Stability of String Configurations Dual To Quarkonium States in Ads/Cft

Nuclear Physics B 793 (2008) 133
www.elsevier.com/locate/nuclphysb
Stability of string configurations

dual to quarkonium states in AdS/CFT
Spyros D. Avramis a,b, , Konstadinos Sfetsos a , Konstadinos Siampos a
a Department of Engineering Sciences, University of Patras, 26110 Patras, Greece
b Department of Physics, National Technical University of Athens, 15773 Athens, Greece
Received 25 June 2007; accepted 24 September 2007

Available online 29 September 2007
Abstract
We extend our earlier work, regarding the perturbative stability of string configurations used for computing the interaction potential of heavy quarks within the gauge/gravity correspondence, to cover a more
general class of gravity duals. We provide results, mostly based on analytic methods and corroborated by
numerical calculations, which apply to strings in a general class of backgrounds that encompass boosted,
spinning and marginally-deformed D3-brane backgrounds. For the case of spinning branes we demonstrate
in a few examples that perturbative stability of strings may require strong conditions complementing those
following by thermodynamic stability of the dual field theories. For marginally-deformed backgrounds, we
find that even in the conformal case stability requires an upper value for the imaginary part of the deformation parameter, whereas in regions of the Coulomb branch where there exists linear confinement we
find that there exist stable string configurations for certain ranges of values of the parameter . We finally
discuss the case of open strings with fixed endpoints propagating in Rindler space, which turns out to have
an exact classical-mechanical analog.
2007 Elsevier B.V. All rights reserved.
1. Introduction
The AdS/CFT correspondence [1] maps the computation of the Wilson-loop heavy quark
antiquark potential in the planar limit of N = 4 SYM to the classical minimal-surface problem
of calculating the action of a string connecting the quark and antiquark on the boundary of AdS5
* Corresponding author at: Department of Physics, National Technical University of Athens, 15773 Athens, Greece.
E-mail addresses: avramis@mail.cern.ch (S.D. Avramis), sfetsos@upatras.gr (K. Sfetsos), ksiampos@upatras.gr

(K. Siampos).
0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.09.025
S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133
and extending into the radial direction [2]. However, when this method is applied to more general
backgrounds [36], one often encounters behaviors that are in sharp contrast with expectations
based on the gauge-theory side [4]. A possible resolution could be that the parametric regions
giving rise to these behaviors are perturbatively unstable and hence unphysical.
Let us summarize the basic relevant facts for the case of N = 4 SYM. In the standard conformal case, dual to a stack of D3-branes, this computation yields the expected Coulomb potential
and all fluctuations about the classical string configuration are found to be stable [7,8]. In extensions to the theory at finite temperature and at the Coulomb branch, dual to non-extremal and
multicenter [912] D3-branes respectively, one expects to find a screened Coulomb potential typical of thermal and Higgsed theories. However, one actually encounters situations where (i) the
potential has a second branch of higher energy than the first, (ii) the potential exhibits a confining
behavior at large distances, and (iii) the screening length is heavily dependent on the location of
the probe string in the internal space. Although these types of behavior are quite counterintuitive,
in a recent work [13] we established the reassuring result that the corresponding parametric regions represent string configurations that are perturbatively unstable, with the physical regions
giving indeed the expected screened Coulomb potential. Our analysis, partly motivated by a mechanical analog, was based on the zero-mode behavior of the differential equations governing
the small fluctuations about equilibrium and led to a formalism by means of which instabilities
may be found by exact or approximate analytic methods. An important fact emerging from this
analysis was the existence of instabilities due to fluctuations of non-cyclic angular coordinates,
often appearing in cases where they are not a priori expected. In particular, these fluctuations
cast the parametric region where confinement appears as unstable.
The analysis of [13] was restricted to diagonal metrics, where all fluctuations obey decoupled differential equations. However, there are physically interesting situations where the gravity
duals where Wilson loops are evaluated contain non-diagonal metrics. A first class of such backgrounds are boosted (non-extremal) D3-brane backgrounds, used for evaluating the potential for
a quarkantiquark pair moving with respect to the thermal medium [1420]. For this case, the potential has a similar form to that in the zero-boost case, namely it is a double-branched function of
the length with the lower branch corresponding to a screened Coulomb potential, but the screening length falls off as the velocity is increased. For this background, a numerical stability analysis
has been presented in [21], indicating that the upper branch is perturbatively unstable. A second
class of such backgrounds are spinning non-extremal D3-branes [9,2224], dual to N = 4 SYM
at finite temperature and R-charge chemical potentials. The behavior is similar to that in the absence of chemical potentials, but the nontrivial dependence of the metrics on certain angles again
calls for a stability analysis. Finally, a third class of such backgrounds are the LuninMaldacena
deformations [25] (see also [26,27]) of D3-brane backgrounds, dual to the LeighStrassler [28]
deformations of N = 4 SYM which break supersymmetry down to N = 1. These backgrounds
are characterized by the two real parameters and . In the usual situation where the quark and
antiquark are not taken to be separated in the internal space, only the -part of the deformation
affects the potential (see [29] for investigations on the effect of -deformations), resulting in
various behaviors ranging from the standard Coulomb behavior (for deformed AdS5 S5 ) to
complete screening and linear confinement [30,31] (for the deformed multicenter D3-branes). To
investigate the significance of these results, a stability analysis is again in order. We note that now
the appearance of a confining potential is not in conflict with expectations from the gauge-theory
side.
This article is organized as follows: In Section 2, we give a brief review of the evaluation of
Wilson loops in the backgrounds under consideration. In Section 3, we analyze small fluctuations
about the classical string configurations and we establish general results which allow us to determine the regions of instability using exact and approximate methods, generalizing the results
of [13] to the non-diagonal case. In Section 4, we apply these results to the gravity duals under
consideration and we identify all unstable regions. In Section 5, we summarize and conclude. In
Appendix A, we present the complete list of angular Schrdinger potentials for the backgrounds
under consideration. In Appendix B, we present the detailed solution of the equation for the
angular fluctuations for a special case, in order to provide a consistency check of the numerical calculation employed in Section 4.3.2. In Appendix C, we apply our results to the problem
of open strings in Rindler space, which turns out to have an exact classical-mechanical analog,
namely the problem of the shape of a soap film stretched between two circular rings.
2. Wilson loops in AdS/CFT
In the framework of the AdS/CFT correspondence, the calculation of the Wilson-loop potential of a heavy quarkantiquark pair proceeds by considering a fundamental string whose
endpoints lie on the two temporal sides of the Wilson loop, and which extends in the radial direction of the dual supergravity background so as to extremize its worldsheet area. This type of
calculation was first considered in [2] for the conformal case and was extended to more general
cases in [35,15,29]. Below, we give a brief review of this procedure, adapted to the case where
the metric has off-diagonal elements in the directions transverse to the quarkantiquark axis and
the B-field has nonzero components.
We consider a general background specified by a metric of the form
ds 2 = Gtt dt 2 + 2Gti dt dxi + Gij dxi dxj + Gyy dy 2 + Guu du2 + Gab da db + ,
(2.1)
where y denotes the (cyclic) coordinate along the spatial side of the Wilson loop, u denotes the
radial direction extending from the UV at u down to the IR at some minimum value umin
determined by the geometry, xi stands for a generic cyclic coordinate, and a stands for a generic
non-cyclic coordinate. We also consider a B-field of the form
B2 = Bai da dxi ,
(2.2)
as is the case with many interesting gravity duals of gauge theories [25,3134]. For future convenience, we introduce the functions
g(u, a ) = Gtt Guu ,
fy (u, a ) = Gtt Gyy ,
fij (u, a ) = Gti Gtj Gtt Gij ,
fab (u, a ) = Gtt Gab ,
h(u, a ) = Gyy Guu .

According to AdS/CFT, the potential energy of the quarkantiquark pair is given by

eiET = W (C) = eiS[C] ,
(2.3)
(2.4)
where S[C] is the action for a string propagating in the supergravity background whose endpoints
trace the contour C. The latter is given by the sum of NambuGoto and WessZumino terms,

1
1
d d
S[C] =
(2.5)
det g b ,
2
2
where g and b stand for the pullbacks

g = G x x ,
b = B x x .
(2.6)
To proceed, we employ the gauge fixing

t = ,
u = ,
(2.7)
we assume translational invariance along t, and we consider the radial embedding

y = y(u),
xi = const,
a = a0 = const,
(2.8)
supplemented by the boundary condition

L
u
= ,
2
(2.9)
appropriate for a quark placed at y = L/2 and an antiquark placed at y = L/2. In the
ansatz (2.9), the constant values a0 of the non-cyclic coordinates a must be consistent with
the corresponding equation of motion. As we shall see later on, this requires that
a g(u, a )|a =a0 = a fy (u, a )|a =a0 = 0;
.
a
(2.10)
For the above ansatz, only the NambuGoto part of the action is nonzero,1 leading to

T
du g(u) + fy (u)y 2 ,
S=
(2.11)
2
where T denotes the temporal extent of the Wilson loop, the prime denotes a derivative with
respect to u while g(u) g(u, a0 ) and fy (u) fy (u, a0 ). Conservation of the momentum
conjugate to y implies that the classical solution ycl (u) satisfies

fy0 F

,
ycl =
(2.12)
fy
where u0 is the value of u at the turning point, fy0 fy (u0 ), the two signs correspond to the two
branches around the turning point, and F is defined as
F=
gfy
.
fy fy0
(2.13)
Integrating (2.13), we express the separation length as

F
1/2
L = 2fy0
du
,
fy
(2.14)
u0
while inserting (2.12) into (2.11) and using (2.4), we obtain the potential energy
1
E=

u0
du F
du g.
(2.15)
umin
1 Nonzero contributions from the WessZumino part can arise when B and/or B are nonzero (see [6]). However,
tu
ty
this occurs in a very restricted class of backgrounds.
When the integrals (2.14) and (2.15) can be evaluated exactly and the first one can be inverted
for u0 , these equations lead to an explicit expression for E = E(L). However, in practice this
cannot be done, except for a few simple cases, and Eqs. (2.14) and (2.15) are rather regarded as
parametric equations for L and E with parameter u0 .
3. Stability analysis
We now turn to a stability analysis of these configurations just discussed, our goal being to
identify all parametric regions which are unstable and for which information obtained using the
gauge/gravity correspondence cannot be trusted. In this section we generalize the results of [13]
to non-diagonal metrics of the form (2.1) and we establish a series of results which will ultimately
allow us to identify the stable and unstable regions by a combination of exact and approximate
analytic methods.
3.1. Small fluctuations
To investigate the stability of the string configurations of interest, we consider small fluctuations about the classical solutions discussed above. In particular, we will be interested in three
types of fluctuations, namely (i) transverse fluctuations, referring to the cyclic coordinates xi
transverse to the quarkantiquark axis, (ii) longitudinal fluctuations, referring to the cyclic
coordinate y along the quarkantiquark axis, and (iii) angular fluctuations, referring to the
non-cyclic coordinates a . The above fluctuations may be parametrized by keeping the gauge
choice (2.7) unperturbed and perturbing the embedding as
xi = xi (t, u),
y = ycl (u) + y(t, u),
a = a0 + a (t, u).
(3.1)
Inserting this ansatz in the action (2.5) and expanding in powers of the fluctuations, we obtain
the series
S = S0 + S 1 + S 2 + ,
(3.2)
with the subscripts corresponding to the respective powers of the fluctuations. The zeroth-order
term gives just the classical action. The first-order contribution reads

fy0 F 1/2
hF
1
1

dt du fy0 y +
S1 =
Gti xi +
g
+
f
a
a y a ,
2
gfy
2F 1/2
2fy2
(3.3)
where a g a g(u, a )|a =a0 and a fy a fy (u, a )|a =a0 . The first term is a surface contribution which is exactly cancelled by a similar term with the opposite sign corresponding to the
contribution of the linear order action coming from the fluctuations of the lower string (cf. (2.12)),
provided we keep the variations of both strings at u = u0 equal. The second term is a total time
derivative and hence is completely irrelevant. Finally, in the third term, the coefficient of a is
just the equation of motion of a and the requirement that it vanish leads indeed to the conditions (2.10). The second-order contribution is written as

gfy
1
hF 1/2
h
1

dt du fij
x
x
y 2
y 2
S2 =
i j +
i j
1/2
3/2
2
2gfy
2F
2F
2F 1/2

hfy0
Gti y xi yx
i + Bai a xi a xi
gfy

hF 1/2
1

a b
2gfy
2F 1/2 a b

fy0 F 1/2
1
+
g
+
a b
a b y
a b ,
4F 1/2
4fy2
+ fab
(3.4)
again with all functions and their a -derivatives evaluated at a = a0 . Writing down the equations of motion of the fluctuations and introducing a harmonic time dependence,
x (t, u) = x (u)eit ,
(3.5)
we obtain the equations

1/2 f
fij d
hfy0
d
ij
2 hF
x
G
y
= 0,
j
u
ti
u
ai
a
du F 1/2 du
gfy
gfy

gfy d
hfy0
d
2 h
+
y
+
i
G
xi = 0,
u
ti
du F 3/2 du
gfy
F 1/2

1/2 f
fy0 F 1/2
1
fab d
d
ab
2 hF
g
+
f
+
b
a b
a b y
du F 1/2 du
gfy
2F 1/2
2fy2

iu Bai xi = 0
(3.6)
for the transverse, longitudinal and angular fluctuations, respectively. We see that the fluctuations
generically couple to each other, satisfying a system of equations of the form

d
du

px
0
0
0
py
0
0
0
p
2
qx
d
+ i T
du
i
i
2 qy
0
i T
0
2 q + r

x
y

= 0,
(3.7)
where the matrices px = (pij ), p = (pab ), qx = (qij ), q = (qab ), r = (rab ), = (ai ) and
the column vector = (i ) are read off from Eqs. (3.6). The problem is defined in the interval
u0 u < ,
u0 > umin ,
(3.8)
and the boundary conditions for the fluctuations, determined by requiring that the string endpoints be held fixed and that the two parts of the string glue smoothly at u = u0 read
lim xi (u) = 0,
lim y(u) = 0,
lim a (u) = 0,
lim (u u0 )1/2 xi (u) = 0,
uu+
0

lim y(u) + 2(u u0 )y (u) = 0,
uu+
0
lim (u u0 )1/2 a (u) = 0.
uu+
0
(3.9)
These boundary conditions follow from a straightforward extension of the proof given in [13]
and is essentially based on the fact that u = u0 and u = present regular singular points of
the system (3.6). Therefore, our stability analysis has reduced to a coupled system of Sturm
Liouville equations, with our objective being to determine the range of values of u0 for which
the lowest eigenvalue becomes negative.
Ideally, one would like to solve the above SturmLiouville equations exactly, obtain the lowest eigenvalue 02 in terms of the parameter u0 and determine the regions where 02 becomes
negative. However, in most cases, this is impossible due to the complexity of the equations. On
the other hand, it turns out that we can obtain useful information by studying a simpler problem,
namely the zero-mode problem of the associated differential operators. Regarding the transverse
and longitudinal fluctuations, we will use our SturmLiouville description to prove that transverse zero modes do not exist while longitudinal zero modes are in one-to-one correspondence
with the critical points of the function L(u0 ). For the angular fluctuations, we will employ an
alternative Schrdinger description which allows us to identify angular zero modes using either
exact or approximate methods.
3.2. Zero modes
To motivate the significance of zero modes for our stability analysis, we consider a Sturm
Liouville system of the form considered earlier on and we let n2 (u0 ) be the corresponding
eigenvalues. By standard results of SturmLiouville theory, these eigenvalues are real and
strictly-ordered in the sense that 02 (u0 ) < 12 (u0 ) < , from which it also follows that different eigenvalues do not cross as we vary u0 . Furthermore, we know that, for sufficiently large
values of u0 , all eigenvalues are positive. The above considerations imply that the first occurrence of instabilities will arise at a value of u0 at which the lowest eigenvalue 02 (u0 ) becomes
negative. Therefore, a necessary condition for the appearance of instabilities is the existence of
(i)
values u0c of u0 for which 02 (u0 ) = 0. To verify that these zero modes really correspond to
points where 02 changes sign, we may solve the full SturmLiouville system near each critical
point u(i)
0c , a task that may be accomplished using perturbative [13] or numerical methods. In fact,
as we shall see later on, in all examples considered in this paper, a zero mode always signifies a
change of sign of 02 , i.e., marks the boundary between a stable and an unstable region.
Restricting to zero modes simplifies our problem tremendously, as the three equations in (3.6)
decouple, reducing to

fij d
d
x
j = 0,
du F 1/2 du

gfy d
d
y = 0,
du F 3/2 du

fy0 F 1/2
d
fab d
1
(3.10)
g
+
b = 0.
a
b
a
b
y
du F 1/2 du
2F 1/2
2fy2
Using these simplified expressions, we will next prove that transverse zero modes do not exist
and that longitudinal zero modes are in one-to-one correspondence with the critical points of the
function L(u0 ), and we will devise analytic methods for seeking angular zero modes. This is a
fairly straightforward extension of the results of [13].
3.2.1. Transverse zero modes
We consider first the case of the transverse fluctuations. Using the definition of F in (2.13),
performing an integration by parts, and expanding about u = u0 , we write the general solution of
the first of (3.10) as

xi =
du
u
gfy
f 1 cj
fy fy0 ij

gfy
gfy 1
1
= 2
fy fy0 fij cj 2 du fy fy0 u
f
cj
fy
fy ij
u

= 2

du fy fy0 u

u0

gfy 1
f
cj
fy ij
g0 fy0 1
1/2
+ O(u u0 ),
fij 0 cj (u u0 )
fy0
(3.11)
where the ci are constants. In order for this zero mode to exist, it must satisfy the first boundary
condition in (3.9), which requires the coefficient of (u u0 )1/2 to vanish. As the matrix fij1
0
turns out not to admit any nontrivial null eigenvectors,2 this is only achieved when all ci = 0,
i.e., for the trivial solution xi = 0. Therefore, the transverse fluctuations do not lead to zero
modes of Eq. (3.6).
3.2.2. Longitudinal zero modes
Turning to the longitudinal fluctuations, the fact that they decouple from the rest in the zeromode analysis implies that the results of [13] for the diagonal case hold in our case as well. In
particular, in [13] it was proven that the longitudinal zero-mode solution can be expressed in
terms of the u0 -derivative of the length function as follows

fy0
g0 fy0
1/2
1/2
y 2
(3.12)
.
(u
u
)
+
L
(u
)
+
O
(u
u
)
0
0
0

3
fy0
fy0
In order for this zero mode to exist, it must satisfy the second boundary condition in (3.9), which
requires the constant term to vanish. Therefore, the solution exists only if
L (u0 ) = 0,
(3.13)
or, equivalently, if [13]

u0
du
u
fy fy0

gfy
fy

= 0,
(3.14)
which requires that the derivative term should change sign at least once as u ranges in the interval
u [u0 , ). Therefore, given the critical points of the length function, we may determine all
values u0c of u0 where the system (3.6) has a zero mode. In view of our earlier remark that
2 We have not a general proof of that statement, but this is indeed the case in all examples of the present paper as well
as of [13].
(a)
(b)
(c)
(d)
Fig. 1. The four generic types of behavior of L(u0 ) and E(L) encountered in AdS/CFT calculations of Wilson loops,
corresponding to (a) no critical points, Coulomb or screened Coulomb potential, (b) no critical points, Coulomb/confining
potential, (c) one critical point, double-branched potential and (d) two critical points, multi-branched potential. The
dashed parts of the curves indicate regions that are unstable under longitudinal perturbations.
zero modes always signify a transition from a stable to an unstable region in our examples,
the longitudinal instabilities for the various configurations of interest are as shown in Fig. 1.
Additional instabilities may arise from angular perturbations, so that part(s) of the curves, stable
under longitudinal perturbations only, might be actually unstable.
3.2.3. Angular zero modes
We finally consider angular fluctuations, which must be treated separately due to the fact that
the non-cyclicity of the angular coordinates induces a restoring force term in the corresponding
SturmLiouville equation, implying that we cannot write a in an explicit integral form, even in
the zero-mode case. On the other hand, if we restrict to situations where the angular fluctuations
are decoupled from the rest (which is indeed the case in most of our examples), the equation
satisfied by each of the fluctuations can be easily brought to a Schrdinger form. This description
allows us to determine the zero modes to quite high accuracy by using approximate methods.
Following [13], we outline the procedure below.
Letting be any of the decoupled non-cyclic angular variables, the SturmLiouville equation
for its fluctuations reads

d
d
p
r = 2 q .
(3.15)
du
du
Employing the change of variables

x=
u
du
q
,
p
= (p q )1/4 ,
we may transform Eq. (3.15) to a standard Schrdinger equation

d2
2 + V (x; u0 ) (x) = 2 (x),
dx
(3.16)
(3.17)
10
with the potential

1/4
V =
=
p d
r
+ 3/4
q
q du
p
q
1/2
d
(p q )1/4
du
r
d2
+ (p q )1/4 2 (p q )1/4 .
q
dx
(3.18)
The problem is defined in the interval

0 x x0 ,
x0 =
u0

q
du
,
p
(3.19)
which is finite, making the fact that the fluctuation spectrum is discrete manifest. The boundary
conditions are simply
(0) = 0,
(x0 ) = 0.
(3.20)
Note that, in general, Eq. (3.16) does not lead to a closed expression for u in terms of x and
therefore it is not always possible to write down the potential as an explicit function of x. In what
follows, we will encounter cases where V (x) can be determined exactly as well as cases where
it may be adequately approximated by solvable potentials.
To develop our approximate methods, we first note that the behavior of the angular
Schrdinger potential V (u; u0 ) in the limits u and u = u0 is given by
V (, u0 ) = au2 + V ,
and
V (u0 ; u0 ) =

2
h0 f0
1 g0 fy0 fy0
1 g0 2
fy0 V0 ,
+
u
0
2
2
8 h0 f0
g0 fy0
2 h0 f0
(3.21)
(3.22)
where a, V and V0 are finite quantities that depend on the parameter u0 . When a 0, the
potential expressed in terms of the variable x rises from a minimum value to infinity in the finite
interval x [0, x0 ], and hence it is reasonable to approximate it by an infinite well given by

1

V , 0 x x0
(V0 + V ), a = 0
2
Vapprox =
(3.23)
.
; V
, otherwise
a = 0
V0 ,
With the boundary conditions (3.18) the energy levels read
n2 =
(2n + 1)2 2
+ V .
4[x0 (u0 )]2
(3.24)
When a = 0, this approximation is valid for both the ground state and the excited states. When
a > 0 however, the approximation is valid only for the ground state as excited states are affected
by the details of the exact potential. In any case, we may determine the critical value u0c by
solving the equation 02 (u0 ) = 0; clearly, a solution to this equation exists only if the average V
is negative at least in a finite range of values of u0 . The infinite-well approximation just described
is to be used for obtaining an indication for the existence of zero modes and for estimating u0c .
In most cases, these estimates are quite close to the exact values, determined through numerical
analysis.
11
4. Applications
Having developed the necessary formalism, we now turn to analyzing the stability properties
of the string configurations of interest in backgrounds of boosted non-extremal D3-branes, spinning D3-branes and marginally-deformed D3-branes, all of which fall in the general category
described by the metrics (2.1). For each case under consideration, we give the formulas for the
length and the potential energy and we perform a stability analysis according to the guidelines
of the previous section. Depending on the problem at hand, the regions of stability are determined by exact or approximate analytic methods, by the numerical solution of certain algebraic
or transcendental equations, or by the numerical evaluation of certain integrals. In all cases, this
represents a considerable improvement, both conceptual as well as practical, over the direct numerical solution of the differential equations governing the fluctuations.
4.1. Boosted non-extremal D3-branes
The first type of metrics we will consider are obtained by a boost of the metric for nonextremal D3-branes along one brane direction transverse to the quarkantiquark axis, say x.
They are given by

u2
2 v4
2 v 2 4
2 4
2
2
2
ds 2 = 2 1 4 dt 2 + 2
dt
dx
+
1
+
+
dy
+
dz
dx
R
u
u4
u4

2
u
du2 + d52 ,
+ R2 4
(4.1)
u 4
where v is the boost velocity and = 1/ 1 v 2 . The metric has a horizon at

uH = ,
(4.2)
while there also exists a velocity-dependent radius u = uH beyond which the string
cannot penetrate (see [1417,19] for discussions). The Hawking temperature of the solution is
given by T = /R 2 . For the stability analysis for this metric, the transverse coordinates
are (x, z, 5 ), with x coupled to the longitudinal coordinate y, while no angular coordinates
exist. This latter fact implies that the position of the string in the internal space can be arbitrary.
The calculation of the Wilson loops of Section 2 for the above metric yields the potential for a
quark and antiquark moving with velocity v with respect to a thermal plasma at a temperature T
and may be used as a crude model for the study of meson dissociation in plasmas. Switching to
dimensionless units by the change of variables
R2
L,
E E,
we find that the length and the potential energy read [15,19]

3/2

u40 2
2 2
1 3 5 1
, , ;
,
L(u0 , ) =
2 F1
2 4 4 u40
(1/4)2
u30
u u,
and
E(u0 , ) =
u0 u0 ,
3/2

2
1 1 1 1
2
1 3 5 1
F
F
,
,
;
,
,
;
u
+
+ 1,
2 1
02 1
4 2 4 u40
2 4 4 u40
(1/4)2
u30
(4.3)
(4.4)
(4.5)
12
Fig. 2. Stability diagram in the (v, u0 ) plane for the case of the boosted non-extremal D3-brane background. The
light-shaded region corresponds to instabilities under longitudinal perturbations and the dotted curve indicates our ap
proximation (4.7) for u0c ( ). The dark-shaded region corresponds to the values u0 < which are inaccessible to the
string. The region u0 < 1 beyond the horizon has been excluded from the diagram.
where 2 F1 (a, b, c; x) is the standard hypergeometric function. The behavior of the length and
the energy is as in Fig. 1(c), with the maximal value of the length, Lc (u0 , ), being a decreasing function of the velocity and satisfying the approximate law Lc (u0 , ) 1/2 Lc (u0 , 1) [15]
indicating an enhancement of the dissociation rate of the quarkantiquark bound state with increasing velocity.
Turning to the stability analysis, our general results imply that the coupled system of the
(y, x) fluctuations has instabilities for u0 < u0c ( ) where u0c ( ) is the critical point of
L(u0 , ). The latter is determined by the solution of (3.13) which leads to the transcendental
equation

1 3 5 1
3 7 9 1
5u40 u40 3 2 2 F1 , , , 4 + 6 u40 2 2 F1 , , , 4 = 0,
(4.6)
2 4 4 u0
2 4 4 u0
which can be obtained numerically. For the physically interesting case 1 (in which we also
have u0 1) we can solve this equation perturbatively in 1/ with the result

6
1
23
u0c ( ) = 31/4 1
(4.7)
.
+
O
15 2 1350 4
The results of the stability analysis just presented are summarized in the diagram of Fig. 2.
Note that the estimate (4.7) for u0c ( ) is remarkably close to the exact numerical result even
for low velocities. Our results are in accordance with those of the numerical analysis presented
in [21].
The above analysis may be extended to the case where the velocity and the axis of the quark
antiquark pair are not perpendicular but form an angle smaller than /2. For such a case, we
may modify our ansatz (2.8) to include a variable x that has nontrivial dependence on u, while in
the special case of motion parallel to the axis we may equivalently boost along the y instead of
the x axis which would mean that the metric (2.1) would include an extra Gty term. Such string
solutions were considered in [15,17,19,20] and the behavior of the length and energy is again
13
as in Fig. 1(c). Hence we expect instabilities under longitudinal perturbations for u0 < u0c ( ),
a fact that we have actually verified by a small-fluctuation analysis which is however too lengthy
to be included here.
4.2. Spinning D3-branes
We next consider the case of spinning (non-extremal) D3-branes, dual to N = 4 SYM theory
at finite temperature and R-charge chemical potentials. These metrics were found in full generality in [9], based on previous results from [22], and their thermodynamical properties were
examined in [24]. In the conventions of [23] which we here follow, the field-theory limit of these
solutions is characterized by the non-extremality parameter and the angular momentum parameters ai , i = 1, 2, 3. Here, we will restrict to two special cases, corresponding to two equal
nonzero angular momenta, a2 = a3 = r0 , and one nonzero angular momentum, a1 = r0 , to which
we will apply our stability analysis.
4.2.1. Two equal nonzero angular momenta
For the case of two equal angular momenta, the metric reads

u4 (u2 r02 cos2 ) 2
4 H
2
1/2
2
2
2
2
ds = H
1 4 dt + dx + dy + dz + H 1/2
du
R
(u4 4 )(u2 r02 )

+ H 1/2 u2 r02 cos2 d 2 + u2 cos2 d32 + u2 r02 sin2 d12
2

2
2 r0
2
2
dt
cos
sin
d
+
cos
d
2
3 ,
R2
(4.8)
where
H=
u2 (u2
R4
r02 cos2 )
(4.9)
and d32 = d 2 + sin2 d22 + cos2 d32 . The metric has a horizon at
uH = ,
and considerations of thermodynamic stability restrict the ratio /r0 according to
1 (CE),
2 (GCE),
(4.10)
(4.11)
where the two values refer to the canonical and grand-canonical ensembles respectively, with the
former giving the lowest bounds in our examples.3 For the stability analysis for this metric, the
transverse coordinates are (x, z, , 1,2,3 ), with 2,3 coupled to the longitudinal coordinate y,
and the only angular coordinate is . The restriction (2.10) on the angular location 0 of the
string allows only trajectories with 0 = 0 and 0 = /2. To keep the discussion to a reasonable
length, we will here restrict to the 0 = /2 trajectories. For the trajectories with 0 = 0, the
longitudinal and angular fluctuations may be analyzed in the same way, using the Schrdinger
potentials given in Appendix A for the latter case.
3 For details on the thermodynamic stability of spinning branes see [24]. For a brief summary for the type of the
spinning D3-branes used in this paper see [35].
14
To examine the Wilson-loop computation, we switch to dimensionless units by setting

u r0 u,
u0 r0 u0 ,
R2
L,
r0
r0
E.
(4.12)
Then, we find that the length and the potential energy are given by the expressions

L(u0 , ) = 2

u40
4
u0
du u

,
(u2 1)(u4 u40 )(u4 4 )
(4.13)
and

E(u0 , ) =
u0
u4 4
du u
(u2 1)(u4 u40 )
uH
u
,
du
2
u 1
(4.14)
which unfortunately cannot be evaluated in closed form.4 Their behavior is as in Fig. 1(c). Turning to the stability analysis, our general results imply that the coupled system of the (y, 2,3 )
(y)
(y)
fluctuations has instabilities for u0 < u0c () where u0c () is the critical point of L(u0 , ). This
value can be obtained by numerically solving Eq. (3.14) which after some algebra takes the
simple form

du
u6 34 u2 + 24
u3 (u2 1)3 (u4 u40 )(u4 4 )
u0
= 0.
(4.15)
In order for a solution to exist, the integrand must change sign

at least once in the interval
u [u0 , ). This requirement leads to the inequality < u0 < 2 cos(/3), where cos =
(y)
1/2 , hence [/2, ]. Taking the upper bound as an equality, i.e., u0c () = 2 cos(/3),
gives a quite good approximation to the exact numerical result, as one can see by inspection of
Fig. 3.
Finally, for the angular fluctuations, the Schrdinger potential reads
V = 1,
(4.16)
implying that the infinite-well approximation is exact in this case. Therefore, the critical value
of u0 beyond which instabilities occur are obtained by using Eq. (3.24), which is in fact exact in
this case, with V = 1 and solving the equation 02 = 0. The resulting equation has the form

u0
du u3
(u2 1)(u4 u40 )(u4 4 )
,
2
(4.17)
and again can be solved numerically to give u()

0c (). The results of this stability analysis are
summarized in the diagram of Fig. 3.
4 Such integrals can be thought of as periods of Riemann surfaces corresponding to algebraic curves. In this case the
genus of the Riemann surfaces is at least two.
15
Fig. 3. Stability diagram in the (, u0 ) plane for the spinning D3-brane with two angular momenta and for the trajectory
with 0 = /2. The two light-shaded regions on the upper right-hand and lower left-hand sides correspond to instabilities
under longitudinal and angular perturbations, respectively. The dark-shaded region corresponds to values of u0 beyond
the horizon. The region to the left of the vertical dashed line corresponds to the values < 1 for which the background is
thermodynamically unstable. The dotted line corresponds to the curve drawn using the approximation described below
(4.15).
4.2.2. One nonzero angular momentum

For the case of one angular momentum, the metric reads

u2 (u2 + r02 cos2 ) 2
4 H
2
1/2
2
2
2
2
ds = H
du
1 4 dt + dx + dy + dz + H 1/2
R
u4 + r02 u2 4

+ H 1/2 u2 + r02 cos2 d 2 + u2 cos2 d32 + u2 + r02 sin2 d12

2 r0 2
2 2 sin dt d1 ,
R
(4.18)
where
H=
R4
u2 (u2 + r02 cos2 )
(4.19)
and d32 is as before. This metric has a horizon at

1
u2H = r02 + r04 + 44 ,
2
and thermodynamic stability restricts /r0 to the range
0.685
(CE),
0.93
(GCE),
(4.20)
(4.21)
for the canonical and grand-canonical ensembles respectively. For the stability analysis, the
transverse and angular coordinates are as before, but now it is 1 that couples to the longitudinal coordinate y. Again, the allowed trajectories have 0 = 0 or 0 = /2, and only the
latter case will be considered here.
16
Fig. 4. Stability diagram in the (, u0 ) plane for the spinning D3-brane with one angular momentum and for the trajectory
with 0 = /2. The light-shaded region corresponds to instabilities under longitudinal perturbations. The dark-shaded
region corresponds to values of u0 beyond the horizon. The region to the left of the vertical dashed line corresponds to
the values 0.685 for which the background is thermodynamically unstable. The dotted line corresponds to the curve
drawn using the approximation described below (4.24).
Using the same rescalings as before, we find that the expressions for the length and energy
read

L(u0 , ) = 2

u40
4
u0
du

,
4
(u4 u0 )(u4 + u2 4 )
and

du
E(u0 , ) =
u0
u4
(u4 u40 )(u4 + u2 4 )
uH
(4.22)
du
u4
u4 4
,
+ u2 4
(4.23)
which again cannot be evaluated in closed form. Their behavior is as in Fig. 1(c).
Turning to the stability analysis, the coupled system of the (y, 1 ) fluctuations has instabilities for u0 < u0c () where u0c () is the critical point of L(u0 , ), obtained by numerically
solving Eq. (3.14) which for our case reads

du
u0
u8 44 u4 44 u2 + 38

= 0.
u4 (u4 u40 )(u4 + u2 4 )3
(4.24)
This equation has a solution only if the integrand changes sign at least once in the interval u
[u0 , ). This leads to the inequality uH < u0 < umax () where umax () is the largest root of the
numerator for which we will not present an explicit expression due to its complexity. As before,
saturation of the upper bound gives a curve defined by u0c () = umax () which gives a quite
good approximation to the exact numerical result, as one can see by inspection of Fig. 4.
Finally, for the angular fluctuations, the Schrdinger potential reads
V = 1,
(4.25)
17
implying that no angular instabilities occur. The results of the stability analysis are summarized
in the diagram of Fig. 4, which is qualitatively similar to Fig. 3 apart from the absence of angular
instabilities.
4.3. Marginally-deformed D3-branes
We finally consider the LuninMaldacena deformations [25] of D3-brane solutions, dual to
the LeighStrassler [28] marginal deformations of N = 4 SYM which break supersymmetry
down to N = 1. These backgrounds are characterized by the complex parameter = +
(where is the type IIB axiondilaton and and are real parameters) which, on the gaugetheory side, represents the complex phase entering the LeighStrassler superpotential (where is
the complexified gauge coupling). In particular, we will consider marginal deformations of the
conformal AdS5 S5 background [25] and of the multicenter backgrounds corresponding to
D3-branes distributed on a sphere and on a disc [31].
4.3.1. The conformal case
Starting from the deformation of the conformal AdS5 S5 background, we rescale the deformation parameters as (, , ) R22 (, , gs ) and we write the deformed metric as
2
2

2
1/2 u
2
2
2
2
2 du
2
ds = H
(4.26)
dt + dx + dy + dz + R
+ d5, ,
R2
u2
2 is the metric on the deformed S5 , given by
where d5,

2
d5,
= d 2 + G sin2 d 12 + cos2 d 2 + G sin2 d22 + cos2 d32
+ G||2 cos4 sin2 sin2 2(d1 + d2 + d3 )2 ,

and the functions G and H are given by

G 1 = 1 + 4||2 cos2 sin2 + cos2 cos2 sin2 ,

H = 1 + 4 2 cos2 sin2 + cos2 cos2 sin2 .
(4.27)
(4.28)
The solution also includes a nonzero B-field which, in the case when the deformation parameter vanishes, is equal to
B2 = cos4 sin 2(d1 + d2 + d3 ) d,

(4.29)
2
and hence is of the form (2.2). For the stability analysis for this and all subsequent metrics,
the transverse coordinates are (x, z, 1,2,3 ), with all of them being decoupled from the longitudinal coordinate y but with the i coupled to each other, while the angular coordinates
are (, ), and are decoupled from each other. The restriction (2.10) on the angular location
(0 , 0 ) of the string allows only for the trajectories (0 , 0 ) = (0, /4), (0 , 0 ) = (0, 0 or /2)
(with bothchoices for 0 leading to equivalent results), (0 , 0 ) = (/2, any), and (0 , 0 ) =
(sin1 (1/ 3 ), /4). All trajectories but the last will also be valid for the multicenter case with
the branes distributed on the sphere and on a disc. We also note that for the trajectories with
0 = /2 the variable becomes a completely decoupled transverse coordinate and hence its
fluctuations are stable. Also, the presence of the nonzero B-field
in (4.29) implies that, for the
trajectories with (0 , 0 ) = (0, /4) and (0 , 0 ) = (sin1 (1/ 3 ), /4), the angular fluctuation and the transverse fluctuations i are actually coupled. However, the coupled terms are
18
proportional to the eigenvalue and hence do not affect at all the results of our analysis which
is based on the zero modes. For (0 , 0 ) = (0, 0 or /2), (/2, any), and i are decoupled.
For the present case, the potential energy E can be calculated explicitly as a function of L
with the result

4 2 R 2 1
E(L) = k( )
(4.30)
,
(1/4)4 L
where k( ) is an angle- and -dependent factor, given explicitly in the examples below, that becomes unity for = 0. The factor in parentheses is the result of [2], giving the standard Coulomb
behavior expected by conformal invariance. We note in passing that this factor is unaffected by
turning on -deformations, unless the quarks are given a separation in the internal deformed
sphere as well [29].
Regarding stability, the transverse/longitudinal fluctuations are obviously stable, while the
Schrdinger potentials for the angular fluctuations all have the form
V, = a, ( )u2
(4.31)
where a, ( ) are angle- and -dependent factors that can be read off the formulas of
Appendix A. The expression for the Schrdinger variable x in terms of u is

1 1 5 u4
1
x = 2 F1 , , , 04
(4.32)
u
4 2 4 u
and, accordingly, the value of the endpoint x0 is
x0 =
(1/4)2 1
.
4 2 u0
(4.33)
Instabilities may occur only when one of a, ( ) is negative. Although Eq. (4.32) cannot be
inverted in terms of u to allow for an analytic solution of the Schrdinger problem, we may obtain
a lower bound for above which instabilities definitely occur. We first note that the infinite-well
approximation is not valid here, as the potential satisfies Eq. (3.21) with a < 0. To obtain our
bound, we consider the limit u/u0 1, where x 1/u and V, a, ( )/x 2 . Then, standard
arguments from quantum mechanics [36] show that for, a, ( ) < 14 , this potential supports an
infinite tower of negative-energy states and the solution becomes unstable for all u0 . This gives
the desired bound on . The results for the allowed trajectories are as follows:
(0 , 0 ) = (0, /4). For this case we have

4 2
.
(4.34)
1 + 2
Instabilities occur only from the fluctuations for > 1/ 15 0.258. Since in the UV all
marginally deformed backgrounds approach (4.26), (4.27), the above bound is universal as long
as the corresponding trajectory remains valid. This is indeed the case for the sphere and disc
brane distributions we consider.
(0 , 0 ) = (0, 0 or /2). Here, we have
k( ) =
1 + 2,
k( ) = 1,
a ( ) =
2 2
,
1 + 2
a ( ) = a ( ) = 4 2 ,
and no angular instabilities occur.
a ( ) =
(4.35)
19
(0 , 0 ) = (/2, any). Here, we have

k( ) = 1,
a ( ) = 4 2 ,
(4.36)
and again no angular instabilities

occur.
(0 , 0 ) = (sin1 (1/ 3), /4). Here, we have
4 2
8 2
k( ) = 1 +
.
,
a ( ) = a ( ) =
3
3 + 4 2
(4.37)
Instabilities occur from both and fluctuations for > 3/28 0.327. This bound does
not survive in the multicenter cases we consider below since the corresponding trajectory is no
longer valid.
Note that the existence of an upper bound for the deformation parameter beyond which stability breaks down is reminiscent of an analogous fact for giant gravitons on the pp-wave
limit
of the deformed background (4.26), (4.27), which exist only for values of < 1/ 12. These
giant graviton solutions can be thought of as integrated perturbations around their infinitesimal
size. This implies that every order in the small size perturbative expansion of the giant graviton
solution is unstable if the above bound is not respected. The different upper value for in this
case should be attributed to the fact that the probes in the giant graviton case refer to D3-branes
and not to strings.
4.3.2. The sphere
We next consider the deformation of the background corresponding to D3-branes distributed
on a sphere of radius r0 . Switching to dimensionless units by using Eq. (4.12), we write the
deformed metric as [31]

u2 r02 cos2 2
ds 2 = H1/2 H 1/2 dt 2 + dx 2 + dy 2 + dz2 + H 1/2
du
u2 r02

+ H 1/2 u2 r02 cos2 d 2 + G u2 r02 sin2 d 12

+ H 1/2 u2 cos2 d 2 + G sin2 d22 + cos2 d32

2
2
2
2
2 2
4
1/2 || Gu (u r0 ) cos sin sin 2
2
(d1 + d2 + d3 ) ,
+H
(4.38)
u2 r02 cos2
with
G 1 = 1 + 4||2 cos2
H = 1 + 4 2 cos2
(u2 r02 ) sin2 + u2 cos2 cos2 sin2

u2 r02 cos2
(u2 r02 ) sin2 + u2 cos2 cos2 sin2

u2 r02 cos2
,
(4.39)
and H is given by (4.9). We also note that here there is a nonzero B-field which for = 0 is
proportional to the expression given in (4.29). Now, the restriction (2.10) on the (0 , 0 ) of the
string allows only for the first three trajectories considered earlier, namely (0 , 0 ) = (0, /4),
(0 , 0 ) = (0, 0 or /2) and (0 , 0 ) = (/2, any).
We next examine the potentials arising in each case in turn:
20
(0 , 0 ) = (0, /4). The length and potential energy read (see Section 6.2 of [31])
L(u0 , ) =
2u0
(u20
1
)(2u20
1+ 2
and
E(u0 , ) =
where

1 + 2

2
2
u0 + 2
1+

,
k=
1
2u20 1+
2
2u20
1
)
1+ 2

2
a , k K(k) ,

1 2
2
a K(k) E(k) + E(c)
K(c) ,
1 + 2
1 + 2

2
1
u0 1+
2

a=
,
1
2
2u0 1+ 2
1
c=
.
1 + 2
(4.40)
(4.41)
(4.42)
Note that these results are independent of the deformation parameter . The reason is that, as
remarked also in the conformal case, we have not separated the quarks in the internal deformedsphere space. Wilson-loop potentials where such a separation was considered, but with vanishing
-deformation, were computed in [29].
For = 0, the behavior of the length and energy is the same as for the undeformed case
(Fig. 1(c)). As is turned on, the length and energy curves are reminiscent of van der Waals
isotherms for a statistical system with u0 , L and E corresponding to volume, pressure and Gibbs
potential respectively (see e.g. [37]). In particular, there exists a critical value of , given by
(y)
cr 0.209 (found analytically in [31]), below which the system behaves like the statistical
system at T < Tcr (Fig. 1(d)) and above which the system behaves like the statistical system at
T > Tcr (Fig. 1(b)). For nonzero , we have the usual Coulombic behavior for large u0 (small L),
while in the opposite limit, u0 1 (large L), the asymptotics of (4.40) and (4.41) lead to the
linear potential
(4.43)
L.
2
To examine the significance of these results, it is crucial to examine the stability of the corresponding string configurations.
Our general results imply that the solution is stable under transverse perturbations. For the longi(y)
(l2)
tudinal fluctuations, the fact that L(u0 ) has two extrema u(l1)
0c ( ) and u0c ( ) for 0 < < cr
(y)
(l1)
and no extrema for > cr leads us to expect longitudinal instabilities in the region u0c ( ) <
(l2)
u0 < u0c ( ) of u0 (cf. Fig. 1(d)) in the first case, and no instabilities (cf. Fig. 1(b)) in the second case.5 To verify that the lowest eigenvalue 02 does indeed change sign at these values, we
performed a numerical analysis whose results are shown in Fig. 5(b) and indeed reproduce the
expected behavior.6 Finally, for the angular fluctuations, the relevant Schrdinger potentials are
given by the complicated expressions in Eqs. (A.13) of Appendix A. Starting from V , we find
that for < 0.71 it is positive while for > 0.71 it develops a negative part which means that
()
it can in principle support a bound state of negative energy. The critical value cr above which
E
5 For > 1/4 the derivative term in (3.14) has definite sign, so (3.14) has no solution for any value of u .
0
6 For this case, we can use a Schrdinger description for the longitudinal fluctuations and apply our infinite-well
(y)
approximation. This does indeed lead to two critical values of u0 for 0 < < cr .
(a)
(b)
21
(c)
Fig. 5. (a) Stability diagram in the (, u0 ) plane for the deformed sphere and the trajectory with (0 , 0 ) = (0, /4).
The three shaded regions correspond to instabilities due to longitudinal, and fluctuations (lightest to darkest). The
origin (0, 1) actually belongs to the stable part of the diagram. Specifically, it is stable on its own right, but nowhere
in its vicinity (for an analogous explicit example see Section 5.4 of [13]). (b) Evolution of the lowest eigenvalue of the
(y)
(y)
longitudinal fluctuations with u0 , plotted for = 0, 0.15 < cr and 0.5 > cr (bottom to top). (c) Same for the angular
( )
( )
( )
fluctuations, for = 0 < cr , 1 > cr and 10 cr (top to bottom).
()
such states can exist is determined in the infinite-well approximation to be cr 0.78, which
()
is quite close to the true value cr 0.805 determined by numerical analysis. As increases
beyond this value, the corresponding critical value u()
0c approaches the value of 1.097 as becomes very large. Turning to the fluctuations, the same reasoning used in the deformation of
the conformal background for the same trajectory carries over to the present case, implying that
()
there exists a critical value cr = 1/ 15 (within our numerical limitations) above which there
occur instabilities for all values of u0 . This value is presumably inherited by the conformal limit
of the potential as analyzed in Section 4.3.1. The above results are summarized in the stability
diagram of Fig. 5.
We also note that, in the limit 1, the potential for the fluctuations reads

V = 2 u2 2 + O 2 ,
(4.44)
and x can be explicitly determined in terms of u by (4.53). In this limit, the problem can be
()
solved exactly (see Appendix B), leading to the critical value u0c 1.09737, by mapping it to
the well studied Lam equation.
From the above we conclude that for a finite narrow range of values of the deformation parameter,
(y)
()
namely for 0.209 cr < < cr 0.258 the linear confining behavior for large distances is
stable, after which it is wiped out first by the instability of the fluctuations and subsequently
(y)
by the ones. Moreover, even for smaller values of , namely for 0 < < cr 0.209, the
confining behavior of potential is stable as we have discussed already.
(0 = 0, 0 = 0 or /2). The length and potential energy are the same as in the corresponding undeformed case, namely they are given by [4]
L(u0 ) =

2u0 k 2
k , k K(k) ,
2
u0 1
(4.45)
and
E(u0 ) =

2u20 1 k 2 K(k) E(k) + 1,
(4.46)
22
where
k=
u0
2u20
,
1
k =
1 k2.
(4.47)
Their behavior is as in Fig. 1(c).

By our general results, there exist longitudinal instabilities occurring for u0 below the critical
value for the undeformed case, namely u0c 1.125. For the angular and fluctuations,
the Schrdinger potentials are given by (A.14) of Appendix A and differ by the ones in the
undeformed case by the positive-definite term 4 2 u2 , while the expression for the Schrdinger
variable x in terms of u is the same as in the undeformed case. The above results imply that,
since the and fluctuations are stable in the undeformed case, they are stable in the present
case as well.
(0 , 0 ) = (/2, any). In this case, the length and potential energy are again the same as in
the undeformed case and are given by [4]

2
1
L(u0 ) =
(4.48)
, k K(k) ,
u0
2
and

u0
E(u0 ) = K(k) 2E(k)
2
where
k=
u20 + 1
2u20
k =

1 k2.
(4.49)
(4.50)
The behavior is as shown in Fig. 1(b). For u0 1 (small L), the behavior is Coulombic, whereas
in the opposite limit, u0 1 (large L), the asymptotics of (4.48) and (4.49) lead to the linear
potential
L
, for L 1.
(4.51)
2
In the undeformed case, the appearance of a linear confining potential is quite unexpected from
the gauge-theory side and the paradox is resolved by the stability analysis of [13] which indicates
that the configurations where this behavior appears are unstable under angular perturbations.
However, for the deformed case, where N = 4 supersymmetry is broken to N = 1, the appearance of a confining potential is expected. Remarkably, this is confirmed by the stability analysis
that follows:
We first note that the transverse and longitudinal fluctuations are manifestly stable, the latter fact
following from the absence of extrema of L(u0 ). For the angular fluctuations, the Schrdinger
potential reads

V = 1 + 4 2 u2 1 ,
(4.52)
E
whence we see that the effect of the deformation is to modify V (which, for the undeformed
case equals 1) by a positive-semidefinite term. It is quite intuitive that the addition of this term
will tend to stabilize the angular fluctuations. To see how this occurs, we note that in the present
(a)
23
(b)
Fig. 6. (a) Stability diagram in the (, u0 ) plane for the deformed sphere and the trajectory with (0 , 0 ) = (/2, any).
The shaded region corresponds to instabilities under angular perturbations. (b) Evolution of the lowest eigenvalue of the
angular fluctuations with u0 , plotted for = 0, 0.25 and 0.5 (bottom to top).
case, the Schrdinger variable x can be explicitly determined in terms of u as
2u20
2k 2 1
1
x=
,
F(, k), = sin
2
u2 + u20
(4.53)
where F(, k) is the incomplete elliptic integral of the first kind and k is the modulus defined in
(4.50). Likewise, the endpoint x0 is determined in terms of u0 by
2k 2 1
x0 = x(u0 ) =
(4.54)
K(k).
2
Then, as shown in detail in Appendix B, the equation for the angular fluctuations takes the form
(B.5) with A and h given by the second of (B.7). For general values of , this equation can be
solved only numerically and the results are as shown in Fig. 6. We verify that the critical point
u0c below which instabilities occur is a decreasing function of which implies that the region
where confinement occurs is gradually stabilized as one increases .
We finally note that for the special values 4 2 = n(n + 1), where n is a positive integer, (B.5)
reduces to the Lam equation and can be solved exactly, as done in Appendix B for the particular
case n = 1. The resulting value of u0c , given in the second of (B.14), agrees with our numerical
results.
4.3.3. The disc
We finally consider the deformation of the background corresponding to D3-branes distributed
on a disc of radius r0 , which is obtained from the deformed sphere background by the analytic
continuation r02 r02 . The possible trajectories are the same as for the sphere. We examine
them in turn:
(0 , 0 ) = (0, /4). The length and potential energy are given by (see Section 6.2 of [31])
L(u0 , ) =
2u0
(u20 +
1
)(2u20
1+ 2
1
)
1+ 2

2
a , k K(k) ,
(4.55)
24
and
E(u0 , ) =

1 + 2
where

2
2
u0 2
1+

,
k=
1
2u20 + 1+
2
2u20

1 2
1
a K(k) E(k) + E(c)
+
K(c) ,
1 + 2
1 + 2

2
u0 + 1 2
a = 2 1+1 ,
2u0 + 1+ 2
c=
.
1 + 2
(4.56)
(4.57)
For u0 1 (small L) we recover the standard Coulombic behavior enhanced by the factor
1 + 2 , while for u0 0 the asymptotics lead to the potential

E
( L)2
,
8E(i )
(4.58)
which shows that there is complete screening at the screening length

Lc = ,
(4.59)
that is invariant under -deformations.

Regarding stability, the solution is stable under transverse and longitudinal perturbations. Turning
to angular fluctuations, the relevant Schrdinger potentials are given by the complicated expressions in Eqs. (A.16) of Appendix A. Starting from the fluctuations, we use the infinite-well
approximation to find that the bottom of the well is raised with increasing and that there is a
critical value cr() > 1/ 15 above which no instabilities exist. The existence of this critical value
can be demonstrated by noting that for 1, the potential takes the form

V 2 u2 + 2 + O 2 .
(4.60)
This potential is positive and, using the techniques of Appendix B, one can then show that
perturbation theory holds about . Thus, no instabilities occur in this limit. For the
fluctuations, the same arguments used for the deformed conformal background for the same
()
trajectory show that there exists another critical value cr = 1/ 15 (again within numerical
limitations and as before inherited by the conformal limit behavior of the solution) above which
there occur instabilities for all values of u0 .
(0 , 0 ) = (0, 0 or /2). In this case we find that the classical solution is the same as in the
undeformed disc at = 0 [4]
L(u0 ) =
and
E(u0 ) =

2u0 k 2
k , k K(k)
2
u0 + 1

2u20 + 1 k 2 K(k) E(k) ,
where
k=
u0
2u20
,
+1
k =
Their behavior is as in Fig. 1(a).
1 k2,
(4.61)
(4.62)
(4.63)
25
By our general results, the solution is stable under transverse and longitudinal perturbations. For
the angular and fluctuations, the Schrdinger potentials are given by (A.17) of Appendix A
and again differ by the ones in the undeformed case by 4 2 u2 , while the expression for the
Schrdinger variable x in terms of u is the same. For the fluctuations, which have instabilities in the undeformed case, the extra term stabilizes the fluctuations and there exists critical
()
value cr above which there are no instabilities. Finally, the fluctuations are stable in the
undeformed case and hence are stable in the present case as well.
(0 , 0 ) = (/2, any). The length and potential energy are given by [4]

2u2<
u2>
, k K(k) ,
2
L(u0 ) =
(4.64)
u0 + u2>
u2 + u2
0
>
and
E(u0 ) =
where now
k=
u20
u20 + u2>
u2> u2<
,
u20 + u2>
K(k)
k =

u20 + u2> E(k) + 1,
1 k2
(4.65)
(4.66)
and u> (u< ) denotes the larger (smaller) between u0 and 1. Their behavior is as in Fig. 1(a) and,
in particular, we have a screened Coulomb potential with screening length
Lc = .
(4.67)
2
By familiar arguments, the solution is stable under transverse and longitudinal perturbations,
whose behavior is insensitive to the deformation. Regarding the angular fluctuations, the
Schrdinger potential (equal to 1 in the undeformed case) now reads

V = 1 + 4 2 u2 + 1 ,
(4.68)
and hence the solution is still stable under angular perturbations.
5. Discussion and concluding remarks
In this paper, we completed the analysis of our earlier work [13] by developing a formalism
for studying the perturbative stability of strings dual to quarkantiquark pairs in a wide class of
backgrounds with non-diagonal metrics and possibly a B-field turned on. In particular, we derived a set of results by means of which the regions where instabilities occur can be determined
by studying the zero modes of the differential equations governing the fluctuations. The simplifying fact making this extension possible is that, although the various fluctuations are generally
coupled, their zero modes actually decouple so that the problem can be treated in an analytic
manner similar to the diagonal case.
The methods developed here were applied to strings in boosted, spinning and marginallydeformed D3-brane backgrounds, dual to quarkonium states. For the case of boosted D3-branes,
we easily recovered the result that instabilities appear for the energetically unfavored long
strings stretching further into the radial direction. For the case of spinning branes, we found regions of instabilities due to both longitudinal and angular fluctuations. Finally, for the case of
26
marginally-deformed D3-branes, we found that angular fluctuations tend to completely destabilize the classical solutions for large enough values of the deformation parameter , except for
some special cases where the classical solutions are unaffected by the deformation and for which
the deformation parameter may actually have a stabilizing effect. In particular, we found parametric regions giving rise to a linear confining potential for which the dual string configurations
are stable.
As the methods developed here are completely general, we expect that they are directly applicable to other backgrounds and, in particular to other gravity duals of gauge theories with
reduced or no supersymmetry. As such backgrounds typically involve a nontrivial dependence
on certain angular coordinates, it would be particularly interesting to determine the parametric
regions for which quarkonium string configurations in these backgrounds are stable under angular perturbations. Extending our formalism to fluctuations of general brane probes should be also
possible.
Acknowledgements
K. Sfetsos and K. Siampos acknowledge support provided through the European Communitys program Constituents, Fundamental Forces and Symmetries of the Universe with contract MRTN-CT-2004-005104, the INTAS contract 03-51-6346 Strings, branes and higher-spin
gauge fields, the Greek Ministry of Education programs YAOPA with contract 89194.
K. Siampos also acknowledges support provided by the Greek State Scholarship Foundation
(IKY).
Appendix A. Angular Schrdinger potentials
Here we collect, for reference purposes, the Schrdinger potentials for angular fluctuations
for all metrics and trajectories considered in the present paper as well as in [13].
A.1. Multicenter D3-branes
The only angular coordinate is . The Schrdinger potentials for the various trajectories are
given below.
Sphere, 0 = 0:
V (u; u0 ) =
2u6 u4 + [6u20 (u20 1) 1]u2 3u20 (u20 1)

.
4u2 (u2 1)2
(A.1)
Sphere, 0 = /2:
V (u; u0 ) = 1.
(A.2)
Disc, 0 = 0:
V (u; u0 ) =
2u6 + u4 + [6u20 (u20 + 1) 1]u2 + 3u20 (u20 + 1)

.
4u2 (u2 + 1)2
(A.3)
Disc, 0 = /2:
V (u; u0 ) = 1.
(A.4)
27
A.2. Spinning D3-branes

The only angular coordinate is as before. The Schrdinger potentials are given below.
Two equal angular momenta, 0 = 0:
V (u; u0 , ) =
1
4u6 (u2
1)2
10

2u u8 + 6 u20 u20 1 + 4 1 u6

3u20 u20 1 + 114 u4 5 2u20 u20 1 1 4 u2

+ 7u20 u20 1 4 .
(A.5)
Two equal angular momenta, 0 = /2:

V (u; u0 , ) = 1.
(A.6)
One angular momentum, 0 = 0:

V (u; u0 , ) =
1
4u4 (u2
+ 1)2
10

2u + 3u8 + 6 u20 u20 + 1 + 4 1 u6

+ 9 u20 u20 + 1 + 4 1 u4 + 3 104 u20 u20 + 1 + 34 u2

5u20 u20 + 1 4 .
(A.7)
One angular momentum, 0 = /2:
V (u; u0 , ) = 1.
(A.8)
A.3. Deformed D3-branes

Now, the angular coordinates are and , with becoming irrelevant for the trajectories
with 0 = /2. The Schrdinger potentials are as follows.
Conformal, (0 , 0 ) = (0, /4):
V (u; u0 , ) =
2 2 2
u ,
1 + 2
V (u; u0 , ) =
4 2 2
u .
1 + 2
(A.9)
Conformal, (0 , 0 ) = (0, 0 or /2):

V (u; u0 , ) = V (u; u0 , ) = 4 2 u2 .
(A.10)
Conformal, (0 , 0 ) = (/2, any):

V (u; u0 , ) = 4 2 u2 .
Conformal, (0 , 0 ) = (sin1 (1/ 3 ), /4):

V (u; u0 , ) = V (u; u0 , ) =
8 2
u2 .
3 + 4 2
(A.11)
(A.12)
28
Sphere, (0 , 0 ) = (0, /4):

2

8 1 + 2 u10 + 2 1 + 2 1 15 2 + 8 4 u8

3 41 2 36 4 u6 + 3 2 1 + 2 u20 1 + 2 u20 1 7 2 u4

9 + 8 2 1 + 2 u20 1 1 u2 + 3u20 1 + 2 u20 1 ,

1
2
2 12
16
1
+
u
V (u; u0 , ) = 2 2
4u (u 1)[(1 + 2 )u2 1]3

2 1 + 2 1 + 35 2 + 18 4 u10 + 5 + 103 2 + 94 4 + 12 6 u8

+ 6 1 + 2 1 + 2 2 u20 1 + 2 u20 1 3 54 2 20 4 u6

13 + 18 2 + 4 4 u20 1 + 2 u20 1 + 1 7 2 u4

+ 8u20 1 + 2 u20 1 + 1 u2 u20 1 + 2 u20 1 .
(A.13)
V (u; u0 , ) =
4u2 [(1 + 2 )u2
1]3
Sphere, (0 , 0 ) = (0, 0 or /2):

V (u; u0 , ) = V (u; u0 ) + 4 2 u2 ,
V (u; u0 , ) = V (u; u0 ) + 4 2 u2 ,
(A.14)
where V (u; u0 ) and V (u; u0 ) are the potentials for the corresponding undeformed case, with
the former given by (A.1) and the latter given by a function which, since the fluctuations in
the multicenter case are stable, admits no negative-energy bound states.
Sphere, (0 , 0 ) = (/2, any):

V (u; u0 , ) = 1 + 4 2 u2 1 .
(A.15)
Disc, (0 , 0 ) = (0, /4):
2

8 1 + 2 u10 2 1 + 2 1 15 2 8 4 u8

3 41 2 36 4 u6 3 2 1 + 2 u20 1 + 2 u20 + 1 7 2 u4

9 + 8 2 1 + 2 u20 + 1 1 u2 3u20 1 + 2 u20 1 ,

1
16 2 1 + 2 u12
V (u; u0 , ) = 2 2
4u (u + 1)[(1 + 2 )u2 + 1]3

+ 2 1 + 2 1 + 35 2 + 18 4 u10 + 5 + 103 2 + 94 4 + 12 6 u8

6 1 + 2 1 + 2 2 u20 1 + 2 u20 + 1 3 54 2 20 4 u6

13 + 18 2 + 4 4 u20 1 + 2 u20 + 1 + 1 7 2 u4

8u20 1 + 2 u20 + 1 + 1 u2 u20 1 + 2 u20 + 1 .
(A.16)
V (u; u0 , ) =
4u2 [(1 + 2 )u2
+ 1]3
Disc, (0 , 0 ) = (0, 0 or /2):

V (u; u0 , ) = V (u; u0 ) + 4 2 u2 ,
V (u; u0 , ) = V (u; u0 ) + 4 2 u2 ,
(A.17)
where V (u; u0 ) and V (u; u0 ) are the potentials for the undeformed case, with the former given
by (A.3) and the latter by a function admitting no negative-energy bound states.
Disc, (0 , 0 ) = (/2, any):

V (u; u0 , ) = 1 + 4 2 u2 + 1 .
(A.18)
29
Appendix B. Angular fluctuations and the Lam equation

For the special cases of (i) the deformed sphere at 0 = 0 in the limit 1 and (ii) the
deformed sphere at 0 = /2 (for any ), the Schrdinger equation for the angular fluctuations
takes a relatively simple form that allows for explicit solutions. In both cases, the Schrdinger
variable x is given in closed form in terms of u by Eq. (4.53), which can be inverted for u to
yield

1
1
1
= 2u20 ( 2u0 x) + 2 ,
u2 = 2u20
(B.1)
6u0
sn2 ( 2u0 x, k) 2
where sn(z, k) is the elliptic Jacobi function and (z) (z | 1 , 2 ) (z, g2 , g3 ) is the
Weierstrass p-function, specified by the half-periods
2 = iK(k ),
1 = K(k),
k2 =
1 + u20
2u20
(B.2)
or, equivalently, by the elliptic invariants

g2 = 4(e1 e2 + e2 e3 + e3 e1 ),
g3 = 4e1 e2 e3 ,
(B.3)
with the ei given by

e1 =
3u20 1
6u20
e2 =
1
,
3u20
e3 =
3u20 + 1
6u20
(B.4)
and satisfying e1 + e2 + e3 = 0. Substituting (B.1) into (4.60) and (4.52) to obtain explicit
expressions for the potentials in terms of x, and making a further change of variables to z = 2u0 x,
we arrive at the following equation

2
d2
2 + A (z) + h (z) = 2 (z), 0 z 1 ,
(B.5)
dz
2u0
subject to the boundary conditions
(0) = 0,
(1 ) = 0,
(B.6)
and to the requirement that (z) be real. Here, A and h stand for the constants
(i)
A = 2,
(ii)
A = 4 2 ,
h=
5
,
3u20
h=
3 + 4 2
.
6u20
(B.7)
In the special cases where A = n(n + 1) (with n being a positive integer) Eq. (B.5) is just the
Lam equation, whose exact solutions are well known (see e.g. [38]). Although the construction
of the specific solutions satisfying the boundary conditions (B.6) is rather cumbersome, it is
instructive to consider the simplest possible case, n = 1, as a consistency check of our numerics.
For this case, the zero-mode equation simplifies to

d2
2 + 2 (z) + h (z) = 0, 0 z 1 .
(B.8)
dz
30
For the analysis that follows, it is convenient to define the quantities

a 1 (h),
3 1 + 2 .
(B.9)
When a = i , i = 1, 2, 3, Eq. (B.8) possesses the linearly-independent solutions

(z a)
,
(z) = exp z (a)
(z)
(B.10)
where (z) (z | 1 , 2 ) and (z) (z | 1 , 2 ) are the Weierstrass zeta and sigma functions.7 When a = i , the linearly-independent solutions are instead

(z + i )
,
i,1 (z) = exp z (i )
(z)

i,2 (z) = (z + i ) + zei i,1 (z).
(B.11)
For these two cases, the general solution reads (z) = C+ + (z) + C (z) and (z) =
Ci,1 i,1 (z) + Ci,2 i,2 (z), respectively. Imposing the condition (0) = 0, we obtain C+ = C
and Ci,1 = (i )Ci,2 , respectively, while using the reality condition for (z) we find that
a must be real. For the case a = i , the condition (1 ) = 0 leads to the equation

(1 a) (a) (1 ) + (1 a) exp 2 a (1 ) 1 (a) 1 = 0,
(B.12)
which is to be solved for a in the square {0 Re a < 21 , 0 Im a < 2|2 |} in the complex
plane, due to the periodicity of the Weierstrass functions under a a +2m1 +2n2 . The trivial
zeros of (B.12) occur at the points a = i , i = 1, 2, 3, where either the first or the second factor
in curly brackets vanishes. However, for these points, we actually have to use the second set of
linearly-independent solutions (B.11). Doing so, we find that the values a = 2,3 are not allowed
due to the reality condition, while for the value a = 1 the boundary condition (1 ) = 0 leads
to the equation 3 (1 ) = e1 1 , which has no solution. Nontrivial roots of Eq. (B.12) can only
arise from the second factor in curly brackets, i.e., they are given by the solutions of the equation
a (1 ) = 1 (a),
(B.13)
subject to the condition a = 1 . Using Eqs. (B.7), (B.2) and (B.9), we can express this as a
transcendental equation for u0c , whose solutions for the two cases of interest are
(i)
u0c 1.09737,
Lc = 1.9312,
(ii)
u0c 1.02676,
Lc = 2.92397,
for 4 2 = 2.
(B.14)
Both results are in precise agreement with our numerical analysis.

Appendix C. Soap films and strings in Rindler space
An example of the calculations considered here and in [13] has been presented recently in [39]
and refers to open strings ending on branes in Rindler space and their stability properties. As
is well known, Rindler space is (a portion of) flat Minkowski space, expressed in coordinates
7 Some properties of these functions, used in the derivation of (B.12), are
(z) = (z),
(z) = (z) (z),
(z + 21,2 ) = (z) + 2 (1,2 ),

(z + 21,2 ) = exp 2(z + 1,2 ) (1,2 ) (z).
31
adapted to an observer at constant acceleration. Such an observer experiences the Unruh effect,
perceiving the inertial vacuum as a state populated by a thermal distribution of particles at a temperature T = /2 , where is the surface gravity. It is then natural to expect that an open string
with fixed endpoints propagating in Rindler space will have similar properties as one propagating
in a black-hole background and, in particular, will exhibit a phase structure of the type shown
in Fig. 1(c). In this appendix we prove the amusing fact that this problem is exactly equivalent
to the classical-mechanical problem of the shape of a soap film suspended between two circular
rings.
The metric for Rindler space has the form
ds 2 = 2 u2 dt 2 + dy 2 + du2 + ,
(C.1)
where u is the radial direction with the Rindler horizon corresponding to u = 0 and y is a generic
spatial direction. The string configuration of interest corresponds to an open string with its two
endpoints located at the same radius u = and separated by a distance L along the y direction.
It is obvious that the formalism of Section 2 readily applies to this setup. Passing to convenient
dimensionless units through the rescalings
2
E,
(C.2)
4 2
inserting the metric components of (C.1) in (2.12), and integrating from u0 to , we find that the
classical solution is the catenary curve of Leibniz, Huygens and Bernoulli,
y
u = u0 cosh ,
(C.3)
u0
u0 u0 ,
u u,
L L,
which is a slice of Eulers catenoid. The integration constant u0 and the energy of the string are
determined by Eqs. (2.12) and (2.13) (without the subtraction term) which for our case read
1

L = 2u0
u0
du
u2 u20
= 2u0 cosh1
1
,
u0
(C.4)
and
1
E = 4
u0
du u2

= 2
u2 u20

1 u20
+ u20 cosh1

1
,
u0
(C.5)
respectively. Eqs. (C.3)(C.5) are exactly the same formulas appearing in Plateaus problem for
a thin soap film of mass density 1 and surface tension 12 stretched between two coaxial rings
of radius 1 that are separated by a distance L, with u0 being the minimal radius reached by
the film (see Eqs. (A.4)(A.5) in Appendix A of [13]), and thus the properties of the solutions
(see e.g. [40]) are the same. Namely, there exists a critical value u0c of u0 and a corresponding
value Lc of L at which L (u0 ) = 0, i.e.,

1
(C.6)
1 u20 cosh1
= 1.
u0
Solving this equation, we find u0c 0.552, whence Lc 1.325 (to compare with the results
of [39], set u0c L/ h). For separations L < Lc , there is a short string (dual to the shallow
catenoid), a long string of higher energy (dual to the deep catenoid), and a configuration of
32
two straight strings (dual to the Goldschmidt solution). For separations L > Lc , only the last
configuration is possible.
Regarding stability, the only relevant fluctuations are the longitudinal fluctuations along y.
By the results of Section 2, instabilities occur for u0 below the critical value where L (u0 ) = 0,
i.e., for u0 < u0c . This can be verified directly by noting that the longitudinal SturmLiouville
equation has the form

u2 + 2u20
2
u2 u20 y (u) +
y (u) = 2 y(u).
u
The zero-mode solution satisfying the boundary condition y(1) = 0 is given by

y(u) cosh1
u
1
u
1

cosh1
+
,
u0
u
0
u2 u20
1 u20
(C.7)
(C.8)
and imposing the boundary condition limuu+ [y(u) + 2(u u0 )y (u)] = 0 leads indeed to
0
Eq. (C.6). Alternatively, one may set x = cosh1
u
u0
and = tanh xy to transform (C.7) to
a Schrdinger equation with the PschlTeller type II potential V =
2
cosh2 x
and the energy
, defined in the interval x [1, cosh1 u10 ] and obeying Neumann and Dirichlet boundary
conditions at the left and right endpoint respectively.8 The solutions are the associated Legendre
functions P1 (tanh x) and Q1 (tanh x), where = i
, and imposing the boundary conditions on
the zero-mode solutions leads again to Eq. (C.6). In both formulations, it is very straightforward
to check that the lowest eigenvalue really changes sign from positive to negative as u0 crosses
u0c from the right. In conclusion, the long strings with u0 < u0c are perturbatively unstable in
the region where they exist.
The qualitative analogy of the soap-film problem with AdS/CFT calculations of Wilson loops
has already been considered in previous work [13,41]. It is a pleasant surprise that this problem
actually has an exact analog in the context of the NambuGoto string.
2
2
References
[1] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 505, hep-th/9803131;
O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183, hep-th/9905111.
[2] J.M. Maldacena, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002;
S.J. Rey, J.T. Yee, Eur. Phys. J. C 22 (2001) 379, hep-th/9803001.
[3] S.J. Rey, S. Theisen, J.T. Yee, Nucl. Phys. B 527 (1998) 171, hep-th/9803135;
A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, Phys. Lett. B 434 (1998) 36, hep-th/9803137;
A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, JHEP 9806 (1998) 001, hep-th/9803263.
[4] A. Brandhuber, K. Sfetsos, Adv. Theor. Math. Phys. 3 (1999) 851, hep-th/9906201.
[5] Y. Kinar, E. Schreiber, J. Sonnenschein, Nucl. Phys. B 566 (2000) 103, hep-th/9811192;
Y. Kinar, E. Schreiber, J. Sonnenschein, N. Weiss, Nucl. Phys. B 583 (2000) 76, hep-th/9911123.
[6] J. Sonnenschein, What does the string/gauge correspondence teach us about Wilson loops?, hep-th/0003032.
[7] C.G. Callan, A. Guijosa, Nucl. Phys. B 565 (2000) 157, hep-th/9906153.
8 This is different than the corresponding Schrdinger equation for the fluctuations of the thin soap film problem (see
[13,42]).
[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]
33
I.R. Klebanov, J.M. Maldacena, C.B. Thorn, JHEP 0604 (2006) 024, hep-th/0602255.
P. Kraus, F. Larsen, S.P. Trivedi, JHEP 9903 (1999) 003, hep-th/9811120.
K. Sfetsos, JHEP 9901 (1999) 015, hep-th/9811167.
D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, JHEP 0007 (2000) 038, hep-th/9906194.
I. Bakas, K. Sfetsos, Nucl. Phys. B 573 (2000) 768, hep-th/9909041;
I. Bakas, A. Brandhuber, K. Sfetsos, Adv. Theor. Math. Phys. 3 (1999) 1657, hep-th/9912132;
M. Cvetic, S.S. Gubser, H. L, C.N. Pope, Phys. Rev. D 62 (2000) 086003, hep-th/9909121.
S.D. Avramis, K. Sfetsos, K. Siampos, Nucl. Phys. B 769 (2007) 44, hep-th/0612139.
K. Peeters, J. Sonnenschein, M. Zamaklar, Phys. Rev. D 74 (2006) 106008, hep-th/0606195.
H. Liu, K. Rajagopal, U.A. Wiedemann, An AdS/CFT calculation of screening in a hot wind, hep-ph/0607062;
H. Liu, K. Rajagopal, U.A. Wiedemann, JHEP 0703 (2007) 066, hep-ph/0612168.
M. Chernicoff, J.A. Garcia, A. Guijosa, JHEP 0609 (2006) 068, hep-th/0607089.
P.C. Argyres, M. Edalati, J.F. Vzquez-Poritz, JHEP 0701 (2007) 105, hep-th/0608118.
E. Cceres, M. Natsuume, T. Okamura, JHEP 0610 (2006) 011, hep-th/0607233.
S.D. Avramis, K. Sfetsos, D. Zoakos, Phys. Rev. D 75 (2007) 025009, hep-th/0609079.
M. Natsuume, T. Okamura, Screening length and the direction of plasma winds, 0706.0086 [hep-th].
J.J. Friess, S.S. Gubser, G. Michalogiorgakis, S.S. Pufu, JHEP 0704 (2007) 079, hep-th/0609137.
M. Cvetic, D. Youm, Nucl. Phys. B 477 (1996) 449, hep-th/9605051.
J.G. Russo, K. Sfetsos, Adv. Theor. Math. Phys. 3 (1999) 131, hep-th/9901056.
S.S. Gubser, Nucl. Phys. B 551 (1999) 667, hep-th/9810225;
R.G. Cai, K.S. Soh, Mod. Phys. Lett. A 14 (1999) 1895, hep-th/9812121;
M. Cvetic, S.S. Gubser, JHEP 9907 (1999) 010, hep-th/9903132;
T. Harmark, N.A. Obers, JHEP 0001 (2000) 008, hep-th/9910036.
O. Lunin, J. Maldacena, JHEP 0505 (2005) 033, hep-th/0502086.
S.A. Frolov, R. Roiban, A.A. Tseytlin, JHEP 0507 (2005) 045, hep-th/0503192;
S. Frolov, JHEP 0505 (2005) 069, hep-th/0503201.
R.C. Rashkov, K.S. Viswanathan, Y. Yang, Phys. Rev. D 72 (2005) 106008, hep-th/0509058;
C.h. Ahn, J.F. Vzquez-Poritz, JHEP 0507 (2005) 032, hep-th/0505168.
R.G. Leigh, M.J. Strassler, Nucl. Phys. B 447 (1995) 95, hep-th/9503121.
R. Hernndez, K. Sfetsos, D. Zoakos, JHEP 0603 (2006) 069, hep-th/0510132;
R. Hernndez, K. Sfetsos, D. Zoakos, Fortschr. Phys. 54 (2006) 407, hep-th/0512158.
C. Ahn, J.F. Vzquez-Poritz, JHEP 0606 (2006) 061, hep-th/0603142.
S.D. Avramis, K. Sfetsos, D. Zoakos, Complex marginal deformations of D3-brane geometries, their Penrose limits
and giant gravitons, 0704.2067 [hep-th].
I.R. Klebanov, A.A. Tseytlin, Nucl. Phys. B 578 (2000) 123, hep-th/0002159.
I.R. Klebanov, M.J. Strassler, JHEP 0008 (2000) 052, hep-th/0007191.
J.M. Maldacena, C. Nez, Phys. Rev. Lett. 86 (2001) 588, hep-th/0008001.
S.D. Avramis, K. Sfetsos, JHEP 0701 (2007) 065, hep-th/0606190.
L.D. Landau, E.M. Lifshitz, Quantum Mechanics (Non-relativistic Theory), Pergamon, London, 1959.
H.B. Callen, Thermodynamics and Introduction to Thermostatistics, second ed., John Wiley & Sons, New York,
1985.
E.L. Ince, Ordinary Differential Equations, Dover Publications, New York, 1956;
E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, Cambridge Univ. Press, Cambridge, 1986.
D. Berenstein, H.J. Chung, Aspects of open strings in Rindler space, 0705.3110 [hep-th].
C. Caratheodory, Calculus of Variations and Partial Differential Equations of the First Order, Amer. Math. Soc.,
1999;
G.A. Bliss, Calculus of Variations, Open Court, Chicago, IL, 1925;
R. Courant, H. Robbins, What is Mathematics?, Oxford Univ. Press, Oxford, 1941;
G.B. Arfken, H.J. Weber, Mathematical Methods for Physicists, sixth ed., Elsevier Academic Press, 2005.
D.J. Gross, H. Ooguri, Phys. Rev. D 58 (1998) 106002, hep-th/9805129.
L. Durand, Am. J. Phys. 49 (1981) 334.

Quaternionic Khler manifolds, constrained instantons

and the magic square: I
Keshav Dasgupta a, , Vronique Hussin b , Alisha Wissanji c
a Rutherford Physics Building, McGill University, Montral, QC H3A 2T8, Canada
b Dpartement de Mathmatiques et Statistique, UdeM, Montral, QC H3C 3J7, Canada
c Centre de Recherches Mathmatiques, UdeM, Montral, QC H3C 3J7, Canada
Received 21 August 2007; accepted 24 September 2007

Abstract
The classification of homogeneous quaternionic manifolds has been done by Alekseevskii, Wolf et al.
using transitive solvable group of isometries. These manifolds are not generically symmetric, but there is a
subset of quaternionic manifolds that are symmetric and Einstein. A further subset of these manifolds are
the magic square manifolds. We show that all the symmetric quaternionic manifolds including the magic
square can be succinctly classified by constrained instantons. These instantons are mostly semilocal, and
their constructions for the magic square can be done from the corresponding SeibergWitten curves for
certain N = 2 gauge theories that are in general not asymptotically free. Using these, we give possible
constructions, such as the classical moduli space metrics, of constrained instantons with exceptional global
symmetries. We also discuss the possibility of realising the Khler manifolds in the magic square using
other solitonic configurations in the theory, and point out an interesting new sequence of these manifolds in
the magic square.
PACS: 11.25.-w; 11.15.-q; 11.25.-j
Keywords: Strings; Branes; Quaternionic manifolds; Gauge theories
* Corresponding author.
E-mail addresses: keshav@hep.physics.mcgill.ca (K. Dasgupta), hussin@dms.umontreal.ca (V. Hussin),

wissanji@crm.umontreal.ca (A. Wissanji).
doi:10.1016/j.nuclphysb.2007.09.026
K. Dasgupta et al. / Nuclear Physics B 793 (2008) 3482
35
1. Introduction
A Riemannian manifold (M, g) is a smooth manifold M endowed with a metric g defined in
T M. The holonomy of such a connected oriented Riemannian manifold belongs to the following list:
SO(n): generic case.

SU(n), U (n) SO(2n): CalabiYau and Khler cases.
Sp(n), Sp(n) Sp(1) SO(4n): Hyper-Khler and quaternionic Khler cases.
G2 SO(7), Spin(7) SO(8).
The above is the so-called Bergers classification theorem [1]. We will be mainly concerned
with the following two holonomies: Sp(n) and Sp(n) Sp(1). Both these groups act on Hn =
R4n where Hn is the right vector space over the quaternions H. The Sp(1) SU(2) factor in
Sp(n) Sp(1) is the group of unit quaternions acting from the right.
The quaternionic Khler manifolds are always Einstein1 for n 2 and are self-dual Einstein
for n = 1. They are considered positive if their metrics are complete and have positive scalar
curvatures. When the scalar curvatures are zero, then the holonomies of these manifolds reduce
to Sp(n) and are called the hyper-Khler manifolds. Thus clearly quaternionic Khler manifolds
are not Ricci flat.
Examples of quaternionic Khler manifolds with positive scalar curvatures are given by compact symmetric spaces classified by Wolf [2] and Alekseevskii [3] and are known as the Wolf
spaces. They are classified by taking centerless Lie group G which form the isometry group of
quaternionic Khler spaces given as conjugacy classes of Sp(1) in G determined by the highest
root of G. These spaces are:

SU(n + 2)
Sp(n + 1)
,
Gr2 Cn+2 =
,
Sp(n) Sp(1)
S(U (n) U (2))

SO(n + 4)
Gr4 Rn+4 =
,
SO(n) SO(4)
E7
E8
E6
,
,
,
SU (6) Sp(1)
Spin(12) Sp(1)
E7 Sp(1)
G2
F4
,
.
Sp(3) Sp(1)
SO(4)
HPn =
(1.1)
Observe that all these spaces are modded by a Sp(1) group as expected. This will be useful later
when we will map our configurations to semi-local defects.
The above examples are all compact. The non-compact duals are symmetric examples of
quaternionic Khler manifolds with negative scalar curvatures. The non-symmetric, non-compact
examples with negative scalar curvatures are also known. However no concrete examples of noncompact non-symmetric positive curvature manifolds are presently known.
In Section 2 we will give some examples of symmetric quaternionic Khler manifolds that
appear in string theory. We will study few representative casesin Sections 2.1 and 2.2and
discuss possible quantum corrections to these spaces. Although most of this is well known, we
will present it in a way so as to connect to latter parts of the paper. Important concepts like c, s
1 By this we mean that the Ricci tensor is proportional to the metric.
36
and r-maps will be introduced in Section 2.2. The connection between c and r-maps, as we will
discuss soon, is the following:
r
Real manifold Khler manifold Quaternionic Khler manifold,

which in the language of supergravity means the following: the moduli space of the scalar fields
in the vector multiplets for a five-dimensional supergravity is a real manifold. Dimensionally
reducing this to four dimension yields a Khler moduli space for the vector multiplets and further
dimensional reduction to three dimensions yields a quaternionic Khler manifold for the hypermultiplets. This way of viewing the connection was described by various authors, for example
[46], which also led to the connection to the magic square of Freudenthal, Rosenfeld and Tits
[7] that we describe at the beginning of Section 4.
Our method of studying the magic square and classifying the quaternionic manifolds is different from what has been attempted so far. We will not analyse using supergravities at all,
instead we will describe the whole system via SU(2) gauge theories with global symmetries G
that resemble sectors of N = 2 SeibergWitten theories [8] in certain parametrisations, but are
not asymptotically free. Most of these theories that we analyse are at strong couplings, and in
certain cases simple YangMills description may not suffice. Nevertheless we will show that
one-instanton moduli spaces could be studied in all these cases, and the corresponding Seiberg
Witten curves could be used to classify the quaternionic spaces. The instantons that we study are
not only constrained instantons [9], but are also semilocal [10].2 The Khler3 and the real spaces
could then be classified by other semilocal defects in the theory for certain choices of global symmetries that we analyse using the so-called sequential gauging. These aspects will be described
in Sections 3 and 4. In Sections 4.1 to 4.4, we will give strong evidence that all the elements of the
magic square [7] can be reproduced starting from certain sectors of N = 2 SU(2) gauge theories
with E6 , E7 , E8 and F4 global symmetries. The case with G2 global symmetry is interesting, and
we study this in Section 3.2 by detailing an explicit construction of the associated quaternionic
space. Normally one would not attach G2 to the magic square, but we show that there is a way
to incorporate the G2 group sequence in the magic square too by adding one extra column.
In Section 4.5 we study another example that has not been discussed in the physics literature
in details. This new sequencing of the magic square follows rather straightforwardly from our
arguments of sequential gauging and could also be added to the magic square by a different
choice of the underlying Jordan algebras [12].
In Section 4.5.2 we discuss the sigma model descriptions of these quaternionic spaces by
analysing the F -functions [5] for all the relevant cases. These F -functions are the prepotential
that determine the Khler spaces associated to the quaternionic spaces. We then use the c-map to
determine metrics of all the quaternionic spaces. Finally, in Section 5 we conclude with a brief
discussion and point out some future directions.
We now begin with the very basics of quaternionic spaces: their role in string theory and gauge
theories.
2. Quaternionic manifolds and string theory
Our first question would be to ask where does the quaternionic manifolds fit in the whole
paradigm of string compactifications. One of the place where these manifolds appear is well
2 In mathematical terminology therefore these instantons are constrained instanton bundles.
3 These Khler spaces have been originally classified in [11].
37
known: the moduli space of sigma models for N = 2 supergravity in four spacetime dimensions.
Imposing only global N = 2 supersymmetry in four dimensions would lead to sigma models with
hyper-Khler target spaces [13]. The N = 2 multiplets on the other hand can be written in terms
of N = 1 multiplets. This should tell us the moduli space structure for the corresponding N = 1
case also. In fact one can now make the following classifications for N = 1 supersymmetry in
four dimensions:
With global supersymmetry the target manifold of a non-linear sigma model can be any
Khler manifold [14].
With local supersymmetry the target manifold of a non-linear sigma model (which is coupled
to supergravity) can only be a restricted Khler type, also known as a Hodge manifold [15].
The second point is easy to show [15]. We can define a Khler potential K in terms of the
chiral superfield i and i . The terms appearing in the N = 1 Lagrangian can be expanded
K
from 3e 3 . The first two relevant terms are

S=

R
i j
d x g gi j + fermions ,
2
4
(2.1)
where gi j (not to be confused with g) is the metric on the moduli space parametrised by the
i the scalar component in the chiral multiplet i .
The Lagrangian (2.1) possesses Khler invariance under a Khler transformation. On a local
patch it is easy to demonstrate. However to demonstrate this globally one has to show how
this transformation can be defined from one patch to another. This gives rise to the consistency
condition on triple junctions. From here one can argue the condition required on the elements
of the second cohomology group of the target manifold H 2 : they have to be even integers [15].
Quantization of Newtons constant also follows directly from here [15].
On the other hand, the classification for N = 2 supersymmetry is more interesting. We discussed this briefly at the beginning of this section. We will now elaborate this in some details. As
before, global and local supersymmetry will have distinct properties:
With global supersymmetry the target manifold of a non-linear sigma model can be any
hyper-Khler manifold [13]. These are 4n-dimensional real Riemannian manifolds with
holonomy group lying in Sp(n).
With local supersymmetry the target manifolds of a non-linear sigma model coupled to supergravity can only be quaternionic Khler manifolds [16]. These manifolds are oriented 4n
real dimensional manifolds with holonomy groups lying in Sp(n) Sp(1). These manifolds
have negative curvatures given by [16]:
R = 64n(n + 2)GN ,
(2.2)
where GN is the Newtons constant and n is an integer. This means that the Newtons constant is fixed for a given manifold and not quantised like the earlier N = 1 cases. It also
means that the global susy couplings are no longer compatible for the local susy case. Only
in the limit GN 0 the local and global cases could be identified.
38
2.1. An example in detail

Let us consider one concrete example where quaternionic target space can be illustrated. As
mentioned above, a sigma model with quaternionic target space has to be coupled to supergravity
to make sense. Global supersymmetry cannot yield a quaternionic target space. Therefore our
four-dimensional Lagrangian can be taken as:

za z b zb
R
z a zd zd
1
a
S = 2 d 4 x g f f za
(2.3)
z
,
2
z z
z c zc
z e ze
which is a FubiniStudy metric on the target space. In fact the way we wrote the Lagrangian only
implies a CPN target because the coordinates za go from a = 1 to a = N + 1. This is a Khler
metric, but still not quaternionic because the Khler potential K is

K = log 1 + za z a ,
(2.4)
where za are summed from a = 1 to a = N because we are in a patch with zN+1 = 1. To convert
(2.3) to quaternionic case, we will first replace all za q a , where qa is a 2 2 matrix given as:
a

q0 + iq3a q2a + iq1a
qa =
(2.5)
,
q2a + iq1a q0a iq3a
where a = 1, . . . , N . This would then convert (2.3) to the following quaternionic analogue:

tr(q q) tr( q q)

R
1
2
4
d x g
S=
tr q q
, (2.6)
2
tr(q q)
tr(q q)

where we have defined tr(q q) as a tr(q a q a ) and similarly the other terms. Such a redefinition to convert (2.3) to (2.6) changes CPN to HPN where
HPN =
Sp(N + 1)
.
Sp(N ) Sp(1)
(2.7)
The quaternionic analogue of CPN i.e., HPN in fact shares the same properties as CPN : the q a
vectors are defined upto a scaling by a quaternion (recall za are only defined upto a complex
scaling). It is also important to note that any 4N + 3 sphere is equivalent to a S 3 fibration over a
quaternionic base HPN . This will be useful soon.
2.2. Structure of the multiplets
The quaternionic sigma model that we discussed above can be shown to appear in string theory
by compactifying type II strings on a CalabiYau three-fold. This leads to N = 2 supersymmetry
in four-dimensional spacetime with the following generic multiplets:
Vector multiplet: (A , 2, 2 ).
Hypermultiplet: (4, 2 ).
Tensor multiplet: (B , 3, 2 ).
Double tensor multiplet: (2B , 2, 2 ).
Vector tensor multiplet: (B , A , , 2 ).
Gravity multiplet: (g , A , 2 ).
39
Where appearing in all these multiplets are real scalars, are Weyl fermions in four dimensions and are four-dimensional gravitinis. Observe that both the double tensor multiplet as
well as the tensor multiplet are dual to the hypermultiplet. Similarly the vector tensor multiplet is
dual to the vector multiplet. Thus the non-trivial four-dimensional N = 2 multiplets are the vector, hyper and the gravity multiplets. Compactifying type IIB theory on a CalabiYau three-fold
gives rise to the following multiplets:
(g , A ) h12 (A , 2) h11 (B , 3) (2B , 2),
(2.8)
where we have ignored the fermionic degrees of freedom. From ten-dimensional type IIB point of
view, the metric fluctuations give rise to (2h21 + h11 ) scalars in four dimensions, the NS and RR
antisymmetric tensors both contribute h11 scalars in four dimensions along with the axio-dilaton
contributing two more scalars. Thus the scalars in the vector multiplets all come from the metric
fluctuations whereas the scalars in the tensor multiplets come partly from the metric fluctuations
and partly from the zero mode fluctuations of the NS and RR two form tensors. Finally the axiodilaton go to the double tensor multiplet. On the other hand, the vectors in the gravity as well as
vector multiplets all come from the zero mode fluctuations of the four-form field. The four-form
fluctuations also contribute h11 antisymmetric tensors that go to the tensor multiplets whereas
the NS and RR two forms both go to the double tensor multiplet. It is also easy to see that once
we dualise the tensor and the double tensor multiplets, we will have one gravity multiplet, h12
number of vector multiplets and (1 + h11 ) number of hypermultiplets. On the other hand, type IIA
theory when compactified on the same CalabiYau will give us the following four-dimensional
multiplets:
(g , A ) h11 (A , 2) h21 (4) (B , 3),
(2.9)
where again we have ignored the fermions. To keep track of the scalars: the hypermultiplet scalars
come from both the metric fluctuations and a zero mode fluctuations of the three-form field. The
vector multiplet scalars come partially from the zero mode fluctuations of the BNS field and
partially from the fluctuations of the metric. The dilaton however goes to the tensor multiplet this
time. On the other hand, the vectors in the vector multiplets do not come from the IIA vectors
but from the zero mode fluctuations of the three form field. In fact the type IIA vector go to
the gravity multiplet. The antisymmetric tensor in the tensor multiplet is the type IIA BNS field.
Observe also that in the dual picture (i.e., dualising the antisymmetric B field) we have one
gravity multiplet, h11 number of vector multiplets and (1 + h21 ) number of hypermultiplets. This
would be exactly the same as the type IIB multiplets if
h11 (IIA) = h21 (IIB),
and h21 (IIA) = h11 (IIB),
(2.10)
which is of course the statement of mirror symmetry at perturbative tree level.

At this point we should also note that the structures of quaternionic manifolds in string theory
are restricted in string compactification. This is easy to see from the fact that some of the scalars
in the hypermultiplets come from the zero mode fluctations of the metric. The moduli space of
these scalars are Khler manifolds and therefore the full quaternionic structure of the hypermultiplet moduli space [6,16]that come from adding RR scalars to the metric fluctuationsshould
have a submanifold that is a Khler manifold. This mapping of a Khler submanifold to the full
quaternionic manifold is called as a c-map4 [5]. Thus, for example, in type IIB on a CalabiYau
4 Or sometime as the s-map [5].
40
manifold the quaternionic space is of real dimension 4(1 + h11 ) with a subspace given by
SU(1, 1)
Mk ,
U (1)
(2.11)
where the first part is parametrised by four-dimensional axion-dilaton i.e., the double tensor multiplet, and the second part is the Khler submanifold. On the other hand, in type IIA theory the
first part of (2.11) comes from the four-dimensional tensor multiplet. Thus clearly the hypermultiplet target space cannot be a generic quaternionic manifold because of the c-map constraint [5].
Furthermore since the dilaton resides in the hypermultiplets, the tree level picture is not correct.
Details of these have been worked out various authors (see for example [1719] and references
therein). In particular, the perturbative corrections are now fully understood, and not just for the
universal hypermultipletas shown by [18] there are no quantum corrections beyond 1-loop due
to a nonrenormalization theorem. Moreover, the complete worldsheet, D1 and D(1) instanton
corrections in IIB as well as half of the D2 instanton effects in IIA have been determined by [18]
together with [19]. The resulting modified moduli spaces are quaternionic in agreement with
unbroken N = 2 supersymmetry.5
2.3. Few more examples
The restriction that we mentioned regarding construction of quaternionic manifolds may pose
a difficulty in having explicit examples. However string theory gives us a very simple way to
construct quaternionic manifolds that are consistent with the c-map:
Construct a vector multiplet Lagrangian in four dimensions. The multiplet is (A , 2, 2)
with the real-scalars forming a Khler target space. Such a Lagrangian coupled to gravity is
well known [16].
Dimensionally reduce this Lagrangian to three spacetime dimensions. The vector multiplet
will give us (A , 3, 2 ) in three dimensions.
Dualise the vector to another scalar via d = dA to convert the vector multiplet to a
hypermultiplet (4, 2 ). The metric on the moduli space of these scalars is exactly quaternionic [6].
The quaternionic metric is also consistent with the c-map because we derived this from the
vector multiplet with a Khler target. Thus the quaternionic manifold will have a submanifold that is Khler, as one would have expected [6].
In fact the above set of steps can be put into a more concrete setting. Consider a simple N = 2
Lagrangian with complex scalars coupled to one forms and gravity. A typical set up is

S4 = d 4 x g R + Ga b a b + cij F i F j + dij F i F j ,
(2.12)
where Ga b is the metric on the moduli spacewhich will be a Khler metric as we discussed
aboveand cij and dij are some coefficients which are functions of the moduli a . The subscript
i, j signify the number of vector multiplets that we couple to gravity.
In this form the Lagrangian (2.12) is almost like a D3-brane action coupled to gravity. However the resulting configuration should not be viewed as a D3 located at a point on a CalabiYau
5 We thank Ulrich Theis for pointing this out to us.
41
because the supersymmetry will not be N = 2 and the dimension of the Khler moduli space
will be fixed. Furthermore the instanton coefficient dij is not quite related to the ten-dimensional
axion. We will however relate a slight variant of this configuration to a D3-brane metric soon.
After a dimensional reduction and subsequent duality, we will get a three-dimensional action
for the hypermultiplets. This is given by:

S3 = d 3 x g R + Ga b a b + Gcd D c D d ,
(2.13)
)
where (, , ,
form the coordinates of a quaternionic space with a metric Gcd spanning the
submanifold specified by the coordinate c . The covariant derivatives D c are with respect to
some connection. This structure of the moduli space can be easily connected to the ones studied
by [17,20].
We can try to make this a bit more precise using the previous form of our action (2.6). Let us
consider the following choice of the quaternion:

0 B
q=
,
(2.14)
C 0
where both B and C are complex numbers (not necessarily independent). The scalar target space
parametrised by the quaternion then will have the following structure:
| C|2 + | B|2 |C C + B B |2
(2.15)
,
|C|2 + |B|2
(|C|2 + |B|2 )2
where we have suppressed the gravity part. Consider now the scenario where B and C appearing
above are complex numbers, but are not independent. They are related by
L=
B = C
(2.16)
as is clear from the quaternionic structure of the q coordinate. Such a choice of B, C would imply
that the Lagrangian (2.15) can be recast as
2| C|2 | S 2C C |2
,
S + S
(S + S )2
where, in our notation, S is not quite an independent variable as it stands. It is given by
L=
(2.17)
S = |C|2 .
(2.18)
The reason for writing (2.17) in the present form is to allude to the subsequent structure that we
will be inferring from string theory.
The string theory examples that have been studied earlier are all non-compact symmetric
spaces with negative curvatures. In fact string theory tells us precisely how S defined above
(2.18) should be modified so as not to change the underlying quaternionic structure. The resulting
metric will be consistent with the target space metric of a tensor multiplet (B , 3, 2) when
dualised to a hypermultiplet in four dimensions. Although this is no way the most generic method
to derive the metric, it does help us to see the subsequent structure. In type IIA this is therefore
a compactification on a CalabiYau three-fold that has no complex structure deformations (more
on this later). Furthermore since dilaton sits precisely in such a multiplet, quantum corrections
are expected to affect the target space metric. After the dust settles, the final answer is a slight
modification of our simple calculation above. The quantity S changes from (2.18) to
S = |C|2 + e2 + i,
(2.19)
42
where is the dilaton sitting in the tensor multiplet, is the corresponding axion (dualised from
the B field in four dimensions) and C, C are the other two scalars in the tensor multiplet.
These are the two scalars that come from type IIA three form in ten dimensions. Similarly the
Khler potential is changed to

K = ln S + S 2|C|2 + quantum corrections ,
(2.20)
which implies that the resulting manifold is also Khler (see [17] for some details). Without
quantum corrections the tree level moduli space for the universal hypermultiplet is given by
MH =
SU(1, 2)
,
U (2)
(2.21)
which is the non-compact analogue of Gr2 (C3 ) because of the negative curvature. Under tree
level quantum corrections the Khler structure of the moduli space is broken [21]. Further corrections to the moduli space come from the two- and five-brane instantons. These and others have
been addressed in [18,19,22] as we discussed briefly before, although a full treatment is far from
complete.
Let us consider another example. This time we compactify type IIA theory on a CalabiYau
threefold with no complex structure deformations (i.e., h21 = 0). Thus in four dimension we will
have the following multiplet structure:
(g , A ) h11 (A , 2) (B , 3),
(2.22)
which is a slight modification of (2.9). As we can see, the universal hypermultiplet is always
there. The moduli space therefore is from the vector multiplet Khler space as well as the universal hypermultiplet, as is given by
SU(2, 1)
,
(2.23)
U (2)
where G is the Khler manifold of dimension h11 . Observe also the fact that there are (1 + h11 )
vectors in this setup (extra one coming from the gravi-photon).
Compactifying type IIB theory on the same CalabiYau gives us (1 + h11 ) hypermultiplets
coupled to gravity (and graviphoton) and no vector multiplets. The quaternionic manifold that
we get here can in fact be derived from the moduli space (2.23) via the c-map. This is given by
h11
M = GKhler
4(h +1)
21
,
Gquaternion
(2.24)
from where we can easily see that the quaternionic space SU(2,1)
U (2) forms a sub-manifold of the
4(h
+1)
21
final irreducible quaternionic space G
. This is the essence of the c-map in the presence
of the universal hypermultiplet.
In the following section we will address the question of classifying quaternionic manifolds
using constrained instantons and SeibergWitten curves, and discuss the emergence of the socalled magic square.
3. On the classification of quaternionic manifolds: standard cases
As discussed in earlier sections, the classification of quaternionic manifolds have been started
in [2,3], and completed finally in [23]. Many of the cases that we studied so far (or have been
addressed in the literature) can be seen to follow from the above framework. For our case we
will try to understand the classification of the compact symmetric quaternionic Khler manifolds
using a different technique. Some aspects of this have been addressed earlier in [24].
43
3.1. Sp(n + 1) quaternionic space

Our first starting point will be the simplest case of Sp(n + 1) quaternionic space.6 As we
will discuss below, the quaternionic space associated with Sp(n + 1) group is special in the
whole classification of quaternionic spaces. The key point that we will follow to classify these
spaces is this: we look for gauge theories with certain global symmetries G (here, for this case,
it is Sp(n + 1)) and find semi-local instanton configurations. The low momentum dynamics of
these theories (by low momenta we mean momenta lower than the masses of the Higgs and the
masses of the photons) can be shown to be sigma models with quaternionic target spaces. Such an
approach was first discussed in [25] (see also [26] for sigma models on Khler target spaces) and
later elaborated in [24]. Here we will try to complete the analysis by detailing the corresponding
gauge theory constructions.
The gauge theory that we are looking for is an Sp(1) SU(2) gauge theory with a global
symmetry G. Clearly this theory resembles closely to a sector of the corresponding Seiberg
Witten theory with global symmetries [8]. To make this precise, let us write the action for our
theory. This is given by the following generic form [25]:

1
S = d 4 x trSU(2) F F + tr D q D q + V tr q q + fermions ,
(3.1)
4
where q is a generic quaternion as described in the previous section, and the trace is over the
global symmetry.7 Obviously, as mentioned above, this is not quite a SeibergWitten theory as it
stands. However once we write the quaternions in terms of complex fields (we show an example
below), the action will resemble a part of the standard N = 2 action with a potential V (a simple
case is the one worked out in [27] for an Sp(1)g Sp(1)l case). In this sense, we can use the
SeibergWitten curves to determine the global properties of this model. A recent example of
semilocal defects like strings in SeibergWitten theory is [28]. Our goal is to study instantons
in the model (3.1), i.e., a sector of, and not quite the actual, SeibergWitten theory. In fact the
analysis of instantons in this theory can be done in two different ways, both leading to the same
result. The first way is to observe that a theory like (3.1) will not allow any non-trivial instantons
if

G
3
(3.2)
= 1,
H
where H is the unbroken subgroup. However instantons are possible when a subgroup of G is
gauged.8 Let us call the ungauged subgroup of G to be Gg H. Then the vacuum manifold M1
of this theory is rather simple. It is given by:
M1 =
G
Sp(n + 1)
=
S4n+3 ,
Gg
Sp(n)
(3.3)
6 A point about notation: we will be considering Sp(n) groups instead of Sp(2n) groups used sometime in the literature.
In our notation therefore Sp(n) group is just the quaternionic unitary group U (n, H). Its a real, compact and simply
connected Lie group of dimension n(2n + 1). In particular Sp(1) SU(2) and we will not distinguish between them in
this paper.
7 Note that q will transform as a fundamental of both the global G and the local SU(2) groups for all choices of G
considered henceforth unless mentioned otherwise.
8 These are the constrained instantons [9] as we will explain below.
44
where, as should be clear from the above analysis, Gg = Sp(n) and we are taking the following
breaking pattern:
Sp(n + 1)g Sp(1)l Sp(n)g Sp(1)g
Z2
Z2
(3.4)
with being the Higgs field. The Higgs field is to be considered as a quaternion and not a
complex number, although we could consider this also to be a complex matrix. The quaternion
that could be used to represent the Higgs field is already pointed out above in (2.5). Thus

a1 a2 |a1 |2
(qa ) =
(3.5)
|a2 |2 a2 a1
is a good representation of the Higgs field in terms of the quaternions (qa ) = (q1 , q2 , . . . , qn ) or in
terms of a . As pointed out in [10,25] (and references therein) this is equivalent to a model with
n + 1 copies of the electroweak scalar sector with an Sp(n + 1) global symmetry in the W = 0
limit.
The second way is to view (3.1), when written in terms of complex coordinates and incorporating other terms, as describing the Higgs branch of SeibergWitten theory. Then the semilocal
instantons can be related to the small instantons described by Witten [29] and GanorHanany
[30] and the vacuum manifold M1 becomes the moduli space of one-instanton. These instantons
are described by embedding SU(2) groups inside the global groups, and therefore the different
SU(2) orientations describe the moduli space of the theory.9 These SU(2) orientations form an
S 3 and the moduli associated with the sizes of these instantons form the radii of the three cycles. In this language these three cycles will be fibered over quaternionic Khler spaces. Such
an approach has been used to study quaternionic Khler manifolds associated with An , Bn , Cn
and Dn groups [34]. The moduli space then is a 3-Sasakian spaces that are Sp(1) fibrations over
quaternionic Khler spaces [35] and is given by:

Mk = R4 R+ Sp(1) f Qk ,
(3.6)
where R4 denotes the four-translation moduli, R+ denotes the size moduli, the subscript k
denotes k-instantons, Qk denotes the quaternionic space associated with k-instantons and the
subscript f denotes non-trivial fibration. In the following we will give a concrete example of
such fibration using mostly the first technique (although in many cases we will alternate between the two techniques10 ). This will prove convenient for theories that may not have a good
9 Observe that, if we view the SeibergWitten theory to be generated by D3/D7 system a la [31], then the gauge
instantons are D(1)-branes inside D3-branes, whereas the small instantons are the bound states of D3-branes with the
D7-branes [32]. If we T -dualise the system then we will have a configuration of D1, D5 and D9-branes. The moduli space
of the small instantons on D9-branes i.e., D5-branes in D9-branes is given via ADHM data by a special hyper-Khler
manifold, or a quaternionic Khler manifold when coupled to supergravity [29]. On the other hand the moduli space of
D1-branes is given by a sigma model with ADHM target space [32,33]. Thus both the pictures describe the same physics.
10 There is also a third way of studying the moduli spaces of these instantons that is slightly different from the above
two approaches (although more related to the second one). This has to do with the fact that N = 2 supersymmetric gauge
theories also have hypermultiplets in the adjoint representations of the gauge groups. Observe that the hypermultiplets
that we considered for the above two cases are all in the fundamental representations of the gauge groups. Combining
these adjoint hypermultiplets with the N = 2 vector multiplets will give us the spectrum of N = 4 gauge theories. In
these theories moduli spaces of instantons will be exactly the same as for the fundamental hypermultiplets if we exchange
the global symmetries with gauge symmetries. Thus N = 4 theories with exceptional gauge symmetries will have the
same moduli spaces of instantons as we study here. Such an approach has been discussed by Stefan Vandoren in the last
45
Table 1
Semilocal theory
SeibergWitten theory
Semilocal instantons
Small instantons
G/H = Vacuum manifold
Higgs branch = Instanton moduli space = Special

hyper-Khler manifold
Gauging an SU(2) subgroup of the global group

3 (G/H ) = 1, 3 (SU(2)) = Z
Embedding SU(2) instanton in the global

group = Orienting SU(2) group in a global group
H = Unbroken subgroup
H = Stability group of the instanton
Quaternions
N = 1 chiral multiplets or N = 2 hypermultiplets
SU(2) gauge group
Microscopic SU(2) group
Quaternionic Khler manifold
Quaternionic Khler manifold
Semilocal strings
Semilocal strings
Mass term in the potential
Massive hypers, mixed CoulombHiggs branch
Lagrangian description (and therefore no well defined Higgs branch) but more importantly the
technique of semilocal defects is ideally suited to study other manifolds in the magic square as
we will discuss soon. In Table 1 we show the precise connection between our semilocal theory,
and the full SeibergWitten theory.
From above table it should be clear that although our theory (3.1) is a small sub-sector of
the original SeibergWitten theory, it has all the necessary ingredients to understand the detailed
aspects of magic square as we will demonstrate soon. The complicacies of the full Seiberg
Witten theory, for example the existence of Coulomb branch or mixed CoulombHiggs branch,
do not effect the analysis that we are going to perform therefore we will continue with our
simpler version (3.1).11 However we will try to demonstrate, whenever possible, how to analyse
the system from the full SeibergWitten theory.
Thus having laid down the possible criteria to construct explicit Sp(n + 1) quaternionic manifolds, there are a few important points to analyse now:
We have to verify whether it is possible to construct a SeibergWitten like theory with
Sp(n + 1) global symmetry. This would be confirmed by the existence of the corresponding SeibergWitten curve for the system. We expect, on generic ground, a curve of the form:
y 2 x 3 a2 x 2 k(z) + a1 xyl(z) + a3 yh(z) a4 xf (z) a6 g(z) = 0,
(3.7)
with ai being constants and k(z), l(z), h(z), f (z) and g(z) are polynomials in z. The coordinate z specifies the complex plane in the corresponding SeibergWitten theory. The above
equation with the right choice of k, l, h, f and g takes the following Weierstrass form that
reflects an Sp(n + 1) global symmetry:

5
zn+1
y = x 3 + xzn+1 + z2n+2
(3.8)
.
4
2
reference of [34]. For most of our analysis in this paper we will not consider the adjoint hypermultiplets as we want to
analyse N = 2 gauge theories only.
11 In fact our model (3.1) does not have a Coulomb branch. So in the corresponding SeibergWitten theory this is the
pure Higgs branch.
46
Using this one can check that the curve12 has the right singularity structure to allow an
Sp(n + 1) global symmetry. A similar curve should then describe the global properties of
our model.
The next step to verify would be the existence of instantons in this model. Clearly existence
of the corresponding curve (3.8) means that we have summed all the instanton contributions
to get the required SeibergWitten curve. However it is instructive to actually construct these
instantons. Out of the various different possibilities of instanton configurations in our system
(because of the matter representations) we will henceforth only concentrate on the so-called
semilocal instantons unless mentioned otherwise. These are the small instantons in the Higgs
branch of the full theory. Their construction is subtle because of two reasons. Firstly the
vacuum manifold being S 4n+3 would imply
3 (M1 ) = 1,
(3.9)
so would disallow instantons. The only allowed instanton configurations therefore would be
the semilocal instantons by gauging an Sp(1) part of the global symmetry.13 We may then
expect that the low momentum dynamics of the theory should be a sigma-model on a certain
quaternionic space, or alternatively the moduli space of the Higgs branch instantons should
be given by the quaternionic space. The structure of the corresponding quaternionic space
can be determined from the following gauge field configuration:
A Aa a =
q q q q
1
,
2
tr(q q)
2gYM
(3.10)
where the sum over repeated indices are implied via the dot product and a are the Pauli
matrices. Now due to the existence of F- and D-terms the low energy effective action will be
a quaternionic manifold HPn as shown in (2.6) when (3.10) is plugged in the action (3.1).
The semilocal instantons in this model have the following structure (see also [25]):

3 S 3 = Z,
S3
S 4n+3 HPn ,
(3.11)
provided certain subtleties are considered. This is the second reason. The subtlety has to
do with the presence of V (tr(q q)) term in the action (3.1), namely, due to Derricks
theorem once the scale invariance is broken by a mass term in the potential, the instantons
all squeeze to zero size. So the semilocal instantons that we are alluding to should exactly be
the constrained instantons of Affleck [9]. These constrained instantons resemble the standard
instanton at short distances only but decay exponentially at the IR [9] (see also [36]). In the
notation of [27], when
S = = 3 = 0,
(3.12)
12 Observe that this is only a genus one curve. For higher local gauge symmetry, for example SU(N ) with N > 2, we
will have a genus N 1 curve. In this paper we will look mostly at the sector of the theory that is given by a genus one
(i.e., N = 2) curve although in the last part of Section 4 we will give some examples of higher genus curves. Generic
case of an SU(N ) gauge theory broken to SU(2) Glocal gauge theory will be studied in the sequel to this paper.
13 One might be wondering about the connection between the curve (3.8) and the contributions from the semilocal
instantons. As is well known all possible instantons should contribute to the path integral to determine the full curve of
the theory [33]. The curve (3.8) is the minimal curve with Sp(n + 1) global symmetry so will have contributions from
the semilocal instantons (which are of course the small instantons in the Higgs branch). The situation gets tricky when
the global symmetry becomes very large (for example En as we will encounter later). In those situations how exactly all
the instantons contribute to give us the full curve will be described elsewhere.
47
Fig. 1.
where are the FI terms, then the instanton allowed are the standard instantons. For the
case when the FI terms are non-zero, to construct constrained instantons all we require is the
maximal subalgebra of the extended Dynkin diagram of Sp(n + 1):
should be expressible as a product of two subalgebras. This fixes the maximal subalgebra for
our case to be sp(n) sp(1). The constrained instantons are exactly of the gauged Sp(1)
SU(2) group. The simplest non-trivial example of such an instanton is for the global group
Sp(2). The quaternionic space associated with this global group is a four sphere S 4 because:
Sp(2)
= S 4 HP1
Sp(1) Sp(1)
(3.13)
and therefore the constrained instantons are non-trivially fibered over the four sphere (this
has also been noticed for a non-stringy example in [25]). For our case when 3 = 0 and all
other FI terms vanishing in V (tr(q q)) of (3.1), the constrained instanton can be explicitly
worked out to be of the following form:
A =
a x
2 2 a
x 2 (x 2 + 2 )
a x
2
a
3 gYM
+ ,
2
2
x
(3.14)
where 2 is the typical size of the instanton in the scale invariant limit (which is of course
the 3 = 0 limit). Observe that we need to also switch on non-zero expectation values for the
quaternions. It can be easily shown that the background values of the quaternions are always
proportional to the FI term 3 so that in the scale invariant limit their expectation values have
to vanish to allow the standard instantons to exist. In Fig. 1 a typical constrained instanton
is shown. We see that the instanton is non-trivially fibered over the quaternionic base HPn
and wraps the three sphere S 3 once at infinity. Over the rest of the space it completes a nontrivial four sphere S 4 in the quaternionic space. This also means that for HP1 we will have
a controlled theoretical way to study the instanton. This is in fact further facilated by the
following group theory identities:
SO(5)
(3.15)
,
SO(4)
which means that this special case could even be studied using real fields. This is indeed the
case, and has been attempted in [25].
HP1 = S 4 =
The above set of procedures was to construct a configuration of the simplest quaternionic
space HPn using constrained instantons. The relevant non-compact extension of the above space
48
is the quaternionic space

Sp(n, 1)
,
Sp(n) Sp(1)
(3.16)
which is more useful to study the moduli spaces in type II theories. Now recall that there is a
natural one-to-one correspondence between quaternionic normal Lie algebras and quaternionic
simply connected normal homogeneous spaces. In fact any normal quaternionic algebra should
contain a one-dimensional quaternionic subalgebra called the canonical quaternionic subalgebra.
The manifold that we studied above (3.16) correspond to the following totally geodesic subalgebra:
C11
Sp(1, 1)
.
Sp(1) Sp(1)
(3.17)
In fact (3.16) is the unique quaternionic algebra whose canonical subalgebra is isomorphic to
C11 [3]. However there is no Khler space associated with (3.16) because there is no c-map.
So (3.16) cannot appear as low energy Lagrangian in type II theories. Thus our construction of
the corresponding compact HPn gives the only legitimate way to study this manifold in string
theory. Below we will show that all the compact versions of the symmetric quaternionic spaces
can be studied using the technique of constrained instantons. In fact we will show how the magic
square appears in this analysis. But first, lets go to the next non-trivial example related to the G2
quaternionic space.
3.2. G2 quaternionic space
The technique that we developed in the previous subsection is universal. We will use the
same procedure of constrained instantons to construct quaternionic manifolds for the G2 cases
also. However instead of repeating the same constructions once again, we will give a concrete
mathematical way to build the quotient space:
G2
,
Sp(1) Sp(1)
(3.18)
so that combining this procedure and the steps elucidated in the previous subsection we will be
able to classify the magic square cases in the next section.
Before going into the details of the specific construction of (3.18) we would like to make
the following comments. The quotient structure of (3.18) should be obvious from the previous analysis, namely, the maximal subalgebra of G2 without an U (1) factor from the extended
Dynkin diagram:
is so(4) su(2) su(2). As this is already expressed in terms of two product group (with an
su(2) factor) we need not go any further. In fact the 7 of G2 then decomposes as14 :
7 (2, 2) + (1, 3),
14 We thank Tom Kephart for discussions on this point.
(3.19)
49
under SU(2) SU(2), where once we give a VEV to (2, 2) one of the global SU(2) (which is
broken) mixes with the broken local SU(2) to give us a diagonal unbroken SU(2). The quotient
space is then clearly (3.18). What remains to study however is the precise embedding of the
SU(2) groups inside a G2 . This will be addressed below.
The next issue is the existence of the corresponding SeibergWitten curve for a global G2
group. We have already laid down the possible curve for any global group G in (3.7). For G = G2
we can choose certain specific functional form for k, l, h, f and g in (3.7) to give us the following
curve:

12a1 zx 4a1 a2 z2 4a12 z2 + 12a3 z2 2
y+
24

x
= x3
a14 z4 + 8 a12 a2 3a1 a3 6a4 z3 + 16a22 z2
48

1
8 8
+
a1 z + 12 a14 a2 3a13 a3 z5 + 48a12 a22 + 216a32 72a12 a4 144a1 a2 a3 z4
864

+ 64a23 288a2 a4 + 864a6 z3 ,
(3.20)
where ai are some constants. The precise mapping of this curve to the G2 Casimirs can be worked
out but we will not do so here as our emphasis is more on the magic square. One can check that
the discriminant is

z6 + O z8
(3.21)
and therefore reflects a global G2 symmetry near the point z = 0. To see the full global symmetry
for other cases one has to generalise the above curve (3.20) further. Examples of these will be
discussed in the next section.
Another point is the existence of third homotopy groups for various coset spaces. For a global
group G broken to a subgroup H SU(2) our first criteria would be to ask the value of the third
homotopy from the exact sequence
3 (H) 3 (G) 3 (G/H) 0,
(3.22)
where both G, H are Lie groups.15 For simple cases dealing with non-exceptional groups this is
easy and well known. The interesting question comes when G is an exceptional group or when
both G and H are exceptional groups. Three rules have been developed to address these questions
[37]:
When both G and H are simple, i.e., when both G and H do not have invariant Lie subgroups,
then
l
(3.23)
,
L
where L is a non-negative integer called the index of a representation DG for the group G.
Similarly l is the index for the corresponding representation DH for the group H. These
3 (G/H) = ZM ,
15 This is crucial because, as mentioned earlier, our theory is only a sector of a bigger theory. Consistency requires that
G to study the instantons. On the other hand, in the full SeibergWitten theory, the
we evaluate the third homotopy of H
instantons are in the Higgs branch and so we would only require to evaluate the third homotopy of the global group G.
For more details see the table of comparison given earlier.
50
indexes are tabulated in details for many representations in [38].16 The idea is to look for a
particular representation (say vector or tensor) for the group G and then look for the same
representation for the group H. The ratio of the corresponding indexes will give us the value
for 3 (G/H). It is interesting to note that as long as we choose the same representations for
both G and H the ratio l/L will always be the same.
If G is simple but H is of the form of H1 H2 Hn with Hi simple, then
3 (G/H) = Z mod every
li
,
L
(3.24)
where (l1 , l2 , . . . , li ) are the collection of n-tuples. In fact H can have an additional Abelian
subgroup without changing the result. Furthermore modding by a discrete subgroup also
does not change the result.
When both G and H are not simple and G is of the form G1 G2 Gn where Gi are
simple,17 then 3 (G/H) consists of n-tuples of the form

(1 , 2 , . . . , n ) mod every
(1) (2)
(n)
li
l
l
, i ,..., i
,
L1 L2
Ln
(3.25)
where (li(1) , li(2) , li(2) , . . . , li(n) ) are the n-tuples associated with the simple groups Hi(1) , Hi(2) ,
. . . , etc., where the Lie algebras g, gi , hi associated with the Lie groups G, Gi , Hi , respec (j )
(j )
tively have the decomposition hi = j hi with the condition hi gj . The Lie algebras
(j )
hi are either isomorphic to hi or {0}. For more details the readers may want to refer to
[37,38].
Therefore the upshot of all these discussions is that the third homotopy groups for coset
spaces can either be 1 or Zp . For exceptional groups the third homotopy groups are all Z.
In fact generically 3 (SU(n))|n2 = Z. Similarly 3 (SO(n))|n3,n=4 = Z and 3 (SO(4)) =
3 (SU(2) SU(2)) = Z Z. This would mean that 3 (G2 /SU(2)) = 1 i.e., the third homotopy group is trivial,18 although this does not mean much because with G2 global symmetry a
Lagrangian description of the system like (3.1) discussed previously is not possible.19 Therefore
to study the constrained instantons in the system we gauge the SU(2) subgroup of the maximal
16 These indexes are represented as I (2) in [38].
rank
17 Additionally allowing Abelian groups as well as discrete moddings.
18 It turns out there are other possible embeddings of an SU(2) group in G , namely that the 7 of G goes to 3 + 2 + 2
2
2
of SU(2) or the 7 of G2 goes to 7 of SU(2). For these two cases 3 (G2 /SU(2)) = Z3 or Z28 respectively. We thank V.P.
Nair for pointing this out to us.
19 It is an issueand we will discuss this again laterfor all theories with exceptional global symmetries. One can
see this from the D3/D7-brane construction of these theories. The fundamental hypermultiplets appear from the strings
connecting the D3-branes with the D7-branes. The gauge symmetries of the seven brane theories appear as global symmetries of the underlying D3-brane theories. For classical Lie groups as gauge or global symmetries, the seven branes
are all D7-branes. However when we have exceptional Lie groups, not all seven branes are D7-branes. Some of them
are SL(2, Z) transform of the D7-branes. Because of that strings connecting the D3 and the seven branes may take nontrivial paths in the u-planes of corresponding SeibergWitten theories [39]. For such strings simple BornInfeld action
may not be easy to write down. Nevertheless such theories exist as can be easily shown from the corresponding F-theory,
or the SeibergWitten curves. Since the curves are constructed by summing up all the instantons, we also know that these
instantons exist. Therefore in this paper we will try to give as much information as possible, for these instantons, that do
not rely on explicit Lagrangian formulations. In the sequel to this paper we will attempt more explicit constructions.
51
SU(2) SU(2) group, or alternativelyviewing this in the Higgs branchwe study the orientations of SU(2) inside G2 . Thus effectively we are studying SU(2) constrained instantons in a
theory with the maximal group. These instantons are non-trivially fibered over the base (3.18).
As we discussed for the Sp(n + 1) case in the previous section we can now describe a possible
quaternionic geometry associated with the constrained instantons. In fact, as before, we need the
sigma model on the non-compact version of the geometry namely, on
G2(+2)
.
SU(2) SU(2)
(3.26)
To determine this we can use the trick of the c-map, that uses the metric of the Khler manifold
to determine the quaternionic manifold. The Khler manifold and the associated F function in
question are [5,6,16]:
MKhler =
SU(1, 1)
,
U (1)
i(X 2 )3
,
F XI =
X1
(3.27)
where X I , I = 1, 2, . . . , n + 1 are the scalar fields corresponding to certain other N = 2 vector

multiplets (including the gravi-photon) and we have introduced the F function to determine the
Khler metric of the manifold MKhler . This F function can be used to determine the Khler
potential K and the metric GAB KAB = A B K in the following way [6,40]:

= ln Z I NI J Z J
with NI J = i(I J F I J F ),
K(Z, Z)
(3.28)
I
{1, Z A } and the Khler metric therefore is the usual form ds 2 =

where Z I = X
X1
KAB dZ A d Z B . Observe that the metric is only positive definite in the region where Z I NI J Z J
is positive definite. Therefore KAB is negative definite [6,40].
It is now time to use the power of the c-map to determine the quaternionic metric for our
case. To build the quaternionic manifold we need 4(n + 1) coordinates. The Z A , Z A contribute
2n coordinates. The other 2n coordinates are denoted as AI , BI , along with two more complex
coordinates , . The c-map then defines the quaternionic metric in the following way [6]20 :

A dB B dA 2
2
2
I J
2
A B
ds = |d| 2e (Re N )I J W W + e
d
4KAB dZ d Z ,
2
(3.29)
where it should be clear that the Khler geometry (3.27) forms a submanifold in the quaternionic
space as expected. The structure of the universal hypermultiplet can also be extracted from (3.29).
The components of the matrix N , and W I are defined as:
NI K N J L X K X L
,
NI J = iI J F
X I NI J X J
I J
2N J K dAK idBJ ,
W I = (Re N )1
(3.30)
where Re N is negative definite. For other details about the properties of N etc the readers
may want to refer to [6,16,40,41]. In the remaining part of this section we will give an explicit
realisation of the quotient space (3.18).
20 We are using the notations of [40].
52
3.2.1. Realisation of the quotient space

G2
G2
SO(4)
, we
To give an explicit realization of the homogeneous space (3.18), i.e., Sp(1)Sp(1)
use the embedding of the exceptional complex Lie group G2 (C) into the complex orthogonal Lie
group SO(7, C). Similar embeddings are valid for the two real forms of G2 , since the compact
NC
group GC
2 (R) is included in SO(7, R) and the non-compact real group G2 (R) in the real Lie
group SO(4, 3). In the following, we will consider only the complex case and so we will omit the
presence of C in the definition of our Lie groups.
The group G2 has been shown [42,43] to be isomorphic to the group of orthogonal transformations SO(7) acting on the vector space C7 and leaving invariant a third-order completely
antisymmetric tensor T . It is completely characterized by the following:
T127 = T154 = T163 = T235 = T264 = T374 = T576 = 1.
(3.31)
Choosing to realize the group SO(7) by matrices G {gab } C77 with determinant equal to 1
that satisfy the orthogonality relation:
G G = I
gab gac = bc ,
(3.32)
we know that G will thus be characterized by 21 independent parameters. The invariance of the
tensor T under such transformations may be written as
G Ta G = gab Tb
Taef gec gf d = gab Tbcd ,
(3.33)
where Ta is the 7 7 matrix which elements are given by (Ta )bc = Tabc . It gives rise to 7
additional constraints on the elements of G and G thus contains the 14 independent parameters
that leads to G2 .
A simple realization of these conditions could be easily seen when we consider the algebra
g2 . It can indeed be realized as the set of orthogonal matrices M o(7) such that M = M
and satisfying the invariance condition
[Ti , M] = aij Ti ,
(3.34)
which can be easily obtained from the relation (3.33) using the usual derivation of the exponential
map which relates the group and algebra elements. We thus find an explicit form of M G2 in
terms of 14 independent parameters as:
0
a12
a13
a14
a15
a16
a17
a12
0
a23
a24
a25
a26
a27
a13
a23
0
a34
a35
a36
a15 + a26
a14
a24
a34
0
a27 + a36
a17 a35
a16 + a25
a15
a25
a35
a27 a36
0
a12 + a34
a13 a24
a16
a26
a36
a17 + a35
a12 a34
0
a14 a23
a17
a27
a15 a26
a16 a25 .
a13 + a24
a14 + a23
0
(3.35)
53
Let us mention that the maximal subalgebra so(4) = su(2) su(2) is easily identified. Indeed,
we first take ai5 = ai6 = ai7 = 0 for i = 1, 2, 3 to reduce the matrix to the form
0
a12
a13 a14
0
0
0
0
a23 a24
0
0
0
a12
0
a34
0
0
0
a13 a23
(3.36)
0
0
0
a14 a24 a34 0
0
0
0
0
a12 a34 a13 + a24
0
0
0
0
0 a12 + a34
0
a14 + a23
0
0
0
0 a13 a24 a14 a23
0
and then take the six remaining independent parameters as a34 a12 = 2x3 , a24 a13 =
2x2 , a14 a23 = 2x1 to get the direct sum decomposition as A B, where:
0
x+3 x+2 x+1 0 0 0
0
x+1 x+2 0 0 0
x+3
0
x+3 0 0 0
x+2 x+1
A = x+1 x+2 x+3

0
0 0 0,
0
0
0
0 0 0
0
0
0
0
0
0 0 0
0
0
0
0
0 0 0
0
x3 x2 x1
0
0
0
0
x1 x2
0
0
0
x3
0
x3
0
0
0
x2 x1
B = x1 x2 x3
(3.37)
0
0
0
0 .
0
0
0
0
2x3 2x2
0
0
0
0
0
2x3
0
2x1
0
0
0
0
2x2 2x1
0
We also see the inclusion of the preceding subalgebra so(4) of G2 in the algebra so(4) so(3)
as a subalgebra of so(7).
3.2.2. Coordinates of the quotient space
We start with the well-known realization of the Grassmannian of nondegenerate three-planes
Gr4 (C7 ) which is isomorphic to SL(7)/Aff(4, 3) where Aff(4, 3) is realized by matrices of the
form

0
G11
,
G11 C44 ,
G22 C33 ,
G0 =
G21 G22
G21 C34 ,
det G11 det G22 = 1.
We then define homogeneous coordinates on Gr4

X
X = z , X, Y C33 , z C3 ,
Y
(3.38)
(C7 )
as
(3.39)
so that SL(7) acts from the left as X = GX with G SL(7) and Aff(4, 3) is thus the isotropy
group of the origin chosen as X0 = (0, 0, I3 ) and X = GX0 . The restriction to SO(7) leads to
the isotropy group SO(4) SO(3) since G0 being orthogonal, it implies G21 = 0. The homogeneous coordinates X = GX0 of SO(7)/(SO(4) SO(3)) satisfy the orthogonality condition:
X X + zz + Y Y = 1,
(3.40)
54
which represents a set of 6 independent equations between the 21 parameters characterizing X .

Since we have

SO(7)
SL(7)
= dim
= 12,
dim
(3.41)
Aff(4, 3)
SO(4) SO(3)
the usual way to reduce further the independent quantities is to use the affine coordinates defined
as
W = XY 1 ,
w = z Y 1 ,
det Y = 0.
Let us now consider the quotient space G2 /SO(4). We have

G2
= 14 6 = 8.
dim
SO(4)
(3.42)
(3.43)
This space can be characterized by the homogeneous coordinates X = GX0 where now G
G2 SO(7) and thus satisfies the relations (3.33). They give rise to supplementary conditions
on the 21 parameters characterizing X . Indeed, we can write
g15 g16 g17

g25 g26 g27

0
g35 g36 g37
X = GX0 = G 0 = g45 g46 g47 .

(3.44)
I3
g55 g56 g57
g65 g66 g67

g75 g76 g77
The relations (3.33) imply, together with (3.31), that:
gab (Tb )76 = ga5 = Taef ge7 gf 6 ,
(3.45)
so the 7 parameters of the first column of X are expressed in terms those of the other columns.
Moreover, we have the orthogonality condition (3.40) which implies 3 more relations between
the remaining parameters:
ga6 ga6 = ga7 ga7 = 1,
ga6 ga7 = 0.
(3.46)
So, the number of independent parameter has been reduced to 11 at this stage. As before, the last
step to reduce further the number of parameters is to use the affine coordinates. The conditions
on W and w that leads a characterization of the quotient G2 /SO(4) are explicitly given in [44].
With this we are now ready to discuss the magic square. We will also show how some of the
aspects that we studied here can be elucidated from the properties of the magic square.
4. On the classification of quaternionic manifolds: The magic square
The magic square in mathematics is used to show the relation between division algebras,
Jordan algebras [12] and Lie algebras. The idea was first developed by Freudenthal, Rozenfeld
and Tits [7] and is introduced to string theory by GunaydinSierraTownsend [4]. The magic
square in mathematics is a 4 4 square with the entries given by elements of the Lie algebras.
The columns of the magic square are defined by the Jordan algebras, whereas the rows are defined
by the division algebras [45]. The division algebras are the real (R), complex (C), quaternion (Q)
and the octonion (O). The columns are labelled by: J 3 (R), J 3 (C), J 3 (Q), J 3 (O) where J 3 (K)
55
is the algebra of 3 3 Hermitian matrices over K. The magic square is then given by:
where Ai , Ci , Di , Ei , F4 are the usual SU, Sp, SO, E6,7,8 and F4 Lie groups respectively (a similar square can be drawn for the corresponding algebras also). The rules for filling up the entry L
of the magic square can be given by the relation (see for example [45]):
L = Der A (A0 J0 ) Der J,
(4.1)
where Der A and Der J are the generators of the automorphism group of the Hurwitz (division)
algebra A and of the algebra J , A0 are the pure imaginary elements of R0 = S 0 , C0 = S 1 , Q0 =
S 3 and O0 = S 7 and J0 are the elements of trace zero of the Jordan algebra J . To make this clear,
we can write the magic square in terms of the dimensions of the Lie algebras in the following
way:
The reason for the magical property of the square can be made clear from the entry-rule given
in (4.1). In terms of last to the first row, we can write the elements of the magic square in the
following way:
For more details see for example [46] (and references therein). The interesting feature of the
magic square is that its symmetric and four of the five exceptional Lie algebras occur in the last
row. In fact one could also add G2 to the magic square by adding an extra column (therefore
some literature also refers the magic square as a 4 5 rectangle). The extra column corresponds
56
to the Jordan algebra R (see figure below):
where other elements of the square are to be filled in the dotted parts. Once we have the Lie
groups, we should ask how to accomodate the quaternionic spaces or Khler spaces in the magic
square. To describe this let us use the column containing G2 and A1 Lie groups as this is the
simplest. In the language of constrained instantons, observe that in the maximal subgroup of G2
i.e., SU(2) SU(2) one of the SU(2) is gauged. This leaves one free SU(2) and the quaternionic
manifold is (3.18) (or (3.26) in the non-compact limit). For the next element of the magic square
i.e., A1 here, we look at the U (1) subgroup of the ungauged SU(2) and gauge it. The resulting
SU(1,1)
space is SU(2)
U (1) or U (1) in the non-compact limit. This reproduces the next element of the magic
square. Finally since we have gauged the remaining U (1) we have nothing else to gauge, so the
other two remaining elements of the magic square are 0 and 0 (see figure above).
Observe however that in the above figure we have ignored a subtlety regarding the c-map for
the G2 case. This has to do with the existence of two different non-trivial F functions for the
corresponding SU(1, 1)/U (1) Khler space [5,6]. This can be illustrated in the following way:
where we see that the same Khler space can give rise to two different quaternionic space. One
of the quaternionic space does not lie in the magic square and is generated by a F function given
by:
2 2
F XI = X1 X2 .
(4.2)
The fact that this is no contradiction is explained in [6]. What we are looking for is the c-map
related to Jordan algebra and this is given by the horizontal arrow.
Thus for the generic case our procedure should now be clear. We are gauging various subgroups as we move along the magic square. We call this sequential gauging. Let us consider a
part of magic square represented by non-compact group elements A, B, C and D in the following
way:
Question now is whether we can determine the corresponding manifolds associated with these
elements of the magic square using the arguments of constrained instantons. The manifold associated with group A is easy. This has to be a quaternionic manifold in such a way that a SU(2)
57
subgroup of the maximal group is gauged. What is the maximal subgroup of A here? This is
exactly given by the next element B of the magic square. Let Bc be the compact version of the
group B. Then the maximal subgroup of A is clearly Bc SU(2) giving rise to the quaternionic
manifold:
A
.
Bc SU(2)
(4.3)
Now question is whether we can determine the next manifold that should be Khler (recall the
c-map constraint). Looking at the next element we find the group C whose compact version is
Cc . What we need now is that the ungauged group B should decompose into Cc and another
subgroup. This is easy to determine from the list of subgroups given in [38]. Let the subgroup be
H1 . This therefore gives us the Khler manifold:
B
,
Cc H 1
(4.4)
whose c-map therefore will be (4.3). Going in this way we can reproduce all the manifolds
associated with the elements of the magic square in the following way:
where the subgroups Hi could in principle be determined from [38]; and the dotted lines are used
to show the connection between the ungauged groups. But the story does not end here because it
turns out that the subgroups themselves are not arbitrary. The quaternionic space was determined
by gauging the SU(2) subgroup. This was related to the constrained instantons. Now what could
be the next subgroup that we can gauge? Clearly this has to be a U (1) subgroup related to
semilocal strings. Similarly we can ask about the next to next subgroup. Since we gauged SU(2)
as well as U (1) we cannot gauge any other group! So our prediction for the magic square will be
H1 = U (1),
H2 = 1,
H3 = 1.
(4.5)
Observe however that there are some subtleties related to these identifications because the third
manifold associated with the group C in the magic square should be a real manifold, so we
might have to consider appropriate complex conjugates of the relevant groups. The final picture
that emerges from all the above consideration is:
which we would verify in the next few examples. A more detailed analysis of the manifolds
other than the quaternionic ones will be presented in the sequel. In the following sections we will
mainly study the quaternionic manifolds associated with En and F4 groups.
58
4.1. E6 quaternionic space

Our first case is to look for a theory with global symmetry G = E6 . To extract the quaternionic
space associated with this group we should study the maximal subalgebra.21 The maximal regular
subalgebra of E6 can be extracted from the extended Dynkin diagram:
and is given by H = su(6) su(2). This immediately tells us two things: One, we are dealing
with a gauge theory with H1 = SU(2) = Sp(1) gauge group, and two, the manifold ME6 is
ME6 =
E6
.
SU(6) Sp(1)
(4.8)
From the analysis that we presented in the previous section and using [37], one can verify that
E6
) = 1, so we need to gauge an SU(2) subgroup. Indeed, as like the previous cases, one
3 ( SU(6)
can find the following decomposition:
2) + (15, 1),
27 (6,
(4.9)
under SU(6) SU(2) subgroup. The SU(2) subgroup that we want to gauge is slightly different.
This subgroup is the diagonal subgroup of the SU(2)g SU(2)l where g, l stand for the global
and local groups respectively. Both the global and the local groups are broken by Higgs expec 2)and therefore an SU(2)g group survives (which
tation valueonce we give a VEV to (6,
we will call SU(2) henceforth). Since SU(2) S 3 , the homotopy classification will tell us that
3 (S 3 ) = Z. These are the constrained instantons, and therefore should have a construction via
the quaternion as we discussed before. These instantons are again non-trivially fibered over the
space (4.8) and therefore exist only as semi-local defects.
Thus we seem to get our required exceptional semilocal defect in this model. However in the
process of deriving this we have ignored a subtlety. This subtlety cannot be seen at the level of
group structure, in the sector of SeibergWitten theory that we study, but is visible when we look
at the corresponding SeibergWitten curve associated to our manifold. Therefore let us construct
the corresponding curve by modifying the G2 curve that we discussed in (3.20). The reason why
21 Notice that in addition to the choice of maximal subalgebras, we also ask for symmetric subalgebras of the groups.
The symmetric subalgebras for various groups have been listed in [38]. For the An , Bn , Cn , Dn cases, they are
su(p + q) su(p) su(q) u(1),
so(p + q) so(p) so(q),
sp(2p + 2q) sp(2p) sp(2q),
(4.6)
where p and q form the various distribution (as even or odd integers). For the En cases one would have
e8 so(16),
su(2) e7 ,
e7 su(8),
su(2) so(12),
e6 sp(8),
su(2) su(6),
e6 u(1),
so(10) u(1),
f4 .
From the list one has to extract out the relevant algebras that we would require for our case.
(4.7)
59
we want to start from G2 and go all the way to E8 is because of the last row of the magic square
which is expressed as a part of the 4 5 rectangle. Since the magic square elements are related,
we will then take (3.20) and add changes so that it eventually becomes the curve for E6 , and then
subsequently for other cases (we have ignored the F4 case for the time being because it will be
shown later to be very close to the E6 case).
Our first modification would be to change the powers of z in (3.20). This modifies the curve
to the following:

12a1 zx 4a1 a2 z3 4a12 z2 + 12a3 z2 2
y+
24

x
4
3
2
=x
a1 + 8a1 a2 + 16a22 z4 24(a1 a3 + 2a4 )z3
48

1
8 8
+
a1 z + 12a14 a2 + 48a12 a22 + 64a23 z6 + 216a32 z4
864

36a13 a3 + 72a12 a4 + 144a1 a2 a3 + 288a2 a4 864a6 z5 ,
(4.10)
with ai arbitrary. To fix the values of ai we have to study the singularity structures carefully. The
discriminant locus of this equation near the points z = 0 can be easily worked out. For us this
will be given by

z8 + O z9 ,
(4.11)
up to an overall numerical factor. To study the singularities at z = 0 the curve (4.10) is not generic
enough. To derive the actual curve we need to manipulate (4.10) further. We will do this in few
steps. First observe that (4.10) can be re-written as:

z4 4
(4.12)
Cz + Dz3 + Ez + F ,
864
where the new coefficients A, . . . , F and Y are defined from (4.10) in the following way:
Y 2 = x 3 xz3 (Az + B) +
12a1 zx 4a1 a2 z3 4a12 z2 + 12a3 z2

,
C = a18 ,
24
A = a14 + 16a22 + 8a12 a2 ,
D = 12a14 + 48a12 a22 + 64a23 ,
Y =y+
F = 216a32 ,
B = 24(a1 a3 + 2a4 ),
E = 36a13 a3 + 72a12 a4 + 144a1 a2 a3 + 288a2 a4 864a6 .

