Z can only grow linearly in the number of boundary edges, which is O( N ). Here we
assume that the boundary spins are kept finite in the limit N , otherwise (18) will not hold.
The weights (6) attain their maxima at s = 0. In the quasi-classical domain 0 < b 1 the
constant W 2 (0) slowly varies between the values

G
W 2 (0)b=0 = e = 1.3385 . . . ,
(19)
W 2 (0)b=1 = 2 = 1.4142 . . . ,
where G = 0.915965 . . . is the Catalans constant. When the parameter b on the unit circle this
constant monotonically increases as arg(b) varies from 0 to /2 and diverges when b i,
W 2 (0) (2)1/2
( 14 )
( 34 )

+ O 1/2 ,
b i.
(20)
Other properties of the FaddeevVolkov model are considered in a separate publication [36].
239
Fig. 3. A set of directed lines going from the bottom to the top of the figure.
3. Z-invariant lattices and rhombic tilings

The above considerations were explicitly carried out for the homogeneous square lattice
(shown in Fig. 1). However, having in mind connections with the discrete conformal transformations considered below, it is useful to reformulate the results for more general Z-invariant
lattices [37]. A brief review of this important notion with an account of some subsequent developments is given below.
Following [38] consider a finite set of L directed lines forming a graph L of the type shown
in Fig. 3. The lines (in this case six) head generally from the bottom of the graph to the top,
intersecting one another on the way. They can go locally downwards, but there can be no closed
directed paths in L. This means that one can always distort L, without changing its topology, so
that the lines always head upwards. Now shade alternative faces of L as in Fig. 4. Form another
graph G by placing a site in each unshaded face, with edges connecting sites belonging to faces
that touch at a corner. These sites and edges of G are represented in Fig. 4 by open circles and
solid lines, respectively. For each intersection of lines in L, there is an edge of G passing through
it, and conversely. The graph L is the medial graph of G. Introduce additional notation. Let
F (G), E(G) and V (G) denote the sets of faces, edges and sites (vertices) of G and Vint (G) the set
of interior sites of G. The latter correspond to interior faces of L (with a closed boundary).
Now we define a statistical mechanical spin model on G. With each line of L associate its
own rapidity variable p . At each site i of G place a spin i R. The edges of G are either of
the first type in Fig. 2, or the second, depending on the arrangement the directed rapidity lines
with respect to the edge. For each edge introduce a rapidity difference variable e defined as

p q,
for a first type edge,
e =
(21)
p + q, for a second type edge,
where p and q are the associated rapidities, arranged as in Fig. 2. Each edge is assigned with the
Boltzmann weight factor We (a b), where a, b are the spins at the ends of the edge.
The partition is defined as an integral over all configurations of the interior spins with the
weight equal to the product of the edge weights over all edges of G,

Z=
(22)
W(i,j ) (i j )
di .
(i,j )E(G)
iVint (G)
240
Fig. 4. The graph G formed by shading alternate faced of Fig. 3 and putting sites on the unshaded faces.
Fig. 5. A pictorial representation of the first inversion relation in (13).
As before, the exterior spins are kept fixed. Taking into account the symmetry (10) one can easily
see that the expression (3) is just a particular case of (22) for the homogeneous square lattice of
Fig. 1.
The partition function (22) possesses remarkable invariance properties [3739]. It remains
unchanged by continuously deforming the lines of L (with their boundary positions kept fixed)
as long as the graph L remains directed. In particular, no closed directed path are allowed to
appear.1 It is easy to see that all such transformations reduce to a combination of the moves
shown in Figs. 57. The special role of the star-triangle (12) and inversion (13) relations is that
they guarantee the invariance of the partition function under these moves. Given that the graphs
L and G can undergo rather drastic changes, the above Z-invariance statement is rather nontrivial.
The partition function (22) depends on the exterior spins and the rapidity variables p1 , p2 , . . . ,
pL . Of course, it also depends on the graph L, but only on a relative ordering (permutation) of
1 Actually, these restrictions can be removed if one properly defines reflected rapidities for downward going lines
(see Section 3 of [38]), but we will not elaborate this point here.
241
Fig. 6. A pictorial representation of the second inversion relation in (13).
Fig. 7. A pictorial representation of the star-triangle relation (12).
the rapidity lines at the boundaries and not on their arrangement inside the graph. Naturally, this
graph can be identified with an element of the braid group. Then the partition function Z can be
regarded as the corresponding representation matrix, acting on the continuous boundary spins.
There are several celebrated appearances of the braid group in the theory of integrable systems,
particularly, in connection with the Jones polynomials [39] and the TsuchiyaKanie monodromy
representations in conformal field theory [40]. In Section 5 we will show that the rapidity graph L
also describes combinatorial properties of integrable circle patterns.
The integral (22) is well defined in the physical regime, when all the rapidity differences (i,j )
lie in the interval 0 < (i,j ) < . It rapidly converges due to the fast exponential decay (11) of
the edge weights. For other values of (i,j ) the partition function (22) is defined with an analytic
continuation from the physical regime. In what follows we will restrict ourselves to the physical
regime only (in this case the graph L cannot contain more than one intersection for the same pair
of lines).
Let the graph
G has a very large number of edges N . We assume that the number of its exterior
sites scales as N and that the boundary spins are kept finite. Following [37] one can show that
the leading asymptotics of the partition function (22) at large N has the form
log Z =
(23)
fedge ((ij ) ) + O( N ),
(ij )E(G)
242
where the function fedge ( ) does not depend on specific details of the lattice. It will be the
same as for the regular square lattice. This result holds for any Z-invariant system with positive
Boltzmann weights for a large graph L in a general position. Then, it follows from (18), that
in the FaddeevVolkov model with our normalization of the weights (6) the edge free energy
vanishes identically,
FV
fedge
( ) 0.
(24)
Consider some additional combinatorial and geometric structures associated with the graph L.
First, if the unshaded faces in the above definition of G are replaced by the shaded ones, one obtains another graph G , which is dual to G. Each site of G corresponds to a face of G and vice
versa. Obviously, both graphs G and G have the same medial graph L. Assign the difference variables e to the edges of G by the same rule (21). Note that there is one-to-one correspondence
between the edges of G and G. Moreover, if e E(G) is of the first type then the corresponding
edge e E(G ) is of the second (and vice versa). In other words for the corresponding edges
e + e = . Let star(i) denote the set edges meeting at the site i. One can show that for any
interior site of G
(25)
(ij ) = 2, i Vint (G).
(ij )star(i)
The similar sum rule holds for the dual graph G ,

