Nucl - Phys.B v.671

Nuclear Physics B 671 (2003) 350
www.elsevier.com/locate/npe
Spinning strings in AdS5 S 5 and integrable systems

G. Arutyunov a,1 , S. Frolov b,1 , J. Russo c,d , A.A. Tseytlin b,c,2
a Max-Planck-Institut fr Gravitationsphysik, Albert-Einstein-Institut, Am Mhlenberg 1,
D-14476 Golm, Germany

b Department of Physics, The Ohio State University, Columbus, OH 43210-1106, USA
c Blackett Laboratory, Imperial College, London, SW7 2BZ, UK
d Departament ECM, Facultat de Fsica, Universitat de Barcelona,
Instituci Catalana de Recerca i Estudis Avanats (ICREA), Spain

Received 1 August 2003; accepted 27 August 2003
Abstract
We show that solitonic solutions of the classical string action on the AdS5 S 5 background that
carry charges (spins) of the Cartan subalgebra of the global symmetry group can be classified in terms
of periodic solutions of the Neumann integrable system. We derive equations which determine the
energy of these solitons as a function of spins. In the limit of large spins J , the first subleading
1/J coefficient in the expansion of the string energy is expected to be non-renormalized to all
orders in the inverse string tension expansion and thus can be directly compared to the 1-loop
anomalous dimensions of the corresponding composite operators in N = 4 super-YM theory. We
obtain a closed system of equations that determines this subleading coefficient and, therefore, the
1-loop anomalous dimensions of the dual SYM operators. We expect that an equivalent system of
equations should follow from the thermodynamic limit of the algebraic Bethe ansatz for the SO(6)
spin chain derived from SYM theory. We also identify a particular string solution whose classical
energy exactly reproduces the one-loop anomalous dimension of a certain set of SYM operators with
two independent R charges J1 , J2 .
2003 Elsevier B.V. All rights reserved.
PACS: 11.25.Tq; 11.27.+d; 02.30.Ik
E-mail addresses: agleb@aei-potsdam.mpg.de (G. Arutyunov), frolov@mps.ohio-state.edu (S. Frolov),

jrusso@ecm.ub.es (J. Russo), tseytlin@mps.ohio-state.edu (A.A. Tseytlin).
1 On leave of absence from Steklov Mathematical Institute, Gubkin str. 8, 117966 Moscow, Russia.
2 Also at Lebedev Institute, Moscow.
0550-3213/$ see front matter 2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2003.08.036
G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350
1. Introduction and summary

Recently, there was a remarkable progress towards understanding AdS/CFT duality in
non-supersymmetric sector of states of string theory on AdS5 S 5 [14], generalizing
earlier work of [57]. On the string side, one identifies semiclassical states described by
solitonic closed-string solutions on a 2-cylinder. They have finite energy and carry SO(6)
(and, in general, SO(2, 4)) angular momentum components Ji . In the limit of large angular
momenta the first subleading term in the expansion of the classical string energies happens
to be protected (i.e., is not renormalized by string corrections) and thus can be matched
onto the dimensions of the corresponding gauge-theory (N = 4 SYM) operators [3]. We
refer to [4] for a more detailed discussion.
In general, according to the AdS/CFT duality the AdS5 S 5 string sigma model
should be equivalent to the N = 4 supersymmetric SU(N ) YM theory with N .
The composite primary operators in this theory are classified in terms of UIRs of the
superconformal group PSU(2, 2 | 4), i.e., by the conformal dimension , two spins S1 , S2
and by the Young tableaux (or, equivalently, by the Dynkin labels) of the R-symmetry
group SU(4). At N = only single-trace operators matter. Thus one should expect that
the energy of the string solutions considered as the function of the angular momenta Ji
should match with the dimensions of the corresponding primary single-trace operators in
the SYM theory.
The bosonic part of the classical string action is a combination of SO(2, 4) and SO(6)
sigma models. The O(n) (or O(p, q)) sigma models are known to be classically integrable
[8], and the same should obviously be true also upon imposition of the conformal gauge
constraints, i.e., for the corresponding classical string theories (for some related work
see [911]). One expects, therefore, a close connection between special classes of string
solutions representing particular semiclassical string states and certain integrable models.
As was already observed earlier, the folded rotating string solutions with one [6,12] or two
[4,7] non-vanishing angular momenta are related to the 1d sine-Gordon model.
Here we will consider a generalization to the case when all the three Cartan
components of the SO(6) angular momentum are non-zero and will find that in this
case the SO(6) sigma model effectively reduces to a special integrable 1d modelthe
Neumann model [13]. The latter describes a three-dimensional harmonic oscillator with
three different frequencies constrained to move on a two-sphere (see, e.g., [1416]).
The class of S 5 rotating string solutions we will be discussing is parametrized by the
angular momenta Ji = (J1 , J2 , J3 ) with the energy being E = E(J1 , J2 , J3 ). To be able to
compare to perturbative conformal dimensions on the SYM side one needs to assume that
1,
Ji2
1
1,
Ji
(1.1)
where is the square of string tension (related as usual to t Hooft coupling). In this
case the AdS5 S 5 superstring corrections to the classical energy will be suppressed
by extra powers of 1/Ji . This happens [2] despite the non-BPS nature of the extended
rotating string states (the only BPS state is a point-like string having only one non-zero
component of Ji [5]) and is due to the fact that the underlying superstring theory has (i)
global supersymmetry and (ii) is effectively massive in this case (with 2d masses 1/Ji ).3
Assuming (1.1), the classical energy can be expanded as4
E = Jtot +

Ji
2
Ji
+ 3 f2
+ ,
f1
Jtot
Jtot
Jtot
Jtot
Jtot J1 + J2 + J3 .
(1.2)
(1.3)
One immediate aim is then to determine the coefficient function f1 (J2 /Jtot , J3 /Jtot ).
Given the analytic dependence of the subleading term in E on and its expected nonrenormalizability on the string side, one may be able, as explained in [2,4], to compare
it directly to the one-loop anomalous dimensions of gauge-theory operators of the type
tr[(1 + i2 )J1 (3 + i4 )J2 (5 + i6 )J3 ] + belonging to irreducible representation
of SU(4) with Dynkin labels [J2 J3 , J1 J2 , J2 + J3 ] (we assume for definiteness
that J3 J2 J1 ).5 Note also that the expected non-renormalization of the 1/Jtot term
to all orders in the inverse string tension predicts (through AdS/CFT duality) that the
corresponding term in N = 4 SYM should be one loop exact.
Finding the spectrum of one-loop anomalous dimensions of such scalar operators with
all three Ji being non-zero should be possible, as in the two-spin case (J3 = 0) in [3],
using the techniques (dilatation operator related to integrable spin chains and Bethe ansatz)
developed in a recent remarkable series of papers [3,1921]. Here we will determine
f1 in several special cases with Ji = 0, thus making string-theory predictions for the
corresponding eigenvalues of the anomalous dimension matrix.
One special three-spin solution was found already in [1,2]: this is a circular string with
J1 = J2 and arbitrary J3 (the stability condition implies J1 + J2 (4n 1)J3 /(2n 1)2
where n is the winding number; in what follows we set n = 1)
(J1 + J2 )
+
2(J1 + J2 + J3 )2

J3
= Jtot +
1
+ ,
2Jtot
Jtot
E = J1 + J2 + J3 +
(1.4)
3 A similar argument explains [7,17] why 2- and higher-loop corrections to the energy of string states in the
BMN [5] sector are suppressed by powers of 1/J .
4 In the cases when there is just one non-vanishing component of the spin as in [6] or the conditions (1.1) are
not satisfied for at least

term O( ).
one of the spins [7,18], the energy may contain also a constant subleading
Under the condition /J 1 such term (which will not be protected against string 1/ -corrections and
thus cannot be easily compared to SYM theory) is much larger that the subleading term in the equation below,
and thus the corresponding SYM operators should have larger anomalous dimensions.
5 The primary operators obtained after diagonalizing the dilatation matrix for the gauge-invariant operators
mentioned above are not superconformal primaries lying on the unitary bound of the continuous (unprotected)
series of UIRs of PSU(2, 2 | 4), as can be seen from the corresponding relation between the conformal dimension
and the Dynkin labels. Their parent superconformal primaries can be found by analyzing representations of
supersymmetry. However, this is immaterial since the anomalous dimensions of primaries and of their SUSY
descendants are equal.
where dots stand for 2 and other subleading terms. The special case of J3 = 0, i.e.,
1
+ , f1 =
(1.5)
J1 + J2
2
corresponds to the simplest two-spin circular string solution [1]. In spite of being unstable,
this solution has its counterpart on the SYM side [3]. Another string state with the same
quantum numbers J1 = J2 , J3 = 0 but lower energy is represented by the stable folded
string solution [4] (which generalizes the single-spin solution of [6] to the two-spin case).
In this case the energy is given by (1.5) with6
E = J1 + J2 + f1
f1 = 0.356 . . .,
and, remarkably, can be matched exactly with the corresponding lowest anomalous
dimension eigenvalue on the SYM side [3].
As in the two-spin case, in general, there will be several three-spin string solutions for
given values of J1 , J2 , J3 having different values of E. The first subleading term in E will
then be expected to correspond to the band of one-loop dimensions of the SYM eigenoperators in the [J2 J3 , J1 J2 , J2 + J3 ] irrep. In particular, there may be several string
solutions with J1 = J2 and small J3 generalizing the circular [1,2] and folded [4] two-spin
solutions, and different from the circular string solution of [2] with E given in (1.4).
We shall see that in spite of the formal integrability of the Neumann model, finding the
explicit form of the three-spin solutions and, in particular, their energies, turns out to be
complicated. Below we shall concentrate on several special cases. In particular, there are
two obvious cases generalizing the two-spin solutions mentioned above: (i) generalization
of the folded two-spin solution to the case of non-zero J3 < J1 = J2 ; (ii) generalization of
the circular two-spin solution to the case of non-zero J3 < J1 = J2 which has less energy
than the circular three-spin solution of [2] (the latter is unstable for small J3 ). In these
and similar cases with J1 , J2 J3 we find the following expression for the energy (to the
leading order in J3 /Jtot )

(0)
(1) J3
E = Jtot +
(1.6)
+ + .
f1 + f1
Jtot
Jtot
Note that the expression (1.4) for the circular three-spin solution of [2] is thus a special
(1)
case of (1.6).7 One of our aims will be to compute the value of the coefficient f1 for
various three-spin solutions. In particular, we will find in Section 3 that the folded string
(1)
solution that generalizes the J1 = J2 , J3 = 0 solution of [4] has f1 = 4.79 . . ..
One may try to find also folded string solutions with J1 = J2 = J3 , which should have
less energy than the circular solution of [2]. Though the latter is stable for J1 = J2 = J3 , it
is likely to be only a local minimum of the energy, i.e., there may be another J1 = J2 = J3
solution with less energy.8
6 This number has a simple origin in terms of values of elliptic functions as mentioned at the end of Section 3.3.
7 Let us mention that a reason for considering linear in J terms in E (i.e., leading deformations of the two3
spin expressions) is that the corresponding leading terms in SYM anomalous dimensions may be possible to
compute by using a perturbation theory near the Heisenberg model Hamiltonian corresponding to the two-spin
case, i.e., to the anomalous dimension matrix for the operators in the [J2 , J1 J2 , J2 ] irrep.
8 In general, there may be several local minima, i.e., stable solutions with the same quantum numbers.
The rest of the paper is organized as follows. In Section 2 we shall present the
ansatz for the general three-spin S 5 rotating string solution and explain its relation to the
Neumann integrable system. This will allow us to reduce the problem to a pair of firstorder differential equations for the two coordinates of S 2 related to 5th-order polynomial
defining a hyperelliptic curve of genus 2.
In Section 3.1 we shall argue that to obtain a non-trivial folded string solution with the
three non-zero spins the string must be bent (i.e., two coordinates of S 2 should have a
different number of folding points). In Section 3.2 we shall derive the general system of
equations that governs the form of the subleading (or one-loop) term f1 in the expression
(1.2) for the energy of the bent string. An equivalent system is expected to follow from the
thermodynamic limit of the algebraic Bethe ansatz for the SO(6) spin chain determining
the one-loop anomalous dimensions on the SYM side [19,21]. In Section 3.3 we shall
study this system in expansion in small J3 and determine the coefficient f1(1) in (1.6) in the
special case of perturbation near two-spin folded string solution of [4] with J1 = J2 .
Section 4 will be devoted to a different class of three-spin solutions which will have
higher energy than folded bent strings for the same values of Ji . Using a combination of
analytic and numerical methods we shall again determine the form of the leading correction
f1 in (1.2) in this case.
In Section 5 we shall consider a two-spin solution of a circular type that generalizes the
circular solution of [1] to the case of unequal S 5 spins (J1 , J2 ). We shall show that, just in
the case of the two-spin folded string in [4], the first subleading term in the corresponding
expression for the energy matches precisely the one-loop anomalous dimensions of a set of
SYM operators with SU(4) Dynkin labels [J2 , J1 J2 , J2 ] which correspond to solutions
of the Bethe ansatz equations in [3] with all Bethe roots lying on the imaginary axis.
This complements the results in [3,4], providing another remarkable test of the AdS/CFT
correspondence.
Similar solutions describing string spinning in AdS5 directions can be analyzed in much
the same way as described in Section 6. In fact, most of the SO(6) case equations have a
direct analog in the SO(2, 4) case or are related by an analytic continuation.
In Appendix A we shall explain how the general solution of the Neumann model can
be written in terms of -functions defined on the Jacobian of a hyperelliptic genus 2
Riemann surface [16]. Appendix B will contain a list of integrals used in Section 3. In
Appendix C we shall study the vanishing of other (non-Cartan) components of the SO(6)
angular momentum tensor for different three-spin solutions which is crucial [1] for their
consistent semiclassical quantum state interpretation (and thus a possibility to establish
a correspondence with particular SYM operators with the same quantum numbers). In
Appendix D we shall describe the two-spin solution corresponding to the straight folded
string without bend points. We will show that such string solution does not allow a
deformation towards a non-zero third spin component.
2. Reduction of O(6) sigma-model to the Neumann system

2.1. Rotating string ansatz and integrals of motion
Let us consider the bosonic part of the classical closed string propagating in the
AdS5 S 5 spacetime. The world-sheet action in the conformal gauge is

5
5 ) (x) x m a x n + G(S ) (y) y p a y q ,
I =
d d G(AdS
a
a
mn
pq
4
R2
.
(2.1)
The two metrics have the standard form in terms of the 5 + 5 angular coordinates:
2

ds AdS = cosh2 dt 2 + d 2 + sinh2 d 2 + sin2 d 2 + cos2 d 2 , (2.2)
5

2

ds S 5 = d 2 + cos2 d32 + sin2 d 2 + cos2 d12 + sin2 d22 .
(2.3)
It is convenient to represent (2.1) as an action for the O(6) SO(4, 2) sigma-model (we
follow the notation of [1])

I=
(2.4)
d d (LS + LAdS ),
2
where
1
1
LS = a XM a XM + (XM XM 1),
2
2
1
1
MN YM YN + 1).
LAdS = MN a YM a YN + (
2
2
(2.5)
(2.6)
Here XM , M = 1, . . . , 6 and YM , M = 0, . . . , 5 are the embedding coordinates of R 6

with the Euclidean metric in LS and with MN = (1, +1, +1, +1, +1, 1) in LAdS ,
respectively. and are the Lagrange multipliers. The action (2.4) is to be supplemented
with the usual conformal gauge constraints.
The embedding coordinates are related to the angular ones in (2.2), (2.3) as follows:
X1 + iX2 = sin cos ei1 ,
X3 + iX4 = sin sin ei2 ,
X5 + iX6 = cos ei3 ,
(2.7)
Y1 + iY2 = sinh sin ei ,

Y3 + iY4 = sinh cos ei ,
Y5 + iY0 = cosh eit .
(2.8)
In the next few sections we will be discussing the case when the string is located at the
center of AdS5 and rotating in S 5 , i.e., is trivially embedded in AdS5 as Y5 + iY0 = ei
with Y1 , . . . , Y4 = 0.
The S 5 metric has three commuting translational isometries in i which give rise to three
global commuting integrals of motion (spins) Ji . Since we are interested in the periodic
motion with three Ji non-zero it is natural to choose the following ansatz for XM :
X1 + iX2 = x1 ( )eiw1 ,
X3 + iX4 = x2 ( )eiw2 ,
X5 + iX6 = x3 ( )eiw3 ,
(2.9)
where the real radial functions xi are independent of time and should, as a consequence of
2 = 1, lie on a two-sphere S 2 :
XM
3

xi2 = 1,
i = 1, 2, 3.
i=1
Then the spins J1 = J12 , J2 = J34 , J3 = J56 forming a Cartan subalgebra of SO(6) are
d 2
xi ( ) Ji .
Ji = wi
2
2
(2.10)
As discussed in [1], to have a consistent semiclassical string state interpretation of these

configurations one should look for solutions for which all other components of the SO(6)
angular momentum tensor JMN vanish.
The spacetime energy E of the string (related to a generator of the compact SO(2)
subgroup of SO(4, 2)) is simply
E = E.
(2.11)
The only non-trivial Virasoro constraint is then (dot and prime are derivatives over and
)

XM
.
2 = X M X M + XM
(2.12)
As a consequence of this relation the energy becomes a function of the SO(6) spins:
E = E(J1 , J2 , J3 ).
(2.13)
On the string theory side of the AdS/CFT duality the problem is thus to classify the
solutions of the string sigma-model subject to the Virasoro constraints with further
determination of their spacetime energy as a function of the spins. Below we shall derive
a closed system of equations which, in principle, allows one to find the energy as a function
of the spins for generic three-spin solutions and therefore to determine the dimensions of
the corresponding gauge theory operators.
Substituting the ansatz (2.9) into the SO(6) Lagrangian (2.5) we get the following 1d
(mechanical) system
3
3

1
1 2
xi wi2 xi2 +
xi2 1 .
L=
(2.14)
2
2
i=1
i=1
10
It describes an (n = 3)-dimensional harmonic oscillator constrained to remain on a unit

n 1 = 2 sphere. This is the special case of the n-dimensional Neumann dynamical system
[13] which is known to be integrable [14].
The Virasoro constraint implies that the energy H of the Neumann system is given by
1
1 2
xi + wi2 xi2 = 2 .
2
2
3
H=
(2.15)
i=1
Solving the equation of motion for the Lagrange multiplier we obtain the following
non-linear equations for xi :
xi = wi2 xi xi
3

2

xj wj2 xj2 .
(2.16)
j =1
The canonical momenta conjugate to xi are

3

i = xi ,
i xi = 0.
i=1
One can think about this dynamical system as being originally defined on the cotangent
bundle T R 3 . Imposing the constraints reduces the phase space to T S 2 . The Dirac bracket
obtained from the canonical structure {i , xj } = ij is
{i , j }D = xi j xj i ,
{i , xj }D = ij xi xj ,
{xi , xj }D = 0.
(2.17)
One can check that (2.16) follows from the Poisson structure (2.17) and the Hamiltonian

1 2
H=
i + wi2 xi2
2
3
(2.18)
i=1

supplemented with the two constraints 3i=1 xi2 = 1, 3i=1 i xi = 0.
The crucial point allowing to solve this model is that the n-dimensional (n = 3 in the
present case) Neumann system has the following n integrals of motion [22]:
Fi = xi2 +
(xi j xj i )2
j =i
wi2 wj2
n

Fi = 1.
(2.19)
i=1
In the present case n = 3 and thus only two of the three integrals of motion are independent.
Moreover, these integrals are in involution with respect to the Poisson bracket (2.17) and
the Hamiltonian is
1 2
H=
wi Fi .
2
3
i=1
(2.20)
11
Thus, any two of these three integrals of motion are enough to integrate this dynamical
system since the motion occurs on a surface of constant integrals.
In order to find the relevant closed string solutions we need also to impose the
periodicity conditions on xi :
xi ( ) = xi ( + 2),
(2.21)
i.e., we are interested in periodic version of the Neumann model.

2.2. First-order system for the ellipsoidal coordinates
It is convenient to describe the phase space of this model in terms of independent
2 + 2 canonical variables rather than the 3 + 3 constrained variables xi , i . One natural
coordinate system on a two-sphere is the angular ( , ) one implied by (2.7) and (2.9):
x1 = sin cos ,
x2 = sin sin ,
x3 = cos .
(2.22)
However, if all the frequencies wi are different and so the Hamiltonian is not spherically
symmetric, it appears advantageous to use the so-called ellipsoidal coordinates [15]. The
ellipsoidal coordinates are introduced as the two real roots 1 and 2 of the following
quadratic equation
x12
w12
x22
w22
x32
w32
= 0.
(2.23)
Assuming w1 < w2 < w3 we can define the range of a (a = 1, 2) as

w12 1 w22 2 w32 .
(2.24)
With this range a cover (1/8)th of the two-sphere corresponding to xi 0. One can think
of the whole sphere as covering the domain (2.24) and branching along its boundary. For
xi 0 we have

(1 w12 )(2 w12 )
(w22 1 )(2 w22 )
,
x2 =
,
x1 =
(2.25)
2
2
2 w2
w21 w31
w21
32

(w32 1 )(w32 2 )
,
wij2 wi2 wj2 .
x3 =
(2.26)
2 w2
w31
32

One can check that 3i=1 xi2 = 1, i.e., we indeed get a parametrization of a two-sphere.
Substituting now this parametrization for xi into Eq. (2.14) we get the following sigmamodel Lagrangian
1
L = gab ( )a b U ( ),
2
where the non-zero components of the two-sphere metric are
2 1
,
4(1 w12 )(w22 1 )(w32 1 )
2 1
g22 =
,
2
4(2 w1 )(2 w22 )(w32 2 )
(2.27)
g11 =
(2.28)
12
and the potential U is very simple

1 2
w + w22 + w32 1 2 .
(2.29)
2 1
Note that in the domain (2.24) the metric gab is non-negative.
Expressing the integrals of motion (2.19) in terms of a one finds a system of two
1st-order equations which can also be obtained (2.27) by solving directly the associated
HamiltonianJacobi problem

P (1 )
d2 2
P (2 )
d1 2
= 4
,
= 4
.
(2.30)
d
(2 1 )2
d
(2 1 )2
U=
Here the function P ( ) is a 5th-order polynomial

P ( ) = w12 w22 w32 ( b1 )( b2 ).
(2.31)
The parameters b1 , b2 are the two constants of motion which can be expressed in terms of
integrals Fi in (2.19) by solving the system of equations

b1 + b2 = w22 + w32 F1 + w12 + w32 F2 + w12 + w22 F3 ,
b1 b2 = w22 w32 F1 + w12 w32 F2 + w12 w22 F3 .
(2.32)
In terms of variables bi the Hamiltonian (2.15) reads as

1
1 2
1
w1 + w22 + w32 b1 b2 = 2 = E 2 .
2
2
2
In what follows we shall assume that
H=
(2.33)
b1 b2 .
(2.34)
In this case (2.30) implies that

b 1 1 b 2 ,
b 2 2 .
(2.35)
Let us note also that the polynomial P ( ) in (2.31) can be interpreted as defining a
hyperelliptic curve of genus 2
s 2 + P ( ) = 0,
(2.36)
with s and being two complex coordinates. Thus, we have found that the most general
three-spin string solutions are naturally associated with hyperelliptic curves.
The system (2.30) allows one to achieve the full separation of the variables: dividing
one equation in (2.30) by the other one can integrate, e.g., 2 in terms of 1 and then obtain
a closed equation for 1 as the function of . In finding the solutions we need also to take
into account the periodicity
conditions (2.21) now viewed as conditions on 1 , 2 .
The spins Ji = Ji in (2.10) expressed in terms of 1 , 2 satisfy the following

relations
J2
J3
J1
+
+
= 1,
w1 w2 w3
(2.37)
2
w1 J1 + w2 J2 + w3 J3 = w12
+ w22
+ w32
d
(1 + 2 ),
2
13
(2.38)
J1
w13
J2
w23
J3
w33
1
w12 w22 w32
2
d
1 2 .
2
(2.39)
To find the energy (2.11) in terms of the spins Ji we need to express the frequencies wi and
the Neumann integrals of motion parameters ba in terms of Ji and use (2.33). After finding
a periodic solution of (2.30), this reduces to the problem of computing the two independent
integrals on the r.h.s. of Eqs. (2.38) and (2.39).
2.3. Moduli space of the multi-spin string solutions
We are thus interested in finding periodic finite-energy solitonic solutions of O(6) sigma
model defined on a 2-cylinder that carry three global charges Ji . They can be parametrized
by the three frequencies wi (or Ji ) as well by the two integrals of motion ba . The five
parameters (wi , ba ) may be viewed as coordinates on a moduli space of these solitons.
Because of the closed string periodicity condition in , general solutions will be
classified by two integer winding number parameters na which will be related to wi and
ba after solving the periodicity condition (2.21). In general, there will be several different
solutions for given values of J1 , J2 , J3 , i.e., the energy of the string E will be a function
not only of J1 , J2 , J3 but also of the values of na .
Depending on the values of these parameters (i.e., location in the moduli space) one
may find different geometric types of the resulting rotating string solutions. In particular,
the string may be folded (with topology of an interval) or circular (with topology of a
circle). A folded string may be straight (as in the one- and two-spin examples considered
in [4,6]) or bent (at one or several points) as in the general three-spin case discussed below
in Section 3. A circular string may have the form of a round circle as in the two-spin
and three-spin solutions of [1,2] or a more general bent circle shape as in the three-spin
solutions of Section 4 below.
Before turning to the S 5 rotation case, it is useful to review how these different string
shapes appear in the case of a closed string rotating in flat R 1,5 Minkowski space. In
orthogonal gauge, string coordinates are given by solutions of free 2d wave equation, i.e.,
= 0.
by combinations of ein( ) , subject to the standard constraints X 2 + X 2 = 0, XX
For a closed string rotating in the two orthogonal spatial planes 12 and 34 and moving
along the 5th spatial direction we find (cf. (2.9))
X0 = ,
X1 + iX2 = x1 ( )eiw1 ,
X3 + iX4 = x2 ( )eiw2 ,
X5 = p5 ,
(2.40)
with
w1 = n1 ,
w2 = n2 ,

x2 = a2 sin n2 ( + 0 ) .
x1 = a1 sin(n1 ),
(2.41)
14
Fig. 1. Image of a physical string at fixed moment of time in the Jacobian (the Liouville torus). The string winds
around the fundamental cycles with winding numbers n1 and n2 .
Here 0 is an arbitrary integration constant, and na are arbitrary integer numbers and the
conformal gauge constraint implies that
2 = p52 + n21 a12 + n22 a22 .
(2.42)
Then the energy, the two spins and the 5th component of the linear momentum are
E=
i.e.,
,

J1 =
n1 a12
,
2
J2 =
n2 a22
,
2
P5 =
p5
,
(2.43)
2
(2.44)
(n1 J1 + n2 J2 ).

To get the two-spin states on the leading Regge trajectory (having minimal energy for given
values of the two non-zero spins) one is to choose n1 = n2 = 1 with 0 = /2.
The shape of the string depends on the values of 0 and n1 , n2 . If 0 is irrational then
the string always has a circular (loop-like) shape. In general, the circular string will
not be lying in one plane, i.e., will have one or several bends. For rational values of 0 the
string can be either circular or folded, depending on the values of n1 , n2 .
Consider as the first example the case of 0 = 0. If n1 = n2 the string is folded and
straight, i.e., have no bends. Indeed, then X1 + iX2 is proportional to X3 + iX4 and thus
one may put the string in a single 2-plane by a global O(4) rotation. If both n1 and n2 are
either even or odd and different then the string is folded and has several bends (in the 13
and 24 planes). For example, if n2 = 3n1 then the folded string is wound n1 times and has
two bends (in this case x2 = x1 (4x12 3)).
Next let us consider the case of 0 = /(2n2 ). If n1 = n2 the string is an ellipsoid,
becoming a round circle in the special case of a1 = a2 [1]. The string is also circular if
n1 is even and n2 is odd. If, however, n1 is odd and n2 is even the string is folded. For
example, if n2 = 2n1 then the folded string is wound n1 times and has a single bend at one
point (in this case x2 = 1 2x12 ).
The structure of the soliton string solutions in curved S 5 case is analogous. Indeed,
the equations of motion of the Neumann system are linearized on the Jacobian of the
hyperelliptic curve (2.36) (see Appendix A). The image of the string in the Jacobian whose
real connected part is identified with the Liouville torus can wind around two non-trivial
cycles with the winding numbers n1 and n2 , respectively (see Fig. 1).
The size and the shape of the Liouville torus are governed by the moduli (wi , ba ).
Specifying the winding numbers n1 , n2 , two of the five parameters (wi , ba ) are then
uniquely determined by the periodicity conditions. The actual shape of the physical string
E=
p52 +
15
at fixed moment of time lying on two-sphere will depend on the numbers n1 , n2 and on the
remaining moduli parameters and may be of (bent) folded type or of circular type.
Let us now study the folded and circular string solutions in turn.
3. Folded string solutions

3.1. Folded bent strings with three spins
Our aim here will be to analyze folded string solutions of (2.30). We shall see that to
have three non-zero spins the folded string must be bent at least at one point.
By definition, a folded closed string configuration is such that for all of the coordinates
(0, ) = X (, ) = 0 (we choose = as a middle
XM (, ) = XM (2 , ), i.e., XM
M
point). In the case of our rotating ansatz (2.9) this leads to
xi ( ) = xi (2 ),
xi (0) = xi () = 0,
i = 1, 2, 3.
If xi vanishes for all i only at the two points, then the string has no bends. Such straight
folded string can carry only one non-trivial component of the spin in flat space, but in the
case of rotation in S 5 it may carry two non-zero spins [4].
To analyze when the folded string can carry three non-zero spins let us use the angular
(, ) parametrization of two-sphere formed by xi (2.22). Let us consider a folded string
stretched along and ,
0 ( ) 0 ,
and assume that

=
= 0,
2
2
0 ( ) + 0 ,
2
2
(3.1)
(0) = () = 0 ,
(0) = () = 0,

=
= ,
2
2
2
(0) = () = 0.
(3.2)
(0) =
0 ,
2
() =
+ 0 ,
2
(3.3)
This configuration is a folded string without bends. The case with two spins considered
in [4] corresponds to 0 = 0, so one may expect that to have a string with a small nonzero J3 one needs to consider a case with small 0 . However, it is possible to see that this
no-bend case is not a genuine three-spin casethere is a global SO(3) rotation that can
be used to eliminate J3 . Indeed, for a consistent semiclassical state interpretation one has
to check that only three components, J1 J12 , J2 J34 , J3 J56 of the SO(6) angular
momentum are non-vanishing. Assuming that the above string configuration exists when
all the frequencies wi are different, the angular momentum conservation of J36 requires
16
the vanishing of the following integral (cf. (2.22))

2
2
d x2 ( )x3 ( ) =
d sin cos sin .
(3.4)
However, it is easy to see that for the folded string configuration (3.3) the integrand here is
positive for any value of . Thus J36 (and also J45 ) do not vanish. We conclude, therefore,
that for the case of all wi different, the above string solution does not exist. Now consider
the case w2 = w3 . As soon as w2 = w3 , one can rotate the folded string to place it entirely
on the equator ( = /2) of S 2 inside S 5 . Then x3 ( ) = 0 and J3 = 0 for the transformed
configuration. The conclusion is therefore that the folded straight string configuration can
only correspond to a two-spin case, i.e., the periodicity conditions should imply w3 = w2 .
The reason why in the no-bend case one is able to rotate the string by a global SO(3)
transformation to set = /2 is that the folded string should be stretching along a geodesic
(i.e., some oblique big circle) of S 2 (, ) part of S 5 . This follows from the fact that both
of the derivatives and vanish only at the same (two ending) points ( = 0, ) of the
folded string (having a wiggle or a bend at some intermediate point of the string would
mean the vanishing of one of the derivatives or there).
We conclude that to find a non-trivial folded string solution with three spins we need to
admit a possibility of bends, i.e., the points on the string where one of the two coordinates
has zero derivative while the other does not (see also Appendix D). For example, if
changes its sign not only at = 0, but also at some 2n other points while does not
that would mean one has n folds in but only one fold in , implying the existence of n
bend points. In what follows we will be considering the simplest and most symmetric case
of a single bend point located in the middle of the folded string, i.e., with vanishing at
0 = /2 and 0 = 3/2; such configuration is expected to have minimal energy for given
values of the three spins.
To have a folded string with a single bend we will thus require

3
x3
xi (0) = xi () = 0,
= x3
= 0.
2
2
In terms of the coordinates 1 and 2 in (2.25) these conditions can be satisfied if

3
= 1
= w22 ,
= 1
= 0.
1
1
2
2
2
2
In view of Eq. (2.30) the second condition is equivalent to

3
= 2
= b2 .
2
2
2
(3.5)
(3.6)
The conditions (3.5), (3.6) mean that there should exist 4 points 1 , . . . , 4 located as
0 < 1 <
,
2
< 2 < ,
2
3
< 4 = 2 1 < 2
2
< 3 = 2 2 <
3
,
2
(3.7)
17
Fig. 2. Bent string. The point A corresponds to = /2, 3/2, the point Bto = 1 , 2 1 and the points
C and D are the turning points where = 0 and = . The turning points C and D are symmetric w.r.t. to the
plane 13. When b2 w32 the bend point A tends to equator and the string itself concentrates around = /2,
i.e., we recover a straight string representing the two-spin solution.
at which 2 = w3 , and x3 = 0. Passing through these points x3 changes its sign. The exact
positions of 1 , . . . , 4 are determined by the values of the parameters wi and ba of the
solution. Note that the middle of the folded string at = /2, 3/2 is not located at x1 = 1,
x2 = x3 = 0. Thus, at = 0 we have 1 = b1 and 2 = b2 . Increasing , both 1 and 2
increase until at = 1 the coordinate 2 reaches the point w32 . Then 2 changes its sign
and 2 begins to decrease. At = /2 the coordinate 1 becomes w22 and 2 reaches the
turning point b2 . Next, 1 begins to decrease and 2 begins to increase until at = 2 the
coordinate 2 reaches the point w32 where 2 changes its sign again. After that both 1 and
2 decrease and at = they reach the turning points 1 = b1 and 2 = b2 (see Fig. 2).
Thus, two derivatives 1 and 2 are both positive on the interval 0 < < 1 , and the
equations of motion (2.30) take the form
d1
d
,
=2
2 1
P (1 )
d2
d
.
=2
2 1
P (2 )
(3.8)
The periodicity conditions (2.21) follow from (3.8) and from the range (2.24), (2.35) of
1 , 2 (these conditions do not depend on 1 )
2
w2
b1
d1 (2 1 )
= ,
P (1 )
w3
2
b2
d2 (2 1 )
= .
P (2 )
(3.9)
The presence of the coefficient 2 in the second equation reflects the fact that we are
considering the single-bend solution. As a consequence of (2.30) we have also the
18
following relation which is valid for any point from the interval 0 2

d1
P (1 )

d = P ( ) .
2
2
(3.10)
Then
2
w2
c1
c1
b1
d1
=
P (1 )
w3
d2
,
P (2 )
d2
P (2 )
b2
c1 1 (1 ),
d1
=
P (1 )
w3
b2
(3.11)
and, therefore,
2
w3
2
b2
d2
=
P (2 )
w2
b1
d1
.
P (1 )
(3.12)
The conditions (3.9) and equations of motion for 1 , 2 (3.8) also imply
2
w3
2
b2
2 d2
P (2 )
w2
b1
1 d1
= .
P (1 )
(3.13)
It will be convenient for the analysis in next subsections to make the following change of
variables 1,2 1,2

1 = w22 w22 b1 1 ,
2 = w32 w32 b2 2 .
(3.14)
Then Eq. (2.30) take the form

w2 (b2 w22 ) 1 (1 1 )(1 t1 1 )(1 u1 1 )(1 v1 1 )
d1 2
= 4 21 2
,
d
(1 u1 1 u2 2 )2
w32

w2 (w2 b1 ) 2 (1 2 )(1 t2 2 )(1 u2 2 )(1 v2 2 )
d2 2
= 4 31 32
.
d
(1 u1 1 u2 2 )2
w32
(3.15)
(3.16)
Here we introduced the parameters

t1 =
t2 =
w22 b1
2
w21
w32 b2
2
w31
> 0,
u1 =
> 0,
u2 =
w22 b1
2
w32
w32 b2
2
w32
< 0,
> 0,
v1 =
v2 =
w22 b1
b2 w22
w32 b2
w32 b1
< 0,
> 0.
(3.17)
(3.18)
19
3.2. System of equations for the leading correction to the energy

Let us now concentrate on the case of solutions which have all three spins very large
(1.1). The energy for such solutions is expected to scale as in (1.2). Our aim is to derive a
closed system of equations which allows one to find the leading-order function f1 in (1.2),
thus determining the one-loop (first order in ) correction to conformal dimensions of the
corresponding dual SYM operators with free-field theory dimension 0 = J1 + J2 + J3 in
SU(4) irreps with Dynkin labels [J2 J3 , J1 J2 , J2 + J3 ]. The same system of equations
should then be expected to emerge from the thermodynamic limit of the algebraic Bethe
ansatz for the SO(6) spin chain derived from the SYM theory [3,19,21].
The two-spin case [4] and the relation (2.33) for the energy of the string suggest that
one should look for solutions which have the following moduli parameters

Ji
1
2
+O
,
+ i
wi2 = Jtot
2
Jtot
Jtot

Ji
1
2
ba = Jtot + a
(3.19)
+O
,
2
Jtot
Jtot
where
Jtot =
Jtot =
(J1 + J2 + J3 ),
(3.20)
is the total spin of the string. We shall assume that in the large Jtot limit the corrections
i and a depend only on the ratios Ji /Jtot = Ji /Jtot . Note that the presence of a linear
O(Jtot ) term in i2 and ba in (3.19) can be ruled out by parity (wi wi , etc.)
considerations.
Then the relation (2.33) implies that

1
2
+ (1 + 2 + 3 1 2 ) + O
,
E 2 = Jtot
(3.21)
Jtot
i.e.,
E = Jtot +

Ji
1
+O
,
f1
3
Jtot
Jtot
Jtot
1
f1 = (1 + 2 + 3 1 2 ).
(3.22)
2
We want therefore to find the expressions for i and a and thus for f1 in terms of the
current ratios Ji /Jtot .
The five parameters i , i can be found by solving a system of five equations that follows in the limit Jtot from the periodicity conditions (3.12) and (3.13), Eqs. (2.37),
(2.38), and one of Eq. (2.10) expressing spins through the parameters. This system can be
written as follows in terms of the hyperelliptic integrals I10 , I20 , I11 , I21 , I22 defined in
Appendix B
2I20 = I10 ,
(3.23)
2I21 = 32 I10 + I11 ,
(3.24)
20
Ji
1 j1 + 2 j2 + 3 j3 = 0, ji
,
Jtot

2 1 2
1
1 + 2 3 +
I0 + 32 I11 (2I22 I12 ) = 0,
2 32
2
1
2t2
d
d1
(1 u1 1 )2 (1 ).
j3 =
d1
(3.25)
(3.26)
(3.27)
Here ij i j and
32 (1 u1 1 u2 2 )
d
=
d1
2 21 (2 2 )
1
(3.28)
,
1 (1 1 )(1 t1 1 )(1 u1 1 )(1 v1 1 )
which follows from Eq. (3.8). We could use any of the other two equations in (2.10) instead
of Eq. (3.27).9 All of the parameters in (3.17), (3.18) and in the above system of equations
depend only on the ratios ji = Ji /Jtot = Ji /Jtot .
Since the integral in (3.27) cannot be computed analytically, there are at least two ways
to proceed for a generic three-spin case. One may try to compute this integral and therefore
to solve the whole system (3.23)(3.27) numerically (see Section 3.4). One may also try to
develop a perturbation theory around the two-spin solution of [4] assuming that the third
spin component J3 is small as compared to the total spin Jtot . Since the two-spin solution
was expressed in terms of the elliptic functions, one could expect that the same should be
true for a perturbative expansion at the vicinity of the two-spin solution. This is indeed
what happens as will be explained in the next subsection.
3.3. Two-spin solution and small J3 expansion
As was mentioned above, the limit b2 w32 should correspond to the two-spin solution.
In this case the folded string stretches along the equator, i.e., it is straight (without bends).
This limit corresponds to a degeneration of the hyperelliptic curve (2.36) governing the
string dynamics into an elliptic one.
To see explicitly how this happens let us change the variables 1,2 1,2 as in (3.14).
In the limit
D w32 b2 0
one finds that t2 , u2 , v2 0 and u1 v1 so that Eqs. (3.15), (3.16) simplify to

d1 2
2
= 4w21
1 (1 1 )(1 t1 1 ),
d

d2 2
t1 2 (1 2 )
2
= 4w21
(1 u1 ) 1
.
d
u1 (1 u1 1 )2
(3.29)
(3.30)
9 Let us also note that we cannot use Eq. (2.38) to determine the leading order correction f because at this
1
order Eq. (2.38) is a consequence of (3.25) and (3.26).
21
Now one can recognize in Eq. (3.29) the differential equation for the elliptic sn function,
i.e., its solution is

2 , t .
1 = sn2 K(t1 ) w21
(3.31)
1
Here K denotes the complete elliptic integral of the first kind.10 In writing Eq. (3.31) we
have also used the fact that 1 is negative on the interval 0 < < /2 and that 1 (0) = 1
due to the identity sn(K(t1 ), t1 ) = 1. Eq. (3.31) defines an elliptic curve with modulus t1 .
Then the second equation (3.30) can be integrated to find 2 . Note that in the two-spin case
the variable 2 is an auxiliary one as the physical coordinates x1 , x2 (x3 = 0, = /2) are
parametrized in terms of 1 only (cf. (2.25), (3.14)). The simplicity of the two-spin case is
thus related to the fact that the equations for 1 and 2 decouple when b2 = w32 .
To determine the parameters (t1 , u1 ) entering Eqs. (3.29) and (3.30) from the periodicity
conditions (3.12) and (3.13) let us study in more detail the relation (3.10). In general, the
latter implies the following two equalities
I1 (1 ) = I2 (2 ) + I2 (b2 ),
I1 (1 ) = I2 (2 ) + I2 (b2 ),
b1 1 c1 , b2 2 w32 ,
c1 1 w22 ,
b2 2 w32 ,
(3.32)
(3.33)
where we have introduced the two period integrals of the Abelian differential of the first
kind:
2
w2
I1 (1 ) =
1
dz
,
P (z)
w3
I2 (2 ) =
2
dz
.
P (z)
(3.34)
Making the change of integration variable z w22 (w22 b1 )1 z for I1 , and z

w32 (w32 b2 )2 z for I2 , and using (3.14) we obtain a more useful form of I1 , I2 given in
Appendix B. In the limit D = w32 b2 0 both integrals can be easily computed and one
finds
2F(u1 , arcsin 1 , t1 )
2 arcsin 2

,
I
, (3.35)
I1 (D 0) =
2 (D 0) =
2 w 2 (b w 2 )
2 w 2 (w 2 b )
w21
w
1
32 2
2
31 32
3
where F denotes the incomplete elliptic integral of the third kind. Now (3.32) allows one
to solve for 2 in terms of 1 , namely

t1
2
2 (1 ) = cos
(3.36)
(1 u1 ) 1
F(u1 , arcsin 1 , t1 ) .
u1
Since a ba implies that a 1 we find the following transcendental equation relating
the parameters u1 and t1 :

u1
.
F(u1 , t1 ) =
(3.37)
(1 u1 )(u1 t1 )
10 We recall the definitions of the elliptic integrals in Appendix B.
22
Then Eq. (3.36) can be written as

F(u1 , arcsin 1 , t1 )
2
2 (1 ) = cos
.
F(u1 , t1 )
(3.38)
This relation is valid for all values of 2 from the interval [0, 1]. In particular, one
recognizes that 2 (1) = 2 (0) = 1 and 2 (c1 ) = 0, where c1 0.56862. This dependence
of 2 on 1 is a reflection of the U -shaped form of our string.
Let us now consider the periodicity condition (3.13):
2I21 I11 = ,
(3.39)
where we have introduced the following period integrals of the other Abelian differential
of the first kind:
2
w2
I11 =
b1
w3
z dz
,
P (z)
I21 =
b2
z dz
.
P (z)
(3.40)
The same change of variables as for (3.32) allows one to compute these integrals in the
limit D 0 with the result

w32
2 w 2 (w 2 b )
w31
1
32
3
w22
2 w4
u1 w21
32

pK(t1 ) + (u1 p)F(u1 , t1 ) = ,
2
(3.41)
where p = (w22 b1 )/w22 > 0. With the help of (3.37) the last equation reduces to

2 .
w21
K(t1 ) =
(3.42)
2
Eqs. (3.37) and (3.42) completely determine the parameters t1 , u1 of the two-spin solution
in terms of the frequencies w1 , w2 . Note that the requirement 1 (/2) = 1 (3/2) = 0
produces the same Eq. (3.42) for t1 .
In the limit D 0 we can compute the integral
/2
1
(1 + 2 ) d = (2I22 I12 ),
2
(3.43)
where
2
w2
I12 =
b1
z2 dz
,
P (z)
w3
I22 =
b2
z2 dz
.
P (z)
(3.44)
As a result,
/2
2
2
d (1 + 2 ) =
E(t1 ),
w1 + w32 + w21
2
0
(3.45)
23
where E is the complete elliptic integral of the second kind. Then in the limit D 0
Eqs. (2.37) and (2.38) reduce to the following equations
J1
J2
+
= 1,
w1 w2
(3.46)

2
2 E(t ),
w21
(3.47)
1
with Eq. (2.39) being their consequence. Summarizing, we have shown how the three-spin
hyperelliptic solution degenerates to the two-spin elliptic one, the later being completely
determined by the system (3.37), (3.42), (3.46) and (3.47). The energy of the two-spin
solution is determined from
w1 J1 + w2 J2 w22 =
2
t1 .
E 2 = 2 = w12 + w21
(3.48)
In the two-spin case with J1 = J2 J , J3 = 0 we know that [4]

t1 = 0.826115 + ,
w1 = 2J
0.272922
,
J
0.272922
.
J
We can then use (3.37) to determine w3 . The result is
w2 = 2J +
(3.49)
1.70597
.
(3.50)
2J
Finally, it is now easy to compute the energy of the three-spin string up to the term linear
in j3 J3 /Jtot . To this end we should expand the system of Eqs. (3.23)(3.27) in D and use
the two-spin solution as the zero-order approximation. Then this system reduces to a linear
system which can be readily solved. We find that the parameters i , 1 , u1 , t1 in (3.17),
(3.19) have the following expansion in j3 :
u1 = 0.777383,
2
w32
= 2.32025,
w3 = 2J +
u1 = 0.777383 + 0.424835j3,
J3
,
Jtot
t1 = 0.826115 + 0.183849j3,
1 = 1.09169 3.86759j3,
2 = 1.09169 2.95629j3,
3 = 3.41194 0.20351j3,
1 = 0.712032 4.11054j3.
D w32 b2 = 3 2 = 6.12528j3,
j3
(3.51)
Using these values of the parameters we find for the energy

2
2
+ 1 + 2 1 + D = Jtot
+ 0.712032 + 3.41194j3,
E 2 = Jtot
(3.52)
i.e.,
2
E 2 = Jtot
+ 0.712032 + 3.41194
J3
.
Jtot
Thus the energy has the form (1.6) with positive coefficients

J3
1 + 4.79183
.
E = Jtot + 0.356016
Jtot
Jtot
(3.53)
(3.54)
24
Table 1
Parameters for string configurations with different values of angular momenta
w12
w22
w32
b1
b2
c1
J1
J2
J3
E 2
23.52
34.65
47.63
1
25
49
49
49
49
25.80
36.89
49.88
4
28
52
53
55
51
28.29
39.40
52.40
9
33
57
59
64
55
23.87
35.01
47.99
1.14
25.14
49.14
49.06
49.01
49.36
27.81
39.00
51.97
4.64
28.64
52.64
53.36
55.09
52.27
24.70
35.82
48.80
2.31
26.31
50.31
50.61
51.33
50.08
2.20
2.75
3.21
0.26
1.30
1.82
1.50
1.16
2.69
2.32
2.86
3.32
0.59
1.56
2.13
1.96
1.59
2.18
0.48
0.38
0.47
1.33
2.55
3.35
3.97
4.96
2.30
1.06
0.94
0.96
3.45
2.88
2.83
3.42
4.47
1.92
(0)
The coefficient f1 = 0.356016 is the same as in the two-spin solution [4],11 while
(1)
f1 = 4.79183 is the string-theory prediction for the term linear in J3 in the one-loop
anomalous dimension of the dual CFT operator.
3.4. Comments on folded string solutions with general values of J1 , J2 , J3
Let us now study more general folded string solutions which can be far from the
two-spin configuration. This can be done numerically as follows. One starts with three
values w1 , w2 , w3 as input parameters and solves the two periodicity conditions (3.12),
(3.13), for the unknowns b1 and b2 . Then one determines the parameter c1 = 1 (/4)
by solving numerically Eq. (3.11) (or, equivalently, Eq. (3.11)). With these values
w1 , w2 , w3 , b1 , b2 , c1 one can then compute
2
d
2
(1 + 2 ) =
2
b2
2
0
w3
1
z2 dz
P (z)
w2
b1
z2 dz
,
P (z)
(3.55)
d
1
1 2 =
2
w2
b1
zz2 (z)(z2 (z) z) dz
P (z)
(3.56)
and find J1 , J2 , J3 from Eqs. (2.37)(2.39). The parameters of different solutions obtained
in this way are shown in Table 1.
The first three entries are cases close to the two-spin case, which are indeed consistent
with Eq. (3.53) for the energy. The input values of w1 , w2 , w3 are obtained from Eqs. (3.51)
using J3 = 0.5, Jtot = 5 for the first entry, and J3 = 0.4, Jtot = 6, J3 = 0.5, Jtot = 7, for
the second and third entries. According to Eq. (3.53), the perturbation-theory values for the
correction to the energy
2
= E 2
E 2 E 2 Jtot
(0)
11 This coefficient has a simple origin: f

1
solution of K(t) = 2E(t).
= (1/ 2 )(2t 1)[K(t)]2 , where t = 0.826115 is the (unique)
25
found in the expansion in powers of J3 /Jtot are, respectively, E 2

= 1.04, E 2
= 0.96,
2
E = 0.94. They agree with the results in Table 1.

The values of J1 , J2 , J3 are also in good agreement with direct perturbation theory
results. Small differences are expected, in view of the higher-order corrections in powers
of J3 /Jtot and in view of the fact that Jtot is not very large (the results of Table 1 represent
summation of all terms in 1/Jtot expansion, while the (3.53) contains only the leading
correction term). Other cases in Table 1 are far from the two-spin case, i.e., have J3 of the
same order as (or larger than) J1 , J2 .
In general, a random choice of w1 , w2 , w3 may not correspond to a folded string
solution. For large Ji , there are no folded string solutions when the wij2 are not small
compared to the wi2 .
2 , w 2 , but increasing
We have considered some cases with the same values of w31
21
values of wi . They exhibit the following interesting fact. The differences b1 w12 and
b2 w22 , c1 w12 are always the same. The difference in the energy E 2 approaches some
asymptotic value as Ji increase.12 For the particular entries of Table 1, one observes that
as the differences wi2 wj2 increase, b1 gets closer to w12 and b2 gets closer to w22 .
In conclusion, there exist folded string solutions for diverse values of J1 , J2 , J3 . In the
case when J3 is smaller than J1 , J2 , the numerical calculation reproduces the perturbation
theory results of the previous subsection.
4. Three-spin string solutions of circular type

4.1. String solutions of circular type in ellipsoidal coordinates
Let us start with recalling that our parameters are assumed to satisfy in general the
conditions (2.24), (2.35), i.e.,
w12 1 w22 2 w32 ,
b 1 1 b 2 2 .
(4.1)
As was discussed in Section 3, to describe a folded string we should consider b1 lying in

the same range as 1 , and b2 lying in the same range as 2 , i.e.,
w12 b1 w22 ,
w22 b2 w32 .
(4.2)
To find a circular string solution we should relax at least one of the conditions (4.2), i.e.,
to assume that either b1 or b2 do not belong to the corresponding intervals. Thus there are
two different cases to be considered
(I)
w12 < b2 < w22 ,
(II)
b1 < w12 .
Let us start with the case (I). We have then two options for the value of b1 :
(i)
w12 < b1 < w22 ,
or (ii)
b1 < w12 .
12 For very large values of J the difference E 2 E 2 J 2 should depend only on the ratios J /J .
i
i tot
tot
26
In what follows we shall consider the case (i), because the case (ii) appears to similar to
(II).
When w12 b1 w22 , we have
b 1 1 b 2 ,
w22 2 w32
and there are many different circular string configurations with these values of parameters.
Let us first consider the simplest one corresponding to
b1 = b2 b,
i.e., 1 = b.
Making the change of variable 2 2 ,

2
2 ,
2 = w32 w32
we get from (2.30) the following equation for 2

w2
d2 2
2
= 4w31
2 (1 2 )(1 t2 ), t 32
.
2
d
w31
(4.3)
Assuming the initial condition 2 (0) = 0, the solution of this equation is

2

(4.4)
2 = sn w31
, t .
According to this formula the function 2 oscillates between 1 and 1 as goes around
the string. Hence, if we want our solution to describe a circular type string
with a winding
2 , i.e.,
number n, the real period of the function sn should be equal to (2/n) w31

2 = 4nK(t).
2 w31
(4.5)
We used the fact that the elliptic function 4K is the real period of the function sn. Since
Eq. (4.5) appears to be similar to (3.42) it may not be apparent why we are dealing with
strings of circular type rather then with multifolded strings. To clarify this point let us write
down the expressions for the physical-space coordinates xi in (2.25):

2
2 1/2
w32
b w12 w31
w22 b
2
x1 =
,
x
=
(1 2 )1/2 ,
2
2
2
2
w21
w31
w21

w32 b 1/2
2 .
x3 =
2
w31
When 2 changes from 0 to 1 all xi remain non-negative. However, x2 , x3 can acquire
zero value, i.e., can reach the boundary of the coordinate patch xi 0 and, therefore, they
may change sign. Change of the sign means that we go to another coordinate patch on S 2 .
: one finds that
This is indeed the case as one can see by computing the derivatives x2,3
x2 changes its sign when 2 crosses 1, while the sign of x3 changes when 2 crosses zero.
Thus, the shape of the resulting string configuration appears to be of a circular type.
Conversely, for the folded string configurations of Section 3, the coordinate x3 is zero,
while x1 remains strictly positive as 1 changes from 0 to 1. We see once again that the
27
shape of the string depends essentially on the relation between the parameters of the moduli
space (cf. Section 2.3) which explicitly occur in the expressions for the coordinates xi in
terms of the variable 2 : xi = xi (2 ; wi , ba ).
It is useful to note that in the limit w3 w2 the periodicity condition (4.5) reduces to
2 = n2 and thus we recover the circular three-spin solution found in [1]:
w31

(4.6)
2 sn(n, 0) = sin n.
We want to emphasize that neither the solution (4.4) nor the periodicity condition (4.5)
depend on b. This parameter occurs only in the expressions of xi through 2 and it tells us
how many spins our solution has. For generic b we have a three-spin solution. A two-spin
solution arises in the limit b w12 (or b w22 ). In this limit, x1 = 0, i.e., J1 = 0, while

x2 = 1 2 ,
(4.7)
x3 = 2 .
The spins J2 and J3 are then easy to compute
2
J2 = w2

d
w2
E(t)
(1 2 ) =
t 1+
,
2
t
K(t)
(4.8)

w3
d
E(t)
2 =
1
,
2
t
K(t)
(4.9)
2
J3 = w3
0
where we made use of the explicit solution (4.4). These two equations can be used to
2 /w 2 can be found from (4.5). The
express w2,3 through J2,3 , while the parameter t = w32
31
energy is then given by
2
w32
.
(4.10)
t
We will not go here into the detailed analysis of the two-spin circular type solution above,
we will do it in Section 5, where a comparison with the gauge theory results will be
presented.
Let us consider now the limiting case when w3 w2 . We also assume that w12 < b1
and therefore b1 < 1 < b2 . First we perform the following change of variables
E 2 = w22 +
1 = b2 b21 1 ,
2
2 = w32 w32
2 ,
b21 b2 b1 .
Then taking the limit w3 w2 we get from (2.30) the following equations for 1 , 2

d1 2
= 4h1 (1 1 )(1 t1 1 ),
(4.11)
d

t1 2 (1 2 )
d2 2
= 4h(1 u1 ) 1
,
(4.12)
d
u1 (1 u1 1 )2
where we introduced the parameters
0 < t1 =
b21
< 1,
b2 w12
u1 =
b21
< 0,
b2
w22
h = b2 w12 > 0.
(4.13)
28
These equations have the same form as (3.29) and (3.30) governing the two-spin solutions
for the folded string but there is an important difference. Now the variable 2 enters the
parametrization of the physical coordinates xi :
x12 =
x32 =
b2 w12
2
w21
w22 b2
2
w21
(1 t1 1 ),
x22 =
w22 b2
2
w21
(1 u1 1 )(1 2 ),
(1 u1 1 )2 .
(4.14)
It is easy to see that the corresponding shape of the string is again of a circular type.
As was pointed out in Section 2.3, generic winding string solitons are parametrized by
we assume that the variable
two
integers (n1 , n2 ). To construct a corresponding solution

1 oscillates between 1 and 1 with period (2/n1 ) h. As above, this gives an equation
that determines the parameter t1 :

h.
K(t1 ) =
(4.15)
2n1
Then Eq. (4.12) can be integrated to give

d
t1
,
2 = sin h(1 u1 ) 1
u1
1 u1 1 ( )
(4.16)
where the initial condition 2 (0) =

0 was assumed.
Suppose now that the variable 2 performs n2 oscillations between 1 and 1 as
runs from 0 to 2 . For = 2 the integral on the r.h.s. of (4.16) can be easily evaluated
by a change of variables, and we obtain the second elliptic function equation to determine
the parameter u1 :

u1
n2
.
F(u1 , t1 ) =
(4.17)
2 n1 (1 u1 )(u1 t1 )
Eqs. (4.15) and (4.17) generalize (3.42) and (3.37) for arbitrary winding numbers n1 and
n2 . At this point it should be mentioned that imposition of the relation w3 = w2 implies that
the three spins Ji are not any more independent but rather satisfy an additional constraint.
Indeed, our solution is governed by four parameters w1 , w2 , b1 , b2 which obey the five
equations, three of them determine the spins Ji through these parameters and the other
two are the periodicity conditions (4.15) and (4.17). Therefore, one of the equations is in
fact a constraint for Ji .
The case of the round-circle string can be recovered by taking first the limit t1 0
and then u1 0. In this double limit Eqs. (4.15) and (4.17) reduce to
b2 w12 = n21 ,
n1 = n2 .
(4.18)
Let us consider now the case (II) and show that it is quite similar to the case (I). This
time we have the following range of parameters
b1 w12 < w22 < b2 < w32 ,
(4.19)
29
and thus
w12 1 w22 ,
b2 2 w32 .
(4.20)
To demonstrate that we are dealing with string solutions of circular type we consider the
limit b2 w32 . Performing a change of variables

2
1 ,
2 = w32 w32 b2 2 ,
1 = w12 + w21
(4.21)
and taking the limit b2 w32 one obtains the same system of Eqs. (4.11) and (4.12)
parametrized by
t1 =
2
w21
w12 b1
< 0,
u1 =
2
w21
2
w31
> 0,
h = w12 b1 > 0.
The coordinates xi depend only on the variable 1

x1 = 1 ,
x2 = 1 1 ,
x3 = 0.
(4.22)
(4.23)
Applying the same arguments as above we conclude that the form of the string is of a
circular type.
One can easily develop a perturbation theory around the circular type solutions by using
the equations expressing spins through wi and ba . Assuming the same ansatz for these
parameters as in (3.19) one can obtain a system of equations which should determine the
leading-order correction to the energy in (1.2), i.e., a one-loop anomalous dimension for
the corresponding SYM operators. Again, an equivalent system of equations should follow
from the Bethe ansatz approach on the field-theory side.
4.2. String solutions of circular type in spherical coordinates
As was already discussed, when all the frequencies wi are different, the ellipsoidal
coordinates allow one to separate the variables and formulate a closed system of equations
that determines the energy E as a function of spins Ji . However, when two of the three
frequencies coincide, the U(1) subgroup of O(3) in (2.1) is restored and it is useful to adopt
the spherical coordinates. To make a contact with [1] in this and the following subsection
we relabel x1 x3 , x3 x2 and consider the case when
w1 = w2 > w3 .
(4.24)
The spherical coordinates are global and thus particularly appropriate for the study of
solutions with non-minimal energy that represent strings of circular type. Here we shall
first present the equations for the spherical coordinates and then relate them to the above
discussion of the circular type strings in the ellipsoidal coordinates.
The equations of motion for the spherical coordinates and follow from the action
(2.1) and the metric (2.3), and in the case w1 = w2 take the form

sin2 = 0,
+

2
1
2 = 0.
sin 2 w13
2
(4.25)
30
Then
=
c
sin2
c2
c = const,
(4.26)
cos
1 2
+ w13
sin 2 = 0,
2
sin
3
(4.27)
and the conformal gauge constraint reduces to

2 = 2
c2
sin2
w12 sin2 w32 cos2 .
This can be rewritten as

2

2
x 2 = w13
x a a+ x 2 ,
x x3 = cos ,
where the constants a are

1 2
2
2
2 w 2 2 4c 2 w 2 .
w
+
w
a =
13
1
3
13
2
2w13
(4.28)
(4.29)
(4.30)
Clearly, to have two different turning points a and a+ we have to impose a condition13
0 < a 1. Note also that a+ = 1 for c = 0. The expressions of the integral of motion c
and the energy of the system in terms of the turning points a are
2
c2 = w13
(1 a+ )(1 a ),
2
2 = w12 + (1 a a+ )w13
.
(4.31)
To establish a connection with the discussion of the circular type strings in terms of the
ellipsoidal coordinates, we introduce a new variable
=
x 2 a
.
a+ a
Then Eq. (4.29) takes the same form as (4.11) with parameters
2
,
h = a w13
t1 =
a+ a
< 0.
a
(4.32)
As to Eq. (4.26), we rewrite it as

2 =
1
c2
(1 a )2 (1 u1 )2
(4.33)
with
0 < u1 =
a+ a
< 1.
1 a
(4.34)
Using (4.31), one can establish a direct correspondence between Eqs. (4.33) and (4.12).
The periodicity conditions are then given by (4.15), where n1 = 1 and by (4.17). In this
way one recovers the ellipsoidal coordinate description of the circular type strings.
13 This implies w < w , w < .
3
1
3
31
4.3. Energy of circular type strings

In this subsection we shall find numerical solutions of the periodicity conditions for
the circular strings and calculate their energy in the special case of two equal frequencies
(4.24).
Let us define the two turning angles 1 , 2 by
a = cos2 1 ,
a+ = cos2 2 ,
so that
c = w13 sin 1 sin 2 ,
w13 =

w12 w32 .
The string extends from = 1 (which is the value closer to the equator point = /2) to
= 2 .
The string energy can be determined from
2 2

E 2 = 2 = w13
(4.35)
sin 2 cos2 1 + w12 .
Our aim is to express E as a function of J3 and J1 + J2 , i.e., of
J
J3
J ,
J J1 + J2 .
J3 ,
The parameters w1 , w13 are given by

J2
J w3
,
w13 = w3
1,
w1 =
w3 J3
(w3 J3 )2
(4.36)
(4.37)
so we have three unknowns w3 , 1 , 2 and three equations:

w3
J3
K2 (1 , 2 ),
n = K3 (1 , 2 ),
(4.38)
=
w13
where K1 (1 , 2 ), K2 (1 , 2 ), K3 (1 , 2 ) are the integrals which in the previous subsection
where computed in terms of the elliptic functions,
w13 = K1 (1 , 2 ),
2
K1 (1 , 2 ) =
1
2
K2 (1 , 2 ) =
1
sin
d
,
(cos2 2 cos2 )(cos2 cos2 1 )
(4.39)
sin cos2
,
d
(4.40)
2
d
K3 (1 , 2 ) = sin 1 sin 2
1
(sin )1
(4.41)
The system (4.38) can be reduced to a system of two equations and two unknowns 1 , 2
by noting that
J3
K2 (1 , 2 )
.
=
w3 K1 (1 , 2 )
32
Using (4.37), the second and third equations of (4.38) become:

J2
K2 (1 , 2 ) = J3
n = K3 (1 , 2 ),
2 1.
1 (1 ,2 )
J32 K
1
K2 (1 ,2 )
(4.42)
These can be solved numerically for 1 , 2 . Note that for large J3 the square root on the
right-hand side of the second equation must be very small. This gives a hint for the values
of the parameters 1 , 2 which solve these equations. We shall always consider the case
when J3 , J 1.
The general features which result from numerical analysis are the following. The above
system has solutions for all possible n = 1, 2, 3, . . . . For given n, there is a minimal value
of J3 /J. A lower bound can be obtained for large n. For n 1, K3 must be large, which
implies that the angles 1 , 2 are close to /2. In this region, one can approximate the
integrals, leading to the bound J3 /J > 1/4n2 .
When J3 /J is small, the solution has maximum eccentricity, with 1 near /2 and 2
near 0. As J3 /J increases, the angle 1 increases and 2 decreases, until some critical value
of J3 /J where 1 = 2 . For higher values of J3 /J, there is no solution.
At the critical value, one has a circular string. Indeed, since ( ) in this case is constant,
then
= n = const
and the solution reduces to the round-circle string solution of [1,2]. The critical value of
J3 /J depends on n. To compute it, we take the limit 2 1 and large J3 of the equations
in (4.42). This implies J/J3 K1 /K2 1. For 1 = 2 0 , one has K2 /K1 cos2 0 ,
and K3 /(2 cos 0 ), so the equations become
cos 0 =
1
,
2n
tan2 0 =
J
.
J3
(4.43)
One also has w13 = n. The second equation can also be written as sin2 0 = J/(J + J3 ),
which coincides with the expression for the angle 0 of the circular solutions obtained in
[2]. The first equation cos 0 = 1/(2n) implies that for given n, there is a single value of
J/J3 which gives circular solutions
J
= 4n2 1.
J3
(4.44)
This is in contrast to [2], where there are circular solutions for any J/J3 at fixed n. The
origin of this extra condition can be traced back to the condition K1 = w13 coming from
imposing periodicity of ( ). In the circular solution of [2] with = 0 = const, the
condition cos 0 = 1/(2n) does not appear, since constant is already periodic without
need to impose extra conditions. The origin of this extra condition can also be understood
),
by considering perturbations around the = const solution. Expanding = n + (
( ) = 0 + ( ), and substituting into the general equation (4.27), we find to first order
+ (2n cos 0 )2 = 0.
(4.45)
33
This has solution ( ) = a sin(m ), m 2n cos 0 . Imposing periodicity of 1 , we find

that m must be an integer. The solution with a single winding in is in fact m = 1,
or cos 0 = 1/(2n), which is precisely the condition obtained above by taking the limit
2 1 on the general solution.
Thus, for a given winding n in ( ), there exist perturbations of the round-circle
solutions only if a circular string is located at a special angle cos 0 = 1/(2n) (i.e.,
for special values of J/J3 ): only such discrete set of circular strings admit regular
deformations.
Returning to the general case, an important feature of the solution is that in the infinite
J3 , J limit with fixed J/J3 , the correction to the energy
E 2 E 2 (J + J3 )2 ,
E 2 = 1 E 2
is a function (as in (3.21)) of the ratio J/J3 only

J
2
2
E = (J + J3 ) + f
.
J3
(4.46)
Just as in the folded string case in Section 3.2, this limit effectively singles out the leading
correction to the energy (4.35) (cf. (1.2))

J
f
E = J + J3 +
(4.47)
+ .
2(J + J3 ) J3
A summary of numerical results for n = 1 is given in Table 2.
For J/J3 = 3, we see that the angles 1 and 2 are nearly equal. This means that is
approximately constant, so this is the case of the circular string mentioned above lying at
= 1.0476 = 0.3334 /3.
For the circular string solution with = nconst, the energy and the angle are given
by [2]
2
Ecirc
= (J + J3 )2 +
n2 J
+ ,
J + J3
sin2 0 =
J
+ ,
J + J3
(4.48)
where dots stand for terms which vanish at large J3 , J. In the case J/J3 = 3, n = 1, we find
2 = 0.75, and = /3, which is in full agreement with the case J/J = 3 of Table 2.
Ecirc
0
3
As a function of J/J3 , E 2 can be roughly approximated by a straight line. A more
accurate fitting is by adding a (J/J3 )2 correction (see Fig. 3):

J
J2
J
E 2 = f
(4.49)
= c0 + c1 + c2 2 ,
J3
J3
J3
c0
(4.50)
c1
c2 = 0.0002.
= 0.524,
= 0.068,
This formula has a different structure as compared to the round-circle case, Eq. (4.48).14
The string configurations are in general very different. In particular, as explained above,
for given n there is a minimum value of J3 /J, so these solutions do not include the circular
14 It is also possible to fit the data in terms of an expansion in, say, J/J or J /J , J = J + J . The formula
tot
3 tot tot
3
(4.49) is given as a simple book-keeping of the numerical data.
34
Table 2
Energies of string configurations with n = 1, lying between angles 1 and 2 . A few cases with equal J/J3 are
included to illustrate explicitly that for these large values of J, J3 the angles and energies depend only on the ratio
J/J3
J
J3
J /J3
925
900
850
600
400
525
350
1500
500
450
2000
800
700
1500
100
100
100
75
50
75
50
250
100
100
500
200
200
500
9.25
9
8.5
8
8
7
7
6
5
4.5
4
4
3.5
3
1.57061
1.57055
1.57038
1.57010
1.57010
1.56879
1.56879
1.56485
1.56894
1.56740
1.56427
1.56427
1.43237
1.04763
0.17367
0.17874
0.18987
0.20253
0.20253
0.23406
0.23406
0.27827
0.34622
0.39740
0.47176
0.47176
0.59732
1.04677
E 2
1.316
1.291
1.242
1.194
1.194
1.097
1.097
1.002
0.910
0.866
0.824
0.824
0.785
0.750
Fig. 3. Fitting of E 2 as a function of J/J3 by E 2 = c0 + c1 J/J3 + c2 J2 /J32 .
solution with J3 = 0 surrounding the equator = /2 as a limit. The present solution with
single (n = 1) winding exists only for J/J3 lying in the interval between the minimal value
at J/J3 = 3 and maximal value J/J3
= 10.
As a result, we get the following expansion for the energy of the corresponding string
state with charges J and J3 :

J
J2
c0 + c1 + c2 2 + O 2 .
E = J + J3 +
(4.51)
2(J + J3 )
J3
J3
35
Let us comment on solutions with higher winding numbers.15 For n = 2 the critical
value at which the string becomes circular is J/J3 = 4n2 1 = 15, and 0 = 1.32 (this can
be compared with the value 0 = /3
= 1.047 appearing at n = 1, J/J3 = 3). For n = 3,
the critical value is J/J3 = 35, and 0 = 1.40, etc. The larger is n, the angles 1 , 2 get
closer to the value = /2. In particular, for n = 2 one has solutions with:
(i) J/J3 = 10, 1 = 1.559, 2 = 0.791, E 2
= 3.62;
(ii) J/J3 = 12, 1 = 1.495, 2 = 1.053, E 2
= 3.69.
The case (i) may be compared with the case J/J3 = 4 in Table 1. For a similar value of
1 , the string is closer to the equator, i.e., the higher is n, the less is the eccentricity of the
circular string.
5. Circular string with two angular momenta: a non-trivial test of AdS/CFT duality
Here we would like to complement the work of Ref. [4], which considered the case
of the folded string solution with two spins in S 5 , by discussing the analogous case of
the circular type two-spin solution and comparing it with the corresponding SYM results
of [3].
In Section 4 of Ref. [3] a set of SYM operators with SU(4) Dynkin labels [J2 , J1 J2 ,
J2 ] was found, whose one-loop anomalous dimensions correspond to solutions of the
Bethe equations with all Bethe roots lying on the imaginary axis. The one-loop anomalous
dimension was determined to be
J2
1
J1 J2
,
f (D),
D
=
= J1 + J2 +
(5.1)
2Jtot
2 Jtot 2(J1 + J2 )
where
f (D) = 1 + 8D 2 + 24D 4 + 96D 6 + 408D 8 + .
(5.2)
Our aim here is to demonstrate that the corresponding dual string configuration is described
by a circular type solution (4.3), (4.4).
To make a comparison with the SYM results straightforward it is convenient to redefine
the variables wi which obeyed w1 w2 w3 as follows: w1 w3 , w3 w1 . Then the
expressions (4.8), (4.9) and (4.10) transform into

E(t)
E(t)
w1
w2
1
,
J2 =
t 1+
J1 =
(5.3)
t
K(t)
t
K(t)
and
E 2 = w22 +
2
w12
,
t
(5.4)
15 In general, one expects that for a string with fixed shape (which requires fixed , ) the energy grows
1 2
as E 2 n2 . . . . In the present case, different winding numbers give different angles 1 , 2 , so this simple n2
dependence does not appear.
36
2 /w 2 < 1. These equations are supplemented by the

where the parameter t = w12
13
periodicity condition (4.5) with n = 1, which now reads as

2
w12
.
K(t) =
(5.5)
2
t
Note that in spherical coordinates ( , ) the equations of motion describing our circular
2 sin 2 = 0. It is worthwhile to mention that
type string are = /2 and + (1/2)w12
after the obvious rescaling (1/2) the last equation describes a motion of the plane
2 (1/t sin2 ),
pendulum in the gravitational field. Integrating it once, we get 2 = w12
where t appears as an integration constant. If t > 1, then 1/t = sin2 0 and this solution
describes the folded string extending from 0 to 0 . For t < 1, when there is no turning
point where = 0, the solution will describe a circular string extending all the way around
the equator = /2 with from 0 to 2 . In the limit t 0, this solution will approach
the circular type string with J1 = J2 . Thus the parameter t provides an interpolation
between the circular and the folded string configurations. Clearly, the rotatory motion of
the pendulum requires more total energy than the oscillatory phase and this explains why
the energy of the circular string is bigger than that of the folded one.
We want to find the leading correction to the energy (5.4) in the limit of the large total
spin Jtot for a generic circular type configuration with J1 = J2 and to determine thereby
the one-loop anomalous dimension
of the corresponding SYM operators. To this end we
assume the following large Jtot = Jtot expansions
2
w12 = Jtot
+ 1 ,
2
w22 = Jtot
+ 2 ,
t t (D),
a = a (D),
where the variable D is defined in (5.1). In the limit of large Jtot the energy (5.4) is

1
1
2 + (1 2 ) .
E = Jtot +
2Jtot
t
(5.6)
(5.7)
Expanding Eqs. (5.3) and (5.5) it is easy to determine the 1,2 as the functions of the
variable t:

4
K(t) (t 1)K(t) + E(t) ,
2

4
2 = 2 K(t) E(t) K(t) ,
1 =
(5.8)
while t is determined as the function of D from the following equation:

D=
1 1
E(t)

.
t
2 tK(t)
(5.9)
The last equation can be easily solved by the power series ansatz and one obtains
t (D) = 16D 128D 2 + 736D 3 3584D 4 + 15808D 5
65024D 6 + 253888D 7 951296D 8 + 3446272D 9 + .
(5.10)
37
Inserting now this expansion in Eq. (5.8) we first determine 1 and 2 and then the energy
(5.7). The final result reads as

4
K(t)E(t)
E = Jtot +
(5.11)
2Jtot 2
and one obtains the following well-defined perturbative expansion of the energy
E = Jtot +

1 + 8D 2 + 24D 4 + 96D 6 + 408D 8 + + .
2Jtot
(5.12)
Quite remarkably, as in the folded string case in [4], this exactly reproduces the full series
(5.1), (5.2) for the anomalous dimension of the set of operators corresponding to imaginary
Bethe roots in [3].16
For D = 0 (i.e., for J1 = J2 ), one recovers the expression for the correction to the energy
(anomalous dimension) /(2Jtot ) corresponding to the circular string solution discussed
in [1]. The opposite, BPS-like limit D 1/2, i.e., J2 0, is ill-defined. This is in
contrast to the folded string case, where J2 0 leads to shrinking of the folded string
to the BPS particle (the corresponding dual operator with = J1 transforms in the BPS
irrep [0, J1 , 0]). Obviously the circular string winding around the equator of S 2 cannot be
contracted to a point particle.
We conclude this section by emphasizing that the Neumann dynamical system and its
string-like solutions encode the information about the full (all-loop) anomalous dimension
of the corresponding gauge theory operators. In fact, our treatment above can be extended
in a straightforward manner to determine the subleading terms in the energy which should
correspond to the two- (and higher-)loop corrections to the anomalous dimensions of the
dual SYM operators. This might help to understand, from the field-theoretic point of view,
the integrability of the quantum SYM theory beyond the one-loop level.
6. Strings rotating in AdS5

In this section we briefly discuss multi-spin strings rotating in AdS5 [1], emphasizing
the similarity with the case of rotation in S 5 .
We use the same rotation ansatz as in the S 5 case (2.9) written in terms of the embedding
coordinates (2.8) as follows
Y1 + iY2 = y1 ( )eiw1 ,
Y3 + iY4 = y2 ( )eiw2 ,
Y5 + iY0 = y3 ( )eiw3 .
(6.1)
The real radial functions yi are independent of time and should, as a consequence of
MN YM YN = 1, lie on a two-dimensional hyperboloid:
ij yi yj y12 y22 + y32 = 1.
(6.2)
16 The relative position of the parameters (w , b ) describing the moduli space of string solitons is therefore
i a
reflected on the field theory side in the way the Bethe roots are distributed on the complex plane. Folded and
circular type strings correspond to double contour roots and imaginary roots, respectively.
38
Note that because of the definition of Y5 + iY0 in (2.8), t = w3 , and, therefore, the 3rd
frequency is now w3 = . Just as it was in the S 5 case, the three O(2, 4) spins
S1 = S12 ,
S2 = S34 ,
S3 E = S56
forming a Cartan subalgebra of SO(2, 4) are given by

Si =
2
wi
d 2
yi ( ) Si ,
2
(6.3)
and satisfy, because of (6.2), the following relation

S1
S2
E
= 1, = w3 .
(6.4)
w1 w2
The effective 1d mechanical system describing this class of rotating solutions in AdS5 has
the following Lagrangian (obtained from (2.6) after an overall sign change, LAdS L)
1
1
ij yi yj 1).
L = ij yi yj wi2 yi yj + (
(6.5)
2
2
Comparing Eqs. (6.5), (6.3) and (6.4) with the corresponding Eqs. (2.14), (2.10) and (2.37)
for the S 5 case, we see that the relation to the S 5 case is through the analytic continuation17
x1 iy1 ,
x2 iy2 .
(6.6)
The ellipsoidal coordinates a defined as in (2.23) now provide the parametrization of the
two-sheeted hyperboloid y32 y12 y22 = 1. Taking into account the analytical continuation,
we get the following relations between yi and a
y12 =
y32 =
(w12 1 )(w12 2 )
2 w2
w21
13
(w32 1 )(w32
2 w2
w31
32
2 )
y22 =
(w22 1 )(w22 2 )
2 w2
w21
32
It is not difficult to check that the equations of motion for a and the Hamiltonian of the
effective 1d system have the same form (2.30), (2.33) as in the S 5 case. In terms of a the
only difference between the S 5 and AdS5 cases is in the range of their allowed values.
Another essential feature of the string motion in AdS5 space is that in this case the
Hamiltonian (2.33) coincides with the r.h.s. of the only non-trivial Virasoro constraint
Y ), and, therefore, has to vanish (recall that = w )
MN (YM YN + YM
3
N
2 + w12 + w22 b1 b2 = 0.
(6.7)
This constraint, together with (6.3) and periodicity conditions will lead to the expression
for the energy as a function of the two (in general, unequal) SO(2, 4) spins,
E = E(S1 , S2 ).
17 The angular coordinates in (2.2), (2.3) are related by i.
39
The previously known examples of the one-spin folded string solution (with w1 = 0,
w2 = 0) [6,12] and the circular string solution (with w1 = w2 = 0) [1] suggest that the
frequencies wi should be chosen as
= w3 w1 w2 .
(6.8)
We shall also assume for definiteness that

1 < 2 ,
b1 < b2 .
(6.9)
Then one can show that

w12 1 w22 2 ,
1 b 1 2 b 2 .
(6.10)
Just as it was in the S 5 case, a folded string solution exists if b1 and b2 belong to the same
wi2 intervals as 1 and 2 , respectively, i.e., if
w12 1 b1 w22 2 b2 .
(6.11)
A two-spin folded string solution exists only if the string is bent. The periodicity conditions
for a bent folded string are similar to the ones in the S 5 case. The relations between spins
and the energy also have the same form as the S 5 case relations (2.37)(2.39), with the
replacement
J1 S1 ,
J2 S2 ,
J3 E.
It would be interesting to analyze the resulting system of equations in the limit of large
spins S1 , S2 . This limit seems to correspond to a long folded string with a large bend.
Two-spin string solutions of circular type exist only if both b1 and b2 belong to the same
interval as 2 , i.e.,
w12 1 w22 b1 2 b2 .
(6.12)
Again, there are two simple cases in which the solutions can be analyzed in detail:
(i) w1 = w2 , and (ii) b1 = b2 . The simplest round-circle string solution found in [1]
corresponds to the case w1 = w2 and b1 = b2 .
7. Concluding remarks
In this paper we have developed a unified treatment of the rotating string solutions
in AdS5 S 5 based on the integrability of the Neumann dynamical system. We have
shown that generic multi-spin solutions are naturally associated to the hyperelliptic genus
2 Riemann surface. The shape of the closed string at fixed time may be of a folded (straight
or bent) string and of a circular type. This depends on the two winding numbers n1 , n2 and
on the relative values of the parameters describing the solution moduli space.
We have also studied perturbation theory around the simplest two-spin solutions in the
direction of the non-zero third spin component and derived a leading correction to the
energy in this case. This enabled us to make an explicit prediction for the 1-loop anomalous
dimensions of the corresponding gauge-invariant operators in N = 4 SYM theory (see
40
(3.54)). One may hope that a simple picture of the constrained harmonic oscillator motion
linearizing on the Liouville torus may be also uncovered in the equations governing the
algebraic Bethe ansatz for gauge-invariant operators on the SYM side (cf. [3,21]).
It would be interesting also to see if and how the hyperellipticity of the general
three-spin solutions is related to the more complicated nature of the dual CFT operators.
Indeed, in comparison to the elliptic two-spin solutions where the dual operators
(tr[(1 + i2 )J1 (3 + i4 )J2 ] + ) are made out of hypermultiplets (in the N = 2
language), the operators dual to genuine three-spin hyperelliptic string solitons will also
mix (beyond the one-loop level) with the operators from the field-strength multiplet.
As we have shown above, a rotating string in the AdS5 space is described by a noncompact version of the Neumann dynamical system which has a simple interpretation in
terms of the harmonic oscillator constrained to move on the 2d hyperboloid. It is desirable
to study integrability of this system in more detail, and, in particular, to determine the
energy as a function of spins E = E(S1 , S2 ) for long string configurations. A most
natural interpretation of the corresponding dual SYM operators will be in terms of nonlocal Wilson loops [23], and one may hope to shed some light on their integrable structures
(see, e.g., [24] and references therein). One specific open problem is to study a folded bent
string solution with two equal spins S1 = S2 which should have lower energy than the
circular solution found in [1].
It would be very important to try to go beyond semiclassics and identify the string
states and field-theoretic operators for small values of the spins and energy (dimension).
One possible direction could be to develop a string-bit model type approximation of
continuous string world-sheet (cf. [25]). Indeed, in the CFT an elementary field contributes
a quantum of dimension and spin to a composite operator. Long composite operators
are then viewed as made of many quanta, and a wave approximation corresponds to
considering excitations of the continuous string world-sheet. To understand the AdS/CFT
correspondence beyond semiclassical approximation one needs to find an analogue of
quantum of energy and spin on the string side.
A related problem is to see if and how the integrability of the classical bosonic
SO(2, 4) SO(6) sigma model can be extended to the AdS5 S 5 GreenSchwarz
superstring sigma model (for a recent work in this direction addressing integrability of
the classical supercoset [26] sigma model see [27]).
Acknowledgements
We are grateful to N. Beisert, M. Staudacher and S. Theisen for discussions of some
related issues. The work of G.A. was supported in part by the European Commission RTN
Programme HPRN-CT-2000-00131 and by RFBI Grant N02-01-00695. The work of S.F.
and A.T. was supported by the DOE Grant DE-FG02-91ER40690. The work of A.T. was
also supported in part by the PPARC SPG 00613 and INTAS 99-1590 Grants and the
Royal Society Wolfson award. The work of J.R. is supported in part by the European
Communitys Human potential Programme under the Contract HPRN-CT-2000-00131,
and by MCYT FPA, 2001-3598 and CIRIT GC 2001SGR-00065. J.R. would like to thank
41
Imperial College for hospitality during the course of this work and acknowledges also the
support of PPARC SPG 00613 Grant.
Appendix A. General solution of the Neumann system in terms of -functions

The general solution of the n = 3 Neumann system can be formulated in terms of the
theta functions defined on the Jacobian of the hyperelliptic genus 2 Riemann surface [16].
Introduce the theta-functions with characteristics :

exp 2iz(n + ) + i (n + )2 + 2i (n + ) ,
[](z) =
(A.1)
nZ 2
where z = (z1 , z2 ) and n = (n1 , n2 ). Here is a period 2 2 symmetric matrix with the
positive definite imaginary part. The characteristic is a 2 2 matrix = ( , ) made
of two columns and .
The two normalized Abelian differentials of the 1st kind (here a, b = 1, 2)

(A.2)
b = ab ,
b = ab
Aa
Ba
can be written in the form

a = pa 1 + ea 2 ,
(A.3)
where
d
d
,
2 =
,
s
s
and s( ) is determined by (2.36). The differentials a , a = 1, 2, satisfy

Aab = b ,
Bab = b
1 =
Aa
(A.4)
(A.5)
Ba
with ab = (BA1 )ab . The normalization condition (A.2) relates the coefficients ea , pa to
Aab , Bab (see also below).
In particular, when , the leading coefficient of the differentials a are ea which
are the frequencies of oscillation on the Jacobian. The solution in [16] reads (i = 1, 2, 3)
xi2 ( ) =
Here
1
1 =
2 [2i1 ](z0 + 12 ie ) 2[2i1 ](0)

2 [0](z0 + 12 ie ) 2[0](0)

0
,
0

3 =
1
2
1
2
(A.6)

,
5 =
0
0
1
2
1
2

(A.7)
are the half-periods and z0 is a constant vector (the initial condition) which, without loss
of generality, we choose to be zero. The specific coefficient 1/2 in front of e in Eq. (A.6) is
related to the fact that the canonical variable a conjugate to ba obeys the linear equation
42
of motion
1
a = {H, a } = ,
2
(A.8)
where H is given by Eq.

(2.33).
Finally, the identity 3k=1 xk2 = 1 is due to the Frobenius formula (here g = 2)
g+1 2

[2k1 ](z) 2 [2k1 ](0)
k=1
2 [0](z) 2[0](0)
= 1.
(A.9)
We do not know the constant vector ea explicitly, but we can impose the periodicity
condition: xi ( + 2) = xi ( ). In general, solutions (A.6) are not periodic functions of
and achieve periodicity we have to require
ea = ab mb + na ,
(A.10)
where ma and na are vectors with integer components. In addition, if we are interested in
the real solutions, then the motion occurs on the real connected component of the Jacobian
identified with the Liouville torus. In this case ea is real and we obtain the condition
ea = na .
(A.11)
Since e1 = A21/(det A), e2 = A11 /(det A), the latter condition reduces to
n1 A11 + n2 A21 = 0,
n1 A12 + n2 A22 = .
(A.12)
Written in the integral form these conditions are

n1
A1
d
+ n2
s
d
= 0,
s
(A.13)
A2
and

n1
A1
d
+ n2
s
d
= .
s
(A.14)
A2
In this way we have demonstrated that the general periodic solitons are characterized by
two integers (n1 , n2 ).
Thus, when goes from 0 to 2 the image of the string in the Jacobian winds around
the real circles A1 and A2 with winding numbers n1 and n2 , respectively. The different
periodicity conditions discussed in the main text can be obtained by picking in Eqs. (A.13)
and (A.14) the concrete values for (n1 , n2 ) and specifying the cycles in terms of the branch
points (wi , ba ). It would be interesting to see how the various elliptic solutions discussed
in the paper can be directly obtained from (A.6) by degenerating the period matrix .
43
Appendix B. Basic integrals

Here we recall the definitions of K(t), E(t) and F(u, t) that are the complete elliptic
integrals of the first, the second and the third kind, respectively:
1
K(t) =

0
dx
(1 x 2 )(1 tx 2 )

1 1
,
;
1;
t
,
F
2 1
2
2 2

1
1 tx 2
1 1
E(t) =
dx = 2 F1 , ; 1; t ,
2
2 2
1 x2
(B.1)
(B.2)
1
F(u, t) =
0
(1 ux 2 )
dx
(1 x 2 )(1 tx 2 )
(B.3)
where for K(t) and E(t) we also provided their expressions in terms of the Gauss
hypergeometric function.
The incomplete elliptic integral of the third kind is given by
sin

F(u, , t) =
dx
(1 ux 2 ) (1 x 2 )(1 tx 2 )
(B.4)
In Section 3.3 we used the following integrals
1
I1 =
2 w 2 (b w 2 )
w21
32 2
2
dz
,
z(1 1 z)(1 t1 1 z)(1 u1 1 z)(1 v1 1 z)
(B.5)
2
I2 =
2 w 2 (w 2 b )
w31
1
32
3
1
0
dz
,
z(1 2 z)(1 t2 2 z)(1 u2 2 z)(1 v2 2 z)
(B.6)
as well as
2
w2
I10 =
b1
=
dz
P (z)
1
2 w 2 (w 2 D)
w21
32
32
1
dz
z(1 z)(1 t1 z)(1 u1 z)(1 v1 z)
44
1
1
2 w2
w21
32
dz
z(1 z)(1 t1 z)(1 u1 z)
1
+
2 w4
2 w21
32
dz
,
z(1 z)(1 t1 z) (1 u1 z)2
(B.7)
w3
I20 =
b2
=
dz
P (z)
1
1
2 w 2 (w 2
w31
32
3
b1 )
2 w 2 (w 2 b )
w31
1
32
3
dz
z(1 z)(1 t2 z)(1 u2 z)(1 v2 z)

2 + w2 ) + w2 w2 )
D((w32 b1 )(w31
32
31 32
2 w 2 (w 2 b ))3/2
4(w31
1
32
3
(B.8)
w2
I11 =
b1
=
z dz
P (z)
1
2 w 2 (w 2 D)
w21
32
32
[w22 (w22 b1 )z] dz

z(1 z)(1 t1 z)(1 u1 z)(1 v1 z)
= w22 I10 I11 .
(B.9)
Here D = w32 b2 and
I11 =
1
w22 b1
2 w 2 (w 2 D)
w21
32
32
u1

2
w21
1
0
Du1
dz z
z(1 z)(1 t1 z)(1 u1 z)(1 v1 z)
dz z
z(1 z)(1 t1 z) (1 u1 z)

2 w2
2 w21
32
1
dz z
.
z(1 z)(1 t1 z) (1 u1 z)2
(B.10)
45
Also,
2
w3
I21 =
b2
=
dz z
P (z)
1
1
2 w 2 (w 2 b )
w31
1
32
3
= w32 I20
dz (w32 Dz)
z(1 z)(1 t2 z)(1 u2 z)(1 v2 z)
I21 ,
(B.11)
where
I21 =
1
D
2 w 2 (w 2
w31
32
3
b1 )
dz z
z(1 z)(1 t2 z)(1 u2 z)(1 v2 z)
.

2 w 2 (w 2 b )
2 w31
1
32
3
(B.12)
(B.13)
Similarly, we define
2
w2
I12 =
b1
=
dz z2
P (z)
1
1
2 w 2 (w 2 D)
w21
32
32
dz (w22 (w22 b1 )z)2

z(1 z)(1 t1 z)(1 u1 z)(1 v1 z)
= w24 I10 2w22 I11 + I12 ,
(B.14)
where
I12 =
1
(w22 b1 )2
2 w 2 (w 2 D)
w21
32
32
u2 w 2
1 32
2
w21
1
0
Du2
+ 1
2
2 w21
dz z2
z(1 z)(1 t1 z)(1 u1 z)(1 v1 z)
dz z2
z(1 z)(1 t1 z) (1 u1 z)
1
dz z2
,
z(1 z)(1 t1 z) (1 u1 z)2
(B.15)
46
and
2
w3
I22 =
b2
=
dz z2
P (z)
1
1
2 w 2 (w 2 b )
w31
1
32
3
= w34 I20
I22 =
2w32 I21
dz (w32 Dz)2
z(1 z)(1 t2 z)(1 u2 z)(1 v2 z)
+ I22 ,
D2
2 w 2 (w 2 b )
w31
1
32
3
1
0
(B.16)
dz z2
z(1 z)(1 t2 z)(1 u2 z)(1 v2 z)
3D 2
0.

2 w 2 (w 2 b )
8 w31
1
32
3
Then the periodicity conditions and the integral
(B.17)

d (1 + 2 ) in Section 3.2 take the form
2I20 = I10 ,
(B.18)
2
I10 2I21 + I11 = ,
2I21 I11 = w32
2
2I21 = w32
I10
or
+ I11 ,
(B.19)
/2
1
1
1 4
1
2
d (1 + 2 ) = I22 I12 = w32 w32
I0 w32
I11 + (2I22 I12 ).
2
2
2
2
0
(B.20)
We also have the following periodicity condition
2
w2
=
b1
d1 [2 (1 ) 1 ]
P (1 )
2
= w32
I0 + I11

2 w 2 (w 2 D)
w21
32
32
1

2
w21
1
0
1
0
d1 2 (1 )
1 (1 1 )(1 t1 1 )(1 u1 1 )(1 v1 1 )
2
dz
K(t1 ).
=
z(1 z)(1 t1 z)
2
w21
(B.21)
47
Appendix C. Vanishing of non-Cartan components of angular momentum

The SO(6) momentum components JMN written in terms of the 6 embedding
coordinates XM in (2.7) are
d
(XM XN XN XM ).
=
2
2
JMN
(C.1)
They are conserved in time by virtue of the SO(6) symmetry of the Lagrangian (2.5).
The diagonal components J1 = J12 , J2 = J34 and J3 = J56 are clearly non-zero for the
rotating string ansatz (2.9) (as long as all wi = 0), being proportional to an integral (2.10)
of a positive definite quantity xi2 .
The question we would like to address is what are the conditions for the vanishing of all
other components of JMN . First, assume that w1 = w2 . Then one has, in particular,

d
J13 = w1 sin(w1 ) cos(w2 ) w2 sin(w2 ) cos(w1 )
x1 ( )x2 ( ).
2
2
2
(C.2)
Since J13 must be time-independent on shell, it follows that 0 (d/(2))x1( )x2 ( )

must vanish on the solution of the string equations of motion. Similarly, J14 , J23 , J24 must
also vanish, since they are proportional to the same integral over . One reaches analogous
conclusions in the cases w1 = w3 and w2 = w3 .
The solution of Section 3 where the string is folded in both and is possible only if
all wi are different, and so all extra JMN components are necessarily zero there.
The only case that may in principle lead to non-zero values for the non-Cartan
components of JMN is when some two of the three frequencies happen to be equal. If
w1 = w2 , then J13 and J24 are automatically zero. For the components J23 and J14 one
finds (using (2.7))

d
d
J23 = J14 = w1
x1 ( )x2 ( ) = w1
sin2 cos sin ,
2
2
2
(C.3)
which may potentially give a non-zero result. Similarly, if w1 = w3 ,

d
d
x1 ( )x3 ( ) = w1
sin cos cos ,
J25 = J16 = w1
2
2
2
(C.4)
and if w2 = w3

d
d
x2 ( )x3 ( ) = w2
sin cos sin .
J45 = J36 = w2
2
2
2
(C.5)
48
Let us study the values of these components in some special cases. First, let us consider
a string configuration which is folded in : when goes from 0 to let vary between
1 and 2 and between 0 to , and when goes from to 2 let vary from 2
to 1 and from to 2 . Then the integral (C.3) vanishes because the contribution
of the integration region 0 < < /2 cancels against the contribution of the integration
region from 3/2 < < 2 , while the contribution of /2 < < cancels against that
of < < 3/2. In the same way, the integral (C.5) also gives zero as the integration
region 0 < < cancels with the region < < 2 , where sin has the opposite
sign. The only non-obvious case is the vanishing of the integral (C.4). If there is an
extra symmetry such as, e.g., ( ) = /2 ( ) (as in the cases where the string
extends from the equator to both sides in a symmetrical way), then the integral vanishes
by similar symmetry arguments. If ( ) is not directly related to ( ) in a simple
way, this integral needs to be performed explicitly. For the circular solution considered
in Section 4, with = c/sin2 , J25 and thus the integral (C.4) must vanish because there
w1 = w3 . We have checked independently that (C.4) is indeed zero in this case.
Appendix D. Straight folded string
Here we describe the two-spin solution realized by the folded string without bend points.
As will follow from our consideration, such solution is rigid in a sense that it does not allow
a deformation in the direction of the non-zero third spin component.
We assume the same range of parameters as in Section 3, i.e.,
w12 < b1 1 w22 < b2 2 w32
(D.1)
and perform the same change of variables (3.14) with the subsequent two-spin limit
b2 w32 , so that Eq. (2.30) will be reduced to the system (3.29), (3.30). However, this time
we assume the existence of the two turning points, at = 0 and at = , and no bend
points. This implies that both derivatives a are negative on the interval 0 /2 and
positive for /2 . The periodicity conditions describing this situation are
I1 (1 ) = I2 (2 ),
(D.2)
I21 I11 = ,
(D.3)
where I1,2 are defined in (3.34) while I11 and I21 are given by (3.40). Considerations
similar to those in Section 3.3. allow one to determine 2 as function of 1 :

2 F(u1 , arcsin 1 , t1 )
.
2 (1 ) = sin
(D.4)
2
F(u1 , t1 )
This time 2 is the monotonic function of 1 on the interval [0, 1]. In particular, 2 (0) = 0
and 2 (1) = 1.
In addition, the parameters u1 , t1 should obey the following two equations

u1
,
F(u1 , t1 ) =
(D.5)
2 (1 u1 )(u1 t1 )

2 .
K(t1 ) =
(D.6)
w21
2
49
Comparison with (4.15) and (4.17) shows that the image of our solution on the Liouville
torus is characterized by the winding numbers n1 = n2 = 1.
The relations between wi and Ji retain the same form (3.46) and (3.47) as they do not
depend on 2 . Solution of Eq. (3.42), which is the same as (D.6), and (3.46), (3.47) for
w1 , w2 and t1 , is given by Eq. (3.49). By using this solution we can now infer the value of
u1 from Eq. (D.5). One finds u1 = , which implies that w2 = w3 .
It turns out that the perturbation theory around u1 = is ill-defined as this is essential
singularity. Moreover, the function 2 (1 ) ceases to be regular in the limit u1 as 2
(1 ) = 1 for 0 < 1 1 and 2 (1 ) = 0 for 1 = 0. Therefore, we conclude that our folded
string is rigid in the sense that it cannot be bent to acquire a small amount of the third
spin component J3 .
It should be emphasized that the variable u1 does not enter either the relations between
wi and Ji or the expression (3.48) for the energy. This parameter and the function 2 play
only an auxiliary rle for the two-spin solutions. However, they both arise in degeneration
of a certain hyperelliptic solution and, therefore, they point out a direction in which an
elliptic two-spin solution can(not) be deformed.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
S. Frolov, A.A. Tseytlin, Multi-spin string solutions in AdS5 S 5 , hep-th/0304255.

S. Frolov, A.A. Tseytlin, Quantizing three-spin string solution in AdS5 S 5 , hep-th/0306130.
N. Beisert, J.A. Minahan, M. Staudacher, K. Zarembo, Stringing spins and spinning strings, hep-th/0306139.
S. Frolov, A.A. Tseytlin, Rotating string solutions: AdS/CFT duality in non-supersymmetric sectors, hepth/0306143.
D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-Yang
Mills, JHEP 0204 (2002) 013, hep-th/0202021.
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, Nucl.
Phys. B 636 (2002) 99, hep-th/0204051.
S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S 5 , JHEP 0206 (2002)
007, hep-th/0204226.
K. Pohlmeyer, Integrable Hamiltonian systems and interactions through quadratic constraints, Commun.
Math. Phys. 46 (1976) 207;
M. Luscher, K. Pohlmeyer, Scattering of massless lumps and nonlocal charges in the two-dimensional
classical nonlinear sigma model, Nucl. Phys. B 137 (1978) 46;
H. Eichenherr, M. Forger, On the dual symmetry of the nonlinear sigma models, Nucl. Phys. B 155 (1979)
381.
B.M. Barbashov, V.V. Nesterenko, Relativistic string model in a spacetime of a constant curvature,
Commun. Math. Phys. 78 (1981) 499;
B.M. Barbashov, V.V. Nesterenko, Introduction to the Relativistic String Theory, World Scientific,
Singapore, 1990, p. 249.
H.J. De Vega, N. Sanchez, Exact integrability of strings in d-dimensional de Sitter spacetime, Phys. Rev.
D 47 (1993) 3394;
M. Zyskin, On gauge fields: strings duality as an integrable system, hep-th/0004105.
G. Mandal, N.V. Suryanarayana, S.R. Wadia, Aspects of semiclassical strings in AdS(5), Phys. Lett. B 543
(2002) 81, hep-th/0206103.
H.J. de Vega, I.L. Egusquiza, Planetoid string solutions in 3 + 1 axisymmetric spacetimes, Phys. Rev. D 54
(1996) 7513, hep-th/9607056.
C. Neumann, De problemate quodam mechanico, quod ad primam integralium ultraellipticorum classem
revocatur, J. Reine Angew. Math. 56 (1859) 4663.
50
[14] A.M. Perelomov, Integrable Systems of Classical Mechanics and Lie Algebras, Springer-Verlag, Berlin,
1990.
[15] O. Babelon, M. Talon, Separation of variables for the classical and quantum Neumann model, Nucl. Phys.
B 379 (1992) 321, hep-th/9201035.
[16] D. Mumford, Tata Lecture Notes on Theta Functions, Springer-Verlag, Berlin, 1994.
[17] A.A. Tseytlin, Semiclassical quantization of superstrings: AdS5 S 5 and beyond, Int. J. Mod. Phys. A 18
(2003) 981, hep-th/0209116.
[18] J.G. Russo, Anomalous dimensions in gauge theories from rotating strings in AdS5 S 5 , JHEP 0206 (2002)
038, hep-th/0205244.
[19] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super-YangMills, JHEP 0303 (2003) 013, hepth/0212208.
[20] N. Beisert, C. Kristjansen, M. Staudacher, The dilatation operator of N = 4 super-YangMills theory, hepth/0303060;
N. Beisert, The complete one-loop dilatation operator of N = 4 super-YangMills theory, hep-th/0307015.
[21] N. Beisert, M. Staudacher, The N = 4 SYM integrable super-spin chain, hep-th/0307042.
[22] K. Uhlenbeck, Equivariant harmonic maps into spheres, in: Proceedings of Tulane Conference on Harmonic
Maps.
[23] M. Kruczenski, A note on twist two operators in N = 4 SYM and Wilson loops in Minkowski signature,
JHEP 0212 (2002) 024, hep-th/0210115;
Y. Makeenko, Light-cone Wilson loops and the string/gauge correspondence, JHEP 0301 (2003) 007, hepth/0210256.
[24] A.V. Kotikov, L.N. Lipatov, V.N. Velizhanin, Anomalous dimensions of Wilson operators in N = 4 SYM
theory, Phys. Lett. B 557 (2003) 114, hep-ph/0301021;
A.V. Belitsky, A.S. Gorsky, G.P. Korchemsky, Gauge/string duality for QCD conformal operators, hepth/0304028.
[25] H. Verlinde, Bits, matrices and 1/N , hep-th/0206059;
D. Vaman, H. Verlinde, Bit strings from N = 4 gauge theory, hep-th/0209215;
U. Danielsson, F. Kristiansson, M. Lubcke, K. Zarembo, String bits without doubling, hep-th/0306147.
[26] R.R. Metsaev, A.A. Tseytlin, Type IIB superstring action in AdS5 S 5 background, Nucl. Phys. B 533
(1998) 109, hep-th/9805028.
[27] I. Bena, J. Polchinski, R. Roiban, Hidden symmetries of the AdS5 S 5 superstring, hep-th/0305116.

The BRST cohomology of the N = 2 string

I. Kriz 1
Department of Mathematics, University of Michigan, Ann Arbor, MI 48109-1109, USA
Received 18 July 2003; accepted 27 August 2003
Abstract
In this paper, we prove by rigorous calculation that BRST cohomology and old covariant
quantization give the same spectra for the N = 2-supersymmetric string. This confirms an old
conjecture, but the precise mechanism by which it happens is more subtle than assumed, and may
shed new light on BRST at extended supersymmetry.
PACS: 11.25.Hf; 11.25.Mj; 11.30.Pb; 04.65.+e
1. Introduction
The N = 2 string (see, e.g., [1,2,912]) has 4 real bosons and fermions, coupled with
N = 2 supergravity (1 graviton, 2 gravitinos and a U (1)-gauge field). Old covariant
quantization (OCQ) for the N = 2 string was carried out, e.g., in [1012] and gives a
physical spectrum which has only one state for each momentum on the mass shell. The
BRST complex for this theory was developed in [1,2,9]. However, the cohomology of
that complex was never calculated directly. In this paper, we show that the N = 2 BRST
cohomology gives the same spectrum as OCQ up to ghost degeneracy similarly as is the
case of N = 0 and N = 1 strings, thus confirming a long standing conjecture.
Despite the analogous nature of the results, computing the BRST cohomology of the
N = 2 string is quite different from the N = 0 and N = 1 cases: at N = 0 and N = 1, (as
shown by Polchinski [13, Section 4.4]), the ghost and antighost mode towers are matched
by the BRST differential precisely with one tower of bosons (and fermions for N = 1)
E-mail address: ikriz@math.lsa.umich.edu (I. Kriz).
1 The author was supported by NSF grant DMS 0305853.
doi:10.1016/j.nuclphysb.2003.08.035
52
I. Kriz / Nuclear Physics B 671 (2003) 5166
each, thus cancelling two towers of bosons (and fermions), which gives the lightcone gauge
spectrum.
When we try to argue the same way for the N = 2 string, however, we arrive at a puzzle.
Just as in the N = 0 and N = 1 cases, Virasoro ghosts/antighosts cancel two bosonic
towers, and the two pairs of supersymmetry ghosts/antighosts cancel the four fermionic
towers. But then we are left with the two remaining bosonic towers and the ghost/antighost
pair coming from the U (1)R-symmetry. There is no way analogous to the N = 0 and
N = 1 cases of cancelling the U (1)-ghosts/antighosts with bosons, since the R-symmetry
acts trivially on bosons! This, in fact, originally led the author to erroneously believe
that the BRST cohomology of the N = 2 string was intrinsically different from its OCQ
spectrum, and that the BRST spectrum had more modes.
It turns out, however, that the remaining bosons and U (1)-ghosts do cancel for a
more subtle reason; consequently, the BRST and OCQ spectra of the N = 2 are the
same up to 4 degenerate ghost 0-modes at each chirality, as expected. The cancellation
mechanism involves a complicated interplay of counterterms which certainly does not
seem to have been known before. In fact, the N = 2 BRST differential has 27 terms
(23 after a transformation), and calculating general differentials directly seems almost
impossible (for illustration, just a few explicit examples of lowest weight differentials are
given in Appendix A). In fact, the author knows no way of rigorously computing the N = 2
BRST cohomology without a spectral sequence. Introducing this technique, and carefully
showing how it applies to the N = 2 BRST case, is the main purpose of this paper.
The present paper is organized as follows: in Section 2, we review the superconformal
algebra of constraints KN and introduce a notation which is well suited for what follows. In
Section 3, we write down explicitly the N = 2 BRST complex in our notation. In Section 4,
we calculate, as a warm-up case, the BRST cohomology of the N = 0 and N = 1 strings
(which is essentially only a slight simplification of Polchinskis argument [13, Section
4.4]). Our main result, the calculation of the BRST cohomology of the N = 2 string, is
given in Section 5.
Notation
For our purposes, it is crucial to reconcile mathematical and physical notation. For a
physical notation closest to our purposes, we refer the reader to [2]. We give here a short
dictionary between the mathematical terms of the present paper and the language of [2].
The action constraints of [2, Eqs. (2.3a)(2.3c)] are (super-versions of ) Euler equations
of the corresponding variation problem; they determine the classical solution. In the
quantized case, the constraints determine the energy-stress tensor, or, as one says in
mathematics, the superconformal algebra action. In our case, the superconformal algebra
is called K2 . We review the superconformal algebras KN , but see [5] for a more detailed
mathematical discussion. In [2], the constraints in the present case are denoted by , S i , T
and later, in the same order, L, Gi , T . This notation coincides with the notation used today
(cf. [6]), except there T is denoted by J . We use the notation G0 , G1 , G2 , G{12} (the exact
translation is in the next section).
53
The massive mode generating fields are, in [2], denoted by i,a (bosons) and bi,a
(fermions), i, a = 1, 2. In this paper, D = 4 (real) plays such a special role that we name
these fields by individual letters x, y, z, t (bosons) and , , , (fermions).
The ghost generating fields are, in [2], denoted by , 1 , 2 , and the antighost fields
From our point of view, these fields are a part of a general
are denoted by , 1 , 2 , .
picture, and we call the ghost fields c12 , c1 , c2 , c0 and the antighost fields b12 , b 1 , b 2 , b 0
(in the same order).
2. A review of the N -superconformal Lie algebra KN

We begin with a review of the N -superconformal algebra KN (see [47,11]). As a superLie algebra, KN can be described as the free complex vector super-space on basis GIr
where I {1, . . . , N}, r !I2 mod 1 in the NS sector and r Z in the Ramond sector (!I
is the number of elements I ). It saves some sign conventions if we identify a sequence
I = (i1 < < ik ) with the set |I | = {i1 , . . . , ik }, and extend the definition of GIr to nonincreasing sequences by
Gr (I ) = sign( )GIr
for a permutation on {1, . . . , N}. Now the super-Lie bracket on KN is defined as follows
(we always write [ ] for the Lie bracket, regardless of parity):
For I = (i1 , . . . , ik ), J = (j1 , . . . , j& ), |I | |J | = , I J = (i1 , . . . , ik , j1 , . . . , j& ),
I J
J
Gr , Gs = r(2 &) s(2 k) GIr+s
.
For I , J as above, but with |I | |J | = {ik } = {j1 }, I J = (i1 , . . . , ik1 , j2 , . . . , j& ),
I J
I J
Gr , Gs = Gr+s
.
When !(|I | |J |) > 1,
I J
Gr , Gs = 0.
We will be mostly interested in the case N = 2, where these symbols relate to standard
{i}
{12}
notation [6] by Gr = 2Lr , Gr = Gir , Gr = iJr . However, even there we will see a
benefit to having a systematic approach to the description of the Lie bracket.
The superalgebra KN can be thought of as the finite energy part of the superalgebra
of infinitesimal deformations of N -super-conformal structure on a boundary component
of an N -superconformal surface (cf. [7]). We recall that an N -superconformal surface
is an (1|N)-dimensional complex supermanifold which possesses an atlas consisting of
coordinate patches (z|1 , . . . , N ) satisfying the following equations: consider the N -tuple
of differential operators
D = (D1 , . . . , DN )T ,
where
Dk =
+ k ,
k
z
k = 1, . . . , N.
54
Considering another set of local coordinates (z|1 , . . . N ), we have

N
N

.
j Di j
(Di j )Dj + Di z
Di =
z
j =1
j =1
The condition is that
D = M D,
where
Mij = Di j ,
which leads to the equations
Di z
N

j Di j = 0,
i = 1, . . . , N.
(1)
j =1
To find possible infinitesimal deformations of coordinates, we must consider the linearized

equations (1), which are
Di z = i +
N

j Di j .
(2)
j =1
Solutions of the linearized equations are (by definition) vector fields. One finds that a
(topological) basis of the space of solutions, written in vector field notation, is of the form
GI
n+ |I2| 1
= (|I | 2) zn I
|I
|1
N

+ zn
(1)j Ikj
nzn1
I i
,
z
ikj
i
j =0
j =1
(3)
where (i1 ,...,ik ) = 1 k and, for I = (i1 , . . . , ik ), Ij = (i1 , . . . , ij , . . . , ik ). These choices

of GIr s obey the Lie relations stated above. Here the numbers n which occur vary with the
boundary conditions imposed in the super-directions, although they are always of the form
n Z + I for some numbers I . We shall not need a detailed analysis of the boundary
conditions: in this paper, we will restrict attention to the R and NS sectors described in the
beginning of this section.
3. The N = 2 BRST complex
In this section, we shall describe the BRST complex Hp of the N = 2 free string at
a given momentum p. Here and from now on we shall work in the chiral theory, which
can be interpreted as calculating the spectrum of the free open N = 2 string. In the closed
string case, both chiralities are present, which in our language means simply calculating
the cohomology of a tensor product of two copies of Hp , which follows by the Knneth
theorem.
55
In the chiral case, Hp is a Z/2 Z/2-graded commutative algebra. This means that two
homogeneous elements u, v of bidegrees (d1 , d2 ) and (e1 , e2 ), di , ei Z/2, satisfy
uv = (1)d1 e1 +d2 e2 vu.
(4)
This is the DeligneMorgan sign convention [3]. In fact, Hp , where p is a quadruple of

numbers (X0 , Y0 , Z0 , T0 ), is a free (Z/2 Z/2)-graded commutative algebra on generators
xn , yn , zn , tn
of bidegree (0, 0), n < 0, n Z,
r , m , r m ,
of bidegree (0, 1),
(5)
r 0,
m < 0,
1
in the NS sector and r, m Z in the R sector,
r, m Z +
2
0
12
bn0 , cm
, bn12, cm
1
2
br1 , cm
, br2 , cm
of bidegree (1, 0), n < 0, m 0,
(6)
(7)
of bidegree (1, 1),
m 0,
1
in the NS sector and r, m Z in the R sector.
r, m Z +
(8)
2
The generators (5), (6) are called matter generators and the generators (7), (8) are called
FadeevPopov ghost generators. Additional numerical invariants we need to keep track of
are the weight which on a monomial in the generators (5)(8) is the sum of the negatives
of the subscripts and the ghost number which is 0 on matter generators and 1, 1 on the
bs and cs, respectively.
Now to construct operators Hp Hp , one introduces generating operators
r < 0,
xn , yy , zn , tn , r , r , r , r , bn0 , cn0 , br1 , cr1 , br2 , cr2 , bn12 , cn12 ,
(9)
in the NS sector and r Z in the R sector without

where n Z, r Z +
positivity restrictions. The operator generators whose indices satisfy the (non)-positivity
requirements of (5)(8) act on Hp by multiplication on the left (and thus can be naturally
identified with the corresponding generators of Hp ). The remaining operator generators
act by 0 on the vacuum 1 with the exception of x0 , y0 , z0 , t0 , which act as multiplication
by numbers X0 , Y0 , Z0 , T0 . The action of operator generators on the entire Hp is then
determined by the operator commutation relations
1
2
[xn , xn ] = [yn , yn ] = [zn , zn ] = [tn , tn ] = n,

[r , r ] = [r , r ] = [r , r ] = [r , r ] = 1,
I I
cr , br = 1,
(10)
and all other commutators of the generators (9) which are not permutations of (10) are 0.
Here the bidegrees of operator generators are specified by the same formulas as for the
generators of Hp , and the commutator is defined on elements a, b of bidegrees (d1 , d2 )
56
and (e1 , e2 ) as
[a, b] = ab (1)d1 d2 +e1 e2 ba.
(11)
Physically, a structure of the kind just described naturally arises in quantization.

Mathematically, the construction also has a better explanation, which will be given in
Section 5 below. For now, it is important that the superconformal algebra K2 acts on Hp .
It turns out helpful to write its operators as
GIr = m GIr + gh GIr
(12)
where m stands for matter and gh stands for ghost. The formula for the BRST differential
is then

1
I I
I I
Q = Qm + Qgh , Qm =
(13)
m Gr cr , Qgh =
gh Gr cr .
2
I,r
We have
0
m Gr
I,r
:xn+r xn : + :yn+r yn : + :zn+r zn : + :tn+r tn :
1
m Gr
n:n+r n : n:n+r n : n:n+r n : n:n+r n :,

=
n+r xn + n+r yn + n+r zn + n+r tn ,
(14)
(15)
2
m Gr
n+r xn + n+r yn n+r zn + n+r tn ,
(16)
12
m Gr

n
I
gh Gr
:n+r n : + :n+r n :,

J
.
GIr , bsJ cs
(17)
(18)
J,s
Here : : denotes the operation of normal ordering. A monomial is said to be in normal

order if all its generators with negative subscripts are to the left of all generators with
positive subscripts. Then : : is the operation of bringing a monomial to normal order using
the relations (4), not (10). In (18), the symbol [GIr , bsJ ] is defined by computing [GIr , GJs ]
and replacing Gs by bs in the result. The formula (13) is then a semiinfinite version
of Lie algebra cohomology, although it hides anomaly cancellation and mass-shell shift
cancellation which one has to deal with in the quantum case (in particular, mass-shell
shifts are trivial for the N = 2 string). In the present paper, the only thing we will use is
that the resulting formula is consistent; for reasoning on the criticality of the D = 4 N = 2
string, and mass-shell shift cancellations in this case, we refer the reader to [11].
Now explicitly, (14)(18) give

0
0
0
0
:xn+r xn :cr
+ :yn+r yn :cr
+ :zn+r zn :cr
+ :tn+r tn :cr
Qm =
0
0
0
0
n:n+r n :cr
n:n+r n :cr
n:n+r n :cr
n:n+r n :cr
1
1
1
1
+ n+r xn cr
+ n+r yn cr
+ n+r zn cr
+ n+r tn cr
57
2
2
2
2
n+r xn cr
+ n+r yn cr
n+r zn cr
+ n+r tn cr
12
12
+ :n+r n :cr
+ :n+r n :cr
(19)
and
Qgh =

0
0 0
1
1 0
2
2 0
(r n):br+n
cn
cr : + (r 2n):br+n
cn
cr : + (r 2n):br+n
cn
cr :
1 0 1 1
12 12 0
12 2 1
2
12 1
2n:br+n
cn cr : + :br+n
cn cr : + (r n)br+n
cn cr + br+n
cn
cr
2
1 0 2 2
1
12 2
+ :br+n
cn cr : br+n
cn
cr .
(20)
2
The operator Q satisfies

QQ = 0,
which can be checked by direct calculation. Thus, Hp is a cochain complex. The purpose of
this paper is to calculate its cohomology, at least with some mild genericity restrictions
on p. We will find it advantageous to introduce the following base change in the matter
operators:
1
xn = (xn + iyn ),
2
1
zn = (zn + itn ),
2
1
n = (n + in ),
2
1
n = (n + in ),
2
1
yn = (xn iyn ),
2
1
tn = (zn itn ),
2
1
n = (n in ),
2
1
n = (n in ).
2
(21)
This changes (12) to

[xn , yn
] = [zn , tn
] = n,

[n , n
] = 1.
(22)
Again, all commutators of the new matter operator generators which are not permutations
of (22) vanish. This transforms (19) to

0

0
Qm =
2:xn+r
yn
:cr
+ 2:zn+r
tn
:cr

0

0
(2n + r):n+r
n
:cr
(2n + r):n+r
n
:cr

1

1

1

1
+ n+r
yn
cr
+ n+r
xn
cr
+ n+r
tn
cr
+ n+r
zn
cr

2

2

2

2
+ in+r
yn
cr
in+r
xn
cr
+ in+r
tn
cr
in+r
zn
cr

12

12
+ i:n+r
n
:cr
+ i:n+r
n
:cr
.
This completes the definitions we need.
(23)
58
4. A warm-up case: the N = 0 and N = 1 string BRST spectra

In this section, we will introduce our main method, and will apply it to a simpler case:
the N = 0 and N = 1 critical free strings. In these cases, the BRST cohomology has
been calculated by Polchinski [13, Section 4.4]. The BRST complexes are as follows:
In the N = 0 case, only K0 acts. Then Hp has a set of 26 sequences of generators
(x1 )n , . . . , (x26)n , n Z, n < 0 and
Qm =
26

0
:(xi )n+r (xi )n :cr
.
i=1 n
The definition of Qgh changes by omitting all summands containing brI or crI with I = ,
and adding the mass shell shift term
2ac00
(24)
with a = 1 (needed to get QQ = 0). To keep our notation, we shall put xn = (x25 )n ,
yn = (x26 )n , and define xn , yn by (21).
In the N = 1 case, K1 acts. Then Hp has a set of 10 sequences of bosonic generators
(x1 )n , . . . , (x10)n , n < 0, n Z, and 10 fermionic generators (1 )r , . . . , (10 )r , r < 0,
r Z + 12 in the NS sector and r Z in the R sector, and one has
Qm =
10

0
0
1
:(xi )n+r (xi )n :cr
n:(i )n+r (i )n :cr
+ (i )n+r (xi )n cr
.
i=1 n
The definition of Qgh is got from (20) by omitting all summands containing brI or crI
where I = , {1} and adding the term (24) where a = 1/2 in the NS sector and a = 0
in the R sector. Again, to keep our notation, we put xn = (x9 )n , yn = (x10 )n , n = (9 )n ,
n = (10 )n and define xn , yn , n , n by (21).
Now to calculate the BRST cohomology, we use a spectral sequence (see [8] for a
general reference). To get a spectral sequence, we introduce a decreasing filtration F i on
Hp (i.e., F i Hp F i+1 Hp ). This is done by assigning a filtration degree to each generating
operator as follows:
|xn | = 1
|yn | = 1
0
b = 1,
n
for n = 0,
0
c = 1
n
for n = 0,
for all n,
and additionally in the N = 1 case:

for all n,
|n | = cn1 = 1
1

|n | = bn = 1 for all n.
The other generating operators (xi )n , n 24 (N = 0 case) and (xi )n , (i )n , n 8 (N = 1
case) have filtration degrees 0.
Now the filtration degree of a monomial in (xi )n , n 24, xn , yn (N = 0 case) and
(xi )n , (i )n , i 8, xn , yn , n , n (N = 1 case) is the sum of filtration degrees of its factors,
59
and the filtration degree of a linear combination of such monomials is the minimum of
the filtration degrees of its summands. F i Hp is the submodule of all elements of filtration
degree i.
The main point now is that since the summands of Q have filtration degree 0, and
also the relations (22) have filtration degree 0, so Q is compatible with filtration in
the sense that QF i Hp F i Hp . Now the E0 -term of the spectral sequence is E0 Hp , the
associated graded object of Hp , which is isomorphic to Hp . The differential d0 is given by
the summands of Q of filtration degree 0, which are

0
1
Q0 =
(25)
2y0 xr cr
+
x0 r cr
r=0
(omit the second summand in the N = 0 case). From this it follows that, assuming
x0 , y0 = 0, up to scalar multiple we have
0

xr
,
d0 : br
1
d0 : br
(r = 0),

r

0
d0 : yr
cr
,

d0 : r
(r = 0),
1
cr
(26)
(omit the last two differentials for the N = 0 case), and so d0 is a differentiation of algebras,
i.e.
d0 (uv) = d0 (u)v + (1)d ud0 (v),
where d is the cohomological degree of u. But (26) is the standard Koszul complex
differential, and its cohomology cancels all the generators (26). Therefore,

E1 Hp = Sym (xi )n , i = 1, . . . , 8 (i )n , i = 1, . . . , 8 c00
(27)
in the N = 1 case and

E1 Hp = Sym (xi )n , i = 1, . . . , 24 c00
(28)
in the N = 0 case (same restrictions on n as above).

Now the differential d1 on E1 is given by terms of Q of filtration degree 1, which are

2:(xi )n (xi )n :c00
Q1 = 2x0 y0 c00 2ac00 +
iD2 n
0
1
2n:(i )n (i )n :c00 2n:bn0 cn
:c00 2n:bn1 cn
:c00 ,
where D = 10 for N = 1, D = 26 for N = 0, omit the and

effect of (29) is multiplication by

D

2
xi c00
2w 2a +
b1 -term
(29)
for N = 0. Now the
(30)
i=1
where w is the weight of the monomial to which Q1 is being applied. Thus, E2 Hp is the
summand of (27), (28) of weight w where
2w 2a +
D

i=1
xi2 = 0
60
(the mass shell condition). Since E2 Hp is concentrated in filtration degree 0, 1, there is no

room for higher differentials, and the spectral sequence collapses to E2 Hp , i.e., E2 Hp is
the associated-graded object of the BRST cohomology.
5. The BRST cohomology of the N = 2 string
To treat the case of N = 2, we will need to understand better the algebraic structure
described in Section 3. Concretely, we shall consider the associative algebra A+ with
generators (9) and relations (10). Even more mathematically precisely, A+ is a completion
of this algebra in the sense that we allow infinite sums of monomials in normal order. Now
the BRST complex at a given momentum p = (X0 , Y0 , Z0 , T0 ) is
Hp = A+ /Ip
where I = Ip is the completed left ideal (infinite sums of monomials in normal order
allowed) spanned by brI , cnI , xn , yn , zn , tn , n , r , n , r , r 0, n > 0, x0 X0 , y0 Y0 ,
z0 Z0 , t0 T0 .
Our calculation will be performed under the assumption X0 , Y0 , Z0 , T0 = 0 (where
these are defined from X0 , Y0 , Z0 , T0 by a formula analogous to (21)). For this reason, we
will, instead of the algebra A+ , work with the algebra

A = A+ (x0 )1 , (y0 )1 , (z0 )1 , (t0 )1 .
Now the point of this discussion is that one can consider the BRST differential directly on
A, where it has the form
du = [Q, u],
u A.
This is a differential because [Q, Q] = 0, and additionally is a differentiation of algebras,

i.e.
d(u1 u2 ) = d(u1 )u2 + (1)d1 u1 d(u2 )
(31)
where d1 is the cohomological degree (= ghost number) of u1 . The reason Q induces a

compatible differential on A/I is that Q I .
Now we shall introduce a complete decreasing filtration F i both on A and Hp = A/I
which are compatible in the sense that the projection
: A A/I
(32)
satisfies (F i A) F i A/I . Additionally, we will have

Q F 0 A.
(33)
Now under these circumstances, one always has two accompanying spectral sequences on
A and A/I , which are a spectral sequence of (non-commutative) differential graded (DG)
algebras and differential graded modules (which means that formulas analogous to (31) in
both the DG algebra and module sense will continue to hold for the higher terms Er and
differentials dr of the spectral sequence). All this was true in the N = 0, N = 1 cases also,
61
but in the N = 2 case, there are actual higher differentials to calculate, and the present
technique is a tool for calculating differentials on non-linear monomias in Hp : we will see
that such terms play a crucial role.
The filtration is given both on A and A/I again by specifying filtration degrees of
generators and then extending to all elements in an analogous manner as above. We put
|x | = 1 (n = 0),
0n

c = | | = c1 = 1 for all n,
n
n
n
0

b = | | = b 1 = 1 for all n,
n
n
n

|y | = 1 for n = 0,
2n 12
2 12
c = c = 2,
b = b = 2
n
n
n
n
for all n.
zn , tn , n , n
The remaining matter generators

have filtration degree 0. Simple verification
shows that (33) indeed holds. To start out the calculation of the spectral sequence, the d0
differential is, again, given by [Q0 , ?] where Q0 is the sum of the terms of Q of filtration
degree 0. We see from (20), (23) that (25) still holds, so d0 is still given by (26), and d0
acts as an algebra differential on monomials of normal order. Thus, both E0 (A), E0 (A/I )
are Koszul complexes. This shows that the cohomology E1 (A) is the algebra defined in
the same way as A, except only on the generators
zn , tn , n , n , bn12 , cn12 , br2 , cr2 , c00 .
(34)
Also, E1 (A/I ) is the quotient of E1 (A) by the ideal defined in the same way as I , except
on the generators (34) only. In other words, additive bases of E1 (A), E1 (A/I ) (in the
completed sense in the case of E1 (A)) are monomials in normal order in the generators
(34), with the same subscript restrictions as in A, A/I . As before, d1 is now given by
commutator with the summands of Q of filtration degree 1, the non-zero terms of which
are

0 0
c00 2nn n
c00 2nbn0 cn
c0
Q1 = 2x0 y0 c00 + 2zn tn
1 0
2 0
12 0
2nbn1 cn
c0 2nbn2 cn
c0 2nbn12cn
c0 .
The effect on an element u A, A/I is, again, left multiplication by

2w(u) + x02 + y02 + z02 + t02 c00 ,
(35)
where w is the weight. Therefore, E2 (A) is simply the sum of the kernel and cokernel of
(35). In E2 (A/I ), one can say more explicitly that (35) leaves intact precisely all terms of
E1 (A/I ) which have

1 2
X0 + Y02 + Z02 + T02 = 0
2
and kills all others; this is, again, the mass shell condition.
Now to calculate d2 , we shall use the full force of our algebra machinery. First, let
us recall how a d2 (or any higher dn ) is calculated in a spectral sequence associated
with a decreasing filtration of a cochain complex C: we select an element u C whose
pq
representative in E0 C supports no differentials d0 , . . . , dn1 , and hence survives to En C.
w(u) +
62
Then we find an element v F >p C such that

d(u + v) F p+n C.
Then the representative of d(u + v) in En C is dn u. The element v can be chosen to be
a linear combination of elements whose representatives in En C have filtration degrees
p + 1, . . . , p + n 1 (i.e., only p + 1 for n = 2). Summands of dn (u) can come both from
u or v; the element v will be called the counterterm.
We claim that the d2 -differential on E2 (A) is given by

12

2
in
cr+n
itn
cr+n
,
d2 r =
(36)

2
12
izs crs
is cs+r
,
d2 r =
(37)

12 2
2in+r
tn
+ i(r n)br+n
cn ,
d2 br2 =
(38)

2
2
i:n+r
n
: :icn
br+n
:,
d2 br12 =
(39)

2
irn+r
cn
,
d2 zr =
(40)

2
irn+r
cn
,
d2 tr =
(41)
d2 cr12 = d2 cr2 = 0.
(42)
Recall that (E2 (A), d2 ) is a (non-commutative) DGA, so this determines the value of d2 on
any monomial. Further, since E2 (A/I ) is a quotient of E2 (A), the differential on E2 (A/I )
is induced.
To prove (36)(42), we use the procedure described above. To calculate d2 r , consider
commutators of r with all summands Q2 of Q of filtration degree 2. The summands
of Q2 which give non-trivial commutators are

0

1

2
12
(r n)r
n
cr+n
+ r
tn cr+n
+ ir
tn cr+n
+ in r
cn+r ,
which gives the commutator

0

1

2
12
(r n)n
cr+n
tn
cr+n
itn
cr+n
+ in cn+r
.
(43)
The last two terms, of filtration degree 2, survive to (36). The first two terms of (43) are of
filtration degree 1, and therefore must be cancelled by a counterterm. The counterterm is
r n
1

n
(44)
xr+n
tn
n+r .

2x0 (r + n)
x0
(Note that c00 is mass-shell related, so we may assume r + n = 0.) Now we must correct
our candidate of d2 r by adding the commutator of the sum Q1 of terms of Q of filtration
degree 1 with (44). But one easily checks that every monomial thus obtained contains
one of the matter generators n , xn as a factor, and thus vanishes in E2 (A). Thus, the
counterterm contributes nothing in this case. (37) is analogous.
In the case of (38), the summands of Q2 giving non-trivial commutators are

2

2

2
yn
cr
in+r
xn
cr
+ in+r
tn
cr
in+r

2
12 1 2
2
0 2
in+r
zn
cr
+ (n r)br+n
cn cr (n 2r)br+n
cn
cr ,
63
giving the commutator

in+r
yn
in+r
xn
+ in+r
tn

12 1
2
0
in+r
zn
+ (n r)br+n
cn (n 2r)br+n
cn
.
(45)
12 c 1 , b 2 c 0 require counterterms which

The first term is in F0 A, and dies. The terms br+n
n r+n n
12 , b 2 x which, when commuted with Q , die similarly
are scalar multiples of br+n
1
n r+n n
x is in F A and hence dies, unless n = 0. In that case,
as above. Also, the term in+r
0
n
we have ir x0 of filtration degree 1, which requires adding the counterterm
ibr1 .
(46)
When computing the commutator of Q1 with (46), the terms giving a non-trivial
commutator are

1

1

1
in+r
yn
cr
+ in+r
xn
cr
+ in+r
tn
cr

1
1
0 1
+ in+r
zn
cr
i(n 2r):br+n
cn
cr :
1 0 1 1
12 2 1
+ i :br+n
cn cr : + i(r n)br+n
cn cr .
2
From the resulting commutator with (46), the terms which survive to E2 (A) are

12 2
in+r
tn
+ in+r
zn
+ i(r n)br+n
cn
which, when added to the third and fourth term of (45), gives (38).
To get (39), the terms of Q2 which produce a non-trivial commutator with br12 are

12

12
12 12 0
2
12 1
i:n+r
n
:cr
+ i:n+r
n
:cr
2rbr+n
cr cn + br+n
cr
cn ,
producing the commutator

12 0
2
1
n
: + i:n+r
n
: + 2rbr+n
cn br+n
cn
.
i:n+r
(47)
The first summand of (47) dies, the second survives to E2 (A). The last two terms must
be cancelled by counterterms (n = 0 in the third term to avoid the mass shell term). The
counterterms are
r 12
b x
(n = 0)
x0 n r+n n
1 2

n
.
+ br+n
(48)
x0
The first term of (48), when commuted with Q1 , dies in E2 A as above, while the second
term has non-trivial commutator with the summands of Q1
1

2
n xs csn in xs cns
.
The only cases which survive are s = 0 in the first summand (which we need for cancelling
the relevant term of (47), and s = 0 in the second summand, which leads to
2
2
bn+r
:.
i:cn
64
Adding this to the second term of (47) gives (39).

To get (40), the terms of Q2 producing commutators with zr are

1

2
2zn+r
tr
cn0 + n+r
tr
cn
+ in+r
tr
cn
,
leading to the commutators

0

1

2
2rzn+r
cn
rn+r
cn
irn+r
cn
.
(49)
The last term survives to (40), the first term for n = 0 and second term need counterterms

, , but both of these when commuted with Q
which are multiples of zn+r
xn
1
n+r n
only produce terms which die in E2 (A) because of matter factors of the form x or .

are
Concretely, the summands of Q1 producing a non-trivial commutator with zn+r
xn
1
2
+ is yn csn
s yn cns
leading to
1

2

zn+r
ins csn
zn+r
,
ns cns
are
and summands of Q1 producing a non-trivial commutator with n+r
n
0
1
+ n xs cns
,
(s n)s n csn
where s n = 0 in the first summand and s = 0 in the second summand. This produces
0

1

(s n)s csn
n+r
+ xs cns
n+r
which dies as promised. This proves (40). (41) is analogous.

For (42), only ghost summands of Q give non-trivial commutators with cr2 , cr12 . Some
counterterms are needed to cancel the commutators, but none produce non-zero terms in
d2 .
Now by looking at (36)(42), it is not immediately obvious what the cohomology is.
To settle that, we will introduce an additional filtration on (E2 (A), d2 ), (E2 (A/I ), d2 ) and
use the resulting spectral sequence(!). To simplify notation, set := d2 .
The filtration follows a now familiar pattern: it is a decreasing filtration defined as
above by specifying filtration degrees of generators. The formula is
2 12
b = b = | | = 1,
|zn | = 1 if n = 0,
r
r
r
2 12
c = c = | | = 1,
|t | = 1
if n = 0.
r
i ,
One easily checks that

so the filtration is compatible with the differential.
As above, the differential 0 on E0 E2 (A) is given by the degree 0 summands of .
Comparing with (36)(42), we see that
0 r = t0 cr2 ,
0 br2 = 2it0 r ,
and all other 0 s on generators are 0. Thus both (E0 E2 (A), 0 ), (E0 E2 (A/I ), 0 ) are
Koszul complexes, and their cohomologies E1 E2 (A), E1 E2 (A/I ) have as additive bases
the sets of normal-ordered monomials in zn , tn , bn12 , cn12 with the same values of the indices
65
as in the definition of A, A/I . Moreover, E1 E2 (A) is an algebra (same relations as A),

and E1 E2 (A/I ) is its factor by a left ideal.
Now one sees that 1 = 0, so E2 E2 (A) = E1 E2 (A), E2 E2 (A/I ) = E1 E2 (A/I ). Now
let us calculate 2 (zr ), by the same method as above. Thus, we apply first and get

2
cn
.
ir
n+r
But this is of filtration degree 0, so we must add the counterterm
r

:.