(4.13)
Secondly, that the curve (4.12) does not fully capture the E6 singularities completely can be
easily demonstrated (see also [47]). The dimensionality of x, Y, z, etc. can be worked out from
the equation
dSW dx
=
,
(4.14)
dz
Y
where SW is the SeibergWitten differential. We can then break the E6 global symmetry to
SO(10) U (1) such that the fundamental 27 decomposes as
27 = 16+1 + 102 + 1+4 ,
(4.15)
60
where the subscripts denote the U (1) charges. This would then imply that the coefficient of x in
(4.12) should have a z2 term [47]. Similar conclusion can be extracted by further breaking the
global symmetry to D4 SO(8) where we know that z2 should exist (see Eq. (2.16) in [48]).
Therefore if we redefine x, Y to x,
y as:
3
x = xz 2 ,
y = Y z2 ,
(4.16)
where the redefinition makes sense because we are not analysing the z = 0 points, then (4.12)
can be written as

1 4
1
2
3
2

3
y = x x(Gz
+ A z + B) +
Cz + Dz + Ez + F 1 log z + , (4.17)
864
2
where A and G are the minimal changes to (4.12). Observe that we can assume A A without
a loss of generality.
The new curve (4.17) is almost the one discussed in [47] with the exception of the additional
log z terms. These terms could be ignored for our case as we want to realise the pure E6 global
symmetry.22 To complete the picture we need to derive the explicit form for G, A and ai (i =
1, 2, 3, 4, 6). These are given in terms of E6 Casimirs defined in the following way [50]:

n
n n
pn (xj ) =
(4.18)
C{ni } x1n1 x2n2 x4 3 x5n4 x6 5 x8 6 ,
{ni }
where the operators xi are defined in terms of the Cartan subalgebra of E6 and n, ni are integers
satisfying the following algebraic equation:
n {2, 5, 6, 8, 9, 12} = n1 + 2n2 + 4n3 + 5n4 + 6n5 + 8n6
(4.19)
and C{ni } are integers. The sum is over all possible integer solutions of the above Eq. (4.19). As
an example the Casimir p6 will be defined via the following values of the coeffcients C{ni } :
C000010 = 1,
C410000 = 1062,
5
C011000 = ,
4
C030000 =
23
,
8
177
C600000 = 4680, (4.20)
,
C100100 = 60,
2
where one can get the full list in [50]. Using these Casimirs one can easily determine the coefficients G, A and ai by comparing the curve (4.17) with the one given in [47]. They are given
by:
C201000 = 15,
C220000 =
p2
2p5
,
A =
,
a1 = 25/8 33/8 2.328,
3
3
a32
298 2
101
13 2
49
19
32
=
p12
p 2 p8
p23 p6 +
p6
p26
p2 p52 ,
4
135
18225
218700
405
1049700
3645

1
11
p8
7
a4 =
p24
p 2 p6 +
25/8 33/8 a3 ,
2 10368
1080
45
G=
a2 = +
b2 1
b
1.837

+
1.355,
9a 2 3a
22 It is not clear to us what singularities would the additional log z dependent terms would imply. Of course additional
singularities besides E6 have been observed for certain F-theory curves in [49], but there the singularities were simple.
a6 =
a13 a3 a12 a4 a1 a2 a3 a2 a4 p22 p5

8
+
+
+
p9 ,
24
12
6
3
18
21
61
(4.21)
where using
c = 576p6 56p23 144 6,
a = 64,
b = 221/4 37/4 260.237,
we can define appearing in the definition of a2 above as

6

3
1
c
c 2
b
1
2
b
2 b3
3
4
,
=
3
3
2 27 a
a
2
27 a
a
3a
(4.22)
(4.23)
which would imply that a2 is a negative definite quantity. From the above we can also determine
the proportionality constant between A and A . This is given by
48a22
2p5
17/4
+2
37/4 a
2p5
,

2
48a2 + 302.8a2 + 29.37
2 + 36 6
(4.24)
where a2 can be extracted from above. This therefore completes the full analysis of the Seiberg
Witten curve for the system.
The subtlety that we were alluding to earlier lies in the realisation of the subalgebra (or the
subgroup (4.9)) associated with the E6 symmetry that would be used to determine the quaternionic manifold directly from the curve (4.17). Knowing the discriminant we can in principle
extract the corresponding subalgebra associated with the global group G = E6 provided the background space is specified. However the issue is more intricate because:
There is no Lagrangian description of the system with exceptional global symmetry. In fact
existence of the curve does not guarantee that the system is a SYM theory in some limit.
Even if there exist some suitable description, the system is at strong coupling [49] where a
controlled analytical calculation cannot be done. Furthermore due to large number of flavors
the theory is not asymptotically free.
All these issues might still be resolved if we embed our gauge theory in some stringy setup. There are various possibilites here. We might embed it in a F-theory set-up much like the
one discussed in [24,27,31,49,51,52], etc., or in a M-theory set-up like [53].23 Using any of
these cases, all we need is that the eight singularities decompose into a bunch of six and two
singularities giving rise to the discriminant and subgroup

E6 SU(6) SU(2),
(z a)6 z2 + b
(4.25)
which is of course the maximal subgroup for our case. Once the global symmetry is broken, a
Lagrangian description is possible when the system is embedded in a F-theory set-up. In F-theory,
analysing the curve however leads to the following subalgebra:
su(5) su(2) u(1)
(4.26)
23 In fact, since our theory is just a sector of the SeibergWitten theory, all the subtleties afflicting the original theory
will not have much effect on our analysis. Furthermore the SeibergWitten curve is the only output that we will be using
for our case.
62
instead of the subalgebra associated with the decomposition (4.9). This is almost the maximal
subalgebra that we wanted, but not quite.24 In fact su(6) is broken to su(5) u(1). Thus this is
the closest we come to getting the full structure of the coset space directly from type IIB string
theory (or F-theory).25 In fact what we need is that the 6 of SU(6) should decompose under
SU(5) U (1) as:
6 51 + 15 ,
(4.28)
which would form the ungauged maximal subgroup. The associated monodromy matrix is then
clearly

1 1
,
(4.29)
1 2
which leaves one of the dyonic point in the monodromy matrix and determines the rest of the
SU(6) generators non-perturbatively. The surviving diagonal SU(2) is now gauged according to
our earlier discussion.26
The above construction therefore gives us the constrained instanton configurations associated
with global symmetry E6 that are fibered over the quaternionic base ME6 (4.8). However, as in
the previous sections, this is not quite the manifold that we are looking for. We should aim for
the non-compact version of (4.8), i.e.,
V (1, 2)
E6(+2)
,
SU(6) SU(2)
(4.30)
where +2 in the bracket denote the difference between the number of compact and non-compact
generators. The corresponding Khler space associated with (4.30) can be constructed by gauging
subgroups of SU(6) according to our scheme. The relevant subgroup of SU(6) for us is SU(3)
SU(3) U (1) under which 6 decomposes as:
6 = (1, 3)1 + (3, 1)+1 ,
(4.31)
where by modding A5 by the corresponding subgroup gives rise to the following Khler space:
SU(3, 3)
,
SU(3) SU(3) U (1)
(4.32)
24 When our theory is embedded in the full SeibergWitten theory the same subtlety should show up in determining the
Higgs branch. However in the absence of a proper Lagrangian description this may not be easy to implement.
25 The full configuration on the other hand can be determined in the following way: First we decompose the E adjoint
6
in terms of the subalgebra (4.26) as
78 = (24, 1)0 + (1, 1)0 + (1, 3)0 + (10, 2)3 + (5, 1)6 + c.c.,
(4.27)
and 5 with U (1) charges 3 and 6,

where the subscripts refer to the U (1) charges and the c.c are associated with 10
respectively. Secondly, having given the decomposition, the rest of the discussion now should follow the familiar line
developed in the series of papers [39,49]. We will not elaborate on this aspect as the readers can look up the details in
those papers. It will simply suffice to mention that the non-trivial configuration required to get the full group structure
lies in the process of brane creation via the HananyWitten effect [54] leading to strings with multiple prongs [5557]
that fill out the rest of the group generators [39].
26 Recall that before combining the SU(2) part of the unbroken global group with the local SU(2) gauge symmetry we
3 2
.
expect a monodromy matrix of the form
2 1
63
where SU(3, 3) is the non-compact version of SU(6). Observe also that (4.32) is exactly of the
form (4.4) with H1 = U (1) and Cc = SU(3) SU(3). Furthermore, under a c-map (4.31) does
give us (4.30) once the F -function is specified. We will specify the F -function a bit later. Looking
now into the magic square for the E6 sequence:
where the vertical sequence is shown to emphasise how the curves were constructed, and the
horizontal sequence is constructed by various maps: c, r etc., we can easily argue the various
manifolds associated with the horizontal elements of the magic square using the technique of
partial gauging of the subgroups discussed in the previous section. This will give us the following
sequence:
where the third term in the sequence has H2 = 1 and the fourth term has H3 = 1 as predicted in
(4.5). With this sequencing structure we can now determine the sigma-model metric associated
with the constrained instantons fibered over the quaternionic base (4.10) (or (4.30) in the noncompact limit). The quaternionic metric is always of the form (3.29) which is derived from the
corresponding Khler metric (3.28). All we need to complete the picture for the E6 case would
be the F -value. We will present a detailed analysis of this in Section 4.5.2 including a generic
derivation for all possible cases.
Before we end this section, notice that we have not yet checked whether there is some semilocal soliton that could be fibered over the space (4.32) much like the quaternionic examples
studied so far. For this we have to study the associated vacuum structure. Whether this theory
could be studied in the same moduli space as the present ones needs to be investigated. It is of
course highly suggestive that there are semilocal string like defects because 1 (U (1)) = Z and
using the exact sequence for Lie group G and its subgroup H:

G
G
0 2
(4.33)
1 (H) 1 (G) 1
0,
H
H
one can easily argue that for G = SU(n) = SU(6) and H = SU(3) SU(3) (or in fact for any
generic Lie subgroups [58]):

SU(6)
SU(6)
1
(4.34)
= 0 = 2
,
SU(3) SU(3)
SU(3) SU(3)
showing that there could only be semilocal defects. We will however leave a detailed study of
this for future investigations.
64

Let us now turn towards the next group G = E7 . The extended Dynkin diagram of E7 :
can be cut in different ways to give rise to various maximal regular subalgebras of E7 . They are
given by
su(8),
spin(12) su(2),
su(6) su(3),
e6 u(1),
(4.35)
where spin(12) actually comes from so(12) with some identification between the generators.
From the set of steps that we mentioned earlier, we can immediately ignore the subalgebras su(8), su(6) su(3) and e6 u(1) and therefore the associated groups SU(8), SU(6)
SU(3), E6 U (1) as they cannot be realised in the present scenario (recall that the gauge group
is SU(2)).27 The above consideration immediately gives us the corresponding unique coset manifold for the global symmetry E7 as
ME 7 =
E7
.
Spin(12) Sp(1)
(4.37)
Our previous consideration will require us to view this as a homogeneous quaternionic Khler
manifold. The SU(2) constrained instantons are fibered over this manifold because the third
E7
homotopy group of the vacuum manifold is trivial i.e., 3 ( SO(12)
) = 1. But then again such a big
global symmetry will not allow a Lagrangian description of the system, so to make any concrete
statements we have to analyse the maximal subgroup SO(12)SU(2) associated with the system.
However as before, analysing the corresponding SeibergWitten curve will tell us that the
actual subgroup realised perturbatively is different from SO(12) SU(2) or Spin(12) Sp(1).
To see this we will study the theory in few steps. Firstly, the breaking pattern for the 56 of E7 is:
56 = (12, 2) + (32, 1)
(4.38)
under SO(12) SU(2). Giving a VEV to (12, 2) the broken global SU(2) can combine with the
broken local SU(2) and give us the unbroken global group SU(2) Sp(1). This is the Sp(1) that
appears in (4.37). Furthermore once we have the coset space (4.37) we have to analyse the rest
of the coset spaces from the magic square column:
To analyse the coset space (4.37) let us determine the curve associated with E7 by deforming
the E6 curve (4.17) that we determined earlier. Our first attempt to determine the curve using the
27 Observe however that the third homotopy groups of SU(2) and SU(3) are both given by

3 SU(2) = 3 SU(3) = Z
(4.36)
and therefore SU(3) theory can also allow non-trivial constrained instantons. It would be interesting to study the manifold
associated with this setup.
following values of the variables in (3.7):

{l, k, h, f, g} = z, z2 , z3 , z3 , z5 ,
65
(4.39)
can only tell us the discriminant behavior at z = 0. To determine the curve at any generic point
z = 0 we can deform (4.17) to the following curve:

y 2 = x 3 x 2z3 + Mz2 + N z + P

1
1 4
3
(4.40)
Qz + Rz + Sz + T 1 log z + ,
+
864
2
where M, N, . . . , etc., are written in terms of SO(12) Casimirs (see [47] for details). The discriminant locus that we can realise here will be:

z9 + O z10
(4.41)
and therefore would show an E7 singularity. On the other hand, we would not be able to realise the maximal SO(12) SU(2) subgroup here. The curve (4.40) will reflect the following
subalgebra:
su(6) su(2) u(1)
(4.42)
where the SU(2) is the same SU(2) symmetry that gets broken completely to give us an unbroken
global SU(2) in (4.37). Also as expected the 12 and 32 of SO(12) decomposes as:
12 = 6+1 + 61 ,
+1 ,
32 = 1+3 + 13 + 151 + 15
(4.43)
under SU(6) U (1). The monodromy matrix is now different from (4.29) that we had earlier for
the E6 case. It is given by

2 3
,
(4.44)
1 2
although the same dyonic point is enclosed.
The two monodromy matrices (4.29) and (4.44)

differ by the monodromy matrix 10 11 as expected.
As before the manifold (4.37) is not quite the quaternionic manifold that we are looking for.
Our aim is to get the non-compact version of this. Therefore using the compact and non-compact
generators of E7 we can construct the following manifold:
E7(5)
(4.45)
,
SO(12) SU(2)
which is the required quaternionic manifold falling in the classification of Alekseevskii [3]. In
this classification the manifold (4.45) is known as V (1, 4) manifold, and should be generated
from a Khler space via the c-map. So the question is: can we derive the Khler space associated with (4.45) using our argument of partial gauging? From the technique developed in earlier
sections, we have to look for the U (1) subgroup of the ungauged group in the global symmetry.
Here the ungauged group is SO(12) whose subgroup is clearly SU(6) U (1). Therefore from
the sequencing of the magic square, we can predict the Khler space to be:
SO (12)
,
(4.46)
SU(6) U (1)
which when acted by the c-map will generate (4.45). The other coset spaces associated with the
magic square can also be easily generated using the arguments of the previous sections. The final
66
magic square sequence for E7 will be given by:
which is consistent with the classification [3]. Observe that to go from the second element from
the left of the sequence to the third element we use the r-map. This is universal for the whole
magic square.
Our next exceptional global symmetry that we want to study here is E8 . This is straightforward (modulo some subtlety that we mention below) from all the considerations of the previous
sections. The extended Dynkin diagram is now given by:
From here the relevant allowed maximal subalgebras are

e7 su(2),
so(16),
su(5) su(5),
su(3) e6 ,
su(9),
(4.47)
out of which only two of them, namely, so(16) and e7 su(2) are also symmetric subalgebras.
We can now easily ignore the SO(16) subgroup because we are looking for constrained instantons
8
associated with the SU(2) group. Again constrained instantons exist because 3 ( E
E7 ) = 1. The
248 of E8 then decomposes as28 :
248 = (1, 3) + (133, 1) + (56, 2),
(4.48)
under E7 SU(2) subgroup. Once we give an expectation value to (56, 2) we can break both
the local and global SU(2)s to give us an unbroken global SU(2). Therefore the final symmetry
group E7 SU(2) is completely global and we can now gauge the SU(2) subgroup. Constrained
instantons can exist for the SU(2) theory, and they are fibered over the base manifold:
E8
,
E7 SU(2)
(4.49)
which gives us the quaternionic Khler manifold associated with E8 global symmetry.
There are few other details we could consider parallel to the details associated with other En
groups studied above. First is the existence of SeibergWitten curve for E8 global symmetry that
could be described here by deforming the E7 curve (4.40). This deformation is simple and is
28 Observe that 248 is the dimension of the adjoint representation of E . This is the smallest representation of E . There
8
8
is no smaller fundamental representation. This would meanfrom our earlier analysis of the potential in (3.1)we can
no longer use the argument of the quaternion q being in fundamental of the global E8 . However since there is no simple
Lagrangian formulation of this theory, an absence of fundamental representation may not pose an issue in constructing
the vacuum manifold. Indeed as we will show below, there is a possible alternative way to verify that the moduli space
of these instantons do not change when we work with the adjoint representation of E8 . We will deal with this issue in
more details in the sequel to this paper.
explained in [47]. The curve therefore is:

T2 T 4
y 2 = x 3 z2 T2 + O z2 x 2z5 + z4 T6 +
+ + O z3 ,
6
67
(4.50)
where Ti are SO(16) Casimirs. For more details the readers can refer to [47]. The manifest
subalgebra that one gets from analysing the curve is neither so(16) not e7 su(2) rather it is:
su(7) su(2) u(1),
(4.51)
which in turn means that the breaking pattern of E7 global symmetry is not directly to (4.51) but
through an intermediate su(8) subalgebra. In terms of the corresponding groups this is:
E7 SU(8) SU(7) U (1),
(4.52)
under which 56 and 133 should be decomposed. The associated monodromy matrix containing
the same dyonic point is:

3 5
,
(4.53)
1 2
under the decomposition (4.52). Using this monodromy matrix one can construct the other generators of E7 non-perturbatively.
As before the quaternionic manifold of interest is not quite (4.49). We have to look for the
non-compact version of this. This is given by:
E8(24)
E7 SU(2)
(4.54)
and is known as V (1, 8) manifold in the classification of Alekseevskii [3]. The associated Khler
manifold should have the necessary U (1) coset as predicted in (4.5). Gauging the U (1) will
correspond to the semilocal strings. The Khler manifold therefore is:
E7(25)
,
E6 U (1)
(4.55)
which under c-map will reproduce (4.54). Similarly (4.55) should come from the r-map of a real
coset space according to (4.5). The final sequence therefore should be:
which is again consistent with the existing classification [3]. In addition to the above scheme,
observe that the generators of the En exceptional groups appearing in the magic square can be
alternatively formulated in the following way [45]:
E6 = SO(8) + SU(3) + 6 7 = 28 + 8 + 6 7 = 78,
E7 = SO(8) + Sp(3) + 12 7 = 28 + 21 + 12 7 = 133,
E8 = SO(8) + F4 + 24 7 = 28 + 52 + 24 7 = 248,
(4.56)
where the existence of SO(8) = Spin(8) has to do with the underlying triality symmetry [45] and
the Lie groups in (4.56) are precisely the F4 , C3 and A2 groups appearing in the magic square.
68
Finally, before ending this section, let us come back to the issue of E8 representation that
we discussed briefly at the beginning. An alternative way to verify that we have the correct oneinstanton moduli space is to use the adjoint hypermultiplets of N = 2 gauge theory. The E8
global symmetry can be enhanced to E8 gauge symmetry by changing the SeibergWitten curve
(4.50) to a new one. The curve for this case takes the following general form [59]:
y+
2
+ PR (x; uj ) = 0,
y
(4.57)
where PR is a polynomial in x of order dim (R), and R is the adjoint representation of E8 ; y in

(4.50) and y differ at most by the polynomial PR . The other terms occuring in (4.57) are defined
as follows: h where h is the dual Coxeter number of E8 and is the PauliVillars scale.
The functions uj , j = 1, 2, . . . , 8 are the fundamental Casimirs of E8 with the top Casimir u8
has degree h. By changing (4.50) to (4.57) we have actually enhanced the susy to N = 4. Now
it is well known that for E8 small instantons in N = 4 gauge theory the moduli space is indeed
given by (4.49), thus confirming our above analysis.
4.4. F4 quaternionic space
The final example of exceptional global symmetry is F4 whose properties are not very different from all the other En examples that we have been studying so far. In fact F4 symmetry is
very close to the exceptional E6 symmetry. One hint comes from the folding relation between
the Dynkin diagrams of E6 and F4 :
Such similarity between the Dynkin diagrams is also reflected in the corresponding Seiberg
Witten curves near z = 0 point. The curves for F4 and E6 have the following structures:
where we have referred to only the highest order polynomials for a given coefficient. Clearly the
singularity structures of both the curves would then be very similar. Indeed the discriminant of
F4 curve is given by:

z8 + O z9 ,
(4.58)
which is identical to the E6 curve (4.11). The distinction between the two curves come from
analysing points z = 0. The fundamental representation of F4 is 26-dimensional whereas the
fundamental representation of E6 is 27, so they differ by a singlet. The maximal subalgebras of
69
F4 can be extracted from the extended Dynkin diagram of F4 :
by cutting the diagram at various points. This will give rise to the following subalgebras:
so(9),
su(3) su(3),
su(2),
sp(3) su(2),
g2 su(2),
(4.59)
out of which we will only keep sp(3) su(2) subalgebra because we want to keep the symmetric
subgroups. Clearly the group G2 SU(2) corresponding to the maximal subalgebra g2 su(2)
is not symmetric, and therefore we will not quotient F4 by this subgroup. Under Sp(3) SU(2)
subgroup the 26 of F4 decomposes as
26 = (6, 2) + (14, 1).
(4.60)
Giving VEV to (6, 2) we can break the global and local SU(2)s to have an unbroken SU(2).
F4
Since 3 ( Sp(3)
) = 1, the constrained instantons will be fibered over the following quaternionic
manifold:
F4
,
Sp(3) SU(2)
(4.61)
which is a compact manifold by construction. The manifold that we are concerned about is the
non-compact version of (4.61). This is given by:
F4(+4)
,
Sp(3) SU(2)
(4.62)
which is also known as V (1, 1) manifold in the classification of Alekseevskii [3]. The Sp(3) part
of the subgroup Sp(3) SU(2) used for quotienting F4 is ungauged. To construct the relevant
Khler manifold associated with (4.62) we need the symmetric subgroup of Sp(3). From [38]
we see that there is one unique subgroup: SU(3) U (1) U (3) containing a U (1). This means
that for a theory with Sp(3) global symmetry semilocal strings can exist by gauging the U (1)
subgroup. This immediately gives us the corresponding Khler manifold associated with (4.62):
Sp(3, R)
SU(3) U (1)
(4.63)
from which (4.62) can be generated by a c-map. The real manifold associated with (4.63) can
be similarly constructed by looking into the symmetric subgroup of SU(3) that does not have
an U (1) factor. This subgroup is SO(3) [38], and therefore the magic square sequence for F4
symmetry will be:
where SL(3, R) is the non-compact group associated with the compact group SU(3). The Khler
manifold (4.62) is the real image of the second coset from the left of the magic square. It is also
interesting to note that the 52 of F4 can be connected to spin(8) SO(8) in the following way:
F4 = SO(8) + SO(3) + 3 7 = 28 + 3 + 3 7 = 52,
(4.64)
70
which is much like (4.56) described earlier. Finally, to determine the sigma-model description of
the quaternionic manifold (4.61) or (4.62) we will need the F function that describes the metric
of the Khler manifold (4.63). This will be determined in Section 4.5.2.
4.5. Other examples of quaternionic spaces
After describing the complete magic square in terms of constrained instantons and possible
other semilocal solitons, let us now use the same procedure to study other coset spaces in string
theory.
4.5.1. Example 1: U (p) local symmetry and SU(n + p) global symmetry
Our first example is for a U (p) gauge theory with a global symmetry SU(n+p). The extended
Dynkin diagram for such a symmetry is
which will give us a symmetric subgroup of SU(n) SU(p) U (1) [38]. The existence of the
extra U (1) factor commuting with SU(n) group can be directly explained from the corresponding
gauge theory dynamics (see [34] for details).
The above theory can also be realised in the SeibergWitten setup by slightly modifying the
present scenario. First of all we need a genus g = p 1 curve instead of genus one curves that
we have been studying so far. The construction of such a curve is very well known [60] so we
will be brief. The curve for N = 2 U (p) gauge theory with SU(n + p) global symmetry is [60]:

2
p
n

2
p
pi
pn
ni
si x
+
gi x
pn x n+p ,
y = x +
(4.65)
i=2
i=0
where is the PauliVillars scale and (si , gi ) are some constants that depend on the parameters
of the theory. The exponent of is evaluated as:
I (RA )I (RM ) ,
(4.66)
where I(RA ), I(RM ) are the Dynkin indices of the adjoint representations of vector multiplet
and representations of matter hypermultiplets respectively [61].
The vacuum manifold of this theory will be a Stiefel manifold Vn+p,p [25] which is a space
of p-frames in Cn+p . This is isomorphic to SU(n+p)
SU(n) . At low energy the sigma model target space
therefore will be given by the following manifold:
CG(n, p)
SU(n + p)
,
SU(n) SU(p) U (1)
(4.67)
which is nothing but the manifold constructed by modding out U (p) gauge orbits from the Stiefel
manifold. This immediately implies:
Vn+p,p U (p) f CG(n, p),
(4.68)
where the subscript f implies non-trivial fibration. Thus the Stiefel manifold is a U (p) bundle
over a Grassmanian manifold. The quaternionic extension of the above case is to consider the
71
complex Grassman manifold CG(n, 2). This is denoted as Gr2 (Cn+2 ) in (1.1). For our purpose,
however, we need the non-compact version of this space. This is given by:
SU(n, 2)
.
SU(n) SU(2) U (1)
(4.69)
The constrained instantons will be non-trivially fibered over (4.69) in the theory. The manifold
(4.69) can be mapped to the corresponding Khler space by gauging a U (1) subgroup of the
unbroken group. The Khler space corresponding to (4.69) is:
U (n 1, 1)
,
U (n 1) U (1)
(4.70)
where (4.70) and (4.69) are related by a c-map as expected. Observe that the unbroken subgroup
in (4.70) is U (n 1) SU(n 1) U (1). To get the corresponding real manifoldthat could
be related to (4.70) by an inverse r-mapwe need a subgroup of U (n 1) that does not have a
SU(2) or an U (1) factor. This is not possible, so our simple rule tells us that there could be no
non-zero-dimensional real space associated with (4.70). This can be confirmed (see for example
[23]). The sequence therefore is:
which fits into Alekseevskii classification [3] as well as the recent completion [23]. Notice that
for n = 1 there is no Khler space.
4.5.2. Example 2: SU(2) local symmetry and SO(p + q) global symmetry
Our next example is almost self-explanatory. This is a SU(2) SeibergWitten theory with
SO(p +q) global symmetry. The symmetric subgroup of SO(p +q) from any of the two extended
Dynkin diagrams (related to Bn and Dn ):
is SO(p) SO(q). Therefore taking SO(7) global symmetry, or more appropriately, SO(3, 4)
global symmetry we can easily find constrained instantons in the theory that are fibered over the
following quaternionic space:
SO(3, 4)
.
SU(2) SU(2) SU(2)
(4.71)
The steps to generate Khler space associated with semilocal strings is also evident: we have
to mod the non-compact version of SO(4) global symmetry by U (1) U (1) symmetry. The
manifold therefore is

SU(1, 1) 2
(4.72)
,
U (1)
72
so that gauging one of the U (1) we can get semilocal strings in our theory. Manifolds (4.72)
and (4.71) are related by a c-map. The real manifold associated with (4.72) is clearly SO(1, 1).
The sequence therefore is:
which fits consistently with the de WitVan Proeyen completion [23] of Alekseevskiis classification [3].
Let us consider one more example that is in the same vein as our previous example. For
this case we take p = q = 4 so that our non-compact global symmetry is SO(4, 4). Clearly
the maximal (and symmetric) subgroup is SO(4) SO(4) [SU(2)]4 , so that the constrained
instantons are fibered over the following quaternionic manifold:
SO(4, 4)
(4.73)
SO(4) SO(4)
where we have, as usual, gauged a SU(2) subgroup of the maximal group. The ungauged subgroup therefore is SO(4) SU(2) [SU(2)]3 whose non-compact version would be [SU(1, 1)]3 .
To determine the Khler manifold we have to gauge an U (1) subgroup of [SU(1, 1)]3 so that we
are studying semilocal strings. The Khler manifold will have more or less the same coset structure as (4.72) discussed above because the ungauged subgroups are of the same form as above.
Following this trend, the sequence of manifolds that we now expect are:
which again fits perfectly with the classification of [23]. The zero-dimensional manifold in the
last box of the sequence is expected because the real manifold does not have a coset structure.
In fact so long as p 4, q 4 we do not expect to get a non-zero-dimensional manifold. This
should give us a hint that if we choose a more generic global symmetry from the start, then maybe
we could get a non-trivial manifold in the last box of the corresponding sequence. This is indeed
the case if we choose p = P + 4, q = 4 with P any integer. The sequence of manifolds are rather
straightforward to determine and they are of the following form:
where we see that we do get a manifold in the last box from which we can get the real, Khler
and quaternionic manifolds by various possible mappings. Needless to say, the above sequence
fits with the classifications of [3,23].
4.5.3. Example 3: New sequence of Khler manifolds in the magic square
Our final example is a rather curious one. Let us look at the third row of the magic square
containing the elements associated with E7 , etc.:
73
By construction these are all Khler manifolds that are related to the corresponding semilocal
strings (observe the U (1) quotients). An inverse r-map to each of these cosets will give us the
corresponding real manifolds that we studied in the earlier sections. For example for the unbroken
E6 subgroup of (4.55) has the following symmetric subgroups:
F4 ,
SU(6) SU(2),
SO(10) U (1),
Sp(4),
(4.74)
out of which we have used F4 to construct the real manifold 6(26)

F4 . The other subgroup SU(6)
SU(2) was used in a different example to construct a quaternionic manifold (which is of course
unrelated to this sequence of magic square). So we can ask the following question: what if instead
of (4.55) we want to construct coset space associated with SO(10) U (1) symmetry? This would
mean that we are again looking for semilocal strings for a U (1) gauge theory with E6 global
symmetry. For such a case the associated coset space will be:
E6(14)
,
SO(10) U (1)
(4.75)
which was first conjectured by [4]. Here we see that there is a natural way to justify29 the existence of such coset space! But this is not the end of the story. Let us look at the next element in
the above row of the magic square. The symmetric subgroups of SU(6) are:
Sp(3),
SU(4) SU(2) U (1),
SU(4),
SU(3) U (3)
(4.76)
SU (6)
where Sp(3) was used earlier to build a real space Sp(3) whereas SU(3) U (3) was used in a
different sequence of the magic square to construct a Khler manifold (4.32). Out of the remaining ones we can build a new non-compact coset space:
SU(4, 2)
,
SU(4) SU(2) U (1)
(4.77)
which in fact does exist in supergravity literature as target space of some sigma model of N = 2
supergravity. Thus a new sequence, not realised directly in the magic square, will be:
which could in principle be embedded in the magic square using the Rozenfeld-Tits constructions [7]. For some more details about these U (1) quotients the readers may want to look up [26].
4.6. A note on holomorphic F -functions
In the previous subsections we discussed the issue of F -functions that could be used to determine the metric on the quaternionic Khler manifolds. In this section we will complete the
29 One can view the coset (4.75) as a plane in the sense of projective geometry. The elements of this plane belong to
certain Jordan pair such that one can define points and lines along with an incidence relation among them. It turns out
that the group E6(14) acts transitively on points and the stability group of a fixed point is SO(10) U (1), thus realising
the correspondence between the plane and the coset space (4.75) (see [62] for details).
74
Table 2
Rank 2 homogeneous special real spaces and their corresponding rank 3 and rank 4 Khler and quaternionic spaces
respectively associated with the magic square
X(P , q)
Real
H (P , q)
Khler
V (P , q)
Quaternionic
X(1, 1)
X(1, 2)
X(1, 4)
X(1, 8)
SL(3, R)/SO(3)
SL(3, C)/SU(3)
SU (6)/Sp(3)
E6 /F4
H (1, 1)
H (1, 2)
H (1, 4)
H (1, 8)
Sp(3)/U (3)
SU(3, 3)/SU(3) SU(3) U (1)
SO (12)/SU(6) U (1)
E7 /E6 U (1)
V (1, 1)
V (1, 2)
V (1, 4)
V (1, 8)
F4 /Sp(3) SU(2)
E6 /SU(6) SU(2)
E7 /SO(12) SU(2)
E8 /E7 SU(2)
analysis by postulating the procedure to determine the F -function for any given Khler manifolds. Although the following analysis is standard (see for example [5,23]) the F -functions for
En and F4 cases have not been explicitly presented anywhere.30
Throughout this section, we use the canonical parametrization introduced by [23] and the third
reference of [16] but where all indices are shifted by one unit in order to fit our notation. The
indices A, B, C = 2, . . . , n + 1 have been decomposed into indices 2, 3, and m, where and
m take respectively q + 1 and r values.
From [23], we know the form of the cubic functions C(h) in terms of scalar fields hA associated to the real manifolds of rank 1 and 2:
3 1 2
C(h) = dABC hA hB hC = h2 h2 h
2

1 3 3
1 m 2
3
3
2
m n
(4.78)
+
h
3( )mn h h h ,
h
+
h 3h
2
2
2
with {3, . . . , n + 1} and where the gamma matrices ( )mn are viewed as r r matrices
generating a (q + 1)-dimensional Clifford algebra denoted C(q + 1, 0).
The coefficients dABC can also be used to describe Khler manifolds.31 By imposing the
following conditions on the symmetric tensor dABC [23]:
1
d333 = ,
2
d3 = d3mm = 0,
dmm = 0,
(4.79)
we construct the holomorphic functions F (X I ), in terms of complex variables X I , associated to

Khler manifolds that are in the image of an r-map:
XA XB XC
F (X I ) = idABC
X1

i 2 3 1 2 2
1 3
= 1 X X X
+ X 3 + 3( )mn X X m X n .
2
X
2
(4.80)
As explained in the third reference of [16], these conditions constrain the allowed values of q to
1, 2, 4 and 8. Since r = 2q and n = 3(q + 1) for Khler manifolds, these are exactly the spaces
corresponding to the magic square with n = 6, 9, 15, 27 [4].
These Khler manifolds are respectively associated to the Jordan algebras J 3 (R), J 3 (C),
3
J (H), and J 3 (O). They were classified in [5]: the Khler H (P , q) spaces generate quaternionic
30 See however Eqs. (3.38) to (3.42) in the recent paper [63]. We thank Sergio Ferrara for pointing this to us. It will be
interesting to relate these values to the ones that we determine here.
31 Note that the use of the canonical parametrisation defines the tensor d
ABC up to arbitrary O(n 1) rotations.
75
Table 3
C(q + 1, 0) represents real Clifford algebras, K(n) are n n matrices with entries over the field
K and Dq+1 represents the real dimension of an irreducible representation of the Clifford algebra
q
C(q + 1, 0)
Dq+1
1
2
4
8
R(2)
C(2)
H (2) H (2)
R(16) {R R}
2
4
8
16
Table 4
Classification of generating matrices of Clifford algebras
Algebra
Matrix dimensions
Generating matrices
C2
C3
C4
C5
C6
C7
C8
C9
22
22
44
44
88
88
16 16
16 16
i2 , i3
1 , 2 , 3
1 , 2 , 3 , 4
i1 , i2 , i3 , i4
ij , j = 1, 2, 3, . . . , 6
j , j = 1, 2, 3, . . . , 7
j , j = 1, 2, 3, . . . , 8
ij , j = 1, 2, 3, . . . , 9
V (P , q) spaces [3] under c-map. This in turn emerge from the real X(P , q) manifolds under the
r-map [23] (see Table 2 for a list of relevant coset spaces).
The trivial case q = 0 with n = 3, which is also generated by the above restriction, is part
of the Khler K(P , P ) space and is associated with W (P , P ) quaternionic manifolds. P and P
represents the multiplicity of each irreducible representations of the Clifford algebras which are
listed in Table 3:
We restrict our study to q > 0 cases. In order to classify all F -functions associated to H (P , q),
one needs to consider all gamma matrices generating a (q + 1)-dimensional real Clifford algebra
with positive metric. This classification was done in [64], see Table 4 above for the relevant
cases.
Solutions are characterised by specifying the multiplicities P and P of each irreducible representations of the Clifford algebras. In all cases we will discuss, we will consider P = 0 and
P = 1. The generating matrices i used in the above table are simply the Pauli matrices:

0 1
0 i
1 0
1 =
(4.81)
,
2 =
,
3 =
.
1 0
i 0
0 1
The i matrices are the Dirac Gamma matrices made out of the sigma matrices in the standard
way. The i matrices are in turn made of the i matrices in the following way:

0
ij
I
0 I4
0
j =
,
6 = 4
, j = 1, 2, 3, 4, 5.
,
7 =
ij
0
I4 0
0 I4
(4.82)
Finally the j are similarly constructed using the j matrices in exactly the same way as above
with j running from j = 1, 2, . . . , 7. The other two matrices 8 and 9 are constructed by I8
like 6 and 7 , respectively.
We associate Cn with C(q+2) . Hence, C3 is associated to R(2) C(q + 1, 0)q=1 , C4 to C(2),
etc. This association allows us to generate a (q + 1)-dimensional Clifford algebra with r r
76
basis that satisfy simultaneously the condition imposed in (4.79) on the gamma matrices, i.e.,
( )mm = 0. Thus, say we have 1 , 2 , 3 and we impose the condition ( )mm = 0, the term
( )mn will therefore be equal to zero when it comes to 3 and we will be left with two (2 2)
matrices, i.e., 1 , 2 to span R(2) as required.
We are now ready to construct the F (X I ) holomorphic functions for each Khler spaces
associated to the magic square. For the Khler space associated with G2 coset we already gave
the F -function in (3.27), and for the coset associated with Sp(n + 1) we know that there is no
Khler space (see Section 3.1 for details).
Khler space H (1, 1):
For H (1, 1), q = 1, r = 2 and n = 6. Hence A, B, C = 2, . . . , 7. In addition, {4, . . . , 7}
and takes exactly q + 1 values say = 4, 5 whereas m {4, . . . , 7} and takes r values for
instance m = 5, 6. The quantity takes all values in {3, . . . , 7}. The matrices generating the
Clifford algebra R(2) C3 would be 1 , 2 according to the previous argument and we shall
rename them . Furthermore F (X I ) iX 1 F (X I ):
3 1 2
1 2
1 3
F XI = X2 X2 X3 X2 X7 + X3
2
2
2
4 m n
5 m n
+ 3(4 )mn X X X + 3(5 )mn X X X .
(4.83)

For H (1, 2), q = 2, r = 4, n = 9, {4, . . . , 10} and takes 3 values say 4, 5, 6 and m
{4, . . . , 10} takes 4 values say 7, 8, 9, 10. {3, . . . , 10} and the Clifford algebra would be
generated by 2 , 3 , 4 which we rename :
3 1 2
2
1
1 3
F X I = X 2 X 2 X 3 X 2 X 10 + X 3
2
2
2
4 m n
+ 3(4 )mn X X X + 3(5 )mn X 5 X m X n + 3(6 )mn X 6 X m X n .
(4.84)

For H (1, 4), q = 4, r = 8, n = 15, {4, . . . , 16} and takes 5 values say 4, 5, 6, 7, 8 and
m {4, . . . , 16} takes 8 values say 9, 10, 11, 12, 13, 14, 15, 16. {3, . . . , 16} and the Clifford algebra would be generated by five (8 8) elements of C6 , i.e., ij with j = 1, . . . , 5
which we rename i :
3 1 2
2
1
1 3
F X I = X 2 X 2 X 3 X 2 X 16 + X 3
2
2
2
4 m n
8 m n
+ 3i(4 )mn X X X + + 3i(8 )mn X X X .
(4.85)

For H (1, 8), q = 8, r = 16, n = 27, {4, . . . , 28} and takes 9 values and m {4, . . . , 28}
takes 16 values. {3, . . . , 28} and the Clifford algebra would be generated by nine (16
16) elements of C9 , i.e., j with j = 1, . . . , 9 which we rename :
3 1 2
2
1
1 3
F X I = X 2 X 2 X 3 X 2 X 28 + X 3
2
2
2
4 m n
12 m n
+ 3(4 )mn X X X + + 3(12 )mn X X X .
(4.86)
77
The above analysis therefore summarises all the F -functions that we need to determine the
Khler spaces. To get the corresponding quaternionic spaces, we use the metric given in (3.29)
for each of the four cases. With this therefore we have the complete picture of all the quaternionic
and the Khler manifolds in the magic square.
5. Summary, discussions and future directions
In this paper we hopefully gave a new way to study the magic square in mathematics and
string theory that is not based on the dimensional reduction of supergravity theories. Our method
relies on the existence of constrained instantons in certain N = 2 gauge theories with exceptional global symmetries. These theories are not asymptotically free and are at strong coupling.
This means that a simple YangMills description may not suffice and we might even lack a Lagrangian description of these theories. Nevertheless we have ample evidence that these theories
exist: via SeibergWitten curves, F-theory and possible quaternionic formulations of low energy
descriptions.
Viewing them as sectors of SeibergWitten theories, the exceptional global symmetries form
non-trivial fixed points of renormalisation group flows. This is well known and they lead to the
following sequence of theories:
E8 E7 E6 D4 A2 A1 {0}.
(5.1)
Our idea of sequential gauging is partially motivated by the above sequence. The SU(2) constrained instanton which is also a semilocal instanton for our case is constructed by gauging
an SU(2) subgroup of the global group. The U (1) part of ungauged global groupthat also
contains the monodromy associated with a dyonic pointis then further gauged to construct
semilocal strings in the model. These two process give us quaternionic and Khler spaces that
are related by a c-map. Once we have these spaces, the real space associated to the Khler space
can be easily constructed.
Our whole analysis therefore depends on the existence of one instanton moduli space in these
theories. In the absence of a proper Lagrangian description we cannot give a concrete construction of these instantons solutions of course, but moduli space can still be constructed. Existence
of SeibergWitten curves also means that we have added all the instanton contributions in the
path-integral. Recall that the instanton contributions to the SeibergWitten prepotential FSW can
be written as:

1
eS k(4NF ) ,
FSW = Fclassical + Fone-loop +
(5.2)
2i
k=1M
where Mk is the moduli space of k-instantons, is the volume form, S is the instanton action,
NF is the number of flavors and is the same PauliVillars scale that we used earlier. It is
therefore an interesting question to ask how instantons in these gauge theories with En , F4 global
symmetries give us the right SeibergWitten curves. Note however that if one breaks the En
symmetry by giving masses to quarks and keeping the gauge coupling finite, one may hope to
get a convergent expression for the instanton partition function. However, to show that an analytic
continuation to the En symmetric point would make sense, requires more work.32 We leave this
aspect for future work.
32 We thank Nikita Nekrasov for comments on this.
78
Another interesting direction is to look for theories with exceptional gauge symmetries. Incidentally one instanton moduli spaces will be the same for these theoriesits just an embedding
of SU(2) in exceptional gauge groups33 but the corresponding curves will be different. We gave
one example before. Another example would be a theory with F4 gauge symmetry. Such a theory
with one massless hypermultiplet has the following SeibergWitten curve34 [61]:

2

y 2 = x 2 b12 x 2 b22 x 2 b32 x 2 b42 x 4 12 ,
(5.3)
where bi are the projections of the weights: (1000), (1100), (0111) and (0011). It would be
interesting to study these theories with more than one massless hypermultiplets.
One final issue is the classification of de-Wit and Van-Proeyen [23] that completes Alekseevskiis classification of quaternionic manifolds [3]. We have shown that we can reproduce all
of Alekseevskiis symmetric manifolds and few more of de-Wit and Van-Proeyen also. However
we have not investigated enough to see whether we could reproduce all other manifolds in the
classification of [23]. In fact its an interesting question to ask whether these manifolds have a
coset structure like the other manifolds in the classification. We leave this for future work.
Acknowledgements
We would like to thank Rhiannon Gwyn for initial collaboration; Sergio Ferrara, Mark Hindmarsh, Sheldon Katz, Joseph Minahan, Nikita Nekrasov, Savdeep Sethi, Ulrich Theis, David
Tong, Tanmay Vachaspati and Alexei Yung for helpful correspondences and especially Ori
Ganor, Tom Kephart and V.P. Nair for important clarifications and discussions. The work of K.D.
is supported by a NSERC grant. The work of V.H. is partially supported by a NSERC grant and
the work of A.W. is supported in part by a NSERC grant and in part by the university research
grant.
References
[1] M. Berger, Sur les groupes dholonomie des varits connexion affine et des varits riemannienes, Bull. Soc.
Math. France 83 (1955) 279;
M. Berger, Sur les groupes dholonomie homognes des varits riemannienes, C. R. Acad. Sci. Paris A 262 (1966)
1316.
[2] J.A. Wolf, Complex homogeneous contact manifolds and quaternionic symmetric spaces, J. Math. Mech. 14 (1965)
1033.
[3] D.V. Alekseevskii, Riemannian manifolds with exceptional holonomy groups, Funct. Anal. Appl. 2 (1968) 97;
D.V. Alekseevskii, Compact quaternion spaces, Funct. Anal. Appl. 2 (1968) 106;
D.V. Alekseevskii, Classification of quaternionic spaces with transitive solvable group of motions, Math. USSR
Izv. 9 (1975) 297.
[4] M. Gunaydin, G. Sierra, P.K. Townsend, Exceptional supergravity theories and the magic square, Phys. Lett. B 133
(1983) 72;
D.V. Alekseevskii, The geometry of N = 2 MaxwellEinstein supergravity and Jordan algebras, Nucl. Phys. B 242
(1984) 244;
D.V. Alekseevskii, Gauging The D = 5 MaxwellEinstein supergravity theories: More on Jordan algebras, Nucl.
Phys. B 253 (1985) 573.
[5] S. Cecotti, Homogeneous Khler manifolds and T algebras in N = 2 supergravity and superstrings, Commun. Math.
Phys. 124 (1989) 23;
33 Recall (E ) = (F ) = Z.
3 n
3 4
34 This could also be derived from (4.57) with appropriate polynomial P .
R
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
79
S. Cecotti, S. Ferrara, L. Girardello, Geometry of type II superstrings and the moduli of superconformal field theories, Int. J. Mod. Phys. A 4 (1989) 2475.
S. Ferrara, S. Sabharwal, Quaternionic manifolds for type II superstring vacua of CalabiYau spaces, Nucl. Phys.
B 332 (1990) 317;
S. Ferrara, S. Sabharwal, Dimensional reduction of type II superstrings, Class. Quantum Grav. 6 (1989) L77.
H. Freudenthal, Beziehungen der E7 und E8 zur Oktavenebene, I, Indag. Math. 16 (1954) 218230;
H. Freudenthal, Beziehungen der E7 und E8 zur Oktavenebene, II, Indag. Math. 16 (1954) 363368;
H. Freudenthal, Beziehungen der E7 und E8 zur Oktavenebene, III, Indag. Math. 17 (1955) 151157;
H. Freudenthal, Beziehungen der E7 und E8 zur Oktavenebene, IV, Indag. Math. 17 (1955) 277285;
H. Freudenthal, Beziehungen der E7 und E8 zur Oktavenebene, VIX, Indag. Math. 21 (1959) 165201, 447474;
H. Freudenthal, Beziehungen der E7 und E8 zur Oktavenebene, X, XI, Indag. Math. 25 (1963) 457487;
B.A. Rosenfeld, Geometrical interpretation of the compact simple Lie groups of the class En , Dokl. Akad. Nauk.
SSSR 106 (1956) 600603 (in Russian);
J. Tits, Le plan projectif des octaves et les groupes exceptionnels E6 et E7 , Bull. Acad. Roy. Belg. Sci. 40 (1954)
2940.
N. Seiberg, E. Witten, Electricmagnetic duality, monopole condensation, and confinement in N = 2 supersymmetric YangMills theory, Nucl. Phys. B 426 (1994) 19, hep-th/9407087;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 485, Erratum;
N. Seiberg, E. Witten, Monopoles, duality and chiral symmetry breaking in N = 2 supersymmetric QCD, Nucl.
Phys. B 431 (1994) 484, hep-th/9408099.
I. Affleck, On constrained instantons, Nucl. Phys. B 191 (1981) 429.
A. Achucarro, T. Vachaspati, Semilocal cosmic strings, Phys. Rev. D 44 (1991) 3067;
A. Achucarro, T. Vachaspati, Semilocal and electroweak strings, Phys. Rep. 327 (2000) 347, hep-ph/9904229.
I.I. Pjateckii-Sapiro, The structure of j -algebras, Izv. Akad. Nauk SSSR Ser. Mat. 26 (1962) 453;
I.I. Pjateckii-Sapiro, Am. Math. Soc. Transl. 55 (1966) 207;
S.G. Gindikin, I.I. Pjateckii-Sapiro, E.B. Vinberg, Homogeneous Khler manifolds, in: Geometry of Homogeneous
Bounded Domains, IME, Italy, 1968, pp. 387;
E.B. Vinberg, S.G. Gindikin, I.I. Pjateckii-Sapiro, Classification and canonical realization of complex homogeneous
bounded domains, Trudy Moskov. Mat. Obshch. 12 (1963) 359.
P. Jordan, ber eine klasse nichtassoziativer hyperkomplexer algebren, Nachr. Ges. Wiss. Gttingen 569 (1932)
569575.
L. Alvarez-Gaume, D.Z. Freedman, Geometrical structure and ultraviolet finiteness in the supersymmetric sigma
model, Commun. Math. Phys. 80 (1981) 443.
B. Zumino, Supersymmetry and Khler manifolds, Phys. Lett. B 87 (1979) 203.
E. Witten, J. Bagger, Quantization of Newtons constant in certain supergravity theories, Phys. Lett. B 115 (1982)
202.
J. Bagger, E. Witten, Matter couplings in N = 2 supergravity, Nucl. Phys. B 222 (1983) 1;
B. de Wit, A. Van Proeyen, Potentials and symmetries of general gauged N = 2 supergravity: YangMills models,
Nucl. Phys. B 245 (1984) 89;
E. Cremmer, C. Kounnas, A. Van Proeyen, J.P. Derendinger, S. Ferrara, B. de Wit, L. Girardello, Vector multiplets
coupled to N = 2 supergravity: Super-Higgs effect, flat potentials and geometric structure, Nucl. Phys. B 250 (1985)
385.
A. Strominger, Loop corrections to the universal hypermultiplet, Phys. Lett. B 421 (1998) 139, hep-th/9706195;
H. Gunther, C. Herrmann, J. Louis, Quantum corrections in the hypermultiplet moduli space, Fortschr. Phys. 48
(2000) 119, hep-th/9901137;
R. Bohm, H. Gunther, C. Herrmann, J. Louis, Compactification of type IIB string theory on CalabiYau threefolds,
Nucl. Phys. B 569 (2000) 229, hep-th/9908007.
D. Robles-Llana, F. Saueressig, S. Vandoren, String loop corrected hypermultiplet moduli spaces, JHEP 0603 (2006)
081, hep-th/0602164.
D. Robles-Llana, M. Rocek, F. Saueressig, U. Theis, S. Vandoren, Nonperturbative corrections to 4D string theory
effective actions from SL(2, Z) duality and supersymmetry, Phys. Rev. Lett. 98 (2007) 211602, hep-th/0612027;
D. Robles-Llana, F. Saueressig, U. Theis, S. Vandoren, Membrane instantons from mirror symmetry, arXiv:
0707.0838 [hep-th].
N. Berkovits, W. Siegel, Superspace effective actions for 4D compactifications of heterotic and type II superstrings,
Nucl. Phys. B 462 (1996) 213, hep-th/9510106;
N. Seiberg, E. Witten, Gauge dynamics and compactification to three dimensions, hep-th/9607163;
80
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
B.R. Greene, D.R. Morrison, C. Vafa, A geometric realization of confinement, Nucl. Phys. B 481 (1996) 513,
hep-th/9608039;
I. Antoniadis, S. Ferrara, R. Minasian, K.S. Narain, R**4 couplings in M- and type II theories on CalabiYau
spaces, Nucl. Phys. B 507 (1997) 571, hep-th/9707013.
I. Antoniadis, R. Minasian, S. Theisen, P. Vanhove, String loop corrections to the universal hypermultiplet, Class.
Quantum Grav. 20 (2003) 5079, hep-th/0307268.
S.V. Ketov, Summing up D-instantons in N = 2 supergravity, Nucl. Phys. B 649 (2003) 365, hep-th/0209003;
S.V. Ketov, Instanton-induced scalar potential for the universal hypermultiplet, Nucl. Phys. B 656 (2003) 63, hepth/0212003;
S.V. Ketov, D-instantons and matter hypermultiplet, Phys. Lett. B 558 (2003) 119, hep-th/0302001;
L. Anguelova, M. Rocek, S. Vandoren, Quantum corrections to the universal hypermultiplet and superspace, Phys.
Rev. D 70 (2004) 066001, hep-th/0402132;
N. Halmagyi, I.V. Melnikov, S. Sethi, Instantons, hypermultiplets and the heterotic string, arXiv: 0704.3308 [hepth].
B. de Wit, A. Van Proeyen, Special geometry, cubic polynomials and homogeneous quaternionic spaces, Commun.
Math. Phys. 149 (1992) 307, hep-th/9112027.
P. Chen, K. Dasgupta, K. Narayan, M. Shmakova, M. Zagermann, Brane inflation, solitons and cosmological solutions: I, JHEP 0509 (2005) 009, hep-th/0501185;
K. Dasgupta, H. Firouzjahi, R. Gwyn, Lumps in the throat, JHEP 0704 (2007) 093, hep-th/0702193.
M. Hindmarsh, R. Holman, T.W. Kephart, T. Vachaspati, Generalized semilocal theories and higher Hopf maps,
Nucl. Phys. B 404 (1993) 794, hep-th/9209088.
K. Higashijima, M. Nitta, Supersymmetric nonlinear sigma models as gauge theories, Prog. Theor. Phys. 103 (2000)
635, hep-th/9911139;
M. Arai, M. Nitta, Hyper-Khler sigma models on (co)tangent bundles with SO(n) isometry, Nucl. Phys. B 745
(2006) 208, hep-th/0602277;
Y. Isozumi, M. Nitta, K. Ohashi, N. Sakai, Construction of non-Abelian walls and their complete moduli space,
Phys. Rev. Lett. 93 (2004) 161601, hep-th/0404198.
K. Dasgupta, J.P. Hsu, R. Kallosh, A. Linde, M. Zagermann, D3/D7 brane inflation and semilocal strings, JHEP 0408
(2004) 030, hep-th/0405247.
A. Hanany, D. Tong, Vortices, instantons and branes, JHEP 0307 (2003) 037, hep-th/0306150;
A. Hanany, D. Tong, On monopoles and domain walls, Commun. Math. Phys. 266 (2006) 647, hep-th/0507140;
M. Shifman, A. Yung, Non-Abelian semilocal strings in N = 2 supersymmetric QCD, Phys. Rev. D 73 (2006)
125012, hep-th/0603134;
M. Eto, et al., On the moduli space of semilocal strings and lumps, arXiv: 0704.2218 [hep-th].
E. Witten, Small instantons in string theory, Nucl. Phys. B 460 (1996) 541, hep-th/9511030.
O.J. Ganor, A. Hanany, Small E8 instantons and tensionless non-critical strings, Nucl. Phys. B 474 (1996) 122,
hep-th/9602120.
A. Sen, F-theory and orientifolds, Nucl. Phys. B 475 (1996) 562, hep-th/9605150;
T. Banks, M.R. Douglas, N. Seiberg, Probing F-theory with branes, Phys. Lett. B 387 (1996) 278, hep-th/9605199.
M.R. Douglas, Branes within branesee, in: Cargese 1997, Strings, Branes and Dualities (1995) 267, hep-th/9512077;
M.R. Douglas, Gauge fields and D-branes, J. Geom. Phys. 28 (1998) 255, hep-th/9604198.
N. Dorey, T.J. Hollowood, V.V. Khoze, M.P. Mattis, S. Vandoren, Multi-instanton calculus and the AdS/CFT correspondence in N = 4 superconformal field theory, Nucl. Phys. B 552 (1999) 88, hep-th/9901128;
N. Dorey, T.J. Hollowood, V.V. Khoze, M.P. Mattis, The calculus of many instantons, Phys. Rep. 371 (2002) 231,
hep-th/0206063.
I. Antoniadis, B. Pioline, Higgs branch, hyper-Khler quotient and duality in SUSY N = 2 YangMills theories,
Int. J. Mod. Phys. A 12 (1997) 4907, hep-th/9607058;
L. Anguelova, M. Rocek, S. Vandoren, Hyper-Khler cones and orthogonal Wolf spaces, JHEP 0205 (2002) 064,
hep-th/0202149;
B. de Wit, M. Rocek, S. Vandoren, Gauging isometries on hyper-Khler cones and quaternion-Khler manifolds,
Phys. Lett. B 511 (2001) 302, hep-th/0104215;
B. de Wit, M. Rocek, S. Vandoren, Hypermultiplets, hyper-Khler cones and quaternion-Khler geometry,
JHEP 0102 (2001) 039, hep-th/0101161;
S. Vandoren, Instantons and quaternions, hep-th/0009150.
A. Swann, Hyper-Khler and quaternionic Khler geometry, Math. Ann. 289 (1991) 421.
G. t Hooft, Computation of the quantum effects due to a four-dimensional pseudoparticle, Phys. Rev. D 14 (1976)
3432;
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
81
G. t Hooft, Phys. Rev. D 18 (1978) 2199, Erratum;

Y. Frishman, S. Yankielowicz, Large order behavior of perturbation theory and mass terms, Phys. Rev. D 19 (1979)
540.
A.P. Balachandran, V.P. Nair, N. Panchapakesan, S.G. Rajeev, Low mass solitons from fractional charges in QCD,
Phys. Rev. D 28 (1983) 2830.
W.J. MacKay, J. Patera, Tables of Dimensions, Indices, and Branching Rules for Representations of Simple Lie
Algebras, Marcel Dekker, New York, 1981;
R. Slansky, Group theory for unified model building, Phys. Rep. 79 (1981) 1.
M.R. Gaberdiel, B. Zwiebach, Exceptional groups from open strings, Nucl. Phys. B 518 (1998) 151, hepth/9709013.
M. Rocek, C. Vafa, S. Vandoren, Quaternion-Khler spaces, hyper-Khler cones, and the c-map, math.DG/0603048.
B. de Wit, A. Van Proeyen, Special geometry, cubic polynomials and homogeneous quaternionic spaces, Commun.
Math. Phys. 149 (1992) 307, hep-th/9112027.
E. Cartan, Oeuvres compltes, GauthierVillars, Paris, 1952.
J.E. Humphreys, Introduction to Lie Algebras and Representation Theory, Springer, NY, 1972.
J. Beckers, V. Hussin, P. Winternitz, Nonlinear equations with superposition formulas and the exceptional group
G2 . I. Complex and real forms of g2 and their maximal subalgebras, J. Math. Phys. 27 (1986) 2217;
J. Beckers, V. Hussin, P. Winternitz, Nonlinear equations with superposition formulas and the exceptional group
G2 . II. Classification of the equations, J. Math. Phys. 28 (1987) 520.
G.M. Dixon, Division Algebras: Octonions, Quaternions, Complex Numbers and the Algebraic Design of Physics,
Kluwer Academic, 1994.
J.M. Landsberg, L. Manivel, The projective geometry of Freudenthals magic square, math.AG/9908039;
A. Sudbery, Division algebras, J. Phys. A: Math. Gen. 17 (1984) 939;
E.J. Lohmus, E. Paal, L. Sorgsepp, Nonassociative Algebras in Physics, Hadronic Press, 1994.
J.A. Minahan, D. Nemeschansky, An N = 2 superconformal fixed point with E(6) global symmetry, Nucl. Phys.
B 482 (1996) 142, hep-th/9608047;
J.A. Minahan, D. Nemeschansky, Superconformal fixed points with E(n) global symmetry, Nucl. Phys. B 489
(1997) 24, hep-th/9610076.
A. Sen, F-theory and orientifolds, Nucl. Phys. B 475 (1996) 562, hep-th/9605150.
K. Dasgupta, S. Mukhi, F-theory at constant coupling, Phys. Lett. B 385 (1996) 125, hep-th/9606044.
W. Lerche, N.P. Warner, Polytopes and solitons in integrable, N = 2 supersymmetric LandauGinzburg theories,
Nucl. Phys. B 358 (1991) 571.
N. Seiberg, IR dynamics on branes and spacetime geometry, Phys. Lett. B 384 (1996) 81, hep-th/9606017.
K. Dasgupta, C. Herdeiro, S. Hirano, R. Kallosh, D3/D7 inflationary model and M-theory, Phys. Rev. D 65 (2002)
126002, hep-th/0203019.
E. Witten, Solutions of four-dimensional field theories via M-theory, Nucl. Phys. B 500 (1997) 3, hep-th/9703166.
A. Hanany, E. Witten, Type IIB superstrings, BPS monopoles, and three-dimensional gauge dynamics, Nucl. Phys.
B 492 (1997) 152, hep-th/9611230.
O. Aharony, A. Hanany, B. Kol, Webs of (p, q) 5-branes, five-dimensional field theories and grid diagrams,
JHEP 9801 (1998) 002, hep-th/9710116.
K. Dasgupta, S. Mukhi, BPS nature of 3-string junctions, Phys. Lett. B 423 (1998) 261, hep-th/9711094.
A. Sen, String network, JHEP 9803 (1998) 005, hep-th/9711130.
J.A. Bryan, S.M. Carroll, T. Pyne, A texture bestiary, Phys. Rev. D 50 (1994) 2806, hep-ph/9312254.
E.J. Martinec, N.P. Warner, Integrable systems and supersymmetric gauge theory, Nucl. Phys. B 459 (1996) 97,
hep-th/9509161;
A. Klemm, W. Lerche, P. Mayr, C. Vafa, N.P. Warner, Self-dual strings and N = 2 supersymmetric field theory,
Nucl. Phys. B 477 (1996) 746, hep-th/9604034;
W. Lerche, N.P. Warner, Exceptional SW geometry from ALE fibrations, Phys. Lett. B 423 (1998) 79, hepth/9608183.
A. Klemm, W. Lerche, S. Yankielowicz, S. Theisen, Simple singularities and N = 2 supersymmetric YangMills
theory, Phys. Lett. B 344 (1995) 169, hep-th/9411048;
A. Klemm, W. Lerche, S. Yankielowicz, S. Theisen, On the monodromies of N = 2 supersymmetric YangMills
theory, hep-th/9412158;
P.C. Argyres, M.R. Plesser, A.D. Shapere, The Coulomb phase of N = 2 supersymmetric QCD, Phys. Rev. Lett. 75
(1995) 1699, hep-th/9505100;
A. Hanany, Y. Oz, On the quantum moduli space of vacua of N = 2 supersymmetric SU (N (c)) gauge theories,
Nucl. Phys. B 452 (1995) 283, hep-th/9505075.
82
[61] U.H. Danielsson, B. Sundborg, Exceptional equivalences in N = 2 supersymmetric YangMills theory, Phys. Lett.
B 370 (1996) 83, hep-th/9511180.
[62] P. Truini, G. Olivieri, L.C. Biedenharn, The Jordan pair content of the magic square and the geometry of the scalars
in N = 2 supergravity, Lett. Math. Phys. 9 (1985) 255.
[63] S. Ferrara, E.G. Gimon, R. Kallosh, Magic supergravities, N = 8 and black hole composites, Phys. Rev. D 74 (2006)
125018, hep-th/0606211.
[64] C.P. Poole Jr., H.A. Farach, PauliDirac matrix generators of Clifford algebras, Found. Phys. 12 (1982) 719;
D. Li, C.P. Poole, H.A. Farach, A general method of generating and classifying Clifford algebras, J. Math. Phys. 27
(1986) 1173.

Supersymmetry relations and MHV amplitudes

in superstring theory
Stephan Stieberger a, , Tomasz R. Taylor a,b,c
a Max-Planck-Institut fr Physik, Werner-Heisenberg-Institut, 80805 Mnchen, Germany
b Arnold-Sommerfeld-Center for Theoretical Physics, Ludwig-Maximilians-Universitt Mnchen,
80333 Mnchen, Germany

c Department of Physics, Northeastern University, Boston, MA 02115, USA
Received 16 August 2007; accepted 28 September 2007

Available online 10 October 2007
Abstract
We discuss supersymmetric Ward identities relating various scattering amplitudes in type I open superstring theory. We show that at the disk level, the form of such relations remains exactly the same, to all
orders in , as in the low-energy effective field theory describing the 0 limit. This result holds in
D = 4 for all compactifications, even for those that break supersymmetry. We apply SUSY relations to
the computations of N-gluon MHV superstring amplitudes, simplifying the existing results for N 6 and
deriving a compact expression for N = 7.
PACS: 11.25.Wx; 11.30.Pb; 12.38.Bx
Keywords: Gauge theories; Gluon scattering; Superstrings; Conformal field theory
1. Introduction
For the last thirty years, multi-gluon amplitudes and their supersymmetric variants have
been extensively studied in the framework of quantum relativistic theory of gauge fields. These
amplitudes are very important from both theoretical and experimental points of view because
they describe the scattering processes underlying hadronic jet production at high energy collid* Corresponding author.
E-mail address: stieberg@mppmu.mpg.de (S. Stieberger).

doi:10.1016/j.nuclphysb.2007.09.033
84
S. Stieberger, T.R. Taylor / Nuclear Physics B 793 (2008) 83113
ers. The modern tools for amplitude computations1 include helicity techniques, string-inspired
color ordering and the use of supersymmetry [35] and recursion relations [6]. More recently,
a substantial progress has been achieved by employing some elements of twistor theory [7]. In
particular, a new type of recursion relations has been constructed, inspired by twistor string theory [8]. Here, similarly to the case of color ordering, string theory comes as a handy tool.
Multi-gluon scattering is also a process of considerable theoretical interest in the framework
of full-fledged superstring theory, and would have a great experimental relevance if the string
mass scale turned out to be low enough to be reached at LHC [9,10]. In this case, some effects
due to the off- and on-shell propagation of the string Regge excitations could be detected in jet
cross sections. The parameter that controls the size of such effects is the Regge slope . As
pointed out in [9,10], a consistent low string mass (small ) scenario must necessarily involve
large extra dimensions [11], therefore massive string excitations come together with Kaluza
Klein states. However, the effects of Regge states dominate over the effects of extra dimensions
for a wide range of parameters, notably in the weak string coupling regime [12].
In two recent papers [13,14], we initiated a systematic study of multi-gluon scattering
processes in open superstring theory, at the semiclassical, disk level of the string worldsheet.
We focussed on one particular, maximum helicity violating (MHV) gluon helicity configuration,
because in the 0 limit, the respective amplitude is described by the well known, simple
formula [15] (written in the notation of [1,2]):

)
AYM (1 , 2 , 3+ , 4+ , . . . , N + ) = ( 2gYM )N2 Tr T a1 T aN M(N
(1.1)
YM ,
where gYM is the properly normalized YangMills coupling constant and
(N )
MYM = i
1 24
.
1 22 33 4 N 1
(1.2)
For N 6 external gluons, we presented the amplitudes in a factorized form, as a product

)
of the zero slope (YangMills) MHV amplitude M(N
YM times a string formfactor involving
(N 3)! generalized hypergeometric functions of kinematic invariants. We argued that the soft
and collinear factorization properties, combined with the Abelian limit, are completely sufficient
to determine all N -gluon MHV amplitudes. Nevertheless, the increasing complexity of string
formfactors has not yet allowed us writing a compact expression describing MHV string amplitudes with arbitrary numbers of gluons.
In this work, we show that the computations of N -gluon string amplitudes can be simplified
by utilizing supersymmetric Ward identities, exactly in the same way as in = 0 gauge theories [3]. In Section 1, we prove that at the disk level, the well-known supersymmetry (SUSY)
relations [3] between amplitudes involving gluons, gluinos and gauge scalars are valid to all orders in . Since the gluonic disk amplitudes are completely determined by the four-dimensional
(D = 4) spacetime part of the underlying world-sheet superconformal field theory (SCFT),
these relations can be used to simplify them even if supersymmetry is broken by the compactification. In particular, for any compactification of open superstring theory, the N -gluon MHV
disk amplitude can be expressed in terms of an amplitude involving N 4 gluons and 4 scalars.
In Section 2, we use this relation to rewrite the results of [13,14] in a non-factorized way, but
in terms of primitive (hypergeometric) integrals. This non-factorized form of amplitudes looks
1 For reviews, see [1] and [2].
85
more promising for constructing a string generalization of MHV rules [8] and it could potentially lead to some soluble recursion relations [16]. In Section 3, we demonstrate the power of
SUSY relations by deriving a compact expression for the seven-gluon MHV string amplitude. In
Section 4 we compute various four-point amplitudes involving scalars, gauginos and vectors and
present an explicit example of a SUSY relation.
2. Spacetime supersymmetry relations and string amplitudes
In superstring theory, the vertices creating gauge bosons, gauginos and gauge scalars are related by supersymmetry transformations. Here, we use the operator product expansion (OPE)
rules in order to evaluate the corresponding contour integrals, with SUSY transformations generated by the appropriate insertions of (spacetime) supercharge operators. We use these transformations in order to derive supersymmetric Ward identities for some amplitudes related to
N -gluon scattering.
This section is divided into four parts. In the first part, we use the so-called doubling
trick [17] to rewrite the commutators of (spacetime) SUSY generators with open string vertices, inserted at the boundary of a disk world-sheet, as the contour integrals on the sphere. In
the second part, we apply these contour integrals in order to determine the SUSY commutators
for the vertices creating full gauge supermultiplet. In the third part, we explain how to use contour deformations in order to derive supersymmetric Ward identities relating various superstring
scattering amplitudes. Finally, in the last part we construct the helicity basis for string vertex
operators, and show that their SUSY transformations agree with the known results in the 0
(YangMills) limit.
2.1. Extended spacetime supersymmetry and SUSY variations on the disk
In the following, we establish SUSY transformation rules on the open string disk world-sheet.
We consider type II or type I superstring compactifications on an internal six-dimensional manifold with 2N or N extended supersymmetry charges in D = 4 spacetime, respectively. In
type II superstring theory, one can construct, on a closed string world-sheet, two sets of N
holomorphic charges QI and Q I , I = 1, . . . , N , with independent actions on the left- and rightmoving closed string modes, respectively. The supercharges QI of the left-moving sector are
given by the contour integrals (fixed time lines on the closed string world-sheet)

dz I
dz I
QI =
(2.1)
Q I =
V (z),
V (z),
2i
2i
of the holomorphic supercurrents VI (z) and VI (z). In D = 4, in the (1/2)-ghost picture [18],
these are given by
1
VI (z) = 1/4 e 2 (z) S (z) I (z),

VI (z) = 1/4 e 2 (z) S (z) I (z),
1
(2.2)
where S , S are the spin fields with the indices (or )

denoting negative (positive) chirality in
four dimensions. The Ramond fields I belong to an internal SCFT with c = 9. Finally, is the
scalar bosonizing the superghost system. The OPEs of the internal Ramond fields I are [19]
(cf. also [20])
86
I (z) J (w) = (z w)3/4 I J + (z w)1/4 J I J (w) + ,

I (z) J (w) = (z w)1/4 I J (w) + O (z w)3/4 ,
(2.3)
with the dimension one currents J I J and dimension 1/2 operators I J . Additional material on
OPEs is given in Appendix A, where we also summarize some basic facts about type II SUSY
I are constructed in the same way from the
algebra. The right-moving supercharges Q I and Q
anti-holomorphic currents VI (z) and V I (z), respectively.

In type I open superstring theory the left- and right-movers are tied together by the world
sheet boundaries and only the total charge
QI = QIa + Q I ,
(2.4)
acting on the open string modes, is preserved. The combination (2.4) enjoys then the open string
SUSY algebra. On an open string world-sheet, some boundary conditions have to be imposed on
the fields. On the disk, which is conformally equivalent to the upper half plane H+ , the following
boundary conditions are imposed on the supersymmetry currents:
VI (z) = VI (z),
z = z .
(2.5)
It is convenient to extend the definition of the currents VI to the full complex plane by using the
so-called doubling trick [17]:

VI (z), z H+ ,
VI (z) =
(2.6)
VI (z), z H .
On the disk H+ a fixed time integral is represented by a half-circle around z = 0 in the upper half
plane. Now, the line (fixed time) integrals on the disk H+ can be combined to full closed contour
integrals on the sphere. In this way, the conserved supersymmetry charges (2.4) can be rewritten
as

dz I
d z I
dz I
V (z) +
V (z) :=
V (z),
QI =
(2.7)
2i
2i
2i
where on the l.h.s., the z and z integrals are over semicircles in the upper and lower half-planes
(H+ and H ), respectively, while the integral on the r.h.s. is over a circle on the full complex
plane (Riemann sphere), see Fig. 1. Hence when converting from type II to type I we can use
the fields from the left-moving closed string sector, extend their definition to the full complex
plane in lines of (2.6) and consider contour integrals in the full complex plane as in Eq. (2.7).
A similar procedure can be applied in the presence of Dp-branes that can further reduce the
Fig. 1. Fixed time lines C1 and C2 on the disk H+ and their extensions to a closed loop C on the sphere.
87
amount of conserved supercharges, however for our purposes, D = 4 type I compactifications

are completely sufficient.
In order to determine the variation of an arbitrary open string vertex operator O(z) (inserted
at the boundary z = z ) under infinitesimal SUSY transformations generated by
I ( I )
QI (I , I ) = QI (I ) + Q
one needs the commutators

I

Q (I ), O(z) :=
I
with QI (I ) = I QI , Q I ( I ) = I Q
(2.8)
dw I
V (w)O(z),
2i I
(2.9)
C (z)
where C (z) is a circle of radius 0 surrounding the point z and V is the supercurrent extended to the full complex plane according to Eq. (2.6). Thus the integral (2.9) effectively picks
up the residuum at w = z, hence it can be determined by using OPE. One also needs similar
commutators with Q I ( I ).
2.2. SUSY transformations of open string vertices
For our purposes, it is completely sufficient to consider a maximally supersymmetric, toroidal
compactification of type I (or type IIA/B) superstring, with N = 4 SUSY in D = 4. Therefore
in the following, we shall specialize to the this case, i.e. I, J = 1, . . . , 4. For N = 4, the internal
fields I have an explicit realization as pure exponentials [19]:
i
1 = e 2 (H1 +H2 +H3 ) ,

i
3 = e 2 (H1 +H2 H3 ) ,
2 = e 2 (H1 H2 H3 ) ,
i
4 = e 2 (H1 H2 +H3 ) .
(2.10)
Furthermore, the fields I J in (2.3) become complex fermions i = eiHi , i+ = eiHi :

12 = 34 = eiH1 ,
13 = 24 = eiH2 ,
14 = 23 = eiH3 .
(2.11)
IJ
The fields
are anti-symmetric w.r.t. to their internal indices I and J .
The massless open string modes on the Dp-brane world-volume form an N = 4 gauge vector
multiplet: three complex scalars i , four gauginos I and one vector A in the adjoint representation of the gauge group. The scalar vertex, in the (1)-ghost picture, reads
V a,j (z, k) = g T a e j eikX ,
(1)
with the complex fermions j =

(1/2)-ghost picture, are
j = 1, . . . , 3,
1 ( 2j +2
2
i 2j +3 ). The gaugino vertex operators, in the
(1/2)
(z, u, k) = g T a e/2 u S I eik X ,
(1/2)
(z, u, k) = g T a e/2 u S I eik X ,
Va,I
V a,I
(2.12)
I = 1, . . . , 4.
(2.13)
The gauge boson vertex operator in the (1)-ghost picture reads:

VAa (z, , k) = gA T a e eikX .
(1)
(2.14)
The open string vertex couplings are

g = (2 )1/2 gYM ,
g = (2 )1/2 1/4 gYM ,
gA = (2 )1/2 gYM
(2.15)
88
Table 1a
Index structure (I, i) and J of the supersymmetry variations [QI (I ), V i ]
Q1
Q2
Q3
Q4
1+
2+
3+
0
0
4
3
0
4
0
2
0
3
2
2
1
3
0
1
0
4
0
0
1
0
0
Table 1b
I ( I ), V i ]
Index structure (I, i) and J of the supersymmetry variations [Q
1
Q
2
Q
3
Q
4
Q
1+
2+
3+
1
0
0
0
1
0
0
0
1
0
0
4
3
0
4
0
2
0
3
2
0
for the scalar, gaugino and vector, respectively [21]. The D = 4 gauge coupling gYM can be
expressed in terms of the ten-dimensional gauge coupling g10 and the dilaton field 10 through
the relation gYM = g10 e10 /2 [21]. In the above definitions, T a are the ChanPaton factors accounting for the gauge degrees of freedom of the two open string ends. Furthermore, the on-shell
constraints k 2 = 0, k/u = 0 are imposed.
The vertices creating gauge bosons, gauginos and gauge scalars are related by N = 4 SUSY
transformations (2.8). The sixteen Grassmann parameters I , I
are chiral spinors of both
SL(2, C) Lorentz group and the internal SO(6) R-symmetry group. Here, we use the OPE rules
in order to evaluate the corresponding contour integrals (2.9). As an example,

I
Q ( I ), Va,J

1
dw 1 (w)
a
= gA T
I S (w) I (w)e 2 (z) u S (z) J (z)eikX
e 2
2i
w=z
gA
= Ta
2
w=z
(z)
dw e(z)
I J
u
eikX
I
2i (w z)1/4
(z w)0 (w z)3/4

gA
= T a e I J I
u eikX ,
2
where we used the OPEs given in (2.3) and in Appendix A. All remaining commutators, obtained
in a similar way, are collected below.
Applying the transformation (2.9) on the scalar vertex operators (2.12), we obtain:

I
1

Q (I ), V a,i = g T a e 2 k I S J eikX ,

I
1
S J eikX .
Q ( I ), V a,i = g T a e 2 k I
(2.16)
The respective internal indices J for the Ramond fields (gaugino vertices) are given in Tables 1a
and 1b.
For gaugino vertices (2.13), we obtain

I
Q (I ), Va,J = g T a e (I u) I J eikX ,
89
gA a I J
I ( I ), Va,J =
Q
u eikX ,
T e I
2
I
gA a I J
Q (I ), V a,J = T e I u eikX ,
2
I

( I ), V a,J = g T a e ( I u)
Q
I J eikX ,
(2.17)
with the complex fermions I J defined in (2.11). Finally, for the vectors (2.14), we obtain
g

I
1
Q (I ), VAa = T a e 2 k I S I eikX ,
2
I
g a 1

( I ), VAa = T e 2 k I S I eikX ,
Q
(2.18)
) and (

).
) = 12 (

with ( ) = 12 (

The results (2.16)(2.18) can be rewritten without referring to a particular ghost picture, in
the following form:
I

Q (I ), V a,i (z, k) = V a,J (z, v,

k), v = k I ,
I

( I ), V a,i (z, k) = Va,J (z, v, k), v = k I

Q
,

I
Q (I ), Va,J (z, u, k) = (I u)V a,j (z, k),

I
( I ), V a,J (z, u,
k) = ( I u)V
a,j (z, k),
Q

I
1 IJ

k) = VAa (z, , k), = I u ,
Q (I ), V a,J (z, u,
2
I

( I ), Va,J (z, u, k) = 1 I J VAa (z, , k), = I

Q
u ,
2
I

1
Q (I ), VAa (z, , k) = Va,I (z, v, k), v = k I ,
2
I

1
( I ), VAa (z, , k) = V a,I (z, v,
Q
(2.19)
k), v = k I .
2.3. Contour deformations and SUSY Ward identities

In this section we consider the action of the SUSY generator (2.1) on a correlation function of
N open string states with vertex operators Vi (zi ) inserted at the boundary of the disk. We choose
a contour C that surrounds all vertex positions zi and consider the integral (see Fig. 2):
dw I
V (w)V1 (z1 )V2 (z2 ) VN (zN ) .
W :=
(2.20)
2i I
C
The SUSY current VI (w) has conformal dimension one. Therefore at infinity the correlator must
behave like w 2 . Since the contour C may be deformed to infinity, we conclude W = 0. On
the other hand, analyticity allows to deform the contour to a sum over contours Cl encircling
each of the points zl and we may rewrite (2.20) as:

N

dw I
W=
V (w)Vl (zl ) Vl+1 (zl+1 ) VN (zN ) . (2.21)
V1 (z1 ) Vl1 (zl1 )
2i I
l=1
Cl
90
Fig. 2. Closed contour C and N contours Cl around the points zl .
In the above expression, each SUSY variation (2.9) gives rise to a SUSY-transformed vertex
operator (depending on I ) discussed in the previous subsection. Hence the above equation gives
rise to non-trivial relations between different correlators:
N

V1 (z1 ) Vl1 (zl1 ) I QI , Vl (zl ) Vl+1 (zl+1 ) VN (zN ) = 0.
(2.22)
l=1
Since Eq. (2.22) holds for arbitrary spinors , the coefficients of 1 and 2 on its l.h.s., as well
),
must vanish. In this way, SUSY
as the coefficients of 1 and 2 in the relations involving Q(
Ward identities can be derived, relating various scattering amplitudes [35]. One of such Ward
identities will be applied below to the computations of N -gluon amplitudes. We also use this
equation in Section 4.2 to relate various four-point string amplitudes.
2.4. Helicity basis and comparison with 0 limit
It is well known from field-theoretical computations that scattering amplitudes simplify in
the so-called helicity basis, that is by considering processes with definite left- and right-handed
polarizations states (helicity or +, respectively) of external particles. Furthermore, such a basis
is very natural in the context of supersymmetry. In order to construct the string vertex operators
describing helicity eigenstates, it is convenient to express the corresponding wave functions in
terms of two-component, chiral spinors. We introduce the following shorthand notation for the
wave functions of chiral fermions:
u (k) = k ,
u (k) = k .
(2.23)
Then the on-shell momentum vectors k, with k 2 = 0, factorize as

k k = k k ,
k
k
= k k .
(2.24)
Note that the inverse relation is

1
1
= k
k = k
.
2
2
(2.25)
The bracket products of two spinors associated to momenta p and q are defined as
p q = p q ,
[p q] = p q .
(2.26)
91
We also use the same symbols for products involving arbitrary spinors: p p, etc. In this
notation, the polarization vectors of left- and right-handed bosons are [1,2]:
(k, r) =
k r
2
,
[k r]
+ (k, r) =
r k
2
,
r k
(2.27)
where r is an arbitrary reference vector.

We introduce the following notation for the vertices describing and + helicity states of the
N = 4 gauge vector multiplet2 :

J (z, k) = VJ z, u , k ,
g (z, k) = VA z, , k ,
i (z, k) = V i (z, k),

i+ (z, k) = V i+ (z, k),
J + (z, k) = V J (z, u , k),
g + (z, k) = VA z, + , k .
(2.28)
Note that for scalars, the above notation is just a matter of convention. Our goal is to simplify the
SUSY transformations (2.19) of these vertices.
First, we consider the transformations of scalars (2.16). They can be simplified by using

k S = ku S
(2.29)
and

S = [ k]u S .
k
Next, we consider the gaugino transformations (2.17). We will prove that
u (k) = 2 [ k]
(2.30)
(2.31)
with the provision that, as in our case, belongs to the gluon vertex operator (2.14) inserted
in an on-shell scattering amplitude. Then the contractions involving amount to replacements
p q , where p and q are some light-like vectors. Then the l.h.s. of Eq. (2.31) becomes
k = [ q]p k.
(2.32)
On the other hand, after substituting the polarization vector (2.27) into the r.h.s. of Eq. (2.31), we
obtain
p k[r q]
[r ]
(2.33)
2 [ k] = [k ]
= [ q]p k +
p k[k q],
[k r]
[k r]
where we used Schoutens identity. After reinstating in the second term on the r.h.s. of
Eq. (2.33), p k[k q] k , one finds that it is equivalent to a contribution of the longitudinal
part of a polarization vector, therefore it does not contribute to the amplitude. Then Eq. (2.31)
follows from Eqs. (2.32) and (2.33). Similarly, one finds
u (k) = 2 k + .
(2.34)
Finally, we consider the gluon transformations (2.18). The term

1
k S = k k
S k k S
2
2 Here, we skip the gauge indices.
(2.35)
92
vanishes for the right-handed polarization vector + , cf. Eq. (2.27), while for the left-handed
polarization

k S = 2 ku (k)S .
(2.36)
Similarly, one finds

+ k S = 2 [ k]u (k)S .
(2.37)
Note that all transformations involve one of the factors

(, k) = k,
+ (, k) = [ k].
After collecting all above formulae, Eq. (2.19) can be rewritten as

I

Q (I ), i (z, k) = (I , k)J + (z, u,
k),
I

i
+
J
Q ( I ), (z, k) = (I , k) (z, u, k),

I
Q (I ), J (z, k) = (I , k) j (z, k),
I

Q ( I ), J + (z, k) = + (I , k) j (z, k),
I

Q (I ), J + (z, k) = (I , k) I J g + (z, , k),
I

Q ( I ), J (z, k) = + (I , k) I J g (z, , k),
I

Q (I ), g (z, k) = (I , k)I (z, k),

I
Q ( I ), g + (z, k) = + (I , k)I + (z, k),
(2.38)
(2.39)
where the l.h.s. and r.h.s. indices are paired according to Eq. (2.11) and Table 1, keeping in mind
that and are associated to and + gaugino helicities, respectively, while = and
= + to and + scalars.
As far as practical applications of SUSY relations to the computations of gluon amplitudes
are concerned, there is no need to employ the full N = 4 algebra. It is often sufficient to consider
the gauge multiplet of N = 2 SUSY generated by QK = (Q1 , iQ2 ) [3]. Such N = 2 SUSY
transformations, parameterized by a Dirac spinor (K , K ), are generated by:
K
K
QK (K , K ) = K
Q + K
.
Q
(2.40)
The N = 2 gauge multiplet consists of the gluon A, two gauginos L = (1 , i2 ) and the scalar
1 . The SUSY transformations of the respective vertex operators can be extracted from
Eq. (2.39). They are:
K

Q (K , K ), (z, k) = i KL (K , k)L (z, k),
K

Q (K , K ), L (z, k) = KL (K , k)g (z, k) i KL (K , k) (z, k),
K

Q (K , K ), g (z, k) = (K , k)K (z, k).
(2.41)
The field-theoretical relations that should be compared with our SUSY transformations of
string vertices are the transformations of the creation and annihilation operators, summarized
in Ref. [3]. Indeed, Eq. (2.41) agree with Eq. (4) of [3], to all orders in . Since the contour
manipulations described at the beginning of this section are equivalent to applying SUSY Ward
identities, all field-theoretical ( = 0) SUSY relations between various amplitudes remain valid
in full-fledged superstring theory.
93
The main focus of this work are the N -gluon MHV disk amplitudes. We will use the wellknown relation [13]
A(1 , 2 , 3+ , 4+ , . . . , N + )

1 22
A (k1 ), (k2 ), + (k3 ), + (k4 ), g + (k5 ), . . . , g + (kN ) ,
=
3 42
(2.42)
a consequence of Eqs. (2.41), now guaranteed to hold to all orders in . It allows replacing four
gluons by four scalars, which is expected to yield a considerable simplification because the scalar
vertices are much easier to handle than the gluonic ones. We would like to stress that at the disk
level, the N -gluon MHV amplitude on the l.h.s. of Eq. (2.42) is universal to all compactifications
while the amplitude on the r.h.s. involves the scalar member of N = 2 gauge multiplet, and can
be evaluated in an arbitrary N 2 compactification. For our purposes, it is most convenient to
use the toroidal, N = 4 compactifications, with the gauge scalar associated to one of the three
complex planes.
We will be first considering the so-called partial amplitudes, associated to one particular
gauge group (ChanPaton) factor Tr(T a1 T aN ), as in Eq. (1.1), and later explain how to obtain
full amplitudes.
3. Disk scattering of scalars and gluons
In this section, we compute the string amplitudes for the scattering of four scalars and N 4
gluons at the disk level. All vertex operators are inserted at the boundary of the disk. In the
notation of Refs. [13,14], the partial amplitude associated to Tr(T a1 T aN ) ChanPaton factor
takes the form:

A a1 , , a2 , , a3 ,+ , a4 ,+ , Aa5 , . . . , AaN
N

1
dzk
= VCKG

z1 <<zN
k=1
V a1 , (z1 , k1 )V a2 , (z2 , k2 )V a3 ,+ (z3 , k3 )V a4 ,+ (z4 , k4 )
N4

VAal (zl , l , kl ) .
l=1
(3.1)
The vertex operators for the scalars, in (1)- and zero-ghost pictures, are (cf. also (2.12))
V a, (z, k) = g T a e eik X (z) ,

g
(0)
V a, (z, k) =
T a iZ + 2 (k) eik X (z) ,
(2 )1/2
(1)
V a,+ (z, k) = g T a e eik X (z) ,

g
(0)
V a,+ (z, k) =
T a i Z + 2 (k) eik X (z) ,
(2 )1/2
(1)
(3.2)
respectively. The vertex operator for the gauge boson has been already written before in the
(1)-ghost picture in Eq. (2.14), and in the zero-ghost picture it takes the form:

gA
T a iX + 2 (k) eik X (z) .

VA(0)
(3.3)
a (z, , k) =
(2 )1/2
94
In order to cancel the background ghost charge on the disk, two vertices in the correlator (3.1)
will be inserted in the (1)-ghost picture, with the remaining ones in the zero-ghost picture.
Furthermore, in Eq. (3.1), the factor VCKG accounts for the volume of the conformal Killing
group of the disk after choosing the conformal gauge. It will be canceled by fixing three vertex
positions and introducing the respective c-ghost correlator. The correlator of vertex operators
in (3.1) is evaluated by performing all possible Wick contractions. It decomposes into products
of the two-point functions on the boundary of the disk:
2
2
X (z1 )X (z2 ) =
,
X (z1 )X (z2 ) = 2 ,
z12
z12

ik X (z ) ik X (z )
1

1 e
2 = |z |2 k1 k2 ,
e(z1 ) e(z2 ) =
,
e
12
z12

1
(z1 ) (z2 ) =
,
(z1 ) (z2 ) =
,
(z1 ) (z2 ) = 0,
z12
z12

2 ) = 2 ,
Z(z1 )Z(z2 ) = 0.
Z(z1 ) Z(z
2
z12
(3.4)
Because of the PSL(2, R) invariance on the disk, we can fix three positions of the vertex operators. A convenient choice respecting the integration region z1 < < zN is
z1 = z = ,
z2 = 0,
z3 = 1,
(3.5)
2 .
c(z1 )c(z2 )c(z3 ) = z
which implies the ghost factor

The remaining N 3 vertex positions
z4 , . . . , zN take arbitrary values inside the integration domain 1 < z4 < < zN < .
In order to correctly normalize the amplitudes, some additional factors have to be taken into
account. They stem from determinants and Jacobians of certain path integrals. On the disk, the
net result of those contributions is an additional factor of
CD2 =
1
2 2
2gYM
(3.6)
which must be included in all disk correlators [21].

3.1. Five-point amplitudes
For N = 5, Eq. (3.1) yields the following correlator:
(0)
(1)
(0)
(1)
V a1 , (z1 , k1 )V(0)
a2 , (z2 , k2 )V a3 ,+ (z3 , k3 )V a4 ,+ (z4 , k4 )VAa5 (z5 , 5 , k5 ) .
Proceeding as outlined above, we obtain

A a1 , , a2 , , a3 ,+ , a4 ,+ , Aa5
1

3
= 2(2 )gYM
Tr T a1 T a2 T a3 T a4 T a5 VCKG
5

s12 z34
1 (5 k4 )
sij
dzk
|zij |
z35
z45 z24 z13 z14 z23
i<j
z1 <<z5 k=1

(5 k1 ) (1 s24 )
s24
s14
(5 k2 ) (1 s14 )
+
+
+
+
,
z15 z24 z24 z13
z14 z23
z14 z25 z14 z23
z13 z24
(3.7)
(3.8)
95
with sij 2 ki kj [13,14]. We set z1 , z2 , z3 as in (3.5) (taking into account the ghost factor) and
use the parameterization
z4 = x 1 ,
z5 = (xy)1
(0 < x, y < 1),
(3.9)
3 2
4 ,z5 )
det( (z
(x,y) ) = x y .
with the corresponding Jacobian

Then Eq. (3.8) becomes
a , a , a ,+ a ,+ a
A 1 , 2 , 3 , 4 ,A 5

3
Tr T a1 T a2 T a3 T a4 T a5 (5 k1 )K1 + (5 k2 )K2 + (5 k4 )K3 ,
= 2(2 )gYM
(3.10)
with the three functions

1 1
s24
I(x, y)
K1 = dx dy 1 s24 +
,
x y(1 xy)
0
1
1
K2 =
dx

dy
1
K3 = s12

1 s14
I(x, y)
+ s14
,
x
(1 xy)
1
dx
dy
0
(1 x) I(x, y)
,
x(1 y) (1 xy)
(3.11)
where
I(x, y) = x s2 y s5 (1 x)s3 (1 y)s4 (1 xy)s1 s3 s4 ,
(k
(3.12)
and si
subject to the cyclic identification i + 5 i. The functions (3.11) intei + ki+1
grate to Gaussian hypergeometric functions 3 F2 [14]. Actually, K3 may be expressed in terms of
K1 and K2 :
s5
s1 s3 + s5
K3 = K1 +
(3.13)
K2 .
s4
s4
Thus as expected, -dependence of the amplitude describing the scattering of four-scalars and
one gluon is determined by two independent functions, exactly as many as in the five-gluon case
[14,22,23].
The low-energy behavior of the amplitude (3.10) is determined, up to the order O( 2 ), by
the following expansions:

1
s3
s3 s4 s4 s5 s32
(2) s1 + s3
+
+ ,
K1 =
s2 s2 s5
s2
s2
s5

1
s4 s5
+ ,
K2 = (2) s1 + s3 +
s2
s2

s1
s1 s5 s12
(2)
+
+ .
K3 =
(3.14)
s2 s4
s2
s4
The functions {K1 , K2 } can be expressed in terms of another two-element basis, previously
used in [13,14]. There, we defined two functions:
)2 ,
1
f1 =
1
dx
dy x
0
1 1
1
I(x, y),
f2 =
1
dx
dy (1 xy)1 I(x, y).
(3.15)
96
Expressed in terms of {f1 , f2 }, the functions K1 and K2 read:

K1 = (s5 s3 )f1 s1 f2 ,
K2 = s5 f1 s1 f2 .
(3.16)
The result (3.10) can be further simplified by choosing k4 as the reference vector r for the polarization vector 5 , so that 5 k4 = 0, see Eq. (2.27). Then, for the positive polarization of the
gluon,
1 4 1[1 5]
1 4 2[2 5]
,
5+ k2 =
.
5+ k1 =
(3.17)
4
5
2
2 4 5
As
a result of combining Eq. (3.10) with the SUSY relation (2.42), after factorizing out
( 2gYM )3 Tr(T a1 T a5 ), we obtain the five-gluon MHV amplitude:

1 22
4 1[1 5]K1 + 4 2[2 5]K2 .
2
3 4 4 5
The above result should be compared with the factorized form derived in [14]

(5)
A(1 , 2 , 3+ , 4+ , 5+ ) = V (5) (sj ) 2iP (5) (sj )(1, 2, 3, 4) MYM ,
A(1 , 2 , 3+ , 4+ , 5+ ) =
(3.18)
(3.19)
where:
1
V (5) (si ) = s2 s5 f1 + (s2 s3 + s4 s5 s1 s2 s3 s4 s1 s5 )f2 ,
2
P (5) (si ) = f2 .
(3.20)
Indeed, it is a matter of simple spinor manipulations to show that Eqs. (3.18) and (3.19) agree
upon using the relations (3.16) between the two sets of basis functions, {K1 , K2 } and {f1 , f2 }.
Unlike the factorized amplitude (3.19), the new form (3.18) is expressed directly in terms of
primitive integrals (3.11), without resorting to the definitions of formfactor functions V and
P , see Eq. (3.20). Although for five gluons it is not a dramatic simplification, we will see that for
six gluons and more, the non-factorized form is more suitable.
3.2. Six-point amplitudes
Now we turn to N = 6. For that case in (3.1), we compute the following correlator:
(0)
(0)
(0)
(0)
(1)
(1)
V a1 , (z1 , k1 )V a2 , (z2 , k2 )V a3 ,+ (z3 , k3 )V a4 ,+ (z4 , k4 )VAa5 (z5 , 5 , k5 )VAa6 (z6 , 6 , k6 ) .
(3.21)
With the choice (3.5) and the parameterization
z4 = x 1 ,
z5 = (x y)1 ,
z6 = (x y z)1 ,
(z4 ,z5 ,z6 )
det( (x,y,z) ) = x 4 y 3 z2 . The amplitude
(3.22)
the Jacobian
(3.21) becomes

a , a , a ,+ a ,+ a
A 1 , 2 , 3 , 4 , A 5 , Aa6

4
Tr T a1 T a2 T a3 T a4 T a5 T a6
= 2(2 )2 gYM

(5 k3 )(6 k2 )K1 + (5 k2 )(6 k3 )K2 + (5 k1 )(6 k2 )K3 + (5 k1 )(6 k3 )K4
+ (5 k2 )(6 k1 )K5 + (5 k3 )(6 k1 )K6 + (5 k3 )(6 k4 )K7 + (5 k4 )(6 k3 )K8
+ (5 k2 )(6 k4 )K9 + (5 k4 )(6 k2 )K10 + (5 k1 )(6 k4 )K11 + (5 k4 )(6 k1 )K12

+ (5 6 )K13 ,
(3.23)
where the integrals
1
K1 =
1
dx
K2 =
dx
dz
1
1
dx
1
dy
dz
1
1
1
K4 =
dx
dy
1
K5 = s34
I(x, y, z)
,
xy(1 z)

s24
I(x, y, z)
dz 1 s24 +
,
x y(1 z)(1 xyz)
dy
0
1
1
dz
0
dy
0

1 s14
I(x, y, z)
+ s14
,
x
(1 z)(1 xyz)
1
dx
0
1
dx
1
1
dy
K6 =

1 s14
I(x, y, z)
dz
+ s14
,
x
(1 z)(1 xy)
1
K3 = s34
1
dy
1
97
I(x, y, z)
,
xyz(1 z)

s24
I(x, y, z)
dz 1 s24 +
,
x yz(1 z)(1 xy)
(3.24)
and the remaining seven integrals are displayed in Eq. (B.1) of Appendix B. All integrands contain the common factor
I(x, y, z) = x s2 y t2 zs6 (1 x)s3 (1 y)s4 (1 z)s5
(1 xy)t3 s3 s4 (1 yz)t1 s4 s5 (1 xyz)s1 +s4 t1 t3 ,
(k
(k
(3.25)
where si =
and tj =
subject to the cyclic identification
i + ki+1
j + kj +1 + kj +2
i + 6 i.
The integrals (3.24) and (B.1) integrate to multiple Gaussian hypergeometric functions, more
precisely they represent triple hypergeometric functions [23]. Although the result (3.23) uses
thirteen functions, only six of them are linearly independent, exactly as many as in the six-gluon
case [23]. For our purposes, it is convenient to choose {K1 , . . . , K6 } as the basis. The remaining
seven functions are then expressed in terms of these functions in Eq. (B.2)
The low-energy behavior of the amplitude (3.10) is determined, up to the order O( 2 ), by
the following expansions:

1
s4 + s5 + s6 t1 t2 s1 + s3
t 1 t2
+ (2) 1
+ ,
K1 =
s2 s5
s2
s5
s2 s5

1
s4 + s5 + s6 t1 t2 s1 + s3
t 1 t2
+ (2)
+
+
+ ,
K2 =
s2 s5
s2
s5
s2 s5

s32
s3
s3
s3 t1
s3 s6
+ (2)
+ ,
K3 =
s2 s5 t2
s2 s2 s5 s5 t2
s2 t2
1
s3
K4 =
+
s2 s5 s2 s5 t2

s32
s3 + s6 t2 s1 + s3
s3 t1
s3 s6
t 1 t2
+ (2)
+
+
+ ,
s2
s5
s2 s5 s5 t2
s2 t2
s2 s5
)2
)2 ,
98
s3
s3
s2 s5 t2 s2 s6 t2

s32
s32
s3 s3 s4
s3 t1
s3 s5
s3 s6
+ (2) +
+
+
+
+
+
+ ,
s2 s2 s6 s2 s5 s5 t2
s2 t2
s6 t2
s2 t2
1
1
s3
s3
+
K6 =
s2 s5 s2 s6 s2 s5 t2 s2 s6 t2

s3 + s5 + s6 + t2 s1 + s3 s3 + t3
+ (2) 1
s2
s5
s6

2
2
s3
s3
s3 s4
s3 t1
s3 s5
s3 s6
s4 t2
t 1 t2
+
+
+
+
+
+
+ .
s2 s6 s2 s5 s5 t2
s2 t2
s6 t2
s2 t2
s2 s6 s2 s5
K5 =
(3.26)
At this point, we specify the result (3.23) to the case of two positive-helicity gluons, as in the
amplitude entering the SUSY relation (2.42). We choose k4 as the reference vector r for both
polarization vectors, so that i k4 = 0, see Eq. (2.27). Then:
1 4 j [j i]
i+ kj =
.
2 4 i
i+ j+ = 0,
(3.27)
We also define:
(a, b) = 2(5+ ka )(6+ kb ) =
4 a4 b[a 5][b 6]

.
4 54 6
(3.28)
As a result of combining Eqs. (3.21) and (2.42), after factorizing out ( 2gYM )4 Tr(T a1 T a6 ),
we obtain the six-gluon MHV amplitude:
A(1 , 2 , 3+ , 4+ , 5+ , 6+ )
1 22
(3, 2)K1 + (2, 3)K2 + (1, 2)K3
= 2
3 42

+ (1, 3)K4 + (2, 1)K5 + (3, 1)K6 .
(3.29)
The above result should be compared with the factorized form derived in [14]:

k=5

(6)
+ + + +
(6)
k Pk (si , ti ) M(6)
A(1 , 2 , 3 , 4 , 5 , 6 ) = V (si , ti ) 2i
YM
(3.30)
k=1
where
1 = (2, 3, 4, 5),
2 = (1, 3, 4, 5),
4 = (1, 2, 3, 5),
The functions
of:
V (6)
1
F1 =
and
1
dx
3 = (1, 2, 4, 5),
5 = (1, 2, 3, 4).
(6)
Pk ,
1
dy
dz
(3.31)
k = 1, . . . , 5, [13,14] are certain (complicated) linear combinations

I(x, y, z)
,
xyz
1
F2 =
1
dx
1
dy
1
1
1
1
1
1
F3 =
dx
0
dy
0
I(x, y, z)
dz
,
1 xyz
F4 =
dx
0
dy
0
dz
I(x, y, z)
,
z(1 xy)
dz
yI(x, y, z)
,
(1 xy)(1 yz)
99
1
F5 =
1
dx
1
dy
1
1
1
F6 =
dx
0
dy
0
dz
I(x, y, z)
,
(1 xy)(1 xyz)
dz
I(x, y, z)
,
(1 yz)(1 xyz)
(3.32)
weighted by coefficients which are polynomial in the kinematic invariants.

In order to compare Eqs. (3.29) and (3.30), one needs relations between K- and F -functions.
These can be obtained by integrating by parts combined with other manipulations described
in [14]. One finds, for instance,
s2 s5 K3 = s2 s3 s6 F1 + s3 s6 (s4 + s5 t1 )F2 + s3 (s4 + s5 t1 )(s1 + s3 s5 t3 )F6

+ s3 s1 s2 (s4 + s5 t1 )(s3 s5 + s6 + t1 t3 ) + s2 (s5 + t3 ) F3
+ s3 (s4 + s5 t1 )(s3 + s5 t1 + t3 )F4
+ s3 (s4 + s5 t1 )(s1 + s3 s5 + t1 )F5 .
(3.33)
The remaining five elements of the K-basis are expressed in terms of {F1 , F2 , F3 , F4 , F5 , F6 } in
Appendix B. After some tedious algebra one finds that the new expression (3.29) does indeed
agree with the results of [13,14]. The use of K-basis imposed by SUSY relations leads to a
much simpler expression, as a linear combination of primitive integrals (3.24) weighted by
single-term twistor-like coefficients.
3.3. Seven-point amplitudes
Finally, we consider the case of N = 7. We evaluate the following correlator:
(0)
(0)
(0)
(0)
(1)
(1)
(0)
V a1 , (z1 )V a2 , (z2 )V a3 ,+ (z3 )V a4 ,+ (z4 )VAa5 (z5 )VAa6 (z6 )VAa7 (z7 ) .
(3.34)
Here again, we make the choice (3.5) and use the parameterization
z4 = x 1 ,
z5 = (xy)1 ,
z6 = (xyz)1 ,
z7 = (xyzw)1 ,
5 4 3 2
4 ,z5 ,z6 ,z7 )
with the corresponding Jacobian det( (z
(x,y,z,w) ) = x y z w . Even
tum conservation to eliminate the scalar products (5 k6 ), (6 k7 ) and (7 k5 ),
(3.35)
after using momen-
5 k6 = 5 k1 5 k2 5 k3 5 k4 5 k7 ,
6 k7 = 6 k1 6 k2 6 k3 6 k4 6 k5 ,
7 k5 = 7 k1 7 k2 7 k3 7 k4 7 k6 ,
(3.36)
the amplitude (3.34) still involves many3 kinematic factors, each of them multiplied by a certain Euler integral or multiple hypergeometric function. However, all integrals can be expressed
in terms of the 24-element basis {K1 , . . . , K24 } written in Appendix C. As we will see below,
3 After applying (3.36) the amplitude (3.34) involves 111 different kinematical factors. Each of them is multiplied by a
certain integral or hypergeometric function. However, some of those kinematical factors appear with the same function.
In fact, only 73 different integrals or functions appear in (3.34). These 73 functions may be expressed in terms of a
basis, whose dimension is 24. As in the five- or six-point case this basis is completely specified by the MHV part of the
amplitude (3.34).
100
similarly to the six-gluon case, this basis appears naturally in the MHV part of the amplitude
(3.34).
At this point, we specify the correlator (3.34) to the case of three positivehelicity gluons, as
in the amplitude entering the SUSY relation (2.42). We choose k4 as the reference vector r for
all polarization vectors, so that i k4 = 0, see Eq. (2.27). Then:
1 4 j [j i]
i+ kj =
.
2 4 i
i+ j+ = 0,
(3.37)
We also define:
4 a4 b4 c[a 5][b 6][c 7]

(a, b, c) = 2 2(5 ka )(6 kb )(7 kc ) =
.
4 54 64 7
(3.38)
For i k4 = 0, the correlator (3.34) gives rise to the partial amplitude

A a1 , , a2 , , a3 ,+ , a4 ,+ , Aa5 , Aa6 , Aa7
24

5
= 2(2 )3 gYM
Tr T a1 T a2 T a3 T a4 T a5 T a6 T a7
KI KI ,
(3.39)
I =1
with 24 kinematics KI andintegrals KI to be specified below. After using the SUSY relation
(2.42) and factorizing out ( 2gYM )5 Tr(T a1 T a7 ), the correlator (3.34) yields the seven-gluon
MHV amplitude
A(1 , 2 , 3+ , 4+ , 5+ , 6+ , 7+ ) = 3
24
1 22
KI KI ,
3 42
(3.40)
I =1
where the kinematic factors KI are as follows:

K1 = (1, 2, 1) + (7, 2, 1),
K2 = (1, 3, 1) + (7, 3, 1),
K3 = (2, 2, 1) + (2, 5, 1),
K4 = (3, 3, 1) + (3, 5, 1),
K5 = (2, 1, 2) + (7, 1, 2),
K6 = (2, 3, 2) + (7, 3, 2),
K7 = (1, 1, 2) + (1, 5, 2),
K8 = (3, 3, 2) + (3, 5, 2),
K9 = (3, 1, 3) + (7, 1, 3),
K10 = (3, 2, 3) + (7, 2, 3),
K11 = (1, 1, 3) + (1, 5, 3),
K12 = (2, 2, 3) + (2, 5, 3),
K13 = (2, 1, 1) + (2, 1, 6),
K14 = (3, 1, 1) + (3, 1, 6),
K15 = (1, 2, 2) + (1, 2, 6),
K16 = (3, 2, 2) + (3, 2, 6),
K17 = (1, 3, 3) + (1, 3, 6),
K18 = (2, 3, 3) + (2, 3, 6),
K19 = (3, 2, 1),
K20 = (2, 3, 1),
K21 = (3, 1, 2),
K22 = (1, 3, 2),
K23 = (2, 1, 3),
K24 = (1, 2, 3).
(3.41)
The 24 basis functions Ki are given as Euler integrals, e.g.:

1
K1 = s34
1
dx
1
dy
1
dz
dw
0
I(x, y, z, w)
,
xyw(1 z)(1 wz)
(3.42)
101
with
I(x, y, z, w) = x s2 y t2 zt6 w s7 (1 x)s3 (1 y)s4 (1 z)s5 (1 w)s6
(1 xy)t3 s3 s4 (1 yz)t4 s4 s5 (1 xyz)s4 t3 t4 +t7
(1 zw)t5 s5 s6 (1 yzw)s5 +t1 t4 t5 (1 xyzw)s1 t1 +t4 t7 ,
(3.43)
si = (ki + ki+1 )2 and tj = (kj + kj +1 + kj +2 )2 subject to the cyclic identification i + 7 i.

All 24 integrals are displayed in Appendix C. The 18 functions K1 , . . . , K18 seem like straightforward generalizations of the six-point functions (3.24). Indeed, they are related to them in
certain soft-boson limits: for example, in the k7 0 limit, K1 s17 K3 of N = 6. On the other
hand, the remaining functions K19 , . . . , K24 seem to have a different character.
3.4. From partial to full amplitudes
So far, we focussed only on one partial amplitude, associated to the color factor Tr(T a1 T aN).
In Ref. [14], we gave a prescription for constructing any partial amplitude A (1 , 2 , 3+ , 4+ ,
. . . , N + ), associated the color factor Tr(T a (1) T a (N) ) with the indices ai permuted by an
arbitrary permutation . One simply factorizes out 124 and applies to all momenta inside the
remainder:

A(1 , 2 , 3+ , 4+ , . . . , N + )
+ +
+
4
.
A (1 , 2 , 3 , 4 , . . . , N ) = 1 2

1 24
ki k (i)
(3.44)
(N )
MYM
(1.2)
For an amplitude written as in [14], as the product of the zero-slope MHV amplitude
times a string formfactor, this is a completely trivial operation, however with the non-factorized
form of amplitudes presented in this section, it requires more care.
4. Four-point string amplitudes and supersymmetry relations
In the previous section we have seen, that string amplitudes may be written much simpler by
making use of the SUSY Ward identities (2.22). In this section we demonstrate this for the fourpoint string amplitude involving scalars, gauginos and vectors from the N = 4 vector multiplet.
After computing the latter we apply the results of Section 2.3 to generate relations among them.
4.1. Four-point disk scattering of scalars, gauginos and gluons
Here we compute in D = 4 the tree-level four-point string amplitude involving scalars, gauginos and vectors from the N = 4 vector multiplet. The world-sheet of the string S-matrix is
described by a disk with all external states a created through vertex operators V a at the boundary of the disk. The color-ordered part of the amplitude of interest takes the form:

A a1 , a2 , a3 , a4
4

1
= VCKG
(4.1)
dzk V a1 (z1 )V a2 (z2 )V a3 (z3 )V a4 (z4 ) .
z1 <<z4
k=1
The vertex operators for the gauginos are given in (2.13), while the vertex operators for the scalars
and vectors are given in (3.2). The four-point correlator in the integrand of (4.1) is evaluated by
102
performing all possible Wick contractions. Some of the relevant correlators are

1/2
S (z1 )S (z2 ) = z12 ,
S (z1 )S (z2 ) = 0,

1 1/2 1/2
S (z1 )S (z2 ) (z3 ) = z13 z23 ,
2

S (z1 )S (z2 ) (z3 ) = 0,
(4.2)
and:
1
1
1
1
e 2 (z1 ) e 2 (z2 ) e 2 (z3 ) e 2 (z4 ) = (z12 z13 z14 z23 z24 z34 )1/4 .
(4.3)
Again, PSL(2, R) invariance on the disk allows to fix three vertex positions according (3.5),
2 . The remaining vertex position z :=
which implies the ghost factor c(z1 )c(z2 )c(z3 ) = z
4
1/x takes arbitrary values inside the integration domain 0 < x < 1 respecting the integration
region z1 < < z4 in (4.1).
Four gauginos
For the four-gaugino amplitude involving four gauginos of the same chirality in (4.1) we
compute the correlator
(1/2)
(1/2)
(1/2)
(1/2)
Va1 ,I (z1 , u1 , k1 )Va2 ,J (z2 , u2 , k2 )V a3 ,K (z3 , u3 , k3 )Va4 ,L (z4 , u4 , k4 ) ,
(4.4)
with the vertices (2.13). We need the following spacetime fermion correlator:

S (z1 )S (z2 )S (z3 )S (z4 )
z13 z14 z12 z14 + z12 z13
=
(z12 z13 z14 z23 z24 z34 )1/2
z14 z23 z12 z34
=
.
(z12 z13 z14 z23 z24 z34 )1/2
(4.5)
In the last step we have applied the Fierz identity = + . Furthermore, the
correlator for the internal Ramond fields is:
I
J = K = L.
(z1 ) J (z2 ) K (z3 ) L (z4 ) = (z12 z13 z14 z23 z24 z34 )1/4 , I =
(4.6)
With the choice (3.5), the correlators (4.3), (4.5) and (4.6) we arrive at the partial amplitude of
four gauginos:

A I J K L

(s) (u)
2
u(u1 u2 )(u3 u4 ) s(u1 u4 )(u2 u3 ) ,
= (2 )gYM
(4.7)
(1 + s + u)
with the Mandelstam invariants s = 2 k1 k2 , t = 2 k1 k3 and u = 2 k1 k4 . On the other hand,
for the string amplitude with two gauginos of opposite chirality
(1/2)
(1/2)
(1/2)
(1/2)
Va1 ,I (z1 , u1 , k1 )Va2 ,J (z2 , u2 , k2 )V a3 ,K (z3 , u 3 , k3 )V a4 ,L (z4 , u 4 , k4 ) ,
(4.8)
we need the correlator
S (z1 )S (z2 )S (z3 )S (z4 ) = (z12 z34 )1/2 ,
(4.9)
103
which may be derived by using the identity = 2 and the correlator:
I
(z1 ) J (z2 ) K (z3 ) L (z4 )

IK JL

I LJ K
z13 z14 z23 z24 1/4

+
=
.
z12 z34
z13 z24
z14 z23
(4.10)
With the choice (3.5), the correlators (4.3), (4.9) and (4.10) the partial amplitude becomes:

A I J K L

(s) (u) I K J L
2
= (2 )gYM
(4.11)
(u1 u2 )(u 3 u 4 )
u + t I L J K .
(1 + s + u)
Two gauginos and two scalars
Now we compute the amplitude of two gauginos and two scalars by calculating in (4.1) the
correlator
(1/2)
(1/2)
(1)
(0)
Va1 ,I (z1 , u1 , k1 )V a2 ,j (z2 , k2 )V a3 ,K (z3 , u 3 , k3 )V a4 ,l+ (z4 , k4 ) .
(4.12)
Here the index j includes the two cases j + and j . After deriving the set of correlators

I
1
z14 z32 1/2
I
j
j+
,
(z1 ) (z3 ) (z2 ) (z4 ) =
3/4
z24 z13 z12 z34

(a): (j, I ) (1, 3), (1, 4), (2, 2), (2, 4), (3, 2), (3, 3) ,

I
1
z12 z34 1/2
(z1 ) I (z3 ) j (z2 ) j + (z4 ) =
,
3/4
z24 z13 z14 z32

(b): (j, I ) (1, 1), (1, 2), (2, 1), (2, 3), (3, 1), (3, 4) ,
I
1/4
(z1 ) K (z3 ) j (z2 ) l+ (z4 ) = z13 (z12 z14 z32 z34 )1/2 ,

(c): (j , l, I, K) (1, 2, 3, 2), (1, 3, 4, 2), (2, 3, 4, 3), (2, 1, 2, 3),

(3, 1, 2, 4), (3, 2, 3, 4) ,

(j +, l, I, K) (1, 2, 1, 4), (1, 3, 1, 3), (2, 3, 1, 2), (2, 1, 1, 4),

(3, 1, 1, 3), (3, 2, 1, 2) ,
(4.13)
with (4.2) the resulting partial amplitude becomes

(s) (u)

2
k4 u1 u 3
A I j K l+ = (2 )gYM
(1 + s + u)

u,
case (a),
s, case (b),
t,
case (c),
(4.14)
for the three cases in (4.13).

Two gauginos, one vector and one scalar
Next, we compute the four-point amplitude of two gauginos, one vector and one scalar. In
(4.1) we compute the correlator:
(1/2)
(1/2)
(1)
(0)
Va1 ,I (z1 , u1 , k1 )Va2 ,J (z2 , u2 , k2 )VAa3 (z3 , 3 , k3 )V a4 ,l+ (z4 , k4 ) .
(4.15)
104
For the spacetime fermions we need the correlator:

S (z1 )S (z2 ) (z3 ) (z4 )
1
(z12 z13 z14 z23 z24 )1/2
= z34

1
1

z13 z14 + z12 z14 + z12 z13
2
2
1/2
z12
1 (z13 z24 + z14 z23 ) 1
=
.
2 z34 (z12 z13 z14 z23 z24 )1/2
2 (z13 z14 z23 z24 )1/2
(4.16)
=
The last step follows from the relations
and + = 2 .

The necessary correlator for the internal fields is:
I
1/4
(z1 ) J (z2 ) l+ (z4 ) = z12 (z14 z24 )1/2 (I, J, l+) I,
(4.17)
with I = {(1, 2, 1), (1, 3, 2), (1, 4, 3), (2, 1, 1), (3, 1, 2), (4, 1, 3)}. With (4.16) and (4.17) the
partial amplitude of (4.1) becomes

(s) (u)
2
A I J A l+ = 2 gYM
(1 + s + u)

(u t)(u1 u2 )(3 k4 ) s u1 u2 3 k4 ,
(4.18)
with (I, J, l+) I.

Four scalars
Finally we compute the string S-matrix of four scalars. With the vertex operators of (2.12) in
(4.1) we compute the correlator:
(0)
(0)
(1)
(1)
V a1 ,i (z1 , k1 )V a2 ,j (z2 , k2 )V a3 ,k+ (z3 , k3 )V a4 ,l+ (z4 , k4 ) .
(4.19)
After some algebra and using (4.6) we find the following expression for the partial amplitude:

(s) (u) ik j l
2
A i j k+ l+ = 2gYM
u + t il j k s.
(1 + s + u)
(4.20)
4.2. Supersymmetry relations of string amplitudes

By applying the results of Section 2.3 in this part we derive relations between the string
amplitudes we have computed in the previous section. Inspection of those amplitudes shows that
they all have the same prefactor encoding the -dependence of the amplitude. Indeed here we
show that they all related through supersymmetry transformations. With (2.16) we may derive
a SUSY relation between a four gaugino amplitude and two amplitudes involving two gauginos
and two scalars.
In analogy to (2.20) we start with the contour integral

dw M
(1/2)
(1/2)
V (w)Va1 ,I (z1 , u1 , k1 )Va2 ,J (z2 , u2 , k2 )
2i M
C
(1/2)
(0)
V a3 ,K (z3 , u 3 , k3 )V a4 ,l+ (z4 , k4 ) ,
(4.21)
105
where C is a closed contour in the complex plane encircling all four vertex positions z1 , . . . , z4 .
With the arguments of Section 2.3 the integral (4.21) vanishes. On the other hand, by analyticity
in (4.21) we may deform the contour to the other four vertex operators with the SUSY operator
acting on one vertex operator, respectively (cf. Eq. (2.19)):

M
(1/2)
(1)
Q (M ), Va1 ,I (z1 , u1 , k1 ) = M u1 V a1 ,i (z1 , u1 , k1 ),
M

(1/2)
(1)
Q (M ), Va2 ,J (z2 , u2 , k2 ) = M u2 V a2 ,j (z2 , k2 ),
M

1
(1/2)
(1)

Q (M ), V a3 ,K (z3 , u 3 , k3 ) = MK VAa3 (z3 , 3 ),
3 = M
u 3 ,
2
M

(1/2)
(0)

Q (M ), V a4 ,l+ (z4 , k4 ) = V a4 ,L (z4 , u 4 , k4 ),
u 4 = k4 M
.
(4.22)
Inserting these results into (2.22) gives a sum of the following four correlators
(1)
(1/2)
(1/2)
(0)
M u1 V a1 ,i (z1 , u1 , k1 )Va2 ,J (z2 , u2 , k2 )V a3 ,K (z3 , u 3 , k3 )V a4 ,l+ (z4 , k4 )

(s) (u)

2
k4 M u1 u2 u 3 u I K J L + t I L J K ,
= 2 gYM
(1 + s + u)
(4.23)
where we have used (4.14). Similarly, the second correlator becomes:

(1/2)
(1/2)
(1)
(0)
M u2 Va1 ,I (z1 , u1 , k1 )V a2 ,j (z2 , k2 )V a3 ,K (z3 , u 3 , k3 )V a4 ,l+ (z4 , k4 )
2
= 2 gYM

(s) (u)

k4 M u2 u1 u 3 u I K J L + t I L J K .
(1 + s + u)
(4.24)
With (4.18) the third correlator gives:

(1/2)
1
(1/2)
(1)
(0)
MK Va1 ,I (z1 , u1 , k1 )Va2 ,J (z2 , u2 , k2 )VAa3 (z3 , 3 )V a4 ,l+ (z4 , k4 )
2

(s) (u)

2
= gYM
(u t)u1 u2 (3 k4 ) s u1 u2 3 k4 MK ,
(1 + s + u)
3 = M
u 3 ,

(s) (u)

u 3 sM u2 u1 u 3 MK ,
k4 uu1 u2 M
(1 + s + u)
(I, J, l+) I.
2
= 2 gYM
(4.25)
Finally, with (4.11) the last correlator becomes:

(1/2)
(1/2)
(1/2)
(1/2)
Va1 ,I (z1 , u1 , k1 )Va2 ,J (z2 , u2 , k2 )V a3 ,K (z3 , u 3 , k3 )V a4 ,L (z4 , u 4 , k4 )

(s) (u)
2

= 2 gYM
.
u1 u2 [u3 u4 ] u I K J L + t I L J K , u 4 = k4 M
(1 + s + u)
(4.26)
For M = K the third correlator (4.25) vanishes. In that case summing up (4.23), (4.24) and (4.26)
gives

(s) (u) I K J L
u + t I L J K
(1 + s + u)

u 3 ,
k4 M u1 u2 u 3 + M u2 u1 u 3 u1 u2 M
2
2 gYM
106
Table 2
Index structure for the SUSY relation (4.28)
M
(I, J, K, L, i, j, l)
(I, J, K, L, i, j, l)
(I, J, K, L, i, j, l)
1
1
1
1
1
1
1
1
1
(2, 2, 2, 2, 1, 1, 1+)
(2, 3, 3, 2, 1, 2, 1+)
(2, 3, 2, 3, 1, 2, 2+)
(2, 4, 4, 2, 1, 3, 1+)
(2, 4, 2, 4, 1, 3, 3+)
(3, 2, 3, 2, 2, 1, 1+)
(3, 3, 3, 3, 2, 2, 2+)
(3, 2, 2, 3, 2, 1, 2+)
(3, 4, 4, 3, 2, 3, 2+)
1
1
1
1
1
1
2
3
4
(3, 4, 3, 4, 2, 3, 3+)
(4, 3, 3, 4, 3, 2, 3+)
(4, 4, 4, 4, 3, 3, 3+)
(4, 2, 4, 2, 3, 1, 1+)
(4, 2, 2, 4, 3, 1, 3+)
(4, 3, 4, 3, 3, 2, 2+)
(1, 1, 1, 1, 1, 1, 1+)
(1, 1, 1, 1, 2, 2, 2+)
(1, 1, 1, 1, 3, 3, 3+)
2
2
2
2
3
3
3
3
4
4
4
4
(1, 3, 3, 1, 1, 3+, 1+)

(1, 4, 4, 1, 1, 2+, 1+)
(3, 1, 3, 1, 3+, 1, 1+)
(4, 1, 4, 1, 2+, 1, 1+)
(1, 2, 2, 1, 2, 3+, 2+)
(1, 4, 4, 1, 2, 1+, 2+)
(2, 1, 2, 1, 3+, 2, 2+)
(4, 1, 4, 1, 1+, 2, 2+)
(1, 2, 2, 1, 3, 2+, 3+)
(1, 3, 3, 1, 3, 1+, 3+)
(2, 1, 2, 1, 2+, 3, 3+)
(3, 1, 3, 1, 1+, 3, 3+)
which vanishes due to the Fierz relation:

M u2 u1 u 3 = M u1 u2 u 3 + M
u 3 u1 u2 .
(4.27)
To conclude, to all orders in we have proven the following SUSY identity

M u1 A i J K l+ + M u2 A I j K l+ + A I J K L = 0|u

4 =k4 M
(4.28)
for an arbitrary spinor M and M = K. Table 2 shows the combination of indices, for which
all the three correlators (4.23), (4.24) and (4.26) are non-vanishing, but (4.24) vanishing due to
M = K.
On the other hand, in the case M = K the third correlator (4.25) contributes in (2.22). However, in that case one of the three correlators (4.23), (4.24) or (4.26) vanishes because the
corresponding SUSY transformation (4.22) gives a vanishing contribution. In either case the
three non-vanishing residua sum up to zero due to (4.27). E.g. for M = K = 2, I = 1, J = 2 and
l = 1 the SUSY transformations (4.22) yield i = 1, L = 1 but vanishing SUSY transformation
for the variation of Va2 ,J . Hence the total contribution of residua to (4.21) is:

(s) (u)

u 3
k4 sM u1 u2 u 3 tu1 u2 M
(1 + s + u)

uu1 u2 M
u 3 + sM u2 u1 u 3 ,
2
2 gYM
which vanishes according to (4.27). Eventually (2.22) gives rise to the following identity:

1

2 u1 A 1+ 2 2 1+ + A 1 2 A 1+ + A 1 2 2 1 = 0 u =k4 . (4.29)
4
2
2

u
3 =M
3
5. Concluding remarks
The main result of this paper is the extension of the well-known SUSY Ward identities relating
the scattering amplitudes of particles with different spin but belonging to the same supermultiplet,
to type I open superstring theory at the disk level. For arbitrary compactifications, the form of
107
such relations remains exactly the same as in the low-energy effective field theory describing the
0 limit. Although we studied explicitly only the case of gauge supermultiplets it is clear
that our results extend to matter supermultiplets.
Here, we focussed on one particular application of SUSY relations: to the computations of
N -gluon MHV amplitudes. In this case, four gluons can be replaced by scalars, leading to significant simplifications. By using this method, we were able to reproduce all known results for N 6
and to derive a compact expression for the seven-gluon MHV amplitude. It is very interesting
that the use of supersymmetry dictates certain choice of (N 3)! elements of the basis of boundary integrals over the vertex positions, such that MHV amplitudes can be represented as linear
combinations of these basis functions weighted by very simple twistor-like coefficients. It seems
that such a representation is more natural than the factorized form discussed previously in [13,
14]. We believe that it is also more suitable for studying the recursive structure for arbitrary N .
There are several extensions and applications of our results that deserve further studies. In
particular, it would be interesting to see if SUSY relations lead to a novel representation of
superstring amplitudes also for non-MHV configurations (that appear for N 6) [16]. Furthermore, a generalization of SUSY relations to world-sheets with string loops would be useful for
understanding how supersymmetry is realized at the level of quantum-corrected, or perhaps even
exact, scattering amplitudes.
Acknowledgements
This work is supported in part by the European Commission under Project MRTN-CT-2004005104. The research of T.R.T. is supported in part by the US National Science Foundation
Grants PHY-0242834 and PHY-0600304. T.R.T. is grateful to Dieter Lst, to Arnold Sommerfeld
Center for Theoretical Physics at Ludwig-Maximilians Universitt, and to Max-Planck-Institut
fr Physik in Mnchen, for their kind hospitality. S.S. is indebted to CERN and Northeastern
University in Boston for warm hospitality and financial support. He would like to also thank the
Galileo Galilei Institute for Theoretical Physics in Firenze for hospitality and INFN for partial
support during completion of this work. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views
of the National Science Foundation.
Appendix A. Operator product expansions and extended supersymmetry
In this appendix we present the N holomorphic spacetime supersymmetry currents and their
operator product expansions. We adapt to the notation and spinor algebra of the book of Wess
and Bagger. In particular, spinor indices are raised and lowered with the anti-symmetric tensors
and . Besides spinor products are defined to be = ( = ) for some

spinors , (,
).
The D = 4 supersymmetry currents in the (1/2)-ghost picture have already
been given in (2.2). The internal Ramond fields I , which belong to an internal superconformal
field theory with c = 9, have conformal dimensions 3/8, while the spacetime spin fields S
have conformal dimensions 1/4 w.r.t. the holomorphic stress tensor T (z). The supersymmetry
currents in the (+1/2)-ghost picture take the form:

1
2
1
I
1/4 e
I
I
V (z) = i
(A.1)
S X + S .
(2 )1/2
2
The dimension 11/8 internal conformal field I appears in the operator product expansion (OPE) of the field I with the internal supercurrent: I (z)TFint (w) = i(2 )1/2 (z
108
w)1/2 I (w) + . The full supercurrent is TF = i(2 )1/2 X + TFint . The supersymmetry currents VI have conformal dimension one. The relevant OPEs for the spin fields are in
D = 4 [24]:
eq1 (z) eq2 (w) = (z w)q1 q2 e(q1 +q2 )(w) + ,
1
S (z)S (w) = (z w)0 (w) + ,

2
1
(w) + ,
S (z)S (w) = (z w)0
2
S (z)S (w) = (z w)1/2 + ,
S (z)S (w) = (z w)1/2 + ,

1
1

(z)S (w) = (z w)1/2 S (w) (z w)1/2 (w) (w) S (w)
2
2
+ ,
1
1
(z)S (w) = (z w)1/2

S (w) (z w)1/2 (w) (w)
S (w)
2
2
+ .
(A.2)
Furthermore we need the following two OPEs:

1
S (z) (w) (w) = (z w)1 S (z) + ,
2

1
S (z) (w) (w) = (z w)1 S (z) + .

2
(A.3)
It is convenient to represent4 the spin fields S , S as exponentials of two free bosons H 1 , H 2 :

, with = ( 1 , 1 ) and
S = eiH , S = ei H
= ( 12 , 12 ), respectively. Their correspond2
2
ing OPEs are:

i
j
i
eis1 H (z) eis2 H (w) = ij (z w)s1 s2 ei(s1 +s2 )H (w) + O (z w)s1 s2 +1 .
(A.4)
The OPEs of two supercharges (2.1)

iX I
1/2
QI (z)Q J (w) =
(z w)1/4
(z) J (w) + ,
(2 )1/2
2(2i)2

1/2
I
J
e(w) (z w)3/4 I (z) J (w) +
Q (z)Q (w) =
(2i)2
(A.5)
and the OPEs (2.3) of the internal Ramond fields I reproduce the spacetime supersymmetry
algebra
I J
I J
Q , Q = I J P ,
(A.6)
Q , Q = Z I J ,
with P = i
dw X
w=z 2i 2
and the central charges Z I J = 1/2
dw (w) I J
(w)
w=z 2i e
(in the
4 In these bosonized expressions we neglect cocycle factors which are required to obtain SO(4) covariant correlation
functions [25].
109
(1)-ghost picture) of the extended supersymmetry algebra. The latter correspond to the compactified fields Z i .
Appendix B. Material for six-point function
In this appendix we present some additional material for the six-point function (3.23), which
has been computed in Section 3.2. In addition to the integrals (3.24) the remaining seven integrals
Ki determining the complete seven-point function (3.23) are:
1
K7 =
1
1
dx
0
dy
0
1
K8 =
dz
1
dx
1
dy
dz
1
1
1
K9 =
dx
dy
1
1
1
K10 =
dx
0
dy
0
1
K11 =
dx
0
K12 =
1
1
0

s13
I(x, y, z)
dz 1 s13 +
,
x (1 z)(1 yz)

s13
I(x, y, z)
dz 1 s13 +
,
x (1 y)(1 z)
dy
0

dz
1
dy
s12 (1 x)
I(x, y, z),
x(1 y)(1 z)(1 xyz)
1
1
dx
s12 (1 x)
I(x, y, z),
x(1 z)(1 xy)(1 yz)
dz
0

1 s23
I(x, y, z)
+ s23
,
x
xy(1 z)(1 yz)

1 s23
I(x, y, z)
+ s23
,
x
xyz(1 y)(1 z)

(1 s14 )(1 s23 )
K13 = dx dy dz (1 s13 )(1 s24 ) +
x2
0
0
0

s12 s34 s14 s23 s13 s24 I(x, y, z)
+
.
x
y(1 z)2
1
1
1
(B.1)
The six-point amplitude (3.23) is completely specified by the basis of six functions {K1 , . . . , K6 },
introduced in (3.24). Therefore the above integrals (B.1) can be expressed in terms of this basis:
(s4 + s5 t1 )t1 K7 = (s1 + s6 t3 )(s3 + s4 t1 t3 )K1 + (s1 + s4 t1 t3 )
(s3 + s6 t2 t3 )K2 + (s5 + s6 t2 )(s1 + s6 t3 )K3
(s5 + s6 t2 )(s1 + s4 t1 t3 )K4 + s6 (s3 + s6 t2 t3 )K5
s6 (s3 + s4 t1 t3 )K6 ,
s4 t1 K8 = (s3 + s4 t3 )(s1 + s6 t3 )K1 (s1 + s4 t3 )(s3 + s6 t2 t3 )K2
(s5 + s6 t2 )(s1 + s6 t3 )K3 + (s5 + s6 t2 )(s1 + s4 t3 )K4
s6 (s3 + s6 t2 t3 )K5 + s6 (s3 + s4 t3 )K6 ,
110
(s4 + s5 t1 )t1 K9 = (s3 + s4 t3 )(s1 + s6 t3 )K1 + (s1 + s4 t1 t3 )

(s3 + s6 + t1 t2 t3 )K2 + (s5 + s6 t2 )(s1 + s6 t3 )K3
(s5 + s6 t2 )(s1 + s4 t1 t3 )K4 + s6 (s3 + s6 + t1 t2 t3 )K5
s6 (s3 + s4 t3 )K6 ,
s4 t1 K10 = (s3 + s4 t3 )(s1 + s6 t1 t3 )K1 (s1 + s4 t1 t3 )(s3 + s6 t2 t3 )K2
(s5 + s6 t2 )(s1 + s6 t1 t3 )K3 + (s5 + s6 t2 )(s1 + s4 t1 t3 )K4
s6 (s3 + s6 t2 t3 )K5 + s6 (s3 + s4 t3 )K6 ,
(s4 + s5 t1 )t1 K11 = (s3 + s4 t3 )(s1 + s6 t3 )K1 + (s1 + s4 t1 t3 )
(s3 + s6 t2 t3 )K2 + (s5 + s6 t1 t2 )(s1 + s6 t3 )K3
(s5 + s6 t1 t2 )(s1 + s4 t1 t3 )K4
+ s6 (s3 + s6 t2 t3 )K5 s6 (s3 + s4 t3 )K6 ,
s4 t1 K12 = (s3 + s4 t3 )(s1 + s6 t3 )K1 (s1 + s4 t1 t3 )(s3 + s6 t2 t3 )K2
(s5 + s6 t2 )(s1 + s6 t3 )K3 + (s5 + s6 t2 )(s1 + s4 t1 t3 )K4
(s6 + t1 )(s3 + s6 t2 t3 )K5 + (s6 + t1 )(s3 + s4 t3 )K6 ,
t1 K13 = (s3 + s4 t3 )(s1 + s6 t3 )K1 (s1 + s4 t1 t3 )(s3 + s6 t2 t3 )K2
(s5 + s6 t2 )(s1 + s6 t3 )K3 + (s5 + s6 t2 )(s1 + s4 t1 t3 )K4
s6 (s3 + s6 t2 t3 )K5 + s6 (s3 + s4 t3 )K6 .
(B.2)
In (3.33) we have expressed the function K3 w.r.t. to the basis {F1 , F2 , F3 , F4 , F5 , F6 }, introduced in (3.32). Similarly the remaining five basis elements are expressed in terms of the basis
{F1 , F2 , F3 , F4 , F5 , F6 }:

s2 s5 K1 = s2 s6 t2 F1 + s6 s1 s2 + s2 s5 (s4 + s5 t1 )(s5 t2 ) F2
+ (s4 + s5 t1 )(s5 t2 )(s1 s3 + s5 + t3 )F6

+ s12 s2 + s1 s2 (s4 + s5 + t1 t2 + t3 )

(s5 t2 ) s2 (s5 + t3 ) (s4 + s5 t1 )(s3 s5 + s6 + t1 t3 ) F3

(s4 + s5 t1 ) s1 s2 + (s5 t2 )(s3 + s5 t1 + t3 ) F4

+ s12 s2 (s4 + s5 t1 )(s3 s5 + t1 )(s5 t2 ) + s1 (s4 + s5 t1 )(s5 t2 )

+ s2 (s4 t1 t3 ) F5 ,

s2 s5 K2 = s2 s6 t2 F1 + s6 s1 s2 + (s4 + s5 t1 )(s5 t2 ) F2 + (s4 + s5 t1 )(s5 t2 )

(s1 + s3 s5 t3 )F6 + s12 s2 + s53 s3 (s4 + s5 t1 )(s5 t2 ) t1 (s6 + t1 )t2
+ s1 s2 (s4 t1 + t2 t3 ) s52 (s6 + 2t1 + t2 t3 )

(s2 t1 )t2 t3 + s4 (s5 t2 )(s5 s6 t1 + t3 ) + s5 s6 (t1 + t2 ) + t1 (t1 + 2t2 )

(t1 + t2 )t3 + s2 (s6 + t3 ) F3

+ (s4 + s5 t1 ) s1 s2 + (s5 t2 )(s3 + s5 t1 + t3 ) F4

+ s12 s2 + (s4 + s5 t1 )(s3 s5 + t1 )(s5 t2 ) s1 (s4 + s5 t1 )(s2 + s5 t2 )

+ s1 s2 t3 F5 ,
111

s2 s5 K4 = s2 s6 (s3 t2 )F1 + s6 s1 s2 + (s4 + s5 t1 )(s3 t2 ) F2 + (s4 + s5 t1 )

(s3 t2 )(s1 + s3 s5 t3 )F6 + s12 s2 s32 (s4 + s5 t1 ) + s3 (s4 + s5 t1 )
(s5 s6 t1 + t2 ) + s3 (s2 + s4 + s5 t1 )t3 (s4 + s5 t1 )t2

(s5 s6 t1 + t3 ) s1 s2 (s3 s4 + t1 t2 + t3 ) s2 s5 (s6 t3 ) + t2 t3 F3

+ (s4 + s5 t1 ) s1 s2 (s3 t2 )(s3 s5 + t1 t3 ) F4 + (s4 + s5 t1 )

(s3 s5 + t1 )(s3 t2 ) s12 s2 s1 (s4 + s5 t1 )(s2 + s3 t2 ) + s1 s2 t3 F5 ,
s2 s5 K5 = s2 s3 (s5 + s6 )F1 s3 s6 (s4 + s5 t1 )F2 + s3 (s4 + s5 t1 )

(s5 + t3 s1 s3 )F6 + (s3 s5 + t1 t3 )F4 + s3 s1 s2 + (s4 + s5 t1 )

(s3 s5 + s6 + t1 t3 ) s2 (s5 + t3 ) F3 + s3 (s1 s3 + s5 t1 )
(s4 + s5 t1 )F5 ,

s2 s5 K6 = s2 (s5 + s6 )(s3 t2 )F1 + s6 (s4 + s5 t1 )(t2 s3 ) s1 s2 s6

+ s2 s5 (s6 t3 ) F2 + (s4 + s5 t1 )(s3 t2 )(s5 + t3 s1 s3 )F6

(s4 + s5 t1 ) s1 s2 (s3 t2 )(s3 s5 + t1 t3 ) F4

+ s1 s2 (s3 s4 + t1 t2 + t3 ) s12 s2

+ (s3 t2 ) (s4 + s5 t1 )(s3 s5 + s6 + t1 t3 ) s2 (s5 + t3 ) F3

+ s12 s2 (s4 + s5 t1 )(s3 s5 + t1 )(s3 t2 )

+ s1 (s4 + s5 t1 )(s3 t2 ) + s2 (s4 t1 t3 ) F5 .
(B.3)
Appendix C. Material for seven-point function

In this appendix we present some additional material for the seven-point function (3.40),
which has been computed in Section 3.3. The complete amplitude is specified by 24 integrals
{K1 , . . . , K24 }, whose integrands contain the common factor (3.43). The integrals may be written
in the following way
1
Ki =
1
dx
1
dy
1
dw Pi I(x, y, z, w),
dz
0
with the following 24 polynomials:

s34
s24
1
P1 =
,
P2 = 1 s24 +
,
xyw(1 z)(1 zw)
x yw(1 z)(1 zw)(1 xyz)

s34
s24
1
P3 =
,
P4 = 1 s24 +
,
xyzw(1 z)(1 w)
x yzw(1 z)(1 w)(1 xy)

s34
1 s14
z
P5 =
,
P6 =
+ s14
,
xy(1 z)(1 zw)
x
(1 z)(1 zw)(1 xyz)

s34
1 s14
1
P7 =
,
P8 =
+ s14
,
xy(1 z)(1 w)
x
(1 z)(1 w)(1 xy)

s24
1
P9 = 1 s24 +
,
x y(1 z)(1 zw)(1 xyzw)
112

z
1 s14
+ s14
,
P10 =
x
(1 z)(1 wz)(1 xyzw)

1
s24
P11 = 1 s24 +
,
x y(1 z)(1 w)(1 xyzw)

1
1 s14
+ s14
,
P12 =
x
(1 z)(1 w)(1 xyzw)
s34
,
P13 =
xyz(1 w)(1 zw)

1
s24
P14 = 1 s24 +
,
x yz(1 w)(1 xy)(1 zw)
s34
,
P15 =
xy(1 w)(1 zw)

1
1 s14
P16 =
+ s14
,
x
(1 w)(1 xy)(1 zw)

1
s24
,
P17 = 1 s24 +
x y(1 w)(1 zw)(1 xyz)

1
1 s14
+ s14
,
P18 =
x
(1 w)(1 zw)(1 xyz)

and:

1 s14
1
1
s24
P19 =
+ s14
1 s24 +
x yz(1 w)
x
1 wz

1
(1 x)s47
,
+
x(1 yzw) w(1 z)(1 xy)

1
s34
1 s14
+ s14
P20 =
x
(1 zw)(1 xyz) (1 w)xyz

1
(1 x)s47
,
x(1 yzw)(1 xyz) w(1 z)

1
1
1 s14
s24
+ s14
1 s24 +
P21 =
x
(1 w)
x y(1 zw)

1
(1 x)s47
,
x(1 yzw) (1 z)(1 xy)

z
s34
s24
+
P22 = 1 s24 +
x y(1 zw)(1 xyz) xy(1 w)

1
z(1 x)s47
,
x(1 xyz)(1 yzw) (1 z)

1
1
1 s14
s34
+ s14
+ s34
P23 =
x
(1 w)
xy 1 zw

(1)
(1 x)s47
,
x(1 yzw) (1 z)(1 xyzw)

(C.1)

1
z
s24
s34
s34
P24 = 1 s24 +
x y(1 w)
xy 1 zw

z
(1)
s47
.
s47
x (1 yzw) (1 z)(1 xyzw)
113
(C.2)
References
[1] M.L. Mangano, S.J. Parke, Multiparton amplitudes in gauge theories, Phys. Rep. 200 (1991) 301, hep-th/0509223.
[2] L.J. Dixon, Calculating scattering amplitudes efficiently, hep-ph/9601359.
[3] S.J. Parke, T.R. Taylor, Perturbative QCD utilizing extended supersymmetry, Phys. Lett. B 157 (1985) 81;
S.J. Parke, T.R. Taylor, Phys. Lett. B 174 (1986) 465, Erratum.
[4] M.T. Grisaru, H.N. Pendleton, Some properties of scattering amplitudes in supersymmetric theories, Nucl. Phys.
B 124 (1977) 81.
[5] M.T. Grisaru, H.N. Pendleton, P. van Nieuwenhuizen, Supergravity and the S matrix, Phys. Rev. D 15 (1977) 996.
[6] F.A. Berends, W.T. Giele, Recursive calculations for processes with n gluons, Nucl. Phys. B 306 (1988) 759.
[7] E. Witten, Perturbative gauge theory as a string theory in twistor space, Commun. Math. Phys. 252 (2004) 189,
hep-th/0312171.
[8] F. Cachazo, P. Svrcek, E. Witten, MHV vertices and tree amplitudes in gauge theory, JHEP 0409 (2004) 006, hepth/0403047.
[9] J.D. Lykken, Weak scale superstrings, Phys. Rev. D 54 (1996) 3693, hep-th/9603133.
[10] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, New dimensions at a millimeter to a Fermi and
superstrings at a TeV, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[11] I. Antoniadis, A possible new dimension at a few TeV, Phys. Lett. B 246 (1990) 377.
[12] S. Cullen, M. Perelstein, M.E. Peskin, TeV strings and collider probes of large extra dimensions, Phys. Rev. D 62
(2000) 055012, hep-ph/0001166.
[13] S. Stieberger, T.R. Taylor, Amplitude for N -gluon superstring scattering, Phys. Rev. Lett. 97 (2006) 211601, hepth/0607184.
[14] S. Stieberger, T.R. Taylor, Multi-gluon scattering in open superstring theory, Phys. Rev. D 74 (2006) 126007, hepth/0609175.
[15] S.J. Parke, T.R. Taylor, An amplitude for n gluon scattering, Phys. Rev. Lett. 56 (1986) 2459.
[16] S. Stieberger, T.R. Taylor, in preparation.
[17] A. Hashimoto, I.R. Klebanov, Scattering of strings from D-branes, Nucl. Phys. B (Proc. Suppl.) 55 (1997) 118,
hep-th/9611214.
[18] D. Friedan, E.J. Martinec, S.H. Shenker, Conformal invariance, supersymmetry and string theory, Nucl. Phys. B 271
(1986) 93.
[19] T. Banks, L.J. Dixon, Constraints on string vacua with spacetime supersymmetry, Nucl. Phys. B 307 (1988) 93.
[20] S. Ferrara, D. Lst, S. Theisen, World sheet versus spectrum symmetries in heterotic and type II superstrings, Nucl.
Phys. B 325 (1989) 501.
[21] J. Polchinski, String Theory, Cambridge Univ. Press, 1998 (Sections 6 and 12).
[22] L.A. Barreiro, R. Medina, 5-field terms in the open superstring effective action, JHEP 0503 (2005) 055, hepth/0503182;
R. Medina, F.T. Brandt, F.R. Machado, The open superstring 5-point amplitude revisited, JHEP 0207 (2002) 071,
hep-th/0208121.
[23] D. Oprisa, S. Stieberger, Six gluon open superstring disk amplitude, multiple hypergeometric series and Euler
Zagier sums, hep-th/0509042.
[24] T. Banks, L.J. Dixon, D. Friedan, E.J. Martinec, Phenomenology and conformal field theory or can string theory
predict the weak mixing angle? Nucl. Phys. B 299 (1988) 613.
[25] V.A. Kostelecky, O. Lechtenfeld, W. Lerche, S. Samuel, S. Watamura, Conformal techniques, bosonization and tree
level string amplitudes, Nucl. Phys. B 288 (1987) 173.

Extra-dimensional relaxation of the upper limit

of the lightest supersymmetric neutral Higgs mass
Gautam Bhattacharyya a, , Swarup Kumar Majee b,c ,
Amitava Raychaudhuri b,c
a Saha Institute of Nuclear Physics, 1/AF Bidhan Nagar, Kolkata 700064, India
b Harish-Chandra Research Institute, Chhatnag Road, Jhunsi, Allahabad 211019, India
c Department of Physics, University of Calcutta, 92 A.P.C. Road, Kolkata 700009, India
Received 4 June 2007; accepted 28 September 2007

Abstract
The upper limit on the mass of the lightest CP-even neutral Higgs in the minimal supersymmetric standard
model is around 135 GeV for soft supersymmetry breaking masses in the 1 TeV range. We demonstrate that
this upper limit may be sizably relaxed if supersymmetry is embedded in extra dimensions. We calculate,
using the effective potential technique, the radiative corrections to the lightest Higgs mass induced by the
KaluzaKlein towers of quarks and squarks with one and two compactified directions. We observe that the
lightest Higgs may comfortably weigh around 200 GeV (300 GeV) with one (two) extra dimension(s).
PACS: 12.60.Jv; 11.10.Kk; 14.80.Cp
Keywords: Supersymmetry; Extra dimension; Higgs mass
1. Introduction
The most general symmetries of local relativistic quantum field theories include supersymmetry, a phenomenological version [1] of which is awaiting a final judgement within the next few
years as the Large Hadron Collider (LHC) turns on. Indeed, one of the most coveted targets of the
LHC is to capture the Higgs boson, and supersymmetry, admitting chiral fermions together with
E-mail address: gautam.bhattacharyya@saha.ac.in (G. Bhattacharyya).

doi:10.1016/j.nuclphysb.2007.09.032
G. Bhattacharyya et al. / Nuclear Physics B 793 (2008) 114130
115
their scalar partners in the same representations, tacitly provides a rationale for treating the Higgs
as an elementary object [2]. Furthermore, through the removal of the quadratic divergence that
plagues the ordinary Higgs mass, phenomenological supersymmetry has emerged as a leading
candidate of physics beyond the standard model (SM). A key signature of the minimal version
of supersymmetry is that the lightest Higgs boson mass obeys an upper bound ( 135 GeV, see
below)a prediction which will be put to test during the LHC run. Now, supersymmetry is an integral part of string theory which attempts to provide a quantum picture of all interactions. Since
string theory is intrinsically a higher-dimensional theory, a reanalysis of some 4-dimensional
(4d) supersymmetric wisdom in the backdrop of extra dimensions might provide important clues
to our search strategies. Considering that the Higgs is the most-wanted entity at the LHC, in this
paper we address the following question which we believe is extremely timely: What is the upper
limit of the lightest CP-even neutral Higgs mass if the minimal supersymmetric standard model
(MSSM) is embedded in extra dimensions? We consider the embedding first in one and then in
two extra dimensions.
Let us first discuss why this is an important issue. Recall that MSSM has two Higgs doublet superfields (H1 and H2 ), and supersymmetry does not allow the scalar potential to have
independent quartic couplings. Gauge interactions generate them through supersymmetry breaking D-terms and the effective quartic interactions are written in terms of the gauge couplings.
This makes the Higgs spectrum partially predictive, in the sense that at the tree level the lightest neutral Higgs (h) weighs less than MZ (m2h < MZ2 cos2 2, where tan is the ratio of two
vacuum expectation values (VEVs)). However, mh receives quantum corrections which, due
to the large top quark
Yukawa coupling and for heavy stop squarks, can become as large as
m2h (3GF m4t / 2 2 ) ln(m2t /m2t ), where mt is an average stop squark mass [3,4]. The upper
limit on mh is then pushed to around 135 GeV for squark mass in the O(TeV) range. Notice that
the non-observation of the Higgs boson at LEP2 has already set a lower limit mh > 114.5 GeV
[5,6], which is satisfied only if a sizable quantum correction elevates the Higgs mass beyond
the tree level upper limit of MZ . This implies (i) lower values of tan , which is usually chosen
in the range 1 < tan < mt /mb , are disfavoured, and (ii) the squark mass mt has to be in the
TeV range, which also sets the scale of a generic soft supersymmetry breaking mass MS . The
MSSM prediction of a light Higgs is also in line with the indication coming from electroweak
precision tests that the neutral Higgs should weigh below 199 GeV1 [7]. The so-called little
hierarchy problem then arises out of an order of magnitude mass splitting between the Higgs
and the superparticles.
Adding a gauge singlet superfield (N ) in the MSSM spectrum and coupling it with H1,2 via
the superpotential NH1 H2 helps to ease the tension. Not only does this next to minimal version
of supersymmetry (the so-called NMSSM [8]) help to address the problem, it also generates
a tree level quartic coupling in the scalar potential which modifies the tree level upper limit on
mh through m2h < MZ2 cos2 2[1 + 22 tan2 2/(g 2 + g 2 )] (see [9]). Assuming to be in the
perturbative regime, i.e., g, g , one basically obtains a new contribution MZ2 sin2 2 to
the tree level m2h . This way the low tan regime can be revived. Since many supersymmetric
1 This indirect upper limit as well as the LEP2 direct search lower limit of m > 114.5 GeV apply, strictly speaking,
h
for the SM Higgs. However, in the decoupling limit of the MSSM (large mA leading to full-strength ZZh coupling),
which is the region of interest in the present paper, the above limits continue to hold.
116
couplings depend on tan , search strategies alter in a significant way if the disfavoured low
tan region is thus resurrected.2,3
In this paper we adopt a different approach which also revives the low tan region. We stick
to the MSSM particle content, but embed it in a higher dimension compactified at the inverse TeV
scale [12]. Although we argued in the beginning that string theory provides a rationale for linking
the two ideas, namely, supersymmetry and extra dimension, establishing any rigorous connection
between the two at the level of phenomenological models is still a long shot. Here we take a
bottomup approach: we first outline what has already been studied in the phenomenological
context of TeV scale extra-dimensional scenarios, and then illustrate what we aim to achieve in
this paper.
(1) Consider first scenarios with one extra dimension (with inverse radius of compactification around a TeV) but without supersymmetry. A typical model is the universal extradimensional scenario (UED) where all particles access the extra dimension [13]. Constraints
on this scenario from g 2 of the muon [14], flavour changing neutral currents [1517],
Z bb decay [18], the parameter [13,19], other electroweak precision tests [20], implications from hadron collider studies [21], etc., imply that R 1 300 GeV. A recent
inclusive B Xs analysis sets a stronger constraint R 1 600 GeV [22]. These scenarios
are motivated from many phenomenological angles. They could lead to a new mechanism of
supersymmetry breaking [23], address the fermion mass hierarchy in an alternative way [24],
provide a cosmologically viable dark matter candidate [25], stimulate power law renormalization group running [12,26], admit substantial evolution of neutrino mixing angles defined
through an effective Majorana neutrino mass operator [27], etc.4
(2) Our object of interest is a supersymmetric theory (e.g. MSSM) but embedded in a higher
dimension. Here we ask the following question: What would be the shift in the Higgs mass
due to radiative effects induced by extra dimensions? In the kind of scenarios we consider, the
SM bosons along with their superpartners access the higher-dimensional bulk. Additionally,
SM fermions of one or more generations together with their superpartners also do so. From a
4d perspective, all the states which access the bulk will have KaluzaKlein (KK) towers. The
zero modes, i.e., those states which do not have any momenta along the extra coordinates,
are identified with the standard 4d MSSM spectra. Now, not only the top quark and the stop
squarks would contribute to the radiative correction to m2h , their KK partners would do so as
well. As it turns out, the radiative correction driven by the KK states has the same sign as the
one from the zero modes. As a result, m2h becomes larger and thus the upper limit on mh
is pushed to higher values beyond the usual 4d MSSM limit of around 135 GeV. As we shall
see, in the absence of any leftright scalar mixing, the new contribution coming from KK
modes is to a good approximation proportional to R 2 (m2t m2t )/n2 . This fits our intuition
that the KK contribution falls with higher KK modes and vanishes both when R 0 and in
the limit of exact supersymmetry. We can interpret the result in two ways. Either, we take
large tan and O(TeV) squark mass that yielded the 4d supersymmetry limit 135 GeV,
in which case the new upper limit shoots up by several tens of GeV. Or, we may admit
lower tan and/or accommodate lighter zero mode squarks which were hitherto disfavoured
2 Low tan is preferred by electroweak baryogenesis as well [10].
3 The constraint arising from perturbativity of couplings can be evaded if the Higgs is charged under an asymptotically
free gauge group [11].

4 Ultraviolet cutoff sensitivity in different kinds of TeV scale extra-dimensional models has been dealt in [28].
117
in the 4d context. Either way, the Higgs phenomenology gets an interesting twist which is
intuitively comprehensible and analytically tractable, owing largely due to the fact that we
are here dealing with only one additional parameter, namely, the radius of compactification.
Moreover, the top quark mass which appears with fourth power in the expression of m2h is
now known to a precision better than ever (mt = 170.9 1.8 GeV [29]).
As mentioned before, we have considered the embedding of 4d supersymmetry in one as well
as two extra dimensions. There are quite a few advantages of considering a 6d gauge theory [30]
even in a non-supersymmetric scenario: (i) number of fermion generations is restricted to three,
or a multiple of it, if the global SU(2) gauge anomaly has to cancel [31], (ii) proton decay is
adequately suppressed, which is difficult to achieve in 5d UED, thanks to a discrete symmetry
that survives as a subgroup of the 6d Lorentz group [32], (iii) observed neutrino masses and
mixings can be nicely explained [33], and (iv) KK vector modes offer better opportunities to be
explored [34]. We shall see that qualitatively the KK contributions to the radiative corrections
of mh from 5d and 6d theories are similar, the quantitative estimates differ due to the different
density of KK states in the two cases. In 5d, the KK states are spaced as n/R (modulo their zero
mode masses) where n, an integer, is the KK number, whereas in 6d, a similar expression holds
except n2 j 2 + k 2 , where j and k are two different sets of KK numbers corresponding to the
two compactified directions.
Section 2 is basically a review of the standard derivation of the upper limit of the lightest
neutral Higgs in conventional 4d MSSM in the effective potential approach. This paves the way,
in Section 3, to upgrade the above derivation for accommodating contributions from the KK
modes of the top quark and squarks in 5d and 6d scenarios. In Section 4, we shall comment
on the numerical impact of the higher KK modes on the lightest neutral Higgs mass and its
consequences. We shall draw our conclusion in the final section.
2. MSSM neutral Higgs spectrum in 4 dimensions
2.1. Tree level mass relations
MSSM requires two Higgs doublets for three good reasons: (i) to avoid massless charged
degrees of freedom, (ii) to maintain analyticity of the superpotential, and (iii) to keep the theory
free from chiral anomaly, which requires two Higgs doublets with opposite hypercharges.
We denote these two complex scalar doublets as
0
+
H2
H1
H1 =
(1)
=
,
H
,
2
H1
H20
whose SU(2) U(1) quantum numbers are (2, 1) and (2, +1) respectively. H10 couples with
down-type quarks and charged leptons, while H20 couples with up-type quarks. This guarantees
natural suppression of flavour-changing neutral currents in the limit of exact supersymmetry. The
tree level potential involving these two doublets is given by
V = m21 |H1 |2 + m22 |H2 |2 + m212 (H1 H2 + h.c.)
2 1 2
2
1
+ g 2 H2 a H2 + H1 a H1 + g |H2 |2 |H1 |2 ,
(2)
8
8
where m21 , m22 and m212 are soft supersymmetry breaking mass parameters, g and g are the SU(2)
and U(1) gauge couplings, and a (a = 1, 2, 3) are the Pauli matrices. Note that the quartic
118
coupling is related to the gauge couplings. The part involving the neutral fields is given by
2
2

2 2 2
1
V0 = m21 H10 + m22 H20 m212 H10 H20 + h.c. + g 2 + g 2 H10 H20 .
(3)
8
After spontaneous symmetry breaking the minimum
the following two VEVs:
of V0 involves
0
0
2
2
H1 = v1 and H2 = v2 . The combination v = v1 + v2 = ( 2GF )1/2
246 GeV sets the
Fermi scale.
Now, out of the eight degrees of freedom contained in the two Higgs doublets three are absorbed as the longitudinal modes of the W and the Z bosons, while the remaining five modes
appear as physical states. Of these five states, two are charged (H ) and three are neutral
(h, H, A). Our present concern is the neutral sector of which (h, H ) are CP-even, while A is
CP-odd. From the separate diagonalisation of the CP-odd and CP-even neutral mass matrices
two important relations emerge:
2m212
v2
(4)
, where tan = ,
sin 2
v1

2
1
m2h,H = m2A + MZ2 m2A + MZ2 4m2A MZ2 cos2 2 ,
(5)
2
where, by definition, h is the lighter of the two CP-even Higgs. These in turn give rise to the
following sum rule and inequality:
m2A =
m2h + m2H = m2A + MZ2 ,
(6)
mh min(mA , MZ )| cos 2| min(mA , MZ ),
(7)
i.e., at the tree level (i) the lighter of the two CP-even Higgs (h) weighs less than MZ , and (ii)
the CP-odd Higgs (A) is heavier than h but lighter than H .
2.2. Radiative corrections
We shall now discuss how the above tree level relations are affected by quantum loops [3,4].
We shall confine our discussion on the correction to mh only, and that too at the one-loop level.
We note two important points:
1. Radiative corrections to mh are dominated by the top quark Yukawa coupling (ht ) and the
masses of the stop squarks (t1 , t2 ). For large values of tan , the contributions from the
b-quark sector also assume significance. We shall ignore loop contributions mediated by
lighter quarks or the gauge bosons.
2. The tree level Higgs mass is protected by supersymmetry. In the limit of exact supersymmetry, the entire quantum correction vanishes. So radiative corrections to mh will be controlled
by MS .
Three different approaches have been adopted in the literature to calculate the radiative corrections to mh : (i) effective potential technique, (ii) direct diagrammatic calculations, and (iii)
renormalisation group (RG) method, assuming MS MZ and fixing the quartic coupling proportional to (g 2 + g 2 ) at that scale and then evolving down to weak scale. In this paper, we shall
follow the effective potential approach primarily for the sake of conveniently including the effect
of new physics later.
119
We first start with an RG-improved tree level potential V0 (Q) which contains running masses
m2i (Q) and running gauge couplings gi (Q). The full one-loop effective potential is now given
by
V1 (Q) = V0 (Q) + V1 (Q),
(8)
where, in terms of the field dependent masses M(H ),

1
M 2 (H ) 3
4
Str
M
(H
)
ln
.
V1 (Q) =
2
64 2
Q2
(9)
The Q-dependence of V1 (Q) cancels against that of V0 (Q) making V1 (Q) independent of Q
up to higher loop orders. The supertrace in Eq. (9), defined through

(1)2Ji (2Ji + 1)f m2i ,
Str f m2 =
(10)
i
has to be taken over all members of a supermultiplet and where m2i m2i (H ) is the fielddependent mass eigenvalue of the particle i with spin Ji . As an example, the contribution from
the chiral multiplet containing the top quark and squarks is given by

m2
m2

3
3
3
3
m2t
t1
t2
4
4
4
+
m
2m
,
m
ln
ln
ln
Vt =
(11)
t
t1
t2
2
2
32 2
Q2
Q2
Q2 2
where the overall factor of 3 comes from colour. Note that mti and mt in Eq. (11) are field
dependent masses. Even though hb ht , the contribution from the bottom supermultiplet turns
out to be numerically significant in the large tan region. Vb can be written analogously to
Vt with the appropriate replacements of top and stop masses by bottom and sbottom masses
respectively.
We now explicitly write down the field dependent mass terms. This simply means a replacement of vi by Hi0 (i = 1, 2) wherever vi appear in the expression of masses. The field dependent
top and bottom quark masses are given by
2
2
m2b (H ) = h2b H10 .
m2t (H ) = h2t H20 ;
(12)
The field dependent stop and sbottom squark mass matrices are written as

ht (At H20 + H10 )
m2Q + h2t |H20 |2
2
,
Mt (H ) =
m2U + h2t |H20 |2

ht (At H20 + H10 )
and

Mb2 (H ) =
m2Q + h2b |H10 |2
hb (Ab H10 + H20 )
hb (Ab H10 + H20 )
m2D + h2b |H10 |2
(13)

.
(14)
In Eqs. (13) and (14) mQ , mU and mD are soft supersymmetry breaking masses, At and Ab are
trilinear soft supersymmetry breaking mass dimensional couplings, and is the supersymmetry
preserving mass dimensional parameter connecting H1 and H2 in the superpotential. We take
both trilinear and the couplings to be real. We have neglected the D-term contributions which
are small, being proportional to gauge couplings. The squark masses appearing in Eq. (11) are
obtained from the diagonalisation of Eq. (13).
We now consider the radiative correction to the CP-odd scalar mass matrix. The one-loop
corrected mass matrix square, obtained by taking double derivatives of the full potential with
120
respect to the pseudo-scalar excitations, can be written as

2

tan
1
m12 + .
M2(odd) =
1
cot
(15)
The radiative corrections generated as a consequence of supersymmetry breaking are contained

in = t + b , which is given by
t(b) =
h2t (b) At (b) 2

3
f mt (b ) f m2t (b ) ,
2
2
2
1 1
2 2
32 [m m ]
t1 (b1 )
(16)
t2 (b2 )
where

m2
f m2 = 2m2 ln 2 1 .
(17)
Q
The zero eigenvalue corresponds to the massless Goldstone boson which is eaten by the Z boson.
The massive state is the pseudo-scalar A whose radiatively corrected mass square is given by
2(m212 + )
(18)
.
sin 2
The Q-dependence of mA cancels in Eq. (18) up to one-loop order. In any case, we shall treat
the radiatively corrected mA as an input parameter.
Now we are all set to calculate the radiative corrections in the neutral CP-even mass eigenvalues. The one-loop corrected mass matrix square is obtained by taking double derivatives of the
full potential with respect to the scalar excitations and is given by

2
MZ cos2 + m2A sin2 (m2A + MZ2 ) sin cos
M2(even) =
(m2A + MZ2 ) sin cos MZ2 sin2 + m2A cos2

3
11 12
+
(19)
,
4 2 v 2 12 22
m2A =
where ij = tij + bij . The individual ij s are explicitly written below:

m4
(At + cot ) 2 2 2
t11 = 2t
g mt , mt ,
1
2
m2t m2t
sin
1
2

m2
m4 (At + cot )
At (At + cot ) 2 2
t1
,
+
g
m
,
m
ln
t12 = 2t
t1
t2
m2t m2t
m2t
m2t m2t
sin
1
2
2
1
2
2
4 m2 m2
m
(A
+
cot
)
m
2A
t
t
t t
t
ln 21
t22 = 2t ln 1 4 2 +
mt
m2t m2t
mt
sin
1
2
2

At (At + cot ) 2 2 2
+
g mt , mt ,
1
2
m2t m2t
b11

m4b
ln
cos2
m2 m2
2
2Ab (Ab + tan) mb1
=
+
ln 2
m4b
m2 m2
m
b1
b2
b2

2

Ab (Ab + tan )
+
g m2b , m2b ,
2
2
1
2
m m
b1
b1
b2
b2
m2

m4b (Ab + tan)
Ab (Ab + tan) 2
b1
2
ln
,
+
g
m
,
m
b1
b2
cos2 m2 m2
m2
m2 m2
b1
b2
b2
b1
b2

m4b
(Ab + tan ) 2 2
b22 =
g mb , m2b ,
2
2
2
1
2
cos
m m
121
b12 =
b1
(20)
b2
where

m2 + m22 m21
g m21 , m22 = 2 12
ln
.
m1 m22 m22
(21)
Two points deserve mention at this stage:

(1) While the leading log contribution appears in 22 for the top sector, the same appears in 11
for the bottom sector. This happens because the right-handed top quark couples to H2 while
the right-handed bottom quark couples to H1 . In the absence of any leftright scalar mixing,
these leading logs are the only radiative contributions.
(2) Ignoring the leftright scalar mixing, the radiative shift to the Higgs mass square coming
from the topstop sector turns out to be m2h = (3/4 2 v 2 )t22 sin2 (3m4t /2 2 v 2 )
ln(m2t /m2t ), where mt = mt1 mt2 is an average stop mass. This is the expression we quoted
in the Introduction.
3. Radiative corrections due to extra dimensions
Let us first consider just one extra dimension which is compactified on a circle of radius R. We
further consider a Z2 orbifolding identifying y y, where y denotes the compactified coordinate. The orbifolding is crucial for reproducing the chiral zero modes of the observed fermions.
After the compactified direction is integrated out, the 4d Lagrangian can be written in terms of
the zero modes and their KK partners. For illustration, we first take a non-supersymmetric scenario and look into the KK mode expansion of gauge boson, scalar and fermion fields from a 4d
perspective. Each component of a 5d field is either even or odd under Z2 . The KK expansions
are given by,
2
2 (n)
ny
A (x, y) =
(x)
+
A (x) cos ,
A(0)
R
2R
2R
n=1
2
ny
(n)
A5 (x, y) =
A5 (x) sin ,
R
2R n=1
2 (0)
2 (n)
ny
(x, y) =
(x) cos ,
(x) +
R
2R
2R n=1

2
ny
ny
t
(n)
(n)
Q(x, y) =
QL (x) cos
(x) + 2
+ QR (x) sin
,
b L
R
R
2R
n=1

2
ny
ny
(n)
(n)
TR (x) cos
+ TL (x) sin
T (x, y) =
tR (x) + 2
,
R
R
2R
n=1
122

ny
ny
(n)
(n)
bR (x) + 2
B(x, y) =
+ BL (x) sin
.
BR (x) cos
R
R
2R
n=1
(22)
The complex scalar field (x, y) and the gauge boson A (x, y) are Z2 -even fields with their
zero modes identified with the SM scalar doublet and a SM gauge boson respectively. The field
A5 (x, y) is a real pseudoscalar field transforming in the adjoint representation of the gauge group,
and does not have any zero mode. The fields Q, T , and B denote the 5d quark SU(2) doublet and
SU(2) singlet states of a given generation (e.g. third generation) whose zero modes are identified
with the 4d SM quark states. We draw attention to two points at this stage: (i) even though the Z2
orbifolding renders the zero mode fermions to be chiral, the KK fermions are vector-like; (ii) the
KK number n is conserved at all tree level vertices, while what actually remains conserved at
all order is the KK parity, defined as (1)n . As is well known, the kinetic terms in the extra
dimension give rise to the KK masses in 4d. For a flat extra dimension the KK masses are added
in quadrature with the corresponding zero mode masses both for fermions and bosons. A generic
expression for the nth mode mass, where m0 is the zero mode mass, is given by
m2n = m20 +
n2
,
R2
n = 0, 1, 2, . . . .
(23)
We now discuss the supersymmetric version of the theory. A 5d N = 1 supersymmetry from a

4d perspective appears as two N = 1 supersymmetries forming an N = 2 theory. For the details
of the hypermultiplet structures of this theory, we refer the readers to [12]. Our concern in this
paper is to calculate the radiative contribution to mh coming from the KK partners of particles
and superparticles. We now proceed through the following steps.
(1) Let us first recall that the N = 2 supersymmetry prohibits any bulk Yukawa interaction involving three chiral multiplets.
interaction is considered to be localised at a

The Yukawa

brane, like (ht5 /3/2 ) d 4 x dy (y) d 2 (H2 QT + h.c.), where the residual supersymmetry is that of N = 1, ht5 is a dimensionless Yukawa coupling in 5d and the cutoff scale.
This localisation has a consequence in the counting of KK degrees of freedom that contribute to the Higgs mass radiative correction. The delta function ensures that those fields
which accompany the sine function after Fourier decomposition do not sense the Yukawa
interaction.
(2) As in the case of 4d (zero mode) supersymmetry, here too the dominant effect arises solely
from the third generation quark superfields, only that now we have to include the contributions from their KK towers. We shall continue to ignore contributions from the gauge
interactions or those from the first two quark families, as they are not numerically significant.
We might as well formulate a scheme in which the first two generation of matter superfields
are brane-localised and only the third generation superfields access the bulk.5 Keeping this
in mind, we displayed the Fourier decomposition of only the third generation superfields in
Eq. (22).
5 If all the three matter generations are bulk fields, then the theory become non-perturbative too soon, unless 1/R >
5.0 1010 GeV [26]. If only one generation accesses the bulk and the other two are confined to a brane, then the validity
of the theory extends further, allowing even a perturbative gauge coupling unification, we checked, around E 40/R.
123
(3) In our scheme MS and R are independent parameters, although we take them to be of the
same order.6 Towards the end of Section 4, we briefly remark on the numerical implications
of any possible connection between MS and R.
(4) The KK equivalent of Eq. (11), which captures the KK contribution arising from the top
quark chiral hypermultiplet, is then given by
Vtn =

m2n
m2n

m2t n
3
3
3
3
t1
t2
4
4
4
+
m
2m
, (24)
m
ln
ln
ln
n
n
t
t2n
2
2
2
32 2 t1
Q2
Q2
Q2
where the field dependent KK masses are given by m2tn = m2t + n2 /R 2 , m2tn = m2t + n2 /R 2 ,
1
and m2t n = m2t + n2 /R 2 . The field dependence is hidden inside the zero mode masses, as illustrated in Eqs. (12), (13) and (14). The corresponding contribution triggered by the bottom
quark hypermultiplet, Vbn , can be written mutatis mutandis.
(5) We now calculate the KK loop contribution to the neutral scalar mass matrix. The procedure
will be exactly the same as that followed for the 4d MSSM scenario in the previous section.
Since we are going to treat the radiatively corrected physical mA as an input parameter, we
concentrate only on the CP-even mass matrix. We first take another look at the expressions
of the different ij , assembled in Eq. (20), calculated in the context of the 4d MSSM. The
prefactors like m4t or m4b originated by the action of double differentiation on the field dependent squark or quark masses. Recall that the squark and quark masses are (quadratically)
separated by the soft supersymmetry breaking mass-squares which are not field dependent.
So, irrespective of whether we double-differentiate the squark or quark masses we get either
the top or bottom quark Yukawa coupling.7 In the same way, the KK mass-squares are separated from the zero mode mass-squares by a field independent quantity n2 /R 2 . Therefore,
the expressions for (ij )n , the radiative corrections from the nth KK level, continue to have
the zero mode quark masses m4t or m4b as prefactors, but now the arguments of the other
functions contain the corresponding KK masses.
We are now all set to write down the expressions for different (ij )n for n = 0. They are given
by
t11
n
(At + cot )
2
m2tn m2tn
sin
m4t
2

g m2tn , m2tn ,
1

m2n
t n

m4 (At + cot )
At (At + cot ) 2
t1
2
12 = 2t
,
+
g
m
,
m
ln
t1n
t2n
m2tn
m2tn m2tn
sin m2tn m2tn
1
6 This is in contrast to other higher-dimensional supersymmetric scenarios in which both the superpartner masses and
the scale of electroweak symmetry breaking arising from quantum loops are set by 1/R, where R is the distance between
the brane at which top quark Yukawa coupling is localised and the brane where supersymmetry is broken [35]. Higher
order finiteness of the Higgs mass, where supersymmetry is broken in the bulk by ScherkSchwarz boundary conditions
[36], has been discussed in [37].
7 This also indicates that by fixing the first and second generation matter superfields at the brane we have not made any
numerically serious compromise as otherwise their contributions would have been adequately suppressed on account of
their small Yukawa couplings.
124
t22
n
2
m2n m2n
2At (At + cot ) mt1n
t1 t2
+
ln
ln
m4t n
m2tn m2tn
m2tn
sin2
1
2
2

At (At + cot ) 2 2
+
g mtn , m2tn ,
1
2
m2tn m2tn
m4t
n
b11

m4b
ln
cos2
m2 n m2 n
m2 n
2Ab (Ab + tan )
b
=
+
ln 21
4
2
2
mbn
m n m n
m n
b1
b2
b2

2

Ab (Ab + tan )
+
g m2b n , m2b n ,
2
2
1
2
m n m n
b1
b1
b2
b2

m2n
b n

m4b (Ab + tan )
Ab (Ab + tan ) 2
b1
2
12 =
,
+
g
m
,
m
ln
b1n
b2n
cos2 m2 n m2 n
m2 n
m2 n m2 n
b1
b2
b2
b1
b2
2

4

b n
mb
(Ab + tan )
g m2b n , m2b n .
22 =
cos2
1
2
m2 n m2 n
b1
(25)
b2
(t )n
and (b )n matrices to the one-loop corrected (from zero modes

Now we have to add the
only) mass matrix in Eq. (19), sum over n, and then diagonalise to obtain the eigenvalues m2h and
m2H . The KK radiative corrections decouple in powers of (R 2 /n2 ). To provide intuition to the
expressions in Eq. (25), we display below the approximate formulae for (t )n in leading powers
of (R 2 /n2 ):
4
2
t n
1 R4
mt
(At + cot ) ,
11 =
4
2
6 n sin
2 4
t n
R
mt
12 =
(At + cot ),
n2 sin2
2 4
t n

R
mt 2
22 =
(26)
mt + m2t 2m2t + 2At (At + cot ) .
2
2
1
2
n sin
Similar expressions for (b )n can be written, with appropriate replacements like mt mb ,
cot tan , etc. So, in the absence of any leftright scalar mixing, the KK contribution to
m2h is controlled by R 2 (m2t m2t )/n2 and its higher powers.
3.1. Six-dimensional scenario
For the 6d scenario we follow the compactification on a chiral square, as done in [30], which
admits zero mode chiral fermions. The two extra spatial coordinates (y1 , y2 ) are compactified
on a square of side length L, such that 0 < y 1 , y 2 < R( L). The boundary condition is the
identification of the two pairs of adjacent sides of the squares such that the values of a field at
two identified points differ by a phase ( ). Nontrivial solutions exist when takes four discrete
values (n/2) for n = 0, 1, 2, 3 and the zero modes appear when n = 0. What matters to our
calculation in this paper is the structure of the KK masses, a generic pattern of which is given by
m2j,k = m20 +
j 2 + k2
,
R2
(27)
125
Table 1
6d scenario mass spectrum in (1/R) units, neglecting the zero mode mass
(j, k)
(1, 0)
mj,k
(1, 1)
(2, 0)
2
(2, 1) or (1, 2)
(2, 2)
2 2
(3, 0)
3
(3, 1) or (1, 3)
10
(3, 2) or (2, 3)
13
(4, 0)
4
where j, k are integers such that j 0 and k 0. We display in Table 1 the KK mass spectrum
(neglecting the zero mode mass m0 for simplicity of presentation while in the actual calculation
we do keep it). The formalism we developed for 5d will simply go through for 6d. More concretely, the structure of Eqs. (24) and (25) would remain the same in 6d, only that one should
now read n (j, k). The numerical impact in the two cases obviously differ, as we shall witness
in the next section.8
4. Results
In this section we explore the consequences of the extra-dimensional contributions to the
Higgs mass encoded in the exact one-loop expressions in Eq. (25). But to start with, to get a
feel for the numerical impact of the extra dimensions, consider the scenario pared down to its
bare minimum by assuming that left-right scalar mixing ingredients are vanishing, i.e., = At =
Ab = 0. This leads to two degenerate stop squarks: m2t = MS2 + m2t . Then, for a moderate tan ,

MS2
3m4t
m2h (n = 0)
ln
1
+
;
2 2 v 2
m2t
3m4t (MS R)2
.
(28)
2 2 v 2 n2
Indeed, non-zero trilinear and terms would complicate the expressions, yet Eq. (28) provides
a good intuitive feel for our results displayed through the different plots. The expected decoupling of extra-dimensional effects in the 1/R limit is transparent in Eq. (28), leaving the
logarithmic dependence on the supersymmetry scale, MS .
As stressed already, the primary emphasis in this work is to examine the effect of extra dimensions on the upper bound of mh . In 4d supersymmetry it is usual to choose the pseudoscalar
Higgs mass, mA , as a free parameter and exhibit mh as its function. This has been done for the
extra-dimensional MSSM models in Figs. 1 (5d case) and 2 (6d case). Let us discuss them in
turn.
In these and the subsequent figures, the parameters involved are chosen as follows:
(a) mQ = mU = mD MS , which is a common soft supersymmetry breaking mass. Several
values of MS have been chosen in the figures to depict its impact.
(b) The trilinear scalar couplings At and Ab are varied in the range [0.81.2]MS . This results
in bands in the figures. We have found that the results are not particularly sensitive to and we
hold it fixed at 200 GeV. Also, sign flips in the trilinear couplings do not change the results.
(c) The stop and sbottom (zero mode) mass eigenvalues are calculated from the diagonalisation
of matrices in Eqs. (13) and (14) after setting the Higgs fields to their VEVs. For a chosen value
of tan and MS , those eigenvalues will vary in a range in accord with the variation of At and Ab
stated above.
m2h (n = 0)
8 Admittedly, the 6d sum is logarithmically sensitive to the cutoff. The low-lying KK states we include reflect the
dominant contribution to the Higgs mass shift. We thank Anindya Datta for raising the 6d divergence issue.
126
Fig. 1. The variation of mh with mA in the 5d MSSM for different choices of the supersymmetry breaking scale (MS )
and the compactification radius (R). The width of each band corresponds to the variation of At and Ab in the range
(0.81.2)MS (see text).
Fig. 2. As in Fig. 1 but for 6d MSSM.
(d) Since we are interested in probing the upper limit of the lightest Higgs, we maximize its tree
level mass as much as possible. For displaying our results we have fixed tan = 10, a moderate
value for which the tree level mh is almost close to MZ .
In Fig. 1 we have displayed the result in the mh mA plane for only one extra dimension. The
dependence of mh on mA in the MSSM case is mimicked in the extra-dimensional case and mh
settles at its upper limit for mA greater than about 150 GeV. In the left panel, MS has been fixed
at 500 GeV. As anticipated, larger the value of 1/R smaller is the extra-dimensional impact. The
4d MSSM case corresponds to 1/R . The width of each band reflects the variation of the
trilinear parameters in the zone mentioned above. For the chosen supersymmetry parameters, the
127
Fig. 3. The dependence of mh on tan (zoomed for the low values) in 5d (left panel) and 6d (right panel) MSSM. The
width of each band corresponds to the variation of At and Ab in the range (0.81.2)MS .
maximum value of mh is a little below 125 GeV for the 4d MSSM case while for the extradimensional situation it is enhanced to above 135 (130) GeV for 1/R = 600 GeV (1 TeV). In
the right panel, the dependence on MS is exhibited holding 1/R at 1 TeV. Clearly, a larger MS
results in bigger radiative correctionsrecall Eq. (28)both from the zero mode as well as from
the KK modes.
Fig. 2 is a 6d version of Fig. 1. While the pure 4d MSSM band remains the same, the KK
radiative effects are larger now due to the denser KK spectrum in the 6d case, specified by two
sets of integers j and k, as shown in Table 1. Quantitatively, for an 1/R of 600 GeV (1 TeV), mh
can now be as heavy as 195 (155) GeV, to be compared with 125 GeV in 4d MSSM for these
parameter values.
As mentioned earlier, the current lower bound on mh of 114.5 GeV excludes low values
of tan in the 4d MSSM. It is expected that in the extra-dimensional scenarios some of this
excluded range of tan will make it into the allowed zone. In Fig. 3, we have shown the variation
of mh with respect to tan (for low values) to illustrate this effect. For the 5d case (left panel),
1/R of even 1.2 TeV eases the tension somewhat while for 1/R of 600 GeV the effect is very
prominent. For 6d (right panel), the extra-dimensional contributions are further enhanced and the
restriction on tan is essentially entirely lifted. We should recall that tan enters in the Higgs
couplings to other particles and so the above result has significant bearing on collider searches
of supersymmetry.
So far, we have exhibited results for a few choices of the compactification scale, 1/R. Fig. 4
demonstrates how the KK-induced radiative correction depends on 1/R for the 5d (left panel) and
6d (right panel) scenarios. If the Higgs boson is detected at the LHC then using these figures one
can gain a handle on 1/R dependent on the supersymmetry parameters like MS . The decoupling
behaviour as 1/R increases is in agreement with expectation.
We have also studied, in passing, the possibility that the soft supersymmetry breaking scale
arises from compactification (e.g. through the ScherkSchwarz mechanism [36]). Let us suppose
MS = C/R, where C is an order one dimensionless constant. Since we are interested in weak
scale supersymmetry breaking, we keep 1/R around a few hundred GeV to a TeV. In this region,
the radiative correction roughly depends on MS and R only through their product ( C), and for
128
Fig. 4. The dependence of mh on 1/R for different choices of MS for 5d (left panel) and 6d (right panel) cases. The ratio
6 between At (= Ab ) and MS maximises the trilinear contribution (see, Drees, Godbole, Roy in [1]).
a choice of C [0.52.0], the upper limit on the lightest Higgs mass turns out to be in the range
mh (150230) GeV (5d) and (200450) GeV (6d).
It may bear mentioning again that in these calculations we have retained the loop contributions
from the t and b quarks only. The other quarks and gauge bosons make negligible impact. Also,
we have dealt only with real MSSM parameters and limited our studies up to one-loop KK contributions. We have not, therefore, included either the two-loop improvements of the 4d MSSM
calculations or the numerical effects of the phases associated with complex MSSM parameters
in our discussions (for a recent survey, see [38]).
5. Conclusions
One of the virtues for which supersymmetry stands out as a leading candidate of physics beyond the SM is that it sets an upper bound on the Higgs mass. The lightest neutral Higgs mass,
mh , could at most be MZ at the tree level, but is pushed further obeying a definite relation, obtained from quantum corrections, involving mh , mt and the stop squark mass, mt. The sensitivity
of this correction to mt is only logarithmic. Consequently, a firm prediction results, namely, that
mh 135 GeV in MSSM for mt O(1 TeV). This is regarded as a critical test of supersymmetry and is naturally high on the agenda of the upcoming LHC experiments. In this paper, we
have probed how much this upper limit could be relaxed, should the MSSM be embedded in one
(S 1 /Z2 ) or two (T 2 /Z4 ) extra dimensions. We highlight our main findings:
(1) The KK towers of the top quark and stop squarks provide a positive contribution to m2h
raising it by several tens of GeV. If we ignore leftright scalar mixing and assume moderate tan (510), then using Eq. (28) and summing over all the KK modes, we obtain
m2h (KK) (60 GeV)2 (MS R)2 . This is a 5d result. Including the leftright scalar mixings, i.e., non-zero and trilinear parameters, somewhat enhances the magnitude of the
correction (see Fig. 1). As in the case of 4d MSSM, here too the size of the correction is
controlled by the large top Yukawa coupling.
(2) If we consider a 6d theory with two extra dimensions compatcified on a chiral square, whose
motivations have been mentioned earlier, the correction gets sizably enhanced (see Fig. 2),
129
compared to 5d, due to a denser packing of KK states, which are now fixed by two independent KK numbers.
(3) Non-observation of a Higgs boson weighing below 114.5 GeV disfavours low tan in 4d
MSSM. Some part of this region can be revived by extra-dimensional embedding (see Fig. 3).
(4) The 4d MSSM relationship between the lightest neutral Higgs mass and the stop squark mass
is extremely profound in the sense that its specific form does not depend on the supersymmetry breaking mechanism. If supersymmetry is embedded in extra dimension(s) and, with
some cooperation from Nature, the KK states happen to be light enough to mark their imprints on the LHC data recorder, then the relationship between the stop mass and the Higgs
mass alters in a numerically significant way (see Fig. 4).
Acknowledgements
G.B. acknowledges hospitality at the Abdus Salam ICTP, Trieste, during a part of the work,
and thanks G. Senjanovic for his critical remarks. G.B. also recognizes the contribution of the
ever refreshing Osmizzas towards this paper and thanks Angelo for accompanying him there
during his stay at Trieste.
References
[1] See e.g., H.E. Haber, G.L. Kane, The search for supersymmetry: Probing physics beyond the Standard Model, Phys.
Rep. 117 (1985) 75;
H.P. Nilles, Supersymmetry, supergravity and particle physics, Phys. Rep. 110 (1984) 1;
M. Drees, R. Godbole, P. Roy, Theory and Phenomenology of Sparticles: An Account of Four-Dimensional N = 1
Supersymmetry in High Energy Physics, World Scientific, 2004;
H. Baer, X. Tata, Weak Scale Supersymmetry: From Superfields to Scattering Events, Cambridge Univ. Press,
Cambridge, UK, 2006;
A. Djouadi, The anatomy of electro-weak symmetry breaking. II: The Higgs bosons in the minimal supersymmetric
model, hep-ph/0503173.
[2] For a recent review, see, J. Ellis, G. Ridolfi, F. Zwirner, Higgs boson properties in the standard model and its
supersymmetric extensions, hep-ph/0702114. Search of the Higgs Particle, C. R. Acad. Sci., France, in press.
[3] J.R. Ellis, G. Ridolfi, F. Zwirner, Phys. Lett. B 257 (1991) 83;
J.R. Ellis, G. Ridolfi, F. Zwirner, Phys. Lett. B 262 (1991) 477;
Y. Okada, M. Yamaguchi, T. Yanagida, Prog. Theor. Phys. 85 (1991) 1;
H.E. Haber, R. Hempfling, Phys. Rev. Lett. 66 (1991) 1815;
A. Brignole, Phys. Lett. B 281 (1992) 284;
M.S. Berger, Phys. Rev. D 41 (1990) 225;
J.F. Gunion, A. Turski, Phys. Rev. D 39 (1989) 2701.
[4] M. Carena, J.R. Espinosa, M. Quiros, C.E.M. Wagner, Phys. Lett. B 355 (1995) 209, hep-ph/9504316;
M. Carena, M. Quiros, C.E.M. Wagner, Nucl. Phys. B 461 (1996) 407, hep-ph/9508343;
H.E. Haber, R. Hempfling, A.H. Hoang, Z. Phys. C 75 (1997) 539, hep-ph/9609331;
S. Heinemeyer, W. Hollik, G. Weiglein, Eur. Phys. J. C 9 (1999) 343, hep-ph/9812472.
[5] R. Barate, et al., LEP Working Group for Higgs boson searches, Search for the standard model Higgs boson at LEP,
Phys. Lett. B 565 (2003) 61, hep-ex/0306033.
[6] S. Schael, et al., ALEPH Collaboration, Search for neutral MSSM Higgs bosons at LEP, Eur. Phys. J. C 47 (2006)
547, hep-ex/0602042.
[7] J. Alcaraz, et al., LEP Collaboration, A combination of preliminary electroweak measurements and constraints on
the standard model, hep-ex/0612034.
[8] See e.g., U. Ellwanger, C. Hugonie, Eur. Phys. J. C 25 (2002) 297, hep-ph/9909260, and the references cited therein.
[9] M. Drees, Int. J. Mod. Phys. A 4 (1989) 3635.
[10] M. Bastero-Gil, C. Hugonie, S.F. King, D.P. Roy, S. Vempati, Phys. Lett. B 489 (2000) 359, hep-ph/0006198;
See also, M. Carena, C.E.M. Wagner, in: G.L. Kane (Ed.), Perspectives on Higgs Physics II, World Scientific,
Singapore, 1998, and references therein.
130
[11] P. Batra, A. Delgado, D.E. Kaplan, T.M.P. Tait, JHEP 0402 (2004) 043, hep-ph/0309149.
[12] K. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292;
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55, hep-ph/9803466.
[13] T. Appelquist, H.C. Cheng, B.A. Dobrescu, Phys. Rev. D 64 (2001) 035002, hep-ph/0012100.
[14] P. Nath, M. Yamaguchi, Phys. Rev. D 60 (1999) 116006, hep-ph/9903298.
[15] D. Chakraverty, K. Huitu, A. Kundu, Phys. Lett. B 558 (2003) 173, hep-ph/0212047.
[16] A.J. Buras, M. Spranger, A. Weiler, Nucl. Phys. B 660 (2003) 225, hep-ph/0212143;
A.J. Buras, A. Poschenrieder, M. Spranger, A. Weiler, Nucl. Phys. B 678 (2004) 455, hep-ph/0306158.
[17] K. Agashe, N.G. Deshpande, G.H. Wu, Phys. Lett. B 514 (2001) 309, hep-ph/0105084.
[18] J.F. Oliver, J. Papavassiliou, A. Santamaria, Phys. Rev. D 67 (2003) 056002, hep-ph/0212391.
[19] T. Appelquist, H.U. Yee, Phys. Rev. D 67 (2003) 055002, hep-ph/0211023.
[20] T.G. Rizzo, J.D. Wells, Phys. Rev. D 61 (2000) 016007, hep-ph/9906234;
A. Strumia, Phys. Lett. B 466 (1999) 107, hep-ph/9906266;
C.D. Carone, Phys. Rev. D 61 (2000) 015008, hep-ph/9907362.
[21] T. Rizzo, Phys. Rev. D 64 (2001) 095010, hep-ph/0106336;
C. Macesanu, C.D. McMullen, S. Nandi, Phys. Rev. D 66 (2002) 015009, hep-ph/0201300;
C. Macesanu, C.D. McMullen, S. Nandi, Phys. Lett. B 546 (2002) 253, hep-ph/0207269;
H.-C. Cheng, Int. J. Mod. Phys. A 18 (2003) 2779, hep-ph/0206035;
A. Muck, A. Pilaftsis, R. Rckl, Nucl. Phys. B 687 (2004) 55, hep-ph/0312186.
[22] U. Haisch, A. Weiler, Phys. Rev. D 76 (2007) 034014, hep-ph/0703064.
[23] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[24] N. Arkani-Hamed, M. Schmaltz, Phys. Rev. D 61 (2000) 033005, hep-ph/9903417.
[25] G. Servant, T.M.P. Tait, Nucl. Phys. B 650 (2003) 391, hep-ph/0206071.
[26] G. Bhattacharyya, A. Datta, S.K. Majee, A. Raychaudhuri, Nucl. Phys. B 760 (2007) 117, hep-ph/0608208.
[27] A. Deandrea, J. Welzel, P. Hosteins, M. Oertel, hep-ph/0611172.
[28] P. Dey, G. Bhattacharyya, Phys. Rev. D 70 (2004) 116012, hep-ph/0407314;
P. Dey, G. Bhattacharyya, Phys. Rev. D 69 (2004) 076009, hep-ph/0309110.
[29] E. Brubaker, et al., Tevatron Electroweak Working Group, hep-ex/0608032.
[30] B.A. Dobrescu, E. Ponton, JHEP 0403 (2004) 071, hep-th/0401032;
G. Burdman, B.A. Dobrescu, E. Ponton, JHEP 0602 (2006) 033, hep-ph/0506334.
[31] B.A. Dobrescu, E. Poppitz, Phys. Rev. Lett. 87 (2001) 031801, hep-ph/0102010;
N. Borghini, Y. Gouverneur, M.H.G. Tytgat, Phys. Rev. D 65 (2002) 025017, hep-ph/0108094;
M. Fabbrichesi, R. Percacci, M. Piai, M. Serone, Phys. Rev. D 66 (2002) 105028, hep-th/0207013.
[32] T. Appelquist, B.A. Dobrescu, E. Ponton, H.U. Yee, Phys. Rev. Lett. 87 (2001) 181802, hep-ph/0107056;
R.N. Mohapatra, A. Perez-Lorenzana, Phys. Rev. D 67 (2003) 075015, hep-ph/0212254.
[33] T. Appelquist, B.A. Dobrescu, E. Ponton, H.U. Yee, Phys. Rev. D 65 (2002) 105019, hep-ph/0201131. See also,
Mohapatra and Perez-Lorenzana in [32].
[34] G. Burdman, B.A. Dobrescu, E. Ponton, Phys. Rev. D 74 (2006) 075008, hep-ph/0601186.
[35] N. Arkani-Hamed, L.J. Hall, Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81, hep-ph/0102090;
See also, R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.
[36] J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61;
J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60.
[37] A. Delgado, M. Quiros, Nucl. Phys. B 607 (2001) 99, hep-ph/0103058;
A. Delgado, G. von Gersdorff, M. Quiros, Nucl. Phys. B 613 (2001) 49, hep-ph/0107233;
R. Barbieri, L.J. Hall, Y. Nomura, Nucl. Phys. B 624 (2002) 63, hep-th/0107004.
[38] S. Heinemeyer, W. Hollik, H. Rzehak, G. Weiglein, arXiv: 0705.0746 [hep-ph].