(kl) = 2, k Vint G .
(26)
(kl)star(k)
There is a dual statistical mechanics model with spins i attached to the sites i V (G ). Its
partition function is defined as

W(i,j
di ,
Z =
(27)
i
j
)
(i,j )E(G )
iVint (G )
where the spins at the exterior sites of G are fixed. Evidently, it possesses exactly the same invariance properties under the deformations of L, as the partition function (22). Using the standard
arguments [41] based on the duality transformation (14) one can relate (27) with (22).
Consider yet another graph L , dual to L. The set of sites of L consists of those of G and G .
They are shown in Fig. 8 by white and black dots, respectively. The edges of L always connect
one white and one black sites. The faces of L correspond to the vertices of L. Since the latter
are of the degree four, all faces of L are quadrilateral. The edges of G and G are diagonals of
these quadrilateral (see Fig. 9). Remarkably, the graph L admits a rhombic embedding into the
plane. In other words this graph can be drawn so that all its edges are line segments of the same
length and, hence, all its faces are rhombi, as shown in Fig. 8. The corresponding theorem has
been recently proven in [42]. It states that such embedding exists if and only if (a) no two lines of
L cross more than once2 and (b) no line of L crosses itself or periodic. Note, that in the physical
regime these conditions are obviously satisfied.
Assume the edges of the quadrilaterals to be of the unit length and consider them as vectors
in the complex plane. To precisely specify a rhombic embedding one needs to provide angles
2 The lines of L are called train tracks (!) in [42].
243
Fig. 8. A rhombic embedding of the graph L .
Fig. 9. Two types of rhombi.
between these vectors. A rapidity line always crosses opposite (equal) edges of a rhombus. Therefore, all edges crossed by same rapidity line p are equal to each other. They are determined by
one vector p , |p | = 1. Thus, if the original rapidity graph L has L lines, there will be only L
different edge vectors. Choose them as pk = eipk , where p1 , p2 , . . . , pL are the corresponding
rapidity variables. Each face of L is crossed by exactly two rapidity lines pk and p . To this face
associate a rhombus with the edges pk and p , as shown in Fig. 9. Its diagonals are edges of G
and G . The rhombus angles are precisely the difference variables e and e assigned to these
edges (this is true for both types of rhombi shown in Fig. 9). Glue two rhombi together if they are
adjacent. In the physical regime all their angles will be in the range from 0 to . So the rhombi
will have a positive area and, thus, will not overlap. The sum rules (25) and (26) guarantee that
the resulting surface is flat with no cusps at the sites of L .
4. Quasi-classical limit
Consider the limit which corresponds to the quasi-classical limit of the model. The
parameter becomes large when b 0 or b . For definiteness assume b 0, then
244
(2b)1 . The weight function W (s) acquires a very narrow bell-shaped form and rapidly decays
outside a small interval |s| < b/ . The integral (22) can be then calculated with a saddle point
method. The asymptotic expansion of the Boltzmann weight (6) reads

2
1
, b 0,
A(
|)
+
B(
|)
+
O
b
= exp
W
(28)
2b
2b2
where
A( |) = A(| ) =
i
0

1 + e +i
log
d.
e + ei
(29)
Note that A(|) 0. It is a smooth even function of with the minimum value A(|0) = 0
reached at = 0. It is simply related to the Euler dilogarithm function

A(|) = i Li2 e
i Li2 e
+i
x
,
Li2 (x) =
log(1 x)
dx.
x
(30)
The constant term in (28) reads

1

A( |) +
B( |) =
2
2
z dz
.
sin z
(31)
Estimating the integral (22) for small b one gets

log Z =
where
A[] =

1
A (cl) + B (cl) + O b2 ,
2
2b
A((ij ) |i j )
(32)
(33)
(ij )E(G)
and

2
A[]
1

B[] = log det
+
2
i j
B((ij ) |i j ).
(34)
(ij )E(G)
Here we use rescaled spin variables = {1 , 2 , . . .}, given by i = 2bi . The symbol (cl)
denotes the stationary point of the action (33), defined by the classical equations of motion