:n+r n
t0
(50)
Note that if and only if both n + r, n have the same sign (are both non-positive or nonnegative, which happens for finitely many but more than 0 terms, which in addition have
the same sign and hence cannot cancel), this will produce a non-zero multiple of cr12 by
using commutators to bring the of (50) to normal order (see (36), (37)). Thus, we see that
2 (zr ) = Cc12 ,
C = 0.
(51)
Now consider 2 (br12 ). Applying first, we obtain

2
2
n
: :cn
br+n
:.
i
:n+r
This is cancelled by a non-zero multiple of the counterterm

2
n
bn+r
.
b 2 ), by bringing
But as above, we see that when n + r, n have the same sign, (n
n+r
terms to normal order, produces non-zero multiples (with the same sign) of tr . Thus, we
see that

2 br12 = Dtr , D = 0.
(52)
Now (51), (52) show that (E2 E2 (A), 2 ), (E2 E2 (A/I ), 2 ) are again Koszul complexes,
and that their cohomology is

c012 Ker(d1 ) Coker(d1 ) c0 ,
(53)
12 0
2
2
2
2
c0 , c0 if x0 + y0 + z0 + t0 = 0 and 0 else,
(54)
respectively. We see that the spectral sequences and also the d-spectral sequences must
collapse to (53), (54), for the only remaining terms are in filtration degrees 0, 1, 2. This is
what we claimed.
Appendix A. A few low weight explicit differentials

It is quite difficult to write down explicitly all the N = 2 BRST differentials on Hp
even in the lowest non-trivial weight w = 1 because of proliferation of counterterms. To
illustrate the difficulty, and the great simplification the spectral sequence provides, we list,
as an example, some (not all) of the differentials in weight 1 in the NS sector. We use the
66
matter generators (9); the generators (21) were introduced to work better with the filtration,
but the generators (9) give more symmetric explicit formulas. We have

1
0
1
2
2
= 2 b1/2
c1/2
+ b1/2
c1/2
x0 x1 y0 y1 z0 z1 t0 t1 ,
Qb1

1
1
2
2
0
c1/2
b1/2
c1/2
,
Q x0 y1 y0 x1 + z0 t1 t0 z1 b1/2
= 4c1
12
2
1
1
2
Qb1
= 1/2 1/2 + 1/2 1/2 + b1/2
c1/2
b1/2
c1/2
,
12
Q(1/2 1/2 x0 y1 + y0 x1 ) = Q(1/2 1/2 z0 t1 + t0 z1 ) = c1
,

2
1
2
1
Q z0 b1/2 1/2 + b1/21/2 t0 b1/2 1/2 + b1/21/2
2

2

1
1
x0 b1/2
1/2 + b1/2
1/2 + y0 b1/2
1/2 + b1/2
1/2

12
1 2 2 2

c012 b1/2
b1/2
z02 + t02 x02 + y02 b1
= 2(1/2 1/2 1/2 1/2 x0 y1 + y0 x1 + z0 t1 t0 z1 ).
References
[1] A. Bilal, BRST approach to the N = 2 superconformal algebra, Phys. Lett. B 180 (3) (1986) 255.
[2] R. Bogojevic, Z. Hlousek, The BRST quantization of the O(2) string, Phys. Lett. B 179 (1986) 69.
[3] P. Deligne, J.W. Morgan, Notes on supersymmetry (following J. Bernstein), in: Quantum Fields and Strings:
A Course for Mathematicians, Vol. 1, Amer. Math. Society and the Inst. for Advanced Study, 1999, pp. 41
98.
[4] J. Distler, P. Nelson, Semirigid supergravity, Phys. Rev. Lett. 66 (5) (1991) 1955.
[5] D. Fatori, V.G. Kac, Classification of finite simple Lie conformal superalgebras, J. Algebra, in press.
[6] B.R. Greene, hep-th/9702155.
[7] I. Kriz, On the N -superconformal algebra, in preparation.
[8] J. McCleary, A Users Guide Spectral Sequences, in: Cambridge Stud. Adv. Math., Vol. 58, Cambridge
Univ. Press, Cambridge, 2001.
[9] S. Mathur, S. Mukhi, BecchiRouetStoraTyutin quantization of twisted extended fermionic string, Phys.
Rev. D 36 (2) (1987) 465.
[10] S. Mathur, S. Mukhi, The N = 2 fermionic string: path integral, spin structures and supermoduli on the
torus, Nucl. Phys. B 302 (1988) 130148.
[11] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge Univ. Press, Cambridge, 1987.
[12] H. Ooguri, C. Vafa, Geometry of N = 2 strings, Nucl. Phys. B 361 (1991) 469518.
[13] J. Polchinski, String Theory, Vol. 1, Cambridge Univ. Press, Cambridge, 1999.

Remarks on stable and quasi-stable k-strings

at large N
A. Armoni a , M. Shifman a,b
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b William I. Fine Theoretical Physics Institute, University of Minnesota, Minneapolis, MN 55455, USA 1
Received 16 July 2003; received in revised form 11 August 2003; accepted 19 August 2003
To the memory of Ian Kogan
Abstract
We discuss k-strings in the large-N YangMills theory and its supersymmetric extension. Whereas
the tension of the bona fide (stable) QCD string is expected to depend only on the N-ality of the
representation, tensions that depend on specific representation R are often reported in the lattice
literature. In particular, adjoint strings are discussed and found in certain simulations. We clarify this
issue by systematically exploiting the notion of the quasi-stable strings which becomes well-defined
at large N. The quasi-stable strings with representation-dependent tensions decay, but the decay rate
(per unit length per unit time) is suppressed as 2 F(N) where F(N) falls off as a function of N.
It can be determined on the case-by-case basis. The quasi-stable strings eventually decay into stable
strings whose tension indeed depends only on the N-ality.
We also briefly review large-N arguments showing why the Casimir formula for the string tension
cannot be correct, and present additional arguments in favor of the sine formula. Finally, we comment
on the relevance of our estimates to Euclidean lattice measurements.
PACS: 11.15.Pg; 12.38.Gc
1. Introduction
In confining theories, such as the YangMills theory, non-supersymmetric or supersymmetric (N = 1 gluodynamics), heavy (probe) color sources in the fundamental representaE-mail address: adi.armoni@cern.ch (A. Armoni).
1 Permanent address.
doi:10.1016/j.nuclphysb.2003.08.021
68
A. Armoni, M. Shifman / Nuclear Physics B 671 (2003) 6794
tion are connected by color flux tubes, fundamental QCD strings. The fundamental string
tension is of the order of 2 where is the dynamical scale parameter of the gauge theory
under consideration, and its transverse size is of the order of 1 . Both parameters are
independent of the number of colors (in what follows the gauge group is assumed to be
SU(N )) and, besides , can contain only numerical factors.
Significant effort has been invested recently in studies of the flux tubes induced
by color sources in higher representations of SU(N ), mostly in connectionbut not
exclusivelywith high-precision lattice calculations, see e.g., [17] and references therein.
The composite flux tubes attached to such color sources are also known as k-strings,
where k denotes the N -ality of the color representation under consideration. The N -ality
of the representation with upper and m lower indices (i.e., fundamental and m antifundamental) is defined as k = | m|.
Some qualitative features of the k-strings are well understood. The most important
feature is that the string tension does not depend on the particular representation of
the probe color source, but only on its N -ality. Indeed, the particular Young tableau of
the representation plays no role, since all representations with the given N -ality can be
converted into each other by emitting an appropriate number of soft gluons. For instance,
the adjoint representation has zero N -ality; therefore, the color source in the adjoint can
be completely screened by gluons, and the flux tube between the adjoint color sources
should not exist. The same is true for any representation with N fundamental or N antifundamental indices. Theoretical ideas regarding confinement in lattice gauge theories and
the role of N -ality are extensively reviewed in Ref. [8].
At the same time, one can find in the literatureespecially, devoted to latticesreports
on the adjoint string tension at intermediate distances, measurements of distinct string
tensions for symmetric and anti-symmetric representations of one and the same N -ality,
and so on. For instance, in Ref. [6] which deals with SU(3) gauge group, the string tensions
(at distances up to 2 fm) are measured for representations 8 of N -ality 0, 3 and 6 of N ality 2, 10 of N -ality 3, and for some other representations. The question is: is there a
conflict between the above results and the general principles?
A part of the apparent contradiction is due to semantics. There is an objective difficulty
too: quantitative analysis is hindered by the fact that seemingly there is no small expansion
parameter in the k-string problem, that could play a guiding role.
In this paper we demonstrate that, using 1/N as such a parameter, one cannot only
resolve the above conflict, but put the analysis on a relatively quantitative basis. (Here N
is the number of colors.) We assume the t Hooft scaling, i.e., g 2 1/N , so that is kept
fixed.
Gauge theories at strong coupling have no expansion parameter other than 1/N . Within
the t Hooft framework, the 1/N expansion is abstracted from the Feynman graphs.
However, since one in fact uses only the topology of the graph, the result obtained is valid to
all orders in the gauge coupling Ng 2 and is, thus, non-perturbative. The leading term in the
1/N expansion corresponds to the planar t Hooft graphs, the next-to-leading to the t Hooft
graphs with one handle, and so on. This featurethe topological character of the 1/N
expansionis inherited and fully incorporated into string theory. The string perturbation
theory is the expansion in gst . For closed strings it is gst2 that enters, which corresponds to
a 1/N 2 expansion in gauge theories. Non-perturbative stringy effects (D-branes), that are
69
Fig. 1. A flux tube for the 2-string.
exponential in 1/gst , may give rise to exponential terms of the type exp (N) or exp (N 2 )
in gauge theories.
At large N the problem of composite strings becomes, in essence, a weak coupling
problem. A crucial starting observation is that at N = the fundamental strings (i.e.,
strings that connect a quarkanti-quark pair in the fundamental representation) become free
and non-interacting, while their interaction emerges only at the level of 1/N 2 corrections
[9].
In the remainder of this paper we will adopt the following convenient convention:
the length of the strings under consideration will be chosen to be N -independent (but
certainly much larger than 1 ). The N -dependence is reserved for time duration of the
measurement.
We view a k-string as a bound state of k elementary fundamental strings (see Fig. 1).
This standpoint is natural in the gauge theory/supergravity correspondence [10] where the
k-string world sheet (Wilson loop) is described as k-coincident elementary world sheets
[11,12]. Accordingly, in field theory one can define the k-string Wilson loop as k coincident
fundamental Wilson loops. In doing so, one will deal with the probe source which belongs
to a reducible representation with k upper indices. If the (Euclidean) observation time is
sufficiently large, all excited-string contributions will die out, and one will measure the
ground-state string tension. Alternatively, for the study of quasi-stable strings, one can
single out a particular representation of the probe source, by considering the Wilson loop
in the appropriate representation.
The question we address below is the life-times of quasi-stable strings and tension
splittings. Our approach is based on a Minkowskian (quasiclassical) picture of string
interactions which is especially appropriate for the description of various tunneling
transitions. Some elements of this picture can be traced back to the 1970s [13]. The validity
of the quasiclassical approximation is justified by the smallness of 1/N . We then translate
our results in the Euclidean language appropriate for lattice simulations.
Using general properties of weakly coupled systems we will show that the decay rates
of the quasi-stable strings (per unit length per unit time) scale as 2 F (N). The
dependence F (N) is either N 2 in the case of the adjoint string, exp(N 2 ) is the case
of a decay from a pure higher (i.e., non-anti-symmetric) representation into the ground
70
state (the anti-symmetric representation), or exp(N) when k N (the saturation limit).

The common reason behind the N -dependence is that all these processes are non-planar in
their nature and, therefore, in the large-N limit they will never occur (an infinite amount of
time is needed to observe them).
For translating our results in the Euclidean language it is important to understand into
which particular final state quasi-stable strings decay. We will comment on this issue too.
We expand our previous results [9] on the k-string tensions by further analyzing string
interactions at large/small distances. In particular, we explain, from a somewhat new angle,
the impossibility of the Casimir scaling2 for the bona fide k-strings3 and what may mimic
the Casimir scaling in lattice measurements at intermediate distances.
The organization of the paper is as follows: Section 2 presents short theoretical and
lattice-data reviews on Wilson loops and k-strings. In Section 3 we estimate the decay
rates of quasi-stable strings in various situations. In Section 4 we introduce the notion of
the saturation limit and discuss the relevance of the sine formula in this limit. Section 6
is devoted to oversatuarted strings. In Section 7 we provide a physically transparent
explanation of the interaction term k 3 /N 2 in the expansion of the sine formula. In Section 8
we introduce and discuss the tension deficit for composite strings. In Section 9 we suggest a
tentative explanation why the Casimir scaling could be observed at intermediate distances.
Section 10 is devoted to k-strings in the orientifold field theories. In Section 11 we
discuss the impact of our results on the Euclidean lattice calculations. Finally, Section 12
presents concluding comments and discussion of relevance of our estimates to N = 3.
A toy (exactly solvable) model illustrating k 3 /N 2 in the expansion of the sine formula in
the tension can be found in Appendix A.
2. Preliminaries: theoretical background, lattice data

The purpose of this section is to define and review the notion of stable (quasi-stable)
k-strings and review the way their tensions are measured. We also briefly summarize our
previous paper [9].
2.1. k-stringstheoretical background
Suppose we wish to measure a long-distance quarkanti-quark potential in pure SU(N)
gauge theory or in its minimal supersymmetric extension, supersymmetric gluodynamics.
We assume that the quarkanti-quark pair is in a specific representation R. The expected
long-distance potential is
V = L.
(1)
The stable string tension should not depend on the specific representation R of the
quarkanti-quark pair, but rather on its N -ality k, since a soft gluon can transform a
2 The very term Casimir scaling was introduced in Ref. [1].
3 By bona fide k-strings we mean the stable strings with all k indices anti-symmetrized.
71
Fig. 2. (a) A screening of an adjoint Wilson loop by a dynamical adjoint loop. (b) The same process described in
t Hooft double index notation. The process in non-planar.
representation R into a representation R within the same N -ality. For this reason we can
evaluate the string tension by considering a reducible representation with a given N -ality
k instead of a specific representation R, provided this reducible representation contains R.
For instance, one can consider, as a probe source, an ensemble of k heavy quarks (in the
fundamental representation) nailed in close proximity from each other. Another source,
at distance L from the former, will be composed of k heavy anti-quarks. Since the fully
anti-symmetric representation is expected to have the lowest energy, the string attached
to a source in a reducible representation will evolve into the anti-symmetric string after
a certain time . The time is definitely than a typical inverse splitting between the
energies of the anti-symmetric string and other representations. In fact, sometimes it may
be much larger.
By the same token, if the probe charges have N -ality zero, even if they are connected
by a string at time zero, it will inevitably evolve into a no-string state since such charges
can be totally screened. For instance, the adjoint probe heavy quark is screened by a gluon.
Another example: if one has N fundamental quarks separated from N fundamental antiquarks by a large distance L eventually each of the two N -quark ensembles will develop
string junctions (a baryon vertex), and there will be no string connecting the two ensembles.
A graphic illustration is presented in Fig. 2. Suppose that one calculates the expectation
value of a Wilson loop in the adjoint representation (i.e., induced by an adjoint probe
quark). Dynamically, a Wilson counter-loop in the adjoint representation will be
formed thus screening the non-dynamical probe quark. In physical terms, a gluon lump
is produced which combines with the probe adjoint source to form a color singlet.
The fundamental string, by definition, is the one that connects a fundamental heavy
quark with an anti-quark. The interaction between the fundamental strings is via the
glueball exchanges. The process is non-perturbative in the coupling g 2 N . Although
one uses the Feynman graphs in the t Hooft representation in order to analyze the N dependence, by no means it is implied that the results thus obtained are perturbative in the
gauge coupling and correspond to few-gluon exchange. As is standard in the t Hooft
framework, since the 1/N analysis is based only on topology of the graphs (e.g., planarity
vs. non-planarity), the results are expected to fully represent the non-perturbative strong
coupling gauge dynamics.
Since we deal with the interactions between color-singlet objects and the interaction is
non-planar, it will be controlled by O(1/N 2 ), see Ref. [9]. At N = the interaction
vanishes, and we have k free fundamental strings with k = kf . A sample 1/N 2
interaction between two fundamental Wilson loop is depicted in Fig. 3(a). The fact that
72
Fig. 3. The interaction of two fundamental strings: (a) field-theory picture(two)gluon exchange;
(b) string-theory pictureexchange of a closed string between two world sheets.
the corrections to the free strings relation k = k1 run in powers of 1/N 2 was used [9] to
exclude the Casimir scaling hypothesis. On general grounds one can assert that at distances
1 the fundamental strings attract each other [9], while at short distances there is a
repulsion (see the end of this section), so that composite k-strings develop.
This picture is supported by string theory. String theorists prefer to call the Wilson loop
a string world sheet. The realization of the k-string will be simply k coincident elementary
(fundamental) strings, or more precisely, a bound state of k elementary string world sheets,
see Fig. 3(b).
Within the AdS/CFT correspondence [10] we can elevate this model to a quantitative
level [11,12]. Indeed, the AdS/CFT correspondence is a natural framework for calculating
the k-string tension, since in the dual string picture we work at strong t Hooft coupling.
In addition, 1/N is represented by gst . In this framework the Wilson loop on the AdS
boundary is described by a fundamental world sheet that extends inside the bulk AdS
[11]. The value of the Wilson loop is the area of the minimal surface, given simply by
the NambuGoto action
W = exp(SNG ).
(2)
Similarly, the k-string Wilson loop (6) is described by k coincident world sheets. In the
supergravity approximation, gst = 0, we obtain k = k1 [12], as expected. In order to
calculate 1/N corrections, one has to consider string interactions, namely to go beyond the
lowest order in gst . It is clear, however, that in the closed string theory (and, in particular,
in type IIB) the expansion parameter is gst2 = 1/N 2 .
The fact that the interaction between the fundamental strings is proportional to 1/N 2
and, thus, vanishes at N is explained in detail in Ref. [9] Moreover, from the
same work we know that at distances 1 the interaction is attractive. An attractive
potential between the fundamental strings is also obtained [14] in lattice strong-coupling
calculations. What is the nature of the inter-string interaction at distances of the order of
the fundamental string thickness?
Logically there are two options. Either the parallel strings attract at all distances (then
k closely situated 1-strings will glue together forming a structureless flux tube carrying k
units of N -ality), or the attraction gives place to repulsion at shorter distances (then the
k-string will have a substructure in the transverse plane reminiscent of that of a nucleus).
The large-N expansion of the k-string tension suggests that it is the latter option that is
73
Fig. 4. The spatial slice of two overlapping fundamental strings.
realized in QCD. Indeed, on general grounds, with no model dependence, one can show
that at N
1 and k
1

k
k2
1 2
kf
N
(see Section 4.1). The above k dependence of the binding energy has no natural
explanation in the picture of a forced collectivization inevitable under the assumption
that the parallel 1-strings attract at all distances. At least, we are unaware of such an
explanation and were unable to obtain it, in spite of several attempts.
At the same time, the repulsion at shorts distances, that naturally leads to a nucleus-type
structure of the slice of the k-string, and explains the above k-dependence of the binding
energy (see Section 7) is an immediate consequence of the following consideration.
The trace of the energymomentum tensor in pure YangMills theory or in QCD

with massless fermions has the form (the scale anomaly)
b a ,a
G G
,
32 2
where Ga is the operator of the gluon field strength tensor (b is the first coefficient of
the Gell-MannLow function). One can use (x) as a local probe of the energy density.
Indeed the value of G2 , in the presence of a Wilson loop W (C), measures the string
tension [15],

1

W (C)
(3)
d 3 x (x), W (C) c,eucl = 2 R,
=
where the subscript c stands for connected part. The chromomagnetic part is presumably
negligible for the static flux tubes attached to static color sources. This implies that the
energy density is proportional to the expectation value of the operator E 2 (x), where E is
the chromoelectric field, and the arrow is used to represent both vectorial and color indices.
Then the string tension (the energy per unit length in the z direction) is given by the integral

d 2 x E 2 (x) ,
where the integral runs in the perpendicular plane.
Two overlapping fundamental flux tubes are depicted in Fig. 4. Now

2

2

2 d x Ef1 (x) + Ef2 (x) = 2f + 2 d 2 x Ef1 (x)Ef2 (x) .
74
The fluxes are fixed by the given (static) color sources. This implies that the interference
term is positive. It is not difficult to see that it is also suppressed by 1/N 2 , because
generically E f1 is orthogonal to Ef2 in the color space; only the Cartan components of
the chromoelectric field are important. Thus, if two parallel 1-strings overlap, the energy
per unit length is larger than 2f by 2 /N 2 , i.e., overlapping flux tubes repel each other.
Certainly, the arguments presented above are somewhat quasiclassical. We believe,
however, that they are qualitatively correct; they certainly lead to a self-consistent overall
picture. For further discussion see Section 7.
2.2. A mini-review of the lattice literature on k-strings
The Euclidean/lattice formulation of the problem is usually given through the expectation value of a rectangular Wilson loop,

a a
W(C) = tr exp i A TR dx
(4)
,
where TRa are the generators of SU(N) in the given representation, which may or may
not be irreducible. Sometimes, for practical purposes, it is more convenient to consider
a correlation function of two Polyakov lines. Assume that the (Euclidean) time direction
is compactified, and one defines two Polyakov lines in the time direction, separated by
a distance L. The Polyakov line is defined through the integral similar to (4), with the
generators in the given representation R. If time interval is large enough, the measurement
of the correlation function of two Polyakov lines must yield exp(R LT ).
The generators of the reducible representation with N -ality k are given by tensor
products of the fundamental representation. For example, in the case of N -ality 2 we have
=
+ .
(5)
Therefore we can evaluate the k-string tension by considering the following Wilson loop

k

Wk -string(C) Wfund (C) .
(6)
Physically, the above definition (6) represents k-fundamental coincident Wilson loops. It
is instructive to think of it as of k fundamental probe sources placed at one point. The
resulting composite probe source is in the reducible representation of the N -ality k, which
includes a mixture of irreducible representations, starting from fully anti-symmetric, up to
fully symmetric.
One of the main tasks of this paper is to bridge a gap between theoretical studies of
k-strings from the string theory side and large-N field theory side on the one hand, and
lattice studies, on the other hand. Therefore, it is natural to give a brief summary of some
lattice works devoted to k-strings.
Let us start with analytic studies, namely the strong coupling expansion. Since the
seminal work of Wilson [16] it is known that in the YangMills theory with no dynamical
matter in the fundamental representation, the expectation value of a large Wilson loop
induced by a heavy (non-dynamical) fundamental quark will trivially exhibit an area law

tr exp i A dx exp( A).
(7)
75
Here the string tension, (ln g 2 )a 2 , where a is the lattice cite and g is the gauge
coupling, g .
The area law for the fundamental Wilson loop is also well established in the continuum
limit, through numerical simulations. To detect the area law there is an obvious necessary
condition on the area of the loop,
A
2 .
(8)
The slope in front of the area, Eq. (7)the fundamental string tensionis measured to a
reasonable accuracy. Moreover, a perpendicular slice of the fundamental string was studied
too. A typical transverse dimension of the fundamental string is 0.7 fm.
On the other hand, an area law has been also detected for the adjoint Wilson loop for
contours satisfying the condition (8). For instance, typical distances in Ref. [6] were 1.5
to 2 fm. The adjoint string tension was measured. The reported ratio adj /f is close to the
Casimir formula which for SU(3) yields 9/4. (Please, note that this number is larger than
2; we will return to this point in Section 9.) In fact, the adjoint strings emerge practically
in all lattice studies. So far, only one work [5] reports an observation of the adjoint string
breaking.4
A linear potential between an adjoint probe quark and anti-quark implies the existence
of a quasi-stable string. It is certain that for asymptotically large areas, the area law
must give place to the perimeter one; the linear potential must flatten off at the level
corresponding to the creation of two gluelumps [5]. The effect is not seen, however, in
the existing simulations. The question to ask is: what is the critical size/time needed for
detecting the adjoint string breaking?
Now, the very same dynamical adjoint counter-loop as in Fig. 2 (or a few counterloops) can transform external probe quarks in a given representation R to a different
representation R with the same N -ality. In physical terms, if we have, say, an antisymmetric source Q[ij ] , we can convert it into an object with two symmetric indices {ij }
by adding a soft gluon. It is clear, then, why one expects the k-string tension to depend
on the N -ality, and not on the specific representation R. The genuine stable string for
the given N -ality is expected to be attached to the probe source in the anti-symmetric
representation. The probe sources in other representations with k (upper) indices give rise
to quasi-stable strings whose tension is supposed to depend not only on the N -ality, but on
the particular representation. If the probe sources have k > [N]/2, we will call such strings
oversaturated. Representations which have both upper and lower indices can be viewed
as bound states of k-strings and some number of adjoint strings. Obviously, they are also
quasi-stable. We will refer to such strings as bicomposites.
Lattice measurements of the Wilson loops or Polyakov line correlators seem to defy the
above argument. They yield string tensions which depend on the particular representation
under consideration rather than on the N -ality of the representation. For instance, in
Ref. [3] which treats the SU(4) and SU(6) cases and measures the anti-symmetric
and symmetric Wilson loops, distinct tensions are obtained for the anti-symmetric and
symmetric two-index representations. It was argued [3] that the symmetric string is not
4 This work is based on a technique different from all other calculations.
76
stable and that it will have to decay into two fundamental strings. In light of the above,
again, the most crucial question is what contour sizes are needed to exhibit the decay of
the symmetric string into anti-symmetric. It is clear that the answer depends on the decay
rate of the symmetric string.
Returning to the tensions of the stable (anti-symmetric) strings, there is no consensus in
the lattice literature on the k-string tension. The Casimir scaling is often reported. In [2] is
it argued that in three dimensions the k-string tension is very close to the Casimir scaling.
A similar claim is made in [7] for four-dimensional SU(3) theories, where the Casimir
behavior is found for symmetric representations. On the other hand, there are dedicated
studies [3] which favor the sine formula for the anti-symmetric strings.
As we have already mentioned, the Casimir scaling behavior for a stable string [2]
is in contradiction with our analytical considerations [9]: the string tension should be
expandable in powers of 1/N 2 rather than 1/N .
3. Decay rates of the quasi-stable strings: case study

In this section we will give systematic estimates of the quasi-stable string decay rates.
The strings are assumed to live in the Minkowski time. As was mentioned, the length of the
strings under consideration will be chosen to be N -independent (but certainly much larger
than 1 ). The N -dependence is reserved for time duration. We will calculate the N and
representation dependence of the decay rates per unit time per unit length of the string. We
will assume that N
1 to infer regularities in this limit, and then speculate as to which
extent our results may survive extrapolation to N = 3. The question as to how these rates
are reflected in the lattice measurements of the Wilson loops/Polyakov lines is deferred till
Section 11.
3.1. Fundamental string in QCD with dynamical quarks
To warm up, let us start our consideration from SU(N) YangMills theory with, say,
one dynamical (Dirac) quark in the fundamental representation. In this context, the title of
this section might seem provocative, since, as everybody knows, dynamical quarks screen
the fundamental probe color charges, and the string does not exist in this case.
Remember, however, that at large N the dynamical quark production is suppressed by
1/N . The quark loops decouple at N = ; in this limit the fundamental stable string
exists, and its tension is well defined and equal to a numerical constant times 2 . At large
but finite N the string is quasi-stable. The probability of its breaking per unit length per
unit time was found long ago [13],
f 2 N 1 .
(9)
This follows from the fact that planar graphs with one quark-loop insertion are suppressed
by the factor N 1 .
After the string is broken by the dynamical quarkanti-quark pair it decays into two
(here Q is the probe non-dynamical
stringsone attached to the pair Qq,
another to Qq
77
Fig. 5. (a) A dynamical pair of soft quarks near the external charges. (b) Creation of a dynamical pair from the
vacuum in the process of the string breaking. The dynamical quarks are hard.
quark, while q is a dynamical one). Let us have a closer look at the final state (the reason
why we dwell on this issue will become clear later).
If we limit ourselves to the classical approximation, we will have to conclude that the
final-state strings are typically long, so that the overall energy carried by these two strings
is f L mod , i.e., the same as the energy stored in the original string before breaking.
Indeed, the qq
pair is produced from the vacuum in a soft manner, locally, at a certain
point on the string. Typically, the momenta of q and q at the time of production are of the
order of . Larger momenta are highly unprobable since they can be obtained only from
exponentially suppressed tails of the wave function.
In the large-N limit, the configuration obtained after the production of the qq
pair
evolves in time with no further breakings. Let us have a closer look at the halves
connected
of the broken string. Each consists of two well separated quarks (say, Qq)
by a long well developed string carrying energy scaling as L. The probability for this
configuration to materialize as a ground-state meson is exponentially suppressed. In typical
resonances, strings have sizes 1 (strictly speaking, they cannot be called strings,
rather sausages). A long string resides in exponentially suppressed tails of the meson
wave functions.
On the other hand, exactly this configurationtwo quarks connected by a long well
developed stringis typical for highly excited meson states (radial excitations). Being
highly excited, they can be treated quasiclassically. In such mesons the light quark is
smeared around the heavy one at a large distance, which parametrically grows with the
excitation number. The energy stored in the gluonic degrees of freedom is proportional
to this distance. If it were not for the 1/N suppression, this meson would decay into the
meson plus multiple glueballs.
ground-state Qq
As was mentioned, quantum-mechanically, there is a certain overlap of the final state
depicted in Fig. 5(a). This
described above with the ground state mesons Qq and Qq
projection must be small, however. The smallness is unrelated to N ; it is controlled by the
parameter L. We will return to the issue of the smallness of this projection in Section 11.
Eq. (9) implies a particular relation between the mass of a highly excited meson and its
decay width, namely,
n = CMn N 1 ,
C 0.5.
(10)
Here n is the excitation number, n

1. The above relation was numerically studied in the
t Hooft model and was confirmed to be valid [17].
3.2. Adjoint string
Now, consider pure SU(N) YangMills theory (no dynamical quarks) with the probe
heavy sources in the adjoint representation. The screening of the sources occurs via
creation of a pair of gluons. At the hadronic level, the string breaking is equivalent to
78
produces a pair of (color-singlet) mesons of the type

the statement that the operator QQ
QG. Here Q is the field of the probe heavy quark while G stands for the gluon. It is easy
to see that the amplitude of this process is suppressed by 1/N , while the probability of the
string breaking (per unit length per unit time) is 2 /N 2 (see Fig. 2(b)),
adj 2 N 2 .
(11)
The adjoint string is quasi-stable, much in the same way as the one in Section 3.1. The
difference is that now the decay rate is suppressed by N 2 rather than N 1 . Although the
distinction is quantitative rather than qualitative, it may entail consequences for numerical
simulations. Indeed, for large N , the available lattice sizes may be insufficient to place
Wilson contours large enough to observe the adjoint string decay (see, however, Ref. [5]).
What can be said about the adjoint string tension adj ? On general grounds

adj = 2f 1 + O N 2 .
(12)
Since at distances 1 there is an attraction between the fundamental strings, the
O(N 2 ) term on the right-hand side must be negative. To measure this term in the
Euclidean lattice simulations one would need to have the contour area A > 2 N 2 .
However, the adjoint string will decay at parametrically smaller areas, see Section 11.1.
Therefore, at asymptotically large N the adjoint string tension is not measurable through
existing lattice methods. At N = 3 the splitting between adj and 2f may be measurable,
though, if the interplay of numerical factors is favorable.
3.3. Bicomposites
In this section we will discuss quasi-stable strings combining features of the adjoint
strings and k-strings, namely, we will consider the probe source of the type
i ik ik+1
Qj1
(13)
One can either deal with the reducible representation or pick up an irreducible one, say, by
anti-symmetrizing all upper indices. In fact, the situation is very similar to that discussed
in Section 3.2, as long as k does not scale with N .
In this case the (quasi-stable) string tension does not depend on the N -ality k, but, rather,
on the index ,
= n + m,
(14)
where n is the number of fundamental and m anti-fundamental indices of the probe field
Q. For the probe source (13) = k + 2.
The tension of the given bicomposite is

bicomp = (2 + k)f 1 + O N 2 .
(15)
It decays into a k-string through the gluelump creation, Fig. 2. The decay rate is given
by Eq. (11) and is suppressed by 1/N 2 . Since the binding energy is O(1/N 2 ) and the
critical contour area scales as ln(N 2 /C), the determination of the binding energy of this
quasi-stable string is impossible at asymptotically large N .
As k grows, on combinatorial grounds one expects the decay rate to increase, and loose
its 1/N 2 suppression when k becomes of the order of N/2.
79
3.4. k-strings and their excitations
In this section we will consider probe color sources of the type Qi1 ,i2 ,...,ik or Qi1 ,i2 ,...,ik ,
with k fundamental or anti-fundamental indices. The values of k to be discussed in this
section are of the order of N 0 . The values of k N 1 are considered in Sections 4 and 5.
In the general case, this is a reducible representation of SU(N ). To single out irreducible
representations, one must perform symmetrization (anti-symmetrization) according to the
given Young tableau. Not to overburden the subsequent discussion, for the time being we
will limit ourselves to k = 2 (higher values of N -ality will be considered in Sections 4.1
and 5). In this case there are two Young tableauxfull symmetrization and full antisymmetrization, Q{ij } and Q[ij ] . One can view the color source Qij as the source
composed of two fundamental heavy quarks, Qi and Qj . If Qi and Qj are placed close to
each other, a flux tube which connects them with two corresponding anti-quarks is called
2-string, see Fig. 1. In the general case of k fundamental quarks, we would be dealing with
the k-strings.
In Ref. [9] we presented arguments that at distances 1 in the transverse direction
the fundamental strings attract each other, that the lowest-energy state belongs to the antisymmetric string, and that the tension splitting between the symmetric quasi-stable string
and anti-symmetric string scales as 2 N 2 . Numerical evidence also suggests (see [1,
3]) that the symmetric quasi-stable 2-string has a larger tension than the anti-symmetric
2-string.
If one studies the probe charges in the reducible representation, as in Fig. 1, containing
a mixture of the symmetric and anti-symmetric quasi-stable 2-strings, it is quite obvious
that one will need the time interval
N 2 1 to resolve the tension splitting. In the
Euclidean language, at T
N 2 1 , the excited (symmetric) string contributions dies off.
A question which is more subtle and interesting is as follows: assume, one starts from
the pure (excited) state, i.e., one considers the Wilson loop for Q{ij } ; then what is its decay
time?
We want to show that the symmetric string does not decay into the anti-symmetric one
to any finite order in 1/N 2 . The decay rate is exponential. Indeed, in order to convert the
symmetric color representation into anti-symmetric one has to produce a pair of gluons.
This takes energy of the order of . However, the string is not entirely broken, rather it
is restructured, with the tension splitting 2 N 2 . To collect enough energy, the gluon
creation should take place not locally, but, rather at the interval of the length 1 N 2 .
This is then a typical tunneling process.
In fact, the decay rate can be found quasiclassically. The world sheet of the symmetric
string is shown in Fig. 6(a). Its decay proceeds via a bubble creation, see Fig. 6(b). The
world sheet of the symmetric string is two-dimensional false vacuum, while inside the
bubble we have a true vacuum, i.e., the surface spanned by the anti-symmetric string. The
tension differencein the false vacuum decay problem, the vacuum energy differenceis
E 2 N 2 , while the energy T of the bubble boundary (per unit length) is T . This
means that the thin wall approximation [18,19] is applicable, and the decay rate (per
80
Fig. 6. The world-sheet of a 2-string. (a) A purely symmetric string (denoted by horizontal lines). (b) The decay
into an anti-symmetric string (vertical lines) via an expanding bubble.
unit length of the string per unit time) is [18,19]

T 2
2
exp
2 exp N 2 ,
(16)
where is a positive constant of the order of unity. Once the true vacuum domain is created
through tunneling, it will expand in the real time pushing the boundaries (i.e., the positions
of the gluons responsible for the conversion) toward the string ends.
4. k-strings: saturation limit and the sine formula

The strings attached to the sources which have k fundamental indices (with no antifundamental), or vice versa are generically referred to as k-strings. The stable k-strings
(or bona fide k-strings) are those attached to the fully anti-symmetric sources. Others are
quasi-stable k-strings. In this section we will assume that k scales as N 1 at large N .
4.1. Saturation limit: generalities
In this section we will address the issue of the stable k-string tension in a limit which
we call saturation,
N ,
k ,
k
x
N
fixed.
(17)
We will combine the knowledge obtained from (i) symmetry properties; (ii) the saturation
limit; (iii) N limit with k fixed; and (iv) 1/N 2 expansion to prove that the general
formula for the k-string tension in the saturation limit is as follows:
k
= f f (x),
k
(18)
where the function f can be expanded in even powers of x,

f (x) = 1 + a1 x 2 + a2 x 4 + .
We then comment on how this peculiar k dependence can be physically understood.
(19)
81
The central element of the proof is the ZN symmetry, i.e., the fact that the bona fide
string tension k does not change under the replacements
k N k,
and k k + N.
(20)
The function f is dimensionless and, generally speaking, depends on two parameters, k

and N . Eq. (20) implies that k may enter only in a very special way, namely, through
powers of | sin x|. Thus, in general one can write
k =

f
c1 (N)| sin x| + c2 (N)| sin x|2 + c3 (N)| sin x|3 + ,
(21)
where c (N) are coefficients.

Now, we make use of the facts that (a) the saturation limit should be smooth, and (b)
for fixed k the expansion parameter is N 2 rather than N 1 . Then we conclude that the
coefficients c2+1 = O(N 1 ) while c2+2 = O(N 0 ) (here = 0, 1, 2, . . .). Moreover, from
the condition (iii) above it follows that c1 (N) = N .
Omitting terms vanishing in the saturation limit, we get
k =

f
N | sin x| + C3 | sin x|3 + ,
(22)
which is equivalent to Eq. (19) (here C3 and possibly C5 , C7 and so on are numerical
coefficients). Numerical lattice evidence suggests [20] that C3 is strongly suppressed.
Assuming that only C3 exists, and fitting the results for the k-strings, Del Debbio et al.
obtain [20]
C3 = 0.01 0.02.
At N = and k fixed Eq. (22) yields that k = k1 . Including the next-to-leading term
in the x-expansion we get

k2
k = f k 1 C 3 2 .
N
(23)
The term f C 3 k 3 /N 2 can be interpreted as the binding energy. The 1/N 2 factor is well
understood, see Ref. [9]. A challenging question is to understand why the next-to-leading
correction to k scales as k 3 . Alternatively, one can say that the binding energy per one
fundamental string in the k-string compound scales as k 2 . Experience based on numerous
discussions with our colleagues tells us that the first guess is, rather, that it is the first power
of k that must appear in the binding energy per one fundamental string.
To get a better idea, let us turn to the supersymmetric k-wall tension where the exact
result is known. In this case the k 3 -dependence is very general and is a consequence of
the ZN symmetry (see Appendix A). Apparently, the ZN symmetry is also instrumental
for k-strings. For the U(1) k-strings (i.e., the AbrikosovNielsenOlesen vortices), the k
behavior of k /k is expected to be different.
A physical picture underlying string dynamics that explains Eq. (23) is presented in
Section 7.
82
4.2. How exact is the sine formula (in the saturation limit)?
The tension of the k-string, k , a crucial parameter of the confinement dynamics, is
under intense scrutiny since the mid-1980s. Most frequently were discussed two competing
hypotheses: (a) the Casimir scaling and (b) the sine formula originally suggested by
Douglas and Shenker [21] (for extensive reviews and representative list of references see,
e.g., [22,23]). The Casimir scaling hypothesis was recently shown [9] to be inconsistent
with the large-N expansion. In the same paper we argued that the sine formula should
hold; in our terms the sine formula can be presented as follows:
sin x
.
(24)
x
It amounts to keeping only the first term in the general expansion (22). The suppression
of higher terms in this expansion cannot be inferred on general grounds alone; it is a
dynamical statement following from the model [9].
In the previous section we wrote down the most general expression for the k-string
tension (21), that is compatible with the ZN symmetry of the problem and 1/N 2 nature of
corrections. In the saturation limit the expression simplifies further to (22). It is tempting
to speculate that the first term, the sine, controls the dynamics in the saturation limit (17),
namely, that the k-string tension is exactly given by
f (x) =
k
.
(25)
N
While there is no proof that the relation (25) is exact, there are arguments suggesting
that it might be exact or, at least, present a good approximation (although we hasten to
add that a parameter controlling this approximation is yet to be discovered). Below we will
summarize theoretical evidence in favor of the above assertion.
The QCD string is not obviously a BPS object in N = 1 gluodynamics. Therefore, in
principle, it could get higher-order corrections in powers of sine, as in (22). The statement
that Eq. (25) is exact is equivalent to the statement that the QCD-string approaches a BPS
status in the saturation limit. Below we explain why a BPS QCD string should satisfy a
sine relation. An additional general argument is given in Appendix A.
The first hint in favor of the sine formula comes from the large-N analysis of N = 2
SYM theory softly broken to N = 1 SYM by the adjoint mass term m, due to Douglas
and Shenker [21]. In slightly broken N = 2 SYM the QCD string is BPS-saturated to the
leading order in m; therefore, no surprise that the sine formula k = mN sin k/N was
found. Details are as follows. In the softly broken N = 2 theory Douglas and Shenker
found that
k = 2 N sin
k mN 2 Mk,k+1 ,
where Mk,j are the masses of the BPS W bosons,
Mk,j = |mk mj |.
Here

(k 1)
k
sin
.
mk N sin
N
N
(26)
83
At large N , to the leading order in 1/N (keeping k/N fixed), one then arrives at the sine
formula,
Mk,k+1
sin
,
N
N
k mN sin
k
.
N
At order m2 the strings cease to be BPS, and, accordingly, the m2 correction to the string
tension was found [24] to defy the sine formula.
The second evidence comes from MQCD. Here again the sine formula is obtained [25],
without an a priori reason. Though the MQCD theory is not QCD, it is not clear why
the obtained result is the exact sine. A possible reason is that MQCD possesses more
symmetries than N = 1 SYM, but no such a symmetry is known at present.
The third hint is the derivation of the k-string tension via supergravity [26]. Here the
result is model-dependent. For the KlebanovStrassler background the sine formula was
found to be an excellent approximation, but not exact. It was found to be exact for the
MaldacenaNuez background. While both backgrounds are conjectured to be a dual
description of pure N = 1 gluodynamics in the far infrared, they have different ultraviolet
contents. For this reason, presumably, different results were obtained. Note that the above
arguments do not apply to pure N = 1 SYM theory. Moreover, only the first argument is
valid within field theory per se.
The last argument in favor of the sine formulaa field-theoretic oneis the relation
between the k-string tension and the k-wall tension in N = 1 gluodynamics [9]. It is
known that the BPS domain wall tensions in N = 1 gluodynamics are exactly given by
the formula
k
(27)
.
N
If one accepts the picture advocated in [9], that domain walls are built from a network of
QCD-strings connected by baryon junctions, one immediately arrives at the sine formula,
for a detailed discussion see Ref. [9]. In this picture, the QCD-string effectively becomes a
BPS object in the saturation limit. The suggestion of the walls built of the string network
is admittedly a model, albeit motivated by various arguments. The most crucial question
we see at the moment is whether the expansion parameter controlling the accuracy of this
model is numerical or related to some well-defined limits, such as large N , saturation limit,
and so on. In the absence of the analytic answer to this question one may resort to lattices,
see the remark after Eq. (22). While the saturation limit is clearly difficult to achieve on
the lattice, the effort is worthwhile.
Tk = N 2 3 sin
5. Decays of quasi-stable strings in the saturation limit

In Section 3.4 we have discussed the decays of the quasi-stable k-strings with a
relatively small k, i.e., k N 0 , for instance, those attached to the source Q{i1 i2 } . In this
section we will deal with a generic representation obtained from the probe color sources
of the type Qi1 ,i2 ,...,ik or Qi1 ,i2 ,...,ik , with k fundamental or anti-fundamental indices and
k N . This will correspond to x not close to zero or in the consideration of Section 4.
84
The reducible representation Qi1 ,i2 ,...,ik can be decomposed in a number of irreducible
representations. For k < N the number of different types of representations of SU(N )
equals to5 P (k), where P (k) = number of partitions of the integer k,
P (k)
exp( 2k/3)
,
4k 3
k
1.
(28)
All strings except that attached to the fully anti-symmetric sources are quasi-stable.
Their tensions are expected to cover densely the shaded area in Fig. 9. The tensions of
neighboring strings may be expected to differ by an exponentially small amount. The
decay rate of a given quasi-stable string into a certain close-lying string will be enormously
suppressed. The density of final states is exponentially high, however. We are interested in
inclusive probability, i.e., the probability of decay of a given quasi-stable string into any
state, with the possible subsequent cascading into the stable anti-symmetric string.
To do the estimate we will fuse possible final states into one effective final string
(say, anti-symmetric), ignoring the fact that the string tensions are spread practically
continuously. We will estimate the decay rate by evaluating a direct single-leap tunneling
into the effective final string.
In Section 3.4 we explained that the conversion of the symmetric two-index string into
the anti-symmetric one can be considered as a quasiclassical tunneling process which, in
turn, can be treated through bubble formation [19]. It was important that the tension
difference between the initial and final strings was suppressed by 1/N 2 while the mass
of the gluelumps produced was not suppressed by 1/N factors. Under these conditions the
approximation of thin domain wall [18,19] is applicable, the exponential factor in the decay
rate does not depend on dynamical details, and is fully determined by two parameters: the
tension difference and the gluelump mass.
Now we want to extend our estimates to cover the case of k indices where k is
specified in Eq. (17). As we will see, again the thin-wall bubble formation is the adequate
approximation. This method guarantees a correct evaluation of the exponent in the
exponential factor determining the inclusive decay rate of a given quasi-stable string. The
inclusive summation over the final state is a built-in feature of the method. Particular details
of the summation affect only the pre-exponential factor and are, thus, subleading.
A typical decay of a given k-string is characterized by a rearrangement of many indices;
in the saturation approximation the number of indices rearranged scales as N . Then the
tension deficit between the initial (quasi-stable) string and a typical decay product is of the
order of unity. This implies, in turn, that we must set E k2 N2 (for the definition
of E see Section 3.4).
The above rearrangement is accompanied by the production of k gluon lumps, so that
T k N. As a result, the decay rate of a generic quasi-stable k-string turns out to be
of the order of
gen 2 exp( N).
5 We are grateful to Haris Panagopoulos for pointing out to us this formula.
(29)
85
The suppression is still exponential but less severe than in Eq. (16). This enhancement of
the decay rate is in one-to-one correspondence with the enhancement of the binding energy
in the saturation limit k/N = const, N .
6. Oversaturated k-strings
For definiteness we will assume in this section the number of colors N to be even.
(Consideration of odd N can be carried out in parallel.) Let us define
N
(30)
.
2
The stable k-string tension grows as a function of k, reaching its maximum at k = k . It is
clear that for k > k it is energetically expedient to split an N 1-string cluster and annihilate
it, remaining with an ensemble of (N k) 1-strings. This is a tunneling process, however.
Quasi-stable strings with k = k + do exist ( > 0). Our task here is to evaluate their
lifetime with respect to the decay into stable strings with k = k .
For simplicity let us start from the probe source
k =
Q[i1 ik ]
with k = k + 1.
(31)
The string created by this source will be composed of (k + 1) 1-strings, and will decay
into the stable k-string with k = k 1. How does this decay occur?
This process is schematically depicted in Fig. 7 where we took N = 6. Two N -string
junctions (baryon vertices) are produced due to tunneling, converting a part of the string
between them from k = k + 1 into k = k 1. An estimate of the decay rate can be
obtained from exactly the same consideration as in Section 3.4. The tunneling occurs
through a bubble creation. The N -string junction scales as N; therefore, T N, see
Eq. (16). At the same time the gain in the energy density (i.e., E in Eq. (16)) is equal to
21 2 . As a result,

(k + 1) (k 1) exp N 2 ,
(32)
where is a positive numerical coefficient of the order of unity.
Fig. 7. The decay of a 4-string into a 2-string via creation of the baryon junctions in SU(6) gauge theory.
86
In fact, there are reasons to believe that the decay rate will be further strongly suppressed
by a large transverse size of the k-string (see Section 7), but we will ignore this effect for
the time being. It will be discussed elsewhere.
The same estimate (32) is valid for > 1, but not scaling with N .
7. Physical picture beyond the sine formula for stable k-strings

In Ref. [9] we suggested a dynamical model leading to the sine formula for the stable
k-string tensions. Irrespective of this model, the argument of Section 4.1 shows that the
expansion of the k-string tension has the form (23). While the 1/N 2 factor in the first
correction is perfectly clear, it is instructive to get a physical understanding of the k 3 scaling
of the 1/N 2 correction.
First of all let us note that at N
1 the 1-string interactions are weak, and a
quasiclassical picture is applicable. In order to reconcile the facts that both the flux of the
k-string and its tension (in the leading order) grow as k we have to conclude that the total
transverse area of the k-string scales as k at k
1, see Fig. 8. Thus, the k-string presents
an ensemble of loosely bound 1-strings, with k non-intersecting cores of 1-strings. The
interaction occurs only at the periphery, in the gaps. This is very similar to the structure of
molecules.
There are of the order of k gaps altogether, each having the area of the order of 2 .
Let us compare the chromoelectric fields at the periphery of the 1-string core in two cases:
(i) isolated 1-string; (ii) 1-string which is bound inside the k-string. It is clear that the
distortion of the field at the periphery in passing from (i) to (ii) is of the order of k/N . The
energy per one gap (which is proportional to the chromoelectric field squared) thus scales
as (k/N)2 . Given that the number of the gaps scales as k, we naturally obtain the k 3 /N 2
regime for the first correction in the k-string tension.
Within this picture the expansion parameter (k/N)2 , with the overall k factor, occurs
naturally, leading to the scaling parameter x introduced in Section 4.1.
The linear growth of the string cross section with k at large k is a prediction following
from our analysis.
We check the above picture in a toy supersymmetric model in Appendix A.
Fig. 8. A slice of a k-string.
87
Fig. 9. The tension defect. The two dashed curves in the upper side of the plot present oversaturated strings.
8. Tension defect vs. the scaling variable x

In this section we will discuss the binding energy of stable and quasi-stable k-strings in
the saturation limit. It is convenient to introduce a quantity analogous to the mass defect in
nuclear physics. We will call it tension defect Tk ,
Tk = 1
k
.
f Min{k, N k}
If the sine formula is correct for the stable k-strings (up to 1/N corrections), then
1
x
at x < /2,
Tk = 1 sin x
( x)1 at x > /2,
(33)
(34)
see Fig. 9. The presence of higher powers of sine in Eq. (22) may somewhat change the
precise form of the curve, but qualitatively its shape should remain as in Fig. 9. For stable
strings, the maximal tension defect is 1 (2/), achieved at x = /2.
The area between the curve (34) and the horizontal axis is covered by quasi-stable
strings. Indeed, for the given N -ality k we have P (k) distinct representation, see Eq. (28),
starting from the fully anti-symmetric, through mixed symmetries, up to fully symmetric.
P (k) is exponentially large. We expect that the tension splittings are roughly the same.
Then it is obvious that the string tensions of the quasi-stable strings are separated from
each other by exponentially small intervals.
Under the circumstances it would be appropriate to replace the discrete tension spacings
by a continuous distribution (T ; x), which for each given x gives the density of strings
per interval dT . Calculating (T ; x) is a very interesting dynamical problem.
9. More on the Casimir scaling

In Section 2.2 we mentioned that: (a) many lattice measurements detect quasi-stable
strings, with no hint on their breaking; and (b) observe the Casimir scaling in the ratio of
the tension of the given string to that of the fundamental string. For instance, in Ref. [6]
88
Fig. 10. Various flux tubes. (a) Two separated fundamental strings. (b) A short adjoint string. (c) A long adjoint
string.
ample data are presented regarding the adjoint and a number of other quasi-stable strings
in SU(3). No deviations from the Casimir scaling are seen withing the achieved accuracy.
According to the Casimir formula,
adj 9
=
f
4
(SU(3) Casimir).
(35)
The adjoint string is built of two fundamental ones, and at distances 1 they attract,
so that placing these two strings at an appropriate distance one must and can get the above
ratio less than 2. Even if we take the separation distance between the two 1-strings involved
very large, the ratio will equal 2. How can one get this ratio larger than 2?
Typical string sizes on the lattice are 2 fm, while the transverse size of the
fundamental string is 0.7 fm. As we have argued above, the transverse size of the
composite string must be even larger. Thus, the length/thickness ratio is in fact not large.
If one could measure the tension of the configuration depicted in Fig. 10(a), because of
the attraction, one would get less than 2f . The only way to get more than 2f is to force the
fundamental strings to overlap. Then, because of the repulsion of the overlapping fluxes,
(prior to the string decay), see the end of Section 2, one will get > 2f for the energy per
unit length.
j
This is precisely what happens if one considers the probe quark of the form Qi , see
j
l the overlap has to be
Fig. 10(b). Because of the insufficient separation between Qi and Q
k
substantial. For bona fide long strings (Fig. 10(c)) the overlap would be insignificant, and
will affect only the part of the potential which does not scale with L as L1 .
Still, the question remains: why a sausage-like configuration (Fig. 10(b)), which
develops at intermediate distances, is characterized by energy per unit length which follows
the Casimir formula?
Here we suggest a tentative semi-empiric answer. It is known since the 1970s [29]
that basic properties of low-lying quark mesons6 are to a large extent determined by
the gluon condensate. The coefficient in front of the gluon condensate is proportional
to the quadratic Casimir operator. Therefore, as long as the higher condensates are
not very important numerically, the masses of the lowest mesons will be related to
6 By low-lying we mean the lowest radial excitations in each channel with the given spin and parity.
89
the quadratic Casimir operators. These latter masses, in turn, are determined by short
strings.
If this explanation is correct, approximate Casimir scaling should take place at
intermediate distances when the overlap between constituents of the composite string is
significant.
10. Orientifold theories vs. supersymmetric gluodynamics

The orientifold gauge theory is obtained from SU(N ) supersymmetric gluodynamics by
twisting the color indices of the fermion field. The fermion sector of the orientifold gauge
theory consists of two Weyl spinors, in the color representations
and

.
One can combine them in one Dirac spinor. Recently it was shown [27] that in the
planar limit N = the orientifold gauge theory is equivalent to N = 1 supersymmetric
gluodynamics in its bosonic sector. (For a general introduction to the topic of planar
equivalence see Ref. [28].)
In connection with the k-string tensions, the above equivalence a priori might seem
rather puzzling. Indeed, the only dynamical fields in supersymmetric gluodynamics are
those belonging to the adjoint representationthey can screen no probe color charge with
a non-vanishing N -ality. At the same time, the anti-symmetric two-index quarks of the
orientifold theory do screen probe color charges with the even N -ality, 2, 4, 6, etc. This
seemingly results in distinct dynamical patterns.
In fact, there is no paradox. Let us remember that the above equivalence takes place
only in the limit N = . Non-planar effects are certainly different in the supersymmetric
and orientifold (parent and daughter) theories. Breaking the stringthats how the probe
color-charge screening is seen in the hadronic languageis a non-planar effect.
Thus, we expect then the Wilson loops of the parent and daughter theories to be related,
only provided their area, being large compared to 2 , does not scale with N . In this case,
as we already know from the previous sections (see also Section 11), there is no time for
screening to occur in either theories. We hasten to add, however, that for such contours
the accuracy of the measurement of the k-string tension is such that sub-leading in 1/N
corrections cannot be detected, and we will arrive at a rather trivial formula
k = 1 ,
(36)
where k is the k-string tension, 1 f and the parameter is introduced in Eq. (14).
Detecting 1/N corrections on the right-hand side of Eq. (36) would require contours with
area growing with N , which would invalidate the planar equivalence.
11. Implications of our results for lattice measurements

Throughout the paper we discussed the decays of quasi-stable strings into stable strings
within the Minkowskian picture. Lattice measurements are mainly performed within the
90
Euclidean approach. Most commonly lattice theorists measure Euclidean Wilson loops or
correlators of the Polyakov lines (see, e.g., Ref. [8]). It is natural to ask what is the impact
of our results on the Euclidean lattice measurements. Below we will discuss two typical
examples. Other examples can be treated in a similar manner.
11.1. Decay of the adjoint string from the Euclidean standpoint
Let us first discuss the adjoint string decay into a broken string. In Section 3.2 we
estimated the life-time of the adjoint string as 1 N 2 . When one studies the adjoint
string by calculating the Euclidean Wilson loop (in the adjoint representation), one arrive
at a two-term formula7
1
W eadj A + 2 CeP ,
(37)
N
where the numerical coefficient C is N independent but is very small (see the end of
Section 3.1 for the relevant discussion). The suppression of C is determined by geometrical
factors, the appropriate parameter being L. The numerical parameter in the second
term in Eq. (37) is of the order of . In fact, it represents the gluelump mass, which is
expected to be rather large, in the ballpark of 0.8 to 1 GeV. It is quite obvious that for the
asymptotically large contours the perimeter term (representing the adjoint string breaking)
always wins. However, given the pre-exponential suppression and a relative largeness of ,
at intermediate sizes there will be a competition between the area term and the perimeter
term.
The perimeter behavior is non-planar and, therefore, it is suppressed by 1/N 2 . In
addition, as was mentioned, there is an additional suppression due to the fact that the final
state producing the perimeter law is not a state with highly excited mesons but, rather, that
with the lowest-mass mesons. By equating the area term with the perimeter term we arrive
at an estimate for the (Euclidean time) duration of the process [14]
2
N
, C 1.
T 1 ln
(38)
C
The critical contour size needed to see the adjoint string breaking grows with N , albeit
rather slowly.
11.2. Decay (conversion) of the excited k-string from the Euclidean standpoint
The prime emphasis in this paper was on the decay of the excited k-strings. Let us
consider for definiteness 2-stringssymmetric and anti-symmetric. The Minkowskian
description of the decay of the symmetric string into the anti-symmetric one is presented in
Section 3.4. Here we will discuss the impact of our result on the Euclidean measurement
of the Wilson loop in the symmetric two-index representation of SU(N ).
The two-term formula (38) now has to be modified, since, after the breaking of the
symmetric string, the anti-symmetric one is formed and, therefore, both terms exhibit the
7 We are grateful to J. Greensite for an insightful discussion of this issue. See also Ref. [8]. %
91
area law,
W e N Ceanti-symm A + esymm A

2
2
2
= eanti-symm A Ce N + e A/N .
2
(39)
As was explained in Section 3.4, the transition into the stable (anti-symmetric) state
is exponentially suppressed, since it is a tunneling process. It is also suppressed by a
numerical (geometrical) factor C (C 1) which is N -independent. In the second line
of Eq. (39) we use the fact that the tension difference is
2
.
(40)
N2
To guarantee that the first term wins (i.e., the decay of the symmetric string into antisymmetric becomes visible) one needs the area

Acrit 2 N 4 (ln C)N 2 , C 1.
(41)
symm anti-symm =
The numerical value of the coefficient is unknown. Given the extremely steep N 4
dependence of Acrit it will be extremely difficult to measure the symmetric string
conversion into the anti-symmetric one within the existing lattice methods, unless is
abnormally small, which, a priori seems quite unlikely.
Eq. (40) implies that the area necessary for measuring symm (more exactly, the
split between symm and 2f ) must scale as 2 N 2 . This is especially obvious if
the measurement is carried out through the Wilson loop in the reducible two-index
2 .
representation, or Wfund
12. Discussion and conclusions
In this paper we started developing a systematic description of quasi-stable and stable
composite QCD strings (k-strings) based on the 1/N expansion. The paper consists of
three main parts. First, we considered various quasi-stable strings and found their decay
rates (per unit time per unit length). These results are summarized in Table 1. The picture
that we developed here is the physical Minkowskian one. We have viewed the quasi-stable
strings in their relation to quasi-stable mesons that can break due to non-planar interactions.
The key observation was that at large N , the probability is small and hence the life-time of
the string (and, correspondingly, excited resonances) is large.
Intuitively it is clear that the longevity of the quasi-stable strings at large N must imply
that detecting the breaking of such strings in the Euclidean lattice simulations must require
Table 1
Decay rates (per unit length per unit time) of quasi-stable strings
Fundamental string Adjoint string Excited k-strings Excited k-strings Over-saturated strings
(QCD with
(pure YM
(k small)
(saturation
dynamical quark)
or SYM)
limit)
= W/( L)
2 /N
2 /N 2
2 e N
2 e N
2 e N
92
large areas. In fact, this is the raison detre of the adjoint strings, of the symmetric strings
with the tension distinct from that of anti-symmetric, and so on. Numerous observations of
these quasi-stable strings at intermediate areas are reported in the lattice literature, with
very few (if at all) observations of the string breaking.
The central question iswhat does it mean, an intermediate area? To answer this
question we need to connect our Minkowskian picture to the Euclidean approach based
on lattice measurements using rectangular Wilson loops or Polyakov line correlators. In
Section 11our second main partwe gave estimates of the contour areas needed in
order to detect the string breaking (conversion). The N dependence of the critical area
for the adjoint string is logarithmic. In contrast, to see the conversion of excited k-strings
into anti-symmetric one would need to deal contour areas growing as powers of N .
Finally, the third main part of this paper is devoted to the tensions of the composite
strings. We summarized old and presented some new arguments in favor of the sine formula
for the tensions of the stable (fully anti-symmetric) k-strings. We made estimates regarding
the splittings between the tensions of the quasi-stable strings and the tension of the stable
string. We introduced the notions of the tension defect, saturation limit, and oversaturated
strings. We tried to explain why the Casimir scaling which in many instances predicts
R /f > , where is introduced in Eq. (14), is physically senseless, on the one hand,
and still, nevertheless, many existing lattice measurements seem to confirm the ratio of the
Casimir operators for R /f .
Now, we would like to say a few words on the actual QCD (N = 3) string. In the
actual world 1/N = 1/3, and one may wonder which aspects of (or conclusions from) our
analysis are applicable for such not-so-small value of the expansion parameter, if at all.
Besides, for the SU(3) color group, only one stringthe fundamental oneis truly stable.
We have seen that in some cases the relevant parameter is N 2 10. Although we do not
know numerical coefficients, it is reasonable to expect that the conclusions based on this
parameter will survive, at least, at a semi-quantitative level.
Acknowledgements
We thank J. Ambjorn, P. de Forcrand, S. Deldar, L. Del Debbio, J. Greensite, K. Konishi,
A. Kovner, C. Korthals Altes, B. Lucini, M. Lscher, H. Meyer, H. Panagopoulos, P. Rossi,
M. Teper, and E. Vicari for numerous enlightening and fruitful discussions. Stimulating
communications with D. Khmelnitskii and V. Shevchenko are gratefully acknowledged.
A.A. thanks Imperial College, Oxford University and Niels Bohr Institute where a part of
this work was carried out, for warm hospitality.
The work of M.S. is supported in part by DOE Grant DE-FG02-94ER408.
Appendix A
This appendix can be viewed as a supplement to Section 7 where we argued that
the k 3 /N 2 behavior of the k-string binding energy is natural. Here we illustrate this
93
argument by an exactly solvable model, which exhibits some general features of the kstring ensemble.
Consider supersymmetric WessZumino model with one complex scalar field and
one complex two-component spinor. In four dimensions this is the theory of one chiral
superfield with the minimal supersymmetry (N = 1, four supercharges), while in two
and three dimensions the supersymmetry is extended. While the concrete form of the
superpotential is unimportant, it is important to choose it to be ZN invariant. An
appropriate choice is

N
1
N+1
W=
(A.1)
.
2
N +1
Irrespective of the number of dimensions (two, three or four), the central charge is

2
N
2
(n + k) exp i
k .
Zn,n+k =
exp i
2
N
N
(A.2)
The mass of the BPS object built of k constituents is

k
(A.3)
.
N
This is an exact formula.
It is easy to see that the only ingredients important for the derivation are: (i) the
holomorphic nature of the central charge; and (ii) ZN symmetry. At N and k N
the mass defect is
Mk = N sin
2 k3
,
(A.4)
6 N2
i.e., the same as discussed in Section 7. In the case at hand, however, the problem is exactly
solvable, everything is known, and the question of the physical interpretation is easy.
The BPS object built from k constituents (i.e., k elementary solitons) has mass (A.4).
Each elementary soliton is separated from its neighbor by a gap, in which the field
deviates from its vacuum value only slightly. It is not difficult to see that the mass defect is
saturated in the inter-soliton gaps, due to a modification of the soliton tails far from their
cores. There are k such gaps, and in the gaps is typically k/N . Then the mass defect
produced by each gap scales as k 2 /N 2 . The overall mass defect scales as k k 2 /N 2 .
Mk kM1 =
References
[1] L. Del Debbio, M. Faber, J. Greensite, . Olejnik, Phys. Rev. D 53 (1996) 5891, hep-lat/9510028;
M. Faber, J. Greensite, . Olejnik, Phys. Rev. D 57 (1998) 2603, hep-lat/9710039.
[2] B. Lucini, M. Teper, Phys. Rev. D 64 (2001) 105019, hep-lat/0107007.
[3] L. Del Debbio, H. Panagopoulos, P. Rossi, E. Vicari, Phys. Rev. D 65 (2002) 021501, hep-th/0106185;
L. Del Debbio, H. Panagopoulos, P. Rossi, E. Vicari, JHEP 0201 (2002) 009, hep-th/0111090.
[4] V.I. Shevchenko, Y.A. Simonov, Phys. Rev. Lett. 85 (2000) 1811, hep-ph/0104135.
[5] S. Kratochvila, P. de Forcrand, String breaking with Wilson loops?, hep-lat/0209094;
S. Kratochvila, P. de Forcrand, Observing string breaking with Wilson loops, hep-lat/0306011.
94
[6] S. Deldar, Phys. Rev. D 62 (2000) 034509, hep-lat/9911008;

S. Deldar, The string tension of SU(3) representations, Ph.D. Thesis, Washington University, St. Louis.
[7] G.S. Bali, Phys. Rev. D 62 (2000) 114503, hep-lat/0006022.
[8] J. Greensite, The confinement problem in lattice gauge theory, hep-lat/0301023, Prog. Part. Nucl. Phys.,
submitted for publication.
[9] A. Armoni, M. Shifman, On k-string tensions, domain walls in N = 1 gluodynamics, hep-th/0304127, Nucl.
Phys. B, submitted for publication.
[10] 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.
[11] J.M. Maldacena, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002.
[12] D.J. Gross, H. Ooguri, Phys. Rev. D 58 (1998) 106002, hep-th/9805129.
[13] A. Casher, H. Neuberger, S. Nussinov, Phys. Rev. D 20 (1979) 179.
[14] L. Del Debbio, M. Faber, J. Greensite, Nucl. Phys. B 414 (1994) 594, hep-lat/9301021.
[15] V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 191 (1981) 301, see
Appendix B;
For further discussions see: C. Michael, Nucl. Phys. B 280 (1987) 13;
H.G. Dosch, O. Nachtmann, M. Rueter, String formation in the model of the stochastic vacuum and
consistency with low-energy theorems, hep-ph/9503386;
M. Shifman, ITEP Lectures on Particle Physics and Field Theory, Chapter 2, Section 11.
[16] K.G. Wilson, Phys. Rev. D 10 (1974) 2445.
[17] B. Blok, M.A. Shifman, D.X. Zhang, Phys. Rev. D 57 (1998) 2691, hep-ph/9709333;
B. Blok, M.A. Shifman, D.X. Zhang, Phys. Rev. D 59 (1999) 019901, Erratum;
See also M. Shifman, Quark-hadron duality, in: M. Shifman (Ed.), Boris Ioffe Festschrift at the Frontier of
Particle Physics, in: Handbook of QCD, Vol. 3, World Scientific, Singapore, 2001, p. 1447, hep-ph/0009131.
[18] I.Y. Kobzarev, L.B. Okun, M.B. Voloshin, Yad. Fiz. 20 (1974) 1229 (in Russian), Sov. J. Nucl. Phys. 20
(1975) 644.
[19] M.B. Voloshin, False vacuum decay, in: A. Zichichi (Ed.), Proceedings of the International School of
Subnuclear Physics, Erice, Italy, July 1995, in: Vacuum and Vacua: the Physics of Nothing, World Scientific,
Singapore, 1996, pp. 88124.
[20] L. Del Debbio, H. Panagopoulos, E. Vicari, private communication.
[21] M.R. Douglas, S.H. Shenker, Nucl. Phys. B 447 (1995) 271, hep-th/9503163.
[22] M.J. Strassler, Nucl. Phys. B (Proc. Suppl.) 73 (1999) 120, hep-lat/9810059.
[23] M.J. Strassler, Millenial messages for QCD from the superworld and from the string, in: M. Shifman (Ed.),
Boris Ioffe Festschrift at the Frontier of Particle Physics, in: Handbook of QCD, Vol. 3, World Scientific,
Singapore, 2001, p. 1859.
[24] R. Auzzi, K. Konishi, New J. Phys. 4 (2002) 59, hep-th/0205172.
[25] A. Hanany, M.J. Strassler, A. Zaffaroni, Nucl. Phys. B 513 (1998) 87, hep-th/9707244.
[26] C.P. Herzog, I.R. Klebanov, Phys. Lett. B 526 (2002) 388, hep-th/0111078.
[27] A. Armoni, M. Shifman, G. Veneziano, Exact results in non-supersymmetric large-N orientifold field
theories, hep-th/0302163, Nucl. Phys. B, submitted for publication.
[28] M.J. Strassler, On methods for extracting exact non-perturbative results in non-supersymmetric gauge
theories, hep-th/0104032.
[29] M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 147 (1979) 385;
M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 147 (1979) 448.

Softer hard scattering and noncommutative

gauge-string duality
Soo-Jong Rey a,b , Jung-Tay Yee c,d
a School of Natural Sciences, Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540, USA
b School of Physics & BK-21 Physics Division, Seoul National University, Seoul 151-747, South Korea
c Institute for Theoretical Physics, University of Amsterdam, Valckenierstraat 65,
1018 XE Amsterdam, The Netherlands

d School of Physics, Korea Institute for Advanced Study, Seoul 130-012, South Korea
Received 11 June 2003; received in revised form 31 July 2003; accepted 21 August 2003
Abstract
We study exclusive scattering of hadrons at high energy and fixed angle in (nonconformal)
noncommutative gauge theories. Via gauge-string duality, we show that the noncommutativity
renders the scattering soft, leading to exponential suppression. The result fits with the picture that,
in noncommutative gauge theory, fundamental partons consist of extremely soft constituents and
hadrons are made out of open Wilson lines.
In this paper, we pose the question: what are the fundamental degrees of freedom in
noncommutative gauge theory? and find an answer to it. Study of noncommutative gauge
theory [1,2], motivated prominently because the theory describes D-brane worldvolume
dynamics under B-field background in string theory, has brought us many surprises.
First, what one might counted as an ultraviolet (UV) effect is sometimes transmuted to
an infrared (IR) effect. Diagrammatically, origin of this so-called UVIR mixing [3] is
quite reminiscent of the channel duality between open and closed strings. Second, the
quantum dynamics is describable entirely in terms of open Wilson lines [4] (see also [5]).
The open Wilson lines [6] are nonlocal, gauge-invariant operators, and behave as a dipole
under a strong magnetic field, analogous to fundamental strings. Both features point to
Work supported in part by the KRF Overseas Research Grant, the KOSEF Interdisciplinary Research Grants,
the KOSEF Leading Scientist Grant, and the Stichting FOM.
E-mail addresses: sjrey@gravity.snu.ac.kr (S.-J. Rey), jungtay@science.uva.nl (J.-T. Yee).
doi:10.1016/j.nuclphysb.2003.08.030
96
S.-J. Rey, J.-T. Yee / Nuclear Physics B 671 (2003) 95102
the possibility that partons in the noncommutative gauge theory are not gluons but some
extended objects. We will attempt to answer the question posed by utilizing the gaugestring duality, and by examining exclusive processes such as high-energy, fixed-angle
scattering [7].
For commutative gauge theory, Polchinski and Strassler [8] recasted the fixed-angle
scattering of glueballs entirely in terms of gauge-string duality [9], and have shown that
the known power-like behavior [7] at high-energy regime is retractible. In terms of the
gauge-string duality, the glueball scattering in the gauge theory side is described by the
dilaton scattering in the bulk string theory side. At high-energy, as the latter process is
described by the VirasoroShapiro amplitude, one might naively expect that the scattering
process is exponentially suppressed.
The above expectation turns out incorrect [8]. By the holography principle in anti-de
Sitter space and the superposition principle in quantum mechanics, the scattering amplitude
in gauge theory is given by a coherent sum of the string scattering amplitude over the
bulk location where the scattering takes place. For this, because of the holographic UVIR
relation [10] in the gauge-string duality, one finds that the sum is dominated by the process
taking place near the boundary. Key feature here is that, because of the warp factor, the
local inertial momentum p is mapped to the conserved, gauge theory momentum p via
p m = (R/st u)pm , so, for any value of p, one can locate a bulk position u where p becomes
O(1/st ), leading to O(1) contribution of the string amplitude. This explains why the
glueball scattering as deduced via the gauge-string duality actually behaves power-like
at high-energy, and reveals point-like, hard parton nature of the fundamental degrees of
freedom in commutative gauge theory.
From the gauge theory viewpoint, one can portray the question posed from a different
angle: is it possible to find a field-theoretic deformation of commutative gauge theory
so that low-energy dynamics remains unmodified but high-energy parton content is
dramatically modified? The answer is affirmatively yes, and is achieved by defining
the theory on a noncommutative spacetime. Our result would be considered as a direct
evidence for the assertion.
For N = 4 supersymmetric gauge theory with noncommutativity on (23)-plane turned
on, the background fields in the dual string theory is given by [12]
2

2

1
R2 2
2
2 u
2
2
2
2
2
dx2 + dx3 + 2 du + R d5 ,
ds = st 2 dx0 + dx1 +
R
1 + a 4 u4
u
4

1

1

1
G0123u =
u st4 u4 1 + a 4 u4 ,
e2 = gst2 1 + a 4 u4 ,
gst
R
2
2

1 2st 4 4
1
NS
RR
B23
(1)
a u .
= st 4 u4 1 + a 4 u4 ,
C01
=
R
gst
Here, the curvature radius is given by Rst , where R is related to the string coupling
parameter and number of colors by R = (4gstN)1/4 . The parameter a 4 = 2 /R 4 sets
the noncommutative scale. In the coordinates adopted, the noncommutativity is denoted
by [x2 , x3 ] = i. As we are interested primarily in studying how the noncommutativity
affects high-energy fixed-angle scattering processes, we will restrict the kinematics so that
the scattering takes place in the noncommutative subspace. The momentum p in local
97
inertial coordinates and the conserved momentum p at the holographic boundary are then
related by the geometry Eq. (1) as
2
1/2
u2
R /
m
+ 2
pm .
p m = em pm =
(2)
st
u2
R /
In the commutative regime, u R/ , and the relation reduces to the same as in the
conformal gauge theory [10] p m (R/st u)pm . As such, for a high gauge-momentum p, it
was always possible to find a region in the bulk, where the string-momentum
p is of order
unity. On the other hand, in the noncommutative regime, u R/ , and the relation is
the gauge-momentum
given by p m (u/st R)p. Thus, for a fixed string-momentum p,
p is inversely proportional to u, precisely the opposite to the commutative regime.
So, what makes the story so interesting for noncommutative gauge theory is that, for the
very high gauge-momentum p, there is no u-region in the bulk where the inertial stringmomentum p can be made small enoughthe conversion factor in Eq. (2) is bounded
below. This will become the key reason why noncommutative gauge theory behaves stringlike at high energy. For the scattering at high enough energy, one cannot always adjust
u-location of the scattering center so that p remains small. The local inertial stringmomentum p has to grow in proportion to the gauge-momentum p. Thus, up to a finite
rescaling, high-energy processes in gauge theory corresponds to high-energy processes in
string theory as well.We will see momentarily that the string theory processes take place at
the location u R/ . Notice that, at this location, the proportionality factor in Eq. (2)
becomes independent of the t Hooft coupling parameter.
We now investigate the high-energy, fixed-angle limit of exclusive scalar glueball
scattering, and compare the result with the one for commutative gauge theory. In exclusive
processes, in- and out-states are created by gauge-invariant operators. In noncommutative
2 with
gauge theory, this requires adjoining the leading-twist glueball operator Fmn
open Wilson lines [6]. The simplest exclusive process would be the elastic two-body
scattering, and the scattering amplitude in gauge theory is obtainable by convoluting the
corresponding string amplitude over the wave function of the bulk state, e.g., the dilaton
for scalar glueballs. Thus,

ANCYM (p) =
du
0
S5
5 )
d5 g(10) Astring(p,
i (p i , u),
4
(3)
i=1
u) is the normalized wave-function of the dilaton field in the bulk, and g(10)
where (p,
is the volume factor of the ten-dimensional spacetime (in type IIB string frame). From
2 3
2 4
4 1
Eq. (1), g(10) = 10
st R u (1 + u /R ) . The string scattering process taking place in
the bulk is complicated as background fieldsmetric, dilaton, NSNS 2-form, and RR 2and 5-formsare turned on. In particular, contribution of the RR field background is of
the same order as that of the NSNS fields, since the former has strength of order 1/gst .
With currently available methods, analytic computation of string scattering amplitude
in such situation is impossible. Nevertheless, in the limit R , the background is
reduced to the one with flat spacetime, constant dilaton, and vanishing NSNS and
RR field strengths. Thus, at leading-order in 1/R, the string scattering amplitude can
be approximated by the VirasoroShapiro amplitude.
98
Consider a large but fixed gauge-momentum pm . When converted into the inertial
string-momentum p m , the conversion factor depends on the u-coordinate. At the minimum
of the
conversion factor, both the mean-value and the variance of the u-coordinate is set
by R/ . Thus, by letting the noncommutativity /2st 2 1 but R/ O(1),
one can always find a situation that the scattering process is pinned at a fixed u-position
and the wave function is spread out typically only over the string scale st . In such a
case, combined with the argument given in the previous paragraph, we can adopt the
VirasoroShapiro amplitude, which refers to the scattering process in a flat ten-dimensional
background, as the leading-order string scattering amplitude. We recall that the Virasoro
Shapiro amplitude is given by

14 2st s 14 2st t 14 2st u
2 6
= gs st
Astring(p)

Kc ,

1 + 14 2st s 1 + 14 2st t 1 + 14 2st u
where Kc is a kinematical factor. For high-energy and fixed-angle scattering, s and
t with s /t held fixed, and the string scattering amplitude exhibits a soft behavior

2

f ( ) st s
2 6 2 3
2
gs st st s sin exp
.
Astring(p)
(4)
2
Here, we have used the mass-shell condition s + t + u = 0, and defined the scattering angle as t = s sin2 2 and u = s cos2 2 . The parametric function f ( ) =
sin2 2 ln(sin2 2 ) cos2 2 ln(cos2 2 ) is a positive-definite function of in the entire range
[0, ].
What becomes modified dominantly by turning on the noncommutativity is the
The wave-function was
normalized wave-function of the bulk scalar field, (u, 5 ).
computed already in [6,11], and is given by (u, p) u2 ei 2 (+1) H (2)(, log(u)),

where H (2) denotes the second kind of the Mathieu function, and refers to so-called
Floquet coefficient (which depends implicitly on the momentum p). For foregoing analysis,
we have found it sufficient and more illuminating to understand asymptotic behavior of
the normalized wave-function, (u, 5). The wave function obeys the dilaton equation
MN e 2(u) (x, u, ) = 0. Fourier transforming with respect

of motion: M g(10) g(10)
N
5
5
to x m , taking S-wave mode on S5 , and rescaling as (u, p) = u 2 (u, p), we recast the
equation of motion into a form of the Schrdinger equation:
u2 (u, p) + K2 (u, p)(u, p) = 0,

4 2
R p
15 1
2
2
where K (u, p) =
(5)
+
+ |p| .
4 u2
u4
Classical turning point u = uc of Eq. (5) is determined by K(uc , p) = 0. At high-energy
limit
|p| we are interested in, it turns out the turning point is located at uc u =
R/ , and this is precisely the scale set by the noncommutativity, as seen in Eq. (2).
Asymptotic form of the wave function is obtainable via the WKB method. The result is
1/2 u5/2 e |p|(uu ) ,

u u ,
C(p)|p|

2 |1/2 u3/2 cos |pR 2 | 1 1 , u u ,
2C(p)|pR
(u, p)
(6)
u
u
4
1

16
5/2
3
h (p)Ai h (p)(u u ) ,
u u .
2 C(p)u
99
Here, C(p) is a momentum-dependent normalization constant, h(p) is defined as h(p)

[u K2 (u, p)]u=u , and Ai(x) refers to the Airy function. In the high-energy limit, the coefficient function h(p) is approximated as h(p) 4p2 5/2 /R. The normalization constant C(p) is determined by requiring that the wave-function
correctly normalized
obeys
00
commutation relations, and this results in the condition: 0 du g(10)
g(10) | |2 = 1. As
| |(u, p) in Eq. (6) is exponentially damped for large u and rapidly oscillatory for small u,
the normalization integral is dominated by the region near u u . Evaluating the integral
by the saddle-point method, we determined C(p) in terms of the coefficient function h(p)
1
as |C(p)h 6 (p)|2 (u3 /8st R 4 ).
It now remains to collect all parts and put them together. Re-expressing all momentum
dependence in terms of gauge-momentum Eq. (2), we obtain the string scattering amplitude
Eq. (4) in the high-energy limit as

Astring(p) gst2 6st (s)3 sin2
u2 u2
+
u2 u2
3

u2 u2
f ( ) u2 + u2 s
.
exp
2
(7)
With the above WKB results, the convolution integral Eq. (3) is not doable analytically,
so we shall evaluate it via the saddle-point method. For large and small values of u, the
integrand in Eq. (3) is exponentially suppressed, as we see from Eq. (7). The suppression is
dominantly due to the string scattering amplitude Eq. (7), and the normalized dilaton wave
function renders further, sub-leading suppression at large u. Thus, leading contribution to
the convolution integral Eq. (3) comes from the region around the saddle-point u = u .
Near the saddle-point, the Airy function yields O(1) contribution. Consequently, we find
that exponentially soft behavior dominates the gauge theory amplitude at high-energy:
1
s

(g 2 N) 2
(s)3 sin2 ef ()
for |p| ,
ANCYM (p) YM 2
(8)
N
where subdominant terms are omitted. Notice that, in Eq. (8), dependence on the string
scale st is cancelled out while dependence on the gauge theory coupling parameters is
identical to the commutative counterpart [8], as anticipated. As Eq. (8) refers to the leadingorder result at large noncommutativity, 2st , naive commutative limit of Eq. (8) is not
permitted. Redoing the saddle-point analysis at finite , however, one readily finds that
Eq. (8) interpolates smoothly to the commutative result of [8]: from Eq. (8), the exponential
suppression disappears, and the pre-exponential part is replaced by a constant (up to subleading corrections).
Eq. (8) is the main result of this paper. What is most remarkable of the result Eq. (8) is
that, in noncommutative gauge theory, the exclusive scattering amplitude is extremely soft
at high-energy, damped exponentially. In fact, the scattering amplitude is the same as that of
the string
theory except the role of the string scale st is now played by the noncommutative
scale . The stringy behavior is pronounced most for
the energymomentum exceeding
the value set by the noncommutativity scale: |p| 1/ . According to the gauge-string
duality, the dilaton scattering in the string theory is described by the glueball scattering
in the gauge theory. In noncommutative gauge theory, it is known that the interpolating
2 (x) but the one adjoined by the open Wilson
operator of the glueball is not just Fmn
line [6]. The size d of the open Wilson line grows linearly with the center-of-mass moment-
100
um p: d = p. It then implies that, in noncommutative gauge theory, stringy behavior

of the partons are manifested
by the open Wilson lines, whose size grows larger at stronger
noncommutativity: |d| . As such, hadrons adjoined with such large open Wilson
lines are interpretable as a sort of closed string modes emergent in noncommutative gauge
theory.
In [6], Gross et al. also reported soft behavior of high-energy scattering in noncommutative gauge theory. There, however, despite apparent similarity of the result with ours,
the exponential suppression bears no connection to the exponential suppression in the dual
string scattering amplitude. This is because their exponential suppression originates from
the bulk-to-boundary propagator (u u region of Eq. (6)) and was entirely within the
supergravity approximation, under which no stringy features would remain. In contrast,
in our result Eq. (8), exponential suppression originates dominantly from the full-fledged
string scattering amplitude Astring (dilaton wave function served mainly to provide the requisite kinematical parameters and coupling constant of the dual gauge theory). We emphasize importance of string theoretic computation in the bulk. Had we taken a supergravity
approximation and replaced Eq. (4) by the four-point dilaton scattering amplitude, the result would come out glossily different and incorrect at high-energy. For that reason, in
our result, correspondence of exponential suppression between the string theory and the
noncommutative gauge theory is exact.
Obviously, next task is to compute high-energy scattering amplitude directly from the
noncommutative gauge theory and to confirm the stringy behavior for the scattering among
four glueball operators involving open Wilson lines. In [13], such processes were studied,
but exclusively at leading order at weak coupling regime. An interesting issue for future
study is to establish a suitable interpolation from weak to strong coupling regime, and
relate to our result.
We showed explicitly, in terms of gauge-string duality, how closed string-like
behavior appears in noncommutative gauge theory. Though, from purely field theoretical
consideration, the appearance of closed string behavior in noncommutative field theories
is rather surprising, we have learned that exclusive high-energy processes offers an ideal
set-up for answering the reason why via the gauge-string duality. A further hint may be
learned from the full-figured ten-dimensional geometry of D3-brane before taking the nearhorizon limit. The geometry is claimed [14] to be dual to a UV-flow of the four-dimensional
N = 4 supersymmetric gauge theory once special irrelevant operators are turned on. If,
tentatively, the holographic boundary is placed at asymptotic infinity (which is the tendimensional flat spacetime), one obtains
the local inertial momentum p is related to the
gauge theory momentum p as p = 1 + 4st R 4 /u4 p, we see that the resulting relation
is qualitatively similar to Eq. (2): if we take large u 1/(Rst ), p is proportional to p.
This means that closed string modes do survive, and the holographic theory would behave
precisely like IIB string theory in ten dimensions.
Summarizing our results, we have demonstrated that noncommutative gauge theory exhibits dramatically modified elastic scattering behavior from commutative counterparts. At
energymomentum transfer well below the noncommutativity scale, both theories are indistinguishable. As energymomentum transfer is increased toward the noncommutativity
scale, both theories exhibit parton-like elastic scattering behavior. Underlying to the behav-
101
ior is attributable to the kinematics of conformal invariance at energy scale far above any
nonperturbative scales and not to underlying dynamics of partons. At energymomentum
transfer well above the noncommutativity scale, the two theories display dramatically different elastic scattering behaviorwhile commutative gauge theory continue exhibiting
parton-like power-law behavior, noncommutative gauge theory begins to expose nonlocal
and string-like constituents. As the behavior originates from departure of the bulk asymptotic geometry from anti-de Sitter and as the departure is precisely the feature underlying
emergence of dipole-like excitations at very high-energy scale, one may interpret quanta
created by the open Wilson lines as the fundamental constituents of the noncommutative
gauge theory.
More recently, Polchinski and Strassler [15] extended the gauge-string duality for
commutative gauge theory to deep inelastic scattering processes at large t Hooft coupling,
and found that evolution of the structure functions is more rapid than those at small t Hooft
coupling. This implies that there are no hard-momentum carrying partons inside hadrons
but only wee partons. In light of the result of this work, it would be quite interesting to
extend the study of [15] to noncommutative gauge theory and examine if high-energy
evolution of the structure functions is modified further than the wee-parton behavior
beyond the noncommutativity scale.
Acknowledgements
We thank D.J. Gross, J. Polchinski, L. Susskind, and C. Thorn for interesting
conversations during the QCD and String Theory Workshop at the Institute for Nuclear
Theory at Seattle. This work was carried out while S.J.R. was a Member at the Institute for
Advanced Study. He thanks the School of Natural Sciences for hospitality and the Funds
of Natural Sciences.
References
[1] A. Connes, M.R. Douglas, A. Schwartz, JHEP 9802 (1998) 003.
[2] N. Seiberg, E. Witten, JHEP 9909 (1999) 032.
[3] S. Minwalla, M. Van Raamsdonk, N. Seiberg, JHEP 0002 (2000) 020;
M. Van Raamsdonk, N. Seiberg, JHEP 0003 (2000) 035;
H. Liu, J. Michelson, Nucl. Phys. B 614 (2001) 279.
[4] Y. Kiem, et al., Phys. Rev. D 65 (2002) 026002;
Y. Kiem, et al., Eur. Phys. J. C 22 (2002) 781;
Y. Kiem, et al., Eur. Phys. J. C 22 (2002) 757;
Y. Kiem, et al., Nucl. Phys. B 641 (2002) 256;
Y. Kiem, et al., Phys. Rev. D 65 (2002) 046003.
[5] H. Liu, Nucl. Phys. B 614 (2001) 305;
A. Armoni, E. Lopez, Nucl. Phys. B 632 (2002) 240;
M. Van Raamsdonk, JHEP 0111 (2001) 006;
L. Jiang, E. Nicholson, Phys. Rev. D 65 (2002) 105020;
E. Nicholson, Phys. Rev. D 66 (2002) 105018.
[6] S.J. Rey, R. von Unge, Phys. Lett. B 499 (2001) 215;
S.R. Das, S.J. Rey, Nucl. Phys. B 590 (2000) 453;
102
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
D.J. Gross, et al., Adv. Theor. Math. Phys. 4 (2000) 893;

S.J. Rey, hep-th/0207108;
S.J. Rey, J. Korean Phys. Soc. 39 (2001) s527.
S.J. Brodsky, G.R. Farrar, Phys. Rev. D 11 (1975) 1309;
C.G. Callan Jr., D.J. Gross, Phys. Rev. D 11 (1975) 2905.
J. Polchinski, M.J. Strassler, Phys. Rev. Lett. 88 (2002) 031601.
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231.
S.J. Rey, J.T. Yee, Eur. Phys. J. C 22 (2001) 379;
J.M. Maldacena, Phys. Rev. Lett. 80 (1998) 4859;
S.J. Rey, S. Theisen, J.T. Yee, Nucl. Phys. B 527 (1998) 171.
S.R. Das, B. Ghosh, JHEP 0006 (2000) 043.
A. Hashimoto, N. Itzhaki, Phys. Lett. B 465 (1999) 142;
J.M. Maldacena, J.G. Russo, JHEP 9909 (1999) 025.
A. Dhar, Y. Kitazawa, JHEP 0102 (2001) 004.
K. Intriligator, Nucl. Phys. B 580 (2000) 99;
U.H. Danielsson, et al., JHEP 0005 (2000) 028.
J. Polchinski, M.J. Strassler, JHEP 0305 (2003) 012, hep-th/0209211.

Observing string breaking with Wilson loops

Slavo Kratochvila a , Philippe de Forcrand a,b
a Institute for Theoretical Physics, ETH Zrich, CH-8093 Zrich, Switzerland
b CERN, Theory Division, CH-1211 Genve 23, Switzerland
Received 20 June 2003; received in revised form 29 July 2003; accepted 11 August 2003
Abstract
An uncontroversial observation of adjoint string breaking is proposed, while measuring the static
potential from Wilson loops only. The overlap of the Wilson loop with the broken-string state is
small, but non-vanishing, so that the broken-string groundstate can be seen if the Wilson loop is long
enough. We demonstrate this in the context of the (2 + 1)d SU(2) adjoint static potential, using an
improved version of the LscherWeisz exponential variance reduction. To complete the picture we
perform the more usual multichannel analysis with two basis states, the unbroken-string state and the
broken-string state (two so-called gluelumps).
As by-products, we obtain the temperature-dependent static potential measured from Polyakov
loop correlations, and the fundamental SU(2) static potential with improved accuracy. Comparing
the latter with the adjoint potential, we see clear deviations from Casimir scaling.
PACS: 11.15.Ha; 12.38.Aw; 12.38.Gc
1. Motivation
Quarks are linearly confined inside hadrons by a force called the strong interaction.
Therefore, we cannot see single quarks: this is the basis of the stability of the matter we
are formed of. One can study this force by analysing the energy between a static colour
charge and a static anticharge. Unlike in the case of the electromagnetic force, this energy
is, as a consequence of linear confinement, squeezed into a long flux tube. This flux tube
is a string-like object. Therefore, one can ask whether this string actually breaks when
it reaches a certain length. This breaking of the string corresponds to the screening of
the static charges by a virtual matterantimatter pair created from that very energy stored
E-mail addresses: skratoch@itp.phys.ethz.ch (S. Kratochvila), forcrand@itp.phys.ethz.ch (P. de Forcrand).
doi:10.1016/j.nuclphysb.2003.08.014
104
S. Kratochvila, P. de Forcrand / Nuclear Physics B 671 (2003) 103132
in the string. The energy of the groundstate of the system, the so-called static potential,
completely changes its qualitative behaviour as a function of the distance between the two
static charges and can therefore be used to detect string breaking. There are two main
situations where string breaking can be studied: (i) when one deals with static charges in
the fundamental representation, which can only be screened by other fundamental particles,
such as dynamical quarks or fundamental Higgs fields; (ii) when one considers static
charges in the adjoint representation which can be screened by the gluons of the gauge
field. To avoid the simulation of costly dynamical quarks or of Higgs fields, we simply
consider here adjoint static charges. The bound state of a gluon and an adjoint static charge
is called a gluelump. Therefore, the breaking of the adjoint string leads to the creation of
two of these gluelumps.
Three approaches have been used to measure the static potential and study string
breaking:
Correlation of Polyakov loops, at finite temperature [1].
Multichannel ansatz (also known as variational ansatz) using two types of operators:
one for the string-like state and one for the broken-string state [2,3].
Wilson loops [4,5].
String breaking has been seen using the first two methods, but no clear signal has been
observed using the third one. The failure of the Wilson loop method seems to be due to
the poor overlap of the Wilson loop operator with the broken-string state. It has even been
speculated that this overlap is exactly zero [6]. This is why we have a closer look at this
problem, taking advantage of recent, improved techniques to measure long Wilson loops.
Preliminary results have been presented in [7].
In Section 2 we recall notions about the static potential and its relation to the Wilson
loop. In Section 3, we take the three methods into more detailed consideration, to be able
to discuss our results in all approaches. We explain the difficulties of measuring string
breaking using Wilson loops only. In Section 4, we present the techniques we applied,
such as adjoint smearing and improved exponential error reduction for Wilson loops. We
also discuss the methods used to analyse our data. For all three methods, results are shown
in Section 5, followed by the conclusion.
2. Static potential
We consider a pure SU(2) gauge system with a static charge and a static anticharge
separated by a distance R. Although we mainly focus on measuring the static potential
using Wilson loops only, we also consider the multichannel ansatz. Therefore, here we
discuss the issue of measuring the groundstate and excited-state energies in a more general
way.
The Hilbert space of the Hamiltonian is spanned by its orthonormal eigenbasis (n) (R).
The corresponding energies are E0 < E1 < , where E0 , the groundstate energy, is
called the static potential. If we knew these eigenstates, we could extract the energies by
105
measuring their time evolution:

(n) (R, T ) = (n) (R)T T (n) (R) = eEn (R)T ,
(1)

(n) (R) e En (R)
(n) (R)|. Since we do not
where T is the transfer operator, T =
n=0 |
know these eigenstates explicitly, let us consider an arbitrary linear combination (R). We
can expand this state in the eigenbasis

(n)

(R) =
(R)(R) (n) (R) .
(2)
n
The temporal evolution is given by

T

(R) (n) (R) 2 eEn (R)T
(R) =
(R)T
cn eEn (R)T .
n
(3)
Let us introduce a finite set of states i (R) which we know how to measure. We set

(R) =
ai (R)i (R).
(4)
To be explicit, let us expand Eq. (3) and define the correlation matrix Vij (R, T )
T

i (R)
Vij (R, T ) = j (R)T
so that
T

(R) =
(R)T
aj (R)ai (R)Vij (R, T ).
(5)
(6)
i,j
A lemma, which is proven in [8], states for the eigenvalues (n) (R, T ) of the correlation
matrix Vij

(n) (R, T ) = f (n) (R)eEn T 1 + O eT En ,
(7)
T
where f (n) (R) > 0 and En = min |En Em |. In general, the correction term cannot
m =n
be neglected and it will play an essential role in our study. The way to determine the
eigenvalues (n) (R, T ) and estimate the energies En is described in Section 4.3.
The finite set of states i (R) is created by applying some operators to the vacuum.
These states are chosen to model the expected ones, the unbroken-string and the brokenstring state. We can build a string-like state s (R) by a spatial line Ss (R) of links of length
R where s denotes the number of spatial smearing iterations (see Section 4.1), or a brokenstring state G (R) of length R, using G(R), a clover discretisation of F around the
two static charges (see Section 4.4). The correlation matrix is then
T

i (R) , i, j = s, G,
(8)

Ss Ss (R, T )
=
GSs (R, T )

Ss G(R, T )
.
GG(R, T )
(9)
106
T

s (R) = Ss Ss (R, T ) = W (R, T )
s (R)T
Fig. 1. The Wilson loop channel. Information about the groundstate energy, the so-called static potential, can be
extracted.
Using the diagonalisation procedure1 (see Section 4.3) one can reconstruct (n) (R, T )
(Eq. (1)) for small n, the lowest energies, hence the static potential, and the overlaps of the
string-like state and the broken-string state with the corresponding eigenstates. In previous
studies [2,3], this correlation matrix has been used in the multichannel ansatz to show
string breaking. We will confirm these results. But since the multichannel ansatz has been
objected to (see next section), we will also show that the full information about the static
potential can be extracted using only the GG-channel or only the Ss Ss (R, T ) channel (the
Wilson loop W (R, T ) (see Fig. 1)). Using the above notation, this corresponds to setting
as = 1, the other ai = 0. Eq. (3) is now given by

s (R)T T s (R) = Ss Ss (R, T ) = W (R, T )

s (R) (n) (R) 2 eEn (R)T =
=
(10)
cn eEn (R)T .
n
We can truncate this sum of exponentials at n = l if all the states we neglect are strongly
suppressed. For all k > l, we demand
ck (R) (Ek (R)El (R))Tmin(R)
e
(11)
cl (R)
which defines a Tmin (R) implicitly. In particular, we consider the following two-mass
ansatz (l = 1)
1
W (R, T ) = c0 eV (R)T + c1 eE1 (R)T ,
T > Tmin ,
(12)
where c0 and c1 are the overlaps of the state |s with the groundstate, which is the static
potential, and the first excited state, respectively.
3. The three approaches

The adjoint Wilson loop shows a good overlap with the unbroken-string but not with
the broken-string state. This is natural: the Wilson loop observable creates a static quark
antiquark pair together with a flux tube joining them. Therefore, the broken-string state,
1 Throughout this paper, we call the multichannel ansatz the approach to obtain eigen-energies and -states
using a multichannel approach, i.e., measuring correlations between states created by a finite set of operators. We
call the diagonalisation procedure the numerical procedure we use to extract information about eigen-energies
and -states from a given correlation matrix of type Eq. (5).
107
which consists of two isolated gluelumps, has poor overlap with the flux-tube state, hence
with the Wilson loop. For R < Rb the unbroken-string state is the groundstate, therefore
in Eq. (12) c0 is large compared to c1 . In this regime, additionally E1 > V by definition,
therefore the second exponential in Eq. (12) is negligible for T > Tmin . An important issue
is what happens for R larger than Rb . The broken-string state becomes the groundstate.
Its energy V (R) is smaller than the unbroken-string state energy E1 (R) (level crossing
has occurred). But for small T , the unbroken-string state still dominates over the brokenstring state because c0 c1 . Of course, this domination holds only up to a temporal extent
T = TP , where TP is the turning point defined by the equality of both terms on the righthand side of Eq. (12). For T > TP , the broken-string groundstate becomes visible in the
exponential decay of Eq. (12). The value of this TP is crucial to be able to detect string
breaking using Wilson loops only.
TP =
ln c1
c0
.
E1 (R) V (R)
(13)
Based on the heavy quark expansion, the strong coupling model of Ref. [9] leads to
estimates of the ratio cc10 and of TP . As an example we consider the distance R = 12a =
1.228(1) fm, whereas string breaking occurs at Rb 10a = 1.023(1) fm. (The lattice
spacing a is calculated in the beginning of Section 5.) The energy of the broken-string
state is V (R) 2 M(Qg), where M(Qg) = 1.03(2)a 1 is the mass of a gluelump
(measured independently in Section 4.4). Within this model, the ratio cc10 is around cc10
eM(Qg)R 2 105 . Therefore, the turning point is estimated to be
R=12a
TP(est) =
M(Qg)R
42a
E1 (R) 2M(Qg) R=12a
(14)
using the value E1 (R = 12a) = 2.34(1)a 1 of our results in advance. Wilson loops of size
12a 42a, i.e., about 1.2 fm 4.3 fm, or larger are needed to observe string breaking
according to this model.
Based on a topological argument, Ref. [6] even suggests that there may be no overlap
at all: c0 = 0 for R > Rb . Adding matter fields gives rise to the formation of holes in the
world sheet of the Wilson loop, reflecting pair creation. The average hole size leads to two
different phases of the world sheet. In the normal phase, holes are small and the Wilson
loop still fulfills an area law, W (R, T ) e RT , where is the string tension renormalised
by the small holes. This phase corresponds to the unbroken-string case and a screening of
the static charges cannot be observed. The other phase is called the tearing phase, where
holes of arbitrarily large size can be formed. As a consequence, the Wilson loop follows
a perimeter law, W (R, T ) ecT . This corresponds to the broken-string state, since the
groundstate energy remains constant. [6] speculates that the Wilson loop is in the normal
phase, and analyticity prevents it from changing phase, so that string breaking cannot be
seen. However, there are limits to this argument, which are discussed by the same group
in [10] in the context of a Z2 gauge-Higgs system. Nevertheless, this conjecture might
explain why, for instance, the authors of [5] could not observe string breaking even at a
distance R 2Rb . Note however, that their temporal extent was T 3.
108
A simple argument gives a necessary condition for the observation of string breaking if
we only use ordinary (non-smeared) Wilson loops W (R, T ). If R > Rb , where Rb is the
string breaking distance, but T < Rb , we can relabel the R and T directions, so that now
R < Rb and T > Rb . In that case, string breaking is not visible since the new spatial extent
is R < Rb , and the Wilson loop must still fulfill the area law. Therefore, both sides R and
T should be larger than Rb . Replacing ordinary Wilson loops by spatially smeared ones
may relax this requirement somewhat.
As already mentioned in the previous section, an improved determination of the static
potential can be achieved by a variational superposition of the unbroken-string and the
broken-string states. In other words, the multichannel ansatz enlarges the operator space
to a multichannel approach. The unbroken-string state is realised via the flux tube of the
Wilson loop, the broken-string state is modelled by considering two gluelumpsseparated
static charges surrounded by gluons which screen the interior colour charge. We end up
with the two-channel transfer matrix Vij of Eq. (9).
Nevertheless, this method has been criticised [11]. One may claim that string breaking is
built into the multichannel ansatz due to the explicit inclusion of both states. Moreover, the
behaviour in the continuum limit ( ) must be considered. If the off-diagonal element
Ss G(R, T ) = GSs (R, T ) is zero in this limit, string breaking does not actually happen: the
Wilson loop does not communicate with the broken-string state; the eigenvalues merely
cross each other at R = Rb . It is only if the off-diagonal elements are different from zero,
that the eigenstates are a mixture of both states. A small overlap at different -values has
indeed been confirmed by the mixing analysis in the multichannel ansatz, as can be seen in
Refs. [2,3], which deal with this issue. However, whether this overlap vanishes or not for
a 0 still has to be checked in detail. Here, we do not address this question of continuum
extrapolation due to technical difficulties: the method we use is more efficient at smaller .
Therefore we consider only one -value, which we choose as small as possible while
staying in the scaling region.
The Wilson loop method and the multichannel ansatz work at zero temperature, where
the question to be answered is: what is the groundstate of a system with two static
charges? In the context of the Polyakov loop method we have contributions of temperaturedependent effects, and the question to be answered is different: what is the free energy of a
system with two static charges coupled to a heatbath? Nevertheless, it is an interesting issue
and, as a by-product, we can also measure the correlations of Polyakov loops, which are
t 1 (t )
in our case of adjoint charges P (x) = 13 Tr N
i=0 Uadj (x, i). This results in a temperaturedependent potential VT (R) [12]:

P (0)P (R) = eVT (R)/T .
(15)
The flattening of the potential VT (R) at large R has been seen in QCD with
dynamical fermions, although in practice only at temperatures close to or above the critical
deconfinement temperature. Unlike the multichannel ansatz, this method builds in no
prejudices about the structure of the groundstate wave function.
109
4. Technical details
We are considering Wilson loops in the adjoint representation (for a definition see the
following subsection). The choice of the representation has a direct impact on the static
potential, which is at lowest order in perturbation theory in 2 + 1 dimensions [13]

g02
ln R + O g04 / .
(16)
2
C2 is the value of the quadratic Casimir operator of the representation of the static charges,
i.e.,
VP (R) C2
fundamental representation: C2 (F )122 = 34 122 ;

adjoint representation: C2 (A)133 = 2133 .
The important point is that, in the regime of perturbation theory, i.e., at small distances
R, at lowest order, the adjoint static potential Vadj (R) differs from the fundamental static
potential Vfund(R) by a factor 83 . Assuming, for simplicity, that the ratio remains the same
at larger R (this issue is discussed in Section 5.5), the adjoint potential is much more
difficult to measure than the fundamental one: the Wilson loop is W (R, T ) eV (R)T ,
therefore the signal decreases much faster with R or T in the adjoint representation. This
is the price to pay if we consider adjoint static charges instead of fundamental static charges
in order to avoid the simulation of dynamical quarks. We need a sophisticated method of
exponential error reduction (see Section 4.2) to detect very small signals: the magnitude of
each measured Wilson loop is Wi (R, T ) O(1) while the average to be detected, as it will
turn out, is W (R, T ) O(1040). Using ordinary methods, 1080 measurements would be
needed.
4.1. Adjoint smearing
It is very desirable to reduce contributions from excited states (n =0) to the Wilson loop
W (R, T ): the turning point Euclidean time TP (Eq. (13)) is reduced, and the accuracy on
the groundstate potential is greatly improved. To this end, we smear adjoint links spatially.
In SU(2), a matrix Ufund in the fundamental representation can be mapped onto a 3 3
real link matrix Uadj in the adjoint representation by
1

Uadj (Ufund) = li Ufund,ij j k Ufund,kl
(17)
,
2
where the are the Pauli matrices; , = 1, . . , 3; i, j, k, l = 1, . . , 2.
The smearing can be done by setting the new adjoint link as the SO(3) projection of the
old link plus a weighted sum of the spatial staples:

4

Adjoint Spatial Staplei ,
Uadj (x) = ProjSO(3) Uadj (x) +
(18)
i=1
where we choose = 0.5. We consider three different smearing levels: 15, 30 and 60
iterations of Eq. (18). For details and usage, see Section 4.2.3. We define our projection
110
of an arbitrary matrix A onto A SO(3) by maximising Tr A A. This can be performed

using the singular value decomposition: every M N matrix A (M N ) can be written
as the product of a column-orthogonal M N matrix U , a diagonal N N matrix with
positive or zero elements (the singular values), and an orthogonal N N matrix V .
A = U V .
(19)
Since in our case M = N = 3, both U and V are elements of SO(3), and we get the
projection of A onto SO(3) by
= U V .
A = ProjSO(3)(A)
(20)
For completeness we also describe the fundamental smearing procedure used in

Section 4.4. Instead of considering adjoint links, we use the same method but in the
fundamental representation:

4

Spatial Staplei ,
U (x) = ProjSU(2) U (x) +
(21)
i=1
where we take = 0.5 and project back onto SU(2) by

U (x)
U (x) = ProjSU(2) U (x) =
.

det U (x)
(22)
4.2. Exponential error reduction

An adjoint R by T Wilson loop in the (x, t)-plane consists of two adjoint spatial transporters of length R, which we call L(0) and L(T ) , and two temporal
1 (t )
sides of length T , which we write explicitly as U(0) = Ti=0
Uadj (x, y, t + i) and
0
(t )
U(R) = j =T 1 Uadj (x + R, y, t + j ).2 The expectation value of the Wilson loop can be
written as

1
[DU ] Tr U(0) L(T ) U(R)L(0) eS[U ] .
W (R, T ) =
(23)
Z
An exponential error reduction is possible because of the locality of the action, which in
our case is the Wilson plaquette action. The main idea is to write the average of a product
as a product of averages.
4.2.1. Multihit method
One possibility to reduce the variance of the Wilson loop observable, is to reduce the
variance of a single link contribution. The multihit method [14] takes the average of many
samples of one particular link with all other links held fixed. As we will show now, all the
(t )
in Eq. (23) can be treated like this for Wilson loops with a spatial
temporal links3 Uadj,k
2 To make clearer the distinction between the two temporal sides of the Wilson loop, we use separate indices
i and j for the two running time-coordinates.
3 The situation is different considering spatial links: since we have smeared them spatially, we cannot hold
all the other links fixed.
111

extent R 2. In the first step we split the action as S = S [U ] + k Sk [Uk ]. Sk [Uk ] is the
local part of the action that contains the fundamental link Uk corresponding to Uadj,k . The
multihit method can be applied also to Uadj,l if Sk [Uk ] does not depend on Ul for k = l.
Therefore, since we use the Wilson plaquette action, this condition is satisfied if R 2.
We can then apply the multihit method on all time-like links and Eq. (23) can be rewritten
as
T 1

1

dUi Uadj,i
eSi [Ui ] L(T )
[DU ] Tr
Z
i=0
0

Sj [Uj ]
L(0)eS [U ]
dUj Uadj,j e
j =T 1
1
=
Z

0

T 1

[DU ] Tr
U adj,i L(T )
U adj,j L(0)eS[U ] ,
i=0
(24)
j =T 1
where the multihit average is given by the one-link integral

dUi Uadj,i eSi [Ui ]
U adj,i =
.

dUi eSi [Ui ]
(25)
In simple cases, as in pure SU(2), the multihit average can even be calculated analytically.
Namely, Si [Ui ] = 12 Tr Ui W , where W is the sum of the four (in 3d) fundamental
staples, and
I3 (w)
,
U adj,i = W adj
(26)
I1 (w)
where w = det W , W = W/w is the projection of the staple-sum onto SU(2) and W adj
represents W in the adjoint representation via Eq. (18).
Since the variance of each time-like link entering W (R, T ) is reduced, the variance
reduction in W (R, T ) is exponential in T . The coefficients have been estimated in [15].
For fundamental Wilson loops, the reduction is about (0.82 )T = e0.45 T , and for adjoint
loops about (0.52)T = e1.39 T .
4.2.2. Multilevel method
Although the multihit method was revolutionary in 1983, the error reduction was not
strong enough to enlarge measurable Wilson loops to temporal extents T sufficient to be
able to observe string breaking. In Section 3, we suggested a heuristic argument, that T
should be at least as large as the string breaking distance, which in our case is at Rb 10a.
The heavy quark expansion even results in an estimation of TP(est) 42a (see Eq. (14)).
Lscher and Weisz generalise the multihit method from single time-like links to link
link correlators T(R, t = na) [16]. Using our notation from above,
(t )
(t )
T(R, t = na) = Uadj
(x, y, t + i = na) Uadj
(x + R, y, t + j = na) .
(27)
112
A single Wilson loop Ws (R, T ) can be written, using the tensor multiplication defined by

T(R, na)T R, (n + 1)a = T(R, na) < T R, (n + 1)a <
(28)
as

Ws (R, T ) = L(0) T(R, 0)T(R, 1a) . . . T R, (T 1)a L(T ) .
(29)
Just like in the multihit method where we considered the average links U i , here time
slice expectation values of a linklink-correlator T(R, na), denoted by [ ], can be obtained
by sampling over the corresponding sublattice which is in this case the time-slice at time
t = na. This sublattice can be studied independently of the surrounding lattice provided
the spatial link variables at the boundaries are held fixed. This is a consequence of the
time-locality of the gauge action. Using a self-explanatory notation

1
[DU ]sub T(R, na)eS[U ]sub ,
T(R, na)
(30)
Zsub
the expectation value of the Wilson loop can be written in the form

W (R, T ) = L(0) T(R, 0) T(R, 1a) T R, (T 1)a L(T ) . (31)
The restriction of fixed spatial links at the boundaries becomes manifest in the fact that
only the temporal links on the time-slice at time t = na are allowed to be updated when
evaluating [ ] with Monte Carlo methods. For our project, this is a severe obstacle which
limits the efficiency of the exponential error reduction. It can be circumvented by using a
hierarchical scheme based on identities like

T(R, na) T R, (n + 1)a

= T(R, na)T R, (n + 1)a = T(R, na) T R, (n + 1)a .
(32)
The impact on the Monte Carlo method is, that we are also allowed to sample over
spatial links on the time-slice (n + 1)a since the boundary of the so-called second-level
average [ ] now consists of the spatial links in the time-slices na and (n + 2)a. The
possibility to use this two-level scheme allows us to measure long Wilson loops almost up
to the desired accuracy. We actually implement the following three-level scheme illustrated
in Fig. 2, where T is restricted to be a multiple of 4:

T(R, na) T R, (n + 1)a
W (R, T ) = L(0)
"
!

T R, (n + 2)a T R, (n + 3)a
(33)
L(T ) .
As a result of a rough optimisation process in the error reduction, based on minimising

the CPU time versus the error, we choose the following parameters (see also [17]): the
innermost averages [T(R, na)] are calculated from 10 sets of time-like Multihit-links,
heatbath and overrelaxation
each obtained after n1 = 10 updates. The updates alternate

steps in the proportion 1 : 4. The second-level averages, [T(R, na)][T(R, (n + 1)a)] are
calculated from n2 = 160 averages of [T(R, na)] and [T(R, (n + 1)a)], separated by 200
113
Fig. 2. Hierarchical scheme. Using the three-level method as described in the text decreases the statistical error
exponentially. While a one-level approach only allows to sample over the temporal links, a multi-level approach
also makes it possible to update the interior spatial links, which improves the error reduction.
updates of the spatial links on time-slice (n + 1)a. Finally, the third-level averages are
calculated from n3 = 165 second-level averages separated by 200 updates of the spatial
links on time-slice (n + 2)a.
n1 = 10 seems rather small, butcan be explained using the confinement-deconfinement
phase transition. For the ratio Tc / [18] finds a value
Tc
= 1.065(6)
(34)
with periodic boundary conditions. The corresponding critical temporal extent (in lattice
units) for our coupling = 6.0 is then
(p)
Nc
a( = 6.0)
3.76.
Tc
(35)
As a rule of thumb, one can estimate that a slice with fixed boundary conditions with a
(p)
temporal extent of 12 Nc will be confined [19]. On the first level, we deal with time
slices of extent 1, a high temperature regime, corresponding to the deconfined phase. Then
the link-link-correlator has a large finite value, even at large R, and its determination is
easy: n1 = 10 multihit averages are sufficient, and increasing n1 further does not reduce
the final error as 1/ n1 . On the next level, time slices of extent 2 are in the confined phase.
The signal then decreases exponentially at large R. We adjust n2 so that the signal to noise
ratio is about 1 for the distance R = 13a which we are interested in. On the third level, the
situation becomes more complicated and the best choice of all three parameters can only
be found using optimisation, to minimise CPU time versus error. We find that a three-level
scheme is more efficient than a one- or two-level scheme.
In Eq. (33), we have a product of N tensors, the three-level link-link-correlators with
a temporal extent 4a. Just like in the case of the multihit method, the variance of each
tensor is reduced by a factor (R). Thus, the variance of the Wilson loop average can be
reduced by as much as (R)N for an effort N(R)2 . Variance reduction exponential in
N = T /4 is achieved.
4.2.3. Improved spatial transporter
Using the above technique, we are able to reduce exponentially the error coming from
the temporal links of the Wilson loop. But there is still the intrinsic noise, coming from
114
Fig.
3. The improved spatial transporters. After each calculation of second-level averages

[T(R, na)][T(R, (n + 1)a)] , we form staple-shaped transporter including smeared spatial links at the
time-slice (n + 2)a.
the frozen spatial links at time-slices T = 0 mod 4, which is relatively large [17]. How
can we decrease it? This is what we do: to provide additional error reduction also in the
spatial transporter, we replace the spatial transporters L(0) and L(T ) with stapleshaped transporters, which are constructed in the following way (see Fig. 3):
(i) After each calculation of second-level averages, denoted as | |, we form the smeared
, at time-slice (n + 2)a. (ii) We multiply them with the second-level
spatial links,
,
. (iii) This procedure is repeated
averages to obtain the staple-shaped transporter
n3 = 165 times, each time after updating the spatial links on time-slice [(n + 2)a] during
the calculation of the third-level average. These error-reduced staple-shaped transporters
replace the naive spatial transporters L(0) and L(T ) and by contracting them with
none, one, two, etc., third-level-averages one obtains Wilson loops at a fixed R with
temporal extent T = 4, 8, 12, etc.
4.3. Diagonalisation procedure
Given a set of states i and the correlation matrix Vij (R, T ) as introduced in
Section 2, one can approximate the eigenstates correlation matrix defined in Eq. (1),
get information on the eigenstates (n) and extract the lowest energies En using a
diagonalisation procedure.
For a given separation R, the correlation matrix is defined in Eq. (5) as
T

i (R) .
(36)
A naive determination of the lowest energies En is obtained by looking for a plateau in
(n) (R,T )
the ratio of eigenvalues (n)
of Vij (R, T ) for increasing T (see Eq. (7)):
(R,T +1)
En (R, T ) = lim ln
T
(n) (R, T )
.
+ 1)
(n) (R, T
(37)
This simple method works very well, especially in the multichannel ansatz. Nevertheless,
we want to increase the signal of the desired state as much as possible. For a finite basis, for
small T , the eigenstates change with T . To improve T -convergence, we apply variational
115
diagonalisation which consists of solving the generalised eigenvalue problem

(n)
(n)
Vij (R, T )vj (R, T , T0 ) = (n) (R, T , T0 )Vij (R, T0 )vj (R, T , T0 ),
T > T0 . (38)
The eigenvalues (n) (R, T , T0 ) also fulfill Eq. (7) but their coefficients f (n) (R) are
enhanced by construction, compared to the previous ones. Once the eigenvectors v(n) and
eigenvalues (n) are known, one may approximate the eigenstates as a superposition of the
operator states using
(n)
(n) (R, T ) = N (n) (R, T , T0 )
vj (R, T , T0 )j (R, T )
aj(n) (R, T , T0 )j (R, T ),
(39)
where the constants N (n) are chosen such that the (n) are normalised to unity.
It is well known from the literature, that the groundstate energy can be extracted nicely
using a single-mass ansatz even in the broken-string regime R > Rb , starting with T0 = 0,
if broken-string state operators are included in the multichannel (variational) basis. Indeed,
we can confirm this observation. The situation is different when one uses a pure Wilson
loop operator basis, at R > Rb . Because the overlap of the Wilson loop with the brokenstring state is so weak, one must choose T0 very large to ensure that the lowest-lying
eigenstates at T0 and T are both broken-string states. Then, the high sensitivity of Eq. (38)
to statistical noise renders the analysis delicate: the matrix Vij (R, T0 ) may not be invertible.
As a trade-off, we choose a small value of T0 , T0 = 4a, but must use a two-mass ansatz to
account for all data points.
4.4. Gluelumps
In order to fully implement the multichannel ansatz, we must consider the two-gluelump
correlator, which is in the notation of Eq. (9)
= GG(R, T ).
(40)
The broken-string state can be described by the presence of two gluelumps, each formed by
the coupling of adjoint glue to an adjoint static charge [15]. It is of interest to measure the
mass and the correlator of gluelumps. A simple way to probe the gluon field distribution,
denoted as , around the adjoint static charge is the so-called clover-discretisation of F ,
denoted as Cm, = C1, , . . . , C4, in Fig. 4.
1
Smeared Spatial Plaquette (m, , )i
4
4
Cm, =
i=1
0
= Cm, 12x2
+ iCm,
.
(41)
The anti-Hermitian part of the clover Cm approximates F

1
Cm, Cm, = Cm,

= g0 a 2 F + O a 4 ,
2i
(42)
116
Fig. 4. Interacting gluelumps at distance R. Four clovers are stuck on a link-link-correlator tensor of temporal
extent T .
where the are the Pauli matrices and we average over the four oriented spatial plaquettes
that share a corner with one end of the time-like line. We choose the orientation of Fig. 4
(all four staples clockwise) to obtain the lowest-lying gluelump mass [5]. The plaquettes
are built using fundamentally smeared links (as described in Eq. (21)).
To measure the gluelump mass, one considers only one gluelump, e.g., the left side
(C1 C3 ) in Fig. 4. The gauge-invariant operator CC(T ) is an adjoint time-like line of
length T using Eq. (17), A (T ), which is coupled at the two ends to the clovers (C1 and
C3 ).
CC(T ) = C1 (0)A (T )C3 (T ).
(43)
The adjoint time-like line can be measured using the multihit method or by applying the
multilevel idea to this particular problem. The mass M(Qg) can then be extracted using
CC(T ) eM(Qg)T ,
(44)
The overlap with the lowest-lying state is enhanced by using smeared links to build the
clover observable Eq. (41). We would like to mention at this point that the gluelump
mass by itself has no real physical meaning, since it contains a UV-divergence in the
continuum limit due to the self-energy of the time-like links. Only the difference between
this divergent mass and another similarly divergent one, like the static potential, makes
physical sense. As a consequence, the string breaking distance Rb remains constant in
physical units, while the energy of the level-crossing diverges as .
To measure the correlation of two gluelumps, one has to consider the full object in
Fig. 4. The four clovers C1 , . . . , C4 are measured as described above. The correlation
GG(R, T ) of two gluelumps separated by a distance R can be measured by contracting
the four clovers to the link-link-correlator tensors with temporal extent T , T(R, T ). The
same tensors, obtained with the multilevel algorithm and used for Wilson loops, are also
used here.
GG(R, T ) = C2 (0)C1 (0)T (R, T )C3 (T )C4 (T ),
(45)
GG(R, T ) can be used in two ways: on the one hand, as mentioned in the beginning of
this section, it is a contribution to the multichannel matrix; on the other hand, we can try
117
to extract, from it alone, the energies of the unbroken-string and of the broken-string states
since presumably the two-gluelump correlator has projection on both states.
GG(R, T ) g0 eV (R)T + g1 eE1 (R)T ,
T > Tmin ,
(46)
where V (R) is the static potential and E1 (R) the first excited state energy. The operator has
obviously a good overlap with the broken-string state, but a poor one with the unbrokenstring state. This situation mirrors that of the Wilson loop, described in the beginning
of Section 3. For R < Rb , the unbroken-string state is the groundstate. Therefore g0 is
expected to be small compared to g1 , and the first excited state, with the energy of two
gluelumps, is dominating for small T . Since the groundstate will be visible for large
temporal extents only, we need a two-mass ansatz to describe the correlator. At large
distances, R Rb , the broken-string state is the groundstate and also the dominating one
(g0 g1 ), therefore a single-mass g0 eV (R)T will suffice.
In the case of Wilson loops, we attach improved spatial transporter to the link-linkcorrelators. Here, we use non-improved clovers for simplicity. Therefore, we have more
statistical noise, which makes it difficult to extract the groundstate, if the turning point is
large. This is the case, for distances R close to but below the string breaking distances Rb .
For details see Section 5.2.
4.5. Multichannel ansatz
To complete the multichannel ansatz, we must also consider the off-diagonal elements
in Eq. (9), denoted Ss1 G(R, T ) and GSs2 (R, T ).
Ss1 G(R, T ) = Ls1 (0) T (R, T )C3 (T )C4 (T ),
(47)
GSs2 (R, T ) = C2 (0)C1 (0)T (R, T )Ls2 (T ) .
(48)
To extract their values at T = 0 mod 4, we use the non-improved spatial transporter

Ls1 (0) and Ls2 (T ), where we smeared the links beforehand using s1 respectively s2
smearing iterations. The clovers are denoted as C1 , . . . , C4 . We attach them to the linklink-correlators T(R, T ). The complete multichannel matrix is
Vij (R, T ) =
S15 S15
S30 S15
=
S60 S15
GS15

Ss1 Ss2 (R, T )
=
GSs2 (R, T )
S15 S30
S30 S30
S60 S30
GS30
S15 S60
S30 S60
S60 S60
GS60
S15 G
S30 G
.
S60 G
GG
Ss1 G(R, T )
GG(R, T )
(49)
This 4 4 matrix can be analysed using the diagonalisation procedure described in

Section 4.3. We end up with enhanced signals for the three lowest-lying states, plus
effective information about higher states.
118
4.6. Polyakov loops

According to Eq. (15) we can extract a temperature-dependent potential VT (R). The
correlator of two adjoint Polyakov loops can be easily built by using the link-link correlator
tensors and the tensor-multiplication defined in Eq. (28)

P (0)P (R) =

T(R, na) T R, (n + 1)a
!

T R, (n + 2)a T R, (n + 3)a
"
=e
VT (R)/T
(50)
5. Results
We are using a 3d-lattice with extent (48a)2 64a at inverse coupling =
4
ag 2
= 6.0.
The lattice spacing a can be obtained from the Sommer scale r0 4 [20], which is defined by
r02 Ffund (r0 ) = 1.65.
(51)
Setting r0 = 0.5 fm and comparing with the lattice result for r0 /a, one obtains a =
0.1022(1) fm. A description of our procedure to extract the fundamental force Ffund is
given in Section 5.5.
We present our results in the following order:
(1)
(2)
(3)
(4)
(5)
The static potential and excited states extracted from Wilson loops only.
The static potential extracted from the two-gluelump correlator.
The static potential and excited states obtained from the multichannel ansatz.
The temperature-dependent potential obtained from the Polyakov loop correlators.
The comparison of fundamental and adjoint potentials and the issue of Casimir scaling.
We have analysed 44 configurations, which appear to be statistically uncorrelated. To

extract the statistical errors we apply the jackknife method.
4 The phenomenological interpretation of this scale is valid only for QCD. When extracting the lattice spacing
a from the Sommer scale r0 = 0.5 fm, the resulting value of a depends on the ansatz chosen for the potential,
and on the fitting range for the force. We chose ansatz Eq. (52), which includes a perturbative logarithmic term,
because not including this term causes an unacceptably bad fit. Fitting the force over the interval 3a R 7a,
one obtains r0 /a is 4.890(1), and a = 0.1022(1) fm. This value changes by 0.5% under a variation of the fitting
interval. Similar ambiguities arise when one tries to extract the lattice spacing from the string tension ( a 2 ).
Setting = 440 MeV (0.45 fm)1 , one obtains a = 0.1136(1) fm, with a systematic variation of about
0.5% with the fitting range.
119
5.1. Wilson loops only

We measure both the fundamental and the adjoint potential between two static charges.
In the first case, string breaking cannot occur since the system does not contain particles
which can screen charges in the fundamental representation. Nevertheless, we can compare
our results with accurate data available in the literature [4,5]. We will also need these values
later on, to discuss the issue of Casimir scaling (see Section 5.5). In the case of adjoint static
charges, string breaking should occur. A summary of our results is given in Fig. 5, where
we show the fundamental potential multiplied by the Casimir ratio 83 (see Eq. (16)) and
the adjoint static potential. It can clearly be seen that the adjoint static potential remains
approximately constant for R 10a proving string breaking at Rb 10a. The unbrokenstring potential is also shown.
The horizontal line at 2.06(3)a 1 represents twice the mass of a gluelump, whose
evaluation is described in Section 4.4. This is the expected value of the static potential
of the system, when the string is broken, since the broken-string state is modelled by the
presence of two gluelumps whose interaction is screened.
Excited states are not visible for the fundamental case since the shortest Wilson loops we
consider have a minimal temporal extent of T = 4a and excited states are already strongly
suppressed. But they are clearly seen in the adjoint case for distances larger than the string
breaking distance since the Wilson loop has very good overlap with the unbroken-string
Fig. 5. The adjoint and fundamental static potentials V (R) (the latter multiplied by the Casimir factor 83 ) versus R
using Wilson loops only. The adjoint static potential remains approximately constant for R Rb 10a proving
string breaking. The unbroken-string state energy is also drawn. The horizontal line at 2.06(3)a 1 represents
twice the mass of a gluelump.
120
state which is an excited state for R > Rb . More about excited states can be found in
Section 5.1.2.
5.1.1. Static potential
We start our discussion with the extraction of the fundamental static potential Vfund (R).
We consider only one level of fundamental smearing (30 iterations of Eq. (21)) of the
fundamental spatial links and do not consider a Wilson loop matrix in the sense of Eq. (54).
A single-mass ansatz works nicely at all R in the temporal range Tmin = 12a T 60a,
where we have no measurable contribution of excited states. The extracted Vfund (R) is in
full agreement with the literature.
From the static potential we can extract the string tension . This gives us a crosscheck
with previous determinations [21] and a way to express the lattice spacing in physical units.
We use a string-motivated ansatz
R
+ R.
(52)
a
R
The Coulombic ln Ra term follows from 3d perturbation theory (see Eq. (16)). The 1/R
term follows from the bosonic string model. is a universal constant with value =
24 (d 2) in d dimensions [22]. The linear term describes confinement, and is the string
tension.
We fit all parameters and find for the string tension = 0.0625(5)a 2. This value is
stable and in full agreement with [21]. Using our ansatz Eq. (52), cannot be reliably
extracted by a global fit of the static potential. Using instead5
V (R) V0 + l ln
2 V (R) 3
R ,
R 2
the extracted tends to the universal value
=
(53)
24
0.131: = 0.126(12) and remains

R=6a
stable for R 6a albeit with larger errors.

