Nucl - Phys.B v.660

Nuclear Physics B 660 (2003) 324
www.elsevier.com/locate/npe
Time- and space-dependent backgrounds

from non-supersymmetric strings
E. Dudas a,b , J. Mourad a,c , C. Timirgaziu a
a Laboratoire de Physique Thorique, 1 Universit de Paris-Sud, Bt. 210, F-91405 Orsay cedex, France
b Centre de Physique Thorique, Ecole Polytechnique, F-91128 Palaiseau, France
c Fdration de Recherche APC, Universit de Paris 7, Paris, France
Received 26 September 2002; accepted 18 March 2003
Abstract
We investigate maximally symmetric backgrounds in non-supersymmetric string vacua with Dbranes and O-planes localized in the compact space. We find a class of solutions with a perturbative
string coupling constant in all regions of spacetime. Depending on the particular model, we find either
a time evolution with a big-bang type singularity or a space-dependent background with generically
orbifold singularities. We show that the result can be interpreted as a supersymmetric bulk with some
symmetries broken by the boundaries. We also discuss an interesting connection to Lorentzian and
Euclidian orbifolds.
2003 Published by Elsevier Science B.V.
1. Introduction and summary of results

String theory provides a natural setting to address cosmological issues like the
cosmological constant problem or the fate of the big-bang singularity. Whereas the first
problem still awaits for a qualitatively different perspective, the second one led over the
last ten years to more explicit proposals like the pre-big-bang model [1], the ekpyrotic
scenario [2] or brane-world models [3]. In the second model, objects with negative tensions
were important for its realization. In addition to the positive tension branes, orientifolds
allow also for negative tension objects: the O-planes. This renders them candidates for
rich possibilities of cosmological backgrounds. The goal of this paper is to study these
possibilities.
E-mail address: mourad@lyre.th.u-psud.fr (J. Mourad).
1 Unit Mixte de Recherche du CNRS (UMR 8627).
0550-3213/03/$ see front matter 2003 Published by Elsevier Science B.V.

doi:10.1016/S0550-3213(03)00248-7
E. Dudas et al. / Nuclear Physics B 660 (2003) 324
The string models we consider are vacua with D-branes and orientifold planes, [4,5]
and with broken supersymmetry. Orientifold models with D-branes/O-planes with broken
supersymmetry in various sectors of the theory were proposed in the last few years [69].
The classical background of such vacua has typically lower symmetry and was explicitly
worked out in some particular examples [10,14]. An obvious and important worry about
such constructions is the issue of classical and quantum stability and their fate. The fact
that some of these constructions [8,9] are tachyon-free in all moduli space of the theory is
a promising feature, but gives only a partial answer to the stability question. On the other
hand, it is clear that typically, as soon as supersymmetry is broken, D-branes/O-planes
start to interact, curve the internal or the non-compact space and generically produce a
time-dependent configuration.
The explicit models we consider are nine-dimensional string models. At the effective
field theory level, supersymmetry breaking is described by non-BPS configurations of
D-branes and O-planes, as well as a one-loop bulk cosmological constant.
The maximally symmetric classical background of these models generically depends on
two coordinates, which, according to the details of the models, can be (t, y) or (z, y), where
t is the time, y is the compact coordinate orthogonal to the branes and z is a non-compact
coordinate parallel to the branes. We concentrate on solutions which are perturbative in the
string coupling constant so that the classical solution receives small quantum corrections.
In the time-dependent case, we show by an adequate choice of coordinates that one of the
solutions has a static bulk and hence has as much symmetries as the supersymmetric bulk,
but the boundaries are moving and thus break the invariance under time translation. The
solutions are characterized by a big-bang (or big-crunch) singularity which is due to the
collision of the two boundaries. Interestingly, this solution can be interpreted also as an
orbifold by a boost of a static and supersymmetric background. The boost parameter on
the Lorentzian side is related to the branes and O-planes and the one-loop cosmological
constant. We also show that in models with NSNS tadpoles, the usual NSNS tadpole
condition is replaced by a sum rule, which is basically the boosted version of the static
tadpole condition. In the space-dependent case, the boundaries join at a conical singularity
and also break some of the bulk symmetries. We show that this background can be
considered as an orbifold by a two-dimensional rotation of the supersymmetric one.
One application of our work is to the big-bang type cosmology. In this respect,
our results have similar features to the pre-big-bang [1] and ekpyrotic [2] scenarios.
Another possible application is to the FischlerSusskind mechanism [20]. The relation we
find between the classical solutions of non-supersymmetric orientifolds and Lorentzian
orbifolds suggests a deepest relation at the quantum level. In this respect, a severe
instability of Lorentzian orbifolds was recently pointed in [11] along with some possible
ways out [11,12]. Irrespective of the final fate of Lorentzian orbifolds, we believe that
connections between seemingly unrelated vacua can be useful for a better understanding
of perturbative and non-perturbative aspects of string theory. We hope to come back to this
issue in the future.
The paper is organized as follows. Section 2 describes the various classical and
perturbative solutions for string models with D-branes and O-planes in nine dimensions
and their relation to Lorentzian and Euclidian orbifolds. Section 3 discusses some of
their applications to the big-bang and big-crunch cosmology, whereas Section 4 presents
some explicit string examples. Appendix A contains the technical details involved in the
resolution of the equations of motion and a more complete set of various classical solutions,
including the ones with strong string coupling in some regions of spacetime. Appendix B
gives some details on the explicit string models under consideration.
2. Time- and space-dependent backgrounds of non-supersymmetric strings

We consider a generic form of the effective action of the type II string containing D8branes (and eventually O8-planes in the case of orientifold models). We include also the
bulk one-loop cosmological constant 1 which typically arise in most non-supersymmetric
strings. The resulting effective action in the string frame reads

1
1
1
10
2
2
2
2
F 2 1
S = 2 d x G e
R ()
2
2
2 10! 10

(1)
d 9 x T0 e + q0 A9
d 9 x T1 e + q1 A9 ,
y=0
y=R
where is the dilaton, A9 the RR nine-form coupling to D8-branes and O8-planes and
is the induced metric. For simplicity of the discussion we placed all branes either at the
origin y = 0 or at y = R of a compact coordinate y of radius R. Since supersymmetry is
broken, we do not assume any relation between the RR charges qi and the NS tensions Ti .
We consider in the following string vacua of the type [8,9], where the RR charge
is globally cancelled, as required by consistency arguments, but the NSNS tadpole
conditions are violated and there is an induced one-loop cosmological constant
q0 + q1 = 0
(RR tadpole conditions),
T0 + T1 = 0 (uncancelled NSNS tadpoles),

1 = 0.
(2)
The classical field equations have no solution with SO(9) symmetry, in agreement with
various arguments presented in the literature [10]. We search here for solutions depending
on the compact coordinate y and on another coordinate, which can be the time t or
another space coordinate z. We restrict ourselves in that section to solutions which
are smoothly connected to the supersymmetric ones [19] and have a perturbative string
coupling throughout the spacetime. We leave to Appendix A the explicit derivation of the
complete set of classical solutions.
2.1. Cosmological solutions
The general form of such a solution is of the form

kx 2 2
dx dx + e2B(t,y) dt 2 + dy 2 ,
ds 2 = e2A(t,y) 1 +
4
= (t, y),
F10 = f (t, y) 10 ,
(3)
where 10 is the ten-dimensional volume form. The eight-dimensional metric at fixed y

and t is a maximally symmetric one: k = 0 for a flat eight-dimensional space, k = 1 for a
closed 8-sphere and k = 1 for an open 8-hyperboloid.
The equations of motion of the nine-form can be readily solved and the solution, in the
Einstein frame, reads
f = q0 2 e5/2 (y),
(4)
where (y) is an odd 2R-periodic function and (y) = 1 when y is between 0 and R.
Since most of our results ask for the existence of negative tension objects, we mostly refer
in the following to orientifold models [4] and consequently we choose the ten-form to be
odd under y y. Since the nine-form potential has no physical degrees of freedom, we
replace it in the Lagrangian by its classical expression (4). As a result, we get an effective
one loop cosmological constant given by
e = 1 +
q02 2
.
4
(5)
We define for later convenience 0 = T0 2 and 1 = T1 2 . The Einstein and dilaton

equations for the remaining functions are given by2
5
t t + yy 8At t + 8Ay y 5 2e e2B+5/2 = S,
2
2 y2
7At t + 7Ayy 28A2t + 28A2y Bt t + Byy t +
21ke2B2A
4
4
+ 2 e e2B+5/2 = S,
2 y2
28ke2B2A
8At t 36A2t + 28A2y + 8At Bt + 8Ay By t
4
4
+ 2 e e2B+5/2 = 0,
2 y2
+ 28ke2B2A
8Ayy + 28A2t 36A2y + 8At Bt + 8Ay By t
4
4
2 e e2B+5/2 = S,
t y
= 0,
Aty + At Ay At By Ay Bt +
16
where we have defined

S = e5/4+B 0 (y) 1 (y R) .
(6)
(7)
(8)
(9)
(10)
(11)
All the functions A, B, and are even in y and 2R-periodic. The sources in the
right-hand side of the equations determine the y derivatives of these functions at 0 and R
as

0
1
Ay 0+ , t = eB(0,t )+5(0,t )/4,
Ay R , t = eB(R,t )+5(R,t )/4, (12)
16
16
2 The notations we use are = 2 /t 2 , etc.
tt
and

By 0+ , t = Ay 0+ , t ,

By R , t = Ay R , t ,

y 0+ , t = 20Ay 0+ , t ,

y R , t = 20Ay R , t .
(13)
The boundary conditions (13) imply the relations

(y, t) = 20A(y, t) + (t),
B(y, t) = A(y, t) + b(t),
(14)
where and b are for the time being arbitrary functions of time. The relations (14) have
the virtue of reducing the boundary conditions to (12). Using the relations (14) in Eq. (10),
one gets
24Ab5/4
e
(15)
= 0,
yt
which is readily solved by
e24Ab5/4 = F (y) + G(t).
(16)
The relations (14) and (16) allow to transform the partial differential equations (6)(9) into
ordinary differential equations for b, , F , and G. The boundary conditions (12) translate
into
30

31
,
F R =
.
F 0+ =
(17)
2
2
The equations resulting from the substitution in (6)(9) of the relations (14) and (16)
are discussed in some details in Appendix A. We show there that for k = 0 it is possible to
solve exactly and to find all the solutions of the equations of motion. There are two classes
of solutions. The first one, denoted (a) in Appendix A, is characterized by
b 2 = 2 ,
= 0,
F
F = 9 e ,
2
b + 2 G = 0,
G
(18)
(19)
where is a positive constant. In (17), (18), and (19) the prime (the dot) denote
differentiation with respect to y (t). There is a second class of solutions, called (b) in
Appendix A, which are characterized by G = 0. It is shown in Appendix A that the
solutions in the first class are the only ones with a perturbative value of the string coupling
in the whole spacetime. In addition, they are the only ones which are smoothly connected
to the supersymmetric solutions [19]. Notice the remarkable fact that Eqs. (18) and (19)
are of first order, even though they are not derived from BPS-type conditions.
The boundary conditions are not always compatible with Eq. (19). Evaluating (19) at
the origin and at R we get the two conditions
1
1
,
T12 q12 + 4 2 .
(20)
2
These conditions are necessary but, as we will see, not sufficient to insure the existence of
a solution in this class. Before analyzing in detail the solutions for the different models,
notice that if the effective cosmological constant e is negative Eq. (20) are automatically
T02 q02 + 4
satisfied. This is to be compared to the sum rules [15] where staticity and compactness
impose severe restrictions (or fine tuning) on the tensions.
The SUSY configuration is a special case of the above system: it corresponds to a
vanishing one-loop cosmological constant, 1 = 0, and to the equality of the RR charges
and NS charges q0 = T0 = T1 . This implies that = 0, F is constant and the space is
flat, k = 0.
If neither of the two conditions (20) are satisfied then the only solutions for the
background of the form (3) have singularities in the string coupling and are displayed
for completeness in the appendix. However, a non-singular solution with the same number
of isometries exists, it amounts to interchange the time coordinate with one of the eight
coordinates x, called z later on. If one the conditions (20) is true but not the other, then
one has to look for a solution with lower symmetries. There are three qualitatively distinct
cases to consider, depending on the value of the effective one-loop cosmological constant
(5):
(i) If e > 0, then the solution has the form3

3 e
sh |y| + ,
F (y) =
where the boundary conditions (17) determine the parameters and

ch() =
T0
,
(2 e )
ch(R + ) =
T1
.
(2 e )
(21)
(22)
Notice that (22) can have a solution only if the tensions in y = 0 and y = R have opposite
signs. Therefore we need objects of negative tension in the theory, which in our explicit
string examples later on are orientifold planes.
The final solutions of the classical field equations, in the Einstein frame, read

3 e t
24A
b0 +50 /4
e
=e
e sh |y| + ,
G0 +

3 e t
e sh |y| + ,
e24B = e24t +25b0+50 /4 G0 +

5/6
3 e t
5b0 /60 /24

.
e sh |y| +
e =e
(23)
G0 +
We did carefully keep track in (23) of the integration constants 0 , b0 , and G0 . Notice that
for G0 < 0, there are singularities in the (t, y) plane. We restrict in this section to safer
values G0
0. Notice that for T1 < |T0 |, we get < 0 and therefore there are singularities
if G0 + (3 e /) exp(t) sh() 0, whereas for T1 > |T0 |, is positive and therefore
there are no singularities in the compact space. In the limit of vanishing and , we recover
the supersymmetric solution [19].
3 There is also the solution F F in (21). This is equivalent, however, to a reflection y y + R which
exchanges the two fixed points, accompanied by the replacement R.
Let us consider in more details the resulting spacetime metric and choose for simplicity
b0 = 0, 0 = 0. We find

1/12

3 e t
2
e sh |y| +
ds = G0 +
dx dx + e2t dt 2 + dy 2 .
(24)
Finally, by making the change of variables
1
1
T = et ch(y + ),
X = et sh(y + ),
we get the spacetime metric

ds 2 = [G0 + 3 e X]1/12 dx dx dT 2 + dX2 ,
(25)
(26)
when y > 0. The Z2 identification y y is mapped in terms of the coordinates (X, T ) to

a parity X times a boost K with a parameter 2. This means that the orientifold operation
acts in the (T , X) plane as
= X K2 .
(27)
In addition, the identification of points on the circle y = y + 2R results in (T , X)

coordinates in the orbifold identification

T
ch(2R) sh(2R)
T
,
(28)
X
sh(2R) ch(2R)
X
which is nothing but a two-dimensional boost K2R with a velocity v = th(2R) in the
(T , X) space.
The final result (26) is quite surprising. Indeed, (26) coincides with the spacetime metric
obtained in [19] in the supersymmetric type I string with N D8-branes at the origin
X = 0 of a compact coordinate of radius R and 32 N D8-branes at X = R. The
supersymmetric PolchinskiWitten solution and our non-supersymmetric solution appear
to be two different orbifolds of the same ten-dimensional background. More precisely,
the SUSY solution uses the translation group, while the non-SUSY one uses the twodimensional Lorentz group.
The metric (26) and the identifications (27), (28) allow a simple physical interpretation
of our configuration in the (X, T ) coordinates. Indeed, the fixed points of the two
orientifold operations are
:

K2R :
X = th T ,
X = th(R + )T .
(29)
Consequently, the negative tension O-planes and the branes located at the origin move with
a constant velocity
v0 = th ,
(30)
in the static background (26), whereas the positive tension branes and O-planes at y = R
move at a constant velocity
v1 = th(R + ).
(31)
10
Moreover, the boundary conditions (22) encode the dynamics of the two boundaries in the
condition

2
T0 1 v0 + T1 1 v12 = 0.
(32)
The interpretation of (32) is quite simple. In the supersymmetric case, the branes and
O-planes are at rest and (32) reduces to the standard NSNS tadpole condition T0 + T1 = 0.
In the case with broken supersymmetry, the NSNS tadpoles are boosted according
to the velocity of the branes and O-planes in the background (26). The boost is the
one appropriate for a Lagrangian density, since the dilaton field in (1) couples to the
Lagrangian, instead of the energy.
An interesting particular example of the above results is the one in which all global
tadpoles are cancelled and we have a negative bulk cosmological constant
T0 + T1 = 0,
1 < 0.
In this case it turns out that (22) is satisfied for

R
R
T1
,
ch
=
= ,
2
2
2 e

R
R
v0 = th
,
v1 = th
.
2
2
(33)
(34)
Therefore even in the absence of disk NSNS tadpoles, one-loop cosmological constant is
sufficient to generate a constant velocity dynamics, which can be interpreted in terms of
orientifolds of a Lorentzian orbifold. Notice, however, that in this case the metric and the
dilaton can be singular between the O-planes. From this perspective, models with NSNS
tadpoles seem to be the only ones free of singularities in the compact space.
(ii) If e < 0, then the conditions (20) are verified and the solution for F is given by

3 e
ch |y| + ,
F (y) =
(35)
where and are determined by

T0
T1
,
sh(R + ) =
.
(2 e )
(2 e )
The solution of the classical field equations reads

3 e t
24A
b0 +50 /4
e ch |y| + ,
e
G0
=e

3 e t
e24B = e24t +25b0+50 /4 G0
e ch |y| + ,

5/6
3 e t
5b0 /60 /24

e ch |y| +
e =e
.
G0
sh() =
(36)
(37)
By the same change of variables (25) and by setting for simplicity b0 = 0 = 0, we get for
y > 0 the spacetime metric

ds 2 = [G0 3 e T ]1/12 dx dx dT 2 + dX2 .
(38)
11
As in the case (i), the Z2 identification y y is mapped in terms of the coordinates

(X, T ) to a parity times a boost with a parameter 2, while the identification of points
on the circle y = y + 2R gives for the (T , X) coordinates the identification (28). The
boundary conditions (36) in this case imply the following condition on the spacetime
boundary velocities

T0
T1
(39)
1 v02 +
1 v12 = 0.
v0
v1
Notice that in this case it is possible to obtain solutions to (39) with tensions of the same
sign in the two fixed points, provided that the two velocities (30), (31) have opposite signs.
(iii) If e = 0, then there are time-dependent solutions provided that the tensions at
y = 0 and y = R have opposite signs. The final form of the solution is

1/12

ds 2 = G0 + F0 e(t |y|)
dx dx + e2(t +b0) dt 2 + dy 2 ,

5/6
e = e0 G0 + F0 e(t |y|)
(40)
,
where F0 is a constant and the parameter in (40) is determined by the condition
eR =
T1
.
T0
(41)
In this case, by introducing the coordinates (25) with = 0, we find the spacetime
metric

1/12

dx dx dT 2 + dX2 ,
ds 2 = G0 + F0 X
(42)
where we introduced the light-cone coordinates X = T X. Notice from (41) that by
taking T0 very small and negative T0 0 , we can generate an infinite boost parameter .
2.2. Static solutions
One may wonder whether there are similar solutions which do not depend on time but
on an additional space coordinate z, that is with the ansatz

ds 2 = e2A g dx dx + e2B dz2 + dy 2 ,
(43)
where g is the eight-dimensional flat, dS or AdS metric and with A, B, and functions
of y and z. Here also we have two classes of solutions. The first one is also characterized
by a finite string coupling and is smoothly connected to the supersymmetric solution [19].
In this case, a similar analysis as before shows that the eight-dimensional metric g is
flat and, with the replacement t z Eqs. (16), (17), and (18) hold true. The non-trivial
modification is a crucial sign in Eq. (19) which becomes
F 2 + 2 F 2 = 9 2 e .
(44)
If we allow for a negative effective one loop cosmological constant, we see at once that
(44) cannot be satisfied and the time-dependent solutions are the only ones. Evaluating
12
(44) at the origin and at R we get the two conditions

1
1
(45)
,
T12 q12 + 4 2 .
2
Unlike in (20), these conditions are necessary and sufficient to insure the existence of a
solution. The explicit form of the solution in this case is

3 e
sin |y| + ,
F (y) =
(46)
where the boundary conditions (17) determine the parameters and

T02 q02 + 4
cos() =
T0
,
(2 e )
cos(R + ) =
T1
.
(2 e )
The final solutions of the classical field equations, in the Einstein frame, read

3 e z
24A
b0 +50 /4
e sin |y| + ,
G0 +
=e
e

3 e z
e sin |y| + ,
e24B = e24z+25b0+50 /4 G0 +

5/6
3 e z
5b0 /60 /24

e sin |y| +
G0 +
.
e =e
(47)
(48)
This solution is continuously connected to the supersymmetric solution [19] in the limit of
vanishing and .
The analog of Eq. (32) in this case is
T0
T1
+
= 0.
cos cos(R + )
(49)
Notice that, contrary to the previous time-dependent case (32), Eq. (49) does not give any
restriction on the tensions.
The z coordinate is non-compact and the Planck mass in this background is infinite.
There are singularities for z = (or z = for the negative branch solution z z).
Depending on the sign of G0 and the numerical values of and , this solutions can also
have singularities at a finite distance from the origin in the (z, y) plane.
Interestingly, this solution can also be related to the supersymmetric solution [19].
Indeed, by the change of coordinates
1 z
1
e sin(y + ),
Z = ez cos(y + ),
we get for y > 0 the spacetime metric

ds 2 = [G0 + 3 e Y ]1/12 dx dx + dY 2 + dZ 2 ,
Y=
(50)
(51)
which is the one derived by Polchinski and Witten [19], except that here the Y coordinate
is non-compact. The periodicity y = y + 2R reflects in the new coordinate system (Z, Y )
in the orbifold identification

Z
cos(2R)
Y
sin(2R)
sin(2R)
cos(2R)

Z
,
Y
13
(52)
which is nothing but a two-dimensional rotation R in the (Z, Y ) space, with an angle
= 2R. The orientifold identification y y is mapped in terms of the coordinates
(Z, Y ) to a parity Y times a rotation R2 with an angle 2
= Y R2 .
(53)
The metric (48) and the orbifold and orientifold operations (52), (53) allow a physical
interpretation of our configuration in the (Z, Y ) coordinates. Indeed, the fixed points of the
two orientifold operations are
:

R2R :
Y = tan Z,
Y = tan(R + )Z.
(54)
Consequently, the orientifolds and branes located at the origin have an angle 0 = with
respect to the Z axis, whereas the branes and orientifolds at y = R are at an angle
1 = R + . In the new coordinate system, the interpretation of (49) is quite simple.
It corresponds to the dilaton NSNS tadpole condition obtained from the action (1), when
we correctly take into account in the BornInfeld action the different rotation of the two
sets of D-branes and O-planes. Notice that the orbifold identification (52) imply that the
two-dimensional (Z, Y ) plane has singularities. For discrete values of the rotation angle
= 1/(NR) or M/(NR) with M and N coprime, these singularities are of a conical type
and the resulting model before the orientifold operation (54) is the non-compact C/ZN
orbifold model.
2.3. Freely-acting Lorentzian orbifold models
It was shown in [11] that Lorentzian orbifolds are unstable, since adding sources
changes completely the spacetime which collapses into a large black hole. It was also
pointed out in [11,12] that making the orbifold boost to be freely-acting by combining it
with a shift in an additional coordinate can cure this problem in some cases. We would like
here to point out that such freely-acting boost operations can also be obtained from some
specific non-supersymmetric orientifold models. For definiteness we restrict ourselves
again to the case of the two-dimensional boost. Let us start with type IIB, orbifolded by
(1)F , where (1)F is the spacetime fermion number and a shift acting simultaneously
in two compact coordinates y1 , y2
y1 = y1 + R1 ,
y2 = y2 + R2 .
(55)
This orbifold breaks completely supersymmetry and generates a negative one-loop bulk
cosmological constant 1 , but has no fixed points. We now orientifold by the operation
= y1 , where y1 is a parity in y1 . The fixed points under will generate
O8-planes. By consistency, we must introduce D8-branes in the theory.4 The low-energy
4 Perturbative orientifold models of this type can be explicitly constructed [21].
14
effective action is of the form (1). By the chain of arguments discussed in the previous
paragraphs, the classical background of this model is equivalent to a two-dimensional
orbifold operation, supplemented by a shift
T + X e2R (T + X),
y2 y2 + R2 ,
T X e2R (T X),
(56)
where (T , X) are defined in terms of (t, y1 ) as in (25). In addition, the orientifold operation
acting in the (T , X, y2 ) coordinates includes a boost, = X K2 , analogously to (27).
This type of smoothing of the singularity for the usual Milne space turns out to be
not sufficient to cure the instability problem discussed in [11] and presumably also in the
context discussed here. However, in other examples, like the null-brane orbifold [12], it
does help. It would therefore be interesting to find non-supersymmetric vacua related in
the sense described in the previous section to the null-brane orbifold [12].
A similar construction can be applied to models with a background depending on
two space coordinates (48). This can help in order to give a simple meaning to the case
where the parameter is generic, by combining an irrational angle in the (Z, Y ) plane
with a translation along an additional circle. This actually defines a Melvin-type model
[22]. There seems therefore to be a surprising connection between non-supersymmetric
orientifolds with NSNS tadpoles and Melvin-type string models.
3. Cosmological applications
The spacetime diagram representing the solution (26) is depicted in Fig. 1. The bulk
is the region between the two dashed lines which represents the two moving boundaries.
At T = 0 there is a big-bang type singularity and the two boundaries coincide. Near this
region of spacetime the effective field theory approach we are pursuing is not valid due
to the higher-order corrections. Note, however, that the string coupling is small in this
region so that higher genus corrections are expected to be negligible. With the help of (26),
Fig. 1. The spacetime diagram of the cosmological solution. In the region T < 0 the time reversed evolution is
drawn. The spacetime is contained in the cone defined by the two orientifold planes, denoted by dash-dotted lines.
15
the proper distance separating the two fixed points under the orientifold involution is easily
calculated and is given in the Einstein frame by

8
[G0 + 3 e v1 T ]25/24 [G0 + 3 e v0 T ]25/24 ,
25 e
(57)
where we recall that v0 , v1 are the velocities of the two boundaries (30), (31).
The distance becomes arbitrarily small for small T signalling a breakdown of the
classical effective description. In the lower part of the diagram in Fig. 1 we draw the time
reversed solution which represent two collapsing boundaries. Assuming, as in [1,2] that
string corrections allow a smooth transition between the two branches, one suppresses the
big-bang type singularity and gets an eternal spacetime. This could offer a new perspective
on the proposals of [1,2], since our string model realizes a scenario similar to the one
proposed there and hopefully the string corrections are easier to handle.
Since matter and YangMills interactions are confined to the boundaries, the relevant
metric to consider, from the point of view of an observer on the branes, is the induced
one on the boundaries. Let us for simplicity suppose that = 0. The metric induced at the
origin y = 0 = X is just a flat metric, and the one induced at y = R is given by

1/12

ds 2 = G0 + 3 sh(R) e T
dx dx d T2 + dX2 ,
(58)
where we performed the change of coordinates T = T / ch(R). The geometry of

spacetime experienced by an observer on the boundaries does not necessarily reveal the
spacetime singularities of the full ten-dimensional geometry. This is particularly obvious
for the induced metric on y = 0 which is completely flat and does not see the singularity
at X = T = 0. On the other boundary, where the metric is (58) the observer experiences
an expanding universe with a singularity in the past at T = G0 /(3 sh(R) e ).

However, this singularity is just an illusion since the metric looses its validity before that
= 0, where the true singularity takes place.
time at T
We now turn to examine the backgrounds of different non-supersymmetric models.
4. Explicit string examples

4.1. The SO(N) USp(32 N) model
The model we consider here is an orientifold of the nine-dimensional ScherkSchwarz
deformation of type IIA strings [17]. There are several consistent orientifold operations
already considered in the literature [7]. The example we are studying here is based on the
involution [16] = (1)FL y , where (1)FL is the left world-sheet fermion number
and y is the parity in a compact coordinate y. The virtue of this projection, similar
in spirit with the one used in type O orientifolds [18], is that it eliminates the closed
string tachyon present in the ScherkSchwarz compactification. The fixed points of the
orientifold operation are y = 0, containing an O8+ -plane with (16, 16) units of RR
and NSNS charges and y = R containing an O8 -plane of charges (16, +16).
16
As shown in Appendix B, RR tadpole cancellation in this model asks for a net number of
N = 32 D8-branes placed in fixed positions on the compact coordinate y. A nice feature of
this model, which was one original motivation for studying it, is that if all D8-branes are at
the origin, the open massless spectrum is supersymmetric with gauge group SO(32), while
if all D8-branes are placed in the other fixed point y = R, the massless open spectrum
precisely coincides with the one of the 10d model [8], with a gauge group USp(32).
Moving continuously the branes from one fixed point to the other interpolates therefore
continuously between the supersymmetric SO(32) and the non-supersymmetric5 USp(32)
type I strings. Of course, this statement is strictly speaking incorrect, since the closed (bulk)
sector has softly broken supersymmetry governed by the radius 1/R, but for large enough
radius the main supersymmetry breaking appears due to the simultaneous presence of D8branes and O8 -planes, which breaks supersymmetry at string scale, if they are in top of
each other, or if not at a scale proportional to the distance between them.
One of the puzzles of this model which motivated our investigation is the interaction
pattern between D-branes and O-planes of this model. In flat space, the O8+ -plane at y = 0
and the O8 at y = R repel each other, a rather surprising feature, whereas O-planes and
anti-O-planes of the same type attract each other. Moreover, the D-branes do not interact
with the O8+ -planes by supersymmetry, whereas they are attracted by the O8 -planes.
Consequently, if we start with an initial configuration of D8-branes on top of O8+ -planes,
with massless supersymmetric spectrum and gauge group SO(32), the dynamics of the
system seems to push the D8-branes to move towards y = R and the final state of the
system is the USp(32) non-supersymmetric string!
In analogy with Section 2, we place N D8-branes at y = 0 and the rest 32 N at y =
R. The effective action of the system is (1), with 0 = (16 N) 2 T8 , 1 = (48 N) 2 T8
and q0 = T0 . There
is a bulk one-loop cosmological constant, which is small in the largeradius limit R , such that the effective cosmological constant is positive e > 0 and
we are in the case (i) of Section 2. To start with, we neglect the one-loop cosmological
constant. The equations are readily solved and the solution is (23) with = 0. In this case,
according to (30), the boundary at the origin has zero velocity v0 = 0, whereas (32) fixes
the boost parameter by the equation 0 ch(R) = 1 . This relation shows that solutions
with the symmetry displayed in (3) exist provided that the number of branes at the origin
is less than 16 (0 > 0) and therefore the tension due to the orientifolds and the branes at
the origin must be negative.
We leave for future work a complete analysis of the dynamics of a probe brane in
this background. Here we just note that, since the bulk is supersymmetric, a particular
solution for the position of the test brane is X = const. In this case, if the one-loop bulk
cosmological constant is small, the velocity of the supersymmetric boundary at y = 0 is
much smaller than the positive velocity of the non-supersymmetric boundary at y = R.
Therefore, for this particular solution the test brane will stay closer to the supersymmetric
boundary, despite the naive flat space arguments presented above.
5 In this case, supersymmetry is however non-linearly realized in the open sector [13].
17
4.2. Other models

The solutions we found in Section 2 can be used for other string models with broken
supersymmetry.
An interesting model to analyze is the nine-dimensional model with 16 O8+ -planes at
the origin and 16 anti-orientifold O8+ -planes at the other fixed point y = R. In fact, this
model is a different orientifold [7] of the same ScherkSchwarz deformation of the type II
string. The model do not need branes for consistency, since the RR charges add up to zero.
In addition, the tachyon is not removed by the orientifold projection for any values of the
radius. Let us however add an equal number N of braneanti-brane pairs, the anti-branes
being placed at the origin and the branes at y = R, in order to avoid the occurrence of
open-string tachyons. For this configuration, we obtain
T0 = (N 16)T8,
T1 = (N 16)T8,
q0 = (N + 16)T8 ,
q1 = (N + 16)T8 .
(59)
The bulk one-loop cosmological constant in this model is small in the large radius limit and
therefore the conditions (45) are satisfied. We obtain therefore solutions of the type (48),
with the parameter = 1/R. In the (Z, Y ) coordinate system, the orbifold identification
(52) becomes trivial and therefore we get a regular two-dimensional non-compact plane.
Since the braneanti-brane pairs are not needed by the consistency of the theory, we can
also analyze the particular case N = 0. The classical background can be easily deduced
from Section 2. In fact this case corresponds to 0 = 1 and 0 > 0. If we take into
account the negative one-loop cosmological constant we can get solutions of the form (37),
but only provided that the effective cosmological constant e is negative. This corresponds
however to a radius R of the order the string scale, where our effective theory analysis is
not reliable.
Another model which can be similarly analyzed is the T-dual of the USp(32) nonsupersymmetric string compactified on a circle. The model has 16 O8 -planes at each
fixed point and 32 D8-branes which for simplicity can be distributed among the fixed
points, N at y = 0 and 32 N at y = R. Here, the one loop bulk cosmological constant
is zero. A quick inspection of (19) supplemented with the boundary conditions (17) shows
that there are no solutions for any values of N .
We can also analyze the case of the supersymmetric type I string supplemented by a
certain number N of braneanti-brane pairs. In order to avoid open-string tachyons, we
place all 32 + N D8-branes at the origin and all N D8-anti-branes at y = R. In this case,
we have
T0 = (N + 16)T8,
T1 = (N 16)T8 ,
q0 = (N + 16)T8 ,
q1 = (N + 16)T8,
(60)
and the bulk one-loop cosmological constant is zero. The conditions (20) in this case are
violated, while (45) are satisfied. Therefore, a solution exists depending on two space
coordinates (y, z) given explicitly by (48), with = 0 and determined by sin(R) =
(16 N)/(N + 16).
18
Acknowledgements
We are grateful to C. Charmousis, G. DAppollonio, and G. Pradisi for useful
discussions. E.D. was supported in part by the RTN European Program HPRN-CT-200000148.
Appendix A. Equations of motion and more solutions

In this appendix we look for solutions to the equations of motion (6)(9) supplemented
with the boundary conditions (17). We will be able to find all the solutions with k = 0 and
special solutions with k = 1.
The combination (8) (9) + 2/5(6) using the relations (14) gives
+ 8A + 140ke2b = 0.
(A.1)
From (16) we get A in terms of F and G and their derivatives. The previous equation
becomes

1
G
5
+
(A.2)
+ b + + 140ke2b = 0.
3 G+F
4
In this equation only F depends on y. If we assume that at least one among 0 and 1
= 0. We have to
is not zero than due to (17) F cannot be constant. This implies that G
consider two cases: (a) = 0 and (b) G = 0.

Case (a) Eq. (A.2) with = 0 gives k = 0, so only flat eight-dimensional spaces are
possible in this case. The combination (7) + (9) gives
+ G)(F
4b + (b G
+ G)1 = 0.
(A.3)
+G
= 0, so we have b = , being a constant and
This equation gives b = 0 and b G
+G
= 0.
G
(A.4)
Next, consider Eq. (8). After using (A.4), it gives

+ G)2 F 2 2 F 2 9 2 e + 2F (G
+ G) = 0.
(G
(A.5)
This can hold provided that

+ G = 0,
G
F 2 2 F 2 = 9 2 e .
(A.6)
We have obtained all Eqs. (19) and (18). It is now possible to verify that the other equations
are identically satisfied. This case leads to solutions where the dilaton never diverges
and they are smoothly connected to supersymmetric solutions in the limit of zero boost
(rotation) and was discussed in great detail in the text.
19
= 0, we can assume without loss of generality that G = 0 because an

Case (b) Since G
eventual constant G can be absorbed into F . Eq. (A.2) now becomes

1 5
+
(A.7)
b + + 140ke2b = 0.
3
4
In this case (16) implies

1 5
A=
b+ ,
24
4
A =
F
,
24F
(A.8)
so that A (A ) depends only on t (y). The four Eqs. (6)(9) have thus the form
Fi (y) + Gi (t) = 0,
i = 1, . . . , 4,
(A.9)
for some functions Fi and Gi . This implies that both must be constants, that is
Fi = a i ,
Gi = ai ,
(A.10)
where ai are four constants. The y-dependent part of the equations is quite simple. For
instance, Eq. (8) gives
F 2 9a3 F 2 = 9 2e .
(A.11)
Inserting this equation in the other ones Fi determines the constants a1 , a2 , and a4 in terms
of a3 . The remaining equations read
25a3
,
8
1
a3
49
A + 8A 2 + A b + b + ke2b = ,
8
8
2
a3
2
2b
A + 8A + 7ke = ,
8
A + b)
=
(A + b ) + 8A(
(A.12)
(A.13)
(A.14)
25a3
.
32 7A 2 b 2 + 48A b + 7 25ke2b =
(A.15)
2
Notice that all the equations are not independent since for instance (A.12) 8(A.13) +
7(A.14) is identically zero.
Let us first consider the k = 0 case. If we define h = exp ((b + 5/4)/3), then
Eq. (A.14) reads
h a3 h = 0.
(A.16)
A direct consequence of this equation is

h 2 a3 h2 = E,
(A.17)
where E is a constant. Eq. (A.7) now can be solved as

=
c
,
h
(A.18)
20
where c is a constant. Eqs. (A.12) and (A.13) are identically satisfied and Eq. (A.15) gives
E = c2 /8. If a3 = 2 /9 > 0, then the solution reads

3c
(t t0 )
h(t) = sh
(A.19)
.
3
2 2
The spacetime metric and the string coupling are given by

ds =
2
(t t0 )
sh
3

1/12
3
F (y)

6
+5 2

(t
t
)
)
(t
t
0
0
2
2
th
dt + dy ,
dx dx + sh
3
6
e = e5b0 /60 /24
(t t0 )
sh
3
5/2

(t t0 ) 2
th
F (y)5/6 ,
6
(A.20)
where t0 is an integration constant and the parameters and are determined by (22).
The physical time-region can be taken, for example, between t0 and . The final solution
depends, as in the solutions discussed in the text, on the sign of the effective cosmological
constant. For the three different values of e , the y-dependent function F (y) is

3 e
sh |y| + ,
e > 0, F (y) =

3 e
ch |y| + ,
e < 0, F (y) =
e = 0, F (y) = F0 e|y| ,
(A.21)
where the parameters , are determined by (22) if e > 0, by (36) if e < 0 and by (46)
if e = 0.
These solutions have large string coupling close to the big-bang singularity. Moreover,
they are not smoothly connected in the = = 0 limit to the supersymmetric solution
[19].
In the case a3 = 2 /9 < 0, the solution is

3c
(t t0 )
h(t) = sin
(A.22)
.
3
2 2
The solution exists only when the effective cosmological constant is positive. The final
solution in this case is
5/2
2 2
(t
t
(t
t
)
)
0
0
+
ctg
e = e5b0 /60 /24 cos
3
6
4
5/6

3 e
sin |y| +

1/12
(t t0 ) 3
e
cos
sin |y| +
ds =
3

2/3

+5 2

(t
t
(t
)
)
0
0
ctg

+
dx dx + cos

3
6
4

2
2
,
dt + dy
21
(A.23)
where the parameters and are determined here by (47).

We were able to find particular curved solutions, k = 0 of the system (A.12)(A.15) for
k = 1. The solution has a constant b and
3
e2b = a3 .
(A.24)
7
The solution, which exists only for a3 > 0, is explicitly given by

1/12
x 2 2
2
2
50 /4 5t
2
2
1
dt + dy
ds = e
e F (y)
,
dx dx +
4
21
2 = 9 16a3,
e = e0 /24 F (y)5/6 et /6,
(A.25)
(A.26)
where for the three different values of e , the y-dependent function F (y) is given by
(A.21).
Contrary to the flat solutions discussed in the text, all the backgrounds discussed in the
appendix cannot be transformed by a change of coordinates into static ones.
There are also static solutions, depending on coordinates (z, y), with z being parallel
to the branes. There are also two cases to consider, and the analog of case (a) was already
discussed in the text. In case (b), Eqs. (A.12)(A.15) still hold provided we flip the sign in
the right-hand side of the equations. Here k = 1 corresponds to the dS eight-dimensional
space, whereas k = 1 corresponds to the AdS eight-dimensional space. The solutions
have a form similar to the one appearing in (A.20) and (A.23) with the time-dependence
replaced by a z-dependence such that hyperbolic functions are replaced by trigonometric
functions and vice-versa.
Appendix B. Nine-dimensional non-supersymmetric orientifold
In this appendix we give the technical details for the string model [16] considered in
Section 4. It is simpler to construct its T-dual version. We start from a ScherkSchwarz
deformation of IIB which is obtained by modding IIB on a circle of radius r by (1)F ,
where F is the spacetime fermion number and acts on a circle S 1 as an asymmetric shift
yR yR + r, yL yL r.
The resulting torus partition function is

1
T = |V8 S8 |2 m,n + |V8 + S8 |2 (1)n m,n
2

1
+ |O8 C8 |2 n,m+ 1 + |O8 + C8 |2 (1)n m+ 1 ,n ,
(B.1)
2
2
2
22
where the SO(8) characters are defined as

34 + 44
34 44
24 14
,
V
=
,
S
=
,
8
8
24
24
24
with i the Jacobi functions and the Dedekind function and
( m + nr )2 ( m nr )2
q 4 r q 4 r .
m,n =
O8 =
C8 =
24 + 14
,
24
(B.2)
(B.3)
m,n
The resulting model interpolates between the IIB theory in the limit r 0 and the 0B
theory in the limit r .
After performing a rescaling of the radius r 2r the torus amplitude reads

T = |V8 |2 + |S8 |2 2m,n + |O8 |2 + |C8 |2 2m+1,n

8 )
8 + C8 O
8 )
(V8
S8 + S8 V
1 + (O8 C
1 .
2m,n+ 2
2m+1,n+ 2
Next, we consider the orientifold obtained by gauging the discrete symmetry =

(1)FL , where is the standard world-sheet parity operator and (1)FL is the worldsheet fermion number. The resulting Klein bottle is given by

1
K = (V8 S8 )P2m (O8 C8 )P2m+1 ,
2
m2
with Pm = m q 4R2 . The tachyon is anti-symmetrized in the Klein bottle and removed
from the spectrum. Notice that the limit where the radius goes to infinity reduces to the
closed sector of the OB model.
The tadpoles are obtained from the transverse channel amplitude
5
5
= 2 (V8 S8 )Wn (V8 + S8 )(1)n Wn = 2 (V8 W2n+1 S8 W2n ),
K
(B.4)
2
2
n2 R2
where Wn = n q .
We can see that there is no NSNS tadpole, but there is a RR tadpole. The cancellation
of the tadpoles will require the introduction of D9-branes. In a T-dual spacetime
interpretation, the model contains 16 O8+ - and 16 O8 -planes.
The open sector amplitudes for the D9-branes are
2
(m+1/2)2
N1 + N22 m22
q R + N1 N2
q R2
(V8 S8 ),
A=
2
m
m
m2
N1 N2
N1 + N2 m22
M =
(B.5)
(1)m q R2 V8 +
q R S8 .
2
2
m
m
The RR tadpole cancellation asks for N1 + N2 = 32 and the resulting gauge group
is SO(N1 ) USp(N2 ). In the T-dual version we have N1 D8-branes at the origin and
N2 = 32 N1 branes at the other fixed point. When N2 = 0 the open spectrum is that
of the supersymmetric type I, whereas the N1 = 0 open spectrum is the one of the USp(32)
non-SUSY model.
Notice that in the limit R the RR tadpole cancellation condition is modified and
the resulting gauge group is U (32).
23
References
[1] G. Veneziano, Phys. Lett. B 265 (1991) 287;
M. Gasperini, G. Veneziano, Astropart. Phys. 1 (1993) 317, hep-th/9211021.
[2] J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, Phys. Rev. D 64 (2001) 123522, hep-th/0103239;
J. Khoury, B.A. Ovrut, N. Seiberg, P.J. Steinhardt, N. Turok, Phys. Rev. D 65 (2002) 086007, hepth/0108187;
N. Seiberg, hep-th/0201039;
V. Balasubramanian, S.F. Hassan, E. Keski-Vakkuri, A. Naqvi, hep-th/0202187;
L. Cornalba, M.S. Costa, hep-th/0203031;
N.A. Nekrasov, hep-th/0203112;
L. Cornalba, M.S. Costa, C. Kounnas, Nucl. Phys. B 637 (2002) 378, hep-th/0204261;
A.J. Tolley, N. Turok, hep-th/0204091;
C.P. Burgess, F. Quevedo, S.J. Rey, G. Tasinato, I. Zavala C., hep-th/0207104.
[3] H.A. Chamblin, H.S. Reall, Nucl. Phys. B 562 (1999) 133, hep-th/9903225;
P. Binetruy, C. Deffayet, D. Langlois, Nucl. Phys. B 565 (2000) 269, hep-th/9905012;
A. Kehagias, E. Kiritsis, JHEP 9911 (1999) 022, hep-th/9910174;
P. Brax, A. Falkowski, Z. Lalak, Phys. Lett. B 521 (2001) 105, hep-th/0107257;
For a review and more references, see, e.g., D. Langlois, hep-th/0209261.
[4] A. Sagnotti, in: G. Mack, et al. (Eds.), Cargese 87, Non-Perturbative Quantum Field Theory, Pergamon
Press, Oxford, 1988, p. 521;
M. Bianchi, A. Sagnotti, Phys. Lett. B 247 (1990) 517;
M. Bianchi, A. Sagnotti, Nucl. Phys. B 361 (1991) 519;
G. Pradisi, A. Sagnotti, Phys. Lett. B 216 (1989) 59;
M. Bianchi, G. Pradisi, A. Sagnotti, Nucl. Phys. B 376 (1992) 365;
P. Horava, Nucl. Phys. B 327 (1989) 461;
J. Dai, R.G. Leigh, J. Polchinski, Mod. Phys. Lett. A 4 (1989) 2073.
[5] J. Polchinski, TASI lectures on D-branes, hep-th/9611050;
C. Angelantonj, A. Sagnotti, Open strings, hep-th/0204089.
[6] J.D. Blum, K.R. Dienes, Nucl. Phys. B 516 (1998) 83, hep-th/9707160;
J.D. Blum, K.R. Dienes, Nucl. Phys. B 520 (1998) 93, hep-th/9708016.
[7] I. Antoniadis, E. Dudas, A. Sagnotti, Nucl. Phys. B 544 (1999) 469, hep-th/9807011;
I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 553 (1999) 133, hep-th/9812118;
R. Blumenhagen, L. Gorlich, Nucl. Phys. B 551 (1999) 601, hep-th/9812158;
C. Angelantonj, I. Antoniadis, K. Forger, Nucl. Phys. B 555 (1999) 116, hep-th/9904092.
[8] S. Sugimoto, Prog. Theor. Phys. 102 (1999) 685, hep-th/9905159.
[9] I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 464 (1999) 38, hep-th/9908023;
C. Angelantonj, Nucl. Phys. B 566 (2000) 126, hep-th/9908064;
G. Aldazabal, A.M. Uranga, JHEP 9910 (1999) 024, hep-th/9908072;
G. Aldazabal, L.E. Ibanez, F. Quevedo, JHEP 0001 (2000) 031, hep-th/9909172;
C. Angelantonj, I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 572 (2000) 36, hepth/9911081;
C. Angelantonj, R. Blumenhagen, M.R. Gaberdiel, Nucl. Phys. B 589 (2000) 545, hep-th/0006033.
[10] E. Dudas, J. Mourad, Phys. Lett. B 486 (2000) 172, hep-th/0004165;
R. Blumenhagen, A. Font, Nucl. Phys. B 599 (2001) 241, hep-th/0011269;
C. Charmousis, Class. Quantum Grav. 19 (2002) 83, hep-th/0107126;
R. Rabadan, F. Zamora, hep-th/0207178.
[11] G.T. Horowitz, J. Polchinski, hep-th/0206228.
[12] G.T. Horowitz, A.R. Steif, Phys. Lett. B 258 (1991) 91;
G.T. Horowitz, A.R. Steif, Phys. Rev. D 42 (1990) 1950;
H. Liu, G. Moore, N. Seiberg, JHEP 0206 (2002) 045, hep-th/0204168;
H. Liu, G. Moore, N. Seiberg, hep-th/0206182;
J. Simon, JHEP 0206 (2002) 001, hep-th/0203201;
M. Fabinger, J. McGreevy, hep-th/0206196.
24
[13] E. Dudas, J. Mourad, Phys. Lett. B 514 (2001) 173, hep-th/0012071;

G. Pradisi, F. Riccioni, hep-th/0107090;
I. Antoniadis, K. Benakli, A. Laugier, Nucl. Phys. B 631 (2002) 3, hep-th/0111209;
M. Klein, hep-th/0205300.
[14] J.H. Schwarz, E. Witten, JHEP 0103 (2001) 032, hep-th/0103099.
[15] G.W. Gibbons, R. Kallosh, A.D. Linde, JHEP 0101 (2001) 022, hep-th/0011225;
F. Leblond, R.C. Myers, D.J. Winters, JHEP 0107 (2001) 031, hep-th/0106140.
[16] E. Dudas, J. Mourad, Nucl. Phys. B 598 (2001) 189, hep-th/0010179.
[17] R. Rohm, Nucl. Phys. B 237 (1984) 553;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 (1989) 75;
E. Kiritsis, C. Kounnas, Nucl. Phys. B 503 (1997) 117, hep-th/9703059;
C.A. Scrucca, M. Serone, JHEP 0110 (2001) 017;
For recent work and extensive list of references, see, for example, C.A. Scrucca, M. Serone, JHEP 0110
(2001) 017, hep-th/0107159;
D.M. Ghilencea, H.P. Nilles, S. Stieberger, New J. Phys. 4 (2002) 15, hep-th/0108183.
[18] A. Sagnotti, hep-th/9509080;
A. Sagnotti, Nucl. Phys. B (Proc. Suppl.) 56 (1997) 332, hep-th/9702093.
[19] J. Polchinski, E. Witten, Nucl. Phys. B 460 (1996) 525, hep-th/9510169.
[20] W. Fischler, L. Susskind, Phys. Lett. B 171 (1986) 383;
W. Fischler, L. Susskind, Phys. Lett. B 173 (1986) 262.
[21] M. Berg, A. Sagnotti, private communication;
G. Pradisi, private communication.
[22] M.A. Melvin, Phys. Lett. 8 (1964) 65;
G.W. Gibbons, D.L. Wiltshire, Nucl. Phys. B 287 (1987) 717, hep-th/0109093;
F. Dowker, J.P. Gauntlett, G.W. Gibbons, G.T. Horowitz, Phys. Rev. D 52 (1995) 6929, hep-th/9507143.

Nonlinear supersymmetry in quantum mechanics:

algebraic properties and differential representation
A.A. Andrianov a,b , A.V. Sokolov b
a INFN, Sezione di Bologna, Via Irnerio 46, 40126 Bologna, Italy
b V.A. Fock Institute of Physics, Sankt Petersburg State University, 198504 Sankt Petersburg, Russia
Received 20 January 2003; accepted 12 March 2003
Abstract
We study the nonlinear (polynomial, N-fold, . . .) supersymmetry algebra in one-dimensional
QM. Its structure is determined by the type of conjugation operation (Hermitian conjugation or
transposition) and described with the help of the Hamiltonian projection on the zero-mode subspace
of supercharges. We show that the SUSY algebra with transposition symmetry is always polynomial
in the Hamiltonian if supercharges represent differential operators of finite order. The appearance
of the extended SUSY with several (complex or real) supercharges is analyzed in details and it is
established that no more than two independent supercharges may generate a nonlinear superalgebra
which can be appropriately specified as N = 2 SUSY. In this case we find a nontrivial hidden
symmetry operator and rephrase it as a nonlinear function of the Hamiltonian on the physical state
space. The full N = 2 nonlinear SUSY algebra includes central charges both polynomial and
nonpolynomial (due to a symmetry operator) in the Hamiltonian.
2003 Elsevier Science B.V. All rights reserved.
PACS: 03.65.Ge; 03.65.Fd; 11.30.Pb
Keywords: Supersymmetric quantum mechanics; Nonlinear supersymmetry; Quasi-solvability; Hidden
symmetry
1. Introduction
Supersymmetric quantum mechanics [1,2] has been well proven as providing efficient
nonperturbative methods to explore new isospectral quantum systems [39] (see reviews
[1014] and references therein) and to design nuclear potentials with required properties
E-mail addresses: andrianov@bo.infn.it (A.A. Andrianov), sokolov@mph.phys.spbu.ru (A.V. Sokolov).
0550-3213/03/$ see front matter 2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0550-3213(03)00232-3
26
A.A. Andrianov, A.V. Sokolov / Nuclear Physics B 660 (2003) 2550
[15,16] as well as to find the SUSY induced hidden dynamical symmetries [1719] and,
more specifically, to search for new exactly or quasi-exactly [20,21] solvable problems in
QM [1113,17,19,2227].
When being written in the fermion number representation the one-dimensional SUSY
QM assembles a pair of isospectral Hamiltonians h+ and h into the matrix Schrdinger
operator, a super-Hamiltonian,

+
2 + V1 (x)
h
0
0
2 I + V(x),
=
H=
(1)
0 h
0
2 + V2 (x)
where d/dx. The isospectral connection between components of the super-Hamiltonian is provided by intertwining relations with the help of CrumDarboux (see [30] and
references therein) differential operators q ,
h+ q + = q + h ,
q h+ = h q ,
which, in the framework of SUSY QM, are components of the supercharges,

0 q+
2 = 0.
= 0 0 ,
,
Q
Q2 = Q
Q=
0 0
q
0
(2)
(3)
The isospectral shift (2) entails the conservation of supercharges or the supersymmetry of
the super-Hamiltonian,
= 0,
[H, Q] = [H, Q]
(4)
which represents the basis of the SUSY algebra. However its algebraic closure is given, in
general, by a nonlinear SUSY relation,
= P(H ),
{Q, Q}
(5)
where P(H ) is a function of the super-Hamiltonian. In this extended form the polynomial
(or higher-derivative) SUSY algebra was systematically introduced in [28,29]: its supercharges were realized by certain N th order differential operators,
qN =
N
wk (x) k ,
wN
= const (1)N .
(6)
k=0
The coefficient functions wk (x) are, in general, complex and sufficiently smooth. If
they are real then from the hermiticity of the Hamiltonians and from Eqs. (2) it follows
that, q (q + ) = (q + )t with the notations: for the Hermitian conjugation and t for
transposition. But in the complex case the four types of SUSY algebra can be introduced,
based on four intertwining operators and, respectively, four supercharges,

= Q ,
q = q+ ,
Q
t
c = Qt ,
qc = q + ,
Q

t
qc+ = q = q + ,
(7)
Qc = Q ,
27
where obviously stands for the complex conjugation of coefficient functions. These
algebras are generated by the pairs:
Q),
A1 = (Q,
Qc ),
A3 = (Q,
c , Q),
A2 = (Q
c , Qc ).
A4 = (Q
(8)
These sets are eventually united in the complex, nonlinear N = 2 supersymmetry (see
Section 5).
Recently the polynomial (or higher-derivative) SUSY algebra has attracted much
interest [3146] being a natural algebraic realization of the ladder [4,9] or dressing
chain [47,48] algorithms. It was rediscovered under the name of nonlinear SUSY [26,
42,46] and of N -fold SUSY [23,4345]. Perhaps the label of nonlinear SUSY (which
we adopt further on) serves better to reflect its essence as one could easily produce
nonpolynomial examples by means of limiting procedure applied to a polynomial SUSY
algebra when supercharges would become pseudo-differential operators or by considering
SUSY algebras with complex-valued supercharges (Sections 3 and 5).
Meantime it was claimed in [45] that there exists a N -fold SUSY which generalizes the
polynomial SUSY in a nontrivial way. To be precise the theorem formulated and proven in
[45] states the following.
Theorem 1 (AoyamaSatoTanaka). Let n (x) (n = 1, . . . , N) be two sets of N linearly
independent functions, zero-modes of the supercharge components (6),

qN = qN+ .
qN n = 0,
(9)
Then the following propositions hold 1
(1) The Hamiltonians h have finite matrix representations when acting on the set of
functions n (x),

Snm
h n =
(10)
m .
m
(2) With the help of the N N matrices

S , the SUSY algebra closure2 takes the general
3
form,

+ +
+
h
+
q
0
det
M

N
1
N
,
Q, Q =
(11)

0
det M
+ 2 qN+
N h
where
Q=
0
0
qN+
0

M
N (E) EI S ,
(12)
1 The first proposition is a necessary condition for the Hamiltonian system to be quasi-exactly solvable [20,21]
and it was investigated recently [4951] within the notion of conditional symmetry.
2 The Mother Hamiltonian in the terminology of [44,45].
3 It is just an algebra of A type with Q
= Q .
1
28
and 1,2
are differential operators of lower orders N1,2 N 1 generating at least
one more SUSY algebra (intertwining relations) for the same super-Hamiltonian H ,
Eq. (1), [H, P1,2 ] = [H, P1,2 ] = 0 with supercharges,

+
0 1,2
+
,
P1,2 = (P1,2 ) ,
1,2
= 1,2
.
P1,2 =
(13)
0
0
(3) In the case of nonvanishing 1,2

the SUSY algebra (11) coexists at least with one
polynomial SUSY of lower order N N1,2 < N .
(4) If for given h the N -fold supercharges are uniquely determined then 1,2
must be
zero and the superalgebra closure leads to the polynomial SUSY of order N ,

Q, Q = det M+
(14)
N (H ) = det MN (H ) PN (H ).
Certainly Theorem 1 is helpful to make links to quasi-exact solvability of particular

quantum Hamiltonians [25]. However it does not explain the origin of generators of
small supersymmetries P1,2 . It does not elucidate also the relationship between them and
S . As well it does not give a hint on what is the maximal
between the matrices
S+ and
order of a coexisting polynomial SUSY and how many supercharges may commute with a
given Hamiltonian. From (11) it is evident that the operators 1+ qN and 2 qN+ are related
to conserved symmetries of the Hamiltonians h . However neither genuine Hermitian
symmetry generators nor the very meaning of such symmetries have not been obtained
in the framework of Theorem 1.
The main aim of the present work is to clarify the above mentioned, missing points and
furthermore to prove the following.
(1) For a given pair of isospectral systems intertwined by differential operators of order
N there is always a choice of certain intertwining operators with real coefficients
(not necessarily unique) which lead to supercharges of a polynomial SUSY of the
same order N . Thereby the N -fold SUSY of [44,45] for a given quantum system
always coexists with a polynomial SUSY of the same order (and possibly few other
polynomial SUSY of different orders).
(2) The complex extension of nonlinear SUSY may bring a SUSY algebra different from
a polynomial one just containing nontrivial symmetry operators. These differential
operators of odd order can be replaced by nonpolynomial functions of a Hamiltonian
being defined in the Hilbert space spanned on eigenfunctions of the Hamiltonian. Thus
for the same Hamiltonian one can simultaneously introduce the nonlinear SUSY in
both a polynomial and a nonpolynomial form. In particular it covers the propositions
of Theorem 1.
commuting with
(3) Among the (infinite) variety of supercharges of type Q (or of type Q)
a given Hamiltonian one can systematically find the optimal set of (no more than) two
basic SUSY generators which are differential operators of even and odd order with real
represent
coefficients. Respectively all other supercharges of type Q (or of type Q)
a linear combination of basic supercharge(s) with coefficient(s) polynomial in the
Hamiltonian components. For two essentially independent supercharges the symmetry
operator is unique up to a multiplier polynomial in the Hamiltonian.
29
(4) There is a more efficient formulation of Theorem 1 which manifestly uses the emerging
dynamical symmetry for a super-Hamiltonian with two supercharges and uniquely
specifies the relationship between q , 1 and 2 and between the matrices
S+ and

S .
(5) For isospectral systems with two independent supercharges the notion of irreducibility
for polynomial SUSY formulated in [29] does not characterize firmly potentials and
the same system may be well described by a more reducible and less reducible SUSY
algebra.
All theorems and constructions are exemplified by means of an exactly solvable system
of second order.
2. Superalgebras with transposition symmetry

The superalgebras with real coefficient functions in the differential representation
of supercharges as well as the A2,3 complex superalgebras have the transposition
c = Qt . The following theorem is valid for these superalgebras (compare
symmetry, Q
with Theorem 1).
Theorem 2 (SUSY algebras with T-symmetry). Let us again introduce two sets of N
linearly independent functions n (x) (n = 1, . . . , N) which represent complete sets of
zero-modes of the supercharge components (6),
t
qN n = 0,
(15)
qN = qN+ .
Then:
(1) the Hamiltonians h have finite matrix representations when acting on the set of
functions n (x),

h n =
(16)
Snm
m ;
m
c = Qt takes the polynomial form,

(2) the SUSY algebra closure with Q

Q, Qt = det EI S+ E=H = det EI S E=H PN (H ),
(17)
irrespectively on whether the supercharge of order N is unique or there exist several

supercharges for a given super-Hamiltonian H .
S+
We stress that the matrix S is the same as in Theorem 1, S =

S whereas the matrix
4
+

is different from S due to Eq. (15).
4 From the definitions (9) and (15) it follows that the spectra of two matrices S+ and
S+ are mutually complex
conjugated.
30
The proof of the first statement of Theorem 2 is analogous to that one of Theorem 1.
Namely, one has to act by the operator intertwining relations (2) on the zero-mode
functions n (x),
qN h n = h qN n = 0,
(18)
and conclude therefrom that h n
is also a zero-mode solution, i.e., can be expressed as a

linear combination (16) of a complete set of n (x).
The proof of the second part Theorem 2 is based on the properties assembled into the
lemma.
Lemma 1. Let
1 , . . . , N be two sets of eigenvalues of matrices S as being introduced in
the above formulated theorem. Then there exist two sets of first-order differential operators
r1 , . . . , rN such that:
(1) they have the canonical form,

rl = + l (x),
l = 1, . . . , N,
(19)
where the functions l (x) may be complex and/or singular at some points;
(2) the factorizations hold,
qN+ = r1+ rN+ ,
qN = r1 rN ;
(3) the chain relations take place,

t
t
rl rl +
l = rl+1 rl+1 + l+1 hl ,
t
rN rN +
N = h hN ,
t
r1 r1 +
1 = h h0 ;
(20)
l = 1, . . . , N 1,
(21)
(4) the intermediate Hamiltonian operators have the Schrdinger form,
2
h
l = + vl (x),

2

2

vl (x) = l (x) l (x) +

l = l+1 (x) l+1 (x) + l+1 ,
but, in general, with complex and/or singular potentials;

(5) the intertwining relations are valid,
t
t
rl hl1 = h
.
h
l rl
l1 rl = rl hl ,
(22)
(23)
The proof of Lemma 1 is based on the quasi-diagonalization of matrices S , i.e., on

their reduction to the Jordan canonical form
S which is block-diagonal and contains the
Jordan cells with eigenvalues on the main diagonal and unities on the upper subdiagonal.
This diagonalization can be realized by nondegenerate linear transformations of the
zero-mode sets which induce the similarity transformations of matrices S ,
l =
N

m=1

lm
m ,

1

S = S .
h l =
N

Slm
m ,
m=1
(24)
31
Certainly it is sufficient to elaborate the factorization of the operator qN . The last line in
the matrix
S+ contains the Hamiltonian eigenvalue +
N,
h+ N
= +
N N .
(25)
Next we define,
rN N
= 0,

)
( N
.

(26)
From Eq. (25) it follows that

t
+
h+ h+
N = rN rN + N .
(27)
Furthermore, the intermediate Schrdinger-like Hamiltonian can be introduced,

2
t
+
2
h+
+ N + +
N1 = rN rN + N = + N
N,
which is obviously involved in the intertwining relations with
t +
t
+
h+
rN hN1 = h+
N rN .
N1 rN = rN hN ,
(28)
h+
N,
(29)
The combination of Eqs. (15) and (26) yields the factorization,
qN = qN1
rN .
(30)
Indeed if (15) and (26) are valid then in (6)

w0 (x) =
N
wk (x)k (x),

k1
k (x) N
N ,
(31)
k=1
where we have introduced the notation for the derivative of coefficient functions to make
a clear distinction from the differential operator . Therefore
qN =
N
k1

wk (x) k + N
N .
(32)
k=1
Respectively the factorization is realized in each component,

k2

k

k1
k1
km2
+
=
m+1 + .
N
(33)
m=0
Now it is straightforward to show that:

(1) the intertwining relation holds,
h qN1
= qN1
h+
N1 ;
(34)
(2) the functions,

n = rN n ,
1 n N 1,
(35)
form the complete, linearly independent set of solutions of the equation qN1
= 0;
32
(3) the matrix

S+
N1 which is uniquely determined from relations,
h+
N1 n =
N1
+

SN1 nm m
,
(36)
m=1
in fact, is derived from

S+ after deleting of the last column and the last line: thereby
+

the matrix SN1 still has the Jordan canonical form and its spectrum consists of
+
+
1 , . . . , N1 .
The first two statements are direct consequences of the basic relations (2) and (15)
respectively, whereas the third one can be obtained when acting by the operator rN on the
definition (24) of the matrix
S+ .
Thus we have reduced the factorization problem of order N to the latter one of order
N 1 having proved the statements of the lemma on this step. Evidently one can proceed
recursively further on and prove completely Lemma 1 by induction.
In turn, the proof of Theorem 2 uses the factorization and intertwining relations provided
by Lemma 1,
t
t
t
t
+
qN+ qN = rN r1 r1 rN = rN r2 h+
1 1 r2 rN

t
t
= rN r2 r2 rN h+ +
1
+

+
+
= = h+ +
1 h N = det EI S E=h+ ,
t
t
qN qN+ = r1 rN rN r1
t
t
+
hN1 +
= r1 rN1
N rN1 r1

+
+
= = h +
1 h N = det EI S E=h .
(37)
The same relation can be derived for the factorization of the operator qN+ into a product of
rl+ using the Jordan form for the matrix
S . It leads to the equivalent representation of the
polynomial algebra,

qN qN = h
1 h N = det EI S E=h .
(38)
As the relations (37) and (38) are operator ones they hold for any values of spectral
parameter h = #. Therefore the eigenvalues of matrices S+ and S and their
corresponding degeneracies coincide.
In general, the eigenvalues of matrices S are complex. For real components of
supercharges qN , the complex eigenvalues obviously appear in complex conjugated pairs
providing the real polynomial PN (x) but for complex supercharge components qN the
resulting polynomial contains complex coefficients. In next sections we examine the
nonuniqueness of a complex supercharge describing a given Hermitian super-Hamiltonian.
33
3. Several supercharges and extended SUSY

The nonuniqueness of supercharges was mentioned for the first time5 in [52]. It

was observed that for a Hermitian Hamiltonian H the conserved supercharges Q, Q
with complex intertwining components qN always generate two SUSY algebras: one for
and another one for their imaginary parts P , P where the
their real parts K, K
corresponding labels are referred to the real and imaginary parts of coefficients in the
+ ipN
,
differential intertwining operators qN = kN
1
Q = K + iP ,
= [H, K]
= [H, P] = 0.
[H, Q] = [H, K] = [H, P ] = [H, Q]
(39)
t
t

Evidently the conjugated operators can be defined uniquely, K = K = K , P = P = P ,
One can always
independently on what a choice is taken from (8) for the operators Q, Q.
employ the normalization (6) of the senior derivative in qN on a real constant. Then the
second supercharge P appears to be a differential operator of lower order N1 < N .
The appearance of the second supercharge conventionally implies the extension of
SUSY algebra. To close the algebra one has to include all anticommutators between
supercharges, i.e., the full algebra based on two supercharges K and P with real
intertwining operators. Two supercharges generate two polynomial SUSY,

N (H ),
N1 (H ).
P,P = P
K, K = P
(40)
This SUSY algebra has to be embedded into a N = 2 SUSY.6
The closure of the extended, N = 2 SUSY algebra is given by
+

pN1 kN
0

,
P,K R =
+
0
kN
pN1
+

kN pN1
0

=
K, P R
+ .
0
pN
k
1 N
(41)
= R = Rt are differential operators of

Evidently the components of operators R, R
N + N1 order commuting with the Hamiltonians h , i.e., they are symmetry operators.
However, in general, they are not polynomials of the Hamiltonians h and this symmetry
imposes certain constraints on potentials.7
5 See the related section in the E-archive version of [52] as it was eliminated from the final journal paper under
the severe pressure of a referee.
6 There is a misinterpretation concerning the classification of extended SUSY in QM. The conventional N = 1
whereas the N = 2 SUSY should employ two
SUSY deals with non-Hermitian nilpotent supercharges Q, Q
j satisfying the extended SUSY algebra with certain central charges. We
pairs of nilpotent supercharges Qj , Q
are grateful to A. Smilga for the discussion of this point. However there are papers (see discussion in [14])

where the SUSY QM algebra is defined as N = 2 SUSY in terms of Hermitian supercharges Q1 = Q + Q,
We would like to stress that an elementary SUSY charge is nilpotent carrying fermion quantum
Q2 = i(Q Q).
numbers. Moreover a nontrivial dynamics cannot be obtained with one real SUSY charge (as it was recently
mentioned in [14]).
7 Such type of symmetries in one-dimensional QM and their possible relation to the Lax method in the soliton
theory was discussed in [30,49].
34
commute each to other. Moreover the

N (H ), P
N1 (H ), R, R
All four operators P
Hermitian matrix describing this N = 2 SUSY,

N (H )
R
P
,
Z(H ) =

N1 (H )
R
P

N P
= 0,
N RR
det Z(H ) = P
(42)
1
is degenerate. Therefore it seems that the two supercharges are not independent and by
their redefinition (unitary rotation) one might reduce the extended SUSY to an ordinary
N = 1 one. However such rotations cannot be global and must use nonpolynomial, pseudodifferential operators as parameters. Indeed, the operator components of the central
charge matrix Z(H ) have different order in derivatives. Thus, globally the extended
nonlinear SUSY deals with two sets of supercharges but when they act on a given
eigenfunction of the Hamiltonian H one could, in principle, perform the energy-dependent
rotation and eliminate a pair of supercharges. Nevertheless this reduction can be possible
only after the constraints on potentials have been resolved.
and the
Let us find the formal relation between the symmetry operators R, R
Hamiltonian. These operators can be decomposed into a Hermitian and an antihermitian
parts,
+

1
0
b
,
B (R + R)
0 b
2
+

1
0
i e
.
iE (R R)
(43)
0 e
2
The operator B is a differential operator of even order and therefore it may be a polynomial
of the Hamiltonian H . But if the operator E does not vanish identically it is a differential
operator of odd order and cannot be realized by a polynomial of H .
The first operator plays essential role in the one-parameter nonuniqueness of the SUSY
algebra. Indeed, one can always redefine the higher-order supercharge as follows,
( ) ( )
( ) (H ),
K ,K
K ( ) = K + P ,
(44)
=P
N
keeping the same order N of polynomial SUSY for arbitrary real parameter . From (44)
one gets,
( ) (H ) P
N (H ) 2 P
N (H ),
2 B(H ) = P
1
N
(45)
thereby the Hermitian operator B is a polynomial of the Hamiltonian of the order Nb

N 1. Let us use it to unravel the Hamiltonian content of the operator E,
N (H )P
N1 (H ) B 2 (H ),
E 2 (H ) = P
(46)
which follows directly from (42) and (43). As the (nontrivial) operator E(H ) is a
differential operator of odd order Ne it may have only a realization nonpolynomial in H
being a square root of (46) in an operator sense. This operator is certainly nontrivial if the
sum of orders N + N1 of the operators kN

and pN
is odd and therefore Ne = N + N1 . For
1
an even sum N + N1 we cannot in general make any definite conclusion concerning the
35
nontriviality of E(H ). However it will be shown in Section 7 that if the symmetry operator
and pN
an optimal set of independent
is nonzero then for any choice of the operators kN
1
supercharges (possibly of lower orders) can be obtained which is characterized by an odd
sum of their orders.
The existence of a nontrivial symmetry operator E commuting with the Hamiltonian
results in common eigenstates which however are not necessarily physical wave functions.
In general they may be combinations of two solutions of the Shrdinger equation with
a given energy, the physical and unphysical ones. But if the symmetry operator E
is Hermitian in respect to the scalar product of the Hilbert space spanned on the
eigenfunctions of the Hamiltonian H then both operators have a common set of physical
wave functions. This fact imposes quite rigid conditions on potentials.
In particular, for intertwining operators with sufficiently smooth coefficient functions
having constant asymptotics at large coordinates the symmetry operator E has the similar
properties and is evidently Hermitian. In this case one has nonsingular potentials with
constant asymptotics at large x and therefore a continuum energy spectrum of H with
wave functions satisfying the scattering conditions. Thus the incoming and outgoing states,
in (x) and out (x), at large x are conventionally represented by combinations of plane
waves which are solutions of the Schrdinger equation for a free particle,
(x)|x exp(ikinx) + R(kin ) exp(ikinx),

(x)|x+ 1 + T (kout) exp(ikoutx),
(47)
where the reflection, R(kin ), and transmission, T (kout), coefficients are introduced. Since
the symmetry is described by a differential operator of odd order which at large x tends to
an antisymmetric operator with constant coefficients the eigenstates of this operator at large
coordinates approach to individual plane waves exp(ikx) with opposite eigenvalues
kf (k 2 ) and cannot be their combinations. Hence the eigenstate of the Hamiltonian
with a given value of the operator E may characterize only the transmission and cannot
have any reflection, R(kin ) = 0. We conclude that the corresponding potentials V1,2 in (1)
inevitably belong to the class of transparent or reflectionless ones [53].
As the symmetry operator E is Hermitian its eigenvalues are real but, by construction,
its coefficients are purely imaginary. Since the wave functions of bound states of the system
H can be always chosen real we conclude that they must be zero-modes of the operator
E(H ),
E(H )i = E(Ei )i = 0,
N (Ei )P
N1 (Ei ) B 2 (Ei ) = 0,
P
(48)
which represents the algebraic equation on bound state energies of a system possessing
two supersymmetries. Among solutions of (48) one reveals also a zero-energy state at the
bottom of continuum spectrum. On the other hand one could find also the solutions which
are not associated to any bound state. The very appearance of such unphysical solutions is
accounted for by the trivial possibility to replicate supercharges by their multiplication on
the polynomials of the Hamiltonian and it is discussed in Section 6.
36
4. Example: N = 2, N1 = 1
Let us examine the algebraic structure of the simplest nonlinear SUSY with two
supercharges,
k 2 2f (x) + b(x)
f (x),
p + (x),
(49)
induced by the complex supercharge of second order in derivatives [52]. The supersymme and P , P prescribe that
tries (39) generated by K, K
2
f
f
d
V1,2 = 2 = 2f + f 2 +
a,
2f
2f
4f 2
2
f
f
d
b = f 2
(50)
+
+
,
2f
2f
4f 2
where , f are real functions and a, d are real constants. The related superalgebra closure
and P , P takes the form,
for K, K
= (H + a)2 + d,
{K, K}
{P , P} = H,
(51)
the latter one clarifies the role of constants a, d.

The compatibility of two supersymmetries is achieved on solutions of the following
equations:
2
f
f
d
2
= 2f + 0 ,
(52)
f +
a = 2 = (2f + 0 )2 ,
2f
2f
4f 2
where 0 is an arbitrary real constant. The latter one represents a nonlinear secondorder differential equation which solutions are parameterized by two integration constants.
Therefore as it was advertised the existence of two SUSY constrains substantially the class
of potentials for which they may hold.
Let us use the freedom to redefine the higher-order supercharge (44) for eliminating
the constant 0 in (52). After this simplification Eq. (52) is integrated into the following,
first-order one,
= 2f,
(f )2 = 4f 4 + 4af 2 + 4G0 f d 4 (f ),
(53)
where G0 is a real constant.

The solutions of this equation are elliptic functions which can be easily found in the
implicit form,
f(x)
f0
df
= (x x0 ),
4 (f )
where the lower limit of integration f0 and x0 are real constants.

It can be shown that they may be nonsingular in three situations.
(54)
37
(a) The polynomial 4 (f ) has four different real roots f1 f2 f3 f4 and f0 is

chosen between two roots f2 and f3 . The corresponding potentials are periodic. This
case will not be examined here.
(b) 4 (f ) has three different real roots and the double root /2 is either the maximal one
or a minimal one,

2 2
2
f+
# , 0 < # < 2.
4 (f ) = 4 f
(55)
2
2
Then there exits a relation between constants a, d, G0 in terms of coefficients , #,

2

3
3 2
< 0,
G0 = 2 # ,
# .
d = 2
a=#
(56)
2
4
Besides the constant f0 is taken between the double root and a nearest simple root.
(c) 4 (f ) has two different real double roots which corresponds in (55), (56) to G0 = 0,
2 = # > 0, a = #/2, d = # 2 /4. The constant f0 is taken between the roots.
The corresponding potentials V1,2 are well known [53] and in the cases (b) and (c)
are reflectionless, with one bound state at the energy ( 2 #) and with the continuum
2
spectrum starting from
. Respectively the scattering wave function is proportional to
exp(ikx) with k = E 2 .
In particular, in the case (b) the potentials coincide in their form and differ only by shift
in the coordinate,
2#
,
(1,2)
ch ( #(x x0 ))
1
#
(1,2)
x
= x0 ln
,
4 # + #
V1,2 = 2
and in the case (c) one of the potential can be chosen constant,

2
2
2
.
V1 = ,
V2 = 1 2
ch ((x x0 ))
(57)
(58)
For these potentials one can illustrate all the relations of extended SUSY algebra.
The initial algebra is given by the relations (51). The first, polynomial symmetry
operator turns out to be constant, B(H ) = G0 when taking into account (49) and (53).
The second symmetry operator reads,

3
3
3
E(H ) = i I aI + V(x) + V (x) ,
(59)
2
4
in terms of the potential (1). From the identity (46) or directly from Eq. (59) one derives
with the help of Eqs. (53) and (56) that,

E 2 (H ) = H (H + a)2 + d G20 = (H Eb )2 H 2 ,
(60)
where Eb = 2 # is the energy of a bound state. Thus some of the zero modes of E(H )
characterize either bound states or zero-energy states in the continuum. We remark that in
38
the case (c) only the Hamiltonian h has a bound state. Hence the physical zero modes of
E(H ) may be either degenerate (case (b), broken SUSY) or (one of them) nondegenerate
(case (c), unbroken SUSY).
The square root in (60) can be carried out,

E(H ) = (H Eb ) H 2 .
(61)
We notice that the symmetry operator (59), (61) is irreducible, i.e., the binomial (H Eb )
cannot be stripped off (the exact meaning of this operation see in Section 6). Indeed the
elimination of this binomial would lead to an essentially nonlocal operator. The sign of
square root in (61) is fixed from the conventional asymptotics of scattering wave functions
exp(ikx) and the asymptotics V1,2 2 by comparison of this relation with Eq. (59).
When taking Eq. (61) into account one finds the nonpolynomial relations of the
extended SUSY algebra,

K, P = K , P = G0 i(H Eb ) H 2 .
(62)
The Hermitian matrix Z(H ), Eq. (42) is built of the elements (51) and (62) and evidently
cannot be diagonalized by a unitary rotation with elements polynomial in H . Thus the
algebra must be considered to be extended in the class of differential operators of finite
order.
It remains to clarify the very nonuniqueness of the higher-order supercharge, namely,
its role in the classification of the polynomial SUSY. For arbitrary in (44) one obtains

( ) ( )
= H 2 + 2a + 2 H + a 2 + d + 2 G0 = (H + a )2 + d ,
K ,K

1
1
d = d + 2 G0 a 2 4 4 ,
a = a + 2 ,
(63)
2
4
2
where 4 (f ) is defined in Eq. (53).
One can see that the sign of d , in general, depends on the choice of . For instance, let
us consider the case (b) when

1
d = ( + )2 ( )2 4 2 # .
(64)
4
Evidently if lies in between the real roots of the last factor in (64) then d is positive and
otherwise it is negative. But two real roots always exist because 2 > #. Thereby the sign
of d can be freely negative or positive without any change in the Hamiltonians. Hence in
the case when the polynomial SUSY is an extended one, with two sets of supercharges, the
irreducibility or reducibility of a polynomial SUSY algebra does not signify any invariant
characteristic of potentials.
5. Complex SUSY algebras

If the intertwining operators q have complex coefficients in (6) then we deal with two
supercharges which we adopt to be independent (see Section 7 for its exact definition). One
can split again the complex supercharge Q into a real, K and an imaginary, P counterparts
39
as in Eq. (39) and normalize them so that the intertwining operator in K has a higher
order in derivatives. Two SUSY algebras with transposition symmetry, A2 and A3 , are
polynomial in virtue of the theorem of Section 2. In terms of real supercharges they have
the following structure,
N (H ) P
N1 (H ) + i2B(H ),
c } = P
{Q, Q
N (H ) P
N (H ) i2B(H ),
=P
{Qc , Q}
(65)
(66)
c = Q = K + iP ,
Q
t
Qc = Q = K iP ,
= K iP .
Q
(67)
Two more algebras, A1 and A4 can be built with a Hermitian closure according to Eq. (8).
In particular, the algebra A1 (used in Theorem 1, Section 1) is completed by the following
closure,
N1 (H ) 2E(H ).
N (H ) + P
=P
{Q, Q}
(68)
Respectively the algebra A4 is completed by the relation,

N (H ) + P
N1 (H ) + 2E(H ).
c } = P
{Qc , Q
(69)
When the symmetry operator E(H ) is nontrivial they are essentially nonpolynomial (see
(46)).
We conclude that for complex intertwining operators the same pair of isospectral
Hamiltonians may be induced both by the polynomial SUSY algebra (40) (or (65), (66))
and by the nonpolynomial one (68), (69).
All four superalgebras Am generated by (Q1 , Q2 ) (Q, Qc ) can be assembled into the
extended N = 2 SUSY algebra,

N (H ) + (I 1 )P
N1 (H ) 2 2B(H ) 3 2E(H ) ,
j } = (I + 1 )P
{Qi , Q
(70)
ij
with i, j = 1, 2. It is equivalent, of course, to the algebra (40) and (41).
Let us illustrate such an algebra using the N = 2, N1 = 1 example of Section 4. Thus
the intertwining operator q + = k + + icp+ is composed from the operators (49) where the
constant c of mass dimension is introduced from dimensional reasons. Respectively,

j } = (I + 1 ) (H + a)2 + d + (I 1 )c2 H
{Qi , Q

2 2cG0 3 2c(H Eb ) H 2 ij ,
(71)
i.e., is manifestly nonpolynomial in respect to the Hamiltonian H .
It remains to clarify the relationship between the Hermitian algebra A1 determined
in Theorem 1 by Eq. (11) and that one given by (68). For this purpose we relate
the representation (11) to the algebra with transposition symmetry, Eq. (17). In order
to establish the exact correspondence the upper and lower components in the matrix
have to be treated differently. Namely for the upper component q + q the suitable
{Q, Q}
N N
+
, yields,
decomposition, qN+ = (qN )t + 2ipN
1

t
+
+
q = det EI
S+ E=h+ + 2ipN
q ,
qN+ qN = qN qN + 2ipN
1 N
1 N
(72)
40
where from one reproduces the upper component of Eq. (11) after the identification
+
. We stress that the matrix
S+ does not coincide with S+ from Eq. (16) because
1+ 2ipN
1
+
+ t
in (72) qN = (qN ) = (qN ) . But in appropriate bases of mutually complex conjugated
functions they may be related by complex conjugation,
S+ = (S+ ) .
Similarly the lower component can be transformed into the form (11) by means of the
and reads,
decomposition, qN = (qN+ )t 2ipN
1

t
+
+
qN qN+ = qN+ qN+ 2ipN
(73)
q = det EI
S E=h 2ipN
q ,
1 N
1 N
where from one obtains the lower component of Eq. (11) after the identification 2
= 1 . The matrix
S is exactly as S in Theorem 2, Eq. (16).
2ipN
1
The AST decomposition (72), (73) is certainly equivalent to the representation
(68) but supplies both the polynomial part and the nonpolynomial symmetry operator
with imaginary contributions which eventually assemble into 2e (h ). Thereby the
Hermitian symmetry operator E(H ) nonpolynomial in H is hidden in Eqs. (72), (73). This
is why we give our favor to the representation (68) in the analysis of supersymmetries with
several supercharges.
6. Strip-off problem
The pair of two supersymmetries analyzed in Sections 3 and 5 may rigidly determine
the class of potentials V1,2 in (1) to a specific, transparent type of them contracting the
freedom in their choice from a functional one to a parametric one. On the other hand, there
and pN
are related by a
exists a trivial possibility when the intertwining operators kN
1
factor depending on the Hamiltonian,

kN
(74)
= F h pN
=
p
F h ,
N
1
1
where F (x) is assumed to be a polynomial. Obviously in this case the symmetry operator
E(H ) identically vanishes and the appearance of the second supercharge does not result in
any restrictions on potentials.
More generally the orders of polynomial superalgebras and some of the roots of
associated polynomials may not be involved in determination of the structure of the
and pN
be reducible to some lower-order
potentials. In particular, let the operators kN
1
ones kN and pN ,
1

kN = Fk h kN = kN
(75)
pN
= Fp h p N
Fk h ,
= p N
Fp h ,
1
1
and N1 , N
1 are simultaneously odd or even and
where the numbers in the pairs N , N
)/2 and (N1 N
1 )/2. Then evidently the
Fk (x), Fp (x) are polynomials of order (N N
superalgebra generated by kN and p N equally well characterizes the super-Hamiltonian

1
system with the same potentials.
We have come to the problem of how to discern the nontrivial part of a supercharge
and avoid multiple SUSY algebras generated by means of dressing (75). It can be
systematically performed with the help of the following theorem.
41
Theorem 3 (Strip-off). Let us admit the construction of the theorem on SUSY algebras
S+ ) generated on the
with T-symmetry. Then the requirement that the matrix
S (or
+
subspace of zero-modes of the operator kN

(or kN
) contains n pairs (and no more)
of Jordan cells with equal eigenvalues l and the sizes l of a smallest cell in the lth
+
(or kN
) to be
pair is necessary and sufficient to ensure for the intertwining operator kN
represented in the factorized form:
kN
= kN

n

l h
l
(76)
l=1
n
8

where kN
l=1 l which cannot be
are intertwining operators of order N = N 2
decomposed further on in the product similar to (75) with Fk (x) = const.
S+ is similar. It
We shall perform the proof of the theorem for
S only as its proof for
is based on the lemma and two remarks.
Remark 1. The matrices
S cannot contain more than two Jordan cells with the same
eigenvalue because otherwise the operator h would have more than two linearly
independent zero-modes (not necessarily normalizable).
+
Lemma 2. In order that the intertwining operator kN
could be factorized,

+
+
h ,
= kN2
kN
(77)
+
with kN2
being another intertwining operator of order N 2, it is necessary and sufficient
for the matrix
S to contain two Jordan cells with the same eigenvalue .
Proof. The requirement of this lemma is sufficient because if

S contains two Jordan
+
cells with the same eigenvalue then the kernel of kN includes two linearly independent
solutions of the equation h = . When repeating the way of proof of Theorem 2 one
+
can derive that kN
is factorized in the form,

+
+
2 (x) 1 (x) ,
= kN2
kN
(78)
+
where kN2
is a differential operator of order N 2 and the functions 1 (x) and 2 (x)
are chosen to provide the equal kernels of operators h and ( 2 )( 1 ). As a
differential operator of second order with the unit coefficient at 2 is uniquely determined
by two linearly independent elements of its kernel one concludes that
h = ( 2 )( 1 ),
and therefore (77) is valid. At last, from the relations,

+
+
+
+
kN2
h ,
h h = kN
h = h+ kN
= h+ kN2
+
one obtains that the operator kN2
is intertwining.
8 In this theorem the intertwining operators k , k and the parameters may also be complex.
l

N N
(79)
(80)
42
+
The condition (77) is also necessary as in this case the kernel of kN
includes two linearly
independent solutions of the equation h = which induce two different Jordan cells
with the same eigenvalue.
Remark 2. Within Lemma 2 let us put the lower lines of two Jordan cells with the
eigenvalue into the ith and j th line (i < j ) of
S respectively and introduce the functions,

+
1 l i 1,
h l ,
+

+
h l+1 , i l j 2,
l (x) =
(81)

+
h l+2
, j 1 l N 2,
+ form the basis of the kernel of the operator k + which supply
where the functions m
N
the matrix
S with the Jordan form. Evidently the functions l+ (x) are linearly
+
,
independent because in the opposite case a nontrivial linear combination of 1+ , . . . , i1
+
+
+
+
+
+
, . . . , , , . . . , must be a combination of zero-modes and . Thus they
j 1
i+1
j +1
+
form the complete set of solutions of the equation kN2
= 0 which comes from (77). The

matrix SN2 which describes the Hamiltonian action on the zero-mode subspace { k+ (x)}
+
of the intertwining operator kN2
, can be produced from the matrix
S by deleting both
the ith and j th columns and lines. Thereby it has a Jordan form.
Now from Lemma 2 and Remarks 1 and 2 one derives all statements of Theorem 3.
This theorem naturally supplements Theorem 2 as it entails the essential identity of the
S+ (up to transposition of certain Jordan cells).
Jordan forms
S and
Let us illustrate Theorem 3 by the example: the matrix
S for the intertwining operator
+
k3 with Jordan cells of different size having the same eigenvalue. It is generated by the
operators,
p = + ,
h = p p + ,

k3+ = p+ p p+ = p+ h .
(82)
Respectively:
2+ : p+ 2+ = 0
0, p p+ 3+ = 0
3+ : p+ 3+ =
+ : p p+ + = +
1
Thus,
1

S = 0
0 0
0
0 .
h 2+ = 2+ ,
h + = + ,
3
h 1+
(83)
= 1+ + 2+ .
(84)
As a consequence of Theorem 3 one finds that the supercharge components cannot be

N (x) in the SUSY algebra closure (40) does
factorized in the form (75) if the polynomial P
not reveal the degenerate zeroes. Indeed the SUSY algebra closure contains the square of
polynomial F (x), for instance,

+

+
h ,
F h = Fk2 h P
kN = Fk h kN
kN
N
kN

43
(85)
(x) is a polynomial of lower order, N

N 2. Therefore each zero of the
where P
N
polynomial Fk (x) will produce a double zero in the SUSY algebra provided by (85).
Thus the absence of double zeroes is sufficient to deal with the SUSY charges nonfactorizable in the sense of Eq. (75). However it is not necessary because the degenerate
zeroes may well arise in the ladder (dressing chain) construction giving new pairs of
isospectral potentials (see, for instance, [29] for the polynomial SUSY of second order).
Further on we consider only the stripped-off supercharges. In this case the existence
of two intertwining operators results in more equations on their coefficient functions and
thereby on the potentials in h . When they are compatible they rigidly dictate the form of
potentials leaving only the parametric freedom for their choice.
Still the stripped-off supercharges do not necessarily represent an optimal set of them
and provide the optimal structure of the symmetry operator E(H ).

k p made of
Let us illustrate it with the sample intertwining operators t = pN
1 N N1
two stripped-off supercharges with components kN and pN1 . If the differential operators
kN
and pN
have even and odd order respectively and the operators b are non-zero the
1
or pN
because the components of symmetry
operators t cannot be stripped off till kN
1
operator e = 2 (kN pN1 pN1 kN ) are surely nontrivial and therefore the operators

pN1 and pN
k are not polynomials of the Hamiltonian. One can see it manifestly for
kN
1 N
the supercharges (49) of the example in the case (b) when G0 = 0.
has
Meantime the system composed of two supercharges with components t and pN
1
the symmetry realized by the operator Et (H ) with components,
1
1

et = i t pN
k
=
i
p
t
p
p
k pN1 pN1
N
N
N
N
1
1
1
1 N
2
2

N1 h ,
= e P
(86)
Eq. (41)
where to obtain it the commutation of the super-Hamiltonian with operators R, R,
N1 (H )
has been used. Thus the symmetry operator Et (H ) contains a polynomial factor P
which zeroes are not in general related to any bound state energies. The lesson is that the
symmetry operator must be stripped-off in addition to the intertwining operators in order
to avoid unphysical zeroes of the function E(E). In order to perform the minimization of
the symmetry operator one can employ Theorem 3 and analyze the Jordan form of the
relevant Hamiltonian projection

S
e on the zero-mode subspace of the operator e . Then
the elimination of pairs of equal eigenvalues from different Jordan cells would essentially
reduce the spectrum of the projected Hamiltonian towards a set of its bound state energies.
On the other hand the intertwining operators t represent a linear combination of kN
and pN
with coefficients depending on the Hamiltonian,
1

N1 h k .
t = 2b h pN
(87)
P
N
1
N do not have common roots one may find the intertwining
If the polynomials b and P
1
and pN
of lower order
operators t which cannot be stripped off till a combination of kN
1
in derivatives (see again the case (b) in Section 4).
44
However one can easily build the equivalent supercharges t = t 2b (h )pN

1
which can be stripped off till kN

. Thus in order to construct the optimal basis of
supercharges one should not only factorize out the polynomials of the Hamiltonian but also
examine their various linear combinations with coefficients polynomial in the Hamiltonian.
7. Optimization of supercharges
As it follows from the previous discussion the existence of several supercharges is
controlled by a nontrivial symmetry operator. If there are several SUSY generators the
necessity arises: (a) to introduce the notion of (in)dependence of intertwining operators;
(b) to find out how many independent supercharges can coexist; (c) to define an optimal
basis of intertwining operators.
Let us extrapolate the relation (87) and define the operators qi , i = 1, . . . , n to be
dependent if and only if the polynomials i (y) exist such that not all of them are vanishing
and
n

i h qi = 0.
(88)
i=1
If the relation (88) results in i (y) = 0 for all i the corresponding SUSY generators are
independent. Evidently the (in)dependence of qi+ entails the (in)dependence of qi and
vice versa.
The following theorem plays a key role in resolution of how many independent
supercharges can commute with a given Hamiltonian.
Theorem 4 ((In)Dependence of supercharges). Consider two nontrivial intertwining
operators qi , i = 1, 2 with transposition symmetry qi+ = (qi )t which in general may have
complex coefficients and let us normalize them in accordance to (6). Then the strippedoff intertwining operators qi coincide if and only if the symmetry operator made of qi
vanishes, q1+ q2 q2+ q1 = 0 (or equivalently q1 q2+ q2 q1+ = 0).
The proof of this theorem is based on the lemma.
Lemma 3. If q1+ q2 q2+ q1 = 0 (or equivalently q1 q2+ q2 q1+ = 0) then the sets of

Jordan cells (and their sizes) in the matrices
S
1 and S2 for the operators q 1 and q 2 are
identical.
To prove its validity we first remark that evidently the symmetry operator q1+ q2 q2+ q1
for stripped-off operators q1 and q2 also vanishes in virtue of intertwining relations and
because the factor polynomials for qi+ and qi are equal. Therefore the operator
+
b12
= q1+ q2 = q2+ q1 =

1
+
q1 + q2+ q1 + q2 q1+ q1 q2+ q2
2
(89)
45
is a polynomial of the Hamiltonian h+ (compare with (43)). Hence, according to

+
Theorem 3 the matrix
S
12 on the basis of zero-modes of the operator b12 contains two
(and evidently no more) Jordan cells of the same size for each eigenvalue.
Next, from the Theorems 2 and 3 one concludes that the spectrum of the matrix
S
12

joins the spectra of
S
and
S
for
the
operators
q
and
q
.
Moreover
since
the
latter
ones
1
2
1
2
are stripped off the related matrices
S
i include one Jordan cell only for each eigenvalue.
At last, taking into account that ker(qi ) ker(qj+ qi ) one can derive that for each

eigenvalue of
S+
i a related Jordan cell with the same eigenvalue exists in S12 and its size is
+
no less than that one of the Jordan cell in
Si .

Thus the number and sizes of Jordan cells in matrices
S
1 and S2 are the same.
As a consequence the orders of differential operators q1+ and q2+ are the same and they
can be combined to form another intertwining operator of a lower order, + = q1+ q2+ . If
+ = 0 one can strip off and normalize this operator + + then apply Lemma 3 to the
pair of operators + , q1+ which symmetry operator is again trivial. As a result we prove
these operators to have the same order in the contradiction with our initial construction
unless the operator + = 0 and thereby q1+ = q2+ . The latter completes the proof of
Theorem 4.
As a corollary of Theorem 4 one finds that for the stripped-off operators q1+ and q2+ of
different order the symmetry operator q1+ q2 q2+ q1 = 0.
Two other consequences of Theorem 4 concern the structure of symmetry operators and
their uniqueness:
(a) any symmetric (self-transposed) symmetry operator B + = (B + )t , B + h+ = h+ B + is
a polynomial of the Hamiltonian, the latter is obtained by substituting a pair of the
operators B + and 1 (unity) instead of qi , i = 1, 2, in the formulation of Theorem 4;
(b) any two antisymmetric symmetry operators ei+ = (ei+ )t , ei+ h+ = h+ ei+ , i = 1, 2,
are dependent, i.e., being stripped off coincide, that follows from Theorem 4 after
substituting them instead of qi , i = 1, 2.
Now one can answer the question about a maximal number of independent supercharges
and their relative oddness. First let us prove that the number of supercharges cannot exceed
two. Assume that there exist three independent intertwining operators qi+ , i = 1, 2, 3.
Then their pairwise symmetry operators qi+ qj qj+ qi , i = j are antisymmetric and nontrivial but dependent in virtue of the consequence (b), qi+ qj qj+ qi = ij (h+ )e+ . When
multiplying these relations on kl one can assemble the following identity,

q1+ 13 h q2 12 h q3 q2+ 13 h q3+ 12 h q1
+
= 0 = q1+ q23
q23
q ,

1
+
+
+
q2 13 h q3 12 h q23
,
(90)
+
where the operator q23
is nontrivial due to independence of q2+ and q3+ . Evidently two
+
intertwining operators q1+ and q23
satisfy the requirements of Theorem 4 and therefore
are dependent in contradiction with the initial assumption. Thus we have proved that the
maximal number of independent supercharges is two.
46
Next let us consider two normalized independent operators qi+ , i = 1, 2, of orders N1

and N2 such that N1 > N2 and the sum of their order N1 + N2 is even. Then evidently the
operator
(N N )/2

q3+ = q1+ h+ 1 2 q2+
(91)
is independent of q2+ and has the order N3 less than N1 . If the sum of orders N2 + N3 is
again even one may normalize q3+ and apply the above algorithm to derive a lower order
independent SUSY generator until the sum of orders became odd. Thus one can always
construct the basis of two intertwining operators containing an even and an odd one.
Finally one can search for a set of minimal operators k , p just solving the system for
two independent intertwining operators,

qi = i h k + i h p , i = 1, 2;
(92)
with coefficients i (h ), i (h ) polynomial in the Hamiltonian.
Indeed,
(a) among all intertwining operators q + there exist an unique real operator of lowest order
p+ normalized according to (6);
(b) among all intertwining operators q + independent of p+ there exist a real operator of
lowest order k + normalized according to (6);
(c) with the help of the algorithm (91) one can prove that one of the operators p+ and k +
is of even order and another one is of odd order;
(d) an arbitrary intertwining operator q is always decomposed in the form (92) in the
unique way which can be proven by a consequent application of the algorithm (91)
and taking into account that the one of the terms k and p is of even order and
another one is of odd order.
q
Thus the set of p+ and k + form an optimal basis of intertwining operators. As all
= (q + )t the same results are translated to the set of p and k .
8. More about symmetry operators

In the previous section we have proven that the antisymmetric symmetry operator in
each component is unique after being stripped off. But the optimization of supercharge
basis may not guarantee the minimal form of components of the symmetry operator.
The uniqueness of decomposition (92) allows to formulate a necessary condition for the
symmetry operator e made of the minimal operators k , p to be stripped off further
on. Namely if a polynomial f (h ) can be factorized out of the symmetry operator e
N (h ), P
N1 (h ), b (h ). It follows
then the same polynomial appears as a multiplier in P
from the relations,

1
N h p ,
ie k = k p p k k = b h k P
2

1
N1 h k b h p .
ie p = k p p k p = P
(93)
2
47
One can give a more detailed description of each component of the symmetry operator
for a particular class of potentials with the help of the lemma.
Lemma 4. Assume that: (a) the Hamiltonian ha commutes with an antisymmetric real
operator Ra of order 2n + 1 which cannot be stripped off ; (b) this Hamiltonian has at
least one bound state; (c) the wave function 0 characterizes a bound state with the lowest
energy E0 .
Then there exist a nonsingular Hamiltonian hb , a nonsingular real operator ra =
(0 /0 ) and a nonsingular antisymmetric real operator Rb of order 2n 1 which
cannot be stripped off such that:
ha = rat ra + E0 ,
Ra = rat Rb ra ,
hb = ra rat + E0 ,
ra ha = hb ra ,
Rb hb = hb Rb .
(94)
(95)
The proof of relations (94) is standard for the SUSY QM [4]. The partial factorization
b ra with a nonsingular differential operator R
b of order 2n is provided by the
of Ra = R
b hb
b = R
equations Ra 0 = 0 and ra 0 = 0. Evidently it is an intertwining operator, ha R
due to Eqs. (94). As the Hamiltonian hb does not have the level E0 the latter relation entails
t 0 = 0. Hence the factorization takes place R
t = R t ra with a nonsingular differential
R
b
b
b
t
operator Rb of order 2n 1. From the intertwining relations it follows that the operator Rbt
is a symmetry operator. At last its antisymmetry under transposition can be easily derived
from the similar property of Ra . The operator Rb cannot be stripped off if Eqs. (95) hold
and the operator Ra has been stripped off already.
From Lemma 4 one can obtain a certain relationship between the number of bound
states of the Hamiltonian and the structure of the symmetry operator. Namely, suppose that
the Hamiltonian h0 has n bound states with energies El , El+1 > El and commutes with
a antisymmetric real operator R0 of order 2n + 1 which cannot be stripped off. Then the
(normalized) symmetry operator can be factorized,
t
R0 = r0t rn1
rn1 r0 ,
rl + l ,
(96)
with nonsingular real superpotentials l . Respectively the ladder (dressing chain) relations
hold,
hl+1 rl = rl hl ,
l = 0, . . . , n 1,
t
hl rl1 rl1
+ El1 = rlt rl + El ,
h0 = r0t r0
+ E0 ,
t
hn = rn1 rn1
l = 1, . . . , n 1,
+ En1 ,
(97)
and the hidden symmetry operators arise for each intermediate Hamiltonian,
t
rn1 rl ,
Rl = rlt rn1
Rl hl = hl Rl ,
l = 0, . . . , n.
Rn = ,
(98)
Evidently the Hamiltonian hn describes a free particle and therefore the Hamiltonian
system with a hidden symmetry can be related to the free-particle system.
48
In the case (b) of Section 4 each component of the symmetry operator can be also
represented in the canonical factorized form (96),

1,2
1,2
3
3
e = i 3 a + V1,2 V1,2
(99)
= i

,
2
4
1,2
1,2

with the help of the bound-state wave functions, 1,2 = c V1,2 2 . In the case (c) the
potentials (58) exemplify Lemma 4 as one of them is constant.
One can guess that the above relationship between the Hamiltonian h0 and the symmetry
operator R0 is quite general because the algorithm of Lemma 4 helps to transform the
system with n bound states and with a symmetry operator to a system with n 1 bound
states and a symmetry operator of order lower in two units. After one removes all bound
states with this algorithm the remaining Hamiltonian is still reflectionless and thereby the
scattering coefficients are trivial corresponding to a free-particle system.
9. Concluding remarks
We have established that:
(a) for supercharges of finite order the polynomial SUSY can be always realized when the
supercharges are related by transposition;
(b) in certain cases (for instance, for complex intertwining operators) several supercharges
may commute with the super-Hamiltonian which may yield a nontrivial hidden
symmetry of such a system;
commuting
(c) the maximal number of independent supercharges of type Q (or of type Q)
with a given super-Hamiltonian is two and for an extended superalgebra there exists an
optimal set of two real supercharges with components of a minimal order in derivatives
one of which is an odd-order operator and another one is an even-order operator;
(d) the hidden symmetry operator E(H ) defined in (43) is unique up to a multiplier
polynomial in the super-Hamiltonian and among zero-modes of this operator there
are all bound states of the super-Hamiltonian.
Finally we mention possible extensions of the theorems and results of this paper. First
of all it seems to be straightforward to apply them to the Hamiltonians with complex
potentials [54] as the SUSY algebra with transposition symmetry is well defined for such
Hamiltonians. The application to matrix Hamiltonians [18] is less trivial but certainly
interesting as well as a generalization on multidimensional QM [4,17].
Acknowledgements
One of us (A.A.) is grateful to F. Cannata and J.-P. Dedonder for useful discussions.
This work was supported by the Grant RFBR 02-01-00499.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
H. Nicolai, J. Phys. A: Math. Gen. 9 (1976) 1497.

E. Witten, Nucl. Phys. B 188 (1981) 513.
F. Cooper, B. Freedman, Ann. Phys. (N.Y.) 146 (1983) 262.
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, JETP Lett. 39 (1984) 93;
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, Phys. Lett. A 105 (1984) 19;
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, Theor. Math. Phys. 61 (1985) 1078.
M.M. Nieto, Phys. Lett. B 145 (1984) 208.
B. Mielnik, J. Math. Phys. 25 (1984) 3387.
D. Fernndez, Lett. Math. Phys. 8 (1984) 337.
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, M.I. Eides, Phys. Lett. A 109 (1985) 143;
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, M.I. Eides, Theor. Math. Phys. 61 (1985) 965.
C.V. Sukumar, J. Phys. A: Math. Gen. 18 (1985) L57;
C.V. Sukumar, J. Phys. A: Math. Gen. 18 (1985) 2917;
C.V. Sukumar, J. Phys. A: Math. Gen. 18 (1985) 2937.
L.E. Gendenshtein, I.V. Krive, Sov. Phys. Usp. 28 (1985) 645.
A. Lahiri, P.K. Roy, B. Bagchi, Int. J. Mod. Phys. A 5 (1990) 1383.
F. Cooper, A. Khare, U. Sukhatme, Phys. Rep. 251 (1995) 267.
G. Junker, Supersymmetric Methods in Quantum and Statistical Physics, Springer, Berlin, 1996.
R. de Lima Rodrigues, hep-th/0205017.
D. Baye, Phys. Rev. Lett. 58 (1987) 2738.
R.D. Amado, F. Cannata, J.-P. Dedonder, Phys. Rev. C 41 (1990) 1289;
R.D. Amado, F. Cannata, J.-P. Dedonder, Phys. Rev. C 43 (1991) 2077;
R.D. Amado, F. Cannata, J.-P. Dedonder, Int. J. Mod. Phys. A 5 (1990) 3401.
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, Phys. Lett. A 201 (1995) 103;
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, Theor. Math. Phys. 104 (1995) 1129;
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, solv-int/9605007;
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. 32 (1999) 4641.
A.A. Andrianov, F. Cannata, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. 30 (1997) 5037.
F. Cannata, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. 35 (2002) 1389.
A.V. Turbiner, Commun. Math. Phys. 118 (1988) 467.
M.A. Shifman, Int. J. Mod. Phys. A 12 (1989) 2897.
C.M. Bender, G.V. Dunne, J. Math. Phys. 37 (1996) 6.
H. Aoyama, H. Kikuchi, I. Okouchi, M. Sato, S. Wada, Nucl. Phys. B 553 (1999) 644.
B. Bagchi, F. Cannata, C. Quesne, Phys. Lett. A 269 (2000) 79.
R. Sasaki, K. Takasaki, J. Phys. A: Math. Gen. 34 (2001) 9533.
S.M. Klishevich, M.S. Plyushchay, Nucl. Phys. B 606 (2001) 583;
S.M. Klishevich, M.S. Plyushchay, Nucl. Phys. B 616 (2001) 403.
P. Dorey, C. Dunning, R. Tateo, J. Phys. A: Math. Gen. 34 (2001) 5679;
P. Dorey, C. Dunning, R. Tateo, J. Phys. A: Math. Gen. 34 (2001) L391.
A.A. Andrianov, M.V. Ioffe, V.P. Spiridonov, Phys. Lett. A 174 (1993) 273.
A.A. Andrianov, F. Cannata, J.-P. Dedonder, M.V. Ioffe, Int. J. Mod. Phys. A 10 (1995) 2683.
V.B. Matveev, M.A. Salle, Darboux Transformations and Solitons, Springer, Berlin, 1991.
V.G. Bagrov, B.F. Samsonov, Theor. Math. Phys. 104 (1995) 1051.
B.F. Samsonov, Mod. Phys. Lett. A 11 (1996) 1563.
A. Gangopadhyaya, U. Sukhatme, Phys. Lett. A 224 (1996) 5.
U. Sukhatme, C. Rasinariu, A. Khare, Phys. Lett. A 234 (1997) 401.
A. Das, S.A. Pernice, Mod. Phys. Lett. A 12 (1997) 581.
D.J. Fernndez C., Int. J. Mod. Phys. A 12 (1997) 171.
G. Junker, P. Roy, Ann. Phys. 270 (1998) 155.
D.J. Fernndez C., V. Hussin, B. Mielnik, Phys. Lett. A 244 (1998) 309;
D.J. Fernndez C., V. Hussin, quant-ph/0011004.
J.O. Rosas-Ortiz, J. Phys. A 31 (1998) 10163.
49
50
[40] B. Bagchi, A. Ganguly, D. Bhaumik, A. Mitra, Mod. Phys. Lett. A 14 (1999) 27.
[41] D.J. Fernndez C., J. Negro, L.M. Nieto, Phys. Lett. A 275 (2000) 338;
D.J. Fernndez C., R. Muos, A. Ramos, quant-ph/0212026.
[42] M.S. Plyushchay, Int. J. Mod. Phys. A 15 (2000) 3679;
M.S. Plyushchay, Phys. Lett. B 485 (2002) 187.
[43] H. Aoyama, M. Sato, T. Tanaka, M. Yamamoto, Phys. Lett. B 498 (2001) 117;
H. Aoyama, N. Nakayama, M. Sato, T. Tanaka, Phys. Lett. B 519 (2001) 260;
H. Aoyama, N. Nakayama, M. Sato, T. Tanaka, Phys. Lett. B 521 (2001) 400.
[44] H. Aoyama, M. Sato, T. Tanaka, Phys. Lett. B 503 (2001) 423.
[45] H. Aoyama, M. Sato, T. Tanaka, Nucl. Phys. B 619 (2001) 105;
T. Tanaka, hep-th/0212276.
[46] S.M. Klishevich, M.S. Plyushchay, Nucl. Phys. B 628 (2002) 217.
[47] A.P. Veselov, A.B. Shabat, Funct. Anal. Appl. 27 (1993) 81.
[48] V.E. Adler, Funct. Anal. Appl. 27 (1993) 140.
[49] R. Zhdanov, J. Math. Phys. 37 (1996) 3198.
[50] W. Fushchych, A. Nikitin, J. Math. Phys. 38 (1997) 5944.
[51] H.-D. Doebner, R. Zhdanov, math-ph/9809021.
[52] A.A. Andrianov, F. Cannata, M.V. Ioffe, D.N. Nishnianidze, quant-ph/9902057;
A.A. Andrianov, F. Cannata, M.V. Ioffe, D.N. Nishnianidze, Phys. Lett. A 266 (2000) 341.
[53] S. Flgge, Practical Quantum Mechanics, Springer, Berlin, 1971;
B.N. Zakhariev, A.A. Suzko, Direct and Inverse Methods, Springer, Berlin, 1990.
[54] A.A. Andrianov, F. Cannata, J.-P. Dedonder, M.V. Ioffe, Int. J. Mod. Phys. A 14 (1999) 2675.

Thermalization of fermionic quantum fields

Jrgen Berges a , Szabolcs Borsnyi a,b , Julien Serreau a
a Universitt Heidelberg, Institut fr Theoretische Physik, Philosophenweg 16, 69120 Heidelberg, Germany
b Etvs University, Department of Atomic Physics, H-1117 Budapest, Hungary
Received 2 January 2003; received in revised form 21 March 2003; accepted 26 March 2003
Abstract
We solve the nonequilibrium dynamics of a (3 + 1)-dimensional theory with Dirac fermions
coupled to scalars via a chirally invariant Yukawa interaction. The results are obtained from a
systematic coupling expansion of the 2PI effective action to lowest nontrivial order, which includes
scattering as well as memory and off-shell effects. The dynamics is solved numerically without
further approximation, for different far-from-equilibrium initial conditions. The late-time behavior is
demonstrated to be insensitive to the details of the initial conditions and to be uniquely determined by
the initial energy density. Moreover, we show that at late time the system is very well characterized
by a thermal ensemble. In particular, we are able to observe the emergence of FermiDirac and
BoseEinstein distributions from the nonequilibrium dynamics.
1. Introduction
The abundance of experimental data on matter in extreme conditions from relativistic
heavy-ion collision experiments, as well as applications in astrophysics and cosmology has
lead to a strong increase of interest in the dynamics of quantum fields out of equilibrium.
Experimental indications of thermalization in collision experiments and the justification of
current predictions based on equilibrium thermodynamics, local equilibrium or (non)linear
response pose major open questions for our theoretical understanding [1,2]. One way
to resolve these questions is to try to understand quantitatively the far-from-equilibrium
dynamics of quantum fields, without relying on the assumption of small departures from
equilibrium, or on a possible separation of scales that forms the basis of effective kinetic
E-mail addresses: j.berges@thphys.uni-heidelberg.de (J. Berges), mazsx@cleopatra.elte.hu (S. Borsnyi),
serreau@thphys.uni-heidelberg.de (J. Serreau).
doi:10.1016/S0550-3213(03)00261-X
52
J. Berges et al. / Nuclear Physics B 660 (2003) 5180
descriptions [3]. In contrast to close-to-equilibrium approaches, the quantum-statistical

fluctuations of the fields are not assumed to be described by a thermal ensemble. Moreover,
one goes beyond the range of applicability of the usual gradient expansion and dilute-gas
approximation.
In contrast to thermal equilibrium, which keeps no information about the past,
nonequilibrium dynamics poses an initial value problem: time-translation invariance
is explicitly broken by the presence of the initial time, where the system has been
prepared. The question of thermalization investigates how the system effectively looses
the dependence on the details of the initial condition, and becomes approximately
time-translation invariant at late times. According to basic principles of equilibrium
thermodynamics, the thermal solution is universal in the sense that it is independent
of the details of the initial condition and is uniquely determined by the values of the
(conserved) energy density and of possible conserved charges.1 There are two distinct
classes of universal behavior, corresponding to BoseEinstein and FermiDirac statistics,
respectively.
The description of the effective loss of initial conditions and subsequent approach
to thermal equilibrium in quantum field theory requires calculations beyond so-called
Gaussian (leading-order large-N , Hartree or mean-field type) approximations [46].
Similar to the free-field theory limit, these approximations typically exhibit an infinite
number of additional conserved quantities which are not present in the underlying
interacting theory [7,8]. These spurious constants of motion constrain the time evolution
and lead to a nonuniversal late-time behavior [9]. It has recently been demonstrated in
the context of scalar field theories, that the approach to quantum thermal equilibrium can
be described by going beyond these approximations [9,10]. In particular, this has been
achieved by using a systematic coupling expansion [10] or a 1/N expansion to nextto-leading-order [9,11] of the two-particle irreducible generating functional for Greens
functions, the so-called 2PI effective action [1215]. In this context, the corresponding
equations of motion have been shown to lead to a universal late time behavior, in the sense
mentioned above, without to have recourse to any kind of coarse-graining.
In this work, we study the far-from-equilibrium time evolution of relativistic fermionic
fields and their subsequent approach to thermal equilibrium. Nonequilibrium behavior of
fermionic fields has been previously addressed in several scenarios [5,6,1619], including
the full dynamical problem with fermions coupled to inhomogeneous classical bosonic
fields [20]. Here we go beyond these approximations and compute the nonequilibrium
evolution in a (3 + 1)-dimensional quantum field theory of Dirac fermions coupled to
scalars in a chirally invariant way. As a consequence, we are able to study the approach
to quantum thermal equilibrium. For the considered nonequilibrium initial conditions we
find that the late-time behavior is universal and characterized by FermiDirac and Bose
Einstein distributions, respectively. The results are obtained from a systematic coupling
expansion of the 2PI effective action to lowest nontrivial (two-loop) order, which includes
scattering as well as memory and off-shell effects. The nonequilibrium dynamics is solved
1 Here we consider closed systems without coupling to a heat bath or external fields, which could provide
sources or sinks of energy.
53
numerically without further approximations. We emphasize that, given the limitations of a

weak-coupling expansion, this is a first-principle calculation with no other input than the
dynamics dictated by the considered quantum field theory for given initial conditions.
In Section 2, we review the 2PI effective action for fermions, which we use to derive
exact time evolution equations for the spectral function and the statistical two-point
function in Section 3. We discuss the Lorentz structure of our equations in Section 4 and
exploit some symmetries in Section 5. In Section 6, we specify to a chiral quark model
for which we solve the nonequilibrium dynamics. The initial conditions are discussed
in Section 7 and some details concerning the numerical implementation are given in
Section 8. The numerical results are presented and discussed in Sections 9 and 10. We
attach two appendices discussing in detail the quasiparticle picture we used to interprete
some of our results.
2. 2PI effective action for fermions

We consider first a purely fermionic quantum field theory with classical action

) ,
S = d4 x i (x)[i/
mf ]i (x) + V (,
(2.1)
for i = 1, . . . , Nf flavors of Dirac fermions i , a mass parameter mf and an interaction

) to be specified below. Here / , with Dirac matrices ( =
term V (,
0, . . . , 3). Summation over repeated indices and contraction in Dirac space is implied. All
correlation functions of the quantum theory can be obtained from the corresponding twoparticle-irreducible (2PI) effective action . For the relevant case of a vanishing fermionic
background field the 2PI effective action can be written as [13]
[D] = i Tr ln D 1 i Tr D01 D + 2 [D] + const.
(2.2)
The exact expression for the functional 2 [D] contains all 2PI diagrams with vertices
) and propagator lines associated to the full connected two-point
described by V (,
function D. In coordinate space the trace Tr includes an integration over a closed time path
C along the real axis [21], as well as integration over spatial coordinates and summation
over flavor and Dirac indices. The free inverse propagator is given by
1
iD0,ij
(x, y) = (i/
mf )C4 (x y)ij .
(2.3)
The equation of motion for D in absence of external sources is obtained by extremizing

the effective action [13]
[D]
= 0.
Dij (x, y)
(2.4)
According to (2.2) one can write (2.4) as an equation for the exact inverse propagator
1
(x, y) ij (x, y; D),
Dij1 (x, y) = D0,ij
(2.5)
54
with the proper self-energy

ij (x, y; D) i
2 [D]
.
Dj i (y, x)
(2.6)
Eq. (2.5) can be rewritten in a form, which is suitable for initial value problems
by convoluting it with D. One obtains the following time evolution equation for the
propagator:

(i/
x mf )Dij (x, y) i ik (x, z; D)Dkj (z, y) = iC4 (x y)ij ,
(2.7)
z

where we employed the shorthand notation z = C dz0 d3 z. Note that to keep the
notation clear, we omit Dirac indices and reserve the Latin indices i, j, k, . . . to denote
flavor.
3. Exact evolution equations for the spectral and statistical components of the
two-point function
To simplify physical interpretation we rewrite (2.7) in terms of equivalent equations
for the spectral function, which contains the information about the spectrum of the theory,
and the statistical two-point function. The latter will, in particular, provide an effective
description of occupation numbers. For this we write for the time-ordered two-point
function Dij (x, y) and proper self-energy ij (x, y; D):2

Dij (x, y) = C x 0 y 0 Dij> (x, y) C y 0 x 0 Dij< (x, y),
(3.1)
0

ij (x, y; D) = C x y 0 ij> (x, y) C y 0 x 0 ij< (x, y).
(3.2)
Note that for convenience we omit the explicit D-dependence in the notation for ij>,< .
Inserting the above decompositions into the evolution equation (2.7), we obtain evolution
equations for the functions Dij> (x, y) and Dij< (x, y) as well as the identity

0 Dij> (x, y) + Dij< (x, y) x 0 =y 0 = (
(3.3)
x y )ij ,
which corresponds to the anticommutation relation for fermionic field operators. For later
use we also note the hermiticity property

>
Dj i (y, x) = 0 Dij> (x, y) 0,
(3.4)
and equivalently for Dij< (x, y). Note that here the Hermitean conjugation denotes complex
conjugation and taking the transpose in Dirac space only.
2 If there is a local contribution to the proper self-energy, we write
(nonlocal)
ij (x, y; D) = i (local) (x; D)C4 (x y)ij + ij
(x, y; D),
and the decomposition (3.2) is taken for (nonlocal) (x, y; D). In this case the local contribution gives rise to an
effective spacetime dependent fermion mass term mf + (local) (x; D).
55
Similar to the discussion for scalar fields in Refs. [9,22] we introduce the spectral
function ij (x, y) and the statistical propagator Fij (x, y) defined as3

ij (x, y) = i Dij> (x, y) + Dij< (x, y) ,
(3.5)

1
Fij (x, y) = Dij> (x, y) Dij< (x, y) .
(3.6)
2
The corresponding components of the self-energy are given by4

Aij (x, y) = i ij> (x, y) + ij< (x, y) ,
(3.7)

1
Cij (x, y) = ij> (x, y) ij< (x, y) .
(3.8)
2
With (3.4) the two-point functions have the properties

j i (y, x) = 0 ij (x, y) 0,
(3.9)

Fj i (y, x) = 0 Fij (x, y) 0 ,
(3.10)
and equivalently for Aij (x, y) and Cij (x, y).
Using the above notations the evolution equation (2.7) written for ij (x, y) and
Fij (x, y) are given by
x 0
(i/
x mf )ij (x, y) =
dz Aik (x, z)kj (z, y),
(3.11)
y0
x 0
(i/
x mf )Fij (x, y) =
y
dz Aik (x, z)Fkj (z, y)
dz Cik (x, z)kj (z, y),
(3.12)

x0
where we have taken the initial time to be zero, and y 0 dz y 0 dz0 d3 z. For known selfenergies Eqs. (3.11) and (3.12) are exact. We note that the form of their r.h.s. is identical
to the one for scalar fields [9,22]. To solve the evolution equations one has to specify
initial conditions for the two-point functions, which is equivalent to specifying a Gaussian
initial density matrix.5 We note that the fermion anticommutation relation or (3.3) uniquely
specifies the initial condition for the spectral function:

x y )ij .
0 ij (x, y)x 0 =y 0 = i(
(3.13)
x0
3 Equivalently, one can decompose

i
Dij (x, y) = Fij (x, y) ij (x, y) C x 0 y 0 C y 0 x 0 .
2
4 Besides the dynamical field degrees of freedom D we introduce quantities which are functions of these
ij
fields. These functions are denoted by either boldface or Greek letters.
5 We emphasize that a Gaussian initial density matrix only restricts the initial conditions or the experimental
setup and represents no approximation for the time evolution. More general initial conditions can be discussed
using additional source terms in defining the generating functional for Greens functions.
56
Suitable nonequilibrium initial conditions for the statistical propagator Fij (x, y) will be
discussed below.
4. Lorentz decomposition
It is very useful to decompose the fields ij (x, y) and Fij (x, y) into terms that have
definite transformation properties under Lorentz transformation. We will see below that,
depending on the symmetry properties of the initial state and interaction, a number of
these terms remain identically zero under the exact time evolution, which can dramatically
simplify the analysis. Using a standard basis and suppressing flavor indices we write
1
= S + i5 P + V + 5 A + T ,
2
(4.1)
where = 2i [ , ] and 5 = i 0 1 2 3 . For given flavor indices the 16 (pseudo-)scalar,

(pseudo-)vector and tensor components
S = tr ,
P = i tr 5 ,
V = tr ,
A = tr 5 ,
T = tr ,
(4.2)
are complex two-point functions. Here we have defined tr 14 tr where the trace acts in
Dirac space. Equivalently, there are 16 complex components for Fij , Aij and Cij for given
flavor indices i, j . Using (3.9) and (3.10), one sees that they obey
( )

( )

( )
( )
Fij (x, y) = Fj i (y, x) ,
ij (x, y) = j i (y, x) ,
(4.3)
where = {S, P , V , A, T }. Inserting the above decomposition into the evolution equations (3.11) and (3.12) one obtains the respective equations for the various components
displayed in Eq. (4.2).
For a more detailed discussion, we first consider the l.h.s. of the evolution equations
(3.11) and (3.12). In fact, the approximation of a vanishing r.h.s. corresponds to the
standard mean-field or Hartree-type approaches frequently discussed in the literature
[4,5]. However, to discuss thermalization we have to go beyond such a Gaussian
approximation: it is crucial to include direct scattering which is described by the
nonvanishing contributions from the r.h.s. of the evolution equations. Starting with the
l.h.s. of (3.11) one finds, omitting flavor indices (see also Ref. [16]):

tr (i/
mf ) = i V mf S ,

mf ) = i i A mf P ,
i tr 5 (i/

mf ) = i S + i i T mf V ,
tr (i/

1

mf ) = i i P + # (i T , ) mf A ,
tr 5 (i/
2

tr (i/
(4.4)
mf ) = i i V i V + # (i A, ) mf T .
The corresponding expressions for the l.h.s. of the evolution equation (3.12) for F follow
from (4.4) with the replacement F . Considering now the various component (4.2) of
57
the integrand on the r.h.s. of Eq. (3.11), we find

1
tr[A]
= AS S AP P + AV V , AA A, + AT T , ,
2
1
5 A] = AS P + AP S iA
AT , T , ,
i tr[
V A, + iAA V , + #
4

tr A = AS V + AV S iAP A + iAA P + iAV , T + iAT V ,

1
+ # (AA, T , + AT , A, ),
2

tr 5 A = AS A + AA S iAP V + iAV P + iAA, T + iAT A,

1
+ # (AV , T , + AT , V , ),
2

1
tr A = AS T + AT S # (AP T , + AT , P )
2

i AV V AV V + # (AV , A, AA, V , )

+ i AA A AA A + i AT T , AT T , .
(4.5)
(4.6)
(4.7)
(4.8)
(4.9)
With the above expressions one obtains the evolution equations for the various Lorentz
components in a straightforward way using (3.11). We note that the convolutions appearing
on the r.h.s. of the evolution equation (3.12) for F are of the same form than those
computed above for . The respective r.h.s. can be read off Eqs. (4.5)(4.9) by replacing
F for the first term and A C for the second term under the integrals of Eq. (3.12).
We have now all the relevant building blocks to discuss the most general case of
nonequilibrium fermionic fields. However, this is often not necessary in practice due to
the presence of symmetries, which require certain components to vanish identically.
5. Symmetries
In the following, we will exploit symmetries of the action (2.1) and of the initial
conditions in order to simplify the fermionic evolution equations derived in the previous
section.
Spatial translation invariance and isotropy. We will consider spatially homogeneous and
isotropic initial conditions. In this case it is convenient to work in Fourier space and we
write

0 0
d3 p i p(
e x y ) x 0 , y 0 ; p ,
(x, y) x , y ; x y =
(5.1)
(2)3
and similarly for the other two-point functions. Moreover, isotropy implies a reduction of
the number of independent two-point functions: e.g., the vector components of the spectral
function can be written as

V0 x 0 , y 0 ; p = V0 x 0 , y 0 ; p ,

V x 0 , y 0 ; p = v V x 0 , y 0 ; p ,
58
where p |p|
and v = p/p.

Parity. The vector components V0 (x 0 , y 0 ; p) and V (x 0 , y 0 ; p) are unchanged under
a parity transformation, whereas the corresponding axial-vector components get a minus
sign. Therefore, parity together with rotational invariance imply that

A0 x 0 , y 0 ; p = A x 0 , y 0 ; p = 0.
(5.2)
The same is true for the axial-vector components of F , A and C. Parity also implies the
pseudo-scalar components of the various two-point functions to vanish.
CP -invariance. For instance, under combined charge conjugation and parity transformation the vector component of transforms as

V0 x 0 , y 0 ; p V0 y 0 , x 0 ; p ,

V x 0 , y 0 ; p V y 0 , x 0 ; p ,
and similarly for A0V and AV . The F -components transform as

FV0 x 0 , y 0 ; p FV0 y 0 , x 0 ; p ,

FV x 0 , y 0 ; p FV y 0 , x 0 ; p ,
and similarly for C0V and CV . Combining this with the hermiticity relations (4.3), one
obtains for these components that

Re V0 x 0 , y 0 ; p = Im V x 0 , y 0 ; p = 0,

Re FV0 x 0 , y 0 ; p = Im FV x 0 , y 0 ; p = 0,

Re A0V x 0 , y 0 ; p = Im AV x 0 , y 0 ; p = 0,

Re C0V x 0 , y 0 ; p = Im CV x 0 , y 0 ; p = 0,
(5.3)
for all times x 0 and y 0 and all individual momentum modes.
We stress that a nonequilibrium ensemble respecting a particular symmetry does not
imply that the individual ensemble members exhibit the same symmetry. For instance,
a spatially homogeneous ensemble can be build out of inhomogeneous ensemble members
and clearly includes the associated physics. The most convenient choice of an ensemble is
mainly dictated by the physical problem to be investigated. Below we will study a chiral
quark model with Dirac fermions coupled to scalars in a chirally invariant way. Since in
this paper we will restrict the discussion of this model to the phase without spontaneous
breaking of chiral symmetry, it is useful to exploit this symmetry as well.
Chiral symmetry. The only components of the decomposition (4.1) allowed by chiral
symmetry are those which anticommute with 5 . We therefore have

S x 0 , y 0 ; p = P x 0 , y 0 ; p = T x 0 , y 0 ; p = 0,
(5.4)
and similarly for the corresponding components of F , A and C. In particular, chiral
symmetry forbids a mass term for fermions and we have mf 0.
59
5.1. Equations of motion

In conclusion, for the above symmetry properties we are left with only four independent
propagators: the two spectral functions V0 and V and the two corresponding statistical
functions FV0 and FV . They are either purely real or imaginary and have definite symmetry
properties under the exchange of their time arguments x 0 y 0 . These properties as well
as the corresponding ones for the various components of the self-energy are summarized
below:
V0 , A0V :
imaginary, symmetric;
V , AV :
real, antisymmetric;
FV0 , C0V :
imaginary, antisymmetric;
FV , CV : real, symmetric.
The exact evolution equations for the spectral functions read (cf. Eq. (3.11)):6
i
0 0
0 0 0
x
x ,y ;p
,
y
;
p
=
p
V
V
x 0
x 0
+

dz0 A0V x 0 , z0 ; p V0 z0 , y 0 ; p AV x 0 , z0 ; p V z0 , y 0 ; p ,
(5.5)
y0

V x 0 , y 0 ; p = pV0 x 0 , y 0 ; p
0
x
x 0
+

dz0 A0V x 0 , z0 ; p V z0 , y 0 ; p AV x 0 , z0 ; p V0 z0 , y 0 ; p .
(5.6)
y0
Similarly, for the statistical two-point functions we obtain (cf. Eq. (3.12)):
i

0 0 0
FV x , y ; p = pFV x 0 , y 0 ; p
0
x
x 0
+

dz0 A0V x 0 , z0 ; p FV0 z0 , y 0 ; p AV x 0 , z0 ; p FV z0 , y 0 ; p
y

dz0 C0V x 0 , z0 ; p V0 z0 , y 0 ; p CV x 0 , z0 ; p V z0 , y 0 ; p ,
(5.7)
6 We note that the following equations do not rely on the restrictions (5.3) imposed by CP -invariance: they
have the very same form for the case that all two-point functions are complex.
60

FV x 0 , y 0 ; p = pFV0 x 0 , y 0 ; p
0
x
x 0

+ dz0 A0V x 0 , z0 ; p FV z0 , y 0 ; p AV x 0 , z0 ; p FV0 z0 , y 0 ; p
0
y

dz0 C0V x 0 , z0 ; p V z0 , y 0 ; p CV x 0 , z0 ; p V0 z0 , y 0 ; p .
(5.8)
The above equations are employed below to calculate the nonequilibrium fermion
dynamics in a chiral quarkmeson model.
6. Chiral quarkmeson model

As an application we consider a quantum field theory involving two fermion flavors
(quarks) coupled in a chirally invariant way to a scalar -field and a triplet of
pseudoscalar pions a (a = 1, 2, 3). The classical action reads

1
+ + a a
S = d4 x i/
2

+ g + i5 a a V 2 + 2 ,
(6.1)
where 2 a a and where a denote the standard Pauli matrices. The above action
is invariant under chiral SU L (2) SU R (2) transformations. For a quartic scalar selfinteraction ( 2 + a a )2 , this model corresponds to the well known linear -model
[28], which has been extensively studied in thermal equilibrium in the literature using
various approximations [29]. For simplicity we consider here a purely quadratic scalar
potential:

1
V = m20 2 + 2 ,
(6.2)
2
which is sufficient to study thermalization in this model. We note that this theory has
the same universal properties than the corresponding linear -model. Extending our
study to take into account quartic self-interaction is straightforward. It gives additional
contributions to the scalar self-energies, Eqs. (6.13) and (6.14) below. These contributions
can be found in Refs. [9,11] and we will point out the respective changes below.
6.1. Equations of motion for the scalar field
The 2PI effective action for nonequilibrium scalar fields has been extensively studied
in the literature [911,14,15]. Here, we briefly recall the main features of the scalar
sector and stress those aspects which are relevant for the present paper (for details see,
e.g., Refs. [9,11]). For the model considered here, Eq. (6.1), the 2PI effective action is
61
a functional of fermion as well as scalar propagators.7 The scalar fields form an O(4)vector A (x) ( (x), (x))

and we denote the full scalar propagator by GAB (x, y) with
A, B = 0, . . . , 3. The 2PI effective action (2.2), augmented by the scalar sector, reads
[G, D] =
i
i
Tr ln G 1 + Tr G01 G i Tr ln D 1 i Tr D01 D
2
2
+ 2 [G, D] + const,
with the free scalar inverse propagator

1
iG0,AB
(x, y) = x + m20 C4 (x y)AB .
(6.3)
(6.4)
Similar to the fermionic case discussed above, the equation of motion for the scalar twopoint function is obtained by minimizing the 2PI effective action with respect to G, and
the proper self-energy is given by [9,11,13]
AB (x, y) = 2i
2 [G, D]
.
GBA (y, x)
(6.5)
Under chiral transformations, the matrix G RGR , where R is an O(4) rotation.

Without loss of generality, because of chiral symmetry the effective action [G, D] and its
functional derivatives can be evaluated for G taken to be the unit matrix in O(4)-space:
GAB (x, y) = G (x, y)AB .
(6.6)
The same holds for the corresponding self-energy (6.5). Similarly, in the fermionic sector,
because of chiral symmetry the most general fermion two-point function can be taken to
be proportional to unity in flavor space:
Dij (x, y) = D(x, y)ij .
(6.7)
Similar to the discussion in Section 3, the scalar spectral and statistical two-point functions
are defined as [9,22]

i
G (x, y) = F (x, y) (x, y) C x 0 y 0 C y 0 x 0 ,
2
(6.8)
and equivalently for the spectral and statistical self-energies and F . These are all real
functions and F -like components are symmetric under the exchange of x and y, whereas
the -like components are antisymmetric. The equal-time commutation relation of two
scalar field operators implies [9,22]

(x, y)x 0 =y 0 = 0,
(6.9)
x 0 (x, y)x 0 =y 0 = x y ,
which uniquely specifies the initial conditions for the spectral function. Initial conditions
for the statistical two-point function will be discussed below. Finally, the equations of
7 As emphasized in the previous section, we do not consider the possibility of a broken symmetry in this paper.
Therefore, we restrict the discussion to a vanishing scalar field expectation value.
62
Fig. 1. Two-loop contribution to 2 [G, D]. The solid and dashed lines represent the full fermion (D) and boson
(G) propagators, respectively.
motion for the scalar propagators read [9,22]:

2
x 0
x20 + p 2 + m0 x 0 , y 0 ; p =

dz0 x 0 , z0 ; p z0 , y 0 ; p ,
(6.10)
y0

2
x 0
x20 + p 2 + m0 F x 0 , y 0 ; p =

dz0 x 0 , z0 ; p F z0 , y 0 ; p
y
+

dz0 F x 0 , z0 ; p z0 , y 0 ; p ,
(6.11)
where we have assumed spatially homogeneous initial conditions and Fourier transformed
with respect to spatial coordinates. For known self-energies these are the exact equations
for the theory described by the classical action (6.1) together with (6.2). In the presence
of scalar self-interactions, the self-energy receives in particular a local contribution
i(local)(x)C4 (x y), which amounts to a shift of the bare mass squared appearing in the
above equations. In that case, the exact equations of motion have the same form as above,
with the replacement m20 M 2 (x) = m20 + (local) (x) [9,11].
6.2. 2PI coupling expansion
We consider a systematic coupling expansion of the 2PI effective action (6.3) to lowest
nontrivial order, which includes scattering as well as memory and off-shell effects, such
that thermalization can be studied. The first nontrivial order in a coupling expansion
corresponds to the two-loop contribution depicted in Fig. 1 (cf. also the discussion in
Ref. [24]).8 Using (6.6) and (6.7), one can express the two-loop contribution to 2 directly
in terms of G and D. We obtain for the chirally symmetric theory:

Nf Ns
(2-loop)
2
(6.12)
[G, D] = ig 2
d4 x d4 y tr D(x, y)D(y, x) G (x, y),
2
C
8 We note that this approximation can also be related to a nonperturbative 1/N expansion at next-to-leading
f
order.
63
where Nf = 2 is the number of fermion flavors and Ns = 4 is the number of scalar

components. From there, it is straightforward to compute the spectral and statistical
components of the two-loop self-energies.9 We obtain for the scalar self-energies

d3 q 0 0

V x , y ; q FV , x 0 , y 0 ; p q ,
x 0 , y 0 ; p = 8g 2 Nf
(6.13)
3
(2)

d3 q

F x 0 , y 0 ; p = 4g 2 Nf
FV x 0 , y 0 ; q FV , x 0 , y 0 ; p q
3
(2)

1
V x 0 , y 0 ; q V , x 0 , y 0 ; p q ,
4
(6.14)
and for the fermion self-energies

d3 q
0 0

2
F x 0 , y 0 ; q x 0 , y 0 ; p q
AV x , y ; p = g Ns
(2)3 V

0 0

0 0
+ V x , y ; q F x , y ; p q ,
(6.15)

CV x 0 , y 0 ; p = g 2 Ns

d3 q

FV x 0 , y 0 ; q F x 0 , y 0 ; p q
3
(2)

1 0 0 0 0
V x , y ; q x , y ; p q .
4
(6.16)
Finally, we note that for the case of a nonvanishing scalar self-interaction, the only
additional two-loop contribution gives rise to a local mass shift as described above. In
particular, at this order there is no additional contribution to the direct scattering part, i.e.,
to the r.h.s. of Eqs. (6.10) and (6.11) relevant for thermalization.
7. Initial conditions
The time evolution for the fermions is described by first-order (integro-)differential
equations for F and : Eqs. (5.5)(5.8). As pointed out above, the initial fermion spectral
function is completely determined by the equal-time anticommutation relation of fermionic
field operators (cf. Eq. (3.13)). To uniquely specify the time evolution for F we have to
set the initial conditions. The most general (Gaussian) initial conditions for F respecting
9 The relevant self-energies can be obtained with
2i
2
2 GAB
= 2i
= AB BA = Ns ,
G
GAB G
and similarly for the fermionic self energies ij = ij :

i
2
= Nf .
D
64
spatial homogeneity, isotropy, parity, charge conjugation and chiral symmetry can be
written as

1
f
FV t, t , p t =t =0 = n0 (p),
(7.1)
2

FV0 t, t , p t =t =0 = 0.
(7.2)
f
Here n0 (p) denotes the initial particle number distribution, whose values can range
between 0 and 1 (the definition of effective particle number distribution in terms of
equal-time two-point functions is detailed in Appendix A). At late times, when thermal
equilibrium is approached, this will lead to a canonical description with zero chemical
potential.10
The evolution equations (6.11) and (6.10) for the scalar correlators are second-order
in time and one needs to specify initial conditions for the propagators and their time
derivatives. As for the fermions, the initial conditions for the scalar spectral function is
completely specified by the field commutation relations (cf. Eq. (6.9)). For the scalar twopoint function F we consider (cf. also Refs. [9,22,25])

1
1

F t, t , p t =t =0 =
n0 (p) + ,
#0 (p)
2

t F t, t , p t =t =0 = 0,

1

t t F t, t , p t =t =0 = #0 (p) n0 (p) + ,
(7.3)
2
with an initial particle number distribution n0 (p) and initial mode energy #0 (p) (see
Appendix A).
It is instructive to consider for a moment the solution of the free field equations, which
can be obtained from Eqs. (5.5)(5.8) by neglecting the memory integrals on their r.h.s.
The solution of the fermionic free field equations with the above initial conditions reads

1
f
n0 (p) cos p t t ,
FV t, t , p =
2

1
f
0

n0 (p) sin p t t
FV t, t , p = i
2
and for the spectral functions one obtains

V t, t , p = sin p t t ,
V0 t, t , p = i cos p t t .
(7.4)
One observes that each mode of the equal-time correlator FV (t, t, p) is strictly conserved
in the absence of the memory integrals. Since this correlator is directly related to particle
number (see Appendix A), this means that the latter is conserved mode by mode in this
approximation. Although this is expected in the free field limit, this is of course not the case
in the fully interacting theory. We emphasize that such additional conservation laws do not
10 In particular, for nonzero chemical potential or net charge density the BCS mechanism can lead to the
condensation of Cooper pairs of fermions, which will be discussed elsewhere.
65
only appear in the free field limit, but are a property of mean-field-type approximations,
which include local corrections to the bare mass, but neglect the scattering contributions
described by the memory integrals on the r.h.s. of Eqs. (5.5)(5.8). In these approximations,
the existence of this infinite number of spurious conserved quantities prevents the system
to approach the thermal equilibrium limit at late times. This aspect has been discussed in
detail in the context of scalar field theories in Refs. [9,30]. It is therefore crucial to go
beyond such Gaussian approximations in order to correctly describe in particular the
late-time evolution of the system in the interacting theory.
8. Numerical implementation
We numerically solve the evolution equations (5.5)(5.8) and (6.10), (6.11), together
with the self-energies Eqs. (6.13)(6.16). The structure of the fermionic equations is
reminiscent of the form of classical canonical equations. In this analogy, FV (t, t ) plays
the role of the canonical coordinate and FV0 (t, t ) is analogous to the canonical momentum.
This suggests to discretize FV (t, t ) and V (t, t ) at t t = 2nat (even) and FV0 (t, t )
and V0 (t, t ) at t t = (2n + 1)at (odd) time-like lattice sites with spacing at . This
is a generalization of the leap-frog prescription for temporally inhomogeneous twopoint functions. This implies in particular that the discretization in the time direction is
coarser for the fermionic two-point functions than for the bosonic ones. This leap-frog
prescription may be easily extended to the memory integrals on the r.h.s. of Eqs. (5.5)(5.8)
as well.
We emphasize that the discretization does not suffer from the problem of so-called
fermion doublers [23]. The spatial doublers do not appear since (5.5)(5.8) are effectively
second order in x -space. Writing the equations for F V (t, t , x ) and V (t, t , x ) starting
from (5.5)(5.8) one realizes that instead of first order spatial derivatives there is a
Laplacian appearing. Hence we have the same Brillouin zone for the fermions and scalars.
Moreover, time-like doublers are easily avoided by using a sufficiently small stepsize in
time at /as .
The fact that Eqs. (5.5)(5.8) and (6.10), (6.11) contain memory integrals makes
numerical implementations expensive. Within a given numerical precision it is typically
not necessary to keep all the past of the two-point functions in the memory. A single
PIII desktop workstation with 2GB memory allows us to use a memory array with 470
timesteps (with 2 temporal dimensions: t and t ). We have checked for the presented runs
that a 30% change in the memory interval length did not alter the results. For a typical run
12 CPU-days were necessary.
The shown plots are calculated on a 470 470 323 lattice. (The dimensions refer
to the t and t memory arrays and the momentum-space discretization, respectively.) By
exploiting the spatial symmetries described in Section 5 the memory need could be reduced
by a factor of 30. We have checked that the infrared cutoff is well below any other mass
scales and that the UV cutoff is greater than the mass scales at least by a factor of three.
To extract physical quantities we follow the time evolution of the system for a given
lattice cutoff up to late times and measure the renormalized scalar mass m which is
then used to set the scale. In the evolution equations we analytically subtract only the
66
respective quadratically divergent terms obtained from a standard perturbative analysis.

We emphasize that for the results presented below we use the late-time thermal mass to set
the scale, and not the vacuum mass for convenience.
We made runs for a range of couplings g 2 = 0.491 which show very similar qualitative
behavior. Below, we present plots corresponding to g = 1, for which the time needed to
closely approach thermal equilibrium is the shortest. This allows us to obtain an accurate
thermalization with the lowest numerical cost.
9. Far-from-equilibrium dynamics
In Fig. 2 we present the time evolution of the fermion equal-time two-point function
FV (t, t; p) for three momenta p. Results are given for two very different initial particle
number distributions, which are displayed in the insets (see also Eqs. (7.1)(7.3)). The
(conserved) energy density is taken to be the same for both runs. In this case, since thermal
equilibrium is uniquely specified by the value of the energy density, the correlator modes
should approach universal values at late times if thermalization occurs.
It is striking to observe from Fig. 2 that after a comparably short time, much before the
correlation modes reach their late-time values, the dynamics becomes rather insensitive to
the details of the initial conditions: for a given momentum, the curves corresponding to
the two different runs come close to each other rather quickly. During the slow subsequent
evolution the system is still far away from equilibrium before the approach to the latetime values sets in. From both runs one observes that the characteristic time needed to
effectively loose the information about the details of the initial conditions is much shorter
Fig. 2. The time evolution of the fermion two-point function FV (t, t; p) for three values of the momentum p, in
units of the renormalized scalar thermal mass m. The evolution is shown for two very different initial conditions
with the same initial energy density. One observes that the dynamics becomes rather quickly insensitive to the
initial distributions displayed in the insetsmuch before the modes settle to their final values. The long-time
behavior is shown on a logarithmic scale for t 30 m1 .
67
than the time needed to approach the late-time result. Moreover, the late-time values are
found to be universal in the sense that the different runs agree with each other to very good
precision.
Fig. 3 shows the corresponding behavior of the scalar correlator modes F (t, t; p). The
respective initial particle number distributions in the scalar sector for the two runs are given
in the inset. For the two different runs the modes having the same momenta approach each
other rather slowly as compared to the fermionic sector. However, they reach their final
values on a time scale which is comparable to that observed for the fermions in Fig. 2.
To characterize these time scales in more detail, we consider in Fig. 4 the unequaltime two-point function FV (t, t ; p). As expected, if the system is to become insensitive
to the details of the initial conditions, we observe that the correlation between some time t
and another time t is suppressed for sufficiently large t t . We note that the oscillation
envelope of FV (t, t , p) can be well described in terms of an exponential for sufficiently
late times. The corresponding damping rate approaches a constant value. To estimate the
asymptotic rate we show in Fig. 4 the unequal-time two-point function FV (t, t ; p) as a
function of t t for the late time mt = 600. The fit to an exponential yields the damping
(damp)
rate f
(p = 0.78 m) = 0.03(1) m. We find a moderate momentum dependence of this
(damp)
rate with f
(damp)
(0) = 0.067(1) m and f
(damp)
(0) f
(p > 0).
(damp)
(p)
f
can be related to the width of the Fourier

We emphasize that the rate
transform of the spectral function with respect to the time difference t t . In principle,
the latter involves an integration over an infinite time interval. However, since we consider
an initial-value problem for finite times we know V (t, t , p) only on a finite interval. To
overcome this problem, we fit the data for V (t, t , p) by a 7-parameter formula that is
capable to account for the observed oscillations and damping, but which is more general
Fig. 3. The same as in Fig. 2 but for the bosonic two-point function F (t, t; p) for three different momenta. The
initial particle number distributions for the two runs denote by A and B are displayed in the insets. For the
employed parameters one observes that, in the scalar sector, the time needed to become effectively insensitive to
the initial distributions is comparable to the time scale describing the approach to the universal late-time value.
68
than the usual 2-parameter BreitWigner formula. We perform the Fourier transformation
on the extrapolated data. The resulting function V (, p) is displayed as a function of
frequency in the inset of Fig. 4. One clearly observes a nonzero width of the spectral
function, the value of which may be obtained from a fit to a BreitWigner formula. By
doing so, we obtain a very good agreement of the damping rate inferred from the width of
the spectral function on the one hand and from the linear fit on the log-plot for FV (t, t ; p)
on the other hand.
We can use the fermion damping rate to quantify the time scale characterizing the
effective loss of the details of the initial conditions: comparing with Fig. 2, we observe
(damp)
that the inverse fermion damping rate at p 0 (1/f
= 15(1) m1 ) characterizes
well the time for which the dynamics becomes rather insensitive to the initial distributions.
In contrast, we find that this time scale does not characterize the late-time behavior. For
the latter, we observe to very good approximation an exponential relaxation of each mode
FV (t, t; p) to its universal late-time value. Carrying out the measurement for different
modes, we observe that the corresponding rate is almost independent of momentum and
is given by 1/f(therm) = 95(5) m1 . The corresponding time scale 1/f(therm) is therefore
(damp)
much larger than the characteristic damping time 1/f

.
A similar analysis can be performed for the scalar sector. Here we find at p 0 the
(damp)
corresponding values 1/
= 50(5) m1 and 1/(therm) = 90(5) m1 . One observes
that although the damping rates for fermions and bosons are very different, the respective
thermalization rates are rather similar. For the scalars the thermalization rate is larger than
the damping rate, as is observed above for the fermions. However, the difference is much
less pronounced for the scalars. Similar studies in (1 + 1)-dimensional quantum [9] and
classical [8] scalar theories typically find a substantial difference between damping and
Fig. 4. The fermionic two-point function FV (t, t ; p) as a function of t t at late time t = 600 m1 . One
observes a very good agreement with an exponential behavior. The rate is well described by the width of the
corresponding spectral function in frequency space as shown in the inset (cf. the text for details).
69
thermalization rates. However, this may be a consequence of the stringent phase-space

restrictions and therefore particular to (1 + 1)-dimensional systems.
We finally note that approximate rates describing the early-time behavior may be
determined by an exponential fit to the functions F (t, 0), FV0 (t, 0) and FV (t, 0) in a finite
time interval. These rates depend on time and approach the late-time values given above.
At early times we observe an approximate exponential damping with a rate about twice as
big as the late-time value for the fermions, and about half the late-time rate for the scalars.
10. Approach to quantum thermal equilibrium

In the previous section, we have seen that the out-of-equilibrium evolution of the system
leads to a universal late-time behavior, uniquely characterized by the initial energy density.
We now analyze in detail if quantum thermal equilibrium, characterized by BoseEinstein
and FermiDirac statistics, is approached.11 In thermal equilibrium, the spectral function
and the statistical two-point function are not independent of each other, but are related by
the fluctuationdissipation relation. The latter is an exact relation, which can be stated in
4-dimensional Fourier space as [9,22]

(eq)
F , p = i

1 (eq)
BE
n () +
, p ,
2
(10.1)
for the scalar correlators, where nBE () = 1/[exp(/T ) 1] denotes the BoseEinstein
distribution function. The frequency is the Fourier conjugate of t t (in thermal
equilibrium, time-translation invariance implies that two-point functions only depend on
t t ). The value of the temperature T is determined by the energy density of the system.
The fluctuationdissipation relation for the fermion correlators is given by the equivalent
expression with the replacement (nBE () + 12 ) (nFD () 12 ) in Eq. (10.1) and the
FermiDirac distribution nFD = 1/[exp(/T ) + 1]. The same relations hold as well for
the statistical and spectral components of the self-energies in thermal equilibrium. They
provide an unambiguous way to extract the distribution functions, which are specific
for quantum thermal equilibrium. Out of equilibrium, the spectral and statistical twopoint functions are completely independent in general. However, if thermal equilibrium
is approached at late times, they become related by the fluctuationdissipation relations.
To extract the answer about the late-time distributions, without relying on any
assumptions, we consider the statistical and spectral correlators in Wigner coordinates.
For this we express F (t, t ; p) and (t, t ; p), as well as the corresponding fermion twopoint functions, in terms of the center coordinate X0 = (t +t )/2 and the relative coordinate
11 We emphasize that thermal equilibrium cannot be reached on a fundamental level from time-reversal
invariant evolution equations at any finite time. The results demonstrate that thermal equilibrium can be
approached very closely at sufficiently late time, without again deviating from it for practically accessible times.
For a more detailed discussion of this aspect see Ref. [9].
70
s 0 = t t and write
2X0
X0 ; , p =

0
ds 0 eis X0 + s 0 /2, X0 s 0 /2; p ,
(10.2)
2X 0
and equivalently for the other correlators. Since we consider an initial-value problem, the
time integral over s 0 is bounded by 2X0 (cf. also the detailed discussion in Ref. [22]).
If thermal equilibrium is approached for sufficiently large X0 , then the correlators do no
longer depend on X0 and a Fourier transform with respect to t t can be performed to very
good approximation (cf. the discussion in Section 9). The distribution functions may then
be extracted from the quotient of the Wigner transformed two-point functions. In Fig. 5
we show the respective ratios at late times. Both functions are in good agreement with the
equilibrium distributions nFD () and nBE (), respectively (cf. Eq. (10.1)). We emphasize
that the displayed continuous curves are no separate fits. They are the BoseEinstein and
FermiDirac distribution functions parametrized by the same temperature. In particular,
the value for the temperature is not fitted but has been extracted from the inverse slope
parameter as explained below.
We stress that the above procedure to extract the distribution functions is independent
of any assumption about a quasi-particle picture. However, for many practical purposes it
is very convenient to have an effective description of particle number and mode energy
directly in real timewithout the need of a Fourier transform. An efficient description
is elaborated in Appendix A. The value for the temperature in the thermal equilibrium
Fig. 5. The late-time ratio of the statistical two-point function and the spectral function in frequency
space, both for fermions (FV /V ) and for scalars (F / ). In thermal equilibrium the quotient corresponds
to the BoseEinstein (BE) distribution function for scalars and to the FermiDirac (FD) distribution for
fermionsindependently of any assumption on a quasi-particle picture (see Eq. (10.1)). The BE/FD distributions
are displayed by the continuous curves parametrized by the same temperature. The value of the latter, T = 0.94 m,
is actually not fitted but has been taken from the inverse slope of Fig. 8. This shows the correspondence with the
quasi-particle picture described in the text.
71
distributions of Fig. 5 has been actually measured based on this quasi-particle picture. In
Appendix A we define the effective particle number and energy to be used, both for the
fermionic and bosonic fields. For scalar field theories, this quasi-particle picture has been
successfully employed previously to investigate thermalization [9,22,26].
In Figs. 6 and 7 we show the effective quasi-particle number distributions defined as
1
nf (t, p) = FV (t, t; p)
2
for fermions and
1
+ n(t, p) = #(t, p)F (t, t; p),
2

t t F (t, t ; p) 1/2
#(t, p) =
,
F (t, t ; p)
t =t
(10.3)
(10.4)
(10.5)
for scalars, where (t, p) is the quasi-particle mode energy as discussed in Appendix A.12
The curves correspond to the initial conditions of run A shown in Figs. 2 and 3.
One observes how the effective fermion and boson particle numbers change with time,
the former approaching a FermiDirac and the latter a BoseEinstein distribution. To
emphasize this point, we plot the corresponding inverse slope functions log(1/nf 1)
and log(1/n + 1), which reduce to straight lines for FermiDirac and BoseEinstein
distributions respectively. The associated inverse slopes correspond to the temperature
of the thermal equilibrium distributions. We see in Figs. 6 and 7 that both inverse slope
Fig. 6. The time-dependent fermion quasi-particle distribution nf (t, p) as a function of mode energy p for various
times t. We have plotted the inverse slope function log(1/nf 1), which reduces to a straight line intersecting
the origin when nf (t, p) approaches a FermiDirac distribution. This plot shows the data for run A of Fig. 2.
12 Note that because of chiral symmetry there is no mass term present for the fermions and the corresponding
quasi-particle mode energy is simply p.
72
Fig. 7. The time-dependent boson quasi-particle distribution n(t, p) as a function of mode energy #(t, p) (see
text) for various times. In this case the inverse slope function is log(1/n + 1), which reduces to a straight line in
case of a BoseEinstein distribution. This plot shows the data for run A of Fig. 3.
Fig. 8. Boson and fermion quasi-particle distributions at late times as a function of mode energy (np n(t, p)
and #p #(t, p) for bosons and np nf (t, p) and #p p for fermions). Both inverse slope functions employed
in Figs. 6 and 7 have been applied here in order to demonstrate that the fermions clearly do not follow a
BoseEinstein distribution, and the scalars cannot be characterized by a FermiDirac distribution. In contrast,
for the respective correct statistics the curves lie on top of each other, showing that both fermions and scalars are
described by the same inverse slope parameter. The value of the latter is used in Fig. 5 as the temperature for the
distributions nFD () and nBE ().
functions approach straight lines at late times. The associated temperatures for fermions
and bosons are independent of time and agree very well with each other, as shown in
Fig. 8. For comparison, we display in Fig. 8 the fermion inverse slope function evaluated
73
with the bosonic effective particle number, and vice versa. This illustrates the degree of
sensitivity of the inverse slope functions to the different statistics and in turn the degree of
precision with which we are able to probe thermalization.
11. Conclusions
In this paper we have discussed the far-from-equilibrium dynamics and subsequent
thermalization of a system of coupled fermionic and bosonic quantum fields. We solved
the nonequilibrium dynamics beyond mean-field type approximations by calculating
the complete lowest nontrivial order in a systematic coupling expansion of the 2PI
effective action, which includes direct scattering as well as memory and off-shell effects.
To our knowledge this is the first time that such a calculation is performed without
further approximations. As a result, we show that, for various far-from-equilibrium initial
conditions, the late-time behavior is universal and uniquely determined by the value
of the initial energy density. Moreover, we are able to probe the approach to quantum
thermal equilibrium, characterized by the emergence of FermiDirac and BoseEinstein
distribution functions, with high accuracy. We emphasize that in the present calculation,
besides the limitations of a coupling expansion, there is no other input than the dynamics
dictated by the considered quantum field theory for given nonequilibrium initial conditions.
This work can be extended in many directions. Most of the equations we derive are also
valid in the phase with spontaneous symmetry breaking. Combined with earlier work on
scalar theories [9,11,25], this provides a description of the nonequilibrium dynamics of the
linear -model for QCD with two quark flavors. The model has served for many years as a
valuable testing ground for ideas on the equilibrium phase structure of low energy QCD at
nonzero temperature and density. With the present techniques a quantitative understanding
of the out-of-equilibrium physics of this model is within reach.
Acknowledgements
We thank C. Wetterich for many discussions and collaboration on related work. We are
also grateful to W. Wetzel for his continuous support with computers. Sz.B. acknowledges
the hospitality of the Institute fr Theoretische Physik, Heidelberg. His work was supported
by the short-term scholarship program of the DAAD.
Appendix A. Effective particle number and energy

A particle number can only be strictly defined in the presence of a conserved charge.
However, for physical interpretation it is often convenient to define an effective particle
number even when there is no conserved charge. In particular, besides a total particle
number it is often useful to have a definition of an effective particle number per
(momentum) mode. The latter is typically not conserved in an interacting theory, and in
the context of thermalization one would like to find a time-dependent particle number
74
distribution, which allows one to observe a BoseEinstein or FermiDirac distribution

at sufficiently late times.13 Of course, the notion of an effective particle number is
typically only meaningful if the relevant degrees of freedom or quasi-particles are weakly
interacting.
There are many different ways of how one can introduce the notion of an effective
particle number (for a recent discussion see, e.g., Ref. [27] and references therein).
For example, one can define it as the average energy per mode divided by the energy
of the corresponding mode. In an interacting theory, this procedure can be ambiguous
because the expression of the total energy receives contributions from interactions and
the average energy per mode is not uniquely defined. In this appendix, we discuss an
elegant way to circumvent this difficulty. More importantly for our purposes, the procedure
leads to a definition of an effective particle number density, which indeed allows one
to directly observe the emergence of a FermiDirac or BoseEinstein distribution from
the nonequilibrium dynamics for the theory considered in this paper. In particular, it can
be explicitely shown that the effective particle numbers are always positive and, for the
fermions, smaller than one (see Appendix B). Applying the present construction to the
case of neutral scalar fields, we recover the particle number definition used in previous
studies to exhibit thermalization in purely scalar field theories [9,22].
A.1. Fermions
We start by considering the case of charged fields and construct the effective
particle number from the conserved current generated by the U (1) symmetry.14 The
. Fourier transformed with respect
associated 4-current for each given flavor is
to spatial momenta, the expectation value of the latter can be written as Jf (t, p) =
<
tr[ D (t, t, p)], where the subscript stands for fermions (cf. Eq. (3.1)). In terms of the
equal-time statistical two point function, its temporal and spatial components read:15

Jf0 (t, p) = 2 1 2FV0 (t, t; p) ,
J f (t, p) = 4
v FV (t, t; p).
A nice property is that the above expressions in terms of the full correlators do not contain
any explicit dependence on the interaction part. In order to obtain an effective particle
number, we want to identify these expressions with the corresponding ones in a quasiparticle description with free-field expressions. These are given by:16

0(QP)
Jf
(t, p) = 2 1 + Qf (t, p) ,
13 We emphasize, however, that the definition of an effective particle number distribution is not necessary to
analyse the approach to quantum thermal equilibrium, as is discussed in Section 10.

14 As we shall see below for scalar fields, the expression for the effective particle number one obtains in this
way can be directly applied to the case of neutral fields as well.
15 The constant factor in the temporal component comes from the fact that we define the current without the
standard normal ordering.
16 Here, we have explicitly used our assumptions of parity symmetry and rotational invariance (cf. Section 5),
which imply that the different spin states contribute the same, therefore the factor of two.
75

(QP)
J f (t, p) = 2
v 1 2Nf (t, p) ,
where Qf (t, p) = nf n f is the difference between particle and antiparticle effective
number densities and Nf (t, p) = (nf + n f )/2 is their half-sum. The physical content
of these expressions is simple: the temporal component J 0 directly represents the netcharge density per mode Qf (t, p), whereas the spatial part J is the net current density per
mode and is therefore sensitive to the sum of particle and antiparticle number densities.
Identifying the above expressions, we define
1
(A.1)
Qf (t, p) = FV0 (t, t; p),
2
1
Nf (t, p) = FV (t, t; p).
(A.2)
2
Of course, these expressions are only meaningful for the case that the interacting theory is
well-described by a quasi-particle picture. The important point is that the above procedure
allows one to construct an effective particle number density without knowing a priori which
part of the interaction is to be considered as the dressing of the quasi-particles and which
part describe their residual interactions. We note that the equal-time two-point functions
on the r.h.s. of these equations are real by definition (see Eq. (5.3)). Moreover, using
the anticommutation relation for the fermion fields, it is shown in Appendix B that these
definitions always satisfy 0 Nf (t, p) 1 and 1 Qf (t, p) 1. These properties are
important for the above definitions to be physically meaningful.
Introducing a nonvanishing net charge density per mode was useful for the above
general construction. However, in the present paper, we consider only CP -invariant
systems, which imply that the latter should vanish. Indeed, we see from Eq. (5.3) that
the requirement of CP -invariance imply that our above definition of net charge density per
mode vanishes identically for all times and for all modes. Therefore, the effective particle
and antiparticle numbers are equal and we have
1
nf (t, p) = FV (t, t; p),
2
where nf (t, p) is the effective particle number.
Qf (t, p) = 0,
A.2. Bosons
Following the same lines as above, let us consider for a moment the case of a single
charged scalar field . The 4-current associated the corresponding U (1) symmetry is
i[ ( ) ( )] and, as for the case of fermions, its expectation value has a
simple expression in terms of the equal-time statistical two-point function of the charged
field. In momentum space, it reads:

Jb0 (t, p) = i(t t )F t, t ; p t =t 1,
(t, t; p),
J b (t, p) = 2pF
where, as before, the constant contribution comes from the fact that we define the
current without the usual normal ordering. Here, the statistical two-point function for
76
the charged scalar field is defined as the expectation value of the anticommutator of
two fields operators: F (t, t ; p) = 12 [(t, p ), (t , p )]+ . It has the symmetry property
F (t, t ; p) = F (t , t; p), so that the above expressions are real. The corresponding quasiparticle expressions read:
Jb0(QP) (t, p) = Qb (t, p) 1,

p
J b(QP) (t, p) =
1 + 2Nb (t, p) ,
#(t, p)
where Qb (t, p) and Nb (t, p) have the same meaning as before in terms of effective
particle and antiparticle number densities and where #(t, p) is the quasi-particle energy.
We, therefore, define:

Qb (t, p) = i(t t )F t, t ; p t =t ,
(A.3)
1 + 2Nb (t, p) = 2#(t, p)F (t, t; p).
(A.4)
Note that the r.h.s. of both expressions are real quantities, as they should if the l.h.s. are to
be interpreted as charge and quasi-particle number densities respectively.
It remains to define the effective quasi-particle energy #(t, p).17 For this purpose, we
use the free-field like expression for the average energy per mode, which in the case of a
single charged scalar field can be written as

t t F t, t ; p t =t + # 2 (t, p)F (t, t; p) #(t, p) 2n(t, p) + 1
in terms of the statistical two-point function. Therefore, one obtains for the effective quasiparticle energy:

t t F (t, t ; p)
.
# 2 (t, p) =
(A.5)
F (t, t ; p)
t =t
It is straightforward to show that the combination of equal-time correlators appearing on
the r.h.s. of (A.5) is indeed positive. Moreover, using commutation relations for the scalar
field, one can show that the effective particle number Nb (t, p), as given by Eqs. (A.4)
and (A.5), is a positive quantity (see Appendix B). We note that for the case of CP invariant systems, the effective charge density per mode (A.3) vanishes, as it should.18
When dealing with neutral scalar fields, as it is the case in the present paper, we can
use the same formula derived above. Notice that this is consistent since in that case one
has F (t, t ; p) = F (t , t; p), which directly implies that Qb (t, p) = 0. Therefore, in the
present paper, we use the definition
1
+ n(t, p) = #(t, p)F (t, t; p)
(A.6)
2
together with (A.5) for the boson effective particle number n(t, p) and mode energy #(t, p)
for each individual scalar species [9,22].
17 Note that this was not necessary for the fermionic case, because of our assumption of chiral symmetry, which
prevents an effective mass term. In fact, one can repeat the following argument to obtain for the fermion effective
quasi-particle energy #f (t, p) = p, as employed above.
18 This immediately follows from the behavior of the statistical propagator under CP -transformation:
F (t, t ; p) F (t , t; p).
77
Appendix B
Here we show that the combinations of equal-time two-point functions used in this paper
to define effective particle number densities are always positive and, for the fermionic case,
smaller than one. It is demonstrated to be a simple consequence of the (anti)commutation
relations of field operators.
B.1. Fermions
As a first exercise, we show that for any operators and satisfying the anticommutation relations

, + = 1
and

[, ]+ = , + = 0,
one has

0 1,
(B.1)
where the brackets denote an average with respect to any density matrix. The left inequality
above is trivially obtained, for example, by inserting a complete sum of states between the
operators and . One obtains a sum of positive quantities which is of course positive.
To show the second inequality, we introduce the Hermitian operator N = . Using the
anticommutation relations above, it is easy to see that N 2 = N , from which it follows that

2

?n2 N N = N 2 N2 = N 1 N .
It is clear that ?n2 0 and we have just shown that N 0. One therefore conclude from
the above equation that N 1, as announced.
We now come to our effective particle number densities. Using Eqs. (A.1) and (A.2)
and recalling that Qf nf n f and Nf (nf + n f )/2, we get for the effective particle
and antiparticle number densities:

1
1
nf (t, p) = FV (t, t, p) + FV0 (t, t, p) = tr (0 + v )F (t, t, p) ,
2
4

1
1
n f (t, p) = FV (t, t, p) FV0 (t, t, p) = tr (0 v )F (t, t, p) .
2
4
p ), which satisfy the
In terms of the fermionic field operators (t, p ) and (t,
anticommutation relations

p ) = 0
(t, p ), (t,
+
and

(t, p ), (t, p )

+

p ), (t,
p ) = 0,
= (t,
+
78
one has

1
p ) .
(t, p ), (t,
2
Therefore, one can rewrite (from now on, we drop the explicit t and p dependence):

1
nf = (
) ,
0+v
4

1
0 v ) .
1 n f = (
4
Now we introduce the operator
F (t, t, p) =
0 + v
,
2
in term of which,
=
1
,
nf =
4
4
(B.2)
=1
where we explicitely wrote the sum over Dirac indices. For a given Dirac index, it is
straightforward to check that

, + = 1 0 v = 1
(B.3)
and

[ , ]+ = , + = 0,
where we specialized to the Dirac basis to write the last equality of Eq. (B.3).19 Using
(B.1) for each individual , we conclude from (B.2) that
0 nf (t, p) 1,
for any time t and any momentum p. It is straightforward to repeat the above arguments to
show that
0 n f (t, p) 1.
B.2. Bosons
For scalars, we first use Eqs. (A.5) and (A.6) to rewrite

1
+ n(t, p) = F (t, t; p) t t F (t, t ; p) t =t .
2
Recalling the definition of the statistical propagator in terms of field operators:

F (t, t; p) = (t, p )(t, p ) ,

t t F (t, t ; p) = (t, p )(t, p ) ,
t =t
19 This is more convenient, but not necessary. It is simple to adapt the argument to any basis.
79
where (t, p)
and (t, p)
are the canonical momenta conjugate to (t, p)
and (t, p)

we see that
respectively (e.g., (t, p)
= (t, p)),

1
,
2
where we dropped in the notation the explicit time and momentum dependence of field
operators. The second inequality is nothing but Heisenbergs uncertainty relation, a direct
consequence of the equal-time canonical commutation relations of field operators [31].
n(t, p) 0
References
[1] P. Braun-Munzinger, J. Stachel, J. Phys. G 28 (2002) 1971;
U.W. Heinz, P.F. Kolb, Two RHIC puzzles: Early thermalization and the HBT problem, in: R. Bellwied,
J. Harris, W. Bauer (Eds.), Proceedings of the 18th Winter Workshop on Nuclear Dynamics. EP Systema,
Debrecen, Hungary, 2002, pp. 205216 hep-ph/0204061.
[2] J. Serreau, hep-ph/0209067, Proceedings of Quark Matter 2002, to appear in Nucl. Phys. A;
J. Serreau, D. Schiff, JHEP 0111 (2001) 039;
R. Baier, A.H. Mueller, D. Schiff, D.T. Son, Phys. Lett. B 539 (2002) 46.
[3] For a review see, J.P. Blaizot, E. Iancu, Phys. Rep. 359 (2002) 355, and references therein;
See also: P. Arnold, G.D. Moore, L.G. Yaffe, JHEP 0301 (2003) 030.
[4] F. Cooper, S. Habib, Y. Kluger, E. Mottola, J.P. Paz, P.R. Anderson, Phys. Rev. D 50 (1994) 2848;
D. Boyanovsky, H.J. de Vega, R. Holman, J. Salgado, Phys. Rev. D 59 (1999) 125009.
[5] F. Cooper, V.M. Savage, Phys. Lett. B 545 (2002) 307;
A. Chodos, F. Cooper, W. Mao, A. Singh, Phys. Rev. D 63 (2001) 096010;
D. Boyanovsky, H.J. De Vega, R. Holman, M.R. Martin, Phys. Rev. D 65 (2002) 045007.
[6] J. Baacke, K. Heitmann, C. Ptzold, Phys. Rev. D 58 (1998) 125013;
J. Baacke, C. Ptzold, Phys. Rev. D 62 (2000) 084008.
[7] F. Cooper, S. Habib, Y. Kluger, E. Mottola, Phys. Rev. D 55 (1997) 6471.
[8] G. Aarts, G.F. Bonini, C. Wetterich, Phys. Rev. D 63 (2001) 025012.
[9] J. Berges, Nucl. Phys. A 699 (2002) 847.
[10] J. Berges, J. Cox, Phys. Lett. B 517 (2001) 369.
[11] G. Aarts, D. Ahrensmeier, R. Baier, J. Berges, J. Serreau, Phys. Rev. D 66 (2002) 045008.
[12] J.M. Luttinger, J.C. Ward, Phys. Rev. 118 (1960) 1417;
G. Baym, Phys. Rev. 127 (1962) 1391.
[13] J.M. Cornwall, R. Jackiw, E. Tomboulis, Phys. Rev. D 10 (1974) 2428.
[14] E. Calzetta, B.L. Hu, Phys. Rev. D 37 (1988) 2878.
[15] Y.B. Ivanov, J. Knoll, D.N. Voskresensky, Nucl. Phys. A 657 (1999) 413.
[16] For reviews, see: P. Danielewicz, Ann. Phys. 152 (1984) 239;
For relativistic fermions, see also: H.T. Elze, U.W. Heinz, Phys. Rep. 183 (1989) 81;
D.A. Brown, P. Danielewicz, Phys. Rev. D 58 (1998) 094003.
[17] P.B. Greene, L. Kofman, Phys. Rev. D 62 (2000) 123516.
[18] M. Joyce, K. Kainulainen, T. Prokopec, JHEP 0010 (2000) 029;
K. Kainulainen, T. Prokopec, M.G. Schmidt, S. Weinstock, Phys. Rev. D 66 (2002) 043502.
[19] I.D. Lawrie, D.B. McKernan, Phys. Rev. D 62 (2000) 105032.
[20] G. Aarts, J. Smit, Nucl. Phys. B 555 (1999) 355.
[21] J. Schwinger, J. Math. Phys. 2 (1961) 407;
L.V. Keldysh, Zh. Eksp. Teor. Fiz. 47 (1964) 1515, Sov. Phys. JETP 20 (1965) 1018;
K.C. Chou, Z.B. Su, B.L. Hao, L. Yu, Phys. Rep. 118 (1985) 1.
[22] G. Aarts, J. Berges, Phys. Rev. D 64 (2001) 105010.
[23] I. Montvay, G. Mnster, Quantum Fields on a Lattice, Cambridge Univ. Press, Cambridge, 1994.
[24] K. Kainulainen, T. Prokopec, M.G. Schmidt, S. Weinstock, COSMO-01, hep-ph/0201245.
80
[25]
[26]
[27]
[28]
[29]
J. Berges, J. Serreau, hep-ph/0208070.