Instanton transition in thermal and moduli deformed

de Sitter cosmology
Costas Kounnas a , Herv Partouche b,
a Laboratoire de Physique Thorique, Ecole Normale Suprieure, 1 24 rue Lhomond, F-75231 Paris cedex 05, France
b Centre de Physique Thorique, Ecole Polytechnique, 2 F-91128 Palaiseau cedex, France
Received 26 August 2007; accepted 1 October 2007

Abstract
We consider the de Sitter cosmology deformed by the presence of a thermal bath of radiation and/or
time-dependent moduli fields. Depending on the parameters, either a first or second-order phase transition
can occur. In the first case, an instanton allows a double analytic continuation. It induces a probability to
enter the inflationary evolution by tunnel effect from another cosmological solution. The latter starts with
a big bang and, in the case the transition does not occur, ends with a big crunch. A temperature duality
exchanges the two cosmological branches. In the limit where the pure de Sitter universe is recovered, the
tunnel effect reduces to a creation from nothing, due to the vanishing of the big bang branch. However,
the latter may be viable in some range of the deformation parameter. In the second case, there is a smooth
evolution from a big bang to the inflationary phase.
PACS: 11.10.Wx
Keywords: Cosmology; Inflation; Thermal field theory; Instanton
Research partially supported by the EU (under the contracts MRTN-CT-2004-005104, MRTN-CT-2004-512194,

MRTN-CT-2004-503369, MEXT-CT-2003-509661), INTAS grant 03-51-6346, CNRS PICS 2530 and 3059, and ANR
(CNRS-USAR) contract 05-BLAN-0079-01 (01/12/05).
E-mail addresses: costas.kounnas@lpt.ens.fr (C. Kounnas), herve.partouche@cpht.polytechnique.fr (H. Partouche).
1 Unit mixte du CNRS et de lEcole Normale Suprieure associe lUniversit Pierre et Marie Curie (Paris 6),
UMR 8549.
2 Unit mixte du CNRS et de lEcole Polytechnique, UMR 7644.
doi:10.1016/j.nuclphysb.2007.10.008
132
C. Kounnas, H. Partouche / Nuclear Physics B 793 (2008) 131159
1. Introduction
The recent astrophysical observations indicate that our universe is in a phase of classical
expansion, with a small positive cosmological term, , representing 6070% of the total energy
density. However, at early time, quantum corrections to this trajectory are expected to be of
first significance. For instance, drastic non-perturbative effects could occur as topology changes,
as can be seen in the context of field theory in semi classical approximation. As an example,
a topology change is provided by the instability of the five-dimensional KaluzaKlein (KK)
spacetime [1]. In that case, one observes the nucleation of a finite size bubble of nothing
that is then growing up at the speed of light. The transition is described by tunnel effect in
terms of an instanton configuration allowing a double analytic continuation. This means that
at two different Euclidean times i and f , analytic continuations to real times = i + it and
= f + it are allowed. To be specific, the configuration admits a time independent asymptotic
behavior allowing a continuation at i = + to the KK universe, while another continuation at
f = 0 describes the evolution of a bubble.
Actually, the KK universe is suffering from a first-order phase transition where bubbles appear
instantaneously, grow and coalesce, so that the space evaporates into nothing (after an infinite
time, since the volume is itself infinite). In some sense, some reversed ideas can be invoked
in another example of topology change, namely in Vilenkins scenario of the de Sitter space
creation from nothing [2,3]. In that case, a finite radius S 3 space appears instantaneously by
tunnel effect from a spacetime state that amounts to the empty set. This S 3 bubble is then
following a de Sitter growing up evolution. The transition involves an instanton, whose shape
is an S 4 hemisphere. An analytic continuation on its boundary amounts to gluing a de Sitter
universe, while the fact that the instanton is compact with no other boundary to analytically
continue allows an interpretation in terms of a transition from an empty set.
Concerning the birth of the de Sitter space, the instanton method provides an estimate of the
transition probability equivalent to the one derived from the HartleHawking wave function [4]
approach. The transition amplitude and wave function being proportional to eSE , where SE
is the Euclidean action, the probability of the event is
p e2SE .
(1.1)
A selection principle of the cosmological constant based on the extremum of p leads to a favored value 0+ [5]. However, a present too small cosmological term cannot account for the
60% to 70% of the total energy density, which is necessary to explain the recent dark energy data.
To remedy to this problem, it as been stressed [6,7] that since the de Sitter space has an horizon,
it can be associated a Hawking temperature. Thus, quantum fluctuations of the metric (or any
massless mode introduced in the model) induce a space filling thermal bath. The latter implies
an additional radiation term in the action3 and thus a back reaction on the metric background
that deforms the de Sitter solution (see also [9]). In that case, the modified probability transition
derived from the Euclidean action is maximal for a non-vanishing finite value of . This computation has also been addressed by the authors of [10] and refined in [11], who considered the
Wheelerde Witt equation in the WKB approximation. They found that the tunnel effect is not
connecting the deformed de Sitter expansion to nothing, but to what they called a thermally
excited era. However, the latter is -dependent. This point makes the difference with the pure
3 See [8] for a cosmological scenario based on a cascade of transitions between vacuum and radiation energy.
133
de Sitter space created from a -independent state, namely the empty set, so that extremizing the
transition amplitude with respect to was making sense. In the case of the deformed solution,
applying this selection rule for is thus questionable.
Our first aim is to reconsider the thermally deformed de Sitter spacetime. The previous analysis that involves an instanton transition occurs when the radiation energy density is below some
critical value. This case corresponds to a first-order phase transition. A double analytic continuation on the Euclidean solution implies a probability to connect two cosmological behaviors in
real time. The first one is the de Sitter like evolution, while the second one is an era that begins
with a big bang and ends with a big crunch, when the transition does not occur. If instead the
radiation energy density is above the critical value, the phase transition becomes second order. It
describes a smooth evolution from a big bang to a de Sitter like behavior. In all these solutions,
the temperature is formally infinite when the radius of the universe vanishes. This means that
these evolutions should be trusted as soon as the temperature passes below some upper bound
such as the string or Hagedorn temperature TH of an underlying fundamental string theory.
We then generalize the model by including time-dependent moduli-like fields that imply a
further deformation of the de Sitter solution. Since, these deformations respect the Robertson
Walker (RW) isometries, they can be expressed in terms of the scale factor a of the universe. The
Hubble parameter is taking the form
2
3k CR CM
a
= 3 2 + 4 + 6 ,
a
a
a
a
(1.2)
where the constant and 1/a 2 terms are associated to the cosmological constant and uniform space
curvature. The 1/a 4 monomial accounts for the effective radiation term discussed before, while
the on shell moduli contribution reaches the 1/a 6 contribution. What we find is that the moduli
deformation can be fully absorbed in a redefinition of time. We then describe different cosmological scenarios that depend on CR and CM . These parameters can be expressed in terms of the
cosmological constant, the number of effective massless degrees of freedom, the temperature
at the transition and the slope of the moduli dependence. If this two parameters model is interesting by itself from a formal point of view, its study is also motivated by the fact that it describes
more complicated systems once they are considered on shell. For instance, cases involving a
scalar field with a non-trivial potential and coupled to moduli with non-trivial kinetic terms are
treated in [12].
2. Effective cosmological models in minisuperspace
We focus on models where space is 3-dimensional4 and closed. The number of degrees of
freedom is important, and actually infinite in the context of string or M-theory. However, a subset
of them to be considered in an effective field theory is determined by the ultraviolet cut off set
by the fundamental theory. Some of the simplest class of models that can be constructed are
considering the so-called minisuperspace (MSS) approximation [4,13]. They involve isotropic
and homogeneous spaces only, so that the only gravitational degree of freedom is the scale factor
4 By this, we mean that if the fundamental theory is for instance string (or M-)theory, there are compact dimensions
whose sizes are dynamically constrained to remain of order ls , the string length.
134
of the universe, to which some matter fields can be coupled to. The spacetime metric is thus

ds 2 = 2 N 2 (t) dt 2 + a 2 (t) d32 ,
where 2 =
2
,
3Mp2
(2.1)
Mp is the Planck mass and a(t) is the radius of a 3-sphere. N is a gauge-dependent function
that can be arbitrarily chosen by a redefinition of time.5 (In our conventions, N and t are dimensionless.) In order to include in the effective action the quantum fluctuations of the full metric
and matter degrees of freedom of the theory, we switch on a thermal bath which is created simultaneously with the spacetime. The thermal bath involves the degrees of freedom below the
temperature scale T and is consisting in an extra radiation term in the effective MSS action.
This amounts to take into account the back reaction of the thermal bath on the space-metric, i.e.
deform the de Sitter solution [6,7].
The starting point is a gravity action coupled to some matter fields, whose Lagrangian density
is Lm :

Mp2
4
d x g(R 2) + d 4 x gLm .
S Sg + S m =
(2.2)
16
Assuming the metric (2.1) in the gravity part and freezing all other degrees of freedom, we obtain
the MSS-action,
1
Sg =
2
tf
ti

1
1 a 2 a tf
1 a 2
dt N a 3 2
+ 2 +
,
2 N ti
N a
a
where =
2
.
3
(2.3)
In this expression, a and N have been chosen positive, without loss of generality. Furthermore,
the boundary terms arise from the integration by parts of the second derivative of a in the scalar
curvature R.
The non-vanishing components of the energymomentum tensor which are induced by the
matter fields (and which are invariant under the isometries of the RW spacetime), can be expressed in terms of the energy density and pressure P :
Ttt = 2 N 2 ,
Tij = P 2 a 2 g ij ,
(2.4)
where d32 = g ij dx i dx j is the line element on S 3 . These quantities appear in the equations of
motion of N and a, i.e. the timetime and spacespace Einstein equations:
2
a
N2
3
(2.5)
+ 3 2 = 12 2 4 N 2 + 3N 2 ,
a
a
2
a
N2
a N
a
2 + 2
2 = 12 2 4 N 2 P 3N 2 .
a
aN
a
a
(2.6)
Since N can be fixed to any arbitrary function by a reparameterization of t in the metric (2.1),
this differential system of two unknown functions must be degenerate so that we only have to
5 We shall use in the following the time variable t either when the function N (t) remains unspecified, or when it is
fixed to N (t) 1. We shall instead use another variable, t, when we choose to fix N (t ) = 1/(1 + 1 /a 2 (t )), where
is a constant.
135
solve any of the two equations,6 e.g. the Friedman one (2.5). Let us consider some cosmological
solutions associated to various matter contents and where phase transitions can occur.
3. Pure de Sitter space
With neither matter nor thermal bath, we have = P = 0. Choosing N (t) 1, Eq. (2.5) is
easily solved,
cosh( t)
a(t) =
(3.1)
.
This corresponds to an usual de Sitter evolution where a phase of contraction from an infinitely
past time is followed by an expansion phase when the radius a has reached its minimum.
However, a different cosmological scenario can give rise to an identical de Sitter expansion.
Actually, one can consider the real time evolution for t 0 as the result of a tunnel effect, since an
analytic continuation at ti = if = 0 is allowed. The instanton solution in question is obtained
by substituting t = i in (3.1),7
| cos( )|
aE ( ) =
(3.2)
,
where we find convenient to insist on a conventional positivity

of aE for future purpose. The
Euclidean metric takes the form dsE2 = ( 2 /)( d 2 + cos2 ( ) d32 ), so that the instanton
is a 4-sphere of radius / , if [/2, /2].
Vilenkin, assuming that the range of the Euclidean time [i , f ] = [/2 , 0], considers
this instantonic hemisphere as giving rise to a transition amplitude from nothing (a state of the
spacetime where it is the empty set) to an S 3 space with de Sitter cosmological evolution [2,3].
As we shall see later, this interpretation is very natural in the deformed de Sitter solution which
is induced by the thermal bath. The Euclidean action,
1
SEg =
2
f
d
i
3
NE a E
1
2
NE
a E
aE
2

2 f
a E
1
1 aE
2 + +
,
2 NE i
aE
(3.3)
evaluated on shell, has vanishing boundary terms and takes the value
1
,
3
so that the probability transition
SEg =
(3.4)
p e2/(3)
(3.5)
becomes maximalfor a vanishing

cosmological constant. Considering the second hemisphere of
the S 4 -instanton, i = 0 and f = /2, we note that the previous probability also evaluates
the chances for a contracting de Sitter space to simply vanish instantaneously when it reaches its
minimal radius at t = 0.
6 This is not true for static solutions.
7 For any field f (t), we define in general its Euclidean counterpart f ( ) f (i ) and its derivative f ( ) =
E
E
i f(i ).
136
Another derivation
Before closing this section, we would like to mention an alternative method to solve the Friedman equation (2.5) that will appear useful to express the general deformed de Sitter solution once
the thermal and moduli effects are simultaneously present. Consider
N 2 (t ) =
1
,
1 + 1 /a 2 (t )
(3.6)
and choose = 1 so that the 1/a 2 term of Eq. (2.5) is cancelled out,8
2
a
= .
a
(3.7)
There are two solutions to this equation, whose corresponding metrics are

d t 2
2
2 t
+e
= 2
d32 ,
ds
(3.8)
1 1 e2 t
defined for t > log()/2 and t < log()/2 , respectively. These parameterizations do not
admit analytic continuations to Euclidean time, and actually correspond to half solutions, as
can be seen as follows. Integrating in each case
dt = N(t ) d t,
(3.9)
one can choose the time variable t (corresponding to the gauge N 1) such that
e2 t = cosh2 ( t),
(3.10)
2
to glue the two branches in t in a smooth way, t R (see Fig. 1). In that case, the metrics ds
are taking a RW form in the N 1 gauge (see Eq. (2.1)), with a defined in Eq. (3.1).
4. Thermally deformed de Sitter universe

The quantum fluctuations of the previous background metric have been shown to induce an
additional effective radiation term in the action. This implies a back reaction of the massless
gravitational degrees of freedom on the metric [6,7]. In other words, the solution of the new
equations of motion describes the evolution of a RW closed universe in presence of thermalized
radiation. An explicit derivation of the effective thermal contribution can be found in [6,7], while
the consistency of the semi-classical approach has given a similar result [10]. We would like here
to reconsider these effects by using thermodynamical and precise dynamical constraints to reveal
new features. The results can be applied to any model involving massless bosonic or fermionic
modes (beside the gravitational ones), as long as their vacuum expectation values do not show
up in the effective action.
For bosonic (or fermionic) fluctuating states of masses mb (or mf ) in thermal equilibrium at
temperature T , the general expressions of the energy density T and pressure PT are

T =
IB (mb ) +
IF (mf ),
PT =
IPB (mb ) +
IPF (mf ),
boson b
fermion f
boson b
fermion f
(4.1)
8 As signaled in footnote 5, remind that the dot derivative is here with respect to t, the time variable in the gauge (3.6).
137
Fig. 1. As a function of t, the pure de Sitter scale factor a has two branches. As a function of t , these branches are glued
together in a smooth way.
where

IB(F ) (mb(f ) ) =
dq
0
q 2 E(mb(f ) )
1
e T E(mb(f ) ) 1
B(F )
IP
1
=
3

dq
0
q 4 /E(mb(f ) )
1
e T E(mb(f ) ) 1
(4.2)
and E(mb(f ) ) = q 2 + m2b(f ) . If the system is consisting in nB(F ) massless degrees of freedom,
it is easy to see that the state equation

4
7
T = 3PT =
(4.3)
n B + nF T 4
15
8
is satisfied. These relations combined with the expression of the energymomentum conservation

a
(T + PT ) = 0
T + 3
(4.4)
a
is then implying that
aT = const
(4.5)
and that
T =
T2
1
,
2
4
16 a 4
(4.6)
where T2 /(16 2 4 ) is a positive integration constant. The factor of in its denominator is

chosen for later convenience.
To proceed, we solve the Einstein equations in the case where the energy density and pressure
are the thermal ones. This has been done for Eq. (2.6) in [6,7,10] and [14] in the N (t) 1 gauge.
However, we prefer to deal with Eq. (2.5) as in [9] but in a way similar to what we did at the end
138
of the last section for the pure de Sitter case, since this will be easier to generalize when we take
into account moduli fields. For this purpose, we choose again N (t ) with a form
1
,
1 + 1 /a 2 (t )
N 2 (t ) =
(4.7)
and look for so that the 1/a 4 thermal contribution is cancelled out. This implies that
2 + +
T2
= 0,
4
(4.8)
a condition that admits real solutions as long as the constraint T2 1 is satisfied.

The case T2 < 1
The Friedman equation (2.5) becomes
2
a
+1
+ 2 = ,
a
a
(4.9)
showing that the solution of our present problem can be related to the one of Section 3, the
de Sitter case with no thermal effect, up to a shift in the 1/a 2 curvature term. The hyperbolic
cosine solutions of Eq. (4.9) for the two roots of (4.8) are giving rise to the metrics

2
1 1 T2
d
t
2
2
2
= 2
+
(
t
)
d
cosh
ds
(4.10)
3 ,
u1
2
1
2
cosh ( t )
where
1
1 T2
u=
]0, 1],
1 + 1 T2
(4.11)
if T2 > 0. Since they simply correspond to different gauge choices of functions N , we already
know that they are equal up to a redefinition of time. This means that the potential singularity
2 when the denominator of N (t ) vanishes should be fake. In any case, let us make contact
of ds
with the more intuitive N 1 gauge by choosing a new time variable t satisfying Eq. (3.9). This
will in particular allow us to study in an explicit way various analytic continuations.
2 , one can fix the integration constant such that
For ds+
cosh2 ( t ) = u + (1 u) cosh2 ( t),

(4.12)
2 = 2 (dt 2 + a 2 (t(t)) d 2 ), with
so that the metric is taking the form ds+
3
a = N + cosh2 ( t),
(4.13)
where
N=
(1 T2 )1/4
,

1
1
1 .
2
1 T2
(4.14)
139
2 in the pure thermal case, the scale factor a has two

Fig. 2. When one is considering the time variable t involved in ds
branches. In terms of t , these branches are glued together in a smooth way. At the minimum, one has a = a+ given in
Eq. (4.18).
2 , one can have instead

For ds

cosh2 ( t ) = 1 + u1 1 cosh2 ( t),
(4.15)
1/2
) and t < arccosh(u1/2 )
which has the effect to glue the two branches t > arccosh(u
in a smooth way, with t R (see Fig. 2). This time variable, has the property to explicitly show
2 ds 2 .
that ds
+
This solution is a deformation of the de Sitter solution (3.1), i.e. a contraction phase followed
by an expansion one. As before, the latter can also arise by tunnel effect since the analytic continuation at ti = if = 0 is still valid. Let us consider for a moment the relevant instanton solution
obtained under the substitution t = i ,
aE = N + cos2 ( ),
(4.16)
for the range of Euclidean time /2 0. An important point is that at i = /2,

another example
aE is no longer vanishing as in the birth of the de Sitter space (Ref. [15] presents
of this phenomenon). Instead, a second analytic continuation is allowed, = /2 + i t,

and another cosmological solution
a = N sinh2 ( t),
(4.17)
satisfies (2.5) (in the N 1 gauge), for arcsinh t 0. Actually, this sinh-solution is
connected to the cosh-solution (4.13) via the instantonic one (4.16). The cosmological evolution
makes use of all above mention solutions.
The universe starts from a big-bang era at t 1/2 arcsinh , with an evolution following the sinh-solution till t = 0.
140
Fig. 3. Cosmological scenario of an expanding RW universe in presence of thermal and time-dependent moduli effects
below some critical value. A phase of expansion starts from a big bang and is connected to an inflationary deformed
de Sitter one by an instanton. This describes a first-order transition, where the scale factor jumps from a to a+ .
At t = 0, either (i) the space starts to contract till a big crunch occurs at t = 1/2 arcsinh ,
or (ii) there is a first-order phase transition that changes instantaneously its radius.
In case (ii), the cosmological evolution for later times, t 0, follows the inflationary coshsolution.
Actually, the very early part of these scenarios (and late time in case (i)) are not trustable,
assuming that the fundamental theory is string like. The degrees of freedom grow exponentially at
high temperature, and blow up at the Hagedorn temperature. This effect gives a cut in time before
which the sinh-solution is no more valid (and another cut for late time in case (i)). A stringy
approach is then necessary to cover such ultra hot eras of the universe.
At the transition, the scale factor is jumping,

1 1 2
1 + 1 2

T
T
a =
(4.18)
a+ =
2
2
(see Fig. 3). The transition probability at t = 0 (see Section 5.2) depends on and on the deformation parameter (or equivalently on T2 ). The maximal value of this probability is for finite
and depends on . This is fundamentally different from the pure de Sitter case where the maximal
probability is for = 0.
It is interesting to make a comparison between the pure de
Sitter case and the thermally
deformed one. When T2 0, the big bang occurs at ti T /2 , while we also have a
T /2 at t = 0 . This means that Vilenkins interpretation of the spontaneous creation of the

pure de Sitter space can be seen as the limit when the big bang sinh-cosmological branch has
shrunk to nothing (see Fig. 4). Actually, the instantonic jump a a+
becomes more drastic
as the thermal effects become more negligible, T 1, since a+ 1/ in this limit. On the
141
Fig. 4. Thermal instanton (in bold) compared to the pure de Sitter S 4 one. In
the limit wherethermal effects are negligible,
T 1, the instantonic jump of a is drastic, passing from a T /2 to a+ 1/ . When thermal effects are
getting closer and closer to their critical magnitude, T = 1, the amplitude of the transition from the big bang branch
to the inflationary deformed
de Sitter one is decreasing. At the very precise value T = 1, the instanton solution is a
constant, aE ( ) 1/ 2.
contrary, the transition from the big bang branch to the cosh-inflationary deformed de Sitter
one is smoother and smoother as we approach the critical magnitude of the thermal effects,
T 1. Actually the amplitude of the instanton solution decreases as we approach this value
of the parameter. When T = 1,
the instanton solution is constant, implying that its analytic
continuations are static, a(t) 1/ 2.
An estimate of the transition probability from the big bang branch to the deformed de Sitter
one is p e2SEeff , where
SEeff
1
=
2
f
i
2 f

a E
2
1
1 a E 2
1 aE
3
d NE aE
2 ++ T4 +
2
2 NE i
NE aE
aE
4aE
(4.19)
is the Euclidean effective action that is giving rise to the desired equations of motion. In this
expression,
SEeff is evaluated with the solution (4.16) and has vanishing boundary terms for
/2 0 (see Eq. (5.57)). However, onecould sum in principle the instanton contributions for the series
of ranges (2n + 1)/2 0, n N. The total probability would
take the form ptot = n0 c2n+1 e(2n+1)2SEeff , where c2n+1 are positive constants. Similarly,
the probability to pass from the contracting
of the deformed de Sitter space to the ex phase4nS
=
Eeff
panding one would take the form ptot
c
e
, for positive constants c2n (see Fig. 5
n0 2n
and [16]).
Introducing the temperature at the transition
It is useful to express the deformation parameter T2 (or N and in Eq. (4.14)) in terms of
more intuitive quantities, such as temperatures. First, observe that before and after the transition,
the quantity aT remains constant (see Eq. (4.5)). Since these evolutions are related by a double
analytic continuation, the constants must be equal. This allows us to express aT in terms of either
142
Fig. 5. The inflationary phase of the deformed de Sitter space for t 0 can arise from any contracting phase or big bang
phase for t 0, after a multi-instantonic transition.
the temperatures T+ at t = 0+ or T at t = 0 :

T 1
1 1 T2 .
T=
(4.20)
a 2
Plugging either of these relations in the expression of T in Eq. (4.3) and comparing the result
with Eq. (4.6) imposes a consistency condition to be satisfied by T2 . Its solution involves a mean
temperature Tm satisfying T+ Tm T ,

1/4
1
4
15
2
where Tm =
.
T =
(4.21)
7
2
2
6
2
2
2
4 (nB + 8 nF )
(Tm /T + T /Tm )
This implies that T+ T = Tm2 and that T2 is invariant under the duality transformation
T+
Tm T
(4.22)
.
Tm
T+ Tm
Actually, this operation exchanges the scale factor of the thermally deformed de Sitter evolution,

T 4 T+4
1
1+ m
cosh(2
t), t > 0,
a=
(4.23)
Tm4 + T+4
2
with the big bang/big crunch one,

T 4 T4
1
a=
1+ m
cosh(2 t),
4
Tm4 + T
2
t < 0.
(4.24)
143
In some sense, the instantonic transition between them is inversing the temperature, since
T (t) T for t 0, while T (t) T+ = Tm2 /T for t 0. If one wishes, one can also rewrite
the constants N and of Eq. (4.14) in terms of either T+ or T ,

1 (T /Tm )4
1
1
,
, =
N=
(4.25)
4
(T /Tm )4 1
1 + (T /Tm )
and note that the function N 2 appearing in Eq. (4.7) satisfies N 2 (t ) = 1/(1 T 2 (t )/T2 ) for
2 , respectively.
ds
Focussing on the -dependence of T2
We would like here to consider the dependence on of T2 . This is relevant, for instance,
if one wants to minimize the Euclidean action with respect to (keeping fixed the number of
degrees of freedom and temperatures T ). Doing so was considered as a selection principle for
the cosmological constant in [6,7,10,11,17]. The reason for this was based on the fact that this
action (see Eq. (5.57)) is controlling the probability transition between the two cosmological
evolutions. However, there are two hypothesis for this argument to be considered. The first one
is that the viable universes are the deformed de Sitter ones, i.e. that the big bang/big crunch
evolution is too hot or too short in time. However, this condition implies that T2 is small enough,
which in turn can be translated into a condition on itself (see Eq. (4.21)). The second is that the
probability to start the -dependent big bang evolution is almost constant as a function of .
Eq. (4.21) can be rewritten as
T2 =
where
=
4
+
4
=
,
2
(1 + + /)
(1 + / )2
(4.26)

4 6
7
nB + nF T4 4 ,
15
8
(4.27)
and reproduces the result of [11]. Actually, is the geometric mean of + and ,
4
4
T+
+
Tm
=
=
1,
=
Tm
(4.28)
so that Eq. (4.26) is relevant to derive an expansion of T2 for small + /, whose leading term
reproduces the result of [6,7,10]. Eq. (4.26) goes beyond the small + / approximation by taking
consistently into account the back reaction term T2 that appears in the r.h.s. of Eq. (4.20).
A conceptual difference of our approach compared to that of [6,7,10] is that we insist on the
difference between the two temperatures T , where T T+ . Thus, the only stringy constraint
to not reach the Hagedorn temperature during the cosmological evolution is T TH . In that
case, the temperature T (t) remains small, T (t) TH , for all t > tH (and t < tH in the big
crunch branch), where tH is the Hagedorn time.
The case T2 > 1
We cannot use any more a function N of the form (4.7) to map the Friedman equation of the
present case to the pure de Sitter one. We thus actually solve Eq. (2.5) in a more straightforward
way. In the gauge N 1, one can immediately express a monotonically increasing time t as
144
Fig. 6. Cosmological scenario of an expanding RW universe in presence of thermal and non-trivial time-dependent
moduli effects above some critical value. A phase of expansion starts from a big bang and evolves smoothly toward an
inflationary one. This describes a second-order transition.
a function of a:
a
t (a) =
0
a 0,

1
2a 2 1
+ ti arctanh
,
2
u4 u2 + T2 /4
2 a 4 a 2 + T2 /4
u du
(4.29)
where the argument of the square root is never vanishing and we have chosen the integration
constant ti = (4)1/2 arctanh T1 . Inverting the function t (a), one finds

1
1
1
1 + T2 1 sinh(2 t), t ti arcsinh
a(t) =
(4.30)
.
2
2
T2 1
The cosmological evolution described by this solution starts with a big bang at t = ti , while for
large positive time, it is inflationary. It thus looks similar to the case T2 < 1, both at the beginning
of time and for large t, when the first-order transition has occurred. However, in the present case,
the evolution from the small to very large scale factors is smooth, (see Fig. 6). Actually, the
solution (4.30) has an inflection point arising at t = tinf , where a(tinf ) = ainf ,

T 1
T
1
,
ainf =
,
tinf = arcsinh
(4.31)
T + 1
2
2
associated to a second-order phase transition. As in the case T2 < 1, the solution should be trusted
as soon as the temperature is below TH .
For completeness, we signal that another solution to Eq. (2.5) is found under the time reversal
t t and is thus monotonically decreasing. A de Sitter like universe is contracting since an
145
infinitely past time and evolves in a smooth way around t = tinf toward a branch that ends with a
big crunch.
The case T2 = 1
Taking the limit T2 1 in the solutions found in the cases T2 < 1 and T2 > 1 reaches the
same static solution,
1
where a0 = .
2
a(t) a0
(4.32)
It corresponds to a universe with an S 3 -space of constant radius. Actually, the T2 1 limit

amounts to send the big bang to an infinitely past time (see Eqs. (4.14), (4.17)), while keeping
the instant the transition occurs at t = 0. Similarly, the T2 1+ limit sends ti (see
Eq. (4.30)), while keeping tinf finite. In both cases, the solutions are thus stretched and converge
toward a constant. In the two cases, it is however possible to redefine the big bang time at a
conventional t = 0 by a T2 -dependent shift in the definition of time. Taking only then the limit
T2 1 reaches solutions where the dynamical part of the solutions are not sent to infinity any
more.
To find these new solutions, one can write the Friedman equation (2.5) in the form
2

(a a)
2 = a 2 a0 ,
(4.33)
and derive two monotonically increasing solutions for the time as a function of a,
a
1
t (a) =
u du
,
u2
a02
0 a < a0 ,
(4.34)
a > a0 .
(4.35)
and
1
t (a) =
a
2a0
u du
,
u2 a02
Integrating explicitly these expressions and inverting the functions t (a) reaches then the two
cosmological evolutions:
a(t) = a0 1 e2 t ,
t 0,
(4.36)
and
a(t) = a0 1 + e2
t ,
t R.
(4.37)
The first one starts with a big bang arising at t = 0, converges quickly toward the static solution,
(and should be trusted as soon as T (t) < TH ). The second is a deviation from the static solution
toward the inflationary phase. Two other monotonically decreasing solutions are found under the
time reversal t t in Eqs. (4.36) and (4.37). The four solutions are drawn on Fig. 7.
146
Fig. 7. Cosmological scenario of an expanding RW universe in presence of thermal and non-trivial time-dependent moduli
effects at a critical value. A phase of expansion starts from a big bang and converges quickly toward a static universe.
After a very long time, the universe can either die in a big crunch or enter into an inflationary phase.
5. de Sitter deformation by moduli fields

In this section, we generalize the thermally deformed de Sitter solution by including the effect
of the non-trivial time dependence of some moduli fields. To be specific, our choice of matter
action to be added to (2.3) is

Mp2

d 4 x g g i i ,
Sm =
(5.1)
16
where the i s are the moduli fields with neither potential, nor interactions with the thermal system considered before. Assuming the RW metric (2.1) and the presence of the thermal corrections
discussed before, the modified MSS action becomes
1
Seff =
2
tf
ti
where

T 2
1
1 2
1 a 2
1 a 2 a tf
dt N a 2
+ 2
+
+
,
2 N ti
N a
a
4a 4 6N 2 i
3
(5.2)
T 2
is a new positive constant. The energymomentum tensor can be expressed in terms of

2 2

3T N
1
1 2
=
(5.3)
+ i ,
2
12 2 4 N 2 4 a 4
2 2

T N
1
1 2
P=
(5.4)
+
.
2 i
12 2 4 N 2 4 a 4
The equation of motion of i is trivially solved,
2 2i N
,
i =
3 a 3
(5.5)
147
where 2 2i /3 is an integration constant, whose form is chosen for later convenience. Defining

2i ,
2 =
(5.6)
i
the moduli contributions of and P become

= P =
1
.
27 2 4 2 a 6
(5.7)
To solve the Friedman equation (2.5) in presence of the extra 1/a 6 -term induced by the motion
of the i fields, we follow the method implemented in the previous sections. We look for a real
constant in the definition of N (t ),
N 2 (t ) =
1
,
1 + 1 /a 2 (t )
(5.8)
so that the extra 1/a 6 contribution is cancelled. This can be achieved by choosing such that
P( ) 3 + 2 +
T 2
4
2 = 0.
4
27
(5.9)
This polynomial always has a real positive root (see Appendix A). Using it, Eq. (2.5) is taking
the form
2
2 + + T 2 /4
a
1 +
(5.10)
+
=+
.
2
a
a
a 4
The above equation has the same form with that of the 1/a 4 -thermally deformed de Sitter one
2 -curvature term and the 1/a 4 discussed previously, up to a shift in both coefficients of the 1/a
thermal term. Using a simple rescaling of the radius a (by a factor 1 + ), one can immediately
deduce the solutions from the previous section in terms of the time variable t. This implies in
particular that the analogue of the order parameter T2 of the pure thermal case is now the positive
quantity
=
2
4 2 + 4 + T 2 16
=
,
27 (1 + )2
(1 + )2
(5.11)
where these two definitions of are equivalent as a consequence of Eq. (5.9). In Appendix A,
we shown that the condition < 1 is equivalent to the statement that the parameters T 2 and 2
satisfy the inequalities,

T 2 < 1 and 2 < h T 2 ,
(5.12)
where
h(T 2 ) =
1
2
2

3 2
3
1
1 T + 1
1 T 2
.
4
4
2
(5.13)
This defines a domain in the (2 , T 2 )-plane shown in Fig. 8. At the origin of this domain, one
has = 0, while = 1 on its boundary. Outside, this region, on has > 1.
148
Fig. 8. Phase diagram in the (T 2 , 2 )-plane. In the domain 2 h(T 2 ) defined in Eq. (5.12), a first-order phase transition between two cosmological branches can occur. Outside the domain, there is a solution describing a second-order
phase transition. On the critical curve 2 = h(T 2 ), there are cosmological evolutions asymptotic to a static solution.
Furthermore, one can introduce the temperatures T+ < T in the interior of the domain to replace the two parameters 2
and T 2 . On the boundary curve, these temperatures converge and their common value can be considered as a function
TcT (T 2 ) or Tc (2 ).
The case < 1

Eq. (5.10) admits the cosmological evolution
a(t ) = N + cosh2 ( t ),
where
N =
1 +
(1 )1/4
and

1
1
1 .
2
1
(5.14)
(5.15)
To proceed, we switch to the more intuitive N 1 gauge. This requests a redefinition of the time
variable dictated by Eq. (3.9):

t
cosh2 ( v) +

.
t = dv
(5.16)
where =
2
(1 + ) 1
cosh ( v) + +
0
An explicit form of t as a function of t can actually be determined, however we find more

convenient to work with the above integral representation. The scale factor a(t(t)) takes a simple
form in terms of the function t(t) found by inverting the definition (5.16),

a t(t) = N + cosh2 t(t) .
(5.17)
Some comments are in order:
The slope d t/dt is positive everywhere, and for large t one has t t. This means that the
moduli deformation of the thermal de Sitter solution appears as a mild redefinition of time and
normalization factor.
149
An Euclidean solution is defined via the analytic continuation, t = i , i.e. t = i ,

0

cos2 ( v) +

2
( ) with = dv
. (5.18)
aE ( ) = N + cos
cos2 ( v) + +
The range of Euclidean time is
/2 0, i.e. i 0,
where i is given by
0
i =
(5.19)
cos2 ( v) +
dv
.
cos2 ( v) + +
(5.20)
The boundary terms in Eq. (3.3) vanish.

The Euclidean solution admits (as in the
pure thermal case)two analytic
continuations,
one
for each boundary. The one associated to i , = /2 + i t, i.e. = i + i t,
gives rise to a sinh-type cosmological solution,

0

sinh2 ( v)

2

t(t) with t = dv
.
a t(t) = N sinh
(5.21)
+ sinh2 ( v)
t
The range of time is
1
arcsinh t 0,
0
i.e. ti =
arcsinh
sinh2 ( v)
dv
t 0.
+ sinh2 ( v)
(5.22)
The full scenario is thus similar to the one presented in the previous section. There is a big
bang at t = ti and the universe expands till t = 0, before either contracting or experimenting a
first-order transition to an inflationary phase. At the transition, the scale factor is jumping,

1 1
1+ 1
a+ = (1 + )
.
a = (1 + )
(5.23)
2
2
Thus, the present moduli deformation is generalizing the pure thermal one discussed previously
(see Fig. 3).
The case > 1
Eq. (5.10) admits the t-dependent solution

1 +
1
1
, (5.24)
1 + 1 sinh(2 t ), t ti arcsinh
a(t ) =
2
1
2
we want to express in terms of a time variable associated to the natural N 1 gauge. Integrating
Eq. (3.9), one can define t as a function of t by

t

1 sinh(2 v) + 1

,
t = dv
(5.25)
2
1 sinh( 2v) + 1 + 1+

0
150
where
0
t ti
ti
and rewrite

a t(t) =

1 sinh(2 v) + 1
dv
,
2
1 sinh( 2v) + 1 + 1+

1+
2
(5.26)

1 + 1 sinh 2 t(t) .
(5.27)
As in the previous case, the slope d t/dt is always positive and t t for large t. The solution
is thus a mild deformation of the pure thermal case. It starts with a big bang at t = ti and is
monotonically increasing. For large time, the behavior is inflationary. The evolution can again be
interpreted as a second-order phase transition since there is a unique inflection point (see Fig. 6).
To show this, one can express dt/da as a function of a from the Friedman equation (in the N 1
gauge) and take a derivative. One finds that the inflection point arises at the radius ainf , where
2
ainf
> 0 satisfies
x3
T 2
8
x 2 = 0.
4
27
(5.28)
Finally, a monotonically decreasing cosmological solution is found under the time reversal
t t.
The case = 1
The static solution obtained in the limit 1 in the previous cases is

1 +
a(t) a0 where a0 =
.
2
Beside this constant radius universe, we again have the two t-dependent evolutions
a(t ) = a0 1 e2 t.
To consider them in the N 1 gauge, we integrate Eq. (3.9) in each case and find

t

1 e2 v
t
(t)

,
a t (t) = a0 1 e
where t = dv
2
2 v
1 + 1+
e
(5.29)
(5.30)
(5.31)
for t 0, i.e. t 0, and

a t(t) = a0
1 + e2 t(t)
t
where t =
0

dv
1 + e2
1+
1+
+ e2
(5.32)
for arbitrary t and t. As in the cases = 1, one has d t/dt > 0 and this redefinition of time
does not change the qualitative behavior of the solutions (see Fig. 7). They are monotonically
increasing as in the pure thermal case, while two other solutions obtained under t t are
decreasing.
151
Fig. 9. Critical temperature as a function of T 2 . It is defined on the curve 2 = h(T 2 ).
5.1. Other parameterizations of the solutions when 1

The temperatures at the first-order transition
We would like to replace the parameters 2 and T 2 that appear in the scale factor solution by
the temperatures T at t = 0 . From the definition (4.6) (with T 2 replacing T2 ) and the relation
(4.3), using the fact that aT a+ T+ = a T , one has
T
2 2
,
T , T = Tm
(5.33)
(1 + )(1 1 )
where the temperature Tm has been defined in Eq. (4.21). On the critical curve that is delimiting
the domain (5.12), one has 2 = h(T 2 ). This implies = 1 and thus T+ = T . It is shown in

defined in Eq. (A.4). One can then deduce the critical
Appendix A that in this case, equals +
temperature TcT on the boundary of the domain (5.12) by TcT (T 2 ) T (T 2 , h(T 2 )) (see Fig. 8),
3
2
2 T
TcT T = Tm
(5.34)
,
3 2
1 + 1 4 T
plotted on Fig. 9. The temperature on the critical curve can also be considered as a function of 2 .
This can be done by expressing T 2 = h1 (2 ),

4 4
1 2
2
1/3
1/3
,
(A + B) + (A B)
T =
(5.35)
3 3
2
where
1
A = + 2 ,
8

B=

1
2 2 +
,
4
for 2 1.
(5.36)
152
Defining Tc (2 ) T (h1 (2 ), 2 ) (see Fig. 8), one obtains

Tc 2

= Tm 3
1/4
1 ((A + B)1/3 + (A B)1/3 12 )2
1/4
.
((A + B)1/3 + (A B)1/3 + 12 )2
(5.37)
Since T+ a+ = T a , one has

=
4
,
(T+ /T + T /T+ )2
(5.38)
a relation that generalizes Eq. (4.21) and the duality T+ T . Actually, for Eqs. (5.38) and
(4.21) to have precisely the same form, one can introduce a mean temperature Tm 2 T+ T ,

1/4
T 2
Tm = Tm
(5.39)
,
4 2 + 4 + T 2
where Tm has been defined in Eq. (4.21). However, this definition is less useful than its counterpart in the pure thermal case. This is due to the fact that Tm depends on the point (T 2 , 2 ) of the
parameter space, while in the pure thermal case, 2 = = 0 so that the mean temperature Tm is
independent of the remaining parameter T 2 .
We may express the evolution of the scale factor a(t(t)) in each branch with T ,

T 2 T+2
1
a t(t) =
(5.40)
1 + 2
cosh 2 t(t) , t > 0
2
2(1 A)
T + T+
and

T 2 T2
1
1 + +4
cosh
2 t(t) ,
a t(t) =
2
2(1 A)
T+ + T
t < 0,
(5.41)
where has been expressed in terms of A that involves T and Tm ,

1 + =
1
1A
where A =
1 T+2 T2 /Tm4
.
(T+ /T + T /T+ )2
(5.42)
To obtain the above expression for A, we have used Eq. (5.11) to express T 2 in terms of
and . Then we combine the result with Eqs. (5.33) and (5.38).
Under the duality T+ T , the two cosmological solutions a(t(t)), for t > 0 and t < 0 are
interchanged:

(T+ T )
(5.43)
a t(t) for t > 0 a t(t) for t < 0 .
For completeness we display the constants N , and appearing in Eqs. (5.17), (5.16) and
(5.21), in terms of T ,

1 (T+ /T )2
1
N =
,
(1 A) 1 + (T+ /T )2
=
1
,
(T /T+ )2 1
1 T+2 T2 /Tm4
.
(T /T+ )2 (T+ /T )2
(5.44)
153
Parameterization with T 2 and 2

Let us determine explicit expressions of the scale factor in terms of T 2 and 2 , when 1.
This will be useful in the next section, when we study the probability transition. Since the parameters T 2 and 2 are chosen in the domain (5.12), the polynomial P in Eq. (5.45) has three
real roots (see Appendix A). Noting that P(a ) = 0, as can be checked with Eq. (5.11), the
polynomial P is actually,

2
2
x + a+
.
P(x) = (x ) x + a
(5.45)
Using Cardans formulas, one finds

1

2
4 3T cos
=
1 ,
3
3

1
2
2

2
a =
1 4 3T cos
,
3
3
(5.46)
(5.47)
where

162 + 9T 2 8
= arccos
.
(4 3T 2 )3/2
(5.48)
We note that all values of in [0, ] can be reached. At the origin (T 2 , 2 ) = (0, 0) of the domain
(5.12), one has = , while vanishes on the critical curve 2 = h(T 2 ). The constant N of
2 /a 2 ,
Eqs. (5.17) and (5.21) can be determined by expressing 1 from the ratio a+

1 34 T 2 1/4 2
sin
.
N =
(5.49)
3
3
2 = N 2 and = N 2 give then
The relations a
)
4 3T 2 cos( 3 ) 1
1 4 3T 2 cos( 2
3
=
,
=
.
3 4 3T 2 sin( 3 )
3 4 3T 2 sin( 3 )
For completeness, the temperatures at the transition are also found to be
3
2 T
.
T = T m
2

2
1 4 3T cos( 3 )
(5.50)
(5.51)
5.2. First-order transition probability

When 1, the transition probability at t = 0 between the big bang branch and the deformed
de Sitter cosmology can be estimated by computing the Euclidean action. Changing of integration
variable d = daE /a E in the Euclidean analog of the action (5.2) gives, on shell,
SEeff
1
=
a+
a
daE
aE

2 2

4
2
P aE +
,
27 P(a 2 )
E
(5.52)
154
where P is given in Eq. (5.45) and enters in the Euclidean equation of motion
2
2)
N 2 P(aE
a E
= 6E
aE
2
aE
(5.53)
we have used. Clearly, SEeff is negative. To explicitly evaluate it, we prefer considering its integral
form on and use Eq. (2.5) in Euclidean time to be on shell,
0
SEeff =
3
d NE aE

T 2
1
.
2
4
aE
4aE
(5.54)
In this expression, aE ( ) and NE ( ) d/d are referring to Eq. (5.18). The result can be
expressed in terms of complete elliptic integrals of the first kind, K(k), and second kind, E(k),

1
+
2 1/4
SEeff = 5/4 4 3T
sin
3
3

4 3T 2 cos( 3 ) 1 + 32 T 2
E(k)
(5.55)
K(k) ,
3 4 3T 2 sin( +
)
3
where

sin( )
3
.
k=
sin( +
3 )
(5.56)
To have a better intuition of the behavior of this action as a function of the parameters T 2 and 2 ,
we concentrate on the boundary of the domain (5.12):
On the side 2 = 0 corresponding to the pure thermal case, the action becomes

1 + 1 2

1
T
SEeff =
E(k) 1 1 T 2 K(k) ,
3
2
(5.57)
where
2(1 T 2 )1/4
k=
.
1+
(5.58)
1 T 2
To derive these expressions, one can use the fact that in Eq. (5.46) vanishes and is thus giving rise to an identity satisfied by . Eq. (5.57) reproduces the result of [10].9 At the origin
(T 2 , 2 ) = (0, 0), one has k = 1 and E(1) = 1, so that the result 1/3 of the pure de Sitter case
is recovered. At (T 2 , 2 ) = (1, 0), k = 0 and E(0) = K(0) = /2, so that the action vanishes.
This is consistent with Eq. (5.52) since a+ = a and 2 vanishes.
9 Up to minor misprint errors in [10].
155
On the critical curve 2 = h(T 2 ), the action is taking the form

SEeff
1+
=
9
1 34 T 2

1 34 T 2 1/2
(1 34 T 2 )1/4
(5.59)
If it is explicitly negative for T 2 1, it vanishes for T 2 = 1 only. This shows that even if a+ a
on the critical curve, the term proportional to 2 in the integrand of Eq. (5.52) contributes, due
to the fact that the denominator vanishes. The probability p refers in this case to the transition
(after an infinite time) from the big bang branch (5.31) to the inflationary one (5.32). The quantity
(1 p) refers to the transition (after an infinite time) form the big bang branch (5.31) to its time
reversal, i.e. describing a big crunch. This is the case since considering the solutions (5.17) and
(5.21) of the case < 1, and performing shifts on the origins of times, give rise to (5.31) and
(5.32) in the limit 1 .
The side T 2 = 0 is corresponding to the pure moduli deformation of the de Sitter solution.
From Eq. (5.59), one finds that at the corner (T 2 , 2 ) = (0, 1) of the domain (5.12), the action
is taking the value /9. Since this result is close to the 1/3 at the origin of the (T 2 , 2 )plane, the dependence of the action on 2 seems to be mild along this axis.
This remark happens to remain true for arbitrary fixed T 2 . To understand this, one can note
in Eq. (5.54) that the 2 /a 6 contribution of i2 in the action SEeff has disappeared on shell. Thus,
the only dependence on 2 occurs through the mild deformation appearing in NE . This is
confirmed on the 3-dimensional plot of Fig. 10, where SEeff appears almost constant when 2
varies from 0 to h(T 2 ), at fixed T 2 .
From Fig. 10, one can also see that the transition amplitude p e2SEeff is larger and larger
when T 2 decreases from h1 (2 ) to 0, at fixed 2 . However, the magnitude of p highly depends
on . For instance, p may not be large even for small T 2 , due to the 1/-dependence of SEeff ,
for large .
At fixed and T 2 , switching on 2 increases (see Eqs. (5.50) and (5.48)), and thus 2|ti | (see
Eq. (5.22)), the range of time along the big bang/big crunch evolution. From Eq. (5.51), larger
and larger 2 imply also lower and lower temperature T . As a result, increasing 2 makes this
cosmological branch more and more viable.
Similarly, at fixed and 2 = 0, switching on T 2 has the effect to increase 2|ti |. However,
T varies from 0 to Tc , with the risk for the universe to be too hot. The case 2 = 0 has a
different behavior. When T 2 grows up from 0 to 1, 2|ti | increases as well but the temperature T
varies from + to Tm .10 Along this axis, T 2 is thus making the big bang/big crunch branch
more viable.
Finally, when we reach the limit case corresponding to (T 2 , 2 ) sitting on the critical curve,
after the universe is born with a big bang, it converges quickly toward a static state for a very
10 Note that T + when 2 0 and 2 0, while T 0 when 2 0 and 2 = 0. This ambiguity in the
T
T
definition of T in the pure de Sitter universe is due to the fact that taking 2 = 0 (i.e. = 0) or T 2 = 0 are operations
that do not commute (see Eq. (5.39)).
156
Fig. 10. Euclidean action SEeff as a function of the radiation parameter T 2 and the moduli deformation 2 . The variable
(T 2 , 2 ) spans the domain (5.12). The dependence on 2 is weak, while the absolute value of SEeff decreases when T 2
is switched on. (We have chosen = 1 on this plot.)
long time, i.e. an S 3 of constant radius

1 + 1 3 2

4 T
a a = a0 =
,
3
(5.60)
where a0 has already been defined in Eq. (5.29). However, it finally enters in an inflationary
phase or dies in a big crunch. Also, one can note that in the case > 1, the big bang evolution is
longer and longer when 1+ , since ti (see Eqs. (5.24) and (5.26)), but the final state
on the universe is always inflationary.
6. Conclusions
In this work, we analyze some aspects concerning inflationary solutions and possible transitions between different cosmological behaviors. Our starting point is a bare de Sitter background
deformed by the presence of a thermal bath and the motion of moduli fields. The thermal bath
summarizes the effects of light degrees of freedom in a consistent way and shows up as a CR /a 4
correction in the MSS-action. The moduli contribution gives a CM /a 6 correction to the energy
density.
A systematic analysis of the system involves an almost triangular domain in the deformation
parameter space (T 2 , 2 ). We solve the gravitational equations in all cases and find that a firstorder phase transition can occur inside this domain, while a second-order one arises outside.
157
In the second case, a smooth transition always occurs between the big bang cosmological
evolution (as soon as T (t) < TH ) and the inflationary behavior. In the first case, we calculate
the transition probability between the big bang cosmological evolution (where T T (t) < TH )
and the inflationary branch (where T (t) T+ ). The two solutions are connected into each other
via a gravitational instanton allowing a double analytic continuation. We find, the existence of
a temperature duality T+ T that interchanges the two solutions. Inside the domain of the
parameter space, the origin (T 2 , 2 ) (0, 0) corresponds to the pure de Sitter case where the
big-bang branch disappears and the instanton connects the inflationary universe to nothing. In
the two other extreme cases, (i) the pure thermal deformation (T 2 , 2 ) (1, 0) and (ii) the pure
moduli deformation (T 2 , 2 ) (0, 1), the big bang branch remains almost static for a very long
period of time. However, the probability transition is minimal in case (i) and maximal in case (ii).
Thus, the late future of these evolutions is most probably contracting (till a big crunch occurs) in
case (i), and inflationary in case (ii).
The above analysis and results can be applied to more complex systems where the radiative
and temperature corrections are effectively taking the form of cosmological, curvature and radiation terms during the time evolution [12]. These richer models occur in string effective no-scale
supergravities, i.e. associated to N = 1 string compactifications where N = 1 supersymmetry
is spontaneously broken. First- and second-order phase transitions are again involved [12]. The
stringy origin of these models is giving us the possibility to go beyond the field theory approximation. In particular, it is possible to study the big bang cosmological evolution above the Hagedorn
temperature [18] as well as the stringy analog of the wave function of the universe [19].
Acknowledgements
We are grateful to Constantin Bachas, Ioannis Bakas, Gary Gibbons and John Iliopoulos for
discussions.
The work of C.K. and H.P. is partially supported by the EU contract MRTN-CT-2004-005104
and the ANR (CNRS-USAR) contract 05-BLAN-0079-01 (01/12/05). C.K. is also supported by
the UE contract MRTN-CT-2004-512194, while H.P. is supported by the EU contracts MRTNCT-2004-503369 and MEXT-CT-2003-509661, INTAS grant 03-51-6346, and CNRS PICS 2530
and 3059.
Appendix A
We would like to find in this appendix the phase diagram in the (T 2 , 2 )-plane that is delimiting when the solutions (5.17) and (5.21) for the scale factor are valid, and when it is instead
the evolution (5.24) that is relevant. The two first solutions exist when 1, while the last one
occurs when 1.
First of all, we note that the polynomial P defined in Eq. (5.9) has three roots, whose product
is 42 /27. If two of them are complex conjugate, the third one is thus real positive. If on the
contrary they are all real, one or three of them must be positive. However, since the sum of these
real roots is 1, only one is positive. In any case, we always have a single real positive root, ,
and eventually two other real negative ones.
For T 2 > 4/3, the derivative P of the polynomial defined in Eq. (5.9) is everywhere positive,
so that P has a single root. When
4
T 2 ,
3
(A.1)
158
the derivative P admits two roots,

1
3 2

= 1 1 T .
3
4
(A.2)
) 0, while P( ) 0 (so that P has 3 real roots) if and only if 2

One always has P(+
h(T 2 ), where h(T 2 ) is defined in Eq. (5.13). Actually, this condition implies in particular that
h(T 2 ) 0, which is equivalent to T 2 1. As a conclusion, the polynomial P has three real roots
(one positive and two negative), if the inequalities (5.12) are satisfied, while it has only one real
root in all other cases.
Let us determine when 1. This condition can be translated into a degree two inequality
for that admits solutions if the condition (A.1) is satisfied. In that case, 1 is equivalent to
having

+
,
where

1
4 3T 2 1 .
3
(A.3)
(A.4)

Since having +
< 0 is equivalent to satisfying T 2 > 1, the single real root that P has in this
, ], since it is positive. Thus, 2 > 1 implies > 1.
case is outside the range [
+
T
Let us then concentrate on the case T 2 1. We note that when the parameters 2 and T 2 are
) = 0. This means that the
chosen on the critical curve defined by 2 = h(T 2 ), one has P(+

single real positive root of P is then precisely + . In that case, we have = 1. If instead we
, due to the fact that
take 2 > h(T 2 ), the only real root of P is positive and satisfies > +
P (x) > 0 for any x > 0. We thus have > 1. On the contrary, if 2 < h(T 2 ), the real root of P

for the same reason, so that we have < 1.
which is positive satisfies < +
As a conclusion, the solutions (5.17) and (5.21) corresponding to the case 1 arise when
the point (T 2 , 2 ) belongs to the interior of the domain (5.12) of Fig. 8. Outside this domain, the
evolution (5.24) associated to 1 is the correct one. On the critical curve 2 = h(T 2 ), one
has = 1.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
E. Witten, Instability of the KaluzaKlein vacuum, Nucl. Phys. B 195 (1982) 481.
A. Vilenkin, Creation of universes from nothing, Phys. Lett. B 117 (1982) 25.
A. Vilenkin, The birth of inflationary universes, Phys. Rev. D 27 (1983) 2848.
J.B. Hartle, S.W. Hawking, Wave function of the universe, Phys. Rev. D 28 (1983) 2960.
S.W. Hawking, The cosmological constant is probably zero, Phys. Lett. B 134 (1984) 403.
S. Sarangi, S.H. Tye, The boundedness of Euclidean gravity and the wavefunction of the universe, hep-th/0505104.
S. Sarangi, S.H. Tye, A note on the quantum creation of universes, hep-th/0603237.
S. Watson, M.J. Perry, G.L. Kane, F.C. Adams, Inflation without inflaton(s), hep-th/0610054.
R. Brout, P. Spindel, Tunneling in cosmology and isothermal inflation, Nucl. Phys. B 348 (1991) 405.
R. Brustein, S.P. de Alwis, The landscape of string theory and the wave function of the universe, Phys. Rev. D 73
(2006) 046009, hep-th/0511093.
[11] M. Bouhmadi-Lopez, P. Vargas Moniz, Quantisation of parameters and the string landscape problem, JCAP 0705
(2007) 005, hep-th/0612149.
[12] C. Kounnas, H. Partouche, Inflationary de Sitter solutions from superstrings, arXiv: 0706.0728 [hep-th].
[13] S.W. Hawking, The quantum state of the universe, Nucl. Phys. B 239 (1984) 257.
159
[14] E.R. Harrison, Classification of uniform cosmological models, Mon. Not. R. Astron. Soc. 137 (1967) 69.
[15] A.Y. Kamenshchik, I.M. Khalatnikov, A.V. Toporensky, Nonminimally coupled complex scalar field in classical
and quantum cosmology, Phys. Lett. B 357 (1995) 36, gr-qc/9508034.
[16] A.O. Barvinsky, A.Y. Kamenshchik, Cosmological landscape from nothing: Some like it hot, JCAP 0609 (2006)
014, hep-th/0605132.
[17] H. Firouzjahi, S. Sarangi, S.H.H. Tye, Spontaneous creation of inflationary universes and the cosmic landscape,
JHEP 0409 (2004) 060, hep-th/0406107.
[18] C. Kounnas, H. Partouche, N. Toumbas, J. Troost, in preparation.
[19] C. Kounnas, N. Toumbas, J. Troost, A wave-function for stringy universes, arXiv: 0704.1996 [hep-th].

Partially quenched chiral perturbation theory

in the -regime
Poul H. Damgaard a , Hidenori Fukaya b,
a The Niels Bohr Institute, The Niels Bohr International Academy, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
b Theoretical Physics Laboratory, RIKEN, Wako 351-0198, Japan

Abstract
We calculate meson correlators in the -regime within partially quenched chiral perturbation theory. The
valence quark masses and sea quark masses can be chosen arbitrary and all non-degenerate. Taking some
of the sea quark masses to infinity, one obtains a smooth connection among the theories with different
number of flavors, as well as the quenched theory. These results can be directly compared with lattice QCD
simulations.
PACS: 12.38.Gc
Keywords: Chiral symmetry; Lattice QCD
1. Introduction
In the low energy limit, the dynamics of QCD is described by the pion fields which appear as
pseudo-NambuGoldstone bosons accompanying the spontaneous breaking of chiral symmetry.
Chiral dynamics and chiral perturbation theory (ChPT), play an essential role in understanding
the interactions among the pions themselves, as well as their couplings with the other (heavier)
hadrons and sources.
The fundamental parameters of ChPT are unknown coupling constants in the effective theory.
But they can be determined by non-perturbative and first-principle calculations in the underlying
E-mail addresses: phdamg@nbi.dk (P.H. Damgaard), hfukaya@riken.jp (H. Fukaya).

doi:10.1016/j.nuclphysb.2007.10.010
P.H. Damgaard, H. Fukaya / Nuclear Physics B 793 (2008) 160191
161
theory, QCD. The most important low-energy constants are the chiral condensate , and the pion
decay constant, F , both at leading order. Numerical simulations of lattice QCD give currently
the most promising approach to achieving this non-perturbative determination. Because such
computer simulations necessarily are restricted to finite volumes, it is important to understand
finite-volume effects in the effective field theory.
Near the chiral limit, the finite-volume effects become increasingly significant due to the
diverging correlation length of the Goldstone bosons. In particular, when the pion correlation
length, or the inverse of the pion mass m overcomes the size of the box L,
1
QCD
L
1
,
m
(1.1)
where QCD is the QCD scale, the zero-momentum mode has to be treated non-perturbatively
and the ChPT has to be performed in a way that achieves this in a systematic fashion: the socalled -expansion [14]. New counting rules are needed to order the perturbative expansion in
this case. In units of the ultraviolet cut-off, the expansion parameter can be defined as , where,
unusually, the pion mass m is not treated as being of the same order as pion momentum p.
Instead,
m p 2 2 ,
(1.2)
while the inverse of spacetime volume V and quark mass mq are being of O( 4 ). The particularly important combination mq V is thus treated as of order unity. With this expansion, the
precise analytical predictions for physical observables in the low-energy sector of QCD at finite
volume V are expressed in terms of the low-energy constants at infinite volume. By comparing
numerical results at finite volume with these predictions, one can thus extract the infinite-volume
constants directly from finite-volume simulations, without the need for extrapolations of data
to infinite volume. The closer one gets to the chiral limit, the bigger is the advantage of this
approach.
A few years ago, the predictions for correlation functions of ChPT were extended to the
cases of both quenched QCD and full QCD at sectors of fixed gauge-field topology [5,6]. For
the chiral condensate, the studies were also extended to the partially quenched cases [7,8]. In
quenched QCD these analytical predictions suffer in the -regime from quenched finite-volume
logarithms [9]. Strictly speaking, such logarithms prevent taking the infinite-volume limit, and
basically invalidate the whole chiral expansion in this regime for the quenched theory. The hope
is that in finite ranges of volume, the resulting predictions may still have a certain range of validity. For both quenched and unquenched cases these computations were restricted to the case of
degenerate, light, quarks.
As mentioned above, the great advantage of the predictions for correlators in the -regime is
that they appear almost tailored for numerical lattice computations near the chiral limit. Indeed,
exploratory studies of these correlation functions have already shown the great potential [1016].
In particular, the possibility of using to ones advantage the role played by fixing topology in
finite volume has been clearly demonstrated. Also the analytic handle one has on the quark
mass dependence due to the finite size effects in the -regime has proved helpful in reducing the
systematic errors of lattice simulations. In this paper, we generalize these analytical computations
of correlation functions in the -regime to the partially quenched theory with separate valence
and sea quarks, both of which are taken to be non-degenerate. The chiral condensate and pseudoscalar and scalar meson correlators are calculated as functions of non-degenerate quark masses,
162
topological charge, and the volume of the Euclidean spacetime V . In a separate forthcoming
publication [17] we will present the analogous results for the vector and axial vector channels.
In practice, the -regime is not trivially reached in numerical simulations. It is therefore important to be able to go close to the chiral limit, but still only marginally in the -regime, while
valence quark masses are taken to that regime. In fact, although our aim in this paper is the
-regime, our partially quenched chiral perturbation theory (PQChPT) [18,19] in the -regime
smoothly connects all the theories with a different number of flavors as a function of the sea
quark masses. In this way, our calculation interpolates between the -regime and the more conventional p-regime, and in one kinematical regime also mixes the two expansions. For this reason
we provide expressions not just with the (simpler) -expansion propagators, but the more general expressions. If one of the sea quark masses is taken to infinity in the Nf -flavor theory, it
converges to the (Nf 1)-flavor theory. Even the quenched theory can be obtained by carefully
introducing the flavor singlet field before taking all the sea quark masses to infinity. Of course, the
low-energy constants in addition have an inherent flavor dependence that is beyond our control.
Our results have wide applicability to unquenched lattice QCD studies near the chiral limit
[2022]. One can choose various valence quark masses with a fixed sea quark mass. The partially
quenched condensate and meson correlators can be compared with simulations of heavier quarks
which are perhaps only marginally in the -regime (or beyond) while the valence quark mass is
still in the -regime.
This paper is organized as follows. In Section 2, we describe the leading contribution of
the partition function of PQChPT in the -expansion. We discuss, in particular, the exact nonperturbative integral of the zero-modes [23,24] which plays a crucial role in deriving both the
chiral condensate and meson correlators in this extended theory. As one fundamental building
block of this work, the chiral condensate and its 1-loop level correction are obtained in Section 3. The exact zero-mode integrals in the replica limit are calculated in Section 4. We also
derive a non-trivial identity which follows from the unitarity of the group integrals. In Section 5,
our main results on meson correlators are presented. We plot Nf = 2 connected pseudo-scalar
and scalar correlators as examples. The conclusions are given in Section 6.
2. The partition function of PQChPT
Our starting point is the (Nf + N )-flavor chiral Lagrangian,
L=

F2
Tr U (x) U (x) Tr M U (x) + MU (x)
4
2
m20 2
+
(x) +
0 (x) 0 (x),
2Nc 0
2Nc
(2.1)
where and F are the chiral condensate and the pion decay constant at infinite volume, both in
the chiral limit. In the mass matrix
M = diag(m1 , m2 , . . . , mNf , mv , . . . , mv ),

Nf
(2.2)
we have in mind a situation in which the valence quark mass is always taken in the -regime of
mv V O(1), while the physical sea quark mass mi may vary more freely. Unlike standard
chiral perturbation theory, U (x) is an element of the U (Nf + N ) group, and the flavor-singlet
Tr ln U (x), is introduced explicitly as a physical degree of freedom with
field, 0 (x) iF
2
163
additional constants, m0 and . The number of colors is denoted by Nc . As is well known, in this
partially quenched theory one can normally take the m0 limit without difficulty. In terms of
first replicated and then quenched valence quarks one is then going from U (Nf + N ) to SU(Nf )
in a smooth way. Then 0 can be decoupled from the theory. Of course, trouble arises again if
we consider the theory in a regime where the sea quark masses mi have effectively decoupled.
We will discuss this issue below. Separating the zero-mode, U0 , and the non-zero modes, (x),

U (x) = U0 exp i 2(x)/F ,
(2.3)
we consider three types of expansion of the partition function in a sector of fixed topological
charge :
(1) Both of the valence and sea quarks are in the -regime:

V

ZN
m
,
{m
}
=
dU
d
det
U
exp
U
+
MU
Tr
M
v
i
0
0
+N
0
0
f
2
U (Nf +N )

1
+ d x Tr[ ] + ,
(2.4)
2
where the m0 limit is taken and the singlet 0 is decoupled from the theory.
(2) The sea quarks are in the p-regime (but light enough that we can still disregard effects of the
singlet field):

V

dU0 d det U0 exp

Tr M U0 + MU0
ZNf +N mv , {mi } =
2

U (Nf +N )
2
d 4 x Tr[ ]
+
. (2.5)
Tr
M
2
F2
Here the valence sector is expanded as in -expansion, while the sea sector is considered in
the p-regime. Note that the mass term in the valence sector, mv 2 /F 2 , is of O( 4 ) but not
ignored here, in order to see a smooth transition to the p-regime.
(3) The sea quarks are heavy:

V

m
,
{m
}
=
dU
d
det
U
exp
Tr M U0 + MU0
ZN
v
i
0
0
f +N
2

U (Nf +N )
+ d x Tr[ ]
Tr M 2
2
2
F

m20
(Tr )2
( Tr )2 + .
2Nc
2Nc

(2.6)
As the sea quark mass increases, new terms (the non-zero modes mass term, and the singlet
fields) come in. Thus, Eq. (2.6) is the most general form.
In the following, we implicitly restrict ourselves to sectors of small enough fixed topology
for the expansion to be valid [25]. Indeed, this is an essential assumption so that in Eqs. (2.5) and
(2.6) one needs only the mass term Tr[M 2 ] instead of Tr[M(U0 + U0 ) 2 ]/2.1
1 We thank P. Hernandez for stressing this point.
164
For the zero-mode integrals, one needs exact formulae of the group integrals over U (Nf + N )
and means of taking the replica limit. As described in detail in Ref. [5], if we wish to consider
correlation functions with Nv external valence quarks we must embed the Nv valence quarks in a
theory with N replicated quarks in total (of which N Nv do not couple to the external sources),
and then take the limit N 0. Alternatively, one can consider a theory with Nv additional
bosonic flavors of common mass mv . It is easy to understand in the quark determinant picture
that this limit is equivalent to the replica limit:

N
v

Nf
lim det(D + m)
det(D + mv + Ji ) det(D + mv )NNv
N0
Nv
i
= det(D + m)Nf
det(D + mv + Ji )
.
det(D + mv )Nv
(2.7)
Here the left-hand side is the replica prescription, while the right-hand side is the prescription of
the graded formalism.
The zero-mode partition function of n bosons and m fermions are analytically known [23,24],
j 1

det[i J+j 1 (i )]i,j =1,...,n+m
Zn,m
{i } = n
,
n+m
2
2
2
2
j >i=1 (j i ) j >i=n+1 (j i )
(2.8)
where i = mi V . Here J s are defined as J+j 1 (i ) (1)j 1 K+j 1 (i ) for i =

1, . . . , n and J+j 1 (i ) I+j 1 (i ) for i = n + 1, . . . , n + m, where I and K are the
modified Bessel functions. Of particular importance is the case (n, m) = (1, Nf + 1):

Z1,1+N
x|y, {zi }
f
= N
f
2
i=1 (zi
1
Nf
y2)
2
k>j (zk
zj2 )
K (x)
I (y)
xK+1 (x) yI+1 (y)
det 2
x K+2 (x) y 2 I+2 (y)
I (z1 )
z1 I+1 (z1 )
z12 I+2 (z1 )
I (z2 )
z2 I+1 (z2 )
z22 I+2 (z2 )
(2.9)
where x = mb V (mb denotes the bosonic quark mass), y = mv V and zi = mi V . One notes
that

lim Z1,1+N
(2.10)
x|y, {zi } = Z0,N
{zi } ,
f
f
xy
and therefore,

x|y, {zi } = lim y Z1,1+N

x|y, {zi } .
lim x Z1,1+N
f
f
xy
(2.11)
xy
It is also remarkable that

x|y, {z1 , z2 , . . . , zj 1 , zj , zj +1 , . . . , zNf }

Z1,1+N
f

1N +N +1

x|y, {z1 , z2 , . . . , zj 1 , zj +1 , . . . , zNf } zj f Z0,1 f (zj ) z

= Z1,1+(N
f 1)
(2.12)
165
which is consistent with the intuitive notion that the (Nf 1)-flavor theory can be obtained in the
+N +1
1N
limit of large sea-quark mass, mj , up to a normalization factor [zj f Z0,1 f (zj )]|zj .
This decoupling is of course more general. For example, in the case of {zi (for all i)} one
, the leading partition function of quenched chiral perturbation theory.
obtains Z1,1
For different valence quarks (with non-degenerate valence masses), we will also need
Z2,2+Nf :

x1 , x2 |y1 , y2 , {zi }
Z2,2+N
f
=
(x22
x12 )(y22
y12 )
Nf
2
i=1 (zi
1
y22 )(zi2 y12 )
Nf
2
k>j (zk
K (x2 )
I (y1 )
(x1 )
x1 K+1 (x1 ) x2 K+1 (x2 ) y1 I+1 (y1 )
det 2
x1 K+2 (x1 )
x22 K+2 (x2 ) y12 I+2 (y1 )
zj2 )
I (y2 )
y2 I+1 (y2 )
y22 I+2 (y2 )
I (z1 )
z1 I+1 (z1 )
z12 I+2 (z1 )
(2.13)
Let us now define the propagator [26] of the fluctuation field :

P(ij )(kl) (x y) ij (x)kl (y)

2
il j k (M
ij |x y)
=
2
2
2
ik kl (M
ii |x y) kl G(Mii , Mkk |x y)
(i = j ),
(i = j ),
(2.14)
where the indices i, j, . . . can be taken both in the valence and sea sectors. Here Mij2 = (mi +
mj )/F 2 and

1 eipx
Mij2 |x

,
2 + M2
V
p
ij
p =0
(2.15)
1

eipx (m20 + p 2 )/Nc
2
Mii2 , Mjj
|x
,
G
2 )F(p 2 )
V
(p 2 + Mii2 )(p 2 + Mjj
p =0
(2.16)
f

(m20 + p 2 )/Nc
.
F p2 1 +
2
p 2 + Mff
f =1
(2.17)
Note that for small sea-quark masses, m0 can be taken infinity,

1
eipx
2
Mii2 , Mjj
G
|x =
Nf
2
2
V
2
2
p =0 (p + Mii )(p + Mjj )( f
1
2
p 2 +Mff
(2.18)
becomes
Conversely, in the quenched limit of taking the sea quark masses to infinity, G

1 eipx (m20 + p 2 )/Nc
2
Mii2 , Mjj
G
|x
.
2 )
V
(p 2 + Mii2 )(p 2 + Mjj
p =0
(2.19)
166
3. The chiral condensate

At tree level, the partially quenched chiral condensate is obtained by the logarithmic xderivative of the zero-mode partition function Eq. (2.9) followed by the y x limit,
PQ (x, {zi })
N

1
U0 + U0 vv
= lim
N0 N
v
=
= lim
yx
U0
Nf 2
Z0,N
({zi }) i=1
(zi
f
x2)
x K (x)
(xK (x))
det x 2 +1
x (x K+2 (x))
Nf
2
k>j (zk

x|y, {zi }
ln Z1,1+N
f
x
zj2 )
I (x)
xI+1 (x)
x 2 I+2 (x)
I (z1 )
z1 I+1 (z1 )
z12 I+2 (z1 )
I (z2 )
z2 I+1 (z2 )
z22 I+2 (z2 )
(3.1)
where x = mv V , zi = mi V , and we use Eq. (2.10) for the denominator and the minus
sign in the first line is due to the derivative with respect to the bosonic flavor. The quantity
PQ (x, {zi })/ has two properties that follow immediately from the corresponding statements
about the partition functions:
(1) When one of the sea quark masses is large, the partially quenched chiral condensate reduces
to that of the (Nf 1)-flavor theory;

PQ x, {z1 , . . . , zj , . . . , zNf } /

PQ
Nf flavors

x, {z1 , . . . , zj 1 , zj +1 , . . . , zNf } /.

(3.2)
(Nf 1) flavors
As a particular case, one recovers the fully quenched condensate [5] when all sea quark
masses are large, i.e., when mi (see Fig. 1).
(2) If the valence quark mass mv is equal to one of sea quark masses mj , it is equivalent to the
j th flavor quark condensate in the full theory:
full(Nf ,j )
lim
xzj

PQ (x, {zi })
{zi }
= zj ln ZN
f
({zi })
(for any j ).
(3.3)
The first property follows directly from Eq. (2.12) and the fact that
+Nf +1
Z0,1
/Z0,1
1
(3.4)
in the large mass limit. The second property can be shown explicitly by using Eq. (2.10) and
Nf 2
Nf 2
(zi x 2 ) k>j
(zk zj2 ) is always antisymmetric under a swap of
noting that the product i=1
167
Fig. 1. The sea quark mass dependence of the 2-flavor partially quenched condensate at = 0 (solid). The valence quark
mass is fixed to mv V = 2.0. As expected, we obtain a smooth curve crossing the full 2-flavor theory result (dashed) at
m1 = m2 = mv , and converging to the quenched limit (dotted).
x zj . Together with the equation
K (x)
det
xK+1 (x)
x 2 K (x)
+2
zj I (zj )
zj (zj I+1 (zj ))
zj (zj2 I+2 (zj ))
j th column

I (zj )
zj I+1 (zj )
zj2 I+2 (zj )
j th column
K (x)
= det
xK+1 (x)
x 2 K (x)
+2
I (zj )
zj I+1 (zj )
zj2 I+2 (zj )

zj I (zj )
zj (zj I+1 (zj ))
zj (zj2 I+2 (zj ))
(3.5)
which holds for any j , the statement follows.

For the normalization factor of the correlation functions to be computed in the following
sections we need to evaluate the partition function to the given order in the expansion. This
essentially boils down to an evaluation of the one-loop correction to the partially quenched chiral
condensate. This correction can be calculated in a standard manner:
PQ,1-loop
(x, {zi })

1
1
lim
ln ZN
+N mv , {mi }
f
V N 0 N mv

Nf +N N !
"

1
= lim 1 2
d 4 x vi (x)iv (x)
N 0
F NV
v
i
N
1-loop

1
U0 + U0 vv ,
,
N v
U0
(3.6)
168
1-loop
where denotes the integral over the field, while the zero-mode integral U0
can be
calculated as above but due to the vacuum bubble, the arguments x and zi are shifted, at 1-loop, as
mv V = x x eff
where
and mi V = zi zieff ,
(3.7)

Nf +N N !
"

x eff
1
4
d x iv vi
= lim 1 2
,
N0
x
F NV
v
i

Nf +N !
"
zieff
1
4
= lim 1 2
d x if f i
.
N0
zi
F V
(3.8)
With the Feynman rules given in the previous section, the integral can be performed, and
we obtain [27]

Nf

1 2 2
x eff
2
Miv |0 G Mvv , Mvv |0 ,
=1 2

x
F
i

Nf

zieff
1 2 2
Mij |0 G Mii , Mii2 |0 .
=1 2
(3.9)

zi
F
j
The corrections are UV divergent, and thus need regularization. Note that each of x and zi receive
different one-loop correction in general. In practice, however, the following three special cases
are of our interest:
(1) Both of the valence quarks and the sea quarks are in the -regime:
One can to this order set Mij = 0 for all i and j , and take the m0 limit, which leads to
2
1 Nf 1
x eff zieff
=1 2
=
(0|0)
x
zi
Nf
F
(3.10)
(for all i).
This correction is equivalent to that of Nf -flavor full theory. In dimensional regularization,

1
(0|0)
= 2 + O 1/L4 ,
(3.11)
L
is obtained where 1 is known as the shape coefficient [4]. It depends only on the shape of
the 4-dimensional Euclidean spacetime volume.
(2) The sea quarks are in the p-regime while the valence quarks are still in the -regime (the
mi s are heavy but much smaller than the QCD scale, QCD ):
One can take the m0 limit but should keep mi s finite, which leads to
Nf

1
1 2 1
x eff
=1 2
Miv |0
,
Nf
1
x
V
F
p4 (
)
i
zieff
1
=1 2
zi
F
p =0
p 2 +Mii2
Nf

1

1
Mij2 |0

Nf
2
V
2
2
j
p =0 (p + Mii ) ( j

1
2
p 2 +Mjj
(3.12)
Note that a double pole contribution appears in x eff /x, as an effect of the partially quenching.
169
(3) All the sea quark masses are heavy, mi QCD :

In this case, m0 cannot be large, but one can take the mi limit, which leads to

x eff
1 1 (m20 + p 2 )/Nc
(3.13)
=1+ 2
,
x
V
F
p4
p =0
which agrees with the quenched result.

To summarize this section, the chiral condensate to one-loop order is given by
PQ,1-loop

# $ x eff
x, {zi } = PQ x eff , zieff
,
x
(3.14)
where the analytical functional form of PQ (x, {zi }) is given by Eq. (3.1). When all the quarks
are in the -regime, the one-loop correction is, to this order, constant, and can simply be taken
into account in the Lagrangian by shifting according to the above prescription. The chiral
condensate in the infinite volume limit, , and all the other low-energy constants, are of course
expected to depend on the number of flavors. Matching conditions [2832] can ensure smooth
connections between theories with different number of flavors.
4. Zero-mode integrals in the partially quenched theory
4.1. U (Nf + N ) group integrals in the replica limit
In the -regime, the integral over the zero-mode, U0 (which for simplicity of notation will
be denoted by U in this subsection) has to be done exactly. This is the central difference between the -regime and the p-regime, and we are fortunately able to perform the required group
integrations exactly. Because these technical aspects are so important for the calculation that follows, we first give a detailed outline of how the integrations have been done, and how we employ
the replica formalism in this context. As in Ref. [5] the needed group integrals are conveniently
obtained through the identities that follow from the fact that [det(V M)] ZN f +N (x, {zi }) is
a function of M M only. Since the group integrals are known for diagonal sources, this basically solves the problem. We begin, however, with two simpler cases. They do not require any
special techniques, and can be evaluated straightforwardly by making use of the graded partition
function:

PQ (x, {zi })
1
= lim y ln Z1,1+N
x|y,
{z
}
=
Uvv + Uvv
,
i
f
yx
2

1
1
2
=
x|y, {zi }
Uvv + Uvv
lim y2 Z1,1+N
f
yx
4
ZNf ({zi })
x PQ (x, {zi }) PQ (x, {zi })
where the second term of Eq. (4.2) is defined by

=
PQ (x, {zi }) limyx y x Z1,1+Nf (x|y, {zi })
ZN f ({zi })
(4.1)
(4.2)
(4.3)
170
If we put x = zj (for any j ), this simply amounts to removing the fermion determinant of quark
species j . From the definition (4.2) we then immediately get

x PQ (zj , {zi }) PQ (zj , {zi })
x=zj
full(Nf ,j )

full(Nf ,j )
({zi })
({zi }) 2
= zj
,
(4.4)
+
corresponding to the result in the full theory without partial quenching. Conversely, in the limit
zi (for all i),
que
x (x)
x PQ (x, {zi })
,
{zi }
2
PQ (x, {zi })
1 + 2 ,
{zi }
(4.5)
we recover the results of the quenched theory [5]. Here, in a hopefully obvious notation, we have
full(Nf ,j )
que
denoted the chiral condensates in the full and quenched theories by
and (see
Eqs. (3.2) and (3.3)), respectively.
We next consider a purely imaginary source iJ (with J real) on a diagonal (v, v) element
of M,

V M + J = diag z1 , z2 , . . . , (x + iJ ), x, . . . , x .
(4.6)

Nf
We note that
lim det(V M + J ) = det(V M) (1 + iJ /x),
(4.7)

(V M + J ) (V M + J ) = diag z12 , z22 , . . . , 2 , x 2 , . . . , x 2 ,

(4.8)
N0
and
Nf
where = x 2 + J 2 . By the chain rule, J -derivative and -derivative are related through

J
2
1

=
= 0,
=
.
(4.9)
J J =0 J =0
J 2 J =0 x =x
From the above equations, we obtain (denote ZN f +N (x, {zi }) by Z for simplicity),

1
1
i
det(V M + J ) Z|J =0 = det(V M) +
Z = 0,
Z J
x
Z J

2
1
1 2 Z
( 1)
det(V M + J ) Z|J =0 = det(V M)
+
,
Z J 2
Z J 2 J =0
x2

PQ (x, {zi })
1 1
det(V M + J ) Z|=x = det(V M) 2 +
.
Z x
x
x
(4.10)
This gives us the useful identities

1 Z
1
Uvv Uvv = i
= ,

2
Z J J =0
x

2

1
1 Z
PQ (x, {zi }) 2
2
Uvv Uvv
=
=
+ 2.
4
Z J 2 J =0
x
x
171
(4.11)
(4.12)
In order to calculate the meson correlators in PQChPT, we also need matrix elements which
have different valence flavor indices. For example, for the disconnected correlators, we need

1
Uv1 v1 + Uv1 v1 Uv2 v2 + Uv2 v2
4

1
lim
=
x1 , x2 |y1 , y2 , {zi }
y1 y2 Z2,2+N
f
ZNf ({zi }) y1 x1 ,y2 x2

DPQ x1 , x2 , {zi } .
(4.13)
To derive the analytical expressions for these and others closely related, we first consider two
purely imaginary sources Jv1 v1 = iJ1 and Jv2 v2 = iJ2 along the diagonal:

V M + J = diag z1 , z2 , . . . , (x1 + iJ1 ), (x2 + iJ2 ), x, . . . , x .
(4.14)

Nf
The replica limit of this determinant is

lim det(V M + J ) = det(V M) (1 + iJ1 /x1 )(1 + iJ2 /x2 ),
N 0
and (V M + J ) (V M + J ) takes the form

(V M + J ) (V M + J ) diag z12 , z22 , . . . , 21 , 22 , x 2 , . . . , x 2 ,

Nf
(4.15)
(4.16)
where i = xi2 + Ji2 (i = 1, 2). Again, J -derivative and -derivative are related through the
chain rule,

i
Ji
(4.17)
=
= 0.
Ji Ji =0 i Ji =0
From these equations we get

2

1 2 Z
1 2

= 0,
det(V M + J ) Z
= det(V M)
+
Z J1 J2
x1 x2 Z J1 J2 J =0
J =0

1
2
Uv1 v1 Uv1 v1 Uv2 v2 Uv2 v2 =
.
4
x1 x2
(4.18)
Next let us consider a real off-diagonal source, Jv1 v2 = J , for which the determinant is unchanged in the replica limit:
lim det(V M + J ) = det(V M).
N 0
(4.19)
172
Now (V M + J ) (V M + J ) can be diagonalized as

(V M + J ) (V M + J ) diag z12 , z22 , . . . , 2+ , 2 , x 2 , . . . , x 2 ,

Nf
where
(4.20)
&
%
'
' (J 2 + x 2 + x 2 ) J 4 + 2J 2 (x 2 + x 2 ) + (x 2 x 2 )2
(
1
2
1
2
1
2
2
(4.21)
If we assume that x1 = x2 (the special case x1 x2 [5] can be taken as a limiting case afterwards,
see below), we obtain the following relation between J -derivatives and -derivatives:

2

1
(4.22)
=
x2
x1
.
+ + =x1
=x2
J 2 J =0 x12 x22
We now use

1
x
x
Z
1
2
2
2
Z x1 x2
+
+ =x1 , =x2

PQ
(x1 , {zi })
PQ (x2 , {zi })
1
= 2
x
.
x
1
2
x1 x22
(4.23)
Note that a purely imaginary source Jv1 v2 = iJ gives the same results. Thus, one obtains

2
1
PQ (x1 , {zi })
PQ (x2 , {zi })
Uv1 v2 Uv2 v1 = 2
x
.
x
1
2
4
x1 x22
(4.24)
As a check, if we take the mass-degenerate limit x2 x1 we recover

2
1 PQ (x1 , {zi }) x1 PQ (x1 , {zi })
1
Uv1 v2 Uv2 v1 =
+
,
4
2
x1
(4.25)
which is obtained from the formula in the degenerate case in Eq. (4.22),

2
1
2
1
(4.26)
=
+
,
+
4 + =x1
J 2 J =0 4x1 +
%
where = ( J 2 + 4x12 J )/2.
We finally put two real sources on the off-diagonal elements, Jv1 v2 = J1 and Jv2 v1 = J2 . In
the replica limit the determinant becomes
lim det(V M + J ) = det(V M)(1 J1 J2 /x1 x2 ),
(4.27)
N0
and (V M + J ) (V M + J ) diagonalizes as

(V M + J ) (V M + J ) diag z12 , z22 , . . . , 2+ , 2 , x 2 , . . . , x 2 ,

Nf
(4.28)
where
=
&
%
'
' (J 2 + J 2 + x 2 + x 2 ) (J 2 + J 2 + x 2 + x 2 )2 4(J J x x )2
( 1
1 2
1 2
2
1
2
1
2
1
2
2
173
(4.29)
The relation between J -derivative and -derivative can be worked out as above. Assuming again
that x1 = x2 (the degenerate case can also here be recovered by taking the limit x1 x2 afterwards), we get

1
2
=
x
(4.30)
x
.
2
1
J1 J2 Ji =0 x12 x22
+ + =x1
=x2
In the same way as above, we thus find that the two equations

1 2
det(V M + J ) Z
Z J1 J2
J =0

1
2 Z
= det(V M)
+
,
x1 x2 Z J1 J2 J =0

1
x
x
det(V M + J ) Z|i =xi
2
1
Z x12 x22
+

PQ (x1 , {zi })
PQ (x2 , {zi })
1
= det(V M)
+
x1
,
x2
x1 x2 x12 x22
(4.31)
lead to

1
Uv1 v2 Uv2 v1 Uv2 v1 Uv1 v2
4

PQ (x1 , {zi })
PQ (x2 , {zi })
1
= 2
x
,
x
2
1
x1 x22
(4.32)
where we used that purely imaginary sources Jv1 v2 = iJ1 , Jv2 v1 = iJ2 give the same result.
We summarize all pertinent formulas in Appendix A.
4.2. The unitarity formula
From the above formulae and the requirement of unitarity one finds in the replica limit,

N+Nf

1 = lim
Uv1 i Uiv1
N0

=
i
N
f

lim Uv1 v1 Uv1 v1 + Uv1 v2 Uv2 v1 + + Uv1 vN UvN v1 +
Uv1 i Uiv1
N0

i

= Uv1 v1 Uv1 v1 Uv1 v2 Uv2 v1 +
N1
Nf

i
Uv1 i Uiv1 ,
(4.33)
174
where we have used that the N replicated quarks are all degenerate: xi = mvi V = x. One then
obtains a non-trivial identity,
Nf

full(Nf ,j )
({zi })
PQ (x, {zi })
2
2
xPQ (x, {zi }) zj
+1+ 2 =
. (4.34)
x
x 2 zj2
j =1
It is not difficult to show that the left-hand side of Eq. (4.34) actually vanishes in the quenched
limit, cf. Eq. (4.5).
It is also interesting to consider the limit of the full Nf -flavor theory of Eq. (4.34), in which
all the valence and sea quarks are degenerate: x = z1 = z2 = = z. In that degenerate limit we
have

N
full(Nf ,i)
f (z)
(z1 , z2 , . . .)
1
(4.35)
,
Nf
z1 =z2 ==z
i

Nf
full(Nf ,1)
(z1 , z2 , . . .)
(z) z1
=

z1 =z2 ==z

full(Nf ,1)
(z1 , z2 , . . .)
z2
+ (Nf 1)
,
(4.36)

z1 =z2 ==z
full(i)
s are independent of i or j in the generate case.

where we have used the fact that the zj
We thus have

full(Nf ,1)
z2
(z1 , z2 , . . .)

z1 =z2 ==z

Nf

full(Nf ,1)
(z1 , z2 , . . .)
1
z
(z)
=
1
(4.37)
.

Nf 1
z1 =z2 ==z
Similarly, one obtains

(z1 , z2 , . . .)

full(Nf ,1)
z1
Nf 1
=
Nf
z1 =z2 ==z

x PQ (x, {z, z, . . .})
1
+

N
f
x=z
f (z)

.
(4.38)
The unitarity equation for the degenerate case (x = z1 = z2 = = z), Eq. (4.34) then finally
becomes
Nf

N
N
(z) 2
f (z)
1 f (z)
2
Nf
+1+ 2
Nf
z
z

(Nf2 1) x PQ (x, {z, z, . . .})

=

Nf
x=z

full(Nf ,1)
({z1 , z, z, . . .})
z1
(z)
= (Nf + 1)
(4.39)
,

Nf
z1 =z
175
a relation which is useful when simplifying the expressions for the meson correlators in Section 5.
We finally note that Eq. (4.39) is consistent with an analogous formula obtained from Schwinger
Dyson equations [5] in the full theory.
5. Meson correlators
In this section we present the detailed analytical predictions of partially quenched scalar and
pseudo-scalar correlation functions in the -regime of QCD. Our calculation is done to the lowest
non-trivial order in the -expansion, and by taking limits we can check that we recover both the
fully quenched and Nf = 2 results at sectors of fixed topological index as reported in Ref. [5].
5.1. Partially quenched correlators with two valence quarks
Of most immediate interest are the correlation functions with two light valence quarks, representing the u and d quarks in QCD. The physical case of low-energy QCD with two very light
quarks (the u and the d quarks) and one heavier quark (the s quark) is an example where one
can be in a mixture of the -regime (with respect to the two light quarks) and the p-regime (with
respect to the s quark). Even if the two light quarks correspond, for the given lattice volume,
to the -regime, it may be convenient to recycle the lattice configurations by considering light
valence quarks that are also in the -regime, but just at different mass values. This requires a
comparison with the results of PQChPT in the -regime that we shall present here.
Furthermore, with the resulting formulae as building blocks one can calculate various other
different types of meson correlators. They can correspond not just to different valence and sea
quark masses but also to an arbitrary number of valence and sea quark flavors. For example, the
SU(3)-singlet scalar correlator is obtained by
3
3
3

S 0 (x)S 0 (0) =
qvi qvi (x)qvj qvj (0) d +
qvi qvi (x)qvi qvi (0) c ,
(5.1)
where the expressions in the r.h.s. are defined below. Note that the contributions inside the sums
can differ due to the non-degeneracy of the valence quark masses.
The connected scalar and pseudo-scalar correlators are defined by

qv1 qv2 (x)qv2 qv1 (0) c

2
U (x)v1 v2 + U (x)v2 v1 + U (x)v2 v1 + U (x)v1 v2
8

U (0)v1 v2 + U (0)v2 v1 + U (0)v2 v1 + U (0)v1 v2 ,

qv1 5 qv2 (x)qv2 5 qv1 (0) c

2
U (x)v1 v2 U (x)v2 v1 + U (x)v2 v1 U (x)v1 v2
8

U (0)v1 v2 U (0)v2 v1 + U (0)v2 v1 U (0)v1 v2 ,
(5.2)
(5.3)
where vi denotes the valence flavor index with the quark mass mvi = xi /V . Note that Eqs. (5.2)
and (5.3) have the same structure as the iso-triplet correlators in the 2-flavor theory of which
masses are mv1 and mv2 .
176
The disconnected correlators are similarly defined

qv1 qv1 (x)qv2 qv2 (0) d

2
U (x)v1 v1 + U (x)v1 v1 U (0)v2 v2 + U (0)v2 v2 ,
4

qv1 5 qv1 (x)qv2 5 qv2 (0) d

2
U (x)v1 v1 U (x)v1 v1 U (0)v2 v2 U (0)v2 v2 .
4
(5.4)
(5.5)
Let us begin with the disconnected scalar correlators. To O( 2 ), we find

eff eff

1-loop
x2
x
2
U v 1 v 1 + U v 1 v 1 U v 2 v 2 + Uv 2 v 2
=
1
4
x1
x2

2
Uv1 v2 Uv2 v1 + Uv2 v1 Uv1 v2 P(12)(21) (x)
2
2F

2
Uv1 v1 Uv1 v1 Uv2 v2 Uv2 v2 P(11)(22) (x)
2
2F

2

1-loop
x1eff 2
=
Uv1 v1 + Uv1 v1 Uv2 v2 + Uv2 v2
4
x1

2
Uv1 v2 Uv2 v1 + Uv2 v1 Uv1 v2 (0|x)

2
2F

2
Uv1 v1 Uv1 v1 Uv2 v2 Uv2 v2 G(0,

0|x).
2
2F
(5.6)
We have consistently set mv1 = mv2 = 0 in the NLO contributions, and

x1eff x2eff
1
=
=1 2
x1
x2
F

Nf

2
Miv |0 G(0, 0|0) ,
(5.7)
which of course needs regularization. Note 1-loop indicates the shift xi xieff and zi zieff
in the arguments of the Bessel functions.
In the same way we obtain the disconnected pseudo-scalar correlation function

qv1 5 qv1 (x)qv2 5 qv2 (0) d

1-loop
2 x1eff 2
=
4
x1

2
Uv1 v2 Uv2 v1 + Uv2 v1 Uv1 v2 (0|x)

2
2F

2
Uv1 v1 + Uv1 v1 Uv2 v2 + Uv2 v2 G(0,

0|x).
2
2F
For the connected correlators, we get
(5.8)
177


2

1-loop
2 x1eff 2
Uv1 v2 + Uv2 v1 + Uv1 v2 + Uv2 v1 Uv2 v1 + Uv1 v2
=
4
x1

2

Uv1 i Uiv
P(2i)(i2) (x) + Uv2 i Uiv
P(1i)(i1) (x)
+
2
1
2
2F
i
2
2
2

Uv1 v2 + Uv2 v1 P(22)(22) (x) + Uv22 v1 + Uv1 v2 P(11)(11) (x)
2
4F

2
Uv1 v1 Uv2 v2 + Uv1 v1 Uv2 v2 P(12)(21) (x) + P(21)(12) (x)
2
4F

2
Uv1 v2 Uv2 v1 Uv2 v1 Uv1 v2 P(22)(11) (x)
2
4F

2
Uv2 v1 Uv1 v2 Uv1 v2 Uv2 v1 P(11)(22) (x),
4F 2
(5.9)

qv1 5 qv2 (x)qv2 5 qv1 (0) c

2

1-loop
2 x1eff 2
Uv1 v2 Uv2 v1 + Uv1 v2 Uv2 v1 Uv2 v1 Uv1 v2
=
4
x1

2

Uv1 i Uiv
P(2i)(i2) (x) + Uv2 i Uiv
P(1i)(i1) (x)
2
1
2
2F
i
2
2
2

Uv1 v2 + Uv2 v1 P(22)(22) (x) + Uv22 v1 + Uv1 v2 P(11)(11) (x)
2
4F

2
Uv1 v1 Uv2 v2 + Uv1 v1 Uv2 v2 P(12)(21) (x) + P(21)(12) (x)
2
4F

2
Uv1 v2 + Uv2 v1 Uv2 v1 + Uv1 v2 P(22)(11) (x)
2
4F

2
Uv2 v1 + Uv1 v2 Uv1 v2 + Uv2 v1 P(11)(22) (x),
(5.10)
4F 2

where the summation over the flavors, denoted by i , has to be taken carefully. For example, in
the sum

Uv1 i Uiv
P(2i)(i2) (x)
1
i
N
lim
Nv1 0
Nf
v1

v2

Uv1 i Uiv
P
Uv1 i Uiv
P(2i)(i2) (x)
(x)
+
lim
(2i)(i2)
1
1
Nv2 0

i

Uv1 i Uiv
P(2i)(i2) (x)
1

= Uv1 v1 Uv1 v1 P(21)(12) (x) Uv1 v1 Uv v P(21)(12) (x)
1 1

+ Uv1 v2 Uv2 v1 P(22)(22) (x) Uv1 v2 Uv2 v1 P22 2 2 (x)

178
Nf

Uv1 i Uiv
P(2i)(i2) (x),
1
(5.11)
the index with a prime i is treated as a different flavor from ith quark but with the same value of
the quark mass. With this, we find the correlators


2

1-loop
2 x1eff 2
Uv1 v2 + Uv2 v1 + Uv1 v2 + Uv2 v1 Uv2 v1 + Uv1 v2
=
4
x1
2
Uv1 v1 Uv1 v1 Uv1 v1 Uv v + Uv2 v2 Uv2 v2 Uv2 v2 Uv v
+
1 1
2 2
2F 2

Uv1 v1 Uv2 v2 Uv v Uv v (0|x)

1 1
2
2F 2
2 2

Uv1 v2 Uv2 v1 + Uv2 v1 Uv1 v2 Uv1 v2 Uv2 v1 Uv2 v1 Uv1 v2 G(0,

0|x)
Nf
2
2
Uv1 i Uiv
+ Uv2 i Uiv
Miv |x ,
2
1
2
2F
i

qv1 5 qv2 (x)qv2 5 qv1 (0) c

2

1-loop
2 x1eff 2
Uv1 v2 Uv2 v1 + Uv1 v2 Uv2 v1 Uv2 v1 Uv1 v2
=
4
x1
2

Uv1 v1 Uv1 v1 Uv1 v1 Uv v + Uv2 v2 Uv2 v2 Uv2 v2 Uv v
1 1
2 2
2F 2

+ Uv1 v1 Uv2 v2 + Uv v Uv v (0|x)
1 1
(5.12)
2 2

2
Uv1 v2 Uv2 v1 + Uv2 v1 Uv1 v2 + Uv1 v2 + Uv2 v1 Uv2 v1 + Uv1 v2 G(0,

0|x)
+
2
2F
Nf
2
2
Uv1 i Uiv
+ Uv2 i Uiv
Miv |x ,
2
1
2
2F
(5.13)
where we have explicitly kept the sea quark mass, in order to possibly apply these equations
outside of the -regime.
With aid of the formulae collected in Appendix A and use of the defining equation (4.13) for
PQ
D , one obtains

eff 2

# $
x
= 2 1
DPQ x1eff , x2eff , zieff
x1

2
2
x2 PQ (x1 , {zi }) x1 PQ (x2 , {zi })
2 2
(0|x)
F x1 x22
+
2 2 2
G(0, 0|x),
F 2 x1 x2
(5.14)
179

qv1 5 qv1 (x)qv2 5 qv2 (0) d
eff 2
2
2 x1
=
x1
x1eff x2eff

x2 PQ (x1 , {zi }) x1 PQ (x2 , {zi })
2
2
(0|x)
2 2
F x1 x22

2 2 PQ
D x1 , x2 , {zi } G(0,
0|x),
2
F

eff 2
PQ eff eff

(x1 , {zi }) PQ (x2eff , {zieff })
x
1
= 2 1
x1
x1eff x2eff

2 PQ (x1 , {zi }) PQ (x2 , {zi }) 2 2
+
+ 2+ 2
2F 2
x1
x2

2

2
(0|x)
+ 2DPQ x1 , x2 , {zi } +
x1 x2

PQ

4
(x1 , {zi }) PQ (x2 , {zi })
2
G(0,
0|x)
2F 2 x1 + x2
Nf

full(Nf ,j )
({zi })
2
x1 PQ (x1 , {zi }) zj
2
2F 2
x12 zj2
j
+

full(Nf ,j )

({zi })
x2 PQ (x2 , {zi }) zj
2
Mj2v |x ,

2
2
x2 zj

qv1 5 qv2 (x)qv2 5 qv1 (0) c
eff 2
PQ eff eff

(x1 , {zi }) PQ (x2eff , {zieff })
1
2 x1
+
=
x1
x1eff + x2eff

2 PQ (x1 , {zi }) PQ (x2 , {zi }) 2 2
+
+ 2+ 2
+
2F 2
x1
x2

2 2
(0|x)
2DPQ x1 , x2 , {zi }
x1 x2

PQ

4
(x1 , {zi }) PQ (x2 , {zi })
2
+
G(0,
0|x)
2F 2 x1 x2
Nf

full(Nf ,j )
({zi })
2
x1 PQ (x1 , {zi }) zj
2
2F 2
x12 zj2
+
(5.15)
(5.16)

full(Nf ,j )

({zi })
x2 PQ (x2 , {zi }) zj
2
Mj2v |x .

+ 2
2
x2 zj
(5.17)
180
We have here three types of -correlators,
(0|x)
=
1 eipx
,
V
p2
(5.18)
p =0
(M
iv |x) =
eipx
1
,
2
V
p + zi /F 2 V
(5.19)
p =0
1 eipx (m20 + p 2 )/Nc
G(0,
0|x) =
.
V
p 4 F(p 2 )
(5.20)
p =0
Note that if the sea quarks are much smaller than the cut-off of ChPT, but still in the p-regime,
one can take the m0 limit. It leads to

Nf
ipx

4
Mii2 eipx
1
e
1
=
G(0,
0|x)
(0|x)
+
+
O
Mii , (5.21)

N
f
1
V
Nf
p4
4(
p
)
i
p =0
2
2
i
p +Mii
where the well-known double pole contribution appears due to a mismatch of the sea and valence
quark masses. Further simplification is possible when all the sea quarks are in the -regime,
(M
iv |x) (0|x),
G(0,
0|x)
(0|x).
Nf
(5.22)
An interesting special case is the degenerate limit x1 = x2 , where the above formulae become

eff 2
PQ (x1eff , {zieff })
x
= 2 1
x1

PQ
2
x (x1 , {zi }) PQ (x1 , {zi })
(0|x)
2
x1
F
2 2 2
G(0, 0|x),
F 2 x12

qv1 5 qv1 (x)qv1 5 qv1 (0) d
eff 2
x
2
= 2 1
x1
(x1eff )2

2 x PQ (x1 , {zi }) PQ (x1 , {zi })
(0|x)
2
x1
F
+
2 2 PQ (x1 , {zi })
G(0, 0|x),
F2

eff 2
x PQ (x1eff , {zieff }) 2 2 2
x
= 2 1
2 2 (0|x)
x1
F x1
2 2 PQ (x1 , {zi })
G(0, 0|x)
x1
F2
(5.23)
(5.24)

Nf

full(Nf ,j )

({zi })
2
4
x1 PQ (x1 , {zi }) zj
Mj2v |x ,

2
2
2
2F
x1 zj
j

qv1 5 qv1 (x)qv1 5 qv1 (0) c
eff 2 PQ eff eff
(x1 , {zi }) 2 2 PQ (x1 , {zi })
x
= 2 1
+ 2
(0|x)
x1
F
x1eff
2 2 x PQ (x1 , {zi })
G(0, 0|x)
F2
Nf

full(Nf ,j )

({zi })
2
4
x1 PQ (x1 , {zi }) zj
Mj2v |x .

2
2
2F 2
x1 zj
j
181
(5.25)
(5.26)
5.2. Singlet and flavored meson correlators

To compare with lattice QCD data, the case with arbitrary Nv -flavor valence quarks is interesting. In particular, Nv can be different from Nf , the number of sea quarks. In this paper we have
aimed at the -regime predictions, and of course we always take the valence quarks to be in that
regime. With the qualification mentioned in Section 2, one can also be more general, and consider the sea quarks to be just marginally in the -regime, or even entirely in the p-regime. The
finite volume is an L3 T box, where L and T are the spacial and temporal extents, respectively.
The zero-momentum projections of singlet and (Nv2 1)-plet correlators are then obtained

d 3 x S 0 (x)S 0 (0)

d 3 x Nv qv1 qv1 (x)qv1 qv (0) c + Nv2 qv1 qv1 (x)qv1 qv1 (0) d

PQ eff
eff
x PQ (x1eff , {zieff })
x1eff 2
2 (x1 , {zi })
=L
Nv
Nv
x1

PQ
PQ
(x1 , {zi })
2 4Nv 2
2 x (x1 , {zi })
+
2N
a(t/T )
x1
2F 2
x12

2
PQ (x1 , {zi })
2
2
4Nv 2 b(t/T )
4Nv
x1
2F 2
x1
N

full(Nf ,j )
f
({zi })
Nv 2
4
x1 PQ (x1 , {zi }) zj
+
cj (t/T ),
2F 2
x12 zj2

d 3 x P 0 (x)P 0 (0)

d 3 x Nv qv1 5 qv1 (x)qv1 5 qv1 (0) c + Nv2 qv1 5 qv1 (x)qv1 5 qv1 (0) d

= L3 2
x1eff
x1
2
Nv
PQ (x1eff , {zieff })
x1eff
Nv2
2
(x1eff )2
(5.27)
182

2 4Nv PQ (x1 , {zi })
2F 2

PQ
x (x1 , {zi }) PQ (x1 , {zi })
2Nv2
a(t/T )
x1

PQ
x PQ (x1 , {zi })
2
2 (x1 , {zi })
4N
4N
b(t/T )
v
v
2F 2
Nf

full(Nf ,j )
({zi })
Nv 2
4
x1 PQ (x1 , {zi }) zj
+
cj (t/T ),
2F 2
x12 zj2
(5.28)

d 3 x S a (x)S b (0)

ab
d 3 x qv1 qv1 (x)qv1 qv1 (0) c

2
)

PQ
ab 3 2 x1eff 2 x (x1eff , {zieff }) 2 2 2
2 2 a(t/T )
=
L
2
x1
F x1
2 2 PQ (x1 , {zi })
b(t/T )
x1
F2
*
Nf

full(Nf ,j )
({zi })
4
x1 PQ (x1 , {zi }) zj
2
cj (t/T ) ,
+
2F 2
x12 zj2
(5.29)

d 3 x P a (x)P b (0)

ab
d 3 x qv1 5 qv1 (x)qv1 5 qv1 (0) c

2
)
eff 2 PQ eff eff
(x1 , {zi }) 2 2 PQ (x1 , {zi })
x
ab
a(t/T )
=
+ 2
L3 2 1
2
x1
F
x1eff
2 2 x PQ (x1 , {zi })
b(t/T )
F2
*
Nf

full(Nf ,j )
({zi })
2
4
x1 PQ (x1 , {zi }) zj
cj (t/T ) ,
2F 2
x12 zj2
j
+
(5.30)
where a(t/T ), b(t/T ) and cj (t/T ) are defined by

t
1 2
1
T
2
12

eipx
1
b(t/T ) d 3 x G(0,
0|x) = d 3 x
Nf
V
4
p =0 p ( i
2

a(t/T )
=
d 3 x (0|x)
1 T
=
Nf 2
t
1
T
2
T
2
12
1
)
p +Mii2
(5.31)
Nf

M2
i
ii
Nf2
T3
24
2
2

1
t
t
1
+ O Mii4 ,
T
T
30

cosh(Mj v (T /2 t))
1
Mj2v |x =
d 3x
2Mj v sinh(Mj v T /2) Mj2v T

2
2

T
T3
1
1
t
1 2
t
t
=
+ Mj2v
1
2
T
2
12
24
T
T
30
4
+ O Mj v ,
183
(5.32)
cj (t/T )
(5.33)
where Mj2v = mj /F 2 and Mii2 = 2mi /F 2 .

In Figs. 2, 3 and 4, we plot, as examples, the flavored pseudo-scalar and scalar correlators (we simply denote Eqs. (5.29) and (5.30) as S a (t) and P a (t) for a = b). We use
= (250 MeV)3 , F = 93 MeV, x1eff /x1 = 1 and L = T = 2 fm as inputs.
5.3. WardTakahashi identities
In this subsection, we check that the above results satisfy the WardTakahashi identities under
the chiral rotation of the degenerate Nv valence quarks. For the singlet chiral rotation, one obtains

A0 (x) 2mP 0 (x) 2iNv (x) O(0) = O(0) (x),
(5.34)
for any operator O(x), where
A0 (x) =
Nv

qv (x) 5 qv (x),
P 0 (x) =
v=1
(x) =
Nv

qv (x)5 qv (x),
v=1
1
Tr F F (x),
16 2
(5.35)
and O denotes the chiral variation of O. Note that Eq. (5.34) holds not only in -vacuum but
also in a fixed topological sector. The identities for O(x) = P 0 (x) and O(x) = (x) and their
integration over the volume give an equation

N 2 2 S 0 (x)
d 4 x P 0 (x)P 0 (0) = v2
m
m V
=
PQ
Nv2 2 Nv (x1eff , {zieff })x1eff
+
,
mx1
m2 V
v
where S 0 (x) = N
v=1 qv (x)qv (x), which coincides with Eq. (5.28). Note here we used

dt a(t/T ) = dt b(t/T ) = dt cj (t/T ) = 0.
In the same way, the flavored identity

Aa (x) 2mP a (x) O(0) = a O(0) (x),
for O(x) = P a (x) and O(x) = Aa (x) gives
(5.36)
(5.37)
(5.38)
184
Fig. 2. Pseudo-scalar (top) and scalar (bottom) flavored correlators with both valence and sea quarks in the -regime,
here for = 0. The sea quark masses are fixed to 5 MeV. We use = (250 MeV)3 , F = 93 MeV, x1eff /x1 = 1 and
L = T = 2 fm as inputs.
PQ

S 0 (x) (x1eff , {zieff })x1eff
d 4 x P a (x)P a (0) =
,
=
2Nv m
2mx1
(5.39)
which is also consistent with Eq. (5.30). Here the operator with the superscript a has the
form of qv (x) T a qv (x) with some gamma matrix , as in the conventional notation, where T a
denotes the ath generator of SU(Nv ) group.
One can also confirm the similar identity [33]

d 4 x S a (x)S a (0) =
1
0
S (x) =
2Nv V mv
x1eff
x1

# $
PQ x1eff , zieff ,
2mv1
(5.40)
185
Fig. 3. The same as Fig. 2 but with = 2. Note that the pseudo-scalar and scalar correlation functions almost sum up to
zero.
is consistent with Eq. (5.29). Since the asymptotic form of the partially quenched condensate in
the chiral limit is known [7]
PQ (x1 , {zi }) ||
+ O(x1 ),
x1
it is not difficult to see that the known quenched identities for the Nv = 2 case [33],

4 2
4||
d 4 x P 0 (x)P 0 (0) 4 S a (x)S a (0) 2 2 ,
mv V
mv V

4||
d 4 x P a (x)P a (0) 4 S a (x)S a (0) 2 ,
mv V
also hold in the limit mv 0 in this partially quenched theory.
(5.41)
(5.42)
(5.43)
186
Fig. 4. The sea quark mass dependence of pseudo-scalar (top) and scalar (bottom) flavored correlators with the valence
quark mass fixed at 3 MeV. The parameters are chosen the same as for Figs. 2 and 3. Here we assume that we can still
ignore the m0 and terms for sea quark masses up to 15 MeV.
5.4. Quenched and full degenerate Nf flavor limits

In this subsection, we show how to reproduce the known results of both the fully quenched
theory and the unquenched theory by taking the zi limit and the x = z1 = z2 = = z
limits, respectively. The valence quarks are chosen to be degenerate in the -regime.
First consider the quenched limit of connected correlators Eqs. (5.25) and (5.26),
que

1
= lim
S a (x)S a (0)
zi 2
eff 2
que
x (x1eff )
2 x1
=
x1
2

que
2
2 2
2 (x1 )
G(0,
0|x)
,
(0|x)
x1
2F 2
x12
a
que

1
P (x)P a (0)
= lim qv1 5 qv1 (x)qv1 5 qv1 (0) c
zi 2
eff 2 que eff
(x1 )
2 x1
=
x1
2x1eff

que
2
2
2x (x1 )
G(0, 0|x) .
2 1 + 2 (0|x) +
2F 2
x1
+
187
(5.44)
(5.45)
Noting
(0|x)
=
1 eipx
,
V
p2
p =0

1 1 eipx m20 eipx
G(0, 0|x) =
+
,
V
Nc
p4
p2
(5.46)
p =0
in the quenched limit, one can see that Eqs. (5.44) and (5.45) agree with the quenched results
in [5].
Next we construct Nv = 1 singlet correlation functions in the quenched limit,
0
que
S (x)S 0 (0)

= lim qv1 qv1 (x)qv1 qv1 (0) c + qv1 qv1 (x)qv1 qv1 (0) d
zi

que
x1eff 2 x (x1eff )
2
+ 1 + eff
x1
(x1 )2

que
que
2
4 2 2 (x1 ) 2x (x1 )
(0|x)
2 +
x1
2F 2
x1

que
4 (x1 ) 4 2
2 G(0, 0|x) ,
x1
x1
0
que
0
P (x)P (0)

= lim qv1 5 qv1 (x)qv1 5 qv1 (0) c qv1 5 qv1 (x)qv1 5 qv1 (0) d
= 2
(5.47)
zi
que

x1eff 2 (x1eff )
2
x1
x1eff
(x1eff )2

que
que
2
4 2 2 (x1 ) 2x (x1 )
+
(0|x)
4+ 2
x1
2F 2
x1

que
4x (x1 )
4 2
+
+ 4 + 2 G(0, 0|x) ,
x1
= 2
(5.48)
which are also equivalent to the results of [5].

The full degenerate Nf -flavor limit (x z = z1 = z2 = ) of the connected correlators in
the -regime are
a
full
S (x)S a (0)

1
= qv1 qv1 (x)qv1 qv1 (0) c x=z =z ==z
1
2
2
188

N
2 2 2 2 f (z)
zeff 2 x PQ (x, {zeff , zeff , . . .})
=
+
eff 2F 2 z2
z
2
Nf z
x=z

Nf
PQ
(z)
x (x, {z, z, . . .})
Nf
+
(0|x),

z
x=z
a
full
P (x)P a (0)

1
= qv1 5 qv1 (x)qv1 5 qv1 (0) c x=z =z ==z
1
2
2
eff 2 Nf eff
(z )
z
= 2
z
2zeff

2PQ (z, {z, z, . . .}) 2x PQ (x, {z, z, . . .})
2
+

Nf
2F 2
x=z

Nf
PQ
(z)
x (x, {z, z, . . .})
+ Nf
+
(0|x).

z
x=z

(5.49)
(5.50)
Nf
2 = M 2 = 0, G(0,
0|x) = (0|x)/N
Note here that we set Miv
f , (z)/ is the full degenerate
vv
Nf -flavor condensate defined by Eq. (4.35), and zeff is given by Eq. (3.10). To eliminate the
partially quenched expression, PQ / , we use Eqs. (4.34), (4.38) and (4.39) to obtain
a
full
S (x)S a (0)
Nf eff 2
(z )
zeff 2
=
2(Nf 1) z
Nf eff

N
2
f (zeff )
1 (z )
+ 1 + eff 2
Nf
Nf
zeff
(z )

N
2
2 (3Nf 2) f (z)
2
2
2
+
N
f
Nf
z
z2
2(Nf2 1)F 2
Nf 2

+ Nf2
f (z)
1
+
Nf
f (z)
1 (0|x),
a
full
P (x)P a (0)
eff 2 Nf eff
(z )
2 z
=
z
2zeff

N
2
2
f (z)
2
4
+
3N
+ Nf2 4
N
+
f
f
z
z2
2(Nf2 1)F 2
N

N

f (z) 2
1 f (z)
+ Nf2 + 2
+
(0|x),
Nf
which agree with those in the full Nf -flavor theory.
(5.51)
(5.52)
In the same way, one can see that the singlet correlators
full
0
S (x)S 0 (0)

= Nf qv1 qv1 (x)qv1 qv1 (0) c + Nf2 qv1 qv1 (x)qv1 qv1 (0) d

eff 2
x PQ (x, {zeff , zeff , . . .})
z
= 2
Nf
eff
z
x=z
PQ eff
eff , zeff , . . .})

(z
,
{z
Nf2

N
2
f (z)
2
1
(0|x),
4
N
+
f
z
2F 2
full
0
P (x)P 0 (0)

= Nf qv1 5 qv1 (x)qv1 5 qv1 (0) c Nf2 qv1 5 qv1 (x)qv1 5 qv1 (0) d
eff 2
N
2
f (zeff )
z
2
N
= 2
Nf
f eff 2
z
zeff
(z )

PQ
2
x (x, {z, z, . . .})
2
+
4 Nf 1
(0|x),

2F 2
x=z
189
(5.53)
(5.54)
are also consistent with the known expressions in [5].

6. Conclusions
In this paper, we have discussed partially quenched chiral perturbation theory (PQChPT) in the
-regime, and in the mixed and p-regime. Using the 1-loop improved chiral condensate and its
derivative as building blocks, we have calculated various zero-mode group integrals in the replica
limit. These integrals are necessary for the computation of mesonic correlation functions in the
partially quenched theory. We have also derived a non-trivial identity which is a consequence of
unitarity of the graded or replicated group.
With these zero-mode integrals and the Feynman rules for the non-zero modes, we have calculated the mesonic correlation functions for both connected and disconnected pseudo-scalar and
scalar channels with non-degenerate quark masses, both of the valence and sea kind. Among others, our results can be applied to the mesons that consist of two non-degenerate valence quarks.
For a demonstration, we have plotted the flavored pseudo-scalar and scalar correlators with a realistic choice of input parameters. As expected, they show a non-trivial valence (sea) quark mass
dependence with a fixed sea (valence) quark mass.
These meson correlators were shown to have the correct quenched and degenerate full
Nf -flavor theory limits. We have not addressed the implicit flavor dependence of , or F , the
fundamental parameters in the infinite volume limit. In order to complete the smooth connection
among the theories with different number of flavors, one has to match the value of them [28
32]. The flavor dependence is expected to be rather weak, but the matching is interesting and
important for the future works.
Our results are useful for the analysis of unquenched lattice QCD simulations in many ways.
The various valence quark masses can be used for each set of fixed sea quark masses. Even if the
physical pions are just barely in the -regime, one can put the valence pions very safely in the
190
-regime and compare numerical data with our formulae for the partially quenched correlation
functions.
Acknowledgements
We would like to thank J.J.M. Verbaarschot and K. Splittorff for fruitful discussions.
H.F. wishes to thank the members of Niels Bohr Institute for warm hospitality during his stay.
The work of H.F. was supported by a Grant-in-Aid of the Japanese Ministry of Education
(No. 18840045), and the work of P.H.D. was supported in part by the European Community
Network ENRAGE (MRTN-CT-2004-005616).
Appendix A. Summary of group integrals
Here we summarize the group integrals in the replica limit, which are necessary for the meson
correlators. See Section 4.1 for the details. The formulae for one valence index are,
PQ (x, {zi })
1
=
Uvv + Uvv
,
2
x PQ (x, {zi }) PQ (x, {zi })

1
2
=
Uvv + Uvv
,
4

1
= ,
Uvv Uvv
2
x

1
PQ (x, {zi }) 2
2
=
Uvv Uvv
+ 2,
4
x
x

1
1

2
2
Uvv Uvv =
Uvv + Uvv
Uvv Uvv
4
4
PQ
x (x, {zi }) PQ (x, {zi }) PQ (x, {zi }) 2
+
2.
=
x
x
For two valence indices,

1
Uv1 v1 + Uv1 v1 Uv2 v2 + Uv2 v2 = DPQ x1 , x2 , {zi } ,
4

1
2
,
Uv1 v1 Uv1 v1 Uv2 v2 Uv2 v2 =
4
x1 x2

2 2
Uv1 v1 Uv2 v2 + Uv1 v1 Uv2 v2 = 2DPQ x1 , x2 , {zi } +
,
x1 x2
(A.1)
(A.2)
(A.3)
(A.4)
(A.5)
(A.6)
(A.7)
(A.8)
where DPQ is defined in Eq. (4.13). Similarly,

2 1
2
1
Uv1 v2 Uv2 v1 =
Uv2 v1 Uv1 v2
4
4
1

1
Uv1 v2 Uv2 v1 =
Uv2 v1 Uv1 v2
=
2
2

PQ
(x1 , {zi })
PQ (x2 , {zi })
1
x
,
x
= 2
1
2
x1 x22
(A.9)
2

1 2
Uv1 v2 + Uv2 v1 = 0,
4

1
4

PQ (x1 , {zi })
PQ (x2 , {zi })
1
x1
= 2
x2
,
x1 x22
191
(A.10)
(A.11)
are obtained.
References
[1] J. Gasser, H. Leutwyler, Phys. Lett. B 188 (1987) 477.
[2] H. Neuberger, Phys. Rev. Lett. 60 (1988) 889.
[3] F.C. Hansen, Nucl. Phys. B 345 (1990) 685;
F.C. Hansen, H. Leutwyler, Nucl. Phys. B 350 (1991) 201.
[4] P. Hasenfratz, H. Leutwyler, Nucl. Phys. B 343 (1990) 241.
[5] P.H. Damgaard, M.C. Diamantini, P. Hernandez, K. Jansen, Nucl. Phys. B 629 (2002) 445, hep-lat/0112016.
[6] P.H. Damgaard, P. Hernandez, K. Jansen, M. Laine, L. Lellouch, Nucl. Phys. B 656 (2003) 226, hep-lat/0211020.
[7] P.H. Damgaard, K. Splittorff, Nucl. Phys. B 572 (2000) 478, hep-th/9912146.
[8] P.H. Damgaard, Phys. Lett. B 476 (2000) 465, hep-lat/0001002.
[9] P.H. Damgaard, Nucl. Phys. B 608 (2001) 162, hep-lat/0105010.
[10] L. Giusti, C. Hoelbling, M. Luscher, H. Wittig, Comput. Phys. Commun. 153 (2003) 31, hep-lat/0212012.
[11] W. Bietenholz, T. Chiarappa, K. Jansen, K.I. Nagai, S. Shcheredin, JHEP 0402 (2004) 023, hep-lat/0311012.
[12] L. Giusti, P. Hernandez, M. Laine, P. Weisz, H. Wittig, JHEP 0401 (2004) 003, hep-lat/0312012.
[14] K. Ogawa, S. Hashimoto, Prog. Theor. Phys. 114 (2005) 609, hep-lat/0505017.
[15] H. Fukaya, S. Hashimoto, K. Ogawa, Prog. Theor. Phys. 114 (2005) 451, hep-lat/0504018.
[16] W. Bietenholz, S. Shcheredin, Nucl. Phys. B 754 (2006) 17, hep-lat/0605013.
[17] F. Bernardoni, P.H. Damgaard, H. Fukaya, P. Hernandez, in preparation.
[18] C.W. Bernard, M.F.L. Golterman, Phys. Rev. D 49 (1994) 486, hep-lat/9306005.
[19] S.R. Sharpe, N. Shoresh, Phys. Rev. D 64 (2001) 114510, hep-lat/0108003.
[20] H. Fukaya, et al., JLQCD Collaboration, PoS LAT2006 (2006) 050, hep-lat/0610024.
[21] H. Fukaya, et al., JLQCD Collaboration, Phys. Rev. Lett. 98 (2007) 172001, hep-lat/0702003.
[22] H. Fukaya, et al., arXiv: 0705.3322 [hep-lat].
[23] K. Splittorff, J.J.M. Verbaarschot, Phys. Rev. Lett. 90 (2003) 041601, cond-mat/0209594.
[24] Y.V. Fyodorov, G. Akemann, JETP Lett. 77 (2003) 438, Pisma Zh. Eksp. Teor. Fiz. 77 (2003) 513, cond-mat/
0210647.
[25] F. Bernardoni, P. Hernandez, arXiv: 0707.3887 [hep-lat].
[26] P.H. Damgaard, K. Splittorff, Phys. Rev. D 62 (2000) 054509, hep-lat/0003017.
[27] J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 540 (1999) 317, hep-th/9806110.
[29] P. Hernandez, M. Laine, JHEP 0610 (2006) 069, hep-lat/0607027.
[30] T. DeGrand, R. Hoffmann, S. Schaefer, Z. Liu, Phys. Rev. D 74 (2006) 054501, hep-th/0605147.
[31] T. DeGrand, Z. Liu, S. Schaefer, Phys. Rev. D 74 (2006) 094504, hep-lat/0608019;
T. DeGrand, Z. Liu, S. Schaefer, Phys. Rev. D 74 (2006) 099904, Erratum.
[32] P. Hasenfratz, D. Hierl, V. Maillart, F. Niedermayer, A. Schafer, C. Weiermann, M. Weingart, arXiv: 0707.0071
[hep-lat].
[33] R.G. Edwards, U.M. Heller, R. Narayanan, Phys. Rev. D 59 (1999) 094510, hep-lat/9811030.

A new class of rank breaking orbifolds

G. von Gersdorff
Department of Physics and Astronomy, Johns Hopkins University, 3400 N Charles Street, Baltimore, MD 21218, USA
Received 22 May 2007; received in revised form 10 August 2007; accepted 1 October 2007
Abstract
We describe field-theory T 2 /Zn orbifolds that offer new ways of breaking SU(N ) to lower rank subgroups. We introduce a novel way of embedding the point group into the gauge group, beyond the usual
mapping of torus and root lattices. For this mechanism to work the torus Wilson lines must carry nontrivial
t Hooft flux. The rank lowering mechanism proceeds by inner automorphisms but is not related to continuous Wilson lines and does not give rise to any associated moduli. We give a complete classification of
all possible SU(N ) breaking patterns. We also show that the case of general gauge group can already be
understood entirely in terms of the SU(N ) case and the knowledge of standard orbifold constructions with
vanishing t Hooft flux.
PACS: 11.10.Kk; 11.15.-q; 11.25.Mj; 12.10.-g
Keywords: Orbifolds; Gauge symmetry breaking; Rank reduction
1. Introduction
Orbifolds [1] are one of the most explored avenues in the study of string theory compactifications. Not only do they possess phenomenologically appealing features such as chirality,
reduced supersymmetry, and a built-in gauge symmetry baking mechanism, they are also extremely tractable and provide a welcome starting point to study more complicated vacua though
string theorys many dualities. Notwithstanding, the classification of all orbifold vacua of the
(heterotic) string seems to be an extremely difficult task, and the search of the standard model, or
its supersymmetric extension, in this vast landscape of vacua has only been partially successful.
E-mail address: gero@pha.jhu.edu.
doi:10.1016/j.nuclphysb.2007.10.003
G. von Gersdorff / Nuclear Physics B 793 (2008) 192210
193
A more modest approach, justified in its own right, are orbifold grand unified theories (orbifold GUTs). It is quite conceivable that some of the extra dimensions are larger than others, and
intermediate models with effectively fewer extra dimensions could be realized in nature. In view
of this, a lot of effort has been made to construct five- and six-dimensional models that break the
GUT group by orbifolding down to the SM [37]. Some intermediate 6d models appearing as
particular compactification limits of the heterotic string have been described in Ref. [2].
A challenge in obtaining the standard model gauge group by orbifolding is the fact that the
simplest consistent choices for the twists do not reduce the rank of the gauge group. In heterotic
string theory, the anomaly-free gauge groups have rank 16 while the Standard Model only has
rank 4. Rank reduction usually proceeds through one of the following mechanisms
Continuous Wilson lines [20,21]: A given orbifold vacuum can possess a nontrivial moduli space in the gauge sector, i.e., flat directions in the tree level potential for the extra
dimensional components (A4,5,... ) of the gauge bosons. The latter typically transform in nonadjoint representations of the gauge group left unbroken by the orbifolding. By obtaining
vacuum expectation values they can break the gauge symmetries further, thereby reducing
its rank. From a four-dimensional (4d) point of view, this is nothing but the standard Higgs
mechanism. This idea has been applied in the context of electroweak symmetry breaking and
is often referred to as gauge-Higgs unification [817]. The flatness is lifted at loop level
by a finite and calculable potential [9], giving rise to a discrete set of vacua. Unfortunately,
in many circumstances, the vacuum calculated this way actually corresponds to a particular point in moduli space where the rank of the gauge group is restored [12,13]. Moreover,
some Higgs mass terms localized at the fixed point are unprotected by the surviving gauge
symmetry [14,16] and can destroy the finiteness and predictivity of the model.
GreenSchwarz mechanism: If the unbroken gauge group contains anomalous U (1) factors,
the latter can be spontaneously broken by an orbifold version [18] of the GreenSchwarz
mechanism [19]. This mechanism is realized, e.g., in the model of Ref. [11], where the rank6 group U (3)2 was broken to the Standard Model by the presence of two anomalous U (1)
symmetries.
Additional Higgs multiplets at the fixed points, as, e.g., in Ref. [4].
Outer automorphisms. A particular choice of the gauge twists, corresponding to a symmetry
of the Dynkin diagram of the associated Lie algebra, can break the rank. There are only
finitely many possibilities.
In this paper we want to introduce a new way to break the rank of the gauge group by orbifolding. We will mainly restrict ourselves to T 2 /Zn orbifolds with gauge group SU(N ) and will
comment on generalizations to higher dimensions and other gauge groups in Section 4. An orbifold is specified by the gauge twists associated to translations and rotations of the underlying
torus lattice. The space-time translations commute, and so must the corresponding twists. However, in a pure gauge theory, the fields transform in the adjoint representation, and the twists need
only commute up to an element of the center of the group. This yields nontrivial gauge bundels on
the torus which still have a flat gauge connection (i.e., the corresponding field strength vanishes)
[22].1 The center of SU(N ) is isomorphic to ZN . Hence, there are N physically different disconnected vacua, or, more precisely, the moduli space consists of N disconnected components. The
1 For recent work on SU(N ) gauge bundles in string theory see Ref. [23].
194
nontrivial statement we make in this paper is that one can orbifold these configurations. The rotations do not commute with the translations, and the resulting non-Abelian group structure must
be realized on the associated gauge twists. This puts tight constraints on the allowed Wilson
lines, and it is a nontrivial result that nontrivial toron configurations possess compatible orbifold
twists. In this paper, we both show that such twists exist and give a complete classification. Since
the distinction to the standard orbifold construction is quite essential, let us dwell a little more
on this point. In the standard approach, lattice translations are realized by shift vectors, i.e., the
corresponding holonomies exactly commute and can be realized as elements of the same Cartan
torus. The rotations of the torus lattice are then realized by an element of the Weyl group (rotations of the root lattice). Here, instead, the lattice translations are already realized as rotations
of the root lattice, in a way that makes it impossible to choose a Cartan torus such that both of
them simultaneously become shifts. Consequently, the orbifold twists associated to the rotations
of the torus lattice cannot be related to any symmetry of the root lattice used to define the torus
holonomies.
The paper is organized as follows. In Section 2 we review the nontrivial flat SU(N ) gauge
bundles on the two-torus, give an explicit form for the holonomies, and describe their symmetry
breaking patterns. We also explain how other gauge groups can be treated once the SU(N ) case
is known. These gauge bundles are orbifolded in Section 3. In Section 3.1 we treat first the case
m = 0. This does not involve any new concepts, but we include it here for completeness and
comparison. Also, in Appendix B we compute the moduli space for this case. In Section 3.2 we
calculate the orbifold twists for the generic case, making use of the results obtained in Section 2
and Section 3.1. We discuss some generalizations and applications in Section 4 and present our
conclusions in Section 5.
2. Breaking SU(N) on T 2 : Torons
In this section we would like to recall t Hoofts toron configurations [22]. These are simply
flat SU(N ) gauge bundles on the torus, which can be characterized by their holonomies. Upon
shifts in the torus lattice
zz+
(2.1)
gauge fields are identified up to gauge transformations 2

AM (z + ) = T AM (z)T1 .
(2.2)
It is clearly sufficient to restrict to the two lattice-defining base vectors 1,2 . As lattice translations
commute, the commutator of the two transition functions has to act as the identity.
T1 T2 T11 T21 = e2i N .
m
(2.3)
On the right-hand side we have allowed for a general element of the center of the group, which,
for SU(N ), equals ZN . Such a gauge transformation indeed acts trivially on the adjoint representation the gauge fields transform in. The integer quantity m is called the t Hooft non-Abelian
flux. The boundary conditions, Eq. (2.2), allow for vanishing gauge potentials, hence the vacuum
corresponds to vanishing field strength. We stress that it is in principle possible to simultaneously
2 We make use of the fact that we can choose a gauge where the transition functions are z-independent, see, e.g.,
Ref. [24].
195
diagonalize the matrices T1 and T2 in the adjoint.3 For nonzero m, it is not possible to represent
both Ti as elements of the same Cartan torus. It is, however, possible to choose a Cartan torus
left fixed (though not pointwise fixed) by both Ti . As a consequence, one can realize the Ti as
Weyl group elements w.r.t. the same Cartan subalgebra.
The flux m (more precisely the phase appearing on the r.h.s. in Eq. (2.3)) labels the equivalence
classes of the transition functions and determines the vacua of the theory. We would like to find
the unbroken subgroup for each vacuum, i.e., we are looking for the generators that are left
invariant by the action of the Ti :
Ti T Ti = T .
(2.4)
For fixed m, there is still a continuous degree of freedom in choosing the Ti , even within the
gauge where the transition functions are constants: If, for a particular solution to Eq. (2.3), the
unbroken subgroup H is nontrivial, one can always turn on Wilson lines in the Cartan torus of H
and still obtain a solution with the same value for m. Such an additional Wilson line will lead
to a different subgroup H , however, the rank of H and H must remain the same. This freedom
is related to the fact that each vacuum will in general possess a nonzero moduli space, i.e., flat
directions in the potential for the extra dimensional components of A.
To describe the solutions, one decomposes N and m according to their greatest common
divisor K = g.c.d(N, m). Explicit solutions to Eq. (2.3) are then given by [24,25]
T1 = QN/K 1K ,
(L1)/2
(QL )j k = qL
T2 = (RN/K )m/K 1K ,
(2.5)
j,k1 ,
(L1)/2+j 1
(RL )j k = qL
j,k ,
(2.6)
where qL = exp(2i/L). The index on Q, R and 1 indicates the dimensionality of the matrices
and the Kronecker is assumed to be periodic. The matrices Q and R satisfy
QR = qRQ,
QL = R L = ()L1 1.
(2.7)
Hence, Eqs. (2.5) and (2.6) are a particular solution to Eq. (2.3). It can then be shown that
the twists QN/K and (RN/K )m/K break SU(N/K) completely [24]. Writing the generators of
SU(N ) as
TN {TN/K TK , 1N/K TK , TN/K 1K }.
(2.8)
We immediately read off that the unbroken subgroup is generated by 1N/K TK and, thus,
is SU(K). The most general solution to Eq. (2.3) can then be obtained by replacing the unit
matrices in Eqs. (2.5) and (2.6) with commuting Wilson lines of SU(K), which one can take to
be elements of the same Cartan torus:
T1 = QN/K exp(2iW1 ),
T2 = (RN/K )m/K exp(2iW2 ).
(2.9)
The shift vectors W1 and W2 are elements of the Cartan subalgebra of SU(K). Nontrivial SU(K)
Wilson lines further break SU(K), but do not reduce its rank. In summary, a toron configuration
with SU(N ) flux m can be decomposed into a toron configuration with SU(N/K) flux m/K and
an SU(K) configuration with vanishing flux.
3 For an explicit diagonal basis see Ref. [24].
196
However, we would like to stress here that different SU(K) Wilson lines, strictly speaking,
do not correspond to different physical theories. The reason is that the above mentioned flat
directions are lifted at the quantum level and two such theories will dynamically evolve to the
same vacuum. One can always perform a field redefinition, corresponding to a nonperiodic gauge
transformation that removes the continuous Wilson line but generates a vacuum expectation value
(VEV)
A4 = W1 ,
A5 = W2 .
(2.10)
One sees that such a field redefinition induces a shift along a flat direction. In other words, a
theory with nonzero Wilson line and a given point in the moduli space is equivalent to a vanishing
Wilson line and a shifted point in moduli space. The degeneracy of the flat directions is lifted
at the quantum level. The effective potential clearly only depends on the sum of the Wilson
line induced background, Eq. (2.10), and the explicit background, and the true vacuum of two
theories with different continuous Wilson lines coincide. It is important to realize that there is no
analogous field redefinition that could change the value of m4 : Two vacua with different m are
truly disconnected.
The natural question to ask is whether all this can be generalized to gauge groups other than
SU(N ). This question has been extensively discussed in Ref. [26], see also Refs. [2729]. Here
we only give some heuristic arguments and some examples. A necessary and sufficient condition for the existence of nontrivial t Hooft flux is that the group possesses nontrivial center.5
This is true for the SO(N ) and Sp(2N ) groups, as well as for the exceptional groups E6 and E7 .
The center can always be embedded in suitable SU(N ) subgroups [26], and, hence, the above
construction can be carried out straightforwardly. Trivial examples are the groups SO(3), SO(4),
SO(6), and Sp(2) that are actually isomorphic to some special unitary groups. For a nontrivial example take SO(8) whose center is C = Z2 Z2 . Consider now the maximal subgroup
SU(2)4 SO(8). By inspection of the branching rules for the SO(8) irreducible representations
8v and 8s , one can see that a suitable parametrization of the two Z2 s of the center is
c1 = (1, 1, 1, 1),
c2 = (1, 1, 1, 1),
(2.11)
where 1 represent the center of the corresponding SU(2) factor. The branching of the adjoint
is
28 (3, 1, 1, 1) + (1, 3, 1, 1) + (1, 1, 3, 1) + (1, 1, 1, 3) + (2, 2, 2, 2).
(2.12)
It can be directly verified that C acts trivially on the 28, as it must. For a given c C, particular
solutions for T1 (c) and T2 (c) can now be constructed by making use of the results for SU(2).
While any pair Ti (c) clearly projects out two of the four triplets, the action on the fourfold doublet
requires a more careful analysis. Take, for instance c = c1 , then the standard solution acts on the
(2, 2, 2, 2) as
T1 = 1 1 1 1,
T2 = 3 3 1 1.
(2.13)
The two twists can be diagonalized simultaneously. There are four eigenstates, each transforming
as (2, 2) of the surviving SU(2)2 . One of these eigenstates has unit eigenvalue on both Ti , and
4 By enforcing such a field redefinition to, say, remove the Wilson line T , the other transition function would no longer
1
remain constant.
5 More precisely, the center of the universal cover, which is isomorphic to the fundamental group of the adjoint representation.
197
Fig. 1. The four different Zn orbifold geometries in 6d, corresponding to n = 2, 3, 4, 6 (from left to right). We show
the embedding of the orbifold fundamental domain (shaded) in the torus (thin line) as well as the fixed points (dots).
The shaded regions have to be folded over the center line and the edges (thick lines) have to be identified. The resulting
geometries are pillows with three or four corners. Note that the edges correspond to nonsingular bulk points.
hence the branching rule of the adjoint under the breaking SO(8) SU(2)2 reads
28 (3, 1) + (1, 3) + (2, 2),
(2.14)
corresponding to the breaking

SO(8) SO(5).
(2.15)
It is remarkable that we can obtain a non-regular subgroup of SO(8) by the combination of two
inner automorphisms of SO(8). Each twist Ti breaks G = SO(8) to a regular subgroup Hi (in
this case SU(4) U (1)). However, T2 is not contained in H1 , and, although being an inner
automorphism on G, it acts as an outer automorphisms on H1 . The result is the special subgroup
SO(5) of SU(4) U (1). All other gauge groups can, in principle, be calculated along these lines.
For a list of gauge groups that can be obtained this way we refer the reader to Table 6 in Ref. [29].
3. Breaking SU(N) on the orbifold
The torus lattice has a discrete rotational symmetry that can be modded out to obtain the
T 2 /Zn orbifold. The only discrete rotations possible are of order n = 2, 3, 4, 6. The topology of
the resulting spaces are pillows, see Fig. 1. The two sides of the pillow represent the bulk and
the corners the fixed points. We depict the four possibilities in Fig. 1. Notice that T 2 /Z4 contains
two Z4 and one Z2 singularity and T 2 /Z6 contains one Z2 , Z3 and Z6 singularity each.
In analogy to Eq. (2.2), one now introduces orbifold twists
AM (pz) = P AM (z)P 1 .
(3.1)
where p is the nth root of unity

p = exp(2i/n).
(3.2)
The additional identification leads to new constraints. Besides the obvious, p n = 1, one also
has to take into account that a Zn rotation followed by a translation along some lattice vector,
followed again by the inverse rotation, equals a lattice translation along the rotated vector:
p1 (pz + ) = z + p 1 .
(3.3)
198
The group generated by translations and rotations is know as the space group. The full set of
constraints is thus6
T T T T T+ ,
P
(3.4)
T P Tp ,
(3.5)
P 1.
n
(3.6)
Here we have introduced the equivalence relation defined as equal modulo an element of
the center of SU(N ).7 The most general solution to the first of these constraints has been
presented in the previous section. The main purpose of this paper is to show that there are nontrivial solutions to the other two constraints, given torus Wilson lines with generic m and for any
n = 2, 3, 4, 6.
There is an alternative description to Eqs. (3.4)(3.6), called the downstairs picture, that only
makes reference to the fundamental domain of the orbifold (i.e., the physical space). For any
given fixed point zf of the rotation p k , one can define a rotation around zf :

= 1 p k zf ,
pzf (z) = p k z + ,
(3.7)
where is a lattice vector. Choosing any fundamental orbifold domain, the product over the
rotations around all four (Z2 ) or three (Z3,4,6 ) fixed points equals a pure lattice translation, and
some special combinations even yield the trivial one: choosing the fundamental domains and
fixed points labels as in Fig. 1, one finds
pz4 pz3 pz2 pz1 = 1,
pz3 pz2 pz1 = 1,
n = 2,
(3.8)
n = 3, 4, 6.
(3.9)
Obviously, any cyclic permutation of these relations hold. For n = 2, the anticyclic order also
yields one (but not an arbitrary permutation),8 while for n = 3, 4, 6 the anticyclic order already
yields a nontrivial shift. Again, these relations must be represented by the corresponding twists:
Pz1 Pz2 Pz3 Pz4 1,
Pz1 Pz2 Pz3 1,
n = 2,
n = 3, 4, 6,
(Pzi ) 1,
i
(3.10)
with i being the order of the fixed point zi . By re-expressing the lattice shifts through the
rotations, it can be shown that, conversely, the relations Eq. (3.10) imply Eqs. (3.4)(3.6). In
other words, the downstairs picture (in which we specify the local orbifold twists) is completely
equivalent to the upstairs picture (in which we specify the torus Wilson lines and the basic Zn
orbifold twist). Moreover, the downstairs relations can be further reduced by actually solving
Eq. (3.10) for one of the twists in terms of the others. In the case of Z6 , for instance, the relations
then reduce to
Pz32 1,
Pz23 1,
(Pz2 Pz3 )6 1.
(3.11)
6 For the second relation, notice that the order of the gauge group elements is reversed w.r.t. the space group.
7 Mathematically speaking we are looking for special unitary projective representations of the space group.
8 This can easily be seen by taking the inverse of Eq. (3.8) and using the fact that p 2 = 1.
z
f
199
While the first two relations are always easy to satisfy, the last relation becomes highly nontrivial
if the two twists do not commute. In fact, the product Pz3 Pz2 does not even have to have finite
order. It is possible to generalize the orbifold construction to allow for gauge twists whose order
does not match that of the spacetime twist [5]. Such models then allow for many more rank
breaking possibilities. While the downstairs picture is very useful, in particular for the case of
commuting Wilson lines, in this paper we will mainly stick to the upstairs description. For one,
it makes an important aspect of the new rank breaking mechanism manifest: it can be viewed
as an orbifold of topologically nontrivial torus Wilson lines. Secondly, the nice factorization of
the torus Wilson lines, obvious from Eq. (2.9), carries over to the orbifold twists and presents a
convenient way to classify all possible orbifolds.
The rest of this section is organized as follows. In Section 3.1 we will calculate the SU(N )
breaking on the orbifold in the case of vanishing t Hooft flux. In particular, we will focus on
breakings by continuous Wilson lines, corresponding to the part of the moduli space of the torus
that survives the orbifold projection. In Section 3.2 we will then show how to construct orbifold
twists that fulfill Eqs. (3.5) and (3.6) for generic m and N .
3.1. The case m = 0
In the case m = 0, there exists a well defined scheme [20] to construct solutions to Eqs. (3.5)
and (3.6), by identifying P with a suitable element of the Weyl group, the symmetry group of
the root lattice of the Lie algebra. Such an element induces an algebra automorphism that maps
the Cartan subalgebra onto itself. For a given orbifold twist, the Cartan subalgebra naturally
decomposes into two subspaces: The eigenspaces to unit and non-unit eigenvalues under the
linear map P . The latter give rise to Wilson lines that commute with P , and Eq. (3.5) implies
that they are discrete. For n = 2 one finds
T12 = T22 = 1,
(3.12)
while for n = 3, 4 one has T1 = T2 T , with

T 3 = 1,
n = 3,
T = 1,
n = 4.
(3.13)
For n = 6 there are no discrete Wilson lines.9 Wilson lines not invariant under P can still exist
and can be constructed as follows. Consider the shift vector as a map from the torus lattice to the
root lattice, then we can rewrite Eq. (3.5) as a composition of maps
1
V = V p.
Padj
(3.14)
If the torus lattice can be embedded into the root lattice of the algebra, one can choose V to be
any scalar multiple of that embedding and identify the rotation P 1 with p. The Wilson lines
defined this way are thus continuous and will break the rank [20,21]. We will not make use of
this description in this paper. Rather, we will consider an equivalent description in terms of the
zero modes of A4,5 . Just in the case of the torus, the continuous Wilson lines can be transformed
into background VEVs for these extra dimensional components of the gauge bosons and, hence,
9 In the downstairs picture this can also be easily understood: any commuting triple fulfilling Eq. (3.10) in the case
n = 6 automatically also satisfies Pz2 = Pz21 , Pz3 = Pz31 . Hence, the Z6 twist Pz1 already determines the other two
twists.
200
parametrize the moduli space of the compactification. The advantage of this approach is that we
can represent P as a shift (element if the Cartan torus) rather than a rotation (element of the Weyl
group).
Let us consider the case that the gauge twist is the same at each fixed point (no discrete Wilson
lines). The orbifold shift vector can be taken, without loss of generality, to be of the form [33]
1
V = (k1 , k2 , . . . , kr ),
n
ki > 0,
ki n 1.
(3.15)
As shown in Appendix B, a flat direction exists if and only if there are exactly n 1 entries with
ki = 1 with the remaining ki = 0:
V=

1 r1
0 , 1, 0r2 , 1, . . . , 0rn1 , 1, 0rn .
n
(3.16)
Here, 0r stands for an r-dimensional zero vector (some of the ri may be zero). Notice that this
means, in particular, that the inequality in Eq. (3.15) is saturated. For n = 2, Eq. (3.15) already
implies a shift vector that is either trivial or of the form Eq. (3.16) and, hence, there are always
flat directions for nontrivial V . For generic n, the breaking pattern induced by this shift vector is
SU(N ) H0
n

SU(Ni ) U (1)n1 ,
i=1
n
Ni = N,
(3.17)
i=1
with Ni = ri + 1. There are Nmin = min{Ni } flat directions, which are calculated in Appendix B.
There it is shown that, for vanishing discrete Wilson lines, a generic point in moduli space breaks
SU(N ) according to
SU(N ) H
n

U (Ni Nmin ) U (1)Nmin 1 .
(3.18)
i=1
with
N n,
Nk 1,
Nmin = min{Nk },
n
Nk = N.
(3.19)
k=1
The rank of SU(N ) is reduced by Nmin (n 1).

To complete the classification, one could turn on discrete Wilson lines. The full moduli space
of the Ti = 1 case survives this additional projection if and only if the Ti reside in the Cartan
torus of H. In this case, the unbroken subgroup can be any full-rank subgroup of H. It is possible
that only a subspace of the moduli space survives. However, a complete treatment of these cases
lies outside the scope of the present paper and we will omit it here for brevity. For Z6 there
are no discrete Wilson lines, and our analysis already covers all possible breaking patterns. The
smallest group whose rank can be spontaneously broken in a Z6 orbifold (with vanishing m) is
thus SU(6), with a single modulus breaking all of SU(6).
3.2. Generic m
Our classification of solutions to Eqs. (3.5) and (3.6) for generic m proceeds in two steps. First,
we construct the solution PN,m for m, N coprime, which always breaks SU(N ) completely, as
201
we have seen in Section 2. For arbitrary (N, m), we write the most general solution as
T1 = QN/K exp(2iW1 ),
(3.20)
T2 = (RN/K )
(3.21)
m/K
exp(2iW2 ),
P = PN/K,m/K exp(2iV ),
(3.22)
where Wi are discrete Wilson lines subject to Eqs. (3.12) and (3.13). The shift vectors V and
Wi are elements of the Cartan subalgebra of SU(K). The moduli space of this geometry is then
given by the moduli space of an SU(K) theory with vanishing flux. For trivial discrete Wilson
lines, this moduli space has been given in Section 3.1 and Appendix B.
It remains to be shown that, given the Wilson lines10
T1 = Q,
T2 = R m ,
(3.23)
for m and N coprime, we can actually construct an orbifold twist P that fulfills Eqs. (3.5) and
(3.6). For n = 2, it is very easy to write down such a P . The matrix
Pk = k,
(3.24)
can easily be confirmed to fulfill the requirements. For n = 3, 4, 6, we can choose our lattice to
be generated by 1 = 1 and 2 = p. Relation Eq. (3.5) then implies that for any n = 3, 4, 6, we
must have
P 1 QP R m
(3.25)
as well as
m 1
R Q ,
1
P QP R m ,
m
R Q,
n = 3,
n = 4,
(3.26)
n = 6.
The matrices Q and R have the same eigenvalues, given by the N different N th roots of unity.
As m and N are coprime, the same holds true for R m . As a consequence, one can always find an
SU(N ) matrix U that satisfies
U QU R m .
(3.27)
We choose U as11
1
Uk = N 2 q (k1)m ,
q =e
2i
N
(3.28)
The proof that U indeed satisfies Eq. (3.27) is presented in Appendix A. Moreover, U also
satisfies
U QU R m .
(3.29)
Notice that U can be multiplied by any diagonal SU(N ) matrix from the left without affecting
Eq. (3.27), as R is diagonal. However, Eq. (3.29) will be modified. It can be shown that there is
10 We will drop the indices N and m for the rest of the section.
11 This matrix is known as a Vandermonde matrix. The matrix in Eq. (3.28) should be divided by its determinant to
obtain an SU(N ) matrix, which we omit here for clarity.
202
a diagonal SU(N ) matrix X satisfying12

XQX QR m
X QX QR m .
(3.30)
Multiplying U with X we find

(XU ) Q(XU ) U QR m U R m U R m U R m Q1 ,
(3.31)
where in the first step we used Eq. (3.30), in the second step Eq. (3.29) and in the last one
Eq. (3.27). In a completely analogous fashion one can show that

X U Q X U R m Q.
(3.32)
One concludes that by choosing
XU, n = 3,
1
P = U,
n = 4,
X U, n = 6,
we satisfy both Eqs. (3.27) and (3.29). It remains to be shown that
6
(XU )3 1,
U 4 1,
X U 1.
(3.33)
(3.34)
We again postpone the proof of this to Appendix A.

