Nucl - Phys.B v.762 PDF

Nuclear Physics B 762 (2007) 137
An exact solution of 4D higher-spin gauge theory

E. Sezgin a , P. Sundell b,
a George P. and Cynthia W. Mitchell Institute for Fundamental Physics, Texas A&M University,
College Station, TX 77843-4242, USA

b Department for Theoretical Physics, Uppsala University, Box 803, 751 08 Uppsala, Sweden
Received 3 January 2006; received in revised form 28 June 2006; accepted 28 June 2006
Available online 18 July 2006
Abstract
We give a one-parameter family of exact solutions to 4D higher-spin gauge theory invariant under a
deformed higher-spin extension of SO(3, 1) and parameterized by a zero-form invariant. All higher-spin
gauge fields vanish, while the metric interpolates between two asymptotically AdS4 regions via two dS3 foliated domain walls and two H3 -foliated RobertsonWalker spacetimesone in the future and one in the
pastwith the scalar field playing the role of foliation parameter. All Weyl tensors vanish, including that of
spin two. We furthermore discuss methods for constructing solutions, including deformation of solutions to
pure AdS gravity, the gauge-function approach, the perturbative treatment of (pseudo-)singular initial data
describing isometric or otherwise projected solutions, and zero-form invariants.
2006 Elsevier B.V. All rights reserved.
1. Introduction and summary

Full higher-spin gauge-field equations have been known in D = 4 for quite some time, essentially since the early work of Vasiliev [1] (see [2] for a review), and their generalizations
to higher dimensions have been started to be understood more recently in [35]. These equations are generalizations of pure AdS gravity, in which the metric is accompanied by an infinite
tower of higher-spin fields and special sets of lower-spin fields, always containing at least one
real scalar. In the minimal setting, the physical fields are symmetric doubly traceless Lorentz
tensors 1 ...s of rank s = 0, 2, 4, . . . . The generally covariant form of the equations [6] (with
non-linearly treated metric ds 2 = dx dx g accommodating the rank-two tensor) is highly
* Corresponding author.
E-mail address: per.sundell@teorfys.uu.se (P. Sundell).

0550-3213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.06.038
E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137
non-local with higher-derivative corrections normalized by the AdS mass-parameter [1,2]. The
expansion around the AdS vacuum yields, however, a tachyon and ghost-free spectrum of massless particles governed by Fronsdals equations, with all non-localities occurring via interactions.
In other words, the classical higher-spin gauge theory is weakly coupled in field amplitudes
and strongly coupled in wave-numbers. This state of affairswhich refers to a classical theory
although it appears reminiscent to that of a quantum-effective theoryblurs many basic fieldtheoretic concepts, even at the level of the lower-spin self-couplings, which, for example, no
longer exhibit any well-defined notion of microscopic scalar-field potential [7] or local stress
energy tensor [8]. The non-localities are, of course, not put in by hand, nor by integrating out
microscopic fields. Instead, the complete form of the interactions are governedsomewhat in
the spirit of classical string field theoryby a manifestly higher-spin covariant master-field formulation [1,2] (without ambiguities in case manifest Lorentz invariance [2] and parity invariance
[7,9] is required) based on simple non-commutative twistor variables [1] (see also [1015] for
higher-dimensional generalizations albeit at the free level) related to deformation quantization of
singletons [16,17].
The master-field formulation, which is presented in Section 2, is amenable to finding classical
solutions, while the non-localities make these difficult to interpret using traditional field-theoretic
and geometric tools. Clearly, the finding of exact solutions is therefore highly desirable, not only
for the important role that they may play in uncovering novel physical phenomena, but also
for developing the interpretational language as well as shedding new light on the origin of the
higher-spin field equations themselves in tensionless limits of string theory [1722].
It is primarily with the above motivations in mind that we have sought and found an exact
solution beyond AdS spacetime. In doing so, we have greatly benefited from the work of [2,
23,24] on how to use the master-field formulation to construct solutions in ordinary spacetime.
Exact, massively deformed, 3D vacua are constructed in [23] using techniques for deforming
twistor oscillators, and plane-wave solutions for the free 4D field equations are provided in [24]
using a gauge-function approach, in its turn related to a more general method described in [2,23]
pertinent to finding exact solutions in a systematic manner by means of integrating flows. We
shall relate further to these works when we discuss methods for finding solutions in Section 3.
The solution, which is presented in detail in Section 4, is constructed by exploiting the simplifications taking place at the full master-field level by imposing SO(3, 1)-invariance (without
having to resort to weak-field expansion). As a result, it is globally well-defined on a manifold
with the same topology as AdS4 , covered by two charts with local stereographic coordinates x a
and x a (restricted by 2 x 2 , 2 x 2 < 1 and related by x a = x a /(2 x 2 ) in the overlap region
2 x 2 , 2 x 2 < 0). In the type A model (see Section 2.1), the solution is given locally in the x a chart by
(x) =

1 2 x 2 ,
b1
1 ...s = 0,
ds 2 =
for s = 4, 6, . . . ,
4 2 (d(g1 x))2
,
(1 2 g12 x 2 )2
ab
ab = f (0)
,
(1.1)
(1.2)
where the scale factors (x 2 ; ), g1 (x 2 ; ) and f (x 2 ; ) are given explicitly in (4.63) and (4.66),
and is the inverse radius of the AdS4 vacuum at = 0, where (x 2 ; 0) = f (x 2 ; 0) =
g1 (x 2 ; 0) = 1.
Remarkably, the higher-spin fields vanish, notwithstanding the fact that lower spins in general
source higher spins, so that the field equations can only be truncated to the spin s 2 sector for
very special configurations. In other words, the solution satisfies a highly non-local extension
of scalar-coupled AdS gravity (corresponding to an effective model given in [25]) as well as an

infinite set of consistency conditions for setting the higher-spin fields to zero. These conditions
reflect the invariance under an infinite-dimensional higher-spin extension of SO(3, 1), that we
denote by
hsl(2, C; ) sl(2, C),
(1.3)
with -dependent generators given in (4.71) and (4.72). This motivates seeking solutions based
on invariance under other subgroups of SO(3, 2), which we shall analyze at the linearized level
in Section 5.
The locally defined scalar and metric in (1.1) are components of a section of the higher-spin
gauge-covariant master fields, related to the section in the x a -chart by a gauge transition defined
in the overlap. The physical component fields are thus related by more complicated duality
transformations than standard reparameterizations. As a result, by the Z2 -symmetry, they read
the same in x a and x a -coordinates, and hence the scalar-field duality transformation (in this
particular background) assumes the form of a fractional linear transformation, viz.
x)
(
=
(x)
.
(x)
(1.4)
The global solution therefore remains weakly coupled throughout spacetime, and describes a
Weyl-flat interpolation between two asymptotically AdS4 regions 2 x 2 1 and 2 x 2 1, via
two dS3 -foliated domain walls in 0 < 2 x 2 , 2 x 2 < 1 and two FRW spacetimes with k = 1 in
the overlap region (one in the future and one in the past).
As we shall discuss in Section 6, the cosmological interpretation of our solution requires the
development of geometric and algebraic tools suitable for the description of higher gauge theory.
Some of the properties of the solution aiming at such an interpretation are reported in [25].
2. The minimal bosonic model
2.1. The master-field equations
To describe the higher-spin gauge theory based on the minimal higher-spin algebra hs(4)
SO(3, 2), one introduces a set of auxiliary coordinates (z , z ) together with an additional set of
internal variables (y , y ), that are Grassmann-even SL(2, C)-spinor oscillators defined by the
associative product rules
y y = y y + i
,
y z = y z i
,
(2.1)
z y = z y + i
,
z z = z z i
,
(2.2)
where the juxtaposition denotes the symmetrized, i.e., Weyl-ordered, products. The hermitian
conjugates (y ) = y and (z ) = z obey
y y = y y + i
,
z y = z y i
,
(2.3)
y z = y z + i
,
z z = z z i
.
(2.4)
Equivalently, in terms of Weyl-ordered functions
f(y, y,
z, z ) g(y,
y,
z, z )
2 2 2 2
d d d d i +i
=
e
(2)4
f(y + , y + , z + , z )g(y
+ , y + ,
z , z + ),
(2.5)
where the hats are used to indicate functions that depend on both (y, y)
and (z, z ), while functions
depending only on (y, y)
shall be written without hats. The basic master fields are differential
forms in an extended spacetime with coordinates (x M , z , z ), namely a one-form
+ dz A (x, z, z ; y, y)
+ d z A (x, z, z ; y, y),
A = dx M A M (x, z, z ; y, y)
(2.6)
and a zero-form = (x,

z, z ; y, y).
The full higher-spin equations based on the above algebraic structures were first given in [1].
It can be shown that parity invariance and manifest Lorentz invariance restricts the possible
interactions to two cases referred to as the minimal type A and type B models, in which the
scalar field is even and odd under parity, respectively [7,9]. The resulting master-field equations
read
i
i
F = b1 dz dz + (b1 ) d z d z ,
(2.7)
4
4
D = 0,
(2.8)
where b1 = 1 and b1 = i in the type A and type B models, respectively, and the curvatures are
defined as
F = d A + A A,
]
,
D = d + [A,
(2.9)
with
[f, g]
= f g g (f),
(2.10)
and the functions and defined by

= = exp i y z ,
= exp iy z ,
(2.11)
have the salient properties

f(y, z) = f(z, y),
f = (f).
f(y, z) = f(z, y),
(2.12)
(2.13)
In order to restrict to the minimal bosonic model one imposes further kinematic conditions
= A,
(A)
A = A,
= (),
()
where is the anti-automorphism

f(y, y,
z, z ) = f(iy, i y,
iz, i z ),
= (),
(f g)
= (g)
(f),
and is the involutive automorphism

f(y, y,
z, z ) = f(y, y,
z, z ) = f ,
(f g)
= (f) (g),
(2.14)
(2.15)
(2.16)
The gauge transformations are given by
A = D
,
= [
, ]
(2.17)
The rigid, i.e., x and Z-independent, gauge parameters form the Lie algebra

hs(4) = P (y, y):
(P ) = P = P ,
(2.18)
whose maximal finite-dimensional subalgebra is given by SO(3, 2), with generators (A.3).
To analyze the master-field equations (2.7) and (2.9) formally, one starts from an initial
condition
AM (x; y, y)
= A M |Z=0 ,
Z=0 ,
(x; y, y)
= |
(2.19)
and a suitable gauge condition on the internal connection, such as [6]

A |=0 = 0,
(2.20)
and proceeds by obtaining the Z-dependence of the fields by integrating the internal constraints, viz.
ib1
,
F =
2
perturbatively in a -expansion, denoted by
D = 0,
= +
A =
(n) ,
n=2
A (n)
,
F = 0,
FM = 0,
A M = AM +
n=1
(2.21)
(n)
A M .
(2.22)
n=1
The x-space constraints FMN = 0 and D M = 0 can then be shown to be perturbatively equiv Z=0 = 0, that is
alent to FMN |Z=0 = 0 and DM |
FMN =

(m)
(n)
A M , A N ,
p=1 m+n=p
DM =

(m)
A M , (n)
(2.23)
p=2 m+n=p
where FMN = 2[M AN] + [AM , AN ] and DM = M + [AM , ] . These equations constitute a perturbatively Cartan-integrable system in x-space provided that the full Z-dependence is
included at each order in .
Since (2.23) are written entirely in terms of differential forms they are manifestly diffeomorphism invariant. In fact, they are invariant under homotopy transformations, whereby coordinate directions can be added and removed without affecting the physical content. Thus,
in case x-space is homotopic to a four-dimensional spacetime manifold with coordinates x
( = 0, 1, 2, 3), then one can without loss of generality formulate (2.23) directly on this fourmanifold.
2.2. The spacetime field equations
In order to obtain the physical field equations on generally covariant form, one first has to
Lorentz covariantize (2.23). To this end, one first identifies the full Lorentz generators acting on
the hatted master fields as follows1
1
(0)
M = M + {S , S } ,
(2.24)
2
1 Under -commutation the full generators close schematically as [M, M] M + M.
The extra term proportional

to the field variation does not affect our solutions. We thank C. Iazeolla for pointing out this extra term.
where M = y y z z , and
(0)
S = z 2i A .
(2.25)
One can show that [2]

= [
0 , ]
,
L [
L , ]
L A D
L = [
0 , A ] + A ,

1
L A D
L = [
0 , A ] +
M h.c. ,
4i
(2.26)
(2.27)
(2.28)
(0)
where
L = 4i1 (x)M (h.c.) are the full parameters, and
0 = 4i1 M (h.c.) are
the parameters of canonical Lorentz transformations of the Y and Z oscillators. The canonically transforming component fields are thus obtained by Y and Z-expansion of A , and
A ( 4i1 M h.c.) where is the Lorentz connection with L = +
+ (related conventions are given in Appendix A). Hence, introducing
1
M h.c., M = y y ,
4i
1
K = (M M ) h.c.,
4i
(2.29)
(2.30)
and using the gauge condition (2.20) to simplify K = K |Z=0 , the Lorentz covariant decomposition of the master-gauge field A reads
A = e + + W + K ,

h.c.,
K = i (A A )
Z=0
(2.31)
(2.32)
where e is the vielbein2

1
(2.33)
e y y , e = (a ) e a ,
2i
2
contains the higher-spin gauge fields. Indeed, inserting (2.31) into (2.23), the
and W = W (y, y)
explicit Lorentz connections cancel, and one obtains a manifestly Lorentz covariant and spacetime diffeomorphism invariant set of constraints [2,6].
To describe the higher-spin gauge symmetries in a generally spacetime covariant fashion, one
proceeds using a weak-field approximation in which the higher-spin gauge fields, the scalar field
and the Weyl tensors, including that of spin 2, are treated as weak fields, while no approximation
Z=0 = 0
is made for the vielbein and the Lorentz connection. It can then be shown that D |
contains the field equation for the physical scalar field
e =
= |Y =0 ,
(2.34)
and that F |Z=0 = 0 contains the field equation for the metric
g = e a ea = 22 e e ,
(2.35)
2 The vielbein here is given in the higher-spin frame, where the torsion is non-vanishing. A discussion of the Einstein
frames is given in [25].
and a set of physical higher-spin fields given by the doubly-traceless symmetric tensor fields of
rank s = 4, 6, . . . , given by

s1
s1
s1 s1
1 1

e
W
.
1 2 ...s = 2ie(
(2.36)
)
s
s1

1
y 1 y s1 y 1 y s1
Y =0
The generally covariant physical field equations are given up to second order in weak fields by

2
i
2
(2)
(2)
P
,
+ 2 = P
(2.37)
2
y y Y =0

(2)
(2.38)
R =
J
,
y y
Y =0

( F2 ...s1 ) 1 ... s1
1

(2)
= (
J
,
(2.39)
1
y 2
y s1 ) Y =0
y s1 ) y
(2)
(2)
where the source terms P and J are provided in Appendix C; R , which is defined
in (A.7), is the self-dual part of the full SO(3, 2)-valued curvature d + with given
in (A.6); the higher-spin curvatures are defined by (s = 4, 6, 8, . . .)
F2 ...s1 2 ... s1
= 2[ W]2 ...s1 2 ... s1 (s 2)( )(2 d W3 ...s1 ) 2 ... s1

s( )( W 2 3 ...s1 2 ... s1 ) ;
(2.40)
and the covariant derivatives in (2.37) and (2.40) are given by = d + with being the canonical Lorentz connection. The expression (2.40) contains the auxiliary gauge fields W(s2)(s)
and W(s3)(s+1)
, of which the latter drops out from (2.39), while the former can be expressed
as exexplicitly in terms of the physical fields using curvature-dressed covariant derivatives ,

plained in [6].
3. Construction of solutions
3.1. Perturbative spacetime approach
In the first order of the weak-field expansion, it is consistent to truncate the higher-spin field
equations to that of pure Einstein gravity with cosmological constant plus a free scalar field in a
fixed background metric, viz.

2
+ 22 = 0.
R + 32 g = 0,
(3.1)
These equations are self-consistent, though they do not derive from an action. They provide a
the spin-2 Weyl tensor and all their derivatives are small. In
good approximation if ,
general, such solutions may describe spacetimes that are not asymptotically conformally flat in
any region.
The higher-order corrections from the weak-field expansion yields a perturbative expansion
in and of the form
g = g +

n=2
(n)
g
,
= +

n=2
(2) ,
W =

n=2
W(n) .
(3.2)
The second-order corrections can be determined from (2.37)(2.39), where the higher-spin gauge
(2)
(2)
fields and Weyl tensors do not contribute to P and J to that order. In this sense, exact solutions to ordinary Einstein gravity with cosmological constant may be embedded into higher-spin
gauge theory, albeit that finding the exact solutions in closed form amenable to studies of salient
properties is highly non-trivial. Moreover, there are subtleties related to boundary conditions as
well as the convergence of the perturbative expansion.
The higher-spin field equation (2.39) reads (s = 4, 6, . . .)

F ...s = O (weak fields)2 ,
(3.3)
1
where the Fronsdal-like operator F 1 ...s is covariantized using the background metric g . The
first-order truncation of (3.3) is not self-consistent, i.e., incompatible with linearized higher-spin
gauge symmetry, unless = 0, i.e., g = g(0) where the subscript (0) denotes the AdS4
background with curvature given by (A.12). Thus, the proper way to include higher-spin fields
1 ...s into the first order is to solve
F ...s = 0,
(3.4)
1
including all gauge artifacts, and impose the gauge conditions order-by-order in weak-field expansion using the fully consistent higher-spin field equation (2.39).
In order to switch on higher-spin fields, it is therefore convenient to consider solutions in
which the full Weyl zero-form asymptotes to zero in some region of spacetime. We shall refer
to such solutions as asymptotically Weyl-flat solutions.
The perturbative approach to these solutions is self-consistent in the sense that the linearized
twisted-adjoint zero-form C = C(x; y, y)
obeying dC + [(0) , C] = 0, where (0) is the AdS4
connection, vanishes on the boundary. To demonstrate this, one writes the linearized equation as
(0) C + e a {Pa , C} = 0,

(3.5)
2i
and expands C and (3.5) in y and y,
which shows that C contains linearized Weyl tensors
Ca(s),b(s) obeying (s = 2, 4, . . .)
a
b
c
e(0)
e(0)
(0)a Cb(s1),c(s1) = 0,
e(0)
(0) C(s1),(s) = 0.
(3.6)
These equations are in fact valid in AdSD . The resulting mass-shell condition reads
2

(0) + 2(s + D 3)2 C(s),(s) = 0.
(3.7)
Setting s = 0 one obtains formally the correct scalar-field equation,

2

(0) + 2(D 3)2 = 0,
(3.8)
where = C|Y =0 in D = 4. Splitting x (x i , r), and using Poincar coordinates,

1
2
ds(0)
= 2 2 dr 2 + dx 2 ,
r
1
1 j
1
j
r
r
(0)ir = i ,
(0)rr
= ,
(0)ij = ij ,
(3.9)
r
r
r
one finds that the component fields Ci(s),j (t)r(st) with s t 1, and where the indices are
curved, are given by curls of Ci(s),r(s) , taken using rj , that in their turn obey

2D
2s
s 2 + (D 1)s + 2(D 3)
ij
+
r +
+
r +
i j Ci(s),r(s) = 0.
r
r
r2
(3.10)
9
()
Thus, the linearized spin-s Weyl tensor Ca(s),b(s) consists of two sectors Ca(s),b(s) with scaling
behavior given by (s = 0, 2, 4, . . .)
D1 D5
.
(3.11)
2
2
At higher orders of the weak-field expansion, and due to the higher-derivative interactions
Z=0 = 0, the corrections to the spin-s Weyl tensor may
hidden in the -products in D |
in principle contain lower-spin constructs with a total scaling weight less than s . We shall
not analyze the nature of these corrections in any further detail here, but hope to return to this
interesting issue in a future work.
Another non-local effect induced via the Z-space dependence, is that the Lorentz covariantizations in K , defined in (2.32), may remain finite in the asymptotic region, despite the naive
expectation that the scaling of (A ( A ) )|Z=0 , which is of order 2 , should over-power that of
. While this is a challenging problem to address in its generality, we shall find that already
the relatively simple case of the SO(3, 1)-invariant asymptotically Weyl-flat solution exhibits
an interesting phenomenon whereby K generates a finite Weyl rescaling and contorsion in the
asymptotic region.
Clearly, the weak-field expansion, which is naturally geared towards dressing up solutions
to (3.1), is going to be far from efficient in dealing with general asymptotically Weyl-flat solutions. Especially when the starting point is not a solution to (3.1), it is appropriate to develop
an alternative approach to solving the basic master-field equations (2.7) and (2.9) in which one
makes a maximum use of the fact that the local x-dependence is a gauge choice. As we shall see
next, the Z-space approach indeed does exploit this fact, and provides a powerful framework for
finding exact solutions to higher-spin field equations.
()
r s ,
Ca(s),b(s)
s = s +
3.2. The Z-space approach

In order to construct solutions one may consider the Z-space approach [24] in which the
constraints in spacetime, viz.
F = 0,
F = 0,
D = 0,
(3.12)
are integrated in simply connected spacetime regions given the spacetime zero-forms at a point p,
p,

= |
A
= A |p ,
(3.13)
and expressed explicitly as
A = L 1 L,
A = L 1 (A
+ )L,
= L 1
(L),
(3.14)
where L = L(x,
z, z ; y, y)
is a gauge function, and
p = 1,
L|
A
= 0,

= 0.
(3.15)
The remaining constraints in Z-space, viz.

ib1
2[ A
] + [A
, A
] =

,
F
2
F
A
A
+ [A
, A
] = 0,
D

+ A

+
(A
) = 0,
(3.16)
(3.17)
(3.18)
10
must then be solved given an initial condition

C
(y, y)
=
|Z=0 ,
(3.19)
and fixing a suitable gauge for the internal connection. The natural choice is
A
|C
=0 = 0,
(3.20)
whose compatibility with (2.20) requires

C
=0 = L(x; y, y),
L|
(3.21)
that is, the gauge function cannot depend explicitly on the Z-space coordinates. In what follows,
we shall assume that
L = L(x; y, y).
(3.22)
The gauge fields can then be obtained from (2.31), viz.

e + + W = L1 L K ,
(3.23)
where
K = K |Z=0 ,
K = i L1 A
A
L.
(3.24)
Hence, the gauge fields, including the metric, can be obtained algebraically without having to
solve any differential equations in spacetime.
The local representatives in two overlapping simply connected regions with gauge functions L
are related on the overlap via a gauge transformation with transition function
and L,
g = L1 L.
(3.25)
The resulting transformations of the physical fields, viz.

e + + W

= g 1 e + + W + K g g 1 K g Z=0 g 1 + g,

= g 1 g Z=0
(3.26)
which are a manifest symmetry of the full constraints in x and Z-space, are a non-trivial symmetry from the point of view of the generally covariant spacetime field equations contained
Z=0 = 0 (whose general covariance corresponds to field-dependent
in F |Z=0 = 0 and D |
gauge parameters).
To describe asymptotically Weyl-flat solutions, one chooses the gauge function to be a parameterization of the coset
L(x; y, y)

SO(3, 2)
.
SO(3, 1)
(3.27)
By construction, its MaurerCartan form

L1 dL = (0) = e(0) + (0) ,
(3.28)
11
which means that the gauge fields defined by (3.23) are given asymptotically by the AdS4 vacuum
plus corrections from K . The latter are higher order in the C
-expansion,3 but may nonetheless
contribute in the asymptotic region to the leading dependence on the radial coordinate defined
in (3.9). A concrete exemplification of this subtlety is provided by the asymptotic behavior of the
scale factors in the SO(3, 1)-invariant solution discussed at the end of Section 4.3.
3.3. On regular, singular and pseudo-singular initial conditions
The perturbative approach to (3.16)(3.18) yields a solution of the form

= C
+

n=2

(n) ,
A
=
A
(n) ,
(3.29)
n=1
where the superscript indicate the order in C

. We shall refer to C
as a regular initial condition
provided that its -product self-compositions, viz.

n
C(2n)
(3.30)
= C
(C
) ,

n
C(2n+1)
(3.31)
= C
(C
) C
,
are regular functions. The second-order corrections (C.4) and (C.5) contain the -product composition

C (tz, y)e
(3.32)
ityz C
(y, y)
= C
(ty, y)e
i(1t)yz C
(y, y),
where t is the auxiliary integration parameter used to present the first-order correction (C.3) to
the internal connection. If C
is regular, then it follows that (3.32) is a regular function of (Y, Z)
with t-dependent coefficients that are analytic at t = 1. We shall assume analyticity also at t = 0,
where (3.32) involves only anti-holomorphic contractions resulting in a softer composition
that should not blow up. Under these assumptions there exists an open contour from t = 0 to
t = 1 along which (3.32) is analytic, that can then be used in (C.3) to produce a well-defined
second order correction. This argument extends to higher orders of perturbation theory, and hence
regular initial data yields perturbative corrections that can be presented using open integration
contours [6].
If C
is not regular, and (3.32) has an isolated singularity at t = 1, we shall refer to C
as a
singular initial condition. In this case, a well-defined perturbative expansion can be obtained by
circumventing the singularity by using a closed contour as follows,

dt
t
d
(3.33)
log
Z f (tZ) = dZ f (Z),
2i
1t
and
d

dt
1
t
t log
1 Z dZ f (tZ) = dZ dZ f (Z),
2i
1t
2
(3.34)
3 Using the gauge function (4.1), the spin-s sector in C(x; y, y)

= L1 (x) C
(y, y)
L(x) scales like h2(1+s) , i.e.,
C
contains the regular boundary data scaling like r s in the notation of (3.11) with r h2 .
12
where Z = (z , z ), d
= dZ , d
f1 = d
f2 = 0 and encircles the branch cut from t = 0 to
t = 1. The resulting closed-contour presentation of the perturbative solution can of course also
be used in the case of regular initial data, in which case can be collapsed onto the branch-cut
as to reproduce the open-contour presentation.
Singular initial conditions may arise from imposing symmetry conditions on C
, as will be
exemplified in Section 5.3 in cases where the symmetry refers to unbroken spacetime isometries.
Here the initial conditions are parameterized elementary functions of the oscillators, e.g., combinations of exponentials of the bilinear translation generator Pa contracted with fixed vectors,
that become singular for particular choices of the parameters.
Another interesting type of irregular initial data arises in the five-dimensional and sevendimensional higher-spin models based on spinor oscillators [1012] and the D-dimensional
model based on vector oscillators. Here the initial conditions are regular functions multiplied
with singular projectors whose role is to gauge internal symmetries in oscillator space in order
that the master-field equations contain dynamics. These symmetries are generated by bilinear oscillator constructs, K, and the singular projectors are special functions with auxiliary
1
integral representations, given schematically by 0 ds f (s)esK , such that the analogs of the selfcompositions (3.30) and (3.31) now contain logarithmic divergencies arising in corners of the
auxiliary integration domain. Due to the projection property, these divergencies can be factored
out and written as [4]
1
0
ds
g(s),
1s
(3.35)
where g(s) is analytic and non-vanishing at s = 1. One can now argue that the perturbative
expansion gives rise to analogs of (3.32) resulting in regular functions of (Y, Z) with t -dependent
coefficients involving prefactors of the form
1
0
ds
g(s),
1 ts
(3.36)
that are logarithmically divergent at t = 1. Thus, the open-contour presentation results in welldefined second-order perturbations containing prefactors of the form
1 1
0 0
ds dt
g(s),
1 st
(3.37)
while the closed-contour presentation, which requires analyticity on closed curve encircling
t = 1, does not apply since the singularity at t = 1 is not isolated. We shall therefore refer to
initial conditions of this type as pseudo-singular initial conditions.
The above analysis indicate that the (pseudo-)singular nature of initial conditions is an artifact of the naive application of the -product algebra to (3.31), while the actual perturbative
expansion in Z-space involves a point-splitting mechanism that softens the divergencies. We
plan to return with a more conclusive report on these important issues in a forthcoming paper.
13
3.4. Zero-form curvature invariants

In unfolded dynamics, the local degrees of freedom are dual to the twisted-adjoint element
C
(y, y)
defined by (3.19), consisting of all gauge-covariant derivatives of the physical fields
evaluated at a point in spacetime. At the linearized level, C
is gauge covariant, which means that
if C
= g 1 C
(g), with g a group element generated by hs(4), then C
and C
give rise to
gauge-equivalent solutions.
To distinguish between gauge-inequivalent solutions at the full level, we propose the following
invariants
r C2p ,
C2p
= N T
(3.38)
for p = 1, 2, . . . , where

p ;
= ()
C2p =
(3.39)
2p times
the full traces are defined by

2 2 2 2
d y d y d z d z
r+ f =
f (y, y,
z, z ),
T
(2)4
r f = T
r+ (f );
T
(3.40)
and the normalizations N are given by

1
(3.41)
,
V
where V is the volume of Z-space. As separate components of vary over spacetime, the net
effect is that the invariants C2p

remain constant, though they may diverge for specific solutions.
Let us motivate the above definitions. To begin with, it follows from (2.5) and (2.12) that the
full traces obey

r f(Y, Z) g(Y,
r g(Y,
T
(3.42)
Z) = T
Z) f(Y, Z) ,
N = 1,
N+ =
and
r f (Y ) = Tr f (Y ),
T
(3.43)
where the reduced traces Tr f (Y ) are defined by4

d 4Y
Tr+ f (Y ) =
f (Y ),
Tr f (Y ) = f (0).
(2)2
(3.44)
Moreover, from D = 0, which is equivalent to ( ) = [A , ], it follows that

Cq = [A , Cq ] ,
Cq = ( )q ,
(3.45)
for any positive integer q, which together with (3.42) and the fact that A is an even function of
all the oscillators implies that formally (q = 1, 2, . . .)
r Cq = 0.
dT
4 The odd trace Tr for the Weyl algebra was originally introduced in [26].
(3.46)
14
This property can be made manifest by going to primed basis using (3.14).
Expanding perturbatively,
Cq = ( )q +
Cq(n) ,
(3.47)
n=q+1
and taking q = 2p, one finds that the leading contribution to the charges are given by

p
(2p)
C2p
= Tr () ,
(3.48)
where we have used (3.43), and

p
+(2p)
C2p
= Tr+ () ,
(3.49)
where we have formally factored out and canceled the volume of Z-space.
For

q = 2p + 1 one
r C2p+1 diverges like d 2 z or d 2 z , whose regufinds that the leading order contribution to T
larization we shall not consider here. The higher-order corrections to the charges, may require
additional prescriptions, and, as already mentioned, they need not be well-defined in general.
(2p)
The form of the leading contribution C2p

suggests that the full charges C2p
are well-defined
for general regular initial data. Moreover, these charges can be rewritten on a more suggestive
= (
which imply
form using (2.7), where dz dz = 2 d 2 z, and ()
),
1
F F Cq = d 2 z d 2 z Cq+2 ,
2
so that

(p1)
1
C2p = Tr+
,
F F ()
2 2
(3.50)
(3.51)
which can be used to show that the charges can be written as total derivatives in Z order by order
in the -expansion.
4. An SO(3, 1)-invariant solution
4.1. The ansatz
To find SO(3, 1)-invariant solutions we use the Z-space approach based on (3.14). It is convenient to use a Lorentz-covariant parameterization of the gauge function, viz. [24]

L(x; y, y)
= exp if x 2 x y y + r x 2
with x = (a ) x a and x 2 = x a xa . The function r(x 2 ) is fixed by demanding (L) = L1 ,
so that L SO(3, 2) and hence L1 dL describes the AdS4 vacuum solution.5 Using L1 =
(1 f 2 x 2 )2 exp[if (x 2 )x y y r(x 2 )] one finds r = log(1 f 2 x 2 ). A convenient choice
of f is [24]

2h
ix y y
L(x; y, y)
=
(4.1)
exp
,
2 x 2 < 1,
1+h
1+h

5 To exhibit L SO(3, 2) one can write L(x; y, y)

= exp (i
A + .
1h
1+h
1h2
artanh
x y y ) where exp A = 1 + A + 12 A
where

x = a xa ,
h=
1 2 x 2 ,
x 2 = x a xa
15
(4.2)
corresponding to the vierbein and Lorentz connection

e(0) =
( a ) dxa
,
h2
(0) =
2 ( ab ) dxa xb
,
h2
(4.3)
that in turn gives

2
=
ds(0)
4 dx 2
,
(1 2 x 2 )2
(4.4)
which one identifies as the metric of AdS4 spacetime with inverse radius given in stereographic
coordinates. The inversion x x /(2 x 2 ) maps the space-like regions 0 < 2 x 2 < 1 and
2 x 2 > 1 into each other and the boundary 2 x 2 = 1 onto itself. The future and past time-like
regions are mapped onto themselves, with distant past and future, where 2 x 2 , sent to
the past and future light cones, respectively. The two space-like and the two time-like regions
provide a global cover of AdS4 spacetime, so that beyond the distant past and future lies the
space-like region 2 x 2 > 1, as can be seen more explicitly using the global
parametrization
ds 2 = (1 + r 2 ) dt 2 + (1 + r 2 )1 dr 2 + r 2 d22 where x 2 = 2(1 + sin t 1 + r 2 )1 . Thus, a
global description can be obtained using the additional gauge function,
L = L(x;
y, y),
2 x 2 < 1,
(4.5)
and declaring the overlap with L(x; y, y)

to be given by
x a =
xa
,
2 x 2

2 x 2 = 1/ 2 x 2 < 0,
(4.6)
leading to the transition function L 1 L. The Z2 -symmetry implies that the local representatives
of the full solution will be given by the same functions, with x a x a , as we shall discuss in more
detail in Section 4.3.6
A particular type of SO(3, 1)-invariant solutions can be obtained by imposing
,
] = 0,
[M
(4.7)
D
M
(4.8)
= 0,
are the full Lorentz generators defined in (2.24) and given in the primed basis, and
where M
obeying (2.26)(2.28). Eq. (4.7) combined with (2.26) imply that

=
(u, u),
(4.9)
where
u = y z ,
u = u = y z ,
(4.10)
6 One can describe anti-de Sitter spacetime using the globally well-defined gauge function L = (1
2 x 2 ) exp ix y y where 2 x 2 = 1. The corresponding SO(3, 1)-invariant solution has Lorentz connection and vierbein given by (4.61) and (4.62) with a 2 replaced by x 2 in (4.59). These expressions are ill-defined for |x 2 | > 1, which
means that a globally well-defined solution still requires another coordinate patch.
16
and (
(u, u))
=
(u, u).
Moreover, from (4.8) combined with (2.27) and the -invariance
condition on A , it follows that
S
= z S(u, u),
A
= z A(u, u),
S = 1 2iA.
(4.11)
We next turn to the exact solution of the Z-space equations (3.16)(3.18).

4.2. Solution of Z-space equations
The internal constraints F
= 0 and D

= 0 are solved by
=

(u, u)
S(u, u)
= S(u),
,
b1
(4.12)
where /b1 is a real constant, so that is real in the type A model and purely imaginary in the
type B model.
, given by (3.16), now takes the form

The remaining constraint on F

S , S = 4i 1 eiu .
(4.13)
To solve this constraint, following [23], we use the integral representation

1
S(u) =
ds m(s)e 2 (1+s)u ,
(4.14)
where the choice of contour is motivated by the relation

i
i

z e 2 (1+s)u z e 2 (1+s )u

i

1
= i
y z + s
(y + z) y + z + s(y z) e 2(1ss
)u ,
4
(4.15)
which induces the map (s, s

) ss
from [1, 1] [1, 1] to [1, 1]. As a result (4.13)
becomes
1
1
ds

i
i
ds 1 + (1 ss )u m(s)m(s
)e 2 (1ss )u = 1 eiu ,
4
(4.16)
which can be written as

1
1

i
i
dt g(t) 1 + (1 t)u e 2 (1t)u = 1 eiu ,
4
(4.17)
where g = m m with defined by [23]

1
(p q)(t) =
1
1
ds
1
ds
(t ss
)p(s)q(s
).
(4.18)
17
Replacing iu by 2d/dt acting on the exponential and integrating by parts, we find

1
1

1

i
i
1
1
dt g(t) + (1 t)g(t) e 2 (1t)u (1 t)g(t)e 2 (1t)u 1 = 1 eiu .
2
2
(4.19)
This can be satisfied by taking g to obey

1
g(1) = ,
(1 t)g(t) = (1 t),
2
with the solution
(m m)(t) = g(t) = (t 1) (1 t).

2
Even and odd functions are orthogonal with respect to the product, i.e.,
g(t) +
p( ) q ( ) =
p ( ) q ( ) ,
Therefore
(+)

m m(+) (t) = I0(+) (t)
()

()
m m() (t) = I0 (t) +
p ( ) (t) = p ( ) (t),
= 1.
,
2
t,
2
(4.20)
(4.21)
(4.22)
(4.23)
(4.24)
where

1
(1 t) (1 + t) .
2
()
I0 (t) =
(4.25)
()
One proceeds [23], by expanding m() (t) in terms of I0 (t) and the functions (k 1)

1 (1 )
( )
Ik (t) = sign(t) 2
1
1
ds1
1
1 k1
)
t2

1 (1 ) (log
= sign(t) 2
,
(k 1)!
dsk (t s1 sk )
(4.26)
which obey the algebra (k, l 0)

( )
Ik
( )
Il
( )
= Ik+l .
(4.27)
Thus, given a quantity

p
( )
(t) =
( )
(4.28)
pk k ,
(4.29)
pk Ik (t),
k=0
and defining its symbol

p ( ) ( ) =

k=0
it follows from (4.27) that

q ( ) = p ( ) ( )q ( ) ( ),
p
(4.30)
18
so that (4.23) and (4.24) become the algebraic equations

2
m
(+) ( ) = 1 ,
2
() 2

m
( ) = 1 +
.
2 1 + 12
(4.31)
Therefore
(+)
(t) = I
(+)
(t) + q
(+)

(t) ,
(+)
( ) =

m() (t) = I () (t) + q () (t) ,
q () ( ) =
1 1,
2
1+

1.
2 1 + 12
(4.32)
The physical gauge condition (3.20), which requires

m(t)|=0 = (1 + t),
(4.33)
implies that
m(t) = m(+) (t) m() (t) = (1 + t) + q(t),
(4.34)
q(t) = q (+) (t) q () (t).
(4.35)
To obtain the functions q () (t) explicitly, we first expand

1 k1

1 (1 )
() (log t 2 )
q () (t) = sign(t) 2
qk
,
(k 1)!
(4.36)
k=1
where the coefficients are related to the expansions of the symbols as

q
()
( ) =
()
qk k .
(4.37)
k=1
In the case of q (+) (t), expansion of
(+)
(t) =
1
2
k+1
k=0

1 2 1 yields
k+1
(log t12 )k
k!
In the case of q () (t), we begin by defining

1
q () (t) = sign(t)q log 2 .
t
1
1
= 1 F1 ; 2; log 2 .
4
2
2
t
(4.38)
(4.39)
Thus, from
)
Q(
d q(
)=
0

k=1
()
qk
k!
(4.40)
19

k
). This Laplace-type transand k = 0 d e (k! ) , it follows that q () ( ) = 0 d e Q(
formation can be inverted as
)=
Q(
+i
with

dz e z () 1
q
,
2iz
z
(4.41)

1 () 1
.
> max Re zi : zi pole or branch cut of q
z
z
The function q () (t) is then obtained by differentiation with respect to and the substitution
= log t12 , that is
+i
()
(t) = sign(t)
i

dz e z () 1
.
q
2i
z =log 1
(4.42)
t2
The contour can be closed around the branch-cut that goes from z = 12 to z = 1+
2 , and one
finds

t
1
1
()
q (t) = 1 F1 ; 2; log 2 .
(4.43)
4
2
2
t
In summary, the internal solution is given by

= ,
b1
A
(4.44)
i
= z
2
1
q(t) =
dt q(t)e 2 (1+t)u ,

1 F1

1
1
1
; 2; log 2 + t 1 F1 ; 2; log 2 .
2
2
2
2
t
t
(4.45)
(4.46)
p
Expanding exp( itu
2 ) results in integrals of the degenerate hypergeometric functions times t (p =
0, 1, . . .), which improve the convergence at t = 0. Thus A is a formal power-series expansion
in u with coefficients that are functions of that are well-behaved provided this is the case for
the coefficient of u0 . This is the case for in some finite region around = 0, as we shall see
next.
4.3. The solution in spacetime

Let us evaluate the physical fields in the two regions 2 x 2 < 1 and 2 x 2 < 1 using the two
gauge functions (4.1) and (4.5). We first compute = L1
(L) = b1 L1 L1 , with the
result

=
(4.47)
1 2 x 2 exp ix y y
b1
20
This shows that the physical scalar field is given in the x a -coordinate chart by

Y =Z=0 =
1 2 x 2 , 2 x 2 < 1,
(x) = |
b1
(4.48)
the physical scalar field in

while the Weyl tensors for spin s = 2, 4, . . . vanish. Using instead L,
a
the x -coordinate chart is given by

= 1 2 x 2 , 2 x 2 < 1.
(4.49)
As a result, the two scalar fields are related by a duality transformation in the overlap region
x)
(
=
1

< 0.
2 x 2 = 2 x 2
(x)
,
(x)
(4.50)
Thus, if the transition takes place at 2 x 2 = 2 x 2 = 1, then the amplitude of the physical scalar
never exceeds 2 so the open-string theory of [17], which couples to the Weyl zero-form, is
weakly coupled throughout the solution. We also note that a gauge-invariant characterization of
the solution is provided by the invariants (3.38), that are given by
= (/b1 )2p ,
C2p
(4.51)
+
while C2p
diverges.
(0)
(0)
The gauge fields are defined by the decomposition (2.31) with A = L1 L = e + ,
that is
(0)
(0)
e + W = e
+
K ,
(0)
(4.52)
(0)
with e and denoting the AdS4 vacuum, and

K = i L1 A
A
L h.c.,
(4.53)
which can be rewritten using (4.15) as

K =
1

16i
1
1
dt
dt
q(t)q(t
)

(1 + t)(1 + t
)
h.c.,

2
2

(1 t)(1 t ) V (y, y;
, ; tt )

= =0
(4.54)
where

1tt
V (y, y;
, ; tt
) = L1 ei( 2 u+y+ z) L Z=0 .
(4.55)
Using (B.6), this quantity can be expressed as

V (y, y;
, ; tt
) =
2a 2
4h2
((1 + a 2 )y + 2a y)
exp
i
,
(1 + h)2 (1 tt
a 2 )2
1 tt
a 2
(4.56)
where
a =
x
.
1 + 1 2 x 2
(4.57)
We can thus write

2

1
K = Q 1 + a 2 y y + 4 1 + a 2 a y y + 4a a y y h.c.,
4i
where
1
2
1
Q = 1 a2
4
1
dt
dt
q(t)q(t
)(1 + t)(1 + t
)
.
(1 tt
a 2 )4
21
(4.58)
(4.59)
This function is studied further and evaluated at order 2 in Appendix D. From (4.52) and (4.58)
it follows that all higher-spin gauge fields vanish,
W = 0,
(4.60)
while the vierbein and Lorentz connection are given by

= f (0) ,

2 4
f) 4a 2 e(0) + (1 + a ) 3 dx a xa x ,
e = e(0) (Qf + Q
(1 a 2 )2
(4.61)
(4.62)
where we have used a a = a 2 , and defined

f=
(0)
(0)
1 + (1 a 2 )2 Q
.
2
2
2
|1 + (1 + a ) Q| 16a 4 |Q|2
(4.63)
We identify the vielbein as a conformally rescaled AdS4 metric

ea = f1 dx a + 2 f2 dx b xb x a =
2 d(g1 x a )
,
1 2 g12 x 2
(4.64)
where
2(1 + a 2 )2

2
f) ,
1
4a
(Qf
+
Q
f1 x 2 =
(1 a 2 )2
(1 + a 2 )4
f2 x 2 =
(Qf + Q f),
(1 a 2 )2
(4.65)
and the scale factor is given by

(1 2 g12 x 2 )f1
=
,
2g1

g1 = exp
2
2

f2 (t) dt
.
f1 (t)
(4.66)
x2
In the boundary region, where a 2 1, the double integral in (4.59) diverges at t = t

= 1 while
the prefactor goes to zero, as to produce a finite residue given by
lim Q =
a 2 1
2
.
6
(4.67)
In this limit
lim =
a 2 1
1
1
4 2
3
(4.68)
22
This factor is positive in the type B model, while curiously enough it blows up at a critical value,
within the range (D.6), in the type A model.
By the Z2 -symmetry, the metric in the x a -coordinate chart is given by the same functions
as in the x a -coordinate system (which is not the same as a reparameterization!). In the distant
past and future, where a 2 1, the function Q diverges logarithmically leading to qualitatively
different behavior of the scale factors in the cases of the type A and B models, which is analyzed
in more detail in [25]. Here we content ourselves by observing that any pathological behavior
in the strong-coupling region 2 x 2 < 1 can be removed by going to the dual weakly coupled
frame. Thus, geometrically speaking, the solution interpolates between two asymptotically AdS4
regions at 2 x 2 1 and 2 x 2 1 via an interior given by complicated scale factors (discussed
in [25]) times foliates determined by symmetries, to which we shall turn our attention next.
4.4. Symmetries of the solution
A more general discussion of symmetries of solutions will be given in Section 5. In view
of (5.4), the gauge transformations preserving the primed solution obey
D

= 0,
(

) =

,
(4.69)
where the last condition is equivalent to [

,
] = 0 since
= is constant. The condition (4.69) is by construction solved by the full SO(3, 1) generators, i.e.,
M h.c.,
4i
(4.70)
given by (2.24) and constant .

with M
The solution is also left invariant by additional transformations with rigid higher-spin parameters

,
(4.71)
=0
where the th level is given by

1 ...2m , 1 ... 2n M
1 2 M
2m1 2m M
1 2 M
2n1 2n

=
m+n=2+1
h.c.,
(4.72)
with constant 1 ...2m , 1 ... 2n . These parameters span the solution space to (4.69), provided that
this space has a smooth dependence on . The full symmetry algebra is thus a higher-spin extension of SO(3, 1) SL(2, C), that we shall denote by
hsl(2, C; ) sl(2, C),
(4.73)
and its hermitian conjugate, and we have indicated that in

where sl(2, C) is generated by M
general the structure coefficients may depend on the deformation parameter .
The generators of the SO(3, 1) transformations preserving the spacetime dependent field configuration, that is, obeying (5.1), are by construction given by
L
M
= L1 (x) M
L(x),
(4.74)
23
and are related to the full Lorentz generators M by

L
M = M (x) M ,
M
where M = y y and we have defined

M (v) = L1 1 v M L 1 v = y (v)y (v),
(4.75)
(4.76)
where the transformed oscillators are defined by

y + v y
y (v) =
,
1 v2
(4.77)
and obey the same algebra as the original oscillators, viz.

y y = y y + i
.
(4.78)
and g L be the group elements generated by M

and M L , respectively, then it
If we let g
follows that

L
L(x) g
(4.79)
= g
L(x),
x = x .
The spacetime decomposes under this SO(3, 1) action into orbits that are three-dimensional
hyper-surfaces which describe local foliations of AdS4 with dS3 and H3 spaces in the regions
x 2 > 0 and x 2 < 0, respectively.
5. On other solutions with non-maximal isometry
In this section we discuss the consequences of imposing various non-maximal symmetry conditions on solutions. We first do this in a general setting, and then consider symmetry groups of
dimensions 3, 4 and 6. In this approach, we recover the previous so(3, 1)-invariant solution and
also construct new solutions at the first order in the Weyl zero-form. The latter include domain
walls and rotationally invariant solutions.
5.1. Some generalities
The solution with maximal unbroken symmetry is the AdS4 vacuum = 0, which is invariant under rigid hs(4) transformations, with Z-independent parameters. Let us consider nonvanishing that is invariant under a non-trivial set of transformations with parameters belonging
to7

=
:
= 0, (
) =
=
.
h()
D
= 0, [
, ]
(5.1)
As is the case for hs(4), this algebra closes under -commutation of parameters as well as compositions induced by the associativity of the -product, e.g.,
1
2
3 +
3
2
1 ,
i h().
(5.2)
7 Here we are specializing to invariance under transformations generated by acting only on the internal indices. More
generally, one may consider invariance conditions involving transformations containing also a spinorbit part, acting on
the x-dependence.
24
contains a finite-dimensional rank-r subalgebra gr SO(3, 2) this induces

In the case that h()
as exemplified in (4.71) and (4.72) for hsl(2, C; ).
a natural higher-spin structure h(),
In the Z-space approach
= L 1

L,

= 0,
(5.3)
where
D

= 0,
[

,
] = 0.
Expanding perturbatively in

|Z=0
(5.4)
= C
,
+
(2)
+ ,

=

+
(1)
(5.5)
and assuming that the parameters obeying (5.4) are given up to and including order n 1 (n =
is determined by
1, 2, . . .), then
(n)
(D

)(n) = 0,
(5.6)
which is an integrable partial differential equation in Z-space provided that

[

,
] (n) = 0.
I(n)
(5.7)
Using the Z-space field equations obeyed in the lower orders, one can show that
I(n)
= 0,
(5.8)
so that (5.7) holds if (n = 1, 2, . . .)

I(n)
I(n)
|Z=0 =
[
(p)
, (q)
] Z=0 = 0.
(5.9)
p+q=n
In the first order, the symmetry condition reads

[

, C
] = 0.
We shall denote the stability algebra of C

h(C
) =

: [

, C
] = 0, (

) = (

) =

.
(5.10)
(5.11)
is in general a deformed
As found in the case of hsl(2, C; ), the full symmetry algebra h()
version of h(C ). Moreover, we shall denote the space of all twisted-adjoint elements invariant
under h(C
) by B(C
), i.e.,

B(C
) = C
: [

, C
] = 0,

h(C
) .
(5.12)
Covariance implies that if g is an hs(4) group element, then

h g 1 C
(g) = g 1 h(C
) g.
(5.13)
The spaces B(g 1 C

(g)) and g 1 B(C
) (g) are in general not isomorphic, however,
as there exist special points in the twisted-adjoint representation space where dim B(C
) is less
than the generic value on the group orbit. The simplest example is the point C
= , as will
be discussed below (5.51). Another subtlety, that we shall exemplify below, is related to the
fact that the associativity of the -product implies that if C
is a regular initial condition then
defined by (3.31). Thus, if dim B(C

) is finite then C
must
B(C
) contains the elements C(2n+1)
25
either violate regularity at some level of perturbation theory or be projector-like in the sense that
C2n+1
C
for some finite value of n.
In the case of a solution in which asymptotes to L1 C
L with h(C
) gr SO(3, 2),
where (3.27) is the AdS4 gauge function (3.27), the solution has gr isometry close to the boundary
provided the perturbation theory holds. Since the spacetime field equations are manifestly diffeomorphism and locally Lorentz invariant, the gr -isometry extends to the solution in the interior,
where it acts on the full master fields via parameters
= L1

L. Hence, the integrability
that is in general some
conditions (5.9) must hold, resulting in a full symmetry algebra h()
deformed higher-spin extension of gr with deformation parameters given by C
.
Let us next examine the above features in more detail in some special cases.
5.2. Solutions with unbroken SO(3, 1) symmetry
Acting on the exact SO(3, 1)-invariant solution described in Section 4 with the gauge transformation generated by the group element

g(v) = L 1 v ,
(5.14)
with L given by (4.1), which requires
v 2 < 1,
(5.15)
one finds the gauge-equivalent exact solution given by the zero-form

1 v 2 exp(iyv y),

(v) = g 1 (v)
g(v) =
b1
and the internal connection
i
A
(v) =
2
1

i
dt q (+) (t) q () (t) g 1 (v) z e 2 (1+t)u g(v).
(5.16)
(5.17)
The gauge-transformed solution has

C
(v) =
(v) ,
(5.18)
with stability group h(C

(v) ) generated via enveloping of the generators M (v) given by (4.76).
Thus, the space B(C
(v) ) consists of all elements obeying

y (v)y (v), C
= 0,
(5.19)
amounting to the following second order partial differential equation

4
+ y (v y)
C
+ ( ) = 0.
iy + i(v y)
(v )
(v )
1 v2
This equation can be solved using the ansatz
C
= C
(V ),
V = yv y,
(5.20)
(5.21)
implying the following second-order ordinary differential equation in variable V with constant
coefficients

2

2 d
2 d
+i 1+v
+ 1 C
= 0,
v
(5.22)
dV
dV 2
26
which admits the solutions

i V2
iV
v
C
= 1 e + 2 e
iV
1 e
for v 2 = 0 and v 2 < 1,

for v 2 = 0.
(5.23)
It is not possible to produce any further solutions to (5.19) using (3.31), since exp iV (exp iV )
is proportional to 1, while the -product composition of exp iV and (exp i vV2 ) is divergent.
Thus,

2 for v 2 = 0 and v 2 < 1,
dim B C
(v) =
(5.24)
1 for v 2 = 0.
For v 2 = 0 the space B(C
(v) ) in invariant under v a v a /v 2 . Extending (5.23) to
v 2 > 1,
(5.25)
gives an SO(3, 1)-invariant two-dimensional solution space with stability group generated by
M (v) given by (4.76), where the fact that the denominator in (4.77) is imaginary does not
present an obstacle since the sl(2, C)-doublet oscillators are complex. We stress that h(C
(v) ),
which is the stability group of the twisted-adjoint element C
(v) and not of the parameter v,
remains equal to SO(3, 1) when v is null. Moreover, we note that the additional solution with
super-luminal boost-parameter v a is not gauge-equivalent to
(v) , and that C
is a regular initial
condition, according to the terminology introduced in Section 3.3, if 1 2 = 0, else it is singular.8
Whether the singular or super-luminal cases can be elevated to exact solutions remains to be seen.
The self-dual case v 2 = 1 requires a separate treatment. Here = y + v y is a commuting oscillator giving rise to a three-dimensional translation generator pa = (v b (ab ) +
h.c.) obeying [pa , pb ] = 0 and v a pa = 0. The commuting oscillators obey the reality condition ( ) = v , and transform as doublets under the oscillator realization of the SL(2, R)
SO(2, 1) that leaves va invariant. This leads to an I SO(2, 1)-invariant two-dimensional solution
space B(eV ) to be described next at the level of the linearized field equations (see Eq. (5.47)
together with some other interesting reductions.
5.3. On domain walls, rotationally invariant and RW-like solutions
Here we shall discuss some classes of solutions of considerable interest corresponding to the
rank-r subalgebras gr SO(3, 2) parameterized by
g3 :
Mij = Lai Lbj Mab ,
g4 :
Mij ,
g6 :
Mij ,
(5.26)
1
P = La Pa = La (a ) y y ,
4

Pi = Mab Lb + Pa Lai ,
where and are real parameters and, in all cases,

SO(3, 1)/SO(3) or SO(3, 1)/SO(2, 1), obeying
La La =
= 1,
(5.27)
(5.28)
(Lai , La )
is a representative of the coset
Lai La = 0,
Lai Lj a = ij = diag(+, +,
).
8 Inserting (5.23) into (3.32) one finds an expression with an isolated singularity at t = 1.
(5.29)
27
The SO(3, 2) algebra (A.2) yields [M, M] M. Furthermore, for g4 one finds [M, P ] 0, while
for g6 one finds [M, P ] P , and

[Pi , Pj ] = i 2
2 Mij .
(5.30)
In summary, one has
2
2
SO(3, 1) for
> 0,
= 1,
2
2
g6 = ISO(2, 1) for = 0,
= +1,
SO(2, 2) for 2 2 > 0,

= +1,

SO(3) SO(2),
= 1,
g4 =
SO(2, 1) SO(2),
= +1,

SO(3),
= 1,
g3 =
SO(2, 1),
= +1.
(5.31)
(5.32)
(5.33)
=
|Z=0 that obey the
Next we seek gr -invariant twisted-adjoint initial conditions C
(y, y)
invariance condition (5.9) in the first order, viz.
[

, C
] = 0,

gr .
(5.34)
As discussed in Section 3.2, C

provides an initial condition to (3.16)(3.18) giving rise to a
full master field
related to the spacetime Weyl tensors through = L1
L1 , so that
C = L1 C
L1 are the linearized Weyl tensors, discussed in Section 3.1. We note that C is
invariant under gauge transformations with parameters L1

L, while is invariant under
gauge transformations with deformed parameters

provided that the integrability conditions
(5.9) hold to all orders. In Section 5.4 we shall perform the symmetry analysis in the second
order.
The condition (5.34) decomposes into two irreducible conditions

[Pi , C
] = 0 for g6 ,
[Mij , C ] = 0,
(5.35)
{P , C
} = 0 for g4 .
The first condition can be shown to have the general solution
g3 : C
= C
(P ),
(5.36)
where C
(P ) is a function that we shall assume is analytic at the origin. To arrive at this conclusion one can use the fact that the oscillator realization (A.3) implies M[ab Mc]d = M[ab Pc] =
Mab M b c = Mab P b = 0, or, alternatively, note that the only g3 -invariant spinorial objects are
,
and L .
Thus, the g3 -invariant solution space B(C
), defined by (5.12), is infinite-dimensional, and
indeed closes under (3.31) for regular initial data. The g3 -invariance can be imposed at the full
level using the deformed generators
,
g3 : M ij
= Lai Lbj M ab
(5.37)
where M ab are obtained from (2.24). This results in consistent SO(3)-invariant or SO(2, 1)invariant mini-superspace truncations described by the master equations (3.16)(3.18) with
reduced master fields
P
, , ),

=
(P ; u, u;
(5.38)
28
and
A
= z A 1 + y A 2 + L (z A 3 + y A 4 ),
P
, , ),
A i = A i (P ; u, u;
(5.39)
taken to depend on the oscillators only through the following set of composite variables,
1
P = L y y ,
4
= La (a ) z y ,
1
P
= L z z ,
4
u = y z ,
(5.40)
where L = ( a ) La . We stress that the full g3 symmetry is manifest, assuring that the above
restricted dependence on the oscillators is actually consistent with the master equations (3.16)
(3.18). The reduced models correspond geometrically to local reductions on either a two-sphere
or a two-hyperboloid, with complete symmetry algebra given by deformed enveloping-algebra
extensions of g3 generated by M ij
.
Turning to the second set of conditions in (5.35), we use the lemmas

Li a Lp Mab , P k = ik
Li a Pa P k1 ,
(5.41)

k(k 1)
k2
Li a Pa , P k = Li a Pa 2P k
(5.42)
P
,
8

k(k + 1)
k1
,
P
P , P k = 2P k+1 +
(5.43)
8
to rewrite them as the following second-order differential equations
d 2
d
+
i
+
2
C
(P ) = 0,
g6 :
(5.44)
8 dP 2
dP
d2
+ 2 P C
(P ) = 0.
g4 :
(5.45)
8 dP 2
The resulting initial data that are analytic at P = 0 are given by
SO(3, 1):
ISO(2, 1):
C
(P ) = 1 e4i P + 2 e4i+ P ,
C (P ) = (1 + i2 P )e
4iP
(5.47)
+ (1 i2 )e
C (P ) = (1 + i2 )e
sinh(4P )
SO(3) SO(2): C
(P ) = 1
,
P
sin(4P )
,
SO(2, 1) SO(2): C
(P ) = 1
P
where 1,2 and 1,2 are real constants and

=

.
SO(2, 2):
4i P
(5.46)
4i+ P
(5.48)
(5.49)
(5.50)
(5.51)
We see that in general dim B(C

) = 2 for the g6 -invariant initial data, except at a few special
points where dim B(C
) = 1, while dim B(C
) = 1 for the g4 -invariant initial data.
29
In the case of SO(3, 1), we can identify particular cases of (5.46) with the exact solutions (5.16) and (5.17), namely9
= 1, 1 > v 2 > 0:
a
La = v , + = v 2 ,
v2
La = v a , = v 2 ,
2
v
= 1, v < 0:
a
va
L = 2 ,
2 =
2
b1 (1 v ),
1 = 0,
1 =
2
b1 (1 v ),
2 = 0,
(5.52)
La =
va
v 2
v 2 ,
= v 2 ,
+ =
2 =
2
b1 (1 v ),
1 = 0,
1 =
2
b1 (1 v ),
2 = 0.
(5.53)
On the other hand, as noted in Section 5.2, the initial condition C

= eiyv y with v 2 > 1 and an
arbitrary constant, and the initial conditions with 1 2 = 0 are gauge-inequivalent to the above
exact solutions. In these cases one might expect the full zero-form
to receive Z-dependent
higher-order corrections. We also note that in the perturbative approach, the case of 1 2 = 0 is
singular in the sense defined in Section 3.3 (i.e., inserting C
into (3.32) one finds an isolated
singularity at t = 1), while the case of v 2 > 1 involves poles at t = 1/v 2 < 1 that can be avoided
using suitable contours in the t -plane.
Turning to the I SO(2, 1)-invariant case (5.47), we identify it as a singular initial condition for
a flat domain wall solution. The case with 2 = 0 can be obtained formally as the limit of the dS3
domain wall in which

(1 v 2 )
2
,
fixed.
1 =
v 1,
(5.54)
b
b1
1
It remains to be seen whether the internal connection (5.17) and the spacetime gauge fields are
well-behaved in this limit. In view of (D.6), we expect there to be differences between the limiting
procedures in the type A and the type B model, where is real and imaginary, respectively.
The remaining three types of initial conditions listed above, i.e., the those invariant under
SO(2, 2), SO(3) SO(2) and SO(2, 1) SO(2), are singular for all values of the parameters.
In the singular cases found above, the perturbative expansion can be obtained using the closed
contours given in (3.33) and (3.34). It remains to be seen, however, whether full solutions can
be given explicitly in these cases. Another issue to settle is whether the

-parameters can be
deformed into full

-parameters obeying (5.4), to which we turn our attention next.
5.4. Existence of symmetry parameters to the second order
to the

-parameter associated to the
Let us prove the existence of the second correction
(2)
gr -symmetries discussed above. To do so, we need to establish the integrability of (5.4) to second
order, i.e., verify (5.9) for n = 2, where the relevant quantity reads

[
(1)
, C
] + [

, (2)
] Z=0 ,
I(2)
(5.55)
with

gr given in (5.28) or (5.27), and C
in (5.46)(5.50).
9 The case of v 2 = 0 requires the embedding of SO(3, 1) into SO(3, 2) using (4.76).
30
In fact, it is possible to show that (5.55) vanishes for the more general case with parameter

=

,

1
y y + y y ,
4i
(5.56)
and twisted-adjoint element

C
=
i m fm v(m)(m)
y (m) y (m)
,
(5.57)
m=0
where fm are real numbers; we use the notation y (m) = y 1 y m ; and we have defined
v(m)(m)
= v a1 v am (a1 )1 1 (am )m m .
(5.58)
The first-order correction to the parameter is given by
(1) = (1) (1)

,
(5.59)
with
1
(1) = b1
1
dt
t
dt

itt

itt yz ,
z z + z C
(tt
z, y)e
2
(5.60)
and the second-order correction to the zero-form is given in (C.5), which can be re-written as
(2)
= T B,
(5.61)
where the twisted-adjoint projection map T is defined by

T f = f + f + (f + f) ,
(5.62)
and
B = z
1

dt A (1)
C ZtZ ,
(5.63)
, which is a twisted-adjoint element, can thus be

with A given by (C.3). The quantity I(2)
written as
(1)
I(2)
= T (B 1 + B 2 )|Z=0 ,
(5.64)
where
B 1 = (1)
C
,
(5.65)
.
B 2 = T [

, B]
(5.66)
and
The expansion of B 1 reads
B 1 = b1
1
1
dt
t
dt
(tt
)m i m+n fm fn v (m)(m)
v (n)(n)
m,n
itt
31

itt
yz
e
z z + z z(m) y(m)
y(n) y(n)
.
(5.67)
To calculate the -contributions to T B 1 |Z=0 we let n m + 2 + k and contract m + 2 of the

y -oscillators with the z-oscillators in the first factor. The remaining y -oscillators may contract
z-oscillators in the exponent, so we sum over l such contractions with 0 l k. The resulting
terms have the common l-independent structure

T v (m)(m)
v(m)(k) (m+2+k)
y (k) y(m)
y (m+2+k)

T v (k)(k+2)
(5.68)
= 0,
y (k) y (k+2)
cannot contribute to the twisted-adjoint reprewhere we use (5.58) and the fact that y (k) y (k+2)
sentation. Similarly, to calculate the -contributions we let n m + 1 + k, resulting in

v (m)(k)(m+1+k)
y (k) y(m1)
y (m+1+k)
T v (m)(m)

T v(k)(k+1)
(5.69)
y (k) y (k+1) = 0.
is given
The same type of cancellations occur in T B 2 |Z=0 . Here the -contribution to [

, B]
by
b1
8
1
1
dt
t
dt
(t
v (n)(n)
m,n

y y , z z(m+1) y(m)
eit yz (y(n) y(n)
) ZtZ ,
(5.70)
which we evaluate at Z = 0 using

y y , z f Z=0 = 4 (y) fZ=0 ,
(5.71)
|Z=0 of the form

resulting in contributions to T [

, B]

v(m+1)(k)(m+1+k)
y (k1) y(m)
y (m+1+k)
T v (m)(m)

T v(k)(k+1)
y (k1) y (k+1) = 0.
(5.72)
reads
Finally, the -contribution to [

, B]
b1
4
1
1
dt
0

t
dt
(t
v (n)(n)
m,n

y y , z z(m+1) y(m)
eit yz (y(n) y(n)
) ZtZ ,
which we evaluate at Z = 0 using

y y , z f Z=0 = 2i y f,
(5.73)
(5.74)
32
|Z=0 of the form

resulting in contributions to T [

, B]

T y v (m)(m)
v(m+1)(k)(m+1+k)
y (k) y(m)
y (m+1+k)

T y v(k)(k+1)
y (k) y (k+1) = 0,
(5.75)
which concludes the proof of (5.55).

Interestingly enough, we have found a stronger result, namely that
C
= C
(V ),
V = yv y,
(5.76)
preserves SO(3, 2) to second order in the curvature expansion for general initial conditions
C
(V ). We expect SO(3, 2) to be broken down to some gr with r 6 at third order.
6. Discussion
We have seen that, while the field equations written in traditional form in spacetime are highly
complicated, and indeed not even available explicitly at present, we can the nonetheless solve
them exactly by exploiting their relative simple form in terms of master fields. Doing so, we
have found the first exact solution of higher spin gauge theory in D > 3 other than the anti de
Sitter spacetime.
The solution presented here forms a consistent background for the first-quantized topological open twistor string description of higher spin gauge theory [17]. It would be interesting to
study this (1 + 1)-dimensional field theory which provides a framework for a dual twistor space
interpretation of the solution.
It would also be useful to determine the holographic interpretation of our solution. This requires, however, the knowledge of an on-shell action and a systematic way to label the fluctuation
fields that takes into account the infinite dimensional nature of the underlying symmetries. While
progress has been made in that direction [27] much remains to be done to develop fully the
necessary tools to facilitate the fluctuation analysis.
Next, we note that while our solution is time dependent, its cosmological interpretation is not
straightforward. To begin with, an understanding of the key concepts of horizons and singularities
are very much based on the geodesic equation of motion for test particles, yet we do not know
its higher spin covariant counterpart.
A key feature of the theory relevant in dealing with both holographic and spacetime geometric aspects is the presence of non-localities in the spacetime field equations. These non-localities
should be manageable, however, once we work with master fields. For example, using this approach, we have already found certain zero-form charges that provide a set of labels for classical
solutions. The master field approach should also be used to construct a higher spin dressed version of a spacetime line element, which in turn should be embedded into a manifestly higher spin
covariant infinite-dimensional geometry.
Finally, it is interesting to compare our solution with the SO(3, 1) invariant solution to the
gauged N = 8, D = 4 supergravity found in [28]. To this end, we first observe that the minimal
bosonic models we have studied here are consistent truncations of the higher-spin gauge theory
based on shs(8|4) osp(8|4) [29] which contains, respectively, 35+ + 35 scalars and pseudoscalars in the supergravity multiplet and 1 + 1 scalar and pseudo-scalar in an smax = 4 multiplet
that we refer to as the Konishi multiplet. The truncations to the type A and type B models keep
the Konishi scalar and pseudo-scalar, respectively [30]. Thus we see that an important difference between the solution of [28] and ours is that different scalars have been activated. The full
consequences of this difference remains to be investigated.
33
Acknowledgements
We thank M. Nishimura for participation at an early stage of this work, and we are grateful
to M. Bianchi, J. Engquist, D. Francia, P. Howe, R. Leigh, H. Lu, T. Petkou, C. Pope, P. Rajan,
A. Sagnotti, B. Sundborg and M. Vasiliev for discussions. We have benefited from the mutual
hospitality of our home institutes. The work of E.S. is supported in part by NSF Grant PHY0314712 and the work of P.S. is supported in part by INTAS Grant 03-51-6346.
Appendix A. Conventions and notation
We use the conventions of [29] in which the SO(3, 2) generators obey
[MAB , MCD ] = iBC MAD + 3 more,
MAB = (MAB ) ,
(A.1)
with AB = diag(+++). The commutation relations decompose into

[Mab , Mcd ] = 4i[c|[b Ma]|d] ,
[Mab , Pc ] = 2ic[b Pa] ,
[Pa , Pb ] = iMab .
(A.2)
The oscillator realization is taken to be

1
1
Mab = (ab ) y y + ( ab ) y y ,
Pa = (a ) y y ,
8
4
where the van der Warden symbols obey

a b

a b
= ab + ab ,
= ab + ab ,

1
1
abcd cd = i(ab ) ,
abcd cd = i( ab ) ,
2
2
a
ab ab
a a
= ,
= ,
=
(A.3)
(A.4)
(A.5)
with Minkowski spacetime metric ab = diag(+++), and spinor conventions A =

A ,
A = A
, and (A ) = A .
The SO(3, 2)-valued connection is expressed as

1

y y + y y + 2e
y y .
dx
4i
The components of R = d + are
=
R = d + + e e ,
R = d + + e e ,
R = de + e + e .
(A.6)
(A.7)
(A.8)
(A.9)
Defining
1
= (ab ) ab ,
4

e = (a ) ea ,
2
= ( ab ) ab ,
4
(A.10)
34
where is the inverse radius of the AdS4 vacuum, and converting the spinor indices of the curvatures in the same way, gives
Rab = dab + a c cb + 2 ea eb ,
Ra = dea + a b eb .
(A.11)
It follows that the AdS4 vacuum (0) is characterized by

R(0), = 2 (g(0) g (0) g (0) g (0) ),
R(0) = 32 g(0) .
(A.12)
Appendix B. Gaussian integration formulae

To compose coset representatives we make use of the formula
eiya y eiyby

1
ba
1
1
a b
+b
y + y
y ,
=
exp i y a
y y
det(1 + ab)
1 + ab
1 + ba
1 + a b
1 + ba
(B.1)
where we use matrix notation, e.g., ya y = y a y and y aby

= y a b y , with a =
a ( ) and a
= a ( )
. In deriving this formula, we use (2.5) and perform the integrals
over holomorphic and anti-holomorphic variables independently. In doing so, the exponentials
remain separately linear in integration variables, allowing the use of the identity
2 2
d d i [ +u ( ,)]+i
v ( ,))

( ,)
e
= eiu ( ,)v
,
(B.2)
2
(2)
and its analog for anti-holomorphic variables. Other useful formulae are

1
2ya y + (1 a 2 )y
iya y
iy
iya y
e
e e
=
exp i
,
(1 + a 2 )2
1 + a2

itza y ityz iya y
1
(1 a 2 )tyz + (1 t)ya y
e
e
=
exp
i
.
e
(1 ta 2 )2
1 ta 2
(B.3)
(B.4)
The relation (4.15) follows from the lemma
ei[tyz+y+ z] ei[t yz+ y+
z]
= ei[(t+t 2tt )yz+((1t )+(1t) +t
t
)y+((1t
) +(1t)
+t
t
)z+(+ )(

)]
Finally, to evaluate K on the solution, we need

eiya y ei(syz+y+ z) eiya y Z=0

2a 2
1
((1 + a 2 )y + 2a y)
=
exp
i
,
(1 (1 2s)a 2 )2
1 (1 2s)a 2
. (B.5)
(B.6)
where a 2 = a a .
Appendix C. Second-order corrections to the field equations
The second-order source terms P and J in the field equations are given by [6]
(2)
(2)
P(2) = (W
) W

(1)
+ (2) (e ) e (2) Z=0 ,
+ e (1) e
(C.1)

(1) (1)

(2)
J
= e
e + e , e(2) + e , W (1) + W , e(1) Z=0

(1) + h.c.
+ iR A (1)
+W W ( ),
A
Z=0
35
(C.2)
where the hatted quantities are defined as10

A (1)
ib1
=
z
2
1
t dt (tz, y)(tz,
y),
(C.3)
A (2)
1
= z

(1)
t dt A (1) A
ZtZ
+ z
1

(1)
t dt A (1)
, A ZtZ
1
ib1

t dt (2) ZtZ ,
(C.4)
(2) =
1

A (1)
dt z A (1)
ZtZ

A (1)
+ z A (1)
ZtZ ,
(1)
e
1

= ie

(1)
dt
y , A (1)
+ A , y ZtZ ,
t
(C.5)
(C.6)
W (1) = i
1
dt
t
W (1)
,A
y
0
(2)
e

= ie
1

W
(1)
+ A ,
,
y ZtZ
(C.7)

(2)
dt
y , A (2)
+ A , y ZtZ
t
0

e
1
dt
t
+ A (1)
1

(1)
dt
(1)

A
y , A (1)
+ A , y Zt
Z
t
z
y

(1)

y , A (1)
+ A , y Zt
Z
y

(1)

(1)
+
+
y , A (1)
+ A , y Zt
Z A
z
y

(1)

(1)
(1)
,
y , A + A , y Zt
Z A
z y
ZtZ
10 In Eq. (C.4) the last term was inadvertently omitted in [6].
(C.8)
36
where the replacement (z, z ) (tz, t z ) in a quantity must be performed after the quantity has
been written in Weyl ordered form, i.e., after the products defining the quantity has been performed.
Appendix D. Analysis of the Q-function
Expanding the denominator of (4.59), and splitting into even and odd parts in a 2 , as

Q a 2 = Q a 2 ,
Q = Q+ + Q ,
(D.1)
we find
Q+ = 1 a

2 2
2
1

4 4p
2p (+)
()
dt t q (t) tq (t) ,
a
2p
p=0
Q = 1 a

2 2
(D.2)
2
1

(+)

4
4p+2
2p+1
()
tq (t) q (t) ,
dt t
a
2p + 1
p=0
(D.3)
with q () (t) given by (4.38) and (4.43). Performing the integrals we arrive at
Q+ =
2

(1 a 2 )2 4 4p
1
,
1+
a
4
2p + 1
2p + 3
2p
1 a2
Q =
4
2
(D.4)
p=0

4
4p+2
1
,
1+
a
2p + 3
2p + 3
2p + 1
(D.5)
p=0
that are valid for

3 Re 1.
(D.6)
Expanding in ,
Q+ =

n=2
n Q+,n ,
Q =
2n Q+,2n ,
the first non-trivial contribution is found to be

(1 a 2 )2 d 2 4 2
2
2
Q+,2 =
a
,
3 F2 1, 1, 4; 3, 3; a + 3 F2 1, 1, 4; 3, 3; a
8
da 4
Q,2 =
(D.7)
n=2

(1 a 2 )2
2
2
.
3 F2 2, 2, 4; 3, 3; a 3 F2 2, 2, 4; 3, 3; a
8
(D.8)
(D.9)
We note that parity acts in tangent space by exchanging holomorphic and anti-holomorphic oscillators and, moreover, by exchanging with or in the type A and type B models,
respectively. Since the spin s = 2, 4, . . . Weyl tensors C1 ...2s can be taken to be even under parity, this means that the type B model is invariant under , or, equivalently, under ,
while this need not be, and is indeed not, a symmetry in the type A model.
37
References
[1] M.A. Vasiliev, Consistent equations for interacting gauge fields of all spins in 3 + 1 dimensions, Phys. Lett. B 243
(1990) 378.
[2] M.A. Vasiliev, Higher spin gauge theories: Star-product and AdS space, hep-th/9910096.
[3] M.A. Vasiliev, Nonlinear equations for symmetric massless higher spin fields in (A)dSd , Phys. Lett. B 567 (2003)
139, hep-th/0304049.
[4] A. Sagnotti, E. Sezgin, P. Sundell, On higher spins with a strong Sp(2, R) condition, hep-th/0501156.
[5] X. Bekaert, S. Cnockaert, C. Iazeolla, M.A. Vasiliev, Nonlinear higher spin theories in various dimensions, hepth/0503128.
[6] E. Sezgin, P. Sundell, Analysis of higher spin field equations in four dimensions, JHEP 0207 (2002) 055, hepth/0205132.
[7] E. Sezgin, P. Sundell, Holography in 4D (super)higher spin theories and a test via cubic scalar couplings, JHEP 0507
(2005) 044, hep-th/0305040.
[8] F. Kristiansson, P. Rajan, Scalar field corrections to AdS(4) gravity from higher spin gauge theory, JHEP 0304
(2003) 009, hep-th/0303202.
[9] M.A. Vasiliev, More on equations of motion for interacting massless fields of all spins in 3 + 1 dimensions, Phys.
Lett. B 285 (1992) 225.
[10] E. Sezgin, P. Sundell, Doubletons and 5D higher spin gauge theory, JHEP 0109 (2001) 036, hep-th/0105001.
[11] E. Sezgin, P. Sundell, Towards massless higher spin extension of D = 5, N = 8 gauged supergravity, JHEP 0109
(2001) 025, hep-th/0107186.
[12] E. Sezgin, P. Sundell, 7D bosonic higher spin theory: symmetry algebra and linearized constraints, Nucl. Phys.
B 644 (2002) 303, hep-th/0112100;
E. Sezgin, P. Sundell, Nucl. Phys. B 660 (2003) 403, Erratum.
[13] M.A. Vasiliev, Higher spin superalgebras in any dimension and their representations, JHEP 0412 (2004) 046, hepth/0404124.
[14] M. Cederwall, AdS twistors for higher spin theory, AIP Conf. Proc. 767 (2005) 96, hep-th/0412222.
[15] I. Bars, M. Picon, Single twistor description of massless, massive, AdS, and other interacting particles, Phys. Rev.
D 73 (2006) 064002, hep-th/0512091.
[16] I. Bars, S.J. Rey, Noncommutative Sp(2, R) gauge theories: A field theory approach to two-time physics, Phys. Rev.
D 64 (2001) 046005, hep-th/0104135.
[17] J. Engquist, P. Sundell, Brane partons and singleton strings, hep-th/0508124.
[18] B. Sundborg, Stringy gravity, interacting tensionless massless higher spins, Nucl. Phys. B (Proc. Suppl.) 102 (2001)
113, hep-th/0103247.
[19] E. Sezgin, P. Sundell, Massless higher spins and holography, Nucl. Phys. B 644 (2003) 303, hep-th/0205131;
E. Sezgin, P. Sundell, Massless higher spins and holography, Nucl. Phys. B 660 (2003) 403, Erratum.
[20] I.R. Klebanov, A.M. Polyakov, AdS dual of the critical O(N ) vector model, Phys. Lett. B 550 (2002) 213, hepth/0210114.
[21] A. Sagnotti, M. Tsulaia, On higher spins and the tensionless limit of string theory, Nucl. Phys. B 682 (2004) 83,
hep-th/0311257.
[22] N. Beisert, M. Bianchi, J.F. Morales, H. Samtleben, Higher spin symmetry and N = 4 SYM, JHEP 0407 (2004)
058, hep-th/0405057.
[23] S.F. Prokushkin, M.A. Vasiliev, Higher-spin gauge interactions for massive matter fields in 3D AdS spacetime,
Nucl. Phys. B 545 (1999) 385, hep-th/9806236.
[24] K.I. Bolotin, M.A. Vasiliev, Star-product and massless free field dynamics in AdS(4), Phys. Lett. B 479 (2000) 421,
hep-th/0001031.
[25] E. Sezgin, P. Sundell, On an exact cosmological solution of higher spin gauge theory, hep-th/0511296.
[26] M.A. Vasiliev, Extended higher spin superalgebras and their realizations in terms of quantum operators, Fortschr.
Phys. 36 (1988) 33.
[27] M.A. Vasiliev, Actions, charges and off-shell fields in the unfolded dynamics approach, Int. J. Geom. Methods Mod.
Phys. 3 (2006) 37, hep-th/0504090.
[28] T. Hertog, G.T. Horowitz, Towards a big crunch dual, JHEP 0407 (2004) 073, hep-th/0406134;
T. Hertog, G.T. Horowitz, Holographic description of AdS cosmologies, JHEP 0504 (2005) 005, hep-th/0503071.
[29] E. Sezgin, P. Sundell, Higher spin N = 8 supergravity, JHEP 9811 (1998) 016, hep-th/9805125.
[30] J. Engquist, E. Sezgin, P. Sundell, On N = 1, 2, 4 higher spin gauge theories in four dimensions, Class. Quantum
Grav. 19 (2002) 6175, hep-th/0207101.
Kerrde Sitter black holes with NUT charges

W. Chen, H. L , C.N. Pope
George P. & Cynthia W. Mitchell Institute for Fundamental Physics, Texas A&M University,
College Station, TX 77843-4242, USA
Received 22 February 2006; accepted 19 July 2006
Available online 7 August 2006
Abstract
The four-dimensional Kerrde Sitter and KerrAdS black hole metrics have cohomogeneity 2, and they
admit a generalisation in which an additional parameter characterising a NUT charge is included. In this
paper, we study the higher-dimensional KerrAdS metrics, specialised to cohomogeneity 2 by appropriate
restrictions on their rotation parameters, and we show how they too admit a generalisation in which an
additional NUT-type parameter is introduced. We discuss also the supersymmetric limits of the new metrics.
If one performs a Wick rotation to Euclidean spacetime signature, these yield new EinsteinSasaki metrics
in odd dimensions, and Ricci-flat metrics in even dimensions. We also study the five-dimensional Kerr
AdS black holes in detail. Although in this particular case the NUT parameter is trivial, our investigation
reveals the remarkable feature that a five-dimensional KerrAdS over-rotating metric is equivalent, after
performing a coordinate transformation, to an under-rotating KerrAdS metric.
1. Introduction
The discovery by Kerr in 1963 [1] of the exact metric describing a rotating black hole was
arguably the most important advance in the study of exact solutions in general relativity since
Schwarzschilds discovery in 1916 [2] of the metric describing a static black hole. It was followed
within a few years by the finding of various generalisations, including charged rotating black
holes [3], and then the further inclusion of a cosmological constant [4], NUT parameter [5] and
acceleration parameter [6].
Research supported in part by DOE grant DE-FG03-95ER40917.
E-mail address: honglu@physics.tamu.edu (H. L).

doi:10.1016/j.nuclphysb.2006.07.022
W. Chen et al. / Nuclear Physics B 762 (2007) 3854
39
With the advent of supergravity and superstring theory, interest also developed in studying
higher-dimensional solutions of the Einstein equations. In 1986, the general solution describing
an asymptotically flat rotating black hole in arbitrary dimension was discovered by Myers and
Perry [7]. In dimension D, this has a mass parameter and [(D 1)/2] independent rotation parameters, one for each of the orthogonal spatial 2-planes. Motivated by the study of the AdS/CFT
correspondence in string theory, Hawking, Hunter and Taylor-Robinson constructed the solution
for the five-dimensional rotating black hole with a cosmological constant in 1998 [8]. They also
obtained a special case of the metrics in all higher dimensions, in which there is a rotation in only
one of the [(D 1)/2] orthogonal spatial 2-planes. The general rotating Kerrde Sitter black hole
solution with a cosmological constant in arbitrary dimension, with all [(D 1)/2] independent
rotation parameters ai , was constructed by Gibbons, L, Page and Pope in 2004 [9,10].
In view of the fact that the four-dimensional rotating black hole metrics admit further generalisations where additional non-trivial parameters are present, one might wonder whether such
additional parameters could also be introduced in higher dimensions too. In fact, in a certain special class of higher-dimensional Kerrde Sitter black holes, namely those in which there is just a
rotation in a single 2-plane, a generalisation which includes a NUT parameter as well as the mass
and the (single) rotation parameter has been obtained [11]. It was shown in [12] that this generalisation, which is trivial in five dimensions but non-trivial in dimensions D 6, still exhibits
certain remarkable separability properties for the HamiltonJacobi and wave equations, which
in fact played an important rle in the original discovery of the generalised four-dimensional
solutions.
The purpose of this paper is to present new results we have obtained for much wider classes
of generalisations of the Kerrde Sitter metrics, in which there is a NUT-type parameter as well
as the mass parameter. The cases covered by our new solutions are when the rotation parameters
ai are divided into two sets, in which all parameters within a set are equal. In odd dimensions,
which we discuss in Section 2, we obtain generalised solutions for an arbitrary partition of the
parameters into two such sets. In even dimensions, which we discuss in Section 3, the parameters
are partitioned into one set with a non-vanishing value for the rotation, and the other set with vanishing rotation. In each of the odd- and even-dimensional cases, the net effect is to give a metric
of cohomogeneity 2. In a manner that parallels rather closely the generalisations in D = 4, the
two associated coordinates, on which the metric functions are intrinsically dependent, enter in a
rather symmetrical way. The metrics that we obtain are equivalent to the previously-known Kerr
de SitterTaubNUT metrics in D = 4. In D 6 the extra parameter that we introduce gives rise
to non-trivial generalisations of the Kerrde Sitter metrics. The new parameter is associated with
characteristics that generalise those of TaubNUT like metrics in four dimensions, and so we
may think of it as being a higher-dimensional generalisation of the NUT parameter. In each of
the odd- and even-dimensional cases, we discuss also their supersymmetric limits. In odd dimensions, these yield, after Euclideanisation, new EinsteinSasaki metrics. In even dimensions, the
supersymmetric limit leads to new Rici-flat Khler metrics.
In Section 4, we discuss some global aspects of the new KerrAdSTaubNUT metrics. In
particular, in the case of even dimensions, the introduction of the NUT-type parameter implies that the time coordinate must be identified periodically, in the same way as happens in
the previously-known four-dimensional solutions. By contrast, we find that in odd dimensions
one can define a time coordinate that is not periodic.
In Section 5, we discuss the case of five dimensions in detail. We find that in this case, the
new NUT-type parameter is actually bogus, in the sense that it can be removed by using a scaling
symmetry that is specific to the five-dimensional metric. In the process of showing this, however,
40
we uncover an intriguing and previously unnoticed property of the five-dimensional KerrAdS

metric. We find that it has an inversion symmetry, which implies that the metric with large
values of its rotation parameters is equivalent, after a general coordinate transformation, to the
metric with small values for the rotations. The fixed point of this symmetry occurs at the critical value of rotation that arises in the supersymmetric limit. This corresponds to the case where
the rotation parameter is equal to the radius of the asymptotically AdS spacetime. The inversion symmetry is therefore a feature specifically of the five-dimensional Kerr black holes with a
cosmological constant, and does not arise in the case of asymptotically flat black holes.
The paper ends with conclusions in Section 6.
2. Kerrde Sitter with NUT parameter in D = 2n + 1
2.1. The metric
We take as our starting point the general Kerrde Sitter metric in D = 2n + 1 dimensions,
which was constructed in [9,10]. Specifically, we begin with the metrics written in an asymptotically non-rotating frame, as given in Eq. (E.3) of [9], specialised to the case of odd dimensions
D = 2n + 1. We choose the cosmological constant to be negative, with the Ricci tensor given by
R = (D 1)g 2 g . The constant g is the inverse of the AdS radius. The metric
is described
in terms of n latitude or direction cosine coordinates i , subject to the constraint ni=1 2i = 1,
n azimuthal coordinates i , the radial coordinate r and time coordinate t . It has (n + 1) arbitrary
parameters M and ai , which can be thought of as characterising the mass and the n angular
momenta in the n orthogonal spatial 2-planes.
In order to find a generalisation that includes a NUT-type parameter, we first specialise the
KerrAdS metrics by setting
a1 = a2 = = ap = a,
ap+1 = ap+2 = = an = b.
(1)
We then reparameterise the latitude coordinates as

i = i sin ,
1 i p,
p
i2 = 1,
i=1
j +p = j cos ,
1 j q,
q
j2 = 1,
(2)
j =1
where we have defined

n = p + q,
(3)
and we also then introduce a coordinate v in place of , defined by

a 2 cos2 + b2 sin2 = v 2 .
(4)
It is convenient to divide the original n azimuthal coordinates i into two sets, with p of them
denoted by i and the remaining q denoted by j .
Because of the specialisation of the rotation parameters in (1), the KerrAdS metric will now
have cohomogeneity 2, rather than the cohomogeneity n of the general (2n + 1)-dimensional
KerrAdS metrics. In fact, as we shall see explicitly below, the metric has homogeneous level sets
R S 2p1 S 2q1 , with the metric functions depending inhomogeneously on the coordinates r
41
and v. Remarkably, the form in which the metric can now be written puts the radial coordinate r
and the coordinate v on a parallel footing, and suggests a rather natural generalisation in which a
NUT-type parameter L can be introduced. Rather than writing the metric first without the NUT
contribution and then again with it added, we shall just directly present our final result with
the NUT parameter included. The original KerrAdS, subject to the constraints on the rotation
parameters specified in (1), corresponds to setting L = 0. Our generalised metric including L is
(1 + g 2 r 2 )(1 g 2 v 2 ) 2 2n2 dr 2 2n2 dv 2
dt +
+
a b
U
V
2
2

2
2
2M (1 g v )
2L
(1 + g 2 r 2 )
+ 2n2
dt A + 2n2
dt A
a b
a b
ds 2 =

(r 2 + a 2 )(a 2 v 2 ) 2
di + i2 di2
2
2
a (a b )
p
+
+
i=1
q
(r 2 + b2 )(b2 v 2 )
(b2
a2)

d j2 + j2 d j2 ,
(5)
j =1
where
b(b2 v 2 ) 2
a(a 2 v 2 ) 2
d
+
j d j ,
i
i
a (a 2 b2 )
b (b2 a 2 )
A=
a(r 2 + a 2 )
A =
a (a 2 b2 )
i=1
j =1
p

i=1
b(r 2 + b2 ) 2
j d j ,
b (b2 a 2 )
q
i2 di +
j =1
(1 + g 2 r 2 )(r 2 + a 2 )p (r 2 + b2 )q
2M,
r2
(1 g 2 v 2 )(a 2 v 2 )p (b2 v 2 )q
+ 2L,
V =
v2

p1 2
q1

2n2 = r 2 + v 2 r 2 + a 2
r + b2
,

p1
q1
2n2 = r 2 + v 2 a 2 v 2
b2 v 2
,
U=
a = 1 a 2 g 2 ,
b = 1 b 2 g 2 .
(6)
It is straightforward (with the aid of a computer) to verify in a variety of low odd dimensions
that the metric (5) does indeed solve the Einstein equations R = (D 1)g 2 g , and since
the construction does not exploit any special features of the low dimensions, one can be confident that the solution is valid in all odd dimensions. We have explicitly verified the solutions in
D 9.
As we indicated above, the metric (5) can be re-expressed more elegantly in terms of two
complex projective spaces CPp1 and CPq1 . The proof is straightforward, following the same
steps as were used in [9] when studying the Kerrde Sitter metrics with equal angular momenta.
The essential point is that one can write
p

p
i=1
i=1

2
2
di2 + i2 di2 = d2p1
= (d + A)2 + dp1
,
i2 di = d + A,
(7)
42
2
where dp1
is the standard FubiniStudy metric on CPp1 (with Rab = 2pgab ), and 12 dA
2
locally gives the Khler form J . Note that d2p1
is the standard metric on the unit sphere
p1
2p1
, expressed here as the Hopf fibration over CP
.
S
With these results, and the analogous ones for the tilded coordinates j and j , we find that
(5) can be rewritten as
(1 + g 2 r 2 )(1 g 2 v 2 ) 2 2n2 dr 2 2n2 dv 2

dt +
+
a b
U
V
2
2

2
2
2L
(1 + g 2 r 2 )
2M (1 g v )
dt A + 2n2
dt A
+ 2n2
a b
a b

(r 2 + a 2 )(a 2 v 2 )
2
+
(d + A)2 + dp1
2
2
a (a b )

(r 2 + b2 )(b2 v 2 )
2
2 + d q1
,
+
(d + A)
2
2
b (b a )
ds 2 =
(8)
now with
a(a 2 v 2 )
(d + A) +
a (a 2 b2 )
a(r 2 + a 2 )
(d + A) +
A =
a (a 2 b2 )
A=
b(b2 v 2 )
(d + A),
b (b2 a 2 )
b(r 2 + b2 )
(d + A).
b (b2 a 2 )
(9)
Here A and A are potentials such that the Khler forms of the complex projective spaces CPp1
respectively. Another useful way of
and CPq1 are given locally by J = 12 dA and J = 12 A,
writing the metric is given in Appendix A.
It can be seen from the form of (8) that the metrics have cohomogeneity 2, with principal orbits
on the surfaces where r and v are constant that are the homogeneous spaces R S 2p1 S 2q1 .
The R factor is associated with the time direction, whilst the spheres S 2p1 and S 2q1 arise from
the Hopf fibrations over CPp1 and CPq1 , respectively. The sphere metrics on the principal
orbits are squashed, and so the isometry group of (8) is R U (p) U (q).
We have presented the new solutions for the case of negative cosmological constant, but
clearly these NUT generalisations of KerrAdS will also be valid if we send g ig, yielding NUT generalisations of the Kerrde Sitter metrics. It is also worth noting that even when the
cosmological constant is set to zero, the solutions are still new, representing NUT generalisations
of the asymptotically-flat rotating black holes of Myers and Perry [7].
Written in the form (5) or (8), the metric appears to be singular in the special case where one
sets a = b. This is, however, an artifact of our introduction of the coordinate v, in place of . We
did this in order to bring out the symmetrical relation between r and v, but clearly, as can be seen
from (4), the coordinate v degenerates in the case a = b. This can be avoided by using as the
coordinate instead, and performing appropriate rescalings.
Having written our new KerrAdSTaubNUT metrics in this form, it is clear that we could
obtain more general Einstein metrics by replacing the FubiniStudy metrics d 2 and d 2
p1
q1
on CPp1 and CPq1 by arbitrary EinsteinKhler metrics of the same dimensions, and nor2
2 . In the generalised
malised to have the same cosmological constants as dp1
and d q1
metrics, A and A will now be potentials yielding the Khler forms of the two EinsteinKhler
metrics, i.e., J = 12 dA and J = 12 d A.
43
If we specialise to the case when b = 0, and define a new coordinate = ag 2 t , then the
metric (9) reduces to
2

r 2 + v2 2 r 2 + v2 2
X
a2 v2

(d
+
A)
ds 2 =
dr +
dv 2
dt
X
Y
aa
r + v2
2

2
2
2
Y
(r + a 2 )(a 2 v 2 )
r +a

2
+ 2
(d
+
A)
+
dp1
dt
aa
r + v2
a 2 a
+
2
where d2q1
r 2v2
2
d2q1
,
a2
2 + d 2
= (d + A)
q1
(10)
is the metric of a unit sphere S 2q1 , and

X = 1 + g2r 2 r 2 + a2
2M
,
(r 2 + a 2 )p1 r 2(q1)

2L
Y = 1 g2v2 a2 v2 2
.
2
(a v )p1 v 2(q1)
(11)
The constant L is related to the original NUT parameter by L = (1)q L. A special case of the
metrics (10), namely when p = 1, was obtained in [11,12].
2.2. The supersymmetric limit
Odd-dimensional KerrAdS black holes admit supersymmetric limits,1 which in Euclidean
signature with positive cosmological constant become EinsteinSasaki metrics [13,14] (see also
[15,16] for discussions of how the supersymmetric limit arises in the Lorentzian regime, when
a Bogomolnyi inequality is saturated). We find that an analogous limit also exists for our new
metrics where the NUT charge is introduced. We first set g = i so that the metric has a unit
positive cosmological constant (R = (D 1)g ). We then Euclideanise the metric by sending
t it,
a ia,
b ib,
(12)
define
1 a 2 = ,
1 b2 = ,
1 r 2 = x,
1 + v 2 = y,
M = m n+1 ,
L = n+1 ,
(13)
and then take the limit 0. This leads to the metric

2
( x)( y)
( x)( y)
ds 2 = dt +
(d + A)
(d + A)
( )
( )
x y 2 x y 2 ( x)( y)
( x)( y)
2
2
+
dx +
dy +
dp1
d q1
4X
4Y
( )
( )
1 To be more precise, supersymmetric limits exist in those dimensions, such as D = 5, where there is an appropriate
gauged supergravity theory admitting the KerrAdS black hole as a solution. More generally, in arbitrary odd dimensions,
we use the term supersymmetric limit to characterise the fact that there exists a Killing spinor , satisfying +
(ig/2) = 0.
44

X
( y)
(d + A)
x y ( )

Y
( x)
+
(d + A)
x y ( )
2
( y)
(d + A)
( )
2
( x)
,
(d + A)
( )
(14)
where again J = 12 dA and J = 12 d A are the Khler forms of the CPp1 and CPq1 complex
2
2
and d q1
respectively, and
projective spaces with metrics dp1
X=
2m
(
x)p1 (
x)q1
x( x)( x),
2
+ y( y)( y).
(15)
( y)p1 ( y)q1
It is straightforward to verify that the above metric (14) is an EinsteinSasaki metric in
D = 2n + 1 dimensions. Specifically, this means that the metric admits a Killing spinor ,
satisfying + (i/2) = 0. Note that the metric has the form
Y=
2
2
= (dt + 2A)2 + ds2n
,
ds2n+1
(16)
where
is an EinsteinKhler metric and A is the corresponding Khler potential, in the
2 can be written locally as J = dA. As far as we know, these
sense that the Khler form for ds2n
2 have not been obtained explicitly before. Note
cohomogeneity-2 EinsteinKhler metrics ds2n
2
that one can go to the Ricci-flat limit of ds2n by performing a rescaling that amounts to dropping
the x 3 term and y 3 term in (15).
If we consider the special case where p = n 1 and q = 1, the EinsteinSasaki metrics reduce
to ones that were obtained recently in [17]. This may be seen by defining new parameters by the
expressions
2
ds2n
1
m = (1)N ,
8
where N p 1, and introducing new coordinates defined by
= 4( 2),
x = x ,
= ( ),
y = y ,
t = + 2( ),
= 2 + 2.
1
= (1)N ,
8
= 2,
Defining also X = 4X and Y = 4Y , we obtain, upon substitution into (14), the metric

2x y 2 x y 2 x y 2
2
ds = d 2(x + y)
d +
+
d x +
d y
X
Y

2
2
X
Y
x y
y
x
+
d +
d +
dN2 ,
x y
x
(17)
(18)
(19)
where = d + 12 A, and
Y = 4y 3 + y 2 + 4 y + N .
X = 4x 3 x 2 4 x N ,
(20)
x
y
This is precisely of the form of the EinsteinSasaki metrics that were obtained in Section 4 of
Ref. [17], in the case where the EinsteinKhler base metric in that paper is taken to be CPN .
A detailed discussion of the global structure of these metrics was given in [17], and new complete
D = 7 EinsteinSasaki spaces were obtained.
45
3. Kerrde Sitter with NUT parameter in D = 2n

The Kerrde Sitter metrics in even spacetime dimensions take a slightly different form from
those in odd dimensions. The reason for this is that now there are an odd number of spatial dimensions, and so there can be (n 1) independent parameters characterising rotations in (n 1)
orthogonal 2-planes, with one additional spatial direction that is not associated with a rotation.
Because of this feature, the D = 2n dimensional Kerrde Sitter black holes in general have cohomogeneity n, which can be reduced to cohomogeneity 2 if one sets all the (n 1) rotation
parameters equal. By contrast, in odd dimensions D = 2n + 1 the general metrics have cohomogeneity n, reducing to cohomogeneity 1 if one sets all the rotation parameters equal.
It will be recalled that in Section 2, we were able to generalise the odd-dimensional Kerr
de Sitter to include a NUT parameter by dividing the angular momentum parameters ai into two
sets, equal within a set, thereby obtaining a metric of cohomogeneity 2. Our construction with
the NUT parameter is intrinsically adapted to metrics of cohomogeneity 2, and so this means that
in the present case, when we consider generalising the even-dimensional Kerrde Sitter metrics,
we shall first need to divide the rotation parameters ai into two sets. In one set, the parameters
will be equal and non-zero, while in the other set, the remaining rotation parameters will all be
chosen to be zero.
Our starting point is the expression for the Kerrde Sitter metrics given in Eq. (E.3) of Ref. [9],
specialised to dimension D = 2n. We shall take the cosmological constant to be negative, with
the resulting KerrAdS metrics satisfying R = (D 1)g 2 g . We then set
a1 = a2 = = ap = a,
ap+1 = ap2 = = an1 = 0.
(21)
We then introduce new latitude coordinates i , j and , in place of the i in [9],

i = i sin ,
1 i p,
p
i2 = 1,
i=1
j +p = j cos ,
1 j n p,
np
j2 = 1.
(22)
j =1
In this case, because there are only (n 1) azimuthal coordinates i , we split them into two sets,
which we shall denote by i and j , defined for
i :
1 i p,
j :
1 j q,
(23)
where this time we have defined q such that

p + q = n 1.
(24)
We then introduce a new variable v, in place of , which this time is defined by

a 2 cos2 = v 2 .
(25)
We can now write out the KerrAdS metric of [9,10], subject to the restriction (21), in terms
of the new variables defined above, and, as in the odd-dimensional case we discussed previously,
this allows us to conjecture a generalisation that includes a NUT parameter L as well as the mass
parameter M and angular momentum parameter a. Again, we shall just present our final result, having included the NUT parameter. Thus we obtain the new KerrAdSTaubNUT metric
46
(which we have verified explicitly in D 8)

(1 + g 2 r 2 )(1 g 2 v 2 ) 2 2n3 dr 2 2n3 dv 2
dt +
+
a
U
V

2

2
2
2
2Mr (1 g v )
2Lv (1 + g 2 r 2 )
+ 2n3
dt A 2n3
dt A
a
a
p

(r 2 + a 2 )(a 2 v 2 ) 2
+
di + i2 di2
a 2 a
i=1

q
2
2

2

r v
2
2 2
+ 2 d vq+1 +
d j + j d j ,
a
ds 2 =
(26)
j =1
where
r 2 + a2 2
a2 v2 2
i di ,
A =
i di ,
A=
aa
aa
i=1
i=1

p

U = 1 + g 2 r 2 r 2 + a 2 r 2q 2Mr,

p

V = 1 g 2 v 2 a 2 v 2 v 2q 2Lv.

p1 2q

2n3 = r 2 + v 2 r 2 + a 2
r ,
a = 1 a 2 g 2 ,

p1 2q
2n3 = r 2 + v 2 a 2 v 2
v .
p
(27)
The (2n)-dimensional KerrAdSTaubNUT metrics that we have constructed here can be

seen to be quite similar in structure to the (2n + 1)-dimensional examples that we constructed
in Section 2, in the special case where we set the b parameter to zero. In fact we can re-express
the metrics (26) in terms of a complex projective space and a sphere metric, in a manner that is
closely analogous to (10). This is expressed most simply by making redefinitions as in (7), and
then introducing a new Hopf fibre coordinate = ag 2 t as we did in the odd-dimensional
case. Having done this, we arrive at the metric
2

r 2 + v2 2 r 2 + v2 2
X
a2 v2
ds 2 =
(d
+
A)
dr +
dv 2
dt
X
Y
aa
r + v2
2

2
2
2
Y
(a + r 2 )(a 2 v 2 )
a +r
2
+ A) +
+ 2
(d
dp1
dt
aa
r + v2
a 2 a
+
r 2v2
2
d2q
,
a2
(28)
2 is the metric on the unit sphere S 2q ,

where d2q

X = 1 + g2r 2 r 2 + a2
2Mr
,
+ a 2 )p1 r 2q

2Lv
,
Y = 1 g2v2 a2 v2 2
(a v 2 )p1 v 2q
(r 2
(29)
2
is given locally by J = 12 dA.
and the Khler form J for the CPp1 metric dp1
For the cases with q = 0, there can also be a BPS limit of the solutions, giving rise to Ricci-flat
Khler metrics instead of EinsteinKhler. To do this, we first Euclideanise the metric by setting
47
t it, a ia and set g = i. We then take the following limit

1 a 2 = ,
M = ( )p1 ,
1 r 2 = x ,
1 + v 2 = y ,
L = i p1 ,
(30)
with 0. The metric becomes ds 2 = d s 2 , where d s 2 is a Ricci-flat Khler metric, given by

x y 2 x y 2 (x )( y)
2
dx +
dy +
dp1
4X
4Y

2

2
X
Y
y
x
+
dt +
(d + A) +
dt
(d + A) ,
x y
x y
2
2
,
Y = y( y)
.
X = x(x ) +
(x )p1
( y)p1
d s 2 =
(31)
The Khler 2-form is given locally by J = dB, where

1
(x )( y)
B = (x + y) dt +
(d + A).
2
2
(32)
4. Global analysis
The global analysis of KerrAdS black holes in general dimensions was given in [9,10]. Here,
we study the effect of introducing the NUT charge L. We shall consider the case where v is a
compact coordinate, ranging over the interval 0 < v1 v v2 , where v1 and v2 are two adjacent
roots of V (v) = 0, such that the function V is positive when v lies within the interval. In the
case when L = 0, we would have v1 = a and v2 = b. The coordinate r ranges from r0 to infinity,
where r0 is the largest root of U (r) = 0. It is evident that when L = 0, the geometries of the
constant-time surfaces at fixed radius are modified from their L = 0 form, and thus in particular
the metrics in general no longer approach AdS as r goes to infinity. The discussion now divides
into the cases of D = 2n + 1 dimensions and D = 2n dimensions.
4.1. D = 2n + 1 dimensions
The metric (8) is degenerate at v = v1 and v2 , where V (vi ) = 0. The corresponding Killing
vectors whose norms 2 = g vanish at these surfaces have the form
= 0
+ 1
+ 2
,
t
(33)
for constants 0 , 1 and 2 to be determined. The associated surface gravities are of Euclidean
type, in the sense that

g ( 2 )( 2 )
2
E =
(34)

42
v=vi
is positive. Thus these degenerations are typical of an azimuthal coordinate at a spatial origin.
We can scale the coefficients i so that the Euclidean surface gravity is 1, implying that the
Killing vector generates a closed translation with period 2 . One might conclude that the time
coordinate is periodic, since 0 is non-vanishing. This is indeed the case for the solutions in
even dimensions. However, in odd dimensions the /t term can be removed by making the
48
coordinate transformation
t = t +
b (a 2 v12 )(b2 v22 )b a (b2 v12 )(b2 v22 )a

ab(a 2 b2 )(1 g 2 v12 )(1 g 2 v22 )
(35)
The two Killing vectors whose norms vanish at v1 and v2 are now given by

4L
b
a
i =
+
.
V (vi ) b2 vi2 a 2 vi2
(36)
Both Killing vectors have unit Euclidean surface gravity, implying that they both generate closed
2 translations. Since it does not suffer a periodic identification, t is perhaps a more natural
choice than t for the time coordinate.
The metric also degenerates at r = r0 , and the corresponding null Killing vector has
Lorentzian surface gravity , in the sense that 2 = E2 is positive. Thus r = r0 is an horizon. If we write the null Killing vector in terms of coordinate t, normalised to
0 =
+ 1
+ 2
,
(37)
where 1 and 2 are determined from the condition that 20 = 0 at r = r0 , we find that the surface
gravity is given by
=
(1 g 2 v12 )(1 g 2 v22 )(r02 + a 2 )(r02 + b2 )(1 + g 2 r02 )U (r0 )

2a b (r02 + v12 )(r02 + v22 )[U (r0 ) + 2M]
(38)
If instead we consider the null Killing vector in terms of the original coordinate t , and rescale it
to give
0 =
+ 1
+ 2
,
t
(39)
then the surface gravity is then given by

=
(1 + g 2 r02 )U (r0 )
,
2[U (r0 ) + 2M]
(40)
which is identical to the result for the KerrAdS black hole [9,10] without the NUT parameter.
It is not a priori obvious what the proper normalisation for the asymptotically timelike Killing
vector should be, since the metrics with the non-vanishing NUT parameter are not asymptotic to
AdS.
4.2. D = 2n dimensions
In even dimensions, the introduction of the NUT parameter implies that the time coordinate
is necessarily periodic (as in four dimensions). To see this, we note from the metric (28) that, at
the degenerate surfaces v = v1 and v2 , the Killing vectors whose norms vanish are given by

2
2
2
i =
(41)
.
a vi
+ aa
V (vi )
t

These Killing vectors are normalised to have unit Euclidean surface gravities, and hence they
generate closed translations with period 2 . In the case when L = 0, then v1 = a and v2 =
49
a, so the i do not have /t terms. However, when L = 0 there are necessarily /t terms
appearing in these Killing vectors that generate periodic translations, and so t must be identified
periodically.
5. Inversion symmetry of D = 5 KerrAdS black holes
In this section, we first demonstrate that the NUT parameter L introduced in our general rotating black holes is trivial in the special case of D = 5 dimensions. However, our demonstration
also brings to light a rather remarkable property of the five-dimensional KerrAdS black hole
metric, namely, that it admits a discrete symmetry transformation which shows that the metric
with over-rotation (where the parameters a and b are such that a 2 g 2 > 1 and/or b2 g 2 > 1) is
equivalent to a KerrAdS metric with under-rotation.
We start with the five-dimensional KerrAdS metric written in the (8) with p = 1 and q = 1,
and make the coordinate transformations

ab2 + ag 2 t + a 1 + b2 g 2 ,
ba 2 + bg 2 t + b 1 + a 2 g 2 ,

t t + a 2 b2 + a 2 + b2 ,
(42)
and define r 2 = x and v 2 = y. This leads to the five-dimensional metric
2

dx
Y
dy 2
X
+
(dt + y d)2 +
(dt x d)2
ds 2 = (x + y)
4X
4Y
x(x + y)
y(x + y)
2
a 2 b2
+
dt xy d (x y) d ,
xy
(43)
where

X = 1 + g 2 x x + a 2 x + b2 2Mx

= g 2 x 3 + 1 + a 2 + b2 g 2 x 2 + a 2 + b2 + a 2 b2 g 2 2M x + a 2 b2 ,

Y = 1 g 2 y a 2 y b2 y + 2Ly

= g 2 y 3 1 + a 2 + b2 g 2 y 2 + a 2 + b2 + a 2 b2 g 2 + 2L y a 2 b2 .
(44)
Although, the solution ostensibly has the four independent parameters (M, L, a, b), one can
in fact scale away either M or L in this five-dimensional case. To do this, we set
t
(45)
t = ,
= 3 .
= 5 ,
The metric (43) is invariant under this transformation, if we simultaneously transform the parameters a, b, M and L. Thus we define X = 6 X and Y = 6 Y , where X and Y are defined as in
M and L.
It follows that we shall have
(44) except with tilded parameters a,
b,

2 + 2 a 2 + b2 g 2 = 1 + a 2 + b 2 g 2 ,
6 a 2 b2 = a 2 b 2 ,

4 a 2 + b2 + a 2 b2 g 2 + 2L = a 2 + b 2 + a 2 b 2 g 2 + 2L,

4 a 2 + b2 + a 2 b2 g 2 2M = a 2 + b 2 + a 2 b 2 g 2 2M.
(46)
x = 2 x,
y = 2 y,
M
We can then choose, for example, to set L = 0, and solve the four equations (46) for a,
b,
and . Thus a solution with L = 0 is transformed into a tilded solution with L = 0, and since this
50
latter solution is just of the original five-dimensional KerrAdS form, it follows that the metric
(43), even with L = 0, is also just the five-dimensional KerrAdS metric, but with changed values
for the rotation and mass parameters. It is nevertheless interesting that the Kerrde Sitter black
hole in D = 5 can be put in such a symmetric form.
It should be stressed that the scaling symmetry that we used above in order to show that the
parameter L in the five-dimensional metrics is trivial is very specific to five dimensions. In
particular, it can be seen from (8) that in higher dimensions, when at least one of p or q exceeds
1, the associated metrics on the complex projective spaces will break the scaling symmetry.
Thus, as in the case of the simpler NUT generalisations discussed [11,12], five dimensions is the
exception in not admitting a non-trivial generalisation.
The transformation described above becomes particularly simple if we consider the case of an
asymptotically flat five-dimensional rotating black hole, i.e. when g = 0. In this case, we have
from (46) that = 1 and
a 2 + b 2 + 2L = a 2 + b2 + 2L,
a 2 b 2 = a 2 b2 .
a 2 + b 2 2M = a 2 + b2 2M,
(47)
Thus L + M = L + M, and so we can arrange to have L = 0 by taking M = L + M, implying

that a 2 + b 2 = a 2 + b2 + 2L, together with a 2 b 2 = a 2 b2 . It is worth noting, however, that even
though one can always map into a solution where L = 0, it may, depending upon the original
Although the metric
values for a, b and L, correspond to having complex values for a and b.
(43) would still be real, the metric written back in terms of the original , and t coordinates
would then be complex. Thus although the parameter L is really trivial in five dimensions, its
inclusion can nevertheless allow one to parameterise the solutions in a wider class without the
need for complex coordinate transformations. Similar remarks apply also to the case when g = 0.
There is another interesting consequence of the five-dimensional scaling symmetry discussed
above, namely, that even with the parameter L omitted entirely, the five-dimensional rotating
AdS black hole metrics have a symmetry that allows one to map an over-rotating black hole
(i.e., where a 2 g 2 > 1 or b2 g 2 > 1) into an under-rotating black hole. This can be understood by
again considering the transformations in (46), where we now choose not only L = 0 but also
M,
) (where we
L = 0. The system of equations then admits a sextet of solutions for (a,
b,
assume, without loss of generality, that the signs of the rotation parameters are unchanged):
a = a,
a = b,
1
a = 2 ,
ag
1
a = 2 ,
bg
a
a = ,
bg
b
,
a =
ag
b = b,
b = a,
b
b =
,
ag
a
b = ,
bg
1
b = 2 ,
bg
1
b = 2 ,
ag
M = M,
M = M,
= 1,
= 1,
M
M = 4 4 ,
a g
M
M = 4 4 ,
b g
M
M = 4 4 ,
b g
M
M = 4 4 ,
a g
1
,
ag
1
= ,
bg
1
= ,
bg
1
=
.
ag
=
(48)
The first of these is the identity, the second is merely an exchange of the rles of a and b, whilst
the remaining four, modulo exchanges of the as and the bs, are equivalent and non-trivial.
Taking the third as an example, we see that if the metric is over-rotating by virtue of having
51
a 2 g 2 > 1, then it can be re-expressed, by a change of variables, as a metric which is underrotating. In fact any five-dimensional KerrAdS black hole with over-rotation is equivalent, after
a change of coordinates, to one with under-rotation. Of course, after transforming back into the
original coordinates in which the over-rotating black hole ostensibly exhibited singular behaviour, one would find that the coordinate ranges that actually reveal that it is well-behaved are not
the naive ones that led to the original conclusion of singular behaviour.
It is instructive to rewrite the transformations (48) in terms of the original coordinates of the
five-dimensional KerrAdS metric as given by Hawking, Hunter and Taylor-Robinson in [8]. The
metric is given by
2
2

b cos2
sin2
a sin2
r 2 + a2
ds52 = 2 dt
d
d +
d
a
dt
a
b
a
2

2
2
2
2
2
2
2
2
cos
dr
d
r +b
+
d +
+
b dt
2
b
2

2
2
2
2
2
(1 + g r )
a(r 2 + b2 ) cos2
b(r + a ) sin
+
d
d
,
(49)
ab
dt
a
b
r 22
where

1 2
r + a 2 r 2 + b2 1 + g 2 r 2 2M,
2
r
1 a 2 g 2 cos2 b2 g 2 sin2 ,
2 r 2 + a 2 cos2 + b2 sin2 ,
a 1 a 2 g 2 ,
b 1 b 2 g 2 .
(50)
It satisfies R = 4g 2 g . Taking the transformation in the third line of (48) as an example,

we find that after re-expressing our results back in terms of the quantities in (49), the symmetry
transformation amounts to
1
,
ag 2
1
,
ag
a
1
r,
ag
b
,
ag
M
,
a4g4
b
1
,
t agt + ,
a
g

a 1/2
cos .
cos 1
b
(51)
It is straightforward to see that this transformation leaves the metric in (49) invariant, and that
it therefore allows one to map an over-rotating KerrAdS metric into an under-rotating one. In
other words, if we perform the transformation of parameters given in the first line in (51), then
the metric is restored to its original form by making the general coordinate transformations given
also in (51).
Another way of expressing this result is that for any given values of a and b, and provided
one allows the coordinates to take complex values in general, then there exist real sections of the
complex metric describing KerrAdS black holes with under-rotation, and also real sections of
the same metric that describe KerrAdS black holes with over-rotation.
It is instructive also to re-express the coordinate transformations in (51) in terms of the coordinates y and rather than r and , where y and are the coordinates with respect to which the
52
conformal boundary of the KerrAdS metric is precisely the standard R S 3 Einstein universe,
with a round S 3 factor. They are defined by [8]

a y 2 sin2 = r 2 + a 2 sin2 ,
(52)
b y 2 cos2 = r 2 + b2 cos2 .
Applying the transformations in (51), we find that these imply the coordinate transformations

1
1
2
2
2
2
tan 1 + 2 2 sec2 .
y 2 y sin ,
(53)
g
g y
This result emphasises that the original y = constant boundary, which is the most natural choice
from the AdS/CFT point of view [8,18], is quite different from the y =constant boundary of the
transformed metric.
A number of remarks are in order. First, we note that the symmetry we are discussing, which
can be expressed in terms of dimensionless quantities as ag 1/(ag), exists only in the case
of the rotating black hole with a cosmological constant. In the case of asymptotically-flat black
holes, for which g = 0, there is no inversion symmetry. The inversion symmetry for the fivedimensional KerrAdS black hole is reminiscent of a T-duality symmetry, in the sense that it
implies there is a maximum allowed value for the rotation, namely a 2 g 2 = 1. In fact, this value is
associated with the supersymmetric limit. If one considers the case where a rotation parameter is
becoming very large, i.e., a 2 g 2 1, then it can be seen from (51) that in the limiting case when
a 2 g 2 approaches infinity, the metric will actually approach the pure AdS metric.
It is interesting also to consider the effect on the canonical AdS metric of the transformations
(53) taken in isolation. In other words, we start with the AdS metric

ds 2 = 1 + g 2 y 2 dt 2 +

dy 2
+ y 2 d 2 + sin2 2 d 2 + cos2 d 2 ,
2
2
1+g y
(54)
and impose just the coordinate transformations given in (53) (which are independent of the rotation parameters a and b). Upon doing so, we find that the AdS metric (54) transforms according
to
ds 2

1
dy 2
1 + g 2 y 2 d 2 +
+ y 2 d 2 + sin2 2 g 2 dt 2 + cos2 d 2 .
2
2
2
g
1+g y
(55)
This is identical in form to (54), with the rles of and gt exchanged. It can easily be seen that
in terms of the standard embedding of AdS5 in R4,2 , the transformation (53) corresponds to exchanging the rles of the two timelike embedding coordinates with a pair of spacelike embedding
coordinates.
6. Conclusions
In this paper, we have constructed generalisations of certain Kerrde Sitter and KerrAdS
black holes in all dimensions D 6, in which an additional NUT-type parameter is introduced.
Specifically, the cases where we have obtained the more general solutions are where the rotation
parameters are specialised so that the metrics have cohomogeneity 2. The nature of the generalisation is then analogous to the way in which a NUT parameter can be introduced in the
four-dimensional Kerrde Sitter metrics.
The same procedure can be followed also in five dimensions, but in this case we find that
the additional NUT parameter is trivial, in the sense that it can be absorbed by a rescaling of
parameters and coordinates. However, we also found that there exists a remarkable symmetry of
53
the five-dimensional KerrAdS metrics, in which one can map a solution where one or both of the
rotation parameters are large (the case of over-rotation, where a 2 g 2 > 1 and/or b2 g 2 > 1) into
a solution where the rotation parameters are small (i.e. under-rotation). This means that there
is effectively a maximum rotation possible, corresponding to the supersymmetric case where
a 2 g 2 = 1 or b2 g 2 = 1.
We also studied the supersymmetric limits of the new Kerrde SitterTaubNUT metrics,
showing that after Euclideanisation we can obtain new cohomogeneity-2 EinsteinSasaki metrics
in all odd dimensions D 7, and new cohomogeneity-2 Ricci-flat Khler metrics in all even
dimensions D 6.
Acknowledgements
We than Zhiwei Chong and Justin Vzquez-Poritz for helpful discussions.
Appendix A. Another form for the odd-dimensional metrics
If we perform the same angular redefinitions (42) in the general odd-dimensional metrics (8),
they may be re-expressed as
ds 2 =
r 2 + v 2 2 r 2 + v 2 2 (r 2 + a 2 )(a 2 v 2 )
2
dp1
dr +
dv +
X
Y
a (a 2 b2 )
+
+
(r 2 + b2 )(b2 v 2 ) 2
d q1
b (b2 a 2 )

2
a 2 b2
(r 2 + a 2 )(a 2 v 2 )
2
2 2
A
d
v
v
d
dt
r
r 2v2
aa (a 2 b2 )

(r 2 + b2 )(b2 v 2 ) 2
B
bb (b2 a 2 )
2

b(b2 v 2 )
X
a(a 2 v 2 )
2
A
B
d
dt
+
v
r 2 + v2
a (a 2 b2 )
b (b2 a 2 )
2

b(r 2 + b2 )
Y
a(r 2 + a 2 )
2
A
B ,
dt r d
r 2 + v2
a (a 2 b2 )
b (b2 a 2 )
(A.1)
where we have defined X and Y as

X
U
(r 2
+ a 2 )p1 (r 2
+ b2 )q1
(1 + g 2 r 2 )(r 2 + a 2 )(r 2 + b2 )
2M
2
,
r2
(r + a 2 )p1 (r 2 + b2 )q1
V
Y 2
(a v 2 )p1 (b2 v 2 )q1
=
(1 g 2 v 2 )(a 2 v 2 )(b2 v 2 )
2L
+ 2
.
2
2
p1
v
(a v )
(b2 v 2 )q1
This form can sometimes be useful, since it is expressed in a manifest vielbein basis.
(A.2)
54
References
[1] R.P. Kerr, Gravitational field of a spinning mass as an example of algebraically special metrics, Phys. Rev. Lett. 11
(1963) 237.
[2] K. Schwarzschild, Sitzungsber. Preuss. Akad. Wiss., Kl. Math.-Phys. Tech. 189 (1916).
[3] R. Couch, K. Chinnapared, A. Exton, E.T. Newman, A. Prakash, R. Torrence, Metric of a rotating, charged mass,
J. Math. Phys. 6 (1965) 918.
[4] B. Carter, HamiltonJacobi and Schrdinger separable solutions of Einsteins equations, Commun. Math. Phys. 10
(1968) 280.
[5] J.F. Plebanski, A class of solutions of EinsteinMaxwell equations, Ann. Phys. 90 (1975) 196.
[6] J.F. Plebanski, M. Demianski, Rotating, charged, and uniformly accelerating mass in general relativity, Ann.
Phys. 98 (1976) 98.
[7] R.C. Myers, M.J. Perry, Black holes in higher-dimensional spacetimes, Ann. Phys. 172 (1986) 304.
[8] S.W. Hawking, C.J. Hunter, M.M. Taylor-Robinson, Rotation and the AdS/CFT correspondence, Phys. Rev. D 59
(1999) 064005, hep-th/9811056.
[9] G.W. Gibbons, H. L, D.N. Page, C.N. Pope, The general Kerrde Sitter metrics in all dimensions, J. Geom. Phys. 53
(2005) 49, hep-th/0404008.
[10] G.W. Gibbons, H. L, D.N. Page, C.N. Pope, Rotating black holes in higher dimensions with a cosmological constant, Phys. Rev. Lett. 93 (2004) 171102, hep-th/0409155.
[11] D. Klemm, Rotating black branes wrapped on Einstein spaces, JHEP 9811 (1998) 019, hep-th/9811126.
[12] Z.W. Chong, G.W. Gibbons, H. L, C.N. Pope, Separability and Killing tensors in KerrTaubNUTde Sitter metrics
in higher dimensions, Phys. Lett. B 609 (2005) 124, hep-th/0405061.
[13] M. Cvetic, H. L, D.N. Page, C.N. Pope, New EinsteinSasaki spaces in five and higher dimensions, Phys. Rev.
Lett. 95 (2005) 071101, hep-th/0504225.
[14] M. Cvetic, H. L, D.N. Page, C.N. Pope, New EinsteinSasaki and Einstein spaces from Kerrde Sitter, hepth/0505223.
[15] M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, Rotating black holes in gauged supergravities: Thermodynamics,
supersymmetric limits, topological solitons and time machines, hep-th/0504080.
[16] M. Cvetic, P. Gao, J. Simon, Supersymmetric Kerranti-de Sitter solutions, Phys. Rev. D 72 (2005) 021701, hepth/0504136.
[17] H. L, C.N. Pope, J.F. Vzquez-Poritz, A new construction of EinsteinSasaki metrics in D 7, hep-th/0512306.
[18] G.W. Gibbons, M.J. Perry, C.N. Pope, AdS/CFT Casimir energy for rotating black holes, Phys. Rev. Lett. 95 (2005)
231601, hep-th/0507034.
Production of the top-pions from the

higgslesstop-Higgs model at the LHC
Chong-Xing Yue , Yi-Qun Di
Department of Physics, Liaoning Normal University, Dalian 116029, PR China
Received 26 March 2006; received in revised form 3 June 2006; accepted 6 September 2006
Available online 25 September 2006
Abstract
The top-pions (t0, ) predicted by extra-dimensional descriptions of the topcolor scenario have similar feature with those in four-dimensional topcolor scenario, which have large Yukawa couplings to the
third generation quarks. In the context of the higgslesstop-Higgs (HTH) model, we discuss the production of these new particles at the CERN Large Hadron Collider (LHC) via various suitable mechanisms
(gluongluon fusion, bottombottom fusion, gluonbottom fusion, and the usual DrellYan processes) and
estimate their production rates. We find that, as long as the top-pions are not too heavy, they can be abundantly produced at the LHC. The possible signatures of these new particles might be detected at the LHC
experiments.
2006 Published by Elsevier B.V.
1. Introduction
To completely avoid the problems arising from the elementary Higgs field in the standard
model (SM), various models for dynamical electroweak symmetry breaking (EWSB) have been
proposed, among which the topcolor scenario is attractive because it can explain the large top
mass and provide a possible EWSB mechanism [1]. The common feature of this kind of models
is that topcolor interactions make small contributions to EWSB and give rise to most part of the
top mass, while EWSB is mainly induced by technicolor or other strong interactions.
Recently, the model building along with the TeV-scale extra-dimensional scenario [2] has been
widely studied. Considering the gauge symmetry breaking can be easily achieved by imposing
E-mail address: cxyue@lnnu.edu.cn (C.-X. Yue).