G. Aarts, J. Berges, Phys. Rev. Lett. 88 (2002) 041603.
B. Garbrecht, T. Prokopec, M.G. Schmidt, hep-th/0211219.
M. Gell-Mann, M. Levy, Nuovo Cimento 16 (1960) 705.
See, e.g., R.D. Pisarski, F. Wilczek, Phys. Rev. D 29 (1984) 338;
J. Berges, D.U. Jungnickel, C. Wetterich, Phys. Rev. D 59 (1999) 034010;
O. Scavenius, A. Mocsy, I.N. Mishustin, D.H. Rischke, Phys. Rev. C 64 (2001) 045202.
[30] L.M. Bettencourt, C. Wetterich, Phys. Lett. B 430 (1998) 140.
[31] A. Bohm, Quantum Mechanics: Foundations, Applications, Springer-Verlag, Heidelberg, 1994.

D-branes and the Standard Model

I. Antoniadis a,1 , E. Kiritsis b,d , J. Rizos c , T.N. Tomaras b
a CERN Theory Division, CH-1211, Genve 23, Switzerland
b Department of Physics and Institute of Plasma Physics, University of Crete and FORTH,
71003 Heraklion, Greece

c Department of Physics, University of Ioannina, 45110 Ioannina, Greece
d CPHT, UMR du CNRS 7644, Ecole Polytechnique, 91128 Palaiseau, France
Received 26 November 2002; received in revised form 24 February 2003; accepted 25 March 2003
Abstract
We perform a systematic study of the Standard Model embedding in a D-brane configuration of
type I string theory at the TeV scale. We end up with an attractive model and we study several
phenomenological questions, such as gauge coupling unification, proton stability, fermion masses
and neutrino oscillations. At the string scale, the gauge group is U (3)color U (2)weak U (1)1
U (1)bulk . The corresponding gauge bosons are localized on three collections of branes; two of them
describe the strong and weak interactions, while the last Abelian factor lives on a brane which is
extended in two large extra dimensions with a size of a few microns. The hypercharge is a linear
combination of the first three U (1)s. All remaining U (1)s get masses at the TeV scale due to
anomalies, leaving the baryon and lepton numbers as (perturbatively) unbroken global symmetries
at low energies. The conservation of baryon number assures proton stability, while lepton number
symmetry guarantees light neutrino masses that involve a right-handed neutrino in the bulk. The
model predicts the value of the weak angle which is compatible with the experiment when the string
scale is in the TeV region. It also contains two Higgs doublets that provide tree-level masses to all
fermions of the heaviest generation, with calculable Yukawa couplings; one obtains a naturally heavy
top and the correct ratio mb /m . We also study neutrino masses and mixings in relation to recent
solar and atmospheric neutrino data.
E-mail address: irizos@cc.uoi.gr (J. Rizos).

1 On leave of absence from CPHT, UMR du CNRS 7644, Ecole Polytechnique, 91128 Palaiseau, France.
doi:10.1016/S0550-3213(03)00256-6
82
I. Antoniadis et al. / Nuclear Physics B 660 (2003) 81115
1. Introduction
In a previous work [1,2], a minimal embedding of the Standard Model (SM) was
proposed in a D-brane configuration of type I string theory with large internal dimensions
and low fundamental scale [3,4]. The SU(3) color and SU(2) weak gauge fields were
confined on two different collections of branes. The model correctly accommodated the
right value of the weak angle for a choice of the string scale of a few TeV. It contained
two Higgs doublets and guaranteed proton stability. Among the issues, which were not
addressed, are the fermion masses, neutrino oscillations, and a natural suppression of
lepton number violating processes.
A generic feature of the models studied was that some of the SM states should
correspond to open strings with one end in the bulk, implying the existence of some extra
branes, in addition to the ones used above [1,2]. Starting from the last point, in the present
work we introduce an extra brane in the bulk with a corresponding U (1)b bulk gauge group
[2]. This group is broken by anomalies, leaving behind an additional global symmetry that
will be identified with the lepton number. In order to give masses to the neutrinos, we
introduce a right-handed neutrino in the bulk [5] that carries non-trivial lepton number.
Large neutrino masses are then forbidden by symmetry, while the right-neutrino coupling
suppression required to explain the neutrino oscillation data, is achieved if the bulk has two
dimensions of submillimeter size.
More precisely, in the minimal case of one bulk neutrino, we show that solar
and atmospheric neutrino data can be accommodated using essentially the two lowest
frequencies of the neutrino mass matrix: the mass of the zero mode, arising via the
electroweak Higgs phenomenon, which is suppressed by the volume of the bulk, and
the mass of the first KaluzaKlein (KK) excitation. The former is used to reproduce the
large mixing angle (LMA or even LOW) solution to the solar neutrino anomaly, through
e transitions. The later is used to explain atmospheric neutrino oscillations with an
amplitude which is enhanced due to logarithmic corrections of the two-dimensional bulk
[6]. Compatibility of the two conditions using one bulk right neutrino is possible only if one
introduces a non-orthogonal angle between the two compact bulk dimensions, that leads
simultaneously to a CP violation in the neutrino sector. Atmospheric oscillations contain
however a significant sterile component which seems to be in contradiction with recent
atmospheric data analyses.
We also compute the tree-level Yukawa couplings of the two higgses to the fermions
of the heaviest generation. They are given in terms of the gauge couplings and lead to a
naturally heavy top and a ratio mb /m compatible with the experimental data. Next, we
proceed to a systematic description of the main features that we will use in the following
sections.
The general framework is type I string theory. We shall restrict ourselves to models
in which the closed string sector is supersymmetric, while supersymmetry is generically
broken by the open strings at the string scale [7].2 Within our framework, the minimal
ensemble of D-branes needed in our construction is the following mutually orthogonal
2 Recent progress in constructing type I vacua with structure close to the SM can be found in [810].
83
stacks: a stack of three coincident branes to generate the color group, a second stack
of two coincident branes to describe the weak SU(2)L gauge bosons, and one more
brane to generate the U (1)b bulk discussed above. The resulting gauge group so far is
U (3)c U (2)L U (1)b , with the three U (1) generators denoted by Qc , QL and Qb ,
respectively. To ensure proton stability, we require baryon number conservation with
generator B Qc . The hypercharge Y cannot have a component along Qb , since this
would lead to unrealistically small gauge coupling, and as explained in [1] the correct
assignment of SM quantum numbers requires the presence of an extra abelian factor, named
U (1)1 with generator Q1 , living on an additional brane. This brane should lie on top of the
color or the weak stack of branes, as we argue below.
Since in our framework, supersymmetry is broken by combinations of (anti-)branes
and orientifolds which preserve different subsets of the bulk supesymmetries, any pair
of D-branes Dp and Dp satisfy p p = 0 mod 4. It follows that a system with three
stacks of mutually orthogonal branes in the six-dimensional internal (compact) space
consists, up to T-dualities, of D9-branes with two different types of D5-branes, extended in
different directions. Specifically, the U (1)b lives on the D9-brane, while the U (3)c and
U (2)L are confined on two stacks of 5-branes, the first along say the 012345 and the
other along the 012367 directions of ten-dimensional spacetime. Thus, the (submillimeter)
bulk is necessarily two-dimensional (extended along the 89 directions), and the additional
U (1)1 brane has to coincide with either U (3)c or U (2)L . The parameters of the model
are the string scale Ms , the string coupling gs and the volumes v45 , v67 and v89 of the
corresponding subspaces, in string units.3 In terms of those, the four-dimensional Planck
mass MP is given by
MP2 =
8
v45 v67 v89 Ms2
gs2
(1.1)
and the non-Abelian gauge couplings are

1
1
= v45 ,
2
gs
g3
1
1
= v67 .
2
gs
g2
(1.2)
It follows that
MP2 =
8
v89 Ms2
2
g3 g22
2
v89 Ms2 ,
3 2
(1.3)
where i = gi2 /4 and v89 v89 /(2)2 = R8 R9 for a rectangular torus of radii R8 , R9 .
The U (1)1 gauge coupling g1 is equal to g3 (g2 ), if the U (1)1 brane is on top of the U (3)c
(U (2)L ).
Upon T-duality, one finds two additional realizations: (i) a set of D3-branes (along 0123)
describing U (3)c , and two orthogonal sets of D7-branes along 01236789 and 01234567
describing U (1)b and U (2)L , respectively; (ii) three sets of D5-branes along 012389,
012345 and 012367, giving rise to U (1)b , U (3)c and U (2)L , respectively. In both cases,
relation (1.3) remains intact.
3 Using T-duality, we choose all internal volumes to be bigger than unity, v > 1.
ij
84
The gauge coupling gb of the U (1)b gauge boson which lives in the bulk is extremely
small since it is suppressed by the volume of the bulk v89 . For instance, in the case where
the U (1)b lives on a D9-brane, its coupling is given by
1
1
gs MP2
= v45 v67 v89 =
,
2
gs
8 Ms2
gb
(1.4)
where in the second equality we used Eq. (1.1). Using now the weak coupling condition
2 following from v > 1 in Eq. (1.2), one finds
gs < 1 and the inequality gs > g3,2
ij
Ms
8 Ms
(1.5)
8
< gb <
,
MP
g3 MP
which implies that gb 1016 1014 for Ms 110 TeV.
If the U (1)b gauge boson is light, it will be subject to strong constraints coming from
supernova observations, since it would be copiously produced in various nuclear reactions
leading to supernova cooling through energy loss in the bulk of extra dimensions. The
corresponding process is much stronger than the production of gravitons because of the
non-derivative coupling of the gauge boson interaction [11]. In fact, in the case of n large
transverse dimensions of common radius R, satisfying mA , R 1 T with mA the gauge
boson mass and T the supernova temperature, the production rate PA is proportional to

n
1
T n2
PA gb2 R(T mA ) 2
,
Msn
T
(1.6)
where the factor [R(T mA )]n counts the number of KaluzaKlein (KK) excitations
of the U (1)b gauge boson with mass less than T . This rate can be compared with the
corresponding graviton production
PG
1
Tn
(RT )n n+2 ,
2
MP
Ms
(1.7)
showing that for n = 2 (sub)millimeter extra dimensions, it is unacceptably large, unless

the bulk gauge boson acquires a mass mA 10 MeV.
The paper is organized in seven sections, of which this introduction is Section 1.
In Section 2, we perform a systematic search for models with four sets of branes
corresponding to the gauge group U (3)c U (2)L U (1)1 U (1)b with the minimal
standard model fermion spectrum and a Higgs sector that generates masses for all quarks
and leptons of the heaviest generation. We identify the hypercharge U (1)Y combination
and in Section 3 we perform a renormalization group analysis of gauge couplings to
identify models with low string scale, where the U (1)1 is on top of either the color or
the weak branes. In Section 4, we select four models with string scale in the TeV region,
possessing in addition baryon and lepton number conservation, and we describe their main
phenomenological features.4 They all contain two Higgs doublets that can provide treelevel masses to all fermions. Moreover, apart from the hypercharge, all other Abelian
4 Orientifold models with baryon and lepton number conservation were also constructed in Ref. [9].
85
factors are broken by mixed gauge and gravitational anomalies and become massive at
the string scale. In Section 5, we compute the tree-level Yukawa couplings of the two
higgses to the fermions of the heaviest generation and study predictions for mass relations.
In Section 6, we introduce one right-handed neutrino in the bulk and study the generation
of neutrino masses and neutrino oscillations. Finally, Section 7 contains our summary and
conclusions.
2. Model search
As shown in [1], the minimal D-brane configuration that can successfully accommodate
the Standard Model (SM) consists of three sets of branes with gauge symmetry U (3)c
U (2)L U (1)1 . The first set contains three coincident branes (color branes). An open
string with one end attached to this set transforms as an SU(3)c triplet (or anti-triplet),
but also carries an additional U (1)c quantum number which can be identified with the
(gauged) baryon number. Similarly, U (2)L is realized by a set of two coincident branes
(weak branes) and open strings attached to them from the one end are SU(2)L doublets
characterized by an additional U (1)L quantum number, the (gauged) weak doublet
number. Moreover, consistency of the SM embedding requires the presence of an additional
U (1)1 factor, generated by a single brane. This is needed for several reasons: TeV scale
unification, baryon number conservation, and mass generation for all quarks and leptons
of the heaviest generation. The hypercharge is then a linear combination of the three
Abelian factors, Y = k3 Qc + k2 QL + k1 Q1 , where Qc , QL , Q1 are the charges under
U (1)c , U (1)L , U (1)1 , respectively. It turns out [1] that there exist four possible viable
models that reproduce the weak mixing angle all low energies. They correspond to k3 = 23
(k3 = 13 ), k2 = 12 , k1 = 1 and require the Abelian brane U (1)1 to be on top of the color
(weak) branes, so that g3 = g1 (g2 = g1 ).
In all the above brane configurations there exist states (e.g., the SU(2)L singlet antiquarks) which correspond to open strings with only one of their ends attached to one
of the three sets of D-branes. The other end is in the bulk, and requires the existence
of some additional branes extended in the bulk, carrying extra quantum numbers. In this
work, we consider a minimal extension of the models considered in [1] by introducing
one additional D-brane in the bulk giving rise to an extra Abelian gauge factor U (1)b . As
we will see later, the requirement of baryon and lepton number conservation leads to four
possible models that we are going to study in the next section. However, in this section, we
do not impose this constraint and we systematically explore the possibility of reproducing
the SM spectrum, together with possibly additional Higgs scalars, as open strings stretched
between any two of the four sets of branes. The extension of the Higgs sector is required for
the realization of the electroweak symmetry breaking and mass generation for all fermions
of at least one (the heaviest) generation.
Thus, the total gauge group is
G = U (3)c U (2)L U (1)1 U (1)b
= SU(3)c U (1)c SU(2)L U (1)L U (1)1 U (1)b
(2.1)
86
Table 1
SM particles with their generic charges under the Abelian part of the gauge group U (3)c U (2)L U (1)1
U (1)b
Particle

Q 3, 2, 16

1, 2
uc 3,
3

1, + 1
d c 3,
3

L 1, 2, 12
U (1)c
U (1)L
ec (1, 1, +1)
U (1)1
U (1)b
+1
a1
a2
b1
b2
+1
c1
c2
dL
d1
d2
and contains four Abelian factors. The assignment of the SM particles is partially fixed
from its non-Abelian structure. The quark doublet Q corresponds to an open string with
one end on the color and the other on the weak set of branes. The anti-quarks uc , d c must
have one of their ends attached to the color branes. The lepton doublet and possible Higgs
doublets must have one end on the weak branes. However, there is a freedom related to
the Abelian structure, since the hypercharge can arise as a linear combination of all four
Abelian factors. In a generic model, the Abelian charges can be expressed without loss of
generality in terms of ten parameters displayed in Table 1.
In a convenient parametrization, normalizing the U (N) SU(N) U (1) generators
ab /2, and measuring the corresponding U (1) charges with respect to
as Tr T a T b =
the coupling g/ 2N , the ten parameters are integers: a1,2, b1,2 , c1,2 , d2 = 0, 1, d1 =
0, 1, 2, dL = 0, 2, w = 1 satisfying

|ai | =
|bi | =
|ci | = 1,
|di | = 2.
(2.2)
i=1,2
i=1,2
i=1,2
i=1,2,L
The first three constraints in (2.2) correspond to the requirement that the uc and d c antiquarks, as well as the lepton doublet, must come from open strings with one end attached
to one of the Abelian D-brane sets. The fourth constraint forces the positron ec open string
to be stretched either between the two Abelian branes, or to have both ends attached to the
Abelian U (1)1 brane, or to the weak set of branes. In the latter case, it has U (1)L charge
2 and is an SU(2)L singlet arising from the anti-symmetric product of two doublets. The
parameter w in Table 1 refers to the U (1)L charges of the quark-doublets, that we can
choose to be 1, since doublets are equivalent with anti-doublets. Note that a priori one
might also consider the case in which one of the uc and d c anti-quarks arises as a string
with both ends on the color branes (3 3 = 3 + 6), so that its U (1)c charges would be
2. This, however, would invalidate the identification of U (1)c with the baryon number
and forbid the presence of quark mass terms, since one of the combinations Quc and Qd c
would not be neutral under U (1)c . Hence, this case will not be explored.
The hypercharge can in general be a linear combination of all four Abelian group
factors. However, we restrict ourselves to models in which the bulk U (1)b does not
contribute to the hypercharge, in order to avoid an unrealistically small gauge coupling.
Hence,
Y = k3 Qc + k2 QL + k1 Q1 .
(2.3)
87
The correct assignments for SM particles are reproduced, provided

1
k3 + k2 w = ,
6
2
k3 + a1 k1 = ,
3
1
k3 + b1 k1 = ,
3
1
k 2 + c1 k 1 = ,
2
k2 dL + d1 k1 = 1.
(2.4)
Notice that the second and third of the above equations imply that k1 = 0.
The next step, after assigning the correct hypercharge to the SM particles, is to check
for the existence of candidate fermion mass terms. Here, we discuss only the question of
masses for one generation (the heaviest) and we do not address the general problem of
flavor. To lowest order, the mass terms are of the form Qd c Hd , Quc Hu and Lec He where
Hd , Hu , He are scalar Higgs doublets with appropriate charges. In a generic model, there
are four different candidate Higgs scalar doublets (and their conjugates) H1 , . . . , H4 , with
U (1)L U (1)1 U (1)b charges:

{H1 , H2 , H3 , H4 } = (1, 1, 0), (1, 0, 1), (1, 1, 0), (1, 0, 1) .
(2.5)
It is easy to show that for any hypercharge embedding of the form (2.3) with k1 = 0,
there are at most three of the above Higgs doublets that have the correct hypercharge.
Depending on the parameters of the model, they can be reduced to two. For the generic
charge assignments of Table 1, the required Higgs charges are
Hu = (1, 2, 0, w, a1, a2 ),
Hd = (1, 2, 0, +w, +b1, +b2 ),
He = (1, 2, 0, 1 + dL , c1 + d1 , c2 + d2 ).
(2.6)
Provided the constraints (2.2) are satisfied, both Hu and Hd have the right charges of
(2.5) and correspond to strings stretched between the weak and one of the Abelian branes.
Thus, (2.2) guarantees the existence of tree-level quark masses. On the other hand, the
existence of He depends on the particular choice of parameters, e.g., for c1 + d1 = 2,
He does not exist and a tree-level lepton mass term (Lec H ) is forbidden. The generic
constraint that guarantees tree-level lepton masses is

|ci + di | = |1 + dL | = 1.
(2.7)
i=1,2
Given the smallness of the lepton mass compared to the masses of the quarks, of the
same generation, it would be reasonable to examine also the possibility that the lepton
mass is generated by a higher order term. The next order candidate lepton mass term
is of dimension six, proportional to M12 Lec H H H . The constraint in this case is more
s
complicated and the method we are going to use is the following: for each configuration
88
that satisfies all other constraints except (2.7), we derive explicitly the candidate Higgs
doublets and check the existence of possible fifth order mass terms.5
The hypercharge constraints (2.4) can be easily solved. They require a1 = b1 and
a1 + 2b1
k3 =
(2.8)
,
3(b1 a1 )
(a1 + b1 )
k2 =
(2.9)
,
2(b1 a1 )w
1
k1 =
(2.10)
,
b1 a1
b1 a1 (a1 + b1 )
+
,
c1 =
(2.11)
2
2w
(a1 + b1 )dL
.
d1 = b1 a1 +
(2.12)
2w
The allowed values of (a1 , b1 ) are {(1, 0), (1, 1), (0, 1), (0, 1), (1, 1), (1, 0)}. However, we notice that the solutions with parameters (a1 , b1 , c1 , d1 , k1 ) and (a1 , b1 , c1 ,
d1 , k1 ) are equivalent, since they correspond to a global change of sign Q1 Q1 .
Thus, it is sufficient to search for solutions with (a1 , b1 ) {(1, 0), (1, 1), (0, 1)}. Solving for these choices, we get three allowed hypercharge embeddings:
1
1
a1 = 1, b1 = 1: Y = Qc + Q1 ,
(2.13)
6
2
1
w
(ii) a1 = 1, b1 = 0: Y = Qc + QL + Q1 ,
(2.14)
3
2
2
w
(iii) a1 = 0, b1 = 1: Y = Qc QL + Q1 .
(2.15)
3
2
Case (i) leads to c1 = 1, c2 = 0, d1 = 2, d1 = dL = 0. This is a special solution where the
U (1)b brane decouples from the model since no SM particles are attached to it. It satisfies
(2.7) and thus leads to tree level lepton masses. The solution exists for both w = 1,
as the value of w does not play an important role when k2 = 0. In case (ii), we have
c1 = (1 + w)/2, dL = 0, d1 = 1 or c1 = (1 + w)/2, dL = 2w, d1 = d2 = 0, while case (iii)
leads to c1 = (w 1)/2, dL = 0, d1 = 1 or c1 = (1 + w)/2, dL = 2w, d1 = d2 = 0.
Combining the above three cases with the constraints (2.2) and (2.7), we get 9 distinct
configurations with tree-level quark and lepton masses, displayed in the upper part of
Table 2. Relaxing the constraint (2.7) with the requirement that lepton masses arise through
dimension six effective operators, leads to 6 more distinct models corresponding to the
cases 1015 of Table 2. In deriving these configurations, we have eliminated all models
connected to the ones above by the global charge redefinition Qb Qb .
As we mentioned before, in all the above configurations, we can define the baryon
number B as
1
B = Qc .
(2.16)
3
(i)
5 Here, we check only the conservation of all gauge quantum numbers. In the string context, there may be
additional selection rules for the non-vanishing of the corresponding couplings that are model dependent.
Table 2
Distinct models with lepton masses generated either at tree level Lec H (cases 19), or by dimension six effective operators Lec H H H /M 2 (cases 1015). We
also display the lepton number combination L (when it exists) and the number of Higgs doublets nh , needed to generate quark and lepton masses
a1
a2
b1
b2
c1
c2
d1
d2
dL
1Q + 1Q
6 c
2 1
1
3 Qc 12 QL + Q1
13 Qc + 12 QL + Q1
2Q 1Q + Q
1
3 c
2 L
2Q + 1Q + Q
1
3 c
2 L
1
1
3 Qc 2 QL + Q1
13 Qc 12 QL + Q1
2Q 1Q + Q
1
3 c
2 L
2Q 1Q + Q
1
3 c
2 L
10
11
12
14
1
1
2Q 1Q + Q
1
3 c
2 L
1Q + 1Q 1Q 1Q
2 c
2 L
2 1
2 b
12 Qc + 12 QL 12 Q1 12 Qb
13 Qc + 12 QL + Q1
2Q + 1Q + Q
1
3 c
2 L
1
1
3 Qc 2 QL + Q1
1Q + 1Q 1Q 1Q
2 c
2 L
2 1
2 b
13
nh
12 Qc + 12 QL 12 Q1 12 Qb
13 Qc 12 QL + Q1
2Q 1Q + Q
1
3 c
2 L
15
89
90
As we will argue below, U (1)c gauge invariance is broken by anomalies to a global

symmetry, implying baryon number conservation in type I string perturbation theory. Since
lepton number is also conserved at present energies, we can further examine which of the
above models possess also the lepton number L as a symmetry. In general, L can also be
expressed as a linear combination of all Abelian factors,

pi Qi
L=
(2.17)
i=c,L,1,b
that satisfies
pc + pL w = 0,
pc + a1 p1 + a2 pb = 0,
pc + b1 p1 + b2 pb = 0,
pL + c1 p1 + c2 pb = 1,
dL pL + d1 p1 + d2 pb = 1.
(2.18)
Inspection of (2.18), in conjunction with (2.4) that requires a1 = b1 , implies that lepton
number can only be defined for pb = 0, i.e., only in the presence of the bulk U (1)b .
This is of course expected, since the models without U (1)b have no lepton number [1].
Solving explicitly (2.18) for each one of the cases of Table 2, we find that only four
models, namely 2, 4, 6, 9, incorporate the lepton number as a (gauged) Abelian symmetry.
Its precise definition for each of these models is also presented in Table 2.
3. The weak angle and the string scale

We now come to the determination of the string scale consistent with the low energy
SM data. Following the hypercharge definition (2.3), the low energy data depend on the
couplings g3 , g2 and g1 of the three brane sets U (3)c , U (2)L and U (1)1 . These couplings
are in principle independent, but, as already explained in the introduction, in order to lower
the string scale we have to consider configurations where the U (1)1 brane is on top of
either the U (3)c or the U (2)L stacks. Hence, we have two possible coupling relations at
the string scale
(i):
g3 = g1
or (ii):
g2 = g1 .
(3.1)
In our normalizations, the hypercharge coupling gY at the string scale is expressed as

6k32 4k22 2k12
1
=
+ 2 + 2 .
gY2
g32
g2
g1
(3.2)
Following the one loop coupling evolution (i = gi2 /4),

1
bi
I Ms
1
=
+
ln
,
i (Ms ) i (MZ ) 4
MZ
(3.3)
91
where b3 = 7, b2 = 10/3 + nh /6, bY = 20/3 + nh /6 and nh is the number of scalar

Higgs doublets. The constant I corresponds to a model independent piece of the type I
string thresholds, entering in the identification of the string scale with the ultraviolet
cutoff
of the effective field theory. Its value was computed in Ref. [12] to be I = 1/ e
0.4, where is the Eulers constant. When the string scale is very high compared to
present energies, this represents a small correction compared to the dominant logarithmic
contribution coming from the renormalization group evolution. On the other hand, when
the string scale is low, such a correction becomes important, as it effectively changes
the string scale by roughly a factor of two, and should be taken with caution since it
is of the same order with the (unknown) model dependent part of threshold corrections.
Consequently, we will leave I as a parameter and discuss its possible effects on our
results case by case.
Solving the one-loop renormalization group equations (RGE) for the coupling evolution, the values of the weak mixing angle sin2 W and of the strong coupling a3 at the
Z-mass MZ are related to the couplings at the string scale:
sin2 W (MZ ) =
1
em (MZ ) (kY b2 bY ) I Ms
+
ln
1 + kY
2
(1 + kY )
MZ

1
1
em (MZ )
2
6k3
+
1 + kY
L (Ms ) 3 (Ms )

1
1
2
+ 2k1
,
L (Ms ) 1 (Ms )
1 b2 + bY b3 (1 + kY )
I Ms
1
1
1
log
=
a3 (MZ ) em (MZ ) 1 + kY
2
1 + kY
MZ

2
1
1
1
+
4k2 + 1
1 + kY
L (Ms ) 3 (Ms )

1
1
2
2k1
,
L (Ms ) 1 (Ms )
(3.4)
(3.5)
where kY = 6k32 + 4k22 + 2k12 and em is the electromagnetic coupling.

Given a coupling relation of (3.1), we can use Eqs. (3.4) and (3.5) to determine the string
scale Ms that correctly reproduces the low energy data. Clearly, the solution depends on
|k3 |, |k2 | and |k1 |. According to our previous analysis, there are three classes of models,
which correspond to the three possible hypercharge embeddings (2.13), (2.14) and (2.15):
(i):
(ii):
(iii):
1
|k3 | = ,
6
1
|k3 | = ,
3
2
|k3 | = ,
3
|k2 | = 0,
1
|k1 | = ,
2
1
|k2 | = ,
2
1
|k2 | = ,
2
|k1 | = 1,
|k1 | = 1.
(3.6)
Using (3.4) and (3.5), for each of the embeddings (3.6) and the unification conditions (3.1),
we computed the corresponding string unification scale MU I Ms . In our calculation
92
Table 3
The string unification scale MU and the two independent gauge couplings for the two possible brane
configurations and the various hypercharge embeddings
|k3 |
g1 = g3
g1 = g2
|k2 |
|k1 |
MU (TeV)
g2 (MU )/g3 (MU )
g2 (MU )g3 (MU )
1
6
1
3
2
3
1
2
4.6 1020
1.1
0.21
1
2
1
2
2.4 103
0.76
0.48
7.2
0.65
0.61
1
6
1
3
2
3
1
2
1.5 1022
1.1
0.26
1
2
1
2
0.32
0.57
0.73
we have used the following values for the low energy quantities a3 (MZ ) = 0.119,
sin2 W = 0.231, aem (MZ ) = 1/127.934. The results are presented in Table 3.
In the above calculations we have assumed that the number of doublets nh is the
minimum nh = 2 required by the model. Of course, one can consider models with more
doublets which can be for instance replicas of these two. It would be thus interesting to
examine the dependence of the above results on the number of doublets. To this end we
can extract analytic formulas regarding the unification scale MU . For the case g1 = g3 ,
taking for simplicity k1 = 1 and kL = 12 , we find

3(4 + 7k32 )
MU
1
1
log
1 2 sin2 W (MZ ) 2 1 + 3k32
=
MZ
em (MZ )
3 (MZ )
(3.7)
which implies that at the one-loop MU is independent of the number of doublets. Similarly,
for g1 = g2 and k1 = 1, kL = 12 , we have

50 + 126k32 nh
MU
1
1
log
1 4 sin2 W (MZ ) 6k32
,
=
6
MZ
em (MZ )
3 (MZ )
(3.8)
where we find a very weak dependence. Obviously, the number of doublets affects the
value of the weak gauge coupling at Ms and thus the volume of the bulk through (1.3).
4. The models
So far, we have classified all possible U (3)c U (2)L U (1)1 U (1)b brane models
that can successfully accommodate the SM spectrum. The quantum numbers of each
model as well as the hypercharge embedding are summarized in Table 2. Furthermore,
compatibility with type I string theory with string scale in the TeV region, requires the bulk
to be two-dimensional of (sub)millimeter size, and leads to two possible configurations:
place the U (1)1 brane on top of the weak U (2)L stack of branes or on top of the color U (3)c
branes. These impose two different brane coupling relations at the string (unification) scale:
g1 = g2 or g1 = g3 , respectively. For every model, using the hypercharge embedding
of Table 2, the one loop gauge coupling evolution and one of the above brane coupling
93
conditions, we can determine the unification (string) scale that reproduces the weak angle
at low energies. The results are summarized in Table 3.
According to the results of Section 2, there are three distinct hypercharge embeddings
that correspond to (|k3 |, |k2 |, |k1 |) = {(1/6, 0, 1/2), (1/3, 1/2, 1), (2/3, 1/2, 1)}. Since we
wish to restrict ourselves to models in which supersymmetry is broken at the string scale
(Ms ), we would like Ms to be low, at the TeV scale, to protect the mass hierarchy.
Thus, model 1 of Table 2, with hypercharge embedding (1/6, 0, 1/2), is rejected for
both U (1)1 brane arrangements, since the resulting string scale is too high. Furthermore,
models with hypercharge embedding (1/3, 1/2, 1) lead to Ms 103 TeV for g1 = g3 .
This scale, although much lower than the traditional GUT scale, is rather high for the
stabilization of hierarchy. On the contrary, for g1 = g2 , we find Ms O(1) TeV (when the
universal threshold correction I is taken into account) that lies at the edge of the present
experimental limits. The third embedding (2/3, 1/2, 1) reproduces successfully the low
energy data only for g1 = g3 and a string scale Ms O(10) TeV.
In all configurations of Table 2, the baryon number appears as a gauged Abelian
symmetry. This symmetry is broken due to mixed gauge and gravitational anomalies
leaving behind a global symmetry. Baryon number conservation is essential for low string
scale models, since one needs to eliminate effective operators to very high accuracy in
order to avoid fast proton decay, starting with dimension six operators of the form QQQL
which are not sufficiently suppressed [13]. In addition to baryon number, one should also
assure that the lepton number is a good symmetry of the low energy theory. Lepton number
conservation is also essential for preservation of acceptable neutrino masses, as it forbids
for instance the presence of the dimension 5 operator LLH H . Such an operator would
lead to large Majorana neutrino masses, of the order of a few GeV, in models where
the string scale, typically a few TeV, is too low for the operation of an effective sea-saw
mechanism. Hence, we shall be interested only in models in which the lepton number is
a good symmetry. Indeed, as seen in Table 2, only in four models, namely 2, 4, 6 and 9,
lepton number appears as a (gauged) Abelian symmetry. Being anomalous, this symmetry
will be broken, but lepton number will survive as a global symmetry of the effective theory.
In fact, these four models can be derived in a straightforward way by simple
considerations of the quantum numbers. The quark doublet Q is fixed by non-Abelian
gauge symmetries, while existence of baryon number implies that the anti-quarks uc , d c
correspond to strings stretched between the color branes and one each of the Abelian branes
U (1)1 and U (1)b . Thus, one has two possibilities leading to models that we call A (d c has
one end in the bulk) and B (uc sees the bulk). Existence of lepton number fixes the lepton
doublet as a string stretched between the weak branes and the U (1)b brane, while for each
of the models A and B there are two possibilities for the anti-lepton ec to emerge as a
string stretched between the two Abelian branes, or to have both ends on the weak branes.
Thus, we obtain two additional models that we call A and B . As it can also be seen in the
table, all these models have tree-level quark and lepton masses and make use of only two
Higgs doublets. They also require low energy string scale for some of the brane coupling
conditions. We now proceed to a detailed study of these four models and to an analysis of
their main phenomenological characteristics.
Notice from Table 3 that in both classes of models A and B, the coupling constant
ratio is g2 /g3 0.6 at the string scale, implying through the relations of Section 1 that
94
at least one of the internal compact dimensions along the world-volume of the weak set
of branes must be larger than the string length, by at least a factor of two (in the case of
two large dimensions, or by a factor of four in the case of one). The relevant experimental
signal would be the production of KaluzaKlein excitations for the W bosons and the
other mediators of the electroweak interactions but not of gluons, providing one of the first
indications of new physics [14].
4.1. Models A and A
We consider here the models 2 and 6 of Table 2, hereafter referred as models A and A ,
respectively. They are characterized by the common hypercharge embedding
1
1
Y = Qc QL + Q1
3
2
(4.1)
but they differ slightly in their spectra. The spectrum of model A is

Q(3, 2, +1, 1, 0, 0),
1, 1, 0, 1, 0),
uc (3,
1, 1, 0, 0, 1),
d c (3,
L(1, 2, 0, +1, 0, 1),
ec (1, 1, 0, 0, +1, +1),
Hu (1, 2, 0, +1, +1, 0),
Hd (1, 2, 0, 1, 0, 1),
while in model A the right-handed electron ec is replaced by an open string with both ends
on the weak brane stack, and thus ec = (1, 1, 0, 2, 0, 0). The two models are presented
pictorially in Fig. 1.
Apart from the hypercharge combination (4.1) all remaining Abelian factors are
anomalous. Indeed, for every Abelian generator QI , I = (c, L, 1, b), we can calculate
the mixed gauge anomaly KI J Tr QI TJ2 with J = SU(3), SU(2), Y , and gravitational
Fig. 1. Pictorial representation of models A, A .
95
anomaly KI 4 Tr QI for both models A and A :

0 1 1 1
12
1
3
13
2
12
1
6
53
43
3
K (A) = 2
3
2
0
3
K (A ) = 2
3
2
0
12
12
.
5
(4.2)
KK T
for both models have only one zero

It is easy to check that the matrices
eigenvalue corresponding to the hypercharge combination (4.1) and three non-vanishing
ones corresponding to the orthogonal U (1) anomalous combinations. In the context
of type I string theory, these anomalies are canceled by a generalized GreenSchwarz
mechanism which makes use of three axions that are shifted under the corresponding U (1)
anomalous gauge transformations [15]. As a result, the three extra gauge bosons become
massive, leaving behind the corresponding global symmetries unbroken in perturbation
theory [16]. The three extra U (1)s can be expressed in terms of known SM symmetries:
1
B = Qc ,
3
1
Lepton number: L = (Qc + QL Q1 Qb ),
2
1
PecceiQuinn:
QPQ = (Qc QL 3Q1 3Qb ).
(4.3)
2
Thus, our effective SM inherits baryon and lepton number as well as PecceiQuinn (PQ)
global symmetries from the anomaly cancellation mechanism. Note however that PQ is the
original PecceiQuinn symmetry only in model A , such that all fermions have charges
+1, while Hu and Hd have charges 2 and +2, respectively. In model A, the global
PQ symmetry defined in (4.3) is similar but with lepton charge +3. The reason is that
in model A the fermion-Higgs Yukawa couplings are different, and leptons get masses
from Hu and not from Hd .
The general one-loop string computation of the masses of anomalous U (1) gauge
bosons, as well as their localization properties in the internal compactified space, was
performed recently for generic orientifold vacua [17]. It was shown that orbifold sectors
preserving N = 1 supersymmetry yield four-dimensional (4d) contributions, localized in
the whole six-dimensional (6d) internal space, while N = 2 supersymmetric sectors give
6d contributions localized only in four internal dimensions. The later are related to 6d
anomalies. Thus, even U (1)s which are apparently anomaly free may acquire non-zero
masses at the one-loop level, as a consequence of 6d anomalies. These results have the
following implications in our case:
Baryon number:
(1) The two U (1) combinations, orthogonal to the hypercharge and localized on the strong
and weak D-brane sets, acquire in general masses of the order of the string scale from
contributions of N = 1 sectors, in agreement with effective field theory expectations
based on 4d anomalies.
(2) Such contributions are not sufficient though to make heavy the third U (1) propagating
in the bulk, since the resulting mass terms are localized and suppressed by the
96
volume of the bulk. In order to give string scale mass, one needs instead N = 2
contributions associated to 6d anomalies along the two large bulk directions. In our
models such contributions are indeed in general present and arise from mixed 6d
gauge-gravitational anomalies of two different sources: (i) the generic presence of a
neutral 6d Weyl fermion on the bulk brane which coincides either with the U (1)b
gaugino (in the supersymmetric case) or the goldstino of the non-linearly realized
supersymmetry (in the brane SUSY breaking case [7]); (ii) the contribution of the
right-handed neutrino which arises from a six-dimensional Weyl spinor. As a result,
the third Abelian gauge field U (1)b acquires also a mass of the order of the string scale,
although its gauge coupling is tiny due to the volume suppression (see Eq. (1.5)).
(3) Special care is needed to guarantee that the hypercharge remains massless despite the
fact that it is anomaly free, along the lines of Ref. [17].
The presence of massive gauge bosons associated to anomalous Abelian gauge symmetries
is generic. Their mass is given by MA2 gs Ms2 , up to a numerical model dependent factor
and is somewhat smaller that the string scale. When the latter is low, they can affect low
energy measurable data, such as g 2 for leptons [18] and the -parameter [19], leading
to additional bounds on the string scale.
Note that the global PQ symmetry leftover from U (1)b is spontaneously broken by the
Higgs expectation value giving rise to an unwanted electroweak axion. A possible way out
was suggested in Ref. [1], using an appropriate departure away from the orientifold point.
A plausible extension of the model is the introduction of a right-handed neutrino in the
bulk. A natural candidate state would be an open string ending on the U (1)b brane. Its
charge is then fixed to +2 by the requirement of existence of the single possible neutrino
mass term LHd R . The suppression of the brane-bulk couplings due to the wave function
of R would thus provide a natural explanation for the smallness of neutrino masses. Note
that if the zero mode of this bulk neutrino state is chiral, the anomaly structure of the model
changes: B L becomes anomaly free and as a consequence the associated gauge boson
remains in principle massless. However, as we discussed above, this is not in general true
because of 6d anomalies [17]. In any case, this problem is absent if we introduce a vectorlike bulk neutrino pair
R (1, 1, 0, 0, 0, +2) + Rc (1, 1, 0, 0, 0, 2)
that leaves the anomalies (4.2) intact. Note that Rc does not play any role in the subsequent
discussion of neutrino masses and oscillations.
Coming to the issue of gauge couplings and the string scale, as already explained
we have two different realizations for each model. The first is with g1 = g3 at Ms that
corresponds to a configuration where the U (1)1 brane is placed on top of the color branes.
According to Table 3, this leads to an intermediate string scale Ms 106 GeV, which
appears too high to guarantee the stabilization of hierarchy. The second possibility is to take
the U (1)1 brane on top of the weak branes, leading to g1 = g2 . The required string scale is
now low Ms O(500) GeV (300800 GeV, depending on the threshold corrections), and
could account for the stability of the hierarchy.
97
4.2. Models B and B

Another phenomenologically promising pair of models consists of solutions 4 and 9 of
Table 2, named hereafter B and B , which corresponds to the hypercharge embedding
2
1
Y = Qc QL + Q1 .
3
2
The spectrum is
(4.4)
Q(3, 2, +1, +1, 0, 0),

1, 1, 0, 0, 1),
uc (3,
1, 1, 0, 1, 0),
d c (3,
L(1, 2, 0, +1, 0, 1),
ec (1, 1, 0, 0, +1, +1),
Hu (1, 2, 0, 1, 0, 1),
Hd (1, 2, 0, +1, +1, 0),
for model B, while in B ec is replaced by ec (1, 1, 0, 2, 0, 0). The two models are
represented pictorially in Fig. 2. The four Abelian gauge factors are anomalous. Proceeding
as in the analysis (4.2) of models A and A , the mixed gauge and gravitational anomalies
are
0
0
1
1
1
1 1
1
3
K (B) = 2
3
2
2
3
4
3
12
11
6
3
K (B ) = 2
3
2
43
1
3
12
.
5
(4.5)
It is easy to see that the only anomaly free combination is the hypercharge (4.4) which
survives at low energies. All other Abelian gauge factors are anomalous and will be broken
by the generalized GreenSchwarz anomaly cancellation mechanism, leaving behind
global symmetries. They can be expressed in terms of the usual SM global symmetries
Fig. 2. Pictorial representation of models B and B .
98
as the following U (1) combinations:

1
B = Qc ,
(4.6)
3
1
Lepton number: L = (Qc QL + Q1 + Qb ),
(4.7)
2
1
PecceiQuinn:
QPQ = (Qc + 3QL + Q1 + Qb ).
(4.8)
2
Similarly to the analysis of models A and A , the PQ charges defined above are the
traditional ones only for model B. In model B , the lepton charge is 3, as a result of
the HiggsYukawa couplings to the fermions (see below). The right handed neutrino can
also be accommodated as an open string with both ends on the bulk Abelian brane:
Baryon number:
R (1, 1, 0, 0, 0, +2) + Rc (1, 1, 0, 0, 0, 2).
(4.9)
According to the RGE running results of Table 3, there is only one brane configuration,
for the models under discussion, that reproduces the weak mixing angle at low energies.
This consists of placing the U (1)1 brane on the top of the color branes, so that g1 = g3 ,
which leads to Ms O(10) TeV (717 TeV, depending on the threshold corrections).
5. Fermion masses
Although the general question of quark and lepton masses goes beyond the scope of
this paper, we would like to make here some comments in the context of our constructions.
The Yukawa couplings relevant to fermion masses are constrained by the various U (1)
symmetries and can present interesting patterns.
Model A. The relevant Yukawa couplings are
MA = u Quc Hu + d Qd c Hd + e Lec Hu + LHd R .
(5.1)
Here, charged leptons and up quarks (of the heaviest generation) obtain masses from the
same Higgs (Hu ).
When all Yukawa couplings arise at the lowest (disk) order, it is easy to check that in the
simplest case (absence of discrete selection rules, etc), they satisfy the following relations:
d = 2gs ,
= 2 gb .
u = e = 2 g2 ,
(5.2)
The top and bottom quark masses are given by
mt = g2 v sin ,
mb = g s v cos ,
(5.3)
expectation values (VEVs) of the two

where tan = vu /vd , with vu and vd the vacuum

higgses Hu and Hd , respectively, and v = vu2 + vd2 = 246 GeV. Note that in the case
where the color branes are identified with D3 branes, one has gs = g3 , and in any case
gs g32 . Note also that since the string scale in this model is relatively low, Ms 1 TeV,
there is no much evolution of the low energy couplings from the electroweak to the string
scale. Thus, using the known value of the bottom mass mb 4 GeV, one obtains for the
99
exp
top quark mass mt 162 GeV which is less than 5% below its experimental value mt =
174.3 5.1 GeV. In addition, the Higgs VEV ratio turns out to be large, tan 100. Note
that such a large value is not in principle problematic as in the supersymmetric case, but it
can lead to important higher order corrections.
On the other hand the -mass is of the same order as the top mass, which is unrealistic.
However, there is still the possibility that the lepton Yukawa coupling e vanishes to
lowest order due to additional string discrete selection rules, and is generated by a higherdimensional operator of the form Lec (Hu H H ) providing the appropriate suppression.6
Model A . The Yukawa couplings here are
MA = u Quc Hu + d Qd c Hd + e Lec Hd + LHd R
(5.4)
with the same relation for the tree-level couplings as in (5.2). Using the parametrization in
(5.3) we see that the relation of mt to mb is the same as in model A and the same remarks
apply. Since here the lepton and down quark acquire their masses from the same Higgs,
one obtains the phenomenologically interesting relation: mb /m = gs /g2 = g3 /g2 , when

strong interactions are on D3-branes. Thus, from Table 3, mb /m 1.75 at the (string)
unification scale, which is in the upper edge of the experimentally allowed region at
the Z-mass, 1.46 mb /m |exp 1.75. This relation could replace the successful GUT
prediction mb = m of the conventional unification framework, in low scale string models.
In conclusion model A seems to be able to generate the required hierarchy of masses for
the third generation.
Model B. The relevant trilinear Yukawa couplings are,
MB = u Quc Hu + d Qd c Hd + e Lec Hd + LHu R .
The tree-level Yukawa couplings satisfy

d = 2 g3 ,
e = u = 2gs ,
and we have
mt = g s v sin ,
mb = g3 v cos .
2 gb
(5.5)
(5.6)
(5.7)
The first relation implies again a heavy top, while the bottom to tau mass ratio is
now predicted, with a value mb /m = g3 / gs 1 which is apparently far from its

experimental value. However, in this case, the string scale is relatively high and therefore
one should take into account the renormalization group evolution above the weak scale.
Solving the associated RGEs with the boundary conditions (5.7) and assuming g3 = gs ,
3
we obtain acceptable mb and m masses for Ms 3 10 TeV and tan 80. Note that
the successful prediction of mb and m is related to the condition mb = m at the (string)
unification scale, which in the case of non-supersymmetric Standard Model is obtained at
relatively low energies [20]. Indeed, in Fig. 3, we plot the mass ratio mb /m as a function
of the energy, within the non-supersymmetric Standard Model with two Higgs doublets.
Nevertheless, the resulting value of Ms is still significantly higher than the unification
6 Models with similar properties have been considered in the past in the perturbative heterotic string
framework.
100
Fig. 3. Evolution of the ratio mb /m as a function of the energy for tan = 2 and tan = 80. We have used as
low energy parameters mb = 4 GeV, mtop = 174 GeV, a3 (Mz ) = 0.12, sin2 W = 0.23113.
scale required from the analysis of gauge couplings in Section 3. Moreover, the top quark
mass turns out to be rather high, mt 220 GeV. It is an open question whether this
discrepancy can be attributed to threshold corrections that can be important in the case
of two-dimensional bulk [6].
Model B . The relevant Higgs couplings are given by
MB = u Quc Hu + d Qd c Hd + e Lec Hu + LHu R
while the tree-level Yukawa couplings by

u = 2gs ,
d = 2 g3 ,
= 2 gb
and e =
(5.8)
2 g2 .
(5.9)
Here, as in model A, the and top mass are of the same order and thus in conflict with
experiment. As in model A, vanishing leading order coupling could be a way out.
In the above analysis we have also assumed that only the heaviest generation acquires
masses at the lowest order. The other two are considered to have vanishing trilinear Yukawa
couplings. This property does not follow from the gauge symmetries we considered and
should be attributed either to discrete string symmetries or to additional gauge symmetries
by enlarging the model.6
6. Neutrino physics
One of the challenges of Standard Model extensions is the justification of the smallness
of neutrino masses. The favorite scenario used to rely upon the introduction of right-handed
neutrinos (SM singlets) and their mixing with some extra massive singlets. The suppression
of the neutrino masses is then obtained as a result of the structure of the full mass matrix
(see-saw mechanism). In order for this mechanism to work effectively, the extra singlet
mass should be about ten orders of magnitude higher than the electroweak scale.
101
Among the promising features of D-brane models is a novel scenario to account for
neutrino masses: right-handed neutrinos are assumed to propagate in the bulk while lefthanded neutrinos, being a part of the lepton doublet, live on the brane. As a result, the
Dirac neutrino mass is naturally suppressed by the bulk volume. Adjusting this volume, so
that the string scale lies in the TeV range, leads to tiny neutrino masses compatible with
current experimental data.
The extra-dimensional neutrino mass suppression mechanism described above can be
destabilized by the presence of a large Majorana neutrino mass term. As already mentioned
in Section 3, the lepton-number violating dimension five effective operator LLH H leads,
in the case of TeV string scale models, to a Majorana mass term of the order of a few GeV.
Even if we manage to eliminate this operator in some particular model, higher order
operators would also give unacceptably large contributions, as we focus on models in
which the ratio between the Higgs vacuum expectation value (VEV) and the string scale is
just of order O(1/10). The best way to protect tiny neutrino masses from such contributions
is to impose lepton number conservation. As we have seen in Section 2, we can find
models which successfully accommodate all SM particles and preserve lepton number as
an effective global symmetry in perturbation theory. These are the models A, A and B, B
described in detail in Section 3.
Apart from neutrino masses these theories contain also the ingredients to explain
neutrino oscillations. The right-handed neutrino, being a bulk state, has a tower of Kaluza
Klein (KK) excitations. Their mixing with the ordinary (left-handed) neutrino leads to
oscillation patterns that have to be compared with present solar and atmospheric neutrino
data. There exist extended discussions in the literature [5] regarding the neutrino mass
and oscillation problems in the context of extra-dimensional theories. Among the common
results of these works is that an explanation of the solar neutrino anomaly is possible
provided the small mixing angle (SMA) solution is acceptable. However, recent SNO
results in conjunction with SuperKamiokande data [2124] strongly disfavor the SMA
solution and thus render this higher-dimensional oscillation mechanism problematic,
at least as far as solar neutrino oscillations are concerned. A possible way out is to
introduce three bulk neutrinos and explain the oscillations in the traditional way [25].
The effect of the KK mixing can be eliminated by appropriately decreasing the size
of the extra dimensions and thus increasing the value of the string scale. However, all
these discussions are restricted to the case of effectively one-dimensional bulk. Besides
these phenomenological difficulties, there is also a serious theoretical problem, since onedimensional propagation of massless bulk states gives rise to linearly growing fluctuations
which yield in general large corrections to all couplings of the effective field theory,
destabilizing the hierarchy [6].
Two-dimensional scenarios have not been considered in detail. We will see below
how the above problems can be resolved and discover that a two-dimensional bulk has
enough structure to describe both the solar and atmospheric neutrino oscillations by
introducing a single bulk neutrino pair. On the other hand, recent experiments are also able
to differentiate between the contributions of active and sterile neutrinos to the neutrino
anomaly problems. From this point of view, the KK excitations do not carry any Standard
Model charges and are thus considered as sterile. It is then important to examine if all these
constraints are compatible with our model.
102
As explained in the introduction, our setup incorporates a two-dimensional bulk and we

are going to assume hereby that neutrinos propagate in the full bulk volume which is a twodimensional space. Among the common features of the models considered in Section 3, one
finds tree-level neutrino couplings and mass-terms of the form:
3

i Li Hi R
i=1
3

i vi iL R ,
(6.1)
i=1
where i is a generation index and for each generation i, Hi is one of the available Higgs
doublets Hd or Hu , providing masses to down quarks (models A, A ) or to up quarks
(models B, B ), respectively, with vi = Hi the corresponding VEV. The above couplings
provide a mass to one linear combination (L ) of the weak eigenstates (iL ), while the
other two remain massless. Note, that it would be possible to generate masses for all lefthanded neutrinos by introducing additional bulk neutrino pairs. In this case the number of
free parameters is increased and predictability is lost. Thus, here, we will study the case of
) the mass eigenstates, the weak
a single bulk neutrino pair. Defining NL = (L , 0L , 0L
eigenstates can be written as

Uij NLj ,
iL =
(6.2)
j

where U is a 3 3 unitary matrix with Uj1 = (U 1 )1j = j vj /mD and m2D = 3i=1 2i vi2
is the mass-square of the massive combination (L ). Being of brane-bulk type, the
couplings i are naturally suppressed by the bulk volume v89 (see Section 1) and lead
to a tiny Dirac neutrino mass
v
2 2 Ms
mD =
(6.3)
=
v,
v89 g3 g2 MP

3
2 2
where v =
i=1 hi vi with hi , i = 1, 2, 3, the associated dimensionless Yukawa
couplings and vi the corresponding Higgs VEV (v = Hd , Hu ) depending on the model.
Using typical values for the gauge couplings (see Table 3 of Section 3), vi < v = 246 GeV
and hi /4 = O(1), we obtain mD < 6 103 eV for Ms 10 TeV. This provides an
explanation for the smallness of neutrino masses and is actually the extra-dimensional
version of the see-saw mechanism.
The above picture is simplified because we have neglected the contributions of the tower
of KK neutrino states. Taking them into account, and assuming for simplicity that the two
bulk radii are equal R8 = R9 = R and form an angle /2 , where /2 < < /2, the
mass terms become

m2 /M 2 (k)

(k)
Lm = mD L
(6.4)
k
R +
mk Rc (k) R + c.c.,
k
k
(0)
where R = R and the summation over k extends over all KK momenta. By m2 we
k
denote the mass-square of the KK excitation labeled by momenta k = (k1 , k2 )
m2k =
2

1
k + k22 2k1 k2 sin .
R 2 cos2 1
(6.5)
103
We also use the notation m2k for the mass-square of the kth KK level. is a model dependent
constant bigger than one, associated to the coupling of two NeumannDirichlet (ND) Z2 twisted strings to an untwisted (NN or DD) string [26]. In our models, there are four Z2 twisted coordinates, implying = 16. M plays the role of an ultraviolet (UV) cutoff, which
is normally the string scale Ms , but we prefer using the symbol M because in certain
processes there exists an induced cutoff that can be a few orders of magnitude below Ms .
For instance, this is the case of solar neutrinos, where the production energy is of order of
a few MeV, and thus heavier KK modes are effectively cut off.
The mass terms (6.4) lead to a mixture of the usual left-handed neutrino with the
infinite tower of its KK excitations. A detailed analysis of the eigenstate problem in our
framework is presented in Appendix A where we derive the basic formulas for neutrino
masses and transition probabilities. Due to its complexity, the problem can be either
treated numerically in the general case or analytically using some approximation. The first
approach has the disadvantage of being rather tedious as it involves summations over a
very large number of KK modes, so we will adopt here an analytic perturbative approach.
Concerning the interpretation of neutrino anomalies, there are also two possible treatments:
the first is a direct fit of neutrino data to the transition probability formulas obtained in
our framework. The second is to try to simulate the standard solutions to the solar and
atmospheric neutrino anomaly problems. We will use here the second method, as it is
sufficient for demonstrating the basic features of our model.
Following Appendix A, the mass spectrum of the full system, in the case of two bulk
dimensions, is
m
2k = m2k + rk m2D 2mk /M (1 k ) + ,
2
(6.6)
where rk is the multiplicity of the kth KK level and

k = m2D R 2 cos log M 2 R 2 cos log 2 + sk ,
(6.7)
with sk a volume independent constant. Our solution is based on the assumptions that
mD R 1, as justified by (6.3) and n < 1 that simplifies the formulas involved. Under
these assumptions, and following the analysis in Appendix A, the survival probability for
a neutrino of flavor i is given by

2 m20 L
2
2
2
2 L
3.2ui 0 sin
,
Pi i 1 4ui 1 ui sin
(6.8)
4 E
R 2 4E

where m20 = m2D (1 0 ) and ui = |Ui1 | satisfying the unitarity relation i u2i = 1;
L is the distance that the neutrino travels before being detected and E is the beam energy.
According to the discussion in Appendix A, the survival probability takes this form only
for specific values of the angle :
sin =
p
,
q
p, q Z, |p| < q,
q(3 + (1)q )
,
2(q 2 p2 )
(6.9)
where p, q are relatively prime integers and q = 1 for p = 0. In our approximation, the
m2 L
D
survival probability (6.8) is a superposition of two modes with frequencies: 4E
and
L
.
These
two
frequencies
can
be
considered
as
independent
parameters,
as
the
first
4R 2 E
104
depends on the Yukawa couplings and the Higgs VEVs, while the second depends on
the compactification radius or equivalently on the string scale. The existence of these two
frequencies provides us with the opportunity to fit both solar and atmospheric oscillations
using (6.8). Furthermore, the amplitudes of the two modes depend on ui and 0 defined in
(6.7). These parameters can be used in order to fit the oscillation amplitudes.
In the standard neutrino (two flavor) scenario, one usually explains the solar neutrino
anomaly by e oscillations and the atmospheric neutrino deficit by
oscillations. The formula for the transition probability is:

>m2ij L
,
Pi j = sin2 2ij sin2
(6.10)
4 E
where >m2ij = m2i m2j is the neutrino mass difference in the case of two states mixing.
L
Expressing L in km, E in GeV and >m2 in eV2 , the frequency >m2ij 4E
takes the form
2
1.27 >mij L/E.
Recent analysis of atmospheric neutrino data [23] at 3 c.l. gives 1 103 < >m2atm <
6 103 eV2 and 0.7 < sin2 2atm < 1. Regarding solar data, the situation has dramatically
changed after the latest SNO results: only the LMA and LOW MSW solutions are
acceptable at the 3 c.l. with 2.3105 < >m2LMA < 3.7104 eV2 , 0.6 < sin2 2 LMA <
1 and 3.5 108 < >m2LOW < 1.2 107 eV2 , 0.8 < sin2 2 LOW < 1. Moreover, the
LMA gives a much better fit. The region of the SMA solution (with best fit values
>m2SMA 5 106 eV2 , sin2 2 SMA 2 102 ) is acceptable only at the 5.5 level
and is thus practically excluded [22].
The atmospheric neutrino oscillation frequency is higher than solar solutions, >m2atm >
>m2sol , and thus we have to use the lowest frequency in (6.8) (i.e., m20 ) to simulate
solar neutrino oscillations. Formula (6.8) contains four independent parameters, namely
mD , R, Ms and ue (assuming u = 0 and thus u2e + u2 = 1). Fitting both solar and
atmospheric oscillations requires to leading order in 0 :
(6.11)
= >m2atm ,
R2
m2D = >m2sol ,
(6.12)

2
2
2
4ue 1 ue = sin 2sol,
(6.13)
3.2u2 0 = sin2 2atm .
(6.14)
Neglecting the constant term s0 in the expression (6.7) of 0 , in the limit MR 1, and
assuming = 16, 0 can be written in terms of Ms and mD as

2 2
log
1 m2D MP2
MP
0
(6.15)
, a=
log
,
aMs
2
g3 g2
2a 2 Ms4
where we have assumed that the cutoff is equal to the string scale (M = Ms ). This choice
of the cutoff is suitable for the atmospheric neutrino data, where the oscillation amplitude
is proportional to 0 . In any case, the exact value of the cutoff plays a minor role in
our calculation, due to the fact that it appears always logarithmically. Furthermore, the
expectation value v is related to the rest of the parameters through Eq. (6.3), while the
105
angle enters in the Plank mass definition (1.3)

R 2 cos =
1 MP2
.
4 2 a 2 Ms4
(6.16)
In terms of the integers p, q that have been introduced in (6.9), we can rewrite the last
equation as
3 + (1)q
>m2atm MP2
Wp,q
=
,
4 2 a 2 Ms4
2 q 2 p2
(6.17)
or equivalently
cos =
3 + (1)q
.
2Wp,q q
(6.18)
Thus, the four conditions (6.11)(6.14) together with (6.15) and (6.3), (6.16) fix all four
parameters of the model. Therefore, fitting the atmospheric neutrino frequency (6.11), one
determines the compactification radius
(6.19)
< 6 103 eV2 ,
R2
or 3 < R < 6 m for 1. Choosing for the solar neutrino deficit the preferred LMA
solution, we get from the second condition (6.12) the neutrino mass range
1 103 <
4.8 103 < mD < 7.7 102 eV.
(6.20)
The third condition (6.13) fixes the mixing coefficient u2e and has two possible solutions,
namely, 0.18 < u2e < 0.5 or 0.5 < u2e < 0.82. Choosing u2e 0.18 and u2 0.82 (u = 0),
Eq. (6.14) leads to 0 0.27 (in the case we choose the lowest allowed value of
sin2 2atm ), which lies at the edge of the validity of our perturbative approach. Any other
choice of ui compatible with the constraints leads to bigger values for 0 . This justifies
also the choice u = 0 in order to minimize 0 in (6.14).7 From (6.14) we get the string
scale
8 Ms 13 TeV,
(6.21)
while compatibility with (6.3) requires O(1) values for the Yukawa couplings. It is
interesting that this range for the string scale coincides with the values we found from
the analysis of gauge couplings in Section 3, for the models B and B . Coming to the
angle, we get from (6.17) 0.02 Wp,q 0.2 for the allowed range of >m2atm and we can
easily verify that there exist integers p, q that satisfy (6.17).
Let us now consider the LOW solution to the solar neutrino deficit. Following similar
steps, the four constraints (6.11)(6.14) in this case give
1.9 104 < mD < 3.5 103 eV,
(6.22)
7 Normally, one should repeat the eigenstate analysis of Appendix A numerically in the non-perturbative
region, but from a preliminary analysis we do not expect significant change of our results.
106
with u2e 0.28 and thus u2 0.72 and 0 0.30. The string scale turns out to be slightly
lower in this case:
1.8 Ms 2.2 TeV,
(6.23)
while for the angle we get 20 Wp,q 200. Note that the range of string scale is now
compatible with the values found from the analysis of gauge couplings in models A and
A . In this solution, the left-handed neutrino Yukawa couplings do not have to be of O(1).
Moreover, the practically excluded SMA solution can also be obtained in this
framework. The associated parameters in this case are: u2e 5 103 , u2 = 1 u2e 1,
0 0.2, mD 2 103 eV, Ms 6 TeV, 0.2 Wp,q 1.2. Note that the case = 0
corresponds to p = 0, q = 1 and thus Wp,q = 1. As seen from our results, only the SMA
solution includes this value in the allowed W -range and this is the reason why only this
solution could be reproduced in the case of an orthogonal torus. The LMA and LOW
solutions require a bulk forming a non-orthogonal lattice, corresponding to non-trivial
values of . It is also worth noticing that such non-trivial values of induce CP violation
in the neutrino sector, which is interesting to be further explored.
The mixing of the neutrino zero mode with its KK excitations can lead to a decay of the
left-handed neutrino to these KK modes, considered as sterile from the SM point of view.
In our framework, and to leading order in the 0 expansion, the average conversion rate of
a neutrino of flavor i to sterile is given by (A.19)
Pi s 2u2i >0 .
(6.24)
Constraints (6.12) and (6.14) fix both the above probabilities. Assuming the LMA solution
to the solar neutrino deficit, we get
P s 0.44
(6.25)
for atmospheric and

Pe s 0.05
(6.26)
for solar neutrinos, where in the second case we have assumed a cutoff M 50 MeV. For
the LOW solution, the transition probabilities are similar: P s 0.32, Pe s 0.08.
Note that the decay rate to sterile neutrinos is significant in the case of atmospheric
neutrinos and is negligible in the case of solar neutrinos. This is related to the structure
of our model for neutrino oscillations. The atmospheric neutrino deficit is simulated using
the lightest KK neutrino excitation (which is interpreted from the SM point of view as a
sterile neutrino), while the solar data are explained using the (active form the SM point of
view) zero mode.
Constraints for the conversion of active to sterile neutrinos have been recently examined
in reference [25]. Following their analysis in the case of the LMA solution, the constraint to
the average decay rates for solar neutrinos is Pe s < 0.40 at 90% c.l. which is obviously
satisfied by our model. For atmospheric neutrinos the relative constraint takes the form
>P = P s Pe s < 0.17.
(6.27)
107
Evaluating this constraint in our framework, one finds >P = 0.44 0.10 = 0.34 (where
Pe s = 0.10 in the case of atmospheric due to the higher cutoff in 0 of Eq. (6.7)) which
is by a factor of two higher than the experimental bound. However, one should take into
consideration that our perturbative analysis, focusing on explicitly revealing oscillations,
does not allow to access the region 0 1 where in principle the above rates could
change. As mentioned earlier this region could be studied only numerically. This requires
summation over a huge number of KK modes and at present it appears insoluble even
numerically.
In any case the exact nature of atmospheric neutrino oscillations is expected to be
further examined in the K2K [27] experiment. In case the predictive scenario of a single
bulk neutrino presented here fails to satisfy the sterile production constraints, one should
proceed in the introduction of additional bulk neutrinos and explain oscillations in the
traditional way, that is by zero mode mass difference and not by mixing with the KKs.
Their presence can still lead to sterile production which can be reduced by appropriately
raising the string scale and thus decoupling the KKs [25].
7. Summary and conclusions

In conclusion, we performed a systematic study of the Standard Model embedding
in type I string theory at the TeV scale. We found that the minimum configuration with
interesting phenomenological features requires three sets of D-branes, so that all SM
particles are obtained as open strings stretched among these brane stacks. Two of them
describe respectively the strong and weak interactions, while the third one contains a single
Abelian brane that extends in a two-dimensional bulk of submillimeter size.
The model predicts the correct value of the weak angle for a string scale of a few TeV. It
also contains baryon and lepton number as perturbative global symmetries, ensuring proton
stability and absence of large (Majorana) neutrino masses. On the other hand, it uses two
Higgs doublets that can provide masses to all quarks and leptons. Concentrating on the
heaviest generation, we computed all trilinear Yukawa couplings and studied the resulting
mass relations. We found a naturally heavy top and the mass ratio of bottom quark to tau
lepton close to its experimental value.
Finally, we have studied neutrino masses and oscillations by introducing a single righthanded neutrino state in the bulk. We found that both solar and atmospheric neutrino data
can be explained if the bulk is a non-orthogonal torus forming a non-trivial angle. Solar
oscillations are then explained using the zero-mode, which obtains a tiny mass from the
electroweak Higgs, while atmospheric oscillations use its first KK excitation. However, in
the cases of atmospheric data, it seems to be an excess in sterile production with respect to
current atmospheric data analyses.
Overall, the model looks very promising and deserves further investigation. Particular
directions that have not been discussed are the masses and mixing angles of the two lightest
generations, possible important threshold corrections related to the two-dimensional bulk,
supersymmetry breaking effects in models with brane supersymmetry breaking, as well as
explicit type I string realizations.
108
Acknowledgements
This work was partly supported by the European Union under the RTN contracts HPRNCT-2000-00148, HPRN-CT-2000-00122, HPRN-CT-2000-00131, HPRN-CT-2000-00152,
HPMF-CT-2002-01898 and the INTAS contract No. 99 1 590. E.K., J.R. and T.N.T. would
like to thank the Theory Division of CERN, J.R. the CPHT of Ecole Polytechnique and
T.N.T. the L.P.T. of the Ecole Normale Suprieure for their hospitality. I.A. would like to
thank Guido Altarelli for enlightening discussions.
Appendix A. Neutrino masses and oscillations

We consider the neutrino mass eigenvalue problem that arises when the usual lefthanded neutrino localized on the brane mixes with one pair of right-handed neutrinos
propagating in a two-dimensional bulk. The solution of this problem in the case of onedimensional bulk has already been studied in the literature [5]. A new feature of the
two-dimensional bulk is that the associated KK sums are divergent and a mass scale M
playing the role of the UV cutoff, normally identified with Ms , appears in the mass and
eigenstate expressions. Moreover, we consider neutrino oscillations and derive formulas
for the transition rates of both active and sterile neutrinos.
We will assume for simplicity that the two bulk radii are equal, R8 = R9 = R, but we
will allow for the possibility that the angle between the two compactified directions
is arbitrary /2 < < /2. The masses of the KK excitations, labelled by momenta
n = (n1 , n2 ), are:
m2n =
2

1
n + n22 2n1 n2 sin .
R 2 cos2 1
(A.1)
The KK modes can be ordered according to their mass and labelled by a unique level
number k. Massive levels have in general degeneracy four, apart from particular points that
have higher degeneracy for special values of . In any case, only the direct sum of the states
of each degenerate level couples to the left-handed neutrino. Hence, we can diagonalize in
the degenerate subspace and choose one of the eigenstates, which corresponds to the sum
of the degenerate KK modes. In this basis, the relevant neutrino mass terms take the form
Lm = mD L
mk 2 (k)
c(k) (k)
rk M 2 R +
mk R R + c.c. + decoupled,
k
where R(k) =
(@)
1
@ R
rk
and Rc(k) =
1
rk
c (@)
@ R ,
@ = 1, . . . , rk , with rk the multiplicity
of the KK level with mass mk . The mass terms can be written in matrix form (NLT mNR +
c.c.) with NL = (L , Rc(1) , . . .), NR = (R(0) , Rc(1) , . . .) and m an infinite-dimensional
matrix.
In order to determine the left-handed neutrino mass eigenstates, we consider
m2
m2D A
m2
m1 mD r1 M12
m2
22
M
m
m
r
2 D
2
mm =
..
m2
r m m Mk1
2
k1 k1 D
m2
109
m2
m1 mD r1 M 2
m2 mD r2 M 2
mk mD rk M 2
m21
m22
..
.
..
.
..
.
..
.
m2k
rk mk mD M 2
(A.2)

2m2@ /M 2
where A = @ r@
. In the sequel we will assume that all masses are measured in
string units, which we restore only at the end of our calculations.
The exact eigenvalue equation for the mass of the nth KK level m
n can be written in the
form
m2D
r@ 2m2@
=1
m
2n m2@
@
(A.3)
and the associated eigenstates

Ln =

1 n n
1, c1 , c2 , . . .
Nn
(A.4)
c@n =
mD m@ m2
@
m
2n m2@
(A.5)
with
and
Nn2
= m2D m
2n

@
r@ 2m@
.
(m
2n m2@ )2
2
(A.6)
The above results can be used to express L in the basis of the mass eigenstates
L =
1
n
Nn L
n
and calculate its time evolution

2
1
im
nL n
L ,
L (t) =
exp
Nn
2E
n
(A.7)
(A.8)
110
where E is the neutrino beam energy and L is the distance from the source. Therefore,
using (6.2) we can derive the time evolution of the weak eigenstates

.
iL (t) = Ui1 L (t) + Ui2 0L + Ui3 0L
(A.9)
The transition rate Pi j , that gives the probability for a neutrino of a specific flavor i
produced in the source to be detected as flavor j in the detector, is

1 u2 + u2 T 2 , i = j,

2
i
i

Pi j = iL (0)|j L (t) =
(A.10)
i = j,
u2i u2j |1 T |2 ,
where ui = |Ui1 | and
2
1
im
nL
.
T
exp
2
N
2E
n
n
(A.11)
The formulas for the transition probabilities to active neutrinos are

2

1 u2i + u2i 1 u2i T + T + u4i |T |2 , i = j,
Pi j =

u2i u2j 1 T + T + |T |2 ,
i = j.
(A.12)
Therefore, the transition rate for a neutrino of flavor i to decay into a sterile neutrino is:
Pi s = 1
3

Pi j = u2i 1 |T |2 .
(A.13)
j =1
Using (A.11) we obtain

2
m
2L
T +T
2
= 1 2 sin2 0 2F2 m
n ,
2
4E
N0
2
2

2
|T |2 = 1 4F2 m
20 + F12 m
n + 4F22 m
nm
n ,
(A.14)
(A.15)
where
2 1
m
2L
n =
sinp n .
Fp m
2
4E
Nn
(A.16)
n=0
We can now calculate the average probabilities for a neutrino of flavor i to survive

2
Pi i = 1 u2i + u2i 2 ,
(A.17)
or to be converted to flavor j

Pi j = u2i u2j 1 + 2 ,
or to decay into sterile

Pi s = u2i 1 2 ,
i = j,
(A.18)
(A.19)
111
where
2
1
Nn4
n
(A.20)
and we have averaged over all frequency modes.

For mD = 0, the eigenvalues of the matrix mm form the usual KK tower with masses
mn given by Eq. (A.1). For mD R 1, it is natural to assume that the KK levels are slightly
shifted:
m
2n = m2n + m2n .
(A.21)
Inserting (A.21) into (A.3) and expanding for m2n m2n m2@ , n = @, we obtain to lowest
order:
rn m2D 2mn
,
1 + n
2
m2n =
(A.22)
where
r@ 2m2@
2m@
2
=
m
.
D
2 m2
2 m2
m
m
n
n
@
2

@=n

2
n = m2D
@Z
m@=mn
(A.23)
Due to the presence of the factor 2mn /M (after restoring the M units), we have m2n 0
for mn > M, implying that KK levels above the cutoff M are not shifted. We notice also
2
that in (A.1) we have
1
R 2 cos
in n :
Ms4
MP2
Ms2 . In this limit, we can calculate the leading terms

n = m2D R 2 cos log M 2 R 2 cos log 2 + sn ,
(A.24)
where sn is a constant term independent of the cutoff. Specifically, s0 = C and sn=0 =

log |n21 + n22 2n1 n2 sin | + C, with C a constant of order one. Similarly, for the
normalization coefficients we get to the lowest order in m2n
m20
1
=
+ ,
N02
m2D
(A.25)
(m2n )2
1
=
+ .
Nn2 m2D m2n rn 2m2n
(A.26)
The infinite KK sums F2 , F1 in (A.14), (A.15) can in principle be calculated numerically

using (A.3) and (A.6). However, this requires summation over a huge number of KK
modes. A convenient approximation is to assume n 1 for n > 0. In this region
2
rn 2m2n
2m2n
2
2 mn L
2
2 mn L
F2 mD
(A.27)
= mD
,
sin
sin
4 E
4 E
m2n
m2n
2
n=0
nZ
n=(0,0)
and thus, F2 can be expressed in terms of theta-functions using the formula
112
2mn
2
sin2 (m2n )
n Z2
m2n
4(x log + i)
4(x log + i)
= Im dx 3
3
(1 sin )R 2
(1 + sin )R 2
0

4(x log + i)
4(x log + i)
2
,
+ 2
(A.28)
(1 sin )R 2
(1 + sin )R 2

2
2
where 3 ( ) = nZ ei n and 2 ( ) = nZ ei (n+1/2) . In the case of an orthogonal
L
L
torus ( = 0), F is periodic under 4E
4E
+ R 2 . However, this property is lost for
arbitrary values of . The periodicity, of the survival probability (A.10), is necessary for
interpreting the neutrino anomaly through neutrino oscillations and is in general restored
for rational values of the angle
p
sin = , |p| < q,
(A.29)
q

where p, q are relatively prime integers, and q = 1 for p = 0. Restricting to this

L
L
subspace, F becomes periodic under 4E
4E
+ p,q R 2 , where
2 2
q p
for q = odd,
q
p,q =
(A.30)
q 2 p2 for q = even.
2q
Since F2 is a periodic function, the next question is to compute the corresponding
amplitude. To get an estimation of the amplitude we can evaluate the sum (A.27) at the
half period. We get
1/2
F2
= m2D
2m2n
= 0,
m2n
2
(A.31)
nZ
where
Z2 is one of the sets of (even, odd) and/or (odd, odd) integers of Z2 , depending on
the choice of q. More particularly, for q = 4@ with @ integer,
Z2 is the set of (odd, odd)
2

pairs, for q = 2@ + 1, Z is the set of (odd, even) and (even, odd) pairs and for q = 4@ + 2,

Z2 is the union of the two previous sets. The constant takes approximately the values
(1/4, 1/2, 3/4) for each of the three cases, respectively. Moreover, an upper bound to
the amplitude of F can be derived by replacing the sin2 terms with unity
F2max = m2D
2m2n
= 0 .
m2n
2
(A.32)
nZ
1/2
Hence, the oscillation amplitude lies in the range F2 < < F2max , that is 0 < <
0 with (1/4, 1/2, 3/4). Furthermore, we can proceed to a numerical evaluation of
F for given sin and MR. An explicit example is presented in Fig. 4, where we have
L
119
2 2
4
calculated F as a function of 4ER
2 in the case M R = 10 , = 16, sin = 120 . In the

L
same figure, we have also plotted the function sin2 240
239 4ER 2 (gray line) with amplitude
Fig. 4. Plot of the infinite two-dimensional sum
nZ2
113
n
2h

L hn
2
M 2 R 2 /hn sin2 4E
= cos
2 with hn
(n21 + n22 2n1 n2 sin ) in the case M 2 R 2 = 106 , = 16 and sin = 119
120 . The gray line represents the dominant
L 240
2 R 2 cos log 2 ).
frequency mode sin2 4E
with
amplitude
=
0.8
cos
log(M
2
239R
arising from the fit to the numerical sum data. The numerical evaluation of the sum shows
that the amplitude can in general be approximated by 0.80 . Taking into account
these results, we will assume in Section 6 that the KK sum F can be simulated by the
dominant frequency mode

pq L
q(3 + (1)q )
,
F2 0.8 0 sin2
(A.33)
=
,
pq
R 2 4E
2(q 2 p2 )
where 0 is given in (A.24) and is given in (A.29).
Assuming 0 1, we can drop the terms F22 and F12 < F22 in (A.15). Putting together
(A.12), (A.27) and (A.33), we obtain an approximate expression for the survival probability
Pi i 1 4u2i (1 u2i )(1 0 ) sin2
m20 L
2
2 pq L
3.2ui 0 sin
,
4E
R 2 4E
(A.34)
where
m20 = m2D (1 0 ).
(A.35)
Moreover, for the parameter 2 that enters in the average transition probability formulas
(A.17)(A.19), we have
2
1
N04
= 1 20 + .
(A.36)
114
References
[1] I. Antoniadis, E. Kiritsis, T.N. Tomaras, Phys. Lett. B 486 (2000) 186, hep-ph/0004214.
[2] I. Antoniadis, E. Kiritsis, T.N. Tomaras, Fortschr. Phys. 49 (2001) 573, hep-th/0111269.
[3] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[4] I. Antoniadis, C. Bachas, D. Lewellen, T.N. Tomaras, Phys. Lett. B 207 (1988) 441;
I. Antoniadis, Phys. Lett. B 246 (1990) 377;
I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69, hep-th/9310151;
J.D. Lykken, Phys. Rev. D 54 (1996) 3693, hep-th/9603133.
[5] K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 557 (1999) 25, hep-ph/9811428;
N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, J. March-Russell, Phys. Rev. D 65 (2002) 024032, hepph/9811448;
A.E. Faraggi, M. Pospelov, Phys. Lett. B 458 (1999) 237, hep-ph/9901299;
G.R. Dvali, A.Y. Smirnov, Nucl. Phys. B 563 (1999) 63, hep-ph/9904211;
R.N. Mohapatra, S. Nandi, A. Perez-Lorenzana, Phys. Lett. B 466 (1999) 115, hep-ph/9907520;
R.N. Mohapatra, A. Perez-Lorenzana, Nucl. Phys. B 576 (2000) 466, hep-ph/9910474;
R. Barbieri, P. Creminelli, A. Strumia, hep-ph/0002199.
[6] I. Antoniadis, C. Bachas, Phys. Lett. B 450 (1999) 83, hep-th/9812093.
[7] I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 464 (1999) 38, hep-th/9908023;
G. Aldazabal, A.M. Uranga, JHEP 9910 (1999) 024, hep-th/9908072.
[8] D. Cremades, L.E. Ibanez, F. Marchesano, hep-th/0205074;
D. Cremades, L.E. Ibanez, F. Marchesano, JHEP 0207 (2002) 022, hep-th/0203160;
D. Cremades, L.E. Ibanez, F. Marchesano, JHEP 0207 (2002) 009, hep-th/0201205;
G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, J. Math. Phys. 42 (2001) 3103, hepth/0011073;
G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, JHEP 0102 (2001) 047, hep-ph/0011132.
[9] L.E. Ibanez, F. Marchesano, R. Rabadan, JHEP 0111 (2001) 002, hep-th/0105155.
[10] R. Blumenhagen, V. Braun, B. Kors, D. Lust, JHEP 0207 (2002) 026, hep-th/0206038;
R. Blumenhagen, B. Kors, D. Lust, Phys. Lett. B 532 (2002) 141, hep-th/0202024;
R. Blumenhagen, B. Kors, D. Lust, T. Ott, Nucl. Phys. B 616 (2001) 3, hep-th/0107138;
M. Cvetic, P. Langacker, G. Shiu, Nucl. Phys. B 642 (2002) 139, hep-th/0206115;
M. Cvetic, P. Langacker, G. Shiu, Phys. Rev. D 66 (2002) 066004, hep-ph/0205252;
M. Cvetic, G. Shiu, A.M. Uranga, Phys. Rev. Lett. 87 (2001) 201801, hep-th/0107143;
D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 502 (2001) 209, hep-th/0011289;
D. Bailin, G.V. Kraniotis, A. Love, hep-th/hep-th/0210219;
C. Kokorelis, JHEP 0208 (2002) 018, hep-th/0203187;
C. Kokorelis, hep-th/0210200.
[11] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Rev. D 59 (1999) 086004, hep-ph/9807344.
[12] I. Antoniadis, M. Quiros, Phys. Lett. B 392 (1997) 61, hep-th/9609209.
[13] L.E. Ibanez, F. Quevedo, JHEP 9910 (1999) 001, hep-ph/9908305.
[14] I. Antoniadis, K. Benakli, M. Quiros, Phys. Lett. B 331 (1994) 313, hep-ph/9403290;
For a recent review see: I. Antoniadis, K. Benakli, Int. J. Mod. Phys. A 15 (2000) 4237, hep-ph/0007226,
and references therein.
[15] A. Sagnotti, Phys. Lett. B 294 (1992) 196, hep-th/9210127;
L.E. Ibanez, R. Rabadan, A.M. Uranga, Nucl. Phys. B 542 (1999) 112, hep-th/9808139.
[16] E. Poppitz, Nucl. Phys. B 542 (1999) 31, hep-th/9810010.
[17] I. Antoniadis, E. Kiritsis, J. Rizos, Nucl. Phys. B 637 (2002) 92, hep-th/0204153.
[18] E. Kiritsis, P. Anastasopoulos, JHEP 0205 (2002) 054, hep-ph/0201295.
[19] D.M. Ghilencea, L.E. Ibanez, N. Irges, F. Quevedo, JHEP 0208 (2002) 016, hep-ph/0205083.
[20] H. Arason, D.J. Castano, B. Keszthelyi, S. Mikaelian, E.J. Piard, P. Ramond, B.D. Wright, Phys. Rev. D 46
(1992) 3945;
M.K. Parida, A. Usmani, Phys. Rev. D 54 (1996) 3663.
[21] J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pena-Garay, JHEP 0207 (2002) 054, hep-ph/0204314.
[22]
[23]
[24]
[25]
[26]
P.C. de Holanda, A.Y. Smirnov, hep-ph/0205241.

G. Altarelli, F. Feruglio, hep-ph/0206077.
A.Y. Smirnov, hep-ph/0209131.
H. Davoudiasl, P. Langacker, M. Perelstein, Phys. Rev. D 65 (2002) 105015, hep-ph/0201128.
S. Hamidi, C. Vafa, Nucl. Phys. B 279 (1987) 465;
L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 282 (1987) 13.
[27] K2K Collaboration, M.H. Ahn, et al., hep-ex/0212007.
115

On mixed phases in gauge theories

V.L. Chernyak
Budker Institute for Nuclear Physics, 630090 Novosibirsk, Russia
Received 14 March 2003; accepted 9 April 2003
To Arkady Vainshtein, on occasion of his 60th birthday
Abstract
In many gauge theories at different values of parameters entering Lagrangian, the vacuum
is dominated by coherent condensates of different mutually non-local fields (for instance, by
condensates of electric or magnetic charges, or by various dyons). It is argued that the transition
between these dual to each other phases proceeds through the intermediate mixed phase, having
qualitatively different features. The examples considered include: ordinary YM, N = 1 SYM, N = 1
SQCD, and broken N = 2 SYM and SQCD.
PACS: 11.15.-q
1. SU(Nc )-YM at = 0 and dyons

vac (, Nc ), is
The physics of this theory, and in particular the vacuum energy density E
supposed to be periodic in + 2k . On the other hand, the standard large Nc -counting
rules imply (b0 = 11/3):
vac = Nc2 b0 4 F (/Nc ),
E
4
(1)
with F (z 0) = 1 c1 z2 + c2 z4 + .1 It was first pointed out by E. Witten [1] (see

also [2] for a similar behaviour in the stringy-YM theory) that the Nc -dependence in
Eq. (1) and periodicity in + 2 imply together that the function F (z) should be
non-analytic in its argument. So, for instance, instead of Eq. (1), the more explicit form of
E-mail address: chernyak@inp.nsk.su (V.L. Chernyak).
1 The numerical coefficient c is positive, but it is a dynamical quantity and cannot be determined from general
1
considerations alone.
doi:10.1016/S0550-3213(03)00309-2
V.L. Chernyak / Nuclear Physics B 660 (2003) 116130
vac on will rather look as:

dependence of E

+ 2k
vac = Nc2 b0 4 min
,
f
E
4
Nc
117
(2)
with f (z) being the normal analytic function.

vac (, Nc ) looks as follows. First, it is
The qualitative behaviour of the curve E
symmetric in and periodic in + 2k . Further, it has minimum at = 0
and begins to increase with increasing > 0, as it follows from general considerations of
the Euclidean functional integral determining this theory. It reaches its maximal value at
vac begins to
= . The curve itself is continuous at this point, but there is a cusp so that E
decrease in a symmetric way in the interval < < 2 , reaching the same minimal value
at = 2 .
As for the qualitative behaviour of the topological charge density, Pvac (, Nc ), it follows
vac /d , and looks as follows. First, it is antisymmetric
from the relation: Pvac (, Nc ) d E
in and periodic in + 2k . So, it is zero at = 0 and increases with
reaching its maximum value at = . There is a discontinuity at this point, so that
the curve jumps to the same but negative value as overshoots , and increasing in a
symmetric way reaches zero at = 2 .
vac (, Nc ) along the real The above described non-analytic (cusped) behaviour of E
vac (, Nc ) at large imaginary
axis agrees, in particular, with the asymptotic behaviour of E
values of , i/Nc = /Nc 1, obtained in [3]:

4
2 4

Evac (, Nc ) Nc exp
(3)
.
b0 Nc
vac (, Nc ) is not naturally periodic at 2i . Rather,
It is seen from Eq. (3) that E

it implies that periodic Evac (, Nc ) is analytic in the strip < Re < in the complex
-plane, and is glued then periodically strip by strip.
The natural physical interpretation explaining the origin of the above described cusped
vac ( ) along the real -axis has been proposed in [3], and looks as follows.
behaviour of E
Let us suppose the standard picture of the confinement mechanism to be valid, i.e.,
those of the dual superconductor. By this we imply here the dynamical mechanism with
composite (naturally adjoint) Higgs field which determines the formation of U (1)Nc 1
from the original SU(Nc ), SU(Nc ) U (1)Nc 1 , and besides the Ui (1)-magnetically
charged excitations (monopoles) condense. We will be interested to trace the qualitative
behaviour of this vacuum state in its dependence of . For this, it will be sufficient
to consider the first U (1)-charge only with its monopoles, the dual photon and
corresponding g gluons as if it were the SU(2) theory, because is SU(Nc )-singlet and
all other U (1) charges will naturally behave the same way under variation of .
As has been shown by E. Witten [4], the pure monopole M = (magnetic charge = 1,
electric charge = 0) at = 0 turns into the dyon with charges d1 = (1, /2) at = 0. So,
the coherent condensate of monopoles in the vacuum at = 0 turns into the condensate of
d1 -dyons as starts to deviate from zero, and the vacuum energy density begins to increase
for this reason.
It is a specific property of our system that there are two types of condensates made of the
dyons and antidyons with the charges: {(1, 1/2); (1, 1/2)} and {(1, 1/2); (1, 1/2)},
118
and having the same energy density. This can be seen, for instance, as follows. Let us
start from the pure monopole condensate at = 0 and let us move anticlockwise along the
path: = 0 = . The vacuum state will consist of (1, 1/2)-dyons and (1, 1/2)antidyons. Let us move now clockwise along the path: = 0 = . The vacuum state
will consist now of (1, 1/2)-dyons and (1, 1/2)-antidyons. Because the vacuum energy
density is even under , these two vacuum states are degenerate.2
Besides, these two states belong to the same world as they are reachable one from
another through a barrier, because there are electrically charged gluons, g = (0, 1),
which can recharge these (1, 1/2)-dyons into each other. In contrast, the two vacuum
states, | and | at = 0, are unreachable one from another and belong to different
worlds, as there is no particles in the spectrum capable to recharge the (1, /2)-dyons
into each other.
Thus, the vacuum state becomes twice degenerate at = , so that the level crossing
(in the form of rechargement: {d1 = (1, 1/2), d1 = (1, 1/2)} {d2 = (1, 1/2), d2 =
(1, 1/2)}) can take place if this will lower the energy density at > . And indeed
vac ( ). At > the vacuum is filled now with
it lowers, and this leads to a casp in E
the coherent condensate of new dyons with the charges: d2 = (1, 1 + /2), d2 =
(1, 1 /2). As increases further, the electric charge of these d2 -dyons decreases,
and the vacuum energy density decreases with it. Finally, at = 2 the d2 -dyons (which
were the (1, 1)-dyons at = 0) become pure monopoles, and the vacuum state becomes
exactly as it was at = 0, i.e., the same condensate of pure monopoles and antimonopoles.
We emphasize that, as it follows from the above picture, it is wrong to imagine the
vacuum state at = 2 as, for instance, a condensate of dyons with the charges (1, 1),
degenerate in energy with the pure monopole condensate at = 0.3
2 The existence of two vacuum states at = does not follow from the symmetry considerations alone,
vac ( ) and E
vac ( ) = E
vac ( + 2 k). It is sufficient to give a counterexample. So, let us
vac ( ) = E
like E
consider the SU(2) YangMills together with the Higgs doublet with large vacuum condensate. In this case the
-dependence of the vacuum energy density is due to a rare quasiclassical gas of instantons, and is cos( ). All
the above symmetry properties are fulfilled, but there is only one vacuum state at = . (See also the end of this
section.)
3 In this respect, the widely used terminology naming the two singularity points u = 2 on the N =
2SU(Nc = 2) SYM moduli space as those where monopoles and respectively dyons become massless, is not
quite adequate (and may be dangerous for this reason, leading to wrong conclusions). It is based on quantum
numbers n = (nm , ne ) of corresponding fields, and these quantum numbers are always the same independently of
the point of the moduli space we are staying in, and are not direct physical observables. In contrast, the standard
physical terminology is based on charges g = (gm , ge ) which are the direct physical observables because, by
definition, the Coulomb interaction of two particles is proportional to product of their charges, not quantum
numbers. In distinction from quantum numbers, the values of charges depend on the point of the moduli space,
due to Wittens effect.
To illustrate, let us start from the vacuum u = 2 where, by definition, the massless particles are pure
monopoles and let us move, for instance, along a circle to the point u = 2 . On the way, the former
massless monopole increases its mass because it becomes the d1 = (1, (u)/2 )-dyon (here (u)/2 =
Re (u)). At the same time, the former massive d20 = (1, 1)-dyon diminishes its mass as it becomes the
d2 = (1, 1 + (u)/2 )-dyon. When we reach the point u = 2 , i.e., (u) = 2 , the former dyon becomes
massless just because it becomes the pure monopole here. So, an observer living in the world with u = 2 will
119
Physically, the above rechargement process will appear as a typical first order phase
transition. After overshoots , in a space with the coherent condensate of d1 =
(1, 1/2)-dyons and d1 = (1, 1/2)-antidyons, the bubbles will appear with the coherent
condensate of d2 = (1, 1/2)-dyons and d2 = (1, 1/2)-antidyons deep inside each
bubble, and with a transition region surface (domain wall) through which the averaged
densities of two type dyons interpolate smoothly. These bubbles expand then over all the
space through the rechargement process d1 + d1 d2 + d2 occurring on a surface of each
bubble. This rechargement can be thought as going through a copious production of
charged gluon pairs g + g , so that the underlying processes are: [d1 = (1, 1/2)] + [g =
(0, 1)] [d2 = (1, 1/2)] and [d1 = (1, 1/2)] + [g + = (0, 1)] [d2 = (1, 1/2)].
Some analogy with the simplest Schwinger model may be useful at this point, in
connection with the above described rechargement process. Let us consider first the pure
QED2 without finite mass charged particles, and let us put two infinitely heavy quarks
with the charges /2 (in units of some e0 ) at the edges of our space. It is well known
[5] that this is equivalent to introducing the -angle into the QED2 Lagrangian. As a result,
there is the empty vacuum at = 0, and the long range Coulomb string at = 0. The
vac ( ) = C0 e2 2 , C0 = const, at any 0 < .
vacuum energy density behaves as: E
0
Let us add now some finite mass, m e0 , and of unit charge e0 field to the
Lagrangian. When there are no external charges, this massive charged field can be
integrated out, resulting in a small charge renormalization. But when the above quarks
vac ( ) becomes non-trivial. The charge of the external
are introduced, the behaviour of E
quark tends to 1/2 as approaches . As overshoots it becomes preferable to
produce a pair of -particles, + , from the vacuum. They separate so that to recharge
the external quarks: (1/2) (1/2) (without changing the volume energy), and the
external charges become equal (/2 1) and (/2 + 1) at > . As a result of this
vac ( ) and it begins to decrease at > , so that
rechargement, there appears a cusp in E
vac ( )
the former empty vacuum is reached at = 2 . Therefore, the behaviour of E
2
2
2
2

will be: Evac ( ) = C0 e0 {min k ( + 2k) }, so that Evac ( ) = C0 e0 at 0 , and
vac ( ) = C0 e2 (2 )2 at 2 .
E
0
Let us return however to our dyons. The above described picture predicts also a definite
qualitative behaviour of the topological charge density, P( ). At 0 < < , i.e., in the
condensate of the d1 = (1, /2)-dyons and d1 = (1, /2)-antidyons, the product of
signs of the magnetic and electric charges is positive for both d1 -dyons and d1 -antidyons.
H , E H > 0, and both
Thus, these charges give rise to the correlated field strengths: E
species contribute a positive amount to the mean value of the topological charge density,
so that P1 ( ) > 0 and grows monotonically with in this interval following increasing
electric charge /2 of the dyon.
On the other side, at < < 2 , i.e., in the condensate of the d2 = (1, 1 + /2)dyons and d2 = (1, 1 /2)-antidyons, the product of signs of the magnetic and
electric charges is negative for both d2 -dyons and d2 -antidyons. Thus, both species
also see the massless monopoles (not dyons, and this is distinguishable by their Coulomb interactions between
themselves and with other dyons), exactly as those living in the world with u = 2 .
120
contribute a negative amount to P2 ( ), such that: P2 ( ) = P1 (2 ), and P( ) jumps
reversing its sign at = due to rechargement.
vac ( ) and discontinuous
On the whole, it is seen that the cusped behaviour of E

behaviour of P ( ) appear naturally in this picture of the confinement mechanism in
SU(Nc )-YM theory, and are exactly the same that are expected from simplest general
considerations and were described in the beginning of this section.
Clearly, at 0 < the condensate made of only the d1 = (1, /2)-dyons (recalling
also for a possible charged gluon pair production) can screen the same type dk =
(const(1, /2) + (0, k))-test dyon only (k = 0, 1, 2, . . . ; and the same for the d2 dyons at < 2). So, the heavy quarkantiquark pair will be confined at = .
New non-trivial phenomena arise at = . Because there are two degenerate states,
i.e., the condensates of (1, 1/2)-dyons (and antidyons), a mixed state configuration
becomes possible with, for instance, each condensate filling a half of space only, and
with the domain wall interpolating between them. The simplest reasonings about the
energy scales involved in this domain wall are as follows. The masses of relevant gluons
g = (0, 1) and both dyons (1, 1/2) coexisting together in the core of the domain wall
are naturally YM , and so of the same size will be increase in energy density. Besides,
there are (Nc 1) independent Ui (1) charges. On the whole, therefore, the domain wall
tension will be T Nc 3YM , while its typical width will be 1/YM .
Physically, the above domain wall represents a smeared rechargement, i.e., smeared
over space interpolation of electrically charged degrees of freedom between their
corresponding vacuum values, resulting in a smooth variation of the averaged densities of
both type dyons (1, 1/2) through the domain wall. Surprisingly, there is no confinement
inside the core of such domain wall.
The reason is as follows. Let us take the domain wall interpolating along the z-axis,
so that at z there is the main coherent density of d1 = (1, 1/2)-dyons, and at
z , that of d2 = (1, 1/2)-dyons. As we move from the far left to the right, the
density of d1 -dyons decreases and there is also a smaller but increasing incoherent density
of d2 -dyons. This small amount of d2 -dyons is harmless, in the sense that its presence
does not result in the screening of the corresponding charge. The reason is clear: the large
coherent density of d1 -dyons confines the d2 -dyons so that they cannot move freely and
appear only in the form of rare and tightly connected neutral pairs d2 d2 , with different
pairs fluctuating independently of each other. As we are going further to the right, the
density of these neutral pairs grows and their typical size increases (although they are still
confined), because the main density of d1 -dyons decreases. Finally, at some distance from
the centre of the wall the percolation takes place, i.e., the d2 -dyons form a continuous
coherent network and become released, so that the individual d2 -dyon can travel freely to
arbitrary large distances (in the transverse xy-plane) but only within its network. And in
this percolated region the coherent network of d1 -dyons still survives, so that these two
coherent networks coexist in space and form the new mixed phase with qualitatively
different properties.
This is a general feature, and each time when there will coexist coherent condensates
of two mutually non-local fields, they will try to confine each other, and will resemble the
above described case.
121
The above mixed phase shares some features in common with the mixed state of the
type-II superconductor in the external magnetic field. The crucial difference is that the
magnetic flux is sourceless inside the superconductor, while in the above described mixed
phase there are real dual to each other charges, each type living within its network.
As we move further to the right, the density of d2 -dyons continue to increase while
those of d1 continue to decrease. Finally, at the symmetric distance to the right of the wall
centre the inverse percolation takes place, i.e., the coherent network of d1 -dyons decays
into separate independently fluctuating neutral droplets whose average density (and size)
continue to decrease with increasing z. Clearly, the picture on the right side repeats in a
symmetric way those on the left one, with the d1 and d2 dyons interchanging their roles.4
Let us consider now the heavy test quark put inside the core of the domain wall,
i.e., inside the mixed phase region. Clearly, this region has the properties of the double
Higgs phase. Indeed, because the (two-dimensional) charges of two dyons, (1, 1/2) and
(1, 1/2), are linearly independent, polarizing itself appropriately this system of charges
will screen any external charge put inside, and the quark one in particular.
Finally, if the test quark is put far from the core of the wall, the string will originate from
this point making its way toward the wall, and will be screened inside the mixed phase (i.e.,
the double Higgs) region. It should be emphasized, however, that, as it is clear from the
above explanations, if this test quark is moved further inside the core of the another wall,
then its flux will be screened and nothing will support this string. So, the electric string
cannot be stretched between two such domain walls.
Let us point out finally that the assumption about the confinement property of the
SU(Nc )-YM theory is not a pure guess, as the above discussed non-analytic (i.e., cusped)
vac (, Nc ), is a clear evidence for a phase
behaviour of the vacuum energy density, E
transition at some finite temperature. Indeed, at high temperatures the -dependence of
the free energy density in the gluon plasma is under control and is: T 4 (/T )Nc b0 cos ,
due to rare gas of instantons. It is important for us here that it is perfectly analytic in
, and that the form of its -dependence is T-independent, i.e., it remains to be cos
when the temperature decreases. On the opposite side at T = 0, i.e., in the confinement
phase, the -dependence is non-analytic and, clearly, this non-analyticity survives at small
temperatures as there are no massless particles in the spectrum. So, there should be a phase
transition (confinementdeconfinement) at some critical temperature, Tc , where the
-dependence changes qualitatively.
2. N = 1 SU(Nc ) SYM
In this theory the residual non-anomalous discrete axial symmetry is broken spontaneously, so that there are Nc vacuum states differing by the phase of the gluino conden4 Evidently, if we replace the above domain wall with fixed = by the domain wall of the light axion
field a(z), ma YM , interpolating between a = 0 at z and a = 2 at z with a(z = 0) = ,

all the properties will remain the same in the core, i.e., at |z| (several)1
YM . The main difference will be that
the condensates of d1 = (1, 1/2), and d2 = (1, 1/2)-dyons will turn into condensates of pure monopoles at
corresponding sides of large distances |z| 1/ma (see also the next section).
122
sate [6,7]:

2k
0||0k Nc exp i
.
Nc
3
(4)
Besides, it is widely believed that this theory is confining, similarly to the usual YM-theory.
In what follows, we will suppose that the confinement mechanism here is the same as in
the previous section, i.e., those of the dual superconductor. Our purpose in this section will
be to describe qualitatively the physical properties of domain walls interpolating between
the above vacua and, in particular, their ability to screen the quark charge [8].
For this, let us consider the effective theory obtained by integrating out all degrees of
freedom except for the composite chiral field S = (W2 /32 2 Nc ), S = (, . . .)/32 2 Nc =
( exp{i}, . . .) (i.e., the integration proceeds with the constraint that the field S is fixed,
[8]). The form of the superpotential in the Lagrangian for the field S can then be simply
determined and coincides with those of VenezianoYankielowicz [6]: W S ln(S N /3N ),
resulting in the gluino condensation, see Eq. (4).
Because the field Nc in SYM is the exact analog of in the ordinary YM, the physical
vac ( ) in the YM-theory described in the
interpretation and qualitative behaviour of E
previous section can be transferred now to SYM, with only some evident changes:
vac ( ) U (Nc ), and it is not the vacuum energy density now but rather the scalar
(a) E
potential of the field ;
(b) if we start with the condensate of pure monopoles at = 0, the rechargement d1 =
(1, Nc /2) d2 = (1, 1 + Nc /2) and the cusp in U (Nc ) will occur now at
= /Nc , so that at = 2/Nc we will arrive at the next vacuum with the same pure
monopole condensate but with shifted phase of the gluino condensate.
Let us consider now the domain wall interpolating along z-axis between two nearest
vacua with k = 0 and k = 1, so that (z) 0 at z , and (z) 2/Nc at z .
There is a crucial difference between this case and those described just above where the
field was considered as being spacetime independent, i.e., (z) = const. The matter
is that the system cannot behave now in a way described above (which allowed it to
have a lowest energy U (Nc ) at each given value of (z) = = const): i.e., to be the
pure coherent condensate of d1 -dyons at 0 < /Nc , the pure coherent condensate of
d2 -dyons at /Nc < 2/Nc , and to recharge suddenly at = /Nc . The reason is that
the fields corresponding
to electrically charged degrees of freedom also become functions

of z at q = dz [ddw (z)/dz] = 0. So, they cannot change abruptly now at z = 0 where
dw (z) goes through /Nc , because their kinetic energy will become infinitely large in this
case. Thus, the transition will be smeared necessarily.
The qualitative properties of the domain wall under consideration here will be similar
to those described in the previous section. The main difference is that was fixed at in
Section 1, while Nc dw (z) acts like the axion field, i.e., it varies here between its limiting
values, and the electric charges of dyons follow it.
So, at far left there will be a large coherent condensate of d1 = (1, Nc /2)-dyons
(pure monopoles at z ), and a small incoherent density of d2 = (1, 1 + Nc /2)-
123
dyons.5 The d2 -dyons cannot move freely in this region as they are confined, and appear

as a rare and tightly connected neutral pairs d2 d2 only. Therefore, their presence does not
result in the screening of the corresponding charge. As we move to the right, the density of
d1 -dyons decreases while those of d2 -increases. These last move more and more freely,
but are still confined. Finally, their density reaches a critical value at z = z0 , so that a
percolation takes place and the d2 -dyons form a continuous coherent network within which
the individual d2 -dyons can move freely to any distance (in the transverse xy-plane). At
the same time, there still survives a coherent condensate of d1 -dyons, which still can freely
move individually within their own network.
At the symmetrical point z = z0 to the right of the domain wall centre at z = 0, the
inverse percolation takes place, so that the coherent connected network of d1 -dyons
decays into separate independently fluctuating neutral droplets, whose density (and size)
decreases with further increasing z. At large z we arrive at the vacuum state with a large
coherent condensate of monopoles (former (1, 1)-dyons at large negative z).
Now, let us consider what happens when a heavy quark is put inside the core of the
domain wall. The crucial point is that there is a mixture of all four dyon and antidyon
species (of all Nc 1 types): d1 = (1, Nc /2), d1 = (1, Nc /2), d2 = (1, 1 +
Nc /2) and d2 = (1, 1 Nc /2) in this percolated region, with each dyon moving
freely inside its coherent network. So, this region has the properties of the double Higgs
phase, as here both the d1 - and d2 -dyons are capable to screen corresponding charges.
And because the charges of d1 - and d2 -dyons are linearly independent, polarizing itself
appropriately this mixture of dyons will screen any test charge put inside, the heavy quark
one in particular.
If the test quark is put at far left (or right) of the wall, the string will originate from
this point making its way towards a wall, and will disappear inside the core of the wall,
i.e., in the mixed phase region where the string flux will be screened. And similarly to the
previous section, the string cannot be stretched between two domain walls.
The above described explanation of the physical phenomena resulting in quark string
ending in the wall differs from both, those described by E. Witten in [9] and those proposed
by I. Kogan, A. Kovner and M. Shifman in [10] (see also footnote 3).
3. N = 1 SU(2), NF = 1 SQCD
As previously, we will imply here that there is confinement of electric charges in the
N = 1 pure SYM-theory (see previous sections). Then, there will be three phases in this
N = 1 SQCDtheory, depending on the value of mthe mass parameter of the quark [8].
At small m , there will be the usual electric Higgs phase, with the large quark
1/2 with masses
(5 /m)1/2 , and light quark composite fields (QQ)
condensate QQ
m. The effective low energy Lagrangian for these fields is those of AffleckDine
Seiberg [11].
5 Other possible dyons play no role in the transition we consider, and we will ignore them.
124
The heavy magnetically charged excitations (monopoles) will be confined, and so the
monopoles will appear as rare and tightly bound neutral pairs only, with different pairs
fluctuating independently of each other.6
With increasing m the quark condensate and the monopole mass decrease, while the
density of monopole pairs and their typical size increase. At some value m = c1 the
percolation of the monopole droplets takes place, so that in the interval c1 m c2
there will be the mixed phase (or equivalently, the double Higgs phase) with two infinite
size connected coherent networks of monopoles and quarks, with their averaged densities
being constant over the space and following only the value of m.
There will be screening rather than confinement (although the difference between these
two becomes to a large extent elusory here) of any test charge in this interval of m.
Finally, at m = c2 the quarks become too heavy and cannot support their coherent
condensate anymore, so that this last decays into independently fluctuating neutral droplets
whose density and typical size decrease with increasing m.
At m we arrive at the N = 1 SYM-theory with YM = m1/6 5/6 , and with heavy
(m YM ) quarks which are confined.
5/2 /m1/2 3 /m is small but non-zero even
The chiral quark condensate QQ
YM
in this region, but this small value is unrelated here with the gluon masses and charge
()/m
screening by quarks, and is a pure quantum loop effect of heavy quarks: QQ
2

(similarly to the heavy quark condensate G /m in the ordinary QCD).
4. Broken N = 2 SU(2) SYM
Let us recall the famous solution of this theory by N. Seiberg and E. Witten, with the
low energy Lagrangian (at small ) [12]:

+ Im AD A /4
eVD M
L = d 4 M e VD M + M

i
(5)
d 2 D WD2 + d 2 2 MMA
D + U (AD ) + h.c.
16
6 That there are monopoles in this theory at m can be seen as follows. First, let us consider the
effective Lagrangian obtained by integrating out hard degrees of freedom with high energy scales 0 , 0
(5 /m)1/4 . These include the instanton contributions, as the typical instanton size is Q Q1/2 1/0 .
to the original superpotential mQQ.
Now, the
The instanton will add the AffleckDineSeiberg term 5 /(QQ)
so obtained effective Lagrangian is the appropriate one to look, in particular, for a possible string solution if the
characteristic distances involved in the string formation are larger than .
This is the case at the classical level, and there will be the solution for the AbrikosovNielsenOlesen like
string with the magnetic flux. But because the quarks are in the fundamental representation, the gauge group
is SU(2) which is simply connected and there are no truly uncontractable strings in this theory. This implies
that the above classical string will break up on account of quantum tunneling effects. Physically, this break up
will be realized through the production of a pair of magnetically charged particle and antiparticle, with their
subsequent separation along the string axis to screen the external infinitely heavy monopoles at the string ends.
So, these magnetically charged particles should be present in the excitation spectrum of this theory (even if they
are not well formed).
125
Here M is the monopole field. Because it was not integrated yet, the terms entering
the Lagrangian in Eq. (5) (D , etc.) do not contain the monopole loop contributions and
have no singularity at 0 and U 2 . The field VD is those of the dual photon
and WD is its field strength. Below, it will be convenient for us to consider A and U in
Eq. (5) as functions of the field AD which is a pure quantum field, i.e., has zero vacuum
expectation value. The vacuum state we are dealing with is at U = 2 , with D 1 and
Im(AD A /4) AD AD in Eq. (5) at small .
How the effective Lagrangian for these fields can look if is large in comparison with
? Because at the degrees of freedom which have been integrated out were heavy
(in particular, the charged Higgs fields with their masses , and charged W bosons
with their masses mW YM = 1/3 2/3 ), the N = 2 supersymmetry will be broken
explicitly and -dependence will penetrate the effective Lagrangian. At the same time, it
is not difficult to see that due to: (a) holomorphicity; (b) R-charge conservation (with the
R-charge of equal two); (c) the known limit at , the additional -dependence
cannot appear in the F -terms, and so will appear in the D-terms only.
Besides, restricting ourselves to the terms with no more than two spacetime derivatives,
it will be sufficient for our purposes to write these D-terms for the monopole and AD
fields as the standard kinetic terms multiplied by the c-number Z-factors ZM (/) and
ZH (/), originating from those degrees of freedom which have been integrated out. Let
us denote by L the so obtained Lagrangian.
Recalling that the original theory was N = 2 SYM broken by the mass term of the
Higgs fields, we are ensured that at the Higgs fields become heavy, with their
masses mH , and decouple. So, we end up with N = 1 SYM with the scale parameter:
YM = 1/3 2/3 , and this is the only scale of this theory.
On the other hand, one obtains from L at that7 ZH stays intact, ZH (/
1) 1, in order to have mH , while the values of the dual photon and monopole masses
look as
m2 ZM (/)0|M M|0 ZM (/) ,
2
m2M ZM
(/) 2 .
(6)
Let us combine now Eq. (6) with the additional assumption: there is no massless
particles in the spectrum of N = 1 SYM. Then, this requires:8

ZM (/)
1/3
at .
(7)
It follows now from Eqs. (6), (7):

m mM YM
at ,
(8)
7 In the above described set up, there is no need for the function U (A ) of the U (A ) term in Eq. (5) to be
D
D
exactly the SeibergWitten function. For our purposes and for simplicity, it will be sufficient to keep only three
first terms (i.e., constant, linear and quadratic in AD ) in the expansion of U (AD ) in powers of AD , to ensure that
the adjoint Higgs becomes heavy, mH , and decouples.
8 If (/)1/3 Z (/) 0, the dual photon will be massless (on the scale
YM , i.e., m /YM 0),
M
while if (/)1/3 ZM (/) the monopole will be massless.
126
i.e., both the dual photon and monopole survive in the spectrum of the N = 1 SYM. This
is non-trivial in the sense that one of them or both could become heavy and decouple at
.
As for the value of the monopole condensate, it depends clearly on the normalization
of the monopole field. In the presence of the monopole ZM factor, the old normalization

0|MM|0
is not the natural one. The appropriate normalization is: 0|NN|0
=
1/2
1/2
2
9

0|ZM M ZM M|0, and it has the right scale: 0|NN|0 YM .
On the whole, the above described results were obtained implying that there is no phase
transition in the broken N = 2 SYM theory when going from small to large
(in a sense, at least, that monopoles continue to form the coherent condensate which gives
the mass to the dual photon; on the other hand, restructuring of the spectrum definitely
occurs at ). They show a selfconsistency of this assumption and give a strong support
to the widely accepted expectation that the N = 1 SYM theory is confining, with the
confining mechanism those of the dual superconductor. In other words, when going from
to in the broken N = 2 SYM theory, the external adjoint Higgs of this
theory decouples at and its role is taken by the dynamically formed and condensing
internal composite adjoint Higgs of the N = 1 SYM theory, while monopoles continue
to form the coherent condensate and keep the dual photon massive.
5. Broken N = 2 SU(2), NF = 1 SQCD
The solution of the unbroken N = 2 theory has been given by N. Seiberg and E. Witten
[13]. The original superpotential of the broken N = 2 N = 1 theory has the form (the
kinetic terms are canonical):
a
a Q + 2 ,
+ h 2Q
W = mQQ
(9)
2
are in the 2 and
where the quark fields Q and Q
2 representations of the colour group
SU(2), and is the adjoint Higgs field. The unbroken N = 2 SUSY corresponds to = 0
and h = 1.
The properties of this broken N = 2 theory have been considered previously in [1416].
The most detailed description has been given recently by A. Gorsky et al. in [17], and we
use widely the results of this paper below. For our purposes, we will deal with the special
case of light quarks weakly coupled to the Higgs field:
3/2
m ,
=h
(10)
1,
m
9 It is not difficult to trace the role of higher order terms A2 (A /)k1 in the expansion of U (A )
D
D D
in L in powers of (AD /). After integrating out the heavy field AD , these will give additional terms in
)/. Rescaling the monopole field to
the superpotential in powers of the pure quantum field (MM
1/2
N = ZM M, to have the canonical kinetic term for the field N , these additional terms in the superpotential
2 )/2 , i.e., depending only on the scale YM , as it
will come in powers of the pure quantum field (NN
YM
YM
should be in the N = 1 SYM theory.
127
where is the scale parameter of the original fundamental theory ( = 1 in what follows).
Under the conditions of Eq. (10), one vacuum state decouples and there remain two
physically equivalent vacuum states. So, it will be sufficient to deal with one of them where
the condensates of original fields take the values [17]:

m1/2 ,
QQ
(11)
U = 2 m1/2 ,
m1/2,
while the condensate of the monopole field is

MM
m1/4.
(12)
Under the conditions of Eq. (10), the only freedom remained is the relative value of
and m, and the phase and physical content of this theory depend essentially on this. Indeed:
are heavy and decouple, we are in the pure
(a) at sufficiently small ; the quarks Q(Q)
(1)
1/4
N = 2 SYM theory with eff = m 3/4 , broken by the small U -term. The vacuum
is the SeibergWitten vacuum, i.e., the dominant condensate is the Higgs one, a ,
1/4 . The light monopole
leading to SU(2) U (1), with W masses MW (1)
eff m
field condenses in the low energy U (1)-theory, resulting in the confinement of electric
charges. The lightest particles are the dual photon , its N = 2 partner AD and the
1/2 , all with small masses 1/2 m1/8 . We will call this
monopole composite (MM)
phase the magnetic one;
(b) at sufficiently large ; the Higgs field is heavy and decouples, we are in N = 1,
(2)
1/2 NF = 1 SQCD with eff = 2/5 3/5 and with light quark composite (QQ)
(2)
j
fields, with their masses m eff . Here, the large quark condensate Qi Q
dominates, SU(2) is broken completely and there is confinement of magnetically
charged excitations (monopoles), see footnote 6. The low energy effective Lagrangian
is those of AffleckDineSeiberg [11]. We will call this phase the electric one.
So, unlike the examples considered in previous sections, at the conditions given by
Eq. (10) we have a good control here over the phases of our theory in both limiting cases
of small and large values of , and these phases are dual to each other and are dominated
by coherent condensates of mutually non-local monopole and quark fields. Our purpose
now is to trace in more detail the transition between the magnetic and electric phases at
some value 0 , when going from small 0 to large 0 values of . We
expect that this transition proceeds through the formation of the mixed phase in some
region c1 0 c2 0 (with c1 c2 , but parametrically both c1 c2 O(1)).10
In the magnetic phase region 0 0 we will proceed in the same way as in
the previous section, by retaining only the lightest fields of the dual photon , AD and
10 Here and in other supersymmetric theories, the condensates of chiral superfields are frequently simplest
smooth functions of chiral parameters in the whole parameter space, see, for instance, Eqs. (11), (12). This does
not contradict to possibility for a system to be in qualitatively different (dual to each other) phases at different
values of parameters, because these condensates are not order parameters for these phases. Rather, the masses of
the direct and dual photon (together with W -masses) look more like the order parameters in the electric and
magnetic phases, respectively.
128
monopole M. All quark fields, in particular, are integrated out. Although, see Eq. (10),
h 0 and quarks do not interact directly with the Higgs fields, they interact with massive
charged gluons and gluinos and will give corrections in powers of the characteristic scale
/m3 (/m5/2). Further, being integrated out, the massive gluons and gluinos will
transmit these corrections to the monopole ZM -factor: ZM = ZM (/m5/2). So, the quarks
really decouple only at m5/2 where ZM (/m5/2 0) 1, while at > m5/2 the
quarks influence the physics and ZM = 1.
Similarly, in the electric phase region 0 , after integrating out the Higgs and gauge
fields, the quark ZQ -factor will obtain corrections in powers of the characteristic scale
/3 (m1/2 /2 ), so that ZQ = ZQ (m1/4 /) and the heavy Higgs field really
decouples only at m1/4 : ZQ (m1/4 / 0) 1. At 0 < m1/4 the adjoint Higgs
field still influences the physics, and ZQ = 1.
It is not difficult to see that with the choice:

m1/4

m
0
0
ZM
(13)
ZM 0 /m5/2
,
ZQ
ZQ m1/4/0
,
0
0
all particle masses and all (properly normalized) condensates are matched in the transition
region 0 , as it should be (the gauge couplings are 1 at 0 , see below):
MH MM MQ 0 ,
MW M M m1/4 > 0 ,

2
0 Q
Z0 Q Z0 M
Z 0 M m1/2 .
ZQ
Q
M
M
(14)
(15)
The non-trivial fact is that the number of matching conditions in Eqs. (14), (15) is larger
than the choice of only two numbers in Eq. (13).
As for the value of 0 , it is determined by matching at 0 of two characteristic
1/4 from the magnetic phase region 0 < < , and those of
scales: those of (1)
0
eff = m
(2)
2/5
eff = from the electric phase region > 0 :
2/5
m1/4 0 0 m5/8 .
(16)
It is not difficult to see that this is equivalent to matching of the gauge couplings.
Indeed, in the electric phase region the charged degrees of freedom decouple at the scale
MW = M , determined by the quark condensation, restructuring the spectrum and
1/2 . So, the inverse gauge
decoupling of the light neutral quark composite field (QQ)
2/5
(2)
0
coupling behaves as ln(MW /eff ), which is ln(m1/4/0 ) at 0 (see
Eq. (14)).
In the magnetic phase region there are two characteristic scales: (a) those connected
(1)
with the Higgs condensation and decoupling of W and gives ln(MW /eff ) 1;
and (b) those connected with the monopole condensation and decoupling of the neutral
1/2 . This last gives D ln((1)/M ), which is also
monopole composite field (MM)
eff
0
D 1 at 0 (see Eq. (14)). Therefore, the matching of couplings at 0 gives:
2/5
0 D0 1 ln(m1/4/o ).
So, ZM behaves like ZM = [1 + (/m5/2)]1/5 in the magnetic phase region 0 <
5/8
m , and the monopole mass (or, more exactly, the mass of the monopole composite
1/2 ) is: MM 2/5 m3/8 at m5/2 < < m5/8 , while ZQ behaves like ZQ =
field (MM)
[1 + (m1/4 /)]1 in the electric phase region m5/8 .
129
It is seen also from Eqs. (13) and (16) that there is a kind of duality relation at the
0
0
1/ZQ
. Moreover, at small m5/2 the confinement is
transition point 0 : ZM
very weak and the quark is nearly free, with mass MQ = m and ZQ 1. So, ZM 1/ZQ
holds also at small as well. Therefore, because the behaviour within the same phase
region is smooth, the duality relation ZQ 1/ZM holds in the whole magnetic phase
region 0 < < 0 .
(eff)
(eff)
1
Introducing the effective quark mass as: MQ = ZQ
m, it is then: MQ ZM m,
(eff)
and (MM /MQ ) (/0 )1/5 < 1 at m5/2 < < 0 . This shows that quark is heavier
than monopole in the whole magnetic phase region, and this is self consistent.
As was pointed out above, the example considered in this section has an advantage that
we have a good control over the properties of the magnetic and electric phases at both
sides, m5/8 and m5/8 , of the transition region at m5/8 . As for the properties
of the mixed phase in the transition region, they are similar to those described in previous
sections. In short:
(1) At very small < m5/2 the condensates of the Higgs and monopole fields dominate,
with the coherent Higgs condensate responsible for the W masses MW , and the
coherent monopole condensate responsible for the dual photon mass M . The quarks
are heavy and confined, and there are rare incoherent fluctuations of neutral quark
antiquark pairs;
(2) The density of these neutral quark pairs increases with increasing ;
(3) These quark bags (or strings) percolate at = c1 m5/8 . At c1 m5/8 c2 m5/8 , with
c1,2 O(1), the system is in the mixed phase, where two infinite size connected
coherent networks of electric and magnetic strings (or bags) coexist. In this region
(parametrically): quark condensate Higgs condensate monopole condensate,
quark mass Higgs mass monopole mass, photon mass dual photon mass, etc.
(see Eqs. (14), (15));
(4) At = c2 m5/8 the quark condensate takes over and enforces depercolation of the
connected coherent condensates of the monopole and Higgs fields. I.e., these last
decay into separate independently fluctuating droplets, whose density and typical size
decrease with increasing , while the gluon masses originate now from the coherent
quark condensate;
(5) At > m1/4 the Higgs field becomes too heavy and decouples completely, while the
magnetically charged excitations are heavy and confined into rare small size neutral
pairs.
6. Summary
As has been argued on a number of examples above, the mixed phases exist with their
properties qualitatively different from those of pure phases. And the appearance of mixed
phases is not an exception but rather a typical phenomenon in various gauge theories, both
supersymmetric and ordinary.
130
References
[1] E. Witten, Nucl. Phys. B 149 (1979) 285;
E. Witten, Ann. Phys. (N.Y.) 128 (1980) 363.
[2] E. Witten, Phys. Rev. Lett. 81 (1998) 2862, hep-th/9807109.
[3] V. Chernyak, Preprint BINP 98-61 (1998) (revised) hep-th/9808092.
[4] E. Witten, Phys. Lett. B 86 (1979) 283.
[5] S. Coleman, Ann. Phys. (N.Y.) 101 (1976) 239.
[6] G. Veneziano, S. Yankielowicz, Phys. Lett. B 113 (1982) 321.
[7] M.A. Shifman, A.I. Vainshtein, Nucl. Phys. B 296 (1988) 445.
[8] V. Chernyak, Phys. Lett. B 450 (1999) 65, hep-th/9808093.
[9] E. Witten, Nucl. Phys. B 507 (1997) 658, hep-th/9706109.
[10] I.I. Kogan, A. Kovner, M. Shifman, Phys. Rev. D 57 (1998) 5195, hep-th/9712046.
[11] I. Affleck, M. Dine, N. Seiberg, Phys. Lett. B 137 (1984) 187.
[12] N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19, hep-th/9407087.
[13] N. Seiberg, E. Witten, Nucl. Phys. B 431 (1994) 484, hep-th/9408099.
[14] K. Intrilligator, N. Seiberg, Nucl. Phys. B 431 (1994) 551, hep-th/9408155.
[15] S. Elitzur, A. Forge, A. Giveon, E. Rabinovici, Phys. Lett. B 353 (1995) 79, hep-th/9504080;
S. Elitzur, A. Forge, A. Giveon, E. Rabinovici, Nucl. Phys. B 459 (1996) 160, hep-th/9509130.
[16] S. Elitzur, A. Forge, A. Giveon, K. Intrilligator, E. Rabinovici, Phys. Lett. B 379 (1996) 121, hepth/9603051.
[17] A. Gorsky, A. Vainshtein, A. Yung, Nucl. Phys. B 584 (2000) 197, hep-th/0004087.

Low energy dynamics from deformed conformal

symmetry in quantum 4D N = 2 SCFTs
S.M. Kuzenko a , I.N. McArthur a , S. Theisen b,c
a School of Physics, The University of Western Australia, Crawley, W.A. 6009, Australia
b Max-Planck-Institut fr Gravitationsphysik, Albert-Einstein-Institut, Am Mhlenberg 1,
D-14476 Golm, Germany 1
c CERN Theory Division, CH-1211 Geneva 23, Switzerland
Received 23 October 2002; accepted 12 March 2003
Abstract
We determine the one-loop deformation of the conformal symmetry of a general N = 2
superconformally invariant YangMills theory. The deformation is computed for several explicit
examples which have a realization as world-volume theories on a stack of D3 branes. These include:
(i) N = 4 SYM with gauge groups SU(N), USp(2N) and SO(N); (ii) USp(2N) gauge theory with
one hypermultiplet in the traceless antisymmetric representation and four hypermultiplets in the
fundamental; (iii) quiver gauge theory with gauge group SU(N) SU(N) and two hypermultiplets
in the bifundamental representations (N,
N) and (
N, N). The existence of quantum corrections to the
conformal transformations imposes restrictions on the effective action which we study on a subset
of the Coulomb branch corresponding to the separation of one brane from the stack. In the N = 4
case, the one-loop corrected transformations provide a realization of the conformal algebra; this
deformation is shown to be one-loop exact. For the other two models, higher-loop corrections are
necessary to close the algebra. Requiring closure, we infer the two-loop conformal deformation.
PACS: 11.30.Pb; 11.30.Na
E-mail addresses: kuzenko@cyllene.uwa.edu.au (S.M. Kuzenko), mcarthur@physics.uwa.edu.au

(I.N. McArthur), theisen@aei-potsdam.mpg.de (S. Theisen).
1 Permanent address.
doi:10.1016/S0550-3213(03)00231-1
132
S.M. Kuzenko et al. / Nuclear Physics B 660 (2003) 131155
1. Introduction and summary

Four-dimensional conformal field theories have attracted much attention recently,
mainly due to the possibility to study them via a dual supergravity description; for a
review, see [1]. This was first proposed for maximally supersymmetric N = 4 YangMills
theory [2], which has long been the prime example of a four-dimensional conformally
invariant quantum field theory. Here the dual supergravity theory is type IIB supergravity
compactified on AdS5 S 5 , which is the near-horizon geometry of a stack of D3 branes.
One simple argument in favour of the gauge theorygravity correspondence is provided
by comparison of their symmetries. The isometry group SU(2, 2) SO(6) of AdS5 S 5
coincides with the conformal group of four-dimensional Minkowski space and the Rsymmetry group of N = 4 SYM theory. This agreement can be extended to the full
supergroup SU(2, 2|4), of which the above is the bosonic subgroup. From this comparison
of symmetries it is clear that it is the AdS5 factor of the compactification which is
responsible for the conformal invariance of the dual gauge theory. The fact that the betafunction of the field theory vanishes is reflected by the constancy of the type IIB dilaton.
Subsequently, many generalizations of the original proposal have been considered and dual
supergravity descriptions of conformal and confining gauge theories have been constructed
with various gauge groups and number of supercharges; see, e.g. [3].
In this paper we address the question to what extent one can infer the geometry of the
dual gravity backgroundif anyfrom the conformal field theory. As to the latter, we
restrict our attention to conformally invariant supersymmetric YangMills theories. At the
classical level they are invariant under conformal transformations, c I , with respect to
which the fields I transform in the standard way, specified by their tensorial structure and
their conformal weight. One may then ask whether this symmetry is still manifest (i.e.,
has the same functional form) in the effective action. The issue here is that quantization
requires gauge fixing and the latter can be shown to necessarily break (part of) the
conformal symmetry (see [4,5] and references therein). The effective action will thus not be
invariant under the same conformal transformations which were a symmetry of the classical
action. The change in the gauge fixing condition under conformal transformations can be
undone in the path integral by a compensating field-dependent gauge transformation. The
invariance of the path integral under combined conformal and gauge transformations leads
to modified conformal Ward identities for the effective action. In other words, the effective
action is invariant under deformed conformal transformations,
I
[]
= 0,
I
I = c I +
h L (L) I ,
(1.1)
L=1
and this deformation can, in principle, be computed at each loop order. Of course, the
deformation depends only on the parameters of those transformations which do not leave
the gauge fixing condition invariant, and these transformations are the special conformal
boosts. Having obtained the deformed conformal transformations, this imposes severe
restrictions on the general form of the effective action. Indeed, if the effective action were
133
invariant under the classical conformal transformations, each term in its loop expansion,
[] = S[] +
h L (L)[],
(1.2)
L=1
would be conformally invariant. The deformed conformal transformations, however, mix

different orders in the loop expansion of []. In particular, it turns out that even the
one-loop deformation, (1) I , contains some non-trivial information about the multi-loop
structure of the effective action.
If we now consider the field theory as living on a D3 brane in a ten-dimensional
spacetime, the deviation of the position of the brane from a chosen reference position
is parametrized by the vacuum expectation values of massless scalar fields. For the field
theory this corresponds to going to the Coulomb branch of the vacuum manifold. If the
ambient geometry is non-trivial, the effective action, which is constrained by requiring
it to be invariant under deformed transformations, should provide information about this
geometry. Analysis of the one-loop deformed conformal symmetry on the Coulomb branch
of N = 4 SU(N) super-YangMills theory has been carried out in [68].
When constructing SYM theories and their quantum-deformed symmetry properties we
must, first of all, ensure that the conformal symmetry survives quantization. A necessary
condition for this is the vanishing of the beta-function, which can be arranged by an
appropriate choice of matter fields. In the N = 4 SYM theory there is no freedom and the
theory is completely specified by a choice of gauge group. For N = 2 the beta-function is
one-loop exact and conformal invariance imposes a single condition on the second Casimir
invariants of the matter hypermultiplets [9]. For N = 1 the situation is more complicated
and one generally finds lines of superconformal fixed points in a higher-dimensional
moduli space of vacua [10] (this paper also includes an extensive list of references on
finite N = 1 theories). In the present paper we concentrate on conformally invariant SYM
theories with eight supercharges, i.e., N = 2. For these we will explicitly determine the
one-loop deformation of the conformal transformation properties of the fields.
The remainder of the paper is organized as follows. In Section 2 we first recall
some well-known facts about N = 2 SYM, including its background field quantization.
Following [7,8] we collect all necessary formulas which are needed to compute the
deformed conformal transformation. In Section 3 we do this, in full generality, for an
arbitrary N = 2 SYM at one loop order. Section 4 is devoted to a discussion of the
conformal deformation on the Coulomb branch with vanishing hypermultiplets. Here a
non-renormalization theorem guarantees that the one-loop deformation is exact.2 One test
of this is that the deformed algebra closes without the need to add higher loop contributions.
We then discuss to what extent invariance of the deformed transformation fixes the form
of the effective action on the Coulomb branch. In particular, if we choose the background
fields to correspond to moving one D3 brane away from a stack of N D3 branes, we are
2 This holds in t Hooft gauge. Changing from this gauge to, say, R gauge is equivalent to a non-local
field redefinition in the effective action, which leads to a restructure of the loop expansion accompanied by a
modification of the functional form of symmetries [8].
134
probing the geometry produced by them. This geometry should then determine the general
structure of the effective action.
In Section 5 we specialize to N = 4 theories, which are equivalent to N = 2 SYM
theories with one hypermultiplet in the adjoint representation of the gauge group. The
brane constructions of these theories are known: in the case of a stack of N D3 branes, one
obtains the gauge group SU(N); this is the case considered by Maldacena [2]. If one puts
N D3 branes on top of an orientifold D3 brane, one obtains, depending on the choice of the
orientifold projection, either SO(2N) or USp(2N) gauge groups [11] (in the SO case N can
be a half-integer). The near-horizon geometry of this brane configuration is AdS5 RP5 .
We compute the one-loop exact deformed conformal transformation for all these cases.
In Sections 6 and 7 we consider two examples of N = 2 theories. First, in Section 6,
we study the field theory which one obtains by placing N D3 branes on top of four D7
branes which are coincident with an O7 plane. This gives a USp(2N) gauge group and one
hypermultiplet in the traceless antisymmetric representation and four hypermultiplets in
the fundamental representation [1214]. Here the near-horizon geometry is AdS5 S 5 /Z2
where the Z2 action has a fixed S 3 S 5 [15]. Other examples of superconformal N = 2
theories are obtained by considering stacks of N D3 branes at an R4 /Zk singularity. This
leads to so-called quiver gauge theories [16,17] with gauge group SU(N)k and matter in
the bifundamental representation of adjacent gauge groups in the quiver diagram, which is
a k-gon in this simple case. The near-horizon geometry of this brane configuration is S 5 /Zk
where the Zk action leaves a S 1 S 5 invariant [18]. In Section 7 we consider the simplest
case of such theories, namely, k = 2. The generalization to other ks is straightforward.
While the one-loop deformation in N = 4 theories and on the pure Coulomb branch
of N = 2 theories is exact, this is no longer the case for the mixed branches of N = 2
theories. One consequence of this is that the conformal algebra does not close as long as
the gauge group has finite rank. Requiring closure one can, on the other hand, infer the
form of the higher loop corrections to the conformal transformation of the various fields.
We will determine them explicitly at two loop order.
2. N = 2 SCFTs
We consider a four-dimensional N = 2 superconformal field theory, which describes
the coupling of an N = 2 vector supermultiplet to a massless hypermultiplet in a (possibly
reducible) representation R of the gauge group G. The finiteness condition (which
coincides with the requirement of absence of one-loop divergences) [9] can be given in
the form
trAd W 2 = trR W 2 ,
(2.1)
where the subscript Ad denotes the adjoint representation, and W is an arbitrary complex
scalar field taking its values in the Lie algebra of the gauge group, W = Wa Ta , with
Ta = (Ta ) the gauge group generators.
The N = 2 vector multiplet is composed of a gauge field Vm , adjoint scalars W and
i
= W , and adjoint spinors i and i
W
= ( ) , where i = 1, 2. The hypermultiplet
is described by R-representation scalars and spinors (Qi , , ) and their conjugates
135
i , , ), where Q
i = (Qi ) . The Lagrangian (with g the coupling constant) is3
(Q

1
, W ]2
Dm W + 1 [W
g 2 L = trF F mn Fmn + D m W
4
2

i i
i
i m
i
+ i Dm i [W , i ] + i W,
2
2
i

Ta Qi 2
i {W
, W }Qi Q
i Ta Qj Q
i Dm Qi Q
j Ta Qi + 1 Q
Dm Q
2

i m Dm i m Dm 2 iW + 2 i W

i i
i i
i + i Qi + Q
+ Q
(2.2)
i i Qi ,
2
2
where Dm = m + iVm , and the generators are normalized such that trF (Ta Tb ) = ab in
self-interaction occurs after elimination of the
the fundamental representation. The QQ
auxiliary triplet, Xij = X(ij ) , which belongs to the off-shell N = 2 vector multiplet. This
self-interaction can be rewritten as4
i

i Ta Qj Q
(i Ta Qj ) Q
j Ta Qi + 1 Q
(i Ta Qj ) .
Ta Qi 2 = Q
Q
(2.3)
2
The model (2.2) admits a manifestly N = 2 supersymmetric formulation with finitely
many auxiliaries in terms of constrained superfields [19,20] (in this approach, the
hypermultiplet possesses an intrinsic off-shell central charge). It can also be formulated
in terms of unconstrained superfields, which involve infinitely many auxiliary fields, in
harmonic superspace [21]. In both superfield realizations, the R-symmetry SU(2)R is
manifest. The superconformal symmetry SU(2, 2|2) is manifest in the harmonic superspace
approach.
The moduli space of vacua of the theory under consideration is specified by the
following conditions:
W] = 0,
[W,
WQi = 0,
(i Ta Qj ) = 0,
Q
(2.4)
i = 0 a consequence of the first and second conditions. The solutions to the

with WQ
vacuum Eqs. (2.4) can be classified according to the phase of the gauge theory they give
rise to. In the pure Coulomb phase the rank of the gauge group is unreduced: generically
it corresponds to Qi = 0 and W = 0 and unbroken gauge group U (1)rank(G) . In the
(pure) Higgs phase the gauge symmetry is completely broken; there are no massless gauge
bosons. This requires Qi = 0. In the mixed phases there are some massless gauge bosons
but the rank of the gauge group is reduced.
3 Here and in the following, lower-case Latin letters from the middle of the alphabet, i, j, k, are used to denote
indices of the automorphism group of the N = 2 supersymmetry algebra, or the R-symmetry group SU(2)R .
Such indices are raised and lowered by antisymmetric tensors ij and ij , with 12 = 21 = 1, in the standard
i , (Qi ) = Q
i .
way: Qi = ij Qj , Qi = ij Qj , such that (Qi ) = Q
4 By giving up manifest SU(2) invariance, this potential can be brought to a more familiar form. Defining
R
, and thus Q
1 = Q and Q
2 = Q,
one obtains Q
(i Ta Qj ) Q
a Q|2
(i Ta Qj ) = 2|QT
Q1 Q and Q2 Q
1 D D , where D = Q T Q QT
aQ
.
a
a
2 a a
136
At tree level at energies below the symmetry breaking scale, we have free field massless
dynamics if the N = 2 vector multiplet (Vm , W, . . .) and the hypermultiplet (Qi , . . .) are
aligned along a particular direction in the moduli space of vacua. At the quantum level,
however, exchanges of virtual massive particles produce corrections to the action of the
massless fields. The aim of this work is to determine restrictions on the structure of the low
energy effective action which are implied by quantum conformal invariance of the theory.
We quantize the N = 2 SYM theory (2.2) in the framework of the background field
method (see [22,23] and references therein), by splitting the dynamical variables I =
(Vm , W, Qi , . . .) into the sum of background fields5 I = (Vm , W, Qi , . . .) and quantum
fields I = (vm , w, qi , . . .). The classical action, S[], is invariant under standard Yang
Mills gauge transformations
Vm = Dm ,
W = i[, W ],
Qi = i Qi , . . . ,
(2.5)
which, in a condensed notation [22], read

I = R I a []a ,
(2.6)
with
a [] the gauge generators and a infinitesimal gauge parameters. Upon background quantum splitting, the action S[ + ] is invariant under background gauge transformations
RI
I = R I a []a ,
I = R I a,J J a ,
(2.7)
and quantum gauge transformations

I = 0,
I = R I a [ + ]a .
(2.8)
The background field quantization procedure consists of fixing the quantum gauge freedom
while keeping the background gauge invariance intact by means of background covariant
gauge conditions. The effective action is given by the sum of all 1PI Feynman graphs which
are vacuum with respect to the quantum fields.
The quantum gauge freedom will be fixed by choosing the following background
covariant gauge conditions (often called t Hooft gauge):
a w + iq i Ta Qi iQ
i Ta qi ,
a W iWT
a = D va + iwT
(2.9)
and the gauge fixing functional6

1
SGF [] = 2 d4 x 2 .
(2.10)
2g
In Eq. (2.9), D denotes the background gauge covariant derivative, Dm = m + iVm .
Introducing

W
+ y,
= W
, Q
i = Y
Y=
(2.11)
= Y + y,
Y
Qi
5 For most of this section, the background fields are completely arbitrary. After Eq. (2.21), they will be taken
to be aligned along a particular direction in the moduli space of vacua.
6 A note on notation: while, e.g., in (2.4) W is matrix valued, it is an (adjoint) vector in (2.9). We will freely
switch from one to the other form to simplify expressions.
137
the gauge conditions can be rewritten in the abbreviated form

a y.
a Y iYT
a = D va + iyT
(2.12)
Under the quantum gauge transformations, = [, ] changes as

a (y + Y) + (y + Y)
Ta Y ( )a ,
a = Dm Dm a + YT
(2.13)
with = [, ] the FaddeevPopov operator. To define the effective action, [], let us
introduce the generating functional of connected quantum Greens functions, W [J, ],

iW [J,]
e
= N D Det [, ] exp i S[ + ] + SGF [, ] + JI I . (2.14)
Its Legendre transform,

, = W [J, ] JI I ,
I =
W [J, ],
JI
(2.15)
is related to the effective action [] as follows: [] = [ = 0, ]. In other words,

[] coincides with W [J, ] at its stationary point J = J [] defined by W [J, ]/J = 0.
By construction, [] is invariant under background gauge transformations.
The theory under consideration is conformally invariant, both at the classical and
quantum levels. The classical action does not change under standard linear conformal
transformations of the fields (these transformations differ in sign from those adopted in
[7,8]):
c Vm = Vm + m n Vn + Vm ,
c W = W + W,
1
3
c Qi = Qi + Qi ,
(2.16)
c = + mn Lmn + ,
2
2
where denotes any spinor field, Lmn the Lorentz generators in the spinor representation,
and = m m an arbitrary conformal Killing vector field,
1
1
(2.17)
m m ,
mn (m n n m ).
4
2
At the quantum level, conformal invariance is governed by the Ward identity [7,8]

1
[, ]
[] I

c I []
(2.18)
=
R
[
+
]
[,
][]
,
a
a
I
I =0
m n + n m = 2mn ,
where [] denotes the inhomogeneous term7 in the conformal transformation of [, ]:

c = ( + 2 ) + ,
(2.19)
= 2 m vm .
7 The general solution to the conformal Killing equation (2.17) is m = a m + x m + K m x n + bm x 2
n
2x m (b x), where the parameters a m and K mn = K nm generate Poincar transformations, dilatations and
bm special conformal boosts. By definition, = 2(b x), hence m = 2bm , and, therefore, the right-hand
side in (2.18) is non-vanishing for the special conformal boosts only.
138
In (2.18), the symbol denotes the quantum average in the presence of sources,

F [, ] = eiW [J,] N D F [, ] Det [, ]

exp i S[ + ] + SGF [, ] + JI I .
(2.20)
Eq. (2.18) should be treated in conjunction with the identity [7,8]

I
1
[, ]
I []
I

= + R a [ + ] [, ][, ] a
I
I
[, ] [ , + ] [, ],
,
=0
(2.21)
with I an arbitrary variation of the background fields. This identity allows one to express
the functional derivative [, ]/ at = 0 via []/.
It follows from Eqs. (2.18) and (2.21) that the effective action [] is invariant under
quantum corrected conformal transformations, as described by Eq. (1.1). In this paper we
will evaluate the one-loop quantum deformation, (1) , of conformal symmetry when the
fields are aligned along a particular direction in the moduli space of vacua.
In the above discussion, the fields I have been completely arbitrary. From now on,
the background N = 2 vector multiplet and hypermultiplet will be chosen to be aligned
along a fixed, but otherwise arbitrary, direction in the moduli space of vacua; in particular,
their scalar fields should solve Eqs. (2.4). At later stages in this work, we will be forced to
impose further restrictions on the fields, of the form
Vm = Vm (x)H,
W = W(x)H ;
Qi = Qi (x),
(2.22)
corresponding to a separation of spacetime and internal variables. Here H is a fixed

generator in the Cartan subalgebra, and a fixed vector in the R-representation space
of the gauge group, in which the hypermultiplet takes values, chosen so that H = 0 and
Ta = 0, cf. (2.4). The freedom in the choice of H and can be reduced by requiring
the field configuration (2.22) to be invariant under a maximal unbroken gauge subgroup.
Eq. (2.22) defines a single U (1) vector multiplet and a single hypermultiplet which is
neutral with respect to the U (1) gauge subgroup generated by H . It is worth noting that an
Abelian vector field and six neutral scalars in four spacetime dimensions is what we need
to describe a (static gauge) D3 brane moving in a ten-dimensional spacetime.
Consider background gauge transformations (2.7) with = (x)H which will be referred to as H -gauge transformations. They leave all background fields (2.22) unchanged,
except the Abelian gauge field Vm . By construction, such transformations leave invariant
the gauge-fixed action S[ + ] + SGF [[, ]] and, in fact, each term in its Taylor expansion in powers of the quantum fields, since the background gauge transformations of the
quantum fields are linear and homogeneous.
3. The one-loop deformation

For the purpose of loop calculations, we expand the action S[ + ] in powers of the
quantum fields and combine its quadratic part, S2 , with the gauge fixing functional, SGF .
139
Modulo fermionic contributions, the quadratic action is

S2 + SGF

1
1
W}qi
+ q i Dm Dm qi q i {W,
m + iv m Fm n vn w w
= 2 d4 x v m v
g
2

j Ta q i (qi Ta Qj + Q
j Ta qi )
+ q i Ta Qj + Q

m
w(D
+ 2i v (Dm W)w
m W)v m

i v m qi ,
+ 2i q i v m Dm Qi Dm Q
(3.1)
where is the FaddeevPopov operator at = 0,

W} + Q
i {Ta , Tb }Qi b
)a = Dm Dm {W,
(
a
m

a , }Y.
= D Dm a + Y{T
(3.2)
It is assumed in (3.1) that the background N = 2 vector multiplet and hypermultiplet are
aligned along a particular direction in the moduli space of vacua such as to satisfy (2.4).
The action (3.1) determines the background covariant propagators of quantum fields,
I (x) J (x ), which are required to evaluate (L).
The one-loop deformation of the conformal transformations of the fields can be
computed by minor modification of the method described in [7] for the case of N = 4
SYM. The conformal Ward identity (2.18) can be rewritten in the form

[, ]
[]

0 = c Vma
2 n Dm 1 vn a
Vma
vma =0

n 1
[, ]
[]

+ 2i vn a Ta y + Ta Y
+ c Y
Y
y =0

n
1 [, ]
[]

+ c Y
a + YTa vn a
.
(3.3)
2i yT

y
Y
=0
At the one-loop level,

2 n Dm 1 vn a

b (1) Y (1)YT
b Y),
= m n ab + Dm 1 Dn ab (1)Vnb i Dm 1 ab (YT
(3.4)
and

2i n 1 vn a (Ta y + Ta Y)

b (1)Y + Ta Y 1 ((1) YT
b Y)
= 1 Ta Y 1 ab YT
ab

iTa Y 1 Dm ab (1)Vmb ,
(3.5)
140
where

(1) Vma = 2i n Tc 1 ab vnb (x)vmc (x ) x =x ,

(1) Y = 2i n 1 ab vnb (x)Ta y(x ) x =x ,
(3.6)
(3.7)
and is given by Eq. (3.2). As in the N = 4 case [7], substitution of (3.4) and (3.5) into
(3.3) yields the appropriate one-loop versions of the combination

I + R I a [ + ] 1 [, ][, ] a ,
(3.8)
appearing in the identity (2.21) to allow conversion of derivatives of the effective action
with respect to quantum fields into derivatives of the effective action with respect to
background fields. The conformal Ward identity then takes the one-loop form
[]
[]
+ (c Y + (1) Y)
Vma
Y
[]
+ (1)Y)

+ (c Y
.

Y
0 = (c Vma + (1) Vma )
(3.9)
The one-loop deformations of the conformal field transformations are, therefore, given by
(3.6) and (3.7). The right-hand sides in (3.6) and (3.7) are non-local functionals of Vm
and Y, each of which can be represented as a sum of infinitely many local terms with
increasing number of derivatives of the fields. We are interested in evaluating (1) Vm and
(1)Y to first order in the derivative expansion.
To first order in the derivative expansion,

vmb wc = 2g 2 1 (Dm W) 1 bc = 2g 2 2 (Dm W) bc ,
(3.10)
where the last relation follows from the identity
b , Tc }Y(Dm W)cd ,
c , Td }Y = Y{T
(Dm W)bc Y{T
which is valid since WY = (Dm W)Y = 0 due to (2.4). As a result, we get

(1) Wa = 4ig 2 m trAd Ta 3 Dm W
(3.11)
(3.12)
where 3 | denotes the kernel of 3 at coincident spacetime points. Let us now turn to
the vector field variation. To first order in the derivative expansion, one similarly gets

vma vnb = ig 2 1 ab mn + 2g 2 2 Fmn ab .
(3.13)
When substituted into (3.6), the first term in the vector propagator yields a potentially
divergent contribution. However, it vanishes on symmetry grounds, as

Tc 2 ac = ifcad 2 dc ,
(3.14)
and the mass matrix in is symmetric under interchange of c and d, while fcad is totally
antisymmetric. As a result, Eq. (3.6) leads to

(1) Vma = 4ig 2 n trAd Ta 3 Fnm .
(3.15)
141
Since the expressions (3.12) and (3.15) are already of first order in derivatives, we can
approximate the operator defined in (3.2) by

2
a , Tb }Y,
ab m m ab + M2v ab ,
(3.16)
Mv ab Y{T
with Y constant. Then, a direct evaluation of (3.12) and (3.15) gives

g2 n
(3.17)
trAd Ta M2
v Dn W ,
2
8

g2
(1) Vma = 2 n trAd Ta M2
(3.18)
v Fnm .
8
These relations determine the quantum modification to the conformal transformations of
the bosonic fields of the N = 2 vector multiplet.
It is worth noting that the mass matrix M2v may possess some zero eigenvalues, and,
2
therefore, M2
v is not defined by itself. But in (3.18), for example, Mv occurs in the
2
combination Mv Fnm and the multiplier F projects out all zero eigenvalues of M2v .
Finally, let us analyse the hypermultiplet conformal deformation. We will use underi = (Q
i ). In
lined Greek letters to denote hypermultiplet indices, i.e., Qi = (Qi ) and Q
accordance with (3.7), we have

(1) Qi = 2i m 1 bc vmc (x)qi (x ) x =x (Tb ) .
(3.19)
(1) Wa =
To first order in the derivative expansion, we can approximate vm qi as

j Td qj qi 0 .
vmc qi = 2i 1 cd (Td Dm Qj ) q j qi 0 Dm Q
(3.20)
Here 0 denotes the propagator corresponding to the free quadratic action

1
W}qi
S0 = 2 d4 x q i m m qi q i {W,
g

j Ta q i (qi Ta Qj + Q
j Ta qi ) ,
+ q i Ta Qj + Q
(3.21)
with W and Qi constant. Introducing

qj

q=
,
q = q i , qi ,
j
q
the action can be rewritten in a more compact form

1
S0 = 2 d4 x q m m q q M2h q ,
2g
where the hypermultiplet mass matrix is

i j
2Cij (Ta Qk ) (Ta Qk )
2
Mh =
,
k Ta ) (Q
k T a )
2C ij (Q
i j
k
W} + 2(Ta Qk ) Q
Ta .
= {W,
(3.22)
(3.23)
(3.24)
142
Eq. (3.19) becomes

(1) Qi = 4 m (Tb ) 2 bd

j Td qj qi .
(Td Dm Qj ) q j qi 0 Dm Q
0
(3.25)
Due to the off-diagonal terms in the hypermultiplet mass matrix, the propagator qj qi 0
does not vanish in general. For the background field configurations we will consider below,
it does, however, vanish.
Unlike Eqs. (3.17) and (3.18), the variation (1)Q involves two different propagators
with different mass matrices, M2v and M2h . The transformation (3.25) for the hypermultiplets will have a form similar to (3.17) only if a special relationship exists between the
mass matrices M2v and M2h . The point is that (1) Q proves to be a linear combination of
terms proportional to the following (Euclidean) momentum integral

1
1
d4 k
4
2
2
2
2
(2) (k + M1 ) (k + M22 )
2
M22
M2
1
,
+
=
(3.26)
ln
2
2
2
2
2
2
32 (M1 M2 )
M1
32 (M12 M22 )
with M12 and M22 being some eigenvalues of M2v and M2h , respectively. Only for M1 =
M2 = M is the mass dependence of the form 1/(32 2M 2 ), as in (3.17). This, therefore,
raises the question: under what circumstances are there coinciding mass eigenvalues?
Let the background vector multiplet fields (Vm , W) be of the form (2.22), with H a
given generator in the Cartan subalgebra. It is useful to examine the implications of the
U (1) gauge symmetry (2.7) generated by H . We know that: (i) the background scalar
fields W and Qi are invariant under the H -gauge transformations; (ii) the quadratic action
(3.1) is invariant under the H -gauge transformations. These observations imply

H, M2v = 0,
(3.27)
H, M2h = 0,
and, therefore, the charge operator H and the mass matrices can be simultaneously
diagonalized. One further observes that the derivative interaction terms in (3.1),
i (1)
1 v m Fm n vn ,
L = w(D
m W)v m v m (Dm W)w
2 int
2

i (2)
i v m qi q i v m Dm Qi ,
L = Dm Q
2 int
(3.28)
(3.29)
are invariant under the H -gauge transformations. The H -invariance of L(1)

int implies that
(1)
each term on the right-hand side of Lint involves two quantum fields of opposite H charge; these fields have the same mass due to the first identity in (3.27). Similarly, the
(2)
(2)
H -invariance of Lint implies that each term on the right-hand side of Lint involves two
quantum fields of opposite H -charge. In order for these fields to have the same mass, it is
sufficient to require (using the condensed notation (3.22))

Dm Q v m M2h q = Dm Q M2v v m q,
(3.30)
143
as a consequence of (3.27). The latter requirement is satisfied provided Qi is of the form

(2.22). In this case, the off-diagonal terms in the hypermultiplet mass matrix (3.24) also
vanish.
Thus, when the conditions (2.22) are imposed, considerable simplification of the
expression (3.25) occurs. The vector mass matrix M2v can be diagonalized by an
appropriate choice of basis for the Lie algebra of the gauge group. Each non-zero mass
eigenvalue corresponds to a generator Ta which is broken, in the sense that [Ta , W] = 0
and/or Ta Qi = 0. The deformation (3.25) can be put in the form
(1) Qi =
g 2 m 1

Ta Ta Dm Qi .
8 2
MI2
I
(3.31)
2
2
Here, I labels the different
non-zero eigenvalues MI of the vector mass matrix Mv , and,
for a given I, the sum a is over all broken generators Ta corresponding to the mass
eigenvalue MI2 . To derive this result, one makes use of the fact that when the conditions
(2.22) are imposed, the propagator qj qi 0 in (3.20) vanishes, as there are no offdiagonal blocks in the hypermultiplet mass matrix. Further, as proven above, the interaction
(Td Dm Qj ) in the first term on the right-hand side of (3.20) only mixes hypermultiplets
and vectors of the same mass. Thus the propagator (3.20) decomposes into a sum of
terms, one for each non-zero mass eigenvalue of the vector mass matrix. For a given mass
eigenvalue MI2 , the operator ( 1 )cd takes the form ( 2 + MI2 )1 cd , while
j

1
q qi 0 = ig 2 2 + MI2 j i .
(3.32)
Substituting the resulting expression (3.20) into (3.19), the deformed hypermultiplet
transformation is

3
2
+ MI2
Ta Ta Dm Qi .
(1) Qi = 4ig 2 m
(3.33)
a
In momentum space, this yields integrals of the form (3.26) with M12 = M22 = MI2 .
An analogous result applies for the transformations (3.17), which can be cast in the form
(1) W =

g 2 m 1

Ta , [Ta , Dm W] ,
2
2
8
MI
I
(3.34)
and similarly for the vector transformation (3.18). Of course, the covariant derivatives in
(3.31) and (3.34) coincide with partial derivatives under the (gauge) choice (2.22).
4. Conformal deformation on the Coulomb branch

Relations (3.17) and (3.18) allow us to evaluate the one-loop deformation of conformal
symmetry on the Coulomb branch, where we restrict the analysis in this section to the
case where the scalars in the hypermultiplets are zero,8 i.e., we consider solutions to the
8 This is the generic case, but, e.g., the model discussed in Section 6 also allows for non-vanishing Q without
i
reducing the rank of the gauge group.
144
equations
W] = 0,
[W,
Qi = 0.
(4.1)
In accordance with (3.16), the mass matrix M2v now becomes

1 2
[W, ] = W, [W,
] .
Mv = W,
(4.2)
2
For definiteness, the variations (3.17) and (3.18) will be evaluated for a special superconformal theoryN = 2 super-YangMills with gauge group SU(N) and 2N hypermultiplets in the fundamental [9]. This example captures the general features which we wish
to highlight; in particular, the analysis of this section is directly applicable to the models
considered in Sections 6 and 7 after setting the hypermultiplets to zero. It is worth noting
that the finiteness condition (2.1) for the SU(N) model follows from the following identity
for the group SU(N): trAd = 2N trF .
First we briefly recall some well-known facts concerning the explicit structure of the
Coulomb branch of the theory under consideration (see, for instance, [24]). Up to a gauge
transformation, the general solution to the first equation in (4.1) is

W = 0.
W = diag(W1 , . . . , WN ),
(4.3)
Generically, when all eigenvalues of W are different, the gauge symmetry SU(N) is broken
to U (1)N1 ; then, Q = 0 is the only solution to the second and third equations in (2.4). If
k eigenvalues of W coincide, the gauge group SU(N) is broken to SU(k) U (1)Nk ; the
second and third equations in (2.4) have non-trivial solutions when some eigenvalues of W
vanish.
Here we will be interested in a maximally symmetric field configuration which contains
a single Abelian N = 2 vector multiplet,
1
diag(N 1, 1, . . . , 1),
H=
N(N 1)
(4.4)
and leaves the subgroup SU(N 1) U (1) SU(N) unbroken. With this choice for W
the only solution to the second equation in (2.4) is Q = 0.
For the N = 2 vector multiplet (4.4), the explicit structure of the variations (3.17) and
(3.18) can be read off from similar results for the N = 4 super-YangMills theory [7], see
also Section 5. One obtains
W = WH,
Vm = Vm H,
g 2 (N 1) ( n )n W
g 2 (N 1) ( n )Fnm
V
=
.
(4.5)
m
(1)

8 2
8 2
WW
WW
This result is universal: modulo an overall common factor, which is the number of
broken generators, the same deformation of conformal symmetry occurs for any N = 2
superconformal theory in the Coulomb phase described by a single U (1) vector multiplet.
The deformed conformal transformations (1.1), with the quantum corrections from all
loops and all orders in the derivative expansion taken into account, should realize the
conformal algebra, as can be justified by a direct generalization of the considerations given
in [5,25]. On the Coulomb branch, the one-loop quantum deformation to first order in
(1) W =
145
derivatives,
( n )Fnm
,

4WW
( n )n W
W = c W R 4
,

4WW
Vm = c Vm R 4
(4.6)
(4.7)
with R 4 = g 2 (N 1)/2 2 , realizes the conformal algebra (up to a pure gauge transformation) without the need to take any higher loop corrections into account. Using the nonrenormalization theorem of Dine and Seiberg [26], we will show that the above conformal
deformation is one-loop exact.
It should first be pointed out that the transformations (4.7) and (4.6) coincide with the
AdS5 isometries of a (static gauge) D3 brane embedded in AdS5 S 1 :

2

X
R2
X4
1
4
det
mn + 2 m X n X + Fmn 4 ,
S = 2 d x
g
R2
X
R
X2 = X X ,
(4.8)

and X = (X1 , X2 ) are defined by 2 W = X1 + iX2 . Of course,
where X2 = 2WW
the conformal invariance (4.7) and (4.6) does not uniquely fix the D3 brane action. But
in conjunction with N = 2 supersymmetric non-renormalization theorems, it severely
restricts the functional form of invariant actions.
Let us demonstrate, by analogy with Maldacenas analysis [2], that the conformal
symmetry (4.7) and simple N = 2 non-renormalization theorems allow one to restore part
of the low energy effective action of the form

= d4 x Leff (W, W,
W, W).

[W, W]
(4.9)
It is useful to introduce radial X and angular variables, 2 W = X exp(i). It follows

from (4.7) and (4.6) that ln X can be interpreted as a Goldstone field for partial symmetry
breaking of the conformal group SU(2, 2) to SO(4, 1). Similarly, can be treated as a
Goldstone field corresponding to spontaneous breakdown of the U (1)R factor in the N = 2
R-symmetry group U (1)R SU(2)R , of which the SU(2)R factor leaves W invariant. As
follows, for instance, from the techniques of non-linear realizations, the general form for
Leff is
Leff = X4 +

k
G
c2k Gmn m n ,
G = det(Gmn ),
(4.10)
k=0
with and ck constant parameters, and Gmn the induced metric on the D3-brane,
X2
R2
X2
R2
X
=
+
m Xn X + R 2 m n .
(4.11)
mn
m
mn
R2
X2
R2
X2
Of course, the numerical coefficient of the angular term, m n , in Gmn can be changed
at the expense of modifying the infinite series in (4.10). The F-independent part of the D3
Gmn =
146
brane Lagrangian in (4.8) is of the form (4.10) with c0 = R 4 = 1/g 2 and c2k = 0 for
k 1.
The theory we are considering is N = 2 superconformal, and thus neither the kinetic
term nor the scalar potential receive quantum corrections. These non-renormalization
properties and the fact that the field configuration (4.4) corresponds to a flat direction
imply that the part of Leff with at most two derivatives must coincide with the tree-level
Lagrangian, and, hence,
Leff =

1 m
Wm W + O 4 ,
2
g
(4.12)
implying that c0 = R 4 = 1/g 2 and c2 = 0. We see that the radial part of Leff
is completely fixed by the conformal symmetry (4.7) and N = 2 non-renormalization
theorems [2]. In addition, the DineSeiberg theorem [26] turns out to imply c4 = 0.
Let W (x, ) be the N = 2 chiral superfield strength with W(x) being its leading ( independent) component. The non-renormalization theorem of Dine and Seiberg states
that, in the Coulomb branch of N = 2 superconformal theories, the following manifestly
N = 2 supersymmetric quantum correction to the low energy effective action

c
ln W,
d4 x d4 d4 ln W
(4.13)
is one-loop exact. In components, this functional contains two special bosonic structures
with four derivatives (with F and
F the helicity +1 and 1 components of the field
strength Fmn ):
F F
F
F
,
2

(WW)
2 (W)2
( W)
,
2

(WW)
(4.14)
which precisely match the four-derivative structures appearing in the expansion of the
BornInfeld action (4.8), including the relative coefficient [27]. The explicit value of the
overall coefficient c in (4.13) for the theory under consideration was computed in [27
29] to be c = (N 1)/(4)2 . This confirms that c4 = 0. The coefficients c6 , c8 , . . . ,
in (4.10) may have non-vanishing values, in particular c6 = 0. The point is that the F6
quantum correction in the low energy effective action of the SU(N) gauge theory with
2N fundamental hypermultiplets is known to be different from the one coming from the
BornInfeld action (4.8) [30], and the leading part of the c6 term in (4.10) should occur in
the same superfield functional which contains those F6 terms which are not present in the
BornInfeld action.
In summary, the terms in Leff containing two and four derivatives of the fields coincide
with the corresponding terms in the D3 brane action (4.8), and the four-derivative term in
Leff is one-loop exact. This information suffices to argue that the parameter R 4 in (4.7) and
(4.6) cannot receive two- and higher-loop quantum corrections. Indeed, the transformation
(4.7) mixes the two-derivative and four-derivative terms in Leff , of which the former is
tree-level exact and the latter is one-loop exact.
147
5. N = 4 SYM
In the case of N = 4 SYM, which is just a special N = 2 SYM theory, the
hypermultiplet sector is composed of a single hypermultiplet in the adjoint representation
i = Q
ia . With the use of Eq. (2.4), one
of the gauge group, and hence Qi = Qia and Q
can now show that the off-diagonal elements of the hypermultiplet mass matrix M2h vanish,

Tc Qk a (Tc Qk )b = 0,
(5.1)
while the block diagonal pieces of M2h coincide with the vector multiplet mass matrix,
M2v ,
k

a , Tb }W + Q
i {Ta , Tb }Qi = M2v . (5.2)
W}ab + 2(Tc Qk )a Q
Tc = W{T
{W,
b
ab
As a consequence, Eq. (3.25) becomes

(1) Qia = 4ig 2 m trAd Ta 3 Dm Qi ,
(5.3)
which is of the same form as (3.12) and (3.15).