Let us illustrate these general considerations with the simplest possible example: SU(2). The
only possible nontrivial choice is m = 1. In the adjoint the two Wilson lines read:
T1 = diag(+1, 1, 1),
T2 = diag(1, 1, +1).
(3.35)
For the Z2 case, Eq. (3.24) actually gives the identity for P . It follows that in this case the local
twists are simultaneously diagonal in the adjoint:
Pz1 = diag(+1, +1, +1),
Pz2 = diag(1, 1, +1),
Pz3 = diag(1, +1, 1),
Pz4 = diag(+1, 1, 1).
(3.36)
This only happens in the case N = n = 2. At one fixed point SU(2) is left unbroken, while at
every other fixed point a different U (1) survives. Note that this breaking pattern is qualitatively
different from the usual breaking of SU(2) by continuous Wilson lines, as described in Section 3.1: There the local gauge group is U (1) at all four fixed points. For Z3 we find for the
twist

0 0 1
P = Pz1 = 1 0 0 .
(3.37)
0 1 0
The local twists are now truly non-commutative as can be seen by computing the twists associated
to the other two fixed points:
0 0 1
0 0 1
P T1 T2 = Pz2 = 1 0 0 ,
(3.38)
P T1 = Pz3 = 1 0
0 .
0
12 We give the precise form of X in Appendix A.
203
Geometrically, these twists are SO(3) rotations by 120 around the axes (1, 1, 1), (1, 1, 1) and
(1, 1, 1), respectively. Each axis of rotation defines a U (1) subgroup that remains unbroken
at the corresponding fixed point. It is easy to verify that the product Pz1 Pz2 Pz3 indeed gives the
identity.
In summary, we have seen that the discrete torus Wilson lines that break SU(N ) down to
SU(K), with K any divisor of N , are orbifold compatible, i.e., there exists an orbifold twist that
fulfills Eqs. (3.5) and (3.6), for any n = 2, 3, 4, 6. In comparison to the mechanism of rank reduction described in Section 3.1, there are no moduli associated to this breaking. Before concluding
this section we would like to comment on the inclusion of matter to this scenario. Up to now,
we have only considered pure gauge theory or, more precisely, only fields in the adjoint of the
group. Matter usually transforms in representations that are sensitive to the center of the group
(such as the fundamental) and, hence, potentially destroy some or all of the torus configurations.
On the orbifold it is not uncommon that non-adjoint matter only appears on the fixed points (as,
e.g., in constructions that have extended N = (1, 1) supersymmetry in 6d). Another possibility
is to include other global or local symmetries in the twists to compensate for the nontrivial action
of the center.
4. Applications
In this section we would like to outline some possible application as well as further developments of the orbifold constructions considered in this work.
Two main generalizations, relevant in particular to string theory applications, have to be
discussed: higher dimensional orbifolds and gauge groups other than SU(N ). Fortunately, the
material developed in this work already covers these cases to a large extent. As mentioned at
the end of Section 2, the torus bundles can be generalized almost straightforwardly to the case
of other gauge groups with nontrivial center: the center can be embedded in suitable SU(N )
subgroups and the construction of torus holonomies proceeds as before. Similarly, the orbifold
twists calculated in this paper can be embedded in this SU(N ) subgroup in a totally analogous
manner. As a matter of fact, the center of groups other than SU(N ) is given by an Abelian group
of order 4. Hence, the corresponding twists are particularly simple: they just correspond to the
SU(N ) twists described in this paper with N 4.
The second point concerns higher dimensional orbifolds (based on tori T d with d > 2). Of
course, if the torus lattice upon which the orbifold is based factorizes into a two-dimensional
and a (d 2)-dimensional lattice, we can just choose our Wilson lines T1,2 to lie along these
two dimensions, set the remaining Wilson lines Ti = 1 and use the orbifold twist P calculated in
Section 3.2.13 For a given 2d sublattice, this constitutes a canonical solution to the space-group
constraints, Eqs. (3.4) to (3.6), and the surviving subgroup Hc is the one obtained from the torus
compactification. Clearly, the remaining freedom in choosing the most general solution is in oneto-one correspondence to the possible orbifolds of the topological trivial sector of Hc , and the
final gauge group is a subgroup of Hc . To be more explicit, one can multiply all Ti by mutually
commuting Wilson lines Ti residing entirely in Hc and similarly multiply P by a twist P also
residing in Hc . The set {P , Ti } has to separately fulfill the space-group constraints, it can be
constructed in the standard way [1,20].
13 Most of the Abelian heterotic orbifolds (though not all) possess such a two-dimensional lattice. In addition, some
of the higher dimensional lattices allow for a solution to the space-group constraints. We will stick to the 2d case for
simplicity.
204
Let us illustrate these abstract considerations with a concrete example. The gauge group
SO(32) of the Heterotic String possesses nontrivial bundles on the two-torus: Its center is
Z2 Z2 , and it is possible to embed one of the Z2 in an SU(2) subgroup by considering the
maximal special subgroup SU(2) Sp(16) [30]. It can be shown that the center of SU(2) corresponds to the element of the center of SO(32) that acts as the identity on the spinor representation
(but minus the identity on the vector and conjugate spinor representations). Now take, e.g., a Z3
orbifold on the standard SU(3)3 root lattice and choose the two holonomies to correspond to
any one of the three two-dimensional lattices. The SU(2) orbifold twist described in Section 3.2,
along with the two Wilson lines, constitutes a solution to the space-group constraints, leaving
over an unbroken subgroup Hcan = Sp(16). This solution does not, in general, lead to a modular invariant theory. But we are free to choose further twists and Wilson lines in the topological
trivial sector of Sp(16). By making use of this freedom it is possible to construct modular invariant twists and thus a consistent heterotic orbifold with a rank-eight gauge group. The explicit
construction and classification of the modular invariant models, as well as the calculation of the
twisted and untwisted string spectra will be considered in a future publication.
The gauge group E8 , having trivial center, does not possess any nontrivial pairs. However,
it does have nontrivial triples: three holonomies, any two of which can be conjugated to the
same Cartan torus, but not all three simultaneously.14 The surviving unbroken subgroup is again
rank-reduced. It would therefore be interesting to construct orbifolds based on these nontrivial
torus bundles.15 To do so, one has to solve the corresponding space-group constraints for a threedimensional torus.
Last but not least we would like to comment on a (field-theory) application to supersymmetry breaking in six dimensions. Minimal N = (1, 0) supersymmetry has an R-symmetry group
SU(2)R . One can break SU(2)R and, hence, supersymmetry completely by continuous Wilson
lines in the case of Z2 orbifolds but not for Z3,4,6 . We have shown that it is nevertheless possible
to find discrete Wilson lines that break all of SU(2)R for arbitrary Zn , and such a ScherkSchwarz
mechanism is possible. Within this context it is interesting to notice that no continuous parameter exists that controls supersymmetry breaking, yet the breaking is still soft, as locally at least
N = 1 supersymmetry is preserved at all fixed points. Similar constructions can of course be
applied to break all or part of extended supersymmetry.
5. Conclusions
In this paper we have analyzed orbifolds that break the gauge group SU(N ) to lower rank
subgroups. The rank breaking proceeds through nontrivial toron configurations, meaning the
gauge fields have twisted boundary conditions on the covering torus of the orbifold. These
twisted boundary conditions are of topological nature, characterized by the t Hooft flux, and,
as a consequence, they cannot be transformed into a constant background VEV for any extra
dimensional components of the gauge fields. The main result of this paper is that one can actually orbifold these configurations and that a classification of all possible breakings emerges
from this approach. Torus Wilson lines can break SU(N ) down to SU(K), where K is a divisor of N . We have explicitely constructed orbifold twists that fulfill the space-group constraints,
14 As a matter of fact, SO(N ) with N 7, as well as all expectational groups, possess such triples [26,2931].
15 Note that the asymmetric orbifolds of Ref. [32] are not related to our construction. The twists employed here can
only correspond to symmetric orbifolds in string theory.
205
Eqs. (3.4)(3.6), and preserve SU(K). The remaining freedom in choosing the orbifold twists is
that of an orbifold with SU(K) gauge group and trivial (commuting) torus Wilson lines.
We have also described possible applications of this mechanism. Besides its purely field theoretical interest in orbifold GUT model building, one can easily generalize our results to other
gauge groups and higher dimensional orbifolds. In particular, our mechanism allows for string
model building based on the SO(32) Heterotic String, leading to rank-8 gauge groups at low
energy. Furthermore, by generalizing to non-commuting n-tuples, even E8 model building with
gauge groups of rank lower than eight is conceivable.
Acknowledgements
I would like to thank M. Salvatori for useful email exchange. This work was supported by
grants NSF-PHY-0401513, DE-FG02-03ER41271 and the Leon Madansky Fellowship, as well
as the Johns Hopkins Theoretical Interdisciplinary Physics and Astrophysics Center.
Appendix A. Some technicalities
In this appendix we will prove Eqs. (3.27), (3.29), (3.30) and (3.34). Throughout this section
m and N are coprime integers and q is defined as
q =e
2i
N
(A.1)
Using the property

N
q k = N ,0 ,
(A.2)
k=1
Eqs. (3.27) and (3.29) can be readily verified:

N1
q 2 (k1)im+j (1)m
U QU k =
q
i,j 1
N
i,j
(k1)m N1
2
=q
(N1)(m1)
2
Rm

k
qm
k j
= q (k1)m
N1
2
k

R m k .
(A.3)
In the last step of the second line we have made use of the fact that Eq. (A.2) holds if q is replaced
with q m for m and N coprime. In the last step we have used that (N 1)(m 1) is always even.
The proof of Eq. (3.29) is completely analogous and we will skip it here.
The identity U 4 U 4 1 is also quite easy. For m = 1
4
k
1 k(i1)+i(j 1)+j (r1)+r(1)

1 kiki+rr i+r1 j
q
q
= 2
q
2
N
N
i,j,r
i,r
1 kiki+rr
q k r
q
=
q
i,1r =
= q
k .
N
N r
i,r
(A.4)
206
For m = 1 just replace q q m . Let us now define

Xk = q
k(kN)
2
(A.5)
k .
The matrix X does not have unit determinant, det X = ei

easily be cured by a rescaling. Now calculate:
XQX

k
= q
k(kN) N1 (k+1)(kN+1)
2 +
2
2
N 2 1
3
. As in the case of U , this can
k,1
= q k(N1) k,1 = q(RQ)k (RQ)k .
(A.6)
For m > 1 one just has to replace X X m , which concludes our proof of Eq. (3.30).
To prove the remaining relations in Eq. (3.34) we will need the identity16
Z
N1
(kN/2)2
2
iN .
(A.7)
k=0
Let us start with m = 1.

(XU )3k = N 3/2
k(kN)
(k1)i i(iN)
(i1)j j (j N)
(j 1)
2
2
2
(i+j +k1N/2)2 (j +k1N/2)2 k(kN)

+
2 +j j (j N)
(j 1)
2
2
2
i,j
= N 3/2

i,j
1
= N 1 (i) 2
q (j 1)(lk)+
(1+N/2)2
2
= (i) 2 q
(1+N/2)2
2
k, .
(A.8)
The fact that we have collected a nontrivial phase (i.e., not an integer power of q) is related to
the fact that our matrices X and U are U (N ) as opposed to SU(N ) matrices. This could easily
remedied by a rescaling, without affecting the other relations Eqs. (3.27), (3.29), and (3.30).
Since SU(N ) is a group, it follows that the r.h.s. of Eq. (A.8) has to be an SU(N ) element also.
After a suitable rescaling we thus arrive at the first relation in Eq. (3.34). For the last relation in
Eq. (3.34) we calculate
k(kN)
3
j (j N)
i(iN)
X U k = N 3/2
q 2 (k1)i+ 2 (i1)j + 2 (j 1)
i,j
= N 3/2
(ij k+1N/2)2 (j +k1+N/2)2 k(kN)
+ 2 +j + j (j N)
(j 1)
2
2
2
i,j
1
= N 1 (i) 2
q j (2+k+l+N )+k+lkN
(1N/2)2
2
j
1
2
= (i) q 2N k
(1N/2)2
2
k,2 .
(A.9)
16 The fact that ZZ

= N can be inferred by considering the Discrete Fourier Transformation (DFT) of xk =
2
q (kN/2) /2 . By performing the DFT and its inverse, one finds xk = ZZ/N
xk . We shall not prove the value of the
phase in Eq. (A.7) since it will turn out to be irrelevant (see comment after Eq. (A.8)).
207
For N = 2, this is already proportional to the identity. For N > 2 we square this to find
X U
6
k
= iq 4(1+N/2) k, .
(A.10)
Finally, for m > 1 we can just replace q q m and observe that Eq. (A.7) still holds since m and
N are coprime.
Appendix B. The moduli space for m = 0
In this appendix we would like to calculate the moduli space on the orbifold, in the case m = 0.
To this end, we calculate the scalar zero modes from the projection Eq. (3.16) and subsequently
find those modes that correspond to flat directions in the potential. The potential is coming from
the term

2
= 2g 1 Tr[A4 , A5 ]2 = 4g 1 Tr[A+ , A ]2 ,
V Tr Fij F ij = 2g 1 Tr F45
(B.1)
where g = det gij and we have defined the complex scalars A = A4 iA5 . Notice that the
hermiticity of the Ai implies the reality constraint A+ = A . The orbifold boundary conditions
now read:

1
A (z).
A (pz) = exp 2i V
n
(B.2)
The zero modes correspond to those states where the term in the square brackets in Eq. (B.2) is
integer.
To find these zero modes, note that there are n special roots that have V = 1/n mod Z:
the n 1 simple roots that have ki = 1 in Eq. (3.16), as well as the most negative root (defined
as minus the sum of all simple roots). They all belong to different irreducible representations of
the subgroup H0 defined in Eq. (3.17). By inspection of the remaining roots and their associated
raising and lowering operators,17 one can parametrize the zero modes of A+ as
0
0
..
.
A+ =
0
An
A1
0
0
0
0
A2
0
0
0
0
0
0
..
.
0 An1
0
0
(B.3)
Here the entry in the ith row and j th column is a matrix of dimension Ni Nj . In particular, the
Ai are Ni Ni+1 matrices forming the representation
Ai = (Ni , N i+1 ),
(B.4)
17 The positive roots of SU(N ) are given by { =

k
() + (+1) + + (k) , 0 k r} in terms of the simple
+
roots (i) . The associated creation operator is given by (Ek
)ij = i, j,k+1 .
208
where we have adopted a cyclic convention for the indices. One immediately calculates F+ =
[A+ , A ]
A1 A 1 A n An
A2 A 2 A 1 A1
F+ =
(B.5)
..
An A n A n1 An1
The diagonal blocks are now square matrices of dimension Ni . The vanishing of F+ is a necessary and sufficient condition for the potential
V
n
Tr(Ai A i A i1 Ai1 )2
(B.6)
i=1
to possess a flat direction. One can always use the H0 gauge symmetry to diagonalize all Ai A i .
If there is a flat direction, then in this basis the matrices A i Ai must be diagonal as well. Let us
define Nmin = min{Ni }. Then all matrices Ai A i and A i Ai have at least rank Nmin . One concludes
that if a flat direction exists, without loss of generality one can assume:

Ai A i = A i1 Ai1 = diag a1 , . . . , aNmin , 0Ni Nmin ,
(B.7)
where the ak are real constants. All that remains to show is that there exists a configuration Ai
that fulfills
Eq. (B.7). This can easily be achieved by choosing the first Ni diagonal entries of Ai
equal to a i with all other entries equal to zero. A generic VEV along this flat direction breaks
each SU(Nk ) factor to SU(Nk Nmin ). To obtain the U (1) factors, it is sufficient to find the rank
of the surviving subgroup, i.e., we are looking for the number of Cartan generators that satisfy
[A+ , H ] = 0.
(B.8)
To this end, note that we can view the quantity A+ as a linear map from the Cartan subalgebra
to the subspace of su(N ) generated by those E that are nonzero in A+ . Writing down the
matrix corresponding to that map, it can be read off that it has rank Nmin (n 1). The rank-nullity
theorem then states that the dimension of the kernel of that map is equal to N 1 Nmin (n 1),
which must equal the rank of the surviving subgroup. Thus, the breaking pattern turns out to be
SU(N )
n

SU(Nk ) U (1)n1
k=1
n

U (Nk Nmin ) U (1)Nmin 1 .
(B.9)
k=1
It may be verified that the rank of this group is indeed N 1 Nmin (n 1). Let us summarize
the conditions the different quantities in Eq. (B.9) are subject to:
N n,
Nk 1,
Nmin = min{Nk },
n
Nk = N.
(B.10)
k=1
Let us now turn to shift vectors that are not of the form Eq. (3.16) but still fulfill condition
(3.15). The breaking pattern will still be of the form Eq. (3.17), but now with fewer SU(Nk )
factors. The simple roots and the most negative root still belong to bifundamentals. The important
difference
is that one or more of these bifundamentals cease to have zero modes (some ki > 1
and/or ki < n 1). Removing one or more of the Ai from Eq. (B.5) or (B.6) clearly destroys
the possibility of having flat directions. We conclude that flat directions exist if and only if V is
equivalent to the form Eq. (3.16).
209
References
[1] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678;
L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285;
L.E. Ibanez, H.P. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25;
L.E. Ibanez, J. Mas, H.P. Nilles, F. Quevedo, Nucl. Phys. B 301 (1988) 157.
[2] W. Buchmuller, K. Hamaguchi, O. Lebedev, M. Ratz, Phys. Rev. Lett. 96 (2006) 121602, hep-ph/0511035;
W. Buchmuller, K. Hamaguchi, O. Lebedev, M. Ratz, hep-th/0606187.
[3] Y. Kawamura, Prog. Theor. Phys. 105 (2001) 999, hep-ph/0012125;
Y. Kawamura, Prog. Theor. Phys. 105 (2001) 691, hep-ph/0012352;
G. Altarelli, F. Feruglio, Phys. Lett. B 511 (2001) 257, hep-ph/0102301;
A.B. Kobakhidze, Phys. Lett. B 514 (2001) 131, hep-ph/0102323;
L.J. Hall, Y. Nomura, Phys. Rev. D 64 (2001) 055003, hep-ph/0103125;
A. Hebecker, J. March-Russell, Nucl. Phys. B 613 (2001) 3, hep-ph/0106166;
A. Hebecker, J. March-Russell, Nucl. Phys. B 625 (2002) 128, hep-ph/0107039;
L.J. Hall, Y. Nomura, D.R. Smith, Nucl. Phys. B 639 (2002) 307, hep-ph/0107331;
T. Asaka, W. Buchmuller, L. Covi, Phys. Lett. B 523 (2001) 199, hep-ph/0108021;
L.J. Hall, Y. Nomura, T. Okui, D.R. Smith, Phys. Rev. D 65 (2002) 035008, hep-ph/0108071;
T. Watari, T. Yanagida, Phys. Lett. B 532 (2002) 252, hep-ph/0201086;
A. Hebecker, JHEP 0401 (2004) 047, hep-ph/0309313.
[4] T. Asaka, W. Buchmuller, L. Covi, Phys. Lett. B 540 (2002) 295, hep-ph/0204358.
[5] A. Hebecker, M. Ratz, Nucl. Phys. B 670 (2003) 3, hep-ph/0306049.
[6] G. von Gersdorff, JHEP 0703 (2007) 083, hep-th/0612212.
[7] M. Trapletti, Mod. Phys. Lett. A 21 (2006) 2251, hep-th/0611030.
[8] N.S. Manton, Nucl. Phys. B 158 (1979) 141;
P. Forgacs, N.S. Manton, Commun. Math. Phys. 72 (1980) 15.
[9] Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Ann. Phys. 190 (1989) 233.
[10] N.V. Krasnikov, Phys. Lett. B 273 (1991) 246;
H. Hatanaka, T. Inami, C.S. Lim, Mod. Phys. Lett. A 13 (1998) 2601, hep-th/9805067;
G.R. Dvali, S. Randjbar-Daemi, R. Tabbash, Phys. Rev. D 65 (2002) 064021, hep-ph/0102307;
L.J. Hall, Y. Nomura, D.R. Smith, Nucl. Phys. B 639 (2002) 307, hep-ph/0107331.
[11] I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20, hep-th/0108005.
[12] M. Kubo, C.S. Lim, H. Yamashita, Mod. Phys. Lett. A 17 (2002) 2249, hep-ph/0111327.
[13] G. von Gersdorff, N. Irges, M. Quiros, Nucl. Phys. B 635 (2002) 127, hep-th/0204223;
G. von Gersdorff, N. Irges, M. Quiros, hep-ph/0206029.
[14] C. Csaki, C. Grojean, H. Murayama, Phys. Rev. D 67 (2003) 085012, hep-ph/0210133;
G. von Gersdorff, N. Irges, M. Quiros, Phys. Lett. B 551 (2003) 351, hep-ph/0210134.
[15] G. Burdman, Y. Nomura, Nucl. Phys. B 656 (2003) 3, hep-ph/0210257;
N. Haba, Y. Shimizu, Phys. Rev. D 67 (2003) 095001, hep-ph/0212166;
I. Gogoladze, Y. Mimura, S. Nandi, Phys. Lett. B 560 (2003) 204, hep-ph/0301014;
I. Gogoladze, Y. Mimura, S. Nandi, Phys. Lett. B 562 (2003) 307, hep-ph/0302176;
I. Gogoladze, Y. Mimura, S. Nandi, hep-ph/0311127;
C.A. Scrucca, M. Serone, L. Silvestrini, Nucl. Phys. B 669 (2003) 128, hep-ph/0304220;
K. Choi, et al., hep-ph/0312178.
[16] C. Biggio, M. Quiros, Nucl. Phys. B 703 (2004) 199, hep-ph/0407348.
[17] G. Martinelli, M. Salvatori, C.A. Scrucca, L. Silvestrini, JHEP 0510 (2005) 037, hep-ph/0503179;
N. Irges, F. Knechtli, hep-lat/0604006;
N. Irges, F. Knechtli, hep-lat/0609045.
[18] P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142;
C.A. Scrucca, M. Serone, M. Trapletti, Nucl. Phys. B 635 (2002) 33, hep-th/0203190;
F. Gmeiner, S. Groot Nibbelink, H.P. Nilles, M. Olechowski, M.G.A. Walter, Nucl. Phys. B 648 (2003) 35, hepth/0208146;
G. von Gersdorff, M. Quiros, Phys. Rev. D 68 (2003) 105002, hep-th/0305024;
C.A. Scrucca, M. Serone, Int. J. Mod. Phys. A 19 (2004) 2579, hep-th/0403163.
210
[19] M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117.
[20] L.E. Ibanez, H.P. Nilles, F. Quevedo, Phys. Lett. B 192 (1987) 332;
A. Font, L.E. Ibanez, H.P. Nilles, F. Quevedo, Nucl. Phys. B 307 (1988) 109;
A. Font, L.E. Ibanez, H.P. Nilles, F. Quevedo, Nucl. Phys. B 310 (1988) 764, Erratum;
A. Font, L.E. Ibanez, H.P. Nilles, F. Quevedo, Phys. Lett. B 210 (1988) 101;
A. Font, L.E. Ibanez, H.P. Nilles, F. Quevedo, Phys. Lett. B 213 (1988) 564, Erratum.
[21] S. Forste, H.P. Nilles, A. Wingerter, Phys. Rev. D 72 (2005) 026001, hep-th/0504117;
S. Forste, H.P. Nilles, A. Wingerter, Phys. Rev. D 73 (2006) 066011, hep-th/0512270.
[22] G. t Hooft, Nucl. Phys. B 153 (1979) 141.
[23] R. Rabadan, Nucl. Phys. B 620 (2002) 152, hep-th/0107036;
R. Blumenhagen, G. Honecker, T. Weigand, JHEP 0508 (2005) 009, hep-th/0507041.
[24] M. Salvatori, hep-ph/0611309;
M. Salvatori, hep-ph/0611391;
J. Alfaro, A. Broncano, M.B. Gavela, S. Rigolin, M. Salvatori, JHEP 0701 (2007) 005, hep-ph/0606070.
[25] P. van Baal, Commun. Math. Phys. 85 (1982) 529;
P. van Baal, Commun. Math. Phys. 92 (1983) 1;
P. van Baal, Commun. Math. Phys. 94 (1984) 397;
P. van Baal, B. van Geemen, J. Math. Phys. 27 (1986) 455.
[26] A. Borel, R. Friedman, J.W. Morgan, math.GR/9907007.
[27] C. Schweigert, Nucl. Phys. B 492 (1997) 743, hep-th/9611092.
[28] W. Lerche, C. Schweigert, R. Minasian, S. Theisen, Phys. Lett. B 424 (1998) 53, hep-th/9711104.
[29] V.G. Kac, A.V. Smilga, hep-th/9902029.
[30] E. Witten, JHEP 9802 (1998) 006, hep-th/9712028.
[31] A. Keurentjes, A. Rosly, A.V. Smilga, Phys. Rev. D 58 (1998) 081701, hep-th/9805183;
A. Keurentjes, JHEP 9905 (1999) 001, hep-th/9901154;
A. Keurentjes, JHEP 9905 (1999) 014, hep-th/9902186.
[32] J. de Boer, R. Dijkgraaf, K. Hori, A. Keurentjes, J. Morgan, D.R. Morrison, S. Sethi, Adv. Theor. Math. Phys. 4
(2002) 995, hep-th/0103170.
[33] J. Fuchs, C. Schweigert, Symmetries, Lie Algebras and Representations, Cambridge Univ. Press, 1997.

Multi-branes boundary states with open

string interactions
Igor Pesando a,b
a Dipartimento di Fisica Teorica, Universit di Torino, Via P. Giuria 1, I-10125 Torino, Italy
b INFN, Sezione di Torino, Italy
Received 5 October 2006; received in revised form 13 September 2007; accepted 1 October 2007
Abstract
We derive boundary states which describe configurations of multiple parallel branes with arbitrary open
string states interactions in bosonic string theory. This is obtained by a careful discussion of the factorization
of open/closed string states amplitudes taking care of cycles needed by ensuring vertices commutativity: in
particular the discussion reveals that already at the tree level open string knows of the existence of closed
string.
PACS: 11.25.-w; 11.25.Uv
Keywords: String theory; D branes
1. Introduction and motivations

Since the discovery of the non-perturbative role played by strings with Dirichlet boundary
conditions [1] the boundary state formalism, first used in [2], has been further developed. It has
proven to be useful in a number of situations such as, for example, in reading which supergravity
fields are switched on in presence of branes [3,4] or in analyzing the rolling of the tachyon (see
for example [5]).
All the boundary states used were trivially a superposition of boundary states and they did
not describe any open string interactions (see however Appendix A of the first of [2]): this can
be rephrased technically by saying that the boundary state of N branes was taken to be equal
E-mail address: ipesando@to.infn.it.
doi:10.1016/j.nuclphysb.2007.10.002
212
I. Pesando / Nuclear Physics B 793 (2008) 211245
to N|B where |B is the boundary state of a single brane. This leaves the doubt whether it is
possible to describe with the boundary state formalism open string interactions or a non-trivial
superposition of branes.
The aim of this paper is to show that it is actually simple to describe open/open and
open/closed string interactions within the closed string formalism and that it is also possible
to describe with a unique boundary state a set of several parallel branes interacting among themselves in the open string channel: an example of the amplitudes described by the boundaries we
consider can be seen in Figs. 4, 5. In doing so we notice how the open string knows about the
closed string already at the tree level. We will show this starting from mixed closed/open string
amplitudes and then factorizing them in the closed channel in order to get boundary states which
describe open string interactions expressed using closed string operators: the resulting boundary
state (see Eq. (4.20)) is for equal parallel branes in generic position, i.e. when the branes are
not superimposed and the low energy gauge group is a product of one U (1) for each brane. It is
nevertheless easy to get the non-generic situation with enhanced gauge group (see Eq. (4.25)).
Since it is possible to describe open string interactions with closed string formalism it is natural to ask whether it is possible to do the opposite, i.e. to describe pure closed string interactions
within open string formalism. An almost positive answer is given in [6] where closed string
amplitudes (with a special state insertions) are obtained from an ordered sequence of D-branes
located at imaginary time.
The paper is organized as follows. In Section 2 we fix our notations (see also Appendix A)
by reviewing how to compute mixed open/closed string amplitudes in open string formalism and
we comment also on how ChanPaton factors emerge naturally when branes are moved from a
generic position where they are parallel but non-superimposed to a superimposed configuration.
We then write the vertices which describe the emission of closed string states using only open
string fields and vertices for the emission of open strings hanging between two parallel but separated branes up to cocycles. The cocycles are discussed in Section 3 for both open and closed
string formalism: they are necessary to ensure the vertices commutativity in both closed and open
formalism and allow a consistent factorization of amplitudes. We start computing the cocycles in
closed string formalism since the vertices for the emission of closed string states in open string
formalism must reproduce the same OPEs as in the closed string sector. It is worth stressing that
in order to achieve the full consistency of the theory is necessary not only to introduce cocycles
but also to let y0 (which is the position of the = 0 boundary of the string when we T-dualize to
a configuration where all space directions have DD boundary conditions or the Wilson lines on
the brane at = 0 when all directions are NN) change after the emission of an open string. This
change is also necessary and felt by closed string vertices when all boundary conditions are NN.
Moreover it is interesting to notice that cocycles are not invariant under T-duality and therefore
amplitudes are determined up to an overall sign.
In Section 4 using the classical techniques we perform the factorization of the open string
formalism amplitudes into the closed string channel and we obtain the explicit form of boundary
states describing a bunch of parallel branes interacting in the open channel. These boundary
states can be used, for example, to recover mixed string amplitudes of one closed string state
with many open string states simply by performing the product with the closed string state or to
obtain one loop amplitudes by computing boundaryboundary interaction.
Finally in Section 5 we draw our conclusions.
213
2. Review of mixed closed and open string amplitudes

Interactions among closed and open strings were extensively studied already in the early days
of string theory [7,8]. In particular, in Ref. [8] Ademollo et al. constructed vertex operators for the
emission of a closed string out of an open string, and computed the scattering amplitudes among
Nc closed and No open strings at tree level. The topology of the string world-sheet corresponding
to these amplitudes is that of a disk emitting No open strings from its boundary and Nc closed
strings from its interior. As customary in those days, only Neumann boundary conditions were
imposed on the disk, and no target-space compactification was considered. This formalism was
extended to the case of mixed NeumannNeumann and DirichletDirichlet boundary conditions
in a compactified target space in Ref. [3].
We find now useful to recall here these results also in order to fix our notations.
We consider a bosonic string described by X ( = 0, . . . , D 1) propagating in a partially
. . . , D 1),
compactified spacetime with toroidal compact directions X a X a + 2R a (a = d,
to simplify the notation we will generically set all R a = R. Both compact and non-compact
directions can have either NN or DD boundary conditions. We write the open string coordinates
as a function of left and right components XL (z), XR (z) as

1
XL (z) + XR (z) ,
(2.1)
2
independently of the boundary conditions. In the previous expression Eq. (2.1) XL,R are defined
as
sgn(n)
X (z) = q + y i 2 a ln z + i 2
a zn ,
n=0
L
0
0
|n| n
(2.2)

XR
(z) = S q y0 i 2 a0 ln z + i 2 n=0 sgn(n)

a z n ,
|n| n
X(, ) =
where < arg(z) + , log(z) = log(z), y0 is an operator which commute with everything
and can be traded with a constant (see later for more) and in the second equation we have introduced the diagonal matrix
S
= diag(1, 1, . . . , 1)
(2.3)
in which all the signs +() can be chosen independently. Each sign is fixed to +() when
the corresponding coordinate has NN (DD) boundary condition, explicitly if we consider a NN
direction we take S = 1 and we have (z = ei( + ) = eE +i where E is the Wick rotated time)
X(, ) = q0 + 2 p + oscillators,
(2.4)
while for a DD direction we choose S = 1 and we find

X(, ) = y0 + 2 p + oscillators.
(2.5)
From these expressions it is clear that y0 plays the role of the boundary value in case of a dia |
a
rection with DD boundary condition, i.e. for example in compact directions XDD
=0 = y0 and

a
a
a
a
a

XDD | = = y = y0 + 2(w + )R where R = R is the dual radius, the operator p a has
a
a
spectrum p a = w R+ = (w a + a )R and 0 a < 1 is associated with the difference of position of two branes or, with NN boundary conditions, with the difference of Wilson lines which
can be eventually turned on the = 0 and = branes. It is also very important
out that
to point

the zero mode part of the expansion in the DD sector XDD(zero modes) = y0 i 2 a0 ln z/z is
214
an operator (where a0 = 2 p) and not a c-number as one would deduce from the canonical
quantization, exactly as it happens for the winding number for a closed string propagating on
a torus. This fact is fundamental in the construction of massive W emission vertices [9] and
in the following construction of the cocycles. It is noteworthy to stress once again that y0 is an
operator too but it commutes with everything and therefore it can be traded with its eigenvalues
and treated as a c-number: the same happens with the usual ChanPaton factors and this is not
accidental since y0 is a kind of ChanPaton factor as we discuss below.
The emission from the disk of an open string state with internal quantum numbers is described by a vertex operator V (XL (x); k) up to a possible cocycle to be discussed later. The
momentum NN b.c.,
vector k = (k , ka ) ( = 0, . . . , d 1) must be interpreted as k = distance
i.e.
DD b.c.,
2

momentum in directions with NN boundary conditions and as distance/(2 ) in the ones with
a
a
DD boundary conditions.1 In particular for compact NN directions we have k a = k +
where
R
a
n Z.
The emission of a closed string state with momentum in non-compact direction k = 2kL =
a
a
a
a
2kR , momentum kLa = 12 ( nR + w R ), kRa = 12 ( nR w R ) in compact direction and left and right
quantum numbers L and R , is described by a vertex operator WL ,R (z, z ; kL , kR ): the presence of a boundary on the world sheet imposes a relation between the left and right parts of
WL ,R which are not independent of each other. In fact it is possible to write [3,8]2

WL ,R (z, z ; kL , kR ) = VL XL (z); kL VR XR (z); kR ,
(2.6)
again up to a cocycle. We would like to stress that the vertex operator W depends on a single set
of oscillators (i.e. those of the open string), and that each factor in Eq. (2.6) is separately normal
ordered. This is to be contrasted with the vertex operators describing the emission of a closed
string out of a closed string where there are two distinct sets of oscillators for the left and right
sectors which only share the zero-mode in the non-compact case.
A simple and intuitive reason why it is possible to write closed string vertices using open
string fields is that if we start with the free open string sigma model with whichever boundary
condition we can always couple the graviton as

1
1
2
S=
(2.7)
d X X S =
d 2 g (X) X X
4
4
and this means that the graviton vertex which follows from the weak coupling expansion
g (X) = + h (X) can be expressed through the open string fields X. In particular
the quantum conformal properties required for a vertex implies its factorized form as shown
in Eq. (2.6) for a generic closed string state. From the same argument this result applies to all the
massless closed string excitations and it can be extended to massive states by OPE. It is in this
process that cocycles become unavoidable since otherwise the OPE coefficients in closed and
open formalism would differ by phases.
1 In presence of DD direction this requires a little explanation, see [9] and also our further discussion in Section 3.2.
The string endpoints can be attached on two different branes at a distance in such a DD direction. This changes the
mass shell condition due to the energy coming from the tension of the stretched string while the vertex still must have
conformal dimension one, hence we must explicitly act on the SL(2, R) vacuum with an operator to get this distance.
2 One could wonder why not to use W = V V

R L instead: the two vertices W and W differ in fact by a phase
exp(i2 kL kR ) which is non trivial in presence of compact directions. The answer is that the true vertices need a
cocycle and that the two versions of the vertices must have different cocycles since they both must reproduce the closed
OPEs.
215
Using the complete operators V and W (i.e. with the cocycles included), the tree-level scattering amplitude among No open and Nc closed strings is given by
No
A(No , Nc ) = C0(p) No(p)
Nc Nc

0|T
N
Nc
o 1

1
dxi (xi+1 xi )
dxNo
d 2 zj
dVabc
i=1
No

i=1
Vi (xi ; ki )
Nc

j =1
Wj L ,j R (zj , z j ; kLj , kRj ) |0,
(2.8)
j =1
where dVabc is the volume of the projective group SL(2, R), T denotes the time (radial) ordering prescription, and C0(p) , No(p) and Nc are respectively the normalizations of the disk, of the
open and of the closed vertex operators corresponding to a Dp brane: they are explicitly given in
Appendix A. They are fixed so that the low energy limit of the previous amplitudes reproduces
the corresponding field theoretical amplitudes of a theory living in p + 1 spacetime dimensions
with DNN,c compact ones (see Ref. [10]). In Eq. (2.8) the variables xi s are integrated on the
real axis while the complex variables zj s are integrated on the upper half complex plane. In this
expression we have not used ChanPaton factors since we suppose the branes are in a generic
position as in Fig. 1, i.e. they are parallel but do not overlap (hence we are restricting to the
case of Dp with p < 25) and therefore the gauge group is a product of U (1) factors. Nevertheless it is possible to introduce the ChanPaton factors simply by performing the substitution
V (x; k) V (x; k) when it is necessary, i.e. when the branes are not anymore in a generic
position and they are stacked.
Fig. 1. As long as brane 0 is at y0 and brane 1 at

y0 + 2 p with p = 0 the brane positions are more
informative than ChanPaton factors and are sufficient
to identify the branes. When p = 0 the branes are superimposed and ChanPaton factors are needed since
all branes have the same position.
Fig. 2. When the branes are coincident one must introduce ChanPaton since we cannot anymore label the
branes by their position.
216
The reason why we did not write the ChanPaton factors in Eq. (2.8) is that they are not
necessary when branes are in a generic position. Let us see why. A ChanPaton factor is nothing
but a matrix: in the simplest setting, i.e. for gauge group U (N ), we can take as the ChanPaton
j
matrices lm =
li m
. The matrices have one label l for the brane where the string leaves from
(start brane) and one label m for the brane where the string ends (end brane).
When computing an open string amplitude with ChanPaton factors we first make a product of ChanPaton matrices and this ensures that the end brane of a string is the start brane
of the next one, then we take the trace of this product so that the brane where the last string
ends is the brane where the first string originates. The result of the trace accounts for the
degeneracy of diagram one is considering but in the generic setting this degeneracy is simply one therefore using the naive ChanPaton factors in computing an amplitude involving N
branes in a generic position would yield the wrong degeneracy. In our case where the branes
are in a generic position the role of the two labels can be taken by y0 for the start brane and
y = y0 + 2(w + )R for the end brane. In computing amplitudes we have now to take care
that the end brane of a string is the start brane of the next one since this is not anymore implied by matrix multiplication. The effect of momentum (actually width) conservation in DD
directions in this formalism is to ensure that the sum of all string widths, defined each as
y y0 , is zero: this is actually not a new constraint but a consistency condition. In the limit
where some branes are superimposed, the brane position is not anymore enough to distinguish
among them and it must be supplemented with a label, thus recovering ChanPaton factors.
It is interesting to note that amplitude (2.8) is well defined even when there are some noncompact directions with DD boundary conditions. When some non-compact X have Dirichlet
boundary condition it would seem that the correlator in (2.8) be divergent since it is proportional to non-compact DD 0|0non-compact DD where |0non-compact DD are the would-be momentum
vacua in the non-compact Dirichlet directions. These vacua |0non-compact DD should naively be
normalized as non-compact DD 0|0non-compact DD (0) since the corresponding directions are noncompact but this is not the case. In fact we can normalize non-compact DD 0|0non-compact DD = 1 as
|0non-compact DD is the only finite energy state because all the winding states have energy proportional to R . This justifies the point of view that non-compact DD directions must be
understood as a limit of compactified ones, in particular the spectral decomposition of unity is
then given by

wR wR

1non-compact DD = lim
(2.9)

R
w

where | wR
are the eigenstates of the corresponding pnon-compact DD .
3. The cocycles for vertices
In this section we first determine the cocycles for closed string vertex operators in closed string
formalism so that two closed string vertices commute with each other, then we fix the cocycles for
closed string vertex operators in open string formalism by demanding that the normal ordering of
the product of two closed string vertices be the same in both closed and open string formalism:
this is done in order to be sure that it is possible to describe closed string emission in open
formalism at all mass levels and that open/closed string amplitudes are correctly and consistently
factorizable in the closed channel. We fix also the cocycles for open string emission vertices
so that they commute with closed string ones. In doing so we realize that also y0 , the quantity
217
entering the left and right moving part of the open string fields given in Eqs. (2.2), has to change
after the emission of an open string from the = 0 border and this change is fundamental even
when open string has NN boundary condition (see Section 3.3 for an explicit example).
All these cocycles and the transformation of y0 turn out to be conceptually very important for
a proper amplitudes factorization but of little practical importance as long as one is willing to
drop phases in amplitudes.
3.1. The cocycles for the closed string
As said before the first step is to determine the cocycles in the closed string formalism. A well
defined CFT for closed strings must satisfy at least the following criteria
(1) closed string vertices commute;
(2) a proper behavior under Hermitian conjugation.
The fist request is necessary in order to ensure the mutual locality of closed vertices [11], or in
other words, that vertices obey the spin-statistics theorem. Usually vertices are written without
cocycles as
WL ,R (z, z ; k) = VL (z; kL )VR (z; kR )
(c)
(3.1)
where the right moving VR is a normal ordered functional of the closed string right moving part
sgn(n)
(c)
a Rn z n
XR (z) = qR i2 p R ln z + i 2
|n|
n=0
(3.2)
and similarly for the left moving part

sgn(n)
(c)
aLn zn
XL (z) = qR i2 pL ln z + i 2
|n|
n=0
with z C and

qL , pL = qR , pR = i ,

am
, an = a m
, a n = m,n ,
(3.3)
m, n > 0.
(3.4)
The previous expansions are however not completely exact for the non-compact directions since
in this case the zero modes are common to left and right moving sectors. Because of this the
non-compact left and right part of the vertices are not separately normal ordered but only the
non-zero modes parts are normal ordered separately while the common zero modes are normal
ordered together.
For implementing the wanted properties we look for a solution of the form [11]
(c)
WL ,R (z, z ; k) = c(kL , kR ; pL , pR )VL (z; kL )VR (z; kR )
(3.5)
where the cocycles are given by

c(kL , kR ; pL , pR ) = cL (kL , kR ; pL )cR (kL , kR ; pR )

= ei (BkL +C kR )a pL ei (DkL +E kR )a pR .
a
(3.6)
We choose the coefficients B, C, D, E to be matrices in order to be general. They have possibly

only non-vanishing entries in compact directions, e.g. Ba.b , since only when there are compact
218
directions they are needed. They must be determined, as said before, so that two arbitrary vertices are mutually local, i.e. commute. This is also equivalent to the fact that the radial ordering
of a product of vertices is given by a unique expression which can be derived by analytically
continuing whichever particular ordering of the vertices is chosen to perform the computation.
In order to compute these matrices we consider the ordering of the product of two arbitrary
vertices as
W(c)
(z, z ; k1 )W(c)
(w, w;
k2 ) = ei(c) (k1 ,k2 ) c(kL1 + kL2 , kR1 + kR2 ; pL , pR )
L ,R
L ,R
VL (z; kL1 )VL (w; kL2 )VR (z; kR1 )VR (w;
kR2 )
(3.7)
where we have defined the phase

(c) (k1 , k2 ) = (BkL2 + CkR2 ) kL1 + (DkL2 + EkR2 ) kR1
(3.8)
and then we use the well known formula ( arg(z), arg(w) + , |z| < |w|) (see also Appendix B)

V (z; k1 )V (w; k2 ) an. cont. = V (w; k2 )V (z; k1 )e+i sgn(arg(z)arg(w))2 k1 k2
(3.9)
(c)
to compare with the product of the vertices in the opposite order W(c)
k2 )W , (z, z ; k1 )
L ,R (w, w;
L R
and determine the unknown matrices. From the previous equations, as it is described in Appendix C.1, we can deduce that the (matrix) coefficients in compact directions are
C = D = diag(1 + 2Na ),
B = E = 2 diag(Ma )
(3.10)
where Na , Ma Z. In the rest of the paper we choose

C = D = 1,
B = E = 0
so that the cocycle in Eq. (3.6) simply

(3.11)
becomes3
c(kL , kR ; pL , pR ) = ei (kRa pL kLa pR ) .

a
(3.12)
The action of T-duality on the cocycle of the vertex is not trivial, since it exchanges the
possible solutions available, i.e. a T-duality along the compact direction a gives (Caa , Daa )
(Daa , Caa ), (Baa , Eaa ) (Eaa , Baa ) or Na Na 1, Ma Ma . The question is now
whether we have different theories since different choices of cocycles yield amplitudes which
differ by a phase. This question assumes that we read the actual effective theory vertices, signs
included, from the string amplitudes. At first look we would answer yes since there is no way
of redefining the fields of the effective action in a such a way that phases in front of amplitudes
match, as it is easy to verify in the case of the three tachyons (see Eq. (C.13) in Appendix C).
On the other side also loop corrections are proportional to the same phase of the tree level since
moving the cocycles in front of the string of vertices yields the same numeric phase as in the tree
level case (this phase depends on external momenta only and does not depend on the number
of handles) times an operatorial cocycle like e
i (B
Nc
j =1 kLj +C
Nc
j =1 kRj )pL +
. Now the above
3 It is also possible to choose c(k , k ; p , p ) = ei (kL +kR )a (pL pR )a = ei na w a where w

a is the winding
L R L R
operator if one is willing to have vertices which are Hermitian up to a phase and, hence, have a Zamolodchikov metric
not equal to the unity.
219
operatorial cocycle is identically 1 since the numeric coefficients are zero because they are proportional to the sum of the external momenta which are conserved. Because of this the would-be
different theories we get give different S-matrix elements which nevertheless yield the same transition probability to all orders in perturbation theory. Moreover the physical states created by the
different vertices are the same.
3.2. The cocycles and y0 shift in the open string formalism
A well defined CFT for an openclosed string theory must satisfy at least the following 5
constraints:
(1) the open string vertices at = 0 and = commute;
(2) the open string emission vertex from = 0 commutes with the closed string vertices;
(3) in a similar way the open string emission vertex from = commutes with the closed string
vertices;
(4) the closed string vertices in open string formalism must have the same OPE as in closed
string formalism, or more precisely we want the normal ordering of the product of two closed
string vertices be formally equal to (3.7), in particular we want the closed string vertices to
commute;
(5) a proper behavior under Hermitian conjugation.
The previous constraints can fail only because of phases and these phases depend only on
momenta therefore the problem amounts to find the proper cocycles and, as we discuss later, the
proper shift of the y0 (which enters the definition of left and right moving part of the open string
expansion given in Eq. (2.2)).
Before writing the vertices it is worth stressing once again what discussed in Section 2: we
want the closed string vertices be expressed using the open string operators only, i.e. closed
strings are made of the same stuff of the open string they are emitted. We derive the vertices
in Appendix C.2, here we limit ourselves to write them explicitly

WL ,R (z, z ; kL , kR ) = ei (SkL kR ) VL XL (z); kL VR XR (z); kR ,
(3.13)

V (x; k) = V XL (x)y0 ; k , x > 0
(3.14)
along with the vertices for the emission from the = boundary

V (x; k) = e2ikp V XL (x)y0 ; k , x < 0.
(3.15)
In this case the action of the cocycle simply means that we can substitute [XL (x) y0 ]zero mode =
q 2 ip log(x) q 2 ip log(|x|) and drop the cocycle.
While the previous vertices seem essentially the normal vertices, the closed string vertex does
depend on y0 since
WL ,R (z, z ; kL , kR ) eiy0 (kL SkR ) . . .
(3.16)
and y0 changes along the worldsheet. This happens because we are taking the original point of
view of [8] according to which the stuff of the carrier string may change after the emission
of another open string. Because of this the closed string emission vertices do not change in form
when written using the carrier string fields from which the closed string states are emitted but
220
Fig. 3. Two different pictures of the emission of a string from an incoming one which transforms into an outgoing string.
The incoming string stretches between brane 0 at y0in and brane 1 at y0in + in (in = 2 kin ); the emitted string stretches
between brane 2 at y0out and brane 0 at y0out + emitted = y0in (emitted = 2 kemitted ) and the outgoing string between
brane 2 at y0out = y0in emitted and brane 1 at y0out + emitted + in .
they do change when the actual carrier string fields are expressed using the fields of incoming
carrier string.
The change of y0 can be better appreciated by considering a carrier open string with Dirichlet
b.c. in at least one compact direction. In Fig. 3 we have pictured in two ways the interaction
between an incoming open string stretched between brane 0 and brane 1 with another open
string, the emitted string, stretched between brane 2 and brane 0. The incoming string has DD
b.c. X in ( = 0) = y0 and X in ( = ) = y0 + in with in = 2 kin . The other string, which is
emitted from the incoming one, stretches between X emitted ( = 0) = y0emitted = y0 emitted and
X emitted ( = ) = y0emitted + emitted = y0 . The result of the interaction is the outgoing string with
X out ( = 0) = y0out = y0emitted = y0 emitted and X out ( = ) = y0out + out = y0 + in + emitted .
The latter equation has a simple interpretation: when a string with DD b.c. is emitted from the
= 0 border of the incoming open string the outgoing string has width equal to the sum of the
incoming carrier string in and the emitted string emitted . This is taken care by the momentum p
conservation but the position of the = 0 border of the outgoing string is changed in such a way
the position of the = border is left unchanged, i.e.
y0out = y0 emitted ,
emitted = 2 kemitted .
(3.17)
The main point hence is that the open string changes its b.c. after the emission of a string from
the = 0 boundary and therefore y0 changes too accordingly. Hence the closed string vertex
does change when expressed using the incoming string fields.
Because of the expansions in Eq. (2.2) this change in y0 means
XLout = XLin emitted
(3.18)
out
in
= XR
+ Semitted .
XR
(3.19)
and
221
Hence the closed string vertex changes when inserted before or after the emission of the open
string from the = 0 boundary. The proper commutation relation between a closed string and
an open string emitted from = 0 then reads

out
WL ,R XLout (z), XR
(z) V (x; k) an. cont.

in
= V (x; k)WL ,R XLin (z), XR
(z)

out
(z) e2 ik(kL SkR )
= V (x; k)WL ,R XLout (z), XR

in
= WL ,R XLin (z), XR
(z) V (x; k) an. cont. e2 ik(kL SkR ) ,
(3.20)
where we have used the fact that = 2 k and an. cont. means analytically continued.
3.3. An explicit example
The importance of the cocycle factors and the y0 shift for a full consistency of amplitudes
and the proper understanding of the consequences of commutation relation Eq. (3.20) can be
appreciated computing in the operatorial formalism the one closed tachyonone open tachyon
amplitude in two different ways either
0|WT (kL , kR ; z, z )VT (k; x)|0,
|z| > |x|
(3.21)
0|VT (k; x)WT (kL , kR ; z, z )|0,
|x| > |z|.
(3.22)
or
In the second way we find

0|VT (k; x)WT (kL , kR ; z, z )|0,
|x| > |z|
0|:eik(XL (x)y0 ) ::eikL XL (z) ::eikR XR (z) :|0

= ei(kL SkR )y0 |z z |2 kL SkR (x z)2 kkL (x z )2 kkR

D
+D
NN,c
DD
(2)DNN,nc DNN,nc (k + kL + kR )k+k
L +SkR ,0

= ei(kL SkR )y0 |z z |2 kL SkR |x z|2 k(kL +SkR )

x z k(kL SkR )
DNN,c +DDD

L +SkR ,0
x z

DNN,c +DDD
D
= ei(kL SkR )y0 |z z |1 |x z|2 2(k + kL + kR ) NN,nc k+k
,
L +SkR ,0
(3.23)
where DDD is the number of directions with DD boundary condition, DNN,c (DNN,nc ) is the
number of compact (non-compact) directions with NN boundary condition and we have written

the modulus |z z |... and not (z z )... because of the contribution from the cocycle ei (SkL kR )
of closed string vertex cocycle. The last line is obtained using the momentum conservation and
mass shell conditions kL2 = kR2 = k 2 = 1 and shows how the amplitude is conformal invariant
when ghost contribution is added.
222
Let us now consider the computation for |z| > x. According to our previous discussion it
requires the use of the closed string emission vertex depending on Xout 4

WT (kL , kR ; z, z ) = ei SkL kR :eikL XL

=e
The shift of y0 by
2 k
out (z)
i SkL kR
:e
= 0|:e
=e
out (
z)
ikL (XL (z)2 k)
::e
ikR (XR (z)+2 Sk)
:.
is necessary in computing this correlator, in fact
0|WT (kL , kR ; z, z )VT (k; x)|0,

out (z)
ikL XL
::eikR XR
::e
out (
ikR XR
z)
::e
|z| > |x|

ik(XL (x)y0 )
i2 k(kL SkR ) i(kL SkR )y0
:|0
2 kL SkR
|z z |
(z x)2 kkL (z x)2 kkR
+D
NN,c
DD
L +SkR ,0

= ei(kL SkR )y0 |z z |2 kL SkR |z x|2 k(kL +SkR )

z x k(kL SkR )

ei2 k(kL SkR )
z x
D
+D
NN,c
DD
L +SkR ,0

= ei(kL SkR )y0 |z z |2 kL SkR |x z|2 k(kL +SkR )

x z k(kL SkR )
DNN,c +DDD

L +SkR ,0
x z
(3.24)
where the phase from the shift ei2 k(kL SkR ) is necessary to compensate the phase obtained
analytically continuing (also off shell where k (kL SkR ) = (kL + SkR ) (kL SkR ))
k(kL SkR ) to the region x > |z|. In performing this analytical continuation one has to
( zx
z x )
use

log(z w) an. cont. = log(w z) + i sgn arg(z) arg(w) < arg(z),
arg(w) <
(3.25)
which in this case gives [log(z x)]an. cont. = log(x z) + i and its complex conjugate expression [log(z x)]an. cont. = log(x z ) i .
4. Amplitudes factorization
Since we are trying to ensure that all intermediate expressions be well defined with respect
to vertices commutativity property and to keep track of how and where phases arise, the naive
procedure adopted in [8,12] is not completely valid: details on how this can be accomplished are
given in Appendix D, here we give only the main steps.
4 Notice that even for DD directions where X

zero modes = y0 i 2 a0 ln z/z for closed string vertices there is
a non-trivial z dependent normalization factor coming from the splitting into left and right moving part of the string
coordinates, explicitly:

eikL (x+y0 2 ip ln(z)) eikR (xy0 2 ip ln(z))

= ei(kL +kR )y0 e2 kL kR ln(z) e2 p(kL ln(z)kR ln(z)) ei(kL kR )x .
223
We start considering the correlator (which is denoted by a capital A and is not the amplitude
which needs an integration over the moduli space which we consider later)
A(No , Nc ) =
N
o 1

|xi+1 | |xi |
i=1
0|T
No

Vi (xi ; ki ) dxi
i=1
Nc

WLj ,Rj (zj , z j ; kL/Rj ) d zj |0
2
(4.1)
j =1
where the xs are real (both positive and negative), all zs are in the upper half-plane H+ , the
=D
vacuum state |0 is defined as |0 = =1 |p = 0; 0a and it is normalized as
non-compact NN 2(0),
p = 0|p = 0 = non-compact DD 1,
(4.2)
compact 1.
The normalization of the non-compact directions with Dirichlet boundary condition may seem
strange but one has to keep in mind that in this case only |0non-compact DD has finite energy since
the momentum spectrum is wR
; nevertheless the spectral decomposition of unity is still given
by
wR wR

1non-compact DD = lim
(4.3)

R

w
and because of this the non-compact DD case must be understood as a special decompactification
limit: using the naive result would lead to a wrong result. On this correlator we want to perform
the SL(2, C) transformation5
w=
z+i
zi
(4.4)
which maps x R w(x) = ei , z H+ w(z) with |w(z)| > 1 and z w(z) = (w(z))1
by inserting the corresponding SL(2, C) operator. In this way we move form the graphical representation given in Fig. 4 to the one in Fig. 5.
The final answer is given by the following correlator as discussed in Appendix D
N
N
o
o 1

i0
Vi eii ; ki deii
(l l+1 )
A(No , Nc ) = e 0|T
i=1
No

l=1
cL (kLj , kRj ; p)eikLj y0 VLj (wj ; kLj ) dwj
j =1
Nc

cR (kLl , kRl ; Sp)e
l=1
iSkRj y0

VRl

1
d w l
; kRl
|0
w l
w l2
(4.5)
where we have made explicit the y0 dependence by redefining the XL,R in such a way they
are free of y0 and we have introduced the cocycles cL and cR to ensure that the amplitude can
5 This is not a symmetry of the amplitude.
224
Fig. 4. A 3 parallel branes interaction with 3 open

strings and 1 closed one.
Fig. 5. The same configuration as the previous picture with

the usual disk picture. This representation is obtained after
changing variables as in Eq. (4.4).
be obtained by the analytic continuation of whichever radial order. It is important to stress that
thanks to the cocycle the phase 0 is independent of the different specific cases which arise from
the radial ordering.
4.1. Spectral decomposition of the unity in the disk channel
We can now proceed as in [8,12] and we insert a complete set of states at radial time 1 . Since
we are in the open string channel there is only one momentum flowing for each direction and
therefore we want to insert something like 1 = q |q q| as far as the zero modes are concerned.
The naive guess is that we can insert the usual open string version of the spectral decomposition
of unity

n wR n wR
, q, ,
1naive =
(4.6)
, q, ,

R
R
,q,n,w
with continuous momentum q flowing in non-compact NN directions, momentum Rn flowing in

compact NN directions and winding wR
flowing in compact DD ones, unfortunately this guess
is wrong. To understand what is going on we analyze what flows in compact directions. The
momentum conservation reads
Nc
No

(kLj + SkRj ) +
ki = 0
j =1
(4.7)
i=1
(with S = +1 in NN directions and S = 1 in DD directions) and because we are inserting 1 at

radial time 1 this equation can be split into two different relations

1
Nc SkRj + q = 0,
j =1
|w| < 1 ,
(4.8)
q + Nc kLj + No ki = 0, |ei |, |w| > 1 .
j =1
i=1
In compact NN directions the second equation can be written in a more explicit way as

N
No
Nc
c

R
1 1
nj +
(ni + i ) +
wj q = 0
R 2
2
j =1
(4.9)
j =1
i=1
and in compact DD direction as

N
Nc
No
c

R 1
1
nj +
wj +
(wi + i ) q = 0.
2R
2
j =1
225
j =1
(4.10)
i=1
Since winding w (momentum n) of the closed string states must be unrestricted in NN (DD)
n
directions a first guess would be to sum over q = Rw
2 w Z (q = 2R n Z) but this is not yet
completely right. Choosing these values for q Eq. (4.9) for the NN directions becomes
c
o

1
nj +
(ni + i ) = 0
2
j =1
(4.11)
i=1
c
and because of the factor 12 which multiplies N
j =1 nj in this equation, we get a wrong momentum conservation (and similarly for DD directions). On the other hand summing over all
q = 12 ( Rn + wR
) (both n and w) would give a wrong one loop amplitude. The right answer for
generic Rs, even if strange, is the following (closed string like) spectral decomposition of unity6

, {q, w, n} , {q, w, n}
1=
,q,w,n
No
No

wR
n
i=1 ki
i=1 ki

=
,
q,
+
+

2
2
2
2R
,q,w,n
No
No

ki
ki
wR
n
, q, i=1 + , i=1 +
,
2
2
2
2R
(4.12)
where we have a sum of shifted winding for any NN compact direction

and sum of shifted
momentum for any DD compact direction. In non-compact directions q has to be under dq
stood as the result of decompactification limit, i.e. 2
for any non-compact DD direction and
wR wR
limR w | | for any non-compact NN direction. This result is a clue that already at
tree level open string knows of the existence of closed string. On the other side one could argue
that is not completely surprising since we are computing a mixed openclosed amplitude, even
so we can understand better why this happens by noticing that this spectral decomposition of
unity is not what one naively would expect because, while factorizing, we are separating the left
and right part of closed string vertex operators.
4.2. Factorizing the disk amplitude
Inserting the unity given in Eq. (4.12) into Eq. (4.5) we get
6 Notice the use of the notation { } and ( ) in the definitions of vectors.
226
A(No , Nc ) = ei0 e
Nc
j =1 (kLj SkRj ))
iy0 (
0|T
,q,w,n
No

Nc

(i i+1 )
cL (kLj , kRj ; p)VLj (wj ; kLj ) dwj
j =1

Vi eii ; ki deii , {q, w, n}
i=1
N

c

1
d
w
l
cR (kLl , kRl ; Sp)VRl
; kRl
, {q, w, n}T
|0.
w l
w l2

(4.13)
l=1
We can then take the transpose of the piece containing VR s7 and get

iy ( Nc (k SkRj ))
(i i+1 )
A(No , Nc ) = ei1 e 0 j =1 Lj
N

c

0|T
cL (kLj , kRj ; p)VLj (wj ; kLj ) dwj
,q,w,n
No

j =1

Vi eii ; ki deii , {q, w, n}
i=1
0|T
Nc

cR (kLl , kRl ; Sp)VRl (w l ; kRl ) d w l , {q, w, n} , (4.14)
l=1
where in the last line we have used the transposition rule8

T
1
cR (kL , kR ; Sp)VR
= VR (w;
kR )cR (kL , kR ; Sp)w 2
; kR
w
and we have changed the overall phase 0 1 due to the reordering of cocycles.
Now the last line is invariant under the substitution

(NN)
(NN)
kR , XR SXR
, , q, w, n kR , XR (w),
S, Sq, Sw, Sn ,
so we can perform this substitution and then the renaming

(c) (c)

(NN)
q,
XL , XR , , q, w, n XL , X R , ,
w,
n .
(4.15)
(4.16)
(4.17)
7 In performing this transposition we again have to pay attention to the cocycles which ensure that we get the same
phase in the transposed expression as in the original one.
8 We define the transposition of an amplitude as
T
0|V (z1 ; k1 ) V (zN ; kN )|0

= 0|V (z1 ; k1 ) V (zN ; kN )|0

21
1
1
1 2N
1
= 0|V
; kN
V
; k1
|0
z N
z N
z 1
z 1

2N
21
1
1
1
1
= 0|V
; kN
V
; k1
|0.
zN
zN
z1
z1
Notice in particular that p = p as it results from X reality.
227
Finally we can rewrite Eq. (4.14) as follows

A(No , Nc ) = e
i1
0| 0|T
Nc

cL (kj ; p)cR (kj ; p)V
Lj (wj ; kLj )VLj (w j ; kRj ) d wj
2
j =1
No

(i i+1 )Vi eii ; ki deii
i=1
Nc
iy0 (
j =1 (kLj SkRj ))

, {q, w, n} S ,
{S q,
S w,
S n}
.
(4.18)
,q,w,n
This expression involves two different sets of operators, like a and a which can be interpreted
as the left and right moving operators of a closed string; because of this interpretation V is now
(c)
expressed using left moving closed string XL (z) only.
As it is evident from Eq. (4.18) this open string derived formalism treats in a uniform way
compact and non-compact directions, this is not what happens in closed string formalism where
zero modes sectors differ in compact and non-compact directions. It is however not difficult to
see that the two formalisms actually yield the same answer in zero modes sector only when the
NN case is treated as a decompactification limit as stated before.
4.3. The boundary state
In order to derive our final expression for the boundary state we must now take care of the
last three ingredients entering the complete amplitude given in Eq. (2.8) which were left out in
computing the correlator in Eq. (4.1): the normalization factors, the integration region and the
projective volume. The latter is very easily taken care of by fixing a closed string emission vertex
at z1 = ie w1 = ( 0) and an open one at xNo = + No = 0, this means that in
this limit the conformal group volume can be written as
dVabc =
=
d 2 z1 dxNo
i(z1 z 1 )(xNo z1 )(xNo z 1 )
d 2 w1 d(eiNo )
d 2 w1 d(eiNo )
;
i(|w1 |2 1)(w1 eiNo )(eiNo w 1 1) w1 ieiNo |w1 |4
(4.19)
since the amplitude must be invariant under the gauge fixing it follows that the correlator (when
integrated over the remaining variables) is A(N0 , Nc ) eiNo and hence we can decide to only
fix the closed string vertex at z1 = ie and let the open string vertex be free if we integrate
2 iN
the so partially gauge fixed amplitude over 0 de2 o and, at the same time, we transform the
(i i+1 ) into a periodic P (i i+1 ) to ensure only the proper sequence of the vertices.
No
The normalization in Eq. (2.8) left out in Eq. (4.1) is C0(p) No(p)
NcNc . Since we want to interpret
Eq. (4.18) as a closed string amplitude it should be normalized as NcNc 1 since the boundary
state is not a normal closed string state, were it a normal state we would normalize the amplitude
as C0 NcNc where C0 is the closed string sphere normalization but this is not the proper recipe to
obtain the one point closed string emission from the boundary state which reads |B where
228
| is a generic closed string state.9 It then follows that the boundary state must be multiplied
T
No
No
by C0(p) No(p)
Nc = 2p No(p)
. From this expression it is clear that the insertion of No open string
states implies, as expected, only a multiplicative factor NoNo with respect to the bare boundary
state.
Finally remembering the shift of the compact momenta in the partition of unity given in
Eq. (4.12) we can write the boundary state describing multiple parallel interacting Dp branes
in generic position as

No 2

(c) i
ii
i
Dp, {ki , i }{1iN } = ei(p) N No
P (i i+1 ) de Vi XL e ; ki
o
o(p)
No
ei

i=1 0
i=1 ki (
xL SxR
2
No
= ei(p) 2iNo(p)
ei
|Dp
2
N
o 1

(c)
(i i+1 ) deii Vi XL eii ; ki
i=1 0
No
xL SxR
i=1 ki (
2
|Dp,
(4.20)
(c)
where the open string vertices Vi (XL (eii ); ki ) are functionally the same open string vertex
operators as in Eq. (3.14) but now they functionally depend not on the open string fields but on
the left moving part of the closed string fields X (c) : to stress this we have explicitly added (c) to
the notation.
The operatorial phase is given by

ei(p) = ei pL SpR
(4.21)
with pL = pR in non-compact direction as shown in Appendix D.1).

In the first and second lines of Eq. (4.20) we have introduced a further angle No +1 = 1
because we have used a periodic P and we integrate over all the No positions while in the
second we have set No = 0, used a normal and we integrate over the first No 1 positions.
We have also introduced the usual boundary state

1 2iy0 (q, wR , Rn )
n
wR
wR n
2
e
S , S
|Dp = C0(p)
, q, 2 , R S , S q,
2
2
R
,q,w,n
Tp an Sa n
n=1
=
|Dp0
e
2
with the zero modes part
(4.22)
|Dp0 = dpnc 1 (x y0 )|p = 0

n xL +xR
i wR ( xL xR y )
0
NN
DD .
2
e
|0, 0
ei R ( 2 y0 ) |0, 0
(4.23)
9 In principle the closed string vertex normalization N

c in open string formalism must not be equal to the closed string
vertex normalization Nc in closed string formalism, they are however equal N c = Nc as follows from the request that
product of two closed string vertices be equal in closed and open string formulations.
229
To derive this result we have explicitly used the momentum conservation and SDD = 1 and
SNN = +1,10 moreover we have added DD and NN to remember from which open string boundary condition the pieces were originated and we have supposed to have pnc = DNN,nc 1 =
D 1 DDD DNN,c non-compact spacial directions along the brane and we have defined
No xL SxR
C Nc
T
the brane tension 2p = 0(p)
) in
i=1 ki
2 . The meaning of the insertion of exp(i
2
Eq. (4.20) (which gives a non-vanishing contribution for compact directions only) is to divide
in the proper way the open string momentum in the left and right moving momentum in order
o
to ensure that the closed string momentum emitted from the boundary is exactly N
i=1 ki . This
expression was already almost guessed in [13].
The last point which must be clarified is the integration region. From the discussion after
Eq. (4.4) it is obvious that all the moving closed string states vertices are integrated on the external region of the unity disk |w| > 1, this can seem unusual but the complete amplitude associated
with Eq. (4.18) can nevertheless be written as usual in the old fashion as

4c
(c)
(c)
A(No , Nc ) = ei1 NcNc 1
L1 , R1 |WL2 ,R2 (1, 1) WLN ,RN (1, 1)
c
c
perm. of
{2,...,Nc }

4c
B, {ki , i }{1iNo }

where we have introduced the closed string propagator c =
W (1, 1)zL0 z L0
zL0 1 z L0 1 W ( 1z , 1z )
(4.24)

4
d 2 z L0 1 L 0 1
z
|z|<1 |z|2 z
and used
=
and L1 , R1 |(L0 1) = 0. The somewhat strange factors
are due to the definition of the closed string propagator c . Instead the factors Nc are
due to the fact in the old formalism vertices were given without normalization factors.
4c

4.4. The special case of enhanced symmetry

Eq. (4.20) is valid in the case in which branes are in a generic position as shown in a particular
case in Figs. 4, 5 and hence it does not need any ChanPaton factor as explained below Eq. (2.8)
The case with enhanced symmetry can be obtained when more branes are stacked on each other:
in this case we have to associate the usual ChanPaton factor to each vertex as discussed below
Eq. (2.8). Explicitly each open string vertex acquires a ChanPaton matrix as V (XL (x); k) =
V (XL (x); k) V (XL (x); k) = V (XL (x); k) so that Eq. (4.20) becomes trivially

Dp, {ki , i }{1iN }
o

N 2
o

(c) i
i(p) No
ij
No(p) tr
P (i i+1 ) de Vi XL e i ; ki
=e
ei
No
i=1 0
xL SxR
i=1 ki (
2
|Dp.
(4.25)
10 For example to determine the dependence on y along Neumann compact directions, i.e. on y n in Eq. (4.22) we
0
0
Nc
n k n )
iy0n j =1
(kLj
w n R + Nc k n +
Rj
|
start from Eq. (4.18) whose part of interest can be written as 0| 0|e
n
w
j =1 Lj
2
1 No k n | w n R + Nc k n + 1 No k n where we have used S
N N = +1 then using momentum conservation
i=1 i
j =1 Rj
i=1 i
2
2
2
Nc
n k n ) = w n R from which the dependence on y n in Eq. (4.22) follows immediately. The nonwe get j =1
(kLj
Rj
0

compact Neumann case follows taking the decompactification limit.
230
4.5. The boundary No reggeon vertex

Eq. (4.20) is the boundary state which describes from the closed string point of view the
interaction among No parallel branes interacting through No open strings whose quantum numbers are given and equal to i . We want now to write the generating function of all boundary
states with No open string states, i.e. we want to write the boundary No reggeon vertex. To
this purpose we introduce for any of the No open string states an auxiliary Hilbert space with
vacuum 0ai , pi = 0| and the corresponding three reggeon SciutoDella SelvaSaito vertex [14]
(see also Appendix B) so that
0ai , xi = 0|Si (z)|i , ki = Vi (z; ki )
where we have defined

1
Si (z) = :exp
2
(4.26)

du
X(c) (u + z) u Xi (u) :
2i
(4.27)
u=0,|u|<|z|
then the boundary (4.20) can be written as

No

Dp, {No } = ei(p) N No
o(p)

dzi P arg(zi ) arg(zi+1 )
i=1 |z |=1
i
21
0ai , xi = 0|
1iNo
dui
ui =0 2i
duj
uj =0 2i
log(zi zj +ui uj )Xi (ui )Xj (vj )
1j <iNo

No
1
:exp
2
i=0 u =0,|u |<|z |

i
i
i
ei
No
i=1 pi (
xL SxR
2

dui (c)
XL (ui + zi ) u Xi (ui ) :
2i
|Dp
(4.28)
where we have explicitly performed the necessary contractions to normal order the vertices and
pi is the momentum operator acting in the ith auxiliary open string Hilbert space.
As it is written the previous boundary No reggeon vertex is a linear application defined on
the tensor product of No open string dual Hilbert spaces Hopen

to the closed string Hilbert space
Hclosed

Dp, {No } : HNo Hclosed
open
(4.29)
which enjoys the fundamental property of being the generating function of all boundary states
with No open interactions, in fact given No open string states |i we can compute Eq. (4.20) by

1iNo

|i Dp, {i }{1iNo } = Dp, {No }
|i .
(4.30)
1iNo
Once again Eq. (4.28) is valid in the case in which branes are in a generic position and hence it
does not need any ChanPaton factor as explained below Eq. (2.8).
231
4.6. Computing a mixed disk amplitude using the boundary

As an example we can now apply this formalism to recover the complete one closed tachyon
one open tachyon amplitude whose value is (1) the value given in Eq. (3.23) because the gauge
fixing in Eq. (4.19) gives this extra minus sign. The first step is to write the boundary state as

xL SxR

(c)
Dp, {1To } = No(p) ei(p)
dz eikXL (z) eik( 2 ) |Dp
|z|=1
and the out state as
ipa w eikL XL (w)+ikR XR (w)

kL , kR | = lim 0, 0|e
|w|4
a
then we can compute the desired amplitude as

A(1Tc , 1To ) = kL , kR |Dp, {1To }

Tp
i(k)
No(p)
dz kL , kR |
=e
2
|z|=1
nc 1
ikxL 2 kpL ln z dp
(x y0 )|p = 0
i wR ( xL xR y )
n xL +xR
)
0
NN
DD .
2
e
e
|0, 0
ei R ( 2 y0 ) |0, 0
w
n
(4.31)
In computing the previous amplitude we must consider the non-compact Dirichlet directions as
a limit of compact ones so that we can write

Tp
i(k)
A(1Tc , 1To ) = e
No(p)
dz
2
e
x Sx
ik( L 2 R
|z|=1

DNN,nc
2 kL + kR + k

lim e2 k (kL +k) ln z
1 n
e
2 k (kL +k) ln z
n (k +k) ln z
n
L
e2 k

wn
d (k +k)
L
d
e2 k
ln z

nd
kR 2 k +
2R
1
,0 kL + 2 k +
n
2R
k n + 1 k n wn R ,0 k n + 1 k n + wn R ,0 ei
R
2
d 1 k d + nd ,0
kR
2
2R
2
d + 1 k d + nd ,0
kL
2
2R
ei
,0
ei R y0
wn R n
y0

nd d
R y0
(4.32)
where the first line is from non-compact Neumann directions, the second line from non-compact
Dirichlet directions, the third line from compact Neumann directions and the fourth line from
compact Dirichlet directions. We have used the index n () for (non)-compact NN directions and
d () for (non)-compact DD ones.
The Kronecker s in non-compact Dirichlet directions imply that the closed tachyon momentum is unconstrained while the open tachyon momentum which is interpreted as a distance
between two branes must vanish k = 0.
xL SxR
It is also worth noticing as the insertion of eik( 2 ) in the boundary state is fundamental
for finding a consistent solution of the compact s, explicitly for the Neumann directions we find
n
+ w n = 0; would the insertion not be present we would have
nnclosed + nnopen + n = 0 and wclosed
232
Fig. 6. One brane is gravitationally interacting with 3

parallel branes which interact among themselves with 3
open strings.
Fig. 7. The previous figure interpreted in the open channel.
found an inconsistent (with the translation invariance) linear system, i.e. 12 nnclosed +nnopen + n = 0
from the left moving sector and 12 nnclosed = 0 from the right moving one.
Finally we get

Tp

dz ei(kL SkR )y0 e2 (kL +k)k ln z
A(1Tc , 1To ) = ei(k) No(p)
2
|z|=1

D
D +DNN,c
2(kL + kR + k) NN,nc kLDD
+SkR +k,0

D
D +DNN,c
i(k)
i(kL SkR )y0
= ie
2(kL + kR + k) NN,nc kLDD
C0(p) Nc No(p) e
+SkR +k,0
(4.33)
where we have used the mass shell conditions k 2 = kR2 = kL2 = 1 to write
kL +SkR
2
e2 (kL +k)k ln z = e2 (

= e (k
+k+
kL SkR
2
)k ln z
2 (k Sk )(k +Sk )) ln z
L
R
L
R
= z1
T
(4.34)
N
c
in order to perform the z integration and the relation 2p = 0(p)
from Eq. (4.22) to rewrite the
2
normalization coefficient. Finally because of the momentum conservation 2 kL SkR = 1 the
phase factor is

ei(k) = ei kL SkR = i
(4.35)
which together the factor i already present gives the required factor 1.
4.7. One loop N tachyons amplitudes
As a final example of application of the interacting boundary state we want to compute the
complete one loop planar amplitude having N open string tachyons, normalization included.
Explicitly we compute the gravitational interaction between one Dp brane and N interacting
parallel Dp branes in a generic position as it is depicted in Fig. 6 in a special case, then we
reinterpret this result in the open channel as in Fig. 7. The closed string computation is given by
233
AN tach,1 loop = Dp|c |Dp, {N}
|ki
1iN
= Dp|c |Dp, {ki }{1iN}
2

Tp
N
i(p)
= No(p)
Dp|e
d e (L0 +L0 2) L0 ,L 0
2
2
0
2N 1

ei(p) 2i
0
ei
(i i+1 ) deii
eii eij
2 ki kj
i<j
i=1
(c)() ii
(e )
i ki XL
ei
(c)(+) ij
(e )
j kj XL
ei
i ki (
xL SxR
2
|Dp
(4.36)
where N = 0 and |Dp is the plain boundary state associated with a Dp brane with boundary
conditions dictated by the reflection matrix S. A straightforward computation gives
AN tach,1 loop
2 2N
1
Tp
N
= (2) i No(p)
d
deii (i i+1 )
4
2
2
i=1

DNN,nc
2 DDD,nc /2 2 (y y0 + i ki i )2
DDD,nc
ki
k
e
2

i i
i
1 ( wn R )2 i wR (y n y n + k n )
D
i i i

NN,c
e 2
e 0
n
k
i i
D
DD,c
d
i ki
n=1
ii
1
1 e2 n
n 2
e 2 ( R ) ei R (y y0 +
D2
kd )
i i i

2 ki kj

1 2 cos ij e2 m + e4 m
(1 e2 m )2
i<j

ij 2 ki kj
m=1
(4.37)
where the two terms in the sixth line are due to non-zero modes (the ghosts account for the 2
in the exponent D 2) with ij = i j , in the fourth and fifth lines we have the compact
zero modes contributions and in the third line there are the contributions from non-compact
zero modes: the non-compact contribution from DD directions must again be understood as the
decompactification limit since we otherwise would miss the conservation of distances which is a
Kronecker delta and not a Dirac delta. As before we have used the index n () for (non)-compact
NN directions and d () for (non)-compact DD ones.
In Eq. (4.37) the contribution from non-zero modes was easily computed using
s0 a a ea A eaB es a
0, 0|e
=
|0, 0
ABs0 s
1
e 1s0 s .
1 s0 s
(4.38)
The string amplitude (4.37) for the special case of D25 (S = 1) with all non-compact directions reduces to the well know 1 loop N tachyons open string amplitude:
234
AN tach,1 loop
N
= C1(25) No(25)
(2)26 26

i
n=1
i<j
1
1 e2 n
D2

ki
2
d
2
2N1

0
i=1
di
(i i+1 )
2
ij 1 2 cos ij e2 m + e4 m
sin
2
(1 e2 m )2
2 ki kj
upon the use of momentum conservation, mass shell condition ki2 = 1, 2

and N = 0 to write
N1

deik
N1

de
eii eij
i<j ki
kj = N
2 ki kj
N1

k=1
ik

i +j
ij 2 ki kj
i 2 2 ki kj
e
2i sin
2
i<j
k=1
1i<j N
k=1
(4.39)
m=1

M1

ij 2 ki kj
ij 2 ki kj
1 dk
dk
=
.
2i sin
sin
2
2i
2
2
i<j
k=1
(4.40)
i<j
5. Conclusions
In this article we have shown how to compute boundary states which describe N parallel
branes in generic position, i.e. not superimposed, interacting through N open string states in
Eq. (4.20). We have also computed the corresponding boundary states when the branes are superimposed as in Eq. (4.25).
The result is not completely unexpected but there are some points which are not trivial:
when acting on the trivial boundary state with open string vertex operators we have to substitute Xopen with the left moving part of the closed string XL in order to obtain a T-duality
invariant formulation;
the non-compact Dirichlet directions must be treated as a decompactification limit;
in compact directions we have to equally divide the momenta coming from open string vertices between left and right closed string momenta, as it is explicit in Eq. (4.20).
Another point which is worth noticing is that this derivation is valid for the bosonic free string
but this could be in principle extended to other CFTs since the fundamental starting point is
the possibility of writing the closed string vertices as a product of two open string ones. This
possibility is quite natural when one starts from the open string model as discussed around
Eq. (2.7): in particular this is clear for the bosonic/NSNS gravitational sector which couples
universally in both bosonic and fermionic strings but it is also true for the fermionic and RR
sectors where branes couple only to a subsector of all the possible closed string states.
235
Acknowledgements
The author thanks M. Bill and A. Lerda for discussions and the Niels Bohr Institute for
the hospitality during the completion of this paper. This work supported in part by the European
Communitys Human Potential Program under contract HPRN-CT-2000-00131 Quantum Spacetime. It is also partially supported by the Italian MIUR under the program Teoria dei Campi,
Superstringhe e Gravit.
Appendix A. Conventions
Indices:
= 0 . . . d 1 for non-compact directions and a = d . . . D 1 for compact ones; n () for
(non)-compact NN directions and d () for (non)-compact DD ones.
Amplitudes normalizations.

If the d-dimensional YM action is given by SYM = d d x tr(F F ) with F = A
A igd [A , A ], A = Aa a and tr(a b ) = 12 ab , tr(a [b , c ]) = 2i f abc , it is possible to derive from direct comparison with field theory as in Ref. [10]
No(d1) = gd 2 ,
C0(d1) =
1
,
(gd )2
Ch+1(d1)
(gd )2
=
Ch(d1)
(2)d (2 )d/2
and from unitarity

2
C0(d1) No(d1)
= 2
1

when using o = 0 dx x L0 2 (L0 = p 2 +

1 nan an ). In a similar way it follows from
closed string unitarity
C0 N c2 = 4

d 2 z L0 1 L 0 1

when using as closed string propagator c = 4
z
(L0 = pL2 +
|z|<1 |z|2 z

2
n a n ).
1 nan an , L0 = pR +
1 na
The vertex normalization factors No(d1) , Nc and N c are common to all the states but depend
on the number of compact and non-compact directions.
Vacua normalizations in non-compact directions.
For each non-compact direction of open string with NN b.c. or of closed string: k|k =
2(k k ) with k| = |k , |x = 0 (x)|p = 0.
For each non-compact direction with DD b.c. since only |0
has finite energy: 0|0 = 1
but the spectral decomposition of the unity is 1 = limR w |w w|; For each compact
direction n|m = n,m .
z C < arg(z) , z H+ 0 arg(z) .
Open string modes expansions:

1
XL (z) + XR (z) ,
2
sgn(n)
XL/R (z) = q y0 i 2 a0 ln z + i 2
an zn ,
|n|
n=0
X(z, z ) =
236
where z = eE +i H+ (E is the Wick rotated time) and the logarithm entering the string
expansion is defined
to have a cut at arg(z) = , i.e. < arg(z) + moreover log(z)
log(z) and a0 = 2 p in the NN case.

The commutation relations are (an = an )

am , an = m,n , m, n > 0.
q , p = i ,
The spectrum of the momentum operator is

non-compact NN b.c.,
k
n
n
n
n +0
compact NN b.c.,
R
k = y y
non-compact DD b.c.,
2
d
d
wd R y y0
compact DD b.c.,
+ 2
where y0 (y ) is the position of the brane at = 0 ( = ) and 20n (2n ) is the Wilson
line on the brane at = 0 ( = ).
The closed string modes expansion is given in analogous manner by

1 (c)
(c)
X(z, z ) = XL (z) + X R (z)
2
where now z = e2E +2i C, XL has the same expansion as for the open string but with
(c)
(c)
y0 = 0 and X R expands as XL but with tilded operators, the commutation relations read

qL , pL = qR , pR = i ,
= a Rm , a Rn
= m,n , m, n > 0.
aLm , aLn
(c)
In the non-compact case we identify xL = xR = x and pL = pR = p2 where p has continuum

spectrum while in compact directions pL (pR ) has discrete spectrum given by kL = 12 ( Rn +
wR
1 n
wR
) (kR = 2 ( R )).
Appendix B. Phases and analytic continuations
We start writing the three reggeon SciutoDella SelvaSaito vertex [14] for a single coordinate

du
1
Xaux (u + z)u X(u) :,
S(z) = :exp
(B.1)
2
2i
u=0,|u|<|z|
where the normal ordering is taken with respect to both the auxiliary Xaux
and the usual X(u)
XL (u). Performing the integral over u the previous equation becomes ( 0 1)11

i nan n
S(z) = :exp
(B.2)
Xaux (z) :.
n!
2
n=0
11 In the case of a non-compact direction the left and right moving parts cannot be factorized because of the zero modes
therefore the complete vertex reads:

i
n
n
n
S(z, z ) = :exp
a0 Xaux (z, z ) +
an + a n Xaux (z, z ) :.
n!
2
n=1
237
This vertex can be thought of as a generating functional of all vertices12 :

0a , x = 0|S(z)|, k = V(aux) (z; k).
(B.3)
We use this formalism in order to exam in general the phases occurring when performing the
analytic continuation of the product of two such vertices. We first compute the product of two
SDS vertices
S(z)S(w) = :S(z)S(w)
1
e 2
du
u=0 2i
= :S(z)S(w)e
dv
v=0,|vu|<|zw| 2i
log(zw)2 p (1) p (2)
log(zw+uv)X(1) (u)X(2) (v)
: |z| > |w|
(B.4)
in this expression the log has to be interpreted as a shorthand version of

log(z w + u v) log(z)

1 w n
n=1

1 vu n
n=1
n zw
and we have made use of the well-known expression

X(z)X(w) = :X(z)X(w): 2 log(z w) |z| > |w|
(B.5)
Then we compare this Eq. (B.4) with the analogous product S(w)S(z) for |z| < |w| and we use

log(z w) an. cont. = log(w z) + i sgn arg(z) arg(w) < arg(z),
arg(w) <
(B.6)
we get the desired phase13

S(z)S(w) an. cont. = S(w)S(z)e+i sgn(arg(z)arg(w))2 p0(1) p0(2)
(B.7)
with < arg(z), arg(w) + . In the generic case the phase of the previous equation becomes

sgn arg(z) arg(w) 2(nz nw ) + sgn [z ] [w ] ,
(B.8)
where we have defined arg(z) = [z ] + 2nz with < [z ] and nz Z.

12 This yields the proper expression for the open string vertex only for the emission from = 0 when z = x, but the
right expression of the SDS vertex for emission from the = border is
S ( ) = :S( ) exp(i2 p0 p0(aux) ):
The necessity of the cocycle follows from the obvious request of commutativity of the product of a vertex for the emission
from = 0 border with one from = border.
13 In a similar way we deduce that in the non-compact case the vertices commute

z )
S(z, z )S(w, w)
an. cont. = S(w, w)S(z,
because of X(z, z )X(w, w)
= :X(z, z )X(w, w):
log |z w|.
238
Appendix C. Details on the cocycles computations

C.1. Closed string case
In this section we give the details on the derivation of the closed string cocycles. In Section 3.1
we introduced the phase ei(c) (k1 ,k2 ) as the phase which arises when computing the normal ordering of the product of two closed string vertices W, explicitly
(c)
WL ,R (z, z ; k1 )W(c)
(w, w;
k2 )
L ,R
= ei(c) (k1 ,k2 ) c(kL1 + kL2 , kR1 + kR2 ; pL , pR )
kR2 )
VL (z; kL1 )VL (w; kL2 )VR (z; kR1 )VR (w;
(C.1)
with

(c) (k1 , k2 ) = (BkL2 + CkR2 ) kL1 + (DkL2 + EkR2 ) kR1
(C.2)
where the matrices B, C, D and E may only have non-vanishing entries in compact directions,
i.e. for example Bba .
With the help of the well known formula ( arg(z), arg(w) + , |z| < |w|) (see also Appendix B)

V (z; k1 )V (w; k2 ) an. cont. = V (w; k2 )V (z; k1 )e+i sgn(arg(z)arg(w))2 k1 k2
and by comparing with the analogous product of the vertices in the opposite order
(c)
(w, w;
k2 )WL ,R (z, z ; k1 )
W(c)
L ,R
it is not difficult to see that the constraint we want to impose on the cocycles coefficients in order
to implement the commutativity is14

ei(c) (k1 ,k2 )i(c) (k2 ,k1 )+i sgn(arg(z)arg(w))2 (kL1 kL2 kR1 kR2 )

= ei [(BB

T )k k +(E E T )k k ]
L2 L1
R2 R1
ei [(C D
T )k k (C D T )T k k 2 sgn(arg(z)arg(w))(k k k k )]
R2 L1
L2 R1
L1 L2
R1 R2
= 1. (C.3)
The quantities entering the previous equation can be trivially evaluated, in particular for any
compact direction we have
a b

Ra
Rb
a b
2 kL1
kR2 kR1
kL2 = w1a nb2 b w2b na1 a ,
R
R
a

R
Rb
a b
a b
kL2 kR1
kR2 = w1a nb2 b + w2b na1 a .
2 kL1
(C.4)
R
R
Considering the cases where the momentum and winding are different from zero only in a given
compact direction a we deduce immediately that we have to impose the following restriction on
the coefficients entering the cocycles definition (3.6)

C DT aa = 4Na + 2, Na Z,
(C.5)
14 Because of (B.8) and (C.4) the same constraint holds in the world-sheet Minkowskian version where arg(z) = + .
239
and
Baa , Eaa R.
(C.6)
If we consider the cases where only two compact directions have kL/R = 0 and we choose their
radii not to be equal we deduce that the off-diagonal elements must be identically zero:

C DT ab = 0 (a = b),
Bab = Eab = 0.
(C.7)
A further constraint comes from Hermitian property of the vertex: the Hermitian of a vertex
is given by the following expression

(c)
1 1
(c)
i(c) (k,k) 1
W
, ; k
WL ,R (z, z ; k) = e
(C.8)
|z|2 L ,R z z
from which we find

ei(c) (k,k) = ei [BkL kL +(C +D

= ei [Baa kLa +(C +D

2
T )k k +E k k ]
R L
R R
T )k k +E k 2 ]
aa Ra
Ra La

= ei [B+E +(C +D)]aa ( R ) ei [B+E (C +D)]aa (

na 2
=1
wa R 2
)

ei[BE ]aa na wa
(C.9)
and finally15
Caa = Daa = 1 + 2Na ,
Baa = Eaa = 2Ma
(C.10)
where Ma are arbitrary integers. In the text we will use

Na = 0,
Ma = 0.
(C.11)
Further constraints come from unitarity which requires that any three points amplitude
A(1, 2, 3) be connected with the amplitude A(1, 2, 3) for the CPT conjugate particles
((1) means the CPT conjugate state of 1) by

A(1, 2, 3) = A(1, 2, 3) .
(C.12)
a , k a ))
This equation can be specialized to the case of three tachyons with momenta ki = (ki , (kLi
Ri
(i = 1, 2, 3) and results in
! "

ki kLi kRi .
A(T1 , T2 , T3 ) ei 1ij 3 (c) (ki ,kj ) d
(C.13)
If we consider sequentially configurations with only one winding, two windings different from
zero, analogously for momenta and finally configurations with mixed momenta and windings
and we require a smooth behavior of the cocycles in the large and small Ra limits we get the
results given in the main text.
15 We could also introduce a non-operatorial cocycle also for the closed vertices and then avoid these conclusions but
this is an avoidable complication.
240
C.2. Open string computation

In this appendix we give the details on how to obtain Eqs. (3.13), (3.14) and Eq. (3.15). To
this purpose we write the generic closed string emission vertex in open string formalism with the
cocycles as

WL ,R (z, z ; k) = c(kL , kR , p)VL XL (z); kL VR XR (z); kR
(C.14)
with
c(kL , kR , p) = cnop (kL , kR )cop (kL , kR ; p),

cop (kL , kR ; p) = ei (ukL +vkR )p ,

cnop (kL , kR ) = e
i (kL kL + kL kR +kR kR )
(C.15)
,
= ,
T
= ,
T
(C.16)
where cop (cnop ) is the (non)-operatorial part of the cocycle and kL = kR = k/2 in non-compact
direction. Our aim is now to determine the (matrix) coefficients u, v, , , in order to reproduce
the commutativity, the OPE coefficients in Eq. (3.7) or Eq. (C.1) and the behavior of closed
string vertices under Hermitian conjugation (C.8). It is worth stressing that the previous expression for the cocycles (C.16) contains a contribution from all directions, even the directions with
DD boundary condition since the coefficient in the string modes expansion is an operator and
hence commuting the cocycles with XL/R yields a non trivial phase. If compute the phase which
we obtain normal ordering the product of two vertices as done in (3.7) and we remember that
NN (where X NN is the right moving part with NN boundary conditions) we get
XR = SXR
R
ei(o) (k1 ,k2 ) =
cnop (k1 )cnop (k2 ) i (ukL2 +vkR2 )(kL1 +SkR1 ) i2 SkR1 kL2
e
.
e
cnop (k1 + k2 )
(C.17)
In this expression the first factor of rhs is due to non-operatorial cocycles, the second to the
reordering of the operatorial ones and the last to the reordering of VR (kR1 )VL (kL2 ). We require
then that the closed string phase (C.2) in closed string formalism and the corresponding phase
(C.17) in open string formalism be equal, i.e.
ei(o) (k1 ,k2 ) = ei(c) (k1 ,k2 )
(C.18)
which is the necessary condition for the correct factorization of mixed open/closed string amplitudes in the closed channel. Given this basic constraint (C.18) we can check that the further
constraints which arise from requiring the commutativity of two closed string vertices and the
proper behavior under Hermitian conjugation are identically satisfied. In particular, the phase
which arises when computing the Hermitian of a vertex is (o) (k, k) analogously to what we
have found with the closed string formalism (C.8).
The constraint (C.18) can be explicitly rewritten as

ei [(BkL2 +C kR2 )kL1 +(DkL2 +E kR2 )kR1 ]

= ei [2kL1 kL2 +2kR1 kR2 + kL1 kR2 + kL2 kR1 +(ukL2 +vkR2 )(kL1 +SkR1 )+2SkR1 kL2 ] .
By considering different combinations of windings and momenta as done for the closed string
cocycles after Eq. (C.3) and using Eqs. (C.4) we derive the following constraints:

2 + (1 + S)u + 2S = 0,
(C.19)
= = v = 0
241
(up to a gauge choice which allows to set = = v = 0) for matrix indexes in the non-compact
directions and
(2 + u B)ab
=
(E 2 Sv)ab
= 2 diag(M a + 2D a )
(C.20)
for kL1a kL2b and kR1a kR2b coefficients in compact directions and
T

+ v C
=
(D + + Su + 2S)ab
= 2 diag(M a )
ab
(C.21)
for kL1a kR2b and kR1a kL2b coefficients. The M a are arbitrary integers because of Eqs. (C.4).
Consistency of the two last equations (C.21) for implies that
v (Su)T = 2 + 2S + 4M + 4N.
(C.22)
Next we want to exam the open string/closed string vertices commutativity. To this purpose
we write the open string emission vertex from the = 0 boundary as

V (x; k) = ei ukp
(C.23)
V XL (x)y0 ; k , x > 0
where u is an unknown matrix. The naive way of writing this commutativity condition (which is
wrong!)

WL ,R XL (z), XR (z) V (x; k) an. cont. = V (x; k)WL ,R XL (z), XR (z)
implies that

L +SkR )] = 1
ei [(ukL +vkR )k+2k(kL SkR )uk(k
which can be satisfied when

u + u T (1 + S) 2(1 S) = 0,

u u T + 2 ab = v u T + 2 S ab = 0
(C.24)
(C.25)
(C.26)
respectively for matrix indices in non-compact and compact directions. In particular the equations in the last line arise since Eq. (C.24) must be valid also when Wilson lines are turned on,
i.e. when k = n+
R .
In a similar way we get further constraints requiring that the generic vertex for the open string
emission from the = border

V (x; k) = ei u kp V XL (x)y0 ; k
(C.27)
commute with both closed string and = 0 open string vertices.
The coefficient u can be fixed by requiring the commutativity of = open string vertices
the with = 0 open string vertices; it turns easily out that
u = u 2.
(C.28)
This in turn implies that the commutativity of an open string emission vertex from = with a
closed string vertex does not really yield new constraints but some consistency conditions which
read

u + u T (1 + S) = 0,
(C.29)

u u T ab = v u T S ab = 0.
(C.30)
242
Unfortunately or better correctly the sets of constraints (C.25)(C.26) and (C.29)(C.30) are
inconsistent. This would seem to be a disaster but luckily it is not so since the proper way of
formulation open/closed string commutativity reads, as explained in the main text, is to take into
account the y0 shift so that

out
WL ,R XLout (z), XR
(z) V (x; k) an. cont.

in
= V (x; k)WL ,R XLin (z), XR
(z)

out
(z) e2 ik(kL SkR ) .
= V (x; k)WL ,R XLout (z), XR
As a consequence of this proper commutativity condition all the constraints from open/closed
string vertices commutativity are now consistent and turn out to be Eq. (C.30) because of the
extra phase contribution from the proper commutativity condition.
From Hermiticity conjugation of V vertices we get
u + u T = 0
(C.31)
which gives using Eqs. (C.29), (C.30) and the consistency condition (C.22)
u = u = v = 0,
2M + 2N + 1 + S = 0.
(C.32)
Finally using Eqs. (C.10) we get

T = (1 + 2N + 2M)
T = S,
= (C + 2M)
1
1
= = B + M + 2D = E + M + 2D = M + M + 2D.
2
2
With our choice (C.11) and choosing M = D = 0 we finally find
= S,
= = 0.
(C.33)
(C.34)
It is interesting to notice that the Hermiticity of the open string vertices V (x) does not yield
2

any further constraint since the phase ei2 k which is obtained reordering the cocycle ei u kp
after taking the Hermitian conjugate is precisely the one needed to express the Hermitian as a
i
1
1
function of (x)
= e|x| ; explicitly we get
=
(ei |x|)

2 k 2
1
1
V (x; k) = e
V
; k
x
(x)

i 2 k 2
1
e
V
; k .
=
|x|
(x)
i2 k 2
(C.35)
Appendix D. Changing variables in the open string amplitude

In this appendix we would like to discuss some subtleties in performing the SL(2, C) change
of variable w = z+i
zi , which is not a symmetry, on the correlator (4.1). Naively one would say that
the answer is
N

2
Nc
o

i
i
d wj
1
i
i
(i i+1 ) 0|T
Vi e ; pi de
WLj ,Rj wj ,
; kj
|0
w j
w j2
i=1
j =1
(D.1)
243
where we now use open string vertices with the trivial cocycle because there is not anymore any
difference between emission from real positive axis and real negative one.
If we look more carefully we realize that the two correlators are related in a non-trivial way
by various analytic continuations:
in the SL(2, C) parameters entering the operator realizing the wanted SL(2, C) transformation;
in the order of the operators since it can happen that a SL(2, C) transformation changes the
radial ordering, i.e. it does not preserve the absolute of |z|;
in the log which enters the string expansion since w and w can move independently on
different sheets because the phases of w and w can exceed the range ], ] and the phase
of w can also be not the opposite of w.
Moreover there is a further overall phase ambiguity due to the possible ordering choices of open
string vertices whose Euclidean times are now the same, of which Eq. (D.1) is a particular one.
Since it is very difficult to keep track of these analytic continuations, we are therefore led
to use an indirect way to compute the phase which arises performing the transformation (4.4)
on Eq. (4.1): we compare the explicit expression for (4.1) obtained using the reggeon vertex
formalism on which we perform the change of variables (4.4) with the corresponding explicit
result of the amplitude (No +1 = )
N
o

i0
Vi eii ; pi deii
A(No , Nc ) = e 0|T
i=1
Nc

cL (kLj , kRj ; p)eikLj y0 VLj (wj ; kLj ) dwj
j =1
Nc

cR (kLl , kRl ; Sp)eiSkRj y0 VRl
l=1

1
d w l
; kRl
|0
w l
w l2
(D.2)
where we have made explicit the y0 dependence by redefining the XL,R in such a way they are
free of y0 and we have introduced the cocycles cL and cR 16 to ensure that the amplitude can be
obtained by the analytic continuation of whichever radial order we choose or, said in different
words, commuting VLj and VLl produces a phase

ei ((C kLl kRj kRl C kLj )+(BkLj kLl BkLl kLj )+2 sgn(arg(wj )arg(wl ))kLj kLl )
which is canceled by the phase

ei ((DkLj kRl kRj DkLl )(E kRl kRj E kRj kRl )2 sgn(arg(wj )arg(wl ))kRj kRl )
we get while commuting the corresponding right vertex operators. To derive the last equation we
have made use of the fact that arg(w)
= arg(w); as it was stressed before this relation could not
otherwise have been taken for granted if we had performed the transformation by inserting the
SL(2, C) operators. It is important to stress that thanks to the cocycle the phase 0 is independent
of the different specific cases which arise from the radial ordering.
16 These cocycles are the same as in closed string (3.6) with the substitution of the closed string operators p , p with
L R
the open string operator p.
244
D.1. Determining the phase ei(p)

In order to compute the phase ei(p) we generalize the computation in Section 4.6 and we
compare the 1 closed tachyon-No open tachyons amplitude at the tree level computed in open
and closed string formalism. This is enough to fix the phase since any amplitude involving a
boundary state can be factorized on a one point closed string with boundary amplitude times a
closed string amplitude and phases arise only from momenta contributions.
The 1 closed tachyon-No open tachyons amplitude in open string formalism when all directions are compact reads
No
A(No To , 1Tc ) = C0(p) No(p)
Nc

dxNo
N
o 1
i=2
2 kL SkR iy0 (kL SkR )

dxi (xi+1 xi ) (2)
(xi xj )2 ki kj
2i<j No
2 kj kL
(xj i)
2j No
(xj + i)2 kj kR k
No
L +SkR +
j =1 kj ,0
(D.3)
where we have gauge fixed the SL(2, R) invariance with z = i and x1 +. The contribution

(2) is what is left of dVabc and 22 kL SkR is the closed string tachyon contribution after the
gauge fixing.
Changing variables to the disk ones as described in the main text, i.e. w = z+i
zi so that w(x) =
i
e and using the momentum conservation and mass shell conditions we get
A(No To , 1Tc )

No
Nc i(i)2 kL SkR
= C0(p) No(p)

No
0 j =2
2 kL SkR
eiy0 (kL SkR )

deij (j +1 j )
eii eij
2 ki kj
2i<j No
eij
2 kj kL
2j No
No
L +SkR +
(D.4)
j =1 kj ,0
where we have set 1 = 0. The corresponding amplitude computed using the boundary is given
by
A(No To , 1Tc )
Tp No i(k)
ie
= 2 No(p)
2

No

deij (j +1 j )
0 j =2
22 kL SkR eiy0 (kL SkR )
2i<j No
L +SkR +
No
j =1 kj
.
,0
eii eij
2 ki kj
eij
2 kj kL
2j No
(D.5)
Comparing these two expressions it follows that the generic phase can be written in an operatorial way as

ei(p) = ei pL SpR
245
(D.6)
where the operatorial momenta pL and pR must be identified in non-compact directions.

References
[1] J. Polchinski, Dirichlet-branes and RamondRamond charges, Phys. Rev. Lett. 75 (1995) 4724, hep-th/9510017.
[2] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Adding holes and crosscaps to the superstring, Nucl. Phys. B 293
(1987) 83;
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Loop corrections to superstring equations of motion, Nucl. Phys.
B 308 (1988) 221;
J. Polchinskim, Y. Cai, Consistency of open superstring theories, Nucl. Phys. B 296 (1988) 91.
[3] P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Classical p-branes from boundary state, Nucl.
Phys. B 507 (1997) 259, hep-th/9707068.
[4] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, I. Pesando, Fractional D-branes and their gauge duals,
JHEP 0102 (2001) 014, hep-th/0011077.
[5] A. Sen, Rolling tachyon, JHEP 0204 (2002) 048, hep-th/0203211.
[6] D. Gaiotto, N. Itzhaki, L. Rastelli, Closed strings as imaginary D-branes, Nucl. Phys. B 688 (2004) 70, hepth/0304192.
[7] C. Lovelace, Pomeron form-factors and dual Regge cuts, Phys. Lett. B 34 (1971) 500;
L. Clavelli, J.A. Shapiro, Pomeron factorization in general dual models, Nucl. Phys. B 57 (1973) 490.
[8] M. Ademollo, A. DAdda, R. DAuria, E. Napolitano, P. Di Vecchia, F. Gliozzi, S. Sciuto, Unified dual model for
interacting open and closed strings, Nucl. Phys. B 77 (1974) 189.
[9] I. Pesando, On the effective potential of the Dp Dp-bar system in type II theories, Mod. Phys. Lett. A 14 (1999)
1545l, hep-th/9902181.
[10] P. Di Vecchia, L. Magnea, A. Lerda, R. Russo, R. Marotta, String techniques for the calculation of renormalization
constants in field theory, Nucl. Phys. B 469 (1996) 235, hep-th/9601143;
G. Cristofano, R. Marotta, K. Roland, Unitarity and normalization of string amplitudes, Nucl. Phys. B 392 (1993)
345.
[11] J. Polchinski, String Theory, Cambridge Univ. Press, 1998.
[12] M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Scattering of closed strings from many D-branes, Phys. Lett.
B 400 (1997) 52, hep-th/9702037.
[13] M. Bill, D. Cangemi, P. Di Vecchia, 1996, unpublished.
[14] S. Sciuto, The general vertex function in dual resonance models, Lett. Nuovo Cimento 2 (1969) 411;
A. Della Selva, S. Saito, Lett. Nuovo Cimento 4 (1970) 689.

Anomaly induced effects in a magnetic field

Ignatios Antoniadis a,1 , Alexey Boyarsky a,2 , Oleg Ruchayskiy b,
a Department of Physics, CERN, Theory Division, 1211 Geneva 23, Switzerland
b Ecole Polytechnique Fdrale de Lausanne, Institute of Theoretical Physics, FSB/ITP/LPPC, BSP 720,
CH-1015 Lausanne, Switzerland

Received 14 September 2007; accepted 3 October 2007
Abstract
We consider a modification of electrodynamics by an additional light massive vector field, interacting
with the photon via ChernSimons-like coupling. This theory predicts observable effects for the experiments studying the propagation of light in an external magnetic field, very similar to those, predicted by
theories of axion and axion-like particles. We discuss a possible microscopic origin of this theory from
a theory with non-trivial gauge anomaly cancellation between massive and light particles (including, for
example, millicharged fermions). Due to the conservation of the gauge current, the production of the new
vector field is suppressed at high energies. As a result, this theory can avoid both stellar bounds (which exist
for axions) and the bounds from CMB considered recently, allowing for positive results in experiments like
ALPS, LIPPS, OSQAR, PVLAS-2, BMV, Q&A, etc.
PACS: 12.60.-i; 14.80.-j; 14.80.Mz; 11.10.Kk
Keywords: Anomalies and anomaly cancelations; Millicharged particles; Stellar bounds on new light particles; Effects
in strong magnetic fields
1. Introduction
A number of experiments, studying the properties of vacuum in a strong external magnetic
field (in particular the propagation of light in such a field), have been performed over the recent
E-mail address: oleg.ruchayskiy@epfl.ch (O. Ruchayskiy).

1 On leave from CPHT (UMR CNRS 7644) Ecole Polytechnique, F-91128 Palaiseau.
2 On leave of absence from Bogolyubov Institute of Theoretical Physics, Kyiv, Ukraine.
doi:10.1016/j.nuclphysb.2007.10.006
I. Antoniadis et al. / Nuclear Physics B 793 (2008) 246259
247
years, or are being in operation at the moment [111]. One of the main motivations for these
experiments is the search for a hypothetical particle, the axion, which was originally predicted
[12,13] by the PecceiQuinn mechanism of solving the strong CP problem in QCD [14,15].
The main property of this particle from the laboratory experiments point of view is the coupling
of the axion field a(x) to the photon via the interaction term 14 a(x)F F = a(x)(E H ). More
generally, axion-like-particles (ALP) are pseudo-scalars which interact with F F ,3 but may not
have other properties of the QCD axion needed to solve the strong CP problem.
In [16] and [17,18] another class of effective theories, which also predict effects in the presence of electric and magnetic fields, having completely different particle physics motivation,
were considered. The coupling of axion to F F appears in general from chiral gauge anomalies,
that may be present in theories with chiral couplings of fermions to a gauge field. Although
gauge anomaly makes theory inconsistent, the theories with several chiral fields may be welldefined if the overall anomaly cancels and the overall gauge current is conserved. The fact that
the net vacuum current is conserved does not mean that it is zero. If it is not, the non-trivial
anomaly cancellation between heavy and light fields may give rise to experimental signatures
of the heavy fields which, due to the topological nature of anomalies, are not suppressed by
the mass of the heavy fermions and, therefore, may be observed at low energies [18]. For theories with non-trivial electromagnetic anomaly cancellations, such effects will be proportional
to F F .
In this paper, we discuss in more details the theory considered in [16] and, in particular, show
how the creation of the new light particles, present there, may be suppressed with energy and,
therefore, the parameters of the theory may not be constrained by the stellar bounds, which
severely restrict the parameter space of axions (for a recent review, see e.g. [19]). We describe
the possible origin of the effective Lagrangian of [16] and show that it may appear, for example,
from a theory of millicharged fermions, interacting with the photon and a paraphoton similar
to the theory considered in [20,21]. This allows to have the effects of vacuum dichroism and
birefringence [22,23] and shining light through the wall [24], observable in experiments like
PVLAS [25], OSQAR [7], ALPS [8,26], BMV [9,10], LIPPS [11] avoiding not only the stellar
constraints for the ChernSimons like interaction of the paraphoton, but also the bounds on the
parameters of millicharged particles discussed in [27].
The organization of the paper is as follows. In Section 2, we review the effective field theory
of [16] that introduces a new massive vector boson, coupled to the photon via a ChernSimons
term, and study the propagation of light in a magnetic field. At low energies, this theory behaves like a theory of ALP corresponding to the longitudinal component of the new gauge field.
In Section 3, we modify it at high energies, so that the gauge boson couples to a conserved
current, suppressing the production of its longitudinal component in stars. This modification is
parametrized by effective non-local terms. We then provide an example of a microscopic theory
reproducing these terms, using millicharged fermions. We also analyze the various experimental
constrains and deduce the allowed parameter space that leaves open the possibility of detecting
interesting effects in forthcoming experiments, such as measuring dichroism and birefringence
of light and photon regeneration in the presence of strong magnetic fields. Section 4 contains our
concluding remarks and a discussion on possible derivation of our model from D-brane constructions.
1
3 We denote by F the dual field strength: F
= 2 F .
248
2. Effective theory at low energies and propagation of light in the magnetic field
We start by reminding the model considered in [16]. We consider the effective theory of two
vector fields: the usual photon A and a massive field B (paraphoton)4 :

Slow =
m2
m2
1
1
d x FA2 FB2 + B (D )2 +
(D )2 + 2D D FA
4
4
2
2
4

(1)
where D = d + B, D = d + A and is a dimensionless coupling. The Stuckelberg field

makes the action explicitly gauge invariant under U (1)B gauge transformations, provided that
when B B + . The field B is therefore massive with mass mB . Another
Stuckelberg field , transforming as when A A + , restores the gauge
invariance with respect to the U (1)A gauge group. This means that the photon A is also massive
with mass m . This becomes explicit in the unitary gauge ( = 0, = 0):

Slow,unitary =

m2 2
m2
1
1
d 4 x FA2 FB2 + B B2 +
A + 2A B FA .
4
4
2
2
(2)
Notice, that sending m 0 (and making the field non-dynamical) will make A massless
and add the constraint FA FB = 0 (with the field being a Lagrange multiplier). To make
the analysis simpler, we will assume massless photon and the action (2) supplemented by this
constraint, although our analysis remains unchanged in the case of finite but small enough m .
The ChernSimons (CS) term A D FA remains gauge invariant in this case. Notice, that
in the theory (1) the field B couples to a non-conserved current (JB )low :

low
JB
Slow
= A F ;
B

JB low = F F .
2
(3)
The conserved (Nther) current is a linear combination of (JB )low and . As we will show
below, the property (3) implies that in the theory (1) the longitudinal part of the massive vector
field B behaves as a massive axion with the mass mB and the coupling constant mB /.
2.1. Propagation of light in a magnetic field
Let us consider the equations of motion of theory (1) for the case of the propagation of photons
in strong external magnetic field. We consider light propagation along the z-axis in the magnetic
field H 0 = Fyz = F23 , pointing in x-direction. Under the condition FA FB = 0, the CS term
does not depend on the choice of the background vector potential. For simplicity we choose the
background vector potential, corresponding to the field H 0 , in the form A y = zH 0 . Working
in the unitary gauge (i.e. = 0) for the B-field, we obtain the following system of equations:

B
F = 2B F A F
(4)
,
FB m2B B = A F ,
(5)
B
F F
= 0.
(6)
4 To simplify the notations and avoid proliferation of non-essential -tensors, we will often use differential forms, such
as A D FA instead of 12 A ( + B )F and F F instead of 14 F F .
249
One can construct the solution of these equations as perturbations in small parameter around
(0)
(0)
the zeroth-order approximation: A = a ei(tz) , B = 0.
A simple analysis shows that only the longitudinal part of B (i.e. longitudinal electric
B ) interacts with the propagating photon, with polarization parallel to the external magfield F0z
netic field. To demonstrate the similarities and differences between this model and the axion,
we revert from Eqs. (4)(5) to the system, involving the longitudinal degree of freedom of the
B-field and the photon Ax .5 Namely, one can represent the field B in terms of one scalar degree
of freedom, :

Bz = mzB ,
1
+ 0 B0 , i.e.
B =
(7)
mB
B0 = m0B + B 0 .
Notice that there is a residual uncertainty in the definition of : one can simultaneously shift
+ (t) and B 0 B 0 (t)

mB . Clearly, F0z = z B0 .
To establish the connection between and B 0 we consider the Maxwells equation:
B
iF0z
+ mB z = 2 A y (z)(iAx ),
(8)
which gives
mB
B 0 =
+ 2
i
z
A y (z )Ax (z ) dz .
(9)
Taking the derivative of Eq. (5) we arrive to the first class constraint which massive vector
fields B should obey
B =

4H 0 (iAx )

F F =
.
2m2B
m2B
(10)
Putting relations (7), (9) into the constraint (10) we obtain

4H 0 iAx
2 + m2B =
2imB
mB
z
A y (z )Ax (z ) dz .
(11)
Next, we analyze the e.o.m. for the field Ax :

4H 0 i
4H 0 mB
2Ax =
+
+ 8 2 H 0
mB
i
2mB
+
z A y + 4 2 A 2y Ax .
i
z
A y (z )Ax (z ) dz

(12)
The propagation of light, polarized perpendicularly to the magnetic field (component Ay ), is not
modified in this theory, as follows trivially from the structure of currents in the right-hand side
of Eqs. (4)(5).
5 We choose the gauge in which the plane wave, with polarization vector parallel to the external magnetic field is
described by the potential Ax (t, z).
250
Let us compare Eqs. (11)(12) to the pure axion case:

1
( )F ,

M

1
F F .

2 + m2a =
(13)
4M
For the propagation of linearly polarized light, parallel to the magnetic field H 0 Eqs. (13) reduce
to

2i
2 + m2a =
Ax H 0 ,
M
2iH 0
2Ax =
(14)
.
M
We see that under the identification
F =
2
1
;
M
mB
ma mB
(15)
Eqs. (11)(12) reduce to (14) plus some additional (non-local) terms, depending on the background potential A y .6 These terms, however, are suppressed (as compared to those, depending
1
on the field H 0 ). The easiest way to see it, is to send 0, while keeping the ratio /mB = 2M
fixed. Thus, Eqs. (11)(12) are reduced into those of massless axion. This leads to the dichroism
2
2 H 20 L2 , where L is the distance traveled by the light. For finite mB , corrections suppressed
mB
by powers of
mB
will appear.
3. Possible microscopic origin of the theory

3.1. Decoupling of the longitudinal polarization at high energies
In quantum electrodynamics, if one adds a small photon mass m , all the processes, involving
m
the third (longitudinal) degree of freedom, are suppressed at energies E m as E . This is due
to the fact that the electromagnetic field couples to a conserved current, which is a consequence
of gauge invariance of the theory. On the other hand, if the current is not conserved, at high energies processes involving the longitudinal polarization of the vector boson are equivalent to those
involving the longitudinally coupled scalar due to the so-called Goldstone boson equivalence
theorem [28].
This is what happens in theory (1). Although the theory is gauge invariant under the U (1)B
gauge symmetry, it is realized by simultaneous gauge transformations of the B-field and the
Stuckelberg field . As we saw in Section 2.1, the field B couples to the non-conserved current (3) and therefore its longitudinal polarization behaves as an axion (for mB ).
However, the theory (1) is an effective field theory, valid up to a certain energy scale . In
many cases it naturally happens that for E the theory gets modified in such a way that
the current, to which the B -field couples becomes conserved. Then, all processes involving
emission or absorption of the longitudinal polarization of B are suppressed as ( mB )2 . On the
6 The term 4H 0 mB in Eq. (12) is subleading, as compared to the first term and does not change the results of our
i
analysis in any essential way.
251
other hand, we are interested in the situation, where the field B can be produced at low laboratory energies. This puts restriction on its mass to be mB lab 1 eV. Then at stellar energies
star 1 keV one obtains a suppression of at least (lab /star )2 106 . For smaller values
of mB the suppression is even stronger. The theory of transverse B -field with CS interaction
does not resemble the theory of axion anymore. For example, the production of this field due
to the CS interaction is strongly suppressed by the small value of the dimensionless CS coupling .
To illustrate this idea, assume that in the theory there is an additional particle with mass m ,
interacting with the fields of the theory (1) and giving rise to an effective action of the following
(schematic) form:

m2
1
1
d 4 x FA2 FB2 + B D 2 + A B FA
4
4
2

2

m
1
) B
) .
+
(F
F
(F
F
2 + m2
2 + m2

S=
(16)
In the next subsection we will present an example of a renormalizable field theory, which has
such properties (recall that we always add to this theory the constraint FA FB = 0 to make
it gauge invariant). At low energies (for < m ) one obtains the action (1) (formally taking
m ).
At high energies ( m ) the effective theory is of course non-local. We can neglect the
interacting term proportional to in the action (16) and obtain

S( m )
1

m2
1
1
FA2 FB2 + B D 2 + A B FA B (F F ). (17)
4
4
2
2
At these energies, the field B couples to the conserved current JB ,
JB =

A F (F F ).

2
2
(18)
Therefore, at energies m the production of the longitudinally polarized B -field in theory (16) is suppressed. Of course, for > m the current (18) should be computed directly in
the microscopic theory producing the non-local terms in (16), containing additional particles,
rather than in the non-local effective theory. Next we are going to present an example of such a
microscopic theory.
3.2. Theory with millicharged fermions
Consider a theory with several sets of chiral fermions, whose masses are given by Yukawa
interactions with Higgs fields 1 and 2 :
L=
(i iD
/ i + fi i 1 i ) + (i iD
/ i + i i 2 i ) + h.c.
(19)
i=1,2
The fermions are charged with respect to U (1)A U (1)B , with one of the possible choices of
charges shown in Table 1. Integrating out the fermions for energies below their masses, one
252
Table 1
A simple choice of charges of fermions, which leads to the low-energy effective action (16). The charges are chosen in
such a way that all gauge anomalies cancel. The cancellation of U (1)3A and U (1)3B anomalies happens for any value
of ei , qi . Cancellation of mixed anomalies requires e = 4q e where is related to ei , qi via (21)
3 3
1
U (1)A
U (1)B
L

R
L
R
e1
q1
e1
q1
e2
q1
e2
q1
e3 + e
q2
e3
q2
e3
q2
e3 + e
q2
5 .
Fig. 1. Anomalous contributions to the correlator
Fig. 2. Two graphs, contributing to the ChernSimons terms.
obtains anomalous (CS-like) terms in the effective action7 :

(e12 e22 )q1
FA FA + A B FA .
Scs =
16 2
(20)
Only fermions contribute to the term FA FA (fermions are vector-like with respect to
the U (1)B gauge group and therefore do not interact with ). The contribution to the CS term
A B FA comes from both sets of fermions. The interaction of the B -field with the photon is
very weak due to the small value of , therefore in the action (20) we omitted anomalous terms,
containing more than one power of the B-field.
7 Of course, integrating out these fermions, one obtains also terms leading to a renormalization of charges of the
fields A and B , of the B mass, possible generation of kinetic mixing between A and B , etc. However, in the case
of chiral fermions, one expects additional contributions from anomalous (triangular) diagrams (shown in Figs. 12)
(see e.g. [29]).
253
The relation between the coefficient in front of the CS term and the fermion charges is
dictated by the gauge invariance of the action with respect to U (1)B gauge transformation
q1 (e12 e22 )
.
(21)
16 2
The Higgs field which provides mass to the fermions i is charged with respect to the U (1)B
with charge 2q1 . If it acquires a vacuum expectation value (VEV) |1 | = vB , one has 1 =
5
vB e2iq and the Yukawa term for fields becomes vB i e2iq1 i . The Higgs then provides
mass to the B -field:
=
mB = 2q1 vB
(22)
and Higgss kinetic term becomes that of the field in action (1).
Let us take Yukawa couplings fi i , so that fermions are much heavier that . Consider
energies m < E < m . The resulting expression for the current JB is given by the expression

A F (F F )

2
2
which arises from the effective action

1

Seff,high = B FA FA + A B FA
2
JB =
(23)
(24)
(compare e.g. [30] where similar result for the massless chiral fermions is presented).
3.3. The parameters of the model
Next, we study the restrictions on the parameters of our model. First of all, notice that in
order to expect positive outcome in laboratory experiments, which aim to measure dichroism
and birefringence or photon regeneration (shining light through the wall), we need to have a
mass of the B -field to be within the energy reach of laboratory experiments:
mB lab 1 eV.
(25)
This means that

vB 1 eV/q1 .
(26)
Notice, that the charges of fermions with respect to the paraphoton B are not restricted and
therefore one may have q1 1. This will make vB to be also in the sub-eV region.
When speaking about ALPs, it is convenient to characterize their interaction with the photon
by the dimensionful coupling M (see Eqs. (13)). It follows from Eq. (15) that in terms of the
parameters of our model M mB . Using then relations (21) and (22), we obtain the following
expression for the interaction strength M in terms of the parameters of the microscopic model8 :
M
vB
.
2
e
8 We consider the case when e e e .
1
2
(27)
254
For our analysis of the classical equations of motion to remain valid, the tree-level unitarity bound
for processes involving massive vector bosons should not be saturated at all relevant energies.
The saturation of this bound is reached at energies E mB / which is again M.
There are various restrictions on the charges of new types of fermions with the electromagnetic
field. First, laboratory bounds, coming from the contributions to the Lamb shift [31] and invisible orthopositronium decay [3133] give e < 3 105 , while recent results from TEXONO
experiment put upper bound e < 105 [34]. This translates into

vB
.
M 108 vB = 10 GeV
(28)
eV
This bound is very weak. A stronger bound on the charges of millicharged fermions (e < 106 )
with sub-eV masses comes from the requirement that such fermions do not distort the CMB
spectrum too much [27]. However, this restriction is not applicable in our case as the mass of our
fermions m > 1 eV (see discussion below).
The strongest bounds on charges of fermions with mass below 30 keV comes from limiting
the contribution of these particles to the energy transfer in stars [35] (see also [36]). This gives
e < 1014 . This bound translates into the following lower bound on M:

vB
28
19
.
M > 10 vB = 10 GeV
(29)
eV
For instance, if we take M 109 GeV (the sensitivity expected to be reached by the OSQAR
experiment [7]), this would require vB 1010 . This would imply an extremely light B field
with mB 1010 eV and 1028 (which is still a possibility).
However, the mass mB is not necessarily so small. In our model, the paraphoton field B
would acquire a kinetic mixing with the photon due to the loop corrections coming from light
fermions. Therefore, the mechanism of additional suppression of the coupling of fermions with
the photon in stars, similar to that proposed in [37], is possible. The restriction then becomes
e 1014 ( mstar
)2 , i.e. the stellar bound of [31,35] is weakened by at least six orders of magniB
tude. This would lead to the following bound
5
vB
M 107 GeV
(30)
eV
with mB 1 eV.
We are interested in the situation where at laboratory energies the effective action is given
by (2). For this to be true, the mass m of the fermions should be greater than laboratory
energies lab . At the same time, the mass should be lower than the stellar energies star 1 keV
for (23) to be true. Therefore the fermion mass m lies in the interval 1 eV m 1 keV.
As we saw in Section 3.1, in the limit m /star 0 the massive vector field B is coupled to
a conserved current and therefore its production in stars is suppressed at least as (lab /star )2 .
At higher energies ( star ) additional contributions to the action (24) due to finite m /star
behave as (m /star )2 . These terms contribute to the divergence of the current JB :

m 2

F F .
JB
(31)
star
This gives an additional contribution to the production of the longitudinally polarized B-field,
not suppressed by (mB /)2 . However, in comparison with the low-energy expression (3), there
255
is an additional suppression (m /star )2 . For convenience, in what follows, we will characterize

the interaction strength in terms of an energy dependent coupling constant M():

mB 2
, > m .
M()
(32)
m
To determine the admissible values of m , let us assume that one of the laboratory experiments (light propagation in a magnetic field or regeneration experiments) had positive outcome.
This determines the value of Mlab mB /. As a conservative estimate we take it to be equal
to the laboratory lower bound, reported by PVLAS Collaboration [6]: Mlab 2 106 GeV.
This value is much lower than those coming from the astrophysical bounds: restriction from the
helioseismology M 2 109 GeV [38,39] and restrictions from the horizontal branch (HB)
stars: MHB 1010 GeV (for a review see e.g. [19,36,40]). Demanding that the modification (32)
at star energies M(star ) is greater than these astrophysical bounds (which we will collectively
call Mstar ), one arrives to the following restriction on the mass m :

Mlab
m star
(33)
.
Mstar
As an estimate for star we take the plasma frequency in the stellar interim: star 0.3 keV for
the Sun and star 2 keV for HB stars. As a result, we obtain m 10 eV using solar data or
m 20 eV using restrictions from HB stars.
The recent bound from CAST Collaboration [41,42] MCAST 1010 GeV would give the restriction on the mass of the fermions m few eV. However, it should be noted that the present
CAST bounds do not extend above axion masses 0.02 eV. The model presented here allows
mB 0.02 eV (while still below lab ), therefore the CAST restrictions may not be applicable in
our case.9
Additional restriction on the parameter space may come from the following fact. Integrating
out fermions leads in general to terms suppressed by 1/m (as in any renormalizable field
theory [43]). At laboratory experiment energies with 1 eV, such contributions may still
be significant. In particular, they contribute to the action the following terms (analogous to the
EulerHeisenberg corrections in quantum electrodynamics [44])

4
2 2
2 e
7
)2 + 4 F
S = d 4 x
(34)
(F
F
.
45 m4 16
These terms of course do not contribute to dichroism (as we consider < 2m ), while their
contribution to birefringence is given by
(L)
4
e
m4
H 20 .
(35)
Taking e 108 , m 1 eV (to maximize the value of (35)), and typical values for 1 eV,
L 1 m, and H 0 5 T, one finds a value of birefringence 1020 much below the current
experimental limits.
9 The CAST Phase II will test the range of masses 0.02 eV < m < 1.1 eV [42] and will allow to probe further our
a
model.
256
Thus, the final range of allowed parameters of millicharge fermions ranges from
vB 1 eV;
e 108 ;
q < 1
(36)
to
vB 1010 eV;
e 1014 ;
q < 1
(37)
with the fermion mass lab m < 20 eV.10

We see, that this range of parameters differs from those of Refs. [20,21], which suggested
that the PVLAS experiment can be explained by the existence of millicharged particles with
m 0.1 eV and charges e 106 . Such millicharged particles, however, would introduce
a distortion of CMB energy spectrum [27]. On the other hand, the millicharged particles with
parameters (36)(37) and vector field with parameters (30) can give rise to effects testable in
present and forthcoming experiments (PVLAS, OSQAR, ALPS, LIPPS, BMV, Q&A): shining
light through the wall and dichroism and birefringence of light propagating in strong magnetic
field.11
4. Discussion
In this paper we presented a simple example of a theory in which non-trivial anomaly cancellation between light and heavy sectors gives rise to possible observable effects in optical laboratory
experiments. We showed that the structure of the effective action changes at high energies, suppressing the production of light particles. Therefore the stellar constraints can be significantly
weaker in this case, allowing for relatively strong optical effects, that may be observed in current
or future experiments.
We would like to notice that although we have presented an explicit microscopic theory (19),
which leads to the effective theory (16) we are interested in, the actual fundamental theory could
be rather different. For example, instead of heavy set of fermions in the theory (19), the CS
term may be generated by different mechanisms in the underlying high-energy theory.
Actually, the low-energy effective action (1) or (2) with the CS coupling can arise easily in
D-brane realizations of the Standard Model [29,45]. Indeed, gauge groups come often in unitary
factors containing extra U (1)s. Moreover, the smallness of the mass mB and the CS coupling
of the vector boson B to the photon may be naturally explained in models of large extra dimensions and low string scale [46,47]. For this, B should propagate in the bulk of extra dimensions,
while its mass should arise from localized anomalies due to chiral fermions present for instance
in the intersections of the Standard Model branes with the brane extended in the bulk. Assuming
for simplicity an homogeneous bulk of volume v in string units and identifying the axion with
the string scales M Ms , one has 1/ v and mB M/ v . It follows that mB 1 eV

14
for M 100 TeV, in which case 10 . On the other hand, the microscopic theory with
millicharged fermions is more difficult to accommodate. One could imagine for instance that the
hypercharge (or the electric charge) acquires a tiny mixing with the bulk gauge boson B, creating
millicharges, although it is not clear how to get a consistent setup with natural suppression.
10 The allowed window for m can become higher, if future laboratory experiments strengthen bounds of [6].
11 Note that as Eq. (30) shows, taking v in the sub-eV range allows to have M 105 GeV and thus explaining the
B
effect of dichroism, reported by the PVLAS Collaboration in 2006 [3]. To avoid the stellar constraints this would require
to have m lab 1 eV. However, the contribution of these fermions to dichroism at laboratory energies is subleading
compared to the corresponding contribution of the CS term (2), which differs this model from the one proposed in [20,21].
257
Another example of a theory with non-trivial anomaly cancellation between light and heavy
sectors, is in the presence of extra dimensions, discussed in [17,18]. In this theory a plane wave,
propagating in a strong magnetic field Hx const with polarization parallel to the field, is described by the equation

2 02 H 2
1
z (z)z Ax + 2Ax = EM
Ax + O(0 ).
(z)
M52 2 (z)
(38)
This leads to the massive wave equation

2Ax (x) m2 H Ax (x) = 0
(39)
where the magnetic mass

m2 H em 0 |H |
(40)
depends only on 4-dim quantities; it is not suppressed by the scale of the 5-dim theory M5 . The
wave equation for perpendicular to the magnetic field component remains the same (massless):
2Ay (x) = 0.
This theory is drastically different from the models discussed in this paper (as well as other
models, predicting non-trivial effects in strong magnetic fields). Namely, it does not contain
any new light degrees of freedom. This means, that the regeneration experiments (shining light
through the wall), as well as the measurements of dichroism will produce no results in this case.
However the theory leads to an ellipticity (birefringence) of the linearly polarized light:
0 EM |H |
L.
(41)
2
2
Actually, the static (capacitor) experiment suggested in [18] is also sensitive to the model
described in this paper. As different models, predicting non-trivial results in optical experiments
behave differently in this static experiment, it provides a good way to distinguish among them.
=
m2 H
Acknowledgements
We would like to thank M. Fairbairn, G. Raffelt and M. Shaposhnikov for useful discussions.
This work was supported in part by the European Commission under the RTN contract MRTNCT-2004-503369 and in part by the INTAS contract 03-51-6346. The work of A.B. was (partially)
supported by the EU 6th Framework Marie Curie Research and Training network UniverseNet
(MRTN-CT-2006-035863). O.R. would like to thank Swiss Science Foundation.
References
[1] BFRT Collaboration, G. Ruoso, et al., Limits on light scalar and pseudoscalar particles from a photon regeneration
experiment, Z. Phys. C 56 (1992) 505508.
[2] BFRT Collaboration, R. Cameron, et al., Search for nearly massless, weakly coupled particles by optical techniques,
Phys. Rev. D 47 (1993) 37073725.
[3] PVLAS Collaboration, E. Zavattini, et al., Experimental observation of optical rotation generated in vacuum by a
magnetic field, Phys. Rev. Lett. 96 (2006) 110406, hep-ex/0507107.
[4] PVLAS Collaboration, E. Zavattini, et al., PVLAS: Probing vacuum with polarized light, Nucl. Phys. B (Proc.
Suppl.) 164 (2007) 264269, hep-ex/0512022.
[5] S.-J. Chen, H.-H. Mei, W.-T. Ni, Q&A experiment to search for vacuum dichroism, pseudoscalar-photon interaction
and millicharged fermions, hep-ex/0611050.
258
[6] PVLAS Collaboration, E. Zavattini, et al., New PVLAS results and limits on magnetically induced optical rotation
and ellipticity in vacuum, arXiv: 0706.3419 [hep-ex].
[7] OSQAR Collaboration, P. Pugnat, et al., Optical search for QED vacuum magnetic birefringence, axions and photon
regeneration (OSQAR), CERN-SPSC-2006-035.
[8] ALPS Collaboration, K. Ehret, et al., Production and detection of axion-like particles in a HERA dipole magnet:
Letter-of-intent for the ALPS experiment, hep-ex/0702023.
[9] BMV Collaboration, C. Rizzo, Laboratory and astrophysical tests of vacuum magnetism: The BMV project, 2nd
ILIAS-CAST-CERN Axion Training, http://cast.mppmu.mpg.de, May 2006.
[10] BMV Collaboration, C. Robilliard, et al., No light shining through a wall, arXiv: 0707.1296 [hep-ex].
[11] LIPPS Collaboration, A.V. Afanasev, O.K. Baker, K.W. McFarlane, Production and detection of very light spin-zero
bosons at optical frequencies, hep-ph/0605250.
[12] S. Weinberg, A new light boson?, Phys. Rev. Lett. 40 (1978) 223226.
[13] F. Wilczek, Problem of strong P and T invariance in the presence of instantons, Phys. Rev. Lett. 40 (1978) 279282.
[14] R.D. Peccei, H.R. Quinn, CP conservation in the presence of instantons, Phys. Rev. Lett. 38 (1977) 14401443.
[15] R.D. Peccei, H.R. Quinn, Constraints imposed by CP conservation in the presence of instantons, Phys. Rev. D 16
(1977) 17911797.
[16] I. Antoniadis, A. Boyarsky, O. Ruchayskiy, Axion alternatives, hep-ph/0606306.
[17] A. Boyarsky, O. Ruchayskiy, M. Shaposhnikov, Observational manifestations of anomaly inflow, Phys. Rev. D 72
(2005) 085011, hep-th/0507098.
[18] A. Boyarsky, O. Ruchayskiy, M. Shaposhnikov, Anomalies as a signature of extra dimensions, Phys. Lett. B 626
(2005) 184194, hep-ph/0507195.
[19] G.G. Raffelt, Astrophysical Axion Bounds, hep-ph/0611350, Lecture Notes in Physics, 2006, submitted for publication.
[20] H. Gies, J. Jaeckel, A. Ringwald, Polarized light propagating in a magnetic field as a probe for millicharged fermions, Phys. Rev. Lett. 97 (2006) 140402, hep-ph/0607118.
[21] M. Ahlers, H. Gies, J. Jaeckel, A. Ringwald, On the particle interpretation of the PVLAS data: Neutral versus
charged particles, Phys. Rev. D 75 (2007) 035011, hep-ph/0612098.
[22] L. Maiani, R. Petronzio, E. Zavattini, Effects of nearly massless, spin zero particles on light propagation in a magnetic field, Phys. Lett. B 175 (1986) 359.
[23] G. Raffelt, L. Stodolsky, Mixing of the photon with low mass particles, Phys. Rev. D 37 (1988) 1237.
[24] K. Van Bibber, N.R. Dagdeviren, S.E. Koonin, A. Kerman, H.N. Nelson, An experiment to produce and detect light
pseudoscalars, Phys. Rev. Lett. 59 (1987) 759.
[25] M. Roncadelli, Behind PVLAS, arXiv: 0706.4244 [hep-ph].
[26] A. Ringwald, Photon regeneration plans, hep-ph/0612127.
[27] A. Melchiorri, A. Polosa, A. Strumia, New bounds on millicharged particles from cosmology, Phys. Lett. B 650
(2007) 416420, hep-ph/0703144.
[28] J.M. Cornwall, D.N. Levin, G. Tiktopoulos, Derivation of gauge invariance from high-energy unitarity bounds on
the s matrix, Phys. Rev. D 10 (1974) 1145.
[29] P. Anastasopoulos, M. Bianchi, E. Dudas, E. Kiritsis, Anomalies, anomalous U(1)s and generalized ChernSimons
terms, JHEP 0611 (2006) 057, hep-th/0605225.
[30] A. Smailagic, E. Spallucci, Higher dimensional Schwinger-like anomalous effective action, Phys. Rev. D 62 (2000)
045010, hep-th/0003293.
[31] S. Davidson, S. Hannestad, G. Raffelt, Updated bounds on milli-charged particles, JHEP 0005 (2000) 003, hep-ph/
0001179.
[32] T. Mitsui, et al., Search for invisible decay of orthopositronium, Phys. Rev. Lett. 70 (1993) 22652268.
[33] A. Badertscher, et al., An improved limit on invisible decays of positronium, Phys. Rev. D 75 (2007) 032004,
hep-ex/0609059.
[34] S.N. Grinenko, N.V. Krasnikov, A. Rubbia, Search for millicharged particles in reactor neutrino experiments:
A Probe of the PVLAS anomaly, Phys. Rev. D 75 (2007) 075014, hep-ex/0612203.
[35] S. Davidson, B. Campbell, D.C. Bailey, Limits on particles of small electric charge, Phys. Rev. D 43 (1991) 2314
2321.
[36] G.G. Raffelt, Stars as Laboratories for Fundamental Physics: The Astrophysics of Neutrinos, Axions, and Other
Weakly Interacting Particles, University of Chicago Press, Chicago, USA, 1996.
[37] E. Masso, J. Redondo, Compatibility of CAST search with axion-like interpretation of PVLAS results, Phys. Rev.
Lett. 97 (2006) 151802, hep-ph/0606163.
[38] H. Schlattl, A. Weiss, G. Raffelt, Helioseismological constraint on solar axion emission, Astropart. Phys. 10 (1999)
353359, hep-ph/9807476.
259
[39] J.N. Bahcall, A.M. Serenelli, S. Basu, New solar opacities, abundances, helioseismology, and neutrino fluxes, Astrophys. J. 621 (2005) L85L88, astro-ph/0412440.
[40] G.G. Raffelt, Particle physics from stars, Annu. Rev. Nucl. Part. Sci. 49 (1999) 163216, hep-ph/9903472.
[41] CAST Collaboration, S. Andriamonje, et al., An improved limit on the axionphoton coupling from the CAST
experiment , JCAP 0704 (2007) 010, hep-ex/0702006.
[42] CAST Collaboration, C. Eleftheriadis, Results on axion physics from the CAST experiment at CERN, arXiv:
0706.0637 [hep-ex].
[43] T. Appelquist, J. Carazzone, Infrared singularities and massive fields, Phys. Rev. D 11 (1975) 2856.
[44] W. Heisenberg, H. Euler, Consequences of Diracs theory of positrons, Z. Phys. 98 (1936) 714732,
physics/0605038.
[45] I. Antoniadis, E. Kiritsis, J. Rizos, T.N. Tomaras, D-branes and the Standard Model, Nucl. Phys. B 660 (2003)
81115, hep-th/0210263.
[46] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, The hierarchy problem and new dimensions at a millimeter, Phys.
Lett. B 429 (1998) 263272, hep-ph/9803315.
[47] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, New dimensions at a millimeter to a Fermi and superstrings at a TeV, Phys. Lett. B 436 (1998) 257263, hep-ph/9804398.

DLCQ strings, twist fields and one-loop correlators

on a permutation orbifold
Henry C.D. Cove, Zoltn Kdr, Richard J. Szabo
Department of Mathematics and Maxwell Institute for Mathematical Sciences, Heriot-Watt University,
Colin Maclaurin Building, Riccarton, Edinburgh EH14 4AS, UK
Received 17 July 2007; accepted 3 October 2007
Abstract
We investigate some aspects of the relationship between matrix string theory and light-cone string field
theory by analysing the correspondence between the two-loop thermal partition function of DLCQ strings in
flat space and the integrated two-point correlator of twist fields in a symmetric product orbifold conformal
field theory at one-loop order. This is carried out by deriving combinatorial expressions for generic twist
field correlation functions in permutation orbifolds using the covering surface method, by deriving the oneloop modification of the twist field interaction vertex, and by relating the two-loop finite temperature DLCQ
string theory to the theory of Prym varieties for genus two covers of an elliptic curve. The case of bosonic
Z2 orbifolds is worked out explicitly and precise agreement between both amplitudes is found. We use these
techniques to derive explicit expressions for Z2 orbifold spin twist field correlation functions in the type II
and heterotic string theories.
PACS: 11.25.Hf; 11.25.Sy; 02.10.Yn
Keywords: Matrix string theory; Conformal field theory; Permutation orbifolds; Discrete light-cone quantization;
Finite-temperature string theory
E-mail addresses: h.cove@ma.hw.ac.uk (H.C.D. Cove), z.kadar@ma.hw.ac.uk (Z. Kdr), r.j.szabo@ma.hw.ac.uk

(R.J. Szabo).
doi:10.1016/j.nuclphysb.2007.10.014
H.C.D. Cove et al. / Nuclear Physics B 793 (2008) 260306
261
1. Introduction and summary

Large N matrix field theories obtained as dimensional reductions of maximally supersymmetric U (N ) YangMills theory in ten spacetime dimensions provide non-perturbative descriptions
of M-theory and string theory in various backgrounds, and associated superconformal field theories (see [1] for a review). The best understood example is matrix string theory [24] which takes
the form of maximally supersymmetric U (N ) YangMills theory in two dimensions. In this case
the gauge coupling is inversely proportional to the string coupling, so that the free string limit
corresponds to the infrared limit and the first order interaction term to the least irrelevant operator in the gauge theory. In this strong coupling limit the supersymmetric YangMills theory
approaches a superconformal fixed point which is conjectured to be the supersymmetric sigma
model on the symmetric product orbifold (R8 )N /SN . The spectrum of this orbifold superconformal field theory can be canonically identified with that of the free second quantized type IIA
string [4,5].
This equivalence is demonstrated qualitatively in discrete light-cone quantization (DLCQ) by
matching configurations obtained by gluing different copies of the strings winding around a lightlike circle to twisted sectors of the symmetric product. The quantitative demonstration is given
by matching the torus partition function of the superconformal field theory [5] to the thermodynamic free energy of the free type IIA superstring [6]. It is conjectured [4] that the equivalence
holds generally in the interacting string theory as well. Strings interact by means of splitting and
joining, and the interaction points correspond to insertions of twist field operators in the orbifold
superconformal field theory. It has been recently argued [710] that the structure of the contact
interactions in GreenSchwarz light-cone superstring field theory simplifies within the twist field
formulation of matrix string theory. Unlike the light-cone string field theory, however, the matrix
model provides a full non-perturbative definition of the string dynamics in the large N limit.
In this paper we will investigate this conjectural perturbative correspondence further by examining the relationship between the thermodynamic free energy of type II superstring theory in
DLCQ and correlation functions of the leading irrelevant twist field operators in the symmetric
product orbifold conformal field theory. The Polyakov path integral for the former quantity is
known [11] to truncate the sum over contributing string worldsheets to those which are branched
covers of the spacetime torus arising from the null compactification at finite temperature. The
free energy at the leading non-vanishing order in the string coupling constant is the two-loop
string amplitude which has been calculated in [12]. In order to check the conjecture one needs to
compute the corresponding amplitude in the orbifold conformal field theory, which is given by
the one-loop two-point function of appropriate twist fields. These operators create twisted sectors
out of the vacuum state, in that the local fields of the sigma model acquire non-trivial monodromy
about the twist field insertion points. Computing their correlation functions is thus not straightforward, and a good portion of our analysis will centre around the technicalities involved in these
calculations.
There are several strategies presented in the literature for computing twist field correlation
functions. The stress tensor method was originally introduced in [13] and used to compute Z2
orbifold [1315], and more generally ZN orbifold [16,17], correlation functions on worldsheets
of arbitrary topology, and SN orbifold correlation functions on the sphere [18,19]. In this method
one first determines the twisted Greens function (the n-point function of the stress energy tensor
in the twisted sector) by demanding the correct short distance behaviour and monodromy about
the twist field insertion points. A closely related but more general technique is the covering space
method. It makes direct use of the fact that a monodromy is associated to a covering surface. If a
262
field is multi-valued when transported around a closed curve, then it is well-defined as a singlevalued function on the appropriate cover of the worldsheet without any special points. In this way
the twist field correlation functions can be expressed as vacuum amplitudes of the free conformal
field theory on the covering surfaces. This method was exploited in [2022]. It is also the main
principle behind computing essentially all quantities in permutation orbifolds as shown in [23]
where, in particular, the partition function was given for arbitrary orbifold twist group.
In this paper we use the covering space method for the definition and computation of twist field
correlation functions in symmetric products defined on worldsheets of non-trivial topology. The
vacuum amplitudes of these conformal field theories are known in complete generality, i.e., for
worldsheets of arbitrary genus and arbitrary finite twist group [24]. We generalize these results
to the n-point correlation functions of twist field operators. When the worldsheet has non-trivial
fundamental group and the twist group is non-Abelian, the definition of the corresponding twisted
Greens functions is problematic and the covering space technique is the only possible way to
define the amplitudes. To make these formulae completely explicit, one needs to determine the
dependence of the complex structure of the covering space on that of the worldsheet and the
location of the twist field operator insertions. This is a very difficult problem in the general
case when the covering surface does not admit any conformal automorphisms. We have not
been able to solve this problem in full generality and are not aware of any solution to it for
any specific cases of such a cover. All known computations of twist field correlation functions
are done with respect to covers with automorphisms (this is the case, in particular, for the ZN
orbifolds), or to worldsheets of trivial topology when the covering space can be parametrized
explicitly in terms of the complex coordinate z of the sphere. Nevertheless, using our technique
we are able to determine the bosonic two-point twist field correlation function of the orbifold
R24 Z2 := (R24 R24 )/Z2 and compare it to the appropriate power of the Z2 orbifold twist
field correlation function of the one-dimensional free boson computed in [21], yielding a highly
non-trivial check of our methods. Although throughout we deal only with orbifolds of flat space
Rd , most of our considerations and results apply to more general symmetric products as well.
When writing down generating functions of amplitudes in symmetric products, one has to
sum over all covers of the worldsheet in such a way that only the connected covering surfaces
contribute. This fact lies behind the conjecture that these amplitudes naturally arise in physical
string theories. We generalize the resummation procedure which was done originally for the
torus partition function in [5] and for the Klein bottle amplitude in [25] for the case of closed
strings, and then for the annulus and Mbius diagrams in [26,27] for the case of open strings.
The generalization to the twist field n-point function is possible due to a general combinatorial
formula [25] which is the crux of all of these calculations.
The main technical achievement of the two-loop calculation of [12] was a modification of
the WeierstrassPoincar theory of reduction. Reduction may be described entirely in terms of
the Riemann matrix of periods of a curve, and it has the effect of expressing theta functions
at a given genus in terms of lower-dimensional theta functions. This happens exactly when the
curve in question covers a surface of lower genus (but it may also occur without there being a
covering map). The remarkable feature of this reduction is the simple universal form that the
genus two DLCQ free energy takes in terms of Jacobi elliptic functions on the base torus. For the
contributions from double covers of the torus to the two-loop free energy of the critical bosonic
string, we find perfect agreement between the string free energy and the correlator of twist fields
computed as the appropriate power of the Z2 orbifold twist field two-point function of [21]. We
will find generally that the original genus zero interaction vertex proposed in [4] must be modified
at one-loop order to ensure equivariance under the action of the non-trivial modular group in this
263
instance. Since the structure of the result depends only on the orbifold twist group and not on
the data of the specific string theory, we use this equivalence and the known formulae from [12]
for the two-loop DLCQ free energy of the type II and heterotic strings to derive the two-point
functions of the appropriate spin twist fields in the corresponding Z2 orbifold superconformal
field theories. To the best of our knowledge, these correlation functions have not been previously
computed, and our explicit formulae should be useful for further clarifying the role of the twist
field interaction vertex in light-cone string field theory.
The difficulty in establishing the correspondence is writing down the period matrix of the
covering surface explicitly in terms of the modulus of the worldsheet torus and the branch point
loci. This is achieved in part by elucidating the geometric meaning of the reduced genus two
period matrix. In [12] it was shown that this period depends on two elliptic moduli, one of which
lies in a modular orbit of an unramified (one-loop) cover of the base torus. Here we show that
the second elliptic modulus determines the complex structure of a Prym variety. Prym varieties
arise in special instances of covering surfaces. A theorem due to Mumford asserts that there
are only three types of branched covers which give rise to Prym varieties, namely unramified
double covers, ramified double covers with two branch points, and precisely our instance of
genus two covers over an elliptic curve. When this is in addition a double cover of the torus,
we use the canonical involution of the genus two surface to explicitly construct the dependence
of the periods on the branch points (which are the images of the fixed points of the involution).
This procedure unfortunately does not generalize to higher degree covering surfaces (although
the identification with a Prym variety always holds).
The organisation of the rest of this paper is as follows. In Section 2 we give a general introduction to the theory of bosonic permutation orbifolds, and use the one-loop sigma model to
illustrate the typical combinatorial structure of amplitudes therein. We apply the combinatorial
resummation formula for symmetric products to compute a large class of correlation functions
which are invariant under the action of the twist group. We show how to generalize these formulae to correlation functions of twist field operators, and briefly review the structure of the
DLCQ string partition function. In Section 3 we present detailed and explicit calculations for Z2
orbifolds. In the course of this analysis, we make the generic connection between DLCQ string
theory and the theory of Prym varieties, and also derive the explicit modification of the twist field
interaction vertex for toroidal worldsheets in the symmetric product sigma model. In Section 4
we discuss the technical issues surrounding the generalizations of these results to SN orbifolds
with N > 2. We examine the uniformization construction, which is used to build vacuum amplitudes, in the context of a generic twist field n-point function, and the problem of determining
the period of the genus two covering surface in terms of the branch point data. We also study
the combinatorial expansion in more detail and indicate that, while computable in principle, the
combinatorics become very non-trivial for N > 2. Finally, in Section 5 we describe the modifications of permutation orbifolds required in the presence of fermionic degrees of freedom, and
of twist field correlation functions therein. We then apply these and previous considerations to
derive explicit formulae for the one-loop spin twist field correlation functions in the Z2 orbifold
supersymmetric and heterotic string theories.
2. Correlation functions on permutation orbifolds
In this section we will discuss some general aspects of permutation orbifolds of conformal
field theories, and in particular the case of two-dimensional sigma models on symmetric product
orbifolds of flat space. We first describe the general structure of the partition functions of these
264
models, and then explain the construction of various classes of correlation functions including
those of twist field operators. For the moment we treat only bosonic sigma models explicitly in
order to highlight the essential details, deferring a more detailed analysis of the supersymmetric
and heterotic cases to Section 5. We also explain how these orbifold theories can be interpreted
as string field theories.
2.1. Permutation orbifolds
When a two-dimensional conformal field theory has a discrete symmetry, one can consider
the orbifold theory arising from quotienting with respect
2 to the 2symmetry. The simplest example
1
is the free boson on the circle S1 . Its action 4
d z X is invariant under the reflection
X X. The quotient of the target space is the well-known geometric orbifold S1 /Z2 , and the
coordinate field X can have non-trivial monodromy when encircling a non-contractible cycle
of
the worldsheet . If the radius of the circle is equal to the fundamental string length s = ,
then the resulting orbifold conformal field theory is the square of the critical Ising model [28].
Permutation orbifolds represent a large class of orbifolds where the parent conformal field
theory (whose quotient is taken) has physical Hilbert space H with a discrete symmetry. This
concept was first introduced in [29], and used for the construction of a Z2 orbifold of the E8 E8
heterotic string in [30]. One of their main applications is to the second quantization of string
theory [5], and they have recently been argued [31] to describe new physical string theories at
multiples of the critical dimension. A permutation orbifold of an arbitrary conformal field theory C, by any finite symmetry group G regarded as a subgroup of a symmetric group of some
degree, is a consistent conformal field theory. All of its important quantities (central charge, conformal weights, genus one characters, modular S and T matrices, genus one partition function,
etc.) were worked out originally for cyclic groups in [32], and then generalized to arbitrary finite
groups in [23]. These formulae express a given quantity as a combinatorial expansion, depending
on the twist group G, of the same quantity in the parent theory.
Highest weight states in a permutation orbifold C G := (C)N /G correspond to orbits of
a subgroup G < SN of the symmetric group of degree N acting on the N -fold tensor product of states in the parent theory C.1 In the case that C admits a sigma model description with
embedding coordinate field X M, there is a corresponding sigma model description of C G
on the geometric orbifold M N /G [26]. One introduces N identical coordinate fields X a = X,
a = 1, . . . , N on the worldsheet and allows for G-twisting of them along non-trivial cycles.
For example, on the torus = T with modulus , the boundary conditions of the N coordinate
fields are labelled by two commuting permutations P , Q G < SN such that
X a (z + 1) = X P (a) (z)
and
X a (z + ) = X Q(a) (z),
(2.1)
where in general g(a) denotes the image of the label a under the permutation g G. For a nontrivial pair (P , Q), these boundary conditions are called twisted sectors of the theory. Two pairs
(P , Q) and (gP g 1 , gQg 1 ) with g SN correspond to the same twisted sector, since we can
get from one to the other by relabelling the coordinate fields a g(a).
In general, a twisted sector is given by an equivalence class of homomorphisms from the fundamental group 1 () of the worldsheet to the twist group G. (Since 1 (T) = ZZ, on the torus
1 There is an additional label corresponding to the irreducible character of the double of the stabilizer of the orbit. See
[23] for the precise definition.
265
one specifies a homomorphism by choosing the image in G of the two commuting generators.)
Two homomorphisms , define the same twisted sector, and are said to be equivalent, if they
are related by conjugation as () = g()g 1 for some g SN . The geometric interpretation is provided by the fact that every equivalence class [] of homomorphisms : 1 () SN
determines an unramified cover of degree N over the Riemann surface . The coordinate label a corresponds to the label of a sheet and is called the monodromy homomorphism of the
covering. Conjugation of homomorphisms corresponds to relabelling of the sheets. In the case of
the torus = T with the boundary conditions (2.1), and with the subgroup generated by the pair
of permutations P , Q acting transitively on the set of coordinate labels a = 1, . . . , N , one can
define a single new field X (z) which generates all of the fields X a (z) through the identifications
X (z + m + n ) = X P
m Qn (a)
(z)
(2.2)
with n, m Z and a fixed choice of a. This field is single-valued on a torus which is a cover the
original torus T, whose modular parameter can be determined from the doubly periodic function
X on T.
The modular invariant partition function of the permutation orbifold is determined entirely by
the above data. It is given by [23]2

1
Z G ( ) =
Z ,
(2.3)
|G|
:1 ()G
O ()
where the product runs over the orbits of the image (1 ()) in G and Z( ) is the modular
invariant partition function of the parent conformal field theory on the connected component,
corresponding to , of the cover of given by the homomorphism . (The covering space is
connected if and only if (1 ()) acts transitively in SN .) The summation over defines the
projection onto G-invariant states and ensures modular invariance of the partition function. We
will now explain how to determine (2.3) in practice.
The complex structure of the worldsheet = is encoded by a monomorphism
u : 1 () I , where I is the isometry group of the universal cover U of . For genus g > 1
the latter space is a two-dimensional hyperbolic space, say the upper half plane U = U, and
I = PSL(2, R). The surface equipped with a complex structure can be presented as the
quotient = U/u(1 ()) and its complex structure inherited from U is encoded by the uniformizing group u(1 ()). Given a monodromy homomorphism , the fundamental group
of the corresponding cover is isomorphic to the stabilizer subgroup Ha = 1 () := {
1 () | ( )(a) = a} with fixed a (represented by closed loops based at sheet a). Its index
is equal to the length of the orbit [1 () : Ha ] = | |, which is the number of sheets of the corresponding connected component of . Thus the monodromy homomorphism determines the
topology of the covering space . We can now define the uniformizing group (and hence the
complex structure ) of the cover to be given by u(Ha ) (i.e., = U/u(Ha )), which is a
< 1 (). Note that the
subgroup of u(1 ()) in accordance with the expected property 1 ()
representative of the orbit a can be arbitrarily chosen. This is because Ha = Ha 1 with
1 () for any a, a and conjugate subgroups of 1 () give rise to isometric quotients,
hence determining equivalent surfaces. The homomorphism u is not unique, as it can be com2 It is also expressible as a sesquilinear expansion in the Virasoro characters Tr (q L0 c/24 ), whose form is known in
H
permutation orbifolds [23,33].
266
posed with a modular transformation, but the partition function is modular invariant which makes
the formula (2.3) well defined.
The expression (2.3) is an example of the typical structure of a quantity defined on a Riemann
surface in a permutation orbifold. It is given by a combinatorial expansion (depending only
on G) over the same quantity in the parent theory C defined on all of those surfaces which cover
whose monodromy group is a subgroup of G. Its direct applicability is limited somewhat by
the RiemannHurwitz formula for the genus g of the unramified cover given by
g = N(g 1) + 1.
(2.4)
This implies that, unless g = 1, we would need to know the partition functions of the parent
theory on surfaces of genera higher than g in order to write down the genus g partition function
of the orbifold.
The case g = 1 is, however, much simpler. The universal cover of the torus T is U = C and
I = {Tc | c C} is the group of translations Tc : z z + c of the complex plane. We saw above
that specifying a homomorphism amounts to assigning commuting elements P , Q G for
the generators (, ) of 1 (T) = Z Z. The stabilizer subgroup Ha of any representative of
an orbit a can be characterized by three positive integers s, m, r such that r is the smallest
positive integer satisfying P r (a) = a, 0 s < r and Ha is generated by r , s m . Then the index
of this subgroup is given by | | = rm. The image of Ha under the isomorphism u : (, )
(T1 , T ) determines a subgroup Tr , Ts+m and the corresponding quotient of C is the torus
with Teichmller parameter given by
s + m
.
=
(2.5)
r
The fact that the finite index subgroups of the group Z Z are all isomorphic to the group
itself implies that all unramified covers of the torus are tori. In sigma model language, the path
integral over the multi-valued fields X a on the torus T is constructed by calculating the path
integral over the single-valued field X on the covering torus and summing over every possible X
constructed by different choices of the commuting pair P , Q G. For example, the genus one
partition function of the S3 orbifold is given by [34]

1
1
+1
+Z
Z S3 ( ) = Z( )3 + Z( ) Z(2 ) + Z
6
2
2
2

1
+1
+2
+
(2.6)
Z(3 ) + Z
+Z
+Z
.
3
3
3
3
Note that the individual terms in (2.6) are not modular invariant, but their sum is.
2.2. Symmetric products
Permutation orbifolds whose twist group G is the full symmetric group SN are called symmetric products SymN (C) := (C)N /SN . In this case the formula (2.3) takes into account all
N -sheeted coverings. Starting from a fixed parent theory C and a given worldsheet genus g, the
generating function of partition functions for all N can be written in a closed form thanks to a
combinatorial identity due to Bntay [25]. This identity translates the sum over homomorphisms
in (2.3) to a sum over finite index subgroups of the group = 1 () and is given by

1
Z(H )
1+
(2.7)
Z( ) = exp
,
N!
[ : H ]
N=1
: SN
O ()
H <
267
where is the stabilizer of the orbit and [ : H ] denotes the index of the subgroup H in .
The formula (2.7) holds generally for any finitely generated group and any conjugation invariant function Z (i.e., Z( H 1 ) = Z(H ) for all ) from the set of finite index subgroups
of to a commutative ring R.

The proof of (2.7) is instructive. A given term Z( ) in the sum on the left-hand side of
(2.7) depends only on the equivalence class of the homomorphism . An equivalence class can
be written as
[] =
(2.8)
nk k ,
k=1
0 is its intewhere k is a transitive

equivalence class whose orbits all have length
k and nk
ger multiplicity with k nk = N . One can then rewrite the product Z( ) = k Z(k )nk ,
where k is the stabilizer subgroup of an arbitrary representative of the image of k in SN .
The cardinality of the equivalence class [] can be determined as follows. The total number of
possible elements to conjugate with is |SN | = N!, but not all of these give inequivalent homomorphisms . The permutations which exchange the orbits that have the same SN -action do not
change [], so we have to divide by their number which is nk !. Finally, we have to divide out
the number of cosets k with k 1 = k , which is the index k = [N (k ) : k ] of the
stabilizer k in its normalizer subgroup N (k ). Thus |[]| = N !/ k nk !knk .
One can now rewrite the left-hand side of (2.7) as
N

1
N!
nk
1+
Z(k )
N
nk
N!
k=1 nk !k
N=1
{nk }
n1 ++nN =N

Z(k )nk
=
nk !knk
k=1 nk =0

Z(k )
exp
.
=
k
k=1
(2.9)
k=1
Note that here a summation over conjugacy classes of index k subgroups is implicitly assumed.
The final step consists in rewriting the product of exponentials as the exponential of a sum
over k, and then translating the latter summation into a sum over index k subgroups. There are
[ : N (k )] distinct subgroups in the conjugacy class of k (as k 1 = k if
/ N (k )),
so we need to divide by this number if we wish to sum over all index k subgroups. Then the
resulting factor in the denominator

k : N (k ) = N (k ) : k : N (k ) = [ : k ] = k
(2.10)
is precisely the index of k in and we have arrived at (2.7).
Let us now apply the identity (2.7) to the uniformizing group = u(1 ()) of a compact
Riemann surface = with the definition

Z(H ) := Z H [ :H ]
(2.11)
where Z( H ) is the modular invariant partition function of C defined on the surface H = U/H ,
with U the universal cover of , and is a formal variable which is determined by physical
268
constants in applications. The result is the grand canonical partition function

Sym
N SN
N
Z ( ) = exp
HN Z( ) ,
Z (, ) := 1 +
N=1
(2.12)
N=1
where Z SN ( ) is the partition function for the SN orbifold given by the formula (2.3) and the
operator HN is defined on modular invariant functions by
HN Z( ) =
1
N

Z H .
(2.13)
[ :H ]=N

Note that
the product over the orbits in (2.7) gives a sum for the power of equal to [ :
] = | | = N . This generating function is a sum over all possible (finite-sheeted) covers of
the surface that the parent conformal field theory C is defined on, and its logarithm gives the
restricted sum over connected covers. The operator defined by (2.13) yields a sum over subgroups
H < 1 () of index N , and in the case of the torus = T it coincides with the Hecke operator
acting on the partition function of the parent theory by

1
s + m
HN Z( ) =
(2.14)
Z
.
N
r
rm=N sZ/rZ
2.3. Sigma models at one-loop

Our primary example of a permutation orbifold in this paper will be that of sigma models
on symmetric products of flat space Rd at one-loop order in string perturbation theory. Let us
describe this example explicitly in the case of a single boson X in R. The path integral of the
sigma model conformal field theory on a symmetric product is gotten by considering the grand
canonical partition function

N

1

Sym
N
1
N
a
Z (, ) = 1 +
(2.15)
,
DX DX exp
I X
N!
N=1
where
1
I (X) =
4

d2 z
T
P ,QSN
P Q=QP
(P ,Q)
X(z)X(z)
2i2
a=1
(2.16)
is the bosonic Polyakov action and z = 1 2 , 1 , 2 [0, 1] are complex coordinates on the
torus with respect to the complex structure = 1 + i2 , 1 R, 2 > 0. The sum over commuting
pairs of permutations, specifying monodronomy homomorphisms : 1 (T) SN , is taken over
worldsheet instantons of the field theory labelled by the boundary conditions (2.1). Note that any
metric on the torus can be written as
ds 2 = e2(z) |dz|2 ,
(2.17)
where the scalar field (z) on T is an arbitrary conformal factor.