0550-3213/$ see front matter 2006 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2006.09.005
56
C.-X. Yue, Y.-Q. Di / Nuclear Physics B 762 (2007) 5566
appropriate boundary conditions [3], the higgsless theory [4] was proposed and the topcolor
scenario was reconsidered in extra dimensions [5,6]. For extra-dimensional descriptions of the
topcolor scenario [5], there are two separate 5-dimensional anti de Sitter spaces(AdS5 s) with
each sector having its own corresponding infrared (IR) brane. The light fermions propagate in an
AdS5 sector, while the third generation quarks would propagate in other AdS5 sector, and sources
for EWSB are localized on both of these IR branes. Using the appropriate boundary and matching
conditions, there are three possibilities for EWSB: Higgstop-Higgs, higgslesstop-Higgs, and
higgslesshiggsless. For the higgslesstop-Higgs case, most of EWSB comes from the higgsless
sector, and the top quark gets its mass from a top-Higgs on the other IR brane.
Similarly to the topcolor scenario in four-dimensional space, the higgslesstop-Higgs (HTH)
model [5] predicts the existence of isotriplet scalars, called top-pions (t0, ), which have large
Yukawa couplings to the third generation quarks and sensibly small couplings to the massive
gauge bosons and light quarks. The top-pions get masses at one-loop from gauge interactions,
which can propagate from one boundary to the other. The mass scale is set by the cutoff scale on
the IR brane, so the top-pions should be quite heavy and cannot be abundantly produced at the
present high energy experiments. The aim of this paper is to investigate production of these new
particles (t0, ) at the CERN Large Hadron Collider (LHC) with s = 14 TeV and see whether
the possible signatures of the HTH model can be detected in the near future LHC experiments.
In the rest of this paper, we will give our results in detail. The couplings of the top-pions
t0, to the ordinary particles are given in Section 2, and the decay widths of the possible decay
modes are also estimated in this section. Sections 3 and 4 are devoted to the computation of
the production cross sections of t0 and t at the LHC, respectively. Some phenomenological
analysis are also included in these sections. Our conclusions are given in Section 5.
0,
2. The possible decay modes of the top-pions t
For the HTH model [5], there are two AdS5 spaces intersecting along a codimension one
surface (Planck brane) that would serve as a ultraviolet (UV) cutoff of the two AdS5 spaces. The
t
two AdS5 spaces, denoted as AdSw
5 and AdS5 , are characterized by their own curvature scale Rw
and Rt , respectively. The common UV boundary for each brane is located at the point Z = Rw
or Rt in the coordinate system and each AdS5 space is also cut by an IR boundary located at
or R . The IR brane at R has a higgsless boundary condition and is responsible for
Z = Rw
t
w
most of EWSB, while the top-Higgs h0t on the IR brane at Rt makes very small contributions to
EWSB. Thus, for the HTH model, there is the following relation:
2 = w2 + t2 ,
(1)
where = 246 GeV is the electroweak scale, w represents the contributions of the higgsless
sector to EWSB. The vacuum expectation value (VEV) t of the top-Higgs produces all of the
top mass. To produce the large top mass mt , the VEV t should be very small. Thus, the top-pions
have large Yukawa couplings to the third generation quarks.
The coupling constants of the top-pions t0, to the third generation quarks can be written
as [5]:
mt
mb ,
t0 t t:
(2)
,
t0 bb:
t
t
e
[mt PR + mb PL ],
t t b:
(3)
2SW mW t
57
where SW = sin W , W is the Weinberg angle, and PL(R) = (1 5 )/2 is the left(right)-handed
projection operator. Certainly, there might be the flavor changing (FC) couplings t0 tc and t bc.
However, for the HTH model, the flavor physics is generated on the Planck brane via mixing in
non-diagonal localized kinetic terms, one can choose the free parameters to avoid the presence of
Furthermore, considering the electroweak precision measurethe FC couplings t0 tc and t bc.
ment constraints, the top-pions should be quite heavy. Through out this paper, we will take the
top-pion mass in the range of 400800 GeV. In this case, the main decay channels are t0 t t
and t tb. Thus, we will not consider the FC couplings of t0, in this paper. Compared
to the SM Higgs boson, the couplings of t0 to the third generation quarks are enhanced by the
factor /t , while the tree-level couplings of t0 to the electroweak gauge bosons W and Z are
suppressed by the factor t /:
t emZ
t emW
t0 ZZ:
(4)
g ,
t0 W + W :
g .
S W CW
SW
For the neutral top-pion t0 decays, we will focus our attention on the decay modes: t t, bb,
+
W W , ZZ, gg and . The expressions of the decay widths for these decay channels can be
written as:

0

3m2t
4m2t 3/2
t t t =
(5)
mt 1 2
,
mt
8t2

3m2b
4m2b 3/2
m
,
t0 bb =
(6)
1
t
m2t
8t2
2

e m3t

m4W
4m2W 1/2
m2W
t
+
12
t0 W + W =
(7)
1
4
,
2 m2
m2t
m2t
m4t
16SW
W
2

e m3t
m4Z
4m2Z 1/2
m2Z
t
+
12
1
4
,
t0 ZZ =
(8)
2 m2
m2t
m2t
m4t
32SW
W

2
2s2 m3t m4t
2 2
m2t
t0 gg =
(9)
m
,
1
4
C
0
t
m2t
m2t
9t2 3

2 2m2t
e2 m3t 4

m2t
2
m
t0 =
m
1
4
C
0
t
16 3 3t2 t
m2t
m2t

2 2
m2
m2W
2 1
2
m
3m
.
+3 W
1
2
C
+ t4
(10)
0
W
W
2
m2t
m2t
The expression of the three-point scalar integral C0 (m2i ) can be written as C0 (m2i ) =
C0 (m2t , 0, 0, m2i , m2i , m2i ) [7]. For the decay channels t0 gg and , we have neglected
the contributions of the light quarks.
To obtain numerical results, we need to specify the relevant SM input parameters. We take
2 = 0.2315, = 1/128.8, = 0.112, m =
the SM parameters as mt = 172.7 GeV [8], SW
e
s
Z
91.187 GeV, and mW = 80.425 GeV [9]. Except for these SM input parameters, the partial
decay widths are dependent on the two free parameters t and mt . To make that the HTH model
generates enough large top mass and satisfies the electroweak precision constraints, the topHiggs VEV t should be very small and the top-pion should be quite heavy [5]. As numerical
estimation, we will take t 50 GeV (t / 1/5) and assume that the top-pion mass mt is in
the range of 400800 GeV.
58
Fig. 1. Branching ratios for the prominent decay modes of the neutral top-pion t0 as functions of the mass parameter
mt for t / 1/5.
The branching ratios of the prominent decay channels for the neutral top-pion t0 are pictured
in Fig. 1 as functions of the top-pion mass mt for t / 1/5, in which we have multiplied
0
+
0
the factors 100 and 10 to the branching ratios Br(t0 bb)[Br(

t W W ), Br(t ZZ)]
0
0
and Br(t gg), respectively. Since the value of the branching ratio Br(t ) is largely
small than that of Br(t0 t t), so we have not shown Br(t0 ) in Fig. 1. From Fig. 1 we
can see that, although the decay widths for all of the possible decay channels increase as mt
increasing, the main decay modes of t0 are t t and gg. For 400 GeV mt 800 GeV, the
value of Br(t0 gg) is in the range of 1.33%0.34%. Certainly, the values of the branching
ratios for various possible modes vary as the parameter t / varying. For example, the value of
the branching ratio Br(t0 gg) increases from 1.33% to 1.35% as t / decreasing from 1/5 to
1/8. Thus the neutral top-pion t0 might be abundantly produced at the LHC via the subprocess
gg t0 . The values of the branching ratios for all possible decay modes are given in Table 1
for three values of the top-pion mass mt and the top-Higgs VEV t .
For the charged top-pions t , the mainly decay mode is tb. The decay width of the decay
channel t+ t b can be written as:

t+ t b =
2

3e
1/2 m2t , m2t , m2b

2
2
3
8SW mW mt t

m2t m2t m2b m2t + m2b 4m2b m2t ,
(11)
where (a, b, c) = (a b c)2 4bc. Certainly, the charged top-pions t can also decay to W Z
and W at loop level. However, comparing with the tree-level process t tb, the branching
ratios of the one-loop processes t W Z, W are very small. The rare decay channels for the
charged top-pions t have been studied in the context of topcolor-assisted technicolor (TC2)
models [10].
59
Table 1
The values of leading branching ratios for three values of the top-Higgs VEV t and the mass parameter mt
t (GeV)
mt (GeV)
Br(t t)
Br(bb)
Br(W + W )
Br(ZZ)
Br(gg)
50
400
600
800
96.69%
98.63%
98.47%
0.56%
0.12%
0.09%
0.97%
0.54%
0.74%
0.45%
0.26%
0.36%
1.33%
0.46%
0.34%
40
400
600
800
97.50%
90.09%
99.11%
0.57%
0.12%
0.09%
0.40%
0.22%
0.30%
0.19%
0.11%
0.15%
1.34%
0.46%
0.34%
30
400
600
800
97.89%
99.31%
99.42%
0.57%
0.12%
0.09%
0.13%
0.07%
0.10%
0.06%
0.03%
0.05%
1.35%
0.46%
0.34%
In the following sections, we will use above formulas to study the production of the top-pions
t0, at the LHC.
3. Production of the neutral top-pion t0 at the LHC
From above discussions, we can see that production of the neutral top-pion t0 at the LHC
mainly comes from the gluon fusion (gg t0 ) and the bottom quark fusion (bb t0 ). The
production cross sections at parton level, which are proportional to the decay widths (t0
can be approximately written as:
gg) and (t0 bb),

2 m4
2 2
m2
gg t0 = s t2 1 4 2t C0 m2t 2 ,
(12)
mt
mt
64t

3m2
bb t0 = 2 b .
4t mt
(13)
The production cross section for the process pp t0 + X at the LHC can be expressed as:

pp t0 + X =
1
dx1

dx2 fg/p (x1 , F )fg/p x2 , m2t gg t0
1
+
1
dx1

dx2 fb/p x1 , m2t fb/p (x2 , F ) bb t0 ,
(14)
where fi/p are the parton distribution functions (PDFs) for gluons and bottom quarks evaluated
at some factorization scale F . In our numerical calculation, we will use CTEQ6L PDFs [11]
for the quark and gluon PDFs. Following the suggestions given by Ref. [12], we assume that the
factorization scale F for the bottom quark PDF is of order mt /4 in this paper.
The production cross section (t0 ) of the process pp t0 + X at the LHC with s =
14 TeV is plotted in Fig. 2 as a function of the top-pion mass mt for t / 1/5, in which the
dashed line, dotted line and solid line represent the contributions of the gluon fusion process,
bottom quark fusion process, and both these processes, respectively. From Fig. 2 we can see that
the production cross section (t0 ) mainly comes from the parton process gg t0 . However,
the contributions of the process bb t0 to (t0 ) cannot be neglected, which is in the range
60
Fig. 2. The production cross section (t0 ) at the LHC with

t / 1/5.
s = 14 TeV as a function of the mass parameter mt for
of 93527 fb for 400 GeV mt 800 GeV. This is because the neutral top-pion t0 has a
enhanced coupling with bottom quark pair, g 0 bb = mtb . If we assume the yearly integrated lumi t
nosity Lint = 100 fb1 for the LHC with s = 14 TeV, then there will be 1.3 104 4.8 105
t0 events to be generated per year. Certainly, the numerical results increase as the VEV t decreasing. For example, if we assume t = 30 GeV and 400 GeV mt 800
GeV, there will
be 3.6 104 1.3 106 t0 events to be generated per year at the LHC with s = 14 TeV and
Lint = 100 fb1 .
The production cross section of the top quark pair (t t) has been calculated at next-to-next-toleading (NNL) order in the context of the SM and can be measured to several percent accuracy
at the LHC [13,14]. Thus, it is needed to calculate the corrections of new physics beyond the
SM to the t t production cross section. The neutral top-pion t0 has large coupling with the top
quarks and should generate significant corrections to the t t production cross section at the LHC
via the enhanced gluon and bottom quark fusion mechanisms. The relative correction parameter
R(t t) = (t t)/ SM (t t) with (t t) = HTH (t t) SM (t t), which comes from t0 exchange,
is plotted as a function of the top-pion mass mt for three values of the VEV t in Fig. 3. One
can see from Fig. 3 that the relative correction parameter R(t t) is sensitive to the mass parameter
mt and its value quickly decreases as mt increasing. However, as long as mt 500 GeV
and t / 1/5, the value of R(t t) is larger than 1.9%, which might be detected at the LHC
experiments.
the proTo further discuss the possible signatures of the neutral top-pion t0 , we calculate
duction cross section of the process pp t0 + X + X at the LHC with s = 14 TeV.

Our numerical results show that the production cross section ( ) is very small. Even for
mt = 400 GeV and
t / 1/8, there will be only about 25 events to be generated per year
at the LHC with s = 14 TeV and Lint = 100 fb1 . This is because, compared to other decay
modes, the branching ratio Br(t0 ) is very small. Thus, the possible signals of t0 pre-
61
Fig. 3. The relative correction parameter R(t t) versus the mass parameter mt for t = 30 GeV (solid line), t = 40 GeV
(dashed line), t = 50 GeV (dotted line).
dicted by the HTH model cannot be detected via the process pp t0 + X + X at the
LHC experiments.
4. Production of the charged top-pions t at the LHC
A charged scalar does not belong to the particle spectrum of the SM. The discovery of a
charged scalar would be a clear signal for the existence of new physics beyond the SM. At
the LHC, the main production mechanism of a light charged scalar is via the top quark pair
production processes gg(q q)
t t followed by the top (or anti-top) quark decay, while, for a
heavy charged scalar, the main production process at the leading order comes from the gluon
bottom quark fusion process [15]. The complete NLO QCD corrections have been calculated by
Ref. [16]. In the context of the topcolor-assisted technicolor (TC2) models [17], we have studied
the production of the charged top-pions t via the parton process gb tt and discussed the
possibility of detecting t at the LHC via the tb(tb) decay channels [18]. The charged toppions t predicted by the HTH model have similar features to those from the TC2 models. In
the framework of the HTH model, the production cross section (tt ) for the process pp
gb + X tt + X at the LHC are plotted as a function of mt for three values of the top-Higgs
VEV t in Fig. 4. For 400 GeV mt 800 GeV and t = 50 GeV, the value of the cross
section (tt ) is in the range of 175563 fb.
From the discussions given in Section 2, we can see that, for the charged top-pion t , the
main decay mode is tb. Thus, the associated production of the charged top-pion t with single top quark can easily transfer to the t tb final state. Our numerical results show that, for
mt = 600 GeV and t / 1/5, there will be 2.8 104 t tb events to be generated per year
at the LHC with Lint = 100 fb1 . According the analysis conclusions for the backgrounds of
the process pp t tb + X given by Refs. [15,19], we except that the charged top-pions t
62
Fig. 4. The production cross section (tt ) as a function of the top-pion mass mt for t = 30 GeV (solid line),
t = 40 GeV (dashed line), t = 50 GeV (dotted line).
Fig. 5. The leading order Feynman diagrams for q q t+ t .
+ X in the near future LHC

should be observed via the process pp tt + X t tb(tt b)
experiments.
At the LHC, the charged top-pions t can also be produced in pair production mode. The
main production processes are the usual DrellYan processes q q t+ t (q = u, d, c, s and b)
through the s-channel Z exchange and photon exchange, and the t -channel bottom quark scatting
process bb t+ t , as shown in Fig. 5. Certainly, t+ t production receives an additional
contribution from the gg fusion process gg t+ t at one loop. However, its contributions are
very smaller than those of the tree level processes. Thus, we will ignore the process gg t+ t
in the following estimation.
Using Eq. (3) and other relevant Feynman rules, we can write the invariant amplitude for the
parton process q(P1 ) + q(P
2 ) t+ (P3 ) + t (P4 ) as:
M = M s + Mt .
(15)
2 ) t+ (P3 ) + t (P4 ), the invariant amplitude comes from

For the process b(P1 ) + b(P
Figs. 5(a) and (b):
63
M = ev(P
2 )Q u(P1 )g (P4 P3 )
+ v(P
2 ) (PL gL + PR gR )u(P1 )
+
e2
2
2m2W SW
v
vt
2
e
g
(P4 P3 )
s m2Z SW CW
v(P
2 )(mt PR + mb PL )
q/ + mt
(mt PR + mb PL )u(P1 ).
t m2t
(16)
For u, c, d and s quarks, we only consider the contributions of the s-channel process to the
scattering amplitude, which can be written as:
M = ev(P
2 )Q u(P1 )g (P4 P3 )
+ v(P
2 ) (PL gL + PR gR )u(P1 )
with
2
Q = e,
3
1
Q = e,
3
e
g
(P4 P3 )
2
s mZ SW CW

e
2eSW
8 2
(q = u, c),
gL =
2 SW ,
gR =
4SW CW
3
3CW

e
eSW
4 2
(q = d, s, b).
gL =
2 + SW
,
gR =
4SW CW
3
3CW
(17)
(18)
(19)
Where s = (P1 + P2 )2 and t = (P1 P3 )2 are the usual Mandelstam variables. We have neglected the light quark masses in our calculation, except for the bottom quark mass in the
couplings t tb. Our numerical results are shown in Fig. 6, in which we plot
the production
cross section (t+ t ) for the process pp t+ t + X at the LHC with s = 14 TeV as
a function of the top-pion mass mt for t / 1/5. To comparison, we use the solid line and
dashed line to represent the contributions of the process q q t+ t (q = u, c, d and s) and the
process bb t+ t , respectively. From Fig. 6 one can see that the production cross section of
Fig. 6. The cross section (t+ t ) as a function of the top-pion mass mt for t / 1/5.
64
the charged top-pion pair t+ t mainly comes from the usual DrellYan process q q t+ t
(q = u, c, d and s) through the s-channel Z exchange and photon exchange. The total production cross section (t+ t ) is in the range of 12.54.4 fb for 400 GeV mt 800 GeV
and t / 1/5. The charged Higgs boson pair production at the LHC has been calculated
to next-to-leading order [20]. They have shown that the total cross section for the process
pp H + H + X is smaller than 10 fb for tan 30 and mH 400 GeV. Thus, we expect that the charged top-pions t predicted by the HTH model can be more easy detected at
the LHC via this process than that for the charged Higgs bosons H .
5. Conclusions and discussions
Since extra dimensions can play an active role in the physics of the TeV scales, theories with
extra dimensions have recently attracted enormous attention [21]. Based on SU(2) SU(2)
U (1) gauge theory in slices of two back-to-back AdS5 s, topcolor theories were reconsidered
in extra dimensions [5]. For extra-dimensional descriptions of topcolor theories, there are three
possibilities for EWSB: Higgstop-Higgs, higgslesstop-Higgs, and higgslesshiggsless. The
higgslesstop-Higgs (HTH) model which is similar to TC2 models, predicts the existence of the
top-pions t0, and the top-Higgs boson h0t . Since the top-Higgs boson h0t is mainly responsible
for generating the large top mass and gives small contributions to EWSB, these new particles
(t0, , h0t ) have large Yukawa couplings to the third generation quarks, which might produce
characteristic signatures at various high-energy collider experiments.
In this paper,
top-pions t0, predicted by the HTH model at
we discuss the production of the
1
the LHC with s = 14 TeV and Lint = 100 fb via various suitable mechanisms (gluongluon
fusion, bottombottom fusion, gluonbottom fusion, and the usual DrellYan processes). The
following conclusions are obtained.
(1) The neutral top-pion t0 can be abundantly produced via the gluongluon fusion and bottom
bottom fusion processes. The production cross section (t0 ) is sensitive to the top-pion
mass mt and the top-Higgs VEV t , which mainly comes from the parton process gg t0 .
For t 50 GeV and 400 GeV mt 800 GeV, the value of (t0 ) is in the range
1304800 fb.
(2) The main decay modes of the heavy neutral top-pion t0 are t t and gg, thus the neutral toppion t0 can generate significant corrections to the t t production at the LHC via the process
pp t0 + X t t + X. For t = 40 GeV and 400 GeV mt 600 GeV, the value of
the relative correction parameter R(t t) is in the range of 2.1%14.8%.
(3) For a heavy charged scalar, the dominant production process at the LHC is its associated
production with a top quark via gluon bottom fusion. The LHC has good potential for discovering a heavy charged scalar through this process. Thus,
we estimate the production cross
section of the process pp tt + X at the LHC with s = 14 TeV and Lint = 100 fb1 .
Our numerical results show that the production rate is significantly large and the possible
signals of t might be detected at the LHC through the process pp tt + X in their
tb) decay modes.
t b(
(4) As long as the charged scalar is not too heavy, it is possible to be produced in pair at the near
future LHC experiments. The main production processes are the usual DrellYan processes
and the t -channel bottom quark scattering process. We further estimate the rate of the charged
top-pion pair production at the LHC. Our numerical results show that the production cross
section (t+ t ) is in the range of 9.118.6 fb for t = 30 GeV and 400 GeV mt
65
800 GeV, which might be larger than that for the charged Higgs bosons H (mH 400 GeV
and tan 30) predicted the minimal supersymmetric standard model.
In conclusion, as long as the top-pions t0, are not too heavy (mt < 1 TeV), they can be
abundantly produced at the LHC via various suitable mechanisms. The possible signatures of
these new particles might be detected at the LHC experiments.
At last, we have to say that all of our numerical results are obtained at leading order. The
strong Yukawa couplings can give large corrections to these numerical results. For example, for
the process pp gb + X tt + X, the leading order cross section given in Section 4 can
be enhanced about 80% by the effects of next leading corrections. Thus, the numerical results
given in this paper can only be taken as order-of-magnitude guides. However, we expect that our
estimations can provide the right qualitative picture.
Acknowledgements
This work was supported in part by Program for New Century Excellent Talents in University (NCET-04-0290), the National Natural Science Foundation of China under the Grant
No. 10475037.
References
[1] C.T. Hill, E.H. Simmons, Phys. Rep. 381 (2003) 235;
C.T. Hill, E.H. Simmons, Phys. Rep. 390 (2004) 553, Erratum.
[2] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 429 (1998) 263;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257;
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47.
[3] Y. Kavamura, Prog. Theor. Phys. 103 (2000) 613;
Y. Kavamura, Prog. Theor. Phys. 105 (2001) 691;
A. Hebecker, J. March-Russell, Nucl. Phys. B 613 (2001) 3;
A. Hebecker, J. March-Russell, Nucl. Phys. B 625 (2002) 128;
L.J. Hall, Y. Nomura, T. Okui, D.R. Smith, Phys. Rev. D 65 (2002) 035008;
L.J. Hall, Y. Nomura, Phys. Rev. D 65 (2002) 125012.
[4] C. Csaki, C. Grojean, L. Pilo, J. Terning, Phys. Rev. Lett. 92 (2004) 101802;
Y. Nomura, JHEP 0311 (2003) 050;
R. Barbieri, A. Pomarol, R. Rattazzi, Phys. Lett. B 591 (2004) 141;
C. Csaki, C. Grojean, J. Hubisz, Y. Shirman, J. Terning, Phys. Rev. D 70 (2004) 015012.
[5] G. Cacciapaglia, et al., Phys. Rev. D 72 (2005) 095018.
[6] N. Arkani-Hamed, H.C. Cheng, B.A. Dobrescu, L.J. Hall, Phys. Rev. D 62 (2000) 096006;
M. Hashimoto, D.K. Hong, Phys. Rev. D 71 (2005) 056004.
[7] G. Passarino, M. Veltman, Nucl. Phys. B 160 (1997) 151;
A. Axelrod, Nucl. Phys. B 209 (1982) 349;
M. Clements, et al., Phys. Rev. D 44 (1983) 570.
[8] CDF Collaboration, D0 Collaboration, Tevatron Electroweak Working Group, Combination of CDF and D0 result
on the top-quark mass, hep-ex/0507091.
[9] S. Eidelman, et al., Particle Data Group, Phys. Lett. B 592 (2004) 1.
[10] X.L. Wang, W.N. Xu, L.L. Du, Commun. Theor. Phys. 41 (2004) 737.
[11] J. Pumplin, et al., JHEP 0207 (2002) 012;
D. Stump, et al., JHEP 0310 (2003) 046.
[12] T. Plehn, Phys. Rev. D 67 (2003) 014018;
F. Maltoni, Z. Sullivan, S. Willenbrock, Phys. Rev. D 67 (2003) 093005;
R.V. Harlander, W.B. Kilgore, Phys. Rev. D 68 (2003) 013001;
66
A. Alves, T. Plehn, Phys. Rev. D 71 (2005) 115014.

[13] ATLAS Collaboration, Technical design report, CERN-LHCC-99-15;
CMS Collaboration, Technical proposal, CERN-LHCC-94-38;
G. Weiglein, et al., LHC/LC Study Group, hep-ph/0410364.
[14] M. Beneke, et al., Top quark physics, hep-ph/0003033;
D. Chakrahorty, J. Konigsherg, D. Rainwater, Annu. Rev. Nucl. Part. Sci. 53 (2003) 301;
W. Wagner, Rep. Prog. Phys. 68 (2005) 2409.
[15] A.C. Bawa, C.S. Kim, A.D. Martin, Z. Phys. C 47 (1990) 75;
J.F. Gunion, Phys. Lett. B 322 (1994) 125;
V.D. Barger, R.J.N. Phillips, D.P. Roy, Phys. Lett. B 324 (1994) 236.
[16] S.H. Zhu, Phys. Rev. D 67 (2003) 075006;
T. Plehn, Phys. Rev. D 67 (2003) 014018;
E.L. Berger, T. Han, J. Jiang, T. Plehn, Phys. Rev. D 71 (2005) 115012.
[17] C.T. Hill, Phys. Lett. B 345 (1995) 483;
K.D. Lane, E. Eichten, Phys. Lett. B 352 (1995) 382;
K.D. Lane, Phys. Lett. B 433 (1998) 96;
G. Cvetic, Rev. Mod. Phys. 71 (1999) 513.
[18] C.-X. Yue, Z.-J. Zong, L.-L. Xu, J.-X. Chen, Phys. Rev. D 73 (2006) 015006.
[19] N. Kidonakis, R. Vogt, Int. J. Mod. Phys. A 20 (2005) 3171;
D.P. Roy, hep-ph/0510070.
[20] J.F. Gunion, et al., Nucl. Phys. B 294 (1987) 621;
A. Krause, T. Plehn, M. Spira, P.M. Zerwas, Nucl. Phys. B 519 (1998) 85;
A.A. Barrientos Bendezu, B.A. Kniehl, Nucl. Phys. B 568 (2000) 305;
S. Moretti, J. Rathsman, Eur. Phys. J. C 33 (2004) 41;
H.S. Hou, et al., Phys. Rev. D 71 (2005) 075014;
A. Alves, T. Plehn, Phys. Rev. D 71 (2005) 115014.
[21] C. Csaki, hep-ph/0404096;
C. Csaki, J. Hubisz, P. Meade, hep-ph/0510275.
Heavy quark pair production near threshold

with potential non-relativistic QCD
Antonio Pineda a,,1 , Adrian Signer b
a Departamento dEstructura i Constituents de la Matria, U. Barcelona, Diagonal 647,
E-08028 Barcelona, Catalonia, Spain

b Institute for Particle Physics Phenomenology, Durham, DH1 3LE, England, UK
Received 24 July 2006; accepted 11 September 2006

Available online 24 October 2006
Abstract
We study the effect of the resummation of logarithms for t t production near threshold and inclusive
electromagnetic decays of heavy quarkonium. This analysis is complete at next-to-next-to-leading order
and includes the full resummation of logarithms at next-to-leading-logarithmic accuracy and some partial
contributions at next-to-next-to-leading logarithmic accuracy. Compared with fixed-order computations at
next-to-next-to-leading order the scale dependence and convergence of the perturbative series is greatly
improved for both the position of the peak and the normalization of the total cross section. Nevertheless,
we identify a possible source of large scale dependence in the result. At present we estimate the remaining
theoretical uncertainty of the normalization of the total cross section to be of the order of 10% and for the
position of the peak of the order of 100 MeV.
1. Introduction
The existence of (heavy) quarks with a large mass compared with QCD , like the top, the bottom, and maybe the charm, makes particularly interesting the study of physical processes where
a heavy quark pair is created close to threshold, because accurate experimental information over
the whole near-threshold region may allow for a precise determination of some parameters of the
E-mail address: antonio.pineda@cern.ch (A. Pineda).

1 Permanent address after September 1st: Grup de Fsica Terica and IFAE, Universitat Autnoma de Barcelona,
E-08193 Bellaterra, Barcelona, Spain.

doi:10.1016/j.nuclphysb.2006.09.025
68
A. Pineda, A. Signer / Nuclear Physics B 762 (2007) 6794
Standard Model. For instance, the future International Linear Collider (ILC) offers the opportunity to study the top quark with unprecedented accuracy [13]. To fully exploit the potential of
an ILC in this respect, it is essential that a dedicated measurement of the cross section for the
production of a topantitop quark pair close to threshold is made. Such a threshold scan allows
for an extremely precise measurement of the top-quark mass and yields information on the topquark width and the top-Higgs Yukawa coupling. The analogous threshold scan for the bb sector
is also fundamental for non-relativistic sum rules [48] and may lead to accurate determinations
of the bottom-quark mass [9].
The characteristic feature of heavy-quark pair production close to threshold is the smallness
of the relative velocity v of the heavy quarks in the centre of mass frame. This entails a hierarchy
of scales h s us where the hard scale h is of the order of the heavy quark mass m, the
soft scale s mv is of the order of the typical momentum of the heavy quarks and the ultrasoft
scale us mv 2 is of the order of the typical kinetic energy of the heavy quarks. The presence
of an additional small parameter can be exploited by systematically expanding in the strong
coupling s and v. Thus, in this context, a next-to-leading order (NLO) calculation takes into
account all terms that are suppressed by either s or v relative to the leading-order (LO) result,
whereas a next-to-next-to-leading order (NNLO) calculation includes all terms suppressed by
two powers of the small parameter s v. The coefficients of this perturbative series contain
large logarithms log v. In order to improve the reliability of the calculation, these logarithms
should be resummed. Counting log v 1 in a NLO or NNLO calculation produces a result
of next-to-leading logarithmic (NLL) or next-to-next-to-leading logarithmic (NNLL) accuracy
respectively. There are no leading logarithmic (LL) corrections, thus the LO and LL results are
the same.
The expansion as well as the resummation of the logarithms can be organized most efficiently
by using an effective theory (for a review see Ref. [10]) approach. This is done most conveniently
by using the threshold expansion [11] that allows to separate the full result of an integral into contributions due to the various modes. Denoting the generic integration momentum by k = (k 0 , k),
the modes that are relevant are: the hard mode k 0 k h , the soft mode k 0 k s , the
potential mode k 0 us , k s and the ultrasoft mode k 0 k us . The standard procedure
is to first match QCD to non-relativistic QCD (NRQCD) [12] at the hard scale. This corresponds
to integrating out the hard modes using the threshold expansion. The resulting theory is then
matched to potential NRQCD (pNRQCD) [13,14] by integrating out the soft modes and potential gluon modes. At this stage the theory consists of a non-relativistic quark pair interacting
through potentials and ultrasoft gluons.
Within this framework NNLO calculations have been performed by several groups (for a
review in the case of the top quark see [3]) and quite a few partial results needed for a NNNLO
calculation have been obtained [1521]. Moreover, the existence of the effective theory also
provides the necessary framework on which to use renormalization group (RG) techniques. These
allow to resum the large logarithms, log v, that appear as the ratio of the different scales appearing
in the physical system: log h /s and log s /us . At present, the situation is as follows. The
computation of the heavy quarkonium spectrum is known with NNLL accuracy [22,23] (for the
hyperfine splitting at NNNLL [24,25]). Inclusive electromagnetic decays of heavy quarkonium
have been computed to NLL [23,26] and for the spin-zero, spin-one ratio to NNLL [27]. In
the case of top-quark pair production a renormalization-group improved (RGI) calculation is
available [28,29], using a somewhat different approach, referred to as vNRQCD [30] (in this
theory, soft degrees of freedom are kept dynamical and the matching from QCD to vNRQCD is
carried out directly). However, so far, there is no RGI calculation of heavy-quark pair production
69
within the conventional pNRQCD approach. It is the main purpose of this work to close this gap
and to provide the ingredients to perform such computations in the case of top and bottom-quark
pair production. We will report results on t t production near threshold, as well as elaborate on the
computation of non-relativistic sum rules presented in Ref. [9]. Inclusive electromagnetic decays
of heavy quarkonium will also be considered. These analyses will be complete at NNLO and
include the complete resummation of logarithms at NLL accuracy and some partial contributions
at NNLL.
The importance of higher-order logarithms can be illustrated in the top-quark case. In the onshell scheme, the NNLO corrections turned out to be much larger than anticipated and, moreover,
made the theoretical prediction very strongly scale dependent. If the cross section is expressed
in terms of a threshold mass [3133] rather than the pole mass, the position of its peak is more
stable and can be predicted with a small theoretical error. This will allow to determine the top
threshold mass and ultimately the top MS-mass with a very small error. The situation is much
less favourable regarding the normalization of the cross section. The corrections are huge and the
scale dependence at NNLO is larger than at NLO, indicating that this quantity is not well under
control at NNLO. This affects how accurately one may obtain the top-quark width and the topHiggs Yukawa coupling. It has been shown that the inclusion of the potentially large log v terms
is numerically very important and improves the situation regarding the normalization of the cross
section considerably [28,29]. One of the aims of this paper is to compare with these results. We
also investigate to what extent the resummation of the logarithms is really required and to what
extent the improvement we observe in the RGI results is simply due to the partial inclusion of
higher-order terms. To do so we will produce NNNLO results by re-expanding full RGI results
and dropping terms that are beyond NNNLO. Of course, these results are by no means complete
at NNNLO, but they contain those NNNLO terms that are enhanced by a logarithm. To obtain an
estimate of the importance of resummation, we then compare these NNNLO curves to the full
RGI results. As we will see, this partial inclusion of NNNLO terms does reproduce the bulk of
the RGI result except for rather small values of s . In particular, in the case of the top quark the
difference between fully resummed and partial NNNLO results is small.
The paper is structured as follows: In Section 2 we present our formalism and describe how
to perform a RGI calculations in pNRQCD. Some formulae and technical details of this section
are relegated to the appendix. In Section 3 we apply these results to t t production near threshold. Section 4 contains the application to bottomonium non-relativistic sum rules and inclusive
electromagnetic decay widths. In the final section, we show our conclusions.
2. Effective theory
Within pNRQCD, and , the fields representing the non-relativistic quark and antiquark,
interact through a potential and with dynamical ultrasoft gluons. It is well known that the leading
Coulomb interaction is not suppressed and has to be included in the LO Lagrangian which is
given by

2
2
0
L(0)
=
i
i
pNRQCD
2m
2m

s a
T ,
+ d 3r T a
(1)
r
where T a are the colour matrices and the strong coupling is understood to be evaluated at the soft
scale, s s (s ), unless explicitly indicated otherwise. Subleading effects are incorporated in
70
the Lagrangian as corrections to the potential, V , and as interactions of the heavy quarks with
ultrasoft gluons. For further details we refer to Ref. [10]. If we restrict the accuracy of our analysis
to NNLL, ultrasoft gluons do not appear as physical final states (thought their effect is embedded
in the RGI running of the matching coefficients of the potentials and currents). It follows that
the number of particles is conserved (we only have one heavy quark and one heavy antiquark)
and the problem effectively becomes equivalent to do standard quantum mechanics perturbation
theory. If we restrict ourselves to study the Hilbert space spanned by the heavy quarkantiquark
system in the singlet colour sector we are lead to solve the following equation for the associated
Green function

H (r, p) E G(r, r ; E) = (r r ),
(2)
where
H = Hc + V
(3)
and
Hc =
s
p2
CF ,
m
r
(4)
with CF = (Nc2 1)/(2Nc ) (where the colour factor Nc = 3). The explicit expression of V (the
correction to the Coulomb potential) will be given afterwards. The full Green function G can be
solved iteratively in an expansion in the velocity by performing multiple insertions of V . The
LO solution is the Coulomb Green function Gc and we write
G(r, r ; E) = Gc (r, r ; E) + G(r, r ; E),
(5)
where
Gc (r, r ; E) = r|
1
|r
Hc E
(6)
and
G(r, r ; E) = r|
1
1
V
|r + .
Hc E
Hc E
(7)
G(r, r ; E) is related to the correlators that appear in the total cross section
forthe production
with the centre of mass energy q 2 = s 2m. This
of a heavy quark pair, (e+ e QQ)
is the key quantity we are interested in. The cross section obtains contributions from and Z
exchange. In order to simplify the discussion we ignore the Z exchange in what follows. The
cross section can then be written as
2
4EM
2
(8)
R(s),
eQ
3s
where eQ is the electric charge of the heavy quark, EM the electromagnetic coupling at the hard
scale and the ratio R (e+ e QQ)/

(e+ e + ) is expressed as a correlator of two
heavy-quark vector currents j (x) Q Q(x)

4
Im i d 4 x eiqx 0|T j (x)j (0) |0 .
R(s) =
(9)
s
(s) =
71
In order to compute this correlator, we first express the current in terms of the non-relativistic
two-component spinor fields and ,
Q = c1 i d1 i (iD)2 + ,
Q
(10)
6m2
where the matching coefficients c1 and d1 are normalized to 1 at LO. By using the equations of
motion, Eq. (10) can also be written in the following way

Q = c1 i d1 i0 i + .
Q
(11)
6m
Given the Lagrangian, we solve the corresponding Schrdinger equation and ultimately relate
the imaginary part of the spin one Green function at the origin to R(s) by the equality

2N
24eQ
c
E
c12 c1 d1
Im Gs=1 (0, 0; E),
R(E) =
(12)
s
3m
which is valid with NNLL accuracy (where E s 2m). We note that in evaluating R(E) we
expand the expression in Eq. (12) and drop all terms that are beyond NNLO/NNLL. In particular,
we set c1 = 1 in the second term of the parenthesis.
In order to connect with the notation in Ref. [10], Eq. (12) can also be written as

18Nc
pNR 3
pNR 3 E
Im Gs=1 (0, 0; E) Im fEM S1 + Im gEM S1
,
R(E) =
(13)
2
m
m2 EM
with
2 2
eQ
EM 2
S1 =
c1 ,
3

2 2
eQ

1
EM
pNR
c1 c1 + d 1 .
Im gEM 3 S1 =
3
3
pNR 3
Im fEM
(14)
2.1. Potential
The computation of G(0, 0; E) will lead to divergences. These divergences are regularized
by performing all calculations in momentum space [34] and using dimensional regularization in
D = d + 1 = 4 2 dimensions. Thus, the LO Green function at the origin, Eq. (6), is understood
as

d d p d d p

Gc (p, p ; E),
Gc (r, r ; E)|r=r =0
(15)
(2)d (2)d
denotes the Fourier transform of the Green function and the insertions, Eq. (7), are to
where G
be evaluated as indicated in Eqs. (42) and (43). Using the threshold expansion [11], the Fourier
transform of the leading order Green function can be computed as the sum of all ladder diagrams
with the exchange of potential gluons between the heavy quarks and can be written as
1
E p2 /m
4CF s
+ finite,
+
(E p2 /m)(p p )2 (E p 2 /m)
c (p, p ; E) = (2)d (d) (p p )

G
(16)
72
where we have omitted terms that are finite if Eq. (16) is used in Eq. (15). The Green function at
the origin has a ultraviolet divergence that manifests itself as a pole 1/. Using MS subtraction
and then taking the limit 0 we find [34,35]

s CF m2 1
1
4mE 1
Gc (0, 0; E) =
(17)
+
(1
)
,
+ log
+
E
4
2 2
2
2s
where CF s /(2 E/m ).

Higher-order corrections to the Green function can be computed perturbatively in two steps.
First, the pNRQCD Lagrangian has to be determined to the accuracy needed for the calculation.
Second, higher-order corrections are computed using quantum mechanics perturbation theory,
Eq. (7). Since some of the potentials generate singularities in insertions, we have to start from
the NNLO potential computed in D dimensions [34]. It is essential to manipulate the potential
consistently in D dimensions, since it allows us to use the same hard matching coefficients defined in the MS-scheme obtained in Refs. [36,37]. We then bring the potential in a form that is
more suitable to be combined with the RGI coefficients as presented in Ref. [22]. In particular
we use
2

4
p2 p23 2
CF s d d pi
c (p3 , p4 )
Gc (p1 , p2 )
G
(2)d
m2
(p2 p3 )2
i=1

2
eE 2 12 12 +
CF
=
(1 2)
2
3/2 (1 2)

4
d d pi
2 s2 2
s
c (p3 , p4 )
G
G
(18)
(p
,
p
)
c
1
2
(2)d
m|p2 p3 |1+2
i=1
to eliminate the CF2 term of the non-analytic potential 1/q 1+2 , present in Ref. [34]. The angular
momentum operator is generalized to
2

L2
p p 2 2
(19)
1
2r 3
q2
to be compatible with D-dimensional calculations in momentum space. Note that an insertion
of this operator does not vanish even for an S-wave (see Eq. (A.28)), but the corresponding
contribution could be absorbed into a redefinition of the matching coefficient of the current.
At NNLL, the higher-order corrections to the D-dimensional potential in momentum space,
V , can then be written as
V
p4
s
(2)d (d) (q) 4CF 2s + 4CF 2
3
4m
q
q
1
1

2
2
2
E
e 2 2 +
(1) s
CF CA Ds
(1 )
mq 1+2
3/2 (1 2)

(2)
(2) 2
2CF D1,s p2 + p 2 CF D2,s
p p 2 2
+
1
m2
q2
m2
q2
V = c4
(2)
3CF Dd,s
m2
4CF DS(2)
2 ,s
dm2
j
j
Si1 , S1 Si2 , S2

(2)
73

qr qj
j
Si1 , Sr1 Si2 , S2 rj d 2
q
4CF DS12 ,s
dm2
(2)
6CF DLS,s p i q j i j i j
S 1 , S1 + S2 , S2 ,
m2
q2
(20)
where the colour factor CA = Nc , q = p p and Vs contains the corrections to the static
potential up to NNLL [38,39]. We will set c4 = 1 due to reparameterization invariance. We
would like to stress that the Wilson coefficients are not dimensionless in D = 4.
The non-relativistic reduction of the spin operators of the potential depends on the operators
used to single out the physical state we want to study. In our case we are using the vector currents,
which project to the spin-one state. We also have to be careful to use the same conventions
than those used to obtain the hard piece of the matching coefficients. This produces O() terms
multiplying the S2 operator. In practise one can do the following replacement for the S2 operator
with S2 = 2
4CF DS(2)
2 ,s
dm2
j
j
Si1 , S1 Si2 , S2
(2)

S2 2CF DS 2 ,s
(d 4)(d 1) .
2
4
dm
(21)
For S-wave creation there is no contribution from the potentials proportional to the S12 and L S
operators.
The D-dimensional prescriptions used above are irrelevant for the computation of the heavy
quarkonium mass with NNLL accuracy, which reflects the fact that the leading running of the
Wilson coefficients is scheme independent. On the other hand to specify the D-dimensional prescription of the potential is important once they are introduced into divergent potential loops.
This is relevant if we want to obtain heavy quarkonium sum rules, or to compute the t t production near threshold with NNLL accuracy, and use computations obtained in other places for
the hard matching coefficients. We would like to emphasize however that we could have chosen a different prescription. This would have changed some intermediate-step results but not the
physical results.
Finally, one should note that the potential above is not equal to the potential used in Ref. [34].
Nevertheless, they can be exactly related with each other by field redefinitions (in four and D
dimensions). In particular this means that the hard matching coefficients will be the same in
both cases. This is actually what we expected, since they simply correspond to the effects of the
(integrated out) hard modes.
The RGI coefficients of the various potentials are known to the accuracy required for evaluating the Green function at NNLL [22,39]. Note that the strong coupling is included in the
(1)
(2)
matching coefficients. Thus Ds s2 (1 + (s log v)n ) and DX s (1 + (s log v)n ). For the
Coulomb potential Vs , the exact static potential at two loop [38] receives additional three-loop
LL terms proportional to s3 CA3 (s log v)n [39]. The explicit form of all matching coefficients can
be found in Ref. [22]. However, we have changed the basis of potentials compared to Ref. [22].
(2)
This affects the matching coefficient Dd,s , which now reads (us = 2s /h )

s (s )
1
1
(2)
Dd,s (us ) =
dss (s )
2 + cD (s ) +
dvs (s ) +
3
3
CF

32 CA
s (s )
+
(22)
CF s (s ) log
.
90 2
s (us )
74
For the other potentials, the expressions obtained for their Wilson coefficients in Refs. [22,39]
hold, where one may also find the expressions for the RGI coefficients (cD , dvs , dss , . . .) of the
NRQCD operators. We repeat them in the appendix for ease of reference.
In this paper we also include an additional contribution in V , the electromagnetic Coulomb
term
V V
2
4eQ
EM
q2
(23)
This will give rise to a NLO term EM /v from single potential photon exchange and a NNLO
term from double potential photon exchange. Strictly speaking, the coupling in Eq. (23) should
be evaluated at the soft scale and not at the hard scale, but this effect is beyond NNLL.
This completes all the corrections needed at the Lagrangian level for the computation of the
imaginary part of the Green function with NNLL accuracy. What is left is to obtain the RGI
expressions for cs and ds , where s = 0, 1 labels the spin. The matching coefficient cs is needed
at NNLL whereas ds is only needed at LL.
2.2. Direct ultrasoft effects to cs and ds
Most of the ultrasoft contribution to the running of cs and ds comes in a indirect way, through
the running of the potentials in the anomalous dimensions of the RG equation. Nevertheless,
there are some genuine ultrasoft effects that have not been considered so far. They are due to the
appearance of some energy dependent potentials with the structure
Vus = (Hc E)Z 1/2 ,
(24)
where Z is the normalization correction to the heavy quarkonium propagator. Therefore, Z 1/2
corresponds to the normalization of the field that represents the heavy quarkonium. Vus was
not included in Eq. (20), as there only energy independent potentials were considered. Its effects could be reabsorbed in field redefinitions of the fields that represent the heavy quarkonium.
Therefore, they have no consequences in the spectrum. Nevertheless, these field redefinitions
change the vertex interaction of the heavy quarkonium with photons producing changes in cs
and ds and they have to be considered in our computation. The leading logarithmic corrections
to Z 1/2 were obtained in Eqs. (15)(18) of Ref. [17] (for some partial results see also Ref. [16]).
They produce corrections of order s3 log s . We can obtain the RGI expressions for them and
(n+3)
generate terms of order s
log(1+n) s . Note that the running goes up to the soft scale because one gets expressions of the form r = 0| log r|p and the logarithm gets the scale of p. The
corrections to the Green function due to Eq. (24) read

CA2 CF
4 2
s (s )
1
G1 = s log
(25)
,
0
s (us ) Hc E 4

4s
s (s )
1
1
2 2
log
,
C F + CF C A ,
G2 =
(26)
0
s (us ) mr Hc E
3
2

4
s (s )
1
p
2
log
,
CF .
G3 =
(27)
0
s (us ) m2 Hc E 3
The correction coming from G3 can be included in ds :

16CF
s (us )
ds d s +
log
.
0
s (s )
(28)
75
The correction from G2 is zero. Finally, the correction from G1 could be absorbed in cs :

s2 CA2 CF
s (s )
log
.
cs cs +
(29)
2 0
s (us )
Note also that these changes are equivalent to a change in the NNLO and LO anomalous dimension of the RG equation describing the running of cs and ds respectively.
2.3. Running of ds
pNR
The LL running of Im gEM can be obtained in two steps. In the first step one computes its
soft running. This has been done in Refs. [40,41]. For the explicit result see Eqs. (C.18), (C.19)
in Ref. [41]. The ultrasoft running can be obtained from Eq. (28). Adding everything together
pNR
the LL running of Im gEM reads

16
s (us )
pNR
pNR
pNR
Im gEM 1 S0 (s ) = Im gEM 1 S0 (h )
CF Im fEM 1 S0 (h ) log
,
30
s (h )
(30)

(
)
16
s
us
pNR
pNR
pNR
CF Im fEM 3 S1 (h ) log
.
Im gEM 3 S1 (s ) = Im gEM 3 S1 (h )
30
s (h )
(31)
The last result agrees with the LL result obtained in Ref. [29]. For the matching coefficient d1 as
defined in Eq. (10), this entails

16CF
s (us )
log
.
d1 (us ) = 1 +
(32)
0
s (h )
2.4. Running of cs
The running of cs is not yet known with NNLL accuracy. It is dictated by the solution of the
RG equation (the LO anomalous dimension is zero)
s
d
log cs = cNLO
+ cNNLO
+ .
s
s
ds
(33)
The structure of the solution reads

NLL (
cs (s ) = cs (h )es (h )cs
2
NNLL ( )+
s )+s (h )cs
s
(34)
Expressions for cs (h ) at two loops in the MS can be found in Ref. [36] for s = 1 and (almost
complete) in Ref. [37] for s = 0.
in the basis of potentials used in this paper reads
The expression for cNLO
s

CF2
4
CA CF (1)
(2)
(2)
(2)
=
+
2
3D
+
4D
s(s
+
1)
D
Ds .
cNLO
(35)
s
s
2 ,s
d,s
1,s
s
S
4
3
2
The expression of cNLL
is known [23,26]. For cNNLL
only the spin-dependent term is coms
s
pletely known [27]. Its contribution to the spin-one case reads
cNNLL
(s ) = cNNLL
+
1 ,SD
v
11 2 s (s ) (2)
11
D (s ) CF2 ,
C
72 F s2 (h ) S 2 ,s
72
(36)
76
where cNNLL
corresponds to the result quoted in Ref. [27]. Note that we use a different exv
pression for the spin-dependent term than just cNNLL
. The expression above corresponds to the
v
spin-dependent contribution to the vector current matching coefficient in the MS scheme. Numerically this contribution is small compared with others.
Besides the spin-dependent correction, we have also incorporated the following spinindependent corrections at NNLL order (to these corrections one obviously has to subtract the
spin-dependent piece that has already been included in cNNLL
):
1 ,SD
(a) Those that appear from the exponentiation of the NLL term and formally are NNLL (see
Eq. (34)).
(b) Effects due to the two-loop beta running of s . They produce the following correction:
(s ) =
s2 (h )cNNLL
1 ,b
1
202

s (s )
ds NLO
s cs
s (h )

s (s )
s (h )
ds s (2s /h ) cNLO
s
.
1
s
1
s (2s /h )
(37)
The last term is generated from the fact that in the determination of cNLL
(s ), the relation
s
1
s (2s /h )
,
s (s )
(2 z0 )
(38)
where z0 = s (s )/s (h ), was used. This relation is only true at one loop and has to be corrected by 1 terms if a NNLL accuracy is demanded. The numerical impact of these corrections
is small.
(c) We have also incorporated the corrections proportional to a1 , the one-loop log-indepens
a1 ). These corrections can be deduced from the
dent term, that appears in Vs s (s )(1 + 4
computation of the NLO anomalous dimension. They read
s2 (h )cNNLL
(s ) =
1 ,c
1
a1
20

s (s )

2
ds
CA CF (1)
NLO
2 s
Ds CF
.
cs +
s
2
4
(39)
s (h )
The numerical impact of these corrections is small.

Finally, the inclusion of the electromagnetic corrections produces some corrections to c1 . Counting EM s2 , the one-loop exchange of a hard photon contributes at NNLO and is taken into
account by
c1 (h ) c1 (h )
2
2eQ
EM
(40)
for the spin-one case, whereas for s = 0 we have

c0 (h ) c0 (h )

2
eQ
EM 5
2
2
8
(41)
77
2.5. Green function

Once the RGI potential and current matching coefficients are available, we are in a position
to use standard quantum mechanics perturbation theory to compute the higher-order corrections
to Im[G(0, 0; E)] via insertions of the potentials. This calculation has been done in momentum
space using dimensional regularization. For the terms suppressed by two powers of s v in
V , Eq. (20), it is sufficient to consider a single insertion,
G(0, 0; E) =

4
d d pi
c (p3 , p4 ; E),
Gc (p1 , p2 ; E) V (p2 , p3 )G
(2)d
(42)
i=1
whereas for terms suppressed by only a single power of s v we have to compute double
insertions as well,
G(0, 0; E)

6
d d pi
c (p3 , p4 ; E) V (p4 , p5 )G
c (p5 , p6 ; E). (43)
Gc (p1 , p2 ; E) V (p2 , p3 )G
=
(2)d
i=1
Since the RGI does not alter the structure of the interaction terms, the insertions can be taken
directly from Ref. [34]. In Appendix A we list the integrals that are needed and present the
explicit results obtained in Ref. [34].
In the calculation described so far, the on-shell scheme for the heavy quark mass has been
implicitly assumed. In order to avoid the bad convergence behaviour inherent to this scheme, it is
necessary to rewrite the expressions in terms of a threshold mass which is free of the renormalon
ambiguity. In this paper we will consider the cases of the potential subtracted (PS) mass [32] and
renormalon subtracted (RS) mass [33]. We will use the difference between both schemes as an
indication of the scheme dependence of our results.
3. The case of the top quark
In this section we apply the previous results to the case of top-quark pair production near
threshold at a future linear collider. Due to the large width t of the top quark there are no
bound states and the toponium resonances are smeared, resulting in a smooth curve for the cross
section with a broad peak as the remnant of the would-be 1S bound state. Using perturbation
theory we can reliably compute the cross section as a function of the energy [1]. This may lead
to accurate determinations of the top mass and its total decay width by measuring the position
and normalization of the peak.
The t t pair will be dominantly produced via e+ e , Z t t. The total production cross
section may be written as [3]
,Z
tot (s) =
2

4EM
F v (s)R v (s) + F a (s)R a (s) ,
3s
(44)
where

v

4
4
iqx
v
R (s) =
Im i d x e 0|T j (x)j (0) |0 ,
s

a

4
a
4
iqx
a
Im i d x e 0|T j (x)j (0) |0 ,
R (s) =
s
v
(45)
78
and jv (ja ) is the vector (axial-vector) current that produces a quarkantiquark pair defined by
Eq. (10). With both and Z exchange the prefactors in Eq. (44) are

s 2 (ve2 + ae2 )vq2
2sve vq eq
+
,
F v (s) = eq2
s m2Z
(s m2Z )2
F a (s) =
s 2 (ve2 + ae2 )aq2

(s m2Z )2
(46)
where
f
vf =
T3 2ef sin2 W
,
2 sin W cos W
af =
T3
.
2 sin W cos W
(47)
Here ef is the charge for fermion f , T3 is the third component of weak isospin, W is the
weak mixing angle, and mZ the mass of the Z. Here we will focus on R v , since it gives the
dominant contribution and we are mainly interested in studying the impact of the resummation
of logarithms on the convergence and scale dependence of the perturbative expansion of R v .
A study of R a can be found in Ref. [29].
Since t mEM mv 2 and the propagator of a potential heavy quark scales as v 2 the
effects due to the width of the top quark are LO effects and have to be taken into account
by modifying the propagator E p2 /(2m) E + it p2 /(2m). This amounts to replacing
E E + it in Eq. (1). As noted in Ref. [1], this is only correct at LO. Higher-order electroweak
corrections have a much richer structure and it is not possible any longer to formulate the problem in terms of a t t final state. In particular, at NNLO there are interference effects (between
double and single resonant processes), QED radiation effects and non-factorizable corrections
(non-trivial interconnections between the decay products of the top quarks with the remainder
of the process). Any consistent approach beyond LO has to introduce additional operators with
fields corresponding to the incoming electrons and the decay products of the top quarks, and link
higher-order corrections of the and operators to three-point and higher-point vertices.
Even though there is an effective theory framework available for systematically taking into account these corrections [42], a full explicit calculation of electroweak effects at NNLO is still
lacking. If we are interested in the total cross section only, the situation is somewhat simpler,
since the non-factorizable corrections cancel [43], and the replacement E E + it becomes
correct with NLO accuracy. Furthermore, some of the electroweak corrections have be taken into
account by including them in the matching coefficients [44]. In this article we restrict ourselves
to the usual shift E E + it .
In Fig. 1 the fixed-order results are compared to the RGI results obtained using the procedure
described above and using the PS mass with the subtraction scale F required for the definition
of the threshold
mass set to 20 GeV. The plots were produced with an energy dependent soft scale

2s = 4m E 2 + t2 . In order to obtain a first rough estimate of the theoretical uncertainty we

show the cross sections as bands obtained by variation of the soft scale in the region 30 GeV
s 80 GeV, where these numbers refer to the scale s at E = 0. For the hard and ultrasoft scale
we take our default values h = mPS = 175 GeV and us = 2s /h . Our results are qualitative
consistent, but not equal, to those obtained in Refs. [28,29]. This agreement is not trivial, since
the ingredients included in the NNLL analysis are different in ours and their computations. This
may indicate that, even if the NNLL evaluation is incomplete, the qualitative features will hold in
the complete result. We observe that the scale dependence is much reduced once the logarithms
79
Fig. 1. Threshold scan for t t using the PS mass, mPS (20 GeV) = 175 GeV. The upper panel shows the fixed-order results,
LO, NLO and NNLO, whereas in the lower panel the RGI results LL, NLL and NNLL are displayed. The soft scale is
varied from s = 30 GeV to s = 80 GeV.
are taken into account and reduces from LL to NLL to NNLL. We also note that the size of
the corrections decreases for the RGI results, in particular the NNLL band is much closer to
the NLL band. However, the NNLL band does not overlap with the NLL band, indicating that a
theoretical error estimate relying on the scale dependence alone is too optimistic. Therefore, we
consider other possible source of errors in what follows.
The variation of the soft scale has been stopped at s = 30 GeV which might seem to be a
rather large value. In fact, the dependence of the normalization of the peak as a function of s
is very smooth at NNLL, up to a value of s 25 GeV, where it abruptly changes.2 At NLL,
already for s 30 GeV there is a rather large variation of the normalization. This is illustrated
in Fig. 2 which shows the s dependence of the normalization of the peak at LO/LL, NLO,
2 As we will see a similar pattern also appears for the bottomonium decays.
80
Fig. 2. The normalization of the peak of the RGI threshold cross section as a function of the soft scale s . The vertical
dashed lines show the limits of variation used in Fig. 1.
Fig. 3. The position of the peak of the RGI threshold cross section as a function of the soft scale s . The vertical dashed
lines show the limits of variation used in Fig. 1.
NLL, NNLO and NNLL. As observed in Ref. [20] the situation for small soft scales may be
remedied if multiple insertions of the Coulomb potential are taken into account. Since in our
result these, formally, higher-order contributions are not taken into account, we refrain from
using scales s 30 GeV and take s = 40 GeV as our default value for the soft scale unless
stated otherwise. The NNNLO result also depicted in Fig. 2 is obtained by re-expanding the
full RGI result and keeping only terms that are NNNLO, but dropping those of even higher order.
The difference between this result and the full RGI result is small except for scales s 30 GeV,
which is outside the range we use.
We also present a similar plot to show the scale dependence of the position of the peak in
Fig. 3. From this plot we can see that the NNLL resummation of logarithms significantly improves over the fixed-order NNLO evaluation. The scale dependence is reduced and also the
81
Fig. 4. Dependence of the t t threshold scan on the hard scale h , using the PS mass. At NNLL (NLL) the lower (upper)
curve corresponds to h = 250 GeV, whereas the upper (lower) curve corresponds to h = 100 GeV.
convergence of the perturbative series is better, which is reflected in a smaller difference between the NLL and NNLL result versus the NLO and NNLO result (this also happens for the
normalization of the total cross section). The difference between the NNNLO and NNLL curve
is again very small for s 30 GeV. From this plot we estimate the theoretical error for the determination of the position of the peak (which is related to the determination of the top mass) to
be of the order of 100 MeV.
In the fixed-order calculations the error is usually estimated from the soft-scale dependence,
even though one may think that the dominant uncertainty came from the magnitude of the correction. This is potentially more of a problem after the resummation of logarithms, since the
soft scale dependence is much less severe. Therefore, we have to be more careful with other
sources of uncertainties. In particular, the missing ultrasoft contributions make it important to
consider the dependence on the other scales as well. Since they are correlated, we can only vary
h together with us and, thus, consider in Fig. 4 the dependence on the hard scale h , setting s = 40 GeV. Variation of the hard scale around its natural value h = m by choosing
100 GeV h 250 GeV results in a scale dependence that is considerably larger than the soft
scale dependence. We note that this error is compatible with the magnitude of the difference between the NLL and NNLL result. It is to be expected that the situation improves once all ultrasoft
logarithms at NNLL are taken into account, but at this stage the rather large dependence of the
cross section on h has to be taken into account if a theoretical error is assigned. We also note
that the dependence on h only enters at NLL, thus the LL band in Fig. 4 is simply a line.
Finally we turn to Fig. 5, where we display the effects due to the QED corrections with the
default choice for the scales, h = m and s = 40 GeV. As previously mentioned, the QED
effects enter at NLL (thus there is no effect on the LL curve) and, compared to the desired
accuracy (top-quark mass measurement with an error m 100 MeV), they are large. They
change the normalization by up to 10% and result in a shift in the extracted MS mass of up to
100 MeV, making it mandatory to include them. We note that we have not changed the definition
of the PS mass. Thus the shifts shown in Fig. 5 are physical effects and are not compensated if
the PS mass is related to the MS mass.
82
Fig. 5. Effects of the QED corrections to the t t threshold scan. The hard ans soft scales are chosen as h = mPS =
175 GeV and s = 40 GeV.
We end this section by mentioning that all results presented here can be worked out in different
threshold mass schemes. The qualitative features of the results obtained are similar. We illustrate
this point using the RS and RS mass, again setting F = 20 GeV, and depict the threshold scan
in Fig. 6. Those plots, together with the plot in the PS scheme (lower panel in Fig. 1), allow us
to visualize the effect of using different threshold masses. In principle, the value of the threshold
masses can run from being numerically close to the MS mass (but not too close otherwise power
counting is broken) to being numerically close to the pole mass (but again not too close, otherwise
the renormalon cancellation does not take place). Within this allowed range, the threshold mass
definitions numerically closer to the MS mass have a smaller scale dependence. The price paid
is that the magnitude of the corrections (between different order in perturbation theory) is larger.
For the specific threshold masses we use, the RS mass is the one closest to the MS mass whereas
the RS mass is closest to the pole mass. The PS mass is in the middle, though somewhat closer
to the RS mass. Once experimental data is available it will be useful to consider different mass
definitions to obtain an estimate of the corresponding error.
4. The case of the bottom quark
We can also apply the results of Section 2 to a variety of observables for bb systems. Likely,
the theoretically cleanest observable on which one can use our results is non-relativistic sum
rules, which has already been considered in Ref. [9], where an accurate determination of the
bottom mass was obtained. Here we will consider inclusive electromagnetic decays (either to
e+ e or to ) of the bottomonium ground state for which we will provide analytic formulae.
These can be obtained from the results obtained for the non-relativistic Green function. The
spin-one decay has the following structure

(nS) e+ e

M (nS) 2m 2
CA EM eQ 2 (s=1) 2
n
(0) c1 d1
.
= 16
(48)
3 M (nS)
6m
83
Fig. 6. Threshold scan for t t using the RS and RS mass, mRS (20 GeV) = 175 GeV, mRS (20 GeV) = 175 GeV. The
upper panel shows the RGI results LL, NLL and NNLL with the RS and the lower panel with the RS mass. The soft
scale is varied from s = 30 GeV to s = 80 GeV.
The corrections to the wave function at the origin are obtained by taking the residue of the
Green function at the position of the poles
(s=1) 2 (0) 2

(0) = n (0) 1 + n(s=1) = Res Gs=1 (0, 0; E),
n
E=En
(49)
where the LO wave function is given by

(0) 2 1
(0) =
n
mCF s
2n
(s=1)
3
.
(50)
The corrections to n
produced by V have already been calculated with NNLO accuracy
[5,6] in the direct matching scheme. One can also obtain them in the dimensional regularized MS
scheme with NNLL accuracy by incorporating the RGI matching coefficients. One obtains then
84
the following correction to the wave function

n(s=1) =
s2 CA3
s (s )
log
20
s (us )

n 2
s 3a1 0
+
3L[n] + S[1, n] + 2nS[2, n] 1
+
4
2
3

2 5
+ CF CA Ds(1) L[n] S[1, n] + +
n 4

5
2 3
(2)
2
+ 2CF s D1,s L[n] S[1, n] 2 + +
n 2
8n
(2)

CF2 s DS 2 ,s
2 11
L[n] S[1, n] + +
3
n 12

(2)
3CF2 s Dd,s
2 1
L[n] S[1, n] + +
2
n 2
CF2 (2)
D s
4 2,s

CF2 s2
2 3
3
+ c4
L[n] S[1, n] 2 + +
2
n 2
4n

2
+ s 2 3a12 + 3a2 14a1 0 + 402 21 + 02 2

(4)
8a1 0 n 2 402 n 2 21 n 2 02 n2 4
+
+
+ 24a1 0 L[n]
3
3
3
9
2802 L[n] + 61 L[n]
1602 n 2
L[n] + 2402 L[n]2
3
+ 8a1 0 S[1, n] 2002 S[1, n] + 21 S[1, n]
1202 S[1, n] 802 n 2 S[1, n]
n
3
+ 1602 L[n]S[1, n] + 802 S[1, n]2 + 802 S[2, n] + 16a1 0 nS[2, n]

802 nS[2, n] + 41 nS[2, n]
402 n2 2 S[2, n]
+ 3202 nL[n]S[2, n]
3
+ 1602 nS[1, n]S[2, n] + 402 n2 S[2, n]2 + 2802 nS[3, n] 2002 n2 S[4, n]

2
2 2
2
240 nS2 [2, 1, n] + 160 n S2 [3, 1, n] + 200 n (3) ,
(51)
where
L[n] = log

s n
,
mCF s
S[a, n] =
n

1
,
ka
k=1
S2 [a, b, n] =
n

1
S[b, k].
ka
k=1
(52)
85
For completeness we also give the corrections to the wave function induced by the QED effect,
Eq. (23). They read

2
2
2
3eQ
eQ
3eQ
EM
EM 3a1
EM
n(s=1) n(s=1) +
+
+
CF s
CF
2
CF s2

5 n 2
(53)
+ 0 3L[n] + S[1, n] + 2nS[2, n]
.
2
3
Of course, the corrections to the matching coefficients of the current, Eqs. (40) and (41) have to
2 = 1/9 these QED corrections
be taken into account as well. Due to the additional suppression eQ
are numerically not very important.
From these expressions one can obtain the wave-function correction for the spin zero case
n(s=0) = n(s=1) + ns ,
where
ns

2 2 (2)
4 7
= CF DS 2 ,s s 2L[n] + 2S[1, n]
3
n 3
(54)
(55)
by using the results from Ref. [27]. Therefore, the decay of the pseudoscalar heavy quarkonium
to two photons reads (d0 = d1 )

b (nS)
2

2 2
EM eQ
(s=0) 2

c0 d0 Mb (nS) 2m .
(0)
= 16CA
(56)
n
Mb (nS)
6m
We now perform a phenomenological analysis of these results. We restrict our analysis to the
ground state of bottomonium. For the mass of the b (1S), we use the (1S) mass, which is
consistent to the order of interest. When we perform the numerical analysis, we expand the
expressions (except for the overall factor 1/M2 (1S)/b (1S) ) in Eqs. (48) and (56), and drop
subleading corrections, in order to work strictly at LL, NLL (NLO) and NNLL (NNLO). In
particular, this applies to the relativistic correction proportional to ds , where M (1S)/b (1S) 2m
(0)
is replaced by E1 = mCF2 s2 /4. The expressions above have been written in the pole scheme.
Therefore, we change to a threshold scheme suitable for the renormalon cancellation. We consider three possible schemes: the PS, the RS and the RS scheme with the subtraction scale
F = 2 GeV. The numerical values of the bottom-quark mass to be used in the various schemes
are taken from Ref. [9] and are given by mPS (2 GeV) = 4.515 GeV, mRS (2 GeV) = 4.370 GeV,
and mRS (2 GeV) = 4.750 GeV. Apart from the different numerical values for m to be used, the
first change in the formulae appears (at most) at NNLL (NNLO) and is due to the shift
m mX (f ) + mX (f ),
mX (0)
,
|m |mmX + 3
mX
(57)
(58)
where mX represents a generic threshold mass, (0) is the decay width at lowest order and
the shift in the mass mX mv 2 . We notice that in the RS scheme mRS = 0 at order mv 2 .
Therefore, in this scheme, with the current precision of the calculation, the expressions for are
equivalent to those in the pole mass scheme.
86
Fig. 7. Prediction for the (1S) decay rate to e+ e . We work in the RS scheme.
Fig. 8. Prediction for the b (1S) decay rate to two photons. We work in the RS scheme.
The results for the vector and pseudoscalar decay can be found in Figs. 7 and 8 respectively.
From the numerical analysis, we find that the NNLL corrections are huge, especially for the
b (1S) decay. The result we obtain for this decay is compatible with the number obtained
in Ref. [27]. This is somewhat reassuring, since in that reference the ratio of the spin-one spinzero decay was considered, which was much more scale independent, as well as more convergent
(yet still large) than for each of the decays themselves. This agreement can be traced back to the
fact that, for the spin-one decay, for which we can compare with experiment, we find that the
NNLL result improves the agreement with the data. Overall, the resummation of logarithms
always significantly improves over the NNLO result, the scale dependence greatly improves, as
well as the convergence of the series. On the other hand the problem of lack of convergence of
the perturbative series is not really solved by the resummation of logarithms and it remains as an
open issue. Due to the lack of convergence we refrain from giving numbers (and assigning errors)
for our analysis. In this respect we cannot avoid to mention that, whereas the perturbative series
87
Fig. 9. Prediction for the (1S) decay rate to e+ e at LL, NLL and NNLL for the PS, RS and RS mass.
in non-relativistic sum rules is sign-alternating, is not sign-alternating for the electromagnetic

decays. Finally, we would also like to remark the strong scale dependence that we observe at
low scales, which we believe to have the same origin than the one observed in t t production near
threshold in the previous section.
As for the t t case the NNNLO curve, which contains the logarithmically enhanced NNNLO
terms, follows closely the RGI curve up to a certain value of s . For smaller soft scales, the two
curves start to deviate. However, in the bb case, this deviation takes place already for reasonably
large values of s 2.5 GeV. This seems to suggest that resummation is rather more important
in the bb case. However, we have to keep in mind that some logarithms at NNLL are still missing
in our analysis and this might well affect this conclusion.
We also illustrate the dependence on the threshold mass evaluation in Fig. 9. Compared with
the uncertainty due to the lack of convergence of the perturbative series, the dependence on the
threshold mass is negligible.
5. Conclusions
We have studied the effect of the resummation of logarithms for t t production near threshold
and inclusive electromagnetic decays of heavy quarkonium. This analysis is complete at NNLO
and includes the full resummation of logarithms at NLL accuracy and some partial contributions
at NNLL accuracy.
Compared with fixed-order computations the scale dependence and convergence of the perturbative series is greatly improved for both the position of the peak and the normalization of the
total cross section of t t production near threshold. Nevertheless, we identify a possible source of
a large scale dependence in the result. While the result is very stable with respect to the variation
of the soft scale s it shows a considerably larger dependence on the hard scale h . This might
well be related to the fact that the variation with respect to h is correlated with the ultrasoft scale
variation and the ultrasoft logarithms are not fully included at the NNLL level in our result. At
present we estimate the remaining theoretical uncertainty of the normalization of the total cross
section to be of the order of 10% and for the position of the peak of the order of 100 MeV, based
on the difference between the NLL and NNLL plots, as well as on the hard scale dependence.
88
We note that this estimate of the theoretical error is somewhat larger than in Ref. [45], due to the
fact that we use more possible sources of the theoretical error. We would like to remark that the
use of the RGI provides us also with a significantly improved determination of the top mass.
For the inclusive electromagnetic bottomonium decays the corrections are even larger than
in the top case. In the case of the (1S) e+ e , they bring the final prediction into better
agreement with experiment. For the b , there is no experimental data at present and the
presence of huge corrections make a reliable theoretical prediction difficult.
Finally we remark that the NNNLO part of the RGI results is numerically considerably more
important than the even higher-order terms, in particular for the top case. Thus it would be highly
desirable to combine a full NNNLO evaluation with a complete NNLL result in order to obtain
a satisfactory theoretical prediction of heavy-quark pair production near threshold.
Acknowledgements
The work of A.P. was supported in part by a Distinci from the Generalitat de Catalunya,
as well as by the contracts MEC FPA2004-04582-C02-01, CIRIT 2005SGR-00564 and the EU
network EURIDICE, HPRN-CT2002-00311. A.P. acknowledges discussions with A.A. Penin
and J. Soto.
Appendix A. Insertions
In this appendix we discuss how to obtain the perturbative expansion of the Green function.
For the loop integration measure we use
2 E

d d pi
e
d p i
(A.1)
4
(2)d
such that the MS-scheme corresponds to minimally subtracting the poles in . We start by writing
the potential, Eq. (20), for the case of a spin triplet S wave

CF 2
q2
s (s )
CF s3 CA3
V |3 S1 = 2 s a1 0 log 2
log
s (us )
q
s
q2 60
(2)
2CF D1,s p2 + p 2
p4
c4 3 (2)d (d) (q)
4m
m2
q2

3
2
2
CF s
q
2
2 q
a2 (2a1 0 + 1 ) log 2 + 0 log 2

4q2
s
s
CF DS(2)
2 ,s (d 4)(d 1)
d
m2
m2
(1)
1
2
2

E
( 2 ) ( 12 + )
e
CF CA Ds
(1
mq 1+2
3/2 (1 2)
2
s

(2)
CF D2,s
p2 p 2 2
+
1 ,
m2
q2
(2)
3CF Dd,s
(A.2)
where we have used the explicit expression for Vs . The terms in the first line of Eq. (A.2)
are NLO and we have to consider double insertions of such terms. All other terms are NNLO.
89

(2)
(2)
(1)
Insertions of the potentials with Dd,s , DS 2 ,s , Ds
(2)
and D2,s result in divergences. Thus, the

(2)
corresponding coefficients have to be known in d dimensions. The insertions with c4 , D1,s and
(2)
D2,s can be related to other insertions, discussed below, using the d-dimensional equation

2
p
c (p, p ; E)
E G
m

d d k 4CF s
d (d)

= (2) (p p ) +
(A.3)
Gc (p k, p ; E).
(2)d
k2
For the terms in the second line of Eq. (A.2) we get
4

(2)
2CF D1,s p22 + p23
p3 c4
d (d)
c (p3 , p4 )
(2) (q23 ) +
G
d p i Gc (p1 , p2 )
2
4m3
m2
q23

E
c (p1 , p2 )
c (p1 , p2 ) VEOM G
c (p3 , p4 ),
d p i G
d p i G
=
c4
2m
(A.4)
where we defined qij pi pj , qij |pi pj | and

E2
E
(2)
VEOM = c4
(2)d (d) (q23 ) 2c4 s + 4D1,s CF
2
4m
mq23

E 2 ( 1 ) ( 1 + )
C 2 s 2
c4
(2) e
2
2
.
F 1+2 s + 2D1,s
2
3/2 (1 2)
mq23
2
s
We thus need the following insertions:

2 /2 ]
log[q23
s
c (p1 , p2 )
Il
d p i G
Gc (p3 , p4 ),
2
q23

2 /2 ]
log2 [q23
s
c (p1 , p2 )
d p i G
Il 2
Gc (p3 , p4 ),
2
q23

2 /2 ]
log[q23
1
s
c (p1 , p2 )
Il,c
Gc (p3 , p4 ) 2 G
d p i G
c (p5 , p6 ),
2
q23
q45

2 /2 ]
2 /2 ]
log[q23
log[q45
s
s
c (p1 , p2 )
d p i G
(p
,
p
)
G
Gc (p5 , p6 ),
Il,l
c
3
4
2
2
q23
q45

c (p3 , p4 ),
c (p1 , p2 )G
I1
d p i G

c (p3 , p4 ),
c (p1 , p2 )(2)d1 (p2 p3 )G
d p i G
I
Ina
d p i G(p1 , p2 )
2
s
1+2
q23
G(p3 , p4 ).
(A.5)
(A.6)
(A.7)
(A.8)
(A.9)
(A.10)
(A.11)
(A.12)
All these insertions have been computed in Ref. [34]. For the readers convenience we present
the results here. We write them in terms of

4mE
4mE
log 2 + 2 log 2 + 2 + 2E ,
(A.13)
90
H1 3 F4 (1, 1, 1, 1; 2, 2, 1 ; 1),
(A.14)
H2 4 F5 (1, 1, 1, 1, 1; 2, 2, 2, 1 ; 1)
(A.15)
and the argument of the functions ([x]

d log [x]/dx) and its derivatives is always under
stood to be (1 ) 1 (CF s )/(2 E/m ) unless stated otherwise.
We start by writing the Green function before MS subtraction

1
CF m2 s 1

+ 1 + G ,
Gc (0, 0; E) =
(A.16)
8
2
where G denotes the term of order of the Green function. This term is not explicitly known,
but it is not needed for the calculation of the cross section, even though it results in finite terms
in some of the insertions below, because it cancels in the total sum.
The insertions of the higher-order corrections to the Coulomb potential with terms of
the form logj [q 2 /2s ]/q 2 are computed by taking derivatives with respect to at = 0 of

c (p1 , p2 )(q23 /s )22 G
c (p3 , p4 ). Taking the first derivative yields the insertion of
d p i G
a single logarithm. We get
2 1
m
1
1
2
(A.17)
Il = 2 + 2 + 4 + 12 4 16H1 + 2
.
2

4
64
2
The insertion of a logarithm squared is obtained by taking the second derivative.
2 1
2
2
1
1
2
m
3 + 42
Il 2 = 3 + 2 + +
2
12 3
64
2

2

8 3 + + 4H1 +
12

5 2
1
128
16 2 64 + 2
48

16
2
1
+ 128H2 64(1 )H
32
3
3
+ 16S1 + 52 (3) + 196 +
11 2
,
2
(A.18)
where
S1 =

n=1

2 [n] [1 ]
+ n ] + (1 + n) [1 + n] .
2[n]
2[1
(1 + n)2 [1 + n ]
(A.19)
Double insertions of Coulomb potentials are computed in a similar way. The results read

1

m2

+ + 2S2 ,
+ 2
(A.20)
Il,c = 2
3
2
3
64 CF s
where
S2 =

]
(n + )[n
n=1
(n )3
(A.21)
91
and
m2
64 3 CF s
where
S5 =
1

Il,l = + 4 ( ) + + S2
2
6
4
8
3 8 2 2 2
8
8
4

4
4

3
3
2
2 2
2,
4S5 8S
3
6
2

]
] (n + )[n
]
(n + )[n
2[1
2
+
n(n )
n
n
(n )3
(A.22)
(A.23)
n=1
Inserting 1 results in the square of the Green function

2
I1 = G(0, 0; E) .
Note that I1 obtains a finite contribution due to
computed directly and takes a very simple form

1
m
I = 22 + 2 + 1,
4CF s
G .
(A.24)
The insertion due to the function can be
(A.25)
while the insertion due to the non-analytic potential is more complicated and, written in terms of
3 + log 2
reads
m3 CF s
64 3
1
Ina =
(A.26)

1
1 2
2

+
+
2
log
2
+ 2 2
2 2
4
5 2
+
+ 2G + 4 log2 2 + 4 log 2
.
24
(A.27)
Finally, the insertion due to the L2 operator, Eq. (19), can be related to Ina and I1 with the help
of Eq. (18) and we get
2

p2 p23 2
c (p3 , p4 )
c (p1 , p2 )
IL2
1
G
d p i G
2
q23

C 2 m4 s2 1
1
1
+
3
.
= F
(A.28)
2(4)2 4
22
We are finally in a position to write down the expression for the Green function

c
NNLL = 1 c4 E G
G
(2)
2
3

s=1
eQ
D
CA
EM
s2
s (s )
s
E
E 1,s
2m
2
s s 1+ 2m
c4 + m
s + 4 a1 + CF s + (4)2 a2 + 60 s log s (us )

3
s3
2
2
+ i CF s 0 +
(2a1 0 + 1 ) Il i 2 s CF 02 Il 2
4
4
92

2
2
4eQ
EM
4 4 2 2
2E
i
+
i
i
c4 I
a1 +
I
l,c
l,l
s
F
0
CF s2
4m

(2)
(2) (d 4)(d 1)
2 CF (2)
2 CF
+i
D I 2 +i
I1
3Dd,s + DS 2 ,s
d
m2 2,s L
m2
2

2
C A CF 2
(2)
2 CF s
(1)
i
c4 s + 4D1,s +
(1 )Ds
2m
m
4
2s4 0 CF2
eE
2 ( 12 ) ( 12 + )
Ina .
3/2 (1 2)
(A.29)
We mention once more that we use strictly expanded results. Thus the terms beyond NNLL that
are present in Eq. (A.29) are dropped.
Appendix B. RGI potentials
For ease of reference we explicitly display the RGI potentials not shown in the main body of
the paper (we remind that us = 2s /h ):

q2 s (s )
Vs (s ) = s (s ) 1 + a1 0 log 2
4
s

2
q2 s2 (s )
q
a2 (2a1 0 + 1 ) log 2 + 02 log2 2
s
s 16 2

3
C
s (s )
+ A s3 (s ) log
,
60
s (us )

16 CA
s (s )
(1)
2
+ CF log
,
Ds (s ) = s (s ) 1 +
30 2
s (us )

8CA
s (s )
(2)
log
,
D1,s (s ) = s (s ) 1 +
30
s (us )
(2)
D2,s (s ) = s (s ),

3
dsv (s ) + CF dvv (s ) ,
2CF

s (s )
(2)
DLS,s (s ) =
cS (s ) + 2cF (s ) ,
3
(2)
DS 2 ,s (s ) = s (s )cF2 (s )
(2)
DS12 ,s (s ) = s (s )cF2 (s ),
1
(s ) 0
where (z = [ ss(
] 1 1/(2)s (s ) log( hs ), 0 =
h)
cF (s ) = zCA ,
cS (s ) = 2zCA 1,
(B.1)
11
3 CA
43 TF nf )
93

5CA + 4TF nf 2CA CA + 16CF 8TF nf
9CA
cD (s ) =
z
+
9CA + 8TF nf
4CA + 4TF nf
2(CA 2TF nf )
7CA2 + 32CA CF 4CA TF nf + 32CF TF nf 4TF nf /32CA /3
z
4(CA + TF nf )(2TF nf CA )

13CA
8TF nf 2CA
20 32 CF
6
1z
+
+
+
z
,
9CA
13 13 CA
+
dss (s )
+ dvs (s )
CF

2
s (h ) z0 1
0

2
2C

27CA
5CA + 4TF nf
0
A 1
z 0
s (h )
9CA + 8TF nf 0
4CA + 4TF nf 0 2CA

CA + 16CF 8TF nf 0
z 1
+
2(CA 2TF nf )
= (2CF 3CA )
7CA2 + 32CA CF 4CA TF nf + 32CF TF nf

4(CA + TF nf )(2TF nf CA )
+4T n /32C /3

30
A
z 0 F f
1
30 + 4TF nf 2CA

2C

8TF nf
0
20 32 CF
A 1 +
+
z 0
+
9CA 0 2CA
13 13 CA

13C

60
A
z0 6 1
,
z 0 1
60 13CA

dsv (s )
CA
+ dvv (s ) =
s (h ) z0 2CA 1 .
CF
0 2CA
+
References
[1] V.S. Fadin, V.A. Khoze, JETP Lett. 46 (1987) 525, Pisma Zh. Eksp. Teor. Fiz. 46 (1987) 417;
V.S. Fadin, V.A. Khoze, Sov. J. Nucl. Phys. 48 (1988) 309, Yad. Fiz. 48 (1988) 487.
[2] M. Martinez, R. Miquel, Eur. Phys. J. C 27 (2003) 49, hep-ph/0207315.
[3] A.H. Hoang, et al., Eur. Phys. J. direct C 2 (2000) 1, hep-ph/0001286.
[4] V.A. Novikov, et al., Phys. Rev. Lett. 38 (1977) 626;
V.A. Novikov, et al., Phys. Rev. Lett. 38 (1977) 791, Erratum;
V.A. Novikov, et al., Phys. Rep. 41 (1978) 1.
[5] K. Melnikov, A. Yelkhovsky, Phys. Rev. D 59 (1999) 114009.
[6] A.A. Penin, A.A. Pivovarov, Nucl. Phys. B 549 (1999) 217, hep-ph/9807421.
[7] A.H. Hoang, Phys. Rev. D 59 (1999) 014039, hep-ph/9803454;
A.H. Hoang, Phys. Rev. D 61 (2000) 034005, hep-ph/9905550.
[8] M. Beneke, A. Signer, Phys. Lett. B 471 (1999) 233, hep-ph/9906475.
[9] A. Pineda, A. Signer, Phys. Rev. D 73 (2006) 111501, hep-ph/0601185.
[10] N. Brambilla, A. Pineda, J. Soto, A. Vairo, Rev. Mod. Phys. 77 (2005) 1423.
[11] M. Beneke, V.A. Smirnov, Nucl. Phys. B 522 (1998) 321, hep-ph/9711391.
[12] W.E. Caswell, G.P. Lepage, Phys. Lett. B 167 (1986) 437.
[13] A. Pineda, J. Soto, Nucl. Phys. B (Proc. Suppl.) 64 (1998) 428, hep-ph/9707481;
A. Pineda, J. Soto, Phys. Rev. D 59 (1999) 016005, hep-ph/9805424.
(B.2)
94
[14] N. Brambilla, A. Pineda, J. Soto, A. Vairo, Nucl. Phys. B 566 (2000) 275, hep-ph/9907240.
[15] N. Brambilla, A. Pineda, J. Soto, A. Vairo, Phys. Rev. D 60 (1999) 091502, hep-ph/9903355.
[16] B.A. Kniehl, A.A. Penin, Nucl. Phys. B 563 (1999) 200, hep-ph/9907489;
B.A. Kniehl, A.A. Penin, Nucl. Phys. B 577 (2000) 197, hep-ph/9911414.
[17] N. Brambilla, A. Pineda, J. Soto, A. Vairo, Phys. Lett. B 470 (1999) 215, hep-ph/9910238.
[18] B.A. Kniehl, A.A. Penin, M. Steinhauser, V.A. Smirnov, Phys. Rev. D 65 (2002) 091503, hep-ph/0106135.
[19] A.A. Penin, V.A. Smirnov, M. Steinhauser, Nucl. Phys. B 716 (2005) 303, hep-ph/0501042.
[20] M. Beneke, Y. Kiyo, K. Schuller, Nucl. Phys. B 714 (2005) 67, hep-ph/0501289.
[21] P. Marquard, J.H. Piclum, D. Seidel, M. Steinhauser, hep-ph/0607168.
[22] A. Pineda, Phys. Rev. D 65 (2002) 074007, hep-ph/0109117.
[23] A.H. Hoang, I.W. Stewart, Phys. Rev. D 67 (2003) 114020, hep-ph/0209340.
[24] B.A. Kniehl, A.A. Penin, A. Pineda, V.A. Smirnov, M. Steinhauser, Phys. Rev. Lett. 92 (2004) 242001, hepph/0312086.
[25] A.A. Penin, A. Pineda, V.A. Smirnov, M. Steinhauser, Phys. Lett. B 593 (2004) 124, hep-ph/0403080.
[26] A. Pineda, Phys. Rev. D 66 (2002) 054022, hep-ph/0110216.
[27] A.A. Penin, A. Pineda, V.A. Smirnov, M. Steinhauser, Nucl. Phys. B 699 (2004) 183, hep-ph/0406175.
[28] A.H. Hoang, A.V. Manohar, I.W. Stewart, T. Teubner, Phys. Rev. Lett. 86 (2001) 1951, hep-ph/0011254.
[29] A.H. Hoang, A.V. Manohar, I.W. Stewart, T. Teubner, Phys. Rev. D 65 (2002) 014014, hep-ph/0107144.
[30] M.E. Luke, A.V. Manohar, I.Z. Rothstein, Phys. Rev. D 61 (2000) 074025, hep-ph/9910209.
[31] I.I.Y. Bigi, M.A. Shifman, N. Uraltsev, Annu. Rev. Nucl. Part. Sci. 47 (1997) 591, hep-ph/9703290;
A.H. Hoang, Z. Ligeti, A.V. Manohar, Phys. Rev. Lett. 82 (1999) 277, hep-ph/9809423.
[32] M. Beneke, Phys. Lett. B 434 (1998) 115, hep-ph/9804241.
[33] A. Pineda, JHEP 0106 (2001) 022, hep-ph/0105008.
[34] M. Beneke, A. Signer, V.A. Smirnov, Phys. Lett. B 454 (1999) 137, hep-ph/9903260.
[35] M. Beneke, hep-ph/9911490.
[36] A. Czarnecki, K. Melnikov, Phys. Rev. Lett. 80 (1998) 2531, hep-ph/9712222;
M. Beneke, A. Signer, V.A. Smirnov, Phys. Rev. Lett. 80 (1998) 2535, hep-ph/9712302.
[37] A. Czarnecki, K. Melnikov, Phys. Lett. B 519 (2001) 212, hep-ph/0109054.
[38] Y. Schroder, Phys. Lett. B 447 (1999) 321, hep-ph/9812205;
M. Peter, Phys. Rev. Lett. 78 (1997) 602, hep-ph/9610209.
[39] A. Pineda, J. Soto, Phys. Lett. B 495 (2000) 323, hep-ph/0007197.
[40] G.T. Bodwin, E. Braaten, G.P. Lepage, Phys. Rev. D 51 (1995) 1125, hep-ph/9407339;
G.T. Bodwin, E. Braaten, G.P. Lepage, Phys. Rev. D 55 (1997) 5853, Erratum.
[41] N. Brambilla, D. Eiras, A. Pineda, J. Soto, A. Vairo, Phys. Rev. D 67 (2003) 034018, hep-ph/0208019.
[42] A.P. Chapovsky, V.A. Khoze, A. Signer, W.J. Stirling, Nucl. Phys. B 621 (2002) 257, hep-ph/0108190;
M. Beneke, A.P. Chapovsky, A. Signer, G. Zanderighi, Phys. Rev. Lett. 93 (2004) 011602, hep-ph/0312331;
M. Beneke, A.P. Chapovsky, A. Signer, G. Zanderighi, Nucl. Phys. B 686 (2004) 205, hep-ph/0401002.
[43] V.S. Fadin, V.A. Khoze, A.D. Martin, Phys. Rev. D 49 (1994) 2247;
V.S. Fadin, V.A. Khoze, A.D. Martin, Phys. Lett. B 320 (1994) 141, hep-ph/9309234;
K. Melnikov, O.I. Yakovlev, Phys. Lett. B 324 (1994) 217, hep-ph/9302311.
[44] A.H. Hoang, C.J. Reisser, Phys. Rev. D 71 (2005) 074022, hep-ph/0412258.
[45] A.H. Hoang, Phys. Rev. D 69 (2004) 034009, hep-ph/0307376.
The membrane at the end of the (de Sitter) universe

A. Batrachenko a , M.J. Duff a, , J.X. Lu a,b
a Michigan Center for Theoretical Physics, Randall Laboratory, Department of Physics, University of Michigan,
Ann Arbor, MI 48109-1120, USA

b Interdisciplinary Center for Theoretical Study, University of Science and Technology of China,
Hefei, Anhui 230026, PR China

Received 17 August 2006; accepted 20 September 2006
Available online 16 October 2006
Abstract
The original membrane at the end of the universe corresponds to a probe M2-brane of signature (2, 1)
occupying the S 2 S 1 boundary of the (10, 1) spacetime AdS4 S 7 , and is described by an OSp(4/8)
SCFT. However, it was subsequently generalized to other worldvolume signatures (s, t) and other spacetime
signatures (S, T ). An interesting special case is provided by the (3, 0) brane at the end of the de Sitter
universe dS4 which has recently featured in the dS/CFT correspondence. The resulting CFT contains the
one recently proposed as the holographic dual of a four-dimensional de Sitter cosmology. Supersymmetry
restricts S, T , s, t by requiring that the corresponding bosonic symmetry O(s + 1, t + 1) O(S s, T t)
be a subgroup of a superconformal group. The case of dS4 AdS7 is doubly holographic and may be
regarded as the near horizon geometry of N2 M2-branes or equivalently, under interchange of conformal
and R-symmetry, of N5 M5-branes, provided N2 = 2N52 . The same correspondence holds in the pp-wave
limit of conventional M-theory.
1. Introduction
The original membrane at the end of the universe [110] corresponds to a probe M2-brane
of signature (2, 1) occupying the S 2 S 1 boundary of the (10, 1) spacetime AdS4 S 7 and is
described by an OSp(4/8) superconformal field theory. However, it was subsequently general
Research supported in part by DOE grant DE-FG02-95ER40899.
E-mail addresses: abat@umich.edu (A. Batrachenko), mduff@umich.edu, m.duff@imperial.ac.uk (M.J. Duff),

jxlu@umich.edu (J.X. Lu).
doi:10.1016/j.nuclphysb.2006.09.017
96
A. Batrachenko et al. / Nuclear Physics B 762 (2007) 95111
Table 1
M2-branes with worldvolume signature (s, t) in spacetime signature (S, T )
(S, T )
(s, t)
Bosonic symmetry
Supergroup
(10, 1)
(9, 2)
(9, 2)
(6, 5)
(6, 5)
(5, 6)
(5, 6)
(2, 9)
(2, 9)
(1, 10)
(2, 1)
(3, 0)
(1, 2)
(2, 1)
(0, 3)
(3, 0)
(1, 2)
(2, 1)
(0, 3)
(1, 2)
O(3, 2) O(8)
O(4, 1) O(6, 2)
O(2, 3) O(8)
O(3, 2) O(4, 4)
O(1, 4) O(6, 2)
O(4, 1) O(2, 6)
O(2, 3) O(4, 4)
O(3, 2) O(8)
O(1, 4) O(2, 6)
O(2, 3) O(8)
OSp(4/8)
OSp (4/8)
OSp(4/8)
OSp(4/4, 4)
OSp (4/8)
OSp (4/8)
OSp(4/4, 4)
OSp(4/8)
OSp (4/8)
OSp(4/8)
ized [6,9] to brane worldvolumes with s space and t time dimensions moving in a spacetime
with S s space and T t time dimensions. The brane occupies the boundary of a universe of
constant curvature so that the bosonic symmetry is O(s + 1, t + 1) O(S s, T t). Supersymmetry restricts the values of s, t, S, T to those for which this bosonic symmetry is a subgroup
of a superconformal group, and the resulting superconformal theories have (s + t) 6. For example, as discussed in [6], the possible signatures of M-theory are (10, 1), (9, 2), (6, 5), (5, 6),
(2, 9) and (1, 10) and the possible M2-branes have worldvolume signatures (3, 0), (2, 1), (1, 2)
and (0, 3). The corresponding superconformal groups are given in Table 1.
In view of the connections [11] between the original membrane at the end of the universe and
the AdS/CFT correspondence [1214], and in view of the recent interest in a possible dS/CFT
correspondence [1517], it is natural to re-examine the special case of a (3, 0) brane at the end of
the de Sitter universe1 dS4 . We show that the resulting CFT contains the one recently proposed
as the holographic dual of a four-dimensional de Sitter cosmology [21,22].
In the usual signature all extended objects appear to suffer from worldvolume ghosts because
the kinetic term for the X 0 coordinate enters with the wrong sign. These are easily removed,
however (at least at the classical level) by the presence of diffeomorphisms on the worldvolume
which allow us to fix a gauge where only positive-norm states propagate e.g. the light cone
gauge for strings and its membrane analogues. Alternatively we may identify the d worldvolume
coordinates i with d of the D spacetime coordinates X i (i = 1, 2, . . . , d) leaving (D d)
coordinates X I (I = 1, . . . , D d) with the right sign for their kinetic energy [8]. Of course, this
only works if we have one worldvolume time coordinate that allows us to choose a light-cone
gauge or else set = t .
In the same spirit, we could now require absence of ghosts (or rather absence of classical
instabilities since we are still at the classical level) for arbitrary signature by requiring that the
transverse group SO(S s, T t) which governs physical propagation after gauge-fixing,
be compact. This requires T = t . It may be argued, of course, that in a world with more than
one time dimension, ghosts are the least of your problems. Moreover, in contrast to strings,
unitarity on the worldvolume does not necessarily imply unitarity in spacetime. This is because
the transverse group no longer coincides with the little group. (For example, the (2, 1) object in
(10, 1) spacetime and the (1, 2) object in (9, 2) spacetime both have transverse group SO(8), but
1 In fact the epithet membrane at the end of the universe is even more appropriate in the de Sitter case since the
restaurant in Douglas Adams book Restaurant at the end of the universe [18] is located at temporal infinity.
97
the former has little group SO(9) and the latter SO(8, 1).) Ref. [6] remained agnostic about the
physical significance of non-compact transverse groups, but simply noted that in the compact
case the list of superconformal groups matches those in Nahms classification [23].
However, an attempt to render these theories respectable has been made in a series of interesting papers by Hull and collaborators [15,2430], in which it is shown that these theories with
unconventional signatures are related to the usual signature by means of a timelike T-duality.
Hull denotes the (10, 1), (9, 2) and (6, 5) signatures as M-theory, M -theory and M -theory, respectively. Unlike M-theory, M - and M -theories can admit de Sitter vacua invariant under a
de Sitter supergroup which does not have unitary highest weight representations. This lack of
unitarity representations is reflected in the fact that some of the fields have kinetic terms with
the wrong sign. However, since these theories are related to the conventional ones by a timelike
T-duality they are no worse than conventional theories compactified in the time direction and so
perhaps might make sense after all. Thus Hulls suggestion is both radical and conservative at the
same time. It is radical in invoking spacetimes with unusual signature but conservative in saying
that the only such theories we need worry about are those dual to the conventional theories.
It seems that one is forced to this interpretation if one wants to combine dS/CFT space
with supersymmetry [15]. Alternatively, one can be content to look at non-supersymmetric
vacua [16,17]. In Section 3, we shall explore both possibilities, having first reviewed the AdS
case in Section 2.
Finally in Section 4 we note that the case of dS4 AdS7 is doubly holographic and may be
regarded as the near horizon geometry of N2 M2-branes or equivalently, under interchange of
conformal and R-symmetry, of N5 M5-branes, provided N2 = 2N52 . The same correspondence
holds in the pp-wave limit of conventional M-theory.
2. A probe (2, 1)-brane on the boundary of AdS4 S 7
2.1. The AdS background
We begin by reviewing the original membrane at the end of the universe for which the spacetime is AdS4 S 7 and for which the supermembrane occupies the S 1 S 2 boundary of the AdS4 .
It will be useful to regard the geometry as the near-horizon limit of a stack of M2-branes.
For M-theory in the usual (10, 1) signature, the low-energy effective field theory is D = 11
supergravity with bosonic action

1 2
1
11
A3 F 4 F 4 ,
SM = d x g R F4
(2.1)
48
12
where F4 = dA3 . The M2-brane solution is given by [31]

2
,
ds 2 = H 2/3 dt 2 + dx12 + dx22 + H 1/3 dy32 + + dy92 + dy10
A012 = H 1 ,
(2.2)
where H (y3 , . . . , y10 ) is a harmonic function in the transverse space. For a stack of N2 M2-branes
at y = 0, we take
6
L2
,
H =1+
(2.3)
y
where y 2 = mn y m y n with m running over the spatial indices in the transverse space, and
L2 = (25 2 N2 )1/6 lp with lp being the eleven-dimensional Planck length. The worldvolume has
98
signature (2, 1) and the solution has bosonic symmetry ISO(2, 1) SO(8). It is non-singular at
y = 0 with near-horizon geometry AdS4 S 7 .
2.2. Unitary but non-supersymmetric
In anticipation of finding in the next section a unitary but non-supersymmetric (3, 0) brane on
the boundary of dS4 , we begin by forgetting the D = 11 supergravity origins of the (2, 1) brane
and simply look at its bosonic sector in a background of AdS4 with a 4-form that follows from
the A012 given above. We write the AdS4 metric as
2

2
2
d + sinh2 d 2 + sin2 d 2 cosh2 dt 2
= RAdS
dsAdS
(2.4)
4
4
and the 4-form as
3
F4 = 3RAdS
cosh sinh2 dt d 2 ,
4
(2.5)
where n is the volume form of a unit n-sphere and the AdS4 radius is RAdS4 = L2 /2.
Now consider a probe (2, 1)-brane with a worldvolume action given by the bosonic sector of
the M2-brane [39,40]

1
1
S 2 = T2 d 3
ij i x M j x N gMN (x) +
2
2
1
+ q ij k i x M j x N k x P AMNP (x) ,
(2.6)
3!
where T2 is the membrane tension, i (i = 0, 1, 2) are the worldvolume coordinates, ij is
the worldvolume metric and x M ( ) are the spacetime coordinates (M = 0, 1, . . . , 10). Following [35] we allow for a non-extremality parameter q where the BPS condition corresponds to
q = 1.
The embedding of the membrane in AdS4 is given by
t = 0,
= 1,
= 2,
(2.7)
so that its worldvolume occupies the S 1 S 2 section. Since we are temporarily ignoring the S 7 ,
we focus just on the bosonic radial mode ( ). We substitute (2.4) and (2.5) into (2.6) to find

3
3
S2 = T2 RAdS
(2.8)
d
det(h ij + i j ) + q sinh3
4
S 1 S 2
where from (2.7) and (2.4)

cosh2
0
0
2
h ij =
(2.9)
.
0
0
sinh
0
0
sinh2 sin2
We are interested in the limit, where the brane approaches the boundary of AdS4 . Then
the action (2.8) becomes

1
1
3
3
d h e hij i j + (1 3q)e
S2 = T2 RAdS4
4
4
S 1 S 2

1
e3 (1 q) + O e
8
(2.10)
99
where hij is the metric on S 1 S 2 , ds 2 = dt 2 + d 2 + sin2 d 2 . Making the change of

variable
= ln
2
3
2T2 RAdS
4
6
+ 4T22 RAdS
4
4
(2.11)
where is the 3-dimensional scalar field with canonical dimension 1/2, and ignoring O( 2 )
terms, we have

1
R
(1 q) 6
d 3 |h| hij i j + 2 +
.
S2 =
(2.12)
6
2
16
64T22 RAdS
4
S 1 S 2
This is just the bosonic singleton action [3] with its scalar mass terms and g 6 /6! coupling.
Since the scalar curvature of S 1 S 2 equals 2, we recognize the correct R 2 coefficient for
6
. Bearing in
Weyl invariance [3]. The coupling constant is given by g = 45(1 q)/4T22 RAdS
4
3
2
3
1/2
2
mind that T2 = 1/(2) lp and T2 RAdS4 = (2N2 ) /8 , we have g = 360 (1 q)/N2 . As
discussed by Seiberg and Witten, for q > 1, the system is unstable against emission of branes,
which is perfectly possible in a non-supersymmetric theory.
2.3. Unitary and supersymmetric
When q = 1, the fermionic completion of the worldvolume action (2.6) is kappa symmetric.
This demands that the background metric gMN and background 3-form potential AMNP obey
the classical field equations of D = 11 supergravity. We shall now consider this case with the full
is AdS4 S 7 background:
2

2
2
2
d + sinh 2 d 2 + sin2 d 2 cosh2 dt 2 + 4RAdS
= RAdS
d72 .
ds11
(2.13)
4
We denote the seven angles parameterizing the S 7 as m with m = 1, 2, . . . , 7 and d72 =
d m d n gmn . We can no longer limit ourselves to a single radial mode, but the generalization
to include the S 7 modes is straightforward. To keep things simple, we shall continue to ignore
the fermions, however. Note that the WessZumino, or volume term remains unchanged. This
implies that the relation between and is the same as before. Note also that the radius for
the S 7 in the full metric is twice that of AdS4 . With all the above, we have, for ,

1 2
1 ij
3
2
m
n
d |h| h i j + i j gmn + .
S2 =
(2.14)
2
8
S 1 S 2
If we define
a = a
(a = 1, 2, . . . , 8),
where a are the 1-forms parameterizing the unit S 7 and a a = 2 , then we have

R a b
1 ij
3
S2 =
d |h| h i j + ab ,
2
16
(2.15)
(2.16)
S 1 S 2
where = = (+, +, . . . , +) and which has manifest SO(3, 2) conformal symmetry and
SO(8) R-symmetry. It is just the bosonic sector of the OSp(4/8) SCFT [3]. (Incidentally, this
rectifies a previous inability to derive the correct bosonic mass terms for the membrane at the
100
end of the universe, necessary to identify it with the OSp(4/8) supersingleton action [2,8] on
S 1 S 2 . The singleton actions in [36,37] were defined on a Minkowski background and so
required no mass terms.)
3. A probe (3, 0)-brane on the boundary of dS4 AdS7
3.1. The dS background
We will now seek the analogue of the M2-brane that occurs in the (9, 2) M -theory, whose
field theory limit is a supergravity theory with bosonic action [15]

1 2
1
11
A3 F 4 F 4 .
S M = d x g R + F 4
(3.1)
48
12
Note that the sign of the kinetic term of F4 is opposite2 to that of the action (2.1). As discussed
in [26], the sign of the kinetic term is intimately related with the worldvolume signatures that
can occur. For example, if the sign of the kinetic term of F4 were reversed in (3.1) to give
a Lagrangian R F42 /48 + in 9 + 2 dimensions, there would be a membrane solution with
(2 + 1)-dimensional worldvolume, while the action (3.1) with the opposite sign for the F4 kinetic
term has brane solutions with worldvolume signatures (3, 0) and (1, 2), as we shall see below.
The sign of the F4 kinetic term in actions (2.1), (3.1) is determined by supersymmetry.
This theory has a number of FreundRubin-type solutions involving the de Sitter-type spaces,
including (d + 1)-dimensional de Sitter space dSd+1 , (d + 1)-dimensional anti-de Sitter space
AdSd+1 , the (d + 1)-dimensional hyperbolic space Hd+1 , and the two-time de Sitter space
AAdSd+1 which is a generalized de Sitter space given by (a connected component of) the
coset SO(d, 2)/SO(d 1, 2), with signature (d 1, 2) and isometry SO(d, 2). The solutions
are dS4 AdS7 , AAdS4 S 7 , AdS7 dS4 and AAdS7 H 4 .
There are two solutions of M -theory analogous to the M2-brane: a (3, 0)-brane and a (1, 2)brane. In this paper we consider the (3, 0)-brane with Euclidean worldvolume. Its solution is
given by

ds 2 = H 2/3 dx12 + dx22 + dx32 + H 1/3 dt 2 dt 2 + dy42 + + dy92 ,
A123 = H 1 ,
(3.2)
where H is a harmonic function on the transverse space.