In the following case of the adjoint static potential, the ansatz Eq. (52) does not result
in stable parameters, with or without the Coulombic ln Ra term. Nevertheless, we include
in Fig. 5 a best fit of the adjoint unbroken-string energy in the range 2a R Rb using
this ansatz.
The extraction of the adjoint static potential works well using a single-mass ansatz for
T 4 and R < Rb . At larger distances R, it is a more complicated matter since the string
breaks and the Wilson loop has a poor overlap with the broken-string. This makes the twomass ansatz mandatory. To extract the energies of the groundstate and first excited state,
as shown in the figures, we use the diagonalisation procedure described in Section 4.3. To
illustrate that the static potential V (R) at a fixed R > Rb cannot been determined by a
single-mass, we show in Fig. 6 W (R, T ) = (0) (R, T , T0 ) at R = 11a and R = 12a. Here
we want to make use of the full Wilson loop data without distorting the ratio cc10 of Eq. (12).
We use a simplified version of the diagonalisation procedure to obtain (0) (R, T , T0 ) using
Wilson loops only.
5 We do observe an increase in with increasing R, visible until R 6a. This increase can be understood as
a 1/R-correction to coming from the next-to-leading term in the bosonic string theory [22].
121
Fig. 6. Adjoint Wilson loop data versus T , for R = 11a and R = 12a, obtained from a diagonalisation procedure
Eq. (55), applied to Wilson loops, considering three different levels of smearing. A two-mass ansatz accounts for
all data points. Single-exponentials (dotted lines) do not. At large T , the broken-string groundstate is exposed.
Note how small a signal can be measured.
122
We have three types of staple-shaped transporter

,
same notation, the Wilson loops correlation matrix is

S15 S15 S15 S30 S15 S60
= Ss1 Ss2 (R, T ) = S30 S15 S30 S30 S30 S60 .
S60 S15 S60 S30 S60 S60
as used in Eq. (49). In the
(54)
(1) For a fixed R, we diagonalise the matrix Ss1 Ss2 (R, T0 ), where we choose T0 so that
the overlap with the desired state (e.g., the groundstate) is as large as possible and the
signal still quite accurate. Setting T0 TP is a natural choice. E.g., at R = 12a we
choose T0 = 24a.
(2) We use the eigenvectors v0 (R, T0 ), v1 (R, T0 ) and v2 (R, T0 ), where the corresponding
eigenvalues fulfill (0) (R, T0 ) > (1) (R, T0 ) > (2) (R, T0 ), to project Ss1 Ss2 (R, T ) to
the different states at all T by
(n) (R, T , T0 ) = vn,s1 (R, T0 )Ss1 Ss2 (R, T )vn,s2 (R, T0 ).
(55)
(0) (R, T , T0 ), the largest eigenvalue, contains amplified information about the groundstate and (1) (R, T , T0 ) information about the first excited state. (2) (R, T , T0 ) is some
effective value containing information about the remaining excitations. W (R, T ) =
(0) (R, T , T0 ) has to be analysed with a two-mass ansatz and the ratio cc10 can be extracted.
W (R, T ) = c0 eV (R)T + c1 eE1 T ,
T > Tmin .
(56)
V (R) corresponds to the groundstate, E1 is the first excited state energy. The
groundstate is the broken-string state and therefore, V (R) should be the energy of two
gluelumps E(2Qg)
2M(Qg).
At small temporal extent T of the Wilson loop W (R, T )
we get a larger slope than at large T , as visible in Fig. 6. This can be explained as elaborated
in Section 3: at small T the signal is dominated by the unbroken-string state. The brokenstring state can only be observed once T is large enough since the Wilson loop observable
has a poor overlap with this groundstate. The ratio cc10 quantifies the domination of the
unbroken-string state signal versus the broken-string state and is related to the turning
point TP (see Eq. (13)). We obtain cc10 = 3.1(4) 103 , while the prediction of the
strong coupling expansion [9] was
R=12a
c1
2
c0 R=12a
105 and (see Eq. (14)) TP(est) 42a.
Our numerical determination of the turning point TP
R=12a
= 22(1)a is clearly below this
R=12a
estimate.
Note that we can detect signals down to 1040 , which corresponds to 1080 ordinary
measurements. Previously, only signals down to 107 have been measured, i.e., in a regime
where the unbroken-string state is dominating over the groundstate. This explains why
string breaking has not been observed in Wilson loops up to now.
5.1.2. Excited states
Excited states of the fundamental representation are suppressed too much for us to
measure. But in the adjoint representation, we have clear information about the first excited
state. Using the diagonalisation procedure Eq. (38), one source of information is the twomass fit of W (R, T ) = (0) (R, T , T0 ) at large distances R Rb where the first excited
123
state is the unbroken-string state. Another, related, source of information, also for smaller
R, is W1 (R, T ) = (1) (R, T , T0 ).
In the same manner as for the static potential, we adopt here a two-mass ansatz
W1 (R, T ) = e0 eV (R)T + e1 eV1 (R)T ,
T > Tmin ,
(57)
where V (R) is the static potential and V1 (R) the energy of the first excited unbroken-string
state (as it turns out). In Fig. 7, for distances smaller than the string breaking distance
Rb we see a clear signal of the first excited state. However, at larger distance we expect
contributions from at least three states: the two-gluelump groundstate, the unbroken-string
state, and an excited unbroken- or broken-string state, since one expects a level-crossing of
the latter two depending on the spatial distance R.
Surprisingly, at R = 8a and R = 9a we completely miss the broken-string state. The
explanation lies in the spectrum at R = 8a, 9a. The first three terms entering the expansion
of the Wilson loop Eq. (10) are:
c0 eE0 T groundstate: lowest-lying energy state of the unbroken-string,
c1 eE1 T first excited state: lowest-lying energy state of the broken-string,
c2 eE2 T second excited state: first excited state of the unbroken-string.
Fig. 7. The static adjoint potential V (R) versus R (same as in Fig. 5) and the first excited unbroken-string state
energy using Wilson loops only. We also show the energy E 1 (R) (Eq. (58)) resulting from the relativistic Nambu
string theory. The horizontal line at 2.06(3)a 1 represents twice the mass of a gluelump.
124
The overlap of the Wilson loop operator with the second excited state (unbroken-string), c2 ,
is larger than the overlap with the first excited state, c1 . In addition, the difference between
the two corresponding energies E1 and E2 is small. In our case, the second excited state
indeed dominates over the first in the accessible range of Euclidean times. Therefore, we
miss the broken-string state at R = 8a and R = 9a.
From our fit of the adjoint potential via Eq. (52) we have extracted the string tension .
We then consider the relativistic Nambu string [23], which predicts for the first excited
string energy
#
E 1 (R) =
2 R2

1
+ 2 1
+ const.
24
(58)
This relativistic bosonic string theory prediction agrees with our measured first excited
energy remarkably well, see Fig. 7.
5.2. Gluelumps
We have seen that the Wilson loop operator has an overlap with both statesthe
unbroken-string and the broken-string state. What about the two-gluelump operator?
At large distances R > Rb , a single-mass ansatz is sufficient since the groundstate is
the broken-string state, which has good overlap with the correlator of two gluelumps,
GG(R, T ). Therefore, we cannot measure any signal of the unbroken-string state in this
regime. At distances R < Rb , GG(R, T ) can be analysed using a two-mass ansatz
GG(R, T ) g0 eV (R)T + g1 eE1 (R)T ,
T > Tmin .
(59)
At small R, the turning point TP is small, and the two-mass ansatz Eq. (59) works fine. For
R = 8a and R = 9a, which approaches the string breaking distance Rb , the energies of the
unbroken-string and broken-string state are almost degenerate, and the turning point value
is large. As mentioned in Section 4.4, we used improved spatial transporters to measure
Wilson loops. Here, we use non-improved clovers and have more noise. In addition, we
have only one operator-state and cannot apply a diagonalisation procedure. Therefore, we
have difficulties to measure the subleading groundstate exponential decay in this regime.
We lose the signal of the unbroken-string state before it becomes visible and cannot extract
the unbroken-string groundstate properly.
Fig. 8 shows the agreement between the static potential extracted from Wilson loops
only and that extracted from the two-gluelump correlator. This confirms that GG(R, T )
has a non-vanishing overlap with the unbroken-string state, similar to the fact, that the
Wilson loop has a non-vanishing overlap with the broken-string state.
GG(R, T ) can also be used in the multichannel ansatz as explained in Section 3. It
enters as a diagonal matrix element in the 4 4-matrix Vij (R, T ) Eq. (49). We will now
consider this approach.
125
Fig. 8. Agreement of the static adjoint potential V (R) versus R, extracted from the two-gluelump correlator and
the one extracted from Wilson loops only (same as in Fig. 5, but shifted to the right for clarity). The deviations at
R = 8a and R = 9a are due to a large value of the turning point TP , as explained in the text.
5.3. Multichannel ansatz

The full multichannel matrix Eq. (49) is obtained by including the mixing terms
Si G(R, T ) and GSj (R, T ).
5.3.1. Static potential
We analyse the 4 4 matrix using the diagonalisation procedure Eq. (38) at T0 = 4a,
as described in Section 4.3. As expected, a single-mass ansatz can be applied for all R
using (0) (R, T , T0 = 4a). The results, presented in Fig. 9, agree with the static potential
extracted from Wilson loops only, with improved accuracy for R > Rb .
The three Wilson loop operator states at different smearing levels have a good
overlap with the unbroken-string state, but a poor one with the broken-string state.
The opposite holds for the two-gluelump operator state. To confirm this statement, we
(n)
analyse the overlaps aj (R, T , T0 ) of Eq. (39). We consider the overlap of all three
Wilson loop operator states (j = S15 , S30 , S60 ) and the overlap of the two-gluelump
(0)
operator state (j = G) with the groundstate (n = 0).6 We observe for aG
(R, T , T0 )
an abrupt change from O(103) (R 10a) to O(1) (R > 10a) and vice versa for
(n=0)
6 In a
(R, T , T0 = 4a) of Eq. (39) we fix T = 8a. The results are stable for larger T , although with
j
increasing errors.
126
Fig. 9. The static adjoint potentials V (R) versus R using the multichannel ansatz. The agreement with the static
potential extracted from Wilson loops only (same as in Fig. 5, shifted to the right for clarity) is good. The
unbroken-string state energy is also drawn (dashed line). The horizontal line at 2.06(3)a 1 represents twice
the mass of a gluelump.
aS(0)(R, T , T0 ). This indicates that string breaking actually occurs at a distance slightly
larger than 10a.
5.3.2. Excited states

Considering (1) (R, T , T0 = 4a), we get information about excited states by applying
a two-mass ansatz. For R > Rb , we extract the lowest-lying unbroken-string state as
expected. For R = 8a and R = 9a, the first excited state is the broken-string state. For
R = 7a, the broken-string state and the excited unbroken-string state energies are almost
degenerate. For R < 7a we extract the energy values of the excited unbroken-string state
which are in agreement with the ones extracted using Wilson loops only. For R < 5a
we can no longer extract first excited state energies due to statistical noise. Finally,
considering (2) (R, T , T0 = 4a), for R 7a, we obtain the energy of the second excited
state, namely the excited unbroken-string state. For R < 7a we cannot extract the second
excited energy due to statistical noise. In Fig. 10, we show the static potential extracted
from (0) (R, T , T0 = 4a), the first excited energy, extracted from (1) (R, T , T0 = 4a) and
the second excited energy, extracted from (2) (R, T , T0 = 4a).
127
Fig. 10. The static potential V (R), the first and the second excited states energies using the multichannel ansatz.
We also show the energy E 1 (R) (Eq. (58)) resulting from the relativistic bosonic string theory. The horizontal
line at 2.06(3)a 1 represents twice the mass of a gluelump.
5.4. Polyakov loops

The correlator of two adjoint Polyakov loops allows the extraction of a temperaturedependent potential VT (R) (see Eq. (15))
VT (R)

1
ln P (0)P (R) .
Nt a
(60)
The temperature of our system is T = N1t a 30 MeV since Nt = 64 is the temporal extent
of the lattice and a = 0.1022(1) fm the lattice spacing. Since this temperature is quite low,
contributions of excited states are negligible. Therefore, we observe a good match with the
static potential measured using Wilson loops only (see Fig. 11).
In the regime of string breaking R 10a we see flattening of the potential indicating
string breaking. For R = 12a, the signal becomes very noisy and the average correlator is
negative.
5.5. Casimir scaling
Since we measure the static potentials between fundamental and between adjoint
charges with high accuracy, we can examine the hypothesis of Casimir scaling, which says
that the ratio of the adjoint static potential over the fundamental static potential remains
equal to the Casimir value 83 over a broad range of distances where both potentials grow
128
Fig. 11. Polyakov loop method. The static adjoint potential V (R) versus R extracted from the correlator of two
adjoint Polyakov loops agrees very well with that measured using Wilson loops only (same as in Fig. 5, shifted
to the right for clarity). The flattening of the potential can be observed, although we cannot extract the value at
R = 12a due to large fluctuations.
more or less linearly with distance. Already in Fig. 5, we see clear deviations from Casimir
scaling: the fundamental static potential, rescaled by the Casimir factor 83 , agrees with the
adjoint static potential at small distances R 2a only. This confirms earlier observations
of Ref. [5], also at = 6.0, and of Ref. [24] at = 9.0. Therefore, we apply a more careful
analysis considering forces, defined by
F (rI ) =
V (r) V (r a)
,
a
(61)
where rI is chosen such that the force evaluated from Eq. (61) coincides with the force in
the continuum at tree level [20]. The procedure to determine the explicit values of rI is
described in Appendix A.
In Fig. 12, we show the ratio Fadj (R)/Ffund (R) at two different s. In the regime
of perturbation theory, i.e., at small distances R, this ratio is 83 as expected. At larger
distances, our = 6.0 data show clear deviations. The ratio appears to decrease linearly
with increasing distance7 R. The less precise data at = 9.0 of Ref. [24] seem to confirm
this R-dependence, making it unlikely to be an artifact of the lattice spacing.
7 For R > R , the adjoint string breaks and the force F
b
adj is essentially zero. Large fluctuations at R Rb
induce the large error on the rightmost data point in Fig. 12.
129
Fig. 12. Ratio of forces Fadj (R)/Ffund (R) as a function of the spatial separation. The horizontal line at 83 indicates
the Casimir ratio expected from perturbation theory. We see clear deviation at distances larger than R = 2a. The
ratio seems to decrease linearly while increasing R.
Our results do not necessarily contradict the work of [25], which found accurate Casimir
scaling at large distances. Ref. [25] considers the 4dSU(3) theory, while we consider SU(2)
in 3 dimensions. Rather, what we see might be specific to 3 dimensions.
6. Conclusions
We have demonstrated that string breaking can be observed, using only Wilson loops
as observables to measure the static potential. This demonstration was performed in the
computationally easiest setup: breaking of the adjoint string in the (2 + 1)-dimensional
SU(2) theory. Even in this simple case, the unambiguous observation of string breaking,
at a distance Rb 1 fm, required the measurement of adjoint Wilson loops of area in
excess of 4 fm2 , with state-of-art variance reduction techniques. A similar study in 3 + 1
dimensions with a larger gauge group will be challenging.
The reason for such large loop sizes is as expected: the Wilson loop has very poor
overlap with the broken-string. Even when the static adjoint charges are separated by
R > Rb and the broken-string becomes the groundstate, its contribution to the Wilson loop
area-law is subdominant. The temporal extent T of the Wilson loop must be increased
beyond a characteristic distance, the turning point TP , to weaken the unbroken-string
state signal and reveal the true groundstate. We find TP 2 fm, which explains why
earlier studies of non-Abelian gauge theories, which did not use similar variance reduction
130
methods, failed to detect string-breaking. While large, this turning point value stays well
below the strong-coupling estimate of [9], which would predict a value about twice as
large.
Of course, string-breaking is easy to observe, over a limited Euclidean time extent,
if one uses a multichannel approach where a correlation matrix between unbroken- and
broken-string states is formed and diagonalised, the latter being modelled by a pair of
gluelumps. We reproduce in this case the results in the literature. We also consider the
two-gluelump correlator, which has poor overlap with the unbroken-string state, and show
that the unbroken-string groundstate can be extracted from that correlator alone, if one
allows again for a large Euclidean time extent. Therefore, full information about the adjoint
potential is contained in each of the diagonal elements of the multichannel matrix.
Finally, we looked in detail at the issue of Casimir scaling, by measuring the ratio of
forces Fadj (R)/Ffund (R) as a function of R. We observe clear deviations of this ratio from
the perturbative value 83 , and an apparent linear decrease with R. A consistent crosscheck
at a smaller lattice spacing makes this violation of Casimir scaling unlikely to be a lattice
artifact. The situation, however, may be different in the (3 + 1)d theory.
Acknowledgements
We are grateful to Michele Pepe and Owe Philipsen for discussions and advice. We also
thank Michele Vettorazzo for helpful hints and support and Oliver Jahn for his continuous
help and for reading the manuscript. The calculations for this work were performed on the
Asgard Beowulf Cluster at the ETH Zrich.
Appendix A. rI and the free scalar propagator
We follow [20] and define forces by
V (r) V (r a)
,
a
where rI is chosen such that the force evaluated from Eq. (A.1) coincides with
F (rI ) =
(A.1)
g02
(A.2)
,
2rI
the force in the continuum at tree level. This results in
a
rI =
(A.3)
,
2(G(r, 0) G(r a, 0))
where G(x, y) is the massless scalar lattice propagator in 2 dimensions defined by
F (rI ) = C2
G(x, y) = (x, y).
(A.4)
We simply solve for G(x = n1 a, y = n2 a) by Fourier transform, namely

2
N

cos( 2
1
1 li ni )
N
G(n1 a, n2 a) = 2
,
2
2
N
li )
i=1 2 2 cos(
l1 =1,l2 =1
(A.5)
131
Table 1
(left) The scalar lattice propagator G(r, 0) in 2d. (right) The naive derivation points (r a2 ) compared with rI in
the range we consider
r/a
G(r, 0)
0
1
2
3
4
5
6
7
8
9
10
11
0.0000
0.2500
0.3634
0.4303
0.4770
0.5129
0.5421
0.5668
0.5881
0.6069
0.6237
0.6389
(r a2 )/a
2.5
3.5
4.5
5.5
6.5
7.5
8.5
9.5
rI /a
2.3790
3.4080
4.4333
5.4505
6.4435
7.4721
8.4657
9.4735
where N n1 , n2 is the number of discretisation points, taken sufficiently large that

G(n1 a, n2 a) is known to 4-digit accuracy. Finally, we set the zero-mode contribution so
that G(0, 0) = 0.0. A summary of our results in the range we consider is given in Table 1.
References
[1] E. Laermann, C. DeTar, O. Kaczmarek, F. Karsch, Nucl. Phys. B (Proc. Suppl.) 73 (1999) 447.
[2] C. Michael, Nucl. Phys. B (Proc. Suppl.) 26 (1992) 417;
O. Philipsen, H. Wittig, Phys. Rev. Lett. 81 (1998) 4056;
O. Philipsen, H. Wittig, Phys. Rev. Lett. 83 (1999) 2684, Erratum;
ALPHA Collaboration, F. Knechtli, R. Sommer, Phys. Lett. B 440 (1998) 345;
O. Philipsen, H. Wittig, Phys. Lett. B 451 (1999) 146;
C. DeTar, U.M. Heller, P. Lacock, Nucl. Phys. B (Proc. Suppl.) 83 (2000) 310;
P. de Forcrand, O. Philipsen, Phys. Lett. B 475 (2000) 280;
UKQCD Collaboration, P. Pennanen, C. Michael, hep-lat/0001015;
MILC Collaboration, C.W. Bernard, et al., Nucl. Phys. B (Proc. Suppl.) 94 (2001) 546.
[3] P.W. Stephenson, Nucl. Phys. B 550 (1999) 427.
[4] K.D. Born, et al., Phys. Lett. B 329 (1994) 325;
U.M. Heller, et al., Phys. Lett. B 335 (1994) 71;
SESAM Collaboration, U. Glssner, et al., Phys. Lett. B 383 (1996) 98;
CP-PACS Collaboration, S. Aoki, et al., Nucl. Phys. B (Proc. Suppl.) 73 (1999) 216;
TXL Collaboration, G.S. Bali, et al., Phys. Rev. D 62 (2000) 054503;
B. Bolder, et al., Phys. Rev. D 63 (2001) 074504;
H.D. Trottier, K.Y. Wong, hep-lat/0209048.
[5] G.I. Poulis, H.D. Trottier, Phys. Lett. B 400 (1997) 358.
[6] F. Gliozzi, P. Provero, Nucl. Phys. B (Proc. Suppl.) 83 (2000) 461.
[7] S. Kratochvila, P. de Forcrand, hep-lat/0209094.
[8] M. Lscher, U. Wolff, Nucl. Phys. B 339 (1990) 222.
[9] I.T. Drummond, Phys. Lett. B 434 (1998) 92.
[10] F. Gliozzi, A. Rago, Phys. Rev. D 66 (2002) 074511.
[11] K. Kallio, H.D. Trottier, Phys. Rev. D 66 (2002) 034503.
[12] L.D. McLerran, B. Svetitsky, Phys. Rev. D 24 (1981) 450.
[13] Y. Schrder, DESY-THESIS-1999-021.
132
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
G. Parisi, R. Petronzio, F. Rapuano, Phys. Lett. B 128 (1983) 418.

C. Michael, Nucl. Phys. B 259 (1985) 58.
M. Lscher, P. Weisz, JHEP 0109 (2001) 010.
P. Majumdar, hep-lat/0208068.
J. Engels, F. Karsch, E. Laermann, C. Legeland, M. Ltgemeier, B. Petersson, T. Scheideler, Nucl. Phys. B
(Proc. Suppl.) 53 (1997) 420.
M. Pepe, private communication.
R. Sommer, Nucl. Phys. B 411 (1994) 839.
M.J. Teper, Phys. Rev. D 59 (1999) 014512.
M. Lscher, Nucl. Phys. B 180 (1981) 317.
S. Perantonis, A. Huntley, C. Michael, Nucl. Phys. B 326 (1989) 544.
O. Philipsen, private communication.
G.S. Bali, Phys. Rev. D 62 (2000) 114503.

Z3 orbifolds of the SO(32) heterotic string:

1-Wilson-line embeddings
Joel Giedt
Department of Physics, University of Toronto, 60 St. George St., Toronto, ON, M5S 1A7, Canada
Received 31 March 2003; accepted 22 August 2003
Abstract
We consider compactification of the SO(32) heterotic string on a 6-dimensional Z3 orbifold
with one discrete Wilson line. A complete set of all possible embeddings is given, 159 in all.
The unbroken subgroups of SO(32) are tabulated. The extended gauge symmetry SU(3)3 , recently
discussed by Kim [hep-th/0301177] for semi-realistic E8 E8 heterotic string models, occurs for
several embeddings, as well as other groups that may be of interest in unified string models. The
extent to which extra gauge group factors can be hidden is discussed and compared to the E8 E8
case. Along flat directions where an effective hidden sector exists, the embeddings described here
provide for hidden gauge groups that are not possible in the E8 E8 heterotic string.
PACS: 11.25.Mj
1. Introduction
One of the main motivations for starting with the E8 E8 heterotic string [1] in semirealistic applications is that the second E8 factor would appear to provide a natural source
of a hidden sector in which to break supersymmetry by, say, gaugino condensation [2].
Indeed, in a field theoretic dimensional reduction of the 10-dimensional theory to 4
dimensions, one finds only Planck mass suppressed operators communicating between the
sets of gauge-charged fields coming from the two E8 s.
However, when the underlying string theory is formulated with 6 dimensions compactified on an orbifold [3,4], quantum consistency of the 2-dimensional conformal field theE-mail address: giedt@physics.utoronto.ca (J. Giedt).
doi:10.1016/j.nuclphysb.2003.08.031
134
J. Giedt / Nuclear Physics B 671 (2003) 133147
ory necessitates the addition of twisted sector states. These are states that would not be
present in a field theoretic dimensional reduction of the original 10-dimensional theory.
Quite generally, these twisted states are simultaneously charged under subgroups of both
E8 factors [5]. Thus these states can mediate supersymmetry breaking through gauge interactions and the so-called hidden sector is no longer hidden. Typically one overcomes
this difficulty by breaking the gauge interactions (most often extra U (1)s and SU(2)s) at
or near the string scale so that their interactions are more or less Planck scale suppressed.
This requires a careful choice of flat direction; for examples see [6].
Against this backdrop it is worth reconsidering the aversion to the SO(32) heterotic
string, since: (i) with an appropriate orbifold embedding chiral representations occur,
and (ii) extra gauge symmetries can be hidden the same way that they are hidden in
E8 E8 heterotic orbifolds. It will be seen from the gauge groups obtained below that
most embeddings give rise to a product of non-Abelian simple groups and some U (1)s.
Generically representations will arise that feel any pair of factors. However, if we can
choose a flat direction that renders these states supermassive (say, near the string scale),
then it would seem that we can manufacture an effectively hidden sector that is comparable
to what may be obtained from a 4-dimensional E8 E8 heterotic string construction
via orbifold compactification. Indeed, it is generic in such constructions that many states
do decouple near the string scale, due to the presence of an anomalous U (1), since the
anomaly is cancelled by a counterterm that induces a FayetIliopoulos term [7]. The point
is that in either case, E8 E8 or SO(32), there exist difficulties hiding the hidden sector
and specific assumptions regarding flat directions must be made [8].
It should be remarked however, that we may not want to hide extra gauge groups. Rather,
one may have in mind gauge mediation scenarios (see [9] for a review, and the extensive
references therein) with exotic messenger states that feel both the strong dynamics of the
sector that breaks supersymmetry and the ordinary gauge symmetries of the observable
world. In this case the larger messenger representations that are possible in the SO(32)
string could be advantageous.1
In either event, model builders might wish to enlarge their vistas by considering
orbifolds of the SO(32) heterotic string. In this regard it is useful to have some idea for
a good starting point. Selecting the string scale gauge symmetry is certainly an important
first step, and it is therefore worth knowing what is possible and how to get it. With this goal
in mind, in the present article we enumerate all inequivalent embeddings with 1 discrete
Wilson line for the 6-dimensional Z3 orbifold of the SO(32) heterotic string. Thus we
perform the analogue of previous calculations in the E8 E8 heterotic string; most notably
those in [1012].
Models that contain unbroken gauge groups such as SU(5)2 are starting points for
higher affine level models. It has been described in [13], for instance, how the presence
+ (5,
5) representations can be used to break to the diagonal subgroup, which
of (5, 5)
is realized at level 2, with the requisite Higgses to break further to the Standard Model
1 In the E E heterotic string only twisted states can talk between the two E s. The twisted states must
8
8
8
have smaller E8 E8 root torus winding numbers and hence tend to be in smaller representations of the gauge
group.
135
gauge group.2 However, the authors of [13] have also concluded that for the Z3 orbifolds
it is not possible to get a chiral spectrum for the resulting level k = 2 SU(5) GUT. They
argue that this is because the net number of fermion generations vanishes in constructions
+ (5,
5) representations. After some elementary calculations, we
with the requisite (5, 5)
have reached the same conclusion. Therefore, although some embeddings here do have
SU(5)2 and the requisite representations to obtain a level k = 2 diagonal subgroup with
adjoint Higgses, they are not viable routes to semi-realistic string GUTs, and one should
look elsewhere.
Another sort of unified model has recently been studied by Kim for the E8 E8
heterotic Z3 orbifold [17]. This is the group SU(3)3 . The advantages of this approach over
standard-like constructions are notable (see [17] for further details). We would like to point
out that several of the embeddings listed here also give SU(3)3 in the surviving group; we
leave as a topic for further investigation the phenomenology of these models, which should
share in the advantages noted in [17].
In Section 2 we describe the embeddings of the orbifold action into the sixteen internal
bosons responsible for gauge symmetry in the low-energy theory. In Section 3 we discuss
equivalence relations that are used to reduce the number of embeddings. In Section 4 we
summarize the key elements used in constructing all consistent embeddings. In Section 5
we address the various possible Wilson lines and our results. In Section 5 we state our
conclusions. In Appendix A we discuss an important set of equivalence relations. In
Appendix B we present tables of the 159 embeddings found here.
2. Abelian Z3 embeddings
The sixteen internal left-moving bosons XI (+ )the gauge degrees of freedomare
compactified on the spin(32)/Z2 lattice, which we will denote by . This lattice consists
of all 16-vectors of the form

1
1
,
n1 + , . . . , n16 +
(n1 , . . . , n16 ),
(2.1)
2
2

subject to the constraints nI Z and I nI = 0 mod 2. We remind the reader that spin(32)
is the covering group for SO(32). The SO(32) roots are

1, 1, 014 .
(2.2)
Here, signs are not correlated. Underlining indicates all permulations are to be taken. The
exponent indicates that the entry is repeated 14 times. Analogous notations will be used
below.
The Z3 orbifold is obtained as the quotient of the 6-dimensional SU(3)3 root torus by
simulataneous 2/3 rotations in each of the three complex planes labeled by complex
coordinates zi (i = 1, 3, 5). This twist in the 6-dimensional compact space is embedded
2 For other works on how higher affine level models can be constructed and how they might afford semirealistic string GUTs, see Refs. [1416].
136
into internal gauge degrees of freedom in an Abelian mannerthrough a shift:

zi e2i/3 zi XI (+ ) XI (+ ) + V I .
(2.3)
In addition we allow for the possibility of discrete Wilson lines ai (i = 1, 3, 5) which embed
translations in the 6-dimensional compact space. It suffices to specify this embedding for
three such shifts, due to constraints arising from the torus construction:
zi zi + 1 XI (+ ) XI (+ ) + aiI
i = 1, 3, 5.
(2.4)
The four embedding vectors are subject to consistency relations that follow from worldsheet modular invariance:
3V , 3ai ,
3V 2 Z,
3ai2 Z,
3V ai Z,
3ai aj Z. (2.5)
An infinite number of solutions to these conditions exist. Fortunately only a finite number
of inequivalent possibilities are contained in this set, due to equivalence relations that we
discuss in Section 3.
The massless gauge-charged states are characterized by 16-dimensional winding
vectors. In the untwisted sector we have for states with non-trivial weights with respect
to the Cartan subalgebra:
K ,
K 2 = 2.
(2.6)
The Wilson lines enforce a projection on these states. Only those that satisfy
ai K Z i = 1, 3, 5
(2.7)
survive. Those that do survive fall into three categories:
gauge,
0 mod 3
3V K = 1 mod 3
matter,
1 mod 3 antimatter.
(2.8)
In truth this is a further projection onto states with differing right-moving quantum
numbers.
For the twisted states, corresponding to string states with non-trivial monodromy, we
have weights K which satisfy
4
K 2 = 2NL ,
3
K = K + V +
ni ai ,
K .
(2.9)
i=1,3,5
If left-moving oscillators are excited in the 6-dimensional compact space, we can have
NL = 1/3 or 2/3. (However, NL = 2/3 only has a solution to (2.9) if the embedding
is equivalent to the trivial one; i.e., V = ai = 0.) The integers ni = 0, 1 label fixed
point locations in each of the 3 complex planes. Each twisted state is labeled by a triple
(n1 , n3 , n5 ). Note that (2.5) implies 3K .
137
3. Equivalence relations
The equivalence relations are essentially those already alluded to in [4] and discussed
in detail in [11] for Z3 orbifolds of the E8 E8 heterotic string.
Lattice group equivalence. This merely states that V V + K and ai ai + Ki
yield equivalent embeddings for any choice K, Ki .
Weyl group equivalence. This states that for any SO(32) root e (given in (2.2)) the
simultaneous Weyl reflection
V V (V e)e,
ai ai (ai e)e,
(3.1)
gives an equivalent embedding. It is easy to check that the Weyl group here consists of
permutation of entries, pair-wise sign flips, and compositions of these operations. This is a
considerable simplification over the E8 E8 case where half-integral roots exist that lead
to a more complicated Weyl group. (The half-integral SO(32) weights are not roots.)
Fixed point label equivalences. First, we have ai ai for any of the Wilson lines.
This is just a relabeling ni ni of the fixed points. Second, we have V V ai for
any choice i = 1, 3, 5. This is a relabeling ni ni 1 of the fixed points. For instance,
the twisted sector winding vector is rewritten

K = K + V +
(3.2)
nj aj = K + (V ai ) + (nj 1)ai +
nj aj
j =i
to display that this is nothing but a relabeling of fixed points, keeping in mind ni
ni mod 3. The complete twisted sector spectrum will be the same; only the labeling will be
different. It is easy to check that the projections (2.8) in the untwisted sector are unchanged,
due to (2.7).
4. Building blocks
Recall from (2.5) that the embedding vectors V , ai must satisfy 3V , 3ai .
Taking into account (2.1), the entries of 3V , 3ai are either all integral or all halfintegral. We easily restrict to integral weight lattice vectors 3V , 3ai using the lattice group
equivalence; simply add 3K where K is any half-integral lattice vector. Repeatedly adding
3ej s, where ej s are the roots in (2.2), allows us to lower the magnitudes of entries of
3V , 3ai until no entry has magnitude greater than 2. We can restrict to no more than one
2 appearing in 3V , 3ai using the lattice group equivalence: addition of some 3ej , where
ej is one of the roots (2.2), can be used to eliminate any pair of 2s. The self-consistency
constraints in (2.5), which we find it convenient to write
(3V )2 = 0 mod 3,
(3ai )2 = 0 mod 3,
provide a further restriction, and since K implies

of 3. Thus,
(3V )2 = 0 mod 6,
(3ai )2 = 0 mod 6.
(4.1)
K2
even, we only get even multiples

(4.2)
138
Table 1
Inequivalent twist embeddings
3V
16
0
6 10
1 ;0
12 4
1 ;0
2, 18 ; 07
Untw. matter
none

3 (15, 1) + (6, 20)

3 (66, 1) + (12, 8v )

3 (3, 1) + (3, 26)

3 (36, 1) + (9, 14)

3 (105) + 2(15)
SU(6) SO(20) U (1)
2, 12 ; 013
G
SO(32)

2, 114 ; 0
SU(12) SO(8) U (1)

SU(3) SO(26) U (1)
SU(9) SO(14) U (1)
SU(15) U (1)2
Therefore we find that 3V , 3ai can only belong to the set

16 6 10 12 4 2 13 8 7 14
0 , 1 , 0 , 1 , 0 , 2, 1 , 0 , 2, 1 , 0 , 2, 1 , 0 ,
(4.3)
up to ordering and sign permutations. It follows that these 6 vectors form the basis of all
subsequent analysis.
We fix ordering and signs for the twist embeddings using the Weyl group equivalence.
Then the inequivalent twist embeddings 3V together with their unbroken subgroups of
SO(32) and untwisted matter are given in Table 1. We find it convenient in what follows to
concentrate on how Wilson lines further break the gauge group in the two distinct parts of
the 16-dimensional space, emphasized by the placement of a semicolon in the entries for
3V in Table 1. The first subspace, where 3V I = 0, we will refer to as the non-zero sector.
The second subspace, where 3V I = 0, we will refer to as the zero sector. These are not to
be confused with the usual sectors of the Hilbert space of the underlying conformal field
theory.
5. Wilson line embeddings

Inequivalent Wilson lines 3a1 are obtained by taking sign and ordering permutations of
the six vectors (4.3) subject to the V a1 constraint in (2.5), which we find it convenient to
write
3V 3a1 = 0 mod 3.
(5.1)
In the case of V = 0 we have only the six possibilities listed in Table 4.

For the effects of Wilson lines when V = 0 it is convenient to focus on the non-zero
sector and the zero sector separately. Condition (5.1) constrains the entries of a1 in the
non-zero sector, whereas it places no constraint on the entries of a1 in the zero sector. As
an example we consider 3V = (16 ; 010) in some detail. For the other cases we merely state
our results, except as regards some none-too-apparent equivalences.
In the non-zero sector the possibilities that exist for the corresponding 6 entries of 3a1
are given in Table 2. We have indicated how the SU(6) factor surviving V is broken by the
Wilson line. Equivalence relations have been used to eliminate obvious redundancies. In
Table 2
Entries in the non-zero sector for 3V = (16 ; 010 )

3 a11 , . . . , a16
6 6

0 , 1 , 2, 15

1, 1, 04 , 14 , 1, 0 , 2, 1, 04 , 2, 13 , 1, 0
3 3 3

1 , 0 , 1 , (1)3 , 2, 12 , (1)3 , 2, 12 , 03
2

1 , (1)2 , 02 , 2, 1, (1)2 , 02
139
SU(6) subgroup
SU(6)
SU(4) U (1)2
SU(3)2 U (1)1
SU(2)3 U (1)2
an appendix we show that embeddings with (2, (1)4, 0) are equivalent to embeddings
with (2, 13, 1, 0). This is why we have dropped the former possibility. For the zero
sector we have the set
9

1 , 1 and 1n , 010n n = 0, 1, . . . , 10.
(5.2)
Note that the sign in the first vector cannot be removed by the pairwise sign flips without
disturbing V . In all other cases signs can be removed in the zero sector. In the case
where we have for a1 entries (06 ) in the non-zero sector of V , we must also include the
possibilities of

2, 12, 07 and 2, 18 , 0
(5.3)
in the zero sector of a1 since, lacking 1s in the non-zero sector, we cannot push the
2 into the non-zero sector using lattice group equivalence. That is, in the other cases
we can eliminate an entry of 2 from the zero sector using the lattice group equivalence
3a1 3a1 + 3K where 3K = (3, 05 ; 3, 09).
It is simple to determine the possibilities that are consistent with a given choice for the
non-zero sector, by comparing to the vectors in (4.3) and requiring that the total 3a1 be one
of these up to sign and ordering permutations. For example, if we choose (1, 1, 04) in the
non-zero sector then we have only

1, 1, 04 14 , 06 , 19 , 1 ,
(5.4)
corresponding the second and third vectors in (4.3). These are Embeddings 2.52.7 of
Table 5. As another example if we choose (2, 12, (1)3 ) in the non-zero sector then we
have only

2, 12, (1)3 13 , 07 , 19 , 0 ,
(5.5)
corresponding to the last two vectors in (4.3). These are Embeddings 2.21 and 2.22 of
Table 5.
The possibilities for the zero sector lead to breakings of the SO(20) that survives V .
These are given in Table 3. Of course they are correlated to what occurs in the non-zero
sector. Thus we obtain as a complete list of consistent embeddings for 3V = (16 ; 010) the
entries of Table 5. We also list the non-Abelian part GNA of the surviving gauge group G.
One should add as many U (1)s as are needed to have G a rank 16 group.
140
Table 3
Entries in the zero sector for 3V = (16 ; 010 )

3 a17 , . . . , a116
10
0
9
1, 0
10n n
, 0 n = 2, . . . , 8
1

9
1 , 0 , 2, (1)8 , 0
9

1 , 1

2, 12 , 07
SO(20) subgroup
SO(20)
SO(18) U (1)
SU(10 n) SO(2n) U (1)
SU(9) U (1)2
SU(10) U (1)
SU(3) SO(14) U (1)
Including the cases with a1 = 0, it can be seen from Tables 49 that we have a
total of 159 different embeddings when one discrete Wilson line is permitted. Although
it is plausible some redundancy may yet exist, most of it has been removed using the
equivalence relations described above. Certainly the list is complete, meaning that any
consistent embedding with one discrete Wilson line is equivalent to one of the embeddings
presented here.
6. Conclusions
We have argued that the SO(32) heterotic string provides an interesting starting point for
semi-realistic string phenomenology. We have worked out a complete list of all embeddings
with one discrete Wilson line for the symmetric Z3 orbifold. To our knowledge this
calculation has not been presented previously in the literature. We have addressed most
of the equivalences that relate embeddings. Some of the less obvious equivalences have
been described in detail in an appendix.
We have commented briefly on the prospects for obtaining a hidden sector and for softly
broken supersymmetry via gaugino condensation in this sector. It is our conclusion that
there is every reason to believe that some of the models described here will be viable in
this respect subject to an appropriate choice of flat direction. We leave explicit explorations
of this conjecture for future investigation.
Many avenues for future research present themselves. One interesting possibility has to
do with the phenomenology of the SU(3)3 embeddings, along the lines of [17]. Research
in this direction is in progress and we hope to report on it shortly. Another issue worth
exploration is flat directions that might lead to a level k = 2 SU(5) GUT for the embeddings
that contain SU(5)2 as a proper subgroup, such as those embeddings with SU(6)2 . In this
case, an important question is whether or not it is possible to obtain a chiral spectrum with
respect to the diagonal subgroup. A final issue we would like to mention is whether or not
gauge mediation is viable for any of these models.
It is our hope that we have convinced the reader that the SO(32) heterotic string can
provide intriguing unified models. Further research on the phenomenological possibilities
141
when this is taken as the starting point would certainly be a welcome supplement to what
is a comparatively sparse examination in the existing literature.
Acknowledgement
This work was supported by the National Science and Engineering Research Council of
Canada.
Appendix A. Technical aspects of embedding equivalences

In this appendix we address some of the more technical details in uncovering the
equivalence between different embeddings.
One type of equivalence is particularly important, since it is easy to overlook, but
removes a good deal of redundancy. It is used on the non-zero sector embeddings with
a 2. We show the intermediate step which sketches out the proof of the equivalence:

2, 1m+2 , (1)3n+m 1m+3 , 2, (1)3n+m1 2, 13n+m1 , (1)m+3 .
(A.1)
In the first step we add a pair of 3s using lattice equivalence. In the second step we send
3a1 3a1 . Whatever signs we had in the zero sector are reversed when this is done. In
many cases the original signs in the zero sector can be restored by pairwise sign flips. In
those cases where this is not true, we can always obtain (1p , 1) in the non-zero sector,
again by pairwise sign flips. But both possibilities are included in our tables, so there is
nothing lost by using the equivalence (A.1).
For example, an equivalence of the form (A.1) exists in the embeddings with 3V =
(16 ; 010). In the non-zero sector we have:

2, (1)4, 0 2, 13, 1, 0 .
(A.2)
It is for this reason that (2, (1)4 , 0) does not appear in our tables.
As a second example, consider the embeddings with 3V = (112; 04 ). In this case when
there is a 2 present in the non-zero sector part of 3a1 , we have the range 3V 3a1 =
12, 9, . . ., 9. However, equivalences of type (A.1) need to be accounted for. It is not
too hard to check that:
(i) all cases of 3V 3a1 = 12 are equivalent to one of the cases of 3V 3a1 = 6;
(ii) all cases of 3V 3a1 = 9 are equivalent to one of the cases of 3V 3a1 = 3;
(iii) all cases of 3V 3a1 = 6 are equivalent to one of the cases of 3V 3a1 = 0.
Then we can restrict to 3V 3a1 = 3, 0, . . . , 9. It is easy to check that the possibilities are
those listed in Table 6.
142
Appendix B. Embedding tables

Table 4
3V = (016 )
No.
GNA
3a1
16
0
1.1
6 10
1 ,0
12 4
1 ,0
1.2
1.3
1.4
2, 18 , 07
1.6
SU(6) SO(20)
2, 12 , 013
1.5
SO(32)
SU(12) SO(8)
SU(3) SO(26)
SU(9) SO(14)

2, 114 , 0
SU(15)
Table 5
3V = (16 ; 010 )
No.
2.1
2.3
2.4
2.5
2.6, 2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15
2.16
2.17
2.18
2.20
2.21
2.22
2.23
SO(20) SU(6)
6 6 4
0 ;1 ,0
2.2
2.19
GNA
3a1
6 10
0 ;0

6
0 ; 2, 12 , 07

6
0 ; 2, 18 , 0
1, 1, 04 ; 14 , 06
SO(8) SU(6)2
SO(14) SU(6) SU(3)
SU(9) SU(6)
SO(12) SU(4)2

1, 1, 04 ; 19 , 1

2
1 , (1)2 , 02 ; 12 , 08
2

1 , (1)2 , 02 ; 18 , 02

3
1 , (1)3 ; 010
3

1 , (1)3 ; 16 , 04
3 3 3 7
1 ,0 ;1 ,0
3 3 9
1 ,0 ;1 ,0

4
1 , 1, 0; 1, 09
4

1 , 1, 0; 17 , 03
6 10
1 ;0
6 6 4
1 ;1 ,0

2, 1, 04 ; 1, 09

2, 1, 04 ; 17 , 03
2, 1, (1)2 , 02 ; 15 , 05

2, 12 , (1)3 ; 13 , 07

2, 12 , (1)3 ; 19 , 0

2, 12 , 03 ; 010
SU(10) SU(4)
SO(16) SU(2)4
SU(8) SU(2)5
SO(20) SU(3)2
SU(6) SO(8) SU(3)2
SO(14) SU(3)3
SU(9) SU(3)2
SO(18) SU(4)
SU(7) SU(4)2
SO(20) SU(6)
SU(6)2 SO(8)
SO(18) SU(4)

SU(7) SU(4)2
SO(10) SU(5) SU(2)3
SO(14) SU(3)3
SU(9) SU(3)2
SO(20) SU(3)2
(continued)
143
Table 5 (continued)
No.
2.24
GNA
3a1

2
2, 1 , 03 ; 16 , 04
SU(6) SO(8) SU(3)2

2, 13 , 1, 0; 14 , 06

2, 13 , 1, 0; 19 , 1

2, 15 ; 13 , 07

2, 15 ; 19 , 0
2.25
2.26, 2.27
2.28
2.29
SO(12) SU(4)2
SU(10) SU(4)
SO(14) SU(6) SU(3)
SU(9) SU(6)
Table 6
3V = (112 ; 04 )
No.
3.2
3.3, 3.4
3.5
3.6
3.7, 3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
3.16, 3.17
3.18
3.19
3.20
3.21
3.22
3.23
3.24
3.25
3.26
GNA
3a1
12 4
0 ;0
3.1

12
0 ; 2, 12 , 0
SU(12) SO(8)
SU(12) SU(3)
1, 1, 010 ; 13 , 1

2
1 , (1)2 , 08 ; 12 , 02
3

1 , (1)3 , 06 ; 04
4

1 , (1)4 , 04 ; 13 , 1
5

1 , (1)5 , 02 ; 12 , 02

6
1 , (1)6 ; 04
3 9 3
1 ,0 ;1 ,0

4
1 , 1, 07 ; 1, 03
6

1 , (1)3 , 03 ; 13 , 0

7
1 , (1)4 , 0; 1, 03
6 6 4
1 ,0 ;0
7

1 , 1, 04 ; 13 , 1
8

1 , (1)2 , 02 ; 12 , 02

9
1 , (1)3 ; 04
9 3 3
1 ,0 ;1 ,0

10
1 , 1, 0; 1, 03
12 4
1 ;0

2, 1, 010 ; 1, 03
SU(10) SU(4)
SU(8) SU(2)5
SU(6) SO(8) SU(3)2
SU(4)4
SU(5)2 SU(2)4
SU(6)2 SO(8)
SU(9) SU(3)2
SU(7) SU(4)2
SU(6) SU(3)3
SU(7) SU(4)2
SU(6)2 SO(8)
SU(7) SU(4)2
SU(8) SU(2)5
SU(9) SO(8) SU(3)
SU(9) SU(3)2
SU(10) SU(4)
SU(12) SO(8)

2, 12 , (1)3 , 06 ; 13 , 0

2, 13 , (1)4 , 04 ; 1, 03

2, 15 , (1)6 ; 13 , 0
SU(10) SU(4)
SU(6) SU(3)3
SU(4)4
SU(6)2 SU(3)
(continued on next page)
144
Table 6 (continued)
No.
3.27
3.28, 3.29
3.30
3.31
3.32, 3.33
3.34
3.35
3.36
3.37
3.38, 3.39
3.40

3a1
2, 12 , 09 ; 04

2, 13 , 1, 07 ; 13 , 1

2, 14 , (1)2 , 05 ; 12 , 02

2, 15 , (1)3 , 03 ; 04

2, 16 , (1)4 , 0; 13 , 1

2, 15 , 06 ; 13 , 0

2, 16 , 1, 04 ; 1, 03

2, 18 , (1)3 ; 13 , 0

2, 18 , 03 ; 04

2, 19 , 1, 0; 13 , 1

2, 111 ; 13 , 0
GNA
SU(9) SO(8) SU(3)
SU(7) SU(4)2
SU(5)2 SU(2)4
SU(6) SO(8) SU(3)2
SU(7) SU(4)2
SU(6)2 SU(3)
SU(7) SU(4)2
SU(9) SU(3)2
SU(9) SO(8) SU(3)
SU(10) SU(4)
SU(12) SU(3)
Table 7
3V = (2, 12 ; 013 )
No.
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12, 4.13
4.14
4.15
4.16
3a1
3 13
0 ;0
3 6 7
0 ;1 ,0
3 12
0 ;1 ,0

3
0 ; 2, 12 , 010

3
0 ; 2, 18 , 04

1, 1, 0; 14 , 09

1, 1, 0; 110 , 03
3 3 10
1 ;1 ,0
3 9 4
1 ;1 ,0

2, 1, 0; 1, 012

2, 1, 0; 17 , 06

2, 1, 0; 112 , 1

2, 12 ; 013

2, 12 ; 16 , 07

2, 12 ; 112 , 0
GNA
SO(26) SU(3)
SO(14) SU(6) SU(3)
SU(12) SU(3)
SO(20) SU(3)2
SU(9) SO(8) SU(3)
SO(18) SU(4)
SU(10) SU(4)
SO(20) SU(3)2
SU(9) SO(8) SU(3)
SO(24)
SO(12) SU(7)
SU(13)
SO(26) SU(3)
SO(14) SU(6) SU(3)
SU(12) SU(3)
145
Table 8
3V = (2, 18 ; 07 )
No.
5.3
5.4
5.5
5.6
5.7
5.8
5.9
5.10
5.11, 5.12
5.13
5.14
5.15
5.16
5.17
5.18
5.19
5.22
5.23
5.24
5.25, 5.26
5.27
5.28
5.29
5.30
5.31
5.32
5.33
5.34
5.35, 5.36
5.37
5.38
SU(9) SO(14)
9 6
0 ;1 ,0
5.2
5.20, 5.21
GNA
3a1
9 7
0 ;0
5.1

9
0 ; 2, 12 , 04
1, 1, 07 ; 14 , 03
SU(9) SU(6)
SU(9) SO(8) SU(3)

2
1 , (1)2 , 05 ; 12 , 05
3

1 , (1)3 , 03 ; 07
3

1 , (1)3 , 03 ; 16 , 0
4

1 , (1)4 , 0; 14 , 03
3 6 3 4
1 ,0 ;1 ,0
4

1 , 1, 04 ; 1, 06
4

1 , 1, 04 ; 16 , 1
5

1 , (1)2 , 02 ; 15 , 02

6
1 , (1)3 ; 13 , 04
6 3 7
1 ,0 ;0
6 3 6
1 ,0 ;1 ,0
7

1 , 1, 0; 14 , 03
9 3 4
1 ;1 ,0

2, 1, 07 ; 1, 06

2, 1, 07 ; 16 , 1
2, 1, (1)2 , 05 ; 15 , 02
SU(7) SU(4)2
SO(10) SU(5) SU(2)3
SO(14) SU(3)3
SU(6) SU(3)3
SU(4)4
SU(6) SO(8) SU(3)2
SO(12) SU(4)2
SU(7) SU(4)2
SU(5)2 SU(2)4
SU(6) SO(8) SU(3)2
SO(14) SU(6) SU(3)
SU(6)2 SU(3)
SU(7) SU(4)2
SU(9) SO(8) SU(3)
SO(12) SU(7)

2, 12 , (1)3 , 03 ; 13 , 04

2, 13 , (1)4 , 0; 1, 06

2, 13 , (1)4 , 0; 16 , 1

2, 12 , 06 ; 07

2, 12 , 06 ; 16 , 0

2, 13 , 1, 04 ; 14 , 03

2, 14 , (1)2 , 02 ; 12 , 05

2, 15 , (1)3 ; 07

2, 15 , (1)3 ; 16 , 0

2, 15 , 03 ; 13 , 04

2, 16 , 1, 0; 1, 06

2, 16 , 1, 0; 16 , 1

2, 18 ; 07

2, 18 ; 16 , 0
SU(7)2
SU(5)2 SU(2)4
SO(8) SU(3)4
SO(12) SU(4)2
SU(7) SU(4)2
SO(14) SU(6) SU(3)
SU(6)2 SU(3)
SU(4)4
SO(10) SU(5) SU(2)3
SO(14) SU(6) SU(3)
SU(6)2 SU(3)
SU(6) SO(8) SU(3)2
SO(12) SU(7)
SU(7)2
SU(9) SO(14)
SU(9) SU(6)
146
Table 9
3V = (2, 114 , 0)
No.
3a1
15
0 ;0
6.1
6.2
6.3
6.4, 6.5
6.6, 6.7
6.8
6.9
6.10, 6.11
6.12
6.13, 6.14
6.15, 6.16
6.17, 6.18
6.19
6.20
6.21
6.22, 6.23
6.24, 6.25
6.26
6.27
6.28, 6.29
6.30
3

1 , (1)3 , 09 ; 0
6

1 , (1)6 , 03 ; 0
4

1 , 1, 010 ; 1
7

1 , (1)4 , 04 ; 1
6 9
1 ,0 ;0

9
1 , (1)3 , 03 ; 0
10

1 , 1, 04 ; 1
12 3
1 ,0 ;0

2, 1, 013 ; 1

2, 13 , (1)4 , 07 ; 1

2, 16 , (1)7 , 0; 1

2, 12 , 012 ; 0

2, 15 , (1)3 , 06 ; 0

2, 18 , (1)6 ; 0

2, 16 , 1, 07 ; 1

2, 19 , (1)4 , 0; 1

2, 18 , 06 ; 0

2, 111 , (1)3 ; 0

2, 112 , 1, 0; 1

2, 114 ; 0
GNA
SU(15)
SU(9) SU(3)2
SU(6)2 SU(3)
SU(10) SU(4)
SU(7) SU(4)2
SU(9) SU(6)
SU(9) SU(3)2
SU(10) SU(4)
SU(12) SU(3)
SU(13)
SU(7) SU(4)2
SU(7)2
SU(12) SU(3)
SU(6)2 SU(3)
SU(9) SU(6)
SU(7)2
SU(10) SU(4)
SU(9) SU(6)
SU(12) SU(3)
SU(13)
SU(15)
References
[1] D.J. Gross, J.A. Harvey, E. Martinec, R. Rohm, Phys. Rev. Lett. 54 (1985) 502.
[2] H.-P. Nilles, Phys. Lett. B 115 (1982) 193;
S. Ferrara, L. Girardello, H.-P. Nilles, Phys. Lett. B 125 (1983) 457.
[3] L. Dixon, J. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678.
[4] L. Dixon, J. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285.
[5] J. Giedt, Ann. Phys. (N.Y.) 297 (2002) 67, hep-th/0108244.
[6] A. Font, L.E. Ibez, F. Quevedo, A. Sierra, Nucl. Phys. B 331 (1990) 421;
J.A. Casas, C. Muoz, Phys. Lett. B 209 (1988) 214;
J.A. Casas, C. Muoz, Phys. Lett. B 214 (1988) 63;
G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, Phys. Lett. B 455 (1999) 135;
G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, Int. J. Mod. Phys. A 16 (2001) 425;
G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, J.W. Walker, Mod. Phys. Lett. A 15 (2000) 1191;
G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, J.W. Walker, Nucl. Phys. B 593 (2001) 471, hep-ph/0104091.
[7] M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117;
M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 585;
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
147
J.J. Atick, L. Dixon, A. Sen, Nucl. Phys. B 292 (1987) 109;

M. Dine, I. Ichinose, N. Seiberg, Nucl. Phys. B 293 (1987) 253.
M.K. Gaillard, J. Giedt, Phys. Lett. B 479 (2000) 308, hep-ph/0001219;
M.K. Gaillard, J. Giedt, Nucl. Phys. B 636 (2002) 365, hep-th/0204100;
M.K. Gaillard, J. Giedt, Nucl. Phys. B 643 (2002) 201, hep-th/0206249;
J. Giedt, Mod. Phys. Lett. A 17 (2002) 1465, hep-ph/0204017;
J. Giedt, hep-ph/0208004.
G.F. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 419, hep-ph/9801271.
L.E. Ibez, H.-P. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25;
L.E. Ibez, J. Mas, H.-P. Nilles, F. Quevedo, Nucl. Phys. B 301 (1988) 157.
J.A. Casas, M. Mondragon, C. Muoz, Phys. Lett. B 230 (1989) 63.
J. Giedt, Ann. Phys. (N.Y.) 289 (2001) 251, hep-th/0009104.
G. Aldazabal, A. Font, L.E. Ibez, A.M. Uranga, Nucl. Phys. B 452 (1995) 3, hep-th/9410206;
G. Aldazabal, A. Font, L.E. Ibez, A.M. Uranga, Nucl. Phys. B 465 (1996) 34, hep-th/9508033.
D.C. Lewellen, Nucl. Phys. B 337 (1990) 61;
A. Font, L. Ibez, F. Quevedo, Nucl. Phys. B 345 (1990) 389;
G.B. Cleaver, hep-th/9409096;
S. Chaudhuri, S.W. Chung, J.D. Lykken, hep-ph/9405374;
S. Chaudhuri, S.W. Chung, G. Hockney, J. Lykken, hep-th/9409151;
S. Chaudhuri, S.W. Chung, G. Hockney, J. Lykken, Nucl. Phys. B 456 (1995) 89, hep-ph/9501361.
K.R. Dienes, J. March-Russell, Nucl. Phys. B 479 (1996) 113, hep-th/9604112;
K.R. Dienes, Nucl. Phys. B 488 (1997) 141;
K.R. Dienes, hep-ph/9606467.
Z. Kakushadze, S.H. Tye, Phys. Rev. Lett. 77 (1996) 2612, hep-th/9605221;
Z. Kakushadze, S.H. Tye, Phys. Rev. D 54 (1996) 7520, hep-th/9607138;
Z. Kakushadze, S.H. Tye, Phys. Lett. B 392 (1997) 335, hep-th/9609027;
Z. Kakushadze, G. Shiu, S.H. Tye, Phys. Rev. D 54 (1996) 7545, hep-th/9607137;
Z. Kakushadze, G. Shiu, S.H. Tye, Nucl. Phys. B 501 (1997) 547, hep-th/9704113;
Z. Kakushadze, G. Shiu, S.H. Tye, Y. Vtorov-Karevsky, Int. J. Mod. Phys. A 13 (1998) 2551, hepth/9710149;
Z. Kakushadze, G. Shiu, S.H. Tye, Y. Vtorov-Karevsky, Phys. Lett. B 408 (1997) 173, hep-ph/9705202.
J.E. Kim, hep-th/0301177.

Higgs as a holographic pseudo-Goldstone boson

Roberto Contino a , Yasunori Nomura b , Alex Pomarol c
a Departamento de Fsica Terica C-XI, Universidad Autnoma de Madrid, Cantoblanco, 28049 Madrid, Spain
b Theoretical Physics Department, Fermi National Accelerator Laboratory, Batavia, IL 60510, USA
c IFAE, Universitat Autnoma de Barcelona, 08193 Bellaterra (Barcelona), Spain
Abstract
The AdS/CFT correspondence allows one to relate 4D strongly coupled theories to weakly coupled
theories in 5D AdS. We use this correspondence to study a scenario in which the Higgs appears as
a composite pseudo-Goldstone boson (PGB) of a strongly coupled theory. We show how a nonlinearly realized global symmetry protects the Higgs mass and guarantees the absence of quadratic
divergences at any loop order. The gauge and Yukawa interactions for the PGB Higgs are simple to
introduce in the 5D AdS theory, and their one-loop contributions to the Higgs potential are calculated
using perturbation theory. These contributions are finite, giving a squared-mass to the Higgs which
is one-loop smaller than the mass of the first KaluzaKlein state. We also show that if the symmetry
breaking is caused by boundary conditions in the extra dimension, the PGB Higgs corresponds to
the fifth component of the bulk gauge boson. To make the model fully realistic, a tree-level Higgs
quartic coupling must be induced. We present a possible mechanism to generate it and discuss the
conditions under which an unwanted large Higgs mass term is avoided.
PACS: 12.60.Fr; 12.60.Rc; 11.10.Kk
1. Introduction
Despite its tremendous phenomenological success, the standard model is almost
certainly not a fundamental theory of nature. Quantum instabilities of the Higgs potential
strongly suggest that it must be replaced by some other theory at energies not much
higher than the electroweak scale. Such a theory, for example, can be a supersymmetric
theory [1], a strongly coupled gauge theory [2], or a theory of quantum gravity [3]. In
E-mail address: roberto.contino@cern.ch (R. Contino).
doi:10.1016/j.nuclphysb.2003.08.027
R. Contino et al. / Nuclear Physics B 671 (2003) 148174
149
these theories the quadratic divergence of the Higgs squared-mass parameter is cutoff
either by embedding the Higgs boson into some larger multiplet or by giving it an internal
structure. A physical scale then exists, NP , at which many new particles appear revealing
the underlying symmetry or dynamics.
In this paper we consider a class of theories where the standard-model Higgs boson
arises as a composite particle of some strongly coupled dynamics. The dynamical scale
NP then must be parametrically larger than the scale of the Higgs mass in order to avoid
strong constraints coming from direct and indirect searches of new particles at colliders.
This suggests that the Higgs mass must be protected by some (approximate) symmetry
even below the scale NP . A natural candidate for such a symmetry is an internal global
symmetry under which the Higgs boson transforms non-linearly: the Higgs is a pseudoGoldstone boson (PGB) of the broken global symmetry. This situation is somewhat similar
to that of the pion in QCD, although for the Higgs there are additional requirements. It must
have a sizable quartic self-coupling and appropriate Yukawa couplings to the quarks and
leptons. In this paper we aim to build theories of this kind, which are well under control
as effective field theories, and in which some quantities are even calculable despite the
strongly interacting dynamics.
The basic observation is the following. Suppose we have a strongly coupled gauge
theory that produces the Higgs boson as a composite state. In general, it is quite difficult
to obtain quantitative low-energy predictions in such a theory because of non-perturbative
effects; one can at best derive estimates by using certain scaling arguments. This is indeed
the case if the gauge interaction is asymptotically free, as in QCD. However, it is not
necessarily true if the theory remains strongly coupled in the UV and approaches a nontrivial conformal fixed point. In this case it is possible that the theory, in the limit of
large number of colors, has an equivalent description in terms of a weakly coupled
5D theory defined on the truncated anti-de Sitter (AdS) space [4,5]. This allows us to
construct theories where the Higgs boson is interpreted as a composite state of a strong
dynamics and yet some physical quantities such as the Higgs potential can be computed
using perturbation theory.
The actual implementation of the above idea is quite simple, as far as gauge and Yukawa
interactions are concerned. These interactions explicitly break the global symmetry, but, as
we will see, they do not induce quadratic divergences for the Higgs mass at any loop order.
The global symmetry protects the Higgs mass parameter at high energies and predicts it to
be a one-loop factor smaller than the mass of the lightest resonance NP . This property
makes this class of theories quite interesting.
To make the model fully realistic, however, we must generate a sizable Higgs quartic
coupling. This is crucial not only to obtain a large enough physical Higgs mass, but
also to obtain the needed mas gap between the electroweak scale and NP (as suggested
by experiments). We will propose a mechanism that allows to generate a Higgs quartic
coupling at tree level. This mechanism needs specific assumptions on the form of
the explicit breaking of the global symmetry. However, these are assumptions on the
underlying physics around the Planck scale, and not on the TeV-scale physics which
yields the Higgs boson as a composite particle. Therefore, once the particular pattern of
breaking is assumed (which is not quite unnatural from the viewpoint of the 4D picture),
150
we can compute the Higgs potential generated at loop level through the explicit symmetrybreaking effects.
Since the cutoff of our theory is around the Planck scale, there is no obstacle in
extending the theory beyond NP , up to very high energies. In this respect, our framework
may be viewed as a way to provide a UV completion for little Higgs theories [6] in
which realistic models of the PGB Higgs have been constructed. It might be interesting to
construct a UV completion of the existing little Higgs models [7] using a warped spacetime
as outlined in the present paper.
In the next section we start by defining the framework in more detail. We describe the
basic structure of our theories in terms of both 4D and 5D pictures. An explicit model is
given in Section 3, in which the Higgs boson is identified with a PGB arising from a scalar
field located on the infrared brane. We present a possible mechanism to obtain a sizable
Higgs quartic coupling and discuss the size of quantum corrections to the Higgs potential,
which trigger electroweak symmetry breaking. In Section 4 we consider theories where
the Higgs boson arises from the extra-dimensional component of a gauge field in a warped
5D spacetime. We point out that also in this case the Higgs is interpreted as a composite
PGB in the 4D picture, and we present several realistic models. Conclusions are given in
Section 6.
2. Higgs as a composite pseudo-Goldstone boson

In this section we describe a class of theories where the standard-model Higgs arises as a
composite PGB of a strong interaction, and in which the presence of a weakly coupled dual
description allows the computation of certain quantities. We begin with the 4D description
of our theory, which we also refer to as the holographic theory. In this picture the theory
consists of two sectors. One is a sector of elementary particles that correspond to the
standard-model gauge bosons and (some of the) quarks and leptons. The other is a strongly
coupled conformal field theory (CFT) sector, where the conformal symmetry is broken
at low energies 1/L1 MPl . This sector produces CFT bound states due to the strong
dynamics at the scale 1/L1 . The Higgs will be one of these bound states.
To have a little hierarchy between 1/L1 and the Higgs mass, we require the Higgs to be
a Goldstone boson of the CFT sector. For this purpose, we assume that the CFT has a global
symmetry larger than the standard-model electroweak gauge group SU(2)L U(1)Y . We
find that it must be at least SU(3). If a global SU(3)L symmetry is spontaneously broken
to SU(2)L by the CFT strong dynamics, then 5 Goldstone bosons appear, a doublet and a
singlet under SU(2)L . The doublet will be associated with the Higgs boson. The SU(3)L
is not a symmetry of the standard-model gauge and matter fields which belong to the
elementary sector, so that the couplings of these fields with the CFT explicitly break the
global SU(3)L invariance. A mass for the Higgs boson is generated at loop level, which,
if negative, will trigger electroweak symmetry breaking. The loop factor appearing in the
Higgs mass can give a rationale for the little hierarchy between the electroweak scale and
the compositeness scale 1/L1 , although to perform quantitative computations we must go
to the weakly coupled dual description of the theory.
151
The AdS/CFT correspondence [4], as applied to a spontaneously broken CFT with a

UV cutoff [5], allows us to relate the above scenario to a theory in a slice of 5D AdS. In
this AdS picture the theory is weakly coupled and we can perform explicit calculations.
The metric of the spacetime is given by [8]
ds 2 =

1
dx dx dz2 gMN dx M dx N .
(kz)2
(1)
The 5D coordinates are labeled by capital Latin letters, M = (, 5) where = 0, . . . , 3;

z = x 5 represents the coordinate for the fifth dimension and 1/k is the AdS curvature
radius. This spacetime has two boundaries at z = L0 1/k 1/MPl (Planck brane) and
z = L1 1/TeV (TeV brane): the theory is defined on the line segment L0 z L1 .
The global symmetry of the 4D CFT is realized as a bulk gauge symmetry in the 5D
picture. In the case of a 4D theory where the CFT sector has a global SU(3)L invariance,
the dual theory is a 5D SU(3)L gauge theory. This SU(3)L is spontaneously broken by two
scalars. One living on the Planck brane and the other on the TeV brane. Being separated in
space, these scalars do not see each other at the classical level, so that the theory contains,
in the gaugeless limit, an enlarged global SU(3) SU(3) symmetry. By giving vacuum
expectation values (VEVs) to the two scalars, the global symmetry is spontaneously
broken, SU(3) SU(3) SU(2) SU(2), delivering two sets of 5 Goldstone bosons.
When we gauge the SU(3)L subgroup of SU(3) SU(3), the Goldstone bosons which
parametrize the SU(3)L /SU(2)L space are true Goldstone bosons and are eaten to form
massive gauge vectors. The remaining ones are PGBs, since gauge interactions do not
respect the full global SU(3) SU(3). In a slice of warped space the scalar living on the
Planck brane corresponds, to a very good approximation, to the true Goldstone boson. This
is because its VEV will be naturally of order MPl , much larger than the VEV of the scalar
on the TeV brane. Therefore, the scalar on the TeV brane corresponds to the PGB, which
we identify as the standard-model Higgs boson.
The holographic dual of this 5D setup is thus the 4D theory we described before:
the Higgs corresponds to the composite Goldstone boson of a CFT sector whose global
SU(3)L invariance is spontaneously broken down to SU(2)L by the strong dynamics. An
explicit breaking of the global CFT invariance is communicated by the interactions with
the elementary sector, and a mass for the Goldstone bosons is generated at one loop. The
spontaneous symmetry breaking of the CFT sector corresponds to the TeV-brane breaking
of the 5D theory, while the explicit breaking given by the elementary sector is associated
with the Planck-brane dynamics. This means that any process of the holographic theory
where the explicit breaking is communicated from the elementary sector to the CFT, will
correspond in the AdS picture to some kind of transmission from the Planck brane to the
TeV brane. The mass of the PGB is an important example: non-locality in the 5D theory
implies that it is a calculable and finite effect. This is a crucial ingredient of our class of
theories, which gives a rationale for explaining the little hierarchy. We will come back to
this point in the next section, where we compute the Higgs mass at one loop.
The scalar on the Planck brane can be replaced by boundary conditions that break
SU(3)L to SU(2)L on the Planck brane. The breaking on the TeV brane can also be realized
by boundary conditions. In this case, the PGB corresponds to the fifth component of the
gauge boson as we will see in Section 4.
152
A realistic theory must have Yukawa couplings between the Higgs and quarks and
leptons. It is simple to incorporate this feature in our theories. The fermions must be put
in the bulk (at least one of their chiralities) in representations of SU(3)L and be coupled to
the Higgs on the TeV brane. After the SU(3)L breaking on the Planck brane, we can obtain
the standard-model quarks and leptons as the only massless fermions (before electroweak
symmetry breaking). Large enough Yukawa couplings are obtained if the bulk fermion
masses are in a certain range such that they probe both branes. These Yukawa couplings
then induce a negative one-loop contribution to the Higgs mass term, which can trigger
electroweak symmetry breaking.
Models with the Higgs as a PGB face, however, a significant phenomenological
challenge. The Higgs quartic coupling H is induced only at one-loop level, giving a
physical Higgs mass, m2h = 2H H 2 , below the experimental bound of mh 114 GeV.
This is one of the major obstacles for the realization of the Higgs as a PGB. A realistic
model has to induce H at tree level. The challenge is to induce this coupling without
generating a tree-level mass term; otherwise the little hierarchy between the electroweak
scale and the compositeness scale is lost. Below we present a possible mechanism,
although, as we will explain, it requires further assumptions about the symmetry breaking
physics at high energies to keep the PGB massless at tree level.
3. A model
In this section we present an explicit model that leads to the standard model with
the Higgs boson as a PGB. This represents a concrete example for the general theories
introduced in the previous section. We discuss an alternative possibility in the next
section, where the Higgs is identified with the extra-dimensional component of the gauge
boson, A5 .
We consider a 5D theory in a slice of AdS with a gauge symmetry SU(3)L U(1)X ,
which contains the electroweak SU(2)L U(1)Y as a subgroup. The extra U(1)X is
introduced to give the correct hypercharges to the fermions. All the gauge bosons are
assumed to have Neumann boundary conditions at both branes (we will work in the unitary
gauge A5 = 0). The matter content of the model is the following (we only consider the
third-generation quark sector for simplicity, but the extension to the full standard model is
straightforward). We introduce two bulk fermions Q and D which transform as 31/3 and 30
under SU(3)L U(1)X . Since the bulk fermions are in the Dirac representation, they
are decomposed into the left-handed, L , and right-handed, R , components in terms of
the 4D irreducible representation (Weyl fermion): Q = QL + QR and D = DL + DR . The
boundary conditions for these fields are chosen such that QL and DR (QR and DL ) obey
Neumann (Dirichlet) boundary conditions at the Planck and TeV branes. This implies that
only QL and DR have massless zero modes. We also introduce the boundary fermion UR
on the TeV brane, which transforms as 12/3 under SU(3)L U(1)X .
We assume that SU(3)L U(1)X isbroken on the Planck brane to the standard-model
group SU(2)L U(1)Y , with Y = T8 / 3 + X. This breaking can be caused, for example,
by a scalar S on the Planck brane with quantum numbers 31/3 under SU(3)L U(1)X . The
153
bulk fermion fields are decomposed under the standard-model group as

(S)
(D)
21/6 + Q(S)
Q 31/3 = Q(D)
L 21/6 + QL (12/3 ) + QR
R (12/3 ),
(D)
(S)
(D)
(S)
D(30 ) = DL (21/6) + DL (11/3 ) + DR (21/6 ) + DR (11/3 ).

We assume that the symmetry breaking dynamics at the Planck brane is such that only Q(D)
L
and DR(S) are left as 4D massless fields, and we identify these fields with the standardmodel doublet quark qL and singlet bottom quark bR .1 The standard-model singlet top
quark, tR , is identified with the brane field UR . When the Yukawa couplings on the TeV
(D)
(D)
brane are included, however, qL will be a mixture of QL and DL states, while tR will be
(S)
a mixture of UR and QR states, as we will see below. The color SU(3)C can be introduced
in a straightforward way: Q, D and U all transform as 3 of SU(3)C .
What about the Higgs field? To have the Higgs as a PGB, we introduce a scalar field
on the TeV brane transforming as 31/3 under SU(3)L U(1)X . We assume that gets a
VEV, say by the potential L = (z L1 ) gind ( v 2 )2 where gind is the induced

metric on the brane. We can then parametrize the scalar field as

0
1 0 H
iT a Ga
(31/3 ) = e
(2)
0
,
T a Ga =
,
v H
v +
where the fields H and are PGBs that transform respectively as doublet and singlet of
SU(2)L , and is a real scalar field; v is the VEV of in terms of the 5D metric. We also
define, for later convenience, the VEV in terms of the 4D metric v vL
0 /L1 , which takes
a value of the order of the local cutoff on the TeV brane, IR TeV. The unbroken group
under the VEV can naturally be aligned with that under the S VEV, due to possible
radiative effects relating them. We then find that the PGB field H has the appropriate
SU(2)L U(1)Y quantum numbers, 21/2 , to be identified as the standard-model Higgs
boson.
We now proceed to the interaction terms of the theory. The 5D theory has mass terms
for the bulk fermions:
L QR MD D
L DR ],
L = g [Lkin MQ Q
(3)
where Lkin are the kinetic terms (see Eq. (37) for the explicit expression). These mass terms
control the shape of the wavefunctions of the QL and DR zero modes, and hence the size
of the various low-energy 4D couplings. In addition, we introduce the Yukawa couplings
for the matter fields on the TeV brane:

L UR + D i Q
j DRk . ij k + h.c. ,
L = (z L1 ) gind U Q
(4)
L
1 This situation can be realized, for example, by adding the following couplings on the Planck brane:
R D + S D
D , where U (12/3 ), D (30 ) and D (11/3 ) are extra quarks that marry with
L U + M D
SQ
R
L
L R
R
L
R
(D)
(S)
the unwanted SU(3)L partners of the zero modes of QL and DR . This symmetry breaking pattern can be
more easily obtained if we choose to break SU(3)L through boundary conditions. The details of this breaking,
however, are not relevant for physics at the TeV scale.
154
where i, j, k represent the SU(3)L index. After integrating out the KaluzaKlein (KK)
states, we obtain the following effective Lagrangian:
L4D = ZH |D H |2 + iZQ qL/ qL + i b R / bR + iZU tR / tR
iU fQ H qL tR + iD fQ fD H qL bR + h.c.
(5)
(D)
(S)
Here, fQ (fD ) denotes the value of the zero-mode wavefunction of QL (DR ) at the TeV
brane; one has
(12ci )

L1
1 L1 (1/2ci )
L0
2
,
N0 =
1 ,
fi =
(6)
N0 L0
(1 2ci ) L0
where the sign holds for the zero mode of a left- (right-)handed field and ci = Mi /k for
i = Q, D. The factors ZH , ZQ and ZU arise respectively due to the mixing of H with the
(D)
(D)
heavy KK gauge bosons, the mixing of the zero mode of QL with the KK states of DR ,
(S)
and the mixing of UR with the KK states of QL . These mixings appear when gets a
VEV, and we find
(g5 v)
2
,
ZQ = 1 + |fQ D v|
2 GD (D) ,
R
4k
ZU = 1 + |U v|
2 GQ(S) .
1
=1+
ZH
(7)
(D)
Here, GQ(S) (GD (D) ) is the 5D propagator of Q(S)
L (DR ) evaluated on the TeV brane
L
(,+) (L1 , L1 ; 0) and G (D) = G

(,+) (L1 , L1 ; 0),
at zero 4D momentum: GQ(S) = G
L
R
D
L
(,+) (z, z ; p) can be found in Appendix A. After canonically

where the propagators G
L,R
normalizing the fields we obtain the top and bottom Yukawa couplings
ht =
iU fQ
,
ZH ZQ ZU
iD fQ fD
hb =
.
ZH ZQ
(8)
The Yukawa couplings strongly depend on MQ and MD . For example, the dependence of
the top Yukawa coupling on MQ is given by
(L /L )|cQ |1/2 ,
for |cQ | > 1/2,
0 1

1/2
ht
(9)
log(L1 /L0 )
, for |cQ | = 1/2,
O(1),
for |cQ | < 1/2,
and therefore the theory can be realistic only if |MQ | k/2. It can also be shown that
|MD | k/2 is also needed to obtain realistic Yukawa couplings.
The holographic picture offers a simple explanation for the behavior of Eq. (9). For
cQ > 1/2 the holographic theory consists of a CFT sector coupled to a left-handed
dynamical source L which transforms as a doublet of SU(2)L [9]:
L = LCFT + i L/ L + k 1/2cQ L OR + h.c. + .
(10)
Here, is a dimensionless coupling, and the ellipses stand for higher order operators (MPl suppressed), and gauge and bottom interactions. The coupling L OR induces a tree-level
155
mixing between the elementary source and the CFT bound states; its strength is determined
by the AdS/CFT correspondence, which relates the dimension of OR with the value of cQ :
dim[OR ] = 3/2 + |cQ + 1/2|. For cQ > 1/2 the coupling is always irrelevant, so that the
physical massless eigenstate qL , to be identified with the standard-model quark doublet, is
almost the elementary state L . The quark singlet tR appears in the theory as a composite
CFT state. For cQ < 1/2 the holographic picture is very different. We find that the 4D
theory must contain a right-handed dynamical source R which is a singlet of SU(2)L [9]:
L = LCFT + i R / R + k 1/2+cQ R OL + h.c. + .
(11)
The AdS/CFT correspondence requires dim[OL ] = 3/2 + |cQ 1/2|, and the coupling
between the elementary source and the CFT is again irrelevant for cQ < 1/2. The
physical qL appears now as a composite CFT state, while tR is almost the elementary
state R .
How does the Yukawa coupling between the composite Higgs H and the physical quarks
qL and tR arise? As is clear from Eqs. (10) and (11), the global SU(3)L invariance of the
CFT is not a symmetry of the elementary sector, and the explicit breaking is communicated
to the conformal sector through its coupling with the source L or R . The Yukawa
coupling is then generated only through the insertion of the composite operators OL,R ,
since the process must involve the elementary source. This implies that the top Yukawa
coupling must be proportional to k 1/2|cQ | , and therefore ht (L1 /L0 )1/2|cQ | as
obtained in the 5D picture, Eq. (9).
For 1/2 cQ 1/2 either of the two descriptions, Eq. (10) or Eq. (11), is valid.
In this case the coupling of the CFT to the elementary sector is relevant (marginal for
cQ = 1/2). This implies that the both physical quarks qL and tR are almost composite
states. The only way to generate the top Yukawa is then through the virtual exchange of the
source. The source has now a relevant coupling with the CFT, so that the CFT correction to
the elementary propagator becomes important. By resumming the infinite series of CFT
insertions, one can express this correction as a renormalization of the coupling ; for
example, in the holographic picture of Eq. (10), the conformal invariance gives
2 ()
2 (k)
.
1 + 2 (k)(2 /k 2 )cQ 1/2
(12)
Therefore, for |cQ | < 1/2 and k, one has () (/k)1/2cQ , so that the top
Yukawa coupling, proportional to 2 ()k 12cQ is always of order one. The particular case
|cQ | = 1/2 gives a logarithmic suppression ht [ln(L1 /L0 )]1/2 .
Summarizing, we have obtained an SU(3)C SU(2)L U(1)Y gauge theory with
a massless Higgs boson, H , a massless quark doublet, qL , and two massless singlet
quarks, tR and bR , which have the Yukawa couplings of Eq. (8).2 The fermion content
thus reproduces the third generation quark sector of the standard model. To complete the
standard-model structure, however, we have to discuss the Higgs potential. We address this
remaining issue in the next two subsections.
2 In the model presented here, there is also an extra singlet PGB , which obtains a mass at one-loop level.
156
(a)
(b)
Fig. 1. One-loop corrections to the PGB mass in the holographic theory from the gauge field (a) and an elementary
fermion (b). If the coupling of the source with the conformal sector is relevant, then the fermion propagator in
the diagram (b) must be intended as corrected by an infinite series of CFT insertions.
3.1. One-loop contribution to the Higgs potential

At tree level, the Higgs potential vanishes due to the SU(3)L symmetry on the TeV
brane. At one-loop level, however, an effective potential for H will be induced. In the
holographic picture this comes from the interactions between the CFT and the elementary
fields, which explicitly break the global SU(3)L symmetry. At one loop the relevant
diagrams for the Higgs mass term are those sketched in Fig. 1. The situation is quite similar
to QCD, where the charged pion gets a mass at one loop due to the fact that the coupling to
the photon explicitly breaks the global chiral symmetry. In the 5D picture the effect comes
from loops of bulk fields that propagate from the TeV brane, where H lives, to the Planck
brane, where we have the breaking of SU(3)L . This is a non-local effect and therefore is
finite. The one-loop effective potential is similar to that calculated in Ref. [10].
The gauge contribution to the Higgs potential arises from loops of SU(3)L gauge
bosons. This is given by (the U(1)X contribution can be obtained in a similar way)
3
gind Vgauge() =
Tr
2
n=1
3
Tr
2
n
dp 3 (1)n+1
G M2 ()
p
2
8
n

dp 3
p ln 1 + G M2 () .
8 2
(13)
Here, M2 is the boundary squared-mass matrix of the gauge boson fields in the
background :
M2ab () = 2g52 Ta Tb ,
(14)
and G is a matrix propagator Gab = Ga (p)ab , where Ga (p) are the propagators of the
SU(3)L gauge boson (a = 1, . . . , 8) evaluated on the TeV brane with 4D momentum p:
1 , L1 ; p) where G(z,
z ; p) is the rescaled gauge boson propagator given in
Ga (p) = G(L
Appendix A with m1 = 0 and m0 MPl (m0 = 0) for the gauge bosons of the broken
(unbroken) generators on the Planck brane. The effective potential of Eq. (13) has the
ColemanWeinberg potential form, with the only difference that 4D propagators have
been replaced by 5D propagators. Plugging Eq. (2) into Eq. (13), we can obtain the gauge
contribution to the effective potential of H . In particular, the mass of H for vL1 1 is
157
given by
m2H

9g52 L1 2
dp 3
2
1
=
p GIII (p) GII (p) GI (p) ,
4 L0
8 2
3
3
(15)
where GIII,II,I (p) are the SU(2)L triplet, doublet and singlet components of the SU(3)L
gauge boson propagators. The integrand in Eq. (15) reaches its maximum at p 1/L1 and
is exponentially suppressed for p > 1/L1 . Therefore, m2H is finite and very insensitive to
physics at energies above 1/L1 . Notice that, contrary to supersymmetry, the cancellation
of quadratic divergences in Eq. (15) is due to particles of equal spin. Eq. (15) yields

0.12 2
2
.
mH
(16)
L1
Hence, the PGB mass turns out to be an order of magnitude smaller than the scale 1/L1 ,
which is also smaller than IR . This mass gap is even larger if mH is compared with the
mass of the first KK state, NP /L1 . The contribution of Eq. (16) is positive and cannot
trigger electroweak symmetry breaking by itself.
The top contribution to the effective potential is given by
gind Vtop() = 6

dp 3
p ln 1 + Gi (p)m2i () ,
2
8
(17)
where
m2i
is the boundary squared-mass of Q(i)

L in the background :
m2i () = |U |2 i i ,
(18)
and Gi (p) are the propagators of Q(i)

L (i = D, S) evaluated on the TeV brane at 4D

(,+) (L1 , L1 ; p),
momentum p. These are GD (p) = G(+,+)
(L1 , L1 ; p) and GS (p) = G
L
L
(,+)

where G
(z, z ; p) can be found in Appendix A. This gives a contribution to the Higgs
L
mass of order
m2H
h2t 1
,
2 L21
(19)
where the exact value depends on MQ . The top contribution is negative and, for certain
values of MQ , is larger than the gauge contribution. The top contribution can then be
responsible for the breaking of the electroweak symmetry.
Despite the occurrence of electroweak symmetry breaking, the one-loop effective
potential calculated above cannot by itself lead to a realistic scenario of electroweak
symmetry breaking. A tree-level Higgs quartic coupling is necessary to obtain a large
enough physical Higgs mass and to generate a Higgs VEV a loop factor smaller than 1/L1 .
In the next subsection we will provide a mechanism to generate a Higgs quartic coupling,
which together with the one-loop gauge and top contributions can lead to a realistic theory
of electroweak symmetry breaking.
Before concluding this subsection, it is interesting to analyze the absence of quadratic
divergences in this class of models from a 4D perspective. In the standard model, the
158
dominant contribution to the Higgs mass term comes from the loop of the top quark and it is
quadratically divergent. The cancellation of this divergence then must arise from a loop of
some extra fields. This is indeed what happens in our case. The top quark is realized as the
zero mode of a bulk fermion and it corresponds, in the holographic picture, to a mixture
between the elementary source and the CFT bound states. The physical spectrum then
consists of a massless top quark, plus a series of CFT bound states that form a complete
multiplet of the global SU(3)L . It is the contribution of this tower of additional states that
cancels the divergence of the top loop. This is, therefore, a cancellation involving an infinite
numbers of 4D fields.
The picture described above, however, is not quite illuminating to understand the
finiteness of the top contribution to the Higgs mass. To understand it better, we can perform
a change of basis going from the mass eigenstate basis to the interaction basis, where we
separate an elementary source field from the tower of composite CFT states (the physical
top quark is a mixture of these states). The whole contribution to the Higgs mass then arises
from an exchange of the elementary field (Fig. 1), because only the elementary sector feels
the explicit breaking directly. Now comes the most important point. Since the elementary
field couples only linearly to the CFT sector, the correction to the Higgs mass cannot
proceed through a loop of elementary modes directly coupled to the Higgs, but it must
necessarily involve the strong CFT dynamics. Therefore, a large momentum circulating
in the top loop always flows into the CFT cloud, and consequently the resulting Higgs
mass always involves a form factor F (q 2 ) which encodes the non-perturbative effects
of the CFT. Since F (q 2 ) is suppressed for q larger than the compositeness scale, the
loop momentum integral is cutoff above that scale. This is the reason why the quadratic
divergences are absent in our theory. In this picture, no cancellation between different
divergent contributions is necessary. The Higgs mass correction is finite simply because of
the form factor suppressing the contribution from large virtual momenta: at high energies
the constituents of the composite Higgs become transparent to the short wavelength probe
of the elementary fermion, so that F (q 2 ) 0 for q 2 . As the explicit computation
reveals, the damping is exponential. This strong damping can be understood by recalling
that in the 5D theory the mass correction arises as a finite non-local effect. As such, it
involves a brane to brane propagation, and that explains the exponential suppression for
internal momenta larger than 1/L1 .
A similar scenario, which does not have quadratic divergences for the Higgs, was
considered in Ref. [10]. In that model too, the Higgs is a bound state of the CFT,
and its tree-level mass vanishes since the conformal sector is invariant under a global
supersymmetry. This invariance is explicitly broken by the interactions with the elementary
sector, so that the Higgs is expected to acquire a mass at one loop. The 5D realization
of this scenario consists of a warped supersymmetric setup, where matter and gauge
fields propagate in the bulk of AdS and the Higgs field is localized on the TeV brane.
If supersymmetry breaking is triggered on the Planck brane, the correction to the Higgs
mass will be a finite non-local effect. The corresponding holographic description is almost
the same as in our case: the Higgs mass is generated through the exchange of some
elementary field and it is finite because the CFT strong dynamics exponentially suppresses
contributions from large virtual momenta.
159
Fig. 2. Contributions to the Higgs quartic coupling. The CFT dynamics is represented as a thick gray line, while
a thin black line represents the propagator of the elementary scalar . A cross indicates an SU(3)L breaking
by the CFT.
3.2. A mechanism for the quartic coupling

Although the Higgs potential is generated by radiative corrections, it is not sufficient
to guarantee a successful phenomenology. We need a large O(1) quartic coupling
while keeping the quadratic term smaller than the effective cutoff scale, i.e., the scale that
suppresses higher-dimensional terms in the Higgs potential. Here we present an example
of dynamics providing such a situation.
The basic idea is simpler to understand in the 4D CFT picture. The explicit breaking of
SU(3)L comes from the elementary sector. Hence, in order to generate a quartic coupling
at tree level, the Higgs must mix with some elementary scalar . Since the mass of the
elementary scalar is not protected by any symmetry, it is expected to be of order the cutoff
scale k. This implies that, if we want to generate an unsuppressed SU(3)L breaking
effect in the Higgs sector, the elementary scalar must be coupled to the CFT with a
coupling linear in k: Lint = k O , i.e., the operator O must have dimension 2. If
2 |0 = 0, where O is an operator that creates , the Higgs will have non-trivial
0|O O
interactions with the scalar , and an explicit breaking of SU(3)L will appear in the Higgs
potential at tree level. For example, if is a 6 of SU(3)L , which decomposes under SU(2)L
as a triplet (T ), a doublet (D ) and a singlet (S ), a Higgs quartic coupling is generated
from the diagrams of Fig. 2. We are assuming here that the different SU(2)L components
of have different masses due to the SU(3)L breaking in the elementary sector; otherwise
the quartic coupling is zero because of the non-linearly realized SU(3)L symmetry. In this
theory, however, there is the danger of also generating a quadratic term for H . This comes
from the diagrams of Fig. 3. A way to avoid the generation of a Higgs squared-mass is to
assume that the breaking of SU(3)L in the elementary affects only T and not the other
SU(2)L components, D and S . It is clear that in this case, the diagrams of Fig. 3 respect
SU(3)L and vanish. Therefore, a mechanism of this type must requires a certain pattern of
SU(3)L breaking in the elementary sector.3
Let us see how this idea is implemented in the 5D AdS picture. By AdS/CFT, a scalar
operator of dimension 2 corresponds to a bulk scalar of squared-mass M2 = 4k 2 . Note
that as long as M 4k 2 (and certain conditions for the brane masses are met), is not
tachyonic and does not develop a VEV. The 5D scalar is responsible for communicating
the SU(3)L breaking on the Planck brane to the TeV brane where the Higgs lives.4 The
3 We could have a Higgs quartic coupling without quadratic term if only is present. Nevertheless, in this
T
case the quartic coupling turns out to be negative.
4 Another possibility, which could avoid introducing , is to have a Higgs with a profile in the extra dimension.
160
Fig. 3. Contributions to the Higgs mass. The CFT dynamics is represented as a thick gray line, while a thin black
line represents the propagator of the elementary scalar . A cross indicates an SU(3)L breaking by the CFT.
field must be coupled to the Higgs on the TeV brane. We thus require to transform as
62/3 under SU(3)L U(1)X and have the following coupling:

L = (z L1 ) gind + h.c. ,
(20)
where is a coupling of mass dimensions 1/2. A small deviation from M2 = 4k 2 , as

arising for example from the one-loop correction to the bulk mass of , will not spoil our
mechanism.
By integrating out the field, we find a tree-level Higgs potential in the low-energy
effective theory,
VH,tree = m2H |H |2 +
H
|H |4 .
2
(21)
m2H and H are generated by 5D diagrams similar to those in Figs. 3 and 2 respectively,
where the internal propagators now represent the 5D field with the appropriate SU(2)L
quantum numbers. We find
S G
D ],
m2H = 22 v 2 [G
(22)
2
D 5G
S 3G
T ],
H = 2 [8G
3
(23)
S , G
D and G
T are the rescaled propagators of the SU(2)L singlet, doublet
where G
and triplet components of with end points on the TeV brane, evaluated at zero 4D
momentum. The explicit form of these propagators can be found in Appendix A. For
M2 4k 2 , which we assume here, they are given by

a 1 1 + (m0,a L0 2) ln L1 ,
G
ra
L0
(24)
1
where ra m0,a + m1 + L0 (m0,a 2L1
0 )(m1 + 2L0 ) ln(L1 /L0 ) and a = S, D, T . We
introduced a common scalar mass m1 on the TeV brane for the SU(2)L singlet, doublet and
triplet components of field, but allowed different masses m0,a on the Planck brane. The
values of the masses m0,a are determined by the high-energy SU(3)L -breaking dynamics
on the Planck brane. Here we do not specify a particular pattern for this symmetry breaking,
but rather we look for the parameter region of m0,a (and m1 ) where the successful Higgs
phenomenology is obtained.
161
Inserting Eq. (24) into Eq. (22), we obtain

m2H
22
(m0,D m0,S )v 2 .
rD rS
(25)
We see that, as expected, if the doublet and singlet components of do not feel the
breaking of SU(3)L on the Planck brane, m0,D = m0,S , the resulting Higgs squared-mass
parameter is zero. Assuming this, the quartic coupling, Eq. (23), is given by
H
22
(m0,T m0,S ).
rT rS
(26)
Therefore, for m0,T m0,S , we obtain a sufficiently large Higgs quartic coupling.
We have seen that the desirable Higgs potential is obtained for m0,T m0,S m0,D .5
Although this may appear a rather ad hoc hypothesis, we stress that it is an assumption
about the underlying physics at the Planck scale which is responsible for the symmetry
breaking. We do not explicitly address this sector here, but we expect that there are
some mechanisms realizing (approximately) the situation discussed above. We must
say, however, that even if m0,S = m0,D at tree level due to some specific SU(3)L
breaking pattern on the Planck brane, quantum effects (coming, for example, from gauge
interactions) will modify this relation. Therefore, a Higgs squared-mass will be induced
from Eq. (25), at least, at the quantum level. This contribution is difficult to calculate since
it depends on the Planck-brane fields that break SU(3)L , but it can be estimated to be oneloop factor smaller than v 2 . We can then conclude that the Higgs potential Eq. (21) together
with the one-loop contributions of Eqs. (13) and (17) can lead to a successful electroweak
symmetry breaking with a VEV for the Higgs a loop factor smaller than 1/L1 , and a
physical Higgs mass larger than the experimental bound. The precise determination of H
and mh is, however, not possible here due to the dependence of H on the unknown free
parameters of the model.
4. Pseudo-Goldstone bosons as holograms of A5

In this section we consider the possibility of breaking the gauge symmetry in the warped
extra dimension by boundary conditions, and not by scalar fields on the two branes. We will
show that the holographic picture is almost the same as before and that the holographic
PGB in this case corresponds in the 5D theory to the zero mode of the fifth component of
the gauge boson, A5 . We note that the model presented in the previous section and the one
presented here must coincide in the limit v IR , since the breaking of SU(3)L by the
scalar is equivalent to a breaking by boundary conditions in the limit v [11].
Let us consider the most general case of a bulk gauge group G reduced to the subgroups
H0 and H1 on the Planck and TeV branes, respectively. This corresponds to assigning the
5 An alternative parameter region is m
0,D , m0,S k m0,T , m1 , in which case the tree-level Higgs potential
is given by m2H 2 (m0,D m0,S )(2 ln(L1 /L0 )2 m21 )1 v 2 and H 22 m0,T /(m1 rT ) so that we can have
m2H v 2 , H 1.
162
Fig. 4. The holographic theory consists of a CFT interacting with an elementary sector represented here by the
gauge fields Aa , Aa and a generic field . The Goldstone bosons a are eaten by the gauge fields Aa to form
massive vectors; the remaining a are PGBs.
Fig. 5. The pattern of symmetry breaking.
following boundary conditions to the gauge bosons at the Planck and TeV branes:
Aa (+, +),
T a Alg{H },
Aa (+, ),
T a Alg{H0 /H },
Aa (, +),
T a Alg{H1 /H },
Aa (, ).
(27)
Here, by + () we denote the Neumann (Dirichlet) boundary condition, and H = H0 H1 .

The A5 s have the opposite boundary conditions to those of the corresponding A s.
The holographic 4D theory consists of a CFT sector whose global invariance G
is spontaneously broken down to H1 by strong dynamics, with an order parameter of
O (TeV). External gauge fields weakly gauge the subgroup H0 of G:
L = LCFT
1 2
F
+ A J ,
4g 2
= a, a.
(28)
This situation is somewhat different from the model of Section 3, where the whole G
was gauged in the 4D theory and higgsed down to H0 at high energies. The two scenarios,
however, are indistinguishable from low-energy observers. The gauging of only a subgroup
of the global symmetry G is experienced by the CFT as an explicit breaking of G.
Let us count the number of PGBs present in the theory. The spontaneous breaking in
the CFT sector delivers n = dim(G/H1 ) Goldstone bosons. However, the gauging of H0 ,
Eq. (28), makes part of them, m = dim(H0 /H ), being eaten by the gauge bosons Aa . The
remaining n m are PGBs; they are massless at tree level, but they acquire masses from
radiative corrections due to the explicit breaking of G by the interaction terms of Eq. (28).
163
Only the gauge bosons associated to the symmetry subgroup H = H0 H1 are exactly
massless. Figs. 4 and 5 give a pictorial representation of the holographic theory and of the
symmetry breaking pattern.
In general, any interaction with the elementary sector can communicate the explicit
breaking of G to the CFT sector. The AdS/CFT correspondence prescribes that adding a
generic field in the bulk of AdS with Neumann boundary conditions on the Planck brane
corresponds to modifying the CFT content and adding some elementary 4D field which
couples to the conformal sector through a coupling O (see Fig. 4). Since will come in
a representation of the group H0 (rather than G), which is the symmetry of the elementary
sector, the coupling O is not G-invariant. As the only source of explicit breaking is
represented by those interactions, this necessarily implies that the mass terms for the PGBs
are generated only through processes where elementary fields are exchanged (Fig. 1).
Since the 4D holographic picture and the 5D AdS setup describe the same physics, they
must exhibit the same physical spectrum. Therefore, there must be n m massless scalars
in the 5D theory after KK reduction. We now see in detail how these massless scalars
appear. The 5D gauge Lagrangian is given by
Lgauge = g 2 g KM g LN FKL FMN + LGF ,

(29)
4g5
where the metric gMN is that of Eq. (1) and LGF is the gauge-fixing term. A convenient
choice for LGF is
2
1
LGF =
(30)
g A + zg 55 z (A5 /z) ,
2
2g5
with which all mixing terms between A and A5 cancel [12]. Eq. (29) can be written as

1
Lgauge = 2
A (1 1/ ) A + (z A )2
2g5 kz

A5 2
+ ,
(31)
+ ( A5 )2 z2 z
z
where the ellipses stand for interaction terms; 4D indices are raised/lowered with .
We now perform a KK reduction. We are interested only in the massless scalar spectrum.
This comes from the zero modes of the fifth components of the gauge bosons, A5 (x, z) =
(0)
f0 (z)A5 (x) + , where f0 (z) satisfies

f0 (z)
z
(32)
= 0.
z
The solution to this equation is only compatible with (+, +) boundary conditions, in which
case we obtain

L21 L20
z
.
, where N0 =
f0 (z) =
(33)
N0
2
The components of A5 with (+, +) boundary conditions are Aa5 . There are dim[G/H1 ]
dim[H0 /H ] of them, as expected from the 4D dual picture. Therefore the massless
164
modes of Aa5 must correspond to the PGBs of the 4D CFT [13]. A further check of this
correspondence comes from their wavefunction. It is peaked towards the TeV brane as
expected if the holographic PGBs are really bound states of the CFT. The excited KK
modes of A5 can be eliminated from the spectrum by going to the unitary gauge = .
It is clear from the 5D point of view that a tree-level potential for A5 is absent, because
it is forbidden by gauge invariance. An effective potential is then generated radiatively
L
as a function of the non-local, gauge invariant Wilson line W = Tr P exp(i L10 dz A5 ).
This implies that A5 will get a mass at one loop, which, by non-locality, is finite and
cutoff insensitive. In general, the energy is minimized at a non-zero background value of
A5 , triggering spontaneous breaking of the symmetry H [14]. As in flat space, different
values of the background A5 = f0 (z)ca T a , with ca = const, define physically inequivalent
vacua. This is because one cannot find a continuous gauge parameter (z) = a (z)T a
with the defined boundary conditions (, ) that eliminates A5 by the use of the gauge
transformation A5 U A5 U + U z U . Nevertheless, by relaxing one of the two
boundary conditions, e.g., that on the Planck brane, it is possible to gauge away A5
by the transformation with
z
(z) =
dz A5 .
(34)
L1
Under this gauge transformation, however, the charged bulk fields are also transformed:
(x, z) ei(z) (x, z). This implies that the theory with A5 = 0 is equivalent to that
with nonzero background A5 , but only if we use the redefined bulk fields ei(z)
instead of . The Planck-brane boundary conditions of the fields are then different
from those of the fields , since and differ at z = L0 by a non-trivial gauge phase:
(L0 ) = ei(L0 ) (L0 ).
(35)
This phase is the Wilson line.

The one-loop contributions to the effective potential of A5 are easily estimated as
follows. The appearance of the Wilson line in the vacuum energy requires a contribution
of a bulk field that propagates from one brane (at L0 ) to the other (at L1 ). The energy
involved in this contribution is
Lthen of the order of the inverse of the conformal distance
between the branes, E 1/ L01 dz 1/L1 . Therefore, the mass of A5 is estimated to
z ; p) is the propagator of
0 , L1 ; p)|Ep1/L1 , where G(z,
be m2A5 g52 (L1 /L0 )2 E 4 G(L
the bulk field given in Appendix A. It is interesting to look at the limit L0 0. In this
case the propagators from L0 to L1 for the gauge boson and the graviton vanish, implying
that no effective potential is induced for A5 . We thus find that A5 s are massless in this
limit at all loop orders. Notice that, contrary to the flat space case, the zero modes of A5
are still normalizable modes even though the extra dimension is infinite (see Eq. (33)),
and thus remain in the theory as massless scalars. This is in fact what we expect from
holography. In the 4D picture, the limit L0 0 corresponds to sending the UV cutoff
to infinity. This implies that the 4D low-energy gauge coupling becomes zero (gaugeless
limit), and the gauge bosons that explicitly break G decouple from the theory, making the
PGBs true Goldstone bosons. Note that in the gaugeless limit the number of Goldstone
165
bosons is n = dim(G/H1 ) instead of n m. In the 5D AdS the m extra massless scalars

come from Aa5 ; they have (, +) boundary conditions and admit zero modes for L0 = 0.
The limit L0 0 is subtler when other interactions are present. The point is that
interactions between the CFT and the elementary sector that proceed through a relevant
coupling are not expected to die off when the UV cutoff goes to infinity. This is the case,
for example, of the interaction between an elementary chiral fermion and a CFT operator,
Lint = O,
for dim[O] < 5/2. By AdS/CFT this corresponds to a bulk fermion with mass
|M | < k/2. In the 5D picture the non-decoupling is evident from the fact that the brane
to brane fermionic propagator does not go to zero in the limit L0 0 for |M | < k/2.
4.1. The mass of A5 at one loop
(0)
We present here the calculation of the mass of A5 at one-loop level, which confirms the
statements made above. We consider the simple case of Eq. (27) with H0 = H1 = H , and
concentrate on the contribution from a bulk fermion with a 5D mass M and the following
boundary conditions:

i
Aa (+, +), Aa5 (, ),
L (+, +) Ri (, )
(36)
.
=
Aa (, ), Aa (+, +),
(, ) (+, +)
The relevant diagram at one loop is depicted in Fig. 6. The contribution from other particles
can be easily derived from this result.
The Lagrangian of a 5D fermion with a constant bulk mass M is

i
i M A
0 M A

L = g eA DM (DM ) eA DM M ,
(37)
2
2
M = kz M the inverse vielbein and M = { , i 5 } the 5D Dirac matrices. The
with eA
A
covariant derivative is

1
DM = M + M AB A , B iAM ,
8
(38)
AB are
where the only non-vanishing entries in the spin connection M
a5 = a /z. The
mass correction to the zero mode of A5 is written as
m2A5

= g52 k C(r)
d 4p
(2)4
L1
L0
1
du
(ku)4
L1
dv
1
f0 (u)f0 (v)
(kv)4
L0

Tr 5 iS (+,+) (v, u; p) 5 iS (,) (u, v; p) ,
Fig. 6. One-loop correction in AdS to the A5 zero-mode mass term.
(39)
166
where C(r) is the Dynkin index Tr(T T ) = C(r) for a fermion in the representation r.
Here, f0 is the A5 zero-mode wavefunction, Eq. (33), and S (,) (z, z ; p) denotes the
propagator of a 5D fermion, with boundary conditions (, ) and 4D momentum p,
between the two points z and z along the fifth dimension.
Using the fermion propagator given in Appendix A, one can obtain an expression
for m2A5 in terms of integrals of Bessel functions. In the particular case of integer values
of M /k, the Bessel functions reduce to trigonometric functions so that the integrals
greatly simplify. For example, when M = 0 Eq. (39) becomes, after some algebra:
m2A5

C(r)
= 2 g52 k
L1
v
dv f0 (v)
L0

du f0 (u)
L0
dp
0
p3
.
sinh[p(L1 L0 )]
(40)
The integrals are convergent and the result is finite. The momentum integral involves the
brane to brane propagator, as expected from the previous discussion, and it converges
exponentially. Performing the integrals one obtains the result
m2A5 =
C(r) 2
1
g5 k 2
F (L0 /L1 ),
2
L1 L20
(41)
where the function F (x) is given, for example, by

(1 + x)2
3
F (x)|M =0 = (3)
,
8
(1 x)2
x(1 + x)2
F (x)|M =k =
4(1 x)2
[(x

dt
(42)
t5
sinh t
1)2 t cosh t

1.67 x + O x 2 ,
x 3 (1 + x)2
F (x)|M =2k =
4(1 x)2

dt
0
1
+ (1 + x(2 x + t 2 )) sinh t]
(43)
t9
[(x 1)2 t cosh t + (1 + x(2 x + t 2 )) sinh t]

3(1 x)2 t 3 + 6t 3xt 3 t cosh t

1

+ 9(1 x)4 + 3(1 x)2 1 + x 2 3x t 2 + x 2 t 4 sinh t

12.4 x 3 + O x 4 ,
(44)
for the case of M /k = 0, 1, 2, respectively. We find that the A5 mass correction
is O(1/L21 ) for |M | < k/2, while it receives a strong suppression for |M | > k/2
(F (x)|M =ck x 2|c|1 for |c| > 1/2). This is in agreement with the holographic picture
167
where for |M | > k/2 the CFT operator coupled to the elementary fermion becomes
irrelevant and the Yukawa coupling becomes small as shown in Eq. (9).6
It is interesting to notice that m2A5 is even under a change of the sign of M . From a 5D
perspective this is expected, as a change of the sign in the bulk fermion mass is equivalent
to inverting the chirality, L R. Given the assignment for the boundary conditions of
the fermion, Eq. (36), this chirality inversion corresponds to exchanging the i superscript
with , an operation which leaves Eq. (41) invariant. From a 4D holographic perspective,
on the other hand, the fact that the result does not depend on the sign of M arises as a
consequence of the requirement that the two CFT descriptions in terms of the left-handed
and right-handed sources, Eqs. (10) and (11), are equivalent for k/2 M k/2.
4.2. The standard-model Higgs as a hologram of A5
It has already been noticed [6] that the Higgs as a PGB can be realized as the
discretization of the Wilson-loop in a deconstructed fifth dimension. We have shown here
that the connection can be even more strict: the PGB can be the holographic image of
the fifth component of the gauge field that lives in a warped extra dimension. In fact,
such a Higgs happens to be a composite bound state of a strongly interacting (conformal)
sector, so that different ideas, which previously seemed distinct, merge together into a
single scenario. Moreover, if the bulk group G is simple, one could also pursue the idea
of unification of the standard-model electroweak interactions. If boundary gauge kinetic
terms do not play a major role, the renormalization-group flow of the gauge couplings, in
the general scenario of Eq. (27), will follow the pattern:
H1
G
H = H0 H1
at low
exact unification
apparent unification
at E 1/L0
energy
at E 1/L1
Strictly speaking, the gauge couplings of the holographic theory never become
G-symmetric at high energies. Nevertheless, low-energy observers still find unification
predictions from G as if unification occurred at E 1/L0 [15], because the gauge
couplings in the holographic theory become strong at E 1/L0 and thus their low-energy
values are insensitive to the initial values at high energies.
The most economical choice for a simple electroweak group is SU(3)L . In this case
it is known that the hypercharge normalization does not come out correct: embedding
the Higgs in an adjoint of SU(3)L gives a prediction sin2 W = 3/4 [16], which cannot
be accommodated neither with a unification at TeV nor at MPl . This difficulty, however,
is avoided if we introduce brane-localized gauge kinetic terms such that the low-energy
6 Apparently, Eq. (41) does not seem to give a one-loop suppression of m2 compared with 1/L2 for
A5
1
M < k/2, as g52 k is expected to be large ln(L1 /L0 ) in realistic cases. This is because the 4D Yukawa coupling

is large g52 k for M < k/2, canceling the suppression from the loop factor. On the other hand, for M = k/2

g52 k/ ln(L1 /L0 ) = O(1), so that m2A is one-loop suppressed compared
5
2
with 1/L1 . In general, written in terms of the 4D Yukawa coupling ht , Eq. (41) always yields the result Eq. (19)
regardless of the value of M , i.e., the value of ht .
the 4D Yukawa coupling is given by
168
gauge coupling values are correctly reproduced. In this case the above picture is modified
by large threshold corrections arising either at E 1/L0 or 1/L1 , depending on where
we put the brane-localized kinetic terms. The situation is similar in the case where one
tries to embed also the standard-model SU(3)C into the bulk simple group. The simplest
possibility in this case is SU(6). Although a naive prediction from the SU(6) group theory
does not yield the observed values of the gauge couplings, we can always adjust them
by appropriately choosing the coefficients of brane-localized gauge kinetic terms, which
do not necessarily respect SU(6). Therefore, although these theories are not as predictive
as 4D supersymmetric unified theories, they are not in contradiction with the observed
value of the low-energy gauge couplings.
An important issue for any realistic theory with the Higgs as A5 is to have correct
Yukawa couplings between matter and the Higgs. This issue was studied in flat space in
Ref. [17], and the mechanism considered there can be applied in our warped case without
any essential modification. For example, we can adopt the SU(6) model of [17] (either
supersymmetric or non-supersymmetric) with the gauge group reduced to SU(5) U(1)X
on the Planck brane and to SU(4)C SU(2)L U(1) on the TeV brane. Below we explain
things in the supersymmetric case. The matter fields are introduced as hypermultiplets,
transforming as D(15), U(20) and E(15) (and N (6) for right-handed neutrinos) under
SU(6), together with some brane fields. (In the non-supersymmetric case, these fields are
bulk and brane fermions.) Realistic Yukawa matrices are then reproduced by appropriately
choosing the bulk masses for these fields: c M/k 1/2 for the third generation and
c>
(the size of the 4D Yukawa coupling is given by
1/2 for the first two generations
ln(L1 /L0 ) g, g, and |c| ln(L1 /L0 ) g (L0 /L1 )(|c|1/2) for |c| < 1/2, = 1/2, and
> 1/2, respectively, where g is the 4D gauge coupling). An additional ingredient to the
flat space case is that some components of the bulk multiplets become exponentially
light for c > 1/2. These are the fields with the following boundary conditions: (+, )
and c (, +) (in the 4D superfield notation) and give unwanted vector-like matter with
masses much lighter than the TeV scale. We can, however, make these fields heavy by
and
c on the Planck and TeV branes, respectively,
introducing appropriate fields
c
2
and by coupling them to and through the brane mass terms (z L0 )[]
c c ] 2 . (The corresponding terms in a non-supersymmetric theory are
and (z L1 )[
the brane fermion masses.) Proton decay is potentially dangerous in this theory, because
SU(6)/(SU(3)C SU(2)L U(1)Y ) gauge fields have masses of order TeV, which could
mediate rapid proton decay. However, the structure of the theory allows us to impose the
baryon number: D(1), U(1), E(0), N (0) with appropriate charges for the brane fields.
Therefore, we can make proton absolutely stable. An important difference with respect
to the flat space model is that, in the absence of brane-localized gauge kinetic terms, the
gauge couplings in the present model should unify into SU(5) U(1)X at the TeV scale.
As was explained before, this unwanted prediction is avoided if we introduce TeV-brane
localized gauge kinetic terms such that the low-energy gauge coupling values are correctly
reproduced,7 although it implies a loss of any quantitative prediction about gauge coupling
7 Large brane kinetic terms on the TeV brane may also help to reduce constraints from precision electroweak
measurements [18].
169
unification. Small neutrino masses are obtained either by exponential suppressions of the
neutrino Yukawa couplings or by the seesaw mechanism operated on the Planck brane in
the case of large and small bulk right-handed neutrino masses, respectively.
The last issue toward a realistic theory of the PGB Higgs is the quartic coupling. If
the theory is supersymmetric, as the one described above, the O(1) quartic coupling is
generated through the supersymmetric gauge potential. Supersymmetric theories have two
Higgs doublets at low energies, one of which is the PGB of the global symmetry. The treelevel potential takes the form of VH (|H1 |2 |H2 |2 )2 , and the PGB Higgs corresponds
to the direction H1 = H2 . Supersymmetry can be broken either at the Planck brane [10] or
at the TeV brane [19]. In the former case, the holographic theory is essentially a nonsupersymmetric theory. Supersymmetry is a global invariance only of the CFT sector,
and this partial supersymmetry is responsible for the generation of the tree-level quartic
coupling in the Higgs potential without introducing a mass term. Having a non-zero
higgsino mass, however, will require some additional source of supersymmetry breaking.
On the other hand, if supersymmetry is broken on the TeV brane in the 5D theory, the
holographic theory is a locally supersymmetric theory with supersymmetry broken at the
TeV scale by the CFT dynamics. The scale of supersymmetry breaking in this case should
not be very high, as the tree-level Higgs quartic coupling becomes zero if the second Higgs
boson, which is not a PGB, obtains a large supersymmetry breaking mass.
The quartic coupling in non-supersymmetric theories remains as a difficult issue.
However, we can at least adopt the mechanism considered in the previous section with
a little modification. For example, in the case of an SU(3)L theory, we can introduce a bulk
scalar field , transforming as a 6 under SU(3)L . By introducing a tadpole on the TeV
brane for the SU(2)L singlet component of , we can generate the Higgs quartic coupling
in essentially the same way as discussed in Section 3.2. This possibility seems to indicate
that we can have realistic theories in the non-supersymmetric case as well.
5. Phenomenological scales and comparison with pions in large N QCD

To better understand the present theory of PGBs, it is instructive to look at the different
physical scales of the model from the 4D perspective. This will elucidate the PGB nature
of the Higgs and its similarities with pions in QCD.
Let us consider the case in which the PGB is A5 with the symmetry breaking pattern
of Fig. 5. This is equivalent to the model of Section 3 if v IR , since in this limit the
SU(3)L breaking by reproduces the breaking by boundary conditions. The original 5D
1
scales of the model are IR , 1/g52 , k = L1
0 and L1 . They can be related to 4D physical
quantities in the following way:

g2
,
g g52 k,
(45)
m
,
g2 =
L1
ln(L1 /L0 )
where g measures the strength of the KK coupling, m is the mass splitting of the KK
towers (approximately this is the first-KK mass), and g is the 4D gauge coupling for the
gauge bosons of H . Let us see how the different scales are related. Using naive dimensional
analysis we can estimate IR (the scale at which the 5D gauge theory becomes strongly
170
coupled for an observer on the TeV brane):

IR
24 3 L0 24 2 m
.
g2
g52 L1
(46)
We can define a decay constant f for the PGBs in our theory. We follow the usual
definition: m2W = g 2 f2 , where mW is the mass that the gauge bosons Aa obtain from
the strong dynamics. In the 5D picture this is the mass of the zero-mode gauge bosons with
(+, ) boundary conditions. One finds m2W = 2/(L21 ln(L1 /L0 )), and therefore
2 m
f =
(47)
.
g
Using the AdS/CFT relation g 1/ N [5], we obtain f N as expected in strongly

coupled large N theories [20]. Notice that IR can be larger than the naive value of 4f .
This is because the PGBs in our theory arise from higher-dimensional gauge bosons, and
the 5D gauge invariance improves the high energy behavior of the theory. The smallest
scale in the model is the mass of the PGBs. It appears at loop level, as that of charged
pions in the massless quark limit. We obtained in Section 4 that this is of order
g 2 m2
.
(48)
16 2 2
The value of the PGB mass from 5D shows the same dependence on m as that of pions in
QCD [21].
We then find that, in general, these theories have the following pattern of scales
m2 = m2A5
IR > f > m > m .
(49)
In real QCD this pattern is not completely followed, since the pion decay constant is
smaller than the mass. This could be due to the fact that in QCD we have N = 3, which
is not really a large number. In spite of this, other predictions of large N QCD agree
surprisingly well with the experimental data. Similarly, when the above 5D AdS model is
used for the standard model, one realizes that the pattern of scales in Eq. (49) is not really
fulfilled. This is because g 4 in order to reproduce the 4D gauge coupling values from
g 2 = g2 / ln(L1 /L0 ). Therefore, the theory of KK states (resonances) is very close to the
non-perturbative limit (in the large N expansion). The pattern of scales that we obtain is
similar to real QCD:
IR m > f > m .
(50)
=
is subject to large
We must emphasize, however, that the relation
logarithmic corrections and large dependence on brane-kinetic terms, so it is possible that
g takes smaller values than what are naively obtained from the 4D gauge coupling values,
making the KK theory more perturbative. This is in fact also needed if we do not want to
have large corrections to electroweak observables coming from virtual KK states. These
states couple to the Higgs (also a composite object in these models) with a strength g , and
for the value g 4 and L1 1/TeV, they give a too large deviation from the standardmodel predictions.
g2
g2 / ln(L1 /L0 )
171
6. Conclusions
We have presented a class of models where the standard-model Higgs appears as a

composite PGB from a strongly coupled theory, similar to pions in QCD. We have used
the AdS/CFT correspondence to describe the models in terms of the weakly coupled
dual theory. The dual theory corresponds to a gauge theory in a slice of 5D AdS in
which the bulk gauge symmetry is broken to the standard-model gauge group on both
boundaries. This automatically delivers massless scalars (at tree level) on the TeV brane
that we associate to the standard-model Higgs field. A remnant global symmetry, under
which the Higgs transforms non-linearly, protects the Higgs mass from large radiative
corrections. The Higgs mass is generated at one-loop level through the explicit breaking
of the global symmetry due to the standard-model gauge interactions. We have shown
that this one-loop contribution is not quadratically divergent. In the 5D AdS picture, this
is because of the locality. The Higgs lives on the TeV brane away from the other scalar
that breaks the bulk gauge symmetry (which is located on the Planck brane). Therefore,
the Higgs can learn of this breaking only through bulk fields that propagate from one
brane to the other. This is a non-local effect and thus it is finite. In the 4D CFT picture,
the cancellation of quadratic divergences is understood in a different way. The Higgs is a
composite state of CFT which decouples at high energies from the standard-model fields
that are elementary states. We have calculated the effective potential of the Higgs from
gauge loops and have shown that the Higgs squared-mass is finite and a loop factor smaller
than the first resonance mass. This is a very appealing property, since it gives a rationale
for the electroweak scale smaller than the new physics scale, as experiments seem to
indicate.
If the breaking of the bulk gauge symmetry is due to boundary conditions, the massless
scalar corresponds to the fifth component of the bulk gauge boson. Therefore, we find that
Higgs-gauge unification in warped space is equivalent to a Higgs as a composite PGB. We
have also discussed the similarities and differences of our PGBs with pions in QCD.
Models with a PGB Higgs generically suffer from the absence of a tree-level Higgs
quartic coupling, needed to generate a physical Higgs mass larger than the experimental
bound. We have presented a mechanism that can generate a quartic coupling without
inducing a large quadratic term. This requires a specific assumption about the breaking
of the global symmetry at high energies.
The models constructed here have an important phenomenological difference from little
Higgs models. The global symmetry that protects the Higgs mass is a symmetry of the
strong CFT sector of the theory, but not a symmetry of the standard model. Therefore,
there is no partner of the standard-model fields to form a complete multiplet of the
global symmetry. New electroweak-scale states appear as resonances that are in complete
multiplets of the global group, similar to the situation in QCD. Detecting these resonances
in future colliders will allow us to find the symmetries of the strong CFT, and tell us
about the symmetry that protects the electroweak scale from potentially large radiative
corrections.
172
Acknowledgements
We would like to thank R. Barbieri for useful discussions. R.C. thanks P. Creminelli,
A. Donini, B. Gavela, E. Trincherini and especially R. Rattazzi for discussions. Y.N. is
grateful for valuable conversations with G. Burdman and B. Dobrescu, and A.P. thanks
S. Peris, M. Quiros and E. de Rafael for discussions. The work of R.C. is supported by the
EC under RTN contract HPRN-CT-2000-00148.
Appendix A
In this appendix we give the propagators for a bulk scalar, fermion and gauge fields in
a slice of 5D AdS (see also Ref. [22]). We start by giving the scalar field propagator in the
presence of general brane kinetic and mass terms. The free action for a scalar field is
given by
L1

S=
d x
dz
L0
MN

g g M N M 2

+ (z L0 ) gind z0 gind m0

+ (z L1 ) gind z1 gind m1 .
(A.1)
is given as a solution of
The propagator G

2 2

z ; p) = z (z z ).
z z + zz p2 z2 + 2 G(z,
(A.2)
k

represents the propagator for the rescaled field
Here, = 4 + M 2 /k 2 and G
2
(kz) , which is related to the propagator G for the unrescaled field, , as G =

(kz)2 (kz )2 G.
If has boundary conditions (+, +), the scalar propagator (for the rescaled field) is
given by

ZI

L0

K |p|z<
I |p|z<
G(z, z ; p) =
(XI /XK ZI /ZK )
ZK

XI

K |p|z> ,
I |p|z>
(A.3)
XK

where |p| p2 and z< (z> ) is the lesser (greater) of z and z ; I (x) and K (x) are the
modified Bessel functions. The coefficients XI , XK , ZI and ZK are given by

XI = |p|L1 I1 |p|L1 s/2 z1 |p|2 L21 L1
0 m1 L0 I |p|L1 ,

XK = |p|L1 K1 |p|L1 s/2 z1 |p|2 L21 L1
0 m1 L0 K |p|L1 ,
(A.4)

ZI = |p|L0 I1 |p|L0 s/2 + z0 |p|2 L0 + m0 L0 I |p|L0 ,

ZK = |p|L0 K1 |p|L0 s/2 + z0 |p|2 L0 + m0 L0 K |p|L0 ,
(A.5)
173
where s = 4. The propagator for a field having the odd boundary condition at the Planck
brane (TeV brane) is obtained by taking the limit m0 (m1 ).
Restricting the end points to the TeV brane, z = z = L1 , and taking the zero-momentum
limit, p 0, the scalar propagator of Eq. (A.3) becomes
= z = L1 ; p 0)
G(z

L0 + 2 + m1 L0 + 2 m0 L0 L0 2 1
=
2 2 m1 L0 2 + m0 L0 L1

+ 2 + m1 L0
+ 2 m0 L0 L0 2
1+
.
1+
2 + m0 L0 L1
2 m1 L0
(A.6)
For = 0 (M 2 = 4k 2 ), this is further simplified as

= z = L1 ; p 0)
G(z
(1 + (m0 L0 2) ln(L1 /L0 ))
,
=
1
m0 + m1 + L0 (m0 2L1
0 )(m1 + 2L0 ) ln(L1 /L0 )
(A.7)
giving the propagator used in Section 3.2 (Eq. (24)).

The fermion propagators used in Section 4.1 are given by
S (,) (z, z ; p)

5/2

M (,)
1
(,) ,
= k 2 zz
+
PR GR
/
p + 5 z +
+ PL G
L
2z
kz
(A.8)
(+,+) is
where PL,R = (1 5 )/2 and M is the bulk mass of the fermion. The quantity G
R
given by Eq. (A.3) for = |M /k + 1/2|, s = 1, m0 = M , m1 = M and z0 = z1 = 0.
R with the odd boundary condition at the Planck brane (TeV brane) is
The case of G
L is obtained
reproduced by taking the limit m0 (m1 ). The expression for G

from that of GR by simply making the replacement M M .
Finally, the rescaled gauge boson propagator is given by taking = 1 and s = 2 in
Eq. (A.3). The parameters m0 and m1 then represent brane masses for the gauge boson
induced by spontaneous symmetry breaking caused by brane Higgs fields. The case of
boundary condition breaking at the Planck brane (TeV brane) is reproduced by taking the
limit m0 (m1 ).
References
[1] For a relatively recent review, see S.P. Martin, hep-ph/9709356.
[2] S. Weinberg, Phys. Rev. D 13 (1976) 974;
S. Weinberg, Phys. Rev. D 19 (1979) 1277;
L. Susskind, Phys. Rev. D 20 (1979) 2619.
[3] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[4] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
J.M. Maldacena, 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.
174
[5] N. Arkani-Hamed, M. Porrati, L. Randall, JHEP 0108 (2001) 017, hep-th/0012148.

[6] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Lett. B 513 (2001) 232, hep-ph/0105239.
[7] See, for example N. Arkani-Hamed, A.G. Cohen, E. Katz, A.E. Nelson, JHEP 0207 (2002) 034, hepph/0206021;
N. Arkani-Hamed, A.G. Cohen, E. Katz, A.E. Nelson, T. Gregoire, J.G. Wacker, JHEP 0208 (2002) 021,
hep-ph/0206020;
I. Low, W. Skiba, D. Smith, Phys. Rev. D 66 (2002) 072001, hep-ph/0207243;
D.E. Kaplan, M. Schmaltz, hep-ph/0302049;
S. Chang, J.G. Wacker, hep-ph/0303001;
W. Skiba, J. Terning, hep-ph/0305302;
S. Chang, hep-ph/0306034.
[8] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
[9] R. Contino, A. Pomarol, in preparation.
[10] T. Gherghetta, A. Pomarol, Phys. Rev. D 67 (2003) 085018, hep-ph/0302001.
[11] Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 613 (2001) 147, hep-ph/0104041.
[12] L. Randall, M.D. Schwartz, JHEP 0111 (2001) 003, hep-th/0108114.
[13] R. Contino, P. Creminelli, E. Trincherini, JHEP 0210 (2002) 029, hep-th/0208002.
[14] Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Phys. Lett. B 129 (1983) 193.
[15] W.D. Goldberger, Y. Nomura, D.R. Smith, Phys. Rev. D 67 (2003) 075021, hep-ph/0209158.
[16] N.S. Manton, Nucl. Phys. B 158 (1979) 141.
[17] G. Burdman, Y. Nomura, Nucl. Phys. B 656 (2003) 3, hep-ph/0210257.
[18] H. Davoudiasl, J.L. Hewett, T.G. Rizzo, hep-ph/0212279;
M. Carena, E. Ponton, T.M. Tait, C.E. Wagner, Phys. Rev. D 67 (2003) 096006, hep-ph/0212307.
[19] T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141, hep-ph/0003129.
[20] G. t Hooft, Nucl. Phys. B 72 (1974) 461.
[21] T. Das, G.S. Guralnik, V.S. Mathur, F.E. Low, J.E. Young, Phys. Rev. Lett. 18 (1967) 759;
For an approach in large N QCD, see M. Knecht, S. Peris, E. de Rafael, Phys. Lett. B 443 (1998) 255,
hep-ph/9809594.
[22] T. Gherghetta, A. Pomarol, Nucl. Phys. B 602 (2001) 3, hep-ph/0012378.

Gauged locally supersymmetric D = 3

nonlinear sigma models
Bernard de Wit, Ivan Herger, Henning Samtleben
Institute for Theoretical Physics & Spinoza Institute, Utrecht University, Postbus 80.195,
3508 TD Utrecht, The Netherlands
Abstract
We construct supersymmetric deformations of general, locally supersymmetric, nonlinear sigma
models in three spacetime dimensions, by extending the pure supergravity theory with a Chern
Simons term and gauging a subgroup of the sigma model isometries, possibly augmented with
R-symmetry transformations. This class of models is shown to include theories with standard Yang
Mills Lagrangians, with optional moment interactions and topological mass terms. The results
constitute a general classification of three-dimensional gauged supergravities.
PACS: 04.65.+e
1. Introduction
Supergravity theories with vector gauge fields can usually be deformed by introducing
gauge charges for the various fields. These charges generate a corresponding gauge
group. Supersymmetry then necessitates the presence of additional interactions consisting
of masslike terms and a scalar potential, beyond the standard gauge interactions; the
possible gauge groups are often severely restricted. For theories with a high degree of
supersymmetry, gaugings constitute the only known supersymmetric deformations.
In three spacetime dimensions the situation is special in two respects. First of all,
pure extended supergravity is topological. Nontopological theories are constructed by
coupling supergravity to matter. In three dimensions the obvious matter supermultiplets
E-mail addresses: b.dewit@phys.uu.nl (B. de Wit), i.herger@phys.uu.nl (I. Herger),
h.samtleben@phys.uu.nl (H. Samtleben).
doi:10.1016/j.nuclphysb.2003.08.022
176
B. de Wit et al. / Nuclear Physics B 671 (2003) 175216
are scalar multiplets, so that the resulting Lagrangians take the form of a nonlinear
sigma model coupled to supergravity. These theories have been constructed and classified
in [1]. Supersymmetry leads to stringent conditions on the target space, which can only
be satisfied when the number of supersymmetries is restricted to N 16, implying that
there are at most 32 supercharges. Beyond N = 4 the target space has to be homogeneous.
There exists no theory with N = 7 supersymmetry and beyond N = 8 there are only four
possible theories. They are N = 9, 10, 12 and 16 supersymmetric, and their target spaces
are unique and equal to the symmetric spaces F4(20)/SO(9), E6(14)/(SO(10) SO(2)),
E7(5)/(SO(12) SO(3)) and E8(8)/SO(16), respectively.
Secondly, the gauging of these theories seems impossible at first sight, because of the
lack of vector gauge fields. However, one can introduce a ChernSimons term in three
dimensions, which is topological just as pure supergravity itself, and the corresponding
gauge fields can be coupled to the nonlinear sigma model by gauging a subgroup of the
target space isometries. Such gaugings have been constructed in [24] for N = 16, 8, and
in [5,6] for some Abelian gauge groups for the case of N = 2. The gauging is defined
by the gauge group embedding in the isometry group, which in this case is defined in
terms of a symmetric embedding tensor. The latter defines the so-called T -tensors. The
viability of the gauging depends in a subtle manner on the properties of these T -tensors
and the gauged supergravity models have an elegant mathematical structure (see [7] for the
corresponding analysis of the maximal supergravities in higher dimensions). In this paper
we exhibit this structure and derive the precise conditions for having a consistent gauging.
For N > 3 these conditions amount to the absence of T -tensor components transforming
in a particular irreducible SO(N) representation.
The gauged supergravities come with a scalar potential that allows for supersymmetric
anti-de Sitter groundstates. Therefore these theories can be connected to two-dimensional
superconformal theories that live on the boundary of the anti-de Sitter space. The threedimensional setting may offer advantages when studying the AdS/CFT correspondence,
because the supergravity theory is more amenable to nonperturbative studies, while at the
same time the large variety of two-dimensional superconformal theories has been studied
extensively in the literature. The theories constructed in this paper include the effective
theories that arise in the compactification of higher-dimensional supergravities, as we
will show below. These include the compactifications on spheres [8,9], compactifications
with nontrivial fluxes [1013], as well as the theories whose existence has been inferred
from computing the KaluzaKlein spectra in the context of the AdS/CFT correspondence
[1418].
The results of this paper constitute a complete classification of gauged supergravities in
three dimensions which can be regarded as an extension of the classification of ungauged
supergravities presented in [1]. In both cases the matter supermultiplets comprise scalar and
spinor fields. Because vector fields can always be converted to scalar fields by a suitable
duality transformation, the restriction to such scalar multiplets does not seem relevant.
However, the presence of gauge charges often poses an obstacle for performing duality
transformations. As it turns out, no such obstacle arises in the three-dimensional context.
Below we will indicate how, by introducing compensating fields, every YangMills
Lagrangian can be converted into a Lagrangian that belongs to the class of Lagrangians
discussed in this paper. Therefore the classification of the gauged supergravities presented
177
here, is on a par with the classification of the ungauged supergravities given in [1], albeit
that it is not quite possible to present an exhaustive classification of all possible gauge
groups.
We briefly indicate how the conversion of three-dimensional YangMills Lagrangians
can be done. In this conversion every gauge field is replaced by two gauge fields and a new
scalar field, which together describe the same number of dynamic degrees of freedom as
the original gauge field. Our presentation is a further elaboration of the results of [13,19]
and is completely general. Consider the Lagrangian, in three spacetime dimensions, with
YangMills term quadratic in the field strengths, and moment interactions proportional to
A ,
a gauge covariant operator O

1 A
A
g F (A) + O
(A, ) MAB () F B (A) + O B (A, )
4
+ L (A, ).
L=
AA
(1.1)
A (A)
F
and
denote the non-Abelian gauge fields and corresponding field
Here
strengths, labeled by indices A, B, . . . , and generically denotes possible matter fields
transforming according to certain representations of the gauge group GYM . The structure
constants of this group are denoted by fAB C , so that the field strengths read
A
A
A B C
F
(A) = AA
A fBC A A .
The symmetric matrix MAB () may depend on the matter fields and transforms
covariantly under GYM . The last term, L (A, ), in the Lagrangian is separately gauge
invariant and its dependence on the gauge fields is exclusively contained in covariant
derivatives of the matter fields or in topological mass terms (i.e., ChernSimons terms).
The Bianchi identities and vector field equations take the form
A (A) = 0,
D F

B

B
] (A) + O
]
D[ MAB () F
(A, ) JA (A, ) = 0,
(1.2)
where we use the definitions

A (A) = 1 i g F A (A),
F
2
L (A, )
1
JA (A, ) = i g
,
2
AA
A (A, ) = 1 i g O A (A, ).
O
2
Usually the duality is effected by regarding the field strength as an independent field on
which the Bianchi identity is imposed by means of a Lagrange multiplier. Because the
Lagrangian (1.1) depends explicitly on both the field strengths and on the gauge fields, we
proceed differently and write the field strength in terms of new vector fields BA and the
derivative of compensating scalar fields A , all transforming in the adjoint representation
of the gauge group. The explicit expression
A (A) + O
A (A) = M AB (BB D B ),
F
(1.3)
178
B , should be regarded as a field equation that follows from the new

where MAC M CB = A
Lagrangian that we are about to present. The structure of (1.3) implies that we are dealing
with additional gauge transformations as its right-hand side is invariant under the combined
transformations
BA = D A ,
A = A ,
(1.4)
under which all other fields remain invariant. The corresponding Abelian gauge group, T ,
has nilpotent generators transforming in the adjoint representation of GYM . Obviously, the
A act as compensating fields with respect to T . The combined gauge group is now a
semidirect product of GYM and T and its dimension is twice the dimension of the original
gauge group GYM . The covariant field strengths belonging to the new gauge group are
A (A) and F
A
F
A (B, A) = 2D[ BA] , and transform under T according to F = 0 and
B . The fully gauge covariant derivative of equals
FA = C fAB C F
A
A D A BA = A fAB C AB
D
C BA ,
(1.5)
and is invariant under T transformations.