From the general formulas (2.12) and (2.14) above it follows that the partition function (2.15)
is given by the combinatorial formula

Z Sym (, ) = exp
N=1
where
z( ) =
rm=N sZ/rZ

1 s + m
z
,
N
r
DXeI (X)
269
(2.18)
(2.19)
is the sigma model partition function on the torus with target space R. This gives a sum of the
partition function on a particular torus T over the discrete set of covering tori. The Gaussian
integral (2.19) can be evaluated in terms of a Quillen norm as

vol(T) det 1/2
,
z( ) =
(2.20)
4 2
where is the scalar Laplacian operator on T with respect to the torus metric (2.17), vol(T)
is the volume of the surface T in (2.17), and det denotes the determinant of with zero
modes excluded. At genus one, this determinant has a natural holomorphic splitting and z( )
1/2 det()1/2 over the moduli space of
is a section of the determinant line bundle det()
complex structures on T. The determinant of the Dolbeault operator is the automorphic form
on Teichmller space given by
det = eSL ()/24 ( )2 ,

i/12
(2.21)
2in ) is the Dedekind funcwhere SL () is the Liouville action and ( ) = e

nN (1 e
tion. The partition function (2.19) is thus given explicitly by

1/2

1
1
2 (z)
z( ) = eSL ()/24
(2.22)
d
z
e
.
2

4
|( )|2
T
By replacing z( ) with z( )d
in (2.18) we get the corresponding result for the parent conformal

field theory of a free boson on the target space Rd . Moreover, the combinatorial formula (2.18) is
completely generic and holds for any sigma model partition function on the torus. For example,
we may simply replace z( ) by the appropriate superstring or heterotic string partition functions
at one-loop (with some modifications that we discuss in Section 5).
The formula (2.12) can also be used to compute any correlation function of fields which
are unaffected by the orbifolding. These are the operators which are symmetric under permutations ofthe indices of the scalar field X. Given any function f , we use the notation
Tr f (X) := a f (X a ) for such an operator referring to a diagonal matrix of the N independent fields X a . The (normalized) correlation function is defined by

Sym
Tr f (X)
(, )

N
1
1
N
Tr I (X)
DX DX Tr f (X)e
. (2.23)
:= Sym
1+
N!
Z (, )
N=1
P ,QSN (P ,Q)
P Q=QP
Rather than trying to determine the combinatorics of this amplitude directly, we will calculate
instead the generating function
Sym

Sym

n Sym

n
Tr f (X)
(, ) := e Tr f (X)
(, ) =
(, ) .
n!
n=0
(2.24)
270
Then we can get the correlation function (2.23) by differentiation as

Sym
Z (, )

Sym
.
Tr f (X)
(, ) =

=0
(2.25)
The generating function (2.24) is just the symmetric product partition function of the sigma
model conformal field theory with a shifted action
I (X) = I (X) f (X)
(2.26)
Sym
Z =0 (, ) = 1.
It can thus be calculated by using the combinatorial forand the normalization

mulae (2.12) and (2.14) as above, with the result

N
1
s + m
Sym
exp
z
,
Z (, ) = Sym
(2.27)
N
r
Z (, )
N=1
where
z ( ) =
rm=N sZ/rZ
DXeI (X)
(2.28)
is the sigma model partition function on the torus with respect to the modified action (2.26). To
carry out the differentiation in (2.25), we first calculate

z ( )
(2.29)
= f (X) ( )

=0
where the (unnormalized) expectation values are calculated as Gaussian moments with respect to
the original action (2.16). Combining these results along with the elementary identity dd eF ( ) =
F ( )eF ( ) gives finally

s + m

Sym
N
f (X)
(, ) =
.
Tr f (X)
(2.30)
N
r
N=1
rm=N sZ/rZ
The correlation function of the symmetric operator Tr f (X) in the symmetric product is thus
likewise expressed in terms of the correlation function of the operator f (X) on all unramified
covering spaces over the base torus T. These formulae have natural extensions to higher loops,
but in those instances they require knowledge of the correlation functions of f (X) on all higher
genus Riemann surfaces.
2.4. Twist fields
A twist field P (w) in a generic permutation orbifold C G is a primary field that creates the
vacuum state of a twisted sector at a point w . In a sigma model conformal field theory, its
insertion results in non-trivial local monodromy

X a (z w)e2i P (w) = X P (a) (z)P (w),
(2.31)
where the permutation P is an element of the twist group G < SN . Its effect is to thus make
the local field X multi-valued about the insertion point w . If P = (n) consists of a single
cycle of length n > 1, then the corresponding twist field (n) (w) permutes n copies of C in a
Zn -twisted sector and is a primary field with conformal weight [18]

d
1
(n) =
n
24
n
271
(2.32)
for a d-dimensional boson. The corresponding fields X ai (z), i = 1, . . . , n can then be glued
together into one field X (z) which is identified with a long string of length n.
In the general case, we have seen that twisted sectors are in one-to-one correspondence with
be decomposed into
conjugacy classes of G. The conjugacy class [P
] of an element P SN can
combinations of cyclic permutations as [P ] = n (n)Nn with Nn 0 and n nNn = N . For a
bosonic sigma model in d dimensions, the corresponding twist field has conformal dimension

N
N

Nn
d
Nn (n) =
N
.
P =
(2.33)
24
n
n=1
n=1
An SN -invariant twist field creating the twisted sector [P ] of the permutation orbifold is defined
by averaging over all twist fields in the conjugacy class of P to get
1
gP g 1 (w).
[P ] (w) =
(2.34)
N!
gSN
In this paper we will be primarily interested in correlation functions [P1 ] (w1 ) [Pk ] (wk ) G
of twist field operators in the permutation orbifold C G. These averages are difficult to calculate directly within a path integral formalism, because the twist fields are non-local operators.
However, since these correlation functions are the vacuum functionals with twisted boundary
conditions due to (2.31), it is natural to extend the covering surface principle as in [2022] and
compute them via a generalization of the permutation orbifold partition function (2.3) on a Riemann surface of genus g > 0. Whenever we have twist fields inserted at k distinct points
w := {w1 , . . . , wk } of the worldsheet, a twisted sector is given by a conjugacy class of homomorphisms : 1 (w ) G < SN where w := \ w is the marked Riemann surface with the
k twist field insertion points deleted. It is restricted by admissibility criteria which require that
the images of the generators i of 1 (w ) which are contractible to wi must be simple cycles
of length i > 1 if a Zi twist field (i ) (wi ) is inserted at wi . Each such homomorphism
determines a cover of the worldsheet on which a single new field X (z), defined by a formula
analogous to (2.2), is single-valued. Namely, after going around a curve which is closed on the
marked worldsheet w , one sews the fields X ( )(a) (z) into X (z). Thus, the contribution to the
correlation function from the worldsheet instanton sector determined by the homomorphism
is the free partition function on the cover of determined by .
While the sum arising in the orbifold partition function (2.3) is only over unramified covers of , the twist field correlation functions involve sums over branched covers w where
w := f 1 (w) is the set of pre-images of the set w under the covering map f : . The
RiemannHurwitz formula for the genus g of the covering space with the given monodromy
homomorphism is the general one for covers with ramification given by
g = N(g 1) + 1 +
B
2
with B =
k

(i 1),
(2.35)
i=1
where i is the ramification index given by the length of the cycle of the ith primary twist field.
As before, we have to take into account those homomorphisms whose image does not act
transitively on the coordinate labels a = 1, . . . , N . In this case the simple cycle condition for fixed
length i has to hold for each orbit . This ensures that the genus of the connected component
272
of the cover determined by the action of (1 (w )) on each orbit is equal to g.

We may now
write down a formula analogous to (2.3) for the normalized k-point correlation function of twist
field operators given by
G
k

,w
1
1
,
[Pi ] (wi ) =
Z
(2.36)
|G|
Z G ( )
i=1
:1 (w )G
O ()
where ,w is the complex structure of the covering surface determined by the worldsheet modulus , the stabilizer 1 (w ) , and the branch point loci w.
There are three crucial differences between the formulae (2.36) and (2.3). Firstly, the twist
field correlation functions are not expressed in terms of correlation functions but instead in terms
of partition functions. Secondly, there is a restriction on the admissible homomorphisms to
ensure that they have the prescribed monodromy around the punctures, i.e., (i ) has to be a
simple cycle of length i in each orbit. Thirdly, while the uniformization theorem provided us
with a computational recipe for obtaining the Teichmller coordinate in terms of via knowledge of , it does not apply to the twist field k-point functions. The reason is that parametrizes
the uniformizing group of the compact Riemann surface , which is isomorphic to 1 (), while
the domain of the monodromy homomorphism is 1 (w ) which differs from the domain of
the isomorphism from the abstract group 1 () to the uniformizing group u(1 ()). Therefore,
the complex structure of the ramified cover w is a function of that of the base space , the
locations w of the branch points, and the monodromy homomorphism .
We are also interested in twist field correlation functions on symmetric products. In order
to apply a version of (2.7) we need to pass the constraint, which is imposed on the admissible
homomorphisms in (2.36), to the definition of the function Z(H ). Let us specialize the discussion to the torus = T for definiteness. In this case, the genus of the covering surface is
g whenever its branching number is B = 2(g 1). A standard presentation of the fundamental
group of the marked torus is given by
:= 1 (Tw ) = , , 1 , . . . , k |[, ]1 k = 1 .
(2.37)
To each N -sheeted cover of T there corresponds a conjugacy class of subgroups of of index

N [35], which is the stabilizer of the monodromy homomorphism acting in SN . Note that
the group (2.37) is isomorphic to the free group on k + 1 generators , , 1 , . . . , k1 , and any
subgroup of a free group is also free. This is consistent with the fact [35] that the stabilizer subgroup is isomorphic to 1 ( w ) < 1 (Tw ). To decide when a given finite index subgroup H <
corresponds to a stabilizer subgroup of an admissible homomorphism in (2.36), we proceed
as follows. Let : w be the natural inclusion of surfaces. The induced homomorphism
is then the natural forgetful map. Since H
: 1 ( w ) 1 ()
= 1 ( w ), a formal criterion
for the admissibility of a finite index subgroup H < 1 (Tw ) is given by
H / ker( )
= 1 ().
(2.38)
We can use (2.38) to check whether a given subgroup H is admissible. If the quotient is defined
then H is admissible. This property does not depend
and it yields a group isomorphic to 1 (),
on the conjugacy class of H in 1 (Tw ). We can thus give an implicit definition for the function
appearing in (2.7) as
H,w )
Z(
[ :H ] if H satisfies (2.38),
S
Z(H ) :=
(2.39)
Z N ( )
0 otherwise.
273
We may then apply the formula (2.7) to get the generating function of twist field correlation
functions.
In the following we will apply this formalism to study the perturbation of the sigma model
conformal field theory, on the symmetric product of Rd , by an irrelevant operator of conformal
dimension 3/2. For this, we introduce the bosonic DijkgraafVerlindeVerlinde (DVV) interaction vertex [4,36] which is defined with respect to the Z2 twist field ab (w) corresponding to
the transposition in SN that interchanges the fields X a and X b while leaving all others invariant. These twist fields generate the elementary joining and splitting of strings in the symmetric
product, and they can be built out of standard Z2 orbifold twist operators [13,20]. Then the translationally invariant vertex operator is defined by

N
Vbos =
(2.40)
d(z)
ab (z),
vol(T)
T
1a<bN
where is a coupling constant proportional to the string coupling gs . In contrast to the originally
proposed genus zero case [4,18,36], we will find that the DVV vertex operator at genus one
needs to be defined using a non-constant measure d(z) = d2 z/(z) on the torus T. It will
be determined explicitly in the ensuing sections (as will the coupling constant ) by modular
invariance requirements. When d = 24, the twist field ab (w) is a primary field of conformal
weight 3/2. Starting from the one-loop action (2.16), the interacting symmetric product sigma
model is defined by the action
S
IintN (X) = Tr I (X) + Vbos

(2.41)

with Tr I (X) = a I (X a ).
In this paper we will compute the leading order effect of this perturbation. Using translational
invariance of the sigma model path integral to move one of the branch points to the origin z = 0,
we are thus interested in computing the translationally invariant correlator

S

2 N 2
SN
=
d(z) a1 b1 (z)a2 b2 (0) N .
V V
(2.42)
bos bos
vol(T)(0)
ai <bi T
The computation of the two-point functions in (2.42) specializes the above discussion to the
case g = 1, g = 2, and k = 2. There are two simple branch points with ramification indices
1 = 2 = 2 and = 1 (T \ {z, 0}). Then the logarithm of the generating function (2.7) with
the definition (2.39) is given by a sum over the modular invariant vacuum amplitudes on all
connected N -sheeted genus two covers with two fixed simple branch points. In this case the
first quantized modular invariant partition function for the parent theory is the two-loop version
of (2.20) on (with vanishing Liouville field = 0 for simplicity) given by [37,38]
z(2) ( ) =
(det(Im ))3d/2
,
(4 2 )d/2 |10 ( )|2
(2.43)
where d is the spacetime dimension (d = 26 for the critical bosonic string). Here 10 ( ) is the
genus two parabolic modular form of weight ten with no zeroes or singularities (the Igusa cusp
form), defined on the Siegel half-space U2 = { | Im(11 ) > 0, Im(22 ) > 0, det(Im ) > 0} of
2 2 Riemann period matrices with the boundary component U U consisting ofdiagonal

matrices removed. It can be expressed in terms of the ten genus two theta-constants ab ( ) :=
274
a
b
(0, 0| ) with even binary characteristics a = (a1 , a2 ), b = (b1 , b2 ) Z2 /2Z2 as

ab ( )2 .
10 ( ) = 212
(2.44)
ab0 mod 2
2.5. Thermodynamics of DLCQ strings

In the genus one case = T, the logarithm of the right-hand side of (2.12) coincides with the
free energy of second quantized string theory on the target space M S1 R when the parent
theory is the corresponding conformal field theory on the spacetime M in the free string limit
gs 0 [5,6]. The matching is provided by identifying the modulus of the worldsheet and that of
the spacetime torus, where the second compact direction is timelike and is generated by the trace
taken in computing the free energy amplitude. Its radius is identified with the inverse temperature
. In discrete light cone quantization (DLCQ), the light cone Hamiltonian and momentum are
given by
H = P+
and P = N/R,
(2.45)
where R is the radius of the compactified light-like direction x + S1 and N N0 . The thermo(1)
dynamic free energy FDLCQ is then defined by
(1)
eFDLCQ = Tr e
(P + +P )
2
eN/
2R
TrHN eP
+/
(2.46)
N=0
where HN denotes the sector of the physical Hilbert space with definite total light cone momentum P = N/R. The trace over this subspace can be computed by using the mass-shell relation
P + = H /P , where H is the Hamiltonian for the transverse degrees of freedom along M.
In this way one arrives at the expression (2.12) with the definition (2.14) and := e/ 2R .
The Teichmller parameter of the base torus T on which the string bits live is
4i
.
:=
2R
(2.47)
The qualitative reason for the equivalence is that the second quantized vacuum amplitude is
given by the integral of the conformal field theory partition function over the moduli space of
complex structures, but the only contributing surfaces at one-loop order are those which arise by
winding the string around the compact directions. In other words, only the discretized moduli
space of unramified covers of the torus T is summed over and taking the logarithm eliminates
the disconnected covers.
When M = R24 one finds that the DLCQ partition function for bosonic string theory coincides exactly with the partition function of the symmetric product in the limit N , with the
length ni of a long string identified with the light cone momentum Pi = ni /R for i = 1, . . . , 24.
Checking the equivalence of perturbative bosonic string dynamics and the corresponding interacting symmetric product of R24 beyond the free string limit gs 0 requires computing the
thermal free energy in DLCQ at higher genus and the appropriate amplitudes in the permutation orbifold perturbed by the DVV interaction vertex (2.40). The former amplitudes truncate to
sums over branched covers of the spacetime torus T arising in the null compactification at finite
temperature [11], while the local structure of the operator Vbos matches nicely with the cubic
275
string interaction vertices in light cone Green-Schwarz string field theory [710]. In this setting
the string interactions are generated by sewing together torus worldsheets along branch cuts.
On the DLCQ side, the next-to-leading order contribution is the two-loop free energy which
was computed in [12] with the result

12 N

r 10
(2)
2

FDLCQ , = gs
m
32 2
N2
rm=N
N=2

2
d2 #
(2.48)
10 r,m,s,t , # .
#
12
(2 )
s,tZ/rZ
t=0
This thermal string amplitude is just the weighted integral over a fundamental modular domain
of the genus two bosonic string partition function (2.43) with respect to the modular invariant
integration measure on the space of 2 2 Riemann period matrices with diagonal matrices excluded, but with integration domain restricted to the partially discretized moduli space of genus
two simple branched covers of the torus T with modulus . The integers appearing in (2.48)
can be assembled into the 2 4 matrix

0 0 m 0
(2.49)
M=
r 0 s t
Z)
which determines a homology basis for the cover in which the push-forward f : H1 (,
H1 (T, Z), induced by the holomorphic covering map f : T, is given on a basis of canonical
homology cycles i , i , i = 1, 2 for by
f ( 1 , 2 , 1 , 2 ) = (, )M
(2.50)
with respect to a canonical homology basis (, ) of the base torus. It specifies the way in which
the cycles of the cover wind around the cycles of T. The period matrix U2 \ (U U) of
the cover in this basis is given by the normal form
s+m/

#
rt
r
r,m,s,t , =
(2.51)
rt
#
with # U, and the integration in (2.48) is taken over the standard fundamental domain U
for the action of the genus one modular group SL(2, Z) on # .
The diagonal elements of the period matrix (2.51) naturally capture the modulus of the degree
N = rm unramified cover of the base torus T of modulus ( )1 , along with a second torus
of modulus # . The key feature of the homology basis in which we have expressed the genus
two amplitude (2.48) is that the genus two theta functions appearing in (2.44) admit reduction
to genus one theta functions on these two tori, due to the rational-valued off-diagonal entries of
(2.51). Hence the -dependence of the two-loop free energy is expressible in terms of elliptic
functions, analogously to the one-loop case. Recall that the elliptic Jacobi theta function with
characteristics a, b Z/2Z is defined by

a 2
a
b
ab (z| ) =
(2.52)
exp i n +
+ 2i n +
z+
2
2
2
nZ
along with the Erdlyi notation

1 (z| ) = 11 (z| ) and 2 (z| ) = 10 (z| ),
276
3 (z| ) =
0
0
and 4 (z| ) =
(z| )
0
1
(z| ).
(2.53)
Then one has the decompositions [12]
a
b
r,m,s,t ,

N1

eia2 b2 /2
a2 mt ms + m2 /
b2 n a1
=
n+
(1) b1
2 N
N
N i # n=0

1
n + a2 /2
j
(2.54)
N 2 # ,
N
where j = 2 (respectively j = 3) when the integer a1 mt + b2 N is odd (respectively even). For

notational ease, this formula is written after performing a projective rotation r,m,s,t ( , # )
r,m,s,t ( , # ) along with a reflection in the modulus # .
In this paper we shall present a detailed comparison between the free energy (2.48) and the
integrated (with respect to the branch point loci) two-point correlation function (2.42) of twist
fields corresponding to transpositions, which requires the generalization of the combinatorial
identity (2.12) to coverings with two simple branch points as explained in Section 2.4 above.
While the auxiliary genus one surface of modulus # above is anticipated a posteriori on general
grounds from the WeierstrassPoincar reduction theory for branched covers [12], its geometrical
significance has been hithereto unclear. In the following we will identify this torus explicitly,
which among other things will provide the transformation from the branch point loci to the
modulus # required to match the expressions (2.42) and (2.48), as well as the measure d(z)
and coupling constant required to define the DVV vertex operator (2.40) on an elliptic curve.
3. Z 2 orbifolds
The purpose of this section is to establish the equivalence of the two-point function for the
DVV vertex operator in the symmetric product R24 Z2 with the N = 2 contribution to the genus
two free energy (2.48) of the bosonic DLCQ string. For the former calculation we will exploit the
known formulae [21] for the multi-loop partition functions and twist field correlation functions
on the geometric orbifold S1 /Z2 . For the latter computation we connect the form of the total
reduced free energy (2.48) to the theory of Prym varieties for generic genus two covers of the
torus T of modulus . By a theorem due to Mumford [39], the only coverings that generate Prym
varieties are double covers with at most two branch points, and our case of genus two covers
over an elliptic curve. Our proof puts the covering surface principle sketched in Section 2.4 on
more solid ground, and provides a non-trivial explicit check for the computation of twist field
correlation functions through two rather distinct methods.
3.1. Target space vs. permutation orbifold
For later use, we begin by elucidating the correspondence between the sigma model conformal field theories on the geometric orbifold R24 /Z2 and on the permutation orbifold R24 Z2 . For
this, let us consider the S1 /Z2 target space orbifold of a free boson X compactified on a circle
S1 of radius R, where the group action is the reflection involution X X. On the other hand,
the permutation orbifold S1 Z2 is defined on the tensor product of the S1 conformal field theory
with itself. Labelling the two copies of the boson X by X a , a = 1, 2, the group action of the
permutation orbifold is given by X 1 X 2 , X 2 X 1 . This can be compared to the geometric
orbifold group action by introducing new coordinate fields X = X 1 X 2 , so that the Z2 permutation group now acts as X X . It follows that the permutation orbifold is equivalent
277
to the target space orbifold plus an independent free boson X + on S1 . The partition functions of
the two theories are thus related by
Z Z2 (, R) = z(, R)Zorb (, R),
(3.1)
X+
where z(, R) denotes the partition function of the compactified scalar field
and Zorb (, R)
that of the S1 /Z2 theory.
It is instructive to check the identity (3.1) explicitly at one-loop order in the decompactified
circle theory. The amplitude for the boson X + on S1 is given by the worldsheet instanton sum

R
4 2
R 2 |m m |2
cl
z(, R) = z( )z (, R) :=
(3.2)
,
exp
2
2 |( )|2

m,m Z
where z( ) is the modular invariant amplitude (2.22) for the free boson on the real line (so that
zcl (, R = ) = 1) and henceforth we set the Liouville field = 0. The sum in (3.2) runs over
classical solutions with the given winding numbers around the generating cycles of a canonical
homology basis. For the partition function of the target space orbifold, we note that the oscillator
part z( ) of the partition function (3.2) is independent of the radius R. A monodromy homomorphism for an unramified double cover of a genus one surface is characterized by a binary
pair (, ) (Z/2Z)2 , where 0 (respectively 1) labels periodic (respectively antiperiodic) global
monodromy around the canonical homology cycles (, ) of the base. In the twisted sectors, the
Z2 action X X kills non-trivial instantons at one-loop (as a consequence of the RiemannRoch theorem), while
the quantum parts may be computed by equating the Z2 -twisted partition
function at R = with that of the untwisted S1 theory at the self-dual radius R = 1/

which coincides with the multi-critical AshkinTeller model. The result is [21]

( ) ( ) ( )
1
+
+
,
Zorb (, R) = z(, R) +
(3.3)
2
( ) ( ) ( )
2
where we have denoted the JacobiErdlyi theta constants by i ( ) := i (0| ). Finally, the vacuum amplitude of the Z2 permutation orbifold can be determined from the formula (2.3) as

1
+1
Z2
2
Z (, R) =
(3.4)
z(, R) + z(2, R) + z , R + z
,R .
2
2
2
Clearly the contributions to both sides of the formula (3.1) from the untwisted sector match.
For the contributions from the twisted sectors, we use the identities 3 ( + 1) = 4 ( ) and
2 ( )3 ( )4 ( ) = 2( )3
(3.5)
to derive the elliptic function relation

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

3 ( )3 ( + 1)
1
1
+
=
+

4
|3 ( )( )| |3 ( + 1)( )|
2( )
1
1
1
+
,
=
2 +
2
2
2|(2 )|
|( 2 )|
|( +1
2 )|
(3.6)
2
where in the last line we substituted the identity 3 ( ) = ( +1
2 ) /( + 1) and used |( + 1)| =
|( )|. This equation establishes the R limit of the formula (3.1), for each twisted sector,
which easily generalizes to Z2 orbifolds of Rd by taking appropriate powers.
278
3.2. DLCQ strings on double covers

We now turn to the explicit form of the N = 2 part of the genus two bosonic DLCQ free
energy (2.48) which is given explicitly by

d2 # # 2
g 2 12
F2 = s
(3.7)
10 s ,
,

2

16 16
(2# )12
s=0,1

where the corresponding period matrices read

1
#
#
2
s , := r=2,m=1,s,t=1 , =
12
s
2
12
#

.
(3.8)
By modular invariance it suffices to restrict to the s = 0 contribution. To see this, we define the
SL(2, Z) modular transformation # = # /(2 # + 1). Then the period matrices 1 ( , # ) and
0 ( , # ) are related by the Sp(4, Z) modular transformation

1
0 , # = A1 , # + B C1 , # + D
(3.9)
given by the matrix
1 1
0 1
g=
0 0
0 2
0
0
1
1
0

0
A
=:
0
C
1

B
.
D
(3.10)
Since the integration over # in (3.7) runs over a fundamental domain for SL(2, Z), we can
compensate the omission of the s = 1 term by simply doubling the s = 0 contribution.
Let us now simplify the integrand of (3.7) by working out explicitly the product of theta
constants appearing in the genus two modular form (2.44). Starting from the reduction (2.54)
with N = 2, one has j = 2 when a1 = 1 and j = 3 when a1 = 0, and hence

a1 a2
a # eia2 b2 /2 a1 a2 1
1

=
b1
0
b 0 ,
4 2
4 4 #
2 i #

a2 1 1
a1 a2 1
1
b2 a1

+ 0
+ #
.
+ (1) b1
4
2 2
4
2 4
(3.11)
Using the property

a

1
(z| )
ab z + = (1)ab b+1
(3.12)
2
where b + 1 is understood modulo 2, one can now write down the product of the even genus
two theta constants in (2.44). To simplify the formulae somewhat, in the ensuing calculations
we will use the shorthand notations i := i (0| 21 ), i := i ( 14 | 21 ), i# := i (0| 41 # ) and
i# := i ( 14 | 41 # ).
Then the modular form (2.44) can be expressed as

10 0 , # =
A2 B 2
,
232 ( # )10
(3.13)
279
where

A = 3 3# + 4 4# 2 2# + 1 1# 4 3# + 3 4#

3 3# 4 4# 4 3# 3 4# 2 2# 1 1# ,

B = 3 3# + 4 4# 2 2# + 1 1# 4 3# + 3 4# 1 2# + 2 1# .
(3.14)
(3.15)
= 0 (and
The products (3.14) can be immediately simplified by noticing that 1 =
#
similarly 1 = 0). One finds

A = 2 2 2# 2 3 2 4 2 3# 4 + 4# 4 3# 2 4# 2 3 4 + 4 4 ,
(3.16)
#2
# 2

2
2

#
2
#
2
#
#
#
#
2
B = 1 2 3 4 1 + 2 3 + 4 + 1 2 3 4 1 + 2 3 + 4

+ 1 2 3# 4# 1# 2 + 2# 2 3 2 + 4 2 + 1# 2# 3 4 1 2 + 2 2 3# 2 + 4# 2 .
(3.17)
1 (0| 21 )
Using (3.12) and the parity properties of the theta functions, one notices that 1 = 2 and
3 = 4 . We may thus simplify (3.17) further to
B = 161 2 3 4 1# 2# 3# 4# = 42 2 3 4 2# 2 3# 4# ,
(3.18)
where the second equality is a consequence of the identity for products of theta functions with
identical modulus given by
1 (2z| )2 (0| )3 (0| )4 (0| ) = 21 (z| )2 (z| )3 (z| )4 (z| ),
(3.19)
applied with z = 1/4.

The next step consists in using the modulus doubling identities
2 (0| )2 = 22 (0|2 )3 (0|2 ),
3 (0| )4 (0| ) = 4 (0|2 )2 ,
3 (0| )2 + 4 (0| )2 = 23 (0|2 )2
(3.20)
along with the Jacobi abstruse identity

3 (0| )4 4 (0| )4 = 2 (0| )4
(3.21)
i := i (0| 1 )
i# := i (0| 21 # )
on both i and i# . After introducing the notations

and

AB = 1282 2 3 2 4 2 2# 2 3# 2 4# 2 4 4 2# 4 + 3# 4 4# 4 2 4 + 3 4 .
we find
(3.22)
We now undo the projective rotation 0 0 and the reflection # # that were used to
write (2.54), in order to use theta functions which are convergent on the standard domain of genus
one moduli 2 > 0. This affects only i , because its modulus changes as i (0| 1 ) i (0| 1 ).
The reflection of the off-diagonal elements of the period matrix (2.51) which flips the sign of the
argument of i via (2.54) is easily checked to have no effect on the product (3.22).
The final transformation we perform on the product (3.22) is a modular S transformation on
both i and i# given by

1

= i 4 (0| ),
2 0

1

= i 3 (0| ),
3 0
280

1

4 0
= i 2 (0| ).
Then we can write the modular form (3.13) as

10 0 , #
10 12 12
= 2 #
4
4 4
4 2
4 4
2 2 # 4 + 3 2 4 2 # + 3 2 #
,
(3.23)
(3.24)
where we have used (3.5). Substituting into (3.7) and using (3.21) we arrive at our final form for
the two-loop DLCQ free energy given by
|( )|24
gs2
8(16 2 )12 | |8

4

d2 #
(2 # )6
.

(2# )12 3 ( )4 4 (2 # )4 4 ( )4 3 (2 # )4

F2 =
(3.25)
3.3. Prym varieties

Our next goal is to determine the genus one modulus # explicitly in terms of the branch point
loci on the base torus T. This modulus arose generically from the algebraic WeierstrassPoincar
reduction of the period matrix of the covering surface to the normal form (2.51), which is
a consequence of the fact that the genus two Riemann period matrix in this instance satisfies a
Hopf condition [12]. We will now elucidate the geometrical significance of this modulus for a
generic genus two cover over T of degree N = rm, and then show how in the case of double
covers this geometrical realization determines it explicitly as a function of branch points on the
worldsheet T.
Let f : T be a holomorphic map. Let i , i = 1, 2 be the canonical, normalized Abelian
holomorphic differentials on with the periods

(3.26)
j = ij and
j = ij .
i
On the base elliptic curve T the holomorphic one-form is dz with the periods dz = 1
and dz = . The two sets of differentials are related by the pull-back homomorphism
C) through
f : H 1,0 (T, C) H 1,0 (,
f (dz) = h1 1 + h2 2
(3.27)
for some complex numbers hi . These numbers can be determined by integrating the relation
Z) using (2.50), and with respect to the basis
(3.27) over a canonical homology basis of H1 (,
specified by (2.49) they are given by
h1 = r
and
h2 = 0.
(3.28)
:= H 1,0 (,
C)/H 1,0 (,
) be the principally polarized Jacobian variety of ,
Let Jac()
2
2
It
where = Z Z is the lattice of rank four induced by the period matrix of .
0
can be identified with the Picard group Pic () of isomorphism classes of flat line bundles
281
over , in correspondence with degree zero divisors, and it is isomorphic to the complex two provided by the Abel map
dimensional torus C2 / . There is an embedding of into Jac()
z
A : z (1 , 2 ), which also provides the mapping from divisors to the Jacobian variety. The

theta divisor is the analytic subvariety of the Jacobian defined by the equation 00 (z1 , z2 | ) = 0.
On the base, the Jacobian torus can instead be identified with the elliptic curve T itself and one
has Jac(T)
= T.
It follows from a general property of finite morphisms between smooth projective curves [39]
that the holomorphic map f : T can be factorized by means of a commutative triangle
g
1
(3.29)
f1
T
where f1 : 1 T is an unramified cover. The induced pullback morphisms on the Jacobian tori
is injective. This accounts for
have the properties that ker(f )
= ker(f1 ) and g : 1 Jac()
the first diagonal entry in the period matrix (2.51). The complimentary subvariety to im(f )
=T
in the Jacobian torus C2 / is gotten from the norm morphism
T
f : Jac()
with f (z1 , z2 ) := h1 z1 + h2 z2
(3.30)
which takes the divisor class D of degree zero by applying f to each point of the divisor. The
called the Prym variety
kernel of this morphism is a principally polarized subvariety of Jac()
of the cover and in the present case it is a complex one-dimensional torus C/(Z Z) whose
period is called the Prym modulus. In the basis defined by (2.49), from (3.28) it follows that the
kernel of (3.30) in C2 consists of all points of the form (z1 , z2 ) = ( mr , z) with m Z and z C.
Passing to the quotient C2 / using (2.51) truncates to points (0, z) with the identifications
z z + mr1 + # m2 for any m1 , m2 Z. It follows that the Prym modulus in this basis is given
by
= r #
(3.31)
and we have explicitly identified the second elliptic modulus in (2.51). Using the factorization
(3.29) one shows [39] that the induced theta divisor on ker(f ) is r times the theta divisor
defining its principal polarization, and hence that ker(f ) is a PrymTyurin variety.
So far everything we have said holds generally for any N -sheeted genus two cover of the
torus T. When N = 2, wherein only the r = 2 term contributes in (2.48), the Prym variety possesses a special characterization [40] which enables one to make this construction much more
explicit. Consider the element of the symplectic group Sp(4, Z) given by
0 0 1 0

0 0
A B
0 1
g=
(3.32)
.
=:
1 1 0 0
C D
0 0
1 1
Z) represented by
It induces the change in basis of H1 (,

1 1 0 0
D
B

,
M = M
M
=
with
0 0 1 1
C A
(3.33)
282
and the genus two modular transformation

1
0 , # = A0 , # + B C0 , # + D
with
0
1
=
2
(3.34)

,
(3.35)
where we have used (3.31) with r = 2. From (2.50) it follows that

f ( 1 ) = f ( 2 ) =
and f (1 ) = f (2 ) = .
(3.36)
Integrating both sides of (3.27) in this basis thus gives h1 = h2 = 1, and hence
f (dz) = 1 2 .
(3.37)
What makes the instance of a double cover f : T special is that it has a canonical confor satisfying f = f , which is the involution permuting the sheets
mal automorphism : ,
From (3.36) it follows that
of the cover. It uniquely determines the covering with T = /.
( 1 ) = ( 2 )
and (1 ) = (2 ),
(3.38)
and hence that

(1 ) = 2 .
(3.39)
The holomorphic one-form

= 1 + 2
(3.40)
is called the Prym differential and it is the unique holomorphic differential on the two-sheeted
cover which is odd under the defining involution with () = . It follows from (3.37)
(3.40) and the form (3.35) of the period matrix in this basis that the Prym period is determined
by

= .
(3.41)
1
The Prym differential is normalized with respect to the 1 cycle, while it has vanishing periods
around 1 2 and 1 2 . At the level of Jacobian varieties, the Prym variety ker(f ) is
consisting of degree zero divisor classes which are odd
isomorphic to the subvariety of Jac()
is
under the involution . Note that from (3.37) it follows that the embedding f : T Jac()
isomorphic to the subvariety invariant under .
Similarly to the even holomorphic one-form (3.37), the Prym differential (3.40) may be given
explicitly as the pull-back = f (pr(w1 , w2 )) of a multiplicative differential pr(w1 , w2 ) =
pr(z; w1 , w2 ) dz on the base elliptic curve T with modulus . It is required to have a square
root cut singularity about each of the branch points w1 , w2 T of the cover and to have global
periodicity under z z + m + n for any m, n Z. This uniquely determines the multiplicative
differential on T in terms of JacobiErdlyi elliptic functions as
2
1 (z w1 +w
2 | )
.
pr(z; w1 , w2 ) =
1 (z w1 | )1 (z w2 | )
(3.42)
283
The Prym modulus (3.41) may then be written as

1
1
# = =
2
2
pr(w1 , w2 )
pr(w1 , w2 )
(3.43)
thereby determining the desired explicit dependence of the elliptic modulus # on the branch
point loci. As expected, in the unramified limit w1 w2 wherein the branch cut on T
closes up. It follows from (3.35) that this limit corresponds to approaching a separating boundary
component of moduli space, wherein the genus two Riemann surface degenerates into two
copies of the base torus T.
Thus far we have not accounted for global monodromy of the covering map f : T,
i.e., the above formulas are written in the untwisted sector (, ) = (0, 0). For each twisted sector
(, ) (Z/2Z)2 there is a holomorphic Prym form , which is odd under the involution
and which has non-vanishing periods only around the ( 1 , 1 ) cycles of the homology group
Z). They project onto multiplicative differentials pr, (w1 , w2 ) on T which have square
H1 (,
root cut singularities about the branch points w1 , w2 T. The Prym form corresponding to the
characteristic (, ) can be gotten from the untwisted one via a crossing transformation of the
branch points
w1 w1 + +
to get
and
w2 w2

pr, (z; w1 , w2 ) = pr z; w1 + + , w2
(3.44)
(3.45)
with pr0,0 (w1 , w2 ) = pr(w1 , w2 ). The corresponding Prym modulus is defined by

, =
pr, (w1 , w2 )
pr, (w1 , w2 )
(3.46)
with 0,0 = .
These constructions of Prym varieties and Prym differentials have natural generalizations to
double covers of a genus g surface with k = 2n branch points (n = 0, 1), with genus
g = 2g + n 1 determined by the RiemannHurwitz formula (2.35). In this case the Prym variety is a complex torus of dimension g + n 1. By the RiemannRoch theorem, there are exactly
g + n 1 independent holomorphic one-forms which are odd under the automorphism and
which form a basis for the Prym differentials. The remaining g even ones on are preimages
of the holomorphic differentials on the base space . A further generalization exists to more
C) is
general Abelian automorphism groups of a cover. The action of the group on H 1,0 (,
then always diagonal on a suitable basis of holomorphic differentials and the subspace corresponding to a non-trivial set of eigenvalues are pull-backs of multiplicative elliptic differentials,
whose multiplicative factors are given by these eigenvalues. This is exploited implicitly in the
computation of ZN orbifold twist field amplitudes in [16].
3.4. Correlation functions of twist field operators
We now come to the computation of the two-point function (z) (0) Z2 of Z2 twist fields
(z) = 12 (z) in the R24 Z2 permutation orbifold. We begin by discussing some general aspects
concerning global monodromy in the covering surface construction of Section 2.4. Recall that the
sum appearing in the correlation function (2.36) of interest (computed with the amplitude (2.43))
284
is restricted to the set of admissible monodromy homomorphisms such that each connected
component of the corresponding cover of the base torus T is a surface of genus two. This
is ensured by the requirement that the monodromy of the generators of 1 (Tw ) encircling the
punctures be a simple transposition in each orbit O(). The period matrix ,w depends
on the monodromy only via its stabilizer subgroups, which are the finite index subgroups H <
1 (Tw ) obeying the admissibility criterion (2.38). Consider the stabilizer subgroup H = Ha
of a given sheet a corresponding to a transitive homomorphism : 1 (Tw ) SN . Since it is
isomorphic to 1 ( w ) and since there are 2N 2 preimages of the two branch points of Tw ,
it is a group freely generated by 2N + 1 elements. The kernel of the forgetful homomorphism
is given by the normal closure
: 1 ( w ) 1 ()

N H (1 , . . . , 2N 2 ) = h1 h1 , . . . , h2N 2 h1 h H
(3.47)
of the generators i encircling the ramification points.
When N = 2 the generators i are easily determined. Let us use the presentation 1 (Tw ) =
, , . The generators of 1 ( w ) encircling the ramification points are the (pullbacks of the)
squares of the generators of 1 (Tw ) which encircle the punctures. For N = 2, the preimages of
the punctures are precisely the ramification points, and hence one has

2
ker( ) = N H 2 , [, ] .
(3.48)
There are four homomorphisms with the prescribed monodromy representing the four twisted
sectors (, ) (Z/2Z)2 , and all of them are transitive. There are correspondingly exactly four
admissible subgroups H of index two. Since Z2 is an Abelian group, conjugacy classes of homomorphisms contain only one element. Their precise forms and the corresponding stabilizers
can be determined explicitly.
The simplest example is provided by the admissible homomorphism 1 which sends to
the transposition (1 2) and , both to the identity. Its stabilizer H1 is freely generated by the
words , , 1 , 1 , 2 . We then seek a presentation of the generators 1 , 2 , 1 , 2 ,
of 1 ( w ) such that the quotient by the relations 2 = ([, ] )2 = 1 yields the group 1 ()
with [ 1 , 1 ][ 2 , 2 ] N H1 ( 2 , ([, ] )2 ). For the case at hand, one sees that the assignments
1 = , 1 = , 2 = 1 , 2 = 1 , = 2 suffice. This determines the homomorphism
of fundamental groups f1 , where f is the restriction of the covering map to the marked
surfaces. Since the abelianization of 1 (Tw ) factors through this map, the powers of , in the
canonical homology generators 1 , 2 , 1 , 2 gives the map (2.50). This yields the covering
homology matrix

1 1 0 0
(3.49)
M1 =
0 0 1 1
which obeys the Hopf condition. Reduction of this matrix via an Sp(4, Z) modular transformation
as in Section 3.3 above yields the normal form (2.49) with r = 2, m = t = 1, s = 0. The other
three admissible homomorphisms are similarly treated.
However, the above formalism is sensitive only to the induced homomorphism f between
homology groups rather than homotopy groups, and it is difficult to proceed further with the
explicit construction of the modular invariant amplitude (2.36). We will return to this issue in
some more detail in the next section. Here we shall compute the twist field correlation function
using results of [21] where the correlation functions are computed for a free boson X in the
285
geometric orbifold S1 /Z2 using the covering space method explained in Section 2.4. The twopoint correlation function on the torus T with twist field insertions may be computed from the
path integral over field configurations X on the double cover which are odd under the canonical
involution with X = X mod 2R. As in Section 3.1 above, in each twisted sector (, ) the
amplitude is a product of a radius independent quantum piece and a classical piece. The instanton
configurations on the worldsheet that contribute to the classical part of the correlation function
are analogous to the untwisted ones used in Section 3.1 above. In the homology basis specified
by (3.33), the boundary conditions of the boson X in the given twisted sector are characterized
by the Prym differential , . The classical contribution is then completely analogous to that in
(3.2) with the period equal to the Prym modulus , .
The quantum contributions may be computed by equating the two-loop orbifold amplitude
with that of the circle theory at the self-dual radius as before, with the additional observation
that the twist fields in this correspondence are equivalent to magnetic vertex operators [21]. At
this radius the momentum lattices appearing in the classical partition sums can be built up from
a finite number of square sublattices. A term by term comparison of the chiral blocks gives
an expression for the ratio of a twisted determinant to the untwisted determinant z( ) as the
modulus squared of a holomorphic function of the positions of the branch points on T. In this
way the normalized twist field two-point function on T with the twist characteristic (, ) in the
Z2 target space orbifold of the compactified boson X can be written as [21]

,
2
(z) (0) orb = z c zcl (, , R),
(3.50)
where
c

= E(z)
1/8
a
(0|, )
! b
.
a
z
)
(
(0|
a+
|
)
b+ 2
b
(3.51)
Here we have used translation invariance to fix one of the twist field insertion points at the origin,
and (a, b) = (1, 1) is a fixed arbitrary characteristic. The quantity E(z) is the prime form of the
elliptic curve T given by
E(z) =
1 (z| )
1 (0| )
(3.52)
1 (z| ), and it is the doubly periodic elementary solution of the Laplace equawith 1 (z| ) := z
tion on the torus. The independence of the expression (3.51) on the choice of characteristic (a, b)
is the mathematical statement of the Schottky relations [40] (see Section 3.5 below).
We can now write down the desired amplitude in the permutation orbifold R24 Z2 . For this,
we redefine the independent bosons Xia , i = 1, . . . , 24, a = 1, 2 to Xi = Xi1 Xi2 as in Section 3.1 above. Since the Z2 permutation group acts on the 24 bosons simultaneously, both the
global and local monodromy of the fields Xi+ are trivial, and the twist operators act as the identity on these fields. The path integral over Xi+ thus leads simply to an overall factor z( , R)24 .
On the other hand, the twist operators act as a Z2 twist field simultaneously on all sigma model
fields Xi . It follows that the correct prescription is to raise the geometric Z2 orbifold twist field
correlation function in each sector to the power 24, and then sum over the twisted sectors. The
Xi+ contribution is cancelled in the suitably normalized correlation function by the same factors coming from the partition function (3.1). One should then take the decompactification limit
286
R , wherein zcl (, , R = ) = 1 as before. This gives the two-point function

Z

, 24
1
(z) (0) 2 = lim
(z) (0) orb .
R 2
2
(3.53)
(,)(Z/2Z)
Substituting (2.22) and (3.50)(3.52), and using the identity

1 (0| ) = 2( )3 ,
then leads to the explicit formula

Z2 1 4 2 5/2 12 ab (0| )4 6

(z) (0) =
(z| )( )5
2
2
1
(3.54)
a+ z 24
b+ ( 2 | )

a (0| )2 .
,
b
2
(,)(Z/2Z)
(3.55)
3.5. DLCQ free energy = DVV correlator

We will now prove the main result of this section, establishing the equivalence

Z
42
d(z) (z) (0) 2
F2 =
2 (0)
(3.56)
between the DLCQ free energy on the double cover T given by (3.25) and the translationally invariant correlator (2.42) of the DVV vertex operator determined by the twist field two-point
function (3.55) on R24 Z2 . We begin by observing that the right-hand side of the formula (3.56)
is independent of the twist characteristic (, ) in (3.55). This follows from the fact that one can
get any twisted sector from the untwisted one (, ) = (0, 0) by a crossing transformation (3.44).
Crossing symmetry of the orbifold theory, along with modular invariance at genus one, is the
remnant of genus two modular invariance on the covering space [21]. One can check this invariance explicitly by showing that the z-dependent part of the correlation function (3.50) transforms
under the crossing transformation (3.44) precisely by changing (0, 0) (, ), just like the Prym
modulus according to (3.45).
Next we examine the change of integration variables from the modulus # in (3.25) to the
branch point location in (3.56). For this, we require the Jacobian |d # /dz|2 . The explicit dependence of the Prym modulus on the branch point loci is given by the formula (3.43) with
w1 = z, w2 = 0, but this is not convenient for computing the requisite derivative d/dz. Instead,
it is more useful to use the implicit dependence of the Prym modulus on the branch point z
dictated by the Schottky relations. For zero characteristics (, ) = (0, 0), they are given by
!
!
i ( 2z | )i (0| )
j ( 2z | )j (0| )
(3.57)
=
.
i (0|)
j (0|)
By separating the explicit z and dependences for i = 4 and j = 2, we can write (3.57) as
"
2 (0| )2 ( 2z | )
2 (0|)
(3.58)
=
.
4 (0|)
4 (0| )4 ( 2z | )
Taking the total derivative of the relation (3.58) with respect to z yields
"

2 (0|) d
d 2 (0| )2 ( 2z | )
=
.
4 (0|) dz
dz 4 (0| )4 ( 2z | )
(3.59)
287
We can transform the derivative by using the heat equation

i (z|)
i 2 i (z|)
=0
+
4
z2
to get the form

2 (0|)
i
2 2 (w|)
(w|)
.
=
4
2
4 (0|)
w 4 (w|) w=0
44 (0|) w
We may then use the identity for the derivative of a ratio of theta functions given by

1 (w|)3 (w|)
2 (w|)
= 3 (0|)2
w 4 (w|)
4 (w|)2
to arrive at

2 (0|)
i 3 (0|)3 1 (0|)
.
=
4 (0|)
4
4 (0|)2
(3.60)
(3.61)
(3.62)
(3.63)
The differentiation on the right-hand side of (3.59) is an easy exercise. This calculation can
be repeated starting from the Schottky relation (3.57) with i = 4 and j = 3. The final result is
identical to that above with the replacements 2 3 of theta functions everywhere. In this way
we can finally write
!

!

2 (0| )2 ( 2z | ) 3 (0| )3 ( 2z | )
d 2

2

dz =

2 (0|)
3 (0|)

z

2 (0| )2 3 (0| )2 1 ( 2 | )2
4 (0|)4
.

(3.64)
z 3
4 (0| )
4 ( 2 | ) 1 (0|) 2 (0|)3 (0|)
To compare (3.64) with the elliptic functions appearing in the expressions (3.25) and (3.55)
for (, ) = (0, 0), we exploit the identity (3.19) and the Schottky relations (3.57) again to write
!

2

z

3
abii ( 2z | ) abii (0| )

d
(
|
)
1

2
2

ai
(3.65)
,
dz = (z| ) (0|)2

bi (0|)
1
1
i=1,2
where (ai , bi ) {(0, 0), (0, 1), (1, 0)} are arbitrary characteristics which we will choose conveniently. We can now use the identities (3.5), (3.19) and (3.54) along with
2
2
2
2
2
2
z
z
z
3 4 0 4 3 0 = 1 2 0
(3.66)
2
2
2
to expand the expression (3.65) into
!

8

abii ( 2z | ) abii (0| )
42

d 2
)
1
(

ai

dz = 218 (z| )6

(0|)
1
bi
i=1

2 2
z
z
z
2 3 4 3 0 4 0
2
2
2
# 2
2
$
2
2 4
z
z

.
3
4 0
4
3 0

2
2
(3.67)
288
We have again used (3.57) to infer that every term of the product in (3.67) is independent of the
chosen characteristic (ai , bi ).
Let us now substitute (3.67) into the integral (3.25), recalling that = 2 # . We can again exploit the freedom in choice
combine the theta functions in (3.67) with
of characteristics (ai , bi ) to
the ones i (0|) =: abii (0|) and i (0| ) =: abii (0| ) appearing in (3.25) by re-expressing
Dedekind functions as theta functions
! using (3.5). The simplification effectively amounts to
replacing each factor i (0|) with i ( 2z | )i (0| ). We can use this trick to cancel the difference of theta functions appearing in the integrand of (3.25) by simply doing this replacement
for every term, and remembering that there are in total 40 factors of i (0|) in each term of the
expansion of the fourth power of the difference.
In this way, it is straightforward to see after some inspection that the free energy (3.25) may
be written in terms of an integral over the branch point location on the torus T as
F2 ( ) =
|( )|30
gs2
(32 2 )12 4| |8
!

48

abii ( 2z | ) abii (0| )
2

d z
1

ai

.

(Im (z))12 1 (z| )6
b (0|(z))
T
i=1
(3.68)
It is now clear that with (3.55) the DLCQ free energy function (3.68) can be expressed in the
form (3.56) if we choose the measure
d/2
2
2
d2 z
Im (z)
,
with (z) =
d(z) =
(3.69)
(z)
2
where d = 24 is the spacetime dimension of the permutation orbifold. Using (0) = , the
coupling constant is then given by
!
4gs
= 3
(3.70)
.
|512 |4 2
Note that the coupling (3.70) has the correct infrared behaviour 0 as 2 to ensure that
the interacting sigma model approaches a conformal fixed point in the infrared limit.
From the genus two perspective the origin of the measure (3.69) is clear. It arises from the
Sp(4, Z) modular invariant integration over the moduli space of genus two branched covering
maps f : T. From the genus one perspective it is a consequence of the conformal anomaly,
implying that the local twist field correlation functions depend on the coordinatization chosen on
the Riemann surface T. For the twist field operators the natural choice is the coordinate z of T,
but to induce the modular invariant interactions of strings in the symmetric product a non-trivial
integration measure (3.69) must be adapted. We will see this explicitly in the next section when
we study the action of the mapping class group of the punctured torus Tw .
4. Non-Abelian orbifolds
In this section we address some issues surrounding the extensions of the results of the previous
section to SN orbifolds with N > 2. At this stage, however, we have not succeeded in making the
construction as explicit as for the Z2 orbifold. The main technical obstruction is the combined
non-commutativity of the twist group SN and the fundamental group 1 (Tw ) of the punctured
torus. For twist group Z2 the image of the latter group under a given monodromy homomorphism
289
is of course an Abelian group, enabling explicit constructions. But these constructions become
ambiguous and inconsistent in the non-Abelian case, as one must deal with the full non-Abelian
homotopy group and not just its abelianization to the homology group. We are not aware of
any direct computation of the twist field correlation functions in these specific instances. In the
following we will highlight some of the main technical issues surrounding these calculations in
the higher degree permutation orbifolds, and in particular to what extent the DLCQ free energy
(2.48) can bused to provide an explicit representative for the DVV correlator (2.42) using the
combinatorial formula (2.36). One of the outcomes of this analysis will be a more precise, general
description of the measure d(z) required in the definition of the vertex operator (2.40).
4.1. Uniformization construction
Let us recall the general construction of Section 2.4. A correlation function involving twist
fields alone in any permutation orbifold is defined through the generalized partition function
(2.36). It gives a twist field correlation function on a worldsheet as a sum over twisted sectors,
each characterized by a conjugacy class of monodromy homomorphisms. One term is given
by the partition function of the covering space determined by Hurwitz data, comprising the
monodromy, the complex structure of the worldsheet and the insertion points of the twist field
operators. The issue is how to determine the covering space and its complex structure in terms
of the Hurwitz data. The monodromy in the case of k distinct insertion points on the worldsheet
is a homomorphism : 1 (w ) G < SN , and the general RiemannHurwitz formula (2.35)
for ramified coverings gives the genus g of the covering space. Determining the topological type
of the cover is analogous to the unramified case. The fundamental group of the marked cover
w is given by a stabilizer subgroup Ha < 1 (w ). The index a is the label of a sheet, which is
permuted by the twist group G < SN , and different choices of a result in conjugate subgroups of
1 (w ) corresponding to different choices of pre-image of the base point of 1 (w ) as the base
point of 1 ( w ).
However, it is much more difficult to determine the complex structure of the cover. Recall that the prescription for the unramified case was to choose a uniformizing homomorphism
u : 1 () U such that = U/u(1 ()). Then one needs to restrict u to the stabilizer subgroup of 1 () corresponding to the monodromy homomorphism . But the domain of the
monodromy is 1 (w ) for the ramified case, which is a group distinct from 1 (). Hence it
is not straightforward to extend this uniformization method to the case of branched coverings.
Consider the commutative diagram
w
(4.1)
where the maps and are the canonical inclusions (filling in the deleted points), and f is
the restriction of the covering map f to the punctured surfaces. Passing to the corresponding
pushforwards, this diagram induces a commutative diagram of fundamental groups given by
1 ( w )
1 (w )
1 ()
f
1 ()
(4.2)
290
Let T (k, g) denote the Teichmller space of genus g Riemann surfaces with k punctures. Let
M(k, g) be the mapping class group of the (marked) Riemann surface w acting on T (k, g).
One seeks maps which fit into the commutative diagram
g)
T (k,
T (0, g)
(4.3)
T (k, g)
T (0, g)
associated to the covering and the inclusions such that the vertical arrow on the left is given by
the surjective map U/u(1 ( w ))
= U/u(Ha ) U/u(1 (w )), where u is a uniformizing map
of punctured surfaces. In this way one can incorporate the information from the monodromy
contained in the admissible finite index subgroup Ha . Note that the corresponding complex dimensions of the spaces involved in (4.3) map as
3g 3 + k
3g 3
(4.4)
3g 3 + k
3g 3
for g > 0 (except for dimC T (0, 1) = 1).

The problem rests in the construction of the horizontal arrows of (4.3). Since the pushforward
is a group homomorphism, the image of an element of a uniformizing group u(1 (w )) <
PSL(2, R), which we identify with the complex structure given by U/u(1 (w )) T (k, g), is a
coset and thus not an element in PSL(2, R). Thus even though the quotient of the uniformizing
group of the marked surface by the normal closure of the parabolic generators is isomorphic to
1 () (by the admissibility constraint), it is not a subgroup of PSL(2, R). The same remarks
apply to the map inducing the top horizontal arrow in (4.3). Therefore it is not possible to apply
the method of uniformization which worked for the unramified case, and the forgetful maps (i.e.,
the horizontal arrows in (4.3)) need to be constructed by hand.
Let us specialize to our main problem of interest, where the base space is the torus = T
with k = 2 simple branch points. For an N -sheeted cover of genus g = 2 there are k = 2N 2
preimages of these branch points, so that two of the N preimages of a generic point of the
base coincide for a branch point. The main obstacle in constructing the map T : T (2, 1)
T (0, 1) rests in the fact that a flat torus admits a complete euclidean metric, whereas a punctured
torus admits a complete hyperbolic metric. Thus in order to apply uniformization one needs to
construct a map between the space of flat tori and the space of hyperbolic tori. Let us assume that
the branch points are distinguished points of the flat metric on T. Using the automorphism group
of the torus we may fix the location of one of the branch points at the origin. Then one requires a
bijection T (2, 1) T (0, 1) U, where the second branch point z varies in the complex upper
half plane U. This must be done in such a way that a lift of the mapping class group M(0, 1) =
SL(2, Z) to M(2, 1) acts equivariantly on T (2, 1) with respect to this bijection.
An element of T (2, 1) is a twice punctured hyperbolic torus. Using the uniformizing homomorphism u : 1 (Tw ) PSL(2, R), it can be characterized as a discrete Fuchsian group

u 1 (Tw ) = , , PSL(2, R) | | tr | > 2, | tr | > 2, | tr | = tr[, ] = 2 . (4.5)
The hyperbolic generators , correspond to translation along a canonical homology basis of
the unmarked torus, while and [, ] are the parabolic generators corresponding to the punc-
291
tures.3 Then the complex structure is given by U/u(1 (T)). The subgroup (4.5) contains three
real parameters for each generator, two trace relations for parabolicity and a conjugation symmetry which eliminates three parameters, hence the real dimension of T (2, 1) is 3 3 2 3 = 4,
as anticipated.
The space T (0, 1) U is coordinatized by ordered pairs (, z), where is a genus one modulus and z is a distinguished point on T. The mapping class group M(0, 1)
= SL(2, Z) of the flat
torus acts on these pairs through the generators

1 z
T : (, z) ( + 1, z) and S : (, z) ,
(4.6)

obeying S 4 = (T S)3 S 2 = 1. The modular S-transformation here is defined via analytic continuation along a clockwise oriented path around the origin in the complex z-plane. A lift of these
generators to the mapping class group M(2, 1) of the twice punctured hyperbolic torus is presented in [41] as an action on the generators of (4.5) by

1

and S :
T :
(4.7)
.
1
This lift of SL(2, Z) is not unique. In fact, the modular group M(2, 1) is an extension of SL(2, Z)
by B(2, 1)/M(2,1) , where M(2,1) is the center of M(2, 1) and B(2, 1) denotes the twostranded braid group of the torus [41]. Equivariance of the bijection T : T (2, 1) T (0, 1) with
respect to these actions is then the statement
T T = T T
and
T S = S T .
(4.8)
We have not succeeded in constructing explicitly the required modular equivariant bijections,
and it is not possible to write an algebraic formula [42]. One could try to surpass this problem by
working directly with the hyperbolic presentation of the tori, and the known bijection between the
FenchelNielsen coordinates of Teichmller space and the Fuchsian coordinates parametrizing
the uniformizing group [42]. But there is a great deal of ambiguity in this procedure which prevents an explicit construction, and there is no canonical way to identify the modular parameters
of the torus itself and those corresponding to the branch points.
4.2. Homology construction
Given the technical difficulties encountered above, we now turn to an alternative approach
to determining the complex structure of the cover via the push-forward induced on homology
Z) H1 (, Z), which is provided by the abelianization of the diagram (4.2)
groups f : H1 (,
for the fundamental groups. If a canonical basis is fixed both in the homology group of the base
and that of the cover, then this map is given by a 2g 2g matrix M . This matrix can then
be used to determine the period matrix of the cover in terms of the period matrix of the base
and some additional parameters [12]. For sufficiently low genus, the period matrix uniquely
characterizes the complex structure. We will go through this construction in detail for the relevant
case of the genus two cover f : T for the two point function of twist fields corresponding
to simple branch points. In this case the complex structure on is determined by a canonical
C) H1 (,
Z) C.
map H 1,0 (,
3 We could have equivalently used an independent parabolic generator with the relation [, ] = 1.
292
Let us see first how the matrix representation M of f can be determined and compared to the
construction of Section 2.5. The main difference from the N = 2 case studied at the beginning of
Section 3.4 is that for N > 2 the preimages of the punctures are no longer just the ramification
points, since there are 2N 2 > 2 preimages of the branch points. Let 1 (Tw ) = , , ,
the free group on three generators such that ker( ) is the normalizer N1 (Tw ) ( , [, ] ). In
other words, the image of and are the standard generators of 1 (T), whereas and [, ]
correspond to simple closed curves which are contractible to the branch points. The stabilizer
Ha < 1 (Tw ) corresponding to a monodromy homomorphism is a subgroup of index N in
the case of an N -sheeted cover. It can be presented in terms of 4 + (2N 2) 1 words from
1 (Tw ) which freely generate the group Ha . By identifying Ha with 1 ( w ), this presentation
gives the homomorphism f explicitly. There are N 2 independent elements from Ha which
are conjugate to in 1 (Tw ), and another N 2 elements which are conjugate to [, ] . There
is one further element conjugate to 2 and another one conjugate to ([, ] )2 . This is because
the N 2 generators of 1 ( w ) corresponding to simple closed curves contractible to N 2
preimages of a branch point project to the simple closed curve contractible to the branch point,
whereas the other two generators project to curves with winding number two about each of the
branch points.
The normalizer of these 2N 2 generators in Ha is the subgroup ker( ). One then seeks
4 + 2N 3 generating elements such that 2N 3 are in ker( ) and also the commutator product
[ 1 , 1 ][ 2 , 2 ] of a suitably chosen remaining four. In other words, i , i are representatives
under the map . Due to the
of the cosets that project to a canonical homology basis of 1 ()
commutativity of the diagram (4.2) and the abelianization, the entry Mij of the 2 4 homology
covering matrix is the sum of powers of the ith generator of H1 (T, Z) ( or ) appearing in the
Z) ( 1 , 1 , 2 or 2 ). In this way, the two-point function
expression of the j th generator of H1 (,
may be computed by summing over admissible finite index subgroups Ha < = 1 (Tw ).
Let us look at an explicit example of how this works. For N = 3, there are 16 conjugacy
classes of transitive monodromy homomorphisms, each class containing 6 homomorphisms. Accordingly, there are 16 conjugacy classes of admissible index three subgroups of , each class
having [ : N (k )] = 3 representatives. Consider the admissible monodromy homomorphism
1 given by
1 : (2 3),
(1 2)
and (2 3).
(4.9)
The corresponding three sheeted cover may be depicted schematically as
(4.10)
with the parallelogram representing the base torus T. The sheets 2 and 3 are ramified over the
branch point corresponding to , while the sheets 1 and 2 are ramified over the other branch
point corresponding to [, ] (since 1 : [, ] (12)).
The stabilizer subgroup of = 1 (Tw ) = , , can be presented by4

H1 = g1 , . . . , g7 := , 2 , , 2 1 , 1 1 , 1 , 1 1 .
293
(4.11)
The elements g1 , . . . , g7 generate the group H1 freely. One can then determine the generators of
ker( ) as
g 1 = g3 = ,
g 2 = g6 g5 = 2 1 1 ,
g 3 = g4 g2 g11 g21 g5 = [, ] 1 1 ,

2
g 4 = g1 g41 g71 g6 g7 g3 = [, ] ,
(4.12)
where we have underlined the curves on the base that they are conjugate to. Finally, it is possible
to write down the generators
1 = g7 ,
such
1 = g1 g41 ,
2 = g1
and
2 = g2
(4.13)
that5

H1 /NH1 (g 1 , g 2 , g 3 , g 4 ) = 1 , 2 , 1 , 2 [ 1 , 1 ][ 2 , 2 ] = 1 = 1 ().
(4.14)
We can now count the powers of , appearing in (4.13) to determine the matrix representation
M = M1 of f in (2.50) with

0 1 1 0
(4.15)
M1 =
.
1 0 0 2
The remaining 15 admissible finite index subgroups are similarly treated. All instances provide
a matrix representation M which satisfies the Hopf condition and which leads to the normal
form (2.49) after reduction using the symplectic group Sp(4, Z). However, the map from the
set of admissible finite index subgroups to the set of normal forms (2.49) obeying the Hopf
condition is not unique, and there is a large degree of arbitrariness in this procedure. The reason
is that the partial reduction leading to (2.49) involves only Sp(4, Z) transformations, but not
modular transformations of the base. It may happen that an admissible finite index subgroup Ha
is invariant under an SL(2, Z) transformation of the base (e.g., ), in which case one may
get matrices M leading to period matrices which are not related by a modular transformation on
the cover . Thus it is only onto the set of fully reduced Poincar normal forms of M, which
incorporates a sum over all such SL(2, Z) transformations of the base T, that this reduction map
is unique. However, the reduced moduli space for the Poincar normal form is very complicated
and depends sensitively on number theoretic properties of the degree N [12].
4.3. Equivariance of the DVV correlator
The construction of Section 4.2 above determines the dependence of the 2 2 period matrix H on a given admissible monodromy homomorphism, or equivalently a given admissible
4 In practice it is easier to determine the monodromy homomorphism corresponding to a given presentation of a finite
index subgroup.
5 One can check that the elements (4.13) are independent representatives of the generators of the quotient modulo g ,
1
g 2 and g 4 , except for g 3 = [ 1 , 1 ][ 2 , 2 ].
294
finite index subgroup H < , with H = r,m,s,t in (2.51). At this stage we are faced with the
problem of finding the dependence (either explicit or implicit) of the Prym modulus = r # on
the branch point location z T. The construction of Prym differentials in Section 3.3 does not
carry through to the higher degree branched covers, because for any genus two cover f : T
of degree N 3 there are no non-trivial automorphisms : such that f = f [43].
As any genus two Riemann surface is a hyperelliptic curve, the cover does have a canonical
hyperelliptic involution and its hyperelliptic divisor which is the effective divisor of degree
six consisting of the fixed points of . Then there is a unique involution T : T T of the base
such that f = T f [43]. However, given that the above construction is not invariant under
SL(2, Z) transformations of the base, it is not clear how to exploit the hyperelliptic representa and the corresponding Schottky relations, to determine the branch point dependence as
tion of ,
before. This is further reflected in the fact that the standard constructions of cut Abelian differentials (such as (3.42)) for cyclic orbifolds [16] become ambiguous for non-Abelian monodromy.
We are not aware of any constructions of Prym differentials or Prym moduli for higher degree
genus two covers f : T in terms of branch point loci.
On general grounds it follows that the complex structure on the covering surface is uniquely
determined by the holomorphic map f : T in terms of the moduli and z, but not necessarily in an explicit parametrization. We can use results of [44] to ascertain that the desired
explicit branch point dependence does exist and can be used to give some insight into the modular behaviour of the DVV correlator. One of the advantages of the formalism of Section 4.2 over
that of Section 4.1 above is that one can study equivariance properties in the genus two modular
group Sp(4, Z), rather than in the more complicated mapping class group M(2, 1). For fixed
monodromy given by an admissible finite index subgroup H < , there is a holomorphic map

, z H , z
H : T (0, 1) U U2 ,
(4.16)
which is determined generically in [44] via a sewing construction on twice-punctured tori in
terms of JacobiErdlyi theta functions, Weierstrass functions and Eisenstein series on the base
torus T. The primary difference in our specific case is that the modulus # has a square root
cut singularity at each of the branch points w1 = z and w2 = 0, rather than the logarithmic cut
singularity which arises in [44].
Consider the monomorphism SL(2, Z) Sp(4, Z) given by
a 0 b 0

a b
0 1 0 0

(4.17)
.
c d
c 0 d 0
0 0 0 1
This lift of SL(2, Z) acts in the expected way on the domain of the map (4.16) as

a + b
z
,
.
, z
c + d c + d
(4.18)
For each choice of branch for # , the map H is equivariant with respect to this action of
SL(2, Z) < Sp(4, Z) [44] and there is a commutative diagram
T (0, 1) U
SL(2,Z) .
SL(2,Z)
T (0, 1) U
U2
U2
(4.19)
295
This property determines the equivariance of the DVV correlator (2.42), represented by the genus
two DLCQ free energy (2.48) at a fixed value of the degree N . Since under (4.18) the flat area
form on the torus transforms as d2 z d2 z/|c + d|2 , and since the local twist field correlation functions a1 b1 (z)a2 b2 (0) SN have total scaling dimension 6, the scaling properties of the
measure (z) under (4.18) can be explicitly determined.
Given the remarkable agreement of the N = 2 free energy with the twist field two-point function in the Z2 orbifold, it is natural to extrapolate this correspondence and to take the fixed
N DLCQ free energy integrand in (2.48) as the definition of the local twist field correlation
function a1 b1 (z)a2 b2 (0) SN on the R24 SN permutation orbifold, according to the covering
surface principle of Section 2.4. However, the explicit form of the mapping (4.16) displayed in
[44, Proposition 6.2] is far too complicated for an explicit determination of the required Jacobian
|d # /dz|2 (and furthermore one needs an Sp(4, Z) transformation relating their period matrix to
ours). Moreover, it is difficult to arrive at explicit formulas which are illuminating, as the products
(2.44) of theta functions (2.54) are rather involved for N 3.
5. Fermionic orbifolds
In this final section we will study fermionic extensions of the permutation orbifolds considered
thus far, in particular those orbifold sigma models arising in discrete light-cone quantization of
superstrings and heterotic strings in ten spacetime dimensions. We will describe the modifications
of the covering surface principle and twist field operators of Section 2 required in these cases. The
genus two DLCQ free energy amplitudes in these instances are derived in [12]. Given the success
of the bosonic Z2 orbifold model of Section 3, we will use the appropriately modified versions
of the generic covering space principle of Section 2.4 to compute local one-loop correlation
functions of (spin) twist field operators in supersymmetric and heterotic Z2 orbifolds. To the best
of our knowledge these correlation functions have not been previously computed. The analysis
of this section thus provides a powerful application of DLCQ string theory to producing new
explicit expressions for correlation functions in orbifold superconformal field theories on the
one hand, and for the forms of the leading cubic string interactions in the associated superstring
field theories in ten dimensions on the other hand. Throughout we work in the NeveuSchwarz
Ramond formalism.
5.1. Spin twist fields
Consider the superconformal sigma model on the torus with target space R8 defined by the
action

1
1
i (z) + i (z)
i (z) + i (z) i (z) ,
Xi (z)X
d2 z
I (X, ) =
(5.1)
4
2i2
T
where the real bosonic fields Xi , i = 1, . . . , 8 transform in the eight-dimensional vector representation 8v of the R-symmetry group SO(8), while the components i , i , i = 1, . . . , 8 of the
16-component MajoranaWeyl spinor field transform in the spinor 8s and conjugate spinor 8c
representations of SO(8), respectively. The spinor fields are sections of the twisted spin line bundle ST L over the torus, where L is
areal line bundle over T with flat connection determined
by one of the four spin structures = H 1 (T, Z/2Z) = Z2 /2Z2 and [ST ] Pic0 (T) is chosen to correspond to the theta divisor in the given homology basis (, ). The N = 8 worldsheet
296
supersymmetry of the sigma-model is generated by the fermionic supercurrents

i (z) ,
G (z) = i (z)Xi (z) + (z)X
2
(5.2)
where i are the Spin(8) Dirac matrices.

In the corresponding permutation orbifold, the monodromy conditions on the bosonic fields
X in a given twisted sector (P , Q) are as in (2.1), while the fermion monodromy is given by
a (z + 1) = (1) P (a) (z)
and
a (z + ) = (1) Q(a) (z),
(5.3)
where for simplicity we have omitted a potential extra sign depending on the reference spin
structure [ST ]. This symmetry is compatible with N = 8 worldsheet superconformal invariance
[45], and it means that on the fermionic fields the twist group G is extended to G (Z2 )N . The
consistency condition P Q = QP implies [27] that the spin structure phases in (5.3) are independent of the coordinate label a in the permutation orbifold, and hence that only the diagonal
subgroup of (Z2 )N acts non-trivially on the fermions. The asymmetry between the twistings of
bosons and fermions implies that the modular invariant sum over monodromy homomorphisms
breaks spacetime supersymmetry of the orbifold sigma model.
Generally, the sum over (Z2 )N monodromy in the fermionic sector is weighted by a consistent
set of GSO phases [; ], generically dependent upon the twisted sector : 1 () G, which
are constrained by modular covariancerequirements.
In the untwisted sector () = e, the phase

corresponding to a spin structure = Z2g /2Z2g is the mod 2 index of the Dirac operator
on twisted by the flat line bundle L given by [46]
[; e] = (1)dim H
0 (,S L )
= (1) ,
(5.4)
where
L ) is the number of linearly independent holomorphic sections of the
spin bundle S L . Schematically then, the modification of the formula (2.3) for the partition
function of the supersymmetric permutation orbifold is given by

1
G(Z2 )N
,
( ) = N
[; ]
Z
Z
(5.5)
2 |G|
1
dim H 0 (, S
:1 ()G H (,Z/2Z)
O ()
where Z ( ) is the partition function of the parent superconformal field theory computed with
the global fermionic monodromy determined by the spin structure .
For example, the partition function of the supersymmetric R8 (SN (Z2 )N ) permutation
orbifold on = T can be determined by first calculating the contribution from a given spin
structure (say the RamondRamond sector) to the path integral over the complex fermionic fields,
and then summing over the modular orbits using either of the two GSO projections of type II
string theory. Then the parent partition function appearing in the formula (2.3) is given by [27]
2

1
Z( ) = z 00 ( )4 z 01 ( )4 z 10 ( )4 z 11 ( )4 ,
2
where the +/ sign corresponds to the type IIA/B string amplitude and
2 4
i
4
2
z()( ) =
e 12 (2 1) ei /4
2

1 (1) ei (2n1+ ) 1 (1) ei (2n1 ) .
n=1
(5.6)
(5.7)
297
The corresponding grand canonical partition function (2.12) matches the type II DLCQ free
energy at finite temperature, with (5.6) producing the action of the (restricted) Hecke operator on
the partition function of the first quantized GreenSchwarz superstring [6,11,12]. A completely
analogous correspondence holds for the thermal partition function of type IIB DLCQ superstrings
on the maximally supersymmetric plane wave background in ten dimensions [47].
The operators which create local monodromy in the superconformal sigma model with respect
to the action of SN are products P (z)SP (z) of bosonic and fermionic twist fields. Let us work in
the sector of trivialglobal
Z2 monodromy for the spinor fields, i.e., with the RamondRamond

spin structure = 00 . The other sectors are treated similarly as in [7]. In a Zn -twisted sector
corresponding to a cyclic permutation P = (n), the vacuum state then carries an irreducible
representation of the Clifford algebra for Spin(8). By using an SO(8) triality isomorphism, the
representation space can be taken to be the direct sum 8v 8c . The corresponding components of
i (z)
the 16-dimensional ground state vector are created respectively by the primary spin fields S(n)
and S i (z), i = 1, . . . , 8. They each have conformal dimension [19]
(n)
n
1
(5.8)
+ .
6 3n
To describe the supersymmetric version of the DVV interaction vertex [4], we need another
kind of spin twist field to ensure that the operators generating the basic joining and splitting of
superstrings yield an irrelevant deformation of the superconformal sigma model. The bosonic
twist field ab (z) transposing the fields X a and X b has conformal dimension 1/2 when d = 8
(see (2.32)), as does the fermionic twist field Sab (z) interchanging a and b . To increase the
scaling dimension by 1/2 in a supersymmetric fashion, we use the supersymmetric descendent
given by
of the primary twist field operators (z)S(z)

Q , (z)S (z) + (z)S (z), Q = !i (z)S i (z) =: (z) ,
(5.9)
RR
(n) =
where
Q =
dz
G (z)
2i
(5.10)
are the N = 8 supercharges and the contour integral is taken around the origin z = 0. The dei (z) create the first excited states in the twisted sector [P ].
scendent bosonic twist fields ![P
]
a
b
Since the combination has Ramond boundary conditions under transposition in SN ,
i (z)
the corresponding spin field carries a representation of the Clifford algebra. The twist field Sab
transforms as a vector of SO(8), and it coincides with the standard spin field of the supersymmetric R8 Z2 permutation orbifold which can be constructed explicitly via bosonization of the
fermion fields i [48,49].
The fermionic DVV vertex operator is now defined by

N
Vferm =
(5.11)
d(z)
ab (z).
vol(T)
T
1a<bN
The descendent twist field ab (z) is a primary field of conformal weight 3/2. The interaction
vertex (5.11) is spacetime supersymmetric, SO(8) invariant and describes elementary string interactions [4].
N
The computation of the local twist field correlations functions a1 b1 (z)a2 b2 (0) SN (Z2 ) requires a modification of the covering surface principle of Section 2.4. This is because one should
298
no longer simply close the punctures on the covering space corresponding to the branch points
to get the identity state at those points. Rather, one must insert the operator that creates a Ramond
vacuum at the insertion points in order to give the fermions the correct local monodromy. Thus
in the supersymmetric orbifold theory one uses the same covering spaces as in the case of
the bosonic orbifold, but instead of computing the partition function on one computes a correlation function of spin fields on .6 Schematically, the modification of a generic, normalized
bosonic twist field correlation function (2.36) is given by
G(Z2 )N
k

[Pi ] (wi )
i=1
1
= N
2 |G|

:1 (w )G(Z2 )N
1
N
Z G(Z2 ) ( )
k

S[Pi ] (w i ) ,w ,
(5.12)
O () i=1
where the global (Z2 )N monodromy acts trivially in the bosonic sector and diagonally in the
fermionic sector as in (5.5). A similar prescription for N = 4 supersymmetric orbifold sigma
models is used in [50]. When = T, this will be provided by the corresponding DLCQ free energy through the required modification of the GSO projection at finite temperature which breaks
supersymmetry by making spacetime fermions antiperiodic around the thermal cycle [12].
Similar considerations also apply to the heterotic sigma model on the torus with target space
R8 , which is defined by the action

1
1
i (z) + i (z)
i (z) + A (z)A (z) ,
I (X, , ) =
(5.13)
Xi (z)X
d2 z

4
2i2
T
where the MajoranaWeyl fermion fields A , A = 1, . . . , 32 are SO(8) singlets. The holomorphic
sector of this worldsheet field theory coincides with that of the supersymmetric sigma model
(5.1), while after bosonization of A the antiholomorphic sector coincides with the bosonic sigma
model (2.16) in d = 24 with 16 of the bosons compactified on the Cartan torus of the heterotic
gauge group G = SO(16) SO(16). The heterotic sigma model (5.13) is a superconformal field
theory with chiral (8, 0) worldsheet supersymmetry.
The corresponding (8, 0) supersymmetric permutation orbifold [36,51] is (R8 G) (G
(Z2 )N ). The twist subgroup (Z2 )N acts on the holomorphic sector exactly as in the supersymmetric case. In the antiholomorphic sector, the gauge fermions A are sections of flat real line
bundles L T like i , and so have global fermionic monodromy conditions as in (5.3). In
contrast to the fields i , however, the spin structure phases for A in the permutation orbifold
generally depend on the coordinate label a. Perturbative string interactions are now generated
by the heterotic version of the DVV vertex operator [36,51]. For this, we must explicitly write
worldsheet fields as products of holomorphic and antiholomorphic fields (which was implicitly
understood in all previous formulae). As the holomorphic sector consists of the usual supersymmetric orbifold theory in eight dimensions, the holomorphic part of the vertex is constructed
using the dimension 3/2 spin twist operators ab (z) defined in (5.9). On the other hand, the antiholomorphic sector consists of d = 24 bosons, and since the local monodromies about branch
6 A similar statement is also true in the NSNS sector. In the mixed RNS and NSR sectors, there are no combinations
of a which possess zero modes, so that these sectors have trivial local spin monodromy and the prescription instead
follows that of Section 2.4.
299
points are insensitive to the compactness of the 16 bosons on the Cartan torus, the antiholomorphic part of the vertex is built from the dimension 3/2 bosonic twist fields ab (z) of Section 2.4.
It follows that the heterotic DVV vertex operator is defined by

N
Vhet =
(5.14)
d(z)
( )ab (z).
vol(T)
T
1a<bN
The computation of local twist field two-point functions proceeds by using a formula analogous
to (5.12).
5.2. Supersymmetric DLCQ strings
The genus two DLCQ free energy Fferm ( , ) for type IIA superstrings at finite temperature
is computed in [12]. To write the result, we require
preliminary definitions. The ten even
some

Z/2Z) = Z4 /2Z4 obey a b
reduced, genus two integer characteristics ab = aa12 bb12 H 1 (,
0 mod 2 and are denoted by

0 0
0 0
0 1
0 1
,
2 =
,
3 =
,
4 =
,
1 =
0 0
0 1
0 0
0 1

0 0
0 1
1 0
1 0
5 =
,
6 =
,
7 =
,
8 =
,
1 0
1 0
0 0
0 1

1 0
1 1
9 =
(5.15)
and 0 =
.
1 0
1 1
(2)
We use the shorthand notation i := ( i )( )4 , where the genus two period matrix =
r,m,s,t ( , # ) is given by (2.51). On the last four characteristics in (5.15) we define genus two
functions 6 ( i )( ) of modular weight six by the formulae
6 ( 7 ) = 2 3 5 + 8 9 0 1 4 6 ,
6 ( 8 ) = 7 9 0 1 4 5 + 2 3 6 ,
6 ( 9 ) = 7 8 0 1 2 6 + 3 4 5 ,
6 ( 0 ) = 7 8 9 + 3 4 6 1 2 5 .
(5.16)
Then one has

g 2 4 N 1
(2)
Fferm
, = s
4 64 2
N
m4
N=2
rm=N
m odd
s,tZ/rZ
t=0
2
d2 #
10 ( )
#
4
(2 )

6 ( 7 )( )( 7 )( )4 + 6 ( 8 )( )( 8 )( )4
2
+ 6 ( 9 )( )( 9 )( )4 + 6 ( 0 )( )( 0 )( )4 .
(5.17)
Note that the fermionic contribution to (5.17) consists of a sum of four terms in the Weierstrass
Poincar reduction. We may identify these terms as resulting from the modular invariant sum
over genus one spin structures, as in (5.6). The free energy (5.17) should now be equated to
N
the translationally invariant correlator Vferm Vferm SN (Z2 ) . As in the bosonic case, one is
then faced with the problem of equating the two continuous parametrizations of the partially
discretized genus two moduli space, one in terms of the elliptic Prym modulus = r # and the
300
other in terms of the branch point location z T. This can again be done explicitly for the degree
two contribution to (5.17), corresponding to double covers of the torus T, and used to compute
local spin twist field correlation functions explicitly in each twisted sector of the R8 (Z2 )3
permutation orbifold.
The N = 2 contribution to (5.17) is given by
2 #

d C(0 ( , # )) 2
gs2 4
ferm
F2
(5.18)
=
,
4 64 2
(2# )4 10 (0 ( , # ))

where
C = 6 ( 7 )7 + 6 ( 8 )8 + 6 ( 9 )9 + 6 ( 0 )0 .
(5.19)
We will begin by simplifying the elliptic function (5.19) using the decomposition (2.54) for
N = 2. We use the notation of Section 3.2 throughout.
To simplify the formulae somewhat,
we momentarily omit the overall factor of 1/2 i # in (3.11) and reinstate it at the end of
the calculation. For reference, let us tabulate the ten reduced even genus two theta constants
according to the spin structures (5.15) as
1
3 3# + 4 4#
3 3# 4 4#
4 3# + 3 4#
4 3# 3 4#
23 3#
23 3#
2# 2
2# 2
21 1#
2i1# 1
(5.20)
Since one has the equalities 5 = 6 , 7 = 8 and 9 = 0 for the given reduction, we immediately find that 6 ( 7 ) = 6 ( 8 ) and 6 ( 9 ) = 6 ( 0 ).
After some elementary manipulations we can bring (5.19) into the form

C = 28 2 4 2# 4 3 4 3# 4# 3 2 4 2 3# 2 4# 2 3 4 3# 4

2

2

3 2 4 2 3# 2 4# 2 + 3# 2 4# 2 3 2 4 2
+ 210 2 8 2# 8 1 8 1# 8 + 211 2 4 3 2 4 2 2# 4 3# 2 4# 2 1 4 3 4 1# 4 3# 4

29 1 4 3 4 1# 4 3# 4 2 4 2# 4 3 4 4 4 3# 4 4# 4 .
(5.21)
We can now proceed as in Section 3.2 by doubling the modulus of the theta functions. In addition
to the identities displayed in (3.20), we will also require the doubling identities
1 (z| )2 (z| ) = 1 (2z|2 )4 (0|2 ),
3 (z| )4 (z| ) = 4 (2z|2 )4 (0|2 )
(5.22)
with z = 1/4. We may then take into account that the theta functions with argument z = 1/4
satisfy 3 = 4 and 1 = 2 , and analogously for i# . The calculation is neither difficult nor
illuminating, and the result is
C=
2 8 4 4 #8 #4 #4
,
( # )8 2 3 4 2 3 4
(5.23)
where we have inserted back the factor (1/2 i # )16 and the bar stands for doubled modulus
as in Section 3.2.
301
Let us now perform a modular S transformation (3.23) on the modulus of both types of theta
functions in (5.23). Then the final result for the numerator of the integrand in (5.35) reads

# 2

C 0 , = 234 16 2 # 24 24 4 2 # 8 4 8 .
(5.24)
Substituting (5.24) along with (3.24) into (5.35), and using the abstruse identity (3.21), we arrive
at the final form for the supersymmetric two-loop DLCQ free energy given by

16g 2 8
F2ferm = 2 s 4 4
( )
4
2 #

d
4 (2 # )2
.

#
4
#
4
4
#
4
4
(2 ) 3 ( ) 4 (2 ) 4 ( ) 3 (2 )
(5.25)
Analogously to the bosonic case of Section 3.5, this integral should be matched to the worldsheet
averaged two-point correlation function of spin twist field operators (z) = 12 (z) in the R8
(Z2 )3 permutation orbifold given by

(Z )3
42
ferm
F2
(5.26)
=
d(z) (z)(0) 2 .
2 (0)
T
We recall that, by modular invariance at genus two, the branch point integration in (5.26)
projects all contributions to the correlation function onto the trivial twist sector (, ) = (0, 0), so
3
2)
that the local integrand that we can read off from (5.26) is 4 213 (z)(0) (Z
0,0 . We substitute
(3.69) with d = 8 and (3.70), and recall that the Prym modulus is given by = 0,0 = 2 # .
The crucial observation is that the bosonic contribution to (5.25) involving the difference of theta
functions is identical to that of the purely bosonic case (3.25), due to the universal dimension
independent contribution of the modular form 10 ( ) to the bosonic genus two partition function
(2.43). We can therefore use the same calculation of the Jacobian |d # /dz|2 that was carried out
in Section 3.5, wherein it was shown that
# 2
d 4 # 4
4 4 4

4 3 2 #
dz 3 4 2
a

a
1 ()4 ( )1 6 b ( 2z | ) b (0| ) 24

= 21
(5.27)

1 (z| )
2
ab (0|)2
for an arbitrary fixed characteristic (a, b) = (1, 1). Using the identity (3.54) and recalling the
definition of the prime form (3.52), after a little algebra we can use (5.25)(5.27) to compute

(Z )3
(z)(0) 0,02

8 ab ( 2z | )3 ab (0| )3 8
6
8
3
,
a
= z E(z) 64 () 4 ()

b (0|)6
where
z( ) =
"

4 2 1 4 ( )
2 |( )|2 ( )
(5.28)
(5.29)
302
is the one-loop, first quantized partition

function of the GreenSchwarz superstring in R evaluated with the genus one spin structure 01 .
We can generate from (5.28) the contribution of a generic twisted sector (, ) (Z/2Z)2
to the spin twist field correlation function by using a crossing transformation z z + +
and the corresponding twisted Prym modulus (3.46), along with the transformation formula for
Jacobi elliptic functions given by

i
i

ab z + + | = exp 2 iz (b + ) a+
(5.30)
b+ z
4
2
which is valid for arbitrary a, b Q and , Q. In fact, the z-dependence of the correlation
function (5.28) is identical to that of Section 3.4 (up to an overall power), and hence the twisted
sector two-point function is an appropriate supersymmetric completion of the bosonic correlation
function (3.50) (with R = and d = 8). The final result is
8 16
(Z )3
(z)(0) ,2 = z c ,
(5.31)
where
c

3
= %
8 (, )3 4 (, )
(5.32)

and the twisted bosonic determinant c is given by (3.51). The cubic power in the supersymmetric twisted determinant (5.32) reflects the fact that the effective twist group of the supersymmetric
permutation orbifold is (Z2 )3 .
5.3. Heterotic DLCQ strings
Finally, we come to the thermodynamic, genus two DLCQ free energy Fhet ( , ) for heterotic strings with heterotic gauge group G = Spin(32)/Z2 or G = E8 E8 . The holomorphic
sector consists of the usual chiral superstring contribution at genus two. In the antiholomorphic
sector, the non-compact bosons produce the usual antichiral bosonic contribution, while the com The latter contribution
pactified bosonic fields produce an instanton sum over the root lattice of G.
yields a theta function of the root lattice which is the unique genus two modular form of weight
eight given by
(2)
8 ( ) =
9
( i )( )16 .
(5.33)
i=0
In the notation of Section 5.2 above, one then has [12]

4 N
1
gs2
(2)

Fhet , =
4

2
8 2048 ( )
N
m4
N=2
rm=N
modd
s,tZ/rZ
t=0
d2 # 8 ( )
(2# )4 |10 ( )|2

6 ( 7 )( )( 7 )( )4 + 6 ( 8 )( )( 8 )( )4

+ 6 ( 9 )( )( 9 )( )4 + 6 ( 0 )( )( 0 )( )4 .
(5.34)
303
Again we deal explicitly only with the contribution of double covers to the formula (5.34),
which is given by
F2het

4 2 #

gs2
d 8 (0 ( , # ))C(0 ( , # ))

=
,
8 2048 4 ( )2
|10 (0 ( , # ))|2
(2# )4
(5.35)
where from (5.24) one has

8 12 12 4 4

C 0 , # = 217 2 # 4 2 # 4 .
(5.36)
To simplify the combination of elliptic functions arising in the genus two modular form (5.33),
we follow the same steps as in the bosonic and supersymmetric calculations. Namely, we expand
the terms in the sum over even genus two spin structures in (5.33) using the table (5.20), transform it to a form that is suitable for doubling the moduli of the Jacobi theta functions, write the
doubling identities, and then make an elliptic S transformation. The final result is again conveniently written in terms of theta functions of moduli 2 # and as

#
16 # 16
3 ( )4 3 (2 # )4
10 8
8 0 ,
(5.37)
= 2 4
4 2
PG
,
,
4 ( )4 4 (2 # )4
where PG (x, y) is the symmetric polynomial defined by

PG (x, y) = 256 x 4 y 4 + 1 512 x 4 y 3 + x 3 y 4 + x + y + 1984 x 3 y 3 + xy

+ 288 x 4 x 2 + x 2 y 4 + x 2 + y 2 2016 x 3 y 2 + x 2 y 3 + x 2 y + xy 2

+ x 4 + y 4 + 604 x 3 y + xy 3 + 3654x 2 y 2 .
(5.38)
Substituting (5.36) and (5.37), along with (3.24) and the abstruse identity (3.21), we find that
the heterotic DLCQ free energy is given by
F2het

gs2 4 ( ) 8 4 ( ) 12

= 2
64 4
( )
4
2 #

d
4 (2 # )2

#
4
#
4
4
#
4
4
(
)
(2
)
(
)
(2
)
(2 ) 3
4
4
3

4 (2 # )
(2 # )
12

3 ( )4 3 (2 # )4
PG
,
,
4 ( )4 4 (2 # )4
(5.39)
where we have used the complex conjugation properties i ( )4 = i ( )4 and ( )12 =

( )12 . We equate (5.39) to the integrated two-point correlation function in the heterotic
(R8 G) (Z2 (Z2 )2 ) permutation orbifold given by
F2het

=
42
2 (0)

Z (Z2 )2
d(z) ( )(z)( )(0) 2
.
(5.40)
Using the identities (3.52), (3.54) and (5.27) we then arrive at the two-point function of twist
fields in the untwisted sector given by
304
Z2 (Z2 )2
( )(z)( )(0) 0,0

8
6
z( ) 8 4 ( ) 12
= 32
E(z) 8 ()3 4 ()
2

( )
4

12

4 ab ( 2z | )3 ab (0| )3 8
4 ( )
3 ( )4 3 ()

PG
,

4
4 ( )4 4 ()
a (0|)6
()
(5.41)
with (a, b) = (1, 1), where z ( ) is the supersymmetric partition function (5.29).
The structure of the formula (5.41) can be understood as follows. Generally, the separating
degeneration limit 12 0 of the genus two modular form (5.33) factorizes into the unique
elliptic modular form of weight eight under SL(2, Z) as

8 ( ) = 2 (11 )16 + 3 (11 )16 + 4 (11 )16 2 (22 )16 + 3 (22 )16 + 4 (22 )16
2
.
+ O 12
(5.42)
in the homology basis wherein the period matrix is given by (3.35)
For the covering surface ,
this degeneration limit corresponds to , or equivalently z 0. Since the x y limit of
the symmetric polynomial (5.38) factorizes as

2
PG (x, x) = 64 (x 1)4 + x 4 + 1 ,
(5.43)
we see that the z 0 limit of the two-point function (5.41) factors
into the one-loop heterotic
string partition function on R8 evaluated with the spin structure 01 which is given by
2 4

1
4
4 ( ) 4 2 ( )16 + 3 ( )16 + 4 ( )16
.
zhet ( ) =
(5.44)
2
|( )|16 ( )
2( )16
However, in contrast to the bosonic and supersymmetric twist field correlation functions, for
distinct branch points the two-point function (5.41) does not neatly factor out a component corresponding to the untwisted fluctuation determinant of the heterotic orbifold sigma model. The
reason generally is that the effective twist group is now a semi-direct product SN (Z2 )N acting
on the gauge fermions a . This means that the discrete (Z2 )N gauge symmetry acts in the gauge
sector together with the monodromy conditions of the permutation orbifold, and a disentanglement of the twisted and untwisted determinants arising from integration over the fermion fields
in terms of branch point data as previously is not possible.
For example, by applying a crossing transformation to (5.41) as before one arrives at the
twisted sector two-point functions

Z2 (Z2 )2
z( ) 8 16 4 ( ) 12
( )(z)( )(0) ,
c
= 32
2( )
4 2
12

4 ( , )
3 ( )4 3 ( , )4
PG
,
4 ( )4 4 ( , )4
2( , )
(5.45)

with the supersymmetric twisted determinant c given by (5.32). The extra gauge symmetry
is implemented by O(N) vector reflections of a and holonomies of the corresponding flat real
line bundles L T. The latter phases correspond to Z2 -valued Wilson lines which break the
spacetime heterotic gauge group G to G = SO(16) SO(16). They yield the extra GSO projection
required to match to the spectrum of the free E8 E8 heterotic string [36,51,52] and to light-cone
heterotic string field theory.
305
Acknowledgements
We thank R. Accola, P. Bntay, H. Braden, J.C. Eilbeck, H. Farkas, J. Howie, A. Konechny,
S. Ramgoolam, A. Recknagel, R. Reis and G. Semenoff for helpful discussions and correspondence. This work was supported in part by the Marie Curie Research Training Network Grants
ForcesUniverse (contract No. MRTN-CT-2004-005104) and ENRAGE (contract No. MRTN-CT2004-005616) from the European Communitys Sixth Framework Programme. The work of H.C.
was supported in part by a Postgraduate Studentship from the Engineering and Physical Sciences
Research Council (UK).
References
[1] W. Taylor, Matrix theory: Matrix quantum mechanics as a fundamental theory, Rev. Mod. Phys. 73 (2001) 419462,
hep-th/0101126.
[2] L. Motl, Proposals on non-perturbative superstring interactions, hep-th/9701025.
[3] T. Banks, N. Seiberg, Strings from matrices, Nucl. Phys. B 497 (1997) 4155, hep-th/9702187.
[4] R. Dijkgraaf, E.P. Verlinde, H.L. Verlinde, Matrix string theory, Nucl. Phys. B 500 (1997) 4361, hep-th/9703030.
[5] R. Dijkgraaf, G.W. Moore, E.P. Verlinde, H.L. Verlinde, Elliptic genera of symmetric products and second quantized
strings, Commun. Math. Phys. 185 (1997) 197209, hep-th/9608096.
[6] G. Grignani, G.W. Semenoff, Thermodynamic partition function of matrix superstrings, Nucl. Phys. B 561 (1999)
243272, hep-th/9903246.
[7] R. Dijkgraaf, L. Motl, Matrix string theory, contact terms and superstring field theory, hep-th/0309238.
[8] S. Moriyama, Comments on supersymmetry algebra and contact term in matrix string theory, JHEP 0409 (2004)
013, hep-th/0405091.
[9] I. Kishimoto, S. Moriyama, S. Teraguchi, Twist field as three string interaction vertex in light cone string field
theory, Nucl. Phys. B 744 (2006) 221237, hep-th/0603068.
[10] I. Kishimoto, S. Moriyama, On LCSFT/MST correspondence, hep-th/0611113.
[11] G. Grignani, P. Orland, L.D. Paniak, G.W. Semenoff, Matrix theory interpretation of DLCQ string worldsheets,
Phys. Rev. Lett. 85 (2000) 33433346, hep-th/0004194.
[12] H.C.D. Cove, R.J. Szabo, Two-loop string theory on null compactifications, Nucl. Phys. B 741 (2006) 313352,
hep-th/0601220.
[13] L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, The conformal field theory of orbifolds, Nucl. Phys. B 282
(1987) 1373.
[14] D. Bernard, Z2 -twisted fields and bosonization on Riemann surfaces, Nucl. Phys. B 302 (1988) 251279.
[15] H. Saleur, Correlation functions of the critical AshkinTeller model on a torus, J. Stat. Phys. 50 (1988) 475508.
[16] J.J. Atick, L.J. Dixon, P.A. Griffin, D. Nemeschansky, Multi-loop twist field correlation functions for Zn orbifolds,
Nucl. Phys. B 298 (1988) 135.
[17] A. Lawrence, A. Sever, Scattering of twist fields from D-branes and orientifolds, arXiv: 0706.3199 [hep-th].
[18] G.E. Arutyunov, S.A. Frolov, Virasoro amplitude from the SymN (R24 ) orbifold sigma model, Theor. Math.
Phys. 114 (1998) 4366, hep-th/9708129.
[19] G.E. Arutyunov, S.A. Frolov, Four graviton scattering amplitude from SymN (R8 ) supersymmetric orbifold sigma
model, Nucl. Phys. B 524 (1998) 159206, hep-th/9712061.
[20] S. Hamidi, C. Vafa, Interactions on orbifolds, Nucl. Phys. B 279 (1987) 465513.
[21] R. Dijkgraaf, E.P. Verlinde, H.L. Verlinde, c = 1 conformal field theories on Riemann surfaces, Commun. Math.
Phys. 115 (1988) 649690.
[22] O. Lunin, S.D. Mathur, Correlation functions for M N /SN orbifolds, Commun. Math. Phys. 219 (2001) 399442,
hep-th/0006196.
[23] P. Bntay, Characters and modular properties of permutation orbifolds, Phys. Lett. B 419 (1998) 175178, hepth/9708120.
[24] P. Bntay, Orbifoldization, covering surfaces and uniformization theory, Lett. Math. Phys. 57 (2001) 15, hepth/9808023.
[25] P. Bntay, Symmetric products, permutation orbifolds and discrete torsion, Lett. Math. Phys. 63 (2003) 209218,
hep-th/0004025.
[26] H. Fuji, Y. Matsuo, Open string on symmetric product, Int. J. Mod. Phys. A 16 (2001) 557608, hep-th/0005111.
306
[27] H. Fuji, Open superstring on symmetric product, hep-th/0112116.

[28] P.H. Ginsparg, Applied conformal field theory, in: E. Brzin, J. Zinn-Justin (Eds.), Fields, Strings, Critical Phenomena, North-Holland, 1990, pp. 1168, hep-th/9108028.
[29] A. Klemm, M.G. Schmidt, Orbifolds by cyclic permutations of tensor product conformal field theories, Phys. Lett.
B 245 (1990) 5358.
[30] P. Forgcs, Z. Horvth, L. Palla, P. Vecsernys, Higher level KacMoody representations and rank reduction in
string models, Nucl. Phys. B 308 (1988) 477508.
[31] M.B. Halpern, The orbifolds of permutation-type as physical string systems at multiples of c = 26. I: Extended
actions and new twisted worldsheet gravities, JHEP 0706 (2007) 068, hep-th/0703044.
[32] L. Borisov, M.B. Halpern, C. Schweigert, Systematic approach to cyclic orbifolds, Int. J. Mod. Phys. A 13 (1998)
125168, hep-th/9701061.
[33] Z. Kdr, The torus and the Klein bottle amplitude of permutation orbifolds, Phys. Lett. B 484 (2000) 289294,
hep-th/0004122.
[34] P. Bntay, Permutation orbifolds, Nucl. Phys. B 633 (2002) 365378, hep-th/9910079.
[35] C.L. Ezell, Branch point structure of covering maps onto nonorientable surfaces, Trans. Am. Math. Soc. 243 (1978)
123133.
[36] S.-J. Rey, Heterotic matrix strings and their interactions, Nucl. Phys. B 502 (1997) 170190, hep-th/9704158.
[37] A.A. Belavin, V.G. Knizhnik, A. Morozov, A.M. Perelomov, Two and three loop amplitudes in the bosonic string
theory, Phys. Lett. B 177 (1986) 324328.
[38] G.W. Moore, Modular forms and two-loop string physics, Phys. Lett. B 176 (1986) 369379.
[39] C. Birkenhake, H. Lange, Complex Abelian Varieties, Springer, 2004.
[40] J.D. Fay, Theta Functions on Riemann Surfaces, Springer, 1973.
& C)
[41] J.M. Smyrnakis, Representations of the mapping class group of the two-punctured torus on the space of sl(2,
spin 12 spin 12 KacMoody blocks, Nucl. Phys. B 496 (1997) 630642, hep-th/9611225.
[42] H.M. Farkas, I. Kra, Theta Constants, Riemann Surfaces and the Modular Group, American Mathematical Society,
2001.
[43] E. Kani, Hurwitz spaces of genus 2 covers of elliptic curves, Collect. Math. 54 (2003) 151.
[44] G. Mason, M.P. Tuite, On genus two Riemann surfaces formed from sewn tori, Commun. Math. Phys. 270 (2007)
587634, math.QA/0603088.
[45] L.J. Dixon, P.H. Ginsparg, J.A. Harvey, c = 1 superconformal field theory, Nucl. Phys. B 306 (1988) 470496.
[46] D. Mumford, Tata Lectures on Theta, Birkhuser, 1983.
[47] Y. Sugawara, Thermal amplitudes in DLCQ superstrings on pp-waves, Nucl. Phys. B 650 (2003) 75113, hepth/0209145.
[48] D. Friedan, Z. Qiu, S.H. Shenker, Superconformal invariance in two dimensions and the tricritical Ising model,
Phys. Lett. B 151 (1985) 3743.
[49] D. Friedan, E.J. Martinec, S.H. Shenker, Conformal invariance, supersymmetry and string theory, Nucl. Phys. B 271
(1986) 93165.
[50] O. Lunin, S.D. Mathur, Three-point functions for M N /SN orbifolds with N = 4 supersymmetry, Commun. Math.
Phys. 227 (2002) 385419, hep-th/0103169.
[51] D.A. Lowe, Heterotic matrix string theory, Phys. Lett. B 403 (1997) 243249, hep-th/9704041.
[52] T. Banks, L. Motl, Heterotic strings from matrices, JHEP 9712 (1997) 004, hep-th/9703218.

Implications of canonical gauge coupling unification

in high-scale supersymmetry breaking
V. Barger a , N.G. Deshpande b , Jing Jiang b, ,
Paul Langacker c , Tianjun Li d,e
a Department of Physics, University of Wisconsin, Madison, WI 53706, USA
b Institute of Theoretical Science, University of Oregon, Eugene, OR 97403, USA
c School of Natural Sciences, Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540, USA
d George P. and Cynthia W. Mitchell Institute for Fundamental Physics, Texas A&M University,
College Station, TX 77843, USA

e Institute of Theoretical Physics, Chinese Academy of Sciences, Beijing 100080, PR China
Received 11 April 2007; received in revised form 9 July 2007; accepted 5 October 2007
Abstract
We systematically construct two kinds of models with canonical gauge coupling unification and universal
high-scale supersymmetry breaking. In the first we introduce standard vector-like particles while in the
second we also include non-standard vector-like particles. We require that the gauge coupling unification
scale is from 5 1015 GeV to the Planck scale, that the universal supersymmetry breaking scale is from
10 TeV to the unification scale, and that the masses of the vector-like particles (MV ) are universal and
in the range from 200 GeV to 1 TeV. Using two-loop renormalization group equation (RGE) running for
the gauge couplings and one-loop RGE running for Yukawa couplings and the Higgs quartic coupling, we
calculate the supersymmetry breaking scales, the gauge coupling unification scales, and the corresponding
Higgs mass ranges. When the vector-like particle masses are less than 1 TeV, these models can be tested at
the LHC.
PACS: 11.25.Mj; 12.10.Kt; 12.10.-g
Keywords: Unification; Supersymmetry; Vector-like particle
E-mail address: jing@uoregon.edu (J. Jiang).

doi:10.1016/j.nuclphysb.2007.10.013
308
V. Barger et al. / Nuclear Physics B 793 (2008) 307325
1. Introduction
There is no known symmetry in effective field theory or string theory that can constrain the
cosmological constant CC to be zero. Why the cosmological constant is so tiny compared to
4 ) is a great mystery in particle
the Planck scale MPl or string scale MString (CC 10122 MPl
physics and cosmology. In addition, because the Standard Model (SM) Higgs boson mass is
not stable against quantum corrections, the weak scale, which is about 16 (15) order smaller
than MPl (MString ), presents another puzzle. These are the cosmological constant problem and
gauge hierarchy problem, respectively. Supersymmetry can solve the gauge hierarchy problem
elegantly; however, it can ameliorate but cannot solve the cosmological constant problem.
Because there exists an enormous landscape for long-lived metastable vacua in the type II
string theories with flux compactifications where the moduli can be stabilized and supersymmetry
may be broken [1], we may explain the tiny value of the cosmological constant by the weak
anthropic principle [2], and solve the gauge hierarchy problem simultaneously without invoking
weak scale supersymmetry [3]. Although the strong CP problem is still a big challenge in the
string landscape [4], it can be solved by the well known PecceiQuinn mechanism [5]. The axion
solutions can be stabilized by the gauged discrete ZN PecceiQuinn symmetry [6,7] arising from
the breaking of an anomalous U (1)A gauge symmetry in string constructions [8,9]. The axion
can also be a cold dark matter candidate [7].
One consequence of the string landscape is that supersymmetry can be broken at a high scale
if we have many supersymmetry breaking parameters or many hidden sectors [10,11]. Because
the string landscape is mainly based on type II orientifolds with flux compactifications, the su2
persymmetry breaking soft masses and trilinear A terms are generically about MString
/MPl , at
least in the known models [12,13]. We shall assume universal supersymmetry breaking in this
paper.
Supposing that the cosmological constant and gauge hierarchy problems are indeed solved in
the string landscape, what would the guiding principle for our model building and new physics
search be? In this paper, we consider canonical gauge coupling unification as our main guiding
principle to study new physics in the extensions of the SM, which would be expected in Grand
Unified Theories (GUTs). Achieving the SM gauge coupling unification for high-scale supersymmetry breaking is an interesting question. It is well known that gauge coupling unification cannot
be achieved in the SM with the canonical normalization of the U (1)Y hypercharge interaction,
i.e., the GeorgiGlashow SU(5) normalization [14], unless we introduce additional vector-like
particles at the weak scale [1518]. However, it can indeed be realized at about 101617 GeV for
non-canonical U (1)Y normalizations [19].
In this paper we systematically construct the models with canonical gauge coupling unification and universal high-scale supersymmetry breaking by introducing extra SM vector-like
fermions at the weak scale.1 To avoid the dimension-6 proton decay problem and quantum
gravity effects, we require that the gauge coupling unification scale (MU ) is in the range from
5 1015 GeV to the Planck scale. We also assume that the supersymmetry breaking scale (MS )
can be from 10 TeV to the unification scale. The masses of the vector-like fermions (MV ) could
in principle be arbitrary. However, we restrict our attention to the case of a universal MV in
the range from 200 GeV to 1 TeV. This is motivated by simplicity and because such particles
would be observable at the LHC. Furthermore, in some models there are additional symmetries
1 We do not consider new particles which are chiral with respect to the SM gauge group because of the precision
electroweak constraints [20]. They could, however, be chiral with respect to additional gauge symmetries.
309
which require MV to be generated by the vacuum expectation value of a Standard Model singlet field which is tied to the electroweak scale [21]. To have such gauge coupling unification,
we show that the total contributions to the one-loop beta function of SU(2)L (b2 ) from the
vector-like fermions must be equal to those of SU(3)C (b3 ), i.e., b2 = b3 , and we also obtain the constraint 2/5 b2 b1 13/5, where b1 is the total contribution to the one-loop
beta function of U (1)Y . There are only finite possibilities for b2 b1 due to the quantization of bi . To systematically study gauge coupling unification with high-scale supersymmetry
breaking, we employ the one-loop beta function equivalent relations among the particle sets,
which was originally proposed in Ref. [22]. If the gauge coupling unification can be achieved
in a model with a set of vector-like fermions which have b2 = b3 and b2 b1 = cb , all
the models with gauge coupling unification and the vector-like fermions which have b2 = b3
and b2 b1 = cb can be constructed by adding particles such that the one-loop beta function
equivalent relations hold for the additional particle sets.
We consider two kinds of models. For the first kind, we introduce the standard vector-like
particles whose quantum numbers are identical to those of the SM fermions and their Hermitian
conjugates, the particles in the SU(5) symmetric representation and their Hermitian conjugates,
and the SU(5) adjoint particles. For the second kind, we introduce non-standard vector-like particles which are charged under the SU(3)C SU(2)L and neutral under U (1)Y . These particles
can arise from string constructions [23,24].2 After identifying viable models, we use two-loop
renormalization group equation (RGE) running for the SM gauge couplings and one-loop RGE
running for the Yukawa couplings and the Higgs quartic coupling to calculate the supersymmetry
breaking scales, gauge coupling unification scales, and the corresponding Higgs mass ranges for
the models with simple sets of extra vector-like fermions for MV = 200 GeV and 1 TeV. In the
first kind of models, b2 b1 can only be equal to 6/5 and 12/5, and then the corresponding
supersymmetry breaking scale can only be around 1010 and 1015 GeV, respectively. In the second kind, b2 b1 can be n/5 with n = 2, 3, . . . , 13, and the supersymmetry breaking scale
can be from 105 to 1016 GeV if we include uncertainties from the threshold corrections at the
scales MV , MS and MU . The masses of the vector-like fermions are within the reach of the Large
Hadron Collider (LHC).
We briefly discuss the phenomenological consequences of the models, which will be presented
in detail elsewhere.
This paper is organized as follows: in Section 2, we present our calculation procedure. We
consider canonical gauge coupling unification and the Higgs mass ranges in the models with standard vector-like particles and non-standard vector-like particles in Sections 3 and 4, respectively.
In Section 5, we comment on phenomenological consequences. Our discussions and conclusions
are in Section 6. The renormalization group equations are given in Appendix A, and the two-loop
beta functions for the additional vector-like particles are given in Appendix B.
2. Calculation procedure
We consider models with canonical gauge coupling unification where the supersymmetry
breaking scale is from 10 TeV to the unification scale. In this range the constraints on the electric
dipole moments (EDMs) of the electron and neutron due to the generic CP violations in the supersymmetry breaking soft terms can automatically be satisfied. The cosmological constant problem
2 In some cases these models imply fractional electric charges, and would be allowed only for non-standard cosmologies.
310
and gauge hierarchy problem are assumed to be solved by the string landscape. We assume that
the strong CP problem can be solved by the PecceiQuinn mechanism. The axion can be a cold
dark matter candidate. The additional vector-like fermions could also provide possible cold dark
matter candidates. Similar to the new minimal SM [25], the neutrino masses and mixings can be
explained by the see-saw mechanism by introducing two or three right-handed neutrinos [26],
and the baryon asymmetry can be generated by leptogenesis [27] or other mechanisms.
In supersymmetric models there generically exist one pair of Higgs doublets Hu and Hd .
We define the SM Higgs doublet H , which is fine-tuned to have a weak-scale mass, as
H cos i2 Hd + sin Hu , where 2 is the second Pauli matrix and tan is a mixing
parameter [7,11]. Inspired by the supersymmetry breaking on type II orientifolds with flux
compactifications [12,13], we assume universal supersymmetry breaking at scale MS , i.e., the
gauginos, squarks, sleptons, Higgsinos, and the other combination of the scalar Higgs doublets
(sin i2 Hd + cos Hu ) have a universal supersymmetry breaking soft mass around MS .
We require that the gauge coupling unification scale is higher than 5 1015 GeV so that the
dimension-6 proton decay via exchange of the X and Y gauge bosons can be suppressed, and that
the scale is smaller than the Planck scale (2.4 1018 GeV) so that quantum gravity effects can be
neglected.3 To achieve canonical gauge coupling unification, we introduce vector-like fermions,
and for simplicity we assume that their masses (MV ) are universal and from 200 GeV to 1 TeV
so that they can be observed at the LHC. Our analysis can be easily extended to the cases where
MV takes either non-universal or higher values. The superpartners of these vector-like fermions
(scalar components in the supermultiplets) are assumed to have supersymmetry breaking soft
masses around MS . If MS MU , the canonical gauge coupling unification is realized in the SM
through the introduction of the vector-like particles.
The one-loop bi relevant from MS to MU are given in the following sections. From MV
to MS , the one-loop beta functions bi from the vector-like fermions should be 2/3 of those
for the complete supermultiplets. The renormalization group equations in the SM and the Minimal Supersymmetric Standard Model (MSSM) can be found in Appendix A. The two-loop beta
functions from these extra vector-like fields are given in Appendix B. We consider two-loop
RGE running for the SM gauge couplings and one-loop running for the Yukawa couplings and
the Higgs quartic coupling. For simplicity, we only consider the contributions to the gauge coupling RGE running from the Yukawa couplings of the third family of the SM fermions, i.e., the
top quark, bottom quark and lepton Yukawa couplings. We do not consider the contributions to
the gauge coupling RGE running from the Yukawa couplings of the extra vector-like particles.
We denote the gaugecouplings for U (1)Y , SU(2)L , and SU(3)C as gY , g2 , and g3 , respectively, and define g1 5/3gY . The major prediction in the models with high-scale supersymmetry breaking is the Higgs boson mass [7,19,29]. We can calculate the Higgs boson quartic
coupling at the supersymmetry breaking scale MS
(MS ) =
g22 (MS ) + 3g12 (MS )/5

cos2 2,
4
(1)
and then evolve it down to around the weak scale. The renormalization group equation for the
Higgs quartic coupling is also given in Appendix A. Using the one-loop effective Higgs potential with top quark radiative corrections, we calculate the Higgs boson mass by minimizing the
3 Unification at the string scale ( 5 1017 GeV) for weakly coupled heterotic string theory [23] is considered in [28].
effective potential
Veff = m2h H H

2
3 4 2
h2t (H H ) 3
h H H
,
H H
log
2!
2
16 2 t
Q2
311
(2)
where m2h is the bare Higgs mass squared, ht is the top quark Yukawa coupling, is evaluated at
the scale Q, and Q is chosen to be at the Higgs boson mass [30]. Since the physical Higgs boson
mass is an unknown at the beginning, we solve for it through iterations. We use the one-loop
corrected MS top quark Yukawa coupling [31], which is related to the top quark pole mass by

16 g32
h2t
mt = ht v 1 +
(3)
2
.
3 16 2
16 2
We define i = gi2 /4 and denote the Z boson mass as MZ . In the following numerical
calculations, we use top quark pole mass mt = 171.4 2.1 GeV [32,33], the strong coupling
constant 3 (MZ ) = 0.1189 0.0010 [34]. The fine structure constant EM , weak mixing angle
W and Higgs vacuum expectation value (VEV) v at MZ are [20]
1
(MZ ) = 127.904 0.019,
EM
sin2 W (MZ ) = 0.23122 0.00015,

v = 174.10 GeV.
(4)
3. Models with standard vector-like particles

To achieve canonical gauge coupling unification, we first introduce the vector-like particles
whose quantum numbers are the same as those of the SM fermions and their Hermitian conjugates, particles in the SU(5) symmetric representation and their Hermitian conjugates, and the
SU(5) adjoint particles. Their quantum numbers under SU(3)C SU(2)L U (1)Y and their
contributions to one-loop beta functions, b (b1 , b2 , b3 ) as complete supermultiplets
are

1
1
1
XQ + XQ = 3, 2,
(5)
+ 3, 2, ,
b =
, 3, 2 ,
6
6
5

2
8
1, 2 ,
XU + XU = 3, 1,
(6)
+ 3,
b =
, 0, 1 ,
3
3
5

1
2
1, 1 ,
+ 3,
b =
, 0, 1 ,
XD + XD = 3, 1,
(7)
3
3
5

1
1
3
XL + XL = 1, 2,
(8)
+ 1, 2, ,
b =
, 1, 0 ,
2
2
5

6
XE + XE = (1, 1, 1) + (1, 1, 1),
(9)
b =
, 0, 0 ,
5
XG = (8, 1, 0),
b = (0, 0, 3),
(10)
XW = (1, 3, 0),
b = (0, 2, 0),

18
b =
, 4, 0 ,
XT + XT = (1, 3, 1) + (1, 3, 1),
5

2
2
16
XS + XS = 6, 1,
+ 6, 1,
,
b =
, 0, 5 ,
3
3
5
(11)
(12)
(13)
312

5
5
XY + XY = 3, 2,
+ 3, 2,
,
6
6
b = (5, 3, 2).
(14)
There are three mass scales in our models: the universal mass for the vector-like particles MV ,
the supersymmetry breaking scale MS , and the gauge coupling unification scale MU . The viable
values of b for our choices of scales: 200 GeV MV 1 TeV, 10 TeV MS MU and
5.0 1015 GeV < MU < 2.4 1018 GeV, are limited. At one-loop level, only the relative differences between the beta functions are relevant so that (b1 , b2 , b3 ) is essentially equivalent
to (0, b2 b1 , b3 b1 ), i.e., increasing or decreasing b1 , b2 , and b3 by the same
amount does not affect these mass scales, but they do increase or decrease the strength of the
unified gauge couplings, respectively. As long as we keep b1 less than around 10, the gauge
couplings at the unification scale will remain perturbative.
Let us first study the possible values for b3 b2 . The choices of b3 b2 1 and
b3 b2 1 respectively produce too small and too large values for the SU(2)L SU(3)C
gauge coupling unification scale MU . Assuming MS = MU and the SM gauge couplings at
the weak scale, we show the one-loop SU(2)L SU(3)C unification scale MU for the cases
(b2 = 1, b3 = 0) and (b2 = 1, b3 = 2) in the left plot of Fig. 1. For (b2 = 1,
b3 = 0), MU is smaller than 8 1014 GeV, which is the maximal unification scale for the
case b3 b2 1. For (b2 = 1, b3 = 2), MU is larger than 1020 GeV, which is the minimal unification scale for b3 b2 1. Because the two-loop RGE running can only change the
unification scale by a factor less than 5 for the models we have studied in this paper and b2 and
b3 only take integer values, we obtain4 that b3 = b2 . We also observe that gauge coupling
Fig. 1. Left: The intersections of the upper solid line with the dashed line, solid line and the dotted line show the one-loop
SU(3)C SU(2)L gauge coupling unification scale for (b2 = 1, b3 = 0), (b2 = 1, b3 = 1), and (b2 = 1,
b3 = 2), respectively. Right: Mass scales MS (solid) and MU (dotted) as functions of b2 b1 from one-loop RGE
running. The upper curves correspond to MV = 200 GeV and the lower curves MV = 1 TeV.
4 The argument becomes even stronger for M < M , with M becoming even smaller or larger for b b = 1,
S
U
U
3
2
respectively. On the other hand the argument would be weakened if we allowed MV much larger than 1 TeV, i.e., in that
case b3 = b2 would be allowed.
313
unification including a canonically normalized U (1)Y requires 2/5 b2 b1 13/5 in the

models with high-scale supersymmetry breaking. For b2 b1 > 3, MS is larger than MU ,
and 1/5, 14/5 or 3 cannot be generated from the given particle sets.
In the right plot of Fig. 1 we show the dependence of MS and MU on b2 b1 , based on
one-loop RGE running for the SM gauge couplings. In two-loop RGE running, the actual values
of bs will shift MS and MU away from those indicated by the curves. Curves for both MS and
MU are plotted for MV = 200 GeV and MV = 1 TeV. However, for MU the two dotted curves
are too close to each other to be discerned. The solid curves are for MS , with the upper one for
MV = 200 GeV and the lower for MV = 1 TeV. As we increase b2 b1 , the increase in MU
is gradual, but the increase in MS is very rapid.
Using the constraints on b1 , b2 , and b3 , we are ready to generate the complete sets of
vector-like particles that will ensure canonical gauge coupling unification. Because b3 = b2
and 2/5 b2 b1 13/5, there are only finite possibilities for b2 b1 due to the quantization of bi . We employ the equivalent relations of the one-loop beta function for the particle
sets [22]. If we can achieve canonical gauge coupling unification by introducing one set of the
vector-like fermions with b3 = b2 and b2 b1 = cb , it also holds for one-loop equivalent
sets, defined as those with the same b3 b2 and b2 b1 at one loop, because the two-loop
effects give only small corrections. The complete independent one-loop beta function equivalent
relations for the particle sets are [22]
or XQ + XU + XE 0,
EQV1:
XQ + XU + XE 0,
EQV2:
XD + XL 0,
EQV3:
XW + XG + XY + XY 0,
EQV4:
XQ + XT + XS 0,
EQV5:
2(XD + XD) + XE + XE + XW 0,
(19)
EQV6:
XL + XL + 2(XE + XE) + XW + XG 0,
(20)
EQV7:
XD + XE XU,
EQV8:
3(XE + XE) + 2XW XT + XT ,
EQV9:
XU + XE 2XD + XY,
XD + XL 0,
or
or
(15)
(16)
(17)
XQ + XT + XS 0,
or XD + XE XU ,
or
XU + XE 2XD + XY ,
(18)
(21)
(22)
(23)
where 0 means the zero particle set. Equivalent relations (15)(18) correspond respectively to
10, 5, 24, and 15-plets of SU(5).
The conditions b2 = b3 and 2/5 b2 b3 13/5 for canonical gauge coupling unification are satisfied by the simple sets

4
6 6
b =
, 2, 2 0, ,
,
Z1: XW + 2(XD + XD),
(24)
5
5 5

9
6 6
b =
, 3, 3 0, ,
,
Z2: XW + 3(XD + XD) + (XL + XL),
(25)
5
5 5

9
6 6
b =
, 3, 3 0, ,
,
Z3: XQ + XQ + XU + XU ,
(26)
5
5 5

9
6 6
b =
, 3, 3 0, ,
, (27)
Z4: XQ + XQ + XD + XD + XE + XE,
5
5 5

9
6 6
b =
, 3, 3 0, ,
,
Z5: XG + 3(XL + XL),
(28)
5
5 5
314

Z6:
XG + XW + XL + XL + XE + XE,
b =

9
6 6
, 3, 3 0, ,
,
5
5 5
(29)
XG + XW + XD + XD + 2(XL + XL) + XE + XE,

6 6
14
(30)
, 4, 4 0, ,
,
b =
5
5 5

3
12 12
b =
, 3, 3 0, ,
,
Z8: XQ + XQ + XD + XD,
(31)
5
5 5

3
12 12
b =
, 3, 3 0, ,
,
Z9: XG + XW + XL + XL,
(32)
5
5 5

18
12 12
b =
, 6, 6 0, ,
,
Z10: XT + XT + XW + 2XG,
(33)
5
5 5
all of which either have b2 b1 = 6/5 (sets Zi, i = 1, . . . , 9) or b2 b1 = 12/5 (sets
Z8, Z9, Z10). The sets Zi (i = 2, . . . , 7) may be generated from Z1 , and Z9 and Z10 from Z8
by using one-loop beta function equivalent relations (see Table 1).
With two-loop RGE running for the gauge couplings and one-loop running for the Yukawa
and Higgs quartic coupling, we show the supersymmetry breaking scales, the gauge coupling
unification scales and the corresponding Higgs mass ranges for MV = 200 GeV and 1 TeV in
Table 2. The Higgs boson mass ranges correspond to the variation of tan between 1.5 and 50,
with smaller tan giving a smaller Higgs boson mass, and s and mt with their 1 ranges. For
the same b2 b1 , the actual values of b1 and b2 , as well as the different two-loop beta
functions due to the different additional particle contents can affect RGE running, and hence
these mass scales and Higgs boson masses. This is evident in comparing the Z1, Z3 and Z5
sets. The Z1 set has different b1 from Z3 and Z5, while Z3 and Z5 differ in the two-loop
beta functions due to the different extra particles involved. For Z1 through Z9 the Higgs mass
ranges are from about 119 GeV to 143 GeV for MV = 200 GeV and from about 122 GeV to
145 GeV for MV = 1 TeV. The Higgs mass ranges are from 103 GeV to 143 GeV and from
113 GeV to 145 GeV in the model with the Z10 set for MV = 200 GeV and MV = 1 TeV,
respectively. The Higgs mass ranges are larger (i.e., a lighter Higgs is allowed) in the model with
the Z10 set than the other models because the values of b1 and b2 are larger. In general, for
the models with b2 b1 = 6/5, the supersymmetry breaking scale is around 1010 GeV. For
those with b2 b1 = 12/5, the supersymmetry breaking scale is about 1015 GeV, which can
be considered as the GUT scale up to uncertainties from the threshold corrections at the scales
MV , MS , and MU . For a particular model with the Zi set, the SM gauge couplings, the Higgs
quartic coupling at the supersymmetry breaking scale, as well as the physical Higgs mass will
decrease if we increase MV .
As an example, we show the two-loop RGE running for the SM gauge couplings in the model
with the Z3 set in Fig. 2.
Z7:
Table 1
Equivalent sets and the equivalent relations involved
Equivalent sets
Equivalent relations
Equivalent sets
Equivalent relations
Z2 Z1
Z4 Z3
Z6 Z1
Z9 Z8
2
7
5, 6
1, 6, 7
Z3 Z1
Z5 Z1
Z7 Z6
Z10 Z8
1, 5
2, 5, 6
2
5, 6, 8
315
Table 2
The supersymmetry breaking scales, the gauge coupling unification scales, and the corresponding Higgs mass ranges for
MV = 200 GeV and 1000 GeV in the models with Zi sets of vector-like particles. All masses are in GeV
Z
b2 b1
MV = 200 GeV
MV = 1000 GeV
MS
MU
3.6 1016
mh
MS
MU
mh
3.3 1016
Z1
Z2
Z3
Z4
Z5
Z6
Z7
6/5
6/5
6/5
6/5
6/5
6/5
6/5
2.9 1010
2.5 1010
4.0 1010
5.1 1010
6.6 1010
1.3 1010
9.6 1010
4.4 1016
4.0 1016
4.0 1017
8.1 1016
7.0 1016
9.8 1016
123144
121143
121143
121143
121143
121143
119143
1.8 1010
1.5 1010
2.3 1010
2.9 1010
4.5 1010
8.2 1010
6.7 1010
4.0 1016
3.7 1016
3.7 1016
7.1 1016
6.1 1016
8.1 1016
125145
124145
124145
124145
123145
123145
122145
Z8
Z9
Z10
12/5
12/5
12/5
8.2 1015
3.9 1015
3.0 1015
6.2 1016
9.6 1016
3.1 1017
119143
119143
103143
3.0 1015
1.6 1015
1.4 1015
5.6 1016
8.3 1016
2.1 1017
123145
123145
113145
Fig. 2. Two-loop gauge coupling unification in the model with the Z3 set of vector-like particles. The solid curves are
for MV = 200 GeV, while the nearly overlapping dotted curves are for MV = 1 TeV.
4. Models with non-standard vector-like particles

In string model building, we may have vector-like particles which are charged under SU(3)C
SU(2)L and neutral under U (1)Y [23,24]. Often, such particles also carry hidden sector charges.
Thus, we introduce such non-standard vector-like particles in this section. Their quantum numbers under SU(3)C SU(2)L U (1)Y and their contributions to one-loop beta functions as
complete supermultiplets are
2, 0),
XQ0 + XQ0 = (3, 2, 0) + (3,
1, 0) + (3, 1, 0),
XD0 + XD0 = (3,
b = (0, 3, 2),
(34)
b = (0, 0, 1),
(35)
316
XL0 + XL0 = (1, 2, 0) + (1, 2, 0),

1, 0),
XS0 + XS0 = (6, 1, 0) + (6,
b = (0, 1, 0),
(36)
b = (0, 0, 5).
(37)
We do not consider XU 0 + XU 0, XT 0 or XY 0 + XY 0 because they are equivalent to XD0 +

XD0, XW , and XQ0 + XQ0, respectively.
Note that the states in (34) and (36) have half-integer electric charge, and that the lightest such
particles would be stable. Due to the stringent experimental limits on the natural abundances of
such particles and their bound states [35], they would have to be much more massive than the
reheating temperature after a period of inflation [36]. Thus, for them to exist at the TeV scale the
reheating temperature would have to be extremely low [37].
Since the non-standard vector-like particles are not part of the GUT multiplet, string-scale unification is assumed. In this case, a simple unified group is not necessary and so the hypercharge
can be non-standard. In addition, in the strong coupled heterotic string models or M-theory on
S 1 /Z2 [38,39], and in the intersecting D-brane models on type II orientifolds [24], the string
scale can be around the GUT scale. Therefore, the unification scale in our scenarios can be as
low as the GUT scale.
The additional independent one-loop beta function equivalent relations for the standard and
non-standard particle sets are
NEQV1:
2XD0 + 3XL0 XQ0,
NEQV2:
5XD0 XS0,
NEQV3:
2XD + XE + 2XL0 0,
NEQV4:
5XE + 6XD0 + 6XL0 0,
or
2XD0 + 3XL0 XQ0,
or
(38)
5XD0 XS0,
(39)
2XD + XE + 2XL0 0,
or
or 5XE + 6XD0 + 6XL0 0.
(40)
(41)
We consider the following simple sets of standard and non-standard vector-like particles

3
2 2
NZ1: XD0 + XD0 + XL + XL,
(42)
b =
, 1, 1 0, ,
,
5
5 5

18
2 2
NZ2: 4(XD0 + XD0) + XT + XT ,
(43)
b =
, 4, 4 0, ,
,
5
5 5

2
3 3
NZ3: XL0 + XL0 + XD + XD,
(44)
b =
, 1, 1 0, ,
,
5
5 5

6
4 4
b =
NZ4: 2(XD0 + XD0) + XW + XE + XE,
, 2, 2 0, ,
, (45)
5
5 5
NZ5:
XD0 + XD0 + XL0 + XL0,
NZ6:
XQ0 + XQ0 + XU + XU ,
NZ7:
XQ0 + XQ0 + XD + XD + XE + XE,

7 7
8
, 3, 3 0, ,
,
b =
5
5 5
b = (0, 1, 1),

8
7 7
b =
, 3, 3 0, ,
,
5
5 5
NZ8:
XD0 + XD0 + XD + XD + XW,
NZ9:
XL0 + XL0 + XG + 2(XL + XL),
(46)
(47)
(48)

2
8 8
b =
, 2, 2 0, ,
,
5
5 5

6
9 9
b =
, 3, 3 0, ,
,
5
5 5
(49)
(50)
317
NZ10:
2(XD0 + XD0) + XW,
N Z11:
XS0 + 5(XL + XL),
NZ12:
XD0 + XD0 + XL0 + XL0 + XQ + XQ + XU + XU ,

11 11
9
, 4, 4 0, ,
,
b =
5
5 5
NZ13:
NZ14:
NZ15:
b = (0, 2, 2),
(51)
b = (3, 5, 5) (0, 2, 2),
(52)
(53)
2(XL0 + XL0) + XG + XL + XL,

12 12
3
, 3, 3 0, ,
,
b =
5
5 5
2(XL0 + XL0) + XT + XT + 2XG,

12 12
18
,
6,
6
0,
,
,
b =
5
5 5
XQ0 + XQ0 + XD + XD,
(54)
(55)

b =

2
13 13
, 3, 3 0, ,
.
5
5 5
(56)
With the non-standard vector-like particles, b2 b1 can be n/5, where n = 2, 3, . . . , 13.

The sets with b3 = b2 and b2 b1 = 6/5 are given in the previous Section. Simple estimates of the supersymmetry breaking scales and the unification scales from one-loop RGE
running are already presented in the right plot of Fig. 1. With two-loop RGE running for the SM
gauge couplings and one-loop running for the Yukawa couplings and Higgs quartic coupling included, we present the more reliable supersymmetry breaking scales, gauge coupling unification
scales and the corresponding Higgs boson mass ranges in Table 3. The supersymmetry breaking
scales can be from 105 GeV to the GUT scale if we include the uncertainties from threshold
corrections at MV , MS and MU . In general, the supersymmetry breaking scale will be higher for
the models with larger b2 b1 .
Table 3
Same as Table 2, only for the N Zi sets
NZ
N Z1
N Z2
N Z3
N Z4
N Z5
N Z6
N Z7
N Z8
N Z9
N Z10
N Z11
N Z12
N Z13
N Z14
N Z15
b2 b1
2/5
2/5
3/5
4/5
1
7/5
7/5
8/5
9/5
2
2
11/5
12/5
12/5
13/5
MV = 200 GeV
MV = 1000 GeV
MS
MU
5.7 105
2.4 1016
1.8 106
9.0 106
2.6 108
1.9 109
3.9 1011
4.9 1011
2.7 1012
7.3 1012
1.5 1014
2.2 1013
1.2 1015
2.9 1015
2.0 1015
4.7 1016
2.9 1016
2.6 1016
3.0 1016
3.1 1016
4.3 1016
4.3 1016
4.2 1016
9.1 1016
4.8 1016
3.1 1017
7.4 1016
1.1 1017
3.6 1017
6.6 1016
mh
MS
MU
mh
114139
114140
119142
121143
124144
121144
121144
124144
121143
124144
113142
116143
120143
104143
118143
4.8 105
2.3 1016
1.6 106
1.4 106
1.8 108
1.2 109
2.1 1011
2.5 1011
1.3 1011
3.8 1012
6.5 1013
1.2 1013
4.7 1014
1.2 1015
9.8 1014
1.6 1016
6.4 1016
3.6 1016
2.8 1016
3.0 1016
4.0 1016
4.0 1016
3.9 1016
7.9 1016
4.5 1016
2.3 1017
6.5 1016
9.1 1016
2.4 1017
6.0 1016
114139
107139
115144
122144
125145
124145
124145
126146
124145
126146
119145
121145
123145
113145
123145
318
Fig. 3. Same as Fig. 2, only for the N Z1 set.
We show the two-loop RGE running of the SM gauge couplings in the model with the N Z1
set in Fig. 3. Because of the smaller b2 b1 value, the supersymmetry breaking scale is lower
compared to Fig. 2.
5. Comments on the phenomenological consequences
We now address the problem of how vector-like particles have masses at the electroweak
scale. Since mass terms of vector-like particles are invariant under the Standard Model gauge
group, we are allowed to write terms like XQXQ in the Lagrangian, and the natural scale of
this mass might be the unification scale. This would then lead to a new fine tuning problem.
A natural mechanism to forbid such mass terms is to embed the particles in a larger symmetry
group such that the only mass terms allowed are through Yukawa couplings with a singlet field S,
for example SXQXQ, with S having a VEV at the electroweak scale. This is the mechanism
of mass generation of vector-like down-type quarks based on the group E6 which could arise
from heterotic string compactification. The question of why vector-like particles do not occur in
complete GUT multiplets can be understood by breaking the GUT symmetry via Wilson lines
[40] or orbifold projections [41]. Another consequence of the singlet field S is that we can obtain
a strong first order electroweak phase transition with the presence of the trilinear interaction
SH H in the Higgs potential, similar to the next to the Minimal Supersymmetric Standard Model
[42,43] and the supersymmetric U (1) model [44].
The vector-like fermions can yield rich low energy phenomenology. Models with XD and
XD have received a lot of attention because they naturally occur in heterotic string inspired
E6 models [45]. It is interesting to note that transformation under the Standard Model gauge
group does not uniquely specify all the properties of such vector-like particles. For example,
the superpotential of XD and XD has to be defined before a complete description can be given.
Three possibilities depending on lepton and baryon number assignments are (a) down type quark,
(b) leptoquark, and (c) diquark. For a review of the production and decays of these particles see
319
Refs. [46,47]. They may also be quasi-stable decaying only by higher-dimensional operators [48]
with cosmological and collider implications [47,48]. For the models with XU and XU , there are
new effects in top and charm quark (e.g., D meson) physics, while for the models with XD
and XD, we have new effects in B physics [17,49,50]. Also, models with XQ, XQ, XD and
XD can explain the bottom quark forwardbackward asymmetry (AbFB ) [16]. Neutrino masses
and mixings can be generated if there exist XW and XL/XL, or two XW , or two XL/XL.
The neutral component of XW or XL/XL can be a cold dark matter candidate if there exists
a discrete symmetry and their masses are around the TeV scale [51]. The models with XW ,
XL, and XL, may not only explain the dark matter but also generate the baryon asymmetry
via electroweak baryogenesis [52]. Similar to split supersymmetry, the supersymmetry breaking
scale may not be higher than 1012 GeV in the models with XG and the standard vector-like
quarks because XG cannot decay fast enough via Yukawa couplings in the superpotential to
satisfy cosmological constraints [53]. For models with XG but no other standard vector-like
quarks, the cosmological constraint on XG and the phenomenological consequences deserve
detailed study because XG can be stable at least in some orbifold models. Similarly, whether
the non-standard vector-like particles can decay, and the cosmological constraints on the nonstandard vector-like particles and their phenomenological consequences deserve further detailed
study.
Let us focus on the experimentally viable models which have standard vector-like particles.
Suppose that the axion is the cold dark matter candidate, and we introduce two or three righthanded neutrinos to explain the neutrino masses and mixings and the baryon asymmetry. The
simple models with 1010 GeV-scale supersymmetry breaking are those with Z1 and Z3 sets,
and the simplest with 1015 GeV-scale supersymmetry breaking is the one with the Z8 set. If that
axion does not contribute to the dominant cold dark matter density, and the neutrino masses and
mixings are generated due to the R-parity violating terms [54], the model with the Z2 set is the
simplest which has a dark matter candidate and can explain the baryon asymmetry.
The phenomenological consequences of our models, for example, new effects in the meson
physics, CP violation, and the collider signatures at the LHC will be presented in detail elsewhere.
6. Discussions and conclusions
We studied the canonical gauge coupling unification in the extensions of the SM with universal high-scale supersymmetry breaking by introducing additional SM vector-like fermions. To
avoid the dimension-6 proton decay problem and quantum gravity effects, we require that the
gauge coupling unification scale is from 5 1015 GeV to the Planck scale. We assume that the
supersymmetry breaking scale is below the unification scale, and that the universal masses of
the vector-like fermions are from 200 GeV to 1 TeV. In order to have the canonical gauge coupling unification and to satisfy these requirements and assumptions, we showed that b2 = b3
and 2/5 b2 b1 13/5 for the extra vector-like particles. To systematically construct the
models with canonical gauge coupling unification, we used the technique of the one-loop beta
function equivalent relations for the particle sets. We discussed two kinds of models. The first
kind of models have standard vector-like particles while the second kind of models have standard
and non-standard ones. In the models with simple sets of extra vector-like fermions whose universal masses are 200 GeV and 1 TeV, we presented the supersymmetry breaking scales, gauge
coupling unification scales, and the corresponding Higgs mass ranges. In the first kind of models,
b2 b1 can only be equal to 6/5 and 12/5, and then the corresponding supersymmetry breaking scale can only be around 1010 and 1015 GeV, respectively. In the second kind, b2 b1
320
can be n/5, in which n = 2, 3, . . . , 13, so the supersymmetry breaking scale can be from 105 to
1016 GeV. Because the universal masses for the vector-like fermions are within the reach of the
LHC, these models can definitely be tested at the LHC.
We briefly commented on some phenomenological consequences of these models, which deserve further detailed study.
Acknowledgements
T.L. would like to thank H. Murayama and S. Nandi for helpful discussions on the gauge
coupling unification in the non-supersymmetric models, and thank S. Thomas for useful conversations. J.J. thanks the National Center for Theoretical Sciences in Taiwan for its hospitality,
where part of the work was done. This research was supported by the US Department of Energy under Grants Nos. DE-FG02-95ER40896, and DE-FG02-96ER40969, by the Friends of the
IAS, by the Cambridge-Mitchell Collaboration in Theoretical Cosmology, and by the University of Wisconsin Research Committee with funds granted by the Wisconsin Alumni Research
Foundation.
Appendix A. Renormalization group equations
In this appendix, we give the renormalization group equations in the SM and MSSM. The
general formulae for the renormalization group equations in the SM are given in Refs. [55,56],
and those for the supersymmetric models in Refs. [5761].
First, we summarize the renormalization group equations in the SM. The two-loop equations
for the gauge couplings are
3

gi3
2 d
3
2
(4)
(A.1)
,
Bij gj
di Tr h h
g i = g i bi +
dt
(4)2
j =1
=u,d,e
where t = ln and is the renormalization scale. g1 , g2 and g3 are the gauge couplings for
U (1)Y , SU(2)L and SU(3)C , respectively, where we use the SU(5) normalization g12 (5/3)gY2 .
The beta-function coefficients are
199 27
44

50
10
5
41 19
9
35
b=
(A.2)
, , 7 ,
B = 10
12
,
6
10
6
11
9
26
10 2

17 3
1 3
3 1
, ,2 ,
dd =
, ,2 ,
de =
, ,0 .
du =
(A.3)
10 2
2 2
2 2
Since the contributions in Eq. (A.1) from the Yukawa couplings arise from the two-loop diagrams, we only need Yukawa coupling evolution at one-loop order. The one-loop renormalization
group equations for the Yukawa couplings are
3

d
3
3
(4)2 hu = hu
(A.4)
ciu gi2 + hu hu hd hd + 2 ,
dt
2
2
i=1
3

3 u u 3 d d
2 d d
d
d 2
(4)
(A.5)
ci gi h h + h h + 2 ,
h =h
dt
2
2
i=1
3

d e
3 e e
e
e 2
h =h
(4)
ci gi + h h + 2 ,
dt
2
2
321
(A.6)
i=1
where hu , hd and he are the Yukawa couplings for the up-type quark, down-type quark, and
lepton, respectively. Also, cu , cd , and ce are given by

17 9
1 9
9 9
, ,8 ,
cd =
, ,8 ,
ce =
, ,0 ,
cu =
(A.7)
20 4
4 4
4 4
and

2 = Tr 3hu hu + 3hd hd + he he .
(A.8)
The one-loop renormalization group equation for the Higgs quartic coupling is

9 3 4 2 2 2
d
9 2
(4)2 = 122
g1 + 9g22 +
g1 + g1 g2 + g24 + 42 44 , (A.9)
dt
5
4 25
5
where
2
2
2

4 = Tr 3 hu hu + 3 hd hd + he he .
(A.10)
Next, we summarize the renormalization group equations in the MSSM. The two-loop renormalization group equations for the gauge couplings are
3

gi3
2 d
3
2
(4)
(A.11)
,
Bij gj
di Tr y y
gi = gi b i +
dt
(4)2
j =1
where the beta-function coefficients are

199

25
33
, 1, 3 ,
B = 95
b=
5

du =

26
, 6, 4 ,
5

dd =
=u,d,e
27
5
88
5
25 24 ,
(A.12)
11
5
9 14

14
18
, 6, 4 ,
de =
, 2, 0 .
5
5
The one-loop renormalization group equations for Yukawa couplings are

3
u u
2 d u
u
u u
d d
u 2
(4)
y = y 3y y + y y + 3 Tr y y
ci gi ,
dt
i=1

3

d d
2 d d
d
u u
d d
e e
d 2
y = y y y + 3y y + Tr 3y y + y y
(4)
ci gi ,
dt
i=1

3

d d
2 d e
e
e e
e e
e 2
(4)
c i gi ,
y = y 3y y + Tr 3y y + y y
dt
(A.13)
(A.14)
(A.15)
(A.16)
i=1
where y u , y d and y e are the Yukawa couplings for the up-type quark, down-type quark, and
lepton, respectively. cu , cd , and ce are given by

16
16
13
7
9
cu =
(A.17)
, 3,
,
cd =
, 3,
,
ce =
, 3, 0 .
15
3
15
3
5
322
Appendix B. Two-loop beta functions for the vector-like particles

In this appendix, we present two-loop beta functions contributions to the SM gauge couplings
from the vector-like particles which are introduced in Sections 3 and 4. The general formulae are
also given in Refs. [5561].
The two-loop beta functions (Bij ) from the extra particles in the non-supersymmetric models are
1
3
8
64
64
150
10
15
75 0 15
1
49
B XU +XU = 0 0 0 ,
B XQ+XQ = 10
(B.1)
8 ,
2
1
15
4
75
B XD+XD = 0
XE+XE
2
15
36
25
= 0
0
0
0
B XW = 0
64
3
76
3
3
0
0
16
15
38
3
0
0
0 ,
0
0,
0 0
0
0,
XG
0 0
B XT +XT =
B XQ+XQ =
B XD+XD = 0
XE+XE
4
15
72
25
= 0
0
6
0
0 0 ,
34
3
0 0
0 0,
0 0
0,
6
5
2
5
3
15
2
49
2
40
3
76
3
= 0
8 ,
0 0
128
75
B XU +XU = 0
B XL+XL =
0 0
XG
B
= 0 0
0 0
(B.5)
0 ,
0 0
0 0
0 0 .
0 250
3
(B.4)
(B.6)
38
3
15
68
3
32
15
B XD0+XD0 = 0 0
(B.2)
(B.3)
21 16 ,
0
25
In the supersymmetric models

1
3
16
75
1
5
2
15
8
75
0,
0
48
72
5
128
3
B XY +XY =
XS+XS
0 ,
25
24
5
= 0 0
38
3
0
9
10
49
6
9
50
3
10
0 0
108
0 0 0
128
0 64
75
3
B XS+XS = 0 0 0 ,
8
0 250
3
3
0 0 0
8
B XQ0+XQ0 = 0 49
,
2
76
0 3
3
0 0 0
B XL0+XL0 = 0 49
0,
6
0
B XL+XL =
8
15
16
15
9
25
3
5
0
0
128
15
34
3
9
5
0 ,
(B.8)
7 0,
0 0
0
0 ,
54
(B.7)
(B.9)
(B.10)
323
216
25
48
5
144
5
B XW = 0 24 0 ,
B XT +XT =
48 0 ,
0 0 0
0
0 0
256
25
128
0
15 80
75
3
3
3
B XS+XS = 0 0 0 ,
B XY +XY = 5 21 16 ,
16
3
290
3
0
0
10
3
6
0 0
B XQ0+XQ0 = 0
68
3
[1] R. Bousso, J. Polchinski, JHEP 0006 (2000) 006;

S.B. Giddings, S. Kachru, J. Polchinski, Phys. Rev. D 66 (2002) 106006;
S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, Phys. Rev. D 68 (2003) 046005;
L. Susskind, hep-th/0302219;
F. Denef, M.R. Douglas, JHEP 0405 (2004) 072;
F. Denef, M.R. Douglas, JHEP 0503 (2005) 061.
[2] S. Weinberg, Phys. Rev. Lett. 59 (1987) 2607.
[3] V. Agrawal, S.M. Barr, J.F. Donoghue, D. Seckel, Phys. Rev. Lett. 80 (1998) 1822;
V. Agrawal, S.M. Barr, J.F. Donoghue, D. Seckel, Phys. Rev. D 57 (1998) 5480.
[4] J.F. Donoghue, Phys. Rev. D 69 (2004) 106012;
J.F. Donoghue, Phys. Rev. D 69 (2004) 129901, Erratum.
[5] R.D. Peccei, H.R. Quinn, Phys. Rev. Lett. 38 (1977) 1440;
R.D. Peccei, H.R. Quinn, Phys. Rev. D 16 (1977) 1791.
[6] K.S. Babu, I. Gogoladze, K. Wang, Phys. Lett. B 560 (2003) 214.
[7] V. Barger, C.W. Chiang, J. Jiang, T. Li, Nucl. Phys. B 705 (2005) 71.
[8] 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.B. Green, J.H. Schwarz, P. West, Nucl. Phys. B 254 (1985) 327.
[9] P. Svrcek, E. Witten, JHEP 0606 (2006) 051, hep-th/0605206.
[10] A. Giryavets, S. Kachru, P.K. Tripathy, JHEP 0408 (2004) 002;
L. Susskind, hep-th/0405189;
M.R. Douglas, hep-th/0405279;
M.R. Douglas, hep-th/0409207;
M. Dine, E. Gorbatov, S. Thomas, hep-th/0407043;
E. Silverstein, hep-th/0407202;
J.P. Conlon, F. Quevedo, JHEP 0410 (2004) 039;
M. Dine, D. ONeil, Z. Sun, JHEP 0507 (2005) 014.
[11] N. Arkani-Hamed, S. Dimopoulos, JHEP 0506 (2005) 073.
[12] P.G. Camara, L.E. Ibanez, A.M. Uranga, Nucl. Phys. B 708 (2005) 268;
L.E. Ibnez, Phys. Rev. D 71 (2005) 055005;
A. Font, L.E. Ibanez, JHEP 0503 (2005) 040.
[13] D. Lst, S. Reffert, S. Stieberger, Nucl. Phys. B 727 (2005) 264;
D. Lst, S. Reffert, S. Stieberger, Nucl. Phys. B 706 (2005) 3.
[14] H. Georgi, S.L. Glashow, Phys. Rev. Lett. 32 (1974) 438;
P. Langacker, M.X. Luo, Phys. Rev. D 44 (1991) 817;
(B.12)
B XD0+XD0 = 0 0 0 ,
21 16 ,
0 6 68
0 0 34
3
0 0 0
0 0 0
B XS0+XS0 = 0 0 0 .
B XL0+XL0 = 0 7 0 ,
0 0 0
0 0 290
3
References
(B.11)
(B.13)
(B.14)
324
[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]
J.R. Ellis, S. Kelley, D.V. Nanopoulos, Phys. Lett. B 260 (1991) 131;
U. Amaldi, W. de Boer, H. Furstenau, Phys. Lett. B 260 (1991) 447.
P.H. Frampton, S.L. Glashow, Phys. Lett. B 131 (1983) 340;
P.H. Frampton, S.L. Glashow, Phys. Lett. B 135 (1984) 515, Erratum;
S. Nandi, Phys. Lett. B 142 (1984) 375;
H. Murayama, T. Yanagida, Mod. Phys. Lett. A 7 (1992) 147;
T.G. Rizzo, Phys. Rev. D 45 (1992) 3903;
G.F. Giudice, A. Romanino, Nucl. Phys. B 699 (2004) 65;
G.F. Giudice, A. Romanino, Nucl. Phys. B 706 (2005) 65, Erratum.
D. Choudhury, T.M.P. Tait, C.E.M. Wagner, Phys. Rev. D 65 (2002) 053002.
D.E. Morrissey, C.E.M. Wagner, Phys. Rev. D 69 (2004) 053001.
A. Kehagias, N.D. Tracas, hep-ph/0506144.
V. Barger, J. Jiang, P. Langacker, T. Li, Phys. Lett. B 624 (2005) 233;
V. Barger, J. Jiang, P. Langacker, T. Li, Nucl. Phys. B 726 (2005) 149.
W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.
M. Cvetic, D.A. Demir, J.R. Espinosa, L.L. Everett, P. Langacker, Phys. Rev. D 56 (1997) 2861, hep-ph/9703317;
M. Cvetic, D.A. Demir, J.R. Espinosa, L.L. Everett, P. Langacker, Phys. Rev. D 58 (1998) 119905, Erratum.
T. Li, H. Murayama, S. Nandi, unpublished.
K.R. Dienes, Phys. Rep. 287 (1997) 447.
R. Blumenhagen, M. Cvetic, P. Langacker, G. Shiu, Annu. Rev. Nucl. Part. Sci. 55 (2005) 71.
H. Davoudiasl, R. Kitano, T. Li, H. Murayama, Phys. Lett. B 609 (2005) 117.
H. Fritzsch, P. Minkowski, Phys. Lett. B 62 (1976) 72;
M. Gell-Mann, P. Ramond, R. Slansky, in: F. van Nieuwenhuizen, D. Freedman (Eds.), Supergravity, North Holland,
Amsterdam, 1979, p. 315;
T. Yanagida, in: Proceedins of the Workshop on Unified Theory and the Baryon Number of the Universe, KEK,
Japan, 1979;
S. Weinberg, Phys. Rev. Lett. 43 (1979) 1566;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45;
P.H. Frampton, S.L. Glashow, T. Yanagida, Phys. Lett. B 548 (2002) 119;
V. Barger, D.A. Dicus, H.J. He, T. Li, Phys. Lett. B 583 (2004) 173.
V. Barger, J. Jiang, P. Langacker, T. Li, hep-ph/0612206.
I. Gogoladze, T. Li, Q. Shafi, hep-ph/0602040.
B. Schrempp, M. Wimmer, Prog. Part. Nucl. Phys. 37 (1996) 1, hep-ph/9606386.
H. Arason, D.J. Castano, B. Keszthelyi, S. Mikaelian, E.J. Piard, P. Ramond, B.D. Wright, Phys. Rev. D 46 (1992)
3945;
H.E. Haber, R. Hempfling, A.H. Hoang, Z. Phys. C 75 (1997) 539.
CDF Collaboration, hep-ex/0507091.
Tevatron Electroweak Working Group, E. Brubaker, et al., hep-ex/0608032.
S. Bethke, hep-ex/0606035.
I.T. Lee, et al., Phys. Rev. D 66 (2002) 012002, hep-ex/0204003;
M.L. Perl, E.R. Lee, D. Loomba, Mod. Phys. Lett. A 19 (2004) 2595.
A. Kudo, M. Yamaguchi, Phys. Lett. B 516 (2001) 151, hep-ph/0103272.
I. Gogoladze, T. Li, V.N. Senoguz, Q. Shafi, Phys. Rev. D 74 (2006) 126006, hep-ph/0608181.
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142.
E. Witten, Nucl. Phys. B 471 (1996) 135, hep-th/9602070.
V. Braun, Y.H. He, B.A. Ovrut, T. Pantev, Phys. Lett. B 618 (2005) 252.
Y. Kawamura, Prog. Theor. Phys. 103 (2000) 613;
G. Altarelli, F. Feruglio, Phys. Lett. B 511 (2001) 257;
L. Hall, Y. Nomura, Phys. Rev. D 64 (2001) 055003;
A. Hebecker, J. March-Russell, Nucl. Phys. B 613 (2001) 3;
T. Li, Phys. Lett. B 520 (2001) 377;
T. Li, Nucl. Phys. B 619 (2001) 75.
M. Pietroni, Nucl. Phys. B 402 (1993) 27.
A.T. Davies, C.D. Froggatt, R.G. Moorhouse, Phys. Lett. B 372 (1996) 88;
S.J. Huber, M.G. Schmidt, Eur. Phys. J. C 10 (1999) 473;
S.J. Huber, M.G. Schmidt, Nucl. Phys. B 606 (2001) 183.
[44] J. Kang, P. Langacker, T. Li, T. Liu, Phys. Rev. Lett. 94 (2005) 061801.
[45] V.D. Barger, N. Deshpande, R.J.N. Phillips, K. Whisnant, Phys. Rev. D 33 (1986) 1912;
V.D. Barger, N. Deshpande, R.J.N. Phillips, K. Whisnant, Phys. Rev. D 35 (1987) 1741, Erratum.
[46] J.L. Hewett, T.G. Rizzo, Phys. Rep. 183 (1989) 193.
[47] J. Kang, P. Langacker, B.D. Nelson, arXiv: 0708.2701 [hep-ph].
[48] M. Fairbairn, A.C. Kraan, D.A. Milstead, T. Sjostrand, P. Skands, T. Sloan, hep-ph/0611040.
[49] V.D. Barger, M.S. Berger, R.J.N. Phillips, Phys. Rev. D 52 (1995) 1663, hep-ph/9503204.
[50] N.G. Deshpande, D.K. Ghosh, Phys. Lett. B 593 (2004) 135.
[51] R. Essig, S. Thomas, in preparation.
[52] M. Carena, A. Megevand, M. Quiros, C.E.M. Wagner, Nucl. Phys. B 716 (2005) 319.
[53] A. Arvanitaki, C. Davis, P.W. Graham, A. Pierce, J.G. Wacker, Phys. Rev. D 72 (2005) 075011.
[54] Y. Grossman, S. Rakshit, Phys. Rev. D 69 (2004) 093002.
[55] M.E. Machacek, M.T. Vaughn, Nucl. Phys. B 222 (1983) 83;
M.E. Machacek, M.T. Vaughn, Nucl. Phys. B 236 (1984) 221;
M.E. Machacek, M.T. Vaughn, Nucl. Phys. B 249 (1985) 70.
[56] G. Cvetic, C.S. Kim, S.S. Hwang, Phys. Rev. D 58 (1998) 116003.
[57] V.D. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 47 (1993) 1093.
[58] V.D. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 49 (1994) 4908.
[59] S.P. Martin, M.T. Vaughn, Phys. Rev. D 50 (1994) 2282, and references therein.
[60] M.x. Luo, Y. Xiao, Phys. Rev. Lett. 90 (2003) 011601, hep-ph/0207271.
[61] M.x. Luo, H.w. Wang, Y. Xiao, Phys. Rev. D 67 (2003) 065019, hep-ph/0211440.
325

Final results on oscillation

from the CHORUS experiment
CHORUS Collaboration
E. Eskut, A. Kayis-Topaksu, G. nengt
ukurova University, Adana, Turkey
M.G. van Beuzekom, R. van Dantzig, M. de Jong, J. Konijn, O. Melzer,

R.G.C. Oldeman 1 , E. Pesen, C.A.F.J. van der Poel, J.L. Visschers
NIKHEF, Amsterdam, The Netherlands
M. Gler, U. Kse, M. Serin-Zeyrek, R. Sever, P. Tolun, M.T. Zeyrek

METU, Ankara, Turkey
N. Armenise , F. Cassol, M.G. Catanesi, M. De Serio, M.T. Muciaccia,

E. Radicioni, P. Righini, S. Simone, L. Vivolo
Universit di Bari and INFN, Bari, Italy
A. Blte, K. Winter
Humboldt Universitt, Berlin, Germany
M. Vander Donckt 2 , B. Van de Vyver, P. Vilain, G. Wilquet

Inter-University Institute for High Energies (ULB-VUB) Brussels, Belgium
B. Saitta
Universit di Cagliari and INFN, Cagliari, Italy
E. Di Capua, C. Luppi
Universit di Ferrara and INFN, Ferrara, Italy
doi:10.1016/j.nuclphysb.2007.10.023
CHORUS Collaboration / Nuclear Physics B 793 (2008) 326343
327
Y. Ishii, M. Kazuno, S. Ogawa, H. Shibuya

Toho University, Funabashi, Japan
J. Brunner, M. Chizhov, D. Cussans, M. Doucet, J.P. Fabre, W. Flegel,

I.R. Hristova 3 , T. Kawamura, D. Kolev 4 , M. Litmaath, H. Meinhard,
E. Niu, H. veras, J. Panman, I.M. Papadopoulos, S. Ricciardi 5 ,
A. Rozanov 6 , D. Saltzberg 7 , R. Tsenov 4 , J.W.E. Uiterwijk,
C. Weinheimer, H. Wong, P. Zucchelli
CERN, Geneva, Switzerland
J. Goldberg
Technion, Haifa, Israel
M. Chikawa
Kinki University, Higashiosaka, Japan
E. Arik, A.A. Mailov

Bogazici University, Istanbul, Turkey
J.S. Song, C.S. Yoon

Gyeongsang National University, Jinju, Republic of Korea
K. Kodama, N. Ushida
Aichi University of Education, Kariya, Japan
S. Aoki, T. Hara
Kobe University, Kobe, Japan
G. Brooijmans 8 , T. Delbar, D. Favart, G. Grgoire, J. Hrin,

S. Kalinin 9 , I. Makhlioueva
Universit Catholique de Louvain, Louvain-la-Neuve, Belgium
A. Artamonov, P. Gorbunov, V. Khovansky, V. Shamanov, I. Tsukerman

Institute for Theoretical and Experimental Physics, Moscow, Russian Federation
D. Bonekmper, N. Bruski, D. Frekers, D. Rondeshagen, T. Wolff

Westflische Wilhelms-Universitt, Mnster, Germany
328
K. Hoshino, J. Kawada, M. Komatsu, Y. Kotaka, T. Kozaki,

M. Miyanishi, M. Nakamura, T. Nakano, K. Narita, K. Niu, K. Niwa,
N. Nonaka, Y. Obayashi, O. Sato, T. Toshito
Nagoya University, Nagoya, Japan
S. Buontempo, A.G. Cocco, N. DAmbrosio 10 , G. De Lellis, G. De Rosa,

F. Di Capua, A. Ereditato 11 , G. Fiorillo, A. Marotta, M. Messina 11 ,
P. Migliozzi, V. Palladino, L. Scotto Lavina , P. Strolin, V. Tioukov
Universit Federico II and INFN, Naples, Italy
K. Nakamura, T. Okusawa
Osaka City University, Osaka, Japan
A. Capone, D. De Pedis, S. Di Liberto, U. Dore, P.F. Loverre,

L. Ludovici, A. Maslennikov, M.A. Mazzoni, G. Piredda, G. Rosa,
R. Santacesaria, A. Satta 12 , F.R. Spada
Universit La Sapienza and INFN, Rome, Italy
E. Barbuto, C. Bozza, G. Grella, G. Romano, C. Sirignano, S. Sorrentino

Universit di Salerno and INFN, Salerno, Italy
Y. Sato, I. Tezuka
Utsunomiya University, Utsunomiya, Japan
Available online 1 November 2007
Abstract
The final oscillation analysis of the complete set of data collected by CHORUS in the years 19941997
is presented. Reconstruction algorithms of data extracted by electronic detectors were improved and the
E-mail address: scotto@cern.ch (L. Scotto Lavina).

1 Now at Universit di Cagliari, Cagliari, Italy.
2 Now at CERN, Switzerland.
3 Now at DESY, Hamburg.
4 On leave of absence and at St. Kliment Ohridski University of Sofia, Bulgaria.
5 Now at Rutherford Appleton Laboratory, Oxon, United Kingdom.
6 Now at CPPM CNRS-IN2P3, Marseille, France.
329
data recorded in the emulsion target were analysed by new automated scanning systems, allowing the use
of a new method for event reconstruction in emulsion. CHORUS has applied these new techniques to the
sample of 19961997 events for which no muons were observed in the electronic detectors. Combining
the new sample with the data analysed in previous papers, the overall sensitivity of the experiment to the
appearance is thus improved. In a two-neutrino mixing scheme, a 90% C.L. upper limit of sin2 2 <
4.4 104 is set for large m2 , improving by a factor 1.5 the previously published CHORUS result.
PACS: 14.60.Pq
Keywords: Oscillation; Neutrino; Tau lepton appearance
1. Introduction
The CHORUS experiment was designed to search for oscillations through the observation of charged-current interactions N X followed by the decay of the lepton,
directly observed in a nuclear emulsion target. The experiment aimed at achieving maximum
sensitivity on the effective mixing angle for values of the mass parameter m2 larger than
10 eV2 /c4 . This particular choice was based on the hypothesis that the neutrino mass could
contribute to the solution of the Dark Matter puzzle [1,2]. A short-baseline experiment in the
CERN SPS Wide Band Neutrino Beam [3] was well suited for this search. Another experiment,
NOMAD, used the same beam and searched for appearance using a purely electronic technique [4].
The CHORUS experiment took data from 1994 to 1997. A first phase of data analysis
(Phase I) was performed and no evidence for oscillations was found [57]. Owing to several
improvements in automated emulsion scanning and in the event reconstruction, it was considered
worthwhile to perform a new and more complete analysis (Phase II) of the data collected in the
19961997 period.
In this paper, we report the analysis performed on the events for which no muons were observed in the electronic detector (0 events). In particular, a dedicated search for decays into
three charged hadrons was also performed. Combining this new sample with the data samples
analysed and published in previous papers, i.e. with a muon observed in the electronic detector
(1) for the whole period of data taking and a small 0 sample collected in 19941995, leads
to an improvement of the CHORUS sensitivity to appearance by a factor 1.5 and to
e appearance by a factor 1.2.
Although it is now established that oscillations occur at m2 103 , this paper
intends to show the capabilities of a hybrid emulsion experiment and that the goal for which
CHORUS was designed has been reached.
7 Now at University of California, Los Angeles, USA.
8 Now at Columbia University, New York, USA.
9 Now at Rheinisch-Westfaelische Technische Hochschule, Aachen, Germany.
10 Now at Laboratorio Nazionale del Gran Sasso, Assergi, Italy.
11 Now at Bern University, Bern, Switzerland.
12 Now at Universit di Roma Tor Vergata, Rome, Italy.
Deceased.
330
2. The experimental setup

The CERN SPS Wide Band Neutrino Beam consists essentially of with a contamination
of about 5.1% , 0.8% of e , and 0.2% of e , while the flux of is negligible [8]. The average
neutrino energy is 26 GeV, well above the production threshold.
The CHORUS detector is a hybrid setup which combines a nuclear emulsion target with various electronic detectors such as trigger hodoscopes, a scintillating fibre tracker system, a hadron
spectrometer, electromagnetic and hadronic calorimeters, and a muon spectrometer [9].
Thanks to its micrometric granularity, nuclear emulsion allows a precise three-dimensional
reconstruction of the neutrino interaction vertex as well as of the decay vertices of associated
short-lived particles. It is a powerful detector technique to directly observe decays which, in
this experiment, occur on average at a distance of 1.7 mm from the interaction vertex.
The scintillating fibre tracker system consists of two components: the target tracker (TT),
which is the major tool for locating in the emulsion the region where a neutrino interaction has
occurred, and the diamond tracker (DT), a set of trackers associated with the hadron spectrometer.
The hadron spectrometer measures the charge and momentum of charged particles using an
air-core magnet. The calorimeter is used to determine the energy of electromagnetic and hadronic
showers. The magnetized iron muon spectrometer determines the charge and momentum of
muons.
The emulsion target has an overall mass of 770 kg and is segmented into four stacks. Each
stack consists of eight modules with 36 plates of size 36 cm 71 cm. Each plate has a 90 m
plastic base coated on both sides with a 350 m emulsion layer. Each stack is followed by three
interface emulsion sheets having a 100 m emulsion layer on both sides of a 800 m thick plastic
base, and by a set of TT planes. The TT predicts particle trajectories on the interface emulsion
sheets with a precision of about 150 m in position and 2 mrad in track angle.
The emulsion scanning is performed by computer-controlled, fully automated microscope
stages equipped with a CCD camera and a read-out system called track selector [10,11]. The
last generation of automated microscopes used in CHORUS experiment is the UTS (Ultra
Track Selector) [12]. In order to recognize track segments in the emulsion, a series of tomographic images is taken by focusing at different depths in the emulsion layer. The digitized
images at different levels are shifted according to the predicted track angle and then added.
The presence of aligned grains forming a track is detected as a local peak in the grey-level of
the summed image. The track-finding efficiency is higher than 98% for track slopes less than
400 mrad [11].
3. The event reconstruction
The event reconstruction algorithms have been improved significantly with respect to the
Phase I analysis described in detail in Ref. [7]. Here, we only summarize the essential points.
The event reconstruction starts with the pattern recognition in the electronic detectors. Tracks
are found in the TT, in the calorimeter and in the muon spectrometer. If a muon is detected in
the downstream detectors the event is classified as 1, otherwise it is called 0. Vertices are
reconstructed by the electronic detectors using the points of closest approach of the tracks in
the TT immediately downstream from the target. The main vertex is the most upstream one
and is selected if it contains either a muon or a hadron surviving criteria which differed in the
331
two phases.13 Such particles or the muon are used as so-called scan-back tracks. The impact
points of the scan-back tracks are predicted on the most downstream interface emulsion sheets,
to initiate their follow back in emulsion.
As in previous analyses the event reconstruction in emulsion starts with a procedure called
vertex location. As a first step, all the tracks in the interface emulsion sheets within an area of
1 mm2 centred on each scan-back track prediction are collected.
The correct association of emulsion tracks with TT tracks, both in position and direction,
requires a precise alignment of the interface emulsions with respect to the fibre-trackers. The
alignment parameters are obtained by finding the best match between the full set of TT predicted
tracks and the full set of collected emulsion tracks. Once an emulsion track is associated to a
scan-back track, mainly on the basis of their common direction, it is followed upstream from one
emulsion plate to the next within a greatly reduced scanning area as the resolution improves. In
the emulsion stacks, the scanning area reduces to a square of 50 m side. The plate-to-plate alignment is first obtained by a coarse adjustment of reference marks, and, as the scanning procedure
continues, is refined by aligning track maps measured in subsequent plates. For this alignment
we used tracks of particles generated by neutrino interactions during the whole data taking, of
cosmic rays and of muons coming from neighbouring beams. The scan-back procedure stops
when a searched track is not found in two consecutive plates. The most downstream one is
defined as the vertex plate. This procedure for locating the vertex has an efficiency of 34%
(Table 1).
Once the vertex plate is found, a new scanning technique is applied on the 19961997
data sample. Originally developed for the DONUT experiment [16], this technique, called
NetScan [17], is described in detail in Refs. [13,18] together with the event location procedure. Its application to the 0 sample is also described in the following section.14
3.1. The NetScan technique
Once the vertex plate is identified, the UTS performs a scan of the emulsion volume around
the vertex position, recording, for each event, all track segments within 400 mrad with respect
to the orthogonal direction of the plates. In each plate only the most upstream 100 m part is
scanned. The scanned volume is 1.5 mm 1.5 mm wide and 6.3 mm long in the beam direction,
corresponding to eight emulsion plates. This volume contains the vertex plate itself, the plate
immediately upstream, and the six plates downstream the vertex plate. The plate upstream the
vertex acts as a veto for passing through tracks. The six plates downstream from the vertex act as
decay volume and are used to detect the tracks of the decay daughters. The scan area is centred
on the scan-back track stopping point.
The number of track segments (coming from particles induced by neutrino interactions, cosmic ray particles and muons from neighbouring beams) found in each plate depends on the
position of the NetScan volume with respect to the beam centre and on average is 920.
The task of the NetScan event reconstruction is to select the segments belonging to the neutrino interaction under study, out of this large number of background track segments. For this
13 In Phase I, 0 events, at least one hadron with a momentum less than 20 GeV/c was required. In Phase II the most
isolated hadron was selected regardless of its charge and momentum.