For the (3, 0)-brane, the null-cone y 2 = t 2 + t 2 divides the transverse space into two regions,
and there are two distinct brane solutions, in which the transverse coordinate space is restricted
to the region inside or outside the null cone. In the region y 2 > t 2 + t 2 , a natural choice for the
time-dependent harmonic function is
H =1+
L2
,
(y 2 t 2 t 2 )3
(3.3)
which gives a real solution (3.2) for y 2 > t 2 + t 2 . For y 2 < t 2 + t 2 , we take instead
H =1+
(t 2
L2
.
+ t 2 y 2 )3
(3.4)
2 Interestingly enough, this means that M -theory avoids the objections that are present in M-theory to implementing
Hawkings resolution [19] of the cosmological constant problem [20].
101
In either case, the solution has bosonic symmetry ISO(3) SO(6, 2). The geometry of (3.2) near
y 2 = t 2 + t 2 differs in the two cases.
1/3
For t 2 + t 2 > y 2 , let 2 = t 2 + t 2 y 2 . Then near = 0, H 1/3 L2 / 2 . Setting y =
sinh , t = cosh cos , and t = cosh sin , the metric near = 0 takes the form
ds 2 =
2 dW 2
RdS
W2 2
2
2
4
dx
+
dx
+
dx
1
2
3
2
W2
RdS
4
2

2
d cosh2 d 2 + sinh2 d52 ,
+ 4RdS
4
1/6
(3.5)
1/6
where W = L2 2 /2 and again RdS4 = L2 /2 = RAdS7 /2. This is the metric of dS4 AdS7 .
The region t 2 + t 2 > y 2 of the solution (3.2) then interpolates between the flat space R9,2 and
dS4 AdS7 . Similar analysis shows that the region y 2 > t 2 + t 2 of the solution (3.2) interpolates
between the flat space R9,2 and H 4 AAdS7 .
3.2. Unitary but non-supersymmetric
We now wish to repeat the calculation given in Section 2.2 but with a (3, 0)-brane occupying
the S 3 boundary of dS4 and with a 4-form that follows from A123 above. In terms of global
coordinates, we write the dS4 metric and 4-form as

2
2
dsdS
d 2 + cosh2 d32 ,
= RdS
4
4
3
F4 = 3RdS
cosh3 d 3
4
(3.6)
where RdS4 = L2 /2.

Now consider a probe (3, 0) brane whose action given by

1 ij M
1
S2 = T2 d 3
i x j x N gMN (x) +
2
2
q
+ ij k i x M j x N k x P AMNP (x) ,
3!
(3.7)
where i now runs over 1, 2, 3 and we have once again allowed the presence of a non-extremality
parameter q. The overall sign is chosen so that the spatial derivatives are the same for (3, 0) as
for (2, 1). The embedding of the membrane in dS4 is given by
1 = 1,
2 = 2,
3 = 3,
(3.8)
so that its worldvolume occupies the S 3 section of dS4 which is located at ( ). We denote the
three angles parameterizing the unit 3-sphere of the dS4 background as 1 , 2 and 3 , and denote
its metric by hij .
On the S 3 boundary of dS4 where is large the area term is
3
3
T2 RdS

3
4
d
h e + 3e 2e hij i j + O e .
T2 A =
(3.9)
23
S3
Further the A3 can be solved from F4 given above as

3
A3 = RdS
3 sinh + sinh3 3 ,
4
(3.10)
102
so the WessZumino term is

3
d 3 h 3 sinh + sinh3
qT2 A3 = qT2 RdS
4
S3
3
qT2 RdS
4
d 3
23
3

h e + 9e + O e .
(3.11)
S3
Note that since is a time coordinate, its kinetic term enters with the opposite sign to that of the
spatial coordinate in Section 2.2. Let us re-express in terms of a scalar field to eliminate
the e in the above WessZumino term. This can be achieved via
= ln
2
3
2T2 RdS
4
6
12T22 RdS
4 .
4
(3.12)
Note also the different coefficient in the 4 term relative to that in (2.11). This reflects the fact
that S 3 has a scalar curvature different from S 1 S 2 . Ignoring O( 2 ) terms, the brane action is
now

1
R
6
S2 = d 3 h ()2 + 2 (1 q)
(3.13)
.
6
2
16
64T22 RdS
4
Interestingly enough, the above action is also the one (generalized to a curved boundary) proposed in [21,22] for the holographic dual of a de Sitter cosmology in the context of a dS/CFT
correspondence [1517]. However, the relative sign of the coupling is opposite to that in the
AdS/CFT case of Section 2.2 so that the signal for instability is now q < 1.
3.3. Non-unitary and supersymmetric
In closing this section, we consider a probe super (3, 0)-brane with q = 1 on the boundary
of dS4 but now in the full dS4 AdS7 background. So we consider now other modes of the brane
in addition to the radial one. If we write the metric for AdS7 with unit radius as
2
dsAdS
= d 2 cosh2 d 2 + sinh2 d52 = gij d i d j
7
(3.14)

2
2
2
ds11
d 2 + cosh2 d32 + 4 dsAdS
.
= RdS
7
4
(3.15)
then
Then we have

1
2
1
3 2
3
2
2
i
j
S2 = d h () gij + + O
,
2
2
8
(3.16)
S3
where we have used the relation between and given earlier.

The above is not yet the wanted form. For this, we need to define
t = cosh sin ,
y i = sinh i
t = cosh cos ,
(i = 1, 2, . . . , 6),
(3.17)
103
such that
2 = t 2 + t 2 y2,
(3.18)
where y 2 = y i y i . In the above, i are the one-forms parameterizing the unit 5-sphere satisfying
i i = 1.
One can check explicitly that

dt 2 + dt 2 dy i dy i = d 2 2 d 2 cosh2 d 2 + sinh2 d52 .
(3.19)
If we now denote = (t, t , y i ), then, ignoring O( 2 ) terms, we have the action

1
R
S = d 3 h hab a b + ,
2
16
(3.20)
S3
where we have also set 2 2 and therefore the signature is now = (, , +, +, +, +,

+, +). Once again, the above action has the same form as that given in (2.16). It has manifest
SO(4, 1) conformal symmetry and SO(6, 2) R-symmetry.
4. Dualities between M2- and M5-branes in M-, M - and M -theories
4.1. Double holography in M - and M -theories
An interesting feature of the various M2- and M5-branes appearing in M - and M -theories is
that their near horizon geometries frequently coincide [28]. In the (9, 2) spacetime of Section 3.1,
for example, both the (3, 0) M2-brane and the (5, 1) M5-brane tend to dS4 AdS7 and share the
same superconformal group OSp (4/8) whose bosonic symmetry is O(4, 1)O(6, 2). Note that,
in contrast to the usual (10, 1) signature, both factors are conformal groups, and so the system is
doubly holographic [28]. Further if we accept the duality between the bulk M-theory and the
boundary conformal theory, the direct consequence of this is that the boundary 3-dimensional
conformal theory describes the same physics as the boundary 6-dimensional conformal theory.
The spacetime symmetry and R-symmetry are exchanged in the two pictures. As we shall now
demonstrate, however, the requirement that the spacetime may be regarded as the near horizon
geometry of N2 M2-branes or equivalently, under interchange of conformal and R-symmetry,
of N5 M5-branes, imposes the constraint
N2 = 2N52 .
(4.1)
(Although we focus on the concrete example of (3, 0, )-branes and (5, 1, )-branes in M theory, the conclusion N2 = 2N52 is true also for the other examples of double holography.) The
near-horizon geometry for a stack of N2 (3, 0, )-branes is given by (3.6), whereas the nearhorizon geometry for a stack of N5 (5, 1, )-branes is given by
2
2
RAdS

7
2
2
2
2
2
ds52 = RAdS
d
+
cosh
d
+
sinh
d
d 2 + cosh 2 d32 ,
5
7
4
6
5
cosh
sinh

,
F7 = 6RAdS
5
7
RAdS7 = 2L5 ,
where L5 = (N5 )1/3 lp .
(4.2)
104
To make these two geometries identical, we need to identify the two metrics and relate the
4-form F4 and 7-form F7 via F7 = F4 with the denoting the Hodge dual. This can be achieved
provided
RdS4 = RAdS7 /2,
(4.3)
which implies
L2 = 2L5 .
(4.4)
Since L2 = (25 2 N2 )1/6 lp , this gives the relation N2 = 2N52 , as promised.

The above conclusion can be understood as the consequence of the relation between the two
radii, one (denoted as R4 ) is associated with 4-dimensional manifold and the other (denoted
as R7 ) with the 7-dimensional one. If the 11-dimensional metric ds 2 = ds42 + ds72 describes
the near-horizon geometry of either M2- or M5-branes, we always have R4 = R7 /2. If in a
given theory, the M2- and M5-branes have the same near-horizon bulk geometry, it must require
N2 = 2N52 given the relation between R4 and N2 and that between R7 and N5 .
At present, due to our lack of understanding of the theories with more than one time, it is hard
for us to make further checks on this conjectured CFT3 /CFT6 correspondence. So the question
arises whether there exists a possibility of such a correspondence, in certain limit, from the
conventional one-time M-theory. The answer is actually positive thanks to the recent interest in
pp-wave backgrounds.
4.2. pp-waves in M-theory
Let us recall how the pp-wave limit arises in the conventional M-theory. The near-horizon
geometries for M2- and M5-branes are AdS4 S 7 and AdS7 S 4 , respectively. AdS/CFT correspondence tells us that M-theory on the AdS4 S 7 background is dual to a (2, 1)-dimensional
conformal theory while M-theory on AdS7 S 4 is dual to (5, 1)-dimensional conformal theory.
At first sight, these look very different. In global coordinates, the near-horizon geometry for
M2-branes is
2

2
2
d + sinh 2 d22 cosh 2 d 2 + 4RAdS
d72 ,
ds22 = RAdS
4
4
3
cosh sinh2 d d 2 ,
F4 = 3RAdS
4
RAdS4 = L2 /2,
(4.5)
while the near-horizon geometry for M5-branes is

2
2
RAdS
7
2
2
2
2
2
d42 ,
ds52 = RAdS
d
+
+
sinh
cosh
d
5
7
4
6
cosh sinh5 d d 5 ,
F7 = 6RAdS
7
RAdS7 = 2L5 .
(4.6)
In the above L2 and L5 are the radii of S 7 and S 4 and are related to the numbers of M2- and
M5-branes, respectively, as before.
We are now seeking a limit such that the above two geometries have the same isometries
and the unit 2-sphere in AdS4 of AdS4 S 7 can be identified with a unit 2-sphere in the S 4 of
AdS7 S 4 and the unit 5-sphere in the AdS7 with a unit 5-sphere in the S 7 . This is nothing but
105
the pp-wave limit. Following [41,42], the limit for AdS4 S 7 is

+ 2 1 2

1 2
2
2
+
2
2
2
2
2
2
ds2 = L2 2dx dx + dx
+ d + d2 + d + d5 ,
2
4
3 2
F4 = 3 L32 2 dx + d 2 ,
(4.7)
2
and the limit for AdS7 S 4 is

+ 2 1 2

1 2
2
2
+
2
2
2
2
2
2
ds5 = 4L5 2dx dx + dx
+ d + d2 + d + d5 ,
2
4
6 5 +
6
F7 = 32 2L dx d 5 .
(4.8)
5
d22
In the above, the

is the original one in AdS4 while the d52 is the one in AdS7 . With this
in mind, the above two metrics can be identified provided L2 = 2L5 . From F4 = F7 we get the
same relation L2 = 2L5 , which implies N2 = 2N52 . Therefore the conclusion we draw here is
that there is a correspondence between the six-dimensional (N+ , N ) = (2, 0) superconformal
theory and the three-dimensional N = 8 superconformal theory, in the respective pp-wave limits
of the conventional one-time M-theory. The advantage here is that we can make further checks
beyond what we can do in the M case. By the respective AdS/CFT correspondence, this must
imply that the entropy calculated from the respective bulk theory must also agree with each other
in the pp-wave limit. For this purpose, we consider both the near-extremal configuration for black
M2-branes and that for black M5-branes, in terms of their respective Poincar coordinates [43].
For M2-branes,
d s22 =
L22 2f dr 2
4r 2 2f 2
i
i
dt
+
dx
dx
+ L22 d72 ,
e
+
e
4
r2
L22
e2f = 1
r06
,
(2L2 r)3
(4.9)
with i = 1, 2, and for M5-branes,

d s52 =

dr 2
r 2 2f 2
e dt + dx i dx i + 4L25 e2f 2 + L25 d42 ,
2
r
4L5
(4L5 r0 )3
,
(4.10)
r6
with i = 1, 2, . . . , 5.
We now calculate the black hole entropy for the above configurations, respectively, using the
BekensteinHawking entropy formula S = A/4G11 with A the event-horizon area and G11 the
eleven-dimensional Newtons constant.3 For M2-branes, it is not difficult to find that the event
horizon is located at r+ = r02 /(2L2 ) and therefore
e2f = 1

A=
2
4r+
L22

L72 dx 1 dx 2 7 = r04 L32
dx 1 dx 2 7 .
3 The entropy per unit brane volume for both of the two cases was calculated in [44].
(4.11)
106
So we have the entropy

r04 L32
dx 1 dx 2 7 .
S2 =
4G11
(4.12)
For M5-branes, we have from the above r+ = 2 L5 r0 . By the same token, we have the entropy

r05 L35
dx 1 dx 5 4 .
S5 =
(4.13)
4G11
In general, it is easy to see that the entropy S2 is quite different from the entropy S5 .
AdSp+2 can usually be realized as a hyperboloid in R2,(p+1) . The metric in terms of the global
coordinates covers the entire hyperboloid while the one in terms of Poincar coordinates covers
only one half of the hyperboloid. On the one hand, the metric in terms of the Poincar coordinates
appears naturally as the near-horizon limit of certain brane configuration and the corresponding
near-extremal thermodynamical entropy can be calculated while the one in terms of the global
coordinates does not have these properties. On the other hand, the pp-wave limit is more naturally
taken in the AdS metric in terms of the global coordinates. In order to examine whether there is
a possibility that the two entropies can be set to equal, we need to relate these two-coordinate
descriptions for a given AdS-space in the region for which both descriptions are valid. For a
given AdSp+2 , we have
r = RAdS (cosh sin + p+1 sinh ),
2
rx 0 = RAdS
cosh cos ,
2
rx 1 = RAdS
1 sinh ,
..
.
2
p sinh ,
rx p = RAdS
(4.14)
where (x , r) with
= 0, 1, . . . , p are the Poincar coordinates with (, , i ) are the global
coordinates and i i2 = 1.
For the near-extremal M2-branes, we write d72 = d 2 + cos2 d 2 + sin2 d52 and take
= in the aboveformulas (4.14). Taking the pp-wave limit ( 0), , and
x ( ( /2)/ 2) 2 x at r = r+ , we then have
dx 1 dx 2 7 =
9 L82 2 5
d d 2 5 .
23 r04
(4.15)
Then the entropy S2 given above becomes

L92 V9
23 4G11
which is now independent of the extremal parameter r0 and where

V9 = 2 5 d d 2 5 .
S2 = 9
(4.16)
(4.17)
If we define the entropy in the pp-wave limit as S2 = lim0 S2 / 9 , then we have

S2 =
L92 V9
.
23 4G11
(4.18)
107
For the near-extremal M5-branes, we write d42 = d 2 + cos2 d 2 + sin2 d22 and take
now = in (4.14). Taking the pp-wave limit, , and x ( (/2 )/ 2)

2 x at r = r+ , we then have
15/2
dx 1 dx 2 dx 5 4 = 9
2 6 L5
5/2
r0
2 5 d d 2 5 .
(4.19)
Then the entropy S5 becomes

S5 = 9
26 L95 V9
4G11
(4.20)
which is now also independent of the extremal parameter r0 and where V9 is also given by (4.17).
By the same token, we have
S5 =
26 L95 V9
.
4G11
(4.21)
Given our previous discussion about the identification of the two metrics under the respective
pp-wave limit, we can identify the two volume V9 in S2 and S5 . So the two entropies can be
identified provided L92 /23 = 26 L95 which again gives N2 = 2N52 .
Acknowledgements
We have benefited from conversations with Finn Larsen and Jim Liu. J.X.L. acknowledges
the partial support of a grant from the Chinese Academy of Sciences.
Appendix A
A.1. Large brane for arbitrary d: AdS case
It is instructive to rewrite the action (2.6) as [1]

S2 = T2 d 3 |g| + qT2 F4 ,
W
(A.1)
where we use the same symbols g and F4 for the metric and 4-form pullbacks onto the brane.
W is a 4-manifold whose boundary is the brane worldvolume W . The first term in the action is
actually proportional to the area A of the worldvolume M of the brane. Also with F4 given by
the volume form on AdS4 [3234] in (2.5), the second term can be recognized as proportional to
the volume V of M [1,35], so that

3q
S2 = T2 A
(A.2)
V .
RdS4
Following [35], the results of Section 2.2 may now be generalized to an arbitrary (d 1)-brane
occupying the conformal boundary Md of an arbitrary Einstein (d + 1)-dimensional manifold Wd+1 of negative curvature.
We shall adapt the notation and the signature used in [35] for our convenience. The boundary Md has a natural conformal structure but not a natural metric. Let hij be an arbitrary
108
metric on the boundary in its conformal class. Here the i , i = 0, . . . , d 1 are an arbitrary set of local coordinates on the boundary. There is then a unique way [38] to extend
the i to coordinates on Wd+1 near the boundary, adding an additional coordinate that
tends to infinity on the boundary, such that the metric in a neighborhood of the boundary is

1
ds 2 = r02 d 2 + e2 hij d i d j Pij d i d j + O e2
4
(A.3)
where
Pij =
2(d 1)Rij hij R

,
2(d 1)(d 2)
(A.4)
and Rij is the Ricci tensor of Md , which implies

hij Pij =
R
.
2(d 1)
(A.5)
Generalizing (A.1), a probe (d 1)-brane occupying the conformal boundary M() of the
Wd+1 with the above metric at large , the action for a radial mode can be written as
S = Td1 (Ad qdVd+1 /r0 )
(A.6)
where Td1 is the brane tension, Ad is the area of the conformal boundary M() with respect to
the brane induced metric and Vd+1 is the volume of Wd+1 enclosed by the conformal boundary.
The area Ad and the volume Vd+1 are all calculated explicitly in [35]. The action for a radial
mode is also given there:

qd
S = Td1 Ad
Vd+1
r0
2(d4)
2d
Td1 r0d d
(d2)
8
d h((1 q) d2 (d2)
[()2 + 4(d1)
R 2 ] + O( d2 ))
d
2
2
for d > 2,
=
(A.7)
2
T
r
1
2 h((1 q)e2 2[()2 + R] + R + O(e2 ))

0
4
for d = 2.
In the above, we have made the change of variable

=
2
d2
ln +
4
d2
1
R
(d1)(d2)
+ e2 R
for d > 2,
for d = 2,
where is the d-dimensional scalar field with canonical dimension (d 2)/2.
(A.8)
109
We are interested in , i.e., . From the above, we find

qd
Vd+1
S = Td1 Ad
r0
d
2d
T
r
(d2)
8
2
2
d1 0 d d |h|((1 q) d2
+ (d2)
for d > 2,
2 [() + 4(d1) R ])
2d
=
2

T1 r0
4
d 2 |h|((1 q)e2 + 2[()2 + R] R)
for d = 2,
(A.9)
where we recognize the curvature term as that required for Weyl invariance of the action [3]. It is
also easy to check that for d = 3, we reproduce the singleton action (2.12) derived earlier noting
that r0 = RAdS4 .
A.2. Large brane for arbitrary d: dS case
Once again we can write the action (3.7) as

3
S2 = T2 d g + qT2 F4 ,
W
(A.10)
so that once again it takes the form of area minus volume

3q
V .
S = T2 A
RdS4
(A.11)
Our probe (3, 0)-brane is actually wrapped on the 3-sphere at a given (). In the action (A.11) the kinetic term can be essentially viewed as an area integration with the induced
metric while the WessZumino term is proportional to the volume enclosed by the brane. Here
we check this explicitly and derive the effective action for the large brane for arbitrary dimension
of the de Sitter factor of the spacetime. In general, for dSd+1 with metric

ds 2 = r02 d 2 + cosh2 dd2 ,
(A.12)
we have the area

1
d
Ad = r0 d d
det hab cosh2 a b
det hab

r0d
= d d d ed + de(d2) 2e(d2) hab a b + O e(d4) ,
2
and the volume

d+1
d
d d coshd
Vd+1 = r0
(A.13)
r0d+1
2d

d

d ed + de(d2) + O e(d4)
d+1
d (d2)
r0
e
+ O(e(d4) )) for d > 2,
d d ( d1 ed + d2
2d
=
r03 2
for d = 2.
d ( 12 e2 + 2 + O(e2 ))
4
(A.14)
110
In the above the r0 is the radius of dSd and hab is the metric of the unit d-sphere.
If we express the radial mode in terms of d-dimensional scalar field with canonical dimension (d 2)/2 via

4
2
1
ln (d1)(d2)
d2 R for d > 2,
= d2
(A.15)
e2 R
for d = 2,
where R is the curvature of unit d-sphere and R = d(d 1), we then have
d
2(d4)
2d
8
d2
2
2
r0 d d x h( d2
d2 ))
(d2)
for d > 2,
2 [() + 4(d1) R ] + O(
2d
Ad =
2
r0 2 2
d x h(e 2[()2 + R] + R + O(e2 ))
for d = 2,
4
(A.16)
and
d+1
2(d4)
2d
r0
d d x h( d1 d2 + O( d2 ))
d
2
Vd+1 =
r02 2 1 2
d x h( 2 e + O(e2 ))
4
for d > 2,
(A.17)
for d = 2.
Finally, the action for the large radial mode of a probe (d, 0)-brane on the conformal bound2(d4)
ary S d of dSd+1 -dimensional manifold is given as (ignoring the O( d2 ))

qd
Vd+1
S = Td1 Ad
r0
d
2d
T
r
(d2)
8
2
2
d1 0 d d x h((1 q) d2
+ (d2)
for d > 2,
2 [() + 4(d1) R ])
2d
=
2

T1 r0
for d = 2.
d 2 x h((1 q)e2 + 2[()2 + R] R)
4
(A.18)
In analogy with (A.9), we expect this formula to be valid for arbitrary manifold Wd , not
just dS, but note the change of sign of the kinetic and mass terms relative to the AdS case. When
d = 3, we recover (3.13) as expected.
References
[1] M.J. Duff, Supermembranes: The first fifteen weeks, Class. Quantum Grav. 5 (1988) 189.
[2] E. Bergshoeff, M.J. Duff, C.N. Pope, E. Sezgin, Supersymmetric supermembrane vacua and singletons, Phys. Lett.
B 199 (1988) 69.
[3] M.P. Blencowe, M.J. Duff, Supersingletons, Phys. Lett. B 203 (1988) 229.
[4] H. Nicolai, E. Sezgin, Y. Tanii, Conformally invariant supersymmetric field theories on S p S 1 and super p-branes,
Nucl. Phys. B 305 (1988) 483.
[5] E. Bergshoeff, A. Salam, E. Sezgin, Y. Tanii, N = 8 supersingleton quantum field theory, Nucl. Phys. B 305 (1988)
497.
[6] M.P. Blencowe, M.J. Duff, Supermembranes and the signature of spacetime, Nucl. Phys. B 310 (1988) 387.
[7] M.J. Duff, C. Sutton, The membrane at the end of the universe, New Sci. 118 (1988) 67.
[8] E. Bergshoeff, M.J. Duff, C.N. Pope, E. Sezgin, Compactifications of the eleven-dimensional supermembrane, Phys.
Lett. B 224 (1989) 71.
[9] M.J. Duff, Classical and quantum supermembranes, Class. Quantum Grav. 6 (1989) 1577.
[10] E. Bergshoeff, E. Sezgin, Y. Tanii, Stress tensor commutators and Schwinger terms in singleton theories, Int. J.
Mod. Phys. A 5 (1990) 3599.
[11] M.J. Duff, Anti-de Sitter space, branes, singletons, superconformal field theories and all that, Int. J. Mod. Phys.
A 14 (1999) 815, hep-th/9808100.
111
[12] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2 (1998)
231, hep-th/9711200.
[13] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys. Lett.
B 428 (1998) 105, hep-th/9802109.
[14] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[15] C.M. Hull, Timelike T-duality, de Sitter space, large N gauge theories and topological field theory, JHEP 9807
(1998) 021, hep-th/9806146.
[16] A. Strominger, The dS/CFT correspondence, JHEP 0110 (2001) 034, hep-th/0106113.
[17] A. Strominger, Inflation and the dS/CFT correspondence, JHEP 0111 (2001) 049, hep-th/0110087.
[18] D. Adams, The Restaurant at the End of the Universe, Pan Books, London, 1980.
[19] S.W. Hawking, The cosmological constant is probably zero, Phys. Lett. B 134 (1984) 403.
[20] M.J. Duff, The cosmological constant is possibly zero, but the proof is probably wrong, Phys. Lett. B 226 (1989)
36.
[21] F. Larsen, J.P. van der Schaar, R.G. Leigh, de Sitter holography and the cosmic microwave background, hepth/0202127.
[22] E. Halyo, Holographic inflation, hep-th/0203235.
[23] W. Nahm, Supersymmetries and their representations, Nucl. Phys. B 135 (1978) 149.
[24] C.M. Hull, B. Julia, Duality and moduli spaces for time-like reductions, Nucl. Phys. B 534 (1998) 250, hepth/9803239.
[25] C.M. Hull, Duality and the signature of space time, JHEP 9811 (1998) 017, hep-th/9807127.
[26] C.M. Hull, R.R. Khuri, Branes, times and dualities, Nucl. Phys. B 536 (1998) 219, hep-th/9808069.
[27] C.M. Hull, Duality and strings, space and time, hep-th/9911080.
[28] C.M. Hull, R.R. Khuri, Worldvolume theories, holography, duality and time, Nucl. Phys. B 575 (2000) 231, hepth/9911082.
[29] C.M. Hull, de Sitter space in supergravity and M-theory, JHEP 0111 (2001) 012, hep-th/0109213.
[30] C.M. Hull, New gauged N = 8, D = 4 supergravities, hep-th/0204156.
[31] M.J. Duff, K.S. Stelle, Multi-membrane solutions of D = 11 supergravity, Phys. Lett. B 253 (1991) 113.
[32] M.J. Duff, P. van Nieuwenhuizen, Quantum inequivalence of different field representations, Phys. Lett. B 94 (1980)
179.
[33] A. Aurilia, H. Nicolai, P.K. Townsend, Hidden constants: The theta parameter of QCD and the cosmological constant
of N = 8 supergravity, Nucl. Phys. B 176 (1980) 509.
[34] P.G.O. Freund, M.A. Rubin, Dynamics of dimensional reduction, Phys. Lett. B 97 (1980) 233.
[35] N. Seiberg, E. Witten, The D1/D5 system and singular CFT, hep-th/9903224.
[36] P. Claus, R. Kallosh, J. Kumar, P.K. Townsend, A. Van Proeyen, Conformal theory of M2, D3, M5 and D1 + D5
branes, JHEP 9806 (1998) 004, hep-th/9801206.
[37] G. DallAgata, D. Fabbri, C. Fraser, P. Fre, P. Termonia, M. Trigante, OSp(8/4) singleton action from the supermembrane, hep-th/9807115.
[38] C.R. Graham, R. Lee, Einstein metrics with prescribed conformal infinity on the ball, Adv. Math. 87 (1991) 186.
[39] E. Bergshoeff, E. Sezgin, P. Townsend, Supermembranes and eleven-dimensional supergravity, Phys. Lett. B 189
(1987) 75.
[40] E. Bergshoeff, E. Sezgin, P. Townsend, Properties of the eleven-dimensional supermembrane theory, Ann. Phys. 185
(1988) 330.
[41] M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry, Class. Quantum Grav. 19 (2002) L87L95, hep-th/0201081.
[42] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-YangMills,
JHEP 0204 (2002) 013, hep-th/0202021.
[43] M.J. Duff, H. Lu, C.N. Pope, The black branes of M-theory, Phys. Lett. B 382 (1996) 7380, hep-th/9604052.
[44] I. Klebanov, A. Tseytlin, Entropy of near-extremal black p-branes, Nucl. Phys. B 475 (1996) 164178, hepth/9604089.
Proton decay via dimension-six operators in intersecting

D6-brane models
Mirjam Cvetic , Robert Richter
Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA
Received 10 August 2006; accepted 11 October 2006
Available online 20 November 2006
Abstract
We analyze the proton decay via dimension six operators in supersymmetric SU(5)-grand unified models
based on intersecting D6-brane constructions in type IIA string theory orientifolds. We include in addition
10 interactions. We provide a detailed
to 10 1010 10 interactions also the operators arising from 5 510
construction of vertex operators for any massless string excitation arising for arbitrary intersecting D-brane
configurations in type IIA toroidal orientifolds. In particular, we provide explicit string vertex operators
for the 10 and 5 chiral superfields and calculate explicitly the string theory correlation functions for above
operators. In the analysis we chose the most symmetric configurations in order to maximize proton decay
rates for the above dimension six operators and we obtain a small enhancement relative to the field theory
result. After relating the string proton decay rate to field theory computations the string contribution to the
proton lifetime is pST = (0.52.1) 1036 years, which could be up to a factor of three shorter than that
predicted in field theory.
1. Introduction
Grand unified theories (GUTs) [1] not only give a neat and aesthetic description of our fourdimensional world but also lead to an explanation of electric charge quantization andwith the
aid of supersymmetrypredict the value of sin2 W in very good agreement with the experimental one. Moreover GUTs lead to baryon number violating processes; in particular they predict
proton decay [2] (for a recent review on proton decay see [3]).
E-mail addresses: cvetic@cvetic.hep.upenn.edu (M. Cvetic), rrichter@physics.upenn.edu (R. Richter).

doi:10.1016/j.nuclphysb.2006.10.029
M. Cvetic, R. Richter / Nuclear Physics B 762 (2007) 112147
113
In supersymmetric GUT field theories [4,5] the proton decay can occur either by an exchange
of
heavy SUSY particle which corresponds to a decay via the dimension 5 operator
a2 super
d Q3 L or by a super heavy gauge boson exchange.1 The latter corresponds to a decay via

L . In the simplest supersymmetric GUT models, proton
the dimension 6 operator d 4 Q2 Q
decay mediated via dimension 5 operators dominates and recent computations predict a lifetime
for the proton, which is below the present experimental bounds [68], but [9,10]. The fact that
proton decay has not yet been observed, suggests the existence of some mechanism that suppresses or even forbids these dimension 5 operators [10,11], so that after all the proton decay via
dimension six operators [12] is the most dominant one.
In this paper we investigate proton decay via dimension six operators in supersymmetric
GUT models based on intersecting D6-brane constructions on type IIA string theory orientifolds.
More precisely, we compute the string effects on the protons decay into a pion and a positron
(p 0 e+ ) for supersymmetric SU(5)-GUT-like models arising from intersecting D6-brane
constructions. In SU(5)-GUTs there are two different amplitudes that contribute to this proton
10, where 5 and 10 denote the multiplets of the gauge
decay rate: 10 1010 10 and 5 510
group SU(5). For intersecting D6-brane constructions with supersymmetric SU(5)-GUTs [13
15],2 the latter amplitude was computed in [27], by explicitly calculating the string amplitude
contribution to 10 1010 10 operators. However, even after pushing all the parameters to the
limit, in order to maximize the proton decay rate, the string contribution to it is at most compa 10 in the
rable to the field theory one. In this work, we explicitly evaluate the amplitude 5 510
same class of models.
As in [27], instead of performing the calculation in a specific model, we rather use generic
universal features of intersecting D-brane model constructions which are relevant for determining
the proton decay rate. In general, the amplitude is sensitive to the local structure of the intersection and the way the D6-branes are wrapped around the compact space. Assuming that the size
of the compactified volume is bigger than the string size, the latter effects can be neglected and
the computation can be performed for a local D6-brane configuration where we do not need to
worry about the embedding in the compact space. This approach allows us to make predictions
about the proton decay rate in a general class of intersecting D6-brane orientifold models. In
generic models the matter fields 10 and 5 are not located at the same intersection, which leads
10. In this work, in order to maximize the
to an overall suppression of the amplitude 5 510
effect, we assume the most symmetric case that all the matter arises at intersections that are on
top of each other. Therefore, we rather compute an upper bound for the string contribution to
the proton decay rate in these models than determining the complete amplitude which is model
dependent.
This paper is organized as follows. In Section 2 we describe the local setup in which we work
and derive the conditions on the intersecting angles, in order to obtain the matter fields in the
representation 10 and 5 at the intersection, simultaneously. In Section 3 we apply the prescription,
given in Appendix A, to construct the vertex operators for the matter fields. Section 4 is dedicated
to the computation of the string scattering amplitudes, including their normalization. Section 5
states the results of the numerical analysis, while the details can be found in Appendix B. In
1 We forbid proton decay due to dimension four operators by introducing R-symmetry.
2 For a review on intersecting D-brane constructions see, e.g., [16]; for the original work on non-supersymmetric inter-
secting D-branes, see [1720], and chiral supersymmetric ones, see [14,15] and also [21]. For flipped SU(5) constructions
see [22,23]. For supersymmetric SU(5) GUT constructions within type II rational conformal field theories see: [24,25]
and references therein. For a related study on CalabiYau manifolds see [26].
114
Section 6 we relate the string theory results to the four-dimensional field theory and determine
the implication of the string scattering amplitude to the proton lifetime. Finally in Section 7 we
present our conclusions. In Appendix A we give a detailed description, how to construct properly
vertex operators for strings stretched between two intersecting D-branes.
2. Setup
We want to analyze proton decay which occurs due to dimension 6 operators in a local intersecting D6-brane configuration. Therefore, we have to consider scattering amplitudes of the form
10 and 10 1010 10, where 5 and 10 denote the multiplets of the gauge group SU(5).
5 510
While the latter amplitude was already examined in [27], we will determine the additional con 10.
tribution to the proton decay arising from the amplitude 5 510
Since we shall only consider scattering arising at the local intersection, the first step is to
derive conditions on the angles so that we have at the local intersection matter fields in the 5 and
10 representation, simultaneously. We will show that this condition is satisfied only for particular
regions. For the explicit analysis we shall employ the toroidal orientifold construction and take
the size of the tori larger than the inverse string tension, thus suppressing effects due to the
world-sheet instantons. In this limit we shall calculate the four-point string amplitudes for the
chiral superfields at the D6-brane intersections at the origin of the toroidal orientifold. In that
sense the analysis can be applied as the leading order calculation of string amplitudes for the
states at the same D6-brane intersection within any orientifold construction.
Let us briefly review the main properties of intersecting D6-models. Generically, one has a
number of stacks of D6-branes (Ni denotes the number of D-branes for the ith stack), which fill
the four-dimensional Minkowski space and intersect each other in the internal space. Open string
excitations located at the intersections correspond to four-dimensional chiral fermions transforming in the bifundamental representation (Ni , N j ), while open strings starting and ending at the
same stack of D6-branes transform as seven-dimensional U (Ni ) gauge bosons. In order to make
contact with the real world, one has to compactify the six-dimensional internal space which leads
to additional consistency conditions on the model called the RR tadpole conditions. D-branes act
as sources for the RamondRamond (RR)-charges which need to be canceled due to Gauss law
in the internal compact space [20,28]. Typically one introduces Orientifold six (O6-) planes, not
only because they carry negative RR-charge, but also because they can maintain supersymmetry
in the four-dimensional world, while the introduction of anti-D-branes would break all the supersymmetry. The orientifold action leads to image D6 -branes and open strings stretched between
a D6-brane and its image transform as symmetric or anti-symmetric representation of U (Ni ).
As mentioned in the introduction, we rather investigate the proton decay amplitude in a local
D6-brane configuration than in a specific model. In the following we discuss all the necessary
ingredients for this configuration to obtain a supersymmetric SU(5)-GUT like model [13] (for
the non-supersymmetric case see [18,29]).
As explained above the analysis of the D-brane configuration we compactify the internal dimensions are on a factorizable six-torus T 6 . Later we assume that the compactification volume
is larger than the string scale so that local effects dominate the amplitude and global ones can be
neglected. This assumption also allows us to embed the local D-brane configuration, described
below, into an arbitrary compactification manifold.
The complex coordinates of the factorizable six-torus T 6 = T 2 T 2 T 2 are given by
z1 = x 4 + ix 5 ,
z2 = x 6 + ix 7 ,
z1 = x 8 + ix 9 .
115
In order to construct an SU(5) GUT model we shall consider very symmetric configurations of
D6-branes. We take a stack b of M D6-branes oriented in the 0123468 directions that coincides
with a stack a of 5 D6 branes along the 0123 directions and forms (supersymmetric) intersecting
angles with stack b in the internal toroidal directions. The dimensions 0123 have an interpretation
as a (3 + 1)-dimensional intersecting brane world. Both types of D-branes are wrapped on the
(nI , mI ) cycle of the I th torus. Obviously, the wrapping numbers of the stack b are given by

b: n1b , m1b n2b , m2b n3b , m3b = (1, 0)(1, 0)(1, 0),
(2.1)
while the one from stack a can take the general form

a: n1a , m1a n2a , m2a n3a , m3a .
(2.2)
Given the wrapping numbers, one can compute the intersection angles which are in general given
by (R1I , R2I denote the radii of the I th torus)3
I I
I I
m R
ma R2
I
I
I
ab = a b = arctan I I arctan Ib I2
n a R1
n b R1
and in our case take the simple form (since b = 0)
I I
m R
I
ab = arctan Ia I2 .
n a R1
(2.3)
In order to cancel the RR-tadpoles, we must introduce O6-planes and in particular the orientifold
action R, where is the world-sheet parity and R acts by
R : (z1 , z2 , z3 ) (z1 , z 2 , z 3 ).
This orientifold action forces us to include stacks of image D-branes. Since we chose stack b to
lie on top of the orientifold O6-plane, it is invariant under the orientifold action: for M coincident
branes on top of the O6-plane the R projection leads to the gauge group Sp(2M). For the stack
a we have to introduce an image stack a of 5 D6-branes whose wrapping numbers are given by

a : n1a , m1a n2a , m2a n3a , m3a .
(2.4)
Fermions that arise from strings stretched between a and a transform in the anti-symmetric
representation of SU(5) SU(5), due to the fact that the D-branes intersect at the origin of the
torus. Depending on the sign of the intersection number these fermions transform as 10s or
10s. Fermions in the ab and ab sector transform in the bifundamental representation (5, M)
M)4 again depending on the sign of the intersection number. In general, the intersection
or (5,
number for two intersecting D-branes a and b is given by
Iab =
3

nIa mIb mIa nIb .
(2.5)
I =1
Now we have all the ingredients to determine the conditions the intersection angles I have to satisfy in order to observe matter fields transforming as 5 and 10 at the intersection, simultaneously.
3 Note that with this definition clockwise angles are positive and counter-clockwise negative.
4 M denotes the representation of the gauge group Sp(2M).
116
Using (2.1), (2.2), (2.4) and (2.5) we obtain for the intersection numbers Iab and Iaa
Iab = (1)3
3

Iaa = (2)3
mIa ,
I =1
3

nIa mIa .
(2.6)
I =1
Obviously, the sign of the intersection number depends on the sign of the wrapping numbers.
I by where I is given by (2.3)) we have to
For every angle I (from now on we denote ab
I
ab
distinguish between four different cases
nIa , mIa > 0 which corresponds to an angle with 0 < I < 2 ;
nIa > 0, mIa < 0 which corresponds to an angle with 2 < I < 0;
nIa < 0, mIa > 0 which corresponds to an angle with
nIa , mIa < 0 which corresponds to an angle with 2 < I < .
< I < ;
(2.7)
Since we want to analyze proton decay in a supersymmetric GUT model the choice of the intersection angles I is not arbitrary; the sum has to satisfy [30]
1 + 2 + 3 = 0 mod 2.
(2.8)
This requirement restricts the choice of the angles. First we consider the case that the angles add
up to 0 and later on we also analyze the configuration where the sums of the angles are 2 or
2 . If the sum is equal to 0 then one or two of the angles have to be negative. If only one angle
is negative, let us assume without loss of generality that 3 < 0. Since for all angles |I | ,
we distinguish between four different cases for which we obtain, by applying (2.6) and (2.7), the
intersection numbers and in particular their signs
Iab > 0 and Iaa > 0 for 0 < 1 < 2 , 0 < 2 < 2 , 2 < 3 < 0;
Iab > 0 and Iaa < 0 for 0 < 1 < 2 , 0 < 2 < 2 , < 3 < 2 ;
Iab > 0 and Iaa > 0 for
< 1 < , 0 < 2 < 2 , < 3 < 2 ;
Iab > 0 and Iaa > 0 for 0 < 1 < 2 ,
< 2 < , < 3 < 2 .

For all combinations of I s that fulfill the above stated properties ( I = 0 and one angle is
negative) we see that the strings stretched between D-branes a and b transform as 5 instead of
10 at the
Therefore we do not observe a 4-point interaction of the form 5 510
the desired 5.
intersection.
Analyzing the case of two negative angles (without loss of generality we assume that 1 and
2 are negative) we again distinguish between four different cases
Iab < 0 and Iaa < 0 for 2 < 1 < 0, 2 < 2 < 0, 0 < 1 < 2 ;
Iab < 0 and Iaa > 0 for 2 < 1 < 0, 2 < 2 < 0,
Iab < 0 and Iaa < 0 for
< 1 < 2 ,
< 1 < 0,
< 2 < 0,
< 2 < 2 ,
< 3 < ;
< 3 < ;
< 3 < .
117
Only in the region 2 < 1,2 < 0, 2 < 3 < we observe matter fields transforming as 5
and 10, where strings stretched between the D-branes a and b transform as 5 and strings stretched
between a and a transform as 10.
Let us now turn to the case in which the intersection angles I add up to 2 . Then all the
angles are positive and we have to distinguish between three different configurations (without
loss of generality let us assume that 1 is always bigger than 2 )
Iab < 0 and Iaa > 0 for
< 1 < ,
< 2 < , 0 < 3 < 2 ;
< 1 < , 0 < 2 < 2 , 2 < 3 < ;

< 1 < , 2 < 2 < , 2 < 3 < .
Again only in one region, 2 < 1,2,3 < , we observe matter fields transforming as 5 and 10,
where strings stretched between the D-branes a and b transform as 5 and strings stretched between a and a transform as 10.
Finally, we examine the case in which the angles add up to 2 . Here all three angles have
to be negative and again one has to distinguish between three different cases (without loss of
generality we assume that 1 is smaller than 2 )
Iab > 0 and Iaa > 0 for < 1 < 2 , < 2 < 2 , 2 < 3 < 0;
Iab > 0 and Iaa > 0 for < 1 < 2 , 2 < 2 < 0, < 3 < 2 ;
Iab > 0 and Iaa < 0 for < 1 < 2 , < 2 < 2 , < 3 < 2 .
As in the first case, the analysis shows that strings stretched between D-branes a and b transform
as 5 under the U (5) gauge group. Therefore, at the intersection we do not have any matter fields
transforming as 5.
Summarizing, we determined that only for the two regions 2 < 1,2 < 0, 2 < 3 < and
2 < 1,2,3 < we have matter fields transforming as 5 and 10 at the intersection simultaneously.
In addition to the amplitude 10 1010 10, we have for these two regions only, a non-suppressed
10 to the proton decay rate. In order to compute these two amplitudes
contribution from 5 510
and 10 in the respective configurawe need the corresponding vertex operators to the states 5,
tions, which we determine in the next section.
3. Vertex operators
For different D-brane configurations we have different vacua and therefore different vertex
operators. Knowing the D-brane configuration we can use the prescription given in Appendix A
to obtain the vertex operator for the massless fermion in the R-sector. In this way we can easily
arising from strings stretched between the stacks a and b.
determine the vertex operators for 5,
The vertex operator for 10 requires more effort. The simple approach just to replace the I in
the 5 vertex operator by the double, 2I only works for |I | < 12 ,5 since in the expansion of the
bosonic (A.2) and fermionic degrees of freedom (A.3) the shift number I has to be in the interval
[1, 1]. Therefore if I > 12 we need to find an expression I which lies between 0 and 1 and
describes the D-brane configuration aa . Fig. 1 which shows the D-brane configuration for the
5 From now on we replace by / so that [1, 1].
I
I
I
118
Fig. 1. Intersection angles for the case 12 < 1 < 0, 12 < 2 < 0, 12 < 3 < 1.
case 12 < 1,2 < 0, 12 < 3 < 1. The vertex operator in the ( 12 )-ghost picture for the massless
fermion, arising from a string stretched between D-branes a and b is given by (keep in mind that
1,2 are negative)
1
2

V5 2 (z) = 5 e 2 (z) S (z)
I (z)ei(I + 2 )HI (z) 13 (z)ei(3 2 )H3 (z) eikX(z) .
(3.1)
I =1
Now we turn to the aa sector in which the string state transforms as 10. We see that the intersection angle in the third complex dimension is given by 3 = 2 + 23 . Note that the intersection
angle 3 is negative and lies between 1 and 0, since 3 takes a value between 12 and 1 and
therefore the corresponding vertex operator for the state 10 takes the form
1
V10 2 (z) = 10 e 2 (z) S (z)
3

1+I (z)ei(I + 2 )HI (z) eikX(z) ,
(3.2)
I =1
where the angles I are given by

1 = 21 ,
2 = 22 ,
3 = 2 + 23 .
Notice, that the angles I add up to 2 so that the SUSY condition (2.8) is satisfied. In an
analogous way (look at Fig. 2), we obtain for the other D-brane configuration
1
2
< 1 < 1,
1
2
< 2 < 1,
1
2
< 3 < 1
For this configuration the vertex operator that creates a string stretched between a and b is
1
V5 2 (z) = 5 e 2 (z) S (z)
3

1I (z)ei(I 2 )HI (z) eikX(z) .
(3.3)
I =1
The intersection angles I are given by

1 = 2 + 21 ,
2 = 2 + 22 ,
3 = 2 + 23 .
Obviously, they are all negative, so that the vertex operator which describes the massless aa string in the R-sector takes the form
1
V10 2 (z) = 10 e 2 (z) S (z)
3

I =1
Again the angles I add up to 2.
1+I (z)ei(I + 2 )HI (z) eikX(z) .
(3.4)
119
Fig. 2. Intersection angles for the case 12 < 1 < 1, 12 < 2 < 1, 12 < 3 < 1.
In order to calculate scattering amplitudes we also need the vertex operators for 5 and 10 . We
obtain them by replacing the spin field by the spin field with opposite chirality and at the same
time sending the angles I and I to 1 I and 1 I , respectively (for negative angle we replace
I and I by 1 I and 1 I , respectively). For these two cases we obtain
12 < 1 < 0, 12 < 2 < 0,
1
V5 2 (z) = 5 e 2 (z) S (z)
1
2
< 3 < 1
2

1+I (z)ei(I + 2 )HI (z) 3 (z)ei(3 2 )H3 (z) eikX(z)
(3.5)
I =1
for 5 and
1
V102 (z) = 10 e 2 (z) S (z)
3

I (z)ei(I + 2 )HI (z) eikX(z)
(3.6)
I =1
for 10 .
1
2
< 1 < 1,
1
1
2
< 2 < 1,
1
2
< 3 < 1
V5 2 (z) = 5 e 2 (z) S (z)
3

I (z)ei(I 2 )HI (z) eikX(z)
(3.7)
I =1
for 5 and
1
V102 (z) = 10 e 2 (z) S (z)
3

I (z)ei(I + 2 )HI (z) eikX(z)
(3.8)
I =1
for 10 .
Finally, we will discuss the ChanPaton factors. In a setup without orientifolds strings transform in the bifundamental of U (N ) U (M). As already mentioned above, the introduction of
orientifolds changes the transformation behavior. The full orientifold action on the ChanPaton
120
factors takes the form

1
,
= R T R
where R is given by [31]
0 1N
0
1
0
N 0
R =
0
0
0
0
0 1M
.
1M
0
(3.9)
The choice of N = 5 leads to the following ChanPaton factors for the 10s
0 10 0 0
T
0 0 0
10 = 10
,
0
0 0 0
0
(3.10)
where 10 is an anti-symmetric 55 matrix. For M we choose 1 that leads to a Sp(2) gauge group
on the D-brane b which has two components in the fundamental representation. One component
is associated with the matter field 5 while the other corresponds to the Higgs particle. Their
ChanPaton factors take the form
0
0 0 0
0
0 0 0
0
0
0 5 0
0 0 H
5 =
(3.11)
H =
,
.
0
H T 0 0 0
0 0 0
T5
Here 5 and H are a 5 1 matrices. 10 and 5 denote the usual 10- and 5-dimensional representations of the SU(5) gauge group and H is the 5-dimensional Higgs field in the gauge field
theory.
4. String amplitude
Having derived the vertex operators in the previous section, we have all the ingredients to
compute the scattering amplitudes. Assuming that the compactification volume is larger than
the string scale world-sheet instantons are suppressed and it is sufficient to compute just the

quantum part of the amplitudes. First we will focus on V 5 1 V 5 1 V 101 V 101 and afterwards we
will compute
V 101 V 101 V 101 V 101 ,

2
2 2
2

2
2
which was already examined in [27].
4.1. The amplitude V 5 1 V 5 1 V 101 V 101

2
We start with the region 12 < 1 < 0, 12 < 2 < 0,

4
i=1

dzi V5 1 (z1 )V5 1 (z2 )V101 (z3 )V101 (z4 ) ,

2
1
2
< 3 < 1 and calculate the amplitude
121
where the vertex operators are in the previous section. Note that all the vertex operators are in the
( 12 )-ghost pictures, which guarantees a total ghost charge of 2 on the disk. Plugging in the
vertex operators we see that in order to calculate the amplitude we need the following correlators
4

4

iki X(zi )
e
zij ki kj ,
=
i,j =1
i<j
i=1
4

1
e 2 (z1 ) e 2 (z2 ) e 2 (z3 ) e 2 (z4 ) =
zij 4 ,
i,j =1
i<j

1 1
u 1 u 2 u 3 u4 S (z1 )S (z2 )S (z3 )S (z4 ) = u 1 u2 u 3 u4 z132 z242 ,
(4.1)
where zij denotes zi zj . The correlator involving the four fermionic twist fields takes an easy
form, since we can bosonize the spin fields
4

4

ii H (zi )
e
zij i j .
(4.2)
=
i,j =1
i<j
i=1
The correlator for the bosonic twist fields is more involved. Using the stress energy tensor
method, the quantum part of four bosonic twist fields with two independent angles evaluates
to [32,33]

1 (+)
2

1
(1) (1) z13 z24
z34
I 2 (x),
1 (z1 ) (z2 )1 (z3 ) (z4 ) = z12
(4.3)
z14 z23
with x =
z12 z34
z13 z24
I (x) =
and I (x) is given by

1
B1 (, )G2 (x)H1 (1 x) + B2 (, )G1 (x)H2 (1 x) ,
2
where
( )(1 )
,
(1 + )
G1 (x) = 2 F1 [, 1 , 1; x],
B1 (, ) =
()(1 )
,
(1 + )
G2 (x) = 2 F1 [1 , , 1; x],
B2 (, ) =
H1 (x) = 2 F1 [, 1 , 1 + ; x],
H2 (x) = 2 F1 [1 , , 1 + ; x].
Applying the correlators, the amplitude becomes

4

5 5 10 10
4 (4)
A = i Tr 1 2 3 4 u 1 u2 u 3 u4 (2)
ki

4
i=1
1
dzi
i=1
[I (1 , 1 + 1 , x)I (2 , 1 + 2 , x)I (1 3 , 1 + 3 , x)] 2

,

(z12 z34 ) s+1 (z13 z24 ) t (z14 z23 ) u+1
where s, t and u are the Mandelstam variables

s = (k1 + k2 )2 ,
t = (k1 + k3 )2 ,
u = (k1 + k4 )2 .
122
The conformal Killing group can be used to fix three of the vertex operator positions. A convenient choice is
z1 = 0,
z2 = x,
z3 = 1,
z4 = z = ,
which implies the c-ghost contribution

2
c(0)c(1)c(z ) = z
.
After fixing three positions, we are left with an integral over one world-sheet variable
4

5 5 10 10
4 (4)
A = iCA Tr 1 2 3 4 u 1 u2 u 3 u4 (2)
ki
i=1
1
dx
0
[I (1 , 1 + 1 , x)I (2 , 1 + 2 , x)I (1 3 , 1 + 3 , x)] 2

.

x s+1 (1 x) u+1
In order to obtain the full amplitude we need to sum over all possible orderings

10
10
Atotal = C Tr 51 52 10
+ Tr 52 51 10
4 3
3 4
0
dx U (x)

5 5 10 10
10
C Tr 51 52 10
3 4 + Tr 2 1 4 3
1
dx U (x)
0

5 10 5 10
5 10
C Tr 51 10
3 2 4 + Tr 1 4 2 3

dx U (x),
1
with

C = iCA u 1 u2 u 3 u4 (2)
4 (4)
4

ki
i=1
and
1
[I (1 , 1 + 1 , x)I (2 , 1 + 2 , x)I (1 3 , 1 + 3 , x)] 2

U (x) =
.

x s+1 (1 x) u+1
Calculating the traces for the third term by plugging in the respective ChanPaton factors immediately shows that they vanish. Explicit computation of the traces leads to the identities
5 5
5
5 10 10
5 5
10
10 10
10
10
Tr(51 52 10
3 4 ) = Tr(2 1 4 3 ) and Tr(1 2 4 3 ) = Tr(2 1 3 4 )
and thus the amplitude takes the form
4

4 (4)
Atotal = 2iCA u 1 u2 u 3 u4 (2)
ki
i=1

5 5 10 10

10
Tr 51 52 10
T (1 , 2 , 3 ) , (4.4)
3 4 K(1 , 2 , 3 ) + Tr 1 2 4 3
123
with
1
K(1 , 2 , 3 ) =
0
T (1 , 2 , 3 ) =
dx U (x),
(4.5)
dx U (x).
In the field theory, the first term corresponds to proton decay via a gauge boson, while the second
one describes the proton decay mediated via a Higgs particle, arising from the Yukawa interaction
105 5 H .
Finally we replace the s by the angles
1 = 21 ,
2 = 22 ,
3 = 2 + 23
and obtain for U

1
[I (1 , 1 + 21 , x)I (2 , 1 + 22 , x)I (1 3 , 1 + 23 , x)] 2

.
U (x) =

x s+1 (1 x) u+1
(4.6)
Applying the same procedure for the other sector we obtain
1
2
< 1 < 1,
1
2
1
2
< 2 < 1,
< 3 < 1
The amplitude

4
i=1

dzi V5 1 (z1 )V5 1 (z2 )V101 (z3 )V101 (z4 ) ,

2
takes the form
Atotal = 2iCA u 1 u2 u 3 u4 (2)
4 (4)
4

ki
i=1
5 5 10 10

10
T (1 , 2 , 3 ) , (4.7)
Tr 51 52 10
3 4 K(1 , 2 , 3 ) + Tr 1 2 4 3
with K and T defined in (4.5) and U given by

U (x) = x s1 (1 x) u1
3

1
I (1 I , 1 + 2I , x) 2 .
(4.8)
I =1
4.2. The amplitude V 101 V 101 V 101 V 101

2
Note that in both cases, 12 < 1,2 < 1, 12 < 3 < 0 and 12 < 1,2,3 < 1, the vertex operators for
the matter fields transforming as 10 take the same form. Thus the computation of the amplitude
V 101 V 101 V 101 V 101 is identical for both cases. We use the same correlators stated above except
2
for the one involving the bosonic twist fields, which takes a simpler form, since it involves only
one independent angle [32]

(1)

1
z13 z24
1 (z1 ) (z2 )1 (z3 ) (z4 ) =
(4.9)
L 2 (x)
z12 z14 z23 z34
124
with
L(x) =
1
2 F1 [, 1 , 1, x] 2 F1 [, 1 , 1, 1 x].
sin( )
Plugging in all the correlators and fixing three vertex operator positions we obtain
4

10 10 10 10
10 10 10 10

4 (4)
ki
Atotal = iCA Tr 1 2 3 4 + Tr 1 4 3 2 (2)
i=1
1
u 1 u2 u 3 u4
3

dx x s1 (1 x) u1
L 2 (1 + I , x).
I =1
Finally we replace the I by I and obtain

12 < 1 < 0, 12 < 2 < 0,
1
2
< 3 < 1

10 10 10
10 10 10 10
Atotal = iCA Tr 10
1 2 3 4 + 1 4 3 2
4

4 (4)
ki u 1 u2 u 3 u4 M(1 , 2 , 3 )
(2)
(4.10)
i=1
with
1
M(1 , 2 , 3 ) =
dx
0
1
2
< 1 < 1,
1
2
x s1 (1 x) u1
1
L 2 (1 + 21 , x)L 2 (1 + 22 , x)L 2 (1 + 23 , x)
< 2 < 1,
1
2
(4.11)
< 3 < 1

10 10 10
10 10 10 10
Atotal = iCA Tr 10
1 2 3 4 + 1 4 3 2
4

4 (4)
ki u 1 u2 u 3 u4 M(1 , 2 , 3 )
(2)
(4.12)
i=1
with
1
M(1 , 2 , 3 ) =
dx
0
x s1 (1 x) u1
1
L 2 (21 1, x)L 2 (22 1, x)L 2 (23 1, x)
(4.13)
The V 101 V 101 V 101 V 101 does not involve an Higgs exchange, since couplings of the form
2
10105H are absent due to the U (1) charge conversation [13].

4.3. Normalization
In this section we determine the two undetermined constants CA and CA in the string amplitudes computed above. We will use the fact that even in the low energy limit the integrals
(4.5), (4.11) and (4.13) are convergent in the limit x 0, which corresponds to a gauge boson
125
exchange. Factorizing the amplitude into two three point functions allows us to normalize it. We

start with the amplitude V 5 1 V 5 1 V 101 V 101 and turn later to V 101 V 101 V 101 V 101 .
2

2
2
4.4. The amplitude V 5 1 V 5 1 V 101 V 101

2
We first examine the limit x 0 and will see that even in the low energy limit the integral is
convergent, due to the special kinematics of this problem.
Limit x 0
As x 0 the hypergeometric functions behave like

(a + b)
(a, b)
F (a, b, a + b, 1 x)
ln
,
(a)(b)
x
F (a, b, 1, x) 1,
(4.14)
with
ln (a, b) = 2(1) (a) (b).
Applying (4.14) I takes the form

(, )
1
lim I (, , x) = ln
,
x0
x
where ln (, ) is given by
1
1
1
1
ln (, ) = 2(1) ( ) (1 ) () (1 ).
2
2
2
2
Therefore even for s = t = 0 we obtain for the integral (4.4) a convergent expression in the limit
x0

dx
3/2
(4.15)
ln[1/x]3/2 .
x
0
That allows us to normalize the amplitude by factorizing the amplitude in the limit x 0, where
it reduces to a product of two three-point functions
I
I
7 7

d kd k
i
I J Aj (k1 , k2 , k)Aj (k3 , k4 , k )(k k )
.
A4 (k1 , k2 , k3 , k4 ) =
(4.16)
2
(2)7
k 2 i
The unusual factor of 12 is introduced to take into account the doubling in the ChanPaton factors.
The three-point amplitudes describe the exchange of a gauge boson and are given by
3

A (k1 , k2 , k3 ) = igD6 (2)4 (4)
(4.17)
ki u 1 u2 Tr 51 52 A .
i=1
Here corresponds to the polarization and A denote the ChanPaton factors of the gauge
boson. The latter takes the form
a 0 0 0
0 0 0
a
A =
(4.18)
,
0
0 0 0
0
0 0
126
where the a s are the gauge bosons of U (5) which satisfy Tr(a b ) = 12 ab . The intermediate
state is a massless aa string, which is a gauge boson, that can carry arbitrary momentum p
along the directions of the D-brane a orthogonal to the intersection. In these directions we have
to integrate over

3/2
2

d3 q dx x q s1 = 3/2 ( )3/2 dx x s1 ln(1/x)
0
which tells us that the replacement, going from effective field theory in four dimensions to the
form of the string integrand near x = 0 is no longer

1

dx x s1 ,
s
0
but

d3 q
3/2 ( )1/2
q2 s

3/2

dx x s1 ln(1/x)
.
(4.19)
Performing the integral on the right-hand side of (4.16) and using the replacement (4.19) we
obtain
4
2 5/2

gD

3/2
5 5 10 10

6
(4)
i
Tr 1 2 3 4 u 1 u2 u 3 u4
ki
dx x s1 ln(1/x)
.

1/2
2
i=1
0
(4.20)
This needs to be the same as (4.4) in the limit x 0

10
2iCA Tr 51 52 10
1 u2 u 3 u4
3 4 u
4

3/2
4 3/2 (4)
(2)
ki
dx x s1 ln(1/x)
,
i=1
2 = (2)4 3/2 g to the normalization constant C

which leads us with gD
s
A
6
CA =
gs .
2
(4.21)
For the second amplitude one obtains, following the same procedure, the same normalization
constant.
4.5. The amplitude V 101 V 101 V 101 V 101

2
Note that the amplitude is invariant under the exchange of x and 1 x if one simultaneously
interchanges s and u. Therefore we obtain similar limits for x 0 and x 1. That is not too
surprising taking into account that we expect an exchange of a gauge boson in both limits.
127
Limit x 0 and x 1
Using (4.14) and taking the low energy limit s, t 0 we get for x 0

dx
3/2
ln[1/x]3/2
x
(4.22)
and a similar result for x 1

1
3/2

3/2
dx
.
ln 1/(1 x)
1x
(4.23)

2
2
Following the same procedure as in the case of the amplitude V 5 1 V 5 1 V 101 V 101 we obtain
for normalization constant CA
CA = gs .
(4.24)
5. Numerical analysis
We want to compute the contribution of the amplitude which arises from the four-Fermi interaction in the low energy effective theory. That means that we take the low energy limit and
subtract the s, t and u poles, if present. It turns out that the amplitudes are divergent only in
the limit x . As derived in Appendix B there is no massless exchange in the u-channel.
The s-channel requires more explanation, since in general we expect a massless gauge boson
exchange, which leads to an undesired s-pole. We saw that the integral does not diverge at the
s-pole, since we neglected global effects coming from the internal space. Locally, the internal
dimensions look like a flat space with infinite volume which leads to a vanishing gauge coupling
in four dimensions
2
gYM
1
,
Vint
(5.1)
here Vint denotes the internal volume and gYM is the gauge coupling in four dimensions. Thus
even if we observe a gauge boson exchange, we do not see an s-pole in our effective low energy
theory. In the limit x , which corresponds to a t -pole, the integral is divergent and in order
to obtain the four-Fermi interaction we have to subtract this pole. A detailed discussion of the
numerical analysis of the integrals K, T and M in the amplitudes (4.4), (4.7), (4.10) and (4.12)
can be found in Appendix B, where for simplification we set 1 = 2 = . Table 1 shows the
contribution M for the string amplitude V 101 V 101 V 101 V 101 and the contributions K and T

2
2
arise from V 5 1 V 5 1 V 101 V 101 for different angles . For = 1/3 and = 2/3 we observe
2
a second massless fermion which indicates that we now have N = 2 supersymmetry. Since our
world is chiral we choose in the ranges, given in Table 1.
Note also, that going from the first sector 12 < 1 < 0, 12 < 2 < 0, 12 < 3 < 1 to the
second one 12 < 1 < 1, 12 < 2 < 1, 12 < 3 < 1 and replacing by 1 , simultaneously leads to
the same results for K, T and M. This is not too surprising, since the respective vertex operators
correspond to the same states if you interchange with 1 .
128
Table 1
Contribution to K, T and M for different angles
12 < 1 < 0,
K
12 < 2 < 0,
T
1 < <1
3
2
1 < < 1,
1
2
1 < < 1,
2
2
1 < <1
3
2
0.40
0.42
0.44
6.5
5.7
4.9
5.4
5.1
4.6
10.3
9.4
8.3
0.505
0.51
0.52
1.5
2.0
2.9
1.5
2.1
2.9
M
2.5
3.5
4.9
0.46
0.48
0.49
0.495
4.0
2.9
2.0
1.5
4.0
2.9
2.1
1.5
6.9
4.9
3.5
2.5
0.54
0.56
0.58
0.60
4.0
4.9
5.7
6.5
4.0
4.6
5.1
5.4
6.9
8.3
9.4
10.3
6. Comparison to four-dimensional field theory

In this section we want to compare the amplitude obtained due to massive string states in string
theory with the amplitude on the field theory side. Therefore, we would like to replace all the
string theory parameters such as the string coupling gs or the gauge coupling gD6 by appropriate
expressions using quantities about which we have some knowledge of, such as MGUT and GUT .
We follow closely the analysis of [27].
The action for the gauge fields living on the D6-branes is

1
d7 x Tr Fij F ij ,
2
4gD
6
where the Fij is the YangMills field strength and Tr denotes the trace in the fundamental representation of U (N ). After compactification on R Q the action becomes

VQ
d4 x Tr Fij F ij ,
2
4gD
6
where VQ is the volume of Q. Keeping in mind the usual convention Tr(Qa Qb ) = 12 ab we
finally obtain for the action

VQ
d4 x Tr Fij F ij .
(6.1)
2
8gD
6
On the other hand, the GUT action is given by

1
d4 x Tr Fij F ij ,
2
4gGUT
(6.2)
2 = (2)4 g 3/2
where gGUT is the GUT coupling. Comparing (6.1) and (6.2), along with gD
s
6
2
[34] and GUT = gGUT
/(4), leads to the identification

GUT VQ 2/3

=
(6.3)
.
(2)3 gs
The volume VQ enters into the running of the SU(3) SU(2) U (1) gauge coupling from high
1/3
energies to low energies. Approximately, one can say that VQ
plays the role of the mass scale
unification MGUT in four dimensions. In order to obtain the exact relation between them one
needs to compute the one loop threshold correction to the gauge coupling, which was done for
129
M-theory on a manifold of G2 holonomy [35]6

VQ =
L(Q)
3
MGUT
(6.4)
where L(Q) is a topological invariant, the RaySinger torsion. In [27] it is argued that this
relation holds true in type IIA string theory and thus we finally obtain

GUT L(Q)
2/3
3
(2)3 gs MGUT
(6.5)
We would like to replace all the string parameters in the amplitudes (4.4) and (4.7) in terms of
four-dimensional field theory quantities. Unfortunately, Eq. (6.5) still includes two string parameters L(Q) and gs . The RaySinger torsion L(Q) depends crucially on the compact space and
takes for simple lens spaces values around 8 [35]. In order to neglect higher order loop amplitudes the string coupling gs is better smaller than 1. On the other hand we are interested in the
largest possible contribution to the enhancement and set therefore gs approximately to 1.
6.1. Field theory amplitude
After relating the string parameters to four-dimensional field theory constants, of which we
have some experimental knowledge, we now recall the analysis of proton decay in the SU(5)
GUT model.7 This treatment closely follows [2].
The kinetic energy for an SU(5) gauge theory, involving the gauge field A, the fermionic field
and the fermionic field 10 transforming as 10 under the SU(5) takes
5 , which transforms as 5,
the form
T=
1
2
4gGUT

Tr F 2 (A) + i 5 D 5 + i 10 D 10
(6.6)
with
igGUT
D 5 a = 5 a (A )ab 5 b
2
and
igGUT
igGUT
D 10 ab = 10 ab (A )ac 10 cb (A )bd 10 ad .
2
2
By explicitly using the anti-symmetry of 10 , the latter can be simplified to
D 10 ab = 10 ab
i2gGUT
(A )ac 10 cb .
2
6 An explicit computation for the one loop threshold correction in type IIA string theory was performed in [36], which
leads in the limit gs 1 to an equivalent relation.
7 As done usually we neglect because of the weakness of the Yukawa couplings to light fermions the Higgs mediated
proton decay.
130
The gauge field A can be displayed as a 5 5 matrix
| X1C

a a
1
| X2C
a G
| X3C
A =
W3
X1
X2
X3
Y1
Y2
Y3 | W
B
,
2
+
30
3
3
Y1C
Y2C
Y3C
+
W
W3
where the a are the GellMann matrices, the Ga denote the gluon fields of SU(3) and W+ ,
W , W3 , B are the bosons of the SU(2) U (1). The X and Y are the new gauge bosons that
are contained in SU(5) and do not occur in the standard model. The exchange of these new gauge
bosons leads to baryonlepton number violating processes and therefore allows proton decay.
To make contact to the standard model the SU(5) needs to be broken, which will be achieved
by giving the Higgs field, which transforms under the 24-dimensional adjoint representation of
SU(5) an expectation value. This generates a mass MX of order of 1016 GeV for the gauge
bosons X and Y .
From (6.6) one can easily deduce the effective four-Fermi interactions which lead to proton
decay. Ignoring mixing effects as well as second and third families one obtains for the
Leff =
2

gGUT
2MX2

C
+
+
u L uL 2eL
dL + eR
dR ,
(6.7)
10
where the first factor arises from a 10 1010 10 interaction and the second factor from a 5 510
interaction.
6.2. Comparison
This result (6.7) we want to compare with the string theory contribution. In order to do that we
turn on Wilson lines, that break the SU(5) gauge group into the standard model ones. Assuming
such a mechanism of symmetry breaking exist we compute the traces of (4.4) and (4.10) only for
entries which lead to proton decay. One obtains for (4.4) and (4.7)
4

C
+
4 (4)
5 51010
Atotal = i(2)
(6.8)
ki gs u L uL eR
dR K( ) + T ()
i=1
and for (4.10) and (4.12)

A10101010
total
= 2i(2)
4 (4)
4

i=1

C
+
ki gs u L uL eL
dL M().
(6.9)
131
Comparing the string theory proton decay rate with the one from four-dimensional gauge theory
one obtains
1/3

gs L(Q)2/3 2 MX 4 (K + T )2 + 4M 2
ST (p 0 e+ )
=
.
(6.10)
1/3
MGUT
4
FT (p 0 e+ )
8 2 GUT
Most recent calculations [8] for the proton decay mediated via gauge bosons in an SU(5)-GUT
model gave the lifetime pFT in terms of gauge boson mass MX and GUT
pFT
= 1.6 10
36

4
0.04 2
MX
years
.
GUT
2016 GeV
(6.11)
This leads with the values MX = MGUT = 2 1016 GeV and GUT = 0.04 to a proton lifetime
of 1.6 1036 years. The present lower bound on the proton lifetime for p 0 e+ is 1.6
1033 years [37] and even the next generation proton decay experiments, based on underground
water Cherenkov detectors will reach a lower bound not larger than 1035 years [38]. Therefore
in the near future, unless there is an enhancement to the proton decay amplitude, we will not
observe the proton decay via gauge boson exchange. Using (6.10) and (6.11) the proton lifetime
in the considered type IIA string models is

4

542
MGUT
0.04 4/3
ST
36
p 1.6 10 years
,
2/3
2016 GeV
L4/3 (Q)gs ((K + T )2 + 4M 2 ) GUT
(6.12)
2/3
) +4M )
is the string enhancement factor. Note that in (6.12) the heavy
where L (Q)gs ((K+T
542
gauge boson mass MX , which is model dependent, is absent and the proton lifetime depends
only on MGUT . We also observe an anomalous power of GUT in (6.12) indicating the stringy
nature of the enhancement.
2/3
4/3
)2 +4M 2 )
. As already mentioned earLet us examine the enhancement factor L (Q)gs ((K+T
542
lier the RaySinger torsion is around 8 for lens spaces with small fundamental group. The string
coupling takes values between 0 and 1, but in order to obtain the largest possible enhancement to the proton decay amplitude we assume it is approximately 1. Table 1 shows that M
ranges between 5 and 10, while K + T 1.2 M, leading with the numerical four-dimensional
SU(5) supersymmetric values MGUT = 2 1016 GeV and GUT = 0.04 to a proton lifetime
pST = (0.52.1) 1036 years. We see that although there is in addition to the contribution to
the four-Fermi interaction which in field theory are due to gauge boson exchange, there is also a
contribution due to terms that in field theory arise from Higgs particle exchange, the total string
contribution is not large enough to lead to a considerable enhancement in the proton decay rate.
10 have in contrast to the operators 10 1010 10 a second
The dimension six operators 5 510
proton decay mode; they lead in addition to the decay mode p 0 e+ also to p + .
Plugging in the respective entries in (4.4) leading to the mode p + one obtains
4

C

4 (4)
5 51010
Atotal = i(2)
(6.13)
ki gs u L dL RC dR K( ) + T () .
4/3
i=1
Within the field theory the effective interaction

Leff =
2

gGUT
2MX2

C

u L dL RC dR ,
(6.14)
132
the ratio between the proton decay rates is given by

ST (p + )
=
FT (p + )
1/3
gs L(Q)2/3
1/3
16 2 GUT
2
MX
MGUT
4
(K + T )2 .
(6.15)
For this decay mode the string enhancement to the proton decay rate is even smaller than for the
mode p 0 e+ due to the absence of the 10 1010 10 interaction term. For the same choice
of parameter as above (in addition we assume that MX = MGUT ) the ratio (6.15) takes values
between 0.2 and 0.8.
7. Conclusions
In this paper we computed the local, string contribution to the proton decay rate for supersymmetric SU(5) GUTs based on intersecting D6-brane constructions in type IIA string theory
orientifolds by explicitly calculating the string amplitude contribution to the dimension six operators. If the compactification volume is larger than the string scale, world-sheet instanton effects
are negligible and the local contribution is the dominant one. In the computation presented, we
assumed that the matter fields 5 and 10 are located at the same intersections on top of each other,
and thus the leading string amplitude contributions have no suppressions from area factors. In
this case the amplitudes give the largest possible contribution to the proton decay rate. In contrast to the authors [27], who only considered the amplitude 10 1010 10, we also included the
10 operators.
explicit calculation of the string amplitude for 5 510
As a by-product we explicitly constructed the vertex operators for any massless string excitation at supersymmetric D-brane intersections arising in type IIA toroidal orientifolds. Specifically, by employing explicit string vertex operators for the 10 and 5 chiral superfields, we
calculated explicitly string theory amplitudes contributing to the proton decay via dimension
six operators. In the analysis we chose the most symmetric configurations in order to maximize
proton decay rates for the above dimension six operators and we obtain a small enhancement relative to the field theory result. In contrast to the string amplitude 10 1010 10, where only the
10 there is
gauge boson exchange contributes to the proton decay rate for the amplitude 5 510
an additional contribution corresponding to the proton decay mediated via Higgs particle.
After relating the string theory result to the field theory computations we obtain for the proton
lifetime in type IIA string theory models
pST 1.6 1036 years
542
0.04
2/3
L4/3 (Q)gs ((K + T )2 + 4M 2 ) GUT
4/3
MGUT
2016 GeV
4
,
(7.1)
which has an anomalous power of GUT indicating the string effects. The string enhancement
factor depends on the RaySinger torsion, the string coupling gs and the numerical quantities M,
K and T . Here the quantity M corresponds to the contribution arising from the string amplitude
10, where
10 1010 10, while the sum K + T originates from the string amplitude 5 510
K is the contribution due to the gauge boson exchange and T describes the contribution due
to the Higgs particle exchange. Choosing common values for L(Q), assuming that the string
coupling gs is approximately 1 and plugging in the computed numerical quantities K, M and T
(see Table 1) the proton lifetime (7.1) is pST = (0.52.1) 1036 years, and could lead up to a
factor of three shorter lifetime than that predicted in field theory.
133
Acknowledgements
We would like to thank Carlo Angelantonj, Andre Brown, Peng Gao, Paul Langacker, Tao
Liu and Stephan Stieberger for useful discussions. The work is supported by an DOE grant DEFG03-95ER40917 and by the Fay R. and Eugene L. Langberg Chair.
Appendix A. Vertex operators for intersecting D-branes
This appendix discusses the vertex operators of bosonic and fermionic string states arising in
intersecting D-branes based on the example of intersecting D6-branes. In the following we will
consider D6-branes in flat, non-compact Minkowski space that fill out the first four dimensions
(our actual spacetime) and intersect in the 3rd, 4th and 5th complex plane. Strings that are
stretched between these D-branes have to satisfy special boundary conditions in the internal
dimensions which leads to non-integer mode expansions for the degrees of freedom. In the vertex
operators for the corresponding string configuration on introduces bosonic and fermionic twist
fields to take into account these non-integer mode excitations. These twist fields depend crucially
on the choice of intersecting angles. In this section we will present a instruction to construct the
vertex operators arising from strings stretched between intersecting D-branes in the NS-sector as
well as in the R-sector.
As a first step we deduce the mode expansions for the bosonic and fermionic degrees of freedom. We start with the NS-sector, where strings stretched between the intersecting D-branes
correspond to massive scalars in the four-dimensional spacetime. After deriving the mode expansions we quantize the string, impose the condition for physical states, and obtain the mass
formula. Later we will also deal with strings in the R-sector and show that in this sector we
always have a massless fermion, independent of the choice of the intersection angles, while in
the NS-sector the scalars become massless only for particular choices of angles that match with
the supersymmetry condition. To get an idea of how the vertex operators look like, in particular
in the internal dimensions, we examine the operator product expansions (OPEs) of the bosonic
and fermionic fields with specific string excitations. These OPEs show the same behavior as
the OPEs of the twist fields in orbifold theories [39]. Therefore the vertex operators for strings
stretched between intersecting D-branes will involve bosonic and fermionic twist fields, and
s in the internal dimensions. The exact knowledge of the OPEs of the bosonic and fermionic
fields with the string states allows us to write the vertex operators for the string states in arbitrary
intersecting D-brane configurations.
An open string stretched between two D-branes at an angle I has to fulfill the boundary
conditions [40,41]
X p (, 0) = 0 = X p+1 (, 0),
Xp (, ) + tan(I ) Xp+1 (, ) = 0,
Xp+1 (, ) tan(I )Xp (, ) = 0.
(A.1)
Given these boundary conditions, we can deduce the mode expansion for Z I (we use complex
coordinates Z I = X 2I +2 + iX 2I +3 ) to
134
I
n
n+I
I
n (nI ) z
I
n+
Z I (z, z ) = n (n+II ) znI
Z I (z, z ) =
I
n+
I
+ n (n+
z nI
I)
I
n
+ n (nII ) z n+I
for I = 1, 2, 3.
(A.2)
Upon quantization the only non-vanishing commutator is

I

I
= mn+m I I .
n , m
World-sheet supersymmetry
X p = p
leads to the same modding for the complexified world-sheet fermions (here we already used the
doubling trick)

1
I
r 12 +I
I
I (z) =
I (z) =
(A.3)
r
z
,
r+
z r 2 I .
I
I
n+ 12
n+ 12
Notice that we consider the NS-sector where the fermions are half integer modded. The only
non-vanishing anti-commutator is given by

I
I
= m,n .
mI , n+
I
For positive I (0 < I < 1) the vacuum in the internal dimensions is defined by
I
m
|0 = 0,
I
m 1,
I
r+
|0 = 0,
I
I
m+
|0 = 0,
I
m 0,
I
r+
|0 = 0,
I
1
r ,
2
1
r .
2
(A.4)
The physical state constraint requires annihilation with all the positive modes of the Virasoro
generators Ln , in particular with L0 , which takes the form
L0 =
3

=0
:n n : +
nZ
n:n n :
nZ
3

I =1
I
:m+
I
:+
I mI
mZ
I
(m I ):m+
I
: + 0 .
I mI
(A.5)
mZ
Here n and n denote the excitations in spacetime and 0 is the zero point energy. Using the
fact that the zero mode 0 represents the momentum of the string we manipulate Eq. (A.5) and
obtain a mass formula for the open string in the twisted sector
M =
2
3

=0 nZn
3

I =1
:n n : +
mZ
n:n n :
nZ
I
:m+
I
:+
I mI

mZ
I
(m I ):m+
I
: + 0 .
I mI
(A.6)
135
The zero point energy can be computed from the -function regularization, as we demonstrate in
the following (for one internal dimension only)
0I
0

1/2

[m+I , mI ] +
m=
(r I ){r+I , rI }
m=
1 1
= [1, I ] [1, 1/2 + I ] = + I .
8 2
(A.7)
To get an expression for the vertex operators we need to determine the OPEs of I and I with
some particular excitations. First we examine the vacuum state |0
1
(z)|0 =
I
r 12 +I
2

rI |0 =
r=
zr 2 +I rI |0 zI tI (0),
r=
where tI (0) denotes the excited twist field at the intersection. Similarly we obtain for I (z)|0
I (z)|0 zI tI (0).
Using the same procedure, the OPE of and with the state 1 +I |0 is
2
(z) 1 +I |0 z
I
I 1
tI (0),
(z) 1 +I |0 z1I tI (0).

2
Considering a negative angle I (1 < I < 0) leads to a different definition of the vacuum
1
r ,
2
1
I
I
r+I |0 = 0, r
m+I |0 = 0, m 1,
2
and the zero point energy, calculated in the same way as above, takes the form
I
m
|0 = 0,
I
m 0,
I
r
|0 = 0,
I
(A.8)
1 1
0I = I
(A.9)
8 2
(keep in mind, that the angle I is negative). Again we examine the OPEs of some special physical states with the fermionic fields (z) and (z). For |0 we get
I (z)|0 zI tI (0),
I (z)|0 zI tI (0)
and for 1 I |0

2
(z) 1 I |0 z1+I tI (0),

I
I (z) 1 I |0 z1I tI (0).

2
Before formulating the vertex operators for particular states we also need the OPEs with the
bosonic fields
I (z)|0 zI I (0),
Z I (z)|0 z(1I ) I (0),
Z
Z I (z)|0 zI I (0),
Z I (z)|0 z(1I ) I (0).
For negative angle, we replace I by I = 1 + I .
Now we can start to construct the vertex operators for the respective states. First we consider
the state = 1 3 |0, where 1 , 2 are negative and 3 is positive, which means that the string
2
136
starts at D-brane a and ends at D-brane b (see Fig. 1).8 The mass of this state is given by

1 1
1
1
1
M 2 = 1 2 + 3 + 3 =
I .
2 2
2
2
2
3
I =1
The scalar becomes massless when the sum of the angles adds up to zero. This is in agreement
with the supersymmetry condition. The corresponding vertex operator in the (1)-ghost picture
takes the form
V1 (z) = e(z)
2

I (z)eiI HI (z) 1+3 (z)ei(1+3 )H3 (z) eikX(z) ,
(A.10)
I =1
where the HI s denote the bosonized world-sheet fermion I . Notice that in the case of supersymmetry, when the state becomes massless (k 2 = 0), the conformal weight of the vertex operator
adds up, as required, to one.
The corresponding complex conjugate state is represented by the same excitation as above
but oriented from brane b to brane a. That means that the intersection angles I = I take the
opposite sign as before and therefore the vertex operator is given by
(z)
V1
(z) = e
2

1I (z)eiI HI (z) 3 (z)ei(1+3 )H3 (z) eikX(z) .
(A.11)
I =1
Let us take a closer look at the vertex

operators in the case of supersymmetry, when they carry
a N = 2 world-sheet charge H = 3I =1 HI . The chiral superfield has N = 2 world-sheet
charge +1, while the charge for the complex conjugate partner is 1.
Next, we examine the state = 1 +1 1 +2 1 +3 |0, where 0 < I < 1 for all I .
2
2
2
Again the string is oriented from brane a to brane b (see Fig. 2). Why we denote the state by
rather than becomes clear later. The mass of is given by

3
3
3

1 1
1
1
M2 = +
I
I
+ I = 1
2 2
2
2
I =1
I =1
(A.12)
I =1
and becomes massless, when the sum of the angles is equal to two, again in agreement with the
supersymmetry condition. The vertex operator in the (1)-ghost picture corresponding to this
state takes the form
V(1)
(z) = e(z)
3

I (z)ei(I 1)HI (z) eikX(z) ,
(A.13)
I =1
and as above the requirement that the vertex operator has conformal weight one is satisfied.
The corresponding complex conjugated state is stretched from brane b to brane a and the
intersection angles I = I are all negative. Therefore the vertex operator is given by
V(1) (z) = e(z)
3

1I (z)ei(I 1)HI (z) eikX(z) .
I =1
8 Recall that we count counter-clockwise angles positive.
(A.14)
137

A look at the N = 2 world-sheet charge in the case of supersymmetry ( 3I =1 I = 2) explains
the notation since carries charge 1 while carries +1.
We now turn to the Ramond sector, in which the string excitations between two intersecting
D-branes correspond to spacetime fermions. The mode expansion for the fermionic degrees
of freedom takes the same form as for the NeveuSchwarz (NS)-sector, but now we sum over
integers instead of half integers

1
I
r 12 +I
I
I (z) =
I (z) =
(A.15)
r
z
,
r+
z r 2 I .
I
I
n
z ). The vacuum is defined

Nothing changes for the bosonic world-sheet fields Z(z, z ) and Z(z,
by (0 < I < 1)
I
m
|0 = 0,
I
I
m+
|0 = 0,
I
m 1,
I
r
|0 = 0,
I
r 1,
m 0,
I
r+
|0 = 0,
I
r 0.
(A.16)
With this definition the zero point energy is independently of the choice of angles given by
0I = 0,
(A.17)
and therefore we always have a massless fermion in spacetime. While the mass of the vacuum
is independent on the angles the vertex operator for the vacuum |0 depends crucially on the
choice of angles. Let us therefore examine the OPEs of world-sheet fermions9 with |0 for the
two different situations that we have positive and negative intersecting angles. We obtain for
0 < I < 1
1
I (z)|0 z 2 +I tI (0),
I (z)|0 z 2 I tI (0).
1
For negative angles we must change the definition of the vacuum to

I
m
|0 = 0,
I
m 0,
I
r
|0 = 0,
I
r 0,
I
m+
|0 = 0,
I
m 1,
I
r+
|0 = 0,
I
r 1.
(A.18)
The zero point energy is still zero. But now we obtain different OPEs for the vacuum |0
1
I (z)|0 z 2 +I tI (0),
I (z)|0 z 2 I tI (0).
1
As before for the NS-sector we present for particular states the vertex operators. The first state
we consider is the vacuum state = |0, whose mass is independent of the choice of angles
equal to zero. Assuming, that the intersecting angles 1 , 2 in the first two internal dimensions
are positive and 3 negative, the vertex operator takes the form
1
V 2 (z) = e 2 (z) S (z)
2

I (z)ei(I 2 )HI (z) 1+3 (z)ei(3 + 2 )H3 (z) eikX(z) ,
(A.19)
I =1
where S = e 2 H1 2 H2 denotes the spin field with positive chirality.10 As for the NS-sector the
corresponding vertex operator for the complex conjugated state is simply given by orientation
1
9 The OPE with bosonic world-sheet fields is the same as before for the NS-sector.
10 eH1,2 are the bosonized world-sheet fermions a where a denotes the four dimensional complexified indices.
138
reversal, so that the intersection angles are I = I . Thus the vertex operator in ( 12 )-ghost
picture has the form
1
V 2 (z) = e 2 (z) S (z)
2

1I (z)ei(I 2 )HI (z) 3 (z)ei(3 + 2 )H3 (z) eikX(z) ,
(A.20)
I =1
where S = e 2 H1 2 H2 represents the spin field with opposite chirality as S . Notice that independent of the choice of angles the
vertex operator has as expected conformal weight one. As
expected, in case of supersymmetry ( 3I =3 I = 0) the vertex operators and carry N = 2
world-sheet charge 12 and 12 , respectively.
Finally let us assume that all the intersecting angles I are positive. In that case the vertex
operator for the vacuum state takes a very symmetric form
1
V 2 (z) = e 2 (z) S (z)
3

I (z)ei(I 2 )HI (z) eikX(z) .
(A.21)
I =1
For a similar reason

as in the NS-sector we call this vacuum state rather than , since in case
of supersymmetry ( 3I =1 I = 2) it carries 12 N = 2 world-sheet charge. Following the procedure
described above we obtain for
3

1
12
2 (z)
V (z) = e
S (z)
1I (z)ei(I 2 )HI (z) eikX(z) .
I =1
(A.22)
One can easily check that in case supersymmetry the vertex operator carries as expected N = 2
world-sheet charge H = 12 .
Appendix B. Numerical analysis
Before we extract the low energy limit of the amplitudes, given above, let us take a look
at three different limits, namely x 0, x 1 and x . The first one corresponds
in the field theory to a gauge boson exchange, while the latter one corresponds to a Higgs
boson exchange. In the limit x 1 the type of the exchange particle depends on which

amplitude we examine; it is either a massive particle, for V 5 1 V 5 1 V 101 V 101 or again a
gauge boson for V 101 V 101 V 101 V 101 .

2
2 2
2
10 10 10
10
V 1 V 1 V 1 V 1 .
2
2 2
2
We start with
2
2 2
5
10
5
V 1 V 1 V 1 V 101
2
2
2 2
and turn later to
x0
The limit x 0 was already explored in Section 4 in order to normalize the amplitude. Here
we just state the result for the case that 1 = 2 =
12 < 1 < 0, 12 < 2 < 0,

3/2
0
dx
x
1
2
< 3 < 1
!

"
(1 + 2, 1 4 ) 1
(, 1 + 2 ) 2
ln
,
ln
x
x
(B.1)
139
where ln (, ) is given by
1
1
1
1
ln (, ) = 2(1) ( ) (1 ) () (1 ).
2
2
2
2
1
2
< 1 < 1,

3/2
1
2
dx
x
< 2 < 1,
1
2
(B.2)
< 3 < 1
!

"
(1 + 2, 3 4 ) 1
(, 1 + 2 ) 2
ln
ln
x
x
(B.3)
with the same (, ) as above.

x1
Using the properties of the Hypergeometric function, in particular the transformation law
2 F1 (a, b, c; x) =
(c)(c a b)
2 F1 (a, b, a + b c + 1; 1 x)
(c a)(c b)
(c)(a + b c)
+ (1 z)cab
(a)(b)
2 F1 (c a, c b, c a b 1; 1 x)
and the limit

lim 2 F1 (a, b, c; x) = 1,
x0
we obtain
1 1
lim
I (a, b, x)
x0 2
# (1a)(b)(1+ab)
ba ,
(a)(1b)(ba) (1 x)
(a)(1b)(1a+b)
ab ,
(1a)(b)(ab) (1 x)
a < b,
(B.4)
a > b.
In this limit we do not obtain an integer mode, which tells us that the exchange particle is massive.
The mass depends on the choice of angles, as we will show based on our first case ( 12 < 1 < 0,
12 < 2 < 0, 12 < 3 < 1). Let us assume that the two angles 1 and 2 are equal
1 = 2 = 3 = 2.
(B.5)
In the limit x 1 the amplitude (4.4) takes the form

$
(1 + )(1 + 2 )(3 ) (1 + 2 )(2 + 4 )(1 6 )
( )(2 )(1 + 3 )
(2 )(1 4 )(2 + 6 )
1

(1 x) u+1+6
for > 1/3 and
( )(2 )(2 + 3 )
(1 + )(1 + 2 )(1 3 )

1
(1 x) u36
(B.6)
(2 )(1 4 )(3 + 6 )

(1 + 2 )(2 + 4 )(2 6 )
(B.7)
140
for < 1/3. In the low energy limit (B.6) and (B.7) are proportional to
A
1
u M 2
(B.8)
where M denotes the mass of the exchanged particle and is given by

#
2 + 6
> 13 ,
2
M =
2 6 < 13 ,
(B.9)
which becomes massless for = 1/3. For this choice of angle we observe N = 2 supersymmetry
in the Minkowski-space. Since we focus on models with N = 1 chiral fermion sector, only, we do
not take this limit. For our second amplitude (4.7) we also observe a massive particle exchange
in this limit
$
(1 + 2 )(2 + 4 )(5 6 )
( )(1 + 2 )(3 3 )
(1 )(2 2 )(2 + 3 ) (2 2 )(3 4 )(4 + 6 )

1
(1 x) u5+6
for > 2/3 and

(1 )(2 2 )(1 + 3 )
( )(1 + 2 )(2 3 )
1
(B.10)
$
(2 2 )(3 4 )(3 + 6 )

(1 + 2 )(2 + 4 )(4 6 )
(1 x) u+36
(B.11)
for < 2/3. In our effective low energy theory we integrate out all massive states, so that the part
of the amplitude arising from these string massive state exchanges contribute to the four-Fermi
contact term.
x
At last let us examine the limit x . As mentioned earlier the second terms of (4.4)
and (4.7) give the contribution to the four Fermi interaction arising from the massless Higgs particle exchange. Therefore in the limit x we expect to observe an exchange of a massless
particle.
The hypergeometric functions behave in the limit x
(c)(b a) a (c)(a b) b
x +
x ,
(b)(c a)
(a)(c b)
(c)(b a) a
(c)(a b) b
lim F (a, b, c, 1 x) = eia
x + eib
x .
x
(b)(c a)
(a)(c b)
Hence I (a, b, x) for x takes the form

1
(1)ab x a+b a,b ,
0 < a + b < 1,
1
lim
Ij (a, b, x)
ab
2ab
x 2
1a,1b , 1 < a + b < 2,
(1) x
lim F (a, b, c, x) =
(B.12)
with
a,b =
(1 a)(1 b)(a + b)

.
(a)(b)(1 a b)
(B.13)
Using (B.12) the amplitude (4.4) becomes in the limit x

1
1
1
3

2
dx x t1 .
(2) 2 21 ,121 22 ,122 1+
3 ,123
141
(B.14)
Thus, we observe an exchange of a massless particle, which we identify as the Higgs-particle.

Note that the prefactor in (B.14) is the expected relative factor between the Yukawa couplings in
string and field theory basis [32,33,42].
Applying the limit for our second amplitude (4.7) we obtain
3
(2) 2
3

I =1
1
2
1+
I ,12I
dx x t1
(B.15)
and again we can observe a massless Higgs exchange in this limit.

2
2
B.1. The amplitude V 5 1 V 5 1 V 101 V 101

2
The analysis for both amplitudes, (4.4) and (4.7) is similar, so that we will describe the steps
for the first one and apply these later for the second amplitude. We start by investigating the
integral K(1 , 2 , 3 ) and turn later to T (1 , 2 , 3 ).
K(1 , 2 , 3 )
Since in this interval the amplitude is finite even in the low energy limit, we do not have to
subtract anything. Thus, we can send to zero and obtain
1
K=
dx
0
1

1
I (1 , 1 + 1 , x)I (2 , 1 + 2 , x)I (1 3 , 1 + 3 , x) 2 .
x(1 x)
(B.16)
Let us split the integral (B.16) by using the expression

1
1
1
= +
.
x(1 x) x 1 x
(B.17)
Let us first evaluate the integral starting with the first summand of (B.16) which is given by
1
K1 =
dx
1
1
I (1 , 1 + 1 , x)I (2 , 1 + 2 , x)I (1 3 , 1 + 3 , x) 2 .
x
(B.18)
Substituting et for x we obtain

K1 =

1

2.
dt I 1 , 1 + 1 , et I 2 , 1 + 2 , et I 1 3 , 1 + 3 , et
(B.19)
Mathematica is not able to evaluate this expression numerically since it is hard to maintain numerical precision for large t . Therefore we will split integral (B.19) into the range from 0 to T
and from T to . For the computation of the first region we will use Mathematica to evaluate it numerically, while for the second region we replace the hypergeometric functions by their
142
asymptotic behavior given in (B.1)

T

1

2
dt I 1 , 1 + 1 , et I 2 , 1 + 2 , et I 1 3 , 1 + 3 , et
K1 =
0

+ 3/2
dt

t + ln (1 , 1 + 1 ) t + ln (2 , 1 + 2 )
1
2.
t + ln (1 3 , 1 + 3 )
Let us assume that the two angles 1 and 2 are equal to each other
1 = 2 = 3 = 2.
Then K1 simplifies to
T
K1 =

1

2
dt I , 1 + 2, et I , 1 + 2, et I 1 + 2, 1 4, et

+
3/2
1

1
t + ln (1 + 2, 1 4 ) 2 .
dt t + ln (, 1 + 2 )
(B.20)
Now we turn to the second term we get after splitting the integral. Again we substitute et for x,
set, as above, 1 = 2 = and obtain

K2 =
dt
0

1
et 2
t
t 2
I
,
1
+
2,
e
I
1
+
2,
1
4,
e
.
1 et
As above we have to split this integral into two parts, where we replace the I s by their asymptotic
behavior
T
K2 =
dt
0

1
et 2
2
I , 1 + 2, et I 1 + 2, 1 4, et
t
1e

3/2
dt
T
1
et
t + ln (, 1 + 2 )
1 et

1
t + ln (1 + 2, 1 4 ) 2 .
Applying the same procedure for the other sector we obtain
1
2
< 1 < 1,
1
2
< 2 < 1,
1
2
< 3 < 1
(B.21)
143
In this sector we obtain

T

1

2
K1 = dt I 2 1 , 1 + 2, et I 2 1, 3 4, et
0

+
3/2
1

1
t + ln (2 1, 3 4 ) 2
dt t + ln (1 , 1 + 2 )
(B.22)
and
T
K2 =
dt
0

1
et 2
t
t 2
I
1
,
1
+
2,
e
I
2
1,
3
4,
e
1 et

3/2
dt
T
1
1
et
t + ln (1 , 1 + 2 )
t + ln (2 1, 3 4 ) 2 .
t
1e
(B.23)
The whole integral K( ) is given by the sum of K1 and K2 .

T (1 , 2 , 3 )
Let us now analyze the massive string state contribution to T (1 , 2 , 3 ), where in the field
theory the proton decay takes place via Higgs particle mediation. Thus, in contrast to the numerical analysis for proton decay via a gauge boson exchange we observe a pole that corresponds to
the Higgs exchange. In order to obtain the four-Fermi interaction term due to the massive string
states, we need to subtract this pole before taking the low energy limit.
Let us split the integral (4.5) into two parts (again we assume that 1 = 2 = )
L

1

dx x s1 (1 x) u1 I 2 (, 1 + 2, x)I (1 + 2, 1 4, x) 2
0
+

1

dx x s1 (1 x) u1 I 2 (, 1 + 2, x)I (1 + 2, 1 4, x) 2 .
(B.24)
Now we replace x by 1 ez in the first summand and in the second by

dz
ln(1L)
[I 2 (, 1 + 2, 1 ez )I (1 + 2, 1 4, 1 ez )] 2

(ez ) u (1 ez ) s+1

+
dz
ln(1 L1 )
1
1
[I 2 (, 1 + 2, 1e
z )I (1 + 2, 1 4, 1ez )] 2

(ez ) u (1 ez ) t
To simplify the computation, we break up both terms into two parts

T1
ln(1L)
1
1ez
[I 2 (, 1 + 2, 1 ez )I (1 + 2, 1 4, 1 ez )] 2
dz

(ez ) u (1 ez ) s+1
(B.25)
144
1
2
3
2
+ (2) ,1+2 1+2,14
s1
u+1
1 ez
dz ez
T1
T2
dz
1
1
[I 2 (, 1 + 2, 1e
z )I (1 + 2, 1 4, 1ez )] 2

3
2
(ez ) u (1 ez ) t
ln(1 L1 )
dz
T2
(z + ln (, 1 + 2 ))1 (z + ln (1 + 2, 1 4 )) 2
.

(ez ) u (1 ez ) t
Here we replaced the hypergeometric expressions by their respective limits in the range from T1
to and T2 to . As mentioned above in order to get the four-Fermi interaction contribution,
we need to subtract the 1 t pole and take the low energy limit
# T1
T ( ) = lim

dz
0
ln(1L)
[I 2 (, 1 + 2, 1 ez )I (1 + 2, 1 4, 1 ez )] 2

(ez ) u (1 ez ) s+1

1
2
3
2
+ (2) ,1+2 1+2,14
s1
u+1
1
1 ez
dz ez

t
T1
T2
dz
1
1
[I 2 (, 1 + 2, 1e
z )I (1 + 2, 1 4, 1ez )] 2

3
2

T2
For the second region
%
1
(z + ln (, 1 + 2 ))1 (z + ln (1 + 2, 1 4 )) 2
dz
.

(ez ) u (1 ez ) t
1
2
< 1 < 1,
# T1
0
ln(1L)
3
2
1
2
< 2 < 1 and
1
2
< 3 < 1, T ( ) takes the form

1
T ( ) = lim

(ez ) u (1 ez ) t
ln(1 L1 )
[I 2 (1 , 1 + 2, 1 ez )I (1 + 2, 3 4, 1 ez )] 2
dz

(ez ) u (1 ez ) s+1
1
2
+ (2) 1,1+2 1+2,34
s1
u+1
1
1 ez
dz ez

t
T1
T2
dz
ln(1 L1 )
3
2

T2
1
1
[I 2 (1 , 1 + 2, 1e
z )I (1 + 2, 3 4, 1ez )] 2

(ez ) u (1 ez ) t
%
1
(z + ln (1 , 1 + 2 ))1 (z + ln (1 + 2, 3 4 )) 2
dz
.