A[]
= 0, i = Vint (G).
(35)
i = (cl)
Explicitly, they read
ej + ei +i(ij )
= 1,
ei + ej +i(ij )
(ij )star(i)
i = Vint (G),
(36)
where the product is taken over the star of edges connected to the site i. These are the so-called
cross ratio equations [22]. They arise in connection with the circle patterns [22], which will be
considered in the next section.
245
Fig. 10. A four-edge star on the square lattice. The -variables are placed in the sites of the dual lattice, located in the
shaded areas.
Fig. 11. Faces of the graph G associated with Eqs. (38) and (39).
The cross ratio equations (36) are closely related to the Hirota difference equation [23]. As
an illustration consider the case regular square lattice G shown in Fig. 1. A typical star of four
edges is shown in Fig. 10. The corresponding equation in (36) takes the form
W
N
S

e + e+i
e + e+i
e E + e+i
e + e+i
(37)
= 1,
e + eW +i
e + eN +i
e + eE +i
e + eS +i
where = p q and = . Now place a purely imaginary variable i , 0 Im i < 2 ,
on every site i of the dual lattice. These sites are shown by black dots located in shaded areas
in Figs. 10 and 11. Connect these new variables to the existing variables i by the following
relations. Let N and S be the variables located above and below of a horizontal edge, as in
Fig. 11. Then, we require that
eN S =
eW + eE +i
.
eE + eW +i
(38)
246
Similarly for a vertical edge,
eN + eS +i
eW + eE +i
N S
(39)
e
=
.
eS + eN +i
eE + eW +i
The consistency of these definitions across the lattice is provided by (37). Note that the second
form of (39) is identical to (38) upon interchanging all - and -variables. Therefore it is natural
to associate this universal equation with every face of the graph L (there are two types of faces,
as shown in Fig. 9).
This the famous Hirota equation [23], which was the starting point of various considerations
in [24]. A special feature of the present case is that the -variables are real R, while the
-variables are purely imaginary, |ei | = 1.
eE W =
4.1. Star-triangle relation

Before concluding this section let us discuss some consequences of the Z-invariance. Obviously, this property applies to every term of the quasi-classical expansion (32). As an example,
consider the star-triangle relation (12). Introduce new variables
1 = q + r,
2 = p + q,
3 = p r
(40)
such that
1 + 2 + 3 = 2.
(41)
Substituting (28) into (12) and using the symmetry (10) one obtains

2
d0
1
1
A [] + B [] = exp
A [] + B [] + O b
exp
2b
2b2
2b2
(42)
where
A [] = A(1 |0 1 ) + A(2 |0 2 ) + A(3 |0 3 ),
(43)
A [] = A( 1 |2 3 ) + A( 2 |3 1 ) + A( 3 |1 2 ).
(44)
Expressions for B and B are defined in a similar way, with the function A( |) replaced by
B( |), given by (31). The star-triangle relation (42) implies two non-trivial identities valid for
arbitrary values of 1 , 2 , 3 . The first one is
(cl)

A 0 , 1 , 2 , 3 = A [1 , 2 , 3 ],
(45)
(cl)
where 0
is the stationary point of the integral in (42)

r2 r3 sin 1 + r1 r3 sin 2 + r1 r2 sin 3
(cl)
0 = log
,
r1 sin 1 + r2 sin 2 + r3 sin 3
ri = ei .
(46)
With an account of (30) it is a twelve-term dilogarithm identity, which can be easily verified by
using the definition of the dilogarithm. A geometric meaning of this identity in terms of volumes
of certain polyhedra in the Lobachevskii 3-space is explained in Appendix A. The second identity

2 A []
1
= B [] = (cl) B []
log
(47)
0
0
2
02 0 =0(cl)
247
is an algebraic identity for r1 , r2 , r3 , which follows from elementary (but lengthy) calculations
with the use of the explicit expression (46) and the relation