The field equations corresponding to the new Lagrangian
A

1
A
B + 1 i F
g D A M AB ()D
BA O
D A
2
2
+ L (A, ),
L=
(1.6)
lead to (1.3) and (1.2), as well as to the same field equations as before for the
matter fields . Observe that the Lagrangian is fully gauge invariant up to a total
derivative. Hence, the YangMills Lagrangian has now been converted to a ChernSimons
Lagrangian, with a different gauge group and a different scalar field content. To obtain the
original Lagrangian (1.1), one simply imposes the gauge A = 0 and integrates out the
fields BA .
This paper is organized as follows. In Section 2 we briefly summarize the results
of [1] for the ungauged theories. Subsequently we analyze the possible invariances of
the Lagrangian possibly related to target space isometries. Then, in Section 3, we discuss
the gauging of possible subgroups of the isometry group. We determine the potential
and the masslike terms in the general case and derive the extra conditions that must be
satisfied in order to preserve supersymmetry. This is the central result of this paper. In
Section 4 we analyze these restrictions in detail for N 4. For N > 4 the target spaces are
homogeneous, and their consistent gaugings are discussed in Section 5. We present some
concluding remarks in Section 6. Some technical details are relegated to Appendices A
and B.
2. Nonlinear sigma models coupled to supergravity

In this section we summarize and elaborate on the construction of three-dimensional
nonlinear sigma models coupled to supergravity. For the derivation and conventions we
refer to [1]. The fields of the nonlinear sigma model are scalar fields i and spinor
179
fields i , with i = 1, . . . , d; the supergravity fields are the dreibein ea , the spin-connection
field ab and N gravitini fields I with I = 1, . . . , N . The gravitini transform under the
R-symmetry group SO(N), which is not necessarily a symmetry group of the Lagrangian.
The scalar fields parametrize a target space endowed with a Riemannian metric gij ().
2.1. Target-space geometry
Pure supergravity is topological in three dimensions and exists for an arbitrary number
N of supercharges and corresponding gravitini [20]. Its coupling to a nonlinear sigma
model requires the existence of N 1 almost complex structures f P i j (), labeled by
P = 2, . . . , N , which are Hermitean
gij f Pj k + gkj f Pj i = 0,
(2.1)
and generate a Clifford algebra

f P i k f Qk j + f Qi k f P k j = 2 P Q ji .
(2.2)
From the f P one constructs 12 N(N 1) tensors fijI J = fijJ I = fjIiJ that can act as
generators for the group SO(N)
f P Q = f [P f Q] ,
f 1P = f P 1 = f P ,
(2.3)
where, here and henceforth, I, J = 1, . . . , N . The tensors f I J satisfy (in obvious matrix
notation)
f I J f KL = f [I J f KL] 4 [I [K f L]J ] 2 I [K L]J 1,
f I J ij f KL ij = 2d I [K L]J N,4 I J KL Tr(J ).
(2.4)
Only for N = 4, the tensor Jji is relevant; it is defined by J = 16 P QR f P f Q f R , so that

f P f Q = P Q 1 P QR Jf R .
(2.5)
Furthermore, J satisfies
1 I J KL I J KL
(2.6)
f f ,
24
and has eigenvalues equal to 1. It turns out that J is also covariantly constant, which
implies that the target space is locally the product of two separate Riemannian spaces of
dimension d , where d+ + d = d and d are both multiples of 4. These two spaces
correspond to inequivalent N = 4 supermultiplets. Hence the case N = 4 is rather special,
and the last identity (2.4) can be written as

f I J ij f KL ij = 4 d+ PI+J,KL + d PIJ,KL ,
(2.7)
Jf P = f P J,
J 2 = 1,
Jij = Jj i ,
J=
with projectors
1
1
PIJ,KL = I [K L]J I J KL.
2
4
(2.8)
180
For rigidly supersymmetric nonlinear sigma models, the number of supersymmetries

is equal to N = 1, 2 or 4, and the Lagrangians are manifestly invariant under SO(N) Rsymmetry transformations acting exclusively on the fermion fields through multiplication
with the complex structures. The case N = 3 is not distinct from N = 4, because the
existence of two complex structures necessarily implies the existence of a third one. In
case of N = 4, the Lagrangian is a sum of two separate Lagrangians corresponding to the
d -dimensional target spaces. For N = 3, 4 the target spaces are hyper-Khler.
When coupling to supergravity the Lagrangian and supersymmetry transformations
depend on SO(N) target-space connections denoted by QIi J (). These connections are
nontrivial in view of
1
J ]K
RijI J (Q) i QIj J j QIi J + 2QK[I
(2.9)
= fijI J .
i Qj
2
For local supersymmetry N can take the values N = 1, . . . , 6 and 8, 9, 10, 12 or 16. The
situation regarding SO(N) symmetry is more subtle in this case, as we shall discuss in
due course. The N = 3 theory is no longer equivalent to an N = 4 theory, as it has
only three gravitini. In view of the three almost complex structures, the target space is
a quaternionic space. For N = 4 the target space decomposes locally into a product of two
quaternionic spaces of dimension d . The f I J are covariantly constant, both with respect
to the Christoffel and the SO(N) connections, ij k and QIi J , respectively,
J ]K
D i (, Q)fjIkJ i fjIkJ 2i[k l fjI]lJ + 2QK[I
i fj k = 0.
(2.10)
For N > 2 we are thus dealing with almost complex, rather than with complex, structures.
This implies an integrability condition for the target-space Riemann tensor Rij kl ,
Rij mk f I J m l Rij ml f I J m k = fijK[I fklJ ]K ,
(2.11)
where we made use of (2.9). Contracting (2.11) with f MNkl gives, for general N > 2,
1
Rij kl f I J kl = dfijI J ,
4
(2.12)
so that the target space has nontrivial SO(N) holonomy, while contracting (2.11) with g j l ,
using the cyclicity of the Riemann tensor and the above result, yields (for N > 2)
Rij Rikj l g kl = cgij ,
(2.13)
with c = N 2 + 18 d > 0. Hence the target space must be an Einstein space.1

1 For N = 3 this is in accord with the fact that quaternionic spaces of d > 4 are always Einstein [21]. In the
case at hand, the result also holds true for a d = 4 target space. For N = 4 Eqs. (2.12) and (2.13) read

1
Rij kl f I J kl = d+ PI+J,KL + d PIJ,KL fijKL ,
2

1
1
Rij = 2 + d gij + (d+ d )Jij ,
8
8
and we have a product space of two quaternionic manifolds, which are both Einstein. For N = 2 the target space
is Khler and f I J is a complex structure. The SO(2) holonomy is undetermined.
181
Following [1] we introduce a complete set of linearly independent, antisymmetric,

tensors hij , labeled by indices , that commute with the complex structures, i.e.,
hik f I J k j hik f I J k j = 0.
(2.14)
For N = 2, there is only one tensor f I J which commutes with itself, so that this
decomposition is not meaningful. For N > 2 we must have hij f I J ij = 0. The tensors
hi j generate a subgroup H SO(d) that commutes with the group SO(N) generated by
the tensors f I J . They can be normalized according to hij hij and are covariantly
constant with respect to the Christoffel connection and a new connection i ,
Di ( )hjk i hj k = 0.
(2.15)
The Riemann tensor can be written as (N > 2)

1 IJ IJ

fij fkl + C hij hkl ,
(2.16)
8
where C () is a symmetric tensor. This result implies that the holonomy group is
contained in SO(N) H SO(d) which must act irreducibly on the target space. For
N = 4 this result is modified because of the product structure. Observe that the Bianchi
identities on the curvature tensor are not manifest for the expression on the right-hand side
of (2.16), something that plays an important role in the analysis of [1].The H curvatures
can now be shown to take the form

1
2 [i j ] [i j ] = f C hij ,
(2.17)
8
where we have made use of the structure constants of the group H defined by
Rij kl =
h h h h = f h .
The above result shows that the connections i
f Qi .
(2.18)
can be restricted to the form i
2.2. Lagrangian and invariances

Let us now turn to the Lagrangian and supersymmetry transformations. In the following
it is convenient to adopt an SO(N) covariant notation which allows to select the N 1
almost complex structures from the f I J tensors by specifying some arbitrary unit N -vector
I and identifying the complex structures with J f J I . By extending the fermion fields i
to an overcomplete set iI , defined by

iI = i , f P i j j ,
(2.19)
we can write the Lagrangian and transformation rules in a way that does no longer depend
explicitly on the almost-complex structures. The fact that we have only d fermion fields,
rather than dN , can be expressed by the SO(N) covariant constraint
iI = PIJji j J

1 IJ i
j f I J i j j J .
N
(2.20)
182
The trace of this projector equals PII ii = d, which confirms that the total number of fermion
fields is not altered. We should stress here, that the introduction of iI is a purely notational
exercise and we do not aim at implementing the constraint (2.20) at the Lagrangian level.
At every step in the computation one may change back to the original notation by choosing
i = I iI . The covariant notation does not imply that the theory is SO(N) invariant,
but the covariant setting allows us to treat the N supersymmetries and the corresponding
gravitini on equal footing and it facilitates the various derivations in later sections.
The supersymmetry transformations read
1
e a = 5 I a I ,
2
1
I
= D 5 I gij iI j J 5 J i QIi J J ,
8
1 I iI
i
= 5 ,
2
i

1
iI = I J 1 f I J j
(2.21)
/ j 5 J j jik kI + QIj J iJ ,
2
where the supercovariant derivative i and the covariant derivative D (, Q)5 I are
defined by
1
i = i I iI ,
2

1
D 5 I = + a a 5 I + i QIi J 5 J .
2
(2.22)
Observe that the terms proportional to in iI do not satisfy the same constraint (2.20)
as iI itself, because the projection operator PIJji itself transforms under supersymmetry.
As in [1], we use the PauliKlln metric with Hermitean gamma matrices a , satisfying
a b = ab + iabc c .
Let us now turn to the Lagrangian, which reads

1

1
/ jI
L0 = i e a Ra + I D I egij g i j + N 1 iID
2
2

1
1
+ egij iI I j + j eN 2 Rij kl iI a j I kJ a lJ
4
24

2
1
2
iI j I 2
+ eN
2(N 2) gij iI a j J .
(2.23)
3 gij
48
Here we used the covariant derivatives

1
D I = + a a I + i QIi J J ,
2

1
D iI = + a a iI + j jik kI + QIj J iJ .
(2.24)
2
We emphasize that the above results coincide with the results of [1], written in a different
form. The conversion makes use of (2.11). The Lagrangian and transformation rules are
183
consistent with target-space diffeomorphisms and field-dependent SO(N) R-symmetry

rotations, acting on I , iI and QIi J according to
I = I J ()J ,
iI = I J () iJ ,
QIi J = Di I J ().
(2.25)
Combining the last result with (2.9), one concludes that the f I J should also be rotated,
f I J = 2K[I ()f J ]K .
(2.26)
The target-space diffeomorphisms and field-dependent SO(N) R-symmetry rotations

correspond to reparametrizations within certain equivalence classes, but do not, in general,
constitute an invariance.
In the remainder of this section we discuss the invariances of these models, other than
supersymmetry, spacetime diffeomorphisms and local Lorentz transformations. The target
space may have isometries, generated by Killing vector fields Xi (). Some of them can be
extended to invariances of the full Lagrangian, possibly after including a field-dependent
SO(N) transformation according to (2.25) and (2.26). Hence, we combine an isometry
characterized by a Killing vector field Xi with a special SO(N) transformation whose
parameters depend on Xi () and on the scalar fields. Denoting the infinitesimal SO(N)
transformations by S I J (X, ), we require invariance of the target-space metric, the SO(N)
connections and the almost complex structures, up to a uniform SO(N) rotation, i.e.,
LX gij = 0,
LX QIi J + Di S I J (, X) = 0,
LX fijI J 2S K[I (, X)fijJ ]K = 0.
(2.27)
The Lagrangian (2.23) is then invariant under the combined transformations

i = Xi (),
I = S I J (, X)J ,
iI = j I j Xi + S I J (, X) iJ .
(2.28)
The fermion transformations can be rewritten covariantly

I = V I J (, X)J i QIi J J ,

iI = Dj Xi j I + V I J (, X) iJ j jik kI + QIj J iJ ,
(2.29)
where V I J (, X) Xj QIj J () + S I J (, X). The significance of this result will become

apparent in a sequel. Using (2.9) and (2.10), one verifies that the second and third equation
of (2.27) can be written as
1
Di V I J (, X) = fijI J ()Xj (),
2
K[I
IJk
J ]K
f
(, X).
[i ()Dj ] Xk () = fij ()V
(2.30)
The first equation in (2.30) shows that V I J (, X) can be regarded as the moment map
associated with the isometry Xi . The second equation is just the integrability condition of
184
the first equation, so it is not independent. After contraction with f MNij , it leads to
1
dV I J ,
for N = 2, 4,
f I J ij Di Xj = 2
(2.31)
I J,KL
I J,KL KL
+ d P
d + P+
V , for N = 4.
From combining the above equations it follows that 7V I J = 14 dV I J , where 7 denotes
the SO(N) covariant Laplacian. This result applies to N > 2, with obvious modifications
for N = 4. The above analysis shows that there are no restrictions for N > 2 to extend
an isometry to a symmetry of the Lagrangian. For N = 2 this is different, as the isometry
should be holomorphic, i.e., it should leave the complex structure invariant. In this case
V I J is determined by (2.30) up to an integration constant. This constant is related to an
invariance under constant SO(2) transformations of the fermions.
For N = 4 we note that the complex structures PIJ,KL f KL live in the two separate
quaternionic subspaces. The same is true for PIJ,KL V KL , which according to (2.30)
depends only on the corresponding subspace coordinates. Note, however, that when one
of the subspaces is trivial, say when d = 0, then PIJ,KL V KL corresponds to a triplet of
arbitrary constants. This is a consequence of the fact that the model in this case has a rigid
SO(3) invariance which acts exclusively on the fermions.
The supersymmetry transformations do not commute with the isometries, as one can
verify most easily on the fields i , where one derives

Q (5), G (X) = Q (5 ), with 5 I = S I J (, X)5 J .
(2.32)
The isometries that can be extended to an invariance of the Lagrangian, generate an
algebra g. Denoting {XM } as a basis of generators, we have
XMi i XN XN i i XM = f MN K XK ,
(2.33)
with structure constants f MN K . Closure of the algebra implies that the corresponding
induced SO(N) rotations, S MI J S I J (, XM ), satisfy

M N I J
= f MN K S KI J + XMi i S N I J XN i i S MI J .
S ,S
(2.34)
Using (2.33) and the second equation (2.27), this can be rewritten as
M N I J
1
V ,V
= f MN K V KI J + fijI J XMi XN j ,
2
(2.35)
with V MI J V I J (, XM ). In the case of N = 2 and of N = 4 with d+ d = 0,

the R-symmetry, which acts only on the fermions, is realized as a separate invariance.
Obviously, R-symmetry commutes with the isometry group; for N = 2 the R-symmetry
may define a central extension of the isometry group, while for N = 4 with d+ d = 0,
the invariance group takes the form of a direct product of the isometry group with an
SO(3) factor of the R-symmetry group. This implies that there are generators for which
the Killing vector XMi vanishes and S MI J is constant. We return to this in our discussion
of the specific cases in Section 4.
We now note that the second equation of (2.30) implies that Di Xj 14 fijMN V MN
commutes with the almost complex structures, so that it can be decomposed in terms of the
185
antisymmetric tensors hij that were introduced in (2.14)

1
Di XjM fijI J V MI J hij V M .
4
(2.36)
This result and the results given in the remainder of this section apply only to N > 2. Using
the general result for Killing vectors, Di Dj Xk = Rj kil Xl , we can evaluate the derivative
of V M . Introducing furthermore the notation V Mi XMi , we establish the following
system of linear differential equations:
1
Di V MI J = fijI J V Mj ,
2
1
Di V M j = fijI J V MI J + hij V M ,
4
1
Di V M = C hij V Mj ,
8
(2.37)
where the covariant derivative contains the Christoffel connection as well as the SO(N)
H connections. One can prove that C is covariant under the isometry, i.e., V Mi Di C =
2V M f ( C) . Other than that the integrability of the above equations is guaranteed by
previous results. Furthermore, by substituting the second identity of (2.37) into (2.33), and
by taking the derivative of (2.33) and exploiting previous identities, we derive the following
two equations:

1
f MN K V K i = fijI J V MI J V N j V N I J V Mj + hij V M V N j V N V Mj ,
4
1
f MN K V K = f V M V N + C hij V Mi V N j .
(2.38)
8
The quantities V MI J , V Mi and V M transform according to the adjoint representation
of the invariance group, up to field-dependent SO(N) H transformations, as is shown by
(note that for V Mi this already follows from (2.33))

I J
V N i Di V MI J = f MN K V KI J + V N , V M ,
V N i Di V M = f MN K V K + f V N V M .
(2.39)
We close this section with a few observations regarding the structure of the last
equations (2.38) and (2.39). Let us first note that the following operators
1
DM = ji V Mk Dk + f I J i j V MI J + hi j V M ,
4
(2.40)
acting in the space of target-space tensors, provide a realization of the Lie algebra g
associated with the invariance group, according to
M N
D ,D
(2.41)
= f MN K DK .
186
Eqs. (2.38) can also be encoded in the following algebraic structure. Define the algebra
a {t A } {t I J , t , t i }, as an extension of so(N) h with commutation relations

t , t = f t ,
t I J , t KL = 4 I [K t L]J ,
IJ i 1 IJ i j
i
t ,t = f j t ,
t , t = h j i t j ,
2
i j 1 ij I J 1
t , t = fI J t + C hij t .
4
8

(2.42)
Unless C is an H -invariant tensor, this algebra will be nonassociative (or may

alternatively be realized as a soft associative algebra upon imposing [t , C ] =
2f ( C ) ). Eqs. (2.35) and (2.38) then imply that the map
V : g a,

1
V XM := V M A t A = V M I J t I J + V M t + V M i t i ,
2
(2.43)
defines a Lie algebra homomorphism, i.e., V([XM , XN ]) = [V(XM ), V(XN )]. In

particular, the image of g under V is an associative subalgebra of a. Furthermore, (2.37)
takes the simple form

Di V XM = gij t j , V XM .
(2.44)
When the inverse C exists, one can prove that
1 MI J N I J
V
V
+ V M i V N i 8C V M V M ,
2
equals a constant.
(2.45)
3. Gauged isometries
In this section we elevate a subgroup of the isometries to a local symmetry by making
the parameters spacetime dependent. With increasing N , supersymmetry then poses severe
constraints on the possible choices of gauge groups.
3.1. Gauge group and embedding tensor
A subgroup of isometries (possibly extended with R-symmetry transformations for
N = 2, 4) can be encoded in an embedding tensor MN which defines the Killing vectors
that generate the gauge group by
Xi = gMN M (x)XN i ,
(3.1)
with spacetime dependent parameters N (x) and a gauge coupling constant g. Unless the
gauge group coincides with the full group of isometries, the embedding tensor acts as a
projector which reduces the number of independent parameters. In order that this subset of
187
Killing vectors generates a group, MN must satisfy the following condition

MP N Q f PQ R = fMN P PR ,
(3.2)
for certain constants fMN P , which are subsequently identified as the structure constants
of the gauge group. One can verify that the validity of the Jacobi identity for the gauge
group structure constants follows directly from the Jacobi identity associated with the full
group of isometries. For a semisimple gauge group the embedding tensor is simply the
sum of the CartanKilling forms of the different group factors with different and a priori
unrelated coupling constants. The embedding tensors of nonsemisimple gauge groups may
take more complicated forms [22].
The next step is to introduce gauge fields AM
corresponding to the gauge group
M
parameters (x) and include them into the definition of the covariant derivatives. For
example, we have
Ni
D i = i + gMN AM
,
X
(3.3)
which transforms under local isometries according to

D i D i + gMN M j XN i D j ,
provided we assume the gauge field transformations

M
Q
MN AM
.
+ g fPQ M AP
= MN
(3.4)
(3.5)
The covariant field strengths follow from the commutator of two covariant derivatives, e.g.,
M Ni
[D , D ] i = gMN F
X ,
(3.6)
and take the form

M
M
M P Q
= MN AM
MN F
A g fPQ A A .
(3.7)
The gauge transformations on the fermion fields follow from (2.29) upon substitution of
(3.1). From this we derive the covariant derivatives for the spinor fields

1 a
I
N IJ J
D = + a I + i QIi J J + gMN AM
,
V
2

1
D iI = + a a iI + j jik kI + QIj J iJ
2
i N IJ

(3.8)
I J g ik Dk V N j j J .
+ gMN AM
j V
In view of the commutator (2.32), the covariant derivative on the supersymmetry parameter
acquires also an additional covariantization

1
N IJ J
D 5 I = + a a 5 I + i QIi J 5 J + gMN AM
(3.9)
5 .
V
2
In this section we only make use of the previous results (2.30) that apply to arbitrary N > 0.
The extra minimal couplings (3.3) induce modifications of the supersymmetry variations and of the Lagrangian. As long as we are dealing with first derivatives, these new
188
couplings do not lead to complications as they are controlled by gauge covariance. However, commutators of the new covariant derivatives lead to new (covariant) terms proportional to the field strength (3.7). These terms, which cause a violation of supersymmetry,
take the form

1
1 N iI
N
N IJ I J
I
L = igMN F 5
V
5 + V i 5 .
2
2
They are canceled by introducing a ChernSimons term for the vector fields

1
1 N P Q
N
LCS = ig AM
g
f
A
A

MN
PQ
,
4
3
(3.10)
provided the embedding tensor MN is symmetric and provided we assume the following
supersymmetry transformations
MI J I J

5 + V M i iI 5 I .
MN AM
(3.11)
= MN 2V
At this point one can also derive the field equation for the vector fields, which reads

1 iI
M
i M
jJ
MI J
IJ
M

MN F + 2ie D V i + gij V
Di V j = 0,
12
(3.12)
M denotes the supercovariant curvature.

where F
The embedding tensor is a gauge group invariant element of Sym(g g) and therefore
satisfies fMN Q QP + fMP Q QN = 0, which implies

PL f KL M N K + f KL N MK = 0.
(3.13)
We define the so-called T -tensor (originally introduced in higher-dimensional supergravity [23]) as the image of under the map V from (2.43), i.e.,
T I J,KL V MI J MN V N KL ,
T I J i V MI J MN V N i ,
T ij V Mi MN V N j ,
T i V M MN V N i ,
T V M MN V N ,
T I J V MI J MN V N .
(3.14)
The T -tensor components that carry indices , do not play an important role, as they
do not appear directly in the Lagrangian and transformation rules. For the case of N = 2
these components are not defined. From (3.13), (2.39) and (2.33), it readily follows that
the T -tensor transforms covariantly under the gauged isometries. Furthermore, we note
the following identities
D(i Tj k) = 0,
1
D(i T I J j ) = Tk(i f I J k j ) ,
2
1
1
I J,KL
= fijI J T KLj + fijKL T I Jj .
Di T
2
2
(3.15)
189
The covariance under the gauged isometries also allows the derivation of identities
quadratic in the T -tensors. Two examples of such identities are
T MNi T KLj fijI J + T MNi T I Jj fijKL
= 4T MN,P [I T J ]P ,KL + 4T MN,P [K T L]P ,I J ,
T ki T KLj fijI J + T ki T I Jj fijKL = 4T P [I k T J ]P ,KL + 4T P [Kk T L]P ,I J .
(3.16)
3.2. Constraints from supersymmetry

The supersymmetry variations of the vector fields in (3.3), (3.8) cause additional
supersymmetry variations of order g. The variations linear in the spinor fields are

L = egMN 2V MI J I 5 J + V M i iI 5 I V N j D j .
(3.17)
They should be canceled by introducing mass-like terms

1 I J I J
1
IJ I jJ
I J iI j J
,
Lg = eg A1 + A2j + A3ij
2
2
(3.18)
accompanied by additional modifications of the supersymmetry transformation rules

g I = gAI1J 5 J ,
g iI = gNAJ2 iI 5 J .
(3.19)
I J = A J I . Furthermore,
Obviously the tensors A1 and A3 are symmetric, AI1J = AJ1 I , A3ij
3j i
in view of (2.20), A2 and A3 are subject to the constraints
Jj
KI
PI i AKJ
2j = A2i ,
Jj
IK
PI i A3jJkK = A3ik
.
(3.20)
The variations of (3.18) and the additional variations (3.19) of the original Lagrangian
IJ
take the following form
together cancel the terms (3.17), provided that AI2iJ and A3ij

1
Di AI1J + 2T I J i ,
N
1
J ]K
= 2 2D(i Dj ) AI1J + gij AI1J + AK[I
1 fij
N

+ 2Tij I J 4D[i T I J j ] 2Tk[i f I J k j ] .
AI2iJ =
IJ
A3ij
(3.21)
Here A3 has the required symmetry structure. For the convenience of the reader we also
give the (dependent) result
1
1
IJ
Ik
Di AI2jJ = gik AI1K PKk
(3.22)
Jj NA3ij + Tik PJj .
2
2
In addition, we need to ensure that both A2 and A3 as defined in (3.21) satisfy the projection
constraints (3.20). In view of (3.22), it is sufficient to impose this constraint on A2 , which
implies the following two equations
J )K
f K(Ij i Dj A1 + (N 1)Di AI1J + 2Di T I K,J K = 0,

f K[Ij i Dj AJ1 ]K + 2T J ]K j 2(N 1)T I J i = 0.
(3.23)
190
These have several consequences. By iterating the first equation (i.e., by resubstituting the
result for Di A1 ), we derive

f KIj i Dj 4T J L,KL + (N 2)AJ1 K Di 4T I L,J L + (N 2)AI1J
= f I Jj i Dj AKK
+ I J Di AKK
1
1 .
This result can be combined with the second equation (3.23) to eliminate
a linear constraint for the components T I J i of the T -tensor
(N 4)T I J i + 2fijK[I T
J ]Kj
1 I J KL KLj
f f
T
= 0.
ij
N 1
(3.24)
AI1J
and to find
(3.25)
Applying this constraint to the combination f I J T KL + f KL T I J , the resulting equation

may be integrated to

(N 2) T I J,KL T [I J,KL] 4 I [K T L]M,J M
2
I [K L]J T MN,MN = 0.
+
(3.26)
N 1
A priori, this equation holds up to a covariantly constant term. Because of (2.9), covariantly
constant terms cannot exist, unless they are SO(N) invariant and therefore constant.
However, the above equation does not contain a singlet contribution so that it is in fact
exact for any N .
Vice versa, one can show that the covariant derivative of (3.26) implies (3.25), such that
these two equations are in fact equivalent. It is not straightforward to disentangle various
independent equations, due to the nontrivial properties of the 12 N(N 1) almost-complex
structures f I J . For example, the following equation is not independent
1 [I J KL]
f f
D j T MN,MN
(N 8)Di T [I J,KL]
ij
N 1

+ 5 f M[I f J K ij T L]Mj = 0.
(3.27)
One can systematize this analysis by employing a set of SO(N) projection operators, as
we briefly sketch in Appendix A.
Now we use (3.25) to rewrite the f KI DT J L,KL term in (3.24). Combining the result
with (3.23) to remove the f KI DAJ1 K terms, we may integrate the resulting equation to
obtain
2
T MN,MN I J = (N 2) I J ,
(3.28)
N 1
with an as yet undetermined constant . Putting things together, we have shown that
supersymmetry at linear order in g determines the tensors A1 , A2 , A3 according to (3.21),
(3.28) in terms of the T -tensor (3.14) while the latter satisfies the (equivalent) constraints
(3.25), (3.26).
Before proceeding to the remaining terms in the action and transformation rules, we
take a brief look at the supersymmetry algebra. The supersymmetry commutator leads
to a covariantized translation, and a supersymmetry and Lorentz transformation with
4T I L,J L + (N 2)AI1J
191
parameters proportional to 2 . When switching on the gauge coupling, there is an extra

Lorentz transformation, but more importantly, also a local isometry with parameter given
by

(51 ), (52 ) = G 2V MI J 52I 51J + .
(3.29)
The supersymmetry established so far guarantees the closure of the algebra to that order,
except for the gauge fields which appear multiplied by a coupling constant. Their closure
(up to the field equations (3.12)) implies another constraint, namely,

J )K
MN 2V N K(I A1 + V N i Di AI1J = 0.
(3.30)
It implies that the function A1 is gauge covariant; in particular, its trace is invariant, i.e.,
MN V Mi Di AI I = 0. This is in agreement with Eq. (3.28) since we have already proven
that the T -tensors are gauge covariant. Moreover, Eqs. (3.21) show that the tensors A2
and A3 are covariant as well, because they depend on the T -tensors and A1 and covariant
derivatives thereof. Again we can derive certain identities from (3.30) that involve some of
the T -tensors and A1 , such as
I J,M(K L)M
T I J i Di AKL
A1 = 0,
1 + 2T
T ij Dj AKL
1 + 2T
M(Ki
L)M
A1
= 0.
(3.31)
In order to preserve supersymmetry to order g 2 one determines the corresponding

variations linear in I and iI . They reveal the need for a (gauge invariant) scalar potential
in the Lagrangian
Lg 2 = eV

4eg 2
1
AI1J AI1J Ng ij AI2iJ AI2jJ
N
2

2
4eg
1 ij
IJ IJ
IJ
IJ
ij I J I J
= 2 NA1 A1 g Di A1 Dj A1 2g Ti Tj .
N
2
We note that the variation of the scalar potential is given by

I J Ij J
i Lg 2 = eg 2 3AI1J AI2iJ + NA3ij
A2 ,
(3.32)
(3.33)
by virtue of (3.21), (3.22) and (3.30). In order that all supersymmetry variations of the
potential cancel, the following two quadratic equations must be satisfied

1 I J KL KL
KL
2A1 A1 NAKiL
2 A2i ,
N

kl I K
KJ
KL
kl LK
KL
+ Ng A2k A3lj = PIJji 3AKL
1 A2i + Ng A2k A3li .
I iK J K
2AI1K AKJ
1 NA2 A2i =
J
3AI1K AK
2j
(3.34)
It can be shown after some computation that these relations are a direct consequence of
(3.31) upon using (3.21) and (3.28) and require the integration constant in the latter
equation to vanish.
What remains is to analyze the supersymmetry variations cubic in the fermion
fields. Their cancellation depends almost entirely on the results presented so far. E.g.,
192
supersymmetry variations proportional to 3 cancel provided that

= T I J,KL + T [I J,KL] ,
I [K AL]J
1
(3.35)
which is in agreement with Eq. (3.28). The variations that are proportional to 2 cancel
by virtue of (3.21). We have not verified the cancellation of the supersymmetry variations
proportional to 2 and 3 , but we expect that all these terms vanish by means of the
constraints derived so far. In the case of the maximal N = 16 theory this has been verified
explicitly [3]. Note that the gauging does not induce new variations quintic in the fermion
fields.
The central result of this paper, is that a gauge group G0 with a gauge invariant
embedding tensor MN describing the minimal couplings according to (3.3), is consistent
with supersymmetry if and only if the associated T -tensor (3.14) satisfies the constraint
T I J,KL = T [I J,KL]
4
2 I [K L]J
I [K T L]M,MJ
T MN,MN .
N 2
(N 1)(N 2)
(3.36)
From this constraint all further consistency conditions can be derived. The Lagrangian is
modified by a ChernSimons term (3.10), mass-like fermionic terms (3.18) and a scalar
potential (3.32) and the fermions have additional supersymmetry variations (3.19). Note
that the constraint (3.36) is well-defined even for N = 1 and N = 2, but degenerates into
an identity. The consistency constraint (3.36) has a simple group theoretical meaning in
SO(N): denoting the irreducible parts of T I J,KL under SO(N) by
sym
=1+
+ ,
(3.37)
with each box representing a vector representation of SO(N),2 Eq. (3.36) expresses that
P T I J,KL = 0.
(3.38)
4. The theories with N 4

4.1. N = 1
In this case, the target space is a Riemannian manifold of arbitrary dimension d.
The consistency conditions for the gauged theory simplify considerably; in particular the
quadratic constraints (3.34) become identities.
The tensor A1 has just one component, which defines a function F on the target space.
According to (3.30) A1 is gauge invariant, and so is F ,
MN XN i i F = 0.
(4.1)
2 We use the standard Young tableaux for the orthogonal groups; i.e., the four representations in the
1 N (N 3)(N + 1)(N + 2), and N , respectively.
decomposition (3.37) are of dimension 1, 12 N (N + 1) 1, 12
4
For N = 8, the last representation is reducible, but this does not affect the argument here.
193
Reading off the values for A2 and A3 from (3.21), we obtain

A1 = F,
A2i = i F,
A3ij = gij F 2Di j F + 2Tij ,
(4.2)
with Tij = XiM MN XjN .

As a consequence of (4.1), any subgroup of isometries can be gauged (for example, by
choosing a constant function F ). The gravitino is never charged under the gauge group,
as can be seen directly from (3.8), and the gauging is restricted to the matter sector. The
scalar potential V (3.32) is given by

V = g 2 4F 2 2g ij i F j F ,
(4.3)
i.e., the function F serves as the real superpotential. Supersymmetry (in a maximally
symmetric spacetime) requires the vanishing of A2 (cf. (6.5)), so that the stationary points
of F define (anti-de Sitter) supersymmetric ground states.
There exist deformations of the original theory that are not induced by gaugings. They
correspond to MN = 0 and F = 0 and are described by the Lagrangian (2.23), together
with the mass-like terms (3.18), subject to (4.2), and with the scalar potential (4.3).
4.2. N = 2
The target space is now a Khler manifold. Some (partial) results for Abelian gaugings
have already been obtained in [5,6]. As it turns out, any subgroup of the invariance group
can be gauged. These gaugings share some features with the N = 1 gaugings of fourdimensional supergravity [2426].
Many of the quantities introduced above simplify considerably. It is therefore convenient to introduce the notation
f = f 12 ,
Ti = T 12 i ,
Qi = Q12
i ,
V = V 12 ,
T = T I J,I J = 2T 12,12.
To avoid confusion, we keep using the notation

SO(2) transformations. Further, we have
1
i Qj j Qi = fij ,
2
Di ( )fj k = 0,
(4.4)
12
and
S 12
for the parameters of the
(4.5)
where ijk is the Christoffel connection. For a Khler target space it is convenient to
decompose the d real fields into d/2 complex ones and their complex conjugates, i
j
( i , ) in a basis where fi j = ii , f = i . From the fact that f is Hermitean, it

follows that only the components gi = gj are nonzero, and therefore fi = igi = fi .
The fact that f is covariantly constant then leads to
i gj k = j gi k ,
(4.6)
is the
which implies that the metric can locally be written as gi = i K, where K(, )
Khler potential. Furthermore (3.15) implies
1
Ti = ii T ,
2
Tij = Di j T .
(4.7)
194
The projectors defined in (2.20), decompose into a holomorphic and an antiholomorphic

component

1
1
PIJ = I J i5 I J .
PIJji = ji I J + i5 I J ,
2
2
The Khler potential K is defined up to Khler transformations
K(, )
+ () + (
).
K(, )
(4.8)
(4.9)
A solution of (4.5) is provided by

1
1
Qi = ii K,
(4.10)
Q = i K.
4
4
This solution is not unique as it is subject to field-dependent gauge transformations. By
adopting (4.10) we have removed this gauge freedom, but the Khler transformations now
act on Q in the form of a field-dependent SO(2) gauge transformation with parameter

1
)
.
= i () (
12 (, )
(4.11)
4
Consequently, all quantities transforming nontrivially under SO(2) become now subject
to Khler transformations induced by (4.11). Note that transformations where equals
an imaginary constant, correspond to SO(2) transformations acting exclusively on the
fermions and not on the Khler potential. These transformations constitute an invariance
group of the ungauged Lagrangian and they are in the center of the full group of combined
isometries and SO(2) transformations of the fermions.
According to (2.27) only holomorphic isometries of the target space can be extended
to symmetries of the Lagrangian. Such isometries, parameterized by Killing vector fields
(Xi , X ), preserve both the metric and the complex structure, i.e.,
LX g = LX f = 0.
The invariance of the complex structure implies that Xi and X must be holomorphic
and antiholomorphic, respectively. The invariance of the metric gives rise to the Killing
equations
Di X + D Xi = 0,
Di Xj + Dj Xi = 0.
Because of the holomorphicity of Xi , the second condition is automatically satisfied,

whereas the first condition implies that the Khler potential remains invariant under the
isometry up to a Khler transformation. We write this special Khler transformation in
terms of a holomorphic function S(), i.e.,
= Xi i K X K = 4i(S S).
K(, )
According to (2.30), the function V, defined by
1
1
V = Xi Qi + X Q + S 12 = iXi i K + iX K + S 12 ,
4
4
(4.12)
195
must satisfy the equation

1
i V = igi X .
2
As the right-hand side can be written as a derivative, this equation can now be solved and
we obtain
).
= S() + S(
S 12 (, )
(4.13)
Consequently we have

1
1
V = i Xi i K X K + S + S = iXi i K + 2S.
4
2
For every generator XM of the isometry group we may thus identify a holomorphic
function S M , determined by (4.12) up to a real constant. The particular transformation
(which we denote with the extra label M = 0)
1
V 0 = 1,
S0 = ,
(4.14)
2
constitutes a central extension of the isometry group and generates the SO(2) R-symmetry
group that acts exclusively on the fermions. The closure relation (2.34) yields

XMi i S N XN i i S M =
(4.15)
f MN K S K + f MN 0 .
X0i = 0,
K>0
For a semisimple isometry group, the f MN 0 can be absorbed by suitable constant shifts
into the functions S M (or, equivalently, into V M ).
It is always possible to gauge the R-symmetry group. In that case we have a gauge
field A0 associated with SO(2) transformations of the fermions and the T -tensor must be
manifestly invariant under this group. When the structure constants f MN 0 do not vanish
when projected onto the gauged subgroup of the isometries, and cannot be absorbed by
suitable shifts, the R-symmetry group must be gauged. In that case it is more practical to
choose a given set of functions V M whose algebra will exhibit a certain central charge,
without trying to modify the structure constants by constants shifts. Instead one can then
vary the embedding tensor MN . The reason is that the Lagrangian is invariant under a
M
M 0
change of basis in the Lie algebra, according to V M V M + cM , AM
A + c A ,
and
M0 M0 cN N M
00 00 2c
M0 + c
(M = 0),
M
MN cN ,
(4.16)
where the cM are arbitrary constants with c0 = 0. To see this one observes that the
combinations V M MN V N , V Mi MN V N and V Mi MN AN
remain invariant under
the combined substitutions.
Let us further note that, given a gauge group G0 whose embedding tensor satisfies
(3.13), another solution to (3.13) can always be obtained by the deformation
M0 M0 + M ,
(4.17)
196
provided the generator M XM commutes with the gauge group G0 . This is the threedimensional analogue of the local version of the FayetIliopoulos mechanism in fourdimensional N = 1 supergravity [25].
Let us now determine the various quantities involved. It is convenient to decompose the
22
tensor AI1J in terms of a singlet part A11
1 + A1 and a complex quantity

1 22
12
A A11
1 + iA1 .
2 1
Khler transformations are induced by the SO(2) transformations (4.11)

eK/2W = 2i12 eK/2 W .
eK/2 W
This implies that W transforms under Khler transformations according to

W = ()W.
(4.18)
Imposing the Eqs. (3.23) then leads directly to the following result

22
i A11
i W = W = 0.
1 + A1 + 2T = 0,
The function W can be identified as the holomorphic superpotential. Gauge covariance of
A1 imposes the additional relation

MN XN i Di W + 4iV N W = MN XN i i W + 4iS N W = 0,
(4.19)
with the Khler covariant derivative Di W i W + i KW , implied by (4.18). For
nonvanishing W the moment maps associated with the gauge group generators can be
expressed in terms of the superpotential and are proportional to the corresponding Killing
vectors. As a consequence, one may verify from (4.15) that the structure constants f MN 0
vanish when projected onto the gauge group, i.e., MK N L f KL 0 = 0. The gauging of
the SO(2) R-symmetry group requires W to vanish (because W transforms nontrivially
under SO(2)). Therefore nonzero W implies that we have M0 = 0 = 00 .
22
The integration constant in A11
1 + A1 is finally determined by the quadratic constraints
(3.34) provided that W is nonvanishing. The tensor AI1J is then given by
K/2
W,
A11
1 = T e
K/2
A22
W,
1 = T + e
21
K/2
A12
W.
1 = A1 = e
(4.20)
The tensor A2 can be derived from (3.21) and its components are given by

1
i T + eK/2 Di W ,
2

1
22
21
A2i = iA2i = i T eK/2 Di W .
2
Finally, the tensor A3 can be evaluated from (3.21); this leads to, for example,
12
A11
2i = iA2i =
1
11
A3ij
= eK/2 Di Dj W,
4
1
1
1
A3i11 = Ti gi T + i T .
2
4
2
(4.21)
(4.22)
197
One can verify the consistency of these results by inserting A1 , A2 and A3 into the
quadratic constraints (3.34). Indeed, these cancel by virtue of (4.19). When W = 0 the
22
vanishing of the integration constant in A11
1 + A1 cannot be derived from the quadratic
constraints, but follows instead from the identity (3.35). We note in passing that pure
N = 2 supergravity (without gauging) can have a cosmological constant corresponding
to a constant W and vanishing T . This implies that the gravitino mass matrix is traceless.
An alternative way to generate a cosmological term in pure supergravity makes use of
gauging the R-symmetry group. In that case, T equals a nonzero constant (equal to 00 )
and W = 0; the gravitino mass matrix is then proportional to the identity. This has been
considered in [44].
The scalar potential (3.32) of the gauged theory is given by

V = g 2 4T 2 4g i i T T + 4eK |W |2 g i eK Di W D W .
(4.23)
Note that in three dimensions, the scalar potential is quartic in the moment map V, since
the T -tensor is quadratic in V. This is in contrast with, e.g., four dimensions, where the
scalar potential is quadratic in V.
Analogous to the N = 1 case, there are two kinds of supersymmetric deformations
of the original theory. On the one hand, there are the gaugings, which are completely
characterized by an embedding tensor MN . The above analysis shows that there is no
restriction on the T -tensor, and therefore any subgroup of the invariance group of the theory
is an admissible gauge group, as long as its embedding tensor satisfies (3.13). On the other
hand there are the deformations described by the holomorphic superpotential W , which
are not induced by a gauging. In case both deformations are simultaneously present, their
compatibility requires (4.19).
4.3. N = 3
In this case the target space is a quaternionic manifold. The condition (3.36), from
which we have derived all other consistency constraints, is identically satisfied, so that
each subgroup of isometries can be consistently gauged. The gauging follows uniquely
from the embedding tensor and there are no other deformations. In particular, the scalar
tensors A1 , A2 , A3 are defined by (3.21), (3.28), and the scalar potential is given by (3.32).
4.4. N = 4
For N = 4 the target space is locally a product of two quaternionic manifolds of
dimension d , associated with the positive and negative eigenvalues of the tensor J , whose
real coordinates we denote by i , , respectively. Because the almost-complex structures
commute with J , they decompose into two sets of three almost-complex structures f ;

the only nonvanishing components are fij+P , f P , where denotes the split according to
(P , P = 1, 2, 3)
1
1
1
f +P (J + 1)f P = f P 5 P QR f QR ,
2
2
4
1
1
1

f P (J 1)f P = f P 5 P QR f QR .
2
2
4
(4.24)
198
The two sets of almost-complex structures satisfy the multiplication rule

f +P f +Q = 1 P Q P QR f +R ,

f P f Q = 1 P Q P QR f R .
(4.25)
There is a corresponding decomposition of the SO(4) R-symmetry group

SO(4) = SO(3)+ SO(3) .
(4.26)
Obviously the isometry group splits into the product of the isometry groups of the two

subspaces whose generators we label by XMi and XM , respectively. These N = 4
three-dimensional gaugings share some similarities with the N = 2 gaugings of fourdimensional supergravity [2831], although the precise relation remains to be understood.
Upon reduction to three spacetime dimensions the special Khler and the quaternionic
manifolds that describe the interactions of the four-dimensional vector multiplets and
hypermultiplets, respectively, give rise to the two quaternionic spaces that span the target
space manifold [32,33].
Let us first consider the nondegenerate case d+ d = 0. According to the discussion of
Eq. (2.31), the quantities V MI J (, X) decompose into two triplets denoted by V MP and

V MP , which live in each of the corresponding subspaces. The triplets V are the moment
maps associated with the isometries of the quaternionic spaces [27].
A priori, the embedding tensor MN decomposes into diagonal components, MN
and M
. The latter are, however, severely
N
, and off-diagonal components MN
constrained by the invariance condition (3.13)

PM f MN K N L = P N f MN L MK
,

MN
MN
PM
f
= P
N
f
MK
.
K N L
L
Under (4.26), the

representations
T I J,KL :
T I J,KL
(4.27)
component of the T -tensor (3.14) takes values in the
(1, 1) + (1, 1) + (5, 1) + (1, 5) + (3, 3).
(4.28)
The constraint (3.36) which is necessary and sufficient for the existence of a supersymmetric gauging, implies the absence of the (1, 5) + (5, 1) representation in this decomposition;
in the basis (4.24) it implies
1
T P Q = P Q T RR ,
3
where T P Q = V MP MN V N Q ,

(4.29)

and correspondingly for T P Q . The off-diagonal components, T P Q , which are proportional

to MN , remain unconstrained. Unlike the cases N < 4, it is no longer possible to
gauge any subgroup of the isometry group; the consistency of the gauged theory depends

on the condition (4.29) for the moment maps V MP and V MP and the gauge group
invariant diagonal components MN and M
N
of the embedding tensor, together with
the compatibility relation (4.27) for the off-diagonal components MN of the embedding
tensor. For symmetric quaternionic spaces there are convenient techniques for finding
199
admissible gauge groups, as we will discuss in the next section. However, for nonsymmetric [3436] or even nonhomogeneous quaternionic spaces it remains to directly
analyze Eq. (4.29) in order to determine the possible solutions for MN .
Let us finally discuss the degenerate case in which one of the two quaternionic manifolds
vanishes, i.e., let us assume that d = 0. The Lagrangian (2.23) then admits an additional
global symmetry SO(3) acting exclusively on the fermions. Similar to (4.14) above we
can conveniently incorporate these invariances into our framework by defining three extra
= P = 1, 2, 3, satisfying
generators with label M

XP = 0,

S P , S Q = P QR S R ,
(4.30)
so that the four-by-four matrices S P generate the SO(3) group on the fermions. With
these definitions, (4.29) implies
PQ
= PQ

T P Q = P Q .
(4.31)
Assuming that = 0, the SO(3) gauge transformations on the spinors can combine
with possible target space isometries, provided that the isometry group contains an SO(3)
subgroup. The gauge group decomposes into a direct product G0 SO(3), where the gauge
symmetries associated with G0 correspond to the embedding tensor

MN MP P Q Q
N ,
(4.32)

with P Q the inverse of PQ

.
When PQ
= 0, the gauge group can still include the SO(3) R-symmetry group (or an
SO(2) subgroup thereof) through the mixed components of the embedding tensor. Hence
we distinguish three SO(3) generators labeled by P = 1, 2, 3,

TP = P M XM + P PS P ,
(4.33)
where P M XM denote possible corresponding SO(3) Killing vectors. The presence of

the mixed components P P induces another set of gauge group generators
TP = PQ XQ ,
(4.34)
where the PQ XQ denote three more Killing vectors. From (4.27) it then follows that
the generators TP are mutually commuting and transform as a vector under the SO(3)
isometries (4.33). The SO(3) Killing vectors P M XM must be nonvanishing. Hence the
gauge group consists of a semidirect product of SO(3) with the three-dimensional Abelian
group T generated by the TP, possibly multiplied with another subgroup of the isometry
group. These are the type of theories one obtains upon dimensional reduction of fourdimensional N = 2 supergravity without hypermultiplets and a gauged SU(2) subgroup of
the R-symmetry group [37]. It is possible to restrict the embedding tensor, such that we
gauge only an SO(2) subgroup of the SO(3) R-symmetry group. By similar reasoning as
above, it follows that one must at least have an SO(2) SO(2) gauge group.
200
5. Symmetric spaces
For N > 4, it has been shown in [1] that the target spaces are symmetric homogeneous
spaces G/H such that d = dim G dim H. A list of these spaces is given in Table 1. In this
section, we show that the underlying group structure allows to translate the consistency
condition for admissible gauge groups into a projection equation for the embedding
tensor . This provides an efficient way of classifying and constructing solutions to this
equation which has been applied to the gaugings of maximal supergravity in [2,3].
For a homogeneous target space manifold G/H, the scalar fields are described by means
of a G-valued matrix L, on which the rigid action of G is realized by left multiplication,
while H acts as a local symmetry by multiplication from the right. The latter gauge
freedom may be used to eliminate the spurious degrees of freedom in L and obtain a
coset representative so that L = L( i ) is directly parametrized by the d scalar fields i .
In the case at hand, the group H is a maximal compact subgroup of G and given by
SO(N) H . The generators of the group G constitute a Lie algebra g, which decomposes
into {t M } = {XI J , X , Y A }. The XI J generate SO(N) and the X generate the group H ;
together they span the subalgebra h while the remaining (noncompact) generators Y A
transform in a (not necessarily irreducible) spinor representation of SO(N). The relevant
representations are collected in Appendix B. Expanding the dependence of L( i ) on i
defines
1
L1 i L = QIi J XI J + Qi X + ei A Y A .
(5.1)
2
We note the connections Qi (the ones associated with the X were introduced at the end
of Subsection 2.1). The vielbein ei A may be used to convert curved target-space indices
into flat SO(N) spinor indices. The target-space metric gij and the antisymmetric tensors
fijI J are realized by
gij = ei A ej B AB ,
IJ A B
fijI J = AB
ei ej ,
(5.2)
Table 1
Symmetric spaces for N > 4. The representation R0 from (5.10) is underlined in the decomposition (5.8). Dots
. . . represent zero weights
G/H
Radj
Radj sym Radj
Sp(2,k)
Sp(2)Sp(k)
8k
(2, 0, . . .)
(0, . . .) + (0, 1, . . .) + (0, 2, . . .) + (4, 0, . . .)
SU(4,k)
S(U(k)U(4))
8k
(1, . . . , 1)
(0, . . . , 0) + (1, . . . , 1) + (0, 1, . . . , 1, 0) + (2, . . . , 2)
SO(8,k)
SO(8)SO(k)
8k
(0, 1, . . .)
(0, 0, . . .) + (0, 0, 0, 1, . . .) + (2, 0, . . .) + (0, 2, . . .)
F4(20)
SO(9)
16
52
1 + 324 + 1053
E6(2)
SO(10)U(1)
32
78
1 + 650 + 2430
64
133
1 + 1539 + 7371
128
248
1 + 3875 + 27000
10
12
16
E7(5)
SO(12)Sp(1)
E8(8)
SO(16)
201
IJ 3
with SO(N) -matrices AB
. The matter fermion fields are redefined by converting their
target-space indices into indices associated with the conjugate spinor representation of
SO(N),
1 A I iI
e .
(5.3)
N i AA
All the general formulae obtained above may be conveniently translated, noting that the
projector P from (2.20) factorizes according to
1 A I J B
e
(5.4)
B A ej .
N i AA
The isometries are generated by the left action of G on L(), accompanied by a
compensating H-transformation to remain in the coset representative
Ij
gik PJ k =
XMi i L = t M L LS M ( i ),
S M ( i ) h.
(5.5)
Here S M decomposes into S MI J (these quantities were already introduced in a more

general context in Section 2.2) and S M , belonging to SO(N) and H , respectively.
Subsequently one forms the combinations V M S M + XMi Qi for all components
belonging to h. For any coset space one can show [38] that these V M , together with the
V Mi XMi are subject to a system of linear first-order differential equations, which
includes the generators of H and the curvatures of the connections Qi . For the case at
hand the resulting equations coincide precisely with (2.37). The V Mi can also generally
be expressed in terms of the coset representatives L, and the combined expression for all
the V takes the following form
1
L1 t M L V M A t A = V MI J XI J + V M X + V M A Y A ,
2
(5.6)
where V Mi = g ij ej A V M A . Hence the V span an element in the Lie algebra g, which

coincides with the algebra of the generators of the isometries. At this point we can make
direct contact with the map (2.43), which now defines an isomorphism, corresponding to
the field-dependent conjugation (5.6). In particular, the T -tensor (3.14), given by
TAB = V M A MN V N B ,
(5.7)
where A = {I J, , A}, contains the embedding tensor of the gauge group as = T |V =I .

5.1. Lifting the consistency constraints
We recall that the consistency condition for a supersymmetric gauging takes the form of
a single Eq. (3.38) for the T -tensor, and dictates the absence of the SO(N) representation
3 Only for N = 9 and N = 16 the Y A transform in an irreducible spinor representation of SO(N ). Generically,
the Y A comprise a reducible representation of SO(N ) H (cf. Appendix B for a complete list). Correspondingly,
I J are understood unambiguously as acting separately on the different subspaces and
the matrices I AA , AB
as identity on each H -representation factor. Moreover, they define what we will refer to as the conjugate spinor
E.g., for N = 10, we have Y A = 16+ + 16 , and the fermions transform
representation with associated indices A.
in the conjugate representation 16+ + 16 .
202
in T I J,KL . In order to satisfy this equation on the entire scalar manifold G/H, the structure
(5.7) of the T -tensor shows that the entire G-orbit of the SO(N) representation must
vanish. Consider now the decomposition of the T -tensor under G, to

TAB = AB +
Ri
PRi TAB ,
Radj sym Radj = 1
(5.8)
i
where 1 and Radj are the trivial and the adjoint representation of G, respectively, AB
is the CartanKilling form of G, and sym denotes the symmetrized tensor product.
By PRi we denote the G-invariant projector onto the representation Ri . From the SO(N)
composition of the generators, it is clear that there is a unique irreducible representation
R0 of G appearing in the sum in (5.8) that branches under SO(N) such that it contains the
representation . The condition (3.38) is thus equivalent to
PR0 TAB = 0.
(5.9)
The other SO(N) representations contained in this equation can be obtained explicitly by
successively taking derivatives of (3.38). Because (5.9) is a G-covariant condition, it is also
equivalent to
PR0 MN = 0.
(5.10)
The underlying coset structure thus allows to translate the field-dependent form (3.36)
of the consistency condition into a single condition for the constant embedding tensor
of the gauge group G0 . After identifying the representation R0 , the condition for the
embedding tensor corresponding to a consistent gauging can thus be given in explicit form.
In Table 1, we have collected the decompositions (5.8) and the representations R0 for all
theories with N > 4.4 Given a subgroup G0 G with a corresponding embedding tensor
MN , Eq. (5.10) provides a simple and efficient criterion for checking whether G0 can
be consistently gauged while preserving all supersymmetries. The solutions of (5.10) will
be referred to as admissible gauge groups G0 . In the following two sections, we discuss
some of these solutions, case-by-case for the different N . We close this section with some
general remarks on the solutions of (5.10).
A direct consequence of the projection equation (5.10) is that the CartanKilling form
of G is a solution to this equation as it corresponds to the singlet in the decomposition
of (5.8). Therefore the full isometry group G is always an admissible gauge group. The
potential of the corresponding gauged theory reduces to a cosmological constant because
the dependence on the scalars disappears as a result of the G-invariance of the potential. In
fact all scalars fields may simply be gauged away by means of the gauged isometries.
Apart from this trivial solution, one may distinguish different classes of solutions of
(5.10): (i) compact gauge groups, of which in general there are very few, (ii) semisimple
noncompact gauge groups, (iii) non-semisimple gauge groups, and (iv) complex gauge
groups embedded in the real group G [22]. In the following, we restrict the discussion
to some semisimple solutions of (5.10); nonsemisimple gauge groups may generically be
4 We used the LiE package [39] for computing the decompositions of tensor products and the branching of
representations; throughout this paper we use the corresponding conventions for the Dynkin weights.
203
obtained by boosting the embedding tensors of their semisimple cousins (see [22] for a
detailed discussion in the N = 16 theory).
For a compact gauge group, the components I J A , I J , and AB of the embedding
tensor vanish. Its SO(N) part I J,KL must satisfy (3.36), i.e., it must be of the form
I J,KL = IKL
J + I [K L]J +I J KL ,
(5.11)
with a traceless symmetric tensor I J and a completely antisymmetric tensor I J KL .

Explicit inspection of (5.10) shows that for N > 5 the embedding tensor must moreover
I J KL
satisfy AB
I J KL 0, which implies I J KL 0, except for N = 8 where the fourfold
antisymmetric product of vector representations becomes reducible.
Let us therefore consider in some more detail compact gauge groups with embedding
tensor of the form
I J,KL = IKL
J + I [K L]J .
It is straightforward to verify, that the choice

2 1 Np I J , for I p,
I J =
2 Np I J ,
for I > p,
(5.12)
2p N
,
N
(5.13)
describes the embedding of SO(p) SO(N p) SO(N) as = SO(p) SO(Np) ,

i.e., with opposite coupling constant. This ratio is fixed by the requirement that the
embedding tensor takes the form (5.12). Note that = 0 can only be achieved for even
N = 2p. Likewise, one can check that no product SO(p1 ) SO(pn ) with more than
two factors can be embedded into SO(N) with an embedding tensor of the form (5.12).
This severely restricts the possible choices of compact gauge groups.
5.2. The theories with 8 < N 16
For the theories with N = 9, 10, 12, 16, the physical fields form a single supermultiplet,
out of which the scalars parametrize the exceptional coset spaces F4(20) /SO(9),
E6(14)/(SO(10) U(1)), E7(5)/(SO(12) SU(2)), and E8(8)/SO(16), respectively. The
decomposition (5.8) for all these groups contains three irreducible representations only
(cf. Table 1), so that the embedding tensor of an admissible gauge group (5.10) is entirely
contained in the union of a single G-representation R1 and the singlet
MN = MN + PR1 MN .
(5.14)
This enables one to identify solutions of (5.10) by purely group-theoretical reasoning as

we shall summarize in the following observations.
Let G0 G be a semisimple subgroup of G such that the decomposition of R0 from
(5.10) under G0 does not contain a singlet. Then G0 is an admissible gauge group.
Let G0 = G(1) G(2) G be a semisimple subgroup of G such that the decomposition
of R0 from (5.10) under G0 contains precisely one singlet. Then G0 is an admissible
204
gauge group, provided a fixed ratio of coupling constants of G(1) and G(2) [3]. We
denote its embedding tensor by
(1)
(2)
MN = g1 MN + g2 MN ,
(5.15)
(1)
(2)
where MN
and MN
are the restrictions of the CartanKilling form MN of G
(1)
(2)
onto G and G , respectively.
Let G0 = G(1) G(2) G be a semisimple subgroup of G satisfying the above
assumptions with embedding tensor (5.16). Let moreover GN G be a group such
that the decomposition of R1 in (5.14) under GN contains no singlet. Then the ratio of
coupling constants in (5.16) is given by
dim GN dim G(2) dim G dim(G(2) GN )

g1
=
.
g2
dim GN dim G(1) dim G dim(G(1) GN )
This is shown by contracting (5.16) over GN and over the full group G.
(5.16)
Using these facts, we now give a brief case by case discussion of some of the semisimple
admissible gauge groups for the theories with N > 8.
5.2.1. N = 16
The gaugings of the maximal three-dimensional supergravity have been constructed
and discussed in great detail in [2,3]; we include some of the results here for completeness.
The embedding tensor of an admissible compact gauge groups must take the form (5.12).
However, the explicit decomposition of the T -tensor (B.1) shows that the two singlets
in (B.1) are linearly related (specifically 8I J,I J = 15AA ). Since AB = 0 for a
compact gauge group, this requires = 0. From (5.13) it then follows that the only
compact admissible gauge group is the product SO(8) SO(8) with opposite gauge
coupling constants. The noncompact admissible gauge groups include the SO(p, 8 p)
SO(p, 8 p), but also the exceptional groups F4(20) G2(14), E6(14) SU(3),
E7(5) SU(2), and different real forms thereof (see [3] for a detailed list). They all satisfy
the assumptions leading to (5.16). The ratios of coupling constants for these groups are
straightforwardly derived from (5.16) with GN = SO(16).
5.2.2. N = 12
There are three singlets in the decomposition (B.2) related by a single linear condition.
This leads to a larger variety of compact gauge groups. Roughly speaking, a nonvanishing
in (5.12) may be compensated by switching on the extra SU(2) factor from H =
SO(12) SU(2) in the gauge group, such that the noncompact part AB still remains zero.
For example, the decomposition of (5.8) under H shows that this subgroup itself satisfies
the assumptions leading to (5.16); unlike in the maximal case, H is itself an admissible
gauge group with a fixed ratio of coupling constants. This ratio can be derived from (5.16)
(using for example GN = SU(6, 2)) and gives = SO(12) 3SU(2) . In general, the
admissible compact gauge groups are given by the products SO(p) SO(12 p) SU(2)
with embedding tensor
1
= SO(p) SO(12p) + (6 p)SU(2).
2
(5.17)
205
The relative coupling constant between the two first factors is determined by (5.13), while
the relative factor in front of the last term stems from the relation 3I J,I J = 22 for
compact gauge groups (cf. (B.2)) whose relative coefficient may be fixed from the case
p = 12 given above. Note that (only) for p = 6, the gauge group lies entirely in SO(12)
and the SU(2) factor is not gauged. Among the noncompact admissible gauge groups there
are E6(2) U(1), F4(20) SU(2), G2(2) Sp(3), all of which are maximal subgroups of
E7(5) .
5.2.3. N = 10
Similar to the above, the admissible compact gauge groups in this case are the products
SO(p) SO(10 p) U(1) with embedding tensor
1
= SO(p) SO(10p) + (5 p)U(1) .
(5.18)
3
The relative coupling constants are fixed as in (5.17), using for example GN = Sp(2, 2).
For p = 5, the gauge group lies entirely in the SO(10) and the U(1) factor is not
gauged. Among the noncompact admissible gauge groups there are SU(4, 2) SU(2),
G2(14) SU(2, 1), as well as the simple group F4(20) . All these gauge groups are
maximal subgroups in E6(14) .
5.2.4. N = 9
In this case there is no additional factor in H; however, the explicit decomposition (B.4)
shows that the two singlets appearing are independent. Therefore, again a compact gauge
group does not necessarily require vanishing in (5.12). In particular, the group H = SO(9)
itself is an admissible gauge group. More generally, the admissible compact gauge groups
are the products SO(p) SO(9 p) with embedding tensor
= SO(p) SO(9p).
(5.19)
Among the noncompact admissible gauge groups there are G2(14) SL(2) and Sp(2, 1)
SU(2) which are maximal subgroups of F4(20) .
5.3. The theories with 4 < N 8
For N = 5, 6, 8, the field content of the ungauged theories is given by an arbitrary number k of supermultiplets whose scalars fields parametrize the coset spaces
Sp(2, k)/(Sp(2) Sp(k)), SU(4, k)/S(U(2) U(k)), and SO(8, k)/(SO(8) SO(k)), respectively.
5.3.1. N = 8
The N = 8 ungauged theories have been constructed in [40], their gaugings were
discussed in [4].5 Consistency of the gauging is again encoded in (5.10) with R0 =
5 Note, that our conventions here differ from those used in [4,40] by a triality rotation 8 8 of SO(8) in
v
s
order to fit into the general scheme.
206
(0, 2, 0, 0, . . .).6 Just as in the previous examples, one may consistently gauge the entire
compact subgroup H = SO(8) SO(k) with a fixed ratio between the two coupling
constants. More interesting are the admissible gauge groups that lie entirely in the
I J KL
I J,KL = 0, no longer forces
group SO(8). Note that for N = 8, the condition AB
the entire antisymmetric part [I J,KL] of the embedding tensor to vanish, rather the
embedding tensor of a compact gauge group G0 SO(8) takes the general form
I J,KL = I [K L]J +I J KL ,
(5.20)
1
with a traceless symmetric tensor I J and selfdual I J KL = 24
I J KLP QRS P QRS ,
relaxing (5.12). In the standard way (5.13), the group SO(4) SO(4) may be embedded
with relative coupling constant of 1. A nonvanishing I J KL in (5.20) furthermore allows
to introduce an arbitrary relative coupling constant between the two factors inside of
each SO(4) (see [4] for details). This corresponds to the existence of the one-parameter
family D1 (2, 1; ) of N = 4 superextensions of the AdS group SO(2, 2), which appear as
spacetime isometries.
5.3.2. N = 6
Closer inspection of the decomposition (B.6) of shows that the maximal compact
subgroup H = SU(4) SU(k) U(1) is among the admissible gauge groups with an
embedding tensor which forms a linear combination of the corresponding singlets in
(B.6). Since a compact gauge group requires AB = 0, and the four singlets appearing
in (B.6) are linearly related, there are only two independent coupling constants. Other
compact gauge groups are obtained by replacing the SU(4) factor by one of its subgroups
SO(p) SO(6 p), the embedding tensor taking the form (5.13). The total embedding
tensor is given by
4(k 1) + k(p 3)
U(1) ,
(5.21)
4+k
with a free parameter . The relative coefficients are obtained in a similar way as in (5.17),
(5.18), generalizing (5.16) to products of more factors and using GN = SO(4, k). The only
compact admissible gauge group which lies entirely in SO(6) is its subgroup SO(3)
SO(3).
= SO(p) SO(6p) + SU(k)
5.3.3. N = 5
The explicit decomposition of the T -tensor (B.7) shows that the group Sp(2) SO(5)
as well as its product with the entire Sp(k) and independent coupling constants, are
admissible gauge groups. The embedding tensors are given by a linear combination of
the two corresponding singlets in the decomposition (B.7). Instead of Sp(2) one may
also gauge any of its subgroups SO(4), or SO(2) SO(3), the embedding tensor taking
the form (5.13). From (B.7) it follows moreover that for N = 5 even the completely
6 In [4], a slightly stronger consistency condition had been given, namely simultaneous absence of the
(2, 0, 0, 0, . . .). This is in general not necessary. Restricting to compact gauge groups G0 SO(8) however,
the two conditions turn out to be equivalent.
207
Table 2
Symmetric spaces for N = 4. The representation R0 from (5.10) is underlined
G/H
Radj
Radj sym Radj
Sp(m,1)
Sp(m)Sp(1)
4m
(2, 0, . . .)
(0, 0, . . .) + (0, 1, . . .) + (0, 2, . . .) + (4, 0, . . .)
SU(m,2)
S(U(m)U(2))
4m
(1, . . . , 1)
(0, . . . , 0) + (0, 1, . . . , 1, 0) + (1, . . . , 1) + (2, . . . , 2)
SO(m,4)
SO(m)SO(4)
4m
(0, 1, . . .)
(0, 0, . . .) + (0, 0, 0, 1, . . .) + (2, 0, . . .) + (0, 2, . . .)
14
1 + 27 + 77
28
52
1 + 324 + 1053
40
78
1 + 650 + 2430
64
133
1 + 1539 + 7371
112
248
1 + 3875 + 27000
G2(2)
SO(4)
F4(4)
Sp(3)Sp(1)
E6(2)
SU(6)Sp(1)
E7(5)
SO(12)Sp(1)
E8(24)
E7 Sp(1)
antisymmetrized tensor [I J,KL] may be nonvanishing for a compact gauge group, since
the two (5, 0, 0, 0, . . .) representations in (B.7) may be chosen independently.
5.4. N = 4: the symmetric spaces
Recall, that for N = 4 the target space manifold is a product of two quaternionic spaces
and consistency of the gauged theory is encoded in Eq. (4.29) for the T -tensor to be
satisfied on the entire target space manifold. In case, the quaternionic spaces are symmetric,
one can apply the techniques described in this section to lift equation (4.29) to an algebraic
projection equation on the embedding tensor, and exploit the known decomposition of
the isometry group under SO(4) to perform a similar analysis of the admissible gauge
groups. To this end, we list all quaternionic symmetric spaces [41] together with the
decompositions (5.8) and the representations R0 defining (5.10) in Table 2. We leave the
further study of these gaugings and the admissible gauge groups to the reader.
6. Concluding remarks
We have constructed the general N -extended gauged supergravity theories in three
spacetime dimensions. The gaugings constitute supersymmetric deformations of the
ungauged theories of [1], and are entirely characterized by a symmetric embedding tensor
MN that specifies the gauge group as a subgroup of the full invariance group of the
ungauged theories. This invariance group consists of the target-space isometries, and (for
N = 2, 4) possible R-symmetry transformations. The embedding tensor must generate
a proper subgroup and must be invariant under the gauge group. This is expressed by
the conditions (3.13). Supersymmetry imposes additional conditions on the embedding
tensor, which are expressed in terms of constraints on the so-called T -tensor. For N > 2
these constraints are encoded in (3.36). We have analyzed these constraints for the
208
different values of N . Any subgroup can be gauged for N 3. For N = 1, 2 there exist
supersymmetric deformations that are not induced by a gauging; for N = 2 their presence
may form an obstacle to certain gauge groups. For N 4 there are restrictions on the
embedding tensor and thus on the corresponding subgroups that can be gauged. For N > 4
all target spaces are symmetric. For symmetric target spaces with N 4 the restriction
on the gauge group can conveniently be formulated in terms of the algebraic projection
equation (5.10) on the embedding tensor .
The gaugings require the usual masslike terms and a scalar potential in the Lagrangian,
parametrized by three tensors A1 , A2 and A3 ,

1
1
L = L0 + eg AI1J I J + AI2jJ I j J + A3ij I J iI j J
2
2

2
4eg
1
+
(6.1)
AI1J AI1J Ng ij AI2iJ AI2jJ ,
N
2
where L0 denotes the ungauged Lagrangian (2.23) with the spacetime derivatives extended
by extra covariantizations associated with the gauging, as specified in (3.3), (3.8) and (3.9).
The supersymmetry transformations of the fermion fields acquire extra terms proportional
to A1 and A2 . They read
1
I = D 5 I gij iI j J 5 J i QIi J J + gAI1J 5 J ,
8
i

1 IJ
iI

/ j 5 J j jik kI + QIj J iJ gNAJ2 iI 5 J .
= 1 f I J jD
2
(6.2)
The transformation rules for e a and i remain as given in (2.21), for the vector fields AM
they have been given in (3.11).

For N > 2, the tensors A1 , A2 and A3 are uniquely given in terms of the T -tensor
(cf. (3.14)) by means of (3.21) and (3.28),
4
2
T I M,J M +
I J T MN,MN ,
N 2
(N 1)(N 2)
2
4
2 I J
f M(I j m T J )M m +
f KL j m T KL m ,
AI2jJ = T I J j +
N
N(N 2)
N(N 1)(N 2)
1
J ]K
A3 I J ij = 2 2D(i Dj ) AI1J + gij AI1J + AK[I
1 fij
N

(6.3)
+ 2Tij I J 4D[i T I J j ] 2Tk[i f I J k j ] .
AI1J =
The cases N = 1, 2 require a separate analysis, which was presented in Section 4. For
symmetric spaces with N 4, these results simplify considerably.
All gauged theories exhibit a potential (3.32) for the scalar fields i . In certain cases
this potential is constant and simply constitutes a cosmological term. In applications one
is often interested in stationary points of this potential which give rise to anti-de Sitter or
Minkowski solutions with residual supersymmetries, or in de Sitter solutions. Extremal
points in the maximal N = 16 theory have been analyzed in some detail in [3,42,43]. Here
we give a generalization to arbitrary N of some essential formulae that may enable the
reader to carry out a similar analysis for the theories presented in this paper.
209
Stationary points of the scalar potential are characterized by (3.33), which together with
the second equation of (3.34) implies the following relation at the stationary point,7
kl I K
KJ
3AI1K AKJ
2j + Ng A2k A3lj = 0.
(6.4)
The residual supersymmetry of the corresponding solution (assuming maximally symmetric spacetimes) is parametrized by spinors 5 I satisfying the condition,
AJ2iI 5 J = 0,
(6.5)
which ensures that the fermions iI remain invariant in a bosonic background. Full
unbroken supersymmetry implies that A2 vanishes at the stationary point. From the
gravitino variations one derives the condition,

V0 I
1
1 ij I J I J I
I K KJ J
IJ IJ
A1 A1 Ng A2i A2j 5 ,
A1 A1 5 = 2 5 =
(6.6)
N
2
4g
where V0 is the potential taken at the stationary point. Let us emphasize that the two
conditions (6.5) and (6.6) are in fact equivalent by virtue of the first equation of (3.34), so
that the condition (6.6) suffices for establishing the (residual) supersymmetry. Obviously
the potential must be nonpositive at the stationary point, so that the maximally symmetric
spacetime must be a Minkowski or an anti-de Sitter spacetime.
From the above results it follows
that residual supersymmetries are associated with
eigenvalues of AI1J equal to V0 /4g 2 . The massive gravitini that may arise are
associated with different eigenvalues and span an orthogonal subspace. We will distinguish
the indices associated to this orthogonal subspace by I, J, . . . . The massive gravitini can
be identified by the linear combinations,

V0 I J
1
1
J
jJ
I massive AI1K AK
+
J + AI2jJ j J + AI1K AKJ
1
2j , (6.7)
4g 2
2g
2
which are explicitly restricted to the orthogonal subspace. Imposing the unitary gauge by
the condition
AI2jJ j J = 0,
(6.8)
we can simply determine the fermionic mass matrices by projection,

Mgravitini = gAI1J ,
IJ
Mfermions = gA3ij

+ 6gAKI
2i
g 2 A1
4g 2 A21 1V0

K
L
ALJ
2j .
(6.9)
Observe that the restriction to the indices associated with the orthogonal subspace is crucial
as otherwise the second mass matrix would diverge. To show that this matrix is orthogonal
to the condition (6.8), one makes use of the identity (6.4).
7 In the remainder of this section the tensors A , A and A are constant and equal to their values at the
1
2
3
stationary point.
210
It remains to evaluate the mass matrices for the bosons, which can simply be read off
from the Lagrangian and are equal to,
Mvectors = g 2 MK V Ki V L i LN ,
M2scalars = g 2 Di j V .
(6.10)
The derivatives of the potential may be explicitly evaluated using the various identities
derived previously. Moreover, the form of the vector mass matrix together with the
corresponding kinetic ChernSimons term shows that the physical vector masses are
encoded in the matrix gTij = gMN V M i V N j , so that eventually all mass matrices can
be expressed in terms of the T -tensors. This is a common feature of all gauged supergravity
theories. In case of residual supersymmetry the spectrum (6.9), (6.10) decomposes into
representations of the appropriate superextension of the AdS3 isometry group SO(2, 2) =
SL(2, R) SL(2, R).
Acknowledgements
We thank M. Haack, H. Nicolai, M. Trigiante and S. Vandoren for invaluable
discussions. This work is supported in part by the EC contracts HPRN-CT-2000-00122
and HPRN-CT-2000-00131 and by the INTAS contract 99-1-590.
Appendix A. About SO(N ) projectors
For a symmetric target space, the consistency conditions on the T -tensor combine
into the G-invariant form (5.9). In the general case, these consistency conditions take an
SO(N) covariant form, cf. (3.25), (3.26), etc. Whereas the SO(N) representation content
of (3.26) is obvious (cf. (3.38)), it is complicated to disentangle the independent parts of
the consistency conditions (3.25) for the components T I J i , derived from (3.20), (3.21).
In this appendix, we present a systematic approach to handle these equations by means of
SO(N) covariant projectors.
The tensors f I J from (2.9) define a d-dimensional representation of SO(N). Consider
the space of tensors I Jj of dimension dN 2 . Using the f I J , we can build the following
SO(N) covariant maps on this space

1 K L n
K L n
I J j f J L j n ,
IdKLn
PIKLn
I Jj I J j ,
Jj
N
1
L K n
P0I Jj KLn I J KL jn .
IKLn
(A.1)
Jj I J j ,
N
They satisfy the following set of relations
2 = Id,
P2 = P,
P02 = P0 = P0 = P0 ,
2
2
P0 P = P0 P0 P ,
P P0 = P0 P P0 ,
N
N
1
2
P P = P NP P0 P ,
P0 P P0 = P0 ,
N
N
(A.2)
211
which follow from (2.4). An equivalent form of the last relation is

Jj
Jj
Lm
PKm
I i PLm + PI i PKm =
2 KL Jj
PI i ,
N
(A.3)
which proves to be useful in checking the absence of several cubic fermion terms in the
supersymmetry variation of the Lagrangian.
For a tensor I Jj , consider the following inhomogeneous system of linear equations
P = ,
(Id ) = 2Z.
(A.4)
This system admits the unique solution = T Z if and only if Z satisfies the projection
equation
Z = PZ,
(A.5)
where T and the projector P are given by

P
N(P P P + P )
2(N 2)
N(P0 N(P P0 + P0 P ) + N 2 P P0 P )
,
(N 1)(N 2)
N
N2
(P P ) +
(P P0 NP P0 P ),
N 2
(N 1)(N 2)
with PP = P, and tr P = dN . The necessity of the projection condition (A.5) follows

from 2Z = (Id )P and from using the relation
(Id )P = P(Id )P ,
(A.6)
which follows straightforwardly from (A.2). Likewise, one may verify the relations
(Id )T = 2P,
P T = T = T P,
(A.7)
which ensure that = T Z together with (A.5) solves (A.4).

This algebra can be applied to perform a systematic analysis of the constraint equations
in Section 3. As an example, note that the antisymmetric part of the first equation in (3.21)
together with (3.20) constitutes a system of type (A.4) with Z = N2 T . Its solubility thus
implies the consistency relation
T = PT ,
(A.8)
on T I J i which precisely agrees with (3.25). Since P is a projector, this shows that
Eq. (3.25) indeed describes a closed set of consistency relations with nontrivial solution.
The tensor A2 is given as A2 = N2 T T which agrees with (3.21) upon eliminating A1 by
means of (3.28). The proof again makes use of the constraint (A.8) on T I J i .
212
Appendix B. Explicit decompositions of the T -tensor

The representation content of the T -tensor under the group G for the various values
of N > 4 has been given in Table 1. In this appendix, we give the explicit decomposition
of the T -tensor under the compact group H = SO(N) H . Since the embedding tensor
of the gauge group is obtained as = TV =I , it satisfies the same decomposition. This
has been used in the main text to further analyze the admissible compact gauge groups in
Sections 5.2 and 5.3.
B.1. N = 16
Under SO(16), the adjoint representation of E8(8) decomposes into
XI J :
120,
Y A:
128,
implying that the T -tensor of the gauged theory consists of

T I J,KL = 1 + 135 + 1820,
T AB = 1 + 1820,
T I J,A = 1920,
(B.1)
where the two singlets 1 and the two representations 1820 coincide.
B.2. N = 12
Under SO(12) SU(2), the adjoint representation of E7(5) decomposes into
XI J :
(66, 1),
X :
(1, 3),
Y A:
(32, 2),
implying that the T -tensor consists of

T I J,KL = (1, 1) + (77, 1) + (495, 1),
T = (1, 1),
T I J = (66, 3),
T AB = (1, 1) + (495, 1) + (66, 3),

T I J,A = (32, 2) + (352, 2),
T A = (32, 2),
(B.2)
with a linear relation between the three singlets (1, 1), and where the two representations
in the (66, 3), (32, 2), and the (495, 1), respectively, coincide.
B.3. N = 10
Under SO(10) U(1), the adjoint representation of E6(14) decomposes into
XI J :
450 ,
X :
10 ,
Y A:
16+ + 16 ,
213

T I J,KL = 10 + 540 + 2100 ,
T = 10 ,
T I J = 450 ,
T AB = 10 + 10+2 + 102 + 450 + 2100 ,
T I J,A = 16+ + 16 + 144 + 144 ,
T A = 16+ + 16 ,
(B.3)
10 ,
with a linear relation between the three singlets

and where the two representations in
+
0
0
the 16 , 16 , 45 , and 210 , respectively, coincide.
B.4. N = 9
Under SO(9), the adjoint representation of F4(20) decomposes into
XI J :
36,
Y A:
16,

T I J,KL = 1 + 44 + 126,
T AB = 1 + 9 + 126,
T I J,A = 16 + 128,
(B.4)
where the two 126 representations coincide.

B.5. N = 8
Under SO(8) SO(k), the adjoint representation of SO(8, k) decomposes into

XI J :
28, (0, 0, 0, 0, . . .) ,
X :
1, (0, 1, 0, 0, . . .) ,

Y A:
8s , (1, 0, 0, 0, . . .) ,

T I J,KL = 1 + 35v + 35s + 35c , (0, 0, 0, 0, . . .) ,

T = 1, (0, 0, 0, 0, . . .) + (2, 0, 0, 0, . . .) + (0, 0, 0, 1, . . .) ,

T I J = 28, (0, 1, 0, 0, . . .) ,

T AB = 1, (0, 0, 0, 0, . . .) + (2, 0, 0, 0, . . .) + 35s , (0, 0, 0, 0, . . .)

+ 28, (0, 1, 0, 0, . . .) ,

T I J,A = 8s + 56s , (1, 0, 0, 0, . . .) ,

T A = 8s , (1, 0, 0, 0, . . .) + (0, 0, 1, 0, . . .) ,
(B.5)
with a linear relation between the three singlets (1, (0, 0, 0, 0, . . .)), and where the two
representations in the (1, (2, 0, 0, 0, . . .)), (35s , (0, 0, 0, 0, . . .)), (8s , (1, 0, 0, 0, . . .)), and
(28, (0, 1, 0, 0, . . .)), respectively, coincide.
214
B.6. N = 6
Under SU(4) SU(k), the adjoint representation of SU(4, k) decomposes into

XI J :
15, (0, 0, . . . , 0, 0) ,
X :
1, (0, 0, . . . , 0, 0) + (1, 0, . . . , 0, 1) ,

4, (1, 0, . . . , 0, 0) + 4, (0, 0, . . . , 0, 1) ,
Y A:

T I J,KL = 1 + 15 + 20 , (0, 0, . . . , 0, 0) ,

T = 1, 2 (0, 0, . . . , 0, 0) + 2 (1, 0, . . . , 0, 1) + (0, 1, . . . , 1, 0) ,

T I J = 15, (0, 0, . . . , 0, 0) + (1, 0, . . . , 0, 1) ,

T AB = 1 + 15, (0, 0, . . . , 0, 0) + (1, 0, . . . , 0, 1)

+ 6, (0, 1, . . . , 0, 0) + (0, 0, . . . , 1, 0) ,

T I J,A = 4 + 20, (1, 0, . . . , 0, 0) + 4 + 20, (0, 0, . . . , 0, 1) ,

T A = 4, (1, 0, . . . , 1, 0) + 2 (0, 0, . . . , 0, 1)

+ 4, (0, 1, . . . , 0, 1) + 2 (1, 0, . . . , 0, 0) ,
(B.6)
with a linear relation between the four singlets, a linear relation between the three
representations in the (1, (1, 0, . . ., 0, 1)), (4, (0, 0, . . . , 0, 1)), (4, (1, 0, . . . , 0, 0)), and
(15, (0, 0, . . . , 0, 0)), respectively, and where the two representations (15, (1, 0, . . . , 0, 1))
coincide.
B.7. N = 5
Under Sp(2) Sp(k), the adjoint representation of Sp(2, k) decomposes into

XI J :
10, (0, 0, 0, . . .) ,
X :
1, (2, 0, 0, . . .) ,

4, (1, 0, 0, . . .) ,
Y A:

T I J,KL = 1 + 5 + 14, (0, 0, 0, . . .) ,

T = 1, (0, 0, 0, . . .) + (0, 1, 0, . . .) + (0, 2, 0, . . .) ,

T I J = 10, (2, 0, 0, . . .) ,

T AB = 1 + 5, (0, 0, 0, . . .) + (0, 1, 0, . . .) + 10, (2, 0, 0, . . .) ,

T I J,A = 4 + 16, (1, 0, 0, . . .) ,

T A = 4, (1, 0, 0, . . .) + (1, 1, 0, . . .) ,
where the two representations (10, (2, 0, 0, . . .)) in T I J and T AB coincide.
(B.7)
215
References
[1] B. de Wit, A.K. Tollstn, H. Nicolai, Locally supersymmetric D = 3 nonlinear sigma models, Nucl. Phys.
B 392 (1993) 338, hep-th/9208074.
[2] H. Nicolai, H. Samtleben, Maximal gauged supergravity in three dimensions, Phys. Rev. Lett. 86 (2001)
16861689, hep-th/0010076.
[3] H. Nicolai, H. Samtleben, Compact and noncompact gauged maximal supergravities in three dimensions,
JHEP 0104 (2001) 022, hep-th/0103032.
[4] H. Nicolai, H. Samtleben, N = 8 matter coupled AdS3 supergravities, Phys. Lett. B 514 (2001) 165172,
hep-th/0106153.
[5] N.S. Deger, A. Kaya, E. Sezgin, P. Sundell, Matter coupled AdS3 supergravities and their black strings,
Nucl. Phys. B 573 (2000) 275290, hep-th/9908089.
[6] M. Abou-Zeid, H. Samtleben, ChernSimons vortices in supergravity, Phys. Rev. D 65 (2002) 085016,
hep-th/0112035.
[7] B. de Wit, H. Samtleben, M. Trigiante, On Lagrangians and gaugings of maximal supergravities, Nucl. Phys.
B 655 (2003) 93126, hep-th/0212239.
[8] M. Cvetic, H. Lu, C.N. Pope, Consistent KaluzaKlein sphere reductions, Phys. Rev. D 62 (2000) 064028,
hep-th/0003286.
[9] H. Lu, C.N. Pope, E. Sezgin, YangMillsChernSimons supergravity, hep-th/0305242.
[10] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69108, hepth/9906070.
[11] M. Haack, J. Louis, M-theory compactified on CalabiYau fourfolds with background flux, Phys. Lett. B 507
(2001) 296304, hep-th/0103068.
[12] R. Argurio, V.L. Campos, G. Ferretti, R. Heise, Freezing of moduli with fluxes in three dimensions, Nucl.
Phys. B 640 (2002) 351366, hep-th/0205295.
[13] M. Berg, M. Haack, H. Samtleben, CalabiYau fourfolds with flux and supersymmetry breaking, JHEP 04
(2003) 046, hep-th/0212255.
[14] H.J. Boonstra, B. Peeters, K. Skenderis, Brane intersections, anti-de Sitter spacetimes and dual superconformal theories, Nucl. Phys. B 533 (1998) 127162, hep-th/9803231.
[15] S. Deger, A. Kaya, E. Sezgin, P. Sundell, Spectrum of D = 6, N = 4b supergravity on AdS3 S 3 , Nucl.
Phys. B 536 (1998) 110140, hep-th/9804166.
[16] J. de Boer, Six-dimensional supergravity on S 3 AdS3 and 2d conformal field theory, Nucl. Phys. B 548
(1999) 139166, hep-th/9806104.
[17] A. Fujii, R. Kemmoku, S. Mizoguchi, D = 5 simple supergravity on AdS3 S 2 and N = 4 superconformal
field theory, Nucl. Phys. B 574 (2000) 691718, hep-th/9811147.
[18] J. de Boer, A. Pasquinucci, K. Skenderis, AdS/CFT dualities involving large 2d N = 4 superconformal
symmetry, Adv. Theor. Math. Phys. 3 (1999) 577614, hep-th/9904073.
[19] H. Nicolai, H. Samtleben, ChernSimons vs. YangMills gaugings in three dimensions, Nucl. Phys. B 668
(2003) 167178, hep-th/0303213.
[20] A. Achcarro, P.K. Townsend, A ChernSimons action for three-dimensional anti-de Sitter supergravity
theories, Phys. Lett. B 180 (1986) 8992.
[21] D.V. Alekseevski, Riemannian spaces with unusual holonomy groups, Funktsional. Anal. Prilozhen. 2
(1968) 110.
[22] T. Fischbacher, H. Nicolai, H. Samtleben, Non-semisimple and complex gaugings of N = 16 supergravity,
hep-th/0306276.
[23] B. de Wit, H. Nicolai, N = 8 supergravity, Nucl. Phys. B 208 (1982) 323364.
[24] J. Bagger, E. Witten, The gauge invariant supersymmetric nonlinear sigma model, Phys. Lett. B 118 (1982)
103106.
[25] J.A. Bagger, Coupling the gauge invariant supersymmetric nonlinear sigma model to supergravity, Nucl.
Phys. B 211 (1983) 302.
[26] E. Cremmer, S. Ferrara, L. Girardello, A. Van Proeyen, YangMills theories with local supersymmetry:
Lagrangian, transformation laws and superhiggs effect, Nucl. Phys. B 212 (1983) 413.
[27] K. Galicki, A generalization of the momentum mapping construction for quaternionic Khler manifolds,
Commun. Math. Phys. 108 (1987) 117.
216
[28] B. de Wit, P.G. Lauwers, A. Van Proeyen, Lagrangians of N = 2 supergravity-matter systems, Nucl. Phys.
B 255 (1985) 569.
[29] R. DAuria, S. Ferrara, P. Fr, Special and quaternionic isometries: general couplings in N = 2 supergravity
and the scalar potential, Nucl. Phys. B 359 (1991) 705740.
[30] L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fr, General matter coupled N = 2
supergravity, Nucl. Phys. B 476 (1996) 397417, hep-th/9603004.
[31] L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fr, T. Magri, N = 2 supergravity
and N = 2 super-YangMills theory on general scalar manifolds: symplectic covariance, gaugings and the
momentum map, J. Geom. Phys. 23 (1997) 111189, hep-th/9605032.
[32] S. Cecotti, S. Ferrara, L. Girardello, Geometry of type II superstrings and the moduli of superconformal field
theories, Int. J. Mod. Phys. A 4 (1989) 2475.
[33] J. De Jaegher, B. de Wit, B. Kleijn, S. Vandoren, Special geometry in hypermultiplets, Nucl. Phys. B 514
(1998) 553582, hep-th/9707262.
[34] D.V. Alekseevski, Classification of quaternionic spaces with transitive solvable group of motions, Izv. Akad.
Nauk SSSR Ser. Mat. 39 (1975) 315362, 472.
[35] B. de Wit, A. Van Proeyen, Special geometry, cubic polynomials and homogeneous quaternionic spaces,
Commun. Math. Phys. 149 (1992) 307333.
[36] V. Corts, Alekseevskian spaces, Differential Geom. Appl. 6 (1996) 129168.
[37] B. de Wit, A. Van Proeyen, Potentials and symmetries of general gauged N = 2 supergravityYangMills
models, Nucl. Phys. B 245 (1984) 89.
[38] B. de Wit, Supergravity, in: C. Bachas, M. Douglas, A. Bilal, N. Nekrasov, F. David (Eds.), Unity from
Duality: Gravity, Gauge Theory and Strings, Springer-Verlag, 2003, hep-th/0212245.
[39] M. van Leeuwen, A. Cohen, B. Lisser, LiE, a Computer Algebra Package for Lie Group Computations,
Computer Algebra Nederland, Amsterdam, 1992.
[40] N. Marcus, J.H. Schwarz, Three-dimensional supergravity theories, Nucl. Phys. B 228 (1983) 145162.
[41] J.A. Wolf, Complex homogeneous contact manifolds and quaternionic symmetric spaces, J. Math. Mech. 14
(1965) 10331047.
[42] T. Fischbacher, H. Nicolai, H. Samtleben, Vacua of maximal gauged D = 3 supergravities, Class. Quantum
Grav. 19 (2002) 52975334, hep-th/0207206.
[43] T. Fischbacher, Mapping the vacuum structure of gauged maximal supergravities: an application of highperformance symbolic algebra, PhD thesis, 2003, hep-th/0305176.
[44] J.M. Izquierdo, P.K. Townsend, Supersymmetric spacetimes in (2 + 1) AdS supergravity models, Class.
Quantum Grav. 12 (1995) 895924, gr-qc/9501018.

Noncommutative superspace, supermatrix and

lowest Landau level
Machiko Hatsuda a,b , Satoshi Iso a , Hiroshi Umetsu a
a Theory Division, High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki 305-0801, Japan
b Urawa University, Saitama 336-0974, Japan
Abstract
By using graded (super-)Lie algebras, we can construct noncommutative superspace on curved
homogeneous manifolds. In this paper, we take a flat limit to obtain flat noncommutative superspace.
We particularly consider d = 2 and d = 4 superspaces based on the graded Lie algebras osp(1|2),
su(2|1) and psu(2|2). Jacobi identities of supersymmetry algebras and associativities of star products
are automatically satisfied. Covariant derivatives which commute with supersymmetry generators
are obtained and chiral constraints can be imposed. We also discuss that these noncommutative
superspaces can be understood as constrained systems analogous to the lowest Landau level system.
PACS: 11.30.Pb; 11.10.Nx; 11.25.-w
Keywords: Superalgebra; Supermatrix; Noncommutative space
1. Introduction
Deformation of superspace by introducing noncommutativity has attracted interests
recently. In papers [13], it is discussed that the background of the RR field strengths (in the
first two papers, graviphotons) in string theory gives rise to non anticommutative fermionic
coordinates. (In this paper we use the word noncommutative for the non anticommutativity
of the fermionic coordinates unless there is any confusion.) This phenomenon is similar to
the well-known case of the string theory in the NSNS two form B background, where the
bosonic spacetime coordinates become noncommutative [4,5].
E-mail addresses: mhatsuda@post.kek.jp (M. Hatsuda), satoshi.iso@kek.jp (S. Iso), umetsu@post.kek.jp
(H. Umetsu).
doi:10.1016/j.nuclphysb.2003.08.013
218
M. Hatsuda et al. / Nuclear Physics B 671 (2003) 217242
Noncommutative superspace has been already studied by several papers from the field
theoretic approach. In paper [6], they discussed a possibility that the anticommutators
of fermionic variables are written in terms of the spacetime coordinates so that the
spacetime can be generated as a composite of fermions. Quantum deformations of the
Poincar supergroup were also considered in [7] and then it was shown that the chiral
operators are not closed under star products with fermionic noncommutativity [8]. In the
paper [9], general forms of the deformed superspace in d = 4 are discussed by imposing
the covariance under supertranslations, Jacobi identity and closure of chiral superfields
under the star product. N = 1/2 supersymmetry in d = 4 was proposed very recently [1,3]
and its radiative corrections are also studied [33].
There is another approach to study the noncommutative superspace from supermatrix
models based on graded Lie algebras. This approach is a natural extension of constructing bosonic noncommutative space based on matrix models. In addition to an advantage
that the system can be realized in terms of finite matrices, this matrix model approach is
especially useful to study gauge theories on noncommutative space because noncommutative spacetime and gauge fields on it are unified by single matrices. In this approach,
constructions of the open Wilson lines or background independence of the noncommutative gauge theories become manifest [1012]. Generalizations to supermatrices were first
investigated in the paper by Grosse et al. [13]. These authors studies supermatrix models based on the osp(1|2) graded Lie algebra and constructed supersymmetric actions for
scalar multiplets on two-dimensional fuzzy supersphere. They further studied noncommutative de Rham complex and forms in another paper [14]. Gauge theories on noncommutative superspace can be also constructed [15]. Recently the concept of noncommutative
superspace based on a supermatrix was also introduced in proving the DijkgraafVafa conjecture as the large N reduction by Kawai et al. [16], which motivated us to start the present
work.
In this paper we first construct noncommutative superspace based on graded (super-)Lie
algebras and then take their flat limits to obtain flat noncommutative superspace. Though
this approach is restricted to construct noncommutative superspace with Lie algebraic
structures, there is an advantage that the symmetry is manifest and that Jacobi identities of
the algebras and associativity of star products are automatically satisfied. We also construct
covariant derivatives and impose chiral constraints.
Noncommutative superspace can be also understood as a constrained system analogous
to the lowest Landau level system of particles moving in a constant magnetic field. For
superparticles moving in a magnetic field on superspace, we can introduce two mutually
commutative sets of operators, covariant derivatives and guiding center coordinates. By
imposing the lowest Landau level constraint, superspace coordinates become noncommutative. Such noncommutativity on superspace was first discussed in [6]. Our approach is
similar to theirs in two points; one is to consider de Sitter algebra instead of the Poincar
algebra and the other is to derive noncommutativity through Dirac brackets.
We will here explain a basic idea to induce fermionic noncommutativity by considering
a particle in a generalized magnetic field on superspace. We first introduce supercovariant
derivatives:
Di = D0i eAi (x, ),
(1.1)
D = D0 eA (x, ).
219
(1.2)
Di are bosonic covariant derivatives in spacetime directions and D are fermionic ones in
Grassmannian directions. In the following we write them together as DI = (Di , D ). These
derivatives are familiar in supersymmetric gauge theories [17] or in supergravities [18].
They transform covariantly under local gauge transformations on superspace. If the
derivatives without gauge fields DI0 satisfy
0 0
0
,
DI , DJ = iTI J K DK
(1.3)
we can define field strengths as
iFI J = [DI , DJ } iTI J K DK .
(1.4)
Now we consider a superspace with a constant magnetic field, namely, with a constant
FI J background.1 This is analogous to the two-dimensional system in a constant magnetic
field. A particle moving in a strong magnetic field is restricted in the lowest Landau level
and the guiding center coordinates become noncommutative. In other words, the lowest
Landau level condition can be made second class by imposing Dx = Dy = 0 and the Dirac
brackets between coordinates become noncommutative. In a magnetic field on superspace,
by generalizing the bosonic second class constraints to include fermionic parts of the
covariant derivatives, we can evaluate Dirac brackets for the superspace coordinates. In
general this leads to the noncommutative algebras for the superspace coordinates. In this
paper we actually show it explicitly for some examples.
This way to generate fermionic noncommutativity becomes more important if we
consider gauge theories on superspace based on supermatrix models. In ordinary matrix
models of the IKKT type [19], we expand the basic bosonic degrees of freedom Ai around
cl
cl
some classical solution Acl
i ; Ai = Ai + ai . The classical background Ai defines the
background spacetime and ai are interpreted as gauge fields on this spacetime. If the
classical solutions are noncommutative, we can obtain noncommutative gauge theories
[10,11]. This unification of spacetime and the gauge field into a single matrix Ai can be
generalized to a unification of noncommutative superspace coordinate and fermionic
gauge field a into a single fermionic matrix A = + a . In order to generate such a
fermionic background, we need to consider supermatrix models [15,20] whose components
are again supermatrices. The fermionic noncommutativity can be generated dynamically
as the background of the unified fermionic matrix A . We want to discuss this issue in a
forthcoming paper.
In Section 2, as a warming up exercise, we briefly review the noncommutative (fuzzy)
sphere based on su(2) algebra and its flat limit by the InnWigner contractions. The
noncommutativity on the fuzzy sphere can be understood as the noncommutativity of the
coordinates of a particle restricted in the lowest Landau level on the commutative sphere in
a monopole background. After taking a flat limit, following a general formalism of left and
right SU(2) multiplications on group manifold, we can construct mutually commutative
1 The field strengths must satisfy the (modified) Bianchi identity and the background F
I J are generally
dependent on superspace coordinates because of the nonvanishing TI J K .
220
set of covariant derivatives and guiding center coordinates. Though it is a well-known fact,
we explain it in details to clarify the origin of noncommutativity as a constrained system.
In Section 3, we first construct a noncommutative curved superspace based on a graded
Lie algebra osp(1|2) and take its flat limit. To make a flat limit together with a fermionic
noncommutativity, we need to take an asymmetric scaling limit of two Grassmannian
coordinates. In this limit, only half of the supersymmetry generators can generate space
time supersymmetry. The other half generates non-dynamical supersymmetry. We can
impose a chiral constraint to remove the latter half. After imposing the chiral constraint, the
system becomes essentially a one-dimensional system. In the latter half of the section, we
consider the noncommutative superspace as a constrained system. We first obtain mutually
commutative sets of operators for superparticle in a constant magnetic field on superspace;
supercovariant derivatives and superguiding center coordinates. Then by imposing the
lowest Landau level conditions we obtain the noncommutative relations for the superspace
coordinates.
In Section 4, we consider a noncommutative superspace based on su(2|1) algebra. This
gives again two-dimensional supersphere but with twice as many as supercharges, i.e., four.
We again need to take an asymmetric scaling for Grassmannian coordinates when we take
a flat limit. All procedures can be performed in parallel to the osp(1|2) case.
In Section 5, we consider four-dimensional superspace based on psu(2|2) superalgebra.
This gives a coset supermanifold PSU(2|2)/U (1)2 whose bosonic part is Euclidean
AdS2 S 2 . In this case there are varieties to scale Grassmannian coordinates. We first give
an example of a similar scaling to the two-dimensional case. In this case, we can easily
perform the same calculation as the supersphere cases and obtain covariant derivatives
which commute with supercharges. Among eight supersymmetries, four of them are
dynamical and generate spacetime supersymmetries. Spacetime translations in only two
directions out of four appear in the anticommutators of supersymmetries and after imposing
chiral constraints this system becomes essentially two-dimensional. We also give another
example where all four spacetime translation generators appear in the anticommutators.
In the latter half of this section, we explain briefly that N = 12 , d = 4 noncommutative
superspace [1,3] can be understood as a constrained system.
In Appendix A we review a general method using the Cartan one-forms to obtain
mutually commutative sets of generators, left and right multiplications on the group
manifold. They become covariant derivatives and guiding center coordinates in our cases.
We then explain a method for the generalized InnWigner contractions. In Appendices B
and C, we give some detailed calculations for supercovariant derivatives and global charges
(guiding center coordinates).
2. Fuzzy sphere and flat limit

In this section, we review the bosonic fuzzy sphere as a warming up. A fuzzy sphere
and field theories on it can be constructed by matrix models based on the su(2) Lie algebra.
(For more details, see [2224], for example.) Take representation matrices li of su(2) with
an angular momentum L. The size of the matrices is N = 2L + 1. The noncommutative
coordinates are defined as xi = li . Then the radius of the sphere r is r 2 = 2 L(L + 1).
221
Any N N matrix can be expanded in terms of noncommutative spherical harmonics Ylm

where l runs from 0 to 2L. The coordinates satisfy the noncommutative algebra;
[xi , xj ] = iij k xk .
(2.1)
We now take the noncommutative flat limit of the fuzzy sphere algebra (2.1). This
corresponds to considering the vicinity of the north pole and scaling (x1 , x2 ) coordinates
so that the noncommutativity is fixed. This can be achieved by redefining coordinates as

y =
x =
(2.2)
x1 ,
x2 .
r
r
They satisfy the commutation relation [x,
y]
= i. is a noncommutative parameter.
2.1. Fuzzy sphere as the lowest Landau level
The noncommutative coordinates on the fuzzy sphere can be understood as the guiding
center coordinates on an ordinary sphere in a magnetic monopole at the origin. Let us see
this explicitly. We consider a particle moving on a sphere with radius r in the field of a
monopole put at the origin. The Hamiltonian is given by
1 2
D
(2.3)
2m i
where Di is the covariant derivative Di = i(i ieAi ). In a monopole magnetic field,
the commutator becomes
xk
[Di , Dj ] = iegij k 3 .
(2.4)
r
The Hamiltonian is written in terms of the following deformed su(2) generators [25]
xi
Ki = ij k xj Dk + eg
(2.5)
r
as
H=
Ki2 e2 g 2
.
(2.6)
2mr 2
A consistency condition requires that the allowed values of the total angular momentum
generated by Ki are
H=
L = |eg|, |eg| + 1, . . . .
(2.7)
Each state with a fixed L corresponds to each Landau level state. The lowest Landau level
states have degeneracy 2|eg| + 1. Such degeneracies are described by the eigenvalue of K3
and they are generated by acting the K operator on the highest weight state. To describe
the degeneracy in each Landau level, we can define the guiding center coordinates Xi as
X i = Ki
(2.8)
where is defined so that the radius of X i becomes r. Then

=
+ 1). Since
they commute with the Hamiltonian, they are constants of motion and can be interpreted
r2
2 L(L
222
as the guiding center coordinates of the cyclotron motion on the sphere. In the large g
limit, = r/eg and the first term in Ki can be neglected compared to the second term.
Then X i is identified with the commutative coordinate xi . This means that the cyclotron
radius becomes small in the large magnetic field limit. These guiding center coordinates
are noncommutative;

X i , X j = iij k X k .
(2.9)
This is nothing but the commutation relation (2.1) on the fuzzy sphere.
2.2. Flat limit of fuzzy sphere
In this subsection, we apply the general method (see Appendix A) to obtain covariant
derivatives and guiding center coordinates to the simplest case. We first parametrize the
group manifold of SU(2) by g = exp(iL x). Following the method of the (generalized)
InnWigner contractions, we can obtain a flat limit by taking the following scaling of
the parameter space. We first take the scaling
x sx,
y sy,
z z.
(2.10)
Then if we take up to the second order of s 2 , we can obtain the algebra

Lx[1] , Lz[2] = Ly[1] , Lz[2] = 0,
Lx[1] , Ly[1] = iLz[2] ,

Lz[0] , Lx[1] = iLy[1] ,
Ly[1] , Lz[0] = iLx[1] .
(2.11)
The generator Lz[2] is a center and can be considered as a constant. The generator Lz[0] is
a rotation generator on the two-dimensional plane. If we instead take the scaling of z as
z s 2 z,
(2.12)
the generator Lz[0] disappears and the algebra of the noncommutative plane which is given
by the first line of (2.11) can be obtained. In the following we consider this case.
The covariant derivatives (right multiplication generators) of the algebra generated
by Lx[1] , Ly[1] and the center Lz[2] can be constructed by the Cartan 1-form g 1 dg =
dx m em a Ta as Da = i(e1)a m m . We take the parametrization (x, y, ) for Ta =
(Lx[1] , Ly[1] , Lz[2] ). Then the covariant derivatives are given as
y
+
,
ix 2 i
Dz =
.
i
Dx =
Dy =
,
iy 2 i
(2.13)
(2.14)
They satisfy the algebra

[Dx , Dy ] = iDz .
(2.15)
The global charges (left multiplication generators) are constructed similarly and given
by
x = y ,
K
ix 2i
y= + x ,
K
iy 2i
z= .
K
i
(2.16)
223
These global charges commute with the covariant derivatives.

Since is a coordinate conjugate to the center Lz[2] and / is a generator to multiply
a constant (Lz[2] ) on g, we can fix it as a constant: Dz = i

= 1 = 0. Then the
commutator becomes
[Dx , Dy ] = i 1 .
(2.17)
We define the noncommutative coordinates X, Y (the guiding center coordinates) as

1
y = x Dy ,
X=
x 2
= K
2
iy

1
x = y + Dx .
Y=
(2.18)
y + 2
= K
2
ix
They satisfy the algebra
[X, Y ] = i.
(2.19)
The transformation from (x, y, px , py ) to (X, Y, Dx , Dy ) is familiar in the two-dimensional

system in a constant magnetic field. If we consider a particle constrained in the lowest
Landau level, we impose the lowest Landau level condition (for > 0)
(Dx iDy )|LLL = 0.
(2.20)
This constraint can be made as the second class by imposing Dx = 0, Dy = 0. The Dirac
bracket for these second class constraints is given by
[O1 , O2 ]D = [O1 , O2 ] + i[O1 , Dx ][Dy , O2 ] i[O1 , Dy ][Dx , O2 ].
(2.21)
The Dirac bracket of the original coordinates becomes

[x, y]D = i.
(2.22)
On the other hand using facts that (X, Y ) are equal to (x, y) up to the second class
constraints and they commute with (Dx , Dy ), we have
[x, y]D = [X, Y ]D = [X, Y ] = i.
(2.23)
Therefore the Dirac bracket of the original coordinates gives the noncommutative
coordinate algebra (2.19). In this picture, the noncommutative space is considered as a
space whose phase space degrees of freedom is reduced by the constraint.
3. Noncommutative flat superspace from osp(1|2)

In the following sections we generalize the method in the previous section to
superspaces. A construction of noncommutative space is performed by generalization of
ordinary Lie algebras to super-Lie algebras. In this way, we can systematically construct
supermatrix models and field theories on noncommutative homogeneous superspaces.
This was first studied in [13] for the case of scalar multiplets with osp(1|2) symmetry.
A gauge theory can be similarly constructed [15]. We then take the flat limit with an
224
appropriate scaling of operators. We can also understand the noncommutative superspace

as a constrained system whose phase space dimension is reduced by the second
class constraints. This give an interpretation that the noncommutative superspace is a
supersymmetric analog of the lowest Landau level system. In this section, we take the
osp(1|2) super-Lie algebra. This gives a two-dimensional noncommutative supersphere
with two real supercharges on the fuzzy sphere.
3.1. osp(1|2) algebra and fuzzy supersphere
The graded commutation relations of osp(1|2) algebra are given by

li , lj = iij k lk ,
1

li , v = (i ) v ,
2
1
{v , v } = (Ci ) li ,
2
(3.1)
where C = i2 . The even part of this algebra is su(2) which is generated by li (i =

1, 2, 3) and the odd generators v ( = 1, 2) are su(2) spinors. In this paper, we also
write v1 = v+ and v2 = v . The irreducible representations of osp(1|2) algebra [21] are
characterized by the values of the Casimir operator K 2 = li li + C v v = L(L + 12 )
where quantum number L is called superspin and L Z0 /2. Each representation consists
of spin L and L 12 representations of su(2), |L, l3 , |L 12 , l3 and its dimension is
N = (2L + 1) + 2L = 4L + 1.
The condition K 2 = L(L + 12 ) defines the two-dimensional supersphere. Consider
polynomials (li , v ) of the representation matrices li and v with superspin L. Let us
denote the space spanned by (li , v ) as AL . The osp(1|2) algebra acts on AL by the
three kinds of action, the left action (liL , vL ), the right action (liR , vR ) and the adjoint
action (Li liL liR , V = vL vR ),
liL = li ,
liR = li ,
vL = v ,
(3.2)
vR
(3.3)
Li = [li , ],
= v ,
V = [v , ].
(3.4)
S
We can define supersymmetrized spherical harmonics Ykm
which are generalization of the
ordinary spherical harmonics to the supersphere (see [14] for the details). k can take either
an integer or half an integer value and they are bosonic or fermionic functions, respectively.
Any N N supermatrix can be expanded in terms of the superspherical harmonics as
(li , v ) =
2L

S
km Ykm
,
(3.5)
k=0,1/2,1,...
where the coefficient km for the even (odd) spherical harmonics is Grassmann even (odd).
We can map the supermatrix (li , v ) to a function on the superspace (xi , ) by

S
(li , v ) (xi , ) =
(3.6)
km ykm
(xi , ),
k,m
225
S
where ykm
(xi , ) are ordinary superspherical functions. A product of supermatrices is
mapped to a noncommutative star product of functions. An explicit form of the star product
is given in [26].
In addition to the osp(1|2) generators (li , v ), we can define additional generators with
which they form bigger algebra osp(2|2). These additional generators are

1
1
C v v + 2L L +
,
=
(3.7)
L + 1/4
2

1
d = [ , v ] =
(3.8)
(i ) v li + li v .
2(L + 1/4)
Commutation relations for the additional generators are given by

1

li , d = (i ) d ,
, d = v ,
, li = 0,
[ , v ] = d ,
2

1
1
d , d = (Ci ) li ,
(3.9)
v , d = C .
2
4
The adjoint action of the fermionic generators D = adj d plays a role of the covariant
derivatives. On the other hand, the adjoint action of the original fermionic generators
Q = adj v are interpreted as supersymmetry generators. We will show that they commute
in the flat limit. These additional generators also play an important role in constructing
kinetic terms for a scalar multiplet on the supersphere [13].
The commutative limit is discussed in [13] and the fuzzy supersphere becomes the
ordinary two-dimensional supersphere with two real Grassmannian coordinates. This limit
can be taken by keeping the radius r of the sphere fixed and taking the large L limit.
3.2. Flat noncommutative superspace

We now take a flat limit. Namely we consider the vicinity of the north pole. For the
bosonic generators, we perform the same scaling as (2.2) to obtain the flat noncommutative
coordinates. For the fermionic generators, in
order to keep the noncommutativity, one
possible choice is to scale both of v as L. Then from the osp(1|2) algebra (3.1)
we can read that the anticommutator of v+ and v survives to be a constant. The
other anticommutators among v vanish. But with this choice we can easily see that the
supersymmetry algebras acting on the noncommutative superspace become trivial, i.e.,
the anticommutators between supercharges do not generate translation of the space. All
supersymmetries become nondynamical symmetries. So we need to take another scaling
limit where v+ and v are scaled asymmetrically. We define new superspace coordinates
as

xi =
li for i = 1, 2,
L
1/4

3/4
+ = 2
(3.10)
v+ ,
= 2
v .
L
L
We multiply a constant on for convenience. Since we are considering around the north
pole, the l3 is scaled as L. With the above scaling, the algebra among the coordinates
226
become
[x+ , x ] = 2 or [x1 , x2 ] = i,

+ , = ,
+ , + = x+ ,

x , + = ,
(3.11)
where we have defined x = x1 i x 2 . All the other commutators vanish. By redefining the
coordinates as
+ = + +
1
x+ ,
2
= ,
(3.12)
(3.13)
the noncommutativity of the coordinates is written simply as a canonical form:

[x+ , x ] = 2,
{ + , } = .
(3.14)
The scaling of the additional generators d can be automatically determined from the
scaling (3.10) because they are written in terms of the osp(1|2) generators. The scaled
generators are defined by

1/4
1
1

d+ = + + x+ = + +
d+
(3.15)
= 2
x + ,
L
2

3/4

d = 2
(3.16)
d = = .
L
The anticommutator with is given by

d , = C ,
(3.17)
which is consistent with the scaling of . becomes a constant and commutes with all the
other generators.
Now we define generators of supersymmetry transformations and covariant derivatives
which mutually commute by
1
P = adj x ,
2
Q = adj .
(3.18)
(3.19)
They generate supersymmetry transformations on the superspace coordinates (x,

).
They
satisfy the supersymmetry algebra
[P+ , P ] = 0,
(3.20)
1
[P , Q+ ] = Q ,
2
{Q+ , Q } = 0,
(3.22)
{Q+ , Q+ } = 2P+ ,
(3.23)
{Q , Q } = 0.
(3.24)
(3.21)
227
Q+ is a dynamical supersymmetry and generates spacetime translation into x direction.

But the Q is a nondynamical supersymmetry and its anticommutator vanishes. This is
caused by the asymmetric scaling of the coordinates. Because of this asymmetric scaling,
we cannot take a further limit to obtain an ordinary two-dimensional superspace with two
dynamical supersymmetries.
Covariant derivatives can be defined similarly by the adjoint action of d :
D = adj d .
(3.25)
They anticommute with Q : {D , Q } = 0. Their commutation relations are:

{D+ , D+ } = 2P+ ,
(3.26)
{D+ , D } = {D , D } = 0,
(3.27)
1
[P , D+ ] = D .
(3.28)
2
Functions on the superspace (x,
)
are given as N N supermatrices. Generically they
depend on full set of supercoordinates: (x+ , x , + , ). We can consistently constrain
them by imposing the chiral constraint as
D = [ , ] = 0.
(3.29)
This automatically leads to

Q = 0
(3.30)
and the superfield depends only on (x+ , x , ). This is the chiral superfield and
supersymmetry is generated by Q+ whose anticommutator becomes P+ . To summarize,
on the chiral superfields (x+ , x , ), the algebra of supersymmetry Q+ , translations
P and the covariant derivative D+ is given by
{Q+ , Q+ } = {D+ , D+ } = 2P+ ,
(3.31)
[P , Q+ ] = [P , D+ ] = 0,
(3.32)
[P+ , P ] = 0.
(3.33)
They are written as differential operators on chiral superfields (x+ , x , ):
,
x

,
Q+ =

x
D+ =
+
.

x
P =
(3.34)
(3.35)
(3.36)
We can further constrain the superfield by P :

P = 0.
(3.37)
Then the superfield becomes independent of the x+ coordinate: (x , ), and the system
becomes essentially one-dimensional.
228
3.3. Magnetic field in 2d superspace

The noncommutative superspace can be understood as a system restricted by some
constraints analogous to the lowest Landau level states in the bosonic case. In this
subsection, we first derive mutually commutative set of differential operators acting on
superparticles in the commutative superspace. They correspond to the covariant derivatives
in a magnetic field and the guiding center coordinates discussed in the previous section.
In order to obtain mutually commutative set of generators on the flat space, we
again
begin with the osp(1|2) algebra (3.1), where we denote the generators as li = Li ,
2 v = Q , and take the InnWigner contraction. We parametrize the group manifold

of osp(1|2) by (x, y, , ) as g = exp(ix L) exp(i Q ). We then take the scaling as

+ s +,
s3 ,
x, y sx, sy,
s 2 ,
(3.38)
and take s 0 limit.
If we take up to the second order of s, we have the algebra

Lx[1] , Ly[1] = iLz[2] ,
L[1] , Q+[1/2] = Q[3/2] ,
Q+[1/2] , Q+[1/2] = L+[1] ,

others = 0.
Q+[1/2] , Q[3/2] = Lz[2] ,
(3.39)
Covariant derivatives are defined as the right multiplication on the group manifold
generated by this algebra. They are calculated in Appendix B resulting as
y
1
+
+
,
ix 2i 2
x
1
Dy =
+ +
,
,
Dz =
iy 2i 2i
i

x + iy +
1 +
1
D+ =
+
i
,
i + 2
4i
2
x
y
1
+ ,
D =
i
2
Dx =
(3.40)
satisfying
[Dx , Dy ] = iDz ,
{D+ , D+ } = D+ ,
[D , D+ ] = D ,
{D+ , D } = Dz ,
others = 0.
(3.41)
These commutators are supersymmetric generalization of (2.15). The terms in the r.h.s.
with a Dz term are interpreted as the effect of the background magnetic field.
The global charges (left multiplications) are similarly calculated in Appendix C resulting as
y
x
,
L y =
+
,
L z =
,
L x =
ix 2i
iy 2i
i

x + iy
1
1 +
Q + =
+
i
+
,
i + 2
2 2
x
y
Q =
1
+ + ,
i
2
satisfying

z,
L x , L y = i L
+,
+ , Q + = L
Q
229
(3.42)

,
L , Q + = Q
Q + , Q = L z ,
others = 0.
(3.43)
These global charges commute/anticommute with covariant derivatives including

{Q, D} = 0. These operators are interpreted as the guiding center coordinates in a constant magnetic field on the two-dimensional superspace.
The covariant derivatives (3.40) can be considered as the supercovariant derivatives
(1.1), (1.2) in a constant magnetic field. In (3.40), the terms containing derivatives with
respect to are contributions from the gauge fields. Applying the definition of the field
strength given in the introduction (1.4), the above system has a constant magnetic field in
(x, y) and (+ , ) directions;
Fx,y = B,
F+ , = iB,
(3.44)
where we write Dz = B. In the next subsection we show that this induces the
noncommutativity on superspace.
3.4. Noncommutative superspace as the lowest Landau level system
We now calculate the Dirac bracket of the supercoordinates by imposing the lowest
Landau level constrains. Since is a coordinate conjugate to the center Lz[2] we can fix
it as a constant; Dz = 1 = 0. This leads to the fermionic noncommutative algebra as
well as the bosonic noncommutative algebra (2.17),
[Dx , Dy ] = i 1 ,
{D+ , D } = 1 .
(3.45)
These algebras are shown to induce noncommutative fermionic coordinates as well as

noncommutative bosonic coordinates:
+
[X, Y ] = i,
, = .
(3.46)
The lowest Landau level conditions are
(Dx iDy )|LLL = 0,
(3.47)
D |LLL = 0.
(3.48)
Analogous to the bosonic case the canonical analysis can be performed. Second class
constraints are given by Dx = Dy = D+ = D = 0 and the Dirac bracket for the system is
calculated as (we have dropped terms up to the second class constraints)

[O1 , O2 }D = [O1 , O2 } + i[O1 , Dx }[Dy , O2 } i[O1 , Dy }[Dx , O2 }

[O1 , D+ }[D , O2 } [O1 , D }[D+ , O2 } . (3.49)
Then the Dirac brackets of the original coordinates become
+
[x, y]D = i,
, D = ,
others = 0.
(3.50)
230
We can introduce noncommutative guiding center coordinates which are written in terms
of the global charges (Li , Q ) in such a way that they are equal to the original coordinates
up to the second class constraints. The results are

1
i
X=
x 2
= L y = x Dy + D ,
2
iy
2

1
i
y + 2
= L x = y + Dx + + D ,
Y=
2
ix
2
1
= + + iD ,
+ = + + = i Q
2
+ + L + = + iD+ + D+ + .
= + + = i Q
(3.51)
2
2
2
They satisfy
[X, Y ] = [x, y]D = i,
{ + , } = { + , }D =
(3.52)
and all the other commutators vanish. This algebra is nothing but the canonical
commutation relation (3.14) for the redefined coordinates. The fermionic coordinates
are related to as
= .
(3.53)
4. Noncommutative superspace from su(2|1)

In this section, we consider noncommutative superspace based on su(2|1) superalgebra
(or equivalently osp(2|2) algebra). This gives two-dimensional supersphere with four real
supercharges. This type of noncommutative superspace was studied in [31] though an
explicit relation to our case is not manifest.
The graded commutation relations of su(2|1) algebra are given by


1
li , q = (i ) q ,
2
1

1
q = q ,
q = q ,
B,
B,
2
2
q , q = (i ) li + B,
others = 0.
1
li , q = (i ) q ,
2
(4.1)
This contains an osp(1|2) superalgebra as a subalgebra. There are two Casimir operators.
The second Casimir operator of su(2|1) algebra is given by

1
K 2 = li li B 2 + q q q q .
2
(4.2)
The third Casimir operator is K 3 K 2 B + . We will consider a coset space

SU(2|1)/U (1)2 and this defines a supersphere with four supercharges.
231
Typical irreducible representations are characterized by two quantum numbers, (b, L)

[21]. The eigenvalues of the two Casimir operators are given by K2 = L2 b2 and
K3 = b(L2 b2). (There are other types of irreducible representations but we do not
consider them here.) In terms of the osp(1|2) subalgebra, this representation is decomposed
into two representations with superspin L and L 1/2. Hence the dimension of the
irreducible representation is N = 8L. Any supermatrix with this size can be expanded
in terms of polynomials generated by li , q and q . B can be solved by the third Casimir
Since three li satisfy the constraint given
K 3 and the polynomials do not depend on B.
by the second Casimir, this defines two-dimensional supersphere with four Grassmannian
coordinates.
In order to take a flat limit, we introduce the following superspace coordinates,
1/2
xi =
li for i = 1, 2,
L
1/4
3/4
3/4
1 =
q1 ,
2 =
q2 ,
=
q 1 ,
L
L
L
1/4

1
b =
(4.3)
q 2 ,
B.
2 =
L
L
Again we need to take an asymmetric scaling for the fermionic coordinates. The l3 is scaled
as L since we are considering the vicinity of the north pole on the sphere. In the large L
limit, the algebra among the coordinates becomes

x+ , 1 = 2 ,
x+ , 2 = 1 ,
[x+ , x ] = 2,
2
1
1 , = x ,
1 , = (b + 1),
2
others = 0,
2 , = (b 1),
(4.4)
where x = x1 i x 2 . Since b is a center of the algebra, we set it as a constant b.
Furthermore, by the transformations
1
x 2 ,
2 = 2 ,
2
1
x 1 ,
1 = 1 ,
(4.5)
2 = 2
2
this algebra reduces to a simpler noncommutative algebra for the superspace coordinates;
1 = 1 +
[x+ , x ] = 2,
1 , 1 = (b + 1),

2 , 2 = (b 1),
others = 0.
(4.6)
Now we define the generators of supersymmetry in the flat limit as

1
P = adj(x ),
2

Q = adj ,
Q = adj .
(4.7)
232
Then the following supersymmetry algebra holds

1
[P+ , Q1 ] = Q2 ,
2

2 = 2P ,
Q1 , Q
1
P+ , Q 2 = Q 1 ,
2
others = 0.
(4.8)
There are four supercharges but only half of them generated by Q1 and Q 2 are dynamical
supersymmetries. The other half are non-dynamical supersymmetries and do not generate
spacetime translation.
We next introduce additional operators2 corresponding to the generators Eq. (3.8) in the
osp(1|2) case:

1
(i ) li q + q li B q + q B ,
d
2L
d 1
( ) l q + q l B q + q B .
(4.9)
i
i
i
2L
We can obtain covariant derivatives by taking the scaling limit of these generators:

D = adj d ,
D = adj d ,
1/4
1
1+b

d1 =
x 2 ,
d1 = (1 b)1 + x 2 = (1 b) 1 +
L
2
3/4
d2 = (1 + b)2 = (1 + b) 2,
d2 =
L
3/4
d 1 =
d1 = (1 b)1 = (1 b) 1 ,
L
1/4
1
1 b 1
x .
(4.10)
d2 = (1 + b)2 + x 1 = (1 + b) 2 +
d 2 =
L
2
and satisfy the following algebra:
and Q
The covariant derivatives anticommute with Q

1
1
[P+ , D1 ] = D2 ,
P+ , D 2 = D 1 ,
2
2

D1 , D 2 = 2(b2 1)P .
(4.11)
are supermatrices and generally written as

Functions on the superspace (x,
,
)
functions of supercoordinates: (x , , ). We can constrain the function by imposing
the following constraints,

D 1 = (1 b) 1 , = 0.
D2 = (1 + b)[ 2, ] = 0,
(4.12)
These conditions automatically mean
Q2 = Q 1 = 0,
(4.13)
2 These operators are introduced so that they anticommute with the supersymmetry generators in the flat limit.
There might be possible 1/L corrections to them before taking the flat limit.
233
and the superfield depends only on (x , 2 , 1 ). This is the chiral superfield3 and
supersymmetries are generated by Q1 and Q 2 whose anticommutator becomes P . As
a result, on the chiral superfields, the algebra among generators for supersymmetries Q1
and Q 2 , translations P , and covariant derivatives D1 and D 2 is given by

2 = 2P ,
Q1 , Q
(4.14)
D1 , D 2 = 2(b2 1)P ,
others = 0.
They are written as differential operators:
P =
,
x
,
Q 2 = (1 b)
1
,
Q1 = (1 + b) 1 + 2

x+
2
x+

D1 = (1 + b) (1 b) 1 2
,

x+

.
1
D 2 = (1 b) (1 + b)
2
x+
We can further constrain the superfield by P+ :
P+ = 0.
(4.15)
Then the superfield becomes independent of the x coordinate and the system becomes
essentially one-dimensional system with two supersymmetries.
5. Noncommutative superspace from psu(2|2)

In this section we try to construct a noncommutative superspace in four dimensions
based on psu(2|2) graded algebra. There are various possibilities for the scaling of
Grassmannian coordinates when we take a flat limit. We show two examples. More details
will be discussed in a separate paper.
The psu(2|2) graded algebra is given by


1
1
li , q = (i ) q ,
li , q = (i ) q ,
2
2

1
l i , q = 1 (i ) q ,
l i , q = (i ) q ,
2
2
q , q = (i ) li (i ) l i .
(5.1)
The bosonic part of su(2|2) consists of two sets of su(2) algebra generated by li and li and
eight odd generators transform as spinors both under two su(2)s. Based on this graded
algebra we can construct a four-dimensional superspace PSU(2|2)/U (1)2.
3 It would have been appropriate to parametrize the superspace so that both of the two constraints are antichiral and the constrained superfield becomes apparently chiral. It is merely a problem of notation.
234
5.1. d = 4 flat noncommutative superspace

In order to take a flat limit we first consider the following scaling of the superspace
coordinates:
1/2
1/2
l for i = 1, 2,
li ,
xi =
x i =
i
L
L
1/4
1 1 1 2
1 1 1 2
2 , 1 , 2 , 2 =
q2 , q 1 , q 2 , q 2 ,
L
3/4
1 2 2 2
1 2 2 2
1 , 1 , 2 , 1 =
(5.2)
q1 , q1 , q2 , q 1 .
L
We use a similar asymmetric scaling of the fermionic coordinates to the two-dimensional
cases. The l3 and l3 are scaled as L and cL respectively, where c is an arbitrary constant.
Taking the large L limit the algebra among the coordinates becomes
[x+ , x ] = 2,
[x + , x ] = 2c,

x , 21 = 11 ,
x , 11 = 21 ,
x , 21 = 22 ,
x , 22 = 21 ,
1 1
1 2
2 , 1 = x+ ,
2 , 2 = x + ,
1 1
2 1
1 , 1 = (1 c),
1 , 2 = (1 + c),
1 2
2 2
2 , 1 = (1 + c),
2 , 2 = (1 c),
(5.3)
others = 0.
By introducing the following fermionic coordinates:
1
1 2
x+ 11 +
x + 2 ,
2
2c
1
1 2
x+ 21 ,
22 = 22
11 = 11 +
x + 1 ,
2
2c
= , = , for others,
21 = 21
(5.4)
this algebra is much simplified to satisfy the canonical forms:

[x + , x ] = 2c,
[x+ , x ] = 2,
2 1
1 1
1 , 2 = (1 + c),
1 , 1 = (1 c),
2 2
1 2
2 , 2 = (1 c).
2 , 1 = (1 + c),
(5.5)
We define the generators of supersymmetry in the flat limit as

1
P = adj x ,
2
Q = adj ,
1
P = adj x ,
2
Q = adj .
(5.6)
Then the supersymmetry algebra has the following forms:

1
1

P , Q12 = Q11 ,
P , Q 11 = Q 21 ,
2
2

1 2

1
2
1
2
= Q ,
P , Q
P , Q2 = Q2 ,
2
2
2 1
1 1
1 2
= 2P+ ,
Q2 , Q
Q2 , Q 2 = 2P+ ,
others = 0.
1
235
(5.7)
Only three supercharges Q12 , Q 11 and Q 22 generate dynamical supersymmetries, i.e.,

anticommutators among them become the generators of the spacetime translations. The
other supercharges are generators of nondynamical supersymmetries. In order to construct
covariant derivatives, it is useful to consider the following operators:

1

(i ) li q + q li + (i ) li q + q li ,
d =
(5.8)
2L

1

(i ) li q + q li + (i ) li q + q li .
d =
(5.9)
2L
We can obtain the covariant derivatives which anticommute with the supercharges in the
flat limit:
3/4
1
adj d11 = (1 + c) adj 11 ,
D1 =
L
3/4
D12 =
adj d12 = (1 c) adj 12 ,
L
1/4

1
1 2
1
1
1
1
D2 =
adj d2 = adj (1 c)2 x+ 1 x + 2
L

1+c
1 + c 2
x+ 11
= adj (1 c) 21
x + 2 ,
2
2c
3/4
adj d22 = (1 + c) adj 22 ,

D22 =
L
1/4

1
1
1
D1 =
adj d 1 = adj (1 + c) 1 + x+ 1
L

1 c 2
x+ 1 ,
= adj (1 + c) 11 +
2
3/4
D 12 =
adj d21 = (1 c) adj 21 = (1 c) adj 21 ,
L
1/4
D2 =
adj d12 = (1 c) adj 12 = (1 c) adj 12 ,
L
1/4

1
D 22 =
adj d22 = adj (1 + c) 22 + x + 21
L
c
= adj (1 + c) 22
(5.10)
x + 21 .
2c
236
The covariant derivatives and the generators of spacetime translations satisfy the
following algebra:

1
1
1

P , D21 = D11 ,
P , D 11 = D 12 ,
P , D21 = D22 ,
2
2
2

P , D 22 = D 12 ,
D21 , D 11 = 2 1 c2 P+ ,
2

1 2
others = 0.
D , D = 2 1 c2 P+ ,
2
(5.11)
Field theories based on this algebra in the flat limit have two-dimensional like supersymmetries because only P+ and P+ appear in the right-hand sides of the algebra (5.7) and (5.11).
Next we consider another choice of scaling for the fermionic superspace coordinates:

21 , 12 , 11 , 22 =
11 , 22 , 12 , 21 =

1/4
3/4
1 2 1 2
q2 , q1 , q 1 , q 2 ,
1 2 1 2
q1 , q2 , q 2 , q 1 .
(5.12)
Then in the large L limit the coordinates of the superspace satisfy

[x+ , x ] = 2,
[x + , x ] = 2c,

x , 21 = 11 ,
x+ , 22 = 12 ,
x+ , 12 = 22 ,

x + , 12 = 11 ,
x , 21 = 22 ,
x + , 11 = 12 ,
2 1
2 2
1 1
1 , 1 = x ,
1 , 2 = x ,
2 , 1 = x+ ,
1 1
1 2
1 , 1 = (1 c),
2 , 1 = (1 + c),
2 1
2 2
1 , 2 = (1 + c),
2 , 2 = (1 c).

x , 11 = 21 ,

x , 22 = 21 ,
1 2
2 , 2 = x + ,
(5.13)
We define the generators of supersymmetries and spacetime translations as Q = adj ,
Q = adj , P = 12 adj x and P = 12 adj x . In this case, it can be easily seen that
the anticommutators among four supercharges Q21 , Q12 , Q 11 and Q 22 become four space
time translations P and P . Therefore, noncommutative theories based on this algebra
have four dynamical supersymmetries.
5.2. N =
1
2
noncommutative superspace as a constrained system
Here we briefly explain that the N = 12 noncommutative superspace, which was first
discussed in [1] and further studied in [3], can be understood as a constrained system and
the noncommutative algebra for the coordinates is realized by the Dirac bracket under the
constraints.
237
We begin with a little more general setting. We consider supercovariant derivatives (1.1),
(1.2) with the following gauge field backgrounds:
i
D = f x ,
2

i
D = i + f ,
2

D = + i f .
2

(5.14)
(5.15)
They are natural generalization of the bosonic covariant derivatives in a constant magnetic
field, but the field strengths depend on the superspace coordinates:
F = f ,
F = f ,

i
F = f ,
2

F = if x ,

i
F = f ,
2
F = f .
(5.16)
Imposing D = D = D = 0 as the second class constraints, we can calculate the Dirac

brackets of the superspace coordinates and obtain the noncommutative algebra similar to
that given in [9].
In the following we consider an easier case for simplicity, namely, a case where only f
are nonvanishing. We then impose the second class constraints by only D s and calculate
the Dirac bracket
1
[A, B}D = [A, B} + i[A, D }f
[D , B}
(5.17)
as
1
,
= if

1
x , D = f ,
x , x
1
.
{ , }D = if
(5.18)
This is nothing but the noncommutative algebra given in [1,3]. The noncommutative
1
parameter C in [3] is related to our f as C = if
. If we redefine the coordinate
as
y = x + i ,
they become commutable with the other coordinates

y , y D = y , D = y , D = 0.
(5.19)
(5.20)
In this case where only f is nonvanishing, we can more easily obtain the canonical
pairs on the reduced superspace by a similarity transformation

O exp i O exp i .
(5.21)
238
Then the mutually commutative set of supercovariant derivatives D and global charges
q are given by
i
+ f ,
2

1
q = i 2 + f .
2
The remaining coordinates on the phase space which (anti)commute with D are

1
1
D +
q + 2 = if
= f
D =
(5.22)
(5.23)
(5.24)
and (x , /x , , / ). They satisfy the canonical algebra with

1
.
{ , } = if
(5.25)
The set ( , x , /x , , / ) gives the phase space coordinates for the constrained system we are considering now. When we construct a field theory on the noncommutative space (x , , ), we can introduce the canonical conjugate to as the
adjoint action
/ = i adj q .
Then the covariant derivatives4 and the supercharges can be defined in the same way as
that given in [3].
6. Conclusions and discussions

In this paper, we have constructed noncommutative superspaces based on graded
(super-)Lie algebras. In particular, we consider fuzzy supersphere based on osp(1|2) and
su(2|1) algebras. They give two-dimensional supersphere with two and four real supercharges. We then consider flat limits. In order to take flat limits with the fermionic noncommutativity, we needed to take an asymmetric scaling limit for fermionic coordinates
on superspace. We also obtained covariant derivatives and imposed chiral constraints to
remove half degrees of freedom. This method was generalized to four-dimensional noncommutative superspaces based on psu(2|2) algebra [15]. In this case, there are varieties
to assign scalings to the fermionic coordinates when we take a flat limit. We showed two
examples. One is similar to the two-dimensional cases and we have obtained supersymmetry generators and covariant derivatives. This system is two-dimensional like in a sense that
only two generators of spacetime translation appear in the anticommutators of the supersymmetry generators. The other example is more nontrivial. With this scaling, spacetime
translation generators into the all four directions appear. More details are left for future investigations. It would be also interesting to investigate other scaling limits of the psu(2|2)
or su(2|2) algebras which can give the noncommutative superspace given in the paper [34].
4 It is confusing to use the same word as the operator defined in (5.22). The covariant derivative in (5.22) is
an operator used to define the noncommutative superspace as a constrained system. The covariant derivative here
is an operator acting within the constrained space that anticommutes with the supercharges.
239
We have also investigated these noncommutative superspaces as constrained systems.

This is an analogue of the lowest Landau level system of particles moving in a
constant magnetic field. We obtained two sets of operators, supercovariant derivatives and
superguiding center coordinates. They are obtained by the right and the left multiplications
on the group manifolds. So they are commutative to each other. The lowest Landau level
conditions are generalized by adding fermionic constrains in addition to the ordinary
bosonic condition for the lowest Landau level states. Imposing gauge fixing conditions,
we calculated the Dirac brackets for superspace coordinates to obtain noncommutative
superspace coordinates. This method can be extended to more general cases to obtain more
general noncommutative superspaces. We want to report it in a future publication. Along
this line, it will also be interesting to investigate a supersymmetric generalization of W
algebraic structure, which plays an important role to study the physics of the lowest Landau
level systems [32].
Our construction has an advantage that Jacobi identities of the supersymmetries and the
associativities of the star products are manifest, but it is restricted to the noncommutative
superspaces with Lie algebraic noncommutativity. Namely, the supersymmetry algebras
satisfy Lie algebras. In the papers [1,3], the anticommutators for supersymmetries contain
the second order derivative operators. In order to construct these structures based on a
supermatrix approach, we may need to investigate supermatrix models without the graded
Lie algebraic structures.
Acknowledgements
We would like to thank Drs. H. Fuji, T. Kimura, Y. Kitazawa, M. Sakaguchi and
T. Suyama for discussions. The work of H.U. is supported in part by JSPS Research
Fellowships for Young Scientists. The work of S.I. is supported in part by the Grant-inAid for Scientific Research from the Ministry of Education, Science and Culture of Japan.
Appendix A. Cartan one-forms and the generalized InnWigner contraction

In this appendix, we briefly explain the method of Cartan one-forms to obtain the
left and right multiplications. Then we explain the generalized InnWigner contraction
proposed in [29].
Suppose we have a Lie algebra [Ta , Tb ] = fab c Tc . A group manifold generated by
a
this algebra can be parametrized as g = eix Ta . Cartan one-forms ea are defined by
a e b e c /(2i). If
g 1 dg = iea Ta and satisfy the MaureCartan (MC) equation dea = fbc
a
a
m
1
m
we write them as e = dx em (x), covariant derivatives Da = (e )a m /i generate
the right multiplication g gh and obey the Lie algebra [Da , Db ] = fab c Dc . The left
multiplication generators are similarly obtained from dg g 1 and commute with the right
multiplications.
The generalized IW contraction can be obtained as follows. We rescale the parameter
x a on the group manifold as x a s na x a and take s 0 limit. Since the Cartan one-form
240
ea is written in terms of a polynomial of xs, it can be expanded by s as

s n ea[n] .
ea =
(A.1)
Here we interpret that each ea[n] is a different Cartan one-form corresponding to different
generators Ta[n] . In this sense, this is an expansion [30] rather than a contraction [27,28].
MC equations are satisfied order by order
1 bc
(A.2)
f a eb[l] ec[nl]
2i
and they determine commutation relations between the generators Ta[n] . It is obvious from
the MC equations (A.2) that the commutation relations are closed among generators with
weights [n] less than some fixed number. Jacobi identities are automatically satisfied.
dea[n] =
Appendix B. Supercovariant derivatives

In this appendix, we give a derivation of (3.40). An group element of the noncommutative supertranslation group generated by TA = {Lx , Ly , Lz , Q } satisfying (3.40)
is parameterized as g(x, y, , + , ). Left-invariant Cartan 1-forms are obtained by
g 1 dg = dzM EM A TA as
i
1
Ex = dx d + + ,
Ey = dy + d + + ,
2
2

i
1
Ez = d (dx y dy x) + d + + d + ,
2
2
i
E + = d + ,
E = d + (dx + i dy) + .
2
The coefficients of the Cartan 1-forms are give as
1
0
y2 0 2i +
x
1
0 12 +
0
2
EM A =
0
1
0
0
,
0
i + 1 + i
1
0
2
2
2
0
i +
2
and whose inverse is given as
1
0
1
0
1 M
E A = 0
0
i +
1 +
2
2
0
(B.1)
(B.2)
1
y
2
x2
2i +
1 +
2
2i + 14 (x + iy) +
2i +
(B.3)
Therefore, the supercovariant derivatives are given as

1 M 1
DA = E A
M
i
241
(B.4)
whose components are (3.40).
Appendix C. Global charges

In this appendix, we give a derivation of (3.42). Under the global transformations a
group element g is transformed into g Gg with infinitesimal parameters
g 1 g = g 1 (G 1)g D E A TA = zM EM A TA .
Expression of D E A s are obtained as

i
1 i + Q+
+
+
+
g e
1 g = i i L+ + i (x + iy) Lz
2

i
+ Q+ (x + iy)Q ,
2

+

1 i Q
1 g = i i Lz + Q ,
g e

x

i
g 1 ei Lx 1 g = ix Lx yLz + + Q ,
2

1
y
g 1 ei Ly 1 g = iy Ly + xLz + Q ,
2

1 i Lz
g e
1 g = i Lz .
The global charges are given by

1
1
Q = zM
M = D E A E 1 A M
M .
i
i
References
[1] H. Ooguri, C. Vafa, hep-th/0302109;
H. Ooguri, C. Vafa, hep-th/0303063.
[2] J. de Boer, P.A. Grassi, P. van Nieuwenhuizen, hep-th/0302078.
[3] N. Seiberg, JHEP 0306 (2003) 010, hep-th/0305248.
[4] V. Schomerus, JHEP 9906 (1999) 030, hep-th/9903205.
[5] N. Seiberg, E. Witten, JHEP 9909 (1999) 032, hep-th/9908142.
[6] J.H. Schwarz, P. Van Nieuwenhuizen, Lett. Nuovo Cimento 34 (1982) 21.
[7] P. Kosinski, J. Lukierski, P. Maslanka, hep-th/0011053.
[8] S. Ferrara, M.A. Lledo, JHEP 0005 (2000) 008, hep-th/0002084.
[9] D. Klemm, S. Penati, L. Tamassia, Class. Quantum Grav. 20 (2003) 2905, hep-th/0104190;
D. Klemm, S. Penati, L. Tamassia, math-ph/9804013.
(C.1)
(C.2)
(C.3)
242
[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]
H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Nucl. Phys. B 565 (2000) 176, hep-th/9908141.
N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Nucl. Phys. B 573 (2000) 573, hep-th/9910004.
N. Seiberg, JHEP 0009 (2000) 003, hep-th/0008013.
H. Grosse, C. Klimcik, P. Presnajder, Commun. Math. Phys. 185 (1997) 155, hep-th/9507074.
H. Grosse, G. Reiter, J. Geom. Phys. 28 (1998) 349.
M. Hatsuda, S. Iso, H. Umetsu, in preparation.
H. Kawai, T. Kuroki, T. Morita, hep-th/0303210.
M.F. Sohnius, Phys. Rep. 128 (1985) 39204.
S.J. Gates Jr., M.T. Grisaru, M. Rocek, W. Siegel, Superspace, BenjaminCummings, Redwood City, CA,
1983.
N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, Nucl. Phys. B 498 (1997) 467, hep-th/9612115.
L. Smolin, hep-th/0006137;
T. Azuma, S. Iso, H. Kawai, Y. Ohwashi, Nucl. Phys. B 610 (2001) 251, hep-th/0102168.
A. Pais, V. Rittenberg, J. Math. Phys. 16 (1975) 2062;
A. Pais, V. Rittenberg, J. Math. Phys. 17 (1976) 598, Erratum;
M. Scheunert, W. Nahm, V. Rittenberg, J. Math. Phys. 18 (1977) 155;
M. Marcu, J. Math. Phys. 21 (1980) 1277.
A.Y. Alekseev, A. Recknagel, V. Schomerus, JHEP 9909 (1999) 023, hep-th/9908040.
J. Madore, Class. Quantum Grav. 9 (1992) 69.
S. Iso, Y. Kimura, K. Tanaka, K. Wakatsuki, Nucl. Phys. B 604 (2001) 121, hep-th/0101102.
S.R. Coleman, The magnetic monopole fifty years later, HUTP-82/A032.
A.P. Balachandran, S. Kurkcuoglu, E. Rojas, JHEP 0207 (2002) 056, hep-th/0204170.
E. Inn, E.P. Wigner, Proc. Natl. Acad. Sci. USA 39 (1953) 510;
E. Inn, in: F. Gursey (Ed.), Group Theoretical Concepts and Methods in Elementary Particle Physics,
Gordon and Breach, New York, 1964.
M. Hatsuda, K. Kamimura, M. Sakaguchi, Nucl. Phys. B 632 (2002) 114, hep-th/0202190;
M. Hatsuda, K. Kamimura, M. Sakaguchi, Nucl. Phys. B 637 (13) (2002) 168, hep-th/0204002;
M. Hatsuda, K. Kamimura, M. Sakaguchi, Prog. Theor. Phys. 109 (2003) 853, hep-th/0106114.
M. Hatsuda, M. Sakaguchi, Phys. Rev. D 66 (2002) 045020, hep-th/0205092;
M. Hatsuda, M. Sakaguchi, Prog. Theor. Phys. 109 (2003) 853, hep-th/0106114.
J.A. de Azcarraga, J.M. Izquiero, M. Picon, O. Varela, hep-th/0212347.
C. Klimcik, Commun. Math. Phys. 206 (1999) 587, hep-th/9903202.
S. Iso, D. Karabali, B. Sakita, Phys. Lett. B 296 (1992) 143, hep-th/9209003;
S. Iso, D. Karabali, B. Sakita, Nucl. Phys. B 388 (1992) 700, hep-th/9202012;
A. Cappelli, C.A. Trugenberger, G.R. Zemba, Phys. Lett. B 306 (1993) 100, hep-th/9303030.
R. Britto, B. Feng, S.J. Rey, hep-th/0306215.
N. Berkovits, N. Seiberg, hep-th/0306226.