14 The NetScan technique has been used also for the 1 data sample, mainly to search for charmed particles decays [14,
15].
332
Table 1
Results of the reconstruction of the 0 sample
Stage of reconstruction
Number of events
Interface emulsion scanned

Vertex plate found
NetScan acquisition accepted
Vertex reconstructed
102544
35039
29404
22661
purpose, a first plate-to-plate alignment is performed by comparing the pattern of segments in a

plate with the corresponding pattern in the next upstream plate. Each segment found in one plate
is extrapolated to the next plate where a matching segment is looked for within about 4 m in
position (3 of alignment resolution) and 20 mrad in angle. If none is found, the straight-line
extrapolation is tried one plate further upstream.
A second and more accurate inter-plate alignment is performed using tracks passing through
the entire volume after the connection of all matched segments. These tracks come from muons
associated with the neutrino beam or with charged particle beams in the same experimental area.
After this fine alignment, the distribution of the residuals of the segment positions with respect
to the fitted track has a standard deviation width of about 0.45 m.
After rejection of isolated track segments, typically about 400 tracks remain in the volume.
The majority of these are tracks (mainly Compton electrons and -rays) with a momentum less
than 100 MeV/c. These background tracks are rejected on the basis of the 2 of a straightline fit to the track segments. The final step is the rejection of the tracks not originating from
the scan volume. After this filtering, the average number of remaining tracks is about 40. The
number of connected tracks and the position and slope residuals provide a good measurement
of the alignment quality. About 84% of the events passes the quality cuts (NetScan acquisition
accepted).
The reconstruction algorithm then tries to associate the tracks to vertices with a pair-based
method: after selecting pairs of tracks having minimal distance less than 10 m, a clustering
among them is performed and vertex points are defined. After this clustering, a track is attached to a vertex if its distance from the vertex point (called hereafter impact parameter)
is less than 3 m. At the end of the procedure, one defines a primary vertex (and its associated tracks) and possibly one or more secondary vertices to which daughter tracks are
attached.
About 77% of the events, where the NetScan acquisition was accepted, have a vertex reconstructed in the NetScan volume. The loss is accounted for by the fact that a background track
might have been selected for the scan-back procedure. The numbers of events at the various
stages of the procedure are given in Table 1.
3.2. The decay search
Once tracks and vertices are reconstructed in the NetScan volume, a search for decay topologies is performed. In Fig. 1 a sketch of a Monte Carlo event inside the NetScan volume is
shown: a (not drawn) interacts producing a lepton and other particles. After three plates, the
lepton decays producing one charged particle (kink topology, i.e. an observed abrupt change
of direction in the track).
333
Fig. 1. The NetScan volume and a Monte Carlo event with one-prong (i.e. kink) decay.
In the decay search an emulsion track is called a TT-confirmed track if its trajectory matches
that of a TT reconstructed track, or if it is connected to a sequence of TT hits. A reconstructed
vertex is a TT-confirmed vertex if it contains at least one TT-confirmed track.
The decay search is done in two stages:
(i) an automated search on the whole sample;
(ii) a computer-assisted visual inspection on the sample selected in the previous stage.
After the automated search, 754 events are selected as belonging to one of three categories:
Short decays. Events for which there is only one TT-confirmed vertex in the NetScan volume and at least one isolated TT-confirmed track having an impact parameter with the primary
vertex inside the range [7.2, 100] m.
Long decays. Events with a one-prong decay topology (kink) directly observed in the
NetScan volume. This occurs when the parent track is observed in at least one emulsion plate.
The daughter track must be TT-confirmed and its minimal distance to the parent track is required
smaller than 6 m. In case a primary vertex is reconstructed, it is required to be TT-confirmed
and the parent track must have an impact parameter less than 3 m with this vertex.
Multiprong decays. Events in which NetScan reconstructs at least two vertices, the secondary vertices being candidates for multiprong decays. The quality of the vertices is ranked
according to the number of TT-confirmed tracks originating from them and it is used as a parameter for the selection of multiprong decay topology. The parent angle, in the case of a neutral
particle decay deduced from the line connecting the primary and secondary vertices, should be
within 400 mrad from the beam direction.
The computer-assisted visual inspection aims at clearly establishing the topology of the primary and secondary vertices. To avoid ambiguous topologies, the flight length of the parent track
is required to be larger than 25 m. For vertices which are not both TT-confirmed, a more stringent cut, which is a function of the flight length of the assumed parent, is applied to remove the
random association of two unrelated vertices.
An event is selected if one of the following topologies is detected:
334
Table 2
0 events selected after visual inspection
Topology
Number of events
C1
C3
V2
59
48
99
C1: a track with a kink (interaction or decay) of at least 50 mrad;

C3: a track with a three-prong (interaction or decay) topology;
V2: a two-prong neutral interaction (or decay) is detected.
The numbers of events observed in these topologies are shown in Table 2. Decays of the
lepton are characterized by either one or three charged particles in the final state, C1 or C3. The
events with a V2 topology will be used for background calculations.
3.3. Background evaluation
Physical processes that can mimic a decay are:
+
decays of charmed charged particles like D+ , D+

s and c , if the primary is not identified and the charge of the charmed particle is not determined;
white interactions, i.e. interactions of hadrons without any other visible activity at the vertex
(recoil or Auger electrons);
decays which affect only the C1 topology.
The charmed particle contribution to the background has been evaluated from production
+
cross-section of D+ , D+
s and c in neutrino interactions and their decay branching fractions
into one or three charged particles [19]. The rate of D0 production in neutrino charged-current
interactions has been measured in CHORUS [14]. Using as the normalization the observed number of events with a V2 topology given in Table 2, the number of one and three-prong decays
from charm particles in the 0 sample is given by
NCi = NV2
C+ Ci
,
D0 V2
where i = 1 or 3, C+ /D0 = 1.03 0.16 is the ratio of charged charm and D0 cross-sections [19]
and V2 , C1 and C3 are the efficiencies in detecting the specific topology, including the
branching ratios, computed by a Monte Carlo simulation: C1 /V2 = 0.28 0.08, C3 /V2 =
0.45 0.09. The number of V2 events has been corrected to account for K0 and 0 decays [18].
This yields a prediction of NC1 = 27.2 8.6 and NC3 = 44 11.
One-prong interactions without visible activity (white kinks, WK) were generated and
processed through the full simulation chain assuming a hadron interaction length = 24 m [20].
In the 1 sample, the ratio of three-prong to one-prong interactions, both with visible activity,
was measured to be 11%. This fraction was assumed to be the same also in absence of recoil
or Auger electrons [18].
The results are summarized in Table 3.
335
Table 3
Expected background and observed data for C1 and C3 topologies in the 0 sample
Topology
Charm
White interactions
Other decays
Total BG
Data
0 C1
0 C3
27.2 8.6
44 11
24.9 2.7
2.7 0.3
1.11 0.34
53.2 9.0
47 11
59
48
Table 4
branching ratios, efficiencies and maximum detectable events for C1 and C3 topologies in the 0 sample
Topology
Channel
BR (%)
loc (%)
sel (%)
0 C1
h
e

49.5
17.8
17.4
32.6 0.5
26.9 0.6
5.9 0.3
23.7 0.6
21.2 1.3
24 3
7227
1922
472
9621
3h
15.2
35.8 0.8
43.1 1.5
4443
4443
0 C3
Nmax
3.4. Maximum number of detectable

We define NMAX as the number of events that would be observed if their detection efficiency
and the oscillation probability were equal to 1. It can be evaluated as:

NCC
MAX
= (E ) (E ) dE =
,
N
CC
where (E ) is the neutrino flux, NCC (93807) is the number of charged-current (CC) neutrino
interaction events (used only for the flux determination) detected in the 19961997 run for which
NetScan acquisition is completed, and CC their detection efficiency. / is the ratio
between the average total and CC cross-sections. Efficiencies have been calculated by
Monte Carlo. We define loc as the probability to locate a vertex and sel as the reconstruction
and selection efficiencies for the various decay channels. For each topology, the number Nmax of
events that would be observed if the oscillation probability were equal to 1 is:

loc
sel
BRm m
m
,
Nmax = NMAX
m
where the index m runs over the various decay modes contributing to that topology. Branching
ratios, efficiencies and Nmax are given in Table 4. Note that muonic decays with a misidentified
muon also contribute to the C1 topology.
3.5. Post-scan selection and final data sample
The signal to background ratio, hence the sensitivity to oscillation, can be improved if, in
addition to the topology, other information is used, such as:
the angle between the parent track and the estimated direction of the hadronic shower;
the length of the parent track;
the mean angle of the daughter tracks with respect to the parent track;
when available, the momentum and charge of a daughter particle.
336
The purpose of the post-scan analysis is to maximize the sensitivity to oscillation by an

optimum combination of this information. The optimization was performed on Monte Carlo
simulated samples of signal and background events [18] and led to the subdivision of the samples
into different subclasses, depending on the observed topology.
3.5.1. The C1 topology
For the C1 topology, the transverse momentum with respect to the parent direction is useful
for the rejection of WK background.
Two methods are available for momentum measurements: either with the DT of the hadron
spectrometer or from the multiple Coulomb scattering (MCS) in the emulsion. The DT also
measures the particle charge. The MCS method is used if no DT measurement is available and if
the number of emulsion plates for the MCS measurement is greater than 6. The C1 sample has
thus been divided into three subclasses:
momentum measurement from DT (P DT ) which also yields a charge measurement of the
sign of the charge which is important for the charm background rejection;
momentum measurement from MCS (P MCS );
no momentum measurement.
The first category has the highest sensitivity to search and the strongest background rejection. The events in which the daughter track has a positive charge are rejected. When the
momentum is determined, a cut is applied on the transverse momentum (PT > 250 MeV/c) of
the daughter kink. A momentum-dependent cut on the flight length (LF ) is also applied. Fig. 2
shows a comparison between signal and background in the selected (I) and rejected (II) regions
in the LF versus P plane. All decay vertices beyond 3 mm or with a parent momentum below
2 GeV/c were rejected because white kink interactions are likely to populate these regions. For a
momentum between 2 GeV/c and 5 GeV/c step functions were used for simplicity. The effects
of this selection on the background rejection and efficiency are shown in Table 5.
In addition, the samples have been further subdivided into three ranges, where is
defined as the angle in the transverse plane between the parent track and the mean direction
of the other primary tracks after exclusion of the one which makes the largest angle with the
parent track (the track with a cross in Fig. 3). A comparison between the distributions of in
charm and samples is shown in Fig. 4. The signal is characterized by a larger mean value
of .
3.5.2. The C3 topology
For the C3 topology, the fraction of events for which the three daughter particles have their
momentum measured is too small to lead to a useful discrimination. Instead, use is made of the
fact that the mean angle of the daughter tracks relative to the parent is strongly correlated to
the inverse of the Lorentz boost factor . Thus, the product of the parent flight length LF and
is a good parameter to discriminate particles of different lifetimes (or apparent lifetime in
the case of white interactions). The C3 sample has been divided into short c and long c
subsamples, separated by a cut value LF = 75 m. Fig. 5 shows a comparison of the charm
and samples: the short c sample is enriched in decays. These categories are further divided
as above according to ranges. The sample most sensitive to a signal corresponds to events
with short c and large .
337
Fig. 2. The flight length LF versus daughter particle momentum computed for charm (left), white kink (centre) and
events (right). The momentum is evaluated by DT (top) or MCS (bottom). Only events in region I are selected.
Table 5
Effect of the selection in flight length-momentum plane on the
signal and background, without considering any other cut
Physical process
After rejection of events
WK
Charm
13%
34%
52%
Once the different categories are defined the same subdivision is applied to the candidate
events. This is done according to the blind analysis prescriptions, in order to minimize the
possibility of biases in the experimental result.
Table 6 shows the number of simulated and observed events in C1 and C3 topologies of the
new 0 sample, together with data from the Phase I analysis which consist of two samples [7]:
1 events from the 19941997 data taking. The background reduction is obtained by applying a cut on the transverse momentum (PT > 250 MeV/c) of the kink daughter. The kink must
occur within five plates downstream from the neutrino interaction vertex plate. The estimated
background is 0.1 events and the maximum number of events is Nmax = 5014 (as reported
in Table 2 of [7]). The background is much smaller than for the 0 sample because of the low
probability of a wrong measurement of the charge in the muon spectrometer. No candidate is
observed.
338
Fig. 3. Definition of the angle . The dashed line is the mean direction of the primary tracks, with the exclusion of
the one which has the largest angle relative to the parent particle. The distribution is different for charm, white
interaction or events.
Fig. 4. distributions for simulated 0 C1 events: comparison between (solid line) and charmed events (dashed
line). The area is normalized to 1. The vertical lines show the applied cut dividing the events in three subsamples.
339
Fig. 5. LF distribution for simulated 0 C3 events: comparison between (solid line) and charm (dashed line)
decays. The area is normalized to 1. The vertical line shows the applied cut LF = 75 m dividing the events in two
subsamples.
0 one-prong events from the 19941995 data taking. Table 2 of [7] gives the background
and the maximum number of oscillated events for the whole 0 19941997 sample. The
19961997 data has been reanalysed, as reported in this paper. The 19941995 data correspond
to an estimated background of 0.3 events and a maximum number of oscillated events of
Nmax = 526. No candidate is observed.
The errors on the expected background are evaluated by taking into account the limited statistics of the Monte Carlo sample and the errors on cross-sections and branching ratios.
4. Results
From the number of observed candidates, the expected background and the number of signal
events expected for full oscillation it is possible to compute, for each subsample of Table 6, the
90% C.L. upper limit on the appearance probability using a frequentist statistical approach,
the so-called Feldman and Cousins unified approach [21].
To evaluate the sensitivity of the experiment, we have simulated a large number of experiments taking into account Poisson fluctuations of the expected background. The sensitivity is
then defined as the average of the 90% C.L. upper limits on the appearance probability of all
these experiments.
The last two columns in Table 6 show the sensitivities S of each subsample and an index i
in decreasing order of sensitivity. The subsamples are then combined, with the same prescriptions used by the NOMAD Collaboration [4]. After including the 11th most sensitive subsample,
the global sensitivity changes only at the percent level. Similarly for the e search. In eval-
340
Table 6
The final CHORUS data sample. The first two rows refer to the Phase I analysis, namely to the 1 channel of the whole
data taking (19941997) and to the 0 sample collected in 19941995. The new sample, consisting of the 0 data
collected in 19961997, is divided in C1 and C3 topologies which are further divided in subsamples, as described in
Section 3.5. For each subsample, the following quantities are shown: the expected background; the maximum detectable
number of events, N and Ne respectively from the and e beam components; the number of data events. The
numbers in the last two columns are evaluated for the search (see Section 4) and give the sensitivity S of
each single channel (times 104 ) and an index i which sorts the sensitivities in decreasing order
Background
Ne
1 [19941997 data taking]

0 C1 [19941995 data taking]
0.100 0.025
0.300 0.075
5014
526
55.8
5.85
0
0
53.2 9.0
9621
76.9
59
18.0 2.3
5.00 0.73
6.2 1.4
4.6 1.1
1.20 0.40
3.3 1.0
0.383 0.071
0.087 0.033
0.055 0.012
14.6 1.6
769
708
1406
991
749
1649
546
556
1023
1224
7.67
5.34
13.9
6.75
5.70
12.8
3.62
4.24
7.24
9.77
26
10
7
2
2
3
0
0
0
9
47 11
4443
35.5
48
14.8 5.0
6.4 3.0
1.5 1.5
15.5 5.0
9.8 3.9
1.7 1.5
792
782
1554
386
336
593
6.63
6.12
13.4
2.37
2.40
4.62
17
6
4
8
8
5
Category
(rad)
[0; /2]
[/2; 3/4]
[3/4; ]
Only MCS momentum measured:
[0; /2]
PTMCS > 250 MeV/c
[/2; 3/4]
[3/4; ]
and (P MCS vs. LF ) cut
DT momentum measured (Charge ):
[0; /2]
PTDT > 250 MeV/c
[/2; 3/4]
and (P DT vs. LF ) cut
[3/4; ]
PT < 250 MeV/c or Charge + or in region II
No momentum measured

Short c (< 75 m)
Long c (> 75 m)
[0; /2]
[/2; 3/4]
[3/4; ]
[0; /2]
[/2; 3/4]
[3/4; ]
Data
S
4.9
48
i
1
10
211
66
38
45
34
26
41
38
22
15
12
6
9
5
4
8
7
2
133
89
23
268
237
62
14
13
3
17
16
11
uating the global sensitivity, we thus consider only the 11 most sensitive subsamples for both
searches.
4.1. oscillation
Fig. 6(left) shows the distribution of the upper limits (at 90% C.L.) obtained by simulating
800 experiments. The average value, S = 2.4 104 , corresponds to the CHORUS sensitivity to
appearance.
The 90% C.L. upper limit on the appearance probability obtained from the data is
P ( ) < 2.2 104 ,
indicated by a vertical line in Fig. 6. It is in agreement with the expected sensitivity. In fact, in
absence of signal events the probability to obtain an upper limit of 2.2 104 or lower is 48%,
which means that the number of observed events is compatible with the estimated background.
The above result improves by a factor 1.5 the previously published limit P < 3.4 104 [7].
In a two-neutrino formalism, the above upper limit corresponds to the exclusion region in the
(m2 , sin2 2 ) oscillation parameter plane shown in Fig. 7(left) and to the limit sin2 2 <
4.4 104 for large m2 .
341
Fig. 6. Upper limits obtained at 90% C.L. [21], in absence of signal events, for 800 simulated experiments with the
CHORUS expected background. The average value corresponds to the sensitivity to (left) and e (right)
appearance probability. The vertical line is the 90% C.L. upper limit obtained from the data.
Fig. 7. The CHORUS upper limit on (left) and e (right) oscillation represented in an exclusion plot
in the oscillation parameter plane. CHORUS results are shown as solid lines and are compared with the last results of
NOMAD [4] and CHOOZ [22].
The upper limit obtained by NOMAD [4] is more stringent than the CHORUS one at large
m2 , whereas for m2 lower than 70 eV2 /c4 the upper limit is improved by CHORUS owing
to its higher efficiencies at low neutrino energies.
4.2. e oscillation
The SPS neutrino beam contains a 0.8% e contamination. Assuming that all observable
would originate from this contamination, the above result translates into a limit on the
e appearance probability. The difference in energy between the (E 26 GeV)
and e (Ee 42 GeV) components leads to a different shape of the exclusion plot in the
oscillation parameter plane.
342
Fig. 6(right) shows the upper limit distribution obtained by simulating 800 experiments. The
average value, S = 2.5 102 , corresponds to the sensitivity to the e appearance probability.
The 90% C.L. upper limit on the appearance probability obtained from the data is
P (e ) < 2.2 102 ,
indicated by a vertical line in Fig. 6. In absence of signal events, the probability to obtain an upper
limit of 2.2 102 or lower is 49%. The above result improves by a factor 1.2 the previously
published limit [7]. The improvement is smaller than in the case for two reasons. The
first one is the variation of the expected e contamination provided by a new simulation of the
beam (reduced by a factor 0.85) [23]. Furthermore, in the case the efficiency improved
predominantly at low neutrino energies and the harder e spectrum makes this improvement
less effective. This reason also explains the large difference in the limit relative to NOMAD.
In a two-neutrino formalism, the above upper limit corresponds to the exclusion region in the
(m2 , sin2 2e ) oscillation parameter plane shown in Fig. 7(right) and to the limit sin2 2e <
4.4 102 for large m2 .
5. Conclusions
The final result of the search for appearance with the CHORUS experiment has been presented. A 90% C.L. upper limit of 2.2 104 was set for the appearance probability, improving
by a factor 1.5 the previously published CHORUS result. In a two-neutrino mixing scheme, this
result corresponds to the limit sin2 2 < 4.4 104 for large m2 . With respect to previous CHORUS results, we have also improved by a factor 1.2 the upper limit on the (e )
oscillation probability.
Acknowledgements
We gratefully acknowledge the help and support of the neutrino beam staff and of the numerous technical collaborators who contributed to the detector construction, operation, emulsion
pouring, development, and scanning. The experiment has been made possible by grants from
the Institut Interuniversitaire des Sciences Nuclaires and the Interuniversitair Instituut voor
Kernwetenschappen (Belgium), the Israel Science Foundation (grant 328/94) and the Technion
Vice President Fund for the Promotion of Research (Israel), CERN (Geneva, Switzerland), the
German Bundesministerium fr Bildung und Forschung (contract numbers 05 6BU11P and 05
7MS12P) (Germany), the Institute of Theoretical and Experimental Physics (Moscow, Russia),
the Istituto Nazionale di Fisica Nucleare (Italy), the Promotion and Mutual Aid Corporation for
Private Schools of Japan and Japan Society for the Promotion of Science (Japan), the Korea
Research Foundation Grant (KRF-2003-005-C00014) (Republic of Korea), the Foundation for
Fundamental Research on Matter FOM and the National Scientific Research Organization NWO
(The Netherlands), and the Scientific and Technical Research Council of Turkey (Turkey). We
gratefully acknowledge their support.
References
[1] H. Harari, Phys. Lett. B 216 (1989) 413.
[2] J. Ellis, J.L. Lopez, D.V. Nanopoulos, Phys. Lett. B 292 (1992) 189.
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
343
E.H.M. Heijne, CERN Yellow Report 83-06 (1983).

P. Astier, et al., NOMAD Collaboration, Nucl. Phys. B 611 (2001) 3.
E. Eskut, et al., CHORUS Collaboration, Phys. Lett. B 424 (1998) 202.
B. Van de Vyver, P. Zucchelli, Nucl. Instrum. Methods A 385 (1997) 91.
E. Eskut, et al., CHORUS Collaboration, Nucl. Instrum. Methods A 401 (1997) 7.
S. Aoki, et al., Nucl. Instrum. Methods B 51 (1990) 466.
T. Nakano, PhD thesis, Nagoya University, Japan, 1997.
T. Nakano, in: Proceedings of the International Europhysics Conference on High Energy Physics, Budapest, Hungary, 2001.
M. Gler, PhD thesis, METU, Ankara, Turkey, 2000.
G. Onengut, et al., CHORUS Collaboration, Phys. Lett. B 613 (2005) 105.
G. Onengut, et al., CHORUS Collaboration, Phys. Lett. B 614 (2005) 155.
K. Kodama, et al., Nucl. Instrum. Methods A 493 (2002) 45.
N. Nonaka, PhD thesis, Nagoya University, Japan, 2002.
L. Scotto Lavina, PhD thesis, Universit degli Studi di Napoli Federico II, 2005.
A. Kayis-Topaksu, et al., CHORUS Collaboration, Phys. Lett. B 626 (2005) 24.
A. Satta, PhD thesis, University of Rome, 2001.
G.J. Feldman, R.D. Cousins, Phys. Rev. D 57 (1998) 3873.
M. Apollonio, et al., CHOOZ Collaboration, Eur. Phys. J. C 27 (2003) 331.
I.I. Tsukerman, CHORUS internal note 2000010 (2002).

Percolation of vortices
in the 3D Abelian lattice Higgs model
Sandro Wenzel a , Elmar Bittner a ,
Wolfhard Janke a, , Adriaan M.J. Schakel b
a Institut fr Theoretische Physik and Centre for Theoretical Sciences (NTZ), Universitt Leipzig,
Postfach 100 920, D-04009 Leipzig, Germany

b Institut fr Theoretische Physik, Freie Universitt Berlin, Arnimallee 14, D-14195 Berlin, Germany
Received 7 August 2007; received in revised form 17 October 2007; accepted 19 October 2007
Available online 6 November 2007
Abstract
The compact Abelian Higgs model is simulated on a cubic lattice where it possesses vortex lines and
pointlike magnetic monopoles as topological defects. The focus of this high-precision Monte Carlo study is
on the vortex network, which is investigated by means of percolation observables. In the region of the phase
diagram where the Higgs and confinement phases are separated by a first-order transition, it is shown that
the vortices percolate right at the phase boundary, and that the first-order nature of the transition is reflected
by the network. In the crossover region, where the phase boundary ceases to be first order, the vortices
are shown to still percolate. In contrast to other observables, the percolation observables show finite-size
scaling. The exponents characterizing the critical behavior of the vortices in this region are shown to fall in
the random percolation universality class.
PACS: 11.15.Ha; 12.38.Aw; 64.60.ah; 87.10.Rt
Keywords: Compact Abelian Higgs model; Monte Carlo simulations; Vortex network; Percolation
1. Introduction
The Abelian Higgs model with a compact gauge field formulated on a three-dimensional (3D)
lattice possesses an intriguing phase structure [13]. In addition to the Higgs state where the
E-mail address: janke@itp.uni-leipzig.de (W. Janke).

doi:10.1016/j.nuclphysb.2007.10.024
S. Wenzel et al. / Nuclear Physics B 793 (2008) 344361
345
photon acquires a mass, it exhibits a state in which electric charges are confined. The richness
of the model, which serves as a toy model for quark confinement, stems from the presence of
two types of topological excitations, viz. vortex lines and magnetic monopoles. The latter are
point defects in three dimensions which arise because of the compactness of the U(1) gauge
group. In the pure 3D compact Abelian gauge theory, monopoles are known to form a plasma
that physically causes confinement of electric charges for all values of the inverse gauge coupling
[4]. Being the only parameter present, the pure model therefore possesses only a confinement
phase. The coupling to the scalar theory preserves this confinement state and gives in addition
rise to a Higgs state. For sufficiently small values of the Higgs self-coupling parameter , the
two ground states are separated by a first-order transition [5,6] as sketched in Fig. 1. For
larger than a critical value c , which depends on the value of the gauge coupling, the two states
are no longer separated by a phase boundary across which thermodynamic observables become
singular, as was first shown by Fradkin and Shenker [1] in the limit where fluctuations
in the amplitude of the Higgs field become completely suppressed. In other words, it is always
possible to cross over from one ground state to the other without encountering a thermodynamic
singularity. Because of this, the Higgs and confinement states were thought to constitute a single
phase, despite profound differences in physical properties.
This conclusion is supported by symmetry arguments [7]. The relevant global symmetry group
of the compact 3D Abelian Higgs model (cAHM) is the cyclic group Zq of q elements, where
the integer q denotes the electric charge of the Higgs field. For the doubly charged case (q = 2),
the relevant symmetry group is Z2 , which is in agreement with the known result that the model
undergoes a continuous phase transition belonging to the 3D Ising universality class [1,8]. For
q = 1, this argument excludes a phase transition characterized by a local order parameter in
the spirit of Landau because the group Z1 , which consists of only the unit element, cannot be
spontaneously broken. Stated differently, there exist no local order parameter which distinguishes
the Higgs from the confinement state.
In Ref. [9], we argued that the phase diagram is more refined than implied by this picture.
We conjectured that although analytically connected, the two ground states can be considered
as two distinct phases. The nature of the phase boundary is intimately connected to the distinct
physical properties of the Higgs and confinement phase. For the latter phase to confine electric
charges, the monopoles must form a plasma. This in turn can only happen when the line tension
of the vortex lines, or flux tubes, connecting monopoles and antimonopoles vanishes, so that
they are no longer tightly bound in pairs as in the Higgs phase (for typical configuration plots see
the snapshots in Fig. 1). Since the vortex line tension is finite in the Higgs phase and zero in the
confinement phase, we argued that the phase boundary is uniquely defined by the vanishing of the
vortex line tension, irrespective of the order of the phase transition. The confinement mechanism
operating in the 3D cAHM is essentially the dual superconductor scenario [10,11].
In addition to open vortex lines, each having a monopole and an antimonopole at its endpoints,
the system also possesses closed vortex lines. These are expected to be characterized by the
same line tension as the open lines. Because of the finite line tension, large vortex loops are
exponentially suppressed in the Higgs phase. Upon approaching the phase boundary, the line
tension becomes smaller so that the vortex network can grow larger and the overall line density
increases. Finally, at the phase boundary where the line tension vanishes, vortices can grow
arbitrarily large at no energy cost. The phase boundary between the Higgs and confinement
phase is therefore expected to be marked by a proliferation of (open and closed) vortex lines, as
was first proposed by Einhorn and Savit [12]. The vortices proliferate both in the region where
the transition is first-order and in the region where it is not. A line along which geometrical
346
Fig. 1. Sketch of the phase diagram of the Abelian lattice Higgs model as a function of the Higgs self-coupling parameter and the hopping parameter at fixed inverse gauge coupling . For < c the transition from the confined
to the Higgs phase is of first-order. For > c ordinary observables show no singular behavior and signal a crossover
between the two ground states indicated by the dark region emanating from the point where the first-order transition line
terminates.
objects proliferate, yet thermodynamic quantities and other local gauge-invariant observables
remain nonsingular has become known as a Kertsz line [13]. Such a line was first discussed in
the context of spin clusters in the 2D Ising model in an applied magnetic field.
The same deconfinement transition driven by the proliferation of vortices with a phase boundary consisting of a first-order line which ends in a critical point and then continues as a Kertsz
line was originally proposed by Langfeld for the SU(2) counterpart of the Abelian Higgs model
defined on a 4D lattice [14]. The relevant vortices, forming surfaces in 4D, are center vortices
which carry a flux related to the nontrivial element of the Z2 center of the SU(2) gauge group.
The presence of the Kertsz line in this non-Abelian model has been further numerically investigated and confirmed in Ref. [15]. While the concept of a Kertsz line was originally introduced in
the context of lattice gauge theories in Ref. [16], the interpretation of a deconfinement transition
as driven by percolating vortices was previously put forward in the context of the SU(2) lattice
Higgs model [1719], and the pure SU(2) lattice gauge theory [2022].
The purpose of this Monte Carlo study is to investigate the vortex proliferation scenario suggested in Refs. [9,23] by studying the behavior of the vortex network directly. Because vortices
are geometrical objects, their analysis is amenable to the methods developed in percolation theory [24]. We conjectured in Refs. [9,23] that along the Kertsz line, percolation observables have
the usual percolation exponents. In addition, we expect that the vortex network displays discontinuous behavior in the region where the phase boundary consists of a first-order transition. For
other recent Monte Carlo studies focusing on various aspects of the model see Refs. [2527] and
references therein.
A similar high-precision Monte Carlo study of the behavior of a vortex network was recently
carried out of the 3D XY and the ||4 lattice model [28,29], respectively. The latter model, whose
critical temperature on a cubic lattice is known to high precision, corresponds to taking
in the Higgs model so that the gauge fields become completely ordered. An important observation made in that study was that the overall vortex line density behaves similar to the energy,
and the associated susceptibility similar to the specific heat. The vortex percolation threshold
estimated through finite-size scaling analysis of the overall line density data was found to be
347
perfectly consistent with the critical temperature. However, estimates based on any of the vortex percolation observables used, while being close to that temperature, never coincided with it
within error bars. Possibly, the mismatches are related to the way the vortex networks are traced
out, or to the absence of a stochastic element as present in, for example, the FortuinKasteleyn
definition of spin clusters in the Potts model [30]. In any case, since we use the same percolation
observables and the same (imperfect) rules to trace out vortex networks, we expect to find a similar mismatch, at least in the region where the transition is no longer discontinuous. To remind
the reader of these qualifications, we will sometimes refer to the estimated percolation threshold
as apparent percolation threshold.
The rest of the paper is organized as follows. In the next section, the observables used to investigate the model are introduced. In Section 3, the simulation methods as well as the numerical
tools used to analyze the data are discussed. In that section, also a new tool to effectively collapse
data gathered on lattices of different size is presented. Sections 4 and 5 contain the Monte Carlo
results in the vicinity of the phase boundary for the two regions where it consists of a first-order
transition and a Kertsz line, respectively. In Section 6, the dependence of the location of the percolation threshold is investigated as a function of the parameters of the model. Finally, Section 7
contains a discussion of the Monte Carlo results and our conclusions.
2. Definitions and observables
The Abelian lattice Higgs model with compact gauge field at the absolute zero of temperature
is defined by the action S = Sg + S , with the gauge part

1 cos (x) ,
Sg =
(1)
x,<
where is the inverse gauge coupling parameter, = 1/aq 2 with a the lattice spacing and q the
electric charge of the Higgs field. We exclusively consider the case q = 1. The doubly charged
Higgs field (q = 2), which has an even richer topological structure than the q = 1 case, has
recently been investigated in Ref. [31], where it was found that the monopoles form chains. The
sum in Eq. (1) extends over all lattice sites x and lattice directions , and (x) denotes the
usual plaquette variable (x) = (x) (x) with the lattice derivative (x)
(x + ) (x) and the compact link variable (x) < . The matter part S of the
lattice action is given by

2

2
S =
(x) + 2 (x) 1 , (2)
(x)(x + ) cos (x) q (x) +
x,
where polar coordinates are chosen to represent the complex Higgs field (x) = (x)ei(x) , with
the compact phase (x) < . In Eq. (2), is the hopping parameter, and the Higgs selfcoupling. We study the theory on a cubic lattice, which either is taken to represent a 3D space
or spacetime box, depending on whether one of the dimensions of the lattice is interpreted as
(Euclidean) time.
In addition to measuring field observables such as the total action or energy S, the hopping
energy
Eh

1
(x)(x + ) cos (x) q (x) ,
3
L x,
(3)
348
the square Higgs amplitude, and coslink energy

1
cos (x) q (x) ,
C 3
3L x,
(4)
we especially probe for topological excitations. A gauge invariant vortex line segment j (x)
pointing in the = 1, 2, 3 direction is given by
j (x) = l (x) n (x),
(5)
where l (x) is the integer-valued field related to the electric current along the links of the lattice
via

1
(x) q (x) (x) q (x) 2 ,
l (x) =
(6)
2
and n (x) measures the multiples of 2 with which the plaquette variable is shifted away from
the interval [, ):

1
(x) (x) 2 .
(7)
2
Here, we use the usual modulo operation [a]2 2n which subtracts n multiples of 2
from the variable such that [a]2 takes values in the interval [, ). While l (x) measures the
quantized vorticity, n (x) gives the number of elementary Dirac strings piercing the plaquette
with its normal pointing in the direction.
Monopoles are detected by taking the divergence of j , m(x) = j (x), where m(x) takes
on integer values only. Using these definitions, we record the vortex line and monopole densities

1
1
j (x),
M 3
m(x).
v 3
(8)
L x,
L x
n (x) =
As already mentioned in the Introduction, in the ||4 theory, v behaves similar to the energy and
the associated susceptibility similar to the specific heat [29]. A short summary of results for the
observables not involving vortices can be found in Ref. [9].
The main focus in this paper is on the vortex networks formed by individual vortex lines (see
Fig. 2). Tracing out such a network is ambiguous. We restrict ourselves to the simplest convention by defining a vortex line, which can be either open or closed, as a set of connected vortex
line segments. Line segments are said to be connected if they enter or leave the same lattice
cube. With this convention, four or six line segments entering and leaving a single cube are not
further resolved into separate vortices, but are lumped together into one, self-intersecting vortex.
The size n of a vortex is just the number of links forming the vortex. Vortex line segments with
|j | 2 are not split into distinct segments and are only counted once in the vortex density. The
vortex networks found with the help of these tracing rules are therefore only an approximation
to the true networks. And the observed, or apparent percolation threshold will in general differ
from the true one. We trace out the vortex networks by using a recursive algorithm and employ
standard percolation observables to probe them [24]. Specifically, we determine the percolation
probability P1D by recording when a vortex spans the lattice in an arbitrary direction. This definition is such that a vortex is already said to percolate when it spans the lattice in just one, arbitrary
direction. We also determine the probability P3D that a vortex spans the lattice in all directions,
and the percolation strength P , defined as the size of the largest vortex per lattice site. We have
in addition considered the average vortex size, but this data is not well suited for analyses as we
349
Fig. 2. Networks formed by vortex lines. The small spheres represent monopoles and antimonopoles. Left: Higgs phase,
characterized by a few short vortex lines which can either be closed or open. Right: Confinement phase, characterized by
a large percolating vortex network.
also found in previous studies on simpler models [32]. Finally, we use the susceptibility

(O) = O2 O2 L3
(9)
(without the L3 for P1D and P3D ) and the Binder parameter
B(O) = 1
O4
3O2 2
(10)
of an observable O to probe the phase boundary.

3. Simulation and data analysis techniques
3.1. Monte Carlo
A variety of Monte Carlo methods are applied to efficiently simulate the system in the different parts of the phase diagram. In the first-order transition region, primarily the multicanonical
algorithm (MUCA) [33] is used with a weight iteration as described in Ref. [34]. Weights are
rendered flat in the hopping term (3) of the action to enhance tunneling. In addition to local updates, the Higgs amplitude and the gauge angles are also updated globally [35] to allow for larger
jumps in phase space and thus for shorter tunneling times between the two metastable states. In
the vicinity of the Kertsz line, the gauge fields are updated using the Metropolis algorithm,
while the Higgs field is updated by means of heat-bath and overrelexation algorithms [36].
Measurements are typically taken after each sweep of the lattice and the entire time series
is recorded to allow for error analysis and post-simulation data processing. Table 1 provides an
overview of parameters used in the simulations. An integral part of our analysis are Ferrenberg
Swendsen reweighting techniques [37,38]. These techniques are applied in two distinct flavours.
350
Table 1
Number of sweeps of the lattice of linear size L and the number of different values of considered for each lattice used
in the simulations
Sweeps
Different values
1.1
0.2
0.75
0.025
10, 12, 14, 16, 18, 20, 22, 24, 26, 30, 32, 42
8, 10, 12, 16, 20
8, 10, 12, 14, 16, 18, 20
105 106
105
107
20
20
1 (MUCA)
2.0
0.2
8, 10, 12, 14, 16, 20, 24, 28, 36
106
10
Fig. 3. Data points for the susceptibilities (P1D ) and (C) (multiplied by 400) for = 2.0, = 0.2 and different values
of and their interpolating lines obtained through reweighting. The measurements are taken on a cubic lattice of linear
size L = 36.
In the Kertsz region, we use the multihistogram reweighting form, which means that we combine simulations at different parameters in an optimal way. Standard reweighting with MUCA
weights is applied in the first-order transition region. To apply these techniques in the best possible way we use optimization routines such as the Brent method [39,40] to search for maxima
in susceptibilities or crossing points of Binder parameters. The use of the Jackknife method [41]
on top of these methods allows for error estimation of the results so determined.
As an example of this approach, we show in Fig. 3 the susceptibilities of the percolation
probability P1D and the coslink energy C. Note that the peak height of (P1D ) is exactly 1/4,
2 = P and hence (P ) = P (1 P ) is maximal at P = 1/2. The
since P1D
1D
1D
1D
1D
1D
points in the plot correspond to individual simulations and the lines through the data points are
obtained by reweighting. Different lines in the figure correspond to having different Jackknife
blocks (but the same block in each time series) omitted. Notice that the peak in the percolation
susceptibility is well defined whereas the peak in the coslink susceptibility is rather broad. The
flatness of the peak is reflected in larger error bars on the estimated peak location. We have
carefully checked our methods against known results for limiting cases such as the London limit
, where fluctuations in the Higgs amplitude are completely frozen, and the XY model,
obtained by setting in addition = . A good check whether vortex lines and monopoles are
correctly identified is to perform a gauge transformation under which their locations are invariant.
351
3.2. Finite-size scaling methods

We estimate critical exponents through finite-size scaling analyses. The scaling Ansatz for an
observable O states that in the vicinity of a continuous phase transition, the data obtained for
different values of the tuning parameter and for different lattice sizes L fall onto a (weakly)
universal scaling function fO defined through

OL () = LO / fO tL1/ , t ( c )/c .
(11)
Here, c denotes the critical point, the correlation length exponent, and O is the critical
exponent characterizing the observable O. The variable t is the reduced coupling and measures
the distance from the critical point.
Data collapse is usually a good check whether the right critical exponents and critical couplings are found by other means. With the correct values, the measured data should fall on the
universal curve given by Eq. (11). Here, we reverse this idea. Starting from an initial guess for the
critical exponents and couplings, we compute the rescaled observable O L (x) = OL ()LO /
with x = tL1/ and judge the quality of the collapse in the interval [xmin , xmax ] by introducing
the weight function
xmax

dx O L2 (x) {O L }2 (x) ,
O
2
(12)
xmin
with
OL (x) L OL (x)/nL and nL the number of different lattice sizes2 included in the sum
.
A
perfect
collapse
in the window [xmin , xmax ] would correspond to O = 0, whereas a bad
L
2 . This approach qualifies for implementation as an automated method
collapse has a large O
2 by adjusting the values of the
when combined with optimization algorithms to minimize O
critical parameters. The method requires that data in each point in the interval [xmin , xmax ] of the
rescaled x-axis be compared, most of which has not been measured. A possibility is to interpolate
between data points by using a polynomial expansion of the function fO and to fit its coefficients
2 . We use a different approach in that
to the data OL [42]. The fit then allows for calculating O
we apply the standard reweighting techniques mentioned above to calculate OL at every point
and hence O L at every point x. We feel that this approach is more natural and does not add
further degrees of freedom to the process.
In practice our implementation is as follows. We first reweight the measured data on all system
sizes. We then start with initial values for the exponents and couplings, and apply a minimization
2 is minimized.
algorithm, such as the simplex method [39,43], that varies the parameters until O
Since the algorithm can become locked in a local minimum, the procedure must be repeated
for many starting points. We find that for our purposes, where only three parameters need to be
determined, the routine gives reliable and consistent results. To obtain error estimates, we apply
the Jackknife method on top of the whole process. The effect of correction terms to scaling is
minimized by repeating the procedure for increasingly smaller intervals [xmin , xmax ] around the
point where collapse is attempted. We expect our tool to be useful in other simulation studies as
well. An idea similar to our implementation was recently presented in Ref. [44].
4. First-order transition region
We start our numerical study by investigating the behavior of the vortex network in the firstorder transition region of the phase diagram sketched in Fig. 1. We choose to simulate at =
352
Fig. 4. Left: Time series of a MUCA simulation showing the hopping energy Eh (upper curve) and percolation strength
P (lower data points) for L = 16. The two observables are seen to be intimately connected. To fit in one plot, the
hopping energy is shifted upwards. Right: Correlation histogram of the number of configurations without a percolating
network over hopping energy Eh for system sizes L = 10 and 16. Note the logarithmic scale used on the vertical axis.
1.1 and = 0.025 where the first-order transition is strong enough already on small lattices
so that time consuming simulations on larger systems are not needed. The left plot in Fig. 4
shows time series of the hopping energy Eh and the percolation strength normalized by the
volume as measured in a typical MUCA simulation [33]. Changes in the energy, which reflect the
more or less random walk through the metastable region, are seen to occur jointly with changes
in the percolation strength. The right plot shows the correlation histogram of the number of
configurations without a percolating network versus hopping energy Eh . Larger negative energies
are seen to strongly correlate with the absence of a percolating vortex network while smaller
negative energies strongly correlate with the presence of such a network.
The central thesis we put forward in Ref. [9] is that the phase diagram of the model can
be understood in terms of proliferating vortices. In the first-order region, the location of the
phase transition has been determined to high precision with the help of observables not involving
vortices such as the hopping energy. To determine the location of the percolation threshold in
the infinite-volume limit, we consider the vortex percolation probability P1D and percolation
strength P , and analyze the scaling of the locations per (L) of the maxima in the associated
susceptibilities and Binder parameters with lattice size L. At a first-order transition, per (L) is
expected to scale as

per (L) = per + cL3 + O L6 ,
(13)
with c a constant. We reweight the MUCA time series to obtain the susceptibilities and Binder
parameters in the vicinity of per (L), and use the methods presented above to search for the
peak locations and its error bars. Fig. 5 summarizes our results. It shows the scaling of the percolation thresholds per (L) obtained from the percolation probability P1D and the percolation
strength P together with the critical points c (L) obtained from the hopping energy Eh and the
monopole density M. The per (L)s obtained from the network observables are seen to strongly
depend on the system size, while the c (L)s obtained from (M) are almost independent of
volume. The results for (Eh ) are not included in Fig. 5 as they cannot be distinguished from
those for M on the scale of the figure. For lattice sizes larger than L = 12, all curves converge to
the same point within error bars. This brings us to the important conclusion that the vortex percolation threshold is located precisely at the first-order transition point. Our best estimate for the
353
Fig. 5. Left: Binder parameters B(Eh ) (upper curves) and B(P ) (lower curves) as a function of in the first-order
transition region at = 1.1, = 0.025 measured on lattices of linear size L = 10, 12, 14, 16, 20 (from broad to sharp
peaks). The Binder parameter B(P ) has been divided by 100 so that both parameters fit into one plot. Right: Scaling
plots of the extrema of the various Binder parameters and susceptibilities. The Binder parameters show larger finite-size
corrections. All curves converge within error bars to the same infinite-volume value of per = c = 0.424570(3).
Table 2
Results of fitting the data in Fig. 5 to Eq. (13). The fits starting at L = 8 and L = 10 also include the correction term
1/L6 . Those starting at L = 12 include only the leading term 1/L3 , except for B(P ), where the correction term is still
needed to obtain reasonable results. The number in square brackets denotes 2 per degree of freedom (DOF), 2 /DOF.
Results are grouped according to percolation and non-percolation observables
Observable
L8
L 10
L 12
(P1D )
(P )
B(P )
0.424569(5) [1.6]
0.424572(6) [1.9]
0.424608(10) [10.7]
0.424569(5) [2.0]
0.424572(5) [2.2]
0.424608(10) [13.1]
0.424570(6) [2.3]
0.424573(6) [2.5]
0.424577(6) [1.1]
(M)
(Eh )
B(M)
B(Eh )
0.424567(3) [2.4]
0.424564(5) [2.52]
0.424558(5) [2.6]
0.424565(6) [3.3]
0.424566(5) [3.1]
0.424564(6) [3.1]
0.424558(8) [3.5]
0.424568(5) [3.9]
0.424567(6) [4.2]
0.424564(6) [3.2]
0.424555(5) [3.2]
0.424566(6) [4.0]
transition point, based on both percolation and non-percolation observables, is c = 0.424570(3)

which is determined from the average of the fit results summarized in Table 2. Averaging the estimates for percolation and non-percolation observables separately, we obtain per = 0.424572(4)
and c = 0.424565(4), respectively. The absence of a mismatch between the critical temperature
and the vortex percolation threshold, as was found in the ||4 theory using the same percolation observables [29], is probably because the phase transition is discontinuous here. An abrupt
change in the ground state apparently washes out any inaccuracy caused by an imperfect tracing
out of the vortex network. To establish in an unbiased fashion that the vortex network reflects
the first-order nature of the transition at = 1.1 and = 0.025, we assume nothing about the
transition and apply standard finite-size scaling to the vortex percolation probability P1D as for a
continuous transition. According to Eq. (13), we then expect to find per = 1/d = 1/3. To obtain
an estimate for the percolation threshold per on the infinite lattice and per , we reweight the data
points from the MUCA time series and subsequently apply our collapsing routine with per and
per as free parameters. Fig. 6 shows the input and the result of this procedure. The extracted
values, per = 0.424568(6), per = 0.325(10), are perfectly consistent with a first-order transition precisely at the expected location. For comparison, Fig. 6 also shows the hopping energy
354
Fig. 6. Left: Reweighted hopping energy Eh and percolation probability P1D for various lattice sizes as a function of
the hopping parameter close to the first-order phase transition. The crossing points should provide a second method to
estimate the transition point. The estimates obtained using the two observables agree within error bars. Right: Collapse
of the P1D data with per = 0.424568, per = 0.325. An optimal collapse of the Eh data (not shown) is achieved for
c = 0.424563 and = 0.338.
which is seen to display the same behavior as the percolation probability. Repeating this procedure for the hopping energy data, we obtain c = 0.424563(6) and = 0.338(10). Both these
estimates based on the data collapse analysis are perfectly consistent with the previous estimates
from finite-size scaling. The raw data in Fig. 6 suggest as in [45] that the crossing points of the
curves measured on lattices of different size provide a second method to estimate the transition
point.
5. Kertsz line
We continue our analysis of the vortex network in the region where the transition ceases to
be of first order. In Ref. [9], we postulated that in this part of the phase diagram the Higgs
and confinement phases are separated by a Kertsz line. Along this line vortices proliferate, yet
thermodynamic quantities remain nonsingular across it. The conjecture is based on the numerical
observation that in the Higgs phase, the monopoles are tightly bound in monopoleantimonopole
pairs. The magnetic flux emanating from a monopole is squeezed into a magnetic flux tube (vortex) which ends on an antimonopole. The finite line tension forces the vortex lines to be short. In
the confinement phase, the monopoles are no longer bound in pairs, but form a plasma. For this to
arise, the vortex line tension must vanish. Vortex lines, both open and closed, can then grow arbitrarily long at no energy cost and proliferate. To facilitate comparison with our previous work [9],
we choose the parameters = 1.1 and = 0.2. Fig. 7 (left) shows our results for the vortex line
density as a function of the hopping parameter for lattices of linear size varying from L = 12
to L = 26. The inset gives the corresponding susceptibilities, which display the same remarkable
behavior first observed in the London limit for other observables [8,25]. Namely, the
susceptibility data obtained on lattices of different sizes is seen to collapse without rescaling.
In particular, the maxima of the susceptibilities do not show any finite-size scaling. From these
data we obtain the estimate = 0.5615(7) for the location of the phase boundary. Applying the
same analysis to the hopping energy, we arrive at the estimate = 0.5655(7) which is close, but
does not agree within error bars with the previous estimate. To further investigate this issue, we
consider the monopole density to see if it leads to the same estimate as does the hopping energy.
The results of our initial study [9] indicated that both estimates agree within error bars. However,
355
Fig. 7. Left: Vortex line density and susceptibility (inset) as a function of the hopping parameter for systems of linear
size L = 12, 16, 18, 26, and = 1.1, = 0.2. Right: The susceptibility (P ) as a function of the hopping parameter
for systems of linear size L = 22, 26, 32, 42.
the high-precision Monte Carlo simulations carried out in the present work show that this is not
the case for the monopole density yields = 0.5639(1). In evaluating these results, it should
be kept in mind that none of these susceptibilities diverge in the infinite-volume limit, so that
small discrepancies were to be expected. The change in the ground state is for this reason usually
referred to as a crossover (). The discrepancy between the peak locations of these observables
is shown in Fig. 9.
In Ref. [9], we argued that the phase diagram is more refined than just showing a crossover
between the Higgs and confinement ground states in that a sharp boundary between the two
phases does exist in the form of a Kertsz line across which the vortices proliferate. Moreover,
as we argued partly on the basis of symmetry, the percolation observables in the vicinity of the
Kertsz line should be characterized by the usual percolation exponents. Unlike the observables
previously studied, these observables are expected to show finite-size scaling. Fig. 7 (right) shows
the susceptibility (P ) of the percolation strength as a function of the hopping parameter for
lattices of linear size varying from L = 22 to L = 42. It is indeed observed that this percolation
observable depends on the lattice size even though we considered system sizes larger than those
for which the other observables already reached the infinite-volume limit. That is, in contrast to
the other observables, percolation observables allow for a precise location of the phase boundary.
To estimate the percolation exponents, we study the behavior of the percolation probability
P1D and P3D as well as the percolation strength P in the vicinity of the percolation threshold
per (see Fig. 8). On the infinite lattice, the percolation strength vanishes on approaching the
threshold as P ( per )per , while (P ) diverges as (P ) | per |per . Finally, the
correlation length per , which provides a typical length scale of the vortex network, diverges as
per | per |per . Given the discrepancy found in the context of the ||4 theory, we do not
expect the estimate of the percolation threshold using percolation observables to coincide with
the one based on the vortex line density. We therefore determine the location of the percolation
threshold anew together with the exponent per by studying the finite-size behavior of the susceptibilities of percolation observables. Fig. 9 shows the scaling of the locations of the susceptibility
maxima with 1/L. The observables considered are the percolation probabilities P1D and P3D ,
and the percolation strength P . For comparison, also the data for the hopping energy, which
was used in Ref. [9] to estimate the location of the phase boundary, and the data for the monopole
and vortex densities are included. The percolation data in Fig. 9 are fitted to the function
356
Fig. 8. Percolation probability P1D and percolation strength P as a function of the hopping parameter for systems of
linear size L = 22, 26, 30, 42 at = 1.1, = 0.2.
Fig. 9. Locations of the susceptibility maxima of the percolation probabilities P1D and P3D , and the percolation strength
P as a function of 1/L for = 1.1. The dashed line is a fit of max (P ) to Eq. (14). The horizontal line is through
the corresponding data for the vortex density, while the two sets of data points above this line pertain to the monopole
density and hopping energy.
per (L) = per + cL1/per ,
(14)
using the standard least-squares method. Unfortunately, this approach does not give reliable and
consistent values for per and per when repeated for the different observables and lattice sizes.
The most stable results are obtained from the susceptibility of P , giving per = 0.54141(18)
and per = 0.87(6). Results for P1D and P3D depend too much on the fitting regime and we
conclude that Eq. (14) for these observables is only fulfilled for large lattice sizes. As expected,
the estimate of the location of the percolation threshold does not agree with obtained from
the vortex line density. To improve these estimates, we now apply our collapsing routine, rather
than carrying out additional simulations for other lattice sizes. The method has the advantage that
much more data is used as input for not only the locations of the peak maxima but all the data in
the vicinity as well as the interpolated values obtained from reweighting are included. We take
the Binder parameter B(P ) of the percolation strength together with the scaling Ansatz

B(P ) = h tL1/per , t ( per )/per ,
(15)
where h is a scaling function. Fig. 10 shows the raw data together with the data collapse for
per = 0.54145 and per = 0.881. The good quality of the data collapse is apparent from the 3D
357
Fig. 10. Top left: Binder parameter of the percolation strength P as a function of the hopping parameter . Top right:
Collapse of the data for per = 0.54145 and per = 0.881. The data is obtained on lattices of linear size L = 22, 26, 30, 42.
Bottom: Landscape of the quality of the data collapse over a range of values of per and per . The peak corresponds to
the best collapse and yields our estimates per = 0.54145(2) and per = 0.881(2).
Table 3
Critical exponents of the percolating vortex network across the Kertsz line for = 1.1 and = 0.2 compared with the
standard percolation exponents
Model
per
per
per
per
cAHM
Percolation [46] (see also [47,48])
0.54145(2)
0.881(2)
0.8765(16)
0.43(2)
0.4522(8)
1.76(2)
1.7933(85)
plot in the same figure. Given these estimates for per and per , we next apply the collapsing
routine to determine the exponents per and per from the scaling relations

P = Lper /per f tL1/per ,

(P ) = Lper /per g tL1/per ,
(16)
with f and g scaling functions. Table 3 summarizes our results and compares them with the
random percolation exponents. In agreement with our conjecture [9], the estimates, which as
shown in Fig. 11 lead to a good data collapse, are consistent with standard percolation exponents.
Notice that, in contrast to the findings in the ||4 theory, the vortices appear to proliferate in the
confinement phase after passing through the crossover region (per < ). We expect this to be
related to the presence of monopoles which can act as sources for the vortices. To assess this, we
study in the next section how the relative positions of the percolation threshold and the crossover
region vary with changes in the parameters and .
358
Fig. 11. Left: Collapse of the percolation strength P yielding an exponent per = 0.43(2). Right: The same for the
susceptibility (P ) yielding per = 1.76(2).
6. Fine tuning
Since the vertex percolation threshold per and the phase boundary c coincide in the firstorder transition region, it is expected that as we approach this region, the discrepancy found
in the previous section at = 1.1 and = 0.2 becomes smaller. To verify this, we repeat our
analysis at = 1.1 and = 0.075, which is closer to the first-order transition region. We find that
the difference between the locations of the percolation threshold and the crossover determined
using the overall vortex line susceptibility (v) indeed becomes smaller, changing from about
0.018 at = 0.2 to 0.007 here. Moving in the opposite direction of increasing , we observe
the discrepancy to increase, becoming as large as about 0.1 in the London limit . The
percolation threshold still appears in the confinement phase after passing through the crossover
region. That is, changes in the Higgs self-coupling seem to leave the relative positions of the
crossover and the apparent vortex proliferation threshold unchanged.
We next vary and set = 2.0 and = 0.2. By increasing , one suppresses the monopoles.
They completely disappear in the limit , where the theory looses its compactness. Unfortunately, simulations at = 2.0, = 0.2 are computationally much more challenging than
at = 1.1, = 0.2 as autocorrelation times are much longer. We therefore restrict ourselves to
lattice sizes up to L = 36. Moreover, observables other than percolation observables become less
useful as can be seen from our example in Fig. 3. Whereas the susceptibility of the percolation
probability has a pronounced peak and a clear maximum, the maximum in the coslink susceptibility (C) would be difficult to identify without reweighting. The large error bars obtained for
the peak location of (C) reflect the absence of a pronounced peak.
Our main conclusion of the simulations at = 2.0 and = 0.2 is that the locations of the
apparent percolation threshold and the crossover determined using the overall vortex line susceptibility (v) have changed relative positions. This conclusion is based on Fig. 12 in which
per (L), estimated using different percolation observables, is plotted as a function of 1/L to see
their tendency for L . The first observation is that the data points are much closer to each
other than was the case at = 1.1 (see Fig. 9). As before, per (L) obtained from the susceptibility of the percolation strength (lowest set of data points) shows the largest corrections but
increases monotonically for L . The values per (L) obtained from the percolation probability P1D show less drastic corrections but are more difficult to extrapolate to the infinite-volume
359
Fig. 12. Locations of the susceptibility maxima of various observables as a function of 1/L for = 2.0.
limit. The two sets of data points both extrapolate to a value around per 0.4748 which follows
from fitting (P ) to Eq. (14).
Fig. 12 displays in addition to the locations of the susceptibility maxima of the vortex line
density, also those of the coslink energy and the hopping energy. For L > 16 the heights and
locations of these susceptibilities remain constant, so that, as far as these observables are concerned, the infinite-volume limit is reached. The location of the crossover in the infinite-volume
limit determined using these observables is below all estimates of the percolation threshold. In
other words, while for = 1.1, = 0.2, per < , here the relative positions have changed,
per > , and the vortices proliferate already in the Higgs phase before entering the crossover
region. By continuity we then expect at some value of in the interval 1.1 < < 2.0 the percolation threshold of the vortices to coincide with the location of the crossover. Since varying
physically changes the monopole density, it is tempting to conclude that monopoles impede
the formation of percolating vortex lines, and that by adjusting the monopole density, the location of the apparent vortex percolation threshold can be fine-tuned to coincide with that of the
crossover. The use of the word apparent here is to underscore that vortex networks identified
with our tracing rules are only an approximation to the true networks. The element of impediment introduced by the monopoles possibly plays a similar role as the stochastic element in the
FortuinKasteleyn construction of spin clusters in the Potts model [30].
We finally estimate the critical exponent per for = 2.0. Using the percolation probability as input observable for our collapsing routine, we arrive at the value per = 0.88(1) and
per = 0.4748(1). As a crosscheck we fit the scaling of (P ) with the lattice size L to Eq. (14)
giving per = 0.88(8) and per = 0.47484(15) with a 2 /DOF = 2.1. This is again consistent
with the standard percolation exponent. Although the relative positions of the apparent percolation threshold and the crossover have changed, variations in the inverse gauge coupling parameter
appear not to change the value of this critical exponent.
7. Conclusions
The vortices arising in the compact Abelian Higgs model have been investigated by means
of Monte Carlo simulations on a cubic lattice, and analyzed with the help of observables known
from percolation theory. Because their behavior is more pronounced, percolation observables are
better suited than other observables to probe the phase boundary, both in the first-order transition
360
region as well as in the crossover region of the phase diagram. In the region where the Higgs
and confinement phases are separated by a first-order transition, the vortices percolate right at
the phase boundary. Since the rules applied to trace out the vortices result in networks that are in
general only an approximation to the true ones, it is concluded that the discontinuous first-order
transition is forgivable of the resulting inaccuracies. The vortex network reflects the first-order
nature of the transition in this region of the phase diagram. In the crossover region, the vortices still percolate. The percolation observables show second-order critical behavior along the
Kertsz line that is characterized by the usual percolation exponents. The location of the vortex percolation threshold estimated using percolation observables does not coincide with that of
the crossover estimated using the vortex line density. Ideally, one would expect both to coincide. Also the ||4 theory in 3D, which undergoes a continuous phase transition, shows a similar
behavior [29]. Whereas the estimate based on the line density coincides with the critical temperature, the estimates based on any of the percolation observables considered do not agree within
error bars. This discrepancy may arise because the vortex networks are not correctly traced out
or because a stochastic or impeding element in the construction of a network is missing. The
monopoles appear to play such a role.
Acknowledgements
We wish to thank Arwed Schiller for useful discussions at an early stage of this project. S.W.
acknowledges a PhD fellowship from the Studienstiftung des deutschen Volkes. A.S. is indebted
to Professor H. Kleinert for the kind hospitality at the Freie Universitt Berlin. This work was
partially supported by the Deutsche Forschungsgemeinschaft under grant Nos. JA483/22-1 and
23-1, a computer time grant on the JUMP computer of NIC at Forschungszentrum Jlich under
project number HLZ12, and the EC Marie Curie Research and Training Network ENRAGE
Random Geometry and Random Matrices: From Quantum Gravity to Econophysics, under
grant No. MRTN-CT-2004-005616.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
E. Fradkin, S.H. Shenker, Phys. Rev. D 19 (1979) 3682.

T. Banks, E. Rabinovici, Nucl. Phys. B 160 (1979) 349.
K.C. Bowler, G.S. Pawley, B.J. Pendleton, D.J. Wallace, G.W. Thomas, Phys. Lett. B 104 (1981) 481.
A.M. Polyakov, Nucl. Phys. B 120 (1977) 429.
Y. Munehisa, Phys. Lett. B 155 (1985) 159.
G.N. Obodi, Phys. Lett. B 174 (1986) 208.
A. Kovner, B. Rosenstein, Int. J. Mod. Phys. A 7 (1992) 7419.
G. Bhanot, B.A. Freedman, Nucl. Phys. B 190 (1981) 357.
S. Wenzel, E. Bittner, W. Janke, A.M.J. Schakel, A. Schiller, Phys. Rev. Lett. 95 (2005) 051601.
S. Mandelstam, Phys. Rep. 23 (1976) 245.
G. t Hooft, Nucl. Phys. B 138 (1978) 1.
M.B. Einhorn, R. Savit, Phys. Rev. D 19 (1979) 1198.
J. Kertsz, Physica A 161 (1989) 58.
K. Langfeld, in: M.G. Schmidt (Ed.), Strong and Electroweak Matter 2002, Proceedings of the Workshop on Strong
and Electroweak Matter, Heidelberg, Germany, 25 October 2002, World Scientific, Singapore, 2003.
R. Bertle, M. Faber, J. Greensite, S. Olejnik, Phys. Rev. D 69 (2004) 014007.
H. Satz, Nucl. Phys. A 681 (2001) 3.
M.N. Chernodub, F.V. Gubarev, E.-M. Ilgenfritz, A. Schiller, Phys. Lett. B 434 (1998) 83.
M.N. Chernodub, F.V. Gubarev, E.-M. Ilgenfritz, A. Schiller, Phys. Lett. B 443 (1998) 244.
361
[19] R. Bertle, M. Faber, in: N. Brambilla, G.M. Prosperi (Eds.), Quark Confinement and the Hadron Spectrum V, Proceedings of the 5th International Conference, Gargnano, Italy, 1014 September 2002, World Scientific, Singapore,
2002.
[20] K. Langfeld, O. Tennert, M. Engelhardt, H. Reinhardt, Phys. Lett. B 452 (1999) 301.
[21] M. Engelhardt, K. Langfeld, H. Reinhardt, O. Tennert, Phys. Rev. D 61 (2000) 054504.
[22] K. Langfeld, Phys. Rev. D 67 (2003) 111501.
[23] S. Wenzel, E. Bittner, W. Janke, A.M.J. Schakel, A. Schiller, POS LAT2005 (2005) 248.
[24] D. Stauffer, A. Aharony, Introduction to Percolation Theory, second ed., Taylor and Francis, London, 1994.
[25] M.N. Chernodub, E.-M. Ilgenfritz, A. Schiller, Phys. Lett. B 547 (2002) 269.
[26] M.N. Chernodub, R. Feldmann, E.-M. Ilgenfritz, A. Schiller, Phys. Rev. D 70 (2004) 074501.
[27] S. Takashima, I. Ichinose, T. Matsui, K. Sakakibara, Phys. Rev. B 74 (2006) 075111.
[28] K. Kajantie, M. Laine, T. Neuhaus, A. Rajantie, K. Rummukainen, Phys. Lett. B 482 (2000) 114.
[29] E. Bittner, A. Krinner, W. Janke, Phys. Rev. B 72 (2005) 094511.
[30] C.M. Fortuin, P.W. Kasteleyn, Physica 57 (1972) 536.
[31] M.N. Chernodub, R. Feldmann, E.-M. Ilgenfritz, A. Schiller, Phys. Lett. B 605 (2005) 161.
[32] W. Janke, A.M.J. Schakel, in: Y. Holovatch (Ed.), Order, Disorder and Criticality, Advanced Problems of Phase
Transition Theory, vol. 2, World Scientific, Singapore, 2007, p. 123.
[33] B.A. Berg, T. Neuhaus, Phys. Rev. Lett. 68 (1992) 9.
[34] W. Janke, in: B. Dnweg, D.P. Landau, A.I. Milchev (Eds.), Computer Simulations of Surfaces and Interfaces,
Kluwer, Dordrecht, 2003, p. 137.
[35] K. Kajantie, M. Laine, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 466 (1996) 189.
[36] B. Bunk, Nucl. Phys. B (Proc. Suppl.) 42 (1995) 566.
[37] A.M. Ferrenberg, R.H. Swendsen, Phys. Rev. Lett. 61 (1988) 2635.
[38] A.M. Ferrenberg, R.H. Swendsen, Phys. Rev. Lett. 63 (1989) 1195.
[39] W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterling, Numerical Recipes: The Art of Scientific Computing,
second ed., Cambridge Univ. Press, Cambridge and New York, 1992.
[40] R.P. Brent, Algorithms for Minimization without Derivatives, Prentice Hall, Englewood Cliffs, NJ, 1973.
[41] B. Efron, The Jackknife, the Bootstrap, and Other Resampling Plans, SIAM, Philadelphia, 1982.
[42] L. Wang, K.D.S. Beach, A.W. Sandvik, Phys. Rev. B 73 (2006) 014431.
[43] M. Galassi, et al., GNU Scientific Library Reference Manual, second ed., Network Theory Limitedition, 2003.
[44] S.M. Bhattacharjee, F. Seno, J. Phys. A 34 (2001) 6375.
[45] W. Janke, Phys. Rev. B 47 (1993) 14757.
[46] H.G. Ballesteros, L.A. Fernandez, V. Martin-Mayor, G. Parisi, J.J. Ruiz-Lorenzo, J. Phys. A 32 (1999) 1.
[47] Y. Deng, H.W. Blte, Phys. Rev. E 72 (2005) 016126.
[48] M. Hellmund, W. Janke, Phys. Rev. E 74 (2006) 051113.

Erratum to: Cosmic microwave background,

matterantimatter asymmetry and neutrino masses
[Nucl. Phys. B 643 (2002) 367]
W. Buchmller a , P. Di Bari a , M. Plmacher b,
a Deutsches Elektronen-Synchrotron DESY, 22603 Hamburg, Germany
b Theoretical Physics, University of Oxford, 1 Keble Road, Oxford, OX1 3NP, United Kingdom
Received 20 November 2007

Available online 5 December 2007
As pointed out by Giudice et al. [1], a proper treatment of the real intermediate state contribution to the L = 2 scattering processes also subtracts the term given in Eq. (18). This corrects a
result arising from the subtraction procedure introduced in [2], which has been extensively used
in the literature. Removing the spurious term (18) reduces the washout rate W by a factor of
about 2/3. As a consequence, the numerical factors change in Eqs. (60) and (63):

2.5 103 eV2 1/2
,
M1 2.2(0.4) 109 GeV
m2atm
and
M1 2 10
10

GeV

m
1
.
0.01 eV
A detailed discussion can be found in [3].

References
[1] G.F. Giudice, A. Notari, M. Raidal, A. Riotto, A. Strumia, Nucl. Phys. B 685 (2004) 89.
[2] E.W. Kolb, S. Wolfram, Nucl. Phys. B 172 (1980) 224;
E.W. Kolb, S. Wolfram, Nucl. Phys. B 195 (1982) 542;
J.A. Harvey, E.W. Kolb, D.B. Reiss, S. Wolfram, Nucl. Phys. B 201 (1982) 16.
[3] W. Buchmller, P. Di Bari, M. Plmacher, Ann. Phys. 315 (2005) 305.
DOI of original article: 10.1016/S0550-3213(02)00737-X.

E-mail address: pluemi@mppmu.mpg.de (M. Plmacher).

doi:10.1016/j.nuclphysb.2007.11.030
Nuclear Physics B 793 [FS] (2008) 363395

Analytic Bethe ansatz and Baxter equations

for long-range psl(2|2) spin chain
A.V. Belitsky
Department of Physics, Arizona State University, Tempe, AZ 85287-1504, USA
Received 23 July 2007; accepted 17 September 2007
Abstract
We study the largest particle-number-preserving sector of the dilatation operator in maximally supersymmetric gauge theory. After exploring one-loop Bethe ansatz for the underlying spin chain with psl(2|2)
symmetry for simple root systems related to several KacDynkin diagrams, we use the analytic Bethe anzats to construct eigenvalues of transfer matrices with finite-dimensional atypical representations in the
auxiliary space. We derive closed Baxter equations for eigenvalues of nested Baxter operators. We extend
these considerations for a non-distinguished root system with FBBF grading to all orders of perturbation
theory in t Hooft coupling. We construct generating functions for all transfer matrices with auxiliary space
determined by Young supertableaux (1a ) and (s) and find determinant formulas for transfer matrices with
auxiliary spaces corresponding to skew Young supertableaux. The latter yields fusion relations for transfer
matrices with auxiliary space corresponding to representations labelled by square Young supertableaux. We
derive asymptotic Baxter equations which determine spectra of anomalous dimensions of composite Wilson
operators in noncompact psl(2|2) subsector of N = 4 super-YangMills theory.
PACS: 02.20.Sv; 11.15-q; 11.30.Ly; 75.10.Jm
Keywords: Analytic Bethe ansatz; Lie superalgebra; Solvable lattice model; Baxter equations; Anomalous dimensions
1. Introduction
Low-dimensional integrable structures were long known to emerge in Quantum Chromodynamicsthe theory of strong interaction. Evolution equations describing logarithmic modification of scattering amplitudes in kinematical regimes corresponding to different physical phenomE-mail address: andrei.belitsky@asu.edu.
doi:10.1016/j.nuclphysb.2007.09.009
364
A.V. Belitsky / Nuclear Physics B 793 [FS] (2008) 363395
ena were found to possess hidden symmetries. These include high-energy and large-momentum
transfer asymptotics of the theory. The former refers to Regge behavior of cross sections in
energy variable. In leading logarithmic approximation it is governed by reggeon quantum mechanics, i.e., an interacting system with conserved number of particles. Its Hamiltonian was
identified with the one of a noncompact Heisenberg magnet with SL(2, C) symmetry group [1,2].
The regime of scattering amplitudes with large momentum transfer is endowed with operator
product expansion such that evolution equations are equivalent to CallanSymanzik equations.
The latter is in turn a Ward identity for dilatations which is one of the generators of the conformal group, the symmetry of classical Lagrangian of the theory. The dilatation operator acts on
the space spanned by Wilson operatorscomposite operators built from elementary fields of the
theory and covariant derivatives. At leading order of perturbation theory, the dilatation operator
admits a pair-wise form for a class of quasipartonic operators [3]. In multicolor Nc limit,
only nearest neighbor interactions survive with long-range effects being suppressed by 1/Nc . It
was realized that the dilatation operator for aligned-helicity operators coincides with the Hamiltonian of yet another integrable systemspin chain with SL(2, R) symmetry group [46]. All
four-dimensional gauge theories inherit integrability of one-loop dilatation operator since they
all share the same noncompact sector of operators with covariant derivatives [7,8]. Supersymmetry enhances the phenomenon to a larger set of operators eventually encompassing all operators
in maximally supersymmetric gauge theory as was demonstrated in [813]. The super-spin chain
Hamiltonian inherits the symmetry group of the classical gauge theory and its spectrum can be
found by means of the nested Bethe ansatz [14].
Beyond leading order of perturbation theory in a generic gauge theory many spacetime
charges associated with classical symmetry generators cease to be conserved due to anomalies.
This may potentially lead to breaking of integrability in higher loops. The N = 4 super-Yang
Mills theory, on the other hand, is superconformal to all orders and thus its classical symmetry
persists even when quantum effects are taken into account. However, starting from two-loop
order the dilatation operator becomes long-ranged, namely more then two partons can be simultaneously involved in scattering. This invalidates standard procedures to derive Bethe ansatz
equations. Evidence gathered from multiloop perturbative calculations hinted that integrability
in maximally supersymmetric gauge theory carries on to higher orders and that the spectrum of
composite operators is encoded in a long-range super-spin chain model
[1517]. Bethe ansatz
type equations which depend on the t Hooft coupling constant g = gYM Nc /(2) and which
generate known anomalous dimensions in lowest orders of perturbative expansion were conjectured in Ref. [17]. Integrability of planar N = 4 super-YangMills theory then suggests that
spectra of anomalous dimensions of composite operators can be computed exactly at finite coupling g and the superconformal nature of the theory implies that this provides an exact solution
to it. These Bethe ansatz equations do not properly incorporate wrapping effect when the range
of the interaction becomes as long as the spin chain itself and thus the equations are intrinsically
asymptotic.
In this paper we will continue developing an alternative approach to long-range super-spin
magnets based on transfer matrices and Baxter equations [18] initiated in Refs. [1921]. Transfer
matrices is the main ingredient of this framework. They encode the full set of mutually commuting conserved quantities with Hamiltonian being one of these. Transfer matrices are supertraces
of monodromy matrices in a representation of the symmetry algebra. The Baxter operators themselves are certain transfer matrices with a specialspectral parameter-dependentdimension of
representations in the auxiliary space. Thus, they are quantum generalization of supercharacters
and we should expect a transfer matrix (or its eigenvalues) to be a sum of terms, one per compo-
365
nent of corresponding representation in the auxiliary space. The absence of R-matrices yielding
the putative long-range Bethe equations does not prevent us from formulating transfer matrices
of the model. To accomplish this goal we resort on the analytic Bethe ansatz [22], a techniques
which bypasses the microscopic treatment and relies on general properties of the macroscopic
system like analyticity, unitarity and crossing symmetry.
Recently we have addressed Baxter equation for the closed sl(2|1) subsector of the N = 4
dilatation operator and have shown that it takes the form of a second order finite difference
equation [21] like for the sl(2) long-range magnet [20]. In the present work, we will focus on
the maximal particle-number-preserving psl(2|2) sector [23] of the maximally supersymmetric
gauge theory studied in a number of papers at one [24], two [25] and all [17] orders, where a
similar form of Baxter equations is anticipated [26].
The outline of the paper is as follows. In Section 2 we describe the sl(2|2) subsector of
N = 4 super-YangMills theory which arises as a projection of the theory on the light-cone
and restricting the particle content to N = 2 hypermultiplet. Then in Section 3, we analyze the
short range spin chain describing the spectrum of one-loop dilatation operator. We start with
the distinguished basis and construct transfer matrices with simplest representations in auxiliary
space which suffice to formulate closed Baxter equations for respective nested Baxter functions.
Subsequently we perform a set of Weyl super-reflections with respect to odd roots to obtain
Bethe equations corresponding to KacDynkin diagram with two isotropic fermionic roots which
allows for a natural generalization to all orders of perturbation theory. In Section 3.3, we construct all transfer matrices with symmetric and antisymmetric atypical representations in the
auxiliary space and build a determinant representation for transfer matrices labelled by a skew
Young superdiagram. Subsequently we find finite difference equations for Baxter polynomials.
All previous considerations are generalized in Section 4 to all orders of perturbation within the
framework of asymptotic analytic Bethe ansatz. Several appendices contain details of projection of su(2, 2|4) superconformal symmetry of maximally supersymmetric YangMills theory
to its psl(2|2) subsector, superspace realization of these algebras and SerreChevalley bases for
psl(2|2) corresponding to distinguished and symmetric KacDynkin diagrams. Finally we determine numbers of roots of nested Baxter functions in terms of eigenvalues of Cartan generators.
2. sl(2|2) subsector of N = 4 super-YangMills
To start with, let us specify the subsector of N = 4 super-YangMills theory which will be the
focus of our current study. Recall that the physical field content of the maximally supersymmetric
gauge theory can be accommodated in a single chiral light-cone N = 4 superfield [8,27,28],

i

x , A = z1 A x + A z1 A x + A B AB x
2!
1

1
+ ABCD A B C D x ABCD A B C D z A x .
3!
4!
(2.1)
depending on the bosonic four-vector x = (z, x + , x ) with its minus component being chiral
light-cone coordinate z = x + 12 A A and four Grassmann variables A . Truncating this superfield in one of the superspace coordinates, say 1 , one observes the particle content falls into two
N = 2 superfields [8]

x , A 1 =0 = G x , 2 , 3 + 4 WZ x , 2 , 3 ,
(2.2)
366
with one encoding the N = 2 gauge supermultiplet and another N = 2 WessZumino hypermultiplet [8],

i

G x , 2 , 3 = z1 A x + z1 j j x + j k j k x ,
2
2 3

1
j
2 3 1
WZ x , , = z 4 x + i 4j x + x ,
(2.3)
(2.4)
where the summation runs over the remaining odd directions in superspace, i.e., j, k = 2, 3. It
is straightforward to identify the closed sl(2|2) subsector [23] of the full theory as the WZ
component of the N = 4 light-cone superfield projected on the light-cone, i.e., x = (z, 0, 0 ).
For further use, it is convenient to introduce new notations for components of the hypermultiplet
and odd direction of the N = 2 superspace. Namely, identifying s as 1 = 2 , 2 = 3 , the
gaugino fields as 4 = , = 1 and the su(2) doublet of scalars as follows a = ( 42 , 43 ) =
(X, Z), we get the WessZumino superfield in the form
1
WZ (Z) = z1 (z) + i a a (z) + ab a b (z),
2
(2.5)
depending on the light-cone superspace variable Z = (z, a ). Truncating further in either 1 or

2 variable, one gets the closed sl(2|1) sector recently discussed in Ref. [21]. Thus in the largeNc limit the sl(2|2) sector of the dilatation operator in the maximally supersymmetric YangMills
theory is spanned by single trace operators built from the WessZumino superfields,

O(Z1 , . . . , ZL ) = tr WZ (Z1 ) . . . WZ (ZL ) .
(2.6)
The light-cone superfield WZ (Z) defines an infinite-dimensional chiral representation of the

sl(2|2) algebra with generators realized as differential operators acting on superspace coordinates
as shown in Appendix A. The WessZumino superfield possesses a vanishing conformal spin and
involves a nonlocal operator z1 (0) as an artefact of the light-cone formalism. To overcome
this complication we introduce a regularization1 by setting the conformal dimension of WZ (Z)
to WZ = and taking the physical limit 0 at the end. The representation space V arises
from the expansion of the superfield in Taylor series in even and odd variables around z = a = 0
and is spanned by polynomials in z and a ,

1
V = span 1, zk+1 , a zk , ab a b zk k N .
2
(2.7)
The single-trace L-field light-cone operator (2.6) belongs to the L-fold product of these spaces
VL
. Therefore, the eigenfunctions of the spin chain are classified according to irreducible
components entering this tensor product parameterized by the eigenvalues = [, t, b, L] of
generators of the psl(2|2) Cartan subalgebra, uB (1) automorphism and the length of the operator
(see Appendix B). The local Wilson operators O corresponding to superconformal polynomials
can be projected out from the light-cone operator (2.6) by means of the sl(2|2)-invariant
scalar product,
1 This regularization affects the form of the generators (A.25) which receive additive addenda: L+ L+ = L+ + 2z,

L0 L0 = L0 + , V a,+ V a,+ = V a,+ + 2 a and B B = B + 12 .
O = (Z1 , . . . , ZL )O(Z1 , . . . , ZL )

L
[DZk ] (Z1 , . . . , ZL )O(Z1 , . . . , ZL ),
367
(2.8)
k=1
with the integration measure [29]

1
[DZ] =
(2 + 1)

|z|1
d 2z

2

2
d a d a 1 z z + a a .
(2.9)
a=1
One can easily convince oneself that the polynomials zk a . . ., forming the representation space
V , are orthogonal with respect to the sl(2|2)-scalar product, i.e., zk a . . . |zn b . . .
kn ab . . ..
The nonlocal operator z1 (0) associated with the lowest component in the Taylor expansion
of the superfield WZ (Z) defines an invariant one-dimensional subspace Vnonloc = {1}. Local
Wilson operators belong to the quotient Vloc = V /Vnonloc . With the chosen normalization of
the measure (2.9), one finds that zk |zn
kn / (2 + k) such that the vector belonging to
Vnonloc possesses zero norm and is orthogonal to all states

k+1 1 k 2 k 1 2 k

z , z , z , z k N (D+ )k ,
(D+ )k X, (D+ )k Z, (D+ )k k N ,
(2.10)
belonging to Vloc . These will be identified with excitations propagating on the super-spin chain
which we will turn to next.
3. Short-range spin chain
The eigenvalue problem for the spectrum of one-loop anomalous dimensions of superconformal operators (2.6) can be reformulated in terms of a short-range psl(2|2) quantum super-spin
chain with the one-loop dilatation operator being identified with the nearest-neighbor Hamiltonian of the magnet. The spin chain is integrable and it can be diagonalized be means of the
nested Bethe ansatz [14]. It was observed some time ago [30,31] that for a spin chain model
based on a given symmetry (super-)algebra, the nested Bethe ansatz equations are determined by
simple root systems of the algebra, generally,

App /2
(1)
(p)
u0,k 2i wp
(p)
u0,k + 2i wp
L
=
(p)
(q)
nq
Nr

u0,k u0,j + 2i Apq
q=1 j =1
(p)
(q)
u0,k u0,j 2i Apq
(3.1)
where2 Apq = ( p | q ) is the Cartan matrix and wp are the KacDynkin labels, wp = ( p |)
determined by a weight vector of a representation of the algebra acting on the spin chain sites.
Here Nr = rank(G) is the rank of the algebra G. For projective algebras the upper limit in the
product is Nr = rank(G) + 1, which is 3 for our psl(2|2) sector. As it becomes obvious from the
above equation, there exists several sets of equivalent Bethe ansatz equations reflecting the fact
that there are several choices of simple root systems { p |p = 1, . . . , Nr } for a superalgebra, see
Fig. 1. These simple root systems are related by reflections with respect to odd simple roots
with vanishing bilinear form (|) = 0 which form the Weyl supergroup. For Bethe equations
this corresponds to a particlehole transformations.
2 See Appendix B for details.
368
Fig. 1. A subset of KacDynkin diagrams for psl(2|2). The super-Weyl reflection of the nodes denoted by the label SW
generates the diagram standing to its right.
To diagonalize the short-range psl(2|2) magnet one can use the nested Algebraic Bethe ansatz
[30,32] following Refs. [3335] and construct transfer matrices by a fusion procedure [36].
However, the lack of a systematic procedure to construct long-range integrable spin chains corresponding to gauge theories, binds one has to resort to techniques which bypass the microscopic
treatment and rely on general properties of macroscopic systems. The method of analytic Bethe
ansatz, which is a generalization of the inverse scattering method, was developed to determine
the spectrum of transfer matrices for closed chains [22] and serves the purpose. In this approach,
one uses general properties such as analyticity, unitarity, crossing symmetry, etc., to completely
determine eigenvalues of transfer matrices.
3.1. Transfer matrices and Baxter equations in distinguished basis
Let us start with the distinguished KacDynkin (left-most) diagram in Fig. 1 with BBFF
grading encoded in the Cartan matrix

2 1 0
A = 1 0
(3.2)
1 .
0
1 2
The nested Bethe equations read, according to Eq. (3.1),

u (1)
0,k i
L
=
(1)
u 0,k + i
(2) i L
u 0,k + 2
(2)
u 0,k
1 =
i
2
(3) (3)
Q 0 (u 0,k
(3) (3)
Q 0 (u 0,k
(1) (u (1) + i) Q (2) (u (1) i )

Q
0
0,k
0
0,k
2
(1) (u (1) i) Q (2) (u (1) + i )
Q
0
0,k
0
0,k
2
(1) (2)
(3) (u (2) + i )
Q 0 (u 0,k 2i ) Q
0
0,k
2
(1) (2)
(3) (u (2) i )
Q 0 (u 0,k + 2i ) Q
0
0,k
2
i) Q 0 (u 0,k + 2i )
(2)
(3)
(2) (3)
+ i) Q 0 (u 0,k 2i )
(3.3)
They are written in terms of the Baxter polynomials

(p)
Q 0 (u) =
n p

(p)
u u 0,k ,
(3.4)
k=1
(p)
parameterized by three sets of Bethe roots u 0,k . Here and below in this section all symbols carry
a subscript 0 indicating zeroth order of perturbation theory for the corresponding quantities in
gauge theory. These equations yield one-loop anomalous dimensions for Wilson operators in
psl(2|2) sector of N = 4 super-YangMills theory.
In this section we briefly discuss the construction of eigenvalues for transfer matrices in the
distinguished basis echoing considerations of Ref. [37] adopted to psl(2|2) algebra. Transfer matrices are supertraces of monodromy matrices with certain representation of symmetry algebra in
369
the auxiliary space. Here we will present only the ones with low-dimensional representations in
the auxiliary space which are sufficient to derive closed Baxter equations for nested Baxter polynomials. A full-fledge considerations will be done below for a simple root system corresponding
to the symmetric KacDynkin diagram in FBBF grading with two isotropic fermionic roots, see
(right-most graph) Fig. 1. Being supertraces over a representation of the algebra, the eigenvalue
formulas are expected to be given by a sum of terms, one for each component of the representation. This idea is at the crux of the approach suggested for spin chain based on classical Lie
algebras in Ref. [38,39] and generalized for superalgebras in Ref. [37].
We derive transfer matrices with auxiliary space labelled by a particular Young supertableau.
The Young superdiagrams are different from classical ones in that there is no limitation on the
number of rows [40]. Thus, we introduce elementary Young supertableaux depending on a spectral parameter u

(1)
i
i LQ
0 (u + 2 )
1u = u+
,
2 Q
(1) (u i )
0
2

L (1)
3i
(2)
(u
Q
3i
0
2 ) Q0 (u)
2u = u
,
2
(1) (u i ) Q (2) (u i)
Q
0
0
2

L (3)
3i
Q0 (u 2 ) Q(2)
i
0 (u)
3u = u
,
(3)
(2)
i
2
Q0 (u 2 ) Q0 (u i)

(3)
i
i LQ
0 (u + 2 )
4u = u
,
2 Q
(3) (u i )
0
2
(3.5)
(k)
parameterized in terms of three Baxter polynomials Q 0 entering the nested Bethe ansatz equations (3.3). Each box is labelled by an index with grading 1 = 2 = 0 and 3 = 4 = 1 in accord
with the distinguished KacDynkin diagram in Fig. 1.
We introduce notations for eigenvalues of transfer matrices
t0,(1a ) (u) = t0,[a] (u),
{s}
t0,(s) (u) = t0 (u),
(3.6)
with totally antisymmetric (1a ) and symmetric (s) atypical representations in the auxiliary
space. For the lowest dimensional representations, the eigenvalues of transfer matrices can we
written in terms of the elementary boxes as follows3
{1}
t0,[1] (u) = t0 (u) = 1 u + 2 u 3 u 4 u ,

1 i
1 i
1 i
2 i
2 i
3 i
t0,[2] (u) = 2 u 2i 3 u 2i 4 u 2i 3 u 2i 4 u 2i + 4 u 2i
u+
u+
u+
u+
u+
u+
2
3 i
4 i
+ 3 u 2i + 4 u 2i ,
u+
u+
2
3 Notice that the transfer matrix t

0,[1] (u) with defining fundamental representation (1) in the auxiliary space is given
by the supertrace str[LL (u). . .L1 (u)] of the product of Lax operators L(u) [8,33,34] and its eigenvalues in nested Bethe
ansatz given by Eq. (3.7).
370

{2}
t0 (u) =
12
13
14
u+ 2i u 2i
u+ 2i u 2i
u+ 2i u 2i
11
u+ 2i u 2i
22
23
u+ 2i u 2i
24
u+ 2i u 2i
34
u+ 2i u 2i
(3.7)
u+ 2i u 2i
The right-hand sides of these expressions are free from pole at positions of Bethe roots as can be
easily proved making use of the nested Bethe ansatz equations (3.3). When written explicitly in
terms of Baxter polynomials, the conjugate transfer matrices, i.e., with antichiral representations
in the auxiliary space, can be obtained from these by merely dressing transfer matrices with a
bar and changing the signs in front of imaginary units. The generating function of all transfer
matrices will be given below in Section 3.3 though for the symmmetric KacDynkin diagram.
Using these transfer matrices and their conjugate one may derive closed equations obeyed
by the Baxter polynomials. First, solving the transfer matrix t0,[1] (u) with respect to Q (1)
0 and
(2)
substituting it into t1 (u), one finds the Baxter equation for Q0

i (2)
i
i (2)
i
t0,[1] u +
Q0 u
t0,[1] u
Q0 u +
= 0.
2
2
2
2
(3.8)
(3)
It is a first order finite-difference equation. Analogously, eliminating the polynomial Q 0 from
the transfer matrix t0,[1] by substituting it twice into t0,[2] and shifting its argument in middle of
the way, one finds

i
i
i (1)
(1)
L
t0,[1] u
Q0 (u) + (u + i) t0,[1] u +
Q0 (u i)
t0,[2] (u) t0,[1] u +
2
2
2
(2)

i
i Q
0 (u + 2 ) (1)
+ (u i)L t0,[1] u
(3.9)
Q (u i) = 0.
2 Q
(2) (u i ) 0
0
2

From this one can find a closed equation for the Baxter polynomial Q 0 by merely eliminating
(2) making use of Eq. (3.8). An alternative, symmetric form of the
the ratio of the polynomials Q
0
Baxter equation can be found by deriving first an equation analogous to (3.9) but with conjugate
(2)
transfer matrices and eliminating the ratio of Q 0 s from them. This yields
(1)

i
i
t0,[2] (u) t0,[1] u +
t0,[1] u
2
2

i
i
(1)
t0,[2] (u) t0,[1] u +

t0,[1] u
Q 0 (u)
2
2

i
i
+ (u + i)L t0,[1] u
t0,[2] (u) t0,[1] u +
t0,[1] u
2
2

i
i
L
+ (u i) t0,[1] u +
t0,[2] (u) t0,[1] u +
t0,[1] u
2
2
i
2
i
2

(1) (u + i)
Q
0

(1) (u i) = 0.
Q
0
(1) from t0,[1] and eliminating it from t0,[2] we find an equation
Similarly, solving for Q
0
(3.10)
371

i (3)
(3)
L
t0,[2] (u)Q0 (u) + u t0,[1] u

Q0 (u + i)
2
(2)

i Q 0 (u + 2i ) (3)
L
+ u t0,[1] u
Q (u i) = 0,
2 Q (2) (u i ) 0
0
2
(3.11)
(2) ,
(3) upon the elimination of the polynomials Q
which gives a closed equation for Q
0
0

i
(3)
(3) (u + i)
t0,[2] (u)t0,[2] (u)Q 0 (u) + uL t0,[2] (u)t0,[1] u
Q
0
2

i (3)
Q0 (u i) = 0.
+ uL t0,[2] (u)t0,[1] u +
(3.12)
2
So far we have derived Baxter equations in terms of transfer matrices with antisymmetric
representation in the auxiliary space. The same considerations can be performed with symmetric
(1) ,
transfer matrices. From symmetric matrices we find an analogue to Eq. (3.9) for Q
0

i
{2}
(1)
(1) (u + i)
t0 (u)Q 0 (u) (u + i)L t0,[1] u
Q
0
2
(2)

i Q0 (u + 2i ) (1)
Q (u i) = 0.
(u i)L t0,[1] u
(3.13)
2 Q
(2) (u i ) 0
0
2
(2) , we get yet another Baxter equation
Eliminating Q
0

i (1)
{2}
{2}
{2}
(1)
t0 (u)t0 (u)Q 0 (u) (u + i)L t0 (u)t0,[1] u
Q0 (u + i)
2

i (1)
{2}
Q0 (u i) = 0,
(u i)L t0 (u)t0,[1] u +
2
(3.14)
{2}
cf. Eq. (3.10). Finally, solving the system of transfer matrices t0,[1] and t0 for Q 0 , we get

i
i
i (3)
{2}
(3)
L
t0,[1] u
Q0 (u) u t0,[1] u
Q0 (u + i)
t0 (u) t0,[1] u +
2
2
2
(2)

i Q 0 (u + 2i ) (3)
uL t0,[1] u
(3.15)
Q (u i) = 0,
2 Q (2) (u i ) 0
0
2
(1)
which being solved for Q (2)

0 together with its conjugate gives

i
i
i
i
{2}
{2}
(3) (u)
t0,[1] u
t0 (u) t0,[1] u +
t0,[1] u
Q
t0 (u) t0,[1] u +
0
2
2
2
2

i
i
i
{2}
(3) (u + i)
t0 (u) t0,[1] u +
t0,[1] u
Q
uL t0,[1] u
0
2
2
2

i
i
i
{2}
(3) (u i) = 0.
t0 (u) t0,[1] u +
t0,[1] u
Q
uL t0,[1] u +
(3.16)
0
2
2
2
The similarity of Baxter equations for Baxter polynomials in terms of symmetric and antisymmetric transfer matrices implies that there are consistency relations between them

i
i
i
i
t0,[1] u
= t0,[2] (u)t0,[1] u +
t0,[1] u
,
t0,[2] (u)t0,[1] u +
(3.17)
2
2
2
2
372

{2}
t0 (u)t0,[1]

i
i
i
i
{2}
t0,[1] u
= t0 (u)t0,[1] u +
t0,[1] u
.
u+
2
2
2
2
(3.18)
The validity of these equations can be explicitly verified using Eqs. (3.7).
3.2. Particle-hole transformation
The Bethe equations (3.3) in the distinguished basis are not particularly convenient for generalization beyond leading order of perturbation theory in maximally supersymmetric gauge theory
since the corresponding pseudovacuum state is not protected from quantum corrections in t
Hooft coupling constant. Therefore, it is necessary to transform Bethe and Baxter equations to
the basis with protected pseudovacuum state (see Appendix C). For the underlying superalgebra this reflects non-uniqueness in the choice of the simple root system. The inequivalent root
systems are related by Weyl group of super-reflections SW (G) with respect to odd roots of the
superalgebra [4143], as discussed in Appendix B. In terms of Bethe ansatz equations this is
known as the particlehole transformation [17,24,33,34,4447].
Let us perform a chain of these transformations on the distinguished KacDynkin diagram
yielding non-distinguished one with FBBF grading, Fig. 1, the central node of which will be
the sl(2) subalgebra corresponding to scalar Wilson operators with covariant derivatives. At first
step, we reflect the diagram with respect to the odd root 2 which translates into a duality trans(2)
formation with respect to the fermionic Bethe root u 0,k . At first, one rewrites the Bethe equation
for u (2)
0,k as zeros of the polynomial P2 (u) at the positions of these roots,

(2)
i L (1) (2) i (3) (2) i
(2)
0 = P2 u 0,k = u 0,k +
Q0 u 0,k
Q0 u 0,k +
2
2
2

L

i
i (3) (2) i
(2)
(1)
(2)
u 0,k
Q0 u 0,k +
.
Q 0 u 0,k
2
2
2
(3.19)
(2) (u) from P2 (u) with remaining roots encoded

Obviously, one can extract the polynomial Q
0
into yet another polynomial Q(2)
(u),
such
that
0
(2)
P2 (u) = 2 Q (2)
0 (u)Q0 (u),
(3.20)
(2)
where 2 = i(L n 3 + n 1 ). The number n2 of the dual roots u(2)
0,k of the polynomial Q0 (u) is
related to the ones of the other Baxter functions as n2 = L + n 1 n 2 + n 3 1. Using Eq. (3.20),
(2)
one eliminates the polynomial Q 0 (u) from the Bethe equations (3.3) and deduces Bethe equa-
tions corresponding to the Cartan matrix

0 1
0
A = 1 0 1 ,
0 1 0
with the (middle) KacDynkin diagram in Fig. 1. They read
(2)
1=
(1)
Q0 (u 0,k + 2i )
,
(2) (1)
Q0 (u 0,k 2i )
(2) i L
(2)
i
(3) (2) i
Q (1)
u0,k 2
0 (u0,k + 2 ) Q0 (u0,k 2 )
= (1) (2)
,
(2)
(3) (u(2) + i )
Q 0 (u0,k 2i ) Q
u0,k + 2i
0
0,k
2
(3.21)
373
1=
(2)
(3)
(2)
(3)
Q0 (u 0,k 2i )
Q0 (u 0,k + 2i )
(3.22)
At the next step, we reflect the root system respect to the odd root 1 of the middle Kac
(1)
Dynkin diagram in Fig. 1. To dualize the corresponding fermionic Bethe roots u 0,k , we introduce
yet another polynomial P1 (u) that vanishes, according to the Bethe ansatz equations (3.22), at
(1)
u = u 0,k ,

(1)
i
i
(2)
(1)
(2)
(1)
Q0 u 0,k
.
0 = P1 u 0,k = Q0 u 0,k +
(3.23)
2
2
(1)
The Bethe roots u 0,k do not exhaust all zeros of the polynomial P1 (u) and, therefore, for arbitrary
u it can be rewritten as a product of two polynomials
(u)Q (u),
P1 (u) = 1 Q
0
0
(1)
(1)
(3.24)
(1)
with the second one being the dual Baxter polynomial or order n1 in new Bethe roots u0,k . The
proportionality factor 1 and the power n1 are related to the numbers of parent Bethe roots as
(1)
follows 1 = in2 and n1 = n2 n 1 1, respectively. Again, eliminating the polynomial Q 0 (u),
one gets Bethe equations for the symmetric KacDynkin diagram with two isotropic fermionic
roots in (right-most) Fig. 1, and the Cartan matrix

0 1 0
A = 1 2 1 ,
(3.25)
0 1 0
which read
1=
(1)
i
Q(2)
0 (u0,k 2 )
,
(2) (1)
Q0 (u0,k + 2i )
(2) i L
(1) (2)
(2) (2)
(3) (2)
Q0 (u0,k i ) Q0 (u0,k + i) Q0 (u0,k 2i )
u0,k
= (1) (2) 2
,
(2) 2
(2) (2)
(3) (2)
u0,k + 2i
Q0 (u0,k + 2i ) Q0 (u0,k i) Q0 (u0,k + 2i )
1=
(2)
(3)
(2)
(3)
Q0 (u0,k 2i )
Q0 (u0,k + 2i )
(3.26)
where for conformity of notations we renamed Q 0 (u) = Q0 (u) and u 0,k = u0,k . In what follows, we will dub for brevity corresponding basis symmetric.
(3)
(3)
(3)
(3)
3.3. Transfer matrices in symmetric basis

Let us now construct eigenvalues of transfer matrices in symmetric basis labelled a skew
Young supertableaux [48,49]. Analogously to (3.5) we identify the elementary Young supertableaux with a product of ratios of Baxter polynomials

(1)
i L Q0 (u 2i )
1u = u+
,
2 Q(1) (u + i )
0
2
374

(1)
i L Q0 (u
2u = u+
2 Q(1) (u +
0

L (3)
Q0 (u +
i
3u = u
2 Q(3) (u
0

L (3)
Q
i
0 (u +
4u = u
2 Q(3) (u
0
(2)
i
2 ) Q0 (u + i)
,
(2)
i
2 ) Q0 (u)
(2)
i
2 ) Q0 (u i)
,
(2)
i
2 ) Q0 (u)
i
2)
,
i
)
2
(3.27)
with gradings 1 = 4 = 1 and 2 = 3 = 0 associated with (right-most) KacDynkin in Fig. 1.

Introducing a symbolic notation for the above single-box Young supertableau with a flavor index
and labelled by the spectral parameter u
u = Y0 (, u),
(3.28)
we can write generating functions [3739,50] for eigenvalues of transfer matrices in antisymmetric (1a ) representation

1

1
1 + Y0 (4, u)eiu
1 + Y0 (3, u)eiu 1 + Y0 (2, u)eiu 1 + Y0 (1, u)eiu

a 1 iau
=
t0,[a] u i
,
(3.29)
e
2
a=0
where the powers p = 1 2 of factors in the left-hand side reflect the grading of the Kac
Dynkin diagram; for symmetric (s) representation one finds

1
1

1 Y0 (1, u)eiu 1 Y0 (2, u)eiu
1 Y0 (3, u)eiu
1 Y0 (4, u)eiu

s 1 isu
{s}
t0 u i
.
e
=
(3.30)
2
s=0
Bethe ansatz equations (3.26) imply that these transfer matrices are pole-free.
One can define transfer matrices with auxiliary space labelled by a skew Young superdiagram
Y (m/n) [38,51]. Y (m/n) is obtained by removing a Young superdiagram Y (n) determined by
the partitioning n = {n1 , n2 , . . .} with the usual ordering condition on its elements n1 n2
from a larger superdiagram Y (m) with m = {m1 , m2 , . . .} and m1 m2 such that m n.
One enumerates all boxes of the Young superdiagram Y (m) starting with the top left one with
a pair of integers (j, k), j and k enumerating rows and columns, respectively. On the skew
superdiagram we define a set of admissible skew Young supertableaux Y (m/n) by assigning
a flavor (j, k) index to each box of the diagram Y (m/n) and distributing them according to
the following ordering conditions: (j, k) < (j, k + 1) and (j, k) < (j + 1, k) for any two
adjacent boxes, with weaker conditions when these indices have coincident gradings, namely, for
bosonic grading = 0:
(j, k) (j, k + 1),
(j, k) < (j + 1, k);
(3.31)
(j, k) (j + 1, k).
(3.32)
fermionic grading = 1:
(j, k) < (j, k + 1),
375
Fig. 2. Skew Young supertableau Y (m/n).
Obviously these flavor indices can take four different values, i.e., 1 4 for psl(2|2). Notice
that a Young dsuperiagram defined by the partition m = {m1 , m2 , . . . , mM } can be equivalently
represented as Y (m) = (s1a1 , s2a2 , . . . , sa ) with a1 +a2 + +a = M in case there are coincident
mk s, i.e., s1 = m1 = = ma1 , s2 = ma1 +1 = = ma2 , . . . . A transposed Young superdiagram
is then obtained by reflection across the main diagonal of horizontal and vertical rows. It can be
= ((a1 + + a )s , (a1 + + a1 )s1 s , . . .) where m
= {m
written as Y (m)
1, m
2 , . . .} such
that m
1 = M is the hight of the first column of Y (m), etc.
For the auxiliary space determined by a skew Young supertableau as in Fig. 2, the transfer
matrix can be constructed from the elementary boxes (3.27),

i
t0,Y (m/n) (u) =
(3.33)
p(j,k) Y0 (j, k), u + (m
1 m1 + 2j 2k) ,
2
Y (j,k)Y
where m
1 = M. These transfer matrices are functionally dependent. They satisfy a number of
functional relations known as fusion relations, namely, they admit a determinant representation
{s}
[3739,52] in terms of (anti-)symmetric transfer matrices t0,[a] and t0 ,

i
1 +m
j + n k j k + 1)
t0,Y (m/n) (u) = det t0,[m j n k j +k] u + (m1 m
2
1j,km1

i
{m n +j k}
= det t0 k j
1 mk nj + j + k 1) , (3.34)
u + (m1 m
2
1j,km
1
{s<0}
(u) = 0. For instance for a skew Young superdiagram Y (m /n ) with
with t0,[a<0] (u) = t0

m = {2, 2, 2} and n = {1, 1}, it yields
{1}
1
0
t0 (u 3i2 )
{2}
{1}
t0,Y (m /n ) (u) = det t0 (u i) t0 (u 2i )

1
{4}
{3}
{2}
i
t0 (u)
t0 (u + 2 ) t0 (u i)

3i
t0,[4] (u)
t0,[1] (u + 2 )
.
= det
(3.35)
1
t0,[3] (u 2i )
376
Eqs. (3.34) are a generalization of classical formulas for characters on representation determined
by corresponding Young supertableaux [40].
Using the determinant representation for transfer matrices t0,Y (m/n) (u), one can immediately
find bilinear fusion relations among them [50,53,54]. Making use of the DesnanotJacobi determinant identity, one immediately finds relations for transfer matrices with auxiliary space
{s}
corresponding to rectangular Young superdiagrams Y (m/) = (s a ), t0,(s a ) = t0,[a] :

i {s}
i
{s}
{s+1}
{s1}
{s}
{s}
t0,[a] u +
(3.36)
t0,[a] u
= t0,[a] (u)t0,[a] (u) + t0,[a+1] (u)t0,[a1] (u).
2
2
This is a Hirota bilinear difference equations derived in Ref. [37] and recently discussed in [55].
Similarly to the distinguished basis, it turns out however that out of the entire tower of transfer
matrices, we will need just the lowest-dimensional ones
{1}
t0,[1] (u) = t0 (u) = 1 u + 2 u + 3 u 4 u

(2)

(3)
i L Q0 (u + 2i ) Q0 (u i)
= u
1
(2)
2 Q(3) (u i )
Q0 (u)
0
2

(2)

(1)
i L Q0 (u 2i ) Q0 (u + i)
+ u+
1 ,
(2)
2 Q(1) (u + i )
Q0 (u)
0
2
(3.37)
and
1 i
1 i
1 i
2 i
2 i
3 i
t0,[2] (u) = 2 u 2i 3 u 2i + 4 u 2i + 3 u 2i 4 u 2i 4 u 2i
u+
u+
u+
u+
u+
u+
2
3 i
4 i
+ 3 u 2i + 4 u 2i
u+
u+
2

(3)
3i
Q(2)
0 (u 2 )
L
L Q0 (u + i)
= u (u i)
1 (2)
(3)
Q0 (u i)
Q0 (u 2i )
(2)
(2)
(1)
(3)
Q (u i) Q0 (u + i) (Q0 (u + 2i ) Q0 (u 2i ))2
u2L 0 (1)
(3)
(2)
(2)
Q0 (u)
Q0 (u)
Q0 (u + 2i )Q0 (u 2i )

(2)
(1)
Q0 (u + 3i2 )
L
L Q0 (u i)
+ u (u + i)
1 (2)
,
(1)
Q0 (u + i)
Q0 (u + 2i )
(3.38)
or
{2}
t0 (u) =
12
u+ 2i u 2i
22
13
u+ 2i u 2i
u+ 2i u 2i
14
u+ 2i u 2i
23
u+ 2i u 2i
24
u+ 2i u 2i
34
u+ 2i u 2i
33
u+ 2i u 2i

3i
2)
1
(2)
(2)
i
i
Q(3)
0 (u i) Q0 (u + 2 ) Q0 (u 2 )
(2)
(3)
(1)
Q0 (u 3i2 )
Q0 (u)
Q (u)
+ (u i)L (u + i)L (3)0
(1)
(2)
Q0 (u i) Q0 (u + i) Q0 (u 2i )
= uL (u i)L
(2)
(2)
(3)
Q0 (u + i) Q0 (u 2i ) Q0 (u

1
377

3i
2)
1
(2)
Q0 (u + 2i )
(1)
L
L Q0 (u i)
+ u (u + i)
(1)
Q0 (u + i)
(2)
Q0 (u +
(2)
(2)
Q0 (u + 2i ) Q0 (u +
(2)
Q0 (u 2i )
3i
2)
(2)
Q0 (u + 2i )

1 ,
(3.39)
(k)
and their conjugate, for the derivation of Baxter equations for the polynomials Q0 .
3.4. Baxter equations in symmetric basis
The transfer matrices (3.37), (3.38) and (3.39) and their conjugate can be used to find closed
equations for nested Baxter polynomials analogously to the distinguished basis as we discussed
in Section 3.1.
(2)
First, solving conjugate transfer matrices for the polynomial Q0 , one immediately finds the
(1)
(3)
following relations involving both Q0 and Q0 ,

i
i
i
i
(1)
(3)
(1)
(3)
Q0 u
t0,[1] (u) = Q0 u
Q0 u +
t0,[1] (u),
Q0 u +
(3.40)
2
2
2
2
Q0 (u + i)Q0 (u i)t0,[2] (u) = Q0 (u i)Q0 (u + i)t0,[2] (u),
(1)
(3)
(1)
{2}
(3)
{2}
Q0 (u + i)Q0 (u i)t0 (u) = Q0 (u i)Q0 (u + i)t0 (u).

(1)
(3)
(1)
(3)
(3.41)
(3.42)
They can be further generalized for arbitrary length of Young supertableaux as shown in
Eq. (4.22) for their all-order analogues. The similarity of the last two equations is a consequence
of functional relations between these transfer matrices

i
i
i
i
t0,[1] u +
(3.43)
t0,[1] u
t0,[2] (u) = t0,[1] u +
t0,[1] u
t0,[2] (u),
2
2
2
2

i
i {2}
i
i {2}
t0,[1] u +
(3.44)
t0,[1] u
t0 (u) = t0,[1] u +
t0,[1] u
t (u).
2
2
2
2 0
Now, finding Q0 from t0,[1] and eliminating it from t0,[2] , we get
(2)
(1)

(3)
i Q0 (u i) Q0 (u + i)
t0,[2] (u) + uL t0,[1] u +
2 Q(1) (u + i) Q(3) (u)
0
0
(1)

(3)
Q
Q
(u
i)
(u
+
i)
i
0
0
= 0.
+ uL t0,[1] u
(1)
(3)
2
Q0 (u) Q0 (u i)
(3.45)
Here is when the relations (3.40) becomes important for derivation of an autonomous finite(1)
(3)
difference equations. Using Eq. (3.40), one can eliminate either Q0 or Q0 from (3.45) and its
conjugate, and obtain a form of Baxter equations for these polynomials,

i
(1)
(1)
L
Q0 (u + i)
t0,[2] (u)t0,[2] (u)Q0 (u) + u t0,[2] (u)t0,[1] u
2

i
(1)
+ uL t0,[2] (u)t0,[1] u +
(3.46)
Q0 (u i) = 0,
2
378

(3)
t0,[2] (u)t0,[2] (u)Q0 (u) + uL t0,[2] (u)t0,[1]

i
(3)
Q0 (u + i)
u
2

i
(3)
Q0 (u i) = 0.
+ u t0,[2] (u)t0,[1] u +
2
L
(3.47)
(3)
Since the Baxter function Q0 was not affected by the series of particlehole transformation, its
Baxter equation in symmetric and distinguished (3.12) bases coincide.
{2}
Similar Baxter equations can be derived using the symmetric transfer matrix t0 . Solving the
{2}
(2)
transfer matrices t0,[1] and t0 with respect to Q0 , one finds

(1)

i
i
i Q0 (u i)
{2}
L
t0,[1] u +
t0 (u) + u t0,[1] u +
t0,[1] u
(1)
2
2
2
Q0 (u)
(3)

i Q0 (u + i)
= 0.
+ uL t0,[1] u
(3.48)
(3)
2
Q0 (u)
Deriving a similar equation for conjugate transfer matrices and solving the resulting system with
(1)
(3)
respect to either Q0 or Q0 , one deduces two Baxter equations

i
i
i
i
{2}
{2}
(1)
t0,[1] u
t0 (u) t0,[1] u +
t0,[1] u
Q0 (u)
t0 (u) t0,[1] u +
2
2
2
2

i
i
i
{2}
(1)
L
t0 (u) t0,[1] u +
t0,[1] u
Q0 (u + i)
u t0,[1] u
2
2
2

i
i
i
{2}
(1)
L
t0 (u) t0,[1] u +
t0,[1] u
Q0 (u i) = 0,
u t0,[1] u +
(3.49)
2
2
2

i
i
i
i
{2}
{2}
(3)
t0,[1] u
t0 (u) t0,[1] u +
t0,[1] u
Q0 (u)
t0 (u) t0,[1] u +
2
2
2
2

i
i
i
{2}
t0 (u) t0,[1] u +
t0,[1] u
Q(3)
uL t0,[1] u
0 (u + i)
2
2
2

i
i
i
{2}
(3)
t0 (u) t0,[1] u +
t0,[1] u
Q0 (u i) = 0.
uL t0,[1] u +
(3.50)
2
2
2
(3)
Again, the equations for Q0 in the distinguished and symmetric bases coincide by the same
token as before. Transfer matrices are polynomials in spectral parameter with coefficients determined by local conserved charges of the spin chain. The above equations provide quantization
(1,3)
conditions on these charges and determine the Baxter polynomials Q0 . This data is used
in turn to find the momentum-carrying roots of Q(2)
0 via Eq. (3.37). The latter determines oneloop anomalous dimensions of superconformal harmonics in the decomposition of the composite
light-cone operators (2.6).
4. Long-range spin chain
Now we turn to multiloop generalization of Bethe ansatz equations which describe the spectrum of anomalous dimensions of Wilson operators
in sl(2|2) sector of N = 4 super-YangMills
theory to all-orders in t Hooft coupling g = gYM Nc /(2). As we pointed out in Section 3.2,
the deformation one-loop equations beyond leading order of perturbation theory is achieved in
379
the symmetric basis (3.26). There are several important changes which occur when the longrange effects enter the game. The long-rage spin chain is written in terms of the renormalized
spectral parameter [15]

1
x[u] = u + u2 g 2 .
(4.1)
2
The scattering matrix of momentum-carrying excitations in long-range Bethe equation in the
symmetric basis acquires an additional phase factor due to renormalization of the superconformal
spin and also a nontrivial scattering phase absorbing deviations from strong-coupling calculations. Then the conjectured form of all-order psl(2|2) asymptotic Bethe equations read [17]
1=

(1)
(2)+
n2

xk xj
(1)
j =1 xk
(2)+ L
xk
(2)
xk
(2)
xj

1
n2

xk(2) xj(2)+
=
(2)+
(2)
xj
1
j =k=1 xk
g2
(2)+ (2)
xj
2
g
4xk
n1
(2)+
(1) n3
(2)+
(3)

xk xm
xk xn
(2)
m=1 xk
(3)
(2)+
n2

xk xj
,
1=
(3)
(2)
j =1 xk xj
(1)
xm
(2) (2)+
xj
4xk
(2)
n=1 xk
(3)
xn
(2)+
,xj )
(2)
,xj )
ei(xk
ei(xk
(2)
(2)
(4.2)
where x = x[u ]. The precise form of the dressing factor (xj , xk ) [56] will be irrelevant for
the present study since the construction is purely algebraic and relies on perturbative analyticity.
As in previous discussion we will assume the approach based on conjectured form of nested
Bethe ansatz equations and subsequent use of the analytic Bethe ansatz to find transfer matrices.
The cancellation of the pole will be done only in perturbative, asymptotic sense, thus neglecting
poles generated at finite coupling.
4.1. Transfer matrices
To start with, we define three Baxter polynomials with zeros determined by three sets of Bethe
roots
(p)
(u) =
np

(p)
u uk (g) .
(4.3)
k=1
(p)
The roots uk (g) depend on the t Hooft coupling and admit an infinite perturbative expansion
(p)
(p)
(p)
uk (g) = u0,k + g 2 u1,k + ,
(4.4)
(p)
with the lowest order term being the one-loop Bethe roots u0,k obeying the Bethe ansatz equations (3.26). Following our one-loop considerations in preceding sections, the eigenvalues of
transfer matrices will be built in terms of elements which parameterize components of a Young
supertableaux associated with an auxiliary space. The single-box Young supertableaux, depending on the spectral parameter u and labelled by the flavor index = 1, . . . , 4 with gradings
380
1 = 4 = 1 and 2 = 3 = 0, read
1
1 u = (x + )L e 2 + (x
+ )+ 1 (x + )
2
2 u = (x + )L e+ (x
+)
3 u = (x )L e (x
4 u = (x )L e 2 + (x
(1) (u i )
Q
2
,
(1) (u + i )
Q
2
(1) (u i )
Q(2) (u + i) Q
2
,
(1) (u + i )
Q(2) (u) Q
2
(3) (u + i )
Q(2) (u i) Q
2
,
(3) (u i )
Q(2) (u) Q
2
)+ 1 (x )
2
(3) (u + i )
Q
2
.
(3) (u i )
Q
2
(4.5)
They depend on the dressing factor

(2)
(x) = (x) (x),
(4.6)
which is composed of a trivial one [20]

1
(p) (x) =
1

dt ln Q(p) ( 2i gt)
u2 g 2
,
1
u + gt
1 t2
(4.7)
roughly accounting for renormalization of the superconformal spin in higher orders of perturbation theory, as explained in Ref. [19], and a nontrivial scattering factor corresponding to
(xj , xk ) in long-range Bethe ansatz equations admitting an integral form in terms of the nested
Baxter polynomials [21]

Q(2) ( 2i gt) 1
dt
1 s2
ln
ds
1
i
s t
Q(2) ( 2 gt)
1 t2
1

d
1
g2
g2
1
,
ln 1 +
2i sinh2 ()
4xx[ + gs]
4xx[ gs]
1
(x) = g
(4.8)
C[i,i]
cf. Ref. [57]. Here we also used a notation for a product of the nested Baxter functions and the
trivial dressing which emerges in all formulas
(p)
Q (p) (u) = e 2 0
1
(x[u])
Q(p) (u),
(4.9)
for p = 1, 3. For vanishing gauge coupling g = 0, Eq. (4.5) reduce to (3.27).

Introducing a symbolic notation for the elementary Young supertableaux (4.5)
u = Y(, u),
(4.10)
(1a )
one can write generating functions for transfer matrices with antisymmetric

1

1
1 + Y(4, u)eiu
1 + Y(3, u)eiu 1 + Y(2, u)eiu 1 + Y(1, u)eiu
=

a=0

t[a] x [1a] eiau ,
(4.11)
381
and symmetric (s) finite-dimensional chiral representations,

1
1

1 Y(1, u)eiu 1 Y(2, u)eiu
1 Y(3, u)eiu
1 Y(4, u)eiu
=

t {s} x [1s] eisu .
(4.12)
s=0
In order to write transfer matrices in a concise fashion, we used

i
u[n] u n,
2

x [n] x u[n] ,
(4.13)
with x [1] = x introduced earlier. The transfer matrices t[s] and t {s} are free from poles upon
the use of Bethe ansatz equations. Notice that in our treatment we assume perturbative analyticity
in the spectral parameter u, ignoring dynamical pole generated at finite coupling constant. The
incorporation of these into the analysisa problem which was recently addressed in a related
context in Ref. [58]goes beyond the scope of this paper.
The eigenvalues of transfer matrices with auxiliary space determined by a skew Young supertableau in Fig. 2 are build from the elementary boxes (4.5) using the same algorithm as spelled
out in Section 3.3 and takes the same functional form as Eq. (3.33),

p(j,k) Y (j, k), u[m 1 m1 +2j 2k] .
tY (m/n) (x) =
(4.14)
Y (j,k)Y
These transfer matrices admit a determinant representation in terms of (anti-)symmetric transfer

matrices t[a] and t {s} ,

tY (m/n) (x) = det t[m j n k j +k] x [m1 m 1 +m j +n k j k+1]
1j,km1
det
1j,km
1

t {mk nj +j k} x [m1 m 1 mk nj +j +k1] ,
(4.15)
with boundary conditions t[a<0] = t {s<0} = 0. For auxiliary space labelled by a rectangular Young
supertableaux, the all-order transfer matrices obey Hirota-type equations
{s}
{s}
{s+1}
t[a] (x + )t[a] (x ) = t[a]
{s1}
(x)t[a]
{s}
{s}
(x) + t[a+1] (x)t[a1] (x).
(4.16)
These can be extended to arbitrary Young supertableaux upon proper choice of boundary conditions as was recently discussed for short-range super-spin chains in Ref. [55] generalizing earlier
considerations for classical Lie algebras [59].
4.2. Baxter equations
Finally, let us derive a set of closed equations for the polynomials (4.3). Again it suffices to
consider the lowest-dimensional transfer matrices only. Using the generating functions (4.11)
and (4.12), we obtain explicit form of transfer matrices in defining fundamental representation

Q (3) (u + 2i ) (x ) Q(2) (u i)
1
1
2 + (x )+ 2 (x )
e
e
t[1] (x) = (x )L
Q(2) (u)
Q (3) (u 2i )

i
(2)
(1)
1
+ L Q (u 2 )
+ (x + ) Q (u + i)
+ (x + )+ 12 (x + )
2
+ (x )
(4.17)
e
e
,
(1) (u + i )
Q(2) (u)
Q
2
382
cf. Refs. [17,58], and the rest deferred to Appendix D. Removing excitation associated with
either first Q(1) = 1 or last Q(3) = 1 node of the symmetric KacDynkin diagram in Fig. 1 or
both, we reduce to all-order transfer matrices in either sl(2|1) [21] or sl(2) [20] subsectors of the
theory, respectively. Performing the same steps as earlier in Section 3.4, we find that the Baxter
equations take the form of second order finite-difference equations with coefficients determined
by either antisymmetric,
t[2] (x)t[2] (x)Q (1) (u) + x L e 2 (x)+ 2 + (x) t[2] (x)t[1] (x )Q (1) (u + i)
1
+ x L e 2 (x)+ 2 + (x) t[2] (x)t[1] (x + )Q (1) (u i) = 0,

1
(3)
t[2] (x)t[2] (x)Q (u) + x e

+ xLe
1
1
2 (x)+ 2 + (x)
1
1
2 (x)+ 2 + (x)
(4.18)
(3)
t[2] (x)t[1] (x )Q (u + i)
t[2] (x)t[1] (x + )Q (3) (u i) = 0,
or symmetric transfer matrices and deformed by dressing factors

{2}

(1)
(u)
t (x) t[1] (x + )t[1] (x ) t{2} (x) t[1] (x + )t[1] (x ) Q

1
1
x L e 2 (x)+ 2 + (x) t[1] (x ) t {2} (x) t[1] (x + )t[1] (x ) Q (1) (u + i)

1
1
x L e 2 (x)+ 2 + (x) t[1] (x + ) t{2} (x) t[1] (x + )t[1] (x ) Q (1) (u i) = 0,
{2}

(3)
(u)
t (x) t[1] (x + )t[1] (x ) t{2} (x) t[1] (x + )t[1] (x ) Q

1
1
x L e 2 (x)+ 2 + (x) t[1] (x ) t{2} (x) t[1] (x + )t[1] (x ) Q (3) (u + i)

1
1
x L e 2 (x)+ 2 + (x) t[1] (x + ) t {2} (x) t[1] (x + )t[1] (x ) Q (3) (u i) = 0.
(4.19)
(4.20)
(4.21)
On top of these equations, the product of Baxter polynomials Q(1,3) obey consistency conditions

(3) [a]

(3) [+a]
Q (1) u[+a] Q
(4.22)
u
t[a] (x) = Q (1) u[a] Q
u
t[a] (x),

(1) [+s] (3) [s] {s}
(1) [s] (3) [+s] {s}
Q u
(4.23)
Q u
t (x) = Q u
Q u
t (x),
with (anti-)symmetric transfer matrices, in their turn, obeying functional relations identical to the
one for a short-range magnet (3.43). Once the nested Baxter polynomials are determined from
Eqs. (4.17)(4.21), they generate the spectrum of transfer matrices associated with a skew Young
supertableaux via Eq. (4.15).
5. Conclusions
The main focus of the present study was a closed psl(2|2) subsector of the dilatation operator
in maximally supersymmetric YangMills theory. The sector is encoded into the N = 2 Wess
Zumino supermultiplet embedded into the N = 4 light-cone superfield and obeys autonomous
renormalization group evolution to all orders in t Hooft coupling.
Due to non-uniqueness of simple root systems for superalgebras we chose the one which allows for a straightforward generalization of Bethe equations to all orders of perturbation theory.
We concentrated on a KacDynkin diagram having two isotropic odd nodes which reduces to
noncompact sl(2) sector when the number of fermionic excitations vanishes. We used close relation of transfer matrices to representation theory and analytic Bethe ansatz to construct their form
for auxiliary spaces associated with skew Young supertableaux in terms of Baxter functions. The
latter possess determinant representation in terms of transfer matrices with symmetric (s) or antisymmetric (1a ) atypical representations in the auxiliary space. For zero spectral parameter these
383
relations reproduce well known supercharacter formulas for supergroups. Bethe ansatz equations
ensure that these transfer matrices are pole-free at positions of nested Bethe roots. We have used
these equations in perturbative, asymptotic sense when the dynamical poles at u = g/(2uk ) are
not reachable. Proper incorporation of the latter into the formalism and a proof of their cancellations would contribute to resolution of the notorious wrapping problem for the underlying
long-range spin chain, i.e., when the range of interaction is even or higher than the length of
chain itself.
We have formulated equivalent closed systems of Baxter equations (4.17), (4.18) and (4.19)
or (4.19), (4.20) and (4.21) for nested Baxter functions. Transfer matrices encode a full set of
mutually commuting conserved quantities. The solutions to these sets determine the spectra of
quantized charges and roots of Baxter functions and thus determine spectra of anomalous dimensions of Wilson operators to all orders in gauge coupling constant via the equation
1
(g) = ig
2
1

dt
i
(2) i
(2)
2
1 t ln Q
.
gt ln Q
gt
2
2
(5.1)
Perturbative solutions to these equations in lowest few orders in t Hooft coupling were given for
the sl(2|1) long-range spin chain in Ref. [21] and applies in a straightforward fashion to the case
at hand.
The framework of Baxter equations is advantageous for a number of reasons. While both
Bethe ansatz and Baxter equations produce identical results for models based on representations with highest and/or lowest weight vectors, the Baxter framework applies even when the
pseudovacuum state in the Hilbert space of the chain is absent. For noncompact super-spin
chains, the number of eigenstates is infinite for a finite length of the spin chain and the analysis of spectra in this approach is preferable. Another advantage of Baxter approach within the
present context of the putative long-range magnet is that it clearly demonstrates the limitation of
the asymptotic equations, i.e., their invalidity beyond wrapping order. Since the Baxter equation
is a polynomial equation in the renormalized spectral parameter x[u], by expanding everything
in perturbative series in coupling constant, one gets nonpolynomial terms. However, it is not this
non-polynomiality per se which breaks beyond wrapping order rather it is the fact that by solving the system of equations stemming from coefficients in front of powers of u, one finds that
it becomes overdetermined and inconsistent. This may serve as a starting point to elaborate on
corrections terms which yield all-order anomalous dimensions even for short spin chains. Another problem which deserves a dedicated study is the origin of the dressing phase. Recall that
the eigenvalues of the Baxter operator have a clear physical meaning of the wave functions of the
magnet in separated variables [60], a property which should be preserved to all order in t Hooft
coupling. The crossing symmetry [61,62] implemented in terms of Q yields a relations which
should be understood in physical terms as a certain constraint on the analytical properties of this
wave function. It remains to formulate psu(2, 2|4) Baxter equations for the full theory relying on
the analytic Bethe ansatz, to start with.
Acknowledgement
This work was supported by the US National Science Foundation under grant No. PHY0456520.
384
Appendix A. Superconformal algebra and superspace realization

The N -extended superconformal algebra su(2, 2|N ) contain 15 even charges P , M , D
, SA , S
A
which are two-component Weyl spinors carryand K and 4N odd charges QA , Q
A
ing an su(N ) index A = 1, 2, . . . , N . There are additional bosonic chiral charge R and, in case of
extended N > 1 supersymmetries, also su(N ) charges TA B satisfying the commutation relations
D T D . The nontrivial commutation relations, on top of conventional
[TA B , TC D ] = CB TA D A
C
bosonic su(2, 2) relations, read

B = 2 B P ,
QA , Q
A
i
[QA , D] = QA ,
2
A
A
,
S , P = Q
1
QA , M = QA ,
2

,
QA , K = S
A

i A
SA
, D = S ,
2
1
A
[QA , R] = QA ,
S , M = SA
,
2
A

A
= 2 A K ,
S , R = SA
S , S
,
B
B
B

B C
BD
BC D
TA , QC = tAC QD ,
TA , S = tAD
S ,

A
4
1 BA R 4TB A ,
S , QB = 2iBA D + BA M +
N
(A.1)
AC = A C 1 A C . The remaining (anti-)commutators

and their complex conjugate. Here tBD
D B
N B D
vanish. Here and below we use conventions for Clifford algebra from Ref. [63].
The extended superspace with coordinates X = (x , A , A
) admits a coset manifold parameterization
g(X ) eix
+i
A Q
A
A +i A
Q
(A.2)
with the multiplication law

g(X1 )g(X2 ) = g(X3 ),
(A.3)
and transformed coordinates being

X = x1 + x2 i2A 1A + i1A 2A , 1A + 2A , 1A
+ 2A
.
(A.4)
In this parametrization, a superfield in this superspace is defined as

(X ) = g(X )(0)g 1 (X ).
(A.5)
Using conventional technique of induced representations one easily computes a representation

of generators in the superspace. The center elements are

M , (0) = (0),
D, (0) = id(0),

B
R, (0) = r(0),
TA , (0) = tAB (0),
(A.6)
385
all the rest vanish.4 The representation of generators as differential operators acting on the coordinates X of the superfield ,

G, (X ) G(X ),
(A.7)
is for su(2, 2) bosonic
iP = ,
i
i
A
,
iM = x x i A A A

2
2
1
1
iD = d + x + A A + A
,
A
2
2

A

iK = 2x x x 2 g + 2A
B B
+ 2dx 2ix

4
+ i A
A 4i A
B tBA
1 A
+ ir

N

+ x A 2i B
A B A

+ x A
+ 2i A
B B A ,

(A.8)
and fermionic generators
iQA = A + i A ,
= A
+ i
A ,
i Q A

4
iSA = 2d + r
1 A + A + 4tBA B
N

x A 2iB B
A + 4B A B

2B B
,
ix
A

4
i S A = 2d r
1 A + A 4tAB B
N

B
x
A + 2i B A
+ 4B A

B

B
A ,
2B ix
(A.9)
and finally su(N ) and u(1) generators

i
iTB A = itBA + i A B B
BA C C C
,
A
C
N
iR = ir i A A + i A
,
A
(A.10)
cf. Ref. [64].

4 We assumed here that the superfield is a tensor with respect to SU(N ) group. The superfields of gauge theories
B = 0.
are SU(N ) scalars thus tA
386
A.1. Light-cone reduction of superspace

In this paper we are interested in a closed sl(2|2) subsector of the dilatation operator, which
acts on the light-cone composite operators (2.6) built from N = 2 WessZumino superfield of the
truncated maximally supersymmetric gauge theory. The sl(2|2) subalgebra of the superconformal
algebra is spanned by the following generators
P+ ,
SA
,
M+ ,
D,
K ,
,
S
A
R,
TA B ,
Q+A ,
A
Q
+ ,
M12 .
The vector indices of generators are contracted with the light-cone vectors
the following conditions n2 = n2 = 0 and n n = 1, such that
G n = G + ,
(A.11)
n
and
G n = G .
obeying
(A.12)
While the fermionic generators are obtained with the help of the projectors
1
G = G ,
2
.
= 1
G
G
(A.13)
One can easily convince oneself that actually only one component in each light-cone Weyl spinor
is nonvanishing. This reflect a general phenomenon of the light-cone formalism: a spinor satisfying the Weyl condition can be described by a complex Grassmann variable without a Lorentz
index. Further, it is convenient to introduce the following combinations of the generators
i
i
iP+ L ,
K L+ ,
(D + M+ ) L0 ,
2
2

A,
1 4
1
4
A i 4 8V
iQ1A 8V
iQ
,
1 R M12 B,
A,
1
4 N
2
4
A,+ ,
1 i 4 32V+ .
iS
iS1A 32V
A
A
(A.14)
The representation (A.7) of these generators in the light-cone superspace then reads for the
bosonic sl(2) subalgebra
L = z ,

A

1
1
B
A
2
A 2
B
z + z + B A A
L = 2z + A bB tB + z + A
4
2

1
+ z B B A A ,
2
1
1
L0 = + zz + A A + A A .
2
2
+
The fermionic generators read

1
VA = A + A z ,
2
1
A,
= A + A z ,
V
2
(A.15)

1
=
B +
2

1
+ z + B B A ,
2

V A,+ = ( + b)BA tBA B +

1
B
+ z B A ,
2
VA+
B
( b)A
+ tAB
387

1
B
z B A z + A B B
2

1
1
B
z + B A z + A B B
2
2
(A.16)
and the remaining are

2 A
1
1
A A A ,
B =b+
2
N

1
TB A = tBA + A B B A BA C C C C .
N
(A.17)
Here we used a notation z = x for the coordinate projected on the light cone. The Grassmann
coordinates A and A were redefined compared to the ones in the covariant superspace (A.2) as
follows
4
4
A 8 1A ,
(A.18)
A i 8A1 .
The quantum numbers of the superfields are encoded into the conformal spin and chirality b

1
1
1
4
= (s + d),
(A.19)
b= r
1 h,
2
4 N
2
with the latter being a linear combination of its helicity h = 12 and the R-charge r.
The commutation relations between the generators in the representation (A.15), (A.16) and
(A.17) can be summarized by combining them in an sl(2|N )covariant matrix with components
E 00 = L0
N
B,
N 2
E 0,N +1 = L+ ,
E 0A = VA+ ,
E A0 = V A, ,
E AB = TB A +
E N +1,0 = L ,
E N +1,A = VA
2
B A ,
N 2 B
E A,N +1 = V A,+ ,
E N +1,N +1 = L0
N
B.
N 2
(A.20)
Then the graded commutation relations read in a concise form

AB CD

E ,E
E AB E CD (1)(A+B)(C +D) E CD E AB
= CB E AD (1)(A+B)(C +D) AD E CB ,
(A.21)
where the indices A, . . . run over N + 2 values (0, A, N + 1) and possess the gradings 0 =
N + 1 = 0 and A = 1.
A.2. Projection to psl(2|2) subsector
The superfield (2.1) encoding the field content of maximally supersymmetric YangMills
theory is chiral, thus the dependence on the Grassmann variables A is trivial, i.e., (z, a , A ) =
388
(z + 12 A A , A , 0). This simplifies significantly the form of the generators, which are quoted
in Ref. [8]. The generators on the sl(2|2) subsector of N = 4 theory are found making use of the
superfield truncation (2.2), such that

E z, A 1 =0 = + 4 EWZ (Z),
(A.22)
where E are given in Eq. (A.20) for N = 4 and E are generators in question acting on a function
of Z = (z, a ). One finds the following identification of sl(2|2) generators E , obeying the
commutation relations

+)

)(
E ,E
E E (1)(+
E E
)( +)
= E (1)(+
E ,
(A.23)
to the ones of the truncated N = 2 superconformal light-cone algebra

E 00 = L0 2B,
E a0 = V a, ,
E 30 = L ,
E 0a = Va+ ,
E 03 = L+ ,
E ab = Tb a + 2Bba ,
E 3a = Va ,
E a3 = V a,+ ,
E 33 = L0 2B.
(A.24)
As it is obvious from this formula, the indices of E run over = 0, a, 3 with an su(2) index
taking two values a = 1, 2 and the grading reflects the one of (A.20), i.e., 0 = 3 = 0 and a = 1.
The above generators are realized on WZ superfield (2.5) and read
L = z ,
L+ = z2 z + z a a ,
1
Tb a = a b ba c c ,
2

V a,+ = a zz + c c ,
1
L0 = zz + a a ,
2
V a, = a z ,
Va = a ,
Va+ = z a .
(A.25)
The uB (1) outer automorphism B,

1 1
B = + a a ,
(A.26)
4 4
does not enter the right-hand side of the commutation relation of the generators (A.25). Thus
the sl(2|2) algebra (A.23) of generators (A.24) is a semidirect product uB (1) psu(2|2) with
generators (A.25) forming the projective algebra psl(2|2). Its quadratic Casimir operator is
3
1
C2 =
(1) E E
2
,=0

0

= L + T 0 L0 T 0 + 1 + L+ L + T2 1 T1 2 Va+ V a, V a,+ Va ,
(A.27)
where we introduced a notation for T 0 T1 1 = T2 2 . Notice that this su(2) generator is related
to the one of su(4) internal rotations as T 0 = T2 2 T3 3 .
Appendix B. SerreChevalley bases for psl(2|2)
The projective algebra psl(2|2) has rank two, but it is described by KacDynkin diagrams
having three nodes. This implies that the simple roots of the root system are not linearly inde-
389
pendent. The root system is expressed in terms of weights5 v = ( 0 | 1 , 2 | 3 ) which form a

basis in the dual space of the Cartan subalgebra. The weights obey the conditions
0 + 3 = 0,
1 + 2 = 0
(B.1)
and are endowed with a bilinear form such that

( 0 | 0 ) = 1,
( 1 | 1 ) = 1,
( 0 | 1 ) = 0.
(B.2)
The set of nonzero roots = 0 1 is divided into the set of even 0 and odd 1 roots,
0 = { a b , a b },
1 = {a b , b a }.
(B.3)
The root vectors associated to these roots are

E v v ,
(B.4)
and the Cartan subalgebra is spanned by the elements

E 00 + E 11 ,
E 11 E 22 ,
E 22 + E 33 .
(B.5)
It is obvious from explicit realization that these generators are linearly dependent, exhibiting
peculiarities of projective algebras. There are several choices of simple root systems depending
on choices of Borel subalgebras. Let us discuss two simple root systems used in the main text,
i.e., corresponding to the distinguished and symmetric KacDynkin diagram with two isotropic
fermionic roots.
B.1. Distinguished KacDynkin diagram
The distinguished KacDynkin diagram corresponding to the BBFF grading6 possesses the
following ordering of the basis elements of the dual Cartan subalgebra ( 0 , 3 | 1 , 2 ) and yields
the simple root system7
1
1 = ( 0 3 ),
2
1
2 = ( 3 1 ),
2
1
3 = ( 1 2 ).
2
(B.6)
The Cartan subalgebra is formed by the generators

h1 = E 00 E 33 = 2L0 ,
h2 = E 33 + E 11 = T 0 L0 ,
h3 = E 11 + E 22 = 2T 0
(B.7)
and together with positive and negative root vectors

e1+ = E 03 = L+ ,
e2+ = E 31 = V1 ,
e1 = E 30 = L ,
e2 = E 13 = V 1,+ ,
e3+ = E 12 = T2 1 ,
e3 = E 21 = T1 2 ,
(B.8)
5 Our ordering of basis vectors reflects the grading of the matrix of sl(2|2) generators E .
6 Note that the KacDynkin diagrams for superalgebras are ambiguous. There is yet another distinguished diagram
with FFBB grading.

7 The normalization factors reflect linear dependence (B.1) of basis elements and are introduced in order to have
conventional definition of the Cartan matrix Apq = ( p | q ).
390
form the SerreChevalley basis obeying the algebra

[hp , eq ] = Apq eq ,
[e1+ , e1 ] = h1 ,
[ep , eq ]p=q = 0,
[hp , hq ] = 0,
{e2+ , e2 } = h2 ,
[e3+ , e3 ] = h3 ,
(B.9)
with the Cartan matrix given in Eq. (3.2). From these one can generate the entire algebra. The
linear dependence of Cartan generators results in linear dependence of the roots
1 + 2 2 + 3 = 0.
(B.10)
B.2. Symmetric KacDynkin diagram

The symmetric KacDynkin diagram with two isotropic fermionic roots corresponding to
FBBF grading with the dual basis ( 1 |0 , 3 | 2 ) possesses the simple root system
1
1
2 = ( 0 3 ),
1 = ( 1 0 ),
2
2
The SerreChevalley basis is
h1 = E 11 E 00 = L0 T 0 ,
e3+ = E 32 = V2 ,
e = E 23 = V 2,+ ,
= E 01 = V1+ ,
(B.11)
h2 = E 00 E 33 = 2L0 ,
h3 = E 33 + E 22 = L0 T 0 ,
e+ = E 03 = L+ ,
e+ = E 10 = V 1, ,
1
e1
1
3 = ( 3 2 ).
2
e2 = E 30 = L ,
(B.12)
with their commutation relations determined by the Cartan matrix (3.25),

[hp , eq ] = Apq eq ,
{e1+ , e1 } = h1 ,
[ep , eq }p=q = 0,
[hp , hq ] = 0,
[e2+ , e2 ] = h2 ,
{e3+ , e3 } = h3 .
(B.13)
Appendix C. Excitation numbers

Here we will present the oscillator realization of sl(2|2) algebra which is useful in relating
the number of excitations in nested Bethe ansatz to the eigenvalues of Cartan generators [46].
One introduces [17,23,43,65] bosonic and fermionic raising (a , b , ca ) and lowering (a, b, ca )
operators, which obey the commutation relations

a, a = 1,
(C.1)
b, b = 1,
ca , cb = ab .
The generators then read8
L = ab,
L+ = a b ,
1
1
1
L0 = a a + b b + ,
2
2
2
1
V a, = aca ,
Tb a = ca ca ba cc cc ,
2
1
1
B = a a + b b,
Va+ = a ca ,
4
4
V a,+ = b ca ,
Va = bca ,
(C.2)
8 These have hermitian conjugation properties identical to ones in differential representation (A.25) endowed with
sl(2|2) invariant scalar product (2.8), see Ref. [29].
391
and the oscillators obey the vanishing central charge condition

a a b b + ca ca 1 = 0.
(C.3)
The nested Bethe ansatz for symmetric KacDynkin diagram is built on an L-site first level
pseudovacuum state
|
L = |Z1 Z2 . . . ZL
,
(C.4)
which in the basis of local Wilson operators corresponds to the product Z L (0). There are four
different types of excitations on each spin chain site9 |Z
identified with particle content of the
sl(2|2) subsector as follows,
a b |Z
= |D+ Z
,
c1 c2 |Z
= |X
,
a c2 |Z
= |
,
b c1 |Z
= |
.
(C.5)
The one-site ground state and one-site elementary excitations have the following nonvanishing
chirality and isospin
1
1
1
1
B|
= |
,
(C.6)
B|
= |
,
T 0 |X
= |X
,
T 0 |Z
= |Z
.
4
4
2
2
The ground state of the second level in the nested Bethe ansatz is chosen in terms of noncompact primary excitations D+ Z(0) on each site of the chain. The third vacuum state can be chosen
in term of fermions, either or . Thus the excitations on the spin chain sites are built in terms
of raising operators corresponding to simple roots acting on nodes of the KacDynkin diagram
such that a general state reads schematically

n
|O
= (e3+ )n3 (e1+ )n1 (e2+ )n2 |
L = (V2 )n3 V 1, 1 (L+ )n2 |
L
n n n n n
(c2 )n3 c1 1 a 2 1 b 2 3 |
L .
(C.7)
This formula requires clarifications. First, since the fermionic generators are nilpotent they all
have to act on different spin chain sites, thus their number cannot exceed the number of sites, i.e.,
0 n1,3 L. For instance, for n3 = 0 and n1 = L and n2 n1 , the resulting state corresponds to
the L-fermion operator (D+ )n2 n1 L (0). Second, since both e1+ and e3+ annihilate the vacuum
state |
L , this gives a natural restriction on the number of excitations acting on the same sites,
i.e., n1 n2 n3 . Third, the number of noncompact excitations n2 is unrestricted from above.
Now, the number-of-excitation operators
Na = a a,
Nb = b b,
Nc1 = c1 c1 ,
Nc2 = c2 c2 ,
(C.8)
can be related to generators of the Cartan subalgebra of psl(2|2) and uB (1) automorphism
1
1
Na = L0 2B L,
Nb = L0 + 2B L,
2
2
1
1
Nc1 = T 0 + 2B + L,
(C.9)
Nc2 = T 0 2B + L,
2
2
and yield the following relations between the excitation numbers np and eigenvalues , t and b
of L0 , B and T 0 , respectively, for the state (C.7)
9 This state is created from true vacuum with su(4) oscillators (d , c , c , d ) as |Z
= |
= c2 d2 |0
.
1 1 2 2
34
392
1
n1 = nc1 = t + 2b + L,
2
1
n3 = nc2 = t 2b + L.
2
n2 = na + nc1 = nb + nc2 = t + ,
(C.10)
Appendix D. Transfer matrices for low-dimensional representations

We give here explicit expressions for transfer matrices with low-dimensional auxiliary space.
The transfer matrix conjugate to Eq. (4.17) reads

Q (3) (u 2i ) + (x + ) Q(2) (u + i)
1
+ )+ 1 (x + )
+
L

(x
+
2
e2
e
t[1] (x) = (x )
Q(2) (u)
Q (3) (u + 2i )

Q (1) (u + 2i ) (x ) Q(2) (u i)
1
1
2 (x )+ 2 + (x ) .
+ (x )L
e
e
Q(2) (u)
Q (1) (u 2i )
(D.1)
The generating function (4.11) yields the eigenvalues of antisymmetric transfer matrix t[2]
1
e 2 (x) 2 + (x) t[2] (x)

L Q(2) (u 2i ) Q (3) (u + i) 1 (x [2] )+ 1 (x [2] )

2
e2 +
= xx [2]
Q(2) (u + 2i ) Q (3) (u i)
3i
(2)
(x [2] ) Q (u 2 )
e
Q(2) (u 2i )
(1) (u i) 1 (x [+2] )+ 1 (x [+2] )

L Q(2) (u + 2i ) Q
2
+ xx [+2]
e2 +
(1) (u + i)
Q(2) (u 2i ) Q
(2) (u + 3i )
(1)
(3)
[+2] Q
2L Q (u + i) Q (u i)
2
x
e+ (x )
(3) (u)
(1) (u)
Q(2) (u + 2i )
Q
Q
1
(e 2 + (x) Q(2) (u + 2i ) e 2 (x) Q(2) (u 2i ))2

Q(2) (u + 2i )Q(2) (u 2i )
and its conjugate t[2]

1
e 2 (x) 2 + (x) t[2] (x)

L Q(2) (u + 2i ) Q (3) (u i) 1 (x [+2] )+ 1 (x [+2] )

2 +
e2
= xx [+2]
Q(2) (u 2i ) Q (3) (u + i)
3i
(2)
+ (x [+2] ) Q (u + 2 )
e
Q(2) (u + 2i )
(1) (u + i) 1 (x [2] )+ 1 (x [2] )
[2] L Q(2) (u 2i ) Q
2 +
+ xx
e2
(1) (u i)
Q(2) (u + 2i ) Q
(2) (u 3i )
[2] Q
2
e (x )
Q(2) (u 2i )
(D.2)
x 2L
(1) (u + i)
(3) (u i) Q
Q
(3) (u)
Q
Q (1) (u)
393
(e 2 (x) Q(2) (u 2i ) e 2 + (x) Q(2) (u + 2i ))2

Q(2) (u 2i )Q(2) (u + 2i )
(D.3)
Similarly, one finds from Eq. (4.12) for symmetric t {2}

L (x [2] ) Q(2) (u 3i2 )

1
[2] )+ 1 (x [2] )
2 + (x
2
e
t {2} (x) = x [2] x [+2]
e
Q(2) (u 2i )
(3)

(2) (u + 3i )
(1) (u)
1
Q
Q (u)
[+2] Q
[+2] )+ 1 (x [+2] )
2
2 + (x
2
e
e+ (x )
i
(3) (u i) Q
(1) (u + i)
Q(2) (u + 2 )
Q

3i
(2)
L

(u 2 )
1
[2] Q
[2] )+ 1 (x [2] )
2 + (x
2
e
+ xx [2] e (x) e (x )
Q(2) (u 2i )
(3) (u + i)
Q(2) (u 2i ) Q
i
(3) (u i)
Q(2) (u + 2 ) Q

[+2] L (x) (x [+2] ) Q(2) (u + 3i2 )
1
+ (x [+2] )+ 12 (x [+2] )
+
+
2
e
e
+ xx
e
Q(2) (u + 2i )
(1) (u i)
Q(2) (u + 2i ) Q
,
(1) (u + i)
Q(2) (u i ) Q
(D.4)
and t{2}

(2) (u + 3i )

1
[+2] Q
(x [+2] )+ 12 + (x [+2] )
2
2
t {2} (x) = x [+2] x [2] L e+ (x )
e
Q(2) (u + 2i )
(3)

3i
(2)
(1) (u)
1
Q
Q (u)
(x [2] ) Q (u 2 )
(x [2] )+ 12 + (x [2] )
2
e
e
(3) (u + i) Q
(1) (u i)
Q(2) (u 2i )
Q

[+2] L (x) (x [+2] ) Q(2) (u + 3i2 )
1
(x [+2] ) 12 + (x [+2] )
+
+
2
e
e
+ xx
e
Q(2) (u + 2i )
(3) (u i)
Q(2) (u + 2i ) Q
(3) (u + i)
Q(2) (u 2i ) Q

(2) (u 3i )
L

1
[2] Q
[2] )+ 1 (x [2] )
2
2 (x
2 +
e
+ xx [2] e (x) e (x )
Q(2) (u 2i )
(1) (u + i)
Q(2) (u 2i ) Q
.
(1) (u i)
Q(2) (u + i ) Q
2
References
[1] L.N. Lipatov, Phys. Lett. B 309 (1993) 394;
L.N. Lipatov, JETP Lett. 59 (1994) 596.
[2] L.D. Faddeev, G.P. Korchemsky, Phys. Lett. B 342 (1995) 311.
(D.5)
394
[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]
A.P. Bukhvostov, G.V. Frolov, E.A. Kuraev, L.N. Lipatov, Nucl. Phys. B 258 (1985) 601.
V.M. Braun, S.E. Derkachov, A.N. Manashov, Phys. Rev. Lett. 81 (1998) 2020.
V.M. Braun, S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 553 (1999) 355.
A.V. Belitsky, Phys. Lett. B 453 (1999) 59;
A.V. Belitsky, Nucl. Phys. B 574 (2000) 407.
A.V. Belitsky, A.S. Gorsky, G.P. Korchemsky, Nucl. Phys. B 667 (2003) 3.
A.V. Belitsky, S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Phys. Lett. B 594 (2004) 385;
A.V. Belitsky, S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 708 (2005) 115.
L.N. Lipatov, Evolution equations in QCD, in: S. Boffi, C. Ciofi Degli Atti, M. Giannini (Eds.), Perspectives in
Hadronic Physics, World Scientific, Singapore, 1998, p. 413.
J.A. Minahan, K. Zarembo, J. High Energy Phys. 0303 (2003) 013.
N. Beisert, C. Kristjansen, M. Staudacher, Nucl. Phys. B 664 (2003) 131.
N. Beisert, Nucl. Phys. B 676 (2004) 3.
N. Beisert, M. Staudacher, Nucl. Phys. B 670 (2003) 439.
C.N. Yang, Phys. Rev. Lett. 19 (1967) 1312.
N. Beisert, V. Dippel, M. Staudacher, J. High Energy Phys. 0407 (2004) 075.
M. Staudacher, J. High Energy Phys. 0505 (2005) 054.
N. Beisert, M. Staudacher, Nucl. Phys. B 727 (2005) 1.
R.J. Baxter, Ann. Phys. 70 (1972) 193;
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, London, 1982.
A.V. Belitsky, G.P. Korchemsky, D. Mller, Nucl. Phys. B 768 (2007) 116.
A.V. Belitsky, Phys. Lett. B 643 (2006) 354.
A.V. Belitsky, Phys. Lett. B 650 (2007) 72.
N.Yu. Reshetikhin, Lett. Math. Phys. 7 (1983) 205;
N.Yu. Reshetikhin, Sov. Phys. JETP 7 (1983) 691;
N.Yu. Reshetikhin, Theor. Math. Phys. 63 (1985) 555;
N.Yu. Reshetikhin, Lett. Math. Phys. 14 (1987) 235.
N. Beisert, Phys. Rep. 407 (2004) 1.
N. Beisert, V.A. Kazakov, K. Sakai, K. Zarembo, J. High Energy Phys. 0507 (2005) 030.
B.I. Zwiebel, J. High Energy Phys. 0602 (2006) 055.
A.V. Belitsky, G.P. Korchemsky, 2005, unpublished.
L. Brink, O. Lindgren, B.E. Nilsson, Nucl. Phys. B 212 (1983) 401;
L. Brink, O. Lindgren, B.E. Nilsson, Phys. Lett. B 123 (1983) 323.
S. Mandelstam, Nucl. Phys. B 213 (1983) 149;
S. Mandelstam, Phys. Lett. B 121 (1983) 30.
A.V. Belitsky, S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 722 (2005) 191.
P.P. Kulish, J. Sov. Math. 35 (1986) 2648.
N.Yu. Reshetikhin, P.B. Wiegmann, Phys. Lett. B 189 (1987) 125;
N.Yu. Reshetikhin, Lett. Math. Phys. 14 (1987) 235.
P.P. Kulish, E.K. Sklyanin, J. Sov. Math. 19 (1982) 1596.
F.H.L. Eler, V.E. Korepin, K. Schoutens, Int. J. Mod. Phys. B 8 (1994) 3205.
F.H.L. Eler, V.E. Korepin, Int. J. Mod. Phys. B 8 (1994) 3243.
K. Schoutens, Nucl. Phys. B 413 (1994) 675.
P.P. Kulish, N.Y. Reshetikhin, E.K. Sklyanin, Lett. Math. Phys. 5 (1981) 393;
P.P. Kulish, E.K. Sklyanin, Lect. Notes Phys. 151 (1982) 61.
Z. Tsuboi, J. Phys. A 30 (1997) 7975;
Z. Tsuboi, Physica A 252 (1998) 565;
Z. Tsuboi, J. Phys. A 31 (1998) 5485.
V.V. Bazhanov, N.Yu. Reshetikhin, Int. J. Mod. Phys. A 4 (1989) 115;
V.V. Bazhanov, N.Yu. Reshetikhin, J. Phys. A 23 (1990) 1477.
A. Kuniba, J. Suzuki, Commun. Math. Phys. 173 (1995) 225.
A.B. Balantekin, I. Bars, J. Math. Phys. 22 (1981) 1149.
V.G. Kac, Commun. Math. Phys. 53 (1977) 31;
V.G. Kac, Adv. Math. 26 (1977) 8.
P. Grozman, D. Leites, Czech. J. Phys. 51 (2001) 1, hep-th/9702073.
L. Frappat, P. Sorba, A. Sciarrino, Dictionary on Lie Superalgebras, Academic Press, London, 2000, hep-th/
9607161.
395
[44] F. Woynarowich, J. Phys. C 16 (1983) 6593;

P. Bares, J.M.P. Carmelo, J. Ferrer, P. Horsch, Phys. Rev. B 46 (1992) 14624.
[45] F.H.L. Eler, V.E. Korepin, Phys. Rev. B 46 (1992) 9147.
[46] A. Foerster, M. Karowski, Nucl. Phys. B 396 (1993) 611.
[47] F. Ghmann, A. Seel, J. Phys. A 37 (2004) 2843.
[48] I.V. Cherednik, Funct. Anal. Appl. 20 (1986) 76;
I.V. Cherednik, Duke Math. J. 54 (1987) 563.
[49] W. Fulton, Young Tableaux, Cambridge Univ. Press, Cambridge, 1997.
[50] I. Krichever, O. Lipan, P. Wiegmann, A. Zabrodin, Commun. Math. Phys. 188 (1997) 267.
[51] A.N. Kirillov, N.Yu. Reshetikhin, J. Phys. A 20 (1987) 1565.
[52] A. Kuniba, Y. Ohta, J. Suzuki, J. Phys. A 28 (1995) 6211.
[53] A. Klmper, P. Pearce, Physica A 183 (1992) 304.
[54] A. Kuniba, T. Nakanishi, J. Suzuki, Int. J. Mod. Phys. A 9 (1994) 5215.
[55] V. Kazakov, A. Sorin, A. Zabrodin, Supersymmetric Bethe ansatz and Baxter equations from discrete Hirota dynamics, hep-th/0703147.
[56] N. Beisert, B. Eden, M. Staudacher, J. Stat. Mech. 0701 (2007) P021.
[57] N. Dorey, D.M. Hofman, J. Maldacena, On the singularities of the magnon S-matrix, hep-th/0703104.
[58] N. Beisert, J. Stat. Mech. 0701 (2007) P017.
[59] O. Lipan, P.B. Wiegmann, A. Zabrodin, Mod. Phys. Lett. A 12 (1997) 1369.
[60] E.K. Sklyanin, Lect. Notes Phys. 226 (1985) 196;
E.K. Sklyanin, Quantum inverse scattering method. Selected topics, in: M.-L. Ge (Ed.), Quantum Group and Quantum Integrable Systems, World Scientific, Singapore, 1992, pp. 6397, hep-th/9211111;
E.K. Sklyanin, Progr. Theor. Phys. Suppl. 118 (1995) 35.
[61] R.A. Janik, Phys. Rev. D 73 (2006) 086006.
[62] N. Beisert, The su(2|2) dynamic S-matrix, hep-th/0511082.
[63] A.V. Belitsky, S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Phys. Rev. D 70 (2004) 045021.
[64] S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Front. Phys. 58 (1983) 1.
[65] M. Gunaydin, N. Marcus, Class. Quantum Grav. 2 (1985) L11;
I. Bars, M. Gunaydin, Commun. Math. Phys. 91 (1983) 31;
M. Gunaydin, C. Saclioglu, Commun. Math. Phys. 87 (1982) 159.