0
sin
= log tan .
sin
2
(48)
Note that exactly the same quasi-classical star-triangle relation (42) arises in a related, but
different, context of the chiral Potts model [43].
5. Circle patterns and discrete conformal transformations
The FaddeevVolkov model on a Z-invariant lattice has a remarkable connection with the
circle patterns related to the discrete Riemann mapping theorem [1215]. The classical Riemann
mapping theorem states that for any simply connected open domain in the complex plane (which
is not the whole plane) there exists a biholomorphic mapping from that domain to the open unit
disk. The fact that the mapping is biholomorphic implies that it is a conformal transformation.
Likewise, the discrete Riemann mapping theorem involves discrete analogs of conformal transformations. Below we give a short introduction into this subject.3
In the continuous case the conformal transformations possess two key properties: (i) they preserve the angles and (ii) uniformly rescale all infinitesimal lengths in a vicinity of every point,
where the scale depends on the point. Now one needs to reformulate these properties for a discrete case, when the continuous plane is replaced, for example, by a polygonal lattice. For the
first property one could, naively, suggest that the angles between the edges meeting at the same
site remain unchanged. We will call this the rule (i). The second property is more complicated,
since there is no a lattice analog of the infinitesimal length. Nevertheless, as a first attempt it is
reasonable to try the simplest possibility and suggest that the lengths of the edges attached to the
same site are scaled equally (though the scale can vary from site to site). We will call this the
rule (ii).
Now consider transformations of the ordinary square lattice, implementing both rules at every
site. The rule (i) requires that the 90 angles between the four edges at any site remain unchanged.
The rule (ii) requires that these edges also remain equal in length for every site (though, in principle, the lengths could change from site to site). A very little inspection shows these requirements
only allow a uniform dilatation of the whole lattice (of course, the same would be true for any
polygonal lattice, not just the square one). Obviously, the rules are too restrictive and do not
allow any interesting transformations of the lattice; they need to be relaxed.
A fruitful idea is to work with a bipartite lattice and apply only one rule on each sublattice.
Consider, for instance, the square lattice shown in Fig. 12. Its sites are divided into two sublattices
shown with white and black dots. Denote them as G and G respectively (to unload the
picture the edges of G, connecting the white sites, are not shown). There are circles centered
at every white site and passing through four neighboring black sites. Thus every face of black
lattice, G , is inscribed in a circle. Two circles intersect at two distinct points (not just touching)
only if the corresponding faces are adjacent. In the case of Fig. 12 they always intersect at the
right angle, which is the angle between the edges normal to the circles.
Let us now distort our lattice implementing the rule (i) for all black and rule (ii) for all white
sites. A particular example of such transformation is shown in Fig. 13. As seen from the picture,
3 We are indebted to A. Bobenko and B. Springborn for many illuminating explanations, some of which are used below.
248
Fig. 12. The regular square lattice and an associated circle pattern.
Fig. 13. A circle pattern with the combinatorics of the square lattice.
249
Fig. 14. An isoradial circle pattern obtained from the rhombic tiling of Fig. 8. The symbol denotes the cone angle at
an exterior site.
the circle centered at white sites still pass through all neighboring black sites (rule (ii)), they still
intersect at the 90 angles (rule (i)), but the radii of these circles have changed! It is important
to keep in mind that, by construction, this transformation preserves the combinatorics of the
original lattice. In particular, there is a one-to-one correspondence between the circles on the two
pictures, which preserves their adjacency and intersection angles.
Before formulating equations determining the radii of the circles let us make a few remarks.
Denote by z = {z1 , z2 , . . . , zN } positions of the circle centers in the complex plane on the original
square lattice, where N is the total number of circles. In the case of Fig. 12 this number is 49.
} be the
The circles there cover a square-like region, bounded by a dashed line. Let {z1 , z2 , . . . , zN
centers of the corresponding circles in Fig. 13. The transformed circle pattern covers a disk-like
region, bounded by a large circle, also shown by the dashed line.4 Now define a function f such
that zk = f (zk ), for all zk z. This is a discrete approximation to the continuous Christophel
Schwartz map. When N tends to infinity, f (z) uniformly converges to the latter with an accuracy
O(N 1 ) [14,15,44].
The patterns considered above enjoy the combinatorics the square lattice. All circles there
intersect at the 90 angles. In a general case, one can consider circle patterns with the combinatorics of an arbitrary planar graph G, where the edges are assigned with arbitrary intersection
angles {e }, e E(G), 0 < e < . The centers of the circles correspond to the sites of G (white
sites), such that every face of the dual graph G is inscribed in a circle. Two circles connected
by an edge e E(G) intersect at the angle e assigned to this edge (Fig. 16). Their intersection
4 The pattern of Fig. 13 is uniquely fixed (up to a uniform rotation) by the four-fold axial symmetry and a requirement
that all exterior circles there touch one additional circle.
250
Fig. 15. An integrable circle pattern with combinatorics of the graph G from Fig. 4. The sum of the kite angles at the
center of the circle flower is equal to 2 .
Fig. 16. Two intersecting circles with the centers at the sites j and k. The angle j k is assigned to the edge connecting
these sites.
points are sites of G (black sites). Examples of two different circle patterns associated with same
graph G as in Fig. 4 and having the same intersection angles are shown in Figs. 15 and 14. Thus,
to define the intersection properties of a circle pattern (its combinatorics and intersection angles)
one needs to specify a set {G, {e }}.
The intersection properties are called integrable [45] if and only if the corresponding circle
pattern admits an isoradial realization (where all its circles are the same). Here we will only consider this case. The integrable circle patterns were extensively studied in connection with various
approximation problems for the Riemann mappings which are typically based on a regular lattice
combinatorics [15]. They also connected with integrable discrete non-linear equations [45].
251
The space of all integrable sets {G, {e }} was completely described in [45]. Essentially the
same arguments were already presented in Section 3 throughout the discussion of the Z-invariant
systems. Remind that our starting point there was a rapidity graph L (Fig. 3) with a set of rapidity
variables, one per each line of L. Using these data we constructed a rhombic tiling of the dual
graph L (Fig. 8). From this tiling one can immediately construct an isoradial circle pattern
by drawing circles of the same radius centered at the white and passing though the neighboring
black sites (Fig. 14). Conversely, any isoradial circle pattern defines a certain rhombic tiling from
which one can easily reconstruct [42] the corresponding rapidity graph5 together with all rapidity
values (up to an insignificant overall shift). Thus the intersection properties of any integrable
circle pattern can be completely described in terms of some rapidity graph, arising in the Zinvariant models (and vice versa).
Let us now derive the equations determining the radii of the circles. We assume the same
notations as in Section 3. We consider the circle pattern with the combinatorics of G and identify
the rapidity difference variables (ij ) , (ij ) E(G) with the circle intersection angles. Remind
that these variables obey the sum rules (25) and (26). Every pair of intersecting circles with the
centers i and j , i, j G, and the radii ri and rj is associated with a kite-shaped quadrilateral
shown in Fig. 16. The angles j k and j k bisected by the edge (j k) are given by (in the notations
defined on the figure)
j k =
rj + rk eij k
1
,
log
i
rj + rk eij k
(49)
and kj is obtained by permuting rj and rk . Consider a circle flower consisting of one central
circle and a number adjacent circles (petals), like the one shaded in Fig. 15. Obviously the kite
angles at the center of the flower add up to 2
(ij ) = 2, i Vint (G).