The scalar sector of the RandallSundrum model

Daniele Dominici a , Bohdan Grzadkowski b , John F. Gunion c ,
Manuel Toharia c
a Dipartimento di Fisica, Florence University and INFN, Via Sansone 1, 50019 Sesto F. (FI), Italy
b Institute of Theoretical Physics, Warsaw University, Hoza 69, PL-00-681 Warsaw, Poland
c Davis Institute for High Energy Physics, University of California Davis, Davis, CA 95616-8677, USA
Received 30 July 2002; received in revised form 1 August 2003; accepted 14 August 2003
Abstract
We consider the scalar sector of the RandallSundrum model. We derive the effective potential
for the Standard Model Higgs-boson sector interacting with KaluzaKlein excitations of the graviton
(hn
) and the radion () and show that only the Standard Model vacuum solution of V (h)/h = 0
(h is the Higgs field) is allowed. We then turn to our main focus: the consequences of the curvature H
(where H
is a Higgs doublet field on the visible brane), which causes the
scalar mixing R H
physical mass eigenstates h and to be mixtures of the original Higgs and radion fields. First, we
discuss the theoretical constraints on the allowed parameter space. Next, we give precise procedures
for computing the h and couplings given the physical eigenstate masses, mh and m , and the
new physics scales of the model. Relations among these new-physics scales are discussed and a
set of values not far above the smallest values required by precision electroweak constraints and
RunI data is chosen. A simple result for the sum of the ZZh and ZZ squared couplings relative
to the ZZhSM squared coupling is derived. We demonstrate that this sum rule in combination with
LEP/LEP2 data implies that not both the h and can be light. We present explicit results for the
still allowed region in the (mh , m ) plane that remains after imposing the appropriate LEP/LEP2
upper limits coming from the Higgs-strahlung channel. In the remaining allowed region of parameter
space, we examine numerically the couplings and branching ratios of the h and for several cases
with mh = 120 GeV and m 300 GeV. The resulting prospects for detection of the h and at the
LHC, a future LC and a collider are reviewed. For moderate | |, both the anomalous h gg
coupling and (when mh > 2m ) the non-standard decay channel h can substantially impact
h discovery. Presence of the latter is a direct signature for non-zero . We find that BR(h )
as large as 3040% is possible when | | is large. Conversely, if m > 2mh then BR( hh) is
generally large. Sensitivity to the model-dependent magnitude of the cubic radion potential term is
discussed. We find that detection of a light might require the LC. Detection of a heavy might need
E-mail addresses: dominici@fi.infn.it (D. Dominici), bohdan.grzadkowski@fuw.edu.pl (B. Grzadkowski),

jfgucd@higgs.ucdavis.edu (J.F. Gunion), toharia@physics.ucdavis.edu (M. Toharia).
doi:10.1016/j.nuclphysb.2003.08.020
244
D. Dominici et al. / Nuclear Physics B 671 (2003) 243292
to take into account the hh channel. The feasibility of experimentally measuring the anomalous
gg and couplings of the h and is examined.
PACS: 04.50.+h; 12.60.Fr
Keywords: Extra dimensions; Higgs-boson sector; RandallSundrum model
1. Introduction
The Standard Model (SM) of electroweak interactions describes successfully almost all
existing experimental data. However the model suffers from many theoretical drawbacks.
One of these is the hierarchy problem: namely, the SM cannot consistently accommodate
the weak energy scale O(1 TeV) and a much higher scale such as the Planck mass
scale O(1019 GeV). Therefore, it is commonly believed that the SM is only an effective
theory emerging as the low-energy limit of some more fundamental high-scale theory
that presumably could contain gravitational interactions. In the last few years there have
been many models proposed that involve extra dimensions. These models have received
tremendous attention since they could provide a solution to the hierarchy problem. One
of the most attractive attempts has been formulated by Randall and Sundrum [1], who
postulated a 5D universe with two 4D surfaces (3-branes). In the simplest version, all
the SM particles and forces with the exception of gravity are assumed to be confined to
one of the 3-branes called the visible brane. Gravity lives on the visible brane, on the
second brane (the hidden brane) and in the bulk. All mass scales in the 5D theory are of
the order of the Planck mass. By placing the SM fields on the visible brane, all the order
Planck mass terms are rescaled by an exponential suppression factor (the warp factor)
0 em0 b0 /2 , which reduces them down to the weak scale O(1 TeV) on the visible brane
without any severe fine tuning. A ratio of 1 TeV/MPl (where MPl is the reduced Planck
mass, MPl 2.4 1018 GeV) corresponds to m0 b0 /2 35. This is a great improvement
compared to the original problem of accommodating both the weak and the Planck scale
within a single theory.
In order to obtain a consistent solution to the Einstein equations corresponding to a
low-energy effective theory on the visible brane with a flat metric, the branes must have
equal but opposite cosmological constants and these must be precisely related to the bulk
cosmological constant. The model is defined by the 5D action:

S =
d 4 x dy

g

R
+
16G5
d x ghid (Lhid Vhid ) +

4

d 4x
gvis (Lvis Vvis ),
(1)
where g (,
= 0, 1, 2, 3, 4, where 4 refers to the y coordinate) is the bulk metric
and ghid (x) g (x, y = 0) and gvis (x) g (x, y = 1/2) (, = 0, 1, 2, 3) are the
245
3 1
induced metrics on the branes. We will use the notation 2 = 16G5 = 1/MPl5
. One
finds that if the bulk and brane cosmological constants are related by /m0 = Vhid =
Vvis = 12m0 / 2 and if periodic boundary conditions identifying (x, y) with (x, y) are
imposed, then the 5D Einstein equations lead to the following metric:
ds 2 = e2 (y) dx dx b02 dy 2 ,
(2)
where (y) = m0 b0 [y(2 (y) 1) 2(y 1/2) (y 1/2)]; b0 is a constant parameter

that is not determined by the action, Eq. (1). Gravitational fluctuations around the above
background metric will be defined through the replacements:
+ h (x, y),
b0 b0 + b(x).
(3)
Below we will be expanding in powers of h and eventually b(x)/b0 as well.

The paper is organized as follows. First, in Section 2 we describe the basic framework
for our analysis and derive the effective potential for the SM Higgs-boson sector interacting
with KaluzaKlein excitations of the graviton (hn
) and the radion (). We discuss the need
to retain a full form in order to show that the only consistent minimum of this effective
potential is the standard one. In Section 3, we introduce the curvature-scalar mixing
H
and discuss its consequences for couplings and interactions. Here, H
is the Higgs
RH
field on the visible brane before any rescalings required for canonical normalization. In
Section 4, we detail the phenomenology of the scalar sector, including the particularly
important possibility of h decays, assuming that the new physics scale is large,
= 5 TeV (where specifies the strength of the radion interactions with matter). We
consider detection of the h and at both the Large Hadron Collider (LHC) and a future
linear collider (LC), as well as in collisions at the latter. In Section 5, we discuss
the even more dramatic features that would arise if = 1 TeV, a choice that might
be excluded with additional analysis of RunI Tevatron data and/or precision electroweak
constraints. We summarize our results in Section 6. Appendix A presents a complete
tabulation of the Feynman rules we employ.
There is already an extensive literature on the scalar sector phenomenology of the
RandallSundrum model. Studies in the absence of mixing ( = 0) include Refs. [37].
Some aspects of = 0 phenomenology appear in Refs. [811]. In this paper, we focus
especially on the impacts of the tri-linear couplings that emerge only when = 0 mixing
is present.
2. The effective potential

Our first goal is to determine the effective potential that is defined as a collection
of all non-derivative contributions to the 4D effective Lagrangian density. We wish to
demonstrate that the standard vacuum defined by the stationary point of the Higgs potential
is the unique potential minimum. It turns out that this requires using a very complete form
for the full effective potential.
1 Our M
Pl5 is the same as the M of [2].
246
In order to show that standard 4D gravity is reproduced by the model, and to identify
scalar degrees of freedom related to fluctuations of b0 , let us assume temporarily that h
is only a function of x.2 Integrating the bulk Lagrangian over the 5th dimension one finds
a contribution to the effective action (see, for example, [2]3):

(1 2 )
6
Seff = d 4 x g 2 b R (4) (g) + 2 ( b ) b + ,
(4)
m0
m0
where g (x) denotes the + h part of the metric, R (4) (g) is the 4D Ricci scalar and
b (x) em0 [b0 +b(x)]/2 . Standard 4D gravity is reproduced by requiring
2
1 2
MPl
= 2 0,
2
m0
(5)
18
where 0 em0 b0 /2 is known
as the warp factor and MPl 2.4 10 GeV is the reduced
Planck mass defined as 1/ 8G4 . The canonically normalized massless radion field 0 (x)
is defined by

12 1/2
0 (x)
(6)
b (x) 6 MPl b (x).
2m
0
For the Lagrangian of Eq. (4), the radion is massless and there is no potential leading to
a definite vacuum expectation value for the radion field. This result is already apparent
at the level of the RS solution of the Einstein equations, Eq. (2), where b0 appeared as a
free parameter. Therefore, some potential, V (0 ), for the radion field is necessary [12] in
order to determine its vacuum expectation
value and in consequence stabilize the distance
between the branes: 0 = 6 MPl 0 .

on the visible brane is
The SM action for the Higgs doublet H

D H
b4 V (H
) ,
Svis d 4 x gvis (Lvis Vvis ) = d 4 x g b2 D H
(7)

where we will show that V (H ) must vanish at the potential minimum, implying V (H ) =
1 v 2 )2 , and in the first term the 4 from gvis is partially canceled by the
H
(H
b
2
H
. (In the final form of Eq. (7) and in subsequent equations
b2 from gvis in gvis H
the flat metric will be assumed whenever repeated indices are summed.)
,
Incorporating 0 into the definition of the Higgs doublet by the rescaling H0 0 H
and employing the radion field 0 of Eq. (6), we can rewrite Svis as

0 2
0 4

4
Svis = d x g
(8)
D H0 D H0
V (H0 ) ,
) implies that V (H0 ) = (H H0 1 v 2 )2 with v0 0 v.
where the above form of V (H

0
2 0
Expanding around the (presumed) vacuum expectation values for the radion, 0 +
2 In other words we consider here contributions from the massless zero KaluzaKlein mode, see Eq. (11).
3 We note that our 2 is related to the 2 of [2] by 2 = 2 2 .
247
0 , and for the Higgs-boson, H0 1 (v0 + h0 + ia0), and dropping terms involving the
2
Goldstone boson a0 , one gets the following contribution to the effective action involving
0 and h0 :4

0
0 2 1
h0 h0 6V (h0 )
T (h0 ) +
d 4 x g LSM (h0 )
2
3

0
,
+O
(9)
where LSM (h0 ) denotes the SM piece, T (h0 ) = 4V (h0 ) h0 h0 and the form of
V (h0 ) corresponding to the above V (H0 ) is

h20 2
V (h0 ) = v0 h0 +
(10)
.
2
That V (h0 ) must indeed vanish as h0 0 will be shown shortly.
From the form of the rescaled potential V (h0 ) in Eq. (10), it is clear that even if
the typical scale of the 5D theory is of the order of the reduced Planck scale, v MPl ,
then v0 = em0 b0 /2 v 1 TeV for moderate values of model parameters: m0 b0 /2 35.
Therefore, the existence of the warp factor provides a solution to the hierarchy problem, as
it explains the large ratio of MPl /1 TeV.
Keeping in mind that h (x, y) depends both on x and y, we use the KK expansion in
the extra dimension

n (y)
h (x, y) =
(11)
hn (x)
b0
n
on the visible brane (y = 1/2) to obtain

2

h h h h + O 3
g = 1 + h
2
4
2

1
1 2 n m 1 n m
n
=1+
h h h h
h
W
W
2
n
n,m
n 3
h
+O
,
W
where
(12)
2 b0
2 MPl 0 ,
(13)
n
(1/2)
where we used n (1/2) m0 b0 /0 . The full effective potential for radion plus Higgs is
constructed by using the expansion (12) to obtain the effective potential parts of Eq. (7) and
W
4 We will, of course, be dropping the derivative terms in Eq. (9) when discussing the effective potential.
248
by including a stabilizing potential for the radion parameterized by a radion mass, m0 :5

1 n
1 2 n m 1 n m
h
+
= 1+
h
h h h
W
W
2
n
n,m
brane
Veff

0 4
1 2 2
1+
V (h0 ) + m0 0 ,
(14)
where the shift to the perturbative 0 fluctuation field 0 0 + 0 = + 0 has been

performed. In particular, for the radion stabilization, it is enough to assume some non-zero
vacuum expectation value b0 and to introduce the mass term 12 m20 02 for the fluctuation
field.
Since we will later investigate the vacuum structure of the theory, it will be useful to
restrict ourself to the trace part of hn 14 h n :6
brane
Veff
1 n
1
= 1+
h +
2

W n
4W

0 4
1+
V (h0 ) +

n

hn h m +

1 2 2
m .
2 0 0
(15)
In order to derive the remaining contributions to the effective potential coming from the
gravity fluctuation h , we temporarily drop derivatives of b(x) in Eq. (1) and expand in
powers of h . The leading contribution ( 1 ) to the 5D Lagrangian density reads
[Lbranes + Lbulk]

1
y e4 (y) y h h ,
[b0 + b(x)]
(16)
where Lbulk and Lbranes denote the bulk and brane Lagrangian densities, respectively. Thus,
by proper matching of bulk and brane contributions, we obtain a total derivative, which
vanishes upon integration over y. The values of Vhid and Vvis required to get this total
derivative form are, of course,7 the same as required by the general relativity equations.
After some algebra, the O(1) term is found to be
[Lbranes + Lbulk]

2

1
e4 (y h ) y h y h .
4[b0 + b(x)]
(17)
In order to find the corresponding contribution to the 4D effective potential, one has to
expand h in KK modes and then integrate over y. The KK modes satisfy the following
5 This is the first in a possible series of terms that would in general also include 3 , 4 , . . . terms. These
0
0
terms do not influence our discussion so long as m2 > 0 and the coefficients of the higher terms are such that
0
the potential is stable in the large field strength limit.
6 4D Lorentz invariance requires that the vacuum expectation value of hn be of the form hn .
7 Linear terms in an expansion around a solution of the equations of motion should vanish since the solution
corresponds to an extremum of the action.
249
orthogonality conditions:
1/2
e2 (y)|b(x)=0 n (y) m (y) dy = mn .
(18)
1/2
To apply the orthogonality relations, we will expand Eq. (17) in powers of b(x)/b0. It is
useful to keep in mind the expression for b(x) in terms of the radion fluctuation 0 (x):

b(x)
2
0 (x)
.
=
ln 1 +
b0
m0 b 0
(19)
Since the solution of the hierarchy problem requires 2/(m0 b0 ) 1/35, we will make the
approximation of dropping powers of b(x)/b0 relative to 1. Moreover, we will later expand
in powers of 0 / 1, which provides an extra justification for neglecting b(x)/b0. With
this approximation, after utilizing the orthogonality relations we find the final result for the
KK-graviton mass term8 in the effective Lagrangian density:9
KK
Veff
=

1 2 n n
n
m h h
hn
h ,
4 n n
(20)
where the KK-graviton masses are given by mn = m0 xn 0 , with xn denoting the zeroes
of the Bessel function J1 (x) and 0 em0 b0 /2 . Keeping in mind that the vacuum
expectation value should satisfy hn 14 h n we get
KK
Veff
=
3 2 n 2
m h .
16 n n
(21)
8 Without the expansion in powers of b(x)/b , interpretation of Eq. (17) in terms of a simple mass term for the
0
gravitons would be much more difficult. The hn s could not be interpreted as the physical gravitons obeying the
standard equations of motion for spin 2 particles. In order to recover the canonical graviton degrees of freedom
one would have to redefine hn by a b(x) field-dependent factor. However, since in our case it is legitimate to
expand in powers of b(x)/b0 , and keep only the very first constant term, we will not discuss this issue further
here.
9 If we had used the parameterization of the metric proposed in [9,13]:
2

2 (y) b(x)
( + h ) dx dx b02 1 + 2 e2 (y) b(x) dy 2 ,
ds 2 = e2 (y)2 e
we would not need to expand in powers of b(x)/b0 in order to derive the interaction quadratic in the graviton
field. However, as we have checked, the form of the graviton mass terms is the same in both approaches, therefore
we adopt here the straightforward definition given by Eqs. (2), (3).
250
That completes the determination of the total 4D effective potential:

brane
KK
+ Veff
Veff = Veff

1 n
1 n m
h +
h h +
= 1+
W
2
4
W n m
n

0 4
1
1+
V (h0 ) + m20 02
2
3 2 n 2
m h + ,
16 n n
(22)
where the dots refer to terms of the order of O[( h )3 ]. Restricting

ourself to the
W 1
perturbative regime we will look for the minimum of Veff that satisfies n h n /
and b(x)/b0 1, the latter being equivalent to 0 (x)/ 1. Keeping all the terms10
shown explicitly in Eq. (22) the extremum conditions are as follows:
Veff
=0
h n

1
1 n
1 2 2
0 4
3
h
+
V (h0 ) + m0 0 m2n h n = 0,
1+
2
W
2
8
2W n
(23)
Veff
=0
0

1 n
0 3 V (h0 )
1 n m
2
1+
4 1+
+ m0 0 = 0,
h +
h h
W
2
4
W n m
n
(24)
Veff
=0
h

1 n
0 4 V (h0 )
1 n m
1+
1
+
= 0.
h +
h
h
W
2
h0
4
W n m
n
(25)
W 1:
There is only one solution of Eq. (25) consistent with 0 / 1 and h n /
V (h0 )
namely, h0 |h0 =0 = 0. For consistency of the RS model we must also require that
V (h0 ) = 0. If V (h0 ) = 0, then the visible brane tension would be shifted away from
the very finely tuned RS solution to the Einstein equations. With these two ingredients,
Eq. (24) requires that 0 = 0 at the minimum, implying that we have chosen the correct
expansion point for 0 , and Eq. (23) then leads to h n = 0, i.e., we have expanded about
the correct point in the h n fields. However, it is only if m20 > 0 that 0 = 0 is required
by the minimization conditions. If m0 = 0, then Eq. (23) still requires h n = 0 but all
equations are satisfied for any 0 .
10 Since the KK-graviton mass term originates from contributions of the order of 1/
2 , for consistency we
W
keep the same approximation while expanding g in Eq. (22).
251
n
0
We note that if one were to use the form Veff = (1 + 1

n h + 4 )V (h0 ) +
W
n 1 2 2
3
2 n 2
n
(1 + 1
n h ) 2 m0 0 16
n mn (h ) , then the linear term in h could be used to
W
compensate the linear term in 0 to obtain an extremum that is apparently deeper than the
standard minimum, which minimum turns out to be tachyonic and, therefore, unphysical.
The full form with the positive definite (1 + 0 / )4 factor makes such a deeper extremum
impossible.
Finally, we note that since V (h0 )/h0 = 0 at the minimum (even after including
interactions with the radion and KK gravitons) there are no terms in the potential that
are linear in the Higgs field h0 as shown in Eq. (10). We will return to this observation in
the next section of the paper.
3. The curvature-scalar mixing

Having determined the vacuum structure of the model, we are in a position to discuss
the possibility of mixing between gravity and the electroweak sector. The simplest example
of the mixing is described by the following action [14]:
H
,
S = d 4 x gvis R(gvis )H
(26)
where R(gvis ) is the Ricci scalar for the metric induced on the visible brane, gvis =
is the Higgs field in the 5D context before
b2 (x)( + h ), and we recall that H
and b (x) =
rescaling to canonical normalization on the brane. Using H0 = 0 H
0 (x) as before, one obtains [9]

H
= 6 (x) (x) + h (x) + H H0 .
gvis R(gvis )H
(27)
0
To isolate the kinetic energy terms we again use the expansions
1
H0 = (v0 + h0 ),
2
(x) = 1 +
0
.
(28)
The h term of Eq. (27) does not contribute to the kinetic energy since a partial integration
would lead to h = h = 0 by virtue of the gauge choice, h = 0.
We thus find the following kinetic energy terms:
L=

1
1
1
1 + 6 2 0 0 0 m20 0 h0 + m2h0 h0 6 0 h0 ,
2
2
2
(29)
where
v0 / .
(30)
In the above,
m2h0 = 2v02 ,
(31)
and m20 are the Higgs and radion masses before mixing. Eq. (29) differs from Ref. [8] by
the extra 0 0 piece proportional to .
252
We define the mixing angle by

tan 2 12 Z
m2h0
m20 m2h0 (Z 2 36 2 2 )
(32)
where
Z 2 1 + 6 2 (1 6 ) 36 2 2 .
(33)
In terms of these quantities, the states that diagonalize the kinetic energy and have
canonical normalization are h and with:

6
6
sin h + sin +
cos dh + c,
h0 = cos
(34)
Z
Z
h
0 = cos + sin a + bh.

(35)
Z
Z
(Our sign convention for 0 is opposite that chosen for r in Ref. [9].) To maintain positive
definite kinetic energy terms for the h and , we must have Z 2 > 0. (Note that this
implies that > 0, see Eq. (33), is implicitly required.) The corresponding mass-squared
eigenvalues are11
2
1/2

1 2
m + m2h0 m20 + m2h0 4Z 2 m20 m2h0
.
(36)
2Z 2 0
We will identify the larger of [mh , m ] with m+ . This equation can be inverted to obtain

Z2 2

2
2
4m2+ m2 1/2
2
2
2 2
m+ + m m+ + m
.
mh0 , m0 =
(37)
2
Z2
m2 =
Using the symmetry of the inversion under m2+ m2 , we could equally well write Eq. (37)
using m2h and m2 . Note that for the quantity inside the square root appearing in Eq. (37) to
be positive, we require that:12

m2+
2
Z2
Z 2 1/2
2
+ 2 1
>1+ 2 1
,
(38)
Z
Z
m2
where 1 Z 2 / = 36 2 2 / > 0. In other words, since we will identify m+ with either
mh or m , the physical states h and cannot be too close to being degenerate in mass,
depending on the precise values of and ; extreme degeneracy is allowed only for small
and/or . We also note that

m2h0 + m20 = Z 2 m2+ + m2 ,
(39)
m2h0 m20 = Z 2 m2+ m2 .
This leaves a two-fold ambiguity in solving for m2h0 and m20 , corresponding to which
we take to be the larger. We resolve this ambiguity by requiring that m2h0 m2h in the
11 Note that the quantity inside the square root is positive definite so long as m2 m2 > 0.
h0 0
12 Since m > m by definition, the second solution for the positivity condition is irrelevant.
+
253
0 limit. This means that for m2h0 we take the + () sign in Eq. (37) for mh > m
(mh < m ), i.e., for mh = m+ (mh = m ), respectively.
Given this choice, we complete the inversion by writing out the kinetic energy of
Eq. (29) using the substitutions of Eqs. (34) and (35) and demanding that the coefficients
of 12 h2 and 12 2 agree with the given input values for m2h and m2 . By using Eqs. (39),
it is easy to show that these requirements are equivalent and imply
sin 2 =
12 m2h0
Z(m2 m2h )
(40)
Note that the sign of sin 2 depends upon whether m2h > m2 or vice versa. It is convenient
to rewrite the result for tan 2 of Eq. (32) using Eq. (39) in the form
tan 2 =
12 m2h0
Z(m2 + m2h 2m2h0 )
(41)
In combination, Eqs. (40) and (41) are used to determine cos 2 . Together, sin 2 and
cos 2 give a unique solution for . As a useful point of reference, we note that m2 = 0
corresponds to m20 = 0, m2h0 = Z 2 m2h , sin 2 = 12 Z/, cos 2 = ( 2Z 2 )/,
sin = 6 / , and cos = Z/ .

Using this inversion, for given , , mh and m we compute Z 2 from Eq. (33), m2h0 and
m20 from Eq. (37), and then from Eq. (32). With this input, we can then obtain a, b, c, d
as defined in Eqs. (34) and (35).
Altogether, when = 0 there are four independent13 parameters that must be specified
to completely fix the state mixing parameters a, b, c, d of Eqs. (34) and (35) defining the
mass eigenstates. These are
, , mh , m ,
(42)
where we recall that v0 / with v0 = 246 GeV. Two additional parameters are
required to completely fix the phenomenology of the scalar sector, including all possible
decays. These are
W , m1 ,
(43)
W will determine KK-graviton couplings to the h and and m1 is the mass of the
where
W is fixed in terms of while m1 depends
first KK graviton excitation. The parameter
upon and the curvature parameter, m0 /MPl . We summarize the relations among all
these parameters as given by our earlier formulae:
2 b0
W
2 MPl 0 ,
n
(1/2)
mn = m0 x n 0 ,
W ,
= 6 MPl 0 = 3
(44)
13 Aside from the constraints that derive from requiring that Z 2 > 0 and the constraint of Eq. (38).
254
where 0 MPl = em0 b0 /2 MPl should be of order a TeV to solve the hierarchy problem. In
Eq. (44), the xn are the zeroes of the Bessel function J1 (x1 3.8, x2 7.0). A useful
relation following from the above equations is
m1 = x 1
m0
.
MPl 6
(45)
To set the scale of m0 independently of b0 requires additional argument. One line of

reasoning is that of Ref. [4]. There it is argued that the 3-brane tension, |Vvis| = 12m2 0 with
2m
2
2MPl
, see Eq. (5), should be roughly the same as the tension, 3 , of a D 3-brane
M4
s
in the heterotic string theory: 3 = g(2)
3 , where g 1 is the string coupling constant and
the string scale is Ms gYM MPl . Setting |Vvis| = 3 gives
g2
m0
(46)
YM
0.013,
MPl
6 (2)3/2
using gYM 0.7. Although this precise value should probably not be taken too seriously,
a reasonable range to consider is 0.01 m0 /MPl 0.1. This guarantees that the ratio of
the bulk curvature m0 to MPl5 , m0 /MPl5 [m0 /MPl ]2/3 , is small, as required for reliability
of the RandallSundrum approach.
In choosing parameters for a more detailed phenomenological study of the scalar sector,
we must be careful to avoid current bounds deriving from RunI Tevatron data and from
precision electroweak constraints. These have been examined in Ref. [6]see their Fig. 22.
The smallest possible value for m1 for which it is clear that KK excitation corrections
to precision electroweak observables are not in conflict with existing bounds while at
the same time all RunI bounds on KK excitations are satisfied is m1 = 450 GeV, for
which m0 /MPl 0.05 is required for simultaneous consistency. Inserting these values
into Eq. (45) gives 5.8 TeV. At higher m0 /MPl , the naive RunI Tevatron restriction
becomes much stronger than the precision electroweak constraint. Thus, for example, at
m0 /MPl 0.1 we employ the RunI Tevatron constraint of m1 600 GeV from Fig. 22
of [6] to obtain 4 TeV. In our detailed study, we will employ m0 /MPl = 0.1 and
= 5 TeV, corresponding [see Eq. (45)] to m1 = 750 GeV. We note that this large mass
for the first KK excitation means that light (mass 300 GeV) Higgs bosons and radions
cannot decay into KK excitations. The full phenomenology of this scenario is explored in
Section 4.
Let us consider further the implications of our choice of = 5 TeV. From Eq. (44),
2000
15. This value
this choice gives MPl 0 2.04 TeV and thence 0 2.410
18 0.85 10
W 3 TeV. For
is equivalent to m0 b0 69. Again using Eq. (44), = 5 TeV implies
mh and m we will consider a range of possibilities, but with some prejudice towards m <
MPl
, with
mh . Indeed, in Ref. [9] (see also [15]) it is argued that m0 (backreaction) 035
(backreaction) < 1 needed for consistency of their expansion. Inserting 0 MPl 2 TeV,
as estimated above for = 5 TeV, this would correspond to m0 < 57 GeV. In Ref. [12],
it is argued that m0 TeV where 1 makes the radion stabilization model most
natural. This would again suggest the possibility of quite a light radion. In fact, we shall
find that the case of a light radion eigenstate (which, even after mixing, still roughly
corresponds to small m0 ) presents a particularly rich phenomenology.
255
Although large > 45 TeV is guaranteed to avoid conflict with all existing
constraints from LEP/LEP2 and RunI Tevatron data, it is by no means certain that such
a large value is required. For example, if = 1 TeV,
m1 =
m0
1.55 TeV
MPl
(47)
ranges from 75 GeV to 1.55 TeV as m0 /MPl ranges from 0.05 to 1. For this case,
if we take m0 /MPl to be of order 1, then m1 1.55 TeV and there are no precision
electroweak or RunI constraints. In fact, even RunII would not probe this scenario (see
Fig. 13 of [6]). Of course, m0 /MPl > 0.1 implies large 5-dimensional curvature, implying
that corrections to the naive RS solution might be large. Nonetheless, in Section 5 we shall
present results for = 1 TeV first assuming that m1 is large. However, for = 1 TeV
it is also very interesting to consider small m0 /MPl and, hence, small m1 . As suggested in
Ref. [6], the h1 and subsequent resonances are very narrow for small m1 and might have
been missed at the Tevatron. Further, it is not clear that precision electroweak data rules out
this kind of scenario. In principle, one should perform an analysis of precision electroweak
constraints simultaneously taking into account the KK excitation effects and the radion and
Higgs contributions. Compensation between these two classes of effects might be possible.
Such an analysis is beyond the scope of this paper. However, we find it useful to entertain
several such scenarios at = 1 TeV in order to explore the possible importance of Higgs
decays to KK excitations. Thus, at the very end of Section 5 we will consider the values
m0 /MPl = 0.065 and 0.195 corresponding to m1 = 100 GeV and 300 GeV, respectively.
Referring to Fig. 22 of [6], we see that both choices are well within the RunI Tevatron
nominally excluded area, but would correspond to such narrow KK spikes that they might
have been missed. The first choice also leads to S and T electroweak observable corrections
that are too large on their own and would have to be compensated by other contributions.
The second choice leads to KK excitation corrections to S and T that are small enough to
be acceptable.
We now turn to the important interactions of the h, and hn . We begin with the ZZ
couplings of the h and . The h0 has standard ZZ coupling while the 0 has ZZ coupling
deriving from the interaction 0 T using the covariant derivative portions of T (h0 ).
After rewriting these interactions in terms of the mass eigenstates, the portion of the
ZZ couplings is given by
gmZ
(d + b)
cW
gmZ
(c + a)
gZZ =
cW
gZZh =
gmZ
gZZh ,
cW
gmZ
gZZ ,
cW
(48)
where g and cW denote the SU (2) gauge coupling and cosine of the Weinberg angle,
respectively, and we have adopted a notation in which the gs without the bar denote the
reduced coupling strength relative to SM strength. The W W couplings are obtained by
replacing gmZ /cW by gmW . As noted in [9], there are additional contributions to the ZZh
and ZZ couplings coming from 0 T for the gauge fixing portions of T . These
terms vanish when contracted with on-shell W or Z polarizations, which is the physical
situation we are interested in. In addition, these extra couplings vanish in the unitary gauge.
256
Thus, we do not write these additional terms explicitly. Notice also an absence of Zh tree
level couplings.
Next, we consider the fermionic couplings of the h and . The h0 has standard fermionic
couplings and the fermionic couplings of the 0 derive from 0 T using the Yukawa
interaction contributions to T . One obtains results in close analogy to the V V couplings

just considered:
gf fh =
gmf
gZZh ,
2mW
g f f =
gmf
gZZ ,
2mW
(49)
i.e., the f f couplings are related to the SM couplings by the same factors as are the
V V couplings. These results for the V V and f f couplings are summarized in Fig. 33
of Appendix A.
For small values of , the reduced couplings gZZh and gZZ have the expansions:

gZZ = 1 +

gZZh = 1 + O 2 ,
6 m2
m2h
m2

+O 3 .
(50)
We note that if 1 6 > 0 (i.e., for smaller than the conformal limit of = 1/6), then
it is always possible to choose parameters so that the decouples from f f and V V :
c + a = 0. This is achieved by taking

1
1
1
1
m2h0 ,
m2h0 ,
m2 = min 2 ,
m2+ = max 2 ,
(51)
Z 1 6
Z 1 6
which corresponds to m20 = m2h0 /(1 6 ).
The following simple and exact sum rules (independently noted in [10]) follow from the
definitions of a, b, c, d:

2
2
gf2 fh + g f2 f
gZZh
+ gZZ
2 (1 6 )2
2
2
=
=
g
+
g
=
1
+
R2 .
gmZ 2
gmf 2
ZZh
ZZ
Z2
cW
(52)
2mW
Note that R 2 > 1 is a result of the non-orthogonality of the relations Eqs. (34) and (35).
Of course, R 2 = 1 in the conformal limit, = 1/6. It is important to note that Z 0
would lead to divergent ZZ and f f couplings for the . As noted earlier, this was to be
anticipated since Z 0 corresponds to vanishing of the radion kinetic term before going
to canonical normalization. After the rescaling that guarantees the canonical normalization,
if Z 0 the radion coupling constants blow up: gZZ (c + a) 1/(6 Z) + O(Z).
To have Z 2 > 0, must lie in the region:

4
4
1
1
(53)
1 1+ 2
1+ 1+ 2 .
12
12
As an example, for = 5 TeV, Z 2 > 0 corresponds to the range 3.31 3.47. Of

course, if we choose sufficiently close to the limits, Z 2 0 implies that the couplings, as
characterized by R 2 will become very large. For example, for = 5 TeV, R 2 in Eq. (52)
takes the values 2.48 and 1.96 at = 2.5 and = 2.5, respectively. Thus, it might be
257
appropriate to impose bounds on that keep R 2 moderate in size. However, we will not
do so in our plots. In fact, a bound such as R < 5 results in essentially invisible changes to
our plots, since it is almost always superseded by the constraint of Eq. (38).
Also of considerable phenomenological importance are the h and couplings to gg
and . As shown in [9], these have anomalous contributions in addition to the usual oneloop contributions. (The latter must be computed after rescaling the f f and V V couplings
by (d + b) for the h and (c + a) for the .) These anomalous contributions can very
significantly enhance the gg coupling in particular. The Feynman rules for these vertices
appear in Appendix A and some of their phenomenological implications will be discussed
in the next section.
The final crucial ingredient for the phenomenology that we shall consider is the trilinear interactions among the h and and hn fields. In particular, these are crucial for the
decays of these three types of particles. The tri-linear interactions derive from five basic
sources.
(1) First, we have the cubic interactions coming from
2

1
1
L V (H0 ) = H0 H0 v02 = v02 h20 + v0 h30 + h40 ,
2
4
after substituting H0 =
1 (v0
2
(54)
+ h0 ). Here, the first term above implies that is related to
the bare Higgs mass as in Eq. (31). The h30 interaction can then be expressed as
L
m2h0
(55)
h3 .
2v0 0
(2) Second, there is the interaction of the radion 0 with the stress-energymomentum
tensor trace:

0
0
L
(56)
T (h0 ) =
h0 h0 + 4v02 h20 .
(3) Thirdly, we have the interaction of the KK-gravitons with the contribution to the
stress-energymomentum tensor coming from the h0 field:
1 n
h h0 h0 ,
L h T
(57)

2
W n
where we have kept only the derivative contributions and we have dropped (using the gauge
n
h = 0) the parts of T .
(4) Finally, we have the -dependent tri-linear components of Eq. (27):

2
v0
h0 0 6 2 h0 0 0
6 (x) (x) + h (x) H0 H0 3

v02 n
v0 n
12
h 0 h0 6
h 0 0 ,
(58)
W
W 2
n
n
n
where we have employed hn = 0, used the traceless gauge condition h = 0, and also
used the symmetry of h .
258
(5) Finally, there is the model-dependent 03 interaction strength. We parameterize this

interaction using the form
m2
1
V (0 ) = m20 02 + X3 0 03 + ,
2
2
(59)
where the represents 04 and higher power interactions that do not give rise to cubic
vertices. In general, because of the presence of such terms there is no constraint on the
sign of X3 . We have examined the prediction for X3 in the context of the GoldbergerWise
stabilization mechanism [12] by carrying their procedure out to the 03 level. The result is
of the form given (i.e., having the m20 / structure) with X3 = +3. Other stabilization
mechanisms may produce somewhat different results, but we believe that this is the correct
order of magnitude for such a term. We will present results for X3 = 0 and then examine
sensitivity to X3 by comparing to X3 = 3.
We discuss briefly why several other kinds of tri-linear interactions are absent. First, there
are no hn h0 h0 tri-linear vertices other than that appearing in Eq. (57). Other possible
sources are zero in the gauge we employ. In particular, consider the hn h0 h0 interactions that
arise in Eqs. (8) [after expanding g as in Eq. (14)] from the kinetic energy derivative
terms and from expanding V (H0 ) about the minimum as in Eq. (54). The Lorentz structure
n
n
of these (and other such tri-linear terms) can only be of the form h h20 or h h0 h0 ,
n
both of which are absent in the h = 0 gauge.
There is another generic class of tri-linear interaction term that can arise, involving two
hn s and one . For example, such an interaction arises if we retain the b(x)/b0 term in the
expansion of Eq. (17) [(b0 + b(x))1 b01 (1 b(x)/b0 + )] and use [see Eq. (19)],
b(x)
2 0 (x)
b0 m0 b0 . The resulting contribution to the Lagrangian takes the form

1 1 2 2 n n
n
mn h h
hn
h 0 (x).
4 m0 b0 n
(60)
Using our earlier numerical estimates, the effective coupling for this interaction is of order:
m21
(0.75 TeV)2
0.8 GeV.
2 m0 b0 10 TeV 69
(61)
Keeping this interaction small is a natural result of having a small value of m0 /MPl .
There are actually many other sources of hn hn interactions that could be retained by
a more exact treatment of the various Lagrangian contributions. As another example, in
the reduction to Eq. (4), one approximates g by in obtaining the second term. If one
instead inserts the full expansion of g5 in terms of the h (x, y) fields out to order 2 , and
uses the eigenexpansion of Eq. (11), hn hn 0 interactions are generated with a coefficient
magnitude similar in size to that estimated above. Note that there has been some discussion
of the possible nature of the hn hn coupling in Ref. [16], where it is stated that it can only
appear at one loop. The coupling generated in the ways mentioned above does not conform
to their assumptions. In any case, we saw earlier that the hn KK excitations must be very
massive for choices of m0 /MPl and 45 TeV that clearly satisfy the combined
259
constraints from RunI Tevatron data and precision electroweak constraints. Even for the
= 1 TeV choice discussed earlier, which requires relaxing the naive RunI and precision
electroweak constraints, an m1 value below 100 GeV would be highly improbable. As a
result, h hn hn or hn hn decays (that would be induced by the above interactions
after rotating to the mass eigenstates, and h) are not relevant for the modest mh and m
values explored in the bulk of this paper. Thus, we have not worked out a full expression
for this vertex.
To proceed with the tri-linear interactions enumerated earlier, we substitute for h0 and
0 in terms of the physical h and states using Eqs. (34) and (35), respectively. The results
for the tri-linear vertices generated, after this substitution into the enumerated interactions,
appear in Fig. 34 in Appendix A. Note that the Feynman rules generated are specified in
part by terms containing the parameters m2h0 and m20 ; m2h0 and m20 must be computed
from m2h and m2 using the inversion procedure given earlier. Since the effective potential
shown in Eq. (22) does not contain any interactions linear in the Higgs field, vertices like
2 h and hn h are a clear indication for the curvature-Higgs mixing. As we shall see, they
could also be of considerable phenomenological importance. It is also useful to note that
since the -mixing angle for 1 is proportional to , Eq. (32), the interaction terms,
W and as a result
Eqs. (57) and (58), are suppressed by at least one power of 1/ or 1/
W ). A useful reference is the
the related couplings will be of the order of 1/2 or 1/(
small limits of the couplings. For instance, if we take the k 2 s for the external particles
to be on-shell (i.e., equal to m2h or m2 depending on the external line) the two couplings
that vanish linearly as 0 have the limits:

2

gnh W = 12 3 +
(62)
+O 3 ,
x

3 1 8 12 X3
9
3
2
2
gh = 12 mh x(1 6 ) + 4 21 X3
2
x
x
2
(63)
+O ,
where we have employed the results for gnh and gh given in Fig. 34 of Appendix A and
defined x 1 m2 /m2h .
4. Phenomenology for = 5 TeV
We begin by discussing the restrictions on the h, sector imposed by LEP Higgs-boson
searches. LEP/LEP2 provides an upper limit for the coupling of a ZZ pair to a scalar (s) as
a function of the scalar mass. Because the decays of the h and can be strongly influenced
by the mixing, it is necessary to consider limits that are obtained both with and without
making use of b tagging. The most recent paper on the flavor-blind limits obtained
without b tagging is Ref. [17].14 Next, there is a preliminary OPAL note [19] in which
14 There is a much earlier paper [18] which claims much stronger limits at low scalar masses 20 GeV in the
case where the scalar decays to any of a certain class of modes. In particular, [18] gives increasingly strong limits
260
decay-mode-independent limits on the ZZs coupling are obtained that are considerably
stronger than those of [17], but not as strong as those of [18]. For scalar masses above
60 GeV, the flavor-blind limits of the above references are superseded by the results found
on the LEPHIGGS working group homepage [20], which extend up to ms 113 GeV. We
have chosen to employ [17] for ms < 60 GeV and [20] for 60 GeV ms 113 GeV.
Including the stronger limits of [18] and/or [19] would have no impact on the plots
2
presented. Next, we have the limits on gZZs
obtained using b tagging and assuming that
BR(s b b) = BR(hSM b b). The best limits that we have found are those contained
in [17] for ms < 60 GeV and in [21] for 60 ms 115 GeV. In implementing these
or BR( b b)
compared to
limits, we correct for the difference between BR(h bb)
BR(hSM b b) computed assuming mhSM = mh or mhSM = m , respectively, and using

= 5 TeV.
The first question that arises is whether both the and the h could be light without
either having been detected at LEP and LEP2. The sum rule of Eq. (52) implies that
this is impossible since the couplings of the h and to ZZ cannot both be suppressed.
For any given value of mh and m , the range of is limited by: (a) the constraint of
Eq. (38) limiting according to the degree of mh m degeneracy; (b) the constraint that
2
2
Z 2 > 0, Eq. (33); and (c) the requirements that gZZh
and gZZ
both lie below any relevant
LEP/LEP2 limit. The regions in the (, m ) plane consistent with the first two constraints
are shown in Figs. 1 and 2 for mh = 112 GeV and mh = 120 GeV, respectively, assuming
a value of = 5 TeV. For the most part, it is the degeneracy constraint (a) that defines
the theoretically acceptable regions shown. The regions within the theoretically acceptable
regions that are excluded by the LEP/LEP2 limits are shown by the light (yellow) shaded
regions, while the allowed regions are the darker (blue) shade. For mh = 112 GeV, the
LEP/LEP2 limits exclude a large portion of the theoretically consistent parameter space.
For mh = 110 GeV (not plotted), the sum rule of Eq. (52) results in all of the theoretically
allowed parameter space being excluded by LEP/LEP2 constraints. For mh = 120 GeV,
the LEP/LEP2 limits do not apply to the h and it is only for m 115 GeV and significant
2
gZZ
(requiring large | |) that some points are ruled out by the LEP/LEP2 constraints.
As a result, the allowed region is dramatically larger than for mh = 112 GeV. The precise
regions shown are somewhat sensitive to the choice, but the overall picture is always
similar to that presented here for = 5 TeV. This is illustrated in Section 5, where the
allowed regions for mh = 120 GeV are shown in the case of = 1 TeV.
Next, we discuss the couplings of the h and . We begin with their f f and V V
couplings-squared. These are illustrated in Figs. 3 and 4. There we consider mh = 120 GeV
and = 5 TeV (for which the allowed region was plotted in Fig. 2) and plot (in the upper
2
2
figures) contours of gZZh
(d + b)2 and gZZ
(c + a)2 . As in Eqs. (48) and (49),
2
on gZZ
as m decreases below 8 GeV, the 95% CL limits being 0.005 at m 0 and 0.02 at m 10 GeV
(using the curve in which the scalar is assumed to decay to the final states to which a SM Higgs boson would
decay, but not necessarily with the same branching ratios). The caveat is that gg decays are dominant in this
region and it is unclear whether or not the limits of Ref. [18] apply. In particular, the gg final state might have a
higher multiplicity of pions at modest m than allowed for in the analysis. For this reason, we do not employ the
results of [18]. Even if employed, they do not result in any additional excluded parameter regions in the case of
= 5 TeV.
261
Fig. 1. Allowed regions (see text) in (, m ) parameter space for = 5 TeV and mh = 112 GeV. The dark
portion of parameter space is theoretically disallowed. The light portion is eliminated by LEP/LEP2 constraints
2
2 BR(s bb),
with s = h or s = .
or on gZZs
on the ZZs coupling-squared gZZs
Fig. 2. As in Fig. 1 but for mh = 120 GeV.
these quantities specify the reduced couplings squared of the h and , respectively, to f f
and V V with respect to the squared coupling strength of the SM Higgs boson. The lower
figures show the variation of these couplings with at fixed m = 20, 55 and 200 GeV.
Large enhancements of (d + b)2 are possible for small m < mh as are large suppressions
when m > mh . For the , (c + a)2 is smaller than 1 except for the largest | | values at
262
2
Fig. 3. For mh = 120 GeV and = 5 TeV, we plot the quantity gZZh
= (d + b)2 which specifies the ratio of
the hs f f and V V couplings squared to the corresponding values for the SM Higgs boson, taking mhSM = mh .
In the upper figure we show contours; line colors/textures drawn actually on the boundary should be ignored. The
lower figure presents results for m = 20, 55 and 200 GeV.
high m . Indeed, (c + a)2 1 is the norm in the m < mh portion of parameter space
and for small | | when m > mh . In particular, there is a line along which (c + a)2 = 0
between the paired contour lines corresponding to (c + a)2 = 0.001; these zeroes are also
apparent from the lower plot of Fig. 4.
A coupling of particular interest in testing the nature of electroweak symmetry breaking
is the h3 self coupling. The algebraic form of the h3 coupling appears in Appendix A. The
SM limit for this coupling corresponds to d = a = 1, c = b = 0. In addition, one employs
1 / = 1/v0 . In Fig. 5, we plot the ratio ghhh ghhh /ghSM hSM hSM of the h3 coupling
relative to the corresponding SM value computed for a SM Higgs boson mass equal to
mh . We see that there are typically rather substantial deviations that one could easily probe
at a linear collider. For these plots we have taken X3 = 0 (i.e., no 03 term in the radion
263
Fig. 4. As in Fig. 3, except for the . Note the zeroes in the middle of the allowed ranges.
potential). If we adopt the GoldbergerWise value, X3 = +3, there is no visible difference

as compared to the graphs presented. (Recall that, in our notation, a bar indicates the full
coupling as opposed to the value relative to the SM, which ratio is indicated by a g without
a bar.)
It is also interesting to examine the 3 self coupling. Taking mhSM = m , we plot in
the upper figure of Fig. 6 contours of g g /ghSM hSM hSM ; in the lower figure, we
plot |g |. For these plots we take X3 = 0. In this case, g vanishes for = 0. It is
often negative (i.e., g has the opposite sign compared to ghSM hSM hSM ) and is generally
< 1, implying suppression relative to the SM case. Only for m = 200 GeV and the largest
allowed | | values can the 3 coupling take values comparable to the SM strength. Thus,
if X3 = 0 measurement of g tends to be quite difficult. Of course, the background
diagrams contributing to the same final state (e+ e t t or e+ e Z) will
also be suppressed in comparison to the SM case; they have two f f or V V vertices
264
Fig. 5. For mh = 120 GeV, = 5 TeV and X3 = 0, we plot the ratio, ghhh = ghhh /g hSM hSM hSM , of the h3
self coupling to the SM prediction for the h3SM coupling for mhSM = mh . The first plot gives contours, while the
2nd plot shows results at the fixed values of m = 20, 55 and 200 GeV.
proportional to (c + a)2 , and (c + a)2 is substantially smaller than 1 for much of the
parameter space being considered. The situation is quite different if X3 is significant. For
example, in Fig. 7 we plot g for X3 = +3. Variation as a function of is much less
dramatic and values of order 0.10.2 are typical. For such values, measurement of g will
be difficult but not impossible. Determination of g would ultimately be very important
as a means of probing the radion potential and stabilization mechanism.
A particularly important feature of the above plots is that once mh is large enough
(mh 115 GeV is sufficient) there is a substantial range of values for any m < mh /2
(so that h decays are possible) that cannot be excluded by LEP/LEP2 constraints.
The reverse is also true; allowed parameter regions exist for which hh decays are
possible once m 230 GeV. We now turn to a discussion of branching ratios, including
the h final mode.
265
Fig. 6. For mh = 120 GeV, = 5 TeV and X3 = 0, we plot the ratio, g = g /g hSM hSM hSM , of the 3
self coupling to the SM prediction for the h3SM coupling taking mhSM = m . The first plot gives contours, while
the 2nd plot gives results for |g | at m = 20, 55 and 200 GeV. In the latter plot, g is: < 0 for all at
m = 20 GeV; > 0 for < 1.87 and < 0 otherwise at m = 55 GeV; and < 0 for < 0.2 and > 0 for > 0.2
at m = 200 GeV.
The partial width of the h in which we are most interested is that for h :
(h ) =
2
g h
32mh
(1 4r )1/2 .
(64)
We also give the expression for h hn :

2

g nh
5/2 (1, r , rn )
m3h
h hn =
,
192
rn2
(65)
where rn = m2n /m2h . In the above equation, (1, r1 , r2 ) 1 + r12 + r22 2r1 2r2 2r1 r2 ,
r = m2 /m2h . (Corresponding results apply for hh and hn h.) Expanding in
powers of = v0 / using Eq. (62), we find that (h hn ) m3h /rn2 m7h .
266
Fig. 7. As for the 2nd plot of Fig. 6 except that we take X3 = +3.
It is interesting to investigate Higgs-boson branching ratios for various decay channels

in the presence of the -mixing. If we neglected the gg and anomalous couplings, we
would have

(h all) = (d + b)2SM (h all) + h hn + (h ),
(66)
where SM (h all) is the SM total width. However, the gg width can be quite enhanced
and this must be included. We have done this in the context of a modified version
of HDECAY [22], which includes all relevant radiative corrections to couplings and
branching ratios. In particular, the running b mass decreases the bb branching ratio of the
h, resulting in some increase in BR(h ). We emphasize that the phenomenology of
the h and is only influenced by the model dependent 03 cubic coupling, parameterized
in terms of X3 , through the (h ) and ( hh) partial widths, respectively.
However, these influence only the branching ratios for the h and decaysthe production
rates for the h and depend only on their couplings to SM particles, which are independent
of the cubic couplings and, hence, independent of X3 .
gg, W W , ZZ and as a function
In Fig. 8, we plot the branching ratios for h bb,
of the mixing parameter , taking mh = 120 GeV, = 5 TeV and X3 = 0. Results are
shown for three different m values: 20, 55 and 200 GeV. (The case of m = 55 GeV is one
for which BR(h ) can be significant when mh = 120 GeV.) These plots are limited
to values allowed by the theoretical constraints discussed earlier. We have chosen to not
2
as might be appropriate in order to guarantee
include a more restrictive bound on gZZh
that the h contributions to precision electroweak observables be consistent with S and T
remaining within the usual 95% CL ellipse.15
The most important features of Fig. 8 are the following. First, large values for the gg
branching ratio (due to the anomalous contribution to the hgg coupling) are the norm.
15 Contributions of the to S and T are small. Indeed, referring to Fig. 4, we see that g 2
ZZ < 1 almost
everywhere when = 5 TeV.
267
gg, W W , ZZ and for mh = 120 GeV, = 5 TeV and

Fig. 8. The branching ratios for h decays to bb,
X3 = 0 as functions of for m = 20, 55 and 200 GeV.
268
This suppresses the other branching ratios to some extent. (The anomalous contribution to
the h coupling is less important due to presence of the large W loop contribution in
this latter case.) Second, for m = 55 GeV, BR(h ) is significant at large | | and
suppresses the conventional branching ratios. In general, changes in the branching ratio of
the h with respect to the SM are modest, but nonetheless they are at an observable level, at
least at the LC. Note, however, that the modest BR changes belie the fact that the f f and
V V coupling-squared factor (c + d)2 is often changing dramatically, implying dramatic
changes in h production rates with respect to expectations for a SM hSM .
Results for the branching ratios are plotted in Fig. 9. We observe that the gg decay
is generally dominant over the bb mode and that it has the largest branching ratio until
the W W () , ZZ () modes increase in importance at larger m . Of course, the zero in
2
gZZ
= (c + a)2 has a very large impact.
The above results for the h branching ratios are little modified for X3 = 0 so long as
|X3 | 3. This is because the h branching ratio remains quite small (as we shall
later discuss in more detail) for such X3 . In the case of the , the plotted branching
ratios are not modified simply because hh decays are not allowed for m < 2mh ,
as being considered. If m > 2mh , for parameters such that the partial widths for decays
to SM particles are suppressed, hh could be substantial. However, as we show later,
BR( hh) depends only weakly on the value of X3 .
Also important for h discovery is its total width. In the left-hand window of Fig. 10,
we plot the ratio of the total h width to the corresponding width of a SM Higgs boson
tot , as a function of for m = 20, 55 and 200 GeV. Note
of the same mass, htot/SM
that a substantially larger total width for the h is possible if m is small. In the righttot (for m
hand window, we plot the ratio tot/SM
hSM = m ) as a function of . The is
generally quite narrow. This is true even for m = 200 GeV, for which W W/ZZ decays
are allowed, near the zero in (c + a)2 . The h total width is relatively insensitive to X3 so
long as |X3 | 3, and the total width does not depend at all on X3 for m < 2mh and is
only weakly dependent upon X3 for m > 2mh .
We will now turn to a discussion of the impact of = 0 on the discovery prospects
for the h and at different types of colliders. We will only present results for X3 = 0.
Production rates of the h and are completely independent of X3 . Further, as discussed
above, the h final state branching ratios are only weakly dependent on X3 for |X3 | 3 (the
range suggested by the GoldbergerWise stabilization mechanism). The branching ratios
are, of course, independent of X3 for m < 2mh and are only weakly dependent on X3 for
m > 2mh . Values of |X3 | 10 are required to substantially impact our collider results.
Experimentally, the above results imply that detection of the h at the LHC could be
significantly impacted if | | is large. To illustrate this, we plot in Fig. 11 the ratio of
the rates for gg h , W W h + and gg t th t tb b (the latter two
ratios being equal) to the corresponding rates for the SM Higgs boson. For this figure,
we take mh = 120 GeV, = 5 TeV and X3 = 0 and show results for m = 20, 55
and 200 GeV. In the case of m = 55 GeV, the h decay, discussed in more
detail later, results in a measurable decrease of the SM branching ratios for large | |.
The resulting suppression of the standard LHC modes at the largest allowed | | values
is evident in the W + W h + curves. Another important impact of mixing is
through communication of the anomalous gg coupling of the 0 to the h mass eigenstate.
269
gg, W W () , ZZ () and for mh = 120 GeV, = 5 TeV

Fig. 9. The branching ratios for decays to bb,
and X3 = 0 as functions of for m = 20, 55 and 200 GeV.
270
Fig. 10. The total widths for the h and relative to the value for a SM Higgs boson of the same mass are plotted
as functions of for mh = 120 GeV, = 5 TeV and X3 = 0 taking m = 20, 55 and 200 GeV.
Fig. 11. The ratio of the rates for gg h and W W h + (the latter is the same as that for
to the corresponding rates for the SM Higgs boson. Results are shown for mh = 120 GeV,
gg t th t tbb)
= 5 TeV and X3 = 0 as functions of for m = 20, 55 and 200 GeV.
The result is that prospects for h discovery in the gg h mode could be either
substantially poorer or substantially better than for a SM Higgs boson of the same mass,
depending on and m .16
16 We note that even for parameters such that tot is enhanced relative to tot , the very tiny SM Higgs
h
hSM
width at mh = 120 GeV implies that htot will remain much smaller than the experimental resolution, even in the
important final state.
271
Fig. 12. The ratio of the rates for gg and for W W + (the latter being the same as that
for gg t t t tbb)
= 5 TeV and X3 = 0 as functions of for m = 20, 55 and 200 GeV.
2
At the LC, the potential for h discovery is primarily determined by gZZh
. As shown in
Fig. 3, this reduced coupling-squared (defined relative to the SM value) is often > 1 (and
can be as large as 5), but can also fall to values as low as 0.3, implying significant
suppression relative to SM expectations. The latter suppression is well
within the reach
of the e+ e Zh recoil mass discovery technique at a LC with s = 500 GeV and
L = 500 fb1 . The techniques that have been developed for measuring the total width of
a Higgs boson at the LC indirectly would remain applicable and could reveal the presence
of = 0 mixing through a sizable deviation with respect to the SM prediction.
What about prospects for detection at the LHC? In Figs. 12 and 13, we plot the same
ratios for the as we did for the h in Fig. 11. For this figure, we take mh = 120 GeV,
= 5 TeV and X3 = 0 and show results for m = 20, 55 and 200 GeV in Fig. 12
and for m = 110 and 140 GeV in Fig. 13. For all masses, the gg rate
is generally significantly suppressed relative to the prediction for a SM Higgs boson,
depending upon and m . The dip in the rates is due to a cancellation that zeroes
the coupling, and occurs very close to the point at which the s couplings to vector
bosons and fermions, (c + a)2 , both vanish. Detection of the in gg will
generally be quite difficult. The W W + and gg t t t tb b modes are
generally also quite suppressed relative to SM rates and would probably not be visible.
For m 110 GeV, in addition to the above three modes one can consider the standard
gg ZZ () 4= signal. The ratio for this rate relative to the SM prediction is
plotted in Fig. 14 for m = 110, 140 and 200 GeV. For m = 200 GeV, the very high level
of statistical significance predicted for the SM ZZ final state signal at this mass implies that
detection in this mode should be possible except near the zeroes in the ZZ coupling.
For the m = 110 and 140 GeV cases, the dip region occupies a lot of the allowed range
and suppression is generally present even away from the dip regions. Detection in the 4=
mode would be unlikely.
272
Fig. 13. As in Fig. 12, but for m = 110 and 140 GeV.
Fig. 14. The ratio of the rate for gg ZZ to the corresponding rate for a SM Higgs boson with mass m
assuming mh = 120 GeV, = 5 TeV and X3 = 0 as a function of for m = 110, 140 and 200 GeV. Recall
that the range is increasingly restricted as m becomes more degenerate with mh .
2
At the LC, the potential for discovery is primarily determined by gZZ
. As shown
in Fig. 4, this reduced coupling-squared is typically substantially suppressed relative to
the SM value of 1. Still, because of the very high statistical
significance associated with a
SM Higgs signal in the e+ e Z + Higgs mode for s = 500 GeV and L = 500 fb1 ,
detection of the will be possible except near the zero in the ZZ coupling. As discussed
earlier, the width of the would be much smaller than anticipated. This could be checked
using the techniques that have been developed for measuring the total width of a narrow
Higgs boson at the LC indirectly.
273
Fig. 15. The rates for h bb and bb relative to the corresponding rate for a SM Higgs boson
of the same mass. Results are shown for mh = 120 GeV, = 5 TeV and X3 = 0 as functions of for m = 20,
55 and 200 GeV.
Also of considerable interest is how = 0 would affect prospects for h and detection
at a collider. To assess this, we plot in Fig. 15 the h bb and bb
rates relative to the SM rates evaluated for Higgs mass equal to mh or m , respectively.
In the case of the h, the plot differs only slightly from Fig. 11 for the W W h +
LHC discovery mode. This means that the anomaly contribution to the h coupling is
much smaller than that from the standard fermion and W boson loops. In the case of the
, differences between these bb curves and the corresponding W W
+ curves of Fig. 12 are somewhat larger, especially in the vicinity of the zeroes.
To summarize the results, in the case of the h, for the parameters considered, the rate is
suppressed by at most a factor of 0.5 and would thus be quite sufficient to yield a highly
detectable and accurately measurable signal. The would typically be much more difficult
to discover in collisions. Large dips in the rate occur in the vicinity of the zero in
the b b coupling and branching ratio, which is at the same location as the zero in
gZZ . The gg channel is somewhat less suppressed in the dip region due to
the anomalous contribution to the gg coupling. However, the signal is still small in the
dip regions and this channel would have large backgrounds. Although it would be difficult
to isolate, further study might be warranted.
An important question is whether the deviations due to the anomalous 0 gg and 0
couplings are sufficiently large to be measurable. To quantify this, we plot in Fig. 16 the
ratios17
Rsgg
2 (with anomaly)
gsgg
2 (without anomaly)
gsgg
and Rs
2 (with anomaly)
gs
2 (without anomaly)
gs
(67)
17 Once again, we remind the reader that the g notation refers to the full coupling strength as normally defined,
whereas gs without a bar are reserved for certain coupling ratios.
274
Fig. 16. In the upper plots, we give the ratios Rhgg and Rgg of the hgg and gg couplings-squared including
the anomalous contribution to the corresponding values expected in its absence. Results for the analogous ratios
Rh and R are presented in the lower plots. Results are shown for mh = 120 GeV and = 5 TeV as
functions of for m = 20, 55 and 200 GeV. (The same type of line is used for a given m in the right-hand
figure as is used in the left-hand figure.)
for s = h and s = . These ratios can be determined experimentally. First, (model2

2
independent) measurements of the gZZh
= (d + b)2 and gZZ
= (c + a)2 coupling
+
+
factors for the h and are obtained using e e Zh and e e Z production
at the linear collider. The sgg and s couplings (s = h or ) expected from the
standard fermion and W -boson loops in the absence of the anomalous contribution can
2 and g 2 , including any
then be computed. Meanwhile, the actual couplings-squared, gsgg
s
275
anomalous contribution, can be directly measured using a combination of s bb

and gg s data.
In more detail, we employ the following procedures.
2
(defined relative to the SM prediction at mhSM = ms ) from
First, obtain gZZs
(e+ e Zs) (inclusive recoil technique).
= (e+ e Zs Zbb)/
Next, determine BR(s bb)

(e+ e Zs).
2
from
b
b)/BR(s
b
b).
Then, compute g s
To display the contribution to the s coupling-squared from the anomaly one would
then compute
Rs
2 (from experiment)
gs
2 (from experiment)
gh2SM (as computed for mhSM = ms ) gZZs
. (68)
2 experimentally requires one more step. We must compute (gg

To determine gsgg
s )/BR(s ). To obtain BR(s ), we need a measurement of stot .
Given such a measurement, we then compute
BR(s ) =
2 )
(s )(computed from gs
stot (from experiment)
(69)
2
where the above experimental determination of gs
is employed and the experimental
tot
techniques outlined in [23] are employed for s .
The ratio analogous to Eq. (68) for the gg coupling is then
Rsgg
2 (from experiment)
g sgg
2 (from experiment)
gh2SM gg (as computed for mhSM = ms ) gZZs
(70)
For a light SM Higgs boson, the various cross sections and branching ratios needed for
the s coupling can be determined with errors of order a few percent [23]. We see from
Fig. 16 that for large this level of accuracy is on the edge of being sufficient to detect the
deviation in the case of the h. In the case of the , the expected deviation is typically much
larger, especially near the zeroes in the rates. Indeed, the size of the deviation is largest
when the rate is smallest. A careful study is needed to assess the prospects. For the gg
coupling, errors might be dominated by the accuracy with which the total width can be
determined. Estimates for this error in the case of the SM hSM are in the neighborhood
of 10% for mhSM = 120 GeV [23], decreasing for higher mhSM . Thus, the factor of two
deviations expected in the case of the hgg coupling-squared at the higher values might
well be discernible experimentally. Since the may prove difficult to detect at the LHC,
a much more detailed study is required to see if deviations in the gg coupling due to the
anomalous contribution could be detected.
We will now turn to a more thorough exploration of the parameter regions in which
h decays are large. The branching ratios for h in the case of mh = 120 GeV,
= 5 TeV and X3 = 0 are shown in Fig. 17 for various choices within the allowed
region. The plots show that h decays can be quite important at the largest | | values
when m is close to mh /2. Detection of the h decay mode could easily provide
276
Fig. 17. The branching ratios for h , gg and bb for mh = 120 GeV, X3 = 0 and = 5 TeV as
a function of m for = 2.16, 1.66, 1.16 and 0.66 (left-hand graphs) and for = 0.66, 1.16, 1.66, and
2.16 (right-hand graphs).
the most striking evidence for the presence of = 0 mixing. In order to understand how
to search for the h decay mode, it is useful to know how the decays. In Fig. 17
for the same m and values
we give detailed results for BR( gg) and BR( bb)
+
for which BR(h ) is plotted. (The cc and channels supply the remainder.) For
277
Fig. 18. The branching ratio for h for mh = 120 GeV, X3 = +3 and = 5 TeV as a function of m
for = 2.16, 1.66, 1.16 and 0.66 (left-hand graphs) and for = 0.66, 1.16, 1.66, and 2.16 (right-hand
graphs). The approximate interchange of < 0 and > 0 results relative to Fig. 17 is simply the accidental
prediction of the theory for the parameter choices made.
is always substantial and might make detection of the h 4b

> 0, BR( bb)
and h 2g2b final states possible. The decay mode always has a very
tiny branching ratio and the related detection channels would not be useful.
An important question is whether the h branching ratio depends significantly on
the value of X3 . To illustrate, we plot in Fig. 18 results for BR(h ) for the case of
X3 = +3. The most important point is that the branching ratio is not dramatically increased
in overall magnitude compared to the X3 = 0 resultsit is typically below 0.1 except for
the highest allowed values and m 50 GeV, for which the branching ratio can reach
values of order 0.3 to 0.5. The second remarkable result, peculiar to the above particular
choice of X3 = +3, is the approximate interchange of results for < 0 with those for > 0
relative to the results obtained for X3 = 0.
One will probably first search for the h in the modes that have been shown to be viable
for the SM Higgs boson. We have given in Fig. 11 the rates for important LHC discovery
modes relative to the corresponding SM values in the case of m = 55 GeV. Results for
other m < mh /2 values are similar in nature. We observe that the W W h +
and gg t th t tb b detection modes are generally sufficiently mildly suppressed that
detection of the h in these modes should be possible (assuming full L = 300 fb1
luminosity per detector). The gg h detection mode could either be enhanced or
significantly suppressed relative to the SM expectation. Once the h has been detected in one
b and h bbgg
decay
of the SM modes, a dedicated search for the h bbb
modes will be important. At the LHC, backgrounds for these modes will be substantial and
a thorough Monte Carlo assessment is needed.
2
At the LC, since gZZh
is close to 1 (relative to the SM Higgs value), the h will be readily
detectable using the recoil mass procedure in e+ e Zh events. Once the h mass peak is
detected, it should be possible to delineate in detail the h and branching ratios.
278
Fig. 19. The hh branching ratio is plotted as a function of for mh = 120 GeV and m = 250, 300
and 350 GeV. We have taken = 5 TeV and assumed m1 > m . Results shown are for X3 = 0. Results for
X3 = +3 are nearly identical.
2
As for detection of the at the LC, the most relevant quantity is gZZ
. Detailed plots
of this quantity appear in Fig. 4. These plots indicate that LC detection of e+ e Z
2
. For a significant
using the recoil mass method will require being far from the zero in gZZ
portion of parameter space, it seems quite apparent that the only way to detect the would
be through the h decays.
In order to have substantial BR(h ) it is necessary that mh < 2mW . As mh is
increased above 2mW , the W W and then ZZ modes become strong and overwhelm the
decay mode. For example, for mh = 200 GeV, the largest value found for BR(h ) is
of order 12%, and such values are again achieved when | | is as large as possible and m
is just below mh /2.
Let us now discuss hh decays. For mh = 120 GeV, these are present once
m 240 GeV. When allowed, these decays will be quite strong since the hh coupling
is typically larger than the h coupling away from zeroes in the coupling. In addition,
the decays to f f and V V are typically suppressed compared to those of the h because
of the smaller size of (c + a)2 compared to (d + b)2 when m > mh (see Figs. 3 and 4)
for all but the largest | | values. The importance of the hh decays is illustrated for
mh = 120 GeV, = 5 TeV and X3 = 0 for the cases of m = 250, 300 and 350 GeV
in Fig. 19. Even though m > 2mW in all these cases, BR( hh) is still of order 0.3
0.4 for most of the allowed range not near a zero in the hh coupling. We also have
examined the sensitivity of BR( hh) to X3 . We find very little dependence on X3 . For
example, the plots of Fig. 19 for X3 = 0 are not visibly altered for X3 = +3.
279
5. Phenomenology for = 1 TeV

In this section, we consider the more marginal choice of = 1 TeV. For this case,
we will consider first results obtained assuming m1 is large (> 1 TeV). As discussed
earlier, such large m1 requires large curvature, m0 /MPl O(1), that would presumably
imply significant corrections to the RS ansatz. Nonetheless, the large-m1 results provide
a useful benchmark that might provide a reasonable first approximation in such a case.
We also noted that for = 1 TeV and m0 /MPl 1 large m1 is needed to clearly avoid
any constraints from RunI Tevatron data. We next consider results obtained in two smallcurvature cases: m0 /MPl 0.065 and 0.195, corresponding to m1 = 100 and 300 GeV,
respectively. As discussed, such small m1 values might or might not be inconsistent with
constraints from current RunI Tevatron data and from the S and T electroweak observables.
However, the very interesting physics associated with Higgs decays to KK excitations that
emerges deserves attention just in case this scenario should arise. In our presentation for
= 1 TeV, we focus only on the significant changes as compared to = 5 TeV.
First, we present the allowed region in (, m ) parameter space for mh = 120 GeV in
Fig. 20. Of course, the allowed range is very much reduced compared to = 5 TeV
since is five times larger; see Eq. (53). As compared to Fig. 2, we see that there is a
significant region with lower m , but with m 8 GeV, that is excluded by the LEP/LEP2
limits coming from untagged hadronic events and/or from b-tagged final states. A similar
2
region is not excluded in the = 5 TeV case because the gZZ
coupling for = 5 TeV
is substantially smaller than for = 1 TeV. Returning to the = 1 TeV case, points
2
with m 8 GeV are not excluded because the upper bound on gZZ
coming from
untagged hadronic final states rises very rapidly as one moves to lower masses and there
Fig. 20. As in Fig. 2 but for = 1 TeV. The region of theoretically allowed > 0 values below 0.45 with
55 GeV m 115 GeV that are in the light LEP/LEP2-excluded region will be referred to as the LE region.
280
2
Fig. 21. For mh = 120 GeV and = 1 TeV, we plot contours for gZZ
= (c + a)2 inside the region that can
2
only be excluded if we assume the limits of [18] apply. The two thick lines are the values such that gZZh
= 2.
2
< 2 and is that most likely to be consistent with precision electroweak
The region between these lines has gZZh
data.
are no limits from b-tagged final states. However, we should note that if the limits of
[18] apply (we have assumed they do not because of the dominance of gg decays),
for = 1 TeV the m 8 GeV region would be excluded as well as the m 8 GeV
2
in this
regions shown. As illustrated in Fig. 21, for = 1 TeV the magnitude of gZZ
low-m region is not so very small.
2
2
= (d + b)2 and gZZ
= (c + a)2 are presented in Fig. 22. There,
Contours of gZZh
2
we see that region LE is excluded by LEP/LEP2 data because in this region the gZZ
value gets to be a reasonable fraction of one, the SM value. Globally speaking, the main
difference between the couplings for = 1 TeV versus those for = 5 TeV of Figs. 3
2
is overall much larger in the = 1 TeV case.
and 4 is that gZZ
Next, we present the corresponding graphs related to LHC and collider discovery.
In these collider discovery plots, we have included the full theoretically allowed range of
(, m ) values, including those that are marked as LEP/LEP2-excluded in Fig. 20. For
these graphs, we have chosen to focus on mh = 120 GeV (as for = 5 TeV) and on
the values of m = 50, 65 and 200 GeV. The lowest value still gives a substantial range
of LEP-allowed and will have significant BR(h ). For the middle value, these
decays are forbidden. In all the LHC and collider graphs, m1 is assumed to be large, in
particular large enough that decays of the Higgs or radion to h1 are forbidden.
We should note that = 1 TeV branching ratios are influenced much more by the
choice of X3 than are the = 5 TeV results. In order to facilitate comparison, we will
plot many of same LHC and LC discovery ratios as given for the = 5 TeV, X3 = 0
case. We will later discuss how X3 = 0 influences BR(h ) and, implicitly, the SM h
branching ratios. The main implication of Figs. 2327 is that for = 1 TeV, discovery
at the LHC and in collisions has much better prospects (away from the usual zeroes
in the gg and couplings) than in the case of = 5 TeV. It is still true that the
anomalous contributions to the h and couplings will be hard to isolate.
The greatest interest in the lower value derives from the fact that h decays
can be much more prominent and that h, decays to final states containing the 1st KK
281
Fig. 22. For mh = 120 GeV and = 1 TeV, we plot contours for the quantities (d + b)2 and (c + a)2 . For
(d + b)2 , only the region m 15 GeV is shown. For (c + a)2 , we show the narrow pipe that connects to the
allowed region of very small m ; see Fig. 21.
Fig. 23. The ratio of the rates for gg h and W W h + (the latter being the same as that for
gg t th t tbb)
X3 = 0 and = 1 TeV as functions of for m = 50, 65 and 200 GeV. LEP bounds not imposed.
282
Fig. 24. The ratio of the rates for gg (the higher curves in the 0 region) and for
to the corresponding rates for
W W + (the latter being the same as that for gg t t t tbb)
the SM Higgs boson. Results are shown for mh = 120 GeV, X3 = 0 and = 1 TeV as functions of for
m = 50, 65 and 200 GeV. LEP bounds not imposed.
Fig. 25. The ratio of the rates for gg (the higher curves in the 0 region) and for
to the corresponding rates for
W W + (the latter being the same as that for gg t t t tbb)
the SM Higgs boson. Results are shown for mh = 120 GeV, X3 = 0 and = 1 TeV as functions of for
m = 110 and 140 GeV. LEP bounds not imposed.
excitation h1 become possible. The first point is illustrated in Fig. 28. BR(h ) can
be as large as 50% at the highest allowed | | values. Sensitivity of BR(h ) to X3
is substantial. As illustrated in Fig. 29, for X3 = +3 and < 0, BR(h ) > 0.5 is
entirely possible. This would substantially depress the discovery mode ratios presented
earlier for X3 = 0.
Let us now discuss what happens at higher mh values. The largest value that can be
easily consistent with precision electroweak constraints is mh = 200 GeV. In order to
learn if the h h1 decay
could be significant, we retain = 1 TeV, for which Eq. (44)
W = / 3 = 577 GeV, and choose m1 = 100 GeV (corresponding to
implies that
283
Fig. 26. The ratio of the rate for gg ZZ to the corresponding rate for a SM Higgs boson with mass m
assuming mh = 120 GeV, X3 = 0 and = 1 TeV as a function of for m = 110, 140 and 200 GeV. Recall
that the range is increasingly restricted as m becomes more degenerate with mh . LEP bounds not imposed.
Fig. 27. The rates for h bb and bb relative to the corresponding rate for a SM Higgs boson
of the same mass. Results are shown for mh = 120 GeV, X3 = 0 and = 1 TeV as functions of for m = 50,
65 and 200 GeV. LEP bounds not imposed.
m0 /MPl 0.065, see Eq. (44)). Results for BR(h ) and BR(h h1 ) are plotted
in Fig. 30 as a function of m for selected values of > 0, assuming X3 = 0. The
h branching ratio can be significant, especially for the larger values of allowed
by the theoretical constraint of Z 2 > 0. Certainly, these decays should be searched for
at the LC as their presence would imply non-zero and would allow a measurement
of this very fundamental parameter. The reason for the small size of the h and
h h1 branching ratios is the dominance of the W W and ZZ decay modes. Once
these V V decays become full strength, the and h1 decays will be rare. Sensitivity
of BR(h ) to X3 is significant, as illustrated in Fig. 31, but the h branching
ratio remains quite small (unless extremely large values of X3 are employed).
284
Fig. 28. For various > 0 and < 0 values, the branching ratio for h is plotted as a function of m ,
taking mh = 120 GeV, X3 = 0 and = 1 TeV. Only points not excluded by LEP/LEP2 (the allowed region
of Fig. 20) are plotted. The curves terminate at low m when the LEP/LEP2 limits of are encountered.
As noted earlier, at still larger values of mh precision electroweak constraints become

difficult to satisfy. A future paper will explore this region in more detail. Very roughly,
2 m7 /m4n behavior found earlier in Eq. (65) means that h hn
the (h hn )
h
W
decays can dominate over the W W and ZZ decay modes that grow only as m3h , provided
W (and hence ) is of order a TeV. To illustrate,
that m1 is sufficiently small and that
in Fig. 32 we plot BR(h h1 + h2 ) for mh = 800 GeV as a function of m for a
number of positive values and for the cases of m1 = 100 GeV and m1 = 300 GeV.
(BR(h ) is typically below or of order 0.01 for this large a value of mh .) In
obtaining the results shown, we have assumed that h h1 h1 + decays (for which
vertices do exist but have not been studied in detail in this paper) are unimportant even
though they are kinematically allowed for the mh and m1 choices of this figure. We
also note that BR(h h1 ) > 11BR(h h2 ) for the cases studied, as anticipated
from the (m1 /m2 )4 (3.8/7)4 0.086 scaling noted above. From Fig. 32, we see that
m1 = 100 GeV yields large values of BR(h h1 + h2 ) at small m when is not
285
Fig. 29. For various > 0 and < 0 values, the branching ratio for h is plotted as a function of m , taking
mh = 120 GeV, X3 = +3 and = 1 TeV. Only points not excluded by LEP/LEP2 (the allowed region of
Fig. 20) are plotted. The curves terminate at low m when the LEP/LEP2 limits of are encountered.
small. For m1 = 300 GeV, BR(h h1 + h2 ) is much smaller than for m1 = 100 GeV,
being of order 0.040.4 for small values of m and ranging from 0.05 to 0.60. Results
for < 0 are very similar in nature. Sensitivity to X3 is minimal.
6. Summary and conclusions

We have discussed the scalar sector of the RandallSundrum model. The effective
potential (defined as a set of interaction terms that contain no derivatives) for the Standard
Model Higgs-boson (h0 ) sector interacting with KaluzaKlein excitations of the graviton
n
(h ) field and the radion (0 ) field has been derived. Without specifying its origin,
a stabilizing mass-term for the radion has been introduced. After including this term,
we have shown that only the Standard Model vacuum determined by V (h0 )/h0 = 0
is allowed. Further, we find that consistency of the RS solution requires that the Higgs
286
Fig. 30. The h and h h1 branching ratios as a function of m for mh = 200 GeV, m1 = 100 GeV,
W = 1 TeV, for various choices. Results are plotted only for m values satisfying
X3 = 0 and = 3
LEP/LEP2 bounds (the allowed region of Fig. 20). The curve legend for the right-hand plot is the same as
shown in the left-hand plot.
Fig. 31. As in Fig. 30, but for X3 = +3.
potential vanishes at the vacuum solution. Otherwise, the finely tuned matching required
in the RS model between the bulk and branes would be violated. As a result, for the
correct vacuum solution the effective potential does not contain any terms linear in the
quantum Higgs field. The above results emerge only with a very full treatment of the
effective potential. Truncation of its expansion in powers of the fields can lead to erroneous
conclusions.
Having confirmed that the usually assumed vacuum properties are correct, we pursue
in more detail the phenomenology of the RS scalar sector, focusing in particular on results
287
Fig. 32. We plot the h h1 + h2 branching ratio as a function of m for mh = 800 GeV, X3 = 0 and
W = 1 TeV, in the cases of m1 = 100 GeV and 300 GeV, for various choices (as indicated in the
= 3
right-hand window). Results are plotted only for m values satisfying LEP/LEP2 bounds. In this plot, we have
assumed that h h1 h1 + decays (for which vertices do exist but have not been studied in detail in this paper)
are unimportant even though they are kinematically allowed for the mh and m1 choices of this figure.
contribution to the Lagrangian.

H
found in the presence of a curvature-scalar mixing R H
We delineate the somewhat tricky inversion procedure for determining all the Lagrangian
parameters given the masses of the physical eigenstates h and . A full set of Feynman
rules for the resulting tri-linear interactions among the h, and hn mass eigenstates are
then derived. These include possible contributions from the model-dependent 03 cubic
interactions. We also summarize the Feynman rules for couplings to standard channels:
W W, . . . as well as gg and (including the anomalous contributions to the latter).
b b,
Simple sum rules that relate Higgs-boson and radion couplings to pairs of vector bosons
and fermions are given. Of particular interest is the fact that non-zero induces interactions
linear in the Higgs field: 2 h and hn h. The explicit forms of these interactions must be
obtained using the above-mentioned full treatment of the effective potential as well as a
similarly full treatment of the related derivative terms in the Lagrangian.
We summarize the connections between the parameters of the model and the lower
bounds on the new physics scales among these parameters required by precision
electroweak and Tevatron RunI constraints. We explore the behavior of the couplings in the
range of parameter space allowed by theoretical and existing experimental constraints. In
particular, we derive the regions of parameter space that are excluded by direct LEP/LEP2
limits on scalar particles with ZZ coupling as function of scalar mass. Of particular note
is the fact that the sum rule for ZZh and ZZ squared-couplings noted above implies that
it is impossible for both the h and to be light.
We note that precision electroweak data is most naturally satisfied if the h and masses
are modest in size, 200 GeV. We focus on the case of small to moderate mh and m , and
discuss expectations for h and production/detection at the LHC and a LC in comparison
to the SM Higgs boson. In the regions of parameter space allowed by theoretical and current
288
experimental constraints, we find that LHC detection of the is likely to be quite difficult.
In addition, LHC detection of the h is not guaranteed.
One particularly interesting complication for = 0 is the presence of the non-standard
decay channel h . The h decay could easily be present since in the context
of the RS model there is a possibility (perhaps even a slight preference) for the to be
substantially lighter than the h. In particular, m < mh /2 is a distinct possibility. We
study in detail the phenomenology when m 60 GeV for mh = 120 GeV, for which
h is possible. Sensitivity of these decays to the strength of the 03 cubic potential
is discussed. The importance and feasibility of eventually measuring BR(h ) as a
means of determining the potential responsible for stabilizing the radion is emphasized.
In the main phenomenology section, Section 4, we consider the new physics scales
of = 5 TeV and m1 750 GeV for the first KK resonance, h1 . These values
imply that constraints from precision electroweak data and from RunI Tevatron data are
clearly satisfied. For this case, the h h1 + modes, which are also potentially very
interesting, are forbidden for the moderate mh and m values explored here.
For mh = 120 GeV, for the largest allowed | | values and for m close to mh /2, the
h mode will substantially dilute the rates for the usual search channels. In fact, we
find that BR(h ) could easily be as large as 3040%. Regardless of the magnitude
of BR(h ), detection of this decay would be very important as it provides a crucial
experimental signature for non-zero . Sensitivity of this branching ratio to the parameter
X3 characterizing the 03 interaction is illustrated.
Of course, it is also possible that m > 2mh . Because of the typically large size of the
hh coupling, we find that hh decays will have a large branching ratio even when
m > 2mW .
We give additional details regarding direct detection of the for the portion of
parameter space for which h decays are important. Prospects for direct detection
at the LHC are not encouraging. At the LC, one should be able to detect e+ e Z
using the recoil mass technique; b-tagging is not necessarily reliable due to the possibility
that decays will be dominated by the gg mode.
In addition to the above, we give a first assessment of whether or not the anomalous
contribution to the hgg, h , gg and couplings could be observed experimentally.
Deviations in these couplings-squared due to the anomalous contribution are plotted and
compared to the errors expected from the outlined experimental procedures for extracting
such deviations. Prospects in the case of the h are relatively encouraging.
In a second phenomenology section, Section 5, we consider the case of much lighter
new physics scales set by = 1 TeV. In this case, we consider both large m1 , which
avoids precision EW and RunI constraints, but requires large five-dimensional curvature,
and the small curvature values of m1 = 100 and 300 GeV. It is not clear if these latter
cases are ruled out by precision electroweak and/or RunI Tevatron data. If such low scales
are allowed, the h branching ratio becomes even more prominent for our sample
choice of mh = 120 GeV. For mh = 200 GeV and above, h h1 decays rapidly emerge
and become dominant for mh 500 GeV in the case of m1 = 100 GeV. However, it must
be kept in mind that for such large mh values other new physics must compensate the
consequent large precision electroweak contributions from the Higgs loop graphs.
289
Overall, the RandallSundrum scenario leads to a fascinating extension of the usual

Higgs phenomenology, especially if radion-Higgs mixing is present, as is most naturally
the case.
Note added
In the course of preparing this paper, another article appeared dealing with the variations
of the couplings of the h and and their branching ratios due to the curvature-scalar
mixing [24]. In the latest revision of this paper, v.3, posted to hep-ph on July 2, 2003, the
results presented now appear to be in substantial agreement with our results as originally
presented in our submission to hep-ph of June 20, 2002.
Acknowledgements
B.G. thanks Zygmunt Lalak, Krzysztof Meissner and Jacek Pawelczyk for useful
discussions. J.F.G. would like to thank J. Wells for useful discussions. B.G. is supported in
part by the State Committee for Scientific Research under grant 5 P03B 121 20 (Poland).
J.F.G. is supported by the US Department of Energy and by the Davis Institute for High
Energy Physics.
Appendix A. Feynman rules

In this appendix, we summarize the relevant Feynman rules for the mass eigenstates ,
h and hn . Rules for the V V couplings of the and h, the f f couplings of the and h,
and the tri-linear self-couplings among the , h and hn fields are given. We note that we

i ] j = 2 ij . In this case, the spin sum
are employing a normalization in which [

n
i
i (k) = B
for the h state polarizations is i=1,5 (k)
(k), where

k k
k k
k k
k k
B (k) = 2
2 + 2
2
mn
mn
mn
mn

k k
k k
2
2
2 .
(A.1)
3
mn
mn
The parameters a, b, c, d define the mixing between the = 0 states and the = 0 mass
eigenstates. They are defined in Eqs. (34) and (35). In the = 0 limit, d = a = 1,
c = b = 0. The auxiliary functions for the gg and couplings of the h and (Fig. 33)
are given by

F1/2 ( ) = 2 1 + (1 )f ( ) ,
(A.2)
F1 ( ) = 2 + 3 + 3 (2 )f ( ),
(A.3)
290
gmZ
(d + b)
cW
g mf
(d + b)
2 mW
g mf
(c + a)
2 mW
igmW (d + b)
gmZ
(c + a)
cW
igmW (c + a)

s

icg ab k1 k2 k1 k2 : cg =
F1/2 (i ) 2b3 gr
gf V
4 v
i

ic k1 k2 k1 k2 : c =
ei2 Nci Fi (i ) (b2 + bY )gr
gf V
2 v
i
Fig. 33. Feynman rules for the V V and f f couplings of the scalars h and . Note: Since there are no
pseudoscalars, there are no single V vertices. We have dropped the extra terms related to gauge fixing, see
Ref. [9], as appropriate when considering on-shell W s or Zs or when working in the unitary gauge. For
gg, final states, we give only the on-shell rules. There, i = 4m2i /m2h, where mi is the mass of the internal
loop particle. The auxiliary functions for spin-1/2 and spin-1 loop particles are defined in Appendix
A. The

SU(3) SU(2) U (1) function coefficients are b3 = 7, b2 = 19/6 and bY = 41/6. For cg , the i is over all

colored fermions (assumed to have Nci = 3). For c , the i comprises all charged fermions (including quarks,
i
with Nc = 3 and ei = 2/3 or 1/3, and leptons, with ei = 1 and Nci = 1) and the W boson (with ei = 1 and
Nci = 1). For the h, gf V = gZZh = d + b and gr = b. For the , gf V = gZZ = c + a and gr = a.
for spin-1/2 and spin-1 loop particles, respectively, with

1,
arcsin2 (1/ ),
1
1+ 1 2

f ( ) = ln
=
2
1+
1
4
14 ln 11 i , 1.
1 1
(A.4)
The variable for a given loop is defined as 4m2 /M 2 , where m is the mass of the
internal loop particle and M is the mass of the scalar state, h or . In Fig. 34, the g
couplings used in the branching ratio formulae, Eqs. (64) and (65), are defined. In the
i ghhh

i
bd 12b + d(6 + 1) k12 + k22 + k32 12dm2h
0

3 1 d 3 m2h 3X3 b3 m2
0
i g

i
ac 12a + c(6 + 1) k12 + k22 + k32 12cm2h
0

3 1 c3 m2h 3X3 a 3 m2
0
i gh
291

i
6a (ad + bc) + cd + bc2 k12 + k22

+ c 12ab + 2ad + bc(6 1) k32 4c(2ad + bc)m2h
0

3 1 c2 dm2h 3X3 a 2 bm2
0

i
6b (ad + bc) + cd + ad 2 k12 + k22
i ghh

+ d 12ab + 2bc + ad(6 1) k32 4d(ad + 2bc)m2h
0

3 1 cd 2 m2h 3X3 ab2 m2
0
i g nh k1 k2

1
i
4 3 a( b + d) + bc + cd k1 k2
W
2
i g n k1 k2

i
1
4 3a [a + 2c] + c2 k1 k2
W
2
i g nhh k1 k2

i
1
4 3b [b + 2d] + d 2 k1 k2
W
2
Fig. 34. Feynman rules for the tri-linear vertices in the scalar sector. All momenta are outward flowing. In the
hn vertices, we have made use of the symmetry of hn under .
= 0 SM limit, ghhh = 3m2h /v0 (where mh = mh0 at = 0). We have not derived
detailed results for the hhn hn and hn hn couplings, although, as explained in the text,
such couplings do exist.
References
[1] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
[2] C. Csaki, M. Graesser, L. Randall, J. Terning, Phys. Rev. D 62 (2000) 045015, hep-ph/9911406.
292
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
S.B. Bae, P. Ko, H.S. Lee, J. Lee, Phys. Lett. B 487 (2000) 299, hep-ph/0002224.
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Phys. Rev. Lett. 84 (2000) 2080, hep-ph/9909255.
K. Cheung, Phys. Rev. D 63 (2001) 056007, hep-ph/0009232.
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Phys. Rev. D 63 (2001) 075004, hep-ph/0006041.
S.C. Park, H.S. Song, J. Song, Phys. Rev. D 63 (2001) 077701, hep-ph/0009245.
G. Giudice, R. Rattazzi, J. Wells, Nucl. Phys. B 595 (2001) 250, hep-ph/0002178.
C. Csaki, M.L. Graesser, G.D. Kribs, Phys. Rev. D 63 (2001) 065002, hep-th/0008151.
T. Han, G.D. Kribs, B. McElrath, Phys. Rev. D 64 (2001) 076003, hep-ph/0104074.
M. Chaichian, A. Datta, K. Huitu, Z.H. Yu, Phys. Lett. B 524 (2002) 161, hep-ph/0110035.
W.D. Goldberger, M.B. Wise, Phys. Rev. Lett. 83 (1999) 4922, hep-ph/9907447.
C. Charmousis, R. Gregory, V.A. Rubakov, Phys. Rev. D 62 (2000) 067505, hep-th/9912160.
J.J. van der Bij, Acta Phys. Pol. B 25 (1994) 827;
R. Raczka, M. Pawlowski, Found. Phys. 24 (1994) 1305, hep-th/9407137.
G.D. Kribs, in: R. Davidson, C. Quigg (Eds.), Proc. of the APS/DPF/DPB Summer Study on the Future of
Particle Physics, Snowmass, 2001, hep-ph/0110242.
R. Delbourgo, D.S. Liu, Austral. J. Phys. 53 (2001) 647, hep-ph/0004156.
OPAL Collaboration, G. Abbiendi, et al., Eur. Phys. J. C 7 (1999) 407, hep-ex/9811025.
ALEPH Collaboration, D. Buskulic, et al., Phys. Lett. B 313 (1993) 312.
OPAL Collaboration, preliminary physics note: OPAL Physics Note 495, February 2002.
P. Teixeira-Dias, on behalf of the LEPHIGGS working group, LEP seminar, July 10, 2001.
LEP Higgs Working Group for Higgs boson searches Collaboration, hep-ex/0107029.
A. Djouadi, J. Kalinowski, M. Spira, Comput. Phys. Commun. 108 (1998) 46.
American Linear Collider Working Group Collaboration, T. Abe, et al., in: R. Davidson, C. Quigg (Eds.),
Proc. of the APS/DPF/DPB Summer Study on the Future of Particle Physics Snowmass, 2001, hepex/0106056, see Sections 8.5 and 8.6, and Table 3.2.
J.L. Hewett, T.G. Rizzo, hep-ph/0202155.

HamiltonJacobi method and effective actions

of D-brane and M-brane in supergravity
Matsuo Sato, Asato Tsuchiya
Department of Physics, Graduate School of Science Osaka University, Toyonaka, Osaka 560-0043, Japan
Received 26 May 2003; received in revised form 22 July 2003; accepted 20 August 2003
Abstract
We show that the effective actions of D-brane and M-brane are solutions to the HamiltonJacobi
(HJ) equations in supergravities. This fact means that these effective actions are on-shell actions
in supergravities. These solutions to the HJ equations reproduce the supergravity solutions that
represent D-branes in a B2 field, M2 branes and the M2M5 bound states. The effective actions in
these solutions are those of a probe D-brane and a probe M-brane. Our findings can be applied to the
study of the gauge/gravity correspondence, especially the holographic renormalization group, and a
search for new solutions of supergravity.
PACS: 11.25.Tq; 11.25.Uv; 11.25.Yb; 04.65.+e
1. Introduction
In this paper, we show that the D-brane effective action (the BornInfeld action plus the
WessZumino action) is a solution to the HamiltonJacobi (HJ) equation of type IIA (IIB)
supergravity and that the M-brane effective action is a solution to the HJ equation of 11dimensional (11d) supergravity. This fact means that the effective actions of D-brane and
M-brane are on-shell actions in supergravities. We also show that these solutions to the HJ
equations reproduce the supergravity solutions which represent a stack of D-branes in a B2
field, a stack of M2-branes and a stack of the M2M5 bound states. In fact, we reported
the case of D3-brane in our previous publication [1], and in this paper we generalize our
previous result to the cases of Dp-branes and M-branes.
E-mail addresses: machan@het.phys.sci.osaka-u.ac.jp (M. Sato), tsuchiya@het.phys.sci.osaka-u.ac.jp
(A. Tsuchiya).
doi:10.1016/j.nuclphysb.2003.08.025
294
M. Sato, A. Tsuchiya / Nuclear Physics B 671 (2003) 293324
The D-brane effective action on a curved background is obtained in principle by

calculating the disk amplitude in superstring on the background. The disk amplitude in the
open string picture is translated into the closed string picture as the transition amplitude
between the vacuum and the boundary state representing a probe D-brane. This transition
amplitude should reduce in the 0 limit to an on-shell action in type IIA (IIB)
supergravity, which is a functional of the values of the fields on a boundary, Therefore,
the 0 limit of the D-brane effective action should be a solution to the HJ equation of
type IIA (IIB) supergravity. Considering the gauge invariance, we see that the 0 limit
of the D-brane effective action corresponds to setting the combination of the gauge field
plus the NS 2-form field to be zero. Nevertheless, a nontrivial fact we obtain is that the
D-brane effective action itself is a solution to the HJ equation. Probably this fact comes
from the supersymmetry, since the RR plays a crucial role in our analysis.
The strategy of our analysis is as follows. We reduce type IIA (IIB) supergravity on
S 8p , dropping the fermionic degrees of freedom consistently, and obtain a (p + 2)dimensional gravity. Adopting the radial coordinate as time, we develop the canonical
formalism based on the ADM decomposition for this (p + 2)-dimensional gravity and
obtain the HJ equation originating from the Hamiltonian constraint. We solve the HJ
equation under the condition that the fields be constant on fixed-time surfaces, and find
that the Dp-brane effective action is a solution to the HJ equation and reproduces
the supergravity solution of a stack of Dp-branes in a B2 field. We note here that
the near-horizon limit of this supergravity solution with p = 3 is conjectured to be
dual to noncommutative YangMills (NCYM) [24] and reduces to AdS5 S 5 in the
commutative limit, which is dual to N = 4 super-YangMills [5,6]. In our formulation,
the fixed-time surface whose dimension is p + 1 can be interpreted as the worldvolume
of a probe Dp-brane, and the radial time as the position of the probe Dp-brane. We
also reduce 11d supergravity on S 7 and S 4 , repeat the above steps, and obtain the
M2 and M5 brane effective actions as solutions to the HJ equations, respectively. We
find that these solutions to the HJ equation reproduce the supergravity solutions of a
stack of M2-branes and a stack of the M2M5 bound states, respectively. Furthermore,
by using the SL(2, R) symmetry in type IIB supergravity and the relation of type IIA
supergravity with 11d supergravity, we obtain solutions to the HJ equations that reproduce
the supergravity solutions representing (p, q) strings and (p, q) 5-branes in type IIB
supergravity and NS 5-branes in type IIA supergravity. Note that the near-horizon limit of
supergravity backgrounds reproduced by M2-branes, the M2M5 bound states and NS 5branes are conjectured to be dual to three-dimensional N = 8 superconformal field theory
(a noncommutative version of) six-dimensional N = (2, 0) superconformal field theory
and Little String Theory, respectively [4,5,7].
As we discuss below, the fact that our solutions to the HJ equations are the on-shell
actions around the supergravity backgrounds which conjectured to be dual to various gauge
theories motivates us to further investigate the subject in the present paper.
Indeed, our findings can be applied to the study of the gauge/gravity correspondence.
A well-understood example of the gauge/gravity correspondence is the AdS/CFT correspondence [5,6]; in particular, the correspondence between N = 4 super-YangMills at
the conformally invariant point and type IIB supergravity on AdS S 5 . It is relevant to
investigate whether this kind of correspondence can be extended to N = 4 super-Yang
295
Mills in the Coulomb branch [5,8] or four-dimensional less supersymmetric (N = 0, 1, 2)

gauge theories [9] or higher- (lower-)dimensional supersymmetric gauge theories [5,7,10].
Another relevant problem to be studied from the viewpoint of the gauge/gravity correspondence is a quantum theory of NCYM. Classical aspects of NCYM such as noncommutative instantons [2,11] are well-understood while little is known about quantum aspects. In particular, renormalizability of NCYM has not been established perturbatively
or non-perturbatively. Note that the authors of Ref. [12] verified the renormalizability of
two-dimensional bosonic NCYM by performing a numerical simulation of its lattice version. As we describe shortly, solving the HJ equation and obtaining the on-shell action in
supergravity is doubly important for studying the above issues in the gauge/gravity correspondence. In order to study the gauge/gravity correspondence for the less supersymmetric
gauge theories, we must generalize our analysis in the present paper; we should reduce supergravities in more complicated ways and search for solutions to the HJ equations of
these reduced gravities that reproduce the supergraviy solutions conjectured to be dual to
the less supersymmetric gauge theories.
One application of the HJ method in supergravity to the study of the gauge/gravity
correspondence is to compare the on-shell action of supergravity with the effective action
of the dual gauge theory. Suppose that supergravity on a certain background corresponds to
a large N (noncommutative) gauge theory in which one of the Higgs fields has a nontrivial
vacuum expectation value (vev) and the U (N) gauge symmetry is spontaneously broken
to U (1) SU(N 1). Then, if we interpret the radial time as a position of the probe
D-brane, the on-shell action of supergravity around this background should coincide with
the effective action of the gauge theory via the identification of the radial time with the vev
of the Higgs field. Thus the HJ method is useful for checking this case of the gauge/gravity
correspondence. It is actually conjectured in Ref. [13] that the effective action of N = 4
super-YangMills in the Coulomb branch takes the form of the D3-brane effective action
in the t Hooft limit. It is important to perform a similar calculation of the effective action
in NCYM and to compare the result with the on-shell action obtained in this paper.
The other application is the study of the holographic renormalization group, which is
also useful for establishing the gauge/gravity correspondence. The authors of Ref. [14]
derived the renormalization group equation in the dual gauge theory from the HJ equation
in supergravity. In particular, they found that in their simple examples the lowest term in
the derivative expansion of the on-shell action in supergravity plays a role of the counter
terms and gives the beta functions and the anomalous dimensions in the dual gauge theory.
In this context, the radial time is interpreted as the renormalization scale in the dual gauge
theory. Note that the solutions found in this paper are also the lowest term in the derivative
expansion. One can check whether supergravity on a certain background corresponds to
a large N gauge theory by comparing the beta functions and the anomalous dimensions
given by the on-shell action around the background with those in the gauge theory. Also,
one can examine the structure of the renormalization of NCYM through the holographic
renormalization group. Although it is not obvious whether the lowest term in the on-shell
action is sufficient in more complicated cases we are interested in, our results are at least a
first step to the study of the holographic renormalization group in these cases.
Another application of the HJ method is searching for new solutions and classifying
the solutions in supergravity. Using the solution to the HJ equation obtained in the
296
present paper, we expect to be able to obtain new solutions in supergravity reduced on

higher-dimensional spheres under the condition that the fields depend only on the radial
coordinate. That is, while the supergravity solution representing Dp-branes in a B2 field
is obtained by T-dualizing a tilted smeared D(p 1)-brane solution [15], we can search
supergravity solutions that cannot be obtained by such a T-dualization, as is discussed
below. If we find a solution to the HJ equation of supergravity reduced in a different
way, we expect to be able to obtain a different kind of new solutions of supergravity.
Furthermore, if we find a complete solution to the HJ equation under a certain condition,
we can classify the solutions in supergravity under the condition. Hence, our analysis in
the present paper should be a first step to the classification of the supergravity solutions.
As is well known, a complete solution to the HJ equation that includes as many
arbitrary constants as the number of the degrees of freedom is a generator of the canonical
transformation that makes the new Hamiltonian vanish. If one finds a complete solution to
the HJ equation, one can represent the coordinates as functions of time for an arbitrary
initial condition. Namely, the problem can be completely solved. Although our solution
to the HJ equation is not a complete solution, it includes several arbitrary constants.
We can reduce the original equations of motion in supergravity, which are the second
order differential equations, to the first order ones by using the solution, and obtain as
many conserved quantities as the number of the arbitrary constants. We may solve the first
order equations to find a new solution of supergravity, utilizing these conserved quantities.
Alternatively, we can search for a new solution to the HJ equation that generalizes the
present solution and includes more arbitrary constants so that we may find a new solution
more easily.
The present paper is organized as follows. In Section 2, we develop the Hamilton
Jacobi method in general constrained systems. In Section 3, we perform reductions of
supergravities on higher-dimensional spheres. In Section 4, we develop the canonical
formalism for the reduced gravities obtained in Section 3 to derive the HJ equations.
In Section 5, we find that the Dp-brane effective action is a solution to the HJ equation
of the reduced gravity and reproduces the supergravity solution of a stack of D3-branes
in a B2 field. Section 6 is devoted to the similar calculation in the cases of M2-brane and
M5-brane. In Section 7, using the SL(2, R) symmetry in type IIB supergravity and the
relation between 11d supergravity and type IIA supergravity, we obtain solutions to the
HJ equations that reproduce the supergravity solutions of (p, q) string and (p, q) 5-brane
in type IIB supergravity and of NS 5-brane in type IIA supergravity. Section 8 is devoted
to discussion and perspective. In Appendix A, the equations of motion in supergravities
are listed. In Appendix B, we write down the ansatz for the fields made in performing a
reduction of type IIA (IIB) supergravity on S 8p .
2. HJ method in constrained systems

Let us consider a classical system with n degrees of freedom and k first class constraints.
Suppose the action of the system is given in the canonical form:
t
I=
297

d pi qi H (p1 , . . . , pn , q1 , . . . , qn , ) a fa (p1 , . . . , pn , q1 , . . . , qn , ) ,
t0
(2.1)
where i runs from 1 to n, a runs from 1 to k, H is the Hamiltonian and the a are Lagrange
multipliers. By variating the action with respect to qi , pi and a , one obtains the following
equations of motion:
H
fa
+ a
,
pi
pi
fa
H
pi =
a
,
qi
qi
fa (p1 , . . . , pn , q1 , . . . , qn , ) = 0.
qi =
(2.2)
Let qi = qi ( ) and pi = p i ( ) be a solution to the above equations of motion which

satisfies the boundary conditions qi (t) = xi . Substituting this classical solution to the action
gives the on-shell action, which can be regarded as a function of the boundary positions xi
and the final time t
t
S(x1 , . . . , xn , t) =

d p i q i H (p 1 , . . . , pn , q1 , . . . , qn , ) .
(2.3)
t0
The variation of S with respect to xi and t is given by

q(t),
t t
S = pi (t)q i (t) H p(t),
t
+
t0

H (p,
q,
)
H (p,
q,
)
d p i q i + p i q i
p i
qi .
p i
qi
(2.4)
Using the equations of motion (2.2) and integrating the right-hand side of (2.4) partially,
one obtains

S = pi (t)q i (t) H p(t),
q(t),
t t + pi (t) qi (t)
t
+
d a
t0

fa (p,
q,
)
q,
)
fa (p,
pi +
qi .
pi
qi
(2.5)
Here the last line in (2.5) vanishes, since

0 = fa (p + p,
q + q,
) = fa (p,
q,
).
Noting that qi (t + t) + qi (t + t) = xi + xi , one obtains qi (t) = xi q i (t)t.
Therefore, (2.5) reduces to

S = H p(t),
x, t t + p i (t)xi ,
(2.6)
298
which is equivalent to the equations

S
= H p(t),
x, t ,
t
S
= p(t)
i.
xi
(2.7)
Finally, one obtains the equations satisfied by the on-shell action S, the HJ equations:

S
S
S
,...,
, x1 , . . . , xn , t = 0,
+H
t
x1
xn

S
S
fa
(2.8)
,...,
, x1 , . . . , xn , t = 0.
x1
xn
Suppose a solution to (2.8) is given. Then, the canonical momenta are represented in
terms of xi by using the second equation in (2.7). It follows that the equations of motion
(2.2) reduce to a set of the first order differential equations only for qi . Thus one can
simplify the problem of solving the equations of motion.
It is well known that the HJ equation is in general more powerful as seen in below.
First, note that if S is a solution to the above equations, S + is also a solution to it, where
is a arbitrary constant. Let the above solution possess l arbitrary constants 1 , . . . , l
that are not the trivial additive constant . Then, the following quantities are conserved
quantities:
s =
S
s
(s = 1, . . . , l).
(2.9)
In fact, on one hand,

2S
d
2S
s =
+
xi .
dt
s t s xi
(2.10)
On the other hand, from (2.8) and the arbitrariness of s , we obtain

S
2S
H
2S
0=
,
+H =
+
s t
s t (S/xi ) s xi
0 = a
2S
fa
fa = a
.
s
(S/xi ) s xi
(2.11)
Summing the two equations in (2.11) and using (2.2) and (2.10) gives
d
s = 0.
dt
(2.12)
In particular, when l = n, it follows that i and i are new canonical momentum and
new coordinates which are obtained from the canonical transformations generated by S,
respectively, and are constant with respect to the time. One can completely solve the
problem by using the second equation in (2.7) and (2.9).
299
3. Reductions of type IIA (IIB) and 11d supergravities on higher-dimensional

spheres
In this section, we perform reductions of supergravities on higher-dimensional spheres.
We will regard the fixed-time surface as the worldvolume of the p-brane (Dp-brane or
M2-brane or M5-brane) later. Hence, we will reduce supergravity to a (p + 2)-dimensional
gravity.
First, in order to fix the conventions, we write down the actions of type IIA (IIB) and 11d
supergravities. In this paper, we drop the fermionic degrees of freedom consistently. In the
1
following equations, |Kq |2 = q!
KM1 Mq K M1 Mq for a q-form Kq , where the appropriate
metric is used for the contractions, and Cp+1 is the RR (p + 1)-form. The bosonic part
of type IIA supergravity is given by

1
1
IIIA = 2
d 10 X G e2 RG + 4M M |H3 |2
2
210

1
1 2
|F2 |2 |F
4|
2
2

1
B2 F4 F4 ,
2
(3.1)
410
where
H3 = dB2 ,
Fp+2 = dCp+1
(p = 0, 2),
4 = F4 C1 H3 .
F
(3.2)
The bosonic part of type IIB supergravity is given by

1
1
10
2
M
2
RG + 4M |H3 |
d X G e
IIIB = 2
2
210

1
1
1
2
2
2

|F1 | |F3 | |F5 |
2
2
4

1
C4 H3 F3 ,
+ 2
410
(3.3)
where
H3 = dB2 ,
Fp+2 = dCp+1
(p = 1, 1, 3),
3 = F3 + C0 H3 ,
F
5 = F5 + C2 H3 .
F
(3.4)
One must also impose the self-duality condition

5 = F
5
F
(3.5)
300
on the equations of motion derived from the above action. The bosonic part of 11d
supergravity is given by

1
1 (M) 2
1
11
I11 = 2
d X G RG F4
A3 F4(M) F4(M) ,
2
2
211
1211
(3.6)
where
F4(M) = dA3 .
We list all of the equations of motion and the Bianchi identities in these supergravities in
Appendix A.
Let us consider a reduction of type IIA (IIB) supergravity on S 8p (p = 0, . . . , 7),
where p takes 0, 2, 4, 6 for type IIA supergravity and 1, 3, 5, 7 for type IIB supergravity.
We split the ten-dimensional coordinates XM (M = 0, 1, . . . , 9) into two parts, as XM =
( , i ) ( = 0, . . . , p + 1, i = 1, . . . , 8 p), where the are (p + 2)-dimensional
coordinates and the i parametrize S 8p . We make the following ansatz for the tendimensional metric, which preserves the (p + 2)-dimensional general covariance:
ds10 = GMN dXM dXN = h ( ) d d + e( )/2 d8p .
We assume that the dilaton depends only on
= ( ).
(3.7)
:1
(3.8)
The ansatzes for the other fields are summarized in Appendix A. Here, as an example, we
perform the reduction for the p = 6 case explicitly. The reductions for the other cases can
be performed in the same way as the p = 6 case. The ansatzes for the other fields in the
p = 6 case are
1
1
H ( ) d d d + d ( )/i1 i2 d di1 di2 ,
3!
2!
1
1 /2 1 8
F1 8 ( )i1 i2 di1 di2 ,
F2 = F1 2 ( ) d 1 d 2
e
2!
2!8!
4 = 1 F
( ) d 1 d 4
F
4! 1 4
1 /2
1 6 ( )i i d 1 d 2 di1 di2 .
e 1 6 1 2 F
+
(3.9)
1 2
2!2!6!
By substituting (3.7), (3.8) and (3.9) into the equations of motion and the Bianchi identities
in type IIA supergravity in Appendix A, we obtain the following equations in eight
dimensions
H3 =
1
1
1
1
Rh + 2(8)(8) (8) (8) H12 H 1 2 e d d
2
8
4
2
1 It is sufficient for the purpose of this paper to assume that the fields depend only on p+1 (= r). However,
we consider more general ansatzes such as (3.7) and (3.8) for further developments.
301

1
1
1 5 F
1 2 3 + 1 F
1 5
e2 F F + F
1 2 3 F
2
3!
5!

1
1 7

+ F1 7 F
7!

1
+ e2 h |F2 |2 + |F4 |2 + |F6 |2 + |F8 |2 = 0,
(3.10)
4
3
1
Rh + 4 (8)2 (8)2 4 + 2 |H3 |2
8
2
1
2
e d d + e 2 R (S ) = 0,
(3.11)
2
1
1
1
Rh + 4 (8)2 (8)2 4 + |H3 |2
2
8
2

1
1
+ e d d e2 |F2 |2 + |F4 |2 + |F6 |2 + |F8 |2 = 0,
(3.12)
2
2

1
1 2 F1 2 + 1 e 2 F
1 4 F
1 4
(8) e2+ 2 H + e 2 F
2
4!
1 1 6
+ e2F
(3.13)
F1 6 = 0,
6!

1
1
3 8 1 F
= 0, (3.14)
F
(8) e2 2 d + 1 8 + F1 2 F
2
6!
4!4! 1 4 5 8
1
(8)
1 2 3 H1 2 3 + 1 1 7 d1 F
2 7 = 0,
e 2 F + e 2 F
(3.15)
3!
6!
1 1 2 3
1
(8)
2 5 = 0,
e 2 F
(3.16)
+ e2F
H1 2 3 1 5 d1 F
3!
4!
1
1 5 + e2F
1 5 1 2 3 H1 2 3 + 1 1 5 1 2 3 d1 F1 2 = 0,
(8) e 2 F
3!
2
(3.17)
(8) 1 7
= 0,
e2F
(3.18)
[1 H2 3 4 ] = 0,
(3.19)
[1 d2 ] = 0,
(3.20)
[1 F1 2 ] = 0,
(3.21)
2 5 ] + 5! F[1 2 H3 4 5 ] = 0,
5[1 F
2!3!
2 7 ] + 7! F
[ H ] = 0.
7[1 F
3!4! 1 4 5 5 7
(3.22)
(3.23)
One can easily verify that these equations are derived from I8 , which we will describe
below. Applying this procedure to the other p case, we obtain Ip+2 . We have thus reduced
type IIA (IIB) supergravity on S 8p and obtain the (p + 2)-dimensional gravities. These
reductions are consistent truncations in the sense that every solution of Ip+2 can be lifted
to a solution of type IIA (IIB) supergravity.
302
Let us write down Ip+2 explicitly. We first define Ip+2 by

Ip+2 = d p+2 h

(8 p)(7 p)
8p
2+ 8p
4

Rh + e 2 R (S ) + 4 +
e
16

1 8p
2
1
(8 p) |H3 |2 e 4
|Fn | ,
2
2
n
(3.24)
where
n = Fn + sCn3 H3 ,
F

s=
1, for type IIA (p = 0, 2, 4, 6),

1,
for type IIB (p = 1, 3, 5, 7).
Here n takes 2, 4, . . . , p + 2 for type IIA supergravity and 1, 3, . . . , p + 2 for type IIB
supergravity. I2 is obtained by setting H3 = 0 in (3.24). Then,
Ip+2 = Ip+2 , for p = 0, 1, . . . , 5,

1
1
8
2

I8 = I8 + d h e |d1 | + b0 F2 F6 F4 F4 ,
2
2

1
3 F
5 F1 F
7 ),
I9 = I9 + d 9 h e 2 |d2 |2 + b1 (F
2
(3.25)
(3.26)
(3.27)
where d1 = db0 and d2 = db1 .

We perform reductions of 11d supergravity on S 7 and S 4 in the same way as those of
type IIA (IIB) supergravity. For a S 7 reduction, we make the following ansatzes:
ds11 = h ( ) d d + e
(M)
F4
()
2
d7 ,
1 (M)
F
( ) d 1 d 4 ,
4! 1 4
(3.28)
where , run form 0 to 3. Then, we obtain a 4-dimensional gravity, which is a consistent

truncation of 11d supergravity,

2
7
21
1
7
I4(M) = d 4 h e 4 Rh + e 2 R (S ) + F4(M) ,
(3.29)
8
2
(M)
where F4
= dA3 . For a S 4 reduction, we make the following ansatzes:
ds11 = h ( ) d d + e
(M)
F4
()
2
d4 ,
= F(M)
( ) d 1 d 4
1 4
1 1 7 (M)
e
+
F1 7 ( )i1 i4 di1 di4 ,
4!7!
(3.30)
303
where , run form 0 to 6. Then, we obtain a 7-dimensional gravity, which is also a

consistent truncation of 11d supergravity,
(M)
I7

=
3
4
d h e Rh + e 2 R (S ) +
4
4
1 (M) 2
F
2 4

1 (M) 2
F
,
2 7
(3.31)
(M) = dA6 1 A3 F (M) .

where F
7
4
2
4. Canonical formalism and the HJ equations in the reduced gravities

(M)
(M)
In this section, we develop the canonical formalism for Ip+2 , I4 and I7

in the previous section. First we rename the (p + 2)-dimensional coordinates:
= x
( = 0, . . . , p),
obtained
p+1 = r.
(M)
(M)
Here p takes 2 and 5 for I4 and I7 , respectively. Adopting r as time, we make the
ADM decomposition for the (p + 2)-dimensional metric.
2
= h d d
dsp+2

= n2 + g n n dr 2 + 2n dr dx + g dx dx ,
(4.1)
where n and n are the lapse function and the shift function, respectively. Hence force ,
run from 0 to p.
In what follows, we consider a boundary surface specified by r = const and impose
the Dirichlet condition for the fields on the boundary. Here we need to add the Gibbons
Hawking term [16] to the actions, which is defined on the boundary and ensures that the
Dirichlet condition can be imposed consistently. Then, the (p +2)-dimensional action Ip+2
with the GibbonsHawking term on the boundary can be expressed in the canonical form
as follows. For p = 0, 1, . . . , 5,

Ip+2 =
dr d p+1 x
1
n1
r g + r + r + B r B +
Cn1
r C1 n1
nH n H
Br GB
n

Cr1 n2 GC1n1 n2

(4.2)
304
with
H = e2
8p
4

2 1 2 1
4
2 +
+ + +
2
2
8p
2
(n 1)(n 2)
1 n3
C1 n3 Cn1
+ B
2
n3
8p
4
(n 1)! 2
1
n1
Cn1
L,
2
n
H = 2 + + + B H

(n 1)!
C 1 n4 Hn3 n2 n1 ,
+
C1n1 n1 F 1 n1 + s
(n 4)!3!
n4
GB
= 2 B ,
n2
GC1n1
n2
1
= (n 1) Cn1
(n + 1)! 1 n2 1 2 3
H 1 2 3 ,
(n 2)!3! Cn+1
(4.3)
where
L = e2+
8p
4

8p
4
Rg + 4
2
(8 p)(9 p)
1
+ (8 p) |H3 |2
16
2
6p
1 8p
2
8p
e 4
|Fn | + e2+ 4 R (S ) ,
2
n
(4.4)
I8 and I9 with the GibbonsHawking terms can be rewritten in the form (4.2) with p = 6
and p = 7, respectively, up to the terms including b0 and b1 . The differences will turn out
not to be relevant for our purpose, so that we do not write down the precise canonical forms
of I8 and I9 here.
Note that (4.2) takes the form of (2.1). Indeed, n, n and Cr1 n2 play the roles
of a in (2.1). They give the constraints, H = 0, H = 0 and GCn1 = 0, which are
called the Hamiltonian constraint, the momentum constraint and the Gauss law constraint,
respectively, and correspond to fa = 0 in (2.2). Note that (4.2) has no analogue of H
in (2.1). It follows from the arguments in Section 2 that the HJ equations of this system
are given by
Sp+1
= 0,
r
H = 0,
H = 0,
GCn1 = 0
(4.5)
305
with
S
1
,
(x) =
g(x) g (x)
S
1
,
(x) =
g(x) (x)
n1
1
Cn1
(x) =
(x) =
S
1
,
g(x) (x)
S
1
B (x) =
,
g(x) B (x)
S
1
,
C
g(x) 1 n1 (x)
(4.6)
where r is the boundary value of r, and g (x), (x), . . . , C1 n1 (x) are the boundary
values of the corresponding fields. The first equation in (4.5) indicates that Sp+1 does not
depend on the boundary time explicitly. The last three equations in (4.5) give functional
differential equations for Sp+1 . The third and fourth ones imply that Sp+1 must be invariant
under the diffeomorphism in p + 1 dimensions and the U (1) gauge transformations (see
Appendix C in Ref. [1]). The second one is a nontrivial equation that can determine the
form of Sp+1 . Hereafter, we call this equation the HamiltonJacobi equation.
The canonical form of I4(M) with the GibbonsHawking term is

(M)
I4 = d 4 g r g + r + A3 r A nH n H

Ar GA3
(4.7)
with

2 1
2
2

7
8
4
H = e 4 + 2 + 3 A3
L,
9
63
9
H = 2 + + A3 F (M) ,
GA3 = 3 A3 ,
where
L=e
7
4

5
7
1 (M) 2
7
7
R F4
+ e 4 R (S ) .
2
2
2
(M)
(4.8)
(4.9)
The canonical form of I7 with the GibbonsHawking term is

(M)
I7 = d 7 g r g + r + A3 r A + A61 6 r A1 6

nH n H Ar GA3 Ar1 5 GA16 5
(4.10)
with
H = e

2 1 2 5 2 4
+
9
9
9

2 6! 1 6 2
1 2 3
1 2 3 1 2 3
A6
+ 3 A3
+ 10A6
A 1 2 3 +
L,
2
306
H = 2 + + A3 F (M)

15

+ F (M) 1 6 A 1 2 F(M)
A16 6 ,
3 6
2
GA3 = 3 A3 ,
5
GA16
= 6 A6 1
(4.11)
where

5
L = e R 2
4
(M)
The HJ equations for I4
(M)
and I7
1 (M) 2
F
2 4
1 (M) 2
7
F7
+ e 2 R (S ) .
2
(4.12)
are derived in the same way as the one for Ip+2 .
5. Dp-brane effective action as a solution to the HJ equation

5.1. Dp-brane effective action as a solution
In this subsection, we find a solution to the HJ equation obtained in the previous
(0)
section. We assume that the fields are constant on the fixed-time surface. Let Sp+1 be a
solution to the HJ equation under this assumption. We can drop b0 and b1 consistently in
(0)
(0)
the HJ equations for S7 and S8 . That is, after this simplification, the HJ equations for
S7(0) and S8(0) coincide with the ones derived from (4.2) with p = 6 and p = 7, respectively.
(0)
We see from (4.3), (4.4), (4.5) and (4.6) that Sp+1 satisfies the equation
e
2 8p
4
+ e

(0)
(0)
(0)
(0)
1 Sp+1 2 1
1 Sp+1 1 Sp+1 1
1 Sp+1 2
+
+ g
g g
2
g g g
2
g

(0) 2
(0)
(0)
1 Sp+1
4
1 Sp+1 1 Sp+1
+
+
8p
g
g
g

(0)
(0)
2
Sp+1
1 Sp+1
(n 1)(n 2)
1
C1 n3
+
g B
2
g C1 n3
n
8p
4
2
(0)
(n 1)! 1
Sp+1
6p
8p
+ e2+ 4 R (S ) = 0.
2
g
C
1 n1
n
(5.1)
In what follows, we show that the form

(0)
c
BI
WZ
= Sp+1
+ Sp+1
+ Sp+1
+ p+1
Sp+1
is a solution to (5.1), with

7p
c
= p+1 d p+1 x ge2+ 4 ,
Sp+1
(5.2)

BI
Sp+1
= p+1
WZ
Sp+1
= p+1
307

d p+1 xe det(g + F ),
Cn1 eF

1
Cp+1 + Cp1 F + Cp3 F F + ,
= p+1
2
(p+1)
(5.3)
(p+1)
is an arbitrary constant antisymmetric tensor, and

where F = B + F , F
BI + S WZ is the
p+1 is an arbitrary constant. As we discussed in Section 6 in Ref. [1], Sp+1
p+1
c
effective action of a probe Dp-brane while Sp+1 should be interpreted as the vacuum to
vacuum amplitude and does not contribute to the effective action of the probe Dp-brane.
(p+1)
is interpreted as the U (1) gauge field strength in the world-volume.
Furthermore, F
Noting that
(0)
(0)
BI
Sp+1
1
1 Sp+1
1 Sp+1
(n 1)(n 2)
C1 n3
g B
2
g C1 n3
g B
n
and
WZ
1 Sp+1
= 0,
g g
one can see that the left-hand side of (5.1) can be decomposed into the four parts

8p
8p
L.H.S. of (5.1) = e2 4 (1) + (2) + (3) + e 4 (4)
+ e2+
with
6p
4
R (S
8p )
(5.4)

(1) =

c
c
c
c
1 Sp+1 2 1
1 Sp+1 2
1 Sp+1 1 Sp+1 1
+ g
+
g g
2
g g g
2
g

2
c
c
c
4
1 Sp+1 1 Sp+1
1 Sp+1
,
+
+
8p
g
g
g
c
BI
BI
BI
1 Sp+1 1 Sp+1 1
1 Sp+1 1 Sp+1
(2) = 2g g
+ g
g g g g
2
g g g
c
BI
1 Sp+1 1 Sp+1
1 Sp+1 1 Sp+1
1
+
+ g
2
g g g
g
g
BI
1 Sp+1 1 Sp+1
+
,
g
g

BI
BI
BI
BI
1 Sp+1 2 1
1 Sp+1 1 Sp+1 1
1 Sp+1 2
(3) =
+
+ g
g g
2
g g g
2
g

2
BI
1 Sp+1
+
,
g B
308
2
WZ
(n 1)! 1
Sp+1
(4) =
.
2
g C1 n1
n
(5.5)
(1) and (4) are easily calculated as

(1) =
7p
7p 2
p+1 e4+ 2 ,
4(8 p)
p+1
[ 2 ]
(2k)! [1 2
1 2
(4) = p+1
2k2k ]
2
4k (k!)2 1 2
k=0
F1 2 F2k1 2k F 1 2 F 2k1 2k .
(5.6)
In order to calculate (2) and (3), we introduce the (p + 1) (p + 1) matrices G and B:

(G) = g ,
(B) = F .
Then, we have

c
1 Sp+1 1
2+ 7p
1
4
= p+1 e
,
g g
2
G
c
7p
1 Sp+1
= 2p+1 e2+ 4 ,
7p
1 Sp+1 7 p
=
p+1 e2+ 4 ,
g
4

BI
1
1 Sp+1 1
det(G + B)
1
G
= p+1 e
,
g g
2
det G
G+B GB

BI
1
1
1 Sp+1 1
det(G + B)
B
= p+1 e
,
g B
2
det G
G+B G B

BI
1 Sp+1
det(G + B)
= p+1 e
.
g
det G
(5.7)
Using this notation, we can express each term in (2) and (3) in terms of the trace of the
(p + 1) (p + 1) matrix and calculate (2) and (3) as follows:

7p
det(G + B)
(2) = p+1 p+1 e3+ 4
det G

1
1
1
1
p+1
tr
G
G
G
2
G G+B G B
4

1
1
7p
1
G
G +2
tr
= 0,
2
G +B G B
4
309
1 2
det(G + B)
e2
(3) = p+1
4
det G

1
1
1
1
1
1
G
G
G
G tr
G
G
tr
G+B G B G+B GB
G +B G B

1
1
1
1
+ 2 tr
B
G
B
G
G +B G B G +B G B
1 2
det(G + B)
= p+1
e2
2
det G
p+1
[ 2 ]
(2k)!
1 2
= p+1
e2
[1 2 2k2k ]
2
4k (k!)2 1 2
k=0
F1 2 F2k1 2k F 1 2 . . . F 2k1 2k .
(5.8)
(0)
From (5.1), (5.4), (5.6) and (5.8), we conclude that Sp+1 satisfies the HJ equation (5.1) if
2
p+1
=
4(8 p) (S 8p )
R
= 4(8 p)2
7p
2
2
and p+1
= p+1
.
(5.9)
5.2. Dp-brane in a B2 field

(0)
In this subsection, we see that Sp+1 obtained in the previous subsection reproduces the
supergravity solution representing a stack of Dp-branes in a constant B2 field. First, we
examine the cases in which 2 p 6. For simplicity, let us consider the supergravity
solutions with only Bp1p nonvanishing. These supergravity solutions were constructed in
Ref. [15], and they are also solutions of Ip+2 taking the following forms:

1
1
2
dsp+2
= f 2 dx dx + hab dx a dx b + f 2 dr 2 ,
2
e2 = gst
f
p3
2
e 2 = r 2f 2 ,
h,
1
sin f 1 ,
C01p2 = (1)p+1 gst
Bp1 p = tan f 1 h,
1
C01p = (1)p+1 gst
cos f 1 h,
(5.10)
where
,
= 0, 1, . . . , p 2,
f =1+

Q
r 7p
a, b = p 1, p,
h1 = sin2 f 1 + cos2 .
(5.11)
Note that these supergravity solutions preserve 16 supersymmetries, and reduce to the
ordinary Dp-brane solutions when = 0.
By varying Ip+2 with respect to the canonical momenta, we obtain the relations between
the canonical momenta and the r-derivatives of the fields. Using these relations, we
310
calculate the values of the canonical momenta on the boundary specified by r = r :

p5

f 4
1 8p
7p
00 = 11 = = p2p2 = 2
r f ,
(8 p)r
f + r
2
gst h
p5

f 4
1
Ip+2
7p
2
8p
(8 p)r
f + cos r
hr f ,
p1p1 = pp = 2
2
gst

f 4
4(p 8)r 7p f r 8p r f ,
2
gst h
(7 p)(8 p) 7p p+1
Ip+2
=
r
f 4 ,
Bp1p
= 0,
2h
2gst
p3
(1)p+1 sin 8p p+1

f 4 r f ,
r
(p 1)! gst
(1)p+1 cos 8p p+1
Cp+1 01p =
r
hf 4 r f ,
(p + 1)! gst
Cp1 01p2 =
(5.12)
where

Q
f = 1 + 7p ,
r
h 1 = sin2 f1 + cos2 .
(0)
On the other hand, Sp+1
gives the following canonical momenta:
(0)
1 Sp+1
= g g
g g

1
det(G + B)
1
G
G
,
G
det G
G + B G B

7p
det(G + B)
= 2p+1 e2+ 4 p+1 e
,
det G
7p
1
1
= p+1 e2+ 4 g + p+1 e
2
2
(0)
1 Sp+1
=
g
(0)
7p
1 Sp+1 7 p
=
p+1 e2+ 4 ,
=
g
4
(0)
1 Sp+1
= g g
g B

1
1
1
det(G + B)
B
G
= p+1 e
G
2
det G
G + B G B
[ p+1
2 ]
+ p+1 1 2 p1
k=0
1
C 1 2 p+12k
(p + 1 2k)!2k (k 1)!
F p+12k+1 p+12k+2 F p2 p1 ,
311
Cn1 1 2 ...n1
(0)
Sp+1
1
= g1 1 g2 2 gn1 n1
g C1 2 n1
= p+1
1
(n 1)!2
pn+2
2
pn+2 1 2 p+1 F
!
2
n n+1
F p p+1 .
(5.13)
We substitute the values of the fields in (5.10) into the right-hand sides of (5.13), setting
(p+1)
F
= 0.
It can be verified that the right-hand sides of (5.13) reproduce the right-hand sides of (5.12)
if
p+1 = 16 2p
and p+1 = (1)p p+1 =
cos
(p 7)Q
.
gst
(5.14)
These conditions are consistent with (5.9).

(0)
Next, we compare the value of the on-shell action with that of Sp+1
directly. Substituting
the values of the fields with r = r in (5.10) into (5.3), we obtain
7p
c
Sp+1
=
p+1 Vp+1 r0
,
2
gst
p+1 Vp+1 1
BI
Sp+1
f ,
=
gst cos 0
p+1 Vp+1 1
WZ
f ,
Sp+1
= (1)p+1
gst cos 0

where Vp+1 = d p+1 x. Here, we set
(5.15)
p+1 = 0.
Then, it follows from (5.2), (5.14) and (5.15) that
7p
(0)
c
BI
WZ
+ Sp+1
+ Sp+1
=
Sp+1 = Sp+1
(16 2p)Vp+1 r0
2
gst
(5.16)
We calculate the values of the on-shell actions for (5.10) by substituting (5.12) with r
replaced by r into (4.2). Noting that the constraints in (4.2) are satisfied on shell, we
(0)
reproduce the value of Sp+1
for (5.10) as follows:
r
on-shell
Ip+2
dr d p+1 x
0
g r g + r + r + B r B

1 n1
+
Cn1
r C1 n1
n
312
=
=
2(7 p)(8 p)Vp+1

2
gst
r
dr r 6p
0
2(8 p)Vp+1
2
gst
r 7p .
(p+1)
(0)
with F
Thus, we have shown that Sp+1
(5.17)
= 0 and p+1 = 0 reproduces the supergravity
(0)
solution (5.10) when p+1 and p+1 take the values in (5.14). We also verified that S1
(0)
and S2 reproduce the supergravity solutions representing a stack of D0-branes and of D1branes, respectively. Note that S4(0) also reproduces the near-horizon limit of (5.10) with
p = 3 [1], which is conjectured to be dual to NCYM [3,4].
6. Effective actions of M2-brane and M5-brane as solutions to the HJ equations

6.1. M2-brane case
We assume again that the fields are constant on the fixed-time surface. Let us denote
(0)
a solution to the HJ equation under this assumption by SM2 . It follows from (4.7), (4.8)
(0)
and (4.9) that SM2
satisfies the equation

(0)
(0)
1 SM2 2
1 SM2 2
1
8
g
9
g g
63
g

(0)
(0)
(0)
1 SM2 2
1 SM2 1 SM2
4
7
+3
g
+ e3 R (S ) = 0.
9
g g g
g A
(0) 2
1 SM2
g g
(6.1)
One can easily verify that the form

(0)
c
NG
WZ
SM2 = SM2
+ SM2
+ SM2
+ M2
with

c
= M2
SM2
NG
SM2
= M2
WZ
SM2
= M2
d 3x

d 3x
(6.2)
g e 2 ,
g,
A3 ,
(6.3)
where M2 is an arbitrary constant, satisfies the HJ equation if

2
M2
=
14 (S 7 )
2
2
R
= 196 and M2
= M2
.
3
(6.4)
NG
WZ
Note that SM2
+ SM2
is interpreted as a probe M2-brane effective action as in the case of
the Dp-brane.
313
(M)
The supergravity solution representing a stack of M2-brane is also a solution of I4

which is given by
ds42 = f 3 dx dx + f 3 dr 2 ,
2
f =1+

Q
,
r6
A012 = f 1 ,
e 2 = r 2f 3 ,
(6.5)
(0)
with M2 = 0 reproduces this solution
where , run from 0 to 3. We verified that SM2
when

M2 = M2 = 6Q,
M2 = 14,
(6.6)
which is consistent with (6.4).

6.2. M5-brane case
We adopt again the assumption that the fields are constant on the fixed-time surface. Let
(0)
SM5
be a solution to the HJ equation under this assumption. It follows from (4.10), (4.11)
(0)
and (4.12) that SM5 satisfies the equation
(0) 2
1 SM5
g g

(0)
(0)
1 SM5 2
1 SM5 2 5
1
g
+
9
g g
9
g
(0)
(0)
1 SM5 1 SM5
4
g
9
g g g

2
(0)
(0)
SM5
1 SM5
1
+3
+ 10A1 2 3
g A
g A1 2 3

(0)
2
SM5
3
6!
4
+
+ e 2 R (S ) = 0.
2 A1 6
(6.7)
Let us consider the following form:

(0)
c
BI
WZ
= SM5
+ SM5
+ SM5
+ M5 ,
SM5
with

c
SM5
= M5
d 6x
BI
SM5
= M5
d 6x
(6.8)
g e 4 ,
2
1
1
1
F F F F F F ,
1 + F F +
12
288
96

1
WZ
SM5
(6.9)
= M5
A6 + A3 F3 ,
2
314

(M5)
(M5)
where F = A + F , F is an arbitrary constant completely antisymmetric

tensor, and M5 is an arbitrary constant. We verified that (6.8) satisfies (6.7), up to the
constraint
1 1 6
F4 5 6
3!
= 12
A12 3

2
1
1
1
F F F F F F ,
1 + F F +
12
288
96
(6.10)
if
2
=
M5
16 (S 4 )
R
= 64 and M5 = M5 .
3
(6.11)
The equations of motion satisfied by M5-brane are determined by the spacetime

supersymmetry and the kappa symmetry [1820]. These equations of motion are equivalent
to the equations derived from the M5-brane effective action and the nonlinear self-duality
condition [21,22].2 This M5-brane effective action and the nonlinear self-duality condition
BI + S WZ and (6.10) in our case in which the worldvolume of the M5-brane is
reduce to SM5
M5
the time-fixed surface and the static gauge is taken.
The supergravity solution of the M2M5 bound state is given in Ref. [17], and it is also
(M)
a solution of I7 which takes the following form:
ds62 = f 3 k 3 dx dx + f 3 k 3 ab dx a dx b + f 3 k 3 dr 2 ,
e 2 = r 2f 3 k 3 ,
A012 = sin f 1 ,
(M) = 3 cos Qr
4 f 1 k 1 ,
F
012345r
A345 = tan k 1 ,
(6.12)
where
,
= 0, 1, 2,

Q
f =1+ 3,
r
a, b = 3, 4, 5,
k = sin2 + cos2 f.
(6.13)
(M5)
= 0 and M5 = 0 reproduces this solution when M5 = 8
We verified that (6.8) with F
cos , which is consistent with (6.11).
and M5 = M5 = 3Q
2 For a review, see Refs. [23,24]. Also, for an alternative formulation of the M5-brane effective action, see
Ref. [25].
315
7. (p, q) string and (p, q) 5-brane in type IIB supergravity and NS 5-brane in
type IIA supergravity
7.1. (p, q) string and (p, q) 5-brane
Let us recall the SL(2, R) symmetry in type IIB supergravity. It is convenient for this
purpose to work in the Einstein frame and redefine the RR 4-form. The Einstein metric is
given by
2
G(E)
GMN ,
MN = e
(7.1)
and the new RR 4-form is given by

1
C4new = C4 + C2 B2 .
(7.2)
2
The type IIB supergravity action (3.3) is rewritten in terms of the Einstein metric and the
new RR 4-form:

1
1 M M 1
1 2
j
i
M
|
F
|
d 10 X G(E) RG(E)
IIIB = 2
ij 3
5
3
2 (Im )2
2
4
210

ij
j
+ 2
(7.3)
C4new F3i F3 ,
810
where
= C0 + ie ,

| |2 Re
1
,
Mij =
Im Re
1

B2
C2i =
,
F3i = dC2i ,
C2
5 = dC new 1 ij C i F j .
F
(7.4)
4
2
3
2
One can easily check that the action (7.3) and the self-duality condition (3.5) is invariant
under the following SL(2, R) transformation.
=
a + b
c + d
a, b, c, d real, ad bc = 1,

d c
j
C2i = i j C2 , i j =
,
b a
C4new = C4new ,
(E)
(E)
GMN = GMN .
(7.5)
First, let us see how a solution to the HJ equation corresponding to (p, q) string is
obtained. Noting that (7.1) implies that (E) = , one can rewrite (5.1) with p = 1 in
316
terms of the Einstein metric as follows.