Evaluating (5.3), the result can be combined with Eqs. (3.17) and (3.18) to give the final
expression for the one-loop deformation of conformal symmetry:

g2 n
trAd Ta M2
v Fnm ,
8 2

g2
(1) Wa = 2 n trAd Ta M2
v Dn W ,
8

g2
(1) Qia = 2 n trAd Ta M2
v Dn Qi .
8
(1) Vma =
(5.4)
By construction, the mass matrix M2v is invariant under the R-symmetry group SO(6)R
Qi and Q
i . It is clear that the one-loop deformawhich rotates the six scalars W, W,
tion (5.4) respects the SO(6)R symmetry.
We want to evaluate (5.4) for a general semi-simple gauge group G. For this purpose it
is convenient to use a complex basis for the charged gauge bosons and their scalar partners.
We choose
them in
correspondence with the roots of the Lie algebra. That is
one-to-one

V = a V a Ta = V E + i V i Hi and likewise for the adjoint scalars. Here E is the
generator corresponding to the root normalized as tr(E E ) = tr(E E ) = , , and
Hi are the rank(G) Cartan subalgebra generators. They satisfy the commutation relations
[Hi , E ] = (Hi )E . With all background fields aligned along an arbitrary element H of
the Cartan subalgebra (cf. (2.22) with = H ), the vector-multiplet mass matrix becomes

2

2
Mv = X2 tr H, E [E , H ] = X2 (H ) , ,
(5.5)
+Q
i Qi ). All components of M2v along the Cartan subalgebra directions
with X2 = 2(WW
vanish: the neutral gauge bosons stay massless (adjoint breaking does not reduce the rank
of the gauge group) and they do not mix with the charged vector bosons at the quadratic
level.
148
It is now straightforward to compute, say, (1)W, for which we get from (5.4) (1)W
trAd (H 2 M2
v ). From the commutation relation [H, E ] = (H )E we read off H in the
adjoint representation. Then the factors (H ) cancel and the trace produces a numerical
coefficient equal to the number of roots with (H ) = 0, i.e., the number of broken
generators or, in other words, the number of massive gauge bosons.
To be specific, we consider the classical groups with the background chosen such that
one obtains the breaking patterns SU(N) SU(N 1) U (1), Sp(2N) Sp(2N 2)
U (1), SO(N) SO(N 2) U (1). These regular subgroups are obtained by removing
an extremal node from the Dynkin diagram of the gauge group. In the brane realization of
these theories this corresponds to moving one of the D3 branes away from the orientifold
O3 plane. This implies a particular Cartan subalgebra generator H (for SU(N) it has been
explicitly given in (4.4)) for which (H ) = , where is the fundamental weight
corresponding to the simple root associated with the removed node of the Dynkin diagram.
We then find,
g 2 n
n W,
8 2 X2

2(N 1) for SU(N),
with = 4N 2
for Sp(2N) and SO(2N + 1),
4N 1
for SO(2N),
(1) W =
(5.6)
and likewise for (1) Qi and (1) Vm . As mentioned above, is simply the number of broken
generators. In the SU(N) case, the deformation (1) W was derived in [6,7].
What remains to be shown is that the other components of the background fields
do not receive any one-loop deformation, i.e., that the deformation vanishes for Ta
such that tr(Ta H ) = 0. Indeed, for any direction a the deformation is proportional to
1
(H ) (Ta ) , where the sum is restricted to those roots for which (H ) = 0. This
vanishes if Ta is one of the E (since [E , E ] is never proportional to E ). If Ta = H with

)
tr(H H ) = 0 the deformation is proportional to (H
(H ) . Using the roots and fundamental
weights for the classical Lie algebras (see, e.g., Appendix of [31]) this can be explicitly
shown to vanish.
The results of this section easily generalize to N = 2 theories on the pure Coulomb

branch by simply setting X2 = 2WW.
It is worth discussing the results obtained in this section. To first order in the derivative
expansion, the one-loop deformed conformal transformations are:
Vm = c Vm R 4
W = c W R 4
( n )Fnm
,
+Q
j Qj )
4(WW
( n )n W
,
+Q
j Qj )
4(WW
Qi = c Qi R 4
( n )n Qi
,
+Q
j Qj )
4(WW
(5.7)
with R 4 = g 2 /4 2 and defined as above for the classical groups. These transformations
leave invariant the D3 brane action (4.8), which now involves six scalars X , where =
149
1, . . . , 6, defined by 2 W = X1 + iX2 and 2 Qi = Xi+1 + iXi+2 . The transformations

(5.7) realize the conformal algebra (up to a pure gauge transformation) without the need
to take into account any higher loop quantum corrections. The consideration of Section 4
implies that the parameter R 4 in (5.7) is one-loop exact. Unlike the situation with generic
N = 2 superconformal theories on the Coulomb branch, which we discussed in Section 4,
the low energy effective action in the N = 4 SYM theory is expected to be of the Born
Infeld form (4.8), at least in the large N limit (see [32,33] and references therein). For
this to hold,9 there should exist a host of (yet unknown) non-renormalization theorems in
N = 4 SYM (see, e.g., the discussion in [30]).
6. USp(2N ) SYM with fundamental and traceless antisymmetric hypermultiplets

Here we consider the N = 2 superconformal YangMills theory introduced in [12,13],
which is known to have a supergravity dual on AdS5 S 5 /Z2 [15]. It will be shown that
the structure of the quantum conformal deformation differs significantly from the N = 4
SYM case.
The gauge group is USp(2N) = Sp(2N, C) U (2N), and the theory contains hypermultiplets in two representations of the gauge group: four hypermultiplets10 QF in the
fundamental and one hypermultiplet QA in the antisymmetric traceless representation of
USp(2N). The Lie algebra usp(2N) is spanned by 2N 2N matrices (i ) satisfying the
constraints

0
1N
.
= ,
J = J T =
T J + J = 0,
(6.1)
1N 0
We will use Greek letters to denote the components of USp(2N) spinors, QF =
(Q ), and also make use of the symplectic metric J = (J ) = (J ) for raising and
lowering USp(2N) spinor indices; for example, J and J . The
antisymmetric traceless representation of USp(2N) is realized by second rank tensors of
the form

tr QA = 0,
Q = Q .
QA = Q ,
(6.2)
The hypermultiplets QF and QA transform under USp(2N) as
QF = i QF ,
QA = i[, QA ].
(6.3)
It is worth noting that the symmetric representation of USp(2N) can be identified with the
Lie algebra sp(2N, C), i.e., = ( ) sp(2N, C) iff = . We also note that the
finiteness condition (2.1) is met due to the following property of USp(2N) representations:
trAd W 2 4 trF W 2 trA W 2 = 0.
(6.4)
9 The claim [7,8] that the deformed conformal symmetry (5.7), SO(6)
R invariance and the known N = 4
non-renormalization theorems (which we discussed in Section 4 of the present paper) uniquely fix the scalar part,
Leff (X , m X ), of the low energy effective Lagrangian in N = 4 SYM, is incorrect.
10 The SU(2) indices of the hypermultiplets are suppressed.
R
150
The background fields W, QF and QA are chosen to solve Eqs. (2.4) defining the
classical moduli space. Up to a gauge transformation, the general solution to the first
equation in (2.4) is W = diag(W1 , W2 , . . . , WN , W1 , W2 , . . . , WN ), with the W s
arbitrary complex numbers. Further analysis is restricted to the choice W1 = 0 and
W2 = = WN = 0; in this case, the unbroken gauge subgroup, USp(2N 2) U (1),
is maximal. We will also impose the requirement that QF and QA , which must be
solutions to the second and third equations in (2.4), be invariant under the unbroken group
USp(2N 2) U (1). This leads to11
W
W = diag(1, 0, . . . , 0, 1, 0, . . . , 0),

2
N1
QF = 0,
N1
Q
QA =
diag(N 1, 1, . . . , 1, N 1, 1, . . ., 1).

2N(N 1)
N1
(6.5)
N1
The non-vanishing background fields constitute the bosonic sector of an Abelian N = 2

and a single neutral hypermultiplet, (Qi , Q
j ). Since the
vector multiplet, (Vm , W, W),
background is of the form (2.22), the formulas (3.31) and (3.34) can be applied to
determine one-loop deformations of the conformal transformations once the mass matrices
are known.
The mass matrix M2v for quantum fields in the adjoint representation has 4(N 1)
+ N Q
j Qj and two eigenvectors with the
eigenvectors with the eigenvalue WW
N1

eigenvalue 4WW.
The massive degrees of freedom in the quantum adjoint scalars w are
parametrized in the form

1 Z
R
w=
(6.6)
T ,
2 S Z
with
Z=
0
z1
z2T
0N1
R=
2
r
rT
0N1
S=
2
s
sT
0N1

, (6.7)
where z1 , z2 , r and s are (N 1)-vectors. The fields which form the components of z1 , z2 ,
+ N Q
j Qj , while the scalars
r and s are eigenvectors of M2v with the eigenvalue WW
N1
2

and are eigenvectors of Mv with the eigenvalue 4WW.
The off-diagonal elements of the mass matrix M2h (3.24) for the hypermultiplet in the
antisymmetric representation vanish, since Qi Qi = 0. There are 4(N 1) eigenvectors
+ N Q
j Qj ; the remaining eigenvalues vanish. The
of M2h with the eigenvalue WW
N1
corresponding massive quantum degrees of freedom can be parametrized in the form

1 A B
q=
(6.8)
T ,
2 C A
11 The background fields are chosen to yield correctly normalized kinetic terms in the classical action,

mW + mQ
i m Qi ).
12 d4 x ( m W
g
where
A=
0
a1
a2T
0N1
B=
0 bT
b 0N1
C=
0
cT
c 0N1
151
(6.9)
Again, a1 , a2 , b and c are (N 1)-vectors.

Although some of the hypermultiplet degrees of freedom in the fundamental representation are massive in the chosen background, they decouple.
The deformed conformal transformations (3.31) and (3.34) take the form

g2 n
1
N 1
+
,
(1) Vm = 2 Fnm

+ N Q
j Qj
4
2WW
WW
N1

g2 n
N 1
1
(1) W = 2 (n W)
(6.10)
+
,

+ N Q
j Qj
4
2WW
WW
N1
(1) Qi =
g2 n
N
.
(n Qi )
N j
2

4
WW + N1
Q Qj
(6.11)
The vector multiplet and the hypermultiplet transformations differ due to eigenvectors with

eigenvalue 4WW,
which are present in M2v but absent in M2h .
Unlike the situation in the N = 4 SYM theory, the one-loop deformed transformations
(c + (1) ) specified in (6.10) and (6.11) do not realize the conformal algebra for finite
values of N , but only in a large N limit. To realize the conformal algebra for finite N , it is
necessary to include two and higher-loop deformations to the conformal transformations
of the fields; these can be determined order by order from the requirement of closure of the
conformal algebra. In particular, the two-loop deformation can be shown by this means to
have the form

1 Ng 2 2 n
j (p Qj )(W, Q),
Fnm p Q
(2) Vm =
2
N 4

1 Ng 2 2 n
j (p Qj )(W, Q),
(n W) p Q
(2) W =
2
N 4

1 Ng 2 2 n
(p W)(W, Q),
(n Qi ) p W
(2) Qi =
(6.12)
2
N 4
where
(W, Q) =
1
+
(WW
N j
2
N1 Q Qj )

1
1
.

+ N Q
j Qj
2WW
WW
N1
(6.13)
The field theory under consideration possesses the global symmetries SU(2)R
U (1)R SO(8) SU(2) of which the first two factors denote the R-symmetry. The group
SO(8) rotates the four fundamental hypermultiplets, while SU(2) acts on the antisymmetric
hypermultiplet.12 Since QF = 0, we cannot probe SO(8) for the background chosen. But
12 In the harmonic superspace approach [21], the symmetries SO(8) and SU(2) can be realized as (a combination of flavour and) PauliGrsey transformations of q + hypermultiplets.
152
the symmetries SU(2)R U (1)R SU(2) should be manifestly realized in the low energy
effective action. It is not difficult to see that both the one-loop and two-loop conformal
deformations respect these symmetries.
In the limit N , the one-loop deformations (6.10) and (6.11) lead to conformal
transformations of the form (5.7), which are symmetries of the BornInfeld action (4.8).
On symmetry grounds, when N is finite, the low energy effective action of the field theory
under consideration cannot have the BornInfeld form considered in Sections 4 and 5.
7. KachruSilverstein model
Here we consider the simplest quiver gauge theory [1618]N = 2 super-YangMills
theory with gauge group SU(N) SU(N) SU(N)L SU(N)R and two hypermultiplets,
i , in the representations (N,
Hi and H
N) and (
N, N) of the gauge group. Both Hi and
i carry an index of the automorphism group of the N = 2 supersymmetry algebra. The
H
hypermultiplets transform under SU(N)L SU(N)R as
H = iL H iH R ,
= iR H
iH
L ,
H
(7.1)
where = a ta , with ta the generators of SU(N). For simplicity we take the gauge coupling
constants in the two SU(N) factors to be equal.
Eq. (2.4) specifies the flat directions in massless N = 2 super-YangMills theories.
We are interested in those solutions of (2.4) in the KachruSilverstein model which
allow for non-vanishing hypermultiplet components. Notationally, we now have W =
i ).
WL 1 + 1 WR and Qi = (Hi , H
Up to a gauge transformation, the general solution to the first equation in (2.4),
= 0, is W = WL 1 + 1 WR , with WL and WR diagonal traceless matrices.
[W, W]
The second equation in (2.4), WQi = 0, becomes
HW
L = 0.
WR H
WL H HW R = 0,
(7.2)
The
These equations are obviously solved by any diagonal matrices WL = WR , H and H.

third equation in (2.4), Q(i Ta Qj ) = 0, is now

= 0,
(i ta Hj ) tr H
(i ta H

tr H
j)

tr H t H tr H t H = 0.
(7.3)
(i a
(i a
j)
j)
The moduli space of vacua of the model thus includes the field configuration13
W
diag(N 1, 1, . . . , 1),
N 2(N 1)
Qi
i =
diag(1, 0, . . . , 0),
Hi = H
2
WL = WR =
(7.4)
= H , which appears to solve (7.3) at first sight, is in fact not a solution. The conjugation
13 The choice H

i
i
) = H
i imply (H
i

i , and hence H
i = H
i .
(Hi ) = H
i , (H
i ) = H

rules (Qi ) = Q
i
153
which preserves an unbroken gauge group SU(N 1) SU(N 1) together with the
diagonal U (1) subgroup in SU(N)L SU(N)R associated with the W chosen. In such a
1
j Qj
background, the hypermultiplet mass matrix has 4(N 1) eigenvalues N1
WW+ N1 Q
4(N1) j
and one eigenvalue N 2 Q Qj . The massive degrees of freedom can be parametrized in
the form

i
sTi
siT
i
,
h i =
.
hi =
(7.5)
ri 0N1
ri 0N1
Here, all of the (N 1)-vectors ri , ri , si and s i are eigenvectors of M2h with the
1
j Qj , while the eigenvector with eigenvalue 4(N1)
j Qj
eigenvalue N1
WW + N1 Q
Q
N2
is (i
i )/ 2. This linear combination is orthogonal to the massless eigenvector
(i + i )/ 2 corresponding to an unbroken linear combination of U (1) generators from
SU(N)L and SU(N)R .
The mass matrix for the adjoint scalars and the gauge bosons also has 4(N 1)
1
j Qj and one eigenvalue 4(N1)
j Qj . The adjoint massive
eigenvalues N1
WW + N1 Q
Q
N2
degrees of freedom can be parametrized in the form
!
2(N1)
T i
T
r

s
L
1
L
L
N
,
!
wL =
2
2N
rL + isL
N(N1) L 1N1
!
2(N1)
R
rRT isRT
1
N
.
!
wR =
(7.6)
2
2N
rR + isR
N(N1) R 1N1
The (N 1)-vectors rL , sR , rL and sR are eigenvectors of M2v with the eigenvalue
1
j Qj , while the eigenvector with eigenvalue 4(N1)
j Qj is (L
+ N1 Q
Q
N1 WW
N2
R )/ 2.
1
j Qj couple
The scalars (both adjoint and hypermultiplet) with mass N1
WW + N1 Q
to the gauge bosons of the same mass via a derivative of the correspondingbackground
scalar. However, for the adjoint scalars, the linear combination (L R )/ 2 does not
couple to the gauge bosons, and so there is no contribution from this mass eigenstate to the
quantum corrected transformations.
This is to be contrasted with the hypermultiplet mass
j Qj , which couples to the gauge boson

eigenstate (i i )/ 2 with the mass 4(N1)
Q
N2
j Qj , giving rise to an additional term in the one-loop hypermultiplet
with mass 4(N1)
Q
N2
transformation compared with the adjoint scalar transformation. This is the reverse of
the situation encountered in the USp(2N) example. The one-loop conformal deformations
(3.6) and (3.7) are
(1) Vm =
(1) W =
N
g2 n
Fnm N
,
+Q
j Qj
4 2
WW
N1
g2 n
N
(n W) N
,
+Q
j Qj
4 2
WW
N1
(7.7)
154
(1) Qi =
g2 n
(n Qi )
4 2

N 1
1
.
+
N
j Qj
j
4Q
N1 WW + Q Qj
(7.8)
Again, the one-loop deformed transformations (c + (1)) specified by (7.7) and (7.8)
only provide a realization of the conformal algebra in a large N limit. For finite values of
N, it is necessary to include higher-loop corrections, which can again be determined order
by order in g 2 N by requiring closure of the algebra of conformal transformations of the
fields. Using this procedure, the two-loop conformal deformation is found to be

1 Ng 2 2 n
j (p Qj )(W, Q),
Fnm p Q
(2) Vm =
N 4 2

1 Ng 2 2 n
j (p Qj )(W, Q),
(n W ) p Q
(2) W =
2
N 4

1 Ng 2 2 n
(p W)(W, Q),
(n Qi ) p W
(2) Qi =
(7.9)
N 4 2
where
(W, Q) =
1
N
j
N1 WW + Q Qj

2

4Qj Qj
1
N
j
N1 WW + Q Qj

.
(7.10)
The field theory under consideration possesses the global symmetries SU(2)R
. It is not

U (1)R SU(2), where the group SU(2) mixes the hypermultiplets H and H
difficult to see that both the one-loop and two-loop conformal deformations respect these
symmetries.
Acknowledgements
The work of S.T. is supported in part by GIF, the German-Israeli Foundation for
Scientific Research. He would like to thank the Theory Division of CERN where this
work was completed and A. Armoni and A. Font for valuable discussions. The work of
S.M.K. and I.N.M. is partially supported by a University of Western Australia Small Grant
and an ARC Discovery Grant. S.M.K. is thankful for kind hospitality extended to him at
the Max Planck Institute for Gravitational Physics (Albert Einstein Institute), Golm where
this project was commenced. His work was also supported in part by the Alexander von
Humboldt Foundation and the Australian Academy of Science.
References
[1] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183, hep-th/9905111.
[2] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
J.M. Maldacena, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
[3] I.R. Klebanov, M.J. Strassler, JHEP 0008 (2000) 052, hep-th/0007191;
C.P. Herzog, I.R. Klebanov, P. Ouyang, D-branes on the conifold and N = 1 gauge/gravity dualities, hepth/0205100;
[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]
155
J.M. Maldacena, C. Nunez, Phys. Rev. Lett. 86 (2001) 588, hep-th/0008001;

A. Loewy, J. Sonnenschein, JHEP 0108 (2001) 007, hep-th/0103163.
E.S. Fradkin, M.Y. Palchik, Phys. Lett. B 147 (1984) 86;
E.S. Fradkin, M.Y. Palchik, Phys. Rep. 300 (1998) 1.
E.S. Fradkin, M.Y. Palchik, Conformal Quantum Field Theory in D-Dimensions, Kluwer Academic,
Dordrecht, 1996.
A. Jevicki, Y. Kazama, T. Yoneya, Phys. Rev. Lett. 81 (1998) 5072, hep-th/9808039;
A. Jevicki, Y. Kazama, T. Yoneya, Phys. Rev. D 59 (1999) 066001, hep-th/9810146.
S.M. Kuzenko, I.N. McArthur, Nucl. Phys. B 640 (2002) 78, hep-th/0203236.
S.M. Kuzenko, I.N. McArthur, Phys. Lett. B 544 (2002) 357, hep-th/0206234.
P.S. Howe, K.S. Stelle, P.C. West, Phys. Lett. B 124 (1983) 55.
R.G. Leigh, M.J. Strassler, Nucl. Phys. B 447 (1995) 95, hep-th/9503121.
E. Witten, JHEP 9807 (1998) 006, hep-th/9805112.
O. Aharony, J. Sonnenschein, S. Theisen, S. Yankielowicz, Nucl. Phys. B 493 (1997) 177, hep-th/9611222.
M.R. Douglas, D.A. Lowe, J.H. Schwarz, Phys. Lett. B 394 (1997) 297, hep-th/9612062.
O. Aharony, J. Pawelczyk, S. Theisen, S. Yankielowicz, Phys. Rev. D 60 (1999) 066001, hep-th/9901134.
A. Fayyazuddin, M. Spalinski, Nucl. Phys. B 535 (1998) 219, hep-th/9805096;
O. Aharony, A. Fayyazuddin, J.M. Maldacena, JHEP 9807 (1998) 013, hep-th/9806159.
M.R. Douglas, G.W. Moore, D-branes, quivers, and ALE instantons, hep-th/9603167.
C.V. Johnson, R.C. Myers, Phys. Rev. D 55 (1997) 6382, hep-th/9610140.
S. Kachru, E. Silverstein, Phys. Rev. Lett. 80 (1998) 4855, hep-th/9802183.
R. Grimm, M. Sohnius, J. Wess, Nucl. Phys. B 133 (1978) 275.
M.F. Sohnius, Nucl. Phys. B 138 (1978) 109.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E.S. Sokatchev, Harmonic Superspace, Cambridge Univ. Press,
Cambridge, 2001.
B.S. DeWitt, Phys. Rev. 162 (1967) 1195;
B.S. DeWitt, Phys. Rev. 162 (1967) 1239.
G. t Hooft, in: Proceedings of the 12th Winter School of Theoretical Physics in Karpacz, Acta Univ.
Wratislavensis 368 (1976) 345;
B.S. DeWitt, in: C. Isham, R. Penrose, D. Sciama (Eds.), Quantum Gravity II, Oxford Univ. Press, New
York, 1981, p. 449;
L.F. Abbott, Nucl. Phys. B 185 (1981) 189;
L.F. Abbott, Acta Phys. Pol. B 13 (1982) 33;
D.G. Boulware, Phys. Rev. D 23 (1981) 389;
C.F. Hart, Phys. Rev. D 28 (1983) 1993.
P.C. Argyres, M.R. Plesser, N. Seiberg, Nucl. Phys. B 471 (1996) 159, hep-th/9603042.
M.Ya. Palchik, in: I.A. Batalin, C.J. Isham, G.A. Vilkovisky (Eds.), in: Quantum Field Theory and Quantum
Statistics, Vol. 1, Adam Hilger, Bristol, 1987, p. 313.
M. Dine, N. Seiberg, Phys. Lett. B 409 (1997) 239, hep-th/9705057.
F. Gonzalez-Rey, B. Kulik, I.Y. Park, M. Rocek, Nucl. Phys. B 544 (1999) 218, hep-th/9810152.
E.I. Buchbinder, I.L. Buchbinder, S.M. Kuzenko, Phys. Lett. B 446 (1999) 216, hep-th/9810239.
D.A. Lowe, R. von Unge, JHEP 9811 (1998) 014, hep-th/9811017.
I.L. Buchbinder, S.M. Kuzenko, A.A. Tseytlin, Phys. Rev. D 62 (2000) 045001, hep-th/9911221.
N. Bourbaki, Elements of Mathematics, Lie Groups and Lie Algebras, Springer-Verlag, Berlin, 2002,
Chapters 46.
I. Chepelev, A.A. Tseytlin, Nucl. Phys. B 511 (1998) 629, hep-th/9705120;
I. Chepelev, A.A. Tseytlin, Nucl. Phys. B 515 (1998) 73, hep-th/9709087.
A.A. Tseytlin, in: M. Shifman (Ed.), Yuri Golfand Memorial Volume, World Scientific, Singapore, 2000,
p. 417;
A.A. Tseytlin, hep-th/9908105.

Energy-charge dependence for Q-balls

T.A. Ioannidou a,1 , V.B. Kopeliovich b , N.D. Vlachos c
a Mathematics Division, School of Technology, University of Thessaloniki, Thessaloniki 54124, Greece
b Institute for Nuclear Research of Russian Academy of Sciences, Moscow 117312, Russia
c Physics Department, University of Thessaloniki, Thessaloniki 54124, Greece
Received 4 March 2003; accepted 28 March 2003
Abstract
We show that many numerically established properties of Q-balls can be understood in terms of
analytic approximations for a certain type of potential. In particular, we derive an explicit formula
between the energy and the charge of the Q-ball valid for a wide range of the charge Q.
1. Introduction
As shown by Coleman [1], the existence of Q-balls is a general feature of scalar
field theories carrying a conserved U (1) charge [2]. Q-balls can be understood as bound
states of scalar particles and appear as stable classical solutions (nontopological solitons)
carrying a rotating time dependent internal phase. They are characterized by a conserved
nontopological charge Q (Noether charge) which is responsible for their stability (see, for
example, Refs. [3,4]). These features differentiate the Q-ball interaction properties from
those of the topological solitons since here the charge Q can take arbitrary values in a
specific range, allowing for the possibility of charge transfer between solitons during the
interaction process.
The concepts associated with Q-balls are extremely general and occur in a wide variety
of physical contexts [5]. Q-balls are a generic consequence of the Minimal Supersymmetric
Standard Model (MSSM) [6] where leptonic and baryonic balls may exist. In this context
the conserved charge is associated with the U (1) symmetries leading to baryon and lepton
E-mail addresses: t.ioannidou@ukc.ac.uk (T.A. Ioannidou), kopelio@al20.inr.troitsk.ru
(V.B. Kopeliovich), vlachos@physics.auth.gr (N.D. Vlachos).
1 Permanent address: Institute of Mathematics, University of Kent, Canterbury CT2 7NF, UK.
doi:10.1016/S0550-3213(03)00266-9
T.A. Ioannidou et al. / Nuclear Physics B 660 (2003) 156168
157
number conservation, and the relevant U (1) fields correspond to either squark or slepton
particles. Thus, the Q-balls can be thought of as condensates of either squark or slepton
particles. It has been suggested that such condensates can affect baryogenesis via the
AffleckDine mechanism [7] during the post-inflationary period of the early universe.
Then, two interesting possibilities occur: (i) if the Q-balls are stable and avoid evaporation
into lighter stable particles like protons, they are cosmologically important since they can
contribute to the dark matter content of the universe [8]; (ii) if they are unstable, they
decay in a nontrivial way into baryons protecting them from erasure through sphaleron
transitions [9].
Up till now, comprehensive studies of these objects have been made by using either
numerical simulations [311] or some analytic considerations [1,12,13]. In our approach
we will analytically identify the explicit relation between the energy and the charge of the
Q-balls by using a semi-Bogomolny argument in the energy density. Subsequently, we will
show that similar results can be obtained by using the WoodsSaxon ansatz for describing
the Q-ball profile function; an approach which has been successfully applied to describe
analytically the properties of multiskyrmions in three [14] and two spatial dimensions [15].
This way, some universal properties of Q-balls in the thin-wall limit can be established.
2. Energy of the Q-balls

Although Q-balls can exist in a variety of field theoretical models, we will consider
the U (1) Goldstone model describing a single complex scalar field in three spatial
dimensions with potential U (||). The Lagrangian is

1
L = U || ,
(1)
2
where the potential is only a function of || and has a single minimum at = 0. This is
equivalent of stating that there is a sector scalar particles (mesons) carrying U (1) charge
and having mass squared equal to 12 U (0). The corresponding energy functional is given
by

3
1
1
2
2
|t | + || + U || d x.
E=
(2)
2
2
The model has a global U (1) symmetry leading to the conserved Noether current
1
).
(
2i
The conserved Noether charge Q is

1
d 3 x.
t t )
Q=
(
2i
J =
(3)
(4)
A stationary Q-ball solution has the form

= eit f (r),
(5)
158
where f (r) is a real radial profile function which satisfies the ordinary differential equation
2 df
d 2f
=
2 f + U (f )
r dr
dr 2
(6)
with boundary conditions f () = 0 and f (0) = 0.

This equation can either be interpreted as describing the motion of a point particle
moving in a potential with friction [1], or in terms of Euclidean bounce solutions [16].
In each case the effective potential being Ueff (f ) = 2 f 2 /2 U (f ) leads to constraints
on the potential U (f ) and the frequency in order for a Q-ball solution to exist. Firstly, the
effective mass of f must be negative. If we consider a potential U (f ) which is nonnegative
2 > 0 then one can deduce that < .
and satisfies U (0) = U (0) = 0, U (0) = +
+
2
Furthermore, the minimum of U (f )/f must be attained at some positive value of f , say
2 = 2U (f )/f 2 .
0 < f0 < and existence of the solution requires that > where
0
0
Hence, Q-balls exist for all in the range < || < + .
Thus, the charge Q of a stationary Q-ball solution simplifies to
.
Q = I [f ] = 4

r 2 f 2 (r) dr,
(7)
where I [f ] is the moment of inertia. Numerical and analytical methods have shown that
when the internal frequency is close to the minimal value , the profile function is almost
constant, implying that the charge (7) is large (thin-wall approximation). On the other hand,
when the internal frequency approaches the maximal value + the profile function falls off
very quickly (thick-wall approximation). In the thick-wall approximation the behavior of
the charge Q depends on the particular form of the potential and the number of dimensions
[13]. In the case studied here we show that Q as + .
In order to derive the minimum of the energy of the Q-ball at fixed charge Q, it is
convenient to represent the energy in the form
Q2
+ 4
EQ =
2I [f ]

f 2
+ U (f ) r 2 dr
2
(8)
where the stabilizing role of the Q-ball rotational part is obvious. Note that, without
rotational energy the Q-ball would collapse since EQ 0 as f (r) 0 everywhere except
at the origin r = 0. This is similar to the case of rotating skyrmions when the 4th order
Skyrme term is omitted in the Lagrangian. The zero mode (rotational) quantum correction
to the energy, which is proportional to (I [f ])1 plays a stabilizing role in this case.
The choice of the potential is not unique, the standard requirement is that the function
U (f )/f 2 has a local minimum at some value of f different from zero. Here we will
consider the following potential

2
U (f ) = f 2 1 + 1 f 2 .
Note that + = 2 and = 2 and so that stable Q-balls exist for 2 < < 2.
(9)
159
3. Semi-Bogomolny argument
We now proceed to obtain an ansatz for the profile function by applying a semiBogomolny argument [17] in the energy functional (8), (9). Initially, this approach was
applied to the Skyrme model [18], where it was shown that the lower energy bound is
proportional to the topological charge. Although the model studied here is not a topological
one (i.e., there is no topological charge), the Bogomolny argument can still be applied
and leads to an upper energy bound. This way, the profile function satisfies an exactly
soluble first order differential equation and the corresponding energy and charge density
can be easily derived. The same approach was applied for description of multiskyrmions
properties, as presented in [14,15].
The energy (8), (9) using the Bogomolny argument can be expressed as

1
1 2
2
2 2
+
(f ) + f 1 f
r 2 dr
Q + 4
EB =
2
2

2 2

1
f
2
+
=
+f 1f
r dr 4
2 f f 1 f 2 r 2 dr
Q + 4
2
2

1
Q 4
+

(10)
2 f f 1 f 2 r 2 dr.
2
The equality is satisfied when the total square term is zero which gives the semiBogomolny equation

f = 2f 1 f2 .
(11)
The corresponding profile function has the simple form
1
f (r) =
(12)
, CB > 0
1 + CB exp(2 2 r)
which
boundary conditions f (0) = 1/ 1 + CB and f () =
0, while f (0) =
satisfies the 3/2
2 CB /(1 + CB ) . Note that the asymptotic
value of (12): f exp( 2 r) is in agreement with the equation of motion (6) for = 2 (i.e., for large values of Q). In fact, inside
the Q-ball f 0 and f 1, while outside the Q-ball f = 0 and f = 0 and so (12) describes accurately the profile function of the Q-ball outside and inside its region where the
last term in (10) vanishes. However, (12) does not describe accurately the profile function
of the Q-ball on its surface (the so-called transition region). Although, in the thin-wall
approximation the analytical and numerical results (presented in Table 1) converge as Q
increases since the relative contribution of the surface decreases like Q1/3 at large Q.
For the specific form of the profile function (12) the charge and the energy can be
evaluated explicitly. We find that

3
2
1
ln(CB ) + ln(CB ) + Li3 [CB ] ,
QB =
6
2 2 6
2

2
1
1
2
1
EB =
+
Q+
+ ln(CB ) + ln 1 +
+ Li2 [CB ] ,
2
4
6
2
CB
(13)
z Lin1 (y)
dy and Li1 (y) = ln(1 y) is the polylogarithm function.
where Lin (z) = 0
y
160
Table 1
Comparison of the energy given by (22) obtained from the semi-Bogomolny argument (EB ) with the ones
obtained by numerical simulations (Enum )
Q
47.691
25.703
18.064
15.168
14.065
14.193
16.268
19.256
23.678
34.910
61.603
149.263
722.656
Enum
EB
num
f (0)num
95.5007
51.6331
36.4665
30.7595
28.6076
28.8511
32.7688
38.2651
46.2186
65.9541
111.220
253.840
1140.119
87.9945
49.8406
36.0640
30.7182
28.6585
28.8982
32.7587
38.2417
46.2262
66.0215
111.482
254.515
1141.95
1.99750
1.98997
1.97737
1.95959
1.93649
1.90788
1.87350
1.83303
1.78606
1.73205
1.67033
1.60000
1.51987
0.1530
0.3020
0.4460
0.5798
0.7018
0.8093
0.9000
0.9723
1.0244
1.0555
1.0654
1.0564
1.0345
Next, the equation for the energy EB above has to be minimized with respect to CB
while QB is kept constant. We expect the semi-Bogomolny ansatz to be valid only when
CB 0 where the initial velocity of the trial profile function f (r) tends to zero. In
this region, the logarithms will dominate the dilogarithm and the trilogarithm functions
respectively, since these functions tend to zero like polynomials. Thus, by substituting
z = ln(CB ) for z > 0 one obtains

z 2
+ z2 ,
12 2
2

1
2 + 3z2
+
+z .
QB +
EB =
2
4
6
QB =
(14)
One can explicitly solve in terms of the charge QB given above in order to obtain that
12 2 QB
=
(15)
z( 2 + z2 )
which when substituted into the energy gives
2

6 2 Q2B
z( 2 + z2 )
2 + 3z2
+
+z .
EB =
+
z( 2 + z2 )
4
6
12 2
(16)
Now, upon minimizing the energy with respect to z we find that the frequency and the
charge QB are given by
12(1 + z)
,
2 + 3z2

2 z2 ( 2 + z2 )2
Q2B =
6 + 6z + 2 + 3z2 .
2
2
144( + 3z )
2 = 2 +
(17)
161
Finally, the relation between z and is

2

2
2 2
z= 2
(18)
2 + 2 2 .
1+ 1
12
( 2)
It is obvious that in the limit 2 the parameter z goes to infinity in consistency with
the analytical results (14). Thus, the semi-Bogomolny argument is valid in the thin-wall
approximation.
The elimination of the z variable from the equations that define QB and can be easily
performed. By letting = 2 2 one gets

2
2 2
z=
(19)
1+ 1+
,
12

2
2 + 24 + 18 + [24 + (6 + 2 )] 1 + 12 2
QB =
(20)
,
9 2 3
(2 + )
2 2 2(1 + )
4+
+
EB =
(21)
1+
+
QB .
2
2
12
2 2+
2
It is obvious that QB as 2 (which corresponds to 0) while the upper

energy limit for any value of Q (or ) can be obtained from (21).
Next, by eliminating between QB and EB we get the Q-ball energy-charge dependence
since
32/3 1/3 2/3

5 2/3
(4 + 3 2 )
1/3
EB = 2 QB +
Q
+
Q
B
27/6
211/632/3 B
36 2
4/3 (17 216 2 ) 1/3 5/3 (20 54 2 + 27 4 ) 2/3
QB +
QB
+
2592 21/631/3
1944 25/632/3
1
+ O QB .
(22)
Note that for large values of Q (i.e., for Q1/3 1) the expression (22) gives the upper
energy bound. In Table 1 we compare the results obtained from the semi-Bogomolny
argument by means of (22) with the ones obtained by solving the full second order
differential equations (6) numerically, for different values of Q. The agreement is
impressive and appears to extend far beyond the expected range of validity of the semiBogomolny argument.
Note that although the semi-Bogomolny argument and therefore the corresponding
energy-charge formula (22) holds only for large values of Q the agreement between the
numerical and analytical results holds for a wide range of Q. However, although the
profile (12) describes very well the energy of the Q-ball as a function of the charge Q, its
value at the origin f (0) is always smaller than one and differs considerably from f (0)num
of Table 1.
In Fig. 1 we plot the energy obtained from the semi-Bogomolny argument (22) and
from the numerical simulations (Enum ) for a wide range of Q. The peculiar behavior of
Fig. 1 for small Q can be explained from Fig. 2 where the energy and charge (obtained
162
Fig. 1. Energy-charge dependence for Q-balls. The upper and lower branches correspond to the thick-wall and
thin-wall approximations, respectively.
Fig. 2. The energy and charge (obtained numerically) as functions of the parameter =

4 2 .
numerically) as functions of are presented. It is obvious that in a range of two different

values of energy and charge exist. Finally, in Fig. 3 we plot the charge obtained from the
semi-Bogomolny argument
(20) and from the numerical
simulations (Qnum ) in terms of
the parameter a = 4 2 in the allowed range 2 < < 2.
163
Fig. 3. The charge Q obtained numerically and semi-analytically (20) as a function of the parameter
= 4 2 .
Remark 1. The semi-Bogomolny method can be applied for other potentials of polynomial
type. For example, in [12] the potential is of the form

U (f ) = f 2 1 + (1 f )2
(23)
and the corresponding semi-Bogomolny equation (analogous to (11)) is f = f (1 f )
while its solution
f (r) =
B exp( 2 r)
1+C
(24)
has the same asymptotic behavior for large r as in (12).
4. WoodsSaxon ansatz
A more general form of the ansatz (12) is widely used in nuclear physics describing
nuclear matter distribution inside heavy nuclei, or potential of nucleonnucleus interaction.
This is the so-called WoodsSaxon distribution and the corresponding profile function is
given by
f (r) =
CWS
1 + exp[a(r rQ )]
(25)
In this case, three arbitrary parameters (instead of one) appear in the parametrization,
and thus, more degrees of freedom exist. Here, rQ corresponds to the radius of the
164
Q-ball and 1/a defines the thickness of the shell of the Q-ball. At the origin, the values
of the profile function and its derivative are: f (0) = CWS (1 + earQ )1/2 and f (0) =
aCWS earQ (2 + 2earQ )3/2 . Recall that, due to boundary conditions, f (0) needs to be
very small (in fact, zero). This limit can beobtained when the product arQ is large that is in
the thin-wall approximation where 2 (see below). The field-theoretical motivation
for the WoodsSaxon ansatz was presented in the previous section since for CWS = 1 and
CB = exp(arQ ) the two profile functions given by (25) and (12) coincide.
For Q large, the energy of the Q-balls given by (8), (9) can be approximately evaluated
using the WoodsSaxon ansatz (25). To do so, integrals of the following type are used:

(CWS )2n
1
In =
(26)
r 2 dr = 3 (CWS )2n b2 In0 + 2bIn1 + In2 ,
n
(1 + exp[a(r rQ )])
a
where n = 1, 2, 3 and b = arQ while the moment of inertia becomes: I [f ] = 4I1 . In
general, by letting w = exp[a(r rQ )] while dr = dw/aw, the m-power of the integral In
is defined as

1 (ln w)m
m
dw,
In =
(27)
z (1 + w)n
w0
where w0 = exp(b). After some algebra it can be easily shown that the integrals Inm are
related via the following recursive relation
m
= Inm
In+1
m m1
(ln w0 )m
In
.
n
n(1 + w0 )n
(28)
For w0 1 one gets:

3
I30 b ,
2
2
2
2
1
b
b
b
I11 + L,
I21 + L,
I31 + L + ,
2
2
2
2
3
3
3
b
b
b
I12 ,
(29)
I22
2L,
I32
3L,
3
3
3
where the parameter L = 2(1 1/4 + 1/9 1/16 + ) 1.644 does not contribute in
the leading order expansion of the parameter Q1/3 (which we assume to be smallsee
below). Then the energy of the Q-ball (8), (9) can be approximated as
I10 b,
EQ
I20 b 1,
Q2
2
2V CWS

2
2
4
2 2CWS
+ Ederiv ,
+ V CWS
+ CWS
(30)
where = 1 3b1 , = 1 92 b1 , = 1 + 6Lb2 and V = 43 rQ3 . The derivative term

of the energy: Ederiv = 12 f 2 r 2 dr is going to be neglected (initially).
2 occurs at
The minimization of EQ with respect to V CWS
2
V CWS

min
=
Q
2 + C 4 )
2(2 2CWS
WS
(31)
which determines rQ . The corresponding minimum of the energy is

2
4
EQ 2 Q 2 2CWS
+ CWS
.
Further minimization with respect to CWS gives the following value for the energy

2
EQ 2 Q 2
165
(32)
(33)
at (CWS )2min = /. For large b (of order Q1/3 ), the parameter (CWS )min can be
approximated by (CWS )min 1 + 3/(2b) = 1 + 3/(4b) and in this limit the energy
becomes:

3
,
EQ 2 Q 1 +
(34)
4b
where
(rQ )min
4 2
1/3
Q1/3 = 0.55Q1/3.
Note that Eq. (34) implies that EQ 2 Q as b .

Next the derivative energy contribution is considered:

aCWS 2 2rQ 2L
+ 2 .
Ederiv
rQ +
4
a
a
(35)
(36)
Taking into account the highest order terms in rQ one obtains
b(rQ )min
3Q
.
2Q + +
(37)
4
2 2b
Further minimization with respect to b gives the energy-charge dependence of the Q-balls
up to order O(Q2/3 ):

9 1/3 2/3
EQ 2 Q +
(38)
Q = 2 Q + 1.35702 Q2/3
8 2
which occurs when
1/3

3 2 Q 1/2
12
Q
=
1.563 Q1/3.
bmin =
(39)
rQ
EQ
Note that the O(Q2/3 ) terms of (38) and (22) coincide, and therefore the two analytic
methods
are in good agreement. In addition, from (35) and since b = arQ one gets that
amin = 2 2 which is the value obtained from the semi-Bogomolny argument (12).
To conclude, let us state that in the thin-wall limit the profile function (25) is given
by f (0) 1 + 3/(4b) (or by f (0) 1 + 1/(2b) including higher order corrections).
Moreover, in this limit the WoodsSaxon distribution (25) is close to the exact values
of the Q-ball profile function (obtained numerically) presented in Table 1. In fact, for
Q = 149.26: f (0) = 1.06 and f (0)num = 1.0564 while for Q = 722.66: f (0) = 1.0356
and f (0)num = 1.0345.
166
Remark 2. For the alternative potential given by (23) the semi-Bogomolny function (24)
can be generalized in a similar way as (25) and the corresponding energy is similar to (38)
since

1/3 2/3
Q .
EQ 2 Q +
(40)
3 2
Note that the two expressions differ only by a 2/3 factor before the term O(Q2/3 ).
5. Discussion and conclusions

It has been shown that for specific parametrizations of the scalar field a semi-analytic
treatment for Q-balls exists leading to transparent and simple results. Two kinds of
approximations have been considered: one based on a semi-Bogomolny argument which
gives an exponential-step parametrization for the profile function and the WoodsSaxon
parametrization (the semi-Bogomolny generalization) motivated also by nuclear physics
experience.
The agreement of the results obtained using both approximations with the numerical
ones follows from the fact that the ansatz for the profile function obtained from semianalytic approaches have the correct asymptotic behavior as Q and r are large. This was
not the case in the semi-analytic treatment of the Skyrme model in three or two spatial
dimensions [14,15]. Although the energies obtained were accurate within 0.5% compared
to the exact ones for large values of baryon number, the asymptotic behavior of the profile
function was incorrect [14].
The thickness of the Q-ball surface (i.e., transition region) where the profile f decreases
from f 1 to f = 0 can be estimated by
2
t rQ .
(41)
b
For Q large
(and using the results of Section 4), the thickness is independent of the charge
since t 2. Thus, the large Q-balls can be visualized as spherically symmetric balls with
constant internal energy density
E,V 2.
(42)
These balls are surrounded by a surface of constant thickness and constant average energy
density per unit volume since
E,S
Ederiv
1
=
2
4trQ 4
(43)
in natural units of the model. Therefore it is energetically favorable for small Q-balls to fuse
into a bigger one since the surface of a single big Q-ball is smaller than the sum of surfaces
of several smaller Q-balls, for the same value of Q (or with the same total volume).
Our approach can be extended in lower (and also higher) spatial dimensions in a natural
way. In particular, in the one-dimensional case the energy-charge dependence is
1
E(Q)1D 2 Q +
(44)
2
167
which is in a close agreement (within 1%) with the numerical results obtained in [19]. In
addition, the value of the profile function at the origin is (approximately) given by
f (0)1D 1 +
1
4Q
(45)
where terms of the form exp(arQ ) have been neglected since Q is large (i.e., rQ > 1).
In the two-dimensional case the corresponding results are

1/2
Q
E(Q)2D 2 Q +
(46)
2 2
while

1/2 1
f (0)2D 1 +
(47)
.
4 Q
2
The formulas (44), (46) and (38) indicate that the relative contribution of the surface
energy (Ederiv ) increases as the dimensionality of space increases. This property of Q-balls
can be useful in cosmological applications. In some respect Q-balls are similar to the
multiskyrmions which correspond to bubbles of matter with universal properties of the
shell, where the mass and baryon number density is concentrated [14,15].
We conclude by saying that our arguments can be extended to other types of potentials.
Another important issue concerns the semi-analytic identification of the profile function
satisfying Eq. (6) for (9). We expect to report on this problem soon.
Acknowledgements
The work of V.B.K. is supported by the Russian Foundation for Basic Research, grant
01-02-16615. T.I. thanks the Royal Society and the National Hellenic Research Foundation
for a Study Visit grant.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
S. Coleman, Nucl. Phys. B 262 (1985) 263.

T.D. Lee, Particle Physics and Introduction to Field Theory, Harwood, London, 1981.
J.K. Drohm, L.P. Yok, Y.A. Simonov, J.A. Tyon, V.I. Veselov, Phys. Lett. B 101 (1981) 204.
T.I. Belova, A.E. Kudryavtsev, JETP 68 (1989) 7.
D.K. Hong, J. Low Temp. Phys. B 71 (1998) 483.
A. Kusenko, Phys. Lett. B 405 (1997) 108.
I. Affleck, M. Dine, Phys. Lett. B 249 (1985) 361.
A. Kusenko, M. Shaposhnikov, Phys. Lett. B 418 (1998) 46.
K. Enqvist, J. McDonald, Nucl. Phys. B 538 (1999) 321.
G. Rosen, J. Math. Phys. 9 (1968) 996.
M. Axenides, S. Komineas, L. Perivolaropoulos, M. Floratos, Phys. Rev. D 61 (2000) 085006.
T. Multamaki, I. Vilja, Nucl. Phys. B 574 (2002) 139, hep-ph/9908446.
F. Paccetti Correia, M.G. Schmidt, Eur. Phys. J. C 21 (2001) 181, hep-th/0103189.
V.B. Kopeliovich, JETP Lett. 73 (2001) 587;
V.B. Kopeliovich, J. Phys. G 28 (2002) 103.
168
[15]
[16]
[17]
[18]
[19]
T.A. Ioannidou, V.B. Kopeliovich, W.J. Zakrzewski, JETP 95 (2002) 572, hep-th/0203253.
A. Kusenko, Phys. Lett. B 404 (1997) 285.
E.B. Bogomolny, Sov. J. Nucl. Phys. 24 (1976) 449.
T.H.R. Skyrme, Nucl. Phys. 31 (1962) 556.
R.A. Battye, P.M. Sutcliffe, Nucl. Phys. B 590 (2000) 329.

A construction of G2 holonomy spaces with torus

symmetry
O.P. Santillan
Bogoliubov Laboratory of Theoretical Physics, JINR, 141 980 Dubna, Moscow Region, Russia
Received 20 November 2002; accepted 14 March 2003
Abstract
In the present work the CalderbankPedersen description of four-dimensional manifolds with selfdual Weyl tensor is used to obtain examples of quaternionic-Kahler metrics with two commuting
isometries. The eigenfunctions of the hyperbolic Laplacian are found by use of Backglund
transformations acting over solutions of the Ward monopole equation. The BryantSalamon
construction of G2 holonomy metrics arising as R 3 bundles over quaternionic-Kahler base spaces is
applied to this examples to find internal spaces of the M-theory that leads to an N = 1 supersymmetry
in four dimensions. Type IIA solutions will be obtained too by reduction along one of the isometries.
The torus symmetry of the base spaces is extended to the total ones.
PACS: 11.10.Lm
1. Introduction and the main result

The classification of the possible holonomy groups of Riemannian or pseudoRiemannian manifolds is an old mathematical problem. In [1] Berger presented a list
of the possible restricted holonomy groups of N -dimensional Riemannian manifolds,
but after the completion of this work it remained to prove the existence of metrics
with exceptional holonomies G2 and Spin(7) for the seven- and eight-dimensional cases,
respectively. This was achieved successfully by Bryant in [2]. In general, if a given
Riemannian metric with dimension N admits at least one covariantly constant spinor
satisfying Di = 0 the holonomy group will be SU(n/2), Sp(n/4), G2 or Spin(7), the
last two cases corresponding to seven and eight dimensions. For G2 holonomy manifolds
E-mail address: osvaldo@thsun1.jinr.ru (O.P. Santillan).
doi:10.1016/S0550-3213(03)00235-9
170
O.P. Santillan / Nuclear Physics B 660 (2003) 169193
there is exactly one. The fact that this spinor exist is apparent from the decomposition
8 = 1 7 of the spinor representation of the tangent space SO(7). Equivalently, in such
manifolds one can choose an orthogonal frame ei in which the three octonionic form
= e1 e2 e7 + e1 e3 e6 + e1 e4 e5 + e2 e3 e5
+ e4 e2 e6 + e3 e4 e7 + e5 e6 e7
and its dual are closed [3].
After the Bryant work explicit compact and non-compact G2 holonomy metrics were
constructed in [4,5], and complete ones in [6,7]. Recently, new examples have been found
in [814]. All of them have vanishing Ricci tensor, i.e., they are Ricci-flat, and this implies
that they are vacuum solutions of the Einstein equation. This suggest that the construction
of G2 holonomy manifolds is related with the construction of seven-dimensional self-dual
manifolds. The self-duality condition for the spin connection in seven-dimensions
cabcd
wcd
wab =
2
implies Ricci-flatness and G2 restricted holonomy [17] (analogous considerations hold in
eight dimensions for Spin(7)). Given an ansatz for a metric, this condition gives a system
of equations to be satisfied in order to have a G2 manifold. Although this equations are
nonlinear they have been solved in cases with suitable symmetries in which the system
takes a more simple form [3,18].
The main feature that relates this subject to physics is the presence of one nonzero parallel spinor field , which plays a central role in supersymmetry, string theory and M-theory.
Compactification of the M-theory (or his low energy limit, the eleven-dimensional supergravity) on G2 holonomy manifolds leads to an effective four-dimensional theory with one
supersymmetry, corresponding to such spinor field [15].
There are in the literature examples of weak G2 holonomy [19], which are again
backgrounds of the M-theory that give rise to N = 1 supersymmetry in D = 4. In this case,
there is an spinor field which is not covariantly constant but satisfies Di i . The
Ricci flatness condition is replaced by Rij gij . In the limit 0 one obtain G2 as
restricted holonomy group. Hitchin has shown that under certain conditions is possible to
construct this kind of manifolds starting with an Spin(7) holonomy one [16].
Compactification of the M-theory based on G2 smooth manifolds cannot give rise to
chiral matter. In the smooth case the harmonic KaluzaKlein decomposition of the elevendimensional supergravity is the N = 1 four-dimensional supergravity coupled Abelian
vector multiplets plus chiral multiplets. But the chiral matter fields can emerge only if the
manifold develops a singularity, as pointed out by Witten and Acharya in [20,22]. It turns
out that to obtain a realistic model one should investigate the dynamics of the M-theory
over orbifolds. A modern description of this dynamics over manifolds that are developing
a conical singularities can be found in [21]. Generalizations of the work of Acharya and
Witten on singular G2 spaces were investigated recently in [23], and over spaces with torus
symmetry and only orbifold singularities in [24,25].
Certain G2 metrics with two Abelian isometries have called the attention recently [26],
because an U (1) isometry allows an type IIA superstring interpretation upon dimensional
reduction to ten dimensions. In [24] Anguelova and Lazaroiu have extracted the IIA
171
reduction of M-theory on certain toric backgrounds, and its type IIB duals. Such IIA
solutions corresponds to systems of weakly and strongly coupled D6-branes, while the
duals describes systems of localized and delocalized 5-branes. The G2 spaces presented
in those works have been found applying the BryantSalamon construction with a fourdimensional quaternionic base manifold with torus symmetry [5]; the U (1) U (1)
isometry is extended to the total space.
The four-dimensional toric quaternionic Kahler spaces has been completely described
by Calderbank and Pedersen [48] in terms of eigenfunctions of the hyperbolic Laplacian
with eigenvalue 3/4, namely the solutions of the equation
F + F =
3F
.
4 2
The toric G2 cones utilized in [24] were constructed with the so-called m-pole eigenfunctions which gives rise to spaces that are complete (thus compact) and admits only orbifold
singularities. As such, they seem to be the best candidates with U (1) U (1) isometry for
which the physical analysis of Acharya and Witten can be applied.
One of the purposes of this paper is to describe how to construct nontrivial eigenfunctions of the hyperbolic Laplacian (and, in consequence, examples of toric quaternionic
spaces) starting with solutions of the monopole equation
V + 1 (V ) = 0
described by Ward and Woodhouse in [46]. In fact, in [48] it has been pointed out that the
space of solutions of the Ward monopole equation and the Calderbank Pedersen one are
related by a Backglund transformation, allowing to construct solutions one from another.
The Ward monopole equation has been investigated in the context of (2 + 1) gravity and
it is known that its solutions admits an integral representation in terms of an arbitrary
function of one variable. For this reason it is possible to find nontrivial eigenfunctions of the
hyperbolic Laplacian selecting a function in the integral representations of the monopoles
and performing a Backglund mapping.
The Ward monopoles are also relevant for hyper-Kahler geometry [47]. This is because
it is possible to construct hyper-Kahler spaces with torus symmetry starting with a given
monopole, in analogous manner as in the CalderbankPedersen picture. Moreover, in the
hyper-Kahler limit the BryantSalamon construction gives G2 holonomy spaces which are
globally the Cartesian product of the hyper-Kahler one with R 3 . For this reason one can
construct such spaces starting with a monopole, without using a Backglund mapping. The
reason for which in the hyper-Kahler limit it is obtained a trivial product with R 3 could be
easy to visualize: being R 3 flat the seven-dimensional self duality condition becomes the
four-dimensional one for the base space.
The organization of this paper is as follows: in Section 2 it is presented the group G2
as the group of automorphisms of the octonions. In Section 3 there are reviewed some
basic features about self-dual seven-dimensional manifolds. In Section 4 it is shown that
such manifolds have restricted holonomy group G2 . Section 5 contains a review of the
BryantSalamon construction of G2 metrics with quaternionic-Kahler base spaces. In
Section 6 the Ward and CalderbankPedersen descriptions are reviewed and examples
of quaternionic-Kahler and hyper-Kahler metrics are constructed. The m-pole solutions
172
are also discussed in some detail. The vacuum configurations of the eleven-dimensional
supergravity related to this metrics are founded in Section 7. Dimensional reduction of
such configurations alone one of the isometries is performed to give rise to ten-dimensional
type IIA backgrounds.
To conclude, it should be mentioned that the study of explicit metrics with exceptional
holonomy has also importance in the context of dualities of string theory and M-theory
[2731]. The range of applications of this topic is very wide; some of them can be found
in [3240].
2. The exceptional group G2 and the octonions

The group G2 SO(7) is one of the exceptional simple Lie groups. It is compact,
connected, simply-connected and of dimension 14. It has been proved [41], that G2 is the
group of automorphisms of the octonions O, up to an isomorphism. The octonions (or
Cayley numbers) constitutes the only nonassociative division algebra (the associative ones
are only R, C and H ) and an arbitrary element x O can be written as a linear combination
of the form x = x 0 + x i ei , where the set ei constitute a basis of 7 unit octonions with the
following multiplication rule:
ei ej = cij k ek ,
ei .1 = 1.ei = ei .
x i s
(2.1)
= xi e
The
takes real values. The subspace P of O generated by the elements x
i is
called the space of pure octonions, and the total space can be decomposed as O = R P .
The constants cij k that define the multiplication (2.1) are totally antisymmetric and
c123 = c246 = c435 = c367 = c651 = c572 = c714 = 1,
(2.2)
up to an index permutation. The constants corresponding to another set of indices are

identically zero. From (2.1) and (2.2) it is seen that (e3 e7 )e5 e3 (e7 e5 ) = e1 , which
shows the nonassociativity of the octonion algebra. For this reason the octonions cannot
be represented as a matrices and do not satisfy the Jacobi identities. In other words, the
algebra of O is not a Lie algebra.
It is possible to represent the components of an arbitrary octonion as a 7-dimensional
= (x1 , . . . , x7 ). We define the octonion numbers g x as those with components
vector X
g.X, where g is an arbitrary 7 7 matrix. The statement that the exceptional group G2
is the group of automorphism of the octonions means if x P , gx P if g is any of the
elements of G2 in the fundamental irreducible representation; and that if x.y = z for given
x, y and z belonging to P , gx.gy = gz.
Over O it is defined an internal product ( , ): O O R given by
(ei , ej ) = ij ,
(2.3)
from where it is obtained that

(x, y) = x i y i .
(2.4)
Taking into account all the facts mentioned above it is possible to construct a threeform over a seven-dimensional space V , which is G2 invariant. This form is fundamental
173
in this work because its closure has implications about the holonomy group of V . The
construction follows decomposing the octonion space as O = R P , and defining over P
the bilinear B(x, y) and the internal product , : P P R by the identities
0

x e0 + x, y 0 e0 + y = x 0 y 0 + x, y,
(2.5)
0
0
0 0

0

0
x e0 + x y e0 + y = x y x, y e0 + x y + y x + B(x, y) .
(2.6)
Under an automorphism transformation B(x, y) and , satisfy
B(gx, gy) = gB(x, y),
(2.7)
B(x, y) = B(y, x),
(2.8)
gx, gy = x, y.
(2.9)
From (2.7) and (2.9), it is seen that

(x, y, z) = B(x, y), z = B(gx, gy), gz = (gx, gy, gz).
(2.10)
In other words, the trilinear (x, y, z) = B(x, y), z is G2 invariant. From the rule
(2.1) and the definition (2.6) it follows that components of are

(ea , eb , ec ) = B(ea , eb ), ec = cabc ,
and that
(x, y, z) = cabc x a y b zc .
From the invariance of under the action of G2 it follows that for a given real vector
space V of dimension 7, with e1 , . . . , e7 a basis for V , the three form
(x, y, z) =
1
cabc ea eb ec
3!
(2.11)
is G2 invariant.
3. Self-dual manifolds in 4 and 7 dimensions

The self-duality condition is a familiar concept in quantum field theory [42] and
in general relativity. By definition the curvature tensor Rab = dwab + wac wcb of a
Riemannian metric is self-dual if
1
Rab = (abmn Rmn ,
2
or, in components
(3.1)
1
Rabcd = (abmn Rmncd .
2
Nontrivial solutions of this type has been found in the past [43]. They are called
gravitational instantons if their are Euclidean and with finite energy. It has been shown
174
that in four dimensions a torsion-free metric that satisfies (3.1) will be a vacuum solution
of the Einstein equations without cosmological constant [44].
In seven dimensions (3.1) is generalized as [17]
1
Rab = cabcd Rcd ,
(3.2)
2
where the totally antisymmetric cabcd are defined in terms of the octonion structure
constants (2.2) through the relations:
1 abcdefg
(
cefg .
(3.3)
3!
This generalization has been related to the octonions because the solutions of (3.2) will
be not only vacuum solutions of the Einstein equation without cosmological constant, but
Ricci-flat, which is one of the main features of the G2 manifolds. This follows directly
from (3.2), the antisymmetry of (3.3) and the Bianchi identity Rd[ebc] = 0 as
cabcd =
1
1
Rab = Racbc = cacde Rdebc = cacde Rd[ebc] = 0.
(3.4)
2
2
Indeed, it will be shown in the next section that seven-dimensional self-dual manifolds
have restricted holonomy G2 . For such case the Weyl tensor Cabcd defined by
Cabcd = Rabcd +
R
(gac gbd gad gbc )
(n 1)(n 2)
1
(gac Rbd gad Rbc gbc Rad + gbd Rac )
(n 2)
(3.5)
will be equal to the Riemann tensor. The first one is traceless, in consequence Ricci-flat
manifolds have traceless curvature tensor.
Many authors presents the seven-dimensional self-duality as a property of the spin
connection, namely
1
ab = cabcd cd ,
2
or, equivalently,
(3.6)
cabc bc = 0.
(3.7)
In [45] a clear proof that the definition (3.6) implies (3.2) was given. It is based in the
following identity for the octonion constants
f]
cabcp cdefp = 3cab[de e]b 2cdef [a b]c + 6a[d be c 2cdef [a b]c .
(3.8)
If (3.6) holds it is clear that dba will be self-dual. The self-duality of ca bc follows
using that
1
1
f
cabcd ec de = cabcd cedfg ec g
2
4
1
1
f
f
= cabf e c ec + caefg eb g
2
4
1
f
cbefg ea g + ca bc
4
1
= cabcd ec de + 2ca bc .
2
The identity (3.8) has been used here. From (3.9) is seen that
175
(3.9)
1
cabcd ec de = ca bc
2
which implies the self-duality of R. The definition (3.6) implies (3.2), but the converse is
not necessarily true.
4. The holonomy group of the metric obtained

The purpose of this section is to show that seven-dimensional self-dual manifolds has
restricted holonomy G2 . Holonomy is the process of assigning to each closed curve of a
manifold the linear transformation that measures the rotation resulting when a spinor or
vector field is parallel transported around the given curve. The set of holonomy matrices
constitutes a group, the holonomy group of the manifold. If it is considered only those
curves which are contractible to a point it is the restricted holonomy. In simply connected
manifolds both groups coincides.
From the Berger list [1] it follows that if a seven-dimensional manifold admits only one
covariantly constant spinor it will have G2 restricted holonomy. It will be shown now that
the seven-dimensional self-dual manifolds admits exactly one. The covariant derivative of
a spinor is

1
ab
Di = i iab
(4.1)
.
4
Choosing its components as
= 8
(4.2)
it is obtained
1
Di = iab ab .
4
One of the possible representations of the SO(7) gamma matrices is the antisymmetric and
imaginary given by
a

= i(ca
+ a 8 + a 8 ),

are zero if or are equal to 8, and the octonion constants in
where the constants ca
other case. In this representation ( ab ) = cabc . Using (3.7) it follows that
1
1
Di = iab ab = iab cabc = 0.
4
4
With this election for the gamma matrices, this is the only spinor that satisfies the last
equation. But the result must be independent of the representation, which implies that the
176
seven-dimensional self-dual manifolds admits exactly one covariantly constant spinor and
their holonomy is G2 .
The change of an arbitrary field under infinitesimal parallel transport is
= Gab Aab ,
(4.3)
where Aab is an infinitesimal area element spanned by the closed curve taking into
consideration. Gab = Rabcd cd generate the infinitesimal holonomy group, being cd the
generators of SO(m) in the representation of the field . The restricted holonomy can be
larger than the infinitesimal one.
It has been mentioned in the introduction that for G2 manifolds it is possible to construct
a G2 invariant closed and co-closed three form. For this reason it is needed to check that this
holds for seven-dimensional manifolds with self-dual spin connection. The most natural
candidates to consider are the G2 equivariant 3-form (2.11) and its dual [45]
1
cabc ea eb ec ,
3!
Taking into account the identity
=
1
cabcd ea eb ec ed .
4!
(4.4)
cabp cpcde = 3ca[cd e]b 2cb[cd e]a

it is obtained
d =
1
1
cabc ea eb cd ed = cabc ccdef ea eb ef ed
3!2
3!
1
cade ea ed eb eb = 2d.
3!
From here follows
=
d = 0.
(4.5)
So, is a closed form. Similarly, using (3.8) follows that

1
cabcd ae ee eb ec ed
4!6
1
caefg cabcd fg ee eb ec ed
=
4!12
1
cf bcd f e ee eb ec ed = 2d ,
=
4!3
which implies the closure of .
d =
5. The BryantSalamon construction

A construction of G2 manifolds starting with four-dimensional quaternionic Kahler
manifold as a base space has been given by Bryant and Salamon in [5] and reconsidered
recently in [26]. Those spaces are construct as R 3 bundles over a quaternionic base. In
177
the hyper-Kahler limit the bundle will be trivial, i.e., it will be the global product of the
hyper-Kahler one with R 3 .
Before describe this construction it is convenient to review certain properties of
quaternionic spaces. By definition [53] a quaternionic-Kahler space is a Riemannian space
of real dimension 4N endowed with a metric
ds 2 = g (x) dx dx
i
and a set of three complex structures J satisfying the quaternionic algebra

j
Ji J = ij + (ij k Jk .
(5.1)
The metric is quaternionic Hermitian:

g J i X, J i Y = g(X, Y ),
i = J i . The holonomy group H Sp(n) Sp(1), and if
from where follows that J
the manifold has scalar curvature equal to zero it will be called hyper-Kahler. From the
complex structures J i is possible to construct the hyper-Kahler triplet of 2-forms given by

i
i
dx dx ,
= g J i
i =
and the three local 1-forms

i
,
Ai = mn Jmn
(5.2)
where mn
represents the anti-self-dual part of the spin connection. The hyper-Kahler form
is covariantly closed with respect to the connection Ai ; namely
i = d i + (ij k Aj k = 0.
Defining the SU(2) curvature
F i = dAi + (ij k Aj Ak ,
it follows that
F i = i .
(5.3)
Here denotes the scalar curvature. Any quaternionic metric is an Einstein space with
curvature and
Rmn = 3gmn .
It is important for the purposes of the present work to remark that in four dimensions a
quaternionic-Kahler metric is an Einstein metric with self-dual Weyl tensor
cabcd
Wcd .
Wab =
2
In the hyper-Kahler case = 0, so F i will be flat and the manifold will be Ricci-flat.
The BryantSalamon result is that the following
i
2
1
ds 2 =
(5.4)
du + ( ij k Aj uk + 2|u|2 + c ds42
2|u|2 + c
178
is a G2 holonomy metric. An straightforward proof can be found in Ref. [26], there is

shown that (5.4) has a self dual spin connection and in consequence the restricted holonomy
will be G2 . The metric ds42 corresponds to the quaternionic base space; the total one is
topologically an R 3 bundle with coordinates ui , and c is an integration constant.
In the hyper-Kahler limit = 0 it is possible to gauge away the connection Ai ; the
resulting manifold is the trivial product of R 3 by the hyper-Kahler manifold, which is noncompact and with holonomy G2 . Noncompact backgrounds are of interest in M-theory as
well [26]. But it should be noted that a manifold that is globally the Cartesian product of R 3
with any four-dimensional manifold with self-dual spin connection will be a G2 holonomy
space. The reason is the following: the spin connection obviously will be independent
on the coordinates of R 3 and the condition (3.6) will reduce to the ordinary self-duality
condition in four dimensions; it follows that if the four-dimensional base space is self-dual
the restricted holonomy will be G2 .
Expressing the R 3 part of (5.4) in polar coordinates it is obtained the following
expression

a
b
r2 2
r2
dr 2
4c
2
a i
b j
ds =
dx
dx
+ ds4 ,
+
+
A
+
A
g
1

ab
i
j
4
r4
2
1 4c4
r
where gab and are the metric and the killing vectors of S 2 , respectively. In this coordinate
system is more clear that the metric is asymptotically a cone; in the limit r it is seen
that
ds 2 dr 2 + r 2 d,
where the part related with is independent of r. In other words, for large r (5.4) is a cone
over a six-dimensional space constructed as an S 2 bundle over a quaternionic-Kahler space
with the metric ds42 . This six-dimensional manifold has weak holonomy SU(3) [3]; such
manifolds are of great interest in compactification of the type IIA supergravity.
6. Toric quaternionic geometry in four dimensions

6.1. The Ward and the CalderbankPedersen descriptions
In the last subsection it has been described how to construct G2 holonomy spaces
starting with hyper-Kahler and quaternionic-Kahler ones. The aim of this work is to
investigate the case in which there are two commuting isometries. The hyper-Kahler case
was discussed in [47] and is related with the monopole solution appearing in the context of
the (2 + 1) Einstein gravity [46].
The main result needed here is that the four-dimensional Euclidean metric

2
ds 2 = V d 2 + d2 + 2 d 2 + V1 d + V d
(6.1)
will be hyper-Kahler if and only if the function V is an axisymmetric harmonic function
(AHF), i.e., if it satisfies the monopole equation
V + 1 (V ) = 0.
(6.2)
179
The metric (6.1) has an U (1) U (1) isometry because the two killing spinors / and
/ commutes.
Many properties of the AHF are well know. It has been shown [46] that any solution V
of the equation
V 1 (V ) = 0
can be expressed in integral form as
1
V (, ) =
2

G sen( ) + d.
(6.3)
Here G(x) denotes an arbitrary function of one variable. For a given solution V (, )
the function V (i, ) will be a solution of (6.2), and W (, ) = V (i, ) + V (i, ) will
be a real AHF. This integral representation will be used in the following sections.
In a recent work Calderbank and Pedersen have given a complete description of the
four-dimensional non-Ricci flat Einstein metrics with self-dual Weyl tensor and two
commuting isometries, in terms of certain eigenfunctions of the hyperbolic Laplacian
in two-dimensions [48]. This case is of interest as well, because such metrics will be
quaternionic-Kahler.
Their statement is that for any Einstein-metric with self-dual Weyl tensor and nonzero
scalar curvature possessing two linearly independent commuting Killing fields there exists
a coordinate system in which the metric g has locally the form
g=
F 2 4 2 (F2 + F2 ) d 2 + d2
4F 2
2
[(F 2F ) 2F ]2 + [2F + (F + 2F )]2
,
+
(6.4)
F 2 [F 2 4 2 (F2 + F2 )]
where = d and = (d + d)/ and F (, ) is a solution of the equation

F + F =
3F
4 2
(6.5)
on some open subset of the half-space > 0. On the open set defined by F 2 > 4 2 (F2 +
F2 ), g has positive scalar curvature, whereas F 2 < 4 2 (F2 + F2 ), g is self-dual with
negative scalar curvature.
It follows that quaternionic metrics with torus symmetry has positive signature. The
three 1-forms (5.2) have a remarkable simple expression in terms of F ,

1
d
1
d
+
F + F
A3 =
A1 =
(6.6)
F
,
A2 = ,
F
F
F
and the relation (5.3) holds with = 1.
Both statements presented here can be interpreted as methods to construct quaternionicKahler and hyper-Kahler geometries starting with solutions of two linear equations of
second order, namely (6.2) and (6.5). As we will see, this equations are related by a
180
Backglund transformation, which implies any solution of one of them allow us to construct
a solution of the other one. This fact can be exploited to find examples of quaternionic
metrics.
6.2. The Backglund transformation
To prove that (6.2) and (6.5) are Backglund transformed it is needed to introduce the
Joyce system of equations [48]
(S0 ) + (S1 ) = S0 /,
(S0 ) (S1 ) = 0
(6.7)
where S0 and S1 are unknown functions. Selecting S0 = H and S1 = H the second

equation will be trivial and the first became H + H = H / (H is usually called
a Tod coordinate). Now, taking H = 1/2 F it follows that F satisfies (6.5). Conversely,
selecting S0 = V and S1 = V the first equation is trivial and the second one is
V + 1 (V ) = 0, in other words V is an AHF.
It is seen that a solution V of (6.2) allows construct a solution F of (6.5) integrating the
system
H = V ,
H = V
(6.8)
and defining F = H / 1/2 . Conversely, a solution F of (6.5) allows to construct a solution

V of (6.2) defining H = 1/2 F and integrating (6.8). By construction the equation for
F is the integrability condition for V and vice versa; the relation between them is an
example of a Backglund mapping. This gives a method to construct quaternionic-Kahler
metrics starting with an hyper-Kahler example and vice versa, in both cases there is a torus
symmetry. The representation (6.3) together with (6.8) can be exploited to find nontrivial
examples of quaternionic-Kahler spaces, with only selecting an arbitrary function of one
variable. This is the purpose of the next subsection.
6.3. Examples of quaternionic spaces with torus symmetry
The following are quaternionic metrics constructed starting with an arbitrary function G(x).
Example 1. The trivial four-dimensional toric metric
ds 2 = d 2 + d2 + 2 d 2 + d 2
corresponds to the monopole
V = .
(This AHF holds using G(x) = x Log(x) in the integral expression (6.3).) Eqs. (6.8) are in
this case
H = ,
H = 0,
181
and the eigenfunction F is given by

3/2
.
2
The insertion of the last eigenfunction in (6.4) gives the following metric:
F=
2d 2 + 2d2 2 (1 + 3) + 2 (9 + 7) 2
d
2
2 5
8
16
4 d 2 4 d d.
g=
(6.9)
The inequality F 2 < 4 2 (F2 + F2 ) holds for > 0. Invoking the AlderbankPedersen
theorem, we see that for > 0 the quaternionic-Kahler metric is g. In this case = 1
and the three 1-forms Ai are
A1 =
2d
,
A2 =
2d
,
A3 =
2
2
d + 2 d.
2
The metric constructed in this example is defined for all the positive values of .
Example 2. With the function G(x) = Log(x)x 3 it is found the monopole
V = 3 2 23
and the hyper-Kahler metric

ds 2 = 3 2 62 d 2 + d2 + 2 d 2 +
3 2

2
1
d + 6 2 d .
2
6
The Backglund transformed F results

3
F = 3/2 42 2
4
and the corresponding metric is

g = g d 2 + d2 + g d 2 + g d 2 + 2g d d,
(6.10)
where the components of the metric tensor are

g =
4(84 + 62 2 + 3 4 )
,
( 3 42 )2
g =
84 2 (19 + 5) + 166 (9 + 7) + 6 (1 + 35) + 32 4 (35 + 61)

,
9 5 ( 2 42 )2 (84 + 62 2 + 3 4 )
g =
g =
9 4 ( 2
64(44 + 4 )
,
42 )2 (84 + 62 2 + 3 4 )
32(85 2 3 + 3 4 )
.
9 4 ( 2 42 )2 (84 + 62 2 + 3 4 )
182
As in Example 1, F 2 < 4 2 (F2 + F2 ) for > 0, g is a quaternionic-Kahler metric

and = 1. The three forms Ai are given by
8
4( 2 22 )
d,
d
+
2 42
( 2 42 )
4
A2 =
d,
2
3( 42 )
4
4
d +
d.
A3 =
2
2
2
3( 4 )
3( 42 )
A1 =
The metric (6.10) is singular at 0 and at the lines 2|| = ||.

Example 3. The powers G(x) = x n and G(x) = Log(x)x 2n+1 can be integrated out giving
polynomial solutions of higher degree. For instance G(x) = Log(x)x 5 gives
V = 3 2 23 ,
F=

1 4
8 + 402 2 + 15 4 .
The even powers G(x) = x 2n Log(x) can be integrated too, but the expressions of the
metrics are more complicated by the appearance of logarithm terms. For example, with
G(x) = Log(x)x 2 it is obtained

2
2
2
2
2
,

V = 6 6 1 + 2 + 2 2 22 Log
+ 2 + 2

8 2 2

4 3
4 2 2
2
2
2
2
2

F = 4 + + + + 4 Log
.
3
3
3
+ 2 + 2 )
The expression for the quaternionic metric and the hyper-Kahler one corresponding to
such solutions is very large and difficult to simplify.
Example 4. The function G(x) = ex gives
V = ei I0 () + c.c.,
where In () denotes the modified Bessell function of the first kind, which are solutions of
the equation

2 H () + H () 2 + n2 H () = 0.
The hyper-Kahler space that corresponds to this monopole is:

ds 2 = I1 () d 2 + d2 + 2 d 2 cos()

2

1

d + I1 () + I0 () + I2 () sin() d . (6.11)
+
I1 () cos()
2
The Backglund transformed eigenfunction F is given by
F = I1 ()ei + c.c.
183
and from this solutions it follows the quaternionic metric

dsqk =

(, ) 2
d + d2
4I1 ()2
[2 3/2I1 () cos() 3/2 (I0 () + I2 ()) sin()]2
+
(, )
[2 3/2I1 () cos() + (I0 () + 2I1 () + I2 ()) sin()]2

+
,
(, )
(6.12)
where it has been defined

(, ) = I0 ()2 + 2I1 ()I2 () + I2 ()2 + 4I1 ()2 ctg()2

+ 2I0 () I1 () + I2 ()
and

(, ) = 2 I1 ()2 sin()2 4 2 I1 ()2 cos()2 I1 ()2 sin()2

2
+ I0 () + I1 () + I2 () sin()2 .
It will be quaternionic for the regions of the plane (, ) in which F 2 < 4 2 (F2 + F2 ). In
those regions k = 1. The three one forms Ai are

d
1
tg() I0 () + I2 ()
tg() d,
A1 = 1 + tg() +
2
I1 ()
A2 =
,
A3 =
.
I1 () cos()
I1 () cos()
The radial component of the metric (6.12) shows that some of the singularities are the zeros
of I1 ().
6.4. The m-pole solutions
In the previous subsection it have been constructed quaternionic-Kahler metrics starting
with an arbitrary AHF and solving the Backglund equations. By completeness it will be
discussed here the spaces corresponding to the m-pole solutions investigated in [50] and
[48]. It has been analyzed by Anguelova and Lazaroiu the dynamics of the M-theory on
toric G2 cones constructed with m-pole spaces as base manifolds in [24] and [25]. Such
examples leads to toric Einstein self-dual spaces of positive scalar curvature which are
complete (thus compact) and admitting only orbifold singularities. As such, they seem to
be the best candidates with U (1) U (1) isometry for which the physical analysis given
in [22] can be applied. In the first reference of [48] it has been described the moduli space
corresponding to the 3-pole solutions and has been shown that they encode some wellknown examples appearing in the physics, like the Bianchi type spaces. It will be shown
that the hyper-Kahler metrics corresponding to the 3-pole solution are those discussed in
[47] which gives rise to the 3-dimensional EguchiHanson like EinsteinWeyl metrics after
184
the quotient by one of the isometries. The following exposition follows closely those given
in the references [48].
The basic eigenfunctions F of (6.5) which we need to consider are

()2 + ( y)2
,
F (, , y) =
(6.13)
where the parameter y takes arbitrary real values. Using the Backglund transformation it
is found the basic monopole

V (, , y) = Log y + 2 + ( y)2 .
(6.14)
Being the equations for F and V linear, for any set of real numbers wi the functions
F=
k+1

wi F (, , yj ),
(6.15)
wi V (, , yj )
(6.16)
j =0
V=
k+1

j =0
will be solutions too. For this reason the 2-pole functions given by

1 + 2 + 2
()2 + ( + 1)2
()2 + ( 1)2
F1 =
,
F2 =
are eigenfunctions of the hyperbolic Laplacian. The first one gives rise to the spherical
metric, while the second one gives rise to the hyperbolic metric

2 2

ds 2 = 1 r12 r22
dr1 + dr22 + r12 d12 + r22 d22 .
The relation between the coordinates (r1 , r2 ) and (, ) can be extracted from the relation
(r1 + ir2 )2 =
1 + i
.
+ 1 + i
The hyper-Kahler metrics corresponding to both cases are

2
2

1
ds 2 =
d + d2 + 2 d 2 2 + 2 d +
d ,
2 + 2
2 + 2
(6.17)
and

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

d + d2 + 2 d 2
ds =
2 + ( + 1)2 2 + ( 1)2

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

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

+1
1
d .
d +

(6.18)
2 + ( + 1)2
2 + ( 1)2
185
The general 3-pole solutions are

b + c/m 2 + ( + m)2 b c/m 2 + ( m)2
1
+
.
F= +
By definition m2 = 1, which means that m can be imaginary or real. The corresponding

solutions are denominated type I and type II, respectively. It is convenient to introduce the
EguchiHanson like coordinate system defined by

= R 2 1 cos( ),
= R sin( ),
where takes values in the interval (/2, /2). In this coordinates
F = 1 + bR + c sin( ),

1
b(R b) + c(sin( ) + c)
1 F 2 2 F2 + F2 =
4
R 2 sin2 ( )
(6.19)
(6.20)
and the family of self-dual metrics corresponding to the 3-pole are expressed as
ds 2 =

dS 2
b 2 c2 + (bR cS) dR 2
+
(1 + bR + cS)2
R2 1 1 S 2
1
2
2
2
(1 + bR + cS) (b c + (bR cS))(R 2 S 2 )

2
2
R 1 1 S 2 (bR cS) d + (cR bS) d

+ b R 2 1 S + c 1 S 2 R d

2
+ c R 2 1 S + b 1 S 2 R + R 2 S 2 d .
(6.21)
It has been denoted S = sin( ) here. The expression (6.21) includes some well-known
metrics. Let us focus in the type I case. The formulas (6.19) and (6.20) allows to determine
the domain of definition of the metric (6.4). When b is nonzero for a given value of , F = 0
if R = (1 + c sin( ))/b and ( 14 F 2 2 (F2 + F2 )) = 0 if R = (b 2 + c2 + c sin( ))/b.
The case c = 0 correspond to a bi-axial Bianchi IX metric [51]. The domains of definition
are (, R ), (R , R ), and (R , ). In the first two cases the curvature is negative,
and in the last one positive, and in the two last cases there is an unremovable singularity at
R = R . In the case b = 0 for c > 1 and c < 1 the metric will be of Bianchi VIII type [49].
The case c = 1 corresponds to the Bergmann metric on CH 2 .
For the type II case, the range of R is (1, ) but the moduli space is more complex that
in the type I case. For the lines b = c it is obtained the hyperbolic metric if b < 0 and the
spherical metric if b > 0. If (b, c) = (1, 0) it is obtained the FubbiniStudy metric on CP 2
whereas the points (0, 1), (1, 0) and (0, 1) yield again the Bergmann metric on CH 2 .
Along the lines joining (1, 0) with others we have bi-axial Bianchi metric IX, while along
the lines between (0, 1), (1, 0) and (0, 1) the metric is Bianchi VIII. A more complete
description is given in [48].
186
The triplet of one forms corresponding to this family of metrics is

A1 = A1+ + A1 ,

(R 2 1)(1 S 2 )
2
d,
A =
(1 + bR + cS)
A3 =
d + d
,
(1 + bR + cS)

A1 = A11 + A12 + A13 ,
with
A11 =
(b c/m)(SR m)

(1 + bR + cS) (1 S 2 )(R 2 1) + (SR m)2
2

S R 1
R 1 S2
dS +
dR
2 1 S2
R2 1
and
A12

(b c/m) (1 S 2 )(R 2 1) + (SR m)2
dS
=
+ S dR
R
2(1 + bR + cS) R 2 1
1 S2
and
A13
(b c/m)(R 2 1)1/2(1 S 2 )1/2

2 (1 S 2 )(R 2 1) + (SR m)2

(b c/m) (1 S 2 )(R 2 1) + (SR m)2
4(1 S 2 )1/2 (R 2 1)1/2

dS
(1 S 2 )(R 2 1)
R
+ S dR .
(1 + bR + cS) R 2 1
1 S2
(The sign in (R 2 1) depends only on the metric in consideration, it is + for type I

and for type II.)
The Backglund transformed function V reads

(b + c/m)
m + ( m)2 + 2
V = Log() +
Log
2

+ m + ( + m)2 + 2
(b c/m)
Log
+
.
2
For the type I case this is the potential for an axially symmetric circle of charge, while the
type II case corresponds to two point sources on the axis of symmetry. The hyper-Kahler
metrics obtained are encoded in the following expression

bR + c 1 S 2 2
ds = 2
d + d2 + 2 d 2
R (1 S 2 )
R 2 (1 S 2 )
+
bR + c 1 S 2
R 2 (1 S 2 ) b(R 2 1) 1 S 2 + cRS 2
d .
d +
R 2 (1 S 2 )
187
(6.22)
This manifolds have been investigated recently in [47] and it has been shown that the
quotient of (6.22) with

gives the EguchiHanson type EinsteinWeyl metrics in D = 3.
The continuum limit of the expressions (6.15) and (6.16) are

F (, ) = w(y)F (, , y) dy,
(6.23)

V (, ) = w(y)V (, , y) dy,
(6.24)
where w(y) is a distribution with compact support in R. A choice of w(y) for which at
least one of the integrals (6.23) and (6.24) converges gives rise to an smooth solution. For
instance for w(y) = y/(y 2 + 1)2 it is obtained the following nontrivial monopole

|1i(1i)2 + 2 |
cos 12 Arg(1 2i 2 2 ) Log

|1+i (1i)2 + 2 |
V (, ) =

|(1 i)2 + 2 |

1i(1i)2 + 2
sin 12 Arg(1 2i 2 2 ) Arg

1+i (1i)2 + 2

,
+
|(1 i)2 + 2 |
and from (6.1) follows an hyper-Kahler metric. But (6.23) is divergent for this distribution.
To conclude this subsection it should be mentioned that higher m-pole solutions have
been considered in [24] and [25], and that quaternionic spaces with torus symmetry have
been investigated recently in [52] using the harmonic space formalism.
7. G2 holonomy metrics with torus symmetry and supergravity backgrounds

In this subsection will be constructed the G2 holonomy metrics corresponding to
Examples 1 and 2. After extend them to a vacuum configuration of the eleven-dimensional
supergravity it will be obtained type IIA backgrounds by reduction along one of the
isometries.
In [24] it has been found the explicit form of (5.4) when the base space (and, in
consequence, the total one) has torus symmetry. The expression is
ds 2 =
r2
dr 2
+
U d 2 + U d d + U d 2 + Q d + Q d
h(r)
2

+ g d2 + d 2 + H ,
(7.1)
188

h(r) = 1 4c/r 4,
U11 = g + h(r)
u21 ( 2 + 2 ) + (u2 + u3 )2
,
2F 2
U22 = g + h(r)
u21 + u22
,
2F 2
U12 = U21 = g + h(r)
(u21 + u22 ) + u2 u3
,
2F 2

1
u1 ( du2 + du3 ) (u2 + u3 ) du1 u1 (u3 u2 )A1 ,
Q = h(r)
F

u1 du2 u2 du1 u1 u3 A1
Q = h(r)
,
F

2
H = h(r) |d u |2 + u21 + u22 A1 2A1(u3 du2 u2 du3 ) .
The second rank tensor gab is the metric of the base manifold. The product metric of
(7.1) with M 4
r2
dr 2
2
2
+
ds11
= dsM
+
U d 2 + U d d + U d 2 + Q d + Q d
h(r)
2

+ g d2 + d 2 + H
(7.2)
is a vacuum configuration of the eleven-dimensional supergravity [19]. With the help of
the quantities
U22 Q U12 Q
U11 Q U12 Q
,
2 =
,
det U
det U
h = H + U11 12 + 2U121 2 + U22 22 ,
1 =
1 = ,
2 = ,
the metric (7.2) is expressed in more simple manner as

2
2
ds11
= dsM
4 +

r2
dr 2
+
Uij (di + i )(dj + j ) + h .
h(r)
2
The last expression takes the usual form of the KaluzaKlein ansatz

2
2
4
2
= e 3 D G dx dx + e 3 D d + dx C (x)
ds11
with the dilaton field and the RR 1-form defined by
2

r U11
3
D = Log
,
4
2
U12 d + Q
C=
.
U11
(7.3)
The reduction of (7.3) along 1 gives the following IIA metric:

2

r U11 1/2
dr 2
2
dsA2 =
dsM
4 +
2
h(r)

2

r
2
2
+
det U d + 2(U11Q U12 Q ) d Q + U11 H .
2U11
189
(7.4)
The components of (7.4) are

2
r2
r U11 1/2
g =
h(r)
2
4U11 F

U11 sin2 ( ) U12

sin(2 ) sin() + sin2 ( ) ,
2
2
1/2
2
r
r U11
g =
h(r)U12 cos(),
2
4U11F
2

1/2
2
r 2 h2 (r) 2
r U11
r h(r)
2
g =
cos () ,
2
4
8U11F 2
2

r U11 1/2 r 2 h(r) 2

g =
sin ( )
2
4

2
r 2 h2 (r)
2
,
sin
(
)
sin(
)
+
cos(
)
sin()
8U11 F 2

2

r U11 1/2 r 2 h2 (r)
g =
sin( ) cos() sin( ) + cos( ) sin() ,
2
2
4U11 F
where it have been introduced the spherical coordinates , through the relations
u1 = sin( ) cos(),
u2 = sin( ) sin(),
u3 = cos( ).
The range of this coordinates is [0, ] and [0, 2]; the other components of the
metric are identically zero.
The base metric (6.9) have a singularity at 0. For this case it is obtained
2 2

u1 ( + 2 ) + (u2 + u3 )2
2 (1 + 3) + 2 (9 + 7)
f (x i )
U11 =
+
2h(r)
,
2 5
4
5

2
u1 + u22
f (x i )
8
,
U22 = 4 + 2h(r)
4
4

2
(u1 + u22 ) + u2 u3
f (x i )
8
,
U12 = U21 = 4 + h(r)
4
4

2
2d
Q = h(r) 2 u1 ( du2 + du3 ) (u2 + u3 ) du1 u1 (u3 u2 )
f (x i , dx i )
,
3
190

2 d
1
f (x i , dx i )
du
u
du
u
u
,
u
1
2
2
1
1
3
2

d2

d
f (x i , dx i )
H = h(r) |d u |2 + 4 u21 + u22 2 4 (u3 du2 u2 du3 )
,
Q = h(r)
where x i denotes all the coordinates except and the behaviour for short distances was
evaluated. The dilaton field is given explicitly as
A =
2
2

r (1 + 3) + 2 (9 + 7)
3
Log
4
2
2 5

2 2
u ( + 2 ) + (u2 + u3 )2
f (x i )
+ r 2 h(r) 1
Log
.
4
5
The expression for the RR one form is

C=
2{8 + h(r)[(u21 + u22 ) + u2 u3 ]} d

2 (1 + 3) + 2 (9 + 7) + 4h(r)[u21 ( 2 + 2 ) + (u2 + u3 )2 ]
4 2 h(r)[u1 ( du2 + du3 ) (u2 + u3 ) du1 2u1 (u3 u2 ) d]

2 (1 + 3) + 2 (9 + 7) + 4h(r)[u21 ( 2 + 2 ) + (u2 + u3 )2 ]

f x i .
+
The components of (7.4) diverges in this case for short ,

g
f (x i )
,
5
f (x i )
,
4
f (x i )
,
2
f (x i )
.
f (x i )
,
2
The quaternionic space (6.10) is singular too in the limit 0. Using it as a base space
gives
84 2 (19 + 5) + 166 (9 + 7) + 6 (1 + 35) + 32 4 (35 + 61)
9 5 ( 2 42 )2 (84 + 62 2 + 3 4 )

2 2
u ( + 2 ) + (u2 + u3 )2
f (x i )
,
+ 8h(r) 1
4
2
2
2
9 ( 4 )
5

u21 + u22
f (x i )
64(44 + 4 )
+
8h(r)
U22 = 4 2
,
2
2
4
2
2
4
4
2
2
2
9 ( 4 ) (8 + 6 + 3 )
9 ( 4 )
4
2

(u1 + u22 ) + u2 u3
32(85 43 2 + 3 4 )
+
8h(r)
U12 = U21 = 4 2
9 ( 42 )2 (84 + 62 2 + 3 4 )
9 4 ( 2 42 )2
U11 =
f (x i )
,
4
191

1
u1 ( du2 + du3 ) (u2 + u3 ) du1
3 2 (42 2 )

f (x i , dx i )
8
4( 2 22 )
d
u1 (u3 u2 ) 2
d
+
,
2
2
2
4
( 4 )
3

4
Q = h(r) 2 2
u1 du2 u2 du1
3 (4 2 )

f (x i , dx i )
4( 2 22 )
8
d
u1 u3 2
d
+
,
42
( 2 42 )
3

2

8
4( 2 22 )
H = h(r) |d u |2 + u21 + u22
d
+
d
2 42
( 2 42 )

8
4( 2 22 )
f (x i , dx i )
d
(u
.
2 2
d
+
du
u
du
)
3
2
2
3
42
( 2 42 )
Q = 4h(r)
The dilaton field is expressed through the relation

4
r 2 84 2 (19 + 5) + 166(9 + 7) + 6 (1 + 35) + 32 4 (35 + 61)
D
3
e
=
2
9 5 ( 2 42 )2 (84 + 62 2 + 3 4 )

2 2
u1 ( + 2 ) + (u2 + u3 )2
,
+ 8h(r)
9 4 ( 2 42 )2
from where follows that

f (x i )
,
D Log
5
0.
The behaviour of the RR one-form at short distances results

C f x i , dx i
and the components of the IIA metric diverges as
f (x i )
,
5
f (x i )
,
2
f (x i )
,
4
f (x i )
.
f (x i )
2
A detailed analysis of the singularities of the backgrounds corresponding to the m-pole

solutions and their physical interpretation was given in [24] and [25], the interested reader
may consult those references.
Acknowledgements
I thank A. Isaev for introducing me in the subject, M. Tsulaia and A. Pashnev for much
valuable discussions and E. Ivanov for pointing me out certain features about quaternionic
manifolds. Finally, I would like to thank L. Masperi, B. Dimitrov and D. Mladenov for
constructive critics and encouragement.
192
References
[1] M. Berger, Bull. Soc. Math. France 83 (1955) 279.
[2] R. Bryant, Ann. Math. 126 (1987) 525.
[3] D. Joyce, J. Diff. Geom. 43 (1996) 2;
D. Joyce, J. Diff. Geom. 43 (1996) 329.
[4] G. Gibbons, D. Page, C. Pope, Commun. Math. Phys. 127 (1990) 529.
[5] R. Bryant, S. Salamon, Duke Math. J. 58 (1989) 829.
[6] A. Brandhuber, J. Gomis, S. Gubser, S. Gukov, Nucl. Phys. B 611 (2001) 179.
[7] A. Brandhuber, Nucl. Phys. B 629 (2002) 393.
[8] D. Joyce, Compact Manifolds with Special Holonomy, 1st Edition, Oxford Univ. Press, 2000.
[9] Y. Konishi, M. Naka, Class. Quantum Grav. 18 (2001) 5521.
[10] M. Cvetic, G. Gibbons, H. Lu, C. Pope, Phys. Lett. B 534 (2002) 172.
[11] M. Cvetic, G. Gibbons, H. Lu, C. Pope, Orientifolds and slumps in G2 and Spin(7) metrics, hep-th/0111096.
[12] M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, Phys. Rev. D 65 (2002) 106004.
[13] Z.W. Chong, M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, P. Wagner, Nucl. Phys. B 638 (2002) 459.
[14] M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, Bianchi IX self-dual Einstein metrics and singular G2
manifolds, hep-th/0206151.
[15] J. Gutowski, G. Papadopoulos, Nucl. Phys. B 615 (2001) 237.
[16] N. Hitchin, Stable forms and special metrics, math.DG/0107101.
[17] B. Acharya, M. OLoughlin, Phys. Rev. D 55 (1997) 4521.
[18] D. Page, C. Pope, Class. Quantum Grav. 3 (1986) 249.
[19] M. Duff, B. Nilsson, C. Pope, Phys. Rep. 130 (1986) 1.
[20] B. Acharya, Adv. Theor. Math. Phys. 3 (1999) 227.
[21] E. Witten, M. Atiyah, M-theory dynamics on a manifold of G2 holonomy, hep-th/0107177.
[22] B. Acharya, E. Witten, Chiral fermions from manifolds of G2 holonomy, hep-th/0109152.
[23] A. Brandhuber, P. Berglund, Nucl. Phys. B 641 (2002) 351.
[24] L. Anguelova, C. Lazaroiu, JHEP 0210 (2002) 038;
L. Anguelova, C. Lazaroiu, Enhanced gauge symmetry from toric G2 cones, hep-th/0208177.
[25] L. Anguelova, C. Lazaroiu, M-theory compactifications on certain toric cones of G2 holonomy, hepth/0204249.
[26] K. Behrdnt, G. DallAgata, D. Lust, S. Mahapatra, JHEP 0208 (2002) 027;
K. Behrdnt, Nucl. Phys. B 635 (2002) 158.
[27] M. Aganagic, C. Vafa, Mirror symmetry and a G2 flop, hep-th/0105225.
[28] M. Atiyah, J. Maldacena, C. Vafa, J. Math. Phys. 42 (2001) 3209.
[29] R. Gopakumar, C. Vafa, Adv. Theor. Math. Phys. 5 (1999) 413.
[30] S. Kachru, J. McGreevy, JHEP 0106 (2001) 027.
[31] C. Vafa, J. Math. Phys. 42 (2001) 2798.
[32] I. Bakas, E. Floratos, A. Kehagias, Phys. Lett. B 445 (1998) 69.
[33] H. Partouche, B. Pioline, JHEP 0103 (2001) 005;
P. Kaste, A. Kehagias, H. Partouche, JHEP 0105 (2001) 058.
[34] S.V. Ketov, Summing up D-instantons in N = 2 supergravity, hep-th/0209003.
[35] R. Blumenhagen, V. Braun, JHEP 0112 (2001) 006.
[36] T. Eguchi, Y. Sugawara, Nucl. Phys. B 630 (2002) 132;
K. Sugiyama, S. Yamaguchi, Phys. Lett. B 538 (2002) 173.
[37] A. Belhaj, J. Phys. A 35 (2002) 8903.
[38] A. Misra, JHEP 0210 (2002) 056.
[39] S.L. Shatashvili, C. Vafa, Superstrings and manifolds of exceptional holonomy, hep-th/940702.
[40] B. Acharya, Nucl. Phys. B 524 (1998) 262.
[41] R.D. Schafer, An Introduction to Nonassociative Algebras, Academic Press, New York, 1966.
[42] A. Belavin, A. Polyakov, A. Schwartz, Y. Tyupkin, Phys. Lett. B 59 (1975) 85;
G. t Hooft, Phys. Rev. Lett. 37 (1976) 8.
[43] T. Eguchi, A. Hanson, Ann. Phys. 120 (1979) 82;
S. Hawking, Phys. Lett. A 60 (1979) 81;
V. Belinskii, G. Gibbons, D. Page, C. Pope, Phys. Lett. B 76 (1978) 433.
[44] T. Eguchi, P. Gilkey, A. Hanson, Phys. Rep. 66 (1980) 214.

[45] A. Bilal, J. Derendinger, K. Stefesos, Nucl. Phys. B 628 (2002) 112.
[46] R.S. Ward, Class. Quantum Grav. 7 (1990) L95;
N. Woodhouse, Class. Quantum Grav. 6 (1989) 933.
[47] D. Calderbank, J. Geom. Phys. 36 (2000) 152.
[48] D. Calderbank, H. Pedersen, Selfdual Einstein metrics with torus symmetry, math.DG/0105263;
D. Calderbank, M. Singer, Einstein metrics and complex singularities, math.DG/0206229.
[49] D. Calderbank, H. Pedersen, Ann. Inst. Fourier 50 (2000) 921.
[50] R. Bielawski, A. Dancer, Commun. Anal. Geom. 8 (2000) 727.
[51] H. Pedersen, Math. Ann. 274 (1986) 35.
[52] P. Casteill, E. Ivanov, G. Valent, Phys. Lett. B 508 (2001) 354;
P. Casteill, E. Ivanov, G. Valent, Nucl. Phys. B 627 (2002) 403.
[53] P. Fre, Nucl. Phys. B (Proc. Suppl.) 45 (1996) 59.
193

On the AdS/CFT dual of deconstruction

Ph. Brax a , R.A. Janik b , R. Peschanski a
a Service de Physique Theorique, CEA-Saclay, F-91191 Gif-sur-Yvette Cedex, France
b Jagellonian University, Reymonta 4, 30-059 Krakow, Poland
Received 12 March 2003; accepted 21 March 2003
Abstract
We consider a class of non-supersymmetric gauge theories obtained by orbifolding the N = 4
super-YangMills theories. We focus on the resulting quiver theories in their deconstructed phase,
both at small and large coupling, where a fifth dimension opens up. In particular, we investigate
the rle played by this extra dimension when evaluating the rectangular Wilson loops encoding the
interaction potential between quarks located at different points in the orbifold. The large coupling
potential of the deconstructed quiver theory is determined using the AdS/CFT correspondence and
analysing the corresponding minimal surface solution for the dual gravitational metric. At small
coupling, the potential between quarks decreases with their angular distance while at strong coupling
we find a linear dependence at large distance along the (deconstructed) fifth dimension.
1. Introduction
Recently there has been a renewed interest in extensions of the standard model
which differ from the supersymmetric framework. One of the initial motivations for
the supersymmetric extensions of the standard model is the possibility of preserving a
hierarchy between the weak scale and the unification scale. In supersymmetric theories this
springs from the delicate balance between bosonic and fermionic contributions to radiative
corrections. In particular, at the one loop level the quadratic divergences exactly vanish.
In the context of softly broken supergravity, the supertrace of the square mass matrix is
proportional to the gravitino mass, hence field independent and preserving the high energy
features of unbroken supersymmetry. The deconstructed models [14] offer an alternative
E-mail addresses: brax@spht.saclay.cea.fr (Ph. Brax), ufrjanik@if.uj.edu.pl (R.A. Janik),
pesch@spht.saclay.cea.fr (R. Peschanski).
doi:10.1016/S0550-3213(03)00253-0
Ph. Brax et al. / Nuclear Physics B 660 (2003) 194210
195
to this scenario. In these models the quadratic divergences are absent due to the equivalence
with a fifth-dimensional gauge theory forbidding the appearance of mass terms.
The deconstructed models have a structure highly reminiscent of quiver theories [5].
Indeed by considering the field theory limit of D3-brane configurations in the vicinity
of an orbifold singularity one can construct non-supersymmetric gauge theories with a
product gauge group U (n) and fields in bifundamental representations. These theories
have proved to be useful in building string realizations of the standard model [6]. Though
non-supersymmetric, the quiver theories share another feature with deconstructed models.
Indeed the quadratic divergences vanish exactly [7]. In the deconstructed phase of quiver
theories, where the gauge group is broken to the diagonal gauge group U (n) U (n)D ,
this springs from an underlying custodial supersymmetry. Similarly to the deconstructed
models, the quiver theories in the deconstructed phase are equivalent to a fifth-dimensional
U (n) gauge theory in the large limit.
In this paper we will investigate the properties of (rectangular) Wilson loops for quiver
theories in the deconstructed phase. In particular, we will focus on the interaction potential
between twisted quarks, i.e., quarks corresponding to open strings with end-points in
different sectors of the orbifold cover. In Section 2, we recall some ingredients about quiver
theories. In Section 3, we compute the quark potential at weak coupling. The development
of a fifth dimension is made explicit in the R5 /L2 behaviour of the potential. The L2
dependence signals the propagation of massless degrees of freedom in five dimensions
while the R5 factor is the only dimension-full constant of the five-dimensional theory. In
Sections 4, 5 and 6 we analyse the strong coupling behaviour. In Sections 4, we formulate
the computation of the rectangular Wilson loop in terms of a minimal surface having the
loop as boundary (Wilson surface) using the AdS/CFT correspondence. In Sections 5,
we analyze the geometry of the Wilson surface in and outside the deconstructed region. In
Sections 6, we finally compute the potential and thus the force between quarks along the
deconstructed dimension. We find that the quarks have a linear potential at large (angular)
distance, a property reminiscent of confinement along the fifth dimension.
2. Deconstructing non-supersymmetric quivers

We are interested in certain non-supersymmetric gauge theories whose structure can
be inferred from the world-volume dynamics of D3-branes in the neighbourhood of an
orbifold singularity. The breaking of supersymmetry is due to the orbifolding which does
not preserve the original N = 4 invariance of the low energy dynamics on D3-branes.
Consider the type IIB string theory with a stack of n coinciding D3-branes. It is well
known that the gauge bosons and fermions living on the worldvolume of the D3-branes
form a 4d N = 4 supersymmetric YangMills model with gauge group U (n). The six
transverse dimensions represent, from the point of view of the 4d theory living on the
worldvolume, six extra non-gravitational dimensions. The spectrum and interactions of that
model are the same as the ones obtained by dimensional reduction of the N = 1 U (n )
gauge theory living in 10 dimensions.
One can obtain a theory with fewer supersymmetries than N = 4 by dividing the extra
dimensions by a discrete group Z and embedding this orbifold group into the gauge
196
Fig. 1. Typical quiver diagram. The bosonic () and fermionic ( ) fields corresponding to zero-modes of
open strings with ends on branes are schematically represented by arrows. The figure corresponds to the
non-supersymmetric case a 1 2a4 2a1 .
group U (n ). The spectrum consists of the fields which are invariant under the combined
geometric and gauge actions of Z (see the quiver diagram in Fig. 1).
The field interaction terms are consistently truncated to yield a smaller daughter gauge
theory. The truncation process breaks the gauge group and some (or all) supersymmetries.
The gauge symmetry breaking is dictated by the embedding of the generator of Z into
U (n ). The matrix that represents the gauge action of Z is chosen to be of the form
of a direct sum of unit matrices of dimensions n n, each multiplied, respectively, by
2
i with = e i . The invariant components of the gauge fields fulfill the condition
A = A 1
(1)
where A is a matrix in the adjoint representation of U (n ). This leaves invariant the

subgroup U (n) .
There are four generations of Weyl fermions whose invariant components must obey
the condition
i = ai i 1
(2)
where i = 1, . . . , 4 and
a1 + a2 + a3 + a4 0 mod( ).
(3)
The invariant fermions transform in the bifundamental representations (nl , n l+ai ) of

the broken gauge group where l numbers blocks of the original n n matrices.
Furthermore, one obtains three generations of complex bosons i , i = 1, 2, 3, whose
invariant components fulfil the condition
i = a i i 1 .
(4)
The invariant scalars transform as (nl , n l+a i ) under the broken gauge group. The integers
a i correspond to the orbifold action zi e2i a i / zi on the three complex planes.
The truncated fields have a block structure in the U (n ) mother gauge group
i
lp
= li p,l+a i ,
i
lp
= li p,l+ai .
(5)
197
Supersymmetry is preserved when the group Z is embedded in SU(3)

a 1 + a 2 + a 3 0 mod( ).
(6)
In that case a4 0 and at least one of the fermions can be paired with the gauge bosons,
i.e., becoming a gaugino of N = 1 supersymmetry. We focus on the non-supersymmetric
case a4 = 0.
Let us move a stack of n D3-branes from the origin. Moving the stacks of n D3-branes
from the origin corresponds to a diagonal vev for each li when a i = 0. This is equivalent
to shifting a stack of n branes from the origin by a distance R. Due to the Z action, the
stacks have copies around the fixed point. The gauge group is broken to the diagonal
subgroup U (n)D . This is the deconstructed phase of the quiver theory with a breaking
pattern U (n) U (n)D .
We are interested in the geometry of the orbifold close to the branes. It is convenient
[8,9] to parametrize the orbifold with the coordinates
ai
zi = ri ei(i 1 +i 2 + )
(7)
where the vectors i , i and ai are orthogonal and we normalize 2 = 2 = 1. The flat
metric on the orbifold reads
3
ds 2 =
dzi d z i =
i=1
3
dri2 + R 2 d12 + R 2 d22 +
i=1
a 2R2 2
d ,
2
(8)
where ri = R(1 + xi ) and xi

1. We have defined a 2 = a 12 + a 22 + a 32 and assumed that
the orbifold acts non-trivially on the three complex planes. One recognizes a circle S 1
parametrized by [0, 2] corresponding to the Z orbit of radius
RS 1 =
ls2
,
R5
R5 =
ls2
.
aR
(9)
In the large limit, the vicinity of a stack of n branes corresponds to a cylinder of very
small radius RS 1 .
In this geometry the stack of D3-branes become localized at a point on the circle S 1 . It
is identified with its multiple images under the 2 rotation around the S 1 circle. It is now
easy to analyse the field theory on the stack of D3-branes. Indeed the six dimensions of the
orbifold have been replaced by a product R 3 (SR )2 S 1 where SR is a circle of radius
R. Now the field theory of a stack of n D3-branes localized in R 3 (SR )2 S 1 is a N = 4
U (n) SYM gauge theory corresponding to the massless open strings joining the stack to
itself, see Fig. 1. The massive states of the theory are obtained by wrapping open strings
around S 1 . The mass spectrum is given by
mk =
2k
R5
(10)
which is a KaluzaKlein spectrum of a fifth-dimensional theory compactified on a radius

R5 . The appearance of this extra dimension can be understood using an appropriate
T -duality [9]. Notice that RS 1
ls as soon as R5 ls , i.e., as keeping R
fixed. In the large limit, substringy distances are probed by the D3-branes. It is more
198
appropriate to T-dualize along S 1 . The radius of the new circle becomes the large radius
R5 . Similarly the D3-branes become wrapped D4-branes. On the D4-branes the gauge
theory is a U (n) five-dimensional gauge theory compactified on a circle of radius R5 . This
is the generalization to quiver theories of the deconstructed models.
In the following we will consider the deconstructed quiver theories both at small and
large coupling constant. In particular, we will focus on the Wilson lines. At weak coupling
the Wilson lines can be understood by moving one of the branes of a stack to infinity and
identifying the open string state connecting the stack of n D3-branes to the brane at infinity
as a static quark. Such an open string can wind w times around the orbifold. This is easily
pictured on the orbifold covering space where the string connects D3-branes belonging to
different sectors. We will refer to this situation as twisted quarks.
At strong coupling we will obtain relevant information from the supergravity solution
generated by the stacks of D3-branes. It has to be noticed that the supergravity background
with branes displaced to various points of the circle of radius R has a constant dilaton.
Hence the gauge theory coupling constant does not run with the energy scale (which is
related by the UV/IR principle to the radial distance scale in the bulk geometry). When we
pass to the orbifold the same is obviously true.
However, one has to keep in mind that there is a tachyonic twisted mode at the centre of
the orbifold which leads to an instability [10]. The behaviour of the dilaton may become
non-trivial leading to the fact that a running coupling may be generated at this stage. For the
part of the geometry relevant for the calculation of the static
potential between static quarks
along the deconstructed fifth dimension, and for R , the static potential should not
be affected by this tachyonic instability if the change of geometry is confined to a finite
region, close to the centre of the orbifold [10]. A more detailed study of this instability is
out of the scope of the present paper, but certainly deserves further investigation.
3. Rectangular Wilson loop at small coupling

The appearance of a perturbative extra dimension in the large regime can be
investigated by computing the quarkquark potential in the static approximation [11]. It
is expected that a 1/L2 dependence of the potential in the inter-quark distance L should
appear as required by a five-dimensional interpretation. We will show that this is indeed
the case. Moreover, we will be interested in the potential between quarks belonging to
different twisted sectors of the orbifold. Indeed in the universal cover of the orbifold one
can place quarks in different sectors identified under the orbifold action. We will compute
the potential as a function of the angle between the quarks in the universal cover. In
particular, we focus on an orbifold action on only one plane, i.e.,
a 1 2a4,
a 2 0,
a 3 0 mod( ),
(11)
corresponding to
a1 a4 ,
a2 a4 ,
a3 a4 mod( ).
(12)
In the following we will focus on the simplest situation a 1 1 mod( ), sketched in Fig. 1.
199
Fig. 2. Rectangular Wilson loop with twist. The space evolution of the fields along the Wilson line is
accompanied by a corresponding rotation (twist) between sectors of the orbifold, see text.
In this section we will compute the small coupling Wilson loop for deconstructed quiver
theories [12]
W=

1
m
i
Tr P eig C (Am x Xi u |x|)
,
n
(13)
where C is the Wilson loop contour parametrized by x m , and ui is a unit vector

representing the coupling to the six real scalar fields Xi , i = 1, . . . , 6, spanning the six
extra dimensions. We will concentrate on the case where the contour is associated to a
rotation along one plane in the six extra-dimensional directions. The twist of this contour
is parametrized by an angle / representing a rotation between two sectors of the orbifold,
see Fig. 2.
In our case

u = cos / , sin /
(14)
L
L
in between the two quarks as runs between 0 and L while
u = (1, 0)
(15)
for the static quark sitting at the origin and

u = (cos /, sin / )
(16)
for the static quark at distance L. The Wilson loop contour is represented in Fig. 2.
We also denote by T the time length of the Wilson loop. We will take it to be very large
eventually. The Wilson loop can be computed by going to Euclidean space and expanding
the connected Green functions to second order in the gauge coupling constant

lnW = g 2 T
) ei/
A0 (0, 0)A0(L, ) + (0, 0)(L,
C

+ (0, )(L,
0) ei/
(17)
where the first argument of A0 (0, ) is at the origin of spacetime where one quark sits and
the second is the time parametrizing the time evolution of the quark. We have denoted
200
2
= X1 +iX
. In the deconstructed phase one can decompose the fields into massless and
2
massive modes according to [7]

( 1)/2

(
1)/2

2
2n
2n
(n)
(n)
n cos
sin
Ap =
(18)
p Ap +
p Ap ,
n=0
n=0
where each field Ap represents one of the blocks and the index (n) numbers the
(n)
differents fields. Here we have identified 0 = 1/ 2 and n = 1, n = 0. The fields Ap
(n)
have masses mn while the fields A p have masses m{ n} . Similar expressions hold for
the scalar fields with the same mass spectrum. This fact is due to the underlying custodial
supersymmetry of the non-supersymmetric quiver theories [7]. Using the propagators

L
(n)

d 4 p eip0 +i p
Ap (0, 0)A(n)
(19)
(L,
)
=
p
(2)4 p02 + p2 + m2n
we find that the Wilson loop can be expressed as

lnW = g nT 1 + cos(/ )
T
d G5 ,
(20)
where
G5 =

1
n=0
L
d 4 p eip0 +i p
4
2
(2) p0 + p2 + m2n
(21)
and we have removed the self energy contributions. In the deconstruction regime this leads
to

L
d 5 p eip0 +i p
G5 = R5
(22)
(2)5 p02 + p2 + p52
which is nothing but the five-dimensional propagator
R5
.
+ L2 )3/2
For large T we deduce the interaction potential
G5 =
( 2
g 2 nR5
V=
L2

0
1 + cos /
g 2 nR5
=
{1 + cos / }.
2
3/2
(1 + u )
L2
(23)
(24)
Notice the expected 1/L2 behaviour, characteristic of the opening of the fifth dimension at
weak coupling.
The potential is periodic in / in the orbifold cover. Antipodal points are such that there
is maximal screening with a vanishing interaction potential. As / increases, the potential
decreases and the interaction force between two twisted sectors decreases. Finally the
potential is proportional to the radius of the compactified fifth-dimensional gauge theory
R5 , which sets its dynamical scale.
201
4. The rectangular Wilson loop at strong coupling

4.1. Dual metric deconstruction
We have seen that in the large limit, deconstructed quiver theories become similar
to a fifth-dimensional gauge theories. We will now analyse the models in the large
gauge coupling limit using the AdS/CFT correspondence [13,14]. To simplify the analytic
treatment we shall concentrate on the case where the orbifold group acts on a single
complex plane. The metric due to the presence of the displaced branes is given by
2
= H 1/2 ds42 + H 1/2 ds62 ,
ds10
(25)
where
H =1+

j =1
r04
|r rj |4
(26)
and rj =
is the location of the ith image of the displaced brane. The
complement metric ds62 is defined up to the orbifold identifications. In the case where
the orbifold acts on a single plane we denote by the polar angle in that plane. We are
interested in computing the Wilson line between quarks belonging to different sectors. In
the orbifold cover this amounts to putting quarks in sectors separated by an angle which is
a multiple of 2/ . At strong coupling the relevant regime is the blown-up vicinity of the
displaced brane where
(Re2ij/ , 0, 0)

i=1
r04
|r ri |4
(27)
while the quarks lie at infinity.

There are various relevant regions. Far away from the branes, spacetime becomes
isometric to AdS5 S 5 /Z . The quarks are on the boundary of AdS5 . In the interior of the
six extra dimensions, the geometry departs from the AdS5 behaviour. In particular, there is
a ring around the stacks of branes where the harmonic function signals the presence of a
fifth dimension. The geometrical setting in the orbifold plane has been sketched in Fig. 3.
The strong coupling calculation of Wilson loop expectation values reduces, in the
classical approximation of the AdS/CFT correspondence, to the evaluation [1416] of
the bulk minimal surface area bounded by the Wilson loop contour. In this context, the
(massive) quarks in the static limit are represented by an open string with ends near infinity
in r.
The nature of the minimal surface depends crucially on the harmonic function H and
on the physical length of the Wilson loop L. In the following we will consider the region
1
,
(28)
where x = r/R 1. In that region the harmonic function (26) can be well approximated
[11] by
x
H = r04
r 2 + R2
(r 2 R 2 )3
(29)
202
with no angular dependence, i.e., leading to the existence of two conserved quantities E
and l. Close to the circle containing all the branes, the harmonic function reads
r04 1
(30)
4R 4 x 3
which is valid for 1/
x x where x < O(1) defines a scale limiting the validity
of the approximation. The circle of radius R = R(1 + x ) is the outer edge of the region
where the behaviour of the harmonic function is similar to a five-dimensional harmonic
function decaying with the third power of the distance. We will call the region between R
and R the deconstruction domain.
Now, at very large distances the harmonic function becomes the AdS5 harmonic function
H=
H=
r04
.
r4
(31)

For the sake of clarity in the following calculation and discussion, let us denote by r R
the circle beyond which (31) is valid with a sufficiently good approximation. This will, in
particular, contain the region where the quark sources stand. Between these two regions,
the harmonic function H of (29) interpolates smoothly.
namely, for R < r < R,
4.2. Integrals of motion on the Wilson contour
We are interested in computing the interaction potential, i.e., the rectangular Wilson
loop in the large T limit, when quarks lie in different sectors of the orbifold. Within
the AdS/CFT correspondence scheme the potential is determined by the minimal surface
area swept by a string connecting the boundary quarks [15]. The geometry of this Wilson
surface, in particular, its behaviour in the deconstructed domain, will determine the basic
features of the force between quarks.
By comparison with the original evaluations of Wilson contours [15], there exist some
differences that we have to face. The quark sources, represented by D-branes at infinity
are a priori outside the deconstruction region, see Fig. 3. One thus has to take care of how
and where the minimal Wilson surface is attracted near the set of orbifolded D-branes.
This does require to determine (at least with sufficient accuracy) the Wilson line contour
inside and outside the deconstruction region. Let us thus first write the general equations
determining the minimal surface area.
This area is determined by the surface element
1/2

dS2 = H 1 + r 2 + r 2 2
(32)
d d,
where d/d. This leads to the potential

1/2

1
V (L, / ) =
.
d H 1 + r 2 + r 2 2

2
(33)
Let us note that for each fixed / we have at least two series of strings which contribute to
the Wilson loopone which stretches anticlockwise with f i = / + 2n, and one
which runs clockwise with f i = 2n /. Therefore, the Wilson loop expectation
203
Fig. 3. Wilson surface projected on the fifth dimension. The minimal surface contours are displayed as
seen in the orbifold plane. The full (respectively dashed) curve correspond to the dominant (respectively first
subdominant) contribution to the potential (see Section 4.2).
R: Location radius of the D3 -branes: Thick dashed circle;
R r R : Deconstructed phase region: Hatched ring;
Lower limit of (approximate) validity of the AdS S5 metrics: Dotted circle;
r = R:
/D (respectively /F ): half the angle spanned by the Wilson contour inside (respectively outside) the
deconstruction region (see Section 5).
value is given by the infinite sum

T V (L,/+2n)

+ eT V (L,2(n+1)/) .
W (T L) =
e
(34)
n0
For T only the two first contributions will survive giving the effective potential

Veff (L, / ) = min V (L, / ), V (L, 2 / ) .
(35)
The potential is explicitly periodic. In the following we will just determine the function
V (L, / ).
The potential can be cast into the form

1/2

1
,
V=
(36)
dr 1 + H 1 2 + r 2 2

2
where d/dr. In the following we shall use the Lagrangian

1/2
L = 1 + H 1 2 + r 2 2
,
(37)
204
where H does not depend on either or . This leads to the existence of two integrals of
motion analogous to the energy and the angular momentum
H 1
= cst.,
L
r 2
= cst.
l=
L
This implies that the Lagrangian can be expressed as

l2 1/2
2
L = 1 HE 2
.
r
The equations of motion can then be deduced
E=
EH
,
(1 H E2 l2 /r 2 )1/2
l
1
= 2
.
r (1 H E2 l2 /r 2 )1/2
The total length of the Wilson line will be 2L where
(38)
(39)

L=
Rmin
EH
dr
(1 H E2 l2 /r 2 )1/2
(40)
(41)
and the twist angle, 2/ , with

/ =
Rmin
l
1
dr.
2
2
r (1 H E l2 /r 2 )1/2
(42)
It is important to note that the minimal surface will not reach the circle of branes
but stops at a minimal value Rmin where the Lagrangian L, see (39) goes to infinity.
More precisely, since then dr/d 1/ L, the minimized Wilson contour will follow
a trajectory near Rmin , spanning an angle 2/D inside the deconstruction region, and
2/F outside this region, see Fig. 3.
5. Dual geometry of deconstruction

5.1. Length and twist of the Wilson line in the deconstruction domain
Let us first focus on the contribution to the length LD of the Wilson line in the
deconstruction region.
Considering (41), together with the harmonic function (30), and using an appropriate
rescaling of variable, one obtains
2
r0
LD = 1/2
R
x
1
min
x/xmin
u3/2
du
,
u3 1
(43)
where we require that r = R(1 + x) lies within the deconstruction region. One gets

r4
x
xmin = 2 0 F 2
,
xmin
LD R 2
205
(44)
where
y
F (y) =
1
u3/2
1/2 7/2

du
2
y3 1
y 2 F1 7/6, 1; 3/2|1 y 3 .
3
u 1 3
(45)
Looking for a solution where xmin/x

1 for Rmin to lie in the deconstruction region, we
find to leading order
r04
xmin =
L2D R 2
where
F F () =
F 2,
(46)
2 (2/3)
0.86237.
(1/6)
(47)
Imposing that 1/
xmin
1 implies that the deconstruction regime is characterized by

1/2
F (g 2 n)1/2 R5
R5 .
LD
F g 2 n
(48)
Note that (48) has the meaning of a limitation of the 4-dimensional distance LD over
which the potential calculation can be done within our approximation scheme. Larger 4D
distances will be briefly mentioned in the conclusion.
We find that the integrals of motion (38) can be expressed as
1
2R 2 5/2
,
E = 2 xmin 2
2 /G2 )1/2
(xmin + /D
r0
/D
R
l=
.
2 + / 2 /G2 )1/2
G (xmin
D
(49)
Hence the twist in the deconstruction region, which we choose as the starting value for our
evaluation can be identified with
x
/D =
dx.
(50)
xmin
This leads to
2R 2 3/2
1
E = 2 xmin
,
2 /x 2 )1/2
(1 + /D
r0
/D
R
l=
2 /x 2 )1/2
x (1 + /D
(51)
206
where
x/xmin
u3/2
,
du
u3 1
G=
1
(52)
and G x /xmin diverges linearly as xmin /x

1.
Using the previous results cf. (30), (51), we find that
H E2
2 2
r04 E2
1
l2 /D
<
,
2 /x 2
4R 4 x 3
R 2 x2
1 + /D
(53)
therefore l2 /r 2 is much larger than H E2 for r R provided

/D
1.
x
(54)
Given /D , which will ultimately define the quark separation distance along the
deconstructed dimension, we thus choose the cut-off x in such a way as to verify the
condition (54). This implies that the width of the deconstructed region we consider is
sizeably smaller than the angle covered by the Wilson line in this deconstruction region.
5.2. Length and twist of the Wilson line outside the deconstruction domain
Since the quarks sources are initially placed at infinity, one cannot make definite
conclusions on the interquark potential without studying the outside of the deconstruction
domain. Let us thus discuss the geometrical features of the Wilson surface solution when
r > R . Following the indications of Fig. 3, the discussion may imply two regions, one
where one can use the conformal AdS5 S5 metrics and one transition region
with r > R

R < r < R.
The discussion depends mainly on the range of values of the twist /D one considers
within the deconstruction domain. If we are considering /D large enough, which will
correspond to a large distance in the deconstructed fifth dimension, then the relation (54)
is small
stands for finite values of x . This means that the transition region R < R < R
and thus not so much contributing to the potential.
Using then (53), one considers valid the condition
H E2
l2
.
r2
(55)
This implies that the Lagrangian for r > R becomes

L=
(r 2
r
.
l 2 )1/2
(56)
This is notably different from the case of non-rotating strings in the AdS5 S5 case,
where the H E2 term is the only non-trivial contribution in the Lagrangian. Notice that
the previous approximation is valid in all the region outside the deconstructed domain. The
207
contribution to the string length outside the deconstructed region is

E r 4
LF = 2 3 0
l

R /l
u2 + u2R
(u2
u2R )3
u
du,
(u2 1)1/2
(57)
3 while L scales like

where uR = R/ l. This is negligible for xmin
1 as it scales as xmin
D
1/ xmin . We then obtain that
L LD ,
LF
LD ,
(58)
an approximation which improves for large L.

The situation is different for the bending of the string (see Fig. 3). The bending angle of
the string in the far-away region is given by

2

r
du
/F =
(59)
/2 arctan
1
l
u(u2 1)1/2
r /l
which is a finite quantity. For large values /D /x 1 we find that l = R and

/F /2. Notice that the bending in the far-away region saturates. Hence it is possible
to ascribe a definite initial position to the source quarks in the far-away region in such a
way that the minimal Wilson contour travels over the deconstruction region for a given
value of /D .
If we are now considering /D smaller, which will correspond to smaller distances
in the deconstructed fifth dimension, then the relation (54) requires x
1. This also
is large and will require using the
means that the transition region R < R < R
interpolating function H of (29) for the minimization. Without considering this more
complex minimization problem, let us note that the function (29) contains singular tems
in 1/x 2 and 1/x which are expected to modify the interquark potential obtained from the
1/x 3 term of (30). All in all, this means that our predictions are only valid for the large
angular distance behaviour of the potential.
6. The static interquark potential: results and comments

We can now be more specific and evaluate the various contributions to the static
potential for the large distance regime (large /D ).
The potential from the deconstruction region is simply

Rxmin
2 + / 2 .
VD =
(60)
G2 xmin
D
2
As LD increases this gives
VD =
ng 2 F 2 R5
2L2D
2.
x2 + /D
(61)
208
The first term leads to the expected 1/L2 behaviour for a five-dimensional theory. When
/D x this reads
VD =
ng 2 F 2 x5
,
2L2D
(62)
where we have put x5 = R5 /D . The potential is proportional to the t Hooft coupling

g 2 n while the potential is in (g 2 n)1/2 for the usual AdS case. Moreover, VD is both
proportional to R5 and /D . It is tempting to interpret the combination x5 = R5 /D as a
fifth-dimensional distance between the quarks. In that case, the behaviour of the potential is
reminiscent of confinement along the fifth dimension. It has to be noted that the would-be
string tension
g 2 nF 2
(63)
2L2
depends on the four-dimensional distance between the quarks, but this dependence is valid
only in the limited region L
(g 2 n)1/2 R5 (cf. (48)).
Let us briefly comment on the physical meaning of the potential VD (/D , LD ). If
we had inserted the probe D3-branes (carrying the Wilson loops) at the edge of the
deconstructed region, i.e., at R , we would have had massive W -bosons in the theory
and VD (/D , LD ) would be related to their interaction potential. However, in order to
make the sources infinitely massive, and hence to have the analogues of quarks in the
fundamental representation, we have to move the probe D3-branes and the Wilson loop
away to r . Then the parameter is modified by F , i.e., / /D + /F
according to (59). This drives the angles / to larger values and changes the potential.
Nevertheless the contribution to the potential from the far away region is nearly trivial as is
shown by the following computation. Hence, the main contribution to the potential comes
from the deconstructed region and can be interpreted as the interaction potential as seen by
static quarks on the edge of the deconstructed domain.
For completeness, let us estimate the contribution to the potential from the domain far
outside the deconstruction region. One obtains
T =
/V = VF V0 ,
(64)
R
where VF is the contribution for r

and V0 is a regularization associated with an
infinitely massive straight string. The contribution from the far-away region is
l
VF V0 =
2

r /l

u
1 2 2
r r l .
du
2
(u2 1)1/2
(65)
Notice that this contribution is independent of L. We are particularly interested in the large
/D x region. In that case l = R and therefore this reduces to a constant contribution
to the potential.
Let us end with some comments on our results. Having analysed non-supersymmetric
quiver theories in their deconstructed phase both at small and large coupling, we have
retrieved the expected fifth-dimensional behaviour. In particular, examining the angular
209
dependence of the quark potential we have found a striking difference between the small
and large coupling regimes. In the former, the force between the quarks decreases and
vanishes for antipodal quarks. This implies that static quarks behave like non-interacting
particles as long as they sit at antipodal points in the orbifold. At strong coupling we have
found that the potential increases linearly with the angular distance, mimicking a confining
behaviour along the large compact extra dimension.
These results are valid as long as L
(g 2 n)1/2 R5 . In the deep infrared regime where
L (g 2 n)1/2 R5 , the open string probes a region where x
1/ . In this regime the
harmonic function H becomes identical with the one of a single stack of n D3-branes.
This corresponds to the fact that at very low energy the deconstructed quiver theory is a
N = 4 U (n) gauge theory. As well known from deconstructed models, it is only in a finite
energy range, i.e., a finite interval in L, that the gauge theory looks five-dimensional.
The situation would be entirely different if we had considered R . In that case, the
non-supersymmetry of the configuration is signaled by the presence of a twisted tachyon
deforming the geometry of spacetime. The analysis of the dual gauge theory in this case
is beyond the scope of this paper [10].
Acknowledgements
R.P. wishes to thank Stefan Pokorski for a fruitful discussion. R.J. thanks the Service de
Physique Thorique for hospitality when part of this work was done. R.J. was supported in
part by KBN grant 2P03B09622.
References
[1] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Rev. Lett. 86 (2001) 4757;
C.T. Hill, S. Pokorski, J. Wang, Phys. Rev. D 64 (2001) 105005.
[2] C. Csaki, G.D. Kribs, J. Terning, Phys. Rev. D 65 (2002) 015004;
H.-C. Cheng, K.T. Matchev, J. Wang, Phys. Lett. B 521 (2001) 308;
N. Arkani-Hamed, A.G. Cohen, H. Georgi, hep-th/0108089;
P.H. Chankowski, A. Falkowski, S. Pokorski, JHEP 0208 (2002) 003.
[3] C. Csaki, J. Erlich, C. Grojean, G. Kribs, Phys. Rev. D 65 (2002) 015003;
T. Kobayashi, N. Maru, K. Yoshioka, hep-ph/0110117;
N. Arkani-Hamed, A.G. Cohen, H. Georgi, JHEP 0207 (2002) 020;
Z. Chacko, E. Katz, E. Perazzi, Phys. Rev. D 66 (2002) 095012.
[4] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Lett. B 513 (2001) 232;
H.-C. Cheng, C. Hill, J. Wang, Phys. Rev. D 64 (2001) 095003;
N. Arkani-Hamed, A.G. Cohen, T. Gregoire, J.G. Wacker, JHEP 0208 (2001) 4757.
[5] M.R. Douglas, G. Moore, D-branes, Quivers, and ALE instantons, hep-th/9603167;
M.R. Douglas, B.R. Greene, D.R. Morrison, Nucl. Phys. B 506 (1997) 84;
S. Kachru, E. Silverstein, Phys. Rev. Lett. 80 (1998) 4855.
[6] G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, JHEP 0102 (2001) 047.
[7] P. Brax, A. Falkowski, Z. Lalak, S. Pokorski, Phys. Lett. B 538 (2002) 426;
P. Brax, Z. Lalak, Acta Phys. Pol. B 33 (2002) 2399.
[8] S. Mukhi, M. Rangamani, E. Verlinde, JHEP 0205 (2002) 023.
[9] N. Arkani-Hamed, A.G. Cohen, D.B. Kaplan, A. Karch, L. Motl, JHEP 0301 (2003) 083.
[10] A. Adams, J. Polchinski, E. Silverstein, JHEP 0110 (2001) 029.
210
[11] K. Sfetsos, Nucl. Phys. B 612 (2001) 191.

[12] A.A. Tseytlin, K. Zarembo, Phys. Rev. D 66 (2002).
[13] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 506 (1998) 105;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253.
[14] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183.
[15] J. Maldacena, Phys. Rev. Lett. 80 (1998) 4859;
S.-J. Rey, J. Yee, Eur. Phys. J. C 22 (2001) 379, hep-th/9803001.
[16] D.J. Gross, H. Ooguri, Phys. Rev. D 58 (1998) 106002.

Nonlinear corrections to the DGLAP equations

in view of the HERA data
K.J. Eskola a,b , H. Honkanen a,b , V.J. Kolhinen a,b ,
Jianwei Qiu c , C.A. Salgado d
a Department of Physics, University of Jyvskyl, P.O. Box 35, FIN-40014 University of Jyvskyl, Finland
b Helsinki Institute of Physics, P.O. Box 64, FIN-00014 University of Helsinki, Finland
c Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA
d CERN, Theory Division, CH-1211 Geneva, Switzerland
Received 2 December 2002; received in revised form 12 March 2003; accepted 25 March 2003
Abstract
The effects of the first nonlinear corrections to the DGLAP evolution equations are studied by
using the recent HERA data for the structure function F2 (x, Q2 ) of the free proton and the parton
distributions from CTEQ5L and CTEQ6L as a baseline. By requiring a good fit to the H1 data,
we determine initial parton distributions at Q20 = 1.4 GeV2 for the nonlinear scale evolution. We
show that the nonlinear corrections improve the agreement with the F2 (x, Q2 ) data in the region
of x 3 105 and Q2 1.5 GeV2 without paying the price of obtaining a worse agreement
at larger values of x and Q2 . For the gluon distribution the nonlinear effects are found to play an
increasingly important role at x 103 and Q2 10 GeV2 , but rapidly vanish at larger values of
x and Q2 . Consequently, contrary to CTEQ6L, the obtained gluon distribution at Q2 = 1.4 GeV2
shows a power-like growth at small x. Relative to the CTEQ6L gluons, an enhancement up to a factor
6 at x = 105 , Q20 = 1.4 GeV2 reduces to a negligible difference at Q2 10 GeV2 .
PACS: 13.60.Hb; 12.38.Cy; 12.39.St; 12.38.Bx
E-mail addresses: kari.eskola@phys.jyu.fi (K.J. Eskola), heli.honkanen@phys.jyu.fi (H. Honkanen),

vesa.kolhinen@phys.jyu.fi (V.J. Kolhinen), jwq@iastate.edu (J. Qiu), carlos.salgado@cern.ch (C.A. Salgado).
doi:10.1016/S0550-3213(03)00257-8
212
K.J. Eskola et al. / Nuclear Physics B 660 (2003) 211224
1. Introduction
Parton distribution functions (PDFs), fi (x, Q2 ), are needed for the computation of
inclusive cross-sections of hard, collinearly factorizable, processes in hadronic collisions.
At sufficiently large values of the interaction scale Q2 and the momentum fraction x,
where power corrections are negligible, the scale evolution of the PDFs is predicted
quite accurately by the DokshitzerGribovLipatovAltarelliParisi (DGLAP) evolution
equations [1] derived from perturbative QCD.
In the global analyzes of the PDFs of the free proton, such as in Refs. [2,3], the
lowest order (LO), next-to-leading order (NLO) and next-to-next-to-leading order (NNLO)
parton distributions are extracted within the DGLAP framework using constraints from the
measured cross-sections of various hard processes and from the sum rules. In the proton
case the procedure is well established: the initial distributions given at some initial scale
Q20 are first evolved to larger Q2 by using the DGLAP equations, then a comparison with
the data is made over a wide range of x and Q2 , after which the initial distributions are
iterated until a good global fit to the data is obtained. The initial distributions are thus
the nonperturbative input needed, the element that perturbative QCD cannot predict. The
data from deeply inelastic leptonproton scattering (DIS) play a key role in these analyzes
especially in the region of the smallest values of x, where the DIS data from ep collisions
at DESY-HERA give the only constraints available. Recently, the H1 Collaboration at
HERA [4] has measured the structure function F2 (x, Q2 ) of the proton down to x
3 105 but still in the perturbatively accessible region Q2 1.5 GeV2 . These data have
been included in the recent global analyzes by the MRST [2] and CTEQ [3] collaborations.
In spite of the impressive success of the DGLAP approach, certain problems appear in
the attempts to make the global fits to the H1 data [4] as good as possible simultaneously
in the region of Q2 > 4 GeV2 (large Q2 ) and in the region of 1.5 GeV2 < Q2 < 4
GeV2 (small Q2 ). In the recent NLO analysis MRST2001 [2] both regions are included
and a good overall fit is found but with the expense of allowing for a negative gluon
distribution. Although a negative contribution in the NLO gluon distribution is acceptable
as long as the NLO cross-sections remain positive, the interpretation of the PDFs as
probability or number density distributions becomes obscured.1 On the other hand, the
CTEQ Collaboration emphasizes the large-Q2 region in their global fits: e.g., in the sets
CTEQ5 [6] and CTEQ6 [3] only the region Q2 > 4 GeV2 of the DIS data is included in the
fit and a very good agreement with the data is found. The agreement of the extrapolation to
the small-Q2 region, however, becomes then worse. One is facing the problem of negative
gluon distributions also in the NLO set CTEQ6M [3]: xg(x, Q2 ) is set to zero at the
smallest values of x at scales Q2 1.69 GeV2 . In LO, the negative gluon distributions,
however, are not allowed. The quality of the LO fits to the H1 data (see Table 1 in Section 3)
reflects the problem of a simultaneous fit to the small- and large-Q2 regions: MRST2001
(CTEQ6) fits the small-Q2 (large-Q2) region better.
The problems discussed above are very interesting as they can be a sign of a new QCD
phenomenon: towards smaller values of x and (or) Q2 (but still Q2 2QCD ), gluon
1 See also [5] for a recent discussion of the PDFs not being probabilities.
213
recombination effects are expected to play an increasingly important role. These effects
induce nonlinear power corrections to the DGLAP equations. First of the nonlinear terms
have been calculated by Gribov, Levin and Ryskin in [7] and by Mueller and Qiu in [8].
We shall refer to these corrections as the GLRMQ terms.
Previous studies of the GLRMQ terms in the context of extracting the PDFs of the free
proton can be found, e.g., in [9]. Also other nonlinear evolution equations relevant at high
gluon densities have been derived in the recent years [10], and the structure functions from
DIS have been analyzed in the context of saturation models [11]. In the present work,
however, we shall adopt the framework of collinear factorization with universal PDFs, and
search for the nonlinear GLRMQ corrections on top of the full DGLAP equations. This
allows for a direct comparison of our results with those of the global DGLAP fits [2,3]. In
this manner we can also show more explicitly the need for nonlinear terms in the evolution
equations.
In the DGLAP evolution, the Q2 -dependence of the sea quarks at small values of x
is dictated by the gluon distribution: the larger xg(x, Q2 ), the faster the Q2 evolution
of F2 (x, Q2 ), since in LO F2 (x, Q2 )/ log Q2 (10s /27)xg(2x, Q2 ) [12]. The
GLRMQ terms slow down the Q2 evolution of gluons and sea quarks from the standard
DGLAP behaviour (assuming the same starting distributions). Consequently, the log Q2
slopes of F2 (x, Q2 ) of the H1 data can be reproduced with a larger gluon distribution than
that in the conventional DGLAP case. In this paper, we investigate this interdependence
of the initial conditions at Q20 = 1.4 GeV2 and the effects of the GLRMQ terms by using
the H1 data as a baseline. We demonstrate that inclusion of the GLRMQ terms on top
of the LO DGLAP evolution improves the agreement with the data in the small-x and
small-Q2 region while simultaneously maintains the good fit of the LO sets of CTEQ5
and CTEQ6 to the data at larger values of x and Q2 . We show that the obtained small-x
gluon distribution can still have a power-like growth at the scale Q2 = 1.4 GeV2 , leading
to an enhancement of a factor 6 relative to the CTEQ6L gluons at x = 105 . We also
show explicitly how the large deviations from CTEQ6L reduce to very small differences at
Q2 10 GeV2 . The size of the nonlinear terms, uncertainties and applicability region of
the DGLAP + GLRMQ approach are discussed as well.
2. Nonlinear evolution equations

The GLRMQ corrections [7,8] arise from fusion of two gluon ladders, and they modify
the evolution equations of gluons as

1

xg(x, Q2 ) xg(x, Q2 )
dy 2 (2)
9 s2
y G y, Q2 ,
=
log Q2
log Q2 DGLAP
2 Q2
y
(1)
where the first term is the standard DGLAP result [1], linear in the PDFs. The 2-gluon
density in the second term we model as

x 2 G(2) x, Q2 =
2
1
xg x, Q2 ,
2
R
(2)
214
with R = 1 fm as the radius of the proton. Eq. (1) thus becomes nonlinear in xg. In the
scale evolution of the sea quark distributions the leading nonlinear correction from the
gluon fusion appears as [8]

xq(x, Q2) xq(x, Q2)
3 s2 2 (2)
(3)
x G x, Q2 ,

2
2
2
log Q
log Q
20 Q
DGLAP
where the first term is again from the DGLAP equations and linear in the PDFs. Note that
here we work under an approximation of neglecting contribution from the high twist
gluon distribution GHT (x, Q2 ) [8]. More discussion on this will be given later [13]. The
evolution of the valence quark distributions remains unmodified.
3. Analysis and results

3.1. Effects of the GLRMQ corrections
The emphasis of the CTEQ analysis is in the large-Q2 region where the nonlinearities
should remain small. Since we wish to search for the nonlinear effects in the small-Q2
(small-x) region but also recover the DGLAP evolution at large Q2 , the CTEQ distributions
are an ideal baseline for our analysis.
In Fig. 1, together with
H1 data [4], we plot the Q2 -dependence of the LO structure
the
2
2
function F2 (x, Q ) = q eq [xq(x, Q2) + x q(x,
Q2 )] computed from CTEQ5L [6,14] for
fixed values of x (dashed lines). The agreement with the H1 data is clearly getting worse
towards smaller values of x and Q2 . This trend is clearly visible also in Fig. 2 (dashed
line), where we quantify the quality of the fit in terms of
(xk ) =
2
n(xj )
xk

[F th (xj , Q(xj ) ) F exp (xj , Q(xj ) )]2
i
i
2
2
xj =0.2 i=1
exp
(xj )
[F2 (xj , Qi
(4)
)]2
x
divided by the cumulative number of data points, N(xk ) = xkj =0.2 n(xj ), as a function of
the x of the data. Here n(xj ) refers to the number of the data points with the same value of
x = xj . The 2 /N computed at different regions of Q2 is in turn summarized in Table 1.
Relative to the data [4] (which was not available at the time of CTEQ5) the computed
log Q2 slopes of F2 are too large at small values of x and Q2 . This is caused by a too large
gluon component in the DGLAP evolution at small values of x.
For showing how the GLRMQ terms slow down the scale evolution, we take the PDFs
from the CTEQ5L parametrization at Q2 = 5 GeV2 , and evolve these both downwards and
upwards by using the nonlinear evolution equations. The results are plotted in Fig. 1 (solid
lines). As seen in the figure, at larger values of x and Q2 the nonlinear effects remain small
and do not make the agreement with the data worse. In addition, as shown both by Figs. 1
and 2, at 5 105 x 103 the agreement with the data becomes quite good (i.e., 2 /N
remains constant), clearly improving the situation from the CTEQ5L case.
Another interesting observation from Fig. 1 is that, with the trial initial conditions
above, the nonlinear corrections at x 5 105 and Q2 1 GeV2 obviously become
215
Fig. 1. The scale evolution of the structure function F2 (x, Q2 ) of the free proton for fixed values of x (with
constants added to separate the curves). The dashed curves show the LO DGLAP result from CTEQ5L [6], and
the solid curve the result after the DGLAP + GLRMQ evolution when initial conditions taken from CTEQ5L at
Q20 = 5 GeV2 . The data is from H1 [4] and the error bars are statistical.
too large and they start to dominate the computed evolution, causing negative log Q2
slopes for F2 (x, Q2 ). Clearly, these are not supported by the data. In the computation
we would then have entered a gluon saturation region where also other correction terms
in the evolution equations should be included. This region is thus beyond the scope of the
present DGLAP + GLRMQ approach. The too strong nonlinearities can again be traced
back to a too large gluon component at small x in the initial condition chosen. A more
realistic gluon distribution which evolves according to the DGLAP + GLRMQ equations
should obviously be smaller than CTEQ5L at small values of x at Q2 = 5 GeV2 .
216
Fig. 2. The goodness parameter 2 of the fits of the computed F2 (x, Q2 ) to the H1 data, divided by the number
of data points, as a function of the x of the data. The cumulative number of the data points is increasing to the left
as indicated at the top of the plot: N (x = 0.2) = 2 and N (x = 3.2 105 ) = 133 (see also Table 1). The curves
are the LO DGLAP results from CTEQ5L (long dashed thick line) and CTEQ6L (dotted-dashed thick line), the
DGLAP + GLRMQ result with the initial conditions at Q2 = 5 GeV2 taken from CTEQ5L (densely dotted) and
from CTEQ6L (sparsely dotted), and, our set 1 (solid), set 2a (double dashed) and set 2b (short dashed).
Table 1
The 2 /N of the fit of the computed LO F2 to the H1 data [4] in the small-Q2 and large-Q2 regions. Also the
number of data points in each case is mentioned
Q2 < 4.0 GeV2
N = 29
CTEQ5L
CTEQ6L
MRST2001
This work:
Set 1: Qc < 1.4 GeV
Set 2a: Qc = 1.3
GeV
Set 2b: Qc = 1.4 GeV
Q2 > 4.0 GeV2

N = 104
all Q2
N = 133
31.8
2.72
0.59
1.18
0.93
2.06
7.86
1.32
1.74
1.75
1.58
0.95
0.96
1.05
0.86
1.13
1.17
0.88
217
Fig. 3. As Fig. 1 but for LO DGLAP result from CTEQ6L (dot-dashed) and for the DGLAP + GLRMQ results
with our set 1 (solid), set 2a (double dashed) and set 2b (short dashed).
The H1 data [4] is partly taken into account in the latest set CTEQ6L [3], and the
situation is clearly improved from CTEQ5L. This is shown in Figs. 3 and 2 by the dotdashed curves. In Fig. 4 we have plotted the PDFs at Q2 = 1.4 GeV2 for several cases.
Now, as shown by Fig. 4, the gluon distributions (upper left panel) of CTEQ6L (dotdashed) at small x are smaller than those of CTEQ5L (dashed), causing more modest
log Q2 slopes for F2 at small x. Fig. 2 indicates that the agreement of CTEQ6L with
the H1 data is excellent at x 103 and stays very good also at smaller x. Table 1 again
expresses 2 /N at the large-Q2 region (included in the CTEQ6 global fit) and in the smallQ2 region (not included in the CTEQ6 fit). Although a good agreement with the data is
218
Fig. 4. The parton distribution functions at Q2 = 1.4 GeV2 as obtained in the DGLAP analyzes CTEQ5L [6]
(dashed), CTEQ6L [3] (dot-dashed) and in the present work based on the DGLAP + GLRMQ evolution. In our
set 1 (solid), there is a finite charm contribution, while in the sets 2 (double dashed) charm is zero at this scale.
Notice the enhancement from the CTEQ6L glue. The gluon distributions of set 1 and sets 2 are identical.
found, a closer look at Fig. 3 shows that again the agreement of the pure DGLAP result is
getting worse towards the smallest values of x.
219
We notice that, contrary to CTEQ5L, the CTEQ6L result in Fig. 3 lies above the H1 data
at the smallest values of x. If we now simply took the initial conditions from CTEQ6L
at, say, Q2 = 5 GeV2 , and evolved the distributions downwards by using the nonlinear
equations (which would slow down the evolution), we would make the agreement with the
data in the small-x region worse than the CTEQ6L result. The dotted curve in Fig. 2 shows
the 2 /N for such a run, notice the growing trend towards the smallest x. Encouraged,
however, by the observations with the CTEQ5L + GLRMQ above, we wish to see whether
we could find initial conditions at some Q20 that would lead to at least the same or possibly
even better agreement with the H1 data as the DGLAP result from CTEQ6L.
3.2. New initial distributions
3.2.1. Set 1
We construct the initial distributions at an initial scale Q20 = 1.4 GeV2 by using the
(4)
CTEQ5L and CTEQ6L sets as our guide. Following CTEQ5L we use QCD = 192 MeV
together with the one-loop expression of the strong coupling constant. The change of the
value of QCD at the heavy-quark mass thresholds is taken into account. Throughout the
study we use Qb = 4.75 GeV for the b-threshold, and the c-threshold will be discussed
below. As seen in Fig. 2 (sparsely dotted curve), a good agreement with the H1 data
is found at x 0.01 by taking the CTEQ6L PDFs at 5.0 GeV2 and evolving them
down to 1.4 GeV2 according to the nonlinear equations. The CTEQ5L distributions
evolved down to 1.4 GeV2 from 10.0 GeV2 and 3.0 GeV2 with the GLRMQ corrections
included, give a reasonably good agreement with the H1 data at 104 x 0.01 and at
105 x 104 , correspondingly (not shown). A working initial condition can then be
found by interpolating between these three results. Finally, we make a power-law fit to the
interpolated gluon and light sea-quark distributions at small values of x. The x-slope of the
small-x gluons is tuned to reproduce the measured log Q2 slopes of F2 . This leads to an
initial gluon distribution xg(x, Q20 ) = 3.64 (0.01/x)0.28 at small x.
The initial conditions constructed in this way at Q20 = 1.4 GeV2 are shown in Fig. 4,
labelled as our set 1. The obtained distributions are compared with the CTEQ5L and
CTEQ6L PDFs at the same scale. Notice especially that our gluon distribution at x = 105
at this scale is a factor 6.4 larger than the one in CTEQ6L. In turn, the light sea-quark
distributions at small x lie in between CTEQ6L and CTEQ5L. Due to the procedure
we have chosen, the valence-quark distributions (which evolve according to DGLAP) are
practically the same as in CTEQ5L at x 0.01 and CTEQ6L at x 0.01.
The DGLAP + GLRMQ evolution to higher scales then gives the result shown in Fig. 3
by the solid lines. The corresponding values of 2 /N are again shown as a function of x
in Fig. 2 and its division into small-Q2 and large-Q2 regions in Table 1. We observe that
while the agreement with the H1 data at large values of x and Q2 is maintained practically
as good as in CTEQ6L, the fit in the smallest-x, small-Q2 region is indeed improving.
This, together with the parton distributions at Q20 = 1.4 GeV2 for the nonlinear evolution,
is the main result of this paper. At this point we should also emphasize that we have not
attempted to make a global statistical analysis such as the one by CTEQ in order to further
minimize the 2 . After such a procedure, further improvement on the 2 could well be
anticipated.
220
Fig. 5. The Q2 -dependence of the gluon distribution function at fixed values of x, from CTEQ5L [6] (dashed),
CTEQ6L [3] (dot-dashed) and set 1 of the present work (solid). Notice the logarithmic scales and the absolute
normalization of the curves.
Fig. 5 shows the Q2 -dependence of the gluon distributions at fixed values of x as

obtained in the DGLAP analyzes CTEQ5L (dashed lines) and CTEQ6L (dot-dashed) as
well as from the GLRMQ + DGLAP evolution of the gluons of our set 1 (solid). Notice
on one hand the large difference in the CTEQ5L and CTEQ6L sets, and on the other hand
the slower evolution in the nonlinear case at small values of x and Q2 . We also draw
attention to the fact that in spite of the large (factor 6.4) difference at Q20 = 1.4 GeV2
at x = 105 , our gluon distributions and the CTEQ6L gluons are in fact quite similar at
Q2 5 GeV2 . Coming back to the discussion related to Fig. 1, we also observe from
Fig. 5 that at Q2 = 5 GeV2 the gluons from our set 1 are below the CTEQ5L gluons as
anticipated.
221
3.2.2. Sets 2a and 2b

A detail to study next is the c-quark contribution. The initial distributions in our set
1 were obtained on the basis of distributions which were evolved downwards from 3,
5, and 10 GeV2 according to the nonlinear evolution equations. The c-quarks we treat
as massless quarks which experience the GLRMQ corrections as well. Therefore, the
downwards evolution of the c-quarks is also slower than in the DGLAP case, and their
distributions vanish only somewhat below the threshold scale Qc = 1.3 GeV of the sets
CTEQ6L and CTEQ5L. This explains the fairly large c-distribution at Q20 in our set 1.
We construct two slightly different sets of PDFs with the same initial conditions at
Q20 = 1.4 GeV2 , where xc(x, Q20) = 0. For the first case, called set 2a we follow
CTEQ6L and take Qc = 1.3 GeV
as the c-threshold scale. For the other case, called
set 2b we choose, Qc = Q0 = 1.4 GeV. In these initial conditions, the gluons are not
modified from set 1. However, in order to compensate for the loss of the initial c-quarks
and to recover the agreement with the H1 data, we enhance the other sea-quarks in the
small-x region simply by tuning their power in the power-law fit. The initial distributions
for our sets 2 are shown in Fig. 4 (double dashed). After the nonlinear evolution to higher
scales, the agreement of the set 2a with the H1 data is practically as good as with our set
1 and CTEQ6L, as is shown in Figs. 3 and 2 (double dashed lines). With the set 2b, the
agreement becomes even better, as can be seen from Figs. 2 and 3 (short dashed lines). The
same conclusion is suggested also by the 2 /N computed in the different regions of Q2 in
Table 1. The GLRMQ corrections thus improve the fits to the data in the region of small x
and Q2 without loosing the good fits at larger x and Q2 . Regarding the sensitivity of the
results to the c-threshold, we note that the differences between our sets 2a and 2b are quite
small and could most probably be obtained also by keeping the c-threshold constant while
tuning the gluon distributions at 0.001 x 0.01. This fine-tuning is, however, beyond
the goal of this paper.
4. Discussion and conclusions

It is important to keep in mind the uncertainties and limitations of the present approach.
A constraint not addressed above but used in the global analysis of the PDFs, is momentum
conservation. Due to the nonlinear terms in Eqs. (1) and (3), some momentum from small
values of x is lost in the evolution towards higher scales. Due to this, our sets 1 and 2
overestimate the total momentum at Q20 = 1.4 GeV2 by less than 2%. By Q2 = 100 GeV2 ,
some 2.6% of the initial total momentum is lost. As the emphasis of our study is in the
small-x behaviour of the PDFs, and as the violation remains small, we have not made an
attempt to correct the obtained distributions for the momentum sum rule.
In modelling the 2-gluon density in Eq. (1), we have set the effective radius parameter of
the free proton to R = 1 fm. Depending on the transverse matter density profile assumed
for the free proton, some 20% uncertainty in R can be expected. The nonlinearities
decrease with increasing R, so due to the interplay between the initial conditions and the
scale evolution demonstrated above, a larger R would lead to a smaller enhancement of the
small-x gluons. In order to properly estimate the uncertainty in the initial gluon distribution
caused by the uncertainty in R, the fit analysis performed above should be redone with
222
modified values of R. While this is beyond the scope of the present paper, it is left as a
future task.
The form (xg)2 for the 2-gluon density also neglects possible longitudinal correlations which should suppress the 2-gluon density at sufficiently large values of x. Also the
possible difference of the fractional momenta of the fusing gluons is neglected. It can be
argued, based on a simple picture of a Lorentz contracted proton and wavelengths of partons given by their inverse momentum, that gluons with x < 1/(2mp R) 0.1 overlap with
any other gluons and thus cause finite nonlinear corrections. As the 1-gluon densities are
already decreasing quite rapidly at x 0.1, and as their log Q2 slopes become anyway
small there, we have not attempted to build in any longitudinal correlations of the fusing
gluons. In a larger system, in a big nucleus, this would be necessary, and sensitivity to such
details should be studied.
Another obvious improvement for the present analysis is to consider the GLRMQ terms
added on top of the NLO DGLAP evolution. In that case, the zero (CTEQ6M) or negative
(MRST2001) gluon distributions at small x may become larger than zero.
In the present study we have neglected a possible but small contribution from the higherdimensional gluon distribution GHT introduced in [8]. The distribution GHT can be thought
as a kT2 moment of kT -dependent gluon distribution, and its upper limit estimated as
GHT (x, Q2 ) kT2 g(x, Q2 ) < Q2 g(x, Q2 ). Therefore, this term should be less than the
normal DGLAP term, and remain negligible. Further studies on this interesting question
will follow [13].
Regarding the size of the nonlinear terms, it is quite interesting to notice that in our
results at Q20 = 1.4 GeV2 , the GLRMQ terms in Eq. (1) make about 48% of the full
DGLAP + GLRMQ log Q2 slope of the gluon distribution at x = 105 , and still some
16% at x = 0.01. The extent of nonlinearity decreases, as expected, with increasing Q2 :
at Q2 = 10 GeV2 the GLRMQ contribution to the total xg/ log Q2 is 26% at x = 105
and below 4% at x = 0.01. At the lowest values of Q2 and x probed by the H1 data [4],
we clearly are at the borderline of the applicability of the approach, i.e., close to the
gluon saturation region, where the next terms in the nonlinear evolution equations are
becoming important [1518]. The further nonlinear correction terms obviously enter with
an alternating sign. In the region where the GLRMQ term in Eqs. (1) and (3) becomes as
important as the DGLAP term, inclusion of the further corrections should thus decrease
the net correction. Therefore, the results of the current paper in the region of smallest x
and Q2 studied can be regarded as an upper limit of the small-x gluon distributions.
In conclusion, we have studied the effects of adding the nonlinear GLRMQ corrections
to the LO DGLAP evolution equations, and especially the interplay between the initial
conditions and the nonlinearities. We use the PDF sets CTEQ5L and CTEQ6L as a
baseline, and the recent DIS data from H1 [4] as a constraint. We have shown that
the agreement between the measured and computed structure function F2 (x, Q2 ) can be
improved at small values of x and Q2 while still maintaining the good fit to the data
obtained in the global analyzes at larger values of x and Q2 . The nonlinearities slow
down the scale evolution, so in order to recover the measured log Q2 slopes of F2 (x, Q2 )
of the data, a larger small-x gluon distribution than that in CTEQ6L is needed. For the
gluon distribution the nonlinear effects are found to play an increasingly important role
at x 103 and Q2 10 GeV2 . The nonlinearities, however, vanish rapidly at larger
223
values of x and Q2 . Consequently, contrary to CTEQ6L, the obtained gluon distribution at

Q2 = 1.4 GeV2 shows a power-like growth at small values of x. Relative to the CTEQ6L
gluons, an enhancement up to a factor 6 at x = 105 , Q2 = 1.4 GeV2 reduces to a
negligible difference at Q2 10 GeV2 .
Acknowledgements
We thank N. Armesto, P.V. Ruuskanen, I. Vitev and other participants of the CERN
Hard Probes workshop for discussions. We are grateful for the Academy of Finland,
Project 50338, for financial support. J.W.Q. is supported in part by the United States
Department of Energy under Grant No. DE-FG02-87ER40371. C.A.S. is supported by
a Marie Curie Fellowship of the European Community programme TMR (Training and
Mobility of Researchers), under the contract number HPMF-CT-2000-01025.
References
[1] Yu. Dokshitzer, Sov. Phys. JETP 46 (1977) 1649;
V.N. Gribov, L.N. Lipatov, Sov. Nucl. Phys. 15 (1972) 438;
V.N. Gribov, L.N. Lipatov, Sov. Nucl. Phys. 15 (1972) 675;
G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[2] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 23 (2002) 73, hep-ph/0110215;
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 531 (2002) 216, hep-ph/0201127.
[3] J. Pumplin, et al., hep-ph/0201195.
[4] H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 21 (2001) 33, hep-ex/0012053.
[5] S.J. Brodsky, P. Hoyer, N. Marchal, S. Peigne, F. Sannino, Phys. Rev. D 65 (2002) 114025, hep-ph/0104291.
[6] CTEQ Collaboration, H.L. Lai, et al., Eur. Phys. J. C 12 (2000) 375, hep-ph/9903282.
[7] L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 (1983) 1.
[8] A.H. Mueller, J. Qiu, Nucl. Phys. B 268 (1986) 427.
[9] J. Kwiecinski, A.D. Martin, W.J. Stirling, R.G. Roberts, Phys. Rev. D 42 (1990) 3645.
[10] A.H. Mueller, Nucl. Phys. B 335 (1990) 115;
A.H. Mueller, Nucl. Phys. B 558 (1999) 285;
L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233;
I.I. Balitsky, Nucl. Phys. B 463 (1996) 99;
Yu.V. Kovchegov, Phys. Rev. D 54 (1996) 5463;
J. Jalilian-Marian, A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 55 (1997) 5414;
Yu.V. Kovchegov, A.H. Mueller, Nucl. Phys. B 529 (1998) 451;
Yu.V. Kovchegov, A.H. Mueller, Nucl. Phys. D 55 (1997) 5445;
M.A. Braun, Eur. Phys. J. C 16 (2000) 337;
E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 692 (2001) 583;
A. Kovner, U.A. Wiedemann, Phys. Rev. D 64 (2001) 114002.
[11] K. Golec-Biernat, M. Wusthoff, Phys. Rev. D 59 (1999) 014017;
W. Buchmuller, T. Gehrmann, A. Hebecker, Nucl. Phys. B 537 (1999) 477;
E. Gotsman, E. Ferreira, E. Levin, U. Maor, E. Naftali, Phys. Lett. B 500 (2001) 87;
J. Bartels, K. Golec-Biernat, H. Kowalski, Phys. Rev. D 66 (2002) 014001;
E. Gotsman, E. Levin, M. Lublinsky, U. Maor, hep-ph/0209074.
[12] K. Prytz, Phys. Lett. B 311 (1993) 286.
224
[13] K.J. Eskola, et al., in preparation.

[14] H. Plothow-Besch, Comput. Phys. Commun. 75 (1993) 396;
H. Plothow-Besch, Int. J. Mod. Phys. A 10 (1995) 2901;
PDFLIB: Proton, Pion, Photon Parton Density Functions, Parton Density Functions of the Nucleus, and s ,
Userss Manual, Version 8.04, W5051 PDFLIB 2000.04.17 CERN-ETT/TT.
[15] A.L. Ayala, M.B. Gay Ducati, E.M. Levin, Nucl. Phys. B 493 (1997) 305, hep-ph/9604383.
[16] J. Jalilian-Marian, X.N. Wang, Phys. Rev. D 60 (1999) 054016, hep-ph/9902411.
[17] H. Weigert, Nucl. Phys. A 703 (2002) 823, hep-ph/0004044.
[18] Y.V. Kovchegov, K. Tuchin, Phys. Rev. D 65 (2002) 074026, hep-ph/0111362.

The impact of universal extra dimensions on the

unitarity triangle and rare K and B decays
Andrzej J. Buras a , Michael Spranger a,b , Andreas Weiler a
a Physik Department, Technische Universitt Mnchen, D-85748 Garching, Germany
b Max-Planck-Institut fr Physik, Werner-Heisenberg-Institut, D-80805 Mnchen, Germany
Received 27 January 2003; received in revised form 14 March 2003; accepted 19 March 2003
Abstract
We calculate the contributions of the KaluzaKlein (KK) modes to the KL KS mass difference
0 B
0 mixing mass differences Md,s and rare decays K +
MK , the parameter K , the Bd,s
d,s
+
0
+
KL ,
KL , B Xs,d and Bs,d in the Appelquist, Cheng
,
and Dobrescu (ACD) model with one universal extra dimension. For the compactification scale
1/R = 200 GeV the KK effects in these processes are governed by a 17% enhancement of the
F = 2 box diagram function S(xt , 1/R) and by a 37% enhancement of the Z 0 penguin diagram
function C(xt /1/R) relative to their Standard Model (SM) values. This implies the suppressions of
|Vt d | by 8%, of by 11% and of the angle in the unitarity triangle by 10 . Ms is increased
by 17%. MK is essentially uneffected. All branching ratios considered in this paper are increased
with a hierarchical structure of enhancements: K + + (16%), KL 0 (17%), B
SD (38%), B Xs (44%), Bd (46%) and Bs (72%).
Xd (22%), (KL )
For 1/R = 250 (300) GeV all these effects are decreased roughly by a factor of 1.5 (2.0). We
emphasize that the GIM mechanism assures the convergence of the sum over the KK modes in the
case of Z 0 penguin diagrams and we give the relevant Feynman rules for the five-dimensional ACD
model. We also emphasize that a consistent calculation of branching ratios has to take into account
the modifications in the values of the CKM parameters. As a byproduct we confirm the dominant
O(g2 GF m4t R 2 ) correction from the KK modes to the Z 0 bb vertex calculated recently in the large
mt limit.
E-mail addresses: aburas@ph.tum.de (A.J. Buras), msprange@ph.tum.de (M. Spranger),

aweiler@ph.tum.de (A. Weiler).
doi:10.1016/S0550-3213(03)00250-5
226
A.J. Buras et al. / Nuclear Physics B 660 (2003) 225268
1. Introduction
During the last years there has been an increased interest in models with extra
dimensions. Among them a special role play the ones with universal extra dimensions
(UED). In these models all the Standard Model (SM) fields are allowed to propagate in all
available dimensions. Above the compactification scale 1/R a given UED model becomes
a higher dimensional field theory whose equivalent description in four dimensions consists
of the SM fields, the towers of their KaluzaKlein (KK) partners and additional towers of
KK modes that do not correspond to any field in the SM. The simplest model of this type is
the Appelquist, Cheng and Dobrescu (ACD) model [1] with one extra universal dimension.
In this model the only additional free parameter relative to the SM is the compactification
scale 1/R. Thus all the masses of the KK particles and their interactions among themselves
and with the SM particles are described in terms of 1/R and the parameters of the SM. This
economy in new parameters should be contrasted with supersymmetric theories and models
with an extended Higgs sector.
A very important property of the ACD model is the conservation of KK parity that
implies the absence of tree level KK contributions to low energy processes taking place
at scales 1/R. In this context the flavour changing neutral current (FCNC) processes
like particleantiparticle mixing and rare K- and B-decays are of particular interest. As
these processes appearing in the SM first at one-loop are strongly suppressed, the one-loop
contributions from the KK modes to them could in principle be important.
The effects of the KK modes on various processes of interest have been investigated in a
number of papers. In [1] their impact on the precision electroweak observables assuming a
light Higgs (mH 250 GeV) led to the lower bound 1/R 300 GeV. Subsequent analyses
of the decay B Xs [2] and of the anomalous magnetic moment [3] have shown the
consistency of the ACD model with the data for 1/R 300 GeV. The scale of 1/R as low
as 300 GeV would lead to an exciting phenomenology in the next generation of colliders
[47]. Moreover the cosmic relic density of the lightest KK particle as a dark matter
candidate turned out to be of the right order of magnitude [8]. The related experimental
signatures have been investigated in Ref. [9].
Very recently Appelquist and Yee [10] have extended the analysis of [1] by considering
a heavy Higgs (mH 250 GeV). It turns out that in this case the lower bound on 1/R
can be decreased to 250 GeV, implying larger KK contributions to various low energy
processes, in particular to the FCNC processes. Among the latter only the decay B Xs
has been investigated within the ACD model so far [2] and it is desirable to consider other
FCNC processes.
0 B
0 mixing mass differences
In the present paper we calculate for the first time the Bd,s
d,s
Md,s , the KL KS mass difference, the CP violation parameter K and the branching
ratios for the rare decays K + + , KL 0 , KL + , B Xs,d and
Bs,d in the ACD model with one universal extra dimension. In the forthcoming
paper [11] we will analyze the decays B Xs , B Xs l + l and KL 0 e+ e . In
order to be more general we will include the results for 1/R = 200 GeV that is only slightly
below the lowest value of 1/R = 250 GeV allowed by the electroweak precision data.
227
As our analysis shows, the ACD model with one extra dimension has a number of
interesting properties from the point of view of FCNC processes discussed here. These
are:
GIM mechanism [12] that improves significantly the convergence of the sum over the
KK modes corresponding to the top quark, removing simultaneously to an excellent
accuracy the contributions of the KK modes corresponding to lighter quarks and
leptons. This feature removes the sensitivity of the calculated branching ratios to the
scale Ms 1/R at which the higher dimensional theory becomes non-perturbative
and at which the towers of the KK particles must be cut off in an appropriate way. This
should be contrasted with models with fermions localized on the brane, in which the
KK parity is not conserved and the sum over the KK modes diverges. In these models
the results are sensitive to Ms and the KK effects in Ms,d are significantly larger [13]
than found here.
The low energy effective Hamiltonians are governed by local operators already present
in the SM. As flavour violation and CP violation in this model is entirely governed
by the CKM matrix, the ACD model belongs to the class of the so-called models
with minimal flavour violation (MFV) as defined in [14]. This has automatically two
important consequences.
The impact of the KK modes on the processes discussed here amounts to the
modification of the InamiLim one-loop functions [15]. This is the function S [16]
in the case of Md,s and of the parameter K and the functions X and Y [17] in the
case of the rare decays considered. In the ACD model these three functions depend
only on mt and the single new parameter, the compactification radius R.
The unitarity triangle constructed from |Vub /Vcb |, Md /Ms and the sin 2 extracted
from the CP asymmetry aKS is common to the SM model and the ACD model. That
is, the R-dependence drops out in this construction. Which of these two models, if
any, is consistent with the data can only be found out by analyzing Md and Ms
separately, K and, in particular, the branching ratios for rare K- and B-decays that
depend explicitly on R.
Our paper is organized as follows. In Section 2, we summarize those ingredients of the
ACD model that are relevant for our analysis. In particular, we give in Appendix A the
set of the relevant Feynman rules in the ACD model that have not been given so far in the
literature. In Section 3, we calculate the KK contributions to the box diagram function S
and we discuss the implications of these contributions for MK , Md , Ms , K and the
unitarity triangle. In Section 4, we calculate the corresponding corrections to the functions
X and Y that receive the dominant contribution from Z 0 -penguins and we analyze the
implications of these corrections for the rare decays K + + , KL 0 , KL
In Section 5, we summarize our results and give a
+ , B Xs,d and Bs,d .
brief outlook.
Very recently an analysis of Md,s in the ACD model has been presented in [18]. In
the first version of this paper the result for the function S found by these authors differed
significantly from our result with the effect of the KK modes being by roughly a factor of
two larger than what we find. After the first appearence of our paper the authors of [18]
228
identified errors in their calculation and confirmed our result for S.