(50)
(ij )star(i)
Similarly, a sum of the kite angles at any interior black site (these are angles as in Fig. 16)
should be equal to 2 as well, but this condition is automatically satisfied in virtue of (26).
Introduce logarithmic radii i = log ri of the circles. Substituting (49) into (50) and taking into
account the sum rule (25) one precisely obtains Eq. (36) which arose in the quasi-classical limit
to the FaddeevVolkov model.
The quasi-classical action A[] in (33) corresponds to the fixed boundary conditions, where
the radii of all boundary circles are fixed. They can be chosen arbitrarily. Bobenko and Springborn [27] considered free boundary conditions. Their action
i i + const,
ABS [] = A[]
(51)
iVext (G)
involve cone angles i at the centers of exterior circles (see Fig. 14). Their values are arbitrary
modulo a single constraints
i = 2Nint + 2
e ,
(52)
iVext (G)
eE(G)
where Nint is the number of interior sites of G. In the variation principle for the action (51) one
considers the angles i as fixed and the external radii as independent variables.
5 Note, that in general this graph can have a multiply connected boundary.
252
6. Conclusion
In this paper we displayed some remarkable connections of the theory of integrable quantum
systems to discrete geometry. We have shown that the FaddeevVolkov model provides a quantization of the circle patterns and discrete conformal transformations associated with the discrete
Riemann mapping theorem.
The partition function of the model in the thermodynamic limit is calculated here via the inversion relation method. It would be interesting to verify this result by other means (particularly,
by numerical calculation) since the inversion relation method is based on unproved analyticity
assumptions of the minimal solution (8). At the moment not much else is known about the
FaddeevVolkov model. In particular, the problem of the diagonalization of the transfer matrix
is not yet solved, though some important advances in the related problems in the continuum
sinh-Gordon were made in [46,47].
Finally mention the intriguing relation of the FaddeevVolkov model with the hyperbolic
geometry discussed in Appendix A. The solution (6) is constructed from the non-compact quantum dilogarithm (4). Note, that there exist another solution of YangBaxter equation connected to
this dilogarithm [48]. The latter is related to the link invariants [49] which are also connected with
hyperbolic geometry, particularly, to the volumes of hyperbolic 3-manifolds [50,51]. It would be
interesting to understand these connections further.
Acknowledgements
The authors would like to thank L.D. Faddeev, A.Yu. Volkov, M.T. Batchelor, M. Bortz and
X.-W. Guan, R.M. Kashaev, E.K. Sklyanin and A.B. Zamolodchikov for interesting discussions.
Special thanks to A. Bobenko and B. Springborn for introducing us to a beautiful world of circle
patterns.
Appendix A. Star-triangle relation and hyperbolic geometry
An excellent introduction into the volume calculation in hyperbolic geometry can be found
in [52,53]. As shown in [54,55] the action (33) has a geometric interpretation as the volume
of a certain polyhedron in the Lobachevskii 3-space. Consider the Poincar half-space model
{x, y, z R | z > 0} of the hyperbolic 3-space with the metric ds 2 = (dx 2 + dy 2 + dz2 )/z2 . In
this model hyperbolic planes are represented by hemispheres and half-planes which intersect the
infinite boundary orthogonally in circles and straight lines. Take a circle pattern (for instance,
the one shown in Fig. 14) and imagine it as lying in the infinite boundary (the xy-plane). If
we erect hemispheres over the circles and orthogonal half-planes over the prolonged boundary
edges (these edges are shown by heavy solid lines in Fig. 14), we obtain a set of hyperbolic
planes which bound a polyhedron P . Most of its vertices are at infinity; these include the circle
intersection points (in the xy-plane) and one extra point at the infinite boundarythe intersection
point of the planes raised from the boundary edges.
Let us now calculate the volume V (P ) of the polyhedron P . Obviously, this polyhedron can
be glued from similar shaped hyperbolic hexahedrons, shown in Fig. 17, which are associated
with every pair of intersecting circles. Every such hexahedron can be split into two tetrahedra
with three vertices at infinity. In what follows we will refer to Fig. 16, which corresponds to the
projection of the hexahedron to the ideal, but simplify the notations there taking j = 1, k = 2 and
denoting j k , kj , j k , jk by 1 , 2 , , , respectively. Note that 1 + 2 = . From (49) one
253
Fig. 17. A hexahedron in the hyperbolic 3-space bounded by four infinite vertical planes and two hemispheres.
has
ei1 =
e1 + e2 +i
,
e1 + e2 i
(A.1)
where i = log ri . Introduce the Milnors Lobachevskii function

x
L(x) =
log |2 sin | d.
(A.2)
The volume of the hexahedron in Fig. 17 is given by [54]6

1
2
Vhexahedron (1 , 2 ) = L
+L
2
2

1
1
= 2L
+ A( |1 2 ) + (1 2 )(1 2 ),
2
2
4
(A.3)
where the function A(|) is defined in (29). Then, the quasi-classical action A[ (cl) ] in (32)
(remind that it corresponds to the fixed (Dirichlet) boundary conditions) is given by

(2ij ij ),
A (cl) = 2V (P ) 2V (P0 )
(A.4)
iVext (G) (ij )star(i)
where V (P ) is the volume of the hyperbolic polyhedron P whose projection to the ideal is a
circle pattern with a given set of intersection angles ij and boundary radii ri = ei . Further,
V (P0 ) denotes the value of this volume for the isoradial case (when all circles have the same
6 Note, that our variables , , correspond to 2 , 2 , in [54].
1 2
1
2
254
Fig. 18. Circle patterns corresponding the star (left) and triangular sides of the YangBaxter equation (42).
radius)
V (P0 ) =
L
(ij )E(G)

ij
.
2
(A.5)
Thus, up to boundary terms the action (A.4) coincides with an excess volume of P with respect
to P0 . The result (23) and (24) implies that if the polyhedron
P is glued from a very large
number of hexahedrons, N , this excess volume grows only as N . This is not very surprising:
an arbitrary integrable circle pattern is significantly different from its isoradial counterpart only
at the boundary (indeed, compare Fig. 14 and Fig. 15).
The star-triangle equation has the following interpretation. The volume of the polyhedron
erected over circle pattern corresponding to the star side of (42), shown in Fig. 18, is equal to