S2(0) 2 1 (E)
S2(0) 2
S2(0) 2
1
1
1
1

+
g
(E)
(E)
8 g (E) g
14
g (E) g
g (E) (E)
(0)
(0)
(0)
(0)
S2
S2
S2
S2
1 (E)
1
1
1
1

g
+ (Im )2
(E)
(E)
(E)
(E)
(E)
(E)
2

g g g
g
g
(E) (E)
+ g
g M1
ij
(0)
(0)
S2
S2
1
(E)
7

+ e3 R (S ) = 0.
i
j
(E)
(E)
C
g
g C
(7.6)
(E)
(E)
One can verify that this HJ equation is invariant under (7.5), which implies g = g
and (E) = (E) . Therefore, the SL(2, R) transformed S2(0) is also a solution to the HJ
equation, which clearly reproduces the supergravity solution of (p, q) string.
Second, let us see briefly how a solution to the HJ equation corresponding to (p, q)
5-brane is obtained. We consider the p = 5 case of the reduction in Section 3 with the
ansatz for H3 replaced by
1
H ( )d d d
3!
1 2+ 3
1 7 ( )1 2 3 di1 di2 di3 .
4 H
e
+
(7.7)
7
1
3!7!
Then, we obtain as a consistent truncation a seven-dimensional gravity, and rewrite it in
terms of the Einstein metric and the new RR 4 form:
3 (E)
(p,q)5
= d 7 h e 4
I7

1 (E)
3
1
3
(E)
Rh + e 2 R (S ) + (E) (E)
8
2 (Im )2

1
1 2 1
j
j
i
M
Mij F3i F3 |F
|
F
(7.8)
5
ij 7
7 ,
2
2
2
where
H3 =
3 = F3 + C0 H3 ,
F1 = dC0 ,
F3 = dC2 ,
F
1
1
5 = dC new B2 F3 + C2 H3 ,
F
4
2
2
1
new
7 = dC6 + C
H3 C2 B2 H3 ,
F
4
2
1
new

7 C
H7 = dB6 C0 F
F3 + C2 C2 H3 ,
4
4

H
7
e
F7i =
,
7 e C0 H
7
e F
H3 = dB2 ,
(7.9)
and , Mij and F3i are formally the same in (7.4). (7.8) is invariant under the
transformation (7.5) with
j
F7i = i j F7 .
(7.10)
317
The transformation laws for B6 and C6 are determined by (7.10). Clearly, the HJ equation
derived from (7.8) is invariant under this SL(2, R) transformation. Therefore, by rewriting
(5.2) with p = 5 in terms of the Einstein metric and the new RR 4-form and applying
the SL(2, R) transformation to it, we obtain a solution to the HJ equation reproducing the
supergravity solution of (p, q) 5-brane.
7.2. NS 5-brane in type IIA supergravity
In this subsection, using the relation between 11d supergravity and type IIA supergravity, we see that the NS 5-brane effective action is a solution to the HJ equation of
supergravity and it reproduces the supergravity solution of NS 5-brane. In order to see the
relation to type IIA supergravity, we consider a reduction of 11d supergravity on S 3 S 1 ,
which is different from the one done in Section 3:
2
= GMN dXM dXN
ds11

2
1
1
= h ( ) d d + e 2 2 ( ) d3 + e 2 1 ( ) dX10 ,
(7.11)
1
F4(M) = F(M)
d 1 d 4
4! 1 4
1 1 1 + 3 2
(M)1 7 i i i di1 di2 di3 dX10 ,
4
e4
+
1 7 F
1 2 3
3!7!
(7.12)
where , run from 0 to 6, and X10 parametrizes S 1 . Then, we obtain as a consistent
truncation a seven-dimensional gravity

2
1
3
3
3
3
(M)
= d 4 h e 4 1 + 4 2 Rh + e 2 R (S ) + 1 2 + 2 2
I7
8
8

1 (M) 2 1 (M) 2
,
F4
F7
(7.13)
2
2

where F
7
(M)
(M)
= dA6 12 A3 F4
. Let us consider the form
c
BI
WZ
+ SM5
+ SM5
+ M5 ,
SM5 = SM5
(0)
with
c
SM5
= M5

d 6x
1
1
g e 4 1 + 2 2 ,
(7.14)
(7.15)
BI and S WZ the same in (6.9). One can verify that (7.14) satisfies the HJ equation
and SM5
M5
(M)
under the assumption that the fields are constant on the fixed-time surface, up to
of I7
the constraint (6.10), if
2
2
= M5
.
M5 = 6R (S ) = 36 and M5
3
(7.16)
Following the relation between 11d supergravity and type IIA supergravity [26], we
define the fields in type IIA supergravity in terms of the fields in 11d supergravity as
318
follows.
2
3 (IIA)
h ,
h(M5)
= e
8
1 = ,
3
A3 = C3 ,
We rewrite
I7(M)
(NS5)
I7
4
2 = ,
3
A6 = B6 .
(7.17)
in term of these new fields and obtain
d 7 h

3
1
3
3
e2+ 4 Rh + e 2 R (S ) + 4 + 3
8

3
1 3
1
7 |2 ,
e 4 |F4 |2 e2+ 4 |H
(7.18)
2
2
7 = dB6 + 1 C3 F4 . This action is actually given by a consistent

where F4 = dC3 and H
2
truncation of type IIA supergravity in which the ansatzes for the fields are
1
2
ds10
= h ( ) d d + e 2 ( ) d3 ,
= ( ),
1 2+ 3
1 7 ( )i i i di1 di2 di3 ,
4 H
e
H3 =
7
1
1 2 3
3!7!
1
F4 = F1 4 ( ) d 1 d 4 .
(7.19)
4!
Thus, by rewriting (7.14) in terms of the fields in type IIA supergravity, we obtain the
NS 5-brane effective action (plus the cosmological term) that is a solution to the HJ
equation of I7(NS5) , a reduction of type IIA supergravity. Clearly, this solution reproduces
the supergravity solution of NS 5-brane.
8. Discussion and perspective

(0)
In this paper, we found that Sp+1
is a solution to the HJ equation of type IIA (IIB)
supergravity and it reproduces the supergravity solution representing Dp-branes in a
(0)
constant B2 field. Sp+1 is not a complete solution to the HJ equation though it has
some arbitrary constants. In other words, it should be obtained by making some of the
arbitrary constants in a certain complete solution take specific values. It is interesting to
(0)
see if Sp+1 can be generalized so that it includes more arbitrary constants and to look
for a complete solution. It is relevant to investigate what class of supergravity solutions
(0)
(0)
Sp+1 can reproduce. We verified that S4 does not reproduce the black 3-brane solution,
which preserves no supersymmetry, or the solution of the D3-brane with the wave, which
(0)
preserves 8 supersymmetries, and that S6 does not reproduce the solution of the D1D5
bound state, which preserves 8 supersymmetries.
319
As is clear from the general argument in Section 2, the quantities,

(0)
Sp+1
(p+1)
g B
and
(0)
Sp+1
p+1
BI
WZ
= Sp+1
+ Sp+1
,
(8.1)
are constant with respect to the time, where we take the sign in (5.9) such that p+1 = p+1 .
We verified this statement by an explicit calculation. We also obtain other conserved
(0)
quantities from the SL(2, R) transformed Sp+1
in type IIB supergravity, since it includes
the continuous parameters of SL(2, R). As is discussed in Section 2, we can reduce the
(0)
equations of motion to a set of the first order differential equations by using Sp+1
. We
may simplify these first order equations by using the conserved quantities so that we can
answer the above question and/or find a new solution of supergravity. As we discussed in
the introduction, it is also interesting to consider a reduction more complicated than that on
higher-dimensional sphere and obtain another solution to the HJ equation of supergravity,
which should be relevant to the gauge/gravity correspondence with less supersymmetries.
Finally, we make a remark on the case in which we perform a reduction on T 8p
8p
(R
). Let us consider an ansatz for the metric
1
2
ds10
= h ( ) d d + e 2 ( ) dy i dy i ,
(8.2)
parametrize
or
where , run from 0 to p + 1, i runs from 1 to 8 p and the
R 8p . We make ansatzes for the other fields similar to the ones in the reduction on S 8p ,
and obtain a (p + 2)-dimensional gravity as a consistent truncation. It follows that (5.2)
with p+1 = 0 is a solution to the HJ equation of this (p + 2)-dimensional gravity. This
fact seems to imply that the vacuum to vacuum amplitude vanishes in the reduction on the
flat manifolds.
yi
T 8p
Acknowledgements
We would like to thank T. Nakatsu, M. Nishimura and C. Nunez for discussions, and
M. Bianchi for comments on the holographic renormalization group. A.T. is also grateful
to the members of Center for Theoretical Physics, Massachusetts Institute of Technology,
for their hospitality, where most part of his work was done. The work of M.S. is supported
in part by Research Fellowships of the Japan Society for the Promotion of Science (JSPS)
for Young Scientists (No. 13-01193).
Appendix A. Equations of motion and Bianchi identities

In this appendix, we list explicitly the equations of motion and the Bianchi identity for
type IIA (IIB) and 11d supergravities. The equations of motion for type IIA supergravity
are
1
1
RMN + 2DM DN HML1 L2 HNL1 L2 e2 FML FNL
4
2
320

1 2
L1 L2 L3 + 1 e2 GMN |F2 |2 + |F
4 |2 = 0,
e FML1 L2 L3 F
(A.1)
N
12
4
1
R + 4DM D M 4M M |H3 |2 = 0,
(A.2)
2

1
L1 L2 MN
DL e2 H LMN + FL1 L2 F
2
1
L1 L2 L3 L4 F
L5 L6 L7 L8 = 0,
MNL1 L8 F
(A.3)
2!4!4!
1
ML1 L2 L3 = 0,
DL F LM + HL1 L2 L3 F
(A.4)
6
LM1 M2 M3 1 M1 M2 M3 L1 L7 HL1 L2 L3 F
L4 L5 L6 L7 = 0,
DL F
(A.5)
3!4!
where DM represents the covariant derivative in ten dimensions. The Bianchi identities for
type IIA supergravity are
dH3 = 0,
(A.6)
dF2 = 0,
4 + F2 H3 = 0.
dF
(A.7)
(A.8)
The equations of motion for type IIB supergravity are

1
1
1
L L
ML1 L2 F
L1 L2
RMN + 2DM DN HML1 L2 HN 1 2 e2 FM FN e2 F
N
4
2
4

1 2
3 |2 = 0,
L1 L4 + 1 e2 GMN |F1 |2 + |F
e FML1 L4 F
N
4 4!
4
1
M
M
R + 4DM D 4M |H3 |2 = 0,
2
2 LMN
L L L F
LMN + 1 F
MNL1 L2 L3 = 0,
DL e
+ FL F
H
6 1 2 3
1
L1 L2 L3 = 0,
DL F L HL1 L2 L3 F
6
LMN 1 HL1 L2 L3 F
MNL1 L2 L3 = 0,
DL F
6
M1 M5 = 1 M1 M5 L1 L5 F
L1 L2 L3 L4 L5 ,
F
5!
(A.9)
(A.10)
(A.11)
(A.12)
(A.13)
(A.14)
where DM represents the covariant derivative in ten dimensions again. The Bianchi
identities for type IIB supergravity are
dH3 = 0,
(A.15)
dF1 = 0,
3 F1 H3 = 0,
dF
(A.16)
(A.17)
3 H3 = 0.
5 F
dF
(A.18)
The equations of motion for 11d supergravity are

1 (M) 2
1 (M)
1
(M)L L L
RMN FML1 L2 L3 FN 1 2 3 + GMN R + F4 = 0,
12
2
4
1
M1 M2 M3 L1 L8 FL(M)
F (M)
= 0,
DL F (M)LM1 M2 M3
1 L2 L3 L4 L5 L6 L7 L8
2(4!)2
321
(A.19)
(A.20)
where DM represents the covariant derivative in eleven dimensions. The Bianchi identities
for 11d supergravity is
(M)
dF4
= 0.
(A.21)
Appendix B. Ansatzes for the fields

In this appendix, we write down the ansatzes for the fields except the metric and the
dilaton in the reduction of type IIA (IIB) supergravity on S 8p .
p=0
1
F2 = F ( ) d d .
2
(B.1)
p=1
1
H ( ) d d d ,
3!
3 = 1 F
( ) d d d .
F
3!
F1 = F ( ) d ,
1
H ( ) d d d ,
3!
4 = 1 F
( )d 1 d 4 .
F
4! 1 4
1
F2 = F ( ) d d ,
2
H3 =
(B.2)
p=2
H3 =
(B.3)
p=3
1
H ( ) d d d ,
F1 = F ( ) d ,
3!
3 = 1 F
( ) d d d ,
F
3!
( ) d 1 d 5
5 = 1 F
F
5! 1 5
1 4/5 1 5
e
F1 5 ( )i1 i5 di1 di5 .

5!5!
H3 =
(B.4)
322
p=4
1
1
H ( ) d d d ,
F2 = F ( ) d d ,
3!
2
1
( ) d 1 d 4
4 = F
F
4! 1 4
1 1 6
e
+
F1 6 ( )i1 i4 di1 di4 .
4!6!
H3 =
(B.5)
p=5
1
H ( ) d d d ,
F1 = F ( ) d ,
3!
( ) d 1 d 2 d 3
3 = 1 F
F
3! 1 2 3
1 3/4 1 7
e
+
F1 7 ( )i1 i2 i3 di1 di2 di3 ,

3!7!
5 = 1 F
( ) d 1 d 5
F
5! 1 5
1 3/4
1 5 ( )i i i
e
1 5 1 2 F
5
1 2 3
2!3!5!
1
2
d d di1 di2 di3 .
H3 =
(B.6)
p=6
1
1
H ( ) d d d + d ( )/i1 i2 d di1 di2 ,
3!
2!
1
1 /2 1 8
e
F2 = F1 2 ( ) d 1 d 2
F1 8 ( )i1 i2 di1 di2 ,
2!
2!8!
4 = 1 F
( ) d 1 d 4
F
4! 1 4
1 /2
1 6 ( )i i d 1 d 2 di1 di2 .
(B.7)
e 1 6 1 2 F
+
4
1 2
2!2!6!
H3 =
p=7
1
1
H ( ) d d d + d12( ) d 1 d 2 d1 ,
3!
2
( ) d 1 1 e/4 1 9 F
1 = F
1 9 ( ) d1 ,
F
9!
( ) d 1 d 2 d 3
3 = 1 F
F
3! 1 2 3
1 /4
1 7 ( ) d 1 d 2 d1 ,
e 1 7 1 2 F
+
4
2!7!
5 = 1 F
( ) d 1 d 5
F
5! 1 5
1 /4
1 5 ( ) d 1 d 4 d1 .
e 1 5 1 4 F
5
4!5!
H3 =
(B.8)
323
References
[1] M. Sato, A. Tsuchiya, BornInfeld action from supergravity, Prog. Theor. Phys. 109 (2003) 687, hepth/0211074.
[2] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 32, hep-th/9908142.
[3] A. Hashimoto, N. Itzhaki, Non-commutative YangMills and the AdS/CFT correspondence, Phys. Lett.
B 465 (1999) 142, hep-th/9907166.
[4] J.M. Maldacena, J.G. Russo, Large N limit of non-commutative gauge theories, JHEP 9909 (1999) 25,
hep-th/9908134.
[5] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math.
Phys. 2 (1998) 231, hep-th/9711200.
[6] 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;
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[7] O. Aharony, M. Berkooz, D. Kutasov, N. Seiberg, Linear dilatons, NS5-branes and holography, JHEP 9810
(1998) 004, hep-th/9808149.
[8] M.R. Douglas, W. Taylor, Branes in the bulk of anti-de Sitter space, hep-th/9807225.
[9] For example, I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and
SB-resolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191;
J.M. Maldacena, C. Nunez, Towards the large N limit of pure N = 1 super-YangMills, Phys. Rev. Lett. 86
(2001) 588, hep-th/0008001.
[10] A. Jevicki, Y. Kazama, T. Yoneya, Generalized conformal symmetry in D-brane matrix models, Phys. Rev.
D 59 (1999) 066001, hep-th/9810146.
[11] M.R. Douglas, G. Moore, D-branes, quivers, and ALE instantons, hep-th/9603167;
N. Nekrasov, A. Schwarz, Instantons on noncommutative R 4 , and (2, 0) superconformal six-dimensional
theory, Commun. Math. Phys. 198 (1998) 689, hep-th/9802068;
M. Berkooz, Non-local field theories and the non-commutative torus, Phys. Lett. B 430 (1998) 237, hepth/9802069.
[12] W. Bietenholz, F. Hofheinz, J. Nishimura, A non-perturbative study of gauge theory on a non-commutative
plane, JHEP 0209 (2002) 009, hep-th/0203151.
[13] I.L. Buchbinder, A.Yu. Petrov, A.A. Tseytlin, Two-loop N = 4 super-YangMills effective action and
interaction between D3-branes, Nucl. Phys. B 621 (2002) 179, hep-th/0110173.
[14] J. de Boer, E. Verlinde, H. Verlinde, On the holographic renormalization group, JHEP 0008 (2000) 003,
hep-th/9912012.
[15] J.C. Breckenridge, G. Michaud, R.C. Myers, More D-brane bound states, Phys. Rev. D 55 (1997) 6438,
hep-th/9611174;
M.S. Costa, G. Papadopoulos, Superstring dualities and p-brane bound states, Nucl. Phys. B 510 (1988)
217, hep-th/9612204.
[16] G.W. Gibbons, S.W. Hawking, Action integrals and partition functions in quantum gravity, Phys. Rev. D 15
(1977) 2752.
[17] J.M. Izquierdo, N.D. Lambert, G. Papadopoulos, P.K. Townsend, Dyonic membranes, Nucl. Phys. B 460
(1996) 560, hep-th/9508177;
J.G. Russo, A.A. Tseytlin, Waves, boosted branes and BPS states in M-theory, Nucl. Phys. B 490 (1997)
121, hep-th/9611047.
[18] P.S. Howe, E. Sezgin, Superbranes, Phys. Lett. B 390 (1997) 133, hep-th/9607227.
[19] P.S. Howe, E. Sezgin, D = 11, p = 5, Phys. Lett. B 394 (1997) 62, hep-th/9611008.
[20] P.S. Howe, E. Sezgin, P.C. West, Covariant field equations of the M-theory five-brane, Phys. Lett. B 399
(1997) 4959, hep-th/9702008.
[21] E. Witten, Five-brane effective action in M-theory, J. Geom. Phys. 22 (1997) 103, hep-th/9610234.
[22] M. Cederwall, B. Nilsson, P. Sundell, An action for the super-5-brane in D = 11 supergravity, JHEP 9804
(1998) 007, hep-th/9712059.
[23] E. Sezgin, P. Sundell, Aspects of the M5-brane, in: Conference on Superfivebranes and Physics in 5 + 1
Dimensions, Trieste, Italy, 13 April, 1998, hep-th/9902171.
[24] M.J. Duff, TASI lectures on branes, black holes and anti-de Sitter space, hep-th/9912164.
324
[25] P. Pasti, D.P. Sorokin, M. Tonin, Covariant action for a D = 11 five-brane with the chiral field, Phys. Lett.
B 398 (1997) 41, hep-th/9701037;
I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, Covariant action for the superfive-brane of M-theory, Phys. Rev. Lett. 78 (1997) 4332, hep-th/9701149;
M. Aganagic, J. Park, C. Popescu, J.H. Schwarz, World-volume action of the M-theory five-brane, Nucl.
Phys. B 496 (1997) 191, hep-th/9701166;
I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, On the equivalence of different
formulations of the M-theory five-brane, Phys. Lett. B 408 (1997) 135, hep-th/9703127.
[26] I. Bandos, A. Nurmagambetov, D. Sorokin, The type IIA NS5-brane, Nucl. Phys. B 586 (2000) 315, hepth/0003169.

Free field construction of D-branes in N = 2

superconformal minimal models
S.E. Parkhomenko
Landau Institute for Theoretical Physics, 142432 Chernogolovka, Russia
Received 6 May 2003; received in revised form 21 August 2003; accepted 22 August 2003
Abstract
The construction of D-branes in N = 2 superconformal minimal models based on free-field
realization of N = 2 super-Virasoro algebra unitary modules is represented.
PACS: 11.25.Hf; 11.25.Pm; 11.25.-w
Keywords: Strings; D-branes; Conformal field theory
1. Introduction
The role of D-branes [1] in the description of certain nonperturbative degrees of
freedom of strings is by now well established and the study of their dynamics has lead to
many new insights into string and M-theory [2,3]. Much of this study was done in the large
volume regime where geometric techniques provide reliable information. The extrapolation
into the stringy regime usually requires boundary conformal field theory (CFT) methods.
In this approach D-brane configurations are given by conformally invariant boundary states
or boundary conditions. However, a complete microscopic description these configurations
are well understood only for the case of flat and toric backgrounds where the CFT on the
world sheet is a theory of free fields. Due to this reason boundary state formalism for Dbranes has been subsequently developed and many calculations concerning the scattering
and couplings of closed strings in a D-brane background have been given exactly in [48].
The class of models of rational CFT gives the examples of curved string backgrounds
where the construction of the boundary states leaving a whole chiral symmetry algebra
E-mail address: spark@itp.ac.ru (S.E. Parkhomenko).
doi:10.1016/j.nuclphysb.2003.08.032
326
S.E. Parkhomenko / Nuclear Physics B 671 (2003) 325342
unbroken can be given in principle and the interaction of these states with closed strings
can be calculated exactly. But in practice the calculation of closed string amplitudes in
general CFT backgrounds is available only if the corresponding free field realization of
the model is known. Therefore, it is important to extend free field approach to the case of
rational models of CFT with a boundary.
This problem has been treated recently to the case of SU(2) WZNW model in [9],
where the Wakimoto free field realization of sl(2) KacMoody algebra [1014] has been
used to boundary states construction.
N = 2 superconformal minimal models represent a subclass of rational CFT which
is most important in the string theory applications. They are building blocks in Gepner
models [15] which give an exact solution by CFT methods of the problem of string
propagation on CalabiYau manifolds. The N = 2 minimal models as suggested are fixed
points of N = 2 supersymmetric LandauGinzburg (LG) models [16,17]. This rigorously
justified suggestion gives a possibility to describe string propagation on CalabiYau
manifold also in geometric terms and provides a link between the algebraic structure
encoded in N = 2 minimal models and geometry of the manifold. These mutually
complementary approaches of CFT and LG are very efficient also in investigation of Dbranes on CalabiYau manifolds [1825].
The free field realization of the unitary N = 2 super-Virasoro algebra representations
has been developed by Feigin and Semikhatov in [28]. From the one hand, it gives an
efficient way for correlation functions calculation. From the other hand it is closely related
[27,28] with Wakimoto free field description of SU(2) WZNW model as well as with LG
approach to N = 2 minimal models. The effect of boundaries has already been studied in
LG approach [23,24] and in the context of coset model [26]. Thus, it is important to extend
the free field realization to the case of N = 2 minimal models with boundaries.
In this note we extend the construction of [28] to the case of N = 2 minimal models
with boundaries. In Section 1, we review the free field realization and butterfly resolution
[28] of Feigin and Semikhatov of the unitary modules in N = 2 minimal models and obtain
free field representation for the characters. In Section 2, we construct Ishibashi states of A-,
B-types in Fock modules. In Section 3, we obtain in an explicit form A-, B-type Ishibashi
state for each irreducible N = 2 minimal model module using superposition of Ishibashi
states of Fock modules from butterfly resolution. The coefficients of the superposition are
fixed by imposing BRST invariance condition which is similar to that of the bulk theory.
At the end of the section free field construction of boundary states is represented using
Cardys prescription. In the last section we briefly discuss free field representation for
boundary correlations functions as well as some generalizations of the construction.
2. Free-field realization of N = 2 minimal models irreducible representations

In this section we briefly discuss free-field construction of Feigin and Semikhatov [28]
of the irreducible modules in N = 2 superconformal minimal models.
327
2.1. Free-field representations of N = 2 super-Virasoro algebra

We introduce (in the left-moving sector) the free bosonic fields X(z), X (z) and free
fermionic fields (z), (z), so that its singular OPEs are given by
X (z1 )X(z2 ) = ln(z12 ) + reg,
1
+ reg,
(z1 )(z2 ) = z12
(1)
where z12 = z1 z2 . Then for an arbitrary number the currents of N = 2 super-Virasoro

algebra are given by
1
(z),
G (z) = (z)X (z) (z),

1
J (z) = (z)(z) + X (z) X(z),

1
T (z) = X(z)X (z) + (z)(z) (z)(z)
2

1 2
1 2
X(z) + X (z) ,
2
G+ (z) = (z)X(z)
and the central charge is

1
.
c=3 12
(2)
(3)
As usual, the fermions in NS sector are expanded into half-integer modes:

1
1
[r]z 2 r ,
(z) =
[r]z 2 r ,
(z) =
r1/2+Z
G (z) =
r1/2+Z
G [r]z
23 r
(4)
r1/2+Z
and they are expanded into integer modes in R sector:

1
1
(z) =
[r]z 2 r ,
(z) =
[r]z 2 r ,
rZ
G (z) =
rZ
G [r]z
23 r
(5)
rZ
The bosons X(z), X (z), J (z), T (z) are expanded in both sectors into integer modes:

X[n]z1n ,
X (z) =
X [n]z1n ,
X(z) =
nZ
J (z) =

nZ
J [n]z
1n
T (z) =

nZ
nZ
L[n]z2n .
(6)
328
To describe the modules of N = 2 Virasoro superalgebra in NS sector we define the

vacuum state |p, p
such that
1
r ,
2
X[n]|p, p
= X [n]|p, p
= 0, n 1,
[r]|p, p
= [r]|p, p
= 0,
X[0]|p, p
= p|p, p
,
X [0]|p, p
= p |p, p
.
(7)
It is a primary state with respect to the N = 2 Virasoro algebra

G [r]|p, p
= 0,
r > 0,
J [n]|p, p
= L[n]|p, p
= 0,
j
J [0]|p, p
= |p, p
= 0,
L[0]|p, p
=
n > 0,
h(h + 2) j 2
|p, p
= 0,
4
(8)
where j = p p, h = p + p. The vacuum state |p, p

corresponds to the vertex
operator V(p,p ) (z) exp(pX (z) + p X(z)) placed at z = 0.
We denote by Fp,p the Fock module generated from the vector |p, p
by the fermionic
operators [r], [r], r < 1/2, and bosonic operators X [n], X[n], n < 0.
It is easy to calculate the character p,p (q, u) of the Fock module Fp,p . By the
definition

c
p,p (q, u) = TrFp,p q L[0] 24 uJ [0] .
(9)
Thus we obtain
p,p (q, u) = q
h(h+2)j 2
c
24
4
(q, u)
,
(q)3
where we have used the Jacobi theta-function

1
1 2
(q, u) = q 8
q 2 m um
(10)
(11)
mZ
and the Dedekind eta-function

1
(q) = q 24
1 qm .
(12)
m=1
The N = 2 Virasoro algebra has the following set of automorphisms which is known as
spectral flow [30]
G [r] G
t [r] G [r t],
c
L[n] Lt [n] L[n] + tJ [n] + t 2 n,0 ,
6
c
J [n] Jt [n] J [n] + t n,0 ,
3
(13)
329
where t Z. Note that spectral flow is intrinsic property of N = 2 super-Virasoro algebra

and hence, it does not depend on a particular realization. Allowing in (13) t to be halfinteger, we obtain the isomorphism between the NS and R sectors.
The spectral flow action on the free fields can be easily described if we bosonize
fermions ,

(z) = exp y(z) ,
(z) = exp +y(z) ,
(14)
and introduce spectral flow vertex operator [29]

1
Ut (z) = exp t y + X X (z) .
(15)
The following OPEs

t
(z1 )Ut (z2 ) = z12
:(z1 )Ut (z2 ):,
t
: (z1 )Ut (z2 ):,
(z1 )Ut (z2 ) = z12
1
tUt (z2 ) + r,
X (z1 )Ut (z2 ) = z12
1 t
Ut (z2 ) + r
X(z1 )Ut (z2 ) = z12
(16)
give the action of spectral flow on the modes of the free fields
[r] [r + t],
[r] [r t],
X [n] X [n] + tn,0 ,
X[n] X[n]
t
n,0 .
(17)
The action of the spectral flow on the vertex operator V(p,p ) (z) is given by the normal
ordered product of the vertex Ut (z) and Vp,p (z).
2.2. Irreducible N = 2 super-Virasoro representations and butterfly resolution
The N = 2 minimal models are characterized by the condition that is integer and
2. In NS sector the irreducible highest-weight modules, constituting the (left-moving)
space of states of the minimal model, are unitary and labeled by two integers h, j , where
h = 0, . . . , 2 and j = h, h + 2, . . . , h. The highest-weight vector |wh,j
of the
module satisfies the conditions (which are similar to (8))
G [r]|wh,j
= 0,
r > 0,
J [n]|wh,j
= L[n]|wh,j
= 0, n > 0,
j
J [0]|wh,j
= |wh,j
,
h(h + 2) j 2
|wh,j
.
L[0]|wh,j
=
4
(18)
If in addition to the conditions (18) the relation

G+ [1/2]|wh,j
= 0
(19)
330
is satisfied we call the vector |wh,j

and the module Mh,j chiral highest-weight vector
(chiral primary state) and chiral module, correspondingly. In this case we have h = j .
Analogously, antichiral highest-weight vector (antichiral primary state) and antichiral
module can be defined if instead of (19)
G [1/2]|wh,j
= 0
(20)
is satisfied. In this case h = j . In [28] the highest-weight vectors satisfying (18) are called
massive highest-weight vectors and the vectors satisfying in addition to (19) are called
topological highest-weight vectors. In this paper we prefer to use the terms introduced in
[29].
As we have seen in the preceding subsection, the highest weight vectors (18) can
be realized by the Fock vacuum vectors |p, p
. But the corresponding Fock modules
are reducible with respect to N = 2 super-Virasoro algebra. To construct irreducible
representations one needs to introduce the integer lattice of the momentums:

P = (p, p ) | p, p Z
(21)
and the space

FP =
Fp,p .
(22)
(p,p )P
Following [28] we introduce two fermionic screening currents S (z) and the charges Q
of the currents
S + (z) = exp(X )(z),
Q = dz S (z).
S (z) = exp(X)(z),
(23)
These charges commute with the generators of N = 2 super-Virasoro algebra (2) and act
in the space FP . Moreover they are nilpotent and mutually anticommute
(Q+ )2 = (Q )2 = {Q+ , Q } = 0.
(24)
Due to these properties one can combine the charges Q into BRST operator acting in
FP and build a BRST complex consisting of Fock modules Fp,p FP such that its
cohomology is given by NS sector N = 2 minimal model irreducible module Mh,j . This
complex has been constructed in [28].
Let us consider first free field construction for the chiral module Mh,h . In this case the
complex (which is known due to Feigin and Semikhatov as butterfly resolution) can be
331
represented by the following diagram

..
.
..
.
F1,h+ F0,h+
F1,h F0,h
(25)
F1,h F2,h
F1,h2 F2,h2
..
..
.
.
The horizontal arrows in this diagram are given by the action of Q+ and vertical arrows are
given by the action of Q . The diagonal arrow at the middle of butterfly resolution is given
by the action of Q+ Q (which equals Q Q+ due to (24)). Ghost number operator g of
this complex is defined for an arbitrary vector |vn,m
Fn,m+h by
g|vn,m
= (n + m)|vn,m
,
if n, m 0,
g|vn,m
= (n + m + 1)|vn,m
,
if n, m < 0.
(26)
For an arbitrary vector of the complex |vN

with the ghost number N the differential dN is
defined by
dN |vN
= (Q+ + Q )|vN
,
+
dN |vN
= Q Q |vN
,
if N = 1,
if N = 1,
(27)
and rises the ghost number by 1.

Theorem 1 [28]. Complex (25) is exact except at the F0,h module, where the cohomology
is given by the chiral module Mh,h .
c
The butterfly resolution allows to write the character h (q, u) TrMh,h (q L[0] 24 uJ [0] )
of the module Mh,h as an alternated sum:
h (q, u) = h(l) (q, u) h(r) (q, u),

(1)n+m n,h+m (q, u),
h(l) (q, u) =
n,m0
(r)
h (q, u) =
(1)n+m n,hm (q, u),
(28)
n,m>0
(l)
(r)
where h (q, u) and h (q, u) are the characters of the left and right wings of the
resolution.
To obtain the resolutions for other (antichiral and nonchiral) modules we note first that
all irreducible modules can be obtained from the chiral modules Mh,h , h = 0, . . . , 2, by
332
the spectral flow action [31]. It turns out that one can get all the resolutions by the spectral
flow action also. Indeed, the charges Q commute with spectral flow operator Ut as it is
easy to see from the corresponding OPEs, hence, the resolutions in NS sector are generated
from (25) by the operators Ut , t = h, h + 1, . . . , 1. The resolutions in R sector are
generated from the resolutions in NS sector by the spectral flow operator U1/2 .
To illustrate how it works we consider the Fock module F0,h from the middle of
the resolution. The vector |0, h
(h > 0) represents N = 2 super-Virasoro algebra chiral highest-weight vector |wh,h
of the module Mh,h . The vector [1/2]|0, h
=
(1/ h)G [1/2]|0, h
represents a cohomology class, hence it is in irreducible module
Mh,h . Acting on this vector by the operator U1 we obtain the vector |1/, h 1
, which is
N = 2 super-Virasoro algebra highest-weight vector |wh,h2
of the module Mh,h2 . Analogously, the vector [3/2][1/2]|0, h
= 1/(h(h 1))G [3/2]G [1/2]|0, h
represents another vector from irreducible module Mh,h . Acting on this vector by U2
we obtain the vector |2/, h 2
, which is N = 2 super-Virasoro algebra highest-weight
vector |wh,h4
of the module Mh,h4 . Going by this way further, we arrive at the end the
vector |h/, 0
= Uh (1/ h!)G [1/2 h] G [1/2]|0, h
, which is antichiral highestweight vector |wh,h
of the antichiral module Mh,h . The modules from R sector can be
generated analogously.
3. Ishibashi states in the Fock modules of N = 2 super-Virasoro algebra

In this section we begin to develop free field representation of N = 2 minimal models
Ishibashi states. Thus, it will be implied in what follows that the right-moving sector of the
z), X (z), (
z), (z), and the right-moving N = 2
model is realized by the free-fields X(
super-Virasoro algebra is given by the formulas similar to (2).
There are two types of boundary states preserving N = 2 super-Virasoro algebra [32],
usually called B-type

L[n] L[n]
|B
= J [n] + J[n] |B
= 0,
+

+ [r] |B
= G [r] + G
[r] |B
= 0,
G [r] + G
(29)
and A-type states

L[n] L[n]
|A
= J [n] J[n] |A
= 0,

+
[r] |A
= G [r] + G
+ [r] |A
= 0,
G [r] + G
(30)
where = 1. The Ishibashi states (as well as the boundary states) can be considered as the
linear functionals on the space of states of the N = 2 minimal model. From the other hand,
we have seen in Section 1, that the space of states in the left-moving sector is represented
by the cohomology groups of butterfly resolutions. It is clear that similar construction
can be applied for the right-moving space of states. Therefore, free field construction of
Ishibashi states has to be consistent with the resolutions. This problem of consistence will
be postponed to the next section. In this section we consider the most simple solutions of
(29), (30) in the tensor product of the left-moving Fock module Fp,p and right-moving
333
Fock module Fp,

p . We shall call these states as linear Ishibashi [33] states and denote by
p , , B(A)
.
|p, p , p,
Let us consider B-type linear Ishibashi states in NS sector. They can be easily obtained
from the following ansatz for fermions

p , , B
= 0,
[r] a [r] |p, p , p,

p , , B
= 0,
[r] b [r]
|p, p , p,
(31)
where a, b are the arbitrary constants. Substituting these relations into (29) and using (2)
we find
a = b = ,
1
p = p 1,
p = p ,

1
p , , B
= 0,
X[n] + X[n] + n,0 |p, p , p,

X [n] + X [n] + n,0 |p, p , p,
p , , B
= 0.
(32)
Thus, the linear B-type Ishibashi state in NS sector is given by the standard expression
[5,34,35]

1
|p, p , , B
=
exp X [n]X[n]
+ X[n]X [n]
n
n=1

exp [r][r]
+ [r] [r]
r=1/2

p, p , p , p 1 .
(33)
The closed string cylinder amplitude between such states in NS sector is given by
p2 , p2 , , B|q L[0]c/24uJ [0] |p1 , p1 , , B
= (p1 p2 )(p1 p2 )p1 ,p1 (q, u).

(34)
Note that the state p, p , , B| is defined in such a way to satisfy conjugate boundary
conditions and to take into account the charge asymmetry [3638] of the free-field
realization of the minimal model.
Introducing the new set of bosonic oscillators

v[n] =
X [n] X[n] ,
2

1
u[n] =
(35)
X [n] + X[n] ,
2
one can rewrite the B-type conditions (32) as

p , , B
= 0,
v[n] + v[n]
|p, p , p,

2
n,0 |p, p , p,
p , , B
= 0.
u[n] + u[n]
+
(36)
334
One can consider the coordinates (exp(u), v) as the polar coordinates on a complex
plane and think of the conditions (36)
as Neumann along both of coordinates (u, v). From
of the sum J [0] + J[0] generates

this point of view the bosonic part 2/ (v[0] + v[0])
an action of a circle on the complex plane such that v is an angular coordinate. Then,
B-type states correspond to 2-dimensional D-branes filling the complex plane. One can
equally well to think of the conditions (36) as Dirichlet
ones along (u, v) such that the
and then, B-type

circle action on the complex plane is generated by 2/ (v[0] v[0])
boundary states correspond to D0-branes. There are also two additional interpretations of
the conditions when we consider one of the relations (36) as Neumann and another one
as Dirichlet condition. Thus, one has four possibilities. As we shall see at the end of this
section, mirror symmetry left only two of them
The linear A-type Ishibashi states can be found analogously. We start from the ansatz
for fermions

|p, p , p,
p , , A
= 0,
[r] a [r]

p , , A
= 0,
[r] b [r] |p, p , p,
(37)
where a, b are the arbitrary constants. Then we find
a = ,
b=
1 + p
p =
,
p = p 1,

p , , A
= 0,
X[n] + X [n] + n,0 |p, p , p,

X [n] + X[n]
+ n,0 |p, p , p,
p , , A
= 0.
Hence the linear A-type Ishibashi state (in NS sector) is given by

1
1
exp X[n]X[n] + X [n]X [n]

|p, p , , A
=
n
n=1

1
exp
[r] [r] + [r][r]
r=1/2

1 + p
, p 1
p, p ,
and the corresponding closed string cylinder amplitude coincide with the (34).
One can find that (38) can be rewritten in (u, v) coordinates as

v[n] v[n]
|p, p , p,
p , , A
= 0,

2
u[n] + u[n]
+
n,0 |p, p , p,
p , , A
= 0.
(38)
(39)
(40)
One can consider these relations as Neumann condition along the coordinate u and
Dirichlet condition along v. Then, A-type states correspond to 1-dimensional D-branes
along the rays in complex plane which is in agreement with the results [23,24] obtained in
335
LG approach. But we are free to choose, similar to the case of B-type states, three other
possible interpretations. Because of A-type Ishibashi state (39) can be obtained from Btype Ishibashi state (33) by the mirror involution in the right-moving sector:
v[n]
= v[n],
[r]
= [r],
u[n]
= u[n],
[r] = [r],
(41)
we are left with two possibilities for A-type states: D1-branes along the rays or along the
circles of the complex plane, and we have two possibilities for B-type states: D2-branes or
D0-branes. But, Poincare duality relates to each other the states for each type, so that we
are left with D1-brane for A-type and D0-brane for B-type.
4. Boundary states in N = 2 minimal models
4.1. Ishibashi states in the irreducible modules of N = 2 super-Virasoro algebra
In this section we represent free field construction of Ishibashi states of irreducible
modules Mh,j . The construction uses the linear Ishibashi states (33), (39) as the building
blocks in such a way to be consistent with butterfly resolutions of irreducible modules.
The relations (28), (34) indicate that Ishibashi state of irreducible module has to be given
as a superposition of linear Ishibashi states of the Fock modules. Indeed, let us consider
the following superposition of B-type free-field Ishibashi states

|Mh,h , , B
=
cn,m |n, m + h, , B
n,m0
cn,m | n, m + h, , B
(42)
n,m>0
where the summation is performed over the momentums from the butterfly resolution (25).
It is easy to see that this state satisfies the relations (29). The arbitrary coefficients cn,m
and cn,m can be fixed partly from the condition that closed string cylinder amplitude
between the Ishibashi states gives the characters of irreducible modules. In the free field
realization it is equivalent to the relation
Mh ,h , , B|(1)g q L[0]c/24uJ [0] |Mh,h , , B
= h ,h h (q, u).
(43)
Indeed, using (34) one can find

Mh ,h , , B|(1)g q L[0]c/24uJ [0] |Mh,h , , B

= h ,h
(1)n+m |cn,m |2 n,h+m (q, u)
n,m0

n,m>0

(1)n+m |cn,m |2 n,hm (q, u) .
(44)
336
Comparing with (28) we find

|cn,m |2 = |cn,m |2 = 1.
(45)
Thus, the state (42) is a good candidate for free-field realization (in NS sector) of B-type
Ishibashi state of the chiral module Mh,h . It would be a genuine Ishibashi state if it did not
radiate nonphysical closed string states which are present in the free field representation
of the model. In other words, the overlap of this state with an arbitrary closed string state
which does not belong to the Hilbert space of the N = 2 minimal model should vanish. As
we will see this condition can be formulated as a BRST invariance condition of the state
(42) and it will fix the coefficients cn,m , cn,m up to the common factor.
To formulate BRST invariance condition one has to consider what kind of closed string
states in the free field realization can interact with the state (42). They come from the
product of left-moving and right-moving Fock modules Fn,h+m Fn1/,1hm ,
where n, m 0 or n, m < 0. The left-moving modules of the superposition (42) constitute
the butterfly resolution (25) whose cohomology is given by the irreducible chiral module
Mh,h . What about the Fock modules from the right-moving sector? They do not form the
resolution like (25) due to the relations for the momentums from (32) and (38). Instead the
right-moving Fock modules constitute the dual butterfly resolution:
..
.
..
.
F1 1 ,1h F 1 ,1h
F1 1 ,1h
F 1 ,1h
F1 1 ,1h+ F2 1 ,1h+
F1 1 ,1h+2 F2 1 ,1h+2
..
.
..
.
(46)
The arrows on this diagram are given by the same operators as on the diagram (25). We
define the ghost number operator g of the complex similar to the ghost number operator of
the complex (25). For an arbitrary vector |vn,m
Fn1/,m1h by
g|
vn,m
= (n + m)|vn,m
,
if n, m 0,
g|v
n,m
= (n + m 1)|vn,m
,
if n, m > 0.
(47)
The differentials dN are defined similar to (27). This resolution can also be used for
description of irreducible modules due to
Theorem 2. Complex (46) is exact except at the F1/,1h module, where the cohomology is given by the antichiral module Mh,h .
337
The proof of this theorem is similar to that one is given in [28] for the complex (25).
Thus, the states which can interact with (42) come from the product of resolution (25) and
(46). The tensor product of complexes (25) and (46) constitutes the complex
2
1
+1
0
Ch,h
Ch,h
Ch,h
,
Ch,h
(48)
which is graded by the sum of the ghost numbers g + g and for an arbitrary ghost number
i
is given by the sum of products of the Fock modules from the resolution
i the space Ch,h
(25) and (46) such that g + g = i. The differential D of the complex (48) is defined by the
differentials dN and dN of the complexes (25) and (46)
Di |vN vN
= |dN vN vN
+ (1)N |vN dN vN
,
(49)
where |vN
is an arbitrary vector from the complex (25) with ghost number N , while |vN
is an arbitrary vector from the complex (46) with the ghost number N and N + N = i. It
follows from the Theorems 1 and 2 that the cohomology H of the complex (48) is nonzero
only at grading 0 and it is given by the product of irreducible modules Mh,h M h,h .
The Ishibashi state we are looking for can be considered as a linear functional on the
Hilbert space of N = 2 superconformal minimal model, then it has to be an element of
the homology group H . Therefore, the BRST invariance condition for the state can be
formulated as follows.
Let us define the action of the differential D on the state |Mh,h , , B
by the relation
D Mh,h , , B|vN vN
Mh,h , , B|DN+N |vN vN
,
(50)
N+N
where vN vN is an arbitrary element from Ch,h
. Then, BRST invariance condition
means that
D |Mh,h , , B
= 0.
(51)
Theorem 3. Superposition (42) satisfies BRST invariance condition (51) if the coefficients
cn,m , cn,m of the superposition obey the equations
cn,m = c0,0 ,
if n + m = 2k,
cn,m = c0,0,
if n + m = 2k + 1.
(52)
Thus the coefficients depends on the ghost number g.

Proof. In view of (42), (33) and because of differential D rises the ghost number by 1,
1
have nonzero overlap with Ishibashi
only BRST images of the states |vN v1N
Ch,h
state |Mh,h , , B
. Thus, one needs to show that

Mh,h , , B|D1 |vN v1N
= Mh,h , , B|dN |vN v1N

+ (1)N Mh,h , , B|d1N |vN v1N
= 0.
(53)
It will be implied during the proof that the state D1 |vN v1N
corresponds to the field
(D1 (vN v1N ))(z, z ) which is placed at the center z = z = 0 of the unit disk.
338
(1) Let N 0 and hence vN and v1N belong to the left wings of resolutions (25) and
(46). Using relations (31) and (32) as well as the definitions (27), (23) one can rewrite the
first term from (53) as
Mh,h , , B|dN |vN v1N

= Mh,h , , B| exp(X0 X 0 )Q + exp (X0 X 0 ) Q |vN v1N
,
(54)
X0 , X0 , X0 are the constant modes canonically conjugated to the momentums

where
X[0], X[0],
X [0], X [0]. The operators exp(X0 X 0 ), exp(X0 X 0 ) only shift the
momentums in the components of the superposition Mh,h , , B|. From the other hand we
can decompose the vectors |vN
and |v1N
according to their components in the Fock
modules of the resolution
X0 ,
|vN
= |yN,h
+ |yN1,h+
+ + |y0,h+N
,
|v1N
= |y1N 1 ,1h
+ |yN 1 ,1h+
+ + |y 1 ,1h+(N+1)
.
(55)
Substitution of this decomposition into the each term of Eq. (53) gives (52).
(2) Let N = 1 and hence v1 and v0 belong to the vertexes of the right wings of
resolutions. Using relations (31) and (32) as well as the definitions (27), (23) one can
rewrite the first term from (53) as
Mh,h , , B|d1 |v1 v0

|v1 v0
.
= Mh,h , , B| exp(X0 X 0 ) exp (X0 X 0 ) Q + Q
(56)
Substitution of this decomposition into the each term of Eq. (53) gives the relation from
(52) when n = m = 1.
(3) Let N < 1 and hence vN and v1N belong to the right wings of resolutions.
This case can be treated similar to the case N 0 and hence we obtain (52). It proves the
theorem.
Note that BRST-closed state (42), (52) is not BRST-exact due to (43). Hence, it
represents a homology class from H and it is defined modulo BRST-exact states satisfying
(29). Note also that normalization phase c0,0 cannot be fixed by the BRST invariance
condition. We fix the normalization of each Isibashi state by c0,0 = 1.
Due to the arguments from Section 1 we can obtain free field construction of the
remainder N = 2 minimal model B-type Ishibashi states in NS sector applying to the
Ishibashi state |Mh,h , , B
the spectral flow operators Ut U t , where t = h, h +

1, . . . , 1. Then, the action by the spectral flow operator U1/2 U 1/2 on the Ishibashi states
from the NS sector gives free field construction of B-type Ishibashi states in R sector. It is
also clear that free field construction of A-type Ishibashi states in N = 2 minimal models
can be obtained from B-type Ishibashi states by the mirror involution (41).
4.2. Boundary states
Free field representation of the boundary states in NS or R sector can be constructed by
applying Cardys prescription [39] to free field realized Ishibashi states.
339
Let us denote by S(h,j ),(h ,j ) the S-matrix of modular transformation of the full
characters of N = 2 minimal model in NS sector:

S(h,j ),(h ,j ) h ,j (q,
0),
h,j (q, 0) =
h ,j

jj
(h + 1)(h + 1)
1
exp
,
S(h,j ),(h ,j ) =
sin
(57)
where q = exp(2 ), q = exp(2/ ). Then Cardys formula in NS sector gives the

following boundary states

|Dh,j , , A
=
(58)
D(h,j ),(h ,j ) |Mh ,j , , A
,
h ,j
while in R sector it gives

|Dh,j , , A
D(h,j ),(h ,j ) U 1 U 1 |Mh ,j , , A
,
2
(h ,j )
(59)
where
S(h,j ),(h ,j )
D(h,j ),(h ,j ) =
S(0,0),(h ,j )
(60)
and h = 0, . . . , 2, j = h , h + 2, . . . , h .
Free field B-type boundary states can be obtained from A-type boundary states by the
orbifold projection [26]. In NS sector we obtain

D(h,j ),(h ,0) |Mh ,0 , , B
,
|Dh,j , , B
=
(61)
h
while in R sector we have the following

D(h,j ),(h ,0) U 1 U 1 |Mh ,0 , , B
,
|Dh,j , , B
=
h
(62)
where h = 0, . . . , 2 and is even.

The Ishibashi states in the expressions (58), (59), (61), (62) correspond to the full
characters of the N = 2 Virasoro superalgebra. Free field realization of the standard A(B-)type Ishibashi states |h, j, s, A(B)
, where s = 0, 2 in NS sector and s = 1, 1 in R

sector is given by the relations

Mh,j , +, A(B) = h, j, 0, A(B) + h, j, 2, A(B) ,

Mh,j , , A(B) = h, j, 0, A(B) h, j, 2, A(B) ,
(63)
in NS sector and it is given by

U 1 U 1 Mh,j , +, A(B) = h, j, 1, A(B) + h, j, 1, A(B) ,
2
2

U 1 U 1 Mh,j , , A(B) = h, j, 1, A(B) h, j, 1, A(B) ,
2
in R sector.
(64)
340
Due to the brief discussion of boundary conditions in Section 2 we have the following
geometric interpretation of boundary states in N = 2 minimal model. A-type boundary
states can be considered as D1-branes along the rays or along the circles in the complex
plane and they are Poincar dual to each other. This is in agreement with the results [23,24]
obtained in LG approach. B-type boundary states can be considered as D0- or D2-branes
which are also Poincar dual to each other.
5. Discussion
In this note we represented free field construction of Ishibashi and boundary states
in N = 2 superconformal minimal models using free field realization of N = 2 superVirasoro algebra unitary modules. Each Ishibashi state of the model is given by infinite
superposition of linear Ishibashi states of Fock modules, forming butterfly resolution of
irreducible representation of N = 2 super-Virasoro algebra. It is shown that coefficients
of the superposition are fixed by the BRST invariance condition and the Ishibashi state
constructed this way is BRST closed but not BRST exact and represents thereby a
homology class. We group these free field realized Ishibashi states into the boundary states
of N = 2 minimal model using the solution found by Cardy. Due to BRST invariance the
boundary states do not radiate nonphysical closed string states (which are present originally
in the free field space of states). We found that B-type boundary states corresponds to D0or D2-branes in complex plane target space. A-type D-branes is natural to identify with
rays or circles in complex plane. This identification is in agreement with the results [23,
24] obtained in LG approach but more detail investigation of D-brane geometry in the free
field approach needs to be done. It would be interesting to find geometric interpretation of
the superposition of linear Ishibashi states involved into the free field construction of the Dbranes. In the R sector it is a brane antibrane system due to (52) and it would be interesting
to find the interpretation of this superposition in the context of tachyon condensation [40].
We close with a brief discussion of some directions to develop. At first we would like to
point out that our free field construction can be easily generalized to the case of orientifolds.
The second problem is a free field construction for boundary correlation functions. This
problem is rather technical in fact because all needed ingredients are known. Indeed, free
field representation of the boundary states is obtained in the present note. The free field
description of the irreducible representations, vertex operators and the screening charge
generating quantum group structure of the conformal blocks (note that the fermionic
screenings Q of the butterfly resolution have trivial braiding relations) are known from
[28]. The screening charge is nothing else the standard Wakimoto screening charge (the
one involved in the construction of Felder-type resolution [10]) of SU(2) WZNW model.
In terms of the free fields it is given by the integral of the (bosonic) screening current:

1
QW = dz SW (z), SW = (X + ) exp X X .
(65)
It is easy to check that QW commutes with N = 2 super-Virasoro algebra and fermionic

charges Q . We hope to develop free field representation of the boundary correlation
functions in future publication.
341
The third interesting direction is the generalization of our boundary state construction
to the case of nondiagonal modular invariant partition functions. It is obviously important
to generalize the free field construction of boundary states to the case of Gepner models.
Free field approach is also applies to N = 2 superconformal models with W -algebra
of symmetries. Some of them is believed to coincide with KazamaSuzuki models [41].
N = 2 minimal models is the simplest example of this situation when KazamaSuzuki
model is SU(2) U (1)/U (1) coset and it would be interesting to extend free field
construction of D-branes to more general class of KazamaSuzuki models.
Acknowledgements
I am grateful to B.L. Feigin, A.M. Semikhatov and I.Yu. Tipunin for very useful
discussions. This work was supported in part by Grants RBRF-01-02-16686, RBRF-0015-96579, INTAS-00-00055.
References
[1] J. Polchinski, Phys. Rev. Lett. 75 (1995) 4724, hep-th/9510017;
J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
[2] E. Witten, Nucl. Phys. B 443 (1995) 85.
[3] J. Polchinski, String Theory, Cambridge Univ. Press, Cambridge, 1998.
[4] P. Di Vecchia, M. Frau, M. Pesando, S. Scuito, A. Lerda, R. Russo, Nucl. Phys. B 507 (1997) 259, hepth/9707068.
[5] M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Phys. Lett. B 400 (1997) 52, hep-th/9702037.
[6] M. Billo, P. Di Vecchia, M. Frau, M. Pesando, S. Scuito, A. Lerda, R. Russo, Nucl. Phys. B 526 (1998) 199,
hep-th/9802088.
[7] P. Di Vecchia, M. Frau, A. Lerda, A. Liccardo, Nucl. Phys. B 565 (2000) 397, hep-th/9906214.
[8] F. Hussain, R. Iengo, C. Nunez, C.A. Scrucca, Phys. Lett. B 409 (1997) 101, hep-th/9706186;
F. Hussain, R. Iengo, C. Nunez, C.A. Scrucca, Closed string radiation from moving D-branes, hepth/9710049;
F. Hussain, R. Iengo, C. Nunez, C.A. Scrucca, Interaction of D-branes on orbifolds and massless particle
emission, hep-th/9711021.
[9] S.E. Parkhomenko, Nucl. Phys. B 617 (2001) 198, hep-th/0103142.
[10] D. Bernard, G. Felder, Commun. Math. Phys. 127 (1990) 145.
[11] B.L. Feigin, E. Frenkel, Representations of affine KacMoody algebras and bosonization, in: Physics and
Mathematics of Strings, Vol. 271, World Scientific, Teaneck, NJ, 1990;
B.L. Feigin, E. Frenkel, Commun. Math. Phys. 128 (1990) 161.
[12] Vl.S. Dotsenko, Nucl. Phys. B 358 (1991) 547.
[13] P. Bouwknegt, J. McCarthy, K. Pilch, Phys. Lett. B 234 (1990) 297;
P. Bouwknegt, J. McCarthy, K. Pilch, Phys. Lett. B 258 (1991) 127;
P. Bouwknegt, J. McCarthy, K. Pilch, Commun. Math. Phys. 131 (1990) 125.
[14] A. Gerasimov, A. Morozov, M. Olshanetsky, A. Marshakov, Int. J. Mod. Phys. A 5 (1990) 2495.
[15] D. Gepner, Nucl. Phys. B 296 (1988) 757.
[16] C. Vafa, N. Warner, Phys. Lett. B 218 (1988) 51.
[17] E. Martinec, Phys. Lett. B 217 (1989) 431.
[18] A. Recknagel, V. Schomerus, Nucl. Phys. B 531 (1998) 185, hep-th/9712186.
[19] M. Gutperle, Y. Satoh, D-branes in Gepner models and supersymmetry, Nucl. Phys. B 543 (1999) 73, hepth/9808080.
342
[20] I. Brunner, M.R. Douglas, A. Lawrence, C. Romelsberger, D-branes on the quintic, hep-th/9906200.
[21] M. Naka, M. Nozaki, Boundary states in Gepner models, hep-th/0001037.
[22] J. Fuchs, C. Schweigert, Nucl. Phys. B 530 (1998) 99, hep-th/9712257;
J. Fuchs, C. Schweigert, Nucl. Phys. B 588 (2000) 110, hep-th/0003298.
[23] K. Hori, A. Iqbal, C. Vafa, D-branes and mirror symmetry, hep-th/0005247.
[24] S. Govindarajan, T. Jayaraman, On the LandauGinzburg description of boundary CFTs and special
Lagrangian submanifolds, hep-th/0003242;
S. Govindarajan, T. Jayaraman, T. Sarkar, Worldsheet approaches to D-branes on supersymmetric cycles,
hep-th/9907131.
[25] U. Lindstrom, M. Zabzine, N = 2 conditions for non-linear sigma models and LandauGinzburg models,
hep-th/0209098.
[26] J. Maldacena, G. Moore, N. Seiberg, Geometrical interpretation of D-branes in gauged WZW models, hepth/0105038.
[27] B.L. Feigin, A.M. Semikhatov, I.Yu. Tipunin, Equivalence between chain categories of representations of
affine sl(2) and N = 2 superconformal algebras, J. Math. Phys. 39 (1998) 3865, hep-th/9701043.
[28] B.L. Feigin, A.M. Semikhatov, Free-field resolutions of the unitary N = 2 super-Virasoro representations,
hep-th/9810059.
[29] W. Lerche, C. Vafa, N.P. Warner, Chiral rings in N = 2 superconformal theories, Nucl. Phys. B 324 (1989)
427.
[30] A. Schwimmer, N. Seiberg, Phys. Lett. B 184 (1987) 191.
[31] B.L. Feigin, A.M. Semikhatov, V.A. Sirota, I.Yu. Tipunin, Resolutions and characters of irredicible
representations of the N = 2 superconformal algebra, hep-th/9805179.
[32] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477 (1996) 407.
[33] N. Ishibashi, Mod. Phys. Lett. A 4 (1989) 251.
[34] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 293 (1987) 83.
[35] J. Polchinski, Y. Cai, Nucl. Phys. B 296 (1988) 91.
[36] B.L. Feigin, D.B. Fuks, Functional Anal. Appl. 16 (1982) 114;
B.L. Feigin, D.B. Fuks, Functional Anal. Appl. 17 (1983) 241.
[37] V.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312.
[38] G. Felder, Nucl. Phys. B 317 (1989) 215.
[39] J.L. Cardy, Nucl. Phys. B 324 (1989) 581.
[40] A. Sen, Tachyon condensation on the brane antibrane system, JHEP 9808 (1998) 012, hep-th/9805170.
[41] K. Ito, Int. J. Mod. Phys. A 7 (1992) 4885.

Discreteness of the spectrum of the compactified

D = 11 supermembrane with nontrivial winding
L. Boulton a , M.P. Garca del Moral b , A. Restuccia b
a Departamento de Matemticas, Universidad Simn Bolvar, Apartado 89000, Caracas 1080-A, Venezuela
b Departamento de Fsica, Universidad Simn Bolvar, Apartado 89000, Caracas 1080-A, Venezuela
Received 25 February 2003; received in revised form 29 July 2003; accepted 20 August 2003
Abstract
We analyze the Hamiltonian of the compactified D = 11 supermembrane with nontrivial central
charge in terms of the matrix model constructed in [Phys. Rev. D 66 (2002) 045023, hep-th/0103261].
Our main result provides a rigorous proof that the quantum Hamiltonian of the supersymmetric
model has compact resolvent and thus its spectrum consists of a discrete set of eigenvalues with
finite multiplicity.
1. Introduction
According to the results of [2], the spectrum of the SU(N) regularized supermembrane
on D = 11 Minkowski target space is continuous and it consists on the whole interval
[0, ). Although this property was proven for a regularized model, it leads to a remarkable
interpretation of the supermembrane in terms of a multiparticle theory. It also provides
explicit evidence of how the presence of supersymmetry may change completely the
spectrum of a bosonic discrete Hamiltonian over a compact world volume. The proof
was based on the presence of supersymmetry and on the existence of locally singular
configurations, which do not change the energy of the system.
In contrast, the situation concerning the spectrum of the compactified supermembrane
is somewhat different. Since the closed but not exact modes present in the compactified
case do not fit into an SU(N) formulation of the theory, cf. [4], the SU(N) regularization
E-mail addresses: lboulton@ma.usb.ve (L. Boulton), mgarcia@fis.usb.ve (M.P. Garca del Moral),
arestu@usb.ve (A. Restuccia).
doi:10.1016/j.nuclphysb.2003.08.024
344
L. Boulton et al. / Nuclear Physics B 671 (2003) 343358
of the compactified supermembrane seems not to be possible. In [4], it was suggested

that the spectrum of the compactified supermembrane should also be continuous due to
the existence of string-like spikes as in the noncompactified case. In fact, the presence of
those singular configurations, which do not change the energy of the system, is a common
property of all p-branes, [5]. The results of [6] show that the Hamiltonian formulation
of the super M5-brane also exhibits singular configurations, even when they are neither
present in the known covariant formulation of the theory given in [7] nor in the one
considered in [8].
Reference [9,10], formulate the compactified supermembrane on M9 S1 S1 as
a noncommutative gauge theory. In such case, the Hamiltonian can be given in terms
of a noncommutative super-Maxwell theory plus the integral of the curvature of the
noncommutative connection on the world volume. An explicit proof that this Hamiltonian
allows the presence of string-like spikes in the configuration space of the compactified
supermembrane can be found in [1]. This is in agreement with the arguments of [4]. If,
however, the theory is restricted to a fixed central charge, by describing a sector of the
full compactified supermembrane, the Hamiltonian reduces exactly to a noncommutative
super-Maxwell theory coupled to seven scalar fields representing the directions transverse
to the supermembrane. In this case, cf. [1], there are no string-like spikes in the
configuration space. This result is in agreement with the arguments of [12].
According to [13], if one considers the central charge to be fixed, the bosonic part of
this Hamiltonian is bounded below and it becomes infinite at infinity in every possible
direction on the configuration space. The potential is what we can formally regard as
basin shaped. This property ensures that the resolvent of the bosonic Hamiltonian is
compact and therefore its spectrum consists purely of a set of isolated eigenvalues of
finite multiplicity whose only accumulation point is +. Furthermore, it is possible to
find upper bounds for the asymptotic distribution of eigenvalues, by comparing with the
harmonic oscillator.
Our aim in the present paper is to prove rigorously that the spectrum of the
Hamiltonian of the noncommutative super-Maxwell theory describing the compactified
D = 11 supermembrane with fixed central charge, also consists purely of a set of isolated
eigenvalues of finite multiplicity whose only accumulation point is +. Our proof is based
on two fundamental properties of the case already discussed in [13], the nonexistence
of string-like spikes and the shape of the bosonic potential. Our key argument is the
criterion furnished in Lemma 1, which extends to the supersymmetric case the well-known
fact that the spectrum of a Hamiltonian is discrete, when the potential is bounded from
below and unbounded above in all directions (see [14,15]). The latter played an analogous
fundamental role in the bosonic case discussed in [13]. We choose to state Lemma 1 in
generic form and provide a self-contained proof, to highlight the fact that it might be
applied to any Hamiltonian, as long as the bosonic potential forbids string-like spikes in
the fermionic sector of the theory.
We should emphasize the relationship between our (fixed central charge) model and
the (free) model studied in [24]. The winding supermembrane in the latter case can
be regarded as a free model, since it includes in the configuration space all possible
wrappings, n = 0, 1, 2, . . . , on the compactified target space. In distinction, the present
work concerns only the sector of the full theory which corresponds to a fixed central charge
345
n = 0. This case is also of physical relevance, since the fixed wrapping is a topological
condition with phenomenological applications. The change in character of the spectrum
is formally analogous to the well-known case of the Laplacian acting on domains of the
Euclidean space: on the one hand the operator has pure continuous spectrum when we
choose the domain to be the whole space, on the other the spectrum is purely discrete
when the domain is bounded, cf. [15]. We will comment further on the comparison of
these two models in Section 2.
2. The Hamiltonian of the double compactified supermembrane

Reference [17] is devoted to studying the quantization of the compactified supermembrane in an M9 S 1 S 1 target space, by finding an explicit expression, given in terms
of creation and annihilation operators, of the quantum Hamiltonian in the semi-classical
regime. In particular, there are no vacuum energy corrections to the mass formula, since
the fermionic contribution cancels the bosonic one.
The exact Hamiltonian was computed in [10] (see also [11]), in terms of a symplectic
noncommutative geometry. The symplectic structure which gives rise to the geometry,
arises from the nontrivial central charge originated in the wrapping of the supermembrane
along the compactified directions of the target space. In order to make more transparent
the present exposition, we devote this section to review the main ideas which led to the
construction of this Hamiltonian.
Let us start by considering the D = 11 supermembrane Hamiltonian in the light cone
gauge as in [2]. In this model, the potential is given by
2

V (X) = Xm , Xn , m, n = 1, . . . , 9,
where {Xm , Xn } ( ab / W )a Xm b Xn . Here, the scalar density W appears in the

formulation as consequence of the light cone gauge fixing procedure.
Let be the spatial part of the world volume. We always assume that is a
compact Riemann surface of genus g. If only one of the target space spatial coordinates is
compactified on S 1 , the natural winding condition is given by

(1)
dX = 2mj ,
cj
where mj are integers and cj is a basis of homology on . Analogously, when the target
space has two compactified directions on S 1 S 1 , we may consider Xr , r = 1, 2, as angular
coordinates on each S 1 . In order to have a well-defined map over S 1 , we must impose as
before the condition

dXr = 2mj r , r = 1, 2.
(2)
cj
Assume that the image of under (X1 , X2 ) describes a torus. Then we should impose an
additional constraint. If itself is a generic torus, where w1 and w2 denote the normalized
346
basis of homology on , we have

dXr = m1r w1 + m2r w2 + dar ,
r = 1, 2,
where mj r , j = 1, 2, are the same integers introduced in (2) and dar are exact one-forms.
The requirement that the image of is a torus, may be interpreted as the independence of
the one-forms m1r w1 + m2r w2 for r = 1, 2, i.e.,
det(mj r ) = n = 0.
This condition is equivalent to requiring

Z = (dXr dXs ) rs = 2n = 0.
(3)
The factor 2 corresponds to a normalization of the area of . We remark that, since Z

becomes

g11 dg1 g21 dg2 ,
where G = (g1 , g2 ) U (1) U (1), the integer n is the winding number of the group
U (1) U (1) over .
Notice that the integral in condition (3), also corresponds to a realization of the central
charge of the supersymmetric algebra of the supermembrane. Hence the condition n = 0 is
equivalent to having nontrivial central charge. If the Xr fulfill condition (2), then (3) holds
automatically. However nontriviality of (2), does not necessarily imply nontriviality of (3).
We show below, that the configuration space of the compactified supermembrane with fixed
central charge is completely characterized by the integer n, that is only the determinant of
(mj r ) is relevant.
In order to describe the winding of the supermembrane in terms of maps from onto
S 1 S 1 satisfying (3), together with maps (Xm )7m=1 : R7 , we interpret (3) in terms
of geometrical objects. Let
F := dXr dXs rs
(4)
be a closed two-form on , such that F satisfies (3). Since dF = 0 and F = 2n, there
always exists a U (1) principal bundle over and a one-form connection on it, such that F
is the curvature two-form, [16]. The integer n characterizes the bundle. If n = 0, the bundle
is nontrivial and the connection one-form must have nontrivial transitions. In this bundle,
there are particular connection one-forms satisfying
ab
F F
(5)
ab = n
W
at any point of , where a, b = 1, 2 are indices associated to local coordinates on .
These are the so-called Dirac monopoles over Riemann surfaces. It turns out that these
monopoles together with the constraint Xm = 0, m = 1, . . . , 7, are configurations where
the Hamiltonian of the supermembrane have local minima, [9]. Moreover, there is only
one local minimum for each n.

347
ab is nondegenerate. Any nondegenerate closed two-form,

Condition (5) implies that F
can always be decomposed as
ab = a X
r b X
s rs ,
F
in the sense that there exists a Darboux atlas for such that the above holds on each open
r , r = 1, 2 are harmonic maps over with metric
set. The X
r b X
s rs
gab = a X
the pull-back of the Euclidean metric over S 1 S 1 . Notice that this is the metric arising
r are harmonic one-forms over . If is any
from the supermembrane action. The dX
given torus, we may then consider these one forms as a normalized basis of homology
r are not uniquely determined. In fact, we are allowed to
over . Clearly the maps X
remains invariant.
change the basis by an element of SL(2, Z), the modular group, and F
This change of basis corresponds to a conformal diffeomorphism over . From (5) we
obtain
ab = ab a X
r b X
s rs ,
n W = ab F
which is invariant under this conformal diffeomorphism. Consequently it is an area
preserving diffeomorphism and hence it corresponds to a gauge symmetry of the
supermembrane action.
r , r = 1, 2, are not unique as homotopic maps from onto S 1 S 1 , they
Although X
are all equivalent on the configuration space of the compactified supermembrane. Then one
realizes that the problem of controlling the closed, but not exact, one-forms in the quantum
analysis of the compactified M9 S 1 S 1 supermembrane has been solved. In fact the
r plus a homotopically trivial map to be quantized.
maps Xr decompose as the sum of X

The Xr will then be conveniently incorporated to the general description of the action. In
so doing, we will end up with a formulation of the theory as a symplectic noncommutative
YangMills action.
The zwei-vein
a
ab
era r b X
r,
W
W
allows us to write down all the geometrical objects in the spatial world volume in terms
of the corresponding objects in the tangent space. We can express the curvature F of any
, as
connection over the mentioned bundle (characterized by n) in terms of F
+ f,
F =F
(6)
where f = da satisfies f = 0. The one-form a is a one-form connection on a trivial

U (1) bundle, it has no transitions over and f is an exact two-form.
Let
a
Dr := r a .
W
Let Ar be such that
r + Ar .
Xr = X
348
Then, by computing dX1 dX2 , the decomposition (6) for (4) yields
+ F ,
F = F
where
F = rs Frs ,
Frs = Dr As Ds Ar + {Ar , As }.
It turns out, [9], that under the area preserving diffeomorphism, the residual gauge
symmetry of the supermembrane Hamiltonian in the light cone gauge, Ar , transforms as
Ar = Dr ,
where the covariant derivative
Dr = Dr + {Ar , }.
Then, the term Frs is interpreted as the curvature of a symplectic noncommutative
connection. The results of [11] describe the relationship between this connection and the
ones arising froma noncommutative product on the Weyl algebra bundle.
The condition f = 0 yields

(7)
F = 0.
This allows to write the Hamiltonian of the supermembrane, only in terms of Ar and Xm .
Identity (7) arises by imposing fixed central charge, or, analogously, by considering a fixed
U (1) principal bundle on . Hence, according to [9],

2

H = (1/2 W ) (Pm )2 + (r )2 + (1/2)W Xm , Xn

+

+ W Dr Xm + (1/2)W (Frs )2

(1/8) W n2 Dr r + Xm , Pm
(1/4)

W n F ,
n = 0
(8)
together with its fermionic contribution

W r Dr + m Xm , + { , } .
(9)
Here Pm and r denote the momenta conjugate to Xm and Ar , respectively. By we

denote the Majorana spinors of the D = 11 formulation which may be decomposed in
terms of a complex 8-component spinor of SO(7) U (1).
349
The above Hamiltonian describes a noncommutative Maxwell connection coupled to

the transverse scalar fields to the supermembrane world volume. The first class constraint
generating the area preserving diffeomorphisms realizes as the noncommutative Gauss
constraint. The presence of the integral of the noncommutative curvature is highly relevant,
since it explains why the nature of the spectrum in our model differs from the model
studied in [24]. Indeed, we can use the very same formulation (8) of the Hamiltonian,
which is also valid for the free winding supermembrane. We emphasize that this is the
exact expression for the Hamiltonian and not an approximation. In the free winding case,
Ar is a multi-valued connection over , unlike the fixed central charge case, where it is
a single-valued object. In this sense, our model corresponds to a restriction in the space
of all possible configurations
of the free winding case. In the fixed central charge model
considered here, the term W n F is a total derivative, hence its integral cancels out. In
this case the condition of zero Hamiltonian density in an open region implies both, zero
curvature and hence trivial Ar , and constant Xm . This ensures the absence of singular
configurations. These results were obtained in [1] in explicit manner. On the other hand,
when Ar is multi-valued, the latter term in expression (8) can be nonzero and it is not
difficult to verify that nontrivial singular configurations can arise in general. This is in
agreement with the results of [4].
In [1] we find an SU(N) regularized formulation of the Hamiltonian for a fixed
nontrivial central charge. In the case of free winding, the closed, but not exact, modes seem
not to fit in an SU(N) model, cf. [4]. Unfortunately, the present work does not contribute
to that situation. However, in the case of fixed central charge, the close one forms are
established in unique manner in terms of a given basis of homology. The configuration
space is then described in terms of exact one-forms and the regularization procedure leads
then to a model formulated in terms of SU(N)-valued geometrical objects. The resulting
model for a toroidal supermembrane is given explicitly by

1 0
0
0
0
2
2
P
T
P
T
+
T
+
(P
)
+
(
)
H = Tr
0
0
0
0
m
r
m
m
r
r
2N 3

2
i
n2 m n
2
n2
m
m
+
+
X ,X
TVr , X TVr Ar , X
16 2 N 3
8 2 N 3 N

2 1 2
n2
i
[TVs , Ar ]TVs [TVr , As ]TVr
+
+ n
[Ar , As ] +
16 2 N 3
N
8

i
n
+
Xm , Pm [TVr , r ]TVr + [Ar , r ]
4N 3
N

in
m Xm , r [Ar , ] + [ , ]
+
3
4N

i
r [TVr , ]TVr
(10)
N
subject to the constraint
A1 = A1(a1 ,0) T(a1 ,0) ,
1 ,a2 )
T(a1 ,a2 )
A2 = A(a
2
with a2 = 0.
(11)
350
Here A = (a1 , a2 ), where the indices a1 , a2 = 0, . . . , N 1 exclude the pair (0, 0),
V1 = (0, 1), V2 = (1, 0) and T0 T(0,0) = N I. We agree in the following convention
Xm = XmA TA ,
Ar = AA
r TA ,
Pm = PmA TA ,
r = rA TA ,
where TA are the generators of the SU(N) algebra:

C
TC .
[TA , TB ] = fAB
The condition (11) corresponds to a truncation of a gauge fixing of (8) and (9), when
the geometrical objects are given in a complete orthonormal basis of L2 (). Gauge fixing
conditions of the same kind were first used in [2].
By virtue of the results of [1] (see also [13]), we know that the regularized Hamiltonian
(10) has an associated mass operator with no string-like spikes. The gauge fixing conditions
together with the constraint
m

i
X , Pm = [Tv , r ]Tv + [Ar , r ] + [ , ] = 0
N
allow a canonical reduction of the Hamiltonian, cf. [13]. Therefore the conjugate pairs
a,b
Aa,b
1 , 1 ,
b = 0,
a,0
Aa,0
2 , 2 ,
do not appear in (10). After this reduction the terms |1a,b |2 , b = 0 and |2a,0 |2 become
nontrivial, however, since they are positive, we can bound the mass operator by an operator
without such terms. Notice that the imposition of the gauge fixing conditions (11)
together with the elimination of the associated conjugate momenta, is valid only in the
interior of an open cone K. Our model considers then wavefunctions with support on the
interior of K. In order to show that the spectrum of is discrete, we will consider as an
operator acting on the whole configuration space. The discreteness of the spectrum for the
latter implies the same property for the restriction to any hyper-cone, in particular K. We
notice that the assumption that the quantum problem is formulated only on an open cone
K of all configuration space, with Dirichlet boundary conditions, is also implicit in [2].
According to the results of [13], the bosonic part of the operator has compact
resolvent. Moreover, the bosonic potential is what we can formally call basin shaped.
We will prove in the forthcoming sections that the operator including the fermionic
sector, has compact resolvent and consequently the mass operator of the Hamiltonian (10)
has also a compact resolvent.
351
3. The fermionic potential

We consider the metric
0 1
1 0
1
ab =
..
1
in the light cone. In the 2 + 2 + 7 decomposition of the spacetime indices, =
( + , , 1 , 2 , i ) where

+ 2 2
=
= 0 and + , = 2I.
An appropriate representation is found by considering

= I I88 ,
where
1
(1 i2 ),
2
r
= 3 ir I88 ,
=
and
i
= 3 3 88
,
r = 1, 2,
i = 3, . . . , 9
i T
= i ,
in order to ensure that C i are symmetric. The charge conjugation is antisymmetric and it
is given by
C = 2 1 I88 .
The Majorana condition and the light cone gauge fixing condition associated to the kappa
symmetry
=TC
and + = 0,
respectively, allow to rewrite in terms of a complex spinor of SO(7)

i

=
,
0
0
where and + are anticommuting variables, and , = 1, . . . , 8 are SO(7) spin indices.
The canonical quantization rules for , considering the second class Dirac constraint on
the spinor fields, turns out to be

, = ,

{ , } = , = 0.
(12)
352
In terms of the matrix model, the canonical quantization reads

A
, B = BA ,
A B

, = A , B
= 0.
This allows us to compute the fermionic potential in terms of the fields. We have

m Xm = 2 i m Xm , + i m Xm ,
and in the matrix model
n A
fBC A m XmB C + T C m XmB A .
3
4N
The other terms are constructed in a similar manner. Spinors will be represented by
28D 28D matrices, with D equals to the dimension of the symmetry group, in this case
SU(N). The explicit construction of such matrices can be performed inductively. To avoid
dispersing from our main task, we show this procedure in detail in Section 6.
4. Basin shaped potentials

The criterion below extends the well-known fact that if a potential is bounded from
below and V (x) + as |x| , then the Hamiltonian d2 /dx 2 + V as a closed
operator acting on L2 (Rn ) has discrete spectrum (see for instance [15, Theorem XIII.16]).
For completeness of our exposition, we provide a self-contained proof of this result.
The Hamiltonian we shall consider is as follows. Let
H = + V
acting as a closed operator on L2 (RN ) Cn , where x Inn and V (x) = V (x)
Cnn , x RN . In addition, assume that V is measurable and satisfies V (x) c in the sense
that
V (x)w w c,
x RN , w Cn ,
where c R is constant. This ensures that the operator H can be defined by means of a
symmetric quadratic form and it is bounded from below. Once we have defined rigorously
operator H , the validity of the following criterion is ensured.
Lemma 1. Let vk (x) be the eigenvalues of V (x). If all vk (x) + as |x| , then the
spectrum of H is discrete.
Proof. Without loss of generality we assume that vk (x) 0. Since H is bounded from
below, one can apply the RaleighRitz principle to find the eigenvalues below the essential
spectrum. Let

T ,
,
m (T ) := inf sup
2
L
353
where the infimum is taken over all m-dimensional subspaces L L2 (RN ) Cn . Then the
bottom of the essential spectrum of T is limm m (T ). Notice that if this limit is +,
then the spectrum of T is discrete. Above

, = (x) (x) dx.
Rn
and 2 = , .
The hypothesis of the lemma is equivalent to the following condition: for all c > 0, there
exists a ball S (of possibly very large radius) such that
V (x)w w c|w|2 ,
Let
all x
/ S.
c, x S,
0,
x
/ S.
Then for all smooth and with compact support,
W (x) :=
V (x)(x) (x) c(x) (x) + W (x)(x) (x)

so that

V , (c + W ), .
Thus
m (H ) c + m ( + W )
for all l = 1, 2, . . . .
Since W (x) is a bounded potential with compact support, by Weyls theorem, the
essential spectrum of + W is [0, ). Therefore by RaleighRitz criterion, there exists
> 0 such that
M
m ( + W ) 1,

m M.
Thus, m (H ) c 1 for all m M. Since we can take c very large, necessarily m +

as m increases and the proof of the lemma is complete.
5. Discreteness of the spectrum

By using Lemma 1, we show in this section that the resolvent of the operator is
compact and hence it has discrete spectrum.
To this end, decompose as
= + VB I + VF ,
where VB and VF denote the bosonic and fermionic potentials, respectively. Then VF is
the sum of a linear homogeneous part M(X, A) corresponding to

in
m Xm , r [Ar , ]
4N 3
354
and a constant matrix C corresponding to

n
[T
,
]T
.
r
V
V
r
r
4N 4
Put T := + VB + M(X, A). Since T is Hermitian (a self-adjoint operator in its
domain), (T i) is invertible for all > 0 and we can choose large enough such that
the resolvent (T i)1 satisfies

(T i)1 C1 /2.
Here is the supremum norm for operators acting on Hilbert spaces. Hence

i = T + C i = (T i) I + (T i)1 C .
Because of (T i)1 C 1/2, the latter term is invertible and so

1
( i)1 = I + (T i)1 C (T i)1 .
Since the first term at the right-hand side is bounded, the resolvent of is compact if and
only if the resolvent of T is compact.
We apply Lemma 1 in order to show compactness for the resolvent of T . If we denote by
R the normal vectors in the configuration space so that X = R and A = R, according
to [13, 3], R = 0 is a double zero of VB (R, R) and
VB (R, R) kR 2
for some constant k > 0. The eigenvalues of the matrix
VB + M(X, A)
are the R such that

det VB M(X, A) = 0.
By virtue of the homogenicity of M, must satisfy

VB
I M(, ) = 0,
det
R
R > 0.
Therefore if are the eigenvalues of M(, ), then
= VB (R, R) + R .
Consequently, + whenever R . Notice that V is continuous, hence it is
automatically bounded from below. This ensures that the resolvent of T is compact as
a consequence of Lemma 1.
355
6. Matrix representation of spinors

In this final section we show how to construct a basis , = 0, . . . , n 1 of size
2n 2n satisfying the anticommutative relations

, = ,
{ , } = 0.
(13)
This ensures (12).

The entries of the matrices are either 0 or 1. Adopting a common notation in
combinatorics, the list

(m1 , n1 ) ; . . . ; (mj , nj )
denotes a matrix full of zeros except that it has either +1 or 1 at the entries
(m1 , n1 ), . . . , (mj , nj ). For l = 0, 1, . . . and > 0, let
B0 (l) = (2l + 1, 2l + 2)+ ,

B (l) = 2+1 l + 1, 2+1 l + 1 + 2 s ; . . . ;
+1

2 l + (2m + 1), 2+1l + (2m + 1) + 2 s ; . . . ;

+1
2 l + 2 1 , 2+1 l + 2 1 + 2 s ;
+1

2 l + 2, 2+1 l + 2 + 2 s ; . . . ;

+1
2 l + (2m + 2), 2+1l + (2m + 2) + 2 s ; . . . ;

+1
2 l + 2 , 2+1 l + 2 + 2 s
be blocks of size 2+1 2+1 , where the sign s of the nonzero entry (m, n)s of B (0) is
determined according to the rule
s = (1)#(q)2 (1)m+1 ,
m = 2q + 1 or m = 2q + 2
(14)
and the sign distribution of B (l) for l 1 is the same as that of B (0). The integer #(q)2
is the number of ones in the binary representation of q. Then we construct the desired basis
as the block diagonal matrices

= B (0); . . . ; B 2n /2+1 1 , = 0, . . . , n 1
of size 2n 2n . In order to illustrate this procedure take n = 4, then the basis is

0 = (1, 2)+ ; (3, 4)+ ; (5, 6)+; (7, 8)+ ; (9, 10)+; (11, 12)+; (13, 14)+; (15, 16)+ ,

1 = (1, 3)+ ; (2, 4) ; (5, 7)+ ; (6, 8) ; (9, 11)+ ; (10, 12) ;

(13, 15)+ ; (14, 16) ,

2 = (1, 5)+ ; (3, 7); (2, 6) ; (4, 8)+ ;

(9, 13)+ ; (11, 15); (10, 14); (12, 16)+ ,

3 = (1, 9)+ ; (3, 11); (5, 13); (7, 15)+; (2, 10); (4, 12)+; (6, 14)+; (8, 16) .
356
We now show that the 0 , . . . , n1 satisfy (13). In order to simplify notation, we

represent the product of two matrices whose rows and columns have only one nonzero
entry whose value is 1, as a signed permutation of the group of integer numbers. To be
more precise, it is easy to see that the only nonzero entries of this product are the pairs
(m, o)s with sign s = rt where the entries (m, n)t are nonzero with sign t on the first
matrix, and the entries (n, o)r are nonzero with sign r on the second matrix. This will be
t
r
described pictorially as m n o.
{ , } = 0 and { , } = I. Notice that each
= (m1 , n1 )s ; . . . ; (mj , nj )s ,
where each index mp or np is different from all other indices. This ensures the first identity.
The second identity is consequence of the fact that the only nonzero entries of the product
are represented by
s
mp np mp
and those of by
s
np mp np .
In order to show that { , } = 0 when = , it is enough to check this property
only when 0 < n 1 and = n 1, whereas the other cases follow from an inductive
argument. Furthermore since our basis consists of block diagonal matrices, we only have to
verify how the first block of , B (0), multiplies with the suitable elements of . Notice
that the only possibly nonzero entries of { , } are (m, n) consequence of the sum
s
mon
m p n.
(15)
We claim that this entry will also be zero, due to the fact that the product of signs stuv is
always negative. Let us show this claim.
{0 , n1 } = 0. Put
m = 1,
o = 2,
p = 1 + 2n1
n = 2 + 2n1 ,
in (15). The signs s and v are positive because of all the nonzero entries of 0 are equal to
one. According to the rule (14), t = (1)#(0)2 (1)2 and u = (1)#(0)2 (1)3 . This ensures
that the product of signs is always negative in this case.
{ , n1 } = 0, 0 < < n 1. If m is odd, the indices in (15) are
m = 2q + 1,
o = 2q + 1 + 2 ,
n = 2q + 1 + 2 + 2n1 ,
p = 2q + 1 + 2n1
for 0 q 21 1. Then
s = u = (1)#(q)2 = v,
t = (1)#(q+2
1 )
For the third inequality notice that by construction the sign rule of B (l) for l > 0 copies
the one of B (0). Since the binary representation of q has at most 22 digits, then
357
#(q + 21 )2 = #(q)2 + 1 and thus the product of signs is always negative. If m is even,
for reasons similar to the odd case, also s = u = v and t = s.
This completes the proof of our claim.
{ , } = 0, = . By using arguments involving the diagram (15), this can be shown
in a similar manner as the previous identity. One should take into account that due to the
transposition in the second term, the signs of s and v should be determined not from the
first but from the second entry of the suitable pair.
7. Conclusions
We have shown that the quantum Hamiltonian of the compactified supermembrane on
M9 S1 S1 with nontrivial central charge, that is with irreducible winding, has compact
resolvent and consequently its spectrum consists of a discrete set of eigenvalues with finite
multiplicity.
We have considered a regularized SU(N) model of the compactified supermembrane
with nontrivial central charge. The condition of having a nontrivial winding, determines a
sector of the full compactified supermembrane. In the explicit formulae we use a toroidal
supermembrane, however, the result still holds for other nontrivial topologies (the spherical
supermembrane has been considered recently in [18] in terms of an SU(2) model). Our
approach is valid for the analysis of the compactified supermembrane with target space
M9 S1 S1 . In this case the class of maps defining the configuration space of the
supermembrane are determined by the central charge which becomes proportional to n,
the winding number of the supermembrane. The existence of a nontrivial central charge
leads to a re-formulation of the problem in terms of a symplectic noncommutative superYangMills theory. In particular our regularized Hamiltonian is a consequence of this
construction.
The lemma we have established in Section 5, seems to be the appropriate strategy to
investigate any compactified supersymmetric models where no string-like configurations
are present. The assumptions on the bosonic potential are very mild, we only require the
potential to be measurable, bounded from below and unbounded above in every direction
(if the potential is continuous the unbounded assumption ensures the boundedness from
below). For instance, for a quantum mechanical potential of the form
V = VB (x)I + VF (x) C2
n 2n
where x RL , VB (x) is continuous with the asymptotic behaviour

VB cx2p ,
c > 0,
and the fermionic matrix potential satisfies

VF VF |x=1 xq ,
for all x > R0 with 2p > q, the Hamiltonian of the quantum system has spectrum
consisting exclusively of isolated eigenvalues of finite multiplicity.
358
Acknowledgements
We wish to thank M. Rosas for helping us with the combinatoric aspect of Section 6.
We also wish to thank J. Pea, M. Asorey, T. Ortn, P. Howe and K. Stelle for helpful
discussions.
References
[1] M.P. Garca del Moral, A. Restuccia, On the spectrum of a noncommutative formulation of the D = 11
supermembrane with winding, Phys. Rev. D 66 (2002) 045023, hep-th/0103261.
[2] B. de Wit, M. Lscher, H. Nicolai, The supermembrane is unstable, Nucl. Phys. B 320 (1989) 135159.
[3] B. de Wit, U. Marquard, H. Nicolai, Area-preserving diffeomorphisms and supermembrane Lorentz
invariance, Commun. Math. Phys. 128 (1990) 3962.
[4] B. de Wit, K. Peeters, J.C. Plefka, The supermembrane with winding, Nucl. Phys. B (Proc. Suppl.) 62 (1998)
405411, hep-th/9707261.
[5] R. Helling, H. Nicolai, Supermembranes and M(atrix) theory, AEI-093, hep-th/9809103.
[6] A. de Castro, A. Restuccia, Master canonical action and BRST charge of the M-theory bosonic five brane,
Nucl. Phys. B 617 (2001) 215236, hep-th/0103123.
[7] I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, Covariant action for the superfive-brane of M-theory, Phys. Rev. Lett. 78 (1997) 43324334, hep-th/9701149.
[8] M. Aganagic, J. Park, C. Popescu, J. Schwarz, World-volume action of the M-theory five brane, Nucl. Phys.
B 496 (1997) 191214, hep-th/9701166.
[9] I. Martn, J. Ovalle, A. Restuccia, Stable solution of the double compactified D = 11 supermembrane dual,
Phys. Lett. B 472 (2000) 7782, hep-th/9909051.
[10] I. Martn, J. Ovalle, A. Restuccia, Compactified D = 11 supermembranes and symplectic noncommutative
gauge theories, Phys. Rev. D 64 (2001) 046001, hep-th/0101236.
[11] I. Martn, A. Restuccia, Symplectic connections, noncommutative YangMills theory and supermembranes,
Nucl. Phys. B 622 (2002) 240256, hep-th/0108046.
[12] J.G. Russo, Supermembrane dynamics from multiple interacting strings, Nucl. Phys. B 492 (1997) 205222,
hep-th/9610018.
[13] L. Boulton, M.P. Garca del Moral, I. Martn, A. Restuccia, On the spectrum of a matrix model for the
D = 11 supermembrane compactified on a torus with nontrivial winding, Class. Quantum Grav. 19 (2002)
29512959, hep-th/0109153.
[14] B. Simon, Some quantum operator with discrete spectrum but classically continuous spectrum, Ann. Phys.
(N.Y.) 146 (1983) 209220.
[15] M. Reed, B. Simon, Methods of Modern Mathematical Physics, Vol. 4: Analysis of Operators, Academic
Press, New York, 1978.
[16] B. Kostant, Quantization and unitary representations, I: prequantization, in: Lecture Notes in Mathematics,
Vol. 170, Springer, Berlin, 1970, pp. 87208.
[17] M.J. Duff, T. Inami, C.N. Pope, E. Sezgin, K. Stelle, Semiclassical quantization of the supermembrane,
Nucl. Phys. B 297 (1988) 515.
[18] J. Conley, B. Geller, M.G. Jackson, L. Pomerance, S. Shrivastava, A quantum mechanical model of spherical
supermembranes, hep-th/0210049.

Plane-wave matrix theory from N = 4

super-YangMills on R S 3
Nakwoo Kim, Thomas Klose, Jan Plefka
Max-Planck-Institut fr Gravitationsphysik, Albert-Einstein-Institut,
Am Mhlenberg 1, D-14476 Golm, Germany
Received 18 June 2003; received in revised form 7 August 2003; accepted 13 August 2003
Abstract
Recently a mass deformation of the maximally supersymmetric YangMills quantum mechanics has been constructed from the supermembrane action in eleven-dimensional plane-wave backgrounds. However, the origin of this plane-wave matrix theory in terms of a compactification of
a higher-dimensional super-YangMills model has remained obscure. In this paper we study the
KaluzaKlein reduction of D = 4, N = 4 super-YangMills theory on a round three-sphere, and
demonstrate that the plane-wave matrix theory arises through a consistent truncation to the lowest
lying modes. We further explore the relation between the dilatation operator of the conformal field
theory and the Hamiltonian of the quantum mechanics through perturbative calculations up to twoloop order. In particular, we find that the one-loop anomalous dimensions of pure scalar operators
are completely captured by the plane-wave matrix theory. At two-loop level this property ceases to
exist.
PACS: 12.60.Jv; 11.25.-w; 11.25.Mj; 12.10.-g; 04.62.+v
1. Introduction
At present the most promising candidates for a microscopic description of M-theory are
given in terms of supersymmetric gauge quantum mechanical models, which are believed
to provide a light-cone quantization of M-theory in suitable backgrounds [1]. Concretely
in the case of a flat Minkowski background the associated matrix model is the maximally
supersymmetric YangMills quantum mechanics [2], obtained through the discretization
E-mail addresses: kim@aei.mpg.de (N. Kim), thklose@aei.mpg.de (T. Klose), plefka@aei.mpg.de
(J. Plefka).
doi:10.1016/j.nuclphysb.2003.08.019
360
N. Kim et al. / Nuclear Physics B 671 (2003) 359382
of the Minkowski background supermembrane action in light-cone gauge [3]. In the same
spirit a similar matrix model has been proposed in [4] for M-theory in the maximally
supersymmetric plane-wave background first discovered by KowalskiGlikman [5] in
1984,

3
9
9
2

2 2

2
2
2
dx +
xa +
xi
dx I
ds 2 = 2 dx + dx +
(1)
9
36
I =1
a=1
i=4
with non-vanishing four-form field strength F123+ = . The corresponding matrix model
is given by a mass deformation of the flat space matrix theory and takes the relatively
simple form in the conventions of [6]
S = Sflat + SM
with
(2)

1
1
2
2
I
Sflat = dt tr (Dt XI ) i Dt + [XI , XJ ] + [XI , ] ,
2
4

2

2

1 m
1 m
m
m
SM = dt tr
(Xa )2
(Xi )2 + i 123 + iabc Xa Xb Xc .
2 3
2 6
4
3

Here X and denote bosonic and fermionic Hermitian N N matrices, respectively.

The transverse SO(9) index I = 1, . . . , 9 is decomposed into a = 1, 2, 3 and i = 4, . . . , 9
reflecting the SO(3) SO(6) split of the transverse sector induced by the background
geometry. In the above m denotes the dimensionless parameter m = /(2R), R is the
radius of the compactified eleventh direction and the covariant derivative is given by
Dt O = t O i[, O] with the gauge field . For m 0 ones recovers the usual matrix
model in a flat background.
This model possesses sixteen dynamical and sixteen kinematical supersymmetries
inherited from the maximal supersymmetry of the plane-wave background
XI = 2 I (t),

1
m
m
= iDt XI I + [XI , XJ ] I J + iXa a 123 iXi i 123 (t) + (t),
2
3
6
= 2 (t)
with time-dependent parameters (t) = e 12 123 t 0 and (t) = e 4 123 t 0 .
Plane-wave matrix theory exhibits a number of interesting properties compared to the
model without mass deformation. The usual matrix theory has a continuous spectrum due
to flat potential valleys [7], whereas by virtue of the mass terms the energy spectrum
of the plane-wave matrix theory is discrete [4,8]. Furthermore, the introduction of the
dimensionless parameter m allows for a perturbative study of the spectrum of the planewave matrix theory for m
1 [8]. As noted independently in [6,9], the spectrum contains
an infinite series of protected multiplets, whose energies are exactly given by their free
field value, a property that may be shown using the representation theory of the classical
Lie superalgebra SU(2|4), the symmetry algebra of the M-theory plane-wave, as done in
[9,10]. For further works on the plane-wave matrix model see [11,12].
m
361
Now, whereas the higher-dimensional origin of the usual flat space matrix theory as
the trivial reduction of the maximally supersymmetric D = 4 YangMills theory to 1 + 0
dimensions (or equivalently as the reduction of N = 1 super-YangMills in 1 + 9 to
1 + 0 dimensions) is obvious by taking all fields to be space-independent, a similar
higher-dimensional origin of the plane-wave matrix theory has remained obscure. One
aim of this paper is to fill in this structural gap in the understanding of supersymmetric
gauge quantum mechanics: we shall point out that the plane-wave matrix theory arises
by considering N = 4, D = 4 super-YangMills compactified on a three-sphere and
performing a consistent truncation of the resulting KaluzaKlein spectrum. It turns out
that such a truncation to a finite number of fields is only possible if one drops half of the
vector and fermion zero modes, as we describe in detail. Upon performing this truncation
we obtain a relation between the four-dimensional YangMills coupling constant gYM and
the mass parameter m of the matrix model

m
3
3
=
32 2
.
2
gYM
(3)
Furthermore, as the radial quantization of N = 4 super-YangMills on R S 3 relates the

energy to the dilatation operator of the conformal field theory, one might speculate on
a relation of the spectrum of the truncated model to the scaling dimensions of composite
operators in the full super-YangMills field theory. We shall show that indeed the full one
loop scaling dimensions of scalar operators in the four-dimensional gauge field theory are
reproduced by the massive gauge quantum mechanics (2) in leading order perturbation
theory. This analogy, however, breaks down at the two-loop level.
Having established the relation between the plane-wave matrix model and N = 4,
D = 4 super-YangMills, it is natural to seek for its possible implications on the AdS/CFT
correspondence. The study of pure scalar operators is particularly relevant to the conjecture
of Berenstein, Maldacena and Nastase (BMN) [4] which involves operators whose
conformal dimensions and U(1) R-charges are taken to infinity. Based on holographic
arguments applied to the dual IIB plane-wave superstring it was claimed [13] that the
gauge field theory should have an effective one-dimensional description in the BMN limit.
In fact, it was proposed that this effective quantum mechanical model arises as a Kaluza
Klein reduction of super-YangMills on R S 3 , which is precisely what we are interested
in. The result of this paper shows that the plane-wave matrix model serves this proposal
at one-loop level, but not at higher loops. Put differently and slightly amusingly: the
effective one-dimensional description of BMN gauge theory is the BMN matrix model
in the BMN limitat one-loop level and in the pure scalar sector. Whether next-to-leading
order corrections can be also succinctly translated into the framework of matrix quantum
mechanics is a very interesting problem which lies beyond our scope in this paper.
The paper is organized as follows. In Section 2 we consider the KaluzaKlein reduction
of super-YangMills on a three-sphere, and illustrate how it is related to the plane-wave
matrix model. The effective vertices for pure SO(6) bosonic excitations are constructed
perturbatively in Section 3 up to next-to-leading order in perturbation theory. Our notation
and some details of the computations are given Appendices AE.
362
2. N = 4 super-YangMills on RS 3
The field content of D = 4, N = 4 superconformal YangMills theory consists of
a vector field A , six real scalars i as well as four Weyl spinors A all in the adjoint
representation of the gauge group. The action of the theory on Minkowski space-time may
be obtained through a trivial dimensional reduction of N = 1 supersymmetric YangMills
theory in D = 10 dimensions on a six-torus. When the theory is formulated on a curved
background, superconformal symmetry is retained if the background admits (conformal)
Killing vectors and spinors which generate the superalgebra SU(2, 2|4). In addition there
are deformations of the flat space action and the supersymmetry transformation rules
induced by the curved background, as discussed by [14,15] and [16]. Most notably in
a curved background the scalars become massive through a coupling to the Ricci scalar,
reflecting the fact that in curved spaces the dAlembertian operator alone is not Weyl
invariant.1 In a notation where the SO(1, 3) SO(6) split of the D = 10 Dirac matrices
has been performed, the action reads

1
1
R
1
2
d 4 x |g| tr F F D i D i i2 + [i , j ]2
S= 2
4
2
12
4
gYM

2iA D A + (i )AB A i 2 i , B

i AB (A )T i 2 i , B .
(4)
We use coordinates x = (t, , , ) labeled by , , . . . = 0, 1, 2, 3. When referring to
spatial coordinates xa = (, , ) only, we use the (curved) indices a, b, . . . = 1, 2, 3. The
gauge-covariant derivative is defined as D = i[A , ], where denotes the space
time-covariant derivative. Moreover, the field strength is given by F = A A
i[A , A ]. We use the metric

ds 2 = g dx dx = dt 2 + R 2 d 2 + sin2 d 2 + sin2 sin2 d 2
(5)
on R S 3 , where the radius of the three-sphere R is kept as a free parameter. For this
background the Ricci scalar is R = 6/R 2 . In (4) we have introduced := (1, a ) where
a are the usual Pauli matrices pulled back onto the S 3 , and ClebschGordan coefficients
(i )AB of SU(4) which relate two 4s with one 6 (cf. Appendix A for our conventions and
how the (i )AB are related to the SO(1, 9) Dirac matrices).
The action (4) is invariant under the following modified supersymmetry transformations

A ,
A = 2i A A A
(6)

2

T
i = 2i A i (i )AB B A i 2 i AB B ,
(7)

1
i
A
A = F A D i i 2 (i )AB B [i , j ] i j B B
2
2
1
i i 2 (i )AB B ,
(8)
2
1 In d dimensions the conformally-invariant wave operator is d2 R, where R is the Ricci scalar.
4(d1)
363
Table 1
Spherical harmonics on S 3 that appear in the expansion of spin-0, spin-1/2 and spin-1 fields. Here k = 0, 1, 2, . . .
labels different irreducible representations of SU(2). The index I enumerates the elements of a particular
representation and therefore takes values from 1 to the dimension of the representation
Spin
Irreps
Masses
Scalar spherical harmonics:
kI
Y(0)
(k + 1, k + 1)
1/2
Spinor spherical harmonics:
kI +
Y(1/2)
(k + 2, k + 1)
(k + 1)/R

k + 32 /R
kI
Y(1/2)
(k + 1, k + 2)
kI +
Y(1)
(k + 3, k + 1)
kI
Y(1)
(k + 1, k + 3)
Harmonical functions
Vector spherical harmonics:
(k + 2)/R
where := (1, a ) and := 12 ( ). The last term in (8) represents

the modification with respect to the flat space transformation law. The supersymmetry
parameters A are four Weyl spinors that satisfy either of the two conformal Killing
spinors equations:
i
.
(9)
2R
Counting the components of the Killing spinors and taking into account the degeneracy
of the solutions of (9) as well as the two signs yields the number of 32 independent
real supersymmetry parameters. The conserved charges corresponding to + and are
denoted by QL and QR , respectively. We finally note that the conformal Killing spinors
can also be obtained from the Killing spinors of AdS5 restricted on the boundary R S 3 .
=
2.1. Harmonic expansion on S 3

In order to perform the dimensional reduction we expand the four-dimensional fields
in terms of the spherical harmonics on S 3 , which come in irreducible representations
(mL , mR ) of the isometry group SO(4)
= SU(2)L SU(2)R . The set of harmonics that
appear in the expansion of a particular field depend on its spin and are listed in Table 1.
See also [17] for a useful discussion on this topic.
We work in Coulomb gauge on S 3 , a Aa = 0, and have the following mode
expansions
i (x) =
(k+1)

kI
ikI (t)Y(0)
(x),
(10a)
k=0 I =1
A
(x) =
(k+1)(k+2)

k=0
A0 (x) =
I =1
(k+1)

k=0 I =1
kI
A,kI (t)Y(1/2)
(x),
(10b)
kI
kI (t)Y(0)
(x),
(10c)
364
Aa (x) =
(k+1)(k+3)

k=0
I =1
kI
AkI (t)Y(1)
a (x).
(10d)
kI
are two-dimensional commuting Weyl
Note that the spinor spherical harmonics Y(1/2)
spinors. Upon inserting the above harmonic expansions into the action (4) and carrying out
the integration over the three-sphere one obtains a one-dimensional theory with an infinite
number of fields. In order to determine the mass spectrum of these excitations we shall
perform this integration for the quadratic terms of the action (4). For this we need only
a few properties of the spherical harmonics. They are orthonormalized to

kI lJ
kI lJ
kl I J
Y(1/2) Y(1/2) = kl I J ,
Y(0) Y(0) = ,
S3
S3
kI lJ a
Y(1)a
Y(1) = kl I J ,
(11)
S3
and their eigenvalues of the LaplaceBeltrami operators are

1
kI
k(k + 2)Y(0)
,
R 2
3 kI
i
kI
k+
Y
/ Y(1/2)
=
,
R
2 (1/2)

1
3 kI
2 kI
Y(1/2) = 2 k(k + 3) + Y(1/2) ,
R
4

1
kI
kI
= 2 k(k + 4) + 2 Y(1)a
.
2 Y(1)a
R
kI
2 Y(0)
=
(12a)
(12b)
(12c)
(12d)
The above operators include, of course, only the spatial parts

/ := a a and 2 := a a .
Upon inserting the mode expansions (13) into the quadratic part of the action (4), one
obtains

1
4 2 R 3
(k + 1)2
kI kI
tr ikI ikI
Squadratic = 2
tr
dt
i
i
2
2R 2
gYM
k,I

1
(k + 2)2
kI kI
kI kI
tr A A
+
tr A A
2
2R 2
k,I,

k + 32
kI A,kI
kI A,kI
tr A
tr A
,
+
i
R
k,I,
(13)
where we have made use of the orthonormality conditions, integration by parts and the
properties (12). In the computation of the masses for the vector modes one also needs the
kJ
transversality of the vector spherical harmonics, a Y(1)a
= 0, and the identity
lJ a
lJ a
lJ c
lJ a
a b Y(1)
= [a , b ]Y(1)
= R a cab Y(1)
= Rab Y(1)
=
2 lJ
Y
R 2 (1)b
(14)
365
Fig. 1. KaluzaKlein particle spectrum of N = 4 super-YangMills on R S 3 . The states are labeled by

representations of SU(2)L SU(2)R SU(4). One climbs upward to the left by acting with the supercharge
and upward to the right with QR = (1, 2, 4). The encircled states comprise the consistent
QL = (2, 1, 4)
truncation to the plane-wave matrix theory.
using Rab = R22 gab of S 3 . The obtained mass spectrum is summarized in Table 1.
In Fig. 1 we present the resulting KaluzaKlein mass tower up to mass 3/R. One
may climb up the various states of the tower by acting with the two supercharges
(to the upper-left) and QR = (1, 2, 4) (to the upper-right). Note that
QL = (2, 1, 4)
unlike the trivial dimensional truncation on flat spaces, where superpartners have the same
masses, the superconformal transformations in our curved background geometry relate the
entire tower of KaluzaKlein modes. In other words, we find that the infinite tower of
KaluzaKlein modes is not decomposed into finite-dimensional irreducible representations
of the superconformal algebra. Instead, the entire tower itself is a single irreducible
representation.
On the other hand, if we consider only half of the supercharges, say QL , which together
with the bosonic symmetries generate the subalgebra SU(2|4), the KaluzaKlein modes
are decomposed in terms of finite-dimensional irreducible representations. We see that
(1, 1, 6) + (2, 1, 4) + (3, 1, 1) (encircled in Fig. 1) is the lowest such supermultiplet.
Higher ones in general branch into five irreducible representations under the bosonic
subalgebra SU(2) SU(4), implying that they are all short supermultiplets of SU(2|4). In
particular, the lowest-lying multiplet consisting of three floors is 1/2 BPS and the following
multiplet comprised of four floors is 1/4 BPS, as may be deduced from the results of [10].
An important result of the supersymmetry algebra is that the ground state energy given as
the sum of zero point energies should vanish, which we can easily verify. For instance,
366
for the lowest-lying supermultiplet (1, 1, 6) + (2, 1, 4) + (3, 1, 1) the summation goes as
follows:
3
2
1
+ 3 = 0.
6 8
(15)
R
2R
R
The truncation we are about to perform consists precisely in the restriction to this lowest
lying multiplet.
2.2. Derivation of plane-wave matrix theory
We now aim to show that the infinite tower of KaluzaKlein states can be consistently
truncated to the lowest lying supermultiplet, and that the one-dimensional action, which
governs its dynamics, is precisely the plane-wave matrix model. Hence we restrict the
expansion (10) to the zero modes which are SU(2)R singlets. For these modes the harmonic
, and three Killing vectors V a+
functions are given by a constant, two Killing spinors S+

a .
The two Killing spinors obey the defining equation
i
(a ) S+
,
2R
=
a S+
= 1, 2
(16)
via2
and lead to the three positive chirality Killing vectors Vaa+
S +
a S + = (a ) Vaa+
.
(17)
Their explicit form along with further useful properties may be found in Appendix B. The
truncated mode expansion then takes the form
i (x) = Xi (t),
A
(x) =
2

=1
A
(t)S (x),
A0 (x) = (t),
Aa (x) =
3
Xa (t)Vaa+
(x).
(18a)
(18b)
(18c)
(18d)
a=1
In order to prove that the above ansatz is a consistent truncation of the spectrum, we
have to insert this ansatz into the four-dimensional equations of motion and show that this
leads to a set of equations without contradictions. The equations of motions which follow
from the four-dimensional action (4) are given by

D F + i[i , D i ] + 2 , = 0,
(19a)

1
D D i 2 i + [j , [i , j ]] , i 2 i + T , i 2 i = 0,
(19b)
R

i D i 2 i i , = 0,
(19c)
2 The Killing vectors of opposite chirality V a

a are induced from the opposite chirality Killing spinor bilinear
S
a S
which obey (16) with an opposite sign.
367
where the anticommutator of two fermions is defined as { T , }A := if ABC ( T )B C .

Inserting the ansatz (18) is a straight-forward computation which we comment on in
Appendix C. One finds that the S 3 coordinate dependences of all terms in each equation
exactly matches, and that the D = 4 equations of motion are satisfied provided that the
one-dimensional fields XI (t), (t) and (t) obey

[XI , iDt XI ] 2 , = 0,
(20a)

4
6i
Dt2 Xa + 2 Xa a b c Xb Xc [XI , [Xa , XI ]] 2 , a = 0,
(20b)
R
R

1
Dt2 Xi + 2 Xi [XI , [Xi , XI ]] + , i 2 i T , i2 i = 0,
(20c)
R

3
iDt
(20d)
+ [Xa , a ] Xi , i2 i = 0.
2R
Note that the label of the fermion zero modes A
which used to account for their
degeneracy has turned into a proper (Weyl) spinor index. Having found the equations of
motion (20) for the zero modes, one may try to find a quantum mechanical action which
leads to these equations of motion. It can be checked easily that these equations are derived
from the following one-dimensional Lagrangian

1
1 m 2 2 1 m 2 2 m
1
2
L = tr (Dt XI )
Xa
Xi + iabc Xa Xb Xc + [XI , XJ ]2
2
2 3
2 6
3
4

m
a
2
2i Dt + 2 [Xa , ] + i i Xi ,
2
T i 2 i [Xi , ] ,
(21)
where the radius of the three-sphere R has been traded for a mass parameter m = 6/R.
This Lagrangian is in fact nothing but the plane-wave matrix theory (2) written in terms of
SU(2) SU(4) spinor variables. This dimensional split was performed for instance in [12],
in Appendix A we repeat this split using our conventions and notation for easier reference.
This result also induces a relation of the four-dimensional YangMills coupling constant
gYM to the mass parameter m of the plane-wave matrix model. It is found by requiring the
prefactor of the reduced super-YangMills action (13) to match the unit prefactor of the
matrix model action (2). Taking into account that m = 6/R one finds

3
m
32 2
(22)
= 2 .
3
gYM
3. Perturbation theory
Having established this connection of D = 4, N = 4 super-YangMills to the planewave matrix model one might wonder what additional structures of the field theory carry
over to the matrix quantum mechanics. By virtue of the state-operator map of conformal
field theories one might expect a connection of scaling dimensions of super-YangMills
368
operators on R4 to the energy of corresponding states in the plane-wave matrix model.

As noted in [6,9] one such connection is already manifest: the protected multiplets of
chiral primary operators in the gauge theory, their lightest representative being the pure
multi-scalar operators OCPO = ci1 ...in tr(i1 in ), with ci1 ...in symmetric and traceless,
have vanishing anomalous dimensions, i.e., their scaling dimension is exactly given by
their free-field value OCPO = n. This property is paralleled by the existence of protected
multiplets in the plane-wave matrix theory, with their lightest member again being the
symmetric traceless excitations ci1 ...in tr(ai1 ain )|0 in the SO(6) sector of the matrix
model (2) [6,9]. The energy of these states is similarly not corrected by interactions
and simply given by the free theory value E = n m6 to all orders in perturbation theory.3
So here a complete analogy exists and it is natural to ask whether it extends to nonprotected states in the SO(6) sector as well. That is, is there a general relation between
the scaling dimension in 4d and the quantum mechanical energy? As we will show in the
following this analogy extends to the one loop level in the SO(6) sector, but breaks down
at two loops. Technically what we demonstrate is the precise matching of the effective
quantum mechanical interaction vertex in first order perturbation theory with the one loop
piece of the dilatation operator of D = 4, N = 4 super-YangMills computed in [18,19].
Comparing our second-order perturbation theory result to the recently established two-loop
piece of the dilatation operator [21] we find a discrepancy, although the same structure of
terms does appear. One might still hope that this discrepancy disappears once one considers
the BMN limit of the gauge theory, however, this turns out to not be the case.
Let us then reconsider the perturbative evaluation of energy shifts in the SU(N) planewave matrix model discussed in [6,8]. The Hamiltonian associated to (2) may be split into
a free and an interaction piece. Working in the conventions of [6] the free piece reads
(a = 1, 2, 3; i = 1, . . . , 6; = 1, . . . , 16)

m
m
m
ai ai + aa aa + +
H0 := tr
(23)
6
3
2
with the matrix oscillators

m
3
ai :=
Pi i Xi ,
m
6

3
m
aa :=
Pa i Xa ,
2m
3
:= ,
1
:= (1 i123) ,
(24)
obeying

1
(ai )rs , aj t u = ij st ru rs t u ,
N

1
(aa )rs , ab t u = ab st ru rs t u ,
N

+ 1
1
rs , t u = st ru rs t u .
2
N

(25)
3 Note that only the energy eigenvalue is protected, the eigenstates are renormalized in perturbation theory.
369
The interacting piece of the Hamiltonian is comprised of cubic and quartic terms in the
fields
Hint := V1 + V2 ,
(26)
where
m
iabc tr Xa Xb Xc tr I [XI , ],
3
1
V2 = tr[XI , XJ ][XI , XJ ],
4
and I is a joint index I = (a, i). For m
1 the interaction piece Hint may be treated
perturbatively.4 The first-order energy shift of a state |0 reads
(1)
1 |0 0 |
(1)
E1 = 0 Veff 0 with Veff = V1 V1 + V2 and =
, (27)
E0 H 0
where we assume 0 |0 = 1 and a non-degenerate free energy E0 . As we shall only
(1)
investigate the energy shifts of pure SO(6) excitations we will normal order Veff and
drop all normal ordered terms containing fermions or SO(3) oscillators, as these would
annihilate a pure SO(6) excitation state. We indicate this simplification by an arrow. Using
M ATHEMATICA and F ORM [20] we find
12N
99 3
V1 V1
(28)
N N
: tr ai ai :,
2
4M
M2

13N

99 3
1
V2
N N +
: tr ai ai : +
: tr ai , ai aj , aj :
2
2
2
4M
M
2M

1
1
(29)
: tr ai , aj ai , aj : 2 : tr ai , aj [ai , aj ]:
2
2M
M
introducing the shorthand M := m/3 which controls the perturbative expansion. When
adding these contributions all constants sum to zero

1
1
1
(1)
Veff 2 N: tr ai ai : + : tr ai , ai aj , aj : : tr ai , aj ai , aj :
M
2
2

: tr ai , aj [ai , aj ]:
(30)
V1 =
a manifestation of the vanishing energy shift of the groundstate |0. This expression can be
simplified further as the state |0 acting upon it is gauge invariant, e.g., some multi-trace
excitation. The simplification concerns all terms with a commutator of contracted creation
and annihilation operators: [ai, ai ], i.e., the second term in the above. One splits off this
commutator from the rest and removes the normal ordering between the two factors:

1
1
: tr ai , ai aj , aj : = : tr ai , ai T A tr T A aj aj :
2
2

1
= : tr ai , ai T A : : tr T A aj , aj : N: tr ai ai :,
(31)
2
4 This is easily seen by performing a rescaling X X/m and /m, rendering the Hamiltonian into
the form H = H 0 + 12 V1 + 14 V2 with no m-dependence in H 0 , V1 and V2 .
m
m
370
where T A denote the generators of SU(N) and we note that

1
tr T A AT A B = tr A tr B tr(AB),
(32)
N

1
tr T A A tr T A B = tr(AB) tr A tr B.
(33)
N
The first term in (31) can be dropped for reasons given in the Appendix D and the second
term cancels the two-point interaction that was already present. Hence, we arrive at the
compact result

1
1
(1)
Veff 2 : tr ai , aj ai , aj : : tr ai , aj [ai , aj ]: .
(34)
M
2
We want to compare this 1-loop effective vertex, which determines the first order energy
shifts of states in the plane-wave matrix model, to the 1-loop dilatation operator of N = 4
super-YangMills, which determines the anomalous dimensions of operators. We take the
dilatation operator from [21]. The 1-loop contribution is given in Eq. (1.6):

g2
1
D2 = YM2 : tr[i , j ][i , j ]: : tr[i , j ][ i , j ]: ,
(35)
2
16
where i :=
d
di .
ai =
i ,
This shows that by identifying the scalars of both theories

ai =
i ,
(36)
we have an exact agreement between energy shifts and anomalous dimensions for all pure
SO(6) states/operators at one loop level!
This result is intriguing in view of the fact [22] that the 1-loop dilatation operator
D2 of N = 4 super-YangMills in the strict N limit may be viewed as the
Hamiltonian of an integrable SO(6) spin chain model, hinting at the integrability of N = 4
super-YangMills. This structure thus carries over to the plane-wave matrix model. The
higher-loop corrections to the super-YangMills dilatation operator represent non-standard
deformations of this spin chain model as discussed in [21]. The fact that the 2-loop effective
vertex in the plane-wave matrix quantum mechanics to be discussed below departs from the
super-YangMills dilatation operator indicates that one is facing an alternative deformation
of the spin chain in the plane-wave matrix theory, iff the integrability is indeed conserved
at higher loop level. Including the fermionic and SO(3) excitations one might expect
to uncover an integrable spin chain structure of plane-wave matrix theory based on the
underlying supergroup SU(2|4). It is certainly interesting to pursue these issues further.
Moving on to the 2-loop energy shift we are faced with the evaluation of the effective
(2)
vertex Veff
(2)
E2 = 0 |Veff
|0
(37)
with
(2)
Veff
= V1 V1 V1 V1 + V1 V1 V2 + V1 V2 V1 + V2 V1 V1
+ V2 V2 V1 V1 P V1 V1 V1 V1 P V2 ,
(38)
371
where
1 |0 0 |
,
P = |0 0 |.
(39)
E0 H 0
Here some information on the particular forms of V1 and V2 entered in the derivation of
(38): All terms have been omitted that contain an odd number of fields of one kind, as they
do not contribute in the computation of expectation values. We again project onto pure,
(2)
gauge invariant SO(6) excitations acting on Veff . The resulting expression is, however,
still rather lengthy and has been relegated to Appendix E, Eq. (E.3). Let us then exemplify
our claim of the disagreement at 2-loop level by considering the energy shifts of an explicit
state.
Consider the Konishi state in the plane-wave matrix theory
=
|K :=
1
12(N 2
1)
tr ai ai |0.
(40)
We then find
E0 = K|H0 |K = M,
(41)
12N
(1)
E1 = K|Veff |K =
(42)
,
M2
228N 2
(2)
.
E2 = K|Veff |K =
(43)
M5
This is to be compared to the scaling dimension of the Konishi operator K = tr(i i ) of
N = 4 super-YangMills [23]
2 N
4 N2
3gYM
3gYM
(44)
+ .
4 2
16 4
As the overall constant of the Hamiltonian is not fixed we can only compare the ratios of
the energy shifts Ei to the free energy E0 with the ratios of the anomalous dimensions
i to the engineering dimension 0 .

M
456N 2
24N
E|K =
(45)
.
2+
2
M3
M6
K = 2 +
Inserting the uncovered relation between the plane-wave matrix theory mass parameter
M = m3 and the gauge theory coupling constant gYM of (3) into (44)
2
gYM
1
=
M 3 32 2
(46)
the 1-loop correction indeed agrees.5 Using this relation in the 2-loop energy shift we
would predict the following 2-loop anomalous dimension for the Konishi field from the
5 An alert reader might think that in choosing the relation (46) one does not have an independent check of the
1-loop agreement. This is of course true. The point here is that after choosing (46) all further 1-loop corrections for
higher excited states are fixed and do agree with the corresponding 1-loop super-YangMills scaling dimensions.
This is a consequence of the matching of (34) with (35).
372
matrix quantum mechanics

3g 4 N 2
19
YM 4
8
16
(47)
which is off by a factor of 19/8.

3.1. The BerensteinMaldacenaNastase limit
Recently there has been considerable interest in a novel double scaling limit of SU(N)
D = 4, N = 4 super-YangMills following the work of Berenstein, Maldacena and Nastase
[4] where one takes
J2
(48)
and gYM fixed.
N
Here J is the U(1) R-charge associated to the complex combination of two of the six scalar
fields i , e.g.,
N ,
J ,
with
1
Z = (5 + i6 ).
2
(49)
This limit leads to a dual gauge theory description of the IIB plane-wave superstring. It
represents an extreme reduction of the field theory: only two- and three-point functions
of so-called BMN operators, having large scaling dimensions and U(1) R-charges
0 > J
1, exist [19]. It also appears at this stage, that the only sensible observable of
BMN gauge theory is the dilatation operator. Moreover, holographic arguments indicate
that the gauge theory dual of the IIB plane-wave superstring should be given by a onedimensional model and it was proposed [13] that this effective quantum mechanical model
indeed arises as the KaluzaKlein reduction of D = 4, N = 4 super-YangMills on RS 3 ,
which we have considered in Section 2. As long as one restricts ones attention to the pure
scalar SO(6) sector we have seen that the plane-wave matrix theory serves this purpose
at the one-loop level. In principle, there is the logical possibility that although this match
of energies and scaling dimensions ceases to exist at higher loop level for the full superYangMills theory, it does hold for the BMN sector of the gauge theory at higher loops.
For convenience we further reduce the SO(6) excitations under consideration in the
plane-wave matrix model, to solely two complex combinations

1
Z := a5 + ia6 ,
2

1
:= a3 + ia4 ,
2
1
Z := (a5 ia6 ),
2
1
:= (a3 ia4),
2
(50)
(51)
which will make our formulas more transparent. The energy shifts of excitations made
entirely from the creation operators Z and are then governed by the simple effective
vertices
H0
: tr(Z Z + ):,
2
(52)
373
2
]:,
: tr[Z, ][Z,
(53)
M2
22N
(2)
]:
Veff
: tr[Z, ][Z,
M5

4
[Z, [Z,
]]]:
]]]:
[, [Z,
(54)
+ 5 : tr[Z, ][Z,
+ : tr[Z, ][,
.
M
Let us stress that no limit has been performed yet, Eq. (54) follows directly from (E.3),
upon restricting its action to states given by excitations in gauge-invariant words made of
Zs and s only.
This result may be directly compared to the two-loop structure of the D = 4, N = 4
super-YangMills dilatation operator in the analog scalar sector, which has been computed
in [21] Eq. (5.5)
(1)
Veff
D0 = : tr(Z Z + ):
2

g
]:
D2 = YM2 2: tr[Z, ][Z,

16
4

gYM
]:
+ 2: tr[Z, ][Z,
[Z, [Z,
]]]:
D4 =
4N: tr[Z, ][Z,
2
2
(16 )

]]]:
[, [Z,
+ 2: tr[Z, ][,
(55)
(56)
(57)
The agreement of the one-loop terms was already observed for general SO(6) excitations
(2)
in the last subsection. However, the closeness of the two-loop result D4 to Veff is striking:
the same structure of terms appears, only one relative factor is wrong! Remarkably, not all
terms of (57) are indeed relevant in the BMN limit (48), as shown in [21]. However, it is
precisely the last term in (57) which is irrelevant in the BMN limit when acting on states
(operators) of the form tr(Z p Z J p ) in the BMN limit (48).
So even if one considers the BMN limit (48) of the next-to-leading order effective vertex
of the plane-wave matrix model (54), the two-loop discrepancy to the scaling dimensions
of the four-dimensional gauge theory persists.
The observed two-loop discrepancy of plane-wave matrix theory and N = 4 superYangMills should be understandable in the framework of Wilsonian effective quantum
field theory, where one integrates out all non-SU(2)R -singlets in a perturbative fashion.
The resulting effective action of the SU(2)R singlet modes will start out with the planewave matrix model, but the inclusion of the non-SU(2)R -singlet modes in loops will lead to
a renormalization of the matrix model mass parameter M as well as higher order interaction
terms in the plane-wave matrix model. As is evident from our results at one-loop order
these corrections are not yet effective, but apparently start to contribute beyond this order.
It would be very interesting to study this in detail, in particular what restrictions on these
higher order interaction terms arise from the underlying supersymmetry.
4. Discussion
In this paper we have shown how the mass deformed gauge quantum mechanics of
plane-wave matrix theory arises from the KaluzaKlein reduction of N = 4, D = 4 super-
374
YangMills theory on R S 3 . Whereas the ordinary gauge quantum mechanics is obtained

through a trivial dimensional reduction of any higher-dimensional YangMills theory with
maximal supersymmetry, the mass deformed supersymmetric quantum mechanics at hand
is intrinsically related to four-dimensions, as the symmetry algebra SU(2|4) suggests.
We went on to explore the relation of the spectrum of the plane-wave matrix model to
the scaling dimensions of N = 4, D = 4 super-YangMills operators. At the free theory
level the two are equivalent, provided one considers YangMills operators without higher
space derivatives. Once interactions are turned on one would in general expect a different
behavior of the two quantities. However, for the special class of operators which are
protected from perturbative corrections the analogy persist: the two theories share the
same 1/2-BPS multiplets, whose primaries are given as totally symmetric traceless SO(6)
tensors. We have seen in this paper that the analogy also extends to the one-loop level
for pure scalar operators and their supersymmetry descendants, but ceases to exist beyond
that. It is natural to ask whether this leading-order agreement will turn out to be true for the
entire spectrum of the massive gauge quantum mechanics as well, and we hope to address
this issue in the future.
The discussions presented in Section 2 were purely classical, which implies that we
can in fact start with any supersymmetric field theory with classical superconformal
invariance and consider KaluzaKlein reductions on R S 3 to obtain a mass-deformed
supersymmetric gauge quantum mechanics. Some of them might also be given a M-theory
interpretation. Let us take the example of pure N = 2 YangMills theory, whose trivial
dimensional reduction gives a gauge quantum mechanics with 8 supersymmetries and
SO(5) symmetry. Put on R S 3 , the mass deformation will induce the symmetry breaking
SO(5) SO(3) SO(2). If we add one hypermultiplet in the adjoint representation
one recovers the plane-wave matrix model. If one adds further hypermultiplets in the
fundamental representation one obtains the matrix model of supersymmetric M5-branes
in a plane-wave background, which are extended along four of the SO(6) directions
as well as two light-cone directions. The supersymmetry of the relevant M5-brane
configurations in the plane-wave background is elucidated in [24]. As an alternative avenue
for generalization one can also consider N = 1 field theories with nontrivial target spaces
for chiral multiplets, i.e., gauged nonlinear sigma models. The relevant massive quantum
mechanical models could give matrix models of eleven-dimensional pp-waves with curved
transverse spaces, which are analogs of ten-dimensional solutions studied in [25,26].
In this paper we made use of the specific formulation of N = 4, D = 4 super-Yang
Mills on R S 3 , but it is certainly quite desirable to carry out a systematic study of
supersymmetric field theories on curved backgrounds in various dimensions, which do
not allow parallel spinors. This has been initiated by Blau in [16], where pure N = 1
super-YangMills theories in D = 10, 6, 4, 3 and their trivial dimensional reductions to
intermediate lower dimensions are considered. He finds that by adding mass and Chern
Simons-like terms the actions can be made supersymmetric with respect to generalized
Killing spinor equations on Einstein manifolds.
Finally, let us stress once more the emergence of an integrable SO(6) spin chain
structure at leading order perturbation theory of the plane-wave matrix model, in analogy
to the situation found at one-loop in D = 4, N = 4 super-YangMills by Minahan and
Zarembo [22]. It is certainly worthwhile to explore the origin and implications of this
375
symmetry further, possibly leading to the integrability of the complete model. One would
expect this program to be simpler than in the full four-dimensional case, that is, if it is
performable at all.
Acknowledgements
We would like to thank G. Arutyunov, N. Beisert, H. Nicolai, S. Kovacs, S. Metzger,
A. Pankiewicz, M. Pssel, T. Quella and M. Staudacher for interesting and helpful
discussions.
Appendix A. Dimensional split of Clifford algebras

In this appendix we give the various forms of the fermionic part of the action of
maximally supersymmetric YangMills theory. In 1 + 9 dimensions we use GM -matrices6
satisfying the Clifford algebra {GM , GN } = 2MN 132 with MN = diag(, +9 ). The
1
charge conjugation matrix C10 is symmetric and acts as C10 GM C10
= (GM )T . In ten
dimensions the fermionic field content of N = 1 super-YangMills theory is a single 32!
component complex spinor subject to the Majorana condition := G0 = T C10 as
!
well as the Weyl condition G11 = , where G11 := G0 G9 is the chirality matrix.
With these definitions the fermionic part of the D = 4, N = 4 theory can be written
concisely as
i
1
D + G
i [i , ],
Lferm = G
2
2
(A.1)
where = 0, 1, 2, 3 and i = 4, . . . , 9. We now split the GM -matrices into I -matrices of

SO(9) and a 2 2 matrix factor according to
G0 = i 2 116 ,
G =
I
(A.2)
with I = 1, . . . , 9.
(A.3)
The charge conjugation matrix decomposes as C10 = 1 C9 and leads to a symmetric

charge conjugation matrix in nine dimensions which satisfies C9 I C91 = ( I )T . Due to
the MajoranaWeyl condition the spinor can be written as

L
with L = LT C9
= 2
(A.4)
0
and the Lagrangian takes the form
Lferm = iL Dt L + iL a Da L + L i [i , L],
6 Throughout this appendix all indices are understood as frame indices.
(A.5)
376
where a = 1, 2, 3. We further split the I -matrices to account for SO(3) SO(6):

0
12 i
a 14
0
a
i
=
(A.6)
,
,
=
0
a 14
12 i
0
where a are the three Pauli matrices and the 4 4 matrices i satisfy
i j + j i = i j + j i = 2 ij 14 .
The charge conjugation matrix in this representation is given by

0
i 2 14
,
C9 =
i 2 14
0
allowing one to write the spinor as

A
L = 2 , = 1, 2, A = 1, . . . , 4,
i
A
(A.7)
(A.8)
(A.9)
where A are now four 2-component Weyl spinors. In this notation the Lagrangian reads

Lferm =2iA Dt A 2iA a Da A + A i 2 (i )AB i , B
T

A i 2 i AB i , B .
(A.10)
If one finally adds the unit matrix to the set of Pauli matrices by the definition = (1, a ),
this is exactly the form (4) given in the main text.
When we reduce the field theory to the zero mode excitation in Section 2.2 we end
up with the plane-wave matrix model where the fermions were written in SU(2) SU(4)
notation. This version of the matrix model can easily be obtained from the original form (2)
when the above realization of -matrices (A.6) is used and the matrix model fermions are
written in terms of Weyl spinors

A

.
=
(A.11)

i 2 A
Note that the charge conjugation matrix C9 is no longer unity, which has been used in (2).
We particularly note that

0
i12 14
123
(A.12)
=
,
0
i12 14
which simplifies the fermionic mass term to
m
m
i 123 = A A .
4
2
(A.13)
Appendix B. Killing spinors and Killing vectors

The Killing spinor equation is given by
a S
=
a S
.
2R
(B.1)
377
The hatted index = 1, 2 represents the degeneracy of the solution. When the different
signs are also taken into account the four solutions give the lowest spinor spherical
harmonics as (2, 1) (1, 2) of SU(2)L SU(2)R . In the following we will often suppress
the index, when the given equations hold for both signs. The solutions of (B.1) can be
orthonormalized to

S
S =
(B.2)
and form a complete basis

S S
= .
(B.3)
Moreover, they satisfy

T

S i 2 S = k i 2
with k C, |k| = 1,
(B.4)
and we are free to define them such that k = 1.
S
The Killing vectors can be obtained as bilinears S
of (commuting) Killing
a
spinors. One can show that these bilinears, considered as 2 2 matrices in the indices
and are hermitian and traceless. Therefore they may be expanded into Pauli matrices:
a S = (a ) Vaa
,
S
(B.5)
where a = 1, 2, 3 is a flat index. The Killing spinor equation (B.1) for S implies the
Killing vector equation
+ b Vaa
=0
a Vba
(B.6)
give the six Killing vectors (3, 1) (1, 3)

for the coefficients of this expansion. Hence V a
which are also the lowest vector spherical harmonics of S 3 . Their orthonormality
V aa
Vab = a b
(B.7)
can be shown from the orthonormality of the Killing spinors by making use of the Fierz
identity. As an immediate consequence of (B.6) we have the divergencelessness of the
Killing vectors
a Vaa = 0.
(B.8)
It can be also shown that
a Vba
=
1 a b c b
Va Vbc
2R
(B.9)
as well as
1
a Vba
(B.10)
= abc V a a
.
R
Let us also give an explicit realization of the Killing spinors and vectors. However,
we emphasize that the derivation of the matrix model can be entirely performed without
378
making use of them. With the obvious choice of vielbein

e1 = d,
e2 = sin d,
e3 = sin sin d,
(B.11)
it is a simple matter to find the solutions
S = ei 2 ei 2 ei 2 S0
(B.12)
and
Va1 dxa = cos e1 cos sin e2 sin sin e3 ,
Va2 dxa = sin cos e1 + (cos cos cos sin sin )e2
(cos sin sin cos cos )e3 ,
Va3 dxa
= sin sin e1 + (cos cos sin sin cos )e2

+ (cos cos sin cos sin )e3 .
(B.13)
Appendix C. Reduction of equation of motion

In this appendix we give some details of the derivation of Eqs. (20) for the zero modes
from the 4-dimensional equations of motion (19).
The first Eq. (19a) for = 0 can be written as

Dt a Aa i Aa , Dt Aa i[i , Dt i ] + 2 , = 0.
(C.1)
When the ansatz (18) is inserted one immediately finds (20a) using the orthogonality of
Killing spinors (B.2) and Killing vectors (B.7) as well as the divergencelessness of the
Killing vectors (B.8).
For the spatial components of the same Eq. (19a) one finds

Dt2 Aa 2 Aa + Ra b Ab + Da b Ab + i Ab , 2b Aa a Ab

Ab , [Aa , Ab ] i[i , a i ] [i , [Aa , i ]] 2 , a = 0,
(C.2)
where Rab = R22 gab is the Ricci tensor of S 3 . After inserting (18) we project onto the
Killing vectors. Then we use the identities
1
1
1
abc V aa
V bb V cc
= a bc det(V ) = a bc ,
R
R
R

S
a S V aa
= (b ) Vab V aa
= a
,
2
2 Vaa = 2 Vaa ,
R
and find (20b).
The equation of motion for the scalar field (19b) becomes
V bb a Vbc =
V aa

1
i i i , a Aa + 2i Aa , a i Aa , [i , Aa ]
2
R

[j , [i , j ]] + , i 2 i T , i 2 i = 0.
Dt2 i 2 i +
(C.3)
Here one has to make use of property (B.4), then (20c) immediately follows.
The fermion equation of motion (19c) is split up into

iDt i a a + a [Aa , ] i 2 i i , = 0
379
(C.4)
and projected onto the Killing spinors. Using similar identities as above one obtains (20d).
Appendix D. A useful identity for gauge invariant states

The operator : tr T a [ai , ai ]: annihilates any n-trace state

tr aj1,1 aj1,k tr aj2,1 aj2,k tr ajn,1 ajn,k |0.
1
(D.1)
This follows from the facts that it commutes with a trace of creation operators and
annihilates the vacuum. It commutes since the sum of all Wick contractions vanishes

: tr T a ai , ai : tr aj1 ajk

= : tr T a , ai ai : tr aj1 ajk

= tr T a , aj1 aj2 ajk + tr aj1 T a , aj2 ajk + tr aj1 aj2 T a , ajk

= tr T a aj1 aj2 ajk tr aj1 aj2 ajk T a = 0.
(D.2)
Appendix E. Effective 2-loop vertex
Summing all 2-loop contributions, we find
51N 2
: tr aiai :
4M 5
2
5 : tr ai aj tr ai aj : + : tr ai ai tr aj aj: + : tr ai aj tr ai aj :
M

+ : tr ai ai tr aj aj : + 2: tr ai aj tr ai aj :

N
67
67
5 30: tr ai aj ai aj : + : tr ai ai aj aj : + : tr ai ajaj ai :
M
8
8
61

+ : tr ai aj aj ai : + 17: tr ai ai aj aj : 15: tr ai aj ai aj :
4

1
5 : tr aj , ai , aj ak , ai ak : + : tr aj , [ai , aj ] ak , ai , ak :
M

+ : tr aj , ai , aj ak , ai , ak : + : tr aj , [ai , aj ] ak , ai, ak :

+ : tr aj , ai , aj ak , ai , ak : + : tr aj , ai aj ak , ai , ak :
(2)
Veff
380

+ : tr aj , ai aj ak , ai , ak : + : tr aj , [ai , aj ] ak , ai , ak :

1
+ : tr aj , ai , aj ak ai , ak : : tr[aj , [ai , aj ]] ak , ai , ak : .
2
(E.1)
This can be simplified using Jacobi identities and a reasoning similar to that at one loop,
e.g., we have

: tr aj , ai , aj ak , ai , ak :

= : tr aj , [ai , aj ] ak , ai ak : + N: tr[ai , aj ] ai , aj :

+ : tr ai , an , ai , an T a :: tr T a aj , aj :.
(E.2)
The last line may be dropped by arguments given in Appendix D.
Then we find the somewhat more compact final result
(2)

2
: tr ai ai tr aj aj : + 2: tr ai aj tr ai aj :
M5

N
5 11: tr ai , aj [ai , aj ]: + 17: tr ai ai aj aj : 15: tr ai aj ai aj :
M

4
5 : tr aj , [ai , aj ] ak , ai , ak :
M

1
5 : tr aj , ai , aj ak , ai , ak : + : tr aj , ai , aj ak , ai , ak :
M

+ : tr aj , ai , aj ak , ai , ak : + : tr aj , [ai , aj ] ak , ai , ak :

+ : tr aj , ai , aj ak , ai , ak :

1
: tr[aj , [ai , aj ]] ak , ai , ak : .
(E.3)
2
Veff
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] M. Claudson, M.B. Halpern, Supersymmetric ground state wave functions, Nucl. Phys. B 250 (1985) 689;
R. Flume, On quantum mechanics with extended supersymmetry and nonabelian gauge constraints, Ann.
Phys. 164 (1985) 189;
M. Baake, M. Reinicke, V. Rittenberg, Fierz identities for real Clifford algebras and the number of
supercharges, J. Math. Phys. 26 (1985) 1070.
[3] B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys. B 305 (1988)
545.
[4] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp-waves from N = 4 super-Yang
Mills, JHEP 0204 (2002) 013, hep-th/0202021.
[5] J. Kowalski-Glikman, Vacuum states in supersymmetric KaluzaKlein theory, Phys. Lett. B 134 (1984) 194.
[6] N. Kim, J. Plefka, On the spectrum of pp-wave matrix theory, Nucl. Phys. B 643 (2002) 31, hep-th/0207034.
[7] B. de Wit, M. Lscher, H. Nicolai, The supermembrane is unstable, Nucl. Phys. B 320 (1989) 135.
381
[8] K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Matrix perturbation theory for M-theory on a PPwave, JHEP 0205 (2002) 056, hep-th/0205185.
[9] K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Protected multiplets of M-theory on a plane wave,
JHEP 0209 (2002) 021, hep-th/0207050.
[10] N. Kim, J.H. Park, Superalgebra for M-theory on a pp-wave, Phys. Rev. D 66 (2002) 106007, hepth/0207061.
[11] D.S. Bak, Supersymmetric branes in PP-wave background, Phys. Rev. D 67 (2003) 045017, hep-th/0204033;
G. Bonelli, Matrix strings in pp-wave backgrounds from deformed super-YangMills theory, JHEP 0208
(2002) 022, hep-th/0205213;
K. Sugiyama, K. Yoshida, Supermembrane on the pp-wave background, Nucl. Phys. B 644 (2002) 113,
hep-th/0206070;
K. Sugiyama, K. Yoshida, BPS conditions of supermembrane on the pp-wave, Phys. Lett. B 546 (2002) 143,
hep-th/0206132;
K. Sugiyama, K. Yoshida, Giant graviton and quantum stability in matrix model on PP-wave background,
Phys. Rev. D 66 (2002) 085022, hep-th/0207190;
N. Nakayama, K. Sugiyama, K. Yoshida, Ground state of supermembrane on pp-wave, hep-th/0209081;
S.J. Hyun, J.H. Park, 5D action for longitudinal five branes on a pp-wave, JHEP 0211 (2002) 001, hepth/0209219;
S.J. Hyun, H.J. Shin, Branes from matrix theory in pp-wave background, Phys. Lett. B 543 (2002) 115,
hep-th/0206090;
S.J. Hyun, H.J. Shin, Solvable N = (4, 4) type IIa string theory in plane-wave background and D-branes,
Nucl. Phys. B 654 (2003) 114, hep-th/0210158;
N. Kim, K.Y. Lee, P. Yi, Deformed matrix theories with N = 8 and fivebranes in the pp wave background,
JHEP 0211 (2002) 009, hep-th/0207264;
K.Y. Lee, M-theory on less supersymmetric pp-waves, Phys. Lett. B 549 (2002) 213, hep-th/0209009;
N. Iizuka, Supergravity, supermembrane and M(atrix) model on pp-waves, hep-th/0211138;
J. Maldacena, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Transverse fivebranes in matrix theory, JHEP 0301
(2003) 038, hep-th/0211139;
M. Li, PP-wave black holes and the matrix model, JHEP 0305 (2003) 031, hep-th/0212345;
S. Hyun, J.H. Park, S.H. Yi, 3D N = 2 massive super-YangMills and membranes/D2-branes in a curve d
background, JHEP 0303 (2003) 004, hep-th/0301090;
J.T. Yee, P. Yi, Instantons of M(atrix) theory in pp-wave background, JHEP 0302 (2003) 040, hepth/0301120.
[12] J.H. Park, Supersymmetric objects in the M-theory on a pp-wave, JHEP 0210 (2002) 032, hep-th/0208161.
[13] D. Berenstein, H. Nastase, On lightcone string field theory from super-YangMills and holography, hepth/0205048.
[14] H. Nicolai, E. Sezgin, Y. Tanii, Conformally invariant supersymmetric field theories on S P S 1 and superp-branes, Nucl. Phys. B 305 (1988) 483.
[15] K. Okuyama, N = 4 SYM on R S(3) and pp-wave, JHEP 0211 (2002) 043, hep-th/0207067.
[16] M. Blau, Killing spinors and SYM on curved spaces, JHEP 0011 (2000) 023, hep-th/0005098.
[17] S. Deger, A. Kaya, E. Sezgin, P. Sundell, Spectrum of D = 6, N = 4b supergravity on AdS(3) S(3), Nucl.
Phys. B 536 (1998) 110, hep-th/9804166.
[18] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, PP-wave string
interactions from perturbative YangMills theory, JHEP 0207 (2002) 017, hep-th/0205089;
N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, Operator mixing and the BMN correspondence,
JHEP 0210 (2002) 068, hep-th/0209002.
[19] C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, A new double-scaling limit of N = 4 super-Yang
Mills theory and PP-wave strings, Nucl. Phys. B 643 (2002) 3, hep-th/0205033;
N. Beisert, C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, BMN correlators and operator mixing
in N = 4 super-YangMills theory, Nucl. Phys. B 650 (2003) 125, hep-th/0208178;
N. Beisert, C. Kristjansen, J. Plefka, M. Staudacher, BMN gauge theory as a quantum mechanical system,
Phys. Lett. B 558 (2003) 229, hep-th/0212269.
[20] J.A.M. Vermaseren, New features of FORM, math-ph/0010025.
382
[21] N. Beisert, C. Kristjansen, M. Staudacher, The dilatation operator of N = 4 super-YangMills theory, hepth/0303060.
[22] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super-YangMills, JHEP 0303 (2003) 013, hepth/0212208.
[23] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Anomalous dimensions in N = 4 SYM theory at order g 4 ,
Nucl. Phys. B 584 (2000) 216, hep-th/0003203;
G. Arutyunov, B. Eden, A.C. Petkou, E. Sokatchev, Exceptional non-renormalization properties and OPE
analysis of chiral four-point functions in N = 4 SYM(4), Nucl. Phys. B 620 (2002) 380, hep-th/0103230.
[24] N. Kim, J.T. Yee, Supersymmetry and branes in M-theory plane-waves, Phys. Rev. D 67 (2003) 046004,
hep-th/0211029.
[25] J. Maldacena, L. Maoz, Strings on pp-waves and massive two-dimensional field theories, JHEP 0212 (2002)
046, hep-th/0207284.
[26] N. Kim, Comments on IIB pp-waves with RamondRamond fluxes and massive two-dimensional nonlinear
sigma models, Phys. Rev. D 67 (2003) 046005, hep-th/0212017.