However, we disagree
Bd FBd by a factor of
with their claim that the reduction of the error on the parameter B
three will necessarily increase the lowest allowed value of the compactification scale 1/R
to 740 GeV. We will address this point at the end of Section 3.
2. The five-dimensional ACD model

The five-dimensional UED model introduced by Appelquist, Cheng and Dobrescu
(ACD) in [1] uses orbifold compactification to produce chiral fermions in 4 dimensions.
This is not the case in the models described in [19], where all fermions are localized on
the 4-dimensional brane. However, there are many similarities between these two classes
of models, and some of the issues discussed in this section have already been presented in
detail in [19]. We assume vanishing boundary kinetic terms at the cut off scale. We also
rely on [1,20,21].
2.1. KaluzaKlein mode expansion
The topology of the fifth dimension is the orbifold S 1 /Z2 , and the coordinate y x 5
runs from 0 to 2R, where R is the compactification radius. The orbifold has two fixed
points, y = 0 and y = R. The boundary conditions given at these fixed points determine
the KaluzaKlein (KK) mode expansion of the fields.
A scalar field has to be either even or odd under the transformation P5 : y y, and
therefore

5 + = 0, for even fields
(2.1)
at y = 0, R.
= 0,
for odd fields
These are von Neumann and Dirichlet boundary conditions, respectively, at the fixed
points. The associated KK expansions are
1
1 +
ny
+
(0)
+ (x, y) =
(x) +
(n) (x) cos ,
R
2R
R n=1
1
ny
(x, y) =
(n) (x) sin ,
R
R n=1
(2.2)
where x x , = 0, 1, 2, 3, denotes the four non-compact spacetime coordinates. The
fields (n)
(x) are called KaluzaKlein modes.
A vector field AM in 5 dimensions has five components, M = 0, 1, 2, 3, 5. The orbifold
compactification forces the first four components to be even under P5 , while the fifth
component is odd:

5 A = 0
(2.3)
at y = 0, R.
A5 = 0
229
Hence, the KK expansion of a vector field is

A (x, y) =
1
ny
A(n) (x) cos ,

A(0)(x) +
R
2R
R n=1
1
1 5
ny
A5 (x, y) =
A(n) (x) sin .
R
R n=1
(2.4)
A Dirac spinor in 5 dimensions is a four component object. Using the chirality

projectors PR/L = (1 5 )/2, a spinor = (PR + PL ) = R + L has to satisfy either

5 R+ = 0
5 L = 0
(2.5)
at y = 0, R or
at y = 0, R.
L+ = 0
R = 0
The respective KK mode expansions are

ny
ny
,
+ L(n) (x) sin
R(n) (x) cos
R(0) (x) +
(x, y) =
R
R
2R
R n=1

1
1
ny
ny
(x, y) =
L(0) (x) +
.
+ R(n) (x) sin
L(n) (x) cos
R
R
2R
R
1

n=1
(2.6)
The zero-mode is either right- or left-handed. The non-zero-modes come in chiral pairs.
This chirality structure is a natural consequence of the orbifold boundary conditions.
We can derive Feynman rules for the KK modes by explicitly integrating over the fifth
dimension in the action:
2R

S=
dy L5 =
d x
d 4 x L4 .
(2.7)
Using the KK mode expansions in (2.2), (2.4) and (2.6), the five-dimensional
Lagrangian L5 reduces to the four-dimensional Lagrangian L4 which contains all KK
modes. The field content is arranged such that the zero-modes are the 4-dimensional SM
particles, whereas the higher modes constitute their KK excitations. Moreover there are
additional KK modes that do not correspond to any field in the SM.
2.2. Universal extra dimensions
In the UED scenarios, all fields present in the Standard Model live in the extra
dimensions, i.e., they are functions of all spacetime coordinates.
For bosonic fields, one simply replaces all derivatives and fields in the SM Lagrangian
by their 5-dimensional counterparts. There are the U (1)Y gauge field B and the SU(2)L
gauge fields Aa , as well as the SU(3)C QCD gauge fields. The Higgs doublet is

i +
1 2 + i 1

=
(2.8)
=
,
1 i 3
2 i 3
2
230
where

1
= 1 i 2 .
(2.9)
2
It is chosen to be even under P5 so it possesses a zero-mode. By assigning a vacuum
expectation value to that zero-mode with the substitution v + H , we can give masses
to the fermions. Note that we label all parameters of the 5-dimensional Lagrangian with
a caret. It is convenient to introduce 4-dimensional
parameters that are related to their
5-dimensional counterparts by factors of 2R, see Appendix A.

Fermions living in five dimensions are more involved. In order to write down a Lorentz
invariant Lagrangian, we need one -matrix for each spacetime dimension to satisfy the
Clifford algebra
M N
= 2g MN , M, N = 0, 1, 2, 3, 5.
,
(2.10)
The metric g MN = diag(1, 1, 1, 1, 1) is the natural extension of the flat
Minkowskian metric for one extra space dimension. For the -matrices, we take
= ,
= 0, 1, 2, 3,
= i5 ,
5
(2.11)
with 5 = i 0 1 2 3 . The 5-dimensional kinetic Lagrangian for spinor fields can now be
written as [20]
M M = (i/
5 5 ).
L5 = i
(2.12)
Integrating over the fifth dimension yields

2R

L4 =
dy L5 =
0

n=0

n
,
i/
(n)
(n)
R
(2.13)
where the sign of the KK mass term n/R depends on the choice in (2.5).
Note that the zero-mode remains massless unless we apply the Higgs mechanism. Note
also that all fields in the 4-dimensional Lagrangian receive the KK mass n/R on account of
the derivative operator 5 acting on them. These tree-level masses are shifted by radiative
corrections due to gauge interactions and boundary terms localized at the fixed points [22].
Since these corrections are a two-loop effect on the processes considered, we will use the
tree-level mass relations in our calculations.
2.3. Gauge fixing and Goldstone mixing
In the 5-dimensional ACD model, we can use the same gauge fixing procedure as in
models in which the fermions are localized on the 4-dimensional subspace. We adopt the
various gauge fixing functionals given in [19] and adapt them to the case of the electroweak
gauge group U (1)Y SU(2)L with one Higgs doublet:

g
1
v
3 + 5 B5 ,
GB B , B5 , 3 = B
2

1
g2 a
+ 5 Aa5 ,
GAa Aa , Aa5 , a = Aa v
2
231
(2.14)
where g and g2 are the respective 5-dimensional U (1)Y and SU(2)L gauge coupling
constants. This is a natural extension of the 4-dimensional R -gauge fixing.
With the gauge fixed, we can diagonalize the kinetic terms of the bosons and finally
derive expressions for the propagators. Compared to the SM, there are the additional KK
mass terms. As they are common to all fields, their contribution to the gauge boson mass
matrix is proportional to the unity matrix. As a consequence, the electroweak mixing angle
w is the same for all KK modes, and we have
M
3M
Z
cos w sin w
A
(2.15)
=
,
AM
sin w
cos w
BM

1
W M = A1M iA2M ,
(2.16)
2
where
sw sin w =
g
and cw cos w =
g22 + g 2
g2
g22 + g 2
(2.17)
For the -components, (2.15) and (2.16) already give the mass eigenstates
Z(n) = cw A(n) sw B(n) ,

1 1
2
W(n) = A(n) iA(n) .
2
A(n) = sw A(n) + cw B(n) ,

(2.18)
The zero-modes Z(0) and W(0) have the masses

v
v
MZ =
g22 + g 2 and MW = g2 .
2
2
(2.19)
a , while A
The components Z5(n) and W5(n)
mix with the Higgs modes (n)
5(n) as defined
in (2.15) is a mass eigenstate.
Because of the KK contribution to the mass matrix, the Higgs components and 3
with n = 0 no longer play the role of Goldstone bosons. Instead, they mix with W5 and
Z5 to form, in addition to the Goldstone modes G0(n) and G
(n) , three additional physical
0
scalar modes a(n) and a(n) . The former allow the gauge bosons to acquire masses without
breaking gauge invariance. The propagator terms are diagonalized by the orthogonal
transformations

n 3
1
n
1
3
0
(n) + MZ Z5(n) ,
MZ (n)
a(n)
Z5(n) ,
=
G0(n) =
MZ(n)
R
MZ(n) R
(2.20)

1
n
1
n
G(n) =
+ MW W5(n)
MW (n) W5(n) ,
a(n) =
MW (n)
R
MW (n) R (n)
(2.21)
232
with MZ(n) and MW (n) given in (A.2).

3 and as the Goldstone bosons that give
For the zero-modes, we can identify (0)
(0)
masses to Z(0) and W(0) . With increasing n, the contributions of Z5(n) and W5(n)
dominate
3
0
the Goldstone modes, while (n) and (n) provide the main fraction of a(n) and a(n)
.
The physical Higgs H(n) does not mix with A5(n) . The latter constitutes an additional
unphysical scalar mode which turns out to be the Goldstone mode for A(n) for n 1.
Note that the fields a 0 , a and A5 do not have zero modes. Consequently, the photon
A(0) remains massless.
2.4. FermionHiggs coupling

The Yukawa coupling of the Higgs doublet to the quark fields is a pivotal part of the
Lagrangian concerning chirality. Analogous to the SM, we write
D D + Q
U U i 2 + h.c.,
LqH (x, y) = Q
(2.22)
with the three generation gauge eigenstates

T
Qc
Qt
,
,
,
Qs
Qb

Dd
Uu
D = Ds .
U = Uc ,
Ut
Db
Q =

Qu
Qd
(2.23)
(2.24)
The SU(2) doublets Q are odd under P5 , while the singlets U and D are even. Due to
this assignment, the zero-modes have the same chirality as the quark fields in the SM.
The fermions receive masses both through spontaneous symmetry breaking and the
KK expansion as described in Section 2.2. In order to diagonalize the Yukawa couplings
between fermions of equal charge, we apply the biunitary transformation
QU = SU QU ,
QD = SD QD ,
(2.25)
U = TU U ,
D = TD D ,
(2.26)

so that the resulting mass matrices

v
MU = SU U TU ,
2
v
MD = SD
D TD
2
(2.27)
are diagonal and their eigenvalues (mu , mc , mt ) and (md , ms , mb ) are non-negative. This
step is analogous to the SM and leads to the CKM matrix VCKM = SU SD which is
unique for all KK levels. Next we apply to each flavour f = u, c, t, d, s, b the unitary
transformation

Uf (n)
5 cos f (n) sin f (n)
Uf (n)
(2.28)
=
5 sin f (n)
cos f (n)
Qf (n)
Qf (n)
233
and the same expression with U replaced by D for the down-type quarks. The fields Qf (n) ,
Uf (n) and Df (n) are the mass eigenstates in 4 dimensions. The mixing angle is
mf
, for n 1.
tan 2f (n) =
(2.29)
n/R
In (2.28) we have used 5 to get both eigenvalues positive and equal to

n2
mf (n) =
(2.30)
+ m2f .
R2
In phenomenological applications we have n/R 200 GeV and can therefore set all
mixing angles to zero except for the top quark.
The Yukawa coupling of the leptons to the Higgs doublet is similar to that of the quarks.
The lepton doublets take the form
T
L
Le
L

,
,
,
L =
(2.31)

Le
L
L
and the singlets are

Ee

E = E .
(2.32)
E
Due to their smallness, we can ignore neutrino masses in heavy flavour physics. Hence,
the neutrino singlets do not couple to the other particles and are therefore omitted. Due
to their small Yukawa masses, we can set the mixing angles for all KK excitations of the
leptons to zero. The lepton masses are given by Eq. (2.30).
The Feynman rules can be found in Appendix A.
0 B
0 mixing and K
3. Bd,s
d,s
0 B
0 mixing
3.1. Bd,s
d,s
The effective Hamiltonian for B = 2 transitions in the SM [23] can be generalized to

the ACD model as follows
B=2
=
Heff

2
G2F
M 2 V Vt q B S(xt , 1/R)
16 2 W t b

6/23
(5)
s(5) (b )
J5 Q(B = 2) + h.c.
s (b )
1+
4
(3.1)
Here b = O(mb ), J5 = 1.627,

V A , q = d, s
V A (bq)
Q(B = 2) = (bq)
V A b
(1 5 )q and [23,24]
with (bq)
B = 0.55 0.01
(3.2)
(3.3)
234
describes the short distance QCD corrections to which we will return below.
The function S(xt , 1/R) is given as follows
S(xt , 1/R) = S0 (xt ) +
Sn (xt , xn ),
(3.4)
n=1
where (mt m
t (mt ))
xt =
m2t
,
2
MW
m2n
,
2
MW
n
R
(3.5)
4xt 11xt2 + xt3

3xt3 ln xt
2
4(1 xt )
2(1 xt )3
(3.6)
xn =
mn =
and
S0 (xt ) =
results from the usual box diagrams with (W , t) and (G , t) exchanges with the mt
independent terms removed by the GIM mechanism.
The KK contributions are represented by the functions Sn (xt , xn ) that are obtained by
calculating the box diagrams in Fig. 1 with W(n)

, a(n)
, G
(n) , Qi(n) and Ui(n) (i = u, c, t)
exchanges and multiplying the result by i/4, where 1/4 is a combinatorial factor. We
neglect momenta and masses of external quarks. Denoting the contribution of the sum of
the diagrams corresponding to a given pair (mi(n) , mj (n) ) to Sn (xt , xn ) by F (xi(n) , xj (n) )
with
xi(n) =
m2i(n)
2
MW
(n)
(3.7)
and mi(n) and MW (n) defined in (2.30) and (A.2), and using the unitarity of the CKM
matrix, we have
Sn (xt , xn ) F (xt (n) , xt (n) ) + F (xu(n) , xu(n) ) 2F (xt (n) , xu(n) ).
(3.8)
As with increasing n the modes Qt (n) , Ut (n) , Qu(n) and Uu(n) become increasingly
degenerate in mass and
xt (n) xu(n) 1,
(3.9)
the functions Sn (xt , xn ) decrease with increasing n so that only a few terms in the sum in
(3.4) are relevant. This is seen in Fig. 2 where we show Sn (xt , xn ) as a function of n/R.
We will discuss this in more detail in Section 4.9.
Fig. 1. Box diagrams contributing to Sn (xt , xn ). We suppress the KK mode number.
(a)
235
(b)
Fig. 2. (a) Contribution Sn of the nth KK mode to S(xt , 1/R). The contributions with a dominate, those with
only G and W are negligible and not shown. (b) The functions S(xt , 1/R) and S0 (xt ).
The contributions from different sets of diagrams to the functions F (xi(n) , xj (n) ) are
, a(n)
) is by
collected in Appendix B. It turns out that the contribution from the pair (a(n)
far dominant. We illustrate this in Fig. 2(a). In phenomenological applications it is more
useful to work with the variables xt and xn than with xi(n) . We find
Sn (xt , xn ) =
1
4(xt 1)3 xt

6xn xt 5xt2 12xn xt2 + 15xt3 + 10xn xt3 11xt4 4xn xt4 + xt5

xn
+ 6xn2 + 2xn xt
1 + xn

2
2
3
3
+ 12xn xt 6xn xt 2xt + 14xn xt 2xn2xt3 + 6xt4 2xn xt4

xt + xn
ln
(3.10)
.
1 + xn
2xn (xt 1)3 (3xn + 3xn xt xt ) ln
In Fig. 2(b) we plot S(xt , 1/R) vs. 1/R. For 1/R = 200 GeV we observe a 17%
enhancement of the function S with respect to its SM value given by S0 (xt ). For 1/R =
250 GeV this enhancement decreases to 11% and it is only 4% for 1/R = 400 GeV.
Proceeding as in the SM, we can calculate the mass differences Mq by means of

2
G2F
Bq FB2 MW
B mBq B
S(xt , 1/R)|Vt q |2 ,
(3.11)
q
6 2
q the renormalization group invariant
where FBq is the Bq -meson decay constant and B
parameter related to the hadronic matrix element of the operator Q(B = 2), see [25] for
details. This implies two useful formulae

Bd FBd 2 |V | 2 S(x , 1/R)
B
td
B
t
Md = 0.50/ps
(3.12)
230 MeV
0.55
2.34
7.8 103
Mq =
236
and

Bs FBs 2 |V | 2 S(x , 1/R)
B
ts
B
t
Ms = 18.4/ps
.
270 MeV
0.040
0.55
2.34
(3.13)
The implications of these results for |Vt d |, Ms and the unitarity triangle will be discussed
below.
Finally, a few comments regarding the QCD factor B are in order. This factor as
given in (3.3) has been calculated within the SM including NLO QCD corrections that are
mandatory for the proper matching of the Wilson coefficient of the operator Q(B = 2)
Bs,d and calculated by
with its hadronic matrix element represented by the parameter B
means of non-perturbative methods. As the top quark and the KK modes are integrated out
at a single scale t = O(mt , 1/R), the contributions to B from scales lower than t are the
same for the SM and the ACD model. They simply describe the finite renormalization of
Q(B = 2) from scales O(t ) down to the scales O(mb ). The difference between QCD
corrections to the SM contributions and to the KK contributions arises only in the full
theory at scales t = O(mt , 1/R) as the unknown QCD corrections to the box diagrams in
Fig. 1 can, in principle, differ from the known QCD corrections to the SM box diagrams
[23,24] that have been included in B . As the QCD coupling constant s (t ) is small and
the QCD corrections to the SM box diagrams of order of a few percent, we do not expect
that the difference between the QCD corrections to the diagrams in Fig. 1 and to the SM
box diagrams is relevant, in particular, in view of the fact that the KK contributions amount
to at most 17% of the full result.
For t , s () in the ACD model becomes larger [26] and the QCD corrections
to KK modes with n 1 could in principle be substantial. However, as seen in Fig. 2(a),
these heavy modes give only a tiny contribution to S(xt , 1/R) and can be safely neglected.
3.2. K
The effective Hamiltonian for S = 2 transitions is given in the ACD model as follows
S=2
=
Heff
2
G2F
2
MW
c 1 S0 (xc ) + 2t 2 S(xt , 1/R) + 2c t 3 S0 (xc , xt )
2
16

(3)
(3) 2/9
s ()
J3 Q(S = 2) + h.c.,
s ()
1+
4
(3.14)
where i = Vis Vid , s(3) is the strong coupling constant in an effective three flavour theory
and J3 = 1.895 in the NDR scheme [23]. In (3.14), the relevant operator
Q(S = 2) = (s d)V A (s d)V A ,
(3.15)
is multiplied by the corresponding Wilson coefficient function. This function is decomposed into a charm-, a top- and a mixed charm-top contribution. The SM function S0 (xc , xt )
is defined by
S0 (xc , xt ) = F (xc , xt ) + F (xu , xu ) F (xc , xu ) F (xt , xu ),
(3.16)
237
where F (xi , xj ) is the true function corresponding to the box diagrams with (i, j )
exchanges. One has

3xt2 ln xt
3xt
xt
,
S0 (xc , xt ) = xc ln
(3.17)
xc 4(1 xt ) 4(1 xt )2
where we keep only linear terms in xc 1, but of course all orders in xt .
In view of the comments made after (3.8) and the structure of (3.16), the impact of
the KK modes on the charm- and mixed charm-top contributions is totally negligible and
we take into account these modes only in the top contribution that is described by the
same function S(xt , 1/R) as in the case of Mq . This also means that the KL KS mass
difference, MK , being dominated by internal charm contributions in (3.14) is practically
uneffected by the KK modes.
Short distance QCD effects are described through the correction factors 1 , 2 , 3 and
the explicitly s -dependent terms in (3.14). The NLO values of i are given as follows
[23,27,28]:
1 = 1.45 0.38,
2 = 0.57 0.01,
3 = 0.47 0.04.
(3.18)
The standard procedure allows now to calculate the CP-violating parameter K [25]

K Im t Re c 1 S0 (xc ) 3 S0 (xc , xt ) Re t 2 S(xt , 1/R)
K = C B
exp(i/4),
(3.19)

is a numerical constant. BK is the renormalization group invariant
where C
parameter related to the hadronic matrix element of the operator Q(S = 2), see [25] for
details.
= 3.837 104
3.3. Unitarity triangle in the ACD model

What is the impact of the KK contributions to the function S on the elements of the
CKM matrix and, in particular, on the shape of the unitarity triangle? In order to answer
this question let us recall a few aspects of the unitarity triangle (UT) shown in Fig. 3 and
of the Wolfenstein parametrization [29] as generalized to higher orders in in [30]. The
apex of the unitarity triangle is given by [30]

2
2
,
= 1
.
C = C 1
(3.20)
2
2
Here , A, C and are the Wolfenstein parameters [29]. Moreover, one has

Vus = + O 7 ,
(3.21)
Vub = A3 (C i),
Vcb = A2 + O 8 ,

1
Vt s = A2 + A4 1 2(C + i) ,
(3.22)
Vt d = A3 (1 C i ).
2
The lengths Rb and Rt are given by

|
|Vud Vub
2 1 Vub
2
2
Rb
(3.23)
+ = 1
,
| = C
|Vcd Vcb
2 Vcb

|V V |
1 Vt d
2 + 2 =
Rt t d tb = (1 C)
(3.24)
|Vcd Vcb |
Vcb
238
Fig. 3. Unitarity triangle.
and the angles and of the UT are related directly to the complex phases of the CKM
elements Vt d and Vub , respectively, through
Vt d = |Vt d |ei ,
Vub = |Vub |ei .
(3.25)
The five constraints on the UT that we have at our disposal at present are:
The Rb constraint: as seen in (3.23), the length of the side AC is determined from
|Vub /Vcb |.
K -hyperbola (indirect CP violation in KL ) obtained from (3.19) and the
experimental value for K ( = 0.221):

K = 0.214,
(1 C)A
2 2 S(xt , 1/R) + Pc () A2 B
(3.26)
where Pc () = 0.28 0.05 [27,28] represents the charm contribution that is not
affected by the KK contributions.
0 -mixing constraint (Md ):
Bd0 B
d

|Vt d |
1 |Vt d |
0.041
= 0.86
Rt =
(3.27)
|Vcb |
7.8 103
|Vcb |
with
|Vt d | = 7.8 10
230 MeV

Bd FBd
B

Md
0.50/ps
where Md = (0.503 0.006)/ps [31].

s0 -mixing constraint (Md /Ms ):
Bs0 B

Rt = 0.87
Md
0.50/ps

18.4/ps
,
Ms 1.18

0.55
B

=
2.34
,
S(xt , 1/R)
s FBs
B
d FBd
B
(3.28)
(3.29)
where Ms > 14.4/ps at 95% C.L. [31].

The direct measurement of sin 2 through the CP asymmetry a(KS ) in Bd KS
that is not affected by the KK contributions.
239
The main uncertainties in this analysis

originate in the theoretical uncertainties in the non
d FBd and to a lesser extent in [32]:
K and B
perturbative parameters B

d FBd = 235+33 MeV,
BK = 0.86 0.15,
(3.30)
= 1.18+0.13
B
41
0.04 .

s FBs = (276 38) MeV are substantial. The
Also the uncertainties in |Vub /Vcb | and B
QCD sum rules results for the parameters in question are similar and can be found in [33].
With these formulae at hand let us enumerate a few general properties of the values of
the CKM elements within the ACD model. These are:
|Vus |, |Vcb | and |Vub | are usually determined from tree level decays. As in the ACD
model there are no KK contributions at the tree level, the absolute values of these three
CKM elements are to an excellent approximation the same as in the SM model. From
the point of view of the unitarity triangle (UT), this means that the lengths of its two
sides, AC and CB are common to the SM model and the ACD model. In our numerical
analysis we will use, as in [34],
|Vcb | = (40.6 0.8) 103 ,
|Vus | = = 0.221 0.002,
(3.31)
|Vub |
(3.32)
= 0.089 0.008,
|Vub | = (3.63 0.32) 103 .
|Vcb |
The angle of the UT has been determined recently by the BaBar [35] and Belle [36]
collaborations from the CP asymmetry aKS in B KS with a high accuracy giving
the world grand average [37]
(sin 2)KS = 0.734 0.054 .
(3.33)
As there are no new complex phases in the ACD model beyond the KM phase, the
angle as extracted by means of aKS is common to both models in question.
A similar comment applies to |Vt d | or equivalently the length Rt of the side AB
in Fig. 3, when Rt is extracted from the ratio Md /Ms that is independent of
S(xt , 1/R) as seen in (3.29).
Thus when |Vub /Vcb |, sin 2 from aKS and Md /Ms are used to construct the UT,
there is no difference between the SM and the ACD model as all explicit dependence on
1/R cancels out. This universal UT (UUT) [14] that is valid for all MVF models, as defined
in [14], has recently been determined [34,38].
Now, even though there exists a UUT common to the SM and the ACD model, in view
of the fact that S(xt , 1/R) = S0 (xt ), only one of these two models, if any, will have K ,
Md and Ms that agree with the experimental data. Let us consider Ms first. As seen
in (3.22) |Vt s | is very close to |Vcb | because of CKM unitarity. Therefore, it is common
with an excellent accuracy to both models and consequently
(Ms )ACD S(xt , 1/R)
> 1.
=
(3.34)
(Ms )SM
S0 (xt )
However, this ratio is at most 1.17 and the distinction between these two models will only
Bs F 2 can be calculated to better than 10% accuracy. A very difficult
be possible provided B
Bs
task.
240
The fact that S(xt , 1/R) > S0 (xt ) implies also that with the experimentally known
values of K and Md , one has
|Vt d |ACD < |Vt d |SM ,
ACD < SM
(3.35)
as can easily be inferred from (3.26) and (3.28). In particular, (3.28) implies

|Vt d |ACD
S0 (xt )
.
=
|Vt d |SM
S(xt , 1/R)
(3.36)
Thus |Vt d |ACD can be smaller than |Vt d |SM by at most 8%. In order to determine such
a difference, more accurate information on the unitarity triangle and the non-perturbative
parameters entering K and Md,s is necessary. Similarly ACD can be smaller than SM
by at most 10 degrees.
We illustrate these properties in Fig. 4, where we show the 1/R dependence of |Vt d |,
, C and .
In obtaining these results we used the following procedure. First using the
central value |Vt d |SM = 0.00815 from the SM fit of [34] and mt = 167 GeV we determined
|Vt d |ACD by means of (3.36). The result is shown in Fig. 4(a). Using next the central
(a)
(b)
(c)
(d)
(d) .
Fig. 4. Results for various CKM parameters in the ACD model and in the SM: (a) |Vtd |; (b) ; (c) ;
241
Fig. 5. Unitarity triangle in the ACD model for 1/R = 200 GeV and in the SM.
Table 1
Values and errors for different quantities in the ACD model with 1/R = 200 GeV [43] and in the SM from [34].
The 95% probability regions are given in brackets
Strategy
ACD (1/R = 200 GeV)
SM
0.342 0.027
0.357 0.027
(0.2880.398)
(0.3050.411)
0.197 0.047
0.173 0.046
(0.1020.296)
(0.0760.260)
0.23 0.25
0.09 0.25
(0.700.27)
(0.540.40)
59.5 7.0
63.5 7.0
(Degrees)
(45.374.8)
(51.079.0)
18.6+1.9
1.5
18.0+1.7
1.5
(15.726.2)
(15.421.7)
sin 2
Ms
1
ps

|Vtd | 103
7.80 0.42
8.15 0.41
(6.968.69)
(7.348.97)
values in (3.31) and (3.32) and the 1/R dependence of (Rt )ACD = |Vt d |ACD /(|Vcb |), we
determined , C and as functions of 1/R. We show in Fig. 5 the unitarity triangles
corresponding to the ACD model with 1/R = 200 GeV and the SM model.
In summary, the CKM elements in the ACD model extracted from F = 2 transitions
and the CP asymmetry aKS are not expected to differ substantially from the corresponding
values found within the SM. This is very fortunate as the most recent fits of the UT [34,37,
3942] based on the SM expressions for F = 2 transitions agree very well with the
direct measurement of the angle by means of aKS . With improved data on aKS ,
K and B
Bq F 2 , a constraint on the
and, in particular, Ms and improved values for B
Bq
compactification radius R from F = 2 processes is, in principle, possible. However, for
the time being even the lowest value 1/R = 200 GeV considered by us is consistent with
the present fits of the UT. Setting 1/R = 200 GeV and repeating the analysis of [34], that
uses the bayesian method [39], one finds [43] the values for (C,
),
sin 2, , Ms and
242
|Vt d | in the second column of Table 1. For a comparison we give the corresponding ranges
in the SM. To this end all input parameters of [34] have been used.
Comparing the two columns in Table 1, we observe all the patterns shown in Fig. 4.
However, due to substantial uncertainties in the input parameters, the effects of the KK
modes are partly washed out. In particular, the suppressions of |Vt d | and amount to 45%
rather
than O(10%) that we found in Fig. 4. This is easy to understand. With no uncertainty
Bd FBd , the value of |Vt d | is strongly correlated with S(xt , 1/R) by means of (3.12)
in B

Bd FBd are taken
as Md is known very well experimentally. Once the uncertainties in B
into account, the enhancement of
the function S in the ACD model can be compensated
Bd FBd , implying a smaller suppression of |Vt d | than
by the decrease of both |Vt d | and B
found in Fig. 4. Similar comments can be made in connection with other quantities in
Table 1.

K and do not depend
Bd FBd , B
Clearly, low energy non-perturbative parameters as B
on the compactification scale. On the other hand they are subject to uncertainties and it is
not surprising
that the fit of the unitarity triangle within the ACD model prefers lower
K than in the SM. The true effects of the KK modes can then
Bd FBd and B
values of B
only be clearly seen as in Fig. 4 when the input parameters have no uncertainties. This
exercise shows very clearly the importance of the reduction of the theoretical uncertainties
in connection with the search for new physics.
This discussion makes it clear that it is
impossible to claim, as done in [18], that the
Bd FBd by a factor of three will necessarily
reduction of the error on the parameter B
increase the lowest allowed value of the compactification scale
1/R. The lower bound on
Bd FBd . With decreasing
1/R from Md depends necessarily on the actual value of B

Bd FBd the lower bound on 1/R becomes weaker. One can easily check that decreasing
B

Bd FBd to 200 MeV, still within the present uncertainties, no
the central value for B

Bd FBd is decreased
significant lower bound on 1/R can be obtained even if the error on B
by a factor of three.
In order to illustrate this point we show in Fig. 6 the correlation between 1/R and
Bd FBd for different values of |Vt d | and mt = 167 GeV. To this end we have used
B
(3.12) with Md = 0.503/ps. Clearly on the basis of Md alone it is impossible to
say anything about 1/R. However, even
if |Vt d | is determined through Md /Ms , the
Bd FBd with the maximal and minimal values of
values of 1/R depend sensitively on B

Bd FBd corresponding to maximal and minimal values of 1/R, respectively. Thus a
B

Bd FBd could in principle provide an improved
significantly improved lower bound on B

Bd FBd
lower bound on 1/R. However, if future lattice calculations will find values for B
that are smaller than the present central value, 1/R could be forced to be low in order for
the ACD model to fit the value of Md .
Fig. 6. Correlation between 1/R and
243

B FB for different values of |Vtd |.
B
d
d
4. Rare K- and B-decays

4.1. Preliminaries
We will now move to discuss the semileptonic rare FCNC transitions K + + ,
KL 0 , KL ,
B Xs,d and Bs,d .
Within the SM and the ACD
model these decays are loop-induced semileptonic FCNC processes governed by Z 0 penguin and box diagrams and described by two functions X(xt , 1/R) and Y (xt , 1/R)
for the decays with and in the final state, respectively.
A particular and very important virtue of these decays (with the exception of KL )
is their clean theoretical character [25] that allows to probe high energy scales of the
theory and in particular to measure Vt d and Im t = Im Vts Vt d from K + + and
KL 0 , respectively. Moreover, the combination of these two decays offers one of
the cleanest measurements of sin 2 [44], see [44,45] for more details.
4.2. Effective Hamiltonians for K and B Xs
The effective Hamiltonian for K + + is given in the ACD model as follows
GF
Heff =
2 2 sin2 w

l
+ Vts Vt d X X(xt , 1/R) (s d)V A (l l )V A .

Vcs Vcd XNL
(4.1)
l=e,,
The index l = e, , denotes the lepton flavour. The dependence on the charged lepton
mass resulting from the box diagram is negligible for the top contribution. In the charm
sector this is the case only for the electron and the muon but not for the -lepton. In what
t (mt ) it equals 0.994.
follows we will set the QCD factor X [4648] to unity as for mt m
244
The function X(xt , 1/R) relevant for the top part is given by
X(xt , 1/R) = C(xt , 1/R) + B (xt , 1/R).
(4.2)
Here
C(xt , 1/R) = C0 (xt ) +
Cn (xt , xn )
(4.3)
n=1
Z 0 -penguin diagrams with
the SM contribution C0 (xt ) given by

3xt + 2
xt xt 6
+
.
ln
x
C0 (xt ) =
t
8 xt 1 (xt 1)2
results from
(4.4)
The sum in (4.3) represents the KK contributions that are calculated from the Feynman
diagrams in Fig. 7 as discussed below. Next
B (xt , 1/R) = 4B0 (xt ) +
Bn (xt , xn )
(4.5)
n=1
results from box diagrams with the SM contribution given by the first term and

xt
1
xt ln xt
.
+
B0 (xt ) =
4 1 xt
(xt 1)2
(4.6)
The sum in (4.5) represents the KK contributions that are calculated from the Feynman
diagrams in Fig. 9 as discussed below.
Fig. 7. Penguin diagrams contributing to Cn (xt , xn ). The analytical expressions are listed in Appendix C.
(8) includes the additional diagram with W and (a , G ) interchanged.
245
l
The expression corresponding to X(xt , 1/R) in the charm sector is the function XNL
. It
results from the NLO calculation [49] and is given explicitly in [48] where further details
can be found. As in the case of the charm contributions in the S = 2 Hamiltonian, here
l
for = mc and
the KK contributions are also negligible. The numerical values for XNL
(4)
several values of and mc (mc ) can be found in [48]. For our purposes we need only
MS
( = 0.221)

1 2 e
1
P0 (X) = 4 XNL
(4.7)
+ XNL
= 0.41 0.06,
3
3
(4)
where the error results from the variation of and mc (mc ).

MS
In the case of KL 0 that is governed by CP-violating contributions only the top
contribution in (4.1) matters. Similarly the effective Hamiltonians for B Xs,d are
obtained from (4.1) by neglecting the charm contribution and changing appropriately the
CKM factor and quark flavours. In all these decays the KK modes contribute universally
only through the function X(xt , 1/R).
4.3. Effective Hamiltonians for Bs,d and KL
The effective Hamiltonian for Bs l + l in the ACD model is given as follows:
GF
V A (ll)
V A + h.c.,
Vtb Vt s Y Y (xt , 1/R)(bs)
Heff =
2 2 sin2 w
(4.8)
with s replaced by d in the case of Bd l + l . The charm contributions are fully negligible
here. In what follows we will set the QCD factor Y [4648] to unity as for mt m
t (mt )
it equals 1.012.
The function Y (xt , 1/R) is given in the ACD model by
Y (xt , 1/R) = C(xt , 1/R) + B (xt , 1/R),
(4.9)
with C(xt , 1/R) given in (4.3) and

B (xt , 1/R) = B0 (xt ) +
Bn (xt , xn )
(4.10)
n=1
with the first term representing the SM contribution. The second term results from the box
diagrams in Fig. 9.
The effective Hamiltonian for the short distance contribution to KL is given by
(4.8) with the appropriate change of the CKM factors and quark flavours and a small SM
l
charm contribution YNL [48] in analogy to XNL
in (4.1). The relevant branching ratio will
be given below.
4.4. Z 0 -penguin diagrams
The function Cn (xt , xn ) can be found by calculating the vertex diagrams in Fig. 7 and
adding an electroweak counter term as discussed in detail in [50]. The latter is found
246
by calculating the self-energy diagrams of Fig. 12 that describe flavour non-diagonal

propagation of quark fields and subsequently rotating the quark fields appropriately so
that this flavour non-diagonal propagation does not take place in the new fields.
In the t HooftFeynman gauge for the W(n)

and G
(n) propagators, the contribution
of the diagrams in Fig. 7 to the flavour-changing vertex Z including the electroweak

counterterm is given by
Z = i
g23
V Vt d Cn (xt , xn )s (1 5 )d.
16 2 cos w t s
(4.11)
We neglect the external momenta in Fig. 7 and the masses of external quarks. The function
Cn (xt , xn ) is defined through
Cn (xt , xn ) = F (xt (n) ) F (xu(n) )
(a)
(4.12)
(b)
Fig. 8. (a) Contribution Cn of the nth KK mode to C(xt , 1/R). The functions Fi correspond to the diagrams
given in Fig. 7. (b) The functions C(xt , 1/R) and C0 (xt ).
Fig. 9. Box diagrams contributing to Bn and to Bn .
247
with the functions F (xt (n) ) and F (xu(n) ) representing the contributions of the Qt (n) , Ut (n)
and Qu(n) , Uu(n) modes, respectively,
F (xt (n) ) =
8

i=1
2
1 1 2
s
Fi (xt (n) ) +
Si (xt (n) ).
2 3 w

(4.13)
i=1
Here sw sin w , Fi stand for the contributions of diagrams in Fig. 7 and Si denote the
electroweak counter terms.
The explicit contributions of various sets of diagrams to these functions are given in
Appendix C. Adding up these contributions we find

xt
xt + xn
2
Cn (xt , xn ) =
x 8xt + 7 + (3 + 3xt + 7xn xt xn ) ln
.
1 + xn
8(xt 1)2 t
(4.14)
In Fig. 8(a), we show Cn (xt , xn ) as a function of n/R. In this case the convergence of the
sum of the KK modes is significantly improved by the GIM mechanism so that only a few
first terms in the sum in (4.3) are relevant. We will return to this at the end of this section.
We observe that in contrast to the function S of Section 3 the diagrams involving only W(n)
play the dominant role among the KK contributions.
In Fig. 8(b) we plot C(xt , 1/R) vs. 1/R. For 1/R = 200 GeV we observe a 38%
enhancement of the function C with respect to its SM value given by C0 (xt ). For 1/R =
250 GeV this enhancement decreases to 26% and it is 11% for 1/R = 400 GeV. The
significant enhancement of C is the origin of the enhancements of the branching ratios
discussed below.
4.5. Z 0 b b vertex in the large mt limit
The result in (4.14) can also be used to find the dominant KK contribution to the Z 0 b b
vertex in the large mt limit. As discussed already in [50], in general the calculation of a
low energy effective flavour violating Z 0 d s vertex cannot be directly compared with the
full calculation of the Z 0 b b vertex. However, as pointed out there in the special limiting
case mt MW the two approaches, the direct evaluation of on-shell diagrams and the
operator product expansion considered by us, are equivalent. The reason for this is that in
the limit mt MW with 1/R mt all the other mass scales as mb and external momenta of
b-quarks become negligible compared with mt and 1/R. Consequently with the definition
Zbb = i
g2
b (gL PL + gR PR )b
cos w
(4.15)
the one-loop contributions to the coupling gL can be found in the large mt limit by using
the relation [50]
2
GF MW
gL =
C(xt , 1/R).
2 2
(4.16)
248
In the case of the SM contributions this relation has been already analyzed in detail in [50]
including s to the one-loop Z 0 b b vertex. For the KK contribution we find from (4.14)
C =
Cn (xt , xn )
n=1

=
4 4

m2t m2W 4
m4t 2
m6t 4
m2t 2
mt
2
+
+
R
R 4 + (4.17)
96
4320
864
96m2W
2160m2W
with the dominant contribution represented by the first term, see Appendix D for the
derivation. Retaining only this term and using (4.16) we find
R2
GF
gLKK = m4t
96
2
(4.18)
that agrees with a recent direct calculation of the KK contribution to the Zbb vertex in
[51], see their formula (3.6). Interestingly, while the corrections to the asymptotic result
(4.18) relevant for the Zbb vertex have been found in [51] to be substantial, the result
C =
m4t 2
2
96MW
R2
(4.19)
give a good approximation to the full KK contribution to the function C(xt , 1/R) even for
mt 167 GeV.
4.6. Box diagram contributions
The functions Bn (xt , xn ) and Bn (xt , xn ) can be found by appropriate rescaling of the
box diagrams contributing to F = 2 transitions that we considered in Section 3. It turns
out that the box contributions of the KK modes are tiny. For instance, for 1/R = 200 GeV
and mt = 167 GeV we find B1 (xt , x1 ) = 0.0098 and B1 (xt , x1 ) = 0.0049 with even
smaller values for n > 1 and larger 1/R. Consequently, these contributions can be safely
neglected in comparision with Cn . For completeness we give in Appendix D the analytic
formulae for Bn (xt , xn ) and Bn (xt , xn ).

4.7. The functions X and Y
Neglecting the box contributions of the KK modes we find
X(xt , 1/R) = X0 (xt ) +
Y (xt , 1/R) = Y0 (xt ) +
n=1
Cn (xt , xn ) X0 (xt ) + X,
Cn (xt , xn ) Y0 (xt ) + Y
n=1
(4.20)
(4.21)
249
with (mt = 167 GeV)

3xt 6
xt xt + 2
X0 (xt ) =
+
ln xt = 1.526,
8 xt 1 (xt 1)2

3xt
xt xt 4
+
Y0 (xt ) =
ln xt = 0.980,
8 xt 1 (xt 1)2
(4.22)
(4.23)
summarizing the SM contributions and X = Y representing the corrections due to KK

modes.
In Fig. 10 we plot X(xt , 1/R) and Y (xt , 1/R) vs. 1/R. We observe that due to the
inequality X0 > Y0 the relative impact of the KK modes is larger in the function Y . For
1/R = 200 GeV the functions X and Y are enhanced by 20% and 31%, respectively.
For 1/R = 250 GeV this enhancement decrease to 13%(21%) and are only 6%(9%) for
1/R = 400 GeV.
In Table 2 we give the values of the functions S, C, X and Y for different 1/R and
mt = 167 GeV.
(a)
(b)
Fig. 10. The functions (a) X(xt , 1/R) and X0 (xt ) and (b) Y (xt , 1/R) and Y0 (xt ).
Table 2
Values for the functions S, C, X and Y
1/R (GeV)
200
250
300
400
SM
2.813
2.664
2.582
2.500
2.398
1.099
1.003
0.946
0.885
0.798
1.826
1.731
1.674
1.613
1.526
1.281
1.185
1.128
1.067
0.980
250
4.8. Branching ratios for rare decays

The branching ratios for the rare decays in question can be directly obtained from
[25] by simply replacing the SM functions X0 and Y0 by X(xt , 1/R) and Y (xt , 1/R),
respectively. We have

Br K + +
2
2

Re c
Re t
Im t
P
X(x
,
1/R)
+
(X)
+
X(x
,
1/R)
= +
(4.24)
,
t
0
t
5
5
+ = rK +
3 2 Br(K + 0 e+ )
2 2 sin4 w
8 = 4.31 1011 ,
(4.25)
where we have used [52]

=
1
,
129
sin2 w = 0.23,

Br K + 0 e+ = 4.87 102.
(4.26)
Here i = Vis Vid with c being real to a very high accuracy. rK + = 0.901 summarizes
isospin breaking corrections [53] in relating K + + to K + 0 e+ . P0 (X) is
given in (4.7).
Next,

2

Im t
Br KL 0 = L
(4.27)
X(x
,
1/R)
,
t
5
rK (KL )
L = L
(4.28)
+ = 1.88 1010
rK + (K + )
with + given in (4.25) and rKL = 0.944 summarizing isospin breaking corrections in
relating KL 0 to K + 0 e+ [53].
Next, normalizing to Br(B Xc e ) and summing over three neutrino flavours we find
3 2
Br(B Xs )
|Vt s |2 X2 (xt , 1/R) (0)
=
.
Br(B Xc e ) 4 2 sin4 w |Vcb |2
f (z)
(z)
(4.29)
Here f (z) is the phase-space factor for B Xc e with z = m2c /m2b and (z) = 0.88
[54,55] is the corresponding QCD correction. The factor (0) = 0.83 represents the
QCD correction to the matrix element of the b s transition due to virtual and
bremsstrahlung contributions. In the case of B Xd one has to replace Vt s by Vt d
which results in a decrease of the branching ratio by roughly an order of magnitude. In our
numerical calculations we set f (z) = 0.54 and Br(B Xc e ) = 0.104.
Next, the branching ratio for Bs l + l is given by

Br Bs l + l

2

m2l 2 2
G2F
2
2

= (Bs )
F
m
m
Bs l Bs 1 4 2 Vt b Vt s Y (xt , 1/R), (4.30)
2
4 sin w
m Bs
251
where FBs is the Bs meson decay constant. The formula for Br(Bd l + l ) is obtained
by replacing s by d. The relevant input parameters are [32]

FBs = (238 31) MeV.
FBd = 203+27
(4.31)
34 MeV,
We set also (Bs ) = 1.46 ps and (Bd ) = 1.54 ps [52]. The short distance contribution to
the dispersive part of KL + is given by [46,49]

2

Re c
Re t
P0 (Y ) + 5 Y (xt , 1/R) ,
Br KL + SD =
(4.32)
2 Br(K + + ) (KL ) 8
= 1.733 109 ,
(K + )
2 sin4 w
(4.33)
where we have used Br(K + + ) = 0.634. The charm contribution including NLO
corrections is given by [48]
YNL
(4.34)
= 0.128 0.013.
4
Unfortunately due to long distance contributions to the dispersive part of KL + ,
the extraction of Br(KL + )SD from the data is subject to considerable uncertainties.
The present best estimate reads [56]

Br KL + SD 2.5 109 .
(4.35)
P0 (Y ) =
4.9. GIM mechanism and convergence of the KK sum

The tree-level masses of the KK modes of all particles approach the value n/R for large
KK mode numbers n, see Eqs. (2.30) and (A.2). Due to the unitary CKM matrix the GIM
mechanism [12] comes into play. It suppresses partly the higher KK mode contributions
to the sums in Eqs. (3.4), (4.2), (4.5) and (4.10) and is essential for the determination
of the dominant contributions. For the Sn function the GIM suppression amounts to an
additional factor of 1/n4 for the contributions W W , GG and W G. The contributions from
(t )
diagrams with a are not suppressed due to the second term in the coupling m4 , see
(A.3). This results in a hierarchy of the various contributions to Sn with W W , GG and
W G proportional to 1/n6 , W a and Ga proportional to 1/n4 and the dominant contribution
aa proportional to 1/n2 for large values of n.
For the penguin diagrams the effect of the GIM mechanism on the convergence of the
particular contributions is partly hidden by the subtle cancellation of divergencies among
the different contributions and the two self-energy diagrams. For the combinations plotted
in Fig. 8 we observe that the term corresponding to W : F1 + F2 + F5 is logarithmically
divergent without GIM and shows a 1/n2 behaviour after GIM mechanism has been taken
into account. The a , G : F3 + F4 + F6 + F7 term shows no GIM suppression and is
proportional to 1/n2 for large values of n. The mixed term W , a , G : F8 is constant
for large n before the inclusion of GIM mechanism, but is GIM suppressed by a factor of
1/n2 .
The contributions to Bn and Bn show a 1/n4 behaviour with and without GIM, except
the GW W and HW W terms. They are proportional to 1/n2 without GIM and behave like
252
1/n4 with GIM suppression at work. This different asymptotic behaviour with respect
to the Sn function is due to the appearance of leptons with negligible zero-mode masses
instead of quarks in the box diagrams contribution to the functions G and H , see Eq. (A.3).
4.10. Numerical analysis
As discussed in Section 3, |Vt d |, C and in the SM and in the ACD model differ
from each other. In our numerical analysis we will take this difference into account. Not
taking this difference into account would misrepresent significantly the patterns of the
enhancements of branching ratios in question.
In Fig. 11 we show the branching ratios Br(K + + ) and Br(Bs + ) as
functions of the compactification scale 1/R. As |Vt s |ACD = |Vt s |SM , the enhancement
of Br(Bs + ) is entirely governed by the ratio (Y/Y0 )2 . On the other hand the
dependence of Br(K + + ) on 1/R differs from the one of (X/X0 )2 because of the
additional charm contribution that is independent of 1/R and the fact that Vt d in the ACD
model differs from its SM value. The remaining branching ratios are shown in Table 3.
The 1/R dependence of Br(B Xs ) is governed by the function X2 (xt /1/R), while
the corresponding dependences of Br(B Xd ), Br(KL 0 ), Br(Bd + )
and Br(KL + )SD include also the 1/R dependence of Vt d shown in Fig. 4(a).
As expected, all the branching ratios are significantly enhanced for 1/R 300 GeV. For
1/R 400 GeV, except for Br(Bs + ), the distinction between the predictions of
the ACD model and the SM will be very difficult.
In this numerical analysis we have used the results for the CKM parameters of Fig. 4 and
the central values of all the remaining input parameters as given above. The uncertainties in
these parameters partly cover the differences between the ACD model and the SM model
and it is essential to reduce these uncertainties considerably if one wants to see the effects
of the KK modes in the branching ratios in question. Therefore, a detailed analysis that
includes all uncertainties would be in our opinion premature at present.
Fig. 11. Branching ratio for the decays K + + and Bs + as predicted by the ACD model and the
SM as functions of the inverse radius R of the extra dimension.
253
Table 3
Branching ratios for rare decays in the ACD model and the SM as discussed in the text
1/R

Br K + + 1011

Br KL 0 1011

Br KL + SD 109

Br B Xs 105

Br B Xd 106

Br Bs + 109

Br Bd + 1010
200 GeV
250 GeV
300 GeV
400 GeV
SM
8.70
8.36
8.13
7.88
7.49
3.26
3.17
3.09
2.98
2.80
1.10
1.00
0.95
0.88
0.79
5.09
4.56
4.26
3.95
3.53
1.80
1.70
1.64
1.58
1.47
6.18
5.28
4.78
4.27
3.59
1.56
1.41
1.32
1.22
1.07
4.11. An upper bound on Br(K + + ) in the ACD model

The enhancement of Br(K + + ) in the ACD model is interesting in view of the
results from the AGS E787 Collaboration at Brookhaven [57] that read

11
Br K + + = 15.7+17.5
(4.36)
8.2 10
with the central value by a factor of 2 above the SM expectation. Even if the errors
are substantial and this result is compatible with the SM, the ACD model with a low
compactification scale is closer to the data. As emphasized in [57,58] the central value
in (4.36) implies within the SM a value for |Vt d | that is substantially higher than the
value obtained from the standard analysis of the UT of Section 3. Here we would like
to emphasize that within the ACD model Br(K + + ) is closer to the data in spite of
the fact that |Vt d |ACD < |Vt d |SM . The enhanced Z 0 -vertex represented by the function C is
responsible for this behaviour.
In [48] an upper bound on Br(K + + ) has been derived within the SM. This
bound depends only on |Vcb |, X0 , and Md /Ms . With the precise value for the angle
now available this bound can be turned into a useful formula for Br(K + + ) [58]
that expresses this branching ratio in terms of theoretically clean observables. In the ACD
model this formula reads:

Br K + +
= + |Vcb |4 X2 (xt , 1/R)

2

1
4 P0 (X)
Rt cos +
,
Rt2 sin2 +
|Vcb |2 X(xt , 1/R)
(4.37)
where = 1/(1 2 /2)2 , + = + /8 and Rt is given in (3.29). This formula is

theoretically very clean and does not involve hadronic uncertainties except for in (3.29)
and to a lesser extent in |Vcb |.
In order to find the upper bound on Br(K + + ) in the ACD model we use
|Vcb | 0.0422,
P0 (X) < 0.47,
sin = 0.40,
mt < 172 GeV,
(4.38)
254
Table 4
11 for different values of and 1/R and M = 18/ps (21/ps).
Upper bound on Br(K + + )
in units of 10
s

B FB 190 MeV
The stars indicate the results corresponding to B
d
1.30
1.25
1.20
1.15
1/R = 200 GeV

13.8 12.3
13.0 (11.6)
12.2 (10.9)
11.5 (10.3)
1/R = 250 GeV

12.7 11.3
12.0 (10.7)
11.3 (10.1)
10.6 (9.5)
1/R = 300 GeV
1/R = 400 GeV
12.0 (10.7)
11.3 (10.1)
11.4 (10.2)
10.7 (9.6)
10.1 (9.0)
10.7 (9.6)
10.1 (9.1)
9.5 (8.5)
SM
10.8 (9.3)
10.3 (8.8)
9.7 (8.4)
9.1 (7.9)
where we have set sin to its central value (see (3.33)) as Br(K + + ) depends very
weakly on it. The bound on |Vcb | results from |Vcb | = 0.0406 0.0008 [34]. We used here
two standard deviations as the determination of |Vcb | involves some hadronic uncertainties.
The result of this exercise is shown in Table 4. We give there Br(K + + )max as a
function of and 1/R for two different values of Ms . The range for chosen by us is in
accordance with (3.30) and the recent new analysis in [59] that gives = 1.22 0.07. The
upper bound in the SM given in the last column is lower than the values for 1/R = 400 GeV
by roughly 10%. We observe that for 1/R = 200 GeV and = 1.30 the maximal value for
Br(K + + ) in the ACD model is rather close to the central value in (4.36).
At first sight the 30% enhancement of Br(K + + ) for 1/R = 200 GeV with
respect to the SM values seems to contradict the results in Table 3, where a more
modest enhancement of Br(K + + ) is seen. However, one should realize that now
Md /Ms and not Md alone enters the analysis and the enhancement of the function
X is not accompanied by a suppression of |Vt d |, that with Rt given by (3.29), equals
the one in the SM. The
consistency with the experimental value of Md requires then
Bd FBd . In Table 4 we indicate by a star the cases that require
a sufficiently small B

Bd FBd 190 MeV. Such low values are rather improbable in view of (3.30) and
B
consequently values of Br(K + + ) larger than 12 1011 are rather unlikely even
in the ACD model.
5. Summary and outlook

In this paper we have calculated for the first time the contributions of the Kaluza
Klein (KK) modes to MK , K , Md,s and rare decays K + + , KL 0 ,
KL + , B Xs,d and Bs,d in the Appelquist, Cheng and Dobrescu
(ACD) model with one universal extra dimension. As a byproduct we have given a list
of the required Feynman rules that have not been presented in the literature so far.
The nice property of this extension of the SM is the presence of only a single
new parameter, 1/R. This economy in new parameters should be contrasted with
supersymmetric theories and models with an extended Higgs sector. Taking 1/R =
200 GeV our findings are as follows:
255
The short distance one-loop function S(xt , 1/R) relevant for F = 2 transitions is
larger than the corresponding SM function S0 (xt ) by roughly 17%. This implies on
the basis of K and Md a 8% suppression of |Vt d |ACD with respect to |Vt d |SM
and a decrease of the angle ACD by 10 with respect to SM . On the other hand,
(Ms )ACD is larger than (Ms )SM by 17%, see Section 3 for details. MK is
essentially uneffected.
In order to see whether the modifications of the SM expectations are required by the
data, the comparision of |Vt d | and extracted from K and Md in the SM and in the
ACD model with the universal unitarity triangle constructed by means of |Vub /Vcb |,
Md /Ms and a(KS ) will be important. To this end, the data on these quantities
have to be improved and the uncertainties in the relevant non-perturbative parameters
reduced.
The short distance one-loop function X(xt , 1/R) relevant for the decays K + + ,
KL 0 and B Xs,d is larger than the corresponding SM function X0 (xt )
by roughly 20% due to KK contributions to the Z 0 -penguins. In the case of K +
+ , KL 0 and B Xd this enhancement is partially compensated by
the fact that |Vt d |ACD < |Vt d |SM and ACD < SM . We then find the enhancements of
Br(K + + ), Br(KL 0 ) and Br(B Xd ) over the SM expectations by
16%, 17% and 22%, respectively. As B Xs is governed by the CKM element
|Vt s | that is common to the SM and the ACD model, the enhancement of Br(B
Xs ) amounts to 44%.
The short distance one-loop function Y (xt , 1/R) relevant for the decays Bd,s
and KL ,
is larger than the corresponding SM function Y0 (xt ) by roughly 30%.
As Bs is governed by the CKM element |Vt s | with |Vt s |ACD = |Vt s |SM this
implies a 72% enhancement of Br(Bs )
relative to the SM expectation. In the
case of Bd and KL ,
due to |Vt d |ACD < |Vt d |SM , the corresponding
enhancements amount to 38% and 46%.
As the ACD model belongs to the class of MFV models, general properties of these
models identified in [14,60] are automatically valid here.
For 1/R = 250 (300) GeV all these effects are decreased roughly by a factor of
1.5 (2.0). See Fig. 11 and Table 3.
In short, the signatures of the deviations from the SM expectations are:
The decrease of |Vt d |, , and C.

The increase of Ms .
The increase of all branching ratios considered in this paper with a hierarchical
structure of maximal enhancements: K + + (16%), KL 0 (17%),
B Xd (22%), KL (38%), B Xs (44%), Bd (46%) and
Bs (72%) for 1/R = 200 GeV. For 1/R = 250 (300) GeV these enhancements
are decreased roughly by a factor of 1.5 (2.0).
In the coming years of particular interest will be the improved measurements of
Br(K + + ) and Ms . Indeed, in the MFV models Br(K + + ) can be
predicted as a function of X once the angle and Md /Ms are known. The relevant
256
formula is given in (4.37). For a given value of X, the branching ratio Br(K +
+ ) decreases with increasing Ms . If the present central experimental values for
Br(K + + ) will remain while the experimental error will decrease significantly,
the SM expectations for Br(K + + ) will be significantly below the experimental
data while the corresponding estimates within the ACD model may agree with them due to
X(xt , 1/R) > X0 (xt ), see Table 4.
Another definite prediction of the ACD model is an increase of Ms over the SM
value. This should be contrasted with the prediction of the MSSM at large tan where a
suppression of Ms with respect to the SM value is predicted [61]. Thus, when the ratio
(Ms )exp /(Ms )SM will be known with a sufficient accuracy, we will know whether the
ACD model with a low 1/R or the MSSM with a large tan is ruled out by the data.
A distinction between the ACD model and the MSSM at low tan will also be possible.
In the latter case the supersymmetric effects in F = 2 transitions considered in Section 3
are generally larger than in F = 1 transions so that the branching ratios for K + + ,
KL 0 , KL + , B Xd and Bd are generally suppressed with
respect to the SM while they can be enhanced for special ranges of supersymmetric
parameters in the case of B Xs and Bs [62].
However, the main message from our analysis is the following one. Even for the
lowest compactification scale, 1/R = 200 GeV, considered by us, the ACD model is
consistent with all the available data on FCNC processes analyzed here. No fine tunning
of the parameters characteristic in the flavour sector for general supersymmetric models
is necessary. This is first of all connected with the GIM mechanism that assures the
convergence of the sum over the KK modes in the case of Z 0 penguin diagrams, removing
the sensitivity of the calculated branching ratios to the scale Ms 1/R at which the higher
dimensional theory becomes non-perturbative and at which the towers of the KK particles
must be cut off in an appropriate way.
With the much improved data on the processes calculated in this paper and the
theoretical uncertainties reduced, it should be possible in the future to distinguish the
predictions of the ACD model from the SM ones, provided 1/R < 400 GeV. For higher
compactification scales this distinction will be very difficult with a possible exception of
Bs + .
Whether these findings apply also to B Xs , B Xs l + l and KL 0 e+ e
is an interesting question. As the Z 0 -penguins are enhanced in the ACD model, the
corresponding enhancement of the branching ratios for B Xs l + l and KL 0 e+ e
can easily be calculated by means of the known formulae [25,63,64]. However, such
an analysis would clearly be unsatisfactory as these decays receive also contributions
from the -penguins and magnetic penguins. The KK contributions to -penguins are
unknown, while the corresponding contributions to the magnetic penguins have been
calculated in the context of an analysis of the decay B Xs under the assumption of
[2]. We have seen that this assumption would be correct

the dominance of the scalars a(n)
in the case of S(xt , 1/R) for 1/R 300 GeV, but certainly would misrepresent the KK
contributions to C(xt , 1/R) in view of strong cancellations between different contributions
as discussed in Section 4. Consequently a satisfactory analysis of B Xs , B Xs l + l
and KL 0 e+ e requires a priori the inclusion of all KK contributions. We will address
all these issues in the forthcoming paper [11].
257
The ACD model is an interesting prototype for universal extra dimensional models
using orbifold compactification, but it is obviously a particular model and one may wonder
how much of the results presented here are strictly linked to the specific model. Without
a detailed analysis, which is clearly beyond the scope of our paper, one can only make
the following general observations. In models where part of the field content is confined
to the brane, KK number is not conserved and contrary to the ACD model KK parity is
also violated. The GIM mechanism ceases to work and KK effects are expected to be
significantly larger, see, e.g., [13]. Similarly in models with more than five dimensions in
which the sum over the KK modes generally diverges we expect the KK effects to be larger
than found here.
Acknowledgements
We thank Fabrizio Parodi and Achille Stocchi for providing the numbers in the second
column of Table 1 and Geraldine Servant for interesting comments. This research was
partially supported by the German Bundesministerium fr Bildung und Forschung
under contract 05HT1WOA3 and by the Deutsche Forschungsgemeinschaft (DFG) under
contract Bu.706/1-1.
Appendix A. Feynman rules in the ACD model

In this section, we list all propagators and vertex rules needed for the calculation of the
box and penguin diagrams considered in this paper. The Feynman rules are derived in the
5d R -gauge described in Section 2.3.
It is convenient to define a 4-dimensional counterpart to every parameter in the
5-dimensional Lagrangian, marked there with a caret. The
conversion factors are chosen in
order to eliminate all explicit appearances of factors of 2R in the Feynman rules:
= ,
v=
1
g =
g ,
2R
1
U /D ,
U /D =
2R
2R v,
1
g2 =
g2 ,
2R
(A.1)
where is the Higgs mass parameter.

In order to simplify the notation, we omit the KK indices of the fields and of the
mass parameters defined in (A.3) and (A.4). There is no ambiguity because in one-loop
calculations at least one field is always a zero-mode. In our case, this is the Z boson in the
Z vertices, and the down-type quark or the neutrino in all other vertices. Due to KK parity
conservation, the other two fields have equal KK mode number, i.e., either zero or n 1.
Fermion zero-modes have substantially different Feynman rules than their KK excitations. The fields Qu , Qd , U , D, L , Le and E are always supposed to be (n 1)-modes,
while the zero-modes are labeled u, d, and e. The generation indices are i = u, c, t and
j = d, s, b for the quarks and i, j = e, , for the leptons.
258
The masses of all bosonic particles can be expressed in terms of four independent mass
parameters:
n2
n2
2
,
M
=
+ MZ2 ,
Z(n)
R2
R2
n2
n2
2
2
2
2
MH
MW
(A.2)
(n) = 2 + MW ,
(n) = 2 + 2 .
R
R
We introduce two sets of mass parameters appearing in the fermion-scalar couplings.
(i)
The mass parameters mx are
n
n
m(i)
m(i)
1 = R ci(n) + mi si(n) ,
2 = R si(n) + mi ci(n) ,
n mi
n mi
m(i)
m(i)
(A.3)
3 = MW ci(n) + R M si(n) ,
4 = MW si(n) + R M ci(n) ,
W
W
where MW and the up-type fermion masses mi on the r.h.s. are the zero-mode masses.
The parameters ci(n) and si(n) denote the cosine and sine respectively of the fermion mass
(i,j )
mixing angle i(n) defined in (2.29). The mass parameters Mx are
2
MA(n)
=
(i,j )
(i,j )
= mj ci(n) ,
M2 = mj si(n) ,
m
n
n mj
j
(i,j )
(i,j )
M3 =
(A.4)
ci(n) ,
M4 =
si(n) ,
R MW
R MW
where again MW and the down-type fermion masses mj on the r.h.s. are the zero-mode
masses.
All momenta and fields are assumed to be incoming.
M1
A.1. Propagators
The propagators are:
for scalar fields S = H, a 0 , a , A5 , G0 , G :
=
i
,
k 2 M 2 + iM
with the masses

S
H(n)
0
a(n)
a(n)
MH (n)
MZ(n)
MW (n)
A5(n)
MA(n)
G0(n)
MZ(n)
G
(n)
MW (n)
(A.5)
for gauge bosons V = A, Z, W :

=

i
kk
(1
)
g
,
k 2 M 2 + iM
k 2 M 2 + iM
with the masses

V
A(n)
Z(n)
W(n)
M MA(n) MZ(n) MW (n)
(A.6)
259
for fermion fields F = u, d, Qu , Qd , U, D, , e, L , Le , E:

=
i(/k + m)
,
k 2 m2 + iM
with the masses

F
Qu(n)
Qd(n)
U(n)
D(n)
L() n
Le(n)
E(n)
mu
md
mu(n)
md(n)
mu(n)
md(n)
me
m()n
me(n)
me(n)
(A.7)
A.2. Vertices
The Feynman rules for the vertices are:
g2
g C.
cw MW (n)
2
2
ZW + G : C = sw2 MW
+ cw
2
2
cw
ZW G+ : C = sw2 MW
ZW + a :
C = MW
ZW a + :
C = MW
Za + a :
ZG+ a :
ZG a + :
n2
,
R2
n
.
R
ZG+ G :
n
,
R
n2
,
R2
(A.8)
(A.9)
(A.10)
(A.11)
ig2
(k2 k1 ) C.
2
2cw MW
(n)
2
2
2
2 n
C = cw
sw2 MW
2cw
,
R2
2
n2
2
2
C = 2cw
MW
cw
sw2 2 ,
R
n
C = MW ,
R
n
C = MW .
R
(A.12)
(A.13)
(A.14)
(A.15)
260

= ig2 cw g (k2 k1 ) + g (k1 k3 ) + g (k3 k2 ) .
(A.16)

Z u i ui :

Z i i :

i Qi :
ZQ
i Ui :
ZQ
ig2
(PL CL + PR CR ).
6cw
CL = 3 4sw2 ,
CR = 4sw2 ,
Z dj dj :
CL = 3,
CR = 0,
Z ej ej :

2 ,
CL = 4sw2 + 3ci(n)
i Ui :
ZU
2 ,
CR = 4sw2 + 3ci(n)
CL = 3si(n) ci(n) ,
CR = 3si(n) ci(n) ,
i Qi :
ZU
CL = 3 + 2sw2 ,
CR = 2sw2 ,
(A.17)
CL = 3 + 6sw2 ,
CR = 6sw2 ,
(A.18)
2 ,
CL = 4sw2 + 3si(n)
2 ,
CR = 4sw2 + 3si(n)
CL = 3si(n) ci(n) ,
CR = 3si(n) ci(n) .
(A.19)
(A.20)
ig2
= PL CL .
2
CL = Vij ,
W dj ui :
CL = Vij ,
(A.21)
i dj :
W +Q
CL = ci(n) Vij ,
W dj Qi :
CL = ci(n) Vij ,
(A.22)
i dj :
W +U
CL = si(n) Vij ,
W dj Ui :
CL = si(n) Vij ,
(A.23)
W + i ej :
CL = ij ,
W ej i :
CL = ij ,
(A.24)
W + i Lj :
CL = ij ,
j i :
W L
CL = ij ,
(A.25)
CL = 0,
W Ej i :
CL = 0.
(A.26)
W + u i dj :
W i Ej :
G u i dj :

i dj :
G+ Q

i dj :
G+ U
+

G i Ej :

a i Ej :
G dj ui :
CL = m(i)
1 Vij ,
(i,j )
CR = M1 Vij ,
G dj Qi :
CL = m(i)
2 Vij ,
CR = M2
Vij ,
G dj Ui :
CL = 0,
CR = mj ij ,
G ej i :
CL = 0,
(j )
CR = m1 ij ,
j i :
G L
CL = 0,
CR = m2 ij ,
(i)
CL = m3 Vij ,
CR = M3
Vij ,
(i)
CL = m4 Vij ,
(i,j )
CL = mi Vij ,
CR = mj Vij ,
(i,j )
i dj :
a+U
(PL CL + PR CR ).
(j )
i dj :
a+Q
a + i Lj :
2 MW (n)
(i,j )
G i ej :
G+ i Lj :
g2
CR = M4
261
G Ej i :
a dj Qi :
CL = 0,
(j )
CR = m3 ij ,
CL = M (i,j ) V ,
CL = 0,
CR = m(i) V ,
2 ij

CL = mj ij ,
CR = 0,

(j )
CL = m1 ij ,
CR = 0,

(j )
CL = m2 ij ,
CR = 0,

(i,j )
CL = M3 Vij ,
(j )
CR = m4 ij ,
CR = m3 Vij ,
(i)
a Ej i :
ij
CR = m(i) V ,
1 ij
CL = M (i,j ) V ,
ij
2
a dj Ui :
j i :
aL
(A.27)
CR = mi Vij ,
Vij ,
CL = mj Vij ,

(i,j )
CL = M4
Vij ,
CR = m(i)
4 Vij ,
(A.28)
(A.29)
(A.30)
(A.31)
(A.32)
(A.33)
(A.34)
(j )
CL = m3 ij ,
CR = 0,
(A.35)
(j )
CL = m4 ij ,
CR = 0.
(A.36)
Appendix B. Different contributions to F = 2 box diagrams

The contributions from different sets of diagrams to the functions F (xt (n) , xu(n) ) in (3.8)
are given as follows
FW W (n) =
2
MW
2
MW
(n)
U (xt (n) , xu(n) ),
(B.1)
262
FW G(n) = 2
2 m
MW
t (n) mu(n)
6
MW
(n)

(u)

(t )
(t )
(u)
m1 ct (n) + m2 st (n) m1 cu(n) + m2 su(n)
#(xt (n) , xu(n) ),

U
M 2 mt (n) mu(n)
(t )
FW a(n) = 2 W 6
m3 ct (n)
MW (n)
(B.2)

(u)

(t )
(u)
+ m4 st (n) m3 cu(n) + m4 su(n)
#(xt (n) , xu(n) ),

U
1
2
FGa(n) =
FGG(n) =
Faa(n) =
MW
) (t )
m(t
1 m3
6
MW
(n)
(B.3)
) (t )
(u) (u)
(u)
+ m(t
m1 m3 + m(u)
2 m4
2 m4
U (xt (n) , xu(n) ),
(B.4)
2

) 2
(u) 2 (u) 2
1 MW
) 2
m(t
m1
U (xt (n) , xu(n) ),
+ m(t
+ m2
1
2
6
4 MW (n)
(B.5)
2

(t ) 2
(u) 2 (u) 2
1 MW
(t ) 2
m3
m3
U (xt (n) , xu(n) ).
+ m4
+ m4
6
4 MW (n)
(B.6)
# are defined as follows

The functions U and U
xt2 log xt
xu2 log xu
1
,
+
+
2
(xt xu )(1 xt )
(xu xt )(1 xu )2 (1 xu )(1 xt )
2xt log xt
1 + xt
U (xt , xt ) =
+
,
(1 xt )3 (1 xt )2
xt log xt
xu log xu
1
#(xt , xu ) =
+
+
U
,
2
2
(1 xu )(1 xt )
(xt xu )(1 xt )
(xu xt )(1 xu )
2
#(xt , xt ) = (1 + xt ) log xt +
.
U
3
(1 xt )
(1 xt )2
U (xt , xu ) =
(B.7)
(B.8)
(B.9)
(B.10)
Appendix C. Different contributions to Z 0 -penguin diagrams

The contributions of the diagrams in Fig. 7 to the functions F (xf ) in (4.12) are given
as follows

1 4
4
F1 (xf (n) ) =
cf (n) + sf4 (n) sw2
8
3

2
3
(C.1)
+ ln 2 + hq (xf (n) ) 2xf (n) hq (xf (n) ) ,
2
mf (n)

2
1
3
F2 (xf (n) ) = cf2 (n) sf2 (n) + ln 2 + hq (xf (n) ) + 2xf (n) hq (xf (n) ) ,
4
mf (n) 2
(C.2)
F3 (xf (n) ) =

(f ) 2 (f ) 2 2
4 2
m
s
+
m
c
f (n)
1
3
3 w
16m2W (n)

(f ) 2 (f ) 2 2
4
+ m2
sf (n) sw2
+ m4
3

2
1
+ ln 2 + hq (xf (n) ) 2xf (n) hq (xf (n) ) ,

mf (n) 2
F4 (xf (n) ) =
263
1
2
8MW
(n)
(f )
(f )
(f )
(f )
m1 m2 + m3 m4
(C.3)
cf (n) sf (n)

2
1
(C.4)
+ ln 2 + hq (xf (n) ) + 2xf (n) hq (xf (n) ) ,
mf (n) 2

3 2
1
2
F5 (xf (n) ) = cw
(C.5)
xf (n) hw (xf (n) ) ,
+ ln 2
4
MW (n) 6

2

2
(f ) 2
1
(f ) 2
2
2 n
F6 (xf (n) ) =
1 2sw MW + 2cw 2
m1
+ m2
4
R
16mW (n)

2

(f ) 2 (f ) 2

n
2
2
MW
+ m4
+ 1 2sw2 2 + 2cw
m3
R

2
1
(C.6)
+ ln 2
+ xf (n) hw (xf (n) ) ,
MW (n) 2
MW n (f ) (f )
(f ) (f )
m1 m3 + m2 m4
4
R
8MW (n)

2
1
+ ln 2
+ xf (n) hw (xf (n) ) ,
MW (n) 2

2
mf (n)
(f )
(f )
2
2
2 n
F8 (xf (n) ) =
M
c
c
+
m
s
s
m
f (n)
f (n)
w W
w 2
1
2
4
R
2MW
(n)

n (f )
(f )
+ MW
m3 cf (n) + m4 sf (n) hw (xf (n) ).
R
F7 (xf (n) ) =
(C.7)
(C.8)
Here
=
1
+ ln 4 E ,
M
D = 4 2M
(C.9)
and the functions hq and hw are given by:

ln x
1
+
,
1 x (1 x)2
x ln x
1
+
hw (x) =
.
1 x (1 x)2
hq (x) =
(C.10)
(C.11)
264
Fig. 12. Self-energy diagrams necessary for calculating the electroweak counter term as discussed in [50].
Finally, the contributions from counter terms corresponding to the self-energy diagrams of
Fig. 12 that should be added to the functions Fi are given by

2
2
MW
1 1 + xf (n) 2xf (n) ln xf (n)
1
(n)
S1 (xf (n) ) =

(C.12)
ln
,
+
4
2 1 xf (n)
(1 xf (n) )2
2

2
2
MW
1
1 1 3xf (n) 2xf (n) ln xf (n)
(n)
S2 (xf (n) ) = (1 + xf ) +
ln
.
8
2 1 xf (n)
(1 xf (n) )2
2
(C.13)
Appendix D. Closed form for the C(xt , 1/R) function

We express the logarithms of (4.14) as integrals I1 and I2
xn + xt
= I1 (xt ) I1 (1) + xt 1,
1 + xn
xn + xt
ln
= I2 (xt ) I2 (1)
1 + xn
xn ln
(D.1)
(D.2)
with
1
I1 (a) = a
dy
0
1
I2 (a) = a
dy
0
a 2y
y
,
+ xn
1
.
ay + xn
(D.3)
(D.4)
Next, we interchange integration and summation. The integrands can now be summed
using the relation

b c coth( c ) 1
b
(D.5)
.
=
c + n2
2c
n=1
265
This allows us to derive a closed form for the sum
Cn (xt , xn )
n=1
mW Rxt
(xt 7)xt
16(1 xt ) 16(1 xt )2
1

3(1 + xt )

1/2
coth mW R y xt coth mt R y
dy
y
0

3/2
+ (xt 7) y coth mW R y xt coth mt R y
.
(D.6)
Expanding in (mW R y, mt R y ) and integrating (D.6) leads to the expression in

(4.17).
Appendix E. Different contributions to F = 1 box diagrams
The one-loop amplitude of the KK excitations in the diagrams in Fig. 9 is a sum of
contributions coming from the various bosonic fields in the loop
Gn (xf (n) , xe(n) ) = GW W (n) + GW G(n) + GW a(n) + GGa(n)
+ GGG(n) + Gaa(n) ,
(E.1)
Hn (xf (n) , x(n) ) = HW W (n) + HW G(n) + HW a(n) + HGa(n)

+ HGG(n) + Haa(n).
(E.2)
The explicit results for Bn (xt , xn ) and Bn (xt , xn ) are given by

Bn (xt , xn ) = Gn (xt (n) , xe(n) ) Gn (xu(n) , xe(n) ),
Bn (xt , xn ) = Hn (xt (n) , x(n)) Hn (xu(n) , x(n) ),
(E.3)
where the results for the -box diagrams take the form
GW W (n) =
2
MW
2
MW
(n)
U (xf (n) , xe(n) ),
(E.4)
GW G(n) =
2 m
(e)
1 MW
f (n) me(n)
(f )
(f )
#(xf (n) , xe(n) ),
m1 cf (n) + m2 sf (n) m1 U
6
2
MW (n)
(E.5)
GW a(n) =
2 m

1 MW
f (n) me(n)
(f )
(f )
#
m3 cf (n) + m4 sf (n) m(e)
3 U (xf (n) , xe(n) ),
6
2
MW (n)
(E.6)
GGa(n) =
1 MW
(f ) (f )
(f ) (f ) (e) (e)
m1 m3 + m2 m4 m1 m3 U (xf (n) , xe(n) ),
6
8 MW (n)
(E.7)
266
GGG(n) =
Gaa(n) =
2

(f ) 2 (e) 2
1 MW
(f ) 2
m1
m1 U (xf (n) , xe(n) ),
+ m2
6
16 MW
(n)
2

(f ) 2 (e) 2
1 MW
(f ) 2
m3
m3 U (xf (n) , xe(n) ),
+ m4
6
16 MW (n)
(E.8)
(E.9)
and for the -box
diagrams we get
HW W (n) =
2
1 MW
U (xf (n) , x(n) ),
2
4 MW
(n)
(E.10)
HW G(n) =
2 m
()
1 MW
f (n) m(n)
(f )
(f )
#(xf (n) , x(n) ),
m1 cf (n) + m2 sf (n) m1 U
6
2
MW (n)
(E.11)
HW a(n) =
2 m

1 MW
f (n) m(n)
(f )
(f )
#
m3 cf (n) + m4 sf (n) m()
3 U (xf (n) , x(n) ),
6
2
MW (n)
(E.12)
HGa(n) =
1 MW
(f ) (f ) () ()
(t ) (f )
m1 m3 + m2 m4 m1 m3 U (xf (n) , x(n) ),
6
8 MW
(n)
(E.13)
2

(f ) 2 () 2
1 MW
(f ) 2
m1
m1 U (xf (n) , x(n)),
+ m2
6
16 MW (n)
(E.14)
HGG(n) =
Haa(n) =
2

(f ) 2 () 2
1 MW
(f ) 2
m3
m3 U (xf (n) , x(n) ).
+ m4
6
16 MW (n)
(E.15)
# are defined in (B.7)(B.10). We have taken into account the overall

The functions U and U
minus sign in (4.8).
Summing all the contributions we find
17xt + 18xn xt 9xn xn (26xt + 9xn + 18xn xt )
xn
+
ln
16(xt 1)
16xt
1 + xn
(xn + xt )(9xn + 17xt ) xn + xt
ln
,
1 + xn
16(xt 1)2 xt
xn (3xn + 6xn xt 2xt )
3xn + 5xt 6xn xt
xn
Bn (xt , xn ) =
ln
16(xt 1)
16xt
1 + xn
(3xn 5xt )(xn + xt ) xn + xt
+
ln
.
1 + xn
16(xt 1)2 xt
Bn (xt , xn ) =
References
[1] T. Appelquist, H.-C. Cheng, B.A. Dobrescu, Phys. Rev. D 64 (2001) 035002, hep-ph/0012100.
[2] K. Agashe, N.G. Deshpande, G.H. Wu, Phys. Lett. B 514 (2001) 309, hep-ph/0105084.
[3] K. Agashe, N.G. Deshpande, G.H. Wu, Phys. Lett. B 511 (2001) 85, hep-ph/0103235;
T. Appelquist, B.A. Dobrescu, Phys. Lett. B 516 (2001) 85, hep-ph/0106140.
[4] C. Macesanu, C.D. McMullen, S. Nandi, Phys. Rev. D 66 (2002) 015009, hep-ph/0201300.
(E.16)
(E.17)
[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]
267
H.C. Cheng, K.T. Matchev, M. Schmaltz, Phys. Rev. D 66 (2002) 056006, hep-ph/0205314.
T.G. Rizzo, Phys. Rev. D 64 (2001) 095010, hep-ph/0106336.
F.J. Petriello, JHEP 0205 (2002) 003, hep-ph/0204067.
G. Servant, T.M. Tait, hep-ph/0206071.
H.C. Cheng, J.L. Feng, K.T. Matchev, hep-ph/0207125;
D. Hooper, G.D. Kribs, hep-ph/0208261;
G. Servant, T.M. Tait, hep-ph/0209262;
D. Majumdar, hep-ph/0209277;
G. Bertone, G. Servant, G. Sigl, hep-ph/0211342.
T. Appelquist, H. Yee, hep-ph/0211023.
A.J. Buras, A. Poschenrieder, M. Spranger, A. Weiler, in preparation.
S.L. Glashow, J. Iliopoulos, L. Maiani, Phys. Rev. D 2 (1970) 1285.
J. Papavassiliou, A. Santamaria, Phys. Rev. D 63 (2001) 016002;
J.F. Oliver, J. Papavassiliou, A. Santamaria, hep-ph/0209021.
A.J. Buras, P. Gambino, M. Gorbahn, S. Jger, L. Silvestrini, Phys. Lett. B 500 (2001) 161.
T. Inami, C.S. Lim, Prog. Theor. Phys. 65 (1981) 297.
A.J. Buras, W. Slominski, H. Steger, Nucl. Phys. B 238 (1984) 529;
A.J. Buras, W. Slominski, H. Steger, Nucl. Phys. B 245 (1984) 369.
G. Buchalla, A.J. Buras, M.K. Harlander, Nucl. Phys. B 349 (1991) 1.
D. Chakraverty, K. Huitu, A. Kundu, hep-ph/0212047.
A. Muck, A. Pilaftsis, R. Ruckl, hep-ph/0210410;
A. Muck, A. Pilaftsis, R. Ruckl, Phys. Rev. D 65 (2002) 085037, hep-ph/0110391.
H. Georgi, A.K. Grant, G. Hailu, Phys. Rev. D 63 (2001) 064027, hep-ph/0007350.
J. Giedt, hep-ph/0204315.
H.C. Cheng, K.T. Matchev, M. Schmaltz, Phys. Rev. D 66 (2002) 036005, hep-ph/0204342.
A.J. Buras, M. Jamin, P.H. Weisz, Nucl. Phys. B 347 (1990) 491.
J. Urban, F. Krauss, U. Jentschura, G. Soff, Nucl. Phys. B 523 (1998) 40.
A.J. Buras, Lectures at the International Erice School, August, 2000, hep-ph/0101336.
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47.
S. Herrlich, U. Nierste, Nucl. Phys. B 419 (1994) 292, and U. Nierste, recent update.
S. Herrlich, U. Nierste, Phys. Rev. D 52 (1995) 6505;
S. Herrlich, U. Nierste, Nucl. Phys. B 476 (1996) 27.
L. Wolfenstein, Phys. Rev. Lett. 51 (1983) 1945.
A.J. Buras, M.E. Lautenbacher, G. Ostermaier, Phys. Rev. D 50 (1994) 3433.
LEP Working group on oscillations: http://lepbosc.web.cern.ch/LEPBOSC/combined_results/
amsterdam_2002/.
L. Lellouch, hep-ph/0211359;
D. Becirevic, hep-ph/0211340.
A.A. Penin, M. Steinhauser, Phys. Rev. D 65 (2002) 054006;
M. Jamin, B.O. Lange, Phys. Rev. D 65 (2002) 056005;
K. Hagiwara, S. Narison, D. Nomura, hep-ph/0205092.
A.J. Buras, F. Parodi, A. Stocchi, hep-ph/0207101.
B. Aubert, et al., BaBar Collaboration, hep-ex/0207042.
K. Abe, et al., Belle Collaboration, hep-ex/0208025.
Y. Nir, hep-ph/0208080.
G. DAmbrosio, G.F. Giudice, G. Isidori, A. Strumia, hep-ph/0207036.
M. Ciuchini, G. DAgostini, E. Franco, V. Lubicz, G. Martinelli, F. Parodi, P. Roudeau, A. Stocchi,
JHEP 0107 (2001) 013, hep-ph/0012308.
A. Stocchi, hep-ph/0211245.
A.J. Buras, hep-ph/0210291, and references therein.
A. Hcker, H. Lacker, S. Laplace, F. Le Diberder, Eur. Phys. J. C 21 (2001) 225, http://ckmfitter.in2p3.fr.
Provided by F. Parodi and A. Stocchi.
268
[44] G. Buchalla, A.J. Buras, Phys. Lett. B 333 (1994) 221;

G. Buchalla, A.J. Buras, Phys. Rev. D 54 (1996) 6782.
[45] Y. Grossman, Y. Nir, Phys. Lett. B 398 (1997) 163.
[46] G. Buchalla, A.J. Buras, Nucl. Phys. B 400 (1993) 225.
[47] M. Misiak, J. Urban, Phys. Lett. B 541 (1999) 161.
[51] J.F. Oliver, J. Papavassiliou, A. Santamaria, hep-ph/0212391.
[52] K. Hagiwara, et al., Phys. Rev. D 66 (2002) 010001.
[53] W. Marciano, Z. Parsa, Phys. Rev. D 53 (1996) R1.
[54] N. Cabibbo, L. Maiani, Phys. Lett. B 79 (1978) 109.
[55] C.S. Kim, A.D. Martin, Phys. Lett. B 225 (1989) 186.
[56] G. DAmbrosio, G. Isidori, J. Portols, Phys. Lett. 423 (1998) 385;
G. Isidori, A. Retico, JHEP 0209 (2002) 063.
[57] S. Adler, et al., Phys. Rev. Lett. 79 (1997) 2204;
S. Adler, et al., Phys. Rev. Lett. 84 (2000) 3768, hep-ex/0111091.
[58] G. DAmbrosio, G. Isidori, hep-ph/0112135.
[59] D. Becirevic, S. Fajfer, S. Prelovsek, J. Zupan, hep-ph/0211271.
[60] A.J. Buras, R. Fleischer, Phys. Rev. D 64 (2001) 115010;
A.J. Buras, R. Buras, Phys. Lett. 501 (2001) 223;
S. Bergmann, G. Perez, Phys. Rev. D 64 (2001) 115009;
S. Bergmann, G. Perez, JHEP 0008 (2000) 034;
S. Laplace, Z. Ligeti, Y. Nir, G. Perez, Phys. Rev. D 65 (2002) 094040.
[61] A.J. Buras, P.H. Chankowski, J. Rosiek, . Sawianowska, Nucl. Phys. B 619 (2001) 434;
A.J. Buras, P.H. Chankowski, J. Rosiek, . Sawianowska, Phys. Lett. 546 (2002) 96, hep-ph/0210145.
[62] A.J. Buras, P. Gambino, M. Gorbahn, S. Jger, L. Silvestrini, Nucl. Phys. B 592 (2001) 55.
[63] A.J. Buras, L. Silvestrini, Nucl. Phys. B 546 (1999) 299.
[64] G. Buchalla, G. Hiller, G. Isidori, Phys. Rev. D 63 (2001) 014015.

Double spin asymmetries for large-pT

hadron production in semi-inclusive DIS
Yuji Koike, J. Nagashima
Department of Physics, Niigata University, Ikarashi, Niigata 950-2181, Japan
Received 13 February 2003; accepted 25 March 2003
Abstract
We study the twist-2 double-spin asymmetries for the 2-jets and large-pT hadron (, , etc.)
production in semi-inclusive deep inelastic scattering to O(s ) in perturbative QCD. After deriving
the complete set of the polarized cross-section which is differential with respect to the transverse
momentum, we discuss characteristic features of the azymuthal spin asymmetries, using existing
parton densities and fragmentation functions at COMPASS and EIC energies.
1. Introduction
High-energy experiment with polarized beams and targets has opened a new window
for revealing QCD dynamics and hadron structures. Ongoing RHIC-SPIN, HERMES and
COMPASS experiments are going to provide us with a variety of data disclosing spin
distributions inside the nucleon. The planed electron ion collider (EIC) at BNL is expected
to be a more powerful and sensitive tool for the QCD spin physics.
In this paper, we study the hadron production in polarized semi-inclusive deep inelastic
scattering (SIDIS) off proton. The cross-section formula with integrated transverse
momentum of the final hadron, pT , has been derived in the leading order (LO) [1] and
in the next-to-leading order (NLO) [2] and has been applied to predict spin asymmetries
for polarized production in SIDIS [3]. We extend these studies to derive cross-section
formula for the complete set of the polarized processes which is differential in pT and is
suited, in particular, for the large-pT hadron production. This study should be useful to
get further insight into the parton densities and fragmentation functions from HERMES,
E-mail address: koike@nt.sc.niigata-u.ac.jp (Y. Koike).
doi:10.1016/S0550-3213(03)00259-1
270
Y. Koike, J. Nagashima / Nuclear Physics B 660 (2003) 269288
COMPASS and future EIC experiment. It is also complementary to studies on largepT hadron production in pp collisions at RHIC and HERA-N [46]. As typical SIDIS
processes, we study pion (or 2-jets) and hyperon production. In the leading twist-2 level,
the complete set of the processes is
(i)
e + p e + (pT ) + X
(ii)
T ) + X,
e + p e + (p
(iii)
e + p e + (pT ) + X,
(iv) e + p e + (pT ) + X
or e + 2-jets + X,
or e + 2-jets + X,
T ) + X,
(v) e + p e + (p
(1.1)
where we have used the notation p and for longitudinally polarized, and p and for
transversely polarized hadrons, and we restrict ourselves to electron (or muon) scattering
off proton. In order that a final hadron carries large transverse momentum, another parton
has to be emitted in the opposite direction. This is an O(s ) effect in perturbative QCD
and was investigated in [7,8] for the unpolarized SIDIS with electron and neutrino beams.
Here we extend the analysis to the polarized cases shown above. For small-pT production
in SIDIS, the origin of pT may be ascribed to the nonperturbative intrinsic transverse
momentum of partons confined in or fragmenting into a hadron. Derivation of the crosssection formula based on this idea was performed in [9], which is complementary to the
present study.
As will be shown below, our cross-section formula for (i)(v) applicable in the largepT region diverges as 1/pT2 as pT 0, which is a manifestation of soft and collinear
divergence. For the pT -integrated cross-section, the soft divergence is canceled by the
soft divergence in the virtual correction diagrams and the collinear one is factorized into
the parton densities. Accordingly, O(s ) cross-section at pT 0 is to be interpreted as
distribution with respect to the variable pT . In order to get a finite pT -differential crosssection, one needs to resume the effect of soft gluon radiation as was studied in [10,11].
Derivation of the resummed cross-section for a low-pT region is beyond the scope of the
present study and will be presented elsewhere.
This paper is organized as follows. In Section 2 we present the formalism to calculate
the cross-sections for (1.1) following [8]. Section 3 presents the final analytic formula for
the cross-section. In Section 4, we present a numerical estimate of the spin asymmetry
and discuss its characteristic features, using existing parton distribution and fragmentation
functions in the literature. Section 5 is devoted to summary and conclusion.
2. Formalism
In this paper we are interested in the cross-section for the process,
e(k) + A(pA , SA ) e (k ) + B(pB , SB ) + X.
(2.1)
271
We define 5 Lorentz invariants for specifying the kinematics. The center of mass energy
Sep for the initial electron and the proton A is
Sep = (pA + k)2 2pA k,
(2.2)
where we ignored the masses. Conventional DIS variables

xbj =
Q2
,
2pA q
Q2 = q 2 = (k k )2 ,
(2.3)
are determined by observing the final electron. For the description of kinematics of the
final hadron B, we introduce
pA pB
,
zf =
(2.4)
pA q
and the transverse component of q which is orthogonal to pA and pB ,
pB q
pA q
pA
p .
qt = q
pA pB
pA pB B
qt is a space-like vector and we define its magnitude as

qT = qt2 .
(2.5)
(2.6)
Differential cross-section for the semi-inclusive production (2.1) can be written as

d =
1
d 3 pB
d 3 k e4
L (k, k )W (pA , pB , q),
2Sep (2)3 2pB0 (2)3 2k 0 q 4
(2.7)
where we use the covariant normalization for each state and 1/2Sep is the initial flux. L
is the leptonic tensor defined as

1
Tr (1 5 )/k k/
2
= 2(k k + k k ) g Q2 2i$ k k ,
L (k, k ) =
(2.8)
with our convention for the anti-symmetric tensor $0123 = $ 0123 = 1. The leading twist-2
contribution to the hadronic tensor W can be written as
W (pA , pB , q)
1
1

dx
dz
B (z, pB , SB )H (xpA , pB /z, q)
=
Tr MA (x, pA , SA )M
2
x
z
0

2 (xpA + q pB /z)2 ,
(2.9)
B (z, pB , SB ) are the distribution and fragmentation functions

where MA (x, pA , SA ) and M
for the hadrons A and B, respectively, and H (xpA , pB /z, q) is the corresponding hard
part. Tr indicates the trace over relevant spinor or Lorentz indices. Since we are interested
in the twist-2 cross-sections, pA and pB can be regarded as light-like. To define the
272
complete set of distribution and fragmentation functions, we introduce another set of lightlike vectors n (n2 = 0) and w (w2 = 0) for pA and pB , respectively, by the relation
pA n = 1 and pB w = 1. The complete set of the twist-2 quark distribution for the
proton A is defined as [12,13]

d ix
q
e pA SA | j (0)i (n)|pA SA
MAij (x, pA , SA ) =
2
1
1
1
/ A q(x) + A 5p
/ A q(x) + 5S/ Ap
/ A q(x) + , (2.10)
= p
2
2
2
2 =1
where the transverse spin vector SA = (0, SA ) satisfies SA pA = 0 and 2A + SA
with A the longitudinal component of the spin vector, and + stands for twist-3 or
higher. q(x), q(x) and q(x) are, respectively, spin-average, longitudinally polarized and
transversely polarized (transversity) quark distribution. q is chiral-odd and hence does not
mix with gluon distributions. The gluon distribution is also defined as [12,14,15]

2
d ix
g
e pA SA | tr n G (0)n G (n) |pA SA
MA (x, pA , SA ) = 2
x
2
1
1
= G(x)gA G(x)i$ pA n + ,
(2.11)
2x
2x
where G is the gluons field strength, tr means the trace over the color index for
n n p . G(x) and G(x) are, respectively, spin-average and

G , gA = g pA
A
longitudinally polarized gluon distributions. Notice there is no transversely polarized gluon
distribution. Similarly we define twist-2 quark and gluon fragmentation functions for the
spin-1/2 hadron B [12,16]:

1 d i/z
q

e
0|i (0)|BX BX| j (w)|0
MBij (z, pB , SB ) =
Nc
2
X
1
1
1
/ B q(z)
+ B 5p
/ B q(z)
+ 5S/ Bp
/ B q(z)
+ ,
= p
z
z
z
(2.12)
g (z, pB , SB )
M
B

2z2 d i/z
e
= 2
0| tr w G (0)|BX BX|w G (w) |0
Nc 1
2
X

pB w + ,
= G(z)g
B G(z)i$
(2.13)
where the transverse spin vector SB = (0, SB ) in the quark fragmentation function
2 = 1, and g = g p w w p . Physical
satisfies SB pB = 0 and 2B + SB
B
B
B
meaning of each fragmentation function is in parallel with the distribution functions
defined above, and we specify them by putting on the corresponding distributions.
For the actual calculation of the cross-section, it is convenient to work in the hadron
frame [8]. In this frame q and pA take
q = (0, 0, 0, Q),

Q
Q
pA =
.
, 0, 0,
2xbj
2xbj
(2.14)
(2.15)
Choosing the xz-plane as the hadron plane, pB can be written as

q 2 2qT
q2
zf Q
1 + T2 ,
, 0, T2 1 .
pB =
2
Q
Q
Q
273
(2.16)
In order to write the lepton momentum in the hadron frame, we need to introduce the
azymuthal angle between the hadron plane and the lepton plane in the hadron frame.
Then the lepton momentum can be parametrized as
Q
(cosh , sinh cos , sinh sin , 1),
2
and k = k q with
k =
cosh =
2xbj SeA
1.
Q2
(2.17)
(2.18)
With these definitions, the cross-section for (2.1) can be expressed in terms of Sep , xbj , Q2 ,
zf , qT2 and in the hadron frame. Obviously, is invariant under boost in the q direction,
so that the in the hadron frame is the same, for example, in the center-of-mass system of
the virtual photon and the initial proton A. The expression for these variables in terms of
the lab. frame variables is given in the Appendix A.
Let us proceed to calculate the cross-section. As is shown in (2.16) the transverse
momentum of pB is pT = zf qT in the hadron frame, and we shall derive the cross-section
formula differential in qT2 . Lowest-order diagram in the QCD coupling relevant for the
processes (1.1) is shown in Fig. 1. We follow the method of [8] for the calculation. To
isolate the -dependence in the cross-section, we first expand the hadronic tensor W in
terms of the complete set of the independent tensors. To this end we introduce the following
4 vectors which are orthogonal to each other [8]:
1

q + 2xbj pB ,
Q

q2
1 pB
X =
q 1 + T2 xbj pA ,
qT zf
Q
Y =$
Z X T ,
q
Z = .
Q
T =
(2.19)
These vectors take the form of T = (1, 0, 0, 0), X = (0, 1, 0, 0), Y = (0, 0, 1, 0),
Z = (0, 0, 0, 1) in the hadron frame, which greatly facilitates the actual calculation
presented below. Note that T , X, and Z are vectors, while Y is an axial vector. Since
W satisfies the current conservation, q W = q W = 0, it is easy to see that W
can be expanded in terms of 9 independent tensors, for which we employ the following:
V2 = g + Z Z ,
V4 = X X Y Y ,

V6 = i X Y Y X ,
V1 = X X + Y Y ,
V3 = T X + X T ,

V5 = i T X X T ,
274
(a)
(b)
(c)
Fig. 1. O(s ) cut diagrams contributing to large-pT hadron production in SIDIS. Distribution function is located
in the lower part of each diagram, and show the fragmentation function insertion.

V7 = i T Y Y T ,
V8 = T Y + Y T ,
V9 = X Y + Y X .
(2.20)
W
in terms of these tensors is easily obtained by the

The expansion coefficients of
inverse tensors
Vk for Vk (k = 1, . . . , 9) as
W =
9

Vk
W
Vk
(2.21)
k=1
where

= 1 2T T + X X + Y Y ,
V
1
2
= T T ,
V
2

= 1 T X + X T ,
V
3
2

i
= T X X T ,
V
5
2

= i T Y Y T ,
V
7
2

1
= X Y + Y X .
V
9
2
275

= 1 X X Y Y ,
V
4
2

i
=
V
X Y Y X ,
6
2

= 1 T Y + Y T ,
V
8
2
(2.22)
Note that Vk (k = 1, . . . , 4, 8, 9) are symmetric between and , and the rest are anti
symmetric. Vk (k = 1, . . . , 5) are even under parity, while the rest are odd under parity.
For the processes with unpolarized electrons, i.e., (i)(iii) in (1.1), both L and W are
symmetric tensors, accordingly Vk (k = 1, . . . , 4) contributes to W . For (iv) and (v)

both L and W are anti-symmetric pseudo-tensors due to a 5 in the trace, accordingly
] in (2.21) involves
only Vk (k = 6, 7) contribute to W . Actual calculation of [W V
k
trace over many -matrices, which can be carried out conveniently in the hadron frame by
using the Tracer program.
With these Vk , we can decompose the -dependence of the cross-sections by the

contraction with L in (2.8). Define Ak (k = 1, . . . , 9) as
Ak =
L Vk
Q2
(2.23)
then one obtains

A1 = 1 + cosh2 ,
A2 = 2,
A3 = cos sinh 2,
A4 = cos 2 sinh2 ,
A6 = 2 cosh ,
A7 = 2 cos sinh .
(2.24)
From (2.24), the cross-section for (i), (ii), and (iii) have -dependence of 1, cos , and
cos 2, while (iv) and (v) have only 1 and cos dependence.
For the case of (iii) e + p e + + X, we need to parametrize the spin vector SA
and SB which lie in the plane orthogonal to pA and pB , respectively. We write
SA = cos A X + sin A Y = (0, cos A , sin A , 0),
SB
= cos B cos B X + sin B Y sin B cos B Z
= (0, cos B cos B , sin B , sin B cos B ),
(2.25)
(2.26)
where the second equality in each equation follows only in the hadron frame. In the hadron
frame, A,B are the azymuthal angles of SA,B measured from the hadron plane, and
B is the polar angle of pB measured from the Z-axis which has the Lorentz invariant
276
expression
cos B =
qT2 Q2
qT2
+ Q2
sin B =
2qT Q
qT2 + Q2
(2.27)
In the Appendix A, we present the expression for A and B in terms of lab. frame
variables.
3. Analytic formula for cross-section

By applying the method described in the previous section, we finally obtain the crosssection for the large pT hadron production in DIS in the following form:
1
1

d 5
e2 s
dx
dz
[f D k ]
=
Ak
2
2
2
2
2
x
z
dxbj dQ dzf dqT d 8xbj Sep Q k

qT2
Q2
xmin
1
1
x
zf
1
1
z
(3.1)
where Ak is defined in (2.24), e = e2 /4 is the QED coupling constant, and we

introduced the variables
xbj
zf
x =
(3.2)
,
z = ,
x
z
and

zf qT2
xmin = xbj 1 +
(3.3)
.
1 zf Q2
For a given Sep , Q2 , and qT , the kinematic constraint for xbj and zf is
Q2
< xbj < 1,
Sep
0 < zf <
(3.4)
1 xbj
1 xbj + xbj qT2 /Q2
For a given Q2 , xbj and zf , qT can take

1
1
1
1 .
0 < qT < Q
xbj
zf
(3.5)
(3.6)
Roughly speaking, 0 < qT < Q corresponds to the current fragmentation region and qT >
Q corresponds to the target fragmentation region. As is shown in (2.16), the transverse
momentum of pB in the hadron frame is

1
1
pT = zf qT < zf Q
(3.7)
1
1 .
xbj
zf
277
In (3.1), [f D k ] should read the following form for each process in (1.1)
(CF = 4/3):
(i) e + p e + {, } + X, [7]:

qq
gq
[f D k ] =
eq2 q(x)q(z)
k +
eq2 G(x)q(z)
k
q
qg ,
eq2 q(x)G(z)
k
(3.8)
where
qq
1
1
= 2CF x z
2
Q qT2
qq

2
Q4
2 2
+6 ,
+ Q qT
x 2 z 2
qq
2 = 2 4 = 8CF x z ,

1 2
qq
3 = 4CF x z
Q + qT2 ,
QqT
(3.9)

2 2
Q2
2
1
+ 2 + 10
= x(1
x)
,
x
z
qT2 x 2 z 2 x z

gq
1
gq
gq
x),
2 = 2 4 = 8x(1

Q2
2
gq
3 = x(1
,
2 Q2 + qT2
x)
QqT
x z

qg
1
1
= 2CF x(1
z )
2
Q qT2
qg
(3.10)

2 q 2
2
z
Q4
(1 z )2
T
+6 ,
Q2
+
x 2 z 2
z 2
(1 z )2
qg
2 = 2 4 = 8CF x(1
z ),

z 2 qT2
qg
2 1
2
z )
3 = 4CF x(1
Q +
.
z QqT
(1 z )2
(ii) e + p e + + X:
[f D k ] =

q
qq
eq2 q(x)q(z)
k +
L
(3.11)
gq
eq2 G(x)q(z)
k
L
q
qg

eq2 q(x)G(z)
k ,
L
(3.12)
where
qq
qq
L k = k
gq
L 1 =
(k = 1, 2, 3, 4),
(2x 1){Q4 (x 1)2 qT4 x 2 }

Q2 qT2 x(
x 1)
(3.13)
,
278

gq
gq
L 2 = L 4 = 0,
gq
L 3 =
2{Q2 (x 1) qT2 x}
,
QqT

qg
L 1 = 2CF x z
(3.14)

x + 1) qT4
x 2 x(
2(2x 2 2x + 1) qT2
+
,
+
x 1 (x 1)2 Q4
(x 1)2
Q2
x 2 z qT2
,
x 1 Q2

x 2 qT2 qT
4CF x z
qg
2
L 3 =
.
(
x
1)
+
(x 1)2
Q2 Q
qg
qg
L 2 = 2L 4 = 8CF
(3.15)
(iii) e + p e + + X:

qq
[f D k ] =
eq2 q(x) q(z)
k ,
T
(3.16)
where
qq
T 1 = 4CF cos(A B ),
qq
T 3 = 4CF
Q
cos(A B ),
qT
qq
T 2 = 0,
qq
T 4 = 4CF
Q2
cos(A B ).
qT2
(3.17)
(iv) e + p e + {, } + X:

qq
gq
[f D k ] =
eq2 q(x)q(z)
k +
eq2 G(x)q(z)
k
LO
LO
q
q
qg

eq2 q(x)G(z)
LO k ,
(3.18)
where

qq
LO 6 = 2CF
qq
2

x z qT2
Q
1
+ x z
,
x z
Q2
qT2
Q2 qT2
,
QqT

x 1 Q2
2x 1
2x + 2
,
=
x
z qT2
2Q (x 1)(2z 1)
,
=
qT
z
LO 7 = 4CF x z
gq
LO 6
gq
LO 7

qT4
2CF z 1
x 4
2
,
=
(x 1) +
x 1 z 2
(x 1)2 Q4

4CF x z
x qT
qg
.
LO 7 =
1
x 1
z Q
(3.19)
(3.20)
qg
LO 6
(3.21)
(v) e + p e + + X:
[f D k ] =

q
qq
eq2 q(x)q(z)
k +
OL
279
gq
eq2 G(x)q(z)
k
OL
q
qg

eq2 q(x)G(z)
OL k ,
(3.22)
where
qq
qq
OL 6,7 = LO 6,7 ,
gq
OL 6
gq
(3.23)
Q2
2x 2 2x + 1
x + (x 1) 2 ,
=
x z
qT
OL 7 =
2Q (x 1)(2x 1)
,
qT
z

qT4
2CF z 1
x 4
2
,
=
+ (x 1)
x 1 z 2
(x 1)2 Q4

4CF x z
x qT
qg
OL 7 =
.
1
x 1
z Q
(3.24)
qg
OL 6
(3.25)
Several comments are in order here:

(1) From the above formula one can separate the -dependence of the qT -differential
cross-section for (i)(v) in (1.1) as
dQ2 dx
d
= 0 + cos()1 + cos(2)2 ,
2
bj dzf dqT d
(3.26)
with 2 0 for (iv) and (v). Transverse momentum of partons also gives the same
dependence as a purely kinematic effect [17]. It is, however, interesting to test whether
the above 0(s ) formula gives quantitatively right magnitude for each component.
(2) The cross-section depends on Sep through Ak in (2.24). For xbj Sep /Q2 1
(see (2.18)), we have A1,3,4 A6,7 (-dependence factored out), which results in
the strong suppression of the asymmetry for (iv) and (v) compared to (ii) and (iii) in
(1.1).
(3) In the large-qT SIDIS, contribution from the gluon distribution and fragmentation
functions is of the same O(s ) effect as the quark contribution, so that qT -differential
cross-section is expected to be a more sensitive tool to determine detailed form of the
polarized gluon contribution than the qT -integrated case [8].
(4) Except for (iii), 0 has a nonintegrable 1/qT2 -dependence at qT 0, while 1,2 are
integrable. For the case of transverse polarization (iii), 2 has the 1/qT2 -dependence,
while 0,1 is integrable at qT 0. To get a finite meaningful result for all value of qT ,
one needs to include higher-order effect based on the resummation technique as in the
case of the spin-average case [10,11].
280
4. Numerical estimate
4.1. Azymuthal asymmetry
In this section, we present a numerical estimate of the polarized cross-sections (ii)(v)
in (1.1) in comparison with the spin averaged one (i), using the existing parton distributions
for the nucleon and the fragmentation function for pion and . For this purpose we
consider the azymuthal asymmetries defined as follows:
2
d 5 pol
pol
0 d dQ2 dxbj dzf dq 2 d
0
T
= av ,
1SA SB 2
d 5 av
0
d
2
0
cos(n) S
dQ2 dxbj dzf dqT d
2
A SB
5 av,pol
av,pol
d cos(n) 2 d
n
dQ dxbj dzf dqT2 d
=
2
5
av
d
20av
2
2
0 d
(n = 1, 2),
(4.1)
dQ dxbj dzf dqT d
where av and pol are the spin-averaged and polarized cross-sections and their
av,pol
decomposition 0,1,2 is defined in (3.26). The subscript SA SB in the left-hand side of (4.1)
specify the spin states of the hadrons A and B for each process in (1.1). We use the symbol
O for spin-average, L for longitudinal polarization and T for transverse polarization.
Accordingly, SA SB in (4.1) can be SA SB = OO, LL, T T , LO, OL corresponding to (i),
(ii), (iii), (iv), and (v), respectively. By definition 1OO 1. For all cases, | 1SA SB | < 1
and | cos(n)SA SB | < 1.
The asymmetries (4.1) are still functions of 5 variables, Sep , Q2 , xbj , zf , and qT .
We estimate them at COMPASS (Sep = 300 GeV2 ) and EIC (Sep = 104 GeV2 ) energies
with typical kinematic variables where our perturbative formula are valid. Comparison
of predicted asymmetries with experimental data may be better achieved by integrating
over some of the variables to gain statistics. Here we will not try this procedure, but will
show the nonintegrated azymuthal asymmetry, intending to show a typical dependence
of the asymmetry on each variable. To determine the kinematic variables, we first note
from (2.18) and (2.24) that at large cosh = 2xbj Sep /Q2 1, the spin asymmetry for
(iv) and (v) is strongly suppressed compared with (ii) and (iii), so that a smaller value
of cosh is prefered to get large asymmetry. In addition, if one chooses the same
values of Q2 and cosh at COMPASS and EIC energies, variation of the asymmetry
at different Sep is wholly ascribed to the variation of distribution and/or fragmentation
functions and the partonic hard cross-sections at different xbj . Keeping this in mind we
have chosen (Q2 , xbj ) = (100 GeV2 , 0.4) for the COMPASS energy and (Q2 , xbj ) =
(100 GeV2 , 0.012) for the EIC energy, by which valence and sea region of the parton
densities are mainly probed.
As a reference distribution and fragmentation functions, we use GRV parton density for
the unpolarized nucleon [18], GRSV polarized parton density [19], KKP fragmentation
function for + + [20] and FSV -fragmentation function [3] for polarized and
In all cases we use NLO versions, putting the factorization scale
unpolarized + .
2F = Q2 .
281
Fig. 2. Azymuthal spin asymmetry for the pion production e + e + + X. (a) and (b) show
the qT -dependence of the asymmetry at COMPASS and EIC energies, respectively. (c) and (d) show the
zf -dependence of the asymmetry at COMPASS and EIC energies, respectively. Solid and dashed lines,
respectively, denote those obtained with standard- and valence-scenarios of GRSV polarized parton density.
Dash-doted lines denotes Azymuthal asymmetry cos OO for the unpolarized cross-section.
4.2. Pion or 2-jets production

The process (iv) e + p e + 2-jets + X or e + + X may be useful to get further
insight into the polarized parton distribution in the nucleon. Here we present an estimate
of 1LO and cos LO for the charged pion production. Fig. 2 shows the qT -dependence
((a) and (b)) and zf -dependence ((c) and (d)) of these asymmetries at COMPASS ((a) and
(c)) and EIC ((b) and (d)) kinematics. cos OO is also shown for comparison. For the
spin asymmetry, we showed the results with standard (solid line) and valence (dashed line)
scenarios in the GRSV distribution. At xbj = 0.4, the spin asymmetry 1LO can be as
large as 50%, while cos LO is within a few % level. At xbj = 0.012, these magnitudes
are somewhat reduced, since the polarized parton density is more suppressed in the small
x-region compared to the unpolarized parton density. At xbj = 0.4, the two scenarios
give totally different behavior for 1LO , which can be, in principle, a sensitive test for
the polarized parton density. Typical magnitude of cos OO is larger than cos LO as
expected, since the unpolarized parton density is larger than the polarized one.
4.3. production
Fig. 3 shows the qT -dependence ((a) and (b)) and the zf -dependence ((c) and (d)) of
the azymuthal spin asymmetry for (v) e + p e + + X ((a) and (c) for 1OL , and (b)
282
Fig. 3. Azymuthal spin asymmetry for e + p e + + X at the COMPASS energy. (a) and (b) show the
qT -dependence of 1OL and cos OL , respectively. (c) and (d) show the zf -dependence of 1OL and
cos OL , respectively.
and (d) for cos OL ) at the COMPASS kinematics. Fig. 4 shows the same asymmetries
as Fig. 3 but for the EIC kinematics. In [3], three sets of polarized fragmentation function
for + have been constructed based on three different assumptions at a low-energy
input scale: scenario 1 corresponds to the nonrelativistic quark model picture, where
only polarized s-quark is assumed to fragment into the polarized , i.e., u = d = 0.
Scenario 2 is based on the assumption that the fragmentation function has a similar flavor
decomposition as the distribution function constructed from protons structure function
p
g1 by the SU(3) flavor symmetry, i.e., u = d = 0.2s . Scenario 3 assumes three
flavors contribute equally to the fragmentation process, i.e., u = d = s . In all three
cases, s has positive sign. We calculated the asymmetries with these three sets. In both
kinematics, 1OL can be as large as 50%, while cos OL is within a few% level, varying
with different scenarios for the fragmentation functions.
The pattern of the calculated asymmetry 1OL can be easily understood [3]. At the
COMPASS kinematics with xbj = 0.4, contribution to the asymmetry is basically from
the valence u-quark distribution (see (3.1) and (3.3)). Thus scenario 1 in which only
s-quark fragments into gives negligible asymmetry, while scenarios 2 and 3 gives,
respectively, negative and positive asymmetries, the former being smaller in magnitude.
At the EIC kinematics with smaller xbj (xbj = 0.012), a large contribution comes from
the sea-quark distribution, which is nearly flavor symmetric. Accordingly scenario 1 also
283
Fig. 4. The same as Fig. 3 but for EIC energy.
gives a positive asymmetry with its magnitude smaller than the scenario 3, since only s In scenario 2, u- and d-quark contribution is partially canceled by
quark fragments into .
s-quark contribution, resulting in the small negative asymmetry.
Figs. 5 and 6 show the asymmetry for (ii) e + p e + + X at the COMPASS
and EIC kinematics, respectively. We show the results with the standard scenario for
the GRSV polarized parton density. For comparison unpolarized azymuthal asymmetry
cos OO and cos 2OO are also shown in Fig. 5, the former being essentially the same
as the one for the pion production shown in Fig. 2 as expected. Typical magnitude of
1LL , cos LL and cos 2LL at xbj = 0.4 is a few 10%, a few %, and O(103 ). At
xbj = 0.4, 1LL and cos LL have approximately the same magnitude as 1OL and
cos()OL . At xbj = 0.012, 1LL and cos LL are, respectively, smaller than 1OL and
cos OL , since polarized parton density is more suppressed at smaller x region compared
with unpolarized parton density. The distinction among three scenarios for the polarized
fragmentation function can be understood in the same way as the asymmetry for (v)
e + p e + + X shown in Figs. 3 and 4.
Figs. 7 and 8 show the spin asymmetry for (iii) e + p e + + X at the COMPASS
and EIC kinematics. The calculation is done at cos(A B ) = 1. In the present estimate,
we assume the transversity distribution and the transversity fragmentation function are
equal to the longitudinally polarized distribution and fragmentation function, respectively:
We put q q and q q.
This assumption may be justified at a low-energy scale.
284
Fig. 5. Azymuthal spin asymmetry for e + p e + + X at the COMPASS energy. (a), (b), and (c) show the
qT -dependence of 1LL , cos LL , and cos 2LL , respectively. (d), (e), and (f) show the zf -dependence of
1LL , cos LL , and cos 2LL , respectively. Unpolarized azymuthal asymmetry cos OO is shown in (b)
and (e), and cos 2OO is shown in (c) and (f) for comparison.
285
Fig. 7. Azymuthal spin asymmetry for e + p e + + X at the COMPASS energy. (a), (b), and (c) show
the qT -dependence of 1T T , cos T T , and cos 2T T , respectively. (d), (e), and (f) show the zf -dependence
of 1T T , cos T T , and cos 2T T , respectively.
But even in that case, they should be different at high energy, since their Q2 -evolution
is different, in particular, at small x [21]. One can alternatively estimate the upper bound
based on the Soffers equality as was done for p + p + X [5]. Here we intend to
present a simplest estimate to study characteristic features of the asymmetry by the above
ansatz. One sees from the figures that 1T T is of a few % to 20% at xbj = 0.4 depending
on the scenarios for the fragmentation function. This value is smaller than 1LL , which
is expected since there is no gluon contribution to the transverse spin asymmetry. On the
other hand, cos T T is from a few to 10%, and even cos 2T T is of a few % level, which
are much larger than cos nLL (n = 1, 2). The same feature persists also at xbj = 0.012,
although absolute magnitude of the asymmetry becomes smaller by the same reason for
cos nLL (n = 0, 1, 2). It is interesting to check these peculiar features of the hard crosssection in future experiments and to get information about transversity.
5. Summary and conclusion

In this paper, we have studied the double spin asymmetries for 2-jets and large-pT
hadron production in semi-inclusive DIS with one-photon exchange. The cross-section
formula for the complete set of the spin dependent processes (1.1) has been derived to
O(s ). Separating the dependence on the azymuthal angle between the hadron and lepton
286
planes, the formula can be written as

d 5
= 0 + cos 1 + cos 22 ,
dxbj dQ2 dzf dqT2 d
(5.1)
where the last term is absent for the process with polarized lepton beams. This
decomposition is purely kinematic, while the magnitude of each component depends on
the origin of the transverse momentum of the final hadron. At low-pT , nonperturbative
component such as intrinsic-kT of partons is expected to contribute as studied in [9].
At moderate-pT , higher-order effect represented by the resummed cross-section is
presumably important. Our O(s ) formula should be useful to clarify the importance of
these effects and to see at which pT the experimentally observed spin asymmetry follows
the perturbative QCD prediction.
In order to see the qualitative behavior of each component in (5.1), we calculated the
azymuthal asymmetries using the parton distribution for the nucleon and the fragmentation
functions for and at xbj = 0.4 and xbj = 0.012. We found that the asymmetry is very
sensitive to different scenarios of parton distribution and fragmentation functions as was
expected from the previous studies on SIDIS and pp collisions [35]. Typical magnitude
of asymmetry from each component in (5.1) turns out to be O(0.1)O(0.6), O(102 ) and
O(103 ) from the first to third term in (5.1). For the process (iii) e + p e + + X in
(1.1), however, the second and third contribution in (5.1) is significantly larger, while the
first term is smaller, compared with other spin asymmetries.
As in the case of unpolarized cross-sections, 0 component of the polarized crosssections have 1/qT2 -singularity at low-qT , except for the above process (iii) where 2
287
has the same singularity. To get a meaningful formula at low-qT , we need resummation
technique [10,11], which will be reported in a future publication.
Acknowledgement
This work is supported in part by the Grant-in-Aid for Scientific Research of MonbuKagaku-sho.
Appendix A
In this appendix, we give the expression for the Lorentz invariants and hadron frame
variables in terms of the variables in the lab. frame. (See [8] for the detail.) In the lab.
frame, 4-vectors which appeared in the text can be parametrized as
pA = EA (1, 0, 0, 1),
pB = EB (1, sin B cos B , sin B sin B , cos B ),

k = E(1, 0, 0, 1),
k = E (1, sin , 0, cos ),
q = (E E , E sin , 0, E cos E),

SA = 0, cos AL , sin AL , 0 ,

SB = 0, cos B cos B cos BL sin B sin BL ,

cos B sin B cos BL + cos B sin BL , sin B cos BL ,
(A.1)
L
is the azymuthal angle of the transverse spin vector SA,B which satisfy
where A,B
SA pA = SB pB = 0. In order to give the expression for qT2 , we introduce the polar
angle for pB in the Born amplitude (diagram obtained by removing the gluon line from
Fig. 1(a)), which allows us to write pB (1, sin , 0, cos ). It is easy to show is given
as

Q2
2xbj EA
1
.
cot =
(A.2)
2
Q
xbj Sep
With this we have

qT2
8E 2 4E (2E E )(1 + cos )

2 B
2 B
=
sin
+ sin B sin sin
,
1 cos B
2
2
(A.3)
and
1/2
2

qT2
Q2
Q
Q
Q2
B
1
.
cos =
+
cot2
1
2qT
xbj Sep
xbj Sep Q2
2xbj EA
2
(A.4)
288
Finally, the azymuthal angles cos A and cos B in the hadron frame can be written as

1 1
L

L
cos A = SA X =
EB sin B cos B A E sin cos A , (A.5)
qT zf
1
SB Z
sin B

Q2 + qT2
E sin cos B cos B cos BL sin B sin BL
=
2
2Q qT

(E cos E) sin B cos BL .
cos B =
References
[1] X. Ji, Phys. Rev. D 49 (1994) 114.
[2] D. de Florian, C.A. Garcia Canal, R. Sassot, Nucl. Phys. B 470 (1996) 195;
D. de Florian, R. Sassot, Nucl. Phys. B 488 (1997) 367.
[3] D. De Florian, M. Stratmann, W. Vogelsang, Phys. Rev. D 63 (1998) 5811.
[4] D. De Florian, M. Stratmann, W. Vogelsang, Phys. Rev. Lett. 81 (1998) 530.
[5] D. De Florian, J. Soffer, M. Stratmann, W. Vogelsang, Phys. Lett. B 439 (1998) 176.
[6] C. Boros, J.T. Londergan, A.W. Thomas, Phys. Rev. D 62 (2000) 014021.
[7] A. Mndez, Nucl. Phys. B 145 (1978) 199.
[8] R. Meng, F.I. Olness, D. Soper, Nucl. Phys. B 371 (1992) 79.
[9] P.J. Mulders, R.D. Tangerman, Nucl. Phys. B 461 (1996) 197.
[10] R. Meng, F.I. Olness, D. Soper, Phys. Rev. D 54 (1996) 1919.
[11] P.M. Nadolsky, D.R. Stump, C.P. Yuan, Phys. Rev. D 61 (2000) 014003.
[12] J.C. Collins, D. Soper, Nucl. Phys. B 194 (1982) 445.
[13] R.L. Jaffe, X. Ji, Nucl. Phys. B 375 (1992) 527.
[14] A.V. Manohar, Phys. Rev. Lett. 65 (1990) 2511.
[15] X. Ji, Phys. Lett. B 289 (1992) 137.
[16] R.L. Jaffe, X. Ji, Phys. Rev. Lett. 71 (1993) 2547.
[17] R.N. Cahn, Phys. Lett. B 78 (1978) 269.
[18] M. Glck, E. Reya, A. Vogt, Eur. Phys. J. C 5 (1998) 461.
[19] M. Glck, E. Reya, M. Stratmann, W. Vogelsang, Phys. Rev. D 63 (2001) 094005.
[20] B.A. Kniehl, G. Kramer, B. Ptter, Nucl. Phys. B 582 (2000) 514.
[21] X. Artru, M. Mekhfi, Z. Phys. C 45 (1990) 669;
W. Vogelsang, Phys. Rev. D 57 (1998) 1886;
A. Hayashigaki, Y. Kanazawa, Y. Koike, Phys. Rev. D 56 (1997) 7350.
(A.6)

Electroweak radiative corrections to e+e H
A. Denner a , S. Dittmaier b , M. Roth c , M.M. Weber a

a Paul Scherrer Institut, Wrenlingen und Villigen, CH-5232 Villigen PSI, Switzerland
b Max-Planck-Institut fr Physik (Werner-Heisenberg-Institut), D-80805 Mnchen, Germany
c Institut fr Theoretische Physik, Universitt Karlsruhe, D-76128 Karslruhe, Germany
Abstract
The complete electroweak O() radiative corrections to the Higgs-boson production processes
e+ e l l H (l = e, , ) are calculated in the electroweak Standard Model. For e+ e e e H,
where ZH production and W-boson fusion contribute, both production channels are added coherently.
The calculation of the corrections is described in some detail including, in particular, the treatment of
the Z-boson resonance in the ZH-production channel. The discussion of numerical results focusses
on the total cross section as well as on angular and energy distributions of the Higgs boson. In the
G -scheme, the bulk of the corrections is due to initial-state radiation. The corrections turn out to
reduce the total cross section by 10% for high energies, where the W-boson fusion dominates.
In this region, the corrections depend only weakly on the energy and the production angle of
the Higgs boson. Based on an analysis of the leading universal corrections, a simple improved
Born approximation is introduced. This approximation describes the corrected cross section within
about 3%.
1. Introduction
One of the most important open problems of particle physics is the understanding of the
mechanism of electroweak symmetry breaking. In the electroweak Standard Model (SM) it
is provided by the Higgs mechanism, leading to the prediction of a physical scalar particle,
the Higgs boson. The investigation of the mechanism of electroweak symmetry breaking
and, in particular, of the Higgs boson, will be one of the main concerns at the Large Hadron
Collider (LHC) at CERN. The LHC experiments ATLAS [1] and CMS [2] are sensitive
to the SM Higgs boson over the whole mass range from the present lower experimental
E-mail address: ansgar.denner@psi.ch (A. Denner).
doi:10.1016/S0550-3213(03)00269-4
290
A. Denner et al. / Nuclear Physics B 660 (2003) 289321
limit of 114.4 GeV [3] up to 1 TeV and will discover the Higgs boson, if it exists and
has no particularly exotic properties. Moreover, these experiments will determine various
properties of the Higgs boson, such as its mass, branching ratios, and ratios of its coupling
constants.
However, the complete profile of the Higgs boson can only be studied in the clean
environment of an electronpositron linear collider [47]. In e+ e annihilation there
are two main production mechanisms for the SM Higgs boson. In the Higgs-strahlung
process, e+ e ZH, a virtual Z-boson decays into a Z boson and a Higgs boson. The
corresponding cross section rises sharply at threshold to a maximum of a few tens of GeV
above MZ + MH and then falls off as s 1 , where s is the centre-of-mass (CM) energy
of the e+ e system. In the W-boson fusion process, e+ e e e H, the incoming e+ and
e each emit a virtual W boson which fuse into a Higgs boson. The cross section of the
W-boson-fusion process grows as ln s and thus is the dominant production mechanism for
s MH . The cross section for the similar Z-boson fusion process, e+ e e+ e H, is

about one order of magnitude smaller.
At a linear e+ e collider with a CM energy of 500 GeV and an integrated luminosity
of 500 fb1 , of the order of 104 Higgs bosons can be produced per year [5]. This allows to
measure the Higgs-production cross sections and thus the Higgsgauge-boson couplings at
the level of a few per cent. Consequently, adequate theoretical predictions have to take into
account radiative corrections and the relevant effects of the finite decay width of the Z and
Higgs bosons. For a heavy Higgs, which decays mainly into W-boson pairs, finite-width
effects have been investigated in Refs. [8,9]. On the other hand, the effects of the finite
width of the Higgs boson can be neglected if the Higgs boson is light, i.e., if its width is
small.
The Higgs-strahlung process has been investigated in lowest order in Ref. [10].
Including the Z-boson decay, the process e+ e Z H ffH has been studied by
Bjorken [11]. For the process e+ e ZH, the O() electroweak corrections have been
calculated in the soft-photon approximation in Refs. [1214], and a Monte Carlo algorithm
for the calculation of the real photonic corrections to this process was described in
Ref. [15]. A compact analytical formula for the electromagnetic corrections to the total
cross section can be found in Ref. [13].
The vector-boson fusion processes have been investigated in lowest order in Ref. [16].
The electroweak corrections to e+ e H have attracted a lot of interest recently.
Analytical results for the one-loop corrections to this process have been obtained in
Ref. [17] as MAPLE output, but a numerical evaluation of these results is not yet available.
A first complete calculation of the O() electroweak corrections to e+ e H in the
SM has been performed in Ref. [18]. The contributions of fermion and sfermion loops
in the Minimal Supersymmetric Standard Model have been evaluated in Refs. [19,20]
with seemingly differing results. We have performed a completely independent calculation
of the O() electroweak corrections to the complete process e+ e H in the SM.
First results of our calculation have already been presented in Ref. [21]. There, we have
successfully compared our results for the total cross section with those of Refs. [1820]
and pointed out that the differences between Refs. [19] and [20] are due to different
renormalization schemes and input parameters.
291
In this paper, we present the details of our calculation of the O() electroweak
corrections to the processes e+ e e e H, H, and H. While the first process
gets contributions from W fusion and Higgs-strahlung, the final states with or neutrinos
receive contributions only from Higgs-strahlung off Z bosons. Single hard-photon radiation
is included using the complete matrix element, and higher-order ISR corrections are taken
into account in the leading-logarithmic approximation. The finite Z-boson decay width is
introduced in the constant-width scheme and in a gauge-invariant scheme that is based
on a factorization of the Z resonance in the gauge-invariant set of diagrams related to the
(neutral-current) Higgs-strahlung process.
The paper is organized as follows: in Section 2 the calculation of the virtual, real,
and (higher-order) leading-logarithmic corrections is described, and the leading universal
corrections are discussed. Section 3 contains a discussion of numerical results. The paper
is summarized in Section 4. Further useful information on the calculation is collected in
Appendix A.
2. Calculation of radiative corrections

2.1. Conventions and lowest-order cross section
We consider the processes
e (p1 , 1 ) + e+ (p2 , 2 ) l (k1 ) + l (k2 ) + H(k3 ),
l = e, , ,
(2.1)
where the momenta pi , kj of all particles and the electron helicities i are given in
parentheses. The helicities take the values i = 1/2, but we often use only the sign to
indicate the helicity. The electron mass is neglected whenever possible, i.e., it is kept finite
only in the mass-singular logarithms related to initial-state radiation. This implies that the
lowest-order and one-loop amplitudes vanish unless 1 = 2 . Therefore, we define =
1 = 2 . The particle momenta obey the mass-shell conditions p12 = p22 = k12 = k22 = 0
and k32 = MH2 . For later use, the following set of kinematical invariants is defined:
s = (p1 + p2 )2 ,
sij = (ki + kj )2 ,
i, j = 1, 2, 3,
tij = (pi kj ) ,
i = 1, 2, j = 1, 2, 3.
(2.2)
In lowest order the processes (2.1) proceed via the diagrams shown in Fig. 1. More
precisely, only for electron neutrinos in the final state (l = e) both the ZH-production and
WW-fusion diagrams contribute, while for and neutrinos merely the former exists.
In the calculation of the tree-level amplitude M0 and of the one-loop amplitude M1 ,
which is described in the next section, we separate the fermion spinor chains by defining
standard matrix elements (SME). To introduce a compact notation for the SME, the tensors
ee,
{,
} = v e+ (p2 ){ , } ue (p1 ),
{,
l (k1 ){ , } v l (k2 ),
}=u
292
Fig. 1. Lowest-order diagrams for e e+ H.
e
{,
l (k1 ){ , } ue (p1 ),
}=u
e
{,
} = v e+ (p2 ){ , } v l (k2 )
(2.3)
are defined with obvious notations for the Dirac spinors ve+ (p2 ), etc., and = (1 5)/2
denote the right- and left-handed chirality projectors. Here and in the following, each entry
in the set within curly brackets refers to a single object, i.e., from the first line in the
equation above we have ee, = ve+ (p2 ) ue (p1 ), etc. Furthermore, symbols like p
are used as shorthand for the contraction p . For the ZH-production channel we define
the 26 SME
ZH, = ee, ,{,p1 p2 } ,
M
{1,2}
ZH, = ee, ,{,p1 p2 } ,

M
k1 k2
{3,4}
ZH, = ee, ,{p1 ,p2 } ,

M
{5,6}
k1
ZH, = ee, ,{p1 ,p2 } ,

M
{7,8}
k2
ZH, = ee, ,{p1 ,p2 } ,

M
k1
{9,10}
ZH, = ee, ,{p1 ,p2 } ,

M
k2
{11,12}
ZH, = ee, , .
M
13
(2.4)
For the WW-fusion channel we introduce the following set of 13 SME,

WW = e e,{,k1 p1 } ,
M
{1,2}
WW = e e,{,k1 p1 } ,
M
k2 p2
{3,4}
WW = pe e,{p1 ,k1 } ,
M
{5,6}
2
WW = e e,{p1 ,k1 } ,
M
{7,8}
k2
WW = e e,{p1 ,k1 } ,
M
p2
{9,10}
WW = e e,{p1 ,k1 } ,
M
k2
{11,12}
WW = e e, .
M
13
(2.5)
The tree-level and one-loop amplitudes can be expanded in terms of linear combinations
of SME,
Mn =
13

i=1
ZH, ZH,
Fn,i
+
Mi
13
WW WW
Fn,i
Mi ,
n = 0, 1,
(2.6)
i=1
ZH,
WW
and Fn,i
. This decomposition is unique in D
with Lorentz-invariant functions Fn,i
dimensions, i.e., if only the Dirac equation for the spinors is used. The number of SME
can, however, be reduced further by exploiting the four-dimensionality of spacetime,
which implies relations among the SME given above. In fact, it is possible to express
the set of all 39 SME in terms of two independent SME. We list the expressions of the two
ZH, , and the relations among the SME
independent SME, which can be identified with M
1
in Appendix A.
293
The lowest-order amplitude reads

+ MWW
M0 = MZH,
0 ,
0
(2.7)
where
MZH,
=
0
1
e3 MW
ZH, ,
g M
3
2
2
2cw sw (s MZ )(s12 MZ2 + iMZ Z ) Ze 1
(2.8)
MWW
= le
0
e3 MW
1
WW
M
1
2 )(t M 2 )
2sw3 (t11 MW
22
W
(2.9)
of the electron to the Z-boson,

with the chiral couplings gZe
=
gZe
sw

.
cw 2cw sw
(2.10)
The sine and cosine of the weak mixing angle are fixed by
2
= 1 sw2 =
cw
2
MW
MZ2
(2.11)
For s12 MZ2 , the lowest-order amplitude develops a resonance corresponding to real
ZH production with the subsequent Z l l decay. We, therefore, have included a finite
Z-boson width Z in the denominator, which results from a (partial) Dyson summation of
the corresponding propagator. This procedure is discussed in more detail in the context of
the virtual radiative corrections in the next section.
Finally, the lowest-order cross section reads

1

2
1
(1 + 2P )(1 2P+ )M0 ,
d3
0 =
(2.12)
2s
4
=1/2
where P are the degrees of polarization of the e beams and the phase-space integral is
defined by

3

3

d3 ki
4
(2) p1 + p2
d3 =
(2.13)
kj .
3 2k 0
(2)
i
i=1
j =1
2.2. Virtual corrections
2.2.1. Survey of one-loop diagrams
The virtual corrections can be classified into self-energy, vertex, box, and pentagon
corrections. The generic contributions of the different vertex functions are shown in Fig. 2.
The first three lines contain those diagrams that contribute to all l l H final states, whereas
the diagrams in the last three lines contribute only to e+ e e e H.
The complete set of pentagon diagrams is shown in Fig. 3. The last eight diagrams
contribute only for the e e H final state. The l l ZH and l l H box diagrams are depicted
in Fig. 4, the e+ e ZH box diagrams in Fig. 5, and the e e W+ H box diagrams in Fig. 6.
Those for the e+ e W H box diagrams can be obtained by charge conjugation.
294
Fig. 2. Contributions of different vertex functions to e+ e H.
Fig. 3. Pentagon diagrams for e+ e H.
Fig. 4. l l ZH and l l H box diagrams.
Fig. 5. e+ e ZH box diagrams.
295
296
Fig. 6. e e W+ H box diagrams.
(a)
(b)
Fig. 7. Diagrams for the l l H and e+ e H vertex functions.
The diagrams for the l l H and e+ e H vertex functions are listed in Fig. 7. Fig. 8
shows the Feynman diagrams for the W+ W H vertex function and Fig. 9 those for the
ZZH and ZH vertex functions. Note that in Fig. 9 those diagrams that are obtained from
the diagrams in the first three lines of this figure by reversing the charge flow in the loop
have been suppressed. Most of the diagrams for the self-energies and the l l Z, e+ e Z,
()
and e e W vertex functions can be found in Ref. [22].

All pentagon and box diagrams are UV finite, and also the e+ e H and l l H vertex
functions are finite since we neglect the electron mass everywhere apart from the masssingular logarithms. For the other vertex functions, ZH, ZZH, l l Z, e+ e Z, W+ W H,
()
e e W, and for the ZZ, Z, and WW self energies the corresponding counterterm
diagrams have to be included.
2.2.2. Calculational framework
The actual calculation of the one-loop diagrams has been carried out in the t Hooft
Feynman gauge using standard techniques. The Feynman graphs have been generated
with FeynArts [23] and are evaluated in two completely independent ways, leading to
two independent computer codes. The results of the two codes are in good numerical
agreement (i.e., within about 12 digits for non-exceptional phase-space points). Apart
from the 5-point integrals, in both calculations the tensor coefficients of the one-loop
integrals are algebraically reduced to scalar integrals with the PassarinoVeltman algorithm
[24] at the numerical level. The scalar integrals are evaluated using the methods and
Fig. 8. Diagrams for the W+ W H vertex function.
297
298
Fig. 9. Diagrams for the ZZH and ZH vertex functions.
results of Refs. [2527], where ultraviolet divergences are regulated dimensionally and
IR divergences with an infinitesimal photon mass m . The renormalization is carried out
in the on-shell renormalization scheme, as, e.g., described in Ref. [27].
In the first calculation, the Feynman graphs are generated with FeynArts version 1.0
[23]. With the help of Mathematica routines the amplitudes are expressed in terms of SME
and coefficients of tensor integrals. The output is processed into a Fortran program for
the numerical evaluation. For the evaluation of the tensor 5-point function two approaches
299
have been followed: the usual PassarinoVeltman reduction and the direct reduction to
4-point integrals of Ref. [28]. The results based on the PassarinoVeltman algorithm
become numerically unstable at the phase-space boundary and could only be rescued by a
careful extrapolation out of the numerically safe inner phase-space domains, as described
in Section 2.2.4 of Ref. [29]. The instabilities are due to the occurrence of inverse Gram
determinants in the recursive tensor reduction. The direct reduction of Ref. [28] avoids
such inverse Gram determinants, rendering the results of this approach well behaved near
the phase-space boundary.
The calculation of the virtual corrections has been repeated using the backgroundfield method [30], where the individual contributions from self-energy, vertex, box, and
pentagon corrections differ from their counterparts in the conventional formalism. The
total one-loop corrections of the conventional and of the background-field approach were
found to be in perfect numerical agreement.
The second calculation has been made using FeynArts version 3 [31] for the generation
and FormCalc [32] for the evaluation of the amplitudes. The analytical results of FormCalc
in terms of SME and their coefficients were translated into C++ code, and the interference
of the one-loop with the lowest-order amplitude calculated numerically. However, in the
case of the pentagon diagrams this interference was calculated analytically using FeynCalc
[33]. After evaluation of the fermion traces, the loop momenta appear in the numerator
only in scalar products and can be cancelled against propagator denominators. In this way,
only scalar 5-point functions remain, thereby avoiding inverse Gram determinants and the
related numerical instabilities.
Finally, the contribution of the virtual corrections to the cross section is given by

1

1
virt =
(2.14)
(1 + 2P )(1 2P+ )2 Re M1 M0 .
d3
2s
4
=1/2
2.2.3. Treatment of the Z-boson resonance

At tree level the introduction of the Z-boson decay width in the lowest-order matrix
element M0 of Eq. (2.7) did not pose any problem with gauge invariance or doublecounting of higher-order effects, since the gauge-invariant ZH-production part before
Dyson summation, MZH,
(Z = 0), was simply rescaled as
0
=
MZH,
0
s12 MZ2
s12 MZ2 + iMZ Z
MZH,
(Z = 0).
0
(2.15)
This modification reproduces the correct BreitWigner shape near the resonance and
changes MZH,
(Z = 0) at the relative order O(Z /MZ ) O() away from the
0
resonance. It should also be mentioned that we introduce a fixed Z-boson width (in contrast
to a running width), i.e., the Z-boson mass MZ deviates from the on-shell mass at the twoloop level (see e.g., Ref. [34]).
The issue of gauge invariance, resonances, Dyson summation, and radiative corrections
is rather complex. In fact, no simple general solution for a gauge-invariant treatment of
finite-width effects exists yet. Fortunately, the situation in our case is relatively simple. We
proceed in two different ways.
300
First we apply the so-called naive fixed-width scheme which simply means that each
resonance propagator 1/(s12 MZ2 ) is replaced by 1/(s12 MZ2 + iMZ Z ), while nonresonant contributions are kept untouched. The contribution iMZ Z originates from the
imaginary part of the (transverse part of the) Z self-energy &TZZ on resonance; in order
has to be
to avoid double-counting, the O() contribution thus contained in MZH,
0
subtracted from the one-loop amplitude. In summary, the one-loop amplitude in the fixedwidth scheme reads
MZH,
1,fixed-width

= MZH,
(Z = 0) MZZ-self (Z = 0)
1

&TZZ (s) Re{&TZZ (MZ2 )}

s MZ2
1
1
2 s M 2 +iM
s12 MZ
Z Z
12
Z
&TZZ (s12 ) &TZZ (MZ2 )

s12 MZ2

+ 2ZZZ MZH,
,
0
(2.16)
where we have made the Z self-energy and its corresponding counterterm contributions
explicit. Note that the term Re{&TZZ (MZ2 )} is identical with the mass counterterm MZ2 ,
while &TZZ (MZ2 ) receives its imaginary part i Im{&TZZ (MZ2 )} in the resonant part from the
. Since only complete
subtraction of the iMZ Z contribution already contained in MZH,
0
S-matrix elements exhibit the gauge-invariance properties such as the independence of
gauge-fixing conditions (gauge parameters) and the validity of SlavnovTaylor identities,
this procedure potentially violates gauge invariance.
As a second option, we applied a factorization scheme where the full ZH-production
amplitude before Dyson summation is rescaled similar to MZH,
(Z = 0) in Eq. (2.15).
0
This procedure is analogous to the treatment of the W-boson resonance in pp W ll
as described in Ref. [35]. Of course, also here double-counting of iMZ Z terms has to be
avoided. In summary, the one-loop amplitude in the factorization scheme reads
MZH,
1,fact. =
=
s12 MZ2
s12 MZ2 + iMZ Z
MZH,
(Z = 0) +
1
ZH,
M1 (Z
2
s12 MZ + iMZ Z
ZZ
&T (s) Re{&TZZ (MZ2 )}
+
s MZ2
s12 MZ2
+ 2ZZZ MZH,
.
0
i Im{&TZZ (MZ2 )}
s12 MZ2
MZH,
0
= 0) MZZ-self (Z = 0)
&TZZ (s12 ) &TZZ (MZ2 )

s12 MZ2
(2.17)
Note that this procedure preserves gauge invariance, since the gauge-invariant amplitude
MZH,
(Z = 0) is only rescaled and accompanied by another gauge-invariant term pro1
portional to MZH,
. However, the rescaling puts all non-resonant terms in MZH,
(Z = 0)
0
1
to zero on the resonance. The corresponding error is of the order of the non-resonant terms
in the O() corrections, i.e., of order O(Z /MZ ) O( 2 ).
Within integration errors, both schemes give the same results.
301
2.2.4. Universal electroweak corrections

The electroweak corrections contain large contributions of universal origin. Besides
initial-state radiation (ISR), which is discussed in Section 2.3.3, these consist, in particular,
2
of the corrections associated with the running of and corrections proportional to m2t /MW
that can be associated to the parameter or to the renormalization of the electroweak
mixing angle. By suitable parametrization of the lowest-order matrix elements, these
universal corrections can be incorporated into the lowest order thus reducing the remaining
corrections. This does not only reduce the O() corrections but in general also the higherorder corrections.
The running of (Q2 ) from Q2 = 0 to Q2 = MZ2 is of the order of + 6%. Since
the cross section for e+ e H is proportional to 3 this amounts to an effect of
18% if (0) defines the electromagnetic coupling ((0)-scheme). By parametrizing
the lowest-order cross section in terms of (MZ2 ) ((MZ2 )-scheme) this large effect can be
incorporated into the leading-order expressions. Note that + contains large contributions
from the hadronic vacuum polarization that cannot be calculated in perturbation theory.
Thus, the sensitivity to these effects is absorbed in the lowest-order cross section via (MZ2 )
in the (MZ2 )-scheme. Alternatively the electromagnetic coupling can be deduced from the
Fermi constant G (G -scheme) via
2 s2
2 G MW
w
,
=
(2.18)
where G is measured in muon decay. Consequently, in the transition from the (0)to the G -scheme the constant 3+r has to be subtracted from the relative correction to
a cross section that is proportional to 3 , where +r contains the electroweak radiative
corrections to muon decay. Since +r is about 3% for the empirical value of mt , this
shifts the relative corrections by 9% with respect to the (0)-scheme. Apart from
absorbing +, the quantity +r additionally contains universal corrections proportional
2 that can be associated to the parameter, +r + +c 2 /s 2 3%
to m2t /MW
w w
with + = 3G m2t /(8 2 2 ) [36].

The cross section for e+ e H is dominated by the WW-fusion diagram, which
gets its main contribution from the region of small momentum transfers. Consequently,
the corresponding corrections are determined by the ee W and WWH vertex corrections
for small invariant W masses and depend only weakly on the energy. The correction to the
ee W vertex in the relevant kinematical region is well approximated by +r. It turns out that
this is also the case for the main contributions to the WWH vertex. Thus, parametrizing the
lowest order in terms of G (G -scheme) absorbs a large part of the universal corrections.
Since in the G -scheme all large universal corrections related to the running of and
2 are absorbed, we use this scheme in the following. With
most of the corrections m2t /MW
respect to this scheme, the corrections in the (0)-scheme are shifted by 3+r +9% and
those in the (MZ2 )-scheme by 3(+r +(MZ2 )) 9%.
302
We have extracted the leading mt -dependent corrections in the heavy-top limit in the
G -scheme:

MZH,
1
G -scheme

2
ZH,

3 2sw2
mt M 0
1 6cw
mt
+
g
+
ln
+ O m0t ,
Ze
2
2
mt 4s
sw
3cw sw
MW
gZe
MW
w 8

5 m2t
MWW
+ O m0t .
MWW
1
0
G -scheme mt 32s 2
2
w MW
(2.19)
The leading m2t -enhanced terms of the WW channel agree with the terms derived for the
HWW vertex [37], since in the G -scheme all leading m2t contributions related to the
W-boson coupling to fermions are absorbed in G . In the ZH channel, m2t -enhanced terms
do not only result from the HZZ vertex, but there are also remnants originating from the
renormalization of the Z-boson couplings to fermions. In contrast to the WW channel, in
large top-quark mass.
the ZH channel there are also logarithmic terms ln mt for a
In the physically interesting region of e+ e H, s is of the same order as mt
or larger. Nevertheless, the expressions for mt reproduce the full mt -dependent
corrections rather well for the WW channel, which
is dominated by small momentum
transfers. The situation for the ZH channel, where s is a typical scale for the momentum
transfer, is completely different. There, the expressions for mt do not provide a good
approximation, and we did not succeed in finding a simple approximation for the fermionic
corrections to the ZH channel. This is due to the presence of loop integrals that depend both
on s and m2t . Since, however, the cross section for e+ e H is dominated by the WW
channel we consider it useful to define the following improved Born approximation
5 m2t
non-photonic
dIBA
= d0 d0WW
.
2
16sw2 MW
(2.20)
This cross section is then convoluted with the ISR as given in (2.49) below.
As discussed in Ref. [21] and in Section 3, the bosonic corrections are small for the WW
channel but large for the ZH channel. In order to find the source of these large corrections,
2 which are
we have evaluated the leading bosonic corrections in the limit s MH2 MW
2
2
2
of the order O(s/MW ) O(MH /MW ). We found that these terms are small and cannot
explain the large bosonic corrections in the ZH channel. In the t HooftFeynman gauge
the large corrections arise predominantly from box diagrams involving W-boson exchange
[14]. The enhancement of these diagrams is partially due to the gauge-boson coupling to
the electron which for W bosons is stronger than for Z bosons.
2.3. Real photonic corrections
2.3.1. Matrix-element calculation
The real photonic corrections are induced by the process
e (p1 , 1 ) + e+ (p2 , 2 ) l (k1 ) + l (k2 ) + H(k3 ) + (k, ),
l = e, , ,
(2.21)
303
Fig. 10. Feynman diagrams for e+ e H

.
where k and denote the photon momentum and helicity, respectively. The Feynman
diagrams of this process are shown in Fig. 10. We have evaluated the helicity matrix
elements M of this process using the Weylvan der Waerden spinor technique as
formulated in Ref. [38]. The amplitudes for the helicity channels with 1 = 2 vanish for
massless electrons. The finite-mass effects of electrons and positrons are included in the
treatment of soft and collinear singularities, as described in the next section. In the notation
of Ref. [38] the amplitudes M read
M = MZH,
+ MWW,
,
4
2 e gZe MW
1
1
=
MZH,
3 s2
2
2
2
cw
(p1 + p2 k) MZ (k1 + k2 ) MZ2 + iMZ Z
w
AZH
(p1 , p2 , k1 , k2 , k),
4
2 e MW
1
1
MWW,
= le
3
2
2
2
sw
(p1 k1 ) MW (p2 k2 )2 MW
AWW
(p1 , p2 , k1 , k2 , k)
with the auxiliary functions
AZH
(p1 , p2 , k1 , k2 , k) = p1 k2
p1 p2 p2 k1 + p1 k k1 k

,
p1 k p2 k
ZH
AZH
, (p1 , p2 , k1 , k2 , k) = A (p2 , p1 , k2 , k1 , k) ,
ZH
AZH
, (p1 , p2 , k1 , k2 , k) = A (p1 , p2 , k2 , k1 , k) ,
AWW
(p1 , p2 , k1 , k2 , k)

p1 p2 p2 k1 k1 k(p1 k1 p2 k1 + p1 k p2 k)
= p1 k2
+
2]
p1 k p2 k
p1 k [(p1 k1 k)2 MW
(2.22)
304

k1 k
k2 k(p2 k2 p2 k1 k2 k k1 k)
,
2 ]
p2 k
p2 k [(p2 k2 k)2 MW
WW
AWW
(p1 , p2 , k1 , k2 , k) = A (p2 , p1 , k2 , k1 , k) .
(2.23)
The relations between the A...

... functions that differ only in the photon helicity are a
consequence of the CP symmetry of the process, while the relation associated with a
reversion of both and results from a P transformation. The spinor products are
defined by

q
p
p
q
AB
ip
iq
,
cos sin e
cos sin
pq = 0 pA qB = 2 p0 q0 e
(2.24)
2
2
2
2
where pA , qA are the associated momentum spinors for the light-like momenta
p = p0 (1, sin p cos p , sin p sin p , cos p ),
q = q0 (1, sin q cos q , sin q sin q , cos q ).
(2.25)
The matrix elements (2.22), (2.23) have been successfully checked against the result
obtained with the package Madgraph [39] numerically.
The contribution of the radiative process to the cross section is given by

1

1
M 2 ,
(1 + 2P )(1 2P+ )
=
(2.26)
d
2s
4
=1/2
=1
where the phase-space integral is defined by

3

3

d3 k
d3 ki
d =
kj . (2.27)
(2)4 p1 + p2 k
(2)3 2k 0
(2)3 2ki0
i=1
j =1
2.3.2. Treatment of soft and collinear singularities

Without soft and collinear regulators the phase-space integral (2.26) diverges in the soft
(k0 0) and collinear (pi k 0) phase-space regions. In the following we describe two
procedures of treating soft and collinear photon emission: one is based on a subtraction
method, the other on phase-space slicing. In both cases soft and collinear singularities
are regularized by an infinitesimal photon mass and a small electron mass, respectively.
Numerically both methods agree within the integration errors, which are typically 0.05%
for the total cross section. For the numerical results presented in Section 3, the subtraction
method has been used.
2.3.2.1. The dipole subtraction approach The idea of so-called subtraction methods is
to subtract a simple auxiliary function from the singular integrand of the bremsstrahlung
integral and to add this contribution back again after partial analytic integration. This
auxiliary function, denoted |Msub|2 in the following, has to be chosen in such a way
that it cancels all singularities of the original integrand, which is |M |2 in our case, so
that the phase-space integration of the difference can be performed numerically, even over
the singular regions of the original integrand. In this difference, M can be evaluated
without regulators for soft or collinear singularities, i.e., we can make use of the results
305
of the previous section. The auxiliary function has to be simple enough so that it can be
integrated over the singular regions analytically, when the subtracted contribution is added
again. This part contains the singular contributions and requires regulators, i.e., photon
and electron masses have to be reintroduced there. Specifically, we have applied the dipole
subtraction formalism, which is a process-independent approach that was first proposed
[40] within QCD for massless unpolarized partons and subsequently generalized to photon
radiation of massive polarized fermions [41]. We only need the limit of small fermion
masses, which was worked out in Refs. [41,42] and in which the application of the method
is relatively simple. In order to keep the description of the method transparent, we describe
only the basic structure of the individual terms explicitly and refer to Refs. [41,42] for
details.
In the dipole subtraction formalism the subtraction function is constructed from
contributions that are labelled by ordered pairs ab of charged fermions, so-called dipoles.
The fermions a and b are called emitter and spectator, respectively, since by construction
only the kinematics of the emitter a leads to collinear singularities. Since we only have
charged particles in the initial state, the subtraction function receives two contributions,
which both have emitter and spectator in the initial state,

2
M (p1 , p2 , ki , k)2 =

M
sub
sub,ab (p1 , p2 , ki , k) ,
a,b=1,2
a=b

M
sub,ab (p1 , p2 , ki , k)
2

= e2 g (sub) (pa , pb , k)M (p 1 , p 2 , ki )2
0
ab
(2.28)
with the dipole function

(sub)
(pa , pb , k) =
gab

2
1
1 xab ,
(pa k)xab 1 xab
(2.29)
where
xab =
pa pb pa k pb k
.
pa pb
(2.30)
The modified momenta in Eq. (2.28) depend on ab and are defined as follows. While the
spectator momentum pb is kept fixed, the emitter momentum pa , is rescaled by xab ,
pa = xab pa ,
p b = pb ,
(2.31)
and all other momenta ki are transformed with a Lorentz transformation,
kj = kj ,
(2.32)
where
2Pab Pab,
(Pab + Pab ) (Pab + Pab )
+
,
2 +P P
2
Pab
Pab
ab ab
kj ,
Pab = xab pa + pb .
Pab = pa + pb k =
= g
(2.33)
306
The subtracted contribution can be integrated over the (singular) photonic degrees of
freedom up to a remaining convolution over x = xab . In this integration the regulators m
and me must be retained, and the soft and collinear singularities appear as logarithms in
these mass regulators,
sub (p1 , p2 , P , P+ )
1
=
dx
G(sub) (s, x)
2
=
0

d0 (xp1 , p2 , P , P+ ) + d0 (p1 , xp2 , P , P+ )

(sub)
G (s)
d0 (s, P , P+ ) + d0 (s, P , P+ ) ,
+
2 =
(2.34)
with the universal functions

1 + x2
s
(sub)
(sub)
(s, x) =
(s, x) = (1 x)+ ,
G+
ln
1
,
G
1x +
m2e
2
2

2 2
m
m
1 2 m2e
1
me
2 3
me
(sub)
G+ (s) = ln
ln
+ ln
ln
+ ln
+ ,
s
s
s
2
s
2
s
3
2
1
G(sub)
(2.35)
(s) = .
2
The two cases = +/ correspond to collinear photon emission without/with a spin-flip
of the electron or positron. The ( )+ prescription is defined as usual,
1
dx f (x) + g(x)
1

dx f (x) g(x) g(1) ,
(2.36)
and P are the degrees of polarization of the e beams.