(cl)

1
1
2
3
V = 2L
+ 2L
+ 2L
+ A 0 , 1 , 2 , 3 + boundary term,
2
2
2
2
(A.6)
(cl)
where the action A and 0 are defined in (43) and (46). Similarly, the volume of the polyhedron corresponding to the triangle side of the star-triangle relation in Fig. 18 is given by

1
2
3
1
V = 2L
+ 2L
+ 2L
+ A [1 , 2 , 3 ]
2
2
2
2
+ boundary term
(A.7)
where A is defined in (44). Clearly, the difference between these two volumes is the volume of
an ideal tetrahedron with its vertices located at the circle intersection points in Fig. 18. Indeed,
taking into account (45) and the fact that the boundary terms in (A.6) and (A.7) coincide, one
obtains
Vtetrahedron = V V

1
2
3
1
2
= 2L
+ 2L
+L
2L
2L
2
2
2
2
2

3
L
.
2
(A.8)
Using the identity

2L
2L
= L( ),
2
2
255
(A.9)
one obtains
Vtetrahedron = L(1 ) + L(2 ) + L(3 ),
(A.10)
which is precisely the Milnors formula [52] for the volume of the ideal tetrahedron. We would
like to stress that whereas the volumes V and V depend on the radii of the circles, their
difference does not. It depends only on the angles 1 , 2 and 3 . In (45) this difference is absorbed
into the normalization of the edge weights. Correspondingly, the RHS of (A.8) contains six terms,
one for each edge of the tetrahedron. Clearly, the action (A.4) is invariant under any star-triangle
transformation since volumes of ideal tetrahedrons will cancel out from the difference V (P )
V (P0 ).
Appendix B. Properties of the functions (z) and (z)
Below we use the following notations

1
2
q = eib ,
= b + b1 ,
2

def
1 qkx .
(x, q) =
q = e
ib2

i(b b1 )
q = i exp
,
2(b + b1 )
(B.1)
(B.2)
k=0
The function (z). This function is defined by the integral (4). It has the following properties.
(a) Simple poles and zeros:

poles of (z) = i + imb + inb1 , m, n Z0 ,

zeros of (z) = i + imb + inb1 , m, n Z0 .
(B.3)
(b) Functional relations:

(z)(z) = eiz
2 i(122 )/6

(z ib1 /2)
2zb1
=
1
+
e
.
(z + ib1 /2)
(B.4)
(c) Asymptotics:
(z) 1,
Re(z) ;
(z) eiz
2 i(122 )/6
Re(z) +,
(B.5)
where Im(z) is kept finite.

(d) Product representation:
(z) =
(qe2zb ; q 2 )
1
(qe
2zb ; q 2 )
Im b2 > 0.
(B.6)
(e) Pentagon relation: The function (z) satisfies the following operator identity [3]
(P)(X) = (X)(P + X)(P),
[P, X] =
1
,
2 i
(B.7)
256
where [ , ] denotes the commutator. It can be re-written in the matrix form [56,57]

(x + u) 2 iwx
(u v i)(w + i)
2
dx = ei(1+4 )/122 iw(v+i)
e
,
(x + v)
(u v + w i)
(B.8)
where the wedge of poles of (x + u) must lie in the upper half-plane, the wedge of zeros
of (x + v) must lie in the down half-plane, and the integrand must decay when x .
The function (z). This function is defined by the integral (5). It has the following properties.
(i) Simple poles and zeros:

poles of (z) = 2i + imb + inb1 , m, n Z0 , m + n |m n| = 0 mod 4 ,

zeros of (z) = 2i + imb + inb1 , m, n Z0 , m + n |m n| = 0 mod 4 .
(B.9)
(ii) Functional relations:
(z)(z) = eiz
2 /2i(182 )/12
(z + i)(z i) = (z).
(B.10)
(iii) Asymptotics:
(z) 1,
z ;
(z) eiz
2 /2i(182 )/12
z +,
(B.11)
where Im(z) is kept finite.