Correlated hierarchy, Dirac masses and large mixing

angles
Aseshkrishna Datta, Fu-Sin Ling, Pierre Ramond
Institute for Fundamental Theory, Department of Physics, University of Florida, Gainesville, FL 32611, USA
Received 4 June 2003; accepted 20 August 2003
Abstract
We introduce a new parametrization of the MNS lepton mixing matrix which separates the
hierarchical grand unified relations among quarks and leptons. We argue that one large angle stems
from the charged leptons, the other from the seesaw structure of the neutral lepton mass matrix. We
show how two large mixing angles can arise naturally provided there are special requirements on
the Dirac (Iw = 1/2) and Majorana (Iw = 0) masses. One possibility is a correlated hierarchy
between them, the other is that the Iw = 0 Majorana mass has a specific texture; it is Dirac-like for
two of the three families.
PACS: 14.60.Pq; 12.10.Dm
1. Introduction
The most compelling scheme for physics beyond the Standard Model revolves around
the idea of grand unification between quarks and leptons. This pattern is evident in terms of
the fermion quantum numbers, in view of the unification of the weak and strong coupling
constants at high energy, and more recently with the discovery of light neutrino masses.
The well-documented hierarchy of quarks and charged lepton masses seems to apply as
well in the neutrino sector, although not decisively since their masses are inferred solely
from interference effects, and there could be hyperfine splittings in their spectrum. Yet,
E-mail addresses: datta@phys.ufl.edu (A. Datta), fsling@phys.ufl.edu (F.-S. Ling), ramond@phys.ufl.edu

(P. Ramond).
doi:10.1016/j.nuclphysb.2003.08.026
384
A. Datta et al. / Nuclear Physics B 671 (2003) 383400
recent experiments ([16]) on neutrino mixings show qualitatively different behavior for
quarks and leptons. Quark mixings seems to be approximately the same for charge 2/3 and
charge 1/3 quarks, resulting in a CKM matrix that is approximately the unit matrix, save
for Cabibbo angle effects. On the other hand, two of the three mixing angles in the MNS
lepton mixing matrix are large, and only one is small. This implies that the mechanisms
that determine the lepton mass eigenstates are different for charged and neutral leptons.
There are differences between the fermion masses even at the level of the Standard
Model. The quark and charged lepton masses are Dirac-like and break electroweak
symmetry along Iw = 1/2, but the neutrino masses can be either Dirac-like with the same
Iw = 1/2 structure or Majorana with Iw = 1; in the latter case qualitative differences
may be expected.
As grand unification relates the Dirac matrices of quarks and leptons, one may expect
some if not all of the quark hierarchy to be mirrored in the lepton sector [7]. However,
SO(10) unification brings about the seesaw mechanism for generating neutrino masses
through a new fermion with a large Iw = 0 Majorana mass [8].
In previous works [9,10], based on SO(10) and on the FroggattNielsen (FN)
approach [11] to masses and mixings, we anticipated, before the Super-Kamiokande data,
one large mixing angle between the second and third family neutrinos in the MNS matrix,
and two small mixing angles from the first to the second and third family neutrinos. The
recent data on solar neutrinos proved otherwise: the MNS matrix contains two large mixing
angles, and only one small one.
The purpose of this note is to investigate in simple terms the possible origin of this
unexpected second large angle, without having to ditch cherished notions from grand
unification. In particular, we would like to analyze how large mixing angles are related
to a hierarchical structure in the mass matrix.
We expect a generic (i.e., without any hierarchical structure) 3 3 matrix to be
diagonalized by rotations with three angles which are not particularly small. One can also
imagine quite easily a matrix that is diagonalized by one large mixing and two small ones,
by assuming inequalities between matrix elements of one row and one column with the
remaining ones. Similarly, if all the off-diagonal matrix elements are small enough, the
mixing matrix will nearly the unit matrix and all three mixing angles come out small.
However, the CHOOZ constraint [4] indicates that only one mixing angle in the MNS
matrix is small. This is not possible to arrange in the hierarchical structure alone, and finetuning between various matrix elements is needed.
The MNS matrix is the overlap of two unitary matrices, one from the charged lepton
sector, the other from the neutral sector. The search for a generic mechanism suggests, in
view of the previous remarks, that one large mixing angle comes from the diagonalization
of the charged lepton masses, the other from that of the neutral lepton masses. We argue
below that the large angle found in atmospheric neutrino oscillations comes from the
Iw = 1/2 charged lepton matrix, with a modicum of assumptions. The large angle
recently determined in e oscillations then comes from the special form of the seesaw
mechanism, either through a correlated hierarchy of the Dirac and Iw = 0 Majorana
masses, or through a Dirac-like texture [12] for two of the three the right-handed neutrino
mass matrix. We conclude with some remarks on how these ideas can be incorporated into
a theory of the FroggattNielsen type.
385
2. Neutrino masses just beyond the Standard Model

We begin by reminding the reader how masses and mixings arise in the Standard Model,
stressing their properties in terms of electroweak quantum numbers. The usual Iw = 1/2
electroweak breaking generates quarks and charged leptons masses and mixing angles
through Yukawa couplings. In the quark sectors, it gives rise to Dirac masses through

0
mu 0
M(2/3) = U2/3 0 mc 0 V2/3

(1)
,
0
0 mt

0
md 0
(1/3)
M
(2)
= U1/3 0 ms 0 V1/3
,
0
0 mb
where U and V are unitary matrices which diagonalize them. Not all these matrices are
directly observable: of these four, only the CKM quark mixing matrix
UCKM = U2/3
U1/3 ,
(3)
is observable, and the right-handed matrices V2/3 , V1/3 are not physical.
Similarly in the charged lepton sector,

0
me 0
(1)
M
= U1 0 m 0 V1
.
0
0 m
(4)
In the original version of the Standard Model, neither unitary matrix is observable, and all
three neutrinos are massless. All mass matrices in the quarks and charged lepton sectors
arise from the same Iw = 1/2 electroweak breaking mechanism.
To go beyond the Standard Model, we add one right-handed neutrino for each family,
and the same electroweak breaking generates a Dirac mass for the neutrinos

0
m1 0
(0)
MDirac = U0 D0 V0 = U0 0 m2 0 V0 ,
(5)
0
0 m3
where mi are the neutrino Dirac masses, not to be confused with their physical masses
(unless there is no seesaw). This matrix respects total lepton number Le + L + L , but
violates the relative lepton numbers Le L and L L .
To account for small neutrino masses, it is customary to assign large Majorana masses
to these three right-handed neutrinos. Let M(0)
Majorana be the Majorana mass matrix with
entries of order M, assumed to be much larger than the electroweak breaking scale. This
matrix can be markedly different from the others as it arises not from electroweak breaking,
but from some unknown Iw = 0 sector. After seesaw (M m), the neutral mass matrix
is
(0)
(0)
MSeesaw = MDirac
1
M(0)
Majorana
(0)T
MDirac ,
(6)
386
where T stands for the transpose. Inserting the diagonalization (5), we find
1
(0)
MSeesaw = U0 D0 V0
M(0)
Majorana
V0 D0 U0T = U0 CU T0 ,
where C is the central matrix

C D0 V0
1
(0)
MMajorana
V0 D0 ,
(7)
which is diagonalized by means of a unitary matrix F

C = F D F T ,
(8)
where D is the diagonal matrix

m 1
0
0
D =
0 ,
0
m 2
0
0
m 3
(9)
with the physical neutrino masses as entries. This enables us to write the observable MNS
lepton mixing matrix in the suggestive form
UMNS = U1
U0 F ,
(10)
where F diagonalizes the central matrix of the seesaw. It is similar in form to the CKM
quark mixing matrix in that it contains the overlap between two unitary matrices that
diagonalize the Iw = 1/2 mass matrices, but it also contains a totally new matrix which
comes from the right-handed Iw = 0 neutrino masses.
This parametrization, which separates the effect of the Iw = 1/2 Dirac masses from
that of the Iw = 0 Majorana masses, makes it particularly convenient to discuss the
intuition coming from grand unified theories. With the simplest Higgs structures, these
theories typically relate the Iw = 1/2 mass matrices for quarks and leptons
SU(5) :
M(1/3) M(1)T ,
SO(10) : M
(2/3)
M(0)
Dirac .
(11)
(12)
These imply similar hierarchies between the CKM and MNS matrix, should F be equal
to one, which happens if the neutral mass matrix is purely Dirac-like (as in theories with
bulk right-handed neutrinos). On the other hand, if the seesaw is operative, the effects of
the unitary matrix F must be included.
The Iw = 1/2 mass matrices in the quark and charged lepton sectors show hierarchy,
a feature consistent with grand unified theories. On the other hand, over the last five years,
several experiments have shown that the MNS matrix shares this hierarchy only partially,
as it contains one small and two large mixing angles. We would like to argue that in the
context of GUTs, this may be an indication in favor of the seesaw, with one or both large
angles in the MNS matrix coming from F .
However, the central matrix C already contains some hierarchical information coming
from D0 , the eigenvalues of the hierarchical neutral Dirac mass. Indeed, using grand
387
unification as a rough guide, we expect a hierarchical structure

m1 m2 m3 ,
(13)
but if F is to contain large angles, some of the matrix elements of the central matrix
must be of similar orders of magnitude. This can happen through numerical accidents, or
(0)
generically if MMajorana contains a correlated hierarchy of its own, designed to offset that
of the Dirac masses. The purpose of this paper is to find how F can contain large angles
despite the expected hierarchy of the Dirac masses.
3. The data
Let us see how these theoretical expectations compare with the data. The combined
experimental results favor a scheme with three active neutrinos. The mixing matrix UMNS
has two large angles ( and for atmospheric [1] and solar [2,3] neutrinos, respectively)
and one small one (from the CHOOZ [4] constraint). This enables us to write

UMNS
cos
cos sin
sin sin
sin
cos cos
sin cos

sin
cos

,
(14)
where is a complex number, with the limits [13]

sin2 2 0.85 (99% C.L.),
0.30 tan2 0.65 (95% C.L.),
||2 0.005 (99.73% C.L.).
(15)
The neutrino mass splittings are

m2 = m21 m22
7 105 eV2 ,

m2 = m22 m23
3 103 eV2 ,
coming from the solar and atmospheric neutrino oscillations, respectively. Although
neutrino oscillation experiments do not fix the absolute mass scale of the neutrinos, we
now have very stringent bounds from cosmology. The large scale structure formation and
the latest measurements of the cosmic microwave background by WMAP [14] indicate that
mk 0.71 eV
(95% C.L.).
These are consistent with three possible neutrino mass patterns
(16)
388
1
2
3
2
1
3
Hierarchy
Inverted
Hyperfine
|m1 | |m2 | |m3 |
|m1 |
|m2 | |m3 |
|m1 |
|m2 |
|m3 |
We can use the data to reconstruct the neutrino matrix in all three cases, and look for
patterns [15].
In the hierarchy case, we neglect m1 and m2 , and the Yukawa matrix looks like

0
0
0
0
m3 sin2
(17)
m3 sin cos ,
0 m3 sin cos
m3 cos2
which suggest as a starting point a matrix with entries in the 23 block only

0 0 0
0 ,
0
(18)
where the crosses describe elements of order one. However, in order to obtain from this
pattern the second large angle, this matrix must contain two zero eigenvalues, that is the
sub-determinant must vanish. This type of matrix has been used as a starting point for the
FroggattNielsen approach, with corrections of the order of the Cabibbo angle in previous
works [10,16], and it can lead to a second large mixing angle only if the sub-determinant is
very small [1618]. If the sub-determinant is not small, non-degenerate perturbation theory
applies and the solar neutrino mixing angle comes out small [19].
In the inverted hierarchy case, all entries of the matrix seem to be of the same order of
magnitude, but there is a special case of great interest. If we take m1 m2 and maximal
angles, the matrix looks like

0
(19)
0 0 ,
0 0
which can be used as a starting point for a Cabibbo expansion [20]. In this special case, the
masses of the first two neutrinos follow a nearly Dirac pattern, with a global Le L L
symmetry [21].
Finally, in the hyperfine case where all three masses are nearly degenerate, there is
no discernible pattern in the Yukawa matrix.
389
We would like to see how to reproduce these patterns, starting from conservative
theoretical notions, namely grand unified symmetries.
4. A useful approximation
In the following we would like to argue that the large mixing angle found in atmospheric
neutrino oscillations stems from U1 , the unitary matrix in the charged lepton sector. To
that effect, consider the limit in which the Dirac masses have two zero eigenvalues in
each charged sector, so that only the third family is massive. If we extend the Wolfenstein
parametrization of the CKM matrix to quark and charged lepton masses, expressing the
masses of the lighter families as powers of the Cabibbo angle, this limit can be thought of
as that where the Cabibbo angle is zero. In this limit, we set the two quark mass matrices
as

0 0 0
0 0 0
(2/3)
M
= 0 0 0 = U2/3 0 0 0 V2/3
,
0 0 mt
0 0 mt

0 0 0
0 0 0
(1/3)
M
= 0 0 0 = U1/3 0 0 0 V1/3
.
0 a b
0 0 mb
It follows that
U2/3 = U1/3 = 1,
which shows that the CKM matrix
UCKM = U2/3
U1/3 = 1,
which is a close approximation to data. It also suggests that the quark family mixings in
the Iw = 1/2 sector are hypercharge independent.
Note that we have allowed for non-zero entries in the third row of the charge 1/3 mass
matrix. This still yields two zero eigenvalues and does not affect the left-handed rotations.
In fact we have encountered such a structure in our previous work [10,16,22] based on the
FroggattNielsen formalism, where the FN charges of d 2 and d 3 are naturally the same,
leading to the same degree of suppression.
This conclusion is not limited [23] to the FN formalism, and it also appears in a
completely different approach, a model where the first two families share the same weak
SU(2)1+2 but the third family has a different weak SU(2)3 . Upon diagonal breaking, these
merge into the physical weak SU(2)1+2+3 . This pattern of symmetries naturally arises
with three copies of the electroweak group [24], broken diagonally by a tri-chiral order
parameter. This order parameter merges the three SU(3)s into one but breaks the three
SU(2) into these two. With only a Higgs doublet with respect to SU(2)3 at tree level, this
singles out the third row of the charge 2/3 and 1/3 matrices with order one entries. In
this approximation, two whole families of quarks are massless, and there is no inter-family
mixing.
390
To find the lepton matrices, we rely on the simplest SU(5) pattern which gathers the lefthanded anti-down quarks with the charged leptons in its 5 representation. This suggests we
assign the same FN charges to the left-handed leptons and anti-down quarks, and leads us
to assume that the charged lepton and charge 1/3 mass matrices are transpose of one
another

0 0 0
0 0 0
(1)
= 0 0 a = U1 0 0 0 V1
,
M
0 0 m
0 0 b
leading to

U1 =
1
0
0
0
cos
sin
0
sin
cos

,
tan =
a
.
b
We claim this is the origin of the large mixing angle which describes the oscillations of
atmospheric neutrinos.
At the next level of grand unification, SO(10), the charge 2/3 and the neutral Dirac mass
matrices are supposed to be related, and if one shows hierarchy, so does the other. Hence
in this approximation, we set the neutral Dirac mass matrix

0 0 0
0 0 0
(0)
MDirac = 0 0 0 = U0 0 0 0 V0 .
(20)
0 0 m
0 0 m3
It follows that U0 is the unit matrix and the MNS matrix becomes

1
0
0
UMNS = U1 U0 F = 0 cos
sin F .
0 sin cos
(21)
The unitary matrices U1 and U0 stem from the same Iw = 1/2 electroweak breaking,
but F can be markedly different as it arises but from the unknown Iw = 0 sector.
Assume that in this limit the Majorana matrix has no zero eigenvalue. Since the neutral
Dirac mass matrix is of the form (20), the seesaw neutrino mass matrix is just

m2 0 0 0
(0)
MSeesaw =
0 0 0 ,
M 0 0 1
so that F = 1 and the MNS matrix is simply

1
0
0
UMNS = 0 cos
sin .
0 sin cos
(22)
If the Majorana matrix has zero eigenvalues, the seesaw may not apply, depending on the
relative alignment of the zeros in the Dirac and Majorana matrices, so one can have both
Dirac and Majorana matrices for the neutrinos.
With two massless neutrinos, the family mixing described by does not specify which
of the massless neutrinos the third one mixes into, and F cannot change the structure of
the MNS matrix. It is only by breaking the degeneracy that one can identify the flavor into
391
which the third neutrino mixes. With this caveat in mind, the MNS matrix remains the
same as above. We therefore think it natural to expect one large lepton mixing angle in the
limit with two massless families, even though there is no quark mixing (UCKM = 1).
To lift the degeneracy, degenerate perturbation theory must be used, so that the form of
the perturbation determines the mixing angle between the two hitherto massless families.
In the quark sector, that mixing is Cabibbo suppressed, and in the lepton sector it is of order
one. As our analysis shows, this difference is naturally accommodated if there is a seesaw
mechanism: the large mixing must be due to the F matrix. Given the form of U1 , there
are two possibilities for F

cos
sin 0
cos
sin 0
F = sin cos 0 ,
(23)
F=
0
0
1 ,
0
0
1
sin cos 0
obtained by permuting the second and third entries. The second case is the same as the first
provided that one reflects about /4. Either way, the mixing between the third family
and the others is naturally small since non-degenerate perturbation theory applies: the
smallness of the 13 element is the only remnant of the quark and charged lepton hierarchy
in the MNS matrix!
We conclude that the second large mixing angle between the first two family neutrinos
must arise when the Cabibbo perturbation is turned on. It is the purpose of the next section
to see how this can come about.
5. Correlated hierarchy
In this section, we examine in some detail how the seesaws central matrix can have
elements of the same order of magnitude, thereby cancelling the Dirac hierarchy in the
numerator with that in the Majorana denominator. At first sight, it implies relations between
two hitherto unrelated sectors of the Dirac (Iw = 1/2) and Majorana (Iw = 0) masses,
and speaks further for grand unification: it suggests that the mechanism which creates
hierarchies acts in some sense on the whole grand-unified structure. We present a detailed
analysis only for the (2 2) case, and simply state the results for the realistic (3 3) case.
5.1. The (2 2) case
Since the data can be fitted with only one large mixing angle in F , this case is quite
instructive on its own. To see how a correlated hierarchy can arise generically when the
neutral Dirac matrix is hierarchical, we expand D0 in the Cabibbo angle la Wolfenstein,

0
a
D0 = m
(24)
,
0
1
with a of order one and > 0, and set

1
1
c s
M1
V0 =
V0 (0)
s
c
0
MMajorana
0
1
M2

c
s
s
c

,
(25)
392
where c = cos , s = sin . The central matrix is then

2 2 cs

c2
s2
cs
M1 + M2 a
M1 M2 a
.
C=

s2

cs
cs
c2
M1 M2 a
M1 + M2
(26)
For F to contain generically a large angle, the diagonal entries in C must be at most of
the same order of magnitude as the off-diagonal element. For C given by Eq. (26), two
different cases lead to a large angle.
5.1.1. Case A
In the first situation, assume all the entries in C are of the same order of magnitude
C11 C22 C12 .
(27)
For this to happen, must be small, such that

s = b + ,
c = 1 ,
(28)
and the Majorana masses must show hierarchy,

M1
2 ,
M2
(29)
which must be less than or equal to 2 , in order to get C22 to be of the same order as the
other elements. If > , the matrix C reduces to

m2 a 2 ab
2
,
M1 ab b2
which produces unsuppressed mixing between the first two species. If the Majorana
hierarchy is extreme ( ), there is one zero eigenvalue to lowest order and a de facto
mass hierarchy, but with a large mixing angle. If the Majorana mass hierarchy is less
pronounced (
), the determinant no longer vanishes at lowest order, and the masses of
the two lightest neutrinos are of the same order
m 1 m 2
m2 2
.
M1
(30)
We see that it is possible to obtain large mixing as long as the right-handed masses are also
hierarchical [25]. Moreover the Majorana hierarchy must be twice as severe as the Dirac
hierarchy. The result is

a
cos
sin
F=
,
tan = ,
(31)
sin cos
b
to lowest order in . Because the (3 3) case can always be reduced to the (2 2) case
by setting to zero the mixing angles with the third family, we see that it is natural in this
approach to expect the MNS matrix to contain two large angles, and , while the Ue3
entry is naturally Cabibbo suppressed.
393
5.1.2. Case B
The second situation which generically gives rise to a large angle is when the diagonal
entries in C are small compared to the off-diagonal ones
C11 , C22 C12 .
(32)
This implies a correlation between the angle and the Majorana hierarchy
tan2 =

M1
1 + O
M2
(33)
with > 0 and

s .
(34)
When the inequality is strict, the matrix C reduces to

m2
0 a
,
M1 M2 a 0
which produces maximal mixing between two light neutrinos with the same absolute mass.
In particular, if the right-handed neutrinos are given a large Dirac mass, it corresponds to
M1 = M2 and
/4, therefore naturally ensuring a large mixing angle among the
light neutrinos.
When s , the diagonal elements of the central matrix cannot be neglected, and
the mixing angle is somewhat less than maximal.1
Unlike case A, the angle in case B can be larger than , as long as it is related to the
Majorana hierarchy through Eq. (33). Hence we again have one large mixing angle, but it
may be naturally close to maximal

cos
sin
F=
(35)
,
tan
,
sin cos
4
just by having, to lowest order in , the first two right-handed neutrinos as Dirac partners!
5.2. The (3 3) case
In the realistic (3 3) case, the analysis of the central matrix is more involved. As we
have stated in the introduction, it is not generic to expect a (3 3) matrix to be diagonalized
by one small and two large angles. This does not mean it is impossible, rather it implies
subtle relations among its matrix elements. It is much easier to expect three, one, or no
large mixing angles. Below, we present the results of systematic estimates of the orders of
magnitude of the central matrix elements and possible cancellations along the lines of the
previous section.
1 It is amusing to note that if one diagonal element gets filled as much as the off-diagonal element, the mixing
angle is determined by the celebrated golden mean which corresponds to a mixing angle of 31 close to the
experimental best-fit value, as pleasing to the eye as it may be to Nature?
394
To set our notation, the hierarchical Dirac matrix eigenvalues are

1 +2
0 0
1
0 ,
D0 m
0
0
0 1
(36)
ignoring prefactors. We write the unitary matrix in the central matrix as the product of
three rotations
V0 = R12 R13 R23 ,
neglecting possible phases, and where Rij denotes the rotation in the ij plane with small
(0)
angle ij , so that sin ij sij
ij and cos ij
1; and take MMajorana to be diagonal with
masses Mi . To facilitate the analysis, we further assume that
s13 s12 s23
(37)
are of the same order of magnitude in , as for the CKM matrix. This relation is natural
when the right-handed neutrino Majorana hierarchy is not inverted, i.e.,
M1 M2 M3 .
(38)
5.2.1. Case A
This first case corresponds to a correlated hierarchy between the Dirac and the Majorana
sector, which will lead to a neutrino mass pattern with a normal hierarchy and one large
mixing angle in F . The analysis of the orders of magnitudes of the resulting matrix
elements is tedious although remarkably straightforward. To maintain the goodwill of the
reader, we only present the results:
If the large angle is in the 12 block of F , the structure of the central matrix obeys
C11 C12 C22 C33 ,
(39)
2
2
, C23
C13
(40)
C11 C33 .
We obtain three solutions which all satisfy

M1
22
M2
and s12 2 .
(41)
First solution: the central matrix is dominated by the lightest right-handed neutrino
mass M1
M1
2 22
s23
M3
and s23 1 .
(42)
This leads to a large mixing angle 12 in the 12 block and small mixing angles with the
third family
13 23
1
1.
s23
(43)
395
The physical neutrino masses are hierarchical, with

m 1 m 2
21
m2 21 +22
m 3
.
2
M1
s23
(44)
Second solution: we have the possibility of an inverted hierarchy between M2 and M3

M1 22
2 .
M3
s23
(45)
13 23 s23 1 .
(46)
22
This leads to
For the physical neutrino masses, m1 and m2 are still given by Eq. (44) but
m 3
m2
.
M3
(47)
Third solution: the physical neutrino masses are the same as for the second solution,
but the mixing angles are different

M1
2 2(2 1 )
max 21 +22 , s23
22 .

M3
(48)
This gives
23 13 1 s23 1 +22
M3
.
M1
(49)
If the large angle is in the 13 block of F , the structure of the central matrix obeys
C11 C13 C33 C22 ,
(50)
2
2
, C23
C12
(51)
C11 C22 .
We again get three different possibilities. To obtain the desired MNS matrix, the heaviest
neutrino must be labeled 3 , while the two lighter neutrinos 1 and 2 mix with the large
angle 12 which gives rise to the observed solar neutrino deficit. Following this labeling
convention, we get
First solution:
M2
2
s12
,
M1
M2
2
s23
,
M3
s23 s12 1 +2 .
(52)
This gives
13 23 s12 2
(53)
and
m 1 m 2
m2 2 21 +22
s
,
M2 12
m 3
m2 21
.
M2
(54)
396
Second solution:
M1
2
s12
,
M2
21 +22
s12 s23 1 +2 .

2
s12
M1

min 1, 2 ,
M3
s23
(55)
This gives
13 23 s23 1
(56)
and
m 1 m 2
m2 21 +22
,
M1
Third solution:
M1
21 +22 ,
M3
m 3
m2 2
s .
M1 12
(57)
M2
2
s23
M3
(58)
s12 2 .
(59)
and
s12 s23 1 +2 ,
The mixing angles are now given by

13 23
2
1.
s12
(60)
F has one large rotation in the 23 block. Since this possibility cannot lead to the
right MNS matrix, we will skip it.
We notice that it is possible to obtain one, three, or no large angles in F . To obtain only
two large angles, there must be further fine-tuning conditions among the prefactors, or
partial cancellations in some matrix elements of the central matrix C. Without such partial
cancellations, a generic solution cannot be obtained.
5.2.2. Case B
We now turn to the possibility of partial cancellations in some matrix elements of C,
leading to a pseudo-Dirac spectrum for the neutrinos which mix strongly (1 and 2 ) and
an inverted hierarchy
m1
m2 |m3 |.
(61)
We notice that such spectrum cannot be obtained without partial cancellation.

For C to have a pseudo-Dirac mass in the 12 block, the only condition is that the
element C12 dominates over all the other elements. There is only one solution, which
extends the (2 2) result. We have the partial cancellation
tan2 12 =

M1
1 + O
M2
(62)
397
together with
M1
s12 21 +2 ,
M3
2 s12 2 ,
s13 1 +2 .
s23 1 ,
(63)
This gives
s23
s13
1,
23 + 1
1
1 2
and a pseudo-Dirac pair of absolute mass
13
(64)
m2
m1
m2
21 +2 .
M1 M2
(65)
The angle 12 can be large while keeping s23 and s13 small as long as the partial
cancellation condition Eq. (62) remains satisfied.
It is not possible to get a simple partial cancellation condition that can give rise to a
pseudo-Dirac mass in the 13 block, if we keep the relation (37). This is expected since
such pattern of the Majorana masses has an inverted hierarchy more consistent with the
relations
s23 s12 s13
or s12 s23 s13 .
(66)
With either of these relations, we can indeed obtain a solution with the partial cancellation
condition
tan2 13 =

M1
1 + O .
M3
(67)
For example, with the second relation in Eq. (66), we get

M1 s13 1 +2
2
, s13 2 1 ,
M2 s23
1 +2 s13 1 +2 ,
s23 s13 2 .
(68)
This gives

s23
M1
13 max 1,
1,
1
M2

s23 s13 s23 M1
23 max
1
,
2 s13 2 M2
(69)
and a pseudo-Dirac pair of absolute mass

m2
m1
m2
1 +2 .
M1 M3
(70)
Finally, let us notice that the structure Eq. (19) which naturally leads to a bi-maximal
mixing and an inverted hierarchy of the light neutrinos, cannot be reproduced in the central
matrix if the relation (37) holds.
398
6. Theoretical outlook
Having identified the conditions for two large mixing angles in the MNS matrix, it
remains to be seen if they can be easily realized in credible theoretical schemes.
Froggatt and Nielsen (FN) proposed long ago to parametrize the hierarchy in terms of
effective operators coming from unknown interactions at higher energies, and that their
degree of suppression is determined by extra charges; the higher the charge, the more
suppressed the operator.
This approach which explains hierarchies by adding U (1) symmetries to the Standard
Model has received much attention and shown some success especially when paired with
anomalous GreenSchwarz U (1) symmetries [10,16,22,2628].
This is particularly suited to the Wolfenstein expansion as powers of the Cabibbo angle,
as it relates the power of to the FN charges of the basic quark and lepton fields. In the
simplest form of the theory, these exponents can almost be read-off from the quark mass
and mixing matrices. As an example, after breaking the FN symmetry, the neutral lepton
Dirac mass operator looks like

LTi M(0)
ij Nj ,
(71)
with the hierarchy determined by the charges of the lepton doublets Li and the right-handed
i . The matrix can be expunged from hierarchy by means of diagonal
neutral leptons N
matrices
(0) ,
M(0) = L M
N
(72)
where

= Diag Q1 Q3 , Q2 Q3 , 1 ,
(73)
are the diagonal matrices and Qi are the FN charges of the relevant fields, and the hatted
matrix contains only elements of order one.
If the hierarchy is explained solely in terms of one FN field, one obtains a simple
prediction for the seesaw masses, namely that the hierarchy matrix N coming from the
right-handed neutrinos cancels out and that we are left with only the hierarchy coming
from the lepton doublets. This conclusion does not hold in more complicated FN schemes
with fields of opposite FN charges, but we are assuming the simplest possibility here.
In this case, the F matrix contains no elements suppressed by powers of the Cabibbo
angle, and all its elements are of order one, the MNS matrix contains three large angles, and
not two as indicated by the data. Hence the simplest FN scheme has trouble explaining any
correlated hierarchy, and typically does not produce a MNS matrix with two large mixing
angles. To produce a correlated hierarchy in the FN approach, there must be several FN
fields of charges of different signs.
This seems to favor a textured Majorana matrix, with carefully chosen zeros, either
through a particular texture or through supersymmetric zeroes. This leads to a simple
proposal: use the same FN schemes proposed in our previous models for the Iw = 1/2
masses, but require that the Iw = 0 mass be textured with a Dirac mass for the righthanded neutrinos of the first two families, and a Majorana mass for the third, to zeroth
order in the Cabibbo angle

0 M1 0
M1 0
0 + .
0
0 M3
399
(74)
Then as we have shown in Section 5, this generates a nearly maximal angle between the
first two neutrinos.
The required texture for the Iw = 0 Majorana mass must be correlated with that of
the Iw = 1/2 Dirac mass in such a way that the large angles coming from both sectors
do not rotate the same block; this would produce only one large angle. In particular, if
the charged lepton sector produces a large angle along 23, the large angle in the central
matrix must be either along 12 or 13, but not 23. As argued throughout the paper, the
presence of one small mixing angle in the MNS matrix can be thought as the last remnant
of a hierarchical structure in the lepton sector. Since hierarchical patterns are generic in the
context of grand unified theories and quarklepton unification, our analysis shows that it
is much more natural to attribute the two observed large mixing angles in the MNS matrix
to different sectors [29]. We conclude that the large atmospheric angle must stem from the
charged lepton sector, while the large solar angle must reflect some peculiar behavior of
the right-handed neutrino Majorana sector and the seesaw mechanism.
Acknowledgements
This work is supported by the United States Department of Energy under grant DEFG02-97ER41029.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
Super-Kamiokande Collaboration, Phys. Rev. Lett. 85 (2000) 3999.

Super-Kamiokande Collaboration, Phys. Lett. B 539 (2002) 179.
SNO Collaboration, Phys. Rev. Lett. 89 (2002) 011301.
M. Apollonio, et al., Phys. Lett. B 338 (1998) 383;
M. Apollonio, et al., Phys. Lett. B 420 (1998) 397.
KamLAND Collaboration, Phys. Rev. Lett. 90 (2003) 021802.
K2K Collaboration, Phys. Rev. Lett. 90 (2003) 041801.
G. Altarelli, F. Feruglio, I. Masina, JHEP 0011 (2000) 040;
W. Buchmller, Acta Phys. Pol. B 32 (2001) 37073718;
T. Blazek, S. Raby, K. Tobe, Phys. Rev. D 62 (2000) 055001.
M. Gell-Mann, P. Ramond, R. Slansky, in: Sanibel Talk, CALT-68-709, February 1979, retropreprinted as
hep-ph/9809459;
M. Gell-Mann, P. Ramond, R. Slansky, in: Supergravity, North-Holland, Amsterdam, 1979;
T. Yanagida, in: Proceedings of the Workshop on Unified Theory and Baryon Number of the Universe, KEK,
Japan, 1979.
J.A. Harvey, D.B. Reiss, P. Ramond, Nucl. Phys. B 199 (1982) 223.
P. Bintruy, N. Irges, S. Lavignac, P. Ramond, Phys. Lett. B 403 (1997) 38;
N. Irges, S. Lavignac, P. Ramond, Phys. Rev. D 58 (1998) 035003.
C. Froggatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 277.
400
[12] P. Ramond, R.G. Roberts, G.G. Ross, Nucl. Phys. B 406 (1993) 19.
[13] G.L. Fogli, et al., Phys. Rev. D 66 (2002) 093008;
G.L. Fogli, et al., Phys. Rev. D 67 (2003) 073002.
[14] WMAP Collaboration, astro-ph/0302207;
WMAP Collaboration, astro-ph/0302209.
[15] For an intriguing pattern, see P. Kaus, S. Meshkov, hep-ph/0211338.
[16] F.-S. Ling, P. Ramond, Phys. Lett. B 543 (2002) 29.
[17] J. Sato, T. Yanagida, Phys. Lett. B 493 (2000) 356.
[18] F. Vissani, JHEP 9811 (1998) 025;
F. Vissani, Phys. Lett. B 508 (2001) 79.
[19] F.-S. Ling, hep-ph/0304135.
[20] S.F. King, JHEP 0209 (2002) 011.
[21] K.S. Babu, R.N. Mohapatra, Phys. Lett. B 532 (2002) 77.
[22] F.-S. Ling, P. Ramond, Phys. Rev. D 67 (2003) 115010.
[23] J. Sato, T. Yanagida, Phys. Lett. B 430 (1998) 127;
C.H. Albright, K.S. Babu, S.M. Barr, Phys. Rev. Lett. 81 (1998) 1167;
C.H. Albright, S.M. Barr, Phys. Rev. D 58 (1998) 013002.
[24] M.J. Bowick, P. Ramond, Family cloning of the gauge group, Phys. Lett. B 131 (1983) 367.
[25] A similar analysis was done earlier by A.Yu. Smirnov, Phys. Rev. D 48 (1993) 3264.
[26] M. Green, J. Schwarz, Phys. Lett. B 149 (1984) 117.
[27] L. Ibaez, Phys. Lett. B 303 (1993) 55.
[28] L. Ibaez, G.G. Ross, Phys. Lett. B 332 (1994) 100.
[29] C.H. Albright, S.M. Barr, Phys. Lett. B 461 (1999) 218.

On the origin of the UV-IR mixing in

non-commutative matrix geometry
Sachindeo Vaidya a , Badis Ydri b
a Department of Physics, University of California, Davis, CA 95616, USA
b School of Theoretical Physics, Dublin Institute for Advanced Studies, Dublin, Ireland
Received 26 May 2003; received in revised form 28 July 2003; accepted 19 August 2003
Abstract
Scalar field theories with quartic interaction are quantized on fuzzy S 2 and fuzzy S 2 S 2 to
obtain the 2- and 4-point correlation functions at one-loop. Different continuum limits of these noncommutative matrix spheres are then taken to recover the quantum non-commutative field theories
on the non-commutative planes R2 and R4 , respectively. The canonical limit of large stereographic
projection leads to the usual theory on the non-commutative plane with the well-known singular
UV-IR mixing. A new planar limit of the fuzzy sphere is defined in which the non-commutativity
parameter , beside acting as a short distance cut-off, acts also as a conventional cut-off = 2 in the
momentum space. This non-commutative theory is characterized by absence of UV-IR mixing. The
new scaling is implemented through the use of an intermediate scale that demarcates the boundary
between commutative and non-commutative regimes of the scalar theory. We also comment on the
continuum limit of the 4-point function.
PACS: 11.10.Lm; 11.10.Gh; 11.90.+t; 11.25.-w; 11.15.Bt
1. Introduction and results

Non-commutative manifolds derive their interest not only from the fact that they
make their appearance in string theory (see, e.g., [1] for a review of non-commutative
geometry in string theory), but also because they can potentially lead to natural ultraviolet regularization of quantum field theories. The notion of non-commutativity suggests
E-mail addresses: vaidya@dirac.ucdavis.edu (S. Vaidya), ydri@synge.stp.dias.ie (B. Ydri).

doi:10.1016/j.nuclphysb.2003.08.023
402
S. Vaidya, B. Ydri / Nuclear Physics B 671 (2003) 401431
a graininess for spacetime, and hence can have interesting implications for models of
quantum gravity.
Theoretical research has usually focused on either flat non-commutative spaces like
R2n or non-commutative tori T 2n , or curved spaces that can be obtained as co-adjoint
orbits of Lie groups. In the latter category, attention has mostly focused on using compact
groups leading to non-commutative versions of CP n [24], which are described by finitedimensional matrices and one or more size moduli: for example, the fuzzy sphere is
described by N N matrices and its radius R. (We use descriptions like flat or curved
compact only in a loose sense here.)
Considerable attention has thus been devoted in trying to understand properties of
simple theories written on non-commutative manifolds. In this endeavor, attention has most
often been devoted to theories on non-commutative R2n and T 2n in the case of flat spaces,
and the curved space SF2 (the fuzzy sphere). Theories on the non-commutative flat spaces
generally possess infinite number of degrees of freedom in contrast to those on compact
spaces like SF2 . In either case, a key property of non-commutative theories that is different
from ordinary ones is the nature of the rule for multiplying two functions. For example,
the star-product on R2n (involving the non-commutativity parameter ) is used for noncommutative theories, while ordinary theories use the usual point-wise multiplication. On
the other hand, functions on curved compact non-commutative spaces are simply finitedimensional matrices, and are multiplied by the usual matrix multiplication. This makes
theories on curved non-commutative spaces easier to study numerically (although it must
also be mentioned here that the torus with rational non-commutativity can also be studied
using finite-dimensional matrices [6]).
Working on curved compact spaces also allows us to study the flattening limit, which is
when we take matrix size as well as the length moduli to infinity. For example, the fuzzy
sphere SF2 can be flattened to give us the non-commutative plane. In this limit, we expect
to reproduce the behavior of the theory on the flat manifold. Surprisingly, this limit can be
crafted in a variety of ways.
A simple way to understand this is as follows. All dimensionful quantities can be
expressed in terms of radius moduli, i.e., the length scale that defines the size of the
compact space. Continuum limit usually corresponds to taking the size of the matrices to
infinity, while flattening corresponds to taking large radii. However, there is a large family
of scales available to us in this flattening limit. In other words, there are many ways of
getting a relevant length dimension quantity on the non-compact space. We could scale
both R and N to infinity keeping R/N fixed, where is some number. This corresponds
to a length scale on the plane, and all quantities in the quantum field theory (QFT) on the
plane can be measured with respect to this scale. A priori, one would suspect that different
values of can lead to theories that behave dramatically differently.
As we will argue here, this variety in the choice of scaling gives us a refined probe
to understand the nature of non-commutativity more clearly. In particular, we will show
with two different scaling limits how this works. One corresponds to strongly noncommutative theories, possessing singular properties like UV-IR mixing that makes
it impossible to write down corresponding low-energy Wilsonian actions. The other
corresponds to weak non-commutativity in a sense that we will make precise. Briefly,
these weakly non-commutative theories are defined on a non-commutative plane, but do
403
not exhibit UV-IR mixing. In some sense, these theories mark the edge between noncommutativity and commutativity.
The standard method of investigating perturbative properties of a scalar QFT is
by introducing an ultra-violet cut-off (see for example [19]). Instead of working with
arbitrarily high energies, one works with the partition function of this cut-off theory,
and attempts to study quantities that depend only weakly on the UV cut-off. However,
applying this technique to non-commutative theories is problematic [5]: taking the limit of
small external momentum does not commute with taking the limit of infinite cut-off. This
problem is commonly known as UV-IR mixing.
QFTs on non-commutative curved spaces allow us to implement a finer version of
the above procedure. In addition to the natural UV cut-off (characterized by 1/N where
N is the matrix size), we can introduce an intermediate scale 1/j characterized by an
integer j < N . It is the interplay between j, N and R that we will exploit to
understand the edge between commutativity and non-commutativity.
In this article, we make concrete this set of ideas by applying them to SF2 and SF2 SF2 .
The former is characterized by (2l + 1) (2l + 1) matrices and radius R, the latter by
two copies of the same matrix algebra and two radii R1 and R2 . Flattening these spaces by
taking l and Ri to infinity (in a prescribed manner) gives us non-commutative R2 and R4 ,
respectively.
In particular, we will study two such scalings here. For example, for SF2 , we keep

= R/ l fixed in the first case, and keep = R/ l fixed in the second, as we take l and
R to infinity. The former gives us the usual theory on the non-commutative plane, which
at the one-loop level reproduces the singularities of UV-IR mixing. The latter is a new
limit, and corresponds to keeping the UV cut-off fixed in terms of the non-commutative
parameter .
A short version explaining the new scaling limit appeared in [7].
The fuzzy sphere is described by three matrices xiF = Li where Li s are the generators
of SU(2) for the spin l representation and has dimension of length. The radius R of the
sphere is related to and l as R 2 = 2 l(l + 1). The usual action for a matrix model on SF2
is

[Li , ] [Li , ]
R2
2 2
Tr
S=
(1.1)
+
m
+
V
[]
,
2l + 1
R2
and has the right continuum limit as l . Because of the non-commutative nature of SF2 ,
there is a natural ultra-violet (UV) cut-off: the maximum energy 2max is = 2l(2l + 1)/R 2 .
To get the theory on a non-commutative plane, the usual strategy is to restrict to (say)
the north pole, define the non-commutative coordinates as xaNC xaF (a = 1, 2), and then
take both l and R to infinity in a precisely
specified manner. For example, a commonly
used limit requires us to hold = R/ l fixed as both R and l increase, which gives us a
non-commutative plane with [10,11]
NC NC
x1 , x2 = i 2.
(1.2)
It is easy to see that in this limit, max diverges, while tends to zero. This is the analogous
to the standard stereographic projection.
404
A second scaling limit which is of interest to us here is one in which R and l become
large with non-commutativity parameter given now by = R/ l kept fixed. The above noncommutativity relation becomes simply [x1NC , x2NC ] = iR which means that xaNC s are
now strongly non-commuting coordinates (R ) and hence nonplanar amplitudes are
expected to simply drop out in accordance with [5]. This can also be seen from the fact that
in this scaling (as is obvious from the relation R 2 = 2 l(l + 1)) max no longer diverges:
it is now of order 1/ , and there are no momentum modes in the theory larger then this
value. Alternatively we will also show that
in this limit the non-commutative coordinates
can be instead identified as XaNC = xaNC / l with non-commutative structure
NC NC
X1 , X2 = i 2 .
(1.3)
While this scaling for obtaining R2 is simply stated, obtaining the corresponding theory
with the above criteria is somewhat subtle. Indeed passing from [x1NC , x2NC ] = iR to
(1.3) corresponds in the quantum theory to a re-scaling of momenta sending thus the finite
cut-off = 2 to infinity. In order to bring the cut-off back to a finite value x = x,
where x is an arbitrary positive real number, we modify the Laplacian on the fuzzy sphere
= [Li , [Li , . . .]]so that to project out modes of momentum greater than a certain value
j given by j = [ 2 x l ]. In other words, the theory on the non-commutative plane R2 with
UV cut-off 1 is obtained by flattening not the full theory on the fuzzy sphere but only
a low energy sector. One can argue that only for when x = that the
canonical UVIR singularities become smoothen out. At this value we have j = [2 l ] which marks
somehow the boundary between commutative and non-commutative field theories.
The generalization to non-commutative R4 is obvious. We work on SF2 SF2 and then
take the scaling limit with fixed, which is the case of most interest in this article. By
analogy with (1.1), the scalar theory with quartic self-interaction on SF2 SF2 is
S=
Rb2
Ra2
2la + 1 2lb + 1
(a)

(a)
(b)
(b)
[Li , ] [Li , ] [Li , ] [Li , ]
4 4
2 2
Tra Trb
+
+
+
,
l
Ra2
4!
Rb2
(1.4)
s are the
where a and b label the first and the second sphere, respectively, and L(a,b)
i
generators of rotation in spin la,b -dimensional representation of SU(2), and is a
(2la + 1) (2la + 1) (2lb + 1) (2lb + 1) Hermitian matrix. As la , lb go to infinity,
we recover the scalar theory on an ordinary S 2 S 2 .
Our strategy for obtaining the theory on non-commutative R4 is straightforward: as
discussed in [8,14], we expand of action (1.4) in terms of SU(2) polarization tensors
(for definition and various properties of polarization tensors, see, for example, [16]). Using
standard perturbation theory and a conventional renormalization procedure, we calculate
the two- and four-point correlation functions, and then we scale R, l with fixed.
Actually (and as we just have said), implementation of the new scaling is somewhat subtle,
in that we will need to work not with the full theory on SF2 SF2 but with a suitably
defined low-energy sector. This low-energy sector is selected by projecting out the high
energy modes in an appropriate manner using projection operators, and thus working with
405
a modified Laplacian:
1
j = + (1 Pj ), j = [2 l ],

where Pj is the projector on all the modes associated with the eigenvalues k = 0, . . . , j ,
and is the canonical Laplacian on the full fuzzy sphere SF2 SF2 . The flattening limit
(1.3) is thus implemented on the scalar field theory (1.4) as the limit in which we first take
0 above, then we proceed with R, l keeping = R/ l fixed.
An obvious consequence of our scaling procedure is that the correlation functions are
not singular functions of external momenta.
There is a nice intuitive explanation for using the modified Laplacian. If the momenta
are cut-off at too low a value, the system becomes in the commutative regime, while
if
the cut-off is too close to l, the system remains non-commutative. The choice [2 l ] for
the cut-off is in some sense the edge between these two situations: there is some noncommutativity in the behaviour, but there is no UV-IR mixing. We will have more to say
about this in Section 4.
The paper is organized as follows: in Section 2 we quantize 4 theory on SF2 SF2
and obtain the one-loop corrections to the 2- and 4-point functions. We also define in this
section the precise meaning of UV-IR mixing on S2F S2F and write down the effective
action. Section 3 is the central importance, in which we define continuum planar limits of
the fuzzy sphere. In particular, we show how the singular UV-IR mixing emerges in the
canonical limit of large stereographic projection of the spheres onto planes. We also show
that in a new continuum flattening limit, a natural momentum space cut-off (inversely
proportional to the non-commutativity parameter ) emerges, and as a consequence the
UV-IR mixing is completely absent. Section 3 contains also the computation of the
continuum limit of the 4-point function. As it turns out we recover exactly the planar oneloop correction to the 4-point function on non-commutative R4 . We conclude in Section 4
with some general observations.
2. Effective action on S2F S2F

In this section, we will set up the quantum field theory on SF2 SF2 , making explicit
our notation and conventions. These reflect our intent to consider SF2 SF2 as a discrete
approximation of non-commutative R4 .
(a) (a)
Each of the spheres ( i xi xi = R (a)2 , a = 1, 2) is approximated by the algebra
Mat2li +1 of (2li + 1) (2li + 1) matrices. The quantization prescription is given as usual,
by
n(a)
i =
xi(a)
R (a)
L(a)
i
.
n(a)F
=
i
la (la + 1)
(2.1)
This prescription follows naturally from the canonical quantization of the symplectic
structure on the classical sphere (see, for example, [18]) by treating it as the co-adjoint
orbit SU(2)/U (1). The L(a)
i s above are the generators of the IRR representation la of
406

(a)
(a)
(a)
SU(2): they satisfy [Li , Lj ] = iij k Lk and

3
(a)2
i=1 Li
= la (la + 1). Thus

i
=
ab ij k n(a)F
, n(b)F
.
n(a)F
i
j
k
l(l + 1)
(2.2)
Formally, SF2 SF2 is the algebra A = Mat2l1 +1 Mat2l2 +1 generated by the identity 1 1
(1)
(2)
together with Li 1 and 1 Li . This algebra A acts trivially on the (2l1 + 1)(2l2 + 1)dimensional Hilbert space H = H1 H1 with an obvious basis {|l1 m1 |l2 m2 }.
(a)
(a) (a)
The fuzzy analogue of the continuum derivations Li = iij k nj k are given by the
adjoint action: we make the replacement
(a)
(a)L
L(a)R
.
L(a)
i Ki = Li
i
(2.3)
s generate a left SO(4) (more precisely SU(2) SU(2)) action on the algebra A
The L(a)L
i
(a)L
(a)
(a)R
given by Li M = Li M where M A. Similarly, the Li s generate a right action on
(a)R
(a)
(a)
the algebra, namely Li M = MLi . Remark that Ki s annihilate the identity 1 1 of
the algebra A as is required of a derivation.
In fact, it is enough to set la = lb = l and Ra = Rb = R as this corresponds in the limit
to a non-commutative R4 with a Euclidean metric on R2 R2 . The general case simply
corresponds to different deformation parameters in the two R2 factors, and the extension
of all results is thus obvious (see Eq. (6) of [1]).
In close analogy with the action on continuum S 2 S 2 , we put together the above
ingredients to write the action on SF2 SF2 :

(2)
1 (1) (1)
R4
1
Sl =
Tr
Li , Li , + 2 L(2)
H
i , Li ,
2
2
(2l + 1)
R
R

S (0) + Slint .
(2.4)
+ 2l 2 + V ()
l
This action has the correct continuum (i.e., l , R fixed) limit:

1 (2) (2)
d (1) d (2) 1
(1) (1)
4
S = R
Li Li () + 2 Li Li ()
4
4
R2
R
S2

+ 2 2
+ V () .
(2.5)
While the technology presented here can be applied to any polynomial potential, we will
= 4,l 4 . We have explicitly introduced factors of R wherever
restrict ourselves to V ()
4!
necessary to sharpen the analogy with flat-space field theories: the integrand R 4 d1 d2
has canonical dimension of (Length)4 like d 4 x, the field has dimension (Length)(1) , l
has (Length)(1) and 4,l is dimensionless.
Following [8,13,14], the fuzzy field can be expanded in terms of polarization
operators [16] as
= (2l + 1)
k1
2l

2l

p1

k1 =0 m1 =k1 p1 =0 n1 =p1
k1 m1 p1 n1 Tk1 m1 (l) Tp1 n1 (l)
(2l + 1)
11 T1 (l1 ) T1 (l2 ).
407
(2.6)
11
In our shorthand notation 11 (for k1 m1 p1 n1 ), the quantum numbers from the first sphere
come with subscript 1 (as in (k1 m1 )), as do those for the second sphere.
The Tkm (l) are the polarization tensors which satisfy
1
(a)
K Tk1 m1 (l) = k1 (k1 + 1) m1 (m1 1) Tk1 m1 1 (l),
2
(a)2
(a)
K
Tk1 m1 (l) = k1 (k1 + 1)Tk1 m1 (l),
K Tk1 m1 (l) = m1 Tk1 m1 (l),
3
and the identities

TrH Tk1 m1 (l)Tp1 n1 (l) = (1)m1 k1 p1 m1 +n1 ,0 ,
Tk1 m1 (l) = (1)m1 Tk1 m1 (l).
The field has a finite number of degrees of freedom, totaling to (2l1 + 1)2 (2l2 + 2)2 .
Our interest is restricted to Hermitian fields since they are the analog of real fields
we obtain the conditions k1 m1 p1 n1 =
in the continuum. Imposing Hermiticity = ,
(1)m1 +n1 k1 m1 p1 n1 .
Since the field on our fuzzy space has only a finite number of degrees of freedom, the
simplest and most obvious route to quantization is via path integrals. The partition function

d 11 d 11
int
Z = N D eSl Sl , D =
(2.7)
2
11
for the theory yields the (free) propagator

(1)m2 +n2
k1 k2 m1 ,m2 p1 p2 n1 ,n2
k1 m1 p1 n1 k2 m2 p2 n2 =
.
R2
k1 (k1 + 1) + p1 (p1 + 1) + R 2 2l
(2.8)
The Euclidean 4-momentum in this setting is given by (11) (k1 , m1 , p1 , n1 ) with

square (11)2 = k1 (k1 + 1) + p1 (p1 + 1). For quartic interactions, the vertex is given
by the expression

Slint =
(2.9)
V (11, 22, 33, 44) 11 22 33 44 ,
11
22
33
44
with
4
V1 (1234, km)V2(1234, pn), where
4!

V1 (1234, km) = (2l + 1) TrH1 Tk1 m1 (l)Tk2 m2 (l)Tk3 m3 (l)Tk4 m4 (l)
V (11, 22, 33, 44) = R 4
(2.10)
and similarly for V2 (1234, pn).

2.1. The 2-point function
The energy of each mode k1 m1 p1 n1 is the square of the fuzzy 4-momentum, namely
(11)2 = k1 (k1 + 1) + p1 (p1 + 1). Since m1 = k1 , . . . , k1 and n1 = p1 , . . . , p1 , there
are (2k1 + 1)(2p1 + 1) modes with the same energy for each pair of values (k1 , p1 ),
408
Fig. 1. The 2-point function at 1-loop.
and may thus be thought of as naturally forming an energy shell. Integrating out the high
energy modes (with (k1 = 2l1 , p1 = 2l2 ) in the path integral implements for us the shell
approach to renormalization group adapted to fuzzy space field theories [8].
Integrating out only over the high momentum modes 1f 1f (k = 2l1 , m, p = 2l2 , n),
the terms in the action that contribute to the 2-point function at one-loop are given by

2 2,
3f 3f , 4f 4f ) 1 1 2 2 3f 3f 4f 4f
S2(1) = + 4
V (1 1,
1 1
+2
2 2 3f 3f 4f 4f
2f 2f , 3 3,
4f 4f ) 11 2f 2f 33 4f 4f + .
V (1 1,
1 1 2f 2f 3 3 4f 4f
(2.11)
The ellipsis indicate omitted terms that are unimportant

for
function calculation.

the 2-point

The notation is that of Eqs. (2.6) and (2.9), and 1 1 = 11 1f 1f . The relative factor
in the above is 4 to 2 since there are 4 ways to contract two neighboring fields (i.e.,
planar diagrams) and only two different ways to contract non-neighboring fields (nonplanar diagrams). The relevant graphs are displayed in Fig. 1.
Instead of integrating out only one shell, one could integrate out an arbitrary number of
them. For example, integrating out q 2 shells gives

(q 2 )
2 2,
3f 3f , 4f 4f ) 1 1 2 2 3f 3f 4f 4f
S2 = + 4
V (1 1,
1 1
+2
2 2 3f 3f 4f 4f
2f 2f , 3 3,
4f 4f ) 11 2f 2f 33 4f 4f +
V (1 1,
1 1 2f 2f 3 3 4f 4f
(2.12)
with now

2l1

1f 1f
2l2

k=2l1 (q1) m=k p=2l2 (q1) n=p

1 1

11

1f 1f
while the partition function (2.7) takes the form

Z = N
q 2 2
q2 2

S2 f S2 f
2

q
int
D exp Sl Si S2 f +
+ ,
2
where D =
d 1 1 d 1 1
1 1
(2.13)
409
For l1 = l2 = l, the full one-loop corresponds to integrating over all shells, i.e., q =
2l + 1. The corresponding effective action is
(q 2 )
k m p n 2
1 4,l P
1 1 1 1 .
S2 q=2l+1 = 2
(2.14)
l + NP
l (k1 , p1 )
R 4!
k1 m1 p1 n1
The 2-point function computation readily gives us the renormalized mass:

1 4,l P
l + NP
l (k1 , p1 ) ,
R 2 4!
where the planar contribution given by
2l (k1 , p1 ) = 2l +
Pl = 4
2l
2l

A(a, b),
A(a, b) =
a=0 b=0
(2.15)
(2a + 1)(2b + 1)
.
a(a + 1) + b(b + 1) + R 2 2l
(2.16)
On the other hand, the non-planar contribution is

NP
l (k1 , p1 ) = 2
2l
2l

A(a, b)(1)k1+p1 +a+b Bk1 p1 (a, b),
a=0 b=0
Bab (c, d) = (2l + 1)2
a
c
l
l
l
l

b
d
l
l
where

l
.
l
(2.17)
The symbol { } in Bab (c, d) is the standard 6j symbol (see, for example, [16]). As is
immediately obvious from these expressions, both planar and non-planar graphs are finite
and well-defined for all finite values of l. However, a measure for the fuzzy UV-IR mixing
is the difference between planar and non-planar contributions, which we define below:
1
P
Pl + NP
l (k1 , p1 ) = l + (k1 , p1 ),
2
(k1 , p1 ) = 4
2l
2l

Pl = 6
2l
2l

A(a, b),
a=0 b=0

A(a, b) (1)k1 +p1 +a+b Bk1 p1 (a, b) 1 .
(2.18)
a=0 b=0
Were this difference (k1 , p1 ) to vanish, we would recover the usual contribution to
the mass renormalization as expected in a commutative field theory. The fact that this
difference is not zero in the limit of large IRRs l, i.e., l , is what is meant by UV-IR
mixing on fuzzy S2 S2 . Indeed this may be taken as the definition of the UV-IR problem
on general fuzzy spaces. In fact (2.18) can also be taken as the regularized form of the
UV-IR mixing on R4 . Removing the UV cut-off l while keeping the infrared cut-off
R fixed = 1 one can show that diverges as l 2 , i.e.

(k1 , p1 ) 8l 2
1 1
1 1

dtx dty
Pk1 (tx )Pp1 (ty ) 1 ,
2 tx ty
(2.19)
where, for simplicity, we have assumed l l [12]. (2.19) is worse than the case of two
dimensions (see Eq. (3.20) of [12]), in here not only the difference survives the limit but
410
also it diverges. This means in particular that the UV-IR mixing can be largely controlled
or perhaps understood if one understands the role of the UV cut-off l in the scaling limit
and its relation to the underlying star product on S2F .
2.2. The 4-point function
The computation of higher order correlation functions become very complicated, but
this exercise is necessary if we want to compute, for example, the beta-function. It is
also useful to put forward key features which will be needed (in the future) to study noncommutative matrix gauge theories and their continuum limits. We will only look at the
four-point function here.
Our starting point is (2.13), which tells us that integrating out q 2 shells produces the
following correction to the 4-point function:
(q 2 ) 2
S2
f

2,
3,
5;
4,
6,
7 , 8)
1235 4f 4f 6f 6f 7f 7f 8f 8f
= W (1,
f
f
4 4 7 7 6 6 8 8
f
f
f
f
f
f
f
f f
f

+ 4f 4f 8f 8f f 7f 7f 6f 6f f .
2,
3,
5;
4,
6,
7 , 8)
= W2 (1 1,
2 2,
4f 4f , 6f 6f )W2 (3 3,
5 5,
7f 7f , 8f 8f ) such that
Here, W (1,
4f 4f ), 1235 =
W2 (11, 22, 3f 3f , 4f 4f ) = 4V (11, 22, 3f 3f , 4f 4f ) + 2V (11, 3f 3f , 2 2,
11 22 33 55 , and the notation is that of Eqs. (2.6), (2.9) and (2.12). Inserting the free
propagator (2.8) above yields the 4-point function
(q 2 ) 2
) f
(S2
(q 2 ) 2
f
S2

1 1
2 2
3 3
5 5
R4
4 1 1 2 2 3 3 5 5
4 (1235),
4!
(2.20)
where
4 (1235) =
4
4!
k4 ,k6 =f p4 ,p6 =f
A(k4 , p4 )A(k6 , p6 )
(2k4 + 1)(2k6 + 1)(2p4 + 1)(2p6 + 1)
(1) (1)
(2) (2)
(3) (3)
(4) (4)
81 2 + 161 2 + 41 2 + 81 2 .
(2.21)
The first graph in (2.21) is the usual one-loop contribution to the 4-point function,
i.e., the two vertices are planar. The fourth graph contains also two planar vertices but
with the exception that one of these vertices is twisted, i.e., with an extra phase. The
second graph contains on the other hand one planar vertex and one non-planar vertex,
whereas the two vertices in the third graph are both non-planar. The relevant graphs
are displayed in Figs. 2 and 3. The analytic expressions for i(a) i(a)(k4 k6 ; 1235) =
k6
k4
(a)
m4 =k4
m6 =k6 i (k4 k6 ; 1235) are given by
411
Fig. 2. The planar 4-point function at 1-loop. The second graph has one of its vertices rotated by 180 degrees, it
is still planar.
Fig. 3. The non-planar 4-point function at 1-loop: the 1st graph has 1 non-planar vertex while in the 2nd graph
both vertices are non-planar.
(1)
f 6f )Vi (3 5 4f 6f ),
= (1)m4 +m6 Vi (1 24
(2)
f 6f )Vi (3 4f 5 6f ),
= (1)m4 +m6 Vi (1 24
(3)
f )Vi (3 4f 5 6f ),
f 26
= (1)m4 +m6 Vi (14
(4)
f 6f )Vi (3 5 6f 4f ),
= (1)m4 +m6 Vi (1 24
i
i
i
i
where the lower index in s and s labels the sphere whereas the upper index denotes
the graph, and the notation 4f 4f stands for (k4 , m4 , p4 , n4 ) in contrast with 4f 4f =
(k4 , m4 , p4 , n4 ).
412
By using extensively the different identities in [16] we can find after a long calculation
that the above 4-point function has the form
4 (1235) =

4
(1)
(2)
(3)
(4)
84 (1235) + 164 (1235) + 44 (1235) + 84 (1235) ,
4!
where
(a)
4 (1235) =
(a)
(a)
A(k4 , p4 )A(k6 , p6 )1 (k4 k6 ; 1235)2 (p4 p6 ; 1235),
k4 ,k6 =f p4 ,p6 =f
a = 1, . . . , 4.
(2.22)
The label f stands for the shells we integrated over and hence it corresponds to q 2 =
(2l + 1)2 for the full one-loop contribution. The planar amplitudes, in the first R2 factor
for example, are given by

(1)
k k
k k
(1)k+k4 +k6 k (1235)Ek14k26 (k)Ek34k56 (k),
1 =
k
(4)
1
k k
k k
k (1235)Ek14k26 (k)Ek34k56 (k)
(2.23)
whereas the non-planar amplitudes are given by

k k
k k
(1)k3 +k4 k (1235)Ek14k26 (k)Fk34k56 (k),
1(2) =
k
1(3)
(1)k2 +k3 k (1235)Fk16k24 (k)Fk34k56 (k)

k k
k k
(2.24)
with

k4 l
l
+ 1) (2k1 + 1)(2k2 + 1) k6 l
l ,
k k1 k2

k1 k2 k
k4
k4 k6
Ek1 k2 (k) = (2l + 1) (2k1 + 1)(2k2 + 1)
l
l l
l
k k
Fk14k26 (k) = (2l
k6
l

k
.
l
(2.25)
The fuzzy delta function k (1235) is defined by

k (1235) = (1)m Ckkm
C km
.
1 m1 k2 m2 k3 m3 k5 m5
(2.26)
The justification for this name will follow shortly.

The full effective action at one-loop of the above scalar field theory on S2F S2F is
obtained by adding the two quantum actions (2.14) and (2.20) to the classical action (2.4).
3. Continuum planar limits

We can now state with some detail the continuum limits in which the fuzzy spheres
approach (in a precise sense) the non-commutative planes. There are primarily two limits
of interest to us: one is the canonical large stereographic projection of the spheres onto
413
planes, while the second is a new flattening limit which we will argue corresponds to a
conventional cut-off.
For simplicity, consider a single fuzzy sphere with cut-off l and radiusR, and define
F = x F ix F ) where = R/ l(l + 1). The
the fuzzy coordinates xiF = Li (i.e., x
1
2
stereographic projection [9,10] to the non-commutative plane is realized as
F
F
= 2Rx+
y+
1
,
R x3F
F
y
= 2R
1
F
x
.
R x3F
(3.1)
In the large l limit it is obvious that these fuzzy coordinates indeed approach the canonical
stereographic coordinates. A planar limit can be defined from above as follows:
R2
2 =
= fixed as l, R .
l(l + 1)
In this limit, the commutation relation becomes
NC NC
NC
F
F
y
y
= x
,
y+ , y = 2 2,
(3.2)
(3.3)
where we have substituted L3 = l corresponding to the north pole. The above

commutation relation may also be put in the form
NC NC
xaNC xaF , a = 1, 2.
x1 , x2 = i 2,
(3.4)
The minus sign is simply due to our convention for the coherent states on co-adjoint orbits.
The extension to the case of two fuzzy spheres is trivial.
A second way to obtain the non-commutative plane is by taking the limit
R
= l, R .
(3.5)
l(l + 1)
A UV cut-off is automatically built into this limit: the maximum energy a scalar mode can
have on the fuzzy sphere is 2l(2l + 1)/R 2 , which in this scaling limit is 4/ 2 . There are
no modes with energy larger than this value. To understand this limit a little better, let us
n0 , l|L3 |
n0 , l = l
restrict ourselves to the north pole n = n 0 = (0, 0, 1) where we have
n0 , l = 0, a = 1, 2. The commutator [L1 , L2 ] = iL3 = il, so the nonand
n0 , l|La |
commutative coordinates on this non-commutative plane tangential to the north pole can
be given either simply by xaF as above. This now defines a strongly non-commuting plane,
viz.
F F
xa , xb = il 2 ab .
(3.6)

Or alternatively one can define the non-commutative coordinate by XaNC R xaF ,
satisfying
NC NC
Xa , Xb = i 2 ab .
(3.7)
=
In the convention used here, 12 = 1 and ac cb = ab .

Intuitively, the second scaling limit may be understood
as follows. Non-commutativity
2
introduces a short distance cut-off of the order X = 2 because of the uncertainty re2
lation X1NC X2NC 2 . However, the Laplacian operators on generic non-commutative
414
planes do not reflect this short distance cut-off, as they are generally taken to be the same
as the commutative Laplacians. On the above non-commutative plane (3.7) the cut-off
X effectively translates into the momentum space as some cut-off P = 1 2 . This
2
is because of (and in accordance with) the commutation relations [XaNC , PbNC ] = iab ,
PaNC = 12 ab XbNC , giving us the uncertainty relations XaNC PbNC 2ab . Since one
cannot probe distances less than X, energies above P should not be accessible either,
i.e., [PaNC , PbNC ] = i2 ab . The fact that the maximum energy of a mode is of order 1/
in the second scaling limit ties in nicely with this expectation.
The limit (3.5) may thus be thought of as a regularization prescription of the noncommutative plane which takes into account our expectation of UV-finiteness of noncommutative quantum field theories.
3.1. Field theory in the canonical planar limit
We are now in a position to study what happens to the scalar field theory in the
limit (3.4). First we match the spectrum of the Laplacian operator on each sphere with
the spectrum of the Laplacian operator on the limiting non-commutative plane as follows
a(a + 1) = R 2 pa2 ,
(3.8)
where pa is of course the modulus of the two-dimensional momentum on the noncommutative plane which corresponds to the integer a, and has the correct mass dimension.
However, since the range of as is from 0 to 2l, the range of pa2 will be from 0 to
2l(2l+1)
= l 2 , = 2/ . In other words, all information about the UV cut-off
R2
is lost in this limit.
Let us see how the other operators in the theory scales in the above planar limit. It is not
difficult to show that the free action scales as

2

R 2 a(a + 1) + R 2 b(b + 1) + R 4 2l abma mb
a,b ma ,mb
pa pb a b 2
d 2 pa d 2 pb 2
.
pa + pb2 + M 2 NC
2
(3.9)
l
p p
a b a b
R 4 abma mb , which
The scalar field is assumed to have the scaling property NC
gives the momentum-space scalar field the correct mass dimension of 3 [recall that
[ abma mb ] = M]. The a and b above (not to be confused with the scalar field!) are
the angles of the two momenta pa and pb , respectively, i.e., a = m1a and b = m1b .
a+ 2
b+ 2
This formula is exact, and can be simplified further when quantum numbers as and bs
are large: the a and b will be in the range [, ]. It is also worth pointing out that
the mass parameter M of the planar theory is exactly equal to that on the fuzzy spheres,
i.e., M = , and no scaling is required. This is in contrast with [12] but only due to our
definition of the fuzzy action (2.4).
415
With these ingredients, it is not then difficult to see that the flattening limit of the planar
2-point function (2.16) is given by

Pl
pa pb dpa dpb
P
M 2 = 16
(3.10)
R
pa2 + pb2 + M 2
which is the 2-point function on non-commutative R4 with a Euclidean metric R2 R2 .
By rotational invariance it may be rewritten as

d 4k
4
P
.
M = 2
(3.11)

k2 + M 2
l
We do now the same exercise for the non-planar 2-point function (2.17). Since the external
momenta k1 and p1 are generally very small compared to l, one can use the following
approximation for the 6j -symbols [16]

(1)a+b
b2
a l l
Pa 1 2 , l , a l, 0 b 2l.
(3.12)
b l l
2l
2l
By putting in all the ingredients of the planar limit we obtain the result
M
NP
NP
(k1 , p1 ) l2 = 8
R

0 0

4 pb2
4 pa2
Pk1 1
Pp1 1
.
2R 2
2R 2
pa2 + pb2 + M 2
pa pb dpa dpb
Although the quantum numbers k1 and p1 in this limit are very small compared to l, they
are large themselves, i.e., 1 k1 , p1 l. On the other hand, the angles a defined by
4 p2
2
cos a = 1 2R 2a can be considered for all practical purposes small, i.e., a = Rpa
because of the large R factor, and hence we can use the formula (see, for example, [20],
page 72)

2 a J1 ()
4 a
J2 () + J3 () + O sin
Pn (cos a ) = J0 () + sin
(3.13)
,
2
2
6
2
for n 1 and small angles a , with = (2n + 1) sin 2a . To leading order we then have

2
2

1
2
4 pa2
Pk1 1
da ei cos a pk1 pa .
= J0 pk1 pa =
2
2R
2
0
This result becomes exact in the strict limit of l, R where all fuzzy quantum numbers
diverge with R. We get then

2
(pa dpa da )(pb dpb db ) i 2 pk (pa cos a )
1
e
M NP (pk1 , pp1 ) = 2
pa2 + pb2 + M 2
ei
2p
p1 (pb cos b )
By rotational invariance we can set 2 B pk1 pa = 2 pk1 (pa cos a ), where B 12 = 1.

In other words, we can always choose the two-dimensional momentum pk1 to lie in the ydirection, thus making a the angle between pa and the x-axis. The same is also true
416
for the other exponential. We thus obtain the canonical non-planar 2-point function on the
non-commutative R4 (with Euclidean metric R2 R2 ). Again by rotational invariance, this
non-planar contribution to the 2-point function may be put in the compact form

d 4k
2
2
NP
ei pBk .
M (p) = 2
(3.14)
2
2

k +M
l
The structure of the effective action in momentum space allows us to deduce the star
products on the underlying non-commutative space. For example, by using the tree level
action (3.9) together with the one-loop contributions (3.11) and (3.14) one can find that the
effective action obtained in the large stereographic limit (3.2) is given by

2 k

g42
d 4 p 1 2
d 4k
1
d 4 k ei pB
2
p + M +
2
+
6
(2)4 2
(2)4 k 2 + M 2
(2)4 k 2 + M 2
l
l

2
1 (p)

l
(3.15)
pa pb a b
where g42 = 8 2 4 and 1 (p)
= 4 2 NC
and l . This effective action
can be obtained from the quantization of the action

g2
1
1
d 4 x ( 1 )2 + M 2 12 + 4 1 1 1 1 ,
2
2
4!

4
NC
d p
ipx = 1 and is the canonical (or Moyal
where 1 1 (x NC ) = (2)
4 1 (p)e
Weyl) star product

i 2 y z
f g x NC = e 2 B f (y)g(z)y=z=x NC .
(3.16)
This is consistent with the commutation relation (3.4) and provides a nice check that
the canonical star product on the sphere derived in [21] (also given here by Eq. (2.2))
reduces in the limit (3.2) to the above MoyalWeyl product (3.16). In the above, B is the
antisymmetric tensor which can always be rotated such that the non vanishing components
are given by B 12 = B21 = 1 and B 34 = B43 = 1.
In fact one can read immediately from the above effective action that the planar
contribution is quadratically divergent as it should be, i.e.

d 4k
1
1
1
P
M
=
=
l 2 ,
M P =
(3.17)
2
4 k2 + M 2
2
64
(2)
16
l
whereas the non-planar contribution is clearly finite

1
M NP (p)
32 2

2 k

2

1
2
d 4 k ei pB
2
=
=
+
M
ln
EM
,
(2)4 k 2 + M 2 8 2 E 2 4
M NP (p) =
where E
l
= B P .
(3.18)
417
This is the answer of [5]: it is singular at P = 0 as well as at = 0.

3.2. A new planar limit with strong non-commutativity
As explained earlier, the limit (3.5) possesses the attractive feature that a momentum
space cut-off is naturally built into it. In addition to obtaining a non-commutative plane in
the strict limit, UV-IR mixing is completely absent. But while the new scaling is simply
stated, obtaining the corresponding field theory is somewhat subtle. We will need to modify
the Laplacian on the fuzzy sphere to project our modes with momentum greater than 2 l.
In other words, the non-commutative theory on a plane with UV cut-off is obtained not by
flattening the full theory
on the fuzzy sphere, but only a low energy sector, corresponding
to momenta up to 2 l.
In order to clarify the chain of arguments, we will first implement naively the limit (3.5)
and show that it corresponds to a strongly non-commuting plane. Finite non-commuting
plane is only obtainable if we pick a specific low energy sector of the fuzzy sphere before
taking the limit as we will explain in the next section.
Our rule for matching the spectrum on the fuzzy sphere with that on the noncommutative plane is the same as before, namely a(a + 1) = R 2 pa2 . However because
= 42 . The kinetic part of the action will
of (3.5), the range of pa2 is now from 0 to 2l(2l+1)
R2
scale in the same way as in (3.9), only now the momenta ps
in (3.9) are restricted such
that p . With this scaling information, we can see that the planar contribution to the
2-point function is given by

P
d 4k
4
2
mP 2l = 2
(3.19)
, = .
R
k 2 + 2l
k
We can similarly compute the non-planar contribution to the 2-point function using (3.12).
The motivation for using this approximation is more involved and can be explained as
follows. In the planar limit l, R , it is obvious that the relevant quantum numbers
k1 and p1 are in fact much larger compared to 1, i.e., k1 Rpk1 1 and p1
Rpp1 1, since R l. However, (3.12) can be used only if k1 , p1 l, or equivalently
2pk
2pp
= 1 1 and pl1 = 1 1. This is clearly true for small external momenta pk1
and pp1 , which is exactly the regime of interest in order to see if there is UV-IR mixing.
The condition for the reliability of the approximation (3.12) is then pexternal 1. We will
sometimes refer to this condition as small, the precise meaning of this phrase being
momentum scale of interest is much smaller than 1/ . We thus obtain
k1
l
NP
NP
(k1 , p1 ) l2 = 8
R

0 0

2 pb2
2 pa2
pa pb dpa dpb
P
.
1
P
k
p
1
1
2
2
pa2 + pb2 + 2l
(3.20)
2 p2
Now the angles a s of (3.13) are defined by cos a = 1 2 a , and since p 1, these
angles are still small. They are therefore given to the leading order in p by a = pa +
where the ellipsis indicate terms third order and higher in p. By using (3.13) we again
418
have

2
1
2 pa2
PRpk1 1
da eiR cos a pk1 pa .
= J0 (Rpk1 pa ) =
2
2
(3.21)
Using rotational invariance we can rewrite this as

d 4 k iRpBk
2
NP
m (p) = 2
e
.
k 2 + 2l
(3.22)
k
One immediate central remark is in order: the non-commutative phase contains now a
factor R instead of the naively expected factor of 2 . This is in contrast with the previous
case of canonical planar limit, where the strength of the non-commutativity 2 defined
by the commutation relation (3.4) is exactly what appears in the non-commutative phase
of (3.14). In other words this naive implementation of (3.5) yields in fact the strongly
non-commuting plane (3.6) instead of (3.7). Also we can similarly to the previous case
put together the tree level action (3.9) with the one-loop contributions (3.19) and (3.22) to
obtain the effective action

g42
d 4 p 1 2
d 4k
1
d 4k
1
2
iR pB
k
+
e
p + l +
2
(2)4 2
6
(2)4 k 2 + 2l
(2)4 k 2 + 2l

2
3 (p)
(3.23)
.

As before g42 = 8 2 4 , whereas 3 (p)

= l 3/2 2 ( l p),
2 (p)
= 4 l23 NC ( p ) with
l
p p
a b a b
NC
= R 4 abma mb (in the metric R2 R2 ). It is not difficult to see that
NC (p)
the one-loop contributions mP and mNP (p) given in (3.19) and (3.22) can also be given
by the equations

d 4k
1
l
P
P
m =
m
=
,
4 k 2 + l2
64 2
(2)
NP (p) =
m

l
NP p
m
=
32 2
l
d 4k
1
2 k

ei pB
.
(2)4 k 2 + l2l
(3.24)
P and m
NP (p) are given by
We have already computed that the leading terms in m

2
2
2
P= l
ln
1
+
,
m
l
16 2
2l

2

2
1
NP
2
+ ll ln l El , where E = B p .
m (p) =
8 2 E 2 4
Obviously then we obtain

2
P
2
2
m = 4 l ln 1 + 2 ,
l

mNP (p) = 42l ln l 2 El .
(3.25)
419
If we now require the mass l in (3.9) to scale as 2l = ml (the reason will be clear shortly),
then one can deduce immediately that the planar contribution mP is exactly finite equal
to 42 , whereas the non-planar contribution mNP (p) vanishes in the limit l .
Remark finally that despite the presence of the cut-off in the effective action (3.23),
this effective action can still be obtained from quantizing

g2
1
1
d 4 x ( 3 )2 + 2l 32 + 4 3 3 3 3 ,
(3.26)
2
2
4!
only we have to regularize all integrals in the quantum theory with a cut-off = 2/ .
d 4p
F
(3 3 (x F ) = (2)
ipx = 3 , and the star product is the MoyalWeyl
4 3 (p)e
product given in (3.16) with the obvious substitution R ).
3.3. A new planar limit with finite non-commutativity
Nevertheless, the action (3.23) can also be understood in some way as the effective
action on the non-commutative plane (3.7) with finite non-commutativity equal to 2 .
Indeed by performing the rescaling p p we get
l

g2
d 4 p 1 2
d 4k
1
p + m2 + 4 2
4
4 k 2 + m2
(2)
2
6
(2)
d 4k
1
2 k

ei pB
(2)4 k 2 + m2

2

2 (p)
.
(3.27)
We have already the correct non-commutativity 2 in the phase and the only
thing which
needs a new reinterpretation is the fact that the cut-off is actually given by l and
not by the finite cut-off . (Remark that if we do not reduce the cut-off l again to the
finite value , the physics of (3.27) is then essentially that of canonical non-commutativity,
i.e., the limit (3.5) together with the above rescaling of momenta is equivalent to the
limit (3.2)).
Now having isolated the l-dependence in the range of momentum space integrals in the
effective action (3.27), we can argue that it is not possible to get rid of this l-dependence
merely by changing variables. Actually, to correctly reproduce the theory on the noncommutative R4 given by (3.5) and (3.7), we will now show that one must start with a
modified Laplacian (or alternately propagator) on the fuzzy space [23]. For this, we replace
(a)
the Laplacian = [L(a)
i , [Li , . . .]] (see Eq. (2.3), a = 1, 2) on each fuzzy sphere which
has the canonical obvious spectrum k(k + 1), k = 0, . . . , 2l, with the modified Laplacian
1
j = + (1 Pj ).
(3.28)

Here Pj is the projector on all the modes associated with the eigenvalues k = 0, . . . , j , i.e.
Pj =
j
k

k=0 m=k
|k, mk, m|.
420
The integer j thus acts as an intermediate scale, and using the modified propagator gives
us a low energy sector of the full theory. We will fix the integer j shortly.
With this modified Laplacian, modes with momenta larger than j do not propagate: as a
result, they make no contribution in momentum sums that appear in internal loops. In other

words, summations like 2l
0 (which go over to integrals with range 0 ) now collapse to
j
j
j
0 (where the integrals now are of the range 0 , with j = 2l ).
The new flattening limit is now defined as follows: start with the theory on SF2 SF2 , but
with the modified propagator (3.28). First take 0, then R, l with = R/ l fixed.
j
,
This gives us the effective action (3.27) but with momentum space cut-off l j =
2 l
i.e.

l j

g42
d 4 p 1 2
2
p + m +
2
6
(2)4 2
d 4k
1
4
2
(2) k + m2
l j
l j
d 4k
1
2 k

ei pB
(2)4 k 2 + m2

2

2 (p)
.
(3.29)
This
also tells us that the correct choice of the intermediate scale is j = [2 l ] for which4
l j = . For this value of the intermediate cut-off, we obtain the non-commutative R
given by (3.5) and (3.7).
By looking at the product of two functionsof the fuzzy sphere, we can understand
better the role of the intermediate scale j (= [2 l ]). The fuzzy spherical harmonics Tla ma
go over to the usual spherical harmonics Yla ma in the limit of large l, and so does their
product, provided their momenta are fixed. Alternately, the product of two fuzzy spherical
harmonics T s is almost commutative (i.e., almost the same as that of the corresponding
Y s) if their angular momentum is small compared to the maximum angular momentum
l, whereas it is strongly non-commutative (i.e., far from the commutative regime) if
their angular momenta are sufficiently large and comparable to l. The intermediate cut-off
tells us precisely where the product goesfrom one situation to the other: Working with
fields having momenta much less than [2 l ] leaves us inthe approximately commutative
regime, while fields with momenta much larger than [2 l ] take us in the strongly noncommutative regime. In other words, the intermediate cut-off tells us where commutativity
and non-commutativity are in delicate balance. Indeed by writing (3.29) in the form

l j

2
g
d 4 p 1 2
p + m2 + 4 2
(2)4 2
6
d 4k
1
(2)4 k 2 + m2
l j
l j
d 4k
1
2 k

ei pB
(2)4 k 2 + m2

2

2 (p)


g42
d 4 p 1 2
d 4k
1
2
p

2
+
+
l,j
4
4
2
(2) 2
6
(2) k + 2l,j
j 2 2
d 4k
1
i(
) pB
k
e 2 l
4
2
(2) k 2 + l,j
421

(j ) 2
(p)

3
(2l)
j 3
j
j 2 2
) 2 (
p),
3 3 ). For j [2 l ], (
) 0
2 l
2 l
2 l
4
on a commutative R with cut-off = 2/ . For j [2 l ]
(j )
(2l,j = l2l ( 2 j l )2 , 3 (p)

=(
and this is the effective action

this effective action corresponds to canonical non-commutativity if we insist on the first
line above as our effective action or to strongly non-commuting R4 if we
consider instead
the effective action to be given by the second line. For the value j = [2 l ], where we obtain the non-commutative R4 given by (3.5) and (3.7), there seems to be a balance between
the above two situations and one can also expect the UV-IR mixing to be smoothen out.
To
show this we write first the one-loop planar and non-planar contributions for
j = [2 l ], viz.

d 4k
1
d 4k
1
2 k

NP
,
m
(p)
=
ei pB
.
mP =
4
2
2
4
2
2
(2) k + m
(2) k + m
these integrals by introducing a Schwinger parameter (k 2 + m2 )1 =

We can evaluate
2
d exp((k + m2 )). Explicitly, we obtain for the planar contribution

1
d m2
d (m2 +2 )
2
2
e
1
e
+
e
mP =
16 2
2

m2
1
2 + m2 ln 2
.
=
(3.30)
2
16
m + 2
Obviously the above planar function diverges quadratically as 2 when 0, i.e., the
non-commutativity acts effectively as a cut-off.
Next we compute the non-planar integral. To this end we introduce as above a
Schwinger parameter and rewrite the integral as follows
m
NP
1
(p) =
16 4
4 2
d e
m2 4E
i 2
d 4 k e k 2 E
2
[ 2r ]
2 r
i rs
1
4
16
s!(r 2s)!
r=0
s=0

E 2 s m2 4 E2
2
r2s ,
4
d
e
d 4 k ek (kE)
4i
0
E = B p .
In above we have also used the fact that is small in the sense we explained earlier (i.e.,
E 1) and in accordance with [15] to expand the second exponential around = 0. This
422
is also because the cut-off is inversely proportional to . (In the last line we used the

[ 2r ]
p
r
r
r
r1
identity
p=0
r=0
s=0 As,r2s , [ 2 ] = 2 for r even and [ 2 ] = 2 for
q=0 Aq,pq =
r odd.) It is not difficult to argue that the inner integral above vanishes unless r is even.
Using also the fact that the cut-off is rotationally invariant one can evaluate the inner
integral as follows. We have

2
n
d 4 k ek (kE)

= 4 E (n 1)!!
2
1
(2) 2 +2
n
n 2
q=1

1
1
1
,
(n 2q)!! (2)q+2 2q
where n is an even number given by n = r 2s.

We can now put the above non-planar function in the form
mNP (p) =

1 1 2 E 2N
d m2 4 E2
4
e
2
16
N!
2
N+2
N=0

N
M+1

(2 )P
2
M
e .
CN
(1)M 1
P!
M=0
(3.31)
P =0
N!
M =
(CN
M!(NM)! ). The first term in this expansion corresponds exactly to the case of
canonical non-commutativity where instead of we have no cut-off, i.e.
NP

N
1 1 2 E 2N
d m2 4 E2 M
4
(p) =
e
CN (1)M +
16 2
N!
2
N+2
N=0
M=0

4 2

1
2
1 (2)
2
2
2 E
=
+
m
ln
m
E
I
,
m
+ .
+
8 2 4 E 2
16 2
4
As expected this term provides essentially the canonical UV-IR mixing. As it turns out this
singular behaviour is completely regularized by the remaining N = 0 term in (3.31), i.e.
m
NP

1 (2) 2 4 E 2
(p) =
I
m ,
16 2
4

1

(2 )P 2
d m2 4 E2
1
4
+
e
e
+
16 2
2
P!
P =0

1 (2) 2 4 E 2
=
m ,
I
4
16 2

4 2
4 2
1
(2)
2
2 E
2 (1)
2
2 E
I
m + ,
m + ,
+ I
+ .
16 2
4
4
(3.32)
423
The integrals I (L) (x, y) are given essentially by Hankel functions, viz.

I
(1)
(x, y) =

d x y
1
(1)
=
e
iH0 (2i xy ) + h.c. ,

I
(L)
(x, y) =
d x y
e
L

L1

iL
1 i
x 2
(1)
(1)
2
HL2 (2i xy ) + HL (2i xy ) + h.c. ,
=
xy e
2 L1 y
L > 1.
(1)
Hankel functions admit the series expansion H0(1)(z) = 2i
ln z + and H (z) =
i(1)!
( 2z ) + for > 0 when z 0. In this case the mass m and the external
momentum E are both small compared to the cut-off

= 2/ and thus the dimensionless

mE
m2 E
parameters z xy = 2 or z xy = 2 1 + 2 are also small, in other words
we can calculate, for example, I (1) (x, y) = 2 ln(2 xy ), I (2) (x, y) = 2x ln(2 xy ) + y1
xy
and I (L) (x, y) = (L2)!
[1 (L2)(L1)
] for L 3. Thus the first term N = 0 in the above
y L1
sum (i.e., Eq. (3.32)) is simply given by
mNP (p) =

m2
2
ln
1
+
+ .
16 2
m2
(3.33)
As one can see it does not depend on the external momentum p at all. In the commutative
limit 0, this diverges logarithmically as ln which is subleading compared to the
quadratic divergence of the planar function. Higher corrections can also be computed and
2
E
one finds essentially an expansion in 2 E = E = 2
given by
mNP (p) =

m2
2
ln
1
+
16 2
m2

1 2 E 2(p1)
2 (1)
I
(x,
y)
p1,p2
16 2
p!
2
p=2

1 2 E 2p
1 (2)
+
I (x, y)
p,p2
16 2
p!
2
p=2
1
16 2
4 2 N

E
N=1
I (N+2) (x, y)

1 2 E 2p
p+N,p2
p!
2
p=2
p2
4 2
(1)M
(x = m2 + 2 , y = 4E , p+N,p2 = M=0 M!(p+NM)!
). It is not difficult to find that
the leading terms in the limit of small external momenta (i.e., E/ 1) are effectively
424
given by
m
NP

2

E2
E
m2
2
E
m2
1
+
O
(p) =
ln
1
+
ln
4
1
+
16 2
m2
4 2
2
2

2
E2
E
+
(3.34)
1+O
.
8 2
2
Clearly in the strict limit of small external momenta when E 0, we have E 2 ln E 0

and the non-planar contribution does not diverge (only the first term in (3.34) survives this
limit as it is independent of E) and hence there is no UV-IR mixing. The limit of zero
non-commutativity is singular but now this divergence has the nice interpretation of being
the divergence recovered in the non-planar 2-point function when the cut-off = 2 is
removed. This divergence is however logarithmic and therefore is sub-leading compared
to the quadratic divergence in the planar part.
The effective action (3.29) with j = [2 l ] can be obviously obtained from quantizing the action (3.26) with the replacements 2l m2 , 3 2 2 (XNC ) =
d 4p
NC
(p)e
ipX = 2 and where as before we have to regularize all integrals in the
(2)4 2
quantum theory with a cut-off = 2/ . The star product is the MoyalWeyl product
given in (3.16) with the substitutions , x NC XNC . This effective action can also
be rewritten in the form

g42
1
1 2
4
d x 2 2 + m 2 2 + 2 2 2 2 ,
(3.35)
2
2
4!
which is motivated by the fact that the effective star product defined by

NC
d 4p
d 4k
ipXNC eikXNC
f g X
=
f
(
p)

g(k)e
(2)4
(2)4

= d 4 y d 4 z 4 (y ) 4 (z )f (y y ) g(z z )
is such that
d 4 x f g(x) =
y=z=X NC
(3.36)
d 4p
4 (y ) is not

p).
The distribution
(2)4 f (p)g(
d 4 p ipy
4 (y ) =
4 (y ) tends to
rather
, i.e.,
(2)4 e
the
Dirac delta function 4 (y ) but

the
ordinary delta function in the limit of the commutative plane where the above
product (3.36) also reduces to the ordinary point-wise multiplication of functions. If the
cut-off was not correlated with the non-commutativity parameter , then the limit
would had corresponded to the limit where the product (3.36) reduces to the
MoyalWeyl product given in Eq. (3.16). This way of writing the effective action (i.e.,
(3.35)) is to insist on the fact that all integrals are regularized with a cut-off = 2/ . In
other words the above new star product which appears only in the kinetic part of the action
is completely equivalent to a sharp cut-off and yields therefore exactly the propagator
(3.28) with which only modes can propagate.
We should also remark here regarding non-locality of the star product (3.36). At first
sight it seems that this non-locality is more severe in (3.36) than in (3.16), but as it turns
425
out this is not entirely true: in fact the absence of the UV-IR mixing in this product also
suggests this. In order to see this more explicitly we first rewrite (3.36) in the form

f g XNC = d 4 y d 4 z f (y )g(z )K y , z ; XNC ,

4
4
K y , z ; XNC =
(y y )
(z z )|y=z=XNC .
The kernel K can be computed explicitly and is given by

K y , z ; XNC =

d 4k 4
2

NC
NC

X y + Bk eik(z X ) .
4
2
(2)
For the moment, let us say that and are unrelated. Then, taking to infinity gives [1,5]

K y , z ; XNC =
16
8 det B
2i
1
(z X NC )B 1 (y X NC )
e 2
.
4
(2)
If we have, for example, two functions f and g given by f (x) = 4 (x p) and g(x) =
4 (x p), i.e., they are non-zero only at one point p in spacetime, their star product which
is clearly given by the kernel K(p, p; XNC ) is non-zero everywhere in spacetime. The fact
that K is essentially a phase is the source of the non-locality of (3.16) which leads to the
UV-IR mixing.
On the other hand the kernel K (p, p; XNC ) with finite can be found in two
dimensions (say) to be given by

K p, p; XNC =
1
2
4

da (a + L1 )e
2i
L2 a
2

db (b + L2 )e
2i2 L1 b
with La = XaNC pa , a = 1, 2. If we now make the approximation to drop the remaining

(since the effects of this cut-off were already taken anyway) one can see that the above
integral is non-zero only for + p1 X1NC + p1 and + p2 X2NC + p2
simultaneously. In other words the star product K (p, p; XNC ) of f (x) and g(x) is also
localized around p within an error and is equal to 21 4 there. The star product (3.36) is
therefore effectively local.
Final remarks are in order. First we note that the effective star product (3.36) leads to an
effective commutation
(3.7) in which the parameter 2 is multiplied by an overall
4 relations

4

4
4 (z ), we simply skip the elementary proof. Remark
constant equal to d y d z (y )
also that this effective star product is non-associative as one should expect since it is for all
practical purposes equivalent to a non-trivial sharp momentum cut-off [22].
The last remark is to note that the prescription (3.28) can also be applied to the canonical
limit of large stereographic projection of the spheres onto planes, and inthis case one
can also obtain a cut-off = 2 with j fixed as above such that j = [2 l ]. The noncommutative plane (3.4) defined in this way is therefore completely equivalent to the above
non-commutative plane (3.7).
426
3.4. The continuum planar limit of the 4-point function

We now undertake the task of finding the continuum limit of the above 4-point function
(Eqs. (2.20) and (2.22)) which we expect to correspond to the 4-point function on the noncommutative R4 . This expectation is motivated of course by the result of the last sections
on the 2-point function. As it turns out this is also the case here and as an explicit example
we work out the continuum flattening limit of the planar amplitudes.
(1)
(4)
The planar diagrams are 4 and 4 . First, let us recall that in above the indices 4
and 6 refer to internal momenta whereas 1, 2, 3 and 5 refer to external momenta. Next,
since we are interested in the planar limits (in which R, l ) of the 4-point function,
we can use the asymptotic formula

(1)a+b+d+e
a
b
c
C cde
, l ,
=
(3.37)
d +l e+l f +l
(2l + 1)(2c + 1) af ebdf
which allows us to approximate in the limit the fuzzy delta function (2.26) as follows:
k (1235) = (2l + 1)(2k + 1)(1)k1 +k2 +k3 +k5 +m m1 +m2 +m3 +m5 ,0

k2
k
k5
k
k1
k3
.
m2 + l m1 + l l
m5 + l m3 + l l
(3.38)
We have also used the properties of the ClebschGordan coefficients to obtain the selection
rule m = m1 + m2 = m3 m5 , thus justifying the name. The next selection rule
k k
k k
comes from the fact that the function Ek14k26 (k)Ek34k56 (k) in the planar diagrams (2.23)
is proportional in the large l limit (by virtue of Eq. (3.37)) to Ckk01 0k2 0 Ckk03 0k5 0 (Ckk04 0k6 0 )2 ,
c0
whereas on the other hand these ClebschGordan coefficients are such that Ca0b0
!= 0 only
if a + b + c = even. This means in particular that k + k4 + k6 = even, k + k1 + k2 = even
and k + k3 + k5 = even, and hence one can argue in different ways that one can have, for
example,
k1 + k2 = k3 + k5 ,
k1 + k2 = k4 + k6 ,
k3 + k5 = k4 + k6 .
(3.39)
For obvious reasons we will only focus on this sector. As a consequence of these rules,
(1)
(4)
the planar graphs i and i are equal. Indeed for large l, one can easily show that these
diagrams take the form
i(1) = i(4)
(2l + 1)3 (1)m1 +m2 m1 +m2 +m3 +m5 ,0 m1 +m2 +m4 +m6 ,0 k1 +k2 ,k3 +k5 k1 +k2 ,k4 +k6
ak (1235)S(46; 1235), where
2

k1
k k6 k
S(46; 1235) =
(2k + 1) 4
l
l l
l
k
k2
l
k
l
2
k3
l
k5
l
k
l
2
, (3.40)

2,3,5
where ak (1235) =
i=1 (2ki + 1). As in the case of the 2-point function we have
assumed that the external momenta k1 , k2 , k3 and k5 are such that ki l, i = 1, 2, 3, 5. It is
also expected that the approximation sign becomes an exact equality only in the strict limit.
427
Furthermore from the

properties of the 6j -symbols, only the values 0 k k4 + k6 will
contribute to the sum k . Lastly we have also invoked in (3.40) the fact that for each fixed
pair (k4 , k6 ) which is integrated over in (2.22) the azimuth numbers (m4 , m6 ), although
they are already summed over, conspire such that their sum is m4 + m6 = m1 m2 .
From [16] we can now use the identity

2

k4 l l
k4 k6 k
k6 l l
k l l
=
(1)X1 (2X1 + 1)
,
l
l l
X1 l l
X1 l l
X1 l l
X1
(3.41)
etc. The delta function k1 +k2 ,k4 +k6 makes it safe to treat the internal momenta k4 and k6
as if they were small (recall that k4 and k6 are non-negative integers), and 0 k k4 + k6
means that k can be treated as small as well. One can therefore use the result (3.12) to
rewrite the above equation as

2
k4 k6 k
l
l l

2l

X12
X12
X12
1
=
(2X1 + 1)Pk4 1 2 Pk6 1 2 Pk 1 2 ,
(2l + 1)3
2l
2l
2l
X1 =0
etc. As we have already established, in the large l limit we can approximate this sum by
the integral

2
A B k
l l l

2 px21
2 px21
2 px21
PB 1
Pk 1
,
px1 dpx1 PA 1
2
2
2
0
(3.42)
with (A, B) = (k1 , k2 ), (k3 , k5 ) and (k4 , k6 ). We are obviously using the flattening limit
R
(3.5), i.e., = l(l+1)
, = 2 for reasons which will become self-evident shortly. Using
the result (3.21) we have

2 px21
2 px21
1
PA 1
(3.43)
dA dB eiR px1 (pA +pB ) ,
PB 1
=
2
2
(2)2
=
2R 2
(2l + 1)3
where pA pB = B pA pB , with B 12 = 1, and A , B have the interpretation of angles

between pA and pB , respectively, and the x-axis. Similarly, we have

2 px23
2 px21
2 px22
Pk 1
Pk 1
Pk 1
2
2
2

1
=
(3.44)
d1 d2 d3 eiR pk (px1 +px2 +px3 ) ,
(2)3
where now the angles i s are the angles between the vectors pxi s and the x-axis. Since R
is large, the integrals (3.43) with (A, B) = (k1 , k2 ), (k3 , k5 ) and (k4 , k6 ) are dominated
by those values of px1 , px2 and px3 such that px1 = pk4 + pk6 , px2 = pk1 + pk2 and
428
px3 = pk3 + pk5 , respectively, and correspondingly the integral (3.44) is dominated by
px1 + px2 + px3 = pk4 + pk6 . This is clearly a valid approximation because the conservation
law pk1 + pk2 + pk3 + pk5 = 0 is expected to hold (as we explain below) and because of the
large factor of R appearing in the different phases in (3.43) and (3.44). After we apply the
conservation law we may reinterpret the angles (say) 1 and 2 as the angles made by pk4
and pk6 and the x-axis respectively. Using all these ingredients one can convince ourselves
that the sum over k in (3.40) behaves in the limit as
1
S(46; 1235)
2l + 1
k4
k6
l
l
l
l
2
k1
k2
l
l
l
l

k3
k5
l
l

l
.
l
We now proceed to the task of rewriting this sum in terms of the non-commutative plane
variables. To this end we use the representation (3.21) in the form
Pk24

2 pk26
dk6
dk4
cos(R sin k4 pk4 pk6 )
cos(R sin k6 pk4 pk6 )
1
=
2
2
2

dk4 dk6
cos2 (R pk4 pk6 ),

2 2
where we have used the large R limit to go to the last line, i.e., since the angles 4 = 6 0
dominate the integrals in the limit, the two cosines become essentially equal. We have
also reinforced explicitly the symmetry of (3.40) under the exchange k4 k6 on each
6j -symbol in S above (as is also the case in (3.40)). The k4 and k6 have the natural
interpretation of angles between the vectors pk4 and pk6 , respectively, and the x-axis of the
plane. For the case (A, B) = (k1 , k2 ), we can use

2 pk22
d iR cos pk pk
1 2.
e
Pk1 1
=
2
2
However, here cannot be interpreted as the angle between pk1 (or pk2 ) with any specific
axis, but if 12 is the angle between the two vectors pk1 and pk2 then we can define
x = + 12 , and write

2 pk22
=
Pk1 1
2
2+
12
dx iR sin x pk pk iR cos x pk pk

iRpk1 pk2
1
2e
1 2 e
e
.
2
12
As before, since R is large, the integral is dominated by the value cos x = 0 or x =

can then evaluate the above sum S explicitly and find

1
dk4 dk6
cos(R pk1 pk2 )
S(46; 1235) =
5
(2l + 1)
2 2
2.
We
cos(R pk3 pk5 ) cos2 (R pk4 pk6 ),

where the symmetry of (3.40) under the exchanges k1 k2 and k3 k5 is now explicit.
This is essentially the phase of the planar 4-point function found in [17]. In order to see
429
this fact more clearly, we first show that (3.40) takes now the form
1(1) = 1(4)

ak (1235) dk4 dk6
cos(R pk1 pk2 )

2 2
R4
cos(R pk3 pk5 ) cos2 (R pk4 pk6 )
2 (pk1 + pk2 + pk3 + pk5 ) 2 (pk1 + pk2 + pk4 + pk6 ),
where we have also made the following interpretation of the limiting form of the 2dimensional fuzzy delta function
m
(1) 2
R2
k,k0 m,m0 (pk + pk0 ).
2l + 1
(3.45)
The factor (1) 2 is motivated by (2.8), the factor 2l + 1 is needed in order for (3.45) to
diverge correctly (in the limit) when k = k0 and m = m0 , while the R 2 factor is to restore
the correct mass dimension for the delta function. An identical formula will of course hold
for the other R2 factor, i.e.
(1)
(4)
2 = 2

ap (1235) dp4 dp6
cos(R pp1 pp2 )

R4
2 2
cos(R pp3 pp5 ) cos2 (R pp4 pp6 )
2 (pp1 + pp2 + pp3 + pp5 ) 2 (pp1 + pp2 + pp4 + pp6 ).
By putting the above functions 1(1,4) and 2(1,4) in Eq. (2.22), we easily obtain the 4(1)
(4)
dimensional one-loop planar contributions 4 and 4 and consequently the planar
contribution to the 4-point function P4 . Indeed we have
(4)
(1)
4 (1235) = 4 (1235)
a(1235) 4
=
(p1 + p2 + p3 + p5 )
R4 4

cos(R p1 p2 ) cos(R p3 p5 ) cos2 (R p4 p6 )
d 4 p4
,
2
2
(p42 + ml )((p1 + p2 + p4 )2 + ml )
where the notation (in the metric R2 R2 ) is p42 = pk24 +pp2 4 , d 4 p4 = 14 dpk24 dpp2 4 dk4 dp4 ,
4 (p1 + p2 + p3 + p5 ) = 2 (pk1 + pk2 + pk3 + pk5 ) 2 (pp1 + pp2 + pp3 + pp5 ) and
p1 p2 = pk1 pk2 + pp1 pp2 and a(1235) = ak (1235)ap (1235).
The associated effective action in this case can now easily be computed and we find
the final result (with some minor change of notation, namely, we denote now the internal
momentum p4 as k and denote the external momentum p5 as p4 )
4
g42
d p1 d 4 p2 d 4 p3 d 4 p4 P
(1234) 4(p1 + p2 + p3 + p4 )
4!
(2)4 (2)4 (2)4 (2)4 4
(3.46)
2 (p1 ) 2 (p2 )2 (p3 ) 2 (p4 ),
430
where
P4 (1234) =
32 2
g
3 4
d 4 k cos( 2 p1 p2 ) cos( 2 p3 p4 ) cos2 ( 2 k P )

,
2 + m2 )
(2)4
(k2 + m2 )((P + k)
P = p1 + p2 .
(3.47)
We have employed in above the same definitions as those

of the 2-point function
p1
used in (3.26), namely, g42 = 8 2 4 and 2 (p1 ) = 4 l23 NC (
). However, the
l
non-commutative field NC (p1 ) is now reinterpreted such that we have NC (p1 )
pk pp k p
NC1 1 1 1 = R 4 k1 p1 m1 n1 (2k1 + 1)(2p1 + 1) or, in other words, NC (p1 ) = NC
(p1 ) (2k1 + 1)(2p1 + 1). We notice immediately that Eq. (3.47) is exactly the result of
[17] up to a numerical factor. More precisely, (3.47) is to be compared with the first term
in the expansion of Eq. (5) of Ref. [17] which corresponds to the planar contribution to the
4-point function.
4. Conclusion
We have investigated in some detail the problem of obtaining theories on noncommutative R4 starting from finite matrix models defined on SF2 SF2 . Particular attention
was paid to a new limit that gives a theory on non-commutative R4 with a UV cut-off
proportional to the inverse of the non-commutativity parameter , and without any mixing
between UV and IR degrees of freedom.
The new scaling is implemented via the introduction of an intermediate scale [2 l ].

Intuitively, this intermediate scale carries information about the transition between
commutative and non-commutative regimes of the theory: if we only use modes with
momenta much smaller than this intermediate scale, the theory becomes commutative,
whereas modes with momenta much larger take us the non-commutative regime.
It would be interesting to extend this analysis to theories on SF2 and SF2 SF2 that have
fermionic and gauge [24] degrees of freedom, as well as supersymmetric theories [25]. We
also see no obstacle to using this method to study theories that are obtained from Kaluza
Klein reduction on fuzzy S 4 [26], as well as gauge theory on fuzzy CP2 [27].
Acknowledgements
It is a pleasure to thank A.P. Balachandran, Denjoe OConnor and Peter Prenajder
for numerous discussions. S.V. would like to thank DIAS for warm hospitality during the
final stage of this project. The work of S.V. is supported in part by DOE grant DE-FG0391ER40674.
References
[1] M.R. Douglas, N.A. Nekrasov, Rev. Mod. Phys. 73 (2001) 9771029, hep-th/0106048.
431
[2] A.P. Balachandran, B.P. Dolan, J.-H. Lee, X. Martin, D. OConnor, J. Geom. Phys. 43 (2002) 184204,
hep-th/0107099.
[3] G. Alexanian, A.P. Balachandran, G. Immirzi, B. Ydri, J. Geom. Phys. 42 (2002) 2853, hep-th/0103023.
[4] S.P. Trivedi, S. Vaidya, JHEP 0009 (2000) 041, hep-th/0007011.
[5] S. Minwalla, M.V. Raamsdonk, N. Seiberg, JHEP 0002 (2000) 020, hep-th/9912072.
[6] J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, JHEP 0005 (2000) 023, hep-th/0004147.
[7] S. Vaidya, B. Ydri, hep-th/0209131.
[8] S. Vaidya, Phys. Lett. B 512 (2001) 403411, hep-th/0102212.
[9] G. Alexanian, A. Pinzul, A. Stern, Nucl. Phys. B 600 (2001) 531547, hep-th/0010187.
[10] I. Kishimoto, JHEP 0103 (2001) 025, hep-th/0103018.
[11] A. Perelomov, Generalized Coherent States and Their Applications, Springer-Verlag, Berlin, 1986.
[12] C.-S. Chu, J. Madore, H. Steinacker, JHEP 0108 (2001) 038, hep-th/0106205.
[13] B.P. Dolan, D. OConnor, P. Prenajder, JHEP 0203 (2002) 013, hep-th/0109084.
[14] H. Grosse, C. Klimck, P. Prenajder, Int. J. Theor. Phys. 35 (1996) 231244, hep-th/9505175;
H. Grosse, C. Klimck, P. Prenajder, Commun. Math. Phys. 185 (1997) 155175, hep-th/9507074.
[15] R. Wulkenhaar, hep-th/0206018.
[16] D.A. Varshlovich, A.N. Moskalev, V.R. Khersonsky, Quantum Theory of Angular Momentum, World
Scientific, New Jersey, 1988.
[17] I. Arefeva, D. Belov, A. Koshelev, Phys. Lett. B 476 (2000) 431, hep-th/9912075.
[18] B. Ydri, Fuzzy physics, hep-th/0110006, Ph.D. Thesis.
[19] K.G. Wilson, J. Kogut, Phys. Rep. C 12 (1974) 75.
[20] W. Magnus, F. Oberhettinger, Formulas and Theorems for the Special Functions of Mathematical Physics,
Chelsea, New York, 1949.
[21] A.P. Balachandran, B.P. Dolan, J. Lee, X. Martin, D. OConnor, hep-th/0107099.
[22] B. Ydri, Phys. Rev. D 63 (2001) 025004, hep-th/0003232.
[23] D. OConnor, S. Vaidya, B. Ydri, unpublished.
[24] S. Iso, Y. Kimura, K. Tanaka, K. Wakatsuki, Nucl. Phys. B 604 (2001) 121147, hep-th/0101102.
[25] A.P. Balachandran, S. Kurkcuoglu, E. Rojas, JHEP 0207 (2002) 056, hep-th/0204170.
[26] D. OConnor, J. Medina, in preparation.
[27] Y. Kitazawa, Nucl. Phys. B 642 (2002) 210226, hep-th/0207115.

Lepton electric dipole moments from heavy states

Yukawa couplings
Isabella Masina
Service de Physique Thorique 1 , CEA-Saclay, F-91191 Gif-sur-Yvette, France
Received 19 May 2003; accepted 13 August 2003
Abstract
In supersymmetric theories the radiative corrections due to heavy states could leave their footprints
in the flavour structure of the supersymmetry breaking masses. We investigate whether present
and future searches for the muon and electron EDMs could be sensitive to the CP violation and
flavour misalignment induced on slepton masses by the radiative corrections due to the righthanded neutrinos of the seesaw model and to the heavy Higgs triplets of SU(5) GUT. When this
is the case, limits on the relevant combination of neutrino Yukawa couplings are obtained. Explicit
analytical expressions are provided which accounts for the dependencies on the supersymmetric mass
parameters.
PACS: 13.40.Em; 12.60.Jv; 14.60.St; 12.10.Dm
1. Introduction
In low energy supersymmetric extensions of the Standard Model (SM), unless sparticle
masses are considerably increased, present limits on flavour violating (FV) decays and
electric dipole moments (EDMs) respectively allow for a quite small amount of fermion
sfermion misalignment in the flavour basis and constrain the phases in the diagonal
elements of sfermion masses, involving the parameters and A, to be rather small. The
bounds on the supersymmetric contribution to lepton (L)FV decays and EDMs have the
advantage, as compared to the corresponding squark sector ones, of being not biased
E-mail address: masina@roma1.infn.it (I. Masina).
1 Laboratoire de la Direction des Sciences de la Matire du Commissariat lnergie Atomique et Unit de
Recherche Associe au CNRS (URA 2306).

doi:10.1016/j.nuclphysb.2003.08.018
I. Masina / Nuclear Physics B 671 (2003) 432458
433
by the SM contributionnor by the non-supersymmetric seesaw [1] contribution [2].