In summary, the phase-space integral (2.26) in the dipole subtraction approach reads

1
2

1
2

(1 + 2P )(1 2P+ )
Msub
M
=
d
2s
4
=1/2
+ sub .
=1
(2.37)
2.3.2.2. The phase-space-slicing approach The idea of the phase-space slicing method
is to divide the bremsstrahlung phase-space into singular and non-singular regions, then
to evaluate the singular regions analytically and to perform an explicit cancellation of the
arising soft and collinear singularities against their counterparts in the virtual corrections.
The finite remainder can be evaluated by using the usual Monte Carlo techniques. For
the actual implementation of this well-known procedure we closely follow the approaches
of Ref. [43]. We divide the five-particle phase space into soft and collinear regions by
introducing the cut-off parameters +E and + , respectively. We decompose the real
307
corrections as
d = dsoft + dcoll + d ,finite.
(2.38)
Here dsoft describes the contribution of the soft photons, i.e., of photons with energies
k0 < +E in the CM frame, and dcoll describes real photon radiation outside the softphoton region (k0 > +E) but collinear to the e beams. The collinear region consists of
the two disjoint parts 0 < < + and + < < , where is the polar angle of
the emitted photon in the CM frame. The remaining part, which is free of singularities, is
denoted by d ,finite .
In the soft and collinear regions, the squared matrix element |M |2 factorizes into the
leading-order squared matrix element |M0 |2 and a soft or collinear factor. Also the fourparticle phase space factorizes into a three-particle and a soft or collinear part, so that the
integration over the singular part of the photon phase space can be performed analytically.
In the soft-photon region, we apply the soft-photon approximation to |M |2 , i.e., the
photon four-momentum k is omitted everywhere but in the IR-singular propagators. In this
region d can be written as [27,44]

2
p
d3 k p1
.
2
dsoft = d0 2
(2.39)
k0 p1 k p2 k
4
k0 <+E
|k|2 =k02 m2
The explicit expression for the soft-photon integral can be found in Refs. [25,27]. For
our purpose it is sufficient to keep the electron mass only as regulator for the collinear
singularities. In this limit we obtain

s
s
1 2 s
2
2+E
dsoft = d0
1 ln
ln
+ ln
+
2 ln
.
m
m2e
m2e
2
m2e
3
(2.40)
In the collinear region, we consider an incoming e with momentum pi being split
into a collinear photon and an e with the resulting momentum xpi after photon radiation.
In the asymptotic limit, |M |2 factorizes into the leading-order squared matrix element
|M0 |2 and a collinear factor describing collinear initial-state radiation, as long as + is
sufficiently small. In the collinear region also the four-particle phase space factorizes into
a three-particle phase space and a collinear part, so that the cross section for hard photon
radiation (k0 > +E) in the collinear region reads
coll (p1 , p2 , P , P+ ) =
d0 (xp1 , p2 , P , P+ ) +
where
(coll)
G+
dx
0

12+E/
s
(s, x) =
G(coll) (s, x)

d0 (p1 , xp2, P , P+ ) ,

1 + x2
s+ 2
1
,
ln
1x
4m2e
(coll)
(s, x) = 1 x.
(2.41)
(2.42)
308
Subtracting the soft and collinear cross sections (2.40) and (2.41) from the cross section
of the bremsstrahlung process yields the finite cross section d ,finite . As usual in the
phase-space slicing approach, this subtraction is done in practice by imposing cuts on the
bremsstrahlung phase space, i.e., a photon-energy cut, k0 > +E, and a cut on the angles
between the photon and the beams, + < < + .
For the numerical evaluation the cuts have been chosen as +E = 0.01 GeV and
+ = 0.001. We have checked that the cross section changes only within integration errors
if +E is varied within 106 GeV < +E < 0.03 GeV for fixed + = 0.001 or if + is
varied within 3 105 < + < 0.1 for fixed +E = 0.01 GeV.
2.3.3. Initial-state radiation in O() and beyond
In O() the effect of ISR is entirely contained in the radiative corrections described
above. However, the emission of photons collinear to the incoming electrons or positrons
leads to corrections that are enhanced by large logarithms. In order to achieve an accuracy
at the few 0.1% level, the corresponding higher-order contributions, i.e., contributions
beyond O(), must be taken into account. This can be done in the structure-function
method [45,46]. According to the mass-factorization theorem, the leading-logarithmic (LL)
initial-state QED corrections can be written as a convolution of the lowest-order cross
section with structure functions, and the corresponding differential cross section reads
1
dISR,LL =
1
dx1
dx2 eeLL x1 , Q2 eeLL x2 , Q2

d0 (x1 p1 , x2 p2 ).
(2.43)
Here x1 and x2 denote the fractions of the longitudinal momentum carried by the incoming
electron and positron momenta just before the hard scattering process occurs. This means
that the incoming momenta p before emission of the collinear photon are rescaled by
x1,2 , and the CM frame of the hard scattering process with the lowest-order cross section
d0 (x1 p1 , x2 p2 ) is boosted along the beam axis. The LL structure function including
O( 3 ) terms is given by [46]
exp 12 e E + 38 e e
e
e
LL
2
ee x, Q =
(1 x) 2 1 (1 + x)

1
2
4
1 + 2 e

2 1 + 3x 2
ln(x) + 4(1 + x) ln(1 x) + 5 + x
e
32 1 x

e3
(1 + x) 6 Li2 (x) + 12 ln2 (1 x) 3 2

384

3
1
+
1 + 8x + 3x 2 ln(x) + 6(x + 5)(1 x) ln(1 x)
1x 2

1
+ 12 1 + x 2 ln(x) ln(1 x) 1 + 7x 2 ln2 (x)
2

1
+ 39 24x 15x 2
(2.44)
4
309
with
2
(L 1)
and the leading logarithm

e =
L = ln
(2.45)
Q2
.
m2e
(2.46)
Note that the scale Q2 is not fixed within LL approximation, but has to be set to a typical
scale of the underlying process; for the numerics we use Q2 = s. In (2.44) E is the Euler
constant and (y) the gamma function, which should not be confused with the structure
functions. Note that some non-leading terms are incorporated, taking into account the fact
that the residue of the soft-photon pole is proportional to L 1 rather than L for the initialstate photon radiation.
We add the cross section (2.43) to the one-loop result and subtract the lowest-order and
one-loop contributions dISR,LL,1 already contained within this formula,
1
dISR,LL,1 =
dx1 dx2 (1 x1 )(1 x2 ) + eeLL,1 x1 , Q2 (1 x2 )
+ (1 x1 )eeLL,1
x2 , Q

d0 (x1 p1 , x2 p2 ), (2.47)
in order to avoid double counting. The one-loop contribution to the structure function reads

e 1 + x 2
eeLL,1 x, Q2 =
4 1x +

1 + x2
3
e
.
lim (1 x)
+ 2 ln 0 + (1 x 0)
=
(2.48)
4 00+
2
1x
Note that the uncertainty that is connected with the choice of Q2 enters only in O( 2 ), if
all O() corrections, including constant terms, are taken into account.
non-photonic
of (2.20) with
Finally, we complete the definition of the IBA by dressing dIBA
the ISR structure functions,
1
dIBA =
1
dx1
dx2 eeLL x1 , Q2 eeLL x2 , Q2
0
non-photonic
dIBA
(x1 p1 , x2 p2 ).
(2.49)
2.3.4. Monte Carlo integration

The phase-space integration is performed with Monte Carlo techniques in both
computer codes. The first code employs a multi-channel Monte Carlo generator similar
to the one implemented in RacoonWW [42,47] and Lusifer [9], the second one uses the
adaptive multi-dimensional integration program VEGAS [48].
310
3. Numerical results
3.1. Input parameters
For the numerical evaluation we use the following set of SM parameters [49],
G = 1.16639 105 GeV2 ,
MW = 80.423 GeV,
MZLEP
me = 0.510998902 MeV,
(0) = 1/137.03599976,
= 91.1876 GeV,
ZLEP = 2.4952 GeV,
m = 105.658357 MeV,
mu = 66 MeV,
mc = 1.2 GeV,
mt = 174.3 GeV,
md = 66 MeV,
ms = 150 MeV,
mb = 4.3 GeV,
m = 1.77699 GeV,
(3.1)
which coincides with the one used in Ref. [21]. Since we employ the on-shell renormalization scheme, the weak mixing angle is fixed by (2.11).
As discussed in Section 2.2.4, and if not stated otherwise, we evaluate amplitudes in the
so-called G -scheme, i.e., we derive the electromagnetic coupling = e2 /(4) from the
Fermi constant G according to (2.18). This procedure, in particular, absorbs all sizeable
mass effects of light fermions other than electrons in the coupling G , and the results are
practically independent of the masses of the light quarks. The masses of the light quarks
are adjusted to reproduce the hadronic contribution to the photonic vacuum polarization of
Ref. [50]. In the relative radiative corrections, we use (0) as coupling parameter, which is
the correct effective coupling for real photon emission. We do not calculate the W-boson
mass from G but use its experimental value as input.
As explained in Section 2.2.3, we employ a fixed width in the resonant Z-boson
propagator in contrast to the approach used at LEP to fit the Z resonance, where a running
width is taken. Therefore, we have to convert the on-shell values of MZLEP and ZLEP ,
resulting from LEP, to the pole values denoted by MZ and Z in this paper. The relation
of the two sets of values is given by [34]

2

MZ = MZLEP 1 + ZLEP /MZLEP = 91.1535 GeV,

2

LEP
1 + ZLEP /MZLEP = 2.4943 GeV,
Z = Z
(3.2)
i.e., the difference is formally of two-loop order and numerically hardly visible in the
results presented below.
3.2. Results on total cross sections
In Ref. [21] we have already discussed numerical results for the total cross sections of
the process e+ e H and the corresponding radiative corrections relative to the pure
lowest-order prediction. Particular attention has been paid to the individual contributions
of the various sources of corrections, such as the contributions of the closed fermion loops,
of the ISR effects in O() and beyond, and of the remaining bosonic corrections. In this
discussion the ZH-production and WW-fusion channels have been considered separately,
thereby revealing characteristic differences of the two channels. In the following we
311
continue the discussion of Ref. [21]. We always sum over all three neutrino species, i.e.,
over the processes e+ e e e H, H, and H. Besides the full cross section,
denoted total in the plots, in some cases we also give the cross section resulting from
the ZH-production and WW-fusion channels separately, which are referred to as ZH and
WW contributions, respectively. In the ZH-production channel we sum over the relevant
contributions of all H final states, which is equivalent to multiplying the cross sections
for e+ e H by a factor 3.
Fig. 11 shows the total cross sections for the ZH and WW channels, as well as their
incoherent (ZH + WW) and coherent (total) sums, in improved Born approximation
(IBA) together with theradiative corrections relative to the IBA as functions of the centreof-mass (CM) energy s for the fixed Higgs-boson mass MH = 150 GeV. Analogous
results have been given in Ref. [21] for the pure lowest-order cross sections tree and
the corresponding relative corrections (see Figs. 1 and 2 there). While the absolute cross
sections IBA and tree look qualitatively similar, the relative corrections normalized to
the IBA are systematically smaller as compared to a normalization to the pure lowestorder cross section. This is mainly due to the dominance of the ISR corrections which are
properly taken into account by the IBA. In the WW channel the IBA describes the corrected
cross section
within 1% for CM energies up to 500 GeV and remains still good within
23% up to s 1 TeV. At high CM energies the IBA misses non-universal bosonic
corrections, the size of which grows with energy (compare Fig. 3 of Ref.
[21]). In the ZH
channel the IBA deviates from the corrected cross section by 3% for s 500 GeV, but
even this reasonably good approximation results from accidental compensations between
fermionic and bosonic non-ISR corrections, which are both of the order of 510% but
of opposite sign (see Fig. 3 of Ref. [21]). Above a CM energy of 500 GeV, where ZH
production is suppressed, the IBA becomes even worse, since the dominating Sudakov
2 ) are missing in the IBA.
logarithms, such as ln2 (s/MW
Fig. 11. Improved Born approximation (IBA) for the total cross sections and radiative corrections relative to the
IBA for MH = 150 GeV.
312
Fig. 12. Cross sections and relative corrections as function of the Higgs-boson mass for
s = 500 GeV.
In Fig. 12 the cross sections and their corrections are shown as functions of the
Higgs-boson mass MH for the typical LC energy of s = 500 GeV. For this CM
energy the contribution of WW fusion is much larger than the one of ZH production for
MH 300 GeV. For MH s MZ 409 GeV the ZH-production channel is strongly

suppressed since the Z-boson cannot become resonant there. The total cross section falls
below 1 fb in this domain. For MH 350 GeV the relative corrections to the total cross
section and to the WW-fusion channel are of the order 10% to 20% if normalized to
tree , but only at the level of 2% if normalized to IBA . For the ZH-production channel
the reduction of the relative corrections with respect to IBA is similar. The spikes at
MH = 2MW , 2MZ , 2mt result from thresholds. These singularities are avoided if the finite
widths of the unstable particles are taken into account appropriately (see, for instance,
Ref. [51]).
313
3.3. Results on distributions

The distributions
in the Higgs-boson energy EH (defined in the CM frame) are depicted
in Fig. 13 for s = 500 GeV and MH = 150 GeV. The broad continuous distribution
for EH 250 GeV is almost entirely due to WW fusion, while the resonance peak at
EH 265 GeV corresponds to the Higgs-boson energy in the 2 2 kinematics of the ZHproduction process with an on-shell Z boson. This explains the behaviour of the radiative
corrections, which are dominated by ISR. Since the WW-fusion cross section rises with
energy, the continuous part receives negative corrections. For the WW-fusion channel these
are particularly large for large EH near the phase space boundary. On the other hand, the
Z peak is reduced by ISR and produces a radiative tail for EH values below the peak, since
ISR effectively reduces the scattering energy of the subsequent ZH-production process.
This radiative tail leads to corrections of up to 220%, as can be seen in the inset of the
Fig. 13. Distribution in the Higgs-boson energy EH and corresponding relative radiative corrections for
s = 500 GeV and MH = 150 GeV.
314
lower left plot in Fig. 13. The relative corrections are reduced drastically if normalized to
the IBA, which takes care of the large ISR effects. For EH 250 GeV the corrections to
the total improved-Born cross section vary only weakly with EH .
Fig. 14 illustrates the distribution in the cosine of the Higgs-boson production angle H
(defined in the CM frame) for s = 500 GeV and MH = 150 GeV. The peak behaviour in
the very forward and backward directions is due to the dominant WW contribution, while
ZH production follows a shape roughly proportional to 1 cos2 H . The relative corrections
to the WW contribution depend only weakly on cos H and reflect the same reduction
after normalization to the IBA as the integrated cross sections. The relative corrections
to the ZH contribution become large in the forward and backward directions, where the
corresponding lowest-order cross section is small.
Fig. 14. Distribution in the cosine of the Higgs-boson production angle H and corresponding relative radiative
corrections for s = 500 GeV and MH = 150 GeV.
315
Fig. 15. Distribution in the photon energy E and in the photon polar angle cos in the radiative process
+ for s = 500 GeV and MH = 150 GeV.

e+ e H
Finally, we show the spectra of the photon energy E and the photon polar angle
+
cos
(both defined in the CM frame) of the radiative process e e H + for
s = 500 GeV and MH = 150 GeV (see Fig. 15). In order to make the photon visible,
we impose angular and energy cuts of
( , beam) > 1 ,
E > 0.1 GeV.
(3.3)
For small photon energies the infrared pole, which is the same for the WW and ZH
channels, dominates the spectrum. Thus, WW fusion dominates for small E , since the
lowest-order cross section is much larger for WW fusion than for ZH production. For larger
E , small-angle photon emission is dominant as some kind of remnant of the collinear
pole for forward and backward emission. This means that larger E reflect the lowestorder cross sections at smallerenergies. Since the ZH-production cross section rises with
decreasing CM energies for s 500 GeV, but the WW-fusion cross section falls off
steeply, ZH production takes over the leading role for E 110 GeV.
3.4. Comparison to related work
Adapting the input parameters and the parametrization of the lowest-order matrix
element to those used by Belanger et al. [18], we reproduced the numbers for the total
cross section given in Table 2 of the first paper of Ref. [18]. Note that we switch off the
ISR beyond O() in this comparison. In Table 1 we list for each Higgs-boson mass and
the corresponding calculated W-boson mass the results of Ref. [18]1 together with our
results. The numbers in parenthesis indicate the errors in the last digits. The agreement is
good; the relative differences are with one exception below 104 for the total lowest-order
1 According to F. Boudjema, the numbers for the lowest-order cross section in Table 2 of Ref. [18] have
integration errors of the order of 0.2%. Table 1 contains updated numbers obtained with increased statistics.
316
Table 1
Total cross section in lowest order and including the full O() corrections and the relative corrections for
s = 500 GeV and various Higgs masses for the input parameter scheme of Ref. [18]
MH (GeV)
MW (GeV)
tree (fb)
(fb)
150
80.3767
200
80.3571
250
80.3411
300
80.3275
350
80.3158
61.074(7)
61.076(5)
37.294(4)
37.293(3)
21.135(2)
21.134(1)
10.758(1)
10.7552(7)
4.6079(5)
4.6077(2)
60.99(7)
60.80(2)
37.16(4)
37.09(2)
20.63(2)
20.60(1)
10.30(1)
10.282(4)
4.184(4)
4.181(1)
/tree 1 (%)
0.2
0.44(3)
0.4
0.56(4)
2.5
2.53(3)
4.2
4.40(3)
9.1
9.27(3)
Ref. [18]
This work
Ref. [18]
This work
Ref. [18]
This work
Ref. [18]
This work
Ref. [18]
This work
cross section and below 0.3% for the corrected cross section. The corrections relative to
the lowest-order cross section agree within 0.2%. This is of the order of the statistical
error of Ref. [18], which is about 0.1%. Note that Belanger et al. use (0) to parametrize
the lowest-order cross section. As a consequence their relative corrections are shifted by
3+r +9% compared to those in the G -scheme.
We have also reproduced the cos H and EH distributions in Figs. 1 and 2 of the first
paper of Ref. [18]. We found agreement within the accuracy of these figures.
When considering only fermion-loop corrections, we find agreement with the calculations of Refs. [19,20], once the appropriate renormalization and input-parameter schemes
are adopted. For more details on this comparison we refer to Ref. [21].
4. Summary
We have presented a calculation of the complete electroweak O() radiative corrections
to the single Higgs-boson production process e+ e H in the electroweak Standard
Model. For e+ e e e H, where ZH production and W-boson fusion contribute, both
production channels are added coherently. Two methods for the treatment of the finite
Z-boson width have been introduced. We have taken special care to treat the contributions
from 5-point tensor integrals in a numerically stable way. The complete single-hardphoton matrix elements have been taken into account, where soft and collinear singularities
are treated both in the subtraction and the phase-space slicing methods. Higher-order
ISR corrections have been included in the structure-function approach. The phase-space
integration is performed with Monte Carlo techniques.
We find that the electroweak corrections are of the order of 10% and are dominated
by ISR corrections if the lowest-order matrix element is parametrized with the Fermi
constant G . The non-ISR corrections are at the level of a few per cent in the G -scheme,
but are of the order of 10% in other schemes. At high energies, where the WW-fusion
channel dominates, the electroweak corrections depend only weakly on the energy and the
production angle of the Higgs-boson.
317
Although the sum of ISR and non-ISR is accidentally small in the (0)-scheme, the G scheme is nevertheless preferable for the following reasons. In contrast to the (0) scheme,
it does not suffer from uncertainties arising from the hadronic vacuum polarization at low
energies. Since the cancellation between ISR and non-ISR corrections in the (0) scheme
at one loop is accidental, it cannot be expected that it still holds in higher orders. On the
other hand, the G -scheme resums the leading universal corrections associated with the
running of the electromagnetic coupling and the universal corrections proportional to the
square of the top mass.
We have shown that the corrections to the WW-fusion channel can be described by a
simple improved Born approximation within an accuracy of typically 1% (3%) for CM
energies below 500 GeV (1 TeV). For the ZH-production channel the improved Born
approximation, which is simply based on ISR, approximates the corrected cross section
within 3% up to CM energies of about 500 GeV, but becomes worse at higher energies. In
summary, the approximation can be used to include radiative corrections at the qualitative
level, but a precision analysis will require the inclusion of the full O() correction.
Acknowledgements
We thank the authors of Refs. [1820] for further information about their results and,
in particular, F. Boudjema for sending us more precise numbers. This work was supported
in part by the Swiss Bundesamt fr Bildung und Wissenschaft and by the European Union
under contract HPRN-CT-2000-00149.
Appendix A. Standard matrix elements

ZH, and M
WW
The four-dimensionality of spacetime implies that the SME M
i
i
introduced in Section 2.1 are not all independent; there are linear relations among them.
A simple way to derive the relations with real coefficients is provided by the following
trick. In four dimensions the metric tensor can be decomposed in terms of four independent
orthonormal four-vectors nl ,
g = n0 n0
3
nl nl ,
(A.1)
l=1
where nk nl = gkl . Two convenient choices (j = 1, 2) for the vectors nl are given by
1
1
n1 = (p1 p2 ) ,
n0 = (p1 + p2 ) ,
s
s

t2j t1j
s
p1 +
p2 ,
kj +
n2 =
t1j t2j
s
s
2
0 p1, p2, kj,
n3 = 0 n0, n1, n2, =
st1j t2j
(A.2)
318
with 0 (0 0123 = +1) denoting the totally antisymmetric tensor. Inserting this
decomposition for both j = 1, 2 in all contractions between different Dirac chains
according to ee, , = ee, g , etc., and subsequently using the Dirac equation
and the Chisholm identity
i0 5 = g + g g ,
(A.3)
ZH, to four and all WW SME M

WW to two SME.
reduces all ZH SME M
i
i
Further relations with complex coefficients result from the direct application of
and subsequently using the
the Chisholm identity (A.3) to structures like k
1 p1 k2
decomposition
g =
4

1
Z ij 2pi pj ,
Zij = 2pi pj ,
p3 = k1 ,
p4 = k2 ,
(A.4)
i,j =1
to separate all contractions between 0 tensors and Dirac chains via 0 =

, in particular, this leads to
g 0 . For k
1 p1 k2

0 k1 p1 k2 = g 0 k1 p1 k2 = 2X p1 Z 1 12 + p2 Z 1 22
(A.5)
with
X = 0 p1 p2 k1 k2 = 0 p1, p2, k1, k2, .
(A.6)
(Z 1 ),
In the inverse matrix

the determinant det(Z) occurs, which can be identified with
det(Z) = 16X2 .
Altogether, the linear relations reduce the set of ZH SME to two and the set of WW
SME to one SME. Explicitly, the relations read
ZH,+ ,
ZH, = r ZH, M
M
i
i
1
WW ,
WW = r WW M
M
i
i
1
i = 2, . . . 13,
with
r2ZH,+ = s,
r2ZH, = 0,
r2WW = t11 ,
r3ZH,+ = 0,
r3ZH, = s12 ,
1
r4ZH, = C ,
2
t11
ZH,
r5
= ,
2
1
ZH,
r6
=
A ,
4t12
1
r7ZH, =
A,
4t21
t22
r8ZH, = ,
2
r3WW = 0,
1
r4WW = C,
2
s
WW
r5 = ,
2
1
r6WW =
B ,
4t12
1
r7WW =
B,
4t21
s12
r8WW =
,
2
r9ZH, = 4t11 ,
r9WW = 4s,
r4ZH,+ = 0,
1
A,
4t22
t21
r6ZH,+ = ,
2
t
12
r7ZH,+ = ,
2
1
A ,
r8ZH,+ =
4t11
1
r9ZH,+ =
A,
t22
r5ZH,+ =
(A.7)
A ,
B ,
ZH,+
r10
= 2t21,
ZH,
r10
=
ZH,+
r11
= 2t12,
1
ZH,+
=
A
r12
t11
ZH,
r11
= 4t12 ,
WW
r11
= 4t12,
ZH,
r12
= 4t22 ,
WW
r12
= 4s12 ,
ZH,+
r13
= 4,
ZH,
r13
= 16,
WW
r13
= 16
t12
WW
r10
=
t12
319
(A.8)
and
A = ss12 t12 t21 t11 t22 + 4iX,
B = ss12 + t12 t21 t11 t22 + 4iX,
C = ss12 t12 t21 + t11 t22 + 4iX.
(A.9)
Finally, there are Fierz identities relating the ZH and WW spinor chains, supplementing
the above list of relations by
ZH, ,
WW = M
M
1
1
(A.10)
ZH, . In terms of Weylvan der Waerden

so that all SME can be expressed in terms of M
1
spinor products, which have been defined in Section 2.3.1 (see also Refs. [38]), these SME
read
ZH,+ = 2p2 k2 p1 k1 ,
M
1
ZH, = 2p1 k2 p2 k1 .

M
1
(A.11)
Since the lowest-order matrix elements for right- and left-handed electrons are
ZH, , respectively, the whole virtual one-loop contributions
ZH,+ and M
proportional to M
1
1
to the squared matrix element are of the form

ZH, 2
,

Re M1 M0
(A.12)
= f1 (s, tij , sij ) + f2 (s, tij , sij )X M
1
where fi are functions of scalar products of the external momenta. Note that the fi
ZH, |2 are invariant under the reflection of all outgoing momenta kj on the plane
and |M
1
spanned by the beam axis and the Higgs-boson momentum k3 , while X changes its sign
under this reflection. Consequently, the contribution proportional to X drops out after
integrating over the momenta of the final-state neutrinos, which are not observable.
References
[1] ATLAS Collaboration, ATLAS Detector and Physics Performance Technical Design Report, CERN-LHWW
99-14.
[2] CMS Collaboration, CMS Technical Proposal, CERN-LHWW 94-38.
[3] The LEP Working Group for Higgs Boson Searches, LHWG Note/2002-01.
[4] E. Accomando, et al., ECFA/DESY LC Physics Working Group Collaboration, Phys. Rep. 299 (1998) 1,
hep-ph/9705442.
[5] J.A. Aguilar-Saavedra, et al., TESLA Technical Design Report Part III: Physics at an e+ e Linear Collider,
hep-ph/0106315.
[6] K. Abe, et al., ACFA Linear Collider Working Group Collaboration, ACFA Linear Collider Working Group
Report, hep-ph/0109166.
320
[7] T. Abe, et al., American Linear Collider Working Group Collaboration, in: R. Davidson, C. Quigg (Eds.),
Proc. of the APS/DPF/DPB Summer Study on the Future of Particle Physics, Snowmass 2001, SLAC-R-570,
Resource book for Snowmass 2001, hep-ex/0106055, hep-ex/0106056, hep-ex/0106057, hep-ex/0106058.
[8] E. Accomando, A. Ballestrero, M. Pizzio, in: R. Settles (Ed.), e+ e Linear Colliders: Physics and Detector
Studies Part E, DESY 97-123E, Hamburg, 1997, p. 31, hep-ph/9709277;
E. Accomando, A. Ballestrero, M. Pizzio, Nucl. Phys. B 547 (1999) 81, hep-ph/9807515;
G. Montagna, M. Moretti, O. Nicrosini, F. Piccinini, Eur. Phys. J. C 2 (1998) 483, hep-ph/9705333;
F. Gangemi, G. Montagna, M. Moretti, O. Nicrosini, F. Piccinini, Eur. Phys. J. C 9 (1999) 31, hepph/9811437.
[9] S. Dittmaier, M. Roth, Nucl. Phys. B 642 (2002) 307, hep-ph/0206070.
[10] J.R. Ellis, M.K. Gaillard, D.V. Nanopoulos, Nucl. Phys. B 106 (1976) 292;
B.L. Ioffe, V.A. Khoze, Sov. J. Part. Nucl. 9 (1978) 50, Fiz. Elem. Chastits At. Yadra 9 (1978) 118.
[11] J.D. Bjorken, in: M.C. Zipf (Ed.), Weak Interactions at High-Energy and the Production of New Particles:
Proceedings of the 4th Slac Summer Institute on Particle Physics, SLAC-198, Stanford, 1976, p. 1.
[12] J. Fleischer, F. Jegerlehner, Nucl. Phys. B 216 (1983) 469.
[13] B.A. Kniehl, Z. Phys. C 55 (1992) 605.
[14] A. Denner, J. Kblbeck, R. Mertig, M. Bhm, Z. Phys. C 56 (1992) 261.
[15] F.A. Berends, R. Kleiss, Nucl. Phys. B 260 (1985) 32.
[16] D.R. Jones, S.T. Petcov, Phys. Lett. B 84 (1979) 440;
G. Altarelli, B. Mele, F. Pitolli, Nucl. Phys. B 287 (1987) 205;
W. Kilian, M. Krmer, P.M. Zerwas, Phys. Lett. B 373 (1996) 135, hep-ph/9512355;
E. Boos, M. Sachwitz, H.J. Schreiber, S. Shichanin, Int. J. Mod. Phys. A 10 (1995) 2067.
[17] F. Jegerlehner, O. Tarasov, hep-ph/0212004.
[18] G. Belanger, F. Boudjema, J. Fujimoto, T. Ishikawa, T. Kaneko, K. Kato, Y. Shimizu, hep-ph/0211268;
G. Belanger, F. Boudjema, J. Fujimoto, T. Ishikawa, T. Kaneko, K. Kato, Y. Shimizu, Phys. Lett. B 559
(2003) 252, hep-ph/0212261.
[19] H. Eberl, W. Majerotto, V.C. Spanos, Phys. Lett. B 538 (2002) 353, hep-ph/0204280;
H. Eberl, W. Majerotto, V.C. Spanos, Nucl. Phys. B 657 (2003) 378, hep-ph/0210038;
H. Eberl, W. Majerotto, V.C. Spanos, hep-ph/0210330.
[20] T. Hahn, S. Heinemeyer, G. Weiglein, Nucl. Phys. B 652 (2003) 229, hep-ph/0211204;
T. Hahn, S. Heinemeyer, G. Weiglein, hep-ph/0211384.
[21] A. Denner, S. Dittmaier, M. Roth, M.M. Weber, hep-ph/0301189, Phys. Lett. B, in press.
[22] W.F.L. Hollik, Fortschr. Phys. 38 (1990) 165.
[23] J. Kblbeck, M. Bhm, A. Denner, Comput. Phys. Commun. 60 (1990) 165;
H. Eck, J. Kblbeck, Guide to FeynArts 1.0, University of Wrzburg, 1992.
[24] G. Passarino, M. Veltman, Nucl. Phys. B 160 (1979) 151.
[25] G. t Hooft, M. Veltman, Nucl. Phys. B 153 (1979) 365.
[26] W. Beenakker, A. Denner, Nucl. Phys. B 338 (1990) 349.
[27] A. Denner, Fortschr. Phys. 41 (1993) 307.
[28] A. Denner, S. Dittmaier, hep-ph/0212259, Nucl. Phys. B, in press.
[29] W. Beenakker, S. Dittmaier, M. Krmer, B. Plmper, M. Spira, P.M. Zerwas, Nucl. Phys. B 653 (2003) 151,
hep-ph/0211352.
[30] A. Denner, S. Dittmaier, G. Weiglein, Nucl. Phys. B 440 (1995) 95, hep-ph/9410338.
[31] T. Hahn, Comput. Phys. Commun. 140 (2001) 418, hep-ph/0012260.
[32] T. Hahn, M. Perez-Victoria, Comput. Phys. Commun. 118 (1999) 153, hep-ph/9807565;
T. Hahn, Nucl. Phys. B (Proc. Suppl.) 89 (2000) 231, hep-ph/0005029.
[33] R. Mertig, M. Bhm, A. Denner, Comput. Phys. Commun. 64 (1991) 345.
[34] D.Y. Bardin, A. Leike, T. Riemann, M. Sachwitz, Phys. Lett. B 206 (1988) 539;
A. Sirlin, Phys. Rev. Lett. 67 (1991) 2127;
D. Wackeroth, W. Hollik, Phys. Rev. D 55 (1997) 6788, hep-ph/9606398;
W. Beenakker, et al., Nucl. Phys. B 500 (1997) 255, hep-ph/9612260.
[35] S. Dittmaier, M. Krmer, Phys. Rev. D 65 (2002) 073007, hep-ph/0109062.
[36] G. Burgers, F. Jegerlehner, in: G. Altarelli, R. Kleiss, C. Verzegnassi (Eds.), Z Physics at LEP 1, CERN
99-08, Geneva, 1989, p. 7.
321
[37] B.A. Kniehl, M. Steinhauser, Nucl. Phys. B 454 (1995) 485, hep-ph/9508241.
[38] S. Dittmaier, Phys. Rev. D 59 (1999) 016007, hep-ph/9805445.
[39] T. Stelzer, W.F. Long, Comput. Phys. Commun. 81 (1994) 357, hep-ph/9401258;
H. Murayama, I. Watanabe, K. Hagiwara, KEK-91-11.
[40] S. Catani, M.H. Seymour, Phys. Lett. B 378 (1996) 287, hep-ph/9602277;
S. Catani, M.H. Seymour, Nucl. Phys. B 485 (1997) 291, hep-ph/9605323;
S. Catani, M.H. Seymour, Nucl. Phys. B 510 (1997) 291, Erratum.
[41] S. Dittmaier, Nucl. Phys. B 565 (2000) 69, hep-ph/9904440.
[42] M. Roth, Ph.D. Thesis, ETH Zrich 13363 (1999), hep-ph/0008033.
[43] M. Bhm, S. Dittmaier, Nucl. Phys. B 409 (1993) 3;
M. Bhm, S. Dittmaier, Nucl. Phys. B 412 (1994) 39.
[44] D.R. Yennie, S.C. Frautschi, H. Suura, Ann. Phys. 13 (1961) 379.
[45] E.A. Kuraev, V.S. Fadin, Yad. Fiz. 41 (1985) 753, Sov. J. Nucl. Phys. 41 (1985) 466;
G. Altarelli, G. Martinelli, in: J. Ellis, R. Peccei (Eds.), Physics at LEP, CERN 86-02, Geneva, 1986, Vol. 1,
p. 47;
O. Nicrosini, L. Trentadue, Phys. Lett. B 196 (1987) 551;
O. Nicrosini, L. Trentadue, Z. Phys. C 39 (1988) 479;
F.A. Berends, G. Burgers, W.L. van Neerven, Nucl. Phys. B 297 (1988) 429;
F.A. Berends, G. Burgers, W.L. van Neerven, Nucl. Phys. B 304 (1988) 92, Erratum.
[46] W. Beenakker, et al., in: G. Altarelli, T. Sjstrand, F. Zwirner (Eds.), Physics at LEP2, CERN 96-01, Geneva,
1996, Vol. 1, p. 79, hep-ph/9602351.
[47] A. Denner, S. Dittmaier, M. Roth, D. Wackeroth, Nucl. Phys. B 560 (1999) 33, hep-ph/9904472.
[48] G.P. Lepage, J. Comput. Phys. 27 (1978) 192, and CLNS-80/447.
[49] K. Hagiwara, et al., Particle Data Group Collaboration, Phys. Rev. D 66 (2002) 010001.
[50] F. Jegerlehner, DESY 01-029, LC-TH-2001-035, hep-ph/0105283.
[51] T. Bhattacharya, S. Willenbrock, Phys. Rev. D 47 (1993) 4022;
B.A. Kniehl, C.P. Palisoc, A. Sirlin, Nucl. Phys. B 591 (2000) 296, hep-ph/0007002.

Natural R-parity, -term, and fermion mass

hierarchy from discrete gauge symmetries
K.S. Babu, Ilia Gogoladze 1 , Kai Wang
Department of Physics, Oklahoma State University Stillwater, OK 74078, USA
Received 31 December 2002; received in revised form 17 March 2003; accepted 25 March 2003
Abstract
In the minimal supersymmetric Standard Model with see-saw neutrino masses we show how Rparity can emerge naturally as a discrete gauge symmetry. The same discrete symmetry explains the
smallness of the -term (the Higgsino mass parameter) via the GiudiceMasiero mechanism. The
discrete gauge anomalies are cancelled by a discrete version of the GreenSchwarz mechanism. The
simplest symmetry group is found to be Z4 with a charge assignment that is compatible with grand
unification. Several other ZN gauge symmetries are found for N = 10, 12, 18, 36 etc., with some
models employing discrete anomaly cancellation at higher KacMoody levels. Allowing for a flavor
structure in ZN , we show that the same gauge symmetry can also explain the observed hierarchy in
the fermion masses and mixings.
1. Introduction
One of the challenging questions facing supersymmetric extensions of the Standard
Model is an understanding of R-parity which is required for the stability of the proton. In
the minimal supersymmetric Standard Model (MSSM), a discrete Z2 symmetry is usually
assumed. Under this symmetry the Standard Model (SM) particles are taken to be even
while their superpartners are odd. The gauge symmetry of MSSM would allow baryon
number and lepton number violating Yukawa couplings at the renormalizable level which
would result in rapid proton decay. The Z2 R-parity forbids such dangerous couplings.
E-mail addresses: babu@okstate.edu (K.S. Babu), ilia@hep.phy.okstate.edu (I. Gogoladze),
wk@okstate.edu (K. Wang).
1 On leave of absence from: Andronikashvili Institute of Physics, GAS, 380077 Tbilisi, Georgia.
doi:10.1016/S0550-3213(03)00258-X
K.S. Babu et al. / Nuclear Physics B 660 (2003) 322342
323
The assumption of R-parity has profound implications for supersymmetric particle

search at colliders as well as for cosmology. At colliders SUSY particles can only be
produced in pairs, and the lightest SUSY particle (LSP), usually a neutralino, will be stable.
This stable LSP is a leading candidate for cosmological cold dark matter.
Since R-parity is not part of the MSSM gauge symmetry, questions can be raised about
its potential violation arising from quantum gravitational effects. These effects (associated
with worm holes, black holes, etc.) are believed to violate all global symmetries [1]. True
gauge symmetries are however protected from such violations. When a gauge symmetry
breaks spontaneously, often a discrete subgroup is left intact. Such discrete symmetries,
called discrete gauge symmetries [2], are also immune to quantum gravitational effects.
Not all discrete symmetries can, however, be gauge symmetries. For instance, since
the original continuous gauge symmetry was free from anomalies, its unbroken discrete
subgroup should be free from discrete gauge anomalies [3,4]. This imposes a nontrivial constraint on the surviving discrete symmetry and/or on the low energy particle
content [28].
It will be of great interest to see if R-parity of MSSM can be realized as a discrete
gauge symmetry, so that one can rest assured that it wont be subject to unknown quantum
gravitational violations. This is the main question we wish to address in this paper.
A seemingly unrelated but equally profound problem facing the MSSM is an
understanding of the origin of the -term, the Higgsino mass parameter. The parameter is
defined through the superpotential term W Hu Hd , where Hu and Hd are the two Higgs
doublet superfields of MSSM. Since the -term is SUSY-preserving and is a singlet of the
SM gauge symmetry, its natural value would seem to be of order the Planck scale. But
102 GeV is required for consistent phenomenology. It will be desirable, and is often
assumed, that the term is related to the supersymmetry breaking scale. An attractive
scenario which achieves this is the GuidiceMasiero mechanism [9] wherein a bare
term in the superpotential is forbidden by some symmetry. It is induced through a higherdimensional Lagrangian term

Hu Hd Z
L = d4
(1)
,
MPl
where Z is a spurion field which parameterizes supersymmetry breaking via FZ = 0, with
FZ /MPl MSUSY 102 GeV. For this mechanism to work, there must exist a symmetry
that forbids a bare term in the superpotential. Such a symmetry cannot be a continuous
symmetry, consistent with the requirement of non-zero gaugino masses, and therefore must
be discrete.2 It would be desirable to realize this as a discrete gauge symmetry.
The purpose of this paper is show that it is possible to realize ZN symmetries as discrete
gauge symmetries which act as R-parity and which solve simultaneously the -problem
via the GuidiceMasiero mechanism. We make use of a discrete version of the Green
Schwarz mechanism [10] for anomaly cancellation. Simple realizations of R-parity as a
discrete gauge symmetry are possible which also solve the -problem, without enlarging
2 Without the -term and the gaugino mass term, MSSM Lagrangian has two U (1) symmetries, a Peccei
Quinn symmetry and a U (1)R symmetry. The -term breaks the PecceiQuinn symmetry and the gaugino mass
term breaks the U (1)R symmetry down to a discrete subgroup.
324
the particle content of MSSM. The simplest symmetry group we have found is a Z4 . Under
this Z4 all MSSM matter superfields and the gauginos carry equal charge of 1 while
the Higgs superfields have zero charge. Such a simple charge assignment is compatible
with grand unification. This charge assignment is anomaly-free by virtue of the discrete
GreenSchwarz mechanism. Other ZN symmetries with N = 10, 12, 18, 36, etc. are also
identified, some realized at higher KacMoody level. Either lepton parity or baryon parity
can be obtained as a discrete symmetry in this approach, with baryon parity requiring
anomaly cancellation at higher KacMoody levels. By allowing for a family-dependent
structure in ZN , we show how it is possible in our framework to explain the observed
fermion mass and mixing hierarchy in a simple way.
Attempts have been made in the past to derive R-parity as a discrete gauge symmetry
within MSSM. Early analysis [5,6] did not incorporate the see-saw mechanism for neutrino
mass or the GuidiceMasiero mechanism for generating the -term. A recent analysis
which includes these features [7] has found Z9 and Z18 discrete gauge symmetries as
possible candidates for R-parity by demanding these symmetries to be anomaly-free. It
turns out that in these models [7] the Kahler potential violates R-parity at higher order,
leading to cosmologically disfavored lifetime for the neutralino LSP. Furthermore, the
discrete charge assignment in these models was not compatible with grand unification
with the MSSM spectrum. The main difference in our approach is that we make use of
the GreenSchwarz mechanism for discrete anomaly cancellation, which is less restrictive
compared to the straightforward methods. The outcome differs in several ways, notably
in the realization of simpler symmetries (e.g., Z4 ), exact R-parity without cosmological
problems, and compatibility with grand unification. It should be mentioned that enlarging
the particle content of MSSM has been proposed as a solution to the R-parity and
problems [8]. In contrast to such approaches, in our framework, the low energy spectrum
is identical to that of MSSM.
This paper is organized as follows. In Section 2 we review briefly the discrete version
of GreenSchwarz anomaly cancellation mechanism. Section 3 contains our main results.
In Section 3.1 we write down the constraints arising from the Lagrangian of MSSM and
the discrete anomaly cancellation conditions. In Section 3.2 we identify possible discrete
gauge symmetries at KacMoody level 1 which prevent R-parity violating terms. In
Section 3.3 we embed these symmetries to a higher ZN to solve the -problem. Section 3.4
is devoted to solutions based on higher KacMoody levels. In Section 4 we show how a
simple discrete gauge symmetry can explain the fermion mass and mixing angle hierarchy.
Finally we conclude in Section 5.
2. Discrete anomaly cancellation via GreenSchwarz mechanism

Let us first recall the essence of the GreenSchwartz (GS) anomaly cancellation
mechanism for a U (1) gauge symmetry. String theory when compactified to four
dimensions generically contains an anomalous U (1)A gauge symmetry. A subset of the
gauge anomalies in the axial vector U (1)A current can be cancelled via the GS mechanism
in the following way [10]. In four dimensions, the Lagrangian for the gauge boson kinetic
energy contains the terms

i ,
Lkinetic = (x)
ki Fi2 + i(x)
ki Fi F
i
325
(2)
where (x) denotes the string dilaton field and (x) is its axionic partner. The sum i
runs over the different gauge groups in the model, including U (1)A . ki are the Kac
Moody levels for the different gauge groups, which must be positive integers for the nonAbelian groups, but may be non-integers for Abelian groups. The GS mechanism makes
use of the transformation of the string axion field (x) under a U (1)A gauge variation
VA VA + (x),
(x) (x) (x)GS ,
(3)
where GS is a constant. If the anomaly coefficients involving the U (1)A gauge boson and
any other pair of gauge bosons are in the ratio
A1 A2 A3
=
=
= GS ,
k1
k2
k3
(4)
these anomalies will be cancelled by gauge variations of the U (1)A field arising from the
second term of Eq. (2). GS in Eq. (4) is also equal to the mixed gravitational anomaly,
GS = Agravity/12 [11]. All other crossed anomaly coefficients should vanish, since they
cannot be removed by the shift in the string axion field.
Consider the case when the gauge symmetry in four dimensions just below the string
scale is SU(3)C SU(2)L U (1)Y U (1)A . Let A3 and A2 denote the anomalies
associated with [SU(3)C ]2 U (1)A and [SU(2)L ]2 U (1)A , respectively. Then if
A3 /k3 = A2 /k2 = GS is satisfied, from Eq. (4), it follows that these mixed anomalies
will be cancelled. The anomaly in [U (1)2Y ] U (1)A can also be cancelled in a similar
way if A1 /k1 = GS . However, in practice, this last condition is less useful, since k1 is not
constrained to be an integer as the overall normalization of the hypercharge is arbitrary. If
the full high energy theory is specified, there can be constraints on A1 as well. For example,
if hypercharge is embedded into a simple group such as SU(5) or SO(10), k1 = 5/3 is fixed
since hypercharge is now quantized. A1 /k1 = GS will provide a useful constraint in this
case. We shall remark on this possibility in our discussions. Note also that cross anomalies
such as [SU(3)] [U (1)A ]2 are automatically zero in the Standard Model, since the trace
of SU(N) generators is zero. Anomalies of the type [U (1)Y ] [U (1)A ]2 also suffer from
the same arbitrariness from the Abelian levels k1 and kA . Finally, [U (1)A ]3 anomaly can be
cancelled by the GS mechanism, or by contributions from fields that are complete singlets
of the Standard Model gauge group.
The anomalous U (1)A symmetry is expected to be broken just below the string scale.
This occurs when the FayetIliopoulos term associated with the U (1)A symmetry is
cancelled, so that supersymmetry remains unbroken near the string scale, by shifting the
matter superfields that carry U (1)A charges [12]. Although the U (1)A symmetry is broken,
a ZN subgroup of U (1)A can remain intact. Suppose that we choose a normalization
wherein the U (1)A charges of all fields are integers. (This can be done so long as all
the charges are relatively rational numbers.) Suppose that the scalar field which acquires
a vacuum expectation value (VEV) and breaks U (1)A symmetry has a charge N under
326
U (1)A in this normalization. A ZN subgroup is then left unbroken down to low energies.
We shall identify R-parity of MSSM with this unbroken ZN symmetry.
The field that acquires a VEV and breaks U (1)A to ZN can supply large masses of
order the string scale to a set of fermions which have Yukawa couplings involving this
field. Such fields may include Majorana fermions and Dirac fermions. These heavy fields
can carry SM gauge quantum numbers, but they must transform vectorially under the SM.
In order that their mass terms be invariant under the unbroken ZN , it must be that
2qi = 0 mod N (Majorana fermion),
qi + qi = 0 mod N (Dirac fermion),
(5)
where qi are the U (1)A charges of these heavy fermions. The index i is a flavor index
corresponding to different heavy fields. These heavy fermions, being chiral under the
U (1)A , contribute to gauge
anomalies. Their
contribution to the [SU(3)C ]2 U (1)
A gauge
anomaly is given
by
A
=
q
=
(N/2)
p
(Majorana
fermion)
or
A
=
3
3
i i i
i i i
i (qi +

qi ) i = (N) i pi i (Dirac fermion) where i is the quadratic index of the relevant
fermion under SU(3)C and the pi are integers. We shall adopt the usual normalization
of = 1/2 for fundamental of SU(N). Then, for the case of heavy Dirac fermion, one has
A3 = p(N/2) where p is an integer, as the index of the lowest-dimensional (fundamental)
representations is 1/2 and those of all other representations are integer multiples of 1/2.
The same conclusion follows for the case of Majorana fermions for a slightly different
reason.
All real representations of SU(3)C (such as an octet) have integer values of ,
so that i pi i is an integer. Analogous conclusions follow for the [SU(2)L ]2 U (1)A
anomaly coefficient.
If the ZN symmetry that survives to low energies was part of U (1)A , the ZN charges
of the fermions in the low energy theory must satisfy a non-trivial condition: the anomaly
coefficients Ai for the full theory is given by Ai from the low energy sector plus an integer
multiple of N/2. These anomalies should obey Eq. (4), leading to the discrete version of
the GreenSchwarz anomaly cancellation mechanism:
A2 + p2 N2
A3 + p12N
=
k3
k2
(6)
with p1 , p2 being integers. Since GS is an unknown constant (from the effective low
energy point of view), the discrete anomaly cancellation conditions of Eq. (6) are less
stringent than those arising from conventional anomaly cancellations. If GS = 0 in Eq. (6),
the anomaly is cancelled without assistance from the GreenSchwarz mechanism. We shall
not explicitly use the condition that GS = 0, so our solutions will contain those obtained
by demanding GS = 0 in Eq. (6), viz., A3 = p1 (N/2), A2 = p2 (N/2) with p1 , p2
being integers.
In our analysis we shall not explicitly make use of the condition A1 /k1 = A2 /k2 , since,
as mentioned earlier, the overall normalization of hypercharge is arbitrary. However, once
a solution to the various ZN charges is obtained, we can check for the allowed values k1 ,
and in particular, if k1 = 5/3 is part of the allowed solutions. This will be an interesting
case for two reasons. If hypercharge is embedded in a simple grand unification group
such as SU(5), one would expect k1 = 5/3. Even without a GUT embedding k1 = 5/3
327
Table 1
The B L charges of the Standard Model fields along with the unbroken Z6 subgroup after the neutrino see-saw
Field
U(1)BL
Z6
uc
dc
ec
Hu
Hd
1/3
1
1/3
5
1/3
5
1
3
1
3
1
3
0
0
0
0
is interesting. We recall that unification of gauge couplings is a necessary phenomenon in

string theory. Specifically, at tree level, the gauge couplings of the different gauge groups
are related to the string coupling constant gst which is determined by the VEV of the dilaton
field as [13]
k3 g32 = k2 g22 = k1 g12 = gst2 ,
(7)
where ki are the levels of the corresponding KacMoody algebra. In particular, if

k1 : k2 : k3 = 5/3 : 1 : 1, we would have sin2 W = 3/8 at the string scale, a scenario identical
to that of conventional gauge coupling unification with simple group such as SU(5). For
these reasons, we shall pay special attention to the case k1 = 5/3.
An interesting example of a discrete gauge symmetry in the MSSM (or the SM) with
see-saw neutrino masses is the Z6 subgroup of B L. The introduction of the right-handed
neutrino for generating small neutrino masses makes B L a true gauge symmetry. When
the c fields acquire super-large Majorana masses, U (1)BL breaks down to a discrete
Z6 subgroup. The Z6 charges of the MSSM fields arising from B L are displayed in
Table 1. Here we have used the standard notation for the fermion fields (Q and L being
the left-handed quark and lepton doublets, uc , d c being the quark singlets, and ec , c
being the (conjugates of) the right-handed electron and the right-handed neutrino singlets).
To obtain the unbroken Z6 charge, we first multiply the B L charge by 3 so that they
become integers, then observe that the c c Majorana mass term carries 6 units of this
integer B L charge. Thus this Z6 subgroup is left unbroken.
It is worth mentioning that the Z6 symmetry has a Z2 and a Z3 subgroups as well. In
the analysis that follows in the next section we will be making connections with the Z6
subgroup of B L and its Z2 and Z3 subgroups.
Anomalous U (1) symmetry has found applications in addressing the fermion mass and
mixing hierarchy problem [14], doublettriplet splitting problem in GUT [15], the strong
CP problem [16], the problem of SUSY [17] and for SUSY breaking [18].
3. Discrete gauge symmetries in the MSSM

In this section we turn to the identification of discrete gauge symmetries in the
MSSM that can serve as R-parity and simultaneously explain the origin of the term.
We stay with the minimal particle content of MSSM, with the inclusion of the righthanded neutrinos needed for generating neutrino masses via the see-saw mechanism [19].
Anomalies associated with the discrete gauge symmetry will be cancelled by the Green
Schwarz mechanism as discussed in Section 2.
328
Table 2
The matter superfields of MSSM along with their anomalous U (1) charges. is the Grassmann variable. Z is the
spurion field which is responsible for supersymmetry breaking
Q
Field
SU(3)C
SU(2)L
U (1)Y
U (1)A
3
2
1/6
q
uc
3
1
2/3
u
dc
3
1
1/3
d
ec
Hu
Hd
1
2
1/2
l
1
1
1
e
1
1
0
n
1
2
1/2
h
1
2
1/2
h
1
1
0
1
1
0
z
3.1. Constraints from the Lagrangian and discrete anomalies

We have displayed in Table 2 the particle content of MSSM along with their charges
under an anomalous U (1) gauge symmetry. The Grassmanian variable also carries a
charge (equal to ), which allows for the U (1) to be identified as an R symmetry. Z is a
spurion superfield that acquires a non-zero F component and breaks supersymmetry with
FZ /MPl MSUSY 102 GeV. We shall assume family-independent U (1) symmetry in
this section. Any unbroken discrete symmetry must be family-independent to be consistent
with MSSM phenomenology, that is the reason for focusing on such symmetries. In
Section 4 we shall extend this analysis to flavor-dependent symmetries, even in that case,
we will demand that a flavor-independent ZM symmetry is left intact.
The superpotential of the model, including small neutrino masses via the see-saw
mechanism is
W = Quc Hu + Qd c Hd + Lec Hd + L c Hu + MR c c ,
(8)
where MR is the heavy right-handed neutrino Majorana mass. We have suppressed Yukawa
couplings and generation indices, which must be understood.

In order to avoid a supersymmetric term in the Lagrangian, L d 2 Hu Hd ,
so that the magnitude of may be related to the SUSY breaking scale, we impose the
condition
h + h = 2.
(9)
A -parameter of the right order

the GuidiceMasiero mechanism [9]
is induced through
Z
.
Invariance
of this term under the U (1)
via the Lagrangian term, L d 4 Hu Hd M
Pl
symmetry requires
h + h z = 0.
(10)
The gaugino masses arise through the Lagrangian term

Z
d 2 W W
(11)
MPl
once FZ =0 is induced. Combining the invariance of this term with that of the gauge
kinetic term d 2 W W , we see that the spurion field Z must have zero charge under the
U (1).3 It is clear that the simultaneous presence of the gaugino mass term and the term
3 SUSY breaking scalar masses are invariant under the U (1) symmetry and do not provide any constraint.
329
reduces the U (1) symmetry to a discrete subgroup ZN . Therefore, one has to start with a
discrete symmetry ZN with the spurion superfield Z having a charge 0 mod N . Under ZN ,
the conditions Eqs. (9)(11) become
z = 0 mod N,
h + h = 0 mod N,
h + h = 2 mod N,
(12)
which also implies that 2 = 0 mod N .

Based on the invariance of the Yukawa couplings of Eq. (8) under Zn and the conditions
listed in Eq. (12), we obtain the following set of constraint equations:
z = m1 N,
h + h = m2 N,
q + u + h = 2 + m3 N,
q + d + h = 2 + m4 N,
l + e + h = 2 + m5 N,
2n = 2 + m6 N,
l + n + h = 2 + m7 N,
(13)
where mi (i = 1 7) are all integers.

The discrete ZN anomaly coefficients for the SU(3)C and the SU(2)L gauge groups are
3
3
A3 = 3 + 3q + u + d,
2
2

3
1
9
A2 = 5 + q + l + h + h .
(14)
2
2
2
Here we note that the fermionic charge of the uc field, for example, relevant for the anomaly
coefficient, is (u ) since carries charge . A3 and A2 include contributions from
the gauginos as well. We shall cancel these anomalies by applying the GreenSchwarz
mechanism as given in Eq. (6).
Non-zero gauginos masses arise through the VEV FZ = 0 (see Eq. (11)). Let us denote
the ZN charge of FZ to be M. FZ = 0 breaks the original ZN symmetry down to a
subgroup ZM :
ZN ZM .
(15)
(It must be that M > 1 for an unbroken discrete symmetry to survive after SUSY breaking.)
Since M = z 2 = m1 N 2 where m1 is an integer, we have
n1
M
(16)
2
with n1 being an integer. Let N = n0 M where n0 is an integer. Since invariance of the
Yukawa couplings under the ZN symmetry requires invariance under the subgroup ZM ,
we can solve the constraints of Eqs. (13), (14) along with Eq. (6) to determine the various
=
330
charges by first confining to the invariance under the smaller group ZM . Under this ZM ,
a superpotential term Hu Hd will be allowed. Once a solution is found, we can embed
the ZM symmetry into a higher ZN symmetry that would forbid the term. Making
some change of variables, viz., n2 = n0 m2 , n4 = n0 m6 , n5 = n0 p1 , n6 = n0 p2 , and
applying the anomaly cancellation condition of Eq. (6), we obtain the charges of the various
fields from Eqs. (13), (14) as
z = Mn0 ,

M
n2 n6 n4 7n1
n5
+
n2 n3 +
n1 ,
3
2
6
b
3

2n2 + n6 n4 n1
n5
M
+
+
+ n1 ,
h = 3q + M
+
n2 + n3
3
2
6
b
3

M
n2 n6 n4 n1
n5
+
+
+
n2 + n3
+ n1 ,
u = 4q + M
3
2
6
b
3

n2 n6 n4 7n1
n5
M
n1 ,
d = 2q + M
+
n2 n3 +
3
2
6
b
3

M
n2 n6 2n1
n5
+
n2 + n3
+
+ n1 ,
l = 3q + M
3
3
b
3

n2 n6 n4 5n1
n5
2M
n1 ,
e = 6q + M
+
n2 n3 +
3
2
6
b
3

n4 + n1
,
n=M
2
n1
=M .
(17)
2
Here we have defined b k3 /k2 . The ni in Eq. (17) are all integers. A specific choice of
the integers ni will fix the charge assignment explicitly. We note that the terms proportional
to q in Eq. (17) are proportional to the SM hypercharge Y . One can remove these terms
and set q = 0 in Eq. (17) without loss of generality by making a shift of the ZM charges
proportional to Y . The quark doublet Q will then have zero charge under the unbroken
ZM . It should be kept in mind that to each solution we find, one can add ZM charges
proportional to Y to obtain equivalent solutions.
Based on Eq. (17), one can compute the anomaly coefficients under ZM . They are

h = 3q + M
3
A3 = M(n1 n2 ),
2
1
A2 = M(3n1 3n2 + bn6 ).
(18)
2b
Note that from the last of Eq. (17), we have 2 = 0 mod M. So the superpotential
is invariant under ZM . Also, under ZM , one has h + h = 0 mod M, so a term in the
superpotential is allowed by this symmetry. (Such a term will be forbidden when the ZM
symmetry is embedded into a higher ZN symmetry, which we shall do in Section 3.3.)
In order to avoid R-parity breaking couplings, the total charges of the corresponding
superpotential terms should be non-integer multiples of M, which puts extra constraints
331
on the ZM charges. There are four types of R-parity violating terms. Their ZM charges are
given by

M
5n1 2n2 + n6 n4
n5
c c c
+
n2 n3 +
n1 ,
u d d : M
6
3
2
b
3

n1 n4
LLec :
M
,
2

n1 n4
,
M
LHu :
2

n1 n4
QLd c : M
(19)
.
2
It is easy to show that the largest ZM symmetry is Z6k3 from Eq. (17). We shall now
find solutions to these sets of equations for various values of the parameter b = k3 /k2 .
3.2. GreenSchwarz anomaly cancellation at KacMoody level 1
The simplest possibility for the parameters k2 and k3 to take is k3 = k2 = 1, so that
b = 1 in Eqs. (17), (18). This is the case of KacMoody algebra realized at level 1. Since
the constraint equations depend only on the ratio b = k3 /k2 , the case of higher levels
will coincide with that of level 1 as long as the levels are the same for both SU(3)C and
SU(2)L . From a theoretical point of view this case is the most attractive, since it allows for
both SU(3)C and SU(2)L to emerge from the same gauge group as in a GUT. The charge
assignment and possible discrete symmetries for this case k2 = k3 are shown explicitly in
Table 3.4
The procedure we have followed to obtain Table 3 is as follows. First we set b = 1.
Then we choose a set of integers ni in the range ni (05). Any ni larger than or equal to
6 (or any negative ni ) can be absorbed into the mod M piece. The highest ZM symmetry
is then found to be Z6 . For every choice of the integer set ni we demand that the R-parity
breaking couplings of Eq. (19) be forbidden. (This requires n1 n4 to be an odd integer
Table 3
MSSM charge assignment when k2 = k3 . denotes the charge of the gaugino and the Grassmanian variable .
When = 0, the ZM acts as an R-symmetry. Also shown are the anomaly coefficients (A2 , A3 )
Model
ZM
(A2 , A3 )
I
II
III
IV
Z2
Z2
Z6
Z6
2
2
6
6
1
1
5
5
1
1
1
1
2
2
2
2
1
1
5
5
1
1
3
3
1
1
1
1
1
1
5
5
1
0
3
0
(2, 2)
(1, 1)
(6, 6)
(3, 3)
4 Among the solutions, we remove those which either are conjugate of the listed solution or can be realized
as linear combination of the known solution and hypercharge. For example, in Model III and IV, we have chosen
q = 0 mod 6. The charge q need not be actually zero. Since there exists an unbroken U (1)Y hypercharge which
is anomaly-free, one can always take a linear combination of Model III (or IV) with U (1)Y to find equivalent
solutions with q = 0 mod M. This comment also applies to the models listed in all the other tables.
332
and that the charge of d c should be non-zero under ZM .) Then we solve for the charges
of the various fields, setting q = 0 as mentioned above. If the GreenSchwarz anomaly
cancellation condition is satisfied we accept the solution. Upto overall conjugation and
shifts proportional to hypercharge, the complete set of solutions is as given in Table 3. We
have also listed the anomaly coefficients (A2 , A3 ) in Table 3.
Several remarks are in order about the results shown in Table 3.
(i) Models I and II differ only in the value of . In the effective low energy Lagrangian,
what matters is 2, which is the same for both models. Although the two models look
identical from low energy point of view, their embedding into a high scale theory will
not be the same. This is the reason for listing them separately. We shall see that when
ZM is embedded into a higher symmetry ZN so as to forbid a large term, Models I
and II will look different. Similar remarks apply to Models III and IV;
(ii) The Z2 symmetries in Table 3 (I and II) are actually subgroups of the Z6 symmetry
(III and IV). Their embedding into ZN will, however, lead to different solutions. Note
also that the Z3 subgroup of Z6 does not show up as a solution, since that would allow
for lepton number violation;
(iii) The Z6 symmetric solutions of Table 3 are actually identical to the Z6 subgroup of
B L shown in Table 1 which can prevent R-parity violation in MSSM [20]. This
can be recognized by taking linear combinations of Models III, IV and hypercharge.
Suppose we normalize hypercharge so that all MSSM fields have integer values
(with Q field having Y
= 1). Take now the combination 3(I V ) + Y
denoted by Y
(mod 6). This redefined charge is identical to the Z6 subgroup of B L of Table 1.
The Z2 models are identified as the Z2 subgroups of IR3 . We conclude from our
systematic analysis that even with GS anomaly cancellation, the only allowed discrete
symmetries at the ZM level (which admits a superpotential term) are the subgroups
of B L. This will, however, be not the case when ZM is embedded into ZN . Note
also that the anomaly coefficients A2 and A3 in Table 3 are all equal to M/2, so GS
mechanism is not playing a role in anomaly cancellations. This remark will also be
different in the ZN embedding.
3.3. Embedding ZM into ZN and solving the -problem
We recall that the original ZN symmetry broke down to a subgroup ZM once the spurion
field Z acquired a VEV along its F -component. At the level of ZM , a superpotential
-term is allowed. Now we turn to the task of identifying the original ZN symmetry
needed for explaining the term. We look for the simplest higher symmetry into which
the ZM solutions of Table 3 can be embedded. Each of the model in Table 3 has a different
embedding into ZN .
Consider the embedding of Model I in Table 3 into ZN . The smallest ZN group that
contains a Z2 subgroup is Z4 . This embedding is shown in Table 4. There are two possible
charge assignments indicated as Models Ia and Ib. These models are obtained as follows.
First we choose the value of to be either 1 or 3 under Z4 (since it must reduce to 1 under
Z2 ). These two values correspond to the two models in Table 4. Then we demand that a
bare term in the superpotential is prevented by the Z4 symmetry. That determines the
333
Table 4
Embedding of the Z2 symmetry of Model I of Table 3 into Z4 symmetry
Z4
(A2 , A3 )
Ia
Ib
4
4
1
1
3
3
4
4
3
3
1
1
1
1
3
3
1
3
(1, 3)
(3, 1)
Table 5
Model 1a recast with a shift proportional to hypercharge
Z4
(A2 , A3 )
Ia
(3, 1)
to be either (1, 3) or (3, 1). It turns out that the charges in the latter case are
charges (h, h)
the conjugates of the former, and so we discard it. Then we set the charge n to be either 1
or 3, consistent with it being 1 under Z2 subgroup. This fixes the charges of all fields. For
each case the anomaly coefficients A2 and A3 are computed. If the anomalies are cancelled
by the GS mechanism, we accept the solution. Only two solutions are found to survive, as
displayed in Table 4.
Note that the discrete Z4 anomalies are cancelled by the GS mechanism. Individually
A2 and A3 are not multiples of N/2 = 2, but the two coefficients differ only by N/2 = 2.
We conclude that this simple solution would not have been possible without GS anomaly
cancellation.
The models of Table 4 can be recast in a very simple form by forming the linear com
(mod 4)}, or {Ib + Y
(mod 4)} with Y
being the integer values of SM

bination {Ia +Y
hypercharge. We display in Table 5 Model Ia recast in this form. The charge assignment
is very simple, all matter fields of MSSM carry charge 1 under Z4 , while the Higgs
superfields carry charge 0. The gauginos also have charge 1. The contribution to the Z4
anomaly from the matter fields are the same for A2 and A3 (the number of color triplets
is the same as the number of SU(2)L doublets in MSSM). While the gluino contributes an
amount equal to 3 to A3 , the sum of the SU(2)L gaugino (= 2) and the Higgsinos (= 1)
add to A2 = +1. We see that A2 and A3 differ by N/2 = 2, signaling anomaly cancellation
via GS mechanism.
The charge assignment shown in Table 5 is clearly compatible with grand unification.
The KacMoody level associated with hypercharge will be k1 = 5/3 with a GUT
embedding. Gauge coupling unification is then predicted, since sin2 W = 3/8 near the
string scale. This is true even if there were no covering GUT symmetry.
Now we turn to Model II of Table 3. Embedding this Z2 into a Z4 is not viable, since
a large term cannot be prevented in that case. The next simplest possibility is Z6 , which
also does not work as the Z6 anomalies do not cancel. The simplest embedding is found to
be into a Z10 with the charge assignment as shown in Table 6.
In the models of Table 6, one might consider a Z5 subgroup of Z10 . This subgroup is
sufficient to forbid the -term in the superpotential W , as well as to prevent dangerous
R-parity violating couplings in W . With invariance only under Z5 , the term uc d c d c
will
zero charge. A Lagrangian term arising from the Kahler potential L
4 have
2 ) will then be allowed. Once F acquires a non-zero VEV, this term
d (uc d c d c Z /MPl
Z
334
Table 6
Z10 embedding of Model II
Z10
(A2 , A3 )
IIa
IIb
10
10
1
7
7
9
4
8
3
1
7
9
3
1
7
9
2
4
(6, 6)
(2, 2)
Table 7
Z12 embedding of Model III
Z12
(A2 , A3 )
IIIa
IIIb
IIIc
IIId
12
12
12
12
5
11
5
11
7
1
7
1
8
8
8
8
11
5
11
5
9
3
9
3
1
7
1
7
11
5
11
5
3
3
9
9
(9, 9)
(9, 9)
(3, 3)
(3, 3)
Table 8
Z18 embedding of Model IV
Z18
(A2 , A3 )
IVa
IVb
18
18
11
5
13
1
14
8
17
11
15
15
1
7
17
11
6
6
(9, 9)
(18, 9)

will lead to a superpotential term L d 2 (MSUSY /MPl )uc d c d c . Such a term violates
R-parity, although very weakly. Signals of such a weak violation will be unobservable in
collider experiments. However, this scenario will not fit well with cosmological constraints.
The LSP will decay through this induced R-parity violating Yukawa coupling , which
has a strength of order MSUSY /MPl 1015 . We can estimate the lifetime of the LSP
to be [( )2 MLSP /(8)]1 104 s. Such a lifetime falls into the cosmologically
disfavored range and would be in violation of nucleosynthesis constraints. (This situation
is analogous to the gravitino problem of supergravity, but is slightly worse, since the LSP
mass is expected to be order 100 GeV, rather than a TeV for the gravitino, making the LSP
lifetime somewhat longer than that of the gravitino.) We consider the Z5 solution to be
unacceptable for this reason. Since Z5 symmetry does not contain a Z2 subgroup, exact
R-parity could not be defined after SUSY breaking, unlike in the case of Z10 model.
In Tables 7 and 8 we show the simplest embedding of Models III and IV into Z12 and
Z18 respectively. The procedure we have adopted is identical to that for Models I and II.
As in the case of the Z10 model of Table 6, we may consider taking a Z9 subgroup
of Z18 in Table 8. However, since Z9 does not contain Z2 or Z6 as subgroups, after
SUSY breaking, small R-parity violating Yukawa couplings of the type W LLec will
be generated from the Kahler potential with coupling constants of order 1015 . Such
couplings would violate constraints from big bang nucleosynthesis since the lifetime of
the LSP will be of order 104 s. We shall not consider the Z9 subgroup any further.
335
3.4. Discrete anomaly cancellation at higher KacMoody level

Thus far we have assumed the parameter b k3 /k2 = 1. This is the case when the
KacMoody levels for SU(3)C and SU(2)L are the same, the simplest possibility being
k3 = k2 = 1. It is also possible that k2 and k3 are not the same. It is not clear to us how
easy it is to construct string models with different values for k3 and k2 . Although it might
appear less attractive theoretically, it is nevertheless a logical possibility. In this section we
analyze discrete anomaly cancellation for values of k3 /k2 = 1.
From a technical point of view it appears to be difficult to construct models with levels
higher than 3 in string theory [21]. Motivated by this observation, we shall confine our
discussions to k2 and k3 being less than or equal to 3. This allows for the cases when
b k3 /k2 = 1, 2, 1/2, 1/3, 2/3 and 3/2. The case of b = 1 has already been analyzed in
the previous section, so we turn to the other cases.
From the solution Eq. (17) which applies to ZM invariance (that allows a bare term,
but forbids all R-parity violations), a few simplifications can be found. The case where
b = 1/2 and b = 1/3 are identical to the case of b = 1. This is because the b-dependent
terms only contribute to the various charges proportional to n5 in Eq. (17). But this n5
contribution can be absorbed into the n2 term in all equations. No new solutions will
then be generated under ZM . Similarly, it is easy to see that the cases b = 2/1 and
b = 2/3 are equivalent under ZM . And the case where b = 3/1 becomes identical to the
case of b = 3/2. Among these equivalent cases under ZM , we shall only consider one
possibility. Although it is possible that when the resulting models are embedded into a
higher symmetry ZN , new models at higher levels may emerge, we shall not pursue it
here.
We shall then focus on the case where b k3 /k2 = 2, and b = 3 for anomaly
cancellation at higher KacMoody level. Following the same procedure as in the previous
section, we obtain the corresponding discrete symmetry and charge assignment. The
solutions for the case of k3 /k2 = 2, which is the same for k3 /k2 = 2/3, are shown in
Table 9. The discrete ZM symmetry is Z6 in this case. Note that the discrete GS mechanism
cancels the gauge anomalies of Z6 . For example, A2 = 9/2, A3 = 6 is anomaly free since
with k2 = 1, k3 = 2, the cancellation condition is that 2A2 and A3 differ by an integer
multiple of N/2 = 3 (see Eq. (6)).
The two Z6 models of Table 9 have been embedded into the simplest possible ZN model
in Tables 10 and 11. The ZN symmetries are found to be Z12 and Z18 . The discrete gauge
anomalies are cancelled by GS mechanism, as before. Take Model Vd, for example, which
has A2 = 3/2, A3 = 9 under Z12 . 2A2 and A3 differ by 6, which is an integer multiple of
N/2 = 6.
Table 9
Discrete symmetries and the corresponding charge assignment when k3 /k2 = 2 or 2/3
Model
ZM
(A2 , A3 )
V
VI
Z6
Z6
6
6
2
2
4
4
5
5
5
5
3
3
4
4
2
2
3
0
(9/2, 6)
(3/2, 3)
336
Table 10
Embedding of Z6 of Model V into Z12
Z12
(A2 , A3 )
Va
Vb
Vc
Vd
12
12
12
12
8
2
2
8
4
10
10
4
5
5
11
11
11
5
11
5
3
9
3
9
10
4
4
10
2
8
8
2
3
3
3
3
(9/2, 9)
(9/2, 9)
(3/2, 9)
(3/2, 9)
Table 11
Z18 -embedding of the Z6 Model VI
Z18
(A2 , A3 )
VIa
VIb
18
18
2
14
10
16
11
5
5
17
15
15
4
10
14
8
6
6
(9/2, 18)
(27/2, 9)
Table 12
Discrete symmetries and the charge assignments for k3 /k2 = 3 or 3/2
Model
ZM
(A2 , A3 )
VII
VIII
Z18
Z18
18
18
1
1
17
17
10
10
7
7
9
9
17
17
1
1
9
18
(6, 18)
(15, 9)
Table 13
Z36 -embedding of the Z18 Model VII
Z36
(A2 , A3 )
VIIa
VIIb
36
36
1
19
35
17
28
10
25
7
9
9
17
35
19
1
9
9
(33, 27)
(6, 27)
Table 14
Embedding of Z18 of Model VIII into Z90
Z90
(A2 , A3 )
VIIIa
VIIIb
90
90
55
55
71
53
46
28
25
25
9
27
53
89
37
1
54
72
(89, 27)
(42, 36)
The next case is when b = k3 /k2 = 3, which gives at the ZM level the same models
as b = 3/2. In Table 12 we list the allowed ZM models, with M = 18. Table 13 has
the embedding of Model VII into Z18 that prevents a large term, Table 14 has the
embedding of Model VIII, the simplest possibility for which being Z90 . In all cases the
discrete anomalies are cancelled by the GS mechanism.
It should be mentioned that at the level of ZM , it is easy to realize an R-parity that
allows for lepton number violation, but conserves baryon number. Rapid proton decay
will be prevented in this case. Lepton number violating processes and neutrino masses do
provide some constraints, but these are much less stringent.
337
Table 15
Examples of a Z4 symmetry that allows for baryon number violation without dangerous lepton number violations.
The Z4 anomalies are cancelled by GS mechanism at levels k2 = 1, k3 = 2
Z4
(A2 , A3 )
A
B
4
0
2
2
2
2
1
3
1
3
1
3
2
0
2
0
1
3
(1/2, 3)
(5/2, 5)
Consider the Z6 models of Table 3 (Models III and IV). Suppose we impose invariance
only under the Z3 subgroup of Z6 with the same charge assignment as in Table 3. The
lepton number violating couplings LHu , LLec , and QLd c all have charge 3 under the
Z6 symmetry, so with only Z3 invariance imposed, these couplings will be allowed in
the superpotential. Since the original Z6 symmetry is free from discrete gauge anomalies,
the subgroup Z3 is also free from such anomalies. One cannot, however, embed this Z3
symmetry to any higher ZN in order to explain the -parameter. Consider the LHu term
in the superpotential. ZN invariance of this term would imply l + h = 2 mod N . The last
two relations of Eq. (13) would imply 2 = 0 mod N , implying that a bare term in the
superpotential will be allowed. An alternative explanation for the -term will have to be
found in the case of lepton number violating R-parity.
It is also possible, although somewhat non-trivial, to have baryon number violating Rparity without dangerous lepton number violation. (Neutrino masses violate lepton number
by two units, but that does not result in rapid proton decay.) At the level of ZN we
can show that anomalies associated with such an R-parity will have to be cancelled at
higher KacMoody levels. If the coupling uc d c d c is allowed in the superpotential, we find
that the ZN discrete symmetry has anomalies given by A3 = 3, A2 = 5/2 (3/4)pN
where p is an integer. Imposing the anomaly cancellation condition, Eq. (6), we find =
(m1 k2 + m2 (k3 /2)p)N/(6k3 5k3 ) with m1 , m2 , p being integers. When k2 = k3 = 1,
this relation shows that 2 = 0 mod N , meaning that a bare -term will be allowed in the
superpotential. If we choose k2 and k3 differently, this problem will not arise. Consider
for example, k3 = 2, k2 = 1. A consistent ZN charge assignment corresponding to a Z4
symmetry is shown in Table 15 for this case. This model allows for the coupling uc d c d c ,
while preventing other R-parity violating couplings. The Z4 anomalies are cancelled by
GS mechanism, which in this case reads as 2A2 A3 = m 2n, with m, n being integers.
4. Discrete flavor symmetries and the fermion mass hierarchy

As indicated earlier, there must be an unbroken ZM symmetry which is flavorindependent that survives to low energy scales, to be identified as an R-parity. We can,
however, introduce flavor dependence in the original symmetry, provided that a subgroup
of the flavor group remains unbroken and can be identified as one of the ZM symmetries
of Table 3. In this section we embark on this question. Our aim will be to seek an
understanding of the observed hierarchy in the fermion masses and mixings without
introducing such hierarchy by hand.
338
Anomalous U (1)A symmetry is widely used for the explanation of fermion mass and
mixing hierarchy [22]. The general superpotential in this has the following expression:

S (h1 )ij
S (h2 )ij
S (h3 )ij
W = Qi ucj Hu
+ Qi djc Hd
+ Li ejc Hd
MPl
MPl
MPl

(h4 )ij
(h5 )ij
S
S
+ Li jc Hu
+ ccS
,
(20)
MPl
MPl
where S is an MSSM singlet field with a non-trivial anomalous U (1)A charge. S acquires
a VEV near the string scale and disappears from the low energy spectrum. Here (h )ij is a
set of integers for = 1, 2, 3, 4, 5 and i, j = 1, 2, 3 are the generation indices. We assume
that all the Yukawa couplings are of order one. After S field develops a VEV, near but
somewhat below the string scale, a small parameter 6 = S/MPl 1/5 is generated. This
factor appears in various powers with the Yukawa couplings, explaining the observed mass
and mixing hierarchy [22]. It is possible to suppress all the MSSM Yukawa couplings to
the desired level by choosing appropriate set of U (1) charges [14].
An acceptable flavor texture which gives the correct pattern of fermion masses and
mixings is:

6
4
6 65 63
6 63 63
5
4
2
3
2
2
Dij = 6 6 6 6 p Hd ,
Uij = 6 6 6 Hu ,
3
2
6 6
1
6 1 1
4
2

6 63 6
6 6 6
p
D
3
2
Lij = 6 6 1 6 Hd ,
(21)
ij = 6 1 1 6 a1 Hu ,
3
2
6 1 1
6 6 1
where Uij , Dij , Lij and ijD correspond to up-quark, down-quark, charged lepton and Dirac
neutrino Yukawa matrices resulting from the appropriate powers of the S field in Eq. (20).
The integer p can be either 0, 1 or 2, corresponding to large, medium and small tan ,
respectively.
Once the charged lepton sector and Dirac neutrino sector are constructed, we can
uniquely define the form of the heavy Majorana neutrino mass matrix. In the present
example it is
2
6 6 6
ijM = 6 1 1 6 a2 .
(22)
6 11
As mentioned before, the MSSM superpotential does not possess any unbroken U (1)
symmetry, apart from the gauge symmetry. Therefore, we seek solutions of a ZN discrete
symmetry that would generate the Yukawa matrices of Eq. (21).
ZN invariance of the Yukawa couplings in Eq. (20) imposes constraints in the up-quark
sector given by
qi + uj + h + ps = 2 mod N,
(23)
where p is the power of 6 appearing in the appropriate element of the Uij matrix, which is
equal to the power in the field S. s denotes the U (1) charge of S field. Similar conditions
339
apply for the charges of the other SM fermions as well. By construction, the flavor
structure of the matrices obey the determinant rule, viz., that in any 2 2 sub-block,
the determinant is a homogeneous function of 6. This means that out of the 18 conditions
for the up-quark and the down-quark sector, only 8 will be independent.
We wish to have an unbroken ZM symmetry that is a subgroup of ZN which is flavorindependent. Since the flavon field S has a ZN charge of s, once it acquires a VEV of order
the string scale, the ZN will be broken down to Zs . We shall attempt to embed the ZM of
Table 3 into Zs .
To be specific, let us work out an example with the Z2 model of Table 3 and embed
this Z2 into a higher ZN symmetry that allows for the desired flavor structure. The ZN
symmetry must be Z14+2n for this embedding to be consistent, with n being any integer.
The smallest such symmetry is then Z14 . The flavon field S must carry zero charge under
Z2 and should transform non-trivially under the ZN . The simplest possibility is s = 2.
Now, the Yukawa textures of Eq. (21) makes use of S 6 terms in the superpotential, which
should be different from S 0 . This requirement makes the smallest ZN symmetry to be Z14 .
If this symmetry were Z12 , for example, S 6 will be neutral under Z12 , making the (11)
and the (33) entries of the up-quark Yukawa matrix of the same order. We can generalize
this statement to any low scale discrete symmetry. The corresponding flavor-dependent
symmetry must be ZM(k+1)+n , where ZM is the low scale surviving discrete symmetry, k
corresponds the highest power of the S field in the general superpotential, Eq. (20), and n
is a positive integer. This choice will guarantee the existence of ZM discrete R-parity at
low energy scales.
Three examples of Z14 symmetric models are presented in Table 16. We have chosen the
charge of S to be 2 and fixed the charge of to be 7 in these examples. Discrete anomaly
cancellation is enforced via GS mechanism at KacMoody level 1. We have also imposed
the conditions that the Z14 symmetry forbid all R-parity violating couplings.
The ZN symmetry group would depend on the highest power of 6 appearing in the
fermion Yukawa matrices. If we want to have symmetry smaller than Z14 , we should reduce
the power of S field in Eq. (20). One way is to re-parameterize the value of 6. For example,
if 6 is taken to be of order 1/10, rather than 1/5 as was assumed in Eq. (21), it might suffice
to use cubic powers of S at most. A Z8 discrete symmetry would then suffice to forbid the
R-parity breaking terms. We consider the expansion given in Eq. (21) to be more realistic.
In Table 17 we present three models based on Z28 symmetry that forbid all R-parity
violating couplings, explain the fermion mass and mixing hierarchy via the texture of
Eq. (21) and also solve the -problem via the GuidiceMasiero mechanism. As before,
the discrete gauge anomalies are cancelled by the GS mechanism. We find it remarkable
Table 16
Examples of flavordependent Z14 symmetry which forbids all R-parity breaking terms. i = 1, 2, 3 is the flavor
index and charges in the brackets are in order of 13. We are considering p = 2 and q = 0 in Eq. (21) which
corresponds to medium values of tan 10. We have taken a2 = 0 in Eq. (22) for simplicity
A
B
C
Qi
uci
dic
Li
eic
ic
Hu
Hd
A2
A3
0, 2, 6
4, 6, 10
6, 8, 12
1, 3, 7
13, 1, 5
5, 7, 11
3, 5, 5
11, 13, 13
1, 3, 3
4, 6, 6
6, 8, 8
0, 2, 2
13, 1, 5
9, 11, 1
7, 9, 13
5, 7, 7
5, 7, 7
5, 7, 7
1
13
9
13
1
5
7
7
7
2
2
2
6
13
13
13
13
6
340
Table 17
Examples of flavor-depended discrete Z28 symmetry which prevent R-parity breaking couplings, explain the
origin of the -term via the GuidiceMasiero mechanism, and explain the hierarchy in quark and lepton masses
and mixings via the Yukawa texture shown in Eq. (21). i = 1, 2, 3 is the flavor index and charges in the brackets
are in order of 13. We are considering p = 2 and q = 0 in Eq. (21), which corresponds to medium values of
tan 10. We have taken a2 = 0 in Eq. (22) for simplicity
Qi
A 12, 16, 24
B 22, 26, 6
C 26, 2, 10
uci
dic
Li
eic
ic
Hu
Hd
A2
A3
7, 11, 19
23, 27, 7
7, 11, 19
9, 13, 13
1, 5, 5
9, 13, 13
4, 8, 8
2, 6, 6
18, 22, 22
17, 21, 1
21, 25, 5
17, 21, 1
3, 7, 7
3, 7, 7
3, 7, 7
27
1
13
1
27
15
7
7
7
4
4
4
11
11
11
11
11
25
that a single discrete symmetry can do all these jobs. It may be mentioned that Z28 is
not a large symmetry unlikely to be realized in string theory. For example, if the particle
spectrum contains fields carrying charges of (1, 1/4, 1/7) under the anomalous U (1), and
if a scalar field with charge 1 acquires a VEV, the unbroken ZN symmetry will be Z28 .
5. Conclusion
In this paper we have investigated the possibility of realizing R-parity of MSSM as a

discrete gauge symmetry. Simultaneously we have demanded that this discrete symmetry
should provide a natural explanation for the -term, the Higgsino mass parameter in the
MSSM superpotential, via the GuidiceMasiero mechanism. We have adopted a discrete
version of the GreenSchwarz anomaly cancellation mechanism in our search for discrete
gauge symmetries, which is less constraining than the conventional methods.
We have found simple examples of ZN symmetries that act as R-parity and simultaneously solve the -problem, without extending the particle content of the MSSM. The
simplest example is a Z4 symmetry with a simple charge assignment that is compatible
with grand unification. The GreenSchwarz mechanism plays a crucial role in cancelling
the Z4 anomalies. Other examples of ZN symmetries are provided with N = 10, 12, 18, 36
etc. In some cases the discrete anomalies are cancelled by the GS mechanism at higher
KacMoody levels. We have found that it is easy to realize lepton number violating Rparity as a discrete symmetry, but implementing the GuidiceMasiero mechanism for the
term is difficult in this case. Baryon number violating R-parity can be realized, along
with a natural explanation of the term, but the discrete gauge anomalies are cancelled in
this case at higher KacMoody level.
It has been shown that a simple ZN symmetry can also explain the observed hierarchical
structure of quark and lepton masses and mixings, while preserving the desired R-parity
and the solution to the -problem. Examples of a Z28 symmetry doing all these have been
presented.
341
Acknowledgement
We thank J. Lykken and Ts. Enkhbat for useful discussion. This work is supported in
part by DOE Grant # DE-FG03-98ER-41076, a grant from the Research Corporation and
by DOE Grant # DE-FG02-01ER-45684.
References
[1] S.W. Hawking, Phys. Lett. B 195 (1987) 337;
G.V. Lavrelashvili, V.A. Rubakov, P.G. Tinyakov, JETP Lett. 46 (1987) 167;
S. Giddings, A. Strominger, Nucl. Phys. B 306 (1988) 349;
S. Giddings, A. Strominger, Nucl. Phys. B 321 (1989) 481;
L.F. Abbot, M. Wise, Nucl. Phys. B 325 (1989) 687;
S. Coleman, K. Lee, Nucl. Phys. B 329 (1989) 389;
R. Kallosh, A. Linde, D. Linde, L. Susskind, Phys. Rev. D 52 (1995) 912.
[2] L. Krauss, F. Wilczek, Phys. Rev. Lett. 182 (1989) 1221.
[3] J. Preskill, S.P. Trivedi, F. Wilczek, M.B. Wise, Nucl. Phys. B 363 (1991) 207.
[4] T. Banks, M. Dine, Phys. Rev. D 45 (1992) 1424.
[5] L.E. Ibanez, G.G. Ross, Phys. Lett. B 260 (1991) 291;
L.E. Ibanez, G.G. Ross, Nucl. Phys. B 368 (1992) 3;
L.E. Ibanez, Nucl. Phys. B 398 (1993) 301.
[6] L.E. Ibanez, Phys. Lett. B 303 (1993) 55;
L.E. Ibanez, G.G. Ross, Phys. Lett. B 272 (1991) 260;
L.E. Ibanez, G.G. Ross, Phys. Lett. B 332 (1994) 100.
[7] K. Kurosawa, N. Maru, T. Yanagida, Phys. Lett. B 512 (2001) 203.
[8] E. Ma, Mod. Phys. Lett. A 17 (2002) 535;
E. Ma, Phys. Rev. Lett. 89 (2002) 041801.
[9] G. Guidice, A. Masiero, Phys. Lett. B 206 (1988) 1480.
[10] M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117;
M.B. Green, J.H. Schwarz, Nucl. Phys. B 255 (1985) 93;
M. Green, J. Schwarz, P. West, Nucl. Phys. B 254 (1985) 327.
[11] For a review see P. Ramond, hep-ph/9808488.
[12] M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 589;
J. Atick, L. Dixon, A. Sen, Nucl. Phys. B 292 (1987) 109.
[13] P. Ginsparg, Phys. Lett. B 197 (1987) 139.
[14] P. Binetruy, P. Ramond, Phys. Lett. B 350 (1995) 49;
P. Binetruy, S. Lavignac, P. Ramond, Nucl. Phys. B 477 (1996) 353;
Y. Nir, Phys. Lett. B 354 (1995) 107;
Z. Berezhiani, Z. Tavartkiladze, Phys. Lett. B 396 (1997) 150;
Z. Berezhiani, Z. Tavartkiladze, Phys. Lett. B 409 (1997) 220;
Q. Shafi, Z. Tavartkiladze, Phys. Lett. B 482 (2000) 145;
M. Gomez, et al., Phys. Rev. D 59 (1999) 116009;
J. Feng, Y. Nir, Phys. Rev. D 61 (2000) 113005;
A.S. Joshipura, R. Vaidya, S.K. Vempati, Phys. Rev. D 62 (2000) 093020;
N. Maekawa, Prog. Theor. Phys. 106 (2001) 401;
I. Gogoladze, A. Perez-Lorenzana, Phys. Rev. D 65 (2002) 095011;
T. Ohlsson, G. Seidl, Nucl. Phys. B 643 (2002) 247.
[15] G. Dvali, S. Pokorski, Phys. Rev. Lett. 78 (1997) 807;
J.L. Chkareuli, C.D. Froggatt, I.G. Gogoladze, A.B. Kobakhidze, Nucl. Phys. B 594 (2001) 23;
Q. Shafi, Z. Tavartkiladze, Nucl. Phys. B 573 (2000) 40;
342
[16]
[17]
[18]
[19]
[20]
[21]
[22]
N. Maekawa, T. Yamashita, Prog. Theor. Phys. 107 (2002) 1201;

B. Bajc, I. Gogoladze, R. Guevara, G. Senjanovic, Phys. Lett. B 525 (2002) 189.
K.S. Babu, S.M. Barr, Phys. Lett. B 300 (1993) 367;
J.L. Lopez, D.V. Nanopoulos, Phys. Lett. B 245 (1990) 111;
K.S. Babu, B. Dutta, R.N. Mohapatra, Phys. Rev. D 65 (2002) 016005.
E.J. Chun, J.E. Kim, H.P. Nilles, Nucl. Phys. B 370 (1992) 105;
K. Choi, E.J. Chun, H.D. Kim, Phys. Rev. D 55 (1997) 7010.
G. Dvali, A. Pomarol, Phys. Rev. Lett. 77 (1997) 3728;
P. Binetruy, E. Dudas, Phys. Lett. B 389 (1996) 503;
R.N. Mohapatra, A. Riotto, Phys. Rev. D 55 (1997) 4262.
M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwnehuizen, 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 the Unified Theories,
Baryon Number in Universe, Tsukuba, Japan 1979 (KEK Report No. 79-18), Tsukuba, 1979, p. 95;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
R.N. Mohapatra, Phys. Rev. D 34 (1986) 3457;
A. Font, L. Ibanez, F. Quevedo, Phys. Lett. B 228 (1989) 79;
S. Martin, Phys. Rev. D 46 (1992) 2769.
K. Dienes, J. March-Russell, Nucl. Phys. B 479 (1996) 113;
K. Dienes, A. Faraggi, J. March-Russell, Nucl. Phys. B 467 (1996) 44;
S. Choudhuri, S.W. Chung, G. Hockney, J. Lykken, Nucl. Phys. B 456 (1995) 89;
Z. Kakushadze, G. Shiu, S.H. Tye, Y. Vitorov-Karevsky, Int. J. Mod. Phys. A 13 (1998) 2551;
K.R. Dienes, Phys. Rep. 287 (1997) 447.
C. Froggatt, H. Nielsen, Nucl. Phys. B 147 (1979) 277.

QCD corrections to resonant slepton production

in hadron colliders
Debajyoti Choudhury, Swapan Majhi, V. Ravindran
Harish-Chandra Research Institute, Chhatnag Road, Jhusi, Allahabad 211 019, India
Received 22 August 2002; received in revised form 18 February 2003; accepted 14 March 2003
Abstract
We consider resonant production of sneutrino and slepton at hadronic colliders such as the
Tevatron and the LHC within the context of a R-parity violating supersymmetric model. We present
next to leading order QCD corrections to total cross-sections which originate from both quark- as
well as gluon-initiated processes. For couplings involving only the first generation quarks, the K
factor at the Tevatron can be as large as 1.5 for a 100 GeV sfermion and falls to nearly 1.1 as the
sfermion mass reaches a TeV. At the LHC, the variation is between 1.2 and 1.45 for masses less
than 2 TeV. While the dependence on the parton density parametrization is found to be mild, this
ceases to be true if the strange quark plays a dominant role in the production process. We also study
the renormalization and factorization-scale dependences and find it to be less pronounced for the
NLO cross-sections as compared to the LO. The results obtained in this article are also applicable to
resonance production of any color-neutral scalar.
PACS: 12.60.Jv
1. Introduction
The scalar sector of the Standard Model (SM), while being integral to the validity of
this otherwise eminently successful model, has also been somewhat of an embarrassment.
Quite apart from the fact that the Higgs particle has, till date, defied all attempts at detecting
it, there is the theoretical problem that the mass of this particle is not protected by any
symmetry. Consequently, quantum corrections would tend to drive it to the next higher
E-mail addresses: debchou@mri.ernet.in (D. Choudhury), swapan@mri.ernet.in (S. Majhi),
ravindra@mri.ernet.in (V. Ravindran).
doi:10.1016/S0550-3213(03)00233-5
344
D. Choudhury et al. / Nuclear Physics B 660 (2003) 343361
scale of interaction that the Higgs participates in, an eventuality that, apart from running
counter to the indications coming from electroweak precision measurements [1], would
also lead to a loss of perturbative unitarity. To overcome this as well as certain other lacunae
of the SM, many models going beyond the SM have been proposed. Two of the most
attractive classes of such models comprise those incorporating supersymmetry [2] and/or
grand unification [3] (especially scenarios with a low intermediate scale [4]). Such models,
however, predict, additional particle states, including scalars. What is most interesting is
that the coupling of the first generation SM fermions to these scalars need no longer be
suppressed, thus offering hope for novel signatures.
The last-mentioned feature has, naturally, attracted much attention, especially in the
context of the current and future colliders. Apart from pair-production (determined, in
the most part, by the gauge interactions), an enhanced coupling to fermions opens up
the possibility of resonance production at colliders whether hadronic (pp or pp) [57],
e+ e [8,9] or e p [10]. These studies conclude that not only is discovery guaranteed
for a significantly large part of the parameter space, even a measurement of the coupling
strengths to a reasonable degree of accuracy might be possible.
Most of these analyses, however, have been performed only at the Born level. In view
of the interesting consequences, it is desirable that quantum corrections to such resonance
production processes should be investigated. While this has been done in the context of
ep colliders [11], a similar exercise has not been attempted for hadronic colliders. In this
paper, we seek to rectify this lacuna. Before we embark on such a venture, however, it is
important to note that, in a generic model going beyond the SM, scalars could appear in
many a hue. The quantum numbers are of crucial importance as these would determine
not only the production cross-sections but also the dominant decay channels and hence
the possible modes of discovery. For the sake of concreteness, we shall confine ourselves
to a discussion of sneutrino and slepton production in the context of a R-parity violating
supersymmetric model. While, at first, this may seem to be a very restrictive assumption, in
reality, these constitute very typical examples of colour-neutral scalars. One might as well
have considered a generic multi-Higgs model wherein some of the low-lying states have
an enhanced coupling with the lighter fermions. The prime rationale behind our choice
is that while the scalar masses can be protected naturally in supersymmetric models, the
same is not so straightforward in non-supersymmetric models (grand unified or otherwise).
Moreover, the R-parity violating Minimal Supersymmetric Standard Model (MSSM) being
a richer (low-energy) theory, offers a larger set of possibilities, both in the context of the
neutrino anomalies seen at Kamiokande [12] or Karmen [13] or the unexplained high-Q2
events at Hera [10].
The plan of the paper is as follows. We start this article (Section 2) with a brief
review of the status of R-parity conservation within the MSSM. Section 3 describes the
particular resonance production processes (at the lowest order) that we are interested in.
The formalism and the calculations for the NLO corrections are set out in the following
section. In Section 5, we present the numerical results and a discussion thereof. And finally
we summarize.
345
2. R-parity violation: a mini-review

As is well known, within the SM, both baryon (B) and lepton (L) number conservation
are but accidental consequences of the choice of the particle content.1 In extensions of the
SM, such an accidental occurrence is obviously not guaranteed. For example, in a generic
grand unified theory (GUT), both the gauge and the scalar sector interactions violate each
of B and L. This is potentially catastrophic as a simultaneous breaking of both B and L
could lead to rapid proton decay. Within GUTs, however, gauge boson-mediated proton
decay is naturally suppressed on account of the large gaugeboson masses. On the other
hand, the scalar sector has to be carefully chosen so as to suppress any effective operator
leading to proton decay.
Within the context of the MSSM though, we do not have the option of demanding
the sfermion or gaugino fields to be superheavy. However, a similar suppression can
be achieved by introducing a discrete symmetry, R (1)3(BL)+2S (with S denoting
the spin of the field) [14] that serves to rule out both B and L violating terms in the
superpotential. In addition, this symmetry renders the lightest supersymmetric partner
absolutely stable. The introduction of this symmetry is clearly an ad hoc measure and
is not even strictly essential to rule out proton decay. Hence, it is of interest to consider
possible violations of this symmetry especially since it has rather important experimental
consequences, not the least of which concerns the detection of the supersymmetric partners.
/ p ) terms in the superpotential can be parametrized as
The possible R-parity violating (R
WR/ p = i Li H2 + ij k Li Lj Ekc + ij k Li Qj Dkc + ij k Uic Djc Dkc ,
(1)
Eic ,
Uic ,
Dic
where Li and Qi are the SU(2)-doublet lepton and quark superfields,

the
singlet superfields and Hi the Higgs superfields. Clearly ij k is antisymmetric under the
interchange of the first two indices, while ij k is antisymmetric under the interchange
of the last two. Whereas the first three terms in Eq. (1) violate L, the last term falls
foul of B conservation. To circumvent the constraints imposed by the non-observance
of proton decay, we, thus, need to have at least one of the two sets of couplings to be
vanishingly small. For the purpose of this paper, we assume that B is a good symmetry
of the theory, or in other words all of ij k are zero. This has the added advantage
that all dimension six operators leading to proton decay are suppressed [15] along with
the dimension five ones. Such a scenario might be motivated within certain theoretical
frameworks [15,16] and also renders simpler the problem of preservation of GUT-scale
baryon asymmetry [17]. Although the presence of the other R
/ p terms could, in principle,
affect the baryon asymmetry of the universe, such bounds are highly model-dependent
and can be evaded [18]. For example, in cases where at least one L-violating coupling
involving a particular lepton family is small enough ( 107 ) so as to (almost) conserve
the corresponding lepton flavour over cosmological time scales, such bounds are no longer
effective.
Each of the terms in Eq. (1) has its unique set of consequences, whether in low-energy
phenomenology or in resonance production. For example, while ij k lead to resonant
1 Indeed, non-perturbative effects within the SM itself do break B + L symmetry.
346
sneutrino production in e+ e collider [8,9], a non-zero ij k leads to resonant squark

production in hadronhadron collisions [5,7]. The ij k terms, on the other hand, can lead
to both resonant squark production at an e p facility [10] as well as to resonant slepton
and sneutrino production at a hadronic collider [5,6]. It is this last aspect that we shall
concentrate on.
The relevant part of the Lagrangian can be written in terms of the component fields as
L = ij k dkR iL dj L + dkR dj L iL + (iL )c dj L dkR

+ h.c.
dkR iL u j L dkR uj L iL (iL )c uj L dkR
(2)
Thus, while the squarks behave as leptoquarks in a non-supersymmetric theory, the
sleptons/sneutrinos behave as if they are charged/neutral Higgses in a multi-Higgsdoublet scenario. Clearly, non-zero values for these couplings could lead to rather
striking phenomenological consequences. For example, pair production of squarks that
subsequently decay through an L violating interaction, leads to a final state comprising
a dilepton pair along with jets [19]. More interestingly, the gluino production crosssection is larger and, in addition, can lead to like-sign dileptons, thereby making the signal
stand out even more [20]. Non-observation of such signals thus rules out a relatively
light squark or gluino along with a sizable . However, these analyses can say very
little about sleptons/sneutrinos as the corresponding production cross-sections are much
smaller than those for a squark/gluino. At an e+ e collider though, both pair production
of sleptons/sneutrinos and the corresponding backgrounds are weak processes and hence
such colliders are expected to be better suited for this particular quest. Unfortunately, an
e+ e collider energetic enough to pair-produce sleptons is still very much in the future.
We now turn to the constraints from low-energy phenomenology. Non-zero s can lead,
for example, to additional four-fermion operators that may contribute to meson decays,
neutral meson mixings, some of which may be forbidden otherwise. Since the absence of
tree-level flavour changing neutral current processes lead to rather severe constraints on
the simultaneous presence of more than one [21], we shall henceforth restrict ourselves
to only one non-zero . In Table 1, we list the currently known bounds on several of
these couplings.2 The strongest bound is on 111 and is derived from non-observation
of neutrinoless double beta decay (a) [23]. The others are much weaker and are derived
from (b) upper bound on the mass of the e [16,2426]; data on (c) charged-current
universality [9]; (d) atomic parity violation [27]; (e) and D Kl [25]; and (f)
0 mixing [21]. Since these bounds are derived from effective four-fermion operators,
D 0 D
they typically scale like the mass of the exchanged sfermion.3 Two points may be noted
here. First, many of these bounds are actually applicable only to particular combinations
of couplings and masses and reduce to those in the table only under the assumption of only
one coupling being non-zero. And secondly, in meson decays, most often it is the squark
that is exchanged; hence sleptons/sneutrinos could very well be much lighter without
contradicting the bounds.
2 A more complete list can be found in Ref. [22].
3 Of those listed in Table 1, the only exceptions to this rule are the bounds for

111 and 122 [16,2325].
347
Table 1
The upper bounds on the -type R
/p couplings of interest for a common sfermion mass m = 100 GeV. The
superscripts refer to the specific experiments leading to the constraints and as described in the text
{ij k}
Existing bounds
111
112
121
122
0.001(a)
0.02(c)
0.035(d)
0.02(b)
{ij k}
Existing bounds
{ij k}
Existing bounds
211
212
221
222
0.09(c)
311
312
321
322
0.10(e)
0.10(e)
0.20(f)
0.20(f)
0.09(c)
0.18(e)
0.18(e)
3. Leading order cross-section

The R
/ p interaction Lagrangian, as presented in Eq. (2), allows for the following
resonance production processes at a hadronic collider:
ij k :
:
dj + dk i , dk + dj i ,
dk + u j i , uj + dk i .
(3)
The conjugate processes obviously have identical cross-sections at the Tevatron, though
not at the LHC.
Before we start, we will make a few simplifying assumptions. Since QCD is flavourblind, the form of the strong interaction corrections would be independent of the particular
initial state quark in Eq. (3). Hence, for the sake of simplicity, we choose to develop
the formalism for the case of identical quarks. Furthermore, rather than consider a chiral
coupling to the scalars, we assume that the interaction is purely a scalar one. The chirality
structure can be accounted for at a later stage simply by introducing an extra factor of 1/2.
Note that neither of these assumptions imply a loss of generality.
The leading order process of interest is then
q(p
) + q(p) S(q),
(4)
where p and p are the momenta of incoming quark and antiquark, respectively, and q that
of the outgoing scalar. The amplitude for this process is given by
)u(p),
M (0) = iv(p
(5)
where is the scalar coupling constant. Since we consider only light quarks in the initial
state, we have
(p )2 = p2 = 0,
(6)
whereas the scalar has a (large) mass mq . The cross-section for the process in Eq. (4) is

1
1 (0)2
1
d n1 q
(2)n n (p + p q),
0 =
(7)
M
N 4(p p )
(2)n1 2q0 4
where the factor 1/4 arises from the spin averaging for the incoming quarks and 4(p p )
is the flux factor. In Eq. (7), N denotes the number of colours. Taking the spacetime
dimension n = 4, the above reduces to
0 = 2
1
(1 ),
N 2s
(8)
348
where
=
m2q
s
s = (p + p)2 , m2q = q q.
(9)
4. NLO corrections
The QCD correction to the process of interest has contributions from two different, but
related, sources. First, the quark-pair-initiated process itself receives radiative correction.
To this must be added the contribution arising from radiating off a soft gluon. And secondly,
since our true initial state is not quarks, but (anti)protons, we must include possible
contributions from initial-state gluons as well. We consider each in turn.
4.1. Correction to the q q initiated process
To calculate the QCD radiative correction to this process, we start by computing
the O(gs2 ) corrections to the vertex function and the self energy, where gs is the
QCD strong coupling constant. A prime ingredient for this is the calculation of the
corresponding renormalization constants Z and Z2 . Even on regulating the ultraviolet
(UV) divergences, we would, expectedly, be left with infrared (IR) divergences, part
of which will be cancelled once we take into account the soft gluon bremsstrahlung
contribution. Throughout our calculation we shall use dimensional regularization to
regulate any divergence and the MS prescription for renormalizing the results.
Let us first consider the vertex function M V upto order gs2 . This can be expanded as
M V = M (0) + M V (1),
(10)
where M (0) = iv(p

)u(p), and

1
)t a t a [ (/
p + k/)(/
p + k/) ]u(p)
d n k v(p
M V (1) = gs2 2 n/22
. (11)
n

2
2
(2) [(p k) + i][(p + k) + i][k 2 + i]
( )
The introduction of the (arbitrary) mass scale (called renormalization scale) is necessary
to render the strong coupling constant gs dimensionless in n spacetime dimensions.
The matrices t a are the Gell-Man matrices and satisfy (t a t a )ij = CF ij , where CF =
(N 2 1)/(2N) for SU(N) with N being the number of colours.
On using the equations of motion, the above can be simplified to
)(i)(1) u(p),
M V (1) = v(p
(12)
where
(1) = igs2 CF
with

I{1,,} =

1
2q 2 I1 + 4(p p ) I + ng I ,
(2 )n/22
{1, k , k k }
d nk
.
n

2
(2) [(p k) + i][(p + k)2 + i][k 2 + i ]
(13)
(14)
349
Naive power counting shows that I is logarithmically divergent in 4 dimensions, while

the other two are convergent. The integrals can be evaluated explicitly (for example, using
Feynman parametrization) and the results expressed in terms of gamma functions. The
resultant vertex function is then (3 n 4)

2
q 2 3/2 2 (1 + 3/2) (1 3/2)
s
2
CF
(1) =
(15)
.
(2
+
3)
4
(2 + 3)
42
32
The renormalization constant Z is defined through the relation

Z1 = 1 + (1) UV ,
(16)
and, of course, depends on the way the ultraviolet divergent part (1) |UV is isolated. Within
the MS scheme, it can easily be ascertained to be
3/2
1
8
s
,
CF
Z = 1 +
(17)
4 (1 + 3/2)
4
3
with s gs2 /4 .
The self-energy correction to the Born amplitude (say, to the quark only) can be
expressed as
1
p )u(p),
M S = iv(p
) (/
p
/
where
(/
p ) = gs2
CF
2
( )n/22
(/
p + k/)
d nk
.
n
2
(2) [k + i][(p + k)2 + i]
(18)
(19)
Notice that (/
p ) does not contribute to the amplitude given in Eq. (4) due to the massless
nature of the light quarks. On the other hand, the above equation can be used to determine
the counter term. In the MS scheme, this is found to be
3/2
s
1
2
CT
CF
=1
(20)
.
4 (1 + 3/2)
4
3
Our next task is to compute the virtual contributions to the process given in Eq. (4).
In order to do this, we have to redefine the fields and the coupling constants in terms of
the renormalized ones (and, of course, the renormalization constants Z2 and Z ). This
is equivalent to adding UV counter terms corresponding to the vertex function and selfenergy contribution. Note that self-energy contribution to the amplitude is identically zero
due to the on-shell condition. Hence only vertex function and the counter terms contribute
to the amplitude, hence
M (0)+vir+CT = M (0) + M V (1) + M CT .
It turns out that the effect of the counter term (CT) is

M (0) + M CT = (iR ) Z CT v(p

)u(p),
(21)
(22)
350
where R is the renormalized coupling. The virtual and counter term contributions to the
Born diagram can be expressed as

1
1 (0)+vir+CT 2
1
d n1 q
(0)+vir+CT
M
=
(2)n n (p + p q).

n1
N 4(p p )
(2) 2q0 4
(23)
Substituting Eqs. (17), (20) in Eq. (23), we have

1 2
s
(0)+vir+CT
(1)
1+
CF F
(24)
=
(1 ),
N 2s R
4
where
F (1) =
4
Z CT + 2(1) 1 .
s CF
(25)
Expanding around 3 = 0, and neglecting those terms which vanish in the limit 3 0, we
get
3/2
2
2

1
q
8 6 4
2
(1)
2 q
2
2 + ln 2 ln
2 + . (26)
F =
(1 + 3/2) 4
3 3
3
2
Next, we compute the contribution from the gluon bremsstrahlung to O(s ):
q(p
) + q(p) S(q) + g(k).
(27)
Here, k is the momentum of the out going gluon. The corresponding cross-section is

1
1
1
r = 2
(28)
dP S2 |Mr |2 ,
N 4(p p )
4
where the two body phase-space is
dP S2 =
d n1 q
d n1 k
(2 )n n (k + q p p ),
(2)n1 2k 0 (2)n1 2q 0
and the amplitude Mr is given by

(/k p
/ ) (/
p k/)
u(p) (k).
)t a
+
Mr = igs v(p
(k p )2
(p k)2
(29)
(30)
The phase-space integration can be easily done in the center of mass frame of incoming
quarks wherein the momenta of the particles can be parametrized as
s
s
(1, 0, 0, 1),
p =
(1, 0, 0, 1),
p=
2
2

k = |q|(1, sin , 0, cos ),

q = q0 , |q| sin , 0, |q| cos ,
(31)
s
s
q0 =
(1 + ),
|q| =
(1 ),
2
2
with the square of the matrix element being given by

4
1
4
1
+
+
,
|Mr |2 = 42 gs2 NCF n 2
(32)
1
1+y 1y
(1 )2
351
and y = cos . Substituting the above in Eq. (32) and performing the integration, we have
r =
(4)3/2
1 s
CF 2
N 4s
(1 + 3/2)

2

2
2
4
8
q
2 q
(1 )
+ ln 2 + ln
3
32 3
2

ln(1 )

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

2
2
2
1+
q
1+
ln( ) .
+2
ln 2 2
(1 )+
(1 )
(33)
In the above, the + prescription for a function f (z) is defined as

1
0
f (z)
dz
=
(1 z)+
1
dz
0
f (z) f (1)
.
(1 z)
(34)
Notice that the above integral is divergent as z 1. This is but a manifestation of the
collinear divergence arising due to the masslessness of the incoming quarks, and can be
safely absorbed into unrenormalized parton densities by a suitable counter term. Within
the MS scheme, the said counter term is given by

M 2 3/2 6
4 1 + 2
1
1 s
CF 2R
(1
,
ct =
(35)
N 4s
(1 + 3/2) 42
3
3 (1 )+
where M is the factorization scale. Adding the virtual corrections to the bremsstrahlung
contribution with the collinear counter term (Eq. (35)), we get, upto O(s ),
q =
1 2
(1 )
N 2s R

2

2
q
q
2 2
1 s
CF 2R 3 ln 2 2 +
(1 ) + 3(1 ) ln
+
N 4s
3
M2
2

ln(1 )
1 + 2
q
2
+ 2(1 ) + 4 1 +
(36)
+2
ln
.
(1 ) +
(1 )+
M 2
4.2. Contribution from the gluon initiated process

We now compute the final piece, namely the contribution of the Compton-like process
to order s :
g(k) + q(p) S(q) + q(p ).
The cross-section is given by
1
1
c =
4(p k) N(N 2 1)

dP S2
(37)
1
|Mc |2
2(n 2)
(38)
352
with the amplitude Mc being

Mc = igs u(p
)t

a

(/
p k/) (/
p + k/)
u(p) (k).
+
(p k)2
(p + k)2
(39)
Once again, the two-body phase-space (dP S2 ) integration can be easily done in the center
of mass frame of incoming gluon and quark wherein
|Mc |
= 2 gs2 NCF

4
(n 2)
+ y(1 )(n 2)
4 +
1 1+y

+ 6 + 10 3n n .
(40)
On performing the angular integration, we get

3/2
q2
1
(1 + 3/2)
2
(1 + 3) 43(2 + 3)
4

2
3(1 )
1
2
6(1 ) + 3(3 + )
8 + (1 ) + 43
(1 )3
1+3
3/2
1
s 1 2
1
=
4s N (1 + 3/2) 4

2
1 2
q
+ (1 )2 + 2 + (1 )2 ln 2
3
2

1
1
(1 )2
+ (1 )(7 3) .
+ 2 + (1 )2 ln
(41)
2
c =
2 gs2 1
16s N
With the MS counter term to remove the collinear divergence coming from the massless
incoming gluon being given by
ct =
2 3/2 3/2

1
s 1 2
1
1 2
M
2
R
+
(1
,
4s N
(1 + 3/2) 2
4
3
(42)
we finally have, to O(s ),

g =
2

q
(1 )2
s 1 2 1 2
1 2
2
R
+ (1 )2 ln
ln
+
(1
)
+
4s N
2
M2
2

1
+ (1 )(7 3) .
(43)
4
Note that both the quark and gluon initiated processes, after the mass factorization,
are free of any IR divergences. We use these results for our further analysis after
folding with appropriate parton distributions. For the numerical calculation we made the
renormalization scale and the factorization scale M equal (i.e., = M = F ).
353
5. Results and discussion

Having obtained the analytic expressions in the last section, we now endeavour to see
the numerical size of these corrections. To this end, one needs to define the leading-order
and the next-to-leading-order cross-sections for a pp (or equivalently pp) collider:
LO : convolute the cross-section of Eq. (7) with the appropriate LO quark distributions;
NLO : convolute the cross-section of Eqs. (36), (43) with the appropriate NLO quark and
gluon distributions; s (Q2 ) is calculated from the LEP measurement of s (MZ2 )
using the NLO evolution equation.
Of course an additional factor of 1/2 needs to be included in the cross-sections to account
/ p interactions.
for the chiral nature of the R
Before we start in earnest, a minor digression. Since QCD corrections are flavour-blind,
the value of the coupling is immaterial and only serves to set an overall scale for the
cross-section. To be concrete, we shall choose to work with
= 0.01,
irrespective of flavour and the mass of the sneutrino/slepton. While this may seem to be an
inconsistent choice for a light e or e (see Table 1), this is not quite germane to the issue at
hand.
5.1. Sneutrino production
To begin with, we concentrate on the resonance production of a sneutrino starting
with a d d initial state (at the Born level). In Fig. 1, we plot both the LO and the NLO
Fig. 1. Cross-section for resonant sneutrino production at the Tevatron (lower curves) and at the LHC (upper
curves). The solid (dashed) curves represent the NLO (LO) cross-sections. The R-parity violating coupling i11
has been set to 0.01. The three cases correspond to structure function parametrizations CTEQ5, GRV98 and
MRS98, respectively.
354
Fig. 2. The K-factor for the process involving coupling i11 as a function of the sneutrino mass as calculated for
different parton distributions. The two graphs correspond to the Tevatron and the LHC, respectively.
cross-sections as a function of the sneutrino mass and for three different choices of parton
distributions. The renormalization scale F as described in the previous section has been
chosen to be the same as the sneutrino mass.
On the face of it, the three sets of curves look quite similar, a point that we shall remark
on later. The dependence on the scalar mass is as expected and is occasioned by both the
fall in the parton-level cross-sections and the rapid decrease of the parton densities at highmomentum fractions. The latter effect, understandably, is more pronounced at the Tevatron
than at the LHC.
To parametrize the effect of the NLO corrections, it is common to introduce the K-factor:
NLO
K
(44)
,
LO
which we plot in Fig. 2. Let us concentrate first on the results for the Tevatron. The near
monotonic decrease of K with m is not unexpected. As m increases, we are sampling
increasingly larger values of parton momenta. This has two immediate consequences. For
one, the Compton contribution becomes increasingly irrelevant. But more importantly,
a large m also means that the primary quark is less able to radiate off a gluon. In
fact, as m spp , the K factor falls well below unity. As is well known, the quark
densities fall as the quark momentum fraction (xBj ) approaches unity. This fall is sharper
for NLO densities than for the LO. This actually overcompensates for the enhancement in
the partonic cross-section due to soft gluon emission.
It might seem significant that the K-factors, as calculated with different sets of parton
distributions, vary significantly amongst themselves. This only reflects the dependence
of the cross-sections on our anstz for the parton densities. This assertion is supported
by the fact the differences are more pronounced for smaller sneutrino masses, where the
cross-section receives a larger contribution from relatively low-momenta partons and hence
prone to larger parametrization errors. Interestingly, the difference in the K-factors arise,
355
Fig. 3. The dependence of the cross-sections at the LHC on the value of the factorization scale F . The ratio
R (see Eq. (45)) compares the cross-section to the reference point of F = m . The legends on the graphs
correspond to the ratio F /m . The left and right panels correspond to the NLO and LO cross-sections,
respectively. The CTEQ5 densities have been used.
in a large measure, due to the differences in the LO cross-sections. The relative differences
between the cross-sections, as calculated with the different anstze, actually decrease as
we go from the LO to the NLO calculations, and can be expected to become smaller as
progressively higher order corrections are incorporated.
Turning now to the results for the LHC, we see that, for m 300 GeV, the behaviour
is quite analogous to the case of the Tevatron. Although the fall with the sneutrino mass
seems to be slower, it should be remembered that the graph covers a much smaller range in
m / spp . The behaviour at small masses ( 300 GeV) seems puzzling though. However,
one must realize that the cross-section for such light sneutrinos is dominated by lowmomenta partons. And since existing data does not probe the parton densities unto very
low-xBj , the various anstze naturally have differing predictions. Although it does not
show up in the curves of Fig. 1, again the difference in the K-factor is dominated by
the deviations in the LO cross-sections rather than the NLO ones.
Having explored the dependence of the K-factor on the sneutrino mass and the choice
of parton densities, we now turn to the final unknown viz. the renormalization scale.
Although the most natural scale is that of the sneutrino mass (with many other similar
analyses making this choice too), the exact value of F is somewhat ambiguous. To
quantify the ensuing dependence, we define the ratio
RNLO (F ; m )
NLO (F )
,
NLO (F = m )
(45)
operative within a given parton density parametrization. An analogous expression can also
be defined for the LO cross-sections. In Fig. 3, we exhibit the variation of R(F ; m ) for
both the LO and the NLO cross-sections. As the graph shows, the variation of the crosssection with F is relatively small. Furthermore, the variation reduces as one progresses
from the LO calculation to the NLO. The last observation lends hope that the remaining
scale ambiguity can, presumably, be reduced by adding still higher order corrections. The
356
Fig. 4. Cross-section for resonant sneutrino production at the Tevatron (lower set of curves) and LHC (upper
set). For each set, the solid (dashed) refer to NLO (LO) cross-sections. The respective Born-level initial states are
indicated in each panel. The value of the R-parity violating coupling (see Eq. (46)) has been set to be 0.01 and
the CTEQ5 parametrization has been used.
approximate meeting of the curves in the LO case is but a reflection of the behaviour of the
integrated flux. For the Tevatron, a similar effect is seen at m 85 GeV.
5.1.1. Initial states with strange quarks
We now consider other possibilities for resonance production. Restricting ourselves to
sneutrinos for the moment, note that a non-zero value for some of the other couplings
could lead to alternate tree-level processes such as:
i22 : s + s i ,
i12 or i21 : d + s i .
(46)
Although the chirality structure of the interaction term is different for the two cases in the
second line, it is of no consequence for either the LO cross-section or the NLO corrections.
In Fig. 4, we plot these cross-sections4 for both the Tevatron and the LHC. For an identical
value of the Rp violating coupling, the total cross-section is much smaller than that in
Fig. 1. This is only to be expected as the strange-quark is a part of the sea and consequently
its flux is much smaller than that for the d-quark. Thus, for the LHC, we have the relation
> (d s) > (s d)
> (s s ), while for the Tevatron, the second inequality is replaced
(d d)
by an equality.
As in the previous case, we may once again choose to parametrise the NLO corrections
in the form of a K-factor. And, although we have chosen to present the cross-sections only
for the CTEQ5 parton distributions, it is quite instructive to consider the dependence on
the parametrization. In Fig. 5, we present this for the s s initial state. The wide difference
between the K-factor as calculated within CTEQ5 [28], with those obtained in the context
4 Of course, similar production mechanisms involving a b-quark in the initial state is possible too. However,
owing to the small flux for the b-quark, these are not of much interest phenomenologically.
357
Fig. 5. The K-factor for the processes with coupling i22 at the Tevatron as a function of the slepton (sneutrino)
mass.
of MRS98 [29] or GRV98 [30] may seem to be a matter of concern. Interestingly, unlike
in the case of the d d initial state, the difference in K here cannot be ascribed to the LO
parton distributions. Rather, the blame lies on the NLO parton distributions, in particular
the much larger strange-quark flux in the CTEQ5 parametrization (as compared to GRV98
or MRS98). Although this large deviation is partly offset by a sizable negative contribution
from the Compton diagram, the latter effect is clearly subdominant. While such a discrepancy might seem vexing, it is easier to appreciate once one considers the experimental
inputs in the parton density parametrizations, especially in the large x region. For example, the CTEQ collaboration uses jet measurement data whereas MRS used prompt photon
data. And since our curves for the Tevatron reach unto a much larger x value than those
for the LHC, the difference is more pronounced in the former case. Notwithstanding this
post-facto rationalization, the resultant KCTEQ5 remains uncomfortably large, and more
over, does not approach unity as m spp . This is symptomatic of an inherent problem
in the CTEQ5 parametrization for the heavier sea quark distributions.
The same problem is also reflected in the K-factors for the two other cases of Eq. (46)
namely those with s d and d s initial states. Once again KMRS98 and KGRV98 are quite
similar while KCTEQ5 is significantly different (see Fig. 6).
5.2. Charged slepton production
We finally consider slepton production. Governed by Eq. (2), the relevant piece of the
interaction Lagrangian is readily seen to have the same structure as the piece responsible
for sneutrino production. Once again, various combinations of quarkantiquark pair could
feature in the production process. However, for brevitys sake we shall confine ourselves
to a discussion of only the processes corresponding initiated, at the Born level, by ud (and
d u).
Of course, in the context of the Tevatron, the two cross-sections would be identical,
whereas in the LHC, the former would be slightly larger.
As Fig. 7 demonstrates, the behaviour of the cross-section is identical to that of
sneutrino production in d d collision (although the magnitude is somewhat larger). This
was only to be expected as the flux of the u-quark inside the proton is similar to that for the
d-quark. To be very precise, the valence u-density is approximately a factor of two larger
358
Fig. 6. The K-factor for the processes with coupling i12 or i21 at the Tevatron as a function of the slepton
(sneutrino) mass.
Fig. 7. Cross-section for charged slepton production at the Tevatron (lower set of curves) and LHC (upper set).
For each set, the solid (dashed) refer to NLO (LO) cross-sections. The respective Born-level processes are given
by d + u and u + d + . The value of the R-parity violating coupling has been set to 0.01 and the
CTEQ5 parametrization has been used.
than the valence d, whereas the sea-densities are very similar for the two. Consequently,
we should expect the behaviour of the K-factor to be quite similar again, as is borne out by
Fig. 8. The remarkable similarity between Kud (and hence Kd u too for the Tevatron case)
and Kd d (Fig. 2) is a testimonial to the fact that the only difference between the two cases
arises from the small effect due to isospin violation in the valence quark densities (which,
of course manifests itself primarily in the large scalar mass region).
6. Conclusions
To summarize, we have calculated the NLO corrections to the resonant sneutrino and
slepton production cross-sections (within R
/ p -MSSM) in the context of Tevatron and LHC.
359
Fig. 8. The K-factor for charged slepton production (processes of Fig. 7) at the Tevatron as well as at the LHC as
a function of the slepton mass. The CTEQ5 densities have been used.
We find that for processes controlled, at the Born level, by first-generation quarks, the
ensuing K-factor is a fairly sensitive function of the scalar mass. For masses below
about 1 TeV, the correction can vary between 10% and 50% at the Tevatron with the
correction falling steeply for higher masses. At the LHC, the mass-dependence is reduced
significantly, and for masses less than 2 TeV, varies only between 1.2 and 1.45. While
there is a significant dependence on the structure function used, the effect is much less
pronounced for the NLO calculation than for the LO. This lends us hope that once the
next order effects are incorporated the theoretical ambiguity would reduce to insignificant
levels.
For production processes involving quarks of higher generations, the situation is not so
simple. The K-factors could be much larger, and worse, show a marked dependence on
the particular density parametrization used. This is but a reflection of the fact that these
distributions are known with much less precision and hence vary significantly between
parametrizations. Since the production cross-sections themselves are large enough to be
interesting, an NNLO calculation thus seems to be called for.
Finally, the calculations presented in this paper are not particular to supersymmetric
theories, but can be applied to any color-singlet scalar (pseudoscalar) coupling to a quark
antiquark pair.
Acknowledgements
We would like to thank Anindya Datta for many useful discussions and, more
particularly, for his participation during the early stages of the project. D.C. and V.R. would
like to thank the Theory Division, CERN for hospitality while part of the project was being
carried out. D.C. also acknowledges the Department of Science and Technology, India for
financial assistance under the Swarnajayanti Fellowship grant.
360
References
[1] LEP Electroweak Working Group, in preparation.
[2] H.P. Nilles, Phys. Rep. 110 (1989) 1;
H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75;
S. Dawson, Nucl. Phys. B 261 (1985) 297.
[3] J.C. Pati, A. Salam, Phys. Rev. D 10 (1974) 275;
H. Georgi, S.L. Glashow, Phys. Rev. Lett. 32 (1974) 438;
For reviews, see P. Langacker, Phys. Rep. 72 (1981) 185.
[4] P. Frampton, Phys. Rev. D 42 (1990) 3892;
P.B. Pal, U. Sarkar, Phys. Rev. D 49 (1994) 3721.
[5] S. Dimopoulos, et al., Phys. Rev. D 41 (1990) 2099;
H. Dreiner, G.G. Ross, Nucl. Phys. B 365 (1991) 597;
B.C. Allanach, et al., hep-ph/9906224;
H. Dreiner, P. Richardson, M.H. Seymour, JHEP 0004 (2000) 008, hep-ph/9912407.
[6] J. Kalinowski, et al., Phys. Lett. B 414 (1997) 297, hep-ph/9708272;
J.L. Hewett, T.G. Rizzo, hep-ph/9809525;
H. Dreiner, P. Richardson, M.H. Seymour, hep-ph/9903419;
H. Dreiner, P. Richardson, M.H. Seymour, hep-ph/0001224;
G. Moreau, M. Chemtob, F. Deliot, C. Royon, E. Perez, Phys. Lett. B 475 (2000) 184, hep-ph/9910341;
G. Moreau, E. Perez, G. Polesello, hep-ph/0002130;
G. Moreau, E. Perez, G. Polesello, Nucl. Phys. B 604 (2001) 3, hep-ph/0003012;
S. Abdullin, et al., hep-ph/0005142.
[7] A. Datta, et al., Phys. Rev. D 56 (1997) 3107, hep-ph/9704257;
J.-M. Yang, et al., hep-ph/9802305;
R.J. Oakes, et al., Phys. Rev. D 57 (1998) 534, hep-ph/9707477;
E.L. Berger, B.W. Harris, Z. Sullivan, Phys. Rev. Lett. 83 (1999) 4472, hep-ph/9903549.
[8] S. Dimopoulos, L.J. Hall, Phys. Lett. B 207 (1988) 210;
G.F. Giudice, et al., hep-ph/9602207;
ECFA/DESY LC Physics Working Group, E. Accomando, et al., Phys. Rep. 299 (1998) 1, hep-ph/9705442;
J. Erler, J.L. Feng, N. Polonsky, Phys. Rev. Lett. 78 (1997) 3063, hep-ph/9612397;
J. Kalinowski, R. Rckl, H. Spiesberger, P.M. Zerwas, Phys. Lett. B 406 (1997) 314, hep-ph/9703436.
[9] V. Barger, G.F. Giudice, T. Han, Phys. Rev. D 40 (1989) 2987.
[10] J.L. Hewett, Snowmass summer study, 1990;
J. Butterworth, H. Dreiner, Nucl. Phys. B 397 (1993) 3, hep-ph/9211204;
D. Choudhury, S. Raychaudhuri, Phys. Lett. B 401 (1997) 54;
G. Altarelli, et al., Nucl. Phys. B 506 (1997) 3;
J. Kalinowski, et al., Z. Phys. C 74 (1997) 595;
T. Kon, T. Kobayashi, Phys. Lett. B 409 (1997) 265;
G. Altarelli, G.F. Giudice, M.L. Mangano, Nucl. Phys. B 506 (1997) 29;
J. Ellis, S. Lola, K. Sridhar, Phys. Lett. B 408 (1997) 252;
M. Carena, et al., Phys. Lett. B 414 (1997) 92.
[11] Z. Kunszt, W.J. Stirling, Z. Phys. C 75 (1997) 453;
T. Plehn, et al., Z. Phys. C 74 (1997) 611.
[12] Super-Kamiokande Collaboration, Y. Fukuda, et al., Phys. Rev. Lett. 81 (1998) 1562;
Super-Kamiokande Collaboration, Y. Fukuda, et al., Phys. Lett. B 436 (1998) 33.
[13] KARMEN Collaboration, B. Armbruster, et al., Phys. Lett. B 348 (1995) 19;
D. Choudhury, S. Sarkar, Phys. Lett. B 401 (1996) 87;
B. Zeitnitz, Talk given at NEUTRINO98, Takayama, 1998;
R. Maschuw, Talk given at WIN99, Cape Town, 1999;
D. Choudhury, H. Dreiner, P. Richardson, S. Sarkar, Phys. Rev. D 61 (2000) 095009.
[14] P. Fayet, Phys. Lett. B 69 (1977) 489;
G. Farrar, P. Fayet, Phys. Lett. B 76 (1978) 575.
[15] L.E. Ibanez, G.G. Ross, Nucl. Phys. B 368 (1992) 3.
[16] L.J. Hall, M. Suzuki, Nucl. Phys. B 231 (1984) 419.

[17] A. Bouquet, P. Salati, Nucl. Phys. B 284 (1987) 557;
A.E. Nelson, S.M. Barr, Phys. Lett. B 246 (1990) 141;
E. Roulet, D. Tommasini, Phys. Lett. B 256 (1991) 218;
B.A. Campbell, S. Davidson, J. Ellis, K. Olive, Phys. Lett. B 256 (1991) 457;
W. Fischler, G. Giudice, R.G. Leigh, S. Paban, Phys. Lett. B 258 (1991) 45.
[18] H. Dreiner, G.G. Ross, Nucl. Phys. B 410 (1993) 188.
[19] D0 Collaboration, B. Abbott, et al., Phys. Rev. Lett. 83 (1999) 4476;
D0 Collaboration, B. Abbott, et al., Phys. Rev. Lett. 84 (2000) 2088.
[20] CDF Collaboration, F. Abe, et al., Phys. Rev. Lett. 83 (1999) 2133;
D. Choudhury, S. Raychaudhuri, Phys. Rev. D 56 (1997) 1778.
[21] K. Agashe, M. Graesser, Phys. Rev. D 54 (1995) 4445;
D. Choudhury, P. Roy, Phys. Lett. B 378 (1996) 153;
F. Vissani, A.Yu. Smirnov, Phys. Lett. B 380 (1996) 317.
[22] G. Bhattacharyya, Nucl. Phys. B (Proc. Suppl.) 52 (1996) 83;
H. Dreiner, hep-ph/9707435;
R. Barbier, et al., hep-ph/9810232.
[23] H.V. Klapdor-Kleingrothaus, et al., Prog. Part. Nucl. Phys. 32 (1994) 261;
M. Hirsch, H.V. Klapdor-Kleingrothaus, S.G. Kovalenko, Phys. Rev. Lett. 75 (1995) 17;
M. Hirsch, H.V. Klapdor-Kleingrothaus, S.G. Kovalenko, Phys. Rev. D 53 (1996) 1329;
K.S. Babu, R.N. Mohapatra, Phys. Rev. Lett. 75 (1995) 2276, hep-ph/9506354;
J.W.F. Valle, hep-ph/9509306.
[24] A.I. Belesev, et al., Phys. Lett. B 350 (1995) 263;
C. Weinheimer, et al., Phys. Lett. B 300 (1993) 210.
[25] G. Bhattacharyya, D. Choudhury, Mod. Phys. Lett. A 10 (1995) 1699.
[26] A.S. Joshipura, V. Ravindran, S.K. Vempati, Phys. Lett. B 451 (1999) 98.
[27] S. Davidson, D. Bailey, B. Campbell, Z. Phys. C 61 (1994) 613;
C.S. Wood, Science 279 (1997) 1759;
W.J. Marciano, J.L. Rosner, Phys. Rev. Lett. 65 (1990) 2963;
W.J. Marciano, 1997 INT summer workshop.
[28] H.L. Lai, et al., Eur. Phys. J. C 12 (2000) 375, hep-ph/9903282.
[29] A.D. Martin, et al., Eur. Phys. J. C 4 (1998) 463, hep-ph/9803445.
[30] M. Gluck, E. Reya, A. Vogt, Eur. Phys. J. C 5 (1998) 461, hep-ph/9806404.
361

Predictive framework with a pair of degenerate

neutrinos at a high scale
Anjan S. Joshipura a , Saurabh D. Rindani a , N. Nimai Singh b
a Theoretical Physics Group, Physical Research Laboratory, Navrangpura, Ahmedabad 380009, India
b Department of Physics, Gauhati University, Guwahati 781014, India
Received 16 December 2002; accepted 14 March 2003
Abstract
Radiative generation of the solar scale is discussed in the presence of leptonic CP violation.
We assume that both the solar scale and Ue3 are zero at a high scale and the weak radiative corrections
generate them. It is shown that all leptonic mass matrices satisfying these requirements lead to
a unique prediction cos 2 4 sin2 A |mee |2 for the solar scale in terms of the radiative
correction parameter , the physical solar (atmospheric) mixing angles (A ) and the Majorana
neutrino mass mee probed in neutrinoless double beta decay. This relation is independent of the
mixing matrix and CP-violating phases at the high scale. The presence of CP-violating phases leads
to dilution in the solar mixing angle defined at the high scale. Because of this, bi-maximal mixing
pattern at the high energy leads to large but non-maximal solar mixing in the low-energy theory. An
illustrative model with this feature is discussed.
PACS: 14.60.Pq; 11.10.Hi
1. Introduction
The neutrino masses and mixing pattern implied [1,2] by the neutrino oscillation
experiments requires (i) two large mixing angles and A to explain respectively the
results of the solar and the atmospheric neutrino experiments, (ii) corresponding mass
scales and atm satisfying

3 102
atm
E-mail address: saurabh@prl.ernet.in (S.D. Rindani).
doi:10.1016/S0550-3213(03)00236-0
(1)
A.S. Joshipura et al. / Nuclear Physics B 660 (2003) 362372
363
and (iii) a third mixing angle which is very small 3 9 .

The smallness of 3 compared to other two large angles and that of /atm are two of
the major puzzles [3] in neutrino physics requiring theoretical explanation. One possibility
is to suppose that some symmetry leads to vanishing of or 3 or both and its breaking is
responsible for small values of these parameters. Vanishing of can be a consequence of
lepton like U (1) or more general non-Abelian symmetry such as SU(2)H . The vanishing
of 3 can also be attributed to some symmetry, e.g., Le L L invariance [4] studied
extensively in the literature. One needs to provide a mechanism for symmetry breaking in
order to generate the solar scale. An economical possibility is to suppose that physics at
a high scale leads to vanishing solar scale and the standard weak gauge bosons (and their
superpartners in supersymmetric theory) are responsible for its generation at a low scale
[57]. This provides a well-defined symmetry breaking pattern which can be deduced [8]
by studying the evolution of the neutrino mass matrix through the renormalization group
(RG) equations [9]. This evolution has been studied very extensively [10] with a different
physical motivation.
Let us suppose that neutrino masses mi and the mixing angle 3 generated by physics
at a high scale (e.g., seesaw mechanism) are given by

mi = (m, m, m ),
(2)
sin 3 (U0 )e3 = 0,
where U0 is the neutrino mixing matrix at the high scale. Clearly, a very large class of
the neutrino and the charged lepton mass matrices can lead to Eq. (2). The RG evolution
provides a systematic way to study generation of the solar scale and Ue3 at a low energy in
all these models. It was found in [11] that all models of leptonic mass matrices leading to
Eq. (2) give a unique prediction for the solar scale when CP is conserved,

cos 2 = 4 sA2 |mee |2 + O , 2 .
(3)
Here, mee is the effective neutrino mass probed in neutrinoless double beta decay (0)
and denotes [8] the size of the radiative correction induced by the Yukawa coupling of
the :

m 2 MX
ln
,
c
(4)
4v
MZ
c = 3/2, 1/cos2 in respective cases of the standard model (SM) and the minimal
sypersymmetric standard model (MSSM).
This equation implies a strong correlation between the solar scale and mee . In particular,
it requires that mee should be close to its present limit if the large mixing angle (LMA)
solution [1] is to be reproduced. CP conservation was assumed in the analysis presented
in [11]. When CP violation is allowed, the vanishing of at high scale implies that the
first two masses are degenerate in Eq. (2) only up to a phase. It is well known [8,10,12]
that CP violating phases can alter evolution of the neutrino masses in a non-trivial and
drastic manner. It is thus important to include the effects of such phases. We find that the
CP violating phases , associated with the neutrino masses influence the predicted value
of the solar mixing angle significantly but in such a way that the basic prediction Eq. (3)
obtained in a CP conserving theory remains unaffected. Unlike in Eq. (3), the radiatively
364
generated Ue3 depends on the phases and . In the following, we derive these basic
results and discuss their consequences.
2. Assumptions and results

A complex symmetric neutrino mass matrix can always be diagonalized by a unitary
matrix U . This U can be identified with the MNS [13] matrix when the neutrino mass
matrix is specified in the flavour basis. U is known to be determined by three mixing
angles i and three phases , , and can be parameterized as

U = R23 (2 )R13 3 ei R12 (1 ) diag 1, ei/2 , ei/2 .
(5)
The phase is analogous to the CKM phases and , are the phases associated with the
Majorana masses.
We denote the neutrino mass matrix in the flavour basis at a high scale by M0 and
the corresponding mixing matrix by U0 . The U0 is obtained by choosing 3 = 0 in Eq. (5).
Explicitly,

s1
0
1
0
0
c1
i/2
,
U0 = c2 s1 c2 c1 s2
(6)
0
0 e
s1 s2 c1 s2 c2
0
0
ei/2
where c1 = cos 1 , s1 = sin 1 , etc.
The solar scale vanishes if
U0T M0 U0 = diag(m, m, m ).
(7)
Several points are to be noted in connection with the above equation.

One can choose m and m real and positive without loss of generality. The non-zero
|m 2 m2 | can be identified with the atmospheric scale atm .
Given the mass ordering as in Eq. (7), (U0 )e3 denotes the mixing element probed at
CHOOZ. We assumed it to be zero which amounted to choosing 3 = 0 in Eq. (5). This
then implies that the CKM phase can be assumed zero at a high scale. CP violation is
still present through non-vanishing and .
The neutrino mass matrix M0 can be determined by inverting Eq. (7):
M0 = U0 diag(m, m, m )U0 .
(8)
The matrix U0 is however not unique. Degeneracy of masses in Eq. (7) implies that
U0 is arbitrary up to an orthogonal transformation O12 by an angle 1 in the 12 plane.
Thus U0 and U0 O12 imply the same physics. This arbitrariness in U0 amounts to the
following redefinition of the angle 1 appearing in Eq. (6):

|c1 | c1 c1 s1 s1 ei/2 ,

|s1 | c1 s + s1 c ei/2 .
(9)
1
This freedom implies that the solar angle corresponding to the 12 mixing cannot
be uniquely defined at a high scale. Eq. (9) at the same time does not allow us to
365
completely rotate away the solar angle unless = 2n . The arbitrariness in defining
the solar angle will be removed by the radiative corrections. We can thus set 1 to zero
without loss of generality.
The matrix M0 determined by Eq. (8) is modified by radiative corrections. The
radiatively corrected form of M0 follows from the relevant RG equations. We assume
RG equations corresponding to the SM or the MSSM. The modified neutrino mass matrix
is given [8] in this case by

M0 M Ig It I U0 diag(m, m, m3 ) U0 I ,
(10)
where Ig,t are calculable numbers depending on the gauge and top quark Yukawa
couplings. I is a flavour dependent matrix given by
I diag(1 + e , 1 + , 1 + ).
e, are obtained from Eq. (4) by replacing the tau mass by the electron and the
muon masses, respectively. The physical neutrino masses and mixing are obtained by
diagonalizing the above matrix. We do this approximately by retaining the contribution of
the tau Yukawa coupling and by working to the lowest order in . The effect of radiative
corrections is best seen by going to the basis in which the neutrino mass matrix M0
defined at high scale is diagonal. This is done through a rotation by the original U0 on
Eq. (10)
U T M U0
M
0
i
i
2 cos 2 c1 s1 s22
c2 s2 s1 e 2 + re 2

mIg It 2 cos 2 c1 s1 s22

1 + 2 c12 s22
c2 s2 c1 ei + rei
i
i

i
2
i
r 1 + 2 c2
c2 s2 s1 e 2 + re 2 c2 s2 c1 e
+ re
1 + 2 s12 s22
(11)
m /m
where r
and ( )/2.
Approximate diagonalization of the above matrix is straightforward. We omit the details
of this diagonalization but mention the salient points.
The structure of the upper 2 2 block in Eq. (11) implies that the correction to the solar
mixing angle arising from its diagonalization is not O( ) but O(1) unless = . This
is a well-studied phenomenon [8] which is sometimes referred to as radiative instability
of the mixing angle. This feature is related here to the arbitrariness (Eq. (9)) in defining
the mixing angle at high scale. Perturbation present in Eq. (11) helps in removing this
ambiguity.
The upper 2 2 block of Eq. (11) is diagonalized by

c1 s1 0
U12 = s1 c1 0 ,
(12)
0
0 1
366
with
tan 21 = cos
tan 21 .
2
(13)
TM
U12 resulting after rotation through
The (13) and (23) elements of the matrix U12
U12 are O(m ) and the 33 element is approximately m . The effect of these off2m2
.
diagonal elements is to induce additional mixing of the O($) where $ mm
m
atm
This mixing is quite small for physically interesting mass range m O (eV). Thus a
TM
U12 is performed by two additional rotations with
complete diagonalization of U12
mixing angles which are O($). Explicitly,
T T T T
U23
U13 U12 U0 M U0 U12 U13 U23 diag(m1 , m2 , m3 ).
(14)
The U13 is responsible for generation of the CHOOZ angle while U23 provides a small
radiative correction to the atmospheric mixing angle 2 . We neglect correction to 2 in
the following. U13 is given by

0 s3 ei
c3
.
U13 =
(15)
0
1
0
c3
s3 ei 0
Eq. (14) gives us the complete mixing matrix which can be approximately written as
U U0 U12 U13
c c3 ei
icA s c3 ei(+/2) sA s3 e(i/2)
is ei
c s3 ei( )
cA c ei(+/2) icA s s3 ei(++/2) + sA c3 ei/2
isA s c3 ei(+/2) cA s3 ei(/2) sA c ei(+/2)
isA s s3 ei + cA c3 ei/2
(16)
where

c ei = c1 c1 s1 s1 ei/2 ,

s ei = i c1 s1 + s1 c1 ei/2 ,
sA = s2 + O( ).
3 is radiatively generated and is given by

1+r
cos
|Ue3 | = |s3 | = |s3 c | = sA cA c s
1r

2

m

cos ,
sin 2A sin 2
2atm
where r m /m can be expressed as:

atm 1/2
atm
|r| = 1 2
1
.
m
2m2
(17)
(18)
(19)
The positive (negative) sign in the above equation applies to the case m > m (m < m). All
the phases appearing in Eq. (23) are expressible in terms of the original phases and .
367
Explicitly,

cos 21 1/2
s1
1
sin =
sin
,
2
cos 2
2 c
L = + 2,

atm 2
L
+O
=
.
2
2m2
The mixing matrix in Eq. (16) assumes the following
terms of O(s3 ) are neglected:1

c
1
s
s3 ei
U U0 U12 U13 cA s cA c
sA
0
0
sA s sA c
cA
(20)
(21)
(22)
simple form once non-leading
0
0
i
0
0 eiL /2

.
(23)
It is clear from Eqs. (13), (17) that the radiative corrections have removed the ambiguity
in the choice of the solar angle by fixing the 1 in Eq. (9) to 1 given in Eq. (13). The same
equation also determines the phase as given in Eq. (20). Interestingly enough, the phase
of the second mass state (which was at the high scale) is now determined to be in
Eq. (23). This corresponds to (almost) equal and opposite neutrino masses at the low scale.
The major physical effects of the radiative corrections are generation of the solar scale,
mixing angle 3 and the CKM phase which was absent at the high scale. The solar scale
follows from the eigenvalues of Eq. (11):

2
m1 Ig It m 1 + 2 sA2 s
+ O 2 ,

2
m2 Ig It m 1 + 2 sA2 c
+ O 2 ,

2
+ O 2
m3 Ig It m 1 + 2 cA
(24)
which lead to

m22 m21 = 4Ig2 It2 m2 sA2 cos 2 + O 2 .
(25)
The Majorana mass mee is obtained using Eqs. (23), (24):

2

|mee |2 Uei2 mi Ig2 It2 m2 cos2 2 + O( ).
(26)
The initial phases , appear in the above equation only implicitly through the solar angle
. This is a consequence of the fact that physical neutrinos resulting after the radiative
corrections consist of a pseudo-Dirac pair with (almost) equal and opposite masses.
Eqs. (18), (25) are predictions of the scheme. The common mass m, Ig It of the
degenerate pair can be identified with the electron neutrino mass me probed through the
direct neutrino mass search, e.g., in tritium beta decay [14]. It is also probed through
measurement of the Majorana mass parameter mee [15]. One can in fact eliminate m
from Eq. (25) using Eq. (26). This leads to the prediction (3) already mentioned in
1 In writing this form, we have made a phase rotation on Eq. (16) from the left and right by two appropriate
diagonal phase matrices PL,R . The PL is absorbed in redefining the charged lepton fields giving us the final
Eq. (23).
368
the introduction. This prediction involves only low energy measurable parameters and is
independent of the CP violating phase.
3. Phenomenological consequences
The five observables, namely , , |mee |, me = mIg It and Ue3 , are correlated
through Eqs. (3), (18), (25). We now study the consequences of this correlation. is
negative in case of the MSSM. This implies a negative cos 2 and hence only much
less preferred dark region of the solar parameter space. This excludes the LMA and LOW
solutions in case of the MSSM but the SM can easily allow them. These solutions are
realized only for the specific range in |mee |. This is displayed in Fig. 1 where we show
contours of the and the electron neutrino mass me in the tan2 |mee | plane. The
values of tan2 are restricted to the typical range 0.20.8 allowed by the LMA or
LOW solution. For these values, one obtains a in the required range 104 105 eV2
provided mee 0.11 eV. This value is close to the experimental limit [15]. We also show
the contours corresponding to the electron neutrino mass me equal to 0.5 eV and 2.0 eV
in the same plot. It is seen that me is restricted to lie in the range 0.52 eV in case of the
LMA solution.
Fig. 1. Contours of (dotted), |Ue3 | (dashed) and the electron neutrino mass me (solid) as a function of
tan2 and mee (in eV). The upper (lower) curves correspond to = 104 (105 ) eV2 , |Ue3 | = 0.02 (0.01)
and the me = 2.0 (0.5) eV, respectively.
369
The predicted values of the |Ue3 | are also shown in the figure. |Ue3 | is seen to be
restricted to a typical range 0.0010.02 in the region of the mee tan2 plane allowed
by the LMA solution.
The results presented above are completely in terms of the low energy variables and
do not need any knowledge of the parameters at the high scale. Let us now comment on
possible choices of the high scale parameter. Eqs. (13), (17) are equivalent to the relation
.
(27)
2
The initial value of the solar mixing angle 1 is subject to the arbitrariness noted in Eq. (9).
We see that irrespective of this, any choice of 1 and must satisfy

sin2 2 (0.60.9).
sin2 21 , sin2
(28)
2
sin2 2 = sin2 21 sin2
It is seen that the mixing angle is reduced compared to its value at a high scale.
This is phenomenologically interesting. One of the preferred phenomenological schemes
corresponds to bi-maximal mixing [16] which can arise from symmetry considerations.
However, the present solar data do not favour strictly maximal mixing [1]. Eq. (27) shows
that one can start with bi-maximal mixing at a high scale and radiative corrections would
lead to the desired reduction provided the CP-violating phase is chosen non-zero and
different from at a high scale. The presence of also plays another important role.
= and maximal solar mixing corresponds to vanishing mee with the consequence that
the solar scale arise only at O(2 , ), see Eq. (3). The natural value for the solar scale
lies in the vacuum region in this case [17,18]. An different from alters this and allows
strictly bi-maximal mixing and a non-zero mee at a high scale.
The radiative reduction in the solar angle found here is to be contrasted with a similar
analysis presented recently in [19]. This analysis assumed vanishing Ue3 but a non-zero
at the high scale itself. It was then found [19] that radiative corrections tend to increase the
sin2 2 compared to its value at the high scale. This does not allow bi-maximal mixing at
the high scale in contrast to what is found here.
The analysis presented so far holds for all models satisfying Eq. (2) at a high scale.
While many possibilities exist, let us give an illustrative example which corresponds to
bi-maximal mixing. This is specified by the following neutrino mass matrix in the flavour
basis:

a ib
0
T
M0 = R23 (2 ) ib a
(29)
0
(2 ).
R23

i
0 0 me
The parameters a, b, m and 2 are assumed real.
0 as:
Define the mixing matrix U
1
1 e i/2
0
2
2
0 = ei/4 R23 (2 ) 1
1 e i/2
U
0
2
2
0
ei/2
(30)
370
This matrix diagonalizes Eq. (29):

0 = diag(m, m, m ),
0T M0 U
U
with
b
= ,
2
a

m2 = a 2 + b 2 .
tan
(31)
We thus have maximally mixed degenerate neutrinos at a high scale. The maximal 1
implies maximal 1 through Eq. (13). The solar angle and the low scale CP violating
phase follows respectively from Eqs. (17) and (20):
tan = cot ,
4
+ 2
.
4
It is seen that leads to large solar mixing. Specifically, b 3 a leads to

30 . The radiatively generated solar scale can naturally fall in the LMA region as
already discussed before in the general context. Eq. (29) in this way leads to the required
pattern of neutrino masses and mixing.
The texture presented here is quite similar to the one studied in Ref. [17] which assumed
zero a and a real ib. For the reasons already mentioned, this model leads only to the vacuum
solution after radiative corrections are included. The presence of a and additional phase in
b alters this and allows one to obtain the LMA solution.
It is possible [18] to obtain the above texture in the context of seesaw model by invoking
additional horizontal2 symmetry. One way [3] of realizing the above texture is to assume a
charged lepton mixing matrix with only mixing. This would generate R23 in Eq. (29).
The neutrino mass matrix in the weak basis is then given by the block diagonal form
explicitly displayed in Eq. (29). An explicit model was presented with these features in [18]
in a CP conserving situation. We do not elaborate on it here since a trivial modification3 of
this model incorporating CP violation leads to the mass matrix presented in Eq. (29).
4. Summary
The presently available information on neutrino oscillations can be nicely understood
if two of the neutrinos pair up to form a pseudo-Dirac state. This can be obtained from a
degenerate neutrino pair by means of standard radiative corrections. We discussed basic
predictions of this picture including the important effects of the CP-violating phases. The
scheme presented here has testable predictions: the LMA solution requires mee 0.11 eV
close to its present limit, relatively small Ue3 0.0010.01 and observable 0.52 eV
neutrino mass in beta decay.
2 SU(2) symmetry has also been recently used in [20] to generate bi-maximal mixing pattern.
H
3 This modification amounts to assuming a non-zero vacuum expectation value (vev) for the CP-odd field T 2
and vanishing vev for the CP even field T 1 in the notation of [18].
371
The CP violating phases and associated with the Majorana masses do not effect the
basic prediction (3) of the scheme but play an important role in diluting the solar mixing
angle defined at a high scale. This allows bi-maximal mixing at the high scale and large
but non-maximal solar angle at the low scale in accordance with the demand of the current
solar neutrino results.
Acknowledgement
Nimai Singh wishes to acknowledge local hospitality at Physical Research Laboratory
during initial part of this work.
References
[1] For reviews of various models, see A.Yu. Smirnov, hep-ph/9901208;
R.N. Mohapatra, hep-ph/9910365;
A.S. Joshipura, Pramana 54 (2000) 119;
G. Altarelli, F. Feruglio, hep-ph/0206077;
M.C. Gonzalez-Garcia, Y. Nir, hep-ph/0206077;
A.Yu. Smirnov, hep-ph/0209131.
[2] M. Frigerio, A.Yu. Smirnov, Nucl. Phys. B 640 (2002) 233, hep-ph/0202247;
M. Frigerio, A.Yu. Smirnov, hep-ph/0207366.
[3] S. Barr, I. Dorsner, Nucl. Phys. B 585 (2000) 79.
[4] A.S. Joshipura, Phys. Rev. D 60 (1999) 053002;
R.N. Mohapatra, A. Perez-Lorenzana, C.A. de Pires, Phys. Lett. B 474 (2000) 355;
A.S. Joshipura, S.D. Rindani, Phys. Lett. B 464 (1999) 239;
A.S. Joshipura, S.D. Rindani, Eur. Phys. J. C 14 (2000) 85;
H.S. Goh, R.N. Mohapatra, S.P. Ng, hep-ph/0205131;
L. Lavoura, Phys. Rev. D 62 (2000) 093011;
W. Grimus, L. Lavoura, Phys. Rev. D 62 (2000) 093012;
W. Grimus, L. Lavoura, JHEP 07 (2001) 045;
R.N. Mohapatra, hep-ph/0107274;
R.N. Mohapatra, Phys. Rev. D 64 (2001) 091301;
K.S. Babu, R.N. Mohapatra, Phys. Lett. B 532 (2002) 77.
[5] L. Wolfenstein, Nucl. Phys. B 186 (1981) 147;
S.T. Petcov, Phys. Lett. B 110 (1982) 245;
C.N. Leung, S.T. Petcov, Phys. Lett. B 125 (1983) 461.
[6] A.S. Joshipura, S.D. Rindani, Phys. Lett. B 494 (2000) 114.
[7] A.S. Joshipura, S.D. Rindani, hep-ph/0202064.
[8] P.H. Chankowski, S. Pokorski, Int. J. Mod. Phys. A 17 (2002) 575.
[9] K.S. Babu, C.N. Leung, J. Pantaleone, Phys. Lett. B 319 (1993) 191;
P.H. Chankowaski, Z. Pluciennik, Phys. Lett. B 316 (1993) 312;
S. Antusch, et al., Phys. Lett. B 519 (2001) 238;
S. Antusch, et al., Phys. Lett. B 525 (2002) 130.
[10] J. Ellis, et al., Eur. Phys. J. C 9 (1999) 310;
J. Ellis, S. Lola, Phys. Lett. B 458 (1999) 389;
A. Casas, et al., Nucl. Phys. B 556 (1999) 3;
A. Casas, et al., JHEP 9909 (1999) 015;
372
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
A. Dighe, A.S. Joshipura, hep-ph/0010079;

N. Haba, et al., Phys. Lett. B 489 (2000) 184;
N. Haba, N. Okamura, Eur. Phys. J. C 14 (2000) 347;
E.J. Chun, Phys. Lett. B 505 (2001) 155;
K.R.S. Balaji, et al., Phys. Rev. Lett. 84 (2000) 5034;
K.R.S. Balaji, et al., Phys. Lett. B 481 (2000) 33;
G. Dutta, hep-ph/0202097;
G. Dutta, hep-ph/0203222;
N. Haba, N. Okumura, M. Sugiura, Prog. Theor. Phys. 103 (2000) 367;
S.F. King, N. Nimai Singh, Nucl. Phys. B 591 (2000) 3;
S.F. King, N. Nimai Singh, Nucl. Phys. B 596 (2001) 81.
A.S. Joshipura, Phys. Lett. B 543 (2002) 276, hep-ph/0205038.
N.H.Y. Matsui, N. Okamura, Eur. Phys. J. C 17 (2000) 513;
N. Haba, et al., Prog. Theor. Phys. 103 (2000) 145.
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
J. Bonn, et al., Nucl. Phys. B (Proc. Suppl.) 91 (2001) 273;
Extensive discussion can be found in Y. Farzan, O. Peres, A.Yu. Smirnov, hep-ph/0105105;
F. Feruglio, A. Strumia, F. Vissani, hep-ph/0201291.
H.V. Klapdor-Kleingrothaus, et al., Eur. Phys. J. A 12 (2001) 147;
H.V. Klapdor-Kleingrothaus, et al., Mod. Phys. Lett. A 16 (2001) 2409;
C.E. Aalseth, et al., hep-ex/0202018;
H.L. Harney, hep-ph/0205293;
H.V. Klapdor-Kleingrothaus, hep-ph/0205228.
Y. Nomura, T. Yanagida, Phys. Rev. D 59 (1999) 017303;
S. Davidson, S.F. King, Phys. Lett. B 445 (1998) 191;
R.N. Mohapatra, S. Nussinov, Phys. Rev. D 60 (1999) 013002;
R.N. Mohapatra, S. Nussinov, Phys. Lett. B 441 (1998) 299;
C. Jarlskog, et al., Phys. Lett. B 449 (1999) 240;
M. Jezabek, Y. Sumino, Phys. Lett. B 457 (1999) 139;
Y.-L. Wu, hep-ph/9905222;
C.H. Albright, S.M. Barr, Phys. Lett. B 461 (1999) 218;
C.S. Kim, J.D. Kim, Phys. Rev. D 61 (2000) 057302;
R.N. Mohapatra, A. Perez-Lorenzana, Phys. Lett. B 474 (2000) 355;
A. Ghosal, Phys. Rev. D 62 (2000) 092001;
T. Kitabaysahi, M. Yasue, hep-ph/0006040;
Y. Koide, A. Ghosal, Phys. Rev. D 63 (2001) 037301;
K. Choi, et al., Phys. Rev. D 64 (2001) 113013;
K.S. Babu, S.M. Barr, Phys. Lett. B 525 (2002) 289;
E. Ma, S. Roy, D.P. Roy, Phys. Lett. B 525 (2002) 101;
B. Brahmachari, A.S. Choubey, Phys. Lett. B 531 (2002) 99;
C. Giunti, M. Tanimotot, hep-ph/0207096.
R. Barbieri, et al., Phys. Lett. B 445 (1999) 239.
A.S. Joshipura, S.D. Rindani, hep-ph/0211404, Phys. Rev. D, in press.
T. Miura, T. Shindou, E. Takasugi, hep-ph/02060207.
R. Kuchimanchi, R.N. Mohapatra, hep-ph/0207110;
R. Kuchimanchi, R.N. Mohapatra, hep-ph/0207373.