SLE in self-dual critical Z(N ) spin systems:

CFT predictions
Raoul Santachiara
CNRS-Laboratoire de Physique Thorique, Ecole Normale Suprieure, 24 rue Lhomond, 75231 Paris Cedex 05, France
Received 17 July 2007; received in revised form 16 September 2007; accepted 19 September 2007
Abstract
The SchrammLoewner evolution (SLE) describes the continuum limit of domain walls at phase transitions in two-dimensional statistical systems. We consider here the SLE in Z(N) spin models at their selfdual critical point. For N = 2 and N = 3 these models correspond to the Ising and three-state Potts model.
For N 4 the critical self-dual Z(N) spin models are described in the continuum limit by non-minimal
conformal field theories with central charge c 1. By studying the representations of the corresponding
chiral algebra, we show that two particular operators satisfy a two level null vector condition which, for
N 4, presents an additional term coming from the extra symmetry currents action. For N = 2, 3 these operators correspond to the boundary conditions changing operators associated to the SLE16/3 (Ising model)
and to the SLE24/5 and SLE10/3 (three-state Potts model). We suggest a definition of the interfaces within
the Z(N) lattice models. The scaling limit of these interfaces is expected to be described at the self-dual
critical point and for N 4 by the SLE4(N+1)/(N +2) and SLE4(N+2)/(N +1) processes.
PACS: 11.25.Hf; 05.10.Gg
Keywords: Stochastic growth processes; Conformal field theory; Critical lattice models
1. Introduction
The SchrammLoewner evolutions (SLEs) are random growth processes that generate conformally invariant curves in two dimensions (2D). SLEs yield probability measures on the continuum limit of non-crossing interfaces in 2D statistical lattice models at criticality with conformal
E-mail address: raoul.santachiara@lpt.ens.fr.
doi:10.1016/j.nuclphysb.2007.09.029
R. Santachiara / Nuclear Physics B 793 [FS] (2008) 396424
397
invariance. SLEs has been proved successful to a more complete, and in some case mathematically rigorous, description of fractal curves in critical percolation [1], loop erased walk [2], level
lines of height models [3] and domain boundaries at phase transitions (see e.g. [4] and references
therein). The SLEs consider directly the geometrical characterization of non-local objects in 2D
critical systems and complement the powerful tools provided by the conformal field theories
(CFTs) techniques.
On the other hand, CFTs focus on correlation functions of local operators which are the
scaling limit of lattice variables. A first family of CFTs, the minimal CFTs, calculates these
correlation functions by studying the infinite constraints imposed by the conformal invariance in
2D systems. The Hilbert space of these theories is constructed from the highest weight representations of the Virasoro algebra formed by the generators of the conformal symmetry.
The relation between SLEs and CFTs has been worked out in [57]. In this respect, an important role is played by the boundary conditions changing operators (b.c.c.) which generate the
boundary conditions from which the curve grows [8,9]. The key property of these operators is
to satisfy particular relations under the action of the conformal symmetry generators which lead
to second order differential equations for their correlation functions. The link between the SLEs
and CFTs is derived by comparing these second order equations to the FokkerPlanck equations
coming from the Brownian motion driving the SLE.
The SLEs/CFTs connection is well established in the case of minimal CFTs, which have a
central charge c 1. However, many critical models in condensed matter and statistical physics
posses, in addition to the conformal invariance, symmetries in some internal degree of freedom,
such as the SU(2) spin-rotational symmetry in critical quantum spin chains [10] or the ZN symmetry in spin lattice models [11,12]. In the continuum limit, these additional symmetries are
enhanced by the presence of chiral currents which form, together with the energymomentum
tensor, more structured algebras which present the Virasoro one as a sub-algebra. The space of
local fields occurring in a CFT with additional symmetry, the non-minimal CFT, corresponds
to the representations of the associated chiral algebra. These theories have in general a central
charge c 1.
The connection between SLEs and non-minimal CFTs has been first addressed in [13,14],
where the relation between stochastic evolutions and superconformal field theory was investigated. More recently, the connection between SLE and WessZuminoWitten models, i.e. CFTs
with additional Lie-group symmetries, has been studied by very different approaches [15,16].
In particular, it has been shown in [16] that a consistent SLE approach is possible provided
that an additional stochastic motion in the internal symmetry group space in introduced. For an
SU(2) symmetry, for instance, the SLE describes a process which carries a fluctuating additional
spin 1/2 degree of freedom.
Inspired by these results, we discuss in this paper a possible SLE approach to CFTs with additional ZN symmetry. The conserved currents associated to the extra symmetry, called parafermions, have fractional spin. We consider the first of such parafermionic theories with central charge
c = 2(N 1)/(N + 2) [11]. The Z(2) and Z(3) models coincide with some particular minimal
CFT where the domain wall boundary conditions and the associated b.c.c. operator are known.
We exploit these pieces of information to identify the possible b.c.c. operators for general ZN
symmetry. We show that, under the action of the parafermionic conserved currents, these operators satisfy algebraic relations similar to the ones found in the case of minimal models. From
these CFT predictions and in analogy with the approach proposed in [16], we introduce an additional stochastic motion driven by the action of the parafermionic currents.
398
The strong motivation to study these theories lies on the fact that they describe the continuum
limit of critical lattice models, the N -states generalization of the Ising model which reduce to the
Ising and three state Potts model for N = 2 and N = 3. This allows for the identification of the
SLE interfaces on the lattice and opens the possibility of a direct numerical verification.
The paper is organized as follows. We review the connection between SLE and minimal CFTs
in Section 2. Section 3 provides the extension to CFTs with extra Lie symmetries. In Section 4,
after an introduction to the operator content of the parafermionic theories, we define a stochastic
motion in the ZN internal space. Further, by studying the parafermionic algebra representations,
we derive some algebraic relations related to the b.c.c. operators. In Section 5 we shall discuss
the definition of interfaces on the lattice. We conclude the paper in Section 6.
2. Stochastic Loewner evolutions and conformal field theories
2.1. The SLE /Minimal models connection
2.1.1. Chordal SLE: Definition
Throughout this paper, we consider chordal SLE which describes random curves joining two
boundary points of a connected planar domain. For a comprehensive and detailed introduction
to SLE, see e.g. [4,17,18]. The definition of SLE is most conveniently given in the upper half
complex plane H: it describes a fluctuating self-avoiding curve t which emanates from the origin
(z = 0) and progresses in a properly chosen time t . If t is a simple curve, this evolution is defined
via the conformal map gt (z) from the domain Ht = H/]0,t] , i.e. the upper half plane from which
the curve is removed, to H. In the more general case of non-simple curves, the function gt (z)
produce conformal maps from Ht = H/Kt to H where Kt is the SLE hull at time t . The SLE
map gt (z) is uniquely determined by fixing three real parameters. A conventional normalization
(the so-called hydrodynamic normalization) is such that gt (z) = z + 2t/z + near z = . With
this choice the time t corresponds to the upper half-plane capacity cKt of the hull Kt , cKt = 2t .
The function cKt is a positive quantity and satisfies the additive law cKt g 1 (K ) = cKt + cKt .
t
t
The SLE map gt (z) is a solution of the Loewner equation:
d
2
,
gt (z) =
dt
gt (z) t
gt=0 (z) = z,
(1)
where t is a real valued process, t R, which drives the evolution of the curve. For a system
which satisfies the Markovian and conformal invariance properties, together with the leftright
symmetry, the process t is shown [19] to be proportional to a Brownian motion: E[t ] = 0 and
E[t s ] = min(s, t). In the following, we use the symbol E[. . .] for the stochastic average over
the Brownian motion.
2.1.2. Martingales in SLE processes
In [57,20] it was understood how to associate CFT states with the growing curves of the
Loewner process. To reveal the SLE/CFT connection, the first step is to define stochastic process
such that conditioned correlation function are martingales for this process. The main idea is to
arrange the statistical sum by first summing over all the configurations Ct presenting a trace t
of the interface with fixed shape and capacity t/2 and then summing aver all the possible shapes
of the trace. The first sum defines a conditioned correlation function while the second sum can
be seen as the stochastic mean of this correlation function. To make it more concrete, consider
an observable O of a lattice model defined on the domain H. The statistical sum OH can be
399
written as:

OH = E Ot =
P [Ct ]Ot ,
(2)
where 0t is the statistical average conditioned to the presence of the trace t and P [Ct ] is
the probability of its occurrence. The correlator OH does not depend on the choice of t : the
stochastic mean of the correlator 0t is thus time independent and it is a martingale.
2.1.3. Martingales and conformal correlators
The relation (2) becomes extremely useful at the critical point where the model is expected
to be described in the continuum limit by a conformal field theory. In this case, the
operator O
corresponds generally to a product of, say, l primary fields i (zi ), O({zi }) = li=1 i (zi ) at
positions zi . The image f of under a conformal transformation f (z) is

f
(3)
i = z f (z) i i f (zi ) ,
where i is the conformal dimension of i . The statistical expectation values can then be expressed in terms of CFT correlation functions F({zi })Ht :

O({zi })()(zt )
Ht
Ot F {zi } H =
t
()(zt )
Ht
(4)
where
Ht indicates the conformal correlation function computed in the domain Ht = H/t ,
i.e. the upper half plane with the trace t removed. The fields (zt ) and (), inserted respectively at the tip zt of t and at the infinity, are the b.c.c. operators implementing the boundary
conditions at which the interface anchor.
Using the conformal invariance, the correlation functions F ({zi })Ht in the domain wall Ht
can be expressed as the correlation functions in the upper half plane H:

gt O({zi })()(t )
H
.
F {zi } H =
t
()(t )
H
(5)
Note that the Jacobians coming from the conformal transformation on the fields cancel between the numerator and the denominator in the above expression.
We are now in the position to understand the SLE/CFT connection. Under the SLE, the trace t
evolves and the iterated sequence of infinitesimal conformal mappings gt (z) satisfying Eq. (1)
leads to a Langevin dynamics for the conformal correlator F ({zi })Ht . Imagine evolving the trace
for a time t and then, for an infinitesimal time dt. Using Eq. (1), the variation d(gt i (zi )) =
gt+dt (z ) gt (z ) is given by:
i i
i

g
1
dgt (w) T (w)gt i (zi )
d t i (zi ) =
2i
zi

gt (zi )
i
gt
+
i (zi )
= 2 dt
(6)
(gt (zi ) t )2 gt (zi ) t
where T (z) is the energymomentum tensor. The variation for the s are given by the Ito differential:

d (t ) = t (t ) dt + 2t (t ) dt.
(7)
2
400
Using Eqs. (7)(6) in Eq. (5) and averaging over all the realization of t , one obtains the diffusion
equation:

gt (zi )
i
2
+
+
E F {zi } H .
t E F {zi } H = 2
(8)
t
2
t
t
gt (zi ) t
2
(gt (zi ) t )
i
As previously said, the correlation function F ({zi })Ht is a martingale of the SLE process and
satisfies therefore the following differential equation

gt (zi )
i
2
+
+
2
(9)
t F {zi } Ht = 0.
2
gt (zi ) t
2
(gt (zi ) t )
i
2.1.4. Operator formalism and minimal models

In order to show the consequences of the above relation, it is convenient to introduce the
operators Ln which are the modes of the energymomentum tensor T (z):

1
Ln (z) =
(10)
dw w n+1 T (w)(z).
2i
z
The Ln are the generators of the conformal transformation in the CFT Hilbert space. A primary
operator , satisfying Eq. (3), is annihilated by the positive modes of Ln , Ln = 0 for n > 0, and
the eigenvalue of the zero mode is its conformal weight, L0 = . The Ln satisfy the Virasoro
algebra
c
[Ln , Lm ] = (n m)Ln+m + n2 (n 1)n+m .
(11)
12
All local fields of the theory can be obtained by applying the Ln , n 1, to a primary fields .
Each set of states of the form Lk1 Lkm forms a conformal family [] and corresponds to
the representation of the highest weight of the Virasoro algebra, also called Verma module,
the
primary field corresponding to the highest vector. The descendant state Lk1 Lkm ,
km = n, is said to be the nth level of the module of and its conformal weight is + n.
For c 1, the unitary representations of the Virasoro algebra has central charge [21]:
c=1
6
,
m(m + 1)
m = 2, 3, . . . ,
(12)
and correspond to the simplest family of conformal theories, called minimal models Mm . The
minimal model Mm is characterized by a finite number of Verma modules [r,s ] with conformal
weight:
r,s =
[(m + 1)r ms]2 1

,
4m(m + 1)
r = 1, 2, . . . , m 1, s = 1, 2, . . . , r.
(13)
2.1.5. Null vectors in CFT and SLE

In CFT, the differential equations satisfied by the correlation function can be derived from
the structure of the representation modules and, in particular, from the relations between the
descendant states. One can show that F ({zi })Ht is a solution of (9) if the b.c.c. operator obeys
the condition:

L2 L21 = 0.
(14)
4
401
Eq. (14) means that the descendant states L21 and L2 are linearly dependent. The operator transforms then as a degenerate representation of the Virasoro algebra which has a null state
at level 2. The operator can thus be identified with the operators 1,2 or 2,1 of the minimal
models Mm which are shown to satisfy:

3
2
L2
(15)
L
1,2 = 0,
2(21,2 + 1) 1

3
L2
(16)
L2 2,1 = 0.
2(22,1 + 1) 1
Comparing Eqs. (15) and (16) to Eq. (14) and using Eqs. (12) and (13), one can finally make
explicit that the SLE/CFT connection: an SLE process with parameter describes interfaces in
CFT with central charge:
(6 )(3 8)
,
2
and the boundary conformal operator has scaling dimension:
c =
(17)
(6 )
.
(18)
2
This derivation shows that martingales for SLE processes are closely related with the existence
of null vectors in the appropriate Verma module.
=
3. SLEs for CFTs with extra Lie group symmetries

As mentioned above, the operator content of the minimal models Mm , which have central
charge c 1, corresponds to the space of representations of the Virasoro algebra (11). In this
case, the conformal symmetry determines completely the spectrum of the theory and the differential equations satisfied by the correlation functions.
The applications of the SLEs defined in the previous section are limited to CFTs with central charge c 1. This can be directly understood from the fact that, for a CFT with extended
symmetry, not all the local fields can be obtained by the application of the Ln . In particular, one
expects at each level additional states and the linear relations of the type (15)(16) will include
additional terms.
3.1. SLEs and CFTs with Lie group symmetries
The SU(2) WZW models are among the most important CFTs with additional symmetries.
These CFTs posses an internal SU(2) continuous symmetry which, together with the conformal
one, is realized by the set of chiral currents J a (z), a = +, , 0, with scaling dimension J = 1.
The current algebra, derived from the currentcurrent operator product expansion (OPE), takes
the form:
a b
c
Jn , Jm = fca,b Jn+m
(19)
+ N nn+m ,

where Jna are the current modes, Jna = 1/(2i) z dw w n J a (w), the fca,b are the structure
constants of the su(2) Lie algebra and N is the level of the algebra. The representations of
the KacMoody algebra, leads to a family of CFTs, denoted SU(2)N , with central charge
402
cN = 3N/(N + 2). The local fields j (z) of the theory transform under the action of the chiral currents in the representation space of dimension j (j + 1) of SU(2). Analogously to the
Virasoro case, a WZW primary field is defined by:
Jna j (z) = 0 for n > 0;
J0a j (z) = tja j (z),
(20)
where the tja are the SU(2) generator matrices in the j representation. The conformal weight j
of a primary field j (z) is
j (j + 1)
j (z).
(21)
N +2
Under an infinitesimal conformal dg(z) transformation together with an infinitesimal SU(2)
gauge transformation d a (z), the primary fields j (z) changes as:

1
1
dw dg(w) T (w)j (z) +
dw d a (w) J a (w)j (z).
d(j )(z) =
(22)
2i
2i
L0 j (z) = j j (z) =
In [16], the SLE approach has been extended to these theories. It proposes to describe this
model by a composition of two independent Brownian motions, one in the physical space, defined
as in Eq. (1), and one in the internal SU(2) spin degrees of freedom:
dta (z) =
dta
,
gt (z) t

E ta sb = a,b min(s, t),
(23)
where the variance is an independent parameter of the SLE evolution. The idea is then to
attach to trace t a spin 1/2 degrees of freedom. The interface then undergoes both a standard
SLE evolution in the physical space and a stochastic SU(2) rotation.
The derivation of the SLE/CFT connection is strictly analogous to the one shown before. The
difference here is that the b.c.c. operator (t ) in Eq. (5) carries a spin 1/2, 1/2 (t ) which
transforms under the representation 1/2 . Using Eqs. (1) and (23) in Eq. (22), the Ito formula for
1/2 (t ), written in terms of the modes Ln and Jna , is:

2
a a
a a
d 1/2 (t ) = dt L1 1/2 (t ) + d J1 1/2 (t ) + dt
L + J J
1/2 (t ).
2 1 2 1 1
(24)
The operator O is now a product of WZW primaries and remain fixed while the trace evolves.
This means one has only to consider their variation (6) under the conformal transformation gt
which maps the domain Ht in the domain Ht . The martingale condition (14) takes the form:

2
a a
L2 L1 J1 J1 1/2 = 0.
(25)
4
4
The correlation functions in WZW theories satisfy first order differential equations, called
KnizhnikZamolodchikov equations, which are derived from a relation between descendant
states at the first level:

1
a a
L1 +
(26)
t jm = 0.
J1
N +2
Applying the operator L+1 to Eq. (25) and using the commutation relation (19) together with
a , one fixes a first relation between , and N , /(6 (2
[Ln , Jma ] = mJn+m
1/2 + 1)) =
2/(N + 2). It was shown [16] that a second condition, + = 4, can be obtained by demanding
403
the one-point function of the current to exhibit a simple pole at the tip of the trace zt . The final
result is:
=4
N +2
,
N +3
4
.
N +3
(27)
4. SLEs in Z(N) parafermionic theories

We consider the CFTs with extended ZN symmetries, the so-called Z(N) parafermionic
theories. In particular, we focus on the Z(N) parafermionic theories with central charge c =
2(N 1)/(N + 2) (c > 1 for N > 4). These theories describe the continuum limit of the N states spin models interacting via a ZN invariant nearest-neighbor coupling, at their self-dual
critical points (see Section 5). For N = 2 and N = 3 one finds the well-known Ising and threestate Potts model. The corresponding CFTs has central charge c = 1/2 and c = 4/5 respectively
and coincide with the minimal models M3 and M5 : the operator content of these theories can
also be determined by studying the representations of the Virasoro algebra. In this sense, the Z2
and Z3 symmetry of these models are trivially realized. The CFTs describing the critical point
of the Ising and the three-state Potts are then expected to be described by the SLE in Eq. (1).
The b.c.c. conformal operators implementing the SLE interface boundary conditions have been
identified. We will show that these results admit a natural extension to general N .
4.1. Parafermionic current algebra
We briefly review the Z(N) parafermionic theories with central charge c = 2(N 1)/(N + 2).
These theories were introduced and constructed in [11,22]. Here we shall enounce the main
results in a slightly different manner. The general arguments and the notations used here are
strictly analogous to the ones discussed in a series of papers [2326] where a second series of
parafermionic theories was studied.
Extra ZN group symmetries in two-dimensional conformal field theories are generally realized by a set of holomorphic currents k (z) (k = 1, . . . , N 1), satisfying the following operator
product expansion:
k
k,k
k+k
(z) (z ) =
k
+
(z z ) k k k+k
k (z) k (z ) =
(z z )2k

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

2
k
T (z ) +
c
(28)
(29)
k,k
where
k is the conformal dimension of the parafermions k (z) and k+k are the structure
constants of the algebra. We shall be interested in parafermionic theories in which the dimenk
sions
k of the parafermions { } take the minimal possible values admitted by the associativity
constraint:
k = Nk =
k(N k)
,
N
k = 0, 1, . . . , N 1.
(30)
Note that the above formula is not symmetric under the exchange k k: the field k in (29)
is assumed to have dimension
Nk , in the sense that the indices k referring to the ZN charge
404
are always defined modulo N . Thus,

+
Nk k k ,
Nk k .
(31)
k,k
The structure constants k+k and the central charge c (of the Virasoro algebra) are given by the
expressions:
(k + k )!(N k)!(N k )!
,
k!k !(N k k )!N!
2(N 1)
c=
.
N +2

k,k
k+k
2
(32)
(33)
4.2. Representation space: The ZN sector

The CFTs we are considering describes the self-dual (KramersWannier invariance) critical
point of ZN -invariant lattice model [11]. In addition to the ZN symmetry, the theory possesses the
dual Z N invariance, which is related to the conservation of the antiholomorphic fields (z). The
Hilbert space of a ZN Z N invariant theory, splits into subspaces characterized by the ZN Z N
charges {p, q}. A field {p,q} in one of these subspaces transform as {p,q} (z) exp[2i(pm +
qn)/N ]p,q (z) under global rotations of angles 2m/N and 2n/N . The lattice spin operator
k (see Section 5) and its dual k are described in the continuum limit by fields of charge {k, 0}
and {0, k} respectively. The (anti-)holomorphic currents k (z) ( (z)), which appear in the OPE
of k k (k +
k ), have charges {k, k} ({k, k}). Recently, the lattice holomorphic realization of
the parafermionic currents has been discussed in [27].
It is in general convenient to express the charge k, k as [q , p ] = {k + k , k k }, where now
q and p are defined mod 2N and q + p even. In this notation, the currents k and (z) have
charge [2k, 0] and [0, 2k] respectively.
In the following we concentrate on the action of the holomorphic field k (z), all the results
being valid also for (z). Thus, without losing any generality, one can study the structure of the
ZN representations which we denote with q . Moreover, we find convenient to use a different

convention q for the ZN charges, defined by q = 2q mod 2N . With this choice, the product
k (z) q is a field with charge k q.
For N odd, we consider then the representation fields
q (z, z ),
q = 0, 1, . . . , (N 1)/2 (N odd).
(34)
For N even, it turns out that the modules of the representation corresponding to q with
N/4 < q N/2 are identical to those of 0 q N/4. In order to recover the right
number of ZN representations one has to consider half-integer value of q:
1
(35)
q = 0, , 1, . . . , N/4 (N even).
2
Naturally, the physical meaning of the half-integer charge is recovered in the usual notation q .
q (z, z ),
4.2.1. Currents modes in ZN sector

The currents { k } can be decomposed into mode operators Ak+n , whose action is to change
the ZN charge of the representation fields q :

1
k (z) q (0) =
(36)
Ak q
q (0),
k+q
k k +n k+q +n
n (z)
k =
q 2 k2
mod 1.
N
405
(37)
The value k is the first level in the module of q corresponding to the ZN charge k: it determines thus the level structure of the modules induced by the ZN representation fields. The
q
values of k can be easily obtained by considering the module of the identity whose descendant
states are the chiral currents k (z). The levels of these operators correspond to their conformal dimensions
k . Taking into account that, owing to the Abelian monodromy of the fields
k
(z) in the ZN sector, the level spacing is equal to 1, one can readily obtain from Eq. (30)
k0 = k 2 /N mod 1. The level structure of a generic field q (z), Eq. (37), is then extracted from
q
the module of the identity by inspecting its corresponding submodule. Note also that k+q = 0
q
for k = 2q. This results in the presence of zero modes A0 :

2q
A0
q (0) = hq q (0)
(38)
q with opposite
which associate at each field q , with conformal dimension
q , the field
charge and with the same conformal dimension, q = q . The eigenvalues hq defined in (38)
characterize the representations of the parafermionic algebra together with the conformal dimension the fields q . As usual, primary fields are defined by Ak q +n q = 0 for n > 0. Each
k+q
representation module is then characterized by the two primary fields q which can be obtained one from the other as in (38). An example of the module structure associated to the field
1 and of the action of the currents is given in Fig. 5 for N = 5. Using the formula (37), we
1
have 01 = 1/5 and 2
= 4/5.
Using the expansion (36), the action of the modes in each sector can be given in terms of a
contour integral:

q
1
k
q
A q +n (0) =
(39)
dz (z)k k+q +n1 k (z) q (0).
2i
k+q
C0
In the ZN sector, one can simply consider the action of the currents { 1 } because, via Eq. (28),
they completely determine the full algebra of the other currents { k }, k = 2, . . . , N/2.
The commutation relations of the mode operators can be deduced from Eq. (39) by using
standard techniques in the complex plane. These relations are given in Appendix B.1.
As shown in the Appendix B.1, the representation space contains N/2+1 primary operators
q with q = 0, 1, . . . , N2 for N odd and with q = 0, 12 , 1, . . . , N/4 for N even. The
conformal dimension of these operators turn out to be:
q =
q(N 2q)
,
N (N + 2)
q = 0, 1, . . . ,
N 1
(N odd),
2
1
N
q = 0, , 1, . . . , (N even).
2
4
(40)
As said above, to each primary

corresponds another primary field
of the same dimension.
In the case N = 2 and N = 3 (i.e. Ising and three state Potts model), one has only two primaries of the parafermionic algebra. For N = 2 one finds the identity operator with 0 = 0 and
the operator 1/2 , with dimension 1/2 = 1/16, corresponding to the Ising model spin operator. Analogously, for N = 3 one has, together with the identity operator, the operator 1
with 1 = 1/15. As previously said, the case N = 2 and N = 3 are special because they can
406
be identified with the minimal model M3 and M5 . The operators 1/2 and 1 are identified respectively to the operators 1,2 and 2,3 of the corresponding M3 and M5 Kac tables, see
Eq. (13). As we will show more in detail below, the representation module corresponding to 1
+1 = A1
1 singlet under Z transformacontains one descendant state, (N=3) = A1
3
1/3
1/3
tion (i.e. q = 0). This singlet (energy) operator has dimension = 1/15 + 1/3 = 2/5, it is a
primary of the Virasoro algebra and can be identified with the operator (N=3) = 2,1 of the M5
Kac table.
4.3. Representation space: The disorder sector
A key observation is that the theory we are considering is actually invariant under the dihedral
group DN which includes ZN as a subgroup. This can directly be seen from the symmetry of
Eqs. (28)(32) under the conjugation of the ZN charge, q N q. At the level of the lattice
model (see Section 5), this comes from the invariance of the Hamiltonian under the transformation k (x) k+ (x).
The space of representation thus includes the N -plet of Z2 operators which we denote as:

Ra (z, z ), a = 1, . . . , N .
(41)
In the following, we refer to these operators as disorder operators.
The theory of disorder operators has been fully developed in [22]. A detailed discussion about
the general properties (products, analytic continuations) of these operators is given in [23].
The most important feature of the disorder operators is their non-Abelian monodromy with
respect to the chiral fields { k }. This amounts to the decomposition of the local products
k (z)Ra (0) into half-integer powers of z:
k (z)Ra (0) =
1
n
k +2
(z)
Akn Ra (0),
2
k = 1, 2, . . . , N/2.
(42)
The expansion of the product k (z)Ra (0) (with k = 1, 2, . . . , N/2) can be obtained by an
analytic continuation of z around 0 on both sides of Eq. (42). The result is:
k (z)Ra (0) =
(1)n
n
k +2
(z)
Akn Rak (0),

2
k = 1, 2, . . . , N/2.
(43)
In accordance with these expansions, the mode operators Akn can be defined by the contour
2
integrals

1
n
k
dz (z)k + 2 1 k (z)Ra (0),
A n Ra (0) =
(44)
2
4i
C0
where the integrations are defined by letting z turn twice around the operator Ra (0) at the origin, exactly as described in [22,23]. The commutation relations between the current modes in
this sector can be computed and the dimension of the primary disorder operators is (see Appendix B.4):
R
s =
N 2 + (N 2s)2
,
16(N + 2)
s = 0, 1, . . . , N/2.
(45)

(0)
407
(1)
R
Note that for N = 2 one has the operators Ra and Ra with dimension R
0 = 1/16 and 1 = 0.
This is of course expected as the dihedral group D2 = Z2 , i.e. the cyclic ZN elements and the
reflection Z2 elements of the group DN coincide for N = 2. In particular one has Ra0 = 1/2
while Ra1 coincides with the identity. For N = 3, the disorder sector contains the operators Ra0
R
1
(R
0 = 1/8) and Ra (1 = 1/40). In terms of primaries of the Virasoro algebra they coincide
respectively with the 1,2 and 2,2 operators of the M5 Kac table, see again Eq. (13).
4.4. Stochastic motion in the internal space

We discuss now a possible SLE approach to describe parafermionic theories. In the previous
section we have seen that the action (39) and (44) of the chiral currents modes on the representation fields amounts to a twist either of the ZN charge q for fields q (z) belonging to the ZN
sector, or of the Z2 index a for fields Ra (z) in the disorder sector. The b.c.c. operator is expected
to transform as an operator in the ZN or in the Z2 sector. By analogy to the case of the SU(2)N
WZW [16], we consider an independent stochastic motion in the additional degrees of freedom.
We provide a parafermionic field the following evolution:

1
1
dgt (w) T (w)X(z) +
dw (d )k (w) k (w)X(z),
X(z) X(z) +
(46)
2i
2i
Cz
Cz
or X(z) = Ra (z) and the contour Cz is a general closed contour around z

where X(z) =
(remember that in the R sector the Riemann surface is two-fold). This evolution contains, in
addition to the variation under an infinitesimal conformal transformation dgt (w), a component
with shifted ZN charge q q + k or Z2 charge, a a + k. The (46) can be seen as a motion
in the parafermionic representation module. In the sequel, we shall concentrate on considering
only the action of the most fundamental 1 fields.
The definition of the internal stochastic motion is given in such a way that the corresponding
Ito derivative of (z) takes the form (7) or (24). More precisely, we want to derive a martingale condition as a linear relation between descendants at the level two of some representation
module. This allows for a direct comparison with the CFT results.
In the case the boundary operator (z) coincides with a primary q in the ZN sector, we
define:
q
(d )kt (z) =
(d )kt
q
k +k+q +1
(gt (z) t )

E tk sk = k,k min(s, t).
(47)
The exponent in the denominator has been chosen according to Eq. (36) which encodes the OPE
between the currents k and the fields q . Note that this evolution is very similar to (23) in the
su(2) Lie algebra. This is somehow expected since the Z(N) parafermionic theory corresponds
to the coset SU(2)N /U (1) (see Appendix A).
In the case when (z) transforms as a disorder Ra field, we define, according to Eq. (42):
(d )kt (z) =
(d )kt
k +2
(gt (z) t )

E tk sk = k,k min(s, t).
(48)
We point out that the definition of the stochastic motion given above is by now purely algebraic and based on a formal analogy with the case SU(2). However, the CFT predictions and the
identification of the possible interface on the lattice model motivate the interest of such construction.
408
4.5. Martingale conditions and parafermionic results

As we will discuss in some detail later, the Ising model (N = 2 and c = 1/2) shows in its FK
representation (see 5) a domain wall described by the SLE16/3 . The b.c.c. operator creating such
interface is identified with the operator 1,2 of the M3 Kac table, what is consistent with Eqs. (17)
and (18). In the parafermionic representations, we have seen that this operator corresponds to the
(0)
disorder Ra operator or, equivalently, to the 1/2 operator (remember that N = 2 is special as
D2 = Z2 ). In the three state Potts model (N = 3), the operator 1,2 of the M5 Kac table is again
(0)
identified again with the disorder Ra and it is expected to create an SLE24/5 , in agreement again
with Eqs. (17) and (18). The operator 2,1 , on the other hand, can be associated to an interface
1
described by the SLE10/3 and it corresponds to the (N=3) = A11/3 1 = A1
1/3 operator
mentioned before.
On the basis of these correspondences, it is then natural to identify for general N the b.c.c.
(0)
operator (z) with the disorder primary operator Ra with dimension 0 = (N 2 + N 2)/
1
1
1
(16(N + 2)) and with the ZN singlet (i.e. charge q = 0) operator (N ) = A1
1/N = A1/N
with dimension = 2/(N + 2). Note that this singlet operator is not a primary of the parafermionic algebra but it appears as descendant in the module of 1 (see Appendix B).
Using Eqs. (30), (37) and (46), the stochastic motions (47) and (48) amounts respectively
to the following dynamics for (z) in the two cases when (z) = (z) and (z)
(0)
R = Ra :

d (t ) = dt L1 (t ) + (d )1 A11/N 1 (t ) + (d )1 A1
1/N 1 (t )

2
1
1
+ dt
+
A
A
(49)
L1 + A11/N 1 A1
1/N 1
1/N 1 1/N 1 (t )
2
2
(0)
and (z) R = Ra :

d R (t ) = dt L1 R (t ) + (d )1 A11 R (t ) + (d )1 A1
1 R (t )

2
1
+ dt
L + A1 A1 + A1
1 A1 R (t ).
2 1 2 1 1
The martingale condition, to be compared with Eqs. (14) and (25), now reads:

1
1
L2 L21 A11/N 1 A1
+
A
A
1/N 1
1/N 1 1/N 1 (t ) = 0,
4
4

1 1
+
A
A
L2 L21 A11 A1
1
1 1 R (t ) = 0.
4
4
(50)
(51)
(52)
We can now enounce the main results of this paper. We have studied the representation mod(0)
ules of the and Ra operator. For N 4 one finds that there are at least two independent states
at the second level of these operators. We have found the following relations at the second level
of these modules:

N +1 2
N3 1
1
1
1
A
A
+
A
A
L1
L2
1/N 1 1/N 1 (t ) = 0,
N +2
(N + 2)3 1/N 1 1/N 1
(53)

4/N
3

N
N +2 2
2
1
L2
(54)
L
A1 A1 + A1
1 A1 R (t ) = 0.
N + 1 1 4(N + 2)(N + 1) 1 1
409
The derivation of these relations is shown in Appendix B. We have verified that, for N = 2, the
relation (54) reduces to (15) and for N = 3, Eqs. (53) and (54) are respectively equivalent to
Eqs. (16) and (15). In these cases, which as already said coincide with minimal models M3 and
M5 , all the descendants of the theory can be obtained by applying negative Virasoro modes. In our
case, this amounts to the fact that the states at the second level constructed from the application
of parafermionic modes can be written as a linear combination of L21 and L2 .
Comparing Eqs. (51) and (52) to Eqs. (53) and (54), we obtain, for N 4 and for the two
SLEs evolutions:
1 = 4
N +1
,
N +2
1 = 4
2 = 4
N +2
,
N +1
2 = 24/N
N3
,
(N + 2)3
N3
.
(N + 2)(N + 1)
(55)
(56)
5. Z(N) self-dual critical spin models

Consider a square lattice with the spin variables j at each sites j taking N possible values:

i2
j = exp
(57)
n(j ) , n(j ) = 0, 1, . . . , N 1.
N
The most general ZN -invariant spin model with nearest-neighbor interactions is defined by the
reduced Hamiltonian [2830]:
H [n] =
N/2

m=1

Jm

2mn
1
cos
N
and the associated partition function reads:

Z=
exp
H n(i) n(j ) .
{ }
(58)
(59)
ij
In fact, the Potts model is recovered in the case Jm = J for all m. The model presents, in this case,
a permutational SN symmetry. Another meaningful model is the clock model which is defined
by the partition function (59) with Jm = J m,1 . Defining the Boltzmann weights:

xn = exp H (n) , n = 0, 1, . . . , N 1,
(60)
the general Z(N) spin system is described by N/2 independent parameters x. The Kramers
and Wannier (order-disorder) duality has proven to be a powerful tool for examining the behavior
of these models. The phase diagram of the model (59) for N = 5 states of spin is
shown in
Fig. 2. It has been show that the Z(5) model is self-dual on the line x1 + x2 = 1/2( 5 1).
Here we are interested in the existence of two critical points (x1 , x2 ) and (x2 , x1 ), where x1 =
sin(/20)/ sin(3/20) 0.34 and x2 = sin(/4)/ sin(7/20)x1 0.27, where the model is
completely integrable. The Z(5) parafermionic theory defined in the previous sections describes
the continuum limit of the lattice model at these points. In general, in the self-dual subspace
of (58)(59), which contains also the Potts model the clock model, the Z(N) spin model is
completely integrable at the points [12]:
410
x0 = 1,
xn
n1

k=0
sin
+
N
4N
1
(k + 1)
.
sin
N
4N
(61)
These critical points are described by the Z(N) parafermionic field theories.
5.1. Boundary conditions and SLE interfaces
Consider the three-states Potts model. From Eqs. (17) and (18), as previously mentioned, the
(N=3)
is expected to create an SLE10/3 trace. In this paragraph we suggest a
b.c.c. operator
definition of the interfaces within the lattice models whose scaling limit is described in these
SLE processes. In the following we indicate the possible values of the spins with the letters
(N=3)
has been shown to generate the boundary condition where
A, B, . . . . The b.c.c. operator
the spins are fixed to (say) the value A on the left side of the origin while can take the value B or
C with equal probability on the right side [9]. With this boundary condition, to which we refer
with the short-hand notation A|B + C, there exists a single domain wall between the spin A and
the spins B and C, as shown in Fig. 1. The statistical properties of this interface has recently
been analyzed by numerical means in [31] and are shown to be well described in the continuum
limit by an SLE10/3 .
The generalization to the N -states spin model is straightforward. Indeed, one can show (see
(N )
Appendix C) that the b.c.c. operator produces the boundary conditions A|B + C + D + ,
where the spins on the left of the origin take the value A while the ones on the right take with
(a)
(b)
Fig. 1. (a) Three state Potts model: interface (green curve) between the A spins (black circles) and the B (full blue circles)
and C (empty blue circles). The boundary conditions A|B + C as in the text. (b) The interface defined in the Z(5) spin
model with boundary conditions A|B + C + D + E. (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this paper.)
411
equal probability the N 1 values of spin different from A. As in the case of N = 3, there
exists a single boundary domain between the A spin and the B, C, . . . . Note that the boundary
conditions do not satisfy the reflection symmetry and the measure on the curve should reflect this
asymmetry. However, in the continuum limit the reflection symmetry is expected to be restored
as explicitly shown in [31]. This can be argued from the fact that, in this limit, the statistical
expectation values can be expressed in terms of CFT correlation functions which contain the
b.c.c. operators, see Eq. (4). These correlation functions satisfy indeed the reflection symmetry.
It is interesting to note that one can assign a color to the interface depending on the values of
the spins (B, C, . . .) at the right of the domain wall. The point is that, for N < 4, the Boltzmann
weights xn , n = 1, 2, . . . , N 1, are equal: the energy of one bond does not depend on the
difference between the values of the nearest-neighbor spins. In this sense, the ZN symmetry is
trivially realized and this is essentially the reason why, for N = 2, 3, the correspondent CFT
can be constructed from the conformal invariance alone. However this is not true for N 4
where the Boltzmann weights xn , n = 1, 2, . . . , N/2 are different. In the continuum limit, this
amounts to a more structured chiral algebra with a central charge c 1. From the point of view
of the lattice model, one can observe that for N = 2, 3 the energy cost to have a spin B or C
at one side of the interface is the same. Thus, as N 4, one expects the internal spin degree of
freedom to play an important role in the spatial evolution of the interface. This is somehow in
agreement with the need to introduce an additional stochastic motion in the ZN group. However
a clear the geometrical interpretation of the stochastic evolution in the representation modules
(see Section 4.4) is missing.
From the results (53) and (55), we hypothesize that the geometric properties of this interface
for N 4 are described by an SLE with = 4(N + 1)/(N + 2). In particular, since 4, we
expect the domain under consideration to be a simple curve for each N .
We discuss now the identification of the interface on the lattice associated to the b.c.c. operator R , see Eq. (54). In the Ising and three states Potts model, this operator is associated to an
SLE16/3 and SLE24/5 respectively. The correspondent interfaces on the lattice can be identified
with the help of the FortuinKasteleyn (FK) representations.
The partition function (59) for the N = 2 and N = 3 critical models take the form:

Z=
(62)
x1 + 1 x1 n(i),n(j ) ,
exp
H n(i) n(j ) =
{ }
ij
{ } ij
= ( 3 1)/2 is the critical point of the Ising and of the three

where x1 = 2 1 or x1 = x1
states Potts model respectively. By expanding the product (62) and summing over all the spin
configurations, one obtains the FK random cluster representation:
Mb
b
Z=
(63)
x1
1 x1 N c
C
where the sum is over the subgraphs C of G, C G, the graph G = (V , E) being composed by
the V sites and the E edges of the square lattice. Each graph C is specified by the b bonds on the
edges and the c clusters (= connected component which include the single sites) in a given bond
configuration.
Consider the following boundary conditions: all the edges on the negative axis carry bonds
while all the edges on the positive axis carry no bonds. For the spin variables this means that all
the spins located at the left of the origin take the same value while the spins on right side are
unconstrained, i.e. they are free (F) to take all the N possible values.
412
Fig. 2. Phase diagram of the Z(5) spin model.
In such an arrangement, each graph C presents a cluster growing from the negative axis into
the upper half plane. The boundary of this cluster, starting from the origin, is conjectured (see
for instance [17, Chapter 5] to be statistically equivalent in the continuum limit to the SLE16/3
for N = 2 (Ising model) and to the SLE24/5 for N = 3 (the three states Potts model). This is
in agreement with the properties of the b.c.c. R(N ) operators, N = 2, 3, 4, which have been
shown [32] to be associated to the A|F boundary conditions.
The generalization to the case N 4 is more cumbersome. To our knowledge, the modular
properties of the R-sector characters, and thus the boundary conditions associated to these operators, have not been yet studied for these cases. This motivates a more detailed analysis of
the boundary conformal properties of this sector. However, we point out that at the level of the
parafermionic algebra representations the main features of the R fields, such as the structure
and the degeneracies of the representation module or the current zero modes eigenvalues (see Appendix B.4), directly generalize to each N 2. The formula (45) is an example. It is thus quite
(N )
natural to suppose that the operators R produce the A|F boundary conditions for each N .
Another observation is that it is quite direct to find an alternative formulation of (59) in terms of
colored clusters on a graph. For sake of clarity, we focus our attention on the case N = 5.
The partition function (59) for N = 5 at the critical point xn , see Eq. (61), can be rewritten as:

Z (N=5) =
(64)
x1 + 1 x1 n(i),n(j ) + x2 x1 n(i),n(j )2 .
{ } ij
If
x1
= x2
one recovers the five states Potts model in the usual FK representation.
Expanding the factors in (64), one has to take into account two type of bonds: the bonds
(type 1) connecting spins which take equal values and the bonds (type 2) connecting spins which
differ by an angle 4/5. Then, summing over the spin configurations, the partition function can
be cast in terms of a sum over colored graphs C :
Eb b
b
b
1
2
Z (N=5) =
(65)
x1
1 x1 1 x2 x1 2 5c .
G
Each graph C is specified by the number c of connected component and by the numbers b1 and
b2 of the bonds of type 1 and type 2. Naturally, the sum over the differences of the spins along
413
Fig. 3. Cluster expansion of the Z(5) spin model. The spins at the vertices of a black bond have the same value while
the ones at the vertices of a red bond differ by an angle 4/5. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this paper.)
Fig. 4. (a) A typical cluster configuration of the Z(5) spin model with type 1 (black) and type 2 (red) bonds. The boundary
conditions on the real axes are as in the text. The green curve is the boundary of the cluster growing from the negative
axes. (b) A piece of the interface which crosses red or empty bond. (For interpretation of the references to color in this
figure legend, the reader is referred to the web version of this paper.)
a loop must be zero (modulo 2 ). For instance, a loop formed by three bonds of type 2 and one
bond of type 1 is not permitted. The prime in C indicates that the sum in (65) is taken over the
allowed graphs. A typical configuration is shown in Fig. 3.
By analogy to the Ising and three states Potts model, we consider the boundary conditions in
which the spins have a fixed value on the negative axis while they are free on the positive axis.
With such boundary conditions, there is a cluster of bonds of type 1 growing from the negative
axis, see Fig. 4.
From (56), we conjecture the geometric properties of the boundary of this cluster are described
in the continuum limit by the SLE14/3 .
414
The above discussion can be directly generalized to each value of N . In particular, the
Z(N) spin model admits a random cluster representations with (N 1)/2 types of bonds.
Imposing the conditions on the real axis discussed above, we conjecture that the measure of
the boundary of the cluster growing from the negative real axis is described by an SLE with
= 4(N + 2)/(N + 1).
Finally we remark that, during its evolution, the interface crosses edges without bonds or with
bonds of the type 2, 3, . . . (N 1)/2 which have different energetic costs. Analogously to the
case of the interface generated by the A|B + C + boundary conditions, one can color the
interface depending on the type of bond it crosses. Again, the interface is naturally provided
of an ZN additional internal degree of freedom, in agreement with the approach proposed in
Section 4.4.
6. Conclusions
In this paper we have considered the possibility to extend an SLE approach to CFTs with
additional ZN discrete symmetries. These theories describe the continuum limit of the self-dual
critical Z(N) spin models. The case N = 2 and N = 3 correspond to the Ising and three states
Potts model where the identification of the interface on the lattice and of the correspondent b.c.c.
operators are known. Using these results, we have identified the possible b.c.c. operators for general N . We show that these operators satisfy a two level null vector condition. For N 4, an
additional term enter these relations. This term is obtained from the action of the parafermionic
currents and, for N 4 (c 1), it cannot be derived by the Virasoro modes action. These results led us to propose an additional stochastic motion in the different ZN charge sectors and
in the Z2 sectors which describe the representation modules of the parafermionic algebra. We
have finally discussed the possible interfaces in the lattice to which the CFT/SLE results should
apply. The fact that SLE candidates have been identified in lattice spin models described by nonconformal minimal theories opens the possibility to study the SLE/non-minimal CFT connection
by a numerical approach.
It will be interesting to verify our predictions by numerical studies. Further we believe these
studies can also better clarify the geometrical interpretation of the additional stochastic motion
in the internal space.
Acknowledgements
We would like to thank M. Bauer, J. Cardy, I. Gruzberg, C. Hagendorf, A. Ludwig, V. Riva for
very helpful discussions. We are particularly grateful to D. Bernard and P. Le Doussal for very
useful advises and support all along this work.
Appendix A. Relation between the SU(2)N WZW model and the Z(N) parafermionic
theory
In this appendix we briefly show how the stochastic motion (23) for the SU(2)N WZW
theories can be somehow compared to the one (47) by taking into account that the Z(N)
parafermionic theory corresponds to the coset SU(2)N /U (1). The SU(2)N WZW theory can
then be seen as a combined system of the ZN parafermionic theory and one free massless boson
field (z). The chiral currents generating the KacMoody algebra (19) can be written in terms
of the parafermionic currents and the free field as:
J =
415

1
(z) exp i (z) ,
N
N
(z).
J0 = i
2
(A.1)
The stress-energy tensor T SU(2)N and T Z(N) of the two theories are simply related by T SU(2)N =
T Z(N) + T U (1) , where T U (1) = 1/4()2 is the stress-energy tensor of the free bosonic theory.
All the primary fields j of the W ZW theory can be obtained by a product of parafermionic
fields in the ZN sector and of vertex operators of the bosonic theory:

j
j (z) = j exp i (z) .
(A.2)
N
of the vertex operator exp[i(z)/ N ] is

Using the fact that the conformal dimension vert.
vert.
= 2 /N , one can easily verify that j =
vert.
j + j , see Eqs. (21) and (40). Note that

m
0
the field j (z) in Eq. (A.2) satisfies J0 j = j j . In the following we denote jm (z), with m =
j, j + 1, . . . , j , the field in the SU(2) representation of spin j such that J00 jm = mjm . The
j k
field j
as:
j k
can be expressed in terms of the parafermionic primaries q and of vertex operators

j k
(z) = Akj j exp i (z)
j k
N
(A.3)
where k has been defined in Eq. (37). For instance, the SU(2) operator 10 corresponds to (N ) ,
1 to 1 exp[i(z)].
see Section 4.5, while the operator 1
Using the above relations, one can compare the definitions (23) and (47), and in particular the
exponent in the denominator of the two formulas. This exponent actually determines the current
a mode, as
mode acting on the b.c.c. field under his evolution. In the SU(2)N theory this is the J1
1
it can be seen in Eq. (24). Let us consider for instance the state J1 1 :

1
J1 1 = 1/(2i) dz (z w)1 J (z)11 (w)
w

i
i
dz (z w)1 1 (z) 1 (w) exp (z) exp (w)
N
N
w

1

i
12/N
1
(z) (w) 1 + (z w)(w) + ,
1/(2i) dz (z w)
N
w
(A.4)
where we have used the relations (A.2) and the divergences coming from the OPE in the bosonic
sector:

q + q
q
q

2 qq

N
exp i (z ) + .
exp i (z) exp i (z ) = (z z )
(A.5)
N
N
N

1/(2i)
1
The above expression gives precisely the state A1
1/N 1 multiplied by a bosonic piece. In
0
+
+
1 and the term (J1
J1 + J1
J1 )10 correspond
the same way one can show that the term J1
1
1
1
respectively to A1/N 1 (N ) and to (A11/N 1 A1/N 1 + A1/N 1 A11/N 1 ) (N ) which also appear
in the Ito derivative of the Z(N) b.c.c. parafermionic operator, see Eq. (49).
416
Appendix B. Null vectors in the parafermionic modules

B.1. Commutation relations of parafermionic current modes: ZN sector
The commutation relations of the mode operators can be deduced from Eq. (39) by using
standard techniques in the complex plane. Consider for instance the commutation relation of the
modes of 1 (z) in the ZN singlet sector 0 . We use the short-hand notation { 1 , 1 } 0 to
indicate such relations. One consider the double integral:

1
dz1 dz2 z1n z2m (z1 z2 )2/N 1 1 (z1 ) 1 (z2 ) 0 (0),
(B.1)
2i
C0
C0
where one adds the term (z1 z2 )2/N 1 to make the integrand single-valued. By using Eqs. (28)
(36) and expanding in series the term (z1 z2 )2/N 1 , the Cauchy theorem gives the following
relations for { 1 , 1 } 0 :

1
0
1,1 2
l
1
1
1
0
D(2N
)/N A(3N )/N+n A1/N +m + A(3N )/N+m A1/N +n = 2 A(4N )/N+n+m .
l
(B.2)
The coefficients
Dl
(1 x) =
are defined from the development
Dl x l .
(B.3)
l=0
The relations { 1 1 } 1 , for instance, are derived from the double integral:

1
2/N +n (N2)/N +m
dz1 dz2 z1
z2
(z1 z2 )(N+2)/N 1 (z1 ) 1 (z2 ) 1 (0).
2i
C0
(B.4)
C0
The exponent of the (z1 z2 ) term has been chosen on the basis of the OPE (29). In particular
it allows the modes of the stress energy, appearing at the second order in (29), to enter in the
relation { 1 1 } 1 . One obtains:

1
1
1
l
1
A1/N +n1 A1
D(N2)/N
1/N +m+1 + A3/N +m A3/N +n
l

1 2
N +2
2
=
+n
+ n 1 n+m,0 +
Ln+m 1 .
2 N
N
N
(B.5)
From the above relations, and using the primary condition (36), one can readily obtain the conformal dimension 1 of a primary field 1 . Setting n = m = 0 in Eq. (B.5), one obtains
1 = (N 2)/(N (N + 2)).
For completeness, we give below in a compact form the commutation relations { 1 , 1 } q
and { 1 , 1 } q which determine the space of the representations:
1
l
D(2N
)/N A
q+1
q+2 +s(q)+nl
l=0
= 21,1 A2 q
q+2 +t (q)+n+m
q ,
A1 q
q+1 +m+l
+ A1
q+1
q+1 +s(q)+ml
A1 q
q+1 +n+l
q
(B.6)
1
l
D(2+N
)/N A
q1
l=0
+u(1,q)+nl
A1
q
q1 +m+l
+ A1q+1
q
417
+u(1,q)+ml
A1 q
q+1 +n+l

1 2|q|
2|q|
N +2
=
+n
+n1 +
Ln+m1+u(1,q) q .
2 N
N
N
q
(B.7)
The integers s(q), t (q) and u(1, q) shift the indices of the parafermionic modes:
2
1,
N
2
q
q
t (q) = q+2 2q+1 + 1,
N
N +2
qk
q
,
u(k, q) = q qk
N
q+1
s(q) = q+2 q+1 +
(B.8)
where q+2 is defined in Eq. (37). Finally the algebra is then completed by the commutators
between the parafermions modes A1 and Ln :

A1
q
Ln
q1 +m
Ln A1
q
q1 +m
q =

q 1
1
1 n + m q1 A q
q1 +m+n
q .
(B.9)
Analogously to what we have seen above for the special case of the 1 field, the formula (40)
giving the dimension of the primary fields is obtained by taking n = m = 0 (n = 1, m = 0) in
Eq. (B.7) for q = 0 (q = 0).
B.2. 1 module
In the following we consider in more detail the module of the representation 1 . The structure of the corresponding module is shown in Fig. 5 for the case N = 5.
Fig. 5. Representation module of 1 and the action of the parafermionic modes giving the operator and its second
level descendants.
418
We have seen above that the dimension of the operator are given by the commutation relations
Eq. (B.7) without imposing any degeneracy in the module. This in general is not the case, as it
can be seen in the study of the representation of the minimal models or of other parafermionic
algebras. However, in order to characterize completely the representation one needs to fix the
eigenvalue h1 of the zero mode defined in Eq. (38). We show this is easily obtained by imposing
a degeneracy at level 1/N where the first descendant states in this doublet are the two singlet
+1 and A1
1
states A1
1/N
1/N . Forming the linear combination
0
1
= aA11/N 1 + bA1
1/N
1/N ,
(B.10)
we wish to make it into a primary operator, i.e., to ensure that it is annihilated upon action by
positive index mode operators. In this case it will be sufficient to verify that
0
0
= 0 and A1
A1+1/N 1/N
+1/N 1/N = 0.
(B.11)
0
The degeneracy conditions A1
1/N 1/N = 0 are satisfied if
(1,1 )2 = (1,1 )2 ,
(B.12)
where the matrix element k,k is defined by

k,k 1 Ak1/N Ak1/N 1 .
(B.13)
Using Eqs. (28)(29), one finds:

1,1 = 21,1 h1 ,
1,1 =
2
,
N
(B.14)
where the structure constant 21,1 is given by Eq. (32), and the zero mode eigenvalue h1 , defined
in Eq. (38), is fixed to:
h21 =
2
.
N(N 1)
(B.15)
Once the conformal dimension and the zero eigenvalue (B.15) have been determined, the operator
0
0
1
= A11/N 1 A1
1/N
1/N is a primary operator and it is put to zero, 1/N = 0, to make
the representation irreducible. After imposing the degeneracy at level 1/N , only one singlet state
remains:
(N ) = A11/N 1 = A11/N 1 ,
(B.16)
with dimension
(N) =
2
.
N +2
(B.17)
B.3. Independent states at the second level of the (N ) operator

Using systematically the commutation relations (B.6)(B.7), one can show that at the second level of the (N ) operator there are only two independent states. All the operators at the
419
second level can indeed be obtained by a linear combination of the following two states:
1
(N )
1
1
A1/N 1 A1
and
1/N 1 + A1/N 1 A1/N 1
1

1
1
(N )
A1/N 2 A1
(B.18)
.
1/N + A1/N 2 A1/N
The action of the above parafermionic modes in the 1 module is shown in Fig. 5.
Now, in order to make the connection with SLE, we express the Virasoro operators L21 and
L2 in terms of combination of parafermionic modes. Using the results (B.14) and (38) in the
relations (B.6)(B.7), we can express the L1 operator as:

4 1
A1/N 1 1 + A11/N 1 1 .
(B.19)
N +2
The above relation, expressed in terms of the coset SU(2)k /U (1), corresponds to the wellknown KnizhnikZamolodchikov equations. Starting from Eq. (B.19) and keeping into account
Eqs. (B.14) and (38) and other similar relations coming from the structure of degeneracies of the
1 module, the commutation relations (B.6)(B.7) give the following relations:
L1 (N ) =
(N )
2N 1
1
A1/N 1 A11/N 1 + A1
1/N 1 A1/N 1
2
(N + 2)

4 1
+
1 ,
A1/N 2 1 + A1
1/N
2
N

N 1
(N )
L2
=
A11/N 1 (N )
A1/N 1 A11/N 1 + A1
1/N
1
N +2

4(N + 1) 1
1
+
1 + A1
A
1/N 2 .
N (N + 2) 1/N 2
L21 (N ) =
(B.20)
(B.21)
From the above relation, Eq. (53) is easily obtained. Note that in the case of N = 3 (the module
1 is not present in the N = 2 theory), the parafermionic algebra provides an additional relation
between states at the second level of (3) . Indeed using the relations { 1 1 } 1 and the fact
that 1 1 1 , valid for N = 3, one obtains:
1
1
A14/3 L1 1 = A1
1 A17/3 1 .
7/3
5
15
Using Eqs. (B.22) and (B.19) in Eqs. (B.21) and in (B.22), one gets:

5
L2 L21 (N=3) = 0
6
(B.22)
(B.23)
which is the relation (15) obtained by the study of the minimal model M5 (remember that (N=3)
corresponds to 1,2 of the minimal model Kac table).
B.4. Commutation relations of parafermionic current modes: R-sector
As mentioned in Section 4.3, the reader can refer to [22] and [23] for a complete discussion
about the construction of the disorder sector modules and of the derivation of the commutation

relations { k , k }Ra . The structure of the disorder representation module for the Z5 theory
is shown in Fig. 6. In the calculations of degeneracy we have used two types of commutation
relations: the first one is between the modes of two 1 chiral fields and the second one is between
the 1 and k chiral fields, with k = 2, 3, . . . , N1
2 . Using the expansions Eqs. (43) and (44),
420
Fig. 6. Representation module of Ra fields for N = 5. The arrows show the action of the parafermionic modes. The zero
mode action between the summit of the module are illustrated by the blue arrows. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this paper.)
the { k , 1 } relations have the following form:

k

l
A(nl)/2 A1(m+l)/2 + A1(ml)/2 Ak(n+l)/2 Ra
D12k/N
l

k+1
3+4k/N k,1

= 214k/N k,1
k1 (1)m Ak+1
k+1 A(n+m)/2 a,a + 2
(n+m)/2 U Ra
where the coefficients Dl are defined from the expansion:

Dl x l ,
(1 x) (1 + x) =
(B.24)
(B.25)
and the matrix Uk Ra = Rak changes the R indices. For the { 1 , 1 } relations, which from
Eq. (43) establish the connection with the Virasoro generators, one gets:

1

l
A(nl)/2 A1(m+l)/2 + A1(ml)/2 A1(n+l)/2 Ra
D12/N
l

= 21+4k/N (1)m

N +2
L(n+m)/2 + (n)n+m,0 URa
N
(B.26)
where
n2 (N 2)
.
8
16N
Finally we will use the commutations {T , }Ra :

m
1
Ln , Am/2 = (n + 1)1 n + A1n+m/2 .
2
(n) =
(B.27)
(B.28)
The level structure of the modules of disorder operators is relatively simple. There are only
integer and half-integer levels, and there exists zero modes for all the operators { q } acting on
the N -uplet of disorder operators. From the expansion (44) there are N/2 zero modes Ak0 (with
k
k = 1, 2, . . . , N1
2 ), associated with the parafermion which acts between the N summit of the
421
module:
Ak0 Ra = hk U2k Ra .
(B.29)
This defines the eigenvalues hk . We recall that URa = Ra1 . The eigenvalues hk characterize,
together with the conformal dimension, each representation Ra . Like the case of the ZN sector,
the formula for the conformal dimension (45) together with the zero modes hk are easily derived
by solving the system of equations obtained by setting n = m = 0 in Eqs. (B.24) and (B.26).
(0)
Here we focus our attention on the module of the operator Ra with dimension:
R
0 =
N2 + N 2
.
16(N + 2)
(B.30)
B.5. Independent states at the second level of the R (0) operator

We have analyzed more in detail the structure of the module of the R (0) disorder operator.
Putting the values n = 0, m = 1 in Eqs. (B.24) and (B.26), and using Eq. (B.30) together with
the value of the zero mode y1 ,
y1 = 22(N1)/N N
(B.31)
(0)
characterizing the module of Ra , it can be shown that

Ak1/2 Ra(0) = 0,
k = 1, 2, . . . , N/2,
(B.32)
for each k. This simplifies greatly the computation of the relations at the second level of this
operator, in which we are interested.
At the first level, by setting n = 1, m = 1 in Eq. (B.26) and using Eq. (B.32), one gets:
(0)
L1 Ra(0) = 242/N N A11 Ra+1
(B.33)
.
Combining the above relation and Eq. (B.28) with n = 1 and m = 2, we find:
N +1
(0)
(0)
L21 Ra(0) = 212/N A12 Ra+1 224/N N A11 A1
(B.34)
1 Ra .
N
Finally, to express the operator L2 we use Eq. (B.26) with n = 2 and m = 2, obtaining:
N +2
N
(0)
(0)
L2 Ra(0) = 212/N A12 Ra+1 24/N
(B.35)
A11 A1
1 Ra .
N
+
2
N
Eq. (54) is then directly derived from the above relations.
As in the case of the (N=3) operator, we have verified that, in this approach, we find the
known results for the 2,1 operator of the M5 model (remember that for N = 3, R (0) = 2,1 ).
Indeed, using the fact that 2 = 1 for N = 3, we can extract from Eq. (B.24) with n = 2,
m = 2 the following relation:
1 (0)
1
2/3 3 1
A R (0) .
A1 A1 Ra = 2
(B.36)
3 2 a
Using Eq. (B.36) in Eqs. (B.34) and (B.35), we obtain:

6
L2 L21 Ra(0) = 0
(B.37)
5
in agreement with Eq. (16).
422
Appendix C. Boundary states

One of the main results of the Boundary conformal field theory [8,9] is the bijection between the possible conformally invariant boundary conditions and the bulk primary operators.
In particular, the allowed boundary states are expressed as linear combinations of bulk primary operators. The coefficients of such expansion are directly related to the entries of modular
transformation matrix S. Reminiscent of the coset construction of the parafermionic theories,
ZN = SU(2)N /U (1), it is convenient to classify the bulk operators with the notation |j, m
,
with j = 0, 1, . . . and m = N/2, N/2 + 1, . . . , N/2. The boundary states |l, m
are then
defined by the formula:
|l , m
=
l ,m
Sl,m
l,m
0,0
Sl,m
|l, m
.
(C.1)
The three states Potts model represents a typical example where these results apply. In order to
fix the notation and to facilitate the comparison with the other ZN theories, we reproduce below
the construction of the boundary states for this model. We use the same conventions as in [9].
The table of the primary fields of the Z3 theory with the associated characters is:
Field
|l, m
0
2/3
2/3
1/15
2/5
1/15
|0, 0
|0, 1
|0, 1
|1, 1
|1, 0
|1, 1
1
1
1
(N=3)
1
In the basis (|0, 0

, |1, 0
, |0, 1
, |1, 1
, |0, 1
, |1, 1
) the modular transformation matrix S
l ,m
Sl,m
takes the form:

l
sll
sll
sl
1

m
= sll
Sm
(C.2)
sll
2 sll

3
sll 2 sll
sll
where = exp(i/3) and

1

sin /5 sin 2/5
.
sll =
(C.3)
5 sin 2/5 sin /5
In order to identify their lattice realization, it is important to study the behavior of the boundary
states under a Z3 rotation. Using the formula (C.1), one can observe that under a Z3 rotation the
boundary states transform as:
|0, 0
|0, 1
|0, 1
|0, 0
,
|1, 0
|1, 1
|1, 1
|1, 0
.
(C.4)
Identifying the boundary state |0, 0

with the boundary conditions in which all the spin on the
real axis take the value A, the above transformations suggests the identification of the state |0, 1
(|0, 1
) to the situation in which all the spins on the real axis take the value B (C). In consistence with the above transformations, the state |1, 0
has been shown to correspond to the state
423
in which the spins can take with equal probability the values B or C. From this identification
one concludes that the b.c.c. operator generates the boundary conditions A|B + C discussed
before.
The results shown here for Z3 generalize directly to the ZN theories. The modular transformation properties of these theories have been analyzed in [33]. For sake of simplicity we consider
here the case Z5 . In general, the number of irreducible representation of the Virasoro algebra is
infinite for N > 3. However, there is a system of principal Virasoro primary fields which appear
at the first levels in each module of q . The table of the principal primary fields of the Z5 theory
is the following:
Field
|l, m
0
1
1
2
2
1
(N=5)
1
A12/5 1
1
A1
2/5
2
2
A1
3/5
2
A2
4/5
1
A3/5 2
2
0
4/5
4/5
6/5
6/5
3/35
2/7
3/35
17/35
17/35
2/35
23/35
6/7
23/35
2/35
|0, 0
|0, 1
|0, 1
|0, 2
|0, 2
|1, 1
|1, 0
|1, 1
|1, 2
|1, 2
|2, 2
|2, 1
|2, 0
|2, 1
|2, 2
In the basis (|0, 0

, |1, 0
, |2, 0
, |0, 1
, |1, 1
, |2, 1
, |0, 2
, |1, 2
, |2, 2
, |0, 1
, |1, 1
, |2, 1
,
l ,m
|0, 2
, |1, 2
, |2, 2
), the modular transformation matrix S Sl,m
takes the form:
l

sll
sll
sll
sll
sl

sl
sll
2 sll
1 sll 2 sll
l
1

l
m
l
l
2
4
Sm = sl
(C.5)
sl
sl 2 sll 4 sll

5
l
l
l
l
l
1
2
2
sl sl sl
sl
sl
sll
2 sll
4 sll
2 sll
where = exp(i/5) and

sin /7 sin 2/7 sin 2/7
1
l
sl =
sin 2/7 sin 4/7 sin 6/7 .
7 sin 3/7 sin 6/7 sin 4/7
4 sll
(C.6)
The corresponding boundary states |0, 0

and |1, 0
transforms, under a Z5 transformation, in the
following way:
|0, 0
|0, 1
|0, 2
|0, 2
|0, 1
,
|1, 0
|1, 1
|1, 2
|1, 2
|1, 1
.
(C.7)
424
It is clear that the case Z5 , and ZN in general, represents, from the point of view of the properties of the boundary states, a direct generalization of the case Z3 . It is then natural to identify
the boundary state |1, 0
with the state in which the spins can take with equal probability the
values B, C, D, E. Consequently, the b.c.c. operator is expected to generates the boundary
conditions A|B + C + D + E discussed in Section 5.
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]
S. Smirnov, C. R. Acad. Sci. Paris 333 (2001) 239.

G. Lawler, O. Schramm, W. Werner, Ann. Prob. 32 (2004) 939.
O. Schramm, S. Sheffield, math.PR/0605237.
M. Bauer, D. Bernard, Phys. Rep. 432 (2006) 115.
M. Bauer, D. Bernard, Commun. Math. Phys. 239 (2003) 493.
M. Bauer, D. Bernard, Phys. Lett. B 543 (2002) 135.
J. Cardy, Nucl. Phys. B 240 (1984) 514.
J. Cardy, Nucl. Phys. B 324 (1989) 581.
I. Affleck, F.D.M. Haldane, Phys. Rev. B 36 (1987) 5291.
V.A. Fateev, A.B. Zamolodchikov, Sov. Phys. JETP 62 (1985) 215.
V.A. Fateev, A.B. Zamolodchikov, Phys. Lett. JETP 92a (1982) 37.
J. Rasmussen, Lett. Math. Phys. 68 (2004) 4152.
J. Nagi, J. Rasmussen, Nucl. Phys. B 704 (2005) 475489.
J. Rasmussen, hep-th/0409026.
E. Bettelheim, I.A. Gruzberg, A.W.W. Ludwig, P. Wiegmann, Phys. Rev. Lett. 95 (2005) 170602.
Q. Kager, B. Nienhuis, J. Stat. Phys. 115 (2004) 1149.
J. Cardy, Ann. Phys. 318 (2005) 81.
O. Schramm, Israel. J. Math. 118 (2000) 221.
D. Friedan, Z. Qiu, S. Shenker, Phys. Rev. Lett. 52 (1984) 1575.
V.A. Fateev, A.B. Zamolodchikov, Sov. Phys. JETP 63 (1986) 913.
V.S. Dotsenko, J.L. Jacobsen, R. Santachiara, Nucl. Phys. B 656 (2003) 259.
V.S. Dotsenko, J.L. Jacobsen, R. Santachiara, Phys. Lett. B 584 (2004) 186.
V. Riva, J. Cardy, J. Stat. Mech. 0612 (2006) P001.
A.B. Zamolodchikov, Sov. Phys. JETP 48 (1978) 168.
V.S. Dotsenko, Sov. Phys. JETP 48 (1978) 546.
E. Fradkin, J. Kadanoff, Nucl. Phys. B 170 (1980) FS1.1.
A. Gamsa, J. Cardy, cond-mat/0705.1510.
H. Saleur, M. Bauer, Nucl. Phys. B 320 (1989) 591.
D. Gepner, Z. Qiu, Nucl. Phys. B 285 (1987) 423.

Effective action in a general chiral model:

Next to leading order derivative expansion
in the worldline method
Andres Hernandez a , Thomas Konstandin b, , Michael G. Schmidt a
a Institut fr Theoretische Physik, Heidelberg University, Philosophenweg 16, D-69120 Heidelberg, Germany
b Department of Theoretical Physics, Royal Institute of Technology (KTH), AlbaNova University Center,
Roslagstullsbacken 21, 106 91 Stockholm, Sweden

Abstract
We present a formalism to determine the imaginary part of a general chiral model in the derivative
expansion. Our formalism is based on the worldline path integral for the covariant current that can be
given in an explicit chiral and gauge covariant form. The effective action is then obtained by integrating the
covariant current, taking account of the anomaly.
2007 Published by Elsevier B.V.
PACS: 11.15.Tk; 11.30.Rd
Keywords: Effective action; Chiral models; Worldline path integral
1. Introduction
When discussing the influence of fermions on the dynamics of some theory, e.g. of the Standard Particle Model (SM) in a cosmology setting, it is mandatory to integrate out the fermions.
This procedure in one-loop order already becomes quite involved if arbitrary chiral couplings of
spacetime dependent outer bosonic fields are considered. The extensive work of Salcedo [1,2]
resulted in an effective action to leading order in covariant derivatives of such fields. Basis is
E-mail addresses: a.hernandez@thphys.uni-heidelberg.de (A. Hernandez), konstand@kth.se (T. Konstandin),

m.g.schmidt@thphys.uni-heidelberg.de (M.G. Schmidt).
0550-3213/$ see front matter 2007 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.10.005
426
A. Hernandez et al. / Nuclear Physics B 793 [FS] (2008) 425450
a refined calculation in momentum space, handling of anomalies using the WessZuminoWitten

model (WZW), and last but not least, a very practical shorthand notation.
Such effective actions are quite important in evaluating some models. For example, in Ref. [3]
Smit discussed a form of cold electroweak baryogenesis at the end of electroweak scale inflation [4] which could very well work if the rephasing invariant
J = s12 s2 s3 c1 c2 c3 sin() = (3.0 0.3) 105
(1)
of the Jarlskog determinant is not accompanied by further suppressions through mass ratios.
It was proposed [3] that derivative terms in the effective action that are analytic in the timedependent masses considered non-perturbatively could be very important in non-equilibrium.
Such effects were also observed in Ref. [5]. In the work [3] the fourth order derivative result of
Ref. [1] turned out not to contain CP violation, but the claim was made that higher orders of the
imaginary part of the effective action will do.
Worldline methods in first quantized quantum field theory are ideally adapted for calculating
effective actions: One considers the propagation of a particle in some spacetime dependent
background [6], but in x-space path integral formulation [7]. This method [8,9], also related to
the infinite tension limit of String theory [10,11], was used heavily for the discussion of various
effective actions in one-loop [9,1214] and two-loop [1518] order. For example, the high order
in the inverse mass calculation of Ref. [15] could hardly be done with other methods.
The present paper provides a formalism to determine higher order contributions to the imaginary part of the effective action using the worldline formalism. We are concerned with the
effective action of a multiplet of N Dirac fermions coupled to an arbitrary matrix-valued set
of fields, including a scalar , a pseudoscalar , a vector A, a pseudovector B, and an antisymmetric tensor K . One peculiar feature of the imaginary part of the effective action is that it
cannot be written in a manifest chiral covariant way, due to the presence of the chiral anomaly.
One possibility to arrive at a closed expression for the effective action is to abandon manifest
chiral covariance as it was done in Ref. [19]. The resulting expression is rather complicated and
not well suited for higher order calculations. Alternatively, it was proposed in Ref. [1] to determine the covariant current for which a manifestly chiral covariant expression exists and to take
account of the anomaly when integrating the current to yield the effective action. Following this
idea, we present a worldline path integral formulation of the covariant current.
Before we do so, we review the worldline formalism by discussing the derivation of the
real part of the effective action. A single Dirac fermion in the presence of both a scalar and
pseudoscalar fields in the context of the worldline formalism was first treated in Ref. [20], the
inclusion of a pseudovector in Ref. [21]. In our discussion of the real part of the effective action
we will follow the elegant subsequent work of Refs. [19,22].
Section 2 contains the derivation of the real part of the effective action, the derivation of the
covariant current and the matching procedure to obtain the imaginary part of the effective action.
In Section 3, we briefly reproduce the results in lowest order from Ref. [1]. As a novel result we
present the imaginary part of the effective action in two dimensions in next to leading order in
Section 4.
2. Effective action
We are concerned with the effective action

/ + 5B
/ + i K ,
iW [, , A, B, K] = log Det i i/
+ i 5 + A
(2)
427
and its continuation to Euclidean space. The matrices remain unaffected by the continuation,
but it is useful to introduce the following notation, (E )j ij , (E )4 0 , and (E )5 5 .
After Wick-rotation, t it, one obtains with this new notation
/ i/
E,
A
/ i/
AE ,
B
/ i/
BE,
K (E ) (E ) KE .
(3)
From now on, the E subscript will be suppressed.

The effective action of Eq. (2) now reads
W [, , A, B, K] = log Det[O],
(4)
with the operator O in momentum space defined by

O p
/ i(x) 5 (x) A(x)
/
5B(x)
/
+ K .
(5)
As in Ref. [2,22], the real and imaginary parts of the effective action are analyzed separately

W + iW = log Det[O] + i arg Det[O] .
(6)
A perturbative expansion in weak fields [22] shows that graphs with an even number of 5 vertices are real, and graphs with an odd number of 5 vertices are imaginary. This will prove useful
when the behavior of the effective action under complex conjugation is explored later on.
2.1. Real part of the effective action
Our intention is to obtain a worldline representation for the effective action with manifest chiral and gauge invariance. This is unproblematic for the real part, but it causes certain difficulties
for the imaginary part due to the chiral anomaly. In order to familiarize the reader with the worldline method we review the derivation for the real part in four dimensions as it was presented in
Refs. [19,22].
2.1.1. Construction of a positive operator for the real part of the effective action
In order to use the worldline formalism, one has to rewrite the effective action in terms of a
positive operator, thus obtaining

1
W + = log Det O O .
(7)
2
The problem with this operator is that it contains terms linear in the matrices, what makes the
transition to a path integral of Grassman fields problematic. One way to avoid this problem is by
doubling the fermion system and exchanging the operator O for a Hermitian operator yielding

2
1
1
0 O
+
W = log Det O O = log Det ,
(8)
.
O 0
2
4
Since is Hermitian, one can use the Schwinger integral representation of the logarithm without
any restrictions. One obtains
1
W =
4
+

dT
Tr exp T 2 .
T
(9)
428
At this point, it is natural to introduce six 8 8 Hermitian A matrices. These matrices satisfy
{A , B } = 2AB , with A, B = 1, . . . , 6 and are defined as

0
0 5
0
i14
=
(10)
,
5 =
,
6 =
.
0
5 0
i14 0
For later use we also introduce the equivalent of 5 ,
7 = i
A =
A=1
14
0
0
14

(11)
and 7 anticommutes with all other matrices.

Expressing in terms of these new matrices yields [19],
= (p A ) 6 5 i 5 6 B i 6 K .
(12)
The aim is to turn Eq. (12) into an expression which is manifestly chiral covariant. This can be
achieved by changing to a basis in which i5 6 is diagonal [22] using the following transformation
12 0 0 0

14
0
0 0 0 12
, M =
M 1 i5 6 M =
(13)
.
0 0 12 0
0 14
0 12 0 0
In this basis, takes the form

(p AL
)
1
= M M =
1
s )
5 (iH + 2 K
s )
5 (iH + 12 K
,
(p AR
)
(14)
which is manifestly chiral covariant. Here AL = A + B, AR = A B, H = i , K s =

K i K and K = 12
K have been defined.
The square of constitutes a positive operator which is suitable for the worldline formalism. However, even though this expression contains only even combinations of matrices, the
coherent state formalism cannot yet be used to transform this expression into a fermionic path
integral. In the coherent state formalism, the 5 matrices have to be rewritten as a product of
the other matrices, what would result again in odd combinations. One possible solution of this
problem is to enlarge the Clifford space, replacing the matrices by matrices

0 1
A A = A
(15)
, A [1, . . . , 5].
1 0
The matrix 5 is then independent from the other matrices and the coherent state formalism
with six (instead of four) operators can be used. The doubling of the Clifford space inside the
trace has to be compensated by a factor 12 such that Eq. (9) reads
1
W =
8
+

dT
Tr exp T 2 ,
T
and the operator 2 is given by
(16)
429

1
i
2 = (p A)2 + H2 + K K + F + {H, K } + i[K , K ]
2
2

1
1
+ i 5 D H + {p A , K } 5 D K K K ,
2
4
with enlarged background fields defined by
L

A 0
0
A =
,
H
=
iH
0 AR
iH
0

K =
0
s
iK
s
iK
(17)

.
(18)
A1 ...Ak [A1 . . . Ak ] denotes the antisymmetrized product of k matrices, and the fieldstrength and the covariant derivative have been defined as
F = A A i[A , A ],
D = i[A , ].
(19)
The 2 operator is seen to be manifestly gauge and chiral covariant. It also contains matrices
to even powers only, and is well suited for the worldline path integral representation.
2.1.2. Worldline path integral
With the use of the coherent state formalism [22,23], one can perform the transition from
matrices to a path integral over Grassman fields , with the correspondence A B 2A B
and A B C D 4A B C D , as long as A, B, C, and D are all different. The final form
for the real part of the effective action is
1
W =
8
+
dT
N
T

Dx
D tr Pe
T
0
d L( )
(20)
AP
Here N denotes a normalization constant coming from a momentum integration and AP stands
for antiperiodic boundary conditions, which must be fulfilled by the Grassman variables (T ) =
(0). The Lagrangian is given by
L( ) =
1
x 2 1
+ A A i x A + H2 + K K + 2i 5 (D H + i x K )
4
2
2

+ i F + {H, K } + i[K , K ]
(25 D K + K K ).
(21)
The periodic boundary conditions for the field x( ) suggest to separate the zero modes of the free
d2
field operator d
2 . The fields x( ) are split into a constant part and a dependent part according
T
to x( ) = x0 + y( ), with x0 = 0 and 0 dy( ) = 0, and the measure in the integral is
changed into Dx = Dy d D x0 . The Green function is defined on a subspace orthogonal to the
zero modes. The A fields contain no zero modes, so that the propagators for the y( ) and
A ( ) fields read

(1 2 )2
y(1 )y(2 ) =
|1 2 |,
T
1

A (1 )B (2 ) = AB sign(1 2 ).
(22)
2
This formalism can then be used to determine the real part of the effective action as discussed in
Ref. [19].
430
2.2. Imaginary part of the effective action

As in the case of the real part of the effective action, one requires a positive operator in order to use the Schwinger trick. Even though this is still possible for the imaginary part, gauge
and chiral invariance cannot be manifestly conserved due to the chiral anomaly. For example, in
Refs. [19,22] a parameter is introduced, which breaks the chiral invariance, but leads to a positive operator. However the resulting expression is not appropriate for higher order calculations
since the breaking of manifest chiral invariance leads to a large number of contributions in the
perturbative expansion of the path integral.
The aim of the present work is to present a worldline representation of the effective current
for which a manifestly chiral covariant expression exists. This current can then be integrated
to obtain the effective action [1,24,25]. This integration rather proceeds by matching: First, a
general effective action is proposed, which has the expected chiral and covariant properties. The
functional variation of this action is then matched to the covariant current that is obtained using
the worldline formalism. This method has the advantage that it is both gauge and chiral invariant at each stage of the calculation. The anomaly only leads to additional complications in the
matching procedure of the lowest order contributions as will be discussed in detail in the next
section.
Starting point of our analysis is the functional derivative of the imaginary part of the effective
action in Eq. (6)

1
1
1
1
W = log Det O log Det O = Tr O O .
(23)
2
2
O
O
This expression can be rewritten in terms of a positive operator which can be used to employ
the worldline representation in combination with the heat kernel formula. Incidentally, it can
also be expressed in a manifestly chiral covariant form, what simplifies higher order calculations
tremendously as compared to the formalism presented in Ref. [19].
2.2.1. Construction of a positive operator for the imaginary part of the effective action
The expression in Eq. (23) can be transformed using the operator defined in Eq. (8)

1
0
O
0
1/O
,
W = Tr
O 0
1/O
0
2
(24)
which, with the introduction of a new matrix , can be rewritten as

W =
with

=
1
Tr 1 ,
2
0
O

O
,
0
(25)

14
0
0
14

.
(26)
To produce the positive definite operator 2 in Eq. (25), we multiply and divide by , using the
cyclic property of the trace and the fact that anticommutes with , to obtain
1
Tr( + ) 2
4
1
= Tr [, ] 2 .
4
W =
(27)
431
Since the last factor is a positive operator, it can be reexpressed as an integral, similar to the
expression of the real part of the effective action in Eq. (17), namely
1
W = Tr
4
dT [, ]eT .
2
(28)
As in the case for the real part, the chiral covariance can be made manifest by changing to an
appropriate basis. With the help of the matrix M in Eq. (13), one obtains again
i
= (p A ) 5 H 5 K .
2
The additional factors [, ] read

5
0
1
= 5 ,
M M = =
0 5
(29)
(30)
and for the case = A ,

]
= {A , p A } iD A 5 {A , H}
[ ,
i
+ i5 [A , K ] 5 {A , K }.
(31)
2
To use the coherent state formalism, it is again necessary to enlarge the Clifford algebra and
to replace the matrices by matrices. However, taking into account the factor 5 in Eq. (30)
the imaginary part of the effective action contains only odd combinations of matrices. Thus,
the replacement

0 1
, A [1, . . . , 5],
A A = A
(32)
1 0
has to be compensated by a factor

i
0 12
7 6 = 14 1
.
0
2
2
(33)
The overall factor 7 changes the boundary condition of the fermionic sector from antiperiodic
to periodic as explained in Ref. [22]. This means that the fermionic sector contains zero modes,
which have to be separated in the same way as was done for the bosonic sector.
Including the factor in Eq. (33) to compensate for the doubling of the Clifford space, one
obtains
i
W = Tr
8
dT 7 6 w(T )eT ,
(34)
where 2 is given in Eq. (17), and the insertion due to the commutator yields
1
w(T ) = 5 {A , p A } i5 D A {A , H}
2
i
+ i [A , K ] {A , K }.
2
(35)
432
To transform this expression into a worldline path integral, a similar procedure as for the real
part of the effective action can be followed. Products of matrices can be replaced by Grassman
fields, however in this case the Jacobian of the transformation contains additional contributions
from the zero modes
D D d3 d2 d1 d 1 d 2 d 3 D D
1
= d10 d20 d30 d40 d50 d60 D .
(36)
J
The factor J only includes the Jacobian for the zero modes, while the Jacobian for the orthogonal
modes is absorbed in the normalization of the correlation functions of the A . J can be calculated
from the definition of the Grassman fields in the coherent state formalism [22] and yields in D
dimension

,
J = det
(37)
= (i)(D+2)/2 .
The final result can be expressed as

1
W = tr
8

dT N

Dx
Dw(T )Pe
T
0
d L( )
(38)
The Lagrangian is of the same form as in the real part, Eq. (21),
L( ) =
x 2 1
1
+ A A i x A + H2 K K + 2i 5 (D H + i x K )
4
2
2

1
+ i F + {H, K } 2 5 D K + K K
2
(39)
and the trivial integration over 6 can been carried out, so that the insertion yields
w(T ) = 4i5 A x 2i5 D A 2 {A , H}
+ 2i [A , K ] 2i {A , K }.
The normalization N coming from the momentum integration, satisfies

T
0 d x4
D/2
d D x.
= (4T )
N Dxe
(40)
(41)
The Green function for the bosonic field x is the same as for the real part of the effective action,
Eq. (22), while the Green function of the Grassman fields A differs due to the presence of the
zero modes. The fermionic fields are split according to A ( ) = A0 + A ( ), with A0 = 0
T
and 0 d A ( ) = 0 and the measure turns into D = d1 d2 d3 d4 d5 D . The Green
function for the A fields, defined on a space orthogonal to the zero modes, reads

(1 2 )
1

sign(1 2 )
.
A (1 )B (2 ) = AB
(42)
2
T
These results can be easily generalized to different dimensions. In two dimension, one obtains
an additional overall factor i from the Jacobian of the zero modes and the fermionic measure
reads D = d1 d2 d5 D .
433
2.2.2. The effective density

The effective density is obtained by varying with respect to the H field, so that = 5 H.
In comparison to the worldline representation of the covariant current only the insertion changes
into
i
]
= 5 {H, p A } + [H, H] + [H, K ].
[ ,
2
(43)
The corresponding insertion w(T ) in the path integral reads then

w(T ) = 2i x H + 25 [H, H] + 2i [H, K ].
(44)
Since A carries an index, the effective current is of one order lower than the effective density
and usually results in less terms to calculate. The advantage of the effective density lies in the
matching process, since the factors in the effective density consist of the same type as found in
the effective action. They both combine the same type of object, DH and F , to the same kind
of order, while the effective current combines the terms to a lower order. Besides, there is no
distinction between a consistent effective density and a covariant effective density, as there is for
the effective current, as will be explained in the next section.
2.2.3. Distinction between the consistent and the covariant current
With Eq. (38) an expression for the covariant current which is chiral and gauge covariant was
derived. This current cannot be the variation of the effective action, since the effective action
contains the chiral anomaly, and in fact the covariant current is not a variation of any action.
The reason for this is that performing the variation does not commute with the regularization
procedure we used, namely the Schwinger trick. On the other hand, knowing the chiral anomaly,
one can reproduce the so-called consistent current that denotes the true variation of the effective
action.
To explain the relation between the two currents, we define a general variation

Y = dx Ya (x)
(45)
,
A (x)
so that a gauge variation is given by

= dx (D )(x)
.
A (x)
(46)
Two subsequent variations have then the commutator [Y , ] = [Y, ] and in order to find the
transformation properties of the consistent current, one can apply this commutator to the effective
action
[Y , ]W [A ] = [Y, ] W [A ].
Using the anomalous Ward identity [26]

W [A ] = dx (x)G[A ](x),
(47)
(48)
with G[A ] denoting the consistent anomaly, one can evaluate both sides of Eq. (47) to obtain
434
dx [Y , ](x)
W [A ]
A (x)

= Y dx (x)G[A ](x) dx Y (x)
W [A ].
A (x)
Defining the consistent current as the variation of the effective action

j (x) =
W [A ],
A (x)
one finds

dx Y (x) j (x) = dx Y j (x) , (x) + dx (x)Y G[A ](x).
(49)
(50)
(51)
Since Y was a general variation this leads to

j (x) = j (x) , + dy b (y)
(52)
G[A ](y).
A (x)
This shows that only if the anomaly vanishes, the current transforms covariantly. This relation
can be used to determine the connection between the consistent current, i.e. the true variation
of the action, and the covariant current. The latter is obtained by adding an object P [A ],
called the BardeenZumino polynomial [27], to the consistent current so that the sum transforms
covariantly

j = j + P .
(53)
This implies the following gauge transformation property for the BZ polynomial

P [A ](x) = P [A ], (x) dy (y)

G[A ](y).
A (x)
It is not obvious that such an object exists, but using
(54)
1
(55)
tr (A F + F A + iA A A ),
48 2
and the consistent anomaly [26]

1
i
tr A A A A A ,
G[A ] =
(56)
2
24 2
it can be shown that the definition of P in Eq. (55) provides a unique polynomial in A that
satisfies Eq. (54). The corresponding functions in two dimensions are given by
P [A ] =
1
1
G[A ] =
(57)
tr A ,
tr A .
4
4
As stated above, the path integral in Eq. (38) constitutes a worldline representation of the
covariant current. To obtain the imaginary part of the effective action from the covariant current
one can use the following ansatz
P =
W = gWZW + Wc .
(58)
Here, gWZW is an extended gauged WessZuminoWitten action [24,28,29], which is chosen

to reproduce the correct chiral anomaly, and Wc denotes a chiral invariant part. The variation of
the functional gWZW , consists of a part that saturates the anomaly, namely the BZ polynomial,
and a covariant remainder which has to be added to the variation of Wc to yield the covariant
current.
435
2.2.4. The WessZuminoWitten action

When the effective action is separated into two parts, it is required by the non-covariant part
that it reproduces the anomaly. It is well known that the WZW action has this property.
The ungauged WZW action in four dimension is e.g. of the form

i
1 1
5
abcde
1
1
1
1
(U) =
(59)
d x
tr U a UU b UU c UU d UU e U ,
5
48 2
Q
where Q is a five-dimensional space with boundary Q equal to the R 4 flat Euclidean space. The
matrix U is a unitary matrix, and is usually related to the case where the mass can be expressed
as a constant times that unitary matrix. We are interested in the more general case when the
mass matrix is not of this form which is called extended WZW action. In addition, the presence
of the background gauge fields makes a gauging of the action mandatory. The gauged extended
WZW action can be generally expressed as the integral in five dimensions [24]. Unlike the action
itself, the resulting current turns out to be a total derivative in five dimensions, such that it can be
represented by an integral over the physical four-dimensional space

1
4
1
1
1
gWZW =
d
x
tr
A
H D HH D HH D H
96 2

+ D HH1 D HH1 D HH1 i H1 D H D HH1 , F

i
i
+ H H1 D H, F H1 H1 D HH1 , F H
2
2

2{A , F } 2iA A A ,
(60)
or in two dimensions
gWZW =
1
8

d 2 x
tr A iH1 D H + iD HH1 2A .
(61)
Notice that in both cases the last term in the current denotes the BZ polynomial. The remaining
chiral covariant terms have to be subtracted from the covariant current before it is matched to the
effective action according to the ansatz made in Eq. (58).
3. Lowest order effective action
3.1. Effective covariant current
In order to reproduce the results from Ref. [1], we neglect in this section the antisymmetric
field K . The fields A and H are matrices of some internal group, and we only assume that
H(x0 ) is nowhere singular. With this in mind, we restate our result (38) from the last section in
D dimensions
i D/2
W =
tr
8

dT N
0
with

Dx
P
Dw(T )Pe
T
0
d L( )
(62)
436
x 2 1
+ A A i x A + H2 + 2i 5 D H + i F ,
4
2
w(T ) = 4i5 A x 2i5 D A 2 {A , H}.
L( ) =
(63)
Next, the derivative expansion of the heat kernel is used. In the derivative expansion terms are
classified by the number of covariant indices that they carry, so that D H is of first order,
while F is of second order. The worldline formalism is well suited for this expansion, and
there are two major advantages compared to the more traditional methods used e.g. in Ref. [1].
First, the tedious manipulations using the algebra are avoided. Secondly, the momentum integration is omitted and replaced by the rather trivial integration in space.
The coordinate is split as x( ) = x0 + y( ), and we work in the FockSchwinger gauge [30],
in which A(x) y = 0. In this gauge, expressions remain gauge covariant and the field A can be
expressed in terms of the field strength tensor F by
1
A (x) =
d F (x0 + y)y .
(64)
All background fields can then be expanded around the point x0 in terms of covariant derivatives

X x0 + y( ) = exp y( ) Dx0 X(x0 ),
(65)
where Dx0 refers to the covariant derivative in Eq. (19) with respect to x0 . With the expansion of
the field strength tensor in terms of covariant derivatives and Eq. (64), one can rewrite the field A
as
1
1
1
y y y D D F (x0 ) + .
A (x) = y F (x0 ) + y y D F (x0 ) +
2
3
4 2!
(66)
Since we will not carry out the integration with respect to x0 we use the following notation in
D dimensions
D/2

i
tr d D x0 X.
XD =
(67)
4
It is important to remember that and H anticommute; hence, when the cyclic property of the
trace is used, a minus sign is generated, for example

HF H3 F =
F H3 F H =
H3 F HF

=
H3 F HF .
(68)
After expanding the mass field H(x)2 = H2 (x0 ) + y D H2 (x0 ) + , the field H(x0 ) is
treated non-perturbatively. Since all the fields can be matrices of some internal space the resulting expressions normally cannot be expressed in closed form. For this case we use the
labeled operator notation laid down in Refs. [2,31]. The notation works as follows: In an
expression f (A1 , B2 , . . .)XY . . . , the labels of the operators A, B, . . . denote the position of
that operator with respect to the remaining operators XY . . . . For instance, for the function
f (A, B) = (A)(B), the expression f (A1 , B2 )XY represents (A)X(B)Y . In the case at
hand, the operator appearing in the functions is always m := H(x0 ), such that general functions f can be easily interpreted in the basis where m is diagonal. Using this notation, Eq. (68)
437
can be recast as

HF H3 F =
m1 m32 F F =
m3 m32 F F

=
m2 m31 F F =
H3 F HF .
(69)
This notation can also be used to simplify the matrix valued derivative. Using the definition
f (m1 ) f (m2 )
(f )(m1 , m2 ) :=
(70)
,
m1 m2
it is possible to prove that
D f (m) = (f )(m1 , m2 )D H.
(71)
f (m) = m3
For example, in the polynomial case

one obtains
3
D f (m) = D H = D HH2 + HD HH + H2 D H

m3 m32
D H
= m22 + m1 m2 + m21 D H = 1
m1 m2
= (f )(m1 , m2 )D H.
(72)
As mentioned earlier, non-polynomial expressions are hereby interpreted in a basis where m is

diagonal, so that for m = diag(d1 , . . . , dn )
f (di ) f (dj )
f (m1 ) f (m2 )
X=
Xij .
m1 m2
d i dj
(73)
More general, this suggests the following definition for the case with several variables:
k f (m1 , . . . , mn ) =
k+1 , . . . , mn ) f (m1 , . . . , m
k , . . . , mn )
f (m1 , . . . , m
,
mk mk+1
(74)
where m
k indicates that the corresponding argument is left out.
If all arguments of the functions are of the same type one can further simplify the notation
and use subscripts to refer to the argument of the function, e.g. f (m1 , m2 ) =: f12 and we employ
this notation in the following. Additionally, negative arguments will be denoted by underlining
the corresponding index, f (m1 , m2 ) =: f12 . More applications of the labeled operator notation
can be found in Refs. [1,2].
The path ordering in Eq. (62) is defined by
N
i=1 0
di N!
1
d2
d1
0
N1
dN = N !
0
Separating the Lagrangian (63) into L( ) = L0
T
d1
( ) + H2 (x
dN
0
0 ) + L1 ( ),
N1
(i i+1 ).
i1
(75)
with
x 2 1
+ A A ,
4
2
L1 ( ) = i x A (x) + 2i 5 D H(x) + i F (x) + y D H2 (x0 ) + ,
L0 ( ) =
(76)
the terms of the expansion of H2 (x), except the leading term H2 (x0 ), are attributed to L1 ( ),
and treated perturbatively. Notice that L0 commutes with the rest of the Lagrangian, so that the
438
expansion of the path ordered exponential in Eq. (62) takes the form
Pe
T
0
=e
d L( )
T
0

d L0 ( )
eT H
2 (x )
0
T
+
d1 e(T 1 )H
2 (x )
0

2
L1 (1 ) e1 H (x0 )
T
+
1
d1
d2 e
(T 1 )H2 (x0 )

( )H2 (x )
H2 (x )
0 L ( ) e
2
0 + .
L1 (1 ) e 1 2
1 2
(77)
When performing the integrals, the zero modes have to be saturated and at least a factor
0 0 is required from the Grassman fields in order to contribute. The first term in Eq. (77)
10 D
5
lacks the appropriate factor except in two dimensions, where the first term of the insertion
Eq. (63) already has the appropriate factor. However it contains a factor x which must be contracted with a similar factor to form a Green function, hence it does not contribute and can be
left out. The rest of Eq. (77) can be simplified using the labeled operator notation. Using the
expression m2n to denote H2 (x0 ) in nth position, one obtains
Pe
T
0
d L( )
= e
T
0
T
2
2
2
d L0 ( )
d1 eT m1 1 (m2 m1 ) L1 (1 )
0
T
+
d1
0
1
d2 e
T m21 1 (m22 m21 )2 (m23 m22 )
L1 (1 )L1 (2 ) + .
(78)
The evaluation of the worldline path integral can be summarized as follows: First, all fields in
Eq. (78) and the insertion are expanded around x0 . Next, the functional integration over the
y fields is carried out, generating bosonic Green functions. Then, the integrations are performed saturating the zero modes and generating fermionic Green functions. Finally, the T and
integrations are performed.
Before presenting the actual calculation, we comment on the behavior of the effective action
under complex conjugation. As noted earlier, in any contribution to the imaginary part of the
action the field 5 appears an odd number of times. If one attributes a factor i to the operators F and A one observes that the remaining expressions in the current in Eq. (63) are real.
Accordingly, all expressions in W are real as long as a factor i is attributed to the operator F .
In addition, notice that the effective action has to be an even function in the masses due to chiral
invariance.
In order to showcase the method, we present the lowest order calculation in two dimensions.
The lowest order contribution coming from the first term in the insertion is given by
T
4i5 y (T )
0
d1 eT m1 1 (m2 m1 ) y ( )D H2 A (T ).
2
(79)
439
Performing the y and integrals one obtains

T
i
dT
T m21 1 (m22 m21 )
d1 e
g B (T , 1 )D HA
(m1 + m2 )
2
T
0

i
2
=
J12
(80)
(m1 + m2 )D HA .
2
The second term of the insertion does not contribute at lowest order since it is already of second
order in derivatives but lacks the appropriate fermionic factor to saturate the zero modes. The
third term of the insertion leads only to one contribution of the form
T
2 {A , H}
d1 eT m1 1 (m2 m1 ) (2i 5 D H),

2
(81)
yielding

T
dT
i
T m21 1 (m22 m21 )
d1 e
D HA

(m1 m2 )
2
T
0

i
1
=
J12
(82)
(m1 m2 )D HA .
2
The factor (m1 m2 ) results from the anticommutator in Eq. (81), and the sign change in the
cyclic property of the trace as explained in Eq. (68). The integrals J are given in Appendix A.
The total current is hence given by

W = i
A112 D HA ,
(83)
A112 =
m1 m2 log(m21 /m22 )
1
,
m1 m2 (m1 m2 )(m21 m22 )
(84)
where the function A112 has been defined. This agrees with the results obtained in Ref. [1].
3.2. Effective density
The effective density can be obtained similar as the covariant current, utilizing the insertion
in Eq. (44). Neglecting the antisymmetric tensor K, the insertion is
w(T ) = 2i x H + 25 [H, H].
The contributions to the effective density are

i 1
2
(m1 m2 ) F
J12 (m1 + m2 ) + J12
W =

4

1 5
6
7
J123
(m1 + m3 ) + J123
(m1 + m2 ) J123
(m2 + m3 ) D HD H H
2

i 1
2
F + B123
D HD H H
=
B12
2
(85)
(86)
440
where the functions B12 and B123 are given by

2
m1
1
m1 m2
,
log
2
2
m1 + m2 (m1 + m2 )(m1 m2 )
m22

2
R
L
L
L
= B123
+ B123
log m21 + B231
log m22 + B312
log m22 ,
B123
1
=
B12
(87)
(88)
with
R
B123
=
L
=
B123
1
,
(m1 m2 )(m2 m3 )(m1 + m3 )
(m31 + m1 m2 m3 )
(m1 m2 )(m1 + m3 )(m21 m22 )(m21 + m23 )
(89)
,
(90)
in accordance with Ref. [1].

3.3. Effective action
We proceed and briefly present the derivation of the imaginary part of the effective action
following Ref. [1]. Using the ansatz in Eq. (58), the most general functional for Wc consistent
with chiral and gauge invariance in two dimensions reads

Wc =
N12 D HD H .
(91)
An additional term proportional to F could be added but it can be removed by partial integration.
Notice that N12 is a real function according to the comments made in the last section.
The function N12 has some nontrivial restrictions. First of all, the function N12 is even in m
such that
N(m1 , m2 ) := N12 = N12 .
(92)
Because of the cyclic property of the trace one obtains

N12 = N32 = N21 = N21 ,
and due to the Hermiticity of
(93)
W ,
N12 = N32 = N12 = N21 .

Varying Wc [A, H] with respect to A, one obtains

Wc = i
2(m1 + m2 )N12 D HA .
(94)
(95)
Comparing this to Eq. (83) and adding the covariant contribution in Eq. (61) coming from gWZW
one has
m1 m2 log(m21 /m22 )
1
1
1
2(m1 + m2 )N12 ,
=
2
2
m1 m2 (m1 m2 )(m1 m2 ) 2m1 2m2
(96)
which finally leads to

log(m21 /m22 ) 1 1
1 m1 m2
1
+
.
N12 =
2 m21 m22
2 m21 m22
m21 m22
(97)
441
At higher order, the matching of the effective potential to the current potentially becomes more
intricate. On the other hand, the anomaly only contributes to the leading order, such that the
knowledge of the covariant current (that in higher order coincides with the consistent current)
suffices to determine the effective action.
3.4. Four dimensions
For completeness, we also present the results for the effective action and the effective current
in four dimensions. The matching procedure proceeds the same way as in Ref. [1], and we do
not repeat it here.
The effective current in four dimensions consists out of three terms and reads

i
i
Wd=4
= i
A2123 F D H A3123 D HF
2
2

A41234 D HD HD H A ,
(98)
while the effective density can be written as

i 4
1 3
=

F D HD H
B123 F F + B1234
Wd=4
4
2
i 5
i 6
+ B1234
D HF D H + B1234
D HD HF
2
2
7
B12345
D HD HD HD H H .
(99)
3 , B 4 , B 5 , B 6 , and B 7
The functions A2123 , A3123 , A41234 , B123
1234
1234
12345 are given in Appendix B.
1234
4. Next to leading order effective action in two dimensions

In this section we present as a novel result the imaginary part of the effective action in next
to leading order and two dimensions. Even though the results are rather lengthy, the evaluation
of the worldline path integral involves only very basic integrals such that it can be easily implemented using a computer algebra system.
In two dimensions and in next to leading order, the imaginary part of the effective action takes
the form

i
c
W =
Q12 D D HD D H + P12 F D D H
2
i
+ R 123 D D HD HD H + R 123 F D HD H
2
+ M1234 D HD HD HD H
(100)
At next to leading order the action is chiral invariant and the effective action can hence be immediately obtained by matching with the covariant current that in this order coincides with the
consistent current. These functions must have the following properties
P12 = P12 = P21 ,
Q12 = Q12 = Q21 ,
(101)
442
R 123 = R 123 = R 213 ,
R 123 = R 123 = R 213 ,
M1234 = M1234 = M3412 = M4321 .
(102)
(103)
We have chosen a rather general imaginary effective action at the required order which preserves gauge and chiral invariance, but we have included a larger number of terms than necessary
to perform the matching process with the effective current. In fact, the matching process could
be done with solely the functions Q12 , R 123 , R 123 , and M1234 . Instead, we have decided to include the additional term P12 , in order to have the option of simplifying the action by a judicious
choice of this extra function. For example, the extra function can be used to ensure that all functions remain finite at the coincidence limit, as will be explained later on.
The calculation from the worldline formalism leads to the following contributions to the covariant current
1
2
3
D D D HA + iI12
D F A + I123
D D HD HA
W c = i
I12
4
5
6
+ I123
D HD D HA + I123
D D HD HA + I123
D HD D HA
7
8
9
+ iI123
F D HA + iI123
D HF A + I1234
D HD HD HA

10
11
+ I1234 D HD HD HA + I1234 D HD HD HA .
(104)
The coefficient functions are given in Appendix C. In order to express the current in this form,
partial integration has been used to remove terms of the form DA. In addition, indices that are
contracted with the
tensor have been moved to the left, such that a term of the form D D
yields a sum of terms of the type D D and F .
The contributions from the variation of Eq. (100) can be grouped in three levels, with the first
level having only contributions from Q and P ; the second level from the previous ones and R
the last level with all functions. Adding the contributions from the worldline method and
and R;
the variation of Eq. (100) one obtains for the first level the following two equations
1
,
P21 + (m1 + m2 )Q12 = I12

2
2
2
(m1 + m2 )P12 m1 m2 Q21 = I12
,
(105)
which have the solution

Q12 =
2
I21
m21
m22
P21
.
m1 + m2
(106)
Besides, there arises the following restriction which is satisfied and can serve as a check for the
corresponding terms in the effective current
1
2
(m1 + m2 )I21
= I12
,
1
1
2
2
I12
= I12
, I12
= I12
.
The matching equations for the next level are

3
2 (m1 + m2 )Q21 P21 + Q12 + Q21 + (m1 + m3 )(R 123 + R 312 ) = I123
,

5
2 (m1 + m2 )(Q12 + Q21 ) (m1 + m3 )R 312 2Q12 2Q21 + R 312 = I123 ,

2 (m1 + m2 )P12 (m1 m2 )2 (m1 + m2 )Q21
2P12 + 2(m1 m2 )Q21 + (m1 + m3 )(Q13 + 2Q31 )
7
,
R 123 (m1 + m3 ) + (m1 m2 )(m1 + m3 )R 312 = I123
and their complex conjugates.
(107)
(108)
(109)
(110)
443
The first equation (108) is of the form

3
I123
,
m1 + m3
and a set of solutions to Eqs. (108) and (109) is hence given by

1
1
1
I123
I123
I123
,
R123 =
2 m1 + m3 123 2 m1 + m3 312 2 m1 + m3 231
R 123 R 312 =
R 123 = I231 (m1 m2 )R 123 .
(111)
(112)
(113)
The functions I and I are hereby defined as

3
I123 = I123
+ 2 (m1 + m2 )Q21 P21 Q12 Q21 ,

I123 = I 5 2 (m1 + m2 )(Q12 + Q21 ) + 2Q12 + 2Q21 .
(114)
(115)
123
The last equation (110) leads to a constraint on the I functions that is given in Appendix C.
The function R possesses the required symmetries, and it reproduces the effective current correctly, but it is not necessarily finite in the coincidence limit, m1 m3 . One way of solving this
problem is to choose the function P appropriately which up to this point remained undetermined.
Such a choice is e.g. given by
P12 =
2
I12
,
m1 + m2
Q12 = 0,
(116)
which leaves I as
3
I123 = I123
2
I21
(m1 m2 )(m2 m3 )
2
I31
(m1 m3 )(m2 m3 )
(117)
With this choice, R is finite in the coincidence limit, as can be checked explicitly, and since I is
also finite, so is R.
For the last level, the following three equations hold

1 R 312 (2 + 3 ) (m1 + m3 )R 312 + 2R 312
9
2(m1 + m4 )M1234 = I1234
,

(3 1 ) R312 (m1 + m3 ) + 2 R 312 2R 312 + 2R 423
(118)
10
2(m1 + m4 )M4123 = I1234
,

(1 + 2 ) (m1 + m3 )R 312 + 3 R 312 2R 423
(119)
11
+ 2(m1 + m4 )M1234 ) = I1234
.
(120)
One of these equations can be used to determine M, while the other two lead again to constraints on the I functions. The sum of the three equations has the especially simple form
9
10
11
2(m1 + m4 )M4123 = (1 + 2 + 3 )R 312 + I1234
+ I1234
+ I1234
.
(121)
Since all previous functions in the effective action have been chosen finite in the coincidence
limit, so is M1234 . Eqs. (118) and (120) show that M1234 is finite in the limit m1 m2 , while
Eq. (119) shows that M1234 is finite in the limit m1 m4 . This concludes the discussion of the
next to leading order contributions in two dimensions.
444
5. Conclusions
We presented a worldline formalism to calculate the imaginary part of the covariant current
of a general chiral model in the derivative expansion and our results are best summarized by
Eqs. (62) and (63).
The resulting covariant current can be used to reproduce the imaginary part of the effective
action by integration or matching. The advantage of this approach, compared to explicit formulas
of the effective action, as given e.g. in Ref. [19], consists in the manifest chiral covariance. Chiral
covariance reduces the number of possible contributions to the current enormously and makes
even next to leading order calculations manageable as demonstrated in Section 4 in the case of
two dimensions. Besides the chiral covariance, the use of the worldline formalism is essential
in our approach. The evaluation of the worldline path integral requires neither performing Dirac
algebra nor integrating over momentum space, in contrast to the more traditional methods used
in Ref. [1].
In principle, it is possible to use the presented formalism to determine the effective CP violation resulting from integrating out the fermions of the Standard model. For example, in next to
leading order in four dimensions, an operator of the form DHDHFF could arise from the CP violation in the CKM matrix. Since the mass terms of the fermions are treated non-perturbatively in
the derivative expansion, the resulting effective CP-violating operator is not necessarily proportional to the Jarlskog determinant. The discussion of this question is postponed to a forthcoming
publication.
Acknowledgements
T.K. is supported by the Swedish Research Council (Vetenskapsrdet), Contract No. 6212001-1611. A.H. is supported by CONACYT/DAAD, Contract No. A/05/12566.
Appendix A. Integrals used in the calculation
In this section, the function g denotes the bosonic Green function

g(T , 1 ) = y(T )y(1 ) ,
and

)y(1 ) = 2 A (T )A (1 ) ,
g(T
, 1 ) = y(T
(A.1)
(A.2)
where the last expression does not contain a summation over the index A.
A.1. Integrals in two dimensions
In the calculation in two dimensions the following integrals have been used

1
=
J12
dT
T

2
J12
=
0
d1 eT m1 1 (m2 m1 ) =
2
dT
T
T
0
log(m21 /m22 )
m21 m22
dt1 eT m1 1 (m2 m1 ) g(T

, 1 ) =
2
2
(m21 + m22 )
m1
2
+
log
,
2
2
2
2
2
m1 m2 (m1 m2 )
m22

3
J12
dT
T
dt1 eT m1 1 (m2 m1 ) g(T

, 1 )g(T , 1 )
2
= 3
445
m21 + m22
(m21 m22 )3
(m41 + 4m21 m22 + m42 )

(m21 m22 )4

log
m21
m22
(A.3)
The remaining occurring integrals can be expressed as

4
J12
=
dT
T
dT
T
1
d1
6
J123
=
dT
T
d1
7
J123
=
dT
T
d1
0
2
J12

5
J123
d1 eT m1 1 (m2 m1 ) g(T , 1 ) =
2
d2 eT m1
2
2
2
i=1 i (mi+1 mi )
2
d2 eT m1
2
2
2
i=1 i (mi+1 mi )
2
d2 eT m1
2
m21 m22
2
2
i=1 i (mi+1 mi )
1
2 J12
,
m2 + m3
g(T
, 1 ) =
2
2 J12
,
m2 + m3
6
g(T
, 2 ) = J321
.
(A.4)
A.2. Integrals in four dimensions

In four dimensions the following integrals with three indices are used

8
J123
dT
T2
1
d1
=
0
m23 log(m23 )
(m21 m23 )(m22 m23 )
dT
T2
2
2
i=1 i (mi+1 mi )
(m21 m22 )(m21 m23 )
2
d2 eT m1
m21 log(m21 )
+
9
J123
1
d1
m22 (m21 m23 ) log(m22 )

(m21 m22 )(m21 m23 )(m22 m23 )
,
2
d2 eT m1
2
2
2
i=1 i (mi+1 mi )
g(T
, 1 )
0
2
m21 (m41 m22 m23 ) log(m21 )
m1
+
= 2
(m1 m22 )(m21 m23 )
(m21 m22 )2 (m21 m23 )2
m21 m23 log(m23 )
m2 m2 log(m22 )
+
,
2 1 22
(m1 m2 )2 (m22 m23 ) (m21 m23 )2 (m22 m23 )

T
1
2 2
2
2
dT
10
9
J123 =
d1 d2 eT m1 i=1 i (mi+1 mi ) g(T
, 2 ) = J321
.
2
T
0
0
0
(A.5)
446
The integrals with four indices can be expressed as

11
J1234
dT
T2

12
J1234
dT
T2

13
J1234
dT
T2

14
J1234
dT
T2
1
d1
2
d2
d1
d2
d1
d2
d1
0
d2
0
3
d3 eT m1
2
2
i=1 i (mi+1 mi )
3
d3 eT m1
2
2
i=1 i (mi+1 mi )
3
d3 eT m1
2
2
i=1 i (mi+1 mi )
3
d3 eT m1
2
2
2
i=1 i (mi+1 mi )
8
3 J123
,
m3 + m4
g(T
, 1 ) =
9
3 J123
,
m3 + m4
g(T
, 2 ) =
10
3 J123
,
m3 + m4
12
g(T
, 3 ) = J4123
.
(A.6)
Finally, the integrals with five indices read

15
J12345
dT
T2

16
J12345
dT
T2

17
J12345
18
J12345
dT
T2
1
d1
2
d2
3
d3
d1
d2
d3
d1
d2
d3
dT
T2

19
J12345
=
0
dT
T2
d1
d2
d3
d1
0
d2
0
d3
0
4
d4 eT m1
2
2
i=1 i (mi+1 mi )
4
d4 eT m1
2
2
i=1 i (mi+1 mi )
4
d4 eT m1
2
2
i=1 i (mi+1 mi )
4
d4 eT m1
2
2
i=1 i (mi+1 mi )
4
d4 eT m1
2
2
2
i=1 i (mi+1 mi )
11
4 J1234
,
m4 + m5
g(T
, 1 ) =
12
4 J1234
,
m4 + m5
g(T
, 2 ) =
13
4 J1234
,
m4 + m5
g(T
, 3 ) =
14
4 J1234
,
m4 + m5
g(T
, 4 ) =
12
3 J1234
.
m3 + m4
(A.7)
Appendix B. Results in four dimensions

In this section, we give the coefficient functions for the effective current and the effective
density in four dimensions introduced in Section 3.4.
The functions of the covariant current are given by
A2123 =
447
m1 m2 m1 m3 m2 m3
(m1 + m2 )(m1 m3 )(m2 m3 )
m3 (m1 (m2 m3 ) 2m2 m3 ) log[m21 /m23 ]
+ 1
(m1 + m2 )(m1 m3 )(m21 m22 )(m21 m23 )
+
m32 (m2 (m3 m1 ) + 2m1 m3 ) log[m22 /m23 ]

(m1 + m2 )(m2 m3 )(m21 m22 )(m22 m23 )
(B.1)
A3123 = A1321 ,
(B.2)
and
2
2
2
2
L
L
L
L
A41234 = AR
1234 + A1234 log m1 + A2341 log m2 + A3412 log m3 + A4123 log m4 ,
(B.3)
with
m1 m2 m3 m1 m2 m4 + m1 m3 m4 m2 m3 m4
,
(m1 m2 )(m1 + m3 )(m1 m4 )(m2 m3 )(m2 + m4 )(m3 m4 )
m31 (m31 m1 m2 m3 + m1 m2 m4 m1 m3 m4 + 2m2 m3 m4 )
AL
.
=
1234
(m1 m2 )(m1 + m3 )(m1 m4 )(m21 m22 )(m21 m23 )(m21 m24 )
AR
1234 =
(B.4)
The explicit functions occurring in the effective density are rather lengthy and hence we display them in terms of the integrals presented in the last section.
3
8
9
10
B123
= J123
(m1 + m5 ) J123
(m1 m3 ) J123
(m2 m3 ),
4
11
12
13
14
B1234
= J1234
(m1 + m4 ) J1234
(m1 m2 ) J1234
(m2 + m3 ) + J1234
(m4 + m3 ),
5
11
12
13
14
B1234
= J1234
(m1 + m4 ) J1234
(m1 + m2 ) J1234
(m2 m3 ) + J1234
(m4 + m3 ),
6
4
B1234
= B4321
,
7
15
16
17
B12345
= J12345
(m1 + m5 ) J12345
(m1 + m2 ) + J12345
(m2 + m3 )
18
19
J12345
(m3 + m4 ) + J12345
(m4 + m5 ).
(B.5)
Appendix C. Covariant current in next to leading order

In this section we summarize the contributions to the covariant current in next to leading order.
For the first two levels, they are given by
2
I21
1
I12
=
,
m1 m2
3

2
4
= 4 m21 m22 J12
+ 4(m1 + m2 )2 J12
,
I12

m1 + m3
m1 m2 + 2m3
3
2
4
+8 2
=
2 I21
2 m1 m2 J12
I123
2
2
2
m2 m3
m2 m3
+ 16
5
I123
=2
4
m1 m3 J13
m22 m23
2 )
2 (I21
m1 m2
+3
2
I31
(m1 m3 )(m2 m3 )
2
I31
(m1 m3 )(m1 m2 )
3
I123
7
I132
m1 m2
448

2
2
1 I12
1 I12
3m2 m3
7
2
+
I123
= (m2 + m3 )
2 I12
m1 + m2 312 m1 + m2
m2 + m3

4
4

1 J12
1 J12
+ 4(m2 + m3 ) m1 (m1 m2 m3 ) + m2 m3
m1 + m2 312 m1 + m2

4 )
4 )
1 (m1 m2 J12
1 (m1 m2 J12
8(m2 + m3 )
m1 + m2
m1 + m2
312

4)

2 ((m21 4m1 m2 + m22 )J12
4
+ 4(m2 m3 )
(C.1)
2 (m1 m2 )J12 ,
m2 + m3
and the relations
4
3
= I321
,
I123
6
5
I123
= I321
,
8
7
I123
= I321
.
(C.2)
The contributions to the last level read

9
11
I1234
= I4321
,

m1 + m4 11
1
m1 + m4
5
2 +
I4123 +
(3 )4123 I123
m3 m4
2
m3 m4

m1 + m3 3
m1 + m4
1
I213 2 I12
+ 2 + 3 +
(3 )4123
213
m3 m4
m2 m3

3

m1 + m4
1
1
2 + 3 +
(1 + 2 + 3 )4123 I123
2 I12
+
2
m3 m4

3

m1 + m4
1
1
1 + 2 +
(3 )4123 I321
2 I12
+
321
2
m3 m4
3

3

m1 + m3
2
1
1
I 2 I12
+
I 2 I12
213
432
2(m2 m3 )(m3 m4 ) 213
m2 + m4 432

1
1
I 3 2 I12
,
421
2(m3 m4 ) 421
10
I1234
=
3
4
5
4
4(m3 + m4 )J34
4(m21 m23 )J13
I124
4J124
m3 m4 (m3 m4 )(m21 m24 )

m3 m4
m21 m24

2
2
2
4 ))
1 ((m1 m2 )(I12 8m1 m2 J12
m1 + m2
m2 + m3
2
3
m3 m4
m3 + m4
(m21 m22 )(m1 + m2 )
11
I1234
= 2 I 4
2 8m m J 4 ))
(m3 + m4 )(1 (I12
1 2 12 413
m24 )(m1
4(m2 + m3 )(m1 m4 )
1 J 4
m1 + m2 )(m3 m4 )
m4 )

2
4 (m2 m2 )J 4
(m1 + m23 )J13
4(m1 + m2 )
m2 + m3
2
3 23
2 +
3
m3 m4
m3 + m4
m21 m22
(m21
4(m2 + m3 )(m1 m4 )(m3 + m4 ) (1 J 4 )312 4(m1 + m3 )

1
3 J 5
m1 + m2
m1 m3
m3 + m4
4(m1 + m2 )(m2 + m4 )
3 J 6 .
m23 m24
(C.3)
449
All functions I are finite in the coincidence limit and must fulfill the following constraints
due to the behavior of the terms in the effective action under cyclic permutation and complex
conjugation:
3
(m1 + m3 )I312
2
2
I31
I13
+
m2 m3
(m1 m2 )(m2 m3 ) (m1 m2 )(m1 m3 )
2
2
I32
(m1 + m3 )I23
= 0,
(m1 m2 )(m22 m23 ) (m1 m2 )(m2 m3 )
3
3
I123
+ I321
3
+ I321
+
3
(m1 + m3 )I231
2
I12
m21 m22
m1 m2
+
5
5
+ I123
+ I321
2
I21
(m1 m2 )(m2 m3 )
3
5
8
(m1 + m3 )I312
+ (m1 + m3 )I312
I123
+
2
I31
(m1 m2 )(m2 m3 )
= 0,
2
2I13
m1 m2
2
(3m1 m2 + 2m3 )I23
= 0,
(m1 m2 )(m2 + m3 )
(C.4)
and
3
3
I123
= I123
= 0,
5
5
I123
= I123
= 0,
7
7
I123
= I123
= 0.
(C.5)
References
[1] L.L. Salcedo, Derivative expansion for the effective action of chiral gauge fermions. The abnormal parity component, Eur. Phys. J. C 20 (2001) 161, hep-th/0012174.
[2] L.L. Salcedo, Derivative expansion for the effective action of chiral gauge fermions. The normal parity component,
Eur. Phys. J. C 20 (2001) 147, hep-th/0012166.
[3] J. Smit, Effective CP violation in the Standard Model, JHEP 0409 (2004) 067.
[4] A. Tranberg, J. Smit, Baryon asymmetry from electroweak tachyonic preheating, JHEP 0311 (2003) 016, hepph/0310342.
[5] T. Konstandin, T. Prokopec, M.G. Schmidt, Axial currents from CKM matrix CP violation and electroweak baryogenesis, Nucl. Phys. B 679 (2004) 246, hep-ph/0309291.
[6] J. Schwinger, On gauge invariance and vacuum polarization, Phys. Rev. 82 (1951) 664.
[7] R.P. Feynman, Mathematical formulation of the quantum theory of electromagnetic interaction, Phys. Rev. 80 (1950)
440.
[8] M.J. Strassler, Field theory without Feynman diagrams: One-loop effective actions, Nucl. Phys. B 385 (1992) 145,
hep-ph/9205205.
[9] M.G. Schmidt, C. Schubert, On the calculation of effective actions by string methods, Phys. Lett. B 318 (1993) 438,
hep-th/9309055.
[10] Z. Bern, D.C. Dunbar, A mapping between Feynman and string motivated one loop rules in gauge theories, Nucl.
Phys. B 385 (1992) 145, hep-ph/9205205.
[11] Z. Bern, D.A. Kosower, Efficient calculation of one loop QCD amplitudes, Phys. Rev. Lett. 66 (1991) 1669.
[12] M. Reuter, M.G. Schmidt, C. Schubert, Constant external fields in gauge theory and the spin 0, 12 , 1 path integrals,
Ann. Phys. 259 (1997) 313, hep-th/9610191.
[13] D. Fliegner, P. Haberl, M.G. Schmidt, C. Schubert, Inverse mass expansions from worldline path integralshigher
order coefficients and ordering problems, hep-th/9505077.
[14] D. Fliegner, M.G. Schmidt, C. Schubert, The higher derivative expansion of the effective action by the string inspired
method. Part 1, Z. Phys. C 64 (1994) 111, hep-ph/940101221 .
[15] D. Fliegner, M. Reuter, M.G. Schmidt, C. Schubert, The two-loop EulerHeisenberg Lagrangian in dimensional
renormalization, Theor. Math. Phys. 113 (1997) 1442, hep-th/9704194.
450
[16] H.T. Sato, M.G. Schmidt, C. Zahlten, Two-loop YangMills theory in the world-line formalism and an Euler
Heisenberg type action, Nucl. Phys. B 579 (2000) 492.
[17] M.G. Schmidt, C. Schubert, Worldline Greens functions for multiloop diagrams, Phys. Lett. B 331 (1994) 69,
hep-th/9403158.
[18] M.G. Schmidt, C. Schubert, Multiloop calculations in the string-inspired formalism: The single spinor-loop in QED,
Phys. Rev. D 53 (1996) 2150, hep-th/9410100.
[19] E. DHoker, D.G. Gagn, Worldline path integrals for fermions with general couplings, Nucl. Phys. B 467 (1996)
297, hep-th/9512080.
[20] M. Mondragn, L. Nellen, M.G. Schmidt, C. Schubert, Yukawa couplings for the spinning particle and the worldline
formalism, Phys. Lett. B 351 (1995) 200, hep-th/9502125.
[21] M. Mondragn, L. Nellen, M.G. Schmidt, C. Schubert, Axial couplings on the world-line, Phys. Lett. B 366 (1996)
212, hep-th/9510036.
[22] E. DHoker, D.G. Gagn, Worldline path integrals for fermions with scalar, pseudoscalar and vector couplings,
Nucl. Phys. B 467 (1996) 272, hep-th/9508131.
[23] Y. Ohnuki, T. Kashiwa, Coherent states of Fermi operators and the path integral, Prog. Theor. Phys. 60 (1978) 548.
[24] E. DHoker, E. Farhi, Decoupling a fermion whose mass is generated by a Yukawa coupling: The general case,
Nucl. Phys. B 248 (1984) 59.
[25] T. Konstandin, CP-violation and baryogenesis on electroweak scales, PhD thesis, Universitt Heidelberg, 2005.
[26] R.A. Bertlmann, Anomalies in Quantum Field Theory, Oxford Science Publications, 2000.
[27] W.A. Bardeen, B. Zumino, Consistent and covariant anomalies in gauge and gravitational theories, Nucl. Phys.
B 244 (1984) 421.
[28] J. Wess, B. Zumino, Consequences of anomalous Ward identities, Phys. Lett. B 37 (1971) 95.
[29] E. Witten, Global aspects of current algebra, Nucl. Phys. B 223 (1983) 422.
[30] M.A. Shifman, Wilson loop in vacuum fields, Nucl. Phys. B 173 (1980) 13.
[31] R.P. Feynman, An operator calculus having applications in quantum electrodynamics, Phys. Rev. 84 (1951) 108.

Form factors and correlation functions of an interacting

spinless fermion model
Kohei Motegi , Kazumitsu Sakai
Institute of physics, University of Tokyo, Komaba 3-8-1, Meguro-ku, Tokyo 153-8902, Japan
Abstract
Introducing the fermionic R-operator and solutions of the inverse scattering problem for local fermion
operators, we derive a multiple integral representation for zero-temperature correlation functions of a onedimensional interacting spinless fermion model. Correlation functions particularly considered are the oneparticle Greens function and the densitydensity correlation function both for any interaction strength and
for arbitrary particle densities. In particular for the free fermion model, our formulae reproduce the known
exact results. Form factors of local fermion operators are also calculated for a finite system.
PACS: 05.30.-d; 71.10.Fd; 02.30.Ik
Keywords: Spinless fermion model; Correlation function; Form factor; Bethe ansatz; Integrable system; Heisenberg
XXZ model
1. Introduction
The evaluation of correlation functions has been one of the challenging problems in research
on quantum many-body systems in one dimension, since most of the intriguing phenomena induced by underlying strong quantum fluctuations are theoretically described through correlation
functions. In these systems, it is widely known that the existence of models which are solvable
by means of Bethe ansatz (see [1,2] for example). Though the exact computation of correlation
functions, of course, is still tremendously difficult even in such models, several analytical meth* Corresponding author.
E-mail addresses: motegi@gokutan.c.u-tokyo.ac.jp (K. Motegi), sakai@gokutan.c.u-tokyo.ac.jp (K. Sakai).

doi:10.1016/j.nuclphysb.2007.10.009
452
K. Motegi, K. Sakai / Nuclear Physics B 793 [FS] (2008) 451468
ods have been recently developed to derive manageable expressions for correlation functions,
especially in the spin-1/2 Heisenberg XXZ chain.
In 1990s, correlation functions of the spin-1/2 XXZ chain at zero temperature and for zero
magnetic field have been expressed as multiple integral forms derived by the q-vertex operator
approach [35]. An alternative method combining the algebraic Bethe ansatz with solutions to
the quantum inverse problem has been provided for the XXZ chain in arbitrary magnetic fields
[68] (see also [9] for a recent review). This method can be generalized to the finite-temperature
and/or the time-dependent correlation functions [1014].
In general, one-dimensional (1D) quantum spin systems are mapped to 1D fermion systems
through the JordanWigner transformation. For the spin-1/2 XXZ chain, the corresponding
system is a spinless fermion model with the nearest neighbor hopping and interaction. In the
thermodynamic limit, the bulk quantities in the XXZ chain are exactly the same as those in
the spinless fermion model. Namely, in the 1D quantum systems, the difference of the statistics
between the spin and the fermion does not show up, as long as we concentrate on their bulk
quantities.
The situation, however, is radically changed when the quantities accompanying a change of
the number of particles are considered. For instance, let us consider the (equal-time) one-particle
for the spinless fermion model and the transverse spinspin correlaGreens function c1 cm+1
+
tion function 1 m+1 for the XXZ chain, which are intuitively the same. In fact, due to the
difference of the statistics, or equivalently the non-locality of the JordanWigner transformation,
both the correlation functions exhibit completely different behavior. For instance, one observes
an oscillatory behavior for the one-particle Greens function (referred to as the kF -oscillation,
where kF is the Fermi momentum), which is peculiar to the fermion systems. In contrast, for the
transverse spinspin correlation function, such an oscillatory behavior does not appear.1
As mentioned above, exact expressions for the correlation functions of the XXZ chain have
already been proposed in the form of multiple integral representations. Unfortunately, once the
JordanWigner transformation is performed and the XXZ chain is considered instead of the
spinless fermion model, it is difficult to trace the difference of the statistics in the framework
of multiple integral representations. Namely, to derive a manageable expression of correlation
functions for the spinless fermion model, we must directly treat the fermion system from the
very beginning.
In this paper, introducing the fermionic R-operator [15] which acts on the fermion Fock space,
we directly treat the spinless fermion model without mapping to the XXZ chain. Combining the
method provided in the XXZ chain [7] with solutions to the inverse scattering problem of local fermion operators [16], we derive a multiple integral representing the equal-time one-particle
Greens function and the densitydensity correlation function at zero temperature both for any interaction strengths and for arbitrary particle densities. Our formulae reproduce the known results
for the free fermion model. In addition to the correlation functions, we also compute form factors for local fermion operators, which might be useful for systematic evaluations of the spectral
functions for the spinless fermion model.
This paper is organized as follows. In the subsequent section, we introduce the fermionic
R-operator and the transfer operator constructed by the R-operator. Then we briefly review the
algebraic Bethe ansatz for the spinless fermion model. The scalar product of a Bethe state with
1 In fact, there could exist an oscillatory behavior in the transverse spinspin correlation function, which, however, can
be eliminated by a gauge transformation.
453
an arbitrary state is presented in Section 3. Combining solutions of the inverse scattering problem
with the scalar product, we compute form factors of local fermion operators. Multiple integral
representations for correlation functions are derived in Section 4. In Section 5, using the multiple
integral representations, we explicitly evaluate correlation functions for the free fermion model.
Section 6 is devoted to a brief conclusion.
2. Spinless fermion model
The Hamiltonian of the spinless fermion model on a periodic lattice with M sites is defined
as
H = H0 c
H0 = t
j =1
M

j =1
M

1

nj ,
cj cj +1 + cj+1 cj + 2
1
nj
2

1
1
nj +1
,
2
4
(2.1)
where cj and cj are the fermionic creation and annihilation operators at the j th site, respectively,
satisfying the canonical anti-commutation relations

{cj , ck } = cj , ck = 0,
(2.2)
cj , ck = j k .
Here t and are real constants characterizing the nature of the ground state, and c denotes
the chemical potential coupling to the density operator nj = cj cj . By use of the JordanWigner
transformation
j 1
j 1

1
+
cj = exp i
(2.3)
nk j ,
cj = exp i
nk j , nj = 1 jz ,
2
k=1
k=1
the spinless fermion model (2.1) can be mapped to the spin-1/2 XXZ chain with an external
magnetic field h:
HXXZ = J
M

j =1
j+ j+1 + j++1 j +

M

z z
h z
j ,
j j +1 1
2
2
(2.4)
j =1
where we have changed the variables t and c in (2.1) as t J and c h.

The difference between the spinless fermion model and the corresponding XXZ chain lies
only in the difference of boundary conditions. Therefore, in the thermodynamic limit, the quantities without accompanying a change of the number of particles such as the densitydensity
correlation function n1 nm+1 are exactly the same as those such as a longitudinal spinspin
z
correlation function (1 1z )(1 m+1
)/4 for the XXZ chain. As mentioned in the preceding section, however, the quantities changing the number of particles such as the one-particle
Greens function c1 cm+1

exhibit completely different behavior from those such as the trans
, due to the non-locality of the JordanWigner
verse spinspin correlation function 1+ m+1
for the XXZ chain has already

transformation (2.3). Though the exact expression of 1+ m+1
been presented in [7], it is difficult to convert it to the expression for the spinless fermion model,
as long as we concentrate on the XXZ chain (2.4). Namely, we should treat the original spinless
454
fermion model without mapping to the XXZ chain. In this section, introducing the fermionic
R-operator [15], we directly consider the spinless fermion model (2.1).
The fermionic R-operator is defined by
R12 () = 1 n1 n2 +

sh
sh
(n1 + n2 2n1 n2 ) +
c1 c2 + c2 c1 ,
sh( + )
sh( + )
(2.5)
which acts on V1 s V2 . Here Vj is a two-dimensional fermion Fock space whose normalized

orthogonal basis is given by |0j and |1j := cj |0j , where cj |0j = 0 and s denotes the super
tensor product. Note that this fermionic R-operator satisfies the YangBaxter equation [15]:
R12 (1 2 )R13 (1 )R23 (2 ) = R23 (2 )R13 (1 )R12 (1 2 ).
(2.6)
Identifying one of the two fermionic Fock spaces with the quantum space Hm , we define the
L-operator at the mth site by
Lm () = Ram ( m ),
(2.7)
where m are inhomogeneous parameters assumed to be arbitrary complex numbers. The fermionic monodromy operator is then constructed as a product of the L-operators:
T () = LM () L2 ()L1 ()
= A()(1 na ) + B()ca + ca C() + D()na .
(2.8)
Thanks to the YangBaxter equation (2.6), the fermionic transfer operator defined by
T () = str a T () = a 0|T ()|0a a 1|T ()|1a = A() D(),
(2.9)
constitutes a commuting family: [T (), T ()] = 0, where the dual fermion Fock space is
spanned by a 0| and a 1| with a 1| := a 0|ca and a 0|ca = 0.
The Hamiltonian of the spinless fermion model H0 (2.1) can be expressed as the logarithmic
derivative of the fermionic transfer operator T () in the homogeneous limit m /2:

,
= ch .
H0 = t sh() ln T ()
(2.10)
=m =
2
Due to the YangBaxter equation (2.6), one immediately sees that the following relation is valid
for arbitrary spectral parameters 1 and 2 :
R12 (1 2 )T1 (1 )T2 (2 ) = T2 (2 )T1 (1 )R12 (1 2 ).
(2.11)
This leads to the commutation relations among the operators A(), B(), C() and D() constructing the monodromy operator (2.8). Let us explicitly write down some of these relations,
which we will use later:
B()B() = B()B(),
C()C() = C()C(),
A()B() = f (, )B()A() g(, )B()A(),

D()B() = f (, )B()D() + g(, )B()D(),
C()A() = f (, )A()C() g(, )A()C(),
C()D() = f (, )D()C() + g(, )D()C(),

C()B() + B()C() = g(, ) D()A() D()A() ,
(2.12)
455
where
sh( + )
sh
(2.13)
,
g(, ) =
.
sh( )
sh( )
Let | be the vacuum state for the system (2.1). Obviously | is an eigenstate of the transfer
operator T ():
f (, ) =

T ()| = a() d() |,
a() = 1, d() =
M

f 1 (, m ).
(2.14)
m=1
A general N -particle state | can be constructed by a multiple action of B() on | as

| =
N

B(j )|,
N = 0, 1, . . . , M.
(2.15)
j =1
Utilizing commutation relations in (2.12), one finds that the state (2.15) becomes an eigenstate
of T (), if the complex parameters {j } satisfy the Bethe ansatz equation;
(1)N+1 =
N
M
N
d(j ) f (j , k ) sh(j m ) sh(j k + )
=
.
a(j )
f (k , j )
sh(j m + )
sh(j k )
k=1
k=j
m=1
(2.16)
k=1
k=j
Then the corresponding eigenvalue () (i.e. T ()| = ()|) is given by

() = a()
N

j =1
f (j , ) + (1)N+1 d()
N

f (, j ).
(2.17)
j =1
(1)N+1
Note that the existence of an extra factor

in (2.16) and (2.17) reflects the fermionic
structure of the system, which does not appear for the XXZ chain (2.4). The relation (2.10) leads
to the explicit expression of the energy spectrum per site e0 of the spinless fermion model:

N
t sh2
1
1 N
c
.
e0 =
(2.18)
M
2 M
sh(j + 2 ) sh(j 2 )
j =1
In the thermodynamic limit, where M , N , and N/M is a constant determined

by a function of the chemical potential c , the Bethe ansatz equation (2.16) characterizing the
ground state reduces to the following linear integral equation for the distribution function of the
Bethe roots:

2itot () + K( )tot () d = t ,
(2.19)
.
2
C
In the above, the integral kernel K() and the function t (, ) are respectively defined by
sh
sh 2
,
t (, ) =
.
(2.20)
sh( + ) sh( )
sh( ) sh( + )
The integration contour C = [c , c ] is given by (c ) = 0, where () is the dressed
energy satisfying the following linear integral equation:

+ c .
2i() + K( )() d = t sh()t ,
(2.21)
2
K() =
456
Namely, for 1 < 1 ( = i , > 0), the contour C is defined by an interval on the real
axis. In particular, at c 0, c . On the other hand, for > 1 ( < 0), the Bethe roots
are distributed on the imaginary axis. In particular, at c = 0, c = i/2. Then the energy
density (2.18) explicitly reads

1
e0 = t sh() t ,
tot () d c
nj
2
2
C

c
= t ,
(2.22)
() d
,
2
2
C
where nj is the particle density given by

nj = tot () d.
(2.23)
Note that the above expression of the ground state energy (2.22), the distribution function (2.19)
and the dressed energy (2.21) are exactly the same as those for the XXZ chain [1,2].
For later convenience, let us define the inhomogeneous distribution function (, ) [7] as the
solution of the integral equation:

2i(, ) + K( )(, ) d = t (, ).
(2.24)
C
Correspondingly, the inhomogeneous total distribution function tot (, { }) is defined by

M

1
tot , { } =
(, m ),
M
(2.25)
m=1
which reduces to tot () (2.19), when one takes the homogeneous limit m /2.
In particular, at zero chemical potential c = 0 (corresponding to the half filling case nj =
1/2), one has
2 ch (+ 2 ) , for || < 1, = i,
1q 2n 2 3 (i(+ 2 ),q)
2n(+ 2 )
(, ) =
(2.26)
i
i
e
=
,
n=
n=1
2
ch(n)
2
1+q 2n 4 (i(+ 2 ),q)
for > 1, q = e ,
where n (, q) is the Jacobi theta function.
3. Scalar products and form factors
3.1. Scalar products
Our main aim is to evaluate the correlation function of the spinless fermion model (2.1):
g |O1 Om+1 |g
,
O1 Om+1 =
g |g
(3.1)
457
where |g is the ground state constructed by substituting an appropriate solution {} of the Bethe

ansatz equation (2.16) into (2.15), and g | is the dual state: g | = | N
j =1 C(j ). In this
paper we concentrate on the one-particle Greens function (Oj = cj , cj ) and the densitydensity
correlation function (Oj = nj ). Here and in what follows, we consider the inhomogeneous
model (2.8) for convenience. The correlation functions of the original Hamiltonian (2.1) or (2.10)
can be obtained by taking the homogeneous limit m /2.
To apply the algebraic Bethe ansatz to the computation of (3.1), we must express local fermionic operators in terms of elements of the monodromy operators (2.8). In fact, such an expression
has already been obtained by solving the quantum inverse scattering problem for local fermion
operators [16].
Theorem 3.1. (See [16].) For the inhomogeneous spinless fermion model (2.8) with arbitrary
inhomogeneous parameters m (m = 1, . . . , M), the local fermionic operators acting on the j th
space can be expressed in terms of the elements of the fermionic monodromy operator as
cj
j
1
T (k )B(j )
k=1
nj =
j

cj =
(k ),
k=1
j
1
T (k )D(j )
k=1
j
1
T (k )C(j )
k=1
j

T 1 (k ),
1=
k=1
j

T 1 (k ),
k=1
M

T (k ).
(3.2)
k=1
To proceed the evaluation by use of Theorem 3.1, one needs

to calculate the action of the

operators A, B, C and D on an arbitrary state | = | N
j =1 C(j ), where {} is a set of
arbitrary complex numbers (not necessary the Bethe roots). After simple but tedious computation
utilizing (2.12), one obtains
|
N

C(j )A(N+1 ) =
j =1
|
N

N+1
N
a(b )
b=1
C(j )D(N+1 ) = (1)
j =1
|
N

N+1

j =1 sh(j b + )
C(j ),
|
N+1
j =1 sh(j b )
j =1
j =b
j=
b
N+1
N
N+1

j =1 sh(a j + )
d(a ) N+1
C(j ),
|
j =1 sh(a j )
a=1
j =1
j =a
j =a
C(j )B(N+1 )
j =1
= (1)
N1
N+1
N
k=1 sh(a k + )
d(a )
N+1
k=1 sh(a k )
a=1
k=a
N+1

a =1
a =a
N+1
j =1 sh(j a + )
N+1

a(a )
j =a
C(j ).
|

N+1
sh(N+1 a + )
j =1 sh j a
j =1
j =a,a
j =a,a
(3.3)
458
Note that the extra sign factors (1)N and (1)N+1 appearing in the last two relations reflect
the fermionic nature of the system (cf. (3.7) and (3.8) in [7]). Finally the action of the operator
C() is free.
In the above, one distinguishes the terms containing a(N+1 ) or d(N+1 ) and the others,
which are referred to as the direct and indirect terms,
respectively.
Due to the existence of indirect
N
C(
)
terms, the explicit form of the scalar product | N
j
j =1
j =1 B(j )| is required to
N
compute the expectation value (3.1), where {j }j =1 are solutions of the Bethe ansatz equation
and {j }N
j =1 are arbitrary complex numbers. This can be explicitly evaluated in the same way
as in [1,17]. The difference between the present system and the XXZ chain stems from the
difference of the commutation relations (2.12). Here we write down only the result.
Proposition 3.1. The scalar product between a Bethe state and an arbitrary state
N
N

SN {} {} = |
C(j )
B(j )|
j =1
can be expressed as follows:

SN {} {} = SN {} {}
= (1)
N(N1)
2
where {j }N
j =1 are Bethe roots,
({}|{}) is defined by
(3.4)
j =1
N
N
a,b=1 sh(a
b + )

detN {} {} ,
a<b sh(a b ) sh(b a )

{j }N
j =1 are arbitrary complex parameters.
(3.5)
The N N matrix
N

sh(k a + )
,
j k {} {} = t (j , k ) (1)N+1 d(k )t (k , j )
sh(k a )
(3.6)
a=1
and detN denotes the determinant of an N N matrix.

Reflecting the fermion statistics, the extra factor (1)N+1 enters into the above expression.
This factor, however, is eliminated by that in the Bethe ansatz equation (2.16), when one takes
the limit {} {}, which gives the square of the norm of the Bethe vector:
N

N(N1)
sh(a b + )
SN {} {} = (1) 2 shN ()
detN {} ,
sh(a b )
(3.7)
a,b=1
a=b
where

j k {} = j k
N
d (j )
K(j a ) + K(j k ).
d(j )
(3.8)
a=1
Set {} = {1 , . . . , n } {n+1 , . . . , N } in (3.5). Then the ratio of the determinants of and

can be evaluated in the thermodynamic limit:

1
detN
Mtot a , { }
detn (j , k ),
=
detN
n
a=1
where (, ) and tot (, { }) are defined as (2.24) and (2.25), respectively.
(3.9)
459
3.2. Form factors

Utilizing the scalar product obtained in the preceding subsection, we derive determinant representations of the form factors for a finite system. These formulae might be useful in calculating
the spectral functions of the present model.
The form factors of the local fermion operators are defined by
N
N+1

C(j )cm
B(k )|,
FN m {} {} = |
j =1
k=1
N+1
N

C(k )cm
B(j )|,
FN+ m {} {} = |
j =1
k=1

FNz m {} {} = |
N

C(j )(1 2nm )
j =1
N

B(k )|,
(3.10)
k=1
N+1
N
where {k }N
k=1 and {j }j =1 (or {j }j =1 ) are solutions of the Bethe ansatz equation (2.16). To
evaluate them, first we substitute solutions of the quantum inverse scattering problem for local
fermion operators (3.2) into the above expressions. For instance, FNz (m|{}|{}) can be written
as

FNz m {} {}
= 2|
N

C(j )
j =1
m1

T (l )A(m )
l=1
M

T (l )
l=m+1
N

B(k )| SN {} {} .
(3.11)
k=1

Then, noting that the states B(k )| and | C(j ) are the Bethe states, we remove the
products of the transfer matrices by (2.17). Finally
M the determinant representation of the
applying
scalar product (3.5), and using a simple identity N
j =1 k=1 f (j , k ) = 1 derived by the Bethe
ansatz equation (2.16), we obtain the determinant representations of the form factors. They are
summarized into the following proposition.
Proposition 3.2. The form factors FN (m|{}|{}) and FN+ (m|{}|{}) for local fermion opera
tors cm
and cm are respectively given by the following determinant representations:
N+1
N+1

N(N+1) m1 ({})
j =1 sh(j m + )

FN m {} {} = (1) 2
N
N
m1 ({})
k=1 sh(k m + )
N+1
j <k

FN+ m {} {} =
detN+1 H (m|{}|{})
,

sh(k j ) N
< sh( )

FN m {} {} ,
N+1
N+1
m1 ({})m ({})
N ({}) N ({})
m
m1
(3.12)
N ({}) is
where the coefficient m
N
m
{} =
m
N

j =1 k=1
f (j , k ),
(3.13)
460
and the (N + 1) (N + 1) matrix H (m|{}|{}) is defined by
N+1
N+1

sh
Hjk =
sh(a k + ) (1)N d(k )
sh(a k )
sh(j k )
a=1
a=j
a=1
a=j
for k N,
HjN +1
= t (j , m ).
(3.14)
On the other hand, the form factor FNz (m|{}|{}) for the operator 1 2nm is given by
N
N

N(N1) m1 ({}) sh(j m + )
FNz m {} {} = (1) 2
N ({})
sh(j m + )
m1
j =1
detN (H ({}|{}}) 2P (m|{}|{}))

,
N
j <k [sh(j k ) sh(k j )]
(3.15)
where H ({}|{}) is an N N matrix defined as

sh
N+1
sh(a k + ) (1)
d(k )
sh(a k ) ,
Hj k =
sh(j k )
a=j
a=j
(3.16)
and P (m|{}|{}) is an N N matrix of rank one,
Pj k (m) = t (j , m )
N

sh(a k + ).
(3.17)
a=1
Here we have set a() = 1. Note that, in the derivation of FNz (m|{}|{}), we have used the
orthogonality of the Bethe vectors, and the formula that the determinant of the sum of an arbitrary
N N matrix A and an N N matrix of rank one B can be given by
det(A + B) = det A +
N

l=1

det A ,
(l)
A(l)
jk
Aj k
Bj k
for k = l,
for k = l.
(3.18)
Comparing the form factor for the local spin operator of the third component mz in the XXZ
chain [18], one finds that the extra factor (1)N+1 appears in (3.16). This factor, however, is
canceled by that in the Bethe ansatz equation (2.16). Hence, in the thermodynamic limit where
the distribution of the Bethe roots (2.26) characterizing the ground state is the same as that of the
XXZ chain, FNz (m|{}|{}) (3.15) coincides with the form factor of the mz for the XXZ chain.2
In contrast to this, for the form factor FN (m|{}|{}) (3.12), the factor (1)N in (3.14) can
not be eliminated by the Bethe ansatz equation (2.16). Namely, compared with the form factors
of the m for the XXZ chain, the extra minus sign remains in (3.14). This essential difference
caused by the statistics between the fermion and the spin induces the different behavior of the
and 1+ m+1

.
correlation functions c1 cm+1
2 Note that the overall factor (1)N (N 1)/2 in (3.15) and (3.12) can be eliminated by the normalization of the Bethe
vector (see (3.7)).
461
4. Multiple integral representations for correlation functions

In this section, we derive multiple integrals representing the one-particle Greens function and
the densitydensity correlation function. Inserting the expression of the local fermion operators
cj and cj (3.2) into (3.1), one obtains

=
c1 cm+1
g |C(1 )
m
a=2 (A D)(a )B(m+1 )
m+1
b=1
(A D)1 (b )|g
g |g
(4.1)
Note that c1 cm+1 = c1 cm+1

.
Following [7], for the
densitydensity correlation function, we conveniently introduce an operator Q1,m as Q1,m = m
k=1 nk and consider the expectation value of the generating function
exp(Q1,m ) ( C):

m
1

g | m
a=1 (A e D)(a ) b=1 (A D) (b )|g
exp(Q1,m ) =
(4.2)
.
g |g
Then the densitydensity correlation can be expressed as

1 2 2
n1 nm+1 = Dm 2 exp(Q1,m )
,
2
=0
(4.3)
2 f (m) =
where Dm denotes the lattice derivative defined by Dm f (m) = f (m + 1) f (m) and Dm
f (m + 1) 2f (m) + f (m 1). In fact, there exist several manners to express the density
density correlation by the generating function [7]. For instance, n1 nm+1 is also written by
using exp(Q1,m )nm+1 as

,
exp(Q1,m )nm+1
n1 nm+1 = Dm1
=0

exp(Q1,m )nm+1

m+1
1
g | m
a=1 (A e D)(a )D(m+1 ) b=1 (A D) (b )|g
.
=
(4.4)
g |g
To evaluate the correlation functions (4.1)(4.4),

we must calculate the multiple action of
(A e D)(x) on an arbitrary state | = | N
j =1 C(j ), where {} are arbitrary complex
numbers. Utilizing the commutation relations among A, D and C (2.12), one finds that the action
on a general state can be written as
|
m

A e D (xa )
a=1
p
n=0 {}={+ }{ }
{x}={x + }{x }
|+ |=|x + |=n
n

Nn
Rn {x + } {x } {+ } { } |
C xa+
C
b ,
a=1
b=1
where |+ |, |x + |, etc. denote the number of elements of {+ }, {x + }, etc.; p = min(m, N ).
(4.5)
462
Proposition 4.1. The coefficient Rn ({x + }|{x }|{+ }|{ }) in (4.5) is given by

Rn {x + } {x } {+ } { }

n

+ + mn
Nn

= Sn {x } { } { }
f x+, x
f , x
a x
a
a=1
n

e d xa
b=1
b=1

+ Nn

f xa , xb
f xa ,
,
b
b=1
(4.6)
b=1
where the highest coefficient is expressed as

n
+
+
+ +
a,b=1 sh(xa b + )

Sn {x } { } { } = n
+
+
+
+ detn Mj k
a<b [sh(b a ) sh(xa xb )]
(4.7)
with

+ + Nn
+
f
M j k = a +
a , j
j t xk , j
a=1
n sh(+ x + + )

+ + Nn
j
b
+
+ e d +
t
,
x
,
a
j
j
k
j
+
+
sh(
x
j
b )
a=1
b=1
(4.8)
The proof is completely parallel to that in the XXZ chain [7]. The difference of the expression
stems from the difference of the commutation relation of C and D, which reflects the fermionic
nature of the present system.
If | is a Bethe vector (or equivalently the eigenstate of the transfer operator (2.9)), one
further has the following expression which is directly applicable to the computation of the generating function (4.2).

Corollary 4.1. If | = | N
j =1 C(j ) is a Bethe vector, namely if {} are the solutions
to
the
Bethe
ansatz
equation
(2.16),
the coefficient Rn ({ + }|{ }|{+ }|{ }) of the action
m
D)( ) is given by
(A
e
a
a=1
+ +
Rn { } { } { } { }
n
n mn
Nn

mn

Nn

+
= Sn { + } {+ }
f a+ , b
f
,
f
a
a , b ,
b
a=1 b=1
where
+
n
+
+
a,b=1 sh(a b + )
+
+
+
+
a<b [sh(b a ) sh(a b )]

Sn { } {+ } = n

a=1 b=1
(4.9)
a=1 b=1
detn M j k .
(4.10)
Here the n n matrix M is defined as

n sh(+ + + ) sh(+ + + )

+ +
a
a
j
j
+
e
t
.
M j k = t k+ , +
j
j
k
+
+
+
+
sh(
+
)
sh(
a
a
j
j + )
a=1
(4.11)
463
Note that here we have set a() = 1 and d(a ) = 0, and used the fact that the parameters {}
satisfy the Bethe ansatz equation (2.16).
Now we would like to compute the correlation functions. First let us consider the generating
function (4.2) of the densitydensity correlation n1 nm+1 . Noting that |g is the Bethe vector characterizing
the ground state,

one finds that the numerator in (4.2) can be expressed by
D)( )| / m ( ), where () is the corresponding eigenvalue of the
(A
e
g | m
a
g
a
a=1
a=1
transfer matrix (see (2.17)). Then from Corollary 4.1, Proposition 3.1 and the norm of the Bethe
vector (3.7), the generating function (4.2) is given by a determinant form. By d(a ) = 0, it immediately follows that the determinant form is the same as that for the generating function of the
z
[7]. Thus utilizing the method proposed
longitudinal spinspin correlation function 1z m+1
in [7], we arrive at the following proposition.
Proposition 4.2. In the thermodynamic limit M , the ground state expectation value of the
generating function exp(Q1,m ) is expressed as

n
m
m
n

dzj
1
sh(za b + ) sh(a b )
n
exp(Q1,m ) =
d
2
2i
sh(za b ) sh(a b + )
(n!)
n=0
a=1 b=1
j =1
C

Wn {} {z} detn M j k {} {z} detn (j , zk ) ,
(4.12)
where
n
n

sh(a zb + ) sh(zb a + )
Wn {} {z} =
,
sh(a b + ) sh(za zb + )
(4.13)
n

sh(a j + ) sh(j za + )
.
M j k {} {z} = t (zk , j ) + e t (j , zk )
sh(j a + ) sh(za j + )
(4.14)
a=1 b=1
and
a=1
The contour C depends on the regime and the magnitude of the chemical potential, while
encircles {j }m
j =1 and does not contain any other singularities. In the above expression, the
homogeneous limit j = /2 can be taken trivially. In this case, surrounds the point /2.
Note that here to replace the sums over the partition of the set { } with the set of contour
integrals around { }, we have used the following relation:
n mn

f a+ , b F { + }
{ }={ + }{ } a=1 b=1

| + |=n

n
n m
dzj
1
n!
j =1
2i
n

a=1 b=1, b=a sh(za zb )
F {z} ,
f (za , b ) n n
sh(z
z
+
)
a
b
a=1 b=1
a=1 b=1
n
(4.15)
where F ({z}) is assumed to be a symmetric function of n variables {z}, and to be analytic in the
vicinities of zj = k . As for the sum over the partition of the Bethe roots {} (2.16) characterizing
the ground state, we have used the following relation, which is valid in the thermodynamic limit:

n

+ 1

1
n
tot j , { } F {} ,
F
{
}
=
d
lim
(4.16)
n
M M
n!
+
{}={ }{ }
|+ |=n
j =1
464
where F ({}) is assumed to be a symmetric function of n variables {}, and to be zero when any
two of its variables are the same.
The multiple integral (4.12) representing the generating function of the densitydensity correlation function is exactly the same as that for the longitudinal spinspin correlation function [7].
The densitydensity correlation function n1 nm+1 can be obtained by taking the derivative with
respect to and m as in (4.3).
As for
function exp(Q
1,m )nm+1 in (4.4), the numerator can be written
the generating
m
D)( )D(
(A
e
)|
/
as g | m
a
m+1
g
a=1
a=1 (a ). Applying Corollary 4.1, and then using the second relation in (3.3), one can obtain a determinant form of the generating function.
By the same method as in the derivation of (4.12), the following multiple integral representing
exp(Q1,m )nm+1 can be derived in the thermodynamic limit.
Proposition 4.3. The ground state expectation value of the operator exp(Q1,m )nm+1 in the
thermodynamic limit is written as

exp(Q1,m )nm+1

m
m
n
n

dzj
1
sh(za b + ) sh(a b )
n+1
=
d
2
2i
sh(za b ) sh(a b + )
(n!)
n=0
j =1
n

a=1 b=1
n

sh(a m+1 )
sh(n+1 za + )
Wn {} {z}
sh(za m+1 )
sh(n+1 a + )
a=1
a=1

detn M j k {} {z} detn+1 (j , zk ), . . . , (j , zn ), (j , m+1 ) .
(4.17)
Here the integration contours C and are taken as in Proposition 4.2. In this representation, the
homogeneous limit j = /2 can be trivially taken.
z
)/2) [7], as exNote that the above expression agrees with that for exp(Q1,m (1 m+1
pected.
(4.1).
Finally we would like to consider the equal-time one particle Greens function c1 cm+1
Because the operators B and C enter into the numerator in (4.1), the computation is more
complicated than that for the generating functions of the densitydensity correlation function.
Nevertheless, applying Proposition 4.1 and then utilizing the third relation in (3.3), one finds the
.
method used above is still applicable for the evaluation of c1 cm+1
in the thermodynamic
Proposition 4.4. The equal-time one-particle Greens function c1 cm+1
limit can be represented as a multiple integral as
n+1 m
n+1

m1

dzj

1
a=1 b=1 sh(za b + )
n+1
d dn+2
c1 cm+1 =
n+1 m+1
n!(n + 1)!
2i
a=1 b=2 sh(za b )
n=0
j =1
n+1
n m+1
[sh(n+1 za + ) sh(n+2 za )]
a=1 b=2 sh(a b )
na=1
n m
sh(
+
)
[sh(
a
b
n+1 a + ) sh(n+2 a )]
a=1 b=1
a=1
W n ({}|{z})
detn+1 M j k
sh(n+1 n+2 + )

detn+2 (j , z1 ), . . . , (j , zn+1 ), (j , m+1 ) ,
(4.18)
465
Fig. 1. The integration contours C and C for 1 < 1 (a) and for > 1 (b). The contour C C encircles the
singularities {j }m
j =1 and does not contain any other singularities.
where the function W n ({}|{z}) and the (n + 1) (n + 1) matrix are respectively defined as
n n+1

b=1 sh(a zb + ) sh(zb a + )

Wn {} {z} = n a=1
(4.19)
,
n+1 n+1
n
a=1 b=1 sh(a b + ) a=1 b=1 sh(za zb + )
and

M j k =

t (zk , j ) + t (j , zk ) na=1
t (zk , 1 ) for j = n + 1.
sh(a j +)
sh(j a +)
n+1
sh(j zb +)
b=1 sh(zb j +)
for j n,
(4.20)
The integration contour C for the variables {j }n+1

j =1 is taken as in Proposition 4.2, while
the contour C for n+2 should be taken such that the contour C C surrounds the points
n+2 = z1 , . . . , zn+1 , m+1 which are simple poles of (n+2 , zk ) and (n+2 , m+1 ), and does
not contain any other singularities. The homogeneous limit j /2 can be taken trivially in
the above expression. Correspondingly the contours and C C encircle the point /2.
In Fig. 1, typical examples of the integration contour C are depicted both for the regimes
1 < 1 and > 1. Recalling that the Fermi point c at zero chemical potential (c = 0)
is c = for 1 < 1 and c = i/2 for > 1 (see Section 2), one finds that the
contributions from the right and left edges of C disappear for 1 < 1, and those from
the upper and the lower edges cancel each other for > 1. Using this together with the relation
(+, z) = (, z) which is valid only for c = 0, one arrives at the following representation.
Corollary 4.2. The equal-time one-particle Greens function for the homogeneous case at zero
chemical potential can be expressed as

=
c1 cm+1
m1

n=0
1
n!(n + 1)!
n+1

a=1
n+1
n+1
dzj
sh(za + ) m
n+2
2
d
2i
sh(za 2 )
j =1
a=1

sh(a 2 ) m n+1
[sh(n+1 za + ) sh(n+2 za )]
na=1
sh(a + 2 )
a=1 [sh(n+1 a + ) sh(n+2 a )]
466
W n ({}|{z})
detn+1 M j k
sh(n+1 n+2 )

detn+2 (j , z1 ), . . . , (j , zn+1 ), j ,
,
2
(4.21)
where M j k is defined as (4.20) for j n and M j k = t (zk , 2 ) for j = n + 1.
Compared with the transverse spinspin correlation function 1+ m+1

in [7],3 one finds the
sign in front of the second term of M j k for j n changes.
5. Free fermion model

Using the multiple integral representations derived in the preceding section, we reproduce the
exact expressions of correlation functions for the free fermion model.
Let us consider the densitydensity correlation function n1 nm+1 first. As mentioned before,
the generating functions exp(Q1,m ) (4.12) for n1 nm+1 is the same as that for the spinspin
z
for the XXZ chain. Hence the same method described in [19] can
correlation functions 1z m+1
be applied to derive the exact expression of n1 nm+1 . Here we derive it by using (4.17) and (4.4).
Set = i/2. Then M j k (4.14) in (4.17) is factorized as M j k = 2(e 1)/ sh(2(j zk )),
which significantly simplifies the integral representation. After taking the derivative of (4.17)
with respect to and setting = 0, one finds all the terms except for n = 1 vanish. Substituting
(, z) = i/( sh(2( z))), and taking the lattice derivative, one obtains

i
sh(1 2 ) m (z) m (1 )
2
n1 nm+1 =
(5.1)
dz
d
,
sh(1 z) sh(2 z) ch(21 ) ch(22 )
2 3
C
where
() =
sh( 4 i)
.
sh( + 4 i)
(5.2)
The integral on can be evaluated by considering the residues outside the contour , i.e., at the
points z = 1 and 2 . It immediately follows that

1
1 m (2 ) m (1 )
d2
n1 nm+1 = 2
(5.3)
.
ch(21 ) ch(22 )
Changing the variables ch(2) = 1/ cos k (k [kF , 2 kF ]), where kF [/2, ] is the
Fermi momentum determined by cos kF = 1/ ch c (see Section 2), and identifying d =
dk/(2 cos k), one arrives at
1
n1 nm+1 =
4 2
2k
F
2k
F
dk1
kF

dk2 1 eim(k1 k2 )
kF
1 cos(2kF m)
= nj 2
,
2 2 m2

kF
, ,
2
(5.4)
3 There exists a typo in Eq. (6.13) in [7] for the transverse spinspin correlation function + . It should be
1 m+1
corrected by multiplying by 1.
467
where nj = 1 kF / is the particle density given by (2.23). This expression agrees with the
well-known result. At zero chemical potential c = 0, where the Fermi momentum corresponds
to kF = /2, one finds
n1 nm+1 =
1 1 (1)m
.
4
2 2 m2
(5.5)
Next let us discuss the one-particle Greens function c1 cm+1

(4.18). Since Mj k = 0 for
j n, only the term corresponding to n = 0 survives in (4.18). Namely

ch(1 z) ch(2 z) m (z)
1
dz
d
d
c1 cm+1 =
1
2
sh(z + 4 i) sh(z 4 i) ch(1 2 )
2 3

1
1
.
sh(2(1 z)) sh(2(2 + 4 i)) sh(2(2 z)) sh(2(1 + 4 i))
(5.6)
To evaluate the above integral explicitly, we shift the 2 -contour C to C. Since the integrand is
antisymmetric with respect to the exchange of 1 and 2 , the resulting integral vanishes: we only
have to take into account the poles surrounded by the 2 -contour C C i.e. 2 = i/4, z.
After evaluating the integral with respect to z by the method used in the calculation for the
densitydensity correlation function, one has

1
m ()
c1 cm+1 =
(5.7)
d
.
2
sh( + 4 i) sh( 4 i)
C
Changing the variables as explained before, one finally obtains

=
c1 cm+1
1
2
2k
F
dk e
imk
sin(kF m)
=
,
m
kF
, .
2
(5.8)
kF
This also coincides with the well-known result. In particular at zero chemical potential c = 0
(kF = /2), one sees

sin( 2 m)
.
c1 cm+1 = (1)m
m
(5.9)
6. Summary and discussion

In this paper, correlation functions for the spinless fermion model have been discussed by
using the fermionic R-operator and the algebraic Bethe ansatz. Applying solutions of the inverse
scattering problem, we derived determinant representations for form factors of local fermion
operators. In addition, multiple integrals representing the densitydensity correlation function
and the equal-time one-particle Greens function were obtained both for arbitrary interaction
strengths and particle densities. In particular for the free fermion model, these formulae reduce
to the known exact results.
Before closing this paper, we would like to remark possible generalizations of our formulae.
A generalization to the finite temperature case is of importance. In fact, the quantum transfer
matrix, which is a powerful tool for study of finite-temperature properties for strongly correlated systems in one-dimension, has already been provided for the spinless fermion model [20].
468
Because the algebraic relations among the elements of monodromy operators in the quantum
transfer matrix approach are completely the same as (2.12), the relations derived in this paper are
applicable without essential changes.
It will also be very interesting to compute the spectral function for the spinless fermion model.
Considering several excited states and inserting the corresponding Bethe roots into the determinant representations for the form factors (3.12) and the scalar product (3.5), one can explicitly
compute the spectral function for a finite system. The method developed in the study of the dynamical structure factors for the XXZ chain (see [2126] for example) will also be useful in
computations of the spectral function of the spinless fermion model. Comparison with the arguments in [27] is also interesting.
Acknowledgements
The authors would like to thank H. Boos, F. Ghmann, A. Klmper and M. Shiroishi for
fruitful discussions. This work is partially supported by Grants-in-Aid for Young Scientists (B)
No. 17740248, Scientific Research (B) No. 18340112 and (C) No. 18540341 from the Ministry
of Education, Culture, Sports, Science and Technology of Japan.
References
[1] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation Functions,
Cambridge Univ. Press, Cambridge, 1993.
[2] M. Takahashi, Thermodynamics of One-Dimensional Solvable Models, Cambridge Univ. Press, Cambridge, 1999.
[3] M. Jimbo, K. Miki, T. Miwa, A. Nakayashiki, Phys. Lett. A 168 (1992) 256.
[4] M. Jimbo, T. Miwa, Algebraic Analysis of Solvable Lattice Models, American Mathematical Society, Providence,
RI, 1995.
[5] M. Jimbo, T. Miwa, J. Phys. A 29 (1996) 2923.
[6] N. Kitanine, J.M. Maillet, V. Terras, Nucl. Phys. B 567 (2000) 554.
[7] N. Kitanine, J.M. Maillet, N.A. Slavnov, V. Terras, Nucl. Phys. B 641 (2002) 487.
[9] N. Kitanine, J.M. Maillet, N.A. Slavnov, V. Terras, hep-th/0505006.
[11] F. Ghmann, A. Klmper, A. Seel, J. Phys. A 37 (2004) 7625.
[12] F. Ghmann, A. Klmper, A. Seel, J. Phys. A 38 (2005) 1833.
[13] F. Ghmann, N.P. Hasenclever, A. Seel, J. Stat. Mech. 0510 (2005) P10015.
[14] K. Sakai, J. Phys. A 40 (2007) 7523.
[15] Y. Umeno, M. Shiroishi, M. Wadati, J. Phys. Soc. Jpn. 67 (1998) 1930.
[16] F. Ghmann, V.E. Korepin, J. Phys. A 33 (2000) 1199.
[17] N.A. Slavnov, Theor. Math. Phys. 79 (1989) 1605.
[18] N. Kitanine, J.M. Maillet, V. Terras, Nucl. Phys. B 554 (1999) 647.
[20] K. Sakai, M. Shiroishi, J. Suzuki, Y. Umeno, Phys. Rev. B 60 (1999) 5186.
[21] D. Biegel, M. Karbach, G. Mller, Europhys. Lett. 59 (2002) 882.
[22] J. Sato, M. Shiroishi, M. Takahashi, J. Phys. Soc. Jpn. 73 (2004) 3008.
[23] J.-S. Caux, R. Hagemans, J.-M. Maillet, J. Stat. Mech. (2005) P09003.
[24] J.-S. Caux, J.M. Maillet, Phys. Rev. Lett. 95 (2005) 077201.
[25] R.G. Pereira, J. Sirker, J.-S. Caux, R. Hagemans, J.M. Maillet, S.R. White, I. Affleck, Phys. Rev. Lett. 96 (2006)
257202.
[26] J.-S. Caux, R. Hagemans, J. Stat. Mech. 0612 (2006) P12013.
[27] M. Khodas, M. Pustilnik, A. Kamenev, L.I. Glazman, Phys. Rev. B 76 (2007) 155402.