(ez ) u (1 ez ) t
145
Mathematica is not able to take that limit, however by plugging in different small values for
(keep in mind that the Mandelstam variables s, t and u have to satisfy momentum conservation
s + t + u = 0) we get a stable contribution for T ().
B.2. The amplitude V 101 V 101 V 101 V 101

2

2
2
The analysis is simpler for V 5 1 V 5 1 V 101 V 101 because of the symmetry of the amplitude:
2
after splitting the integral (B.17) both parts give the same contribution, so that we only need to
focus on one part and multiply by a factor of two. Following the same steps as above the integral
M becomes
T
M =2
3 &

dt
I =1

+

1
sin (1 + I ) L 2 (1 + I )
3/2
dt
T
3

1
t + ln (1 + I , 1 + I ) 2 .
I =1
Replacing I by I and assuming that 1 = 2 we get for

12 < 1 < 0, 12 < 2 < 0,
T
M =2
1
2
< 3 < 1

&

1
dt sin (1 + 2 ) sin (1 4 ) L1 (1 + 2 )L 2 (1 4 )

+
3/2
1

1
t + ln (1 4, 1 4 ) 2 ,
dt t + ln (1 + 2, 1 + 2 )
(B.26)
and for
1
2
< 1 < 1,
T
M =2
1
2
< 2 < 1,
1
2
< 3 < 1

&

1
dt sin (2 1) sin (3 4 ) L1 (2 1)L 2 (3 4 )

+
3/2
1

1
t + ln (3 4, 3 4 ) 2 .
dt t + ln (2 1, 2 1)
References
[1]
[2]
[3]
[4]
[5]
H. Georgi, S.L. Glashow, Unity of all elementary particle forces, Phys. Rev. Lett. 32 (1974) 438441.
P. Langacker, Grand unified theories and proton decay, Phys. Rep. 72 (1981) 185.
P. Nath, P.F. Perz, Proton stability in grand unified theories, in strings, and in branes, hep-ph/0601023.
N. Sakai, Naturalness in supersymmetric GUTS, Z. Phys. C 11 (1981) 153.
S. Dimopoulos, H. Georgi, Softly broken supersymmetry and SU(5), Nucl. Phys. B 193 (1981) 150.
(B.27)
146
[6] H. Murayama, A. Pierce, Not even decoupling can save minimal supersymmetric SU(5), Phys. Rev. D 65 (2002)
055009, hep-ph/0108104.
[7] R. Dermsek, A. Mafi, S. Raby, SUSY GUTs under siege: Proton decay, Phys. Rev. D 63 (2001) 035001, hepph/0007213.
[8] J. Hisano, Proton decay in the supersymmetric grand unified models, hep-ph/0004266.
[9] D. Emmanuel-Costa, S. Wiesenfeldt, Proton decay in a consistent supersymmetric SU(5) GUT model, Nucl. Phys.
B 661 (2003) 6282, hep-ph/0302272.
[10] B. Bajc, P. Fileviez Perz, G. Senjanovc, Minimal supersymmetric SU(5) theory and proton decay: Where do we
stand?, hep-ph/0210374.
[11] B. Bajc, P. Fileviez Perz, G. Senjanovc, Proton decay in minimal supersymmetric SU(5), Phys. Rev. D 66 (2002)
075005, hep-ph/0204311.
[12] P. Fileviez Perz, Fermion mixings vs d = 6 proton decay, Phys. Lett. B 595 (2004) 476483, hep-ph/0403286.
[13] M. Cvetic, I. Papadimitriou, G. Shiu, Supersymmetric three family SU(5) grand unified models from type IIA
orientifolds with intersecting D6-branes, Nucl. Phys. B 659 (2003) 193223, hep-th/0212177.
[14] M. Cvetic, G. Shiu, A.M. Uranga, Chiral four-dimensional N = 1 supersymmetric type IIA orientifolds from intersecting D6-branes, Nucl. Phys. B 615 (2001) 332, hep-th/0107166.
[15] M. Cvetic, G. Shiu, A.M. Uranga, Three-family supersymmetric standard like models from intersecting brane
worlds, Phys. Rev. Lett. 87 (2001) 201801, hep-th/0107143.
[16] R. Blumenhagen, M. Cvetic, P. Langacker, G. Shiu, Toward realistic intersecting D-brane models, hep-th/0502005.
[17] G. Aldazabal, S. Franco, L.E. Ibez, R. Rabadn, A.M. Uranga, Intersecting brane worlds, JHEP 0102 (2001) 047,
hep-ph/0011132.
[18] R. Blumenhagen, B. Krs, D. Lst, T. Ott, The standard model from stable intersecting brane world orbifolds, Nucl.
Phys. B 616 (2001) 333, hep-th/0107138.
[19] G. Aldazabal, S. Franco, L.E. Ibez, R. Rabadn, A.M. Uranga, D = 4 chiral string compactifications from intersecting branes, J. Math. Phys. 42 (2001) 31033126, hep-th/0011073.
[20] R. Blumenhagen, L. Grlich, B. Krs, D. Lst, Noncommutative compactifications of type I strings on tori with
magnetic background flux, JHEP 0010 (2000) 006, hep-th/0007024.
[21] C. Angelantonj, I. Antoniadis, E. Dudas, A. Sagnotti, Type-I strings on magnetised orbifolds and brane transmutation, Phys. Lett. B 489 (2000) 223232, hep-th/0007090.
[22] C.M. Chen, G.V. Kraniotis, V.E. Mayes, D.V. Nanopoulos, J.W. Walker, A supersymmetric flipped SU(5) intersecting brane world, Phys. Lett. B 611 (2005) 156166, hep-th/0501182.
[23] C.-M. Chen, V.E. Mayes, D.V. Nanopoulos, Flipped SU(5) from D-branes with type IIB fluxes, Phys. Lett. B 633
(2006) 618626, hep-th/0511135.
[24] T.P.T. Dijkstra, L.R. Huiszoon, A.N. Schellekens, Supersymmetric standard model spectra from RCFT orientifolds,
Nucl. Phys. B 710 (2005) 357, hep-th/0411129.
[25] P. Anastasopoulos, T.P.T. Dijkstra, E. Kiritsis, A.N. Schellekens, Orientifolds, hypercharge embeddings and the
standard model, hep-th/0605226.
[26] R. Tatar, T. Watari, Proton decay, Yukawa couplings and underlying gauge symmetry in string theory, hepth/0602238.
[27] I.R. Klebanov, E. Witten, Proton decay in intersecting D-brane models, Nucl. Phys. B 664 (2003) 320, hepth/0304079.
[28] E.G. Gimon, J. Polchinski, Consistency conditions for orientifolds and D-manifolds, Phys. Rev. D 54 (1996) 1667
1676, hep-th/9601038.
[29] M. Axenides, E. Floratos, C. Kokorelis, SU(5) unified theories from intersecting branes, JHEP 0310 (2003) 006,
hep-th/0307255.
[30] M. Berkooz, M.R. Douglas, R.G. Leigh, Branes intersecting at angles, Nucl. Phys. B 480 (1996) 265278, hepth/9606139.
[31] M. Cvetic, P. Langacker, T.-J. Li, T. Liu, D6-brane splitting on type IIA orientifolds, Nucl. Phys. B 709 (2005)
241266, hep-th/0407178.
[32] M. Cvetic, I. Papadimitriou, Conformal field theory couplings for intersecting D-branes on orientifolds, Phys. Rev.
D 68 (2003) 046001, hep-th/0303083.
[33] D. Lst, P. Mayr, R. Richter, S. Stieberger, Scattering of gauge, matter, and moduli fields from intersecting branes,
Nucl. Phys. B 696 (2004) 205250, hep-th/0404134.
[34] J. Polchinski, String Theory, vol. 2: Superstring Theory and Beyond, Cambridge Univ. Press, Cambridge, 1998.
[35] T. Friedmann, E. Witten, Unification scale, proton decay, and manifolds of G(2) holonomy, Adv. Theor. Math.
Phys. 7 (2003) 577617, hep-th/0211269.
[36]
[37]
[38]
[39]
147
D. Lst, S. Stieberger, Gauge threshold corrections in intersecting brane world models, hep-th/0302221.
Particle Data Group Collaboration, S. Eidelman, et al., Review of particle physics, Phys. Lett. B 592 (2004) 1.
C.K. Jung, Feasibility of a next generation underground water Cherenkov detector: Uno, hep-ex/0005046.
L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, The conformal field theory of orbifolds, Nucl. Phys. B 282
(1987) 1373.
[40] H. Arfaei, M.M. Sheikh Jabbari, Different D-brane interactions, Phys. Lett. B 394 (1997) 288296, hep-th/9608167.
[41] S.A. Abel, A.W. Owen, Interactions in intersecting brane models, Nucl. Phys. B 663 (2003) 197214, hepth/0303124.
[42] M. Bertolini, M. Bill, A. Lerda, J.F. Morales, R. Russo, Brane world effective actions for D-branes with fluxes,
Nucl. Phys. B 743 (2006) 140, hep-th/0512067.
The one-plaquette model limit

of NC gauge theory in 2D
Badis Ydri 1
Department of Physics, Faculty of Science, Badji Mokhtar-Annaba University, Annaba, Algeria
Received 9 August 2006; received in revised form 25 September 2006; accepted 25 October 2006
Abstract
It is found that noncommutative U (1) gauge field on the fuzzy sphere S2N is equivalent in the quantum
theory to a commutative 2-dimensional U (N ) gauge field on a lattice with two plaquettes in the axial
gauge A1 = 0. This quantum equivalence holds in the fuzzy sphere-weak coupling phase in the limit of
infinite mass of the scalar normal component of the gauge field. The doubling of plaquettes is a natural
consequence of the model and it is reminiscent of the usual doubling of points in Connes standard model.
In the continuum large N limit the plaquette variable W approaches the identity 12N and as a consequence
the model reduces to a simple matrix model which can be easily solved. We compute the one-plaquette
critical point and show that it agrees with the observed value = 3.35. We compute the quantum effective
2 in the 1/N expansion using this
potential and the specific heat for U (1) gauge field on the fuzzy sphere SN
one-plaquette model. In particular the specific heat per one degree of freedom was found to be equal to 1
in the fuzzy sphere-weak coupling phase of the gauge field which agrees with the observed value 1 seen in
Monte Carlo simulation. This value of 1 comes precisely because we have 2 plaquettes approximating the
NC U (1) gauge field on the fuzzy sphere.
1. Introduction
Quantum noncommutative (NC) gauge theory is essentially unknown beyond one-loop [1].
In the one-loop approximation of the quantum theory we know for example that gauge models
on the MoyalWeyl spaces are renormalizable [2]. These models were also shown to behave in
E-mail address: ydri@synge.stp.dias.ie (B. Ydri).
1 Current address: Institut fr Physik, Mathematisch-Naturwissenschaftliche Fakultt I, Humboldt-Universitt zu
Berlin, D-12489 Berlin, Germany.

doi:10.1016/j.nuclphysb.2006.10.030
B. Ydri / Nuclear Physics B 762 (2007) 148188
149
a variety of novel ways as compared with their commutative counterparts. There are potential
problems with unitarity and causality when time is noncommuting, and most notably we mention the notorious UVIR mixing phenomena which is a generic property of all quantum field
theories on MoyalWeyl spaces and on noncommutative spaces in general [1,3]. However a nonperturbative study of pure two-dimensional noncommutative gauge theory was then performed
in [5]. For scalar field theory on the MoyalWeyl space some interesting nonperturbative results
using theoretical and Monte Carlo methods were obtained for example in [6]. An extensive list
of references on these issues can be found in [1] and also in [4].
The fuzzy sphere (and any fuzzy space in general) is designed for the study of gauge theories
in the nonperturbative regime using Monte Carlo simulations. This is the point of view advocated
in [7]. See also [810] for quantum gravity, string theory or other different motivations. These
fuzzy spaces consist in replacing continuous manifolds by matrix algebras and as a consequence
the resulting field theory will only have a finite number of degrees of freedom. The claim is that
this method has the advantagein contrast with latticeof preserving all continuous symmetries
of the original action at least at the classical level. This proposal was applied to the scalar 4
model in [11] and to the U (1) gauge field in [12] with very interesting nonperturbative results.
Quantum field theory on fuzzy spaces was also studied perturbatively quite extensively. See for
example [1315]. For some other nonperturbative (theoretical or Monte Carlo) treatment of these
field theories see [16,24].
Another motivation for considering the fuzzy sphere is the following. The MoyalWeyl NC
space is an infinite-dimensional matrix model and not a continuum manifold and as a consequence it should be regularized by a finite-dimensional matrix model. In 2 dimensions the most
2 which is a finite-dimensional matrix model which renatural candidate is the fuzzy sphere SN
duces to the NC plane in some appropriate large N flattening limit. This limit was investigated
quantum mechanically in [14,17]. In 4-dimensions we should instead consider Cartesian prod2 [15], fuzzy CP2 [18] or fuzzy S4 [19]. It is fair to mention here that
ucts of the fuzzy sphere SN
N
an alternative way of regularizing gauge theories on the MoyalWeyl NC space is based on the
matrix model formulation of the twisted EguchiKawai model. See for example [20,21,32].
The goal of this article and others [12,22] is to find the phase structure (i.e. map the different
regions of the phase diagram) of noncommutative U (1) gauge theories in 2 dimensions on the
fuzzy sphere S2N . We consider the fuzzy sphere since it is the most suited two-dimensional space
for numerical simulation because of the obvious fact that it is a well defined object.
There are three phases of U (1) gauge theory on S2N . In the matrix phase the fuzzy sphere vacuum collapses under quantum fluctuations and we have no underlying sphere in the continuum
large N limit. This phenomena was first observed in Monte Carlo simulation in [23] and then
in [12]. In [22] it was shown that the fuzzy sphere vacuum becomes more stable as the mass
of the scalar normal component of the gauge field increases. Hence this vacuum becomes completely stable when this normal scalar field is projected out from the model. This is confirmed in
Monte Carlo simulation in [12].
In the other phase, the so-called fuzzy sphere phase, there are in fact two distinct regions
in the phase diagram corresponding to the weak and strong coupling phases of the gauge field.
The boundary between these two regions is demarcated by the usual third order one-plaquette
phase transition [25]. This is precisely what we observe in our Monte Carlo simulation of the
model with a very large mass of the normal scalar field [12]. This result indicates that quantum
noncommutative gauge theory is essentially equivalent (at least in this fuzzy sphere phase) to
(some) commutative gauge theory not necessarily of the same rank. This prediction goes also
150
in line with the powerful classical concept of Morita equivalence between noncommutative and
commutative gauge theories on the torus [1,21].
In this paper we will give a theoretical proof that quantum noncommutative gauge theory is
equivalents to quantum commutative gauge theory in the fuzzy sphere-weak coupling phase in
the limit of infinite mass of the normal scalar component of the gauge field. More precisely we
will show that the partition function of a U (1) gauge field on the fuzzy sphere S2N is proportional
to the partition function of a generalized 2-dimensional U (N ) gauge theory in the axial gauge
A1 = 0 on a lattice with two plaquettes. This doubling of plaquettes is reminiscent of the usual
doubling of points in Connes standard model [27]. This construction is based on the original
fuzzy one-plaquette model due to [26].
However in the present article we will show that in order to maintain gauge invariance and
obtain sensible answers we will need to introduce two different U (1) gauge fields on the fuzzy
sphere which will only coincide in the continuum large N limit. This doubling of fields is not
related to the above doubling of plaquettes since it disappears in the continuum limit where the
path integral is dominated by the configuration in which the two U (1) gauge fields are equal.
Furthermore we will need in the present article to write down two different one-plaquette actions
on the fuzzy sphere. Linear and quadratic terms in the plaquette variable W are in fact needed in
order to have convergence of the path integral. We will show explicitly the classical continuum
limit of these one-plaquette actions.
Quantum mechanically since the plaquette variable W is small in the sense we will explain
we can show that the model in the large N limit will reduce to a simple matrix model and as
a consequence can be easily solved. We compute the critical point and show that it agrees with
the observed value. We will also compute the quantum effective potential for U (1) gauge field on
2 in the 1/N expansion using this one-plaquette model. This is in contrast with
the fuzzy sphere SN
the calculation of the effective potential in the limit N in the one-loop approximation done
in [22]. The difference between the two cases lies in the quantum logarithmic potential which
is in absolute value larger by a factor of 4 in the 1/N expansion as compared to the one-loop
theory. We will discuss the implication of this to the critical point and possible interpretation of
this result. We will also compute the specific heat and find it equal to 1 in the fuzzy sphere-weak
coupling phase of the gauge field which agrees with the observed value 1 seen in Monte Carlo
simulation. The value 1 comes precisely because we have two plaquettes which approximate the
noncommutative U (1) gauge field on the fuzzy sphere.
This paper is organized as follows. In Section 2 we will briefly comment on the classical
Morita equivalence between noncommutative gauge theories and commutative gauge theories on
the torus. In Section 3 we will rederive the one-loop result of [22] using an RG method. Thus we
will explicitly establish gauge invariance of the S2N -to-matrix critical point. Section 4 contains
the main original results of this article discussed in the previous three paragraphs. In Section 5
we conclude with a summary and some general remarks.
2. The NC torus and Morita equivalence
The strongest argument concerning the equivalence between classical noncommutative gauge
theories and classical commutative gauge theories comes from considerations involving the noncommutative torus and Morita equivalence. In this section we will briefly review this result
following the notations of [1,21].
Any U (N ) gauge model on the noncommutative torus T2 with a nonzero magnetic flux Q
can be shown to be Morita equivalent to a U (N0 ) gauge model on the noncommutative torus T2
151
with zero magnetic flux. The noncommutativity parameter = 2 det is given in terms of
= 2 det by the equation
a b
.
(1)
m + l
The integer l which is the ratio l = N/N0 is the dimension of the irreducible representation of
the Weylt Hooft algebra
=
1 2 = e N iQ12 2 1
(2)
found in the nontrivial solution

i
a (x) = eiai x a
of the model
SYM =
1
4g 2
(3)

d 2 x trN (Fij fij )2 .
(4)
a are constant SU(N ) matrices while ai is a 2 2 real matrix which represents the U (1) factor
of the U (N ) group.
The N N star-unitary transition functions a are global large gauge transformations
whereas fij is a constant curvature on T2 which is equal to the curvature of the U (1) background gauge field ai given by
1
1
ai = Dij x j 1N ,
(5)
D = 2 T
.
2
T
In above we have q/N = m/ l where q = Q12 is the component of the antisymmetric matrix Q
of the non-Abelian SU(N ) t Hooft flux across the different noncontractible 2-cycles of the noncommutative torus. By construction q is quantized, i.e. q Z. Furthermore m is defined by
m = qx , x = gcd(q, N ) where gcd stands for the great common divisor. Since l and m are relatively prime there exists two integers a and b such that al + bm = 1. These are the same integers
a and b which appear in (1).
The period matrix of the dual torus T2 is related to the period matrix of the torus T2 by
= (m + l).
(6)
The dual metric is therefore = T = (m + l)2 1 which is to be compared with the original
metric = T = 1.
The dual action is by the very definition of Morita equivalence equal to the U (N ) gauge action
SYM on T2 , viz.

2
1
SYM = 2 d 2 x trN0 Fij (x )
(7)
4g
where g 2 is given in terms of g 2 by the equation
g 2 = g2
N0
(m + l)2 .
N
(8)
In other words SYM can also be interpreted as a U (N0 ) gauge action on T2 . It is understood that
the star product here is the one associated with the parameter . It is the U (1) background gauge
field ai which is used to twist the boundary conditions on the U (N ) gauge field and hence obtain
152
a nontrivial field configuration. Indeed the curvature f12 of the vacuum gauge configuration ai
on T2 is related to the SU(N ) t Hooft magnetic flux q by the equation
f12 =
2q
1
.
det N + q
(9)
If we turn this equivalence upside down then we can obtain a correspondence between a U (1)
gauge model on a (finite-dimensional) fuzzy torus TJ2 and an ordinary U (N ) gauge model on T 2 .
In particular we remark that if we set = 0, = 1 and N0 = 1 in the above equations then l = N
and m = q and we will have an ordinary U (N ) on a square torus T 2 with nonzero magnetic flux
2
12
q = Nf
2 and a coupling constant g . The dual torus in this case is also square since its period

matrix is given by = N . 1 whereas its noncommutativity parameter becomes
=
b
.
N
(10)
The commutation relation of the NC torus T2 becomes therefore

b

.
z 2 z 1 = z 1 z 2 exp 2i
N
(11)
Since the noncommutativity parameter here is rational we know that this Lie algebra must have
a finite-dimensional N N representation which can be written down in terms of shift and clock
matrices as usual. In other words
z 1 = VN ,
z 2 = (WN )b .
(12)
VN and WN are the canonical SU(N ) clock and shift matrices which satisfy VN WN =
2i
e N WN VN . This is indeed a fuzzy torus, i.e. T2 = TN2 . The coupling constant of the U (1) model
on TN2 is g 2 = g 2 N .
3. The fuzzy sphere S2N
Let Xa , a = 1, 2, 3, be three N N hermitian matrices and let us consider the action
S[Xa ] =
2
N
iN
Tr[Xa , Xb ]2 +
abc Tr[Xa , Xb ]Xc + Tr Xa2 + M Tr Xa2 .
4
3
(13)
This action is invariant under U (N ) unitary transformations Xa U Xa U + . The trace is normalized such that Tr 1 = N . , and M are the parameters of the model. This action is bounded
from below for all strictly positive values of M. For M = 0 this model is also symmetric under
global translations Xa Xa + xa 1N where x is any constant vector. We can fix this symmetry
by choosing the matrices Xa to be traceless.
The classical equations of motion read

Ja [X] = N 2 [Xb , iXab ] + 2Xa + 2M Xb2 , Xa 0,
2 Xab = i[Xa , Xb ] + abc Xc .
(14)
Absolute minima of the action are explicitly given by the fuzzy sphere solutions
Xa = RLa 1n .
(15)
153
La are the generators of SU(2) in the irreducible representation L2 . They satisfy [La , Lb ] =
iabc Lc , L2a = L2 ( L2 + 1) c2 and they are of size (L + 1) (L + 1), viz. N = n(L + 1). R is
the radius of the sphere given explicitly by the solution of the equation

2c2 M
1+
(16)
R 2 R + = 0.
N
N
In particular for = M = 0 we have the solution R = . If we insist that R = then we will
have the constraint
2c2 M
.
= 2M 2 c2 = 2 m2 , = N , m2 =
(17)
N
In general we can show that a solution of (16) exists if and only if is such that
,
m2 ) with
Explicitly we have R = R N = R(

+ 2 4(1 + m2 )

2
.
R ,
m =
2(1 + m2 )
2
.
4(1+m2 )
(18)
In the following we will strictly work with the case R = . We expand around the solution (15)
by writing
Xa = (La + Aa ).
(19)
Aa , a = 1, 2, 3 are N N hermitian matrices which admit the interpretation of being the components of a U (n) gauge field on a fuzzy sphere of size (L + 1) (L + 1). To see this we introduce
the curvature tensor by

2 Fab = i[Xa , Xb ] + abc Xc Fab = iLa Ab iLb Aa + abc Ac + i[Aa , Ab ] .
(20)
We also introduce the normal component of A by

Xa2 2 c2
1
A2a
2
=
xa Aa + Aa xa +
=
(21)
,
2 c2
2
c2
where La = [La , . . .] and xa =
La
c2
are the derivations and coordinate-operators on the fuzzy
We can then check that the action S[Xa ] takes the form

4
4
i
2
Tr Fab
abc Tr Fab Ac [Aa , Ab ]Ac
S[Aa ] =
4N
4N
3
sphere
S2L .
1
1
2 4 m2
Tr 2 4 c2 4 c2 m2 .
N
6
2
(22)
We note that a natural definition of the U (n) gauge coupling constant is given by g 2 = 14 . Also
we note that S0 S[Aa = 0] = 16 4 c2 12 4 c2 m2 . Finally we remark that there is no linear
term in . For completeness we will include a linear term in as follows

4
4
i
2
a ] = Tr Fab

S[A
4N
4N
3
+
2 4 m2
Tr( 0 )2 + S0
N
154

4
4
i
2
Tr Fab
4N
4N
3
+
2 4 m2
4 4 m2
Tr 2
0 Tr + S0 ,
N
N
(23)
S0 = S0 + 2 4 m2 02 . Now in the limit m the field will be equal to a constant given

by 0 .
3.1. The effective action from an RG method
We are interested in the partition function

Z = [dXa ]eS[Xa ] .
(24)
For simplicity we consider U (1) theory so that N = L + 1 and the full fuzzy U (1) symmetry
is given by the gauge group U (N ). The treatment of U (n) is identical. We will also confine our
analysis to the case where the coupling constant is related to m by (17).
We will fix the U (N ) symmetry by diagonalizing the third matrix X3 . This will clearly reduce the original U (N ) symmetry group to its maximal Abelian subgroup U (1)N . Although this
method is not manifestly SU(2)-covariant it is completely gauge invariant since it does not require any extra parameter to be introduced in the model unlike other gauge-fixing procedures.
Thus we will choose a unitary matrix U such that U + X3 U = 3 is a diagonal matrix with eigenvalues A , A = 1, N . We will have the simultaneous rotations U + Xi U = i , i = 1, 2. As it turns
out X3 = U 3 U + can also be thought of as a parametrization of the matrix X3 in terms of its
radial degrees of freedom encoded in 3 and its angular degrees of freedom given by U = ei .
Indeed we can compute the following metric and measure

Tr(dX3 )2 =
d2A + 2
(A B )2 dAB dAB
,
A
[dX3 ] =
A<B
dA
A=1

(A B )

A<B
dab dab

.
(25)
A<B
The partition function becomes (since the integration over the unitary matrix U decouples)

dA eS[i ,A ] ,
Z = [di ]
(26)
A=1
where the action is now given by

log(A B )2 ,
S[i , A ] = SN [a ]
A<B
N
iN
abc TrN [a , b ]c
SN [a ] = TrN [a , b ]2 +
4
3
2
+ TrN 2a + M TrN 2a .
(27)
We are using now the new notation Tr TrN . We stress again the fact that this action is still
symmetric under the Abelian U (1)N transformation i V + i V and 3 3 where V is
given explicitly by
VAB = eiA AB .
(28)
155
Now we adopt the RG prescription of [28] to find the quantum corrections of this action at oneloop. To this end we parametrize the N N matrices a in terms of (N 1) (N 1) matrices
Da , (N 1)-dimensional vectors va and 1-dimensional vectors a as follows

Da va
a =
(29)
.
va a
Since 3 is diagonal we must have v3 = 0 while 3 = N . This method consists in finding
quantum corrections to the action coming from integrating out the 4(N 1) + 3 degrees of
freedom vi and a which we can naturally think of as fluctuations around a background defined
by the matrices Da . Furthermore it is not difficult to argue that this method is also equivalent
to the usual Wilson procedure of integrating out the top modes with spin L = N 1 from the
theory.
To see this more explicitly we write (a )AB = (Da )AB , (i )AN = viA , (i )N A = (viA ) and
(a )N N = a where A, B = 1, . . . , N 1. We check that the Abelian transformations (28) will
act on Da , vi and a as follows
D a W Da W + ,
va+ va+ W + ,
va W va ,
a a ,
(30)
where

(Wa )AB = ei(A N ) AB = eiN V + AB .
(31)
Next we will denote the (N 1)-dimensional trace by TrN1 and compute

TrN [a , b ]2 = TrN1 [Da , Db ]2 + vi+ 4Di Dj + 8Dj Di 4Da2 ij vj + O(3),
iabc TrN [a , b ]c = iabc TrN1 [Da , Db ]Dc 6iij 3 vi+ D3 vj
TrN 2a = TrN1 Da2 + 2vi+ vi + a2
+ O(3),
(32)
(33)
(34)
and
3
2
2

a4 + vi+ 2Da2 ij + 2Di Dj vi + O(3).
TrN 2a = TrN1 Da2 +
(35)
a=1
O(3) stands for cubic or higher order terms. In one-loop approximation it is sufficient that one
keeps only terms up to quadratic powers in the fluctuation fields which are
identified here with the
vi and a degrees of freedom. However we have kept the quartic term 3a=1 a4 in equation (35)
for other purposes which will become clearer shortly. The action SN [a ] is then given by
SN [a ] = SN1 [Da ] + vi+ ij vj +
3

a2 + Ma4 + O(3),
(36)
a=1
where clearly
N
iN
TrN1 [Da , Db ]2 +
abc TrN1 [Da , Db ]Dc
4
3

2
+ TrN1 Da2 + M TrN1 Da2 .
SN1 [Da ] =
The operators ij are given explicitly by

ij = 2 + (2M + N )Da2 ij + (2M N )Di Dj 2i 2 Fij .
(37)
(38)
156
From Eqs. (30) and (36) it is quite clear that the (N 1)-dimensional vectors vi play exactly the
role of (bosonic) quark fields moving in the background of a covariant U (1)N1 gauge field Da .
On the other hand the logarithmic potential (Eq. (27)) takes the form

log(A B )2 =
A<B
By integrating out
log(dA dB )2 +
N1

A<B
vi , vi
log(dA 3 )2 .
(39)
A=1
eff [D , d ] given by
and a we obtain the effective action SN1
i a
eSN1 [Di ,dA ] = eSN1 [Da ]+

eff

A<B
log(dA dB )2

2

+
dvi [dvi ]evi ij vj
i=1

2
2
4 N1
2
i2 Mi4
d3 e3 M3 + A=1 log(dA 3 ) .
[di ]e
(40)
i=1
The effective action reads therefore

eff
[Di , dA ] = SN1 [Da ]
SN1
log(dA dB )2
A<B

2
4 N1
2
log
d e M + A=1 log(dA )

+ Tr2 TrN1 log 2 + (2M + N )Da2 ij

+ (2M N )Di Dj 2i 2 Fij .
(41)
Tr2 is the 2-dimensional trace associated with the remaining U (1) rotational symmetry of the
two matrices D1 and D2 (since D3 is treated differentlydiagonalizedin this approach) and
TrN 1 is the usual trace over the matrices; here D1 and D2 are (N 1) (N 1) matrices.
The sum of the first two terms in (41) is nothing but the action (27) with the replacement
a Da , A dA and TrN TrN1 . Thus

eff
2 M 4 + N1
log(dA )2
A=1
SN1 [Di , dA ] = SN1 [Di , dA ] log
d e


(42)
This action has U (1)N1 gauge symmetry and a U (1) rotational symmetry since the matrix D3
is diagonal. In the partition function the gauge symmetry can be easily enlarged to U (N 1) by
rotating the diagonal matrix D3 (back) to a general form C3 given by D3 = U C3 U + where U
is an (N 1) (N 1) unitary matrix. We will have the simultaneous rotations Di = U Ci U + .
The action SN1 [Di , dA ] becomes given by (13) with the replacement Xa Ca and where the
trace is normalized such that Tr 1 = N 1. We write the above result in the following suggestive
form

2
eff
2 M 4 + N1
log(d
)
A
A=1
d e
SN1 [Da ] = log


(43)
157
eff is precisely the one-loop contribution to the classical action (13) coming from integrating
SN
1
out from the model only one row and one column. In the large N limit we can treat N as a continuous variable and thus we can simply obtain the full one-loop contribution to the classical
action (13) by integration over N of the above result.
3.2. The S2N -to-matrix phase transition

We are interested in particular in verifying the stability of the fuzzy sphere ground state (15)
under quantum fluctuations. We consider therefore the background Da = La where La are the
generators of SU(2) in the irreducible representation N2
2 which are of size (N 1) (N 1).
is the field associated with the fluctuations of the radius R = . The classical potential from (37)
is given by

1 3
N 2 2 2
N 2
4 1 4
2c2

c2
+
V [] =
N +1
4
3
N +1

N 2 (N 1)2 1 c22 4 M 4
+
N + 1 N2 1
N

N 2 4 1 + m2 4 1 3 m2 2
=

+ O(N),
2
4
3
2
(44)
2
where we have used the relations = N , = 2 m2 and M = N2cm2 . We have also kept terms
of order N 2 only which dominates in the large N limit. Since D3 = L3 and (L3 )AB = mA AB
where mA = N2 A we have dA = mA . Thus the last integral in (40) becomes in the large N
limit

e
(2N1) log
d e
4 m2 2 2 m2 4 4 4 N1
2c N + A=1 log( N2 A)2
N
2
ce(2N1) log .
(45)
The constant c is independent of the field . We also compute in the large N limit

N 2 2N
2 2
+
m
c
2 + (2M + N )Da2 = 2 2m2 +
2
N2 1

2

2m2 2
1
N 2 N 2
= 2
2m2 + m2 2
+O
,
4
2
N
N2
(46)

N 2 2N 2
m c2 2 x i x j
2
N 1

2m2
1
N2 N
2
2
=
+ +m
+O
2 xi xj ,
4
2
N
N2
(2M N)Di Dj = 2
xi =
2Li
N 2 2N
(47)
158

N 2 2N
2i Fij = 2i c2
2 ij 3 x3
2
N 1

1
1
2
= i N 1
+O
2 ij 3 x3 ,
2N
N2
2L3
x3 =
(48)
.
N 2 2N
The leading quantum contribution of the effective potential from (43), (45) and (46)(48) is given
by
2 2 2

N
eff
(ij xi xj )
VN1 = (2N 1) log + Tr2 TrN1 log
4
= (2N 3) log + a constant independent of .
(49)
2
As we will check in Appendix A all higher order terms in Eqs. (46)(48) will give vanishingly
small quantum contributions in the large N limit. Thus (49) is the one-loop correction of the
effective potential coming from integrating out one column and one row from the theory. The full
one-loop correction coming from integrating out all columns and rows is obtained by a simple
integration over N . We get

eff
VN1
(50)
= N 2 3N log .
By putting (43) and (50) together we get the effective potential

N 2 4 1 + m2 4 1 3 m2 2
Veff =
(51)

+ N 2 log + O(N).
2
4
3
2
Let us point out here that this potential was derived elsewhere using a completely different (much
simpler) argument involving gauge fixing the original action (13) and then computing the full
one-loop effective action in the background field method. The argument in this article is however
superior from the point of view that it (manifestly) preserves gauge symmetry at all stages of the
calculation since we are not fixing any gauge in the usual sense [22]. See also [23].
It is not difficult to check that the corresponding equation of motion of the potential (51)
admits two real solutions where we can identify the one with the least energy with the actual
radius of the sphere. This however is only true up to a certain value of the coupling constant
where the quartic equation ceases to have any real solution and as a consequence the fuzzy
sphere solution (15) ceases to exist. In other words the potential Veff below the value of the
coupling constant becomes unbounded and the fuzzy sphere collapses. The critical value can be
easily computed and one finds

3
32m2 (1 + m2 )
=
(52)
1+ 1+
9
8(1 + m2 )
and

1
1
m2 2
1
= 1 + m2 4 + 3 +
.
4
2
2
2

Extrapolating to large masses we obtain the scaling behavior
1
4
8
.
=
2
m + 21
(53)
(54)
159
Fig. 1. The phase diagram of the S2N -to-matrix phase transition. The fuzzy sphere phase is above the solid line while the
matrix phase is below it. In this figure s is the Monte Carlo measurement of the critical value .
In other words the phase transition happens each time at a smaller value of the coupling constant
and thus the fuzzy sphere is more stable. This one-loop result is compared to the nonperturbative result coming from the Monte Carlo simulation of the model (13) with the constraint (17)
in Fig. 1. As one can immediately see there is an excellent agreement. In this sense the one-loop
result for the U (1) is exact. Let us finally report that this phase transition was also observed in
4 dimensions on S2L S2L . See the first reference of [15].
4. The small one-plaquette model limit on S2N
We have found in the one-loop calculation as well as in numerical simulation that the presence
of the normal field (21) is what causes the model to undergo the above first order phase transition from the fuzzy sphere to a matrix phase where the fuzzy sphere collapses under quantum
fluctuation. At the level of perturbation theory of the gauge field Aa this shows up in the form
of a compact UVIR mixing phenomena which goes to the usual singular UVIR mixing on the
NC plane in some appropriate planar limit of the sphere. In the large m limit we also have shown
that these two (possibly related) effects disappear [22].
As it turns out there is some signature in Monte Carlo simulation of the model (13) with the
constraint (17) for the existence of another kind of phase transition which seems to be unrelated
to the S2N -to-matrix phase transition and which generically persists even in the large m limit.
This latter phase transition resembles very much the third order one-plaquette phase transition in
ordinary two-dimensional gauge theory. In particular the agreement in the weak regime between
the simulation and the theory (which we will present now) is excellent. Let us also say that this
transition starts to appear when the critical value as m increases becomes less than the value
3.35
and it becomes more pronounced as decreases further away from this value. The new
N
phase transition thus occurs at
3.35
= .
N
(55)
160
Our goal in this section is to give a detailed theoretical model which describes this transition.
This construction is motivated by [24,26].
We start by making the observation that in the large m limit we can set = 0 as one
can immediately see from the action (22) and the partition function (24). Indeed we have for
m

2
2m
2 Tr
g N
g2N
2m2
N2
2
().
(56)
In other words the normal scalar field becomes infinitely heavy (m is precisely its mass)
and thus decouples from the rest of the dynamics. Hence we can effectively impose the extra
constraint Xa2 = 2 c2 on the field Xa in this limit m . In terms of Da = La + Aa = Xa this
constraint reads
=
Da2 c2 1
1
= {xa , Aa } + A2a = 0.
2
2 c2
2 2
The action (22) (if (17) is also satisfied) becomes

2

N2
g N
S = SYM + SCS + S0 +
log
.
2
2m2
(57)
(58)
The YangMills and ChernSimons-like actions are given respectively by

1
2
Tr Fab
,
4g 2 N

1
SCS = 2 Tr abc Fab Dc + Da2 c2
6g N
1
= 2 Tr abc Fab Dc when m .
6g N
SYM =
(59)
In above g 2 = 14 and S0 = S[Aa = 0] = 16 4 c2 12 4 c2 m2 . In the continuum large N

limit the constraint Xa2 = 2 c2 becomes the usual requirement that the normal component of the
gauge field on the sphere is zero, viz. = na . Aa = 0. Moreover the ChernSimons-like action
vanishes in this limit by this same condition = na . Aa = 0 because it will involve the integral
of a 3-form over a 2-dimensional manifold. Hence S = SYM in the large N limit provided
we also impose the condition = 0. In summary if we take the limit m first and then we
take the continuum limit N we obtain a U (1) action on the ordinary sphere, viz.
2

N2
g N
1
d
S S0 =
(60)
log
(iLa Ab iLb Aa + abc Ac )2 .
+ 2
2
2
4
2m
4g
By construction the continuum gauge field (which should be easily distinguished from its corresponding operator on the fuzzy sphere although we are using the same symbol Aa for both
quantities) is strictly tangent. See Appendix B for more detail.
If we study instead the action (23) in the limit m first and then in the continuum limit
N then we will find the action

1
d
4 2
S S0 0 = 2
(61)
(iLa Ab iLb Aa + abc Ac )2 .
4
4g
161
Again the continuum gauge field Aa is strictly tangent. The only difference with the previous
case is the extra piece 4 02 which we pulled out from the action in the process of writing it only
in terms of a tangent gauge field.
We now relate this action with the one-plaquette action. To this end we introduce the 2N 2N
idempotent
1
(62)
2 =1
(12N + 2a La ),
N
where a are the usual Pauli matrices. It has eigenvalues +1 and 1 with multiplicities N + 1
and N 1, respectively. We introduce the covariant derivative Da = La + Aa through a gauged
idempotent D as follows
=
1
D = ,
2
8 c2
2
2
1
+ 2 abc c Fab .
= (1 + 2a Da ) = + a Aa , 2 = 1 +
(63)
N
N
N2
N
Since we are interested in the large m limit we may as well set = 0 in above. Clearly
D has the same spectrum as . Thus it is a continuous deformation of in the sense that there
exists a U (2N) unitary transformation U such that D = U U + . Furthermore if U U T where
T U (N + 1) or T U (N 1) then T T + = and as a consequence D D . So D
is an element of the dN -Grassmannian manifold U (2N )/U (N + 1) U (N 1). We compute
the dimension dN as follows
dN = 4N 2 (N + 1)2 (N 1)2 = 2N 2 2.
(64)
This is exactly the correct number of degrees of freedom in a gauge theory on the sphere without
normal scalar field or with a normal scalar field frozen to some fixed value. The 2 counts the zero
modes which decouple because of commutators.
The original U (N ) gauge symmetry acts on the covariant derivatives Da as
Da = gDa g + ,
g
g U (N ).
(65)
This symmetry will be enlarged to the following U (2N ) symmetry. We introduce a tentative link
variable W (a 2N 2N unitary matrix) by W = D . The extended U (2N ) symmetry will then
act on W as follows W V W V + , V U (2N ). It is clear that this transformation property of
W can only be obtained if we impose the following transformation properties V V + and
D V D V + on and D , respectively. Hence the U (N ) subgroup of this U (2N ) symmetry
which will act on Da as Da gDa g + will also have to act on La as La gLa g + , i.e.
Da = gDa g + ,
g
La = gLa g + ,
g
g U (N ).
(66)
It is not difficult to see that the two sets of gauge transformations (65) and (66) are identical if
we are looking at the action (13) since it only depends on Xa = Da and not on La . However
for the gauge field Aa there is certainly a difference between the two sets of transformations (65)
and (66). Under (65) we have Aa gAa g + + g[La , g + ] whereas under (66) we have Aa
gAa g + . The actions as written in (22) and (23) are invariant only in the first case.
We want thus to modify the definition of the link variable W so that we have (65) and not (66).
In other words under this new definition the fixed background La will not rotate whereas the
gauge field Aa will transform correctly as Aa gAa g + + g[La , g + ]. Towards this end we
162
introduce another covariant derivative Da = La + Aa through the gauged idempotent D given
by
D =
=
1
2
8 c2
1
2
2

+ 2 abc c Fab
.
(1 + 2a Da ) = + a Aa , 2 = 1 +
2
N
N
N
N
(67)
As before we will also set = 0. From the two idempotents D and D we construct the link
variable W as follows
W = D D .
(68)
The extended U (2N) symmetry will then act on W as follows

W V W V +,
V U (2N ).
(69)
This transformation property of W can only be obtained if we impose the following transformation properties D V D V + and D V D V + on D and D , respectively. Hence the
U (N ) subgroup of this U (2N ) symmetry which will act on Da as Da gDa g + will also have
to act on Da as Da gDa g + , i.e.
Da = gDa g + ,
g
g
Da = gDa g + ,
g U (N ).
(70)
Under these transformations the gauge fields Aa and Aa transform as Aa gAa g + + g[La , g + ]
and Aa gAa g + + g[La , g + ] respectively like we want.
Remark also that for every fixed configuration Aa the link variable W contains the same
degrees of freedom contained in D . To see this we will go to the basis in which D is diagonal,
viz.

1N+1
0
D =
(71)
.
0
1N1
In this basis D and W will have the following generic forms

W1 W12
W1
W12
D =
,
W=
.
+
+
W12
W2
W12
W2
(72)
W1 = W1+ is an (N + 1) (N + 1) matrix, W2 = W2+ is an (N 1) (N 1) matrix and W12 is

+
an (N + 1) (N 1) matrix whereas the hermitian adjoint W12
is an (N 1) (N + 1) matrix.
2
+
Since D = 1 or equivalently W W = 1 we must also have the conditions
+
W1+ W1 + W12 W12
= 1,
+
W2+ W2 + W12
W12 = 1,
W1 W12 + W12 W2 = 0.
(73)
Knowing W12 will determine completely the matrix W (or equivalently D ) and hence we have
2(N + 1)(N 1) = 2N 2 2 degrees of freedom which agrees with (64).
163
4.1. The coordinate transformation (A1 , A2 , A3 ) (W, )

The main idea is that we want to reparametrize the gauge field on S2L in terms of the fuzzy
link variable W and the normal scalar field . In other words we want to replace the triplet
(A1 , A2 , A3 ) with (W, ). It is the link variable W which contains the degrees of freedom of the
gauge field which are tangent to the sphere as is shown by the result (64). Thus in summary we
have the coordinate transformation
(A1 , A2 , A3 ) (W, ).
(74)
First we need to show that we have indeed the correct measure. Namely one must show that we
have

dA1 dA2 dA3 = cN dW d,
(75)
where cN is some constant of proportionality which can only depend on N . In order to compute
the measure we will compute the quantity Tr2N (dW )+ dW where Tr2N denotes the (2N 2N )dimensional trace. For this exercise the scalar field will not be assumed to be fixed whereas
the other gauge configuration Aa and its corresponding normal scalar field are supposed to
be some constant backgrounds. From the definition W = D D and Eqs. (63) and (67) one can
easily compute

2
2
2
1
2

W = 1 + a Aa + a Aa + O
N
N
N
N
N2
(76)
or equivalently
dW =

2
2
1
a dAa d + O
.
N
N
N2
(77)
Hence a straightforward calculation yields the measure

Tr2N (dW )+ dW =
8
8
Tr(dAa )2 + 2 Tr(d)2
2
N
N

16
1
3 Tr dd(La Aa + Aa La ) + O
.
N
N3
(78)
By using the identity 2 c2 = Aa La + La Aa + A2a we arrive at the result

Tr2N (dW )+ dW =

8
8
1
2
2
Tr(dA
)
Tr(d)
+
O
.
a
N2
N2
N3
(79)
The correct (more suggestive) way of writing this equation is the following
Tr(dAa )2 =

N2
1
Tr2N (dW )+ dW + Tr(d)2 + O
.
8
N
(80)
In the large N limit it is obvious that this equation implies (75) which is what we desire.
164
4.2. The U (1) gauge action as a linear one-plaquette model

It remains now to show that the enlarged U (2N ) symmetry reduces to its U (N ) subgroup in
the large N limit. The starting point is the 2N -dimensional one-plaquette action with a positive
coupling constant , viz.

N
Tr2N W + W + 2
with the constraints

SP =
W = D D ,
(81)
Da2 c2
.
2 c2
We have the path integral

2
ZP = cN dD d ( )
(82)

dW d ()eSP .
(83)
W =D D
2 is the constant which appears in (75). The extra integrations over and (in other words
cN
D
over Da ) is included in order to maintain gauge invariance of the path integral. The integration
over W is done along the orbit W = D D inside the full U (2N ) gauge group. In above we
have also to integrate over configurations Da and Da such that = 0 and = 0 since we are
only interested in the limit m of the model (23). Furthermore we can conclude from the
result (75) that in the large N limit this path integral can be written as

ZP = dAa ( )
(84)
dAa ()eSP .
W =D D
We need now to check what happens to the action SP in the large N limit. This is done in
Appendix C and one finds

2
N
16
1
32
2
SP =
(85)
2 Tr Aa + 4 Tr i[La , Ab ] i[Lb , Aa ] + abc Ac + O
.
N
N
N5
The constraints Da2 = c2 and Da 2 = c2 (or equivalently = 0 and = 0) become in terms of
the variables A a = 12 Aa 12 Aa and Sa = Da + Da (or equivalently A a = 12 Aa + 12 Aa )
Sa2 + 4A 2a = 4c2 A 2a + {La , A a } + A 2a = 0,
{Sa , A a } = 0 {La , A a } + {A a , A a } = 0.
(86)
In the continuum limit these two constraints becomes na A a = and na A a = 0, respectively. By

using the first constraint we can rewrite the action in the form
N 64
32
SP =
Tr A a La + 2 Tr A 2a
2
N
N

2
16
1
+ 4 Tr i[La , Ab ] i[Lb , Aa ] + abc Ac + O

(87)
.
N
N5
The leading contribution in the action SP as written in Eq. (85) is a simple Gaussian which is
clearly dominated by the configuration A a = 0. As a consequence the full path integration over
165
A a is dominated by A a = 0. This yields a zero action which is obviously not what we want.
Furthermore the path integration over A a diverges since this action (85) does not depend on
these matrices. On the other hand the Gaussian term becomes in equation (87) (after using the
constraint) a quadratic integral over the matrices A a but with a wrong sign since the first term
converges to 0 in the limit (see Appendix B). Thus the path integration over the three matrices A a
will again diverge. The one-plaquette action SP by itself is therefore not enough to obtain a U (1)
action on the sphere in the continuum large N limit.
4.3. A quadratic one-plaquette action
Towards the end of constructing a U (1) action on the fuzzy sphere using the one-plaquette
variable W we add to the action SP the following quadratic one-plaquette action (where is the
corresponding coupling constant)
SP =

N
Tr2N W 2 + W +2 2 .

(88)
Remark the extra minus sign in front of this action, i.e. is a positive coupling constant. As
before we need now to compute the large N limit of this quadratic one-plaquette action. This is
also done in Appendix C and one finds the result
2
N 256
128
512
SP =
Tr A a La + 2 Tr A 2a + 4 Tr A 2a + {A a , La }
2
N
N
N

2

64
1
+ 4 Tr i[La , A b ] i[Lb , A a ] + abc A c + O
(89)
.
N
N5
By putting the one-plaquette actions (87) and (89) together we obtain the total one-plaquette
action

2

512
32
SP + SP =
Tr {A a , La } + A 2a 3 Tr {A a , La } + A 2a
1 N
N
2

16
Tr i[La , A b ] i[Lb , A a ] + abc A c

1 N 3

1
1
+O
(90)
O
.
N 4
N 4
The positive coupling constant 1 is defined in terms of and by
1
1
4
= .
1
(91)
The effect of the dominant terms (the first two terms in the above action) is now precisely what
we want. The path integral over the three matrices A a is given by

32
3N 2

d Aa ()( ) exp
Tr {A a , La } + A 2a
(Aa ) = 2
1 N

512
2 2
3 Tr {Aa , La } + Aa
N

1
1
2N 2
d Aa {xa , Aa } + {Aa , Aa }
=2
2
2 c2
166

1 2 1
1 2
Aa + {xa , Aa } + Aa exp{same}
2
2 c2
2 c2

1
1
2
d A a {xa , A a }
22N {xa , A a }
2
2

32
512 2
exp
Tr A 2a 3 Tr A 2a
.
1 N
N
(92)
In the large N limit the first term in the exponent dominates (see below) and as a consequence the
path integral over the three matrices A a becomes a simple Gaussian. Since the second constraint
inside the integral has the effect of reducing the number of independent matrices A a to just two
we obtain the final result

1
1
32
2
d A a {xa , A a } exp
(A a ) 22N {xa , A a }
Tr A 2a
2
2
1 N

N 2
1
N1
2
.
22N {xa , A a }
(93)
2
32
Another (more correct) way of understanding this result is to note that this path integral is dominated in the large N limit by the configurations A a = 0.
The path integral of the one-plaquette model (with an action SP + SP ) becomes in the large
N limit as follows

eff
1
ZP = d A a {xa , A a } eSP ,
(94)
2
where

N1
16
b ] i[Lb , A a ] + abc A c 2
SPeff = N 2 log
Tr
i[L
,
A
a
8
1 N 3

1
1
+O
O 4 .
N 4
N
(95)
Notice that this action is invariant not only under the trivial original gauge transformation law
A a A a but also it is invariant under the nontrivial gauge transformation A a A a + g[La , g + ]
where g U (N ). This emergent new gauge transformation of A a is identical to the transformation property of a U (1) gauge field on the sphere. Therefore the action SPeff given by the above
equation is essentially the same U (1) action (S S0 ) given in Eq. (60) provided we make the
following identification

16
4
4
4
1
1
16
= 2
(96)

2
2

4
N 1 N
4g
4N 2
between the U (1) gauge coupling constant g on the fuzzy sphere and the one-plaquette model
coupling constant 1 . The action becomes

8g 2
1
d
2
(iLa A b iLb A a + abc A c )2
SPeff = N 2 log
N
4
4g

1
1
+O
(97)
O
.
N 4
N 4
167
Let us remark that in this large N limit in which g is kept fixed the one-plaquette coupling constant 1 goes to zero. Hence the fuzzy sphere action with fixed coupling constant g corresponds
in this particular limit to the one-plaquette gauge field in the weak regime and agreement between the two is expected only for weak couplings (large values of ).
To see this more clearly

we notice that in terms of and the limit 1 = 4 0 is equivalent to the limit 0
for fixed or vice versa, i.e. to the limit 0 for fixed . Furthermore 1 going to 0 is also
equivalent to the limit when both and go to zero. Clearly all these possibilities correspond
to the one-plaquette gauge field in the weak regime.
Finally we remark that the constant term in (60) depends on the mass parameter m. Thus by
comparing between the constant terms in (60) and (97) we can determine m2 as a function of g 2
3 6
3 5
(or equivalently 4 ) and N . We find m2 = 32N g = 32 12N .
4.4. The one-plaquette path integral
Instead of (83) we will therefore consider in the remainder of this article the following (corrected or generalized) one-plaquette path integral

2
dD d ( )
dW d ()eSP +SP .
ZP = cN
(98)
W =D D
In analogy with (72) we decompose the 2N 2N matrices D D and W as follows

(D D )1
W1
(D D )12
W12
D D =
,
W
=
.
+
(D D )+
W12
W2
12 (D D )2
(99)
In particular W1 = W1+ is an (N + 1) (N + 1) matrix, W2 = W2+ is an (N 1) (N 1) matrix

+
and W12 is an (N +1)(N 1) matrix whereas the hermitian adjoint W12
is an (N 1)(N +1)
matrix. Since W + W = 1 we have the conditions
+
W1+ W1 + W12 W12
= 1,
+
W2+ W2 + W12
W12 = 1,
W1 W12 + W12 W2 = 0.
(100)
First we observe that in this basis the metric becomes

Tr2N (dW )+ dW = TrN+1 (dW1 )+ dW1 + TrN1 (dW2 )+ dW2
+ TrN+1 dW12 (dW12 )+ + TrN1 (dW12 )+ dW12 .
(101)
Hence we can immediately conclude that the path integral over W can be rewritten (by neglecting
an overall proportionality factor) as

+
dW
dW12
dW12
dW1
dW2 .
W =D D
W12 =(D D )12
+
W12
=(D D )+
12
W1 =(D D )1
W2 =(D D )2
(102)
Furthermore we can show that in this basis the actions SP and SP take the form
SP =

N

N
TrN+1 W1 + W1+ 2 + TrN1 W2 W2+ 2
(103)
168
and

N

N
TrN+1 W12 + W1+2 2 TrN1 W22 + W2+2 2

2N
2N
+
+
+ TrN+1 W12 W12 + TrN1 W12
W12 .
SP =
(104)
+
(as opposed to the diagonal matrices W1 and W2 )
Thus the off-diagonal matrices W12 and W12

appear only in the action SP .
Let us recall that since the integration over W is done along the orbit W = D D inside
U (2N) and since in the large N limit both D and D approach the usual chirality operator
= na a we see that W approaches the identity matrix in this limit. It is in this sense that W
yields in the continuum large N limit a small one-plaquette model.
Let us now explain how we will approximate the above path integral in the continuum
large N limit. From one hand we have the following limiting constraint W = D D 12N 2
which means that when N we have the behavior W1 = (D D )1 1N+1 , W2 =
(D D )2 1N1 and W12 = (D D )12 0. From the other hand since W must be always
+
a unitary matrix and since the off-diagonal parts W12 and W12
tend to zero the matrices W1 and
W2 become in this approximations (N + 1) (N + 1) and (N 1) (N 1) unitary matrices
respectively (which are close to the identity) in accordance with Eqs. (100). Let us also stress the
fact that the strict limits of W1 , W2 and W12 are independent of D . For example the matrix
W1 goes always to the same limit 1N+1 for all matrices D .
The main approximation which we will adopt in this article consists therefore in replacing the
constraint W = D D with the simpler constraint W 12N by taking the diagonal parts W1 and
W2 to be two arbitrary, i.e. independent of D , unitary matrices which are very close to the
+
identities 1N+1 and 1N1 respectively while allowing the off-diagonal parts W12 and W12
to go
to zero. We observe that by including only W1 and W2 in this approximation we are including
in the limit precisely the correct number of degrees of freedom tangent to the sphere, viz. 2N 2 .
Thus in this approximation the integrations over , and D decouple while the integrations
+
+
are dominated by W12 = W12
= 0. There remains the two independent path
over W12 and W12
integrals over W1 and W2 which are clearly equal in the strict limit since the matrix dimension
of W1 approaches the matrix dimension of W2 for large N . Thus the path integral ZP reduces
(by neglecting also an overall proportionality factor) to

2
ZP ZP (, ) ,
(105)
where

ZP (, ) =

N 2

N
+
+2
dW1 exp
Tr W1 + W1 2 Tr W1 + W1 2 .
(106)
The path integral of a 2-dimensional U (N ) gauge theory in the axial gauge A1 = 0 on a lattice
2
with volume V and lattice spacing a is given by ZP (, )V /a where ZP (, ) is the above

partition function (106) for = , i.e. the partition function of the one-plaquette model Sp =
+
N
Tr(W1 + W1 2). Next we need to understand the effect of the addition of the quadratic one-
+2
2
V /a for
plaquette action Sp = N
Tr(W1 + W1 2). Formally the partition function ZP (, )

any value of the coupling constant can be obtained by expanding the model SP + SP around
2
2 Notice that W = goes to 1

2N independently of any basis.
D D
169
= . Thus it is not difficult to observe that the one-plaquette action Sp + Sp does also lead to
(a more complicated) U (N ) gauge theory in two dimensions. The U (N ) gauge coupling constant
g12 is simply given by
1
1
= N a4 .
2
g1
(107)
Therefore we can see that the partition function ZP of a U (1) gauge field on the fuzzy sphere is
proportional to the partition function of a generalized 2-dimensional U (N ) gauge theory in the
axial gauge A1 = 0 on a lattice with two plaquettes. This doubling of plaquettes is reminiscent
of the usual doubling of points in Connes standard model. The U (1) gauge coupling constant g 2
and the U (N ) gauge coupling constant g12 are related (from (96) and (107)) by the equation

g 2 N 3 a 4 4
=
1 .
(108)
64

g12
It is quite natural to require the two coupling constants g 2 and g12 to be equal which means we
must choose the lattice spacing a such that
a4 =
64
.
N 3 4
(109)
4.5. Saddle point solution

We are therefore interested in the N -dimensional one-plaquette model

N 2

N

+
+2
ZP (, ) = dW exp
Tr W + W 2 Tr W + W 2 .
(110)
Let us recall that dW is the U (N ) Haar measure. We can immediately diagonalize the link variable W by writing W = T DT + where T is some U (N ) matrix and D is diagonal with elements
equal to the eigenvalues exp(ii ) of W . In other words Dij = ij exp(ii ). The integration over
T can be done trivially and one ends up with the path integral
ZP (, ) =

N
di eN SN .
(111)
i=1

2N
The action SN contains besides the Wilson actions 1 Tr(W + W + 2) = 2 N
i=1 cos i and

N
1
2
2N
2
+2 2) =
i=1 cos 2i contributions coming from the usual Vandermonde
Tr(W +W

determinant. Explicitly the total action reads

i j 2 2N
2
2
1
2N
SN =
(112)
cos i
cos 2i +
ln sin
+ .
2N
2
i=j
In the large N limit we can resort to the method of steepest descent to evaluate the path inteN
gral ZP (, ). The partition function will be dominated by the solution of the equation dS
di = 0
which is a minimum of the action SN . Before we proceed to the solution we need to take into account the following crucial property. Since the link variable W tends to one in the large N
limit we can conclude that all the angles i tend to 0 in this limit and thus we can consider instead
170
of the full one-plaquette model action (112) a small one-plaquette model action by including corrections up to the quadratic order in the angles i . We obtain
SN =

1 2
1 (i j )2
i +
ln
+ O 4 .
2
2N
4
i
(113)
i=j
2 is given by
1
1
1
= + .
2
1 12
(114)
For the consistency of the solution below the coupling constant 1 must be negative (as opposed
to the classical model where 1 was assumed positive) and as a consequence the coupling constant 2 is always positive. As it turns out most of the classical arguments of Sections 4.2 and 4.3
will go through unchanged when 1 is taken negative.
Thus in the following quantum theory of the model we will identify the effective one-plaquette
action SPeff with the fuzzy sphere action S S0 (which is to be compared with the classical identification SPeff = S S0 ) and hence we must make the following identification of the coupling
constants
16
1
4
=
=
.
N 2 1 4g 2 4N 2
(115)
This is precisely due as we have said to the fact that 1 becomes negative in the quantum theory.
In the continuum large N limit where 4 is kept fixed instead of 1 we can see that 11 scales
with N 2 and as a consequence
64
.
N 2 4
The saddle point solution must satisfy the equation of motion
i j
2
4
1
cot
sin i sin 2i =
.
N
2
2 = 1 =
(116)
(117)
j =i
The equation of motion (117) takes (in the limit N when all the angles tend to zero) the
form
2i
2 1
(118)
=
.
2
N
i j
j =i
In order to solve the above problem we introduce the potential V ({i }) defined through its first
({i })
i
V (i ) = 2
derivative dVd
2 and also the N N matrix M defined through its eigenvalues i ,
i
i = 1, . . . , N . The trace (z) of the resolvent of M is given by
1
1
1 1
(z) = Tr
(119)
=
.
N
M z N
i z
i
The condition (118) can then be rewritten as follows

2 (z)
1
1 V (z) V (i )
.
(z) + V (z)(z) = R(z)
N
N
z i
i
(120)
171
In the large N limit we can also introduce

a density of eigenvalues ( ) which is positive definite

and normalized to one; ( ) > 0, d ( ) = 1 [N( ) is
the number
of eigenvalues in the range
[ d/2, + d/2]]. Thus the sum will be replaced by i = N d ( ) and one obtain
(z) + V (z)(z) = R(z)

2
d ( )
V (z) V ( )
.
z
(121)
The trace of the resolvent is now given by

(z) =
d ( )
1
.
z
(122)

The density of eigenvalues ( ) should satisfy d ( ) = 1 and ( ) 0 for all angles
. We can easily solve this problem since we can compute
R(z) =
2
2
(123)
(z) =
2i
22 z2 .
2
(124)
and
The solution of the equation of motion is immediately given by

z
1
1
i
(z) = V (z) (z) =
22 z2 .
2
2
2 2
(125)
The function (z) is a multi-valued function of z with branch points at z = z0 = 22 .

Since the potential V has only one minimum at = 0 the density of eigenvalues must have only
one support centered around this minimum. This support is clearly in the range between z0
and +z0 . In terms of the resolvent (z) the density of eigenvalues is defined by
(z) =
(z + i) (z i)
.
2i
(126)
(z + i) is the trace of the resolvent of M computed with a contour in the upper half complex
plane and we choose for it the plus sign, viz. (z + i) = + (z). Similarly (z i) is the trace
of the resolvent of M computed with a contour in the lower half complex plane and we choose
for it the minus sign, viz. (z i) = (z). We obtain therefore
( ) =
1
22 2 .
2
(127)
It
is obvious that
this density of eigenvalues is only defined for angles which are in the range
22 22 . However the value of the critical angle should be determined from the

normalization condition d ( ) = 1. This condition yields the value
=
22 .
(128)
172
4.6. The one-plaquette phase transition

It is quite obvious that the action (113) is an excellent approximation of (112) for all angles i
in the range
1
1
i .
2
2
(129)
The particular value 12 comes from the fact that the expansion of the quadratic one-plaquette
action SP will converge to the original expression only for small i in the above range. The
expansions of the linear one-plaquette action SP and of the Vandermonde action will converge
to the original expressions for i in the range 1 i 1.
The solution (127) with the critical angle (128) is then valid only for very small values of
the coupling constant 2 . Indeed it is only in this regime of small 2 where the fuzzy sphere
action with fixed coupling constant g is expected to correspond to the one-plaquette model as we
have discussed previously. However in order to find the critical value of 2 we need to extend
the solution (127) to higher values of 2 . To this end we note that the action (113) can also be
obtained from the effective one-plaquette model
Speff =
2
eff
2

2
2N
+
Tr Weff + Weff
2 = eff
cos ieff eff .
2 i
2
(130)
For small ieff in the range

1 ieff 1.
(131)
The total effective one-plaquette action becomes

eff
SN
=
1
eff
2
ieff
2
eff
eff 2
4
1 (i j )
ln
+ O eff .
2N
4
The action (132) must be identical to the action (113) and hence we must have (ieff )2 =
From the two ranges (129) and (131) we conclude that ieff = 2i and eff
2 = 42 .
The saddle point solution of the action (130) must satisfy the equation of motion
2
eff
2
sin ieff =
(132)
i=j
ieff jeff
1
cot
.
N
2
eff
2
2
2 i .
(133)
j =i
In the continuum large N limit this equation becomes

2
eff eff
(134)
sin
=
deff (eff ) cot
.
eff
2
eff
2

By using the expansion cot
n=1 (sin n cos n cos n sin n ) we can solve this equa2 =2
tion quite easily in the strong-coupling phase (large values of 2 ) and one finds the solution
(eff ) =
1
1
cos eff .
+
2
eff
2
(135)
173
However it is obvious that this solution makes sense only where the density of eigenvalues is
positive definite, i.e. for eff
2 such that
1
1
0
2
eff
2
eff
2
=2
2 = 0.5.
(136)
This strong-coupling solution should certainly work for large enough values of 2 . However this
is not the regime we want. To find the solution for small values of 2 the only difference with the
above analysis is that the range of the eigenvalues is now [ , + ] instead of [, +] where
is an angle less than which is a function of 2 . It is only in this regime of small 2 where the
fuzzy sphere action with fixed coupling constant g is expected to correspond to the one-plaquette.
In the strong regime deviations become significant near the sphere-to-matrix transition. Finding
the solution in the weak-coupling phase for the effective action (132) is a more involved exercise.
This is done in [25] with the result

eff
2
eff eff
2
(eff ) =
(137)
cos
sin2
,
eff
2
2
2
2

eff
eff
2
sin
(138)
=
.
2
2
It is very easy to verify that the this density of eigenvalues and critical angle will reduce to the
solution (127) and the critical angle (128) when the angles are taken to be very small.
The above computed critical value 2 = 0.5 leads to the critical value of the coupling constant
4 =
64
= 128
2
= 3.36,
(139)
which is to be compared with the observed value

= 3.350.25.
(140)
Indeed for U (1) theory we observe in Monte Carlo simulation of the model (13) with the relation (17) the value = 3.350.25 (or equivalently the value 2 = 0.51 0.15). Indeed for
very large values of the mass parameter m we observe two critical lines (see Fig. 2); the lower
line is the S2N -to-matrix critical line discussed previously. This line comes from the measurement
of the critical value s = from the action. The upper line is the one-plaquette critical line
which we can fit to the curve
0.04
p =
m2
1
2
+ 3.350.25.
(141)
Remark that this curve saturates in the limit m around

the value 3.35. The points p on
Fig. 2 comes from the measurement of the position = N of the peak in the specific heat
which for large values of the mass captures the one-plaquette phase transition. For even larger
values of m the peak disappears and in this case p measures the position where the specific heat
jumps discontinuously to the value 1.
174
Fig. 2. The phase diagram of the one-plaquette phase transition.
4.7. The specific heat and effective potential in 1/N expansion

We are now in a position to compute the quadratic average Q defined by the equation
1 2
Q=
(142)
i .
N
i
We obtain in the fuzzy one-plaquette solution (127) the result

Q=
d ( ) 2 =
2
.
2
(143)
We need also to compute the nonlocal average

i j 2 1
1
2
ln
=
d ( )
d () ln
QN-L =
2
2
2
2N 2
i=j
1 2 S 1
= ln
+ .
2
2
2
S1 is a constant of integration given explicitly by S1 =
ln(x y)2 . The action (113) is therefore given by
1 1 2 S 1
1
SN
= Q + QN-L = + ln
+ .
N
2
2 2
2
2
(144)
4
2
1
1 dx
1
1 x 2 1 dy
1 y2
(145)
Let us recall from Eqs. (105) and (106) that we have actually two identical one-plaquette models
and hence the above action must be multiplied by a factor of 2. Furthermore by comparing
between (94) and (111) we can see that SPeff must be identified with N SN (or twice as much due
to the above factor of 2) whereas we have found that the action S S0 on the fuzzy sphere must
175
Fig. 3. The action for nonzero mass. The fit is given by the second term of Eq. (146).
be identified in the quantum theory with SPeff . In other words the effective action on the fuzzy
sphere is given by

2
2
S = S0 + N 1 + ln
+ S1
2

1
1
N2
= 4 c2 4 c2 m2 N 2 ln 4 + N 2 1 ln
(146)
+ S1 .
6
2
32
In above we have also used equation (116). It is interesting to compare this effective action with
the original effective action (51) obtained in the one-loop. If we set = 1 in (51) then we will
find the same classical action as in the above equation, namely 16 4 c2 12 4 c2 m2 . However
the quantum correction in (51) in terms of is by inspection given by N 2 ln which is different
from the quantum correction in the above equation which is equal to 4N 2 ln .
We also note
that in the large m , then large N limit the above action will be dominated by the
classical mass-dependent term 12 4 c2 m2 . This is precisely what we observe in Monte Carlo
simulation. See Fig. 3.
Finally we need to compute the specific heat. Towards this end we implement the scaling
4
transformations S TS and 4 T . The specific heat is then defined by

2
3 S
.
Cv = T
(147)
T 2 T =1
A straightforward calculation yields the very simple result
Cv = N 2 .
(148)
Again this is what we observe in our numerical simulation of the U (1) gauge field on the fuzzy
sphere in the weak regime. In the strong regime deviations are significant near the sphere-tomatrix transition. See Fig. 4. In this regime of strong couplings the action and specific heat are
176
Fig. 4. The specific heat for very large values of m.
computed using the distribution of eigenvalues (137). We find

1
1
1
S = S0 + N 2 + 2
+ S1
2 82 2

4
1 4
N 2 4 2
1 4
1
2
2
2
N
+ N + S1 ,
= c2 c2 m +
6
2
2 128
64
2
(149)
and

Cv = N 2
4
128
2
.
(150)
We observe then that in the weak regime the specific heat is essentially given by 4c2 = N 2 1
within statistical errors whereas in the strong regime the data does only follow the theoretical oneplaquette prediction away from the S2N -to-matrix transition. This is presumably due to the effects
of the matrix phase which becomes strong near the critical S2N -to-matrix transition. Remark that
the minimum of the specific heat is where the S2N -to-matrix transition happens for large values
of m.
Let us comment further on the quantum effective potential for U (1) gauge field on the fuzzy
2 in this 1/N expansion of the one-plaquette model. As we have said before by comsphere SN
paring the 1/N effective potential (146) with the one-loop effective potential (51) in which we
set = 1 we can see that the classical contribution is the same in both potentials whereas the
quantum correction in (51) in terms of is given by N 2 ln which is different from the quantum
correction 4N 2 ln in (146). This observation allows us to rewrite (or to guess that) equation (146) (should be rewritten) in terms of the radius as follows

N 2 4 1 + m2 4 1 3 m2 2
S=
(151)

4N 2 log .
2
4
3
2
177
Now in contrast to what we have done so far in this article we will choose the mass parameter m
to be proportional to N . The simplest most natural choice is m = mN
. The effects of the large
mass limit will then be included implicitly in the continuum large N limit. This is in fact what
was done in [24]. The above effective potential becomes

N 2 4 m
2 1 4 1 2
S=
(152)
4N 2 log .
2
4
2
It is very easy to verify that this potential admits a local minimum for all values of the coupling
constant .
The minimum value min is found to be given by

1 + 1 + 432m 2
2
min
(153)
=
.
2
In the limit m first and then N (considered in this article) we can see that m

and hence min = 1 as expected. The fuzzy sphere ground state (15) is extremely stable in this
limit and the 1/N potential S (as opposed to the one-loop potential (51)) is completely insensitive
to the S2N -to-matrix phase transition.
5. Conclusion
In this article we have shown explicitly that quantum noncommutative U (1) gauge field on the
fuzzy sphere S2N is equivalent (at least in the fuzzy sphere-weak coupling phase) to a quantum
commutative 2-dimensional U (N ) gauge field on a lattice with two plaquettes.
By using the structure of the fuzzy sphere we have constructed a 2N 2N matrix W given
by Eq. (68) which was shown to contain the correct number of degrees of freedom tangent to the
fuzzy sphere, namely 2N 2 degrees of freedom. The other N 2 degrees of freedom are contained
obviously in the normal scalar field . Indeed we have shown that the gauge field (A1 , A2 , A3 )
is equivalent to (W, ) and dA1 dA2 dA3 dW d. The fuzzy sphere action (22) (or equivalently (13)) written in terms of Aa can therefore be rewritten in terms of W and . We have
shown explicitly that in the limit where we can set = 0 this action (22) and the action SP + SP
given by (81) + (88) will tend to the same continuum limit, viz. a U (1) gauge theory on the
ordinary sphere. As a consequence the partition function of the fuzzy U (1) model in the limit
m can be given either by the large m limit of (24) or by Eq. (98).
Indeed the fuzzy partition function ZP given by Eq. (98) is our starting point. It is found to
be proportional to the partition function of a U (N ) model in the axial gauge A1 = 0 on a lattice
with two plaquettes given by Eqs. (105) and (106). We remark that the U (N ) theory consists of
the canonical one-plaquette Wilson action Sp = N Tr(W1 + W1+ 2) plus a novel quadratic one+2
2
plaquette action Sp = N
Tr(W1 + W1 2) together with the canonical measure dW . This is
in fact the reason why W is called a link variable. The quadratic term was needed in order that
the fuzzy sphere one-plaquette path integral ZP converges to the sphere path integral in the large
N limit. Remark also that the effective actions Sp and Sp involve the N N link variable W1 as
opposed to the original actions SP and SP which involve the 2N 2N link variable W .
The doubling of plaquettes is a natural consequence of the model and it is reminiscent of the
usual doubling of points in Connes standard model. However the (other kind of) doubling of
U (1) fuzzy gauge fields was needed in order to have a gauge invariant formulation of the fuzzy
one-plaquette model. In fact a covariant plaquette variable W can only be constructed out of two
such U (1) fuzzy fields.
178
The main results are given by Eqs. (105) and (106). It is therefore of paramount importance to
find a more rigorous derivation of these two equations. Furthermore it will be very interesting to
show that the large m limit of the path integral (24) and the path integral (98) are also equivalent
for finite N . Their large N equivalence used in this article is confirmed in our Monte Carlo
simulation of the model (which uses (24)) where the measurement of the critical line = 3.36
and the specific heat can be understood in very simple terms using the limit (105) of (98). In
particular the value = 3.36 seen in the simulation is precisely the GrossWadiaWitten oneplaquette 3rd order transition point as calculated in this article from the path integral (105).
Since the plaquette variable W is small, i.e. it approaches 12N in the large N limit, we were
able to show that the model in this limit reduces to a simple matrix model and as a consequence
was easily solved. We computed the critical point and showed that it agrees with the observed
value. We computed also the quantum effective potential and the specific heat for U (1) gauge
2 in the 1/N expansion using this one-plaquette model. In particular
field on the fuzzy sphere SN
the specific heat was found to be equal to 1 in the fuzzy sphere-weak coupling phase of the gauge
field which agrees with the observed value 1 seen in Monte Carlo simulation. The value 1 comes
precisely because we have two plaquettes which approximate the noncommutative U (1) gauge
field on the fuzzy sphere. In the fuzzy sphere-strong coupling phase deviations were found to be
significant near the S2N -to-matrix critical point. It will be very interesting to be able to extend this
one-plaquette model to these large values of the gauge coupling constant g 2 , i.e. to small values
of .
The key to this we believe lies in improving the basic approximations of this article given
in Eqs. (105) and (106).
The most natural generalization of this work should include fermions in two dimensions [30]
and as a consequence take into account topological excitations [29]. The best example which
comes to mind is the Schwinger model [31]. Then one should seriously contemplate going to 4
dimensions with the full might of QCD. Early steps towards this larger goal were taken in the
first reference of [15]. First we need to have a complete control over the phase diagram of the
pure gauge model considered in this article [12].
Acknowledgements
The author Badis Ydri would like to thank Denjoe OConnor, P. Castro-Villarreal and
R. Delgadillo-Blando for their extensive discussions and critical comments while this research
was in progress.
Appendix A. Next-to-leading correction of the effective potential
The next-to-leading contribution (coming from the terms of order N and of order 1 in
Eqs. (46)(48)) is given by

2
8m2
4m2
1
Tr2 TrN1 log 1 2 (ij xi xj ) + 2 1 2 ij
N
N
N

4i
1
1
+
(A.1)
1
1
ij 3 x3 .
N
N
The -dependence is only in the second and third terms inside the logarithm. Let us also remark
that the inverse of the operator
Lij = aij + bxi xj
(A.2)
179
is given by
L1

ij
1
1
1
= ij + xi 2
xj .
a
a x3 1 ab
(A.3)
then we can see that the inverse L1 does not exist because of the zero
If a = b = 1 N2 4m
N2
eigenvalue of x3 . This can be traced to the fact that the rotational U (1) symmetry (unlike gauge
symmetry) cannot be restored back to the original full SU(2) invariance.
Thus we regularize L as follows

2
4m2
Lij = 1 2 (ij xi xj ).
(A.4)
N
N
Furthermore the value of in the matrix phase is expected to be very close to the classical
value 1. This means in particular that 1 1 is a small number and thus it is a very good expansion
parameter in this model. The vertex is given by

8m2
1
1
1
Vij =
(A.5)
1 2 ij + 4i 1
1
ij 3 x3 .
N
N
This is small both in N1 expansion and in 1 1 expansion. We need to evaluate

1
Tr2 TrN1 log 1 + L1 V
N
r+1
(1)

1
=
TrN1 L1 V i i L1 V i i L1 V i i .
r
r 1
1
2
2
3
r
N
(A.6)
r=1
1
Since (1) L1 and V are (N 1) (N 1) ordinary matrices, (2) the trace N1
TrN1 becomes
the ordinary integral in the limit and (3) the operators xa go over to the coordinates na on the
sphere we conclude that the only non-vanishing term in the N limit is the order r = 1 term
in the above equation. We have

1
Tr2 TrN1 log 1 + L1 V
N

1
1
1
1
1
= TrN1 L V i i + = 4i 1
xj x3 +
ij 3 TrN1 xi 2
1
1
N
N
x3 1 +

x2 1
16
1
= 2 1
(A.7)
TrN1 2 3 2 + .
N
x3 1 +
This term is clearly going to zero in the limit. In particular the contribution of the zero eigenvalue
1
of x3 is going to zero as N12 1
.
We can check that the quantum correction of the effective potential coming from the terms of
order O( N1 ) in equations (46)(48) are also going to zero in the limit. Hence the full quantum
correction to the effective potential is given by (49).
Appendix B. The star product on S2L
The coherent states on S2 are constructed as follows. Let us introduce the 2-dimensional rank
one projector P 1 = 12 + na 2a . The requirement P 12 = P 1 implies the condition n2 = 1. At the
2
180
north pole we have n0 = (0, 0, 1) and the projector becomes P 10 which projects onto the state
2

|
n0 , 12 = 10 . In other words P 10 = |
n0 , 12
n0 , 12 |. A generic point n on S2 is obtained by the
2
rotation g such that n = g n0 . The corresponding state is |

n, 12 = g|
n0 , 12 and the corresponding
projector is precisely P 1 which can also be rewritten as P 1 = |
n, 12
n, 12 |.
2
The irreducible representation L2 of SU(2) can be obtained from the symmetric product of L
copies of the fundamental representation 12 . The L2 -representation of the element g SU(2) is
L
given by the matrix U ( 2 ) (g) defined by

L
U ( 2 ) (g) = g s s g,
(B.1)
L times.
The (L + 1)-dimensional rank one projector P L which defines the

2
L
2 -coherent
state |
n, L2 is
given as the L-fold symmetric tensor product of the level 12 projector P 1 , viz.
2

L
L
n, = P 1 s s P 1 , L times.
P L n,
2
2
2
2
2
(B.2)
n, L2 = U ( 2 ) (g)|
n0 , L2 where |
n0 , L2 is
The coherent state |
n, L2 can also be constructed as |
the coherent state defined by the projector P L0 |
n0 , L2
n0 , L2 | = P 10 s s P 10 .
2
To any N N matrix (where N = L + 1) we associate an ordinary function L (

n) on S2
given by

L L
L (
(B.3)
n) = n, n, .
2
2
The product of two matrices is mapped to the star product L L (
n) defined by

L L
L L (
n) = n, n, .
2
2
(B.4)
We can show that

L L (
n)
=
L

(L k)!
k=0
k!L!
Ka1 b1 Kak bk
L (
n)
L (
n),
na1
nak
nb1
nbk
(B.5)
where
Kab = ab na nb + iabc nc .
Using these coherent states we can compute

L L
L
n, La n,
= na ,
2
2
2

L
L
= La L (
n),
n, [La , ]n,
2
2
where La = iabc nb c and

1
d
Tr =
L L (
n).
N
4
(B.6)
(B.7)
(B.8)
181
As another example we will compute N22 Tr La Aa which appears in the expansion (87) of the
one-plaquette action. We have immediately
2
1
Tr La Aa = 2 Tr(La Aa + Aa La )
N2
N

d
N 1
=
na (Aa )L + (Aa )L na .
2N
4
(B.9)
It must be clear that (Aa )L is the function which corresponds to the N N matrix Aa . In this
article since we are mostly working with the N N matrices Aa it is more easier to denote the
corresponding functions (Aa )L (in the very few places which appear) by the same symbol Aa
without fear of confusion. By using the star product (B.5) we obtain the result

1
1
2
d
Tr La Aa =
+ a Aa na a
4
N
N
N2

1
3
d
1+
+ a (Aa na ) .
=
(B.10)
4
N
N
This is an exact formula where is defined by = na Aa . In the limit becomes exactly
the normal component of Aa and therefore Aa na is precisely the tangent gauge field on
the sphere. Hence we can see directly that d a (Aa na ) = 0. Furthermore since is
a constant equal to 0 in the limit we can conclude that we have the final result
2
Tr La Aa = 0.
N2
(B.11)
Appendix C. The continuum limits of the one-plaquette actions SP and SP

We need to check what happens to the action SP in the large N limit. We have
SP =

N
Tr2 W + W + 2 .
(C.1)
We will introduce the covariant matrices A a and Sa defined respectively by 2A a = Da Da =

Aa Aa and Sa = Da + Da = 2(La + A a ) where A a is a gauge field defined by the matrices
2
2A a = Aa + Aa . The measure becomes therefore dAa dAa = 23N d A a d A a . We start with the
expansion

1
3
1
2
D = 2 (abc c Fab ) +
(C.2)
(abc c Fab ) + O
N
2N 4
N6
and a similar expansion for D . We can now compute the first nonvanishing covariant terms in
Tr2N W to be

1
1

) 2 (abc c Fab )
Tr2N W = Tr2N 2 (abc c Fab
N
N
3
3
2
(abc c Fab )2 +
(abc c Fab
)
+
2N 4
2N 4

1
1

+ 4 (abc c Fab ) (abc c Fab
)+O
(C.3)
.
N
N6
182
Explicitly we have (by reducing the 2N -dimensional trace Tr2N to the N -dimensional trace Tr)
the following first contribution
Tr2N + h.c. =
2
32
Tr(1 + 4Da Da ) + h.c. = 4N 2 Tr A 2a .
2
N
N
(C.4)
Next we have

1

Tr2N 2 (abc c Fab ) + h.c.
N
4

= 4 Tr(abc Dc Fab
+ abc Dc Fab
+ 4iDa Db Fab
) + h.c.
N

8

+ abc Dc Fab
+ i[Da , Db ] i[Db , Da ] Fab
= 4 Tr abc Dc Fab
N
8

.
= 4 Tr Fab Fab
(C.5)
N
4i[A
a , A b ]. Similarly we can obtain
In above the matrices Fab are defined by Fab = Fab + Fab

8
1
Tr2N 2 (abc c Fab ) + h.c. = 4 Tr Fab Fab .
(C.6)
N
N
Finally we need to evaluate the following three terms

3
3
2
(abc c Fab )2 +
(abc c Fab
)
SP = Tr2N
2N 4
2N 4

1

)
N

3
3
2
= Tr2N
(abc c Fab )2 +
(abc c Fab
)
2N 4
2N 4

1
1

) + 4 [ , abc c Fab ] (abc c Fab
) .
N
N
(C.7)
We start by computing the last piece. To this end we use the identity
[ , abc c Fab ] =

1
4i
2
.
a {Db , Fab } + abc [Dc , Fab ] = 2ia xbD , Fab + O
N
N
N
(C.8)
O( N1 ) stands for all other subleading terms which will yield corrections of the order of
1
or higher to the action. The operators xaD are covariant coordinates on the fuzzy sphere
N5
defined by xaD = Da / c2 . It is clear that in the large N limit xaD na which are
the usual coordinates on the ordinary sphere. Thus the only difference between xaD and the
usual coordinates xa = La / c2 on the fuzzy sphere is that under U (N ) gauge transformaD

D
+
tions we have xa gxa g as opposed to xa which remain fixed. However since = 0 the
operator {xbD , Fab } tends in the continuum limit to 2nb Fab which vanishes identically. Hence
[ , abc c Fab ] = O( N1 ) and thus we obtain

3
3
2
SP = Tr2N
(abc c Fab )2 +
(abc c Fab
)
4
2N
2N 4

1
1

.
)+O
(C.9)
N
N5
To evaluate the other terms we use the following remarkable identity

4
1
1
+ 2 abc c Fab + i{Da , A b } i{Db , A a }
= 2 Da Da +
4
N
N

+ i{Da , Ab } i{Db , Aa }

1
4 N2
2
2Aa [Aa , Sa ]
= 2
4
2
N

1
+ 2 abc c Fab + i{Sa , A b } i{Sb , A a }
N
183
(C.10)
or equivalently
[ , ] =
4i
4
abc c {Da , Db } + 2 [Da , Da ]
2
N
N
(C.11)
and
{ , } = 2
16 2
2
Aa + 2 abc c Fab .
2
N
N
(C.12)
We see immediately that since we are already at order N14 we can set in Eq. (C.9) the following
1 and 1. Thus we obtain

3
3
2
(abc c Fab )2 + 4 (abc c Fab
)
SP + h.c. = Tr2N
4
N
N

2
1

+ 4 (abc c Fab )(abc c Fab
)+O
N
N5

2
4
1
2

= 4 Tr 3Fab
(C.13)
+O
+ 3Fab
+ 2Fab Fab
.
N
N5
The one-plaquette action SP becomes (by putting the contributions (C4)(C6) and (C.13) together)
N
8
32

)
2 Tr A 2a 4 Tr Fab (Fab + Fab
SP =
N
N

2
1
4
2

.
+ 4 Tr 3Fab + 3Fab + 2Fab Fab + O
(C.14)
N
N5
) + Tr(3F 2 + 3F 2 + 2F F ) = Tr(F
2
We remark that 2 Tr Fab (Fab + Fab
ab ab
ab Fab ) +
ab
ab
)[A
a , A b ] and thus
8i Tr(Fab + Fab
N
4
32
2
)
2 Tr A 2a + 4 Tr(Fab Fab
SP =
N
N

32i
1

+ 4 Tr(Fab + Fab )[Aa , Ab ] + O

(C.15)
.
N
N5
= i [S , S ] + 2i[A
a , A b ] + abc Sc and Fab F = i[Sa , A b ] +
By using the results Fab + Fab
ab
2 a b
i[Aa , Sb ] + 2abc Ac we have
184
4
32i
2

Tr(Fab Fab
) + 4 Tr(Fab + Fab
)[A a , A b ]
4
N
N

4
= 4 Tr 2[Sa , A b ][Sb , A a ] 2[Sa , A b ]2 + 8A 2a
N

+ 16iabc A c [Sa , A b ] 4[Sa , Sb ][A a , A b ] 16[A a , A b ]2 .
(C.16)
We recall that Sa = 2La + 2A a and that all commutators [A a , A b ], [A a , A b ] and [A a , A b ] are of

order N1 and hence lead to terms of order N15 in the action in the limit. With this approximation
the transformation laws A a g A a g + and A a g A a g + + g[La , g + ] become A a A a and
A a A a + g[La , g + ], respectively. Thus we obtain
4
32i
2

Tr(Fab Fab
) + 4 Tr(Fab + Fab
)[A a , A b ]
N4
N
16
= 4 Tr 2[La , A b ][Lb , A a ] 2[La , A b ]2 + 2A 2a
N

1
+ 8iabc A c [La , A b ] 4[La , Lb ][A a , A b ] + O
.
N5
The one-plaquette action takes therefore the form

2
N
16
1
32
2 Tr A 2a + 4 Tr i[La , A b ] i[Lb , A a ] + abc A c + O
.
SP =
N
N
N5
(C.17)
(C.18)
We find now the continuum limit of the quadratic action

SP =

N
Tr2N W 2 + W +2 2 .

(C.19)
We have

2
2
Tr2N W 2 = Tr2N ( )2 + 2 Tr2N I2 + 4 Tr2N I4
N
N

1
1
+ 4 Tr2N I22 + O
N
N6
(C.20)
where

I2 = (abc c Fab ) (abc c Fab
)
(C.21)
and
3
3
2

I4 = (abc c Fab )2 + (abc c Fab
) + (abc c Fab
) (abc c Fab ).
2
2
Straightforward computation using Eq. (C.10) gives skip
2
2
16
N
16
Tr2N ( )2 + h.c. = 4 Tr
4A 2a + 4 Tr[A a , Sa ]2
2
N
N

2
8
8
2
4 Tr {Sa , A b } {Sb , A a } + 4 Tr Fab
N
N
16
64 2
= 4N + 4 Tr 4 A 2a N 2 A 2a + 4 Tr[A a , Sa ]2
N
N

2
8
8
2
4 Tr {Sa , A b } {Sb , A a } + 4 Tr Fab
.
N
N
(C.22)
(C.23)
185
Explicitly we have

2
8
Tr {Sa , A b } {Sb , A a }
N4
16
= 4 Tr [Sa , A b ]2 [Sa , A b ][Sb , A a ] 2[Sa , Sb ][A a , A b ]
N

+ 4A 2a Sb2 4A b Sb Sa A a

16
Tr [Sa , A b ]2 [Sa , A b ][Sb , A a ] 2[Sa , Sb ][A a , A b ] 4A 2a
4
N

64
64 2
+ 4 Tr A b Sb Sa A a + 4 Tr 4 A 2a N 2 A 2a .
(C.24)
N
N
In the last line above we have used the constraint Sa2 = 4c2 4A 2a . By using the second constraint
Sa A a = A a Sa we can rewrite this equation as
=

2
8
Tr {Sa , A b } {Sb , A a }
4
N

16
= 4 Tr [Sa , A b ]2 + [Sa , A b ][Sb , A a ] + 2[Sa , Sb ][A a , A b ] + 4A 2a
N

16
64 2
4 Tr[A a , Sa ]2 + 4 Tr 4 A 2a N 2 A 2a .
N
N
Thus we obtain the final exact expression
(C.25)
Tr2N ( )2 + h.c.
128
512 2
8
2
= 4N 2 Tr A 2a + 4 Tr A 2a + 4 Tr Fab
N
N
N

16
+ 4 Tr [Sa , A b ]2 + [Sa , A b ][Sb , A a ] + 2[Sa , Sb ][A a , A b ] + 4A 2a .
N
The next computation is to find
(C.26)

2
2

Tr2N I2 = 2 Tr2N ( )2 (abc c Fab ) + ( )2 (abc c Fab
) .
2
N
N
(C.27)
We use Eq. (C.10) in the form = I + abc c Jab = I + abc c ( N12 Fab + Ni 2 Kab ). The definition of the operators I , Jab = Jba and Kab = Kba is of course obvious. Thus we can
compute

2
4
2
+
V
Fab .
Tr
(
)
(
F
)
+
h.c.
=
Tr

+
V
2N
abc
c
ab
abc
c
c
N2
N2
The operator Vc is defined in terms of I and Jab as follows
(C.28)
Vc = 2iabd Jab Jcd + abc (I Jab + Jab I ).
(C.29)
It is easy to check that the contribution of the first term 2iabd Jab Jcd is of order
whereas the contribution of the second term abc (I Jab + Jab I ) is given by

1
32
.
4 Tr Fab Fab + O
N
N5
1
N5
at least
(C.30)
186
The final result is

2
32
1

Tr
I
+
h.c.
=
Tr
F
(F
+
F
)
+
O
.
2N
2
ab
ab
ab
N2
N4
N5
(C.31)
Next we have to compute the following

2
Tr2N I4
N4

2
3
3 2
2
= 4 Tr2N
)
( ) (abc c Fab )2 + ( )2 (abc c Fab
2
2
N

+ ( )2 (abc c Fab
)(abc c Fab ) + (abc c Fab
)[ , abc c Fab ] .
Since [ , abc c Fab ] is of order
1
N
(C.32)
we obtain
2
2
3
3 2
2
Tr2N I4 = 4 Tr2N
)
( ) (abc c Fab )2 + ( )(abc c Fab
2
2
N4
N

1

+ ( )2 (abc c Fab
)(abc c Fab ) + O
.
N5
(C.33)
In above we can also make the approximations , 1 since we are already at order
Hence we obtain
1
.
N4
2
Tr2N I4 + h.c.
N4

2
1
2

= 4 Tr2N 3(abc c Fab )2 + 3(abc c Fab
) + 2(abc c Fab
)(abc c Fab ) + O
N
N5

2

8
1
2

= 4 Tr2N 3Fab
(C.34)
+O
+ 3Fab
+ 2Fab Fab
.
N
N5
Finally we need to compute
1
Tr2N I22 + h.c.
N4

1

) (abc c Fab
)
= 4 Tr2N (abc c Fab ) (abc c Fab ) + (abc c Fab
N

) + 2 (abc c Fab )[ , abc c Fab
] + h.c.
+ 2( )2 (abc c Fab )(abc c Fab

1
2
2

) + 2(abc c Fab )(abc c Fab
) +O
= 4 Tr2N (abc c Fab )2 + (abc c Fab
N
N5

2

8
1
2

= 4 Tr2N Fab
(C.35)
+O
+ Fab
+ 2Fab Fab
.
N
N5
By putting Eqs. (C.26), (C.31), (C.34) and (C.35) together the quadratic one-plaquette action
becomes
N
512 2
128

SP = 2 Tr A 2a + 4 Tr A 2a
N
N

16
+ 4 Tr [Sa , A b ]2 + [Sa , A b ][Sb , A a ] + 4A 2a + 2[Sa , Sb ][A a , A b ]
N
8
64i
2

Tr(Fab Fab
) + 4 Tr(Fab + Fab
)[A a , A b ]
4
N
N

128
1
.
4 Tr[A a , A b ]2 + O
N
N5
187
(C.36)
As before if we drop all commutators [A a , A b ], [A a , A b ] and [A a , A b ] (since they are of order

1
1

N and hence lead to terms of order N 5 in the action) then the limit of SP reduces to
2
N 256
128
512
Tr A a La + 2 Tr A 2a + 4 Tr A 2a + {A a , La }
SP =
2
N
N
N

2
64
1
.
+ 4 Tr i[La , A b ] i[Lb , A a ] + abc A c + O
(C.37)
N
N5
References
[1] R.J. Szabo, Phys. Rep. 378 (2003) 207.
[2] C.P. Martin, F. Ruiz Ruiz, Nucl. Phys. B 597 (2001) 197;
For a treatment of the same problem on the 4D NC torus see: T. Krajewski, R. Wulkenhaar, Int. J. Mod. Phys. A 15
(2000) 1011;
See also M. Hayakawa, hep-th/9912167.
[3] S. Minwalla, M. Van Raamsdonk, N. Seiberg, JHEP 0002 (2000) 020.
[4] M.R. Douglas, N.A. Nekrasov, Rev. Mod. Phys. 73 (2001) 977.
[5] W. Bietenholz, F. Hofheinz, J. Nishimura, JHEP 0209 (2002) 009.
[6] S.S. Gubser, S.L. Sondhi, Nucl. Phys. B 605 (2001) 395;
J. Ambjorn, S. Catterall, Phys. Lett. B 549 (2002) 253;
W. Bietenholz, F. Hofheinz, J. Nishimura, JHEP 0406 (2004) 042;
W. Bietenholz, F. Hofheinz, J. Nishimura, Nucl. Phys. B (Proc. Suppl.) 119 (2003) 941.
[7] B. Ydri, Fuzzy physics, PhD thesis, 2001, hep-th/0110006;
H. Grosse, C. Klimcik, P. Prenajder, Commun. Math. Phys. 180 (1996) 429;
H. Grosse, C. Klimcik, P. Prenajder, Int. J. Theor. Phys. 35 (1996) 231;
D. OConnor, Mod. Phys. Lett. A 18 (2003) 2423;
C. Klimcik, Commun. Math. Phys. 199 (1998) 257.
[8] J. Madore, Class. Quantum Grav. 9 (1992) 6988;
J. Hoppe, MIT PhD thesis, 1982;
J. Hoppe, S.T. Yau, Commun. Math. Phys. 195 (1998) 6777.
[9] A.Y. Alekseev, A. Recknagel, V. Schomerus, hep-th/0003187;
A.Y. Alekseev, A. Recknagel, V. Schomerus, hep-th/9812193.
[10] S. Iso, Y. Kimura, K. Tanaka, K. Wakatsuki, Nucl. Phys. B 604 (2001) 121.
[11] X. Martin, JHEP 0404 (2004) 077;
X. Martin, Mod. Phys. Lett. A 18 (2003) 23892396.
[12] D. OConnor, B. Ydri, Monte Carlo simulation of NC gauge field on the fuzzy sphere, hep-lat/0606013.
[13] S. Vaidya, Phys. Lett. B 512 (2001) 403;
B.P. Dolan, D. OConnor, P. Prenajder, JHEP 0203 (2002) 013;
B.P. Dolan, D. OConnor, P. Prenajder, JHEP 0402 (2004) 055;
T. Imai, Y. Kitazawa, Y. Takayam, D. Tomino, Nucl. Phys. B 665 (2003) 520.
[14] S. Vaidya, B. Ydri, Nucl. Phys. B 671 (2003) 401431, hep-th/0209131;
C.S. Chu, J. Madore, H. Steinacker, JHEP 0108 (2001) 038;
B.P. Dolan, D. OConnor, P. Prenajder, JHEP 0203 (2002) 013, hep-th/0109084.
[15] P. Castro-Villarreal, R. Delgadillo-Blando, B. Ydri, JHEP 0509 (2005) 066;
W. Behr, F. Meyer, H. Steinacker, JHEP 0507 (2005) 040;
T. Imai, Y. Takayama, Nucl. Phys. B 686 (2004) 248;
T. Azuma, S. Bal, K. Nagao, J. Nishimura, JHEP 0509 (2005) 047;
T. Azuma, S. Bal, K. Nagao, J. Nishimura, JHEP 0407 (2004) 066.
188
[16] H. Steinacker, JHEP 0503 (2005) 075;

T. Azuma, S. Bal, K. Nagao, J. Nishimura, hep-th/0405277.
[17] B. Ydri, Mod. Phys. Lett. A 19 (2004) 22052213;
B. Ydri, Nucl. Phys. B 690 (2004) 230248.
[18] G. Alexanian, A.P. Balachandran, G. Immirzi, B. Ydri, J. Geom. Phys. 42 (2002) 2853;
H. Grosse, A. Strohmaier, Lett. Math. Phys. 48 (1999) 163179;
H. Grosse, H. Steinacker, Nucl. Phys. B 707 (2005) 145.
[19] J. Medina, D. OConnor, JHEP 0311 (2003) 051;
Y. Kimura, Nucl. Phys. B 637 (2002) 177.
[20] T. Eguchi, H. Kawai, Phys. Rev. Lett. 48 (1982) 1063.
[21] J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, JHEP 0005 (2000) 023.
[22] P. Castro-Villarreal, R. Delgadillo-Blando, B. Ydri, A gauge-invariant UVIR mixing and the corresponding phase
transition for U (1) fields on the fuzzy sphere, hep-th/0405201;
P. Castro-Villarreal, R. Delgadillo-Blando, B. Ydri, Nucl. Phys. B 704 (2004) 111153.
[23] T. Azuma, S. Bal, K. Nagao, J. Nishimura, JHEP 0405 (2004) 005.
[24] H. Steinacker, Nucl. Phys. B 679 (2004) 66.
[25] D.J. Gross, E. Witten, Phys. Rev. D 21 (1980) 446453;
See also S.R. Wadia, EFI preprint EFI 80/15, March 1980.
[26] P. Prenajder, private communication.
[27] A. Connes, J. Lott, Nucl. Phys. B (Proc. Suppl.) 18 (1990) 29;
J.M. Gracia-Bondia, J.C. Varilly, J. Geom. Phys. 12 (1993) 223;
C.P. Martin, J.M. Gracia-Bondia, J.C. Varilly, Phys. Rep. 294 (1998) 363.
[28] E. Brezin, J. Zinn-Justin, hep-th/9206035.
[29] S. Baez, A.P. Balachandran, S. Vaidya, B. Ydri, Commun. Math. Phys. 208 (2000) 787;
A.P. Balachandran, S. Vaidya, Mod. Phys. A 16 (2001) 17;
H. Grosse, C. Klimcik, P. Prenajder, Commun. Math. Phys. 178 (1996) 507;
H. Grosse, C.W. Rupp, A. Strohmaier, J. Geom. Phys. 42 (2002) 5463;
H. Aoki, S. Iso, K. Nagao, hep-th/0312199.
[30] A.P. Balachandran, T.R. Govindarajan, B. Ydri, Mod. Phys. Lett. A 15 (2000) 1279, hep-th/0006216;
H. Aoki, S. Iso, K. Nagao, Phys. Rev. D 67 (2003) 085005;
A.P. Balachandran, G. Immirzi, Phys. Rev. D 68 (2003) 065023;
A.P. Balachandran, G. Immirzi, Int. J. Mod. Phys. A 18 (2003) 5981;
U. Carow-Watamura, S. Watamura, Commun. Math. Phys. 183 (1997) 365382;
U. Carow-Watamura, S. Watamura, Int. J. Mod. Phys. A 13 (1998) 32353244.
[31] B. Ydri, JHEP 0308 (2003) 046;
P. Prenajder, J. Math. Phys. 41 (2000) 2789;
H. Aoki, S. Iso, K. Nagao, Phys. Rev. D 67 (2003) 065018;
H. Grosse, P. Prenajder, Lett. Math. Phys. 46 (1998) 61, hep-th/9805085.
[32] J. Volkholz, W. Bietenholz, J. Nishimura, Y. Susaki, PoS LAT2005 (2006) 264;
W. Bietenholz, A. Bigarini, F. Hofheinz, J. Nishimura, Y. Susaki, J. Volkholz, Fortschr. Phys. 53 (2005) 418;
W. Bietenholz, F. Hofheinz, J. Nishimura, Y. Susaki, J. Volkholz, Nucl. Phys. B (Proc. Suppl.) 140 (2005) 772.
Electroweak symmetry breaking and precision tests

with a fifth dimension
Giuliano Panico a , Marco Serone a , Andrea Wulzer b,
a ISAS-SISSA and INFN, Via Beirut 2-4, I-34013 Trieste, Italy
b IFAE, Universitat Autnoma de Barcelona, 08193 Bellaterra, Barcelona, Spain
Received 7 June 2006; received in revised form 15 September 2006; accepted 31 October 2006
Abstract
We perform a complete study of flavor and CP conserving electroweak observables in a slight refinement
of a recently proposed five-dimensional model on R 4 S 1 /Z2 , where the Higgs is the internal component of
a gauge field and the Lorentz symmetry is broken in the fifth dimension. Interestingly enough, the relevant
corrections to the electroweak observables turn out to be of universal type and essentially depend only on
the value of the Higgs mass and on the scale of new physics, in our case the compactification scale 1/R. The
model passes all constraints for 1/R 4.7 TeV at 90% C.L., with a moderate fine-tuning in the parameters.
The Higgs mass turns out to be always smaller than 200 GeV although higher values would be allowed,
due to a large correction to the T parameter. The lightest non-SM states in the model are typically colored
fermions with a mass of order 12 TeV.
1. Introduction
The idea of identifying the Higgs field with the internal component of a gauge field in
TeV-sized extra dimensions [1], recently named gauge-Higgs unification (GHU), has received
a revival of interest in the last years. Although not new (see e.g. [2]), only recently it has been
realized that this idea could lead to a solution of the Standard Model (SM) instability of the
electroweak scale [3].
E-mail address: wulzer@ifae.es (A. Wulzer).

doi:10.1016/j.nuclphysb.2006.10.032
190
G. Panico et al. / Nuclear Physics B 762 (2007) 189211
Five-dimensional (5D) models, with one extra dimension only, are the simplest and phenomenologically more appealing examples. When the extra dimension is a flat S 1 /Z2 orbifold,
however, the Higgs and top masses and the scale of new physics, typically given by the inverse
of the compactification radius 1/R, are too low in the simplest implementations of the GHU
idea [4]. Roughly speaking, the Higgs mass is generally too light because the Higgs effective
potential is radiatively induced, giving rise to a too small effective SM-like quartic coupling. The
top is too light because the Yukawa couplings, which are effective couplings in these theories,
are engineered in such a way that they are always smaller than the electroweak gauge couplings,
essentially due to the 5D Lorentz symmetry. Finally, the scale of new physics is too low, because
in minimal models there is no way to generate a sizable gap between the electroweak (EW) scale
and 1/R.
It has recently been pointed out that it is possible to get rid of all the above problems by
increasing the Yukawa (gauge) couplings of the Higgs field with the fermions, by assuming a
Lorentz symmetry breaking in the fifth direction [5] (see also [6] for other interesting possibilities in a warped scenario). In this way, the Yukawa couplings are not constrained anymore to
be smaller than the electroweak gauge couplings and we have the possibility to get Yukawas of
order one, as needed for the top quark. Stronger Yukawa couplings lead also to a larger effective Higgs quartic coupling, resulting in Higgs masses above the current experimental bound of
mH > 115 GeV. Furthermore, again due to the larger Yukawa couplings, the Higgs effective potential is totally dominated by the fermion contribution. The latter effect, together with a proper
choice of the spectrum of (periodic and antiperiodic) 5D fermion fields, can lead to a substantial
gap between the EW and compactification scales. In non-universal models of extra dimensions
of this sort, where the SM fermions have sizable tree-level couplings with KaluzaKlein (KK)
gauge fields, the electroweak precision tests (EWPT) generically imply quite severe bounds on
the compactification scale. At the price of some fine-tuning on the microscopic parameters of
the theory, however, one can push 1/R in the multi-TeV regime, hoping to get in this way a
potentially realistic model.
The SM fermions turn out to be almost localized fields, with the exception of the bottom quark,
which shows a partial delocalization, and of the top quark, which is essentially delocalized. This
effect generally leads to sizable deviations of the Z bL bL coupling with respect to the SM one.
Another distortion is introduced in the gauge sector, in order to get the correct weak mixing angle
(see [7] for a recent discussion on the difficulty of automatically getting the correct weak-mixing
angle in 5D GHU models). The main worrisome effect of the distortion is the departure [5] of
the SM parameter from one at tree-level.
Aim of this paper is to consider a variant of the model constructed in [5] and to compute the
flavor conserving phenomenological bounds. The main new ingredient is an exact discrete Z2
symmetry, that we call mirror symmetry. It essentially consists in doubling a subset of bulk
fields , namely all the fermions and some of the gauge fields, in pairs 1 and 2 and requiring
a symmetry under the interchange 1 2 . Periodic and antiperiodic fields arise as suitable
linear combinations of 1 and 2 and the symmetry constrains the couplings of the Higgs with
periodic and antiperiodic fermions with the same quantum numbers to be equal. When this is
not the case, as in [5], the contributions to the Higgs mass-term of periodic and antiperiodic
fields must be finely canceled in order to get a hierarchy between the compactification and EW
scale. This adjustment was shown in [5] to be the main source of fine-tuning. Thanks to the
mirror symmetry, on the contrary, a natural partial cancellation occurs in the potential. We will
see that a difference of an order of magnitude between the compactification and EW scale, with
1/R 1 TeV, is completely natural in our model. Such a scale is still too low to pass EWPT, but
191
we significantly lower the required amount of fine-tuning. Using the standard definition of [8],
thanks to the mirror symmetry, we get a fine-tuning of O(1%), an order of magnitude better
than the O(1h) found in [5]. If quantified with a more refined definition [9] which takes into
account the presence of a possible generic sensitivity in the model, but under some physically
motivated assumptions, the fine-tuning turns out to be less, of about O(10%). Another important
difference with respect to the model of [5] is the choice of the bulk fermion representations
under the electroweak gauge group, which leads to a drastic reduction of the deviation to the SM
Z bL bL coupling. This reduction is so effective, that the Z bL bL distortion becomes negligible in
our set-up.
Interestingly enough, all SM states are even under the mirror symmetry, so that the lightest
odd particle is absolutely stable. In a large fraction of the parameter space of the model, such a
state is the first KK mode of an antiperiodic gauge field. The mirror symmetry represents then
an interesting way to get stable non-SM particles in non-universal extra-dimensional theories,
where KK parity is not a suitable symmetry. Along the lines of [10], it could be interesting to
investigate whether such state represents a viable dark matter (DM) candidate.
Even assuming flavor and CP conserving new physics effects only, it is necessary to add 18
dimension 6 operators to the SM in order to fit the current data (see e.g. [11,12]). Among such
18 operators, it has recently pointed out that only 10 are sensibly constrained [13]. They can be
T , U , V , X, W and Y introduced in [14], the
parametrized by the seven universal parameters S,
b parameter [15], and other two parameters which describe the deviation of the up and down
quark couplings to the Z boson. Due to the way in which light SM fermions couple in our model
and to the above suppression of the Z bL bL deviation, it turns out that only the four parameters S,
T , W and Y are significant, like in a generic universal model as defined in [14]. Out of the several
T , W and Y depend essentially only on mH and 1/R. The result of
parameters in our theory, S,
our fit is reported in Fig. 3, which represents one of the main results of this paper. We find that
1/R 4 TeV with a Higgs mass which can reach up to 600 GeV. Such high values of the Higgs
mass are permitted because a relatively high value of the T parameter can be compensated by the
effects of a heavy Higgs (see [16] for a discussion of a similar effect in the context of universal
extra dimensions (UED) [17]). Although allowed, high values of mH are actually never reached
in our model, which predicts 100 GeV mH 200 GeV. As can be seen from Fig. 3, our model
can successfully pass the combined fit of all electroweak precision observables.
The paper is organized as follows. In Section 2 we introduce the model, emphasizing the main
points in which it differs from that of [5]. In Section 3 we show our results for the combined electroweak observables, including a brief discussion on the fine-tuning of the model. In Section 4
we draw our conclusions and in the appendices we collect some useful technical details.
2. The effective Lagrangian
The 5D model we consider is a slight refinement of the one proposed in [5]. The main essential feature is the introduction of a Z2 mirror symmetry. The gauge group is taken to be
G = SU(3)w G1 G2 , where Gi = U (1)i SU(3)i,s , i = 1, 2, with the requirement that the
Lagrangian is invariant under the Z2 symmetry 1 2. The periodicity and parities of the fields
on the S 1 /Z2 space are the same as in [4,5] for the electroweak SU(3)w sector, whereas for the
Abelian U (1)i and non-Abelian colored SU(3)i,s fields we now have (omitting for simplicity
vector and gauge group indices):
A1 (y 2R) = A2 (y),
A1 (y) = A2 (y),
(2.1)
192
where = 1, 5 = 1, denoting by Greek indices the 4D directions. The unbroken gauge

group at y = 0 is SU(2) U (1) G+ , whereas at y = R we have SU(2) U (1) G1 G2 ,
where G+ is the diagonal subgroup of G1 and G2 . The Z2 mirror symmetry also survives the
compactification and remains as an exact symmetry
of our construction. It is clear from Eqs. (2.1)
that the linear combinations A = (A1 A2 )/ 2 are respectively periodic and antiperiodic
on S 1 . Under the mirror symmetry, A A , so we can assign a multiplicative charge +1
to A+ and 1 to A . The massless 4D fields are the gauge bosons in the adjoint of SU(2)
U (1) SU(3)w , the U (1)+ and gluon gauge fields A+ and a charged scalar doublet Higgs
4,5,6,7 .
field, arising from the internal components of the odd SU(3)w 5D gauge fields, namely Aw
The SU(3)+,s and SU(2) gauge groups are identified respectively with the SM SU(3)QCD and
SU(2)L ones, while the hypercharge U (1)Y is the diagonal subgroup of U (1) and U (1)+ . The
extra U (1)X gauge symmetry which survives the orbifold projection is anomalous (see [4,5])
and its corresponding gauge boson gets a mass of the order of the cut-off scale of the model.1
An Higgs VEV, i.e. a VEV for the extra-dimensional components of the SU(3)w gauge fields,
induces the additional breaking to U (1)EM . Following [4,5], we parametrize this VEV as
Aw y =
2 7
t ,
g5 R
(2.2)
where g5 is the 5D charge of the SU(3)w group and t a its generators, normalized as 2 Tr t a t b = ab
in the fundamental representation. With this parametrization, the associated Wilson line is
7
W = e4it .
(2.3)
We also introduce a certain number of couples of bulk fermions (, ), with identical quantum numbers and opposite orbifold parities. There are couples (1 , 1 ) which are charged under
G1 and neutral under G2 and, by mirror symmetry, the same number of couples (2 , 2 ) charged
under G2 and neutral under G1 . No bulk field is simultaneously charged under both G1 and G2 .
t ,
t ) in the antifundamental representation of
In total, we introduce one pair of couples (1,2
1,2
b
b
SU(3)w and one pair of couples (1,2 , 1,2 ) in the symmetric representation of SU(3)w . Both
pairs have U (1)1,2 charge +1/3 and are in the fundamental representation of SU(3)1,2,s . The
boundary conditions of these fermions follow from Eqs.
(2.1) and the twist matrix introduced
in [4]. In particular, the combinations = (1 2 )/ 2 are respectively periodic and antiperiodic on S 1 .
Finally, we introduce massless chiral fermions with charge +1 with respect to the mirror
symmetry, localized at y = 0. As explained in [4,5], as far as electroweak symmetry breaking
is concerned, we can focus on the top and bottom quark only, neglecting all the other SM matter fields, which however can be accommodated in our construction. Mirror symmetry and the
boundary conditions (2.1) imply that the localized fields can couple only to A+ . Hence, we
have an SU(2) doublet QL and two singlets tR and bR , all in the fundamental representation of
SU(3)+,s and with charge +1/3 with respect to the U (1)+ gauge field A+ .
The most general 5D Lorentz breaking effective Lagrangian density, gauge invariant and mirror symmetric, up to dimension d < 6 operators, is the following (we use mostly minus metric
and ( 5 )2 = 1):
L = Lg + L + (y)L0 + (y R)L ,
(2.4)
1 In an interval approach, one can get rid of U (1) by imposing Dirichlet boundary conditions for it at one of the
X
end-points of the segment. In fact, the two descriptions are equivalent and become identical in the limit .
193
with

1
1
2
i
2
i5
i
i5
Fi5 F
Tr Gi G s Tr Gi5 G Fi F
Lg =
2
4
2
i=1,2
1
Tr F F w2 Tr F5 F 5 ,
(2.5)
2

/ 4 (Ai ) ka D5 (Ai ) 5 ia + ai iD
/ 4 (Ai ) ka D5 (Ai ) 5 ia
L =
ia iD
i=1,2 a=t,b
Ma ai ia + ia ia ,
L iD
L0 = Q
/ 4 (A+ )QL + tR iD
/ 4 (A+ )tR + bR iD
/ 4 (A+ )bR
t

t
b
b
t
t
+ e1 QL + + e1 QL + + e2 tR + + e2b bR +b + h.c. + L 0 .
(2.6)
(2.7)
In Eq. (2.5), we have denoted by Gi = DAi,s the gluon field strengths for SU(3)i,s and for
simplicity we have not written the ghost Lagrangian and the gauge-fixing terms. For the same
reason, in Eqs. (2.6) and (2.7) we have only schematically written the dependencies of the covariant derivatives on the gauge fields and in Eq. (2.7) we have not distinguished the doublet
and singlet components of the bulk fermions, denoting all of them simply as +t and +b . Notice
that + is the only bulk fermion that can have a mass-term mixing with the localized fields,
since mixing with and is forbidden by mirror symmetry and the relevant components of
+ vanish at y = 0 due to the boundary conditions. Extra brane operators, such as for instance
localized kinetic terms, are included in L 0 and L . Additional Lorentz violating bulk operators
like 5 , y or iD
/ 4 5 can be forbidden by requiring invariance under the inversion
of all spatial (including the compact one) coordinates, under which any fermion transforms as
0 . This Z2 symmetry is a remnant of the broken SO(4, 1)/SO(3, 1) Lorentz generators.
Notice that our choice of U (1) charges allows mixing of the top quark with a bulk fermion in the
3 while the bottom couples with a 6 of SU(3)w . In [5], the choice of taking bulk fermions neutral
under the U (1) led to the opposite situation. As we will see, this greatly reduces the deviation
from the SM of the Z bL bL coupling, which becomes negligible.
Strictly speaking, the Lagrangian (2.4) is not the most general one, since we are neglecting
all bulk terms which are odd under the y y parity transformation and can be introduced if
multiplied by odd couplings. If not introduced, such couplings are not generated and thus can
consistently be ignored.
It is now possible to better appreciate the reason why we have introduced the above mirror symmetry. In terms of periodic and antiperiodic fields, the mirror symmetry constrains the
Lorentz violating factors for periodic and antiperiodic fermions to be the same: k+ = k k,
k+ = k k for both the 3 and 6 representations. Finding an exact symmetry constraining these
factors to be equal is important, since it was found in [5] that the electroweak breaking scale was
mostly sensitive to the ratio k+ /k , which implied some fine-tuning in the model. Thanks to the
mirror symmetry, the fine-tuning is significantly lowered as we will better discuss in Section 3.1.
All SM fields are even under the mirror symmetry. This implies that the lightest Z2 odd (antiperiodic) state in the model is absolutely stable. This is a very interesting byproduct, since in a (large)
fraction of the parameter space of the model such state is the first KK mode of the A gauge
field. The latter essentially corresponds to the first KK mode of the hypercharge gauge boson in
the context of UED [17], which has been shown to be a viable DM candidate [10]. Hence, it is
not excluded that A can explain the DM abundance in our Universe.
194
A detailed study of the model using the general Lagrangian (2.4) is a too complicated
task. For this reason, we take ka = ka which considerably simplifies the analysis2 and set
w = 1. The latter choice can always be performed without loss of generality by rescaling
the compact coordinate, and hence the radius of compactification as well as the other parameters of the theory.3 Moreover, we neglect all the localized operators which are encoded in
L 0 and L . The latter simplification requires a better justification that we postpone to Section 2.2.
2.1. Mass spectrum and Higgs potential
The localized chiral fermions, as explained at length in [4,5], introduce gauge anomalies involving the U (1) gauge field AX . The effect of such anomalies is the appearance of a large
localized mass term for AX at y = 0, whose net effect is to fix to zero the value of the field
at y = 0: AX (0) = 0. This complicates the computation of the gauge bosons mass spectrum,
which is distorted by this effect. The Lorentz violating factor 2 also distorts the spectrum
of the KK gauge bosons, so that the analytic mass formulae for the SU(3)w U (1)+ gauge
bosons are slightly involved. They are encoded in the gauge contribution to the Higgs potential
which we will shortly consider. For = 1, the mass spectrum has been computed in [5]. The
only unperturbed tower is the one associated to the W boson, whose masses are mW + n/R
where
mW = ,
(2.8)
R
is the SM W boson mass. The mass spectra of the KK towers associated to the U (1) gauge field
A and to the gluons are trivial and given by mn = (n + 1/2)/R for U (1) , mn = ns /R
for SU(3)+,s and mn = (n + 1/2)s /R for SU(3),s .
The spectrum of the bulk-boundary fermion system defined by the Lagrangian (2.4) is determined as in [5], once one recalls the interchange between the fundamental and symmetric
representations of SU(3)w , which simply lead to the interchange t b in Eq.
(2.18) of [5]. Let
us introduce, as in [4,5], the dimensionless quantities i = RMi and ia = R/2eia . In the
limit 1t , 2t 1, 1b = 2b = 0, 1, one gets
2t /kt
(2.9)
sinh (2t /kt )
which gives a top mass a 2 factor lighter than in [5]. We deduce from Eq. (2.9) that kt 23 is
needed to get the top mass in the correct range.
For a large range of the microscopic parameters, the bulk-boundary fermion system also gives
the lightest new particles of our model. Such states are colored fermions with a mass of order Mb ,
and, in particular, before electroweak symmetry breaking (EWSB), they are given by an SU(2)
triplet with hypercharge Y = 2/3, a doublet with Y = 1/6 and a singlet with Y = 1/3. For
mt kt mW
2 It should be emphasized that there is no fine-tuning associated to k/k (contrary to k /k ) and thus this choice
+
represents only a technical simplification.