(iv) Product representation:
(z) =
(q 2 e2zb ; q 4 ) (qez/(2) ; q 2 )
,
1
(q 2 e2zb ; q 4 ) (qez/(2) ; q 2 )
Im b2 > 0.
(B.12)
References
[1] A.Y. Volkov, Quantum Volterra model, Phys. Lett. A 167 (1992) 345355.
[2] L. Faddeev, A.Y. Volkov, Abelian current algebra and the Virasoro algebra on the lattice, Phys. Lett. B 315 (1993)
311318.
[3] L. Faddeev, Currentlike variables in massive and massless integrable models, in: Quantum Groups and Their Applications in Physics, Varenna, 1994, in: Proceedings of the International School of Physics Enrico Fermi, vol. 127,
IOS, Amsterdam, 1996, pp. 117135.
[4] M. Jimbo (Ed.), YangBaxter Equation in Integrable Systems, Advanced Series in Mathematical Physics, vol. 10,
World Scientific, Teaneck, NJ, 1989.
[5] L. Faddeev, Modular double of a quantum group, in: Confrence Mosh Flato 1999, vol. I (Dijon), in: Mathematical
Physics Studies, vol. 21, Dordrecht, Kluwer, 2000, pp. 149156.
[6] L.D. Faddeev, R.M. Kashaev, A.Y. Volkov, Strongly coupled quantum discrete Liouville theory. I. Algebraic approach and duality, Commun. Math. Phys. 219 (2001) 199219.
[7] S. Kharchev, D. Lebedev, M. Semenov-Tian-Shansky, Unitary representations of Uq (sl(2, R)), the modular double,
and the multiparticle q-deformed Toda chains, Commun. Math. Phys. 225 (2002) 573609.
[8] A.Y. Volkov, Noncommutative hypergeometry, Commun. Math. Phys. 258 (2005) 257273.
[9] A.G. Bytsko, J. Teschner, Quantization of models with non-compact quantum group symmetry: Modular XXZ
magnet and lattice sinh-Gordon model, J. Phys. A 39 (2006) 1292712981.
[10] K. Stephenson, Circle packing: A mathematical tale, Not. Am. Math. Soc. 50 (2003) 13761388.
[11] K. Stephenson, Introduction to Circle Packing. The Theory of Discrete Analytic Functions, Cambridge Univ. Press,
Cambridge, 2005.
257
[12] W.P. Thurston, Invited talk at the international symposium on the occasion of the proof of the Bieberbach conjecture,
Purdue University, 1985.
[13] B. Rodin, D. Sullivan, The convergence of circle packings to Riemann mapping, J. Diff. Geom. 26 (1987) 349360.
[14] Z.-X. He, O. Schramm, The C convergence of hexagonal disc packings to Riemann map, Acta Math. 180 (1998)
219245.
[15] O. Schramm, Circle patterns with the combinatorics of the square grid, Duke Math. J. 86 (1997) 347389.
[16] W.P. Thurston, Three-Dimensional Geometry and Topology, vol. 1, Princeton Univ. Press, 1997.
[17] A. Marden, B. Rodin, On Thurstons formulation and proof of Andreevs theorem, Lect. Notes Math. 1435 (1990)
103115.
[18] A.F. Beardon, K. Stephenson, The uniformization theorem for circle packings, Indiana Univ. Math. J. 39 (1990)
13831425.
[19] Z.-X. He, Rigidity of infinite disk patterns, Ann. Math. 149 (1999) 133.
[20] A. Bobenko, U. Pinkall, Discrete isothermic surfaces, J. Reine Angew. Math. 475 (1996) 187208.
[21] F. Nijhoff, H. Capel, The discrete Kortewegde Vries equation, Acta Appl. Math. 39 (1995) 133158, KdV 95,
Amsterdam, 1995.
[22] A.I. Bobenko, Y.B. Suris, Integrable systems on quad-graphs, Int. Math. Res. Not. (2002) 573611.
[23] R. Hirota, Nonlinear partial difference equations. III. Discrete sine-Gordon equation, J. Phys. Soc. Jpn. 43 (1977)
20792086.
[24] L. Faddeev, A.Y. Volkov, Hirota equation as an example of an integrable symplectic map, Lett. Math. Phys. 32
(1994) 125135.
[25] P.B. Wiegmann, A. Zabrodin, Conformal maps and integrable hierarchies, Commun. Math. Phys. 213 (2000) 523
538.
[26] R. Hirota, Nonlinear partial difference equations. II. Discrete-time Toda equation, J. Phys. Soc. Jpn. 43 (1977)
20742078.
[27] A.I. Bobenko, B.A. Springborn, Variational principles for circle patterns and Koebes theorem, Trans. Amer. Math.
Soc. 365 (2004) 659689.
[28] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Infinite conformal symmetry in two-dimensional quantum
field theory, Nucl. Phys. B 241 (1984) 333380.
[29] C. Itzykson, H. Saleur, J.-B. Zuber (Eds.), Conformal Invariance and Applications to Statistical Mechanics, World
Scientific, Teaneck, NJ, 1988.
[30] E.W. Barnes, Theory of the double gamma function, Philos. Trans. Roy. Soc. A 196 (1901) 265388.
[31] L.D. Faddeev, R.M. Kashaev, Quantum dilogarithm, Mod. Phys. Lett. A 9 (1994) 427434.
[32] S. Lukyanov, A. Zamolodchikov, Exact expectation values of local fields in the quantum sine-Gordon model, Nucl.
Phys. B 493 (1997) 571587.
[33] Y.G. Stroganov, A new calculation method for partition functions in some lattice models, Phys. Lett. A 74 (1979)
116118.
[34] A.B. Zamolodchikov, Z4 -symmetric factorized S-matrix in two spacetime dimensions, Commun. Math. Phys. 69
(1979) 165178.
[35] R.J. Baxter, The inversion relation method for some two-dimensional exactly solved models in lattice statistics, J.
Stat. Phys. 28 (1982) 141.
[36] V.V. Bazhanov, V.V. Mangazeev, S.M. Sergeev, Exact solution of the FaddeevVolkov model, arXiv:0706.3077.
[37] R.J. Baxter, Solvable eight-vertex model on an arbitrary planar lattice, Philos. Trans. R. Soc. London, Ser. A 289
(1978) 315346.
[38] R.J. Baxter, Functional relations for the order parameters of the chiral Potts model, J. Stat. Phys. 91 (1998) 499524.
[39] V. Jones, On knot invariants related to some statistical mechanical models, Pacific J. Math. 137 (1989) 311388.
[40] A. Tsuchiya, Y. Kanie, Vertex operators in the conformal field theory on P1 and monodromy representations of the
braid group, Lett. Math. Phys. 13 (1987) 303312.
[41] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, London, 1982.
[42] R. Kenyon, J.-M. Schlenker, Rhombic embeddings of planar quad-graphs, Trans. Amer. Math. Soc. 357 (2005)
34433458 (electronic).
[43] V.V. Bazhanov, On star-triangle relation in the chiral Potts model (2007), in preparation.
[44] D. Matthes, Convergence in discrete Cauchy problems and applications to circle patterns, Conf. Geom. Dynam. 9
(2005) 123.
[45] A.I. Bobenko, C. Mercat, Y.B. Suris, Linear and nonlinear theories of discrete analytic functions. Integrable structure
and isomonodromic Greens function, J. Reine Angew. Math. 583 (2005) 117161.
[46] S. Lukyanov, Finite temperature expectation values of local fields in the sinh-Gordon model, Nucl. Phys. B 612
(2001) 391412.
258
[47] A. Zamolodchikov, On the thermodynamic Bethe ansatz equation in the sinh-Gordon model, J. Phys. A 39 (2006)
1286312887.
[48] R.M. Kashaev, On the spectrum of Dehn twists in quantum Teichmueller theory, math.QA/0008148.
[49] K. Hikami, Generalized volume conjecture and the A-polynomialsthe NeumannZagier potential function as a
classical limit of quantum invariant, math.QA/0604094.
[50] R.M. Kashaev, The hyperbolic volume of knots from the quantum dilogarithm, Lett. Math. Phys. 39 (1997) 269.
[51] H. Murakami, J. Murakami, The colored Jones polynomials and the simplicial volume of a knot, Acta Math. 186
(2001) 85104.
[52] J. Milnor, Hyperbolic geometry: The first 150 years, Bull. Amer. Math. Soc. (N.S.) 6 (1982) 924.
[53] .B. Vinberg, Volumes of non-Euclidean polyhedra, Usp. Mat. Nauk 48 (1993) 1746.
[54] B.A. Springborn, Variational principles for circle patterns, math.GT/0312363.
[55] B.A. Springborn, A variational principle for weighted Delaunay triangulations and hyperideal polyhedra,
math.GT/0603097.
[56] B. Ponsot, J. Teschner, ClebschGordan and RacahWigner coefficients for a continuous series of representations
of Uq (sl(2, R)), Commun. Math. Phys. 224 (2001) 613655.
[57] R.M. Kashaev, The non-compact quantum dilogarithm and the Baxter equations, J. Stat. Phys. 102 (2001) 923936.