On conformal, SL(4, R) and Sp(8, R) symmetries

of massless fields
M.A. Vasiliev
I.E. Tamm Department of Theoretical Physics, Lebedev Physical Institute,
Leninsky prospect 53, 119991 Moscow, Russia
Abstract
The sp(8, R) invariant formulation of free field equations of massless fields of all spins in AdS4 available
previously in terms of gauge invariant field strengths is extended to gauge potentials. As a by-product,
free field equations for a massless gauge field are shown to possess both su(2, 2) o(4, 2) and sl(4, R)
o(3, 3) symmetry. The proposed formulation is well-defined in the AdS4 background but experiences certain
degeneracy in the flat limit that does not allow conformal invariant field equations for spin s > 1 gauge
fields in Minkowski space. The basis model involves the doubled set of fields of all spins. It is manifestly
invariant under U (1) electricmagnetic duality extended to higher spins. Reduction to a single massless field
contains the equations that relate its electric and magnetic potentials which are mixed by the conformal
transformations for s > 1. We use the unfolded formulation approach recalled in the paper with some
emphasis on the role of ChevalleyEilenberg cohomology of a Lie algebra g in g-invariant field equations.
This method makes it easy to guess a form of the 4d sp(8, R) invariant massless field equations and then to
extend them to the ten-dimensional sp(8, R) invariant spacetime. Dynamical content of the field equations
is analyzed in terms of cohomology.
PACS: 11.15.-q; 11.25.Hf; 11.25.Tq; 11.30.Ly
Keywords: Higher-spin gauge theory; Conformal symmetry
E-mail address: vasiliev@lpi.ru.

doi:10.1016/j.nuclphysb.2007.10.017
470
M.A. Vasiliev / Nuclear Physics B 793 [FS] (2008) 469526
1. Introduction
It was argued by Fronsdal in the pioneering work [1] that the tower of free higher-spin
(HS) fields of all spins in four dimensions admits sp(8, R) symmetry. The conformal symmetry su(2, 2), that acts individually on a massless field of a fixed spin, extends to sp(8, R) that
mixes states of different spins. Based on this observation Fronsdal addressed two fundamental
questions: what is a minimal space where sp(8, R) is geometrically realized and what are manifestly sp(8, R) invariant field equations that describe 4d massless fields of all spins?
He answered the first question by showing that the relevant space is ten-dimensional with
real symmetric matrices X AB = X BA as local coordinates [1] (A, B, . . . = 1, . . . , 4 are 4d Majorana spinor indices). This is Lagrangian Grassmannian M4 Sp(8, R)/P where P is the
parabolic subgroup of Sp(8, R) that results from crossing out the right node of the Dynkin diagram of sp(8, R)
. Using two-component spinor notation, A = (, ),
, , . . . = 1, 2, , , . . . = 1, 2, the ten-dimensional matrix space extends Minkowski coordi

nates x to X AB = (X , X , X ) where X and X are additional six coordinates
that form an antisymmetric Lorentz tensor. Note that the relevance of this space to the description of massless fields was rediscovered by Bandos et al. in [2]. More generally MM denotes the
Lagrangian Grassmannian with local coordinates X AB = X BA , A, B = 1, . . . , M (in this paper
M(M+1)
we do not distinguish between the parabolic space and its big cell R 2 ).
The form of the dynamical variables and field equations in M4 was obtained later in [3] where
it was shown that the tower of all 4d massless integer and half-integer spins can be described,
respectively, by a single scalar C(X) and spinor CA (X) in M4 that satisfy the field equations

2
2
(1.1)
C(X) = 0,
X AB X CD X CB X AD

(1.2)
CC (X)
CA (X) = 0.
AB
X
X CB
These equations possess no gauge symmetry because the fields C(X) and CA (X) describe gauge
invariant objects like scalar (spin 0), Maxwell field strength (spin 1), Weyl tensor (spin 2) and
their HS generalizations.
The infinite towers of fields that appear in the Sp(8, R) invariant consideration are precisely
the HS multiplets of the 4d nonlinear HS gauge theory [4,5]. This is not accidental because the
original argument of Fronsdal in favor of Sp(8, R) was based on the prominent FlatoFronsdal
theorem [6] stating that the tensor product of two singletons, where Sp(8, R) acts in a natural
way, is equivalent to the set of massless fields of all spins as a sp(4, R) o(3, 2)-module. On the
other hand, the states of the HS gauge theories can also be understood as resulting from tensoring
singletons [7]. (Recall that singletons are conformal scalar and spinor fields in three dimensions.)
The fields C(X) and CA (X) can be interpreted as hyperfields in the hyperspace M4 that
allow to describe all 4d massless fields at once. M4 plays for a HS multiplet a role analogous to
that of superspace for supersymmetric theories. The concise form of Eqs. (1.1) and (1.2) makes
it tempting to look for a formulation of the full nonlinear theory in this formalism. Note that, as
mentioned in [3], the 4d nonlinear HS models of [4,5] can indeed be interpreted as possessing a
spontaneously broken Sp(8, R) symmetry.
Theories in MM have been studied in a number of papers from different perspectives [821].
In particular, an attempt to formulate a nonlinear theory in this framework was undertaken in [15].
One reason why it was hard to go beyond free theory is that C(X) and CA (X) describe gauge
471
invariant curvature tensors. It was not clear so far how to describe gauge field potentials like spin
two metric tensor in the sp(8, R) covariant way and in M4 . As gauge potentials play the key
role in any nonlinear field theory including YangMills theory, Einstein gravity, supergravity and
nonlinear HS gauge theories [2225,4,5,26,27] (see also [2830] for reviews of nonlinear HS theories and more references), to proceed towards a nonlinear HS theory in M4 one has to introduce
the gauge potentials in the Sp(8, R) invariant framework. This is the primary aim of this paper.
For the first sight, apart from being interesting, the problem may look unsolvable. Indeed, at
any rate the project is to find an sp(8, R) invariant formulation of free massless fields described in
terms of gauge potentials. Since sp(8, R) contains the conformal symmetry su(2, 2), this should
result in a conformal invariant formulation of free massless field equations in terms of potentials.
There is a lot of studies of conformal field equations in the literature starting from the seminal
work of Dirac [31] (see, e.g., [3238] and references therein). It is known however that the free
field equations in terms of potentials are not conformal for spins s > 1 in flat space. Actually, the
complete list of free conformal invariant equations in flat space available in [38] does not contain
the 4d massless equations in terms of gauge potentials except for the 4d Maxwell equations, i.e.,
spin one.
On the other hand this looks unnatural because the equations in terms of gauge invariant field
strengths are conformal invariant and the space of states of 4d massless equations admits the
action of the conformal symmetry. Aiming at preserving the sp(8, R) symmetry, we have to find
out what goes wrong with conformal symmetry in terms of potentials.
The results of this paper show that the sp(8, R) covariant formulation is consistent in the
AdS4 background but experiences certain degeneracy in the flat limit to Minkowski space, in
which either the form of the equations breaks down leading to (anti)selfdual equations (insisting
on the symmetry) or the symmetry transformation law on the fields in Minkowski space blows up
(insisting on the full massless equations). The same happens with the conformal symmetry that
can be defined in AdS4 but not in Minkowski space. Note that this is not the first time when the
AdS curvature resolves a no-go statement. Analogous phenomenon occurs for HS interactions in
HS gauge theories [25].
Our model exhibits manifest electricmagnetic (EM) duality symmetry extended to all HS
fields as a u(1) subalgebra of sp(8, R). A closely related property is that it contains two sets
of fields of all spins related by the EM duality symmetry. This doubling also plays a role in
the conformal transformations that mix the two species of gauge fields. The reduction to the
undoubled set of fields in which every massless field of spin s > 0 appears in one copy is also
possible. In this case, the field equations relate electric and magnetic potentials of spins s 1.
An interesting feature of this dynamical system is that the conformal transformations mix electric
and magnetic potentials for s > 1.
Apart from the conformal embedding of the AdS4 symmetry sp(4, R) o(3, 2) su(2, 2)
o(4, 2) sp(8) a different embedding sp(4, R) o(3, 2) sl(4, R) o(3, 3) sp(8, R) exists.
This simple observation has a surprising output that the gauge theories in AdS4 exhibit sl(4, R)
o(3, 3) symmetry at the free field level even for a single massless field of a fixed spin. This raises
an intriguing question whether the sl(4, R) extends to nonlinear 4d models, including both HS
theories and lower spin (super)gravity-like theories.
The organization of the rest of the paper is as follows. Sections 29 remind the reader some
known facts about HS field equations and unfolded dynamics approach extensively used throughout this paper. Namely, in Section 2 we recall the Sp(8, R) invariant formulation of the dynamics
of massless fields in terms of gauge invariant field strengths. In Section 3 the unfolded formulation of the field equations of HS gauge fields in AdS4 is summarized. The flat limit that reproduces
472
the standard on-shell formulation of HS dynamics in Minkowski space is discussed in Section 4.

In Section 5 we summarize relevant elements of the unfolded dynamics approach with some emphasize on the role of ChevalleyEilenberg cohomology. General strategy of searching unfolded
formulation of a g-symmetric field-theoretical model is outlined in Section 6. In Section 7 we
interpret Minkowski and AdSd geometries in terms of flat connections of o(d 1, 2). In Section 8
we extend this analysis to Sp(8, R), focusing main attention on the group manifold Sp(4, R) as
a ten-dimensional generalization of AdS4 . In Section 9 we introduce the star-product formalism
underlying the unfolded formulation of the HS dynamics and recall the pure gauge representation
of the flat Sp(2M) connection found in [10].
The original part of the paper starts in Section 10 where the Fock modules appropriate for
the Sp(8, R) invariant description of HS gauge fields are introduced. Conformal invariant unfolded field equations for massless fields of all spins are analyzed in Section 11, where we prove
the formal consistency of the conformal field equations, analyze their dynamical content, global
symmetries and specificities of the flat limit. In particular we show in this section that the proposed field equations are invariant under the EM duality transformation and that the conformal
symmetry cannot be preserved in Minkowski space because the special conformal part of the
field transformation blows up in the flat limit. In Section 12 these results are extended to sp(8, R)
while the gl(4, R) symmetry of the equations is considered in Section 13. The detailed study of
the dynamical content of the proposed equations within the cohomology approach is done
first for the 4d case in Section 14 and then for the case of matrix space in Section 15. Sections 14
and 15 can be skipped by the reader not interested in details of the formalism. Conclusions and
perspectives are discussed in Section 16.
R) symmetry
2. 4d massless fields and Sp(8,R
The key observation is [39,40] (see also [41]) that the generating function
C(b|x) =
=

1
CA ...A (x)bA1 bAk
k! 1 k
k=0
n,m=0
1

C ... , ... (x)b1 bn b 1 b m
n!m! 1 n 1 m
can be used to describe all 4d massless fields by virtue of the equation

2
+
C(b|x) = 0,
x
b b
(2.1)
where x are Hermitian coordinates of Minkowski spacetime (x = n x n , n = 0, 1, 2, 3,

n are four 2 2 Hermitian matrices) and b , b are auxiliary commuting spinor variables.
The interpretation of the components C1 ...n , 1 ... m (x) is as follows. Those that carry both
primed and unprimed indices (i.e., mn = 0) are expressed by (2.1) via spacetime derivatives
of the holomorphic and antiholomorphic components that carry only primed or only unprimed
indices, respectively, i.e.,
C1 ...m+k , 1 ... m (x)
= (1)m 1 1 m m Cm+1 ...m+k (x),
=
,
x
(2.2)
473
where the indices k and (independently) k are symmetrized. The formula in the conjugated
antiholomorphic sector is analogous.
The holomorphic and antiholomorphic fields describe a scalar field C(x) for s = 0, selfdual
and anti-selfdual components of the spin one Maxwell field strength, C (x) and C , (x), selfdual and anti-selfdual components of the Weyl tensor for spin two, C1 ...4 (x) and C 1 ... 4 (x),
and so on. The system (2.1) decomposes into an infinite set of subsystems for the fields of definite

helicities according to different eigenvalues of the helicity operator H = 12 (b b b ).
b
Apart from expressing auxiliary fields in terms of spacetime derivatives of the (anti)holomorphic fields via (2.2), Eq. (2.1) imposes the massless field equations on the latter
C(b, 0|x) = 0,
b[ x ]
=0
C(0, b|x)

b [ x ]
(2.3)
and
2C(0, 0|x) = 0.
(2.4)
It is easy to check that (2.2), (2.3) and (2.4) exhaust the content of (2.1).
Eq. (2.1) can be interpreted as the covariant constancy condition

D C(b|x) = 0
(2.5)
for the field

C(b|x) = C(b|x)|0,
(2.6)
that takes values in the Fock module generated from the vacuum state
aA |0 = 0
(2.7)
of the algebra of oscillators

B
aA , bB = A
,
[aA , aB ] = 0,

bA , bB = 0.
(2.8)
The generators of sp(8, R) are realized in this module as

1
PAB = aA aB ,
2
LA B =
1
a A b B + b B aA ,
2
1
K AB = bA bB .
2
(2.9)
Note that LA B form gl(4|R).

The covariant derivative
D = d + dx P
(2.10)
is a particular flat sp(8, R) connection, i.e.

D = d + w,
w = hAB PAB + B A LA B + fAB K AB ,
D 2 = 0.
(2.11)
The connection (2.10) is flat because the generators P commute to themselves. This choice of
the flat connection corresponds to Cartesian coordinates in Minkowski space.
As explained in Section 5.1, that the 4d massless equations (2.1) have the form of a covariant
constancy condition with the covariant derivative in a sp(8, R)-module V (here V is the space
of functions of b) implies their invariance under the global sp(8, R) symmetry. The generators
of sp(8, R) act on the dynamical fields as differential operators with coefficients polynomial in x
(see [3] and Section 9 for explicit field transformations).
474
The conformal symmetry su(2, 2) extended to u(2, 2) by the helicity generator is the subalgebra of sp(8, R) spanned by the generators

1
1

L = a , b ,
L = a , b ,
P = a a ,
2
2

K = b b
(2.12)
with the respective gauge fields h , , , f . The dilatation and helicity generators D
and H are
1
D = L + L ,
(2.13)
2
1
H = L L .
(2.14)
2
The helicity operator H is central in u(2, 2). This is expected because it takes a fixed value
on any su(2, 2)-module with definite helicity. H is the generator of EM duality transformations
which is the manifest symmetry of this formulation [3].
The extension of the dynamical equations to M4 is achieved by replacing the unfolded equation (2.1) with

2
+ A B C(b|X) = 0,
(2.15)
X AB
b b
where X AB = X BA are symmetric matrix coordinates associated with the generalized momentum PAB . This is the equivalent extension of the unfolded 4d massless equations to M4 . Indeed,

the part of Eq. (2.15) with A = , B = and X = x is just Eq. (2.1) while the equations

that contain extra six coordinates X and X reconstruct the dependence on these coordinates
in terms of the generating function of 4d massless fields C(b|x). (See also Section 5.5.)
On the other hand, one can interpret Eq. (2.15) differently by observing that they express all
components of C(b|X) that contain two or more oscillators bA via derivatives over the hyperspace coordinates X AB of the two dynamical fields which are polynomials of zeroth and first
degree of bA . Thus C(X) + CA (X)bA are dynamical fields in M4 while all other components
are auxiliary fields expressed by (2.15) via X-derivatives of the dynamical fields. Namely, for
C(b|X) =

1
CA ...A (X)bA1 bAn ,
n! 1 n
n=0
,
(2.16)
X AB
where [ n2 ] is the integer part of n2 and CAn2[ n ] (X) is either C(X) or CA (X).
2
Eqs. (1.1) and (1.2) are consequences of (2.15). Eqs. (1.1), (1.2) and (2.16) exhaust all restrictions on C(b|X) imposed by (2.15). This proves that Eqs. (1.1) and (1.2) in M4 are equivalent
to the 4d massless field equations for all spins. That all 4d massless fields are described by

only two hyperfields is because spin is carried by the spinning coordinates X and X in the
hyperspace M4 (see also [16]).
CA1 ...An (X) = (1)[ 2 ] A1 A2 A2[ n ]1 A2[ n ] CAn2[ n ] (X),
n
AB =
3. Higher spin gauge fields in AdS4

In this section we recall the unfolded form of 4d free HS field equations proposed in [39,40].
It is based on the frame-like approach to HS gauge fields [42,43] where a spin s HS gauge field
475
is described by the set of 1-forms

1 ...k , 1 ... l = dx n n1 ...k , 1 ... l ,
k + l = 2(s 1).
The HS gauge fields are self-conjugated 1 ...k , 1 ... l = 1 ...l , 1 ... k . This set is equivalent
to the real 1-form A1 ...A2(s1) symmetric in the Majorana indices A, that carries an irreducible
module of the AdS4 symmetry algebra sp(4, R) o(3, 2).

The AdS4 space is described by the Lorentz connection , and vierbein e . Altogether they form the sp(4, R) connection w AB = w BA that satisfies the sp(4, R) zero curvature
conditions
R AB = 0,
R AB = dw AB + w AC wC B ,
(3.1)
where indices are raised and lowered by a sp(4, R) invariant form CAB = CBA
AB = AA CAB ,
AA = C AB AB ,
B
CAC C BC = A
.
In terms of Lorentz components w AB = ( ,

the AdS4 equations (3.1) read as
R = 0,
R = 0,

, e
(3.2)
, e
) where 1 is the AdS4 radius,
R = 0,
(3.3)
where
R = d + + 2 e e ,
R = d +

+ e
R = de + e +

(3.4)
e ,
e .
(3.5)
(Two-component indices are raised and lowered as in (3.2) with CAB replaced by the twocomponent symplectic forms
or
.)
The unfolded equations of motion of a spin-s massless field read as

0 2s1 2s
H
C1 ...2s ,
R1 ...n , 1 ... m = n0 H 2s1 2s C 1 ... 2s + m
n + m = 2(s 1)
(3.6)
and
D tw C1 ...n , 1 ... m = 0,
D tw C 1 ...n , 1 ... m = 0,
n m = 2s,
m n = 2s.
(3.7)
Here the HS field strength and twisted adjoint covariant derivative have the form
R1 ...n , 1 ... m = D L 1 ...n , 1 ... m + ne1 m+1 2 ...n , 1 ... m+1

+ men +1 1 1 ...n+1 , 2 ... m ,
D tw C1 ...n , 1 ... m = D L C1 ...n , 1 ... m
+ (e C1 ...n , 1 ... m + nme1 1 C2 ...n ,2 ... m ),

D tw C 1 ...n , 1 ... m = D L C 1 ...n , 1 ... m

+ (e C ...n ,
1

1 ... m
+ nme1 1 C 2 ...n ,2 ... m ),
(3.8)
(3.9)
476
where the indices and are (separately) symmetrized and D L is the Lorentz covariant derivative
D L = d + ,

D L = d + .

H = H and H = H are the basis 2-forms built of the vierbein 1-form e

H = e e ,

H = e e .
(3.10)
(3.11)
Formulae simplify in terms of the generating functions

A(y, y|x)

n,m=0
1

y . . . yn y 1 y m A1 n , 1 ... m (x)
n!m! 1
(3.12)
R, etc. In particular, we have

with A = , C, C,
R(y, y|x)
= D ad (y, y|x)

= D (y, y|x)
e
y + y (y, y|x),
y
y

2

D tw C(y, y|x)
= D L C(y, y|x)
+ e y y + C(y, y|x),
y y
L
where the Lorentz covariant derivative D L takes the form

D A(y, y|x)
= dA(y, y|x)
y +
y
A(y, y|x).
y
y
(3.13)
(3.14)
(3.15)
As a consequence of the zero curvature equation (3.1) which is true for AdS4 geometry, the
covariant derivatives D ad and D tw are flat, i.e.,
ad
2 tw
2
D
= D
= 0.
These conditions are necessary for the consistency of Eqs. (3.6) and (3.7) (i.e., the compatibility
with d 2 = 0) and guarantee the gauge invariance of the field strength (3.13) and, therefore, the
free HS field equations (3.6) under Abelian HS gauge transformations
(y, y|x)
= D ad
(y, y|x).
(3.16)
It is important that the consistency of the equations is not spoiled by the C-dependent terms in
(3.6). As explained in more detail in Section 5.4, this means that these terms correspond to a
ChevalleyEilenberg cohomology of sp(4, R).

In Eqs. (3.6) and (3.7), a spin s field is described by the set of gauge 1-forms 1 ...n , 1 ... m (x)

with n + m = 2(s 1) (for s 1) and 0-forms C 1 ...n , 1 ... m(x) with n m = 2s along with

their conjugates C 1 ...n , 1 ... m (x) with m n = 2s. Indeed it is easy to see that the field equations (3.6) and (3.7) for such a set of fields with some s form an independent subsystem.
The dynamical massless fields are
C(x) and C(x)

for two spin zero fields,
C (x) and C (x) for a massless spin 1/2 field,
1 ...s1 , 1 ... s1 (x) for an integer spin s 1 massless field,
1 ...s3/2 , 1 ... s1/2 (x) and its complex conjugate 1 ...s1/2 , 1 ... s3/2 (x) for a halfinteger spin s 3/2 massless field.
477
All other fields are auxiliary, being expressed via derivatives of the dynamical massless fields by
Eqs. (3.6) and (3.7).
The key fact of the unfolded form of free massless field equations is the so-called Central
On-Shell Theorem [40] that states that the content of Eqs. (3.6) and (3.7) is just that they express
all auxiliary fields in terms of derivatives of the dynamical fields and impose the massless field
equations on the latter in the standard form of Fronsdal [44] and Fang and Fronsdal [45]. To
make the paper as self-contained as possible we sketch the proof of Central On-Shell Theorem
in Section 14 using the -cohomology technics.
The meaning of Eqs. (3.6) and (3.7) is as follows. Eqs. (3.7) provide the AdS4 deformation of
(2.1). They remain independent for spins s = 0 and s = 1/2 and partially independent for spin
one but become consequences of (3.6) for s > 1. Eqs. (3.6) express the holomorphic and antiholomorphic components of spin s 1 0-forms C(y, y|x)
via derivatives of the massless field gauge

1-forms described by (y, y|x).
This identifies the spin s 1 holomorphic and antiholomorphic

components of the 0-forms C(y, y|x)
with the Maxwell tensor, on-shell RaritaSchwinger curvature, Weyl tensor and their HS generalizations. In addition, Eqs. (3.6) impose the standard field
equations on the spin s > 1 massless gauge fields so that the field equations (2.3) become their
consequences by virtue of Bianchi identities. The dynamical equations for spins s 1 are still
contained in Eqs. (3.7). (For more detail see e.g. [28] and also Section 14.)
Although the system (3.6) and (3.7) is consistent at the free field level, to extend it to the nonlinear case one has to double the set of HS fields [7,40,4,5].1 This can be achieved by introducing
the fields
ii (y, y|x),
C i1i (y, y|x),
i = 0, 1
such that ii (y, y|x)
are selfconjugated, while C 01 (y, y|x)
and C 10 (y, y | x) are conjugated to

one another,
ii (y, y|x)
= ii (y,
y|x),
C i1i (y, y|x)
= C 1ii (y,
y|x).
The unfolded system for the doubled set of fields is

= H
R ii (y, y|x)
2
2
1ii
C
(0,
y|x)
+
H
C i1i (y, 0|x),

y y
y y
D tw C i1i (y, y|x)

= 0.
(3.17)
(3.18)
Note that now all components of the expansions of C i1i (y, y|x)
contribute to Eqs. (3.17) and

(3.18), while in (3.6) and (3.7) with the single HS 1-form (y, y)
only parts of the components
of C(y, y)
and C(y,
y)
contributed.
In the standard formulation of the 4d nonlinear HS gauge theory [5] (see [28] for a review)
the doubling of the fields is due to the dependence on the Klein operators k and k that flip
chirality [47]
k 2 = 1,
k 2 = 1,
ky = y k,
= y k,
ky
k y = y k,
k y = y k,
= 0.
[k, k]
1 Note that, as discussed in [7,40] (see also [28]), the full N = 2 supersymmetric nonlinear HS system can be truncated
to subsystems with reduced sets of fields. In particular, truncating out fermions, it is possible to consider a system with
bosonic fields of all spins in which every integer spin appears once and further to the minimal system considered in some
detail in [46], in which every even spin appears just once.
478
The fields are 1-forms
y, y|x)
(k, k;
=
j ij (y, y|x)
(k)i (k)
ij =0,1
and 0-forms
y, y|x)
C(k, k;
j C ij (y, y|x).
(k)i (k)
ij =0,1
Now both the adjoint and twisted adjoint covariant derivative result from different sectors of the
adjoint covariant derivative in the Weyl algebra extended by the Klein operators [47].
Massless fields are those with
y, y|x)
y, y|x),
(k, k;
= (k, k;
C(k, k; y, y|x)
= C(k, k; y, y|x).
The fields with the opposite oddness in the Klein operators are topological, carrying at most a
finite number of degrees of freedom per an irreducible subsystem [48]. We will see in Section 11
how this pattern of HS fields emerges in the sp(8, R) invariant formulation. In particular the
topological field sector also plays a role in the model we focus on in this paper.
4. Flat limit
To take the flat limit it is necessary to perform certain rescalings. To this end let us intro
duce notations A and A0 so that the spectrum of the operator (y y y ) is positive on
y
A+ (y, y|x),
negative on A (y, y|x)
and zero on A0 (y, y|x).
Having the decomposition

A(y, y|x)
= A+ (y, y|x)
+ A (y, y|x)
+ A0 (y, y|x),
(4.1)
we introduce a new field

y|x)
A(y,
= A+ (y, y|x)
+ A (y, y|x)
+ A0 (y, y|x).
(4.2)
(Note that A0 (y, y|x)
= A0 (y, y|x).)
For the rescaled variables, the flat limit 0 of the

adjoint and twisted adjoint covariant derivatives (3.13) and (3.14) gives
y|x)
y|x)
Dflad A(y,
= D L A(y,

+ y A + (y, y|x)
,
e y A (y, y|x)
y
y

y|x)
y|x)
Dfltw A(y,
= D L A(y,
+ e
A(y, y|x).
y y
(4.3)
(4.4)
The flat limit of the unfolded massless equations results from (3.17) and (3.18) via the substitution of D L and e of Minkowski space and the replacement of D ad and D tw by Dflad and Dfltw ,
respectively (recall that R ii = D ad ii ). The resulting field equations describe free HS fields in
Minkowski space. Let us stress that the flat limit prescription (4.2), that may look somewhat
unnatural in the two-component spinor notation, is designed just to give rise to the theory of
Fronsdal [44] and Fang and Fronsdal [45] (for more detail see Section 14).
479
Note that, although the contraction 0 with the rescaling (4.2) is consistent with the free
HS field equations, it turns out to be inconsistent in the nonlinear HS theory because negative
powers of survive in the full nonlinear equations upon the rescaling (4.2), not allowing the
flat limit in the nonlinear theory. This is why the Minkowski background is unreachable in the
non-linear HS gauge theories of [25,4,26]. The reason why conformal symmetry blows up in
the flat limit is that the translation generators in the sector of HS gauge fields , that respect the
described limiting procedure, disagree with their standard identification in the conformal algebra
(for more detail see Section 11.3).
The flat limit of the equation D tw C = 0 just reproduces Eq. (2.1) which underlies the original
extension of the HS dynamics from four to ten dimensions. The main question addressed in this
paper is how to include the HS gauge 1-forms into the manifestly sp(8, R) invariant formalism
and then into the ten-dimensional formulation. This is most naturally achieved within the unfolded formulation. In fact, the 4d Eqs. (3.17) and (3.18) do have the unfolded form. To proceed,
we now summarize relevant properties of the unfolded formulation using the 4d HS system as
the basis example.
5. Unfolded dynamics
5.1. Unfolded equations
Let M d be a d-dimensional manifold with coordinates x n (n = 0, 1, . . . , d 1). (For d = 4

we use the Hermitian coordinates x .) By unfolded formulation of a linear or nonlinear system
of differential equations and/or constraints2 in M d we mean its equivalent reformulation in the
first-order form3

dW (x) = G W (x) ,
(5.1)
where d = dx n x n is the exterior differential on M d , W (x) is a set of degree p -differential
forms and G (W ) is some degree p + 1 function of the differential forms W
G (W ) =
f 1 ...n W 1 W n ,
n=1
where the coefficients f 1 ...n satisfy the (anti)symmetry condition f 1 ...k k+1 ...n =
(1)pk+1 pk f 1 ...k+1 k ...n (extension to the supersymmetric case with an additional boson
fermion grading is straightforward) and G satisfies the condition
G (W )
(5.2)
=0
W
(the derivative W is left) equivalent to the generalized Jacobi identity on the structure coefficients
m

(n + 1)f [1 ...mn f mn+1 ...m } = 0,
(5.3)
G (W )
n=0
2 By constraints we mean equations like dA = B that express auxiliary fields like B in terms of derivatives of other
fields like A imposing no differential equations on the latter.

3 The idea of this approach was suggested and applied to the analysis of interacting HS gauge theory in [39,40] while
the name unfolded formulation was given somewhat later in [49].
480
where the brackets [} denote appropriate (anti)symmetrization of indices i . Strictly speaking,

formal consistency demands (5.3) to be satisfied only at p < d for a d-dimensional manifold
Md where any d + 1-form is zero. Given solution of (5.3) it defines a free differential algebra
(FDA) [5053]. We call a free differential algebra universal if the generalized Jacobi identity
is true independently of d. The HS FDAs that appear in HS gauge theories and, in particular,
those discussed in this paper belong to the universal class. Unfolded formulation is a covariant
multidimensional extension of the first-order reformulation of ordinary differential equations.
Universal FDAs have the distinguished property that the operator W is well-defined irre(W )
spectively of on what it acts, i.e. F
is defined for any F (W ) built of wedge products of
W
differential forms. For non-universal FDAs this is not true. Actually, the condition
W1 Wk = 0,
p1 + + pk > d
(5.4)
may lead to a contradiction upon formal differentiation. For instance, differentiating (5.4) over
all Wi involved leads to the contradiction 1 = 0. In other words, for non-universal FDAs the
space of W is constrained by the relations (5.4). Correspondingly, only vector fields tangent to
the constraint surface are allowed.
For universal FDAs, Eq. (5.1) is invariant under the gauge transformation
W (x) = d (x) + (x)
G (W (x))
,
W (x)
(5.5)
where the gauge parameter (x) is an arbitrary (p 1)-form (0-forms W (x) do not give
rise to gauge symmetries). This property of universal FDAs makes the unfolded formulation
an efficient tool for the study of gauge invariant dynamical systems. Since unfolded equations
are formulated in terms of the exterior algebra, this approach respects diffeomorphisms thus
providing a natural framework for the study of models that contain gravity.
5.2. Vacuum
An important class of universal FDAs is in the one-to-one correspondence with Lie algebras.
Indeed, let w be a set of 1-forms. If no other forms are involved (e.g., all of them are consistently
set equal to zero in a larger system) the most general expression for G (w), that has to be a
w w . The consistency condition (5.3) then becomes the Jacobi
2-form, is G (w) = 12 f
of a Lie algebra g. The unfolded equations (5.1) impose
identity for the structure coefficients f
the flatness condition on the connection w
1
dw + f
w w = 0.
(5.6)
2
The transformation law (5.5) gives the usual gauge transformation of the connection w
w (x) (x).
w (x) = D (x) = d (x) + f
(5.7)
A flat connection w(x) is invariant under the global transformations with the covariantly constant parameters
D (x) = 0.
(5.8)
This equation is consistent by (5.6). Therefore, locally, it reconstructs (x) in terms of its values (x0 ) at any given point x0 . (x0 ) are the moduli of the global symmetry g that is now
recognized as the stability algebra of a given flat connection w(x).
481
This example is of key importance because this is how g-invariant vacuum fields appear in
the unfolded formulation. In particular, Eq. (3.1) of AdS4 spacetime is of this type. The same
happens for the general case. Typically, an unfolded system that contains 1-forms w associated
with some Lie algebra g admits a flat connection w as its natural g-symmetric vacuum solution.
In the perturbative analysis, w is assumed to be of the zeroth order because it contains the
background metric that is usually non-degenerate, being of order zero. The flat connection w is
then referred to as vacuum connection.
Let us stress that this way of description of the background geometry is coordinate independent. A particular form of w (x) is not needed for the analysis unless one is interested in explicit
solutions in a specific coordinate system. The only important condition is that a flat vacuum
connection has to be chosen so that its part associated with the generators of translations in g

(i.e., vielbein sometimes also called soldering form, which is the 1-form e in the discussion
of Section 3) is non-degenerate. The ambiguity of the choice of a particular vacuum form w is
up to the gauge transformations (5.7). In particular, the coordinate choice ambiguity, which of
course preserves the flatness property, is reproduced by the gauge transformations (5.7). In fact,
this is the general property of the unfolded dynamics where a diffeomorphism generated by an
infinitesimal vector field n (x) can be realized as the field-dependent gauge transformation (5.5)
with the gauge parameter of the form
(x) = n (x)
W (dx, x).
dx n
(5.9)
5.3. Free fields and ChevalleyEilenberg cohomology

Let us now linearize the unfolded equation (5.1) around some vacuum flat connection w of a
Lie algebra g, that solves (5.1). To this end we set
W = w + ,
(5.10)
are differential forms of various degrees that are treated as small perturbations and
where
enter the equations linearly. Consider first the sector of forms i (x) of a given degree p (e.g.,
0-forms) within the set (x). Then Gi is bilinear in w and , i.e., Gi = w (T )i j j . In
this case the condition (5.2) implies that the matrices (T )i j form a representation T of g in a
vector space V where i (x) takes its values (index i). The corresponding equation (5.1) is the
covariant constancy condition
Dw i = 0
(5.11)
with Dw d + w being the covariant derivative in the g-module V .

Eqs. (5.6) and (5.11) are invariant under the gauge transformations (5.5) with
i (x) = (x)(T )i j j (x).
(5.12)
Once the vacuum connection is fixed, the system (5.11) is invariant under the global symmetry g
with the parameters satisfying (5.8). This simple analysis allows useful applications.
Firstly, one observes that, by unfolding, any g-invariant linear dynamical system turns out to
be reformulated in terms of g-modules. This allows for full classification and explicit derivation
of g-invariant equations by studying various g-modules. In particular, the full list of conformal
invariant equations in flat spacetime of any dimension has been obtained this way in [38]. In
this paper we will derive the manifestly sp(8, R) covariant form of the massless field equations
in terms of gauge potentials just by guessing appropriate sp(8, R)-modules.
482
Secondly, it follows that if g is a larger Lie algebra that acts in V , g g End V , it is also
a symmetry of (5.11) simply because any flat g-connection is the same time a flat g-connection.
As a result, g = End V is the maximal symmetry of (5.11) (of course, modulo possible subtleties
in the infinite-dimensional case).
Thirdly, this gives an efficient tool for the derivation of the explicit form of the symmetry
transformation laws via (5.8) and (5.12).
For example, once Eq. (2.1) is re-understood in the form (2.5), (2.10), (2.11), from the general
analysis it follows that it is sp(8, R) invariant. To derive the explicit form of the global sp(8, R)
transformation we solve the covariant constancy condition for the global symmetry parameter
(x)

gl (a, b|x) = exp [x a a ]

0 (a, b) exp x a a ,
1
1
0 (a, b) =
AB bA bB +
A B bA aB +
AB aA aB ,
2
2

where bA = (b , b ) and aA = (a , a ). This gives
1
1

gl (a, b|x) =
AB bA bB +
A B bA aB +
AB aA aB x
0 B a aB x
0 B a aB
2
2

+ x
bD a
D x bD a
D x .
The desired transformation law is then obtained by restricting the variation

C(b|x) =
gl (a, b|x)C(b|x)
(5.13)
and using (2.2) for the auxiliary fields

to the dynamical holomorphic fields C(b, 0|x) or C(0, b|x)
that appear on the r.h.s. of (5.13) to derive terms with x-derivatives of the dynamical fields in the
transformation law. Note that this way it is possible to derive explicit form of the transformation
law of the full HS algebra in a far more complicated AdS case (for more detail see [3,10] and
Section 9).
Suppose now that a (x) and i (x) are forms of different degrees, say, pa pi = k > 0. Then,
in the linearized approximation, one can consider functions G polylinear in the vacuum field
w but still linear in the dynamical fields
Ga (w, ) = fa1 ...k+1 i w 1 w k+1 i .
(5.14)
(Note that the case of k = 1 with Ga (w, ) (5.14) independent of w is also possible. It
corresponds to the so-called contractible FDAs [50] and is dynamically empty because the corresponding unfolded equation just expresses a pi -form i via the lower degree forms and their
derivatives. As such it is not considered in this paper.)
Let i be a 0-form. The equation for i is then always a covariant constancy condition (5.11).
The consistency condition (5.2) applied to (5.14) then literally implies that fa1 ...k+1 i w 1
w k+1 is a ChevalleyEilenberg cocycle of g with coefficients in Vl Vr where Vl is the module where Ga takes values while Vr is the module conjugated to that of i . Coboundaries are
dynamically empty because, as is easy to see, they can be removed by a field redefinition. Thus,
in the unfolded formulation, the ChevalleyEilenberg cohomology classifies possible nontrivial
mixings of higher form fields with 0-forms. This type of mixing is of most importance in the
context of known HS theories.
More generally, one can imagine the equations with the terms of the type of (5.14) that involve
p > 0 forms on the r.h.s. The consistency condition then is that the cohomology algebra with
483
respect to the natural product of the elements f (V1 , V2 ) and f(V1 , V2 ) with V2 = V1 is nilpotent
in the sense that (5.2) is true.
The free HS system (3.1), (3.6) and (3.7) has unfolded form. It is consistent in the sense
of (5.2). The terms on the r.h.s. of Eqs. (3.6) describe the ChevalleyEilenberg cohomology of
sp(4, R) with coefficients in the corresponding infinite-dimensional modules. Let us note that
without these cohomological terms, i.e., relaxing the r.h.s. of Eq. (3.6), the sector of 1-forms
would become dynamically trivial (any solution of the zero curvature equation is pure gauge in
the topologically trivial situation). The gluing with 0-forms via (3.6) and (3.7) makes the 1-form
gauge potential dynamically nontrivial and, the same time, expresses the 0-forms in terms of
derivatives of the gauge potentials (except for spin zero and spin 1/2 fields that have no associated
gauge potentials because of the second derivatives over y and y on the r.h.s. of (3.17)).
Note that the statement that g extends to a larger symmetry algebra g that acts in the gmodules i , which is true in the 0-form sector, may not be true in the other sectors in presence
of the cohomological terms because it is not a priory guaranteed that a g-cohomology extends to
a g-cohomology.
To large extent our analysis of the sp(8, R) symmetry in Section 12 amounts to

the analysis of the extension of the sp(4, R) HS cohomology to sp(8, R).
It is important that in presence of the cohomological terms (5.14) the system remains invariant
under the global symmetry g. Indeed, the system (5.6) along with
d = G (w)
(5.15)
is formally consistent and therefore gauge invariant. The global symmetry g is still the part
of the gauge symmetry that leaves invariant the vacuum fields w . Its action on is deformed
however by the cohomological terms (5.14) according to (5.5). In addition the system is invariant
under the Abelian gauge transformations (5.5) associated with p > 0 form fields among .
For example, the system (3.1), (3.6) and (3.7) is invariant under the HS gauge transformations
(3.16) associated with the 1-form connections (y, y|x),
which are Abelian gauge transformations for free massless fields of spins s 1 whose form depends on the vacuum AdS fields

w AB = ( , , e ) that enter via the flat covariant derivative D ad . Also the system (3.1),
(3.6) and (3.7) is invariant under the global sp(4, R) symmetry that leaves invariant the AdS4
vacuum fields w AB . The formula (5.5) applied to Eq. (3.6) gives the sp(4, R) transformation law
gl (y, y|x)

=
(x)y +
(x)y +
(x) y + y (y, y|x)
y
y
y
y
+ 2
(x)e
2
2
1ii

C
(0,
y|x)
+
2
(x)e
C i1i (y, 0|x),

y y
y y
(5.16)
gl C(y, y|x)

(x) y y +
=
(x)y +
(x)y
C(y, y|x),
+
y
y
y y
(5.17)

where the parameters of global sp(4, R) transformations

(x),
(x) and
(x) satisfy the
sp(4, R) covariant constancy conditions (5.8). The C-dependent terms in (5.16) generalize to HS
fields the well-known formula that represents diffeomorphisms in gravity as the deformation of
the o(d 1, 2) (or Poincar) gauge transformations by the Riemann tensor dependent term (see
484
e.g. [54]). Thus, the unfolded form of field equations reproduces necessary deformation terms
automatically via the ChevalleyEilenberg cohomology deformation.
Finally, let us note that the condition that the mixture of differential forms via Chevalley
Eilenberg cohomology is perturbatively nontrivial, i.e., that the terms (5.14) do not accidentally
vanish because of unlucky choice of the vacuum connection w, may impose an additional restriction on the latter. We shall see below that it is this condition that gives preference to AdS4
geometry in the models of interest because the relevant terms (5.14) turn out to be non-degenerate
in the AdS4 case but trivialize in the Minkowski case.
5.4. Dynamical content via cohomology
In the unfolded dynamics approach, dynamical fields (i.e., those that are neither pure gauge
nor auxiliary (see footnote 2)), their differential gauge symmetries (i.e., those that are not Stueckelberg (= shift) symmetries) and differential field equations (i.e., those that are not constraints),
are characterized by the so-called cohomology [55] (see also [3,29]). The aim of this subsection is to recall briefly the main idea of this method to make it possible to explain in Section 6
the general strategy of the search and investigation of the sp(8, R) invariant equations. To keep it
short, the consideration of this subsection is general and formal. It will be applied to the analysis
of the unfolded HS field equations first in Section 14, explaining in some detail the dynamical
content of the 4d unfolded HS equations and, in particular, the Central On-Shell Theorem, and
then in Section 15 to the case of M4 . Sections 14 and 15 will provide examples clarifying the
formal scheme sketched in this subsection.
-cohomology is a perturbative concept that emerges in the linearized analysis. Eq. (5.1)
linearized by (5.10) gives
D (x) = 0,
(5.18)
where D is some differential built of the order zero flat vacuum connection w . To fulfill the
consistency condition (5.2), D should square to zero
D2 = 0.
(5.19)
Generally, D is the generalized covariant derivative that, in addition to usual connection-like

terms linear in the vacuum connection w , contains all ChevalleyEilenberg-type terms polylinear in the w . For a p -form with p 1, the linearized gauge transformation is
= D
,
(5.20)
where
is a (p 1)-form gauge parameter.
For a meaningful dynamical interpretation of Eq. (5.18), a space V , where fields take
their values, should be endowed with some grading G such that its spectrum is bounded from
below. Typically G counts a rank of a tensor (equivalently, a power of an appropriate generating
polynomial) and eventually is associated with the order of spacetime derivatives of dynamical
fields. Suppose that
(x)
D = D 0 + + + ,
(5.21)
where [G, ] = , [G, D0 ] = 0 and + is a sum of some operators of positive grade. From
(5.19) it follows that 2 = 0. Provided that acts vertically (i.e., does not differentiate x n ), cohomology of determines the dynamical content of the dynamical system at hand. Namely, as
485
shown in [55], for a p-form that takes values in a vector space V , H p+1 ( , V ), H p ( , V )
and H p1 ( , V ) characterize, respectively, differential equations, dynamical fields and differential gauge symmetries encoded by Eq. (5.18).4
The meaning of this statement is quite simple. From the level-by-level analysis of Eqs. (5.18)
and (5.21) it follows that all fields that do not belong to Ker are auxiliary, being expressed by
(5.18) via derivatives of the lower grade fields. Those that are exact can be gauged away by the
Stueckelberg part of the gauge transformation (5.20), associated with the part of D in (5.21).
The fields that remain belong to the cohomology of . These are dynamical fields. Analogously
one analyzes the dynamical content of the gauge transformations and field equations. (For more
detail see e.g. [29].)
Usually originates from the part of the covariant derivative of a spacetime symmetry
algebra, that contains vielbein required to be non-degenerate thus providing a frame of 1-forms
at any point x0 of spacetime. The non-degeneracy of the vielbein allows to express as many as
possible auxiliary fields via spacetime derivatives of the dynamical fields.
The role of the ChevalleyEilenberg cohomology terms in the unfolded equations is that they
fill in unwanted cohomologies with auxiliary Weyl tensor-like variables to avoid too strong
differential consequences of the unfolded system. For example, relaxing the C-dependent terms
on the r.h.s. of (3.17) would imply that all HS gauge 1-forms are pure gauge, which condition is
too strong to describe nontrivial dynamics.
In principle, the leftover cohomology responsible for the HS field equations can also be
glued with the additional 0-form fields. This corresponds to the situation with H p+1 ( , V ) = 0
where no differential equations on the dynamical variables are imposed at all. In this case, the
unfolded equation (5.1) just expresses the Bianchi identities for some constraints on auxiliary
fields. Unfolded systems of this type are referred to as off-shell. As discussed in [56], they are
useful for the Lagrangian formulation of a dynamical system.
Let us stress that the cohomology analysis applies both to linear and to nonlinear systems
treated perturbatively. In particular, nonlinear equations are off-shell once their linearization is
off-shell.
5.5. Properties
Let us summarize briefly the main properties of unfolded dynamics. First of all, the method
is universal in the sense that any dynamical system can in principle be unfolded. This statement is analogous to the text-book fact that any system of ordinary differential equations can be
reformulated in the first-order form.
Indeed, let w = e0a Pa + 12 0ab Mab be a vacuum gravitational gauge field taking values in some
spacetime symmetry algebra g. Let C (0) (x) be a given spacetime field satisfying some dynamical equations to be unfolded. Consider for simplicity the case where C (0) (x) is a 0-form. The
general procedure of unfolding goes schematically as follows. For a start, one writes the equation
(1)
(1)
D0L C (0) = e0a Ca , where D0L is the covariant Lorentz derivative and the field Ca is auxiliary.
Next, one checks whether the original field equations for C (0) impose any restrictions on the first
L C (0) might vanish on-mass-shell (e.g. for
derivatives of C (0) . More precisely, some part of D0m
(1)
Dirac spinors). These restrictions in turn impose some restrictions on the auxiliary fields Ca . If
(1)
these constraints are satisfied by Ca , then these fields parameterize all on-mass-shell nontrivial
4 Let us note that higher H p ( , V ) describe so-called syzygies of the field equations.
486
components of first derivatives. One continues by writing analogous equation for the first-level
(1)
(2)
(2)
auxiliary fields D0L Ca = e0b Ca,b , where the new fields Ca,b parameterize the second derivatives
of C (0) . Once again one checks (taking into account the Bianchi identities) which components of
(2)
the second level fields Ca,b are non-vanishing provided that the original equations of motion are
satisfied. This process continues indefinitely, leading to a chain of equations having the form of
some covariant constancy condition for the chain of fields Ca(m)
1 ,a2 ,...,am (m N) parameterizing all
on-mass-shell nontrivial derivatives of the original dynamical field. By construction, this leads to
(m)
a particular unfolded equation. The set of fields Ca1 ,a2 ,...,am realizes some g-module V . The full
infinite chain of equations becomes a single covariant constancy condition D0 C = 0, where D0
is the g-covariant derivative in V .
If one starts with some gauge field like, for example, the fluctuational part of the metric tensor,
analogous analysis determines a form of the Stueckelberg shift gauge transformations that subtract Stueckelberg field components to be introduced to describe a system in terms of differential
forms. (For instance in gravity, the local Lorentz symmetry results this way as the Stueckelberg
symmetry that removes the extra components of the vielbein 1-form compared to the metric tensor.) The correspondence between p 1 forms and gauge symmetries in the unfolded dynamics
approach uncovers the pattern of local and global symmetries associated with a given gauge
field. In particular, the pattern of the linearized 4d HS algebras was deduced this way in [43].
These results were then used in [57,58,47] to find infinite-dimensional non-Abelian HS algebras
that underly the nonlinear 4d HS theories and in [40,4,5] to construct full nonlinear HS field
equations as a nonlinear deformation of the unfolded system (3.17) and (3.18).
Other important properties of the unfolded formulation include:
Manifest gauge invariance and invariance under diffeomorphisms (i.e., coordinate independence) due to using the exterior algebra formalism is perfectly suited for the study of gauge
invariant theories in the framework of gravity and, in particular, HS gauge theories.
In the topologically trivial situation, degrees of freedom are concentrated in 0-forms 0i (x0 )
at any x = x0 . This is a consequence of the Poincar lemma: the unfolded equations express
all exterior derivatives in terms of the values of fields themselves modulo exact forms that
can be gauged away by the gauge transformation (5.5). What is left is the constant part of
the 0-forms.
This simple observation has a consequence that, to describe a system with an infinite number
of degrees of freedom, it is necessary to work with an infinite set of 0-forms that form an
infinite-dimensional module of the spacetime symmetry g. In fact, the module carried by
0-forms turns out to be dual (complex equivalent) to the space of single-particle states in the
respective QFT.
On the other hand, if the unfolded formulation of a system operates with a finite set of 0forms, the system is topological, describing at most a finite number of degrees of freedom.
In particular, the topological dynamical systems of [48] mentioned in the end of Section 3
are just of this type. A typical example of such a system is Eq. (5.8) on the global symmetry
parameters. In fact, the covariant constancy condition on the 0-forms in the topological sector
of Section 3 coincides with Eq. (5.8) on the HS global symmetry parameters.
Unfolded formulation admits natural realization of higher derivative infinite symmetries as
endomorphisms of the infinite-dimensional modules of 0-forms.
ChevalleyEilenberg cohomology of a Lie algebra g underlying the unfolded formulation of
one or another system is responsible for nontrivial mixture of differential forms of differ-
487
ent degrees, thus making gauge fields associated with (p > 0)-forms dynamically nontrivial.
Note that the corresponding cohomology turns out to be nontrivial even for simple Lie algebras associated with the spacetime symmetries like o(d, 2), sp(M, R), etc., just because it
has coefficients in infinite-dimensional g-modules.
Unfolded formulation unifies various dual versions of the same system. The difference
results from the ambiguity in what is chosen to be dynamical or auxiliary fields, the nomenclature governed by the choice of the grading G and . Different gradings lead to different
interpretations of the same unfolded system in terms of different dynamical fields that satisfy
seemingly unrelated differential equations. The key point is that if two dynamical systems
give rise to the same unfolded system, they are equivalent.5 We conjecture that all dual descriptions of a given dynamical system D are contained in its maximally extended projective
unfolded version P (D). By a P (D)-projective unfolded system we mean such a maximal
unfolded formulation of D that (i) any unfolded description of D is a subsystem of P (D) and
(ii) P (D) does not decompose into two independent subsystems one of which is an unfolded
formulation of D. Note that P (D) may require a larger set of differential form variables.
For example, the extension of the set of Weyl 0-forms C(y, y|x)
by the HS gauge 1-forms

(y, y|x)
leads to the unfolded system (3.17), (3.18) that extends the HS equations (2.3)
in terms C(y, y|x)
to those in terms of HS gauge potentials. The unfolded system (3.17),

(3.18) is not projective however. It can be further extended without changing its dynamical
content by replacing 1-forms ii and 0-forms C i1i by forms of all odd and even degrees,
respectively. The resulting system is likely to be projective for the doubled set of 4d massless
fields of all spins.
The concept of P (D)-projective unfolded system is homological in nature. General analysis
of this interesting and important issue lies beyond the scope of this paper and will be given
elsewhere.
One of the striking features of the unfolded formulation based on universal FDAs is that,
to some extent, it makes the spacetime x-dependence artificial. The dynamics is entirely
encoded in the form of the function Gi (W ). In particular, unfolded formulation allows one
to extend spacetime without changing dynamics simply by letting the differential d and
differential forms W to live in a larger space
d = dx n n + dx n n ,
x n
x
x
dx n wn dx n wn + d x n w n ,
d = dx n
where x n are some additional coordinates. For a universal unfolded system such a substitution neither spoils the consistency nor changes the local dynamics still determined by the
0-forms at any point of (any) spacetime. Alternatively, one observes that the unfolded system in the x space remains a subsystem of that in the enlarged space while the additional
equations reconstruct the dependence on the additional coordinates in terms of solutions of
the original system (of course, this consideration is local).
This property is not only practically useful allowing to introduce easily appropriate hyperspaces [3,60], but is likely to have deep meaning encouraging to reconsider a role of such
5 It is worth to note that, as shown in [59], differential duality relations between dual systems in flat space may
sometimes become algebraic in AdS geometry. This phenomenon is analogous to the AdS resolution of the flat space
degeneracy discussed below: some operators to be interpreted as components of that degenerate in flat background
may be non-degenerate in the AdS background.
488
fundamental concepts as local event and metric tensor in a fundamental theory. An illuminating manifestation of this issue comes from the analysis of 4d HS physics formulated in
M4 in [8], where it was argued that the concepts of local event, spacetime dimension and
metric tensor have dynamical origin. In this respect, the unfolded dynamics [40] has some
similarity with the matrix model approach to string theory [6163], having however the advantage of being covariant. (See also [64,65] where matrix models are linked to HS theory
in a somewhat different fashion.)
The unfolded formulation approach provides an efficient tool for the analysis of gauge invariant interactions. The problem reduces to the unification of the zero-order vacuum field
w and first-order dynamical gauge fields into a single field W by (5.10) and to searching
for a nontrivial deformation of G (W ) that respects the consistency condition (5.2) and reproduces correctly the linearized dynamics. In fact, the results of this paper form a basis for
the future search of nonlinear sp(8, R) invariant HS gauge theories.
Note that there is a great similarity between the unfolded formulation approach and the prolongation technics in the analysis of partial differential equations (see e.g., text book [66]) that
operates with jet spaces designed to describe higher derivatives to make it possible to rewrite a
partial differential system in the first order form. The important novelty is due to the extension
to differential forms as dynamical variables in the unfolded dynamics approach, that results in
the properties listed above. Note also that the unfolded formulation admits a nice interpretation
[56,67] in terms of L strong homotopy algebra [68].
6. Strategy
Equipped with the unfolded dynamics approach, the strategy of the analysis of free gsymmetric dynamical equations may be as follows.
Fix a flat (vacuum) connection w of g that gives a non-degenerate vielbein as the 1-form connection associated with the generators of translations in the respective spacetime symmetry
subalgebra s g (e.g., Poincar, (A)dSd , or conformal).
Guess a set of g-modules where the variables , that are differential forms of different
degrees, take their values and find ChevalleyEilenberg cohomology (5.14) of g with the coefficients in the respective g-modules. It is this step that fixes a particular dynamical system
encoded by the unfolded equations (5.18).
Introduce a grading G of g-modules where the take their values, that gives rise to the
decomposition (5.21) of the generalized covariant derivative. Since, the ambiguity in the
choice of G results in dual descriptions of the same model, to describe a system in terms
of some preferable dynamical variables one has to choose G appropriately. Analyze the
cohomology to figure out what are differential gauge symmetries, dynamical fields and
differential field equations encoded by the unfolded equations (5.15).
Let M be some s-invariant manifold and w be a vacuum connection of s with a nondegenerate vielbein. A larger symmetry g of an unfolded system at hand may admit no
geometric interpretation in M. To find an equivalent formulation of the same unfolded
dynamics in a larger spacetime M where g acts geometrically it suffices to extend the
exterior differential d from M to M and to replace w by a flat connection of g that contains
non-degenerate basis 1-forms (generalized vielbein) in M. Provided that the Chevalley
Eilenberg cohomology terms were defined with respect to g, the resulting unfolded equations
489
in M remain consistent and describe the same dynamics. Keeping the grading G unchanged,
one finds the new operator in M. The dynamical content (i.e., symmetries, field variables
and field equations) of the unfolded system in M is uncovered via the analysis of the cohomology of the new . Although the resulting dynamical variables and field equations in M
may differ from those of the original system in M (pretty much as superfields and superfield
equations of supersymmetric theories look differently from their component counterparts),
the resulting system in M is guaranteed to be g-invariant and (locally) equivalent to that
in M.
In the rest of the paper we systematically implement this approach. Although it still has some
research freedom in guessing appropriate g-modules, for distinguished systems like the one explored in this paper this part of the project is not too hard because the choice is usually quite
limited if not obvious. The benefit is that the g-symmetry is guaranteed and the rest of the analysis is straightforward. Most notably, the g-invariant differential equations are derived (rather than
guessed) via the analysis of cohomology and the equivalence of the equations in different
spaces is automatic.
7. Conformal geometry
To describe a conformal system in d dimensions in the unfolded dynamics approach one
should first of all fix a non-degenerate flat connection of the conformal algebra o(d, 2). In terms
of Lorentz (i.e., o(d 1, 1)) irreducible components, the o(d, 2) has the generators Pn , Lnm =
Lmn , D, K n . The o(d, 2) connection 1-form w and curvature 2-form R are
w = hn Pn + nm Lnm + fn K n + bD,
R = R n Pn + R nm Lnm + rn K n + rD,
(7.1)
where
R n = dhn + n m hm b hn ,
R nm = dnm + n k km hn f m + hm f n ,
r = db + hn fn ,
r n = df n + n m f m + b f n .
Here the 1-form hn = dx m hm n is identified with vielbein. It is required to be non-degenerate in
the sense that det |hm n | = 0. nm is Lorentz connection. fn and b are gauge fields for special
conformal transformations and dilatation, respectively.
The conformal gauge transformations are
hn = D L
n
n m hm +
hn
n b,
nm = D L
nm hn
m +
n f m + hm
n
m f n ,
b = d
+ hn
n
n fn ,
f n = D L
n
n m hm
f n +
n b,
where
n (x),
mn (x),
n (x) and
(x) are gauge parameters of translations, Lorentz transformations, special conformal transformations and dilatations, respectively.
490
The interpretation of these fields is as follows (for more detail see, e.g., [69,35]). The 1-form
b can be gauge fixed to zero
b=0
(7.2)
by a special conformal gauge transformation with the parameter

n (x). (Here one uses that hn is
non-degenerate.) The leftover gauge symmetries are local translations, Lorentz transformations
and dilatations.
Imposing the condition that the o(d, 2) curvatures are all zero
R=0
(7.3)
has the following consequences. R n = 0 in the gauge (7.2) is the usual zero torsion condition
that expresses the Lorentz connection nm in terms of the vielbein hn . The condition R nm = 0
requires the Weyl tensor C nm,kl (h) to be zero and expresses the symmetric part f(nm) of f n =
dx m fm n in terms of the Ricci tensor of h. The antisymmetric part of fm n is zero by virtue
of r = 0 in the gauge (7.2). rn = 0 holds by virtue of Bianchi identities. Thus, in terms of the
vielbein, (7.3) implies that Weyl tensor is zero. All other equations contained in (7.3) are either
constraints or consequences of the other equations.
The translation of these results into the cohomology language is as follows. The grading
G is just the conformal dimension in o(d, 2) induced by D, i.e., hn has grading 1, nm and
b have grading zero and fn has grading +1. is the h-dependent part the covariant derivative where hn is treated as the respective part of the vacuum connection. (hn = dx n in Cartesian
coordinates.) H 0 ( ) describes linearized diffeomorphisms (all other gauge transformations are
Stueckelberg). H 1 ( ) describes the conformal class of metrics (perturbatively, second rank
traceless tensors). H 2 ( ) describes the Weyl tensor (i.e., the only nontrivial differential equation in (7.3) is that the Weyl tensor is zero). We leave it to the reader to check details of this
correspondence as a useful exercise.
Taking into account that, locally, any two o(d, 2) flat connections are related by a o(d, 2)
gauge transformation and that o(d, 2) gauge transformations contain local dilatations of the metric, a simple consequence of this analysis is that the metric tensor is conformally flat iff the Weyl
tensor is zero.
In Cartesian coordinates, the Minkowski space solution of (7.3) is w = dx n Pn . Another important example of conformally flat space is provided by the anti-de Sitter geometry. Indeed, the
AdSd algebra o(d 1, 2) can be realized as the subalgebra of o(d, 2) spanned by the generators
Pn = Pn + 2 Kn and Lnm . Choosing a flat connection of o(d 1, 2), namely en and nm , gives
us a flat connection of o(d, 2) with hn = en , fn = en , b = 0 and the same Lorentz connection
nm . This Ansatz solves (7.3) for the conformal algebra once en and mn solve the zero curvature equations for o(d 1, 2). As a by-product, this gives a coordinate independent proof of the
fact that AdSd is conformally flat.
In the 4d case one can use two-component spinor notation. In these terms, the su(2, 2)

o(4, 2) connections are h , , , b and f . Extending su(2, 2) to u(2, 2) by adding a
the u(2, 2) flatness conditions read as
central helicity generator with the gauge connection b,
R = dh h h = 0,
R = df + f +

f = 0,

R = d + f h
R
= d
+
= 0,

f h
= 0.
(7.4)
(7.5)
(7.6)
(7.7)

491
Here the traceless parts L and L of and describe the Lorentz connection
while their traces contain the gauge fields b and b according to
1
1
+ , b = .
2
2
The AdS4 geometry is described by
b=
h = e ,
f = e ,
(7.8)
b = b = 0.
(7.9)

e
The u(2, 2) flatness conditions are solved by the vierbein

and Lorentz connection
and

that satisfy the zero curvature conditions (3.3). This Ansatz for the vacuum connection will
be used later on to describe 4d conformal invariant systems in the unfolded dynamics approach.
8. Generalized conformal geometry
The example of o(d, 2) conformal symmetry admits a natural generalization to sp(2M, R)
treated as generalized conformal symmetry with the group manifold Sp(M, R) as an analog of
AdSd . (Note that the case of M = 2 reproduces the usual 3d case with AdS3 Sp(2, R) and
sp(4, R) so(3, 2).) This interpretation of Sp(M, R) was discussed in [70,3,10,11].
The components of the sp(2M, R) gauge connection (2.11) are real. Generalized conformally
flat background geometry is described by a sp(2M, R) flat connection that satisfies
R AB = dhAB C A hCB C B hCA = 0,
(8.1)
RAB = dfAB + A fCB + B fCA = 0,
(8.2)
RA = dA + A C fAC h
(8.3)
CB
= 0.
The Cartesian coordinates in MM are associated with the particular flat connection
hAB = dX AB ,
A B = 0,
fAB = 0.
A different choice of the sp(2M, R) flat connection describes the group manifold Sp(M, R).
The group of motions Sp(M, R) Sp(M, R) of Sp(M, R) results from the left and right action
of the group on itself. Let two flat sp(M, R) connections satisfy
RAB
= dwAB
wA
wCB
= 0,
(8.4)
where indices are raised and lowered by a Sp(M, R) invariant symplectic form CAB = CBA
according to (3.2). Locally, they admit the pure gauge representation
(X) = UAC
(X) dUB C (X),
wAB
where UA B (X) is an arbitrary Sp(M, R) valued matrix that satisfies

UA C (X)UB D (X)CCD = CAB ,
UA C (X)UB D (X)C AB = C CD .
Setting
1
+
,
w wAB
2 AB
and using (8.4), we observe that
AB =
eAB =
1 +

w + wAB
2 AB
RAB = dAB + A C CB + 2 eA C eCB = 0,
(8.5)
492
rAB = deAB + A C eCB + B C eCA = 0.
(8.6)
To embed sp(M, R) sp(M, R) into sp(2M, R) we express the connections of sp(2M, R) in

terms of those of sp(M, R) sp(M, R) as follows
hAB = eAB ,
fAB = eAB ,
A B = A B .
(8.7)
Then the sp(2M, R) flatness conditions (8.1)(8.3) hold as a consequence of (8.5) and (8.6).
We will use the sp(2M, R) flat vacuum connection (8.7) in the analysis of the Sp(2M, R)
invariant field equations in Sp(M, R). The important difference between the Minkowski-like
geometries and their AdS-like counterparts is that in the former case the special conformal connections f n and f AB vanish while in the latter case they are non-degenerate by virtue of (7.9) and
(8.7). This property will be of crucial importance for the physical interpretation of the usual and
generalized HS conformal equations because they contain some f -dependent terms that should
be non-degenerate for a consistent perturbative interpretation.
9. Star-product and vacuum symmetry
Instead of working in terms of oscillators (2.8), it is convenient to use the star-product in the
algebra of polynomials of commuting variables aA and bA

1
A
A
(f g)(a, b) = 2M f (a + u, b + t)g(a + s, b + v)e2(sA t uA v )
d M u d M t d M s d M v.
(9.1)
The star-product defined this way, often called Moyal product, describes the product of symmetrized (i.e., Weyl ordered) polynomials of oscillators in terms of their symbols. The integral is
normalized so that 1 is the unit element of the algebra

1
A
A
e2(sA t uA v ) d M u d M t d M s d M v = 1.
2M
Eq. (9.1) defines the associative algebra with the defining relations

A B
a A , b B = A B ,
[aA , aB ] = 0,
b ,b = 0
([a, b] = a b b a). The following useful formulae hold
1
1
,
bA = bA
,
A
2 b
2 aA
1
1
,
bA = bA +
.
aA = aA
2 bA
2 aA
The star-product realization of the generators of sp(2M, R) is
aA = aA +
1
1
PAB = aA aB ,
K AB = bA bB .
2
2
sp(2M, R) extends to the superalgebra osp(1|2M, R) by adding the supergenerators
(9.2)
(9.3)
LA B = a A b B ,
(9.4)
QA = a A ,
(9.5)
S B = bB .
Using the oscillator realization of sp(M, R) sp(M, R) sp(2M, R) we can set

1
w(a, b|X) = B A (X)aA bB + eAB (X) a A a B + 2 bA bB ,
2
(9.6)
493
where AB and eAB satisfy the flatness conditions (8.5) and (8.6) to ensure that w(a, b|X) satisfies the vacuum flatness condition
dw + w w = 0.
(9.7)
= aA bC CCA with the commutation relations

Let us introduce the oscillators A

A , B = 2CAB ,
A , B = 0.

= 12 A
B are the generators of sp + (M, R) sp (M, R) sp(2M, R).
Then TAB
A useful viewpoint is that w(a, b|X) takes values in the infinite-dimensional star product
algebra of various polynomials of aA and bB , which is the HS symmetry algebra as a Lie superalgebra. (In accordance with the spin-statistics relationship, the bosonfermion Z2 grading
counts the oddness of a number of spinor indices, i.e., w(a, b|X) = (1)(w) w(a, b|X).)
As explained in Section 5.2, any fixed vacuum solution w0 of (9.7) breaks the local HS symmetry to its global stability subalgebra with the infinitesimal parameters
0 (a, b|X) that satisfy the
equation
d
0 + [w0 ,
0 ] = 0.
(9.8)
Solving (9.7) in the pure gauge form

w0 (a, b|X) = g 1 (a, b|X) dg(a, b|X),
where g(a, b|X) is some invertible element of the star-product algebra, g
we solve (9.8) as
0 (a, b|X) = g 1 (a, b|X) (a, b) g(a, b|X),
(9.9)
g 1
g 1
g = 1,
(9.10)
where an arbitrary X-independent star-product element (a, b) describes free parameters of the
global HS symmetry. Note that the explicit form of the HS transformations depends on a chosen
coordinate system encoded by w0 (a, b|X).
As shown in [10], the star product realization of the pure gauge representation (8.5) is given
by the formula
g(X) = g + (X) g (X) = g + (X)g (X),
where
M

22

B ,
g (X) =
exp 2f AB [U ]A
det U + 1
(9.11)
(g )

22

(X) =
B
exp 2f AB [U ]A
det U + 1
(9.12)
and we have set = 1 for simplicity. Here U AB (X) is some sp(M, R) valued function of local
coordinates and we use the notations

U 1 AB
1 + f AB
AB
AB
f [U ] =
,
U [f ] =
.
U +1
1f
Note that the gauge functions g (X) (9.11) are chosen so [10] that the corresponding flat
connection w is bilinear in the oscillators aA , bB , i.e., it indeed takes values in sp (M, R).
494
A particularly useful choice is that with U + = (U )1 = U , giving

g(X) =

2M
exp f AB [U ](aA aB + bA bB ) .
det U + 1
As noted in [10], the ambiguity in the function f AB (X) parameterizes the ambiguity in the
choice of local coordinates of Sp(M, R). (For particular coordinate choices see [10].)
Once the vacuum solution w0 is fixed in the pure gauge form (9.9) with some gauge function g, it is easy to find the gauge parameter
0 (a, b|X) of the leftover global symmetry. Indeed,
let the generating parameter (a, b; , ) in (9.10) be of the form = 0 exp(aA A bA A )
where 0 is an infinitesimal constant while A and A are constant parameters. Substitution of
(9.11) into (9.10) gives [10]
0 (a, b; , |X) = g 1 g = 0 exp aA A bA A ,

(9.13)
where

1 + 2 f 2 (X) B
2f (X)
B
A =
B
B ,
1 2 f 2 (X) A
1 2 f 2 (X) A

1 + 2 f 2 (X) B
2f (X)
B
2
B .
A =
B
1 2 f 2 (X) A
1 2 f 2 (X) A
Now, any HS global symmetry parameter, which is some x-dependent star-product polynomial,
can be obtained by differentiation of
0 (a, b; , |X) over A or/and A .
This simple procedure demonstrates the efficiency of the unfolded dynamics approach. Analogously to the flat space example (5.13), to obtain the form of the HS transformation of massless
fields in Sp(M, R) one has to act by the parameter
0 (a, b; , |X) on a module where the
fields take their values. The corresponding sp(8, R)-modules, which the same time form the
HS symmetry-modules, are introduced in the next section.
R) Fock modules
10. Sp(8,R
To extend the free sp(8, R) invariant equation (2.15) on HS Weyl field strengths to the sector
of gauge fields we have to identify a sp(8, R)-module where 4d HS gauge fields of Section 3
take their values. Since the HS Weyl 0-forms were described in Section 2 in terms of the Fock
module (2.6), (2.7) in which the sp(8, R) is realized by bilinears of oscillators, a natural option
is to use the same realization of sp(8, R), changing however the Fock module by changing its
vacuum.
Fock vacua projectors can be realized in terms of the star-product algebra. For example, with
the help of (9.2) and (9.3) one finds that the Fock vacuum |0, 0| defined by aA |0, 0| = |0, 0|
bA = 0 is realized as the exponential |0, 0| = 2M exp(2aA bA ), where the normalization factor
is chosen so that |0, 0| |0, 0| = |0, 0|. Alternatively, one can consider Fock vacua projectors
defined with respect to different sets of creation and annihilation operators. Demanding their
Lorentz invariance and definite scaling dimension with respect to the Lorentz generators in (2.12)
and the dilatation generator (2.13), respectively, we set

b |10| = |10| a = 0,
|10| = 4 exp 2a b :
(10.1)

|01| = 4 exp 2a b :
(10.2)
a |10| = |10| b = 0,

0|
= |1
0|
a = 0,
0|
= 4 exp 2a b : b |1
|1
(10.3)

1|
= 4 exp 2a b :
|0
495
0|
= |1
0|
b = 0.
a |1
(10.4)
We have
1|
= 16 exp 2aA bA :
|0, 0| = |01| |0
aA |0, 0| = |0, 0| bB = 0,

1|
= 16 exp 2 a b a b :
|1, 0| = |10| |0
(10.5)

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

0|
= 16 exp 2 a b a b :
|0, 1| = |01| |1
(10.6)

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

0|
= 16 exp 2aA bA :
|1, 1| = |10| |1
(10.7)
bB |1, 1| = |1, 1| aA = 0.
(10.8)
Correspondingly, we introduce two-component oscillators i , j and

respectively, the annihilation and creation operators of the vacuum |i, j |
0 = a ,
0
=b ,
1 = b ,
1
= a ,
0 = a ,

= b ,
0
i ,

j

1 = b ,
that are,
(10.9)

1 = a .
(10.10)
Note that
i = 1i ,

i = 1i .
(10.11)
|i , j |
It should be noted that the Fock vacua |i, j | cannot be star-multiplied with
in the class
of regular functions if i = i and/or j = j . This is not accidental. Indeed, if, say, T = |0, 0||1, 1|
existed, it would satisfy aA T = T aA = 0. Taking into account (9.2) and (9.3), from here it
follows that T = (a). (Of course, this can be directly derived from (9.1).) However T does not
belong to the star-product algebra because T T = (0).
On the other hand, different Fock spaces originating from |i, j | form well-defined modules
of the star-product algebra. This is sufficient for the analysis of the free HS dynamics of this
paper. To go beyond the free field level one has to handle potential difficulties of co-existence of
different sectors of the star-product algebra associated with the different Fock modules. We hope
to come back to this interesting question elsewhere.
Now we introduce four 0-form modules

Cij (i , j |X) = Cij (i , j |X) |i, j |
(10.12)
and four 1-form modules

ij (i , j |X) = ij (i , j |X) |i, j |,
ij (i , j |X) = dX U U ij (i , j |X),
XU
(10.13)
will be specified later on depending on the problem

where the meaning of the coordinates
under study. More precisely, |Cij (i , j |X) and |ij (i , j |X) are sections of the Fock fiber
bundles over a spacetime manifold with local coordinates X, that can be either a 4d spacetime
or one of the ten-dimensional spacetimes M4 or Sp(4, R).
The generating fields |Cij (i , j |X) and |ij (i , j |X) form sp(8, R)-modules with the
generators (9.4). This allows us to define the sp(8, R) covariant derivatives Dij in the module induced from the vacuum |i, j |. Note that |Cij (i , j |X) and |ij (i , j |X) also form modules
496
of su(2, 2) with the generators (2.13), (2.12), of osp(1|8, R) with the supergenerators (9.5) and
of the infinite-dimensional HS superalgebra whose generators are various (i.e., not only bilinear)
polynomials of aA and bA . Let us stress that, because the vacua |i, j | are Lorentz invariant and
have definite scaling dimensions, so defined sp(8, R)-modules consist of towers of Lorentz multispinor fields with definite scaling dimensions. (If vacua were not Lorentz invariant, the resulting
Lorentz algebra-modules could be infinite-dimensional.)
In our construction we postulate that the oscillators a and b are complex conjugated to

a and b , respectively. With this convention the conjugation , that singles out the real form
sp(8|R) of sp(8|C), acts as follows

LA B = LA B ,
K AB = K AB .
(PAB ) = PAB ,
From the definition of the Fock modules |i, j | it follows then that |i, j | = |j, i|, i = i ,

i = i . Correspondingly, the following reality conditions are imposed
ij (i , j |X) = j i (j , i |X),
Cij (i , j |X) = Cj i (j , i |X).
The original sp(8, R) invariant form of the HS Eqs. (2.15) is the covariant constancy condition
D00 C00 (b|X) = 0 which is the analogue of flat space limit of the 4d equation (3.18) in Cartesian
coordinates. Our aim is to extend Eqs. (3.17) and (3.18) first to su(2, 2) and then to sp(8, R)
symmetric formulations. To make it possible to use general properties of the unfolded formulation, this will be done for generic su(2, 2) and sp(8, R) flat connections in Sections 11 and 12,
respectively. As a result, the obtained systems will be proven to have global su(2, 2) and sp(8, R)
symmetries that act both on the 1-form HS gauge fields and on the 0-form field strengths.
11. Conformal invariant unfolded massless equations
11.1. Consistent equations
The u(2, 2) sp(8, R) covariant derivatives in the various Fock modules are defined by
D con |Cij (ij |X) where
D con = d + w,
w = h a a + a b + a b + f b b .
Here the traceless parts L and L of and , respectively, describe the Lorentz
connection while the traces describe the gauge fields b and b (7.8). For the generating functions
(10.12) this defines the covariant derivatives Dijcon ,

con

Dij Cij (i , j |X) |i, j | = D con Cij (i , j |X) ,
which have the form

1
1

con
= D L a
+ 1 + b + 1
D10
2
a
2
b

+ h a f
b ,
a
b

1
1

con
L
D01 = D + b
+ 1
+1
a
2
b
2
a

+ h a f
b ,
b
a
(11.1)
(11.2)

con
D00
497

1
1

= D + b
+ 1 + b
+1
2
b
2
b
L
2

+ f b b ,
b b

1
1

con
D11
= D L a
+ 1 a
+1
2
a
2
a
+ h
+ h a a + f
2
,
a a
(11.3)
(11.4)
where D L is the Lorentz covariant derivative (3.15).

From the form of covariant derivatives Dijcon it follows in particular that the operators of helicity H (2.14) and dilatation D (2.13) act on the respective modules ij = ij or ij = Cij as
follows

1

H10 (a, b) = a
(11.5)
+b
+ 2 10 (a, b),
2
a
b

1

=
+
a
+
2
01 (b, a),
b
H01 (b, a)
(11.6)
2
b
a

= 1 b b 00 (b, b),
H00 (b, b)
(11.7)
2
b
b

1
= a
a
11 (a, a),
H11 (a, a)
(11.8)
2
a
a

1

b
D10 (a, b) = a
(11.9)
10 (a, b),
2
a
b

1
=
a
b
01 (b, a),
D01 (b, a)
(11.10)
2
b
a

= 1 b + b + 2 00 (b, b),
D00 (b, b)
(11.11)
2
b
b

1
= a
+ a
+ 2 11 (a, a).
D11 (a, a)
(11.12)
2
a
a
Now we are in a position to write the conformal invariant unfolded system of equations
con
= h h
D10
10 (a, b)
2
2

C
(0,
b)
+
f
f
C11 (a, 0),
00

a a
b b
(11.13)
con
D01
01 (b, a)
= h h
2
2
f
C
(b,
0)
+
f
C11 (0, a),
00
b b
a a
con
= f f b b C10 (0, b)
+ f
D00
00 (b, b)

con
11 (a, a)
= h h a a C10 (0, a)
+ h
D11
Dijcon Cij = 0.
f b b C01 (b, 0),

h a a C01 (a, 0),
(11.14)
(11.15)
(11.16)
(11.17)
498
This system decomposes into two independent subsystems. One contains the 1-forms ii and
0-forms Ci1i while another one contains i1i and Cii . As will be explained in Section 11.2,
the subsystem (11.15), (11.16) and (11.17) with i = j is topological while the subsystem (11.13),
(11.14) and (11.17) with i + j = 1 describes massless fields of all spins.
The important property of the system (11.13)(11.17) is that it is consistent for any flat u(2, 2)
i.e., Eq. (11.13) along with
connection. Let us for definiteness consider the sector of 10 (a, b),
con
= 0,
C00 (b, b)
D00
con
D11
C11 (a, a)
= 0.
(11.18)
Because vacuum connections are such that Dijcon squares to zero, the proof of consistency is
con to the r.h.s. of (11.13) gives zero by virtue
equivalent to checking that the application of D10
of Eqs. (11.18). The analysis of the two terms on the r.h.s. of (11.13) is independent of each
other. Since they are exchanged by the Chevalley automorphism that exchanges translations and
special conformal transformations, we only consider the h-dependent term in (11.13), which is
con (11.1) are proportional to a while
not a coboundary because the h-dependent terms in D10
the h-dependent terms on the r.h.s. of (11.13) are a-independent. The proof of cocyclicity is
elementary and consists of the following observations:
Since the whole setup is Lorentz covariant, the L , L -dependent terms in the consistency
conditions cancel out.
con contains derivative over a of
The f -dependent terms cancel out because the one in D10
con C (b, b)
the a -independent expression on the r.h.s. of (11.13) while the other one in D00
00
disappears upon setting b = 0.
The h3 -terms vanish because
3
(11.19)

=0
b b b
as a result of antisymmetrization of the three two-component indices , , .

con and D con and cancel each other while the terms
The terms with are the same in D10
00

that result from the differentiation of h h by virtue of (7.4) compensate those that
2

result from the commutator of the covariant derivative with (equivalently, the L
h h h
b b
weight of h h compensates that of ).

b b
Finally, the terms with also cancel out. Namely the differential parts trivialize either
con ) or setting b = 0 (D con ). The constant terms,
because of differentiating a constant (D1,0
0,0

which are different, then exactly cancel the result of differentiation of h h . (Equiva

lently, the L weight of h h is compensated by the difference of the weights of the
Fock vacua |1, 0| and |0, 0|.)
The cocyclicity of all other terms on the r.h.s. of (11.13)(11.16) is checked analogously.
The essentials of the construction include
(i) The matching between the number of values of two-component indices and the form degree
on the r.h.s. of (11.13)(11.16). This allows us to use the identity (11.19).
(ii) The twist to C1ij or Ci1j of the modules glued to ij via the r.h.s. of (11.13)(11.16)
leads to the shifts of the vacuum helicities and conformal dimensions (i.e., the constant terms in
the operators acting on ij in (11.5)(11.12)) of the respective Fock vacua that compensate those
carried by a, b or a
and b
on the r.h.s. of (11.13)(11.16).
499
11.2. Dynamical content

con (11.1) and (11.2) preserve a degree
First of all we observe that the covariant derivatives Di1i
of a polynomial on which it acts. This means that the Fock modules induced from the vacua
|i, 1 i| decompose into infinite sums of finite dimensional u(2, 2)-modules carried by homogeneous polynomials. On the other hand, from the form of covariant derivatives Diicon (11.3) and
(11.4) it follows that the Fock modules induced from the vacua |i, i| decompose into infinite
sums of infinite-dimensional u(2, 2)-modules.
As explained in Section 5.5, local degrees of freedom of a system are carried by 0-forms. We
conclude that, the fields Ci1i and, therefore, by virtue of (11.15) and (11.16), ii describe an
infinite set of topological systems each carrying at most a finite number of degrees of freedom
equal to the dimension of the space of polynomials of an appropriate degree. Note that the sector
of gauge fields of this type was originally analyzed by far more complicated Hamiltonian methods in [48] with the same conclusion that these fields are of topological type. (In [48] these fields
were called auxiliary to emphasize that they do not carry field-theoretical degrees of freedom.)
On the other hand, each irreducible subsystem in the sector of Cii and i1i describes an
infinite number of degrees of freedom. These are massless fields of various spins. Since the gauge

massless fields 10 (a , b ) are complex conjugated to 01 (b , a ), the system of Eqs. (11.13),
(11.14) and (11.17) at i = j describes the set of massless fields in which every spin appears
twice. This pattern matches that of the AdS4 HS theories as discussed in Section 2, although the
mechanism of the doubling is different.
The realization of the helicity operator H on different Fock modules is given in (11.5)(11.8).
Correspondingly, the eigenvalues of the helicity operator on the HS gauge 1-forms are
H10 = s10 ,
H01 = s01 ,
where spin s is defined according to (3.6), i.e.,

y + y
= 2(s 1)(y, y|x).
(y, y|x)
y
y
(We assume that spin is non-negative while helicities s may have any sign.) On the HS Weyl
0-forms, the eigenvalues of the helicity operator are
HC00 = sC00 ,
HC11 = sC11 ,
where signs are determined by those of the corresponding eigenvalues of the operators (11.7)
and (11.8) (cf. (3.6), (3.7)). We see that the u(1) symmetry generated by H rotates two species
of fields of all spins with the spin-dependent phases.
into the real and imaginary parts

Let us expand i1i (y, y|x)
10 (y, y|x)
= 1 (y, y|x)
+ i2 (y, y|x),
= 1 (y, y|x)
i2 (y, y|x),
01 (y, y|x)
(11.20)
In the AdS4 case, where the u(2, 2)

where i are real in the sense i (y, z |x) = i (z, y|x).

connections are realized by those of sp(4, R) u(2, 2) according to (7.9), both h = 0 and
f = 0, allowing to express the 0-forms Cii via the spacetime derivatives of the dynamical
massless fields. In this case, Eqs. (11.13) and (11.14) take the form

= H
R1 (y, y|x)
2
2
C
(0,
y|x)
+
H
C1 (y, 0|x),
1

y y
y y
(11.21)
500

R2 (y, y|x)
= H
2

y y
C2 (0, y|x)
H
2
y y
C2 (y, 0|x),
(11.22)
where R1,2 (y, y|x)
have the form of the linearized HS curvatures (3.13) and
2
+ C11 (y, y|x)
,
C00 (y, y|x)
2
2
C2 (y, y|x)
=
C11 (y, y|x)
C00 (y, y|x)

2i
satisfy the twisted adjoint covariant constancy condition
C1 (y, y|x)
=0
D tw Ci (y, y|x)
(11.23)
(11.24)
with D tw (3.14) and the reality conditions C1 (y, z ) = C1 (z, y),

C2 (y, z ) = C2 (z, y).
Eqs. (11.21), (11.22) and (11.24) are equivalent to the unfolded massless equations (3.17) and
(3.18) in the AdS4 background with the identification
11 = 2 ,
00 = 1 ,
C 10 = C1 C2+ + C10 C20 ,
C 01 = C1+ + C2 + C10 + C20 ,
where the labels +, and 0 refer to the decomposition (4.1). Thus, in the AdS4 background, the
conformal invariant equations (11.13), (11.14) and (11.17) with i = j amount to the standard
unfolded field equations for the doubled set of free massless fields of all spins.
11.3. Global symmetries and EM duality
As a consequence of general properties of unfolded equations, the massless equations (11.13),
(11.14) and (11.17) are invariant under the u(2, 2) global symmetry that consists of the su(2, 2)
conformal symmetry and u(1) EM duality transformation generated by the helicity operator H.
Since H is central in u(2, 2), from (5.8) it follows that the global symmetry parameter of EM
duality transformation remains x-independent.
The transformation law of the gauge 1-forms consists of the Lie-algebraic transformation in
the module carried by the gauge fields and the additional terms resulting from the Chevalley
Eilenberg cohomology terms via (5.5). From Eqs. (11.13) and (11.14) it follows that the transformation law is

=
gl (a, a,
10 (a, b)
b, b|x)
10 (a, b)

2

+ 2
(x)h

C00 (0, b)
b b

2

C11 (a, 0) |1, 0|,
+
(x)f
(11.25)
a a

01 (a, b)
=
gl (a, a,
b, b|x)
01 (b, a)

2

+ 2
(x)h C00 (b, 0)
b b

2
C11 (0, a)
|1, 0,
+
(x)f
a a
(11.26)
501
where
(x) and
(x) are parameters of global translations and special conformal transformations, respectively. All other symmetry parameters of the conformal algebra enter only
through the original (i.e., C-independent) module transformation law
|. The transformation
law of the Weyl 0-forms does not deform

C00 (b, b|x)
C00 (b, b|x)

=
gl (a, a,
b, b|x)
,

C11 (a, a|x)
C11 (a, a|x)
=
gl (a, a,
b, b|x)
.
The precise form of the transformation law depends on a chosen vacuum connection and,
in particular, on a coordinate system parameterized by the function U AB (X) of Section 9. For
a broad class of vacuum connections, the global symmetry parameter

gl (a, a,
b, b|X)
can be
obtained from (9.13).
It is important to note that the full conformal symmetry does not act individually on every
massless field, mixing together the two copies of fields of equal spins. This happens because of
the gauge field sector where the complex conjugated gauge fields i1i with i = 0,1 transform
differently. This in particular implies that the real fields i (11.20) are mixed by the su(2, 2)
transformations as well as by the EM duality transformations. Correspondingly, the 0-forms Ci
(11.23) are also mixed by the u(2, 2) transformations. This agrees with the fact that the EM
duality transformations cannot act locally on a single gauge potential. On the other hand, that
the AdS4 symmetry sp(4, R) acts individually on the each of the two subsets of i and Ci with
i = 1 or 2 leads to two decoupled sp(4, R) invariant HS systems in (11.21), (11.22) and (11.24).
Somewhat surprisingly, we shall see in Section 13 that, for any spin, the sp(4, R) symmetry of
each of these subsystems extends to sl(4, R) o(3, 3) sp(8, R).
Conformal invariant truncation to the undoubled set of massless fields can be obtained as
follows. Since the generalized Weyl tensors C00 and C11 are self-conjugated, each carrying a
u(2, 2) (in fact, sp(8, R)) module, it is consistent with the u(2, 2) symmetry to set Cii = 0 for
some i. Let us, for example, set
C11 = 0.
(11.27)
This implies that the fields Ci (11.23) are linearly dependent, namely C2 (y, y)
= iC1 (y, y).
Either of Eqs. (11.21) or (11.22) along with the respective covariant constancy condition (11.24) for C1 or C2 describes the set of massless fields of all spins. However, the other
one then has different interpretation. Let, say, 1 (y, y|x)
be chosen as independent electric HS

gauge connection. Eq. (11.21) expresses the generalized Weyl tensors in terms of derivatives
of 1 (y, y|x).
Then the meaning of Eq. (11.22) with C11 = 0 is that it defines the gauge field
2 (y, y|x)
in terms of C1 (y, y|x).
If the form of the r.h.s. of Eqs. (11.21) and (11.22) was the
same, the corresponding equations meant that a linear combination of 1 (y, y|x)
and 2 (y, y|x)
would be pure gauge. However, the r.h.s. of Eqs. (11.21) and (11.22) have different relative signs
of the holomorphic and antiholomorphic parts. As a result, (11.22) expresses 2 (y, y|x)
as the
EM dual of 1 (y, y|x),
i.e., the potential 2 (y, y|x)
is magnetic.
In terms of the complex gauge fields i1i , the (anti)holomorphic components (C11 (0, a))
C11 (a, 0) describe (anti)selfdual components of the complexified s = 1 Maxwell field strength
(C ) C , s = 2 Weyl tensor (C 1 ... 4 ) C1 ...4 and their HS counterparts. As a result, the condition (11.27) implies that the HS field strength is (anti)selfdual. The (anti)selfduality condition
imposed in the Minkowski signature is consistent because the gauge field 10 is complex, having
01 as its complex conjugate. This complex (anti)selfduality condition relates the field strength
of one of the two spin s real fields contained in 10 to the dual field strength of the other one.
502
Let us stress that this relationship is non-local in terms of gauge potentials, being expressed by a
differential duality equation that follows from (11.27).
For example, for the spin one case in tensor notation, (11.27) implies the condition
1
j
i
i
i
Fnm
(11.28)
=
ij nm pq Fpq , Fnm
= n m
m ni ,
2
where i, j = 1, 2 label two real components of the complex spin one potential part 10 (0, 0) of
1 and F 2 is consistent in
(
ij =
j i ,
12 = 1). The relationship (11.28) between Fnm
10 (a, b)
nm
ij
Minkowski case due to the factor of
which becomes the imaginary unit in complex notation.
Thus, the unfolded equations (11.13), (11.14) and (11.17) with C11 = 0 or C00 = 0 describe
the undoubled set of all spins in conformal and duality invariant way. In this case, the unfolded
HS equations describe both electric and magnetic HS potentials. The conformal transformations
resulting from the -dependent terms in (11.25) and (11.26) mix the HS electric and magnetic
potentials. In agreement with the known conformal invariance of (distinguished) 4d spin one
models, the spin one case is degenerate, allowing conformal transformation that acts only on
the electric potential. Technically, this happens because the spin one connection carries trivial
su(2, 2)-module so that the -dependent part of the transformation (11.25) and (11.26) is absent
in this case, i.e., the conformal transformation of spin one results only from the C-dependent
terms in (11.25) and (11.26).
It is tempting to speculate that the restrictions C11 = 0 or C00 = 0 may result from some
dynamical mechanism in a full theory with spontaneously broken HS and sp(8, R) symmetries,
that freezes degrees of freedom of the magnetic phase at least below some mass scale, leaving
us only with the half of degrees of freedom corresponding to the electric phase. The EM duality
can then be expected to be a true local symmetry of a full sp(8, R) invariant nonlinear HS gauge
theory. This can even be true in the HS models with non-Abelian YangMills symmetries considered in [7,40] and in the sector of spin two massless fields that can appear in the HS theories
in many copies, carrying color indices.6 It would be interesting to interpret from this perspective
the gravitational duality studied e.g. in [7782].
11.4. Flat limit
From (11.9)(11.12) it follows that the full chain of field variables in the unfolded formulation
of a massless field of definite spin contains components with the conformal weights from
to . The dynamical HS fields are those with minimal absolute values of conformal weights,
namely D = 1 for spin zero, D = 3/2 for spin one-half, D = 0 for spin s 1 bosonic fields
and D = 12 for spin s 3/2 fermionic gauge fields (for more detail see the -cohomology
analysis of Section 14). All other fields in the chain are auxiliary, being expressed via derivatives
of the dynamical fields by the unfolded equations. Namely, the difference of the modules of
conformal dimensions of a certain auxiliary field A and related dynamical field D equals to the
highest number of derivatives in the resulting expression A( k (D)). The property that the module
of the conformal dimension rather than the conformal dimension itself matters, is intimately
related to the relevance of the AdS background: the mismatch of dimension is compensated by
the powers of the AdS radius 1 .
6 The no-go statements on the existence of models with several interacting spin two fields [71,72] are avoided analogously to the no-go statements on HS interactions [73,74] due to using the AdS4 background and infinite sets of HS
fields.
503
Let us consider more closely the translation and special conformal transformations of the
gauge field 10 . From (11.1) it follows that, discarding the C-dependent terms,

= 2
(x)a
gl (x) b 10 (a, b|x)
+ O(C).
10 (a, b|x)
gl
a
b
In terms of components (3.12) this gives
10 1 ...n , 1 ... m (x)

= 2 n
gl n (x)10 1 ...n1 , 1 ... m (x) + m
gl m (x)10 1 ...n , 1 ... m1 (x) .
(11.29)
With the identification (7.9), the l.h.s. of the unfolded equation (11.13) has the form (3.13).
Abusing notation, the expressions for auxiliary fields resulting from the unfolded system are
[ nm ] dyn

10 (n, m)(x) = 1 D L 2 10 (n0 , m0 )(x),
|n0 m0 | 1,
n0 + m0 = n + m,
where (n, m)(x) symbolizes 10 1 ...n , 1 ... m (x) and we only keep track of the highest deriv
ative terms. In particular, in the case of spin two, the vierbein is the dynamical field while

the Lorentz connection , is auxiliary.
The transformation law (11.29) then implies

dyn
dyn
dyn
10 (x) 1
gl (x)D L 10 (x) +
gl (x)D L 10 (x) .
(11.30)
This transformation law is ill-defined in the flat limit 0. It is possible to rescale the gener

ators of the conformal algebra P 1 P , K K without affecting their com

mutation relations. This induces the rescaling of the parameters

,
1

as well as of the corresponding vacuum connections. This ambiguity can be used to compensate
the factor of 1 in front of one of the two terms on the r.h.s. of (11.30). In particular, choosing
= we obtain instead of (11.30)
10 (x)
gl (x)D L 10 (x) + 2
gl (x)D L 10 (x).
dyn
dyn
dyn
(11.31)
In the flat limit, this transformation law is well defined in the sector of translations but blows up in
the sector of special conformal transformations. This is why spin s > 1 massless field equations
formulated in terms of gauge fields in Minkowski space (in particular, linearized gravity) respect
Poincar symmetry but are not conformal invariant.
A closely related fact is that the rescalings procedure of Section 4, that leads to the conventional flat limit description of HS fields in terms of potentials, breaks down conformal invariance.
The algebraic reason for this is that the Poincar covariant derivative in (4.3) is defined so that
Poincar translations on A + and on A are generated, respectively, by the translation and special
conformal transformations of the original conformal module. Clearly, this realization of translations is incompatible with the standard realization of the Poincar algebra as a proper subalgebra
of the conformal algebra. On the other hand, the Lorentz, duality and dilatation transformations
survive in so defined flat limit because they act on homogeneous polynomials of the oscillators
and commute with the rescalings.
On the other hand, the naive flat limit of Eqs. (11.13), (11.14) and (11.17) with f = 0
at 0 gives rise to conformal invariant equations that is hard to interpret. Actually, in this
system C11 is not any longer expressed via the gauge fields i1i , thus becoming an independent
field. As a result, the system decomposes into two parts. One contains the 1-forms i1i along
504
with the 0-forms C00 while another one is the covariant constancy condition (11.17) on the 0form C11 . The latter however does not make sense in terms of any conventional formulation of
massless fields because it does not express higher components of the expansion of C11 (a, a)
via
the spacetime derivatives of the lower ones, imposing the condition that the lowest component
is a constant. From perspective of -cohomology analysis, this awkward picture results from
the degeneracy of the operator which is zero in this case. As a result, the system gets a
form of an infinite set of topological-like field equations which, in fact, is hard to interpret.
This makes the naive limit ill-defined eventhough the subsystem that contains i1i and C00 is
more tractable if the antiholomorphic field 10 (0, a|x)
(which is the antiholomorphic part of the

Lorentz connection in the spin two case) is chosen as dynamical.
As mentioned in Section 11.3, the spin one case is degenerate because there are no auxiliary
gauge connections, i.e., both terms on the r.h.s. of (11.29) are absent. The nontrivial transformation originates from the C-dependent terms on the r.h.s. of (11.25) and (11.26) where the spin
one components of the 0-forms Cii identify with the Maxwell tensor. These can be rescaled independently to get rid of the -dependence, preserving conformal invariance in the flat limit in
agreement with the fact of conformal invariance of the 4d spin one gauge theory in Minkowski
space.
R) invariant massless equations
12. sp(8,R
The u(2, 2) invariant unfolded equations (11.13)(11.17) admit an extension to the sp(8, R)
invariant form. The fields are still described in terms of the modules |i1i (x) and |Cii (x). The
vacuum covariant derivative (2.11) is defined in star-product notation by (9.4), i.e.,
1
1
W = hAB aA aB + B A aA bB + fAB bA bB .
(12.1)
2
2
A particular form of the sp(8, R) covariant derivative depends on a chosen Fock vacuum

D Cij (i , j |X) = Dij Cij (i , j |X) |i, j |.

D = d + W,
For example,
2
1
1
1
D00 = d + hAB A B + B A bB A + C C + fAB bA bB .
2
b b
b
2
2
(12.2)
Here B A are gauge fields of gl(4, R) that act on homogeneous polynomials of bA . The vacuum |0, 0| forms a one-dimensional gl(4, R)-module. The gl(4, R) sp(8, R) is an extension of
the conformal embedding gl(2, C) u(2, 2), where gl(2, C) contains Lorentz transformations,
dilatations and duality transformations.
Note that, every vacuum |i, j | has the glij (4, R) invariance with the generators bilinear in
the respective creation and annihilation generators. The subalgebras glij (4, R) sp(8, R) are
different for different i, j , having gl(2, C) as the maximal common subalgebra. As a result,
gl(2, C) remains the maximal manifest symmetry of the construction.
The explicit form of the derivatives Dij , that extend the conformal covariant derivatives
(11.1)(11.4) to the sp(8, R) case, can be obtained from (12.2) by renaming the oscillators according to (10.9) and (10.10). Eq. (11.17) still has the covariant constancy form
Dij Cij = 0.
(12.3)
The extension of (11.13)(11.16) is

ij Ci1j (i , 1j )
Dij (i , j ) =
1j =0

+ ij C1ij (1i , j ) =0 ,

1i
505
(12.4)
where
ij =
W (0, i , j , j ) W (0, i , j , j ),
4
i
i
ij =

W (i , i , 0, j )
W (i , i , 0, j )
4
j

(12.5)
(12.6)
with W (aA , bB ) (12.1) represented as W (i , i , j , j ). The tensor product symbol in (12.4)

means that W (i , i , j , j ) is realized as a sum of operators that act both on the (i, j )-module
ij or both on the (i, j )-module and on the (1 i, j )-module
and on the (i, 1 j )-module for
for ij . Namely, by (10.11),
j = j + 1j ,
i = 2i ,
i = 2i
j = 2j ,
j = 2 j
ij term and
in the
i = i + 1i ,
in the ij term in (12.4). Here k and l are understood as derivatives over k and l (precise
signs follow from the definitions (10.9) and (10.10)) with the convention that the differentiation
is done before 1j or 1i are set to zero in (12.4).
The proof of consistency of the unfolded system (12.3) and (12.4) is relatively simple but still
miraculous.
ij ) (12.5) contains connections that have posFirst of all we observe that the operator ij (
itive i i (j j ) grading. Let us for definiteness, consider the case i = 1, j = 0. Using

(10.9) and (10.10) we find that the connections that contribute to 10 are
1
h a a ,
2
and
h a a ,
a b ,

10 C00 (b, b)

b=0

1

=
h
a + + h + b
2
b
b

1

.
+ b C00 (b, b)
h a + + h

2
b
b
b=0
The analysis is greatly simplified in terms of the antiholomorphic Fock space vector

1|,

|0
|10 = 10 C00 (b, b)

(12.7)
(12.8)
(12.9)
b=0
where denotes the restriction of the star-product to the subalgebra generated by the oscillators
carrying primed indices. It can be equivalently rewritten as

1|
,
|0
|10 = g g C00 (b, b)
(12.10)
b=0
506
where

1
g = h
a + + h ,
2
b

h = h a + b .
(12.11)
Note that, because of the wedge product, the antisymmetrization with respect to the indices
implies that g g is symmetrized in the primed indices carried by h (12.11). This allowed
us to replace g g by its -Weyl symbol g g in (12.10).
In the sector where only the fields (12.7) are present, the consistency requires that

1

H=
h a a g g + h a g g
2
b b
g g h
C00 (b, b)
b

a + h C00 (b, b)
+ dh
(12.12)
b
should vanish (the terms resulting from dh do not contribute to this sector). Taking into account
that dh = 2h( h) + , where dots denote terms that contain fields of non-positive a b
grading, we find that the terms resulting from the non-commutativity of the star-product in the
second and third terms in the first and second lines of (12.12) compensate the last term. As a
result, we obtain that

=0
C00 (b, b)
H g g g a
b
by antisymmetrization of three two-component indices of g due to the wedge product.
It remains to consider the part of the consistency condition that contains vacuum connections
of non-positive a b grading. These include
1

h a a , a b ,
2
1

a b ,
f b b .
2
a b ,
1

f b b ,
2
f b b ,
Since (12.8) is independent of these fields, they contribute to the consistency condition at most
linearly. All such terms cancel out trivially except for the dilatation field contained in a b
which, however, cancels out just as in the conformal algebra case considered in Section 11. This
concludes the analysis of the consistency of the term with 1,0 . The analysis of the terms with
ij is analogous modulo renaming the oscillators.
other ij and
Since, the consistency of the unfolded equations (12.4) and (12.3) has been verified for arbitrary flat sp(8, R) connection, from the general argument of Section 5.3 it follows that the system
(12.3), (12.4) is invariant under the global sp(8, R) transformations.
If vacuum connection is chosen to belong to su(2, 2) sp(8, R), the system (12.3), (12.4)
amounts to the conformal system (11.13)(11.17) of Section 11, which is therefore also shown
to be sp(8, R) invariant. As shown in Section 11, with this choice of the vacuum fields it describes
the doubled set of field equations for all massless fields described by i1i and Cii plus an infinite
set of topological fields, each carrying a finite number of degrees of freedom, described by ii
and Ci1i . The interpretation of the sp(8, R) invariant equations (12.3), (12.4) depends, however,
507
on the choice and interpretation of the vacuum fields associated with a chosen flat sp(8, R) connection. In Section 13 we show that the roles of the massless and topological fields are exchanged
if the nonzero vacuum fields are associated with the subalgebra sl(4, R) o(3, 3) sp(8, R).
By the general argument of Section 5.5, to extend the obtained sp(8, R) invariant unfolded
equations to M4 one has to replace the four-dimensional exterior differential by the tendimensional one

AB
dx
dX
X AB
x
simultaneously extending a 4d flat sp(8, R) connection 1-form to M4 . As in the 4d case, the flat
limit degeneracy in the Cartesian coordinates with fAB = 0 is resolved in the AdS-like tendimensional space Sp(4, R). The dynamical interpretation of the resulting equations in Sp(4, R),
i.e., which field components are dynamical, auxiliary, Stueckelberg, etc., is most conveniently
elucidated by the analysis of cohomology. This is done in Section 15.
As discussed in the beginning of this section, the manifest (i.e., linearly acting) symmetry of
the system (12.3), (12.4) consists of the usual Lorentz symmetry plus dilatation and duality transformations which altogether form gl(2, C) = sl(2, C) R u(1). Because it is small enough,

Minkowski coordinates X and spinning coordinates X and X have different appearance
in the full sp(8, R) invariant system lifted to M4 or Sp(4, R), i.e., the equations for the gauge
fields break the manifest gl(4, R) symmetry of the equation on the generalized Weyl tensor C00
down to gl(2, C). A related point is that the analysis of the role of different sp(8, R) curvatures
along the lines of the analysis of conformal field strengths sketched in Section 7, which answers
the question which of the component of the HS field strengths can be set to zero as constraints and
which are zero by virtue of field equations or/and Bianchi identities, turns out to be more complicated in M4 . (This is analyzed in Section 15 in terms of cohomology.) Correspondingly,
an M4 (or Sp(4, R)) analog of the holonomy group is not expected to be larger than GL(2, C).
R) invariant massless equations
13. gl(4,R
The generators LA B (2.9) span gl(4, R) sp(8, R). L and L include the Lorentz generators, H (2.14) and D (2.13). These span gl(2; C) = gl(4, R) u(2, 2). Note that the generators
H and D exchange their roles compared to the conformal case of su(2, 2). Now D (2.13) is the
central element of gl(4, R), that characterizes different irreducible subsystems in the gl(4, R)
invariant equations.
In addition, gl(4, R) contains the generators

L = a b ,
L = a b .
These are analogues of the translation and special conformal transformation generators P and

K of the conformal algebra. The important difference is, however, that P and K are

self-conjugated while L and L are conjugated to each other. This implies in particular that
although

L , L = 0,
L , L = 0,
neither L nor L are translation generators of a Poincar subalgebra of gl(4, R).

One reason why gl(4, R) symmetry might have been missed in field-theoretical models is that

it does not allow lowest weight unitary modules because neither L nor L can serve as
step operators. A related property is that GL(4, R) admits no induced modules to define induced
508
GL(4, R) action on tensor fields. On the other hand, gl(4, R) does act on the lowest weight unitary
sp(8, R)-modules and therefore can act on relativistic fields and their single-particle quantum
states.
Now we observe that
P = L + L
L = L
,
L = L
along with the Lorentz generators span the AdS4 subalgebra o(3, 2) sp(4, R) gl(4, R) (for
simplicity we set = 1 in the rest of this section). This observation leads to the alternative
interpretation of the proposed sp(8, R) invariant equations, as gl(4, R) invariant equations in
AdS4 associated with the embedding o(3, 2) sp(4, R) gl(4, R) sp(8, R).
In the oscillator realization we have P = a b + a b . So defined P acts in finite
dimensional spaces of homogeneous polynomials of the modules ii and Cii . On the other hand,
i1i and Ci1i now decompose into the infinite sum of infinite-dimensional sp(4, R)-modules.
This means that the subsystem of field equations for ii and Ci1i now describes massless fields
in AdS4 while that for i1i and Cii describes an infinite set of topological fields. We see that
the massless and topological fields exchange their roles depending on which sp(4, R) subalgebra
of sp(8, R) is identified with the AdS4 symmetry.
Note that the original form of the nonlinear massless field equations of [5] (see also [28])
in which the massless fields are self-conjugated as presented in Section 3 is most naturally related to the gl(4, R) invariant version of the equations. The free gl(4, R) invariant system admits
a reduction to the subsystem that describes a single massless field of any spin. Algebraically,
this is because, in the gl(4, R) invariant case, spin is characterized by the eigenvalues of the
non-compact generator D (cf. (11.9)(11.12)) that admits one-dimensional real modules rather
than by the compact generator H with minimal two-dimensional modules as in the conformal
case. Flat limit can be taken using the rescaling procedure of Section 4. However, analogously
to su(2, 2), the gl(4, R) symmetry does not survive in the flat limit, thus being invisible in
Minkowski space.
Equivalently, the gl(4, R) covariant description can be obtained by keeping the same oscillator
generators as in the conformal case but changing the conjugation conditions to a = b , b =
a , which implies that |i, j | = |1 j, 1 i| and, therefore, ij = 1j,1i for ij = ij or
Cij . With these reality conditions we obtain that 00 = 11 , 01 = 01 and 10 = 10 . In this
setup, the sets of higher spin and topological fields remain the same as in the su(2, 2) case but
the reality conditions change.
As shown in Section 11.3, the free equations for a single massless field of a fixed spin admit
the action of su(2, 2) that involves dual gauge potentials in the transformation law. This means
that, allowing nonlocal field transformations of this kind, free field equations of a massless field
of a given spin are invariant under both su(2, 2) and gl(4, R) and, therefore, under their closure
gl(4, C). Note that gl(4, C) acts locally on the doubled sets of massless fields with the natural
realization of a pair of real fields as a single complex field. Let us stress that gl(4, C) algebra does
not belong to sp(8, R) that acts individually on every Fock module. Rather, gl(4, C) sp(8, C)
where sp(8, C) mixes two Fock modules that describe massless fields in a chosen vacuum realization.
An interesting project for the future is to look for nonlinear gl(4, R) invariant models in AdS4 .
Since gl(4, R) acts individually on fields of different spins, the problem can be analyzed, e.g. for
spin two, i.e., AdS4 gravity or supergravity. We hope to come back to this intriguing question
elsewhere.
509
14. analysis in Minkowski space

14.1. Grading
An appropriate grading G of the Fock modules in the unfolded HS equations is G = |D| =
where |D| results from the dilatation generator D via replacing its eigenvalues by their
absolute values. In other words, the grading G of a field equals to the absolute value of its
scaling dimension. So defined G is diagonalizable and bounded from below. Abusing notation,
from (11.9)(11.12) we obtain for ij = Cij or ij

1

G00 (b, b) =
+b
(14.1)
b
+ 2 00 (b, b),
2
b
b

1
=
+ a
+ 2 11 (a, a),
a
G11 (a, a)
(14.2)
2
a
a

= 1 a b 10 (a, b),
G10 (a, b)
(14.3)
2 a
b

1
= b a
01 (b, a).
G01 (b, a)
(14.4)
2
b
a
1
A
2 |aA b |,
The grading G treats symmetrically the parts A+ and A in the decomposition (4.1) underlying the flat limit and leads to the standard description [44,45] of 4d massless fields.
14.2. 0-forms
Let us analyze Eq. (2.1) on the 0-form C(b|x). The grading (14.1) implies that
= dx
2
.
b b
That 2 = 0 is the consequence of anticommutativity of the differentials dx . As expected, the

dynamical fields in H 0 ( ) are holomorphic and antiholomorphic
H 0 ( ):
b).
C(b) + C(
H 1 ( ) is of the form
H 1 ( ):

+ b b E ,
dx b E (b) + b E (b)
(14.5)
are arbitrary polynomials of their arguments and E is a constant. It is easy

where E (b), E (b)
to see that the elements in (14.5) are closed but not exact. We leave it as an exercise to
and E parameterize the
the reader to check that (14.5) describes full H 1 ( ). E (b), E (b)
l.h.s. of Eqs. (2.3), (2.4). Indeed, the unfolded equations (2.1) demand all derivatives of the 0form dynamical fields, that turn out to be closed because the dynamical fields themselves are
closed, to be zero except for those that are exact to be absorbed by C aux with some
auxiliary fields C aux (for more detail see e.g. [29]). This has the consequence that Eqs. (2.3),
(2.4) follow from (2.1). Since (14.5) describes the full cohomology, they provide the full list of
differential equations imposed by the unfolded equations (2.1) on the dynamical fields.
To analyze the content of the 4d massless field equations (3.17) and (3.18) for gauge potentials
first of all we observe that all fields C(b, 0|x) and C(0, b|x),
except for the scalar field C(0, 0|x)
510
and spinor field linear in bA , are expressed via the gauge fields by Eq. (3.17) thus becoming
auxiliary fields. To take this into account it is convenient to redefine where acts on
the direct sum of spaces of 0-forms C and 1-forms so that the terms on the r.h.s. of (3.17) become (C). As a result, all auxiliary fields C(x) that correspond to spins s 1 disappear from
H 0 ( ). Correspondingly, their field equations become consequences of the Bianchi identities
for the equations (3.17) for s > 1, thus disappearing from H 1 ( ).
The case of spin one is special. Here, Eq. (3.17) is just the definition of the Maxwell field
strength C and C in terms of potentials while the spin one equation is still a part of the co

homology (14.5), dx (b E b b E b ), where E is an arbitrary Hermitian bispinor
(i.e., Lorentz vector) that parameterizes the l.h.s. of the second pair of the Maxwell equations.

The first pair associated with dx (b F b + b F b ) becomes the Bianchi identity and
1

disappears from H ( ).
The sector of spin 0 and 1/2 is unaffected by the transition from to . As a result, we
conclude that the relevant part of the cohomology in the sector of 0-forms C is
H 0 ( , C):
H 1 ( , C):
C + b C + b C ,

dx b E + b E + b b E + b E b b E b ,
(14.6)
(14.7)
where C, C , C and E, E , E , E parameterize, respectively, dynamical fields of spin 0

and 1/2 and the l.h.s. of the dynamical equations of spin 0, 1/2 and 1. Now we are in a position
to analyze the sector of gauge fields.
14.3. 1-forms
Dynamics of spins s 1 is described by the gauge 1-forms . As shown in [43,40] (without
using the cohomology language, however) the dynamical fields are 1 ...n , 1 ... m with n = m
for bosons and |n m| = 1 for fermions, which are the frame-like counterparts of the Fronsdals double traceless boson and triple -transverse fermion metric-like fields, respectively. Field
equations for massless fields of all spins s > 1 are contained in the sector of gauge 1-forms. Let
us show how these facts are reproduced in terms of cohomology.
In the case of 1-forms, the grading operator is of the type (14.3)
1
G = |n n|,
n = y

, n = y
.
y
y
(14.8)
The operator is
(A) = A + A,

+ y A+ (y, y|x)
,
A = e y A (y, y|x)
y
y
weyl
(14.9)
where is the part of responsible for gluing the Weyl 0-forms C to the field strengths of
the gauge 1-forms via the terms on the r.h.s. of (3.17). Note that defined as the e-dependent
part of (4.3) respects the decomposition (4.1). Equivalently,
weyl
= (n n 2) + (n n 2) +
weyl
(14.10)
where
= e
y ,
y
= e
y
y
(14.11)
511
and
(m) = 1(0),
m 0 (m < 0).
(14.12)
Although and do not anticommute, squares to zero because ( )2 = ( )2 = 0 and

the step functions guarantee that the parts of associated with and act in different spaces.
The nontrivial cohomology of is concentrated in the subspaces of G-grades 0, 1/2 and 1.
This follows from the fact that the operators and (14.11) act as the exterior differentials

y and with = e y and = e y in the spaces of functions of y and y ,
y
respectively. As a result, by Poincars lemma, the cohomology is concentrated in the sectors
where differs from or , that is where the step functions differ from a constant. Also let
weyl
us note that, in the gauge field sector, the difference between and due to matters
only in the computation of H 2 ( ) because the Weyl 0-forms in (3.17) contribute to the sector
of 2-forms.
H 0 ( ) is easy to compute. The nontrivial cohomology appears in the subspaces of grades
G = 0 or 1/2 where acts trivially because of the step functions in (14.10). So,

1

H 0 ( ):
(y, y)
(14.13)
=
y1 . . . yn y 1 y m
1 n , 1 ... m .
2n!m!
|nm|1
...
1 ...n , 1 m (x), |n m| 1 are parameters of differential gauge symmetry transformations of

spin s = 1 + 12 (n + m) massless fields. For integer spins with n = m = s 1, the corresponding
spin s gauge symmetry parameter is equivalent to a rank s 1 symmetric traceless Lorentz tensor.
This agrees with the standard Fronsdal formulation [44]. For half-integer spins, n = s 3/2,
m = s 1/2 or m = s 3/2, n = s 1/2. The corresponding spin s gauge symmetry parameter
is equivalent to a rank s 3/2 symmetric -transversal tensor-spinor in tensor-spinor notation.
This agrees with the FangFronsdal theory [45].

Eq. (14.13) implies that all gauge parameters
1 ...n , 1 ... m (x) with |n m| > 1 are of
Stueckelberg type, i.e., the associated gauge field transformations contain algebraic shifts that
gauge away some components of the HS connections. In particular, the linearized local Lorentz

symmetry parameters
(x) and
(x), which allow to gauge away the antisymmetric part of
the spin two vierbein, are of this class.
H 1 ( ) describes those components of the 1-form connections that are neither auxiliary (i.e.,
being expressed in terms of other connections by algebraic constraints that can be imposed in
terms of HS curvatures, la zero-torsion constraint in gravity) nor Stueckelberg (i.e., cannot be
gauge fixed to zero by algebraic shift gauge symmetries). It is not hard to see that H 1 ( ) is
concentrated in the subspace with G = 0 for bosons and G = 1/2 for fermions, where acts
trivially so that all elements are closed in this sector. To compute H 1 ( ) it is thus enough to
factor out the exact part, which is equivalent to gauging away the Stueckelberg components
of the gauge fields. This gives the following results.
In the bosonic case
H 1 bos ( ):
(y, y)
= e
G() = 0,
2

+ e y y (y, y),
(y, y)
y y
i.e., spin s 1 dynamical fields identify with the 0-forms (y, y|x)
and
n(y, y|x)
= n(y,
y|x)
= s(y, y)(x),
(14.14)
(y, y|x)
that satisfy
512
n (y, y|x)
= n
(y, y|x)
= (s 2) (y, y|x).
These describe two irreducible components of the spin s Fronsdal double traceless symmetric
tensor field. In particular, in the spin two sector, H 1 bos ( ) describes a rank two symmetric
tensor in terms of two-component spinors. This is the fluctuational part of metric equivalent to
the fluctuational part of the vierbein modulo linearized local Lorentz gauge symmetry.
In the fermionic case
H 1 fer ( ):
+ (y, y)
+ (y, y),
+ (y, y)
= e
(y, y)
= e
2

y y
2

y y
(n n)
(y, y)
= (y, y),
(14.15)
+ e y y 3+ (y, y),
2 (y, y)

+ (y, y)
+ e y y 3 (y, y).
y 2
1+ (y, y)
+ e y
1 (y, y)
+ e y
Here 1+ (y, y|x),
2 (y, y|x)
and 3+ (y, y|x)
and their conjugates 1 (y, y|x),
2+ (y, y|x)
and
3 (y, y|x)
describe three irreducible components of the FangFronsdal triple -transverse symmetric tensor-spinor fermionic field. For a half-integer spin s we have
n1 = (s 1/2)1 ,
n
1 = (s 1/2)1 ,
n3 = (s 2 1/2)3 ,
n
3 = (s 2 1/2)3 .
n2 = (s 1 1/2)2 ,
n
2 = (s 1 1/2)2 ,
Finally, H 2 ( ) classifies differential equations on the dynamical fields contained in the 4d

unfolded HS system. H 2 ( ) consists of the generalized Weyl part parameterized by the r.h.s.
of (3.17) and the Einstein cohomology that represents the l.h.s. of the massless field equations.
weyl
Since the generalized Weyl tensor part has already been taken into account by in (14.9),
H 2 ( ) consists of the Einstein cohomology.
Let us start with the simpler fermionic case. H 2 fer ( ) consists of the grade 1/2 2-forms R,
which are automatically closed, modulo exact 2-forms
R exact = W,
R exact = W ,
(n n)W
= 3W,
(n n)W = 3W .
Elementary computation shows that H 2 fer ( ) is

+ E (y, y),
(n n)E
(y, y)
= E (y, y),
H 2 fer ( ) = E + (y, y)
(14.16)

2

= H
+ y E2+ (y, y)
+ H y y E3 (y, y),
E + (y, y)

E (y, y)
y y 1
y

2

+
E (y, y)
= H
E
(y,
y)
+
y
E
(y,
y)
+ H y y E3+ (y, y),
1
2
y y
y

where H and H are the basis 2-forms (3.11) and
nE1 = (s 1/2)E1 ,
nE
1 = (s 1/2)E1 ,
nE3 = (s 2 1/2)E3 ,
nE
3 = (s 2 1/2)E3 .
nE2 = (s 1 1/2)E2 ,
nE
2 = (s 1 1/2)E2 ,
513
As a linear space, Einstein cohomology H 2 fer ( ) is isomorphic to H 1 fer ( ) (14.15). This

is expected because the massless field equations are Lagrangian, i.e., there are as many equations
as field variables.
Let us now consider the bosonic case. Here the sectors of G = 0 and 1 should be analyzed.
It is easy to see that

H 2 ( )G=0 = 0,
(14.17)
which means that any 2-form with G = 0, which is automatically closed, is exact. For spin
one this means that the respective part of Eq. (3.17) with G = 0 is a constraint that expresses the

Maxwell stress tensor C , C via derivatives of the spin one gauge potential. For spins s 2,
(14.17) implies that the 4d 1-form gauge fields (y, y)
with G = 1 are auxiliary being expressed
(modulo pure gauge components) via derivatives of the dynamical fields by the zero torsion condition R(y, y)|
G=0 = 0 that imposes no differential equations on the dynamical Fronsdal fields.
In particular, in the spin two sector, (14.17) allowed to impose the standard zero-torsion condition, that expresses Lorentz connection via derivatives of the vierbein imposing no restrictions
on the latter.
Now consider the part of the cohomology H 2 ( ) in the G = 1 sector. We have to find such
2-forms with n n = 2 and with n n = 2 that
+ = 0.
(14.18)
The nontrivial cohomology consists of solutions of this condition with = 0 and = 0

(otherwise, and are exact). The key point is that, to cancel out in (14.18), the terms
coming from and should carry equivalent representations of the Lorentz algebra eventhough
they support polynomials of y and y with n n = 2 and n n = 2, respectively.
It is not hard to see that this is possible if the cohomology space is spanned by the polynomials

Ei (y, y)
that contain as many y as y , i.e., G(Ei (y, y))
= 0. The appropriate Ansatz is
2
E2 (y, y),
y y
(14.19)
2

E3 (y, y)
+ H y y E4 (y, y).
y y
(14.20)
= H y y E1 (y, y)
+ H
= H
Using the identities

H e =
H +
H ,

H e =
H
H ,
where H = 13 e e e are 4d basis 3-forms, we obtain

+ = H y y y + 3 (E1 E4 )
y

2

+ y
1 (E2 E3 ) .
y
y y
(14.21)
This implies that (14.19) and (14.20) describe a nontrivial cohomology provided that E2 = E3 =
E and E1 = E4 = E , where E (y, y)
is an arbitrary grade zero polynomial while E(y, y)
should
be at least of fourth order to contribute.
514
For any integer spin s 2, the Einstein cohomology

2
2
+ H E(y, y)
H 2 bos ( ) = H
y y
y y

+ H y y + H y y E (y, y)
(14.22)
is described by the polynomials E(y, y)

and E (y, y)
that satisfy
nE(y, y)
= nE(y,
y)
= sE(y, y),
nE (y, y)
= nE
(y, y)
= (s 2)E (y, y).
Einstein cohomology is responsible for the dynamical field equations for any integer spin
s 2: the condition that the part of the HS curvature 2-form parameterized by E(y, y)
and
E (y, y)
is zero, which is true by the unfolded equations (3.17), imposes the Fronsdal field equations on the dynamical fields of spins s 2. This agrees with the condition that there are as
many dynamical equations as field variables. In terms of Lorentz tensors, E(y, y)
and E (y, y)
are equivalent to rank s and rank s 2 traceless tensors, respectively.

=0
The spin one part of E is exact because from (14.21) it follows that () = ()
in this case. This conforms the fact that the spin one dynamical equations are described by the
0-form cohomology (14.7).
The pattern of H p ( ) proves the so-called Central-On-Shell theorem [40] that states that
Eqs. (3.17) and (3.18) are equivalent to the standard 4d massless field equations plus an infinite
set of constraints that express infinitely many auxiliary fields via derivatives of the dynamical
massless fields. Let us stress that the Central-On-Shell theorem is true both in 4d Minkowski

space and in AdS4 . Actually, although the frame 1-form e (x) that enters does depend on
a chosen geometry and coordinate system, the analysis of the cohomology only uses that the
vacuum connection is flat and the vacuum frame field is non-degenerate, forming a frame in the
space of 4d 1-forms.
15. analysis in M4
15.1. and
The grading operator G, which is independent of the choice of a base manifold, is given by
(14.1)(14.4). is the grade 1 part of the covariant derivative. Let ij denote in the
module ij . We obtain
2
1
00 = hAB A B ,
2
b b
2
1
11 = fAB
,
2
aA aB

1 2

+h
a + h a a (n n 2)
h
01 =
2
b b
b

2

1
f
b + f b b (n n 2),
+
f
2
a a
a
(15.1)
(15.2)
(15.3)
10 =

2

1

f
f
b + f b b (n n 2)
2
a a
a

2
1

a a (n n 2),
h
+

+h
a + h
2
b b
b
515
(15.4)
where n, n and the step function (n) are defined in (14.8) and (14.12). Because i1i are
asymmetric with respect to primed and unprimed indices, the manifest symmetry is GL(2, C)
which consists of the Lorentz symmetry SL(2, C), dilatations and duality transformations.
Recall that in the case (8.7) of Sp(4, R), both hAB and fAB are non-degenerate. The naive flat
limit with fAB 0 is degenerate. However, because the operators i1i are defined differently
in the sectors with n n > 0 and n n < 0, the fields can be rescaled so that i1i remain nondegenerate in the flat limit. The result extends the e-dependent part of the Minkowski covariant
derivative (4.3) to M4 .
Once both hAB and fAB are expressed in terms the vielbein eAB by (8.7), from (15.1)(15.4)
it is clear that, up to renaming the oscillator variables, there are two essentially different
operators. One in the sector of 00 and 11 and another one in the sector of 01 and 10 . To
simplify notations we call the former and the latter . Denoting the respective oscillators as
y A we have
= eAB
(15.5)
= t (n n 2) + t (n n 2),
(15.6)
y A y B
where
t = + + 0 + ,
= e
2
y y
= e
0 = e
2
,
y y
Note that
i j
, = 0,
t = + + 0 + ,
y ,
y
0 = e
i j
, = 0,
(15.7)

+ = e y y ,
(15.8)
+ = e y y .
(15.9)

y ,
y
i, j = , 0, +
and, therefore (t )2 = (t )2 = 0. Although {t , t } = 0, ( )2 = 0 because the step functions in

(15.6) imply that the parts of associated with t and t act in different subspaces.
The dependence on has been removed by a field redefinition along the lines of Section 3,
which also was used to adjust convenient coefficients in (15.7). In the rescaled variables y A ,
appears in front of the +ij part of the covariant derivative. Correspondingly, the flat space
field equations in M4 result from setting = 0 and dropping the terms with +ij . In the flat case
of M4 one can use Cartesian coordinates with
eAB = dX AB ,
D tw;fl = d + ,
D ad;fl = d + .
(15.10)
The dynamical content of the unfolded field equations in Sp(4, R) and M4 is determined
by the cohomology of and . We have to calculate H 0 ( ) and H 1 ( ) to identify the
independent fields and field equations in the twisted adjoint 0-form sector and H 0 ( ), H 1 ( )
and H 2 ( ) to identify the gauge parameters, dynamical fields and gauge invariant combinations
516
of derivatives of the dynamical fields that either represent the l.h.s. of field equations or identify
with the generalized Weyl tensors via the ChevalleyEilenberg cohomology terms.
We start with H p ( ) extending the results of [3] to the AdS-like case of Sp(M, R).
15.2. Weyl 0-form sector
In the case of MM , the unfolded equations in question are (2.15). They are equivalent to
D tw;fl C = 0. H 0 ( ) = C + y A CA describes the dynamical fields C(X) and CA (X)bA in MM .
H 1 ( ) is
H 1 ( ) = eAB y C y D EAB,CD + eAB y C EAB,C ,
where EAB,CD and EAB,C represent the l.h.s. of the field equations (1.1) and (1.2) and satisfy
EAB,CD = EBA,CD = EAB,DC ,
EAB,C = EBA,C ,
E(AB,C)D = 0,
E(AB,C) = 0.
Eqs. (1.1) and (1.2) in M4 were originally derived this way in [3]. The analysis in the curved
Sp(M, R) background is analogous. is still given by (15.5), where eAB is the generalized
vielbein of Sp(M, R) introduced in Section 8. The symmetry type of the equations and leading
derivative terms remain the same as in M4 . The exact form of Eq. (1.1) is deformed by the 2 dependent lower-derivative terms. These appear due to the + part of the covariant derivative
associated with the nonzero special conformal connection (8.7)
1
+ = 2 eAB bA bB , eAB = CF A CGB eF G ,
4
where CAB is the Sp(M, R) invariant antisymmetric form.
Also, usual derivatives have to be replaced by the Lorentz-like Sp(M, R) covariant derivatives
associated with the connection A B
DfA (X) = dfA (X) + A B (X)fB (X),
eAB = dX AB eAB AB ,
D = dX AB DAB ,
DAB = eAB AB DAB .
As a result, the Sp(2M, R) invariant deformation of Eqs. (1.1) and (1.2) to Sp(M, R) reads as
1
(DAB DCD DCB DAD )C(X) 2 (CAC CBD + CBC CAD )C(X) = 0,
2
DAB CC (X) DCB CA (X) = 0.
(15.11)
(15.12)
15.3. Gauge 1-form sector

In this subsection we analyze H p ( ) that determines dynamical content of the field equations
in the sector of 1-form connections.
15.3.1. Auxiliary lemmas
The following useful lemmas will be used in what follows.
Lemma 1. H p ( + , P ) = 0 at p = 0, 1, 2 if P is the space of polynomials of y with =
1, 2, . . . , m > 1.
517
Proof. H 0 ( + , P ) = 0 because Ker + = 0 on the space of polynomials (product of any two

nonzero polynomials never gives zero polynomial).
The case of H 1 ( + , P ) can be analyzed as follows. Let (y) be a 1-form (y) =
e 1 ...n y1 yn . The + closedness condition y y (y) y y (y) = 0 is equivalent to

(n) (n) = 0,
where we use the convention that lower (upper) indices denoted by the same letter are symmetrized and a number of symmetrized indices is indicated in parentheses. Contracting twice the
indices with we obtain

(n + m + 1)(n + m) 2 (n)
(15.13)
(15.14)
= 4n (n1) + n(n 1) (n2) .

One more contraction gives
(n + m 2) (n1) = (n 1) (n2) .
2
Now one observes that since a + -exact 1-form has the form e y y (y), a cohomology class
can be fixed by setting (n2) = 0. Then from (15.14) and (15.13) it follows that (y) = 0,
i.e., H 1 ( + , P ) = 0. The proof that H 2 ( + , P ) = 0 is analogous. 2
From Lemma 1 follows:
Lemma 2. H p (t , P ) = H p (t , P ) = 0 at p = 0, 1, 2.
Proof. Consider the sector of p-forms 0 expandable into the wedge product of p 1-forms e .
From the condition that is t closed it follows that 0 is + closed. By Lemma 1 it follows
that 0 is + exact. Therefore, one can choose a representative of H p (t , P ) = 0 with 0 = 0
in the purely e sector. Then one considers the sector of p-forms 1 that contains p 1 e ,
repeating the analysis. The process continues till one proves that H p (t , P ) = 0. Analogously
one proves that H p (t , P ) = 0. 2
Corollary. As a consequence of Lemma 2 it follows that H p ( , P ) is concentrated in the subspace with G 1 where + and + do not act independently.
Note that the case with n n = 2 is still nontrivial because here both Im t and Im t
belong to the space with G = 0 and can cancel each other thus extending Ker compared to
Ker t Ker t . (Analogous phenomenon occurred in the 4d analysis of the bosonic sector of
H 2 ( ) in the end of Section 14.3.)
15.3.2. H 0 ( )
H 0 ( ) is easy to compute. Here the key observation is that the sectors with n n = 2 do
not talk to each other. As a result, H 0 ( ) is described by the same formula (14.13) as in the
4d case, i.e., the true gauge symmetry parameters in the matrix space are described by the same

set of multispinors as in Minkowski space. In particular, all gauge parameters
1 ...n , 1 ... m (X)
with |n m| > 1 are of Stueckelberg type with the field transformations that contain algebraic
shifts gauging away some components of the HS 1-form connections.
518
15.3.3. H 1 ( )
The computation of H 1 ( ) is also based on the fact that it is concentrated in the subspace
where is identically zero, i.e., |n n|
1.
In the bosonic case, it is necessary to check that the possible extension of H 1 ( ) due to
extension of Ker compared to Ker t Ker t does not take place. The proof of this fact,
which is elementary but somewhat lengthy, we leave to the reader. The idea of the proof is
illustrated below by the analysis of H 2 ( ) which results in nontrivial cohomology.
Just as in the 4d case, the ambiguity in adding -exact 1-forms is used to get rid of Stueck
elberg components of the 1-forms of the form e (y, y).
As a result,
H 1 ( ):

(y, y)
= e (y, y)
+ e (y, y)
+ e (y, y),
and (y, y)
are arbitrary fields with |n n|
1 while e (y, y)
has
where (y, y)
the same content (14.14) and (14.15) as in the 4d theory. We conclude that, apart from the
4d Fronsdal gauge fields, the formulation in M4 requires the additional fields (y, y)
and
(y, y),
that describe components of the 1-form connection along the spinning directions
in M4 . Since the system in M4 is by construction equivalent to that in Minkowski space, the
additional dynamical fields in M4 are related by their field equations to the 4d HS fields.
To figure out the form of nontrivial field equations we have to compute H 2 ( ). We start with
the simpler fermionic case.
15.3.4. H 2 ( ): Fermions
H 2 fer ( ) consists of the 2-forms R = eAB eCD RAB,CD (y, y)
with n n = 1 (which
are all -closed) modulo -exact 2-forms
R exact+ = t W + ,
W
R exact = t W ,
(n n)W

(15.15)
= 3W .
where
are arbitrary 1-forms such that

In the e e sector the analysis repeats that of the 4d case. The nontrivial class is represented by the Einstein cohomology and Weyl cohomology. The Weyl cohomology is represented
+3/2
by the y independent 0-form C y y y and its conjugate that describe spin 3/2 (cf. (3.6))
Thus, this part of H 2 fer+ is

2
2 fer+

=H
+ y +2 (y, y)
H0

1 (y, y)
y y
y

2
+ H y y 3 (y, y)
(15.16)
+ C +3/2 (y) ,
y y

= i (y, y)
and the 2-forms H and H are defined in (3.11).
where (n n)(
i (y, y))
H02 fer is given by the complex conjugated expression.
Important difference compared to the 4d case is that, by Lemma 1, the Weyl 0-forms associated with spins s > 3/2 do not correspond to a nontrivial cohomology in the case of M4 . This
means that although the generalized Weyl tensors appear on the r.h.s. of (12.4), in M4 this is a
consequence of Bianchi identities applied to the lower spin field equations. This conclusion is
consistent with the fact [3] that, in M4 , different spins correspond to modes of the hyperfields
C(X) and CA (X) with respect to extra spinning directions. In other words, it is not possible to
restrict to zero C(y|X) of some power in y without restricting its dependence on the spinning
coordinates.
519
Having fixed the representatives H02 fer+ and H02 fer in the form (15.16) and complex con
jugated, we have fixed the part of -exact 2-forms (15.15) with W0 = e W0
. The other
W
+
e
remain
to
be
factored
out
in
the
analysis
of the recomponents W1 = e W1
1

maining sectors of H 2 fer ( ). The analysis of the sectors e e and e e amounts

effectively to the 4d analysis of H 1 ( ) where the leftover components of the 1-forms e W1

and e W1
are treated as the 0-form Stueckelberg parameters. As a result, this part of
2
fer
( ) is
H
H12 fer+ = e

+
+ e y 2 (y, y)
+ e y y 3+ (y, y),
(15.17)
1 (y, y)
y y
y

(y, y),
= e i
+ i
and analogously for the complex conjuwhere i (y, y)
(y, y)
e
gated.

Finally, the sectors of e e and e e fully belong to H 2 ( )

H22 fer = e e 1
(y, y)
+ e e 2
(y, y).
To summarize, in the fermionic case H 2 fer ( ) = H 2 fer+ ( ) H 2 fer ( ), where

2 fer ( ) are complex conjugated and
) and H
H 2 fer+ (
H 2 fer ( ) = H22 fer H12 fer H02 fer .

All components of H 2 fer ( ) except for the C-dependent Weyl cohomology term in (15.16)
correspond to the l.h.s. of differential field equations imposed by the unfolded equations (12.3)
and (12.4) on the dynamical HS gauge fields in M4 . Because H 2 fer ( ) is concentrated at the
lowest grade, all these equations are of first-order. Their explicit form results from projecting to
zero the components of the HS field strengths 2-forms in M4 that belong to H 2 fer ( ).
15.3.5. H 2 ( ): Bosons
In the bosonic case nontrivial cohomology can be concentrated in the sectors with G = 0 or 1.
Let us start with the case of G = 0, denoting this part of the cohomology H02 bos ( ).
In the 4d analysis, the H02 ( ) was zero. The meaning of this result for spin s > 1 was that
the 1-form gauge connections (y, y)
with (n n)(y,
y)
= 2(y, y)
were auxiliary fields
expressed by the zero torsion condition R(y, y)|
G=0 = 0 via derivatives of the dynamical fields
module pure gauge Stueckelberg components. In particular, in the spin two sector, the fact that
H02 ( ) = 0 makes it possible to impose the zero-torsion condition in 4d gravity.

The situation in M4 is different. The part of H02 bos ( ) in the sector of e e is still zero
as in four dimensions but the other sectors are nonzero. Analogously to the fermionic case, it is

easy to see, that fixing to zero the cohomology class in the sector of e e implies that the

nonzero class of H02 bos ( ) in the sectors of e e and e e has the form analogous to
that of H 1 ( ) (14.14) with the dynamical fields and replaced by 1-forms with coefficients

in e and e . Finally, the part of H02 bos ( ) in the sector free of e remains unrestricted. To
summarize,
+ e
H02 bos ( ) = (y, y)
2

+ e y y W (y, y),
W (y, y)
y y
520
where

(y, y)
= e e (y, y)
+ e e
()

+e
(y, y)

e (y, y),
()
W () (y, y)
= e W (y, y)
+ e W (y, y).
This part of H 2 bos ( ) is responsible for the field equations that relate additional field components in H 1 ( ) to the usual 4d massless fields associated with H 1 ( ).
Now let us consider the part of H 2 bos ( ) with G = 1 which we denote H12 bos ( ). It is

nonzero due to the phenomenon discussed in Corollary. The analysis of the 4d sector e e
is identical to that of Section 14.3. The question is how this cohomology extends to M4 . In
principle, it might happen that the Einstein cohomology disappears or shrinks to a smaller vector
space in M4 which would imply that most of the 4d field equations become consequences of
the Bianchi identities in the larger spacetime, applied to the extended zero-torsion condition
and/or to a subsystem of the field equations. Indeed, such a phenomenon occurred in the analysis
of the equations on the Weyl 0-forms in [3] where the infinite-dimensional cohomology in 4d
shrinks to a finite-dimensional one in M4 in agreement with the fact that the infinite set of 4d
field equations (2.3) and (2.4) for massless fields of all spins amounts to the finite system of
Eqs. (1.1) and (1.2) in M4 . This does not happen to the Einstein cohomology, however. Namely,
as in 4d Minkowski space. The detailed analysis
it is still parameterized by E(y, y)
and E (y, y)
is straightforward although somewhat annoying. The final result is that the nonzero components
of H12 bos ( ) are
= + + ,
where
+
+
+
+ = 2+ + 1+ + 0+ + 4d
+ 1
+ 2
,
+ 1
+ 2
= 2 + 1 + 0 + 4d
and

+
4d
+ 4d
= H

2
2
H E(y, y)
y y
y y

+ H y y H y y E (y, y),
4
y E
E + e e
n+4
y n

2
2
+ e e
y y
2(n + 2)E n(n 1)E
y y
n(n + 1)(n + 2)
2
2(n 1)
(n + 2)E + nE ,
+ e e y y
y y n(n + 1)(n + 2)
1+ = e e y y y y
4
2
4

E + e e y
E
y y y y n 2
y n + 2

2
2

(n + 2)(n + 3)E + 2nE
+ e e y y
y y n(n + 1)(n + 2)
1 = e e
(15.18)
+ e e
y y
+
1
= 2e e

2(n + 3)
nE + (n + 2)E ,
n(n + 1)(n + 2)
4
1
4

E + e e y
E
y y y y n 2
y n + 2
+ e e y y
+ e e
y y
521

2
2
2nE + (n + 2)(n + 3)E
y y n(n + 1)(n + 2)

2
2(n + 3)
y y
nE + (n + 2)E ,
y y
n(n + 1)(n + 2)
2
4

E e e y E
n+4
y n

2
2

+ e e
y y
2(n + 2)E n(n 1)E
y y
n(n + 1)(n + 2)

= e e y y y y
1
+ 2e e
y y

2
n1
nE + (n + 2)E ,
y y n(n + 1)(n + 2)

n1
(n + 2)E + nE ,
(n + 1)(n + 2)
2
n+3
(n + 2)E + nE ,
2 = e e
y y n(n + 1)
2
n+3

+
= e e
(n + 2)E + nE ,
2
y y n(n + 1)

n1

2
= e e y y
(n + 2)E + nE ,
(n + 2)(n + 1)
2+ = e e y y

3
n+3
y
(n + 2)E + nE
y y y
n(n + 2)(n + 1)
n1

+ 2e e y y y
(n + 2)E + nE ,
y n(n + 2)(n + 1)
0+ = 2e e
n1
(n + 2)E + nE
y n(n + 2)(n + 1)

3
n+3

(n + 2)E + nE ,
+ 2e e y
y y y n(n + 2)(n + 1)

0 = 2e e y y y
where n is defined in (14.8). (Note that the signs in (15.18) differ from those in (14.22) because
of the sign difference in the definition of (14.10) and (15.6).)
To impose HS field equations one has to set E and E to zero. The structure of H12 bos ( ) suggests that this can be done in a variety of equivalent ways by setting to zero any of the components
of the curvature 2-forms that contain E and E . All differently looking field equations imposed
by setting to zero different components of the curvatures proportional to E and E are equivalent
by virtue Bianchi identities along with the grade zero field equation R0 = 0, G(R0 ) = 0.
522
Let stress that the cohomology analysis in Sp(4, R) is identical to that in M4 because the
form of at any given point does not change. This means in particular that one has as many
symmetries, dynamical fields and field equations in Sp(4, R) as in M4 . The precise form of
the field equations may of course be different because of the appearance of the + operators as
discussed in Section 15.2.
16. Conclusion
The main result of this paper is that free equations for massless fields of all spins in AdS4
admit sp(8, R) covariant formulation not only in terms of gauge invariant field strengths [3],
but also in terms of the gauge potentials. The key point is that the formulation is well-defined
in the AdS4 background but experiences certain degeneracy in the flat limit that does not allow
sp(8, R) and conformal invariant formulations of spin s > 1 gauge fields in flat Minkowski space.
There are two alternatives for the flat limit procedure. One leads to standard flat space massless
field equations but blowing up special conformal symmetry transformations. Another one keeps
the conformal transformation well defined, but the limiting flat space field equations is hard to
interpret.
The formulation in terms of gauge potentials is needed to reach the sp(8, R) covariant formulation at the action level and/or at the interaction level (even on-shell). The obtained results
provide the starting point for the sp(8, R) covariant study of the interacting HS theory. That
sp(8, R) symmetry may play a role in the nonlinear HS theory has been already observed in [3]
in the relation with the doubling of auxiliary spinor variables in the nonlinear HS field equations
of [5,28]. However, the formalism developed in this paper is different from that of [5,28] because
physical degrees of freedom are described here by Fock modules rather than by twisted adjoint
module as in [5,28]. In fact, the proposed formulation is close to that of the 2d HS model of [75]
that was also formulated in terms of Fock modules. It is tempting to extend this analogy to the
sp(8, R) invariant formulation of 4d HS theory.
An interesting direction for the future investigation is to study the models in the matrix spaces
MM with M > 4. Note that the dynamics of 6d and 10d conformal free massless fields in terms
of gauge invariant field strengths has been understood [2,3,8,16] in the matrix spaces M8 and
M16 , respectively. The results of this paper indicate how this theory can be extended to the level
of gauge fields, although the details remain to be elaborated.
Surprisingly, the obtained results may even shed some light on the structure of conventional
field-theoretical models like gravity. In particular, we have shown that linearized gravity in AdS4
exhibits the gl(4, R) o(3, 3) symmetry. It would be interesting to see whether and how the
gl(4, R) symmetry extends to the nonlinear gravity in AdS4 . A promising starting point is the
MacDowellMansouri action [76].
An interesting feature of the proposed model is that it has manifest EM duality symmetry
along with its HS generalization as the u(1) part of sp(8, R). In the unreduced model, that describes two infinite sets of massless fields of all spins, this symmetry rotates two species of a spin
s gauge field as a complex field. The reduction to the system of massless fields in which every
spin appears once also respects the duality transformation because it still has two sets of the
gauge fields that are dual to one another by virtue of the field equations. Hopefully, the proposed
formulation may be helpful for the further analysis of duality in the models that contain gravity
like those studied in [7782].
The main tool for the study of HS fields applied in this paper is the unfolded formulation
which is a covariant first-order reformulation of a dynamical theory in any dimension [40,49].
523
The unfolded formulation is perfectly suited for elucidation of symmetries, dynamical content
and equivalent formulations of a theory, including those in extended (super)spaces with extra
(super)coordinates. In this paper, the unfolded formulation insures the sp(8, R) invariance, determines the precise form of sp(8, R) field transformations and allows the straightforward extension
of the sp(8, R) invariant HS gauge theory in AdS4 to the ten-dimensional spacetime with the
coordinates X AB (A, B = 1, . . . , 4), which is the group manifold Sp(4, R) in the AdS like case.
The original ten-dimensional formulation of HS theory in terms of gauge invariant Weyl
0-forms [3] was manifestly covariant under GL(4, R) transformation of the spinor indices
A, B, . . . , 1, . . . , 4 so that all coordinates X AB appeared on equal footing. The manifest symmetry of the proposed extension to the HS model formulated in terms gauge fields turns out
to be reduced to GL(2, C) GL(4, R). This happens because different sectors of fields in the
theory respect different subalgebras gl(4, R) sp(8, R) which have gl(2, C) as the maximal

common subalgebra. As a result, the spacetime coordinates X and spinning coordinates X

and X have different appearance in the theory. Nevertheless, sp(8, R) acts geometrically on
X AB and remains a symmetry of the system. Let us note that the fact that the manifest symmetry
between spacetime and spinning coordinates is lost in the full theory indicates that a generalized holonomy group in the spacetimes with matrix coordinates should be GL(2, C) rather than
GL(4, R) which case was tested in [15].
As a first step towards a new version of nonlinear HS theory it is interesting to check whether
the proposed formulation exhibits an infinite-dimensional conformal HS symmetry that contains
sp(8, R) as a finite-dimensional subalgebra. As explained in Section 5.3, one way to check this
is to extend the sp(8, R) ChevalleyEilenberg cohomology to the full HS symmetry. Technically,
this is equivalent to the conformal extension of the AdS4 analysis of [40] to the first order in the
Weyl 0-forms, that accounts all HS 1-form connections.
More generally, a nonlinear extension of the free HS theory formulated in this paper can
go far beyond the original 4d HS gauge theory we started with just because it will allow a
formulation in the ten-dimensional spacetime. Actually, as observed in [13], different types of
sp(8, R) invariant fields in MM are visualized as usual fields that live in spacetimes of different
dimensions 1 10 so that the resulting theory may provide a dynamical theory of different types
of branes in the ten-dimensional spacetime. Let us note that, for higher rank (brane) solutions,
sp(8, R) extends to higher sp(2n , R) including sp(32, R) and sp(64, R) that have been argued
long ago to be symmetries of M theory [8388]. A deep relationship between HS theories and
M theory is also indicated by the recent papers [20,89].
Acknowledgements
This research was supported in part by INTAS Grant No. 03-51-6346, RFBR Grant No. 05-0217654. LSS No. 44012006.2. The author is grateful to J. de Azcarraga, I. Bandos, E. Skvortsov,
D. Sorokin, K. Stelle and S. Theisen for useful discussions. Also I am grateful to M. Eastwood
for hospitality at IMA during the 2006 summer programme on Symmetries and Overdetermined
Systems of Partial Differential Equations where this work was initiated and to H. Nicolai and
M. Staudacher for hospitality at AEI Max Planck Institute, Golm, where most of this work has
been done.
References
[1] C. Fronsdal, Massless particles, ortosymplectic symmetry and another type of KaluzaKlein theory, preprint
UCLA/85/TEP/10, in: Essays on Supersymmetry, in: Mathematical Physics Studies, vol. 8, Reidel, 1986.
524
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
I. Bandos, J. Lukierski, D. Sorokin, Phys. Rev. D 61 (2000) 045002, hep-th/9904109.

M.A. Vasiliev, Phys. Rev. D 66 (2002) 066006, hep-th/0106149.
M.A. Vasiliev, Phys. Lett. B 243 (1990) 378.
M. Flato, C. Fronsdal, Lett. Math. Phys. 2 (1978) 421;
M. Flato, C. Fronsdal, Phys. Lett. B 97 (1980) 236.
S.E. Konstein, M.A. Vasiliev, Nucl. Phys. B 312 (1989) 402;
S.E. Konstein, M.A. Vasiliev, Nucl. Phys. B 331 (1990) 475.
M.A. Vasiliev, Relativity, causality, locality, quantization and duality in the sp(2m) invariant generalized space
time, hep-th/0111119;
M. Olshanetsky, A. Vainshtein (Eds.), Contribution to the Marinovs Memorial Volume, World Scientific, 2002.
I.A. Bandos, Phys. Lett. B 558 (2003) 197, hep-th/0208110.
V.E. Didenko, M.A. Vasiliev, J. Math. Phys. 45 (2004) 197, hep-th/0301054.
M. Plyushchay, D. Sorokin, M. Tsulaia, JHEP 0304 (2003) 013, hep-th/0301067;
M. Plyushchay, D. Sorokin, M. Tsulaia, GL flatness of OSp(12n) and higher spin field theory from dynamics in
tensorial spaces, hep-th/0310297.
M.A. Vasiliev, Higher-spin theories and Sp(2M) invariant spacetime, hep-th/0301235.
O.A. Gelfond, M.A. Vasiliev, Theor. Math. Phys. 145 (2005) 35, hep-th/0304020.
A.A. Zheltukhin, D.V. Uvarov, Phys. Lett. B 565 (2003) 229, hep-th/0304151.
I. Bandos, P. Pasti, D. Sorokin, M. Tonin, JHEP 0411 (2004) 023, hep-th/0407180.
I. Bandos, X. Bekaert, J.A. de Azcarraga, D. Sorokin, M. Tsulaia, JHEP 0505 (2005) 031, hep-th/0501113.
E. Ivanov, J. Lukierski, Phys. Lett. B 624 (2005) 304, hep-th/0505216.
S. Fedoruk, E. Ivanov, Class. Quantum Grav. 23 (2006) 5195, hep-th/0604111.
S. Fedoruk, E. Ivanov, J. Lukierski, Phys. Lett. B 641 (2006) 226, hep-th/0606053.
P. West, Phys. Lett. B 650 (2007) 197, hep-th/0701026.
E. Ivanov, Nonlinear realization in tensorial superspaces and higher spins, hep-th/0703056.
A.K. Bengtsson, I. Bengtsson, L. Brink, Nucl. Phys. B 227 (1983) 31;
A.K. Bengtsson, I. Bengtsson, L. Brink, Nucl. Phys. B 227 (1983) 41.
F.A. Berends, G.J. Burgers, H. van Dam, Z. Phys. C 24 (1984) 247;
F.A. Berends, G.J. Burgers, H. van Dam, Nucl. Phys. B 260 (1985) 295.
A.K.H. Bengtsson, I. Bengtsson, Class. Quantum Grav. 3 (1986) 927.
E.S. Fradkin, M.A. Vasiliev, Phys. Lett. B 189 (1987) 89;
E.S. Fradkin, M.A. Vasiliev, Nucl. Phys. B 291 (1987) 141.
M.A. Vasiliev, Phys. Lett. B 567 (2003) 139, hep-th/0304049.
X. Bekaert, N. Boulanger, S. Cnockaert, JHEP 0601 (2006) 052, hep-th/0508048.
M.A. Vasiliev, Higher spin gauge theories: Star-product and AdS space, in: M.A. Shifman (Ed.), The Many Faces
of the Superworld, World Scientific, 2000, hep-th/9910096.
X. Bekaert, S. Cnockaert, C. Iazeolla, M.A. Vasiliev, Nonlinear higher spin theories in various dimensions, in:
Proceedings of the First Solvay Workshop on Higher-Spin Gauge Theories, Brussels, May 2004, hep-th/0503128.
A. Sagnotti, E. Sezgin, P. Sundell, On higher spins with a strong Sp(2, R) condition, in: Proceedings of the First
Solvay Workshop on Higher-Spin Gauge Theories, Brussels, May 2004, hep-th/0501156.
P.A.M. Dirac, Ann. Math. 37 (1936) 429.
R. Marnelius, Phys. Rev. D 20 (1979) 2091.
W. Siegel, Int. J. Mod. Phys. A 4 (1989) 2015.
R.R. Metsaev, Mod. Phys. Lett. A 10 (1995) 1719.
E.S. Fradkin, A.A. Tseytlin, Phys. Rep. 119 (1985) 233.
S. Ferrara, C. Fronsdal, Conformal fields in higher dimensions, hep-th/0011215.
P. Arvidsson, R. Marnelius, Conformal theories including conformal gravity as gauge theories on hypercone, hepth/0612060.
O.V. Shaynkman, I.Yu. Tipunin, M.A. Vasiliev, Rev. Math. Phys. 18 (2006) 823, hep-th/0401086.
M.A. Vasiliev, Ann. Phys. (N.Y.) 190 (1989) 59.
Y. Eisenberg, Phys. Lett. B 225 (1989) 95.
M.A. Vasiliev, Sov. J. Nucl. Phys. 32 (1980) 855 (p. 439 in English translation).
M.A. Vasiliev, Fortschr. Phys. 35 (1987) 741.
C. Fronsdal, Phys. Rev. D 18 (1978) 3624;
C. Fronsdal, Phys. Rev. D 20 (1979) 848.
525
[45] J. Fang, C. Fronsdal, Phys. Rev. D 18 (1978) 3630;

J. Fang, C. Fronsdal, Phys. Rev. D 22 (1980) 1361.
[46] E. Sezgin, P. Sundell, JHEP 0207 (2002) 055.
[47] E.S. Fradkin, M.A. Vasiliev, Mod. Phys. Lett. A 3 (1988) 2983.
[48] M.A. Vasiliev, Nucl. Phys. B 307 (1988) 319.
[49] M.A. Vasiliev, Class. Quantum Grav. 11 (1994) 649.
[50] D. Sullivan, Publ. Math. IHES 47 (1977) 269.
[51] R. DAuria, P. Fre, Nucl. Phys. B 201 (1982) 101;
R. DAuria, P. Fre, Nucl. Phys. B 206 (1982) 496, Erratum.
[52] P. van Nieuwenhuizen, Free Graded Differential Superalgebras, in: M. Serdaroglu, E. Inn (Eds.), Group Theoretical Methods in Physics: Proceedings, in: Lecture Notes in Physics, vol. 180, Springer-Verlag, 1983.
[53] R. DAuria, P. Fre, P.K. Townsend, P. van Nieuwenhuizen, Ann. Phys. 155 (1984) 423.
[54] P. van Nieuwenhuizen, Phys. Rep. 68 (1981) 189;
P. van Nieuwenhuizen, Supergravity as a YangMills theory, in: G. t Hooft (Ed.), 50 Years of YangMills Theory,
World Scientific, hep-th/0408137.
[55] O.V. Shaynkman, M.A. Vasiliev, Theor. Math. Phys. 123 (2000) 683, Teor. Mat. Fiz. 123 (2000) 323, hepth/0003123.
[56] M.A. Vasiliev, Int. J. Geom. Methods Mod. Phys. 3 (2006) 37, hep-th/0504090.
[57] E.S. Fradkin, M.A. Vasiliev, Dokl. Acad. Nauk 29 (1986) 1100;
E.S. Fradkin, M.A. Vasiliev, Ann. Phys. 177 (1987) 63.
[58] M.A. Vasiliev, Fortschr. Phys. 36 (1988) 33.
[59] A.S. Matveev, M.A. Vasiliev, Phys. Lett. B 609 (2005) 157, hep-th/0410249.
[60] J. Engquist, E. Sezgin, P. Sundell, Nucl. Phys. B 664 (2003) 439, hep-th/0211113.
[61] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hep-th/9610043.
[62] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, Nucl. Phys. B 498 (1997) 467, hep-th/9612115.
[63] L. Susskind, Another conjecture about M(atrix) theory, hep-th/9704080.
[64] M. Hanada, H. Kawai, Y. Kmura, Prog. Theor. Phys. 114 (2006) 1295, hep-th/0508211;
M. Hanada, H. Kawai, Y. Kmura, Prog. Theor. Phys. 115 (2006) 1003, hep-th/0602210.
[65] T. Saitou, JHEP 0606 (2006) 010, hep-th/0604103;
T. Saitou, JHEP 0707 (2007) 057, arXiv: 0704.2449 [hep-th].
[66] P.J. Olver, Applications of Lie Groups to Differential Equations, Springer Verlag, New York, 1993.
[67] G. Barnich, M. Grigoriev, RST extension of the non-linear unfolded formalism, hep-th/0504119.
[68] T. Lada, J. Stasheff, Int. J. Theor. Phys. 32 (1993) 1087, hep-th/9209099.
[69] M. Kaku, P.K. Townsend, P. Van Nieuwenhuizen, Phys. Lett. B 69 (1977) 304;
M. Kaku, P.K. Townsend, P. Van Nieuwenhuizen, Phys. Rev. Lett. B 39 (1977) 1109;
M. Kaku, P.K. Townsend, P. Van Nieuwenhuizen, Phys. Rev. D 17 (1978) 3179.
[70] I. Bandos, J. Lukierski, C. Preitschopf, D. Sorokin, Phys. Rev. D 61 (2000) 065009, hep-th/9907113.
[71] C. Cutler, R. Wald, Class. Quantum Grav. 4 (1987) 1267.
[72] N. Boulanger, T. Damour, L. Gaultieri, M. Henneaux, Nucl. Phys. B 597 (2001) 127, hep-th/0007220.
[73] C. Aragone, S. Deser, Phys. Lett. B 86 (1979) 161.
[74] F.A. Berends, J.W. van Holten, B. de Wit, P. van Nieuwenhuizen, J. Phys. A 13 (1980) 1643.
[75] M.A. Vasiliev, Phys. Lett. B 363 (1995) 51.
[76] S.W. MacDowell, F. Mansouri, Phys. Rev. Lett. 38 (1977) 739;
S.W. MacDowell, F. Mansouri, Phys. Rev. Lett. 38 (1977) 1376, Erratum;
F. Mansouri, Phys. Rev. D 16 (1977) 2456.
[77] S. Deser, C. Teitelboim, Phys. Rev. D 13 (1976) 1592.
[78] J.A. Nieto, Phys. Lett. A 262 (1999) 274, hep-th/9910049.
[79] M. Henneaux, C. Teitelboim, Phys. Rev. D 71 (2005) 024018, gr-qc/0408101.
[80] B. Julia, J. Levie, S. Ray, JHEP 0511 (2005) 025, hep-th/0507262.
[81] S. Deser, D. Seminara, Phys. Lett. B 607 (2005) 317, hep-th/0411169;
S. Deser, D. Seminara, Phys. Rev. D 71 (2005) 081502, hep-th/0503030.
[82] R.G. Leigh, A.C. Petkou, Gravitational duality transformations on (A)dS4 , arXiv: 0704.0531 [hep-th].
[83] P.K. Townsend, hep-th/9507048;
P.K. Townsend, hep-th/9712004.
[84] I. Bars, Phys. Rev. D 54 (1996) 5203, hep-th/9604139;
I. Bars, Phys. Rev. D 55 (1997) 2373, hep-th/9607112;
I. Bars, Phys. Lett. B 403 (1997) 257, hep-th/9704154.
526
[85]
[86]
[87]
[88]
[89]
M. Gunaydin, D. Minic, Nucl. Phys. B 523 (1998) 145, hep-th/9802047.

M. Gunaydin, Nucl. Phys. B 528 (1998) 432, hep-th/9803138.
S. Ferrara, M. Porrati, Phys. Lett. B 458 (1999) 43, hep-th/9903241.
P. Horava, Phys. Rev. D 59 (1999) 046004, hep-th/9712130.
P. Horava, C. Keeler, JHEP 0706 (2007) 031, arXiv: 0704.2230 [hep-th].

Stability of String Configurations Dual To Quarkonium States in Ads/Cft

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stability of String Configurations Dual To Quarkonium States in Ads/Cft

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 793 (2008) 133

Stability of string configurations

Received 25 June 2007; accepted 24 September 2007

E-mail addresses: avramis@mail.cern.ch (S.D. Avramis), sfetsos@upatras.gr (K. Sfetsos), ksiampos@upatras.gr

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

fy (u, a ) = Gtt Gyy ,

fij (u, a ) = Gti Gtj Gtt Gij ,

fab (u, a ) = Gtt Gab ,

h(u, a ) = Gyy Guu .

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

where g and b stand for the pullbacks

To proceed, we employ the gauge fixing

we assume translational invariance along t, and we consider the radial embedding

supplemented by the boundary condition

Integrating (2.13), we express the separation length as

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

y = ycl (u) + y(t, u),

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

we obtain the equations

lim (u u0 )1/2 xi (u) = 0,

lim (u u0 )1/2 a (u) = 0.

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

the first of (3.10) as

or, equivalently, if [13]

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

we may transform Eq. (3.15) to a standard Schrdinger equation

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

with the potential

The problem is defined in the interval

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

where v is the boost velocity and = 1/ 1 v 2 . The metric has a horizon at

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

To examine the Wilson-loop computation, we switch to dimensionless units by setting

(u2 1)(u4 u40 )

u3 (u2 1)3 (u4 u40 )(u4 4 )

In order for a solution to exist, the integrand must change sign

and again can be solved numerically to give u()

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

4.2.2. One nonzero angular momentum

and d32 is as before. This metric has a horizon at

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

(u4 u40 )(u4 + u2 4 )

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

+ G||2 cos4 sin2 sin2 2(d1 + d2 + d3 )2 ,

B2 = cos4 sin 2(d1 + d2 + d3 ) d,

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

and no angular instabilities occur.

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

(0 , 0 ) = (/2, any). Here, we have

and again no angular instabilities

(0 , 0 ) = (sin1 (1/ 3), /4). Here, we have

(u2 r02 ) sin2 + u2 cos2 cos2 sin2

(u2 r02 ) sin2 + u2 cos2 cos2 sin2

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

Their behavior is as in Fig. 1(c).

S.D. Avramis et al. / Nuclear Physics B 793 (2008) 133

case, the Schrdinger variable x can be explicitly determined in terms of u as

lim (u u0 )1/2 xi (u) = 0,

lim (u u0 )1/2 a (u) = 0.

(z) = (z) (z),