3 Strictly speaking, in a UV completion where the theory is coupled to gravity and the Lorentz violation is, for instance,
due to a flux background [5], w cannot be rescaled away by redefining the radius of compactification, since the latter
becomes dynamical and w essentially corresponds to a new coupling in the theory. In our context, however, R is simply
a free parameter.
195
the typical values of the parameters needed to get a realistic model, the mass of these states is of
order 12 TeV.
Let us now turn to the computation of the Higgs effective potential. The fermion contribution in presence of bulk-to-boundary mixing and Lorentz violating factors is the same as
in [5]. The gauge contribution is slightly different, because of the parameter 2 , which in [5]
was set to unity, for simplicity. The one-loop gauge effective potential is however readily
computed using the holographically inspired method of [18] (see also [19] for a treatment of
fermions in this context), which also gives part of the mass spectrum. We refer the reader
to Appendix A for a brief review of such technique, applied in our context. The only gauge
fields which contribute to the Higgs potential are Aw , associated to SU(3)w , and A+ , associated to U (1)+ . Before EWSB, it is trivial to integrate out the bulk at tree-level. In the mixed
momentum-space basis, and in the unitary gauge, one finds the following holographic Lagrangian:

1
Lholo = P
Tr q cot(2Rq)A w (q)A w (q)
2

+ q csc(2Rq)A w (q)P A w (q)P

+ q cot(2Rq/)A + (q)A + (q)
2

+ q csc(2Rq/)A+ (q)A+ (q) ,

(2.10)

where q = q q , P = q q /q 2 is the standard transverse projector, P = diag(, ,
+) is the orbifold projection matrix and A are the holographic gauge fields, as defined in
Appendix A. Since the Higgs is a Wilson line, its VEV can be removed from the bulk
through a non-single valued gauge transformation [20], which we can choose in such a way
that the boundary conditions at y = R remain unchanged. In this new basis, we can obtain the Lagrangian after EWSB directly from Eq. (2.10) by replacing Aw W 1/2 Aw W 1/2 ,
or P P W , where W is the Wilson line as defined in Eq. (2.3). Notice that all the dependence of the action on , in the rotated field basis, comes from the boundary condition at y = 0. When integrating out at tree-level the bulk and the y = R boundary, then,
we are not neglecting any contribution to the one loop effective potential. After obtaining
the holographic (-dependent) Lagrangian we have to impose the boundary conditions, i.e.
to put to zero the Dirichlet components of the holographic fields. We normalize the surviving (electroweak) gauge bosons in a non-canonical way (as in [14]) and then parametrize
3

2 4
1
1
a
4
A w =
(2.11)
W ta + W t8 ,
W ,
A + =
g5
g 5
3
a=1
where W 4 indicates the hypercharge gauge boson and g 5 the (common, due to mirror symmetry)
U (1)1,2 charges. We finally get an action of the form:
Lholo =
1 i
W (q)i,j W ,j (q),
2
i,j
(2.12)
196
Fig. 1. Distribution of min for uniformly distributed input parameters in the ranges 0.5 < < 2, 0.25 < t,b < 1.25,
2 < kt < 3, 0.75 < kb < 1.5, 0.75 < t1,2 < 2.5, 0.1 < b1,2 < 0.45.
from which the gauge contribution to the one loop effective potential is found to be
Vg () =
3
2
3
=
2

d 4q
log Det()
4
(2)

2
d 4q
log cos(2) cos(2qR)
4
(2)

d 4q
log 4g5 2 sin (qR) cos(qR/) cos(4) cos(2qR)
4
(2)

2
+ 3g5 sin(qR/) cos(qR) 3 + cos(4) 4 cos(2qR)
3
2
+ -ind. terms.
(2.13)
Note that the above integrals are divergent but they can be made finite (after rotating to Euclidean
momentum) by adding suitable -independent terms which we have not written for simplicity. The zeroes of the first square bracket in Eq. (2.13) correspond to the aforementioned W
tower while the second one provides the mass-equation for the neutral gauge bosons sector.
Notice that for = 1 the Z boson tower cannot be separated from the ones associated to the
photon and to AX ; all of them arise from the zeroes of the second term of Eq. (2.13). The
solution at q = 0 corresponds to the physical massless photon while the first non-trivial one
determines the Z boson mass. We have checked that the above mass equation can be also obtained, after a long calculation, by directly computing the KK wave functions for the gauge
bosons.
The full Higgs effective potential is dominated by the fermion contribution. The presence of
bulk antiperiodic fermions, whose coupling with the Higgs are the same as for periodic fermions
due to the mirror symmetry, allows for a natural partial cancellation of the leading Higgs mass
terms in the potential, lowering the position of its global minimum min , as discussed in [5],
Section 2.3. This can be seen from the histogram in Fig. 1, which shows the distribution (min )
for random (uniformly distributed) values of the input parameters, chosen in the ranges which
give the correct order of magnitude for the top and bottom masses. We are neglecting the points
with unbroken EW symmetry (min = 0), which are about half of the total.
197
Fig. 2. Higgs and top masses for points with min < 0.05. Different colors label different values of kt . The region among
the two vertical black lines corresponds to the physical top mass.
From the effective potential we can determine the Higgs mass, which reads

g52 R 2 V
2

.
mH (min ) =
8 2 =min
(2.14)
In Fig. 2 the Higgs mass is plotted versus the top one and different colors label different ranges
of the input parameter kt . We see how realistic values of mt can be obtained already for kt = 2.
The two vertical black lines in Fig. 2 identify the region of physical top mass. This is taken to be
around 150 GeV, which is the value one gets by running the physical top mass using the SM RGE
equations up to the scale 1/R. From Fig. 2 we see that the Higgs mass lies in the [100, 200] GeV
range. It is then significantly lighter than what found in [5].
2.2. Estimate of the cut-off and loop corrections
The model presented here, as any 5D gauge theory, is non-renormalizable and hence has a
definite energy range of validity, out of which the theory enters in a strong coupling uncontrolled
regime. It is fundamental to have an estimate of the maximum energy which we can probe
using our effective 5D Lagrangian (2.4). First of all, we have to check that is significantly
above the compactification scale 1/R, otherwise the 5D model has no perturbative range of
applicability. Once the order of magnitude of is known, we can also estimate the natural values
of the parameters entering in the Lagrangian (2.4). We define as the energy scale at which the
one-loop vacuum polarization corrections for the various gauge field propagators is of order of
the tree-level terms. This can be taken as a signal of the beginning of a strong coupling regime.
The relevant gauge fields are those of SU(3)i,s and SU(3)w , the Abelian ones associated to
U (1)1,2 typically giving smaller corrections for any reasonable choice of 2 . Due to the Lorentz
violation, however, the transverse (A ) and longitudinal (A5 ) gauge bosons couple differently
between themselves and with matter, so that they should be treated separately. We have computed
the one-loop vacuum polarizations neglecting the effect of the compactification, which should
give small finite size effects, using a PauliVillars (PV) regularization in a non-compact 5D
()
()
(5)
(5)
space. We denote by s , s , w and w the resulting cut-offs (due to the mirror symmetry,
the cut-offs associated to the two SU(3)s gauge groups coincide). A reasonable approximation is
198
to assume that the gauge contribution is approximately of the same order as that of the fermions
associated to the first two generations. In this case, by taking kt 23, kb 1, which are the
typical phenomenological values, we get4

() s 1 12 24
6
+
s R
,
1 ()
s
6 2 kt
kb
R
(5) s 1
3s2
(12kt + 24kb ) s2 (5)

,
s R
s
6 2
R

() w 1 12 60
5
+
,
1 ()
w R
w
6 2 kt
kb
R
(5) w 1
4
w R
(2.15)
(12kt + 60kb ) 1 (5)
.
w
6 2
R
As can be seen from Eq. (2.15), () and (5) (both for the strong and weak coupling case) scale
respectively as k and 1/k in the Lorentz violating parameters. This is easily understood by noting
that the loop factor scales as 1/k [5]. Thus () 1/ k k and (5) 1/(k 2 ) k 1/k.
Due to the group factor of the SU(3)w symmetric representation, the electroweak and strong
interactions give comparable values for the cut-off. Independently of the strong interactions,5 the
cut-off in the model is quite low: 4/R. The doubling of the fields due to the mirror symmetry
has the strongest impact, since it decreases the electroweak cut-off estimate by a factor of 2. As
far as the Lorentz violation is concerned, we see that Eq. (2.15) gives, for kt = kb = s = 1,
(5)
s 3/R. In the absence of Lorentz breaking and mirror symmetry, therefore, the final cut-off
would have been as low as 3/R. Even though Eq. (2.15) only provides an order-of-magnitude
estimate, we can use it to place upper bounds on the allowed values of the Lorentz violating
parameters kt and kb . To ensure 4/R, we impose kt 3 and kb 1.2.
Once we have an estimate for the value of the cut-off in the theory, we can also give an
estimate of the natural size of the coefficients of the operators appearing in the Lagrangian (2.4).
In particular, since we have neglected them, it is important to see the effect of the localized
operators appearing in L 0, when their coefficients are set to their natural value. At one-loop
level, several localized kinetic operators are generated at the fixed-points. In the fermion sector,
which appears to be the most relevant since it almost completely determines the Higgs potential,
we have considered operators of the form6

i
L 0, =
a0,
4 i ,
(2.16)
i i/
i
indicates here the components of the bulk fields which are non-vanishing at the y = 0
where
or y = R boundaries. Such operators are logarithmically divergent at one-loop level and thus
i
not very sensitive to the cut-off scale; anyhow, computing their coefficients a0,
using a PV
i
4 Notice that the estimates (2.15) differ significantly from those reported in [4,5], since here we have taken into account
the fermion multiplicities. In addition, the PV regularization gives rise to a 5D loop factor in the vacuum polarization
diagram which is 24 2 and not 24 3 , as taken in [4,5] and naively expected.
5 Since 2 is essentially a free parameter in our considerations, we could anyhow increase (5) by taking 2 signifis
s
s
cantly larger than unity.
6 Among all possible localized operators, those with derivatives along the internal dimension require special care and
are more complicated to handle [21]. It has been pointed out in [22] that their effect can however be eliminated by suitable
field redefinitions (see also [23]).
199
regularization and setting 4/R, they turn out to be of order 103 R. By including these
terms in the full Lagrangian, we have verified that the shape of the effective potential, the Higgs
and the fermion masses receive very small (of order per-mille) corrections. Localized operators
can however sizably affect the position of the minimum of the effective potential when it is tuned
to assume small values. We will came back on this in Section 3.1.
Finally, let us comment on the predictability of the Higgs mass at higher loops. Although
finite at one-loop level, the Higgs mass will of course develop divergencies at higher orders,
which require the introduction of various counterterms (see [24] for a two-loop computation of
the Higgs mass term). The crucial point of identifying the Higgs with a Wilson line phase is the
impossibility of having a local mass counterterm at any order in perturbation theory. This is true
both in the Lorentz invariant and in the Lorentz breaking case. For this reason, the Higgs mass
is computable. In the Lorentz invariant case, it weakly depends on the remaining counterterms
needed to cancel divergencies loop by loop. In the Lorentz breaking case, the Higgs mass still
weakly depends on higher-loop counterterms, with the exception of w2 , which is an arbitrary
parameter directly entering in the Higgs mass formula. However, as we have already pointed
out, w always enters in any physical observable together with R, so that it can be rescaled
by redefining the radius of compactification. Thanks to this property, the Higgs mass remains
computable also in a Lorentz non-invariant scenario.
3. Phenomenological bounds
A full and systematic analysis of all the new physical effects predicted by our theory is a quite
complicated task, mainly because it mostly depends on how flavor is realized. For simplicity,
we focus in the following on flavor and CP conserving new physics effects. Due to the strong
constraints on flavor changing neutral currents (FCNC) and CP violation in the SM, this is a
drastic but justified simplification, since flavor does not play an important role in our mechanism
of EWSB. In fact we even did not discuss an explicit realization of it in our model (see however
[4,25] for possible realizations and [5] for an order of magnitude estimate of the bounds arising
from FCNC). The only exception to this flavor universality is given by the third quark family,
that has to be treated separately.
The analysis of flavor and CP conserving new physics effects requires in general the introduction of 18 dimension 6 operators in the SM to fit the data (see e.g. [11,12]). It has recently been
pointed out in [13] that, out of these 18 operators, only 10 are sensibly constrained. In a given
basis (see [13] for details), 7 of these operators are parametrized by the universal parameters S,
T , U , V , X, W and Y introduced in [14] extending the usual S, T , U basis [26]. They are defined
in general in [14], starting from the inverse propagators i,j of the linear combinations of gauge
bosons (not necessarily mass eigenstates of the Lagrangian) that couple to the SM fermions. As
in the examples discussed in [14], in our model these gauge fields coincide with the holographic electroweak bosons which appear in Eq. (2.12). The remaining 3 operators are parametrized
by the distortion gb (or the b parameter [15]) of the Z bL bL coupling and other two parameters
which describe the deviation of the up and down quark couplings to the Z boson. As before, we
put a hat superscript to distinguish the holographic field from its corresponding 5D field or mass
eigenstate. Due to the way in which SM fermions couple in our model, the later two parameters
are totally negligible. Being all light fermions almost completely localized at y = 0 (see Fig. 7
in Appendix B for a quantitative idea of this effect), their couplings with the SM gauge fields are
universal and not significantly distorted. As in other models based on extra dimensions [6,27,28],
a possible exception is the bottom quark coupling (the top even more, but its coupling to the Z is
200
at the moment practically unconstrained), which is instead typically distorted by new physics.
In the original version of our model considered in [5], indeed, gb turned out to give one of the
strongest constraints on the model. As anticipated before and as we will now see in some detail,
the simple idea of reversing the mixing of the localized bottom and top quark with respect to [5]
strongly reduces gb .7
To be more precise, the distortion of the Z bL bL coupling is due to two effects. One of those
is the mass-term mixing of the bL with the KK tower of the SU(2)w triplet and singlet fermions
coming from the bulk field in the rep. 6 of SU(3)w . The second distortion is a consequence of the
Both
delocalization of bL , which then also couples to gauge bosons components orthogonal to Z.
effects are proportional to the mixing parameters 1t,b . Interestingly enough, at leading order in
= mW R, the former distortion exactly vanishes in the (very good) approximation of neglecting
the bottom mass. This can be understood by noting that the bL mixes with the component of an
SU(2)w triplet which has T3 = 1 and with a singlet (T3 = 0). The mixing with the triplet state
leads to an increase (in magnitude) of gb , whereas the singlet makes it decrease. The two effects
turn out to exactly compensate each other. The distortion due to the delocalization of bL is instead
non-vanishing. An explicit computation along the lines of the one reported in Appendix A gives,
at leading order in :
2
g SM 2 2 1i
i
F
,
gb = gb gbSM = bb
(3.1)
ki
2i
4Z1 cos2 W
i=t,b
where F (x) is defined in Eq. (A.8) and Z1b is the wave function normalization factor needed to
canonically normalize the bL field, defined in Eq. (2.18) of [5] (recall t b with respect to [5]).
We have verified the validity of Eq. (3.1) by computing it both holographically and from a more
direct KK analysis. In the latter case, the distortion is caused by the KK tower of the AX gauge
field, which mixes to the lowest mass eigenstate of the Z boson and couples to bL in the bulk. It is
worth to notice that the two results agree completely, but only when one sets the Z boson on-shell
in the holographic approach, as it should be done. This result is expected, if one considers that at
the pole, the Z gauge boson is totally given by its lowest mass eigenstate Z.8 In the holographic
approach, as shown in Appendix A in a toy example, one would find gb to be identically zero
if it is computed at zero momentum. Since F (x) is a positive function, gb < 0. In the typical
range of parameters of phenomenological interest, gb 2 2 and it comfortably lies inside the
experimentally allowed region (at 1 form the central value) if 3 102 . As we will now
discuss, stronger bounds on arise from universal corrections and non-universal corrections to
Z bL bL can be neglected in our global fit.
T , U , V , X, W and Y . From the inverse gauge propagators
We are left with the 7 parameters S,
i,j in Eq. (2.12), one easily finds, at tree-level and at leading order in ,
S = 23 2 2 ,
4 4
X = 14
45 tan(w ) ,
T = 2 2 ,
2
2
w )+1+2 cos(2w ) 2 2
(3.2)
,
Y = sin (
2 4 4
9 2 cos2 (w )
U
=
W = 13 2 2 .
6 6
V = 14
45 ,
7 We would like to thank G. Cacciapaglia for this observation.
8 Strictly speaking, due to the Z width, there will still be left a combination of the other KK eigenstates, but this effect,
not visible at tree-level, is totally negligible.
201
Fig. 3. Constraints coming from a 2 fit on the EWPT. The contours represent the allowed regions in the (1/R, mH )
plane at 68%, 90% and 99% confidence level (2 d.o.f.). The shaded band shows the experimentally excluded values
for the Higgs mass (mH < 115 GeV). The blue dots represent the predictions of our model for different values of the
microscopic parameters (only points with the correct top and bottom masses are plotted).
Since has to roughly be of order 102 , we see from Eq. (3.2) that the 3 parameters U , V
and X are totally negligible. This is actually expected for any universal theory [14]. Notice that
the universal parameters essentially depend only on , the other parameter entering is , which
affects only Y . We have verified that radiatively generated localized gauge kinetic terms give
negligible effects in Eq. (3.2) and can thus be consistently neglected.
In Fig. 3, we report the constraints on the Higgs mass and the compactification scale due to
all electroweak flavor and CP conserving observables, obtained by a 2 fit using the values in
Eq. (3.2).9 To better visualize the results, we fixed the value of the parameter to the Lorentzsymmetric value ( = 1), hence we used a 2 fit with 2 d.o.f. The fit essentially does not depend
on , as can be checked determining by a minimization of the 2 function, which gives a plot
which is almost indistinguishable from that reported in Fig. 3. Due to their controversial interpretation (see e.g. [29]), we decided to exclude from the fit the NuTeV data. Nevertheless, we
verified that the inclusion of such experiment leaves our results essentially unchanged. From
Fig. 3 one can extract a lower bound on the compactification scale 1/R 45 TeV (which
corresponds to 0.0160.02) and an upper bound on the Higgs mass which varies from
mH 600 GeV at 1/R 4 TeV to mH 250 GeV for 1/R 10 TeV. Notice that the values
W and Y (for = 1) found in Eq. (3.2) are exactly those expected in a generic 5D theory
for S,
with gauge bosons and Higgs in the bulk, as reported in Eq. (20) of [14]. On the contrary, the T
parameter in Eq. (3.2) is much bigger and is essentially due to the anomalous gauge field AX .
9 We would like to thank A. Strumia for giving us the latest updated results for the 2 function obtained from a
combination of experimental data (see Table 2 of [14] for a detailed list of the observables included in the fit), as a
function of the universal parameters of Eq. (3.2).
202
Fig. 4. Inverse sensitivity for those points which pass the 2 test.
A heavy Higgs is allowed in our model because of this relatively high value of the T parameter,
which compensates the effects of a high Higgs mass.
Analyzing the models obtained by a random scan of the microscopic parameters, we found
that the bound on the compactification scale can be easily fulfilled if a certain fine-tuning on the
parameters is allowed. On the other hand, the bounds on the Higgs mass does not require any
tuning.
3.1. Sensitivity, predictivity and fine-tuning
As expected, the constraints arising from the electroweak observables require a quite stringent
bound, 1/R 4.5 TeV, on the size of the compactification scale. Differently from previously
considered scenarios of GHU in flat space (see e.g. [4]), our model can satisfy such a bound (see
Fig. 3) for suitable choices of the microscopic input parameters. It is clear, however, that a certain
cancellation (fine-tuning) must be at work in the Higgs potential to make its minimum = mW R
as low as 0.018.
The fine-tuning is commonly related [8] to the sensitivity of the observable (, in our case)
with respect to variations of the microscopic input parameters.10 We define

log
,
C Max
(3.3)
log I
as the maximum of the logarithmic derivatives of with respect to the various input parameters I .
The fine-tuning, defined as f = 1/C, is plotted in Fig. 4 for the points which pass the 2 test.
For 1/R 4.5 TeV, the fine-tuning of our model is 1%.
According to a more refined definition [9], the fine-tuning should be intended as an estimate of
how unlike is a given value for an observable. This definition is not directly related to sensitivity,
even though it reduces to f = 1/C in many common cases. In this framework, the fine-tuning
is computed from the probability distribution () of the output observable for reasonable
10 In the particular case at hand, the sensitivity of is related to the cancellation of the Higgs mass term in the effective
potential. As a consequence, one expects the amount of tuning to scale roughly as 2 , as confirmed numerically.
203
Fig. 5. Bayesian measure of the fine-tuning at fixed (black) and not fixed (red) top mass. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version in this article.)
distributions of the input microscopic parameters. One defines [9]

()
,
(3.4)
()

where11 () = d [()]2 . The choice of the range of variation of the input parameters12
is the strongest ambiguity of the procedure, and must reflect physically well motivated assumptions. The distribution shown in Fig. 1, which we will use, has been obtained with quite generic
ranges, which however reflect our prejudice on the physical size of the top and bottom quark
masses. Even though this does not make an important difference (see Fig. 5), we can make our
assumption more precise by further restricting to points in which the top (not the b, due to the
lack of statistic) has the correct mass. The straightforward application of Eq. (3.4) gives rise to
the fine-tuning plot of Fig. 5. At 1/R = 4.5 TeV this prescription gives a fine-tuning of 10%. One
could also relax the assumption of having EWSB. Since the fraction of points without EWSB is
about one half of the total, one could argue, very roughly, that the resulting fine-tuning will increase of a factor two and be of order 5%. This result strongly depends on the assumption we did
on the fermion mass spectrum. In practice, what we observe is a certain correlation among the
requirements of having a massive top and an high compactification scale, in the sense that points
with massive top (and light b) also prefer to have small minima.
Independently on how we define the fine-tuning, however, the high value of C at small
represents a problem by itself. The high sensitivity, indeed, makes unstable against quantum
corrections or deformations of the Lagrangian with the inclusion of new (small) operators. As
described in Section 2.2, we have considered the effects on the observables of localized kinetic
terms for bulk fermions. We found that such operators, which lead to very small corrections to
all other observables considered in this paper, can completely destabilize the compactification
scale at small . A similar effect is found when including in the effective potential the very
f=
11 Our definition of is not precisely the one given in [9]. The difference however disappears when, as in our case,
the input variables are uniformly distributed.
12 We have checked that our results are independent on the detailed form of the probability distribution of the input.
This is because the ranges which we consider are not so wide. Uniform distributions have been used to derive the results
which follow.
204
small contributions which come from the light fermion families. For 0.02 the corrections
which come in both cases are of order 50%, so that the compactification radius is effectively not
predicted in terms of the microscopic parameters of the model.
4. Conclusions
In this paper we presented a realistic GHU model on a flat 5-dimensional orbifold S 1 /Z2 . The
model is based on the one outlined in [5], in which an explicit breaking of the SO(4, 1) Lorentz
symmetry down to the usual 4-dimensional Lorentz group SO(3, 1) was advocated in order to
overcome some of the most worrisome common problems of GHU models on flat space.
The main new feature of the present model is the presence of a Z2 mirror symmetry, which
essentially consists in a doubling of part of the bulk fields, contributing to a reduction of the finetuning. It also allows to have a stable boson at the TeV scale, which could prove to be a viable
dark matter candidate. We have also shown how, due to a suitable choice of the bulk fermion
quantum numbers, the deviations from the Z bL bL SM coupling can be made totally negligible.
T , W and Y defined in [14], and
The new physics effects can be encoded in the 4 parameters S,
thus our model effectively belongs to the class of universal theories. From a combined fit, we
derived a lower bound on the compactification scale 1/R 4 TeV with allowed (although never
reached, see Fig. 3) Higgs masses up to 600 GeV. Our model is compatible with the experimental
constraints if a certain tuning in the parameter space, of order of a few %, is allowed. Such a
fine-tuning is certainly acceptable, but it is nevertheless still too high to claim that our model
represents a solution to the little hierarchy problem. However, it should be emphasized that the
fine-tuning we get is of the same order of that found in the MSSM.
Acknowledgements
We would like to thank G. Cacciapaglia for participation at the early stages of this work and
A. Strumia for providing us the complete 2 function used in this work and for useful discussions. We would also like to thank C. Grojean, A. Pomarol, R. Rattazzi and A. Romanino for
interesting discussions. This work is partially supported by the European Communitys Human
Potential Programme under contract MRTN-CT-2004-005104 and by the Italian MIUR under
contract PRIN-2003023852.
Appendix A. The holographic approach
Deriving 4D effective theories from 5D models is usually done by integrating out the massive
KK eigenstates, keeping the lightest states of the KK towers as effective degrees of freedom.
When the extra space has a boundary at y = 0, an alternative and useful holographic possibility
is to use the boundary values (x)

= (x, 0) of the 5D fields as effective degrees of freedom and
to integrate out all the fields components (x, y) with y = 0 [18]. As long as (x, 0) has a nonvanishing component of the light mode of the KK tower associated to (x, y), the holographic
and KK approaches are completely equivalent. In warped 5D models, such a prescription closely
resembles the one defined by the AdS/CFT correspondence [30], where roughly speaking (x)
is the source of a CFT operator and (x, y) is the corresponding AdS bulk field. Although in
flat space such interpretation is missing, it has been emphasized in [18] that it is nevertheless a
useful technical tool to analyze various properties of 5D theories. The holographic approach, in
205
this paper, has been used to perform the global fit to EW observables. Along the lines of [14], it
significantly simplifies the analysis.
We think it might be useful to collect here a few technical details on how to use this approach in our particular context. To keep the discussion as clear as possible, and to emphasize
the key points, we consider in this appendix a simple 5D Lorentz invariant model on a segment of length L, which resembles some aspects of our construction. Let AM be an U (1)
bulk gauge field, and a couple of bulk charged fermions and qR a chiral 4D fermion
localized at y = 0, which mixes with L . The gauge field components A and A5 satisfy respectively Neumann and Dirichlet boundary conditions at y = L, whereas for the fermions we
have R (L) = L (L) = 0. For what concerns the bosonic fields, boundary conditions at y = 0
are simply imposed by the choice of the dynamical holographic fields to be retained. The holographic approach to fermions is a bit more complicated since one has to modify the action by
introducing suitable localized mass-terms [19]. With the convention 5 L = +L , the action
reads

1
1
D
dy d 4 x FMN F MN + (i
/ 5 + iD
/ 5 ) M + h.c.
4
2
0

1
+ d 4 x qR iD
/ 4 qR + eqR L (0) + e L (0)qR + (0)(0)
(0)(0) ,
2
L
S=
(A.1)
and one can check that, by performing generic variation of the bulk fermion fields, the boundary conditions R (0) = e/2qR and L (0) = 0 automatically arise as equations of motion.
The field which we will retain are A = A (0) and qR . Integrating out at tree-level the remaining degrees of freedom is equivalent to solve their equation of motion with fixed holographic fields ( A = qR = 0 in the action variation) and to plug the solution back into
the action. Varying Eq. (A.1) with the holographic prescription simply gives the standard 5D
Dirac and Maxwell equations plus the above mentioned boundary conditions for the fermions.
It is useful to adopt a mixed momentum-space basis (p, y) for the fields. In the unitary
gauge A5 vanishes and the longitudinal component of A is decoupled, while the equations for
the transverse part At = P A (P = p p /p 2 ) reduce to (p 2 + y2 )At (p, y) = 0.
Imposing the boundary condition y A (L) = 0, one easily gets
A (p, y) =
cos[p(L y)] t
A (p),
cos[pL]
(A.2)

where p = p p . One proceeds analogously for fermions. It is a simple exercise, given the
boundary conditions at y = L, to determine (p, y) and (p, y) in terms of qR . One gets

R = 2e sin[(Ly)]
/ qR ,
L = 2e cos[(Ly)]
sin[L] qR ,
sin[L] p
(A.3)
= e cos[(Ly)] Mq ,
L = 0,
R
sin[L]
where = p 2 M 2 . Plugging back the solutions (A.2) and (A.3) into the action (A.1), we can
get the holographic Lagrangian. At quadratic level
1
1
Lholo = (p)A t A t + Zq qRp
/ 4 qR ,
2
2
(A.4)
206
where we defined
e2 cot L
(A.5)
.
2
The holographic procedure can also be used to study interaction terms. Again, one has simply
to substitute into the action (A.1) the solutions (A.2) and (A.3). It is useful to see how this works
by computing the distortion to the U (1) gauge coupling g of qR to A due to the massive modes.
This interaction is encoded in the following Lagrangian term:
(p) = p tan pL,
L
L
(3)
=g
Zq = 1

+ q, y)A
dy (p
/ (q, y)(p, y) + (p + q, y)A
/ (q, y)(p, y)
g
+ qR (p + q)A
(A.6)
/ (q)qR (p).
2
In order to simplify our computation, let us consider the kinematic configuration in which the
fermion is on-shell, p 2 = (p + q)2 = 0, and q 2 = 2 M 2 . By direct substitution and after a
bit of algebra, one gets the following cubic interaction, up to terms of order 2 /M 2 :

g
e 2 2
(3)
L =
(A.7)
F (ML) qR (p + q)A
/ t (q)qR (p),
Zq +
2
8M 2
where Zq is computed at zero momentum and we have defined
F (x) 1
1
coth x + coth2 x.
x
(A.8)
As expected, the form of the interaction vertex at 2 = 0 has precisely the same structure as the
quadratic term in (A.4), as required by gauge invariance, which forbids any correction to g. At
quadratic order in the gauge boson momentum, however, we get a correction
g =
ge2 2
F (ML).
8Zq M 2
(A.9)
Notice that Eq. (A.9) coincides exactly with Eq. (3.1), when fixing 2 = m2Z and taking into
account the SM coupling of the bL . This is a further proof that no corrections arise in our model
from the mixing of bL with fields with different isospin quantum numbers, the only effect being
given by the partial delocalization of the field, resulting in couplings with the massive gauge
fields A (y), with y = 0.
The generalization to the non-Abelian case and fields in different representations of the gauge
group does not present any conceptual problem. When a Wilson line symmetry breaking occurs,
like in our model, the most convenient thing to do is to go in a gauge in which the VEV for
A5 vanishes and the boundary conditions for the various fields are twisted at y = 0. In this way,
the effect of the twist simply amounts to a redefinition of the holographic fields, as discussed in
Section 2.1.
Appendix B. Fermion wave functions
In this appendix we turn to the more standard analysis of the KK mass eigenstates and examine
the wave functions of the fermions arising from the mixing of the bulk fields with the fields
localized at y = 0. The purpose of this analysis is to illustrate how much the SM fermions are
207
Fig. 6. Top (red) and bottom (blue) quark wave function profiles (right-handed components). The areas below the lines
represent the amount of the delocalized part of the fields. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version in this article.)
Fig. 7. Delocalized part of the quark wave functions. The fraction of bulk wave function is shown for the various quarks.
localized in our model. As we will see, in fact, the top quark is not localized at all, whereas the
bottom is only partially localized (see Figs. 6 and 7). In order to simplify the discussion, we
focus on the wave functions of right-handed singlets before EWSB. Left-handed localized fields
are more involved since they couple to two different bulk fields, as shown in Eq. (2.7).
The relevant quadratic Lagrangian describing the coupling of a localized right-handed fermion
with the bulk fields is easily extracted from the full Lagrangian (2.4). It reads
4 k5 5 ) M( + )
4 k5 5 ) + (i/
L = (i/
4 qR + eqR L + e L qR ],
+ (y)[qR i/
(B.1)
where and are the singlet components of the periodic bulk fermions +t and +b , respectively
for qR = tR and qR = bR . For simplicity of notation, we denote the bulk-to-boundary mixing
parameter simply as e in both cases. Due to the latter and the bulk mass M, the equations of
(n)
motion for , and qR are all coupled with each other. The 4D KK mass eigenstates L,R (x)
of Eq. (B.1) will then appear spread between the fields , and qR . In other words, we expand
208
the various fields as follows:

(n)
q = n gn R ,
(n)
(n)
L,R = n fL,R (y)L,R ,
(n)

(n)
L,R = n fL,R (y)L,R ,
(B.2)
(n)
(n)
where fL,R (y) and fL,R (y) are the wave functions along the fifth dimension and gn are constants. The fields satisfy the 4-dimensional Dirac equation
i/
4 (n) = mn (n) ,
(n)
(B.3)
(n)
where (n) = L + R , and mn is the mass of the nth state. By solving the equations of
motion, one finds two distinct towers of eigenstates: an unperturbed tower with only massive
fields and a perturbed one containing a right-handed massless field. We will also see how the
mass spectra found here with the usual KK approach match with the one deduced from Eq. (A.5)
using the holographic approach.
B.1. The unperturbed tower
This tower has no component along the localized fields and is entirely build up with bulk
fields. The wave functions are analytic over the whole covering space and the left-handed ones
vanish at y = 0. The mass levels are given by

2
kn
,
mn = M 2 +
(B.4)
R
but with n 1, without the n = 0 mode. The corresponding wave functions are
(n)
gn = 0,
(n)
fR =
sin( Rn y),
mn R
fL = 0,
f(n) = nk/R
cos( Rn y).
f(n) = 1 sin( n y),
R
L
(B.5)
mn R
Although e does not explicitly appear in Eqs. (B.4) and (B.5), its presence constrains the lefthanded fields to vanish at y = 0 and thus it indirectly enters in the above expressions. When e
vanishes, indeed, one recovers the usual mode n = 0.
B.2. The perturbed tower
This tower is distorted by the coupling with the localized fields. Its massive levels are given
by the solutions of the transcendental equation

2
n
,
=
tan R
(B.6)
k
Rkn

where n (m2n M 2 ). Notice that the masses defined by Eq. (B.6) exactly coincide, for
k = 1, with the zeroes of Zq in Eq. (A.5). The mass equation has a tower of solutions with
mn > M, whose wave functions are13
f (n) = mn gn cos((yR)n /k) ,

L
sin(Rn /k)
f(n) = 0,
2Rkn
(n)
fR =
gn sin((yR)n /k) ,

2Rk sin(Rn /k)
f(n) = Mgn cos((yR)n /k) .
R
sin(Rn /k)
2Rkn
209
(B.7)
The constants gn in Eq. (B.2) are determined by imposing the canonical normalization of the
4D fields. One gets

gn = 1 +
1/2

2
Rm2n
M 2 n2
cot(R
/k)
.
+
k
n
n
2Rk 2 n2 sin2 (Rn /k)
(B.8)
Notice that some care has to be used in taking the limit 0 in the above expressions, since
one encounters apparently ill-defined expressions in Eqs. (B.7) and (B.8).
B.3. The zero mode
The zero mode is of particular interest, being identified with a SM field. Its wave function is
(0)
(0)
fR = g0 sinh((yR)M/k)
,
fL = 0,
2Rk sinh(RM/k)
(B.9)
f(0) = g0 cosh((yR)M/k) ,
f(0) = 0,
L
2Rk
sinh(RM/k)
where again g0 is determined by the normalization conditions and equals

g0 = 1 +
1/2
2
.
coth(RM/k)
RMk
(B.10)
The constant g0 indicates how much of the zero mode field is composed of the localized field.
When 1 and g0 1, as expected, the wave function of the massless state is mostly given
by the localized field, the bulk components (B.9) being proportional to and thus small. On the
contrary, for large mixing 1, one has g0 1 and the localized field is completely dissolved
into the bulk degrees of freedom. In the latter case, then, the localization of the R and R
components of the zero mode is essentially determined by the adimensional quantity RM/k.
The bulk wave functions at y = R are suppressed by a factor exp(RM/k) with respect to
their value at y = 0 so that, independently of , for large values of RM/k the chiral field is still
localized at y = 0.
Summarizing, the zero mode is localized at y = 0 for small mixing with the bulk fields and
arbitrary bulk masses, or for large bulk masses and arbitrary mixing. The requirement of having
the correct top mass after EWSB implies a large mixing and a small bulk mass, giving rise to
a delocalized top wave function. In Fig. 6 we illustrate the wave function profiles of the righthanded top and bottom quark components for typical acceptable values of input parameters,
before EWSB.
The mass spectrum and fermion wave functions for the antiperiodic fermions and is
straightforward, since they do not couple with the localized fields. Their wave functions can be
13 In Eq. (B.7) and in Eq. (B.9) we report the wave functions for 0 y R, the continuation for generic y can be
obtained using the properties of the fields under parity and under translation.
210
written as
(n)
fL =
f(n) =
L
1
2R
1
2R
cos( n+1/2
R y),
(n)
fR =
sin( n+1/2
R y),
f(n) =
R
M(n+1/2)k/R
sin( n+1/2
R y),
mn 2R
M(n+1/2)k/R
cos( n+1/2
R y),
mn 2R
(B.11)
where n 0 and stands for the two towers of mass eigenstates, both with masses given by

k(n + 1/2) 2
.
mn = M 2 +
(B.12)
R
After EWSB, all above relations are clearly modified by O() effects. In particular, a small
fraction of the SM fields is now spread also among the bulk fermion components whose mixing
were forbidden by SU(2)L U (1)Y symmetry. We do not write the modification to the mass
formula (B.6) for the deformed tower, the form of which can be deduced from the zeroes of the
integrand of Eqs. (2.15) and (2.16) of [5], with the replacement t b.
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[2] D.B. Fairlie, Phys. Lett. B 82 (1979) 97;
D.B. Fairlie, J. Phys. G 5 (1979) L55;
N.S. Manton, Nucl. Phys. B 158 (1979) 141;
P. Forgacs, N.S. Manton, Commun. Math. Phys. 72 (1980) 15;
S. Randjbar-Daemi, A. Salam, J. Strathdee, Nucl. Phys. B 214 (1983) 491;
N.V. Krasnikov, Phys. Lett. B 273 (1991) 246;
H. Hatanaka, T. Inami, C. Lim, Mod. Phys. Lett. A 13 (1998) 2601, hep-th/9805067.
[3] 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;
I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20, hep-th/0108005;
M. Kubo, C.S. Lim, H. Yamashita, Mod. Phys. Lett. A 17 (2002) 2249, hep-ph/0111327;
C. Csaki, C. Grojean, H. Murayama, Phys. Rev. D 67 (2003) 085012, hep-ph/0210133;
G. Burdman, Y. Nomura, Nucl. Phys. B 656 (2003) 3, hep-ph/0210257;
N. Haba, M. Harada, Y. Hosotani, Y. Kawamura, Nucl. Phys. B 657 (2003) 169, hep-ph/0212035;
N. Haba, M. Harada, Y. Hosotani, Y. Kawamura, Nucl. Phys. B 669 (2003) 381, Erratum;
N. Haba, Y. Shimizu, Phys. Rev. D 67 (2003) 095001, hep-ph/0212166;
K.W. Choi, et al., JHEP 0402 (2004) 037, hep-ph/0312178;
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, Phys. Rev. D 69 (2004) 075006, hep-ph/0311127;
C.A. Scrucca, M. Serone, L. Silvestrini, A. Wulzer, JHEP 0402 (2004) 049, hep-th/0312267;
N. Haba, Y. Hosotani, Y. Kawamura, T. Yamashita, Phys. Rev. D 70 (2004) 015010, hep-ph/0401183;
C. Biggio, M. Quiros, Nucl. Phys. B 703 (2004) 199, hep-ph/0407348;
K.Y. Oda, A. Weiler, Phys. Lett. B 606 (2005) 408, hep-ph/0410061;
Y. Hosotani, M. Mabe, Phys. Lett. B 615 (2005) 257, hep-ph/0503020;
A. Aranda, J.L. Diaz-Cruz, Phys. Lett. B 633 (2006) 591, hep-ph/0510138;
G. Cacciapaglia, C. Csaki, S.C. Park, JHEP 0603 (2006) 099, hep-ph/0510366.
[4] C.A. Scrucca, M. Serone, L. Silvestrini, Nucl. Phys. B 669 (2003) 128, hep-ph/0304220.
[5] G. Panico, M. Serone, A. Wulzer, Nucl. Phys. B 739 (2006) 186, hep-ph/0510373.
[6] K. Agashe, R. Contino, A. Pomarol, Nucl. Phys. B 719 (2005) 165, hep-ph/0412089;
K. Agashe, R. Contino, Nucl. Phys. B 742 (2006) 59, hep-ph/0510164.
[7] B. Grzadkowski, J. Wudka, 5-dimensional difficulties of gauge-Higgs unifications, hep-ph/0604225.
[8] R. Barbieri, G.F. Giudice, Nucl. Phys. B 306 (1988) 63.
[9] G.W. Anderson, D.J. Castano, Phys. Lett. B 347 (1995) 300, hep-ph/9409419.
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
211
G. Servant, T.M.P. Tait, Nucl. Phys. B 650 (2003) 391, hep-ph/0206071.

Z. Han, W. Skiba, Phys. Rev. D 71 (2005) 075009, hep-ph/0412166.
C. Grojean, W. Skiba, J. Terning, Phys. Rev. D 73 (2006) 075008, hep-ph/0602154.
G. Cacciapaglia, C. Csaki, G. Marandella, A. Strumia, The minimal set of electroweak precision parameters, hepph/0604111.
R. Barbieri, A. Pomarol, R. Rattazzi, A. Strumia, Nucl. Phys. B 703 (2004) 127, hep-ph/0405040.
G. Altarelli, R. Barbieri, F. Caravaglios, Nucl. Phys. B 405 (1993) 3.
I. Gogoladze, C. Macesanu, Precision electroweak constraints on universal extra dimensions revisited, hepph/0605207.
T. Appelquist, H.C. Cheng, B.A. Dobrescu, Phys. Rev. D 64 (2001) 035002, hep-ph/0012100.
R. Barbieri, A. Pomarol, R. Rattazzi, Phys. Lett. B 591 (2004) 141, hep-ph/0310285.
R. Contino, A. Pomarol, JHEP 0411 (2004) 058, hep-th/0406257.
Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Phys. Lett. B 129 (1983) 193;
Y. Hosotani, Ann. Phys. 190 (1989) 233.
F. del Aguila, M. Perez-Victoria, J. Santiago, JHEP 0302 (2003) 051, hep-th/0302023.
F. del Aguila, M. Perez-Victoria, J. Santiago, Effective description of brane terms in extra dimensions, hepph/0601222.
A. Lewandowski, R. Sundrum, Phys. Rev. D 65 (2002) 044003, hep-th/0108025.
N. Maru, T. Yamashita, Two-loop calculation of Higgs mass in gauge-Higgs unification: 5D massless QED compactified on S 1 , hep-ph/0603237.
G. Martinelli, M. Salvatori, C.A. Scrucca, L. Silvestrini, JHEP 0510 (2005) 037, hep-ph/0503179.
M.E. Peskin, T. Takeuchi, Phys. Rev. D 46 (1992) 381.
K. Agashe, A. Delgado, M.J. May, R. Sundrum, JHEP 0308 (2003) 050, hep-ph/0308036.
G. Burdman, Y. Nomura, Phys. Rev. D 69 (2004) 115013, hep-ph/0312247;
G. Cacciapaglia, C. Csaki, C. Grojean, J. Terning, Phys. Rev. D 71 (2005) 035015, hep-ph/0409126.
G. Altarelli, M.W. Grunewald, Phys. Rep. 403404 (2004) 189, hep-ph/0404165.
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.
Magnetic dipole operator contributions to the photon

energy spectrum in B Xs at O(s2)
H.M. Asatrian a , T. Ewerth b, , A. Ferroglia c , P. Gambino d , C. Greub b
a Yerevan Physics Institute, 375036 Yerevan, Armenia
b Institute for Theoretical Physics, University of Berne, CH-3012 Berne, Switzerland
c Physikalisches Institut, Albert-Ludwigs-Universitt, D-79104 Freiburg, Germany
d INFN, Sezione di Torino & Dipartimento Fisica Teorica, Universit di Torino, I-10125 Torino, Italy
Received 14 August 2006; accepted 2 November 2006

Abstract
We compute the O(s2 ) contributions to the photon energy spectrum of the inclusive decay B Xs
associated with the magnetic penguin operator O7 . They are an essential part of the ongoing NNLO calculation of this important decay. We use two different methods to evaluate the master integrals, one based on
the differential equation approach and the other on sector decomposition, leading to identical results which
in turn agree with those of a recent independent calculation by Melnikov and Mitov. We study the numerical relevance of this set of NNLO contributions in the photon energy spectrum and discuss the change of
bottom quark mass scheme.
1. Introduction
More than a decade after their first direct observation, radiative B decays have become a key
element in the program of precision tests of the Standard Model (SM) and its extensions. The
inclusive decay B Xs is particularly well suited to this precision program thanks to the
low sensitivity to non-perturbative effects. The present experimental world average [1] for the
branching ratio of B Xs has a total error of about 6% and agrees well with the SM prediction,
that is subject to a considerably larger uncertainty [2]. In view of the final accuracy expected at
the B factories, about 5%, the SM calculation needs to be improved. It presently includes next* Corresponding author.
E-mail address: tewerth@itp.unibe.ch (T. Ewerth).

doi:10.1016/j.nuclphysb.2006.11.002
H.M. Asatrian et al. / Nuclear Physics B 762 (2007) 212228
213
to-leading order (NLO) perturbative QCD corrections as well as the leading non-perturbative
and electroweak effects (see e.g. [3] for a complete list of references). The calculation of next-tonext-to-leading order (NNLO) QCD effects is currently under way and is expected to bring the
theoretical accuracy to the required level.
Among the NNLO QCD contributions, the following have already been computed: (i) all
relevant O(s2 ) Wilson coefficients [4]; (ii) the relevant three loop anomalous dimension matrix
[5,6]; (iii) the O(s2 NF ) corrections to the matrix elements of the operators O1 , O2 , O7 , O8 [7];
(iv) those matrix elements related to the dominant electromagnetic dipole operator O7 [810].
Among the pieces still missing to date, there are two rather important and difficult ones. The first
concerns the four-loop anomalous dimension matrix elements that describe the O(s3 ) mixing of
the four-quark operators into O7 and O8 ; preliminary results on this have been recently presented
[11]. The second concerns the three-loop, O(s2 ), matrix elements of the four-quark operators O1
and O2 . These loops contain the charm quark, and the ambiguity associated to the choice of scale
for its mass is the main source of uncertainty at NLO [2]. A NNLO calculation of these matrix
elements is therefore crucial to reduce the overall theoretical error on the branching ratio [12].
Other missing contributions, like the O(s2 ) matrix elements of O8 , may also prove essential to
reach a high theoretical precision.
In addition to the total branching fraction, the photon energy spectrum in B Xs is also
a useful observable: it receives both perturbative and non-perturbative contributions and is important for several reasons: (i) the precise knowledge of the spectrum is necessary to predict the
measured branching ratio since experiments apply a lower cut of 1.82.0 GeV on the photon
energy in the B center-of-mass frame; (ii) it is almost insensitive to new physics; (iii) from the
moments of the truncated spectrum one can extract relevant information on the parameters of
the heavy quark expansion. In particular, a precise value of the b quark mass can be extracted
from the mean value E of the photon energy [13,14]; (iv) the measured spectrum gives direct information on the Shape Function that encodes the QCD dynamics in the endpoint region,
where non-perturbative contributions dominate and perturbative corrections must be resummed
[15]. The NNLO Sudakov resummation was first completed in [16] and there has been recent
progress towards NNLO in the multi-scale OPE approach [17]. A detailed knowledge of the
Shape Function is useful for the determination of |Vub | from inclusive semileptonic B decays.
The perturbative contributions to the photon energy spectrum are known at O(s ) [18] and
O(s2 0 ) [19] since several years. In this paper we present a calculation of the O(s2 ) photon
spectrum induced by the magnetic operator O7 . This operator gives the dominant contribution to
the spectrum at O(s ). Our results, which were obtained by using two different methods, confirm
those of a recent paper [10].
Our paper is organized in the following way. In Section 2 we present our analytical results
for the unnormalized photon energy spectrum and compare with the literature. In this section we
also compare numerically the NLO and the NNLO spectra, and study the impact of a change of
the bottom quark mass scheme. Section 3 is devoted to a detailed description of the techniques
used for our calculation. Finally, we give a brief summary in Section 4.
2. Results and applications
2.1. Analytical results for the photon energy spectrum
Within the low-energy effective theory the partonic b Xs decay rate can be written
as
214

parton
b Xs
E
=
>E0
2 eff
G2F em m
2b ()m3b
Vtb Vts
Ci ()Cjeff () Gij (E0 , ),
4
32
i,j
(2.1)
where mb and m
b () denote the pole and the running MS mass of the b quark, respectively,
Cieff () the effective Wilson coefficients at the low energy scale, and E0 the energy cut in the
photon spectrum. Here we give the result for the function G77 (E0 , ) corresponding to the selfinterference of the electromagnetic dipole operator

e
m
()
s
P
b
F
O7 =
(2.2)
b
R
16 2
including O(s2 ) terms as required for NNLO accuracy. Introducing the dimensionless variable
2E
(2.3)
,
mb
this function can be written as an integral over the (rescaled) photon energy spectrum
dG77 (z, )/dz, i.e.,
z=
1
G77 (E0 , ) =
dG77 (z, )
dz,
dz
(2.4)
z0
where z = z0 corresponds to E = E0 . When working to the required precision, the photon

energy spectrum can be separated into three different parts,
dG77 (z, ) dG1n
77 (z, )
=
,
dz
dz
4
(2.5)
n=2
corresponding to the n particles in the final state, namely the b s (n = 2), b s g (n = 3),
b s gg and b s q q (both n = 4, q {u, d, c, s}) transitions. Furthermore, each individual
contribution can itself be written in the form
dG12
77 (z, )
= f2 ()(1 z),
dz
dG13
77 (z, )
= f3 ()(1 z) + R3 (z, ),
dz
dG14
77 (z, )
= f4 ()(1 z) + R4 (z, ).
dz
Consequently, the complete result for the photon energy spectrum is of the form
(2.6)
dG77 (z, )
= F ()(1 z) + R3 (z, ) + R4 (z, ),
(2.7)
dz
with F () = f2 () + f3 () + f4 (). We stress here that the calculation of dG77 (z, )/dz boils
down to the determination of R3 (z, ) and R4 (z, ), because the coefficient F () in front of the
delta-function can then be fixed by the requirement that the total integral
1
G77 (0, ) =
0
dG77 (z, )
dz
dz
(2.8)
215
yields the result obtained in [8], which was recently confirmed in [9].
As we discuss the details of the calculation in Section 3, we immediately present our final
result for the photon energy spectrum (including also the order s0 and s1 pieces):

dG77 (z, )
s () 2
s ()
(1)
CF H (2) (z, )
= (1 z) +
CF H (z, ) +
dz

+ O s3 ,
(2.9)
where
H (2) (z, ) = CF H (2,a) + CA H (2,na) + TR NL H (2,NL) + TR NH H (2,NH) .
(2.10)
The numerical values of the color factors are CF = 4/3, CA = 3, and TR = 1/2. Furthermore, NL
and NH denote the number of light (mq = 0) and heavy (mq = mb ) quark flavors, that is the total
number of quark flavors is NF = NL + NH . The functions H (1) and H (2,j ) (j = a, na, NL, NH)
appearing in (2.9) and (2.10) are given by

5 2
ln(1 z)
H (1) =
+
+ L (1 z)
4
3
1z +

7
z+1
7 + z 2z2
1
(2.11)
ln(1 z) +
,
4 1z +
2
4

29 2
1
+
L + L2 (1 z)+C(z)L
H (2,a) = 8.10798 +
8
3
2
3
1 ln (1 z)
21 ln (1 z)
+
+
2
1z
8
1z
+
+

2

2
69
5
(3) 67
ln(1 z)
1
+
+
+
+
16
6
1z +
12
2
32 1 z +
1 + 2z 2z2 + z3 3
z3 4z2 + 4z + 1
ln (1 z) +
B(z)
4z
24(1 z)
8z6 46z5 + 64z4 3z3 27z2 + 21z 9
A(z)
+
24(1 z)3 z

9 + 5z + 7z2 + 5z3 3z4 + z5 (z2 + 8z 11) ln z

+
+
ln2 (1 z)
24z
8(1 z)
2
(3z 1 z2 ) z6 4z5 8z4 + 61z3 74z2 + 13z + 3
+
ln(2 z)
+
12
12z(1 z)
32z4 156z3 + 98z2 + 95z + 35

ln(1 z)
+
48
11 2z 9z2 + 2z3
Li2 (1 z) ln(1 z)
4(z 1)
32z5 + 144z4 + 68z3 + z2 297z 36
+
96z
2 (z5 3z4 21z3 + 41z2 + 19z 6) z3 + 3z2 + 10z 16
+
(3)
72z
4(1 z)
+
216
z6 4z5 8z4 + 61z3 74z2 + 13z + 3

12z(1 z)
2
3 + 2z 2z + z3
ln(1 z) Li2 (z 1)
+
2(z 1)
11 + 4z 17z2 + 4z3
Li3 (1 z) 2(1 z)2 Li3 (z 1)
4(z 1)
11 8z z2
Li3 (z),
+
4(1 z)
(2,na)
(2.12)

11
211 11 2
11 2
+
L + L (1 z) C(z)L
= 13.7256 +
36
18
12
6
2

2
ln(1 z)
11 ln (1 z)
95
+
+
+
8
1z
144
12
1z +
+

2
1
(3) 905 17
z(z 2)2 + 1
+
+
+
B(z)
4
288
72
1z +
48(z 1)
(z + 1)(15 57z + 73z2 29z3 + 2z4 )
A(z)
48(z 1)3

z6 4z5 2z4 + 54z3 74z2 + z 6
ln(1 z) ln(2 z) + Li2 (z 1)
+
24(z 1)z

(z + 2) 3
(z 1)2 3
ln (1 z)
z 5z2 + 9z 35 ln2 (1 z)
8
48
2 2
(z z + 3) 12z5 156z4 + 57z3 + 545z2 + 74z + 72
ln(1 z)
+
24
144z

(z 3)z
z3 2z2 + 2z 3
Li2 (1 z) +
Li2 (z 1) ln(1 z)
4
4(z 1)
+
12z4 138z3 628z2 + 659z + 671

288
2 (z5 3z4 3z3 + 34z2 24z + 3)
+
144z
z2 + 3z + 5
(z 3)z
Li3 (1 z) + (z 1)2 Li3 (z 1) +
(3),
+
2
8
+
(2.13)
with

A(z) = 2 ln(1 z) (z 1)2 + ln z(2 z) + Li2 (z 1)2 (z 1)2 ,

1
+ 2 ln(2 z) 6 ln2 (1 z) 2 ln2 (2 z) + 2
B(z) = 24 Li3
2z

z
z
+ 12 Li3 (z) 12 Li3
+ 12 Li3
15 (3),
2z
z2

ln(1 z)
7
1
1
1
C(z) =
+
+ (1 + z) ln(1 z) + 2z2 z 7 .
4 1z +
1z + 2
4
(2.14)
217
Finally,

(2,NL)

631 91 2 (3)
1 2
14 2 2
2
=
+
+
+
+
L + L (1 z)+ C(z)L
432
216
3
9
9
3
3

1 ln2 (1 z)
13 ln(1 z)
85 2
1
2
1z
36
1z +
72 18
1z +
+

2
2
z 3
2
1+z 2
+
Li2 (z)
ln (1 z) (1 + z)
6(1 z)
6
4
36
38z2 55z 49
6z3 25z2 z 18
ln(1 z) +
,
36z
72
(2.15)

3563 29 2 1
1
14 2 2
(3) +
+
L + L2 (1 z)
648
54
3
9
9
3
2
+ C(z)L .
(2.16)
3
z
Here, L = ln(/mb ), Li3 (z) = 0 dx Li2 (x)/x, (3) is the Riemann zeta-function, and [lnn (1
x)/(1 x)]+ , with n = 0, . . . , 3, are plus-distributions defined in the standard way.
We also worked out the normalized photon energy spectrum 1/G77 (0, )dG77 (z, )/dz, and
after setting = mb in our result, we find complete agreement with a recent paper by Melnikov
and Mitov [10].
H (2,NH) =
2.2. Comparison of NLO and NNLO results

The top frame of Fig. 1 shows the dependence of the spectrum dG(z, mb )/dz on the rescaled
photon energy z at NLO (dashed curve) and NNLO (solid curve). Also plotted is the NLO contribution supplemented by those NNLO terms which are proportional to s2 0 (dotted curve).
These terms, already worked out in [19], are often called BrodskyLepageMackenzie (BLM)
terms [20]. They arise from the contributions s2 NL , after replacing NL 30 /2 according
to the procedure of naive non-abelianization [21]. As seen in the figure, the BLM terms provide
the dominant part of the O(s2 ) corrections. As pointed out in [8], one should stress that this statement refers to the scheme where the decay width is written as in (2.1), i.e. with the combination
m
2b ()m3b explicitly appearing.
The behavior of the non-BLM corrections was discussed in detail in [10] and since we agree
with their findings, we do not repeat this discussion. However, when comparing the top frame of
our Fig. 1 with Fig. 1 of [10], the reader should bear in mind that the spectrum shown in the latter
is normalized to the fully integrated quantity G77 (0, mb ), followed by a consistent expansion in
powers of s . As a result the NLO and BLM curves are equal in both figures, but the NNLO
approximation lies below the BLM curve in the top frame of our Fig. 1, while it lies above the
BLM approximation in the case of the normalized spectrum shown in Fig. 1 of [10].
1
In the bottom frame of Fig. 1 we display the quantity z0 [dG77 /dz] dz as a function of the
(rescaled) photon energy cutoff z0 . As can be seen, the difference between the NNLO and the
BLM result is larger than in the top plot. This is due to the fact that the endpoint contributions
at z = 1 enter this integrated quantity, for which the difference between the BLM approximation
and the full NNLO contributions is somewhat larger. But still the BLM terms provide a good
approximation also in this case.
218

Fig. 1. The spectrum dG77 (z, mb )/dz as a function of z (top) and 1z [dG77 /dz] dz as a function of z0 (bottom) at NLO
0
(red dashed curve) and NNLO (blue solid curve) for NL = 4, NH = 1 and s (mb ) = 0.22. The black dotted curve is the
BLM approximation of the NNLO result. (For interpretation of the references to colour in this figure legend, the reader
is referred to the web version of this article.)
2.3. Change of bottom mass scheme

The above results have been obtained for a pole b quark mass, that is known to be affected by
a leading infrared renormalon and generally leads to slowly converging perturbative expansions.
In order to consider a change of scheme for this parameter, it is convenient to use the normalized
photon energy spectrum, namely
f (z, )
dG77 (z,)
dz
G77 (0, )

s ()
s () 2
(1)
= (1 z) +
CF f (2) (z, ) + O s3 .
CF f (z) +
(2.17)
X
X
A redefinition of the b quark mass mb mX
b mb , where mb is the b mass in the scheme X,
X
X
X
leads to a rescaling of the variable z z (1 + mb /mb ). A meaningful comparison of different
mass schemes can be made at the level of moments of the spectrum or fraction of events. They
are physical observables and are expected to have small and computable non-perturbative corrections. Several calculations of these quantities exist [14,22,23], but they do not always include
219
all the information coming from the O(s2 ) calculation of the spectrum. In the following we drop
the dependence from all formulas, set = mb , and use s (mb ) = 0.22 in the numerics.
Let us consider for instance the fraction R(E0 ) of events with photon energy above a cut E0 .
Neglecting as usual contributions from operators other than O7 , the perturbative expansion for
this quantity,
z0
R(E0 ) = 1
dz f (z),
where z0 =
2E0
,
mb
(2.18)
depends on the mass scheme adopted. In particular, the integral above receives contributions that
start at O(s ). In the pole mass scheme with mb = 4.8 GeV and E0 = 1.8 GeV one obtains
R(1.8 GeV)|pole = 1 0.0145 0.0085 0.0019 + = 0.9751 + ,
(2.19)
where we have listed separately the tree-level component, the NLO term, the BLM result, and
the non-BLM terms at NNLO. As typical in b decays, the large value of the BLM term suggests
an optimal scale for s of O(12 GeV). The inherently large uncertainty in the determination of
the pole mass implies a significant uncertainty in R(E0 ). For instance, using mb = 5.0 GeV we
obtain R(1.8 GeV)|pole = 0.9811.
Let us repeat the same calculation in a different scheme X for the b mass. Apart from the
change in the numerical value of the mass in the evaluation of z0 , there is an extra NNLO (nonBLM) term
R X (E0 ) =

mX
s
CF z0X Xb f (1) z0X .
mb
(2.20)
To illustrate the numerical difference due to a change of scheme in R(E0 ), we consider here
the kinetic mass of the bottom quark [24], also employed in [14]. At O(s2 ) the kinetic mass is
related to the pole mass by [25]
mkin
b (kin ) mb

2kin
s (mb )
4
=
CF
kin +
3
2mkin
b (kin )

2

2
13
4
s
4 1 2kin
CF 0
ln
kin
CA
3 2
mb
6
12
3

2

2
kin
13
13 1 2kin
+ 0
ln
CA
.
12 2
mb
6
12
2mb
(2.21)
Here kin is a Wilsonian cutoff that factorizes soft and hard gluons; the standard choice is
kin = 1 GeV. Inserting (2.21) in (2.20) and employing mkin
b (1 GeV) = 4.6 GeV, as suggested
by recent fits to the semileptonic moments [13], we obtain
R(1.8 GeV)|kin = 1 0.0196 0.0118 0.0025 + 0.0042 +
= 0.9703 +
(2.22)
where the last contribution has a positive sign and comes from (2.20). The result in the kinetic
scheme is compatible with the one in the pole scheme (2.19), that is expected however to have
larger higher order corrections. Eq. (2.22) can be compared with the simple estimate given in
220
[10]: the non-BLM correction has the same magnitude as quoted in [10], but differs for the sign
once the additional shift (2.20) is taken into account. As suggested in [10], one can write the perturbative expansion in terms of s at a scale lower than mb , improving the apparent convergence
of the series, and increasing (by up to a factor 2.5) the relative importance of the non-BLM
term. A consistent implementation of the kinetic scheme has been presented, up to O(s2 0 ),
in [14]: besides the use of the kinetic bottom mass, it involves a proper definition of the nonperturbative matrix elements of higher-dimensional operators and a cut on soft-gluon radiation
that modifies the shape of the spectrum close to the end-point. A complete implementation of
these effects is beyond the scope of the present paper, but we expect (2.22) to show the likely
impact of the O(s2 ) photon spectrum on the kinetic scheme calculation of R(E0 ).
Finally, we consider the truncated first moment of the normalized spectrum in the pole and
kinetic mass schemes, neglecting all non-perturbative corrections and the contributions from
operators other than O7 . It can be written in the on-shell scheme as

1
z0
3
mb
1 dz (1 z)f (z) 1 + dy f (y) + O s .
E E >E0 =
(2.23)
2
z0
Using the same numerical inputs as above, the pole scheme expansion is
mb
[1 0.0246 0.0242 + 0.0037 0.0004] = 2.291 GeV,
E E >1.8 GeV
(2.24)
2
where we have listed separately the tree-level, NLO, BLM, NNLO non-BLM, and the contribution of the interference (last) term in (2.23). The relative weight of the BLM contribution is even
larger than in the previous example, as expected since the first moment probes a region of smaller
gluon energies.
A change of bottom mass scheme in the first moment leads to the extra term

1
(1) X
mX
s X
b
X
X
(1)
z0 + dz (1 z)f (z) ,
1 CF
z 1 z0 f
E E >E0 =
2
0
z0X
(2.25)
mX
b
O(s2 )
O(s2 ).
now includes the

term, and the r.h.s. must be expanded up to
The
where
scheme change affects already the NLO calculation and at NNLO contributes also to the BLM
term. The numerical result in the kinetic scheme is
E E >1.8 GeV
mkin
b (1 GeV)
[1 0.0234 0.0234 + 0.0038 0.0005
2
+ 0.0293 + 0.0296 0.0035 0.0016] = 2.324 GeV,
(2.26)
where the last four entries are the NLO, BLM, NNLO non-BLM mass shifts, and the mixed
term in (2.25). Once again, the result is consistent with the one in the on-shell scheme, that is
however subject to a larger uncertainty; it is also consistent with the experimental value by the
Belle Collaboration [26]: E E >1.8 GeV = 2.292 0.026 0.034 GeV. We stress that our calculation in the kinetic scheme is not complete and that non-perturbative effects have not been
included (see [14]). We agree with [10] for the part included in that paper, which does not properly consider the change of scheme. Our result for the non-BLM NNLO contribution to (2.26)
is mkin
b (1 GeV)/2 (0.0038 0.0005 0.0035 0.0016) 4 MeV. Even if we rescale it to
account for a lower s scale, this has the opposite sign and is much smaller than the estimate in
221
[10]. It can be expected to shift the b quark mass extracted from the first moment by less than
15 MeV, well beyond present sensitivity. On the other hand, it is clear that a complete NNLO
analysis of the B Xs moments based on the spectrum presented here will have a smaller
perturbative uncertainty.
3. Details of the calculation
Our goal is to calculate the NNLO corrections to the quantity dG77 (z, )/dz introduced in
(2.4), and, as stated in the previous section, to do so we only have to calculate contributions with
three and four particles in the final state. Hence, we restrict the following discussion to the cases
n = 3 and 4 [see (2.5)].
We start by considering the general expression for the decay rate of a massive b quark with
momentum pb into n = 3, 4 massless final-state particles with momenta ki ,
n

n

d d1 ki
1
d (d)
1n =
ki |Mn |2
(2)
pb
2mb
(2)d1 2Ei
i=1
i=1

n1

1
d d ki 2 0
=
(2)n
d ki k i
2mb
2
i=1

2

n1
n1

pb
(3.1)
ki
ki0 |Mn |2 ,
pb0
i=1
i=1
where the squared Feynman amplitude |Mn is always understood as summed over final spin-,
polarization- and color states, and averaged over the spin directions and colors of the decaying b
quark. For n = 4 it also includes a possible factor of 1/2 if there are two identical particles in the
final state. Furthermore, d = 4 2 denotes the spacetime dimension that we use to regulate the
ultraviolet, infrared and collinear singularities. We also stress that we used the Feynman gauge
for the gluon propagator.
Since we are interested in the photon energy spectrum, we fix the photon energy by inserting
in the phase-space integrand on the r.h.s. of (3.1) a factor

pb p
= 2mb (pb + p )2 (1 + z)m2b .
E
(3.2)
mb
|2
Then, after performing the phase-space integrations over the full range, 1n coincides with
dG1n
77 (z, )/dz given in (2.4) up to a constant factor.
The optical theorem relates the b quark decay rate we are interested in to the imaginary parts
of three-loop b quark self-energy diagrams. The contribution of a specific physical cut of the
three-loop diagrams to the imaginary part of the b quark self-energy can be evaluated using the
Cutkosky rules [27], resulting in precisely the same integrals that appear on the r.h.s. of (3.1).
The various contributions to dG1n
77 (z, )/dz can therefore be easily represented in terms of cut
three-loop diagrams. Figs. 25 show the physical cuts that give non-vanishing contributions to
dG1n
77 (z, )/dz.
The individual contributions to the decay rate (or, equivalently, each cut-diagram) can now
be evaluated with the technical tools usually employed in the calculation of multiloop Feynman
diagrams. To this end, we rewrite all delta functions appearing in the phase-space representation
(3.1) as well as the one from the kinematic constraint in (3.2) as a difference of propagators
222
Fig. 2. Diagrams contributing to 1n for n = 3 and 4 at O(s2 ). We draw here only the non-Abelian diagrams. Thick
lines represent the massive b quark, thin lines the massless s quark, wavy lines photons and curly lines gluons. Dashed
lines separate the original Feynman diagrams which enter the squared amplitude |Mn |2 . The dashed blue lines indicate
cuts contributing to the b s gg process, and the dashed orange lines indicate cuts contributing to the b s g process.
Possible leftright reflected diagrams are not shown. See text for more details. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 3. Same as in Fig. 2 for diagrams with a closed light quark loop. Similar diagrams with a closed gluon loop are also
present. The dashed blue lines indicate cuts contributing to the b s q q (q {u, d, c, s}) and b s gg processes.
Concerning diagrams with closed ghost loops, see text. Note that three-particles cuts of the diagrams above are zero in
dimensional regularization. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
Fig. 4. Same as in Fig. 2 for diagrams including a closed bottom quark loop. The dashed orange lines indicate cuts contributing to the b s g process. Four-particles cuts of these diagrams are kinematically not allowed. (For interpretation
of the references to colour in this figure legend, the reader is referred to the web version of this article.)
[28,29],
q m
2

1
1
1
=
.
2i q 2 m2 i0 q 2 m2 + i0
(3.3)
The expressions for 1n can then be reduced algebraically by means of the systematic Laporta
algorithm [30], based on the integration-by-part identities (IBPs) first proposed in [31,32]. Basi-
223
Fig. 5. Same as in Fig. 2 for Abelian diagrams without a closed quark loop. The color code is the same as in Figs. 24.
cally, the Laporta algorithm allows one to express the various contributions to the integral in (3.1)
(i.e. the various cut diagrams in Figs. 25) as linear combinations of a small number of simpler
integrals, usually referred to as the master integrals (MIs) of the problem. The coefficients in
front of the MIs are ratios of polynomials of the kinematic variables present in the problem; in
the case under study, the only kinematic variable is z. The reduction procedure is simplified by
the fact that all the integrals in which one of the cut propagators disappears from the integrand, or
224
is raised to a negative power, are zero, because in this cases the i0 prescription becomes irrelevant. A further comment concerning the application of the IBPs to the calculation of phase-space
integrals is in order. It is not often stressed that the propagators originating from phase-space
integrations via (3.3) are always accompanied by a factor (q0 ), where q0 is the energy flowing
in the propagator. In building the IBPs it is necessary to take the derivative of the integrand of a
given diagram with respect to the integration momenta. When the derivative /q0 acts on (q0 ),
the latter transforms into a delta function and hence terms like

1
1
1
1
)
=
(q
(q0 ) 2
(3.4)
0
q m2 i0 q 2 m2 + i0
q
2 + m2 + i0 q
2 + m2 i0
are produced. Obviously, the i0 prescription is of no relevance here and the two terms in the
parenthesis cancel. Thus terms where the derivative acts on theta functions do not contribute to
the IBPs. Delta functions enforcing kinematic conditions, like the one in (3.2), are not multiplied
by theta functions. We made use of two automatic implementations of the Laporta algorithm:
AIR [33] based on Maple [34] and an independent Mathematica [35] code.
Once the reduction to MIs has been completed, the propagators introduced via (3.3) are
reconverted into delta functions and the MIs are evaluated. After introducing appropriate parameterizations for the three- and the four-particle phase-space [36,37] (and Feynman parameters
in order to perform the loop integration if the MI includes a closed loop without cut propagators),
the simpler integrals can be calculated analytically after expanding in . To obtain analytic expressions for the more difficult ones, it is convenient to employ the differential equation method
[38] as described in detail in [28,29]. As an independent check we also computed all the MIs
numerically by means of the sector decomposition method [39] (see [9] for further details).
The color factors for the individual diagrams shown in Figs. 25 are listed in Table I of [8].
The contribution of the diagrams which are not leftright symmetric must be multiplied by a
factor 2. Notice that we also included diagrams which arise by replacing the quark loop in Fig. 3
by a gluon loop. The four-particle cuts of these diagrams can be calculated in two ways: (a) using
the complete expression for the polarization sum of the two transverse gluons, or, (b) replacing
the polarization sums of the gluons by the negative of the metric tensor; in this case, the emission
of ghost particles (corresponding to four-particle cuts through diagrams with a closed ghost loop)
has to be taken into account. We convinced ourselves in [9] that the two procedures lead to the
same result. In the present paper we used procedure (b). As pointed out in [8], the contributions
of the diagrams with a closed gluon/ghost loop can be obtained from those involving a closed
light quark loop by replacing (in the Feynman gauge)

2
5
.
+ +O
TR NL CA
(3.5)
4 2
The diagrams in Figs. 25 involve IR, collinear and UV singularities, that appear in the intermediate steps of the calculation as poles in the dimensional regulator . Some of the cuts generate
not only single, but also double poles; the latter, however, cancel in the sum of all the cuts belonging to a given b quark self-energy diagram. The sum of all diagrams has a single pole that
gets removed when adding the counterterm contributions.
We conclude this section by summarizing the renormalization procedure adopted; it coincides
with the one employed in [8,9]. As explained after (2.6), our calculation can be restricted to
z = 1. In this case all counterterms of O(s2 ) (i.e. those contributing to H (2) defined in (2.10))
are induced by renormalizing the various parameters and wave functions in the process b s g.
To get the counterterms proportional to the color factors CA and TR NL , only the renormalization of the strong coupling constant is needed. For the term proportional to TR NH we need the
225
renormalization of the strong coupling constant and that of the gluon wave function. In order to
renormalize the contribution proportional to CF , one needs also to consider the renormalization
of the operator O7 , the renormalization of the bottom and strange quark wave functions, and of
the bottom quark mass. The renormalization constants, which are only needed to O(s ) precision
in our calculation, are fixed as follows:
(i) the strong coupling constant is renormalized using the MS scheme:
s ()

+ O s2 ,
ZMS = 1 + TR NF K(1) + CA K(2)
K(1) =
1
11
, K(2) =
;
3
12
(3.6)
(ii) the b quark mass contained in the operator O7 and the Wilson coefficient C7eff are renormalized in the MS scheme in our calculation. In the present application we only need the product
of these two renormalization constants which is given by
MS MS
Zm
Z77 = 1 + CF K77
b

s ()
+ O s2 ,
K77 =
1
;
4
(3.7)
(iii) all the remaining external fields and the b mass are renormalized in the on-shell scheme.
The on-shell renormalization constants for the b quark mass and wave function are given by

s ()
3
3
+ O s2 , Kmb = 1 L ,
4
2

s ()
OS
Z2b
= 1 + CF K2b
+ O s2 , K2b = Kmb .
OS
Zm
= 1 + CF Kmb
b
(3.8)
For the wave-function renormalization factor of the gluon-field we find

Z3OS = 1 + TR NH K3

s ()
+ O s2 ,
K3 =
1
2
L .
3 3
(3.9)
The s-quark field renormalization constant is equal to one. [Notice that in our approach the
effects of the diagrams in Fig. 4 are taken into account through the renormalization of the gluon
wave function.]
Consequently, we find (z = 1)
H (2,a) = H (2,a,bare) + (2K77 + K2b )H (1,) Kmb H (1,m,) ,
H (2,na) = H (2,na,bare) + K(2) H (1,) (z),
H (2,NL) = H (2,NL,bare) + K(1) H (1,) (z),

H (2,NH) = K(1) + K3 H (1,) ,
(3.10)
where the quantities H (2,i,bare) (i = a, na, NL) are the unrenormalized contributions to the photon energy spectrum corresponding to the diagrams given in Figs. 2, 3 and 5. The function
H (2,NH,bare) vanishes in our approach as just explained. The functions H (1,) and H (1,m,) include terms of order and their explicit expressions are collected in Appendix A.
226
4. Summary
A precise calculation of the photon energy spectrum in the inclusive B Xs decay is crucial
for comparing theoretical predictions with measurements. In this paper we calculated the NNLO
QCD corrections to the photon energy spectrum induced by the magnetic penguin operator O7 .
This subset of contributions is likely to provide the leading correction to the photon energy
spectrum at O(s2 ). The present calculation represents an independent check of the results of
[10], with which we find complete agreement.
The numerical relevance of the results and some application have already been studied in [10]
for the normalized photon energy spectrum: in particular, the effect of the non-BLM corrections
is minor with respect to the dominant BLM component [19]. This statement also holds for the
unnormalized photon energy spectrum dG77 /dz shown in the top frame of Fig. 1, as well as for
1
the integrated quantity z0 [dG77 /dz] dz shown in the bottom frame of this figure, provided the
decay width is written to be proportional to m
b ()2 m3b as in (2.1).
Furthermore, we investigated the impact of the change of the bottom quark mass from the pole
to the kinetic scheme on observables like the first moment of the spectrum E E >E0 and the
fraction of events with photon energy above E0 , R(E0 ). Apart from the change of the numerical
values of the b quark mass, these observables receive the additional shifts presented in (2.20) and
(2.25). We find that the results in the two schemes are compatible, although the pole scheme is
expected to have larger higher order corrections.
In the calculation we employed a number of techniques usually applied to the calculation of
multiloop Feynman diagrams, such as the Laporta algorithm, the differential equation method for
the analytic evaluation of master integrals, and sector decomposition as an independent numerical
check of the analytically evaluated master integrals. Several technical aspects of the calculation
were discussed in detail in Section 3. The same techniques can now be applied to the calculation
of the subleading O(s2 ) contributions (O7 , O8 ) and (O8 , O8 ) to the photon energy spectrum of
the inclusive radiative B decay.
Acknowledgements
H.M.A. is partially supported by the ANSEF No. 05-PS-hepth-0825-338 program, C.G.
and T.E. by the Swiss National Foundation as well as RTN, BBW-Contract No. 01.0357 and
EC-Contract HPRN-CT-2002-00311 (EURIDICE). P.G. is supported by MIUR under contract
2004021808-009. A.F. is grateful to E. Remiddi for a useful discussion on the use of IBPs in
problems involving cut propagators, and to J. Vermaseren for his kind assistance in the use of the
algebraic manipulation program FORM [40].
Appendix A. Functions H (1,) and H (1,m,)
Here we collect the functions H (1,) and H (1,m,) including O() terms as required by renormalization. For z = 1, that is with terms proportional to (1 z) dropped, they are given by

ln(1 z) 7 1
1+z
7 + z 2z2
H (1,) (z) =
ln(1 z) +
(1 + 4L )
1z
41z
2
4

2

7 ln(1 z) 3 ln2 (1 z)
7
1
2 Li2 (1 z) +
+
+
1z 3
2
4 1z
2 1z

7 ln z
7 + z 2z2
3
+ (1 + z) ln2 (1 z)
ln(1 z) + 2 ln z
21z 4
4

2

7 + z 2z2
Li2 (1 z) +
+ O 2 ,
+ (1 + z)
6
2
227
(1,m,)
(A.1)
ln(1 z)
6
(z) = 3
+
+ (1 + z) ln(1 z) (6 + 2z) (1 + 4L )
1z
1z

9 ln2 (1 z)
ln(1 z)
ln z
2
12
12
+
2
2 1z
1z
1z
(1 z)

2
8
3
3
2
+ (1 + z)
Li2 (1 z) +
(1 + z) ln2 (1 z)
1z
6
1z 2

+ (6 + z) ln(1 z) + (12 + 4z) ln z 6 2z + O 2 .
(A.2)
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]
See: Heavy Flavour Averaging Group, http://www.slac.stanford.edu/xorg/hfag/.

P. Gambino, M. Misiak, Nucl. Phys. B 611 (2001) 338, hep-ph/0104034.
A.J. Buras, M. Misiak, Acta Phys. Pol. B 33 (2002) 2597, hep-ph/0207131.
M. Misiak, M. Steinhauser, Nucl. Phys. B 683 (2004) 277, hep-ph/0401041.
M. Gorbahn, U. Haisch, Nucl. Phys. B 713 (2005) 291, hep-ph/0411071.
M. Gorbahn, U. Haisch, M. Misiak, Phys. Rev. Lett. 95 (2005) 102004, hep-ph/0504194.
K. Bieri, C. Greub, M. Steinhauser, Phys. Rev. D 67 (2003) 114019, hep-ph/0302051.
I. Blokland, A. Czarnecki, M. Misiak, M. Slusarczyk, F. Tkachov, Phys. Rev. D 72 (2005) 033014, hep-ph/0506055.
H.M. Asatrian, A. Hovhannisyan, V. Poghosyan, T. Ewerth, C. Greub, T. Hurth, Nucl. Phys. B 749 (2006) 325,
hep-ph/0605009.
K. Melnikov, A. Mitov, Phys. Lett. B 620 (2005) 69, hep-ph/0505097.
M. Czakon, U. Haisch, M. Misiak, in preparation;
U. Haisch, talk at the workshop Flavour in the LHC era, see http://flavlhc.web.cern.ch/flavlhc/.
H.M. Asatrian, C. Greub, A. Hovhannisyan, T. Hurth, V. Poghosyan, Phys. Lett. B 619 (2005) 322, hep-ph/0505068.
O. Buchmuller, H. Flacher, Phys. Rev. D 73 (2006) 073008, hep-ph/0507253;
C.W. Bauer, Z. Ligeti, M. Luke, A.V. Manohar, M. Trott, Phys. Rev. D 70 (2004) 094017, hep-ph/0408002;
M. Neubert, Phys. Rev. D 72 (2005) 074025, hep-ph/0506245.
D. Benson, I.I. Bigi, N. Uraltsev, Nucl. Phys. B 710 (2005) 371, hep-ph/0410080.
I.I.Y. Bigi, M.A. Shifman, N.G. Uraltsev, A.I. Vainshtein, Int. J. Mod. Phys. A 9 (1994) 2467, hep-ph/9312359;
M. Neubert, Eur. Phys. J. C 40 (2005) 165, hep-ph/0408179;
J.R. Andersen, E. Gardi, JHEP 0506 (2005) 030, hep-ph/0502159.
E. Gardi, JHEP 0502 (2005) 053, hep-ph/0501257.
T. Becher, M. Neubert, Phys. Lett. B 633 (2006) 739, hep-ph/0512208;
T. Becher, M. Neubert, Phys. Lett. B 637 (2006) 251, hep-ph/0603140.
A. Ali, C. Greub, Z. Phys. C 49 (1991) 431;
A. Ali, C. Greub, Phys. Lett. B 259 (1991) 182;
A. Ali, C. Greub, Phys. Lett. B 361 (1995) 146, hep-ph/9506374;
N. Pott, Phys. Rev. D 54 (1996) 938, hep-ph/9512252.
Z. Ligeti, M.E. Luke, A.V. Manohar, M.B. Wise, Phys. Rev. D 60 (1999) 034019, hep-ph/9903305.
S.J. Brodsky, G.P. Lepage, P.B. Mackenzie, Phys. Rev. D 28 (1983) 228.
M. Beneke, V.M. Braun, Phys. Lett. B 348 (1995) 513, hep-ph/9411229.
J.R. Andersen, E. Gardi, JHEP 0506 (2005) 030, hep-ph/0502159.
M. Neubert, Phys. Rev. D 72 (2005) 074025, hep-ph/0506245.
I.I.Y. Bigi, M.A. Shifman, N. Uraltsev, A.I. Vainshtein, Phys. Rev. D 56 (1997) 4017, hep-ph/9704245;
I.I.Y. Bigi, M.A. Shifman, N. Uraltsev, A.I. Vainshtein, Phys. Rev. D 52 (1995) 196, hep-ph/9405410.
228
[25] A. Czarnecki, K. Melnikov, N. Uraltsev, Phys. Rev. Lett. 80 (1998) 3189, hep-ph/9708372.
[26] Belle Collaboration, P. Koppenburg, et al., Phys. Rev. Lett. 93 (2004) 061803, hep-ex/0403004;
See also BaBar Collaboration, B. Aubert, et al., hep-ex/0607071.
[27] R.E. Cutkosky, J. Math. Phys. 1 (1960) 429;
M.J.G. Veltman, Physica 29 (1963) 186;
E. Remiddi, Helv. Phys. Acta 54 (1982) 364.
[28] C. Anastasiou, K. Melnikov, Nucl. Phys. B 646 (2002) 220, hep-ph/0207004.
[29] C. Anastasiou, L.J. Dixon, K. Melnikov, F. Petriello, Phys. Rev. D 69 (2004) 094008, hep-ph/0312266.
[30] S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033.
[31] F.V. Tkachov, Phys. Lett. B 100 (1981) 65.
[32] K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
[33] C. Anastasiou, A. Lazopoulos, JHEP 0407 (2004) 046, hep-ph/0404258.
[34] Maple 9.5, Copyright 2004 by Maplesoft.
[35] Mathematica 5.0, Copyright 2003 by Wolfram Research.
[36] A. Gehrmann-De Ridder, T. Gehrmann, G. Heinrich, Nucl. Phys. B 682 (2004) 265, hep-ph/0311276.
[37] C. Anastasiou, K. Melnikov, F. Petriello, Phys. Rev. D 69 (2004) 076010, hep-ph/0311311.
[38] E. Remiddi, Nuovo Cimento A 110 (1997) 1435.
[39] T. Binoth, G. Heinrich, Nucl. Phys. B 680 (2004) 375, hep-ph/0305234.
[40] J.A.M. Vermaseren, math-ph/0010025.
Supersymmetry, attractors and cosmic censorship

Jorge Bellorn, Patrick Meessen, Toms Ortn
Instituto de Fsica Terica UAM/CSIC, Facultad de Ciencias C-XVI, C.U. Cantoblanco, E-28049 Madrid, Spain
Received 29 June 2006; received in revised form 22 October 2006; accepted 6 November 2006
Abstract
We show that requiring unbroken supersymmetry everywhere in black-hole-type solutions of N = 2,
d = 4 supergravity coupled to vector supermultiplets ensures in most cases absence of naked singularities. We formulate three specific conditions which we argue are equivalent to the requirement of global
supersymmetry. These three conditions can be related to the absence of sources for NUT charge, angular
momentum, scalar hair and negative energy, although the solutions can still have globally defined angular
momentum and non-trivial scalar fields, as we show in an explicit example. Furthermore, only the solutions
satisfying these requirements seem to have a microscopic interpretation in string theory since only they
have supersymmetric sources. These conditions exclude, for instance, singular solutions such as the Kerr
Newman with M = |q|, which fails to be everywhere supersymmetric. We also present a re-derivation of
several results concerning attractors in N = 2, d = 4 theories based on the explicit knowledge of the most
general solutions in the timelike class.
1. Introduction
In spite of the impressive progress made during the last few years in the study of supersymmetric black-hole solutions, there are important questions that remain unanswered or whose answer
is unclear. For instance, we know how to construct many supersymmetric black-hole-type solutions, but many of them are singular. Some of these become regular when string corrections are
taken into account and for all the regular black hole solutions we seem to have a string theory
model that accounts for its entropy. How are the other singular solutions to be understood? How
E-mail addresses: jorge.bellorin@uam.es (J. Bellorn), patrick.meessen@cern.ch (P. Meessen),

tomas.ortin@cern.ch (T. Ortn).
doi:10.1016/j.nuclphysb.2006.11.004
230
J. Bellorn et al. / Nuclear Physics B 762 (2007) 229255
can it be that they are supersymmetric and yet there is no string theory model for them? Or, if
there is, why are they singular?
The main goal of this paper is to try to answer this question by giving a set of conditions that
supersymmetric black-hole-type solutions must satisfy in order to be admissible in the context of
N = 2, d = 4 supergravity coupled to vector supermultiplets. Admissible solutions will be regular and will describe one or several black holes in static equilibrium, even though the system may
have a finite global angular momentum, as is for example the case in the solution constructed in
Ref. [1]. Furthermore, we expect only admissible solutions to have a microscopic string theory
model. We will argue that the non-admissible solutions are, in general, not truly supersymmetric in the sense that will be explained later on and the conditions of admissibility can be seen
as conditions for a solution to be everywhere supersymmetric. For instance: the KerrNewman
solution with equal charge and mass, which is singular but nevertheless commonly believed to
be supersymmetric, is non-admissible according to our criteria. We will show that it fails to be
supersymmetric at the singularity, where the sources might be located. Equivalently we can say
that the KerrNewman field with M = |q| is caused by non-supersymmetric sources. This explains why it is not described by any supersymmetric string theory model. We will also show
that, generically, rotating sources are not allowed by supersymmetry and that regular, supersymmetric solutions with angular momentum are always composite objects made out of several static
black holes in equilibrium. The angular momentum has its origin in the dipole momenta of the
electromagnetic fields corresponding to the distribution of charged black holes. Something similar happens for scalar fields: supersymmetric configurations satisfying our conditions can have
non-trivial scalar fields but cannot have sources.
In order to prove these results, we will make use of the explicit knowledge of the most general
solutions of N = 2, d = 4 supergravity coupled to vector multiplets, which have recently been
classified in Ref. [2].1 All the asymptotically flat supersymmetric black hole solutions seem to
belong to the timelike class, and, although they coincide with the solutions found in Ref. [4], the
general formalism will allow us to make further progress in their understanding. In particular,
we will use the Killing spinor identities (KSIs) [5,6], which can be understood as integrability
conditions for the Killing spinor equations, in order to study supersymmetry at the singular points
where the sources of these solutions should be located.
The final ingredient will be the attractor equations of N = 2, d = 4 supergravity [7,8,10,11]:
these provide us with information about the sources thought of as being placed at the attractor
points. In fact, we will find interesting relations between KSIs and attractor equations, the former
showing explicitly that
(1) supersymmetry always requires the absence of the kind of scalar hair called primary in
Ref. [12], and that
(2) when the attractor equations are satisfied there are no sources whatsoever for scalar hair.2
These results can be viewed as an extension of those of Ref. [14] in which it was observed
that supersymmetry seems to act as a cosmic censor for static black-hole-type configurations but
not for the stationary ones, such as the KerrNewman M = |q| solution.
1 In this paper we will not consider the coupling to hypermultiplets. The classification of the supersymmetric solutions
with both vector multiplets and hypermultiplets is considered in Ref. [3].

2 If there is more than one basin of attraction, contrary to what is assumed in this article, this last conclusion might
change due to the area codes [13].
231
This paper is organized as follows: in Section 2 we review the timelike class of supersymmetric solutions of N = 2, d = 4 supergravity coupled to vector multiplets. First, we review how all
the solutions in this class can be constructed from a symplectic vector of real harmonic functions
and then in Section 2.1 we derive the KSIs that, by assumption of supersymmetry, all solutions
must satisfy. In Section 2.2 a re-derivation of some of the main results involving N = 2, d = 4
supersymmetric black-hole attractors, taking advantage of the actual and explicit knowledge of
all the solutions of this kind, which has helped us to improve some of the presentations existing
in the literature and prove some new results.
In Section 3 we study how the KSIs constrain the possible sources and singularities of blackhole-type solutions and the interplay with the attractor equations in a general way. The main
result of this section will be the formulation of three conditions that express the existence of
supersymmetry everywhere in the solutions, including, particularly, the locations of the sources.
These conditions should ensure the regularity of the admissible solutions and we study in very
close detail several examples in Section 4. Section 5 contains our conclusions.
2. Timelike BPS solutions of N = 2, d = 4 SUEGRA
It was recently shown in Ref. [2] that all the supersymmetric solutions in the timelike class
of N = 2, d = 4 supergravity coupled to n vector multiplets3 can be constructed by setting the
2n = 2(n + 1) components of a real, symplectic vector I = (I , I ) equal to 2n = 2(n + 1) real
functions harmonic on 3-dimensional Euclidean space4

I
I
(2.1)
,
m m I = m m I = 0, = 0, 1, . . . , n.
I
This real section I enters the theory as the imaginary part of the section V/X, where V is the
covariantly-holomorphic canonical section defining special geometry:

V | V L M L M = i,
L
1
V=
(2.2)
M
Di V = (i 2 i K)V = 0,
Di V | V = 0.
X on the other hand is proportional to the complex, scalar bilinear constructed out of the
Killing spinors: supersymmetry and consistency of the solutions imply that it can be expressed
in terms of I, see e.g. Ref. [2] or Eq. (2.7).
Eqs. (2.1) are sometimes known as the generalized stabilization equations, the standard stabilization equations having the same form but with the harmonic functions (I , I ) replaced by
magnetic and electric charges, e.g. (p , q ).
The real part of V/X, denoted by R (R , R ) can, in principle, be written in terms of the
real harmonic functions, which is usually referred to as solving the stabilization equations. In
theories with a prepotential, the homogeneity properties of the prepotential allow us to write
M /X =
F (L /X)
.
(L /X)
(2.3)
3 These solutions were first found in slightly different form in Ref. [4] and the procedure followed in Ref. [2] shows
that they are the only solutions in this class.

4 If the functions are not harmonic, the field configurations are still supersymmetric, but are not solutions of the equations of motion.
232
Taking the imaginary part of this equation, we have

I (R , I ) = I ,
(2.4)
which implicitly defines R (I , I ), although solving these equations can be extremely hard and
in general the explicit solution is unknown.
The real part of Eqs. (2.3) and the above solutions give straightforwardly the functions
R (R (I , I ), I ).
Having the complete symplectic section V/X entirely given in terms of the real harmonic
functions, one can construct the fields of the solutions as follows:
(1) The n complex scalar fields Z i are given by the quotients
Zi =
Ri + iI i
Li /X
= 0
.
0
L /X R + iI 0
(2.5)
(2) The metric has the form

ds 2 = 2|X|2 (dt + )2
1
dx i dx i ,
2|X|2
i, j = 1, 2, 3,
(2.6)
where
1
= R|I,
2|X|2
(2.7)
and is a time-independent 1-form on Euclidean 3-dimensional space satisfying the equation

(d)mn = 2mnp I|p I.
(2.8)
(3) The symplectic vector of field strengths and their duals F = (F , F ) is given by
F =

1
d[RV ] [dI V ] ,
2
V = 2 2|X|2 (dt + ).
(2.9)
The Killing spinors of these solutions have the form

I = X 1/2 I 0 ,
I 0 = 0,
I 0 + i0 I J J 0 = 0,
(2.10)
which implies
I + i0 ei I J J = 0,
1/2

ei = X/X
.
(2.11)
Observe that we can write

X=
for any .
L (Z, Z )
,
R + iI
(2.12)
233
2.1. Killing spinor identities

All supersymmetric configurations satisfy the Killing spinor identities relating the Einstein
equations E , the Maxwell equations E , the Bianchi identities B and the scalar equations
of motion E i [2,5,6]

Ea a I 4i E V I J J = 0,
(2.13)
i I J
i I

E + 2i E
(2.14)
/ U J = 0,
where E is the symplectic vector (B , E ).
In the timelike case, they lead to the following identities in an orthonormal frame
E ab = a 0 b 0 E 00 ,
1

V/X E a = |X|1 E 00 a 0 ,
4
a 1 i

Ui E = e Ei a 0 .
2
These equations imply directly
E 0m = 0,
E mn = 0,
(2.15)
(2.16)
(2.17)
m
V E = 0,

Ui E m = 0.
Further, the r.h.s. of Eq. (2.15) is real, and this leads to two important identities:
0
I E = 0,

E 00 = 4 V E 0 .
(2.18)
(2.19)
(2.20)
2.2. Attractor equations

It is well known that, in general, the scalar fields of the black-hole solutions of these theories
have certain attractor values that depend solely on the electric and magnetic charges and which
are attained at the event horizons irrespectively of the chosen asymptotic values [7,8].5 The
attractor values are those which extremize a specific function; furthermore, the absolute value
squared of the central charge for the attractor values is essentially the horizon area [10,11]. Here
we are going to rederive these results using our notation and to relate them to the KSIs. We
also want to improve the previous derivations by making explicit use of the knowledge of all the
supersymmetric configurations.
Let us consider single, static, asymptotically flat, spherically symmetric, black-hole-type solutions of N = 2, d = 4 supergravity coupled to vector multiplets: they are given by real harmonic
functions of the form
q
I = I + ,
(2.21)
r
which is the general choice compatible with the assumptions. The metric can be conveniently
written in spherical coordinates as

1 2
2
dr + r 2 d(2)
.
ds 2 = 2|X|2 dt 2
(2.22)
2
2|X|
5 If there are multiple attractor regions, it might happen that there is some residual dependency on the asymptotic
values. Here we assume there to be only one attractor region.
234
This metric describes black holes if

1 r
2M
1 +
,
grr =
r
2|X|2
(2.23)
is always finite for finite r, whence M, which is the mass, must be positive. Further, we have to
require
1 r0 A
> 0,
2|X|2
4r 2
(2.24)
which imposes the existence of an event horizon with area A > 0 at r = 0 instead of a naked
singularity.
The existence of attractors (fixed points) of the scalar fields follows from the fact that in
supersymmetric configurations, the scalars satisfy first-order differential equations, as follows
immediately from the Killing spinor equations associated to the gaugino supersymmetry transformation rule:
Zi I + I JG
/ i+ J = 0.
I i = i/
(2.25)
To derive the needed first-order equations, we first use the time-independence of the solutions
i m m Z i I 4 I J Gi+ 0m m 0 J = 0,
and then the known constraint Eq. (2.11) as to obtain

m Z i 4ei Gi+ 0m m I = 0 m Z i = 4ei Gi+ 0m .
(2.26)
(2.27)
Going over to curved indices, the equation takes the form
dZ i
= 2 2Gi+ tr /X .
dr
(2.28)
The self-duality of Gi+ allows us to express the Gi+ tr component in terms of the Gi+ :

2|X|2 i+
G ,
Gi+ tr = i Gi+ = i 2
(2.29)
r sin
which combined with
i

i

Gi+ = T i F + = G ij Dj V F = G ij Dj V F ,
(2.30)
2
2
leads to

X
dZ i
G ij Dj V F .
=2 2 2
(2.31)
dr
r sin
Since the form of all the fields in terms of I(r) is in principle known, we can try to find a
more explicit form for this equation: using the general form of the vector fields Eq. (2.9) and of
I(r), Eq. (2.21), we find
1
dI
q
= sin .
F = r 2 sin
dr
2
2
After substituting this into Eq. (2.31), one ends up with
dZ i
= 2XG ij Dj Z ,
d
(2.32)
(2.33)
235
where 1/r and where

Z(Z, q) V|q,
(2.34)
is the central charge of the theory [15]. Observe that the presence of the factor X in the r.h.s.
is crucial for it to have zero global Khler weight, just as the l.h.s. Further observe that the
r-dependence is only through the scalars Z i (r)!
The r.h.s. of this system of differential equations depends only on the scalar fields Z i , and,
i at
thus, it is an autonomous system of ordinary differential equations6 that has fixed points Zfix
the values at which the r.h.s. vanishes
Di Z|Z i =Z i = 0.
(2.35)
fix
i as
If the solution of this system of equations exists, it gives the fixed values of the scalars Zfix
functions of the electric and magnetic charges only
i
i
Zfix
= Zfix
(q),
(2.36)
i do not occur in the above differential equation. The

since the asymptotic values (moduli) Z
fixed values are reached by the scalars at the value = , i.e. r = 0, which is where the event
horizon would be, as discussed at the beginning of this section and in what follows.
The fixed values may or may not be admissible, i.e. they may or may not belong to the defi are admissible and
inition domain of the complex coordinates Z i . If the asymptotic values Z
i
the fixed values Zfix (q) are not, there must be a singularity between r = and r = 0, which
will induce a curvature singularity. We will require both the asymptotic and the fixed values to
be admissible. These aspects will be discussed in Section 3.
i = Z i have constant scalar
Black-hole solutions whose scalars take the asymptotic values Z
fix
fields, and are called doubly extreme black holes. These values are the ones that extremize, not
the central charge, but the zero-Khler-weight combination eK/2 Z:

Di Z|Z i =Z i = eK/2 i eK/2 Z Z i =Z i = 0.
(2.37)
fix
fix
2.2.1. Consequences of the existence of attractors

There are no more scalar fields in the theory, but in the timelike supersymmetric solutions
there is another scalar object7 that satisfies a first-order differential equation: X. From the Killing
spinor equation associated to the gravitino supersymmetry transformation rule it is possible to
derive [2]
D X = iT + V ,
(2.38)
where V is the timelike Killing vector constructed from the Killing spinor. The graviphoton
field strength can be written in the form
T + = V|F ,
(2.39)
6 The use of the variable = 1/r is essential in this argument. It is easy to see that the derivatives of the scalar fields
of typical black-hole solutions w.r.t. to r do not vanish at r = 0, while their derivatives w.r.t. do.
7 In previous derivations in the literature the absolute value |X| = eU is considered, but then the Khler weights and
the reality properties of the two sides of the equations derived are different.
236
and, together with

V F = 2 |X|2 R ,
the equation for X becomes

D X = 2i V |X|2 R .
(2.40)
(2.41)
Dividing both sides by X and expanding the r.h.s. using V/X = R + iI we get
D X
= 2i|X|2 R| R 2 |X|2 I|R 2|X|2 I| R.
X
Now, from Eq. (2.7)
2I| R = |X|2 2R| I,
(2.42)
(2.43)
and we get
D X
= 2i|X|2 R| R 2|X|2 R| I.
X
Finally, using
R| R = I| I,
which is proved in Appendix A, we arrive at8

D X 1 = 2 V I .
(2.44)
(2.45)
(2.47)
This equation is valid for all supersymmetric configurations in the timelike class. For those considered in this section we arrive at the equation we were looking for:
D X 1 = 2Z .
(2.48)
The real and imaginary parts of this equation are

d(grr )
= 2 e Z /X = 2R|q,
d

d
+ Q = |X|2 2 m Z /X = 2I|q = 2I |q.
d
(2.49)
(2.50)
For the spherically symmetric solutions under consideration vanishes and this requires the
phase of X to be covariantly constant, i.e.
I|q = I |q = 0.
(2.51)
We will later show that this is equivalent to the requirement that the NUT charge vanishes. Since
there is only dependence on , the phase of X can simply be gauged away by means of a Khler
transformations. The phase of Z is then also constant, whence Z/X is real, which can be used
8 Observe that the compatibility between Eq. (2.7) and the following equations requires the identity
R|I = R| I,

to hold. For theories admitting a prepotential, this is done in Appendix A.
(2.46)
237
to write
d|X|1
(2.52)
= 2|Z|.
d
The sign is the sign of R|q and we can argue that it has to be positive if the mass is going to
be positive: if we take Eq. (2.49) at = 0 (r = ), we find that the mass of the solution is given
by the linear combination of charges and moduli
M = R |q.
(2.53)
Observe that there is no a priori guarantee that M > 0: this is a condition that has to be imposed
independently as to avoid singularities. We will do so and will only consider the positive sign
above; Eq. (2.52) is then the expression found in the literature.
If we take another derivative of Eq. (2.49) and use Eq. (2.52), we find

i
d 2 (grr )
d|X 1 |
1 d|Z|
2
1 dZ
=2
|Z| + 2|X|
= 4|Z| + 2|X|
i |Z| + c.c. . (2.54)
d
d
d
d 2
Now, at = fix = 0 we have Z i = Zfix and dZ i /d = 0, and the above equation takes on the
form
A
(2.55)
= 4|Zfix |2 .
2
Again, there is no a priori guarantee that |Zfix | = 0, which therefore is another condition that has
to be imposed independently as to avoid singularities. Actually, even though in this expression
A is basically an absolute value, the positivity of A is only guaranteed if the scalar fields take
admissible values, the mass is positive etc.
These identities allow us to find two interesting expressions for |Zfix |. Expanding the two
sides of Eq. (2.49) as a power series in we find

A
dR
(2.56)
=2
q .
2
d =0
Using the expressions in Appendix A we get [9,10]

1 dR
1
|Zfix |2 =
q = q T M(Ffix )q,

2 d =0
2
where
M(F )
m F + e F m F 1 e F
e F m F 1
(2.57)
m F 1 e F
m F 1
A direct computation of |Zfix |2 gives

2

q .
|Zfix |2 = Vfix |q = q|Vfix Vfix
The matrix of this bilinear is

M M

|Vfix Vfix =
L M
M L
L L

.
(2.58)
(2.59)

.
(2.60)
fix
We can use the relation

1
L L = m(N )1| f i G ii f i ,
2
(2.61)
238
taking into account that at the fixed point the second term in the r.h.s. will not contribute, and
that only its symmetric part will contribute, to get [9,10]
1
|Zfix |2 = q T M(Nfix )q.
2
(2.62)
So far we have checked that the coefficient of the 2 term of grr is given by the value of
the central charge at the fixed point but, if there are terms of higher order in in grr there will
not be a regular horizon. We can, however, see that taking another derivative of grr w.r.t. at
= 0 will give zero if the attractor equations (2.35) are satisfied and the same will happen for
higher derivatives.
Summarizing we can say that the attractor equations (plus the positivity of the mass, which
is not guaranteed) seem to be sufficient conditions to have regular, static, spherically symmetric
black holes.
Finally, observe that Eq. (2.53) plus the identification, which will be established later on,
between the NUT charge and the linear expression of the charges
N = I |q,
lead to a complex BPS relation

M + iN = (V/X) q .
(2.63)
(2.64)
We will argue that supersymmetry requires N to vanish, whence the above relation reads
M = 2 |Z |,
(2.65)
which is the standard BPS relation between mass and central charge. Of course, only the positive
sign will be admissible.
3. Relations between the N = 2, d = 4 KSIs, attractors and sources
The equations of motion9 for supersymmetric configurations of supergravity theories satisfy
certain relations known as Killing spinor identities (KSIs), which can also be derived from the
integrability conditions of the Killing spinor equations [5,6]. We have unbroken supersymmetry
wherever the Killing spinors exist, and these exist, locally, wherever the KSIs are satisfied. Thus,
if we are to have unbroken supersymmetry everywhere we must demand the KSIs to be satisfied
everywhere. In this section we are going to study the consequences of demanding the black-hole
solutions of N = 2, d = 4 supergravity to be everywhere supersymmetric.
The KSIs of N = 2, d = 4 supergravity are given in Eqs. (2.13) and (2.14) and they lead
to Eqs. (2.15)(2.20) for configurations in the timelike class. Since we are going to consider
configurations that solve the equations of motion, it may seem that the KSIs are automatically
satisfied. However, most solutions have singularities at which the equations of motion are not
satisfied, i.e. one has E() = J (). The r.h.s. of the equations of motion at the singularities
can be associated to sources for the corresponding fields and the KSIs are then understood as
relations between the possible sources of supersymmetric solutions: the KSIs put constraints on
possible sources of supersymmetric solutions.
9 By equation of motion E() of a given field we will mean here the l.h.s. of the equation of motion S/ = E() =
0. This slight abuse of language should lead to no confusions.
239
Let us consider from this point of view the KSIs Eqs. (2.15)(2.20): the first of them,
Eq. (2.15), tells us that the components E 0m and E mn of the Einstein equations must vanish
automatically for supersymmetric configurations and they must do so everywhere if the solutions are everywhere supersymmetric. This means that the sources J 0m and J mn of the Einstein
equation must vanish identically everywhere
J 0m = J mn = 0.
(3.1)
Hence, singular (delta-like) sources are not allowed, and in particular this means that no localized
sources of angular momentum are allowed.
Any singular contributions to J 0m and J mn must originate in the R 0m components of the
Ricci tensor; more precisely, they come from the term m (d)mn , where is the 1-form that
appears off-diagonally in the metric of the timelike supersymmetric solutions of N = 2, d =
4 supergravity Eq. (2.6). Therefore, using Eq. (2.8) and defining the complex 3-dimensional

vector W

1
= (Wm ) V/X|m I ,
W
(3.2)
m(Wm ) = mnp (d)np = I|m I,
4
we can translate the above KSIs, Eqs. (3.1), to the condition
= 0,
W)
m(
(3.3)
which has to be imposed everywhere. Actually, only the singular parts of this equation have
to be taken into account since, dealing with solutions, the finite parts must be canceled in the
equations of motion by other finite contributions. Therefore, from now on we will ignore all
finite contributions to this equation.
Let us consider the real and imaginary parts of Eq. (2.16), namely Eq. (2.20) and (2.19). The
real part gives us two important pieces of information: first, it tells us that the component J 00
of the source of the Einstein equation is related to component J 0 of the source of the combined
Maxwell and Bianchi equations E a. If the electromagnetic fields have only one static point-like
source at r = 0, E t 1 q (3) (
x )/ |g|, then using the fact that Z/X is real (see Eq. (2.51) and
2
the previous discussion)

E 0t = 2 2 |Z||r=0 (3) (
(3.4)
x )/ |g|,
which shows that, if the attractor equations are satisfied, the source for the Einstein equations is
just |Zfix (q)|. The sign is related to the positivity of R|q, which is, as was discussed before,
associated to the positivity of the mass etc. This is the only value admissible by supersymmetry,
since we can understand this source as a source of energy. However, if the scalars take nonadmissible values we will find the wrong sign or a zero at r = 0 and supersymmetry will be
broken at the source: we will have to require that the attractor equations are solved by admissible
values of the scalars.
The second piece of information we can obtain from the real part concerns the spacelike
components of the electromagnetic sources. Combined with the spacelike components of the
imaginary part, Eq. (2.19), we get the condition

V/X J m = 0.
(3.5)
Let us now consider the time component of the imaginary part of the KSI Eq. (2.16),
Eq. (2.19):
t
I J = 0.
(3.6)
240
To find the physical meaning of this condition we use the explicit form of the symplectic
vector of vector field strengths F for timelike BPS solutions Eq. (2.9):

t m m I
.
J = E = dF = |X|2 (m m I)V =
(3.7)
|g|
2
This result tells us that the KSIs Eq. (3.5) are always satisfied and that the KSI Eq. (3.6) is
equivalent to the condition
I|m m I = m(m Wm ) = 0,
(3.8)
which is nothing but the integrability condition for the equation determining , which now has
to be satisfied everywhere as a consequence of demanding unbroken supersymmetry everywhere.
For the point-like sources considered above, these equations take the form

I|qA (3) x xA
|g| = 0.
(3.9)
A
The consequences of imposing this condition were first studied by Denef and Bates in
Refs. [16,17] in the context of general N = 2, d = 4 supergravity, but was studied earlier by
Hartle and Hawking in Ref. [18] in the context of IsraelWilsonPerjs (IWP) solutions of the
EinsteinMaxwell theory. As shown by Tod in Ref. [19] these are precisely the timelike solutions
of pure N = 2, d = 4 supergravity and a special case of the general problem that we are going to
study. Hartle and Hawking were motivated, not by supersymmetry, but rather by the prospect of
finding regular solutions describing more than one black hole. They were, in particular, worried
about possible string singularities related to NUT charges. These singularities can be eliminated
by compactifying the time coordinate with certain period [20], but at the price of losing asymptotic flatness. Let us consider a possible string singularity parametrized by z and choose polar
coordinates , around it. If one considers the integral of the 1-form that appears in the metric
along a loop of radius R enclosing the possible string singularity at two different points z1 and
z2 , denoted by I (R, z1,2 ), one can use Stokes theorem to derive

I (R, z1 ) I (R, z2 ) = d = 2 3 m W,
(3.10)
2
where
is a surfaces whose boundaries are the loops of radius R at z1,2 . In the zero radius
limit 2 is a closed surface that crosses the possible string singularity at z1 and z2 and we
have

2 lim R (R, z1 ) (R, z2 ) = 2 3 m W = d3 m W
R0
=2
d 3 x m(m Wm ),
(3.11)
where 3 = 2 . Thus, m(m Wm ) = 0 implies that is singular on the string somewhere between z1 and z2 . These singularities are related to the presence of NUT sources,
since we can define the NUT charge contained in 3 as the integral of d over 2 =
3 :

2
8N = d = d = 2 d 3 x m(m Wm ).
(3.12)
2
241
Thus, the condition m(m Wm ) = 0, required by supersymmetry, is equivalent to the absence

of sources of NUT charge.
Hartle and Hawking argued that the only solutions in the IWP class with no NUT charge
(and no singularities) were the MajumdarPapapetrou solutions [21,22] which are regular and
static. We will review their arguments in Section 4.1.4 and show that there are indeed non-trivial
solutions that satisfy the KSIs and have no NUT charges, apart from the MajumdarPapapetrou
ones; they all have negative total mass, which causes other naked singularities to appear.
Thus, if we include positivity of all masses among the requirements necessary to have supersymmetry, the only supersymmetric black-holes-type solutions of pure N = 2, d = 4 supergravity will indeed be the MajumdarPapapetrou solutions. We will have to consider more general
N = 2, d = 4 theories in order to be able to have stationary solutions such as the one found in
Ref. [1], that satisfy the KSIs and have positive mass. This will be done in Section 4.2.
Next, let us consider the KSI Eq. (2.17) which relates the sources of the scalar fields with
those of the vector fields. If we consider only point-like sources and call A the scalar charge at
xA , this equation implies, at each sources
A = 2ei Di Z|xA .
(3.13)
As mentioned before, the scalar sources are completely determined by the electric and magnetic
charges and the asymptotic values of the scalar fields. This is known as secondary scalar hair
[12]. Primary scalar hair correspond to completely free parameters as in the Einstein-scalar solutions of Ref. [23] or in the solutions of Ref. [24] which may be embedded in N = 4, d = 4
supergravity. Neither of these solutions is supersymmetric (nor regular) and the above KSI explains just why.
But there is more to the above KSI: it shows that the existence of attractors at the sources
implies total absence of scalar sources, either of primary or secondary type. Since this seems
to be necessary in order to have regular event horizons, this KSI implies that there will not be
supersymmetric black holes with scalar hair in these theories. Unfortunately, it seems possible to
have singular supersymmetric solutions with primary scalar hair.
We can summarize the results obtained in this section as follows: we have identified a series
of requirements necessary to avoid singularities in supersymmetric black-hole-type solutions
of N = 2, d = 4 supergravity coupled to vector multiplets, which can be associated to having
unbroken supersymmetry everywhere (including the sources).
I The conditions
= 0,
W)
m(
= 0,
W)
m(
(3.14)
(3.15)
have to be satisfied everywhere in order to have supersymmetry everywhere. They ensure

the absence of string singularities associated to source of NUT charge and other singularities
associated to sources of angular momentum We stress that, when dealing with solutions, all
finite contributions to the first equation should be ignored and the second equation can only
have singular terms in the l.h.s.
II The mass has to be positive. Actually, the masses of each of the sources of the solutions
should be positive. They cannot be rigorously defined in general (for multi-black-hole solutions), but they can be identified with certain confidence in the supersymmetric configurations at hands [25].
242
III The attractor equations (2.35) must be satisfied at each of the sources for admissible values
of the scalars and the value of the central charge at each of them must be finite. As we have
seen, the first condition is equivalent to the total absence of scalar sources.
The last two conditions are associated to the finiteness and positivity of grr outside the
sources. Since grr eK , it would be finite and positive as long as the scalar fields take
admissible values within their domain of definition. All the zeroes of grr can be related to singularities of the scalar fields. Imposing that the scalar fields take admissible values everywhere
is too strong a condition, since it is almost equivalent to directly impose absence of singularities
in the metric.
The conditions that we have imposed are, however, heuristically equivalent: for a single blackhole solution the conditions of asymptotic flatness and positivity of the masses ensure positivity
of grr in the limit r . The third condition ensures positivity in the r 0 limit and,
furthermore, ensures that there will be a horizon of finite area. Since there are no reasons to
expect singularities at finite values of r, the positivity and finiteness should hold for all finite
values of r. The same should happen in multi-black-hole solutions.
4. N = 2, d = 4 attractors, KSIs and BPS black-hole sources
Now we want to apply the results of the previous sections to several examples of black-holetype solutions of N = 2, d = 4 supergravity theories, demanding the three conditions formulated
in the introduction and checking the regularity of those solutions that satisfy them. We are going
to start with the simplest theory.
4.1. Pure N = 2, d = 4 supergravity
This theory has n = 1, no scalar fields, and it is given by the prepotential
i 0
2
F0 = iX 0 .
X
2
This implies that the components of the symplectic section V are constant
L0 = iM0 = ei / 2,
F =
(4.1)
(4.2)
and X is not related to any Khler potential, but
1
ei
ei
=
.
X = L0 /X
2
2(R0 + iI 0 )
(4.3)
The central charge is constant and given by
iei
iei
Z = p 0 iq0 q.
(4.4)
2
2
The attractor equations do not make sense because Z is already moduli-independent.
The timelike supersymmetric configurations of this theory were first found by Tod in his
pioneering paper Ref. [19], belong to the family of solutions found by Perjs, Israel and Wilson
(IWP) [26,27]; they are completely determined by the choice of a single complex, harmonic
In the framework of general N = 2, d = 4 theories, the solutions
function that we denote by I.
of pure N = 2, d = 4 supergravity are given by just two real harmonic functions I 0 and I0 , the
243
components of the real symplectic vector I. The relation between I and I is

I = I 0 iI0 .
(4.5)
Observe that
iei
X = ,
(4.6)
2I
and therefore 2X coincides with the function V of Ref. [19] and is the inverse of the complex
harmonic function.
It is convenient to use the complex formulation of this theory. In it, the symplectic product
where the tilde
of two real symplectic vectors x, y can be written in the form x|y = m(x y)
indicates complexification (x = x 0 ix0 etc.). Further, electricmagnetic duality rotations of the
We would like to
symplectic vectors is equivalent to multiplication by a global phase x = ei x.
stress that the metric is invariant under these transformations.
Using Eq. (4.1) one finds that R, the real part of V/X is the symplectic vector

I0
2.
= i I grr = 1 = R|I = |I|
R=
(4.7)
R
I0
2|X|2
Finally,
= I
I.
W
(4.8)
It was argued by Hartle and Hawking [18] that the only regular black hole solutions in the
IWP family are the static MajumdarPapapetrou solutions that describe several charged black
holes in static equilibrium. We are going to see that these are in fact the only solutions which
are everywhere supersymmetric (condition I) and that demanding positivity of the masses of the
components (condition II) is enough to have regular black holes (condition III plays no rle here).
4.1.1. Single, static black hole solutions
The complex harmonic function I adequate to describe a static, spherically symmetric, extreme black hole with magnetic and electric charges p 0 and q0 is
q
I = I + ,
r
q p 0 iq0 ,
(4.9)
and asymptotic flatness requires |I | = 1. Since I is just a phase that can be taken to be unity
by an electricmagnetic duality rotation. Then,

|q|
2
2 = 1 + 2 e(I q)
grr = |I|
+ 2 .
r
r
The mass is given by

M = e I
q = R |q,
(4.10)
(4.11)
and the equations of motion and supersymmetry seem to allow for it to be positive or negative.
When M is negative |I|2 will vanish for some finite value of r, giving rise to a naked singularity.
In the limit r 0, which makes sense if M is positive, we find that the area of the 2-spheres of
constant t and r is finite and equal to
A = 4|q|
2 = 8|Z|2 .
(4.12)
244
Observe that, in general,
|M| = 2 |Z|,
(4.13)
even though these solutions are usually understood to be supersymmetric.