Nucl - Phys.B v.784 PDF

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Nucl - Phys.B v.784 PDF

Încărcat de

Drepturi de autor:

Formate disponibile

Nuclear Physics B 784 (2007) 121

Chiral symmetry breaking from intersecting D-branes

E-mail address: harvey@theory.uchicago.edu (J.A. Harvey).

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

q+1 (L) = q+1 L3q ,

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

The quartic interaction due to single gluon exchange is proportional to

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

is proportional to the (q + 1)-dimensional massless propagator over a distance L in the directions

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

There is also a RR flux through the (8 q)-sphere (2.7).

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

Using (2.10) and (2.7) in (2.9) leads to the action

where we have defined

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

The relation between the two scales (2.17) can be rewritten as

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

with the dynamically generated mass mf given by

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

5. Lower-dimensional color branes

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

E. Antonyan et al. / Nuclear Physics B 784 (2007) 121

Nuclear Physics B 784 (2007) 2235

Linear relations and their breakdown in high energy

Received 21 May 2007; accepted 1 June 2007

E-mail addresses: jcclee@cc.nctu.edu.tw (J.-C. Lee), yyang@phys.cts.nthu.edu.tw (Y. Yang).

J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235

J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235

2. String compactified on torus

The left and right momenta are defined to be

J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235

where p |p| and q |q| and

With this setup, the center of mass energy E is

and T-duality symmetry implies conservation of winding number

One can easily calculate the following kinematic relations

sL,R (k1L,R + k2L,R )2 ,

tL,R (k2L,R + k3L,R ) ,

uL,R (k1L,R + k3L,R )2 ,

2.2. Four-tachyon scatterings with NR = NL = 0

J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235

where we have used  = 2 for closed string propagators

(1 12 12 )(1 + n + 12 )(1 + m + 12 )

(2 + k1R k2R + k2R k3R )

3. High energy behaviors

J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235

vertex of the amplitude is

J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235

J.-C. Lee, Y. Yang / Nuclear Physics B 784 (2007) 2235

d 2 z1 d 2 z2 d 2 z3 d 2 z4 eik1L X (z1 )eik1R X (z1 )

X T L L i 2 X P L eik2L X (z2 ) X T R R i 2 X P R eik2R X (z2 )

eik3L X (z3 )eik3R X (z3 )eik4L X (z4 )eik4R X (z4 )

After the standard SL(2, C) gauge fixing, we get

By using Eqs. (3.5) to (3.10), the amplitude can be written as

where p |p| and q |q| and

where we have used = 2 for closed string propagators

(1 12 12 )(1 + n + 12 )(1 + m + 12 )

(2 + k1R k2R + k2R k3R )

(2 + k1R k2R + k2R k3R NR )

E 12(NL +NR ) sin

( t2R 1)( u2R 1) ( t2L 1)( u2L 1)

sin(sR /2)( t2R 1)( u2R 1)( t2L 1)( u2L 1)

r |r = (r r ),

The symmetric part

where (x) is the Euler gamma-function, we have

Here (x) = (x)/ (x) and