Stringy instability of topologically non-trivial AdS

black holes and of de Sitter S-brane spacetimes
Brett McInnes
Department of Mathematics, National University of Singapore,
10 Kent Ridge Crescent, Singapore 119260, Singapore
Received 30 May 2002; received in revised form 6 January 2003; accepted 31 March 2003
Abstract
Seiberg and Witten have discussed a specifically stringy kind of instability which arises in
connection with large branes in asymptotically AdS spacetimes. It is easy to see that this instability
actually arises in most five-dimensional asymptotically AdS black hole string spacetimes with
non-trivial horizon topologies. We point out that this is a more serious problem than it may at
first seem, for it cannot be resolved even by taking into account the effect of the branes on the
geometry of spacetime. (It is ultimately due to the topology of spacetime, not its geometry.) Next,
assuming the validity of some kind of dS/CFT correspondence, we argue that asymptotically de Sitter
versions of the HullStromingerGutperle S-brane spacetimes are also unstable in this topological
sense, at least in the case where the R-symmetries are preserved. We conjecture that this is due
to the unrestrained creation of late branes, the spacelike analogue of large branes, at very late
cosmological times.
1. Large brane instability and topologically non-trivial AdS black holes

The fact that asymptotically AdS black holes can have topologically non-trivial event
horizons [13] has attracted considerable attention in connection with the AdS/CFT
correspondence. In five dimensions [4], apart from the spherical, flat, and hyperbolic
possibilities for the structure of the event horizon, Milnors [5] prime decomposition
of compact orientable 3-manifolds suggests further candidates, and black hole spacetimes
with some of the corresponding event horizons have been constructed in a remarkable
paper of Cadeau and Woolgar [6].
E-mail address: matmcinn@nus.edu.sg (B. McInnes).
doi:10.1016/S0550-3213(03)00268-2
374
B. McInnes / Nuclear Physics B 660 (2003) 373388
In [7], Witten remarks that if one wishes to generalize the AdS/CFT correspondence to
bulk manifolds which are merely asymptotically AdS, then (in the Euclidean formulation
at least) one must take care that the scalar curvature at conformal infinity should remain
non-negative. For otherwise the massless scalars of the CFT will undergo runaway
behaviour. For AdS space itself, the (Euclidean) boundary is a conformal sphere with its
canonical structure, represented by a metric of positive scalar curvature; the latter prevents
a runaway through the conformal coupling term. Notice that it is not claimed that all
CFTs misbehave on spaces of negative curvaturethis is of course not the case. The
instability arises only for a subclass of CFTs arising holographically, as for example
in the AdS/CFT correspondence. (This point is emphasized in [8]; see also [9].) Note that
the two-dimensional case is particularly delicate here; but we stress that, in any case, none
of the CFTs discussed in this work will be defined on a manifold of dimension less than
three.
The bulk version of this runaway was explained in detail in [10]; the instability arises
from the nucleation and growth of large branes, that is, in the five-dimensional case,
3-branes which are generated at large distances, near the boundary. The relevant
Euclidean action for a (D 1)-brane in a (D + 1)-dimensional asymptotically AdS space
is, in the notation of [10],

TrD
2D
8
D2 2
2
g (1 q) D2 +
+
R
()
S = D0
2
(D 2)2
4(D 1)

2(D4)
+ O D2
(1)
.
Here D is strictly greater than 2that is, this equation, and consequently the rest of our
discussion, is only valid for spacetimes of dimension at least 4 (and consequently only
for conformal boundaries of dimension at least 3). The brane carries a charge q under a
background antisymmetric field; the field tends to infinity as the conformal boundary
is approached, and R is the scalar curvature of that boundary. In the BPS case, the action
will be unbounded below if R is negative: there will be unstable production of branes
near the boundary. The case where R is zero is more delicate; one expects instability in
some cases but not in others, depending on higher-order corrections [8]. As the boundary
metric is only defined modulo conformal factors, and as scalar curvature is not a conformal
invariant, perhaps we should clarify. A conformally invariant formulation is most easily
given in terms of the scalar fields on the boundary. These are governed by the conformal
Laplacian, defined (for D > 2) by
D2
R,
(2)
4(D 1)
where is the ordinary Laplacian. The conformally invariant form of the stability
condition is that this operator should be non-negative [9]. For this to hold, it must be
possible to choose a conformal gauge such that R (a function of position in general) is
everywhere non-negative. In the case of a compact boundary, this is easier to understand
because it is then always possible [11] to choose a conformal gauge such that R is
constant. This is usually, but not invariably [12], also possible for complete, non-compact
Riemannian manifolds.
CON = +
375
Now consider a five-dimensional Euclidean AdS black hole with metric

g(EAdS5 ; hyper) = f5 (r) dt dt + f5 1 (r) dr dr

+ r 2 hij H 3 / dx i dx j
(3)
with
f5 = 1
16G
3r 2 Vol(H 3 / )
r2
,
L2
(4)
where is the mass parameter and hij [H 3 / ] dx i dx j is the metric on the threedimensional horizon, H 3 / . Here H 3 is the three-dimensional hyperbolic space of
constant sectional curvature 1, and is an infinite discrete group acting on H 3 in such a
way that the quotient is compact. The metric at infinity (that is, the canonical representative
of the induced conformal structure on the boundary) is

4 = dt dt + L2 hij H 3 / dx i dx j .
(5)
The scalar curvature of this metric is 6/L2 . Thus, large branes apparently grow
uncontrollably at large distances in this spacetime. Before commenting on this, let us take
note of the topology of the space on which 4 is defined. First, as usual for Euclidean
black holes, we have to identify time periodically, so t is a parameter on a circle. Next,
the topology of hyperbolic space is that of R3 , so we see finally that the topology of the
boundary is that of S 1 R3 / . Of course, the conformal boundary of Euclidean AdS space
itself has a very different topology, namely that of a sphere. Thus we see that a spacetime
can be asymptotically AdS, and yet have a conformal infinity with a topology which differs
very radically from that of the boundary of AdS space.
It is clear that the hyperbolic AdS black hole is not stable as a string background.
However, we wish to argue that it is by no means obvious, from this alone, that such
spacetimes must be rejected as string backgrounds. Instability, in reality, rarely leads to
indefinitely increasing perturbations: it is usually ultimately self-limiting, and we should
assume that this is the case here unless we can prove the contrary. In using the Seiberg
Witten criterion in the above way, we are ignoring the effect of the branes on the spacetime
geometry. This is a good approximation at short distances, but eventually, at extremely
large distances from the event horizon, the branes must have such an effect. In the
Lorentzian picture, the geometry of the spacelike sections will begin to evolve in response
to the branes, and, as can be seen from the form of Eqs. (3) and (5) (in which the geometry
of the spacelike sections is clearly influenced by the geometry at infinity and vice versa),
this means that the geometry at infinity must also change. This will in general change
the scalar curvature, so the SeibergWitten instability condition might cease to hold. In
physical language, one would say that the back-reaction of the branes on the geometry
eventually causes the spacetime to settle down to a new, approximately static state. But
now we have an obvious consistency check: clearly, the scalar curvature of the boundary of
the (Euclidean version of) the final spacetime geometry must be positive or zero; otherwise
we would have a contradiction. In the Euclidean language of [10]: in order for large-brane
instability to limit itself through back-reaction, it must be possible to deform the Euclidean
version of the spacetime in such a way that the scalar curvature at infinity changes from
376
negative to non-negative values. In short, unless we can prove otherwise, we should assume
that the real significance of SeibergWitten instability for this spacetime is that formulae
(3) and (4) are not to be trusted at extremely large distances, so that the metric 4 does not
accurately reflect the geometry of the boundary in this case. This greatly complicates the
study of these black holes from the AdS/CFT point of view, but in itself it does not rule
them out as string backgrounds.
Circumstantial evidence that something of this kind can happen comes from the study
of asymptotically AdS black holes with flat, toral event horizons. Intuitively, one would
expect that AdS itself should be the thermal background for any asymptotically AdS black
hole, and this is certainly the case when the horizon is spherical. But it has been argued
[13] that it is not so when the horizon is a torus. The argument is based on a conjecture,
due to Horowitz and Myers [14], that the AdS soliton is the lowest energy metric among
all metrics which are asymptotic to it sufficiently rapidly. The (Euclidean) metric of the
soliton is given, in n + 1 dimensions, by

r0n 1
L2
r2
r0 n
2
dr dr + 2 1 n d d
g(AdS; soliton) = r dt dt + 2 1 n
r
r
L
r
+ r2
n2
di di .
(6)
i=1
Here r0 and L are positive constants, r is a radial coordinate satisfying r > r0 , and and the
i are angular coordinates of various periodicities. In the Euclidean case, t too is periodic.
Thus the conformal infinity of the soliton has the structure of an n-dimensional torus, T n .
This is of course the same as the conformal infinity of an asymptotically AdS black hole
with a flat (toral) horizon, and the black hole metric does approach that of the soliton
sufficiently rapidly for the HorowitzMyers conjecture to apply. If the latter is correct,
then, as argued in [13], we should certainly use the soliton as the thermal background for
toral asymptotically AdS black holes, not AdS itself.
There is now quite impressive evidence in favour of the HorowitzMyers conjecture
[1517] and for the claim that the soliton is the correct thermal background for toral
black holes [18,19]. The resulting picture of these toral-boundary spacetimes is very
satisfactory; for example, they exhibit a well-defined confinement/deconfinement transition
similar to, but interestingly different from, the more familiar transition which is known to
occur [20] in the case of the AdSSchwarzschild black hole. There is certainly no hint
of any kind of runaway on the boundary or large-brane instability in the bulk. Now the
conformal structure at infinity is represented by an exactly flat metric. But suppose that
we slightly perturb the geometry of either the toral black hole or the soliton at some point
deep in the bulk. Suppose that this perturbation causes the boundary conformal structure
to change so that it is no longer the conformal structure represented by the flat metric.
Then, using techniques to be explained below, one can show that the scalar curvature of
the resulting metric on the boundary cannot be everywhere positive or zero. That is, the
slightest perturbation of the boundary conformal geometries of these spacetimes renders
them unstable in the SeibergWitten sense.
We suggest that this can be reconciled with the good thermodynamical behaviour
of these spacetimes by invoking back-reaction as above. That is, we predict that the
377
(Lorentzian) geometry evolves in such a way that the (Euclidean version of) the final
metric corresponds to a boundary metric with a scalar curvature that has been driven back
towards zero. (It cannot become positive everywheresee below.) If this prediction proves
to be false, then the implication is that the AdS soliton is unstable against arbitrarily small
fluctuations, which seems very unlikely. (Note that it follows from the main theorem stated
in the next section that every metric of zero scalar curvature on a torus is perfectly flatthat
is, if the scalar curvature vanishes on a torus, so does every curvature component. This does
not imply, however, that the final metric after back-reaction effects is necessarily identical
to that given in Eq. (6). For it is known [21] that when the boundary is topologically a
torus, the boundary conformal structure does not uniquely determine the bulk metric, not
even when the bulk is an Einstein manifold.)
The question, then, is this: can the manifold S 1 R3 / , which we obtained above
as the Euclidean boundary of the five-dimensional hyperbolic AdS black hole spacetime,
be given a metric of non-negative scalar curvature? Intuition may suggest that it should
be impossible to find a metric of positive scalar curvature on R3 / , but, before we trust
intuition, note that it is certainly possible [22, p. 123], to find a metric of constant negative
scalar curvature on the 3-sphere S 3 . The point is that the scalar curvature is just the
average of all of the curvature components. Thus, if even one component can be made
sufficiently negative at each point, the scalar curvature on S 3 will be negative at each point.
The reader can visualize this by imagining a deformation of the ordinary 3-sphere such
that in some directions at each point the geometry becomes saddle-like, while remaining
sphere-like in other directions. (This corresponds to modifying the metric, but not the
topology, of the 3-sphere.) Similarly, while it is of course impossible to force all of the
sectional curvatures on R3 / to be positive, one would be surprised to find that this cannot
be arranged for the average. Even if it is indeed impossible to construct a positive scalar
curvature metric on R3 / , it might still be possible to construct a warped product (or
even more general) metric on S 1 R3 / with positive scalar curvature, since warping
can certainly change the sign of scalar curvaturecompare the positive scalar curvature
product metric dx dx + d d + sin2 ( ) d d on (0, ) S 2 with the negative
scalar curvature warped metric dx dx + sinh2 (x)[d d + sin2 ( ) d d]. Clearly
we need some powerful mathematical technique to settle this.
Such a technique exists, and in the next section we give a very brief overview of it.
The upshot is that it is actually impossible to define a metric of positive or zero scalar
curvature on this manifold: no matter how we deform it, the scalar curvature remains
resolutely negative. Thus, these black hole spacetimes are unstable (if we embed them
in string theory) in a very radical way. The only escape route is to suppose that the
branes actually change the topology of spacetime, a possibility that we shall also consider.
Finally, in Section 3 we show that, if some kind of dS/CFT correspondence is valid,
then any asymptotically de Sitter version of the HullStromingerGutperle S-brane
spacetime is also radically unstable in the same sense, at least if we maintain the ansatz
used to obtain all known examples of S-brane spacetimesthat the R-symmetries are
preserved. We believe that this is due to late S-branes, the spacelike analogue of large
branes.
378
2. Topologically induced instability

In this section we introduce a geometric technique which, in view of the importance of
scalar curvature in SeibergWitten instability, is the natural one for our purposes. The main
reference is [23].
A smooth map f : X Y from one Riemannian manifold to another is said to be #contracting if for any piecewise smooth curve C in X,

L f (C) # L[C],
(7)
where L denotes the length. A map which takes every point of X to a single point of Y
evidently satisfies this condition for any positive #. To avoid this trivial case, we consider
only mappings of non-zero degree (see [23, p. 303]; note that mappings may have to
be constant at infinity to make sense of this if X happens to be non-compact). An
n-dimensional manifold M is said to be enlargeable if, given any Riemannian metric
on M, for any positive # there exists an orientable covering manifold which admits an
#-contracting map (with respect to the pull-back metric) of non-zero degree onto the unit
n-sphere. Notice that enlargeability is a topological condition; we can nevertheless think
of an enlargeable manifold as one which, like a torus but unlike a sphere, has arbitrarily
large covering spaces. Simple examples are provided by compact manifolds, such as tori
and the underlying manifolds of compact hyperbolic spaces, which admit a metric of nonpositive sectional curvature. Thus, R3 / above is enlargeable. The following properties of
enlargeable manifolds are important. (See [23, p. 306].)
(1) The product of any compact enlargeable manifold with a torus of any dimension is
again enlargeable.
(2) The connected sum of a compact enlargeable manifold with any compact manifold is
again enlargeable.
(3) If the scalar curvature of a Riemannian metric on a compact enlargeable spin manifold
is non-negative, then the metric must be flat.
(Recall that the connected sum of two manifolds is obtained by removing small balls
from each, and then joining the two along the resulting boundaries.) Now point (1) informs
us that S 1 R3 / is enlargeable. Point (3) then implies, since this space (like all products
of a circle with an orientable compact 3-dimensional manifold) is a spin manifold, that
there can be no metric of positive scalar curvature on this manifold, not even if we allow
warped products. Furthermore, if the scalar curvature were zero, then the metric would
have to be flat. But this is not possible: S 1 R3 / does not have the topology of a flat
manifold. (One says that is homotopically atoroidal [22, p. 158] and that, as the name
suggests, means that cannot occur as part of the fundamental group of a manifold which
can be flat [24].) Thus, the scalar curvature of S 1 R3 / cannot be everywhere positive
or zero, no matter how the manifold is deformed.
We conclude that once large branes begin to develop in the hyperbolic AdS black
hole spacetime, nothing can rein in the instability: for, no matter how the branes deform
the spacetime, the scalar curvature at infinity can never become everywhere positive or
zero. The instability is induced topologically [25]. We are forced to conclude that these
379
spacetimes simply cannot arise as solutions of string or M-theory. In fact, this conclusion
is actually consistent with the findings of [13], where the authors remark that no analogue
of the AdS soliton seems to exist in the hyperbolic case. It is now clear why this is so: there
can be no well-behaved ground state, analogous to the AdS soliton, for these radically
unstable spacetimes. (We agree with these authors that pure AdS is not the correct ground
state to use in the hyperbolic case; this is argued most convincingly in the introduction to
[16].)
More generally, Milnors [5] prime decomposition theorem states that any compact
orientable 3-manifold M 3 can be expressed as a connected sum in the following way:

M 3 = 1 #2 # # S 1 S 2 # S 1 S 2 # #K1 #K2 # ,
(8)
where each i is a manifold covered by a homotopy 3-sphere, where # denotes the

connected sum, and where each Ki is an EilenbergMacLane space of the form K(, 1).
We shall assume the truth of the Poincar conjecture, so that the reader can interpret
homotopy 3-sphere as S 3 . Then the i are just quotients of S 3 by (completely known)
finite groups; for example, the real projective space S 3 /Z2 is one possibility. (See [26] for a
readily accessible statement of the classification.) A K(, 1) space is just a 3-dimensional
manifold whose only non-trivial homotopy group is its fundamental group. It is not yet
proven that all compact K(, 1) spaces are enlargeable, but all known examples are so,
and furthermore it is known [23, p. 324], that no such manifold can accept a metric of
positive scalar curvature, and that the same is true of the connected sum of a K(, 1) with
any other compact manifold. Therefore, in view of properties (2) and (3) listed above, it is
reasonable to conjecture that all compact K(, 1) spaces are enlargeable.
Assuming the truth of this, we see that any compact orientable 3-manifold is enlargeable
if it has at least one Ki in its Milnor decomposition. Thus we find that compact orientable
3-manifolds fall into three categories: the six well-understood [24] manifolds which can be
flat (see [27] for an accessible review), the enlargeable manifolds of non-flat topology,
and manifolds of the form 1 #2 # #(S 1 S 2 )#(S 1 S 2 )# . Every member of this
last class has a metric of constant positive scalar curvature [28,29]. Assume that, as is
the case in physically interesting examples, the topology of the black hole spacetime is
a product of the event horizon topology with R2 . Then the conformal boundary of the
corresponding Euclidean asymptotically AdS black hole spacetime in five dimensions will
be a product of S 1 with a member of one of these classes, and so property (1) above
implies that the scalar curvature of the conformal boundary cannot be non-negative unless
the event horizon is either flat or of the form 1 #2 # #(S 1 S 2 )#(S 1 S 2 )# . Hence
we conclude that large-brane instability rules out all five-dimensional asymptotically AdS
black holes (as string backgrounds) except those with flat or 1 #2 # #(S 1 S 2 )#(S 1
S 2 )# event horizons. (Strictly speaking, this is under the assumption that the spacetime
topology has the above product form. We conjecture, however, that it is true in general
that the conformal boundary of (the Euclideanized version of) a black hole spacetime
with an enlargeable event horizon is necessarily itself enlargeable.) The flat case we have
considered already, and the AdSSchwarzschild black hole can have any i as event
horizon. It would be interesting to exhibit a black holeperhaps one should really say
black stringwith, for example, 1 #2 #(S 1 S 2 ) as event horizon. In any case, these
380
are the only remaining possibilities for event horizons of string theoretic asymptotically
AdS black holes in five dimensions.
Throughout this discussion, we have assumed that while the nucleation of large branes
can change the geometry of spacetime, it cannot change the topology. But it is well known
that D-branes can in fact change the topology of 10- or 11-dimensional spacetimes in
string/M-theory [30,31]. However, that kind of topology change usually involves changes
of topology among the members of some family of spacetimes as a parameter is changed.
Even in the cases where the topology change occurs within one spacetime, as in [32], it
occurs along some spacelike dimension of a higher-dimensional spacetime. In our case, in
order to affect the topology at infinity, the branes would have to change the topology of
the spacelike sections as they evolve in time in the Lorentzian version of the spacetime.
(Again, to see this, consider Eqs. (3) and (5): the topology of the boundary is controlled
by that of the spatial sections.) But classical theorems of Geroch [33] and Tipler [34]
state that spacetimes in which topology changes occur due to temporal evolution are
singular or contain closed timelike worldlines. Admittedly, Horowitz [35] has argued that
the singularities are not necessarily very drastic in cases where the topology change is
very simple. Even if we accept that argument, however, the topology change here would
necessarily be very extremefrom a compact hyperbolic space, for example, with its
infinite and very complex fundamental group, to a sphere. It is most unlikely that such
a major topological change can be effected without inducing non-innocuous singularities
or causality violations. In short, it is conceivable that large branes can avert a catastrophic
runaway by modifying the topology of spacetime, but it is highly likely that this has to
be paid for by some kind of equally catastrophic gravitational collapse in the region far
outside the event horizon. Thus it seems clear that changing topology cannot save the
situation, though the details of this deserve further investigation.
We wish to impress on the reader the extreme nature of the restriction imposed on black
hole horizons by SeibergWitten instability. For there is a definite sense in which compact
orientable 3-manifolds having no K(, 1) component in their Milnor decomposition are
a tiny minority. This instability rules out a large number of apparently acceptable
spacetimes. We now argue that, if there is a dS/CFT correspondence, it has a similar
effect in the case of asymptotically de Sitter S-brane spacetimes.
3. Late brane instability of asymptotically de Sitter S-brane spacetimes

( Throughout this section, for technical reasons connected with the form of the S-brane
metric we shall consider below, all spacetimes will be four-dimensional; therefore all
event horizons will be two-dimensional and all CFTs will be defined on three-dimensional
conformal boundaries.)
The above discussion of the SeibergWitten instability was given, as in [10], in the
Euclidean formulation. To repeat: a negative scalar curvature on the boundary gives rise
to a scalar field runaway which is the AdS/CFT counterpart of the unstable emission
of large branes in the bulk. When we turn to the dS/CFT correspondence ([3642];
see [43,44] for relevant recent discussions), we find that, since conformal infinity is now
spacelike, the CFT is automatically defined on a space of Euclidean signature, even if the
381
bulk is Lorentzian. We can therefore reasonably hope that the lessons we learned in the
Euclidean AdS case will apply here: a CFT on a boundary with scalar curvature which is
not positive or zero will suffer a runaway in this case too. (We should perhaps stress that
the dS/CFT correspondence is by no means supported by as much evidence as its AdS
counterpart. However, for our purposes, the precise details, and even the exact validity, of
the dS/CFT correspondence are not essential. All we require is that a runaway in a CFT
at infinity should be reflected in some kind of bulk instability. We certainly expect this to
hold true even if, as Susskind [45] has suggested, the dS/CFT correspondence can only be
approximately valid. In fact, anything else would signal a basic failure of the holography
principle in the de Sitter context.)
Now in fact non-singular asymptotically de Sitter spacetimes are well-behaved from
this point of view. For the main result of [46] states the following. Suppose that an
asymptotically de Sitter spacetime is future asymptotically simple, in the sense that every
future-directed inextendible null geodesic has an end point on future null infinity; this
essentially just means that there are no singularities. Then, if the dominant energy condition
(DEC) holds, the spacetime is globally hyperbolic with compact Cauchy surfaces having
finite fundamental groups. This result implies that future and past conformal infinity are
also compact with finite fundamental groups. (Notice that the DEC is essential here:
the spacetime given in [47] is asymptotically de Sitter and future asymptotically simple,
but conformal infinity has an infinite fundamental group. Of course, this spacetime does
not satisfy the DEC.) In the case of a four-dimensional asymptotically de Sitter space,
this compactness in turn means that (assuming orientability) we can use the Milnor
decomposition; and clearly there can be no K(, 1) factors in this case, since such
manifolds always have infinite fundamental groups. (This does not mean that the scalar
curvature must necessarily be positiverecall that the scalar curvature can be negative
even for a 3-spherebut it does mean that the manifold can be deformed so that the scalar
curvature becomes positive, that is, all these manifolds admit a metric of positive scalar
curvature [28,29].) Thus, there is no danger of SeibergWitten instability for non-singular
asymptotically de Sitter spacetimes.
Of course, this result does not apply to spacetimes that are asymptotically de Sitter but
not future asymptotically simple. To see why such spacetimes could be of considerable
interest, consider the spacelike branes introduced by Hull [48] and analysed in detail in
[49] (see also [5052]; see [53] for more recent references). An explicit four-dimensional
S-brane metric is given by [49]
d 2
+ 2 dz2 + R 2 dH22 ,
2
(9)
where dH22 denotes a locally hyperbolic metric (of constant curvature 1), and where
and R are simple functions of the time coordinate . (The hyperbolic term ensures the
preservation of the R-symmetries.) This metric is asymptotically flat in the remote past and
future. Its Penrose diagram [54,55] is obtained simply by means of a ninety degree rotation
of the Penrose diagram of a Schwarzschild black hole, with the understanding that each
point represents a locally hyperbolic space instead of a sphere. Thus the spacetime has nonspacelike future and past conformal boundaries and is nakedly singular. (The significance
of this last property is discussed in [54,55]. See also [56,57] for an interesting interpretation
382
of the singularities.) Thus we have an important example of a spacetime which is not future
asymptotically simple.
As our Universe is probably [58] asymptotically de Sitter rather than asymptotically
flat, it is very important to ask whether it is possible to modify the S-brane geometry so
that the conformal boundary becomes spacelike. If such asymptotically de Sitter S-brane
spacetimes exist, their Penrose diagrams will be square, with naked singularities to each
side as above, but with horizontal (spacelike) upper and lower boundaries. That is, their
Penrose diagrams will be obtained from a ninety degree rotation of the diagram of an AdS
black hole, just as the Penrose diagram of the above metric was obtained from a ninety
degree rotation of the diagram of a Schwarzschild black hole. Let us try to construct an
explicit example of a metric with such a Penrose diagram. (The intention here is not to
find an S-brane spacetime, but rather to give a concrete example of a metric with such
a Penrose diagram, so that the structure of future conformal infinity can be determined
for any spacetime having a diagram of this type. Following [54,55], we maintain the Rsymmetries by assuming that the horizons are hyperbolic.)
The Lorentzian four-dimensional anti-de Sitter black hole with a hyperbolic horizon has
metric

g(AdS4 ; hyper) = f4 (r) dt dt + f4 1 (r) dr dr + r 2 hij H 2 / dx i dx j
(10)
with
f4 = 1
r2
8G
+
;
r Vol(H 2 /) L2
(11)
note that the event horizon is now a compact Riemann surface, H 2 /, where is a
discrete infinite group acting freely and properly discontinuously on the hyperbolic plane,
H 2 . (The conformal boundary is then a three-dimensional manifold with topology given
by the product of the Riemann surface topology with a line.) The Penrose diagram, as
stated above, is a square with horizontal black and white hole singularities and with
vertical timelike infinities. Now simply rotate this diagram through 90 degrees. The new
diagram has spacelike, horizontal infinities and timelike, naked singularities to either side:
it resembles what in the West would be called a Chinese lantern. This is the desired nakedly
singular asymptotically de Sitter spacetime; the AdS event horizon has been transformed
to the dS cosmological horizon. Notice that, in Eq. (10), the spacelike and timelike roles
of r and t are exchanged inside the event horizon as usual. We can use this to deduce the
metric of the rotated spacetime as follows. If denotes the radius of the event horizon,
then set r = # and substitute this into Eq. (11). The result is, up to first order in #,
f4 =
8G#
2 Vol(H 2 /)
2#
.
L2
(12)
Now we want the geometry just inside the AdS event horizon to resemble the geometry
just outside the dS cosmological horizon. If the parameters are adjusted so that the latter
has radius , then we have r = + # in this case, and so we see that the desired metric
is obtained simply by reversing the signs of the last two terms in Eq. (11). That is, the
383
dS analogue of the AdS black hole with hyperbolic event horizon is the nakedly singular
spacetime with metric

g(dS4 ; hyper) = j (r) dt dt + j 1 (r) dr dr + r 2 hij H 2 / dx i dx j , (13)
where
j = 1 +
8G
r2
.
r Vol(H 2 /) L2
(14)
Of course we are not suggesting that an asymptotically de Sitter S-brane metric will have
exactly this form: no doubt there will be differences. The point, however, is that, in view
of the way we constructed it, we can expect the asymptotic behaviour of this spacetime to
mimic that of an asymptotically de Sitter S-brane spacetime. Since we propose to use the
dS/CFT correspondence, this is all we need.
Now in fact the metric (13) has already been discussed in the dS/CFT literature: it
was introduced in [59] as an example of a topological de Sitter spacetime. In fact, the
hyperbolic de Sitter spacetime was found to be particularly well-behaved, in that it has a
positive CardyVerlinde Casimir energy [6062]. Thus, the situation here is particularly
favourable from the dS/CFT point of view. Let us therefore investigate the structure of
conformal infinity. Outside the cosmological horizon, the coordinate r in Eq. (13) becomes
timelike, and so the geometry of conformal infinity is revealed by letting r tend to infinity.
In view of the way we derived the metric, it is no surprise to find that the canonical
representative of the conformal structure is almost precisely the same as in the case of
the Euclidean hyperbolic AdS black hole, that is, it is given by

3 = dt dt + L2 hij H 2 / dx i dx j .
(15)
Here 3 is a metric on a three-dimensional Euclidean manifold (since t has become
spacelike) with scalar curvature 2/L2 . The full conformal boundary consists of two
copies of this manifold, one in the infinite past, the other in the infinite future; this is
analogous to the fact that the conformal boundary of de Sitter space itself consists of two
3-spheres in the infinite past and future.
Now, examining 3 more closely, we notice two crucial features. The first is initially
surprising: this metric is geodesically complete, despite the fact that the naked singularities
are eternal and so are still present at infinity. The reason for this is that while the
spacetime has no event horizon, it does still have cosmological horizons which, as in any
asymptotically de Sitter spacetime, continuously contract around a given point. Thus, at
late times, the regions of future infinity which can be affected by the singularities becomes
steadily smaller, until, at infinity itself, they have shrunk to two points. These two points do
have to be excised, but the conformal freedom at infinity can be used to push the holes
off to an infinite distance. (One hole is at t = , the other at t = .) In fact, this
is an example of a classical construction due to Nomizu and Ozeki [63], whose theorem
states that, given any Riemannian metric on any manifold, there exists a conformally
related metric which is geodesically complete. The fact that this nakedly singular spacetime
has a conformal infinity which is complete is a beautiful physical implementation of the
NomizuOzeki construction: the strong de Sitter expansion has indeed pushed the holes
off to an infinite spatial distance (at temporal infinity).
384
Now from the dS/CFT point of view, this is just as we would wish. For the
correspondence, like its AdS/CFT counterpart, is supposed to relate a gravitational theory
in the bulk to a non-gravitational theory on the boundary. A non-gravitational theory
should be defined on a non-singular, geodesically complete space. The good behaviour
of the geometry at infinity confirms our belief that the dS/CFT correspondence can be
used here, in the sense that any misbehaviour of the boundary CFT cannot be ascribed to
singularities of the boundary metric. (Related observations were made in [64].) For this
reason, we shall henceforth only consider complete metrics on the boundary.
This discussion brings us to the second point: since, for reasons explained above,
t runs from to +, it is clear that the boundary is not compact. (Indeed, this is
why the question of completeness is an issue herea compact manifold is geodesically
complete with respect to any Riemannian (not Lorentzian) metric.) Before dealing with
the consequences of this, let us contrast the situation here with pure de Sitter space, with
its metric

1

r2
r2
g(dS4 ) = 1 2 dt dt + 1 2
(16)
dr dr + r 2 d22,
L
L
where d22 is the unit 2-sphere metric. Letting r tend to infinity as above, we obtain
dt dt + L2 d22 ,
(17)
which appears to be the metric on the non-compact cylinder R S 2 . Of course this is not
correct: the future boundary is a three-sphere. The point is that the polar coordinates do not
cover the origin or its antipode, and the excision of these two points from S 3 does indeed
produce a topological cylinder; the de Sitter expansion then implements a NomizuOzeki
completion, producing the metric (17). Evidently, the non-compactness here is a mere
coordinate effect. In the case of Eq. (15), however, the excisions are genuinely necessary,
so that future infinity is genuinely non-compact; it is a three-dimensional manifold with
topology R (R2 /). Thus we see that while the local geometry of conformal infinity
here is the same as that of the conformal infinity of a (Euclidean) hyperbolic AdS black
hole spacetime, the global structure is quite different. But the methods we used to establish
topological instability in the AdS case were based on the compactness of the boundary. The
failure of asymptotic simplicity in this case produces, as the AnderssonGalloway theorem
[46] would lead us to expect, a non-compact boundary. It follows that it is no longer clear
that the scalar curvature on the boundary cannot be positive or zero everywhere. We must
therefore reconsider the consequences of sign conditions on the scalar curvature of the
boundary, corresponding to singular asymptotically de Sitter metrics like the one given by
Eq. (13).
The scalar curvature of the boundary metric used here (as representative of the
conformal structure) is 2/L2 , so we can expect unstable behaviour for the CFT on the
boundary. What is the bulk counterpart of this instability? A comparison with the AdS
case suggests an answer: at extremely late times, beyond the horizon (in the uppermost
diamond of the Penrose diagram, where the metric is not static), we can expect spacelike
branes to be emitted (that is, to appear suddenly). We interpret the boundary runaway
as the dS/CFT dual of unstable production of these late branes, which are the dS/CFT
385
analogue of the AdS/CFT large branes. Of course, the question now is this: can backreaction bring the instability under control by modifying the geometry of the spatial
sections, so that future conformal infinity actually has positive or zero scalar curvature,
with late branes mediating the transition? As we saw, the methods used in Section 2 to
prove that the analogous AdS boundary cannot accept positive or zero scalar curvature do
not work here, so the answer requires new techniques.
Again, these techniques do exist, but they are somewhat more abstruse than in the
compact case, so we shall not attempt to summarize them; the reader may consult [23,
pp. 313326]. The main result we need may be stated as follows. Let M be any compact
enlargeable spin manifold, and let g be any metric on R M such that the manifold is
geodesically complete. (Recall that there is a physical motivation for requiring this last
condition.) Then the scalar curvature of g cannot be everywhere positive. Since R2 / is
enlargeable (it has a metric of non-positive sectional curvature), we see that there is indeed
no complete metric of positive scalar curvature on a three-dimensional manifold with the
topology that we have here, R (R2 /), despite the non-compactness. In fact, we have a
stronger result in this particular case. Suppose that the scalar curvature of a complete metric
on this manifold satisfies R 0 everywhere. Then a theorem of Kazdan [65] states that the
Ricci tensor, if it is not identically zero, can be used to deform the metric so that the scalar
curvature becomes everywhere strictly positive. Thus R 0 everywhere on R (R2 /)
implies that the Ricci tensor must vanish. But since R (R2 /) is three-dimensional, this
in turn means that the metric is flat. However, once again, this manifold has the wrong
topology to be flat. We conclude that there is no complete metric with scalar curvature
everywhere non-negative on this boundary manifold.
Once again, then, we are forced to conclude that no matter what effect late branes
have on the spacetime geometry, their unstable production cannot be halted: it is simply
impossible to deform a metric on R (R2 /) in a way that would achieve this. Again,
as in the previous section, processes which might change the topology of the spacelike
sections would lead either to causality violation or to new singularities, and we would not
normally interpret this as a sign that stability had been restoredjust the reverse. Clearly,
such spacetimes cannot occur either in dS/CFT or in whatever fundamental theoryone
hopes it is string or M-theoryunderlies dS/CFT. If anything resembling dS/CFT is valid,
then asymptotically de Sitter S-brane spacetimes preserving R-symmetries are unstable.
4. Conclusion
We have seen that SeibergWitten instability imposes a strong and unexpected
constraint on the geometry of the event horizon of an asymptotically AdS black hole
embedded in string theory. All known asymptotically AdS black holes are SeibergWitten
unstable in string theory, except those with horizons built up (by connected sums) from
S 1 S 2 and quotients of spheres and tori. In fact, we believe that the word known in this
statement can be dropped; this depends on the purely topological conjectures that (a) all
EilenbergMacLane spaces are enlargeable and (b) the Euclidean conformal boundary of a
black hole spacetime necessarily inherits enlargeability from the horizon. Counterexamples
to either of these conjectures, in the highly improbable event that any can be found, must
386
have an extremely complex topology and are most unlikely to be of physical interest. Thus
it seems that, for string theory on AdS backgrounds, event horizon geometries cannot be
much more complicated than in the asymptotically flat case. Surprisingly, we were able to
establish this even if we allowed the large branes in the bulk to act back on the spacetime
geometry. No matter how strong the back-reaction may be, the SeibergWitten criterion
continues to hold if it holds initially. Only when conditions become so extreme that the
topology of the spatial sections changes can there be any possibility of back-reaction
bringing the large branes under control; but, by then, new singularities (or, worse still,
causality violations) are generated, which presumably indicates that stability has been lost
in any case.
We have also investigated the possibility of generalizing the asymptotically flat S-brane
solutions of [49] to asymptotically de Sitter S-brane spacetimes. Again we find, assuming
that some kind of dS/CFT correspondence is valid, and assuming that the R-symmetries
are preserved, that such solutions are unstable in a way which is immune to back-reaction.
In [54,55], the consequences of including tachyonic matter in the effective dynamics
of S-branes was studied, while maintaining the R-symmetries as usual. The main
consequence is the replacement of the Milne horizon by a spacelike singularity
representing an S-brane. The spacetime otherwise evolves in an orderly way, though the
timelike singularities are not resolved. As one would expect on the basis of the results
obtained here, future conformal infinity remains non-spacelike. An asymptotically de Sitter
version of this spacetime would (because the timelike singularities are still eternal, and
the hyperbolic part of the metric is still present) have the same kind of future conformal
infinity as we discussed above, and we therefore predict that it would be unstable, with
or without back-reaction. In fact, we predict that physically reasonable de Sitter S-brane
spacetimes can only be obtained by explicitly breaking the R-symmetries. Unfortunately,
it is not known how to obtain explicit solutions in this case; the difficulties are discussed in
[54,55]. An alternative approach would be to try to generalize the S-brane solutions with
flat transverse spaces, mentioned in [57]. Unfortunately, the SeibergWitten criterion in the
case where the boundary scalar curvature is zero is not fully understood. Perhaps a higherderivative approach (see for example [66] for references) can elucidate the higher-order
terms in the large brane action.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
J.P.S. Lemos, Cylindrical black hole in general relativity, Phys. Lett. B 353 (1995) 46, gr-qc/9404041.
S. Aminneborg, I. Bengtsson, S. Holst, P. Peldan, Class. Quantum Grav. 13 (1996) 2707, gr-qc/9604005.
R.B. Mann, Black holes of negative mass, Class. Quantum Grav. 14 (1997) 2927, gr-qc/9705007.
D. Birmingham, Topological black holes in anti-de Sitter space, Class. Quantum Grav. 16 (1999) 1197,
hep-th/9808032.
J. Milnor, A unique decomposition theorem for 3-manifolds, Am. J. Math. 84 (1962) 1.
C. Cadeau, E. Woolgar, New five-dimensional black holes classified by horizon geometry, and a Bianchi VI
braneworld, Class. Quantum Grav. 18 (2001) 527, gr-qc/0011029.
E. Witten, AdS/CFT correspondence and topological field theory, JHEP 9812 (1998) 012, hep-th/9812012.
E. Witten, Connectedness of the boundary in the AdS/CFT correspondence, http://online.itp.ucsb.
edu/online/susy_c99/witten/.
387
[9] E. Witten, S.-T. Yau, Connectedness of the boundary in the AdS/CFT correspondence, Adv. Theor. Math.
Phys. 3 (1999) 1635, hep-th/9910245.
[10] N. Seiberg, E. Witten, The D1/D5 system and singular CFT, JHEP 9904 (1999) 017, hep-th/9903224.
[11] J.M. Lee, T.H. Parker, The Yamabe problem, Bull. Am. Math. Soc. 17 (1987) 37.
[12] Z.R. Jin, A Counterexample to the Yamabe Problem for Complete Noncompact Manifolds, in: Lecture Notes
in Mathematics, Vol. 1306, Springer, 1988.
[13] S. Surya, K. Schleich, D.M. Witt, Phase transitions for flat AdS black holes, Phys. Rev. Lett. 86 (2001) 5231,
hep-th/0101134.
[14] G.T. Horowitz, R.C. Myers, The AdS/CFT correspondence and a new positive energy conjecture for general
relativity, Phys. Rev. D 59 (1999) 026005, hep-th/9808079.
[15] G.J. Galloway, S. Surya, E. Woolgar, A uniqueness theorem for the AdS soliton, Phys. Rev. Lett. 88 (2002)
101102, hep-th/0108170.
[16] G.J. Galloway, S. Surya, E. Woolgar, On the geometry and mass of static, asymptotically AdS spacetimes,
and the uniqueness of the AdS soliton, hep-th/0204081.
[17] G.J. Galloway, S. Surya, E. Woolgar, Non-existence of black holes in certain < 0 spacetimes, grqc/0212079.
[18] D.N. Page, S. Surya, E. Woolgar, Positive mass from holographic causality, Phys. Rev. Lett. 89 (2002)
121301, hep-th/0204198.
[19] D.N. Page, Phase transitions for gauge theories on tori from the AdS/CFT correspondence, hep-th/0205001.
[20] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[21] M.T. Anderson, Boundary regularity, uniqueness and non-uniqueness for AH Einstein metrics on 4manifolds, math.DG/0104171.
[22] A.L. Besse, Einstein Manifolds, Springer, 1987.
[23] H.B. Lawson, M.-L. Michelsohn, Spin Geometry, Princeton Univ. Press, 1989.
[24] J.A. Wolf, Spaces of Constant Curvature, Publish or Perish, 1984.
[25] B. McInnes, Topologically induced instability in string theory, JHEP 0103 (2001) 031, hep-th/0101136.
[26] R. Lehoucq, J. Weeks, J.-P. Uzan, E. Gausmann, J.-P. Luminet, Eigenmodes of 3-dimensional spherical
spaces and their application to cosmology, Class. Quantum Grav. 19 (2002) 4683, gr-qc/0205009.
[27] G.I. Gomero, M.J. Reboucas, Detectability of cosmic topology in flat universes, gr-qc/0202094.
[28] R. Schoen, S.-T. Yau, On the structure of manifolds with positive scalar curvature, Manuscripta Math. 28
(1979) 159.
[29] D. Joyce, Constant scalar curvature metrics on connected sums, math.DG/0108022.
[30] P.S. Aspinwall, B.R. Greene, D.R. Morrison, Multiple mirror manifolds and topology change in string
theory, Phys. Lett. B 303 (1993) 249, hep-th/9301043.
[31] P.S. Aspinwall, B.R. Greene, D.R. Morrison, CalabiYau moduli space, mirror manifolds and spacetime
topology change in string theory, Nucl. Phys. B 416 (1994) 414, hep-th/9309097.
[32] B.R. Greene, K. Schalm, G. Shiu, Dynamical topology change in M-theory, J. Math. Phys. 42 (2001) 3171,
hep-th/0010207.
[33] R. Geroch, Topology in general relativity, J. Math. Phys. 8 (1967) 782.
[34] F. Tipler, Singularities and causality violation, Ann. Phys. 108 (1977) 1.
[35] G.T. Horowitz, Topology change classical and quantum gravity, Class. Quantum Grav. 8 (1991) 587.
[36] C.M. Hull, Timelike T-duality, de Sitter space, large N gauge theories and topological field theory,
JHEP 9807 (1998) 021, hep-th/9806146.
[37] M.-I. Park, Statistical entropy of three-dimensional Kerrde Sitter space, Phys. Lett. B 440 (1998) 275,
hep-th/9806119.
[38] M.-I. Park, Symmetry algebras in ChernSimons theories with boundary: canonical approach, Nucl. Phys.
B 544 (1999) 377, hep-th/9811033.
[39] E. Witten, Quantum gravity in de Sitter space, hep-th/0106109.
[40] A. Strominger, The dS/CFT correspondence, JHEP 0110 (2001) 034, hep-th/0106099.
[41] M. Spradlin, A. Strominger, A. Volovich, Les Houches lectures on de Sitter space, hep-th/0110007.
[42] A.J.M. Medved, A holographic interpretation of asymptotically de Sitter spacetimes, Class. Quantum
Grav. 19 (2002) 2883, hep-th/0112226.
[43] V. Balasubramanian, J. de Boer, D. Minic, Exploring de Sitter space and holography, hep-th/0207245.
[44] F. Leblond, D. Marolf, R.C. Myers, Tall tales from de Sitter space II: field theory dualities, hep-th/0211025.
388
[45] L. Susskind, Quantum gravity in dS-space?, http://www.pims.math.ca/science/2002/pnwstring/susskind/

susskind.html.
[46] L. Andersson, G.J. Galloway, dS/CFT and spacetime topology, hep-th/0202161.
[47] B. McInnes, The dS/CFT correspondence and the big smash, JHEP 0208 (2002) 029, hep-th/0112066.
[48] C.M. Hull, De Sitter space in supergravity and M-theory, JHEP 0111 (2001) 012, hep-th/0109213.
[49] M. Gutperle, A. Strominger, Spacelike branes, JHEP 0204 (2002) 018, hep-th/0202210.
[50] C.-M. Chen, D.V. Galtsov, M. Gutperle, S-brane solutions in supergravity theories, Phys. Rev. D 66 (2002)
024043, hep-th/0204071.
[51] M. Kruczenski, R.C. Myers, A.W. Peet, Supergravity S-branes, JHEP 0205 (2002) 039, hep-th/0204144.
[52] V.D. Ivashchuk, V.N. Melnikov, Exact solutions in multidimensional gravity with antisymmetric forms,
Class. Quantum Grav. 18 (2001) R1, hep-th/0110274.
[53] K. Hashimoto, P.-M. Ho, J.E. Wang, S-brane actions, hep-th/0211090.
[54] A. Buchel, P. Langfelder, J. Walcher, Does the tachyon matter?, Ann. Phys. 302 (2002) 78, hep-th/0207235.
[55] A. Buchel, J. Walcher, The tachyon does matter, hep-th/0212150.
[56] C.P. Burgess, F. Quevedo, S.-J. Rey, G. Tasinato, I. Zavala, Cosmological spacetimes from negative tension
brane backgrounds, JHEP 0210 (2002) 028, hep-th/0207104.
[57] F. Quevedo, G. Tasinato, I. Zavala, S-branes, negative tension branes and cosmology, hep-th/0211031.
[58] P.J.E. Peebles, B. Ratra, The cosmological constant and dark energy, astro-ph/0207347.
[59] R.-G. Cai, Y.S. Myung, Y.-Z. Zhang, Check of the mass bound conjecture in the de Sitter space, Phys. Rev.
D 65 (2002) 084019, hep-th/0110234.
[60] R.-G. Cai, CardyVerlinde formula and asymptotically de Sitter spaces, Phys. Lett. B 525 (2002) 331, hepth/0111093.
[61] R.-G. Cai, CardyVerlinde formula and thermodynamics of black holes in de Sitter spaces, Nucl. Phys.
B 628 (2002) 375, hep-th/0112253.
[62] A.J.M. Medved, dS-holographic C-functions with a topological, dilatonic twist, Phys. Rev. D 66 (2002)
064001, hep-th/0202193.
[63] K. Nomizu, H. Ozeki, The existence of complete Riemannian metrics, Proc. Am. Math. Soc. 12 (1961) 889.
[64] A.M. Ghezelbash, R.B. Mann, Action, mass and entropy of Schwarzschildde Sitter black holes and the de
Sitter/CFT correspondence, JHEP 0201 (2002) 005, hep-th/0111217.
[65] J.L. Kazdan, Deformation to positive scalar curvature on complete manifolds, Math. Ann. 261 (1982) 227.
[66] S. Nojiri, S.D. Odintsov, (Anti-)de Sitter black holes in higher derivative gravity and dual conformal field
theories, Phys. Rev. D 66 (2002) 044012, hep-th/0204112.

Inflationary cosmology from STM theory of gravity

Mauricio Bellini
Consejo Nacional de Investigaciones Cientficas y Tcnicas (CONICET), Departamento de Fsica,
Facultad de Ciencias Exactas y Naturales, Universidad Nacional de Mar del Plata,
Funes 3350, (7600) Mar del Plata, Buenos Aires, Argentina
Received 11 June 2002; received in revised form 22 January 2003; accepted 14 March 2003
Abstract
I study the power-law and de Sitter expansions for the universe during inflation from the STM
theory of gravity. In a de Sitter expansion the additional dimension is related to the cosmological
constant = 3/ 2 . I find from experimental data that the mass of the inflaton field is m2 = 2/(3 2 ).
The interesting in this case is that the inflaton field fluctuations are related to the fifth coordinate. In
power-law expansion, the fifth coordinate () appears to be a dimensionless constant. Here, the
-value depends on the initial conditions. I find the 5D line element for this inflationary expansion,
which is a function of the classical component of the inflaton. But the more important result here
obtained is that in both cases there is not dynamical compactification during inflation.
PACS: 98.80.Cq; 04.20.Jb; 11.10.Kk
1. Introduction
The extra dimension is already known to be of potential importance for cosmology.
There is a class of five-dimensional (5D) cosmological models which reduce to the
usual four-dimensional ones, on hypersurfaces defined by setting the value of the extra
coordinate equal to a constant [1]. In these models, physical properties such as the
mean energy density and pressure of matter are well defined consequences of how the
extra coordinate enters the metric [2]. That is, matter is explained as the consequence
of geometry in five dimensions. The physics of this follows from a mathematical result,
which is that Einsteins equations of general relativity with matter are a subset of the
KaluzaKlein (KK) equations in apparent vacuum. Hence, cosmology is a subject of pure
E-mail address: mbellini@mdp.edu.ar (M. Bellini).
doi:10.1016/S0550-3213(03)00234-7
390
M. Bellini / Nuclear Physics B 660 (2003) 389400
geometry in five-dimensional KaluzaKlein gravity known as spacetimematter (STM)

theory. This theory is supported by more local astrophysics [3]. There is a canonical class
of five-dimensional metrics, most often discussed in the form due to Gross with Perry [4]
and Davidson with Owen [5]. During the last years there were many attempts to construct
a consistent brane cosmology [6]. However, after a detailed investigation of this possibility
a series of no-go theorems have been proven [7]. There is a main difference between STM
and brane-world (BW) [8] formalisms. In the STM theory of gravity we don not need to
insert any matter into the 5D manifold by hand, as is commonly done in the BW formalism.
In the STM theory of gravity the 5D metric is an exact solution of the 5D field equations
in apparent vacuum [11]. The interesting here, is that matter appears in four dimensions
without any dimensional compactification, but induced by the 5D vacuum conditions.
On the other hand, inflationary cosmology is one of the most reliable concepts in modern
cosmology. The first model of inflation was proposed by A. Starobinsky in 1979 [12].
A much simpler inflationary model with a clear motivation was developed by Guth in
the 80s [13], in order to solve some of the shortcomings of the big bang theory, and in
particular, to explain the extraordinary homogeneity of the observable universe. However,
the universe after inflation in this scenario becomes very inhomogeneous. Following a
detailed investigation of this problem, A. Guth and E. Weinberg concluded that the old
inflationry model could not be improved [14].
These problems were sorted out by A. Linde in 1983 with the introduction of chaotic
inflation [15]. In this scenario inflation can occur in theories with potentials such as
V () n . It may begin in the absence of thermal equilibrium in the early universe, and it
may start even at the Planckian density, in which case the problem of initial conditions
for inflation can be easily solved [16]. According to the simplest versions of chaotic
inflationary theory, the universe is not a single expanding ball of fire produced in the big
bang, but rather a huge eternally growing fractal. It consists of many inflating balls that
produce new balls, which produce more new balls, ad infinitum.
BW cosmology has been studied very recently within [9] and without inflation [10].
However, inflationary cosmology from the STM theory, still remains without a detailed
study. The aim of this paper is the study of inflation from the STM theory of gravity. More
exactly, I am interested in the study of de Sitter and power-law expansions for the universe
during inflation from the STM theory developed by P. Wesson [17]. This paper is organized
as follows. In Section 2, I study the STM basic equations. In Section 3, I develope the
classical and quantum field dynamics during the inflationary stage using the semiclassical
approach developed in [18]. Furthermore, de Sitter and power-law expansions for the
universe are examined from the STM theory of gravity and the results are contrasted with
experimental data. Finally, in Section 4 some final comments are developed.
2. Basic STM equations

Following the idea suggested by Wesson and co-workers [11], in this section I develope
the induced 4D equation of state from the 5D vacuum field equations, GAB = 0 (A, B =
0, 1, 2, 3, 4), which give the 4D Einstein equations G = 8GT (, = 0, 1, 2, 3). In
391
particular, we consider a 5D spatially isotropic and flat spherically-symmetric line element

dS 2 = e(,t ) dt 2 + e(,t ) dr 2 + e (,t ) d 2 ,
(1)
where dr 2 = dx 2 + dy 2 + dz2 and is the fifth coordinate. We assume that e , e and

e are separable functions of the variables and t. The equations for the relevant Einstein
tensor elements are
2

3
3
3 2 3
3
+
+e
+
,
G 0 = e
4
4
2
2
4

3 3 3
0
3
+
G 4=e
,
2
4
4
4

2

3 2
+ +
+
Gi i = e +
4
2
4
2
2
4
2

2

3
+
,
+ + +
+e
4
2
4
2
2
4

2
3 2 3
3
4
3
3
G 4=e
+e
,
+
+
2
2
4
4
4
0
(2)
(3)
(4)
(5)
where the overstar and the overdot denote, respectively, / and /t , and i = 1, 2, 3.
Following the convention (, +, +, +) for the 4D metric, we define T 0 0 = t and
T 1 1 = p, where t is the total energy density and p is the pressure. The 5D-vacuum
conditions (GA
B = 0) are given by [17]
3
8Gt = e 2 ,
4

3 2

8Gp = e
,

2
4

2

3 2
3
3

+
+
e
=e
.
4
4
2
2
4
(6)
(7)
(8)
Hence, from Eqs. (6) and (7) and taking = 0, we obtain the equation of state for the
induced matter

4
+
1
t .
p=
(9)
3 2
2 0 and |/
2 | 1 (or zero), this equation describes an inflationary
Notice that for /
2
universe. The equality / = 0 corresponds with a 4D de Sitter expansion for the

universe.
In this paper I explore the possibility to obtain inflation for the metrics (1) with the
restrictions
(),
(, t),
(t),
(10)
392

where e is a separable function of and t. The conditions (10) imply that = = 0.

Furthermore, in that follows we restrict our study to inflationary models on 5D metrics
where all the coordinates are independent. The choice (10) implies that only the spatial
sphere and the fifth coordinate have respectively squared sizes e and e that evolve with
time.
3. Inflationary universe
The dynamics of a scalar field minimally coupled to gravity is described by the
Lagrangian density

1
R
+ g , , + V () ,
L(, , ) = g
(11)
16 2
where R is the 4D scalar curvature and g is the determinant of the 4D metric tensor. If the
spacetime has a FriedmanRobertsonWalker (FRW) metric, ds 2 = d 2 + a 2 ( ) dr 2 , the
resulting equations of motion for the field operator and the Hubble parameter, are
1 2
+ 3H + V () = 0,
a2

4
1 2
2
2
H =
+ 2 + 2V () ,
3Mp2
a
(12)
(13)
where, in the following the overdot represents the derivative with respect to and
V () dV /d. The expectation value is assumed to be a constant function of the spatial
variables for consistency with the FRW metric. The discussion of the inflationary stage
is usually done through the classical analysis of the above equations, which in the case
of a homogeneous scalar field and null spatial curvature reduce to a system of two first
order equations, even without any slow-roll assumption [19]. The period in which a > 0
and thus the universe has an accelerated expansion is the inflationary stage, and models
are discarded or not depending on if they provide enough inflation or not. When the
inflation ends the field starts oscillating rapidly and its potential energy is converted into
thermal energy. This is the general scheme of the inflationary scenario without considering
the quantum effects. In the usual formulation of this approach the slow-roll regime is
assumed. Instead, we avoid here the use of such an assumption and consider a consistent
semiclassical expansion of the theory. To obtain this we decompose the inflaton operator
in a classical field c plus a quantum correction , = c + , such that = c and
= 0. The field c is defined as the solution to the classical equation of motion
=
c + 3H c + V (c ) = 0,
(14)
where we have assumed that the classical field is spatially homogeneous, in agreement
with the hypothesis of an inflationary regime. The evolution of the quantum operator is
given by
1
1
V (n+1) (c ) n = 0.
2 2 + 3H +
(15)
a
n!
n
At the same time the squared Hubble parameter can be expanded as

4
1 2 1 (n+1)
2
2
2
n
H = Hc +
V
(c ) ,
+ 2 +
3Mp2
a
n!
n
393
(16)
where
Hc2 =

4
2
c + 2V (c ) ,
2
3Mp
(17)
is the classical Hubble parameter. If the quantum fluctuations are small, hence the inflation
is driven by the classical field. The effect of the classical field c is to change the
environment in which the field evolves, and in particular the mass of the particles.
in Eq. (15),
Making a first order expansion in Eq. (16) for , we obtain H Hc = a/a
which reduces to
1
2 2 + 3Hc + V (c ) = 0.
(18)
a
From Eqs. (14) and (17), we can describe the classical dynamics of the Hubble parameter
and the inflaton field by the relations
c =
Mp2
4
Hc ,
Mp2 2
Hc .
4
In these terms the potential accounting for such a dynamics is given by:

3Mp2
Mp2 2
Hc
V (c ) =
.
Hc2
8
12
H c = Hc c =
(19)
(20)
(21)
Eqs. (19) and (20) define the classical evolution of spacetime, determining the relation
between the classical potential and the inflationary regimes. On the other hand Eq. (8)
defines the quantum dynamics of the field , whose classical character was studied
in [18,20]. In this framework, the total energy density and the pressure (neglecting the
terms 2 /2, because they are very small on cosmological scales), are given by
c2
+ V (c ),
2
2
p = c V (c ),
2
such that the equation of state for supercooled inflation is

2Hc
p =
+
1
t .
3Hc2
t =
(22)
(23)
(24)
From Eqs. (9) and (24) one obtains the Hubble parameter as a function of for inflationary

models with = = 0
Hc = /2.
(25)
394
Furthermore, from Eqs. (25) and (21) we can write the scalar potential for inflationary
models that follows from the 5D metric (1)

3Mp2 2 2
+ .
V c (t) =
(26)
2
3
Eq. (25) shows that models with Hc = const. (like a de Sitter one) are described with
potentials
3Mp2 2
V c (t) =
.
2
(27)
3
To map Eq. (8) in a wave equation, we can make the transformation = e 2

3
a2

k02
1 2
= 0,
a2
a2
d H
(28)
where k02 = a 2 ( 94 Hc2 + 32 H c V c ). Now, we have a scalar field with a time-dependent

mass. It can be described in terms of a set of modes

1

d 3 k ak ei k.r k (t) + h.c. ,
=
(29)
3/2
(2)
where we have made explicit use of the spatial homogeneity of the FRW metric. The
operators ak and ak satisfy the canonical commutators:

ak , ak = k k ,

[ak , ak ] = ak , ak = 0,
(30)
(31)
and the equation of motion for the modes k (t) are given by
k + k2 k = 0,
(32)
with k2 = a12 (k 2 k02 ). Here, k0 ( ) separates the large scale (unstable) sector (k 2 < k02 )
and the short scale (stable) one (k 2 > k02 ). When k0 surpasses k, the temporal oscillation
of the mode ceases. During the expansion of the universe, more and more new fluctuations
suffer this transformation. These quantum fluctuations with wave number below k0 are
generally interpreted as inhomogeneities superimposed to the classical field c . They are
responsible for the density inhomogeneities generated during the inflation.
In order that and satisfy a canonical commutation relation, [(r , t), (r , t)] =
i(r r ), the normalization of the modes k (t) must be chosen
k k k k = i,
(33)
where the asterisk denotes the complex conjugate. In the following subsections I will study
two particular cases of inflation from the STM theory of gravity; de Sitter and power-law
expansions.
395
3.1. de Sitter expansion

2 = 0, = = 0
The special case e = 2 in Eq. (26) (see Eqs. (25) and (27)), with /
in Eq. (1), gives a de Sitter expansion for which t + p = 0, so that c = 0 , V (c ) = V0
and Hc ( ) = H0 are constant [18]. It implies that
H02 =
8
V0 .
3Mp2
(34)
This case corresponds exactly to a scalar field with a mass m2 =
d 2V
d 2 =0
background with a constant Hubble parameter H0 and a scale factor

model the cosmological constant gives the vacuum energy density
,
8G
such that is related with the fifth coordinate by means of
in a de Sitter
a(t) eH0 .
For this
t =
(35)
= 3/ 2 .
(36)
The 5D metric for a de Sitter expansion of the universe is [1]
2
dS 2 = d 2 + 2 e2 /(3 ) dr 2 + d 2 .
(37)
Hence, the 4D de Sitter inflationary universe can be seen as a 5D metric with the fifth
constant coordinate and unitary size. Eq. (35) shows that the vacuum on the metric (37)
induces the cosmological constant , which depends on the value of the fifth coordinate
and generates the expansion of the universe.
On the other hand, the wave number k0 is

9 m2
.
k0 = H0 a( )
(38)
4 H02
The equation of motion for the modes in a de Sitter expansion can be written explicitely

k ( ) + k 2 e2H0 9 H02 + m2 k ( ) = 0,
(39)
4
and its general solution is

k
k
k ( ) = A1 H(1)
+ A2 H(2)
,
H0 a
H0 a
(40)

(1)
(2)
where H and H are the Hankel functions and = 9/4 m2 /H02 < 3/2. The long
time behavior of the mean square of reproduces the exact value as calculated by standard
quantum field methods [22]. For a massive inflaton (i.e., with m/H0 1), once we taken
into account the normalization condition for the modes (33), we have

a (32)( ) 2
3 ()H021
SH
232 3
k0
dk 32
k
,
k
(41)
396
where 3() is the gamma function with -argument. Hence, the scale invariant spectrum
ns = 1 (i.e., = 1/2) implies that m2 = 2H02 = 2/(3 2 ), and

3 2 (1/2) 2
,
2 SH
(42)
24 3
where 3 is the gamma function. Hence, the fifth coordinate also should be related to the
inflaton fluctuations for a scale invariant power spectrum.

3.2. Power-law inflation

To study a power-law expansion of the universe, we can make e = s , e = a 2 (t) r
and e = At n , in Eq. (1). Hence, the 5D-line element can be written as
dS 2 = s dt 2 + a 2 (t) r dr 2 + At n d 2 ,
(43)
where the scale factor is a t p . From the 5D-vacuum conditions (6)(8), we find
s = n = 2,
r=
2p
,
p1
A = (p 1)2 ,
(44)
such that we obtain the Ponce de Leon metric [1]

2p
dS 2 = 2 dt 2 + a 2 (t) p1 dr 2 + (p 1)2 t 2 d 2 .
(45)
This metric is an exact solution of the 5D field equations in apparent vacuum, which in
terms of the Ricci tensor are RAB = 0. The induced metric h on these hypersurfaces is
given by
ds 2 = h dx dx = d 2 + a 2 ( ) dr 2,
(46)
where = t is the cosmic time which corresponds to the spatially flat FRW metric. With
this representation, the scale factor is
a( ) =
p(2p)
p1
p.
(47)
I am interested in the study of power-law inflationary dynamics from the metric (45). In
this case p > 1, and the equation of state if matter is interpreted as a perfect fluid, is

1 + 6p
p =
(48)
t .
6p
From Eqs. (20) and (26), one obtains

Hc
2
2
.
Mp2
(49)
Replacing in Eq. (21) the scalar potential can be obtained as a function of (, t).
Furthermore the constant p is related with the properties of matter

2
t
p=
(50)
,
3 t + p
and the temporal evolution for c ( ) can be obtained from Eq. (19)

Mp p
H0
ln
.
c ( ) =
2
p
Hence, due to = t, we obtain the value of the constant
397
(51)
p M2p p 0
=
(52)
e
,
H0 t0
where (H0, t0 , 0 ) are respectively the values of (Hc , t, c ) when inflation starts. This
implies that should be strongly dependent of initial conditions. Note that p is related
to the properties of matter such as the energy density t and pressure p. In the inducedmatter interpretation of KK theory, these are the result of the geometry in five dimensions,
insofar as they are functions of the extra part of the metric and partial derivatives of the
metric coefficients with respect to the extra coordinate. The field equations of the theory in
terms of the Ricci tensor are RAB = 0. These appear to describe empty five-dimensionnal
space. But they in fact contain the Einstein equations G = 8GT .
Finally, taking into account the results above, the Ponce de Leon metric (45) can be
written as
2p

p
2p
p
4
( )
e Mp c dr 2
dS 2 = d 2 + p1
H0
2
2 1

1
p
4 p Mp c ( )
+
e
d 2 ,
(53)
p1
H0
where (p, c , ) are given respectively by Eqs. (50)(52). Notice the c ( )-dependence of
the metric (53), which make the difference between power-law inflationary scenarios and
other cosmological models with p < 1, studied in [1]. From Eq. (53) can be view that the
size of the fifth coordinate grows with , but the 5D vacuum here requiered implies that
= const. Hence, for a power-law inflationary expansion, the 4D universe that expands
from a metastable 4D vacuum with a scalar potential given by Eq. (26), is realy a 5D
universe in a true vacuum with a metric given by Eq. (53). The differential equation (32),
has the solution

(1) k 1p H0
(2) k 1p H0
+
A
,
H
H
k ( ) = A1
(54)
2
p
0
0 (p 1)0p
(p 1)0

(1)
(2)
9
where H and H are the Hankel functions and = 94 p2 15
2 p + 4 (p 1). From
the normalization condition (33), we obtain the adiabatic vacuum

k 1p H0
H(2)
k ( ) =
(55)
p .
2 (p 1)0
(p 1)0

(x/2)
The Hankel functions take the small-argument limit H(2,1)[x]x 1 3(1+)
i 3()(x/
2) . Hence, the super-Hubble squared fluctuations 2 |SH are

41 3 2 ()(p 1)21 p(32)2+1
SH
p(32)+1
H03+2
0
k0
dk 32
k
,
k
(56)
398

p(2p)
1/2 1p and 3() is the gamma function with where k0 ( ) = p1 [ 94 p2 15

2 p + 1]
k (t )
argument. The integral controlling the presence of infrared divergences is 0 0 dk k 2(1) ,
which has a power spectrum P()2 k 32 . We can evaluate the expression (56), and
we obtain
(32)
2

(p 1)21
41 3 2 ()[ 94 p2 15
2 p + 1]
SH
1p(32)
H03+2 0
(3 2)
p(2p)(32)
p1
12 .
(57)
The spectral index ns = 3/2 1 agrees with the experimental data [21] for 1/2
(i.e., for p 3.08). For this value of , we obtain a time independent 2 |SH , so that the
SH metric fluctuations in (53) remains constant for a scale invariant power spectrum.
4. Final comments
In this paper we studied the evolution of the early universe from the STM theory
of gravity. In this framework inflation should be a consequence of the expansion
generated by a cosmological constant (variable in power-law inflation and constant in a
de Sitter expansion) induced by a generalized 5D spatially flat, isotropic an homogeneous
background metric where the fifth dimension becomes constant as a consequence of the
5D vacuum state. The condition to obtain a inflationary expansion is

4
p =
+
1
t t ,
3 2
or respectively
1,
2
= 0,
2
for power-law and de Sitter expansions where = 2Hc .

In the models here studied all of the components of the RiemannChristoffel tensor for
the 5D metric are zero. Despite this, the models 4D part is not flat, since the 4D Ricci
scalar is non-zero. Hence, we see that while the inflationary universe may be curved in
4D, it is flat in 5D. Thus, the 5D STM theory of gravity describes an inflationary universe
consisting of localized singular sources embedded in a globally 5D flat cosmology.
We have restricted our study to 4D spatially flat FRW metrics, which are relevant to
inflationary cosmology. We studied the power-law and de Sitter models of inflation. We
find that (a) for a de Sitter expansion the additional dimension isrelated to the cosmological
constant = 3/ 2 and the Hubble parameter is H0 = 1/ 3 2 . Hence, we inquire
that has G1/2 unities. From the experimental data obtained from BOOMERANG-98,
MAXIMA-1 and COBE DMR [21], we obtain that m2 2 = 2/3. This implies that, in
a de Sitter expansion with a 5D vacuum state, gives nearly the inverse of the mass
of the inflaton field. Hence, for m2 (108 1012)Mp2 , the resulting fifth coordinate
will be of the order of 2 (108 1012)Mp2 . Note that the inflaton field fluctuations
are also related to the fifth coordinate (see Eq. (42)). Hence, in the induce-matter
399
interpretation of KK theory, matter field fluctuations also appear as a consequence of the

five-dimensional geometry. (b) For the power-law expansion, the fifth coordinate appears
to be a dimensionless constant, which, in principle, could be considered as 1 (notice that
we are considering c = h = 1). The interesting here is that the -value depends on the
initial conditions during power-law inflation (see Eq. (52)). Furthermore, the main result
of this paper is the metric (53) which describes the 5D line element in power-law inflation.
Notice the dependence with the spatially isotropic component of the inflaton field, c .
Furthermore, it is very important the fact that the squared size of the fifth coordinate e
grows with time in a power-law inflationary universe.
Acknowledgements
I would like to acknowledge CONICET (Argentina) and Universidad Nacional de Mar
del Plata for financial support in the form of a research grant.
References
[1] J. Ponce de Leon, Gen. Relativ. Gravit. 20 (1988) 539.
[2] P.S. Wesson, Phys. Lett. B 276 (1992) 299.
[3] P.S. Wesson, Astrophys. J. 394 (1992) 19;
P.S. Wesson, Astrophys. J. 436 (1994) 547;
P.S. Wesson, Astrophys. J. 440 (1995) 1.
[4] D.J. Gross, M.J. Perry, Nucl. Phys. B 226 (1983) 29.
[5] A. Davidson, D. Owen, Phys. Lett. B 155 (1985) 247.
[6] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
I. Antoniadis, N.A. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690.
[7] R. Kallosh, A. Linde, JHEP 0002 (2000) 005;
K. Behrndt, M. Cvetic, Phys. Rev. D 61 (2000) 101901;
G.W. Gibbons, N.D. Lambert, Phys. Lett. B 488 (2000) 90;
J. Maldacena, C. Nunez, Int. J. Mod. Phys. A 16 (2001) 822.
[8] S.S. Seahra, P.S. Wesson, Class. Quantum Grav. 19 (2002) 1139.
[9] S. Nojiri, S.D. Odintsov, JHEP 0112 (2001) 033.
[10] R. Mansouri, F. Nasseri, Phys. Rev. D 60 (1999) 123512.
[11] P. Wesson, Gen. Relativ. Gravit. 22 (1990) 707;
P. Wesson, Astrophys. J. 394 (1992) 19;
P. Wesson, H. Liu, P. Lim, Phys. Lett. B 298 (1993) 69.
[12] A.A. Starobinsky, JETP Lett. 30 (1979) 682;
A.A. Starobinsky, Phys. Lett. B 91 (1980) 99.
[13] A. Guth, Phys. Rev. D 23 (1981) 347.
[14] A. Guth, E.J. Weinberg, Nucl. Phys. B 212 (1983) 321.
[15] A.D. Linde, Phys. Lett. 129 (1983) 177.
[16] For a review of inflation the reader can see, A.D. Linde, Particle Physics and Inflationary Cosmology,
Harwood Academic, Chur, 1990;
A.D. Linde, Phys. Rep. 333 (2000) 575.
[17] For a review of STM theory the reader can see, J.M. Overduin, P.S. Wesson, Phys. Rep. 283 (1997) 303.
[18] M. Bellini, H. Casini, R. Montemayor, P. Sisterna, Phys. Rev. D 54 (1996) 7172.
400
[19]
[20]
[21]
[22]
E.J. Copeland, E.W. Colb, A.R. Liddle, J.E. Lidsay, Phys. Rev. D 48 (1993) 2529.
H. Casini, R. Montemayor, P. Sisterna, Phys. Rev. D 59 (1999) 063512.
A.H. Jaffe, et al., Phys. Rev. Lett. 86 (2001) 3475.
S. Habib, Phys. Rev. D 46 (1992) 2408.

Erratum
Erratum to: Massless higher spins and holography

[Nucl. Phys. B 644 (2002) 303]
E. Sezgin a , P. Sundell b
a Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
b Department for Theoretical Physics, Uppsala Universitet, Sweden
Received 25 March 2003
1. In Eqs. (4.26)(4.28) the second equal signs should be replaced by plus signs.
2. In Eq. (4.30) the factors of 1/2 should be removed.
3. The two paragraphs above Eq. (5.12), the first one of which starts with The above
identification . . . should be replaced by the following:
The connection between the leading Regge trajectory at small ls and the bilinear HS
currents in the SYM theory at small has also been made by the authors of [34] who
give long string solutions to the worldsheet sigma model in the limit ls /R 1. These
solutions describe states on the leading Regge trajectory with spins s R 2 / ls2 1 and
AdS energies E s + R 2 / ls2 log(sls2 /R 2 ) s, which couple to bilinear HS currents in the
SYM theory. These operators arise in the OPE of Wilson lines making up the boundaries
of worldsheets of infinitely long strings on the leading Regge trajectory. The leading
2 2
Regge
trajectory also contains short strings which have spins s R / ls and energies
E s R/ ls . Since E/s 1 for long strings, and E/s 1 for short strings, the long ones
cannot decay into large number of short ones. Furthermore, in any Regge trajectory there
should be long string with large spins and asymptotically small anomalous dimensions,
suggesting that string interactions in the limit s R 2 / ls2 1 have a consistent description
in terms of long strings with E s. Upon increasing ls /R, we expect a short string state
with fixed s to become long for large enough ls /R. In the limit in which 0 on the
SYM side, there should exist a hs(2, 2|4) invariant worldsheet sigma model describing
closed string interactions in the bulk corresponding to the free SYM theory.
doi of original article:10.1016/S0550-3213(02)00739-3.

E-mail addresses: sezgin@physics.tamu.edu (E. Sezgin), p.sundell@teorfys.uu.se (P. Sundell).
doi:10.1016/S0550-3213(03)00267-0
Nuclear Physics B 660 [FS] (2003) 407435

Schrdinger invariance and spacetime symmetries

Malte Henkel a , Jrmie Unterberger b
a Laboratoire de Physique des Matriaux, 1 Universit Henri Poincar Nancy I,
B.P. 239, F-54506 Vanduvre ls Nancy Cedex, France

b Institut Elie Cartan, 2 Dpartement de Mathmatiques, Universit Henri Poincar Nancy I,
B.P. 239, F-54506 Vanduvre ls Nancy Cedex, France

Abstract
The free Schrdinger equation with mass M can be turned into a non-massive KleinGordon
equation via Fourier transformation with respect to M. The kinematic symmetry algebra schd
of the free d-dimensional Schrdinger equation with M fixed appears therefore naturally as a
parabolic subalgebra of the complexified conformal algebra confd+2 in d + 2 dimensions. The
explicit classification of the parabolic subalgebras of conf3 yields physically interesting dynamic
symmetry algebras. This allows us to propose a new dynamic symmetry group relevant for the
description of ageing far from thermal equilibrium, with a dynamical exponent z = 2. The Ward
identities resulting from the invariance under confd+2 and its parabolic subalgebras are derived and
the corresponding free-field energymomentum tensor is constructed. We also derive the scaling
form and the causality conditions for the two- and three-point functions and their relationship with
response functions in the context of MartinSiggiaRose theory.
Keywords: Schrdinger invariance; Conformal invariance; Ward identity; Energymomentum tensor; Parabolic
subalgebra; Response function; Ageing
1. Introduction
The Schrdinger group Sch(d) is defined as the following set of spacetime transformations in d space dimensions
1 Laboratoire associ au CNRS UMR 7556.

2 Laboratoire associ au CNRS UMR 7502.
doi:10.1016/S0550-3213(03)00252-9
408
M. Henkel, J. Unterberger / Nuclear Physics B 660 [FS] (2003) 407435
t t =
t +
,
t +
r r =
Rr + vt + a
,
t +
= 1,
(1.1)
where R is a rotation matrix. Consider the free Schrdinger equation

2
2
(1.2)
2 = 2M
2 = 0,
t
t
r
r
where the mass m is a constant. In 1972, Niederer [44] showed that the maximal kinematic
invariance group of (1.2) is the Schrdinger group Sch(d), while Lie had already shown in
1882 that the free diffusion equation is invariant under Sch(d) [41]. The action of Sch(d)
on the space of solutions of (1.2) is projective, that is, the wave function = (t, r)
transforms into

(t, r) (Tg )(t, r) = fg g 1 (t, r) g 1 (t, r)
(1.3)
2mi
(the companion function fg is explicitly known [44]). Independently, Hagen [21] showed
that the free-field action from which (1.2) can be derived as equation of motion is
Schrdinger-invariant. In this paper, we shall mainly consider the Lie algebra schd of the
Schrdinger group Sch(d). In particular, one has the following set of generators for sch1 :
X1 = t ,
Y1/2 = r ,
Y1/2 = tr Mr,
1
x
X0 = tt rr ,
2
2
M
X1 = t 2 t tr r
r 2 2xt,
2
M0 = M,
time and space translations,

Galilei transformation,
dilatation,
special Schrdinger transformation,
phase shift.
(1.4)
Here, M = im and x is the scaling dimension of the wave function on which the
generators of sch1 act. For a solution of the free Schrdinger equation (1.2) one has
x = d/2. The non-vanishing commutators of sch1 are

n
m Yn+m ,
[Xn , Xn ] = n n Xn+n ,
[Xn , Ym ] =
2
[Y1/2 , Y1/2 ] = M0
(1.5)
(n, n {1, 0}, m {1/2}). The Schrdinger group has been introduced [21,44] as
a non-relativistic analogue of the conformal group in d dimensions (whose Lie algebra
will be denoted here confd ). Indeed, it was argued by Barut [1] that the Schrdinger Lie
algebra schd could be obtained from the conformal algebra confd+1 . . . by a combined
process of group contraction and a transfer of the transformation of mass to the coordinates. The projective unitary irreducible representations of the Schrdinger group
Sch(d) are classified in [46]. In particular, Sch(1) is isomorphic to the semi-direct product
Sl(2, R) H (1), where Sl(2, R) comes from the exponentiation of the X-generators, and
the 1-dimensional Heisenberg group H (1) from the exponentiation of Y1/2 and M0 .
Schrdinger invariance has been considered in a wide variety of situations, for example,
celestial mechanics [13], non-relativistic field theory [3,5,12,24,37,38,43] and/or nonrelativistic quantum mechanics [15,34,36], hydrodynamics [23,35,39,45] or dynamical
409
scaling [2527], see also references therein. Our interest in this dynamical symmetry comes
from the consideration of situations of dynamical scaling such that the correlators of field
operators i transform covariantly under dilatations (with b constant)

z

1 b t1 , br 1 n b z tn , br n = b x1 xn 1 (t1 , r 1 ) n (tn , r n ) ,
(1.6)
where z is the dynamical exponent. Such a dynamic scaling behaviour occurs in many
physical situations, for example, critical dynamics or else in the phase-ordering kinetics
undergone by a spin system quenched from a disordered initial state to a temperature
T < Tc , where Tc > 0 is the critical temperature (see, e.g., [4,9] for reviews). Eq. (1.6) is
compatible with Galilei invariance only if z = 2. By analogy with conformal invariance [2],
one may ask whether a generalization of (1.6) to a local scale invariance with spacetimedependent rescaling factors b = b(t, r) is possible. Indeed, it has been shown recently
that infinitesimal generators of local scale transformations with any given value of z can
be constructed [31]. In turn, admitting local scale invariance as a hypothesis of dynamics
leads to explicit expressions for two-point functions which can be tested in specific models.
These phenomenological predictions have so far been confirmed at the Lifshitz points
of spin systems with competing interactions [28,49] and in the non-equilibrium ageing
behaviour of ferromagnetic spin systems [6,8,1820,30,32,33,47].
These experimental confirmations provide some credibility to the idea of local scale
invariance. However, an understanding of the origin of local scale invariance is still
lacking. For example, in the present tests of local scale invariance the values of the scaling
dimensions xi are still considered as free parameters. An approach similar to the one of 2D
conformal invariance as initiated in [2] and which allows, among other things, to obtain
the xi from the representation theory of the Virasoro algebra, to find all n-point functions
and furthermore to list the universality classes, is not available (see, e.g., [11,29] for
introductions to conformal invariance). As a first step in this direction, we shall undertake
here a case study of the simplest case, namely Schrdinger invariance, which is realized
for z = 2 [26].
In conformal field theory, the central object is the energymomentum tensor and the
main tool the conformal Ward identities it satisfies. In order to prepare an analogous study
for a Schrdinger-invariant system, we shall go here through an exercise in classical nonrelativistic field theory. In Galilei-invariant field theories, a technical problem comes from
the fact that wave functions transform under projective representations (as given by the
companion function fg and parametrized by the non-relativistic mass m [44]) rather than
under true representations. However, Giulini [16] pointed out that by treating the mass m
as a dynamical variable and going over to the new field defined by (we merely write here
the single-particle case)

1
(t, r) =
(1.7)
d eim (, t, r)
2
R
one obtains a true unitary representation T g of the Galilei group (see [17] for a discussion
on the interpretation of the Bargmann superselection rule). In Section 2, we shall show that
his result generalizes to the full Schrdinger group Sch(d) as well as to a certain infinitedimensional extension thereof.
410
Treating the mass m as a dynamical variable from the outset allows further insights.
In Section 3, we shall derive a precise relationship between the Schrdinger algebra
and the Lie algebra of the conformal group. In particular, we shall show that for the
complexified Lie algebras, one has schd confd+2 . We shall show that the maximal
kinematic invariance group of the d-dimensional free Schrdinger equation with variable
mass is the (d + 2)-dimensional conformal group. We also discuss the relevance of
the parabolic subalgebras of (confd+2 )C for physical applications, notably to ageing in
simple magnets. Finally, we correct the claims of Barut and show that his would-be group
contraction from confd+1 to schd should rather be viewed as a projection from (confd+2 )C
d distinct from schd . In Section 4, we consider the Schrdinger

to a new subalgebra alt
Ward identities which must be satisfied by the energymomentum tensor. The improved
energymomentum tensor satisfying the resulting symmetry requirements on the level of
classical field theory will be explicitly constructed, for theories built either from fields
(t, r) or from fields (, t, r). We also show how to generalize these considerations such
as to make them applicable to ageing phenomena, where time-translation invariance no
longer holds. In Section 5, we discuss the resulting two- and three-point functions and
show that they satisfy the causality conditions required for their physical interpretation
as response functions. We conclude in Section 6. In Appendix A, we provide details on
the non-relativistic limit of the conformal algebra. In Appendix B, we derive the causality
conditions satisfied by the two- and three-point functions. Appendix C reviews basic facts
about parabolic subalgebras.
2. Extended Schrdinger transformations

We consider the following infinite-dimensional extension of sch1 , denoted by S1 , and
spanned by the generators {Xn , Ym , Mn } with n Z and m Z + 1/2. They are of the
following form [26]
n+1 n
x
n(n + 1)
t rr (n + 1) t n
Mt n1 r 2 ,
2
2
4

1 m1/2
m+1/2
t
r m +
rM,
Ym = t
2
Mn = Mt n .
Xn = t n+1 t
(2.1)
Here, the constant x is the scaling dimension of the field (t, r) on which these generators
act, see below. The generators satisfy the following non-vanishing commutation relations

n

m Yn+m ,
[Xn , Ym ] =
[Xn , Xn ] = n n Xn+n ,
2

[Xn , Mn ] = n Mn+n ,
(2.2)
[Ym , Ym ] = m m Mm+m .
Extensions to d > 1 are straightforward, see [31]. The special case M = 0 of S1 was
rediscovered later as a kinematic symmetry of the 1D Burgers equation [35]. In particular,
the Xn satisfy the commutation relations of the Virasoro algebra without central charge.
Indeed, one might consider the generators Xn , Ym and Mn as the components of associated
411
conserved currents. Then the conformal dimensions of these currents, as measured by X0 ,

are dim X = 2, dim Y = 3/2 and dim M = 1. In other terms, just as the finite-dimensional
Schrdinger algebra was a semi-direct product of sl(2, R) with a Heisenberg algebra, the
Lie algebra S1 is a semi-direct product of a Virasoro algebra without central charge
(extending sl(2, R)) and of a two-step nilpotent (that is to say, whose brackets are central)
Lie algebra generated by the Ym and Mn , extending the Heisenberg algebra.
We shall assume that the action of the generators (2.1) describes the transformation of a
non-relativistic field (t, r) of mass M. It is straightforward to integrate the infinitesimal
transformations through formal exponentiation. From the generators Xn we find the
following coordinate transformations (where (t ) is a non-decreasing function of t )

t = t ,
(2.3)
r = r t .
Here and in the sequel the dot will denote the derivative with respect to the time variable.
The field transforms into such that

) 2
x/2
M (t
(t, r) = t
(2.4)
exp
r
t ,r .
)
4 (t
The infinitesimal generator Xn in Eq. (2.1) is recovered from (t) = t t n+1 in the limit
0.
Similarly, exponentiation of Ym gives the coordinate transformations
t = t ,
r = r (t)
and the field transforms as

1

(t, r) = exp M t t r t
t , r .
2
(2.5)
(2.6)
The infinitesimal generator Ym is recovered from (t) = t m+1/2 in the limit 0.

Finally, exponentiation of Mn merely changes the phase of

(t, r) = exp M (t) (t, r)
(2.7)
without any changes in the coordinates.
In the sequel, it will be useful to work with the Fourier transform of the field and of the
generators with respect to M. We define the new field as follows

1
(t, r) =
(2.8)
d eiM (, t, r).
2
R
As a consequence of (1.2), this field satisfies the equation of motion, provided

lim (, t, r) = 0
2i
2
2
+ 2 =0
t
r
which for the sake of brevity we shall also call a Schrdinger equation.
(2.9)
412
The generators of S1 act on the Fourier transform (, t, r) as follows:

n+1 n
x
n(n + 1) n1 2
t rr (n + 1) t n + i
t
r ,
2
2
4

1 m1/2
r ,
Ym = t m+1/2 r + i m +
t
2
Mn = it n .
Xn = t n+1 t
(2.10)
Let us write out for later use the action on of those generators of sch1 which contain a
phase shift
Y1/2 = tr + ir ,
Galilei transformation,
i
X1 = t 2 t tr r + r 2 xt, special Schrdinger transformation,
2
M0 = i ,
phase shift.
(2.11)
It turns out that the field transforms in a simpler way than the field considered
above. The complicated phases acquired by the field are replaced by a translation of the
new internal coordinate , and at most a scaling factor remains. Under the action of the
Xn , we find

) 2

i (t
r = r t ,
=
t = t ,
(2.12)
r
)
4 (t
and the field transforms into such that
x/2
(, t, r) = t
,t ,r .
(2.13)
Similarly, exponentiation of Ym gives the coordinate transformations

i
(2.14)
= + ir t t t
2
and the field (, t, r) = ( , t , r ) transforms trivially. The absence of a phase factor
under the Galilei transformation Y1/2 (where (t) = vt) was observed before by Giulini
[16].
Finally, the Mn merely give time-dependent translations in the coordinate .
Summarizing, we have shown: the algebra S1 acts on fields (t, r) with a fixed
mass as a projective representation. Conjugation by Fourier transformation with respect
to M changes this into a true representation of the same algebra, now acting on functions
(, t, r).
t = t ,
r = r (t),
3. Relation between Schrdinger and conformal transformations

In this section, we shall investigate the relationship between the Schrdinger Lie algebra
sch1 and the conformal Lie algebra conf3 .
We start from the three-dimensional massless KleinGordon equation
( ) = 0.
(3.1)
413
The Lie algebra of the maximal kinematic group of this equation is the conformal algebra
conf3 , with generators
P = ,
M = ,
K = 2 + 2x ,
D = + x
(3.2)
(, {1, 0, 1}) which represent, respectively, translations, rotations, special transformations and the dilatation. Here x is the scaling dimension of the field . If is a solution
of (3.1), x = 1/2.
It is well known that conf3 is isomorphic to so(4, 1). An explicit identification is given
by the so(4, 1) generators J as follows, where now 1 , 3
J = iM ,
J2,3 = iD,
i
J2 = (P + K ),
2
i
J3 = (P K ).
2
(3.3)
Next, we define the physical coordinates as3

i
1
1
1
t = (0 + i1 ),
= (0 + i1 ),
r=
(3.4)
2
2
2
and (, t, r) = ( ). Then the KleinGordon equation (3.1) reduces to the Schrdinger
equation (2.9)

2
2
+ 2 (, t, r) = 0.
2i
(3.5)
t r
It follows that the generators of sch1 are linear combinations (with complex coefficients)
of the above conf3 generators, so that we have an inclusion of the complexified Lie algebra
(sch1 )C into (conf3 )C . Explicitly
X1 = i(P1 iP0 ),

Y1/2 =
2
P1 ,
i
i
1
X0 = D + M10 ,
2
2

i
(M11 + iM01 ),
Y1/2 =
2
i
X1 = (K1 + iK0 ),
4
M0 = P1 + iP0 . (3.6)
Four additional generators should be added in order to get the full conformal Lie algebra
(conf3 )C . We take them in the following form
N = iM10 = tt + , time-phase symmetry,

i
(M11 iM01 ) = r + irt , dual Galilei transformation,
V =
2
3 We use the short-hand
i = ei/4 throughout.
414
i
i
W = (K1 iK0 ) = 2 rr + r 2 t x,
4
2
dual special transformation,

i
K1 = 2 tr t 2 r r 2 + 2i t r 2xr,
V+ =
2
transversal inversion.
(3.7)
The generators V and W are, up to constant coefficients, the complex conjugates of Y1/2
and X1 , respectively, in the coordinates , hence, their names. The complex conjugation
becomes the exchange t in the physical coordinates (, t, r).
In order to understand these results, we recall a few basic facts from the theory of Lie
algebras [40]. The Lie algebra (conf3 )C is simple and of rank 2. The generators D and N
span a Cartan subalgebra. We now show that the generators of sch1 are root vectorsin
other words, they are common eigenvectors of the commuting generators N and D. Let e1
and e2 be the linear forms on RN RD defined by ei (N) = i,1 and ei (D) = i,2 . Recall
that e is a root if there is a non-zero element Z of (conf3 )C such that [N, Z ] = 1 Z
and [D, Z ] = 2 Z . Since our complexified conformal algebra is isomorphic to so(5, C),
its set of roots is of type B2 . One finds that = + , where
+ = {e2 , e1 + e2 , e1 , e1 e2 },

= + = e2 , (e1 + e2 ), e1 , e1 + e2 .
(3.8)
The elements of + are called positive roots and the elements of are called negative
roots. The root vectors can be identified explicitly
Xe2 = Y1/2 , Xe1 +e2 = X1 , Xe1 = Y1/2 ,
Xe2 = V+ , X(e1 +e2 ) = X1 , Xe1 = V ,
Xe1 e2 = M0 ,
X(e1 e2 ) = W.
(3.9)
These results are summarized in Fig. 1. Each of the points in the diagram indicates a
root space. They are labelled by the corresponding generators from (conf3 )C , according
to Eq. (3.9).
We may use this information to list some interesting subalgebras of (conf3 )C , using the
notion of parabolic subalgebras explained in Appendix C. If two root spaces Z1 , Z2 have
coordinates (i1 , j1 ), (i2 , j2 ) in Fig. 1, then the root space [Z1 , Z2 ] will have coordinates
(i1 + i2 , j1 + j2 ). Therefore, subalgebras may be easily obtained by taking a subset of the
points in Fig. 1 which is invariant under the addition of coordinates. Furthermore, the Weyl
group of the conformal Lie algebra (conf3 )C is given by the discrete set of transformations
(e1 , e2 ) (e1 , e2 ) or (e1 , e2 ) (e2 , e1 ). On the physical coordinates, they can be
implemented by the action of an element of the conformal group and will therefore give
isomorphic (conjugate) subalgebras.
The Weyl group is generated by w1 : (e1 , e2 ) (e2 , e1 ) and w2 : (e1 , e2 )
(e1 , e2 ). Both appear as simple symmetries on Fig. 1 below. On the physical coordinate
space, these act as (, t, r) ( , t , r ) such that
w1 :
= +
w2 :
= t ,
ir 2
, t = 1/t ,
2t
t = , r = r .
r = r /t ,
(3.10)
415
Fig. 1. Roots of B2 and their relation with the generators of the Schrdinger algebra sch1 . The double circle in
the centre denotes the Cartan subalgebra h.
The inversion w1 belongs to the Schrdinger group as can be checked from Eq. (2.12),
while the duality w2 is a conformal transformation.
We arrive this way at the following list of standard parabolic subalgebras, see
Appendix C for details. There are three of them, up to conjugation. We shall also mention
at the same time natural subalgebras of these, with one or two generators of the Cartan
subalgebra removed, that contain essentially the same information:
(1) the algebra age1 generated by X0 , Y1/2 , M0 , Y1/2 , X1 . Adding the generator N , we
1 := age1 CN , which is the minimal standard parabolic

obtain the algebra age
subalgebra (see Appendix C), namely, it is the sum of the Cartan subalgebra and of
the positive root spaces. It is also possible to dismiss altogether the generator X0 or
replace it by any linear combination of X0 and N ;
(2) the Schrdinger algebra sch1 . One may also add to it the generator N . We call
1 := sch1 CN the parabolic subalgebra thus obtained;

sch
(3) the algebra alt1 generated by D, Y1/2 , M0 , Y1/2 , V+ , X1 . As before, one may add the
1 := alt1 CN .
generator N and obtains thus the parabolic subalgebra alt
1 and alt
1 are maximal non-trivial subalgebras of (conf3 )C ,

Note that the algebras sch
as expected from the theory of parabolic subalgebras. This is easily checked from the
1 and alt
1 . In the first case,
1 is the intersection of sch

commutators. On the other hand, age
the generator X1 is taken out and in the second case, the generator V+ .
It may be interesting on physical grounds to introduce also the images of these algebras
under the Weyl symmetry (e1 , e2 ) (e1 , e2 ). This gives the following new subalgebras:
416
(4) the algebra generated by M0 , Y1/2 , V+ , X1 and any linear combination of X0 and N
(or both);
(5) the algebra generated by X0 +N, M0 , Y1/2 , W, V+ , X1 ; one may also add the generator
N.
The consideration of these subalgebras might yield physically interesting applications.
For example, it has been recently shown that the response functions of simple ferromagnetic systems undergoing ageing after a quench from some initial state to a temperature
below the equilibrium critical temperature Tc are determined from their covariance under
the infinitesimal transformations contained in aged [30,31]. Since ageing phenomena do
break time-translation invariance generated by X1 , a subalgebra without this generator is
needed. The existence of the subalgebra alt suggests that the response functions might also
transform covariantly under the conformal generator V+ .
A new type of application is suggested by the fourth and fifth subalgebras. In contrast
to sch1 , not only time-translation invariance (X1 ), but also space translation invariance
(Y1/2 ) is broken. The breaking of both of these would be a necessary requirement to
describe ageing processes in disordered, e.g., glassy systems. It remains to be seen whether
the larger algebra, with both V+ and W present, or the smaller one with only V+ , is realized
in physical systems. Tests of this possibility are currently being performed and will be
described elsewhere.
Another interesting way (motivated by an analogy with the scheme of group contractions) to look at how these subalgebras sit inside the conformal algebra is to consider a
family of linear maps depending on a parameter c, that gives in the c limit a kind
of projection of conf3 onto any of these subalgebras. This is easy to construct by using
conjugation by exp((aN + bD) log c) for adequate a, b, which multiplies each of the root
vectors above by a certain power of c. For instance, in the new coordinates
= c2 ,
t = t,
r = cr,
(3.11)
cij ,
so the X-generators
any root vector with coordinates (i, j ) in Fig. 1 is multiplied by
and N are preserved, Y1/2 are multiplied by c, M0 by c2 , while the other generators go to
1 . Similarly, if one sets

zero. So, in a certain way, one has a projection conf3 sch
= c2+ ,
t = c t,
r = cr
( > 0),
(3.12)
then X1 is multiplied by c , X0 and N remain constant, and Y1/2 , Y1/2 and M0 are
multiplied, respectively, by c, c(1+) and c2+ , while the other generators go to zero.
1 . In physical applications,
Therefore, in the c limit, one has a projection conf3 age
one usually considers families of c-dependent maps such that c can be interpreted as the
speed of light. Indeed, Barut [1] had claimed that a group contraction from confd+1 to schd
were possible. We shall reconsider his argument in Appendix A. Working with the masses
as dynamical variables from the very beginning, we shall show that his procedure rather
1 in the non-relativistic limit.

gives a map conf3 alt
One may also form the infinite-dimensional algebra S1 := S1 CN , which has
besides Eq. (2.2) the following non-vanishing commutators

1
Ym ,
[Ym , N] = m +
[Mn , N] = (n + 1)Mn . (3.13)
[Xn , N] = nXn ,
2
417
1 and in S, the generator M0 is no longer central.

Note that, both in sch
1
In order to see which of these generators form, indeed, a symmetry algebra of the free
Schrdinger equation (2.9), consider the 1D Schrdinger operator
S = 2i t + r2 .
(3.14)
It is straightforward to check that

1 n1
n3 n n2 2 2
t
t

r ,
[S, Xn ] = (n + 1)t n S in(n + 1) x
2
2

1 m3/2 2
r ,
[S, Ym ] = 2 m2
t
4
[S, Mn ] = 2nt n1 2 ,
[S, V ] = 0 = [S, N],
[S, V+ ] = 2(1 2x)t 4rS,
[S, W ] = i(1 2x)r 2 S.
(3.15)
Therefore, under the action of the conformal algebra (conf3 )C any solution of the
Schrdinger equation S = 0 with a scaling dimensions x = 1/2 is mapped onto another
solution of (2.9). Restricting to the subalgebra sch1 , we recover the well-known invariance
of the free Schrdinger equation with fixed mass [21,44].
The results of this section can be extended to an arbitrary space dimension d, but we
shall not discuss this here.
4. Schrdinger-invariant free fields

Having dealt with the algebraic aspects, we now study how the physical action may
transform under an extended Schrdinger transformation. In particular, we are interested
in the consequences for the Schrdinger Ward-identities linking the components of the
energymomentum tensor. We shall check our results for free Schrdinger-invariant fields.
As we have seen above, it is useful to study Schrdinger invariance in a setting where
the mass M is treated as an additional variable from the outset. We shall, therefore, study
two types of action. The first one is the usual one, see, e.g., [14], with M fixed

.
Sa = dt dr La , La = M
(4.1)
t
t
r r
The equation of motion for Sa gives (1.2). The conjugate field satisfies the same
equation with M replaced by M. The mass M plays therefore the role of a quantum
number and we associate with a mass M and with a mass M := M.
On the other hand, treating M as a variable [16], we use the field = (, t, r) and
consider the free action

2
Sb = d dt dr Lb , Lb = 2i
(4.2)
+
t
r
with Eq. (2.9) as equation of motion.
418
4.1. We first consider the transformation of the action Sa under the extended
Schrdinger transformation. In view of the results of Section 3, we take x = 1/2 as the
scaling dimension of the fields , . The finite coordinate changes generated from the
exponentiated generators exp Xn and exp Ym are given by two functions (t) and (t),
see Section 2. The transformation of the fields , under these is explicitly known. We
find the change of the action Sa Sa = Sa + Sa , where

1
2
X Sa = dt dr M2 r t , t ,
2

Y Sa = dt dr M2 t 2r t ,
(4.3)
respectively. Here,
...

2

(t) 3 (t)
(t), t =
2 (t)
(t)
(4.4)
is the Schwarzian derivative. Consequently, Sa = 0 if (t) is at most linear in t and (t)

is a Mbius transformation. These transformations make up exactly the Schrdinger group
Sch(1) as defined in Eq. (1.1), as expected [21,44].
We now discuss the Ward identities which follow from the hypothesis of Schrdinger
invariance. Denote by = (t, r) the vector of spacetime coordinates. Consider an arbitrary
infinitesimal coordinate transformation = + () such that a field operator
= (t, r) may also pick up a phase = (t, r). Then the simplest possible way a local
translation-invariant action may transform to leading order in is given by

S = dt dr (T + J )
(4.5)
which defines the energymomentum tensor T and the current vector J , with , =
0, 1, . . . , d. If we had included an extra term J , the invariance under constant phase
shifts generated by M0 would have immediately implied that J = 0. Now, S = 0 under
translations by construction. Furthermore, scale invariance implies the modified trace
identity [21,26,37]
2T00 + T11 + + Tdd = 0.
(4.6)
For Galilei transformations, the phase = Mr and invariance of the action S implies
T0a + MJa = 0,
a = 1, . . . , d.
(4.7)
Furthermore, using rotation invariance, one gets Tab = Tba with a, b, = 1, . . . , d. Now,
quite analogously to conformal invariance, see [11,29], the Ward identities (4.6), (4.7)
imply full Schrdinger invariance of the action. Take d = 1 for simplicity, then 0 = t 2 ,
1 = tr and = 12 Mr 2 . Thus for a special Schrdinger transformation

X1 S = dt dr (2T00 + T11 )t (T01 + MJ1 )r = 0.
(4.8)
Therefore, for sufficiently local interactions such that (4.5) is valid, invariance under
spatio-temporal translations, phase shifts, dilatations and Galilei transformations is
419
enough to guarantee invariance under the full Schrdinger group. Previously, this was
only proven for vanishing masses M = 0 [26].
For later use, we list how S should transform according to (4.5) under the generators
Xn , Ym of the extended Schrdinger algebra S1 . We find

1
2 ...
Y ()S = dt dr MrJ0 ,
X()S = dt dr Mr J0 ,
(4.9)
4
where
1
M
x
X() = (t)t (t)rr
(t)r 2 (t),
2
4
2
Y () = (t)r M(t)r
(4.10)
and by comparison with the infinitesimal transformations of Sa which can be read from
Eq. (4.3) we identify
J0 = 2M .
(4.11)
To finish, we now construct T and J explicitly from the free-field action Sa . From
the canonical recipe, we would obtain
La
La
+
,
T = La +
( )
( )
La
La
J =
( )
( )
(4.12)
Using the equations of motion, these may be shown to satisfy the conservation laws
T = J = 0 and all Schrdinger Ward identities with the only exception of the
trace condition Eq. (4.6). This can be remedied along the lines of [7] by constructing the
improved tensor
T = T + B ,
(4.13)
where B is antisymmetric in the two first variables, B = B . If we take Ba00 =

d
4 ( a + a ) with a = 1, . . . , d and B = 0 unless () = (a00) (up to

symmetries), then we get a classically conserved energymomentum tensor, satisfying all
required Ward identities, which reads

dM
d
1 r r ,
T00 =
t t +
2
2

d

d
a t + a t a,t + a,t ,
Ta0 = 1
4
4

T0a = M a a ,

Taa = 2a a r r M t t ,
Tab = a b + b a .
(4.14)
The current J need not be improved and we have

J0 = 2M ,
Ja = a a .
(4.15)
420
In particular, we recover (4.11). For a physical interpretation, it is better to divide La

by 2M. We then recover the usual interpretation of J0 as a probability density, Ja as a
probability current, T00 as an energy density, T0a as a momentum density and Ta0 and Tab
as energy and momentum currents, respectively.
Notice that (2M)1 T00 coincides in d = 2 dimensions and with t replaced by z, with
the energymomentum tensor T (z) of a complex fermionic field, see [11, p. 147].
4.2. Similarly, we now treat the transformation of the free-field action Sb under the 3D
conformal group Conf (3), whose Lie algebra generators are listed in Section 3. Recalling
the transformation of the field from Section 2, and setting x = 1/2, the action changes
into Sb Sb = Sb + Sb where

1 2 2
X Sb = d dt dr r t , t
,
2

2

Y Sb = d dt dr t 2r t
(4.16)
.

It is also esay to see that S = 0 under the other generators of conf3 . This means that Sb
is invariant under the 3D conformal group, in agreement with the conclusion drawn in
Section 3 from the infinitesimal transformations.
As before, we now discuss the Ward identities. To straighten notation, we construct a
vector with components
1 = ,
0 = t,
1 = r1 ,
...,
d = rd
(4.17)
and write the derivatives = / . Under an infinitesimal transformation

=
+ ( ), the action is assumed to transform to leading order as

S = d dt dr T
(4.18)
and we proceed to write down the Ward identities, restricting to d = 1 for simplicity. Again,
S is translation-invariant par construction. Dilatation invariance implies
2T00 + T11 = 0.
(4.19)
Invariance of S under the three generators N , Y1/2 and V coming from the 3D conformal
rotations (see Section 3) yields the following Ward identities, respectively
1
T00 = 0,
T1
T01 iT11 = 0,
1
T1
iT10 = 0
(4.20)
and it follows that T has 5 independent components. Next, we consider the three remaining
generators X1 , W and V+ which come from the 3D special conformal transformations. The
components of are easily read from the generators and we have

X1 S = d dt dr 2T00 + T11 t + T01 iT11 r = 0,

1

1

+ T11 + T1
iT10 r = 0,
W S = d dt dr 2T1

V+ S =
d dt dr
421

1

1
T1
+ T00 + T11 r + T01 iT11 i + T1
iT10 it = 0.
(4.21)
This merely translates the well-known result of conformal invariance, namely, that
translation, rotation and scale invariance imply full conformal invariance [7,11,29], to
the formulation at hand. Furthermore, considering the transformation of S under the
infinitesimal action of the generators X and Y of the extended Schrdinger algebra, we
read off from the Fourier transform of (4.10) and find

i
...
Y () S = i d dt dr r T01 .
X()S =
(4.22)
d dt dr r 2 T01 ,
4
As before, we can compare this with the infinitesimal form of the transformation of the
free-field action Sb in Eq. (4.16) and identify

2
T01 = 2i
(4.23)
which is the analogue of Eq. (4.11).

Therefore, the current J0 or the component T01 , respectively, generate the change of
the action under an infinitesimal extended Schrdinger transformation. We point out that
in 2D conformally invariant classical field theories, no such term is present. Namely, the
conformal generators 9n = zn+1 z and 9n = zn+1 z with n Z can be rewritten as
Xn = 9n + 9n and Yn = i(9n 9n ) where the time t and space r were introduced
from the complex coordinates z, z via z = t + ir, z = t ir. The projective conformal
Ward identities T00 + T11 = 0 and T01 = T10 follow as usual. Discarding any conformal
anomalies, these identities are sufficient to show that Xn S = Yn S = 0 for all n Z.
We finish by constructing the energymomentum tensor explicitly. The canonical
energymomentum tensor is given by
T = Lb +
Lb

( )
(4.24)
and may be written in a matrix form (here for the d = 1 case, where labels the columns
and we write = /, . . .)
2it2
2ir t
r2
r2
2i r .
T = 2i2
(4.25)
2
2r 2r t r 2i t
It reproduces (4.23), is classically conserved and satisfies all Ward identities (4.20) coming
from the 3D conformal group, but not scaling identity Eq. (4.19). To correct this, define
the improved tensor [7]
+ B
,
T = T
(4.26)
antisymmetric in in , which has the same divergence as T

. Choosing
with B
1

2
0
2 r
1
B10
B11
= 2i t ,
= 2 r ,
=
0
, (4.27)
B10
i
2i t
0
2
422
the improved energymomentum tensor reads (where the equations of motion were used)
i t i t r2
3it2 it t
i[3t r t r ]
1
i t i t r2
i[3 r r ]
3i2 i
T =
2
2
3 r r
3t r t r
2[i t i t r ]
(4.28)
which manifestly satisfies the conditions Eqs. (4.19), (4.20) and is conserved. Of course,
this is just a particular case of the construction of the Belinfante tensor in 3D conformal
theory (see [7,11]). As pointed out long ago [7], the consideration of the improved energy
momentum tensor in Poincar-invariant interacting theories is of particular interest, since
satisfying the conformal Ward identities implies that the elements of T remain finite in
the limit of a large cut-off for renormalizable interactions. Their result should translate,
through the inclusion of schd into confd+2 described in Section 3, to non-relativistic
interacting theories and one should be able to avoid this way difficulties [22] with the
finiteness of the elements of T which may arise in Galilei-invariant field theories with
fixed masses.
4.3. Having studied so far the full conformal algebra (conf3 )C , we now consider
its subalgebras. In particular, we inquire about the status of the Ward identities for the
subalgebras sch1 and age1 .
We consider first the free-field action Sb , formulated in terms of scaling operators
= (, t, r). From Eq. (4.21) it is clear that the Ward identities coming from dilatation
and Galilei invariance are sufficient to prove also special Schrdinger invariance, i.e.,
X1 S = 0. The question raised is thus settled for sch1 .
A new aspect arises, however, for age1 . In that case, time translation invariance is no
longer assumed, but all elements of age1 keep the line t = 0 invariant. Consequently,
the transformation (4.18) of the action under infinitesimal transformations might be
generalized to

S = d dt dr T +
(4.29)
d dr U ,
(t =0)
where the second integral is restricted to the boundary t = 0.

Then, and now specializing to d = 1, translation invariance in r and in yields
U 1 = U 1 = 0. Although U 0 is not fixed, it does not contribute to S, since 0 = 0 at
t = 0 for all elements of age1 . From dilatation and Galilei invariance, the Ward identities
T00 + 12 T11 = 0 and T01 iT11 = 0 follow and consequently for a special Schrdinger
transformation generated by X1 , we have

i

d dr r 2 U 1 = 0.
X1 S = d dt dr 2T00 + T11 t + T01 iT11 r +
2
(t =0)
(4.30)
This means that the validity of special Schrdinger invariance mainly depends on having a
z = 2 scale invariance and Galilei invariance, while time-translation invariance is not really
required.
423
Similarly, if we had chosen instead to work with a fixed mass M, the transformation
(4.5) of the action should be generalized as follows

dr (U + V )
S = dt dr (T + J ) +
(4.31)
(t =0)
and it is now straightforward to see that again special Schrdinger invariance X1 S = 0

follows. Note that it follows in particular form phase-shift invariance that V should
be 0.
Our result is that special Schrdinger invariance holds as a consequence of scale
and Galilei invariance, even in the absence of time-translation invariance, provided only
that the dynamics is local in the sense of Eqs. (4.29) or (4.31). It is well-known that
ageing ferromagnets undergoing phase-ordering kinetics after being quenched to a fixed
temperature T below criticality is scale invariant with a dynamical exponent z = 2 [4].
If we accept that these systems are also Galilei-invariant, we can conclude that these
coarsening systems must also be invariant under special Schrdinger transformations. This
explains why special Schrdinger invariance could be successfully tested in these systems,
see [20,30,32,33,47].
5. Response functions
Having described the relationship between conformal and Schrdinger transformations,
we now discuss the consequences for the two- and three-point functions. Let a ( a ) =
a (a , ta , r a ) be a scalar and quasi-primary conformal scaling operators with scaling
dimension xa . We can always take the a to be real. It is well known that [50]

1 (1 , t1 , r 1 )2 (2 , t2 , r 2 )

= 1 ( 1 )2 ( 2 ) = 0 x1 ,x2 | 1 2 |2x1

i (r 1 r 2 )2 x1
= 0 x1 ,x2 (t1 t2 )x1 1 2 +
,
(5.1)
2 t1 t2
where 0 = 4x1 0 and 0 is a normalization constant. In order to understand the physical
meaning of the result (5.1), we rewrite it in terms of scaling operators a (t, r) with fixed
mass Ma 0, using Eq. (2.8). As shown in Appendix B, we find, provided x1 > 0

1 (t1 , r 1 )2 (t2 , r 2 )

1
=
d1 d2 eiM1 1 +iM2 2 1 (1 , t1 , r 1 )2 (2 , t2 , r 2 )
2
R2

M1 (r 1 r 2 )2
t2 )(t1 t2 )
exp
,
2
t1 t2
(5.2)
where 0 is again a normalization constant (proportional to 0 ) and is the Heaviside
function. The functional form of had been derived before from the requirement
1
(t1
= 0 x1 ,x2 (M1 M2 ) M1x
1
x1
424
that it transform covariantly under the realization of the Schrdinger group Sch(d) with
fixed masses [26]. We stress that the causal prefactor (t1 t2 ) is not a consequence
of covariance under that realization, but rather had to be put in by hand [26] to comply
with the requirement that 1 2 should decay to zero for large distances r = |r 1 r 2 |.
Furthermore, causality as implied in (5.2) fits perfectly with the interpretation of =
as response function in the context of MartinSiggiaRose theory [42] and suggests

the identification of the complex conjugate scaling operator with the response operator
conjugate to the order parameter scaling operator .
,
An analogous result holds for three-point functions. Recall the well-known result of
conformal invariance [50]

1 ( 1 )2 ( 2 )3 ( 3 ) = C123 | 1 2 |x12,3 | 2 3 |x23,1 | 3 1 |x31,2 ,
(5.3)
where xab,c := xa + xb xc and C123 is an operator product expansion (OPE) coefficient.
As shown in Appendix B we find, provided x1 > 0 and x2 > 0

1 (t1 , r 1 )2 (t2 , r 2 )3 (t3 , r 3 )
= C12,3(M1 + M2 M3 )
(t1 t3 )(t2 t3 )(t1 t2 )x12,3 /2 (t1 t3 )x13,2 /2 (t2 t3 )x23,1 /2

M1 (r 1 r 3 )2 M2 (r 2 r 3 )2
exp
2
t1 t3
2
t2 t3

1 [(r 1 r 3 )(t2 t3 ) (r 2 r 3 )(t1 t3 )]2
,
12,3
2
(t1 t2 )(t2 t3 )(t1 t3 )
(5.4)
where C12,3 is a constant related to the OPE coefficient C123 and 12,3 (v) is a scaling
function (see Eq. (B.11) for an integral representation). As we had seen before for the twopoint function, the functional form of this result is in complete agreement with the one
found previously from Schrdinger invariance alone for scaling operators a with fixed
masses Ma 0 [26]. However, the causality conditions t1 > t3 and t2 > t3 expected for a
response function are now automatically satisfied and need no longer be put in by hand. The
conditions on the masses in (5.2), (5.4) are nothing but examples of the standard Bargmann
superselection rules.
In summary, we have seen how to reconstruct the physically more appealing expectation
values and from their conformal forms as given by , , where
we have used right away the forms (5.1), (5.3) which follow from covariance under the full
conformal algebra (conf3 )C . We now ask to what extent the form of the two-point function
is already determined by one of the subalgebras as obtained in Section 3.
In order to do so, we shall begin by deriving the constraints on the two-point function
(for simplicity in d = 1 dimensions)

F = 1 (1 , t1 , r1 )2 (2 , t2 , r2 )
(5.5)
1 = {X0,1 , Y1/2 ,
coming from its invariance under the minimal parabolic subalgebra age
M0 , N}. It is convenient to introduce the variables u = t1 t2 and v = t1 /t2 . From
space and phase translation invariance, we have F = F (, u, v, r), where = 1 2 and
425
r = r1 r2 . Next, the covariance conditions yield the following equations

1
x1 + x2
X0 F = uu rr
F = 0,
2
2
Y1/2 F = (ur + ir )F = 0,
NF = (uu + )F = 0,

i
X1 F = u2 u uvv urr + r 2 ux1 F = 0.
2
(5.6)
Using the first and second of these again, the last condition X1 F = 0 simplifies into

x2 x1
F = 0.
vv +
(5.7)
2
Therefore, we have the factorization F = F0 (v)F1 (, u, r), where F1 satisfies the same
conditions found in the time-translation invariant case. We easily find

1 (1 , t1 , r1 )2 (2 , t2 , r2 )
(x2 x1 )/2

t1
i (r1 r2 )2 (x1 +x2 )/2
(x1 +x2 )/2
1 2 +
(t1 t2 )
= 0
(5.8)
t2
2 t1 t2
with some normalization constant 0 . Comparing with the conformal result Eq. (5.1), we
observe the absence of the constraint x1 = x2 on the scaling dimensions and the presence
of a factor which explicitly breaks time-translation invariance.
If we had merely required invariance under a proper subalgebra of the minimal parabolic
1 , we would not have had sufficiently many conditions in order to fix the
subalgebra age
two-point function completely. However, if we restrict ourselves to age1 by leaving out
only the generator N , although the two-point function = t x1 E( + ir 2 /2t) then
contains the undetermined scaling function E, the form of the response function =
0 t x1 exp(r 2 /(2M1t)) is still completely fixed, up to a mass-dependent normalization
constant 0 = 0 (M1 ).
1 to one of the two maximal parabolic

Next, we consider the extension of age
1
subalgebras of conf3 studied in Section 3 and in Appendix C. First, we consider sch
by adding the generator X1 of time translations. This simply enforces x1 = x2 and we
1 by adding the generator V+ .

recover the conformal result (5.1). Second, we consider alt
We obtain

V+ F = 2uru 2vrv 2 r r 2 + 2i u r 2rx1 2t2 (ru + i r ) F
= 2u(v 1)1 (ru + i r )F
(5.9)
and where in the second line the covariance conditions Eqs. (5.6), (5.7) were used. From
(5.8) it can be seen that indeed V+ F = 0 and the two-point function transforms covariantly
1 . In this case, there is no constraint on x1 and x2 .

under alt
The case of an arbitrary space dimension d is treated in the same way by simply
replacing the ra by r a . In conclusion, we have found two distinct forms of the two-point
function, namely, (5.1) and (5.8). The form of the two-point function is completely fixed if
If in addition the dynamic symmetry algebra contains

the symmetry algebra contains age.
426
the two-point function reduces to the form

the extended Schrdinger algebra sch,
(5.1) as obtained from full conformal invariance or, alternatively, the causal form
of Eq. (5.2). On the other hand, if time-translation invariance is absent, the algebra of
or age
and the two-point function is given by

dynamical symmetries may be given by alt
(5.8). The causal form with fixed masses is then given by, provided x1 + x2 > 0

1 (t1 , r 1 )2 (t2 , r 2 )
1(x +x )/2
= 0 (M1 M2 ) M1 1 2

(x2 x1 )/2

t1
M1 (r 1 r 2 )2
. (5.10)
(t1 t2 )
(t1 t2 )(x1 +x2 )/2 exp
t2
2
t1 t2
0) = (t; r) (s; 0),
This form for the response function R(t, s; r) = (t; r)(s;
including the causality condition t > s, has been confirmed in ageing phenomena in several
models of simple ferromagnets quenched to a temperature T < Tc below criticality, in
particular the 2D and 3D GlauberIsing model [30,32,33], the spherical model with a nonconserved order parameter [6,8,19,30,47] and, last but not least, the free random walk [10].
6. Conclusions
Our study of Schrdinger invariance as a dynamical spacetime symmetry was motivated
by the known explicit confirmation of some of its consequences in a few specific and nontrivial models. In order to understand better the origin of such a symmetry, a useful starting
point is the analysis of the associated classical free-field theory of which the Schrdinger
equation is the EulerLagrange equation of motion. Going through this exercise, it became
apparent to us that the mass M in this equation should be considered as a dynamical
variable on the same level as space and time coordinates. This leads us to the following
results:
(1) the usual projective representation of the Schrdinger group becomes a true
representation, via conjugation by Fourier transformation with respect to M;
(2) a new relation between the Schrdinger Lie algebra schd and the complexified
conformal Lie algebra (confd+2 )C is found, namely,
schd (confd+2 )C .
(6.1)
Some subalgebras of (confd+2 )C closely related to parabolic subalgebras may play a rle
in physical applications, notably to ageing phenomena in spin systems. We leave the
elaboration of this to future work.
We also reconsidered an old claim [1] that schd could be obtained as a non-relativistic
limit from a conformal Lie algebra. If the mass M is treated as a dynamic variable from
1 = sch1 ;
the beginning, we find in the non-relativistic limit c instead (conf3 )C alt
(3) the Ward identities which express the invariance of an action under conformal
transformations, or merely under the Schrdinger subalgebra schd or else age, are derived
from the locality assumption Eqs. (4.29) or (4.31). An important open question is the
427
extension of schd to an infinite-dimensional algebrapossibly S1 or natural extensions

thereof to d spatial dimensionssuch as to include those terms which would describe
the contributions of anomalies coming from quantum fluctuations. In order to prepare
such a study, we explicitly constructed the conserved energymomentum tensor (and also
the probability current) which satisfy these Schrdinger (or conformal) Ward identities.
Further work along these lines is in progress.
When applying these considerations to ageing, we must take into account that timetranslation invariance does no longer hold. This can be dealt with by allowing for a
boundary term in the action and coming from the t = 0 initial conditions. Schematically,
we have
spatial translation invariance
phase shift invariance
Galilei invariance
(6.2)
special Schrdinger invariance,
scale invariance with z = 2
locality
where locality is understood in the sense of Eqs. (4.29), (4.31). We emphasize that while
time-translation invariance is not really needed, Galilei invariance is a necessary condition
for having an invariance under the action of the special Schrdinger transformation
generated by X1 .
We have also seen that the classical free-field action is not invariant under the algebra S1 ,
in contrast with the situation found for two-dimensional conformal invariance;
(4) tests of the predictions of Schrdinger invariance can be carried out by considering
the following response functions

(t1 )
= (t1 ) (s) ,
= (t1 )(s)
h(s)

(t1 )(t2 )
= (t1 )(t2 )(s)

= (t1 )(t2 ) (s) ,
R3 (t1 , t2 , s) =
h(s)
R2 (t1 , s) =
(6.3)
where is the MartinSiggiaRose response operator associated with the order parameter
scaling operator and h is the conjugate magnetic field. Here t1 , t2 are observation times
and s is a waiting time. Our results (5.2), (5.4), (5.10) suggest the identification =
of the response operator with the complex conjugate in the formalism at hand. In
particular, the causality conditions t1 > s and t2 > s required for an interpretation of R2
and R3 as response function were derived in a model-independent way.
In several spin systems undergoing ageing, the resulting scaling form (5.10) of the twopoint response function R2 has been fully confirmed, see [20,3033,47]. However, we are
not aware of any tests of (5.2), (5.4) in equilibrium critical dynamics. It appears significant
that covariance under at least the minimal parabolic subalgebra of the conformal algebra is
required in order to fix the two-point function completely. Tests of the three-point response
R3 would be very interesting and might provide an answer to the open question which
or age
is the relevant one for the description of

of the two parabolic subalgebras alt
ageing phenomena. This will also require the explicit inclusion of the effects of the initial
conditions into the analysis, where work is in progress [48].
428
Appendix A. On the non-relativistic limit

We briefly describe the non-relativistic limit of the conformal Lie algebra confd and the
relation to the Schrdinger Lie algebra schd . This was discussed by Barut [1] long ago. He
started from the massive KleinGordon equation

1 2
2 2
M
(A.1)
M (t, r) = 0,
+
c
c2 t 2 r r
where c is the speed of light. This equation is invariant under the conformal group in (d +1)
dimensions, provided the mass M is transformed as well [1]. After the substitution
1
t Mc + t
(A.2)
c
the non-relativistic limit c reduces (A.1) to the Schrdinger equation. However, the
c limit of the conformal generators (3.2) in d + 1 dimensions (and with 0 = ct
and a = ra , a = 1, . . . , d) do not commute with the Schrdinger operator. Therefore,
Barut argued that one may . . . fix M, but change the transformation properties of t and
r in such a way that we obtain symmetry operations for the Schrdinger operator. . . [1]
and in this way, a contraction confd+1 schd is claimed to be achieved. However, that
procedure appears rather ad hoc and it might be useful to reconsider that derivation in
somewhat more detail, treating M as a dynamical variable from the outset and avoiding
any ill-defined changes of transformation properties.
Starting again from (A.1), we define a new function (u, t, r) through

1
du eiMu (u, t, r)
M (t, r) =
(A.3)
2
R
which satisfies the equation of motion (if limu (u, t, r) = 0)

2
1 2
+c
(u, t, r) = 0
+
c2 t 2 r r
u2
(A.4)
and the contact with the (d + 2)-dimensional massless KleinGordon equation and the
associated conformal generators (3.2) is reached by defining ( ) = (u, t, r) where
1 = u/c, 0 = ct and a = ra , a = 1, . . . , d. Next, we define the wave function
(, t, r) := (u, t, r),
:= u + ic2 t
(A.5)
which is the exact analogue of Baruts substitution (A.2) in t and have from (A.4)

2
1 2
+
2i
(A.6)
(, t, r) = 2 2 (, t, r) = O c2
t r r
c t
which reduces to the Schrdinger equation in the c limit. Next, we rewrite the
generators of confd+2 . For brevity, we specialize to d = 1. For the translations, we find

P1 = icM0 ,
P0 = c M0 + O c2 ,
P1 = Y1/2 . (A.7)
For the rotations, we obtain

M01 = c Y1/2 + O c2 ,

M10 = iN + O c2 .
429

M11 = ic Y1/2 + O c2 ,
(A.8)
The dilatation becomes D = (2X0 + N) and for the special conformal transformations we find

K1 = 2ic X1 + O c2 ,
K0 = 2c X1 + O c2 ,

K1 = V+ + O c2 .
(A.9)
Therefore, having treated the masses as dynamical variables from the beginning, we
1 (and in general
rather have a projection of the complexified algebras (conf3 )C alt
(confd+2 )C altd ) when taking a non-relativistic limit. We stress that the non-relativistic
limit procedure actually throws out the time-translation operator X1 . On the other hand,
the operators V+ and N remain part of the algebra in the c limit. These results are
incompatible with Baruts claim that sch1 would be obtained in the non-relativistic limit,
since time translations belong to the Schrdinger group.
Appendix B
We derive the causal representations Eqs. (5.2) and (5.4). Because of rotation invariance
and since the a are scalars, it is enough to consider the case d = 1 explicitly. Beginning
with the two-point function, we introduce for the phase variables a center-of-mass
coordinates = 1 + 2 and relative coordinates = 1 2 . We then have

i
i
1
1 2 =
d d e 2 (M1 M2 ) e 2 (M1 +M2 ) 1 2
4
R2
= (M1 M2 )
d eiM1 1 2
= (M1 M2 )x1 ,x2 0 t x1 Mx11 1
= (M1 M2 )x1 ,x2 0 t

where
I=
R+i
du ux1 eiu
x1

iM1 r 2 x1
d ei +
2 t

M1 r 2
I,
2 t
Mx11 1 exp
(B.1)
(B.2)
M1 r 2
2 t
and r = r1 r2 , t = t1 t2 and 0 = 4x1 0 . The only singularity of the integrand is

the cut along the negative real axis. Provided the negative real axis is not crossed, the
430
integration contour can be arbitrarily shifted and, therefore, I = I (x1 ) depends only on the
2
1 r
sign of M
2 t .
1 r
Consider the case M
2 t < 0, which implies t < 0 because of the physical convention
M1 0. Then the contour of integration may be taken as R i with > 0 and can be
closed by a semi-circle in the lower half-plane. Using polar coordinates u = Rei , the
contribution Iinf of the lower semi-circle of radius R can be estimated in a standard fashion
2

|Iinf | R
1x1
R sin
d e
2R
1x1
/2
d e(2R/) R x1
(B.3)
and, therefore, vanishes as R , provided x1 > 0. It follows that I = 0 for t < 0 which
proves (5.2).
The three-point function is treated similarly. Introduce center-of-mass and relative
coordinates
= 2 3 ,
= 1 3 ,
Then

1 2 3
= (2)
3/2
= 1 + 2 + 3 .
(B.4)

d1 d2 d3 exp(iM1 1 iM2 2 + iM3 3 )1 2 3
R3
(M1 + M2 M3 )
=
d d eiM1 iM2 1 2 3 .
(B.5)
R2
Next, we set, using rab = ra rb and tab = ta tb

u= +i
2
r13
,
2t13
u = + i
2
r23
2t23
(B.6)
and
v=
2
r2
r2
r12
1 [r13 t23 r23 t13 ]2
+ 23 13 =
2t12 2t23 2t13 2
t12 t23 t13
(B.7)
and find

x
/2 x
/2 x
/2
1 2 3 = C12,3(M1 + M2 M3 ) t12 12,3 t23 23,1 t13 13,2

2
2
M1 r13
M2 r23
exp
(B.8)
I,
2 t13
2 t23
where C12,3 = iC12,3 2(x1+x2 +x3 1/2) / and

x /2 x /2

I=
du
du eiM1 uiM2 u u u + iv 12,3 u 12,3 ux13,2 /2 .
r2
R+i 2t13
13
r2
R+i 2t23
(B.9)
23
Without restriction of the generality, we can take t12 > 0. Otherwise, the roles of u and
u in the following discussion will be exchanged. The integrand in I has fixed cuts on the
431
Fig. 2. Integration contours in (B.9) if t23 > 0, indicated as oriented full lines. The thick black lines indicate the
fixed cuts on the negative real axis and the grey lines indicate the moving cuts which occur for u u + iv
2 /2t
real negative. On the left the contour is shown when one first integrates over u, with = r13
13 and
2 /2t v. On the right one integrates first over u , with a = r 2 /2t and b = r 2 /2t + v.
= r23
23
23
13
23
13
negative real axes of both u and u and a movable cut which arises if u u + iv is real
negative. From (B.7) follows the important inequality
v
2
r2
r23
13 .
2t23 2t13
(B.10)
First, we consider the case t23 > 0. Then t13 > 0 as well and from (B.7) one also has
v 0. In Fig. 2, we show the integration contours in the complex plane, when the integral
over u or u , respectively, is performed first. Because of the inequality (B.10), the contours
never cross the moving cuts. Therefore, if one integrates first over u, the contour can be
moved freely in the upper half plane above the movable cut and consequently, I must be
2 /2t . On the other hand, if one integrates first over u , the contour can
independent of r13
13
2 /2t . Therefore,
be moved freely between the singularities and I is independent of r23
23
I = I (v; M1 , M2 , x1 , x2 , x3 ) and we have the integral representation

x /2 x /2

I=
du
du eiM1 uiM2 u u u + iv 12,3 u 23,1 ux13,2 /2 (B.11)
R+i
R+i
,
with
0, at least if v > 0.
Second, we consider the case t23 < 0. We shall show that I = 0 provided x2 > 0. It
is convenient to carry out first the integration over u . The contour together with the cuts
is shown in Fig. 3. The position of the contour relative to the cuts follows from (B.10)
and allows one to close the contour in the lower half plane (the relative position of the
cuts depends on the values of the coordinates but does not matter). We now show that the
integral

x y iu
du + i u
u e
J=
(B.12)
R+i
vanishes for all real values of , and such that > and if x + y > 0. That will
imply
I = 0 under the stated conditions. In order to see this, consider the contour integral

( + i u )x u y e iu The contour C runs along R + i and is closed by a
du
C
semi-circle of radius R in the lower half-plane. The condition > ensures that the cuts
432
Fig. 3. Integration contour in (B.9) if t23 < 0. The notation is the same as in Fig. 2.
lie on the outside of C and, therefore, the contour integral vanishes. We introduce polar
coordinates u = Rei and estimate the contribution Jinf of the lower arc

|Jinf | R
1xy
d eR sin B x/2 ,
(B.13)
where B is defined below. Because of the estimate

+ i i
|| 2
i i
|| 2
B := 1
e
e
+ 1+
3
1
1+
R
R
R
R
(B.14)
which holds for R sufficiently large, we have
|Jinf | 3x/2R 1xy
d eR sin 3x/2R xy
(B.15)
which tends to zero as R provided x + y > 0, hence, J = 0. Finally, the exponents

x, y in J are related to the scaling dimensions x + y = (x12,3 + x23,1)/2 = x2 > 0.
In conclusion, the integral I vanishes if t12 > 0 and t23 < 0 and will therefore contain
a factor (t23 ). By symmetry between t1 and t2 , the case t12 < 0 will produce a factor
(t13 ) provided x1 > 0. The scaling function 12,3 (v) in Eq. (5.4) can be identified with
the integral I in (B.11) and the assertion is proven.
Appendix C. On parabolic subalgebras

We recall here the definition of parabolic subalgebras of a complex simple Lie algebra g,
see [40]. The presentation is somewhat simpler than in the general case since for complex
g, the compact part and the non-compact part of the Cartan decomposition may be chosen
to be the same up to multiplication by i.
433
Let g be a complex simple Lie algebra of rank r, h a complex Cartan subalgebra of g

and the associated root system. One chooses a basis of simple roots = {1 , . . . , r }
root, then g g
and denotes by + the associated set of positive roots. If is a
will be the corresponding root space. We shall also use the notation n = + g for the
subalgebra of g made up of all positive root spaces.
The minimal standard parabolic subalgebra s0 of g (associated with the given choice
of h and + ) is defined as
s0 = h n = h
g .
(C.1)
A standard parabolic subalgebra of g is a subalgebra s g containing s0 . More

generally, a parabolic subalgebra is defined to be a subalgebra of g containing a conjugate
of s0 .
The standard parabolic subalgebras s of g can be classified by means of a powerful
result: they are in one-to-one correspondence with the subsets s of . Given s ,
here is how one constructs the associated parabolic subalgebra s. Let hs be the set of H h
such that (H ) = 0 for all s , or in other words the orthogonal in h of s ; ms the
centralizer of hs in g, that is, the set of elements in g that commute with all elements in hs ;
finally ns n the sum of all root spaces associated with positive roots + that are not
identically zero on hs . Then hs , ms and ns are subalgebras of g, with hs h ms , and s
is the direct sum
s = ms ns .
(C.2)
It is easily checked from the above definitions that Eq. (C.2) yields a Lie subalgebra of g,
and that is in s if and only if g ms in which case g is also included in ms .
So, in particular, one also sees quite easily that s is indeed a standard parabolic subalgebra.
Since the standard parabolic subalgebras are non-conjugate, it is straightforward to see how
this classification extends to a classification of all parabolic subalgebras.
The main motivation for the definition of parabolic subalgebras is that, if G is a simple
group with Lie algebra g, the pieces that appear in the Plancherel formula for G all come
from representations induced from certain standard parabolic subgroups of G (integrating
standard parabolic subalgebras).
The case g = sl(r + 1, C) is particularly illuminating. Take h to be the space
of diagonal matrices with vanishing trace. Let 1 = diag(1, 1, 0, . . . , 0), . . . , r =
diag(0, . . . , 0, 1, 1) form a basis of the set of roots, with the usual identification of h
with its dual. We choose a subset i1 , . . . , is (s r) of . One says that i connects with j
(1 i < j r + 1) if i , i+1 , . . . , j 1 s . This defines the connected components of
the set {1, . . . , r + 1}. Then diag(x1 , . . . , xr+1 ) hs if and only if x1 + + xr+1 = 0 and
xi = xj whenever i and j are connected. Therefore, a diagonal matrix H with vanishing
trace is in hs if and only if its entries situated in a single connected component of
{1, . . . , r + 1} are all equal. So, ms is the set of block-diagonal matrices with vanishing
trace that do not mix the different components.
434
Table 1
Construction of the standard parabolic subalgebras s of the complex Lie algebra g = (conf3 )C
s
hs
ms
ns
{1 }
CN
h CY1/2 CV+
CX1 CM0 CY1/2
1
age
1
alt
{2 }
{1 , 2 }
C(N D)
{0}
h CX1 CX1
g
CY1/2 CY1/2 CM0

{0}
1
sch
(conf3 )C
For illustration, take the example r = 4 and s = {1 , 3 , 4 }. An element H hs can

be written as a diagonal matrix
x1
x1
with 2x1 + 3x2 = 0.

H =
(C.3)
x2
,
x2
x2
In turn, elements M ms and N ns can be written as

M1 0
0 N1
M=
,
N=
,
0 M2
0 0
(C.4)
where M1 , M2 and N1 are, respectively, 2 2, 3 3 and 2 3 matrices and tr M1 + tr M2 =

0.
It is fairly easy to find the standard parabolic subalgebras in the case studied in Section 3,
namely, g = so(5, C). Referring to the notations of that section, the chosen basis of simple
roots is
1 = e2 ,
2 = e1 + e2 .
(C.5)
In Table 1 we list all possible s , hs , ms , ns and s.

Let us explain how to find the entries in Table 1 in the case where s = {1 } = {e2 }.
Then
!

hs = N + D ! 1 (N + D) = 0 = {N} = CN,
(C.6)
where we recall from Section 3 that ei (N) = i,1 and ei (D) = i,2 . So
!

ms = X g ! [X, hs ] = 0 = h ge2 ge2 = h CY1/2 CV+ .
(C.7)
To find ns , we must determine all positive roots which vanish on hs . In this case, the only
such positive root is e2 . Consequently, with the set + of positive roots given by (3.8)
ns = ge1 +e2 ge1 ge1 e2 = CX1 CY1/2 CM0 .
References
[1] A.O. Barut, Helv. Phys. Acta 46 (1973) 496.
[2] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
(C.8)
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
435
O. Bergman, Phys. Rev. D 46 (1992) 5474.

A.J. Bray, Adv. Phys. 43 (1994) 357.
G. Burdet, M. Perrin, Lett. Nuovo Cimento 4 (1972) 651.
P. Calabrese, A. Gambassi, Phys. Rev. E 65 (2002) 066120.
C.G. Callan, S. Coleman, R. Jackiw, Ann. Phys. 59 (1970) 42.
S.A. Cannas, D.A. Stariolo, F.A. Tamarit, Physica A 294 (2001) 362.
M.E. Cates, M.R. Evans (Eds.), Soft and Fragile Matter, Proceedings 53rd Scottish University Summer
Schools in Physics, IOP, Bristol, 2000.
L.F. Cugliandolo, J. Kurchan, G. Parisi, J. Physique I 4 (1994) 1641.
P. di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, Heidelberg, 1997.
C. Duval, in: H. Doebner (Ed.), Conformal Groups and Related Symmetries, in: Lecture Notes in Physics,
Vol. 261, Springer-Verlag, Berlin, 1986, p. 162.
C. Duval, G. Gibbons, P. Horvathy, Phys. Rev. D 43 (1991) 3907.
L.. Gergely, Ann. Phys. 298 (2002) 394.
P.K. Ghosh, Phys. Rev. A 65 (2002) 012103.
D. Giulini, Ann. Phys. 249 (1996) 222.
D. Giulini, quant-ph/9906108.
C. Godrche, J.M. Luck, J. Phys. A 33 (2000) 1151.
C. Godrche, J.M. Luck, J. Phys. A 33 (2000) 9141.
C. Godrche, J.M. Luck, J. Phys.: Condens. Matter 14 (2000) 1589.
C.R. Hagen, Phys. Rev. D 5 (1972) 377.
C.R. Hagen, Phys. Rev. D 5 (1972) 389.
M. Hassane, P.A. Horvathy, Ann. Phys. 282 (2000) 218.
T. Haugset, F. Ravndal, Phys. Rev. D 49 (1994) 4299.
M. Henkel, Int. J. Mod. Phys. C 3 (1992) 1011.
M. Henkel, J. Stat. Phys. 75 (1994) 1023.
M. Henkel, G.M. Schtz, Int. J. Mod. Phys. B 8 (1994) 3487.
M. Henkel, Phys. Rev. Lett. 78 (1997) 1940.
M. Henkel, Phase Transitions and Conformal Invariance, Springer-Verlag, Heidelberg, 1999.
M. Henkel, M. Pleimling, C. Godrche, J.-M. Luck, Phys. Rev. Lett. 87 (2001) 265701.
M. Henkel, Nucl. Phys. B 641 (2002) 405.
M. Henkel, M. Paessens, M. Pleimling, cond-mat/0211583, Europhys. Lett. (2003), in press.
M. Henkel, M. Pleimling, cond-mat/0302482.
F.J. Herranz, M. Santander, J. Phys. A 35 (2002) 6601.
E.V. Ivashkevich, J. Phys. A 30 (1997) L525.
R. Jackiw, Phys. Today 25 (1) (1972) 23.
R. Jackiw, Ann. Phys. 201 (1990) 83.
R. Jackiw, S.-Y. Pi, Phys. Rev. D 42 (1990) 3500;
R. Jackiw, S.-Y. Pi, Phys. Rev. D 48 (1993) 3929, Erratum.
O. Jahn, V.V. Sreedhar, math-ph/0102011.
A.W. Knapp, Representation Theory of Semisimple Groups: An Overview Based on Examples, Princeton
Univ. Press, Princeton, NJ, 1986.
S. Lie, Arch. Math. Nat. Vid. (Kristiania) 6 (1882) 328.
P.C. Martin, E.D. Siggia, H.H. Rose, Phys. Rev. A 8 (1973) 423.
T. Mehen, I.W. Stewart, M.R. Wise, Phys. Lett. B 474 (2000) 145.
U. Niederer, Helv. Phys. Acta 45 (1972) 802.
L. ORaifeartaigh, V.V. Sreedhar, Ann. Phys. 293 (2001) 215.
M. Perroud, Helv. Phys. Acta 50 (1977) 233.
A. Picone, M. Henkel, J. Phys. A 35 (2002) 5575.
A. Picone, Thse de doctorat.
M. Pleimling, M. Henkel, Phys. Rev. Lett. 87 (2001) 125702.
A.M. Polyakov, Sov. Phys. JETP Lett. 12 (1970) 381.

Organizing boundary RG flows

Stefan Fredenhagen
CPHT Ecole Polytechnique, F-91128 Palaiseau Cedex, France
Abstract
We show how a large class of boundary RG flows in two-dimensional conformal field theories
can be summarized in a single rule. This rule is a generalization of the absorption of the boundary
spin-principle of Affleck and Ludwig and applies to all theories which have a description as a coset
model. We give a formulation for coset models with arbitrary modular invariant partition function
and present evidence for the conjectured rule. The second half of the article contains an illustrated
section of examples where the rule is applied to unitary minimal models of the A- and D-series, in
particular, the 3-state Potts model, and to parafermion theories. We demonstrate how the rule can be
used to compute brane charge groups in the example of N = 2 minimal models.
1. Introduction
The study of renormalization group (RG) flows in two-dimensional quantum field
theories is an important subject in condensed matter physics and statistical mechanics,
and it also plays a vital role in string theory. In systems with boundaries or defects, there
are flows generated by boundary fields which only affect the boundary condition and leave
the theory in the bulk unchanged. In string theory, such flows describe the dynamics of
D-branes in a given closed string background.
How do we find boundary RG flows? For a given boundary perturbation of a boundary
conformal field theory (BCFT), we have various tools at our disposal. In some cases when
we perturb by a field which is only slightly relevant, we can apply perturbation theory [1].
If the perturbation is integrable, we may use exact integral equation techniques (like the
thermodynamic Bethe ansatz). A method which can always be employed is the truncated
E-mail address: stefan@aei.mpg.de (S. Fredenhagen).

doi:10.1016/S0550-3213(03)00226-8
S. Fredenhagen / Nuclear Physics B 660 [FS] (2003) 436472
437
conformal space approach where we truncate the Hilbert space to a finite dimensional space
and compute RG flows numerically.
All these tools have helped to get a substantial knowledge about boundary RG flows.
To organize the informations, we need general, model-independent principles. One such
principle is the g-conjecture of Affleck and Ludwig [1] which states that the boundary
entropy g always decreases along a RG flow. Although very important, the g-conjecture
is not a constructive principle: it only tells us which flows are possible and which are not.
In the case of WZNW models, we have a constructive principle at hand, namely, the
absorption of the boundary spin-principle of Affleck and Ludwig [2]. This rule is easy to
formulate and describes a large class of flows.
A generalization of this rule to fixed-point free coset models was proposed in [3].
The formulation there was for coset models with a charge-conjugated modular invariant
partition function and boundary conditions of Cardy type. Here, we shall present a
formulation that is applicable for all maximally symmetric boundary conditions in coset
models with any modular invariant. Furthermore we shall work out some arguments
supporting the proposal, and employ the rule in a number of examples.
The structure of the paper is as follows: we start with an introduction to coset models and
their maximally symmetric boundary conditions in Section 2. Although this is essentially
a review of [4,5], we hope to clarify the role of the modular invariant when we relate coset
boundary conditions to those of WZNW models. In Section 3, we formulate the absorption
of boundary spin-principle and its generalization to coset models and discuss the relation
to perturbative calculations and the compatibility with the g-conjecture. Through a number
of examples, we present the rule at work in Section 4. We shall make extensive use of a
geometric interpretation of the boundary conditions as branes to visualize the RG flows.
Whenever we are aware of results on boundary RG flows in the specific examples, we
compare them to the predictions of the rule. At the end of Section 4 the rule is used to
determine the charge group of branes in N = 2 minimal models. In the appendix we collect
the complete results for the critical and tricritical Ising model as well as the 3-state Potts
model.
2. Boundary conditions in coset models

The coset construction [6] allows to access a great variety of rational conformal field
theories. Boundary conditions in these models have been investigated in the past. Most
of the work was concentrated on maximally symmetric boundary conditions, i.e., those
where the boundary conformal field theory admits the action of the coset chiral algebra
Untwisted boundary conditions in coset models with charge-conjugated modular
g /h.
invariant partition function are already covered by the seminal paper of Cardy on boundary
conditions in rational CFTs [7]. The generalization to twisted boundary conditions and
more general modular invariants has been worked out in [4,5]. Symmetry breaking
boundary conditions in coset models have been first considered in [8,9] relying on previous
work in WZNW models [10,11]. In the -model approach, boundary conditions in gauged
WZNW-models have been studied in [1214]. Recently there has been also some work on
boundary conditions in asymmetric cosets [9,15,16].
438
We give a short introduction to coset models to set up our notation. Subsequently, we

discuss some general properties of maximally symmetric boundary conditions. The section
ends with a review on the construction of boundary states from known boundary conditions
in the product theory with chiral algebra g h.

2.1. Coset construction
From now on let h g denote some simple subalgebra of the simple Lie algebra g
(the generalization to semi-simple Lie algebras is straightforward). We want to study the
associated g /h coset model. A more precise formulation of this theory requires a bit of
preparation (more details can be found, e.g., in [17]).
Induced from the embedding of h in g, there is an embedding of the affine Lie algebra
h k into g k . The level k is related to k by the embedding index xe , k = kxe . We shall label

the sectors Hhl of the affine Lie algebra h k with labels l Rep(h k ). Note that the sectors
of the numerator theory carry an action of the denominator algebra h k g k and under this
action each sector Hgl decomposes according to

Hgl =
(1)
H(l,l ) Hhl .
l

Here we have introduced the infinite dimensional spaces H(l,l ) which we want to interpret
as sectors of the coset chiral algebra. The latter is usually hard to describe explicitly, but at
least it is known to contain a Virasoro field with modes
Ln = Lgn Lhn .
(2)
One may easily check that they obey the usual exchange relations of the Virasoro algebra
with central charge given by c = cg ch .

Note that some of the spaces H(l,l ) may be trivial simply because a given sector Hhl
of the denominator theory may not appear as a subsector in a given Hgl . This allows to
introduce the set

E = (l, l ) Rep(gk ) Rep(h k ) | H(l,l ) = 0 .
Furthermore, some of the coset spaces labeled by different pairs (l, l ) and (m, m )
correspond to the same sector of the coset theory. Therefore, we label coset sectors by
equivalence classes [l, l ] of pairs.
There is an elegant formalism to describe these selection and identification rules which
is applicable in almost all coset models.1 It involves the so-called identification group Gid
which contains pairs (J , J ) of simple currents. It is a subgroup of the direct product of
the simple current groups of g k and h k . A simple current J of g k is an element in Rep(gk )
which has the property that the fusion product of J with any other representation l contains
1 The known exceptions all appear at low levels of the involved affine Lie algebras, see, e.g., the Maverick
cosets [18].
439
exactly one sector m =: J l Rep(gk ),

NJ l m = m,J l .
To formulate the selection rules in coset models, we introduce the monodromy charge
QJ (l) of l with respect to J in terms of conformal weights,
QJ (l) = hJ + hl hJ l mod Z.
The monodromy charge appears when a simple current J acts on the modular S-matrix,
SJ lm = e2iQJ (m) Slm .
We are now prepared to formulate selection and identification rules in terms of the
identification group Gid of simple currents:
A pair (l, l ) is allowed, i.e., (l, l ) E, if QJ (l) = QJ (l ) for all (J , J ) Gid .
Two pairs (l, l ) and (J l, J l ) label the same sector, i.e.,

H(l,l )
= H(J l,J
l)
At this point we want to make one assumption, namely, that all the equivalence classes we
find in E contain the same number N0 = |Gid | of elements, in other words, Gid acts fixedpoint free. This holds true for many important examples and it guarantees that the sectors
= E/Gid .
of the coset theory are simply labeled by the equivalence classes,2 i.e., Rep(g/h)
It is then also easy to spell out explicit formulas for the fusion rules and the S-matrix of the
coset model. These are given by

g
h
N[j,j ][k,k ] [l,l ] =
(3)
Nj k m Nj k m ,
(m,m )(l,l )
S[l,l ][m,m ] = N0 Slm Sl m ,

g
(4)
where the bar over the second S-matrix denotes complex conjugation.
2.2. Maximally symmetric boundary conditions
Let us turn now to coset models with a boundary and summarize some general properties
of maximally symmetric boundary conditions to be prepared for the concrete analysis in
Section 2.3.
We want to impose conditions along the boundary gluing left- and right-moving fields
together with a suitable automorphism of the coset chiral algebra. The corresponding set
.
of elementary boundary conditions is denoted by Bg/h
2 For more general cases, there are further sectors that cannot be constructed within the sectors of the
numerator theory.
440
Because of the specific gluing conditions, the annulus partition function involving the
decomposes into coset characters,
boundary conditions , Bg/h
Z (q) =
n[l,l ] (l,l ) (q)
[l,l ]
, these
with non-negative integers n[l,l ] . For a complete set of boundary conditions Bg/h
numbers are known to form a representation of the fusion algebra [19,20],

n[l,l ] n[j,j ] =
N[l,l ][j,j ] [k,k ] n[k,k ] .
[k,k ]
The integers n have the further properties

n[0,0] =
and
n[l,l ] = n[l + ,l + ] ,
where l + labels the representation conjugate to l.
2.3. Boundary conditions from WZNW models
In the last subsection we have been rather general. It is possible to relate the analysis
of boundary conditions in coset models to the investigation of boundary conditions in the
product theory with chiral algebra g h [4,5,21]. Before we enter the detailed description,
let us sketch our general procedure: we specify a modular-invariant partition function of
the coset model and from that we construct a partition function for the product theory.
In the resulting theory we impose gluing conditions involving a gluing automorphism
g (h )1 where we assume that g restricts to h and h = g |h . The corresponding
boundary conditions of the product theory can be projected to boundary conditions in the
coset model by certain selection rules:
partition function Z g/h
partition function Z gh
gluing
automorphism

g h 1
set of boundary conditions
Bg/h
selection
rules
set of boundary conditions
Bgh
Let us become more specific. We start with a coset model with the partition function

Z g/h (q, q)
=
Z[l,l ],[m,m ] [l,l ] (q) [m,m ] (q)
[l,l ],[m,m ]
441
with some non-negative integers Z[l,l ],[m,m ] . To this model we associate a product theory
g h with partition function (following [22])

=
Z[l,l ],[m,m ] l (q) m (q) m (q)
l (q).
Z gh (q, q)
l,l ,m,m
Note the exchange of the h-labels l and m which is necessary to guarantee modular
invariance of the product theory.
In this theory we want to analyze maximally symmetric boundary conditions. To this
end, we glue the left- and right-moving currents J (z), J(z) of the g and h theory with a
gluing automorphism induced from g (h )1 along the boundary,
J (z) = (J)(z),
for z = z .
Assume now that we have solved the problem of finding all maximally symmetric
boundary conditions in the theory, i.e., we have a set of boundary conditions Bgh
(l,l ;)
specified by the boundary couplings

| =
. The corresponding boundary state is
(l,l ;)

(l, l ; ) .
Sl0 Sl 0

(l,l ;)
Here, (l, l ; ) labels an Ishibashi state in the sector (l, l ) and is an additional multiplicity
index in the range
= 1, . . . , Z[l,1 (l + )][(l + ),l ] .
At this point we want to make an important assumption. We assume that we can find a
basis of Ishibashi states s.t. the action of a simple current on the boundary couplings is
given by a pure phase factor which only depends on ,
(J l,(J
+ )l ;)
= e2iQ(J ,(J + )) () (l,l ;) ,
for (J , J ) Gid .
This is certainly true in all examples that we considered, but it is unclear whether this
assumption holds in general (this problem has already been mentioned in [4,5]).
We are now prepared to write down a set of boundary conditions for the coset model.
satisfying the selection rule

For any Bgh
Q(J ,(J + )) () = 0,
for all (J , J ) Gid ,
(5)
we define a boundary condition in the coset model which we also label by ,

+
([l,l ];) := N0 (l,(l );).
[l, l ]
(6)
The Ishibashi states of the coset model are labeled by equivalence classes
of pairs
together with a multiplicity index running from 1 to Z[l,l ],[(l + ),(l + )] . Note that the
multiplicity of the coset Ishibashi state [l, l ] and the product Ishibashi state (l, (l + ))
coincide so that we can use the same label on both sides of Eq. (6).
([l,l ];)
fulfill the completeness conditions
It is straightforward to verify that the

([l,l ];) ([m,m ];) = [l,l ],[m,m ] ,
(7)
442
and

([l,l ];)
([l,l ];)
= , .
(8)
([l,l ];)
Furthermore, Cardys condition which says that the annulus coefficients

g/h
n[l,l ]
([j,j ];)
([j,j ];) g/h

S[l,l ][j,j ]
g/h
S[0,0][j,j ]
([j,j ];)
are non-negative integers, is satisfied, and

g/h
gh
n[l,l ] = n(l,(l )) .
(9)
2.4. The factorizing case

In this section we want to deal with the situation of a factorizing modular invariant
partition function in the coset theory. We should state more clearly what we mean by
id
factorizing, namely, that the associated modular invariant of the product theory is a G
simple current orbifold of a direct product of a g and a h modular invariant,

g
h
Zl J m Zl J m l (q) l (q) m (q)
m (q),
Z gh =
id
(J ,J )G
l,l ,m,m
QJ (l)+QJ (l )=0
id = {(J , J )|(J , J 1 ) Gid }. The boundary conditions B in the simple

where G
gh
current orbifold can be obtained from pairs (L, L ) of boundary conditions L Bg and
L Bh of the g and h theory, respectively. These pairs are subject to identification
rules, and fixed-points can occur (even if Gid acts fixed-point free on the sectors). These
orbifold fixed-points can be easily resolved.3 We shall label the boundary conditions by
equivalence classes of pairs = [L, L ], always remembering the possible orbifold fixedpoint resolution.
When we want to obtain boundary conditions in the coset model along the lines of
Section 2.3, we only have to impose in addition the selection rules (5) on the boundary
conditions = [L, L ].
We arrive at the final conclusion thatin the factorizing caseboundary conditions
of the coset model are obtained from pairs of boundary conditions of numerator and
denominator theory by suitable identification and selection rules. This result has been
formulated first in [4]. There, the authors took a direct way not involving boundary
conditions in the simple current orbifold. In practice this can simplify things: it is not
always necessary to do the fixed-point resolution in the orbifold step, because it may
happen that many of the resolved boundary conditions do not survive the selection rules.
Still we think that the detour via the simple current orbifold has conceptual advantages.
Firstly, it shifts all problems with fixed-point resolution to the orbifold step. Secondly, it
3 In string theory these resolved boundary conditions are called fractional branes.
443
fits in the general framework of Section 2.3 which is also applicable in the non-factorizing
case.
In all our examples in Section 4, we shall encounter the factorizing case. The
formulation of the rule for boundary RG flows in Section 3, however, is more general
and can be also used in the non-factorizing case.
3. RG flows: a simple rule

3.1. Generalized AffleckLudwig rule
To motivate the rule for boundary RG flows in coset models, we shall first review shortly
the original proposal of Affleck and Ludwig for the absorption of the boundary spin in the
Kondo model.
The Kondo model is designed to understand the effect of magnetic impurities on the
low temperature conductivity of a conductor. Usually a decreasing temperature will result
in an increasing conductivity, because the scattering with phonons is reduced (Matthiesens
rule). In some cases, however, when magnetic impurities are present, the conductivity
reaches a maximum and starts to decrease again. This phenomenon is explained by the
coupling of the electrons to the magnetic impurities. The electrons tend to screen the
impurity, and this coupling increases when temperatures become low.
Let us say that the conductor has electrons in k conduction bands. We can build several
currents from the basic fermionic fields: the charge current, the flavor current, and the spin
current J(y). The latter gives rise to a su(2)k current algebra. The coordinate y measures
the radial distance from a spin S impurity at y = 0 to which the spin current couples.4 This
coupling is
Hpert = R J (0),
(10)
where R ( = 1, 2, 3) is a (2S + 1)-dimensional irreducible representation of su(2), is

the coupling constant.
The operator Hpert acts on the tensor product V S H of the (2S + 1)-dimensional
quantum mechanical state space of our impurity with the Hilbert space H for the
unperturbed theory described by a Hamiltonian H0 .
When the boundary spin is large (2S > k), the low temperature fixed point of the Kondo
model appears only at infinite values of (under-screening). On the other hand, the fixed
point is reached at a finite value = of the renormalized coupling constant if 2S k
(exact- or over-screening, respectively). In the latter case, the fixed points are described
by non-trivial (interacting) conformal field theories. Affleck and Ludwig [2,23] found an
elegant rule to determine these strong-coupling fixed-points. The spectrum at the fixedpoint is given by

ren
NSj l j (q).
trV S Hl q H0 +Hpert = =
(11)
j
4 We only consider the case of a single isolated impurity.
444
Here, H0 = L0 + c/24 is the unperturbed Hamiltonian, and the superscript ren stands for
renormalized. By S we label a dominant highest-weight representation of su(2). V S
denotes the corresponding module of the finite dimensional Lie algebra su(2), and Hl is an
irreducible sector of the su(2)k -theory. The formula (11) is the content of the absorption
of boundary spin-principle by Affleck and Ludwig [2,23].
It is straightforward to generalize these considerations to an arbitrary simple Lie algebra
g. The space Hl can be any of the g k -irreducible subspaces in the physical state space H of
the theory. Formula (11) means that our perturbation with some irreducible representation
S interpolates continuously between a building block dim(V S ) l (q) of the partition
function of the UV-fixed point (i.e., = 0) and the sum of characters on the right-hand
side of the previous formula,

dim V S l (q)
(12)
NSj l j (q).
j
In [3] it was proposed to generalize the absorption of boundary spin-principle to coset

models. The suggested rule is

(13)
bS + S NS l j (l,l ) (q)
NSj l (j,j ) (q).
S ,l
respectively.
Here, S, l and j label dominant highest-weight representations of g and h,
The coefficients bSS are the branching coefficients describing the decomposition of V S ,

the corresponding representation of the finite Lie algebra g, into representations V S of h,

VS =
(14)
bSS V S .
The embedding of affine Lie algebras h g guarantees that these representations can again

be identified with highest-weight representations HS of h.

The flows (13) are generated by fields coming from the coset sectors

H(0,l ) , where V l V h .
(15)
Here, labels the integrable highest-weight representation which is built from the adjoint
representation of the Lie algebra g. The adjoint representation l = of h can be omitted
from the list (15) if it occurs only once in the decomposition of .
To see that (13) is really a generalization of (12), we should recover the flows (12) when
specializing to the trivial subgroup {e} of g. The primed label can then be omitted and the
branching coefficient is just the dimension of the representation space V S .
3.2. Simple rule for boundary RG flows in coset models
From the stated rule (13) formulated in terms of characters we can infer a rule for a flow
between superpositions of boundary conditions. Before we discuss how this is done, let us
present the result.
s.t.
Choose a representation S Rep(g) and a boundary condition Bgh
Q(J ,(J + )) () + Q(J ,(J + )) (S, 0) = 0,
for all (J , J ) Gid .
445
Then there will be a RG flow between the following coset boundary configurations X
and Y ,

(S, 0)
=: Y.
X := 0, S + h
(16)
to define a superposition of the form
Here, we introduced the shorthand notation
gh

:=
l, l
n(l,l ) ().
The label (0, S + |h ) has to be understood as

:=

bS + S (0, S )
0, S + h
where bS + S are the finite branching coefficients defined in (14). The flow (16) is generated
by fields from the coset sectors (15).
To derive (16) from (13), we introduce an arbitrary spectator boundary condition .
The annulus partition function ZX (q) can be decomposed into combinations of characters
that appear on the left side of (13) just using the fact that the annulus coefficients form a
representation of the fusion algebra. The expression that we obtain from applying (13) can
then be rewritten as ZY , i.e., we find the result
ZX ZY ,
for arbitrary boundary conditions .
In the remainder of this section we want to give two arguments to support our claim.
First, we want to relate the rule to results from a perturbative analysis in the limit when
some levels are large. Then we shall present evidence that the rule is compatible with the
g-conjecture of Affleck and Ludwig.
For a general coset theory with semi-simple numerator and denominator there occur
different levels kr for the simple constituents of the numerator which then determine the
levels in the denominator. Assume that we take some of the levels to very large values of
the order of a common scale k 1. In the limit k , there are many coset fields whose
conformal weight approaches h = 1 (the difference to 1 being of the order 1/k). The RG
flows induced by such fields can be studied by perturbative techniques. One way is to use
the method of effective actions. Here, the couplings of the boundary fields are combined
into matrices A which are interpreted as fields in an effective theory determined by an
action S[A]. The equations of motion for A are precisely the fixed-point equations = 0.
In [21,24] the effective action for untwisted boundary conditions in coset models has been
constructed to leading order in 1/k building upon earlier works in WZNW models [25].
The generalization to twisted boundary conditions has been worked out in [26] using
results of [27].
A special class of solutions for all, untwisted and twisted, boundary conditions [26] has
precisely the form (16),

(S, 0),
0, S + h
446

g
but here we have to restrict S to representations s.t. the conformal weight hS in the g -theory
is of order 1/k. The rule (16) thus extrapolates the perturbative results to arbitrary values
of the levels.
The g-conjecture of Affleck and Ludwig states that the boundary entropy g decreases
along a boundary RG flow X Y ,
gX > gY .

The boundary entropy gX for a superposition X = X (with X N0 ) of boundary
conditions occurring with multiplicity X is defined as the sum of the g-factors of the
single boundary conditions,
gX =
X g =
0
X .
S00
The ratio gX /gY for the conjectured flow (16) is given by

gX
=
gY
gh
S bS + S n(0,S ) g
gh
n(S,0) g
gh 0
S bS + S n(0,S )
.
gh 0
n(S,0)
We simplify this expression by using the fact that the vector ( 0 ) is an eigenvector of the
matrix (n(l,l ) ) with eigenvalue S(l,l )0 /S00 and obtain a result which does only depend
on S,
gX
=
gY
g h
S bSS S00 SS 0
.
g h
SS0 S00
Hence, if our conjectured rule (16) and the g-conjecture are correct, we obtain the
following inequality for quantum dimensions of g and h (S = 0),

S
SS
SS 0
h
S00
>
SS0
g
S00
(17)
This inequality can be used to test our proposal.

For diagonal cosets g k g l /gk+l the inequality is satisfied. This follows from the
g
g
fact that the quantum dimension SS0k /S00k of a fixed representation S is a monotonically
increasing function of the level k. Unfortunately, for general coset models we have not
found a proof yet. However, numerical checks have been performed in a large number of
coset models, all in accordance with the conjecture. Furthermore, when we take some levels
to be large, we can confirm the inequality in a perturbative calculation (see Appendix A).
This ends our discussion of the general properties of the rule (16). We have given some
evidence by showing that the rule is consistent with the perturbative results and, although
not completely proven, with the g-conjecture. The next section will provide more evidence
coming from specific examples where the rule is able to reproduce a number of known
flows.
447
4. Examples
In this section we shall present the rule (16) at work in a number of examples. In all these
examples we shall first give the field content of the theory (coset sectors) and the boundary
conditions by specifying identification and selection rules. Then we shall formulate the
annulus coefficients for the coset boundary theories. The boundary conditions for the
associated product theory are obtained by forgetting the selection rules on them, and the
corresponding annulus coefficients are related to the ones from the coset model by (9).
We shall introduce a pictorial representation of the boundary conditions as branes in some
target space. This is followed by an application of the rule (16) to identify flows which
are visualized as brane processes. For some models we collected the complete results in
Appendix B.
4.1. Minimal models, A series
The unitary minimal models can be constructed as diagonal cosets of the form su(2)k
su(2)1 /
su(2)k+1 with an integer k 1. The modular invariant partition functions for these
models are completely classified [2830] (see also [17]); in this subsection we shall deal
with the A series which is sometimes denoted as (Ak+1 , Ak+2 ).
The sectors of the theory are labeled by three integers [l1 , l2 , l ] in the range l1 =
0, . . . , k; l2 = 0, 1; l = 0, . . . , k + 1. Selection rules force the sum l1 + l2 + l to be even,
and there is an identification [l1 , l2 , l ] [k l1 , 1 l2 , k + 1 l ] between admissible
labels.5 The adjoint field from the sector [0, 0; 2] that induces the flow described by the
rule (16) has conformal weight h = (k + 1)/(k + 3).
In the A-series we are in the Cardy case, i.e., the boundary conditions are labeled by
triples [L1 , L2 , L ] taking values in the same range as the sectors including selection and
identification rules.
The annulus coefficients are just given by the fusion rules N (k) and N (k+1) of su(2)k
and su(2)k+1 respectively,

(k+1) J
n[l1 ,l2 ,l ][L1 ,L2 ,L ] [J1 ,J2 ,J ] = Nl1 L1 J1 Nl L

(k)
(k)
(k+1)
+ Nkl1 L1 J1 Nk+1l L J .
We now want to give a pictorial representation of the boundary conditions. Coset

models can also be formulated as non-linear -models on a background geometry which is
essentially given by the space G/Ad H where we divide the group G by the adjoint action
of the subgroup H . The boundary conditions can then be described by certain subspaces
(branes) onto which the boundary of our two-dimensional world-sheet is mapped [12,13].
One should be aware that this geometrical interpretation is only valid for large values of
the level. If one, however, views the pictures just as a nice tool to illustrate the boundary
conditions, we can profitably employ them for arbitrary levels.
In the case of minimal models we describe the background as a solid cylinder with
squeezed ends [21]. The boundary conditions are represented by branes, extended objects
of dimension 0,1 and 2. Let x be the coordinate along the axis of the cylinder, z the
5 The relation to the usual Kac labels (r, s) is r = l + 1 and s = l + 1.
1
448
Fig. 1. The geometrical representation of boundary conditions in the minimal model A-series.
coordinate along the squeezed ends and y a third coordinate perpendicular to the others
(see Fig. 1). All branes [L1 , L2 , L ] are located along surfaces of constant z and are
maximally extended in the y-direction. In x they stretch between two values xmin and
xmax . The boundary conditions with L = 0, k + 1 are represented as points at the top or
bottom of the cylinder, the ones with L = 0, k are one-dimensional objects stretching in
y-direction (see Fig. 1). We shall give explicit formulas for z, xmin , xmax . The coordinate z
lies in the range [0, 1], and the coordinate x takes values in [0, k],

L
,
for L2 = 1,
z = k+1 L
1 k+1 for L2 = 0,

k

L ,
xmin = L1
k+1

k
k

xmax = max L1 +
L , 2k L1
L .
k+1
k+1
The rule (16) describes a large number of flows for many different starting configurations. We shall concentrate here on two types of starting points: a single boundary condition
and a superposition of boundary conditions of the form [0, L2 , L ].
Assume that we want to study flows starting from the boundary condition [L1 , L2 , L ]
with 1 L k. To apply our rule (16), we have to find a boundary condition and a
boundary spin S s.t.

= [L1 , L2 , L ].
0, S + h
On the one hand we can set = [L1 , L2 , 0] and S = (L , 0). This corresponds to the flow
(k)
[L1 , L2 , L ]
(18)
NL1 L J [J, L2 , 0].
J
On the other hand, the choice = [k L1 , 1 L2 , 0] and S = (k + 1 L , 0) leads to the

flow
(k)
NL1 L 1 J [J, 1 L2 , 0].
[L1 , L2 , L ]
(19)
J
The number of elementary boundary conditions appearing on the r.h.s. of (18) and (19)
can be different depending on the values of L1 , L . Assume that L1 + L k which we
449
can always achieve by using the identification rules. For L1 < L we find a superposition
of L1 + 1 boundary conditions in both flows (18) and (19). These flows are illustrated in
Fig. 2. For L1 L there are L + 1 boundary conditions on the r.h.s. of (18), but in (19)
we find a superposition of L boundary conditions (see Fig. 3).
The first of the flows, (18), has been analyzed in perturbation theory for large k in [31],
the second one, (19), cannot be seen in this limit. Nevertheless, both flows are known to
exist [3234]. They are generated by the [0, 0, 2] field (in Kac labels (1, 3)) and differ by
the sign of the perturbation. This is in agreement with our general statements (15) on the
boundary fields generating the flow.
Now let us choose a superposition of boundary conditions with a trivial first label. We
set S = (L1 , 0) (1 L1 k) and = [0, L2 , L ] in (16) and obtain
(k+1)
NL1 L J [0, L2 , J ] [L1 , L2 , L ].
(20)
On the other hand, we could choose = [0, L2 , k + 1 L] and S = (k + 1 L1, 0) which
leads to
(k+1)
NL1 L J [0, L2 , J ] [L1 1, 1 L2 , L ].
(21)
The two flows are illustrated in Fig. 4. Again, perturbation theory for large k can only see
the first of these flows [21,35].
Fig. 2. A pictorial representation of the flows (18) and (19) for L1 < L , L1 + L k: a single brane can flow to
a superposition of point-like branes.
Fig. 3. A pictorial representation of the flows (18) and (19) for L1 L , L1 + L k: a single brane can flow to
a superposition of point-like branes.
450
Fig. 4. Pictorial representation of the flows (20) and (21): a superposition of string-like branes can flow to a single
brane.
Fig. 5. Ising model: flows from the free boundary condition to spin up or spin down.
4.2. Critical Ising model

The simplest model in the unitary minimal A-series is the critical Ising model with
k = 1. There are three boundary conditions, the free one and two with fixed spin (up
or down) at the boundary. Our geometrical picture reduces to a cushion-like background
where the fixed boundary conditions are point-like objects at the top and bottom whereas
the free boundary condition is a string-like object sitting precisely in the middle of the
cushion (see Fig. 5).
Starting from the free condition, the system can be driven into a theory with fixed spin.
These are precisely the two flows (18), (19). They are depicted in Fig. 5. Flows starting
from a superposition of boundary conditions can be found in Appendix B.
4.3. Tricritical Ising model
The second model in the unitary minimal series is the tricritical Ising model with central
charge c = 7/10. Once more, the flows triggered by the (1, 3)-field as analyzed in [32]
are correctly reproduced by (18) and (19). There are, however, more flows known which
correspond to a perturbation with other fields [36]. As the rule depends on the specific coset
construction, it is possible to find additional flows by choosing different coset realizations
of the same theory. For the tricritical Ising model, such alternative realizations do exist.
One is given by (E7 )1 (E7 )1 /(E7 )2 . When we apply our rule to this coset construction,
it reproduces the two known flows caused by the (3, 3)-field. In Kac labels they read
(2, 2) (3, 1),
(2, 2) (1, 1),
and they are depicted together with the other flows in Fig. 6. These two flows also
451
Fig. 6. Boundary RG flows in the tricritical Ising model induced by the (1, 3)-field (13 ) and the (3, 3)-field
(33 ).
appear in higher minimal models [37] where we do not know a coset realization for
the (3, 3)-perturbations. This may be related to the observation that the tricritical Ising
model seems to be the only theory in which the considered perturbations are integrable
[37]. Nevertheless, recovering flows from the exceptional E7 coset construction can be
considered as an important check of the conjectured rule.
There are more realizations of the tricritical Ising model as coset model [38], but only
for one of them our rule predicts flows starting from single boundary conditions. This is
the construction as a so(7)1 /(G2 )1 coset model. The flows found there coincide with the
(1, 3)-flows (19), i.e., with those flows found in the SU (2) construction that cannot be seen
in the perturbative approach.
In Appendix B we collected the complete results, i.e., all flows described by the rule
(16) including those that start from a superposition of boundary conditions, for the su(2)
and the E7 construction.
4.4. Minimal models, D-series
For the minimal models with k 3, there is in addition to the diagonal modular invariant
(A-type) another modular invariant giving rise to the D-series of minimal models. Up to
some exceptional values of k, these form all possible modular invariants for the minimal
models. Depending on k being even or odd, we distinguish between minimal models of
type (D k+4 , Ak+2 ) and (Ak+1 , D k+5 ), and we shall discuss these two classes of models
2
2
separately.
4.4.1. (D, A): k even
Boundary conditions in the (D, A) models are labeled by triples [L1 , L2 , L ] where

L = 0, . . . , k + 1 and L2 = 0, 1 lie in the usual ranges whereas L1 takes the values
0, 1, . . . , k/2 1, [k/2, +], [k/2, ]. The sum L1 + L2 + L of boundary labels has to
452
be even.6 We have the identifications [L1 , L2 , L ] [L 1 , 1 L2 , k + 1 L ] where

k
k
k
,
,
for
L
=
1
2
2 , and 2 odd,
L1 =
(22)
L1 , otherwise.
The annulus coefficients can be written as a combination of the annulus coefficients nD,k
for the su(2) model at level k with D-type partition function and the fusion rules of su(2)1
and su(2)k+1 ,
n[l1 ,l2 ,l ][L1 ,L2 ,L ] [J1 ,J2 ,J
(1)
]
(k+1) J
J1
J2
= nD,k
l1 L1 Nl2 L2 Nl L
+ nD,k
l1 L1
J1
(1)
(k+1)
Nl2 (1L2 ) J2 Nl (k+1L ) J .
Here, nD,k is given by
J1
nD,k
l1 L1
(k)
(k)
Nl1 L1 J1 + N(kl1 )L1 J1 , for J1 , L1 = k/2,
(k) J1
for J1 = k/2,
L1 = [k/2, ],
Nl1 k/2 ,

(k)
= N k/2 ,
for J1 = k/2, ,
L1 = k/2,
l L

1 1
for J1 = L1 = k/2, ,
l1 mod 4 ,

l1 2 mod 4 ,
for J1 = k/2, ,
L1 = [k/2, ].
(23)
The geometry of the minimal models of the (D, A)-series is the cylinder of the A-series
divided by the reflection at the plane at x = k/2 (see Fig. 7). The branes which are
symmetric with respect to the reflection split into two fractional branes.
We are only going to discuss flows starting from a single boundary condition
[L1 , L2 , L ] with L = 0, k +1. Because of the identification rules we are allowed to choose
L k/2. We distinguish three cases:
L1 + L < k/2:
In (16) we choose = [L1 , L2 , 0] and S = (L , 0) and find the flow
(k)
NL1 L J [J, L2 , 0].
[L1 , L2 , L ]
(24)
Fig. 7. The geometry of the (D, A)-minimal model is the cylinder of the A-minimal model modulo the reflection
(indicated by the arrow) at the x = k/2-plane.
6 When we use L only as a numerical value and not as a label, we forget about the possible signs from
1
fixed-point resolution.
453
Alternatively, we choose = [L1 , 1 L2 , 0] and S = (k + 1 L , 0) and obtain

(k)
[L1 , L2 , L ]
(25)
NL1 ,(L 1) J [J, 1 L2 , 0].
J
The flows are illustrated in Fig. 8.

k/2 L1 + L , L1 = k/2:
As in the previous case, we set = [L1 , L2 , 0] and S = (L , 0). The rule (16) then
leads to
(k)

kJ
[L1 , L2 , L ]
NL1 L J + NL(k)
[J, L2 , 0]

1L
J < k/2
k/2
NL(k)

1L
(k)
NL1 L k/2
k
k
2,+
2,
, L2 , 0

, L2 , 0 .
(26)
We find a second flow for = [L1 , 1 L2 , 0] and S = (k + 1 L ),

(k)

kJ
[L1 , L2 , L ]
NL1 (L 1) J + NL(k)
[J, 1 L2 , 0]
1)
(L
1
J < k/2
k

, 1 L2 , 0

(k)
NL1 (L 1) k/2 k2 , , 1 L2 , 0 .
(k)
NL1 (L 1) k/2
2,+
(27)
We have depicted the flows in Fig. 9.
Fig. 8. The flows (24) and (25) illustrated in the half-cylinder geometry of the (D, A)-minimal models.
Fig. 9. The flows (26) and (27) in the (D, A)-minimal model. Some of the branes appear with multiplicity 2. Note
that the point-like branes sitting on the fixed-plane come in pairs of fractional branes indicated by + and .
454
L1 = [k/2, +] (analogously for [k/2, ]):

Here we find the flows
(k)
N k J [J, L2 , 0]
[L1 , L2 , L ]
L
J < k/2
k

2 , + , L2 , 0 L = 0 mod 4,
k

,
L
,
0
L = 2 mod 4,
2
2
0
L odd
(28)
and
[L1 , L2 , L ]
(k)
N(L 1)k/2 J [J, L2 , 0]
J < k/2

k
2 , + , L2 , 0 L = 1 mod 4,

k

2 , , L2 , 0 L = 3 mod 4,
0
L even.
(29)
One example for the described flows with L = 2 can be found in Fig. 10.
4.4.2. (A, D): k odd
In the (A, D)-series we label the boundary conditions by triples [L1 , L2 , L ] where
k+1
k+1
L1 = 0, . . . , k, L2 = 0, 1, and L = 0, . . . , k1
2 , [ 2 , +], [ 2 , ]. Selection rules force
the sum L1 + L2 + L to be even, triples [L1 , L2 , L ] and [k L1 , 1 L2 , L ] are identified.
Here, L is defined as in (22) with k replaced by k + 1.
The annulus coefficients are given by the fusion rules of su(2) at level k and 1 and by
the annulus coefficients nD,k+1 of the su(2)k+1 model with D-type modular invariant (see
(23)),
n[l1 ,l2 ,l ][L1 ,L2 ,L ] [J1 ,J2 ,J
(k)
(1)
]

(k)
(1)
J
J
= Nl1 L1 J1 Nl2 L2 J2 nD,k+1
+ Nl1 (kL1 ) J1 Nl2 (1L2 ) J2 nD,k+1
.

l L
lL
The geometry of the (A, D)-minimal models is obtained from the cylinder geometry
of the A-series by dividing out the reflection at the center (see Fig. 11). A brane which is
symmetric under this reflection splits into two fractional branes.
Fig. 10. The flows (28) and (29) for L = 2. In contrast to the flows in Fig. 9 all branes only appear with
multiplicity 1. Note that the sign of the fractional brane at the RG end-point depends on the value of L .
455
As in the (D, A)-case we look for flows starting from a single boundary condition
[L1 , L2 , L ] with L = 0. We have to distinguish two cases:
L = k+1
2 :
When we choose = [L1 , L2 , 0] and S = (L , 0) in (16), we find

NLk 1 L J [J, L2 , 0].
[L1 , L2 , L ]
(30)
Similarly, for = [k L1 , 1 L2 , 0] and S = (k + 1 L , 0) we obtain

NLk 1 (L 1) J [J, 1 L2 , 0].
[L1 , L2 , L ]
(31)
These flows are shown in Fig. 12.

L = [ k+1
2 , ]:
By setting = [L1 , L2 , 0] and S = ( k+1
2 , 0) we find a flow for a superposition of two
boundary conditions,
(k)

k+1
L1 , L2 , k+1
(32)
N k+1 J [J, L2 , 0].
2 , + L1 , L2 , 2 ,
J
L1
An example, for this flow is shown in Fig. 13.
Fig. 11. The (A, D)-minimal model is described by the cylinder geometry of the A-series modulo the reflection
at the center (indicated by the arrows).
Fig. 12. The flows (30) and (31) in the (A, D)-minimal model.
Fig. 13. The flow (32) in the (A, D)-minimal models starting from a superposition of two fractional branes +
and .
456
4.5. Parafermion series

The parafermion series can be realized by the cosets su(2)k /u 2k . The sectors of the
theory are labeled by pairs [l, l ] where l = 0, . . . , k and l is a 2k-periodic integer for
which we usually choose the range l = k + 1, . . . , k. Selection rules force the sum l + l
to be even, and the pairs [l, l ] and [k l, l + k] are identified. The fields which appear as
perturbing fields in our rule have conformal weight h = (k 1)/k.
The maximally symmetric branes in parafermion theories come in two classes: the
untwisted branes (A-branes), and the twisted branes (B-branes). The untwisted branes are
the usual Cardy branes and carry labels [L, L ] from the same set as the sectors. The
annulus coefficients are given by

n[l,l ][L,L ] [J,J ] = l +L J mod 2k NlL
(k) J
(k)
+ l +L J +k mod 2k Nl(kL) J ,
where N (k) are the fusion rules of su(2)k .

In the limit of large levels k, the parafermion models can be described by a non-linear
-model on a disc with non-trivial metric (sometimes this geometry is called bell). We
want to use this picture to visualize boundary conditions as branes. The untwisted branes
[0, L ] appear as point-like objects sitting at k special equidistant points on the boundary
of the circle. The other untwisted branes [L, L ] are one-dimensional objects that stretch
between these points (see Fig. 14). These pictures have been introduced in [10].
Let us now apply the rule (16) to untwisted boundary conditions. The general result is
(k)
NSL J [J, L ]
[L, L S] [L, L S + 2] [L, L + S]
J
L, L , S
L + L
with
+ S even. An example of a flow with L = 2, S = 1 is
for any label
graphically presented in Fig. 15. Particularly interesting is the case L = 0 where the end
configuration consists only of a single boundary condition:
[0, L S] [0, L S + 2] [0, L + S] [S, L ].
Such a flow is shown in Fig. 16.
In addition to the untwisted (Cardy) boundary conditions there are twisted ones which
involve a non-trivial automorphism . In the u(1) part it acts as reflection,
(J )(z) = J (z),
Fig. 14. A generic untwisted brane [L, L ] in the parafermion model and the geometric interpretation of the labels
of the brane. The possible positions of the point-like branes of type [0, L ] are also indicated.
457
Fig. 15. Two untwisted branes in the parafermion model with L = 2 flow to a configuration of a L = 1 and a
L = 3 brane.
Fig. 16. Four point-like branes condense into a single extended brane.
on the numerator su(2) it acts only as an inner automorphism. The twisted boundary
conditions have first been constructed in [10].
These boundary conditions are labeled by pairs [L, L ] where L = 0, . . . , k is an integer
coming from the numerator part, and the sign L = comes from the twisted U (1).
Selection rules force L to be even in combination with the sign L = +, and odd if it
comes with L = . As L is determined by L, we shall often leave it out and write [L, ].
Furthermore, there is an identification between pairs, [L, +] [k L, (1)k ]. For even k,
the pair [k/2, L ] is a fixed-point of this identification, and the corresponding boundary
condition has to be resolved into two elementary boundary conditions [k/2, L ; ].
The annulus coefficients (before fixed-point resolution) are given by

n[l,l ][L,L ] [J,J ] = NlL J nl L J + NklL J nk+l L J ,

(k)
where the coefficients nS L J

1,
+
nl = nl + =
0,

0,
nl + = nl + =
1,
(k)
for the twisted U (1) read

l even,
l odd,
l even,
l odd.
The resolution of the fixed-point for L = k/2 is straightforward.

In our geometric picture the brane [0, +] appears as point-like object in the center of
the disc, and the branes [L, ] are two-dimensional discs placed at the origin (see Fig. 17).7
The rule (16) applied to the twisted parafermion branes (ignoring again the fixed-point
resolution) describes a flow from a superposition of S + 1 identical boundary conditions
7 Note that from the point of view of closed strings as derived in [10], the smallest twisted brane is not point-
like, but has a small, non-zero radius. We do not want to discuss these differences any further, as the only purpose
of the geometrical pictures here is to visualize the boundary conditions and the boundary RG flows.
458
Fig. 17. A generic twisted brane [L, ] in the parafermion model. For L = 0 the brane becomes point-like. If k
is even we find two branes for L = k/2 which cover the whole disc.
Fig. 18. Two of the smallest twisted branes condense into a larger disc.
[L, ] to some other configuration,

(k)
(S + 1)[L, ]
NSL J [J, ].
J
An example for L = 0, S = 1 is shown in Fig. 18.

There is a different realization of the parafermion series, namely, as diagonal coset
models su(k)1 su(k)1 /
su(k)2 . Here, the adjoint field which induces the flows has
conformal weight h = 2/(k + 2). In this realization, we can even find flows starting from
single boundary conditions, e.g., a one-dimensional brane at the boundary of the disc flows
to a point-like one. We leave it at these general words here, but we shall work out the flows
for k = 3 in Section 4.6.
4.6. 3-state Potts model
The 3-state Potts model is a square lattice model where at each site i there is an angular
variable i taking values 0, 2/3. The interaction is given by the classical Hamiltonian

H = c
cos(i j ),
i,j
the sum running over nearest neighbor pairs. When the model is at its critical coupling,
it can be described by a conformal field theory. Introducing a boundary into the problem,
one can show that there are 8 possible boundary conditions [39,40]. These are the free
boundary condition, the three different fixed boundary conditions, three mixed boundary
conditions (one of the three spin states is forbidden at the boundary) and one additional
boundary condition whose interpretation in the classical Potts model is not as simple as
for the others (see [39] for details). We use the nomenclature of [39] and call the boundary
conditions F , A, B, C, AB, BC, AC and N (for new), respectively.
459
Fig. 19. Pictorial representation of boundary conditions in the 3-state Potts model.
The CFT describing the critical 3-state Potts model is a minimal model of central charge
c = 4/5. It can be obtained by various coset constructions: it belongs, e.g., to the minimal
D-series for k = 3 and to the parafermion series also for k = 3. In addition to these
two realizations, we shall review the construction as a diagonal su(3) coset. In all these
realizations we determine flows between boundary conditions using the rule (16). In this
section, we shall see the rule in action in examples with twisted boundary conditions.
We start with the construction as a
su(2)3
u 6
coset that we already encountered in the discussion of parafermion theories in Section 4.5.
The untwisted branes are labeled by pairs [L, L ] where the labels L and L lie in the range
L = 0, 1, 2, 3 and L = 2, 1, 0, 1, 2, 3. Selection rules force the sum L + L to be even,
and the pairs [L, L ] and [3 L, L 3] label the same brane. These are the usual Cardy
branes, and there are six of them in the model. We adopt the geometric interpretation from
Section 4.5. In this interpretation, three branes are points on the boundary of the disc and
correspond to the three fixed boundary conditions A, B, C. The other three describe mixed
boundary conditions AB, BC, AC and are represented as lines (see Fig. 19).
The remaining two boundary conditions can be constructed as twisted branes. They are
labeled by pairs [L, ] where L = 0, . . . , 3 is an integer coming from the numerator part,
and the sign comes from the twisted U (1). Selection and identification rules leave us
with the two boundary conditions [0, +] [3, ] and [1, ] [2, +]. In our geometric
picture the brane [0, +] appears as point-like object in the center of the disc, and the brane
[1, ] is a two-dimensional disc placed at the origin (see Fig. 19). They are the free
and the new boundary condition, respectively. Table 1 gives an overview of boundary
conditions in this particular model.
Now, we want to apply our rule (16) to determine RG flows. We first observe that the
rule does not describe flows starting from a single boundary condition. Instead, we shall
analyze all possible flows for superpositions of two boundary conditions. In all these cases
the boundary spin triggering the flow is S = 1.
We start with untwisted branes. Applying the rule (16) for = [L = 0, L = 1], we find
the flow
A B = [0, 0] [0, 2] [1, 1] = AB,
460
Table 1
Boundary conditionsin the 3-state Potts model
in three different coset constructions. The g-factors are given in
terms of N 4 = (5 5 )/2 and 2 = (1 + 5 )/2
Boundary label from
su(2)3
u(1)
3
su(2)3
su(2)1
su(2)4
su(3)1
su(3)1
su(3)2
[0, 0]
[0, 2]
[0, 2]
[1, 1]
[1, 3]
[1, 1]
[1, ]
[0, +]
[0, 0, 0]
[0, 0, 2+]
[0, 0, 2]
[2, 0, 2]
[2, 0, 0]
[2, 0, 2+]
[1, 0, 1]
[3, 0, 1]
[(0, 0), (0, 0), (0, 0)]

[(0, 0), (0, 1), (2, 0)]
[(0, 0), (1, 0), (0, 2)]
[(0, 0), (1, 0), (1, 0)]
[(0, 0), (0, 0), (1, 1)]
[(0, 0), (0, 1), (0, 1)]
[0, 0, 0; ]
[0, 0, 1; ]
g-factor
Notation from [39]
N
N
N
N 2
N 2
2
N
2 3
N
N 3
A
B
C
AB
BC
AC
N
F
As one could already infer from symmetry arguments, there are also the flows B C
BC and A C AC. Starting instead with = [1, 2] we find
AB BC = [1, 1] [1, 3] [0, 2] [2, 2] = B AC,
Analogous results can be obtained for permutations of the letters A, B, C.