For this solution Eq. (3.14) is automatically satisfied, while Eq. (3.15) takes the form

(3)
= 4 m I
W)
m(
(4.14)
q (
x ) = 0.
We can, either
(1) Adopt the point of view proposed in this paper that the integrability condition has to be
satisfied everywhere (condition I), whence impose the condition

m I
(4.15)
q = I |q = 0.
q|
ei . The complex harI is just a phase and this condition determines it: I = q/|
monic function becomes

|q|
I =e 1
(4.16)
.
r
The overall phase ei is irrelevant for our problem (it can always be eliminated by an
electricmagnetic duality rotation that does not change the metric), but the relative sign
between the two terms, which is the sign of the mass,
M = |q|
= |Z|,
(4.17)
is important since the minus sign leads to naked singularities. We take the positive sign as
to comply with condition II. We can the integrate the equation for everywhere. The above
condition, however, implies the vanishing of the r.h.s. of the equation and, therefore, also
that of . Thus, after imposing conditions I and II we obtain a solution which is static and
spherically symmetric and has a regular horizon if M > 0; or
(2) We can accept this singularity, ignoring condition I, arguing that, after all, the harmonic
functions are already singular at that point10 and proceed to integrate the equation and obtain
which, in spherical coordinates, takes the form
= 2N cos d,
where N is NUT charge and it is given by
q = I |q |M + iN | = 2 |Z |.
N = m I
(4.18)
(4.19)
The metric is no longer static, but stationary, and contains either wire singularities or closed
timelike curves plus TaubNUT asymptotics.
It is clear that by imposing conditions I and II, these pathologies are avoided. Furthermore, in
the microscopic models of black holes constructed in the framework of string theory there seem
to be no configurations that give rise to macroscopic NUT charge (nor to negative masses). The
agreement between spacetime supersymmetry and the microscopic string theory models on this
point, together with the elimination of pathologies is encouraging and we will see that it applies
to more cases.
10 We have seen that the solution can, nevertheless, be regular at that point, which is the event horizon.
245
4.1.2. Single black hole solutions with a dipole term

Let us now consider harmonic functions adequate to describe rotating supersymmetric black
holes. We can add angular momentum to the previous solution by adding a dipole term to its
complex harmonic function which becomes:
q
)
1,
I = I + + (m
r
r
(4.20)
0 ) is a symplectic vector of dipole magnetic and electric momenta. When they

where m
= (m
0, m
are parallel we can take them to have only z component and, then, in spherical coordinates
q m
cos
I = I +
.
r
r2
is
The corresponding (which exists except at the singularities of I)

sin2
sin2
= 2N cos + 2J 2 + m q m
d.
r
r3
(4.21)
(4.22)
N is the NUT charge and is given again by Eq. (4.19). The new features are J , the z component
of the angular momentum, given by

J = m I
(4.23)
m
= I |m, ,
which does not have a conventional name but vanishes when N = J = 0.
and m(q m)
Let us now analyze the KSIs Eqs. (3.14) and (3.15) (condition I). In the general case they take,
respectively, the form

1

1 i m
1 m
1 = 0,
2 m q m
(4.24)
r
r
r
r

(3)
1
q (
x ) + m I
m (3) (
x ) + m q m (3) (
x)
m I

r
1
1
x ) m(q
m ) + m m
x ) = 0,
m (3) (
+ (3) (
(4.25)
r
r
and are satisfied if

N = m I
q = I |q = 0,

= I |m
= 0,
J = m I m

= q|m
= 0,
m q m

n ] = m[m |mn ] = 0,
m m
[m m
(4.26)
(4.27)
(4.28)
(4.29)

and where we have taken into account
where we have defined the differential operator m m
Eq. (4.23) to identify the angular momentum.
The first condition is, again, the absence of sources of NUT charge. The second condition is
the absence of sources of angular momentum. The third and fourth conditions are automatically
satisfied in this theory if the first two are.
In this case, these conditions are not enough to eliminate all the singularities introduced by the

1 |2 in the grr component of
dipole term since the above conditions do not cancel terms like |m
r
the metric and we no longer find a regular 2-sphere in the r 0 limit. However, we are going to
246
argue that, although technically possible, dipole terms should not be allowed in I because their
only possible origin is a distribution of point-like charges and it is the fundamental distribution of
point-like charges that we have to consider in the above equations and not the field they produce
at distances larger than its size. It is in these conditions that imposing supersymmetry everywhere
is equivalent to cosmic censorship.
Indeed, from the point of view of the electromagnetic fields, the magnetic dipole momenta,
for instance, can have two fundamental origins: dipole momenta in a distribution of magnetic
monopoles or fundamental dipole momenta that can be seen as stationary electric currents. In
standard electrodynamics the first possibility is experimentally excluded (see, e.g. Ref. [28]) but
in N = 2, d = 4 supersymmetric configurations it is the only one allowed (see Eq. (3.7)).
4.1.3. The supersymmetric KerrNewman solution
Therefore we must only consider distributions of static point-like charges. We will do so in a
moment, but there is an interesting example of rotating black-hole-type solution which must be
considered before: it is given by the complex harmonic function

q
I = I + , r x 2 + y 2 + (z i)2 ,
(4.30)
r
which is known to lead to the (ultra-extreme) supersymmetric KerrNewman solution with
angular momentum around the z axis; as is known it has naked singularities, as all 4-dimensional
supersymmetric rotating black-holes [29]. This is the prototype of solution for which supersymmetry does not act as a cosmic censor as proposed in [14]. Generalizations of this solution
in some other N = 2, d = 4 theories have been constructed in Ref. [4].
The asymptotic expansion of I
q i qz
I I + 3 + ,
(4.31)
r
r
corresponds to a charge distribution with only two independent parameters: and q.
The magnetic (electric) dipole momentum is equal to the product of and the electric (magnetic) charge
and the infinite number of non-vanishing higher momenta depend also on these few parameters.
According to the point of view advocated here this solution should not be considered because
it corresponds to the far field of a very charge distribution. As we are going to see, condition I is
enough to exclude it.
Finding the sources of the solution associated to the above complex harmonic function is
very complicated. To start with, I is singular on the ring x 2 + y 2 = 2 , z = 0 but it is also
discontinuous on a disk bounded by the ring. (See e.g. [30], whose results we are going to use
here. See also Refs. [31,32].)
Eqs. (3.14) and (3.15), which express condition I, take, respectively, the form

1
2

m I q e( C) + e I q m( C) + |q|
(4.32)
m C = 0,
r

1
+ e I
+ |q|
C)
C)
q e(
q m(
2 m
C = 0,
m I
(4.33)
r
where we have defined
C
(x, y, z i)
.
[x 2 + y 2 + (z i)2 ]3/2
(4.34)
247
The curl and divergence of C have been carefully computed in Ref. [30] in a distributional
sense, i.e. as integrals of their products with test functions. For us it is enough to known that
= m(
= 0,
C)
C)
e(
(4.35)
vanishes for vanishing . We are left with

C)
and that m(

1
= 0,
C)
e I
q + |q|
2 e
m(
r

1
= 0.
C)
q |q|
2 m
e(
m I
r
(4.36)
(4.37)
= 0, which requires = 0
C)
The only way to satisfy the first condition is to have m(

(no sources of angular momentum). Since e( C) = 0 always, the only way to satisfy the
q)
second condition is to have m(I
= 0 as before (no sources of NUT charge) and m 1r = 0
which also requires = 0.
Thus, imposing supersymmetry everywhere is equivalent, yet again, to requiring absence of
sources of NUT charge and angular momentum. In the supersymmetric KerrNewman solution
all the angular momentum originates in that source11 and, thus, that solution and its naked singularities can be excluded from the class of everywhere supersymmetric solutions of N = 2,
d = 4 supergravity. Again, supersymmetry acts as a cosmic censor and, most importantly, there
is agreement between the macroscopic description of black holes provided by supergravity and
the microscopic models provided by string theory in which there seems to be no way of having
angular momentum without breaking supersymmetry.
Therefore, we must only consider distributions of point-like charges, which correspond to
complex harmonic functions of the form
I = I +

A
qA
,
|
x xA |
(4.38)
from which dipole (and higher) momenta arise only in asymptotic expansions:

( A qA xA ) x
A qA
+ ,
+
I I +
|
x|
|
x |3
and may give rise to non-vanishing angular momentum

= I |m,
J = m I

m
=
qA xA ,
m
(4.39)
(4.40)
but not to non-vanishing NUT charge.

q = I |q = 0, q =
qA .
N = m I
(4.41)
We are going to look for this kind of solutions in pure N = 2, d = 4 supergravity next, recovering the (negative) Hartle and Hawking result [18]. We will have to look for them in more
general N = 2, d = 4 theories.
11 We are going to see that there are solutions with angular momentum and no elementary sources of angular momentum.
248
4.1.4. Solutions with two black holes

Let us consider, to start with, just two poles
I = I +
q1
q2
+
.
|
x x1 | |
x x2 |
(4.42)
Asymptotic flatness requires |I | = 1. The condition Eq. (3.14) is automatically satisfied and
(3.15) takes the form

q2 |q1 (3)
q1 |q2 (3)
I |q1 +
(4.43)
x x1 ) + I |q2 +
x x2 ) = 0,
(
(
|
x1 x2 |
|
x2 x1 |
which leads to the two equations
q2 |q1
= 0,
|
x1 x2 |
q1 |q2
= 0,
I |q2 +
|
x2 x1 |
I |q1 +
(4.44)
each of which expresses the absence of sources of NUT charge at x1 and x2 . The antisymmetry
of the symplectic product implies the consistency condition
I |q1 + q2 = 0,
(4.45)
which means that the total charge of the two objects satisfies the same condition (no global NUT
charge) as the charge of just one.
Expanding asymptotically I and using the above constraints we find that this two-body system
has a total mass and angular momentum given by

M=
(4.46)
R |qA
MA ,
A
(
x2 x1 )
= q1 |q2
.
J = I |m
(4.47)
|
x2 x1 |
Observe that there is total angular momentum even though there are no sources of angular
momentum.
There are two types of solutions to these equations required by condition I:
(1) Each objects charge satisfies the condition for single independent objects I |qA = 0
which requires q2 |q1 = 0. In this theory this means that the phases of I , q1 and q2 are
such that

sA |qA |
i
I =e 1+
(4.48)
,
|
x xA |
A
where sA = 1. The total mass is given by the formula Eq. (4.11)

M = e I
qA = R
qA =
sA |qA |,
A
(4.49)
and the angular momentum vanishes ( vanishes).

These are the MajumdarPapapetrou solutions [21,22]. Only the solutions with all sA = +1
are regular, but one could argue that only those correspond to objects that would have positive
masses MA = |qA | if they were isolated [25]. This is the meaning of condition II.
249
These solutions describe two charged, static black holes in equilibrium with their event horizons placed at x1 and x2 which are really 2-spheres of finite areas equal to 4|q1 |2 and
4|q2 |2 . They are, as argued by Hartle and Hawking, and as we are going to see, the only
regular black-hole-type solutions in the whole IWP family [18].
(2) I |qA = 0 and we have two objects that cannot exist independently in the vacuum I
(i.e. we have a bound state). The distance between them is fixed by the condition of absence
of sources of NUT charge to be
q1 |q2
(4.50)
.
I |q1
The sign of the r.h.s. can always be made positive by flipping the sign of I , which is
irrelevant for the moduli and for solving Eq. (4.45). Thus, this equation always has a solution.
However, when all the above conditions have been satisfied, the total mass of the solution is
negative. The simplest way to see this is by first making I = 1 by a duality rotation that
does not change the metric. After the duality rotation one finds qA = MA + iNA , meaning
that they are complex combinations of the masses and NUT charges of each object. Using
N2 = N1 , the above condition takes the form

N 1 M2 N 2 M1
M1 + M2
= N1 1 +
= 0,
N1 +
(4.51)
|
x2 x1 |
|
x2 x1 |
which has solution only for vanishing NUT charges or for negative total mass M1 + M2
which violates condition II and produces naked singularities. Thus, we cannot simultaneously satisfy conditions I and II for bound states with q1 |q2 = 0.
|
x2 x1 | =
This result can be generalized to solutions with more poles: let us consider first the 3-pole
harmonic function
q1
q2
q3
+
+
.
I = I +
(4.52)
|
x x1 | |
x x2 | |
x x3 |
The integrability condition leads to three equations (one to cancel the NUT charge at each
pole) which can be written as a linear system for the NA s:
M3
2
1
1
(1 + M
M
M
" #
r12 + r14 )
r12
r13
N1
M3
M2
M1
M2
(4.53)
r12
(1 + r12 + r23 )
r23
N2 = 0.
N3
M3
M3
M1
M2
r13
r23
(1 + r13 + r23 )
It is easy to see that the determinant of the matrix is +1 plus terms linear and quadratic
in the masses, all with positive sign. It will never vanish if all the masses are positive. This
argument can be easily generalized to a higher number of poles and, therefore we conclude that
the only solutions satisfying conditions I and II are the MajumdarPapapetrou solutions. This
result should be read in a positive sense: no singular solutions are allowed by the conditions
proposed in the introduction, even if only static solutions are allowed in this simple theory. To
find solutions with angular momentum satisfying conditions IIII we need to consider theories
with scalars.
4.2. General N = 2, d = 4 supergravity
The setup of our problem in general N = 2, d = 4 theories is similar to pure supergravity
case. Let us first consider spherically-symmetric, static, single black-hole-type solutions with
250
magnetic and electric charges p and q . They are determined by a symplectic vector of 2n real
harmonic functions

q
I
I
p
I=
(4.54)
= I + , q
, I
.
I
q
I
r
We assume that the stabilization equations have been solved and R (I) has been found in
order to be able to construct the fields of the solutions.
The n complex scalars are constructed using the general formula Eq. (2.5). The moduli (the
i ) are complicated functions Z i (I ) of these
values of the n complex scalars Z i at infinity, Z

2n + 2 real constant components of I . One of the components of I can be determined
as a function of the remaining 2n + 1 by imposing asymptotic flatness of the metric, that is,
R |I = 1, and another one can be determined by imposing condition I, since Eq. (3.8) implies
N = I |q = 0.
(4.55)
It should always be possible to give the 2n real moduli any admissible value within their definition domain with the remaining 2n unconstrained real components of I . This is difficult
to prove explicitly due to the complicated and theory-dependent relations between I and the
i , but it is safe to assume that in general it is possible.
moduli Z
Let us turn to condition II. The positivity of the masses, which is given by the general expression Eq. (2.53) has to be imposed by hand and, although this can always be done, it is a non-trivial
constraint on the charges and moduli. The positivity of the masses can be also understood as part
of a stronger requirement that the scalar fields take values only within their definition domain
for all values of r. Actually, this requirement should suffice to ensure the finiteness of grr for
r = 0.
The finiteness of grr for r = 0 is not enough to have a black hole and condition III has to be
imposed to find a finite horizon area at r = 0.
If we want to describe more than one black hole we have to use harmonic functions with two
point-like singularities:
I = I +
q1
q2
+
.
|
x x1 | |
x x2 |
(4.56)
Again, one of the components of I is determined by imposing asymptotic flatness. Condition

I now leads to the two equations Eqs. (4.44) which should determine another component of I
and the parameter |
x1 x2 | if q2 |q1 = 0. The question now is whether these solutions can be
obtained while maintaining the positivity of the masses (condition II)
Mi R |qi > 0,
(4.57)
and solving the attractor equations for each of the singularities of the harmonic functions. We
have no general answer to these questions and, what we are going to do is to study how the
three conditions can actually be imposed in a particularly simple example and suffice to ensure
regularity of the solutions.
4.2.1. A toy model with a complex scalar field
We are going to consider the n = 2 theory with prepotential
F = iX 0 X 1 .
(4.58)
251
This theory has only one complex scalar

iX 1 /X 0 ,
(4.59)
in terms of which the period matrix is given by

0
(N ) =
0 1/
(4.60)
and, in the X 0 = i/2 gauge, the Khler potential and metric are
K = ln m ,
G = (2 m )2 .
(4.61)
The reality of the Khler potential requires the positivity of m . Therefore, parametrizes the
coset SL(2, R)/SO(2) and can be identified with the axidilaton and this theory is a truncation of
the SO(4) formulation of N = 4, d = 4 supergravity.
The symplectic section V is
i
1

V=
(4.62)
,
2( m )1/2 i
1
and the central charge is

Z , , q = V|q =
1

1
p iq0 q1 + ip 0 .
1/2
2( m )
(4.63)
The attractor equation is

d 1 1
p iq0 q1 + ip 0 = = 0,
fix
d m
and has the general solution
fix =
p 1 + iq0
,
q1 ip 0
(4.64)
(4.65)
which is admissible (belongs to the definition domain of ) if

m fix = p 0 p 1 + q0 q1 > 0.
The central charge at the fixed point of the scalar takes the value

q1 + ip 0
Zfix = i
p 0 p 1 + q0 q1 ,
|q1 + ip 0 |
(4.66)
(4.67)
and it is always finite for fix = 0.

Solutions with a single black hole
Let us now consider solutions with
q
I = I + .
r
(4.68)
252
In this theory the stabilization equations can be easily solved and they lead to

0 1
I grr = R|I = 2 I 0 I 1 + I0 I1 ,
R=
1
0
(4.69)
which shows that the area of the horizon (if any) is related to |Zfix |2 above according to the
general formula Eq. (2.55).
We also have
=i
L1 /X I 1 + iI0
=
,
L0 /X I1 iI 0
(4.70)
which implies that the 4 harmonic functions are not entirely independent but have to satisfy
m = I 0 I 1 + I0 I1 > 0,
(4.71)
which ensures that, if there are no pathologies that make a black-hole interpretation of the solution impossible, the attractor equations will always have solutions and Zfix = 0. Thus, we will
not have to worry about condition III but only about the positive definiteness of m .
The only possible pathologies (negative mass and presence of NUT charge) are clearly
avoided by imposing conditions I and II, which is always possible and presents no difficulties.
Solutions with two black holes
Let us now consider solutions of the form
q1 q 2
I = I +
+ , ri |
x xi |.
r1
r2
(4.72)
Our goal is to find a configuration (i.e. a set of asymptotic values I and charges q1,2 ) that
satisfy conditions IIII. The previous discussions indicate how this has to be done and which
formulas need to be applied. There is no systematic procedure to find such a configuration but it
is not too difficult to find one:
1
q
q
I0 = + + ,
2 r1 r2
1
8q 8q
+
,
I1 = +
r1
r2
2
4q
I0 = ,
r2
1
q
q
I1 = + ,
4 2 r1 r2
where q > 0 in order to guarantee Eq. (4.71). The metric component
9 2q 10 2q 16q 2 8q 2 40q 2
+
+ 2 + 2 +
,
grr = 1 +
r1
r2
r1 r2
r1
r2
(4.73)
(4.74)
is finite everywhere outside r1,2 = 0, and therefore, so is m . In particular the mass of each
of the two objects is positive
M1 = 9q/ 2,
(4.75)
M2 = 5 2q,
M = M1 + M2 = 19q/ 2,
253
and in the r1,2 0 limits we find spheres of finite areas

A1
= 16q 2 = 2|Zfix,1 |2 ,
4
A2
= 8q 2 = 2|Zfix,2 |2 .
4
(4.76)
The total horizon area is

A1 A2
A
=
+
= 24q 2 < 2|Zfix,tot |2 = 64q 2 ,
4
4
4
(4.77)
which is the area of the horizon of a single black hole having the sum of the charges of the two
black holes.
For this configuration
q2 |q1 = 12q 2 ,

I |q1 = I |q2 = q/ 2,
(4.78)
so, choosing
r12 = |
x2 x1 | = 12 2q,
(4.79)
we satisfy condition I (no NUT charges). The system has nevertheless angular momentum given
by the general formula Eq. (4.47):

|J | = q2 |q1 = 12q 2 .
(4.80)
5. Conclusions
We have formulated three conditions that supersymmetric black-hole-type solutions have to
satisfy in order to be supersymmetric everywhere, including at the sources. We have shown how
these conditions constrain the possible sources by, basically, excluding those with NUT charge,
angular momentum, negative energy and scalar hair, which seemingly cannot be modeled in
string theory. We arrived at a picture in which if an observer far away from one of the globally
supersymmetric configurations we have considered, detects angular momentum and non-trivial
scalar fields he/she will only find static electromagnetic sources in equilibrium when approaching
the system.
These conditions and this picture should be improved by considering quantum corrections.
Another interesting course of action would be to consider regularity of black-hole solutions in
N > 2 theories, e.g. [33], and investigate the rle played by the attractor [11].
It is also clear that the situation in d = 5 is completely different as there are regular rotating
supersymmetric black holes for which microscopic string theory models are known [34]. Work
on these issues is already in progress [35].
Acknowledgements
T.O. would like to thank Renata Kallosh for pointing the authors towards Refs. [16] and [17]
and for useful conversations, Roberto Emparan for his explanations concerning rotating blackhole configurations and many other useful comments and, finally, M.M. Fernndez for her long
standing support. This work has been supported in part by the Spanish Ministry of Science and
Education grant BFM2003-01090 and the Comunidad de Madrid grant HEPHACOS P-ESP00346.
254
Appendix A. Proofs of some identities

Let us consider the generalized stabilization equations derived from Eq. (2.3). Differentiating
the imaginary part of that equation (i.e. Eq. (2.4)), we get
dX F dX F
2i
= dR m F + dI e F ,
dI = d m F =
(A.1)
where we have used X = R + iI . Using the invertibility of the imaginary part of F we

get
dR = m F dI m F e F dI .
(A.2)
On the other hand, differentiating the real part of Eq. (2.3)
dX F + dX F
2
= dR e F dI m F ,
dR = d e F =
and, substituting our previous result for
(A.3)
dR

dR = e F m F dH m F + e F m F e F dI .
We can write all these results in the form

m F 1 e F
m F 1
dR =
dI,
( m F + e F m F 1 e F ) e F m F 1

m F 1
m F 1 e F
dR,
dI =
m F + e F m F 1 e F e F m F 1
(A.4)
(A.5)
(A.6)
from which we can read identities such as

R
R
I
=
=
,
I
I
R
R
I
R
=
=
,
I
I
R
R
R
I
=
=
,
I
I
R
R
R
I
= =
.
I
I
R
We can now prove Eq. (2.46): taking the derivative of R as a function of I we have

R
R
I
+
I
R|I =
I
I

R
R
R
R
= I I
I
I
I
+
I
,
I
I
I
I
and using now the above relations between partial derivatives

R
R
R
R
R|I = I I
+ I
+ I
I I
.
I
I
I
I
(A.7)
(A.8)
(A.9)
Given that the real section R is homogeneous of first order in the Is

I
R
R
+ I
= R ,
I
I
R
R
+
I
= R ,
I
I
(A.10)
which proves the identity.

Similarly, expanding the r.h.s. of Eq. (2.45) we get

R
R
R
R
R
R dI +
R
R dI ,
R| R =
I
I
I
I
255
(A.11)
and using the identities between partial derivatives and the fact that the real section I is homogeneous of first order in R, we arrive at the result we wanted.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
H. Elvang, R. Emparan, D. Mateos, H.S. Reall, JHEP 0508 (2005) 042, hep-th/0504125.
P. Meessen, T. Ortn, Nucl. Phys. B 749 (2006) 291, hep-th/0603099.
M. Hbscher, P. Meessen, T. Ortn, hep-th/0606281.
K. Behrndt, D. Lst, W.A. Sabra, Nucl. Phys. B 510 (1998) 264, hep-th/9705169.
R. Kallosh, T. Ortn, hep-th/9306085.
J. Bellorn, T. Ortn, Phys. Lett. B 616 (2005) 118, hep-th/0501246.
S. Ferrara, R. Kallosh, A. Strominger, Phys. Rev. D 52 (1995) 5412, hep-th/9508072.
A. Strominger, Phys. Lett. B 383 (1996) 39, hep-th/9602111.
A. Ceresole, R. DAuria, S. Ferrara, Nucl. Phys. B (Proc. Suppl.) 46 (1996) 67, hep-th/9509160.
S. Ferrara, R. Kallosh, Phys. Rev. D 54 (1996) 1514, hep-th/9602136.
S. Ferrara, R. Kallosh, Phys. Rev. D 54 (1996) 1525, hep-th/9603090.
S.R. Coleman, J. Preskill, F. Wilczek, Nucl. Phys. B 378 (1992) 175, hep-th/9201059.
G.W. Moore, hep-th/9807087.
R. Kallosh, A.D. Linde, T. Ortn, A.W. Peet, A. Van Proeyen, Phys. Rev. D 46 (1992) 5278, hep-th/9205027.
A. Ceresole, R. DAuria, S. Ferrara, A. Van Proeyen, Nucl. Phys. B 444 (1995) 92, hep-th/9502072.
F. Denef, JHEP 0008 (2000) 050, hep-th/0005049.
B. Bates, F. Denef, hep-th/0304094.
J.B. Hartle, S.W. Hawking, Commun. Math. Phys. 26 (1972) 87.
K.P. Tod, Phys. Lett. B 121 (1983) 241.
C. Misner, J. Math. Phys. 4 (1963) 924.
S.D. Majumdar, Phys. Rev. 72 (1947) 390.
A. Papapetrou, Proc. R. Irish Acad. A 51 (1947) 191.
A.I. Janis, E.T. Newman, J. Winicour, Phys. Rev. Lett. 20 (1968) 878.
A.G. Agnese, M. La Camera, Phys. Rev. D 49 (1994) 2126.
D.R. Brill, R.W. Lindquist, Phys. Rev. 131 (1963) 471476.
Z. Perjs, Phys. Rev. Lett. 27 (1971) 1668.
W. Israel, G.A. Wilson, J. Math. Phys. 13 (1972) 865.
J.D. Jackson, Yellow report CERN-77-17 SPIRES entry reprinted, in: V. Stefan, V.F. Weisskopf (Eds.), Physics and
Society: Essays in Honor of Victor Frederick Weisskopf, AIP Press/Springer, New York/Berlin, 1998, p. 236.
E. Bergshoeff, R. Kallosh, T. Ortn, Nucl. Phys. B 478 (1996) 156, hep-th/9605059.
G. Kaiser, J. Phys. A 37 (2004) 8735, gr-qc/0108041.
E.T. Newman, Phys. Rev. D 65 (2002) 104005, gr-qc/0201055.
A. Gsponer, gr-qc/0405046.
M. Cvetic, D. Youm, Phys. Rev. D 53 (1996) 584, hep-th/9507090;
M. Cvetic, A.A. Tseytlin, Phys. Rev. D 53 (1996) 5619, hep-th/9512031;
M. Cvetic, A.A. Tseytlin, Phys. Rev. D 55 (1997) 3907, Erratum;
M. Cvetic, C.M. Hull, Nucl. Phys. B 480 (1996) 296, hep-th/9606193.
J.C. Breckenridge, R.C. Myers, A.W. Peet, C. Vafa, Phys. Lett. B 391 (1997) 93, hep-th/9602065.
J. Bellorn, P. Meessen, T. Ortn, in preparation.
Azimuthal asymmetries in DIS as a probe

of intrinsic charm content of the proton
L.N. Ananikyan, N.Ya. Ivanov
Yerevan Physics Institute, Alikhanian Br. 2, 375036 Yerevan, Armenia
Received 16 October 2006; accepted 6 November 2006
Abstract
We calculate the azimuthal dependence of the heavy-quark-initiated O(s ) contributions to the lepton
nucleon deep inelastic scattering (DIS). It is shown that, contrary to the photongluon fusion (GF) component, the photonquark scattering (QS) mechanism is practically cos 2-independent. We investigate the
possibility to discriminate experimentally between the GF and QS contributions using their strongly different azimuthal distributions. Our analysis shows that the GF and QS predictions for the azimuthal cos 2
asymmetry are quantitatively well defined in the fixed flavor number scheme: They are stable, both parametrically and perturbatively. We conclude that measurements of the azimuthal distributions at large Bjorken x
could directly probe the intrinsic charm content of the proton. As to the variable flavor number schemes,
the charm densities of the recent CTEQ and MRST sets of parton distributions have a dramatic impact on
the cos 2 asymmetry in the whole region of x and, for this reason, can easily be measured.
PACS: 12.38.-t; 13.60.-r; 13.88.+e
Keywords: Perturbative QCD; Heavy flavor leptoproduction; Intrinsic charm; Azimuthal asymmetries
1. Introduction
The notion of the intrinsic charm (IC) content of the proton has been introduced over 25 years
ago in Refs. [1,2]. It was shown that, in the light-cone Fock space picture [3,4], it is natural to
expect a five-quark state contribution to the proton wave function. The probability to find in a
E-mail addresses: lev@web.am (L.N. Ananikyan), nikiv@uniphi.yerphi.am (N.Ya. Ivanov).

doi:10.1016/j.nuclphysb.2006.11.008
L.N. Ananikyan, N.Ya. Ivanov / Nuclear Physics B 762 (2007) 256283
257
nucleon the five-quark component |uudcc

is of higher twist since it scales as 1/m2 where m is
the c-quark mass [5]. This component can be generated by gg cc fluctuations inside the proton
where the gluons are coupled to different valence quarks. Since all of the quarks tend to travel
coherently at same rapidity in the |uudcc
bound state, the heaviest constituents carry the largest
momentum fraction. For this reason, one would expect that the intrinsic charm component to be
dominate the c-quark production cross sections at sufficiently large Bjorken x. So, the original
concept of the charm density in the proton [1,2] has nonperturbative nature and will be referred
to in the present paper as nonperturbative IC.
A decade ago another point of view on the charm content of the proton has been proposed in
the framework of the variable flavor number scheme (VFNS) [6,7]. The VFNS is an approach
alternative to the traditional fixed flavor number scheme (FFNS) where only light degrees of
freedom (u, d, s and g) are considered as active. It is well known that a heavy quark production
cross section contains potentially large logarithms of the type s ln(Q2 /m2 ) whose contribution
dominates at high energies, Q2 . Within the VFNS, these mass logarithms are resummed
through the all orders into a heavy quark density which evolves with Q2 according to the standard
DGLAP [810] evolution equation. Hence the VFN schemes introduce the parton distribution
functions (PDFs) for the heavy quarks and change the number of active flavors by one unit when
a heavy quark threshold is crossed. We can say that the charm density arises within the VFNS
perturbatively via the g cc evolution and will call it the perturbative IC.
Presently, both perturbative and nonperturbative IC are widely used for a phenomenological
description of available data. (A recent review of the theory and experimental constraints on the
charm quark distribution can be found in Refs. [11,12]. See also Appendix C in the present paper.) In particular, practically all the recent versions of the CTEQ [13] and MRST [14] sets of
PDFs are based on the VFN schemes and contain a charm density. At the same time, the key
question remains open: How to measure the intrinsic charm content of the proton? The basic
theoretical problem is that radiative corrections to the fixed order predictions for the production
cross sections are large. In particular, the next-to-leading order (NLO) corrections increase the
leading order (LO) results for both charm and bottom production cross sections by approximately
a factor of two at energies of the fixed target experiments. Moreover, soft gluon resummation of
the threshold Sudakov logarithms indicates that higher-order contributions are also essential.
(For a review see Refs. [15,16].) On the other hand, perturbative instability leads to a high sensitivity of the theoretical calculations to standard uncertainties in the input QCD parameters. For
this reason, it is difficult to compare pQCD results for spin-averaged cross sections with experimental data directly, without additional assumptions. The total uncertainties associated with the
unknown values of the heavy quark mass, m, the factorization and renormalization scales, F
and R , QCD and the PDFs are so large that one can only estimate the order of magnitude of
the pQCD predictions for production cross sections [17,18].
Since production cross sections are not perturbatively stable, it is of special interest to study
those observables that are well-defined in pQCD. A nontrivial example of such an observable was
proposed in Refs. [1922] where the azimuthal cos 2 asymmetry in heavy quark photo- and
leptoproduction has been analyzed.1 In particular, the Born level results have been considered
[19] and the NLO soft-gluon corrections to the basic mechanism, photongluon fusion (GF),
have been calculated [20,21]. It was shown that, contrary to the production cross sections, the
1 The well-known examples are the shapes of differential cross sections of heavy flavor production which are sufficiently stable under radiative corrections.
258
azimuthal asymmetry in heavy flavor photo- and leptoproduction is quantitatively well defined
in pQCD: The contribution of the dominant GF mechanism to the asymmetry is stable, both
parametrically and perturbatively. Therefore, measurements of this asymmetry would provide
an ideal test of pQCD. As was shown in Ref. [22], the azimuthal asymmetry in open charm
photoproduction could have been measured with an accuracy of about ten percent in the approved
E160/E161 experiments at SLAC [23] using the inclusive spectra of secondary (decay) leptons.
In the present paper we study the IC contribution to the azimuthal asymmetries in heavy quark
leptoproduction:
X ).
l() + N(p) l( q) + Q(pQ ) + X[Q](p
(1)
Neglecting the contribution of Z-boson as well as the target mass effects, the cross section of the
reaction (1) for unpolarized initial states may be written as

d3 lN
em 1 y 2
2
2
2
=
x,
Q
+
x,
Q
+
x,
Q
cos 2
T
L
A
dx dQ2 d (2)2 xQ2 1

+ 2 (1 + )I x, Q2 cos .
(2)
The quantity measures the degree of the longitudinal polarization of the virtual photon in the
Breit frame [24],
=
2(1 y)
,
1 + (1 y)2
(3)
and the kinematic variables are defined by

S = ( + p)2 ,
y=
pq
,
p
Q2 = q 2 ,
Q2 = xy S,
x=
=
Q2
,
2p q
4m2
.
S
(4)
The cross sections i (i = T , L, A, I ) in Eq. (2) are related to the structure functions Fi (x, Q2 )
as follows:

Fi x, Q2 =

Q2
i x, Q2
2
8 em x

Q2
F2 x, Q2 =
2 x, Q2 ,
2
4 em
(i = T , L, A, I ),
(5)
where F2 = 2x(FT + FL ) and 2 = T + L . In Eq. (2), T (L ) is the usual N cross section describing heavy quark production by a transverse (longitudinal) virtual photon. The third
cross section, A , comes about from interference between transverse states and is responsible
for the cos 2 asymmetry which occurs in real photoproduction using linearly polarized photons
[19,20,22]. The fourth cross section, I , originates from interference between longitudinal and
transverse components [24]. In the nucleon rest frame, the azimuth is the angle between the
lepton scattering plane and the heavy quark production plane, defined by the exchanged photon
and the detected quark Q (see Fig. 1). The covariant definition of is

Q2 1/x 2 + 4m2N /Q2
r n
cos =
(6)
,
sin =
n ,
r 2 n2
2 r 2 n2
259
Fig. 1. Definition of the azimuthal angle in the nucleon rest frame.
r = p q ,
n = q p pQ .
(7)
In Eqs. (4) and (6), m and mN are the masses of the heavy quark and the target, respectively. Usually, the azimuthal asymmetry associated with the cos 2 distribution, A2 (, x, Q2 ), is defined
by

A2 , x, Q2 = 2cos 2 , x, Q2
d3 lN ( = 0) + d3 lN ( = ) 2 d3 lN ( = /2)
d3 lN ( = 0) + d3 lN ( = ) + 2 d3 lN ( = /2)
2

A (x, Q2 )
2 + R(x, Q )
=
A
x,
Q
,
=
T (x, Q2 ) + L (x, Q2 )
1 + R(x, Q2 )
=
where d3 lN ()
d3 lN
(, x, Q2 , )
dx dQ2 d

cos n , x, Q2 =
and the mean value of cos n is
2
0
(8)
lN
2
d cos n dxddQ
2 d (, x, Q , )
.
2
d3 lN
2 , )
d
(,
x,
Q
2
0
dx dQ d
(9)
In Eq. (8), the quantities R(x, Q2 ) and A(x, Q2 ) are defined as

L
FL

R x, Q2 =
x, Q2 =
x, Q2 ,
T
FT

FA
A
x, Q2 = 2x
x, Q2 .
A x, Q2 =
2
F2
(10)
(11)
Likewise, we can define the azimuthal asymmetry associated with the cos distribution,
A (, x, Q2 ):

A , x, Q2 = 2cos , x, Q2 =
2 d3 lN ( = 0) 2 d3 lN ( = )
d3 lN ( = 0) + d3 lN ( = ) + 2 d3 lN ( = /2)

1 + R(x, Q2 )
2 (1 + ) I (x, Q2 )
= AI x, Q2 (1 + )/2
,
=
2
2
T (x, Q ) + L (x, Q )
1 + R(x, Q2 )
(12)
260
where
I
FI

AI x, Q2 = 2 2
x, Q2 = 4 2 x
x, Q2 .
2
F2
(13)
Remember that y 1 in most of the experimentally reachable kinematic range. Taking also into
account that = 1 + O(y 2 ), we find:

A , x, Q2 = AI x, Q2 + O y 2 .
A2 , x, Q2 = A x, Q2 + O y 2 ,
(14)
So, like the 2 (x, Q2 ) cross section in the -independent case, it is the parameters A(x, Q2 ) and
AI (x, Q2 ) that can effectively be measured in the azimuth-dependent production.
In this paper we concentrate on the azimuthal asymmetry A(x, Q2 ) associated with the
cos 2-distribution. We have calculated the IC contribution to the asymmetry which is described
at the parton level by the photonquark scattering (QS) mechanism given in Fig. 2. Our main
result can be formulated as follows:
Contrary to the basic GF component, the IC mechanism is practically cos 2-independent.
This is due to the fact that the QS contribution to the A (x, Q2 ) cross section is absent (for
the kinematic reason) at LO and is negligibly small (of the order of 1%) at NLO.
As to the -independent cross sections, our parton level calculations have been compared with
the previous results for the IC contribution to 2 (x, Q2 ) and L (x, Q2 ) presented in Refs. [25,
26]. Apart from two trivial misprints uncovered in [25] for L (x, Q2 ), a complete agreement
between all the considered results is found.
Since the GF and QS mechanisms have strongly different cos 2-distributions, we investigate the possibility to discriminate between their contributions using the azimuthal asymmetry
A(x, Q2 ). We analyze separately the nonperturbative IC in the framework of the FFNS and the
perturbative IC within the VFNS.
The following properties of the nonperturbative IC contribution to the azimuthal asymmetry
within the FFNS are found:
The nonperturbative IC is practically invisible at low x, but affects essentially the GF predictions at large x. The dominance of the cos 2-independent IC component at large x leads to
a more rapid (in comparison with the GF predictions) decreasing of A(x, Q2 ) with growth
of x.
Contrary to the production cross sections, the cos 2 asymmetry in charm azimuthal distributions is practically insensitive to radiative corrections at Q2 m2 . Perturbative stability of
the combined GF + QS result for A(x, Q2 ) is mainly due to the cancellation of large NLO
corrections in Eq. (11).
pQCD predictions for the cos 2 asymmetry are parametrically stable; the GF + QS contribution to A(x, Q2 ) is practically insensitive to most of the standard uncertainties in the QCD
input parameters: R , F , QCD and PDFs.
Nonperturbative corrections to the charm azimuthal asymmetry due to the gluon transverse
motion in the target are of the order of 20% at Q2 m2 and rapidly vanish at Q2 > m2 .
We conclude that the contributions of both GF and IC components to the cos 2 asymmetry
in charm leptoproduction are quantitatively well defined in the FFNS: They are stable, both
parametrically and perturbatively, and insensitive (at Q2 > m2 ) to the gluon transverse motion
261
in the proton. At large Bjorken x, the A(x, Q2 ) asymmetry could be a sensitive probe of the
nonperturbative IC.
The perturbative IC has been considered within the VFNS proposed in Refs. [6,7]. The following features of the azimuthal asymmetry should be emphasized:
Contrary to the nonperturbative IC component, the perturbative one is significant practically
at all values of Bjorken x and Q2 > m2 .
The charm densities of the recent CTEQ and MRST sets of PDFs lead to a sizeable reduction
(by about 1/3) of the GF predictions for the cos 2 asymmetry.
We conclude that impact of the perturbative IC on the cos 2 asymmetry is sizeable in the whole
region of x and, for this reason, can easily be detected.
Concerning the experimental aspects, azimuthal asymmetries in charm leptoproduction can,
in principle, be measured in the COMPASS experiment at CERN, as well as in future studies at
the proposed eRHIC [27,28] and LHeC [29] colliders at BNL and CERN, correspondingly.
The paper is organized as follows. In Section 2 we analyze the QS and GF parton level predictions for the -dependent charm leptoproduction in the single-particle inclusive kinematics. In
particular, we discuss our results for the NLO QS cross sections and compare them with available
calculations. Hadron level predictions for A(x, Q2 ) are given in Section 3. We consider the IC
contributions to the asymmetry within the FFNS and VFNS in a wide region of x and Q2 . Some
details of our calculations of the QS cross sections are presented in Appendix A. An overview
of the soft-gluon resummation for the photongluon fusion mechanism is given in Appendix B.
Some experimental facts in favor of the nonperturbative IC are briefly listed in Appendix C.
2. Partonic cross sections
2.1. Quark scattering mechanism
The momentum assignment of the deep inelastic lepton-quark scattering will be denoted as
l() + Q(kQ ) l( q) + Q(pQ ) + X(pX ).
(15)
Taking into account the target mass effects, the corresponding partonic cross section can be
written as follows [30]
em y 2
1 + 4z2
d3 lQ
=
2,Q (z, ) (1 ) L,Q (z, )
dz dQ2 d (2)2 zQ2 1

+ A,Q (z, ) cos 2 + 2 (1 + ) I,Q (z, ) cos .
(16)
In Eq. (16), we use the following definition of partonic kinematic variables:
y=
q kQ
,
kQ
z=
Q2
,
2q kQ
m2
.
Q2
(17)
In the massive case, the (virtual) photon polarization parameter, , has the form [30]
=
2(1 y z2 y 2 )
.
1 + (1 y)2 + 2z2 y 2
(18)
262
Fig. 2. The LO (a) and NLO (b) and (c) photonquark scattering diagrams.
At leading order, O(em ), the only quark scattering subprocess is

(q) + Q(kQ ) Q(pQ ).
The Q cross sections,
are:
(0)
k,Q
(19)
(k = 2, L, A, I ), corresponding to the Born diagram (see Fig. 2a)

(0)
2,Q (z, ) = B (z) 1 + 4z2 (1 z),
(0)
L,Q (z, ) = B (z)
4z2
1 + 4z2
(1 z),
(0)
(0)
A,Q
(z, ) = I,Q (z, ) = 0,
(20)
with
B (z) =
2
(2)2 eQ
em
Q2
(21)
z,
where eQ is the quark charge in units of electromagnetic coupling constant.

To take into account the NLO O(em s ) contributions, one needs to calculate the virtual
corrections to the Born process (given in Fig. 2c) as well as the real gluon emission (see Fig. 2b):
(q) + Q(kQ ) Q(pQ ) + g(pg ).
(22)
(1)
(1)
The NLO -dependent cross sections, A,Q and I,Q , are described by the real gluon emission
(1)
only. Corresponding contributions are free of any type of singularities and the quantities A,Q
(1)
and I,Q can be calculated directly in four dimensions.

(1)
(1)
In the -independent case, 2,Q and L,Q , we also work in four dimensions. The virtual contribution (Fig. 2c) contains ultraviolet (UV) singularity that is removed using the on-mass-shell
regularization scheme. In particular, we calculate the absorptive part of the Feynman diagram
263
which has no UV divergences. The real part is then obtained by using the appropriate dispersion
relations. As to the infrared (IR) singularity, it is regularized with the help of an infinitesimal gluon mass. This IR divergence is cancelled when we add the bremsstrahlung contribution
(Fig. 2b). Some details of our calculations are given in Appendix A.
The final (real + virtual) results for Q cross sections can be cast into the following form:
s
(1)
2,Q (z, ) =
CF B (1) 1 + 4 (1 z)
2

1 + 2
2 Li2 r 2 + 4 Li2 (r)
2 + 4 ln 1 + 4 ln r +
1 + 4

2
+ 3 ln (r) 4 ln r + 4 ln r ln(1 + 4) 2 ln r ln

s
1
1
+
1 3z 4z2 + 6z3 + 8z4
CF B (z)
2
3/2
2
4
(1 + 4z )
[1 (1 )z]

8z5 + 6z 3 18z + 13z2 + 10z3 8z4 + 42 z2 8 77z + 65z2 2z3

+ 163 z3 1 21z + 12z2 1284 z5
2 ln D(z, )
+
1 + 4z2

1 + z + 2z2 + 2z3 + 2z 2 11z 11z2 + 82 z2 (1 9z)
8(1 + 4)2 z4 4(1 + 2)(1 + 4)2 z4 ln D(z, )
(1 z)+
(1 z)+
1 + 4z2
(1)
L,Q
(z, ) =
s
2
(1 z)
CF B (1)
1 + 4

1 + 2
4
ln r +
2 Li2 r 2 + 4 Li2 (r)
2 + 4 ln
1 + 4
1 + 4

+ 3 ln2 (r) 4 ln r + 4 ln r ln(1 + 4) 2 ln r ln

s
1
z
+ CF B (z)
(1 z)2
(1 + 4z2 )3/2 [1 (1 )z]2

z 13 19z 2z2 + 8z3 22 z2 31 39z + 8z2
2z2 ln D(z, )
83 z3 (10 7z) 324 z4
[3 + 3z + 16z]
1 + 4z2
8(1 + 4)z4 4(1 + 2)(1 + 4)z4 ln D(z, )
(1 z)+
(1 z)+
1 + 4z2
(1)
A,Q (z, ) =
(23)

s
z(1 z)
1
CF B (z)
1 + 2(4 3z) + 82 z
2
3/2
2
[1 (1 )z]
(1 + 4z )
2 ln D(z, )
[2 + z + 4z] ,
+
1 + 4z2
(24)
(25)
264
z
s
1
(1)
I,Q (z, ) = CF B (z)
2
2
(1 + 4z ) [1 (1 )z]3/2
8 2

(1 z)(1 + 2z) 4z 10 10z z2 + 2z3

82 z2 25 29z + 8z2 963 z3 (3 2z) 1284 z4

+ 8 z[1 (1 )z] 1 z2 + z(13 11z) + 42 z2 (7 4z) + 163 z3 .
(26)
In Eqs. (23)(26), CF = (Nc2 1)/(2Nc ), where Nc is number of colors, while

1 + 4 1
1 + 2z 1 + 4z2
.
,
r = D(z = 1, ) =
D(z, ) =
2
1 + 4 + 1
1 + 2z + 1 + 4z
(27)
The so-called plus distributions are defined by

g(z) + = g(z) (1 z)
1
(28)
d g( ).
0
For any sufficiently regular test function h(z), Eq. (28) gives
1
a
k

1

ln (1 z)
lnk (1 z)
lnk+1 (1 a)
dz h(z)
= dz
h(z) h(1) + h(1)
.
1z
1z
k+1
+
(29)
To perform a numerical investigation of the inclusive partonic cross sections, k,Q (k =

(n,l)
T , L, A, I ), it is convenient to introduce the dimensionless coefficient functions ck,Q ,
k,Q , ,
2 (2 )
eQ

em s
4s
m2
n=0
n
n
l=0

(n,l)
ck,Q (, ) lnl

2
,
m2
(30)
where is a factorization scale (we use = F = R ) and the variable measures the distance
to the partonic threshold:
=
s
1z
1=
,
2
z
m
s = (q + kQ )2 .
(0,0)
(31)
(0,0)
Our analysis of the quantity cA,Q (, ) is given in Fig. 3. One can see that cA,Q is negative
at low Q2 (1 1) and positive at high Q2 (1 > 20). For the intermediate values of Q2 ,
(0,0)
cA,Q (, ) is an alternating function of .
(0,0)
Our results for the coefficient function cI,Q (, ) at several values of are presented in
(0,0)
Fig. 3. It is seen that cI,Q is negative at all values of and . Note also the threshold behavior
of the coefficient function:
(0,0)
+ O().
cI,Q ( 0, ) = 2 2 CF
(32)
1 + 4
(0,0)
This quantity takes its minimum value at m = 1/4: cI,Q ( = 0, m ) = 2 CF /(2 2 ).
(0,0)
265
(0,0)
Fig. 3. cA,Q (, ) and cI,Q (, ) coefficient functions at several values of .
2.2. Comparison with available results

For the first time, the NLO O(em s ) corrections to the -independent IC contribution have
been calculated a long time ago by Hoffmann and Moore (HM) [25]. However, authors of
Ref. [25] do not give explicitly their definition of the partonic cross sections that leads to a
confusion in interpretation of the original HM results. To clarify the situation, we need first to
derive the relation between the leptonquark DIS cross section, d lQ , and the partonic cross sections, (2) and (L) , used in [25]. Using Eqs. (C.1) and (C.5) in Ref. [25], one can express the
HM tensor R in terms of our cross sections 2,Q and L,Q defined by Eq. (16) in the present
paper. Comparing the obtained results with the corresponding definition of R via the HM cross
(2)
(L)
(given by Eqs. (C.16) and (C.17) in Ref. [25]), we find that
sections and

2,Q (z, ) B (z) 1 + 4z2 (2) (z, ),

2 B (z) (L)
(z, ) + 2z2 (2) (z, ) .
L,Q (z, )
1 + 4z2
(33)
(34)
Now we are able to compare our results with original HM ones. It is easy to see that the LO cross
sections (defined by Eqs. (37) in [25] and Eqs. (20) in our paper) obey both above identities.
(2)
(1)
Comparing with each other the quantities 1 and 2,Q (given by Eq. (51) in [25] and Eq. (23)
in this paper, respectively), we find that identity (33) is satisfied at NLO too. The situation with
longitudinal cross sections is more complicated. We have uncovered two misprints in the NLO
expression for (L) given by Eq. (52) in [25]. First, the r.h.s. of this equation must be multiplied
by z. Second, the sign in front of the last term (proportional to (1 z)) in Eq. (52) in Ref. [25]
must be changed.2 Taking into account these typos, we find that relation (34) holds at NLO as
well. So, our calculations of 2,Q and L,Q agree with the HM results.
Recently, the heavy quark initiated contributions to the -independent DIS structure functions,
F2 and FL , have been calculated by Kretzer and Schienbein (KS) [26]. The final KS results are
2 Note that this term originates from virtual corrections and the virtual part of the longitudinal cross section given by
Eq. (39) in Ref. [25] also has wrong sign. See Appendix A for more details.
266
expressed in terms of the parton level structure functions H 1 and H 2 . Using the definition of H 1
q
and H 2 given by Eqs. (7), (8) in Ref. [26], we obtain that
q
q
s B (z) H 1 ( , )
T ,Q (z, )
,
2 1 + 4 1 + 4z2

1 + 4 q
s
H ( , ),
2,Q (z, )
B (z)
2
1 + 4z2 2
(35)
where T ,Q = 2,Q L,Q and L,Q are defined by Eq. (16) in our paper and =
z(1 + 1 + 4 )/(1 + 1 + 4z2 ). To test identities (35), one needs only to rewrite the NLO
q
q
expressions for the functions H 1 ( , ) and H 2 ( , ) (given in Appendix C in Ref. [26]) in
terms of variables z and . Our analysis shows that relations (35) hold at both LO and NLO.
Hence we coincide with the KS predictions for the Q cross sections.
However, we disagree with the conclusion of Ref. [26] that there are errors in the NLO expression for (2) given in Ref. [25].3 As explained above, a correct interpretation of the quantities
(2) and (L) used in [25] leads to a complete agreement between the HM, KS and our results
for -independent cross sections.
As to the -dependent DIS, pQCD predictions for the Q cross sections A,Q (z, ) and
I,Q (z, ) in the case of arbitrary values of m2 and Q2 are not, to our knowledge, available in the literature. For this reason, we have performed several cross checks of our results
against well-known calculations in two limits: m2 0 and Q2 0. In particular, in the chiral limit, we reproduce the original results of Georgi and Politzer [32] and Mndez [33] for
I,Q (z, 0) and A,Q (z, 0). In the case of Q2 0, our predictions for 2,Q (s, Q2 0)
and A,Q (s, Q2 0) given by Eqs. (23), (25) reduce to the QED textbook results for the Compton scattering of polarized photons [34].
2.3. Photongluon fusion
The gluon fusion component of the semi-inclusive DIS is the following parton level interaction:
X ).
l() + g(kg ) l( q) + Q(pQ ) + X[Q](p
(36)
Corresponding leptongluon cross section, d lg , has the following decomposition in terms of the
helicity g cross sections:
d3 lg
em 1 y 2
=
2,g (z, ) (1 ) L,g (z, )
2
dz dQ d (2)2 zQ2 1

+ A,g (z, ) cos 2 + 2 (1 + ) I,g (z, ) cos ,
(37)
where the quantity is defined by Eq. (3) with y = (q kg )/( kg ).

At LO, O(em s ), the only gluon fusion subprocess responsible for heavy flavor production
is
).
(q) + g(kg ) Q(pQ ) + Q(p
Q
3 In detail, the KS point of view on the HM results is presented in PhD thesis [31], pp. 158160.
(38)
267
Fig. 4. The LO photongluon fusion diagrams.
The g cross sections, k,g (k = 2, L, A, I ), corresponding to the Born diagrams given in Fig. 4
have the form [35,36]:

1 + z

s
(0)
B (z) (1 z)2 + z2 + 4z(1 3z) 82 z2 ln
2,g (z, ) =
2
1 z

1 + 4z(1 z)( 2) z ,
(0)
2s
1 + z
+ (1 z)z ,
B (z)z 2z ln
1 z

1 + z
s
(0)
A,g (z, ) = B (z)z 2 1 2z(1 + ) ln
+ (1 2)(1 z)z ,
1 z
(0)
L,g (z, ) =
(0)
I,g
(z, ) = 0,
where B (z) is defined by Eq. (21) and the following notations are used:

m2
4z
Q2
,
= 2,
z = 1
.
z=
2q kg
1z
Q
(39)
(40)
symmetry
Note that the cos dependence vanishes in the GF mechanism due to the Q Q
which, at leading order, requires invariance under + [37].
(1)
(1)
As to the NLO results, presently, only -independent quantities 2,g
and L,g are known
exactly [38]. For this reason, we will use in our analysis the so-called soft-gluon approximation
for the NLO g cross sections (see Appendix B). As shown in Refs. [20,21,39], at energies
not so far from the production threshold, the soft-gluon radiation is the dominant perturbative
mechanism in the g interactions.
3. Hadron level results
3.1. Fixed flavor number scheme and nonperturbative intrinsic charm
In the fixed flavor number scheme,4 the wave function of the proton consists of light
quarks u, d, s and gluons g. Heavy flavor production in DIS is dominated by the gluon fusion
4 This approach is sometimes referred to as the fixed-order perturbation theory (FOPT).
268
mechanism. Corresponding hadron level cross sections, k,GF (x, ), have the form
1
k,GF (x, ) =
dz g(z, F ) k,g (x/z, , F )
(k = 2, L, A, I ),
(41)
= x(1 + 4),
(42)
where g(z, F ) describes gluon density in the proton evaluated at a factorization scale F . The
(0)
(k = 2, L, A, I ), are given by Eqs. (39). The NLO results,
lowest order GF cross sections, k,g
(1)
k,g , to the next-to-leading logarithmic accuracy are presented in Appendix B.

We neglect the q(q)
fusion subprocesses. This is justified as their contributions to heavy
quark leptoproduction vanish at LO and are small at NLO [38].
In the FFNS, the intrinsic heavy flavor component of the proton wave function is generated by
gg QQ fluctuations where the gluons are coupled to different valence quarks. In the present
paper, this component is referred to as the nonperturbative intrinsic charm (bottom). The prob is of higher twist since it scales
ability of the corresponding five-quark Fock state, |uudQQ,
as 2QCD /m2 [5]. However, since all of the quarks tend to travel coherently at same rapidity in
bound state, the heaviest constituents carry the largest momentum fraction. For
the |uudQQ
this reason, the heavy flavor distribution function has a more hard z-behavior than the light
parton densities. Since all of the densities vanish at z 1, the hardest PDF becomes dominant
at sufficiently large z independently of normalization.
Convolution of PDFs with partonic cross sections does not violate this observation. In particular, assuming a gluon density g(z) (1 z)n (where n = 35), we obtain that the LO GF
contribution to F2 scales as (1 )n+3/2 at 1, where is defined by Eq. (42). In the case
of Hoffman and Moore charm density (see below), the LO IC contribution is proportional to
(1 ) at 1. It is easy to see that, independently of normalizations, the IC contribution to
be dominate over the more soft GF component at large enough x.
For the first time, the intrinsic charm momentum distribution in the five-quark state |uudcc
was derived by Brodsky, Hoyer, Peterson and Sakai (BHPS) in the framework of a light-cone
model [1,2]. Neglecting the transverse motion of constituents, they have obtained in the heavy
quark limit that
c(z) =

N5 2
z 6z(1 + z) ln z + (1 z) 1 + 10z + z2 ,
6
(43)
1
where N5 = 36 corresponds to a 1% probability for IC in the nucleon: 0 c(z) dz = 0.01.
Hoffmann and Moore (HM) [25] incorporated mass effects in the BHPS approach. They first
introduced a mass scaling variable ,
1 + 1 + 4
2ax

, a=
,
=
(44)
2
1 + 1 + 4N x 2
where N = m2N /Q2 . To provide correct threshold behavior of the charm density, the constraint
< 1 was imposed where
=
2a x

,
1 + 1 + 4N x 2
x =
1
.
1 + 4 N
(45)
(1)
269
(0+1)
Fig. 5. The quantity (A,QS /2,QS )(x, ) in the FFNS with the HM [25] charm density (left panel) and in the VFNS
with the CTEQ5M charm distribution function (right panel).
Resulting charm distribution function, c(, ), has the following form in the HM approach:

c( ) c( ), ,
c(, ) =
(46)
0,
> ,
with c( ) defined by Eq. (43). Corresponding hadron level cross sections for the (c + c)
production, k,QS (x, ), due to the heavy quark scattering (QS) mechanism, are

k,QS (x, ) =
dz
1 + 4 2 /z2
c+ (z, ) k,c (/z, )
(k = 2, L, A, I ),
(47)
). The LO and NLO expressions for the

where the charm density c+ (z, ) c(z, ) + c(z,
partonic cross sections k,c (z, ) are given by Eqs. (20) and (23)(26), respectively.
Note also that, in the FFNS, the full cross section for the charm production, k (x, ), is simply
a sum of the GF and IC components:
k (x, ) = k,GF (x, ) + k,QS (x, )
(k = 2, L, A, I ).
(48)
Let us discuss the FFNS predictions for the hadron level asymmetry parameter A(x, Q2 ) de(1)
(0+1)
fined by Eq. (11). First we consider the ratio (A,QS /2,QS )(x, ), i.e., the mere IC contribution
to A(x, Q2 ). In Fig. 5, the x-behavior of this quantity at various values of is presented. One
(1)
(0+1)
can see that the ratio (A,QS /2,QS )(x, ) is negligibly small (of the order of 1%) practically at
all values of Q2 > m2 . Note that this fact is independent of the charm density we use (see, for
instance, the right panel in Fig. 5), but is only due to the smallness of the partonic cross section
(1)
A,c (z, ) [30]. So, the quantity A,QS (x, ) is exactly zero at LO5 and negligibly small at NLO.
This implies that the IC contribution has no practically cos 2-dependence and, for this reason,
we will neglect both A,c (z, ) and A,QS (x, ) cross sections in our further analysis.
Fig. 6 shows A(x, ) as a function of x for several values of variable : 1 = 1, 4, 10, 20, 50
and 100. We display both LO and NLO predictions of the GF mechanism as well as the analogous results of the combined GF + QS contribution. The azimuthal asymmetry due to the mere
LO GF component is given by solid line. The NLO GF predictions are plotted by dashed line.
5 Remember that the LO quantity (0) (z, ) vanishes for the kinematic reason, see Eqs. (20).
A,c
270

Fig. 6. Azimuthal asymmetry parameter A(x, ) in the FFNS at several values of in the case of 01 c(z) dz = 1%.
The following contributions are plotted: GF(LO) (solid lines), GF(LO) + kT -kick (dotted lines), GF(NLO) (dashed lines),
GF(LO) + QS(LO) (dash-dotted lines) and GF(NLO) + QS(NLO) (long-dashed lines).
The LO and NLO results of the total GF + QS contribution are given by dash-dotted and longdashed lines, respectively. In our calculations, the CTEQ5M [40] parametrization of the gluon
distribution function is used and a 1% probability
for IC in the nucleon is assumed. Through
out this paper, the value F = R = m2 + Q2 for both factorization and renormalization
scales is chosen. In accordance with the CTEQ5M parametrization, we use mc = 1.3 GeV and
4 = 326 MeV [40].
One can see from Fig. 6 the following basic features of the azimuthal asymmetry, A(x, ),
within the FFNS. First, as expected, the nonperturbative IC contribution is practically invisible at
low x, but affects essentially the GF predictions at large x. Since, contrary to the GF mechanism,
the QS component is practically cos 2-independent, the dominance of the IC contribution at
large x leads to a more rapid (in comparison with the GF predictions) decreasing of A(x, ) with
growth of x.
271
The most remarkable property of the azimuthal asymmetry is its perturbative stability. In
Refs. [20,21], the NLO soft-gluon corrections to the GF predictions for the cos 2 asymmetry
in heavy quark photo- and leptoproduction was calculated. It was shown that, contrary to the
production cross sections, the quantity A(x, ) is practically insensitive to soft radiation. One
can see from Fig. 6 in the present paper that the NLO corrections to the LO GF predictions
for A(x, ) are about few percent at not large x. This implies that large soft-gluon corrections
(LO)
(LO)
to A,GF
and 2,GF
(increasing both cross sections by a factor of two) cancel each other in the
(NLO)
(NLO)
ratio (A,GF /2,GF )(x, ) with a good accuracy. In terms of so-called K-factors, Kk (x, ) =
(NLO)
(LO)
(k
/k )(x, ) for k = 2, L, A, I , perturbative stability of the GF predictions for A(x, )
is provided by the fact that the corresponding K-factors are approximately the same at not large
x: KA,GF (x, ) K2,GF (x, ).
Comparing with each other the dash-dotted and long-dashed curves in Fig. 6, we see that the
NLO corrections to the combined GF + QS result for A(x, ) are also small. In this case, three
reasons are responsible for the closeness of the LO and NLO predictions. At small x, where the
nonperturbative IC contribution is negligible, perturbative stability of the asymmetry is provided
by the GF component. In the large-x region, where the IC mechanism dominates, the azimuthal
asymmetry rapidly vanishes with growth of x at both LO and NLO because the QS component
(1)
(0)
is practically cos 2-independent, A,c
(x, ) A,c (x, ) = 0.6 At intermediate values of x,
where both mechanisms are essential, perturbative stability of A(x, ) is due to the similarity of
the corresponding K-factors: K2,GF (x, ) K2,QS (x, ).7
Another remarkable property of the azimuthal asymmetry closely related to fast perturbative
convergence is its parametric stability.8 The analysis of Refs. [19,21] shows that the GF predictions for the cos 2 asymmetry are less sensitive to standard uncertainties in the QCD input
parameters (m, R , F , QCD and PDFs) than the corresponding ones for the production cross
sections. We have verified that the same situation takes also place for the combined GF + QS
results.
Let us discuss how the GF predictions for the azimuthal asymmetry are affected by nonperturbative contributions due to the intrinsic transverse motion of the gluon in the target. Because
of the relatively low c-quark mass, these contributions are especially important in the description
of the cross sections for charmed particle production.
To introduce kT degrees of freedom, kg p + kT , one extends the integral over the parton
distribution function in Eq. (41) to kT -space,
d g(, F ) d d2 kT f (kT )g(, F ).
(49)
The transverse momentum distribution, f (kT ), is usually taken to be a Gaussian:

f (kT ) =
2
ekT /kT
.
kT2
2
(50)
6 Although the ratio (A(NLO) /A(LO) )(x, ) is sizeable at sufficiently large x, the absolute values of the quantities
A(LO) (x, ) and A(NLO) (x, ) become so small that it seems reasonable to consider the asymmetry as equally negligible
at both LO and NLO and treat the predictions as perturbatively stable.
7 Note however that this similarity takes only place at intermediate values of x where both GF and QS components are
essential. In the low- and large-x regions, the factors K2,GF (x, ) and K2,QS (x, ) are strongly different.
8 Of course, parametric stability of the fixed order results does not imply a fast convergence of the corresponding series.
However, a fast convergent series must be parametrically stable. In particular, it must be R - and F -independent.
272

Fig. 7. The LO predictions for A(x, ) in the FFNS at several values of and Pc = 01 c(z) dz. Dash-dotted curves
describe the GF(LO) + QS(LO) contributions with Pc = 5%, 1%, 0.1% and 0.01%. Solid lines correspond to the case
when Pc = 0.
In practice, an analytic treatment of kT effects is usually used. According to Ref. [41], the kT smeared differential cross section of the process (1) is a two-dimensional convolution:

2
2
kick
d4 lN
d4 lN
ekT /kT
1
2
(
p

)
=
d
k
k
(51)
.
p

QT
T
QT
T
2
dx dQ2 dpQT d
kT2 dx dQ2 dpQT d
The factor 12 in front of kT in the r.h.s. of Eq. (51) reflects the fact that the heavy quark carries
away about one half of the initial energy in the reaction (1).
Values of the kT -kick corrections to the LO GF predictions for the cos 2 asymmetry in the
charm production are shown in Fig. 6 by dotted curves. Calculating the kT -kick effect we use
kT2 = 0.5 GeV2 . One can see that kT -smearing for A(x, Q2 ) is about 2025% in the region of
low Q2 m2 and rapidly decreases at high Q2 .
In Fig. 7, the dependence of the asymmetry A(x, ) on the nonperturbative intrinsic charm
content of the proton is presented. We plot the LO predictions for A(x, ) as a function of x for
273
1
several values of the variable and quantity Pc = 0 c(z) dz describing a probability for IC in
the nucleon. Dash-dotted curves describe the GF(LO) + QS(LO) contributions with Pc = 5%, 1%,
0.1% and 0.01%. Solid lines correspond to the case when Pc = 0. Comparing with each other
Figs. 6 and 7, one can see that even a 0.1% contribution of the nonperturbative IC to the proton
wave function could be extracted from the cos 2 asymmetry at large enough Bjorken x.
3.2. Variable flavor number scheme and perturbative intrinsic charm
(0)
One can see from Eqs. (39) that the GF cross section 2,g (z, ) contains potentially large
(1)
logarithm, ln(Q2 /m2 ). The same situation takes also place for the QS cross section 2,Q (z, )
given by Eq. (23). At high energies, Q2 , the terms of the form s ln(Q2 /m2 ) dominate
the production cross sections. To improve the convergence of the perturbative series at high
energies, the so-called variable flavor number schemes (VFNS) have been proposed. Originally,
this approach was formulated by Aivazis, Collins, Olness and Tung (ACOT) [6,42].
In the VFNS, the mass logarithms of the type sn lnn (Q2 /m2 ) are resummed via the renormalization group equations. In practice, the resummation procedure consists of two steps. First,
the mass logarithms have to be subtracted from the fixed order predictions for the partonic cross
sections in such a way that in the asymptotic limit Q2 the well-known massless MS coefficient functions are recovered. Instead, a charm parton density in the hadron, c(x, Q2 ), has to
be introduced. This density obeys the usual massless NLO DGLAP evolution equation with the
boundary condition c(x, Q2 = Q20 ) = 0 where Q20 m2 . So, we may say that, within the VFNS,
the charm density arises perturbatively from the g cc evolution.
In the VFNS, the treatment of the charm depends on the values chosen for Q2 . At low
2
Q < Q20 , the production cross sections are described by the light parton contributions (u, d, s
and g). The charm production is dominated by the GF process and its higher order QCD corrections. At high Q2 m2 , the charm is treated in the same way as the other light quarks and
it is represented by a charm parton density in the hadron, which evolves in Q2 . In the intermediate scale region, Q2 m2 , one has to make a smooth connection between the two different
prescriptions.
Strictly speaking, the perturbative charm density is well defined at high Q2 m2 but does
not have a clean interpretation at low Q2 . Since the perturbative IC originates from resummation
of the mass logarithms of the type sn lnn (Q2 /m2 ), it is usually assumed that the corresponding
PDF vanishes with these logarithms, i.e. for Q2 < Q20 m2 . On the other hand, the threshold
constraint W 2 = (q + p)2 = Q2 (1/x 1) > 4m2 implies that Q0 is not a constant but live
function of x. To avoid this problem, several solutions have been proposed (see e.g. Refs. [43,
45]). In this paper, we use the so-called ACOT( ) prescription [43] which guarantees (at least at
Q2 > m2 ) the correct threshold behavior of the heavy-quark-initiated contributions.
Within the VFNS, the -independent charm production cross sections have three pieces:
2 (x, ) = 2,GF (x, ) 2,SUB (x, ) + 2,QS (x, ),
(52)
where the first and third terms on the right-hand side describe the usual (unsubtracted) GF and
QS contributions while the second (subtraction) term renders the total result infra-red safe in
the limit m2 0. The only constraint imposed on the subtraction term is to reproduce at high
energies the familiar MS partonic cross section:

MS
(z).
lim 2,g (z, ) 2,SUB (z, ) = 2,g
(53)
0
274
Evidently, there is some freedom in the choice of finite mass terms of the form n (with a
positive n) in 2,SUB (z, ). For this reason, several prescriptions have been proposed to fix the
subtraction term. As mentioned above, we use the so-called ACOT( ) scheme [43].
According to the ACOT( ) prescription, the lowest order -independent cross section is
(LO)
2 (x, ) =

s 2F
(0)
(0)
dz g(z, F ) 2,g (x/z, )
(/z)
ln 2 B (x/z)Pgc
+ B (x)c+ (, F ),
(54)
(0)
(0)
where Pgc is the LO gluonquark splitting function, Pgc ( ) = [(1 )2 + 2 ]/2, and the LO
(0)
GF cross section 2,g is given by Eqs. (39). Remember also that = x(1 + 4) and c+ (, F ) =
c(, F ) + c(,
F ).
The asymptotic behavior of the subtraction terms is fixed by the parton level factorization
theorem. This theorem implies that the partonic cross sections d can be factorized into processdependent infra-red safe hard scattering cross sections d , which are finite in the limit m 0,
and universal (process-independent) partonic PDFs fai and fragmentation functions dnQ :

fai ( ) d ( + i n + X) dnQ (z).
d ( + a Q + X) =
(55)
i,n
In Eq. (55), the symbol denotes the usual convolution integral, the indices a, i, n and Q denote
partons, pi = pa and pQ = zpn . All the logarithms of the heavy-quark mass (i.e., the singularities in the limit m 0) are contained in the PDFs fai and fragmentation functions dnQ
while d are IR-safe (i.e., are free of the ln m2 terms). The expansion of Eq. (55) can be used to
determine order by order the subtraction terms. In particular, for the LO GF contribution to the
charm leptoproduction one finds [6]

(0)
(0)
(1)
, ln 2F /m2 k,QS (z/ ) (k = 2, L, A, I ),
k,SUB z, ln 2F /m2 = fgc
(56)
(1)
(0)
where fgc (, ln(2F /m2 )) = (s /2) ln(2F /m2 )Pgc ( ) describes the charm distribution in
the gluon within the MS factorization scheme.
One can see from Eq. (56) that the azimuth-dependent GF cross sections A,GF and I,GF do
not have subtraction terms at LO because the lowest order QS contribution is -independent. For
this reason, the cos 2-dependence within the VFNS has the same form as in the FFNS:
A(LO) (x, ) =
(0)
dz g(z, F ) A,g
(x/z, ).
(57)
Fig. 8 shows the ACOT( ) predictions for the asymmetry parameter A(x, ) at several values
of variable : 1 = 1, 4, 10, 20, 50 and 100. For comparison, we plot also the LO GF predictions
(solid curves). In the ACOT( ) case, we consider the CTEQ5M (dotted lines) and CTEQ4M
(dashed curves) parametrizations of the gluon and charm densities in the proton. Corresponding
values of the charm quark mass are mc = 1.3 GeV [40] (for the CTEQ5M PDFs) and mc =
1.6
GeV [44] (for the CTEQ4M PDFs). The default value of the factorization scale is F =
m2 + Q2 .
One can see from Fig. 8 the following properties of the azimuthal asymmetry, A(x, ), within
the VFNS. Contrary to the nonperturbative IC component, the perturbative one is significant
275
Fig. 8. Azimuthal asymmetry parameter A(x, ) in the VFNS at several values of . The following contributions
are plotted: GF(LO) (solid curves), GF(LO) SUB(LO) + QS(LO) with the CTEQ5M set of PDFs (dotted curves) and
GF(LO) SUB(LO) + QS(LO) with the CTEQ4M set of PDFs (dashed curves).
practically at all values of Bjorken x and Q2 > m2 . The perturbative charm contribution leads to
a sizeable decreasing of the GF predictions for the cos 2-asymmetry. In the ACOT( ) scheme,
the IC contribution reduces the GF results for A(x, ) by about 30%. The origin of this reduction
is straightforward: the QS component is practically cos 2-independent.
The ACOT( ) predictions for the asymmetry depend weakly on the parton distribution functions we use. It is seen from Fig. 8 that the CTEQ5M and CTEQ4M sets of PDFs lead to very
similar results for A(x, ). Note that we give the CTEQ5M predictions at low x only because of
irregularities in the CTEQ5M charm density at large x.
We have also analyzed how the VFNS predictions depend on the choice of subtraction prescription. In particular, the schemes proposed in Refs. [26,45] have been considered. We find
that, sufficiently above the production threshold, these subtraction prescriptions reduce the GF
results for the asymmetry by approximately 3050%.
One can conclude that impact of the perturbative IC on the cos 2 asymmetry is essential in
the whole region of Bjorken x and therefore can be tested experimentally.
276
4. Conclusion
In the present paper, we consider the azimuthal dependence in charm leptoproduction as a
probe of the IC content of the proton. Our analysis is based on the fact that the GF and QS components have strongly different cos 2-distributions. This fact follows from the NLO calculations
of both parton level contributions. In the framework of the FFNS, we justify the most remarkable
property of the hadron level azimuthal cos 2 asymmetry: The combined GF+QS predictions
for A(x, Q2 ) are perturbatively and parametrically stable. The nonperturbative IC contribution
(resulting from the five-quark |uudcc
component of the proton wave function) is practically
invisible at low x, but affects essentially the GF predictions for the asymmetry at large Bjorken
x. We conclude that measurements of the cos 2 asymmetry at large x could directly probe the
nonperturbative intrinsic charm.
Within the VFNS, charm density originates perturbatively from the g cc process and obeys
the DGLAP evolution equation. Presently, charm densities are included practically in all the
global sets of PDFs like CTEQ and MRST. Our analysis shows that these charm distribution
functions reduce dramatically (by about 1/3) the GF predictions for A(x, Q2 ) practically at all
values of x. For this reason, the perturbative IC contribution can easily be measured using the
azimuthal cos 2-distributions in charm leptoproduction.
The VFN schemes have been proposed to resum the mass logarithms of the form
sn lnn (Q2 /m2 ) which dominate the production cross sections at high energies. Evidently, were
the calculation done to all orders of s , the VFNS and FFNS (without nonperturbative IC) would
be exactly equivalent. There is a point of view advocated in Refs. [6,7] that, at high energies,
the perturbative series converge better within the VFNS than in the FFNS. There is also another
opinion [46,47] that the above logarithms do not vitiate the convergence of the perturbation
expansion so that a resummation is, in principle, not necessary. Our analysis of the azimuthal
dependence in leptoproduction indicates an experimental way to resolve this problem. First,
contrary to the production cross sections, the azimuthal cos 2-asymmetry is well defined numerically in pQCD. Second, sufficiently above the production threshold (i.e., at small enough
Bjorken x), the LO VFNS predictions for the cos 2-asymmetry differ by more than 30% from
the corresponding FFNS ones. Third, nonperturbative contributions (like the intrinsic gluon motion in the target) cannot compensate for this difference at non-small Q2 where the VFNS is
expected to be adequate. Therefore measurements of the azimuthal distributions in charm leptoproduction would make it possible to clarify the question whether the VFNS perturbative series
for A(x, Q2 ) converges better than the FFNS one.
Acknowledgements
We thank S.J. Brodsky for stimulating discussions and useful suggestions. We also would like
to acknowledge interesting correspondence with I. Schienbein. This work was supported in part
by the ANSEF grant 04-PS-hepth-813-98 and NFSAT grant GRSP-16/06.
Appendix A. Virtual and soft contributions to the quark scattering
In this appendix we reproduce some results of Hoffmann and Moore for the -independent
QS cross sections, and correct two misprints uncovered in Ref. [25]. We work in four dimensions, in the Feynman gauge and use the on-mass-shell renormalization scheme. We compute the
277
absorptive part of the Feynman diagram (which is free of the UV divergences) and then restore
the real part using the appropriate dispersion relations.
In the on-mass-shell scheme, the renormalized fermion self-energy vanishes like (p Q m)2
which means that the second and third diagrams in Fig. 2c do not contribute to the cross section
when the external quark legs are on-shell, p Q m. The first graph in Fig. 2c describes the NLO
corrections to the quark-photon vertex function:

g(Q2 )
(q) = f Q2
q ,
2m
(A.1)
where = 12 ( ) while f (Q2 ) and g(Q2 ) are the quark electromagnetic formfac(0)
tors. At the lowest order = .
V , is obtained from the interference term between
The virtual leptonquark cross section, lQ
the virtual and the Born amplitude. The result can be written in terms of the electromagnetic
formfactors as:
V
d2 lQ
dz dQ2

em B (z)
(1 z) 1 + (1 y)2 2z2 y 2 f Q2 + y 2 g Q2 .
zQ2
(A.2)
Taking into account the definition of the HM cross sections, (2) and (L) , given by Eqs. (33),
(34) and (16), we find that corresponding virtual parts are:

(2)
(L)
1V
(A.3)
z, Q2 = 2(1 z)f (1) Q2 ,
1V z, Q2 = (1 z)g (1) Q2 ,
where f (1) (Q2 ) and g (1) (Q2 ) are the NLO corrections to the electromagnetic formfactors. For
(2)
(L)
and 1V
, we use exactly the same notations as in Ref. [25].
the NLO HM cross sections, 1V
In the on-mass-shell renormalization scheme, the renormalized vertex correction vanishes
as the photon virtuality goes to zero, f (1) (0) = 0. This is a consequence of the Ward identity
and the fact that the real photon field (like the massive fermion one) is unrenormalized in first
order QCD. To satisfy the condition f (1) (0) = 0 automatically, we should use for f (1) (q 2 ) the
dispersion relation with one subtraction. The second formfactor, g (1) (q 2 ), has no singularities.
For this reason, we use for g (1) (q 2 ) the dispersion relation without subtractions:
f
(1)
2 q 2
q =

4m2
dt Im f (1) (t)
,
t (t q 2 i0)
(1)
1
=

4m2
dt Im g (1) (t)
.
t q 2 i0
(A.4)
Calculating the imaginary parts of the formafactors and restoring their real parts with the help of
Eqs. (A.4) yields

s
m
1 + 2
f (1) Q2 = CF 1 +
ln r ln
1
mg
1 + 4

1
1 2
1 + 4
1 ln r
1 + 2
Li2 (r) +
,
+ ln r + ln r ln
+
+
12 4
2
4 1 + 4
1 + 4
(A.5)
2
ln r
s
(1)
.
g Q = CF
(A.6)
1 + 4
In Eqs. (A.5) and (A.6), CF = (Nc2 1)/(2Nc ), where Nc is number of colors, and r is defined
(1)
by Eq. (27). Taking into account that (1) = (s /em )CF QED , we see that Eqs. (A.5), (A.6)
reproduce the textbook QED results.
278
It is now straightforward to obtain the virtual contribution to the longitudinal cross section.
Combining Eqs. (A.3) and (A.6) yields:
s
ln r
(L)
.
1V z, Q2 = CF (1 z)
1 + 4
(A.7)
Comparing the above result with the corresponding one given by Eq. (39) in Ref. [25], we see
(L)
that the HM expression for 1V has opposite sign. Note also that this typo propagates into the
(L)
final result for 1 given by Eq. (52) [25].
(L)
(z, Q2 ),
Calculation of the bremsstrahlung contribution to the longitudinal cross section, 1B
(L)
is also straightforward. We coincide with the HM result for 1B (z, Q2 ) given by Eq. (49) in
(L)
Ref. [25]. However there is one more misprint in the HM expression for 1 : The r.h.s. of
Eq. (52) [25] should be multiplied by z.
(2)
In the case of 1 (z, Q2 ), the situation is slightly more complicated due to the need to take
into account the IR singularities. One can see from Eq. (A.5) that f (1) (Q2 ) has an IR divergence
which is regularized with the help of an infinitesimal gluon mass mg . This singularity is cancelled when one adds the so-called soft contribution originating from the real gluon emission.
For this purpose we introduce another infinitesimal parameter z, (mg /m) z 1. The full
(2)
bremsstrahlung contribution, 1B
, can then be splitted into the soft and hard pieces as follows:

(2)
(2)
1soft z, Q2 = (z + z 1)1B z, Q2 ,

(2)
(2)
1hard z, Q2 = (1 z z)1B z, Q2 ,
(A.8)
where (1 z z) is the Heaviside step function. The soft cross section should be calculated
in the eikonal approximation, pg 0, taking into account the infinitesimal gluon mass mg . As
a result, the sum of the virtual and soft contributions is IR finite:
(2)
(2)
1V + 1soft

1 + 4
s
1 + 2
ln r + 2 ln 1
= CF (1 z) 2 ln(z) 1 +
ln r
2
1 + 4

3
1 + 2
Li2 r 2 + 2 Li2 (r) + ln2 r 2 ln r ln r ln + 2 ln r ln(1 + 4) .
+
2
1 + 4
(A.9)
(2)
Adding to the above expression the hard cross section 1hard defined by Eq. (A.8), we reproduce
(2)
in the limit z 0 the full result for 1 given by Eq. (51) in Ref. [25].
Appendix B. NLO soft-gluon corrections to the photongluon fusion
This appendix provides an overview of the NLO soft-gluon approximation for the photon
gluon fusion mechanism. We present the final results for the parton level cross sections to the
next-to-leading logarithmic (NLL) accuracy. More details can be found in Refs. [20,21,39].
To take into account the NLO contributions to the GF mechanism, one needs to calculate the
virtual O(em s2 ) corrections to the Born process (38) and the real gluon emission:
) + g(pg ).
(q) + g(kg ) Q(pQ ) + Q(p
Q
(B.1)
279
The partonic invariants describing the single-particle inclusive (1PI) kinematics are
s = 2q kg = s + Q2 = S ,
t1 = (kg pQ )2 m2 = T1 ,
s4 = s + t1 + u1 ,
u1 = (q pQ )2 m2 = U1 ,
(B.2)
where is defined by kg = p and s4 measures the inelasticity of the reaction (B.1). The corresponding 1PI hadron level variables describing the reaction (1) are
S = 2q p = S + Q2 ,
T1 = (p pQ )2 m2 ,
S4 = S + T1 + U1 ,
U1 = (q pQ )2 m2 .
(B.3)
The exact NLO calculations of the unpolarized heavy quark production in g [48,49], g
[38], and gg [5053] collisions show that, near the partonic threshold, a strong logarithmic enhancement of the cross sections takes place in the collinear, pg,T 0, and soft, pg 0, limits.
This threshold (or soft-gluon) enhancement has universal nature in the perturbation theory and
originates from incomplete cancellation of the soft and collinear singularities between the loop
and the bremsstrahlung contributions. Large leading and next-to-leading threshold logarithms
can be resummed to all orders of perturbative expansion using the appropriate evolution equations [5456]. The analytic results for the resummed cross sections are ill-defined due to the
Landau pole in the coupling strength s . However, if one considers the obtained expressions as
generating functionals of the perturbative theory and re-expands them at fixed order in s , no
divergences associated with the Landau pole are encountered.
Soft-gluon resummation for the photongluon fusion has been performed in Ref. [39] and
checked in Refs. [20,21]. To NLL accuracy, the perturbative expansion for the partonic cross
sections, d2 k,g /dt1 du1 (k = T , L, A, I ), can be written in a factorized form as
s 2
d2 k,g
(s , t1 , u1 )
dt1 du1
Born
= Bk,g
(s , t1 , u1 )
(s + t1 + u1 ) +

s CA n
n=1
with the Born level distributions
Born
Bk,g

K
(n)
(s , t1 , u1 ) ,
(B.4)
given by

2

u1
s
m2 s
t1
s (m Q2 /2) Q2
+
+4
+
,
u1
t1
s
t1 u1
t1 u1
s
2

m2 s
8Q s
Born
2
(s , t1 , u1 ) = eQ
em s
,
BL,g
s
s
t1 u1

2

m2 s
Q2
s
m s
Born
2
(s , t1 , u1 ) = eQ
em s 4
+
,
BA,g
s
t1 u1
t1 u1
s

1/2

t1 u1
t1 u1 s
2Q2 2m2 s
Born
2
2
2
m
1
.
BI,g (s , t1 , u1 ) = eQ em s 4 Q
t1 u1
s
t1 u1
s 2
(B.5)

2
BTBorn
,g (s , t1 , u1 ) = eQ em s
Note that the functions K (n) (s , t1 , u1 ) in Eq. (B.4) originate from the collinear and soft limits. Radiation of soft and collinear gluons does not affect the transverse momentum of detected
particles and therefore the azimuthal angle . For this reason, the functions K (n) (s , t1 , u1 ) are
the same for all helicity cross sections k,g (k = T , L, A, I ). At NLO, the soft-gluon corrections
280
to NLL accuracy in the MS scheme are

K (1) (s , t1 , u1 )

2

ln(s4 /m2 )
1
u1
2CF
=2
1 + ln
1
(1 + Re L ) + ln
2
s4
s
t
C
m
4 +
1
A
+

2
u1
+ (s4 ) ln
(B.6)
ln
,
2
m
m2
where we use = F = R . In Eq. (B.6), CA = Nc and CF = (Nc2 1)/(2Nc ),where Nc is
number of colors, while L = (1 2m2 /s){ln[(1 )/(1 + )] + i} with = 1 4m2 /s.
The single-particle inclusive plus distributions are defined by

lnl (s4 /m2 )

s4

1
lnl (s4 /m2 )

l+1
= lim
(s4 ) +
ln
(s4 ) .
0
s4
l+1
m2
+

(B.7)
For any sufficiently regular test function h(s4 ), Eq. (B.7) gives
max
s4

l

ln (s4 /m2 )
ds4 h(s4 )
s4
+
max
s4

=
0

lnl (s4 /m2 )
1
ds4 h(s4 ) h(0)
+
h(0) lnl+1 s4max /m2 .
s4
l+1
(B.8)
In Eq. (B.6), we have preserved the NLL terms for the scale-dependent logarithms too. Note
also that the results (B.5) and (B.6) agree to NLL accuracy with the exact O(em s2 ) calculations
of the photongluon cross sections T ,g and L,g given in Ref. [38].
To investigate the scale dependence of the results (B.4)(B.6), it is convenient to introduce
for the fully inclusive (integrated over t1 and u1 ) cross sections, k,g (k = T , L, A, I ), the dimen(n,l)
sionless coefficient functions ck,g
defined by Eq. (30). Concerning the NLO scale-independent
(1,0)
(1,0)
coefficient functions, only cT ,g and cL,g are known exactly [38,57]. As to the -dependent
coefficients, they can by calculated explicitly using the evolution equation:
d k,g (z, Q2 , 2 )
=
d ln 2

d k,g z/, Q2 , 2 Pgg ( ),
(B.9)
min
where z = Q2 /s , min = z(1 + 4), k,g (z, Q2 , ) are the cross sections resummed to all orders in s and Pgg ( ) is the corresponding (resummed) AltarelliParisi gluongluon splitting
function. Expanding Eq. (B.9) in s , one can find [20,39]
(1,1)
ck,g (z, ) =
1
4 2
1
min

(0,0)
(0)
d b2 (1 ) Pgg
( ) ck,g (z/, ),
(B.10)
281
where b2 = (11CA 2nf )/12 is the first coefficient of the (s )-function expansion and nf is
the number of active quark flavors. The one-loop gluon splitting function is:

1
(0)
( ) = lim
+
+ (1 ) (1 ) + (1 ) ln CA
Pgg
0
1
+ b2 (1 ).
(B.11)
With Eq. (B.10) in hand, it is possible to check the quality of the NLL approximation against
exact answers. As shown in Ref. [21], the soft-gluon corrections reproduce satisfactorily the
threshold behavior of the available exact results for 1. Since the gluon distribution function supports just the threshold region, the soft-gluon contribution dominates the photonhadron
cross sections k,GF (k = T , L, A, I ) at energies not so far from the production threshold and at
relatively low virtuality Q2 m2 .
Appendix C. Nonperturbative IC and relevant experimental facts
The most clean probe of the charm quark distribution function (both perturbative and nonperturbative) is the semi-inclusive deep inelastic lepton-proton scattering, lp l cX. To measure
the nonperturbative IC contribution, one needs data on the charm production at sufficiently large
Bjorken x. The only experiment which has investigated the large x domain is the European Muon
Collaboration (EMC) [58] where the decay lepton spectra have been used to detect the produced
charmed particles. In Ref. [59], a re-analysis of the EMC data on F2c (x, Q2 ) have been performed
using the NLO results for both GF and QS components. The analysis [59] shows that a nonperturbative intrinsic charm contribution to the proton wave function of the order of 1% is needed
to fit the EMC data in the large x region. This value of the nonperturbative IC is consistent with
the estimates based on the operator product expansion [5]. Note however that the EMC data are
of limited statistics and, for this reason, more accurate measurements of charm leptoproduction
at large x are necessary.
It is also possible to extract useful information on the IC from diffractive dissociation
processes such as p pJ / on a nuclear target. Comprehensive measurements of the pA
J /X and A J /X cross sections have been performed in the fixed target experiments
NA3 at CERN [60] and E886 at FNAL [61]. According to the arguments presented in Refs. [62
64], the IC contribution is predicted to be strongly shadowed in the above reactions that is in a
complete agreement with the observed nuclear dependence of the high Feynman xF component
of the J / hadroproduction.
A non-vanishing five-quark Fock component |uudcc
leads to the production of open charm
states such as c (cud) and D (cd)
with large Feynman xF . This may occur either through
a coalescence of the valence and charm quarks which are moving with the same rapidity or
via hadronization of the produced c and c.
As shown in Refs. [65,66], a model based on the
nonperturbative intrinsic charm naturally explains the leading particle effect in the pp DX
and pp c X processes that has been observed at the ISR [67] and Fermilab [68,69].
As to the high-xF hadroproduction of open bottom states like b (bud), corresponding cross
sections are predicted to be suppressed as m2c /m2b 1/10 in comparison with the case of charm
production. Evidence for the forward b production in the pp collisions at the ISR energy was
reported in Refs. [70,71].
Rare seven-quark fluctuations of the type |uudccc
c
in the proton wave function can lead to
the production of two J / [72] or a double-charm baryon state at large xF and low pT . Double
J / events with a high combined xF 0.5 have been detected in the NA3 experiment [73]. An
282
+ (3520) with mean x 0.33 has been reported

observation of the double-charmed baryon cc
F
by the SELEX Collaboration at FNAL [74].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
S.J. Brodsky, P. Hoyer, C. Peterson, N. Sakai, Phys. Lett. B 93 (1980) 451.

S.J. Brodsky, C. Peterson, N. Sakai, Phys. Rev. D 23 (1981) 2745.
S.J. Brodsky, Light-front QCD, hep-ph/0412101.
S.J. Brodsky, Few-Body Systems 36 (2005) 35.
M. Franz, V. Polyakov, K. Goeke, Phys. Rev. D 62 (2000) 074024.
M.A.G. Aivazis, J.C. Collins, F.I. Olness, W.-K. Tung, Phys. Rev. D 50 (1994) 3102.
J.C. Collins, Phys. Rev. D 58 (1998) 094002.
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 438.
Y.L. Dokshitzer, Sov. Phys. JETP 46 (1977) 641.
G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
J. Pumplin, Phys. Rev. D 73 (2006) 114015.
S.J. Brodsky, B. Kopeliovich, I. Schmidt, J. Soffer, Phys. Rev. D 73 (2006) 113005.
J. Pumplin, D.R. Stump, J. Huston, H.L. Lai, P. Nadolsky, W.K. Tung, JHEP 0207 (2002) 012.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 604 (2004) 61.
N. Kidonakis, Phys. Rev. D 73 (2006) 034001.
N. Kidonakis, Phys. Rev. D 64 (2001) 014009.
M.L. Mangano, P. Nason, G. Ridolfi, Nucl. Phys. B 373 (1992) 295.
S. Frixione, M.L. Mangano, P. Nason, G. Ridolfi, Nucl. Phys. B 412 (1994) 225.
N.Ya. Ivanov, A. Capella, A.B. Kaidalov, Nucl. Phys. B 586 (2000) 382.
N.Ya. Ivanov, Nucl. Phys. B 615 (2001) 266.
N.Ya. Ivanov, Nucl. Phys. B 666 (2003) 88.
N.Ya. Ivanov, P.E. Bosted, K. Griffioen, S.E. Rock, Nucl. Phys. B 650 (2003) 271.
SLAC E161, http://www.slac.stanford.edu/exp/e160, 2000.
N. Dombey, Rev. Mod. Phys. 41 (1969) 236.
E. Hoffman, R. Moore, Z. Phys. C 20 (1983) 71.
S. Kretzer, I. Schienbein, Phys. Rev. D 58 (1998) 094035.
A. Deshpande, R. Milner, R. Venugopalan, W. Vogelsang, Annu. Rev. Nucl. Part. Sci. 55 (2005) 165.
See also http://www.bnl.gov/eic for information concernig the eRHIC/EIC project.
J.B. Dainton, M. Klein, P. Newman, E. Perez, F. Willeke, hep-ex/0603016.
L.N. Ananikyan, N.Ya. Ivanov, hep-ph/0609074.
I. Schienbein, hep-ph/0110292.
H. Georgi, H.D. Politzer, Phys. Rev. Lett. 40 (1978) 3.
A. Mndez, Nucl. Phys. B 145 (1978) 199.
U. Fano, Phys. Rev. 93 (1954) 121.
J.P. Leveille, T. Weiler, Phys. Rev. D 24 (1981) 1789.
A.D. Watson, Z. Phys. C 12 (1982) 123.
J.P. Leveille, T. Weiler, Nucl. Phys. B 147 (1979) 147.
E. Laenen, S. Riemersma, J. Smith, W.L. van Neerven, Nucl. Phys. B 392 (1993) 162.
E. Laenen, S.-O. Moch, Phys. Rev. D 59 (1999) 034027.
H.L. Lai, et al., Eur. Phys. J. C 12 (2000) 375.
L. Apanasevich, et al., Phys. Rev. D 59 (1999) 074007.
M.A.G. Aivazis, F.I. Olness, W.-K. Tung, Phys. Rev. D 50 (1994) 3085.
W.-K. Tung, S. Kretzer, C. Schmidt, J. Phys. G 28 (2002) 983.
H.L. Lai, et al., Phys. Rev. D 55 (1997) 1280.
M. Kramer, F.I. Olness, D.E. Soper, Phys. Rev. D 62 (2000) 096007.
W.L. van Neerven, hep-ph/0107193.
M. Buza, Y. Matiounine, J. Smith, W.L. van Neerven, Eur. Phys. J. C 1 (1998) 301.
R.K. Ellis, P. Nason, Nucl. Phys. B 312 (1989) 551.
J. Smith, W.L. van Neerven, Nucl. Phys. B 374 (1992) 36.
W. Beenakker, H. Kuijf, W.L. van Neerven, J. Smith, Phys. Rev. D 40 (1989) 54.
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
P. Nason, S. Dawson, R.K. Ellis, Nucl. Phys. B 303 (1988) 607.

H. Contopanagos, E. Laenen, G. Sterman, Nucl. Phys. B 484 (1997) 303.
E. Laenen, G. Oderda, G. Sterman, Phys. Lett. B 438 (1998) 173.
N. Kidonakis, G. Oderda, G. Sterman, Nucl. Phys. B 531 (1998) 365.
B.W. Harris, J. Smith, Nucl. Phys. B 452 (1995) 109.
J.J. Aubert, et al., Nucl. Phys. B 213 (1983) 31.
B.W. Harris, J. Smith, R. Vogt, Nucl. Phys. B 461 (1996) 181.
J. Badier, et al., Z. Phys. C 20 (1983) 101.
M.J. Leitch, et al., Phys. Rev. Lett. 84 (2000) 3256.
S.J. Brodsky, P. Hoyer, Phys. Rev. Lett. 63 (1989) 1566.
P. Hoyer, M. Vanttinen, U. Sukhatme, Phys. Lett. B 246 (1990) 217.
S.J. Brodsky, P. Hoyer, A.H. Mueller, W.K. Tang, Nucl. Phys. B 369 (1992) 519.
V.D. Barger, F. Halzen, W.Y. Keung, Phys. Rev. D 25 (1982) 112.
R. Vogt, S.J. Brodsky, Nucl. Phys. B 478 (1996) 311.
P. Chauvat, et al., Phys. Lett. B 199 (1987) 304.
E.M. Aitala, et al., Phys. Lett. B 495 (2000) 42.
E.M. Aitala, et al., Phys. Lett. B 539 (2002) 218.
M. Basile, et al., Nuovo Cimento A 65 (1981) 408.
G. Bari, et al., Nuovo Cimento A 104 (1991) 1787.
R. Vogt, S.J. Brodsky, Phys. Lett. B 349 (1995) 569.
J. Badier, et al., Phys. Lett. B 114 (1982) 457.
A. Ocherashvili, et al., Phys. Lett. B 628 (2005) 18.
283
Nuclear Physics B 762 [PM] (2007) 285300
Partition functions of reduced matrix models

with classical gauge groups
H. Itoyama a,b , H. Kihara b, , R. Yoshioka a
a Department of Mathematics and Physics, Graduate School of Science, Osaka City University,
3-3-138, Sugimoto, Sumiyoshi-ku, Osaka 558-8585, Japan

b Osaka City University, Advanced Mathematical Institute (OCAMI),
3-3-138 Sugimoto, Sumiyoshi, Osaka 558-8585, Japan

Received 15 September 2006; received in revised form 18 October 2006; accepted 2 November 2006
Abstract
We evaluate partition functions of matrix models which are given by topologically twisted and dimensionally reduced actions of d = 4 N = 1 super-YangMills theories with classical (semi-)simple gauge
groups, SO(2N ), SO(2N + 1) and USp(2N ). The integrals reduce to those over the maximal tori by semiclassical approximation which is exact in reduced models. We carry out residue calculus by developing a
diagrammatic method, in which the action of the Weyl groups and therefore counting of multiplicities are
explained obviously.
1. Introduction
In 1996 Banks, Fischler, Shenker and Susskind (abbreviated BFSS) suggested the equivalence between 11-dimensional M-theory and the N limit of the supersymmetric matrix
quantum mechanics describing D0-branes [1]. The action of their model is obtained by the reduction of d = 10 N = 1 super-YangMills theory with gauge group SU(N ) [2]. Ishibashi,
Kawai, Kitazawa and Tsuchiya (IKKT) proposed a zero-dimensional matrix model with manifest ten-dimensional N = 2 super-Poincar invariance [3]. The action of their model is given by
reduction to zero dimension of the N = 1, d = 10 super-YangMills action with gauge group
E-mail addresses: itoyama@sci.osaka-cu.ac.jp (H. Itoyama), kihara@sci.osaka-cu.ac.jp (H. Kihara),

yoshioka@sci.osaka-cu.ac.jp (R. Yoshioka).
doi:10.1016/j.nuclphysb.2006.11.001
286
H. Itoyama et al. / Nuclear Physics B 762 [PM] (2007) 285300
G = SU(N )/ZN . We will call it IKKT action. Hirano and Kato showed that the IKKT action
is topological [4]. Topological field theory is introduced in [5]. In 1998 Moore, Nekrasov and
Shatashvili (MNS) [6], being motivated by the existence of D-particle bound states [711], computed the partition functions of the zero-dimensional supersymmetric matrix models as the deficit
terms of the Witten indices [1214],1

1
Z=
(1.1)
[dX][d ]eS .
vol(G)
Here S is the IKKT action and we denote bosonic matrices by X and fermionic matrices by
collectively.
The partition functions of matrix models are expressed as functional integrals with the actions reduced from higher (4, 6 and 10) dimensional gauge theories to their zero-dimensional
counterparts. MNS treated the topologically twisted models. The action given by reduction to
zero dimension of the topologically twisted N = 1, d-dimensional super-YangMills action with
gauge group G is denoted by MNS(d, G) action. MNS obtained the value of the integral for the
MNS(4, SU(N )/ZN ) action by using the Cauchy determinant formula. They also obtained the
results for d = 6 and d = 10 with gauge group G = SU(N )/ZN . Kostov et al. studied the partition functions or the correlation functions of the models reduced to various dimensions [15,16].
Suyama and Tsuchiya calculated the exact partition function of the IIB matrix model with gauge
group SU(2) [17]. Sugino et al. have developed the improved Gaussian and mean field approximation method for the reduced YangMills integrals [18]. Austing and Wheater discussed the
finiteness of the SU(N ) bosonic YangMills matrix integrals [19]. Dorey et al. claimed that in a
certain limit the D-instanton partition function reduces to the functional integral of N = 4 U(N )
supersymmetric gauge theory for multi-instanton solutions [20]. Their review on the calculus of
many instantons is helpful.
2. Preliminaries
Generalizations of Eq. (1.1) to orthogonal and symplectic groups are discussed and partial results are obtained [2123]. In this article we evaluate matrix integrals for the MNS(4, G) actions
in cases of G = SO(2N ), SO(2N + 1) and USp(2N ).
We use the following notations. Let G be a Lie group. G is the Lie algebra of G and C G
is a Cartan subalgebra of G. Below, we list some examples of Cartan subalgebras for classical
groups.
(1)
(2)
(3)
(4)
SU(N ): C = {i; = diag(1 , . . . , N ), 1 + + N = 0, i R};

SO(2N): C = {; = 1 J N J, i R};
USp(2N): C = {; = 1 J N J, i R};
SO(2N + 1): C = {; = 1 J N J (0), i R},
where J = i2 . Let be a root system associated with C. We denote the dual space of C by C .
Let C and C . We define the inner product , = (). The dual basis ei is defined by
1 van Baal attempted to deal with the orbifold singularities in the moduli space of flat connections for supersymmetric
gauge theories on the torus. The vacuum valley parametrized by the Abelian zero-momentum modes and the effective
Hamiltonian requires modification due to a singularity in the non-adiabatic behavior at the orbifold singularities.
287
Table 1
Orders of centers and the Weyl groups for classical groups
AN 1
BN
CN
DN
G
#W
#ZG
SU(N )
N!
N
SO(2N + 1)
N !2N
1
USp(2N )
N !2N
2
SO(2N )
N !2N 1
2

AN 1 = (ei ej ), (i < j ) ,

BN = (ei ej ), (ei + ej ), ek (i < j ) ,

CN = (ei ej ), (ei + ej ), 2ek (i < j ) ,

DN = (ei ej ), (ei + ej ), (i < j ) ,
(2.1)
ei , = i for classical groups:

G = SU(N ),
G = SO(2N + 1),
G = USp(2N),
G = SO(2N),
In (2.1) AN , BN , CN and DN are the root systems associated with the Lie algebras of SU(N + 1),
SO(2N + 1), USp(2N ) and SO(2N ), respectively. The index N is the rank of the root system.
We denote the Weyl group by W and the center of the group G by ZG . We summarize the orders
of W and ZG for AN , BN , CN and DN series in Table 1.
The number of set X is denoted by #X. Each root defines a hyperplane , = 0 in the
vector space C. These hyperplanes divide the space C into finitely many connected components
called the Weyl chambers. These are open, convex subsets of C.
We now consider the partition function Z of the model with a gauge group G, where is a
root system associated with G,

1
Z =
[dX][d ]eS ,
vol(G)

1
1
(M, N = 1, 2, 3, 4).
S = Tr [XM , XN ]2 M [XM , ]
(2.2)
4
2
Here XM , are G-valued and the measures [dX], [d ] are G-invariant measures in this article.
We choose two matrices X3 and X4 and arrange them into a complex matrix = X3 + iX4 .
According to the prescription of MNS [6] the functional integrals reduce to the integral of . In
addition can be integrated over G. In reducing the integral on from G to C we obtain the
integral

r
,
d
1 #ZG
.
Z = r
(2.3)
E #W
, E
2 1
=1
Here E is a deformation parameter associated with the global symmetry SO(2) and r is the rank
of . The integrand is a rational function of i and the degree of the denominator is equal to that
of the numerator. Naively the integrals diverge, so we must regularize and renormalize the integrals. We cutoff the integrals by introducing a parameter temporarily. We add an integral along
the upper half-circle with radius in every i plane as a counter term. Then the renormalized
partition function becomes

r
,
1 #ZG
d
R
.
Z = r
(2.4)
E #W
, E
2 1
=1
288
We work on the residue calculations. We shift E to the pure imaginary direction to avoid the
poles. The nontrivial contributions come from points in which at least r divisors {a ,
E}a=1,...,r take zero value. Such a point is a solution of a system of linear equations {, E =
0} , ( : subset, # r). The roots are separated into two kinds, positive and negative
roots by a certain partial order. The collections of positive and negative roots will usually be
denoted + and , respectively. The integrand can be regarded as a function of , (
+ ),

r

#Z
, 2
1
d
G

R
= r
.
Z
(2.5)
E #W
, 2 E 2
2 1
=1
We will carry out these integrals for AN , BN , CN and DN .