Choosing = [0, ] in (16) yields the flow
2 F = 2 [0, +] [1, ] = N,
If we set = [1, +], the resulting flow is

2 N = 2 [1, ] [0, +] [2, +] = F N,
These are all flows provided by the rule (16) for superpositions of two boundary conditions.
The field responsible for the flows comes from the coset sectors H(0,2) and has conformal
weight h = 2/3. This can be concluded from our general prescription in Section 3 (see
Eq. (15)).
We now turn to the description of the Potts model as diagonal su(2) coset,
su(2)3 su(2)1
su(2)4
where the modular invariant is obtained from charge-conjugated modular invariants in the
numerator, the denominator su(2)4 contributes a D4 modular invariant. The perturbing
field comes from the adjoint sector [0, 0, 2] and has conformal weight h = 2/3.
461
We find four boundary conditions L1 = 0, 1, 2, 3 in the su(2)3 part and two boundary
conditions L2 = 0, 1 in the su(2)1 part. The su(2)4 part has a D4 modular invariant. There
are four boundary conditions which we label by L = 0, 1, 2+, 2. The coefficients of the
corresponding boundary states in terms of Ishibashi states can be found, e.g., in [20].
Identification and selection rules leave us with eight boundary conditions for the 3-state
Potts model. They are given in Table 1. Applying our rule, we observe first that we find the
same flows involving superpositions of two boundary conditions that we discussed in the
parafermion construction. In addition we find flows relating free and new boundary
conditions with the others, namely (for superpositions of maximally three boundary
conditions):
= F A =
= F AB =
= N AB =
= N AC B =
= 2 F AB =
,
,
= 2 N AC B =
= A B C F =
= A B C N =
= AB BC AC N =
= AB BC AC F N =
,
.
Let us finally discuss the construction of the Potts model as

su(3)1 su(3)1
su(3)2
coset. Its sectors are labeled by three su(3) weights

(l1 , l2 ), (m1 , m2 ), (l1 , l2 ) ,
462
where li , mi , li are non-negative integers (Dynkin labels) obeying

0 l1 + l2 1,
0 l1 + l2 2,
0 m1 + m2 1,
2(l1 + m1 l1 ) + l2 + m2 l2 = 0 mod 3.

The sectors are identified according to the field identification

(l1 , l2 ), (m1 , m2 ), (l1 , l2 )

(1 l1 l2 , l1 ), (1 m1 m2 , m1 ), (2 l1 l2 , l1 ) .
What remains are 6 sectors. According to the standard Cardy construction, these give rise
to 6 boundary conditions which are listed in Table 1 along with their g-factors. Before we
go to construct the remaining two boundary conditions, we want to look for RG flows.
Let us start with the boundary condition AB and exhibit what flows are predicted by
(16). We choose the perturbation S = ((0, 0), (0, 1)) and find the flow

AB = (0, 0), (1, 0), (1, 0) (0, 0), (0, 0), (0, 0) = A,
The spin S = ((0, 1), (0, 0)) leads to

AB = (0, 0), (1, 0), (1, 0) (0, 0), (0, 1), (2, 0) = B,
Analogously, we find BC B, BC C and AC A, AC C. These constitute all

flows from single boundary conditions described by the rule. For a superposition of two
boundary conditions we find flows of the form
AC B
A,
.
The two remaining boundary conditions can be obtained from twisted gluing conditions
using an automorphism which interchanges the two Dynkin labels of the su(3) theories.
In the su(3)1 there is only one sector left invariant under this automorphism, in the su(3)2
theory there are two. In total we find two twisted boundary conditions
[0, 0, 0; ] and [0, 0, 1; ],
there are no selection or identification rules in this example. We can calculate their gfactors (see Table 1) and identify the two boundary conditions as the new and the free
boundary condition, respectively.
Again, we want to investigate what flows are described by the rule (16). Let us start with
the new boundary condition and try the perturbation S = ((1, 0), (0, 0)). This leads to
N = [0, 0, 0; ] [0, 0, 1; ] = F,
463
We can identify the field that drives the described flows. From our general prescription (15), we conclude that the perturbing field is [(0, 0), (0, 0), (1, 1)] and has conformal
weight h = 2/5.
Let us compare our results with the work of Affleck et al. [39] (see also [41]). They
find several flows driven by fields of conformal weight h = 2/3 and h = 2/5. The flows
they find are all reproduced by our rule. For single boundary conditions we find exact
coincidence, for superpositions our rule suggests further flows that have not been analyzed
in [39].
Fig. 20 summarizes part of the results for boundary RG flows in the 3-states Potts model
obtained by the rule (16). The complete results can be found in Appendix B.
Fig. 20. Some of the boundary RG flows found in the 3-states Potts model. The vertical ordering of the
configurations is done according to the g-factors. The conformal weight of the field responsible for a flow is
quoted.
464
4.7. N = 2 minimal models

As last example we choose the supersymmetric parafermion theories, the N = 2
minimal series. They can be constructed as cosets su(2)k u 4 /u 2k+4 . The sectors of the
theory are labeled by triples [l1 , l2 , l ] where8 l1 = 0, . . . , k and l is a (2k + 4)-periodic
integer with standard range l = k 1, . . . , k + 2. The third label l2 can take the values
l2 = 1, 0, 1, 2. Selection rules force l1 + l2 + l to be even and the triples [l1 , l2 , l ] and
[k l1 , l2 2, l (k + 2)] label the same sectors.
The discussion of boundary conditions is analogous to the parafermion case (Section 4.5). There are untwisted boundary conditions (A-branes), labeled by triples
[L1 , L2 , L ], and twisted ones (B-branes) [10]. As many results can be directly translated
from the parafermion case, we are not going to repeat the whole picture here, but restrict
our discussion now to untwisted, even (L2 = 0, 2) boundary conditions.
Geometrically they can be displayed by straight, oriented lines (orientation depends on
the label L2 ) stretched between k + 2 special punctures on a disc [10]. The smallest lines
connecting two neighboring points have a label L1 = 0, k.
Let us see what flows are described by the rule (16). Let us choose L1 , L2 , L , S s.t.
L1 + L2 + L + S is even. Then we find the flow
[L1 , L2 , L S] [L1 , L2 , L S + 2] [L1 , L2 , L + S]
(k)
NSL1 J [J, L2 , L ].
(33)
We show some examples for k = 3 in Figs. 21 and 22.
Fig. 21. Some flows in the N = 2 minimal model for k = 3.

8 Usually the sectors are labeled by the triples (l, m, s) where l corresponds to l , m to l and s corresponds
1
to l2 . We choose a different order here and put all labels that belong to the numerator theory to the front.
465
Fig. 22. Flows (33) in the N = 2 minimal model for k = 3: the first one is obtained by setting S = k, L1 = 1, the
second one by setting S = k, L1 = 0.
At the end of this section, we want to show how these results can be used to determine
the group of brane charges in the N = 2 minimal models. We assign charges Q[L1 ,L2 ,L ]
to the boundary conditions s.t. they are conserved during RG flows. From the rule we
see immediately that all charges can be expressed by the charges Q[0,L2 ,L ] of boundary
conditions with L1 = 0 (just set L1 to zero in the flow (33)). This means that for the even
untwisted boundary conditions we have at most the charge group Z2k+4 , one copy of Z
for every boundary condition with L1 = 0. Now, the rule implies more constraints on the
charges. It turns out that in the end we remain with the relations
Q[0,0,L k1] + Q[0,0,L k+1] + + Q[0,0,L +k+1] = 0,
Q
[0,0,L ]
= Q
[0,2,L ]
(34a)
(34b)
To find (34a), it is sufficient to consider the flows with S = k and L1 = 1 in (33) (see,
e.g., the first flow of Fig. 22). The second relation (34b) results from flows with S = k and
L1 = 0 (see the second flow of Fig. 22) combined with (34a). One can then show that other
flows do not give any further constraints.
To summarize we find that the charge group of even untwisted branes is Zk+1 . This
result has already been obtained in [10]. It coincides with the computation of RR-charges
in [42].
5. Conclusions
In this work we presented a proposal for a simple rule for boundary RG flows in coset
models. A specialized version of the rule for untwisted Cardy boundary conditions had
been announced earlier [3]. Evidence for the rule comes from three directions. First, the
rule is in concordance with perturbative results. Second, there are strong arguments that all
flows described by the rule satisfy the general g-theorem of Affleck and Ludwig. Third,
in a number of examples, including non-trivial exceptional coset constructions, the rule is
able to reproduce flows that have been obtained by different means.
In the last section of the paper, we presented the broad range of application of the rule
in different coset constructions. Having a rich knowledge of possible boundary RG flows
466
in particular models, we can start to determine quantities that are invariant under RG flows.
Interpreting the RG flows as dynamical processes between brane configurations in string
theory, the determination of invariants leads to the computation of D-brane charge groups.
For WZNW models such a computation has been performed in [43,44] using the rule of
Affleck and Ludwig. We showed how the generalized rule can be used to determine brane
charges in N = 2 minimal models. The same method can be applied to other coset models,
in particular, to other KazamaSuzuki models and compared to the RR-charges that have
been calculated in [42].
Let us mention two open issues that one could consider in the future. In the perturbative
regime we see a lot more flows than are described by our rule. What happens to them when
we take the level to finite values? Another problem which is almost unexplored concerns
boundary RG flows for non-maximally symmetric boundary conditions. In the perturbative
regime, we have some (though very limited) informations from non-symmetric solutions of
the effective action found in [27]. It would be interesting to study these boundary conditions
and their relevance for brane charges further.
Acknowledgements
I would like to thank A. Cappelli, G. DAppollonio, K. Graham, A. Recknagel,
T. Quella, C. Schweigert and especially V. Schomerus for their comments and useful and
stimulating discussions.
Appendix A. Compatibility with g-conjecture

In this appendix we want to show that the inequality (17) which describes the
compatibility of our proposed rule with the g-conjecture is satisfied when we take the
levels to be large. For convenience, let us rewrite the inequality here and introduce the
abbreviations gL and gR for left- and right-hand side of the relation,
gL :=

S
bSS
SS 0
h
S00
>
SS0
g
S00
=: gR .
(A.1)
We shall expand both sides of the relation to the order 1/k 2 where k is the level which is
sent to large values. Note that we not necessarily have to take all levels to be large as long
as the representation S is trivial w.r.t. the algebras that stay at small levels.
Let us consider the simplest case when g and h are simple Lie algebras with level k and
k = kxe , respectively. Here, the integer xe denotes the embedding index of the embedding
h ; g. When we want to expand the expressions in (A.1), we can make use of the formula

g

S

SS0
2
(A.2)
1 2 g CS + O 1/k 3
g = dim V
6k
S00
which can be found, e.g., in [17, Eq. (13.175)]. Here, g is the dual Coxeter number of g,
V S is the representation space of the representation labeled by S, and CS is the quadratic
467
Casimir,

1
trV S T T .
S
dim V
T are the generators of the Lie algebra g. To relate the Casimir of a g-representation with
that of a representation of h we use the formula
CS =

1 dim g
trV S Tm T m ,
trV S T T =
xe dim h
(A.3)
where the Tm are the generators of h embedded in g.

Now we are prepared to check the inequality (A.1). We expand the l.h.s. according to
the formula (A.2) and obtain

2
bSS dim V S 1 2 gh CS + O 1/k 3 .
gL =
6k

S
After some manipulations and applying the relation (A.3) we arrive at

xe dim h
2
CS + O 1/k 3 .
gL = dim V S 1 2 gh
6k
dim g
Inserting k = kxe and using the fact that dim g > dim h as well as g > gh /xe we can
finally conclude that gL > gR in the order 1/k 2 .
The proof for semi-simple Lie algebras can be done essentially along the same lines. Let
(n)
us sketch the procedure in the example of a coset model of the form g(1)
k1 gkn /hk
where the level of the denominator is determined by k = k1 x1 + + kn xn . Using the
expansion formula (A.2), we reduce our problem to proving the inequality
n

gi 2 gh 2
tr Ti > 2 tr
Ti .
k
k2
i=1 i
(A.4)
The left-hand side can be estimated from below as before,

n

g
gi 2
1
h

>
tr
T
tr Ti 2 .
i
2
x
k
x
k
k
j
j
i
i
i=1 i
i
(A.5)
We show now that this estimate is good enough to prove (A.4), i.e., that the difference

j kj xj
2
:=
tr Ti 2 tr
Ti
ki xi
i

is positive. By introducing ai := ki xi / j kj xj we can rewrite in a manifestly positive
form,
2
n1

ai
ai
1

=
tr
(T1 + + Tn1
) + Tn Ti ,
1 + an
an
ai
i=1
which completes the proof.
468
Appendix B. Tables for flows in specific models

B.1. Critical Ising model
Boundary conditions: 0 (free), + (spin up), (spin down).
1 su(2)2
Coset realization su(2)su(2)
: Perturbing field has h = 1/2. Flows resulting from (16)
3
starting from:
a single boundary condition
0 +,
.
a superposition of two boundary conditions
+ 0,
+,
.
B.2. Tricritical Ising model
Boundary conditions:
Symbol
Conf. weight h of correspectively field
g-factor
0
0+
0
d
3/2
0
7/16
3/5
1/10
3/80
a 0.5127
a
0.5127
2 0.7251
a
b/a 2 0.8296
b/a 2 0.8296
b/a 1.173
Here, a 4 =
5 5
40 ,
b2 =
5+ 5
2 .
2 su(2)1
: Perturbing field has h = 3/5. Flows resulting from (16)
3
starting from:
d 0,
+ ,
0+ 0,
+,
0 0,
;

0 d
d,
0,
+ ,
0+ 0,
0+
,
0,
d,
0,
+,
0+,
d,
0;
469
a superposition of three boundary conditions

2 d 0 d,
0 0+,
2 0 + d,
0,
2 0+ d,
0+;
1 (E7 )1
Coset realization (E7 )(E
: Perturbing field has h = 1/10.
7 )2
Flows resulting from (16) starting from:
d +,
;
0+ 0 0,
0 d 0+,
0;
a superposition of four boundary conditions

+ 0+ 0 d,
+,
,
2 0 2 d 0;
a superposition of six boundary conditions

+ 2 0+ 2 0 0+,
0,
2 0 4 d d;
B.3. Three-state Potts model

Boundary conditions: A, B, C, AB, BC, AC, F, N (see Table 1). Note that the model
has a Z3 -symmetry which acts on the boundary conditions as A B C A, AB
BC AC AB, the boundary conditions F and N are fixed. We shall only write out
flows modulo this symmetry, e.g., the flow A B AB stands also for B C BC
and C A AC.
1 su(2)3
: Perturbing field has h = 2/3.
4
Flows resulting from (16) starting from:
F BC,
A,
N BC,
A BC;

2 F N,
BC,
2 N F N,
A BC,
A B AB,
AB AC A BC;
470

A B C N,
F,
AB,
A,
AB BC AC F N,
N,
A BC,
AB,
3 F N,
F,
3 N F N,
N;
a superposition of six boundary conditions

2 A 2 B 2 C N,
2 AB 2 BC 2 AC F N;
3
Coset realization su(2)
u(1) : Perturbing field has h = 2/3. Flows resulting from (16)
starting from:
A B AB,
2 F N,
AB AC A BC,
2 N F N;

A B C AB,
3 F N,
AB BC AC A BC,
3 N F N;
a superposition of four boundary conditions

2 A B C A,
4 F F,
2 AB BC AC AB,
4 N N;
su(3)1 su(3)1
: Perturbing field
su(3)2
Coset realization
starting from:
AB A,
B,
has h = 2/5. Flows resulting from (16)
N F ;

A BC AB,
AC,
A,
B,
C,
F N N,
F ;

A 2 BC BC,
AB,
AC,
F 2 N N.
471
References
[1] I. Affleck, A.W.W. Ludwig, Universal noninteger ground state degeneracy in critical quantum systems,
Phys. Rev. Lett. 67 (1991) 161164.
[2] I. Affleck, A.W.W. Ludwig, The Kondo effect, conformal field theory and fusion rules, Nucl. Phys. B 352
(1991) 849862.
[3] S. Fredenhagen, V. Schomerus, On boundary RG-flows in coset conformal field theories.
[4] H. Ishikawa, Boundary states in coset conformal field theories, Nucl. Phys. B 629 (2002) 209232.
[5] H. Ishikawa, T. Tani, Novel construction of boundary states in coset conformal field theories, Nucl. Phys.
B 649 (2003) 205242.
[6] P. Goddard, A. Kent, D.I. Olive, Virasoro algebras and coset space models, Phys. Lett. B 152 (1985) 88.
[7] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[8] T. Quella, On the hierarchy of symmetry breaking D-branes in group manifolds, JHEP 12 (2002) 009.
[9] T. Quella, V. Schomerus, Asymmetric cosets.
[10] J. Maldacena, G.W. Moore, N. Seiberg, Geometrical interpretation of D-branes in gauged WZW models,
JHEP 07 (2001) 046.
[11] T. Quella, V. Schomerus, Symmetry breaking boundary states and defect lines, JHEP 06 (2002) 028.
[12] K. Gawedzki, Boundary WZW, G/H, G/G and CS theories, Ann. Inst. H. Poincar 3 (2002) 847881.
[13] S. Elitzur, G. Sarkissian, D-branes on a gauged WZW model, Nucl. Phys. B 625 (2002) 166178.
[14] T. Kubota, J. Rasmussen, M.A. Walton, J.-G. Zhou, Maximally symmetric D-branes in gauged WZW
models, Phys. Lett. B 544 (2002) 192198.
[15] M.A. Walton, J.-G. Zhou, D-branes in asymmetrically gauged WZW models and axial-vector duality, Nucl.
Phys. B 648 (2003) 523541.
[16] G. Sarkissian, On D-branes in the NappiWitten and GMM models.
[17] P.D. Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, in: Graduate Texts in Contemporary
Physics, Springer, New York, 1999.
[18] D.C. Dunbar, K.G. Joshi, Maverick examples of coset conformal field theories, Mod. Phys. Lett. A 8 (1993)
28032814.
[19] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, On the classification of bulk and boundary conformal
field theories, Phys. Lett. B 444 (1998) 163.
[20] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Boundary conditions in rational conformal field
theories, Nucl. Phys. B 570 (2000) 525589.
[21] S. Fredenhagen, V. Schomerus, D-branes in coset models, JHEP 02 (2002) 005.
[22] T. Gannon, M.A. Walton, On the classification of diagonal coset modular invariants, Commun. Math.
Phys. 173 (1995) 175198.
[23] I. Affleck, A.W.W. Ludwig, Exact conformal field theory results on the multichannel Kondo effect: Single
fermion Greens function, selfenergy and resistivity, UBCTP-92-029.
[24] S. Fredenhagen, V. Schomerus, Brane dynamics in CFT backgrounds.
[25] A.Y. Alekseev, A. Recknagel, V. Schomerus, Brane dynamics in background fluxes and non-commutative
geometry, JHEP 05 (2000) 010.
[26] S. Fredenhagen, D-brane dynamics in curved backgrounds, Ph.D. Thesis, Humboldt University, Berlin,
http://edoc.hu-berlin.de/abstract.php3/dissertationen/fredenhagen-stefan-2002-09-16, 2002.
[27] A.Y. Alekseev, S. Fredenhagen, T. Quella, V. Schomerus, Non-commutative gauge theory of twisted
D-branes, Nucl. Phys. B 646 (2002) 127157.
[28] A. Cappelli, C. Itzykson, J.B. Zuber, Modular invariant partition functions in two dimensions, Nucl. Phys.
B 280 (1987) 445465.
[29] A. Cappelli, C. Itzykson, J.B. Zuber, The ADE classification of minimal and A1(1) conformal invariant
theories, Commun. Math. Phys. 113 (1987) 1.
[30] A. Kato, Classification of modular invariant partition functions in two dimensions, Mod. Phys. Lett. A 2
(1987) 585.
[31] A. Recknagel, D. Roggenkamp, V. Schomerus, On relevant boundary perturbations of unitary minimal
models, Nucl. Phys. B 588 (2000) 552564.
[32] L. Chim, Boundary S-matrix for the tricritical Ising model, Int. J. Mod. Phys. A 11 (1996) 44914512.
[33] C.-R. Ahn, C. Rim, Boundary flows in the general coset theories, J. Phys. A 32 (1999) 25092525.
472
[34] F. Lesage, H. Saleur, P. Simonetti, Boundary flows in minimal models, Phys. Lett. B 427 (1998) 8592.
[35] K. Graham, On perturbations of unitary minimal models by boundary condition changing operators,
JHEP 03 (2002) 028.
[36] I. Affleck, Edge critical behaviour of the 2-dimensional tri-critical Ising model, J. Phys. A 33 (2000) 6473
6480.
[37] K. Graham, I. Runkel, G.M.T. Watts, Minimal model boundary flows and c = 1 CFT, Nucl. Phys. B 608
(2001) 527556.
[38] P. Bowcock, P. Goddard, Virasoro algebras with central charge c < 1, Nucl. Phys. B 285 (1987) 651.
[39] I. Affleck, M. Oshikawa, H. Saleur, Boundary critical phenomena in the three-state Potts model.
[40] J. Fuchs, C. Schweigert, Completeness of boundary conditions for the critical three-state Potts model, Phys.
Lett. B 441 (1998) 141146.
[41] I. Affleck, M. Oshikawa, H. Saleur, Quantum Brownian motion on a triangular lattice and c = 2 boundary
conformal field theory, Nucl. Phys. B 594 (2001) 535606.
[42] W. Lerche, J. Walcher, Boundary rings and N = 2 coset models, Nucl. Phys. B 625 (2002) 97127.
[43] S. Fredenhagen, V. Schomerus, Branes on group manifolds, gluon condensates, and twisted K-theory,
JHEP 04 (2001) 007.
[44] P. Bouwknegt, P. Dawson, D. Ridout, D-branes on group manifolds and fusion rings, JHEP 12 (2002) 065.

K-matrices for 2D conformal field theories

Eddy Ardonne a,1 , Peter Bouwknegt b , Peter Dawson c
a Institute for Theoretical Physics, University of Amsterdam, Valckenierstraat 65,
1018 XE Amsterdam, The Netherlands

b Department of Physics and Mathematical Physics and Department of Pure Mathematics,
University of Adelaide, Adelaide, SA 5005, Australia

c Department of Physics and Mathematical Physics, University of Adelaide, Adelaide, SA 5005, Australia
Received 10 January 2003; accepted 7 March 2003
Abstract
In this paper we examine fermionic type characters (Universal Chiral Partition Functions) for
general 2D conformal field theories with a bilinear form given by a matrix of the form K K1 .
We provide various techniques for determining these K-matrices, and apply these to a variety of
examples including (higher level) WZW and coset conformal field theories. Applications of our
results to fractional quantum Hall systems and (level restricted) Kostka polynomials are discussed.
1. Introduction
Two-dimensional conformal field theories can be studied in a variety of ways. In this
paper, we will pursue the quasiparticle description, which has attracted a lot of attention
recently. In a quasiparticle description, the characters of the conformal field theories are
of the fermionic sum type. It has been conjectured that all these fermionic sums are of a
form which goes under the name of the Universal Chiral Partition Function (UCPF), see,
for instance, [6,9,12], and references therein. In general, the statistics of the quasiparticles
is fractional and interpolates between Fermi and Bose statistics. Moreover, to describe
general CFTs, we need to be able to incorporate the effect of the non-trivial fusion
rules of the fields, which can be done by allowing for so-called pseudoparticles. These
E-mail addresses: ardonne@uiuc.edu (E. Ardonne), pbouwkne@physics.adelaide.edu.au,
pbouwkne@maths.adelaide.edu.au (P. Bouwknegt), pdawson@physics.adelaide.edu.au (P. Dawson).
1 Present address: Department of Physics, University of Illinois at Urbana-Champaign, Loomis Lab of Physics,
1110 W. Green Street, Urbana, IL 61801-3080, USA.
doi:10.1016/S0550-3213(03)00223-2
474
E. Ardonne et al. / Nuclear Physics B 660 [FS] (2003) 473531
pseudoparticles do not carry any energy and are essential in describing the non-Abelian
statistics which is found in the CFTs with non-trivial fusion rules.
Fractional statistics can be described in terms of the Haldane exclusion statistics [27].
If we allow for new types of particles, such as the pseudoparticles, the same is true for
the non-Abelian statistics, see [26] and [9]. The exclusion statistics is defined in terms of
the exclusion statistics parameters of the particles. The parameters are intimately related to
the Universal Chiral Partition Functions, as it is these parameters which lie at heart of the
UCPF, via the so-called K-matrix, which contains all the (mutual) statistics parameters. In
this paper, we will determine the K-matrices related to the affine Lie algebra CFTs, in a
particular basis. This basis was first proposed in the context of the fractional quantum Hall
states.
The topological properties of (fractional) quantum Hall states are also encoded in
matrices, which turned out to be the same as the K-matrices alluded to in the above. In
the Abelian states, the entries correspond to the coupling parameters of the ChernSimons
fields which appear in the effective action of the quantum Hall system (see, in particular
[47], and references therein). The ChernSimons term effectively changes the statistics of
the matter fields, making the relation between with the exclusion statistics plausible. More
details on this relation can be found in [3]. The basis used in the description of certain
classes of non-Abelian quantum Hall states is found to be useful in the context of general
affine Lie algebra CFTs as well.
One of the reasons that this basis is useful relates to the presence of a duality, which
relates the electron-like particles to the quasiparticles (the notion of electron-like and
quasiparticles will be explained in Section 2.1.4). Moreover, there is no mutual statistics
between these two types of particles. As this structure simplifies the study of the conformal
field theories, we will use this type of basis throughout this paper.
One of the main themes in this paper will be the determination of the K-matrices for the
affine Lie algebra CFTs. We will develop a scheme which is used to find the general Kmatrices. The main idea is to use Abelian coverings of the (in general non-Abelian) CFTs,
and project out some degrees of freedom (see also [13]). Having obtained the K-matrices,
we will propose a scheme to obtain the K-matrices for conformal field theories which are of
the coset form. We will address the diagonal cosets, as well the parafermion CFTs, related
to the affine Lie algebra CFTs. Another application are the Kostka-polynomials (see, e.g.,
[34,35], and references therein), which can also be described in terms of the K-matrices.
In more detail, the outline of this paper is as follows. We start with a general introduction
to the role of the K-matrix in 2D conformal field theories in Section 2. We will review some
results concerning the Universal Chiral Partition Function and the relation with exclusion
statistics. The structure of the basis of quasiparticles which will be used throughout
this paper is explained. We will end Section 2 by explaining the relation between the
pseudoparticles and the fusion rules of CFTs. In Section 3 we will explain the tools we will
use in determining the K-matrices for a general affine Lie algebra. The idea is to embed the
level-k affine Lie algebra in k copies of the level-1 version, and project out certain degrees
of freedom, by using what we call a P-transformation. In Section 4, we will explicitly give
the K-matrices for all the simple (untwisted) affine Lie algebras. We will apply these results
to obtain K-matrices for cosets in Section 5. Finally, in Section 6, we will present some
new results on level restricted Kostka polynomials related to affine Lie algebras. Some of
475
the details are presented in the appendices. Appendix A deals with some notational issues,
and explicitly gives all the Cartan matrices and there inverses. Appendix B and Appendix
C deal with the K-matrices for so(5)1 and G2,1 , respectively, while Appendix D relates
two different bases for sl(3)k .
2. K-matrices for 2D conformal field theories

2.1. The UCPF and exclusion statistics
Quasiparticles play an important role in the description of 2-dimensional conformal
field theories (CFTs). The exclusion statistics of these particles is closely related to
characters for CFTs, or more precisely, the Universal Chiral Partition Function (UCPF).
2.1.1. Quasiparticle basis
We will start the discussion by introducing quasiparticle bases for two-dimensional
conformal field theories, and in particular (truncated) partition functions based on these
bases. In CFTs, the quasiparticles take the form of chiral vertex operators (i) (z) (i =
1, . . . , n), which intertwine between irreducible representations of the chiral algebra. By
applying the modes of these operators on a set of vacua |, one finds (in general) an over
complete basis, which, by using suitable restrictions on the modes (s1 , . . . , sN ), can be
turned into a maximal, linearly independent set of states
i
i2
N
s
s
i1 |.
N
2 s1
(2.1)
The grand canonical partition function is obtained by taking the trace over this basis

N
L0
i
.
P (z; q) = Tr
(2.2)
zi q
i

Ni is the number operator for the quasiparticles (i) and L0 = i si . Furthermore, zi =
ei is a (generalized) fugacity and q = e . To find the one particle grand canonical
partition functions i , we will use truncated partition functions, see [44]. In particular,
one defines the truncated partition function PL (z; q) by restricting the trace over the states
(2.2) in such a way that the modes s of the quasiparticles of species i satisfy s Li
(L = (L1 , . . . , Ln )). In the limit of large L one has

PL+ei (z; q)
i zi q Li ,
PL (z; q)
(2.3)
where ei is the unit vector in the i-direction. By using a recursion relation for the truncated
partition function PL (z; q) (which can be obtained from the basis (2.1)) and the limit (2.3),
one finds relations for the one particle partition functions i (for more details, see [9,12]).
For all the CFTs which were investigated by means of a quasiparticle basis as discussed in
this section, the equations determining i are of the form (2.14), and thus the quasiparticles
satisfy so-called exclusion statistics, see Section 2.1.3.
476
2.1.2. The universal chiral partition function

It has been conjectured (see [6], and references therein) that the characters of all the
irreducible representations of (rational) conformal field theories can be written in the form

m 1

((I K) m + u)
i
mKm+Qm
i
P (z; q) =
(2.4)
,
zi q 2
mi
m
which goes under the name of the Universal Chiral Partition Function (UCPF) (or
fermionic-type character). The matrix K is a symmetric n n matrix, I is the n n
identity matrix and Q and u are n-vectors. The sum is over the non-negative integers
m1 , . . . , mn . The restrictions denoted by the prime are (in general) such that the coefficients
of the q-binomials are integers. These q-binomials are defined by

m

(q)M
M
=
(2.5)
,
(q)m =
1 qk .
m
(q)Mm (q)m
k=1
Depending on the parameters ui , the associated particles are of certain type. For physical
particles ui = , while pseudoparticles have ui < . Note that in the limit ui the
ith q-binomial reduces to 1/(q)mi due to

1
M
=
lim
(2.6)
.
M m
(q)m
As will become clear below, pseudoparticles do not carry energy. They come about in
theories with a non-Abelian symmetry, and in a sense they serve as bookkeeping devices
for the internal structure of the theory.
It was conjectured in [9,26] that the UCPF (2.4) is the partition function of a set of
particles satisfying exclusion statistics. To be able to make this connection with exclusion
statistics, we will take a closer look at truncated versions of the UCPF, and continue with
a discussion on exclusion statistics and the relation between the two.
Suppose that the truncated partition function PL (z; q) takes the form of a finitized
UCPF2

m 1

(L + (I K) m + u)i
PL (z; q) =
zi i q 2 mKm+Qm
(2.7)
.
mi
m
One can then derive recursion relations for these truncated characters by using the qbinomial relation

M 1
M
M 1
(2.8)
.
=
+ q Mm
m1
m
m
This leads to the recursion relations [3,8]
PL (z; q) = PLei (z; q) + zi q 2 Kii +Qi +ui +Li PLKei (z; q).
1
(2.9)
2 While this is the case for many examples, in general the finitized UCPF corresponding to a set of
(quasi)particles may differ from (2.7) by terms q n with n = O(Li ). This will, however, not affect the conclusion.
477
After dividing by PL (z; q), setting q = 1, taking the large L limit and using relation (2.3),
one finds
Kji
1 = 1
(2.10)
j
,
i + zi
j
or equivalently,
i 1 Kij
j = zi .
i
(2.11)
These relations are known as the IsakovOuvryWu (IOW) (2.14) equations, which give
the one particle partition functions for a system of particles which obey exclusion statistics;
this will be addressed in the next section. For more details on this issue, we refer to [3] and
references therein.
In the case of WZW Conformal Field Theories, i.e., CFTs with affine Lie algebra
symmetry, it is known that in many cases (see [10,28,39,48], and references therein) the
(chiral) partition function can be written in the form
(k)
P (z; q) =
M (q)M() (z; q),
(2.12)
(k)
(q) are the so-called level-k truncated Kostka polynomials, M() (z; q) their
where M
k limit (with fugacity parameter z). Having found an expression for the K-matrices
of these CFTs will thus give a natural guess for an explicit expression of these level-k
truncated Kostka polynomials. We will explore this further in Section 6.
For completeness, let us recall the value of the central charge cALA , of a CFT with affine
Lie algebra symmetry g at level k,
cALA =
k dim g
,
k + h
(2.13)
where h is the dual Coxeter number corresponding to g.

For convenience, throughout this paper we will denote the (untwisted) affine Lie algebra
at level k, corresponding to a finite-dimensional Lie algebra Xn , by Xn,k , rather than by
(Xn(1) )k which is more common in the literature.
2.1.3. Exclusion statistics
The starting point of the discussion on exclusion statistics will be an ideal gas of
particles which satisfy fractional (exclusion) statistics [27].
The one particle grand canonical partition functions i for a set of quasiparticles
obeying fractional exclusion statistics can be obtained from the IOW equations [31]
i 1 Kstij
j = xi ,
i
(2.14)
where Kst is the statistics matrix and xi = zi q = ei e the fugacity. Here, i is the
chemical potential of species i and the energy. Under the assumption of a symmetric
478
matrix Kst , the one particle distribution functions follow:

ni () = xi
log
j
=
xj
log i
.
xi
(
)
x
j
xi =e i
xi =e(i )
j
(2.15)
These distribution functions are in general interpolations between the BoseEinstein and
FermiDirac distribution functions.
The discussion above holds in the case of Abelian statistics, but can be generalized to the
non-Abelian case [2,3]. Non-Abelian statistics arises when quasiparticle operators (chiral
vertex operators, see below) in the underlying CFT have non-trivial fusion rules. The effect
of these fusion rules can be taken into account via so-called pseudoparticles, which do
not carry any energy (i.e., q = 1). Note that for all the cases we consider, a formulation in
which the pseudoparticles have x = 1 is possible. In fact, we only consider formulations
in which x = 1 for the pseudoparticles. More on the relation between fusion rules and
pseudoparticles can be found in Section 2.2.
We will now turn to the question of how to calculate the central charge of a system
of quasiparticles satisfying exclusion statistics with statistics matrix Kst (and speak of the
central charge associated to the matrix Kst ). First, we consider an Abelian system, i.e., a
system without pseudoparticles. In that case, the central charge is given by
1
6
cCFT = 2
dx
tot (x),
x
(2.16)
where tot (z) denotes the product

i (xj = x).
tot (x) =
(2.17)
By using the IOW-equations the central charge of Eq. (2.16) can be rewritten in the form
(see, for instance, [12])
6
cCFT = 2
(2.18)
L(i ),
where the i s are solutions of the central charge equations

st
i = (1 j )Kij ,
(2.19)
and L(z) is Rogers dilogarithm

1
L(z) =
2
z
0

log(1 y)
log y
+
dy
.
1y
y
(2.20)
The presence of pseudoparticles gives rise to a reduction of the central charge. This
reduction can be calculated in a similar way, by considering the central charge equations
restricted to the pseudoparticles. For future convenience, we will denote the statistics
matrix restricted to the pseudoparticles by K . The central charge equations become
(the prime denotes the restriction to the pseudoparticles)

(K )
1 j
ij ,
i
=
479
(2.21)

giving rise to a reduction 62 j L(j
). The central charge becomes

6
cCFT = 2
L(i )
L j
.
(2.22)
This formula agrees with the central charge calculated from the asymptotics of the UCPF
(2.4) (see, e.g., the discussion in [3]).
To summarize the above, we note that the truncated UCPFs in the large L limit give
rise to one particle partition functions (2.11), which are of the form of the IOW-equations
(2.14), with statistics matrix Kst = K. Thus the K-matrix of the UCPF can be interpreted
as a matrix which describes the statistical interactions between the (quasi)particles.
The other important point was that in all the cases where conformal field theories were
studied by means of quasiparticle bases, Eqs. (2.3) which determine i were shown to be
of the form of the IOW-equations.
We end this section by discussing the so-called quantum Hall basis, which turns out to
be very convenient for determining and studying K-matrices for conformal field theories.
2.1.4. The quantum Hall basis
A convenient basis for WZW conformal field theories was first proposed in the context
of the quantum Hall effect [18]. (This basis is also very natural from the mathematical
point of view as it is closely related to the existence of generalizations of the Durfee square
formula in combinatorics [8].) The electron-like particles (with unit charge and spin- 21
and (fractionally) charged quasiparticles (sometimes called quasiholes) are chosen to form
a basis. It was found that a basis could be chosen in such a way that the statistics matrix
Ke for the electron-like particles, and the matrix Kqp for the quasiparticles are each others
inverse
Kqp = K1
e ,
(2.23)
while, furthermore, there is no mutual statistics between the quasiparticles and electrons,
i.e.,
K = Ke Kqp.
(2.24)
This is a very important observation, which will have many consequences. Though this
basis was first proposed in the context of the Laughlin and Jain states [18], it was
soon realized that a basis with a similar structure could be constructed for the nonAbelian generalizations of the Abelian quantum Hall states [2,3,26]. These non-Abelian
generalizations are based upon WessZuminoWitten conformal field theories. In this
paper, we will determine bases for general WZW conformal field theories. In the next
section, we will review and develop some techniques which are needed to perform this
task. Here, we will first explore some consequences of the duality between the electron
and quasiparticle sector.
480
In the description of the quantum Hall effect, the quantum numbers of the particles play
an important role, as they are used to calculate physical properties. The most important are
the charge and spin quantum numbers, which are usually grouped in the so-called charge
and spin vectors, t and s, respectively (see, for instance, [47]). Denoting a general vector
for the electron (quasiparticle) sector by qe (qqp ) we have
qqp = K1
e qe .
(2.25)
The filling fraction and the spin filling are given by the expressions
T
1
= tTe K1
e te = tqp Kqp tqp ,
T
1
= sTe K1
e se = sqp Kqp sqp .
(2.26)
These quantities are important physically; from a mathematical point of view they are
interesting, as they are conserved by the W- and P-transformations of Section 3. In a sense,
these transformations are constructed in such a way that they have this property.
Let us explore some consequences of the duality, in particular Eqs. (2.23) and (2.24).
We will focus on the thermodynamic properties first and have a closer look at the IOWequations (2.14). We will denote the one particle distribution functions for the electron-like
particles and quasiparticles by i and i , respectively. The corresponding fugacities are
given by yi and xi . Thus, the i and i are the solutions to the equations
i 1 (Ke )ij
j
= yi ,
i
i 1 (Kqp )ij
j
= xi .
i
(2.27)
Now Eq. (2.23) leads to the following relations

i =
i 1
,
i
xi =
(Ke )1
ij
yi
(2.28)
Another important feature of the basis described in this section is that the presence of
pseudoparticles in the quasiparticle matrix Kqp is accompanied by the presence of so-called
composite particles in the electron matrix Ke . The reason for this will become clear in
Section 3. In general, the matrix Ke contains a few electrons (particles with unit charge
and spin up or down), with fugacities y. In addition, there are composite particles, with
fugacities y li , where the li are positive integers. The quantum numbers of the composites
in the electron sector are integer multiples of the quantum numbers of the electrons. In
the presence of composites in the electron sector, there will be pseudoparticles in the
quasiparticle sector. Pseudoparticles have x = 1, and as a consequence, pseudoparticles
will have all quantum numbers equal to zero. In principle, the fugacity of pseudoparticles
might be of the more general form xi /xj (compare Eqs. (3.19) and (3.20)), but in all
cases we will consider, this will not be the case. Also, physical particles with all quantum
numbers trivial might occur, but again, we will not encounter such a situation in this paper.
In the following, we will only encounter the situation where the electron sector
has composites, but no pseudoparticles, while the quasiparticle sector does contain
pseudoparticles, but no composites. Thus, we will assume that the quasiparticle matrix
481
has the following form:

K
K
Kqp =
,
K
K
KT = K ,
KT = K ,
KT = K ,
(2.29)
where K denotes the statistic matrix for the physical (as opposed to pseudo) quasiparticles and K the mutual statistics between the pseudo- and physical particles.
In the presence of composites and pseudoparticles, we have to generalize the definition
of tot (see Eq. (2.17)) to

l
tot (x) =
(2.30)
i xj = x lj i .
i
With this definition, the central charge is still given by Eq. (2.16). In the absence of
pseudoparticles, the central charge associated to the system Ke Kqp , is simply given
by the rank n of the matrix Ke (see, for instance, [3]). To show this, we take a look at the
central charge equations

K
i = (1 j ) qp ij .
i = (1 j )Keij ,
(2.31)
j
Now because of the fact that Kqp = K1

e , the solutions to these equations i and i are
simply related by i = 1 i . We find the central charge to be

6
6
L(i ) + L(1 i ) = 2 nL(1) = n,
cCFT = 2
(2.32)
by using the dilogarithm relation

2
(2.33)
.
6
In the case pseudoparticles are present, we again have a simple subtraction (see Eq. (2.22),
the prime denotes the restriction to the pseudoparticles)
6
L(j
).
cCFT = n 2
(2.34)
L(z) + L(1 z) = L(1) =
It is important to note that the knowledge of the K-matrix is not enough to specify
the theory completely. In addition, one has to know, or rather specify, which particles are
pseudoparticles. So two theories can have the same K-matrix, but differ in the particle
content and thereby (for instance) have different central charge. We will encounter this
situation frequently, namely as we discuss the K-matrices for CFTs with affine Lie algebra
symmetry, in cases the Lie algebra is non-simply-laced.
2.2. Pseudoparticles and fusion rules
There is an intimate connection between the pseudoparticle K-matrix K and the
fusion rules of a CFT, which can be used as a consistency check or guiding principle
482
on the construction of K-matrices. To explain this connection, consider a CFT with fusion
rules Nij k , i, j, k = 1, . . . , ). The incidence matrix of the fusion graph i , corresponding
to taking consecutive fusions with the field i, is given by the matrix Ni with components
(Ni )j k = Nij k . Hence, if Pij k (M) denotes the number of paths of length M on the fusion
graph i beginning at j and ending at k we have

Pij k (M) = (Ni )M j k .
(2.35)
Thus we find a recursion relation

Pij k (M) =
Pij l (N)Pil k (M N),
(2.36)
for each 0 N M, with initial condition Pij k (0) = j k . These recursion relations,
however, involve paths beginning and ending at arbitrary points. To derive a recursion
relation for fixed j and k we apply the characteristic equation of Ni , i.e., the )th order
polynomial equation for Ni arising from the eigenvalue equation, to Pij k (M). If the
characteristic equation is given as
)

an (Ni ))n = 0,
a0 1,
(2.37)
n=0
then, by using (2.36) for N = 1, we find the recursion relation

)

an Pij k (M n) = 0.
(2.38)
n=0
That is, a recursion relation for fixed j and k and with coefficients independent of j and
k. Different solutions of (2.38), determined by different initial conditions, correspond to
different choices of j and k.3 In particular, asymptotically the number of paths is given by
(max )M , where max is the largest eigenvalue of Ni .
On the other hand, according to the UCPF assumption, the number of paths P (M) of
length M on the fusion graph i is given in terms of the q 1 limit of the UCPF (2.4),
i.e.,
((I K ) m) + L
i
i
,
PL =
(2.39)
mi
mi
where Li = ai M + ui and K the pseudoparticle K-matrix. The numbers ai are fixed

(only depend on the sector i), in fact they arise as the part of the K-matrix describing the
coupling of the pseudoparticles to physical particles, while ui is determined by begin and
end point of the path. (The q-analogue of Eq. (2.39) is related to (level restricted) Kostka
polynomials and will be discussed in Section 6.) The numbers PL satisfy the recursion
relations (cf. (2.9))
PL = PLei + PLK ei ,
(2.40)
3 In fact, for specific initial conditions, the solution might actually satisfy simpler recursion relations obtained
by factorizing the characteristic equation and taking a subset of the factors.
483
where ei is the unit vector in the ith direction. In principle, the recursion relations (2.40)
can be manipulated to yield a recursion relation for P (M) Pai M , the quantity of interest.
Ideally, this recursion relation should be the same as (2.38). In practice, however, one
finds that one corresponds to a factor of the other due to the fact we are dealing with
specific initial conditions. In practice, it is easier to study the recursion relations (2.40)
in the large M limit, where they reduce
ato the IOW-equations (2.11). These can then be
used to derive an equation for = i i i which should correspond to the characteristic
equation for the eigenvalues of Nij k , i.e., Eq. (2.37). In particular, the largest root of the
equation determining should be equal to max .
Moreover, note that while the recursion relations corresponding to graphs on i depend
on the sector i, they should all derive from one and the same pseudoparticle matrix K
(they just differ in the choice of ai ). This puts extra constraints on the possible choices of
K , given a set of fusion rules Nij k . Unfortunately, this still does not suffice to uniquely
associate a pseudoparticle K with a set of fusion rules Nij k as is illustrated, for instance,
by the matrix
4 2
K = 32 43 ,
(2.41)
3
which arises both in A2,2 and F4,1 (see Sections 4.3.1 and 4.3.6), while these two theories
clearly have different fusion rules. This is because additional information is present in the
coupling of pseudoparticles to the physical particles (i.e., the numbers ai ). Conversely,
given a pseudoparticle K-matrix leading to the correct fusion rules, one can always
construct other K-matrices giving rise to the same recursion relations by extending the
matrix symmetrically. An example of this will be given in Section 2.3.
Finally, given a set of fusion rules Nij k , we can compute the modular S-matrix, since
this is the matrix which simultaneously diagonalizes all matrices Ni [46]. Since the T matrix acts diagonally on the characters of the CFT with values exp(2i(hi c/24)), we
can find constraints on the conformal dimensions hi and the central charge c from the
condition (ST )3 = 1 (when S 2 = 1) or (ST )6 = 1 (when S 4 = 1).
The central charge constraint in particular can be compared to the central charge (2.34)
arising from a particular choice of pseudoparticle K-matrix. Obviously, the constraints on
which fusion rules correspond to which pseudoparticle K-matrix derived this way are much
weaker than those arising from the comparison of the above recursion relations.
2.3. Simple examples
Let us illustrate the considerations of the previous section in a few examples.
Consider a CFT with two primary fields 1 and and non-trivial fusion rule = 1,
i.e.,

0 1
,
N =
(2.42)
1 0
which has eigenvalues = 1 and is diagonalized by

1 1 1
,
S=
2 1 1
(2.43)
484
which satisfies S 2 = 1. We find that (ST )3 = 1 yields the condition

h =
1
1
mod ,
4
2
(2.44)
while
1 mod 8 for h = mod 1,

4
c=
(2.45)
7 mod 8 for h = 3 mod 1.
4
Clearly, A1,1 is an example of the first possibility, while E7,1 is an example of the second.
Since c is necessarily an integer, one would conclude that as far as this calculation is
concerned no pseudoparticles are necessary. The characteristic equation for N is given by
2 1 = 0 and leads to the recursion
P (M) = P (M 2),
(2.46)
which is trivially solved by P (2M) = P (0) and P (2M + 1) = P (1). Again, this does not
require pseudoparticles, since the fusion paths are obviously unique.
Now consider A1,k for generic level k. The fusion matrix of the generating field 2 is
given by the incidence matrix of the Dynkin diagram of Ak+1 (see, for example, [23]). The
characteristic equation is thus given by

[(k+1)/2]

k+1j
(2.47)
(1)j
k+12j = 0,
j
j =1
and has roots (see, e.g., [23])

j
j = 2 cos
, j = 1, . . . , k + 1.
k+2
For example, the characteristic equation at the first few levels is given by
k = 1,
2 1 = 0,
k = 2,
(2 1) = 0,
k = 3,
4 32 + 1 = (2 + 1)(2 1) = 0,
k = 4,
5 43 + 3 = (2 3)(2 1) = 0.
On the other hand, the pseudoparticle K-matrix for A1,k , is known to be K =

while a = ( 12 , 0, . . . , 0). This leads to, e.g.,
k = 2,
2 1 = 0,
k = 3,
2 1 = 0.
(2.48)
(2.49)
1
2 Ak1 ,
(2.50)
which, in general, corresponds to a factor of (2.49) as discussed in Section 2.2.

As a final example consider a CFT with two primary fields 1 and and fusion rule
= 1 + , i.e.,

0 1
.
N =
(2.51)
1 1
485
The characteristic equation is given by

2 1 = 0,
(2.52)
with roots = 12 (1 5 ). The constraints on h and c, arising from the modular matrices,
are (see, e.g., [23], Exercise 10.16)
c 12h = 2 mod 8,
(2.53)
while
h=
m
mod 1,
5
m = 1, 2, 3, 4.
(2.54)
G2,1 is an example of a solution for m = 2 (c = 14/5, h = 2/5), while F4,1 is an example

of a solution for m = 3 (c = 26/5, h = 3/5). Examples of m = 1, 4 solutions can be found
among the minimal (non-unitary) models.
The characteristic equation (2.52) leads to the recursion relation
P (M) = P (M 1) + P (M 2),
(2.55)
the solutions of which are (generalized) Fibonacci numbers. Clearly, the recursion relation
(2.55) arises from the pseudoparticle matrix (cf. (2.40))
K = (2),
(2.56)
with a = (1).
The central charge subtraction corresponding to (2.56) is, according to (2.34), given by

3 1
2
6
L
5
= ,
(2.57)
2
2 2
5
which is not the correct subtraction for either G2 or F4 . We can however double the
subtraction while, at the same time, keeping the recursion relation, by a symmetric
doubling of (2.56), i.e., by making a 2 2 matrix with entries that sum to 2 in all
columns and rows and which is such that the solution to the IOW-equation is identical
for all components, e.g.,
4 2
K = 32 43 ,
(2.58)
3
with, a = (a1 , a2 ) where a1 +a2 = 1. This case is relevant for (F4(1) )k=1 (see Section 4.3.6).
(1)
To get a subtraction of 6/5, as needed for (G2 )k=1 , we need to do a symmetric tripling
such as
1 12 21
K = 12 1 12 .
(2.59)
1
2
Cf. Section 4.3.7.
1
2
486
3. Composite and dual composite construction

As is well known in the context of the quantum Hall effect, the K-matrices describing
the Abelian quantum Hall states are not unique, but are in fact determined up to similarity
transformations. These similarity transformations can be thought of as changing the basis
for the description. Moreover, the physical properties such as the filling fraction are not
changed by this transformation. Also the central charge is left unchanged.
A similar situation occurs when we want to view the K-matrices as the data for a
general (i.e., non-Abelian) CFT. There exist transformations of the K-matrices, which
leave the corresponding characters unchanged. Therefore, the K-matrices related by such
a transformation correspond to the same theory. A prime example will be described in
Section 3.2 and the dual version in Section 3.3. At first sight, this might be a disturbing
observation because we would like to have a unique description of the theory. However, the
situation can be used in our advantage, for instance, in the construction of the K-matrices
for general affine Lie algebra CFTs, as will be pointed out in Section 3.4.
3.1. W-transformations
To describe the well-known W-transformations (see, for instance, [47]), we will use the
notation of the fqH basis (as we will do in the rest of this section). Of course, it is applicable
to all Abelian quantum Hall systems. So we have a K-matrix Ke and the quantum number
vectors qe (the dual data is obtained by applying Eqs. (2.23) and (2.25)). Let W be an
SL(n, Z) matrix, where n is the rank of K. The W-transformation takes the form

e = W Ke WT ,
qp = W1 T Kqp W1 ,
K
(3.1)
K
while
q e = W qe ,
T

q qp = W1 qqp .
(3.2)
Indeed, physical quantities of the form qTe K1

e qe , such as the filling fraction are invariant
under this transformation. Also, the central charge, which is given by n for the Abelian
states, is not changed. In the non-Abelian case, we can also apply these W-transformations,
however, to conserve the central charge, we can only use those transformations which do
not change the pseudoparticle part of the K-matrix.
In the following, we will concentrate on constructions based on character identities (so
we view the K-matrices as matrices containing CFT data). In addition, we will show that
extended matrices obtained in this way can be used to make a reduction of the theory,
which turns out to be closely related to the W-transformations described above. We will
use the results of this section extensively in the remainder of this paper, in particular in
Section 4, where we will obtain the K-matrices for general affine Lie algebra CFTs.
3.2. Composite construction
The basic transformation one can do on a K-matrix, leaving the theory invariant, is
the composite construction [3]. The effect of this transformation is to add a particle, which
487
is the composite of two particles already present in the theory. The quantum numbers of
this composite particle are just the sum of the quantum numbers of the two constituent
particles. In order to keep the theory unchanged, one has to increase the mutual exclusion
statistics of the two constituent particles. In a sense, they avoid one another more, while
the gap is filled by the composite particle.
To make this more precise, consider the IOW-equations (2.14) with a symmetric matrix
e = K
T ), fugacities y and quantum numbers qe
Ke (i.e., a12 = a21 and K
e

a11 . . . a1n
a11 a12 aT1
.
.. a
T
Ke = ..
=
21 a22 a2 ,
.
e
a1 a2 K
an1 . . . ann

qe,1
y1
qe = qe,2 .
y = y2 ,
(3.3)
y
q e
If we define the operation C12 , corresponding to adding a composite of the quasiparticles 1
and 2 to the system, by
a11
a12 + 1
a11 + a12
aT1
a21 + 1
a22
a21 + a22
aT2
C12 Ke =
(3.4)
a11 + a21 a12 + a22 a11 + 2a12 + a22 aT + aT ,
1
2

a1
a2
a1 + a2
K
and
y1
y
C12 y = 2 ,
y1 y2
y
qe,1
qe,2
C12 qe =
,
qe,1 + qe,2
q e
(3.5)
then the two systems are equivalent, at least at the level of thermodynamics. The action of
the general Cij is defined, as above, by a suitable permutation of the rows and columns.
The solutions {i } to the IOW-equations defined by (Ke , y) and {
i } defined by (K
e , y
) =
(Cij Ke , Cij y) are simply related by
i + j 1
,
j
i j
,
n+1 =
i + j 1
i =
j =
i + j 1
,
i
k = k ,
k = i, j, n + 1.
(3.6)
Note that, in particular, it follows i =

i
n+1 and j =
j
n+1 such that tot =
tot .
Also, from i =
i
n+1 and j =
j
n+1 one sees that the original one particle partition
functions for i and j , receive contributions from the new particles i and j , respectively, as
well as from the composite particle n + 1. The operation Cij has the effect that states in the
spectrum containing both particles i and j get less dense (their mutual exclusion statistics
is bumped up by 1), while the resulting gaps are now filled by the new composite particle.
A consistency check on the equivalence of the systems described by (Ke , y) and
(K
e , y
) = (Cij Ke , Cij y) is the fact that both lead to the same central charge. It was shown
in [9] that this is in fact a consequence of the five-term identity for Rogers dilogarithm.
488
For completeness, we repeat the argument here. It is not hard to check that the solutions
to the Eqs. (2.19), with Ke and Cij Ke , which we will denote by i and i
, respectively, are
related by
i
=
i (1 j )
,
1 i j
n+1
= i j ,
j
=
k
= k ,
j (1 i )
,
1 i j
k = i, j, n + 1.
(3.7)
The equivalence of the central charge for both matrices follows from

y(1 x)
x(1 y)
+L
+ L(xy),
L(x) + L(y) = L
1 xy
1 xy
(3.8)
which is the five-term identity for Rogers dilogarithm.

Finally, we note that the composite transformation (3.4) can be derived from the
following character identity, which is a special case of the q-PfaffSaalschtz sum (see
[24])
M1
m1

M2
M1 m2 M2 m1
q (m1 m)(m2 m)
=
m2
m1 m
m2 m
m0

M1 + M2 (m1 + m2 ) + m
.
m
(3.9)
If one inserts this identity at the (i, j )th entry in the UCPF of Eq. (2.4), one finds, after
shifting the summation variables mi mi m and mi mi m, another UCPF, based
on the data (Cij K, Cij y).
The form (3.9) is used for the composite construction on two pseudoparticles. Taking
the limit M1 (M1 , M2 ) by using Eq. (2.6), gives the appropriate identity for the
composite construction applied to a physical and a pseudoparticle (two physical particles),
respectively.
3.3. Dual composite construction
Using the logic of the fqH basis, one might expect that upon inverting the extended
matrix Cij Ke , one should find a matrix, which is related to Kqp = K1
e by a character
identity as well. This turns out to be the case.
1
We will denote this transformation by Dij , thus we define Dij Kqp = (Cij K1
qp ) .
After performing this transformation, the quasiparticles corresponding to i and j have
become pseudoparticles. This is necessary, because otherwise the central charge of the
= Cij Ke Dij Kqp would have been increased by one with respect
transformed system K
to Ke Kqp , because the rank of the K-matrices is increased by one. The presence of the
extra pseudoparticles reduces the central charge by precisely the right amount, to keep the
total central charge the same (see below).
489
The action of Dij on a symmetric matrix Kqp , in the case of two (physical) particles,
can be described in the following way:

1
1
ab1
1
a b
Kqp =
,
,
D12 Kqp =
1
1
cb1
b c
a b 1 c b 1 (1 + b)2 ac
(3.10)
where = 2 (a 2b + c). In addition, in the transformed formulation, the particles 1 and
2 are pseudoparticles. When, in the original formulation, the particles i and j are physical,
it is easily verified that the reduction of the central charge, in the transformed formulation,
due to the particles i and j is in fact equal to one. This is precisely the value needed to give
the transformed system the same central charge as the original formulation, as was to be
expected.
The action of D12 on the fugacity and quantum number vectors xT = (x1 , x2 ) and
T
qqp = (qqp,1 , qqp,2 ) is given by
x 1/
1
D12 x =
x2
x2 1/
x1
(1+bc)/ (1+ba)/
x1
x2
1
D12 qqp =
qqp,1 qqp,2
qqp,2 qqp,1
(1 + b c)qqp,1 + (1 + b a)qqp,2

.
(3.11)
If we have x1 = x2 = x and hence, qqp,1 = qqp,2 = qqp , as will always be the case in this
paper, we find

1
0
D12 x = 1 ,
(3.12)
D12 qqp = 0 .
x
qqp
From a character identity point of view, the transformation (3.10) is based on the qbinomial doubling formula

M +N
M N
=
(3.13)
.
q (Mp)(Nq)
n
p
q
pq=Mn
Indeed, considering the UCPF for two physical particles with Kqp as in Eq. (3.10), i.e.,
Z=
q 12
am21 +cm22 +2bm1 m2
(q)m1 (q)m2
q 12
am21 +cm22 +2bm1 m2
(q)m1 +m2

m1 + m2
,
m1
(3.14)
and then applying (3.13) with

M = (b c)m1 + (1 + b a)m2 ,
N = (1 + b c)m1 (b a)m2 ,
n = m1 ,
(3.15)
490
to the q-binomial in (3.14), results in the UCPF based on D12 Kqp of (3.10), with the
identifications
m
1 = p,
m
2 = q,
m
3 = m1 + m2 ,
(3.16)
and where the first two particles in D12 Kqp are pseudo.
The general case can be derived from (3.13) as well, and is described in the following
way. Again, we will focus on the case where we let D work on the first two particles. In
addition, we will assume that both those particles are physical. For ease of presentation,
we now define = 2 (b11 2b12 + b22 ), 1 = 1 + b12 b11 and 2 = 1 + b12 b22 .
Using similar notation as in Eq. (3.3), we take (the symmetric) Kqp , the fugacities and
quantum numbers
b
11
b12
Kqp = b21
b22
b1
b2
bT1
bT2 ,
qp
K

x=
x1
x2
x

,
qqp =
qqp,1
qqp,2
q qp

.
(3.17)
The dual composite construction, applied on the first two particles is given by
D12 Kqp =
1
1
1
b1 b2
T
bT
1 b2
T
bT
2 b1
(1 + b12 )2 b11 b22
b2 b1
2 b1 + 1 b2
T
2 bT
1 + 1 b2
qp + (b1 b2 )i (b1 b2 )j
K
ij
(3.18)
The first two particles have become pseudoparticles, while the extra particle is a physical
particle. Note that this construction based on the character identity Eq. (3.13) only works
in the case that the particles on which it is applied are physical particles. We have not
found a character identity for the case where the dual composite construction is applied to
two pseudoparticles. However, we will show below that also in that case the central charge
works out alright, so we suspect that there is indeed a character identity relating the two
systems.
The action of the dual composite construction on the fugacities and quantum number
vectors is given by
x1 1/
D12 x =
x22 1/
x1
,
/ /
x12 x21
x1 (b1 b2 )i /
xi x2
qqp,1 qqp,2
1
qqp,2 qqp,1
D12 qqp =
.
2 qqp,1 + 1 qqp,2
q qp + (b1 b2 )(qqp,1 qqp,2 )
(3.19)
491
Again, specifying to the situation where x1 = x2 = x and qqp,1 = qqp,2 = qqp , as holds in
all the cases we consider, we find
1
0
1
0
D12 qqp =
D12 x = ,
(3.20)
.
x
qqp
x
q qp
The solutions {i } to the IOW-equations defined by (Kqp , x) and {
i } defined by
(K
qp , x
) = (Dij Kqp , Dij x) are, as was the case for the composite construction (compare
(3.6)), related in a simple way
i =
i j 1
,
j 1
n+1 = i j ,
j =
i j 1
,
i 1
k = k ,
k = i, j, n + 1.
(3.21)
Using the relations (3.21) it is not hard to show that the IOW-equations based the two
systems (Kqp , x) and (Dij Kqp, Dij x) are in fact equivalent. We also find that tot =
tot
by using the fact that the particles i and j are pseudoparticles after the dual composite
construction has been applied. The composite particle which is created is a physical
(pseudo) particle if particles i and j are physical (pseudo) in the original description.
From Eq. (3.21) it follows that the dual composite construction cannot be applied on a
physical and pseudoparticle. In that case,
tot cannot be made equal to tot . Note that such
a restriction does not apply to the composite construction of Section 3.2. Though we do
not quite understand this difference, it will not affect any results in this paper.
Let us now focus on the central charge, and look at the case in which all the particles
are physical particles first. Because the rank of the transformed matrices is increased by
one, we need that the two created pseudoparticles reduce the central charge by one. This
is easily verified. Also, because the central charge of the matrix Cij Ke equals the central
charge of Ke , we need to find the result that the central charge related to Dij Kqp without
the pseudoparticle subtraction equals the central charge related to Kqp plus one. To show
this, we need to relate the solutions to Eqs. (2.19), which we denote by i and i
for Kqp
and Dij Kqp , respectively. The relations are given by
i
=
i
,
i + j i j
n+1
= i + j i j ,
j
=
j
,
i + j i j
k
= k ,
k = i, j, n + 1.
(3.22)
Because of the relation between the central charges, we require the following dilogarithm
identity

y
x
+L
+ L(x + y xy) L(1),
L(x) + L(y) = L
x + y xy
x + y xy
(3.23)
which is easily derived from Eq. (3.8) by applying Eq. (2.33) to each term, and making the
change of variables (x 1 x, y 1 y).
The argument above not only shows that the central charge works out correctly in the
absence of pseudoparticles. It can also be used to show that the reduction of the central
charge increases by one if we apply the dual composite construction on pseudoparticles.
492
What remains to be checked is the central charge if we apply the composite construction
to physical particles, while pseudoparticles are present. For this, we need to compare the
central charge equations for the original pseudoparticles with the ones where the additional
two pseudoparticles are present. Though non-trivial, one can convince oneself that the
solutions to the central charge equations of the original pseudoparticles do not change,
while the solutions for the two pseudoparticles which are introduced add up to one and
therefore increase the reduction by one, which gives the correct result.
3.4. P-transformations
In this section, we will discuss a transformation which is based on the (dual) composite
construction. This construction is very useful in determining K-matrices for general affine
Lie algebra CFTs. We will motivate this construction by using a simple example, which
captures the essence of the method. In the end, this P-transformation is very similar
to the W-transformations described in Section 3.1, with one important difference. After
applying a P-transformation, some of the physical quasiparticles have transformed into
pseudoparticles. One of the consequences of this is a reduction of the central charge.
As we will use the P-transformations mainly as a tool to obtain K-matrices for levelk affine Lie algebras from the direct sum of k level-1 algebras, we will explain the
construction using the simplest case. Afterwards, we will present the general case. In the
next section, we will use the results obtained here to find the K-matrices we are after.
3.4.1. The case sl(2)2
The goal in this section is to obtain the K-matrices for the sl(2)2 affine CFT, which
describes the MooreRead (or Pfaffian) quantum Hall state. The corresponding matrices
are known, see, for instance, [2,3,45]. Let us recall the K-matrices for the (bosonic) = 1
case, which corresponds to sl(2)2

2 2
1
2 =
Ksl(2)
=
(3.24)
,
t
.
e
e
2 4
2
The first particle can be identified with the (bosonic) electron, while the second is a
composite of two electrons. In the quasiparticle sector

1 12
0
1
2 = K1 =
=
K
t
=
,
t
Ksl(2)
(3.25)
qp
e
1 ,
qp
e
e
1
12
2
2
where the first particle is a pseudoparticle. The K-matrices for the general MooreRead
1
state, at filling fraction = M+1
, are obtained by applying the so-called shift map, which
is described in detail in [3]. Though the theory for general M has the same central charge,
the theory does not have the underlying sl(2)2 structure anymore, but rather a deformation
(along the charge direction) of this. In this paper, we concentrate on the M = 0 case
throughout; the K-matrices for general M are obtained by applying the shift map as
indicated above. Note that the pseudoparticle matrices K are unchanged under this shift
map.
The main idea is now to obtain these sl(2)2 matrices via an embedding of sl(2)2 in
sl(2)1 sl(2)1 (which we will call an Abelian covering, see also [13]). By introducing
493
a composite, and projecting out some degrees of freedom, we obtain the K-matrices for
sl(2)2 . In physical terms, we start from two, uncoupled, quantum Hall layers with filling
= 12 (these are in fact bosonic Laughlin states). In a sense, this state is a covering state
for the MooreRead state at filling = 1. By increasing the interactions between the two
layers, one might encounter a phase transition to the MooreRead state, as described in
[30]. The bosons form pairs, and condense. In the terminology of an effective Landau
Ginzburg theory (see [22]), the difference of the gauge fields describing the bosons acquires
a mass, and decouples from the spectrum. This is the Meissner effect.
On the level of the K-matrices, we can describe this in the following way. We first
introduce the composite of the two bosonic particles, and afterwards simply delete (or
project out) one of the original bosons. So we actually reduced the theory, as required.
We start with the direct sum of two sl(2)1 K-matrices

2 0
1
cover
=
,
te =
,
Ke
(3.26)
0 2
1

1
1
0
Kcover
(3.27)
tqp = 21 .
= 2 1 ,
qp
0 2
2
Now, applying the composite and dual composite constructions (Eqs. (3.4) and (3.18)) on
these matrices gives the following, equivalent description

2 1 2
1

te = 1 ,
Ke = 1 2 2 ,
(3.28)
2 2 4
2

1
0 12
0
1
qp = 0
tqp = 0 .
K
(3.29)
1 ,
2
12
12
1
2
3
4
Note that the first two particles of the quasiparticle matrix are pseudoparticles. To obtain
the sl(2)2 matrices, we have to project out one of these pseudoparticles, by putting it into
the vacuum state. In addition, we discard one of the original bosons.
However, while projecting out one of the bosons in the electron sector simply
e , projecting out one of the
corresponds to deleting the respective row and column in K
pseudoparticles is more subtle, due to the negative coupling between the pseudoparticles
qp .
and the physical particle in K
qp of (3.29)
For explicitness, consider the UCPF corresponding to K
q 12
2

m1 +m22 (m1 +m2 )m3 + 34 m23
1
(q)m3
2 m3
m1
2 m3
m2
(3.30)
qp , the vacuum state

Due to the minus-sign in the coupling between particles 2 and 3 in K
1
for particle 2 is not achieved for m2 = 0, but rather for m2 = 2 m3 . Hence, rather than just
qp , we need to set m2 = 1 m3 in the bilinear form. This results in
omitting particle 2 from K
2
qp m = m2 + m2 (m1 + m2 )m3 + 3 m2
mT K
1
2
4 3
494
2

1
1
3
1
+
(3.31)
m3 m1 + m3 m3 + m23 = m21 m1 m3 + m23 ,
2
2
4
2
which precisely corresponds to the matrix Kqp of (3.25).
To summarize, the results of projecting out degrees of freedom in Eqs. (3.28) and
(3.29), gives rise to the K-matrices of Eqs. (3.24) and (3.25). One of the key points of this
section is that there is an elegant way of going from K-matrices for the (Abelian) coverings
(Eqs. (3.26) and (3.27)) to the K-matrices of sl(2)2 , by what we call a P-transformation.
This also hold for the general case, as we will show below. We find

2 = P Kcover PT ,
2 = P1 T Kcover P1 .
Ksl(2)
Ksl(2)
(3.32)
e
qp
e
qp
m21
The vectors containing the quantum numbers (denoted by qe and qqp ) transform as
T

q qp = P1 qqp .
q e = P qe ,
(3.33)

1 0
1
1
In the above, we have to take P = 1 1 , and hence (P1 )T = 0 1 . A few remarks need
to be made here. First of all, the P-transformation described by Eqs. (3.32) and (3.33)
closely resembles the W-transformation, as they act on the K-matrices in the same way
(compare (3.1)). However, there are a few important differences. As we explained above,
upon applying a P-transformation, we introduced a pseudoparticle in the quasiparticle
sector. This is important, as the presence of a pseudoparticle changes the theory. For
instance, the central charge is reduced, in the case at hand by 1/2, which is precisely
the difference in central charge between sl(2)1 sl(2)1 and sl(2)2 (given by c = 2 and
c = 3/2, respectively). So the P-transformation actually changes the theory, while the Wtransformation is a basis transformation, which does not change the theory.
In the remainder of this section, we will show how a P-transformation works on a
general K-matrix. These results are used in the next section to find the K-matrices for
the general affine Lie algebra CFTs, in a similar way as we constructed the sl(2)2 matrices
above.
3.4.2. The general case
In this section, we will relate the introduction of a composite (in the electron sector), and
the corresponding transformation in the quasiparticle sector to a general P-transformation.
For notational simplicity, consider introducing a composite of particles 1 and 2 in a general
symmetric K-matrix as given by Eq. (3.4). Now, suppose we delete particle 2 from the
e , q e ) given by
resulting matrix C12 Ke , we then find a new K-matrix system (K

a11
a11 + a12
aT1
q1
T
T
e = a11 + a21 a11 + 2a12 + a22 a1 + a2 ,
K
q e = q1 + q2 . (3.34)
q

a1
a1 + a2
K
e , q e ) and (Ke , qe ) as
Notice that we can write the relation between (K
e = P K e PT ,
K
with

P=
1
1
0
0 0T
1 0T
0 I
q e = P qe ,
(3.35)

.
(3.36)
495
Now consider the dual composite construction D12 Kqp (see Eq. (3.18)). In analogy with
Eq. (3.30), putting the second pseudoparticle in its vacuum state amounts to setting
.
m2 = ( 1)m1 + 2 m3 (b2 b1 ) m
(3.37)
Substituting this in the quadratic form yields, after a lengthy calculation,

qp m,
mT (D12 Kqp) m mT K
qp is given by
where K
b11 2b12 + b22

b12 b22
Kqp =
b12 b22
b22
b1 b2
b2
which is related to Kqp by

qp = P1 T Kqp P1 ,
K
with

1 T

=
1 1
0 1
0 0
0T
0T
I
(3.38)
bT1 bT2
bT2 ,

K
(3.39)
(3.40)

,
(3.41)
qp in
in accordance with Eq. (3.36). It is important to note that the first particle of K
Eq. (3.39) is a pseudoparticle. The presence of this pseudoparticle causes the reduction
qp with respect to Ke Kqp . Of course, this is
e K
of the central charge of the system K
to be expected when degrees of freedom are projected out.
Summarizing, a P-transformation acts on the K-matrices and quantum number vectors
(denoted by qe and qqp ) as follows:

qp = P1 T Kqp P1 ,
e = P K e PT ,
K
K
(3.42)
and
q e = P qe ,
T

q qp = P1 qqp ,
(3.43)
where in addition, some of the quasiparticles have been transformed into pseudoparticles.
In Section 4.1 we will repeatedly use the (dual) composite construction combined with
the projecting out of degrees of freedom to determine K-matrices for a variety of CFTs.
Rather than specifying the particles to which we consecutively apply this construction we
will simply state the required resulting P-transformation, and specify which quasiparticles
have become pseudoparticles.
Because the P-transformations take the form (3.42), properties such as the filling
fraction (see (2.26)), are not changed upon performing the P-transformation. Of course, the
statistics properties are changed in a profound way, because the induced pseudoparticles
lead to non-trivial fusion rules as described in Section 2.2. In turn, this leads to the nonAbelian statistics of the physical quasiparticles (see, for instance, [38]).
One important remark needs to be made before closing this section. In the construction
of the K-matrices, we will use the (dual) composite construction via the P-transformation.
496
We will always apply the dual composite construction to identical (quasi)particles. Hence,
the quantum numbers of the quasiparticles (and also their electronic equivalents) are the
same. Moreover, we will always have aii = ajj and bii = bjj . As a result, it does not
matter which of the electron-like particles is projected out. If aii = ajj , the two different
projections are related by P
= PT . The general form for P we use in this paper will be
discussed in the next section (see, in particular, Eq. (4.13)).
4. K-matrices for affine Lie algebras

One of the main themes of this paper is the identification of the K-matrices for general
affine Lie algebra CFTs. We will work in the so-called quantum Hall basis, as described
above. In [3] (see also [2]), the K-matrices corresponding to the sl(2)k and sl(3)k CFTs
were derived. Here, we will give an alternative construction of the k > 1 cases directly from
the k = 1 cases, which can be found in [2]. This construction is based on the embedding
of the level-k theory in the direct sum of k level-1 theories. By applying composite and
dual composite constructions, we introduce pseudoparticles. After projecting out some of
these, we have reduce the theory to the level-k theory. We will phrase all of this in terms
of the P-transformations of the previous section. Apart from the sl(2)k and sl(3)k theories,
we will also use this construction for the other (simply-laced) affine Lie algebra cases,
and provide a few non-trivial checks to show that we indeed found the correct K-matrices.
The non-simply-laced cases can be obtained by embedding the level-1 affine algebras into
simply-laced algebras, and performing a similar construction as outlined above.
4.1. Constructing the matrices
We will use the techniques described in the previous section to construct the K-matrices
for general affine Lie algebras.
In the this section, we will describe how this works in detail for the simplest examples,
which have all the characteristics of the general case. Motivation of this construction can
be found in the previous section. In Sections 4.2 and 4.3 we will present the results for the
K-matrices for general affine Lie algebra CFTs.
4.1.1. Example: the case sl(2)k
Let us illustrate the construction for the level k > 2 generalizations of the MooreRead
states, the so-called ReadRezayi states [41]. The covering state in this case is the direct
sum of k level-1 theories (instead of just 2 for the MR case). So we have

2
1
2
1
,
tcover
..
.
=
=
Kcover
(4.1)
..
e
e
.
.
2
(Here, and in the following we use the convention that empty entries contain zeroes, if not
implied otherwise by dots.) We also indicated the charge vector, containing the charge
quantum numbers of the particles, as the transformation behavior of the quantum numbers
497
under the P-transformation clearly shows that composites are introduced. To obtain the Kmatrices for sl(2)k , describing the ReadRezayi states, we need to introduce all types of
composites, from a pair up to a cluster made out of the k original particles. Thus P takes
the following form:
1 1
.
P=
(4.2)
... . . . . . .
1
This leads to following matrix Ke and charge vector te (by using Eqs. (3.42) and (3.43))
2 2 2 2

1
2 4 4 4
2 4 6 6 ,
te =
Ke =
(4.3)
. . . .
... ,
.. .. ..
. . ...
k
2 4 6 2k
which are indeed correct for the sl(2)k theory. The dual sector is simply obtained by using
the duality relations (2.23), (2.25). Alternatively, we can apply the dual P-transformation
on the dual (i.e., the inverse) of the covering matrix Eq. (4.1). The corresponding P-matrix
is
1 1
..
1 T
.
=
P
(4.4)
,
..
. 1
1
from which we find
1 12
1
1
2
Kqp =
..
12
12
1
12
1
2
1
2

0
...
tqp =
0 .
(4.5)
1
2
From the matrix equation (4.4) we read of that the first k 1 particles are pseudoparticles.
These results are in perfect agreement with the results of [3,26].
4.1.2. Example: the case sl(3)k
As an example of a case where the rank n of the affine Lie algebra is greater than 1,
we show that a similar construction can be carried out to obtain the K-matrices related to
the sl(3)k CFT. This is the underlying theory of the non-Abelian spin-singlet quantum
Hall states as defined in [5]. Finding the K-matrices when the rank n > 1 is somewhat
more complicated than for n = 1. The K-matrices for the sl(3)k CFT were obtained in [3].
There, the basis was chosen in such a way that all the particles in the electron sector had
the same sign for the charge. The reason for this choice was that the electron operators (for
498
spin up and spin down) appearing in the construction of the quantum Hall state have the
same sign of the charge. These electron operators are associated to the roots 1 and 2
of sl(3). From mathematical point of view, it is more natural to work with 1 and 2 , as
the resulting K-matrices have a simpler structure. So here we will present the results using
the (mathematically) more natural formulation, based on the positive roots. In Appendix
D, we will explain the precise relationship between the two descriptions. Essentially, the
relation is a W-transformation on the physical particles, which leaves the pseudoparticles
unchanged. This is required, because the pseudoparticles are related to the fusion rules of
the affine Lie algebra and they also determine the central charge. The K-matrix for the
electron sector at level 1 takes the form in the representation chosen here

2 1
1
1
,
te =
,
se =
.
Ke =
(4.6)
1 2
1
1
In the other formulation, used in [3], the off-diagonal elements of Ke are 1, while the role
of te and se is interchanged.
The K-matrix in Eq. (4.6) is the building block of the covering matrix, from which we
construct the level-k K-matrices
2 1
1 2
2 1
cover
,
1
2
=
Ke
..
2 1
1 2

1
1
1
1

1
1

cover
1 ,
1.
s
=
=
tcover
(4.7)
e
e
.
.
..
.
.

1
1
1
1
At this point, we need to specify the matrix P, which is used to project to the K-matrix
for the sl(3)k theory. However, because we have n = 2 in this case, we can construct the
composites (up to order k) in different ways. We will first state the form which gives the
correct result, and comment on the other possibilities afterwards. The P-transformation
which gives the correct central charge is given by
I2
I2 I2
,
P=
(4.8)
... . . . . . .
I2 I2 I2
where I2 is the 2 2 identity matrix. The resulting K-matrix has the following form
(explicit forms of the Cartan matrix A2 of A2 and the symmetrized Cartan matrix M1
k
of Bk can be found in Appendix A)
2
1
2
1
.
.
Ke = A2 Mk = .
2
1 2
2 1
1 4
2 2
..
..
.
.
1 4
1
2
2
4
..
.
2
..
.
2
1
4
2
..
.
2(k 1)
1
2
2
4
..
.
(k 1)
2
1
4
2
..
.
2(k 1)
(k 1)
2(k 1)
499
1
2
..
.
,
(k 1)
1 2 2 4 (k 1) 2(k 1) (k 1) 2(k 1)
2 1 4 2 2(k 1) (k 1)
2k
k
1
2k
(4.9)
while the charge and spin quantum numbers are given by
1
1
2 ,
te =
.
.
.
k
k

1
1

2

2
se =
. .
.
.

k
k
(4.10)
It is not to hard to see that introducing the composites can be done in different ways. For
instance, we could move some of the 1s in the lower-triangular part of the matrix P of
Eq. (4.8) to the corresponding place in the upper-triangular part. If done systematically,
we still would introduce all the composites, so the resulting quantum numbers would be
the same. Luckily, all the essentially different possibilities result in different K-matrices,
which have different central charge associated to them. So we can pick the, presumably,
correct description by looking at the central charge and perform further checks to assure the
validity of the chosen matrices. In all the cases we encountered, only one P-transformation
gave rise to a rational central charge (as far as the numerical checks could tell), which
indeed was the central charge corresponding to the affine Lie algebra CFT. We refer to
Section 4.3 for more details on the checks of the central charge associated to the Kmatrices. Whether or not the other possibilities correspond to (non-rational) CFTs is not
clear at the moment.
The K-matrices and quantum numbers for the quasiparticle sector are obtained similarly
as in the sl(2)k case, by applying the dual P-transformation to the dual of the covering.
Now, the transformation matrix becomes the inverse transpose of Eq. (4.8)

1 T
I2
I2
I2
..
..
I2
I2
(4.11)
500
with the results

1
A2 Ak1
1
1
Kqp = A2 Mk =
A1
2
0
..
.
sqp = 0 .
1
1
0
...
tqp = 0 ,
1

3
1
3
A1
2 ,
A1
2
(4.12)
The K-matrix is to be compared with the matrix (7.23) in [3]. Note that part of the K-matrix
corresponding to the 2(k 1) pseudoparticles is the same in both cases. So, because we
know the two descriptions are related (see Appendix D), we can say that by using the
method of the P-transformations, we were able to obtain correct K-matrices for the sl(3)k
theory. One important check is the central charge. Because the pseudoparticles are the
same in both formulations, the central charge is also equal. In Section 4.3, the quasiparticle
matrices for all simple affine Lie algebra CFTs will be given. The electron matrices are
specified in Section 4.2. Before we come to that, we will first describe in detail how to
construct the general K-matrices, using the P-transformations and suitable coverings.
4.1.3. The general case
Using the knowledge obtained in the previous section, we go on and propose a scheme to
obtain the K-matrices for general affine Lie algebra CFTs. We will first concentrate on the
simply-laced cases, and discuss the non-simply-laced cases afterwards. As we discussed
the case of sl(3), which has all the essential ingredients, in detail in the previous section,
we will be brief here. We saw that in the case of sl(3)1 , we could use the particles related
to the simple roots as the basis of the electron sector. Simple roots are roots which cannot
be written as a sum of two positive roots. A Lie algebra of rank n has n simple roots,
and their scalar products define the Cartan matrix. So we found that the K-matrix for the
electron sector of sl(3)1 was the Cartan matrix. In the following, we will assume that this
is the case for all the simply-laced affine Lie algebras. What we need to do further to obtain
the level-k K-matrices is construct the covering theory, which is just the direct sum of k
level-1 theories, and apply the correct P-transformation. The form of the P-transformation
is similar to the sl(3) case, where the rank is the only thing which needs to be changed. So
we find P for the simply-laced cases
In
In
P= .
..
In
In
..
.
..
In
In

1 T
In
In
In
..
..
In
In
(4.13)
501
Applied to the covering matrix we find the result Ke = P (An Ik ) PT = An Mk . See

Section 4.2 for an explicit example. Of course, An can be replaced by the Cartan matrix
of any other simply-laced algebra, Dn or En . The K-matrix for the quasiparticle sector is
1
1
obtained by applying (P1 )T to the dual covering A1
n Ik , resulting in Kqp = An Mk .
1
T
From the form of (P ) we find that the first n(k 1) particles are in fact pseudoparticles.
These matrices will be given explicitly in Section 4.3. For now, we note that the central
charge associated to these systems does indeed have the correct value. More on this can be
found in Section 4.3.
Let us now focus our attention to the non-simply-laced case. The idea is to apply the
same construction as for the simply-laced cases. However, we need to find the correct
starting point, that is, the level k = 1 formulation. The non-simply-laced affine algebras
have non-trivial fusion rules already at level-1, so we already need pseudoparticles at
level-1. This is also reflected in the central charge, which is non-integer. To find the Kmatrices, we embed the non-simply-laced algebra in a simply-laced one, and basically
do the same construction as before: project out some degrees of freedom by introducing
pseudoparticles. As an example, we quote the case for so(5)1 , which is related to the spincharge separated quantum Hall states of [4] (see also [9,11]). There, the K-matrices for the
so(5)1 were obtained from the so(6)1 K-matrices using the construction outlined above. It
turns out that in general, the matrices for the non-simply-laced affine Lie algebras are equal
to the (simply-laced) affine Lie algebra in which they are embedded. The difference is the
presence of pseudoparticles in the non-simply-laced cases, as described above. Alternative
descriptions are possible, e.g., for G2,k we have an alternative description (which is used
in connection with the corresponding parafermion CFT), where the k = 1 K-matrix has
a couple of sign changes in comparison to the Cartan matrix of the algebra used for the
embedding, see Appendix C.
To check that we indeed found the correct matrices, we will provide another way to
obtain the K-matrix for non-simply-laced CFTs at level one. This time, we directly use
the exclusion statistics parameters of the electron-like operators, corresponding to the root
lattice of the algebra. It is important to know the exclusion statistics of the corresponding
parafermions (which are part of the electron operators, see Section 5.3 and also [25]), but
we can borrow results from the literature here. We will show how this works for the case
so(5)1 in Appendix B, while G2 at level-1 can be found in Appendix C. The other nonsimply-laced cases can be obtained in a similar way.
Having identified the k = 1 K-matrices for the non-simply-laced algebras, we can go
on, and take the direct sum of k of the level-1 matrices, and do exactly the same Ptransformations as in the simply-laced case. Because the covering matrices for the nonsimply-laced cases are identical to the ones used for the corresponding simply-laced cases,
the resulting K-matrices will be identical as well. The only difference is the number of
pseudoparticles, as there will be more pseudoparticles in the non-simply-laced case. So,
specifying the nature of the particles is the only way to tell the difference between the two.
It is important to note that in the P-transformation, (dual) composites are made only out of
identical particles. We never have the situation where a physical particle is paired with a
pseudoparticle, in accordance with the results of Section 3.3.
502
4.2. The matrices Ke

The building blocks of all the K-matrices are the Cartan matrices An , Dn , En and their
inverses. In addition, we need the symmetrized Cartan matrix of Bn , which we denote by
Mk , and its inverse. All these matrices can be found explicitly in Appendix A.
From Section 4.1.3, we have the results that for the simply-laced cases An,k , Dn,k and
En,k the matrices Ke take the form An Mk , Dn Mk and En Mk , respectively. As an
example, we will give the result for D4,2 explicitly
2 1 0
0
2 1 0
0
1 2 1 1
1 2 1 1
0 1 2
0 1 2
0
0
0 1 0
2
0 1 0
2
.
Ke = D4 M2 =
(4.14)
0
4 2 0
0
2 1 0
1 2 1 1
2 4 2 2
0 1 2
0
0 2 4
0
0
For the non-simply-laced cases, we have to take the Cartan matrix corresponding to affine
Lie algebra which we used for the embedding. We find that the matrices Ke are Dn+1 Mk ,
A2n1 Mk , E6 Mk and D4 Mk for Bn,k , Cn,k , F4,k and G2,k , respectively.
4.3. The matrices Kqp
The matrices Kqp can be obtained from Ke by a simple inversion (see (2.23)). In the
following, we will explicitly give these matrices, and indicate which particles are in fact the
pseudoparticles. With this knowledge, one can calculate the central charge corresponding
to Ke Kqp by using Eq. (2.34). As this is hard to do analytically in general, we determined
the central charge numerically for some low values of (n, k). All the cases up to rank n = 10
have been checked up to level k = 20. We found that the central charge corresponding to
the matrices was equal to the central charge of the CFTs up to 1020 or better. The central
charge of an affine Lie algebra CFT is given by (cf. (2.13))
cALA =
k dim Xn
,
k + h
(4.15)
where dim Xn is the dimension and h the dual Coxeter number of the Lie algebra Xn .
Both can be found in Appendix A for every simple Lie algebra.
In the following, we will denote the ith column of the matrix M by (M)i . Recall that
the quasiparticle matrices are of the form (see Eq. (2.29))

K
K
,
Kqp =
K
K
KT = K ,
KT = K ,
KT = K ,
where denotes the pseudoparticles, and the physical quasiparticles.
(4.16)
503
4.3.1. The case An,k

The quasiparticle matrix Kqp for sl(n + 1)k is given by
1
An Ak1
1
Kqp = A1
M
=
n
k
A1
n
A1
n .
(4.17)
A1
n
In particular, the pseudoparticle matrix is given by K = A1

n Ak1 .
4.3.2. The case Bn,k
As already pointed out, we need to an embedding to obtain the Bn,1 description first.
This is done for so(5) in Appendix B, where we used D3,1 for the embedding. In general,
we need Dn+1,1 . We find that we need one extra pseudoparticle, which corresponds to
the first node of the Dynkin diagram of Dn+1 . This extra particle has exclusion statistics
parameter 1, which gives a reduction of the central charge by 12 , which is indeed the
difference of the central charge of the theories Dn+1,1 and Bn,1 . At general level we find
1
that Kqp = D1
n+1 Mk , which is characterized by
2D1
n+1
D1
n+1
D1
n+1
2D1
n+1
..
.
..
..
D1
n+1
D1
n+1
2D1
n+1
1 T
Dn+1 1
1
Dn+1 1
(4.18)
where we see explicitly that there is an extra pseudoparticle next to the D1

n+1 Ak1 part.
Accordingly, the matrix K is the inverse Cartan matrix of Dn+1 , with the first row
and column omitted (denoted by D1
n+1 |/1 )
2
2
.
.
.
K = D1
|
=
/
1
n+1
2
1
1
2
3
..
.
3
..
.
2
3
..
.
n1
3
2
3
2
n1
2
n1
2
..
. .
n1
2
n1
4
3
2
3
2
..
.
n1
2
n+1
4
n1
4
(4.19)
n+1
4
Finally, we have
0
1
D
K =
n+1 2
1
12
D1
n+1 n+1 ,
12
(4.20)
504
where 0 stands for the (column) vector with all zeroes (of length (n + 1)(k 2) in this
case). Putting the parts together, we find

K K
1
= D1
Kqp =
(4.21)
n+1 Mk .
KT K
To put emphasis on the fact that the pseudoparticle matrix is bigger that the one for the
Dn+1 CFT, we gave the matrices K etc. explicitly, as we will do for all non-simplylaced cases.
4.3.3. The case Cn,k
In this case, we need A2n1,k as the theory for the embedding. For level k = 1 we need
the particles corresponding to the even nodes to be pseudoparticles. These will be the extra
pseudoparticles for k > 1, giving n 1 extra pseudoparticles. We again will specify the
matrix Kqp by its parts K , etc.
2A1
A1
2n1
2n1
1
..
1
A
.
2n1 2A2n1
..
..
.
.
A1
2n1
=
A1
2A1
2n1
2n1

T
A1
2n1 2
..

A1
2n1 2
T

A1
2n1 2n2
1
K = A2n1 1

A1
2n1 3
Ke
0
1
A2n1 2n1 .

A1
2n1 2n2
2A1
n1
(4.22)
(4.23)
The matrix Ke , which contains the coupling between the physical particles and the extra
pseudoparticles, is described most easily by specifying its entries explicitly. Let us first
recall the elements of the inverse Cartan matrix of A2n1 (compare with Appendix A)
1
(2i 1)(2j 1)
A2n1 2i1,2j 1 = min(2i 1, 2j 1)
,
2n
i, j = 1, . . . , 2n 1.
(4.24)
Then we have

(2i 1)2j
,
Ke i,j = min(2i 1, 2j )
2n
i = 1, 2, . . . , n, j = 1, 2, . . . , n 1.
(4.25)
For the matrix K we have

(2i 1)(2j 1)
, i, j = 1, 2, . . . , n.
(4.26)
2n
Note that the elements of the matrix describing the extra pseudoparticles (Ke e )i,j =
)
1
min(2i, 2j ) (2i)(2j
2n , where i, j = 1, . . . , n 1 is indeed equal to 2An1 .
K
i,j
= min(2i 1, 2j 1)
505
4.3.4. The case Dn,k

As we already used the matrix corresponding to Dn+1,k in the case of Bn,k , we will be
brief here:
1
Dn Ak1
D1
n .
Kqp = D1 M1 =
(4.27)
n
D1
n
D1
n
So we have n(k 1) pseudoparticles, and n physical ones.

4.3.5. The cases En,k with n = 6, 7, 8
For En,k , we simply have a similar result as for the other simply-laced cases:
1
En Ak1
1
E1
Kqp = E1
n .
n Mk =
E1
n
(4.28)
E1
n
so the n(k 1) pseudoparticles couple via En Ak1 .

4.3.6. F4,k
The embedding used this time is based upon E6,k . Now we expect to have two extra
pseudoparticles, based on the level-1 case (cf. (2.58), Section 2.3), which turns out to be
true. The couplings of these extra pseudoparticles are related to the nodes 1 and 5 (see
Appendix A). For general k, we have the pseudoparticle matrix
1
E1
2E6
6
..
E1 2E1
.
6
6
..
..
1
.
.
E
6
,
K =
(4.29)
1
1
1
1
E6
2E6
E6 1 E6 5
T

4
2
E1
6 1
3
3
1 T
2
4
E6 5
3
3
while the physical particles have
10
4 83 2
3
4 6 4 3
K =
8 4 10 2 .
3
(4.30)
Physical and pseudoparticles are coupled via
0
0
0

1
1

E1
6 2 E6 3 E6 4
K =
4
5
2
3
3
4
3
5
3
0
1
E6 6
1
1
(4.31)
506
Again, if we combine the physical and extra pseudoparticles in the right way, we find the
matrix E1
6 .
4.3.7. G2,k
Finally we come to the last case, which is G2,k . This case is special in the sense that if
we use a similar procedure as we used in all the other cases, we find a description in which
the number of physical particles does not equal the rank of the algebra, as was the situation
in the other cases. This will have consequences as we consider the related parafermions in
Section 5.3. In Appendix C we will provide a different description of G2,k , which does have
two physical particles. For now, we will just use the description based on the K-matrices
for D4,k , in which we embed G2,k . It turns out that we need three extra pseudoparticles,
leaving only one physical particle. Note that the coupling of the extra pseudoparticles is
given by Eq. (2.59) in Section 2.3.
2D1
D1
4
4
..
D1 2D1
4
4
..
..
1
.
.
D
4
1
1
1
1
1
K =
D4
2D4
D4 1 D4 3 D4 4 ,

T
1
1
D1
1
4
2
2
1
1 T
1
1
D4 3
1
2
2
1 T
1
1
D4 4
1
2
2
(4.32)

K = 2 ,
(4.33)
0
1
D4 2
.
K =
(4.34)
1
1
1
5. K-matrices for coset conformal field theories

Having identified the K-matrices for the affine Lie algebra CFTs, one might hope to find
K-matrices for more general CFTs. An obvious class to look at are the coset conformal
field theories, as most CFTs can be written in a coset form. In this section, we will provide
K-matrices for a class of coset CFTs. In our search for the K-matrices for coset CFTs, we
will be mainly guided by the central charge. We can test our results by comparing to known
coset K-matrices. For diagonal cosets of simply-laced affine Lie algebras, the results of the
K-matrices are due to McCoy and co-workers. See, for instance, [6].
Having obtained a scheme, we will apply it to the cosets so(2n)k /so(2n 1)k with
k = 1, 2, where the latter is the non-trivial one. The parafermionic cosets are dealt with in
507
Section 5.3, as they require a different approach. This already shows that the scheme we
found is by no means unique, but useful anyway.
5.1. Diagonal cosets
As said, the central charge is an important quantity to keep in mind in determining the
K-matrices for the cosets. Let us take a look at the general coset G/H , where H G is
maximal. Let us assume that both G and H are of the form Ke Kqp , with equal rank
n. Also, both quasiparticle matrices can contain pseudoparticles. So the central charge of
these theories (denoted by c(G) and c(H )) is given by

c(G) = n c K (G) ,
(5.1)
c(H ) = n c K (H ) ,
where c(K (G)) denotes the central charge corresponding to the pseudoparticle matrix of
G. Let us further assume that all the pseudoparticles which appear in K (G) also appear
in K (H ). This restricts the applicability of the construction, but still covers a large class
of cosets. Now the argument of the central charge suggests to take the pseudoparticle Kmatrix of H , and change the pseudoparticles which do not appear in the pseudoparticle
matrix of G into physical particles. The central charge corresponding to this matrix is
c(K (H ))c(K (G)). This indeed equals the central charge of the coset theory, which
is given by c(G) c(H ). Note that the matrix we propose for the coset theory is not of
the form K K1 . This is in fact consistent with known results for K-matrices of coset
conformal field theories, as we will discuss below. This construction does work for the
cosets of the type Xn,k Xn,l /Xn,k+l , where Xn is a simply-laced Lie algebra. Indeed,
using this, we reproduce the results of McCoy for these diagonal cosets, see, for instance,
[6].
The construction above is in fact more generally applicable as we will show in
the next subsection, where we will show a non-trivial example based on the coset of
so(2n)k /so(2n 1)k .
5.2. so(2n)k /so(2n 1)k
Applying the construction above to the coset so(2n)k /so(2n1)k at level k = 1, we find
the K-matrix K = (1), which is obviously the correct result for this c = 12 CFT. Another
coset with c 1 is the case k = 2, which has c = 1. We find the following K-matrix
1 1 12 12
1
..
K=
(5.2)
,
1
1
2Dn
1
2
where only the first particle is physical. As mentioned, this matrix yields the correct central
charge c = 1 by construction. That it indeed describes the correct c = 1 CFT can be seen
508
as follows. Applying the dual composite construction to

1
1
1 2n
2n
K=
,
1
1
1 2n
2n
where both particles are physical, we find
1
12
12
n
D12 K = 12
1 n2 .
2
12
n
2
(5.3)
(5.4)
n
2
Now applying the composite construction to the two pseudoparticles in (5.4) (n 2) times
we find (5.2). On the other hand, the UCPF based on (5.3), summed over m1 + m2
0 mod 2n, equals the c = 1 u(1)-character
1 n(2n1)k 2
q
,
(5.5)
(q)
kZ
by using the Durfee square identity (see, e.g., [1])

2

1
qm
=
.
(q)
(q)m (q)m
(5.6)
m0
So we indeed find that the matrix (5.2) describes a c = 1 conformal field theory, namely,
the free boson compactified on a circle.
In addition to this non-trivial example, also the equivalence used in the theory of G2 holonomynamely, between so(7)1 /G2,1 and the tricritical Ising modelworks, if we
take the G2 (level k = 1) description of Appendix C. We find the K-matrix

1 12
,
K=
(5.7)
12
1
with one physical and one pseudoparticle. This is indeed the K-matrix corresponding to
the minimal model with c = 7/10.
5.3. Parafermions
Generalized parafermionic conformal field theories were defined by Gepner [25] as a
generalization of the Zk parafermions of [49]. The generalized parafermion theories can
be viewed as cosets based on general affine Lie algebras (ALAs) and u(1) theories
pf
Xn,k =
Xn,k
,
u(1)n
(5.8)
where n is the rank of the Lie algebra Xn , and k the level. The central charge of the
parafermion CFT (5.8) is given by
cpf = cALA n,
(5.9)
where cALA is the central charge of the corresponding affine Lie algebra theory (see
Eq. (2.13)). The parafermion cosets (5.8) are somewhat different in comparison to the
509
diagonal cosets of Section 5.1, and need to be treated differently. Before we come to the
discussion of the K-matrices, we first fix some notations concerning the parafermion fields,
following [25].
The primary fields of the theory are labeled by a (highest) weight and a charge
, which is also an element of the weight lattice, and is defined modulo kML , i.e., k
times the long root lattice. To obtain a complete, independent set of parafermion fields,
one has to impose the following restrictions. The charge must be accessible from by
subtracting roots (including 0 ) from . Furthermore, the (proper) external automorphisms
(see [23]) of the affine Lie algebra give rise to field identifications
()
+
(0) ,
(5.10)
where (0) denotes the image of the affine weight k0 under .

An important check on the K-matrices for the parafermionic CFTs is based on the
relation between the parafermionic partition functions and the string functions c of the
corresponding affine Lie algebras [25]
,
Zpf
(5.11)
= ()n c ,

k
where = q 1/24
k=1 (1 q ) is the Dedekind function. As an example, we will express
,
with = (0, . . . , 0) 1 in terms of UCPFs based on the Kthe partition function Zpf
matrices for the parafermion CFTs. Using Eq. (5.11), we can check our results against the
known (tabulated) string functions.
We will use the matrices Ke of the corresponding affine Lie algebras as a starting
point for obtaining the parafermionic matrices Kpf . The matrices Ke correspond to the
(elementary) electron-like particles and composites (up to order k) of these elementary
particles. The operators corresponding to these (elementary) particles have the form
1 :ei : ,
(5.12)
where = (1 , . . . , n ) is a set of bosonic fields, which correspond to the u(1) degrees of

freedom and determine the quantum numbers of the particles via the constants i . For the
1 = 1, for M ( a long
order k composites, the parafermion fields are trivial, i.e., k
L
root), in which case only the vertex operator part remains.
In this section, we are interested in the K-matrices for the parafermionic CFTs. These
can be obtained from the matrices Ke of the corresponding affine Lie algebra theories
by subtracting from the particles which have a non-trivial parafermion field 1 the
part of the exclusion statistics which corresponds to the vertex operator :ei :. This
can be done by hand by calculating the exclusion statistics of the vertex operators.
Actually, because there are always particles which do not have a parafermion field (or
equivalently, a trivial parafermion field), this can be done by applying what we will call
an X-transformation. Such a transformation is like a W-transformation. However, the
matrices associated to an X-transformation are not SL(r, Z) matrices, but rather SL(r, Q).
This is because the quantum numbers of the largest composites (which are the particles
with trivial parafermion fields) are k times the quantum numbers of the particles in the
k = 1 formulation. In general, the non-zero non-diagonal entries take the form l/k, with
l = 1, . . . , k 1. Explicitly, in the case of the Zk = sl(2)k /u(1) parafermions we find the
510
following
X=
1k
1
1
..
2k
.. .
.
k1
1 k
1
(5.13)
For more general parafermions, the matrices are (a little) more complicated. In fact, each
entry of the matrix (5.13) becomes an n n matrix. Although fractions appear in X,
the quantum numbers of the particles after the transformation are still integers, because
the largest composite is of order k. More precisely, the X-transformation is such that
all the quantum numbers of the transformed particles are in fact zero; in a sense all the
vertex operators containing the chiral boson fields are stripped of from the parafermionic
fields. The transformed matrix Ke splits in two pieces, namely, a part containing the order
k composites and the part corresponding to the parafermions 1 , which is the matrix
we are looking for. We will denote this matrix by Kpf . In the quasiparticle sector, the
pseudoparticles will completely decouple from the physical quasiparticles and hence the
, where K
is a deformed quasiparticle
transformed matrix is of the form K K
matrix. So we conjecture that the K-matrices for parafermionic CFTs are given by the
inverse of the pseudoparticle matrix K , of the corresponding affine Lie algebra CFT
Kpf = K1
.
(5.14)
A first check on the proposed matrices is the corresponding central charge. The central
charge corresponding to the matrices K is given by
c = (n + p)k cALA ,
(5.15)
where p is the difference in rank between the affine algebra under consideration and the
one used to build the K-matrices (thus for simply-laced algebras, p = 0). The rank of the
matrix K is (k 1)(n + p) + e, where e is the number of extra pseudoparticles needed
for the non-simply-laced algebras. Thus we have the following result for the central charge
of matrices Kpf
cpf = cALA n (p e).
(5.16)
For all the affine algebras, except G2,k , the K-matrices of Section 4.3 have p = e, so we
obtain the correct result of Eq. (5.9). However, we also find that the construction above
does not work for the description of G2,k as given in Section 4.3.7, because there the
number of physical quasiparticles is 1 instead of 2, which is the rank of G2 . Luckily, there
exists another way to represent the G2,k affine Lie algebra, which does have two physical
quasiparticles. The inverse of the pseudoparticle matrix therefore has the correct central
charge. The corresponding K-matrices can be found in Appendix C. It has been checked
that this G2,k parafermion K-matrix does give rise to the corresponding string functions
(for k = 2, 3).
511
5.3.1. The case so(5)2 as an example

As an example, we will discuss the characters of the parafermionic theory associated to
so(5)2 .
The conjectured pseudoparticle K for so(5)2 is given by Eq. (4.18) with n = 2, k = 2
1 1 1 1
2
2
1 T
1
1 2
1
1
D3 1
D3 11
K =
(5.17)
=

1 .
3
1
1
D1
2D1
2
2
2
3 1
3
3
1
12
1
2
2
The K-matrix which is supposed to describe the so(5)2 parafermions is simply the inverse
of the pseudoparticle matrix, where it is assumed that all particles are physical
2 1
0
0
3
1
12 12
2
Kpf =
(5.18)
.
0 12
1
0
0 12
The UCPF based on this K-matrix, namely,

=1
Zpf
q 12 mKpf m

=
,
i (q)mi
(5.19)
with m a 4-dimensional vector, is the sum over string-functions

=1
Zpf
=
()l c .
(5.20)
(0,0)
The sum over runs over the independent parafermion fields (
(where we assume
1 ,2 )
(0,0)
that the first root is the short root). The various string-functions c(1 ,2 ) are obtained by
restricting the sum in Eq. (5.19). Explicitly, we have
c(0,0) =
1
pf
q 1/12 q 2 mK m

,
(q)2
i (q)mi
(5.21)
res()
where
res() =
2m1 + m2 + 2m3 = 0 mod 4,
m3 + m4 = 0 mod 2, for = (0, 0);
2m + m + 2m = 0 mod 4,
1
2
3
m + m = 1 mod 2, for = (2, 0);

3
2m1 + m2 + 2m3 = 2 mod 4,
m
3 + m4 = 0 mod 2, for = (0, 2);
2m1 + m2 + 2m3 = 1 mod 4,

m3 + m4 = 0 mod 2, for = (0, 1).
(2,0)
(5.22)
The string functions c(1 ,2 ) can be obtained by using a shift vector; more explicitly, by
changing the power of q in Eq. (5.19) to 12 m Kpf m
, where m
= (m1 1, m2 , m3 , m4 ).
512
We have not yet found similar expressions for the other (independent) string functions,
(0,1)
(1,0)
such as c(
and c(
.
1 ,2 )
1 ,2 )
5.3.2. Cases checked
The cases for which we checked that the conjectured matrices do give the string
functions c1 include all the affine Lie algebras up to rank n = 3 and level k = 2. In addition,
we also checked so(5)3 , so(8)2 , E6,2 , E7,2 , E8,2 and F4,2 . The checks were performed by
numerically calculating the partition functions up to a certain order in q, depending on the
dimension of the K-matrix. These results were compared to the weight-multiplicity tables
of Kass et al. [33]. Note that despite the fact that for the higher rank algebras the checks
were performed to rather low order in q, we believe that the formulas hold to all orders
in q.
As an example, we give the K-matrix associated to the F4 parafermions at level k = 2:
3 1
0
0 1
0
1 1
2
1
2
pf
K (F4,2 ) = 1
4
1
12
12
12
12
12
12
12
12
3
2
12
12
12
.
1
(5.23)
Explicitly, the relation between the parafermionic character based on the matrix in
Eq. (5.23), namely,

Z1
pf
2 mK m
i (q)mi
{mi }
1
(5.24)
and the string functions is as follows. Upon splitting the character in pieces containing
powers of q which differ by integers, one finds

q 12 mKm
1
n

1
1

q ; nN ,
= q 6 (q)4 c(0,0,0,0)
+ 3c(0,0,0,2)
i (q)mi
{mi }

q 12 mKm
1
1

= 12q 6 (q)4 c(1,0,0,0)
i (q)mi
{mi }

q 12 mKm
1
1

= 24q 6 (q)4 c(0,0,1,0)
(q)
m
i
i
{mi }

q 12 mKm
1
1

= 24q 6 (q)4 c(0,0,0,1)
(q)
m
i
i
{mi }
(5.25)

1
q n+ 2 ; n N ,
(5.26)

1
q n+ 4 ; n N ,
(5.27)

3
q n+ 4 ; n N .
(5.28)
513
The primes on the sums denote the restriction to the powers of q as indicated. The various
numerical constants for the string functions c1 are the number of independent fields of the
form 1
which have the same conformal dimension as the field 1 .
6. Application to level restricted Kostka polynomials
In Section 2.2 we have argued that there exists an intimate relation between the fusion
rules of a CFT and the pseudoparticle K-matrix as both count paths on the fusion diagram.
In fact, there exists a natural q-deformation of the number of fusion paths giving rise to
the so-called level truncated Kostka polynomial. This deformation shows up as part of the
UCPF expression for the characters of WZW models, as conjectured in Section 2.1.2. One
would thus expect that the level truncated Kostka polynomials can be expressed as UCPFs
with the K-matrices found in this paper.
Concretely, if i = i , i = 1, . . . , r, denotes the field corresponding to the ith
fundamental weight of g, the multiplicity of the field in the fusion rule
n
1 1 rnr
(6.1)
n1
nr
is given by N1 . . . Nr 0 . By associating a power of q to each path, determined through
the crystal graph of g, we obtain a q-deformation of this number. This is referred to as the
(dual) level-k truncated Kostka polynomial (or truncated q ClebschGordan coefficient) of

(k)
(k)
g and we will denote it by M (q) where = i ni i . An explicit expression of M (q)
for k is known (see, e.g., [12] and references therein) and originates in Bethe-ansatz
techniques [36]. Explicit UCPF type expressions for finite k are known for g = sl(n) (see
[42] for the most general result and also [7,15,28,34]) and so(5)1 [12]. In [29], UCPF
type expressions for Kostka polynomials for general (non-twisted) affine Lie algebras were
conjectured. Proofs for some of these conjectures and expressions for some twisted cases
can be found in, for instance, [42] and [40]. The relation between the K-matrices used in
these expressions and the ones brought forward in this paper is not clear at the moment.
We are gratefull to Ole Warnaar for bringing these references to our attention.
(k)
(q) should be closely related to
According to the UCPF conjecture, M

1 mK m+nK m
1
1
q 2 nK n 2 n K n
q2
m

((I K
i

(K n)i + ui
,
mi
) m)i
(6.2)

where = n
i i and = i ni i . (We have set Q = 0 as we are only discussing paths
starting at the identity representation.)
In the simply-laced case, it has been conjectured before (see, e.g., [10] and references
(k)
(q) can indeed be written in terms of the UCPF based on Kqp =
therein) that M
1
X1
n Mk . Here we will focus on a specific non-simply-laced example, namely so(5)
at levels k = 1, 2. We defer a general investigation to future work. An explicit recipe for
(k)
(q) for g = so(5), at level 1, was given in [48]. Explicit formulae for the
computing M
514
level k = 1 case were given in [12]. Concretely,

3 2 1 2
1
1 2
n
(1)
M(0,0),(n1,n2 ) (q) = q 2 n1 +n1 n2 + 8 n2
q 2 m1 m1 n2 2 2 ,
m1
m1
1
n1 + n2 + m1 even, n2 even,
2

1 2
3 2 1
(1)
M(1,0),(n1,n2 ) (q) = q 2 n1 +n1 n2 + 8 n2 2

1 2
2 m1 m1 n2

1 2
2 m1 m1 n2
m1
1
n1 + n2 + m1 odd, n2 even,
2

1 2
3 2 3
(1)
M(0,1),(n
(q) = q 2 n1 +n1 n2 + 8 n2 8
1 ,n2 )

m1
1
(n2 + 1) + m1 even, n2 odd.
2
The above formulae are of the UCPF form with
0 12
1
1
K= 0
,
1
12
2
3
4
1
2
2 n2
m1

+ 1)
,
m1
2 (n2
(6.3)
(6.4)
which is to be compared to the B2,1 quasiparticle K-matrix of Section 4.3.2, given by
1
1
1
2
2
1
3
K = 1
(6.5)
.
2
1
2
4
1
4
4
3
4
While the pseudoparticle part of Eqs. (6.4) and (6.5) agree, the K-matrices obviously differ
in the physical particle part. Both K-matrices are reminiscent of so(6), but while (6.5) has
physical particles inherited from the 4 and 4 of so(6), Eq. (6.4) contains physical particles
inherited from the 4 and the 6 of so(6). Since 6 = 5 1 under so(5), the matrix (6.4)
does indeed seem to be better suited to describe general (truncated) Kostka polynomials
for so(5), although we expect that the so(5) Kostka polynomials can also be expressed in
terms of a UCPF based on (6.5). Unfortunately, it seems that Eq. (6.4) does not have a
straightforward higher level generalization.
Therefore, motivated by the decomposition of finite-dimensional irreducible representations W(n1 ,n2 ,n3 ) of so(6) into those of so(5) under the regular embedding so(5) so(6),
i.e.,
W(n1 ,n2 ,0)
=
n1

W(n1 l,n2 ) ,
(6.6)
l=0
we introduce
(1)
M
(0,0),(n1 ,n2 ) (q) =

n1

n
1
k=0
(1)
M(0,0),(n1k,n2 ) (q).
(6.7)
515
(1)
Inserting the expression for M(0,0),(n1k,n2 ) (q), and changing k n1 k in the

summation, we find
1

3 2 1 2 1

1 2
n1
2 n2
(1)
2 k +kn2 + 8 n2 + 2 l 2 ln2
(q)
=
q
.
M
(6.8)
(0,0),(n1 ,n2 )
l
k
k,l;k+l+n2 /2 even
Now, let p = k l, then

(1)
M
(0,0),(n1 ,n2 ) (q) =
1 2 1
3 2
2 p + 2 pn2 + 8 n2
p k,l;kl=p
q 2p
2 + 1 pn + 3 n2
2 8 2
2
kl
2 n2
n1
k

n1 + 12 n2
,
n1 p
(6.9)
where, in the last step, we have used a finite version of the Durfee square formula (see [8]).
Finally, letting p n1 p, we find

1
3 2 1 2
1 2
1
+
n
n
(1)
1
2
n
+n
n
+
n
p
pn
pn
1
2
1
2
2

8 2
2
M (0,0),(n1,n2 ) (q) = q 2 1
(6.10)
q2
.
p
p
(1)
A similar computation can be given for the other sectors M(n

,n
),(n ,n ) (q) of (6.3). Now,
1 2
1 2
Eq. (6.10) is of the UCPF form with
1 12
1
1
K = 1
(6.11)
,
1
2
12
1
2
3
4
which has the same K and K parts as (6.4), but differs in the coupling K .
Now consider the so(5), level k = 2 case. As an ansatz we take the pseudoparticle
matrix of Section 4.3.2 (see also Eq. (5.17)), and the physical particles of Eq. (6.11), and
adjust the coupling between them. Specifically, let
1
1 12 12
0
0
1
2
1
1
1 12
1
1
3
1
3
2
1
2
2
2
4
K=
(6.12)
3
1
1
1 .
1
1
2 4
2
2
2
0
1
1 12 12
1
2
3
1
0
12 34 14
2
4
Note that this matrix is not invertible, as is the case for the matrix in Eq. (6.11). Thus,
Eq. (6.2) reads explicitly
1
1
2
2
2
(2)

M
(q) = q 2 n1 +n1 n2 + 8 n2 2 n1 +n1 n2 8 n2
(n1 ,n2 ),(n1 ,n2 )

1 2
1
2
2
2 1
q 4 2m1 +4m2 +3m3 +3m4 2 m1 (2m2 +m3 +m4 )+m2 (m3 +m4 )+ 2 m3 m4
516

1
q 2 n1 (2m2 +m3 +m4 ) 4 n2 (2m2 +3m3 +m4 )
1

(2m2 + m3 + m4 ) + u1
2
m1

m1 (m2 + m3 + m4 ) + n1 + 12 n2 + u2
m2
1

m1 (2m2 + m3 + m4 ) + 12 n1 + 34 n2 + u3
2
m3
1

m (2m2 + m3 + m4 ) + 12 n1 + 14 n2 + u4
2 1
,
m4
(6.13)
with some appropriate restriction on the summation over (m1 , . . . , m4 ).

Numerical evidence suggests the following conjecture (cf. (6.7))
(2)

M
(q) =
(n1 ,n2 ),(n1 ,n2 )
n
1
n1

n1
k
k=0
(2)
M(n
l,n
),(n
1
l=0
1 k,n2 )
(q),
(6.14)
or equivalently,
(2)
M(n
,n
),(n ,n ) (q) =
1 2
1 2
n1

(1) q
1
2 k(k1)
k=0
n1
k
n1

(2)
(1)l M
(n l,n
),(n
l=0
1 k,n2 )
(q),
(6.15)
where the vectors u in (6.13), for given (n

1 , n
2 ), are given in Table 1.4 The summation
restrictions are such that 2m2 + m3 + 3m4 2((n1 n
1 ) + (n2 n
2 )) mod 4, and

n

n1 + n22 + m1 n
1 + 22 mod 2.
Again, the conjectured formula (6.14) is strongly reminiscent of the decomposition of
finite-dimensional irreducible representations (6.6). This suggests that while the procedure
of Section 4.3.2 does produce a pseudoparticle K-matrix leading to the correct central
charge, it still overcounts the number of fusion paths. This overcounting can also be seen
by applying the analysis of Section 2.2, as the pseudoparticle K-matrix does not give rise to
Table 1
The vectors n
and u for the so(5)2 Kostka polynomials
(n
1 , n
2 )
(u1 ; u2 , u3 , u4 )
(0, 0)
(1, 0)
(0; 0, 0, 0)

0; 1, 12 , 12
1 1 3
0; 2 , 4 , 4
(0, 1)
4 We have not been able to find the u-vectors corresponding to the remaining integrable highest weight
modules at level 2, i.e., (n
1 , n
2 ) = (2, 0), (1, 1) and (0, 2).
517
the same recursion relations as the so(5)2 fusion rules. For this reason we also expect that
the so(5)2 characters, when written in UCPF form using the K-matrices of Section 4.3.2,
will need alternating sign corrections.
7. Discussion
In this paper, we proposed a scheme to obtain the K-matrices for the CFTs with affine
Lie algebra symmetry. This construction was based on character identities, which were
applied to certain Abelian covering states. After projecting out some degrees of freedom,
the K-matrices were obtained. Subsequently, these K-matrices were used to obtain the Kmatrices of coset CFTs. Also, they appeared in some expressions for the level-k restricted
Kostka polynomials.
It would be interesting to investigate if the K-matrices obtained here indeed are the
central objects in the Kostka polynomials related to a general affine Lie algebra. An
interesting open question is whether similar K-matrices can be used for more general CFTs,
such as the twisted affine Lie algebras (and their parafermions), which were studied in
[16] and [17]. Another interesting class of theories which might be addressed in a similar
fashion are the affine Lie superalgebras and the related parafermions (see, for instance,
[14] and [32] for the case osp(1|2)).
Most of our consistency checks on whether we obtained the correct K-matrices were
based on the fact that the central charge worked out correctly. Even though this proved
to be an extremely restrictive guide, the ultimate verification of course relies on the
construction of the CFT characters in the UCPF form using these K-matrices. While
we have proved this in special cases, and did numerical checks in others, a complete
verification requires tools beyond the scope of this paper, and will require proving a host of
new q-identities. A systematic approach towards a full proof will undoubtedly benefit from
a better algebra-geometric understanding of the role of K-matrices (see, e.g., [10,1921]
for some initial studies).
Note added
In an earlier version of this paper we referred to the Kostka polynomials of Section 6
as generalized Kostka polynomials to indicate the generalization of the standard An
Kostka polynomials to general simple Lie algebras. In order to avoid confusion with the
generalized Kostka polynomials, introduced independently by Schilling and Warnaar
[43] and by Kirillov and Shimozono [37] (cf. [40] for a discussion), which are more general
than the Kostka polynomials which are the subject of this paper, we will simply refer to the
polynomials in this paper as (level restricted) Kostka polynomials. We thank Ole Warnaar
for communication on these points.
518
Acknowledgements
E.A. would like to thank the Department of Physics and Mathematical Physics at the
University of Adelaide, where most of this work was carried out, for hospitality. P.B.
acknowledges financial support from the Australian Research Council. The research of
E.A. was supported in part by the foundation FOM of the Netherlands and by the National
Science Foundation through the grant DMR-01-32990.
Appendix A. Cartan matrices and their inverses

In this appendix, we will list of the Cartan matrices of the simple Lie algebras, to clarify
the conventions used in this paper. In addition, we will give some other properties, namely
the dimension and the dual Coxeter number. Other properties can be found, for instance,
in [23].
In the Cartan matrices, the empty entries correspond to zeros, unless otherwise implied
by the dots. Even though we only use matrices corresponding to simply laced Lie algebras,
we will give the Cartan matrices of all the simple Lie algebras, for completeness. We will
denote the Cartan matrix corresponding to the Lie algebra Xn by Xn .
An : The Cartan matrix for An is given by
2 1
1 2 1
1 2
An =
..
..
..
1
2
1 2
n
n1
n2
n 1 2(n 1) 2(n 2)
n 2 2(n 2) 3(n 3)
1
.
=
A1
..
..
n
n+1
..
.
.
2
4
6
1
2
3
Bn :
2 1
1
Bn =
2
1
.
1
1
2
..
.
..
..
.
1
2 2
1 2
(A.1)
...
...
...
..
.
...
2
1
4
2
6
3
..
..
.
.
.
2(n 1) n 1
n1
n
(A.2)
1
1
1
Bn = .
..
1
1
2
1
2
2
..
.
1
2
3
..
.
..
.
2
1
3
2
1
2
3
..
.
1
2
3
..
.
n1
2
n
2
519
n 1 n 1
(A.3)
Cn :
2 1
1 2 1
1 2
Cn =
..
1
1
1
Cn =
..
.
1
1
1
2
2
..
.
2
2
1
2
3
..
.
3
3
..
.
..
..
.
1
1
2
2
1
2
3
..
.
n1
n1
1
1
2
3
2
.. .
.
n1
(A.4)
2
n
2
Dn :
2 1
1 2 1
1 2
..
Dn =
.
1
1
..
D1
n = .
1
2
1
2
2
..
.
2
1
2
3
..
.
3
1
2
3
2
3
2
..
.
..
..
.
1
,
1
2 1 1
1 2
0
1 0
2
1
1
1
2
3
..
.
n2
n2
2
n2
2
1
3
2
..
.
n2
2
n
4
n2
4
3
2
..
.
.
n2
2
n2
4
n
4
(A.5)
520
E6 :
2 1 0
0
1 2 1 0
0 1 2 1
E6 =
0
0 1 2
0
0
0 1
0
0 1 0
4 5 6 4
5 10 12 8
6 12 18 12
1
E1
=
6
3 4 8 12 10
2 4 6 5
3 6 9 6
0
0
0
0
0 1
,
1 0
2
0
0
2
2 3
4 6
6 9
.
5 6
4 3
3
(A.6)
E7 :
2 1 0
0
0
0
0
1 2 1 0
0
0
0
0 1 2 1 0
0
0
0 1 2 1 0 1 ,
E7 = 0
0
0
0 1 2 1 0
0
0
0
0 1 2
0
0
0
0 1 0
0
2
3 4 5 6 4 2 3
4 8 10 12 8 4 6
5 10 15 18 12 6 9
E1
6
12
18
24
16
8
12
=
.
2
4 8 12 16 12 6 8
2 4 6 8 6 4 4
3 6 9 12 8 4 7
E8 :
2
1
E8 =
0
0
0
1 0
0
2 1 0
1 2 1
0 1 2
0
0 1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1 0
0
0
,
2 1 0 1
1 2 1 0
0 1 2
0
1 0
0
2
(A.7)
2
3
1
E8 =
6
2
3
3
6
8
10
12
8
4
6
4
8
12
15
18
12
6
9
5
10
15
20
24
16
8
12
6
12
18
24
30
20
10
15
521
4 2 3
8 4 6
12 6 9
16 8 12
.
20 10 15
14 7 10
7 4 5
10 5 8
(A.8)
F4 :
2 1 0
1 2 2
F4 =
0 1 2
0
0 1
0
0
,
1
2
3
F1
4 = 2
1
3
6
4
2
4
8
6
3
2
4
.
3
2
(A.9)
G2 :

G2 =
2
1

3
,
2
G1
2 =
2
1

3
.
2
(A.10)
In Table 2 we list some of the properties of the simple Lie algebras. The black nodes in
the Dynkin diagrams correspond to the short roots.
In addition to the Cartan matrices given above, we will frequently use the symmetrized
Cartan matrix of Bk , which we denote by M1
k . Explicitly, we have
1
1
Mk = 1
.
..
1
1
2
2
..
.
1
2
3
..
.
..
.
1
2
3
,
..
.
2
1
=
M1
k
1
2 1
1
2
..
.
..
..
.
1
2 1
1 1
(A.11)
The simple Lie algebras are labeled by Xn , where n is the rank, and X can be
A, B, . . . , G. As we will only be dealing with the untwisted affine Lie algebras, we will use
the notation Xn,k , rather than (Xn(1) )k , which is more common in the literature. Sometimes,
we will use the notation sl(n)k , so(2n 1)k , sp(2n)k and so(2n)k for the infinite series of
untwisted affine Lie algebras. Here, and in the rest of the paper, the level is denoted by k.
Blackboard bold, such as A is used for matrices, while vectors are in boldface, such as
Q. If we want to specify a column of a matrix, say A, we use the notation (A)c , where the
integer c denotes the column we want to specify. In bilinear forms such as mT K m, we
will frequently omit the transposition symbol T .
522
Table 2
Some properties of the finite-dimensional simple Lie algebras
dim Xn
An
n(n + 2)
n+1
Bn
n(2n + 1)
2n 1
Cn
n(2n + 1)
n+1
Dn
2n(n 1)
2n 2
E6
78
12
E7
133
18
E8
248
30
F4
52
G2
14
Xn
Dynkin diagram
Appendix B. Obtaining the so(5)1 matrices

The electron matrix for so(5)1 can be obtained by using knowledge about the root
diagram and the associated parafermions (see [25] for general parafermion theories). We
will anticipate that it is in fact possible to use a quantum Hall type of basis for this theory.
So we define a set of electron operators, where the vertex operator part is chosen in such
a way that the spin and charge are such that we actually have electron-like operators. The
matrix Ke is obtained via the connection with the exclusion statistics, i.e., we calculate the
associated exclusion statistics parameters of these electron operators. From [25] we obtain
that at level k = 1, the short roots of so(5) come with a parafermion operator, which is
in fact the Majorana fermion , which has the same exclusion statistics parameter as a
fermion, namely, 1. The root diagram of so(5) is given in Fig. 1. The electron operators we
,
take to be part of the quantum Hall basis correspond to
s and c . These operators
take the form (at level k = 1)
i
= :e 2 (c +s ) : ,
s = :e
2s
:,
c = :ei
2c
:,
(B.1)
where s and c are spin and charge bosons, respectively, chosen according to the spin and
charge direction indicated in Fig. 1. From these operators, we infer the following exclusion
523
Fig. 1. The roots () and weights () of so(5).
statistics matrix
2 1
Ke = 1 2
1 0
1
0 ,
2
1
te = 0 ,
2
1
se = 2 .
0
(B.2)
We should comment on a few things here. First of all, the matrix we found is equal to the
Cartan matrix of so(6), which relates to the so-called covering state of the state related
to so(5). This is analogous to the situation of the MooreRead state, which is related to
a two-layer state. So we could have started from this K-matrix, and performed a similar
construction as was done in Section 3.4.1 to find the K-matrices for the MooreRead state.
This would lead to the same matrix (B.2). In addition, in the quasiparticle sector, there is a
pseudoparticle, just as in the MooreRead case. The matrix for the quasiparticle sector can
simply be obtained by inverting the matrix (B.2). As said, it is important to notice that the
particle in the quasiparticle sector which has trivial quantum numbers, is to be considered
as a pseudoparticle. Otherwise, we would not obtain the correct central charge, and hence,
not the correct description. We find
1 12 12
0
0
tqp = 0 ,
sqp = 1 .
Kqp = 12 43 14 ,
(B.3)
1
3
1
1
0
2
To obtain the K-matrices for so(5) at general level, we take k copies of the level-1
formulation, and do a similar construction as described in Section 3.2. This gives the result
of Section 4.3.
Appendix C. The case G2,k

In Section 4.3 we found that the K-matrices for the affine Lie algebra G2,k are special
in the sense that the number of physical quasiparticles is not equal to the rank of this
algebra (which is 2), if we use the standard construction of Section 4.1. Here, we will
find another way of describing this theory, which does have two physical quasiparticles.
524
Fig. 2. The roots of G2 .
We will start by deriving the K-matrices for level k = 1, in a similar way as we did for
so(5)1 in Appendix B. We continue by explaining how to obtain the K-matrices for general
level k. This is a little different from Section 4.1, as the P-transformation which is needed
is different.
The root lattice for the Lie algebra G2 is given in Fig. 2. In fact, it is not possible to pick
four electron-like operators, such that the K-matrix is the Cartan matrix of the enveloping
algebra so(8), but we will stay as close as possible.
The short roots come with two types of parafermions, 1 and 2 , which belong to the
Z3 parafermion theory. The operators needed to form the quantum Hall basis are

i
i
= 1 :exp c + s : ,
6
2

3
i
c = :exp i c + s : ,
6
2

i
i
= 2 :exp c s : ,
6
2

3
i
c = :exp i c s : ,
6
2
(C.1)
(C.2)
where c,s are the charge and spin boson. As the K-matrix for the Z3 parafermions is given
by
pf
4
KZ3 =
3
2
3
2
3
4
3

,
(C.3)
and the statistics parameters due to the vertex operators of the spin and charge bosons are
easily calculated, we find the following data for the electron sector of the G2,k=1 theory
2
0
Ke =
1
0
0
1 0
2 1 0
,
1 2 1
0
1 2
1
1
te =
3 ,
3
1
1
se =
1 .
1
(C.4)
By the duality construction, we find the dual data

0
1 12
1 12
1
1
1
1 2
0
tqp =
Kqp = 2
,
1,
1 1
2 1
1
12
1 1
1
2
525
0
0
sqp =
1 ,
1
(C.5)
where the first two particles are pseudoparticles, which reduce the central charge, and take
care of the non-Abelian statistics. Note that we do not use the usual ordering of the Cartan
matrix (compare Appendix A), because in the quasiparticle sector, we want the first to
particles to be the pseudoparticles.
Picking the operators associated to the right roots is crucial in finding a basis for the G2
affine Lie algebra. The way we have chosen them here gives a description which does give
the right central charge, and has two physical quasiparticles.
We would like to comment on the difference between the pseudoparticle matrices
for the two descriptions of G2,1 . If we apply the composite construction on the 2 2
pseudoparticle matrix of this appendix, we indeed find the pseudoparticle matrix (at level 1)
of Section 4.3. This matrix also appeared in Section 2.2, Eq. (2.59). So the pseudoparticles
are equivalent in both cases.
We now proceed by constructing the matrices for level k. As usual, the covering is of
the form Ke Ik . The required P-transformation turns out to be of the form (compare with
Appendix D)
I4 Ju4 Ju4
..
..
Jl I
.
.
4
P
= 4
(C.6)
,
.. . . . .
.
. Ju4
.
Jl4
where
Ju4
and
Ju4 =
Jl4
1
Jl4
I4
are given by
Jl4 =
0
0
1
1
1
(C.7)
Because Ju4 + Jl4 = I4 , all composites up to order k are formed. To display the resulting
matrix, it is most convenient to reorder the particles in the order of increasing quantum
numbers (this is not done automatically, because of the form of the P-transformation). To
conveniently display the permuted K-matrix for the electron sector, we define a modified
Cartan matrix of D4
a 0 b 0
0 a c 0
M(a, b, c) =
(C.8)
b c a b .
0 0 b a
526
Then, the electron K-matrix for G2,k can be described by

M(2, 0, 1) M(2, 0, 1)
G
Ke 2,k
M(2, 1, 1)
M(4, 2, 2)
M(2, 0, 1) M(4, 0, 2)
..
..
.
..
.
.
.

=
M 2 min(i, j ), max(i + j k, 0), min(i, j )
.
.
.
..
..
..
M(2, 1, 1) M(4, 2, 2)
M(2k, k, k)
(C.9)
To make this a little more clear, we give the result for k = 2 explicitly
2 0
0 0
2 0
1 0
0 2 1 0
0 2 1 0
0 1 2 0
1 1 2 1
0 0
0 2
0 0
1 2
.
Ke =
2 0
1 0
4 0
2 0
0 2 1 0
0 4 2 0
1 1 2 1
2 2 4 2
0
(C.10)
The quasiparticle sector for k > 2 is characterized by the following matrices (compare with
Section 4)

2D1
D1
D1
4
4
4 1
..
D1
1
.
2D
4
4
..
..
1
.
.
D
4
K =
, (C.11)
1
1
1
D4
2D4
D4 3
1 T
1
0
D4 1
1 T
D4 3
0
1

1 T
D4 2 0 1
K =
(C.12)
,
1 T
1
D4 4
2 0

2 0
,
K =
(C.13)
0 1
tqp = (0, . . . , 0; 1, 1),
(C.14)
sqp = (0, . . . , 0; 1, 1).
(C.15)
So, although the form of the K-matrix differs from the general description, we still find
that all the elements are related to the (inverse) Cartan matrix of the Lie algebra D4 .
Now that we have a description of G2 which does have two quasiparticles (for every k),
we can use the same conjecture (5.14) to find the K-matrices for the parafermions, namely,
the parafermion theory G2,k /[u(1)]2 . So, without giving the explicit form, it is found that
527
the parafermion K-matrix KG2 = K1

does have the right properties. It gives the correct
central charge, and reproduces the string functions as described in Section 5.3.
For the case k = 1, we indeed find that the parafermions associated to G2 are the Z3
parafermions. At level k 2 we find the K-matrices of the G2 parafermions, which for
k = 2 is given by
5
1
3
4
1
0
pf
1
3
1
2
pf
KG2 ,k=2 =
0
3
2
3
3
5
3
12
3
2
3
12
12
12
2
3
4
3
8
3
4
3
2
4
2
0
.
0
(C.16)
3
8
3
Note the asymmetry between the parafermions 3 and 4.
Appendix D. Relating different bases

In Section 4.1 we pointed out that the K-matrices for sl(3)k found in [3] differ from the
ones we presented here. The reason for this was also given. In [3], all the particles in the
electron sector were chosen such that their charge all had the same sign. Consequently, the
K-matrix for level-1 was based on the roots 1 and 2 . This resulted in the following
K-matrix and quantum number vectors

2 1
1
1
(k=1)
=
,
te =
,
se =
.
Ke
(D.1)
1 2
1
1
In this appendix, we will explain in detail the relation between this approach and the
one followed in this paper. The matrix (D.1) can also be used to obtain K-matrices for
sl(3)k . This formulation is different, but can be related to the one obtained in Section 4.1.
We will first show that we can construct the sl(3)k K-matrices found in [3] using the Ptransformations. We then explicitly relate the two constructions.
So, let us begin with the covering matrix based on Eq. (D.1), which is constructed
(k=1)
in the usual way, by taking a direct sum of k copies: K
cover
= Ke
Ik . Now the
e
P-transformation is different than the one used in Section 4.1. It will be such that all
composites up to order k are formed (for both spin up and spin down particles). However,
P is not lower triangular, but instead we have
I2 Ju2 Ju2
.
..
Jl I
. ..
2
P =
(D.2)
.
.. . . . .
u
.
.
.
J2
Jl2 Jl2 I2

Here, Ju2 = 10 00 and Jl2 = 00 01 . The transformed K-matrix P
K
cover
P
T is most easily
e
described after a suitable permutation of the particles, which orders the particles according
528
to their quantum numbers; as indicated before, all composites (up to order k) are formed,
because Ju2 + Jl2 = I2 . The quantum numbers after applying the P-transformation to the
covering and the permutation to order them, are given by t
e = (1, 1, 2, 2, . . ., k, k) and
s
e = (1, 1, 2, 2, . . ., k, k). The K-matrix becomes
2
0
.
Ke =
.
2
1
0
2
0
2
..
.
2
0
4
0
..
.
0
2
0
4
..
.
..
.
0
2
1
2
4
1
4
2
1
4
2
4
2
0
4
1
..
.
0
2
1
4
..
.
2
1
4
2
..
.
1
2
2
4
..
.
. (D.3)
2(k 1)
k2
2(k 1)
k1
k2
2(k 1)
k1
2(k 1)
2(k 1)
k1
2k
k
k1
2(k 1)
k
2k
This matrix is to be compared with Ke of Eq. (4.9). The diagonal part of the 2 2 blocks is
the same, namely, 2 min(i, j ), where i, j label the blocks. The off-diagonal parts are given
by max(k i j, 0). The inverse is found to be (again, after a suitable permutation of the
particles)
Kqp =

T
A1
2 1
0

A1
2 1
0
0
0
A1
2 Ak1
0
0
0
0
0
T

A1
2 2
2
3
0
1
,
A2 2
(D.4)
2
3
which is to be compared with Kqp of Eq. (4.12). To relate the two descriptions, we make use
of the fact that we know how to relate the matrices for k = 1. The difference is the use of 2
in the description detailed in Section 4.1 and 2 in the description of this appendix and
2 1
(k=1)
.
[3]. Recall that the K-matrix for level k = 1 from Section 4.1 is given by Ke
= 1
2
So we find that we can relate the two K-matrices for level-1 by a W-transformation, which
0
(k=1)
(k=1)
. Because we also know how to
= W Ke
WT , where W = 10 1
is given by Ke
transform the coverings into the corresponding K-matrices for sl(3)k , we can relate the two
descriptions in terms of a W-transformation. Apart from the extra permutations which are
involved, the calculation is straightforward, and we find the relation K
e = We Ke WTe ,
with (dropping the subscript 2)
Jl
Ju
..
.
.
..
...
. .
We =
.
..
..
.
Ju
Jl
Ju
Ju
..
.
Ju
Ju
Ju
Jl
529
(D.5)
. .
where . . . stands for

Jl Ju
Ju Jl
if k is odd and for
Jl
Ju
I
u
l
J
J
if k is even. Note that W1
e = We . For the quasiparticle sector we have a similar relation,
K
qp = Wqp Kqp WTqp . But because we needed the extra permutations, we do not have the
T
relation Wqp = (W1
e ) . This only holds for the case at hand if we undo this permutation.
Instead, we have
Wqp =
I
..
(D.6)
I
Ju Ju Ju Ju Jl
Note that in going from the one formulation to the other, we are only transforming the
physical quasiparticles, the pseudoparticles are not changed. This should be the case, as
the pseudoparticles govern the fusion rules and the central charge.
Let us end this discussion by mentioning that the formulation for sl(3)k of the type of
Eq. (D.3) can be generalized to arbitrary affine Lie algebra CFTs. The relations between the
description in this paper is precisely analogous to the relation for sl(3) as described in this
appendix. The only difference would be in the form of the matrices Ju and Jl . However,
they still would only have non-zero elements on the diagonal, subject to the constraint
Ju + Jl = I.
References
[1] G.E. Andrews, Partitions: Yesterday and Today, New Zealand Mathematical Society, Wellington, 1979.
[2] E. Ardonne, P. Bouwknegt, S. Guruswamy, K. Schoutens, K-matrices for non-Abelian quantum Hall states,
Phys. Rev. B 61 (2000) 1029810302, cond-mat/9908285.
530
[3] E. Ardonne, P. Bouwknegt, K. Schoutens, Non-Abelian quantum Hall statesexclusion statistics, Kmatrices and duality, J. Stat. Phys. 102 (2001) 421469, cond-mat/0004084.
[4] E. Ardonne, F.J.M. van Lankvelt, A.W.W. Ludwig, K. Schoutens, Separation of spin and charge in paired
spin-singlet quantum Hall states, Phys. Rev. B 65 (2002) 041305(R), cond-mat/0102072.
[5] E. Ardonne, K. Schoutens, New class of non-Abelian spin-singlet quantum Hall states, Phys. Rev. Lett. 82
(1999) 50965099, cond-mat/9811352.
[6] A. Berkovich, B. McCoy, The universal chiral partition function for exclusion statistics, in: M.T. Batchelor,
L.T. Wille (Eds.), Statistical Physics on the Eve of the 21st Century, in: Series on Advances in Statistical
Mechanics, Vol. 14, World Scientific, Singapore, 1999, pp. 240256, hep-th/9808013.
[7] A. Berkovich, B. McCoy, A. Schilling, RogersSchurRamanujan type identities for the M(p, p
) minimal
models of conformal field theory, Commun. Math. Phys. 191 (1998) 325395, q-alg/9607020.
[8] P. Bouwknegt, Multipartitions, generalized Durfee squares and affine Lie algebra characters, J. Austral.
Math. Soc. 72 (2002) 395408, math.CO/0002223.
[9] P. Bouwknegt, L.-H. Chim, D. Ridout, Exclusion statistics in conformal field theory and the UCPF for WZW
models, Nucl. Phys. B 572 (2000) 547573, hep-th/9903176.
[10] P. Bouwknegt, N. Halmagyi, q-identities and affinized projective varieties, II. Flag varieties, Commun. Math.
Phys. 210 (2000) 663684, math-ph/9903033.
[11] P. Bouwknegt, K. Schoutens, Non-Abelian electrons: SO(5) superspin regimes for correlated electrons on a
two-leg ladder, Phys. Rev. Lett. 82 (1999) 27572760, cond-mat/9805232.
[12] P. Bouwknegt, K. Schoutens, Exclusion statistics in conformal field theorygeneralized fermions and
spinons for level-1 WZW theories, Nucl. Phys. B 547 (1999) 501537, hep-th/9810113.
[13] A. Cappelli, L. Georgiev, I. Todorov, Parafermion Hall states from coset projections of Abelian conformal
theories, Nucl. Phys. B 599 (2001) 499530, hep-th/0009229.
[14] J.M. Camino, A.V. Ramallo, J.M. Snchez de Santos, Graded parafermions, hep-th/9805160.
[15] S. Dasmahapatra, R. Kedem, T. Klassen, B. McCoy, E. Melzer, Quasi-particles, conformal field theory, and
q-series, Int. J. Mod. Phys. B 7 (1993) 36173648, hep-th/9303013.
(2)
[16] X.-M. Ding, M.D. Gould, Y.-Z. Zhang, Twisted sl(3, C)k current algebra: free field representation and
screening currents, Phys. Lett. B 523 (2001) 367376, hep-th/0109009.
[17] X.-M. Ding, M.D. Gould, Y.-Z. Zhang, Twisted parafermions, Phys. Lett. 530 (2001) 197201, hepth/0110165.
[18] R. van Elburg, K. Schoutens, Quasi-particles in fractional quantum Hall effect edge theories, Phys. Rev.
B 58 (1998) 1570415716, cond-mat/9801272.
[19] B. Feigin, M. Jimbo, R. Kedem, S. Loktev, T. Miwa, Spaces of coinvariants and fusion products, I. From
equivalence theorem to Kostka polynomials, math.QA/0205324;
B. Feigin, M. Jimbo, R. Kedem, S. Loktev, T. Miwa, Spaces of coinvariants and fusion products, II. Affine
sl2 character formulas in terms of Kostka polynomials, math.QA/0208156.
[20] B. Feigin, R. Kedem, S. Loktev, T. Miwa, E. Mukhin, Combinatorics of the
sl2 spaces of coinvariants I,
math-ph/9908003;
B. Feigin, R. Kedem, S. Loktev, T. Miwa, E. Mukhin, Combinatorics of the
sl2 spaces of coinvariants II,
math.QA/0009198;
B. Feigin, R. Kedem, S. Loktev, T. Miwa, E. Mukhin, Combinatorics of the
sl2 spaces of coinvariants III,
math.QA/0012190.
[21] S. Fishel, I. Grojnowksi, C. Teleman, The strong Macdonald conjecture, math.RT/0107072.
[22] E. Fradkin, C. Nayak, K. Schoutens, LandauGinzburg theories for non-Abelian quantum Hall states, Nucl.
Phys. B 546 (1999) 711730, cond-mat/9811005.
[23] P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, New York, 1997.
[24] G. Gasper, M. Rahman, Basic Hypergeometric Series, Cambridge Univ. Press, Cambridge, 1990.
[25] D. Gepner, New conformal field theories associated with Lie algebras and their partition functions, Nucl.
Phys. B 290 (1987) 1024.
[26] S. Guruswamy, K. Schoutens, Non-Abelian exclusion statistics, Nucl. Phys. B 556 (1999) 530544, condmat/9903045.
[27] F.D.M. Haldane, Fractional statistics in arbitrary dimensions: a generalization of the Pauli principle, Phys.
Rev. Lett. 67 (1991) 937940.
531
[28] G. Hatayama, A. Kirillov, A. Kuniba, M. Okado, T. Takagi, Y. Yamada, Character formulae of

sln -modules
and inhomogeneous paths, Nucl. Phys. B 536 (1999) 575616, math.QA/9802085.
[29] G. Hatayama, A. Kuniba, M. Okado, T. Takagi, Y. Yamada, Remarks on fermionic formula, Contemp.
Math. 248 (1999) 243291, math.QA/9812022.
[30] T.-L. Ho, The broken symmetry of two-component = 1/2 quantum Hall states, Phys. Rev. Lett. 75 (1995)
11861189, cond-mat/9503008.
[31] S.B. Isakov, Generalization of statistics for several species of identical particles, Mod. Phys. Lett. B 8 (1994)
319327;
A. Dasnires de Veigy, S. Ouvry, Equation of state of an anyon gas in a strong magnetic field, Phys. Rev.
Lett. 72 (1994) 600603, hep-th/9306039;
Y.-S. Wu, Statistical distribution for generalized ideal gas of fractional-statistics particles, Phys. Rev. Lett. 73
(1994) 922925.
[32] P. Jacob, P. Mathieu, Graded parafermions: standard and quasi-particle bases, Nucl. Phys. B 630 (2002)
433452, hep-th/0201156.
[33] S. Kass, R.V. Moody, J. Patera, R. Slansky, in: Affine Lie Algebras, Weight Multiplicities and Branching
Rules, Vol. 2, University of California Press, Berkeley, 1990.
[34] A.N. Kirillov, Dilogarithm identities, Prog. Theor. Phys. Suppl. 118 (1995) 61142, hep-th/9408113.
[35] A.N. Kirillov, Ubiquity of Kostka polynomials, math.QA/9912094.
[36] A.N. Kirillov, N.Yu. Reshetikhin, The Bethe ansatz and the combinatorics of Young tableaux, J. Sov.
Math. 41 (1988) 925;
A.N. Kirillov, N.Yu. Reshetikhin, Representations of Yangians and multiplicities of occurrence of the
irreducible components of the tensor product of representations of simple Lie algebras, J. Sov. Math. 52
(1990) 31563164.
[37] A.N. Kirillov, M. Shimozono, A generalization of the KostkaFoulkes polynomials, J. Algebraic Combin. 5
(2002) 2769, math.QA/9803062.
[38] G. Moore, N. Read, Nonabelions in the fractional quantum Hall effect, Nucl. Phys. B 360 (1991) 362396;
G. Moore, N. Read, Fractional quantum Hall effect and nonAbelian statistics, Prog. Theor. Phys. Suppl. 107
(1992) 157166, hep-th/9202001.
[39] A. Nakayashiki, Y. Yamada, On spinon character formulas, in: H. Itoyama, et al. (Eds.), Frontiers in
Quantum Field Theories, World Scientific, Singapore, 1996, pp. 367371.
[40] M. Okado, A. Schilling, M. Schimozono, Crystal bases and q-identities, Contemp. Math. 291 (2001) 2953,
math.QA/0104268.
[41] N. Read, E. Rezayi, Beyond paired quantum Hall states: parafermions and incompressible states in the first
excited Landau level, Phys. Rev. B 59 (1999) 80848092, cond-mat/9809384.
[42] A. Schilling, M. Shimozono, Fermionic formulas for level-restricted generalized Kostka polynomials and
coset branching functions, Commun. Math. Phys. 220 (2001) 105164, math.QA/0001114.
[43] A. Schilling, S.O. Warnaar, Inhomogeneous lattice paths, generalized Kostka polynomials and An1
supernomials, Commun. Math. Phys. 202 (1999) 359401, math.QA/9802111.
[44] K. Schoutens, Exclusion statistics in conformal field theory spectra, Phys. Rev. Lett. 79 (1997) 26082611,
cond-mat/9706166.
[45] K. Schoutens, Exclusion statistics for non-Abelian quantum Hall states, Phys. Rev. Lett. 81 (1998) 1929
1932, cond-mat/9803169.
[46] E. Verlinde, Fusion rules and modular transformations in conformal field theory, Nucl. Phys. B 300 (1988)
360376.
[47] X.-G. Wen, Topological orders and edge excitations in fractional quantum Hall states, Adv. Phys. 44 (1995)
405, cond-mat/9506066.
(1)
[48] Y. Yamada, On q-ClebschGordan rules and the spinon character formulas for affine C2 algebra, qalg/9702019.
[49] A.B. Zamolodchikov, V.A. Fateev, Nonlocal (parafermion) currents in two-dimensional conformal quantum
field theory and self-dual critical points in ZN -symmetric statistical systems, Sov. Phys. JETP 62 (1985)
215225.

Characteristic polynomials of complex random

matrix models
G. Akemann, G. Vernizzi
Service de Physique Thorique, CEA/DSM/SPhT Saclay, Unit de recherche associe au CNRS,
F-91191 Gif-sur-Yvette Cdex, France
Received 13 December 2002; received in revised form 11 February 2003; accepted 7 March 2003
Abstract
We calculate the expectation value of an arbitrary product of characteristic polynomials of complex
random matrices and their Hermitian conjugates. Using the technique of orthogonal polynomials in
the complex plane our result can be written in terms of a determinant containing these polynomials
and their kernel. It generalizes the known expression for Hermitian matrices and it also provides
a generalization of the Christoffel formula to the complex plane. The derivation we present holds
for complex matrix models with a general weight function at finite-N, where N is the size of the
matrix. We give some explicit examples at finite-N for specific weight functions. The characteristic
polynomials in the large-N limit at weak and strong non-hermiticity follow easily and they are
universal in the weak limit. We also comment on the issue of the BMN large-N limit.
1. Introduction
The study of the spectral statistical properties of random matrix models naturally
leads to consider characteristic polynomials of random matrices. Because of the wide
applicability of random matrices, the analysis of characteristic polynomials has been
approached from different fields of physics and mathematics, such as quantum chaos [1],
quantum chromodynamics (QCD) [2] or the study of the statistical distribution of the zeros
of the Riemann zeta-function [3] to name a few of them.
The expectation value of ratios of random matrix determinants can be used as a
generating functional for resolvents and thus for all eigenvalue correlation functions. In
the supersymmetric approach the determinants are expressed as integrals over fermionic
E-mail addresses: akemann@spht.saclay.cea.fr (G. Akemann), vernizzi@spht.saclay.cea.fr (G. Vernizzi).
doi:10.1016/S0550-3213(03)00221-9
G. Akemann, G. Vernizzi / Nuclear Physics B 660 [FS] (2003) 532556
533
and bosonic variables and we refer to [4] for a review (see also [5] for recent results). An
alternative approach is the replica method, where new exact results have been obtained
by [6] and [7] (see also [8]). A third way which we also adopt in the following, is the
method of orthogonal polynomials. Arbitrary ratios of characteristic polynomials have
been calculated recently in [8,9], including a rigorous universality proof for the Unitary
Ensemble [10]. In this article we will focus on products of characteristic polynomials
only, but it would be desirable to generalize the above mentioned methods to complex
matrix models. In principle products alone contain already all informations about the
eigenvalues. When the matrices are Hermitian, products of characteristic polynomials
exhibit an interesting duality property, which interchanges the size of the matrix with
the number of determinants [11,12]. Moreover, in the application to QCD they enjoy a
direct interpretation as partition functions including quark masses [2]. They also give the
orthogonal polynomials and the kernel with respect to a weight function which includes
such quark mass terms. Consequently all massive correlation functions can be obtained
from them [13].1 In fact, many of the results for characteristic polynomials have been first
obtained in the QCD-related literature in the large-N limit, without explicitly calling them
products of characteristic polynomials but massive partition functions. For the unitary
ensemble of random matrices they were first calculated in [14] including a universality
proof. For the chiral unitary ensemble the partition functions were derived in [15] for
finite-N and infinite-N and shown to be universal [16]. For the other symmetry classes of
chiral orthogonal and symplectic ensembles the massive partition functions were derived
in [17] and [18], and their universality has been proved [18]. The corresponding non-chiral
results were obtained in [19]. Results for finite-N and infinite-N were independently found
for all three non-chiral symmetry classes in [20] and [11].
Up to now such results were little explored for random matrix models with complex
matrices having complex eigenvalues. Our aim is to provide an answer at finite-N and
large-N for products of characteristic polynomials where one has to distinguish between
the matrix and its Hermitian conjugate now. Since the introduction of complex matrix
models by Ginibre [21] they have been studied in the context of a two-dimensional
Coulomb plasma (see [22] for a review). Recent developments in several fields have
renewed the interest in complex matrix models. They have been related, for example,
to the SaffmanTaylor instability on the interface of two-dimensional fluids [23], and
correlation functions of complex eigenvalues at large distance were calculated in [24].
The correspondence between supersymmetric YangMills theory and string theory has
also boosted the interest in expectation values of traces of complex matrices [25]. There,
the complex matrix model belongs to the Ginibre ensemble and is a tool for computing
operators in a peculiar large-N limit, the so-called BerensteinMaldacenaNastase (BMN)
limit [26]. The computational techniques range from combinatorics, character expansion,
orthogonal polynomials, up to large-N loop equation techniques [27] or collective field
theory [28].
1 Here we are not considering the relation to finite volume partition functions of QCD, but only to random
matrix models partition functions.
534
Matrices with complex eigenvalues occur also in QCD when a chemical potential for
the quarks is introduced [29]. As QCD belongs to the chiral symmetry class, a chiral
complex eigenvalue model has been introduced and solved in [30]. It successfully describes
QCD lattice data with chemical potential [31]. The non-chiral complex model with unitary
symmetry, corresponding to three-dimensional QCD, was previously studied in [32].
There, the massive partition functions as well as all correlation functions were determined
under the hypothesis that the mass terms provide a non-negative definite measure. Here,
we will generalize these results to arbitrary products of characteristic polynomials, without
any restriction on the sources.
The technique we use in the present article is based on general orthogonal polynomials
in the complex plane. Many important properties that usually hold for such polynomials
on the real axis (e.g., the existence of a three-step recursion relation or the Christoffel
Darboux formula) are no longer valid in the complex plane in general. Our results will rely
solely on the existence of the orthogonal polynomials.
The article is organized as follows. After some definitions we first state and prove our
main result in Section 2 for any product of characteristic polynomials of complex matrices
and their Hermitian conjugate. As a nice consequence we also find a simple determinant
expression for products of characteristic polynomials of Hermitian matrices, when the
number of factors is odd. In Section 3 we give an interpretation of our findings. In some
special case they give massive orthogonal polynomials and their kernel, which leads to
all massive eigenvalue correlation functions. We also make contact to previous results
in [32]. In Section 4 we provide some explicit examples of orthogonal polynomials in
the complex plane at finite-N . We will also extend the duality [11,12] between products
of characteristic polynomials of different matrix size to complex matrices. Section 5
is devoted to the microscopic large-N limit. The known results for the asymptotics of
orthogonal polynomials and their kernel at weak and strong non-hermiticity directly apply
to the asymptotics for the characteristic polynomials. Also, the proof of universality
at weak non-hermiticity [33] can be extended to our case. We note however that at
strong non-hermiticity only the kernel has a smooth microscopic large-N limit (and not
the polynomials). Before the conclusions we also briefly comment in Section 6 on the
informations about the BMN large-N limit that can be extracted from the products of
characteristic polynomials.
2. Products of characteristic polynomials

We define our complex matrix model partition function and the corresponding
expectation values by

dJ dJ w J, J ,
ZN
(2.1)

1
dJ dJ w J, J O J, J ,
ON
(2.2)
N
Z
respectively. Here J is a general complex matrix of size N N and we integrate over all
independent matrix elements. We restrict ourselves to real non-negative weight functions
535
w(J, J ). In general a complex matrix can be diagonalized by using two unitary matrices.
Alternatively it is possible to transform it into a triangular form by a single unitary
transformation, the Schur decomposition
J = U (Z + R)U 1 ,
Z = diag(z1 , . . . , zN ),
U U (N), zi C,
(2.3)
where R is a strictly upper triangular matrix with complex entries. We assume that the
above partition function can be written solely in terms of the complex eigenvalues zi ,
i = 1, . . . , N , and that the weight w(J, J ) factorizes. We thus can write

N
2
2

d zi w zi , z i N (z) C(N)ZN ,
N = C(N)
Z
(2.4)
D i=1
where we have used that the Jacobian of the transformation (2.3) yields the Vandermonde
determinant denoted by
N (z)
N

(zi zj ) =
i>j
det
1i,j N
j 1
zi
(2.5)
The constant C(N) contains the integral over the unitary group U (N) as well as the integral
over the upper triangular matrix R. The latter drops out also in other expectation values, as,
for example, in the determinants we consider below. In Eq. (2.4) D denotes the domain in
the complex plane over which we integrate the complex eigenvalues. Examples for weight
functions satisfying the above conditions are2

w J, J = exp tr V J, J , with V J, J = gJ J + V (J ) + V (J ) . (2.6)
Here V (J ) can be taken, for example, as an arbitrary complex polynomial and g is a real
coupling constant. The domain D can be the full complex plane or any bounded domain,
depending on the weight function. For instance, if

w J, J = tr J J r 2 or w J, J = tr J J r 2 ,
(2.7)
then the domain is obviously a circle or a disk of radius r around the origin in the complex
plane, respectively.

If the weight function w(z, z ) is such that all moments exist, D d 2 z w(z, z )|z|2n < ,
n N, then the orthogonal polynomials Pn (z)

(2.8)
d 2 z w z, z Pk (z)Pl (z) = hk kl ,
D
exist as well and are uniquely determined by the monic normalization Pk (z) = zk +
O(zk1 ) with norm hk . Note that the coefficients of Pl (z) are complex in general. However
l (z) can
if both w(z, z ) and the domain D are symmetric under the exchange z z then P
2 In our conventions we will not put a factor of N into the exponent in order to have exact results for finite-N .
When taking the large-N limit it can be reestablished by scaling zi N zi and by suitably rescaling the
coupling constants in the potential V .
536
be chosen with real coefficients. The theory of orthogonal polynomials in the complex
plane (and their asymptotics) is a well-established subject of mathematics; for further
details, we refer the interested reader to the mathematical literature (e.g., [34]).
We can easily express the eigenvalue partition function Eq. (2.4) in terms of the norms
hk by replacing the Vandermonde determinant Eq. (2.5) with a determinant of the monic
polynomials Pk (z) and using the orthogonality relation (2.8):
ZN = N!
N1

hk .
(2.9)
k=0
In order to state our results we also introduce the orthonormal polynomials
Pk (z) hk 2 Pk (z).

1
(2.10)
With these preliminaries we can turn to the main result of this article.
Theorem 1. Let {vi ; i = 1, . . . , K} and {ui ; i = 1, . . . , L} be two sets of complex numbers
which are pairwise distinct among each set. Without loss of generality we assume K L,
where the empty set with L = 0 is permitted as well. Together with the definitions and
restrictions on the weight function made above, the following statement holds:3

K
L

det[vi J ]
det u j J
j =1
i=1
N+K1
=
i=N
1
2
hi
N+L1
j =N
K (v)L (u)
1
2
hj
det

B vl , u m ,
(2.11)
for m = 1, . . . , L,
for m = L + 1, . . . , K.
(2.12)
1l,mK
where we have defined:

N+L1

Pi (vl )Pi (um )
i=0
B vl , u m
PN+m1 (vl )
Before presenting the proof of Theorem 1 let us make a few remarks. The reason why we
restrict ourselves to the case K L is that the other case K L can be obtained simply
by complex conjugating the above equations. The restriction to pairwise distinct sets of
parameters vi or ui can also be lifted. In the limit vi vj the Vandermonde determinant
in the denominator vanishes as well as the determinant in the numerator. Hence, in this limit
the row containing vj has to be differentiated with respect to vj and then we set vj = vi
there. The same procedure can be iterated when vi is degenerate of order m. The result is
given by substituting the kth row of Eq. (2.12) for k = 1, . . . , m with its (k 1)th derivative
at vi (and dropping the corresponding arguments in the Vandermonde determinants in the
denominator). The same argument applies for the case with degenerate u i . We will give an
example for L = 0 in Eq. (4.19).
3 The following notation is understood: (x) = (x) = 1 and N1 h = 1.
0
1
i=N i
537
We also wish to mention that the object occurring in the first L columns of the matrix
B(vl , u m ) is, apart from the weight factors, nothing else than the kernel of orthogonal
polynomials defined as:
KN

1 N1
v, u w v, v w u, u 2
Pi (v)Pi (u)

i=0

1

w v, v w u, u 2 N v, u .
(2.13)
Here we have also explicitly defined the bare kernel N (v, u)

as it occurs in Theorem 1 and
subsequent formulas. We note that this kernel cannot be written in terms of polynomials
of order N and N 1 alone because for orthogonal polynomials in the complex plane the
analog of the ChristoffelDarboux formula, Eq. (2.18) below, does not hold in general (for
an example, see [33]).
Proof. The proof will be done in two steps. First we will prove the case K = L for arbitrary
L. Then we consider the case K > L using induction. For K = L we have:

L
L

det[vi J ]
det u j J
j =1
i=1

L
N

1
vj zi u j zi
N (z)N z
=
d 2 zi w zi , z i
ZN
j =1
D i=1

N

1
2
=
d zi w zi , z i
ZN
D
i=1
N+L (z1 , . . . , zn , v1 , . . . , vL ) N+L (z1 , . . . , z n , u 1 , . . . , u L )

L (v1 , . . . , vL )
L (u 1 , . . . , u L )

N
2

j 1
j 1

1
d zi w zi , z i
det
i
det
i
,
=
ZN L (v)L (u)
1i,j N+L
1i,j N+L
i=1
D

{z, v},
with
{z, u}
N+L1

N
2

hi
i=0
=
d zi w zi , z i
det
Pj 1 (i )
ZN L (v)L (u)
1i,j N+L

D i=1
Pj 1 (i )
det
N+L1
hi
=
N!L (v)L (u)
i=N
1i,j N+L

N
2

d zi w zi , z i
D i=1
det
1i,j N+L
N+L

Pr1 (i )Pr1 (j ) .
r=1
(2.14)
In the second step we have put the products of differences between vj and zi into a
larger Vandermonde determinant of size N + L and between u j and z i into a complex
538
conjugated one. The additional factors of differences between vj s and u j s have been
divided out in terms of smaller Vandermonde determinants of size L. The Vandermonde
determinant can be written as a determinant of powers, Eq. (2.5), where we have introduced
a larger set of variables 1,...,N+L and 1,...,N+L . The properties of determinants allow
j 1
to replace the powers i
first by the monic polynomials Pj 1 (i ) and then by the
orthonormal ones times their norms, which have been taken out of the determinants. They
partially cancel the normalization ZN from Eq. (2.9). We then apply that det[A] det[B] =
det[AB T ] and thus obtain a single determinant of the bare kernel Eq. (2.13). Now
we make use of Theorem 5.2.1 in [35], which can be applied as follows. The kernel
K
satisfies the hermiticity condition KN (, )
= KN (, ) and the chain relation
N2(, )
d z KN (, z )KN (z, )
= KN (, ).
Then

d 2 zn det KN zi , z j = (c n + 1)
(2.15)
det
KN zi , z j ,

1i,j n
1i,j n1
where c = d 2 z KN (z, z ) = N . We can thus successively integrate out all z1,...,N using
Eq. (2.15). The successive use produces a factor N! and we thus arrive precisely at
Eq. (2.11) for K = L, with the matrix B only consisting of the bare kernels.
Next we prove Eq. (2.11) for K > L by induction in K L. The starting point of
induction is already satisfied for K = L. For the induction step K K + 1 we proceed as
follows:

K+1
L

det[vi J ]
det u j J
j =1
i=1

N
1
ZN K+1 (v)L (u)

d 2 zi w zi , z i
D i=1
(1) (z1 ) P
(N) (zN )P (N+1) (v1 ) P (N+K) (vK )
(1) P
{0,...,N+K}
(N+K+1) (vK+1 )
P

(1) P (1) (z1 ) P (N) (zN )P (N+1) (u1 )
{0,...,N+L1}
N+K
=
i=N
hi2
P (N+L) (uL )
N+L1
j =N
hj2
K+1 (v)L (u)

PN+K (vK+1 )
K

j =1
det
1l,mK
PN+K (vj )

B vl , u m
det
m=1,...,K
l=1,...,j 1,K+1,j +1,...,K

B vl , u m
.
(2.16)
We repeated the first steps in Eq. (2.14) until we have the product of two determinants of
monic polynomials and their conjugate, containing the variables {z1,...,N , v1,...,K+1 } and
539
{z1,...,N , u 1,...,L }, respectively. We have explicitly spelled out these determinants as sums
over permutations. The index (N + K + 1) of the polynomial P (N+K+1) (vK+1 ) can
take values in {0, . . . , N + K}. For (N +
+ 1) = N + K
the polynomial PN+K (vK+1 )
K
L
K
multiplies exactly the expectation value i=1 det[vi J ] j =1 det[u j J ]N for which
Theorem 1 holds by using the assumption of induction. The other values for the index
(N + K + 1) < N + K can be obtained by simply exchanging the argument vK+1 with the
one of the corresponding polynomial of argument vj =1,...,K . Since this is a pair permutation
in it will produce a sign for all terms. As a final step, in the last line of Eq. (2.16) we
move the row containing vK+1 from the position j to the lowest row of position K, without
changing the order of the other rows in the determinants. This produces a factor (1)Kj .
Then we see that the sum of all determinants is just the Laplace expansion of a matrix
of size K + 1 with respect to the last column, and that this matrix is nothing else than
B(vl , u m ) in Eq. (2.12). We have thus proved Theorem 1 for K + 1.
Let us emphasize that Eq. (2.11) in Theorem 1 also holds for partition functions defined
as eigenvalue integrals such as Eq. (2.4) and which do not necessarily have a representation
in terms of complex matrices. Examples for such models are the chiral complex matrix
model as recently proposed in [30] which we will discuss again in Section 4. For this
model no explicit matrix realization is known so far. Another example are weight functions
containing mixed higher powers of z and z such as w(z, z ) = exp[(zz)k ], k 2. Such
weight functions can be only written down in terms of complex matrices which are normal,
[J, J ] = 0, as normal matrices can be diagonalized without an upper triangular matrix as
in Eq. (2.3). However, such terms with higher powers in |z|2 may be necessary in the
weight function to write down convergent integrals.
After having completed the proof let us compare to the known results for products of
characteristic polynomials of Hermitian matrices J = H = H . Since in our proof we have
only assumed the existence of an eigenvalue representation for the partition function and
the existence of orthogonal polynomials on some domain D, it holds also for arbitrary
Hermitian matrix models. We only
have to restrict D to the real line and replace complex
integrals d 2 z by real integrals dx. Because of J = J the splitting into two sets of
determinants becomes immaterial here. In particular for L = 0 we find back the known
result (see, e.g., in [11,12,20])
K

1
det
(2.17)
det[vi H ] =
PN+m1 (vl ) .
K (v) 1l,mK
i=1
The product of K characteristic polynomials is thus given by a K K determinant of the

orthogonal polynomials. In [10] it was shown that for even K = 2k this can be further
simplified to a k k determinant. We obtain the same result from setting K = L = k in
our Eq. (2.11) in Theorem 1 and using that for orthogonal polynomials on the real line the
ChristoffelDarboux identity holds,

N1

hN PN (x)PN1 (y) PN (y)PN1 (x)
(2.18)
,
Pi (x)Pi (y) =
hN1
xy
i=0
540
with x, y real numbers.4 We thus have
k

det[vi H ] det[ui H ]
i=1
N+k1
hi
i=N
(hN+k1 )k k (v)k (u)

N+k (vl )PN+k1 (um ) PN+k (um )PN+k1 (vl )
P
det
.
1l,mk
vl um
(2.19)
As a simple corollary we can thus show from Eq. (2.11) in Theorem 1 that also for and odd
number K = 2k + 1 of products of characteristic polynomials the K K determinant in
Eq. (2.17) can be reduced down to size (k + 1) (k + 1).
Corollary. Let the definitions (2.1)(2.5) and (2.8)(2.10) be valid as well for Hermitian
matrices H with real eigenvalues integrated over a real domain D. Take the parameters
(real or complex) {vi ; i = 1, . . . , k + 1} and {ui ; i = 1, . . . , k} to be all pairwise distinct.
From Eq. (2.11) in Theorem 1 the following relation follows:
k+1

k

det[vi H ]
det[uj H ]
j =1
i=1
N+k1
hi
i=N
det
(hN+k1 )k k+1 (v)k (u) 1l,mk+1
where we have defined

B(vl , um )

B(vl , um ) ,
N+k (vl )P
N+k1 (um )P
N+k (um )P
N+k1 (vl )
P
vl um
PN+m1 (vl )
for m = 1, . . . , k,
for m = k + 1.
(2.20)
(2.21)
It is clear that from Eq. (2.11) in Theorem 1 for real Hermitian matrices we can get
many equivalent formulas for determinants of size in between Eq. (2.17) and Eqs. (2.19)
or (2.20), respectively, by splitting the total number of parameters in different pairs K
and L.
3. Applications
In this section we give an interpretation for some particular products of characteristic
polynomials we calculated in Eq. (2.11). From the beginning of our investigations we have
assumed that the weight function has to be non-negative definite on the domain D in order
4 Formula (2.18) holds algebraically also for x, y C and P (x) being the analytic continuation to the
k
complex plane of the polynomials which are orthonormal on the real line. However, we emphasize that in this
N1
case the sum i=0 Pi (x)Pi (y) is not a kernel in the complex plane, because the Pk are not longer orthonormal
in the complex plane (for an explicit example see Eq. (4.7)).
541
to make the construction of the polynomials via GramSchmidt possible. The examples
given for w(z, z ) in Eqs. (2.6) and (2.7) obviously satisfy this requirement. We will now
use the results from our Theorem 1 to explicitly construct orthogonal polynomials for
more complicated weight functions containing determinants or in other word weights with
products of eigenvalues as prefactors.
From the theory of orthogonal polynomials defined on the real line the following is
known. Choosing any parameter vj R in Eq. (2.17) for Hermitian matrices the right-hand
N (vj ), which is orthogonal with
side can be interpreted as a polynomial in
vj of order N , P
respect to the weight w(K1) (x) w(x) i=j (vi x). This holds provided that the new
weight is non-negative definite on the domain D. Note that under this condition, Eq. (2.17)
immediately follows from the Christoffel theorem [36]. An example is when D R and
all vi are purely imaginary, coming in complex conjugate pairs. This would correspond to
the weight for the random matrix partition function of QCD in three dimensions.
In order to make such an interpretation of the characteristic polynomial in the complex
plane possible we have to choose K = L + 1 in Eq. (2.11) and to take all ui = vi ,
i = 1, . . . , L. It is easy to convince oneself by going to an eigenvalue basis that the
following matrix representation holds:

L

(L)
det[vi J ] det vi J
.
PN (z) det[z J ]
(3.1)
i=1
This gives a polynomial orthogonal with respect to the non-negative definite weight
L

det[vi J ] det vi J ,
w(L) J, J w J, J
(3.2)
i=1
and the complex conjugate polynomial PN(L) (z) is simply obtained by taking the Hermitian
conjugate of Eq. (3.1). Please note the exact analogy with a similar formula for orthogonal
(L)
polynomials on the real line (see, e.g., [36]). In order to satisfy the normalization PN (z) =
N
z + we define

det[z J ] L
i=1 (det[vi J ] det[vi J ]) N
(L)

.
PN (z)
(3.3)
L


Let us give an example for L = 1:
N+1 (z, v)
PN+1 (v) N+1 (v, v)
PN+1 (z)
(1)
PN (z) =
,
(v z)N+1 (v, v)
(3.4)
where we have used the bare kernel N+1 (v, u)

Eq. (2.13) and the superscript (L = 0) has
been dropped. We can also calculate the norm of the polynomial given by
N+2 (v, v)
,
N+1 (v, v)
and we obtain for the orthonormal polynomials

(1)
hN = hN+1
PN(1) (z) =
N+1 (v) N+1 (v, v)P

N+1 (z)
N+1 (z, v)P

.
(v z) N+1 (v, v)
N+2 (v, v)
(3.5)
(3.6)
542
We have not been able to prove a similar expression for the orthonormal polynomials at
general L. We expect to obtain the following result for the norm of PN(L) (z):
(L)
hN = hN+L
deti,j =1,...,L [N+L+1 (vi , vj )]

,
deti,j =1,...,L [N+L (vi , vj )
(3.7)
which we have checked for L = 1, 2, 3, 4, 5. This would give us PN(L) (z) in a form very
similar to [37].
After having determined the orthogonal polynomials for weight functions w(L) (J, J )
we can use them to obtain the respective kernel as defined in Eq. (2.13):
(L)

1 N1
(L)
z, u = w(L) z, z w(L) u, u 2
Pi (z)Pi (u).
(L)
KN
(3.8)
i=0
While we have an explicit expression for L = 1 in Eq. (3.6) we would still have to calculate
(L)
the norms h(L)
N of the polynomials PN (z) which are of the conjectured form Eq. (3.7).
However, we can also directly read off the kernel from Eq. (2.11) in Theorem 1. It is known
that not only the polynomials have a matrix representation as in Eq. (3.3) but also the kernel
itself. Choosing the same weight Eq. (3.2) one has to insert two more determinants instead
of only one as for the polynomials:

1
KN(L) z, u = w(L) z, z w(L) u, u 2

det[z J ] det[u J ] L
i=1 (det[vi J ] det[vi J ]) N1
. (3.9)

hN L
This can be shown along the same lines as in [38] where the same statement was made for
Hermitian matrices. Note the different matrix size of the expectation value for numerator
and denominator. We also give the example with L = 1 explicitly as it follows from
Eq. (2.11):

N+1 (z, v)

N+1 (z, u)

(1) (1)
1 N+1 (v, u)
N+1 (v, v)
(1)
2
.
KN z, u = w z, z w u, u
(3.10)
(v z)(v u)
N+1 (v, v)
While Eqs. (3.8) and (3.9) are equivalent by definition, for general L it is easy to see
that already for L = 1 this yields a highly non-trivial identity, comparing Eq. (3.10) and
Eq. (3.8) with the polynomials Eq. (3.6) inserted. From a practical point of view the second
form Eq. (3.9) which can be directly deduced from Eq. (2.11) in Theorem 1 is much more
convenient as it is only a ratio of two determinants. Compared to this in Eq. (3.8) we have
to perform a sum over ratios of determinants of growing size.
Usually the aim of calculating orthogonal polynomials and their kernel is to evaluate
correlation functions of eigenvalues defined as
(L)
RN (z1 , . . . , zk )
N!
1
ZN (N k)!
N

D l=k+1
d 2 zl
N

i=1
2

w(L) zi , z i N (z) .
(3.11)
543
Using standard orthogonal polynomials techniques [35] they are given by

(L)

(L)
RN (z1 , . . . , zk ) = det KN zi , z j .
(3.12)
i,j =1,...,k
Therefore, from Eq. (3.9) we can immediately read off all k-point correlation functions.
(L)
They are given by a k k determinant of the kernel KN (zi , z j ) which is itself again a
ratio of determinants of size (L + 1) (L + 1) over L L. There is an even more compact
expression for the k-point functions which has been introduced in [32]. It uses the fact that
the correlation functions with weight w(L) (J, J ) from Eq. (3.2) can be entirely expressed
in terms of higher n-point correlation functions with weight w(J, J ) at L = 0:5
(L)
RN
(z1 , . . . , zk ) =
(0)
RN+L
(z1 , . . . , zk , v1 , . . . , vL )
(0)
(3.13)
RN+L (v1 , . . . , vL )
This result reduces the k-point correlation function to a single ratio of determinants of size
(L)
(L)
(k + L) (k + L) over L L. While for the 1-point function RN (z) = KN (z, z ) it is easy
to see that Eqs. (3.13) and (3.12) together with Eq. (3.9) agree perfectly, for higher (k > 2)point functions their equivalence requires highly nontrivial identities among determinants.
The fact that several different (but equivalent) determinant formulations exist, depending
on the way they are derived, is a common phenomenon also for correlations functions
of Hermitian matrices. In some cases their equivalence can be directly shown [39] (see
also [40]). Let us mention that a mapping of correlation functions with different weight
functions as in Eq. (3.13) has already been very useful for matrix models with real
eigenvalues [18,32,41].
4. Explicit results for finite-N

In the first subsection we will give some explicit examples for weight functions and
their corresponding orthogonal polynomials and kernel at finite-N . These are the objects
that then have to be inserted into Eq. (2.11) in Theorem 1. In particular our second example
will be studied also in Section 5, when taking the large-N limit. In the second subsection
we derive a finite-N duality relation between characteristic polynomials of different matrix
size.
4.1. Examples for orthogonal polynomials in the complex plane
We start with the Ginibre ensemble where the weight function is defined on the full
complex plane D = C as

wG J, J
N

1

1
exp tr J J =
exp zi zi .
i=1
5 Note the typo in the index of the denominator in Eq. (2.12) of Ref. [32].
(4.1)
544
As one can easily convince oneself the orthogonal polynomials are monic powers,
Pj (z) = zj ,
hj = j !.
(4.2)
We thus have from the orthonormal polynomials Pj (z) = zj (j !)1/2 a simple expression
for the kernel
(v u)
2(N, v u)

N1
i
N v, u =
(4.3)
= exp v u
,
i!
2(N)
i=0

where 2(a, z) = z dt t a1 et is the incomplete gamma function. Actually, all domains
and weight functions in the complex plane which are invariant under the rotation
z ei z, [0, 2] (i.e., they are functions only of the radial coordinate = |z|),
give orthogonal polynomials which are monic powers, the only difference being the
normalization coefficients. For instance, Pj (z) = zj are indeed orthogonal polynomials
with respect to the flat measure wconst(z, z ) = 1 on the disk Dr = {z C: |z| < r} and
on the circle Cr = {z C: |z| = r}, with normalization coefficients hj = r 2j +2 /(j + 1)
and hj = 2r 2j +1 , respectively. Of course such a difference propagates to the explicit
expression for the bare kernels:
j
N1

1
v u
1 1 a N (1 + N aN)
(j + 1) 2
= 2
, in Dr ,
N v, u = 2
r
r
r
(1 a)2
j =0

1
N v, u =
2r
N1

j =0
v u
r2
j
=
1 1 aN
,
2r (1 a)
in Cr , a =
v u
.
r2
(4.4)
As a consequence of the results in Section 3 we can

also immediately read off explicitly the
polynomials for the weight function w(L) (z, z ) = L
) from
i=1 det[(vi z)(vi z )]w(z, z
Eq. (3.3) (which is not rotationally invariant). For L = 1 they are given by:

N (v v)
(zv)
i
i
N+1
N j

v N+1 N
j +1 (v, v)
z
i=0 hi z
i=0 hi
(1)
N
. (4.5)
=
v
PN (z) =
N (v v )i
v N+1 (v, v)
(v z)
i=0
hi
j =0
Note that Eq. (4.5) holds for any set of monic orthogonal polynomials. Moreover, note
that we started from polynomials for L = 0 with real coefficients Eq. (4.2) while those
of the new polynomials with L = 1 are now complex. This example shows a well-known
remarkable fact. When D is a real interval there is a one-to-one correspondence between
the distribution of the eigenvalues and the distribution of the zeros of the orthogonal
polynomials. However, in the complex plane such a close correspondence breaks down.
For instance, Eq. (4.2) shows that all the zeros are atthe origin z = 0, whereas the spectral
density is located on a whole disk of radius r N in the complex plane. This is a
particular case of a general theorem according to which all the zeros of the complex
orthogonal polynomials lie inside the convex hull of the support [34].
The following example is for the weight

2
1
2
wH J, J exp
(4.6)
tr
J
J
+
J
, [0, 1],
J
2
1 2
545
which is again defined on D = C and it is not rotational

invariant when = 1. Here
the complex matrix is parameterized as J H + iA (1 )/(1 + ) and H and A
are Hermitian matrices with equal Gaussian weight. The parameter measures the
degree of non-hermiticity, a concept which was introduced in [42]. In particular this
allows to take the Hermitian limit by sending 1. In this limit lim 1 wH (J, J )
(Im J ) exp[ tr(Re J )2 ]. This possibility makes the difference between the complex
matrix model and two-matrix models particularly transparent. In the former we have
introduced polynomials of a complex variable Pj (z) as well as their complex conjugate.
While we can smoothly take the limit of z becoming real here, this is not possible in the
two-matrix model solved by the method of bi-orthogonal polynomials. The orthogonal
polynomials for the weight in Eq. (4.6) can be shown to be Hermite polynomials [43],
Pj (z) =

j
z
2
,
Hj
2
2

hj = j ! 1 2 .
(4.7)
Using that the Hermite polynomials have real coefficients their kernel is given by
N (z, u)
=
N1
1 j z u
1
Hj
Hj
.
2
2
1 2 j =0 j ! 2
(4.8)
An other example is for a bounded domain which is not rotationally invariant, e.g., the
ellipse

1 i
ell
2 1 i
, [0, 2] , r > c 0.
Dr = z C: |z| < re + c r e
(4.9)
2
The shape of the ellipse is parameterized by two parameters, r and the semi-focal length c.
Obviously the minor/major axis lengths are equal to r c2 r 1 , r + c2 r 1 , respectively.
It is easy to see that in this case the monic orthogonal polynomials with respect to a flat
measure wconst(z, z ) = 1, are given by the Chebyshev polynomials of the second kind:

j
2 2j +2
z
c
c
,
hj = 2j +2
r 2j +2
Pj (z) =
(4.10)
.
Uj
2
c
2
(j + 1)
r
It is easy to check that Eq. (4.10) in the limit c 0 correctly reproduces the case for the
circular disk Dr . The kernel reads:

4 N1
c2j (j + 1)
v
u
N v, u =
(4.11)
U
Uj
.
j
r 2j +2 (c2 r 1 )2j +2
c
c
j =0
The final example we want to give are the generalized Laguerre polynomials in the
complex plane as recently introduced in [30]. For this chiral complex eigenvalue model
no simple matrix representation is known.6 Therefore, we have to slightly modify our
definitions (2.4) and (2.8) in order to be formulated in terms of complex eigenvalues. This
model deserves particular attention due to its physical applications [30,31].
6 A model including two complex matrices that are simultaneously diagonalizable can be constructed though.
546
The partition function is given by

ZNch

N
2

2
d zi wch zi , z i N z2 ,
(4.12)
D i=1
with D being the full complex plane and the weight

1
2
2
2
wch z, z |z|2a+1 exp
z
+
z
|z|
,
1 2
2
1
a > R.
2
(4.13)
We could also allow for more general weight functions and domain D, the crucial point for
the model to be chiral being the squared argument inside the Vandermonde determinant.
The orthogonal polynomials are then necessarily polynomials in z2 :

hk kl = d 2 z w(z, z )Pkch z2 Plch z2 .
(4.14)
D
The characteristic polynomials consequently have to be defined in terms of the eigenvalues

NK
L

2

2
2
2
vj zi
u l z i
.
(4.15)
i=1 j =1
l=1
The whole proof of Eq. (2.11) in Theorem 1 goes through as before, replacing all arguments
by their square on the right-hand side.
The finite-N orthogonal polynomials for our example Eq. (4.13) are given by the
generalized Laguerre polynomials [33]
2

ch 2
k
k
a z

.
Pk z = (1) (2 ) k!Lk
(4.16)
2
Their norm can be calculated to be
2(a + k + 1)k!
hk = f a ( )22k
,
2(a + 1)

a 3
1
3
a
2
2 2+4
1
,
Pa+ 1
f ( ) = d z wch z, z = 2 a +
2
2
1 2
(4.17)
where Pa+1/2 (x) is the Legendre function. The bare kernel is thus reading

N z, u =
2
N1
u
1 2(a + 1)k! 2k a z2
Lk
Lak
.
a
f ( )
2(a + k + 1)
2
2
(4.18)
k=0
4.2. A duality relation

In this subsection we wish to prove a duality relation that relates products of
characteristic polynomials of different (finite) matrix size. This generalizes the relation
found in [11,12] to complex matrices. The proof goes back to a relation between
determinants of Hermite polynomials [11] which we will use. Therefore, our duality
547
relation will only hold for the weight Eq. (4.6) having Hermite polynomials in the complex
plane.
To begin with we consider Eq. (2.11) in Theorem 1 at L = 0 where only polynomials
and no kernels appear. Next we take the limit of equal arguments vi = v, i = 1, . . . , K.
We obtain

(l1)

1
det
v
det[v J ]K N = K1
(4.19)
PN+m1 (v) .
1l,mK
j =1 j !
This looks exactly as for the Hermitian matrix model (e.g., in [11]). Next we insert the
orthonormalized Hermite polynomials Eq. (4.7) for the chosen weight Eq. (4.6) and obtain

det[v J ]K N

NK+K(K1)/2
K(K1)/2
v
2
1
(l1)
HN+m1
= K1
2
det
2
1l,mK
2
j =1 j !

K(K1)/2

NK
1
1
v
2
. (4.20)
= K1
HN+m+l2
det
2
2
1l,mK
2
j =1 j !
In the first step we have taken out all factors of the determinant
stemming from the monic
(l1)
normalization and the differentiation, where HN+m1 (v/ 2 ) denotes the derivative with
respect to the full argument v/ 2 . In the second step we have used the following relation
for the Hermite polynomials,
HN (x) = 2xHN (x) HN+1 (x),
(4.21)
which holds algebraically for any x being real or complex. We have thus shifted the
derivative into the index of the polynomials, taking out all signs. We can now use the
following relation among determinants of Hermite polynomials as being shown in [11]7
det1l,mK [HN+m+l2 (x)]
det1l,mN [HK+m+l2 (ix)]
= (i)KN
.

K1
(2)K(K1)/2 j =1 j !
(2)N(N1)/2 N1
j =1 j !
(4.22)
This relation was proved in [11] for real x only. However, the relation is purely algebraic
as it relates two polynomials after multiplying out the
two determinants. We can, therefore,
also allow for x to be complex, in particular x = v/ 2 . Note also the factor of i in the
argument on the right-hand side.
Replacing the determinant of size K K in Eq. (4.20) by a determinant of size N N
with the help of Eq. (4.22), we arrive at the following remarkable identity:

det[v J ]K N = (i)KN det[iv J ]N K ,
(4.23)
interchanging the power of the product with the matrix size. It is identical to the
corresponding equation in [11] replacing the complex matrix J and argument v by the
respective Hermitian matrix H and real argument x. The normalization in Eq. (4.23) can
7 In [11] this relation was shown for the determinant of C (x) 2n H (x), but the factors of 2 can be easily
n
n
taken out. We further note a typo in Eq. (3.16) there.
548
be easily checked taking the limit v . Another peculiar case is N = K at v = 0. Still,

the duality relation is satisfied as both expectation values which are now equal vanish for
odd N = K due to symmetry.
The relation Eq. (4.23) trivially extends to the Ginibre ensemble Eq. (4.1), which can
be obtained by taking 0 on the Hermite polynomials Eq. (4.7). It is because in the
Ginibre case all expectation values of powers of J vanish. The same argument applies
for all weights with spherical symmetry and monomials as orthogonal polynomials, as in
Subsection 4.1. We suspect that the duality also holds for other weight functions in C. For
other examples on R see [44].
5. The microscopic large-N limit and universality

The explicit examples given in the last section are a good starting point for the analysis
of the
asymptotic large-N limit. To this aim we first rescale the complex eigenvalues by
zi Nzi in order to have N explicitly in the exponent of the weight functions Eqs.
(4.6) or (4.13), to which we will restrict in this section. Then we focus on the microscopic
large-N limit where the correlations of eigenvalues at small distance O(|zi zj |) 1 are
considered. For spectral correlators at a distance of O(1) (the so-called macroscopic limit)
we refer to [22,24,27].
Furthermore, one has to distinguish two different ways of taking the microscopic
N limit [42]. Starting from a general weight such as in Eq. (2.6) or simply from
the Ginibre ensemble Eq. (4.1) we will always end up in a regime which is called strong
non-hermiticity. Introducing an additional parameter [0, 1] as in Eq. (4.6) will allow us
to reach also the regime of weak non-hermiticity, when taking simultaneously N and
(1 ) 0 while keeping the product fixed. The latter limit introduced in [42] permits
to smoothly interpolate between correlations of real eigenvalues on one hand and complex
eigenvalues in the strong non-hermiticity limit on the other hand.
Let us begin with the weak non-hermiticity limit. Here we focus on a weight function
of the form of Eq. (2.6), which is in terms of eigenvalues

d

1
1
g
2k
w z, z exp N
z2k + z 2k
|z|2 z2 + z 2 +
. (5.1)
2
2
2k
1 2

k=2
It generalizes the weight Eq. (4.6).8 Here the g2k R are real coupling constants and
[0, 1] measures the degree of non-hermiticity as explained after Eq. (4.6). The weak
8 Strictly speaking the integral over the weight function Eq. (5.1) cannot be made convergent for any choice of
signs of the coupling constants g2k . Consequently one would either have to restrict the domain D to be compact
or to introduce a small term :|z|2d+2 to achieve convergence. However, in the weakly non-Hermitian large-N
limit the support of the eigenvalues does not only become compact but also gets projected onto the real line.
The correlators depend on complex variables only microscopically. For that reason we ignore the aforementioned
technical subtleties.
non-hermiticity limit is defined by taking [42]

lim N 1 2 2
549
(5.2)
to be constant. At the same time we rescale the complex eigenvalues according to

Nz = N(Re z + i Im z) ,
(5.3)
which is the microscopic limit at the origin since z 0 while N . It is important to

note that the scaling is done with the same power of N as for real eigenvalues, in contrast
to the strong non-hermiticity limit (see Eq. (5.9) below). This has important consequences
on the asymptotic limit of the orthogonal polynomials as well.
We will now give the asymptotic results for the orthonormal polynomials of Eq. (5.1)
and their kernel as they were derived in [33]:

lim Pn z =
n,N
N
1

cos(u(t) ) n even,

3 1
2
n
2
= 2 Nu (t) 1 2 2 e 4 u(t )
(5.4)
t= .
sin(u(t) ) n odd,
N
In [33] the full kernel Eq. (2.13) including the weights was given. Using the fact that the
asymptotic limit of the weight function reads

2
w z, z exp 2 (Im z)2 ,
(5.5)
we obtain for N (z1 , z 2 ) appearing in Eq. (2.11) in Theorem 1

1 (0)

2
1
du 2 u2
1
2
2 .
2
e
=
,
z
=
=
cos
u
lim
N
1
2
1
N N 2
N
N
(5.6)
Here we have introduced the spectral density of real eigenvalues t (x) corresponding to the
Hermitian limit of the potential in Eq. (5.1): for an explicit expression see, e.g., Eq. (3.10)
in [33]. The function u(t) in Eq. (5.4) is related to the spectral density by
tt (0) u(t).
(5.7)
Finally the asymptotic of the norm of the polynomials is:

u (t)
hN
(5.8)
.
The remaining N -dependence of the norm hN N! in the prefactors of Eq. (2.11) has to
be canceled by a suitable normalization. Inside the determinant it is needed when passing
from the monic to the orthonormal polynomials to make the asymptotic limit of the latter
well defined. To summarize we have determined the weakly non-Hermitian large-N limit
of the characteristic polynomials as stated in Eq. (2.11) in Theorem 1, when inserting
Eqs. (5.4), (5.6) and (5.8). Hence, it is also universal.
550
In [42] the weak non-hermiticity limit away from the origin was also analyzed, for the
Gaussian weight of Eq. (4.6). The corresponding
kernel is exactly of the same form of

Eq. (5.6), with replacing 1 (0) 1 X2 /4 which is the Wigner semicircle density
at the point
of investigation X R. While the norms hN still behave as in Eq. (5.8)
with u(t)
=
2t the asymptotic polynomials will acquire an additional oscillating phase
cos( 2t( + NX)) as we now microscopically rescale the deviation from point X:
z = X + /N , with z, C. In other words, their asymptotic limit does no longer exist.
Therefore, only for K = L a smooth limit is found at weak non-hermiticity away from
the origin (a similar situation is found below at strong non-hermiticity). Furthermore the
question of universality away from the origin remains open.
We now turn to the microscopic limit at strong non-hermiticity where we keep [0, 1)
fixed while N . The microscopic origin scaling limit is defined here by letting z 0
and N keeping
N z = N (Re z + i Im z) ,
(5.9)
constant. The reason for the different scaling is that now the eigenvalues will become
dense on a compact domain which truly extends into the complex plane. This has an
immediate consequence as the asymptotic large-N of the orthogonal polynomials no longer
exists. This can be seen in the simplest example of the Ginibre ensemble Eq. (4.1) or
of the bounded domains Eq. (4.4). The polynomials PN (z) = zN /N! from Eq. (4.2) will
certainly no longer have a smooth large-N expansion. The same problem occurs for the
Hermite
plane. Expanding, for example, the even powers,
polynomials in
the complex
H2j (z N/2 ) cos[ 4j + 1 z N/2 ] it can be seen that while the weak scaling limit
Eq. (5.3) is smooth (keeping j/N = t fixed) the strong limit from Eq. (5.9) does not exist.
While on the real line the existence of an asymptotic expansion of the polynomials and
the kernel are directly related through the ChristoffelDarboux formula Eq. (2.18) this is
no longer the case here, since such a relation does not hold in general in the complex
plane. However the asymptotic limit of the kernel Eq. (4.3) still exist and it is given by
the exponential function being smooth. The result for the weight in Eq. (4.6) reads after
splitting off the weight factors [42]:

1
2
1
N z1 = , z 2 =
lim
N N
N
N

1
1
2 2
exp
=
(5.10)
.
1 2
2
(1 2 )
(1 2 )
2 1
For 0 we recover the exponential Ginibre kernel [21]. For 0 < < 1 all eigenvalue
correlations calculated from Eq. (5.10) coincide with those of the Ginibre
ensemble after
a trivial rescaling [42]. Taking into account the rescaled norms hN 1 2 we can
read off from Eq. (5.10) the asymptotic strongly non-Hermitian large-N limit of our
characteristic polynomials Eq. (2.11) for K = L. The question of universality of Eq. (5.10)
for more general weight functions remains an open question.
The situation in the macroscopic large-N limit is different. It has been shown recently
that the connected two-point resolvent is universal while the higher connected three- and
four-point functions are non-universal [24]. This is in contrast to the Hermitian matrix
551
model, where it is well known that all connected k-point functions starting to lowest order
at O(N 22k ) are universal in the macroscopic large-N limit, including all higher order
corrections in 1/N 2 [45].
The same analysis for the microscopic large-N limit at weak and strong non-hermiticity
can be done for the chiral ensemble of complex eigenvalues introduced in Eqs. (4.12) and
(4.13). The asymptotic expressions for the norms, polynomials and kernel in the weak
limit, and the norms and kernel in the strong limit have already been explicitly given in
[33] and we will not repeat them here. We only stress that they are essentially different
from those given above, in both limits at weak and strong non-hermiticity. We find the
same phenomenon as above that only the case K = L at strong non-hermiticity makes
sense asymptotically. At weak non-hermiticity there are no such restrictions. The issue of
universality for the chiral ensemble remains an open problem.
We finish this section with a remark: in the mathematical literature the large-N
is known as Bergman kernel function or Szeg
macroscopic limit of our kernel KN (v, u)
kernel function, according to the case where the weight function is defined on a domain
in the complex plane or on its boundary. Both these kernels are objects of fundamental
importance in the theory of conformal mapping [34].
6. The BMN large-N limit

In this section we apply the results of Sections 3 and 4 to the study of objects like
tr J K tr J K N in the limit where both N and K become large. Such quantities have
become of interest in the correspondence between pp-wave strings and supersymmetric
YangMills theory [25]. In this case the expectation values are evaluated with respect to
the weight of the Ginibre ensemble Eq. (4.1). It has been found that besides the usual
(macroscopic) large-N limit, one can consider also K, N such that K 2 /N is finite.
This limit is called BMN limit after [26]. In a sense in this limit the expectation value
becomes part of the matrix model action and the weight is no longer truly Gaussian as in
Eq. (4.1). It would be very interesting to study more general insertions to the matrix model
action and to see if such models permit a large-N BMN limit. In this section we will study
the weight Eq. (4.6) for the Hermite polynomials as an example.
It is convenient to introduce products of resolvent operators such as

1
1
1 1 l
tr
tr J tr J m ,
W z, u tr
(6.1)
=
z J u J
zu
zl u m
l,m=0
in order to generate expectation values of products of traces of powers of matrices.

However, at finite-N only few closed formulas for a small number of traces are known
so far, such as [25]

l

2(N + l + 1) 2(N + 1)
tr J tr J m = lm
(6.2)
,
2(N)
2(N l)
for the Ginibre weight Eq. (4.1). More general multi-point resolvents are known only in
the asymptotic large-N limit so far [27].
552
The products of characteristic polynomials we have calculated in Eq. (2.11) are

generating functions that may be used to extract information about the spectrum. In fact
they generate all the elementary symmetric functions cr of the eigenvalues:
det[v + J ] = v
N

cr v r ,
r=0
N

cr
zi1 zir ,
r = 1, . . . , N.
(6.3)
i1 <<ir

r for the Hermitian conjugate. The
Similarly we have det[u + J ] = u N N
r=0 cr u
elementary symmetric functions cr can be expressed by the power sums
pr
N

zir = tr J r ,
r = 1, . . . , N.
(6.4)
i=1
Namely, the two sets are related by

1
det [Sij ],
cr =
r! 1i,j,r
p
Sij =
if j i,
if j = i 1,
if j < i.
j i+1
j
0
(6.5)
As a first example we can apply Eq. (2.11) in Theorem 1 for K = L using the finite-N
result Eq. (4.3) for the Gaussian weight:

N

1 j
v u
det[v J ] det u J N = hN N+1 v, u = N!
j!
j =0
N

cl cm

N
Nm
(v)Nl u
.
(6.6)
l,m=0
Comparing coefficients we obtain

cl cm

N
= lm
N!
.
(N l)!
In the large-N limit it yields

1

1 l
l(l 1)
+
cl cl N = N l 1
3(l 1)2 l 1 2 +
2N
12 2
N

2h
l
= Nl 1 +
ah h + ,
N
(6.7)
(6.8)
h=1
and consequently an expansion in l 2 /N exists. We note that the expansion goes in powers
of 1/N instead of the usual genus expansion in powers of 1/N 2 . This is due to the fact
that
expectation values of all combinations pi1 pir pj1 pjr N with
contains
r Eq. (6.7)
r
i
=
j
=
r. So far such expectation values were not known explicitly for
s
s
s=1
s=1
finite-N in a closed form. If we could manage to disentangle all expectation values of
553
pr and p r alone, the individual terms would all have a genus expansion in 1/N 2 , with
different prefactors in N , however. The fact that the BMN limit of Eq. (6.7) exists is quite
remarkable. This holds for the Ginibre ensemble. We wish to mention however, that for a
more general weight function the expansion is generically in powers of 1/N . For example,
the next to leading order of the free energy in 1/N can be calculated [24].
The difficulty to extract formulas like Eq. (6.2) from Eq. (6.7) is that each elementary
symmetric functions cr depends on all lower powers p1 , . . . , pr through
the determinant in
Eq. (6.5). The fact that the cr can be constructed recursively, ncn = nr=1 (1)r1 pr cnr ,
is also not enough to extract each single term. We have, therefore, not been able to derive
new finite-N expectation values of traces alone, that would generalize Eq. (6.2). However,
we have explicitly checked in few cases that Eq. (6.7) leads to Eq. (6.2).
We will now address the question whether the existence of the BMN limit is a generic
property of complex matrix models or if it is specific to the Ginibre ensemble. Therefore,
we look at the ensemble with weight Eq. (4.6) of Hermite polynomials. In this model
single powers of traces tr J l do not vanish any longer for = 0. The simplest non-trivial
characteristic polynomial is given by

det[v J ] N
= PN (v) =

N
v
2
,
HN
2
2
(6.9)
where we have used Eq. (4.7). In order to read off the coefficients from Eq. (6.3) by
expanding the right-hand side we distinguish between even and odd N . For N = 2n we
obtain

n

det[v J ] 2n =
c2n2l 2n (v)2l
l=0

n

n+l
nl n 2(n + 1/2) 2l
v ,
=
(1) (2 )
l 2(l + 1/2)
(6.10)
l=0
where we have used the fact that our weight is even. Repeating the same analysis for
N = 2n + 1 and using an identity for the Gamma function we arrive at

N!
l
c2l N =
(6.11)
,
c2l+1 N = 0,
2 l!(N 2l)!
which holds for any N . Looking at Eq. (6.7) we can see that again the limit l, N with
l 2 /N fixed exists. Such a large-N limit thus seems to be a generic property of complex
matrix models. We note however, that the expectation value Eq. (6.11) has to be properly
normalized since both l and 1/ l! vanish for 0 < < 1 at large-l.
7. Conclusions
We have studied products of characteristic polynomials of complex matrices and
their Hermitian conjugate. For Hermitian matrix models several equivalent determinant
formulas hold for characteristic polynomials. They contain orthogonal polynomials and
554
their kernel, which are related through the ChristoffelDarboux formula. For complex
matrices this is no longer the case. The number of characteristic polynomials and their
conjugate completely fixes the result which in general is a determinant of both the
orthogonal polynomials in the complex plane and their respective kernel. If we only
consider characteristic polynomials of complex matrices without their conjugate our result
reduces to a determinant over polynomials only, which is identical to the result of the
Hermitian matrix model. This does not come as a surprise since our proof only uses
the existence of orthogonal polynomials in the complex plane and holds despite of the
breakdown of the three-step recursion relation and the ChristoffelDarboux identity.
Next we gave an interpretation of our results in terms of orthogonal polynomials in
the complex plane, with weights including determinants, and their corresponding kernels.
We have provided explicit examples for complex polynomials at finite-N such as Hermite,
Laguerre or Chebyshev polynomials in the complex plane. From them we have constructed
orthogonal polynomials for a class of more general weights.
At finite-N we proved a duality relation for the weight with Hermite polynomials
in the complex plane. It relates expectation values of different powers of characteristic
polynomials with respect to a different matrix size, generalizing a known duality for the
Gaussian Hermitian matrix model.
We have also studied the microscopic large-N limit of characteristic polynomials
when the arguments are rescaled with the matrix size N . In the weak non-hermiticity
limit all products of characteristic polynomials are universal for a large class of weight
functions. The strongly non-Hermitian large-N limit is smooth only for products with an
equal number of characteristic polynomials and their Hermitian conjugate. This is again
a peculiar implication of the breakdown of the ChristoffelDarboux identity. Although in
general the orthogonal polynomials do no longer have a smooth asymptotic limit, their
kernel still can and does have a smooth limit. The issue of universality of the large-N
kernel at strong non-hermiticity remains open at present.
Furthermore we have argued that the existence of the BMN large-N limit is a generic
phenomenon for complex matrix models. In using the fact that characteristic polynomials
generate the elementary symmetric functions we have shown that for quite general
expectation values such a limit exists. In principle from our finite-N results one can
compute iteratively expectation values of single powers of traces alone. However, it would
be more useful to find a closed expression for that, if any.
It would be also very interesting to extend our analysis to include negative powers of
moments in the expectation values. We suspect that a result similar to that of Hermitian
matrices holds, where the Cauchy transformation of the orthogonal polynomials occurs.
This would provide us with an alternative way to evaluate complex eigenvalue correlations
at finite-N and infinite-N .
Acknowledgements
We wish to thank A. DAdda, F. David and M.L. Mehta for useful discussions and
P. Forrester for correspondence. This work was supported by the European network on
555
Discrete Random Geometries HPRN-CT-1999-00161 EUROGRID and by a Heisenberg

fellowship of the Deutsche Forschungsgemeinschaft.
References
[1] A.V. Andreev, B.D. Simons, Phys. Rev. Lett. 75 (1995) 2304.
[2] E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 (1993) 306, hep-th/9212088;
J. Verbaarschot, Phys. Rev. Lett. 72 (1994) 2531, hep-th/9401059.
[3] J.P. Keating, N. Snaith, Commun. Math. Phys. 214 (2000) 57.
[4] T. Guhr, A. Mller-Groeling, H.A. Weidenmller, Phys. Rep. 299 (1998) 190, cond-mat/9707301.
[5] Y.V. Fyodorov, Nucl. Phys. B 621 (2002) 643, math-ph/0106006;
Y.V. Fyodorov, E. Strahov, Nucl. Phys. B 630 (2002) 453, math-ph/0201045.
[6] E. Kanzieper, Phys. Rev. Lett. 89 (2002) 250201, cond-mat/0207745.
[7] K. Splittorff, J.J.M. Verbaarschot, Phys. Rev. Lett. 90 (2003) 041601, cond-mat/0209594.
[8] Y.V. Fyodorov, E. Strahov, Nucl. Phys. B 647 (2002) 581, hep-th/0205215;
Y.V. Fyodorov, G. Akemann, Comment on the article Replica limit of the Toda lattice equation by
K. Splittorff, J.J.M. Verbaarschot, cond-mat/0210647.
[9] Y.V. Fyodorov, E. Strahov, An exact formula for general spectral correlation function of random Hermitian
matrices, math-ph/0204051.
[10] E. Strahov, Y.V. Fyodorov, Universal results for correlations of characteristic polynomials: RiemannHilbert
approach, math-ph/0210010.
[11] M.L. Mehta, J.-M. Normand, J. Phys. A: Math. Gen. 34 (2001) 1, cond-mat/0101469.
[12] P.J. Forrester, N.S. Witte, Commun. Math. Phys. 219 (2001) 357, math-ph/0103025.
[13] P.H. Damgaard, Phys. Lett. B 424 (1998) 322, hep-th/9711110;
G. Akemann, P.H. Damgaard, Nucl. Phys. B 528 (1998) 411, hep-th/9801133;
G. Akemann, P.H. Damgaard, Phys. Lett. B 432 (1998) 390, hep-th/9802174.
[14] P.H. Damgaard, S.M. Nishigaki, Phys. Rev. D 57 (1998) 52995302, hep-th/9711096.
[15] T. Wilke, T. Guhr, T. Wettig, Phys. Rev. D 57 (1998) 6486, hep-th/9711057.
[16] P.H. Damgaard, in: Proceedings of Quantum Chromodynamics 98, Paris, 1998, pp. 296303, hepth/9807026.
[17] T. Nagao, S.M. Nishigaki, Phys. Rev. D 62 (2000) 065006, hep-th/0001137.
[18] G. Akemann, E. Kanzieper, Phys. Rev. Lett. 85 (2000) 1174, hep-th/0001188.
[19] T. Nagao, S.M. Nishigaki, Phys. Rev. D 63 (2001) 045011, hep-th/0005077.
[20] E. Brzin, S. Hikami, Commun. Math. Phys. 214 (2000) 111, math-ph/9910005;
E. Brzin, S. Hikami, Commun. Math. Phys. 223 (2001) 363, math-ph/0103012.
[21] J. Ginibre, J. Math. Phys. 6 (1965) 440.
[22] P.J. Forrester, Phys. Rep. 301 (1998) 325.
[23] O. Agam, E. Bettelheim, P. Wiegmann, A. Zabrodin, Phys. Rev. Lett. 88 (2002) 236802, cond-mat/0111333.
[24] P. Wiegmann, A. Zabrodin, Large scale correlations in normal and general non-Hermitian matrix ensembles,
hep-th/0210159;
A. Zabrodin, New applications of non-Hermitian random matrices, cond-mat/0210331.
[25] C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, Nucl. Phys. B 643 (2002) 3, hep-th/0205033;
N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, JHEP 0207
(2002) 017, hep-th/0205089.
[26] D. Berenstein, J. Maldacena, H. Nastase, JHEP 0204 (2002) 013, hep-th/0202021.
[27] B. Eynard, C. Kristjansen, JHEP 0210 (2002) 027, hep-th/0209244.
[28] R. de Mello Koch, A. Jevicki, J.P. Rodrigues, Collective string field theory of matrix models in the BMN
limit, hep-th/0209155.
[29] M.A. Stephanov, Phys. Rev. Lett. 76 (1996) 4472, hep-th/9604003.
[30] G. Akemann, Phys. Rev. Lett. 89 (2002) 072002, hep-th/0204068;
G. Akemann, The solution of a chiral random matrix model with complex eigenvalues, J. Phys. A, special
issue, in press, hep-th/0204246.
556
[31] G. Akemann, T. Wettig, Matrix model correlation functions and lattice data for the QCD Dirac operator with
chemical potential, hep-lat/0301017.
[32] G. Akemann, Phys. Rev. D 64 (2001) 114021, hep-th/0106053.
[33] G. Akemann, Phys. Lett. B 547 (2002) 100, hep-th/0206086.
[34] H. Stahl, V. Totik, General Orthogonal Polynomials, Cambridge Univ. Press, London, 1992.
[35] M.L. Mehta, Random Matrices, 2nd Edition, Academic Press, London, 1991.
[36] G. Szeg, Orthogonal Polynomials, in: American Mathematical Society, Colloquium Publications, Vol. 23,
American Mathematical Society, Providence, 1975.
[37] F. Abild-Pedersen, G. Vernizzi, On the microscopic spectra of the massive Dirac operator for chiral
orthogonal and chiral symplectic ensembles, hep-th/0104028.
[38] P. Zinn-Justin, Commun. Math. Phys. 194 (1998) 631, cond-mat/9705044.
[39] G. Akemann, P.H. Damgaard, Nucl. Phys. B 576 (2000) 597, hep-th/9910190.
[40] H.W. Braden, A. Mironov, A. Morozov, Phys. Lett. B 514 (2001) 293, hep-th/0105169.
[41] P.J. Forrester, T. Nagao, J. Phys. A 34 (2001) 7917.
[42] Y.V. Fyodorov, B.A. Khoruzhenko, H.-J. Sommers, Phys. Lett. A 226 (1997) 46, cond-mat/9606173;
Y.V. Fyodorov, B.A. Khoruzhenko, H.-J. Sommers, Phys. Rev. Lett. 79 (1997) 557, cond-mat/9703152.
[43] P. Di Francesco, M. Gaudin, C. Itzykson, F. Lesage, Int. J. Mod. Phys. A 9 (1994) 4257, hep-th/9401163.
[44] P.J. Forrester, N.S. Witte, math-ph/0201051;
P.J. Forrester, N.S. Witte, math-ph/0204008.
[45] J. Ambjrn, J. Jurkiewicz, Y. Makeenko, Phys. Lett. B 251 (1990) 517;
J. Ambjrn, L. Chekhov, C.F. Kristjansen, Yu. Makeenko, Nucl. Phys. B 404 (1993) 127, hep-th/9302014.

Dynamical correlations for circular ensembles

of random matrices
Taro Nagao a , Peter J. Forrester b
a Department of Physics, Graduate School of Science, Osaka University, Toyonaka, Osaka 560-0043, Japan
b Department of Mathematics and Statistics, University of Melbourne, Victoria 3010, Australia
Received 23 August 2002; accepted 7 April 2003
Abstract
Circular Brownian motion models of random matrices were introduced by Dyson and describe the
parametric eigenparameter correlations of unitary random matrices. For symmetric unitary, self-dual
quaternion unitary and an analogue of antisymmetric Hermitian matrix initial conditions, Brownian
dynamics toward the unitary symmetry is analyzed. The dynamical correlation functions of arbitrary
number of Brownian particles at arbitrary number of times are shown to be written in the forms of
quaternion determinants, similarly as in the case of Hermitian random matrix models.
PACS: 05.20.-y; 05.40.-a; 02.50.Ey
Keywords: Random matrix; FokkerPlanck equation; Quaternion determinant
1. Introduction
Circular ensembles of random matrices were introduced by Dyson as the simplest
possible models of complex energy spectra [1,2]. Dyson further introduced a Brownian
motion model to describe the parametric correlation of the energy levels [3]. His Brownian
dynamics is specified by the FokkerPlanck equation
p
= Lp,

N

W
1
L=
+
,
j j
j
j =1
E-mail address: nagao@sphinx.phys.sci.osaka-u.ac.jp (T. Nagao).

doi:10.1016/S0550-3213(03)00292-X
(1.1)
558
T. Nagao, P.J. Forrester / Nuclear Physics B 660 [FS] (2003) 557578
where 1 , 2 , . . . , N are the locations of Brownian particles, is the time variable and
W =
N

logej el .
(1.2)
j <l
The stationary distribution of this Brownian dynamics is

pS = eW =
N

e j e l ,
(1.3)
j <l
satisfying pS / = 0. Using a perturbation theory, Dyson showed that it is a natural

model to describe the parametric eigenparameter correlation of symmetric unitary, unitary
and self-dual quaternion unitary random matrices, corresponding to = 1, 2, and 4,
respectively.
In order to solve the FokkerPlanck equation, a transformation of the FokkerPlanck
operator L into a Hermitian operator H is useful. The Hermitian operator H is defined as
1
eW/2 LeW/2 = (H E0 ),
(1.4)
with a constant E0 and explicitly written as

H=
N
N

2
1
(/2 1)
.
+
2
2
4
j
sin [(j l )/2]
j <l
j =1
(1.5)
It is known as the CalogeroSutherland Hamiltonian on the unit circle. This Hamiltonian

has a complete set of orthogonal eigenfunctions
(1 , . . . , N ) = eW/2 P(2/) (z1 , . . . , zN ),
zj = eij ,
(1.6)
with eigenvalues E [46]. Here = (1 , . . . , N ) represents a partition with non-negative

(2/)
(z1 , . . . , zN ) is a particular polynomial known as
integers 1 2 N and P
the Jack polynomial.
For the imaginary time Schrdinger equation
1
= (H E0 ),
(1.7)
the Green function solution can be written as

G(H ) (1 , . . . , N ; 1 , . . . , N ; )
(1 , . . . , N ) (1 , . . . , N )
e(E E0 )/ ,
=

|

(1.8)
where

| =

d1
2

dN (1 , . . . , N ) ,
(1.9)
559
and is the complex conjugate of . According to (1.4), the Green function solution
G(1 , . . . , N ; 1, . . . , N ; ) of the FokkerPlanck equation (1.1) is given by
G(1 , . . . , N ; 1, . . . , N ; )
=
eW (1 ,...,N )/2 (H )
G (1 , . . . , N ; 1, . . . , N ; ).
eW (1 ,...,N )/2
(1.10)
In this paper we focus on the case = 2, in which the interaction term in (1.5) vanishes,
and calculate the dynamical correlation functions for typical initial conditions. Then G(H )
is a determinant of one particle Green functions and (1.10) reads
G(1 , . . . , N ; 1, . . . , N ; )

N

sin j 2 l
j >l
where (z = ei , w = ei ),

1/2
1 z
g(, ; ) =
sin
j l
2
w n
n ,
n= z e
2 w
1
w
n ,
n= z e
2

det g(j , l ; ) j,l=1,2,...,N ,
N even,
(1.11)
(1.12)
N odd,
and

n =
(n (1/2))2/2,
N even,
n2 /2,
N odd.
(1.13)
We remark that the following discussion does not depend on the particular form (1.13) of
n , provided they are positive.
Let us suppose that the initial distribution of the eigenvalues is one of the followings
(note that U () can be a function with a complex value):
N

ij eil |,
U (j ) N
j
=1
j <l |e
P N/2
j =1 U (j ) (j j +N/2 )
N/2 i
il 4
j
j <l |e e | (assuming N even),

p0 (1 , . . . , N ) [N/2] U ( )2 ( +
(1.14)
j
j
j +[(N+1)/2] )
P
j =1
[N/2]
j <l |eij eil |2 |eij eil |2
1, N even,
[N/2]

([(N+1)/2] ) j =1 |1 eij |2 , N odd.

Here we have introduced a sum P over the permutation of j s in order to make the
initial distribution totally symmetric. [x] is the largest integer not exceeding x. The first and
second of these initial conditions can be derived as eigenvalue distributions of symmetric
unitary and self-dual quaternion unitary random matrices, respectively. The third one, for
which the weight function U () is assumed to be symmetrical about the origin, is an
analogue of the eigenvalue distribution of antisymmetric Hermitian random matrices.
560
The probability distribution function p of the eigenvalues is calculated from the initial
condition and the Green functions as

p 11 , . . . , N2 ; 1 ; 12 , . . . , N2 ; 2 ; . . . ; 1M , . . . , NM ; M
1
=
N!

d10

dN0 p0 10 , . . . , N0
M

G 1l1 , . . . , Nl1 ; 1l , . . . , Nl ; l l1 ,
0 = 0.
(1.15)
l=1
For N even, performing the integration gives

p 11 , . . . , N2 ; 1 ; 12 , . . . , N2 ; 2 ; . . . ; 1M , . . . , NM ; M
=i
N(N1)/2
N

i(N1)jM /2
j =1

Pf Fj11
l j,l=1,2,...,N
N

i M
M
e j eil
j >l
M1

k
det gjk+1
.
l
j,l=1,2,...,N
(1.16)
k=1
Here Pf means a Pfaffian. The matrices g mn and F mn are defined as

m n
gjmn
l = g j , l ; m n ,
(1.17)
m n
Fjmn
l = F j , l ; m , n ,
(1.18)
and
where
F (, ; , )

=

d U ()U ( ) g(, ; )g( , ; ) g( , ; )g(, ; ) ,
F (, ; , )

2

= d U () g(, ; ) g( , ; ) g( , ; ) g(, ; ) ,
F (, ; , )

= d U ()2
0

1
g(, ; )g( , ; ) g( , ; )g(, ; ) ,
2 sin
(1.19)
561
for each of the three initial conditions (1.14), respectively. For N odd, performing the
integration similarly yields

p 11 , . . . , N2 ; 1 ; 12 , . . . , N2 ; 2 ; . . . ; 1M , . . . , NM ; M
=i
N(N1)/2
N

i(N1)jM /2
j =1
11
[Fj l ]j,l=1,2,...,N
Pf
[fl1 ]l=1,...,N
Here
N

i M
M
e j eil
j >l
[fj1 ]j =1,...,N
0
M1

k
det gjk+1
.
l
j,l=1,2,...,N
(1.20)
k=1
fjm = f jm ; m ,
(1.21)
with

f ( ; ) =
U ()g(, ; ) d,
f ( ; ) = U (0)g(, 0; ),
(1.22)
for the first and last of the initial conditions (1.14), respectively.
Now we are in a position to define multilevel dynamical correlation functions

11 , . . . , m1 1 ; 1; 12 , . . . , m2 2 ; 2 ; . . . ; 1M , . . . , mMM ; M

1
(N!)M
1
1
M
=
dm1 +1 dN dmM +1 dNM

CN M
(N
m
)!
l
l=1
1

1
2
2
M
M
p 1 , . . . , N ; 1 ; 1 , . . . , N ; 2 ; . . . ; 1 , . . . , N ; M ,
(1.23)
where the normalization constant CN is defined as

1
1
M
CN = d1 dN d1 dNM

1
1
2
2
1 , . . . , N ; 1 ; 1 , . . . , N ; 2 ; . . . ; 1M , . . . , NM ; M .
(1.24)
Let us review the history of the study on multilevel correlation functions of random
matrices and explain the purpose of this paper. The quaternion determinant formulas for
the static correlations (M = 1, 1 = 0) with U ( ) = 1 was discovered by Dyson [7].
His result was generalized to the case with a real weight function U ( ) by Nagao and
Wadati [8]. The dynamical correlations were first explored for Hermitian random matrix
models by Pandey and Mehta [9,10] and then extended to the unitary matrix case with
U ( ) = 1 by Pandey and Shukla [11]. They were able to derive quaternion determinant
forms for equal time (M = 1, 0 1 < ) correlation functions. Recently Nagao and
Forrester succeeded in evaluating the dynamical correlation functions with general M for
Hermitian random matrices [12,13]. In this paper we deal with unitary random matrices
and generalize both of Nagao and Wadatis, and Pandey and Shuklas results to show how
the multilevel dynamical correlation functions with general M can be written in quaternion
determinant forms.
562
2. Dynamical correlation functions

2.1. Quaternion determinant
Let us begin with an introduction of a quaternion determinant, a determinant of a matrix
with quaternion elements [7,1417]. We define a quaternion as a linear combination of four
basic units {1, e1 , e2 , e3 }:
q = q 0 + q e = q 0 + q 1 e1 + q 2 e2 + q 3 e3 .
(2.1)
Here q0 , q1 , q2 , and q3 are real or complex numbers. The first part q1 is called the scalar
part of q. The quaternion multiplication is associative but in general not commutative: the
multiplication rule of the four basic units are
1 ej = ej 1 = ej ,
1 1 = 1,
j = 1, 2, 3,
e12 = e22 = e32 = e1 e2 e3 = 1.
(2.2)
(2.3)
A quaternion q has a dual q defined as

q = q0 q e.
(2.4)
A matrix Q with quaternion elements qj l also has a dual matrix Q = [qlj ]. The quaternion
units can be represented as 2 2 matrices

0 1
1 0
,
1
,
e1
1 0
0 1

0 i
i 0
,
e3
.
e2
(2.5)
i 0
0 i
Now we are in a position to define a quaternion determinant Tdet. For a self-dual Q
it is written as
(i.e., Q = Q),
Tdet Q =
l

(1)Nl (qab qbc qda )0 .
P
(2.6)
Here P denotes any permutation of the indices (1, 2, . . . , N) consisting of l exclusive

cycles of the form (a b c d a). Note that (1)Nl is the parity of P . The
subscript 0 has a meaning that we take the scalar part of the product over each cycle. If all
the elements of Q are scalars, then everything is commutable and a quaternion determinant
becomes an ordinary determinant.
2.2. The case N even
Throughout this paper we adopt a notation that z is the complex conjugate of z (note
that z is not necessarily the complex conjugate of z). Let us define (z = ei , w = ei )
n ( ; ) = z Rn ( ; ),
563
z Rn ( ; ),

n ( ; ) = F (, ; , ) w Rn (; ) d,
n ( ; ) =

n ( ; ) =
F (, ; , ) w Rn (; ) d,
(2.7)
where Rn ( ; ) = zn + cn n1 zn+1 + + cn n zn and Rn ( ; ) = zn + cn n1 zn1 +

+ cn n zn are arbitrary polynomials (R0 ( ; ) = R0 ( ; ) = 1).
We introduce matrices D mn , I mn , and S mn as
m n

Djmn
l = D j , l ; m , n ,

m n
Sjmn
l = S j , l ; m , n
m n

Ijmn
l = I j , l ; m , n ,
(2.8)
with
D(, ; , )
=
(N/2)1

k=0
1 k ( )
e
k ( ; )k ( ; )
rk ( )

ek+1 ( ) k ( ; )k ( ; ) ,
(2.9)
I (, ; , )
=
(N/2)1

k=0
1 k ( )
e
k ( ; )k ( ; )
rk ( )

ek+1 ( ) k ( ; )k ( ; ) + F (, ; , ),
(2.10)
and
S(, ; , )
=
(N/2)1

k=0
1 k ( )
k ( ; )k ( ; )
e
rk ( )

ek+1 ( ) k ( ; )k ( ; ) .
Moreover we define
mn
Sj l gjmn
l ,
=
Sjmn
mn
l
Sj l ,
m > n,
m n.
Then we have the following theorem.
(2.11)
(2.12)
564
Theorem 1. The probability distribution function (1.16) with N even can be written as a
quaternion determinant

p 11 , . . . , N2 ; 1 ; 12 , . . . , N2 ; 2 ; . . . ; 1M , . . . , NM ; M
= i N(N1)/2
(N/2)1

rj (M ) Tdet B ,
, = 1, 2, . . . , M.
(2.13)
j =0
Each block B is an N N quaternion matrix. Its quaternion elements are represented

as

Sj l
Ij l
.
Bj l =
(2.14)
S
D
jl
lj
Starting from (1.16), we can follow a similar argument as in Refs. [12,13] to prove
Theorem 1.
By Schmidts orthogonalization procedure, we can specify the polynomials Rn ( ; )
and Rn (, ) so that they satisfy the following skew orthogonality relation:

z Rm ( ; ), w Rn (; ) = z Rn ( ; ), w Rm
(; ) = rm ( )mn ,

z Rm ( ; ), w Rn (; ) = 0,
z Rm ( ; ), w Rn (; ) = 0,
(2.15)
where

f ( ), g() =

d F (, ; , )f ( )g().
(2.16)
Explicit determinant formula for the skew orthogonal polynomials Rn ( ; ) and R (, )

are given by (n 1) [8]
n

z
J n n1
J n0
J n 1
J n n
n1

n1
n1
n1
0
n1
1
n1
n
z

J
J
J
J
.

.
.
.
.
..
..
.

.
.
.
.
.
.
.
.
.
.

1 .
Rn (z) = Dn 0
, (2.17)
J 0 n1
J00
J 0 1
J 0 n
z
.

..
..
..
..
..
..
..

.
.
.
.
.
.

zn J n n1 J n 0 J n 1 J n n
and
n
z
n+1
z
.
.
.
1
Rn (z) = Dn 0
z
.
..

zn
J n1 n
J n n
..
.
..
.
J n1 n
..
.
J 1 n
..
.
J n1 0
J n 0
..
.
J 1 0
..
.
J n1 0
J n1 1
J n 1
..
.
..
.
J n1 1
..
.
J 1 1
..
.

J n1 n1

J n n1

..

.

,
J 1 n1

..

.

n1
n1
J
(2.18)
565
where

J mn = zm+(1/2), wn+(1/2) ,
(2.19)
and
J n1 n1

..

0 .n1
Dn = J
..

.
n n1
J
..
.
..
.
J n1 0
..
.
J n 0
J00
..
.
J n1 1
..
.
J 0 1
..
.
J n 1
..
.
..
.
J n1 n

..

.

0
n
.
J

..

.

J n n
(2.20)
2.3. The case N odd

Let us next consider the case with N odd. We define
n ( ; ) =
F (, ; , )Rn (; ) d,

n ( ; ) =
F (, ; , )Rn (; ) d.
(2.21)
Here Rn ( ; ) and Rn ( ; ) are arbitrary functions provided that

N

i

e j eil
j >l
R(N1)/2 (1 ; ) R(N1)/2
(2 ; )

.
.

..
..

R ( ; )
N
R1 (2 ; )

1 1
ei(N1)j /2 R0 (1 ; )
=
R0 (2 ; )
R ( ; )
j =1
R1 (2 ; )
1 1

..
..

.
.

R
(
;
)
R
(N1)/2 1
(N1)/2 (2 ; )
..
.
..
.
R(N1)/2
(N ; )

..

.
R1 (N ; )
R0 (N ; ) .
R1 (N ; )

..

.

R
( ; )
(N1)/2
(2.22)
Then matrices D mn , I mn , and S mn are introduced as

m n
Djmn
l = D j , l ; m , n ,
m n

Sjmn
l = S j , l ; m , n ,
m n
Ijmn
l = I j , l ; m , n ,
(2.23)
566
where
D(, ; , )
=
(N1)/2

k=1
1 k ( )
Rk ( ; )Rk ( ; )
e
rk ( )
ek (
)

Rk ( ; )Rk ( ; ) ,
(2.24)
I (, ; , )
=
(N1)/2

k=1
1 k ( )
e
k ( ; )k ( ; )
rk ( )

ek ( ) k ( ; )k ( ; )
+
1
1
0 ( ; )f ( ; )
0 ( ; )f ( ; ) + F (, ; , ),
s0 ( )
s0 ( )
(2.25)
and
S(, ; , )
=
(N1)/2

k=1

1 k ( )

k ( ; )Rk ( ; ) ek ( ) k ( ; )Rk ( ; )
e
rk ( )
1
R0 ( , )f ( ; ).
s0 ( )
As before we set
mn
Sj l gjmn
l ,
Sjmn
=
mn
l
Sj l ,
(2.26)
m > n,
m n,
(2.27)
and find the following theorem.

Theorem 2. The probability distribution function (1.20) with N odd can be rewritten as

p 11 , . . . , N2 ; 1 ; 12 , . . . , N2 ; 2 ; . . . ; 1M , . . . , NM ; M
=i
N(N1)/2
s0 (M )
(N1)/2

rj (M ) Tdet B ,
, = 1, 2, . . . , M.
(2.28)
j =1
Each block B is an N N quaternion matrix the elements of which are represented as

Ij l
Sj l
Bj l =
(2.29)
.
S
D
jl
lj
Theorem 2 can be proven from (1.20) by proceeding as in Refs. [12,13].
567
We now introduce a set of polynomials n ( ; ), n ( ; ) for n 1 and 0 ( ; ) =

satisfying

(; ) = rm ( )mn ,
m ( ; ), n (; ) = n ( ; ), m

m ( ; ), n (; ) = 0,
m ( ; ), n(; ) = 0,
(2.30)
0 ( ; )
for m, n 1 and

n ( ; ), 0(; ) = 0 ( ; ), n(; ) = 0,

0 ( ; ), n (; ) = n ( ; ), 0(; ) = 0,
(2.31)
for 0 n (N 1)/2. Here the bracket is defined in (2.16).

Let us define
m
z , n (; ), n > 0,
K mn =
zm , n (; ), n < 0,
(2.32)
and
m ( ; ), n (; ),
m ( ; ), n (; ),
mn
L =
( ; ), (; ),
n
m
( ; ), (; ),
m
n
m, n > 0,
m > 0, n < 0,
m < 0, n > 0,
m, n < 0.
(2.33)
Then, starting from (z = ei )

1 ( ; ) = a1 + z,
1
1 ( ; ) = a1 + ,
z
(2.34)
we can recursively construct the polynomials as (n 2)

n ( ; )
(n1)
( ; )

K n n1
zn

( ; ) Ln1 n1
n1

..
..

.
.

1
L1 n1
+ Dn1
1 ( ; )
( ; )
L1 n1
1

..
..

.
.

n1 ( ; ) Ln+1 n1
= a n 0
and
n ( ; )
= an 0(n1) ( ; )
..
.
Kn1
Ln1 1
..
.
..
.
L1 1
L1 1
..
.
Ln+1 1
K n 1
Ln1 1
..
.
L1 1
L1 1
..
.
Ln+1 1
..
.
..
.

1
n+1
L

1
n+1
L

..

.

Ln+1 n+1
(2.35)
K n n+1
Ln1 n+1
..
.
568

n

K n n+1
z
n1 ( ; ) Ln+1 n+1

..
..

.
.

1
( ; )
1 n+1
+ Dn1
L
1
( ; )
L1 n+1
1

.
..
..

.

( ; ) Ln1 n+1
n1
..
.
..
.
K n 1
Ln+1 1
..
.
L1 1
L1 1
..
.
Ln1 1
K n 1
Ln+1 1
..
.
L1 1
L1 1
..
.
..
.
..
.
Ln1 1

K n n1
n+1
n1

L

..

.

1
n1
.
L

L1 n1

..

.

n1
n1
L
(2.36)
Here
0(n1) ( ; )

1
K 0 n1

( ; ) Ln1 n1
n1

..
..

.
.

1 ( ; )
L1 n1
= Dn1
1

L1 n1
1 ( ; )

..
..

.
.

n1 ( ; ) Ln+1 n1
..
.
K0 1
Ln1 1
..
.
..
.
L1 1
L1 1
..
.
Ln+1 1
K 0 1
Ln1 1
..
.
L1 1
L1 1
..
.
Ln+1 1
..
.
..
.

1
n+1
L

1
n+1
L

..

.

Ln+1 n+1
(2.37)
K 0 n+1
Ln1 n+1
..
.
and
Ln1 n1

..

.

L1 n1
Dn1 = 1 n1
L .

..

n+1 n1
L
..
.
..
.
Ln1 1
..
.
L1 1
L1 1
..
.
Ln+1 1
Ln1 1
..
.
L1 1
L1 1
..
.
Ln+1 1
..
.
..
.

1
n+1

L
.
1
n+1

L

..

.

Ln+1 n+1
Ln1 n+1
..
.
(2.38)
Setting
((N1)/2)
0 ( ; ) = 0
( ; ),
(2.39)
we find that all of the skew orthogonality conditions (2.30) and (2.31) are satisfied. Note
that there is an ambiguity in the determination of the skew orthogonal polynomials due to
the constants an and an .
Let us now introduce (n 0)
sn ( ) =
d f ( ; )n( ; ),
sn ( ) =
d f ( ; )n ( ; ).
(2.40)
569
Then Rn ( ; ) and R ( ; ) satisfying (2.22) can be constructed as (n = 1, 2, . . . , (N

1)/2)
sn ( )
0 ( ; ),
s0 ( )
sn ( )
0 ( ; ),
Rn ( ; ) = n ( ; )
s0 ( )
Rn ( ; ) = n ( ; )
(2.41)
and
R0 ( ; ) = R0 ( ; ) = 0 ( ; ).
(2.42)
2.4. Quaternion determinant expressions

Using the orthogonality relations introduced above, it can be readily proven in both the
cases N even and odd that

mp
pn
Bj i Bil di
0, m < p, p > n,
B mn 1 00 , m < p, p = n,
0j l0 0 mn
B l , m = p, p > n,
00 10 jmn
1 0
Bj l + Bjmn
, m = p, p = n,
0
1
0 0 mn
0 0 1 0 mn

mn 1 0
mn 0 0
= 0 1 Bj l + Bj l 0 0 0 0 Bj l Bj l 0 1 , m > p, p < n,
00
10
mn 1 0
mn
B mn
m > p, p = n,
l + Bj l 0 0 0 0 Bj l ,

00 10 jmn
1
0
0
0
mn
mn
B
,
m
= p, p < n,
+
B
l
jl 0 1
0
00 10 jmn
1j l0 0 mn
Bj l 0 0 Bj l , m > p, p > n,
01
0 0
mn 1 0
m < p, p < n.
Bj l 0 0 Bjmn
l 01 ,
(2.43)
Moreover we can easily find

mm
Bjj
djm = N.
(2.44)
We then arrive at the following theorem which has the central importance in the evaluation
of the dynamical correlations.
Theorem 3. In terms of the quaternion elements Bjmn
l , we define quaternion rectangular
matrices
B mn B mn
11
1L
.. .
..
..
Qmn
(2.45)
.
JL =
.
.
mn
mn
BJ 1 BJ L
570
Let us further define a self-dual quaternion matrix Q consisting of Qmn

J L as
Q11
n1 n1
Q = ...
QK1
nK n1
..
.
Q1K
n1 nK
..
. .
KK
QnK nK
(2.46)
We denote the j th row (column) of the mth row (column) block of the quaternion matrix Q
as (m, j ) row (column). The (m, j ) row and (m, j ) column contain the variable jm . If the
(m, j ) row and (m, j ) column are removed, the resulting smaller matrix Qm
j is still selfdual and does not contain the variable jm .
Using the above definitions, we have

dn Tdet Q = (N n + 1) Tdet Q
n .
(2.47)
The proof of Theorem 3 is found by following the strategy given in Ref. [12].
Successive application of Theorem 3 leads to the quaternion determinant expression for
the multilevel dynamical correlation function as

11 , . . . , m1 1 ; 1; 12 , . . . , m2 2 ; 2 ; . . . ; 1M , . . . , mMM ; M = Tdet B (m , m ) ,
, = 1, 2, . . . , M,
(2.48)
where each block B (m , m ) is obtained by removing the m + 1, m + 2, . . . , N th

rows and m + 1, m + 2, . . . , N th columns from B .
3. Dynamical correlation within the unitary symmetry

In this section we consider the limit j , j = 1, 2, . . . , M, with all j l ,
j, l = 1, 2, . . . , M, fixed. In this limit it is expected that the dynamical correlations describe
the Brownian dynamics within the unitary symmetry. We can conveniently take this limit
by using the following summation formulas.
Let us express the skew orthogonal polynomials with = 0 as (z = ei )
Rn ( ; 0) =
n
nj zj ,
nn = 1,
j
nj
z ,
nn
= 1.
j =n
Rn ( ; 0) =
n

j =n
(3.1)
571
Then it can be seen that

Rn ( ; ) = en+1
Rn ( ; ) = en
n
nj zj ej+1 ,
j =n
n
j j
nj
z e
,
(3.2)
j =n
with
rn ( ) = en en+1 rn (0),
(3.3)
the proof of which comes from an identity

z Rm ( ; ), w Rn (; ) = em en+1 z Rm
( ; 0), wRn (; 0) =0 .
(3.4)
Let us introduce an inverse expansion of (3.2) (n 1)
zn en+1 =
n

j =0
zn en =
1
nj ej Rj ( ; ),
z
n1
nj ej+1 Rj ( ; ) +
n

j =0
j =0
nj ej Rj ( ; ) + z
n1
nj ej+1 Rj ( ; ).
(3.5)
j =0
Using (3.9) and (3.5), we can readily derive (jj = 1, n 0)

en
(1/2) +1
+(1/2)
n ( ; ) =
rn (0)
z
e
n +
z
e
n ,
2
n
n+1

n+1
e
n ( ; ) =
rn (0)
z(1/2)e+1 n +
z+(1/2)e n .
2
n+1
n
(3.6)
Putting (3.5) into (1.19) and comparing the result with (3.6) lead to
F (, ; , )

1 k ( )
=
k ( ; )k ( ; )
e
rk ( )
k=0

ek+1 ( ) k ( ; )k ( ; ) ,
(3.7)
so that
I (, ; , )

1 k ( )
=
k ( ; )k ( ; )
e
rk ( )
k=N/2

ek+1 ( ) k ( ; )k ( ; ) .
(3.8)
572
Substitution of (3.6) into (2.11) results in

S(, ; , ) = S1 (, ; , ) + S2 (, ; , ),
(3.9)
i
where (z = ei , z = e )
1
S1 (, ; , ) =
2

1/2
z
z
N/2

n=(N/2)+1
n
z

en ( ) ,
z
(3.10)
and
S2 (, ; , )
1
=
2
(N/2)1

k=0

=N/2
z(1/2)e+1 k + z+(1/2)e k

ek+1 z Rk ( ; )

+ z(1/2)e+1 k + z+(1/2)e k

ek z Rk ( ; ) .
(3.11)
Let us assume that n = n+1 and n+1 > n for n 1. This assumption is consistent
with (1.13). In the asymptotic limit j , j = 1, 2, . . . , M, with all j l , j, l =
1, 2, . . . , M, fixed, it can be seen from (3.9) that

S(x, y; , ) S1 (x, y; , ) + O e((N/2)+1 N/2 ) .

(3.12)
We can further derive the asymptotic relations Rn (x; ) O(1), Rn (x; ) O(1),
n (x; ) O(e2n+1 ) and n (x; ) O(e2n+1 ) from (3.2) and (3.6) so that

I (x, y; , ) O e(N/2)+1 ( + ) ,
(3.13)
D(x, y; , ) O eN/2 ( + ) .
Therefore in the asymptotic limit we obtain a quaternion determinant in which Sjmn
l are
replaced by

S1 (jm , ln ; m , n ) gjmn
l , m > n,
mn
j l =
(3.14)
m n,
S1 (jm , ln ; m , n ),
and the other elements are set to zero. Then the quaternion determinant becomes an
ordinary determinant

1
2
M
; 1 ; x12, . . . , xm
; 2 ; . . . ; x1M , . . . , xm
; M
x11 , . . . , xm
M
1
2
11

12 (m1 , m2 ) 1M (m1 , mM )
(m1 , m1 )
21

22
2M
(m2 , m2 ) (m2 , mM )
(m2 , m1 )
,
=
(3.15)
..
..
.
..

..
.
.
.

M1

(mM , m1 ) M2 (mM , m2 ) MM (mM , mM )
where each block (m , m ) is obtained by removing the m +1, m +2, . . . , N th rows
and m + 1, m + 2, . . . , N th columns from . This is the unitary symmetry version of
a similar determinant formula [18,19] known for the Brownian dynamics within Hermitian
symmetry.
573
3.2. The case N odd

In the case with odd N , we introduce expansions (n 1, z = ei ),
n
n ( ; ) = en
n ( ; ) = e
nj zj ej ,
j =n+1
n

n
j j
nj
z e
,
(3.16)
j =n+1
= 1 and
with nn = nn
(N1)/2
0 ( ; ) = e0
0j zj ej .
(3.17)
j =(N1)/2
As before it can be easily proven that

rn ( ) = en en rn (0),
(3.18)
sn ( ) = en sn (0),
(3.19)
and
sn ( ) = en sn (0).
Let us now define

n ( ; ) =
F (, ; , )n (; ) d

0 n (N 1)/2,
n (; 0), 0( ; 0)| =0 ( ; ),
en
2

n ( ; ) =
0,
n (N + 1)/2,
F (, ; , )n (; ) d
0,
en
2
0 n (N 1)/2,
n (; 0), 0( ; 0)| =0 ( ; ),
n (N + 1)/2,
(3.20)
where
( ; ) =
zl el l0 +
l=0
zl el l0 .
(3.21)
l=1
The inverse expansions (n 1)

z n e n =
n
nj ej j ( ; ) +
j =0
n n
n

j =0
n
nj ej j ( ; ),
j =1
nj ej j ( ; ) +
n

j =1
nj ej j ( ; ),
(3.22)
574
and
z 0 e 0 =
(N1)/2
0j ej j ( ; ) +
j =0
(N1)/2
0j ej j ( ; ),
(3.23)
j =1
where
n = n + 1 = n,
if n > (N 1)/2,
n = n = (N 1)/2,
if n (N 1)/2,
(3.24)
lead to
n ( ; )

(0) l l
l l ],
[ l=0 z e
ln +
en rn2
l=1 z e
ln
=

(0)
l el +
l el ],
en rn2
[
z
z
ln
l=n
l=n+1
ln
n ( ; )

(0) l l
l l ],
[ l=0 z e
ln +
en rn2
l=1 z e
ln
=

rn (0)
l l
n
l
e
ln + l=n z e
ln ],
2 [ l=n+1 z e
1n
n
N1
2 ,
N+1
2 ,
1n
n
N1
2 ,
N+1
2 .
Using the formula (3.25) for n ( ; ) and n ( ; ), we can readily derive

f ( ; ) =

n=1

1
1
( ; )sn ( ) n ( ; )sn ( ) +
s0 (0) ( ; ).
rn ( ) n
2
(3.25)
(3.26)
Substituting (3.25) and (3.26) into (2.26) yields

S(, ; , ) = S1 (, ; , ) + S2 (, ; , ),
where
(z = ei , z
(3.27)

= ei )
1
2
S1 (, ; , ) =
(N1)/2

n=(N1)/2
n
z

en ( ) ,
z
(3.28)
and
S2 (, ; , )
=
1
2
+
+
(N1)/2
k=0
=(N+1)/2
0 ( ; )
s0 ( )
(N1)/2
( ; )
0
s0 ( )
k=1

z e k + z e k ek k ( ; )

+ z e k + z e k ek k ( ; )

1
k ( ; )sk ( ) k ( ; )sk ( )

rk ( )
k=(N+1)/2

1
k ( ; )sk ( ) k ( ; )sk ( ) ,
rk ( )
(3.29)
575
where we set 0 = 0 = 0. Finally we put (3.22) and (3.23) into (1.19) to find
I (, ; , )

=
k=(N+1)/2
1 k ( )
e
k ( ; )k ( ; )
rk ( )
ek (
0 ( ; )
s0 ( )

k=(N+1)/2
k ( ; )k ( ; )

1
k ( ; )sk ( ) k ( ; )sk ( )

rk ( )
( ; )
0
s0 ( )
)
k=(N+1)/2

1
k ( ; )sk ( ) k ( ; )sk ( ) .
rk ( )
(3.30)
Assuming that n = n and n+1 > n for n 0, we can now take the asymptotic limit
j , j = 1, 2, . . . , M, with all j l , j, l = 1, 2, . . . , M, fixed. Let us first note that
n (x; ) O(1), n (x; ) O(1), n (x; ) O(e2n ), and n (x; ) O(e2n )
for n 1. We can further obtain an estimation

0 ( ; ) O e(0 +(N+1)/2 ) .
0 ( ; ) O e(0 (N1)/2 ) ,
(3.31)
Then it is straightforward to find

S(x, y; , ) S1 (x, y; , ) + O e((N+1)/2 (N1)/2 ) ,

(3.32)

( + )

( + )
(N+1)/2
(N1)/2
I (x, y; , ) O e
,
D(x, y; , ) O e
. (3.33)
mn
Therefore Ijmn
l and Dj l can be set to zero in the asymptotic limit and, as in the case N
even, a determinant expression (3.15) results. Here the matrix elements jmn
l are again
defined by (3.14), although the definitions of S1 (, ; , ) and g(, ; ) are different
from those in the case N even.
4. The case U ( ) = 1
In this section we deal with the simplest case U ( ) = 1 and explicitly give the formulas
for skew orthogonal polynomials. Though this case was already treated by Pandey and
Shukla, our result is more general than theirs because they evaluated only the equal time
correlations.
4.1.1. Symmetric unitary initial condition
For the bracket defined in (2.16) corresponding to the first of the three initial conditions
(1.14), we can easily derive (z = ei )

zm+(1/2), zn+(1/2) =0 =
4i
m+n+1 0 ,
n + (1/2)
(4.1)
576
which yields
Rn ( ; 0) = zn ,
Rn ( ; 0) = zn ,
rn (0) =
4i
.
n + (1/2)
4.1.2. Self-dual quaternion unitary initial condition

Let us next consider the second of the initial conditions (1.14). Noting

m+(1/2) n+(1/2)
1

z
,z
= 4i n +
m+n+1 0 ,
=0
2
we obtain
Rn ( ; 0) = zn ,
Rn ( ; 0) = zn ,

1
rn (0) = 4i n +
.
2
4.1.3. An analogue of antisymmetric Hermitian initial condition

The third of the initial conditions (1.14) yields
i,
if n m is odd and positive,
m+(1/2) n+(1/2)

0,
if n m is even,
,z
=
z
=0
i, if n m is odd and negative,
(4.2)
(4.3)
(4.4)
(4.5)
which implies (n 1)
Rn ( ; 0) = zn + zn ,
Rn ( ; 0) = zn + zn ,
rn (0) = 2i,
(4.6)
and
R0 ( ; 0) = 1,
r0 (0) = i.
(4.7)
4.2. The case N odd

4.2.1. Symmetric unitary initial condition
In N odd case of the symmetric unitary initial condition, we obtain

4i
m+n 0 ,
zm , zn =0 =
n
m, n = 0,
(4.8)
and

4i
(1)n .
zn , 1 =0 =
n
Then, putting an = an = 0, we can derive (n 1)
n ( ; 0) = zn ,
n ( ; 0) = zn ,
rn (0) =
(4.9)
4i
,
n
(4.10)
and
0 ( ; 0) =
(N1)/2

j =(N1)/2
(z)j .
(4.11)
577
4.2.2. An analogue of antisymmetric Hermitian initial condition

An analogue of the antisymmetric Hermitian initial condition is handled by considering
the inner products
i,
if n m is odd and positive,
m n

if n m is even,
z , z =0 = 0,
(4.12)
i, if n m is odd and negative.
We fix the constants as an = an = (1)n+1 and find (n 1)
n ( ; 0) = zn + zn+1 ,
n ( ; 0) = zn + zn1 ,
rn (0) = 2i,
(4.13)
and
0 ( ; 0) = (1)(N1)/2

1
(N1)/2
z
+ z(N1)/2 .
2
(4.14)
5. Conclusion
In this paper we have derived quaternion determinant expressions for the dynamical
multilevel correlation functions for Dysons Brownian motion of eigenparameters (energy
levels) toward unitary symmetry. As the initial conditions, we assume eigenparameter
distributions of symmetric unitary, self-dual quaternion unitary and an analogue of the
antisymmetric Hermitian random matrices. Any of the quaternion elements is represented
in terms of four functions D(, ; , ), I (, ; , ), S(, ; , ), and g(, ; )
appearing in the two level dynamical correlation functions. Therefore, analyzing the two
level dynamical correlations is enough to understand the behavior of all the multilevel
correlations. The two level dynamical correlations are investigated in Refs. [2022] and
some asymptotic results in the limit N are already known.
References
[1] F.J. Dyson, J. Math. Phys. 3 (1962) 140;
F.J. Dyson, J. Math. Phys. 3 (1962) 157;
F.J. Dyson, J. Math. Phys. 3 (1962) 166.
[2] M.L. Mehta, Random Matrices, 2nd Edition, Academic Press, San Diego, 1990.
[3] F.J. Dyson, J. Math. Phys. 3 (1962) 1191.
[4] B. Sutherland, Phys. Rev. A 5 (1972) 1372.
[5] P.J. Forrester, Nucl. Phys. B 416 (1994) 377.
[6] P.J. Forrester, T. Nagao, Nucl. Phys. B 532 (1998) 733.
[7] F.J. Dyson, Commun. Math. Phys. 19 (1970) 235.
[8] T. Nagao, M. Wadati, J. Phys. Soc. Jpn. 61 (1992) 1903.
[9] A. Pandey, M.L. Mehta, Commun. Math. Phys. 87 (1983) 449.
[10] M.L. Mehta, A. Pandey, J. Phys. A 16 (1983) 2655.
[11] A. Pandey, P. Shukla, J. Phys. A 24 (1991) 3907.
[12] T. Nagao, P. J Forrester, Nucl. Phys. B 563 (1999) 547.
[13] T. Nagao, Nucl. Phys. B 602 (2001) 622.
[14] E.H. Moore, Bull. Am. Math. Soc. 28 (1922) 161.
578
[15] E.H. Moore, R.W. Barnard, General Analysis, Memoirs of the American Philosophical Society, American
Philosophical Society, Philadelphia, PA, 1935.
[16] F.J. Dyson, Helv. Phys. Acta 45 (1972) 289.
[17] M.L. Mehta, Matrix Theory, Les Editions de Physique, 1989.
[18] B. Eynard, M.L. Mehta, J. Phys. A 31 (1998) 4449.
[19] T. Nagao, P.J. Forrester, Phys. Lett. A 247 (1998) 42.
[20] H. Spohn, Interacting Brownian particles: a study of Dysons model, in: G. Papanicolaou (Ed.), Proceedings
of the Workshop at the Institute of Mathematics, Minneapolis, MN, March 1986, J. Stat. Phys. 47 (1987)
669.
[21] P.J. Forrester, Physica A 223 (1996) 365.
[22] P.J. Forrester, T. Nagao, J. Stat. Phys. 89 (1997) 69.

Excited TBA equations II: massless flow from

tricritical to critical Ising model
Paul A. Pearce a , Leung Chim a,1 , Changrim Ahn b
a Department of Mathematics and Statistics, University of Melbourne, Parkville,
Victoria 3010, Australia

b Department of Physics, Ewha Womans University, Seoul 120-750, South Korea
Abstract
We consider the massless tricritical Ising model M(4, 5) perturbed by the thermal operator
1,3 in a cylindrical geometry and apply integrable boundary conditions, labelled by the Kac
labels (r, s), that are natural off-critical perturbations of known conformal boundary conditions.
We derive massless thermodynamic Bethe ansatz (TBA) equations for all excitations by solving,
in the continuum scaling limit, the TBA functional equation satisfied by the double-row transfer
matrices of the A4 lattice model of Andrews, Baxter and Forrester (ABF) in Regime IV. The resulting
TBA equations describe the massless renormalization group flow from the tricritical to critical Ising
model. As in the massive case of Part I, the excitations are completely classified in terms of (m, n)
systems but the string content changes by one of three mechanisms along the flow. Using generalized
q-Vandermonde identities, we show that this leads to a flow from tricritical to critical Ising characters.
The excited TBA equations are solved numerically to follow the continuous flows from the UV to
the IR conformal fixed points.
PACS: 05.50.+q; 11.10.-z
1. Introduction
In integrable Quantum Field Theory (QFT), the Thermodynamic Bethe Ansatz (TBA)
[1,2] continues to be an important method for the study of massive and massless
E-mail addresses: p.pearce@ms.unimelb.edu.au (P.A. Pearce), leung.chim@dsto.defence.gov.au (L. Chim),
ahn@dante.ewha.ac.kr (C. Ahn).
1 Current address: DSTO, Adelaide.
doi:10.1016/S0550-3213(03)00254-2
580
P.A. Pearce et al. / Nuclear Physics B 660 [FS] (2003) 579606
Renormalization Group (RG) flows including the study of excited states [35]. Moreover
the Tricritical Ising Model (TIM), which is the simplest member M(4, 5) of the unitary
minimal series [6] beyond the critical Ising model M(3, 4), remains a rich example for
studying both thermal and boundary flows [710].
In Part I of this series (hereafter referred to as PCAI [11]) we considered the massive
tricritical Ising model M(4, 5) perturbed by the thermal operator 1,3 in a cylindrical
geometry and systematically derived the TBA equations for all excitations by using a lattice
approach. More specifically this was achieved by solving, in the continuum scaling limit,
the TBA functional equation satisfied by the double-row transfer matrices of the A4 lattice
model of Andrews, Baxter and Forrester (ABF) in Regime III [12]. In this paper we turn
our attention to the massless tricritical Ising model which is obtained as the continuum
scaling limit of the A4 lattice model in Regime IV. Our goal is to systematically derive
and study the massless TBA equations which describe the renormalization group (RG)
flow from the tricritical M(4, 5) to critical M(3, 4) Ising model. We apply the methods
developed in PCAI and use the concepts and notations introduced in that paper without
further elaboration.
The layout of the paper is as follows. In Section 2 we discuss the classification of excitations using (m, n) systems including a description of the three mechanisms by which the
string contents change along the flow. In Section 3 we derive in detail the massless TBA
equations in the (r, s) = (1, 1) vacuum sector. We do not discuss the very similar derivation
of the TBA equations in the other 5 sectors. The numerical solution of the TBA equations
to yield continuous flows is discussed in Section 4. We finish in Section 5 with a brief
discussion.
2. Classification of excitations and flows

For small perturbations, the scaling limit of the excitations in Regime IV of the
A4 lattice model are classified by precisely the same (m, n) systems [13,14] as at the
conformal critical point [15] and throughout the massive Regime III [11]. Unlike the
massive case, however, we find that in the massless regime the string content can change
by one of three mechanisms along the flow. This was first observed empirically by direct
numerical diagonalization of a sequence of finite-size transfer matrices approaching the
scaling limit
=
mR
=
lim
N|t| ,
N, t 0
4
= 5/4
(2.1)
for modest values of the system size N and 0 mR < 5 and is confirmed by our numerical
solutions of the TBA equations. Here measures the perturbation strength and t is the
departure-from-criticality variable. The mass m and continuum length scale R usually
occur in the single combination mR. Notice that in the finite-size scaling we use the
Regime III correlation length exponent = 5/4 even though the actual [16] correlation
length exponent in Regime IV is = 5/2.
Let us consider the vacuum sector with boundary condition (r, s) = (1, 1). The excitation energies are given by the scaling limit of the eigenvalues of the double-row transfer
581
matrix D(u), or equivalently the normalized transfer matrix t(u), where u is the spectral
parameter. The two relevant analyticity strips in the complex u-plane are, respectively,
3
< Re(u) < ,

2 < Re(u) < 4
(2.2)
2
2
where = /5 is the crossing parameter. The excitations are classified by their string contents
mi = {number of 1-strings in strip i = 1, 2},

ni = {number of 2-strings in strip i = 1, 2}.
(2.3)
At the conformal critical point (mR = 0) the string contents satisfy the (m, n) system
1
m + n = (Ne1 + Im),
(2.4)
2
where m1 , m2 and N are even, m = (m1 , m2 ), n = (n1 , n2 ), e1 = (1, 0), and I is the A2
incidence matrix with entries Ij,k = |j k|,1 . For the leading excitations m1 , m2 , n2 are
finite but n1 N/2 as N .
As explained in PCAI, an excitation with string content (m, n) is uniquely labelled by
a set of quantum numbers

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

I = I (1) I (2) = I1 , I2 , . . . , Im(1)1 I1 , I2 , . . . , Im(2)2 ,
(2.5)
(i)
where the integers Ij {0, 1, 2, . . .} with i = 1, 2 give the number of 2-strings whose
(i)
(i)
(i)
imaginary parts wk are greater than that of the given 1-string vj . The 1-strings v1 and
(i)
(i)
2-strings w1 labelled by 1 are closest to the real axis. The quantum numbers Ij satisfy
(i)
(i)
ni I1 I2 Im(i)i 0,
i = 1, 2.
(2.6)
For given string content (m, n), the lowest excitation occurs when all of the 1-strings are
further out from the real axis than all of the 2-strings. In this case all of the quantum
(i)
numbers vanish Ij = 0. Bringing the location of a 1-string closer to the real axis by
interchanging the location of the 1-string with a 2-string increments its quantum number
by one unit and increases the energy.
Although we do not make use of it here, we mention in passing that, for the tricritical
Ising model, there exists a bijection [17] between the patterns of 1- and 2-strings classifying
the eigenvalues of the double-row transfer eigenvalues and RSOS paths on A4 . This has
as a consequence that the finitized partition functions (Virasoro characters) satisfy the
same recursions [12] as the one-dimensional configurational sums of the Corner Transfer
Matrices (CTMs).
2.1. Three mechanisms for changing string content
Let t(u) be the normalized double row transfer matrix as in PCAI. In Regime IV the
eigenvalues t (u) of t(u) are doubly periodic meromorphic functions in the period rectangle

9
(i, i),
period rectangle = ,
(2.7)
2 2
582
Fig. 1. Upper part of the period rectangle (/2, 2) ( i, i) in the complex u-plane for the A4 lattice
model in Regime IV showing the putative strips 1 and 2. The schematic location of zeros reflects the complex
conjugation symmetry and the symmetry (2.9). In the scaling limit the imaginary period and it is the
behaviour of the zeros near the indicated scaling edge at Im(u) = /2 that is relevant.
where is related to the departure-from-criticality variable t by

t = exp(2).
(2.8)
However, in Regime IV the A4 lattice model also admits the symmetry

t (u /2 + i) = t (u)
(2.9)
so we can restrict ourselves further to the rectangle (/2, 2) (i, i). In

particular, this means that for mR > 0 the distinction between the two strips effectively
disappearsthe two are joined along the scaling edge at Im(u) = /2 in the scaling limit
(2.1). This allows zeros to move between the putative strips 1 and 2 by crossing the scaling
edge. The combination of strips 1 and 2 in the upper half plane into one extended strip is
shown in Fig. 1. There is a complex conjugation symmetry between the upper and lower
half planes. Note that this is consistent with the picture at the conformal point since in the
limit mR 0 the zeros are infinitely far below (strip 1) or infinitely far above (strip 2) the
scaling edge as .
Let us consider the extended analyticity strip /2 Re(u) 3/2. Empirically,
we find that as mR is increased from 0 the strip 1 zeros approach the scaling edge at
Im(u) = /2 from below while the strip 2 zeros approach the scaling edge from above.
Potentially, this allows zeros to collide or migrate from one strip to the other. In fact we
find that the zero patterns change by one of the following three mechanisms which occur
583
Fig. 2. Schematic representation of the three mechanisms A, B, C by which the string contents change under the
flow. In each case the two circled zeros leave the analyticity strip as indicated by the arrows. In mechanism A two
1-strings leave the strip whereas in mechanisms B and C it is a 2-string that leaves the strip. Note that only the
location of the 1-strings enter the TBA equations.
at the scaling edge:

A. If 1-strings are closest to the scaling edge in both strips 1 and 2 then these 1-strings
collide and move symmetrically in opposite directions parallel to the scaling edge until
they reach Re(u) = 0, . At this point the two zeros no longer contribute and they can
be removed from the analyticity strip. This mechanism applies if Im(1)1 = Im(2)2 = 0.
B. If a 2-string is closest to the scaling edge in strip 2 and a 1-string is closest to the scaling
edge in strip 1 then the 2-string moves to the scaling edge and leaves the analyticity
strip. This mechanism applies if Im(1)1 = 0 and Im(2)2 > 0.
C. Otherwise, if a 2-string is closest to the scaling edge in strip 1, then this 2-string moves
to the scaling edge and leaves the analyticity strip. This mechanism applies if Im(1)1 > 0.
Similar mechanisms are also observed [8] in the boundary flows of the tricritical Ising
(1)
model. If m2 = 0 only mechanisms B and C apply according to whether Im1 = 0 or
(1)
Im1 > 0. If m1 = m2 = 0 then mechanism C applies. In each of the three mechanisms
exactly two zeros are removed from the extended analyticity strip. These three mechanisms are shown schematically in Fig. 2. They are confirmed by the numerics discussed in
Section 4.
2.2. Operator flow
Remarkably, as we explain in this subsection, the empirical rules giving the three
mechanisms A, B, C for changing string contents under the flow suffice to determine a
map in each sector (r, s) between finitized characters for the UV (mR = 0, c = 7/10) and
584
Table 1
Flow of primary operators (r, s) (r , s ) displayed in the A4 Kac table
(r , s )
4
r,s
3
r ,s
7 0
4 32 16
3 1
3 35 80
10
1 3 3
2 10
80 5
7 3
1 0 16
2
4 (1, 3) (1, 2) (1, 1)
1 0
4 12 16
1 0
3 12 16
3 (2, 3) (2, 2) (2, 1)
2 (2, 1) (2, 2) (2, 3)
1 1
1 0 16
2
1 (1, 1) (1, 2) (1, 3)
3 r
1 1
2 0 16
2
1 2 3 r
IR (mR = , c = 1/2) fixed points. Since two zeros leave the analyticity strip under the
flow, this map from the UV to IR takes the form
4,N+2
r,s
(q) r3,N
,s (q),
1 r 2,
1 s , r 3,
1 s 4.
(2.10)
We find that, in the presence of a boundary, the primary operators (r, s) of the tricritical
Ising model flow to primary operators (r , s ) of the critical Ising model as shown
in Table 1. This pattern of flows differs fundamentally from the flow of operators
observed with periodic boundaries. For example, with periodic boundaries, the spinless
operator (, ) = (1/10, 1/10) (1/2, 1/2) whereas in the presence of a boundary
= 1/10 0. These differences arise because the extra zeros associated with the
boundary participate in the flow mechanisms.
4 (q) to 3 (q)
2.3. RG flow from 1,1
1,1
In the vacuum sector (r, s) = (1, 1) the A4 (m, n)N+2 system of the tricritical Ising
model is
1
1
m2 + n2 = m1
m1 + n1 = (N + 2 + m2 ),
(2.11)
2
2
and these relations determine n1 and n2 in terms of m1 and m2 . Similarly, let m and n
be the number of 1- and 2-strings of the critical Ising model in the vacuum (r, s) = (1, 1)
sector satisfying the A3 (m, n)N system
1
1
m + n = (N + m) or n = (N m).
(2.12)
2
2
Then m and n are given by the total number of 1- and 2-strings in the extended strip in the
IR limit

A,
n1 + n2 ,
m1 + m2 2, A,
n=
m=
(2.13)
m1 + m2 ,
n1 + n2 1, B, C.
B, C,
The mechanisms A, B, C determine which energy level flows to which energy level
under the RG flow. As explained in [15] and PCAI, the energies at a conformal point are
585
Table 2
Mapping of the first 22 energy levels in the (r, s) = (1, 1) sector under mechanisms A, B or C. The energies,
string contents and quantum numbers are shown in both the UV and IR regimes. The degeneracies of all the
4 (q) 3 (q)
levels reconstruct the mapping between the tricritical and critical Ising characters 1,1
1,1
Mech.
m1
m2
n2
I
E
C
B
B
B
B
B
B
B
B
B
A
B
B
B
A
B
B
B
B
A
A
B
0
2
3
4
4
5
5
6
6
6
6
7
7
7
7
8
8
8
8
8
8
8
0
2
2
2
2
2
2
2
2
2
4
2
2
2
4
2
2
2
2
4
4
4
0
0
0
0
0
0
0
0
0
0
2
0
0
0
2
0
0
0
0
2
2
0
0
1
1
1
1
1
1
1
1
1
0
1
1
1
0
1
1
1
1
0
0
2
()
(0, 0)
(1, 0)
(2, 0)
(1, 1)
(3, 0)
(2, 1)
(4, 0)
(3, 1)
(2, 2)
(0, 0, 0, 0|0, 0)
(5, 0)
(4, 1)
(3, 2)
(1, 0, 0, 0|0, 0)
(6, 0)
(5, 1)
(4, 2)
(3, 3)
(2, 0, 0, 0|0, 0)
(1, 1, 0, 0|0, 0)
(0, 0, 0, 0)
0
2
2
2
2
2
2
2
2
2
4
2
2
2
4
2
2
2
2
4
4
4
()
(0, 0)
(1, 0)
(2, 0)
(1, 1)
(3, 0)
(2, 1)
(4, 0)
(3, 1)
(2, 2)
(0, 0, 0, 0)
(5, 0)
(4, 1)
(3, 2)
(1, 0, 0, 0)
(6, 0)
(5, 1)
(4, 2)
(3, 3)
(2, 0, 0, 0)
(1, 1, 0, 0)
(1, 1, 1, 1)
0
2
3
4
4
5
5
6
6
6
8
7
7
7
9
7
7
7
7
10
10
12
determined by the string content m and the patterns of zeros in the complex u-plane. In
terms of quantum numbers the precise mapping of energy levels under the flow is given by
(2) (2)

(1) (1)
I = I1 , I2 , . . . , Im(1)1 I1 , I2 , . . . , Im(2)2 I = I1 , I2 , . . . , Im
(2.14)
where
A:
B, C:
Ij = n2 + Ij(1) ,
(2)
Im 1 1+k = n2 Im2 k ,

Ij = n2 1 + Ij(1) ,
(2)
Im 1 +k = n2 Im2 +1k ,
j = 1, 2, . . . , m1 1,
k = 1, 2, . . . , m2 1,
j = 1, 2, . . . , m1 ,
k = 1, 2, . . . , m2 .
(2.15)
(2.16)
Details of this mapping for the first 15 energy levels in the (r, s) = (1, 1) sector are shown
in Table 2. For given string content m, the base energy level Em is determined by the
Cartan matrix C
1 2
2
1
q 2 m1 m1 m2 +m2 , A4 ,
Em
mCm
4
=
q =q
(2.17)
1 2
A3 .
q 2m ,
The base energy occurs when the location of all of the 1-strings are further from the real
axis than the locations of all of the 2-strings. Additional excitation energy is generated by
586
permuting the order of 1- and 2-strings in each strip as dictated by the sum of quantum
numbers Ij in each strip and is given by the Gaussian polynomials or q-binomials

Im1
I1
n
m+n
m+n
=
=
q I1 ++Im
m
m
q
I1 =0 I2 =0
Im =0
(q)m+n
, m, n 0,
= (q)m (q)n
0,
otherwise,
(2.18)
with the q-factorials (q)m = (1 q) (1 q m ) for m 1 and (q)0 = 1. The energy E is

increased by one unit each time a 1-string is brought closer to the real axis by interchanging
its location with the location of an adjacent 2-string. The product of two q-binomials is the
generating function for the conformal spectra with given string content in each strip. The
q-binomials satisfy the properties

n1
n
n
n1
,
=
=
+ qm
m
m
nm
m1

m+n
m+n
= q mn
.
(2.19)
m
m
q
1/q
4 (q) is the generating function for the tricritical Ising conformal spectra
The character 1,1
in the (1, 1) sector. Explicitly, using the recursion (2.19), we decompose it into three terms

1
m1 + n1
m2 + n2
4,N+2
1,1
(q) = q 7/240
q 4 mCm
m1
m2
(m,n)N+2
=q
7/240

1 2
m1 m1 m2 +m22
q2
m1 ,m2 even

m2 + n2 1
m1 + n1 1
m1 1
m2 1

m1 + n1 1
m2 + n2 1
+ q m2
m1 1
m2

m1 + n1 1
m2 + n2
+ q m1
.
m1
m2

(2.20)
These three terms correspond precisely to the energy levels effected by mechanisms A, B
and C, respectively. So simply reading off the conformal energies from the respective zero
patterns after applying each mechanism we find the following mapping between finitized
characters
1 2
4,N+2
1,1
(q) q 1/48
q 2m

q
m1 ,m2 even
n2 (m1 1)+n2 (m2 1)
m1 + n1 1
m1 1
m2 + n2 1
m2 1

1/q
587

m1 + n1 1
m2 + n2 1
m1 1
m2
1/q

m
m
+
n
1
+
n
1
1
2
2
+ q n2 m1 +n2 m2
m1
m2
1/q
1 2
= q 1/48
q 2m
+ q (n2 1)m1 +(n2 1)m2
m even

m1 + n1 1
m2 + n2 1
q n2 (m1 1)
m1 1
m2 1
m1 even

m1 + n1 1
m2 + n2 1
+ q (n2 1)m1
m1 1
m2

m1 + n1 1
m2 + n2
+ q n2 m1
m1
m2
1 2
3 2
q 2m
q 2 m1 mm1
= q 1/48

m even
m1 even

7
(N + m m1 + 2)/2
(m1 2)/2
q m 2 m1 +2
m1 1
m m1 + 1

(N + m m1 )/2
(m1 2)/2
+ q m1
m1 1
m m1

(N + m m1 )/2
m1 /2
+
m1
m m1
1 2 (N + m)/2
3,N
= q 1/48
q 2m
(q).
= 1,1
m
(2.21)
m even
Notice that, after the mapping, the q-binomials of strip 2 are with respect to 1/q. This
is because strip 2 is turned upside down when it is placed on top of strip 1 to form the
extended strip. These q-binomials are naturally replaced by q-binomials in q using (2.19).
All integers are then eliminated in favour of m and m1 using the appropriate relations
which apply to the mechanism corresponding to each of the three terms. The final equality
holds because of the remarkable generalized q-Vandermonde identity which is proved in
Appendix A

3 2
(N + m)/2
=
q 2 m1 mm1
m
m1 even

(m1 2)/2
m 72 m1 +2 (N + m m1 + 2)/2
q
m1 1
m m1 + 1

(N + m m1 )/2 (m1 2)/2
+ q m1
m1 1
m m1

(N + m m1 )/2
m1 /2
+
(2.22)
.
m1
m m1
588
3,N
The finitized Ising character 1,1
(q) is not the usual finitized character. In the limit q 1
it gives the correct A4 counting
(N + m)/2

3,N
lim (q) =
(2.23)
,
= AN+2
4
1,1
m
q1 1,1
m even
whereas the usual finitized Ising character gives A3 counting
N/2
1 2 N/2
1/48
m
=
= AN
lim q
q2
3 1,1
m
m
q1
m even
(2.24)
m even
where A4 and A3 denote the adjacency matrices.

2.4. RG flow in other sectors
The analysis of the flow using the three mechanisms A, B and C can be extended to
each of the sectors (r, s) with r = 1, 2, s = 1, 2, 3. In this way we obtain six generalized
q-Vandermonde identities as follows:
1,1 N even, m even:

2
(N + m)/2
q 3m1 /2mm1
=
m
m1 even

(N + m m1 )/2
m1 /2
m1
m m1

(N + m m1 )/2
(m1 2)/2
+ q m1
m1 1
m m1

(N + m m1 + 2)/2
(m1 2)/2
+ q m7m1 /2+2
;
m1 1
m m1 + 1
(2.25)
3,1 N even, m odd:

(N + m m1 + 1)/2
m1 /2
(N + m + 1)/2
3m21 /2mm1
q
=
m1
m m1
m
m1 even

(N + m m1 + 1)/2 (m1 2)/2
+ q m1
m1 1
m m1

(m1 2)/2
m7m1 /2+2 (N + m m1 + 3)/2
+q
;
m1 1
m m1 + 1
(2.26)
2,2 N even, m even:

2
(N + m)/2
=
q 3m1 /2mm1
m
m1 even

(N + m m1 )/2
m1 /2
m1
m m1
589

(N + m m1 2)/2
m1 /2
m1
m m1 1

m1 /2
m5m1 /2+1 (N + m m1 )/2
; (2.27)
+q
m1 1
m m1 + 1
+ q m+5m1 /2+1
2,1
N odd, m even:

2
(N + m 1)/2
=
q 3m1 /2mm1
m
m1 odd

(N + m m1 2)/2 (m1 + 1)/2
q m1 /2
m1
m m1

(m1 1)/2
m1 /2 (N + m m1 2)/2
+q
m1 1
m m1

(N + m m1 )/2
(m1 1)/2
+ q m3m1 +3/2
;
m1 1
m m1 + 1
(2.28)
1,2
N odd, m even:

2
(N + m 1)/2
=
q 3m1 /2mm1
m
m1 odd

(m1 1)/2
m3m1 +3/2 (N + m m1 )/2
q
m1 1
m m1 + 1

(N + m m1 2)/2
(m1 1)/2
+ q m+2m1 +1/2
m1
m m1 1

(m1 1)/2
m1 /2 (N + m m1 )/2
; (2.29)
+q
m1
m m1
3,2
N odd, m odd:

2
(N + m)/2
=
q 3m1 /2mm1
m
m1 odd

(m1 1)/2
m3m1 +3/2 (N + m m1 + 1)/2
q
m1 1
m m1 + 1

(N + m m1 1)/2
(m1 1)/2
+ q m+2m1 +1/2
m1
m m1 1

(m1 1)/2
m1 /2 (N + m m1 + 1)/2
. (2.30)
+q
m1
m m1
These identities are simplified and proved in Appendix A.
590
3. TBA equations in Regime IV

Recall from PCAI that the normalized double row transfer matrix is defined by

2N
1 (u + 2, p)1 (, p)
D(u),
t(u) = Sr,s (u)S(u) i
1 (u + 3, p)1 (u + , p)
(3.1)
where
S(u) =
1 (2u , p)2
1 (2u 3, p)1 (2u + , p)
(3.2)
and
Sr,s (u) = (1)s hr (u L )hr (u + L )h s (u R )h s (u + R )
(3.3)
with
1 (, p)1 (u + (3 r), p)1 (u + (1 r), p)
,
1 (u, p)1 (u , p)1 (u + 2, p)
4 (, p)4 (u + (3 s), p)4 (u + (1 s), p)
.
h s (u) =
4 (u, p)4 (u , p)4 (u + 2, p)
hr (u) =
(3.4)
(3.5)
Moreover, the normalized transfer matrix satisfies [18] the universal TBA functional
equation
t(u)t(u + ) = I + t(u + 3)
(3.6)
independent of the boundary condition (r, s). For Regime IV, the nome is pure imaginary
with p = ie, .
3.1. UV massless TBA: (r, s) = (1, 1)
In this section we derive the TBA equations for the (r, s) = (1, 1) boundary by solving
the TBA functional equations in the scaling limit for even N . We follow closely the
derivations in [15,19] and PCAI [11]. The derivation for other boundary conditions is
similar. We begin by factorizing the eigenvalue t (u) of t(u) for large N as
t (u) = f (u)g(u)l(u),
(3.7)
where f (u) accounts for the bulk order-N behaviour, g(u) the order-1 boundary
contributions and l(u) is the order-1/N finite-size correction. We will solve for f (u), g(u)
and then l(u) sequentially. For the order-N behaviour the second term on the r.h.s. of the
TBA functional equation (3.6) can be neglected giving the inversion relation
f (u)f (u + ) = 1.
In the physical strip 1, the solution [20] with the required analyticity is

5/2 ) 2N
4 ( 4 5u
2 , |t|
.
f (u) =
5/2 )
4 ( 4 + 5u
2 , |t|
(3.8)
(3.9)
591
The solution in strip 2 satisfies the same inversion relation and is given by the
symmetry (2.9).
In the two analyticity strips labelled by j = 1, 2 we define generically for the functions
h = t, f, g, l the notations

ix
+
, Im(x) < ,
h1 (x) = h
(3.10a)
2
5

ix
h2 (x) = h 3 +
(3.10b)
, Im(x) < ,
5
H1 (x) = 1 + h1 (x),
(3.10c)
H2 (x) = 1 + h2 (x)
and we assume the relevant functions have the scaling form
h i (x) = lim h(x + log N).
N
(3.11)
For example, we see that

log f1 (x) = 42 ex .
(3.12)
As in Regime III, we next have to solve by Fourier series the inversion relations for the
order-1 boundary terms

g1 x i
(3.13a)
g1 x + i
= 1,
2
2

g2 x + i
= G1 (x).
g2 x i
(3.13b)
2
2
To find gi (x) we need to consider the zeros and poles introduced by the prefactor in
(3.1). We find the order-1 zeros of D(u) cancel exactly the poles of S1,1 (u). Taking into
account periodicity, the prefactor exhibits poles at u = 2 + i,, 3
2 + i,, 2 + i,,
, 3
, 11
, 13
,
4 + i,, 4 i 2 , 4 i 2 , 4 i 2 , 4 i 2 and double zeros at u = 2 + i ,,
, 17
,
3 + i ,, 7
4 i 2 , 4 i 2 where = 0, 1. The solution for g1 (x) with this
analyticity but restricted to the strip | Im(x)| < 3/4 is

5/4 ) 2
3 ( ix2 8 , |t|5/4 )3 ( ix2 + 8 , |t|5/4 ) 1 ( ix
2 , |t|
g1 (x) =
(3.14)
.
5/4 )
4 ( ix2 8 , |t|5/4 )4 ( ix2 + 8 , |t|5/4 ) 2 ( ix
2 , |t|
By comparing the expected pattern of zeros and poles of g2 (x i/2)g2 (x + i/2) with
the analyticity of G1 (x), we find that this functional relation is satisfied inside the strip
| Im(x)| < 3
4 . We observe that g2 (x)/g1 (x) is free of zeros and poles in this strip. Similarly
G1 (x) is analytic and nonzero, but only inside a narrower strip | Im(x)| < 4 . Hence in this
smaller strip
log g2 (x) = log g1 (x) + log G1 (x),
(3.15)
where the kernel in Regime IV is given by

(x) =
2 (0, |t|5 )3 (0, |t|5 )3 (ix, |t|5 )

.
22 (ix, |t|5 )
(3.16)
592
Note that the signs of gi (x) are chosen to match the corresponding expressions [15] in the
tricritical limit t 0. They can be determined numerically from the eigenvalues of the
transfer matrix with real u in the physical range 0 < u < /2. The functional equations for
the finite-size corrections

i
i
l1 x +
= T2 (x),
l1 x
(3.17a)
2
2

T1 (x)
i
i
l
=
x
x
+
l2
(3.17b)
2
2
2
G1 (x)
can be converted to Nonlinear Integral Equations (NLIE) by standard techniques [21,22]
where the key input is the analyticity determined by the patterns of zeros. Suppose that
1-strings are located in the upper half plane in the extended strip 1 at {/2 + iv1 , . . . , /2 +
ivm1 } below the scaling edge and at {/2 + iw1 , . . . , /2 + iwm2 } above the scaling
edge. Then the zeros in strip 2 are determined by the symmetry (2.9) and occur at
{3 + i( v1 ), . . . , 3 + i( vm1 )} and {3 + i( w1 ), . . . , 3 + i( wm2 )}.
To account for these zeros, we note that the functions
5v
5v
m1

2 (i x2 i 2j , |t| 2 )2 (i x2 + i 2j , |t| 2 )
5
l1 (x)
j =1
1 (i x2 i
5vj
2
, |t| 2 )1 (i x2 + i
5
5vj
2
2 , |t| )
m2

2 (i x2 i 5w2 k , |t| 2 )2 (i x2 + i 5w2 k , |t| 2 )
5
5wk
5wk
x
x
k=1 1 (i 2 i 2 , |t| 2 )1 (i 2 + i 2 , |t| 2 )
j =1
(3.18)
5v
5v
m1

3 (i x2 i 2j , |t| 2 )3 (i x2 + i 2j , |t| 2 )
5
l2 (x)
4 (i x2 i
5vj
2
, |t| 2 )4 (i x2 + i
5
5vj
2
2 , |t| )
m2

3 (i x2 i 5w2 k , |t| 2 )3 (i x2 + i 5w2 k , |t| 2 )
5
(3.19)
5wk
5wk
x
x
k=1 4 (i 2 i 2 , |t| 2 )4 (i 2 + i 2 , |t| 2 )
are free of zeros and poles inside their respective analyticity strips. The products of elliptic
i
functions satisfy the inversion relations li (x i
2 )li (x + 2 ) = 1, i = 1, 2 and are doublyperiodic.
It follows that
log t1 (x) = log f1 (x) + log g1 (x) + log T2 (x) + C1
+
m1
log
j =1
m2
k=1
log
5
5vj
x
2
2 , |t| )1 (i 2
5
5v
2 (i x2 i 2j , |t| 2 )2 (i x2
1 (i x2 i
5
5vj
2
2 , |t| )
5
5vj
+ i 2 , |t| 2 )
+i
1 (i x2 i 5w2 k , |t| 2 )1 (i x2 + i 5w2 k , |t| 2 )

2 (i x2 i 5w2 k , |t| 2 )2 (i x2 + i 5w2 k , |t| 2 )
(3.20)
593
log t2 (x) = log f2 (x) + log g2 (x) + log T1 (x) log G1 (x) + C2
+
m1
log
j =1
m2
log
k=1
5
5vj
x
2
2 , |t| )4 (i 2
5
5v
3 (i x2 i 2j , |t| 2 )3 (i x2
4 (i x2 i
5
5vj
2
2 , |t| )
5
5vj
+ i 2 , |t| 2 )
+i
4 (i x2 i 5w2 k , |t| 2 )4 (i x2 + i 5w2 k , |t| 2 )

3 (i x2 i 5w2 k , |t| 2 )3 (i x2 + i 5w2 k , |t| 2 )
(3.21)
where (x) is the kernel (3.16). As in [15], the integration constants Ci vanish. So now
taking the scaling limit yields
ex + 2 + 1 ex
2 x
log t1 (x) = 4 e + log
+ k log T2 (x)

ex 2 + 1 ex

m1
1 (1)
+
log tanh (j x log )
2
j =1

1 (2)
+
log tanh (k x log ) ,
2
k=1
ex + 2 + 1 ex
x
log t2 (x) = 4e + log

+ k log T1 (x)
ex 2 + 1 ex

m1
1 (1)
+
log tanh (j + x + log )
2
m2
j =1
m2
k=1

1 (2)
log tanh (j + x + log ) ,
2
(3.22)
(3.23)
where the kernel is

1
.
2 cosh x
In deriving this result we have assumed that the zeros scale as
k(x) =
(1)
vj
(1)
j
log |t| j
+
=
+
,
4
5
2
5
(2)
(3.25)
(2)
k
log |t| k
+
=
+
4
5
2
5
or, more precisely, the scaled locations of the zeros are defined by

5
j(1) =
log
|t|
=
lim
,
5v
5v
lim
j +
j + log
N,t 0
N,t 0
4
N

5
(2)
lim
k =
lim
5wk + log |t| =
5wk + log
.
N,t 0
N,t 0
4
N
wk
(3.24)
(3.26)
(3.27)
(3.28)
594

(1)
(2)
Close to the UV fixed point, when mR is small, we must have j < 0 and k > 0. If we
define the rapidity and pseudo-energies ,i () by

= ti ( log ), i = 1, 2
e,i () = lim ti log
(3.29)
N
N
where = mR/4 we obtain the TBA equations

,2 ( ) )
e + 2 + e
1
log(1 + e
d
,1 () = mRe log
2
cosh( )
e 2 + e
m1
j =1

(1)
(2)
m2
j
,
log tanh
log tanh k
2
2
2
2
k=1
,2 () = mRe

log(1 + e,1 ( ) )
e + 2 + e
1
log
d
2
cosh( )
e 2 + e
m1
j =1

(1)
(2)
m2
j
.
+
log tanh
log tanh k +
2
2
2
2
(3.30)
k=1
(1)
(2)
To find the locations j , k of the 1-strings consider the functional equations

t1 x i
t1 x + i
= 1 + t2 (x),
2
2

t2 x i
t2 x + i
= 1 + t1 (x),
2
2
(3.31)
at x = i/2 + 5vj , x = i/2 + 5wk , respectively. Since the right-hand sides vanish, in
the scaling limit this implies

(1)
i
log = 1 = enj i , j = 1, 2, . . . , m1 ,
t2 j(1)
2

(2)
i
(2)
log = 1 = enk i , k = 1, 2, . . . , m2 ,
t1 k
(3.32)
2
or

i
(1)
(1)
,2 j
= nj i, j = 1, 2, . . . , m1 ,
2

i
(2)
,1 k
(3.33)
= n(2)
k i, k = 1, 2, . . . , m2 ,
2
(1)
(2)
where nj , nk are odd integers. These integers are given by their values [15] in the UV
limit as determined by winding numbers
(1)
(1)
nj = 2(m1 j ) m2 + 1 + 2Ij ,
j = 1, 2, . . . , m1 ,
(2)
n(2)
k = 2(m2 k) m1 + 1 + 2Ik ,
k = 1, 2, . . . , m2 ,
(3.34)

(1)
(1)
(2)
595
(2)
where I = (I1 , . . . , Im1 |I1 , . . . , Im2 ) are the quantum numbers (2.5). These integers
(i)
n8 can change during the flow due to winding of phases. For numerical purposes, a more
(i)
useful form of these auxiliary equations is obtained by replacing with 8 i/2 in the
TBA equations (3.30). Similar equations can obtained for the locations of the 2-strings.
Repeating the same calculation as in Regime III, we find the finite-size corrections to
the scaled energies in Regime IV are

1
RE(R) cosh x
1
log D1 (x) =
+o
(3.35)
2
2N
N
with

RE(R) = mR
m1
(1)
j =1
m2

e
(2)
k
k=1
mR

d e log 1 + e,1 () . (3.36)
In the UV limit mR 0 we recover the critical TBA equations of [15]. Explicitly,

setting = x + log , we find that as 0 the location of the 1-strings scale as
j(1) = yj(1) + log ,
j = 1, 2, . . . , m1 ,
k(2) = yk(2) log ,
k = 1, 2, . . . , m2 .
(3.37)
It follows that in the limit mR 0 the finite-size corrections to the scaled energies are
given exactly by

mi
2
7
1
1
2
(i)
log D1 (u) =
sin 5u
mCm
(3.38)
Ij
2
N
240 4
i=1 j =1
with the finitized partition function

ZN (q) = q
7/240
1
4 mCm
m1 , m2 even
m1 + n1
m1

q
m2 + n2
m2
(3.39)
where the modular parameter is q = exp(sin 5uM/N) for M double rows.

We have chosen to write the TBA equations in terms of both strips 1 and 2. However,
in accord with the existence of a single extended strip, the set of TBA equations (3.30),
(3.33), (3.34) and (3.36) can be written in terms of a single strip by using the symmetry
,2 () = ,1 (). Explicitly, extending strip 1, we obtain the TBA equations

e + 2 + e
1
log(1 + e,1 ( ) )
d
,1 () = mRe log
2
cosh( + )
e 2 + e
8=1
RE(R) = mR

(1)
8
,
log tanh
2
2
8=1
(1)
mR

d e log 1 + e,1 () ,
(3.40)
(3.41)
596
where m = m1 + m2 and
(1)
m+1k
= k(2),
k = 1, 2, . . . , m2 .
The auxiliary equations become

i
(1)
8
,1
= n(1)
8 i,
2
(3.42)
8 = 1, 2, . . . , m,
(3.43)
8 = 1, 2, . . . , m
(3.44)
where
n(1)
8 = m 28 + 1 + 2I8 ,
and the quantum numbers I8 in the extended strip are given by

8 = 1, 2, . . . , m1 ,
n2 + I8(1) ,
I8 =
(1)
n2 Im+18 , 8 = m1 + 1, m1 + 2, . . . , m,
(3.45)
with n2 = m1 /2 m2 . This symmetry between strips 1 and 2 is manifestly broken in

the UV limit mR 0 and the IR limit mR . Also these TBA equations need to be
modified in the intermediate regime for mechanism A levels after collision of the 1-strings.
3.2. IR massless TBA: (r, s) = (1, 1)
For large mR we work with the extended strip 1. In this regime the total number of
1-strings is either m = m1 + m2 for mechanism B, C or m = m1 + m2 2 for mechanism A
with quantum numbers I = (I1 , I2 , . . . , Im ) given by (2.14) to (2.16). Setting
, () = ,1 () + mi = ,2 () + mi
(3.46)
we now obtain the single TBA equation

, ( ) )
e + 2 + e
1

log(1 + e
, () = mRe log
2
cosh( + )
e 2 + e
8=1

8
log tanh
2
2
(3.47)
with scaling energies

RE(R) = mR
8
8=1
The auxiliary equations are

i

,
8 = n8 i,
2
mR

d e log 1 + e, () .
(3.48)
8 = 1, 2, . . . , m,
(3.49)
where
n8 = 2(m 8) + 1 + 2I8 ,
8 = 1, 2, . . . , m.
(3.50)
597
The difference between this equation and (3.44), as reflected in (3.46), arises because of
the mi winding between the previous position of the scaling edge at i/2 (reference
point for = 0) and the new position of the scaling edge at i.
Setting = x + log in the IR limit mR , the locations of the 1-strings scale as
8 = y8 + log ,
8 = 1, 2, . . . , m
(3.51)
and the above equations reduce to the TBA equations of the critical Ising model. Indeed,
the pseudo energies ,i () decouple giving the energy of the usual massless free fermions
, () mRe
(3.52)
so that the auxilliary equations (3.49) become

4ey8 = 2(m 8) + 1 + 2I8 .
(3.53)
In this limit the finite size corrections for the scaled energies are

1
cosh x
m2

1
+ E , E =
log D1 (x) =
+
I8
2
N
48
2
(3.54)
with the partition function

Z1,1 (q) = q 1/48
Ising
2 /2
qm
m even
N+m2
2
(3.55)
4. Massless numerics
The TBA equations of the previous section can be solved numerically by an iterative
procedure. There are, however, some subtleties. The process starts with initial guesses for
the pseudoenergies ,i () and 1-string locations, close to the UV or IR fixed points. The
flow is followed by progressively incrementing or decrementing mR. At each value of
mR, the TBA equations are used to update the pseudoenergies ,i () and then these are
used in the auxiliary equations to update the locations of the 1-strings, and so on, until
a stable solution is reached. Typically, the UV form of the equations is stable for small
values of mR, the IR form is stable for large values of mR and there is an intermediate
range of values of mR for which both forms are stable and converge (with a precision of
five decimals places) to the same values for the scaled energies and the locations of each
of the 1-strings. In all cases these numerical flows confirm the three mechanisms A, B, C.
One numerical difficulty is related to the determination of the location of the 1-strings
(2)
in strip 2. This problem arises because in the UV form of the equations k cannot be
obtained by direct iteration of the auxiliary equation. This problem is solved by inverting
the phases
(1)

(2)
j k
i
+
log tanh
.
2
4
(4.1)
598
Fig. 3. Flow of groundstate energy ceff = 12RE(R)/ vs. log(mR) for periodic and (r, s) = (1, 1) boundary
conditions. The difference arises from the boundary term in the TBA equations. The energy is a decreasing
function of mR for periodic boundary conditions by Zamolodchikovs c-theorem but this theorem does not apply
with fixed boundary conditions.
A more serious problem relates to mechanism A cases for which two 1-strings collide to
form a (short) 2-string with complex coordinates
(1)
(2)
m
, m
i .
1
2
(4.2)
For these cases there is an intermediate range of values of mR which requires a

modification of the TBA equations. Such short 2-strings were studied [23] in the context of
the YangLee scaling theory. Unfortunately, due to instabilities in our equations, we have
been unable to numerically solve the mechanism A equations throughout the intermediate
region.
We present some typical numerical results in a series of figures. In Fig. 3, we compare
the groundstate scaling energy in the (r, s) = (1, 1) sector with the scaling energy for
periodic boundary conditions. In Figs. 4 and 9 we show the flow of scaling energies in
the sectors (r, s) = (1, 1) and (r, s) = (3, 1), respectively. A dashed curve is used to guide
the eye in the intermediate regime of the mechanism A level. In Figs. 5 and 6, we show
the flow of the scaling energy and 1-strings for the mechanism A level in this sector
with string contents (m1 , m2 ) = (4, 2) and UV quantum numbers I = (0, 0, 0, 0|0, 0).
The dashed curves in the intermediate regime are schematic and have not been calculated
from the solution of the TBA equations. For comparison, we show in Figs. 7 and 8, the
flow of the scaling energy and 1-strings for the mechanism B level with string contents
(m1 , m2 ) = (6, 2) and quantum numbers I = (0, 0, 0, 0, 0, 0|1, 1). The flow of the scaling
energies and 1-strings for arbitrary mechanism B and C levels can be calculated throughout
the flow by numerical solution of the TBA equations. The mechanism A levels can be
calculated right up to the point where the two 1-strings collide. Note the linear regimes in
599
4 (q) 3 (q) of scaling energies c /24 = RE(R)/2 vs. log(mR) in the (r, s) = (1, 1)
Fig. 4. Flow 1,1
eff
1,1
sector. The degeneracies of the levels are shown in the margins. The intermediate region of the mechanism A
levels (shown dashed) are schematic and have not been obtained from numerical solution of the TBA equations.
Fig. 5. The scaling energy ceff /24 = RE(R)/2 vs. log(mR) for the mechanism A level in the (r, s) = (1, 1)
sector with string contents (m1 , m2 ) = (4, 2) and UV quantum numbers I = (0, 0, 0, 0|0, 0).
600
Fig. 6. Flow of six 1-strings vs. log(mR) for the mechanism A level in the (r, s) = (1, 1) sector with string
contents (m1 , m2 ) = (4, 2) and UV quantum numbers I = (0, 0, 0, 0|0, 0).
Fig. 7. The scaling energy ceff /24 = RE(R)/2 vs. log(mR) for the mechanism B level in the (r, s) = (1, 1)
sector with string contents (m1 , m2 ) = (6, 2) and UV quantum numbers I = (0, 0, 0, 0, 0, 0|1, 1).
the UV and IR for the flows of 1-strings. This corresponds to the assumed limiting scaling
of the locations of these 1-strings.
5. Discussion
In this paper we have used a lattice approach to derive TBA equations for all excitations
in the massless renormalization group flow from the tricritical to critical Ising model.
The excitations are classified according to string content which changes by one of three
mechanisms A, B, C along the flow and leads to a mapping between finitized Virasoro
characters. With the exception of the intermediate regime for mechanism A flows, the
601
Fig. 8. Flow of eight 1-strings versus log(mR) for the mechanism B level in the (r, s) = (1, 1) sector with string
contents (m1 , m2 ) = (6, 2) and UV quantum numbers I = (0, 0, 0, 0, 0, 0|1, 1).
4 (q) 3 (q) of scaling energies c /24 = RE(R)/2 vs. log(mR) in the (r, s) = (3, 1)
Fig. 9. Flow 3,1
eff
1,3
sector. The degeneracies of the levels are shown in the margins. The intermediate region of the mechanism A
levels (shown dashed) are schematic and have not been obtained from numerical solution of the TBA equations.
602
TBA equations can be solved numerically by iteration. It would be of interest to compare

our results with the results of the Truncated Conformal Space Approximation.
Although, the tricritical Ising model is superconformal, the boundary conditions applied
in this paper break the superconformal symmetry. It would be of interest to investigate the
pattern of the superconformal flows between fixed points corresponding to superconformal
boundary conditions [9,24]. It would also be of interest to extend the analysis of this paper
to the complete flow for periodic boundary conditions.
Acknowledgements
P.A.P. is supported by the Australian Research Council and thanks the Asia Pacific
Center for Theoretical Physics for support to visit Seoul. C.A. is supported in part by
Korea Research Foundation 2002-070-C00025, KOSEF 1999-00018. We thank Giuseppe
Mussardo for discussions.
Appendix A
In this appendix we prove the generalized q-Vandermonde identities of Section 2.4.
We simplify the identities to show they are special cases of general identities obtained by
Bender [25]. We then give an elementary proof of these identities using induction.
To simplify the identities of Section 2.4 we set m1 = 2k or m1 = 2k + 1 depending on
the parity of m1 and set n = (N + m)/2 or n = (N + m 1)/2 as appropriate. This reduces
the six identities to four identities

2
nk
k
n
=
q 6k 2km
1,1 /3,1 :
2k
m 2k
m
k

k1
2k n k
+q
2k 1
m 2k

nk+1
k1
,
+ q m7k+2
2k 1
m 2k + 1
(A.1)

nk
k
n
6k 2 2km
2,2 :
=
q
2k
m 2k
m
k

k
m+5k+1 n k 1
+q
2k
m 2k 1

k
m5k+1 n k
+q
, (A.2)
2k 1
m 2k + 1

2
n
2,1 :
=
q 6k 2kmm+6k+3/2
m
k

nk1
k+1
q k+1/2
2k + 1
m 2k 1

nk1
k
2k
m 2k 1

nk
k
+ q m6k3/2
,
2k
m 2k
+ q k1/2

1,2 /3,2 :
603
2
n
q 6k 2kmm+6k+3/2
=
m
k

nk
k
q m6k3/2
2k
m 2k

nk1
k
+ q m+4k+5/2
2k + 1
m 2k 2

nk
k
.
+ q k1/2
2k + 1
m 2k 1
(A.3)
(A.4)
After some recasting, a surprising mod 3 property emerges in the terms of these identities

2
n
=
q 6k 2km
1,1 /3,1 :
m
k

nk
k
n 3k
3k m

nk
k1
+ q 2k
n 3k + 1
3k 1 m

nk+1
k1
,
+ q m7k+2
(A.5)
n 3k + 2
3k 2 m

2
n
q 6k 2km
2,2 :
=
m
k

nk
k
n 3k
3k m

nk1
k
+ q m+5k+1
n 3k 1
3k + 1 m

nk
k
,
+ q m5k+1
(A.6)
n 3k + 1
3k 1 m

2
n
2,1 :
=
q 6k 2km
m
k

nk1
k+1
q m+7k+2
n 3k 2 3k + 2 m

nk1
k
+ q m+5k+1
n 3k 1
3k + 1 m

nk
k
+
(A.7)
,
n 3k
3k m
604

1,2 /3,2 :
2
n
=
q 6k 2km
m
k

nk
k
n 3k
3k m

nk1
k
+ q 2m+10k+4
n 3k 2
3k + 2 m

nk
k
+ q m+5k+1
.
n 3k 1
3k + 1 m
(A.8)
Setting 8 = 3k, 3k + 1, 3k + 2 mod 3 reduces the four identities to just two identities

n
8/3
(88/3)(8m) n (8 + 1)/3
q
=
1,1 /3,1 /1,2 /3,2 :
,
m
n8
8m
8
(A.9)

n
q (8(8+1)/3)(8m)
=
2,2 /2,1 :
m
8

n (8 + 2)/3
(8 + 1)/3
.
(A.10)
n8
8m
For (n, m) = (0, 0) there is an additional identity

n (8 + 3)/3
(8 + 2)/3
n
.
q (8(8+2)/3)(8m)
=
n8
8m
m
(A.11)
These are in fact special cases2 of identities due to Bender [25].

A.1. Generalized q-Vandermonde identities

n
n (8 + 1 + a)/3
(8 + a)/3
=
,
q (28+2a)/3(8m)
m
n8
8m
8
m a 2n + 1.
(A.12)
Proof. For n = m we have 8 = m so l.h.s. = r.h.s = 1 for m a 2m + 1. We now

proceed by induction on n. Suppose that m a 2n + 1 and (m 1) a + 1 2n + 1,
that is, m a 2n. Then

n+1
n
n
n
n
=
+ q nm+1
=
+ q (n8)+(8m+1)
m
m
m1
m1
m

n (8 + 1 + a)/3
(8 + a)/3
q (28+2a)/3(8m)
=
n8
8m
8
2 We thank George Andrews for pointing this out to us.
605
q ((28+1a)/3+1)(8m+1)

n (8 + 2 + a)/3
(8 + 1 + a)/3
q
n8
8m+1

n (8 + 1 + a)/3
=
q (28+2a)/3(8m)
n8
8

n (8 + 1 + a)/3
(8 + a)/3
+ q n8+1
n8+1
8m

n + 1 (8 + 1 + a)/3
(8 + a)/3
=
q (28+2a)/3(8m)
.
n+18
8m
n8
Now suppose that a = 2n + 1. Then only the terms 8 = n and 8 = n + 1 survive on the
r.h.s. and

(8 + a)/3
(28+2a)/3(8m) n + 1 (8 + 1 + a)/3
q
n+18
8m
8

(8 + 2n + 1)/3
(282n+1)/3(8m) n + 1 (8 + 2n + 2)/3
=
q
n+18
8m
8

n
n+1
n
=
.
=
+ q nm+1
m1
m
m
References
[1] C.N. Yang, C.P. Yang, J. Math. Phys. 10 (1969) 1115.
[2] Al.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695;
Al.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 497;
Al.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 524;
Al.B. Zamolodchikov, Nucl. Phys. B 366 (1991) 122.
[3] M.J. Martins, Phys. Rev. Lett. 67 (1991) 419.
[4] P. Fendley, Nucl. Phys. B 374 (1992) 667.
[5] P. Dorey, R. Tateo, Nucl. Phys. B 482 (1996) 639.
[6] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
[7] F. Lesage, H. Saleur, P. Simonetti, Phys. Lett. B 427 (1998) 85.
[8] G. Feverati, P.A. Pearce, F. Ravanini, Phys. Lett. B 534 (2002) 216223.
[9] R.I. Nepomechie, Int. J. Mod. Phys. A 17 (2002) 3809.
[10] R.I. Nepomechie, C. Ahn, Nucl. Phys. B 647 (2002) 433470.
[11] P.A. Pearce, L. Chim, C. Ahn, Nucl. Phys. B 601 (2001) 539568.
[12] G.E. Andrews, R.J. Baxter, P.J. Forrester, J. Stat. Phys. 35 (1984) 193.
[13] E. Melzer, Int. J. Mod. Phys. A 9 (1994) 1115.
[14] A. Berkovich, Nucl. Phys. B 431 (1994) 315.
[15] D.L. OBrien, P.A. Pearce, S.O. Warnaar, Nucl. Phys. B 501 (1997) 773.
[16] R.J. Baxter, P.A. Pearce, J. Phys. A 16 (1983) 2239.
[17] G. Feverati, P.A. Pearce, Critical RSOS and minimal models I: Paths, fermionic algebras and Virasoro
modules, hep-th/0211185.
[18] R.E. Behrend, P.A. Pearce, D.L. OBrien, J. Stat. Phys. 84 (1996) 1.
[19] P.A. Pearce, B. Nienhuis, Nucl. Phys. B 519 (1998) 579.
606
[20]
[21]
[22]
[23]
[24]
[25]
R.J. Baxter, J. Stat. Phys. 28 (1982) 1.

A. Klmper, P.A. Pearce, J. Stat. Phys. 64 (1991) 13.
A. Klmper, P.A. Pearce, Physica A 183 (1992) 304.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamoldchikov, Nucl. Phys. B 489 (1997) 487.
C. Richard, P.A. Pearce, Nucl. Phys. B 631 (2002) 447470.
E.A. Bender, A generalized q-binomial Vandermonde convolution, Discrete Math. 1 (1971) 115119.

CUMULATIVE AUTHOR INDEX B651B660
Abe, T.
Abed-Pour, N.
Abel, S.A.
Abramowicz, H.
Adam, C.
Adamczyk, L.
Adamus, M.
Aghamohammadi, A.
Aghuzumtsyan, G.
Ahmady, M.R.
Ahn, C.
Akemann, G.
Alkalaev, K.B.
Alonso-Alberca, N.
Andrianov, A.A.
Anselmi, D.
Antoniadis, I.
Antonioli, P.
Antonov, A.
Antusch, S.
Anzivino, G.
Aoyama, S.
Arai, M.
Ardonne, E.
Arneodo, M.
Aschieri, P.
Auzzi, R.
Ayad, R.
Ayala, A.
Azuma, T.
B658 (2003) 3
B655 (2003) 342
B651 (2003) 191
B658 (2003) 3
B657 (2003) 214
B658 (2003) 3
B658 (2003) 3
B655 (2003) 342
B658 (2003) 3
B655 (2003) 221
B660 (2003) 579
B660 (2003) 532
B655 (2003) 57
B651 (2003) 263
B660 (2003) 25
B658 (2003) 440
B660 (2003) 81
B658 (2003) 3
B658 (2003) 3
B658 (2003) 203
B658 (2003) 3
B656 (2003) 325
B652 (2003) 35
B660 (2003) 473
B658 (2003) 3
B651 (2003) 45
B653 (2003) 204
B658 (2003) 3
B651 (2003) 211
B651 (2003) 71
Babu, K.S.
Bagnoud, M.
Bailey, D.S.
Bamberger, A.
Barakbaev, A.N.
Barbagli, G.
Barberis, E.
Barbi, M.
B660 (2003) 322

B651 (2003) 71
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
0550-3213/2003 Published by Elsevier Science B.V.

doi:10.1016/S0550-3213(03)00400-0
Bardakci, K.
Bari, G.
Barreiro, F.
Bartsch, D.
Barvinsky, A.O.
Basile, M.
Basu-Mallick, B.
Bauerdick, L.A.T.
Beenakker, W.
Bgin, L.
Behrens, U.
Beisert, N.
Belitsky, A.V.
Bell, M.
Bellagamba, L.
Bellini, M.
Beneke, M.
Benen, A.
Berezinsky, V.
Berges, J.
Bertolin, A.
Besprosvany, J.
Bhadra, S.
Binoth, T.
Blau, M.
Bloch, I.
Bodmann, B.
Bokel, C.
Bod, T.
Boogert, S.
Boos, E.G.
Boos, H.E.
Brnsen, J.-P.
Borras, K.
Borsnyi, S.
Borunda, M.
Boscherini, D.
Bossard, A.
Botella, F.J.
Bouttier, J.
B652 (2003) 196

B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B654 (2003) 225
B658 (2003) 3
B659 (2003) 437
B658 (2003) 3
B653 (2003) 151
B659 (2003) 365
B658 (2003) 3
B659 (2003) 79
B656 (2003) 165
B658 (2003) 3
B658 (2003) 3
B660 (2003) 389
B651 (2003) 225
B658 (2003) 3
B658 (2003) 254
B660 (2003) 51
B658 (2003) 3
B651 (2003) 211
B658 (2003) 3
B654 (2003) 277
B654 (2003) 135
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 417
B657 (2003) 257
B658 (2003) 3
B660 (2003) 51
B653 (2003) 85
B658 (2003) 3
B651 (2003) 249
B651 (2003) 174
B655 (2003) 313
608
Bouwknegt, P.
Bozhilov, P.
Branco, G.C.
Branco, G.C.
Branco, G.C.
Brax, Ph.
Brock, I.
Brook, N.H.
Brower, R.C.
Brugnera, R.
Brmmer, N.
Bruni, A.
Bruni, G.
Buchbinder, E.I.
Buchbinder, I.L.
Buras, A.J.
Buras, A.J.
Burdman, G.
Bussey, P.J.
Butterworth, J.M.
Bylsma, B.
B660 (2003) 473

B656 (2003) 199
B651 (2003) 174
B657 (2003) 355
B659 (2003) 119
B660 (2003) 194
B658 (2003) 3
B658 (2003) 3
B652 (2003) 127
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B653 (2003) 400
B653 (2003) 64
B659 (2003) 3
B660 (2003) 225
B656 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
Caldwell, A.
Cao, J.
Capua, M.
Cara Romeo, G.
Cardoso, G.L.
Carena, M.
Carli, T.
Carlin, R.
Cartiglia, N.
Catterall, C.D.
Caux, J.-S.
Chankowski, P.H.
Chekanov, S.
Chernyak, V.L.
Chiarini, M.
Chim, L.
Chiochia, V.
Choudhury, D.
Chwastowski, J.
Ciborowski, J.
Ciesielski, R.
Cifarelli, L.
Cindolo, F.
Cirio, R.
Clarkson, R.
Cloth, P.
Cole, J.E.
Collins-Tooth, C.
Contin, A.
Cooper-Sarkar, A.M.
Coppola, N.
Corley, S.
B658 (2003) 3
B651 (2003) 87
B658 (2003) 3
B658 (2003) 3
B652 (2003) 5
B659 (2003) 145
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B651 (2003) 413
B659 (2003) 3
B658 (2003) 3
B660 (2003) 116
B658 (2003) 3
B660 (2003) 579
B658 (2003) 3
B660 (2003) 343
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B652 (2003) 348
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B653 (2003) 45
Cormack, C.
Corradi, M.
Corriveau, F.
Costa, M.
Crittenden, J.
Curio, G.
Cvetic, M.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B652 (2003) 5
B659 (2003) 193
DAgostini, G.
Dal Corso, F.
DallAgata, G.
Damgaard, P.H.
Danilov, P.
Dannheim, D.
Das, A.
Dawson, P.
DeBenedictis, A.
Dedes, A.
de Forcrand, P.
de Forcrand, P.
Delpine, D.
del Peso, J.
De Martino, A.
Dementiev, R.K.
Denner, A.
Denner, A.
Dent, T.
Dent, T.
De Pasquale, S.
Derrick, M.
Deshpande, A.
Devenish, R.C.E.
de Wit, B.
de Wolf, E.
Dhawan, S.
Di Francesco, P.
Dittmaier, S.
Dittmaier, S.
Dittmaier, S.
Dolgoshein, B.A.
Donagi, R.
Dorogovtsev, S.N.
Dotsenko, V.S.
Doyle, A.T.
Drews, G.
Dudas, E.
Dunne, G.V.
Durkin, L.S.
Dusini, S.
B658 (2003) 3
B658 (2003) 3
B652 (2003) 5
B656 (2003) 226
B658 (2003) 3
B658 (2003) 3
B653 (2003) 279
B660 (2003) 473
B653 (2003) 279
B657 (2003) 333
B651 (2003) 125
B655 (2003) 170
B657 (2003) 355
B658 (2003) 3
B654 (2003) 427
B658 (2003) 3
B658 (2003) 175
B660 (2003) 289
B652 (2003) 142
B653 (2003) 256
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B655 (2003) 93
B658 (2003) 3
B658 (2003) 3
B655 (2003) 313
B653 (2003) 151
B658 (2003) 175
B660 (2003) 289
B658 (2003) 3
B653 (2003) 400
B653 (2003) 307
B656 (2003) 259
B658 (2003) 3
B658 (2003) 3
B660 (2003) 3
B654 (2003) 445
B658 (2003) 3
B658 (2003) 3
Eberl, H.
Eguchi, T.
Eisenberg, Y.
Elias, V.
B657 (2003) 378

B657 (2003) 3
B658 (2003) 3
B655 (2003) 221
Ellis, J.
Ellis, J.
Engelen, J.
Engels, J.
Ermolov, P.F.
Eskola, K.J.
Eskreys, A.
Esposito, S.
Evans, T.S.
Evslin, J.
B652 (2003) 259

B659 (2003) 145
B658 (2003) 3
B655 (2003) 277
B658 (2003) 3
B660 (2003) 211
B658 (2003) 3
B658 (2003) 217
B654 (2003) 357
B657 (2003) 139
Fadin, V.S.
Fairbairn, M.
Falk, T.
Faraggi, A.E.
Farakos, K.
Fatibene, L.
Ferrando, J.
Ferrero, M.I.
Feruglio, F.
Figiel, J.
Filges, D.
Fleischer, R.
Forger, M.
Forrester, P.J.
Fortin, J.-F.
Foster, B.
Foudas, C.
Fourletov, S.
Fourletova, J.
Fox-Murphy, A.
Fredenhagen, S.
Fricke, U.
Fromme, L.
Fujimoto, J.
Fusayasu, T.
B658 (2003) 156

B653 (2003) 256
B652 (2003) 259
B659 (2003) 224
B655 (2003) 170
B652 (2003) 348
B658 (2003) 3
B658 (2003) 3
B659 (2003) 359
B658 (2003) 3
B658 (2003) 3
B659 (2003) 321
B659 (2003) 461
B660 (2003) 557
B659 (2003) 365
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B660 (2003) 436
B658 (2003) 3
B655 (2003) 277
B654 (2003) 301
B658 (2003) 3
Gabareen, A.
Gaberdiel, M.R.
Gade, R.M.
Galea, R.
Gallo, E.
Garavuso, R.S.
Gardi, E.
Garfagnini, A.
Garousi, M.R.
Garriga, J.
Gattringer, C.
Geiser, A.
Genta, C.
Georgiev, L.S.
Geyer, B.
Ghilencea, D.M.
Ghosh, P.K.
B658 (2003) 3
B654 (2003) 177
B659 (2003) 387
B658 (2003) 3
B658 (2003) 3
B659 (2003) 224
B653 (2003) 227
B658 (2003) 3
B651 (2003) 26
B655 (2003) 127
B654 (2003) 30
B658 (2003) 3
B658 (2003) 3
B651 (2003) 331
B652 (2003) 408
B653 (2003) 27
B659 (2003) 437
609
Gialas, I.
Giannakis, I.
Gilmore, J.
Ginsburg, C.M.
Giusti, P.
Gladilin, L.K.
Gladkov, D.
Glasman, C.
Gliga, S.
Goebel, F.
Goers, S.
Gogoladze, I.
Golubkov, Yu.A.
Gmez, M.E.
Gmez Nicola, A.
Gomis, J.
Gonalo, R.
Gonzlez, O.
Gttlicher, P.
Govindarajan, T.R.
Grabowska-Bod, I.
Greene, B.R.
Grena, R.
Grijpink, S.
Grinza, P.
Grisaru, M.T.
Gross, D.H.E.
Grzelak, G.
Gubser, S.S.
Guitter, E.
Gupta, K.S.
Gurrieri, S.
Gutsche, O.
Gwenlan, C.
B658 (2003) 3
B654 (2003) 197
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B660 (2003) 322
B658 (2003) 3
B659 (2003) 119
B654 (2003) 357
B659 (2003) 179
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B655 (2003) 300
B658 (2003) 3
B652 (2003) 105
B653 (2003) 204
B658 (2003) 3
B651 (2003) 387
B655 (2003) 250
B654 (2003) 427
B658 (2003) 3
B656 (2003) 23
B655 (2003) 313
B659 (2003) 437
B654 (2003) 61
B658 (2003) 3
B658 (2003) 3
Haas, T.
Haba, N.
Hahn, T.
Hain, W.
Hall-Wilton, R.
Hamatsu, R.
Hamilton, J.
Hanlon, S.
Harada, M.
Harikumar, E.
Hart, J.C.
Hartmann, H.
Hartner, G.F.
Hasenbusch, M.
Hata, H.
Heaphy, E.A.
Heath, G.P.
Heath, H.F.
Heinemeyer, S.
B658 (2003) 3
B657 (2003) 169
B652 (2003) 229
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B657 (2003) 169
B655 (2003) 300
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B659 (2003) 299
B651 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B652 (2003) 229
610
Heinrich, G.
Helbich, M.
Henkel, M.
Hernndez, P.
Heusch, C.
Hilger, E.
Hill, R.J.
Hillert, S.
Hirose, T.
Hochman, D.
Holm, U.
Honkanen, H.
Hosotani, Y.
Huang, C.-S.
Huber, P.
Hughes, V.W.
Hung, P.Q.
Hyun, S.
B654 (2003) 277

B658 (2003) 3
B660 (2003) 407
B656 (2003) 226
B658 (2003) 3
B658 (2003) 3
B657 (2003) 229
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B660 (2003) 211
B657 (2003) 169
B657 (2003) 304
B654 (2003) 3
B658 (2003) 3
B653 (2003) 123
B654 (2003) 114
Iacobucci, G.
Iannotti, L.
Iga, Y.
Inuzuka, M.
Ioannidou, T.A.
Irrgang, P.
Ishikawa, T.
Itoyama, H.
Ivanov, D.Yu.
Ivanov, E.A.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B660 (2003) 156
B658 (2003) 3
B654 (2003) 301
B657 (2003) 53
B658 (2003) 156
B653 (2003) 64
Jacob, P.
Jacobsen, J.L.
Jahn, O.
Jakob, H.-P.
Janik, R.A.
Jansen, K.
Jansen, K.
Janssen, B.
Jegerlehner, F.
Ji, X.
Ji, X.
Jones, T.W.
Joshipura, A.S.
Jurco, B.
B659 (2003) 365

B656 (2003) 259
B651 (2003) 125
B658 (2003) 3
B660 (2003) 194
B656 (2003) 226
B659 (2003) 299
B658 (2003) 281
B658 (2003) 49
B652 (2003) 383
B656 (2003) 165
B658 (2003) 3
B660 (2003) 362
B651 (2003) 45
Kagawa, S.
Kaji, H.
Kalmykov, M.Yu.
Kananov, S.
Kappes, A.
Karshon, U.
Katkov, I.I.
Kato, K.
Katz, U.F.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 49
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B654 (2003) 301
B658 (2003) 3
Kauer, N.
Kawabata, S.
Kawamura, Y.
Kazama, Y.
Kira, D.
Kersten, J.
Ketov, S.V.
Keurentjes, A.
Keurentjes, A.
Khalil, S.
Khein, L.A.
Khorrami, M.
Kim, C.L.
Kim, H.
Kim, J.Y.
Kim, Y.K.
Kind, O.
Kiritsis, E.
Kiritsis, E.
Kisielewska, D.
Kitamura, S.
Klebanov, I.R.
Kleinert, H.
Klimek, K.
Klinkhamer, F.R.
Koffeman, E.
Kohno, T.
Koike, Y.
Kolhinen, V.J.
Konishi, K.
Kooijman, P.
Koop, T.
Kopeliovich, V.B.
Korepin, V.E.
Korthals Altes, C.P.
Korzhavina, I.A.
Kostov, I.K.
Kotanski, A.
Kotsky, M.I.
Ktz, U.
Kounnas, C.
Kowal, A.M.
Kowal, M.
Kowalski, H.
Kowalski, T.
Krakauer, D.
Kramberger, G.
Krmer, M.
Kreisel, A.
Krumnack, N.
Kurihara, Y.
Kuze, M.
Kuzenko, S.M.
Kuzmin, V.A.
B654 (2003) 277

B654 (2003) 301
B657 (2003) 169
B656 (2003) 93
B658 (2003) 3
B658 (2003) 203
B656 (2003) 63
B658 (2003) 303
B658 (2003) 348
B659 (2003) 119
B658 (2003) 3
B655 (2003) 342
B658 (2003) 3
B651 (2003) 143
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B652 (2003) 165
B660 (2003) 81
B658 (2003) 3
B658 (2003) 3
B656 (2003) 23
B651 (2003) 361
B658 (2003) 3
B657 (2003) 214
B658 (2003) 3
B658 (2003) 3
B660 (2003) 269
B660 (2003) 211
B653 (2003) 204
B658 (2003) 3
B658 (2003) 3
B660 (2003) 156
B658 (2003) 417
B655 (2003) 170
B658 (2003) 3
B658 (2003) 397
B658 (2003) 3
B658 (2003) 156
B658 (2003) 3
B652 (2003) 165
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B653 (2003) 151
B658 (2003) 3
B658 (2003) 3
B654 (2003) 301
B658 (2003) 3
B660 (2003) 131
B658 (2003) 3
Labarga, L.
Labes, H.
Laine, M.
Laine, M.
Lainesse, S.
Lamberti, L.
Lammers, S.
Langmann, E.
Lazar, M.
Lee, J.H.
Lelas, D.
Lelli, S.
Lellouch, L.
Levchenko, B.B.
Levi, G.
Levman, G.M.
Levy, A.
Li, L.
Lightwood, M.S.
Lim, H.
Lim, I.T.
Lima-Santos, A.
Limentani, S.
Lindner, M.
Lindner, M.
Ling, T.Y.
Liu, J.T.
Liu, X.
Lockman, W.
Lhr, B.
Lohrmann, E.
Loizides, J.H.
Long, K.R.
Longhin, A.
Lpez, A.
Lopez-Duran Viani, A.
Louis, J.
Lowe, D.A.
Lowe, D.A.
Lozano, Y.
Lukina, O.Yu.
Lukyanov, S.
Lunin, O.
Lupi, A.
Lst, D.
B658 (2003) 3
B658 (2003) 3
B655 (2003) 170
B656 (2003) 226
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B654 (2003) 404
B652 (2003) 408
B658 (2003) 3
B658 (2003) 3
B656 (2003) 37
B656 (2003) 226
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B654 (2003) 466
B658 (2003) 3
B654 (2003) 3
B658 (2003) 203
B658 (2003) 3
B654 (2003) 197
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B651 (2003) 413
B658 (2003) 3
B654 (2003) 61
B652 (2003) 127
B653 (2003) 45
B658 (2003) 281
B658 (2003) 3
B654 (2003) 323
B655 (2003) 185
B658 (2003) 3
B652 (2003) 5
Ma, J.-P.
Maddox, E.
Maggiore, M.
Magill, S.
Majerotto, W.
Majhi, S.
Maktabdaran, G.R.
Mangano, G.
B652 (2003) 383

B658 (2003) 3
B656 (2003) 37
B658 (2003) 3
B657 (2003) 378
B660 (2003) 343
B651 (2003) 26
B658 (2003) 217
611
Mankel, R.
Mann, R.B.
Manousselis, P.
Margotti, A.
Marini, G.
Maroto, A.L.
Martelli, D.
Martn, C.P.
Martin, J.F.
Maselli, S.
Massam, T.
Mastroberardino, A.
Mastrolia, P.
Masuda, T.
Mateos, T.
Mathieu, P.
Mathur, S.D.
Matone, M.
Matsuzawa, K.
Mattingly, M.C.K.
McArthur, I.N.
McCarthy, P.J.
McInnes, B.
McKeon, D.G.C.
Melas, E.
Mellado, B.
Melzer-Pellmann, I.-A.
Menary, S.
Mendes, J.F.F.
Metlica, F.
Metsaev, R.R.
Meyer, A.
Meyer, H.B.
Micu, A.
Miele, G.
Milite, M.
Miller, D.B.
Mirea, A.
Mohammedi, N.
Monaco, V.
Moritz, M.
Moriyama, S.
Moriyama, S.
Morozov, A.
Mourad, J.
Mrenna, S.
Mck, W.
Munehisa, T.
Muramatsu, T.
Musgrave, B.
B658 (2003) 3
B652 (2003) 348
B652 (2003) 5
B658 (2003) 3
B658 (2003) 3
B653 (2003) 109
B654 (2003) 248
B652 (2003) 72
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B657 (2003) 397
B656 (2003) 325
B651 (2003) 291
B659 (2003) 365
B655 (2003) 185
B656 (2003) 78
B658 (2003) 3
B658 (2003) 3
B660 (2003) 131
B653 (2003) 369
B660 (2003) 373
B655 (2003) 221
B653 (2003) 369
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B653 (2003) 307
B658 (2003) 3
B655 (2003) 3
B658 (2003) 3
B658 (2003) 113
B654 (2003) 61
B658 (2003) 217
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B651 (2003) 249
B658 (2003) 3
B658 (2003) 3
B651 (2003) 3
B659 (2003) 179
B657 (2003) 53
B660 (2003) 3
B659 (2003) 145
B654 (2003) 248
B654 (2003) 301
B656 (2003) 93
B658 (2003) 3
Naculich, S.G.
Nagano, K.
Naganuma, M.
B651 (2003) 106

B658 (2003) 3
B652 (2003) 35
612
Nagao, T.
Nagao, T.
Nagashima, J.
Nair, V.P.
Namsoo, T.
Nania, R.
Narayan, M.
Nebot, M.
Nemoz, C.
Nesterov, D.V.
Neubert, M.
Neubert, M.
Nigro, A.
Ning, Y.
Nishigaki, S.M.
Nishimura, T.
Nitta, M.
Nobre, B.M.
Nogueira, F.S.
Nomura, Y.
Notz, D.
Nowak, R.J.
B658 (2003) 373

B660 (2003) 557
B660 (2003) 269
B651 (2003) 313
B658 (2003) 3
B658 (2003) 3
B658 (2003) 254
B651 (2003) 174
B658 (2003) 3
B654 (2003) 225
B651 (2003) 225
B657 (2003) 229
B658 (2003) 3
B658 (2003) 3
B654 (2003) 445
B658 (2003) 3
B652 (2003) 35
B657 (2003) 355
B651 (2003) 361
B656 (2003) 3
B658 (2003) 3
B658 (2003) 3
Oeckl, R.
Oh, B.Y.
Okun, L.B.
Olive, K.A.
Olkiewicz, K.
OLoughlin, M.
Ortn, T.
Ovrut, B.A.
Owen, A.W.
B657 (2003) 107

B658 (2003) 3
B656 (2003) 255
B652 (2003) 259
B658 (2003) 3
B654 (2003) 135
B651 (2003) 263
B653 (2003) 400
B651 (2003) 191
Pac, M.Y.
Padhi, S.
Paganis, S.
Palmonari, F.
Pankiewicz, A.
Papadimitriou, I.
Parenti, A.
Park, I.H.
Park, J.
Patel, S.
Patel, S.
Paul, E.
Pavel, N.
Pawlak, J.M.
Pearce, P.A.
Pelfer, P.G.
Pellegrino, A.
Penati, S.
Peroni, C.
Peschanski, R.
Pesci, A.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B657 (2003) 79
B659 (2003) 193
B658 (2003) 3
B658 (2003) 3
B659 (2003) 179
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B660 (2003) 579
B658 (2003) 3
B658 (2003) 3
B655 (2003) 250
B658 (2003) 3
B660 (2003) 194
B658 (2003) 3
Petropoulos, P.M.
Petrov, A.Yu.
Petrucci, M.C.
Picardi, I.
Pilaftsis, A.
Pisanti, O.
Plucinski, P.
Plmper, B.
Pokrovskiy, N.S.
Polini, A.
Pons, J.M.
Posocco, M.
Proskuryakov, A.S.
Przybycien, M.
Pujols, O.
B652 (2003) 165

B653 (2003) 64
B658 (2003) 3
B658 (2003) 217
B659 (2003) 145
B658 (2003) 217
B658 (2003) 3
B653 (2003) 151
B658 (2003) 3
B658 (2003) 3
B651 (2003) 291
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B655 (2003) 127
Qiu, J.
B660 (2003) 211
Raach, H.
Rago, A.
Rahimi Tabar, M.R.
Rahn, J.T.
Ratz, M.
Rautenberg, J.
Raval, A.
Ravindran, V.
Rebelo, M.N.
Redondo, I.
Reeder, D.D.
Remiddi, E.
Ren, H.-C.
Ren, Z.
Renner, R.
Repond, J.
Rigby, M.
Rindani, S.D.
Rissone, A.
Rivers, R.J.
Rizos, J.
Rizos, J.
Robaschik, D.
Roberts, R.G.
Robins, S.
Rodrigues, E.
Rosiek, J.
Ross, G.G.
Roth, M.
Ruspa, M.
B658 (2003) 3
B651 (2003) 387
B655 (2003) 342
B658 (2003) 3
B658 (2003) 203
B658 (2003) 3
B658 (2003) 3
B660 (2003) 343
B651 (2003) 174
B658 (2003) 3
B658 (2003) 3
B657 (2003) 397
B654 (2003) 197
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B660 (2003) 362
B656 (2003) 37
B654 (2003) 357
B652 (2003) 165
B660 (2003) 81
B652 (2003) 408
B653 (2003) 227
B658 (2003) 3
B658 (2003) 3
B659 (2003) 3
B653 (2003) 3
B660 (2003) 289
B658 (2003) 3
Sabetfakhri, A.
Sacchi, R.
Sadrozinski, H.F.-W.
Sakai, N.
Sakamura, Y.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B652 (2003) 35
B656 (2003) 132
Salehi, H.
Salgado, C.A.
Samtleben, H.
Samukhin, A.N.
Santachiara, R.
Santiago, J.
Santillan, O.P.
Santoso, Y.
Sartorelli, G.
Saull, P.R.B.
Savin, A.A.
Saxena, A.
Saxon, D.H.
Schaefer, S.
Schfer-Nameki, S.
Schagen, S.
Schalm, K.
Schellekens, A.N.
Schepkin, M.G.
Schioppa, M.
Schlenstedt, S.
Schmidke, W.B.
Schneekloth, U.
Schnitzer, H.J.
Schnurbusch, H.
Schupp, P.
Sciulli, F.
Seco, M.
Seiden, A.
Selonke, F.
Seniuch, M.
Serone, M.
Serreau, J.
Sezgin, E.
Shcheglova, L.M.
Shin, H.
Shiu, G.
Shiu, G.
Silva-Marcos, J.
Singh, N.N.
Sjstrand, T.
Skands, P.Z.
Skillicorn, I.O.
Slavich, P.
Sawianowska, .
Sominski, W.
Sloth, M.S.
Smalska, B.
Smilga, A.V.
Smilga, A.V.
Smirnov, F.A.
Smith, W.H.
Soares, M.
Sokolov, A.V.
B658 (2003) 3
B660 (2003) 211
B655 (2003) 93
B653 (2003) 307
B656 (2003) 259
B657 (2003) 355
B660 (2003) 169
B652 (2003) 259
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B655 (2003) 185
B658 (2003) 3
B654 (2003) 30
B654 (2003) 177
B658 (2003) 3
B652 (2003) 105
B653 (2003) 339
B656 (2003) 255
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B651 (2003) 106
B658 (2003) 3
B651 (2003) 45
B658 (2003) 3
B653 (2003) 123
B658 (2003) 3
B658 (2003) 3
B655 (2003) 277
B653 (2003) 85
B660 (2003) 51
B660 (2003) 403
B658 (2003) 3
B654 (2003) 114
B652 (2003) 105
B659 (2003) 193
B652 (2003) 142
B660 (2003) 362
B659 (2003) 243
B659 (2003) 243
B658 (2003) 3
B657 (2003) 333
B659 (2003) 3
B658 (2003) 3
B656 (2003) 239
B658 (2003) 3
B652 (2003) 93
B659 (2003) 424
B658 (2003) 417
B658 (2003) 3
B658 (2003) 3
B660 (2003) 25
613
Solano, A.
Son, D.
Sosnovtsev, V.
Sousa, N.
Spanos, V.C.
Spira, M.
Spranger, M.
Squires, A.
Staiano, A.
Stairs, D.G.
Stanco, L.
Standage, J.
Steele, T.G.
Steer, D.A.
Stefanski, B.
Stifutkin, A.
Stonjek, S.
Stopa, P.
Straub, P.B.
Strumia, A.
Suchkov, S.
Sugawara, Y.
Sundell, P.
Sundman, S.
Suppa, D.
Susinno, G.
Suszycki, L.
Sutton, M.R.
Sztuk, J.
Szuba, D.
Szuba, J.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B653 (2003) 339
B657 (2003) 378
B653 (2003) 151
B660 (2003) 225
B655 (2003) 221
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B655 (2003) 221
B654 (2003) 357
B657 (2003) 79
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B659 (2003) 359
B658 (2003) 3
B657 (2003) 3
B660 (2003) 403
B656 (2003) 344
B651 (2003) 413
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
Talavera, P.
Tan, C-I
Tanaka, H.
Tanaka, T.
Tandler, J.
Tapper, A.D.
Tassi, E.
Tawara, T.
Teixeira, A.M.
Teper, M.J.
Terras, V.
Terrn, J.
Theisen, S.
Thorn, C.B.
Tiecke, H.
Timirgaziu, C.
Tokushuku, K.
Tomaras, T.N.
Trapletti, M.
Trigiante, M.
Tsukerman, I.S.
Tsurugai, T.
B651 (2003) 291

B652 (2003) 127
B654 (2003) 301
B655 (2003) 127
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B659 (2003) 119
B658 (2003) 113
B654 (2003) 323
B658 (2003) 3
B660 (2003) 131
B652 (2003) 196
B658 (2003) 3
B660 (2003) 3
B658 (2003) 3
B660 (2003) 81
B653 (2003) 85
B655 (2003) 93
B656 (2003) 255
B658 (2003) 3
614
Tuning, N.
Turcato, M.
Tymieniecka, T.
B658 (2003)
B658 (2003)
B658 (2003)
3
3
3
Ukleja, A.
Ukleja, J.
Unterberger, J.
B658 (2003) 3
B658 (2003) 3
B660 (2003) 407
van de Ven, A.E.M.

Vasiliev, M.A.
Vasiliev, M.A.
Vzquez, M.
Velasco-Sevilla, L.
Velthuis, J.J.
Veretin, O.
Vernizzi, G.
Vettorazzo, M.
Vissani, F.
Vissani, F.
Vlachos, N.D.
Vlasov, N.N.
Voss, K.C.
Votyakov, E.V.
B657 (2003) 257

B652 (2003) 407
B655 (2003) 57
B658 (2003) 3
B653 (2003) 3
B658 (2003) 3
B658 (2003) 49
B660 (2003) 532
B655 (2003) 170
B658 (2003) 254
B659 (2003) 359
B660 (2003) 156
B658 (2003) 3
B658 (2003) 3
B654 (2003) 427
Wagner, C.E.M.
Walczak, R.
Waldram, D.
Walker, R.
Wang, K.
Wang, M.
Weber, A.
Weber, M.M.
Weiglein, G.
Weiler, A.
Wess, J.
Wessoleck, H.
B659 (2003) 145

B658 (2003) 3
B654 (2003) 61
B658 (2003) 3
B660 (2003) 322
B658 (2003) 3
B658 (2003) 3
B660 (2003) 289
B652 (2003) 229
B660 (2003) 225
B651 (2003) 45
B658 (2003) 3
Whitmore, J.J.
Wichmann, R.
Wick, K.
Wiggers, L.
Williams, D.C.
Wing, M.
Winter, W.
Winterhalder, A.
Wolf, G.
Wu, X.-H.
Wyllard, N.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B654 (2003) 3
B659 (2003) 461
B658 (2003) 3
B657 (2003) 304
B651 (2003) 106
Xiong, Z.
B651 (2003) 87
Yamada, S.
Yamaguchi, S.
Yamashita, T.
Yamazaki, Y.
Yang, J.M.
Yoshida, R.
Youngman, C.
Yuan, F.
Yuan, F.
B658 (2003) 3
B657 (2003) 3
B658 (2003) 3
B658 (2003) 3
B651 (2003) 87
B658 (2003) 3
B658 (2003) 3
B652 (2003) 383
B656 (2003) 165
Zamora Garcia, Y.
Zarnecki,
A.F.
Zawiejski, L.
Zerwas, P.M.
Zeuner, W.
ZEUS Collaboration
Zhautykov, B.O.
Zichichi, A.
Ziegler, A.
Ziegler, Ar.
Zotkin, S.A.
Zoupanos, G.
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B653 (2003) 151
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B658 (2003) 3
B652 (2003) 5

Nucl - Phys.B v.660

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Nucl - Phys.B v.660

Caricato da

Copyright:

Formati disponibili

Nuclear Physics B 660 (2003) 324

Time- and space-dependent backgrounds

Received 26 September 2002; accepted 18 March 2003

1. Introduction and summary of results

0550-3213/03/$ see front matter 2003 Published by Elsevier Science B.V.

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

2. Time- and space-dependent backgrounds of non-supersymmetric strings

(RR tadpole conditions),

T0 + T1 = 0 (uncancelled NSNS tadpoles),

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

where 10 is the ten-dimensional volume form. The eight-dimensional metric at fixed y

We define for later convenience 0 = T0 2 and 1 = T1 2 . The Einstein and dilaton

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

The boundary conditions (13) imply the relations

B(y, t) = A(y, t) + b(t),

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

where the boundary conditions (17) determine the parameters and

5b0 /60 /24

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

we get the spacetime metric

when y > 0. The Z2 identification y y is mapped in terms of the coordinates (X, T ) to

In addition, the identification of points on the circle y = y + 2R results in (T , X)

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

In this case it turns out that (22) is satisfied for

where and are determined by

5b0 /60 /24

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

As in the case (i), the Z2 identification y y is mapped in terms of the coordinates

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

(44) at the origin and at R we get the two conditions

where the boundary conditions (17) determine the parameters and

5b0 /60 /24

we get for y > 0 the spacetime metric

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

in the orbifold identification

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

where we performed the change of coordinates T = T / ch(R). The geometry of

an expanding universe with a singularity in the past at T = G0 /(3 sh(R) e ).

4. Explicit string examples

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

4.2. Other models

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

Appendix A. Equations of motion and more solutions

consider two cases: (a) = 0 and (b) G = 0.

Next, consider Eq. (8). After using (A.4), it gives

This can hold provided that

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

= 0, we can assume without loss of generality that G = 0 because an

A direct consequence of this equation is

where E is a constant. Eq. (A.7) now can be solved as

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

where the parameters and are determined here by (47).

e = e0 /24 F (y)5/6 et /6,

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

where the SO(8) characters are defined as

Next, we consider the orientifold obtained by gauging the discrete symmetry =

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

E. Dudas et al. / Nuclear Physics B 660 (2003) 324

[13] E. Dudas, J. Mourad, Phys. Lett. B 514 (2001) 173, hep-th/0012071;

T0 + T1 = 0 (uncancelled NSNS tadpoles),

(2) With the help of the N N matrices

c = Qt takes the polynomial form,

We stress that the matrix S is the same as in Theorem 1, S =

= R = Rt are differential operators of

commute each to other. Moreover the