3. AN 1 series
Let us reproduce the result for AN 1 . The element of the Cartan subalgebra is a traceless
hermitian matrix as mentioned above. Let (# = k) be a subset of AN 1 ; = {a = a (eia
eja ), ia < ja , a {1}, a = 1, 2, . . . , k}.
We explain how to draw the diagram associated with . First of all, we draw a small
circle for each ei (i = 1, 2, . . . , N). The circle for ei is denoted by {i}. Next, we draw a line
between two circles {ia } and {ja }. The line is endowed with the sign a for each a = 1, . . . , k.
The diagram consists of N small circles and k lines endowed with sign . The line from {i}
to {j } endowed with is denoted by {i j }(i < j ). Some typical diagrams are depicted in
Fig. 1. We obtain a system of linear equations from a subset ;
, E = 0,
(3.1)
The line {ia ja } corresponds to an equation; a (ia ja ) E = 0. To evaluate the integral

we consider diagrams which include just k = N 1 lines. One such diagram corresponds to
a term contributing to the partition function. The term is given as a residue at a zero of such a
system {, E = 0} .
We can show that many of diagrams do not contribute to the partition function. Indeed a
+
+
+
diagram including folded diagrams {i j } {j k},2 loop diagrams {i j } {k l}
+
+
{i l} and branching diagrams {i j } {j k} {j l} as subdiagrams does not

contribute. Let us prove this statement. We consider three circles {i}, {j }, {k} and draw a line
+
{i j }. Because we obtain an equation i j = E from the line, we consider the residue at
Fig. 1. (a) Root ei ej , (b) f : folded diagram, (c) b : branching diagram, (d) l : loop diagram.
2 Folded means that the two lines attached to the same circle are endowed with different signs.
i = j + E,

dj
(i j )2
(j k )2
(i k )2
di
2 1
2 1 (i j )2 E 2 (i k )2 E 2 (j k )2 E 2

dj E (j k + E)2
(j k )2
2 1 2 (j k + E)2 E 2 (j k )2 E 2

dj E
(j k + E)(j k )
.
=
2 1 2 (j k + 2E)(j k E)
+
289
(3.2)
The factors corresponding to {i k} and {j k} in the denominator are divided by factors

+
in the numerator. The factors corresponding to {j k} and {i k} remain. The factor

j k + 2E is obtained from the factor i k + E. This result shows that a diagram including
a folded diagram does not contribute.
We concretely calculate the residue at the zero of {i j E = 0, j k E = 0},

dj
(i j )2
(j k )2
di
(i k )2
E2
. (3.3)
3
2 1
2 1 (i j )2 E 2 (i k )2 E 2 (j k )2 E 2
The factor i k is a sum of the two factors (i j ) and (j k ). Thus the factor i k
is not linearly independent of i j and j k . This argument is easily generalized to a
+
long loop {i1 i2 }{i2 i3 } {ip1 ip }{i1 ip }. The factor i1 ip is a sum
p1
a=1 (ia ia+1 ). Thus there is no solution to {ia ia+1 E = 0 (a = 1, . . . , p 1), i1
ip E = 0} (p > 2). This result includes that there is no contribution from a diagram including
a loop subdiagram either.
+
+
Next we consider four circles {i}, {j }, {k}, {l} and draw two lines {i j } and {j k}.
The corresponding residue calculation is as follows:

dj
(i j )2
di
(i k )2
dk
2
2
2 1
2 1
2 1 (i j ) E (i k )2 E 2
(j k )2
(j l )2
(k l )2
(i l )2
(i l )2 E 2 (j k )2 E 2 (j l )2 E 2 (k l )2 E 2

dk E 2 (k l + 2E)(k l )
.
2 1 3 (k l + 3E)(k l E)
+
(3.4)
Two factors corresponding to {k l} and {i l} remained. We take no account of branching diagrams because of this result. Thus these results imply that there is no contribution from
the diagrams which include one of these three types as a subdiagram.
+
We can draw only straight line configurations like {i i + 1} (i = 1, 2, . . . , N 1) by
+
this prescription. In fact every allowed diagram can be transformed into the diagram {i
i + 1} (i = 1, 2, . . . , N 1) with a Weyl transformation which reorders their indices. In addition,
{ei ei+1 , (i = 1, . . . , N 1)} is a fundamental root system. One might think that only diagrams
constructed from fundamental root systems are relevant for every gauge group. We will show
later that this inference is not correct. Let us continue the remaining calculation for AN 1 . The
residue at the solution to {(i i+1 ) E = 0}i=1,...,N 1 is explicitly calculated. Taking account
of the multiplicity (N 1)! caused by the Weyl group, we finish the calculation of the partition
290
function for AN 1 ,
ZAN1

1 N E N 1 1
= (N 1)! N
E N! 2
N

i<j,j i=1
(zi zj )2
(zi zj )2 E 2

=
1
,
N2
1
zi = (N + 1 2i)E.
(3.5)
2
This result agrees with the original result of MNS, which is derived from the Cauchy determinant
formula. This demonstrates that our diagrammatic method works for AN 1 . In the remainder of
this article we carry out the same calculus for other classical groups.
4. DN series
In this section we develop the diagrammatic method for the DN series properly. The gauge
group associated with the root system DN is SO(2N )/Z2 (ZSO(2N ) = Z2 ). The difference between SO(2N) and SO(2N )/Z2 gives rise to the difference in the multiplicities. We do not take
care of this difference until we consider the multiplicities. The root system DN consists of roots
{(ei ej )} and {(ei + ej )}. We evaluate the renormalized partition function ZDN for the root
system DN ,

N

2
,
1
d
,
ZDN = N
(4.1)
E N !2N 1
,
E
2 1 D
=1
N
where we have omitted the superscript R. The Weyl transformation plays an important role in
this case. The Weyl group consists of permutations i (i), SN and sign flips (ei , ej )
(ei , ej ) (i < j ). The integrand is also invariant under sign flips ei ei . The root system
DN includes two kinds of positive roots ei ej and ei + ej (i < j ).
Now we extend the diagrammatic method. Let us draw circles for ei and lines for the roots as
well as those for SU(N ). To express the difference between two kinds of positive roots, we use
solid lines for ei ej and broken lines for ei + ej . A solid line between {i} and {j } is represented
+
by {i j } and a broken line between these is by {i - -+- - j } as Fig. 2. Each diagram includes
N circles. Because the partition function has N integrations, diagrams with N lines contribute to
the partition function. Every diagram with N circles and N lines must contain at least one loop
subdiagram. The number of loops corresponds to that of connected components of the diagram.
The symmetry induces transformations on diagrams. Two diagrams which can be transformed
each other yield the same contribution. In particular every connected diagram can be transformed
into a diagram with only one broken line.
Let us prove this statement partially. If we encounter a broken line {i - -+- - j }, we change the
+
sign ej . Then the line {i - -+- - j } is transformed into {i j }. These transformations eliminate
almost all broken lines. However if we encounter {i j } and {k - -- - j }, we cannot decrease

the number of broken lines, because the sign flip does not change the number. Though there is no
Fig. 2. (a) ei ej , (b) ei + ej .
291
proof that disconnected diagrams do not contribute, we consider only connected diagrams. This
conjecture is partially obtained from some explicit evaluations. A connected diagram can include
only one loop because two loops imply the disconnectedness. For solid lines the rules in the case
of AN 1 are still valid. The branching {i j }{j k}{j - - - - l} cannot happen because
the sign flip el el transforms that to the branching of solid lines. Thus the possible cases
are {i j }{j k}{j - - - - k}. Here we omitted their signs. Hence the connected diagram
which contributes to the partition function is transformed into a diagram which has only one
broken line in the loop part {i j }{i - - - - j }. We have arrived at our result.
Now we consider a valid connected diagram with one broken line. Its subdiagram consisting
of all of the solid lines is a straight configuration which is the same one as AN 1 . It reflects
that the group SU(N ) is a subgroup of SO(2N ). The following change of variables makes this
situation clear;
1
(1 + 2 + + N ),
i = i x,
N
(0 , 1 , . . . , N 1 ) (1)N 1
1 + 2 + + N = 0,
J=
=
.
(1 , 2 , . . . , N )
N
x = 0 =
(4.2)
Here we separate coordinates into center of mass x and relative coordinates i . Then the
factors, , , are written as,
i j = i j ,
i + j = i + j + 2x.
(4.3)
We pick up residues at i i+1 = +E (i = 1, 2, . . . , N 1). This is the same system of

equations as that appeared in the case of AN 1 . This residue calculation reduces the original
integral to the one over a Weyl chamber W of the subgroup SU(N ). We obtain the solution to
i i+1 = +E,
1
zi zj = (i j )E,
i = zi = (N + 1 2i)E,
2

zi + zj = (N + 1) (i + j ) E.
(4.4)
W from the Weyl chamber W to the

After we pick up residues at these points, the contribution ZD
N
partition function ZDN is given by an integral of x,
N 1

(zi zj )2
E
2
1
1
W
N 1
=
(1)
N
ZD
N
E N N !2N 1
2
N
(zi zj )2 E 2
i<j,j i=1
N
1
N

(zi + zj + 2x)2
dx
2 1 i=1 j =i+1 (zi + zj + 2x)2 E 2
(1)N 1
N ! N 2N 1
N 1
d (N + 1) (2i + 1)
.
(2i)
2 1 2N
i=1
(4.5)
W is not equal to
We have made a change of variable, x = 2x/E + (N + 1). Though ZD
N
the full ZDN , this calculation reveals the proper set of points which contribute to the partition
function ZDN .
In order to evaluate the full contribution, we must determine the multiplicities of contributions.
Multiplicities originate from the transformation properties of the Weyl group. Permutations yield
292
N! terms and sign flip transformations bring the 2m terms for some integer m. The determination
of m is the heart of the problem of multiplicities and we will carry this out now. The integrand in
(4.5) has an even number of poles. A pole x = w has a counterpart w in the x plane. The poles
in plane in (4.5) are:
N = 2p
= 2, 4, 6, . . . , 2p, 2p + 2, . . . , 4p,
N = 2p + 1 = 2, 4, 6, . . . , 2p, 2p + 4, . . . , 4p + 2.
(4.6)
The poles = 2, 4, 6, . . . , 2p in (4.6) correspond to the minus poles in x plane while the remainder to the plus. The contributions of plus poles and minus poles are equivalent if the orientation
of the integration is considered. The integral (4.5) then consists of p different terms. This also
shows that there are diagrams which are not constructed from the fundamental root systems.
Now we return to the calculation of full contribution. We must take care of the orbits of these
points under the Weyl transformation. The pole = 2j represents a point,

(D)
Xj = (1 , . . . , N ) = (j 1)E, (j 2)E, . . . , E, 0, E, . . . , (j N )E .
(4.7)
(D)
This point Xj
is stable under the subgroup Hj of the Weyl group whose order is 2j 1 . The
group Hj is the isotropy group of the point Xj . There are #W/#Hj = N !2N j points which
(D)
(D)
have the same residue as that at the point Xj . In order to compute the full contribution to ZDN ,
we must count the multiplicities of p terms. Let us denote by j -type a point whose residue is
(D)
equal to that of Xj . The existence of the nontrivial isotropy group means that the point with
j 2 is on the boundary of the closure of a Weyl chamber. There are two p-type points in W.
To evaluate the integral (4.5), we collect the residues at = 2, 4, 6, . . . , 2p. If the rank N is even
(N = 2p), the factor 2N j runs from 22p1 to 2p . To calculate the minimal contribution, we find
that the multiplicity is equal to 2pj which is given by dividing 22pj by 2p . If the rank N is
odd (N = 2p + 1), the factor 2N j runs from 22p to 2p+1 . In this case, the multiplicity can be
read off as 2pj which is given by dividing 22pj +1 by 2p+1 . Finally we must take the center
Z2 into account, which yields a factor 1/2. We have carried out this evaluation for all N . The
results are given as follows:
(1) N = 2p case (SO(4p))

2p 2j + 1 (2j 3)!! (4p 2j 1)!!
1
,
22pj
2p
4p 2j (2j 2)!! (4p 2j 2)!!
2 (2p)
p
ZD2p =
(4.8)
j =1
(2) N = 2p + 1 case (SO(4p + 2))

2p 2j + 2 (2j 3)!! (4p 2j + 1)!!
1
.
22pj +1
4p 2j + 2 (2j 2)!! (4p 2j )!!
22p+1 (2p + 1)
p
ZD2p+1 =
(4.9)
j =1
Here p 1. This results disagree with the previous works. The validity of our results will be
examined in Appendix A.
5. BN and CN
Next we evaluate the partition functions for the root systems BN and CN . They are the
root systems with respect to SO(2N + 1) and USp(2N ). Their centers are ZSO(2N +1) = {1}
293
Fig. 3. Root P ei .
and ZUSp(2N ) = Z2 . The factors caused by these groups must be taken into account at the
calculations of multiplicities. The partition functions for these root systems can be treated in
parallel to the DN case in the last section. The root system BN or CN has three types of roots;
{(ei ej ), i < j }, {(ei + ej ), i < j } and {P ei } (P = B, C, B = 1, C = 2). The circles and the lines are the same as those for the DN . Roots P ei are represented by cyclic lines
introduced in Fig. 3.
We regard the cyclic line as a kind of loop subdiagram. A connected diagram also has only
one loop in these cases. So a connected diagram has a cyclic line or a loop {i j }{i - - - - j }.
The transformations on the diagrams can be defined and the reduction of the number of broken
lines can be applied to these cases. The diagrams are transformed into those with zero or one
broken line. The partition function for the root system PN (P = B, C) is,

N

,
1 #ZPN
d
,
ZPN = N
(5.1)
E N !2N
,
E
2 1
=1
PN
where #ZBN = 1 and #ZCN = 2. The variables, x and i , introduced in the case of DN are also
valid,
1
(1 + 2 + + N ),
i = i x,
N
(0 , 1 , . . . , N 1 ) (1)N 1
=
,
1 + 2 + + N = 0,
J=
(1 , 2 , . . . , N )
N
x = 0 =
1
i = zi = (N + 1 2i)E,
2

zi zj = (i j )E,
zi + zj = (N + 1) (i + j ) E.
(5.2)
(5.3)
We can perform the integrals of i in the same manner. Then the contributions from the Weyl
chambers of the subgroups SU(N ) are
N 1

N 1
(2i + 1)
(1)
d
(
3)(
(N
+
1))
ZBWN =
(5.4)
,
2i
N 2N +1
2 1 ( 2(N + 1))
ZCWN =
(1)N 1
N 2N
i=2
N
d
( N 1) ( 2i)
,
( 2i + 1)
2 1 ( 2N 1)
(5.5)
i=1
where = 2x/E + (N + 1) again. For the root system BN , the poles in the -plane are
N = 2p
N = 2p + 1
= 0, 4, 6, 8, . . . , 2p, 2p + 2, . . . , 4p 2, 4p + 2,
= 0, 4, 6, 8, . . . , 2p, 2p + 4 . . . , 4p, 4p + 4.
(5.6)
For the root system CN , the poles in the -plane are

N = 2p
N = 2p + 1
= 1, 3, . . . , 2p 1, 2p + 3, 2p + 5, . . . , 4p 1, 4p + 1,
= 1, 3, . . . , 2p + 1, 2p + 3, . . . , 4p + 1, 4p + 3.
(5.7)
294
These poles make pairs as the case of DN . We must calculate the multiplicities to finish these
calculations. For BN the pole = 2j (j = 0, 2, 3, . . . , p) represents a point,

(B)
Xj = (1 , . . . , N ) = (j 1)E, (j 2)E, . . . , E, 0, E, . . . , (j N )E .
(5.8)
For CN the pole = 2j 1 (j = 1, 2, 3, . . . , [(N + 1)/2]) represents a point,
(C)
Xj
= (1 , . . . , N )

= (2j 3)E/2, (2j 5)E/2, . . . , E/2, E/2, . . . , (2j 1 2N )E/2 .
(5.9)
Orders of isotropy groups Hj(B,C) for Xj(B,C) (1 j ) are 2j 1 and the order of H0(B) for X0(B)
is 2N . These orders determine the multiplicities. Carrying out the residue calculus, we obtain,

(4p 1)!!
1
ZB2p =
22p
2p+1
(4p 2)!!
(2p)2

p

(2j
5)!!
(4p
2j
1)!!
(2j
3)(2p
2j
+
1)
+
(5.10)
22pj +1
,
2j (4p 2j + 2) (2j 4)!! (4p 2j 2)!!
j =2
ZB2p+1

(4p + 1)!!
1
=
22p+1
(4p)!!
(2p + 1)22p+2
+

3)(2p 2j + 2) (2j 5)!! (4p 2j + 1)!!
,
2j (4p 2j + 4) (2j 4)!! (4p 2j )!!
2pj +2 (2j
j =2
ZC2p =
(5.11)

p

1
2pj +1 (p j + 1) (2j 3)!! (4p 2j + 1)!!
,
2
(2p j + 1) (2j 2)!! (4p 2j )!!
(2p)22p+1
(5.12)
j =1

p+1

1
2pj +2 (2p 2j + 3) (2j 3)!! (4p 2j + 3)!!
ZC2p+1 =
2
.
(4p 2j + 4) (2j 2)!! (4p 2j + 2)!!
(2p + 1)22p+2
j =1
(5.13)
These expressions are valid for p 1. We summarize the values of partition functions for B, C
and D for small values of p in Table 2.
Table 2
The absolute values of partition functions for root systems B, C and D
p
D2p
1
8
1
8
3
8
5
16
3
16
11
64
33
256
1
8
71
256
129
512
5
32
75
512
125
1024
483
4096
957
4096
7
32
1127
8192
4279
32768
29953
262144
3621
32768
54195
262144
51501
262144
32589
262144
124765
1048576
224577
2097152
217705
2097152
196805
1048576
754839
4194304
478951
4194304
1844775
16777216
D2p+1
B2p
B2p+1
C2p
C2p+1
295
6. Accidental isomorphisms
We have calculated the partition functions for all classical gauge groups. We must confirm
our results. Let us check some of the well-known correspondence among lower-dimensional Lie
groups. Groups SO(4), SO(5) and SO(6) are locally isomorphic to SU(2) SU(2), USp(4) and
SU(4), respectively. One might think that the values of the partition functions in each pair should
be equal. Let classical groups G1 and G2 be locally isomorphic to each other. The integration
(l)
(1)
(2)
variables for Gl are i (l = 1, 2). The variables {i } do not coincide with {i } and we must
consider the Jacobian. To clarify this argument we construct an explicit isomorphism between
members of each pair.
6.1. SO(4) and SU(2) SU(2)
The root system of SO(4) is D2 and that of SU(2) SU(2) is A1 A1 . Let us construct an
isomorphism which maps D2 = {(1 2 ), (1 + 2 )} onto A1 A1 = {21 , 22 }.
The isomorphism is given as

1 1 1
1
1
=
.
(6.1)
2
2
2 1 1
Thus the Jacobian for the change of variables (1 , 2 ) (1 , 2 ) is 2. So the value of the
integration for SO(4) becomes 2(1/4 1/4). Note that our prescription for the multiplicity has
been defined such that the pre-factor #ZG /#W is canceled. The correspondence between SO(4)
and SU(2) SU(2) has been confirmed.
6.2. SO(5) and USp(4)
The root system of SO(5) is B2 and that of USp(4) is C2 . Let us construct an isomorphism which maps B2 = {(1 2 ), (1 + 2 ), 1 , 2 } onto C2 = {(1 2 ), (1 +
2 ), 21 , 22 }.
The isomorphism is given as,

1 1 1
1
1
=
.
(6.2)
2
2
2 1 1
Indeed under this change of variables, every element in B2 maps to C2 . The Jacobian for the
change of variables (1 , 2 ) (1 , 2 ) is 2. So the value of the integration for SO(5) becomes
2(3/16). The correspondence between SO(5) and USp(4) has been confirmed.
6.3. SO(6) and SU(4)
The root system of SO(6) is D3 and that of SU(4) is A3 .

D3 = (1 2 ), (1 3 ), (2 3 ), (1 + 2 ), (1 + 3 ), (2 + 3 ) ,
(6.3)

A3 = (1 2 ), (1 3 ), (2 3 )

(21 + 2 + 3 ), (1 + 22 + 3 ), (1 + 2 + 23 ) .
(6.4)
296
The isomorphism is given as,

1
1
1 1 1 1
2 =
2 .
1
1 1
2 1 1
3
3
1
(6.5)
Under this change of variables, every element of D3 maps into that of A3 . The Jacobian for this
change of variables (1 , 2 , 3 ) (1 , 2 , 3 ) is 2. So the value of the integration for SO(6)
becomes 2(1/16). The correspondence between SO(6) and SU(4) has been confirmed.
These demonstrations consolidate the correctness of our evaluations.
7. Summary and discussions
We evaluated the partition functions for all classical gauge groups by using diagrammatic
methods. The actions are given as topologically twisted and dimensionally reduced actions of
d = 4, N = 1 super-YangMills theories with classical (semi-)simple gauge groups. Our diagrammatic methods revealed multiplicities of poles which contribute equally. The multiplicities
for points on the boundary of a Weyl chamber were given correctly. With similar manner, the partition functions for dimensionally reduced actions of d = 6 and d = 10, N = 1 super-YangMills
theories might be evaluated.
Our original motivation for carrying out matrix integrals for groups other than SU(N ) stems
from an interesting class of USp and SO matrix models [2427]. It may be that our evaluations
are useful in investigating the dynamical generation of spacetime, which is so far examined in
mean field approximations [28].
Acknowledgements
We are grateful to Yukinori Yasui and Takeshi Oota for useful comments. This work is supported by the 21 COE program Constitution of wide-angle mathematical basis focused on
knots and in part by the Grant-in Aid for scientific Research (No. 18540285) from Japan Ministry of Education.
Appendix A. Direct calculation for D2 , B2 and C2
We present the direct calculations for D2 , B2 and C2 in this appendix.
The root system D2 is related to the Lie group SO(4). The partition function is given by

(1 + 2 )2
1 2
d1
d2
(1 2 )2
.
ZD2 = 2
(A.1)
E 2!22
2 1
2 1 (1 2 )2 E 2 (1 + 2 )2 E 2
The intersections of the lines 1 2 = E contribute to the integral. Fig. 4 shows all intersections.
The residues at all points take the same value. We calculate that at (0, E) which is a intersection of 1 2 = E and 1 + 2 = E

(1 2 )2
E (22 + E)2
(1 + 2 )2
=
,
Res
1 2 =E (1 2 )2 E 2 (1 + 2 )2 E 2
2 42 (2 + E)
E2
E (22 + E)2
=
.
2 =0 2 42 (2 + E)
8
Res
(A.2)
297
Fig. 4. The gray points in this figure contribute to the partition function of D2 .
Fig. 5. The gray points in this figure contribute to the partition function of B2 and the blank points do not.
The root system B2 is related to the Lie group SO(5). The partition function is given by

(1 + 2 )2
1 2
d1
d2
(1 2 )2
ZB2 = 2
2
2
2
E 2!2
2 1
2 1 (1 2 ) E (1 + 2 )2 E 2
12
22
12 E 2 22 E 2
(A.3)
The intersections of the lines 1 2 = E, i = E contribute to the integral. Fig. 5 shows all
intersections.
The residues at all points also take the same value. We calculate that at (2E, E) which is a
intersection of 1 2 = E and 2 = E.

12
22
(1 2 )2
(1 + 2 )2
Res
1 2 =E (1 2 )2 E 2 (1 + 2 )2 E 2 2 E 2 2 E 2
1
2
E
(22 + E)2
,
2 4(2 + 2E)(2 E)
(22 + E)2
E
3E 2
Res
=
.
2 =E 2 4(2 + 2E)(2 E)
8
=
(A.4)
One might pick up positive poles upon 1 integration, which are on the lines, 1 2 = E
and 1 = E. We can find the solutions; (2E, E), (E, 2E) and (E, 2E). Here we select the
positives. We sum up the contributions from the three points. Then the value of the partition
298
Fig. 6. The gray points in this figure contribute to the partition function of C2 and the blank points do not.
function becomes 3 2/2!22 3/8 = 9/32. If we divide the result 9/32 by the order of the
center Z2 , we obtain the result 9/64 which is the same value in the previous works [2123]. Our
results in Table 2 do not agree with this result.
The root system C2 is related to the Lie group USp(4). The partition function is given by

(1 + 2 )2
1 2
d1
d2
(1 2 )2
ZD2 = 2
E 2!22
2 1
2 1 (1 2 )2 E 2 (1 + 2 )2 E 2
2
2
4
42
2 1
.
(A.5)
2
2
41 E 42 E 2
The intersections of the lines 1 2 = E, 2i = E contribute to the integral. Fig. 6 shows
all intersections.
The residues at all points take the same value. We calculate that at (3E/2, E/2) which is a
intersection of 1 2 = E and 22 = E.

412
422
(1 2 )2
(1 + 2 )2
Res
1 2 =E (1 2 )2 E 2 (1 + 2 )2 E 2 4 2 E 2 4 2 E 2
1
2
E
42 (2 + E)
=
,
2 (22 + 3E)(22 E)
2 (2 + E)
3E 2
E
=
.
Res
(A.6)
2 =E/2 2 (2 + 3E/2)(2 E/2)
16
These results support our calculation.
References
[1] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: A conjecture, Phys. Rev. D 55 (1997)
5112, hep-th/9610043.
[2] B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys. B 305 (1988) 545.
[3] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, A large-N reduced model as superstring, Nucl. Phys. B 498
(1997) 467, hep-th/9612115.
[4] S. Hirano, M. Kato, Topological matrix model, Prog. Theor. Phys. 98 (1997) 1371, hep-th/9708039.
[5] E. Witten, Topological quantum field theory, Commun. Math. Phys. 117 (1988) 353.
[6] G.W. Moore, N. Nekrasov, S. Shatashvili, D-particle bound states and generalized instantons, Commun. Math.
Phys. 209 (2000) 77, hep-th/9803265.
[7] E. Witten, Bound states of strings and p-branes, Nucl. Phys. B 460 (1996) 335, hep-th/9510135.
299
[8] P. Yi, Witten index and threshold bound states of D-branes, Nucl. Phys. B 505 (1997) 307, hep-th/9704098.
[9] S. Sethi, M. Stern, D-brane bound states redux, Commun. Math. Phys. 194 (1998) 675, hep-th/9705046.
[10] M. Porrati, A. Rozenberg, Bound states at threshold in supersymmetric quantum mechanics, Nucl. Phys. B 515
(1998) 184, hep-th/9708119.
[11] M.B. Green, M. Gutperle, D-particle bound states and the D-instanton measure, JHEP 9801 (1998) 005, hepth/9711107.
[12] E. Witten, Dynamical breaking of supersymmetry, Nucl. Phys. B 188 (1981) 513;
E. Witten, Constraints on supersymmetry breaking, Nucl. Phys. B 202 (1982) 253.
[13] S. Cecotti, L. Girardello, Functional measure, topology and dynamical supersymmetry breaking, Phys. Lett. B 110
(1982) 39;
L. Girardello, C. Imbimbo, S. Mukhi, On constant configurations and the evaluation of the Witten index, Phys. Lett.
B 132 (1982) 69;
M. Claudson, M.B. Halpern, Supersymmetric ground state wave functions, Nucl. Phys. B 250 (1985) 689;
S.F. Cordes, M. Dine, Chiral symmetry breaking in supersymmetric O(N ) gauge theories, Nucl. Phys. B 273 (1986)
581;
H. Itoyama, Supersymmetry and zero momentum modes, Phys. Rev. D 33 (1986) 3060;
A.V. Smilga, Perturbative corrections to effective zero mode Hamiltonian in supersymmetric QED, Nucl. Phys.
B 291 (1987) 241;
H. Itoyama, B. Razzaghe-Ashrafi, Ground state structure of supersymmetric YangMills theory, Nucl. Phys. B 354
(1991) 85.
[14] P. van Baal, The Witten index beyond the adiabatic approximation, hep-th/0112072.
[15] V.A. Kazakov, I.K. Kostov, N.A. Nekrasov, D-particles, matrix integrals and KP hierarchy, Nucl. Phys. B 557
(1999) 413, hep-th/9810035.
[16] I.K. Kostov, P. Vanhove, Matrix string partition functions, Phys. Lett. B 444 (1998) 196, hep-th/9809130.
[17] T. Suyama, A. Tsuchiya, Exact results in N (c) = 2 IIB matrix model, Prog. Theor. Phys. 99 (1998) 321, hepth/9711073.
[18] S. Oda, F. Sugino, Gaussian and mean field approximations for reduced YangMills integrals, JHEP 0103 (2001)
026, hep-th/0011175;
F. Sugino, Gaussian and mean field approximations for reduced 4D supersymmetric YangMills integral, JHEP 0107
(2001) 014, hep-th/0105284;
J. Nishimura, T. Okubo, F. Sugino, Convergent Gaussian expansion method: Demonstration in reduced YangMills
integrals, JHEP 0210 (2002) 043, hep-th/0205253.
[19] P. Austing, J.F. Wheater, The convergence of YangMills integrals, JHEP 0102 (2001) 028, hep-th/0101071;
P. Austing, J.F. Wheater, Convergent YangMills matrix theories, JHEP 0104 (2001) 019, hep-th/0103159.
[20] N. Dorey, T.J. Hollowood, V.V. Khoze, The D-instanton partition function, JHEP 0103 (2001) 040, hep-th/0011247;
N. Dorey, T.J. Hollowood, V.V. Khoze, M.P. Mattis, The calculus of many instantons, Phys. Rep. 371 (2002) 231,
hep-th/0206063.
[21] W. Krauth, H. Nicolai, M. Staudacher, Monte Carlo approach to M-theory, Phys. Lett. B 431 (1998) 31, hepth/9803117;
W. Krauth, M. Staudacher, Finite YangMills integrals, Phys. Lett. B 435 (1998) 350, hep-th/9804199;
W. Krauth, M. Staudacher, Eigenvalue distributions in YangMills integrals, Phys. Lett. B 453 (1999) 253, hepth/9902113;
W. Krauth, M. Staudacher, YangMills integrals for orthogonal, symplectic and exceptional groups, Nucl. Phys.
B 584 (2000) 641, hep-th/0004076;
M. Staudacher, Bulk Witten indices and the number of normalizable ground states in supersymmetric quantum
mechanics of orthogonal, symplectic and exceptional groups, Phys. Lett. B 488 (2000) 194, hep-th/0006234;
W. Krauth, M. Staudacher, Statistical physics approach to M-theory integrals, cond-mat/0010127.
[22] V.G. Kac, A.V. Smilga, Vacuum structure in supersymmetric YangMills theories with any gauge group, hepth/9902029;
V.G. Kac, A.V. Smilga, Normalized vacuum states in N = 4 supersymmetric YangMills quantum mechanics with
any gauge group, Nucl. Phys. B 571 (2000) 515, hep-th/9908096.
[23] V. Pestun, N = 4 SYM matrix integrals for almost all simple gauge groups (except E(7) and E(8)), JHEP 0209
(2002) 012, hep-th/0206069.
[24] H. Itoyama, A. Tokura, USp(2k) matrix model: F-theory connection, Prog. Theor. Phys. 99 (1998) 129, hepth/9708123;
H. Itoyama, A. Tokura, USp(2k) matrix model: Nonperturbative approach to orientifolds, Phys. Rev. D 58 (1998)
026002, hep-th/9801084;
300
H. Itoyama, A. Tsuchiya, USp(2k) matrix model, Prog. Theor. Phys. Suppl. 134 (1999) 18, hep-th/9904018.
[25] K. Ezawa, Y. Matsuo, K. Murakami, Matrix model for Dirichlet open string, Phys. Lett. B 439 (1998) 29, hepth/9802164.
[26] H. Itoyama, T. Matsuo, Berrys connection and USp(2k) matrix model, Phys. Lett. B 439 (1998) 46, hepth/9806139;
B. Chen, H. Itoyama, H. Kihara, Non-Abelian Berry phase, YangMills instanton and USp(2k) matrix model, Mod.
Phys. Lett. A 14 (1999) 869, hep-th/9810237;
B. Chen, H. Itoyama, H. Kihara, Non-Abelian monopoles from matrices: Seeds of the spacetime structure, Nucl.
Phys. B 577 (2000) 23, hep-th/9909075.
[27] H. Itoyama, R. Yoshioka, Matrix orientifolding and models with four or eight supercharges, Phys. Rev. D 72 (2005)
126005, hep-th/0509146.
[28] J. Nishimura, F. Sugino, Dynamical generation of four-dimensional spacetime in the IIB matrix model, JHEP 0205
(2002) 001, hep-th/0111102;
H. Kawai, S. Kawamoto, T. Kuroki, T. Matsuo, S. Shinohara, Mean field approximation of IIB matrix model and
emergence of four dimensional spacetime, Nucl. Phys. B 647 (2002) 153, hep-th/0204240;
H. Kawai, S. Kawamoto, T. Kuroki, S. Shinohara, Improved perturbation theory and four-dimensional spacetime
in IIB matrix model, Prog. Theor. Phys. 109 (2003) 115, hep-th/0211272.
A microscopic model for the black hole:

Black string phase transition
Borun D. Chowdhury a , Stefano Giusto b, , Samir D. Mathur a
a Department of Physics, The Ohio State University, Columbus, OH 43210, USA
b Department of Physics, University of Toronto, Toronto, Ontario, Canada M5S 1A7

Abstract
Computations in general relativity have revealed an interesting phase diagram for the black holeblack
string phase transition, with three different black objects present for a range of mass values. We can add
charges to this system by boosting plus dualities; this makes only kinematic changes in the gravity computation but has the virtue of bringing the system into the near-extremal domain where a microscopic model
can be conjectured. When the compactification radius is very large or very small then we get the microscopic models of (4 + 1)-dimensional near-extremal holes and (3 + 1)-dimensional near-extremal holes
respectively (the latter is a uniform black string in 4 + 1 dimensions). We propose a simple model that
interpolates between these limits and reproduces most of the features of the phase diagram. These results
should help us understand how fractionation of branes works in general situations.
1. Introduction
Semiclassical physics gives us the entropy and Hawking radiation rates of black holes. In
string theory we can understand extremal and near-extremal holes in terms of branes, and thereby
reproduce the entropy and radiation from a microscopic description.
In this paper we will consider another property of black holes that can be described in classical
general relativity: The transition between black holes and black strings when a transverse circle
E-mail addresses: borundev@mps.ohio-state.edu (B.D. Chowdhury), giusto@physics.utoronto.ca,

giusto@mps.ohio-state.edu (S. Giusto), mathur@mps.ohio-state.edu (S.D. Mathur).
doi:10.1016/j.nuclphysb.2006.11.007
302
B.D. Chowdhury et al. / Nuclear Physics B 762 [PM] (2007) 301343
Fig. 1. (a) A small black hole in a space with a compact circle of length L; (b) The horizon distorts when the mass
is increased so that its radius becomes comparable to L; (c) At still larger masses we get a black string which wraps
uniformly around the compact circle.
(a)
(b)
Fig. 2. The phase diagram for (a) 4 non-compact space directions and (b) 3 non-compact space directions. The vertical
axis measures the dimensionless relative tension and the horizontal axis gives a dimensionless mass parameter. The
solid line denotes the uniform black string, the dashed line denotes the black hole, and the dot-dashed line denotes the
non-uniform black string.
is made large or small. We suggest a simple microscopic picture of this transition, which will
reproduce many of the broad features of the phase diagram of the system.
1.1. The phase diagram
In Fig. 1(a) we depict a small black hole in a spacetime with a compact transverse direction
which we will call z; the length of this transverse circle is L. In Fig. 1(b) we depict a hole with a
larger mass; the horizon now feels a significant distortion from the compactification. In Fig. 1(c)
we have increased the mass still further; the black hole horizon has now turned into a black
string horizon.
The black string has a non-zero tension T along the compact circle. The hole in Fig. 1(b)
will also have some tension in this direction. When we take the hole to be very small, as in
Fig. 1(a), the hole does not notice the compactification, and this tension becomes small too,
vanishing for infinitesimal holes. The relevant parameter here is r0 /L, where r0 is the radius of
the horizon. When this parameter is small the effects of the compactification go away, and the
tension becomes ignorable in the determination of the geometry of the hole.
Through a large number of studies, some analytical and some numerical, an interesting phase
diagram has emerged for this system [126]. In Fig. 2(a) we reproduce this diagram for the case
of 5D; i.e., 4 non-compact space dimensions and the compact circle of length L. In Fig. 2(b)
we give the diagram for 4D; i.e., 3 non-compact space directions and the compact circle. (Our
interest will be in the 4D case, but we give the diagram for the 5D case as well since it is more
completenumerical computations are easier in this case since the gravitational fields fall off
faster at infinity.)
(a)
303
(b)
Fig. 3. The entropies for (a) 4 non-compact space directions and (b) 3 non-compact space directions. The vertical axis
gives the ratio of the entropy to the entropy of the uniform black string, and the horizontal axis gives a dimensional mass.
The three kinds of lines label the three phases in the same way as in Fig. 2.
On the horizontal axis we have the mass, scaled by some constant and termed . On the
vertical axis we have a dimensionless tension n = T L/M. For small we have just the black
hole phase, with n 0 for 0. At very large we have the black string phase, with n
independent of . But the most striking feature of these phase diagrams is that we do not see just
a black hole and a black string; for a range of there is a third branch, called the non-uniform
black string.
In each of these phases we can compute the Bekenstein entropy from the area of the horizon.
In Fig. 3(a), (b) we reproduce the entropy graphs. The horizontal axis again gives the mass, and
on the vertical axis we plot s/sus , where sus is the entropy of the uniform string. Thus the uniform
string branch just gives a horizontal line at height unity. The black hole branch is seen to have
higher entropy at low mass, and lower entropy at high mass. The non-uniform string is seen to
have an entropy that is always lower than the other two branches.
1.2. The microscopic picture
Our goal is to see if this phase diagram can be understood in terms of a microscopic description of black holes. The microscopic picture works best for extremal and near-extremal objects.
This is because supersymmetry prevents strong dependences on coupling and thus allows computations in terms of weakly interacting excitations. So our first goal is to map the phase diagram
of neutral objects to the phase diagram for near-extremal objects. A detailed study of this map
was made in [17,25], and we will use many of their results.
We will work in type IIB string theory. Consider the compactification
M9,1 M4,1 T 4 S 1 .
(1.1)
The compact circle of length L will be chosen from the spatial directions in M4,1 , we will call
it S 1 .
We will add large D1 and D5 charges to the neutral object. The addition of charges is done
by boosting the system in the compact direction S 1 ; this gives a momentum charge P in that
direction. Dualities convert this to a D1 charge along S 1 . We boost again along S 1 , and dualize
to get a D5 charge on T 4 S 1 .
First let L , so we have a black object in 4 + 1 non-compact spacetime dimensions. This
is the near extremal D1D5 black hole studied in [27,28]. In the microscopic description, the D1
304
D5 bound state produces an effective string winding along S 1 with total winding number n1 n5 .
The non-extremal energy of the system is carried by a dilute gas of P , P excitations along the
S 1 direction, running along this effective string. The entropy of these excitations is given by

S = 2 n1 n5 np + n p .
(1.2)
With no net P charge, we have np = n p , and the value of np is given by the non-extremal energy.
So we have a microscopic picture of small black holes, i.e. black holes where r0 /L is small.
Now let us imagine that L is small. Then the effective compactification is
M9,1 M3,1 T 4 S 1 S 1 .
(1.3)
With this compactification we can make an extremal black hole with 4 kinds of charges: D1, D5
as above, P along S 1 , and KK monopoles which have S 1 as their non-trivially fibered circle.
What we have however is two nonzero charges (D1, D5) plus non-extremal energy. Note that
the 4 kinds of charges can be permuted in all possible ways by dualities. Using this fact, we can
write the entropy of this near-extremal system from the expression found in [29], getting

nk + n k .
S = 2 n1 n5 np + n p
(1.4)
If the net P , KK charges are zero then np = n p , nk = n k , and the values of np , nk are determined
by extremising S over all possibilities for given total non-extremal energy. This entropy describes
a black object that is uniformly smeared in all the compact directions of (1.3), so it is a uniform
black string in the direction S 1 .
One can write an expression similar to (1.2), (1.4) for the neutral hole as well [30,31]. But
we have chosen to add charges to our system and look at the near-extremal systems (1.2), (1.4)
because for these cases we also have a microscopic derivation of low energy Hawking radiation
[28,32,33]. Thus not only is the count of states correct, their microscopic description must also
be correct because they lead to the expected dynamics.
Let us return to our problem of interest. For large L, we have the black hole phase, described
by (1.2), and for small L we have the uniform black string phase, described by (1.4). What
happens when we reduce L from large values to small values? We seek a microscopic picture
of the intermediate L regime, which will help us understand the three phases that emerge in the
gravity computation; in particular we wish to understand the non-uniform black string in the
microscopic picture.
1.3. Outline of our computations
We proceed in the following steps:
(A) A basic property of gravity solutions with a horizon is the Smarr relation. This is a scaling
relation that relates the tension T of the solution to the mass, charge and entropy. Since our
microscopic model must describe a black object with a horizon, the Smarr relation must be
satisfied by the model. We thus recall the Smarr relation and sketch its derivation.
(B) The gravity solution has a nontrivial metric in M4,1 . The compact directions T 4 S 1
can be added on trivially to the neutral solution in M4,1 ; i.e. the metric is flat and constant
in these directions. From Einsteins equations one then finds that the tension in these trivial
directions must be nonzero in general. Thus from a microscopic perspective, these directions are
not so trivial: One must wrap branes around these directions to obtain the required tension. The
tension in these trivial directions is also important in getting the correct changes of physical
305
parameters under the boosting which adds charge. We recall the equations that relate old and
new quantities upon boosting; these were derived in [17], but we also supply a physical derivation
valid at weak gravitational coupling since this makes some of the physics more transparent.
(C) We next make a microscopic model that describes the transition from the black hole to
the black string. The entropy (1.2) arises from momentum modes np , n p that are fractionated in
units of 1/n1 n5 by the presence of the D1, D5 branes [34,35]. The entropy (1.4) comes from two
kinds of excitations (momentum and KK monopoles) which are again fractionated by the D1,
D5 brane charges [29]. We make a simple model where a part E1 of the non-extremal energy E
goes to creating entropy of the type (1.2) and the rest E E1 goes to creating entropy of type
(1.4). In this process we assume that a fraction (1 x) of the fractionating charge n1 n5 goes to
fractionating the excitations used in (1.2), and the fraction x goes to fractionating the excitations
for (1.4). We then extremise over E1 , x to find the preferred state of the microscopic system.
We find that below some energy E there is just one phase, the black hole phase, where all
entropy is of type (1.2). But for E1 greater than a certain critical value, there are three extrema
of the entropy. One is a black hole with entropy entirely of type (1.2), one is a uniform black
string which has entropy entirely of type (1.4), and a third exremum where the entropy is partly
of type (1.2) and partly of type (1.4). This last extremum is a saddle point of the entropy, and we
identify it with the non-uniform black string. The tension and entropy graphs for this very simple
microscopic model are given in Section 3, in Figs. 5 and 6 respectively.
(D) While this simple model reproduces the three phases seen in the gravity computations, we
notice one simple qualitative difference between the microscopic and gravity computations: The
microscopic model has a vanishing tension in the black hole phase, while the gravity description
gives a tension that rises from zero as the mass is increased from zero. We argue that this tension
should be found in the microscopic picture if we include the twist operator interaction in the
microscopic CFT, something we have not done in our leading order computation.
An indirect approach to computing the tension of the black hole branch was taken in [25,36],
and we extend this approach here. One computes the tension from the gravity description, and
uses this to read off the changes in excitations in the microscopic description. In the microscopic
description we can imagine integrating out the effects of the KK monopole pairs (which are only
weakly excited at large L) to obtain an effective interaction between the P , P excitations of
the black hole. We first use the leading order gravity result to find this interaction to leading
order in the microscopic picture. Then we use analyticity in the microscopic variables to predict
the tension to next order on the gravity side, and observe agreement with the known gravity
computation at that order.
We thus obtain some understanding of tension in the black hole phase. We note some agreements and also some disagreements between the gravity phase diagram and the microscopic
diagram. We hope to return in future work to exploring further the interactions in the microscopic picture in order to obtain a more accurate phase diagram.
2. Some basic relations
2.1. The Smarr relation
We recall the Smarr relation [7], and sketch a derivation; derivations along these lines can be
found for example in [8].
The full theory of quantum gravity has an inbuilt length scalethe Planck length
lp . But lp
involves h , and the classical theory does not have this length scale. The action
gR does not
306
have any natural length scale, so if we have one solution to the vacuum equations, we can obtain
another by uniformly scaling up all lengths by the same factor. Let the length of the compact
direction z change as
L L + dL.
(2.1)
Consider a black hole in a spacetime with one compact direction z of length L. Let the number of
non-compact spacetime dimensions be d; thus including the compact circle z the total spacetime
dimension is d + 1. In this dimension the mass M scales as Ld2 . (The mass appears in the
(d+1)
(d+1)
metric through the combination GN M, which has units Ld2 ; GN
is held fixed, and thus
d2
M scales as L .) Thus we will get

dL d2
dL
M M 1+
(2.2)
,
dM = M(d 2)
.
L
L
The entropy is proportional to the surface area of the horizon. The horizon is a (d 1)dimensional surface, so it will scale as

dL
dL d1
.
,
dS = S(d 1)
S S 1+
(2.3)
L
L
The charge Q appears in the metric the same way that the mass does, so we have

dL d2
dL
QQ 1+
.
,
dQ = Q(d 2)
L
L
(2.4)
Now assume that the solution satisfies the first law of thermodynamics (this will be true for
all black objects)
dE = T dS + T dL + dQ,
(2.5)
where T is the tension in the z direction and the potential at the horizon. This law will be
satisfied in particular for the changes (2.1)(2.4), since these changes lead from one valid solution
to another. We thus get
E(d 2) = T S(d 1) + T L + Q(d 2)
(2.6)
or
(d 2)(E Q) T L
.
d 1
This is the Smarr relation.
TS =
(2.7)
2.2. Behavior under boosting

Consider the metric of a black object, by which we mean any solution of gravity with a single
connected horizon. We assume that the solution is neutral, i.e. it carries no charge. The metric is
of the form
ds 2 = U dt 2 + dsB2 ,
(2.8)
where U is independent of t and the base metric dsB2 involves only the spatial coordinates. The
horizon is at the location U = 0. We can add one or more trivial directions to this metric; we
307
single out a special one called y for later use:

ds 2 = U dt 2 + dsB2 + dy 2 + dza dza .
(2.9)
We will think of the added directions as compact circles. In particular let 0 y < Ly .
We can start with the solution (2.9) and add charges to it by boosting plus dualities. Thus
writing
t = t cosh + y sinh ,
y = t sinh + y cosh
(2.10)
we get

ds 2 = H 1 U dt 2 + H 2 (dy At dt )2 + dsB2 + dza dza ,
(2.11)
where
H = 1 + (1 U ) sinh2 ,

At = H 1 1 coth .
(2.12)
This process adds momentum charge Py . We can change this to a different kind of charge
by S, T dualities, and add further charges by boosting. We will write down a 2-charge solution in
Appendix A, but for now we will look at the simple case having just Py charge. We can extract
the following properties of the boosted solution; the derivations can be seen from the 2-charge
case discussed in Appendix A.
Let the neutral solution (2.9) have mass M, entropy S, and temperature T . (The temperature
is extracted from the surface gravity at the horizon in the usual way.) The part dsB2 describes a
(d 1)-dimensional non-compact space and a compact direction z, and we will let T denote the
tension in the z direction.
What are the properties of the charged solution (2.11)? Note that y was a compact direction,
and we do not have the boost symmetry (2.10) for compact y. But since the classical solution is
homogeneous in y, we can lift the solution to the covering space of y, apply the boost, and then
recompactify the new coordinate y . Since the boosted solution is again homogeneous in y ,
we can pick any coordinate length to compactify y . Let us choose 0 y < Ly , thus keeping the
length of the compact circle the same as in the neutral solution. Then we find the following [17]:
(a) The mass M of the charged solution (2.11) is given by

d 2n
M = M 1 +
(2.13)
sinh2 ,
d 1
where the relative tension n is defined as
LT
n
.
M
(b) The tension T of the charged solution is the same as that of the neutral solution
T=T.
(2.14)
(2.15)
(c) The charge of the solution is

Q=M
d 2n
sinh cosh .
d 1
(2.16)
(d) The entropy S of the charged solution is

S = S cosh .
(2.17)
308
(e) The temperature T of the charged solution is

T =
T
.
cosh
(2.18)
Note that
T S = T S.
(2.19)
(f) The electric potential at the horizon is

= tanh .
(2.20)
2.3. Understanding the boost relations

The above relations (a)(f) between the neutral and charged solutions are straightforward to
derive in the gravity description of black objects. But our goal is to understand black objects
from a microscopic point of view. From this matter point of view these relations appear somewhat nontrivial. They tell us the thermodynamic properties of a charged system once we know
the properties of the neutral system. But abstractly speaking, there is no general relation between
charged and neutral systems. So we must be looking at very special systems, and our microscopic model must be made to reproduce the properties of such special systems. To understand
these properties better, we now rederive the relations of the above section from a microscopic
perspective.
2.4. The matter stress tensor
Consider the neutral system. Let the total spacetime dimension be D. The Einstein equations
are
1
RAB gAB R = 8GTAB .
2
We can rewrite these as
RAB = 8GTAB ,
(2.21)
(2.22)
where
1
gAB T .
(2.23)
D2
We will now assume that the gravitational field is weak. This will not be the case in general for
the systems we finally consider, but we are only trying to get a physical feeling for the relations
(a)(f) which are already known to be correct, and this weak-field assumption will allow us
to extract some basic ideas in a simple way. Writing gAB = AB + hAB we get (in the gauge
hAB, B = 0)
TAB = TAB
1
1
hAB h,AB = 8GTAB .
(2.24)
2
2
Recall that we had started with a metric of the form (2.8) and added trivial directions to
obtain the metric (2.9). It seems from the gravity point of view that these extra directions play
no role in the physics of the black object. But from a microscopic viewpoint we see a different
309
story. Let the weak gravity system (2.24) describe this situation where we have trivially added
directions y, za , and the metric is nontrivial in the non-compact directions x and the compact
direction z. Consider one of the trivial directions, say y. First, we have homogeneity in y, so
fields do not depend on y. Thus in (2.24) we will get h,yy = 0. Second, we have uniformity of
the y circle, which means that the size of this circle does not change from place to place. This
gives hyy = 0. Eq. (2.24) then gives Tyy = 0, which is
1
1
T = Tyy
[T00 + Tzz + Tyy + Tza za ] = 0.
(2.25)
D2
D2
Note that T = 0 in general, so Tyy = 0 in general. This means that our microscopic model cannot
consist of branes that extend only in the directions xi , z while ignoring the trivial directions y,
za . Branes have a tension only along their worldvolumes, and such a model would give Tyy = 0.
Instead, we must have at least one kind of brane wrapping each of the trivial directions. As an
example consider the Schwarzschild solution in 4 + 1 non-compact dimensions. We can add the
other 5 directions of string theory as trivial directions in the gravity solution. But a microscopic
model for this system can be written in terms of branes and antibranes of type D1, D5, P [30],
and these objects are seen to wrap around the 5 trivial directions.
The mass of the neutral system is

M = [dxi dz dy dza ]T00 .
(2.26)
Tyy
The tension T along the z direction is

[dxi dz dy dza ]Tzz
T =
.
L
(2.27)
First note that under the boost (2.10) we have Tzz = Tzz , so the tension remains unchanged
T=T
(2.28)
which is the relation (2.15). Next, note that

= [dxi dz dy dza ] T00 cosh2 + Tyy sinh2 .
M = [dxi dz dy dza ]T00
(2.29)
In (2.25) the trivial directions y, za are all on the same footing, so Tyy = Tza za . We then find
1
(T00 Tzz ).
d 1
Substituting this in (2.29) we find
Tyy =
M = M cosh2
1
d 2n
(M + T L) sinh2 = M + M
sinh2 ,
d 1
d 1
where we have used the definition n = T L/M. Thus we reproduce (2.13).

The charge Py is

Q = [dxi dz dy dza ]T0y

= [dxi dz dy dza ](T00 sinh cosh + Tyy sinh cosh ).
(2.30)
(2.31)
(2.32)
310
Using (2.30) we find

Q=
d 2n
(d 2)M T L
sinh cosh = M
sinh cosh
d 1
d 1
(2.33)
which agrees with (2.16).

2.4.1. Entropy
Now consider the entropy S. The neutral system was homogeneous in the direction y; we used
this fact in lifting to the non-compact covering space before boosting. The classical entropy is
linear in the coordinate length of the y circle, since the area of the horizon is proportional to this
length. Thus the entropy of the microscopic model must be extensive in the length Ly .
Under boosting
along y we would expect that the length of the y direction would contract by
a factor = 1/ 1 v 2 = cosh ; thus it would fit in a length Ly = Ly / cosh . We could have
chosen any length of recompactification after boosting, but our choice was to keep Ly unchanged.
This means that we have considered a larger amount of matter in the charged caselarger by a
factor cosh . This accounts for the relation
S = S cosh .
(2.34)
2.4.2. The relation = tanh

The potential of the charged system is defined as

M
=
.
Q S
(2.35)
Setting dS = 0 in (2.34) we find

S sinh d +
dS
cosh dM = 0.
dM
Then from (2.31) we find (using (2.36) to eliminate d)

dn
d 2 n M dM
dS M d 2 n

2
2
cosh +
sinh .
dM = dM 1 2
dM S d 1
d 1
(2.36)
(2.37)
From (2.33) we find

dn
cosh dS M d 2 n d 2 n M dM
+
sinh cosh .
dQ = dM cosh 2
sinh dM S d 1
d 1
(2.38)
As it stands, the ration dM /dQ does not simplify. But now we use the Smarr relation for the
neutral system
TS =M
dS M d 2 n
d 2n
= 1.
d 1
dM S d 1
(2.39)
Then we find
=
dM
= tanh .
dQ
(2.40)
2.4.3. The relation T S = T S

We have

dM
T =
.
dS Q
311
(2.41)
Setting dQ = 0 in (2.33) we find

d =
dn
sinh cosh
dM d 2 n M dM
.
M
d 2n
sinh2 + cosh2
Using this we find

dn
d 2 n M dM
sinh2
,
dM = dM 1
d 1
sinh2 + cosh2
dS =
dn
d 2 n M dM
dS
sinh2
dM
cosh .
M
S
M
dM
d 2n
sinh2 + cosh2
(2.42)
(2.43)
(2.44)
As it stands the expression

dM
S
dS
does not simplify. But if we use the Smarr relation (2.39) we find
T S =
(2.45)
T S = T S.
(2.46)
2.5. The near extremal system

Let us now note some results that will be useful in our study of the near extremal system that
we will use. We will for the most part use D1, D5 charges, but will also have occasion to consider
the system with three nonzero chargesD1, D5, P .
2.5.1. Two charges
Let us take the solution (2.9) and add D1D5 charges. We find the solution (we give the
derivation in Appendix A)

1/4 1/4 1
dsE2 = H1 H5
H1 U dt 2 + dy 2 + H5 dsB2 + dza dza ,
G(3) = U 1/2 r H5 coth 5 B dr r H11 coth 1 dr dt dy,
H1
, H1 = 1 + (1 U ) sinh2 1 , H5 = 1 + (1 U ) sinh2 5
e2 =
H5
(Here dsE2 is the Einstein metric.) In this case the analog of (2.31) is

2 n
M = M 1 +
sinh2 1 + sinh2 5 .
3
(2.47)
(2.48)
Since we will be working close to extremality, let us compute the energy above extremality
E M Mex = M Q1 Q5 .
(2.49)
312
The charges Q1 , Q5 are given by the analog of (2.33)

2n
2n
sinh 1 cosh 1 ,
sinh 5 cosh 5 .
Q5 = M
3
3
In the near extremal limit we have 1 , 5
1, so that
Q1 = M
1
sinh i cosh i sinh2 i + ,
2
Using this in (2.49) we get
i = 1, 5.
(2.50)
(2.51)
1+n
.
(2.52)
3
For the near extremal case it is convenient to define the relative tension in a manner similar to
the definition (2.14) but with the mass replaced by the mass above extremality [17]
EM
LT
.
E
From (2.15) and (2.52) we find
r
r=
LT
LT M
3n
=
=
.
E
M E
1+n
(2.53)
(2.54)
We can similarly find the transformation law for the entropy. With two charges the analogue
of (2.17) is
S = S cosh 1 cosh 5 .
(2.55)
For later use let us define a rescaled entropy as

S
S
.
Q1 Q5
(2.56)
From (2.55) and (2.50) we find

cosh 1 cosh 5
S 3
S =
.
M 2 n sinh 1 cosh 1 sinh 5 cosh 5
(2.57)
In the limit of large charges 1 , 5

1 we get
S 3
.
S =
M 2n
(2.58)
2.5.2. Three charges

We will also have occasion to turn on 3 charges, D1, D5, P . We will have the D1, D5 charges
large as above, but the P charge will be arbitrary. The relation (2.48) now becomes

2 n
sinh2 1 + sinh2 5 + sinh2 p
M = M 1 +
(2.59)
3
and we have
Qp = M
2n
sinh p cosh p .
3
(2.60)
313
We define the energy above extremality E again as the energy above the mass of the large charges
D1, D5, and find
2n
1+n
+M
sinh2 p .
(2.61)
3
3
For later use, it will be helpful to write this relation in another way. For the neutral system (i.e.
before any charges were added) we have the Smarr relation
E M Q1 Q5 = M
TS =M
2n
.
3
(2.62)
This implies
3
nM
3
TL
M = TS +
= TS +
.
2
2
2
2
Substituting this in (2.61) we find
E=
TS
TL
cosh 2p +
.
2
2
(2.63)
(2.64)
3. Phase transition: Microscopic model

We are looking at type IIB string theory with the compactification M9,1 M4,1 T 4 S 1
1
S . On these 6 compact directions we can wrap 4 mutually BPS charges:

(i)
(ii)
(iii)
(iv)
D5 branes wrapped on T 4 S 1 .
D1 branes wrapped on S 1 .
Momentum modes P that along S 1 .
KK-monopoles that have S 1 as their nontrivial fiber, and that extend uniformly along
T 4 S1.
The four charges above can be permuted in all possible ways among themselves by S, T
dualities. For this reason we will often list the number of these charges as n1 , n2 , n3 , n4 , instead
of writing n5 , n1 , np , nk .
Work on black hole microscopics has given an understanding of the entropy of black objects
for various compactifications. Let us recall these results.
First let the length L of S 1 be infinite, so that we have only the compactification T 4 S 1 .
Then the mass of the KK charge becomes infinite, and it drops out of the computations and the
entropy is an expression in the three remaining charges n1 , n2 , n3 . The entropy for all such cases
can be derived from the following abstract expression [30]

n2 + n 2
n3 + n 3 .
S = 2 n1 + n 1
(3.1)
There are 6 variables on the right-hand side. We have to vary these variables to achieve a maximum of S, subject to the constraints that we are given the net values of the 3 types of charges
n i = ni n i ,
i = 1, 2, 3
(3.2)
and the total energy

M=
3

i=1
mi (ni + n i ),
(3.3)
314
where mi is the mass of each quantum of type i. Roughly speaking, extremising S tells us that
we create pairs of those charges that are lighter, and that have smaller net charge values; i.e., we
create pairs of the charges that we do not have. In particular, if some charge n i , then the
non-extremal energy will not create pairs for this charge.
Let us list those cases that will be relevant to us in what follows:
(a) Let there be 2 large charges n1 , n2 , n 3 = 0, and a small amount of non-extremal
energy E. Then we find
n3 = n 3 =
E
2m3
(3.4)
and

S = 2 n1 n2 n3 + n 3 = 4 n1 n2

E
.
2m3
(3.5)
(b) Let there be one large charge n1 , n 2 = n 3 = 0, and a small amount of non-extremal
energy E. Then we have to extremise

n3 + n 3 = 8 n1 n2 n3
S = 2 n1 n2 + n 2
(3.6)
subject to
E = 2m2 n2 + 2m3 n3 .
(3.7)
We find
n2 = n 2 =
E
,
4m2
n3 = n 3 =
E
4m3
(3.8)
and
E
S = 2 n1
.
m2 m3
(3.9)
Now let S 1 be finite, so that all 4 charges are relevant. The analog of (3.1)(3.3) are [31]

n2 + n 2
n3 + n 3
n4 + n 4 ,
S = 2 n1 + n 1
(3.10)
n i = ni n i ,
M=
i = 1, 2, 3, 4,
mi (ni + n i ).
(3.11)
(3.12)
i=1
The analogues of cases (a), (b) above are

(a) We have n 1 , n 2 , n 3 , n 4 = 0. We get
n4 = n 4 =
E
2m4
and
S = 4 n1 n2 n3
(3.13)

E
.
2m4
(3.14)
315
(b) We have n 1 , n 2 , n 3 = n 4 = 0. We get

n3 = n 3 =
E
,
4m3
n4 = n 4 =
E
4m4
(3.15)
and
E
S = 2 n1 n2
.
m3 m4
(3.16)
Consider any of these expressions, say (3.5). We can regard the entropy as arising from fractional np , n p pairs. We have n1 n5 np units of fractional P excitations, and n1 n5 n p units of
fractional P excitations. Our interest is in understanding more about how fractionation works.
We will therefore examine different possible fractionations below, and see how well they fit the
physics of the black holeblack string transition.
3.1. Modeling the phase transition
Let us now come to the main computations of this paper. We have seen that in the limit of
large L and small L the entropy of the system is given by (1.2) and (1.4), respectively. We
now wish to make a simple model for the transition between these two possibilities. We have
two large charges D1, D5 (with integer values n1 , n5 respectively); we have added these to the
system to make the system near-extremal. On the gravity side the addition of charges induced
a simple map between neutral and charged systems, so no generality was lost by looking at this
near-extremal case.
Let the energy above extremality be E. We let a part E1 of this energy go to exciting P ,
P pairs. If these were the only excitations present then they would describe a black hole in
4 + 1 non-compact dimensions, and give an entropy of type (1.2), which we have rewritten in the
form (3.5) above. The rest of the energy E E1 will go to entropy of the form (1.4); we have
rewritten this entropy in the form (3.16) above. If all the energy was in this latter form we would
get a uniform black string. The important question now is what happens to the charges n1 , n5 .
There are three possibilities:
(a) A fraction (1 x) of the charges n1 go to the entropy of type (3.5), and the rest x go to the
entropy (3.16); similarly a fraction (1 x) of the charges n5 contribute to (3.5), and the rest x
contribute to (3.16). (We take the fractions x equal for the two charges by the symmetry between
these charges, but we can consider different fractions too.) Thus we have two disjoint systems,
and the fractionation available for the two types of entropy are x 2 n1 n5 and (1 x)2 n1 n5 ,
respectively.
(b) A fraction (1 x) of the product n1 n2 goes to the entropy (3.5) and a fraction x to the
entropy (3.16). Thus the D1 and D5 branes make a bound state with n1 n5 effective degrees of
freedom, and it is these effective degrees of freedom that get partitioned into the two subsystems.
It is useful to define
N = n1 n5
(3.17)
and the partitioning gives n1 n5 = N N to the entropy (3.5) and n1 n5 = N to the entropy (3.16).
(c) All the charges n1 and all the charges n5 contribute to each kind of entropy. Thus the full
product n1 n5 appears in each of the expressions of type (3.5) and (3.16).
316
It turns out that possibility (b) gives a phase diagram that resembles the gravity computation,
and we will adopt this choice below. We will discuss what happens with possibilities (a), (c) at
the end of the section.
Thus adopting possibility (b) we write the entropy as

E1
E E1
+ 2 N
S = 2 2 N N
mp
mk mp

E E1
2 mk
E
.
=
2 N N
+ N
(3.18)
mp
mk
mk
If we introduce the dimensionless quantities
E
,
mk
1 =
E1
,
mk
x=
N
N
we can rewrite the entropy as

2 N mk
s [
1 , x;
],
S=
mp
where
s [
1 , x;
] =
(3.19)
(3.20)
2 1 x
1 + x(

1 ),
(3.21)
where for a given

we should extremize in
1 , x.
Let us now study the properties of the function s for different
. The extrema of
occur at
points where
s
= 0,
s
= 0.
x
(3.22)
There is only one such point:

1
.
2
1
This point lies inside the allowed region
1 [0,
], x [0, 1] for
1 =
1,
x=
1.
(3.23)
(3.24)
The matrix of second derivatives at the extremal point is

(2
1)5/2 (2
1)3/2
1
2 s
8
4
=
.
3/2
1
(x,
1 )
1 (2
1)
1/2
4
(3.25)
2(2
1)
Its determinant is
det
2 s
(2
1)2
=
(x,
1 )
8(
1)
(3.26)
which is negative for

> 1. Thus the extremal point is a saddle point, which is a maximum along
some direction and a minimum along another direction. The value of s at the extremum point is
sc = 2
1.
(3.27)
(a)
317
(b)
Fig. 4. (a) The entropy is plotted on the vertical axis, as a function of x,

1 , for
= 0.5, a value below the point where
the three phase structure appears; there is only one maximum at the boundary of the parameter space; (b) The entropy
graph for
= 2.4, a value at which there are three phases; there are two maxima at the boundary of the parameter space
and a saddle point in the middle.
Since the extremum we have found is a saddle, the actual maxima of the entropy must occur
at the boundary of the allowed parameter space. We plot s against x,
1 in Fig. 4. When
< 1,
as in Fig. 4(a), there is one maximum at the boundary
1 =
,
(3.28)
x = 0 sa = 2
.
This corresponds to having all the energy and entropy in the mode of type (3.5) (we call this
mode (a)). When
> 1 we have the situation in Fig. 4(b), where there are two maxima at the two
ends. One as in (3.28), and the other at
1 = 0,
x = 1 sb =
.
(3.29)
This latter extremum corresponds to having all the energy and entropy in a mode of type (3.16)
(we call this mode (b)).
Both maxima sa and sb are always greater than sc .
To summarize, we have the following situation:
(1) For
< 1, there is only one maximum corresponding to the black hole phase, phase (a).
(2) For
> 1 we have three phases, that we interpret as follows:
(a) black hole phase:

1 =
, x = 0, sa = 2
;
(b) uniform black string phase:
1 = 0, x = 1, sb =
;
(c) non-uniform black string phase:

1 =
1, x = 1/(2
1), sc = 2
1.
Phase (c) is always unstable.
3.2. Computing the tension
The energy above extremality comes from P , P and KK, KK pairs
E = mp (np + n p ) + mk (nk + n k ) = 2mp np + 2mk nk .
(3.30)
Note that mk depends on L but mp does not. Thus the tension comes only from the excitations of
type (b), where both P and KK pairs are excited. Since mk depends quadratically on L we have
mk
mk
=2 .
L
L
(3.31)
318
Thus we get
mk
mk
E Eb
=
= (nk + n k )
= 4nk
.
L
L
L
L
Using (3.15) we get1
T =
E Eb
E Eb mk
=
.
4mk L
L
The rescaled tension, defined in (2.53), is
T L E E 1

1
=
=
.
r=
E
E
In the uniform black string phase

1 = 0 and thus the rescaled tension is
T =4
rb = 1.
(3.32)
(3.33)
(3.34)
(3.35)
In the black hole phase

1 = 0 and the tension vanishes
ra = 0.
(3.36)
In the non-uniform string phase

1 = 1 and our microscopic prediction for the rescaled
tension is
1
rc = .
(3.37)
Note that the value ra = 1 we have obtained for the uniform string tension is the one expected
from gravity (a fact that had already been noted in [30]). This can be seen as follows. The rescaled
tension of the neutral system in the uniform string phase is obtained by setting cz = 0 in (A.13),
and it is given by
1
nus = .
(3.38)
2
Substituting this value in (2.54) gives the rescaled tension for the charged uniform string:
rus = 1.
(3.39)
We thus see that ra = rus = 1.

3.3. Comparison between gravity and the microscopic model
In this subsection we compare various results from gravity with the predictions following
from our simple microscopic model. It will be convenient to introduce rescaled energies as
(5)
16GN
M,
L2
Using the fact that
L2
(5)
16GN
(5)
16GN
E.
L2
Rz2 Ry V
= mk
g2 4
(3.40)
(3.41)
we see that the quantity

defined here coincides with the one introduced in (3.19).
1 It is easy to check that the same result is obtained if we start with the definition T = ( E )
L S,Qi and compute the
requited variation of E keeping S, Qi fixed.
319
3.3.1. Mapping from the charged system to the neutral one

We would now like to compare the graphs for tension and entropy derived from model (b)
with the corresponding graphs obtained from gravity (shown in Figs. 2, 3). The graphs obtained
from the gravity computation describe the neutral system, while the results from microscopics
are for the charged system. We will thus map the variables in the microscopic computation to the
corresponding neutral system, and then plot the results.
The relations required to effect the needed map were given in Section 2.5.1. Eq. (2.54) relates
the relative tension r of the charged case to the relative tension n of the neutral case. Eq. (2.52)
reads as
= 1+n
3 in scaled variables, and relates the energy above extremality of the charged
case to the mass in the neutral case. The entropy of the charged case is S , but it is more convenient to look at S defined in (2.56), which is related to the quantities of the neutral case by (2.58).
It turns out that S exactly equals the quantity s that we have used in our microscopic analysis.
Checking this needs the exact constants relating the Qi to the corresponding ni , but even without
doing this we note that the ratio between S and s is a constant, and this constant will cancel when
we plot ratios of entropies.
Carrying out this map to the neutral system, we find the following for the three phases:
(a) black hole phase:
n = 0,

S
4 2 3/2
=
;
Sus 2 3
(3.42)
(b) uniform string phase:

1
n= ,
2
S
= 1;
Sus
(c) non-uniform string phase:

S
4 2 1 3/2
1
=
.
n= ,
Sus 2
3
(3.43)
(3.44)
3.3.2. Tension and entropy graphs

With the above map to the neutral system we plot the tension and entropy graphs obtained
from the microscopic model in Figs. 5, 6, respectively. We should compare these graphs to the
graphs of Fig. 2(b) and Fig. 3(b), respectively. The first feature we observe is that the microscopic
model exhibits three phases, and the tension and entropies of these phases are in the correct order
Fig. 5. The phase diagram predicted by the leading order microscopic model. The solid line is the uniform black string.
The black hole branch is a horizontal (dashed) line that overlaps with the x-axisthe tension is zero. The dot-dashed
line is the non-uniform string branch.
320
Fig. 6. The entropy graphs for the three phases, with the same coding for the lines as in Fig. 5.
when compared to the gravity graphs; i.e., the tension of the non-uniform string branch is lower
than the tension of the uniform string and higher than the tension of the black hole, and the
entropy of the non-uniform string is always lower than the entropy of the other two phases.
We will now compare the microscopic results to the gravity results in more detail. We note
that the microscopic model gives vanishing tension for the black hole phase while the gravity
computations do not; we will discuss this in the next section.
3.3.3. The GregoryLaflamme point
Let the GregoryLaflamme (GL) point in the phase diagram be the point where the nonuniform string branch starts off from the uniform string branch. In the microscopic model this
point is at
GL = 1,
rGL = 1.
(3.45)
Mapping this to the neutral system we get

GL = 2
GL = 2.
(3.46)
3.3.4. Transition energy

On the gravity side, the energy at which the GL transition happens has been determined
numerically in [2,5,13]. The value of this energy depends on the number of non-compact dimensions, and for d = 4 non-compact dimensions it is given by
(5)
GL
16GN
MGL 3.52.
L2
(3.47)
Since the microscopic model gives GL = 2, we find

(grav)
GL
(micro)
GL
= 1.76.
(3.48)
Thus we do not get perfect agreement on the GL point in our leading order microscopic model.
But we will see below that once we scale energies so that they are all compared to the GL point,
we get good agreement for the tension graph of the non-uniform string.
321
3.3.5. Non-uniform string tension near GL

In [9], from an analysis of the d = 5 numerical solution of [5], it was found that the following
relation describes with good accuracy the non-uniform string phase near to the GL point:

M
TS
1 = x(d)
1 .
(3.49)
TGL SGL
MGL
Let us assume that an analogous relation is valid for a generic dimension d, with a dimensiondependent constant x(d). One can combine this relation with the Smarr identity2
d 2n
d 3
M,
TGL SGL =
MGL
d 1
d 2
to obtain a relation for the rescaled tension
TS =
MGL (d 2)2 x(d)(d 1)(d 3)

(d 1)(d 3)
x(d) 1
+
.
d 2
M
d 2
The M-independent term on the r.h.s. of the above equation vanishes if
n=
(d 2)2
.
(d 1)(d 3)
For d = 5, the numerically derived value for the constant x is
(3.50)
(3.51)
x(d) = x(d)
(3.52)
x(5) = 1.12.
(3.53)
We note that this number is very close to

9
= 1.125.
(3.54)
8
Thus for d = 5 the M-independent term in (3.51) vanishes with good accuracy. We do not know
the numerical value for x(4). It seems however reasonable to conjecture that the M-independent
term vanishes also for d = 4. This happens if
x(5)
4
1.33.
(3.55)
3
We can support this conjecture by expanding (3.51) in powers of M MGL : The first term in this
expansion is known, for d = 4, from the work of [2]:
x(4) = x(4)

1
( GL ) + O ( GL )2 ,
2
Expanding (3.51) we find, for d = 4
n=
n=
= 0.14.
MGL
GL
3
1 3
3
x(4) 1
+ 2 x(4) = x(4) 1
2
M
2
2 2
GL

2
+ O ( GL ) .
(3.56)
(3.57)
Comparison of (3.56) and (3.57) gives

=
3 x(4) 1
2
x(4) = GL + 1 1.33.
2 GL
3
(3.58)
2 In the second equation in (3.50) we have used the value of the uniform string tension in d-dimensions: n = 1/(d
us
2).
322
This confirms, with good accuracy, that x(4) is such that the M-independent term in (3.51)
vanishes. If we use x(4) = x(4)
in (3.51), we then have

n=
1 MGL 1 GL
=
.
2 M
2
(3.59)
In the microscopic model we have found that for the non-uniform string n = 1/. Noting that
GL = 2 in this model (Eq. (3.46)), we can write for the microscopic model
n=
1 GL
2
(3.60)
which is in perfect agreement with (3.59).

3.3.6. The Smarr relation
On the gravity side all black objects satisfy the Smarr relation. In the microscopic model,
the black hole and uniform black string phases correspond to the 4 + 1 dimensional black hole
and the (3 + 1)-dimensional black hole respectively, and the entropies S(E, Qi ) in each case are
known to agree with the gravity computation. Thus the Smarr relation is satisfied for these two
phases.
For the non-uniform string, the microscopic model gives
Sn-us (E, N ) = Sus (E1 , N1 ) + Sbh (E2 , N2 ),
(3.61)
where E = E1 + E2 , N = n1 n5 = N1 + N2 .
It is easy to check that if a system is made of two non-interacting subsystems, i.e.,
E = E1 + E2 ,
Qi = Qi,1 + Qi,2 ,
S = S1 + S2
(3.62)
and if each subsystem satisfies the Smarr relation, then the total system will satisfy the Smarr
relation. Thus it is consistent to identify the saddle point of the microscopic model with the
non-uniform string black string.
Even though satisfying the Smarr relation appears trivial in this leading order microscopic
model, it will be nontrivial if we look at higher order corrections where we need to add interactions that link the two different modes of excitation. We will then not have (3.62), and the
interactions will have to be chosen such that the Smarr relation continues to be satisfied. Note
that a generic field theory model will not satisfy the required Smarr relation. The dimension d of
the spacetime appears in (2.7), and this dimension is in general different from the dimension of
the dual field theory. Thus the Smarr relation appears to be nontrivial from the viewpoint of the
microscopic model.
3.4. Other models of fractionation
At the start of Section 3.1 we had noted three different possible models of fractionation. We
analyzed model (b) above in detail; here we note what happens if we look at models (a) and (c).
3.4.1. Model (a)
The analog of (3.21) is
s = 2(1 x)
1 + x(

1 ).
(3.63)
323
Fig. 7. The tension graph for model (a). Note that the non-uniform string branch starts off at = 0, which is not in accord
with the gravity diagrams (Fig. 2).
Fig. 8. The phase diagram for model (c). There is only one phase for each , which does not agree with the gravity
diagrams (Fig. 2).
Extremising in x,
1 gives the point
2
1
1
x=
(3.64)
1 + 2
1 .
,
1 =
2
1 + 2
The determinant of the matrix of second derivatives is

2

1
2 s
= 1+
det
(3.65)
.
(x,
1 )
2
1
Since this is negative, we have a saddle point, and the actual maxima are again at the endpoints
of the domain of parameters. The black hole branch is at (x = 0,
1 =
), and the uniform black
string branch is at (x = 1,
1 = 0). But note that the non-uniform black string branch, given by
(3.64), is present for all values of
. The tension versus mass graph is given in Fig. 7. We see that
the point where the non-uniform black string branch meets the uniform black string branch is at
= 0, rather than at some positive value of

. Thus the phase diagram has a qualitative difference
from the gravity diagram.
3.4.2. Model (c)
In this case the analog of (3.21) is

s = 2
1 + (

1 ).
(3.66)
Setting s /
1 = 0 gives
1 =
1
2
(3.67)
and 2 s /
12 = 1 < 0. Thus this point is a maximum when it can be attained. Thus for
< 1/2
the maximum is at
1 =
, while for
> 1/2 the maximum is at
1 = 1/2. The phase diagram is
sketched in Fig. 8. We see that there is only one phase for all
, so we have a qualitative difference
from the gravity picture.
324
3.5. A physical picture of model (b)

Having seen above that models (a), (c) give phase diagrams that are less in accord with gravity than model (b), let us look at why a model like (b) is suggested by what we know about
microscopics. Consider first the case where we have the compactification M4,1 T 4 S 1 , and
let there be just one large charge. We let this charge be n5 units of D5; the excitations can then
be (fractional) pairs of D1 and P charges.
First suppose that only one kind of charge pairs is excited. We can let this be P P . But this
situation is very similar to exciting P P pairs on a bound state of D1 branes; the D5 wrapped
on T 4 acts like an effective string in 6D and the P P excitations run on this effective string. The
picture of this bound state is simple: the strands of the effective string join together to make one
long string, and the P P excitations are transverse vibrations of this long string. The vibrations
cause the strands of the long string to separate from each other and spread over some transverse
region.
Now consider the excitation mode where two kinds of charge pairs are excited. This situation
was studied in [29], and it was found that the entropy was exactly reproduced by assuming that
the energy of excitation is carried by a fractional tension D1 brane living in the D5 worldvolume; the tension of the D1 is fractionated by the factor 1/n5 . The string excitations represent
both P P pairs and D1D1 pairs. The entropy of such a string is given by (3.9). It seems plausible
that to get the required fractionation of the tension the D5 branes must be all stuck together,
and the D1 should be dissolved in this D5 bound state to give rise to the effective string with
fractional tension.
Thus we see that in the mode where one kind of charge pairs are excited the D5 branes
separate from each other in carrying the excitation, while in the mode where we have two kinds
of excitations we need the D5 branes to be together. If we are in the process of transition from one
mode to the other, then we will have some excitation in each of these modes. But this suggests
that we will need to partition the D5 branes into two sets: A fraction (1 x) which will contribute
to the first mode and a fraction x which will contribute to the second mode.
Now return to our actual problem, where the compactification is M3,1 T 4 S 1 S 1 and
instead of the D5 charge we have D1D5 charge. When we have the mode with only one kind of
charge pair the entropy is given by (3.5), where the effective string has winding n1 n5 instead of
n5 . When we have two kinds of charge pairs, we just note that by dualities we can map this to
near extremal D5KK, which was also studied in [29]. It was found that we again get a fractional
tension effective string living in the worldvolume made by the D5 and KK branes; all one has to
do is replace n5 with n5 nk , which in our duality frame corresponds to replacing n5 everywhere
with n1 n5 .
Thus extending our intuition from the D5 charge case to the D1D5 charge case we expect that in the phase transition we will need to partition the product n1 n5 into two parts:
a fraction (1 x) for the first mode and a fraction x for the second mode. This is just
model (b).
3.6. Microscopic model at large momentum
So far we have added only two large charges, D1, D5, to bring the system near to extremality.
But we can also add a large P charge by another boost. In this subsection we consider how the
black holeblack string phase transition will look in the near extremal 3-charge system.
325
The transition happens when the horizon radius parameter r0 becomes comparable to the
compactification length L. We have now chosen all the charges D1, D5, P large, so the charge
radii for these charges are all much larger than L. We can thus assume that we are looking at a
situation with three nonzero charges in 3 + 1 non-compact dimensions, with a small amount of
non-extremality. Since the 4 possible charges for this compactification can be freely permuted by
dualities, we can equally well look at a system where the nonzero charges are D1, D5, KK, and
the non-extremality then creates pairs of fractional P P excitations.
This system has been studied before. If the numbers of the three nonzero charges are
n1 , n2 , n3 , then we get an effective string with total winding number n1 n2 n3 . The kind of entropy that we will encounter here was studied in [37].3 If the system is extremal, then we get
entropy from partitioning the effective string in all possible ways into component strings, getting
an entropy S = 2 n1 n2 n3 . If there is some non-extremal energy then pairs of P P excitations

can be created on this effective string in the following way. A fraction f of the effective string
winding creates entropy as
before by breaking up into component strings of different length; this
gives an entropy S = 2 n1 n2 n3 f . The remainder of the effective string joins up into one long
component string (thus having no entropy of its own) but this enables the P P excitations
to be
E
, by
fractionated in units of 1/(n1 n2 n3 (1 f )). This gives an entropy 4 n1 n2 n3 (1 f ) 2m
4
(3.14). We then extremize over f to find out how the effective string will partition itself into its
two kinds of roles.
Let us apply this 3-charge model to our present problem, calling the result model A. The
entropy is thus

E
S = 2 f N + 4 (1 f )N
(3.68)
= 2 N f + 2(1 f )
,
2mk
where N = n1 n2 n3 , and we have again defined
= E/m4 = E/mk (we have reverted to our
duality frame where the nonzero charges are D1, D5, P and the excitations are pairs of KK). The
entropy S has only one extremum, as a function of f , given by
1
2
1
(3.69)
.
=0f =

2
+ 1
f
1f
It is easy to check that this extremum is a maximum. Thus we find a unique maximum for
every value of the energy: According to model A the complicated phase structure we have seen
emerging in Section 3.1 disappears, at large P , and we are left with a unique stable phase for
every E. The value of the entropy at the maximum is

E
+ 1.
Smax = 2 N 2
+ 1 = 2 N 2
(3.70)
mk
We note that this entropy agrees with the extremal 3-charge black hole entropy [39]

Sbh = 2 N
(3.71)
for small energies (

1), and with the near-extremal 3-charge uniform black-string entropy
(Eq. (3.14))

Sus = 2 N 2
(3.72)
3 A review is given in [38].
326
for large energies above extremality (

1). We can also compute the value of the rescaled
tension r predicted by this model. The rescaled tension is defined as

L E
LT
=
r=
(3.73)
,
E
E L S,Qi
where E is the energy above extremality (so that the total energy is given by M = Q1 + Q5 +
Qp + E). Consider the entropy given by (3.70). Since mk L2 , we find that S is kept constant
if the variations of E, L satisfy
E = 2E
L
.
L
(3.74)
Thus
L 2E
= 2.
(3.75)
E L
This agrees with the result from the gravity computation when we have the extremal three large
charges limit. In [25] it was shown that in this limit the r vs.
graphs for the three phases in the
gravity computation all degenerate4 to the line r = 2. Thus, as far as the tension is concerned,
model A gives a prediction which is consistent with gravity.
But this analysis may also look a little puzzling, for the following reason. In the gravity computation the three phases collapse to one horizontal line in the r
diagram, but there should
still be three overlapping phases for each
in the range where the non-uniform string exists. On
the other hand the analysis of model A gives a unique maximum for the entropy for all
, so there
seems to be only one phase.
To emphasize the puzzle still further consider the model (3.21) that we have analyzed in the
case of two large charges D1, D5, and where we did get three phases for a range of
. In this
model we can let np = n p , thus getting a nonzero P charge. All the results derived above for
this model can be extended to the nonzero P case by changing np , n p in the following way to
account for the P charge:
r=
np np e ,
n p n p e .
(3.79)
The limit of large P is given by taking

1. Let us denote the microscopic model obtained in
this way as model B. From the point of view of model B, it is clear that there is a one to one map
4 This can be seen as follows. For the 3-charge system the energy M and the charges are given by

2n
sinh2 1 + sinh2 5 + sinh2 p ,
M = M 1 +
3
2n
Qi = M
sinh i cosh i , i = 1, 5, p.
3
(3.76)
Taking all the charges to be large (i

1) one finds
M = Q1 + Q5 + Qp + M
n
n
E=M .
2
2
(3.77)
Since the absolute tension of the neutral system is the same of the one of the 3-charge system, the rescaled tension is
r=
M
n = 2.
E
(3.78)
327
between the phase diagram we have found at P = 0 and the one at P

1, given by (3.79). In
particular model B still predicts three phases (which all degenerate to the line r = 2 as ).
We thus need to ask: If we go slightly off extremality, how does the single line in the tension
graph of model A split into three closely spaced lines? We expect that we will see this split if
we take into account interactions in the CFT which describes the near extremal 3-charge system.
This is a general feature of adding charges to a system; we simplify the system by going closer to
extremality, but we may need to take into account interactions between the degrees of freedom of
the system to recover all the physics. We will discuss this issue in more detail in the next section
when we talk about the need for interactions to understand the tension in the black hole phase.
4. Tension in the black hole phase
Our microscopic model gives a phase diagram that has several features of the phase diagram
obtained by numerically solving the equations of general relativity. But one important feature
that is missing is the tension of the black hole branch: The microscopic model gives T = 0
while the numerical results show a tension rising from zero as the mass is increased. This tension
was recently related to microscopics in [25]. In this section we discuss why this tension did not
show up in our calculation of the last section, and extract some properties of the tension from
microscopic arguments.
4.1. The source of tension in the black hole phase
4.1.1. The leading order gravity calculation
We first recall the origin of this tension in the gravity picture. A similar calculation can be
found for instance in [21].
Fig. 9(a) shows a black hole with radius much smaller than the compactification length. In
Fig. 9(b) we lift to the covering space, getting an infinite set of images of the black hole. Each
copy of the hole feels the attraction of all other copies, and this creates a negative potential energy
which is the effect of the compactification. At leading order we can just use the Newtonian
potential to compute this energy. In 4 + 1 dimensions the gravitational potential energy between
(a)
(b)
Fig. 9. (a) A small black hole in a space with a compact transverse circle; (b) Lifting to the covering space replaces the
compactification by an infinite set of images of the hole. Gravitational attraction between the images is given by graviton
exchange, which is a one loop of open strings in the dual channel, with all modes of the open string contributing.
328
two bodies of mass M1 , M2 at distance r is

(5)
4GN M1 M2
.
3
r2
The energy between a hole and all its images is
V =
4 GN M 2 1
4 GN M 2
2
=
,
3 L2
9
n2
L2
(5)
Varray =
(4.1)
(5)
(4.2)
n=1
where we have used that

1
2
.
=
6
n2
(4.3)
n=1
The part of this energy that we must attribute to the compactification length L is
(5)
Vint =
Varray
2 GN M 2
=
.
2
9
L2
(4.4)
The tension is given by T = ( E

L )S . Since the black hole horizon is not distorted to leading order,
the entropy is independent of L, and we can just compute the L derivative of Vint to obtain
(5)
4 GN M 2
(4.5)
,
9
L2
where the approximation symbol denotes the fact that we have just computed the interaction
energy at leading order in the size of the black hole.
The above calculation was for neutral holes, but we can obtain Vint for charged holes by
relating them to neutral holes via the relations of Section 2.2. We need large D1, D5 charges, and
finite P charge. Let E0 be the energy above extremality in the absence of the interaction Vint .
With no interaction the tension n vanishes, so that at leading order one has
LT
M
cosh 2p .
(4.6)
3
The tension is invariant under the addition of charges (Eq. (2.15)), so we just get (4.5) with M
obtained through (4.6)
2
(5)
G
E0
LT 4 N2
(4.7)
.
cosh p
L
E0
The momentum charge carried by the hole is

M
sinh 2p
3
(here P is the integer charge).
We can write for the charged hole

E
Vint
.
T =
=
L Q,S
L
Qp mp P
(4.8)
(4.9)
Note that the entropy changes in a known way under addition of charge by boosting (S
S cosh ). From this relation and (4.8) we see that (to the order we are working at) we keep
(a)
329
(b)
Fig. 10. (a) A string stretched between two well separated branes has a closely spaced tower of excitations; (b) When
a large brane charge is added to the system we only get the ground states of open strings, but the interactions between
these encode the original tower of excitations.
Qp , S fixed if we keep M, p fixed. Thus the tension is related to Vint only through variations of
L, and we find for the charged case a relation analogous to (4.4)
(5)
Vint 2
GN
L2
E0
cosh 2p
2
.
(4.10)
4.1.2. The effect in the dual field theory

Let us now ask how the energy (4.10) would be obtained in a field theory calculation. We
have seen that this energy is obtained from the long range gravitational attraction between the
hole and its images. The exchanged
graviton can be seen in a dual channel as a 1-loop open string
diagram, but note that since L
all modes of the open string contribute in the loop integral,
not just the lowest energy modes.
Now let us ask how this interaction would be seen in a dual field theory description. First
look at a simpler system:
An open string stretched between two D3-branes, with the length of
this string being L
; this is pictured in Fig. 10(a). The low energy dynamics of this open
string has a tower of closely spaced vibration modes. Now consider such a string, still stretched
between two D3 branes, in a space with large D3 brane chargei.e., in AdS5 . In the gravity
picture, the string still has a tower of oscillation modes. But in the dual CFT the open string is
a gauge boson, Higgsed because of the separation between the D3 branes at its ends. At weak
coupling this gauge boson has no tower of excitation modes. But at larger coupling, we can sum
a set of ladder diagrams and see that the tower of oscillation modes is reproduced [40]. Thus
we see that the tower of modes in the gravity picture of Fig. 10(a) is reproduced in the CFT by
degrees of freedom that arise from only the ground states of open strings; the price we pay is that
we have to consider all orders of interaction between these open strings.
Let us draw lessons from this example for our own system. In the gravity picture Fig. 9(b) we
have all modes of the open string stretched between images of the black hole. The dual CFT in
our case is complicated, but can be assumed to arise in some way analogous to the field theory of
open strings on branes. To see the vibration modes of the open string in the CFT we thus expect
that we will need to consider all orders of interaction between the basic degrees of freedom of
the CFT.
But in our microscopic computation of the phase transition we considered only the free CFT.
This should explain why we did not see the tension in the black hole phase.
Interactions in the CFT are difficult to study. In the D1D5 CFT they involve the insertion of
a twist operator in the orbifold CFT [41]. With the presence of the KK-monopole degrees of
330
freedom, we will get other interactions as well. If we do not study these interactions, how can we
proceed to study the phase transition?
In [36] a method was developed to get an indirect picture of the CFT effects, and this method
was used further in [25]. We will recall this approach below, and study it a further to extract some
properties of the effective interactions in the CFT.
4.2. Microscopics
First consider the black hole in the limit L . Then we have just a black hole carrying
D1, D5 charges in 4 + 1 dimensions, and this is described by the (1 + 1)-dimensional CFT. The
energy above extremality and the P charge are given by
E0 = mp (np + n p ),
P = np n p .
(4.11)
Me2p
.
6mp
(4.12)
Using (4.6) and (4.8) we find

np =
Me2p
,
6mp
n p =
In [25,36] a mixed approach was taken. Suppose that for L the black hole nonextremal energy is E0 and the charge is P ; we can find np , n p using (4.11). For finite L there is a
lowering of the energy due to the interaction with images, by an amount Vint . Thus for the same
total energy, there will be higher values of np , n p , and a correspondingly higher entropy. This
increase in the entropy can be checked against the gravity prediction. This is a mixed approach
in the sense that Vint is computed from gravity, and then fed into the microscopic computation
through the values of np , n p .
We will also follow such a mixed approach, but proceed in a slightly different way that will
allow us to better extract some quantities of interest. At L the near horizon region of the
D1D5 system is dual to a (1 + 1)-dimensional CFT, in which the left and right movers do not
interact; thus there is no term coupling np to n p . Because the AdS region ends and goes over
to flat space, there is a breaking of conformal symmetry and a coupling of left and right movers
to the graviton field; this cubic interaction was responsible for Hawking emission from the nearextremal D1D5 system [32]. In the present problem we will let the charge radii of the D1, D5
charges be much larger than L, so this changeover from AdS to flat space is not relevant, and
the above coupling is not the leading order interaction between left and right movers. But what
we do have is a compact z direction, and KK monopole pairs can wrap around this circle. When
we integrate out the effect of these pairs we will be left with an interaction between left and
right movers. We will now try to understand these effective interactions in more detail, obtaining
information about them from gravity, the Smarr scaling relation, the behavior under boost etc.
We see from (4.10), using (4.12), that the interaction energy can be written as
(5)
Vint = 8
GN m2p
np n p .
(4.13)
L2
We observe that the interaction energy vanishes in the BPS case (np = 0 or n p = 0), as expected.
The total energy is (to this order)
E = E0 + Vint = mp (np + n p ) 8
2
G(5)
N mp
L2
np n p .
(4.14)
This relation gives the energy of the microscopic model to second order in np , n p .
The horizon is not deformed to leading order, so we still have

S = 2 n1 n5 np + n p .
331
(4.15)
How do we continue the study of interactions to higher order? In the L limit we had a
free CFT and we determined np , n p from the relations (4.11). In the interacting theory there will
be an ambiguity in defining np , n p for a given state. From the gravity perspective we can split the
corrections to the thermodynamics into two kinds of terms: Those that come from nonlinearities
of general relativity and those that arise because the horizon of the black hole is distorted by
the compactification. The latter set of terms do not start till order r010 , where r0 is the horizon
radius. The former set of terms includes the interaction energy Vint derived above as the leading
correction, and other similar terms that follow at higher order, all in integer powers of r02 [23].
We will define np , n p at all orders in the interaction through the relations
P=
Qp
= np n p ,
mp

S = 2 n1 n5 np + n p .
(4.16)
In view of the comments above, this definition is natural to all orders where the horizon is not
distorted (since the internal degrees of freedom giving the entropy S are not affected), and at
higher orders there is no particularly natural scheme to define np , n p , so we might as well use
(4.16). The nontrivial information about the system will then be in the expression for E in terms
of np , n p .
Note that a priori we have E = E(S, Q, L). Roughly speaking, the Smarr scaling relation
will set the L dependence, and the behavior under boosts will set the Q dependence. Thus the
freedom in E will be equivalent to one function of one variable. The coefficients describing
this remaining function we will take from the gravity calculationin principle they can all be
determined numerically, but of course we will be looking at only the first few coefficients, and
these can be understood analytically.
4.3. Interactions at second order
This section is somewhat long, so we first summarize the steps in our computation. In Section 4.3.1 we recall the relations in gravity between the neutral system and the charged one, but
in a notation specially adapted to the study of the black hole branch. In Section 4.3.2 we introduce dimensionless variables to simplify notation. In Section 4.3.3 we argue that because of the
Smarr relation, we can write the dimensionless energy
as a function only of rescaled versions
of the parameters np , n p (these rescaled parameters are called , ). In Section 4.3.4 we derive
our main result: By using the fact that the tension T must remain unchanged under boosting, we
derive the term in
(, )
which takes us to one order higher in , than the expression (4.14).
In Section 4.3.5 we extend this result to the case where we keep the first order in , but all orders
in .
4.3.1. Gravity variables for the small black hole limit
Let us recall the relation (2.64)
TL
TS
cosh 2p +
.
(4.17)
2
2
Since we are working in the phase where we have a small black hole, it will be convenient to use
variables that relate directly to the structure of the hole. The neutral hole in 4 + 1 dimensions has
E=
332
the metric

r2
dr 2
ds 2 = 1 02 dt 2 +
+ r 2 d32 .
r02
r
1 r2
(4.18)
We have
M=
r 2,
(5) 0
8GN
S=
r 3,
(5) 0
2GN
T=
1
,
2r0
TS =
r 2.
(5) 0
(4.19)
4GN
After we compactify the transverse circle z the metric will no longer have the spherically symmetric form (4.18). But we will use the last of the relations (4.19) to motivate the following
definition of r0 even after compactification
2
TS
(4.20)
r .
(5) 0
4GN
Let us substitute this in (4.17). We find
E=
r02
TL
cosh 2p +
.
(5) 2
2
4G
(4.21)
The tension T to leading order has been found in (4.7). Putting this in (4.21) we get

(5)
2GN M 2
r02
cosh 2p +
E
.
(5)
3
L2
4G 2
(4.22)
The Smarr relation for the neutral system gives

M
2n
2
=TS
r .
(5) 0
3
4GN
(4.23)
When L we have n 0, so for small black holes we have

M
2
r .
(5) 0
3
8GN
(4.24)
Substituting this is (4.22) we get

2
(5)

r02
2GN
E=
cosh 2p +
r04 + O r06 .
2
(5) 2
(5)
L
4GN
8GN
(4.25)
4.3.2. Defining dimensionless variables

The above expression can be simplified by factoring out constants to define dimensionless
variables. Thus write
(5)
16GN
E,
L2
Then (4.25) becomes
(5)
q=
16GN
Qp ,
L2

02
4
cosh 2p + 0 + O 06
2
32
0 =
2
r0 .
L
(4.26)
(4.27)
333
and we have (using (2.60), (4.23))

02
sinh 2p .
2
We also define scaled microscopic values:
q=
(5)
16GN
mp np ,
L2
(4.28)
(5)
16GN
mp n p .
L2
(4.29)
4.3.3. The Smarr relation in microscopic variables

We are interested in looking at the general expression for E as a power series in np , n p ; this
expression will encode the effect of interactions in the CFT arising from the compactification of
the direction z. We will now argue that in terms of the dimensionless variables introduced above,
we will have
=
(, ).
(4.30)
The argument proceeds in three steps:

(A) First recall Eq. (2.61). For the neutral system, we have
n = n(M, L).
(4.31)
Using (2.60) to express p , we find that we can write

E = E(M, L, Qp ).
(4.32)
Now consider
S = Sneutral cosh 1 cosh 5 cosh p ,
(4.33)
where
Sneutral = Sneutral (M, L).
(4.34)
Let us define
S
S
.
Q1 Q5
(4.35)
Then we find, for large D1 and D5 charges,

S =
Sneutral cosh 1 cosh 5 cosh p

2n
M 3 sinh 1 cosh 1 sinh 5 cosh 5
Sneutral cosh p
M 2n
3
(4.36)
and we find that we can again write
S = S(M,
L, Qp ).
(4.37)
We can now eliminate M between (4.32) and (4.37) to write

L, Qp ).
E = E(S,
(4.38)
(B) Now consider the Smarr scaling for the (4 + 1)-dimensional gravity solution. The solution
is invariant under
E L2 E,
Q1 L2 Q1 ,
Q5 L2 Q5 ,
Qp L2 Qp ,
S L3 S.
(4.39)
334
Note that under this scaling we will have
S LS.
(4.40)
Thus we can write (4.39) as

E
E S Qp
.
=
,
L2 L2 L L2
(4.41)
(5)
In the above scalings we have regarded GN as a constant, but since this constant has units,
its useful to see how it appears in the physical quantities of interest. All energies appear in the
gravity theory through the combinations
(5)
GN E,
(5)
GN Q1 ,
(5)
GN Q5 ,
(5)
(4.42)
GN Qp ,
etc. The entropy appears in the gravity theory through the area of a horizon, i.e. as
S =
(5)
GN S A,
(5)
GN S
(5)
(G(5)
N Q1 )(GN Q5 )
(4.43)
We can thus write (4.38) as

(5)
(5)
(5)
GN E GN E S GN Qp
.
,
=
L
L2
L2
L2
(4.44)
(C) Now note that

(5)
16GN
Qp = .
L2
(4.45)

2 n1 n5 ( np + n p ) 2( np + n p )
S
S =
=
=
,
n1 m1 n5 m5
m1 m5
Q1 Q5
(4.46)
Also,
where m1 , m5 , mp will be used to denote the masses of individual quanta of D1, D5, P charges.
Now note that
m1 m5 mp =
4G(5)
N
(4.47)
and we find
S
= + .
L
(4.48)
Returning to (4.44) we see that the LHS is just

(upto a numerical constant), and the arguments
on the RHS are (from (4.45), (4.48)) just functions of , .
Thus we establish (4.30).
335
4.3.4. The second order correction in the microscopic picture

We have defined the microscopic parameters np , n p through (4.16), and their rescaled versions
, through (4.29). We now wish to work out the expansion of
(, )
in powers of its arguments,
using what we know from gravity.
At leading order in , we have
= + ,
q = .
(4.49)
We note some other general properties of

. First, it must be symmetric in , . Second, if = 0
then the system is BPS, and the energy is not corrected by the compactification length L being
finite. This means that for = 0 we must have precisely
= ; there are thus no terms in
of
type 2 , 3 , . . . , and similarly there are no terms 2 , 3 , . . . .
The gravity calculation of the next to leading correction to the energy (Eq. (4.14)) is written
in dimensionless form as
1
= + .
2
(4.50)
At next order we would have

1
= + + ( + ).
2
(4.51)
We will now see that we can determine the constant just by using the fact that the tension does
not change under boosting. Note that boosting changes the charge Qp , and will also change ,
. But since the tension is to be left invariant, we get a constraint on expansion coefficients like
in (4.51). Let us carry out this computation.
The tension is given through
LT = L
L2
E=L
(,
)
.
L
L 16G(5)
N
(4.52)
The L derivative is done by keeping S and Qp fixed; from (4.16) this is equivalent to keep np
and n p fixed, and hence , scale as L2 . It follows that for a term in
of the form a b we
will have the contribution to the tension
(5)
16GN
LT 2(a + b 1) a b .
L2
(4.53)
Thus for the terms in (4.51) we will get

= 4 ( + ).
(4.54)
Let us now convert to gravity variables, using (4.27), (4.28). We have

2
4
1
= + = 0 cosh 2p + 0 ,
2
2
32
2
q = = 0 sinh 2p ,
2
(4.55)
(4.56)
336
where we have written the , expansion only to the orders that we will need.5 These relations
invert to give
=
02 2p 04
e
+ ,
4
32
02 2p 04
e
+ .
4
32
(4.57)
Substituting in (4.54) we find

=

04 06
+ (1 8 ) cosh 2p + O 08 .
16 64
(4.58)
Requiring that not depend on p then gives

=
1
8
(4.59)

04
+ O 08 .
16
(4.60)
and
The relation (4.21) rewritten in rescaled variables appears as
02
cosh 2p + .
2
2
(4.61)
Substituting (4.60) we get

02
4
cosh 2p + 0 + C06 + O 08
2
32
(4.62)
with
C = 0.
(4.63)
Thus we find that the coefficient of 06 in

vanishes. The determination of this O(06 ) term
is the main result of the computation above. Determination of this term has given us the next
correction to the energy beyond (4.14). This term was determined by an explicit and detailed
gravity computation in [18,23]. It may appear remarkable that we have found it here by a simple
argument involving the invariance of the tension under boosting, and one might wonder where
the physics of this correction has been fed into our analysis. This physical input is actually
hidden in the assumption that we can expand
in integer powers of , to the order that we are
working. In other words, the vanishing of the O(06 ) term in the energy does not seem to have
a simple interpretation in the gravity computation, but it does have a simple interpretation in the
microscopic description. Thus we find that the numbers np n p defined through (4.16) give a good
definition of quasi-particle excitations to the first few orders in the interaction, and using an
expansion in integral powers of these numbers automatically reproduces the O(06 ) term in the
energy.
5 Since , 2 , if follows from (4.54) that to compute up to order 6 we only need
and q up to order 4 .
0
0
0
337
4.3.5. First order in , all orders in

In the above we have worked to second order in the excitations , which correspond to left
and right movers respectively. We will now extend our microscopic expression for the energy
to include terms with arbitrary orders in , while keeping only the first order in . This means
that we allow a large net P charge, since the left moving excitation is no longer small. Our
technique will be to again use the fact that the tension T does not depend on the boost parameter
p , but now we will take our limits in a slightly different way, as follows.
We will still let 0 be our small parameter, but we are interested in large boosts p . Thus we
will let
0 0,
p ,
02 e2p finite.
Note that we will thus have

02 e2p O 04 ,
(4.64)
(4.65)
etc.
We now have
= +
and

I (),
2
q =
(4.66)

= I () + I () + O 2 .
(4.67)
To understand the limits that we are taking, look at the gravity description of these variables. First
consider the expression for energy in the expansion where 0 is small, and p is finite. From the
way that energy transforms under boosting, we know that
= 02 sinh2 p +
0 (0 ),
(4.68)
where
0 (0 ) is the energy of the system without the P charge; thus
0 (0 ) does not depend on
p . We can thus write

4
02
cosh 2p + 0 + O 06 ,
2
32
q=
02
sinh 2p ,
2
(4.69)
where we have used cosh 2p = 2 sinh2 p + 1 and the terms O(06 ) do not depend upon p . The
latter fact then tells us that even when we take the limit (4.64) the terms dropped in
are O(06 ).
The expression for q is exact.
We now find

2
2
q0
q
0 e2p 1 +
.
q 0 e2p q0 ,
(4.70)
4
2
4
4
We will in general keep terms upto order 04 , but we have written the expression for q only to
leading order since we will not need higher orders. We invert (4.66) to find , to the required
order

1

q
q
1 I (q)
+ O (
q)2 ,
=q+
2
4

1

q
q
1 I (q)
=
(4.71)
+ O (
q)2 .
2
4
338
Using (4.70) and then substituting in (4.67) we have

1

04
q0
q0
1+
1 I (q0 )
=
I (q0 ) + q0 I (q0 ) + O 08 .
16
4
4
Invariance under boost requires

1

q0
q0
1 I (q0 )
1+
I (q0 ) + q0 I (q0 ) = const.
4
4
(4.72)
(4.73)
Moreover when q0 = 0 one should recover the previous result (4.60), so the constant in the above
equation should be unity. The differential equation

1

1 I ()
1+
(4.74)
I () + I () = 1
4
4
with the boundary condition I (0) = 1, has a unique solution given by

1
I () = 1 +
4
(4.75)
(4.75) is the main result of this subsection. Substituting back in

in (4.66) we find that we have
the energy to first order in n p but to all order in np . We have again obtained this result by a
simple argument; the essential physics here was that np can be changed by boosting (which adds
P charge), and properties of black objects change in simple ways under boosting.
Note that the first order expansion of I () in powers of implies = 1/8, in agreement with
our earlier result (4.59).
5. Discussion
Let us summarize the computations of this paper. We have noted that gravity computations
reveal an interesting phase diagram for the black holeblack string transition, with three phases
for a range of r0 /L. We can add charges to the neutral system by boosting plus dualities. This
does not add or lose any information as far as the gravity computation is concerned, since the
neutral and boosted variables are related in a simple and invertible way. But it does map the
problem to a near extremal one, where we can look for microscopic models.
For r0 /L very small we get the near extremal 2-charge system in 4 + 1 non-compact dimensions, and for r0 /L very large we get the near extremal 2-charge system in 3 + 1 non-compact
dimensions. Both these systems have been given a microscopic description in earlier work on
the subject. We propose a simple model that interpolates between these two microscopic models;
the nontrivial step is guessing what kind of fractionation will describe the D1D5 degrees of
freedom in the interpolating regime. We find that the three-phase nature of the phase diagram
is reproduced if we assume that the two large charges n1 , n5 bind to make N = n1 n5 units of
an effective charge, and this product N is then partitioned between subsystems that behave like
one or the other of the two limiting systems mentioned above. We suggest a physical reason why
this kind of fractionation is reasonable for this problem. Note that even though the microscopic
model looks like a superposition of two subsystems (Eq. (3.18)), we have not just superposed
a black hole and a black string; the fact that the product N = n1 n5 partitions between the two
subsystems forces the entire state to be a nontrivial bound state of all the excitations.
The non-uniform string appears as a saddle point between two maxima of the entropy; this
accords with the fact that the entropy of this branch is always lower than the entropy of the
339
other two phases, so it is always unstable. The n graph describing the tension of this branch
appears to have a profile n 1/ from the gravity computation; the microscopic computation
also gives the functional behavior n 1/.
The microscopic computation based on such a simple model does not give any tension for the
black hole branch. We have argued that this tension will be seen only if we include interactions
like the twist interaction that are present in the CFT when we move away from the orbifold
point (where we have a free CFT) to a point in moduli space that has the couplings appropriate
to the actual gravity dual. We did not look at such interaction terms directly, but instead followed
a mixed approach similar to that in [25,36]. Thus we used the knowledge of the leading order
tension in the gravity computation to see what effective interactions we obtain in the CFT. We
were then able to reproduce the next order term in the tension by using the invariance of the
tension under boosting. Since boosting adds charge it changes np , n p and the non-extremal
energy E, and we get a nontrivial condition when we require invariance of the tension T =
E/L. This next order term in the tension is equivalent to a corresponding correction term in the
energy, and this correction is seen to agree with detailed computations of the energy carried out
in [18,23]. The essential physics we used was to assume that we have an expansion of the energy
in integer powers of np , n p to the first few orders; this suggest that the np , n p (defined through
(4.16)) are good definitions of quasi-particles for the system to this order in the interaction.
While our microscopic model gives some general qualitative agreements with the gravity
phase diagram, there are also some differences. First of course is the issue of tension for the black
hole branch mentioned above. Next, consider for example the point where the non-uniform black
string branch starts off from the uniform string branch. In the microscopic model this branch
starts at a mass value that is 0.57 times the value seen in the gravity diagram. It would also be
nice to see in the microscopic model how the black hole branch bends around to join with the
non-uniform string branch, and nature of the physics and the joining point.
In principle all features of the gravity phase diagram are contained in the correct microscopic
dual, so we look upon this phase diagram as a large source of data on the interactions of branes
that make up black holes and other black objects. The phase diagram offers us continuous curves
that the microscopics must fit; further, we are probing a domain where the black object is changing character from one type to another, so we learn about aspects of the CFT that we do not
encounter when looking at just near extremal entropy or low energy Hawking radiation. Thus we
consider the model proposed here as just a first pass at understanding the microscopics of branes
dual to black objects.
Acknowledgements
We thank T. Harmark, N. Obers and B. Kol for explaining many aspects of their work to
us, and A. Saxena, K. Skenderis and Y. Srivastava for many helpful comments. This work was
supported in part by DOE grant DE-FG02-91ER-40690.
Appendix A. Map between neutral and 2-charge hole
In this appendix, starting from the neutral geometry (2.9), we generate a geometry carrying
D1D5 charges: The D1 extends along the direction y and the D5 along y and za , with a =
1, . . . , 4. Of the remaining 4 + 1 directions, one, denoted by z, is compactified on a circle of
length L. We are thus left with d = 4 non-compact directions.
340
The desired 2-charge solution is generated by acting on (2.9) with the following sequence of
boosts and dualities (we denote by By a boost along y and by Tij ... a T-duality in the directions
i, j, . . .)
By
By
Ty
T1234
neutral Py NS1y NS1y , Py D1y , Py D5y1234 , Py

Ty1
NS5y1234 , Py NS5y1234 , NS1y D5y1234 , D1y .
(A.1)
We list the geometries obtained at each step. The notation is as follows: ds 2 denotes the string
metric, is the dilaton, B (2) the NSNS B-field and H (3) its field strength, C (p) is the RR pform gauge potential and G(p+1) the corresponding field strength, B denotes the Hodge dual
with respect to dsB2 .
By :

2
ds 2 = H51 U dt 2 + H5 dy H51 1 coth 5 dt + dsB2 + dza dza ,
H5 = 1 + (1 U ) sinh2 5 .
(A.2)
Ty :

ds 2 = H51 U dt 2 + dy 2 + dsB2 + dza dza ,

(2)
e2 = H51 .
Bty = H51 1 coth 5 ,
(A.3)
By :

2
+ dsB2 + dza dza ,
ds 2 = H51 H11 U dt 2 + H1 dy H11 1 coth 1 dt

(2)
e2 = H51 ,
Bty = H51 1 coth 5 ,
H1 = 1 + (1 U ) sinh2 1 .
S:
(A.4)
1/2

2
H11 U dt 2 + H1 dy H11 1 coth 1 dt

1/2
+ H5 dsB2 + dza dza ,

(2)
Cty = H51 1 coth 5 ,
e2 = H5 .
ds 2 = H5
(A.5)
T1234 :
1/2

2
H11 U dt 2 + H1 dy H11 1 coth 1 dt
ds 2 = H5
1/2
+ H5
(6)
Ctyz1 ...z4
1/2
dsB2 + H5
dza dza ,
1

= H5 1 coth 5 ,
e2 = H51 ,
Grtyz1 ...z4 = r H51 coth 5 ,

(7)
S:
G(3) = U 1/2 r H5 coth 5 B dr.
(A.6)

2

ds 2 = H11 U dt 2 + H1 dy H11 1 coth 1 dt + H5 dsB2 + dza dza ,
H (3) = U 1/2 r H5 coth 5 B dr,
e2 = H5 .
(A.7)
Ty1 :
341

ds 2 = H11 U dt 2 + dy 2 + H5 dsB2 + dza dza ,
H (3) = U 1/2 r H5 coth 5 B dr r H11 coth 1 dr dt dy,
H5
.
e2 =
H1
S:
(A.8)
1/2

H1
ds 2 = (H5 H1 )1/2 U dt 2 + dy 2 + (H5 H1 )1/2 dsB2 +
dza dza ,
H5
H1
.
e2 =
H5
(A.9)
Converting from string to Einstein frame, we find

1/4 1/4 1
dsE2 = H1 H5
H1 U dt 2 + dy 2 + H5 dsB2 + dza dza ,
H1
, H1 = 1 + (1 U ) sinh2 1 , H5 = 1 + (1 U ) sinh2 5 .
e2 =
H5
(A.10)
Energy and tension are derived from the large r behavior of the Einstein metric. Assume that
the neutral metric has the following asymptotic expansion
ct
cz
gzz = 1 + ,
gyy = 1,
gza za = 1.
gtt = 1 ,
(A.11)
r
r
Let M and T be the mass and tension of the neutral system, and define the relative tension
LT
.
M
Einsteins equations relate these quantities to the ct and cz defined above:
n=
M=
L
(5)
16GN
2 (2ct cz ),
T =
2
(5)
16GN
(ct 2cz ),
n=
(A.12)
ct 2cz
.
2ct cz
(A.13)
denote the components of the metric in (A.10), C (2) the RR 2-form gauge field and
Let g
C (6) the dual 6-form field. Their asymptotic expansions are of the form
gtt = 1
ct
,
r

gzz
=1+
cz
,
r

gyy
=1+
cy
gz a za = 1 +
ca
,
r
r

c
c
(2)
(6)
Cty = 1 , Ctyz1 ...z4 = 5 .
(A.14)
r
r
Let M and T be the mass and tension of the charged system and Q1 and Q5 denote the D1 and
D5 charges (with a normalization such that M = Q1 + Q5 for the extremal hole). The relations
linking mass, tension and charges to the asymptotic fall-off of the metric are
M =
L
16G(5)
N
2 (2ct cz cy 4ca ),
T=
2
16G(5)
N
(ct 2cz cy 4ca ),
342
Q1 =
L
(5)
16GN
2 c1 ,
Q5 =
L
(5)
16GN
2 c5 .
(A.15)
From (A.10) we find the relations between the coefficients ci and ci :
3
1
1
3
cz = cz + ct sinh2 1 + ct sinh2 5 ,
ct = ct + ct sinh2 1 + ct sinh2 5 ,
4
4
4
4
3
1
1
1

2
2

2
cy = ct sinh 1 ct sinh 5 ,
cu = ct sinh 1 ct sinh2 5 ,
4
4
4
4
c1 = ct sinh 1 cosh 1 ,
(A.16)
c5 = ct sinh 5 cosh 5 .
We deduce from (A.13), (A.15) and (A.16) that

2 n
M = M 1 +
sinh2 1 + sinh2 5 ,
T=T,
3
2n
Qi = M
(A.17)
sinh i cosh i (i = 1, 5).
3
Let us now consider temperature and entropy. Temperature is proportional to the surface gravity , which is given by

1 tt rr
(A.18)
g g (r gtt )2 r=r
0
4
with r = r0 the location of the horizon. From (2.47), we find that the components of the 2-charge
metric are related as follows to the components of the neutral metric (2.9)
1/4 1/4 1
1/4 1/4

grr
= H1 H5
H1 gtt ,
H5 grr .
gtt = H1 H5
(A.19)
2 =
The surface gravity of the 2-charge system is thus

2 =

1 tt rr
2
g g (r gtt )2 r=r = 2 (H1 H5 )1 r=r =
,
2
0
0
4
cosh 1 cosh2 2
(A.20)
where we have used the fact that gtt = 0 at the horizon and that
Hi |r=r0 = cosh2 i
(i = 1, 5).
(A.21)
Then the temperatures of the neutral and charged holes (T and T ) are related as
T =
T
.
cosh 1 cosh 5
(A.22)
Entropy is proportional to the horizon area. Using again (2.47), we see that the horizon areas
before and after the boost (Ah and Ah ) are related as
Ah = (H1 H5 )1/2 Ah
(A.23)
and thus the respective entropies (S and S ) scale as

S = S cosh 1 cosh 5 .
(A.24)
Note that
T S = T S.
(A.25)
References
[1] R. Gregory, R. Laflamme, Phys. Rev. Lett. 70 (1993) 2837, hep-th/9301052;
R. Gregory, R. Laflamme, Nucl. Phys. B 428 (1994) 399, hep-th/9404071.
[2] S.S. Gubser, Class. Quantum Grav. 19 (2002) 4825, hep-th/0110193.
[3] T. Harmark, N.A. Obers, JHEP 0205 (2002) 032, hep-th/0204047.
[4] B. Kol, JHEP 0510 (2005) 049, hep-th/0206220.
[5] T. Wiseman, Class. Quantum Grav. 20 (2003) 1137, hep-th/0209051;
T. Wiseman, Class. Quantum Grav. 20 (2003) 1177, hep-th/0211028.
[6] B. Kol, T. Wiseman, Class. Quantum Grav. 20 (2003) 3493, hep-th/0304070.
[7] T. Harmark, N.A. Obers, Class. Quantum Grav. 21 (2004) 1709, hep-th/0309116.
[8] B. Kol, E. Sorkin, T. Piran, Phys. Rev. D 69 (2004) 064031, hep-th/0309190.
[9] T. Harmark, N.A. Obers, Nucl. Phys. B 684 (2004) 183, hep-th/0309230.
[10] E. Sorkin, B. Kol, T. Piran, Phys. Rev. D 69 (2004) 064032, hep-th/0310096.
[11] H. Kudoh, T. Wiseman, Prog. Theor. Phys. 111 (2004) 475, hep-th/0310104;
H. Kudoh, T. Wiseman, Phys. Rev. Lett. 94 (2005) 161102, hep-th/0409111.
[12] T. Harmark, Phys. Rev. D 69 (2004) 104015, hep-th/0310259.
[13] E. Sorkin, Phys. Rev. Lett. 93 (2004) 031601, hep-th/0402216.
[14] D. Gorbonos, B. Kol, JHEP 0406 (2004) 053, hep-th/0406002.
[15] O. Aharony, J. Marsano, S. Minwalla, T. Wiseman, Class. Quantum Grav. 21 (2004) 5169, hep-th/0406210.
[16] B. Kol, E. Sorkin, Class. Quantum Grav. 21 (2004) 4793, gr-qc/0407058;
B. Kol, E. Sorkin, Class. Quantum Grav. 23 (2006) 4563, hep-th/0604015.
[17] T. Harmark, N.A. Obers, JHEP 0409 (2004) 022, hep-th/0407094;
T. Harmark, N.A. Obers, Nucl. Phys. B 742 (2006) 41, hep-th/0510098.
[18] D. Karasik, C. Sahabandu, P. Suranyi, L.C.R. Wijewardhana, Phys. Rev. D 71 (2005) 024024, hep-th/0410078.
[19] B. Kol, Phys. Rep. 422 (2006) 119, hep-th/0411240.
[20] T. Harmark, N.A. Obers, hep-th/0503020.
[21] D. Gorbonos, B. Kol, Class. Quantum Grav. 22 (2005) 3935, hep-th/0505009.
[22] J.L. Hovdebo, R.C. Myers, Phys. Rev. D 73 (2006) 084013, hep-th/0601079.
[23] Y.Z. Chu, W.D. Goldberger, I.Z. Rothstein, JHEP 0603 (2006) 013, hep-th/0602016.
[24] B. Kleihaus, J. Kunz, E. Radu, JHEP 0606 (2006) 016, hep-th/0603119.
[25] T. Harmark, K.R. Kristjansson, N.A. Obers, P.B. Ronne, hep-th/0606246.
[26] A. Murugan, V. Sahakian, hep-th/0608103.
[27] C.G. Callan, J.M. Maldacena, Nucl. Phys. B 472 (1996) 591, hep-th/9602043.
[28] J.M. Maldacena, A. Strominger, Phys. Rev. D 55 (1997) 861, hep-th/9609026.
[29] J.M. Maldacena, Nucl. Phys. B 477 (1996) 168, hep-th/9605016.
[30] G.T. Horowitz, J.M. Maldacena, A. Strominger, Phys. Lett. B 383 (1996) 151, hep-th/9603109.
[31] G.T. Horowitz, D.A. Lowe, J.M. Maldacena, Phys. Rev. Lett. 77 (1996) 430, hep-th/9603195.
[32] S.R. Das, S.D. Mathur, Nucl. Phys. B 478 (1996) 561, hep-th/9606185.
[33] I.R. Klebanov, S.D. Mathur, Nucl. Phys. B 500 (1997) 115, hep-th/9701187.
[34] S.R. Das, S.D. Mathur, Phys. Lett. B 375 (1996) 103, hep-th/9601152.
[35] J.M. Maldacena, L. Susskind, Nucl. Phys. B 475 (1996) 679, hep-th/9604042.
[36] M.S. Costa, M.J. Perry, Nucl. Phys. B 591 (2000) 469, hep-th/0008106.
[37] S.D. Mathur, Nucl. Phys. B 529 (1998) 295, hep-th/9706151.
[38] S.D. Mathur, Class. Quantum Grav. 23 (2006) R115, hep-th/0510180.
[39] A. Strominger, C. Vafa, Phys. Lett. B 379 (1996) 99, hep-th/9601029.
[40] I.R. Klebanov, J.M. Maldacena, C.B. Thorn, JHEP 0604 (2006) 024, hep-th/0602255.
[41] O. Lunin, S.D. Mathur, Commun. Math. Phys. 219 (2001) 399, hep-th/0006196;
O. Lunin, S.D. Mathur, Commun. Math. Phys. 227 (2002) 385, hep-th/0103169.
343
Gauge invariant Lagrangian formulation of higher spin

massive bosonic field theory in AdS space
I.L. Buchbinder a , V.A. Krykhtin b, , P.M. Lavrov c
a Department of Theoretical Physics, Tomsk State Pedagogical University, Tomsk 634041, Russia
b Laboratory of Mathematical Physics, Tomsk Polytechnic University, Tomsk 634034, Russia
c Department of Mathematical Analysis, Tomsk State Pedagogical University, Tomsk 634041, Russia
Received 18 August 2006; received in revised form 30 September 2006; accepted 20 November 2006
Available online 4 December 2006
Abstract
In this work we develop the BRST approach to Lagrangian construction for the massive integer higher
spin fields in an arbitrary dimensional AdS space. The theory is formulated in terms of auxiliary Fock
space. Closed nonlinear symmetry algebra of higher spin bosonic theory in AdS space is found and a
method of deriving the BRST operator for such an algebra is proposed. A general procedure of Lagrangian
construction, describing the dynamics of a bosonic field with any spin is given on the base of the BRST
operator. No off-shell constraints on the fields and the gauge parameters are used from the very beginning.
As an example of general procedure, we derive the Lagrangians for massive bosonic fields with spin 0, 1
and 2, containing the total set of auxiliary fields and gauge symmetries.
1. Introduction
In modern theoretical physics various aspects of higher spin field theory are intensively investigated [1]. At present, research in this area is focused on the development of general methods
for constructing Lagrangian formulations, on the study of underlying symmetry structure and
relation to string theory, and on efficient way of describing higher spin fields interaction (see e.g.
[28] for recent progress in massive higher spin theories and [917] for massless ones).
E-mail addresses: joseph@tspu.edu.ru (I.L. Buchbinder), krykhtin@mph.phtd.tpu.edu.ru (V.A. Krykhtin),

lavrov@tspu.edu.ru (P.M. Lavrov).
doi:10.1016/j.nuclphysb.2006.11.021
I.L. Buchbinder et al. / Nuclear Physics B 762 [PM] (2007) 344376
345
In the present paper the general approach to constructing Lagrangians for massive higher spin
fields in AdS space of arbitrary dimension is formulated. Originally, the Lagrangian formulation
for massive higher spin fields in AdS space has been proposed in [18] in terms of constrained
double traceless fields. Our approach is based on the BRSTBFV construction1 [19] (see also the
reviews [20]) which automatically allows one to obtain gauge invariant Lagrangian in terms of
unconstrained fields and gauge parameters.
BFV construction was originally developed for quantization of gauge theories. Gauge theories
are formulated in phase space and characterized by first class constraints Ta satisfying the invoc T where f c are structure functions.
lution relations2 in terms of Poisson brackets {Ta , Tb } = fab
c
ab
The basic ingredient of the BFV construction is a nilpotent BFV charge Q defined as follows
1
c
Pc + ,
Q = a Ta + b a fab
2
(1)
where a and Pa are canonically conjugate ghost variables. The dots mean extra terms which
should be added in order to get the nilpotent charge when the structure functions depend on
phase variables of the theory. After quantization the BFV charge Q is a nilpotent Hermitian
operator3 acting in extended space of states | containing the ghosts. Subspace of physical
states is defined by the equation Q| = 0 up to the transformation | = | + Q| which is
the gauge transformation. It is proved that in the subspace of physical states a unitary S-matrix
exists [19]. We emphasize that the starting point of this construction is a classical Lagrangian
formulation of the gauge theory.
One of the problems of the higher spin field theory is its Lagrangian formulation. As it was
pointed out in the pioneer paper [21], the Lagrangian description of higher spin fields along with
a basic spin-s field involves also auxiliary fields with lower values of spin. The application of
the BRST approach to higher spin filed theory, in principle, allows one to describe its dynamics.
The matter is that a gauge theory is completely determined by its constraints. Since the BRST
charge is constructed with the use of the constraints, it contains all information about dynamics.
Therefore if we are able to express Lagrangian in terms of the BRST charge, we automatically
obtain a consistent gauge invariant Lagrangian formulation of a theory with all auxiliary fields
properly taken into account. Namely, such a situation is realized in free higher spin field models
[23,3033] (see also the earlier application of the BRST approach to the interacting higher spin
filed theory in [34]).
The use of the BRST construction in the context of higher spin field theory differs from its
common use for quantization in the following respects.4 Classical Lagrangian formulation of the
theory is unknown from the very beginning. Moreover the main problem of the theory is to find
such a formulation. Hence, a classical system of constraints is also unknown and classical BRST
charge cannot be constructed. What we know are the conditions on the fields defining irreducible
representations of the Poincar or AdS group with given mass and spin. These conditions are
realized as operators acting in auxiliary Fock space and are interpreted as first class constraints.
1 Following the terminology accepted in string field theory and massless higher spin field theory we call BRSTBFV
construction as BRST one.
2 We do not concern here with the open gauge algebras.
3 Following the accepted tradition we call this operator as BRST one.
4 The application of the BRST construction in higher spin field theory is analogous in some aspects to application of
BRST construction in string field theory [27] (see also the reviews [28]). Relations of higher spin field theory to string
field theory are discussed in [29].
346
Then one demands that the algebra of these operators be closed. This condition allows us to find
a complete system of operators forming a closed algebra. As a result, from the very beginning
we have operators and the formulation looks like quantum, although it is still a classical theory.
Moreover, there is no guaranty that the commutator algebra is a Lie algebra. As we will see later
on, the algebra in the case under consideration is nonlinear and analogous to the W3 algebra
(see e.g. [26]). Besides, it turns out that some of the operators forming the algebra cannot be
interpreted as constraints. Thus, the application of the BRST construction in higher spin field
theory faces many principal problems.
In this paper we develop a gauge invariant approach for deriving Lagrangian for massive
higher spin fields in AdS space of arbitrary dimension. This approach is based on the BRST
construction and automatically yields a gauge invariant Lagrangian for a field of arbitrary spin
containing a complete set of auxiliary fields. As it should be, the gauge invariance of the massive
theory means that the corresponding Lagrangian formulation includes the Stueckelberg fields.
The approach under consideration is a generalization of works [3033] on Lagrangian construction of higher spin fields in arbitrary dimensional Minkowski space. The BRST approach to
higher spin fields in AdS space was first used in [23], where the bosonic massless fields were
considered. It was proved [23] that in four dimensions, obtained Lagrangian theory is reduced,
after elimination of all auxiliary fields, to the Fronsdal Lagrangian [22]. In this paper we develop
a general method of Lagrangian construction for massive higher spin bosonic fields in AdS space
and show that in the massless limit it yields the results presented in [23].
The paper is organized as follows. In Section 2 we introduce an auxiliary Fock space, which
is the base of the BRST construction for higher spin fields and describe differential operators
corresponding to constraints. These constraints define an irreducible massive integer spin representation of the AdS group. In Section 3 we construct a set of operators forming a closed
algebra. It is shown that this algebra is nonlinear (quadratic algebra analogous to the W3 algebra)
and includes two operators which are not constraints. In Section 4 we generalize the method
proposed in [3033]. This generalization demands a deformation of the algebra and construction of a representation for the deformed algebra. Such a representation is found so that the new
expressions for the operators consist of two parts: the initial expression of the operator plus an
additional part. The additional parts for the operators are explicitly derived in Section 5. Then in
Section 6 we discuss the construction of the BRST operator for the deformed nonlinear algebra.
After this in Section 7 we derive the Lagrangian describing propagation of a massive bosonic
field of arbitrary fixed spin in AdS space. It is also shown that this theory is a gauge one and
gauge transformations are written down. In Section 8 we illustrate the procedure by finding the
gauge invariant Lagrangians for massive spin-0, spin-1, and spin-2 fields and their gauge transformations in explicit form. In Section 9 summarize the results obtained. Finally, in Appendix A
we provide information on differential calculus in auxiliary Fock space on a gravitational background. Appendix B is devoted to constructing a representation of the operator algebra given in
Table 2 in terms of creation and annihilation operators. In Appendix C we prove that the constructed Lagrangian correctly reproduces the conditions which define a irreducible representation
of the AdS group.
2. Auxiliary Fock space for higher spin fields in AdS spacetime
Massive integer spin-s representation of the AdS group is realized in a space of totally symmetric tensor fields 1 ...s (x) satisfying the following conditions (see e.g. [4])
2

+ r s 2 + s(d 6) + 6 2d + m2 1 2 ...s (x) = 0,
(2)
1 1 2 ...s (x) = 0,
g
1 2
347
(3)
1 2 ...s (x) = 0,
(4)
R
where r = d(d1)
with R being the scalar curvature and d being the dimension of the spacetime.
Our purpose is to construct a Lagrangian which reproduces these constraints as the consequences
of the equations of motion.
In order to avoid an explicit manipulation with a number of indices it is convenient to introduce auxiliary Fock space generated by creation and annihilation operators with tangent space
indices (a, b = 0, 1, . . . , d 1)

aa , ab+ = ab , ab = diag(+, , . . . , ).
(5)
An arbitrary vector in this Fock space has the form

| =
a1 ...as (x)a +a1 a +as |0 =
s=0
1 ...s (x)a +1 a +s |0,
(6)
s=0
where a + (x) = ea (x)a +a , a (x) = ea (x)a a , with ea (x) being the vielbein. It is evident that

a , a+ = g .
(7)
We call the vector (6) a basic vector. The fields 1 ...s (x) are the coefficient functions of the
vector | and its symmetry properties are stipulated by the properties of the product of the
creation operators. We also suppose the standard relation
ea = 0.
(8)
We want to realize the constraints (2)(4) as some constraints on the vectors |. To do that
one introduces a special derivative operator acting on the vectors |5
ab +
D =
aa ab ,
D |0 = 0.
(9)
Then one can show that

[D , 1 ...s ] = ( 1 ...s ),

ba +
ab
D , a a+ =
ab ,
D , a a =
ab ,

+

D , a + =
a ,
D , a =
a ,

+
+
[D , a ] = a ,
D , a = a ,
(10)
(11)
(12)
(13)
where we suppose that aa = aa+ = 0. The operator D acts on the vectors (6) as
D | =

a1 ...as (x) a +a1 a +as |0 =
1 ...s (x) a +1 a +s |0,
s=0
s=0
(14)
where is the covariant derivative acting on tensor fields with tangent and spacetime indices
by standard rule.
5 Another (equivalent) form of the derivative operators acting on vectors of Fock space and differential calculus based
on such operators are discussed in Appendix A.
348
For latter purposes it is useful to state the relations

1 n (x) a 1 + a n + |0,
D | =
g D D
(15)
n=0
[D , D ] = R ab aa+ ab = R a + a ,
(16)
with R
R being the curvatures (A.16), (A.15). Here and further we use the following
= r(g g g g ).
convention for AdS curvature tensor R = g R

The operator D is used for realization of the constraints in space of vectors |. Let us define
the following operators

d(d 4)
2
2
,
l0 = D + m + r X 4g0 + 6
(17)
4

D 2 = g D D
(18)
D ,
ab ,
X = g02 2g0 4l2+ l2 ,

l1 = ia D ,
1
l2 = a a ,
2
l1+
(19)
= ia D ,
1
l2+ = a + a+ ,
2
+
d
.
2
Using the above relations, one can see that if the following constraints
(20)
(21)
g0 = a+ a +
(22)
l0 | = l1 | = l2 | = 0
(23)
are fulfilled then each component 1 ...s (x) of (6) obeys the conditions describing spin-s field
in AdS space (2)(4). Therefore if the conditions (23) are fulfilled then vector (6) will describe
the fields with arbitrary integer spin s in AdS space.
Let us turn to the algebra generated by operators l0 , l1 , l2 .
3. Algebra generated by the constraints
We develop an approach to Lagrangian formulation for massive higher spin fields based on the
following idea. We treat the conditions (2)(4) or the equivalent conditions (23) as constraints in
some unknown yet Lagrangian gauge theory. Our purpose is to restore the Lagrangian leading to
the given set of first class constraints. The most efficient way to realize this idea is to use general
BRST method.
The basic notion of BRST method is the Hermitian BRST operator which is constructed on the
base of a set of first class constraints. To get the Hermitian BRST operator, the set of constraints
should be invariant under Hermitian conjugation. In the case under consideration, the operator
l0 is Hermitian, but the operators l1 , l2 are not Hermitian. Therefore, to get a set of operators
invariant under Hermitian conjugation we should add two more operators l1+ (20), l2+ (21) to the
operators l0 , l1 , l2 . As a result, the set of operators l0 , l1 , l1+ , l2 , l2+ is invariant under Hermitian
conjugation. However, it is clear that the operators l1+ , l2+ cannot be interpreted as constraints
on the vectors (6) on an equal footing with the operators l0 , l1 , l2 . Indeed, taking the Hermitian
conjugation of (23) we see that they (together with l0 ) are constraints on the bra vector
|l0 = |l1+ = |l2+ = 0.
(24)
349
Nevertheless we will use the operators l1+ , l2+ for BRST construction and show that they do not
contribute to equations of motion for the basic vector (6) obtained from final Lagrangian. This
occurs due to the fact that these operators are multiplied on the annihilation ghost operators in
BRST operator (see relations (55) and (71), some details are discussed in [31]).
The algebra of the operators l0 , l1 , l1+ , l2 , l2+ is open in terms of commutators of these operators. To close the algebra we add to it all the operators generated by the commutators of l0 , l1 ,
l1+ , l2 , l2+ .
Before proceeding in this way let us introduce a more general operator l0 instead of operator
l0
d(d 4)
+ ( 1)rX + rg0
l0 = D 2 + m2 + ( 1)r
(25)
4
and use this operator l0 as a constraint. The operator l0 depends on real parameters , , . It
24
, = 2, = 4. We shall show that the
is evident that l0 (17) is obtained from l0 at = d(d4)
correct on-shell condition (2) can be reproduced at any values of , , , besides the expressions
of the Lagrangian (C.7) will be simpler for l0 at some specific values , , which do not coincide
with l0 (see Appendix C). Also we note that the case of massless bosonic higher spin fields in AdS
space considered in [23] corresponds to the following choice of the coefficients m = = = 0,
= 2.
Now let us close the algebra generated by the operators l0 , l1 , l1+ , l2 , l2+ . As a result we obtain
two more Hermitian operators: g0 (22) and
d(d 4)
r.
4
The algebra of operators (20)(22), (25), (26) is closed and it is given in Table 1, where

[l1 , l0 ] = r 4l1+ l2 + 2g0 l1 l1 + rl1 ,

l0 , l1+ = r 4l2+ l1 + 2l1+ g0 l1+ + rl1+ ,

l1 , l1+ = l0 l + (2 )rX rg0 .
l = m2 +
(26)
(27)
(28)
(29)
In this table and in the subsequent ones the first arguments of the commutators are listed in the
first column, the second arguments are listed in the upper row. The algebra corresponding to
Table 1 is a base for massive integer higher spin field Lagrangian construction in AdS space. We
will call this algebra a massive integer higher spin symmetry algebra in AdS space. We want to
emphasize here two points. First, the operator l commutes with all other operators and therefore
Table 1
Algebra of the operators (20)(22), (25), (26)
l0
l1
l1+
l2
l2+
l0
l1
0
(27)
(27)
0
(28)
(29)
2 rl2
0
l1+
2 rl2+
l1+
(28)
(29)
l1
l2
2 rl2
l1
l2+
g0
l
2 rl2+
0
0
l1+
l1
0
0
l1+
0
g0
2l2
0
g0
0
l1
0
0
l1+
g0
2l2
0
2l2+
0
2l2+
0
0
0
0
0
350
it plays a role of central charge. Second, due to (27)(29), the given algebra is nonlinear, the
right-hand sides of these commutators are quadratic forms in operators.
Since the operators g0 and l are obtained as commutators of a constraint on the ket vector
(23) with a constraint on the bra vector (24), these operators g0 and l cannot be considered as
constraints neither in the space of bra-vectors nor in the space of ket-vectors.
One can show that a straightforward use of BRST construction as if all the operators are the
first class constraints does not lead to the proper equations (23) for any spin (see e.g. [3033] for
the cases of higher spin fields in flat space). This happens because among the above Hermitian
operators there are operators which are not constraints (g0 and l in the case under consideration)
and they bring two more equations (in addition to (23)) onto the physical field (6). Thus we must
somehow get rid of these supplementary equations.
The method of avoiding the supplementary equations consists in constructing new enlarged
expressions for the operators of the algebra given in Table 1 so that the Hermitian operators
which are not constraints will be zero.
4. Constructing the deformations of the symmetry algebra
Our purpose here is to construct new enlarged operators satisfying some deformed algebra so
that the new operators g0 and l become zero or (as was shown in [31]) they contain arbitrary
parameters which are then defined from the condition that the supplementary equations do not
give more restrictions on the basic vectors (6) in addition to (23).
The method we apply in this paper for constructing the enlarged expressions for the operators
is a generalization of the method used in our papers [3033] where the operator algebras were
Lie algebras.
In the case of Lie algebras we acted as follows. We enlarged the representation space introducing new creation and annihilation operators and constructed a new representation of the
corresponding operators so that the expression for any operator was a sum of two parts: the initial
expression for the operator plus a specific part which depends on the new creation and annihilation operators only. As a result, in this new representation the operators which are not constraints
were zero or contained arbitrary parameters whose values would be defined later.6 One more
requirement was that the vector in the enlarged space (including the ghost fields) should be independent of the ghost fields corresponding to the operators which are not constraints [3033].
The generalization of the above method to the case of non-linear symmetry algebra given by
Table 1 is based on deformation of algebra of the enlarged operators in comparison with the
corresponding initial algebra.
One describes the method of deformation. Let us denote all the operators of the algebra given
in Table 1 as li . Then the structure of the algebra looks like7
[li , lj ] = fijk lk + fijkm lk lm ,
(30)
6 Introducing the new creation and annihilation operators in BRST construction for higher spin field theory is anal-
ogous to a conversion procedure (see e.g. [35]) which is used for quantization of constraint systems with second class
constraints.
7 The method can easily be generalized to the case when the algebra has the structure
k ...kN
[li , lj ] = fijk lk + fijkm lk lm + + fij1

where all f s are constants.
lk1 lkN ,
351
where fijk , fijkm are constants. In case of fijkm = 0 we have a Lie algebra and the described
method will be reduced to the method used in the flat space case [3033]. As in the case of
a Lie algebra we enlarge the representation space by introducing the additional creation and
annihilation operators and construct the new operators of the algebra li Li
Li = li + li ,
(31)
where li is the part of the operator which depends on the new creation and annihilation operators
only (and constants of the theory like the mass m and the curvature). It is evident that if we fix the
structure of the new operators (31) then we are not able to preserve the algebra (30). Therefore
there are two possibilities. The first one is to reject (31) and the second one is to deform the
algebra (30). We choose the second possibility and demand that the new operators Li (31) are in
involution relations
[Li , Lj ] Lk .
(32)
Since [li , lj ] = 0 we have

[Li , Lj ] = [li , lj ] + [li , lj ]

= fijk Lk fijkm + fijmk lm
Lk + fijkm Lk Lm fijk lk + fijkm lm
lk + [li , lj ].
(33)
Then in order to provide (32) the last three terms must be canceled. Therefore we put

lk
[li , lj ] = fijk lk fijkm lm
(34)
and as a result we have the deformed algebra for the enlarged operators

[Li , Lj ] = fijk Lk fijkm + fijmk lm
Lk + fijkm Lk Lm .
(35)
Thus we see that the algebra (35) of the enlarged operators Li is deformed in comparison with
the algebra (30) of the initial operators li . The second term in this algebra shows that some of
the structure functions are not the constants and depend on the new creation and annihilation
. In the case of Lie algebra f km = 0 the algebras of the initial operators
operators by means of lm
ij
li , of the additional parts li , and of the enlarged operators Li coincide. Namely, this situation
takes place in the flat space [3033].
Before we go further, we must discuss some details concerning the construction of the algebra
(35). First, we should find the operators (additional parts) li satisfying the algebra (34). Second,
these additional parts corresponding to the operators which are not constraints (g0 and l in the
case under consideration) should contain an arbitrary parameter or vanish in sum with their initial
expression (g0 and l in the case under consideration). Third, the additional part corresponding
to the Hermitian constraint l0 should contain an arbitrary parameter which value will be defined
from the condition of reproducing the correct conditions (23). After the additional parts li are
calculated we proceed as follows: one constructs the BRST operator as if all the operators Li are
the first class constraints and the vector in the enlarged space (including the ghost fields) must be
independent of the ghost fields corresponding to the operators which are not constraints.
In next section we study the algebra (34) of the additional parts of the operators li and find
their explicit expressions.
352
5. Constructing the additional parts of the operators

The procedure described in Section 4 in the case under consideration leads to algebra (34) for
the operators of the additional parts in the form given by Table 2 where

[l1 , l0 ] = r 4l1+ l2 + 2g0 l1 l1 + rl1 ,
(36)
+

+
+
+
+

l0 , l1 = r 4l2 l1 + 2l1 g0 l1 + rl1 ,
(37)

2
+
+

l1 , l1 = l0 l rg0 + ( 2)r g0 2g0 4l2 l2 .
(38)
One must point out that in the case of massless fields considered in [23] m = = = 0,
= 2, and therefore it was possible to construct the additional parts so that l0 = l = l1 = l1+ = 0.
In the more general case considered here the explicit expressions for the additional parts can
be calculated with the help of the method described in8 [24,30]. As a result we get

8r k b2+k b12k+1
l0 = m20 + r b1+ b1 + 2b2+ b2 r(2h 1)b1+
(2k + 1)!
m21
k=0

8r k b2+k b12k+2 1
8r k b2+k b12k
2
+
M
,
2rb1+2
(39)
(2k + 2)! 2
(2k)!
m21
m21
k=0
k=1

8r k (b2+ )k1 b12k
+
+ 2h 1

l1 = m1 b1 b2 + m1 b1
4
(2k)!
m21
k=1

8r k (b+ )k1 b2k+1 M 2

8r k b2+k b12k+1
1
+2
2
1
(40)
+
,
+ m1 b1
2
(2k + 1)!
m1
(2k + 1)!
m21
m21
k=1
k=0

2h 1 + 8r k (b2+ )k1 b12k+1
b1
l2 = b2+ b2 + b1+ b1 + h b2
4
(2k + 1)!
m21
k=1

1 +2 8r k (b2+ )k1 b12k+2 M 2 8r k b2+k b12k+2

(41)
2
,
b1
2
(2k + 2)!
(2k + 2)!
m21
m1
m21
k=1
l1+
k=0
= m1 b1+ ,
(42)
Table 2
Algebra of the additional parts for the operators
l0
l0
l1
l1+
l2
l2+
g0
l
l1
l1+
l2
(36)
(37)
2 rl2
(36)
(38)
0
l1
(37)
(38)
2 rl2
l1
2 rl2+
0
l1+
l1
l1+
g0
2l2
l2+
2 rl2+
l1+
0
g0
0
2l2+
0
8 To be completed a detailed calculation of the additional parts is given in Appendix B.
g0
l
l1
l1+
2l2
2l2+
0
0
0
l2+ = b2+ ,
g0
= b1+ b1
353
(43)
+ 2b2+ b2
(44)
+ h,
d(d 4)
r,
l = m2
(45)
4
where we have denoted
d(d 4)
M 2 = m2 + m20 +
(46)
r + ( 2)h(h 2)r hr.
4
In the above expressions h is a dimensionless arbitrary constant, m0 and m1 are the arbitrary
constants with dimension of mass. In constructing the additional parts (39)(45) we have introduced, in accordance with method given in Section 4, two pairs of new bosonic creation and
annihilation operators satisfying the standard commutation relations

b1 , b1+ = b2 , b2+ = 1.
(47)
The calculated additional parts of operators possess all the necessary properties described in
Section 4. The operators which are not constraints give zero in sum with their initial expressions
for the operator (l + l = 0) or contain an arbitrary parameter (g0 contains an arbitrary parameter h) which value will be defined later from the condition that they do not generate the extra
restrictions on the physical states. The additional part for the Hermitian operator l0 also contains an arbitrary parameter m20 which value will be defined from the condition of reproducing
the correct conditions (23). The massive parameter m1 remains arbitrary, in particular, it can be
expressed through the other massive parameters of the theory
m1 = f (m, r) = 0.
(48)
For example one can put m1 = m or m1 = , where is the inverse radius of the AdS space
2 = r. This arbitrariness does not affect on the equations for the basic vector (6).
We note that the operators of the additional parts do not satisfy the usual properties
(l0 )+ = l0 ,
(l1 )+ = l1+ ,
(l2 )+ = l2+
(49)
if we use the standard rules of Hermitian conjugation for the new creation and annihilation operators
(b1 )+ = b1+ ,
(b2 )+ = b2+ .
(50)
To restore the proper Hermitian conjugation properties for the additional parts we change the
scalar product in the Fock space generated by the new creation and annihilation operators
1 |2 new = 1 |K|2
(51)
with some unknown yet operator K. This operators is defined from the condition that all the
operators must have the proper Hermitian properties with respect to the new scalar product
1 |K l0 |2 = 1 |(l0 )+ K|2 ,
1 |Kl1 |2 = 1 |(l1+ )+ K|2 ,

1 |Kl1+ |2 = 1 |(l1 )+ K|2 ,
1 |Kg0 |2 = 1 |(g0 )+ K|2 ,
1 |Kl2 |2 = 1 |(l2+ )+ K|2 ,

1 |Kl2+ |2 = 1 |(l2 )+ K|2 .
(52)
(53)
(54)
The explicit expression for the operator K can be found using the method given in [25,30,31].
The calculations of the operator K are described in Appendix B.
354
6. The deformed algebra and BRST operator

In this section we find BRST operator and discuss the aspects stipulated by the nonlinearity
of the operator algebra.
The construction of BRST operator is based on following general principles (see the details
in reviews [20]):
(1) The BRST operator Q is Hermitian, Q+ = Q , and nilpotent, Q2 = 0.
(2) The BRST operator Q is built using a set of first class constraints. In the case under consid+
eration the operators L 0 , L1 , L+
1 , L2 , L2 , G0 are used as a set of such constraints.

(3) The BRST operator Q satisfies the special initial condition
+
+
Q |P =0 = 0 L 0 + 1+ L1 + 1 L+
1 + 2 L2 + 2 L2 + G G0 .
(55)
Here 0 , 1+ , 1 , 2+ , 2 , G are ghost coordinates with ghost number gh() = +1 and

P0 , P1 , P1+ , P2 , P2+ , PG are their canonically conjugate ghost momenta with ghost number
gh(P) = 1 satisfying the anticommutation relations

{0 , P0 } = {G , PG } = 1 , P1+ = 1+ , P1 = 2 , P2+ = 2+ , P2 = 1.
(56)
Our purpose here is to construct such an operator Q in an explicit form.
Let us turn to the algebra of the enlarged operators Li . Straightforward calculation of commutators (in accordance with Section 4) allows to find this algebra in the form9 given in Table 3,
where
[L1 , L 0 ] = ( )rL1 + 4rL+ L2 4rl + L2 4rl2 L+
1
+ 2rG0 L1 2rl1 G0 2rg0 L1 ,
(57)

+
+
+
+
L 0 , L+
1 = ( )rL1 + 4rL2 L1 4rl2 L1 4rl1 L2
+
+
+ 2rL+
1 G0 2rl1 G0 2rg0 L1 ,

+
+
L1 , L+
1 = L0 rG0 + 4(2 )r l2 L2 + l2 L2

2(2 )rg0 G0 + (2 )r G20 2G0 4L+
2 L2 .
(58)
(59)
The relations (57)(59) show that the symmetry algebra is nonlinear (quadratic). Using the
commutation relations one can write the right-hand sides of quadratic products of the operators
Table 3
Algebra of the enlarged operators
L 0
L1
L+
1
L2
L 0
L1
0
(57)
(57)
0
(58)
(59)
2 rL2
0
L+
1
2 rL+
2
L+
1
(58)
(59)
L1
L+
1
L2
2 rL2
L1
G0
2L2
L+
2
2 rL+
2
L+
1
G0
2L+
2
G0
L1
2L2
2L+
2
L+
1
L+
2
G0
0
L1
9 In what follows we forget about operator l, since its enlarged expression is zero l + l = 0. See formulas (26) and (45).
355
(57)(59) in various equivalent forms. Each form of writing the algebra can, in principle, lead
to different BRST charges. This means that in order to control the construction of the BRST
operator we should fix an ordering for quadratic products of operators in the algebra. We consider
a most general ordering prescription. All possible ways to order the operators in right-hand sides
of (57)(59) are described in terms of arbitrary real parameters 1 , 2 , 3 , 4 , 5 . The ordered
commutation relations look like
+
+
+
[L1 , L 0 ] = rL1 + (2 1 )rL+
1 L2 + (2 + 1 )rL2 L1 4rl1 L2 4rl2 L1
+ (1 2 )rG0 L1 + (1 + 2 )rL1 G0 2rl1 G0 2rg0 L1
+ (1 2 )rL1 ,

+
+
+
+
+
L 0 , L1 = rL+
1 + (2 + 3 )rL2 L1 + (2 3 )rL1 L2 4rl2 L1 4rl1 L2
(60)
+ (4 3 )rL+
1,

+

L1 , L1 = L 0 rG0 + 4(2 )r l2+ L2 + l2 L+
2 2(2 )rg0 G0
2
+
+ (2 )r G0 5 G0 (2 + 5 )L+
2 L2 (2 5 )L2 L2 .
(61)
+
+
+
+ (1 + 4 )rL+
1 G0 + (1 4 )rG0 L1 2rl1 G0 2rg0 L1
(62)
Before we begin a construction of the BRST operator one points out that the algebra of the
enlarged operators in the case under consideration is very similar to the one considered in [26].
The only difference between our case and [26] consists in the symmetry property of constants
fijkm (35). In [26] this quantity is symmetric in upper indices fijkm = fijmk , but in the present paper
we leave them with arbitrary symmetry by means of introducing the arbitrary parameters i . At
i = 0 we have the symmetrized product of the operators in rhs of (60)(62) and our algebra will
be a partial case of the algebra considered in [26]. It was proved in [26] that the BRST operator
includes the terms of first, thirds and seventh order in ghosts. Therefore one can expect in our
case the analogous terms in the BRST operator plus some extra terms stipulated by the difference
in symmetry of the constants fijkm .
Now let us turn to construction of the BRST operator. Demanding that the BRST operator be
Hermitian (with respect to the new scalar product (51)), we find equations on i . These equations
give us only one arbitrary coefficient which we denote , the others are expressed through it as
follows
= 1 = 2 = 3 = 4 ,
5 = 0.
(63)
In this case the commutators (60)(62) take the form

+
+
+
[L1 , L 0 ] = rL1 + (2 )rL+
1 L2 + (2 + )rL2 L1 4rl1 L2 4rl2 L1
+ (1 )rG0 L1 + (1 + )rL1 G0 2rl1 G0 2rg0 L1 ,

L 0 , L+ = rL+ + (2 + )rL+ L1 + (2 )rL1 L+ 4rl + L1 4rl1 L+
1
+
+
+
+ (1 + )rL+
1 G0 + (1 )rG0 L1 2rl1 G0 2rg0 L1 ,

+

+
L1 , L +
1 = L0 rG0 + 4(2 )r l2 L2 + l2 L2 2(2 )rg0 G0

2
+
+ (2 )r G0 2L+
2 L2 2L2 L2 .
(64)
(65)
(66)
To find the BRST operator we write it as a Hermitian operator which is a seventh degree
polynomial in ghosts and demand its nilpotency. Such a procedure leads to the following result
356
for the BRST operator Q

+
+
+
+
Q = 0 L 0 + 1+ L1 + 1 L+
1 + 2 L2 + 2 L2 + G G0 1 1 P0 2 2 PG

+ G 1+ + 2+ 1 P1 + 1 G + 1+ 2 P1+ + 2G 2+ P2 + 22 G P2+

+ r1+ 0 4l1+ P2 + 4l2 P1+ + 2g0 P1 + 2l1 PG

+ r0 1 4l1 P2+ + 4l2+ P1 + 2g0 P1+ + 2l1+ PG

+
r1+ 0 (2 )L+
1 P2 + (2 + )L2 P1 + (1 )G0 P1 + (1 + )L1 PG

+
+
r0 1 (2 )L1 P2+ + (2 + )L+
2 P1 + (1 )G0 P1 + (1 + )L1 PG

+
+

+
(2 )r1+ 1 G0 PG 2L+
2 P2 2L2 P2 2g0 PG + 4l2 P2 + 4l2 P2

+ r1+ 1 PG + r0 1+ P1 1 P1+ + 22+ P2 22 P2+

+ 3 r0 1 2 P1+ P2+ + 1+ 2+ P1 P2 + 2+ 1 P1+ P2

+ 1+ 2 P2+ P1 + 2+ 1 PG P1 + 1+ 2 P1+ PG .
(67)
It is interesting to compare this result for the BRST operator with the result in [26]. We see that
the BRST operator (67) unlike [26] has no terms of the seventh degree in the ghosts (three ghost
momenta and four ghost coordinates). The matter is that, according to [26] the presence (or the
absence) of such terms in BRST operator depends on the number of operators whose commutators have quadratic rhs. To get the seventh degree terms we need at least four such operators.
In our case we get only three operators whose commutators have quadratic rhs (64)(66). These
commutators are of L 0 , L1 , L+
1 operators, therefore we have only three ghost coordinates 0 ,
1+ , 1 whereas the terms of seventh degree in ghosts demand at least four ghost coordinates.
New point is that in comparison with [26] the terms of fifth degree in ghosts appear in the BRST
operator. Their origin is another symmetry property of the constants fijkm in the case under consideration in comparison with [26]. If we put = 0 (what corresponds to the symmetric ordering
of the operators in rhs of (64)(66)) the BRST operator Q (67) has no terms of the fifth degree
in ghosts as it should be in accordance with [26].
Further on we will show that the arbitrariness in the BRST operator stipulated by the parameter is resulted in arbitrariness of introducing the auxiliary fields in the Lagrangians and hence
does not affect the dynamics of the basic field (6).
7. Construction of Lagrangians
To construct the Lagrangian we use the procedure developed in [31] for massive bosonic
higher spin fields in the flat space. According to this procedure we should extract dependence of
Q (67) on the ghosts G , PG

Q = Q + G ( + h) 2+ 2 PG + 2r0 1 l1+ 1+ l1 PG

+

+ (1 + )r0 1+ L1 1 L+
1 PG (2 )r1 1 (G0 2g0 )PG

+ r1+ 1 PG + 3 r0 1+ 2 P1+ 2+ 1 P1 PG ,
(68)
where
+ h G0 + 1+ P1 1 P1+ + 22+ P2 22 P2+ ,
[, Q] = 0,
(69)
357
+
+
+
+
+
+
Q = 0 L 0 + 1+ L1 + 1 L+
1 + 2 L2 + 2 L2 1 1 P0 + 2 1 P1 + 1 2 P1

+ r1+ 0 4l1+ P2 + 4l2 P1+ + 2g0 P1 + r0 1 4l1 P2+ + 4l2+ P1 + 2g0 P1+

+
r1+ 0 (2 )L+
1 P2 + (2 + )L2 P1 + (1 )G0 P1

+
r0 1 (2 )L1 P2+ + (2 + )L+
2 P1 + (1 )G0 P1

+
+
+
+ (2 )r1+ 1 2L+
2 P2 + 2L2 P2 4l2 P2 4l2 P2

+ r0 1+ P1 1 P1+ + 22+ P2 22 P2+

+ 3 r0 1 2 P1+ P2+ + 1+ 2+ P1 P2 + 2+ 1 P1+ P2 + 1+ 2 P2+ P1 .
(70)
We consider that the ghost operators act on the vacuum state as follows
P0 |0 = PG |0 = 1 |0 = P1 |0 = 2 |0 = P2 |0 = 0
(71)
ghost10
and suppose [31] that the vectors and the gauge parameters do not depend on the
G . It
is this condition which gives us a possibility to find the values of the parameter h in (39)(44).
As a result we have from the equation defining the physical vectors Q | = 0, where | is a
vector in the extended space (including all ghosts except G )

Q| = 0,
( + h)| = 0,
gh | = 0,
(72)

| = Q|,
( + h)| = 0,
gh | = 1,
(73)

| = Q|,
( + h)| = 0,
gh | = 2.
(74)
Since we cannot write a gauge parameter with ghost number 3, the chain of gauge transformation is finite.
Let us discuss the Lagrangian construction for the field with a given value of the spin s. The
middle equation of (72) is equation for defining the possible values of the arbitrary parameter h.
One can see that it takes the values h = s + d2 3 and these values are connected with the
spin of the field. Having fixed a value of the spin we define the parameter h. This value of h is
substituted into the other Eqs. (72)(74) (including operator Q). Let us denote Qs the operator
Q (70) where we substituted s + d2 3 instead of h
Qs = Q|hs+ d 3
2
(75)
and let us denote the eigenvectors of the operator (69) corresponding to the eigenvalue s + d2 3
as |s

d
|s = s + 3 |s ,
Q2s |s 0.
(76)
2
Thus for a given spin-s we have from (72)(74) the equation of motion and the gauge transformation
Qs | s = 0,
(77)
| s = Qs |s ,
(78)
|s = Qs |s
(79)
10 Equivalently, this condition can be written as P | = 0, with operator P

G
G being in involution with Q:
{PG , Q} = 0.
358
and the middle equations of (72)(74) are satisfied by the construction.

Next step [31] is to extract the Hermitian ghosts 0 , P0 in Eqs. (77), (73), (79). We have11
+ P ,
Qs = Q + 0 L
0
1 1 0
(80)
Q = 1+ L1
+
+
+
+
+
+ 1 L+
1 + 2 L2 + 2 L2 + 2 1 P1 + 1 2 P1

+
+
+
+ (2 )r1+ 1 2L+
2 P2 + 2L2 P2 4l2 P2 4l2 P2 ,
= L r+ 4l + P + 4l P + + 2g P + r 4l P + + 4l + P + 2g P +
L
0
0
2
1
1
2 1
0 1
1 2
0 1
1
1
2
+
+
+
+ r1 (2 )L1 P2 + (2 + )L2 P1 + (1 )G0 P1

+
r1 (2 )L1 P2+ + (2 + )L+
2 P1 + (1 )G0 P1

+ r 1+ P1 1 P1+ + 22+ P2 22 P2+

+ 3 r 1 2 P1+ P2+ + 1+ 2+ P1 P2 + 2+ 1 P1+ P2 + 1+ 2 P2+ P1
(81)
(82)
for the operator Qs and

| = |S + 0 |A,
| = |0 + 0 |1 ,
(83)
| = |0
(84)
for the gauge parameters. As a result we have the equations of motion and the gauge transformation
|S Q|A = 0,
Q|S + |A = 0,
L
(85)
0
|S = Q|0 1+ 1 |1 ,
|0 = Q|,
| Q| ,
|A = L
0 0
1
|.
|1 = L
0
(86)
(87)
It is easy to show that the equations of motion (85) can be derived from the following Lagrangian12

0 |S Q|A + A|K Q|S + + 1 |A
Ls = S|K L
(88)
1
which can also be written in a more compact form

Ls = d0 s |KQs | s .
(89)
Now we fix the arbitrary parameter m20 . It is defined from the condition of reproducing the
conditions (23) for the basic vector | (6). The general vector | includes the basic vector |
as follows
| = | + |A ,
(90)
where the vector |A includes only the auxiliary fields13 as its components. One can show,
using the part of equations of motion and gauge transformations, that the vector |A can be
11 Here and further ones assume that we substituted s + d 3 instead of h in all the operators l .
i
2
12 The Lagrangian is defined as usual up to an overall factor.
13 A field is auxiliary if it is associated not only with creation operator a + but at least with one of the ghost operators
0 , 1+ , 2+ , P1+ , P2+ or with one of the additional creation operators b1+ , b2+ . The basic (physical) field is associated
+ only (6).
with the creation operators a
359
completely removed. The details are given in Appendix C. As a result we obtain the equation of
motion for the basic vector (77) in the form

l0 m2 + M 2 | = l1 | = l2 | = 0,
(91)
where M 2 is defined in (46). We see that the Lagrangian reproduces the basic conditions (23) if
M 2 = m2 . It leads to
m20 =
d(d 4)
r ( 2)h(h 2)r + hr.
4
(92)
It is interesting to note that in case of = = 0, = 2 we will have m20 = 0. Such a situation

was considered in [23] where the higher spin massless bosonic fields in AdS space were studied.
One can prove that the Lagrangian (89) indeed reproduces the basic conditions14 (23). The
details of the proof are given in Appendix C. The ratio (89) where the parameter m20 is fixed by
(92) is our final result.
Construction of the Lagrangian describing propagation of all massive bosonic fields in AdS
space simultaneously is analogous to that in the flat space [31] and we do not consider it here.
Let us turn to examples.
8. Examples
8.1. Spin-0 field
For a scalar field with spin s = 0 we have h = 3 d2 . Then the vector | s which satisfies the
condition (76) and has the proper ghost number (72) looks like
| 0 = (x)|0.
Then using (92), (B.32) the relation for Lagrangian (88) gives

L0 = 2 + r(6 2d) + m2 .
(93)
(94)
It is easy to see that this Lagrangian reproduces Eq. (2) for s = 0.

8.2. Spin-1 field
For a vector field we have h = 2 d2 . Then the vector | s and the gauge parameter |s ,
which satisfy the condition (76) and have the proper ghost numbers (72), (73), look like

| 1 = ia + A (x) + b1+ A(x) |0 + 0 P1+ (x)|0,
(95)
|1 = P1+ (x)|0.
(96)
14 The reason why the theories with different parameters , , reproduce the correct equations (2)(4) is that we
can absorb extra terms proportional to r with the help of redefinition of the mass. For the mass m2 will be not shifted
we introduced arbitrary parameter m20 with dimension of mass squared which cancels the extra terms proportional to r.
Having fixed parameter m20 (92) the equations of motion for the basic vector | become independent of the parameters
, , .
360
Using (92), (B.32) we find the Lagrangian (88) for the vector field

L = A 2 r(d 1) + m2 A

m2 2
m2
2
+ 2 A 4r + m A m1 + A
A
m1
m1
(97)
and the gauge transformations (86)

A (x) = ,
A(x) = m1 ,

(x) = 2 + m2 .
(98)
The relations (97) and (98) are the final result for the massive vector field. We obtain the gauge
invariant theory in terms of vector field A and two auxiliary scalar fields A and .
Let us consider the massless limit of this theory. To do that we put m 0 in (97).15 Then the
field A vanishes and we obtain the theory of the massless field s = 1 in AdS space.
Now we show how the Lagrangian (97) is transformed to the conventional form. Let us rescale
the field A
m
A = A A.
(99)
m1
In this case Lagrangian (97) and the gauge transformation (98) are rewritten as

L = A 2 + m2 r(d 1) A

+ A 2 + m2 A m + A mA ,

A (x) = ,
A(x) = m,
(x) = 2 + m2
(100)
(101)
and they become independent of m1 .

One excludes the field (x) from Lagrangian (100) with the help of its equation of motion. It
leads to

1
L = F F (mA A) mA A ,
(102)
2
where
F = A A .
(103)
Now removing the field A with the help of its gauge transformation we arrive at the conventional
form of Lagrangian for spin-1 field (up to an overall factor)
1
L = F F m2 A A .
2
(104)
8.3. Spin-2 field

For spin-2 field we have h = 1 d2 . Then the vector | s and the gauge parameter |s ,
which satisfy the condition (76) and have the proper ghost numbers (72), (73), look like

(i)2 + +
+
+2
+ +
|S1 =
(105)
a a H (x) + b2 H1 (x) ia b1 A (x) + b1 (x) |0,
2
15 Ones remind that arbitrary parameter m arose in Section 5 when we constructed the additional parts of the operators.
1
In general it has no relation to physical mass m and should not tend to zero in massless limit.
|S2 = H (x)|0,

|A1 = ia + H (x) + b1+ A(x) |0,

|1 = ia + (x) + b1+ (x) |0,
361
(106)
|A2 = H2 (x)|0,
|2 = 2 (x)|0.
(107)
(108)
Here |Si , |Ai , |i are defined in (C.1)(C.4).

Using (92), (88), (B.32) we find Lagrangian for the massive spin-2 field in the form

1
L = H 2 + m2 2r H 2( 1)rg H
2

+ (2 + )rg H + g H2 2 H

2

d 2
m2
H1 + 2
+ m2 + 2r(d 1) H1
2
m1

d 2
m2
H1 + 2 + (2 )rH H2
4r
2
m1

m2 2
A + m2 + r(d 1) A A m1 H
m21

2

m2 2m2 + 2r(d 1)
+ m2 + 2r(d 1) m1 A
+ 2
H
1
2
m1
m1

m2
H 2 + m2 + 2r(d 1) + 2r H H
A + H2
m1

1
d 2
m2
H1 + r(2 ) 2
(2 + )rH + r(2 )
2
2
m1

m2
A
+ H H H H +
m1

m2
2m2 + 2r(d 1)
2 A A m1 (H + H1 ) A +
m1
m1

1
d 2
m2
H1 + 2 H .
+ H2 H +
2
2
m1
(109)
Using (86), (92) ones find the gauge transformations

H = + g 2 ,
A = + m1 ,
H1 = 2 ,
= m1 ,
m2
H = +
2 ,
m1
2

H = + m2 + r(d 1) ,

A = 2 + m2 + 2r(d 1) ,
m2
H2 = (2 )r (2 + )r
m1

+ 2 + m2 + 2r(d 1) + r(2 2d + ) 2 .
(110)
(111)
(112)
(113)
(114)
(115)
362
The relation (109) and the above gauge transformations are our final result for massive spin-2
theory in AdS space. We get the gauge theory in terms of basic filed H and a number of
auxiliary fields.
In massless limit m 0 16 in (109) we get Lagrangian for the massless spin-2 field in terms
of the fields H , H , H , H1 and H2 .
Now we rewrite the Lagrangian (109) in more conventional form. Let us make the following
transformation. First, we redefine the fields
m2
d 2
d 2
d 2
H1 + 2 =
H1
H1 ,
2
2
2
m1
m
A = A A,
m1
m
A = A A ,
m1
(116)
m2
= ,
m21
(117)
and the gauge parameters

m
(118)
= .
m1
Then we use the equations of motion for H , A in (109) and remove field H1 with the help of its
gauge transformation. After that the gauge parameters and 2 are not independent
2m
= 0.
(119)
d 2
Next we use the equations of motion for the fields H2 and H in previously obtained Lagrangian.
Finally, after one more rescaling the field
1/2

r
1
+
=
2(d 1)
(120)
d 2 m2
2 +
we arrive at Zinovievs Lagrangian (up to an overall sign) [18]

1
1
L = H 2 + m2 2r H H 2 + m2 + r(d 3) H
2
2

+ H H H

2

d
2
2
m
A + r(d 1) A A +
d 2

2
1/2

m
+ 2(d 1)
+r
mH 2 A
d 2

+ 2m H A H A
(121)
with the gauge transformations

H = + + g
2m
,
d 2
A = + m ,

2
1/2
m
= 2(d 1)
+r
.
d 2
16 We assume as in Section 8.2 that the arbitrary parameter m does not tend to zero in this limit.
1
(122)
(123)
(124)
363
It is easy to see that at

r+
m2
=0
d 2
(125)
Lagrangian (121) splits into a sum of two independent parts with helicities 2, 1 (H , A )
and the scalar field . Thus the field H becomes partial massless at
m2 = (d 2)r
(126)
in accordance with [3].

9. Summary
We have developed the gauge invariant approach to deriving the Lagrangians for bosonic
massive higher spin fields in arbitrary dimensional AdS space. The Lagrangian includes the
Stueckelberg fields, providing the gauge invariance in the massive theory, and the complete set of
the auxiliary fields which should be introduced for Lagrangian formulation in higher spin theory.
The approach is based on BRST construction for special nonlinear symmetry algebra formulated
in the paper.
We begin with embedding the massive higher spin fields into the vectors of auxiliary Fock
space. All such fields are treated as the components of the vectors of the Fock space and the
theory is completely formulated in terms of such vectors. The conditions defining the irreducible
representation of the AdS group with given mass and spin are realized by differential operators
acting in this Fock space. The above conditions are interpreted as the constraints on the vectors
of the Fock space and generate the closed nonlinear symmetry algebra which is the main object
of our analysis. As we proved, derivation of the correct Lagrangian demands an extension and
deformation of the algebra. BRST construction for such nonlinear symmetry algebra is given.
It is shown that the BRST operator corresponding to the extended deformed algebra defines the
consistent Lagrangian dynamics for bosonic massive fields of any value of spin. We constructed
the gauge invariant Lagrangians in terms of Fock space in the concise form for any higher spin
fields propagating in AdS space of arbitrary dimension. It is interesting to point out that the
gauge transformations in the theory under consideration are reducible, the corresponding order
of reducibility is equal to unit. As an example of the general scheme we obtained the gauge
invariant Lagrangians and the corresponding gauge transformations for the component spin-0,
spin-1, and spin-2 massive fields in the explicit form. The Lagrangians for component massive
fields with other spins can be analogously found using the simple enough manipulations with
creation and annihilation operators in the Fock space.
The main results of the paper are given by the relations (88), (89) where Lagrangian for the
massive field with arbitrary integer spin is constructed, and (86), (87) where the gauge transformations for the fields and the gauge transformations for the gauge parameters are written down.
The procedure for Lagrangian construction developed here for higher spin massive bosonic
fields can also be applied to fermionic higher spin theories in AdS background and for fields
with mixed symmetry tensor or tensor-spinor fields (see [25] where bosonic massless fields with
mixed symmetry were considered in Minkowski space). The results obtained open a principle
possibility for derivation of the interacting vertices for massive higher spin fields in AdS space.
364
Acknowledgements
The authors are grateful to R.R. Metsaev, H. Takata, M. Tsulaia, and M.A. Vasiliev for discussions. The work was partially supported by the INTAS grant, project INTAS-03-51-6346, the
RFBR grant, project No. 06-02-16346 and grant for LRSS, project No. 4489.2006.2. Work of
I.L.B. and P.M.L. was supported in part by the DFG grant, project No. 436 RUS 113/669/0-3 and
joint RFBR-DFG grant, project No. 06-02-04012.
Appendix A. Differential calculus in auxiliary Fock space
We consider an arbitrary d-dimensional Riemann space with metric g (x) which can be
b (x)
expressed in terms of the vielbein e
b
g (x) = bc e
(x)ec (x),
a
e
(x)eb (x) = ba ,
g (x)g (x) = ,
eb (x) = g (x)bc ec (x),
bc cd = bd ,
(A.1)
where bc is the Minkowski metric with the signature bc = (+, , . . . , ).

We introduce the creation and annihilation operators ab+ and ab carrying flat-space indices b

+ +

[ab , ac ] = 0,
ab , ac = 0,
ab , ac+ = bc ,
(A.2)
ab |0 = 0,
0|0 = 1,
a b = a b+ = 0,
(A.3)
where |0 is the vacuum vector of the Fock space. Any s-particle vector | s can be presented in
the form

s
= b ...b (x)a b1 + a bs + |0,
N s = s s ,
s
1
N = a b+ ab ,
a b+ = bc ab+ ,
(A.4)
where b1 ...bs (x) is a symmetric tensor field of rank s. It is useful to introduce the creation and
annihilation operators a+ (x) and a (x) by the relations
b
a+ (x) = e
(x)ab+ ,
b
a (x) = e
(x)ab .
(A.5)
They obey the properties

+

a (x), a+ (x) = 0,
a (x), a (x) = 0,

a (x)|0 = 0
a (x), a+ (x) = g (x),
and create the s-particle vector
s
= ... (x)a 1 + (x) a s + (x)|0,
s
1
N = a + (x)a (x),
a + = g (x)a+ (x),
(A.6)

N s = s s ,
(A.7)
where 1 ...s (x) is a symmetric tensor in the Riemann space.

One introduces the operator D acting on any operator tensor with curved-space and with
flat-space indices by the rule

c+

b
D Q
= Q c a ab , Q ,
365
where the dots denote the curved-space and flat-space indices. The is the proper covariant
derivative. It is clear that this operator possesses the Leibniz rule
D (A B) = (D A) B + A (D B).
(A.8)
From the definition of D it follows that it acts on usual tensor fields (which do not depend
on creation and annihilation operators) as the covariant derivative
D 1 2 ...s (x) = 1 2 ...s (x),
D b1 b2 ...bs (x) = b1 b2 ...bs (x).
(A.9)
In particular we have
D g (x) = g (x) = 0.
(A.10)
Let us consider the action of D on the creation and annihilation operators with flat-space indices. Ones obtain

D a b = a b + b c a c d c a d+ ac , a b = b c + c b a c = 0,
D a b+ = 0,
(A.11)
= c . Therefore

D a = eb a b + eb a b d c ed a c+ ab , a d = eb + eb + b c ec a b

= eb a b = 0,
because of
ab
a b+
= 0 and
D a + = 0.
(A.12)
The property of covariant constancy of the creation and annihilation operators (A.11), (A.12)
allows us to consider the operator D as representation of the covariant derivative on vectors
| s if we assume that D |0 = 0. Indeed, we have

D s = 1 2 ...s (x) a 1 + (x) a s + (x)|0

= b1 b2 ...bs (x) a b1 + a bs + |0.
(A.13)
In a similar way we can find the action of the commutator of the covariant derivative on the
vectors (A.7)

[D , D ] s = R a + a s = Rb c a c+ ab s ,
(A.14)
where we used notations
R = + ,
R
d
b
d c
(A.15)
(A.16)
for the curvature tensors.

Appendix B. Calculation of the additional parts
In this appendix we show how the representation of the algebra given in Table 2 can be constructed in terms of some creation and annihilation operators.
Let us consider a representation of this algebra with the vector |0V annihilated by the operators l1 and l2
l1 |0V = l2 |0V = 0
(B.1)
366
and being the eigenvector of the operators l0 , g0 and l

l0 |0V = m20 |0V ,
g0 |0V = h|0V ,
l |0V = m22 |0V ,
(B.2)
where m0 , m2 are arbitrary constants with dimension of mass and h is a dimensionless arbitrary
constant.17 They are the arbitrary constants which must be contained in the additional parts of
Hermitian operators. Next we choose the basis vectors as follows
+ n1
+ n2
l
|n1 , n2 V = 1
(B.3)
l2
|0V ,
m1
where m1 is an arbitrary nonzero constant with dimension of mass. It may be constructed from
the parameters of the theory m1 = f (m, r) = 0. It is easy to find how the operators l , l1+ , l2+ ,
g0 act on these basis vectors
l |n1 , n2 V = m22 |n1 , n2 V ,
(B.4)
l1+ |n1 , n2 V = m1 |n1 + 1, n2 V ,

l2+ |n1 , n2 V = |n1 , n2 + 1V ,
g0 |n1 , n2 V = (n1 + 2n2 + h)|n1 , n2 V .
(B.5)
(B.6)
(B.7)
Calculation of how the rest operators l0 , l1 , l2 acts on the basis vectors is much more difficult.
First we find that
n
l0 |n1 , n2 V = 2 rn2 |n1 , n2 V + l2+ 2 l0 |n1 , 0V ,
(B.8)
+ n2

l1 |n1 , n2 V = m1 n2 |n1 + 1, n2 1V + l2
(B.9)
l1 |n1 , 0V ,
+ n2

l2 |n1 , n2 V = (n2 1 + n1 + h)n2 |n1 , n2 1V + l2
(B.10)
l2 |n1 , 0V .
Thus it remains to calculate how the operators l0 , l1 , l2 act on the basis vectors |n1 , 0V .
To do this we define some auxiliary operators and their action on the vector |0V
K0 g0 2 2g0 4l2+ l2 ,
K0 |0V = h(h 2)|0V ,

+
+
+
K1 K0 , l1 = 4l2 l1 + 2l1 g0 l1+ ,
K1 |0V = m1 (2h 1)|1, 0V .

K2 K1 , l1+ = 2l1+2 + 4l2+ K2 ,
d(d 4)
r + ( 2)rK0 rg0 ,
K2 = l0 + m2 +
4
K2 |0V = M 2 |0V ,
K2 |0V = 4M 2 |0, 1V + 2m21 |2, 0V ,
(B.11)
(B.12)
(B.13)
(B.14)
(B.15)
where
d(d 4)
r + ( 2)rh(h 2) rh.
4
In terms of these operators we have
+
+
l0 , l1 = rK1 + rl1+ ,
l1 , l1 = K2 ,
M 2 = m2 + m20 +
(B.16)
17 In the mathematical literature the representation given by (B.1) and (B.2) is called the Verma module. It explains the
subscript V at the vectors.

K2 , l1+ = 8l2+ rK1 .
367
(B.17)
Using the above formulas we obtain

l0 |n1 , 0V = m0 |n1 , 0V + rn1 |n1 , 0V r(2h 1)
2r
8r k
k=0
where Ckn =
n!
k!(nk)! .
m21
8r k
k=0
m21
n1
C2k+1
|n1 2k, kV
n1
C2k+2
|n1 2k, kV

4rM 2 8r k n1
C2k+2 |n1 2k 2, k + 1V ,
m21 k=0 m21
(B.18)
Substituting (B.18) into (B.8) ones get
l0 |n1 , n2 V = m0 |n1 , n2 V + r(n1 + 2n2 )|n1 , n2 V

8r k n
1
r(2h 1)
C2k+1
|n1 2k, n2 + kV
2
m
1
k=0
2r
8r k
k=0
m21
n1
C2k+2
|n1 2k, n2 + kV

4rM 2 8r k n1
C2k+2 |n1 2k 2, n2 + k + 1V .
m21 k=0 m21
(B.19)
Analogously one can first find

l1 |n1 , 0V

2h 1 8r k n1
= m1
C2k |n1 2k + 1, k 1V
4
m21
k=1

m1 8r k n1
+
C2k+1 |n1 2k + 1, k 1V
2
m21
k=1

M 2 8r k n1
+
C2k+1 |n1 2k 1, kV ,
m1
m21
k=0
l2 |n1 , 0V =

1 2h 8r k n1
C2k+1 |n1 2k, k 1V
4
m21
k=1

1 8r k n1
C2k+2 |n1 2k, k 1V

2
m21
k=1

M 2 8r k n1
2
C2k+2 |n1 2k 2, kV
m1 k=0 m21
(B.20)
(B.21)
368
and then substituting (B.20) and (B.21) into (B.9) and (B.10) respectively we find how operators
l1 and l2 act on the basis vectors |n1 , n2 V
l1 |n1 , n2 V = m1 n2 |n1 + 1, n2 1V

2h 1 8r k n1
+ m1
C2k |n1 2k + 1, n2 + k 1V
4
m21
k=1

m1 8r k n1
+
C2k+1 |n1 2k + 1, n2 + k 1V
2
m21
k=1

M 2 8r k n1
+
C2k+1 |n1 2k 1, n2 + kV ,
m1
m21
k=0
l2 |n1 , n2 V = (n2 1 + n1 + h)n2 |n1 , n2 1V

2h 1 8r k n1
C2k+1 |n1 2k, n2 + k 1V

4
m21
k=1

1 8r k n1
C2k+2 |n1 2k, n2 + k 1V

2
m21
k=1

M 2 8r k n1
2
C2k+2 |n1 2k 2, n2 + kV .
m1 k=0 m21
(B.22)
(B.23)
Now let us turn to construction of a representation of the operator algebra given in Table 2
in terms of creation and annihilation operators. The number of pairs of these operators and their
statistics is defined by the number of operators used in the definition of the basis vectors (B.3).
Thus we introduce two pairs of the bosonic creation and annihilation operators with the standard
commutation relations

b1 , b1+ = 1,

b2 , b2+ = 1,
(B.24)
corresponding to l1 , l1+ and l2 , l2+ respectively. After this we map the basis vectors (B.3) and the
basis vectors of the Fock space generated by b1+ , b2+
n n
|n1 , n2 V b1+ 1 b2+ 2 |0 = |n1 , n2
(B.25)
and find from (B.4)(B.7), (B.19), (B.22), (B.23) form of the operators in terms of the creation
and annihilation operators
l = m22 ,
l2+ = b2+ ,
l1+ = m1 b1+ ,
g0 = b1+ b1 + 2b2+ b2 + h,

l0 = m20 + r b1+ b1 + 2b2+ b2 r(2h 1)b1+
(B.26)
(B.27)

k
b2+k b12k+1
8r
(2k + 1)!
m21
k=0

8r k b2+k b12k+2 1
8r k b2+k b12k
2
+
M
,
2rb1+2
(2k + 2)! 2
(2k)!
m21
m21
k=0
k=1
(B.28)
l1
= m1 b1+ b2
2h 1
+ m1 b1+
4
369

8r k (b+ )k1 b2k
2
k=1
m21
(2k)!

8r k (b2+ )k1 b12k+1 M 2 8r k b2+k b12k+1
1
+
,
+ m1 b1+2
2
(2k + 1)!
m1
(2k + 1)!
m21
m21
k=1
k=0

+

2h 1 + 8r k (b2+ )k1 b12k+1
+

l2 = b 2 b 2 + b 1 b 1 + h b 2
b1
4
(2k + 1)!
m21
k=1

1 +2 8r k (b2+ )k1 b12k+2 M 2 8r k b2+k b12k+2

2
.
b1
2
(2k + 2)!
(2k + 2)!
m21
m1 k=0 m21
k=1
(B.29)
(B.30)
Finally we put the arbitrary constant m22 to m22 = m2 d(d4)

4 r. Thus we have obtained the
expressions of the additional parts (39)(45) for the operator algebra given in Table 2.
Let us find an explicit expression for the operator K which is used in the definition of the scalar
product (51). The defining relations for this operator are given by (52)(54). These relations shall
be satisfied for any | if they shall be satisfied for the basis vectors of the Fock space |n1 , n2 .
Then using arguments of [25,30] one can check that the following operator satisfy (52)(54)
K = Z + Z,
Z=
|n1 , n2 V
n1 ,n2 =0
1
n1 , n2 |.
n1 !n2 !
(B.31)
+2n2
For practical calculations it is useful to note that V n1 , n2 |n1 , n2 V nn1+2n
. For low numbers
1
2
n1 + 2n2 the operator K is
K = |00| +
+

M2 +
M 2 +2
+
b1 |00|b2 + b2+ |00|b12
b
|00|b
+
hb
|00|b
1
2
1
2
2
2
m1
2m1
M 2 M 2 + r(1 2h) +2
b1 |00|b12 + .
m21
2m21
(B.32)
This expression for the operator K is used when we give the examples.
Appendix C. Removing of the auxiliary fields
In this appendix we explain how the conditions (23) on basic vector (6) can be obtained from
equations of motion (85) following from Lagrangian (89).
First we explicitly extract the dependence of the fields and the gauge parameters on the ghost
fields
|S = |S1 + 1+ P1+ |S2 + 1+ P2+ |S3 + 2+ P1+ |S4 + 2+ P2+ |S5
+ 1+ 2+ P1+ P2+ |S6 ,
(C.1)
|A = P1+ |A1 + P2+ |A2 + 1+ P1+ P2+ |A3 + 2+ P1+ P2+ |A4 ,
(C.2)
| = |0 + 0 |1 ,
(C.3)
|0 = P1+ |1 + P2+ |2 + 1+ P1+ P2+ |3 + 2+ P1+ P2+ |4 ,

|1 = P1+ P2+ |5 ,
| = P1+ P2+ |.
(C.4)
(C.5)
(C.6)
370
In this notation the Lagrangian (89) takes the form

+
+
+
L = S1 |K L 0 |S1 L+
1 |A1 L2 |A2 r (2 + )L2 4l2 |S2

r (1 )G0 2g0 |S1 3 r|S1 3r |S1

S2 |K L 0 |S2 L1 |A1 + |A2 L+
2 |A3

+ r (1 )G0 2g0 |S2 + r (2 + )L2 4l2 |S1

+
+ r (2 )L+
1 4l1 |S4 r|S2 + 3r |S2

S4 |K L 0 |S3 L1 |A2 + L+
1 |A3

+

+ r (2 )L1 4l1 |S2 + r (2 )L+
1 4l1 |S5

+
+ (2 )r 2L2 4l2 |A1 (2 )r 2L+
2 4l2 |A4

S3 |K L 0 |S4 L2 |A1 L+
2 |A4

S5 |K L 0 |S5 L2 |A2 |A3 + L+
1 |A4

+ r (2 )L1 4l1 |S4 r (1 )G0 2g0 |S5

+
+ r (2 + )L+
2 4l2 |S6 + r|S5 + 3r |S5

+ S6 |K L 0 |S6 + L2 |A3 L1 |A4 + r (1 )G0 2g0 |S6

r (2 + )L2 4l2 |S5 + 3 r|S6 3r |S6

+

+
+
A1 |K L1 |S1 L+
1 |S2 L2 |S3 |A1 + (2 )r 2L2 4l2 |S4

+
A2 |K L2 |S1 + |S2 L+
2 |S5 L1 |S4

+ A3 |K L2 |S2 + |S5 L1 |S4 + L+
2 |S6

A4 |K L1 |S5 L2 |S3 + L+
1 |S6 + |A4 (2 )r[2L2 4l2 ]|S4 .
(C.7)
Here we see at = = 0 (these values do not correspond to that which give l0 (17)) many terms
in (C.7) disappear and the expression for the Lagrangian is simplified.
The equations of motion which follow from Lagrangian (C.7) are

+
+
+
L 0 |S1 L+
1 |A1 L2 |A2 r (2 + )L2 4l2 |S2

r (1 )G0 2g0 + 3 |S1 3 r|S1 = 0,
(C.8)

+

L 0 |S2 L1 |A1 + |A2 L2 |A3 + r (1 )G0 2g0 + 3 |S2 r|S2

+
+ r (2 + )L2 4l2 |S1 + r (2 )L+
(C.9)
1 4l1 |S4 = 0,

+
+
+

L 0 |S3 L1 |A2 + L1 |A3 + r (2 )L1 4l1 |S2 + r (2 )L1 4l1 |S5

+
+ (2 )r[2L2 4l2 ]|A1 (2 )r 2L+
(C.10)
2 4l2 |A4 = 0,
L 0 |S4 L2 |A1 L+
2 |A4 = 0,

L 0 |S5 L2 |A2 |A3 + L+
1 |A4 r (1 )G0 2g0 3 |S5 + r|S5

+
+ r (2 )L1 4l1 |S4 + r (2 + )L+
2 4l2 |S6 = 0,

L 0 |S6 + L2 |A3 L1 |A4 + r (1 )G0 2g0 3 |S6 + 3 r|S6

r (2 + )L2 4l2 |S5 = 0,
+
+
+
L1 |S1 L+
1 |S2 L2 |S3 |A1 + (2 )r 2L2 4l2 |S4 = 0,
(C.11)
(C.12)
(C.13)
(C.14)

+
L2 |S1 + |S2 L+
2 |S5 L1 |S4 = 0,
371
(C.15)
L2 |S2 + |S5 L1 |S4 + L+

2 |S6 = 0,
+
L1 |S5 L2 |S3 + L1 |S6 + |A4 (2 )r[2L2
(C.16)
4l2 ]|S4 = 0.
(C.17)
These equations of motion and the Lagrangian (C.7) are invariant under the gauge transformations
+
|S1 = L+
1 |1 + L2 |2 ,
(C.18)
|S2 = L1 |1 |2 + L+
2 |3 ,
+
|S3 = L1 |2 L1 |3 |5 (2 )r[2L2

+
+ (2 )r 2L+
2 4l2 |4 ,
|S4 = L2 |1 + L+
2 |4 ,
|S5 = L2 |2 + |3 L+
1 |4 ,
(C.19)
4l2 ]|1
|S6 = L2 |3 + L1 |4 ,
|A1 = (L 0 2 r)|1 + L+ |5 ,
(C.20)
(C.21)
(C.22)
(C.23)

|A2 = (L 0 r)|2 L+
1 |5 + r (2 )L1 4l1 |1

+
r (1 )G0 2g0 |2 r (2 + )L+
2 4l2 |3 ,

|A3 = (L 0 + r)|3 L1 |5 + r (2 + )L2 4l2 |2

+
+ r (1 )G0 2g0 |3 + r (2 )L+
1 4l1 |4 ,
|A4 = (L 0 + 2 r)|4 L2 |5 ,
(C.24)
(C.25)
(C.26)
(C.27)
in its turn the gauge parameters are not uniquely defined but are invariant under transformations
|1 = L+
2 |,
|4 = L2 |,
|2 = L+
1 |,
|5 = L 0 |.
|3 = L1 |,
(C.28)
(C.29)
One can see again that the gauge transformations (C.18)(C.27) are simplified at = = 0.
C.1. The gauge fixing
First we discard the gauge parameter |5 and consider |1 , |2 , |3 , |4 to be independent.
Next we remove dependence of |S1 on the operator b1+ using the parameter |1 and using
the gauge parameter |2 we can remove the field |S3 using some components of |2
2

M
b1 + |2 + .
|S3 = L1 |2 + =
(C.30)
m1
After this we have used parameter |1 completely and we have the restricted parameter |2 :
b1 |2 = 0, i.e. |2 is independent of b1+ . Using this restricted gauge parameter |2 we can
remove dependence of |S1 on b2+ getting |S1 = |. After this the parameter |2 is exhausted.
Next we remove dependence of the vectors |S2 and |S4 on b2+ with the help of the parameters
|3 and |4 respectively. Now we have used all the gauge parameters. Let us show that the rest
auxiliary fields are zero as a consequence of the equations of motion in this gauge.
372
C.2. Removing of the auxiliary fields with the equations of motion

After the gauge fixing we have the gauge conditions
|S1 = |,
b2 |S2 = 0,
|S3 = 0,
b2 |S4 = 0,
(C.31)
where | is the basic vector (6). Let us look at the equation of motion (C.15)
+
l2 | + |S2 L+
2 |S5 L1 |S4 = 0.
(C.32)
Acting on this equation by the operator b2 and using (C.31) we get

b 2 L+
2 |S5 = 0
|S5 = 0.
(C.33)
Acting by b2 on (C.14) we get

b2 |A1 = 2r( 2)|S4
b2 b2 |A1 = 0.
(C.34)
Thus we can decompose |A1

|A1 = |A1 + b2+ |A1 ,
(C.35)
and (C.34) takes the form

|A1 = 2r( 2)|S4 .
(C.36)
Acting twice by b2 on (C.8) we get

b 2 b 2 L+
2 |A2 = 0
b2 |A2 = 0,
(C.37)
thus |A2 does not depend on b2+ .

Let us expand the fields |S2 , |S4 , |A1 , |A1 , |A2 , in powers of b1+
s2

+ k
b1 |S2k ,
|S2 =
|A1 =
k=0
s1

+ k
b1 |A1k ,
k=0
s3

+ k
b1 |S4k ,
|S4 =
(C.38)
k=0
s3

+ k
b1 |A1k ,
|A1 =
k=0
s2

+ k
b1 |A2k .
|A2 =
(C.39)
k=0
Then decomposing the equations of motion in powers of b1+ and b2+ one obtains (here we assume (92)).
C.2.1. Equations of motion (C.8)
+ s
m1 |A1,s1 = 0,
b1
+ s1
m1 |A1,s2 = l1+ |A1,s1 ,
b1
+ k
m1 |A1,k1 = l1+ |A1k + l2+ |A2k + r(2 + )l2+ |S2k ,
b1

+ 0
l0 + 4(2 )l2+ l2 | = l1+ |A10 + l2+ |A20 + r(2 + )l2+ |S20 ,
b1
(C.40a)
(C.40b)
(C.40c)
(C.40d)

+ s1
b1
m1 |S2,s2 = |A1,s1 ,
+ s2
b1
m1 |S2,s3 = |A1,s2 + l1+ |S2,s2 ,
+ k
m1 |S2,k1 = |A1k + l1+ |S2k + 2r( 2)l2+ |S4k ,
b1
+ 0
b1
l1 | = l1+ |S20 + |A10 + 2r( 2)l2+ |S40 ,
+ s2
b1
m1 |S4,s3 = |S2,s2 ,
+ s3
b1
m1 |S4,s4 = |S2,s3 l1+ |S4,s3 ,
+ k
b1
m1 |S4,k1 = |S2k l1+ |S4k ,
+ 0
b1
l2 | = l1+ |S40 |S20 .
k
b2+ b1+
|A1k = 2r( 2)|S4k .
s2
b2+ b1+
|A2,s2 = r(2 )|S2,s2 m1 |A1,s3 ,

s3
b2+ b1+
|A2,s3 = r(2 )|S2,s3 m1 |A1,s4 l1+ |A1,s3 ,
k
|A2k = r(2 )|S2k m1 |A1,k1 l1+ |A1k ,
b2+ b1+
0
b2+ b1+
|A20 = r(2 )|S20 l1+ |A10 .
373
(C.41a)
(C.41b)
(C.41c)
(C.41d)
(C.42a)
(C.42b)
(C.42c)
(C.42d)
(C.43)
(C.44a)
(C.44b)
(C.44c)
(C.44d)
One can show that the solution of equations of motion (C.40)(C.44) is

|S2 = |S4 = |A1 = |A2 = 0.
(C.45)
To see this one should start from the first two equations of (C.40). They give |A1,s1 =
|A1,s2 = 0. Then we go down to next set of Eqs. (C.41). The first two of them give us
|S2,s2 = |S2,s3 = 0. Going down to the subsequent sets of equations (C.42)(C.44) we obtain
one after another that |S4,s3 = |S4,s4 = 0, |A1,s3 = |A1,s4 = 0, |A2,s2 = |A2,s3 = 0.
After this we return to the first set of Eqs. (C.40) and repeat the procedure until we obtain (C.45).
After this we get from (C.40d), (C.41d), (C.42d) that the equations on the basic fields are (23).
Thus now we have
|S1 = |,
|S2 = |S3 = |S4 = |S5 = |A1 = |A2 = 0.
(C.46)
Substituting these solutions into (C.16) we get

L+
2 |S6 = 0
|S6 = 0,
(C.47)
then substituting into (C.17) we obtain

|A4 = 0,
(C.48)
374
and finally substituting into (C.12) one has

|A3 = 0.
(C.49)
Thus we remove all the auxiliary fields and the equations of motion for the basic fields |
are (23).
References
[1] M. Vasiliev, Higher spin gauge theories in various dimensions, Fortschr. Phys. 52 (2004) 702717, hep-th/0401177;
D. Sorokin, Introduction to the classical theory of higher spins, hep-th/0405069;
N. Bouatta, G. Compre, A. Sagnotti, An introduction to free higher-spin fields, hep-th/0409068;
A. Sagnotti, E. Sezgin, P. Sundell, On higher spin with a strong Sp(2) conditions, hep-th/0501156;
X. Bekaert, S. Cnockaert, C. Iazeolla, M.A. Vasiliev, Nonlinear higher spin theories in various dimensions, hepth/0503128.
[2] I.L. Buchbinder, V.A. Krykhtin, V.D. Pershin, On consistent equations for massive spin-2 field coupled to gravity
in string theory, Phys. Lett. B 466 (1999) 216226, hep-th/9908028;
I.L. Buchbinder, D.M. Gitman, V.A. Krykhtin, V.D. Pershin, Equations of motion for massive spin 2 field coupled
to gravity, Nucl. Phys. B 584 (2000) 615640, hep-th/9910188;
I.L. Buchbinder, D.M. Gitman, V.D. Pershin, Causality of massive spin 2 field in external gravity, Phys. Lett. B 492
(2000) 161170, hep-th/0006144.
[3] S. Deser, A. Waldron, Gauge invariances and phases of massive higher spins in (A)dS, Phys. Rev. Lett. 87 (2001)
031601, hep-th/0102166;
S. Deser, A. Waldron, Partial masslessness of higher spins in (A)dS, Nucl. Phys. B 607 (2001) 577604, hepth/0103198;
S. Deser, A. Waldron, Null propagation of partially massless higher spins in (A)dS and cosmological constant
speculations, Phys. Lett. B 513 (2001) 137141, hep-th/0105181;
K. Hallowell, A. Waldron, Constant curvature algebras and higher spin action generating functions, Nucl. Phys.
B 724 (2005) 453486, hep-th/0505255.
[4] R.R. Metsaev, Massive totally symmetric fields in AdS(d), Phys. Lett. B 590 (2004) 95104, hep-th/0312297.
[5] R.R. Metsaev, Mixed-symmetry massive fields in AdS(5), Class. Quantum Grav. 22 (2005) 27772796, hepth/0412311;
R.R. Metsaev, Cubic interaction vertices of massive and massless higher spin fields, hep-th/0512342.
[6] S.M. Klishevich, Massive fields with arbitrary integer spin in symmetrical Einstein space, hep-th/9812005;
S.M. Klishevich, On electromagnetic interaction of massive spin-2 particle, hep-th/9708150;
S.M. Klishevich, Massive fields with arbitrary half-integer spin in constant electromagnetic field, hep-th/9811030;
S.M. Klishevich, Massive fields with arbitrary integer spin in homogeneous electromagnetic field, hep-th/9910228;
S.M. Klishevich, Interaction of massive integer-spin fields, hep-th/0002024.
[7] N. Beisert, M. Bianchi, J.F. Morales, H. Samtleben, Higher spin symmetries and N = 4 SYM, JHEP 0407 (2004)
058, hep-th/0405057;
A.C. Petkou, Holography, duality and higher spin fields, hep-th/0410116;
M. Bianchi, P.J. Heslop, F. Riccioni, More on La Grande Bouffe: towards higher spin symmetry breaking in AdS,
JHEP 0508 (2005) 088, hep-th/0504156;
P.J. Heslop, F. Riccioni, On the fermionic Grande Bouffe: more on higher spin symmetry breaking in AdS/CFT,
JHEP 0510 (2005) 060, hep-th/0508086;
M. Bianchi, V. Didenko, Massive higher spin multiplets and holography, hep-th/0502220.
[8] I.L. Buchbinder, S.J. Gates, W.D. Linch, J. Phillips, New 4d, N = 1 superfiled theory: Model of free massive
superspin-3/2 multiplet, Phys. Lett. B 535 (2002) 280288, hep-th/0108200;
I.L. Buchbinder, S.J. Gates, W.D. Linch, J. Phillips, Dynamical superfiled theory of free massive superspin-1 multiplet, Phys. Lett. B 549 (2002) 229236, hep-th/0207243;
I.L. Buchbinder, S.J. Gates, S.M. Kuzenko, J. Phillips, Massive 4D, N = 1 superspin 1 and 3/2 multiplets and their
dualities, JHEP 0502 (2005) 056, hep-th/0501199;
S. Fedoruk, J. Lukiersky, Massive relativistic models with bosonic counterpart of supersymmetry, Phys. Lett. B 632
(2006) 371, hep-th/0506086.
[9] L. Brink, R.R. Metsaev, M.A. Vasiliev, How massless are massless fields in AdSd , Nucl. Phys. B 586 (2000) 183
205, hep-th/0005136;
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
375
K.B. Alkalaev, M.A. Vasiliev, N = 1 supersymmetric theory of higher spin gauge fields in AdS(5) at the cubic
level, Nucl. Phys. B 655 (2003) 5792, hep-th/0206068;
K.B. Alkalaev, O.V. Shaynkman, M.A. Vasiliev, On the frame-like formulation of mixed-symmetry massless fields
in (A)dS(d), Nucl. Phys. B 692 (2004) 363393, hep-th/0311164;
K.B. Alkalaev, Two-column higher spin massless fields in AdS(d), hep-th/0311212;
O.V. Shaynkman, I.Yu. Tipunin, M.A. Vasiliev, Unfolded form of conformal equations in M dimensions and o(M +
2)-modules, hep-th/0401086;
E.D. Skvortsov, M.A. Vasiliev, Geometric formulation for partially massless fields, hep-th/0601095;
K.B. Alkalaev, O.V. Shaynkman, M.A. Vasiliev, Frame-like formulation for free mixed-symmetry bosonic massless
higher-spin fields in AdS(d), hep-th/0601225.
R.R. Metsaev, Free totally (anti)symmetric massless fermionic fields in d-dimensional anti-de Sitter space, Class.
Quantum Grav. 14 (1997) L115L121, hep-th/9707066;
R.R. Metsaev, Fermionic fields in the d-dimensional anti-de Sitter spacetime, Phys. Lett. B 419 (1998) 4956,
hep-th/9802097;
R.R. Metsaev, Arbitrary spin massless bosonic fields in d-dimensional anti-de Sitter space, hep-th/9810231;
R.R. Metsaev, Light-cone form of field dynamics in anti-de Sitter spacetime and AdS/CFT correspondence, Nucl.
Phys. B 563 (1999) 295348, hep-th/9906217;
R.R. Metsaev, Massless arbitrary spin fields in AdS(5), Phys. Lett. B 531 (2002) 152160, hep-th/0201226.
E. Sezgin, P. Sundell, Analysis of higher spin field equations in four-dimensions, JHEP 0207 (2002) 055, hepth/0205132;
E. Sezgin, P. Sundell, Holography in 4D (super)higher spin theories and a test via cubic scalar couplings, JHEP 0507
(2005) 044, hep-th/0305040;
E. Sezgin, P. Sundell, An exact solution of 4D higher-spin gauge theory, hep-th/0508158.
D. Francia, A. Sagnotti, Free geometric equations for higher spins, Phys. Lett. B 543 (2002) 303310, hepth/0207002;
D. Francia, A. Sagnotti, On the geometry of higher-spin gauge fields, Class. Quantum Grav. 20 (2003) S473S486,
hep-th/0212185;
D. Francia, A. Sagnotti, Minimal local Lagrangians for higher-spin geometry, Phys. Lett. B 624 (2005) 93104,
hep-th/0507144;
A. Sagnotti, M. Tsulaia, On higher spins and the tensionless limit of string theory, Nucl. Phys. B 682 (2004) 83116,
hep-th/0311257;
A. Fotopoulos, K.L. Panigrahi, M. Tsulaia, On Lagrangian formulation of higher spin theories on AdS, hepth/0607248.
F. Kristiansson, P. Rajan, Scalar field corrections to AdS4 gravity from higher spin gauge theory, JHEP 0304 (2003)
009, hep-th/0303202.
X. Bekaert, N. Boulanger, S. Cnockaert, Spin three gauge theory revisited, JHEP 0601 (2006) 052, hep-th/0508048;
N. Boulanger, S. Leclercq, S. Cnockaert, Parity violating vertices for spin-3 gauge fields, Phys. Rev. D 73 (2006)
065019, hep-th/0509118.
G. Bonelli, On the tensionless limit of bosonic strings, infinite symmetries and higher spins, Nucl. Phys. B 669
(2003) 159172, hep-th/0305155;
G. Barnich, G. Bonelli, M. Grigoriev, From BRST to light-cone description of higher spin gauge fields, hepth/0502232.
M. Plyushchay, D. Sorokin, M. Tsulaia, Higher spins from tensorial charges and OSp(N |2n) symmetry, JHEP 0304
(2003) 013, hep-th/0301067;
M. Plyushchay, D. Sorokin, M. Tsulaia, GL Flatness of OSp(1|2n) and higher spin field theory from dynamics in
tensorial space, hep-th/0310297;
I. Bandos, X. Bekaert, J.A. Azcarraga, D. Sorokin, M. Tsulaia, Dynamics of higher spin fields and tensorial space,
JHEP 0505 (2005) 031, hep-th/0501113;
E. Ivanov, J. Lukiersky, Higher spins from nonlinear realizations of OSp(1|8), Phys. Lett. B 624 (2005) 304, hepth/0505216;
S. Fedoruk, E. Ivanov, Master higher spin particle, hep-th/0604111;
S. Fedoruk, E. Ivanov, J. Lukiersky, Massless higher spin D = 4 superparticle with both N = 1 supersymmetry and
its bosonic counterpart, hep-th/0606053.
G. Barnich, M. Grigoriev, A. Semikhatov, I. Tipunin, Parent field theory and unfolding in BRST first-quantized
terms, Commun. Math. Phys. 260 (2005) 147181, hep-th/0406192;
G. Barnich, M. Grigoriev, Parent form for higher spin fields on anti-de Sitter space, hep-th/0602166.
376
[18] Yu.M. Zinoviev, On massive high spin particles in (A)dS, hep-th/0108192;

Yu.M. Zinoviev, On massive mixed symmetry tensor fields in Minkowski space and (A)dS, hep-th/0211233.
[19] E.S. Fradkin, G.A. Vilkovisky, Quantization of relativistic system with constraints, Phys. Lett. B 55 (1975) 224;
I.A. Batalin, G.A. Vilkovisky, Relativistic S-matrix of dynamical systems with bosonic and fermionic constraints,
Phys. Lett. B 69 (1977) 309;
I.A. Batalin, E.S. Fradkin, Operator quantization of relativistic dynamical systems subject to first class constraints,
Phys. Lett. B 128 (1983) 303.
[20] I.A. Batalin, E.S. Fradkin, Operator quantization method and abelization of dynamical systems subject to first class
constraints, Riv. Nuovo Cimento 9 (10) (1986) 1;
I.A. Batalin, E.S. Fradkin, Operator quantization of dynamical systems subject to constraints. A further study of the
construction, Ann. Inst. H. Poincar A 49 (1988) 145;
M. Henneaux, C. Teitelboim, Quantization of Gauge Systems, Princeton Univ. Press, Princeton, 1992.
[21] M. Fierz, W. Pauli, On relativistic wave equations for particles of arbitrary spin in an electromagnetic field, Proc. R.
Soc. London, Ser. A 173 (1939) 211232.
[22] C. Fronsdal, Singletons and massless, integral-spin fields on de Sitter space, Phys. Rev. D 20 (1979) 848856.
[23] I.L. Buchbinder, A. Pashnev, M. Tsulaia, Lagrangian formulation of the massless higher integer spin fields in the
AdS background, Phys. Lett. B 523 (2001) 338346, hep-th/0109067;
I.L. Buchbinder, A. Pashnev, M. Tsulaia, Massless higher spin fields in the AdS background and BRST constructions for nonlinear algebras, hep-th/0206026;
X. Bekaert, I.L. Buchbinder, A. Pashnev, M. Tsulaia, On higher spin theory: Strings, BRST, dimensional reductions,
Class. Quantum Grav. 21 (2004) S1457S1464, hep-th/0312252.
[24] C. Burdik, O. Navratil, A. Pashnev, On the Fock space realizations of nonlinear algebras describing the high spin
fields in AdS spaces, hep-th/0206027.
[25] C. Burdik, A. Pashnev, M. Tsulaia, On the mixed symmetry irreducible representations of the Poincare group in the
BRST approach, Mod. Phys. Lett. A 16 (2001) 731746, hep-th/0101201.
[26] K. Schoutens, A. Sevrin, P. van Nieuwenhuizen, Quantum BRST charge for quadratically nonlinear lie algebras,
Commun. Math. Phys. 124 (1989) 87103.
[27] E. Witten, Noncommutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253.
[28] C.B. Torn, String field theory, Phys. Rep. 175 (1989) 1101;
W. Taylor, B. Zwiebach, D-branes, tachyons and string field theory, hep-th/0311017.
[29] S. Ouvry, J. Stern, Gauge fields of any spin and symmetry, Phys. Lett. B 177 (1986) 335340;
A.K.H. Bengtsson, A unified action for higher spin gauge bosons from covariant string theory, Phys. Lett. B 182
(1986) 321325;
W. Siegel, B. Zwiebach, Gauge string fields from light cone, Nucl. Phys. B 282 (1987) 125;
W. Siegel, Gauging Ramond string fields via OSp(1, 1|2), Nucl. Phys. B 284 (1987) 632.
[30] I.L. Buchbinder, V.A. Krykhtin, A. Pashnev, BRST approach to Lagrangian construction for fermionic massless
higher spin fields, Nucl. Phys. B 711 (2005) 367391, hep-th/0410215.
[31] I.L. Buchbinder, V.A. Krykhtin, Gauge invariant Lagrangian construction for massive bosonic higher spin fields in
D dimensions, Nucl. Phys. B 727 (2005) 536563, hep-th/0505092.
[32] I.L. Buchbinder, V.A. Krykhtin, BRST approach to higher spin field theories, hep-th/0511276.
[33] I.L. Buchbinder, V.A. Krykhtin, L.L. Ryskina, H. Takata, Gauge invariant Lagrangian construction for massive
higher spin fermionic fields, Phys. Lett. B 641 (2006) 386392, hep-th/0603212.
[34] I.G. Koh, S. Ouvry, Interacting gauge fields of any spin and symmetry, Phys. Lett. B 179 (1986) 115;
A.K.H. Bengtsson, BRST approach to interacting higher spin fields, Class. Quantum Grav. 5 (1988) 437;
L. Cappiello, M. Knecht, S. Ouvry, J. Stern, BRST construction of interacting gauge theories of higher spin fields,
Ann. Phys. 193 (1989) 1039;
F. Fougere, M. Knecht, J. Stern, Algebraic construction of higher spin interaction vertices, preprint LAPP-TH338/91.
[35] L.D. Faddeev, S.L. Shatoshvili, Realization of the Schwinger term in the Gauss low and the possibility of correct
quantization of a theory with anomalies, Phys. Lett. B 167 (1986) 225338;
I.A. Batalin, I.V. Tyutin, Existence theorem for the effective gauge algebra in the generalized canonical formalism
and Abelian conversion of second class constraints, Int. J. Mod. Phys. A 6 (1991) 32553282;
E. Egorian, R. Manvelyan, Quantization of dynamical systems with first and second class constraints, Theor. Math.
Phys. 94 (1993) 241252.
New example of infinite family of quiver gauge theories

Takeshi Oota a, , Yukinori Yasui b
a Osaka City University Advanced Mathematical Institute (OCAMI),

b Department of Mathematics and Physics, Graduate School of Science, Osaka City University,

Available online 4 December 2006
Abstract
We construct a new infinite family of quiver gauge theories which blow down to the Xp,q quiver gauge
theories found by Hanany, Kazakopoulos and Wecht. This family includes a quiver gauge theory for the
third del Pezzo surface. We show, using Z-minimization, that these theories generically have irrational Rcharges. The AdS/CFT correspondence implies that the dual geometries are irregular toric SasakiEinstein
manifolds, although we do not know the explicit metrics.
1. Introduction
D3-branes at the tip of CalabiYau cones have been extensively studied. The corresponding
IIB supergravity solution in the near horizon limit takes the form AdS5 X5 , where X5 is a
five-dimensional SasakiEinstein manifold. In dimension five, simply-connected regular Sasaki
Einstein manifolds are classified [1]. Indeed we have S 5 , T 1,1 and the total space Sk of the circle
bundles Sk dPk (3 k 8) where dPk is a del Pezzo surface with a Khler Einstein metric. It
is known that Sk is diffeomorphic to the k-fold connected sum k(S 2 S 3 ). Recently, an infinite
family of irregular toric SasakiEinstein manifolds Y p,q with topology S 2 S 3 was constructed
[2,3]. Especially, Y 2,1 is the horizon of the complex cone over the first del Pezzo surface dP1 .
Also, the existence of irregular toric SasakiEinstein manifolds X p,q (S 2 S 3 ) (S 2 S 3 )
which blow down to Y p,q and Y p,q1 by higgsing was conjectured in [4]. These are considered
E-mail addresses: toota@sci.osaka-cu.ac.jp (T. Oota), yasui@sci.osaka-cu.ac.jp (Y. Yasui).

doi:10.1016/j.nuclphysb.2006.11.024
378
T. Oota, Y. Yasui / Nuclear Physics B 762 [PM] (2007) 377391
Fig. 1. Toric diagram, quiver diagram and brane tiling for dP1 .
Fig. 2. Toric diagram, quiver diagram (model II) and brane tiling (model II) for dP2 .
to be extension of the SasakiEinstein manifold X 2,1 over the second del Pezzo surface dP2 [5].
For other new SasakiEinstein manifolds see [68] and also [9].
The AdS/CFT correspondence states that IIB string theory on AdS5 X5 is dual to a fourdimensional N = 1 quiver gauge theory. Given a toric SasakiEinstein manifold, one can
determine the corresponding quiver gauge theory by using the brane tiling (dimer) construction[1018]. In Figs. 1, 2 and 3, we present some data of quiver gauge theories for the del Pezzo
379
Fig. 3. Toric diagram, quiver diagram (model I) and brane tiling (model I) for dP3 .
surfaces dPk (k = 1, 2, 3), corresponding to Y 2,1 , X 2,1 and S1 [1930]. These quiver theories
have been extended to the general Y p,q [31,32] and X p,q theories [4,33,34].
In this paper we construct an infinite family of N = 1 quiver gauge theories which blow down
to X p,q . This construction generalizes the dP3 quiver gauge theory, corresponding to S1 Z 2,1 .
The dual geometries are irregular toric SasakiEinstein manifolds which are diffeomorphic to
3(S 2 S 3 ). We denote as Z p,q assuming they exist. It should be mentioned that the existence
of Z p,q was suggested in a recent paper [35].
In Section 2 we describe the brane tiling construction of the Z p,q theories. The explicit examples are given in Figs. 514. In Section 3 we determine the R-charges for these theories by using
Z-minimization.
2. Un-higgsing Xp,q
The brane tiling for X p,q was given in [33]. We use the convention of [33] for the brane tiling.
Some data for the X p,q quiver gauge theory are summarized in Fig. 4. We take n = 2q 1,
m = p q, j = 1, 2, . . . , q 1 and k = n + 1, n + 2, . . . , n + m 1. Then, the brane tiling
contains 2q 1 hexagons and p q + 1 cut hexagons. The Ri + Ri+1 + on edges represent
R-charges: at every vertex the sum of R-charges is 2 and for every face the sum of R-charges is
equal to the number of edges minus 2.
We can un-Higgs X p,q to Z p,q by cutting the ith hexagon horizontally (1 i n). This
procedure introduces one new edge with weight w 1 which connects the (i + 2)th white node
and the ith black node. If we replace 0 at (i + 2, i) element of the Kasteleyn matrix for X p,q
with w 1 , we obtain the Kasteleyn matrix for Z p,q . The determinant is modified to contain the
term w 1 z, which corresponds to an additional node (see next section for details):
V6 = (1, w6 ),
w6 = (2, 1).
(2.1)
380
Fig. 4. Brane tiling for X p,q .
The perfect matchings which correspond to the node (2, 1) are given by
(i)
P6 =
i
i +1
i +2
i +3
..
.
..
.
n+m+1
1 ... ...
0
1 0
.. ..
. .
1
0
w 1
i + 1 i + 2 i + 3 ... ... n + m + 1
1
0
.
0
0
1
0
1 0
.. ..
. .
1
(2.2)
381
Fig. 5. Toric diagram, quiver diagram and brane tiling for X 3,1 .
(i)
The matrices P6 (i = 1, 2, . . . , n) have non-zero elements Kab at the edges

connected by
bold lines (see Figs. 7, 10, 12 and 14), and the values Kab are decided according to a rule of the
Kasteleyn matrix. The charge R6 is added to the corresponding edges.
The contents of the minimal toric phase are summarized as follows:
Y p,q
X p,q
Z p,q
Ng
Nf
NW
2p
2p + 1
2p + 2
4p + 2q
4p + 2q + 1
4p + 2q + 2
2p + 2q
2p + 2q
2p + 2q
Here Ng represents the number of gauge groups, Nf the number of bifundamental fields, NW
the number of the interaction terms in the superpotential. These correspond to the numbers of
faces, edges and nodes on the brane tiling, respectively.
In the following we describe the brane tiling construction of the Z 3,1 and Z 3,2 theories explicitly.
2.1. Z 3,1
The brane tiling for X 3,1 is given by Fig. 5. In this case, there is only one hexagon that we
can put a cut in. We have drawn the resulting brane tiling for Z 3,1 in Fig. 6 together with the
corresponding quiver diagram and toric diagram. The perfect matching in Fig. 7 is given by the
382
Fig. 6. Toric diagram, quiver diagram and brane tiling for Z 3,1 .
matrix

(1)
P6 Z 3,1 =
2
3
4
1
0
0
w 1
0
2
0
1
0
0
3
0
0
0
1
4
z
0
.
0
0
(2.3)
2.2. Z 3,2
The brane tiling for X 3,2 is given by Fig. 8. In this case we have three types of cuttings corresponding to three hexagons. These lead to different brane tilings and quiver diagrams for Z 3,2 ,
although their toric diagrams are equivalent, as shown in Figs. 814.
The superpotentials contain the terms with the following degree:
Case 1
Case 2
Case 3
Cubic
Quartic
Quintic
Sextic
5
6
6
4
2
3
1
2
0
0
0
1
These provide an example of toric duality and are connected by Seiberg duality [19,24,25,27].
383
Fig. 7. The perfect matching which corresponds to the node V6 and the R-charge assignment for Z 3,1 .
3. Z-minimization
The toric diagram for Z p,q can be obtained by adding one vertex to the toric diagram for
as shown in Figs. 15 and 16. The six vectors Vi = (1, wi ) for Z p,q are written as
X p,q ,
w1 = (1, p),
w2 = (0, p q + 1),
w4 = (1, 0),
w5 = (2, 0),
w3 = (0, p q),
w6 = (2, 1),
(3.1)
where p and q are integers with 0 < q < p. Let us determine the Reeb vector b = (3, x, y)
for Z p,q using the method of [36], Z-minimization. Then, we obtain x = 3 and the remaining
component y is given by a root of the polynomial
P (y) = 2y 3 9pqy 2 9p 2 (2p 3q)y + 27p 3 (p q).
(3.2)
This polynomial has three real roots y = yi (i = 1, 2, 3). By P (0) = 27p 3 (p q) > 0 and
P (3p) = 27p 3 (p + q 2) < 0 they satisfy y1 < 0 < y2 < 3p < y3 . We find that the correct
384
Fig. 8. Toric diagram, quiver diagram and brane tiling for X 3,2 .
Fig. 9. Toric diagram, quiver diagram and brane tiling for Z 3,2 (case 1).
385
Fig. 10. The perfect matching and the R-charge assignment for Z 3,2 (case 1).
root is a middle one y2 , which is explicitly given by

3pq
+
3p q 2 2q + (4/3)p cos
,
y2 =
(3.3)
2
3
where the angle (0 /2) is calculated as
D
tan =
(3.4)
(q 1)(q 2 2q + 2p)
with
64
4
D = p 3 + p 2 (q + 1)(q 3) 4pq(q 2) q 2 (q 2)2 .
(3.5)
27
3
The R-charges can be computed from volumes of certain calibrated submanifolds i in Z p,q :
Ri =
vol(i )
,
3 vol(Z p,q )
(3.6)
where
vol(i ) =
2(Vi1 , Vi , Vi+1 ) 2
,
(b, Vi1 , Vi )(b, Vi , Vi+1 )
6

vol Z p,q =
vol(i ).
6
i=1
(3.7)
386
Explicitly we have the following volumes

2(p + q 2) 2
,
(3p y2 )2
2 2
vol(3 ) = vol(5 ) =
,
3y2
vol(1 ) =
vol(2 ) = vol(6 ) =
vol(4 ) =
2(p q) 2
y22
2 2
,
3(3p y2 )
(3.8)
and

9p 3 9p 2 q + 6pqy2 2y22 3
vol Z p,q =
.
3y22 (3p y2 )2
(3.9)
It should be noticed that1

vol(Z p,1 ) vol(Z p,2 )
>
> .
3
3
This implies the inequality of the central charge
1>
a=
3N 2
N2
>
,
p,q
4 vol(Z )
4
(3.10)
(3.11)
1 We calculate as vol(Z p,1 )/ 3 = 8(2p 1)/(27p 2 ) < 1, and the successive inequalities can be evaluated by pertur-
bation.
387
where N is the number of D3-branes and N 2 /4 the central charge of N = 4 YangMills theory.
The root y2 (3.3) is generically an irrational number and hence Z p,q are irregular Sasaki
Einstein manifolds. As a special case we have a rational number y2 = 3p/2 for q = 1.
For the toric diagram with six external lines, the third homology is given by H3 (Z p,q ) =
Z3 [4]. It turns out that from the work of Smale [37] Z p,q must be diffeomorphic to
3(S 2 S 3 ).2
The cone of a SasakiEinstein manifold is Ricci-flat Khler, i.e., CalabiYau. As described
in [36], the cone metric in the symplectic coordinates is given by a symplectic potential. We
found that for Y p,q the symplectic potential takes very simple form [39]. At present the Sasaki
Einstein metrics on X p,q and Z p,q are not constructed or even proved to exist for generic
integers p and q. It is expected that the symplectic approach can be useful to study these metrics.
2 In [38] it has been shown that 3(S 2 S 3 ) admits an infinite family of non-regular SasakiEinstein structures. We
do not know whether Z p,q are equivalent to those.
388
Before concluding this paper, we comment on the generating function f for the single-trace
gauge invariant operators [40,41]. In the notation of [40], the toric diagram for Z p,q is obtained by adding the point (1, 1, 1) to the one for X p,q . One more triangle with vertices (1, 0, 1),
(1, 1, 1) and (0, p, 1) gives a new term to the generating function (see Eq. (3.22) and the next
equation of [40]),

f x, y, z; Z p,q = f x, y, z; X p,q +
1

x p y
zp
1 xz 1 x p1
zp
y
(3.12)
Taking the limit [41]

V (b) = lim t 3 f eb1 t , eb2 t , eb3 t ; Z p,q
t0
(3.13)
we have
V (b) =
(p 1)b1 + pb3
b1 b2 (b1 b3 )((p 1)b1 + b2 pb3 )
(p 1)b1 + pb3
,
b1 (b1 + b3 )((p q)b1 + b2 )((q 1)b1 b2 + pb3 )
which reproduces the volume vol(Z p,q )/ 3 for the Reeb vector (b1 , b2 , b3 ) = (0, y2 , 3).
(3.14)
389
Fig. 15. Toric diagram for X p,q .
Fig. 16. Toric diagram for Z p,q .
Acknowledgements
We would like to thank K. Maruyoshi and M. Yamazaki for useful discussions. This work is
supported by the 21 COE program Construction of wide-angle mathematical basis focused on
knots. The work of Y.Y. is supported by the Grant-in Aid for Scientific Research (Nos. 17540262
390
and 17540091) from Japan Ministry of Education. The work of T.O. is supported by the Grant-in
Aid for Scientific Research (No. 18540285) from Japan Ministry of Education.
References
[1] H. Baum, T. Friedrich, R. Grunewald, I. Kath, Twistors and Killing Spinors on Riemannian Manifolds, TeubnerTexte fr Mathematik, vol. 124, Teubner, Leipzig, 1991.
[2] J.P. Gauntlett, D. Martelli, J. Sparks, D. Waldram, SasakiEinstein metrics on S 2 S 3 , Adv. Theor. Math. Phys. 8
(2004) 711734, hep-th/0403002.
[3] D. Martelli, J. Sparks, Toric SasakiEinstein metrics on S 2 S 3 , Phys. Lett. B 621 (2005) 208212, hep-th/0505027.
[4] A. Hanany, P. Kazakopoulos, B. Wecht, A new infinite class of quiver gauge theories, JHEP 0508 (2005) 054,
hep-th/0503177.
[5] A. Futaki, H. Ono, G. Wang, Transverse Khler geometry of Sasaki manifolds and toric SasakiEinstein manifolds,
math.DG/0607586.
[6] M. Cvetic, H. L, D.N. Page, C.N. Pope, New EinsteinSasaki spaces in five and higher dimensions, Phys. Rev.
Lett. 95 (2005) 071101, hep-th/0504225.
[7] M. Cvetic, H. L, D.N. Page, C.N. Pope, New EinsteinSasaki and Einstein spaces from Kerrde Sitter, hepth/0505223.
[8] W. Chen, H. L, C.N. Pope, General KerrNUTAdS metrics in all dimensions, Class. Quantum Grav. 23 (2006)
53235340, hep-th/0604125.
[9] C.P. Boyer, K. Galicki, Sasakian geometry, hypersurface singularities and Einstein metrics, Rend. Circ. Mat.
Palermo (2) Suppl. 75 (2005) 5787, math.DG/0406627.
[10] A. Hanany, K.D. Kennaway, Dimer models and toric diagrams, hep-th/0503149.
[11] S. Franco, A. Hanany, K.D. Kennaway, D. Vegh, B. Wecht, Brane dimers and quiver gauge theories, JHEP 0601
(2006) 096, hep-th/0504110.
[12] S. Franco, A. Hanany, D. Martelli, J. Sparks, D. Vegh, B. Wecht, Gauge theories from toric geometry and brane
tilings, JHEP 0601 (2006) 128, hep-th/0505211.
[13] A. Hanany, D. Vegh, Quivers, tilings, branes and rhombi, hep-th/0511063.
[14] B. Feng, Y.-H. He, K.D. Kennaway, C. Vafa, Dimer models from mirror symmetry and quivering amoebae, hepth/0511287.
[15] A. Hanany, C.P. Herzog, D. Vegh, Brane tilings and exceptional collections, JHEP 0607 (2006) 001, hepth/0602041.
[16] S. Franco, D. Vegh, Moduli spaces of gauge theories from dimer models: Proof of the correspondence, hepth/0601063.
[17] Y. Imamura, Anomaly cancellations in brane tilings, hep-th/0605097.
[18] Y. Imamura, Global symmetries and t Hooft anomalies in brane tilings, hep-th/0609163.
[19] B. Feng, A. Hanany, Y.-H. He, D-brane gauge theories from toric singularities and toric duality, Nucl. Phys. B 595
(2001) 165200, hep-th/0003085.
[20] S. Franco, A. Hanany, F. Saad, A.M. Uranga, Fractional branes and dynamical supersymmetry breaking, JHEP 0601
(2006) 011, hep-th/0505040.
[21] S. Franco, A. Hanany, A.M. Uranga, Multi-flux warped throats and cascading gauge theories, JHEP 0509 (2005)
028, hep-th/0502113.
[22] C. Beasley, B.R. Green, C.I. Lazaroiu, M.R. Plesser, D3-branes on partial resolutions of Abelian quotient singularities of CalabiYau threefolds, Nucl. Phys. B 566 (2000) 599640, hep-th/9907186.
[23] M. Bertolini, F. Bigazzi, A.L. Cotrone, New checks and subtleties for AdS/CFT and a-maximization, JHEP 0412
(2004) 024, hep-th/0411249.
[24] B. Feng, A. Hanany, Y.-H. He, Phase structures of D-brane gauge theories and toric duality, JHEP 0108 (2001) 040,
hep-th/0104259.
[25] C.E. Beasley, M.R. Plesser, Toric duality is Seiberg duality, JHEP 0212 (2002) 076, hep-th/0109053.
[26] B. Feng, A. Hanany, Y.-H. He, A.M. Uranga, Toric duality as Seiberg duality and brane diamonds, JHEP 0112
(2001) 035, hep-th/0109063.
[27] B. Feng, S. Franco, A. Hanany, Y.-H. He, Symmetries of toric duality, JHEP 0212 (2002) 076, hep-th/0205144.
[28] B. Feng, S. Franco, A. Hanany, Y.-H. He, Unhiggsing the del Pezzo, JHEP 0308 (2003) 058, hep-th/0209228.
[29] S. Franco, A. Hanany, Geometric dualities in 4d field theories and their 5d interpretation, JHEP 0304 (2003) 043,
hep-th/0207006.
391
[30] K. Intriligator, B. Wecht, Baryon charges in 4d superconformal field theories and their AdS duals, Commun. Math.
Phys. 245 (2004) 407424, hep-th/0305046.
[31] S. Benvenuti, S. Franco, A. Hanany, D. Martelli, J. Sparks, An infinite family of superconformal quiver gauge
theories with SasakiEinstein duals, JHEP 0506 (2005) 064, hep-th/0411264.
[32] A. Butti, D. Forcella, A. Zaffaroni, The dual superconformal theory for Lp,q,r manifolds, JHEP 0509 (2005) 018,
hep-th/0505220.
[33] A. Butti, A. Zaffaroni, R-charges from toric diagrams and the equivalence of a-maximization and Z-minimization,
JHEP 0511 (2005) 019, hep-th/0506232.
[34] A. Butti, A. Zaffaroni, From toric geometry to quiver gauge theory: The equivalence of a-maximization and Zminimization, Fortschr. Phys. 54 (2006) 309316, hep-th/0512240.
[35] R. Argurio, M. Bertolini, C. Closset, S. Cremonesi, On stable non-supersymmetric vacua at the bottom of cascading
theories, hep-th/0606175.
[36] D. Martelli, J. Sparks, S.T. Yau, The geometric dual of a-maximisation for toric SasakiEinstein manifolds, hepth/0503183.
[37] S. Smale, On the structure of 5-manifolds, Ann. Math. 75 (1962) 3846.
[38] C.P. Boyer, K. Galicki, M. Nakamaye, On the geometry of SasakiEinstein 5-manifolds, Math. Ann. 325 (2003)
485524.
[39] T. Oota, Y. Yasui, Toric SasakiEinstein manifolds and Heun equations, Nucl. Phys. B 742 (2006) 275294, hepth/0512124.
[40] S. Benvenuti, B. Feng, A. Hanany, Y.-H. He, Counting BPS operators in gauge theoriesQuivers, syzygies and
plethystics, hep-th/0608050.
[41] D. Martelli, J. Sparks, S.T. Yau, SasakiEinstein manifolds and volume minimisation, hep-th/0603021.

Nucl - Phys.B v.762 PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Nucl - Phys.B v.762 PDF

Загружено:

Авторское право:

Доступные форматы

Nuclear Physics B 762 (2007) 137

An exact solution of 4D higher-spin gauge theory

College Station, TX 77843-4242, USA

1. Introduction and summary

E-mail address: per.sundell@teorfys.uu.se (P. Sundell).

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

of scalar-coupled AdS gravity (corresponding to an effective model given in [25]) as well as an

Equivalently, in terms of Weyl-ordered functions

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

and a zero-form = (x,

and the functions and defined by

have the salient properties

f(y, z)  = f(z, y),

where is the anti-automorphism

and is the involutive automorphism

The gauge transformations are given by

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

and a suitable gauge condition on the internal connection, such as [6]

The extra term proportional

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

One can show that [2]

where e is the vielbein2

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

= 2[ W]2 ...s1 2 ... s1 (s 2)( )(2 d W3 ...s1 ) 2 ... s1

as exexplicitly in terms of the physical fields using curvature-dressed covariant derivatives ,

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

(0) C + e a {Pa , C} = 0,

Setting s = 0 one obtains formally the correct scalar-field equation,

where = C|Y =0 in D = 4. Splitting x (x i , r), and using Poincar coordinates,

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

3.2. The Z-space approach

and expressed explicitly as

The remaining constraints in Z-space, viz.

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

must then be solved given an initial condition

whose compatibility with (2.20) requires

The gauge fields can then be obtained from (2.31), viz.

The resulting transformations of the physical fields, viz.

By construction, its MaurerCartan form

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

where the superscript indicate the order in C

3 Using the gauge function (4.1), the spin-s sector in C(x; y, y)

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

3.4. Zero-form curvature invariants

the full traces are defined by

and the normalizations N are given by

effect is that the invariants C2p

where the reduced traces Tr f (Y ) are defined by4

Moreover, from D = 0, which is equivalent to (  ) = [A ,  ], it follows that

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

where we have used (3.43), and

The form of the leading contribution C2p

5 To exhibit L SO(3, 2) one can write L(x; y, y)

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

corresponding to the vierbein and Lorentz connection

that in turn gives

and declaring the overlap with L(x; y, y)

E. Sezgin, P. Sundell / Nuclear Physics B 762 (2007) 137

We next turn to the exact solution of the Z-space equations (3.16)(3.18).

, given by (3.16), now takes the form

To solve this constraint, following [23], we use the integral representation

f(y, z) = f(z, y),

(0) C + e a {Pa , C} = 0,

Moreover, from D = 0, which is equivalent to ( ) = [A , ], it follows that

where the th level is given by

The spaces B(g 1 C

for v 2 = 0 and v 2 < 1,