Experimental limits on LFV decays and EDMs are then a direct probe of the flavour and
CP pattern of slepton massessee, e.g., Ref. [3] for a recent collection. From a theoretical
perspective, understanding why CP phases and deviations from alignment are so strongly
suppressed is one of the major problems of low energy supersymmetry, the CP and flavour
problem. On the other hand, precisely because FV decays and EDMs provide strong
constraints, they can share some light on the features of the (possibly) supersymmetric
extension of the SM.
Indeed, it is well known that even if these CP phases and misalignments were absent
or suppressed enoughfrom the effective broken supersymmetric theory defined at MPl ,
they would be generated at low energy by the RGE corrections due to other flavour and CP
violating sources already present in the theory, in particular from the Yukawa couplings
of heavy states like the right-handed neutrinos of the seesaw model [4] and the Higgs
triplets of SU(5) grand unified theories (GUT) [5]. Effects of radiative origin have the
nice features of being naturally small, exactly calculable once specified a certain theory
andmost interestinglyif experiments are sensitive to them, they yield limits on some
combination of Yukawa couplings.2
Recently, it has been stressed that the present (planned) limit on e [6] (
[7]) is sensitive to the leptonslepton misalignments induced by the radiative corrections
in the framework of the supersymmetric seesaw model and that the associated constraints
on neutrino Yukawa couplings have a remarkable feedback on neutrino mass model
building [8,9]. However, LFV decays cannot provide any information on the pattern of
CP violation of slepton masses, while EDMs are sensitive to both LF and CP violations.
Since the present sensitivities to the electron and muon EDMs, de < 1027 e cm [10] and
d < 1018 e cm [11], could be lowered by planned experiments by up to three to five
[12,13] and six to eight [14,15] orders of magnitude respectively, it is natural to wonder
whether lepton EDMs would explore the range associated to LF and CP violations of
radiative origin.
In this work we analyze the predicted range for de and d when the radiative corrections
due to the right-handed neutrinos of the seesaw model and, possibly, the Higgs triplets of
SU(5) GUT provide the main source of CP violation and misalignment in slepton masses.
This allows to extract many informations. If experimental searches could explore this
range, one could obtain limits on the imaginary part of the relevant combination of neutrino
Yukawa couplings. Moreover, the eventual discovery of a lepton EDM in this range might
be interpreted as indirectly suggesting the existence of such a kind of fundamental particles
which are too heavy to be more manifest. On the other hand, finding de and/or d above this
predicted range would prove the existence of a source of CP (and likely also LF) violation
other than these heavy states.
The pure seesaw case has been considered in Ref. [16]see also the related studies
[17,18] on specific seesaw textureswhere it has been pointed out that threshold effects
due to hierarchical right-handed neutrino masses enhance the radiatively-induced Im(Aii ),
2 Of course, several effects of different origin could be simultaneously present, but there would be no reason
for a destructive interference among them to occur.
434
i = e, , . By extensively reappraising this framework, we find that for tan 10 the

amplitude with a LLRR double insertionproportional to tan3 dominates over the
one involving Im(Aii )insensitive3 to tan the exact ratio depending on the particular
choice of supersymmetric mass parameters. We provide general expressions from which
it is easy to recognize the model dependencies and which complete previous analyses.
Lepton EDMs are then strongly enhanced in models with large tan . To see how close
experiments are getting to the seesaw induced lepton EDMs range, we compare the upper
estimates for the radiatively-induced misalignments and CP phases with the corresponding
present and planned experimental limits collected in [3].
The range for a seesaw-induced d turns out to be quite far from present searches and, in
particular, its eventual discovery above 1023 e cm would prove the existence of a source
of CP violation other than the neutrino Yukawa couplings. On the contrary, the present
sensitivity to de explores the seesaw-induced range in models with large tan and small
R-slepton masses, say mR around 100200 GeV. The planned improvements for de would
allow to test also models with mR up to the TeV region and moderate tan . When present
or planned experimental limits turns out to overlap with these allowed ranges, bounds on
the imaginary part of the relevant combinations of neutrino Yukawa couplings are obtained
and plotted in the plane (M 1 , mR ), respectively the bino and the average R-slepton masses
at low energy. These plots allow to check whether any particular seesaw model is consistent
with present data and, if so, which level of experimental sensitivity would test it. In any
case, an experimentally interesting contribution requires hierarchical right-handed neutrino
masses.
This feature is no more necessary when, in addition to the seesaw, a stage of SU(5)
grand unification is present above the gauge coupling unification scale. It is well known
that the main drawback of minimal SU(5) is that the value of the triplet mass, MT ,
required by gauge coupling unification is sizeably below the lower bound on MT derived
from proton lifetime (see for instance [19] for recent reviews and references). In our
analysis we therefore keep the triplet mass as a free parameter. We nevertheless exploit the
minimal SU(5) relations between the doublet and triplet Yukawa couplings since in general
they are mildly broken in non-minimal versions of SU(5). The simultaneous presence
of right-handed neutrinos and heavy triplets turns out to further enhance the amplitude
with the LLRR double insertion over the one with Im(Aii ), essentially unaffected by
triplets. This cannot be derived by theotherwise eleganttechnique based on the allowed
invariants [20].
When right-handed neutrinos and Higgs triplets are simultaneously present, the
predicted range for the radiatively-induced de is already sizeably excluded by the present
experimental limit. This in turn is translated into a strong constraint on the imaginary part
of the combination of Yukawa couplings which is relevant for the LLRR amplitude. Such
a constraint could be hardly evaded even for large values of the triplet mass and unfavorable
supersymmetric mass parameters. The radiatively-induced d in the presence of triplets
should not exceed 1023 e cm, as was the case for the pure seesaw. However, at difference
of the latter case, in the former one d only mildly depends on the spectrum of right3 In Ref. [17] a dependence on tan arises due to the particular class of seesaw textures studied.
435
handed neutrinos. Notice that planned searches for d could constrain (depending on the
triplet mass mass, of course) the imaginary part of the combination of Yukawa couplings
whose absolute value could be independently constrained by the LFV decay .
The paper is organized as follows. In Section 2 we introduce our notations, discuss
the framework and make some preliminary considerations. Section 3 considers the pure
seesaw case by separately analyzing the flavour conserving (FC) and flavour violating (FV)
amplitudes contributing to lepton EDMs. In Section 4 the framework of seesaw and SU(5)
is discussed along the same lines. Concluding remarks are drawn in Section 5. Finally, in
Appendices A and B we collect the RGE in the case of the seesaw, without and with a
minimal SU(5) unification respectively.
2. Framework and method

In this section we recall the expression for the lepton EDMs in the mass insertion
approximation [3,2123,25] to display the supersymmetric mass parameters that are
constrained by the present and projected searches for de and d . We then draw some
preliminary considerations to introduce our procedure to calculate the radiative 1-loop
contribution to these mass parameters. The relevant RGE can be found in Appendices A
and B.
We adopt here the following conventions for the 6 6 slepton mass matrix in the lepton
flavour (LF) basis where the charged lepton mass matrix, m , is diagonal:

Ae vd tan m
m2LL
L
L R
(1)
,
R
Ae vd tan m
m2
RR
where Ae is the 3 3 matrix of the trilinear coupling, the A-term. All deviations from
alignment in this mass matrix are gathered in the matrices, which contain 30 real
parameters (including 12 phases) and are defined as:

m2RR = m2R I + RR ,
m2LL = m2L I + LL ,

Ae vd tan m = A vd tan m + mL mR LR ,
(2)
where mL , mR are average masses for L and R sleptons respectively and A
O(msusym /vd ) are the diagonal elements of Ae , so that LR has only non-diagonal,
flavour violating, elements.
The supersymmetric contributions to di (i = e, , ), can be splitted in two parts,
involving respectively only flavour conserving (FC) or flavour violating (FV) elements
of the slepton mass matrix (1):
di = diFC + diFV ,
M1
e
diFC =
2 4||2 cos2 W

1
mi Im() tan IB + IL IR + I2 vd Im Aii IB ,
2
(3)
(4)
436
diFV =
M1
e
2 4||2 cos2 W

mR mL Im LL LR ii IB,L
+ Im LR RR ii IB,R

+ tan Im LL m RR ii IB
+ Im LR m LR ii IB
(5)
where I Avd /(m tan ), the functions I are defined as in [3] and terms that
are less relevant4 or higher order in the s matrix elements are omitted. Notice that
I for relatively large values of tan , favored in mSUGRA and for which the LF
and CP violations are most likely to be detected. The FC and FV contributions could
result from different seeds of CP violation but could also be correlated in many different
ways in models. Anyway, the experimental limit can be put on both because, due to the
different nature of the many parameters involved, an eventual cancellation between these
contributions appears unnatural. The distinction between the FC and FV contributions is
also phenomenologically relevant: some of the ||s in the FV terms are already constrained
to be smaller than O(1) by LFV decays, while Im() and Im(Aii ) are directly constrained
by the EDMs.
Our aim here is to estimate the radiative contribution to the lepton EDMs induced by
the seesaw interactions, first alone and subsequently accompanied by a stage of SU(5)
GUT. Since the present experimental bounds on LFV decays and EDMs already point
towards family blind soft terms (i.e., sparticles with the same quantum numbers must have
the same soft terms) with very small CP phases, at the cut-off = MPl corresponding to
the decoupling of gravitational interactions we assume real and flavour blind soft terms,
namely in Eq. (1),
m2LL = m2RR = m20 I,
Ae = ye a0 ,
(6)
with real a0 , m0 and term. In the next sections we separately study the FC and FV
contributions to di and obtain explicit approximate expressions for Im(Aii ) and the
products of s in (5). The effects of a more general family independence assumption are
important but not crucial and can be easily included in our analysis. By means of these
general approximated expressions we will:
(i) derive the upper prediction for the radiatively induced leptonic EDMs, stressing the
model dependences;
(ii) compare it with the experimental limits;
(iii) when allowed by the experiment, obtain an upper bound on the imaginary part of the
relevant combination of Yukawa couplings.
A couple of preliminary considerations are in order before presenting the results of the
next sections.
4 E.g., a contribution to d FC of the form of (5), with LR (A v tan m )/(m m ), is in principle

L R
i
d
present, but it is negligible with respect to (4).
437
2.1. On the naive scaling relation

In the limit that all slepton masses are family independent, the FV contribution vanishes
and the FC one is proportional to the mass of the ith lepton (Im(Aii )vd Im(a0 )mi )
leading, except an accidental cancellation5 with the -term amplitude, to the naive
scaling relation
di /dj = mi /mj .
(7)
Then, due to the present experimental limit on de , d could not exceed de m /me
2 1025 e cm, which roughly corresponds to the planned sensitivity and, if the limit on de
were still to be lowered, next generation experiments would have no chance of measuring
d .
Such considerations could provide interesting informations because (7) strictly apply
only to the FC -contribution to the EDM, while in general both Im(Aii ) and the FV
terms may strongly violate it. This is the case for the radiatively induced Im(Aii ). Some
of the FV contributions are instead naturally proportional to a different lepton mass, mk ,
possibly heavier than mi , as discussed in [20,25]and, before, for the quark sector, in [26].
In particular, it will turn out in the next sections that the FV contribution can even take over
the FC one. Hence, a value of d above 2 1025 e cm is a possibility that deserves
experimental tests and, interestingly enough, it would imply the source of lepton EDM
being either the FV contribution or a non-universal Im(Aii ), so providing a remarkable
hint for our understanding of CP violation.
2.2. On the relevant combination of Yukawa couplings
Once a theory is specified, it is relatively easy to list the combinations of Yukawa
couplings appearing in the slepton mass radiative corrections that may contribute to lepton
EDMs. However, the actual calculation is more involved as we now turn to discuss.
As an example, let us look for Im(Aii ) by studying the evolution of Ae in the case
for simplicity. Let us first consider the
of degenerate right-handed neutrino masses, M,
6
case of the pure seesaw and define the hermitian matrices E ye ye , N y y . Notice
that E is real and diagonal in the LF defining basis and N is diagonalized by a unitary
matrix similar to the CKM one with only one phase (even for non-degenerate right-handed
neutrinos). By solving the RGE for Ae , Eq. (A.5), linearly in t3 1/(4)2 ln(MPl /M),
Ae can only be proportional to ye E and ye N , whose diagonal elements are real. At O(t32 )
only ye EE, ye NN , ye NE and ye EN appear, whose diagonal elements are again real.
A potential Im(Aii ) shows up at O(t34 ), through Im(ye N[N, E]N)ii . On the other hand,
when also Yukawa interaction of the SU(5) triplets are present, Eq. (B.7) shows that, at
first order in tT 1/(4)2 ln(MPl /MT ), Ae can be proportional to ye E, ye N but also to
U ye , where U yu yu . The latter have real diagonal elements but allow at O(tT2 ) for a
combination with diagonal imaginary part, namely (U ye N)ii .
5 For dedicated studies on cancellations between amplitudes, also in more general frameworks, see, e.g., [23,
24].
6 Here and in the following Dirac mass terms are always written as f m f .
R f L
438
Accordingly, in order to evaluate the actual coefficients in front of the products of

Yukawa coupling matrices which are likely to have phases, we solve the RGE for the
soft parameters by a Taylor expansion in the small parameters tif = 1/(4)2 ln(Qi /Qf )
associated to the intervals between the successive decoupling thresholds of the various
heavy states. For instance, by integrating the RGE for Ae in the case of SU(5) plus seesaw,
a non-vanishing coefficient is obtained for the combination Im(U ye N)ii at O(tT2 ). Now,
lepton FV transitions and CP phases are naturally defined in the LF basis where ye and
the Majorana masses MR are diagonal and real. This basis is not invariant under the RGE
evolution and one should diagonalize ye and MR again at the lower scale. Therefore, one
has to find out the effect of these final rotations.
We adopt the rotating basis method introduced in Ref. [26], where the RGE are modified
to incorporate the fact that the matrices are defined in the LF basis at each scale, so that
ye and MR are always diagonal. It is worth to stress that, even if the rotations needed to
diagonalize ye are small, their effect could be crucial for Im(Aii ) and the four products
of s in Eq. (5). For instance, in the case of SU(5) plus seesaw it turns out that this
correction exactly cancels the term proportional to tT2 Im(U ye N)ii in Ae . On the contrary,
these rotations can be safely neglected when deriving approximate expressions for the
radiatively-induced LFV decays because the latter are only sensitive to absolute values of
s.
3. Lepton EDMs and seesaw

In this section we consider the predictions for lepton EDMs in the context of
the supersymmetric extension of the seesaw model, namely, we assume the MSSM
supplemented with the seesaw Yukawa interactions and Majorana masses for the righthanded neutrinos as the effective theory valid up to the cut-off = MPl where gravitational
interactions decouple. Starting with real and universal boundary conditions at MPl , we
solve the RGE displayed in Appendix A by expanding in the small parameters defined by
the right-handed neutrino thresholds
t3 =
ln
,
2
(4)
M3
t2 =
1
M3
ln
,
2
(4)
M2
t1 =
1
M2
ln
,
2
(4)
M1
(8)
with the ordering M3 > M2 > M1 . We thus obtain approximate analytic expressions for the
radiatively induced s and Im(Aii ) which depend on m0 , a0 and the Yukawa couplings
defined in the LF basis at the scale . Of course the latter can be immediately translated
into the corresponding ones defined at any scale, e.g., M1 or msusy .
Precisely because of the lepton flavour and CP violating Yukawa couplings, the LF basis
is continuously rotated and rephased with the RGE evolution and, as already discussed,
we handle this by working in a rotating basis. It turns out that the basis transformation
introduces negligible corrections in the FC terms (4) but important ones in the FV
terms (5). Notice also that the seesaw effects stop at the decoupling threshold of the
lightest right-handed neutrino, M1 , and that the RGE evolution of these effects down to
the supersymmetric scales where CP and lepton FV transitions are estimated is generically
439
small and can be neglect for our estimates. We now study separately the FC and the FV
contributions.
3.1. Flavour conserving contribution
Starting in the LF basis at MPl from Ae = ye a0 , the seesaw interactions generate a
Im(Aii ) in the LF basis at M1 . At leading order in the ts and defining y Pa y Na (a =
1, 2, 3) and the right-handed neutrinos projectors P1 = diag(1, 0, 0), P2 = diag(1, 1, 0),
P3 = I, the latter reads:7

Im(Aii ) = 8a0 yei t2 t3 Im(N2 N3 )ii + t1 t3 Im(N1 N3 )ii + t1 t2 Im(N1 N2 )ii ,
(9)
where the Yukawas in the r.h.s. are evaluated at . A similar8 formula were previously
presented in Refs. [1618], but with the various Yukawas involved defined at different
scales. Then, one must be more cautious in drawing general conclusions and the authors
validate theirs with specific numerical examples. Our approach offers the advantage that
the corresponding results become more transparent and allow for an easier estimate of the
effects once a pattern of seesaw parameters is assigned. Anyway, the crucial point [16]
is that the more right-handed neutrinos are hierarchical, the more the FC contribution
increases. Indeed, it vanishes in the limit that right-handed neutrinos are degenerate,
in which case a contribution to Im(Aii ) only appears at fourth order, proportional to
Im(ye N[N, E]N)ii . Notice also that the naive scaling relation is generally violated [16]
by (9).
Eq. (9) displays a linear dependence on the unknown parameter a0 . By defining
Aii |Aii |ei8Ai , then Ai t2 t3 Im(N2 N3 )ii + t1 t3 Im(N1 N3 )ii + t1 t2 Im(N1 N2 )ii
is completely specified by the seesaw parameters. However, as appears from Eq. (4)
the experimental limit on di does not probe directly Ai , rather it gives a bound on
Im(Aii )vd mi Im(ai ) once supersymmetric masses are fixed (and up to unnatural
conspiracies between the various amplitudes). Fig. 1, taken from Ref. [3], show the
present upper limits on | Im(ae )|/mR and the planned ones for | Im(a )|/mR in the plane
(M 1 , mR ), respectively the bino and R-slepton mass at msusy . Notice that these limits are
quite model independent because, apart from mR and M 1 , there is only a mild dependence
on mL , which we have fixed as in mSUGRA for definiteness. Indeed, since the A-term
amplitude in diFC arises from pure bino exchange, it does not involve9 nor tan . In
mSUGRA there is an unphysical region in the plane (M 1 , mR ) corresponding to m20 < 0
and which has been indicated in light grey in the plots. Anyway, the dark grey region and
below is also excluded because mR M 1 , in contrast with the requirement of neutrality
for the LSP.
7 Actually, (9) is an approximation by excess and improves as the involved Yukawa couplings become small.
However, for our estimates it is reliable up to O(1) Yukawa couplings.
8 As far as the comparison with the expressions in Refs. [1618] is possible, we find agreement with the last
papers up to the coefficient of Im(N1 N2 )ii .
9 The function I in Eq. (4) contains a factor ||2 that cancels the one in the overall coefficient. The exact
B
expressions for IB can be found in [3], as well as approximate ones suitable for various pattern of supersymmetric
masses.
440
Fig. 1. Experimental upper bound on | Im(ae )|/mR and | Im(a )|/mR for de < 1027 e cm and
d < 1024 e cm, respectively, [3]. r 1 corresponds to the slope: r M 1 /mR . For the present sensitivity
d < 1018 e cm, the numbers have to be multiplied by 106 . mL is fixed as in mSUGRA.
To find an upper estimate for the seesaw induced diFC , let us evaluate | Im(ai )|/mR
from Eq. (9) by considering only its first term. This situation is representative because
the terms proportional to t1 are negligible when the lightest right-handed neutrino has
smaller Yukawa couplings, as happens in many models and as one would guess from
similarity with the charged fermion sectors. Anyway, the following discussion is trivially
adapted to other cases. For definiteness, we adopt this set of reference threshold values:
M2 = 1012 GeV, M3 = 1015 GeV, = 2 1018 GeV. Since, as demanded by perturbativity,
(N2 N3 )ii O(1),

a0
t2 t3
Im(ai )
O 0.02
.
(10)
3
mR
mR
2 10
This upper estimate is easily adapted to any given model once the relation between
a0 and mR is made explicit. To make the dependence more manifest, let us introduce
2 , where M
1/2 is
the following two mSUGRA situations: (a) a02 = 2m20 ; (b) a02 = M 1/2
the universal gaugino mass at the gauge coupling unification scale. In the first case
(a0 /mR )2 2(1 0.9r 2 ), with

r M1 /mR . The ratio a0 /mR thus displays a mild
excursion as it decreases from 2 down to 0.45 when moving from the vertical left axes,
where r 1, to the joining line of the dark grey region, where r = 1. For case (b),
the situation is opposite and not so mild: (a0 /mR ) 2.5r. In the more natural cases in
between, the ratio a0 /mR should thus be rather stable.
Let us now compare the upper estimate (10) with the experimental bound. As appears
from Eq. (4), the latter improves linearly with the experimental sensitivity to di . Taking
a0 mR , Fig. 1 shows that values around 0.02 for | Im(ai )|/mR would require an
experimental sensitivity to de and d , respectively at the level of 1028 1029 e cm and
441
2 1026 2 1027 e cm, the exact value depending on the particular point of the
(M 1 , mR ) plane. However, to avoid charge and color breaking, the constraint a0 /mR 3
in general applies. Thus, focusing for instance around mR 500 GeV, the FC contribution
cannot exceed 1028 e cm for de and 2 1026 e cm for d , even with highly
hierarchical right-handed neutrinos. Notice that the former value could be at hand of
future experimental searches for de , while the latter is at the very limit of the planned
experimental sensitivity to d . Allowing for smaller mR values, mR 200 GeV, the FC
seesaw upper estimate comes close to the present limit for de while it cannot be more than
2 1025 e cm for d .
The result for d is thus a kind of negative one, since its eventual future discovery
above 2 1025 e cm could not be attributed to the radiative FC contribution of the
seesaw. Even if the situation for deFC appears more optimistic, it is worth to underline that
the previous upper estimate actually applies to models with at least four O(1) neutrino
Yukawa couplings and large CP phases, like those in Ref. [18]. However, as we now turn
to discuss, the FV contribution, underestimated by previous analyses, could drastically
enhance the di upper estimate.
3.2. Flavour violating contributions
In the following we isolate the potentially most important products of s and give
an estimate of their relative magnitude with respect to the FC amplitude. The relevant
approximations for the flavour violating elements, i = j , of the s are:

t2

a
FA (a, a)ij +
mL mR ijLR = a0 mi 2t3 Nij +
(11)
ta tb FA (a, b)ij ,
2
a
a>b
m2R ijRR =
t2
a
FR (a, a)ij +
ta tb FR (a, b)ij ,
(12)
a>b
t2

a
FL (a, a)ij +
ta tb FL (a, b)ij ,
m2L ijLL = 6m20 + 2a02 t3 Nij +
2
a
(13)
a>b
where a, b = 1, 2, 3, the matrix N is defined as t3 N t3 N3 + t2 N2 + t1 N1 and
(b)
FA (a, b) = 15{E, Na } 5{E, Nb } + 12{Na , Nb } + 4 Nb , uE

(a)

D
(a)
+ 2 (Nb (d) + De )Na + (Na (d) + De )Nb + 4
+ D Nb
a0
+ [Na , E] + 4[Nb , Na ] + 7[E, Nb ],
(14)
2
2

2
FR (a, b) = 8 6m0 + 5a0 ye Na ye 6m0 + 2a0 ye Nb ye ,

(15)
2

(b)
2
(a)
2
FL (a, b) = 2 6m0 + 2a0 {3Nb + E, Na } + 2D Nb + Na , uE + 2mHu Nb

+ 4a02 {3Nb + E, Na } + {E, Nb } + 2 GL + 4a0 D (a) Nb ,
(16)
with the Yukawas in the r.h.s. evaluated at the scale . The subscript (d) indicates to take
(a)
only the diagonal elements of the matrix, u(a)
E is defined through [uE , E] = {E, 3E +
Na + De }, and the definition of all the other quantities can be found in Appendix A.
442
By means of the above expressions, the predictions for the imaginary part of the
various product of s can be studied. An imaginary part in the products ( LL LR )ii ,
( LR RR )ii , . . . , only arises at third order in at least two different ts.
3.2.1. The LLRR contribution
Let us firstly discuss the LLRR double insertion, Im( LL m RR )ii , which turns
out to be the quantitatively most interesting one. Since in general I and the phase
of if anyis extremely small so that it can be safely neglected in the discussion,
Im( LL m RR )ii Im( LL m RR )ii . From Eqs. (12), (13) it appears that ijRR ,
ijLL are respectively of second and first order in the ts. Thus, the lowest order is the cubic
and the LLRR contribution reads:

(6m20 + 2a02 )(6m20 + 7/2a02)
Im LL m RR ii = 8mi
m2L m2R

ta tb (ta + tb ) Im(Na ENb )ii .
(17)
a>b
Notice that Im(Na ENb )ii = tan2 Im(Na m2 Nb )ii /m2t , where mt is the top mass. Due to
the hierarchy in m , a potentially important effect could only come from Im(Na i3 Nb3i ).
Notice also that, on the contrary of the FC contribution, (17) does not strongly depend
on a0 .
The relative importance between this amplitude and the FC one can be easily
appreciated by neglecting the terms proportional to t1 , which, as already mentioned, is
justified when the lightest right-handed neutrino has the smallest Yukawa couplings:
diFVLLRR
diFC
Im(N3 m2 N2 )ii IB
tan3 (6m20 + 2a02 )(6m20 + 7/2a02)

(t
+
t
)
. (18)
2
3
a0
m2R m2L
m2t Im(N3 N2 )ii IB
For realistic values, the ratio of the two loop functions is slightly smaller than one. To
obtain a rule of thumb, let us take m0 a0 mL,R and the reference ratio /M2 =
2 106 . Then, unless ad hoc fine-tunings in the structure of N , Im(N3 m2 N2 )ii
m2 Im(N3 N2 )ii , and one finds
diFVLLRR
diFC
0.5
tan3 t2 + t3
103 ew 0.1
(rule of thumb),
(19)
where |ew |2 0.5m2R + 20M 12 is the value of accounting for radiative electroweak
breaking in mSUGRA. Despite being of third order in the ts, the FV amplitude can
take over the FC one thanks to its tan3 dependence. The precise value of this ratio is
displayed in Fig. 2 in the plane (M 1 , mR ). Cases (a) and (b) are separately displayed so
that the behavior for any situation in between can be easily extrapolated. The relevant
generalizations are also reminded. The plots show that the rule of thumb is quite reliable
and allow to extract, for each point of the plane, the value of tan for which the FV
amplitude takes over the FC one. In both cases (a) and (b), this happens for tan 10
the only exception being the region with r 1 for case (a), where tan 20 is required.
443
FV
Fig. 2. Ratio di LLRR /diFC when Im(N3 m2 N2 )ii m2 Im(N3 N2 )ii . We have assumed as reference:
/M2 = 2 106 , mL as in mSUGRA, = ew and tan = 10.
3.2.2. The other contributions

The amplitudes with the other products of s are less important than the LLRR one.
Consider for instance Im( LR RR )ii . The flavour violating elements in RR and LR are
respectively of second and first order, so that the product appears at third order:

mi a0 (12m20 + 7a02)
ta tb (ta + tb ) Im(Na ENb )ii .
Im LR RR ii = 8
m3R mL
a>b
(20)
It vanishes in the limit a0 0, as the FC contribution. Neglecting the terms proportional

to t1 , the ratio of the LRRR amplitude and the FC one reads
diFVLRRR
diFC
= tan
2
12m20 + 7a02
m2R
Im(N3 m2 N2 )ii IB(R)

(t2 + t3 ) 2
.
mt Im(N3 N2 )ii IB
(21)
It is easy to check that this ratio is smaller than one (unless ad hoc fine-tunings in
the structure of N ): the ratio of the two loop functions is slightly smaller than one for
realistic values of the supersymmetric parameters and, taking for instance m0 = a0 and
/M2 = 2 106 , (21) should not exceed 104 tan2 .
For Im( LL LR )ii , no imaginary part can arise at second order in ts because both LL
and LR are proportional to t3 N . At third order there are many contributions and it is
lengthy but straightforward to check that they are proportional to at least two different ts.
This contribution is also proportional to a0 and could be comparable to the FC one but is
in any case smaller than the LLRR one. The expression for the double LR insertion is
also quite involved. It is proportional to a02 and, being also suppressed by a factor m2 /m2L
with respect to the LLRR one, it can be safely neglected.
444
3.3. Predicted range and constrains on Yukawas

For values of tan 10, for which the EDMs are enhanced and thus most likely to be
observed, the FV amplitude with the LLRR double insertion is generically dominant
with respect to all other amplitudes. Then, when tan 10, all the considerations
made previously for diFC actually apply to di when strengthened by the rule of thumb
factor (19). To give an example, if t2 + t3 0.1 and ew , de cannot exceed
0.5 1028(27) tan3 /103 e cm and d 1026(25) tan3 /103 e cm when mR
500(200) GeV. Planned experimental sensitivities to d could then test the seesaw radiative
contribution for models with small mR and/or large tan , in which case the factor (19)
could be up to 5060 so that d should not exceed O(1023) e cm. On the contrary, the
range of the seesaw induced de already overlaps with the present experimental limit for
values of mR up to 1/2 TeV when tan 30.
Barring unnatural cancellations, planned searches for de could thus test each term of the
sum in (17), namely each10

m2
ta tb (ta + tb ) Im Na 2 Nb
(a > b).
(22)
m
11
The effect of an eventual two orders of magnitude improvement for de on the upper limit
on Im(Na m2 Nb )11 /m2 is displayed in Fig. 3 by taking as reference values tan = 30
and ta tb (ta + tb ) = 2 104 . It turns out that planned limits could be severe enough to
Fig. 3. Upper bound on Im(Na m2 Nb )11 /m2 . Solid dashed lines refer to cases (a) and (b).
10 Needless to say, this remains true even if the FC were the dominant amplitude, in which case the bound
would be stronger.
445
test models with hierarchical neutrino Yukawa couplings. This cannot be done by present
limits. For any given seesaw model, it is straightforward to extrapolate from the plot
the level of experimental sensitivity required to test it. Notice also that for d , limits on
Im(Na m2 Nb )22 /m2 as strong as the present ones on Im(Na m2 Nb )11 /m2 would require
a sensitivity to d at the level of 2 1025 e cm.
4. Lepton EDMs, seesaw and SU(5) triplets

We now add to the supersymmetric seesaw model a stage of a minimal SU(5) GUT
above the gauge couplings unification scale, MGUT 2 1016 GeV. Namely, we include
the contribution of the Higgs triplets Yukawa interactions to the RGE evolution of slepton
masses from MPl down to their threshold decoupling scale MT , which is very likely to be
bigger than M3 . Notations are defined in Appendix B.
It is not restrictive to work (at any scale) in the basis where ydT (= ye ) and MR are real
and diagonal and yu = V T du u V , where du are the (real and positive) eigenvalues of yu ,
V is the CKM matrix in the standard parameterization (more on this later) and u is a
diagonal SU(3) matrix. The RGE for the radiatively induced misalignments are written in
Eqs. (47)(50). At first order in t1,2,3 defined in Eq. (8) and
tT
1
MPl
ln
,
(4)2 MT
their solutions at the scale M1 reads:

m2R RR = 6m20 + 2a02 2t1 ye2 + 3tT U ,

m2L LL = 6m20 + 2a02 (3tT + t1 )ye2 + t3 N ,

mL mR RL = a0 6tT U m + 2m t3 N ,
(23)
(24)
where the matrix N is defined as t3 N t3 N3 + t2 N2 + t1 N1 . It is understood that all

the quantities in the r.h.s. of (24) are evaluated at . The small effects of the subsequent
evolution from M1 down to msusy can be neglected in the following discussion. We
explicitly write only the first order terms11 in the ts because, contrarily to the seesaw case,
they already produce a potential imaginary part for the FV contribution, Eq. (5). Instead,
for the diagonal part of Ae this is not the case, as we now turn to discuss.
4.1. Flavour conserving contribution
As anticipated in the simplified discussion of Section 2, a potential candidate for
Im(Aii ), proportional to Im(U ye N)ii , could show up at O(tT2 ). It is lengthy but
straightforward to see that such term is exactly canceled by the effect of rotating the basis.
For the same reason, in the general case with different thresholds M1,2,3 , MT , the overall
11 Of course, Eqs. (24) overestimate the misalignment. However, for our estimates here they are reliable up to
yt y3 1.
446
coefficient of Im(U ye N )ii is zero. Therefore, the second order contribution to Im(Aii )
is just

3
3
tT + t3 t2 Im(N2 N3 )ii +
tT + t3 t1 Im(N1 N3 )ii
Im(Aii ) = 8a0 yei
2
2

+ t2 t1 Im(N1 N2 )ii ,
(25)
namely, the pure seesaw one discussed in the previous section, Eq. (10), with the
substitution t3 (3/2tT + t3 ), due to the fact that above MT also the triplets circulate in
the loop renormalizing the wave functions. tT is naturally expected to be small (triplets will
not decouple much below 1016 GeV) so that eventual higher order contributions involving
tT are expected to be negligible with respect to (25). As a result, a stage of SU(5)-like
grand unification, cannot enhance by much the FC contribution with respect to the pure
seesaw case.
4.2. Flavour violating contribution
On the contrary, products of two s have an imaginary part proportional to t3 tT
Im(U ye N )ii and the FV contribution to di is potentially bigger than the FC one. Most
interestingly, the predicted range for the radiatively induced de turns out to have been
already sizeably excluded by the present experimental bounds. Planned searches for d
would also get close to test the range corresponding to radiatively induced misalignments.
Let us consider in turn the predictions for the imaginary part of the products of s,
Eq. (5), from their expressions given in Eq. (24). As before, the most important contribution
comes out from the LLRR double insertion. Since I and the phase of if anyis
experimentally small enough to be safely neglected in the present discussion, one has at
the lowest relevant order in the ts

(6m20 + 2a02 )2
Im N m U ii .
Im LL m RR ii = 3tT t3
2
2
mR mL
(26)
This FV contribution is potentially much bigger than the FC one because, apart from
Yukawas and numerical coefficients, it is enhanced by a factor (m tan )/(mi a0 ).
The other FV combinations in (5) are:

a 2 Im(N m3 U )ii
,
Im LR m LR ii = 12tT t3 02
mR
m2L
(27)

a0 (6m20 + 2a02 ) Im(N m U )ii
Im LR RR ii = 6tT t3
,
mL
m3R
(28)

m2
Im LL LR ii = R
Im LR RR ii .
2
mL
(29)
The only contribution which does not vanish in the limit a0 0 is the LLRR one. To
compare the amplitudes, (26), (27) and (28), (29) have to be multiplied respectively by
tan and mL mR and also by the appropriate loop functions, which have the same sign
447
and in general are of the same order of magnitude (for more details see [3]). Then, if > 0
the four FV amplitudes have the same sign. However, the double LR insertion is always
negligible with respect to the LLRR one because of the suppression factor m2 /m2L . For
the other amplitudes (28), (29) the suppression factor with respect to (26) is mR,L /( tan )
(actually smaller due to the numerical coefficients). Then, even in the case of < 0,
a reduction of the LLRR amplitude due to accidental cancellations seems unrealistic.
Of course, also the contributions due to different thresholds in the right-handed neutrino
spectrum are present, in exact analogy to what has been discussed in the previous section.
It is instructive to focus on the magnitude and dependencies of the combination
Im(N m U )ii . For de and d , neglecting subleading terms proportional to yc2 , yu2 and
defining Vt d |Vt d |eitd :

me
|Vt d | Im eitd N12 ,
Im N m U 22 m yt2 Vt s Im(N23 )
(30)
m

m

Vt s Im eitd N12 ,
Im N m U 11 m yt2 |Vt d | Im eitd N13 +
(31)
m
where we exploited the fact that Vt s is real in the standard parameterization. The latter
is convenient to stress that the CP phases involved in the above combinations could be
naturally largeas is indeed the case for t d but, of course, any other choice must give
equivalent results.12 The contribution proportional to Im(N12 ) has important suppression
factors. Moreover, |N12 | is independently constrained to be quite small from the present
limits on e . A plot of the present upper limit on |C12 |, with C (4)2 t3 N , in the
plane (M 1 , mR ) can be found in Ref. [9]. The limit were derived for the seesaw but also
applies without significant modifications to the case of SU(5) plus seesaw. As a result,
once fixed MT , experimental searches for de and d represent a test for Im(eitd N13 )
and Im(N23 ), respectively. Although present searches for ( e ) are not able
by now to interestingly constrain |C23 | (|C13 |), eventual experimental improvements would
have an impact on d (de ).13
Notice also that the naive scaling relation is violated according to
|Vt d | Im(eitd N13 )
de
=
d
Vt s
Im(N23 )
(32)
and that the combinations of Yukawas relevant for di are independent on the phases of the
diagonal SU(3) matrix, u . On the contrary, the latter affects (see, for instance, Ref. [27])
the proton decay lifetime due to d = 5 operators, whose most important decay mode in the
case of minimal SU(5) is p K + .

4.3. Predicted range and constraints on Yukawas
To understand how close to the experimental sensitivity is the radiatively induced EDM
range, we plot in Figs. 5 and 8 the upper estimate for the most important products of s. We
12 Had we exploited the freedom of parameterizing V in such a way that V are real numbers, then the CKM
ti
phase of V would have been hidden in the redefinition of N .
13 Work in progress.
448
Fig. 4. Experimental upper bounds [3] on various products of s corresponding to the present sensitivity
de < 1027 e cm.
Fig. 5. Upper estimate for various products of s in SU(5) with seesaw. The reference values /MT = 102 and
/M3 = 2 103 have been taken. Solid dashed lines correspond to cases (a) and (b).
consider a degenerate spectrum of right-handed neutrinos to pick out just the effect of the
triplets, substitute the MPl -values for yt ( 0.7) and the relevant CKM elements and choose
as reference values /MT = 102 and /M3 = 2 103 . Then, the upper estimate follows
by requiring perturbativity, Im(N23 ) 1, Im(eitd N13 ) 1. Solid and dashed lines refer
449
Fig. 6. Present upper bound on Im(eitd N13 ). The reference values tan = 10, /MT = 102 and
/M3 = 2 103 have been taken. Solid dashed lines correspond to cases (a) and (b).
to cases (a) and (b), respectively. We do not show case (b) for Im( LL m RR )ii /m ,
because the predicted value is essentially flat. The upper estimate for Im( LL LR )ii is
not shown, being closely related to that on Im( LR RR )ii (see Eq. (29)). The bounds
on Im( LR m LR )ii /m are not displayed because they are too small to be of potential
interest.
For an easy comparison, we have taken from the sleptonarium [3] the experimental
limits on the same quantities, Figs. 4 and 7. The experimental limits on Im( LR RR )ii
are close to those on Im( LL LR )ii and mildly depend on mL , which has been fixed
as in mSUGRA in the plots. On the contrary, the limits on Im( LL m RR )ii /m are
proportional to 1/( tan ). For definiteness, tan = 10 and = ew have been assumed.
The experimental bounds shown in Figs. 4 and 7 correspond to the present bound de <
1027 e cm and to the planned sensitivity d < 1024 e cm. Since the experimental bounds
are proportional to the bound on di , it is straightforward to extrapolate the sensitivity to d
and de required to test the radiatively induced de and d .
Let us firstly discuss the de range. For our reference values, the upper estimate for
Im( LR RR )11 is O(106 ) and the present bound on de already constrains it to be smaller
in a large region of the plane. Although we already know that they are not dominant, it is
worth to discuss the LRRR amplitude and the similar LLLR one because, as already
mentioned, they are quite model independent. Allowing for higher triplet masses, however,
these amplitudes shift below the present experimental sensitivity. This is not the case for the
LLRR amplitude. The maximum value allowed for Im( LL m RR )11 /m is displayed
in the right panel of Fig. 5 for case (a). In case (b) it is 103 everywhere. Then, the
450
Fig. 7. Same as Fig. 4 but for d < 1024 e cm.
Fig. 8. Same as Fig. 5 but for i = 2.
present experimental bound on de has already explored the radiative range for roughly 23
orders of magnitude.
This can be translated into an upper limit on Im(eitd N13 ), as in Fig. 6. Notice that
this limit is indeed very strong and can be hardly evaded even looking for less favorable
parameters than those taken as reference. To check this, keep, e.g., t3 fixed and try to
451
Fig. 9. The upper bound on Im(N23 ) which would be extracted by improving the present limit by many orders
of magnitude [14,15]. Solid dashed lines correspond to cases (a) and (b).
worsen the limit: a reduction of with respect to ew is unlikely to reduce it more than
one order of magnitude; a suppression by another factor 10 would require a /MT 1.6;
on the other hand, for values of tan larger than 10 the limit linearly improves. As a result,
in the framework of the seesaw accompanied by SU(5), the limit on Im(eitd N13 ) can
be considered robust.
Let us now turn to discuss d . By comparing Figs. 7 and 8, it turns out that the upper
bound on the LRRR insertion would require a sensitivity to d at O(1026) e cm, at
the very limit of planned experimental improvements. Quantitatively, the upper estimate
for Im( LL m RR )22 /m which, for case (b) is 4 103 , is more promising. The
major part of the plane in Fig. 8 could be tested with d at the level of 1024 1025 e cm
and in general d should not exceed O(1023) e cm. As a result, the eventual presence
of triplets does not enhance by much the range for d with respect to the pure seesaw
case. Nevertheless, here too the possibility of constraining Im(N23 ) can be envisaged,
as shown in Fig. 9. Notice that, due to (32), a limit on Im(N23 ) comparable to the
present one on Im(eitd N13 ) would require to improve the d sensitivity down to
5 1027 e cm.
5. Conclusions
Planned experiments might significantly strengthen the limit on de [12,13] and d
[14,15]. Their eventual discovery could be interpreted as an indirect manifestation of
supersymmetry but could not reveal which source of CP (and possibly flavour) violation is
452
actually responsible for the measured effect. Clearly, all sources in principle able to give
the lepton EDM even at an higher level would be automatically constrained while those
which fail in giving the lepton EDM at the desired level would be automatically excluded
from the list of possible candidates.
In this work, we have estimated the ranges for the lepton EDMs induced by the Yukawa
interactions of the heavy neutrinos, both alone and with the simultaneous presence of the
heavy SU(5) triplets. It turns out that the FV LLRR amplitude is in general larger or
comparable to the FC one. So, EDMs are enhanced for large values of tan and do not
strongly depend on a0 .
The pure seesaw, even with large tan and very hierarchical right-handed neutrinos,
cannot account for d above 1023 e cm. Its eventual discovery above this level would
then signal the presence of some source of CP and LF violation other than the neutrino
Yukawa couplings. The heavy triplets Yukawa couplings would be excluded from the
list of possible sources because their additional presence do not significantly enhance the
predicted range for d . Notice, however, that in the latter case a hierarchical right-handed
neutrino spectrum is no more essential to end up with d at an interesting level for planned
searches. From the theoretical point of view, finding d above 1023 e cm would indeed
have a remarkable impact.
Interestingly enough, the present experimental sensitivity to de is already testing
the simultaneous presence of triplets and right-handed neutrinos. Correspondingly,
constrains on Im(eitd N13 ) have been derived which are significant even for quite
large values of the triplet mass and unfavorable supersymmetric masses. Without the
triplets, the radiatively-induced de is close to the present experimental sensitivity only
in models with large tan and small slepton masses. Therefore, an experimental
improvement would eventually provide interesting limits on the imaginary part of
the relevant combination of neutrino Yukawa couplings and right-handed neutrino
masses.
In the present discussion we have been looking for results as general as possible. Indeed,
the specification of any particular seesaw model has been avoided and the attention has
rather focused on the dependencies on the supersymmetric masses and heavy thresholds.
Although some relevant seesaw models deserve a dedicated analysis,14 Figs. 3, 6 and 9
are suitable for a quick check of the status of any given seesaw model with respect to the
present and planned experimental limits on lepton EDMs.
Acknowledgements
We thank C.A. Savoy for useful discussions and collaboration in the early stage of this
work. I.M. acknowledge the CNRS and the A. Della Riccia Foundation for support and
the SPhT, CEA-Saclay, for kind hospitality.
14 See footnote 13.
453
Appendix A. Seesaw
In the basis where charged fermion and right-handed Majorana neutrino masses are
diagonal
1
W ucT yu QHu + d cT yd QHd + ecT ye LHd + cT y LHu + cT MR c ,
2
(A.1)
0
= vd(u) . Soft scalar masses are defined as
where Q = (ud)T , L = (e)T and Hd(u)
2
Lsoft u R m2u u R + dR m2d dR + Q m2Q Q + eR
me eR + L m2L L + R m2 R

+ u R Au u L vu + dR Ad dL vd + eR
Ae eL vd + R A L vu + h.c. .
(A.2)
Let us also introduce the following notations:

yx yx X,
Pa y y(a),
yx yx X
y Pa y
(x = e, u, d),
Na ,
Pa y y Pa N a
(a = 1, 2, 3),
where P2 , P1 project out M3 and M3,2 , respectively, P2 = diag(1, 1, 0), P1 = diag(1, 0, 0),
and P3 = I.
A.1. Running
Defining t
1
(4)2
ln Q, the running of the Yukawa coupling constants is governed by:
(a)

dy
= y(a) 3Na + E + D(a) ,
dt
dye
= ye [3E + Na + De ],
dt

dyu
= yu 3U + D + Du(a) ,
dt
dyd
= yd [3D + U + Dd ],
dt

d(Pa MR Pa )
= 2 Pa MR N aT + N a MR Pa ,
dt
(A.3)
where

3 2
2
= Tr(3U + Na ) 3g2 + g1 I,
5

9 2
2
De = Tr(3D + E) 3g2 + g1 I,
5

16 2
13 2
(a)
2
Du = Tr(3U + Na )
g + 3g2 + g1 I,
3 3
15

16 2
7 2
2
g + 3g2 + g1 I.
Dd = Tr(3D + E)
3 3
15
D(a)
(A.4)
454

(a)
For the trilinear couplings, defining Pa A A
dA(a)
(a)
(a)
(a)
(a) (a)
(a) (a)
= 4N a A(a)
+ 5A Na + 2y ye Ae + A E + D A + 2D y ,
dt
dAe
e + 5Ae E + 2ye y(a) A(a)
= 4EA
+ Ae Na + De Ae + 2De ye ,
dt
dAu
u + 5Au U + 2yu y Ad + Au D + Du(a) Au + 2D u(a) yu ,
= 4UA
d
dt
dAd
d + 5Ad D + 2yd yu Au + Ad U + Dd Ad + 2D d yd ,
(A.5)
= 4DA
dt
where

3 2
2
g
I,
+
3g
M
M
D (a) = Tr 3yu Au + y(a) A(a)
1
2 2
5 1

9
D e = Tr 3ydAd + ye Ae 3g22 M 2 + g12 M 1 I,
5

13 2
16 2
2
g
g
I,
+
3g
+
M
M
M
D u(a) = Tr 3yu Au + y(a) A(a)
3
1
2 2
3 3
15 1

16 2
7 2
Dd = Tr 3yd Ad + ye Ae
g M3 + 3g2 M2 + g1 M1 I.
3 3
15
(A.6)
2 (a)
For soft scalars, defining Pa m2 Pa m

dm2L 2
= mL , E + Na + 2 ye m2e ye + m2Hd E + Ae Ae
dt

(a)
+ 2 y(a) m2 (a)y(a) + m2Hu Na + A(a)
A + GL ,

dm2e
= 2 m2e , E + 4 ye m2L ye + m2Hd E + Ae Ae + Ge ,
dt

dm2 (a)
(a)
= 2 m2 (a), N a + 4 y(a)m2L y(a) + m2Hu N a + A(a)
,
A
dt
dm2Q 2

= mQ , U + D + 2 yu m2u yu + m2Hu U + Au Au
dt

+ 2 yd m2d yd + m2Hd D + Ad Ad + GQ ,

dm2u
= 2 m2u , U + 4 yu m2Q yu + m2Hu U + Au Au + Gu ,
dt

dm2d
= 2 m2d , D + 4 yd m2Q yd + m2Hd D + Ad Ad + Gd ,
dt
where
GL =

6 2 2
g1 M1 + 6g22 M 22 I,
5

Ge =

24 2 2
g1 M1 I,
5
(A.7)
455

2 2 2
32
g1 M1 + 6g22 M 22 + g32 M 32 I,
GQ =
15
3

32 2 2 32 2 2
8 2 2 32 2 2
g1 M1 + g3 M3 I,
g1 M1 + g3 M3 I.
Gd =
Gu =
15
3
15
3
(A.8)
Finally,
dm2Hd
dt
dm2Hu
dt

= 6 Tr yd m2Q yd + yd m2d yd + m2Hd D + Ad Ad

+ 2 Tr ye m2L ye + ye m2e ye + m2Hd E + Ae Ae + GL ,

= 6 Tr yu m2Q yu + yu m2u yu + m2Hu U + Au Au

(a)
+ 2 Tr y(a)m2L y(a) + y(a) m2 (a)y(a) + m2Hu Na + A(a)
A + GL ,
(A.9)
and we define m2 Hd(u) dm2Hd(u) /dt.

For energy scales between and M3 , one has to take a = 3; below M3 and above M2 ,
a = 2; while below M2 and above M1 , a = 1.
Appendix B. SU(5) + seesaw
We adopt the following notation to write matter and Higgs superfields:
c
0
uc3 uc2 u1 d 1
d1
c
c
2
2
c
0
u1
u
d
u3
d2
c
c
10 =
5 = d3c ,
0
u3 d 3
,
u2 u1
2 1
e
u u2 u3
0
ec
d 1
1 = ,
H3d 1
H3d 2
H = H3d 3 ,
H2d
H2d 0
d 2
d 3
ec
(B.1)
H3u 1
H3u 2
H = H3u 3 ,
H2u +
(B.2)
H2u 0
0
where H2d(2u)
vd(u) . The superpotential
1
W AB yu CD H E ?ABCDE + 2 AB ye A H B
4
1
+ 1 y A H A + 1 MR 1 ,
2
(B.3)
456
where A, B, . . . = 1, . . . , 5 and flavour indices are understood, gives rise to (A.1) with
yd = yeT and yu = yuT . For the soft breaking part of the Lagrangian

1
Lsoft = m2 + m2 + m2 + m2h h h + m2h h h +
M5 5 5 + h.c.
2

cT
cT
T
cT
cT
d + e Ae ev
u + d Ae dv
d + A vu + h.c.
+ u Au uv
(B.4)
ec ), = (dc , e, ) and = c and gauginos are
d,
where the scalar fields are = (u c , u,
denoted with 5 . In this way, in the scalar lepton mass matrix, m2LL = m2 , m2RR = m2
.
B.1. Running
Setting
t
1
ln Q,
(4)2
dg5
= 3g53 ,
dt
dM5
= 6g52 M5
dt
and
dye
= 6ye E + 3U ye + ye N + Ge ye ,
dt
dyu
u + 2yu D + Gu yu ,
= 6yu U + 2Ey
dt
dy
= 6y N + 4y E + G y ,
(B.5)
dt
where Ge = (84/5)g52 + 4 Tr(E), Gu = (96/5)g52 + Tr(3U + N), G = (48/5)g52 +
Tr(3U + N). For scalar masses
dm2
dt

2
2
= m2
, 2E + 3U + 4 ye m ye + mh E + Ae Ae

+ 6 yu m2 yu + m2h U + Au Au + G ,
dm2

2
= m2 , 4E + N + 8 ye m2
ye + mh D + Ae Ae
dt

2
+ 2 y m2
y + m h N + A A + G ,
dm2

2
2
= 5 m2
, N + 10 y m y + mh N + A A ,
dt
where G = (144/5)g52M52 I, G = (96/5)g52M52 I. For trilinear couplings
dAe
e + 3UA
e + 8Ae E + 6Au yu ye + Ae N
= 10EA
dt
e ye ,
+ 2ye y A + Ge Ae + 2G
dAu
u + 2EA
u + 9Au U + 4Ae ye yu + 2Au D
= 9UA
dt
u yu ,
+ 4yu yd Ad + Gu Au + 2G
dA
+ 11A N + 4A E + 8y ye Ae + G A + 2G
y ,
= 7NA
dt
(B.6)
(B.7)
(B.8)
(B.9)
457
e = (84/5)g 2 M5 + 4 Tr(ye Ae ), G
u = (96/5)g 2M5 + Tr(3yu Au + y A ),
where G
5
5
= (48/5)g 2M5 + Tr(3yuAu + y A ). Finally, for scalar higgses
G
5

dm2h
= Tr(6U + 2N )m2h + 6 Tr yu m2 yu + yu m2
yu + Au Au
dt

96
+ 2 Tr y m2 y + yT m2 y + A A g52 M52 ,
5
2
dmh
96 2 2
2 + 8 Tr ye m2 ye + ye m2
= 8 Tr(E)m
g M .
ye + Ae Ae
h
dt
5 5 5
(B.10)
References
[1] M. Gell-Mann, P. Ramond, R. Slansky, in: Supergravity, 1979;
T. Yanagida, in: Proceedings of the Workshop on Unified Theory and Baryon Asymmetry of the Universe,
1979;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[2] S.T. Petcov, Sov. J. Nucl. Phys. 25 (1977) 340, Yad. Fiz. 25 (1977) 641 (in Russian).
[3] I. Masina, C.A. Savoy, Nucl. Phys. B 661 (2003) 365, hep-ph/0211283.
[4] F. Borzumati, A. Masiero, Phys. Rev. Lett. 57 (1986) 961.
[5] R. Barbieri, L. Hall, Phys. Lett. B 338 (1994) 212, hep-ph/9408406;
R. Barbieri, L. Hall, A. Strumia, Nucl. Phys. B 445 (1995) 219, hep-ph/9501334;
R. Barbieri, A. Romanino, A. Strumia, Phys. Lett. B 369 (1996) 283, hep-ph/9511305;
A. Romanino, A. Strumia, Nucl. Phys. B 490 (1997) 3, hep-ph/9610485.
[6] R. Bolton, et al., Phys. Rev. D 38 (1988) 2077;
M.L. Brooks, et al., MEGA Collaboration, Phys. Rev. Lett. 83 (1999) 1521, hep-ex/9905013.
[7] I. Hinchliffe, F.E. Paige, Phys. Rev. D 63 (2001) 115006, hep-ph/0010086;
D.F. Carvalho, J.R. Ellis, M.E. Gomez, S. Lola, J.C. Romao, hep-ph/0206148;
J. Kalinowski, hep-ph/0207051.
[8] W. Buchmuller, D. Delepine, F. Vissani, Phys. Lett. B 459 (1999) 171, hep-ph/9904219;
J.L. Feng, Y. Nir, Y. Shadmi, Phys. Rev. D 61 (2000) 113005, hep-ph/9911370;
J. Ellis, M.E. Gomez, G.K. Leontaris, S. Lola, D.V. Nanopoulos, Eur. Phys. J. C 14 (2000) 319, hepph/9911459;
K.S. Babu, B. Dutta, R.N. Mohapatra, Phys. Lett. B 458 (1999) 93;
W. Buchmuller, D. Delepine, L.T. Handoko, Nucl. Phys. B 576 (2000) 445;
J. Sato, K. Tobe, T. Yanagida, Phys. Lett. B 498 (2001) 189, hep-ph/0010348;
J. Hisano, K. Tobe, Phys. Lett. B 510 (2001) 197, hep-ph/0102315;
J.A. Casas, A. Ibarra, Nucl. Phys. B 618 (2001) 171, hep-ph/0103065;
D.F. Carvalho, J. Ellis, M.E. Gomez, S. Lola, Phys. Lett. B 515 (2001) 323, hep-ph/0103256;
S. Davidson, A. Ibarra, JHEP 0109 (2001) 013, hep-ph/0104076;
T. Blazek, S.F. King, Phys. Lett. B 518 (2001) 109, hep-ph/0105005;
S. Lavignac, I. Masina, C.A. Savoy, Phys. Lett. B 520 (2001) 269, hep-ph/0106245;
E.O. Iltan, Phys. Rev. D 64 (2001) 115005, hep-ph/0107107;
S. Lavignac, I. Masina, C.A. Savoy, Nucl. Phys. B 633 (2002) 139, hep-ph/0202086;
A. Masiero, S.K. Vempati, O. Vives, Nucl. Phys. B 649 (2003) 189, hep-ph/0209303;
K.S. Babu, B. Dutta, R.N. Mohapatra, hep-ph/0211068;
S. Pascoli, S.T. Petcov, C.E. Yaguna, hep-ph/0301095;
G.C. Branco, D. Delepine, S. Khalil, hep-ph/0304164.
[9] I. Masina, in: Proceedings of SUSY 02, DESY, Hamburg, 2002, hep-ph/0210125.
458
[10] E.D. Commins, S.B. Ross, D. Demille, B.C. Regan, Phys. Rev. A 50 (1994) 2960;
B.C. Regan, et al., Phys. Rev. Lett. 88 (2002) 071805.
[11] CERNMainzDaresbury Collaboration, Nucl. Phys. B 150 (1979) 1.
[12] C. Chin, et al., Phys. Rev. A 63 (2001) 033401;
J.J. Hudson, et al., hep-ex/0202014;
B.E. Sauer, Talk at Charm, Beauty, CP, 1st Int. Workshop on Frontier Science, Frascati, Italy, October
611, 2002.
[13] S.K. Lamoreaux, nucl-ex/0109014.
[14] R. Carey et al., Letter of Intent to BNL (2000);
Y.K. Semertzidis, et al., hep-ph/0012087.
[15] J. Aysto, et al., hep-ph/0109217.
[16] J. Ellis, J. Hisano, S. Lola, M. Raidal, Nucl. Phys. B 621 (2002) 208, hep-ph/0109125.
[17] J. Ellis, J. Hisano, M. Raidal, Y. Shimizu, Phys. Lett. B 528 (2002) 86, hep-ph/0111324.
[18] J. Ellis, J. Hisano, M. Raidal, Y. Shimizu, Phys. Rev. D 66 (2002) 115013, hep-ph/0206110;
J. Ellis, M. Raidal, Nucl. Phys. B 643 (2002) 229, hep-ph/0206174.
[19] I. Masina, Int. J. Mod. Phys. A 16 (32) (2001) 5101, hep-ph/0107220;
J.C. Pati, hep-ph/0204240;
S. Raby, hep-ph/0211024.
[20] A. Romanino, A. Strumia, Nucl. Phys. B 622 (2002) 73, hep-ph/0108275;
See also: A. Romanino, A. Strumia, Nucl. Phys. B 490 (1997) 3, hep-ph/9610485;
O. Lebedev, Phys. Rev. D 67 (2003) 015013, hep-ph/0209023.
[21] L.J. Hall, V.A. Kostelecky, S. Raby, Nucl. Phys. B 267 (1986) 415;
F. Gabbiani, A. Masiero, Nucl. Phys. B 322 (1989) 235.
[22] T. Moroi, Phys. Rev. D 53 (1996) 6565, hep-ph/9512396;
T. Moroi, Phys. Rev. D 56 (1997) 4424, Erratum.
[23] S. Pokorski, J. Rosiek, C.A. Savoy, Nucl. Phys. B 570 (2000) 81, hep-ph/9906206.
[24] T. Ibrahim, P. Nath, Phys. Rev. D 57 (1998) 478;
T. Ibrahim, P. Nath, Phys. Rev. D 58 (1998) 019901, Erratum;
T. Ibrahim, P. Nath, Phys. Rev. D 58 (1998) 111301;
T. Ibrahim, P. Nath, Phys. Rev. D 60 (1999) 099902, Erratum;
S. Abel, S. Khalil, O. Lebedev, Nucl. Phys. B 606 (2001) 151, hep-ph/0103320;
T. Ibrahim, P. Nath, Phys. Rev. D 64 (2001) 093002, hep-ph/0105025;
T. Ibrahim, P. Nath, hep-ph/0210251.
[25] J.L. Feng, K.T. Matchev, Y. Shadmi, Nucl. Phys. B 613 (2001) 366, hep-ph/0107182.
[26] P. Brax, C.A. Savoy, Nucl. Phys. B 447 (1995) 227, hep-ph/9503306.
[27] T. Goto, T. Nihei, Phys. Rev. D 59 (1999) 115009.

New strategies to obtain insights into CP violation

through Bs DsK , DsK , . . . and
Bd D , D , . . . decays
Robert Fleischer
Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Received 7 April 2003; accepted 8 August 2003
Abstract
Decays of the kind Bs Ds K , Ds K , . . . and Bd D , D , . . . allow us to probe
q0
s + and d + , respectively, involving the angle of the unitarity triangle and the Bq0 B
mixing phases q (q {d, s}). Analysing these modes in a phase-convention-independent way, we
find that their mixing-induced observables are affected by a subtle (1)L factor, where L denotes the
angular momentum of the Bq decay products, and derive bounds on q + . Moreover, we emphasize
that untagged rates are an important ingredient for efficient determinations of weak phases, not
only in the presence of a sizeable width difference q ; should s be sizeable, the combination
of untagged with tagged Bs Ds K , Ds K , . . . observables provides an elegant and
unambiguous extraction of tan(s + ), whereas the conventional determination of s + is
affected by an eightfold discrete ambiguity. Finally, we propose a combined analysis of Bs
Ds K , Ds K , . . . and Bd D , D , . . . modes, which has important advantages,
offering various interesting new strategies to extract in an essentially unambiguous manner.
1. Introduction
The exploration of CP violation through studies of B-meson decays is one of the most
exciting topics of present particle physics phenomenology, the main goal being to perform
stringent tests of the KobayashiMaskawa mechanism [1]. Here the central target is the
unitarity triangle of the CabibboKobayashiMaskawa (CKM) matrix, with its angles ,
and (for a detailed review, see [2]). Thanks to the efforts of the BaBar (SLAC) and Belle
E-mail address: robert.fleischer@cern.ch (R. Fleischer).
doi:10.1016/j.nuclphysb.2003.08.010
460
R. Fleischer / Nuclear Physics B 671 (2003) 459482
q0 Dq u q .
Fig. 1. Feynman diagrams contributing to Bq0 Dq u q and B
(KEK) Collaborations, CP violation could recently be established in the neutral Bd -meson

system with the help of Bd J / KS and similar decays [3]. These modes allow us to
determine sin d , where the present world average is given by sin d = 0.734 0.054 [4],
+4
0 0
implying the twofold solution d = (47+5

4 ) (1335 ) for the Bd Bd mixing phase d ,
which equals 2 in the Standard Model. Here the former solution would be in perfect
agreement with the indirect range following from the Standard-Model CKM fits,
40 d 60 [5], whereas the latter would correspond to new physics [6]. Measuring
the sign of cos d , the two solutions can be distinguished. Several strategies to accomplish
this important task were proposed [7]; an analysis using the time-dependent angular
distribution of the decay products of Bd J /[ + ]K [ 0 KS ] [8,9] is already
in progress at the B factories [10].
An important ingredient for the testing of the KobayashiMaskawa picture is provided
by transitions of the kind Bs Ds K , Ds K , . . . [11] and Bd D , D , . . .
[12], allowing theoretically clean determinations of the weak phases s + and d + ,
respectively, where s is the Bs -meson counterpart of d , which is negligibly small in the
Standard Model. It is convenient to write these decays generically as Bq0 Dq u q , so that
we may easily distinguish between the following cases:

q = s: Ds Ds+ , Ds+ , . . . , us K + , K + , . . . ,

q = d: Dd D + , D + , . . . , ud + , + , . . . .
In the discussion given below, we shall only consider Bq0 Dq u q decays, where at least
one of the Dq , u q states is a pseudoscalar meson. In the opposite case, for example the
Bs0 Ds+ K decay, the extraction of weak phases would require a complicated angular
analysis [1315]. If we look at Fig. 1, we observe that Bq0 Dq u q originates from colourq0 meson may decay into the same
allowed tree-diagram-like topologies, and that also a B
q0 mixing and
final state Dq u q . The latter feature leads to interference effects between Bq0 B
decay processes, allowing the extraction of q + with an eightfold discrete ambiguity.
Since q can be straightforwardly fixed separately [2], we may determine the angle of
the unitarity triangle from this CP-violating weak phase.
In Section 2, we focus on the Bq Dq u q decay amplitudes and rate asymmetries,
and investigate the relevant hadronic parameters with the help of factorization. In this
section, we shall also point out that a subtle factor (1)L arises in the expressions for
461
the mixing-induced observables, where L denotes the angular momentum of the Dq u q

system, and show explicitly the cancellation of phase-convention-dependent parameters
within the factorization approach. After discussing the conventional extraction of q +
and the associated multiple discrete ambiguities in Section 3, we emphasize the usefulness
of untagged rate measurements for efficient determinations of weak phases from Bq
Dq u q decays in Section 4, and suggest several novel strategies. In Section 5, we then
derive bounds on q + , and illustrate their potential power with the help of a few
numerical examples. In Section 6, we propose a combined analysis of Bs Ds u s and
Bd Dd u d modes, which has important advantages with respect to the conventional
separate determinations of s + and d + , offering various attractive new avenues
to extract in an essentially unambiguous manner and to obtain valuable insights into
hadron dynamics. Finally, we conclude in Section 7.
2. Amplitudes, rate asymmetries and factorization

2.1. Amplitudes

The Bq Dq u q decays are the colour-allowed counterparts of the Bs D( ) , D, . . .

and Bd D 0 , D 0 , . . . channels, which were recently analysed in detail in [16,17]. If we
follow the same avenue, and take also the Feynman diagrams shown in Fig. 1 into account,
we may write

0
0 GF
0
q Dq u q
q ,
q Dq u q = u q Dq |Heff B
Bq = vq M
A B
2
(2.1)
where the hadronic matrix element

0
q C1 () + O
q C2 ()
q u q Dq |O
M
Bq
1
2
(2.2)
involves the currentcurrent operators

q (q u )V A (c b )V A ,
O
1
q (q u )V A (c b )V A .
O
2
(2.3)
The CKM factors vq are given by
vs Vus
Vcb = A3 ,

vd Vud
Vcb = A2 1 2 /2 ,
(2.4)
with (for the numerical value, see [18])

1
|Vcb | = 0.83 0.02,
2
and |Vus | = 0.22 is the usual Wolfenstein parameter [19].
On the other hand, the Bq0 Dq u q decay amplitude takes the following form:

A Bq0 Dq u q = u q Dq |Heff Bq0 Dq u q Bq0

q
q
GF
= vq u q Dq | O1 C1 () + O2 C2 ()Bq0 ,
2
A
(2.5)
(2.6)
462
where we have to deal with the currentcurrent operators

q
O1 (q c )V A (u b )V A ,
O2 (q c )V A (u b )V A ,
(2.7)
and the CKM factors vq are given by

vs Vcs Vub
= A Rb e
3
vd Vcd
Vub

A4 Rb
=
ei ,
1 2 /2
with (for the numerical value, see [18])

2 1 Vub
= 2 + 2 = 0.39 0.04.
Rb 1
2 Vcb
(2.8)
(2.9)
If we introduce convention-dependent CP phases through

,
(CP)|F = eiCP (F ) |F
= eiCP (F )|F
(CP)|F
(2.10)
for F {Bq , Dq , uq }, we obtain

q uq ,
(CP)|Dq u q = (1)L ei[CP (Dq )CP (uq )] |D
(2.11)
where L denotes the angular momentum of the Dq u q state. As we shall see below, the
subtle (1)L factor enters in mixing-induced observables, and plays an important role for
the extraction of weak phases from these quantities in the presence of non-trivial angular
momenta, for instance in the case of Bd0 D + . In the literature, this factor does not
show up explicitly in the context of Bq0 Dq u q modes, but it was recently pointed out
in the analysis of their colour-suppressed counterparts in [16,17]. If we now employ, as in
these papers, the operator relations
(CP) (CP) = 1,
(2.12)
q
(CP)Ok (CP)
(2.13)
q
= Ok ,
we may rewrite (2.6) as

GF
A Bq0 Dq u q = (1)L ei[CP (Bq )CP (Dq )+CP (uq )] vq Mq ,
2
(2.14)
where
0
q |Oq C1 () + Oq C2 ()
Bq .
Mq uq D
1
2
(2.15)
It should be noted that also certain exchange topologies contribute to Bq0 Dq u q ,

q0 Dq u q transitions, which werefor simplicitynot shown in Fig. 1. However, these
B
additional diagrams do not affect the phase structure of the amplitudes in (2.1) and (2.14),
and manifest themselves only through tiny contributions to the hadronic matrix elements
q and Mq given in (2.2) and (2.15), respectively. We shall come back to these topologies
M
in Subsection 4.2, noting also how they may be probed experimentally.
463
q0 D
q uq and Bq0 D
q uq processes yields
An analogous calculation for the B
GF
0
q uq =
q D
A B
(2.16)
vq Mq ,
2

GF
q uq = (1)L ei[CP (Bq )+CP (Dq )CP (uq )]
vq Mq ,
A Bq0 D
(2.17)
2
q0 Dq u q modes arise.
where the same hadronic matrix elements as in the Bq0 Dq u q , B
2.2. Rate asymmetries
q0 meson may decay
Let us first consider Bq decays into Dq u q . Since both a Bq0 and a B
into this state, we obtain a time-dependent rate asymmetry of the following form [2]:
q0 (t) Dq u q )
(Bq0 (t) Dq u q ) (B
q0 (t) Dq u q )
(Bq0 (t) Dq u q ) + (B
=
C(Bq Dq u q ) cos(Mq t) + S(Bq Dq u q ) sin(Mq t)

,
cosh(q t/2) A (Bq Dq u q ) sinh(q t/2)
(q)
(2.18)
(q)
where Mq MH ML > 0 is the mass difference of the Bq mass eigenstates BqH
(q)
(q)
(heavy) and BqL (light), and q H L denotes their decay width difference,
providing the observable A (Bq Dq u q ). Before we turn to this quantity in the context
of the untagged rates discussed in Subsection 4.1, let us first focus on C(Bq Dq u q )
and S(Bq Dq u q ). These observables are given by
C(Bq Dq u q ) Cq =
1 |q |2
,
1 + |q |2
S(Bq Dq u q ) Sq =
2 Im q
, (2.19)
1 + |q |2
where
q0 Dq u q )
A(B
q eiq eiCP (Bq )
A(Bq0 Dq u q )
(2.20)
q0 mixing and decay

measures the strength of the interference effects between the Bq0 B
q0 mixing phase
processes, involving the CP-violating weak Bq0 B

SM +2 = O(50 )

(q = d),
q 2 arg Vtq Vt b =
(2.21)
2
2 = O(2 ) (q = s).
If we now insert (2.1) and (2.14) into (2.20), we observe that the convention-dependent
phase CP (Bq ) is cancelled through the amplitude ratio, and arrive at

1
,
q = (1)L ei(q + )
(2.22)
xq eiq
where
xs Rb as ,

xd

2 Rb
ad ,
1 2
(2.23)
464
with
aq eiq ei[CP (Dq )CP (uq )]
Mq
.
q
M
(2.24)
The convention-dependent phases CP (Dq ) and CP (uq ) in (2.24) are cancelled through
the ratio of hadronic matrix elements, so that aq eiq is actually a physical observable.
Employing the factorization approach to deal with the hadronic matrix elements, we shall
demonstrate this explicitly in Subsection 2.3. We may now apply (2.19), yielding

1 xq2
L 2xq sin(q + + q )
,
S
.
=
(1)
Cq =
(2.25)
q
1 + xq2
1 + xq2
If we perform an analogous calculation for the decays into the CP-conjugate final state
q uq , we obtain
D
q0 D
q uq )
A(B

q = eiq eiCP (Bq )
(2.26)
= (1)L ei(q + ) xq eiq ,
0

A(Bq Dq uq )
which implies

1 xq2
q = +
,
C
1 + xq2

Sq = (1)L

2xq sin(q + q )
,
1 + xq2
(2.27)
q uq ) and
q uq ).
q C(Bq D
Sq S(Bq D
where C
It should be noted that q and q satisfy the relation
q q = ei2(q + ) ,
(2.28)
where the hadronic parameter xq eiq cancels. Consequently, we may extract q + in a

theoretically clean way from the corresponding observables. For our purposes, it will be
convenient to introduce the following quantities:
q + Cq
C
= 0,
2
q Cq
1 xq2
C
Cq
,
=
2
1 + xq2

Sq + Sq
2xq cos q
Sq +
= +(1)L
sin(q + ),
2
1 + xq2

Sq Sq
2xq sin q
= (1)L
Sq
cos(q + ).
2
1 + xq2
Cq +
(2.29)
(2.30)
(2.31)
(2.32)
We observe that the factor (1)L is crucial for the correctness of the sign of the mixinginduced observable combinations Sq + and Sq . In particular, if we fix the sign of
cos q through factorization arguments, we may determine the sign of sin(q + ) from
the measured sign of Sq + , providing valuable information. If we consider, for example,
Bs Ds K or Bd D modes, we have L = 1, and obtain a non-trivial factor
465
of (1)1 = 1. On the other hand, we have (1)0 = +1 in the case of Bs Ds K or

Bd D channels. Let us next analyse the hadronic parameter aq eiq with the help of
the factorization approach.
2.3. Factorization
Because of colour-transparency arguments [20,21], the factorization of the hadronic
matrix elements of four-quark operators into the product of hadronic matrix elements of
q0 Dq u q , involving the matrix
two quark currents can be nicely motivated for the decay B
q . Recently, this picture could be put on a much more solid theoretical basis [22].
element M
q0 D
q uq channel entering Mq ,
On the other hand, these arguments do not apply to the B
since there the spectator quark q ends up in the uq meson, which is not heavy (see Fig. 1).
In order to analyse the hadronic parameter aq eiq introduced in (2.24), it is nevertheless
instructive to apply naive factorization not only to (2.2), but also to (2.15), yielding

0
q = a1 u q |(q u )V A |0Dq |(c b )V A
Bq ,
M
(2.33)
fact

0
q |(q c )V A |0uq |(u b )V A
Bq ,
Mq fact = a1 D
(2.34)
where
a1 =
C1 (F )
+ C2 (F ) 1
NC
(2.35)
is the well-known phenomenological colour factor for colour-allowed decays [21], with a
factorization scale F and a number NC of quark colours.
s0 Ds()+ K and B
0 D ()+ , i.e.,
To be specific, let us consider the decays B
d
us = K + , ud = + and Ds = Ds()+ , Dd = D ()+ . Using (2.10) and (2.12), as well as

(CP) q
(1 5 )u (CP) = u
(2.36)
(1 5 )q ,
we obtain
u q |(q u )V A |0 = eiCP (uq ) uq |(u q )V A |0,
(2.37)
q |(q c )V A |0 = eiCP (Dq ) Dq |(c q )V A |0

D
(2.38)
for the pseudoscalar mesons, and

q |(q c )V A |0 = +eiCP (Dq ) Dq |(c q )V A |0
D
(2.39)
for the vector mesons Ds = Ds+ and Dd = D + . If we now use these expressions in
(2.33) and (2.34), we see explicitly that the phase-convention-dependent factor in (2.24)
is cancelled through the ratio of hadronic matrix elements, thereby yielding a conventionindependent result. In the case of the decays Bs Ds K and Bd D , we obtain
as e
is
fact
2
2
2
fDs FB(0)
(MDs )(MBs MK )
sK
2 )(M 2 M 2 )
fK FB(0)
(MK
Bs
Ds
s Ds
(2.40)
466
and
(0)
2 )(M 2 M 2 )
fDd FB (MD

Bd
d
d
,
ad eid fact =
(0)
2
2
2 )
f FBd Dd (M )(MBd MD
d
(2.41)
0 D+
s0 Ds+ K and B
respectively. If we apply heavy-quark arguments to the B
d
modes [21,23], we arrive at

2
2
2
2fDs FB(0)

(MDs )(MBs MK ) MBs MDs
sK
is
,
as e fact =
(2.42)
2 ]
fK s (ws )(MBs MDs )[(MBs + MDs )2 MK

(0)
2 )(M 2 M 2 ) M M
2fDd FB (MD

Bd Dd
Bd
d
id
d
,
ad e fact =
(2.43)
f d (wd )(MBd MDd )[(MBd + MDd )2 M2 ]
q0 Dq transitions, and
where the q (wq ) are the IsgurWise functions describing B
ws =
2 M2
MB2 s + MD
K
s
2MBs MDs
wd =
2 M2
MB2 d + MD
2MBd MDd
(2.44)
In the case of Bs Ds K and Bd D , we obtain accordingly

2
2
fDs FB(1)
2fDs FB(1)

(MD )
(MD ) MBs MDs
sK
sK
s
s
is
as e
=
=
fact
(0)
fK s (ws )(MBs + MDs )
fK A (M 2 )
Bs Ds
and

ad eid fact =
(2.45)
(1)
2 )
fDd FBd (MD
(0)
f ABd D (M2 )
d

2 ) M M
2fDd FB(1) (MD
Bd Dd
d
f d (wd )(MBd + MDd )
(2.46)
respectively, where we have taken the relative minus sign between (2.37) and (2.39) into
account, and
ws =
2 M2
MB2 s + MD
K
s
2MBs MDs
wd =
An important result of this exercise is

q fact = 0 ,
q fact = 180.
2 M2
MB2 d + MD
2MBd MDd
(2.47)
(2.48)
q0 D
q uq ,
q0 Dq u q , in contrast to B
Since factorization is expected to work well for B
(2.48) may in principle receive large corrections, yielding sizeable CP-conserving strong
phases. However, we may argue that we still have
cos q > 0,
cos q < 0,
(2.49)
in accordance with the factorization prediction. This valuable information allows us to fix
the sign of sin(q + ) from (2.31), where the (1)L factor plays an important role, as
we already noted: it is +1 and 1 for Bs Ds K , Bd D and Bs Ds K ,
467
Bd D , respectively. Moreover, it should not be forgotten in this context that xs

is positive, whereas xd is negative because of a factor of 1 originating from the ratio of
CKM factors vd /vd (see (2.23)).
Using, for instance, the BauerStechWirbel form factors [24], we obtain ad = 0.8 and
ad = 1.0; if we take also (2.9) and (2.23) into account, these values can be converted into
xd() = O(0.02), whereas xs() = O(0.4). In Section 6, we shall have a closer look at the
flavour-symmetry-breaking effects, which arise in the ratios as /ad and as /ad .
It is useful to briefly compare these results with the situation of the colour-suppressed

counterparts of the Bq Dq u q decays, the Bs D( ) , D, . . . and Bd D 0 , D 0 , . . .
modes discussed in [16,17]. Here factorization may receive sizeable corrections for each
q0 D 0 fq amplitudes. However, the corresponding hadronic
of the Bq0 D 0 fq and B
matrix elements are actually very similar to one another, so that the factorized matrix elements cancel in the counterpart of aq eiq . Consequently, the thus obtained information on
the sign of the cosine of the corresponding strong phase difference fq appears to be a bit
more robust than (2.49).
3. Conventional extraction of q +
We are now well prepared to discuss the conventional extraction of the CP-violating
phase q + from Bq Dq u q decays [11,12]. As we have already noted, because of
(2.28), it is obvious that these modes and their CP conjugates provide a theoretically clean
extraction of this phase. Using (2.30), we mayin principledetermine xq through

1 Cq
xq = q
(3.1)
,
1 + Cq
where
q =
+1 (q = s),
1 (q = d),
(3.2)
takes into account the minus sign appearing in (2.23) for q = d. Using the knowledge of
xq , we may extract the following quantities from the combinations of the mixing-induced
observables introduced in (2.31) and (2.32):

1 + xq2
L
Sq + = + cos q sin(q + ),
s+ (1)
(3.3)
2xq

1 + xq2
s (1)L
(3.4)
Sq = sin q cos(q + ),
2xq
which allow us to determine sin2 (q + ) with the help of

1
2
2
2 s 2 2 4s 2 .
1 + s+
sin2 (q + ) =
(3.5)
1 + s+
s
+
2
This relation implies a fourfold solution for sin(q + ). Since each value of this quantity
corresponds to a twofold solution for q + , the extraction of this phase suffers, in general,
468
from an eightfold discrete ambiguity. If we employ (2.49) and (3.3), the measured sign of
s+ allows us to fix the sign of sin(q + ), thereby reducing the discrete ambiguity for
the value of q + to a fourfold one. Needless to note that these unpleasant ambiguities
significantly reduce the power to search for possible signals of new physics.
Another disadvantage is that the determination of the hadronic parameter xq through
(3.1) requires the experimental resolution of small xq2 terms in (2.30). In the q = s case, we
naively expect xs2 = O(0.16), so that this may actually be possible, though challenging.1
On the other hand, it is practically impossible to resolve the xd2 = O(0.0004) terms, i.e.,
(2.30) is not effective in the q = d case. However, it may well be possible to measure
the observable combinations Sd + and Sd , since these quantities are proportional to
xd = O(0.02). In this respect, Bd D channels are particularly promising, since
they exhibit large branching ratios at the 103 level and offer a good reconstruction of the
D states with a high efficiency and modest backgrounds [25,26]. In order to solve
the problem of the extraction of xd , which was also addressed in [12], we shall propose
the use of untagged decay rates, where we do not distinguish between initially, i.e., at
0 mesons. Also in the case of q = s, alternatives to (3.1) for an
time t = 0, present Bd0 or B
d
efficient determination of xs are obviously desirable.
4. Closer look at untagged rates

4.1. New strategy employing q
As we have seen in (2.18), the width difference q of the Bq mass eigenstates provides
another observable, A (Bq Dq u q ), which is given by
A (Bq Dq u q ) =
2 Re q
.
1 + |q |2
(4.1)
This quantity is, however, not independent from C(Bq Dq u q ) and S(Bq Dq u q ),
satisfying the relation
2

2
2
C Bq Dq u q
(4.2)
+ S(Bq Dq u q ) + A (Bq Dq u q ) = 1.
Interestingly, A (Bq Dq u q ) could be determined from the untagged rate

0

q (t) Dq u q
Bq (t) Dq u q Bq0 (t) Dq u q + B

0

q Dq u q
= Bq0 Dq u q + B

cosh(q t/2) A (Bq Dq u q ) sinh(q t/2) eq t ,
(q)
(4.3)
(q)
where the oscillatory cos(Mq t) and sin(Mq t) terms cancel, and q (H + L )/2
denotes the average decay width [27]. In the case of the Bd -meson system, the width
difference is negligibly small, so that the time evolution of (4.3) is essentially given by
1 Note that non-factorizable effects may well lead to a significant reduction or enhancement of x .
s
469
the well-known exponential ed t . On the other hand, the width difference s of the Bs meson system may be as large as O(10%) (for a recent review, see [28]), and may hence
allow us to extract A (Bs Ds u s ).
Inserting (2.22) into (4.1), we obtain

L 2xs cos(s + + s )
A (Bs Ds u s ) As = (1)
(4.4)
,
1 + xs2
and correspondingly

L 2xs cos(s + s )

,
A (Bs Ds us ) As = (1)
1 + xs2
(4.5)
which yields

s + As
A
2xs cos s
= (1)L
cos(s + ),
2
1 + xs2

s As

A
2xs sin s
= (1)L
As
sin(s + ).
2
1 + xs2
As
(4.6)
(4.7)
If we compare now (4.6) and (4.7) with (2.31) and (2.32), respectively, we observe that the
same hadronic factors enter in these mixing-induced observables, and obtain
Ss +
= tan(s + ),
As +
As
= + tan(s + ),
Ss
(4.8)
(4.9)
implying the consistency relation

As + As = Ss + Ss .
0
(4.10)
180 ,
Should s take values around

or
as in factorization (see (2.48)), we may extract
tan(s + ) from (4.8), whereas we could use (4.9) in the opposite case of s being close
to +90 or 90 . The strong phase itself can be determined from
tan s =
Ss
As
=
.
As +
Ss +
(4.11)
The values of tan(s + ) and tan s thus extracted imply twofold solutions for s +
and s , respectively, which should be compared with the eightfold solution for s +
following from (3.5). Using (2.49), we may immediately fix s unambiguously, and may
determine the sign of sin(s + ) with the help of the measured sign of Ss + from (2.31),
thereby resolving the twofold ambiguity for the value of s + . On the other hand, the
conventional approach discussed in Section 3 would still leave a fourfold ambiguity for
this phase, as we shall illustrate in Section 5. Finally, we may of course also determine xs
from one of the Ss or As observables.
We observe that the combination of the tagged mixing-induced observables Ss
with their untagged counterparts As provides an elegant determination of s +
in an essentially unambiguous manner. In [13], strategies to determine this phase from
470
untagged Bs data samples only were proposed, which employ angular distributions of
decays of the kind Bs Ds K and are hence considerably more involved. Another
important advantage of our new strategy is that both Ss and As are proportional
to xs . Consequently, the extraction of s + does not require the resolution of xs2 terms.2
On the other hand, we have to rely on a sizeable width difference s , which may be too
small to make an extraction of As experimentally feasible. In the presence of CPs0 mixing, manifesting themselves through a
violating new-physics contributions to Bs0 B
sizeable value of s , s would be further reduced, as follows [29]:
s = sSM cos s ,
(4.12)
where sSM is negative [28]. As is well known, s can be determined through Bs

J / , which is very accessible at hadronic B-decay experiments [25,30]. Strategies to
determine s unambiguously were proposed in [9,16].

In the case of the Bs D( ) , D, . . . modesthe colour-suppressed counterparts of
the Bs Ds u s channels, untagged rates for processes where the neutral D mesons are
observed through their decays into CP eigenstates f provide a very useful untagged rate
asymmetry , allowing efficient and essentially unambiguous determinations of from
mixing-induced observables [16,17]. These strategies, which can also be implemented for
Bd DKS(L) modes, have certain similarities with those provided by (4.8) and (4.9).
However, they do not rely on a sizeable value of q , as is extracted from unevolved
untagged rates, which are also very useful for the analysis of Bq Dq u q modes, as we
shall see below. Since these decays involve charged Dq mesons, the observable has
unfortunately no counterpart for the colour-allowed transitions.
4.2. Employing untagged rates in the case of negligible q
Even for a vanishingly small width difference q , the untagged rate (4.3) provides
valuable information, as it still allows us to determine the unevolved, untagged rate

0

q Dq u q .
(Bq Dq u q ) Bq0 Dq u q + B
(4.13)
Using (2.1) and (2.14), as well as (2.16) and (2.17), we obtain
q uq )
(Bq Dq u q )
(Bq D
1
.
=1+ 2 =
0
q0 D
q uq )
(Bq Dq u q )
xq
(B
If we now employ

0

q D
q uq ,
Bq0 Dq u q = B
which follows from (2.14) and (2.16), we may write
1/2
q uq )
(Bq Dq u q ) + (Bq D
xq = q
,
1
q0 D
q uq )
(Bq0 Dq u q ) + (B
(4.14)
(4.15)
(4.16)
2 A similar feature is also present in the untagged B D K strategy proposed in [13], and in the
s
s
tagged analysis in [14], employing the angular distribution of the Ds , K decay products.
471
offering a very attractive untagged alternative to (3.1), provided we fix the sum of the
Bq0 Dq u q rate and its CP conjugate in an efficient manner. To this end, we may replace
the spectator quark q by an up quark, which will allow us to determine this quantity from
the CP-averaged rate of a charged B-meson decay as follows:3

0

q D
q uq
Bq0 Dq u q + B

q uu ,
= 2Cq2 B + Dq u u + B D
(4.17)
where uu { 0 , 0 , . . .} depends on the choice of uq . For example, we have uu = 0 for
+ , whereas u = 0 for u = + or u = K + . The factor of 2 takes
ud = + or us = K
u
d
s
into account the 1/ 2 factor of the uu wave function, and the deviation of Cq from 1 is
governed by flavour-symmetry-breaking effects, which originate from the replacement of
the spectator quark q through an up quark.
Since B + Dd u u is related to Bd0 Dd u d through SU(2) isospin arguments, we
obtain to a good approximation
Cd = 1.
(4.18)
In addition to the conventional isospin-breaking effects, exchange topologies, which

contribute to Bd0 Dd u d but have no counterpart in B + Dd u u , and annihilation
topologies, which arise only in B + Dd u u but not in Bd0 Dd u d , are another limiting
factor of the theoretical accuracy of (4.18). Although these contributions are naively
expected to be very small, they mayin principlebe enhanced through rescattering
processes. Fortunately, we may probe their importance experimentally. In the case of
()
Bd D () and B + D ()+ 0 this can be done with the help of Bd Ds K
+
()+
0
and B D
K processes, respectively.
Applying (4.17) to the q = s case, we have to employ the SU(3) flavour symmetry. If
we neglect non-factorizable SU(3)-breaking effects, the Cs are simply given by appropriate
form-factor ratios; important examples are the following ones:
(0)
Bs Ds K
2 )(M 2 M 2 )
FBs K (MD
Bs
K
s
Bs Ds K
(0)
2 )(M 2 M 2 )
FB 0 (MD
Bu
2
FB(1)
(MD )
sK
s
(1)
2 )
FB 0 (MD
(4.19)
Also here, we have to deal with exchange topologies, which contribute to Bs0 Ds K
but have no counterpart in B + Ds()+ 0 . Experimental probes for these topologies are
provided by Bs D () processes.
As an alternative to (4.17), we may use

0

d D () +
Bd0 D ()+ + B

0

d Ds() +
= Bd0 Ds()+ + B
(4.20)
()+
3 For simplicity, we neglect tiny phase-space effects, which can be straightforwardly included.
472
and

0

s Ds() K +
Bs0 Ds()+ K + B

0

1
s D () K + ,
= Bs0 D ()+ K + B
where

2
1 2
f
()
(4.21)
2
Dd
fD ()
(4.22)
takes into account factorizable SU(3)-breaking corrections through the ratio of the Dd()
and Ds() decay constants. The decays on the right-hand sides of (4.20) and (4.21) have the
advantage of involving flavour-specific final states f , satisfying A(Bq0 f ) = 0 and
q0 f ) = 0. In this important special case, the time-dependent untagged rates take
A(B
the following simple forms:

0

q (t) f
Bq (t) f Bq0 (t) f + B

= Bq0 f cosh(q t/2) eq t ,
(4.23)

0

0

q (t) f
Bq (t) f Bq (t) f + B
0

q f cosh(q t/2) eq t ,
= B
(4.24)
q0 f) with
and allow an efficient extraction of the CP-averaged rate (Bq0 f ) + (B
the help of

Bq (t) f + Bq (t) f

0

q f cosh(q t/2) eq t .
= Bq0 f + B
(4.25)
Obviously, in the case of q = d, (4.17) is theoretically cleaner than (4.20), providingin
combination with (4.16)a very interesting avenue to determine xd . On the other hand,
the modes on the right-hand side of (4.20) are more accessible from an experimental point
of view, and were already observed at the B factories [31].
Since simple colour-transparency arguments do not apply to Bq0 Dq u q , B +
Dq u u modes, as we noted in Subsection 2.3, expressions (4.19), (4.20) and (4.21)
may receive sizeable non-factorizable SU(3)-breaking corrections. However, there is yet
q uq rate, where
another possibility to exploit (4.13). To this end, we factor out the Bq0 D
factorization is expected to work well [22], yielding
q uq )
(Bq Dq u q )
(Bq D
= 1 + xq2 =
,
0
0

q uq )
(Bq Dq u q )
(Bq D
which implies

q uq )
(Bq Dq u q ) + (Bq D
1.
xq = q
q0 Dq u q ) + (Bq0 D
q uq )
(B
(4.26)
(4.27)
473
In the q = d case, it willin analogy to (2.30)be impossible to resolve the vanishingly

small xq2 term in (4.26). On the other hand, this may well be possible in the q = s case. If
we use

0
s Ds()+ K + Bs0 Ds() K +
B

2

2
fK 0
Bs Ds()+ + Bs0 Ds() + ,
=
(4.28)
2
1
f
expression (4.27) offers a very attractive possibility to determine the values of xs() , where
(fK /f )2 describes factorizable SU(3)-breaking effects. Additional corrections are due
s0 Ds()+ K , but are not present in B
s0
to exchange topologies, which arise in B
()+
Ds . However, as we already noted, their contributions are expected to be very small,
s0
and can be probed experimentally through Bs D () processes. Since the B
()+
() +
0
Ds and Bs Ds rates involve flavour-specific final states, we may efficiently
determine their sum from untagged Bs data samples, with the help of (4.25). In this context,
it should also be noted that these rates are enhanced by a factor of (1 2 )/2 20 with
respect to the Bs Ds() K rates. Moreover, non-factorizable effects are expected to
play a minor role in (4.28) because of colour-transparency arguments, in contrast to (4.19)
and (4.21). Further calculations along [22] should provide an even more accurate treatment
of the SU(3)-breaking corrections. In comparison with (3.1), the advantage of the strategy
offered by (4.27) and (4.28) is the use of untagged rates, which are particularly promising
in terms of efficiency, acceptance and purity, and do not require the measurement of the
time-dependent cos(Ms t) terms in (2.18). Interestingly, the quantity 1 + xs2 , which can
nicely be determined through the combination of (4.27) and (4.28), will play an important
role in Section 6.
As we have seen above, the untagged rates introduced in (4.3) provide various strategies
to determine the hadronic parameters xq , some of which are particularly favourable. In
order to implement these approaches, we must not rely on a sizeable width difference
q . It will be interesting to see whether they will eventually yield a consistent picture of
the xq . Following these lines, we may also obtain valuable insights into hadron dynamics.
5. Bounds on q +
If we keep xq and q as unknown, i.e., free parameters in (2.31) and (2.32), we may
derive the following bounds:

sin(q + ) Sq + ,
(5.1)

cos(q + ) Sq .
(5.2)
On the other hand, if we assume that xq has been determined with the help of the
untagged strategies proposed in Subsection 4.2, we may fix the quantities s+ and s
introduced in (3.3) and (3.4), respectively, providing more stringent constraints:

sin(q + ) |s+ |,
(5.3)

cos(q + ) |s |.
(5.4)
474
Table 1
The mixing-induced observables in the case of L = 0, = 60 , d = 47 , s = 0 , Rb = 0.4 and aq = 1:
the upper half corresponds to factorization, i.e., q = 0 , whereas the lower half illustrates a non-factorization
scenario with q = 40 . Note that we have Cs = 0.724, while the deviation of Cd from 1 is negligibly
small
q
Sq , %

Sq , %
Sq + , %
Sq , %
d
s
d
s
3.89
+59.7
2.22
+67.9
3.89
+59.7
3.74
+23.6
3.89
+59.7
2.98
+45.8
+0.00
+0.00
0.76
22.2
s+ , %
+95.6
+86.6
+73.3
+66.3
s , %
+0.00
+0.00
+18.8
32.1
Interestingly, (5.1) and (5.3) allow us to exclude a certain range of values of q + around
0 and 180 , whereas (5.2) and (5.4) provide complementary information, excluding a
certain range around 90 and 270 . The constraints in (5.1) and (5.2) have the advantage
of not requiring knowledge of xq . On the other hand, because of the small value of xd ,
we may only expect useful information from them in the case of q = s. Once s+ and s
have been extracted, it is of course also possible to determine sin2 (q + ) through the
complicated expression in (3.5), as discussed in Section 3. However, since the resulting
values for q + suffer from multiple discrete ambiguities, the information they are
expected to provide about this phase isin generalnot significantly better than the
constraints following from the very simple relations in (5.3) and (5.4).
It is instructive to illustrate this feature with the help of a few numerical examples. To
this end, we assume = 60 , d = 47 and s = 0 , which would be in perfect agreement
with the Standard Model, as well as Rb = 0.4 and aq = 1. Let us consider the decays
Bd D and Bs Ds K , which have L = 0. As far as q is concerned, we
may then distinguish between a factorization scenario with q = 0 (see (2.48)), and
a non-factorization scenario, corresponding to q = 40 . For simplicity, we shall use
the same hadronic parameters aq eiq for the q = d and q = s cases. The corresponding
mixing-induced observables are listed in Table 1. Let us also assume that d and s will
be unambiguously known by the time these observables can be measured. As we have
already noted, because of the small value of xd , (5.1) and (5.2) do not provide non-trivial
constraints on d + , in contrast to their application to the q = s case.
Let us first focus on the factorization scenario, corresponding to the upper half of
Table 1. Since Sq and s vanish in this case, as these observable combinations are
proportional to sin q , (5.2) and (5.4) imply only trivial constraints on q + . However,
we may nevertheless obtain interesting bounds in this case. For the q = d example, the
situation is as follows: if we employ (2.49) and take into account that xd is negative, the
negative sign of Sd + implies a positive value of sin(d + ), i.e., 0 d + 180 .
Applying now (5.3), we obtain 73 d + 107 from s+ , which corresponds to
26 60 , providing valuable information about . On the other hand, if we use again
that sin(d + ) is positive, the complicated expression (3.5) implies the threefold solution
= 26 43 60 , which covers essentially the whole range following from the simple
relation in (5.3). It is very interesting to complement the information on thus obtained
from Bd D with the one provided by its Bs Ds K counterpart. Using again
(2.49), the positive sign of Ss + implies that sin(s + ) is positive, i.e., 0 s +
475
180 . We may now apply (5.1) to obtain the bound 37 s + 143 from Ss + ;
a narrower range follows from s+ through (5.3), and is given by 60 s + 120 .
Since s = 0 , we may identify these ranges directly with bounds on . On the other hand,
the complicated expression (3.5) implies the threefold solution = 60 90 120 ,
which falls perfectly into the range provided by s+ , which can be obtained in a much
simpler manner. We now make the very interesting observation that the q = s range of
60 120 is highly complementary to its q = d counterpart of 26 60 ,
leaving 60 as the only overlap. Consequently, in this example, the combination of our
simple bounds on d + and s + yields the single solution of = 60 , which
corresponds to our input value, thereby nicely demonstrating the potential power of these
constraints.
Let us now perform the same exercise for the non-factorization scenario, represented by
the lower half of Table 1. In the case of q = d, s+ and s imply 47 d + 133 and
(0 d + 79 ) (101 d + 180), respectively, which can be combined with
each other, taking also d = 47 into account, to obtain (0 32 ) (54 86 ).
On the other hand, if we apply (3.5) and use that sin(d + ) is positive, we obtain the
fourfold solution = 3 26 60 83 . Let us now consider the q = s case. Here Ss +
and Ss imply 27 s + 153 and (0 s + 77) (103 s + 180),
respectively, yielding the combined range (27 s + 77 )(103 s + 153).
Using s+ and s , and taking into account that s = 0 , we obtain the more stringent
constraint (42 71 ) (109 138), whereas (3.5) would imply the fourfold
solution = 50 60 120 130 , providing essentially the same information. We
observe again that the bounds on arising in the q = d and q = s cases are highly
complementary to each other, having a small overlap of 54 71 . Although the
constraint on following from the bounds on q + would now not be as sharp as in the
factorization scenario discussed above, this approach would still provide very non-trivial
information about this particularly important angle of the unitarity triangle.
In Table 1, we have considered a Standard-Model-like scenario for the weak phases.
However, as argued in [6], the present data are also perfectly consistent with the picture
0 mixing.
of (d , ) = (133, 120), corresponding to new-physics contributions to Bd0 B
d
Since we have sin(d + ) sin(d + ) for d 180 d , 180 , the

sign of the (1)L Sd + observable combination allows us to distinguish between the
(d , ) = (47 , 60 ) and (133 , 120) scenarios, corresponding to sin(d + ) = +0.956
and 0.956, respectively. Practically, this can be done with the help of Bd D ()
modes. If we take into account that the xd() are negative, include properly the (1)L
factors and fix the signs of cos d() through (2.49), we find that a positive value of the
Sd() + observables would be in favour of the unconventional (d , ) = (133 , 120)
scenario, whereas a negative value would point towards the Standard-Model picture
of (d , ) = (47 , 60). A first preliminary analysis of Bd D by the BaBar
Collaboration [32] gives
Sd + = 0.063 0.024(stat.) 0.017(syst.),
thereby favouring the latter case.
(5.5)
476
6. Combined analysis of Bs,d Ds,d u s,d modes

As we have seen in the previous section, it is very useful to make a simultaneous analysis
of Bs Ds u s and Bd Dd u d decays. Let us now further explore this observation. Using
(2.31) and (2.32), we may write
as cos s
Ss +
Ls Ld sin(d + )
(6.1)
R = (1)
ad cos d
sin(s + ) Sd +
and

cos(d + ) Ss
as sin s
R = (1)Ls Ld
,
ad sin d
cos(s + ) Sd
respectively, where

1 + xd2
1 2
R
.
2
1 + xs2
(6.2)
(6.3)
Using the results derived in Subsection 4.2, we may easily determine the parameter R,
where the xd2 term is negligibly small, and xs enters only through 1 + xs2 , i.e., a moderate
()
correction. To be specific, let us consider the Bs Ds K channels. If we insert (4.28)
into (4.27), we arrive at
2
s0 Ds()+ ) + (Bs0 Ds() + )

(B
fK
,
R() =
(6.4)
()+
()
f
(Bs Ds K ) + (Bs Ds K + )
where the decay rates can be straightforwardly extracted from untagged Bs data samples
with the help of (4.3) and (4.25). As we have emphasized in Subsection 4.2, nonfactorizable SU(3)-breaking corrections to this relation are expected to be very small.
If we look at Fig. 1, we see that each Bs Ds u s mode has a counterpart Bd Dd u d ,
which can be obtained from the Bs transition by simply replacing all strange quarks
()+
through down quarks; an important example is the Bs0 Ds K , Bd0 D ()+
system. For such decay pairs, we have Ls = Ld , and the U -spin flavour symmetry of
strong interactions, which relates strange and down quarks in the same manner as ordinary
isospin relates up and down quarks, implies the following relations for the corresponding
hadronic parameters:4
as = ad ,
s = d ,
(6.5)
which we may apply in a variety of ways.

Let us first consider a factorization-like scenario, where cos s 1 cos d and
Ss 0 Sd (see Table 1). In this case, (6.2) would not be applicable. However,
we may use (6.1) to determine tan through

sin d S sin s s =0
sin d
tan =
(6.6)
=
,
cos d S cos s
cos d S
4 Note that these relations do not rely on the neglect of (tiny) exchange topologies.
where
S|U spin = R

Sd +
.
Ss +
477
(6.7)
If we follow these lines, we obtain a twofold solution = 1 2 , where we may choose

1 [0, 180 ] and 2 = 1 + 180 ; the theoretical uncertainty would mainly be limited
by U -spin-breaking corrections to as = ad , apart from tiny corrections to cos s = cos d .
If we assumeas is usually donethat lies between 0 and 180, as is implied by the
Standard-Model interpretation of K , which measures the indirect CP violation in the
neutral kaon system, we may immediately exclude the 2 solution. However, since K
may well be affected by new physics, it is desirable to check whether actually falls in the
interval [0 , 180]. To this end, we may use (2.49) and the signs of the Sq + observables,
as we have seen in the examples discussed in Section 5.
Let us now consider a non-factorization-like scenario with sizeable CP-conserving
strong phases, so that we may also employ (6.2), as the Sq observables would no longer
vanish. If we assume that s = d , we may calculate (as /ad )R both with the help of the
Sq + observables through (6.1) and with the help of the Sq observables through (6.2).
The intersection of the corresponding curves then fixes and (as /ad )R. Comparing the
value of (as /ad )R thus extracted with (6.4), we could determine as /ad . If we use the
observables given in the lower half of Table 1, which were calculated for s = d = 40
and as = ad = 1, we obtain the contours shown in Fig. 2, where we have also taken the
bounds implied by (5.1) and (5.2) into account, and have represented the curves originating
Fig. 2. Extraction of assuming s = d for the non-factorization scenario in Table 1; the dashed and dotted
curves were calculated with the help of (6.1) and (6.2), respectively.
478
from (6.1) and (6.2) through the dashed and dotted lines, respectively. We observe that
the intersection of these contours gives actually our input value of = 60 , without any
discrete ambiguity. These observations can easily be put on a more formal level, since (6.1)
and (6.2) imply the following exact relation:
tan d Ss Sd + U spin Ss Sd +

tan(d + )
=
.
(6.8)
tan(s + )
tan s Ss + Sd
Ss + Sd
Consequently, the theoretical uncertainty of the resulting value of would only be limited
by U -spin-breaking corrections to tan s = tan d ; in Fig. 2, they would enter through a
systematic relative shift of the dashed and dotted contours.
Finally, we may also extract without assuming that s is equal to d . To this end, we
use the exact relation

sin(2d + 2 ) Ss 2+ cos2 (s + ) + Ss 2 sin2 (s + )
as

,
R =
(6.9)
ad
sin(2s + 2 ) Sd 2+ cos2 (d + ) + Sd 2 sin2 (d + )
where we have

= sgn Ss + Sd + sin(d + ) sin(s + )
if we assume that cos s and cos d have the same sign, and

= sgn Ss Sd cos(d + ) cos(s + )
(6.10)
(6.11)
if we assume that sin s and sin d have the same sign. Using (6.9), we may calculate
(as /ad )R in an exact manner as a function of from the measured values of the mixinginduced observables Ss and Sd . On the other hand, we have as ad because of
the U -spin flavour symmetry, and may efficiently fix R from untagged Bs data samples
through (6.4), allowing us to determine . Let us illustrate how this strategy works in
practice by considering again an example, corresponding to as = ad = 1, d = 50 and
s = 30 . Moreover, as in Table 1, we choose = 60 , d = 47 , s = 0 and Rb = 0.4,
implying Sd + = 2.50%, Sd = 0.91%, Ss + = 51.7%, Ss = 17.2%. If we
apply (5.1) and (5.2) to the Bs observables, we obtain 31 80 100 133 .
Constraining to this range, the right-hand side of (6.9) yields the solid lines shown
in Fig. 3, where we have represented the measured value of R through the horizontal
dot-dashed line; the three lines emerge if we fix through (6.10), yielding the threefold
solution = 33 60 104 . However, (6.11) leaves only the thicker solid line in the
middle, thereby implying the single solution = 60 . In this particular example, the
extracted value for would be quite stable with respect to variations of (as /ad )R, i.e.,
would not be very sensitive to U -spin-breaking corrections to as = ad . We have also
included the contours corresponding to (6.1) and (6.2) through the dashed and dotted
curves, as in Fig. 2; their intersection would now give = 68 , deviating by only 8 from
the correct value. It should be noted that we may also determine the strong phases s and
d with the help of

Sq
tan q =
(6.12)
tan(q + ),
Sq +
479
Fig. 3. Extraction of with the help of (6.9), yielding the solid lines, for an example with d = 50 and s = 30 ,
as discussed in the text.
providing valuable insights into non-factorizable U -spin-breaking effects.

In comparison with the conventional Bq Dq u q approachesapart from issues
related to multiple discrete ambiguitiesthe most important advantage of the strategies
proposed above is that they do not require the resolution of xq2 terms, since the mixinginduced observables Sd and Ss are proportional to xd and xs , respectively. In
particular, xd has not to be fixed, and xs may only enter through 1 + xs2 , i.e., a moderate
correction, which can straightforwardly be included through untagged Bs rate analyses.
Interestingly, the motivation to measure xs and xd accurately is here related only to the
feature that these parameters would allow us to take into account possible U -spin-breaking
corrections to (6.9) through
as
=
ad

xs
2
.
1 2 xd
(6.13)
After all these steps of progressive refinement, we would eventually obtain a theoretically
clean value of .
For a theoretical discussion of the U -spin-breaking effects affecting the ratio as /ad , we
may distinguishapart from mass factorsbetween two pieces,
as
1 2 ,
ad
(6.14)
480
which can be written for the Bs Ds K , Bd D () systemif we apply the

factorization approximationwith the help of (2.42), (2.43) and (2.45), (2.46) as follows:
()

f d (wd() )
,
1() fact =
()
fK s (ws )

2
fact
2
fDs FB(0)
(MDs )
sK
(0)
2
(MDd )
d
fDd FB
(6.15)
2 fact
(1)
2 )
fDs FBs K (MD
(1)
2
(MD )
d
d
(6.16)
fDd FB
Because of the arguments given in Subsection 2.3, the factorized expression (6.15) for
()
1 is expected to work well. Studies of the light-quark dependence of the IsgurWise
function were performed in [33] within heavy-meson chiral perturbation theory, indicating
an enhancement of s /d at the level of 5%. The application of the same formalism to
fDs /fDd yields values at the 1.2 level [34], which is of the same order of magnitude as
()
recent lattice calculations (see, for example, [35]). In the case of 2 , (6.16) may receive
sizeable non-factorizable corrections, since simple colour-transparency arguments are not
on solid ground, and the new theoretical developments related to factorization that were
presented in [22] are not applicable. Moreover, we are not aware of quantitative studies
of the SU(3)-breaking effects arising from the Bs K , Bd form factors in
(6.16), which could be done, for instance, with the help of lattice or sum-rule techniques.
Following the latter approach, sizeable SU(3)-breaking corrections were found for the
Bs K , Bd form factors in [36]. Hopefully, a better theoretical treatment of
the U -spin-breaking corrections to as /ad will be available by the time the Bq Dq u q
measurements can be performed in practice.
The new strategies proposed above complement other U -spin approaches to extract
[37,38], where the U -spin-related Bs K + K , Bd + system is particularly
promising [25,30,38]. Since penguin topologies play here a crucial rle, whereas these
()
topologies do not contribute to the Bs Ds K , Bd D () system, it will be
very interesting to see whether inconsistencies for will emerge from the data.
7. Conclusions
Let us now summarize the main points of our analysis:
We have shown that Bs Ds K , Ds K , . . . and Bd D , D , . . .
decays can be described through the same set of formulae by just making straightforward
replacements of variables. We have also pointed out that a factor of (1)L arises in the
expressions for the mixing-induced observables. In the presence of a non-vanishing angular
momentum L of the Bq decay products, this factor has properly to be taken into account
in the determination of the sign of sin(q + ) from Sq + .
Should the width difference s be sizeable, the combination of the tagged
mixing-induced observables Ss with their untagged counterparts As offers an
elegant determination of tan(s + ) in an essentially unambiguous manner, which does
not require knowledge of xs . Another important aspect of untagged rate measurements is
481
the efficient determination of the hadronic parameters xq . To accomplish this task, we may
apply various untagged strategies, which do not rely on a sizeable value of q .
We have derived bounds on q + , which can straightforwardly be obtained from the
mixing-induced Bq Dq u q observables, and provide essentially the same information
as the conventional determination of q + , which suffers from multiple discrete
ambiguities. Giving a few examples, we have illustrated the potential power of these
constraints, and have seen that stringent bounds on may be obtained through a combined
study of Bs Ds u s and Bd Dd u d modes.
If we perform a simultaneous analysis of U -spin-related decays, for example of the
()
Bs Ds K , Bd D () system, we may follow various attractive avenues to
determine from the corresponding mixing-induced observables Sq . The differences
between these methods are due to different implementations of the U -spin relations
for the hadronic parameters aq and q . For example, we may extract by assuming
tan s = tan d or as = ad . In comparison with the conventional Bq Dq u q approaches,
the most important advantage of these strategiesapart from features related to discrete
ambiguitiesis that xd does not have to be fixed, and that xs may only enter through 1+xs2 ,
i.e., a moderate correction, which can straightforwardly be included through untagged Bs
rate measurements; an accurate determination of xd and xs would only be interesting for
the inclusion of U -spin-breaking corrections to as /ad . After various steps of refinement,
we would eventually arrive at an unambiguous, theoretically clean value of , and could
also obtainas a by-productvaluable insights into U -spin-breaking effects.
Since Bs,d Ds,d u s,d modes will be accessible in the era of the LHC, in particular at
LHCb, we strongly encourage a simultaneous analysis of Bs and Bd modesespecially of
U -spin-related decay pairsto fully exploit their very interesting physics potential.
References
[1] M. Kobayashi, T. Maskawa, Prog. Theor. Phys. 49 (1973) 652.
[2] R. Fleischer, Phys. Rep. 370 (2002) 537.
[3] BaBar Collaboration, B. Aubert, et al., Phys. Rev. Lett. 87 (2001) 091801;
Belle Collaboration, K. Abe, et al., Phys. Rev. Lett. 87 (2001) 091802.
[4] Y. Nir, WIS-35-02-DPP, hep-ph/0208080.
[5] A.J. Buras, TUM-HEP-435-01, hep-ph/0109197;
A. Ali, D. London, Eur. Phys. J. C 18 (2001) 665;
D. Atwood, A. Soni, Phys. Lett. B 508 (2001) 17;
M. Ciuchini, et al., JHEP 0107 (2001) 013;
A. Hcker, et al., Eur. Phys. J. C 21 (2001) 225.
[6] R. Fleischer, J. Matias, Phys. Rev. D 66 (2002) 054009;
R. Fleischer, G. Isidori, J. Matias, JHEP 0305 (2003) 053.
[7] Ya.I. Azimov, V.L. Rappoport, V.V. Sarantsev, Z. Phys. A 356 (1997) 437;
Y. Grossman, H.R. Quinn, Phys. Rev. D 56 (1997) 7259;
J. Charles, et al., Phys. Lett. B 425 (1998) 375;
B. Kayser, D. London, Phys. Rev. D 61 (2000) 116012;
H.R. Quinn, et al., Phys. Rev. Lett. 85 (2000) 5284.
[8] A.S. Dighe, I. Dunietz, R. Fleischer, Phys. Lett. B 433 (1998) 147.
[9] I. Dunietz, R. Fleischer, U. Nierste, Phys. Rev. D 63 (2001) 114015.
482
[10] R. Itoh, KEK-PREPRINT-2002-106, hep-ex/0210025.

[11] R. Aleksan, I. Dunietz, B. Kayser, Z. Phys. C 54 (1992) 653.
[12] I. Dunietz, R.G. Sachs, Phys. Rev. D 37 (1988) 3186;
I. Dunietz, R.G. Sachs, Phys. Rev. D 39 (1989) 3515, Erratum;
I. Dunietz, Phys. Lett. B 427 (1998) 179;
M. Diehl, G. Hiller, Phys. Lett. B 517 (2001) 125;
D.A. Suprun, C.W. Chiang, J.L. Rosner, Phys. Rev. D 65 (2002) 054025.
[13] R. Fleischer, I. Dunietz, Phys. Lett. B 387 (1996) 361.
[14] D. London, N. Sinha, R. Sinha, Phys. Rev. Lett. 85 (2000) 1807.
[15] M. Gronau, D. Pirjol, D. Wyler, Phys. Rev. Lett. 90 (2003) 051801.
[16] R. Fleischer, Phys. Lett. B 562 (2003) 234.
[17] R. Fleischer, Nucl. Phys. B 659 (2003) 321.
[18] A.J. Buras, TUM-HEP-489-02, hep-ph/0210291.
[19] L. Wolfenstein, Phys. Rev. Lett. 51 (1983) 1945;
A.J. Buras, M.E. Lautenbacher, G. Ostermaier, Phys. Rev. D 50 (1994) 3433.
[20] J.D. Bjorken, Nucl. Phys. B (Proc. Suppl.) 11 (1989) 325;
M.J. Dugan, B. Grinstein, Phys. Lett. B 255 (1991) 583.
[21] M. Neubert, B. Stech, Adv. Ser. Direct. High Energy Phys. 15 (1998) 294.
[22] M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 591 (2000) 313;
C.W. Bauer, D. Pirjol, I.W. Stewart, Phys. Rev. Lett. 87 (2001) 201806.
[23] M. Neubert, Phys. Rep. 245 (1994) 259.
[24] M. Bauer, B. Stech, M. Wirbel, Z. Phys. C 34 (1987) 103.
[25] P. Ball, et al., CERN-TH/2000-101, hep-ph/0003238;
P. Ball, et al., in: CERN Report on Standard Model Physics (and more) at the LHC, CERN, Geneva, 2000,
p. 305.
[26] P. Harrison, H. Quinn (Eds.), The BaBar Physics Book, 1998, SLAC report 504.
[27] I. Dunietz, Phys. Rev. D 52 (1995) 3048.
[28] M. Beneke, A. Lenz, J. Phys. G 27 (2001) 1219.
[29] Y. Grossman, Phys. Lett. B 380 (1996) 99.
[30] K. Anikeev, et al., FERMILAB-Pub-01/197, hep-ph/0201071.
[31] BaBar Collaboration, B. Aubert, et al., BABAR-PUB-02-010, hep-ex/0211053;
Belle Collaboration, P. Krokovny, et al., Phys. Rev. Lett. 89 (2002) 231804.
[32] BaBar Collaboration, B. Aubert, et al., BABAR-CONF-03-015, hep-ex/0307036.
[33] E. Jenkins, M.J. Savage, Phys. Lett. B 281 (1992) 331.
[34] B. Grinstein, E. Jenkins, A.V. Manohar, M.J. Savage, M.B. Wise, Nucl. Phys. B 380 (1992) 369.
[35] UKQCD Collaboration, K.C. Bowler, et al., Nucl. Phys. B 619 (2001) 507.
[36] P. Ball, V.M. Braun, Phys. Rev. D 58 (1998) 094016.
[37] R. Fleischer, Eur. Phys. J. C 10 (1999) 299;
M. Gronau, J.L. Rosner, Phys. Lett. B 482 (2000) 71;
P.Z. Skands, JHEP 0101 (2001) 008.
[38] R. Fleischer, Phys. Lett. B 459 (1999) 306.

CP violation and matter effect for a variable earth

density in very long baseline experiments
Biswajoy Brahmachari a,d , Sandhya Choubey b , Probir Roy c
a Theory Group, Saha Institute of Nuclear Physics, AF/1 Bidhannagar, Kolkata 700 064, India
b INFN, Sezione di Trieste and Scuola Internazionale Superiore di Studi Avanzati, I-34014 Trieste, Italy
c Tata Institute of Fundamental Research, Homi Bhaba Road, Mumbai 400 005, India
d Department of Physics, Vidyasagar Evening College, 39, Sankar Ghosh Lane, Kolkata 700 006, India
Received 21 March 2003; accepted 5 August 2003
Abstract
The perturbative treatment of subdominant oscillation and matter effect in neutrino beams/superbeams, propagating over long baselines and being used to look for CP violation, is studied here for
a general matter density function varying with distance. New lowest order analytic expressions are
given for different flavour transition and survival probabilities in a general neutrino mixing basis and
a variable earth matter density profile. It is demonstrated that the matter effect in the muon neutrino
(antineutrino) flavour survival probability vanishes to this order, provided the depletion, observed for
atmospheric muon neutrinos and antineutrinos at super-Kamiokande, is strictly maximal. This result
is independent of the earth density profile and the distance L between the source and the detector.
In the general variable density case we show that one cannot separate the matter induced asymmetry
from a genuine CP effect by keeping two detectors at distances L1 and L2 from the source while
maintaining a fixed ratio L1 /E1 = L2 /E2 . This needs to be done numerically and we estimate the
asymmetry generated by the earth matter effect with particular density profiles and some chosen
parameters for very long baseline neutrino oscillation experiments.
1. Introduction
Finding CP violation in the neutrino sector [1] is a tantalizing goal waiting to be
attained [2] in forthcoming long baseline experiments with neutrino beams/superbeams
E-mail address: sandhya@he.sissa.it (S. Choubey).

doi:10.1016/j.nuclphysb.2003.08.009
484
B. Brahmachari et al. / Nuclear Physics B 671 (2003) 483497
and future ones at neutrino factories. The discovery and quantitative measurement of
such an effect will not only open a new window for physics beyond the standard model,
but may provide an insight into leptogenesis [3]. From solar neutrino studies and the
reactor experiment by KamLAND, we already know [4] the squared mass difference
and the angle of mixing between concerned mass eigenstate pair 1,2 to be m221
m22 m21 7 105 eV2 and sin2 12 0.3, respectively (see [5] and references therein).
Similarly, atmospheric neutrino studies [6] have revealed the corresponding parameters
for the other mass eigenstate pair 2,3 to be |m232| |m23 m22 | 3 103 eV2 and
sin2 23 0.5. Other reactor experiments [7] have, however, shown that the third possible
mixing angle 13 is very small, consistent with zero. These mysterious numbers have
increased the importance of the task of getting a quantitative handle on CP violation
likely to be associated with the MNS neutrino mixing matrix. This is sought [1,2] to be
achieved by measuring the difference between neutrino and antineutrino beams in the
muonic to electronic flavour transition probability which can be obtained from the general
expression

P (L, E) = P (0) (L) P (0) (L)
(1)
over a large distance L, E being the beam energy and , being the flavour indices. If
|m232|L/E is chosen to be O(1), the effect will be driven dominantly by oscillation.
A major practical problem, associated with the above task, is the occurrence of the
matter effect in neutrino flavour transitions studied in long baseline experiments. These
contribute to P (L) and induce a fake CP violation. The latter needs to be filtered
out, leaving only the genuine CP violation part. However, the matter effect is not merely
a background but deserves to be studied in its own right. In particular, it can yield
valuable information on the sign of m232 , on the MSW resonant enhancement of neutrino
oscillations and on the mixing between neutrinos of the first and the third generations.
It is important therefore to be able to compute the matter effect relevant to any given
experiment looking for CP violation. Several studies [2,811] have been conducted to this
end. Specifically, we shall use the formalism of Arafune et al. [11]developed to treat the
matter effect as well as the subdominant oscillation driven by m221 and 12 perturbatively
to the lowest order relative to the dominant oscillation driven by |m232 | and 23 . Thus
the evolution matrix element S (x) (x)| (0)
, , being flavour indices, could be
calculated to first order in m221 and a, where
E (x)

a(x, E) = 2 2 GF Ne E = 7.56 105 eV2
,
GeV g cm3
(2)
with Ne being the electron density and (x) the mass density of the earth, expressed as
a function of the path length x of the beam. The parameter was assumed in Ref. [11]
485
to be spatially uniform.1 This is not always a realistic assumption [9], especially for
future very long baseline experiments where nonuniformity in earths density profile
cannot be neglected. One possible approach [14] is to write the density function (x),
and correspondingly a(x, E), as an average constant plus a spatially fluctuating part
and expand the latter into Fourier modes, arguing that only the first few modes is
important for foreseeable experiments. There is, however, a large uncertainly [1517] in
the seismological knowledge of the latter; indeed, this feeds back into the average density
parameter. Under the circumstances, it is worthwhile makingwithin the lowest order
perturbative frameworkgeneral statements for an arbitrarily varying a(x, E) that can be
checked by experimental measurements. This is our aim here. Our approach is basically
analytic and the price to pay is to use lowest order perturbation theory. As will be pointed
out in our section on numerical studies, a lowest order perturbative result is often invalid
in very long baseline neutrino experiments. There are, however, sizable domains in the
(L, E) plane where it should be reliable. At least, the importance of the effects considered
is clearly brought out by our analytic considerations.
In this paper we reexamine the treatment of Ref. [11], assuming an arbitrary spatial
dependence in (x) and hence a(x, E), consistent with the approximation |a| |m232|.
We give extended versions of the formulae of Arafune et al., accommodating such a
nonuniform (x) and in a general neutrino mixing basis. Even with such an arbitrary earth
density profile, we find that the matter effect in the muon neutrino (antineutrino) flavour
()
()
survival probability P [ (0) (L)] vanishes if the flavour conversion of atmospheric

and , as observed in super-K, is truly maximal. However, the simple methods,
proposed in Ref. [11] to filter out the matter effect terms, are specific to the uniform earth
density assumption and do not extend to the variable density case in the lowest order of
perturbation theory. We show that, in general, it is not possible to separate the fake matter
induced asymmetry from the genuine CP effect by keeping two detectors at distances L1
and L2 from the source while keeping the ratio L1 /E1 = L2 /E2 = L/E fixed and taking
a linear combination of P [ (0) (L)] and P [ (0) (L)], as proposed for a
constant matter density profile to the lowest perturbative order in subdominant oscillation
and matter effects by Arafune et al. This separation has to be done numerically and in the
last part we compute the asymmetries generated by the earths matter effect for sample
earth matter density profiles keeping specific experimental possibilities in mind.
This paper is organized as follows. Section 2 contains an extension of the lowest order
perturbative calculation of the evolution operator to the variable density case. In Section 3
()
()
we calculate the muon (anti)neutrino flavour survival probability P [ (0) (L)].

()
This calculation is extended to the general flavour transition probability P [ (0)

()
(L)]
in Section 4. Our numerical studies are presented in Section 5. Section 6 contains

a summary of our results.
1 For instance, one could assume [12] a constant density equal to the average density of the PREM profile
[13].
486
2. Evolution operator formalism

The formulae for the amplitude and probability of the transition of (0) to (L) in
vacuum are2

m2i L
A (0) (L) = (L) (0) =
Ui e 2E Ui ,
(3a)
P (0) (L) = 4
Re

Ui Ui
Uj Uj

sin
m2ij L
4E
j >i
+2

j >i
m2ij L

Im Ui Ui
Uj Uj sin
,
2E
(3b)
where m2ij = m2i m2j and the neutrino mixing matrix U is defined in vacuum by
|
= Ui |i
, and i being flavour and mass eigenstate indices, respectively. More
generally, when matter is present, one can decompose the Hamiltonian as
H = H0 + H ,
(4)
where H0 is the unperturbed part containing m232 while H is the perturbation involving
m221 and a(x, E).
In the case of an x-dependent H , we cannot use the replacement procedure of Ref. [18],
but need to use the evolution operator formalism. Thus we write

(x) = S (x) (0) ,
(5)
where the operator S obeys the evolution equation
i
dS
= H0 S(x) + H (x)S(x),
dx
(6)
with
1
U
H0 =
2E
0
0
0

1
H (x) =
U
2E

0
0
0

0
U ,
0
m231

0
0
0
a(x, E) 0
2
0 m21 0 U +
0
0
0
0
0
0
0
(7a)
0
0
0

.
(7b)
To the lowest order of perturbation, (6) can be solved by [11]

S(x) S 0 (x) + S (x),
(8)
2 We shall use L to denote the baseline with respect to a measurement, x to denote any intermediate length
and s for a dummy variable in any integration being performed upto x.
487
with
S 0 (x) = eixH0 ,
(9a)
S (x) = eixH0 (i)
x
ds eisH H (s)eisH .
0
(9b)
We turn now to matrix elements in the flavour basis and work in the approximations
m221 |m232| and |a(S, E)| |m232|. First, it is convenient to define
1
g(v) m231 v, for any variable v.
4
Then we can explicitly write
ig(2x/E)

$
S 0 (x) = + U3 U3
e
1

$ ig(x/E)
= 2iU3 U3
e
sin g(x/E)
A (x/E)
(10)
(11)
for the unperturbed part. The lowest order expression for the perturbed part
x
S (x) = i

ds ei(xs)H0 H (s) esH0
can be rewritten, on using (7a), as

x
S (x) = i

ds Ui exp i diag 0, 0, g 2(x s)/E ii U$ i

$
H (s) Uj exp i diag 0, 0, g(2s/E) jj Uj
.
(12)
In the RHS of (12), H (s) has two additive parts, one constant and one depending on s:
H (s) = H 1 + H a (s),
(13)
with

0
0
0
1
2
H =
U 0 m21 0 U ,
2E
0
0
0

a(s, E) 0 0
1
Ha =
0
0 0 .
2E
0
0 0
1
(14a)
(14b)
Correspondingly, we can take

S (x) = S 1 (x) + S a (x) ,
(15)
488
where H 1 contributes to S 1 and H a to S a . Since H 1 does not depend on s, we may write

S 1 (x) = iUi Ui H 1 Uj Uj
x

ds exp i diag 0, 0, g(2(x s)/E) ii

exp i diag 0, 0, g(2s/E) jj
m221 x
U2 U2
iB (x/E)m221,
2E
where we have used the identity
= i
(16)

m221
i2 j 2 .
Ui H 1 Uj =
2E
Turning to S a (x) , it is convenient to use the result

a(s, E)
U1i U1j .
Ui H a (s) Uj =
2E
The employment of (17) enables us to write S a (x) as
a
(x, E) =
S
Ui Uj
U1i
U1j
2E
x
(17)

(2x 2s)i3 + 2sj 3
a(s, E)
ds exp ig
E
Ui Uj
=
U1i
U1j ija (x, E),
2E
(18)
with
ija (x, E) = i3 j 3 eig(2x/E)
x
x
ds a(s, E) + (1 i3 )(1 j 3 )
x
+ (1 i3 )j 3
ds a(s, E)
0
ds a(s, E)eig(2s/E)
x
+ i3 (1 j 3 )
ds a(s, E)eig(2(xs)/E).
(19)
On using (19) in (18), we obtain

a
S
(x, E) =
i
1 1
2E
x
ds a(s, E)
0
1 U3 U13
E
x
0

ds a(s, E)ei[g(x/E)g(s/E)] sin g(x/E) g(s/E)
1 U3
U13
E
x
489

ds a(s, E)eig(s/E) sin g(s/E)
i
U U3 |U13 |2 eig(x/E)R(x, E)
2E 3
iGa (x, E).
(20)
In (20) we have introduced the real function R(x, E):

R(x, E) e
ig(x/E)
x

ds a(s, E) 1 eig(2s/E)
x
+ eig(x/E)

ds a(s, E) 1 eig(2s/E)
x
2

ds a(s, E) cos g(x/E) cos g(x/E) g(2s/E) .
(21)
Note that R(x, E) becomes a function of x/E for a constant (x), not otherwise.
0 + S1 + Sa
Finally, the flavour matrix element of the evolution operator S = S
can be expressed as
S (x, E) = A (x/E) iB (x/E)m221 iGa (x, E),
(22)
Ga (x, E)
are given by (11), (16) and (20), respectively.

where A (x/E), B (x/E) and
We can now calculate the transition probability P to the lowest order in a and m221 in
the approximations m221 |m231|, |a(s, E)| |m231| stated already. We obtain
0
+ P
,
P = P
(23)
with

0
= A A + 2 Im A B ,
P

a
P
= 2 Im A Ga .
(24)
(25)
3. The flavour survival probability (0) (L)

Using (11), (16), (20) and choosing = = , we have

A (x/E) = 1 2i|U3 |2 eig(x/E) sin g(x/E) ,
x
B (x/E) = |U2 |2
,
2E
|Ue3 |2 |U3 |2 ig(x/E)
Ga (x, E) =
R(x, E),
e
2E
(26a)
(26b)
(26c)
490
with R(x, E) as defined in (21). In the matter free case, a(s, E) = 0 and replacing x by L
we obtain

0
(L/E) = 1 4|U3 |2 sin2 g(L/E) + 4|U3 |4 sin2 g(L/E)
P

L
+ |U2 |2 |U3 |2 sin 2g(L/E) m221.
(27)
E
If we consider this transition to be overwhelmingly driven by a two flavour oscillation,
as done in the super-K analysis [6], we can ignore the third RHS term in (27) to see
0 (L, E), i.e., the flavour transformation and hence mixing for a
that the depletion 1 P
muonic neutrino or antineutrino, would be maximal for |U3 | = 1/ 2. Next, we consider

propagation in matter and keep a and hence M(L). Then we have
0
a
P (L, E) = P
(L/E) + P
(L, E),
(28)
where

|Ue3 |2 |U3 |2
2|U3 |2 1 sin g(L/E) R(L, E).
E
The asymmetry P (L, E), defined in (1), is thus
a
P
(L, E) =
P (L, E) =
(29)

|Ue3 |2 |U3 |2
2|U3 |2 1 sin g(L/E)
E

R + (L, E) R (L, E)

2|Ue3 |2 |U3 |2
2|U3 |2 1 sin g(L/E) R + (L, E).
E
(30)
In (30) R (L, E) is obtained from (21) by using a(s, E) |a(s, E)|, respectively, so that
R (L, E) = R + (L, E). Thus we see that the matter effect
contribution to the muonic
flavour survival probability P (x) vanishes if |U3 | = 1/ 2. Moreover, even in matter
and to the lowest order of perturbation, any nonzero asymmetry P , detected from the
()
()
muon (anti)neutrino flavour survival probabilities P ( ) will signal a deviation

from the condition for the strictly maximal mixing of atmospheric neutrinos at super-K.
This canbe used in future to sensitively probe any deviation of |U3 | from its maximal
value 1/ 2.
4. General oscillation probability in matter: (0) (L)
From (24) and (26a), (26b) we have

0
P
(L/E) 1 4|U3 |2 sin2 g(L/E) + 4|U3 |2 |U3 |2 sin2 g(L/E)

Lm221
+
Re U3 U3 U2 U2
sin 2g(L/E)
E

2

2 Im U3
sin g(L/E) ,
U3 U2 U2
(31)
491
while (25) and (26a), (26c) lead us to the expression

a
(L, E) 2 Im A (L/E)Ga (L, E)
P

1
=
1 |U3 |2 sin2 g(L/E)
E
L

ds a(s, E) sin g(L/E) g(s/E)

1
|U3 |2 |Ue3 |2 sin g(L/E) R(L, E)
E

1
|U3 |2 |U3 |2 2|Ue3 |2 (1 + 1 ) sin g(L/E) R(L, E),
E
with R(L, E) substituted from (21).
We can now discuss two distinct cases.
+
(32)
Case 1: =
= = e
In this case the matter-independent and matter-dependent transition probabilities, cf.

0
P
(L/E) = 1 4|U3|2 1 |U3 |2 sin2 g(L/E)

m221 L
|U2 |2 |U3 |2 sin 2g(L/E) ,
(33a)
E

1
a
(L, E) = |U3 |2 |Ue3 |2 2|U3|2 1 sin g(L/E) R(L, E).
P
(33b)
E
= =e
0 (L) is the same as the RHS of (33a) with = e, but the matterThe expression for Pee
dependent part is
+
a
(L, E) =
Pee

1
|Ue3 |2 sin2 g(L/E)
E
L

ds a(s, E) sin g(L/E) k(s/E)/2

1
|Ue3 |4 sin g(L/E) R(L, E)
E

2
+ |Ue3 |4 |Ue3 |2 1 sin g(L/E) R(L, E).
E
(34)
Case 2: =
Now the matter-independent transition probability is given from (31) by

0
P
(L/E) = 4|U3 |2 |U3 |2 sin2 g(L/E)

1
2m221L
Im sin2 g(L/E) Re sin 2g(L/E) ,
E
2
(35)
492
with
U2
U3 U2 U3
.
(36)
For the matter-dependent part, (32) leads to three cases.

= e, = e
a
(L, E) =
P

2
|U3 |2 |U3 |2 |Ue3 |2 sin g(L/E) R(L, E).
E
(37)

1
|U3 |2 |Ue3 |2 2|Ue3 |2 1 sin g(L/E) R(L, E).
E
(38)

1
|Ue3 |2 |U3 |2 2|Ue3 |2 1 sin g(L/E) R(L, E).
E
(39)
= e, = e
a
Pe
(L, E) =
= e, = e
a
Pe
(L, E) =
Both the CP violating part, proportional to Im in (35), and the matter dependent
a
part P
change sign when one goes from neutrinos and antineutrinos. However, the
former is a function of x/E, while the latter involves bothx/E and x in an unknown
way for a general earth matter density function (x) = (2 2 GF Ne E)1 a(x, E). Only
for (x) = const, can all the transition probabilities become functions of L/E and the
matter dependent part can be eliminated, to the lowest order of perturbation, by [11] taking
L1 (L2 L1 )1 [P (L2 ) P (L1 )]L/E fixed , but this procedure is in general invalid
for a spatially varying a(x, E).
5. Numerical estimate of matter induced asymmetry

In this section we present some numerical results for the matter contribution to the
transition probabilities using the formalism developed in this paper. We choose realistic
neutrino beam energies and both realistic and notional detector baselines and discuss the
conditions for the validity of our perturbative calculation. For numerical studies we have
used the Preliminary Reference Earth Model (PREM) [13] and another Earth model
ak135-F [19]. For the validity of perturbation theory for all intermediate values of s we
must satisfy the conditions
m231 s
a(s, E)s

,
4E
4E
m221s
a(s, E)s
1,
1,
2E
2E

m221s
m231 s
sin2
,
sin2
4E
4E
(40)
(41)
(42)
where 0 s L. Note that conditions in Eq. (41) are required for the expansion of S(x)
in Section 2. Neutrino oscillation experiments either look for disappearance of the initial
493
Fig. 1. The comparison of the matter term against the m231 term as a function of the neutrino baseline and for
different neutrino energies. The matter term is independent of energy and is shown by the black solid line.
neutrino beam or for the appearance of a different flavour in the final neutrino beam at
the detector. Both these effects are maximal when sin2 (m231L/4E) 1 corresponding to
a peak in the transition probability. The Super-Kamiokande atmospheric neutrino data
demands a m231 3 103 eV2 [6] while the mass squared difference associated with
the solar neutrino oscillation has now been confirmed to be around m221 7 105 eV2
by the KamLAND experiment [5]. Thus for all realistic experimental scenarios where the
beam energy is tuned to an oscillation maximum for m231 , oscillations driven by the
solar scale is always subdominant compared to those driven by the atmospheric scale, i.e.,
the condition (42) and the second of conditions (41) are satisfied. For the validity of our
approximation (42) we will therefore always confine ourselves to values of

L/Km
O 102 103 .
E/GeV
(43)
To check our other approximation concerning the matter potential we present in Fig. 1
the comparison of the strength of the matter term vis-a-vis the term involving m231 . The
solid line shows a L L/4E, where a L is the average matter potential for a given neutrino
baseline L defined as
1
a L =
L
L
a(s, E) ds.
(44)
For each experimental baseline L, a L is computed using the PREM model [13]. Since
a(s, E) goes linearly with the neutrino energy E (cf. Eq. (2)), the matter potential term
a L L/4E is independent of the energy of the neutrino beam. It is only a function of
494
Table 1
The magnitude of P a for some specific values of E and L and for the different oscillation channels. Values of L
in km are shown in parentheses after P a . Also shown is the value of the function R(L, E) defined in Eq. (32).
We have taken m231 = 3 103 eV2 , m221 = 7 105 eV2 and |U3 |2 = 0.5. We present results for three
different values of |Ue3 |
E/GeV
1
1
1
E/GeV
3
3
3
E/GeV
5
5
5
E/GeV
7
7
7
Ue3
|Pa (732)|
a (732)|
|Pe
a (732)|
|Pe
0.001
0.01
0.1
5.32 105
5.21 103
5.32 105
5.21 103
Ue3
0.001
0.01
0.1
Ue3
0.001
0.01
0.1
Ue3
0.001
0.01
0.1
5.32 107
5.32 107
5.32 107
5.32 105
5.21 103
1.54
1.54
1.54
|Pae (2500)|
R in GeV
a (2500)|
|Pe
a (2500)|
|Pe
2.16 105
2.12 103
2.16 105
2.12 103
2.16 107
2.16 107
2.16 105
2.12 103
2.16 107
2.16 105
2.08 103
19.42
19.42
19.42
|Pae (3000)|
R in GeV
|Pa (3000)|
a (3000)|
|Pe
a (3000)|
|Pe
5.72 104
5.61 102
5.72 104
5.61 102
5.72 106
5.72 106
5.72 106
5.72 104
5.61 102
5.72 106
5.72 104
5.50 102
37.91
37.91
37.91
|Pae (3500)|
R in GeV
|Pa (3500)|
a (3500)|
|Pe
a (3500)|
|Pe
7.03 104
6.89 102
7.03 104
6.89 102
7.03 104
6.89 102
7.03 106
7.03 106
R in GeV
5.32 107
5.32 105
5.11 103
|Pa (2500)|
2.16 107
|Pae (732)|
7.03 106
7.03 106
7.03 104
6.75 102
52.10
52.10
52.10
the average matter density and hence of the baseline L. The dotted, dashed and longdashed lines show m231L/4E for E = 1, 5 and 10 GeV, respectively. We note that for
neutrinos with E = 3 GeV the approximation given in Eq. (40) works very well for all
baselines shown. However, for neutrino energies equal to or in excess of 10 GeV our
perturbative expansion for the matter term works only for smaller L and breaks down
at higher baselines.
In Table 1 we give a summary of the magnitude of P a for some specific cases. In
Fig. 2 we show P a as a function of the baseline length L for the e transition
and with E = 3 GeV. We have checked that very similar behaviour is exhibited by P a
for the transitions e and . The solid and the dashed lines show the P a
obtained using our formalism for a varying density matter profile for the earth. The dashed
line is for the PREM model [13] while the solid line corresponds to the ak135-F model [19]
for the earth matter profile. The dotted line gives the corresponding values for a constant
density earth with = 3.28 g cm3 . The values of parameters used for generating the
figure and the table are shown in the captions. Comparison of the solid and/or dashed
lines with the constant density dotted line shows that the variation in the density profile
can lead to a change in the oscillation probability. Even at L = 3000 km, we note about
10% difference in P a between the constant density case and the case corresponding to
PREM or ak135-F. At smaller L we note a small difference between P a corresponding to
PREM and ak135-F, which just reflects the fact that ak135-F has more density fluctuation
495
Fig. 2. The CP asymmetry induced by matter effect (P a ) as a function of the baseline L for the e
oscillation channel with E = 3 GeV. The dashed line and the solid line give the P a calculated using
the PREM model and the ak135-F model respectively, while the dotted line is for a constant density earth
with = 3.28 g cm3 . We have taken m231 = 3 103 eV2 , m221 = 7 105 eV2 , |U3 |2 = 0.5 and
|Ue3 |2 = 0.01.
at smaller L than PREM.3 We see that the effect of density variation begins to be significant
for L > 5000 km. A caveat is that out formulae are not quantitatively reliable if L and E
are such that either (40) or the first of conditions (41) breaks down. (The second condition
of (41) is always safely obeyed.) Such is evidently not the case in most of Fig. 1. Work
is in progress to improve upon the first order perturbation theory in a(s, E) so that the
dependence on these conditions is reduced.
6. Conclusions
With the confirmation of neutrino flavour oscillations both in the atmospheric as well
as the solar neutrino sectors, the focus now has shifted to the precise determination of
the oscillation parameters involved. This will be possible in the currently planned and
future long and very long baseline experiments involving conventional superbeams and
neutrino factories. The determination of CP violation in the lepton sector and measurement
of the CP phase will be the most interesting as well as the most challenging goal of
3 It has been pointed out in [17] that even the so-called realistic earth models like PREM and ak135-F neglect
the local density fluctuations which can have an impact on the final oscillation probability and hence on the
CP sensitivity of a specific experiment. However, our expressions for the oscillation probabilities in varying
density matter are completely general and can be applied to any earth matter density profile, local variations
notwithstanding.
496
these experiments. The matter effect induces a fake CP asymmetry even if there is no
intrinsic CP violation in the neutrino sector. Thus a knowledge of matter effect is extremely
important in all neutrino CP violation studies. The matter effect also helps in ascertaining
the sign of the atmospheric neutrino mass squared difference m231 which has a bearing on
the neutrino mass hierarchy with profound theoretical and phenomenological implications.
The effect of earth matter on the survival and transition probabilities were studied earlier
and ways to disentangle the real CP from the fake CP due to matter were discussed.
However most of these studies assumed a constant density of the earth matter.
In this paper we have used the evolution operator formalism to derive the most
general lowest order perturbative expressions for neutrino flavour survival and transition
probabilities in varying density matter. We have worked in a perturbative scheme, where
the oscillation driven by m231 is assumed to be much larger than those driven by m221
and the matter potential a(s, E). We have shown to the lowest nontrivial order that for
a maximal mixing in the sector, the matter effect in the survival probability P
vanishes identically. We have made numerical checks for the validity of our perturbation
approximation and conclude that as long as E 10 GeV, our approximation holds, at least
upto a baseline of L 4000 km. Finally, we have compared the results obtained with the
PREM and ak135-F density profiles with that for a constant density earth matter. Work is
in progress to extend the range of validity of our theory to longer baselines.
Acknowledgements
B.B. acknowledges hospitality of Department of Theoretical Physics, TIFR and of
Theory Division, CERN. S.C. thanks S.T. Petcov for discussions. P.R. is indebted to Theory
Group, SLAC and to Santa Cruz Institute for Particle Physics, UCSC, for their hospitality.
References
[1] N. Cabibbo, Phys. Lett. B 72 (1978) 333;
V.D. Barger, K. Whistant, R.J.N. Philips, Phys. Rev. Lett. 45 (1980) 2084;
S. Pakvasa, in: Proc. XXth Int. Conf. on High Energy Physics, Madison, 1980, p. 1164;
S.M. Bilenky, F. Niedermeyer, Sov. J. Nucl. Phys. 34 (1981) 606, Yad. Fiz. 34 (1981) 1091 (in Russian).
[2] A. de Rujula, M.B. Gavela, P. Hernandez, Nucl. Phys. B 547 (1999) 21, hep-ph/9911390;
C. Albright, et al., hep-ex/0008064;
M. Apollonio, et al., hep-ph/0210192;
Y. Itow, et al., hep-ex/0106019;
D. Ayres, et al., hep-ex/0210005;
M. Lindner, et al., hep-ph/0209083;
K. Dick, M. Freund, M. Lindner, A. Romanino, Nucl. Phys. B 562 (1999) 29, hep-ph/9903308;
A. Cervera, A. Donini, M.B. Gavela, J.J. Gomez Cadenas, P. Hernandez, O. Mena, S. Rigolin, Nucl. Phys.
B 579 (2000) 17, hep-ph/0002108;
A. Cervera, A. Donini, M.B. Gavela, J.J. Gomez Cadenas, P. Hernandez, O. Mena, S. Rigolin, Nucl. Phys.
B 593 (2001) 731, Erratum;
V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. D 65 (2002) 073023, hep-ph/0112119;
S. Geer, hep-ph/0210113;
O. Yasuda, hep-ph/0209127.
497
[3] M. Fukugita, T. Yanagida, Phys. Lett. B 175 (1986) 45;

G. Branco, T. Morozumi, B. Nobre, M.N. Rebelo, Nucl. Phys. B 617 (2001) 475;
A. Joshipura, E. Pascos, W. Rodejohann, JHEP 0108 (2001) 029;
W. Buchmller, D. Wyler, Phys. Lett. B 521 (2001) 291;
T. Endoh, T. Morozumi, A. Purwanto, Nucl. Phys. B (Proc. Suppl.) 111 (2002) 291;
J. Ellis, M. Raidal, Nucl. Phys. B 643 (2002) 229;
S. Davidson, A. Ibarra, hep-ph/0206304;
G.C. Branco, R. Gonzalez Felipe, F.R. Joaquim, M.N. Rebelo, Nucl. Phys. B (Proc. Suppl.) 111 (2002) 303;
P. Frampton, S. Glashow, T. Yanagida, hep-ph/0208157;
T. Endoh, S. Kaneko, S.K. Kang, T. Morozumi, M. Tanimoto, Phys. Rev. Lett. 89 (2002) 231601.
[4] Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 89 (2002) 011301, nucl-ex/0204008;
Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 89 (2002) 011302, nucl-ex/0204009;
S. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Lett. B 539 (2002) 179, hep-ex/0205075;
K. Eguchi, et al., KamLAND Collaboration, Phys. Rev. Lett. 90 (2003) 021802, hep-ex/0212021.
[5] A. Bandyopadhyay, S. Choubey, R. Gandhi, S. Goswami, D.P. Roy, hep-ph/0212146;
A. Bandyopadhyay, S. Choubey, S. Goswami, hep-ph/0302243;
V. Barger, D. Marfatia, hep-ph/0212126;
G.L. Fogli, E. Lisi, A. Marrone, D. Montanino, A. Palazzo, A.M. Rotunno, hep-ph/0212127;
M. Maltoni, T. Schwetz, J.W. Valle, hep-ph/0212129;
J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pena-Garay, hep-ph/0212147;
P.C. de Holanda, A.Y. Smirnov, hep-ph/0212270.
[6] Y. Fukuda, et al., Phys. Rev. Lett. 81 (1998) 1562;
Y. Fukuda, et al., Phys. Rev. Lett. 85 (2001) 3999;
T. Kajita, Y. Totsuka, Rev. Mod. Phys. 73 (2001) 85.
[7] M. Applonio, et al., Phys. Lett. B 466 (1999) 415;
F. Boehm, et al., Phys. Rev. D 64 (2001) 112001.
[8] J.J. Gomez-Cadenas, et al., hep-ph/0105297;
M. Akoi, K. Hagiwara, N. Okamura, hep-ph/0208223.
[9] I. Mocioiu, R. Shrock, Phys. Rev. D 62 (2001) 0153017;
I. Mocioiu, R. Shrock, JHEP 0111 (2001) 050;
V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. D 65 (2002) 073023;
T. Ohlsson, H. Snellman, Eur. Phys. J. C 20 (2001) 507.
[10] M. Freund, M. Lindner, S.T. Petcov, A. Romanino, Nucl. Phys. B 578 (2000) 27.
[11] J. Arafune, M. Koike, J. Sato, Phys. Rev. D 56 (1997) 3093;
J. Arafune, M. Koike, J. Sato, Phys. Rev. D 60 (1999) 119905, Erratum.
[12] M. Freund, T. Ohlsson, Mod. Phys. Lett. A 15 (2000) 867.
[13] A.M. Dziewonski, D.L. Anderson, Phys. Earth Planet. Inter. 25 (1981) 297;
S.V. Panasyuk, Reference Earth Model (REM) webpage http://cfauves5.harvrd.edu/lana/rem/index.html.
[14] M. Koike, J. Sato, Mod. Phys. Lett. A 14 (1999) 1297;
T. Ota, J. Sato, Phys. Rev. D 63 (2000) 093004;
B. Jacobsson, T. Ohlsson, H. Snellman, W. Winter, Phys. Lett. B 532 (2002) 259;
B. Jacobsson, T. Ohlsson, H. Snellman, W. Winter, hep-ph/0209147.
[15] T. Ota, J. Sato, hep-ph/0211095.
[16] L.Y. Shan, B.L. Young, X.M. Zhang, Phys. Rev. D 66 (2002) 053012.
[17] L.Y. Shan, Y.F. Wang, C.G. Yang, X. Zhang, F.T. Liu, B.L. Young, hep-ph/0303112.
[18] K. Kimura, A. Takamura, H. Yokomakura, hep-ph/0203099.
[19] B.N.L. Kennett, E.R. Engdahl, R. Buland, Geophys. J. Int. 122 (1995) 108, webpage http://wwwrses.anu.
edu.au/seismology/ak135/intro.html.

Erratum
Erratum to: Inflationary cosmology from STM

theory of gravity
[Nucl. Phys. B 660 (2003) 389]
Mauricio Bellini
CONICET, Departamento de Fsica, FCEyN, Universidad Nacional de Mar del Plata,
Funes 3350, (7600) Mar del Plata, Argentina
Received 18 August 2003
PACS: 98.80.Cq; 04.20.Jb; 11.10.Kk
Because of an error in the calculation, Eqs. (48) and (50) are incorrect. The equation of
state (48) should be given by

3p 2
p =
t .
3p
On the other hand, Eq. (50) correctly written, is

t
2
.
p=
3 t + p
doi of original article: 10.1016/S0550-3213(03)00234-7.

E-mail address: mbellini@mdp.edu.ar (M. Bellini).
doi:10.1016/j.nuclphysb.2003.08.028

AUTHOR INDEX B671
Armoni, A.
Arutyunov, G.
B671 (2003) 67
B671 (2003) 3
Bellini, M.
Boulton, L.
Brahmachari, B.
B671 (2003) 501

B671 (2003) 343
B671 (2003) 483
Choubey, S.
Contino, R.
B671 (2003) 483

B671 (2003) 148
Datta, A.
de Forcrand, P.
de Wit, B.
Dominici, D.
B671 (2003) 383

B671 (2003) 103
B671 (2003) 175
B671 (2003) 243
Fleischer, R.
Frolov, S.
B671 (2003) 459

B671 (2003) 3
Garca del Moral, M.P.

Giedt, J.
Grzadkowski, B.
Gunion, J.F.
B671 (2003) 343

B671 (2003) 133
B671 (2003) 243
B671 (2003) 243
Hatsuda, M.
Herger, I.
B671 (2003) 217

B671 (2003) 175
Iso, S.
Kim, N.
Klose, T.
Kratochvila, S.
Kriz, I.
Ling, F.-S.
B671 (2003) 383
Masina, I.
B671 (2003) 432
Nomura, Y.
B671 (2003) 148
Parkhomenko, S.E.
Plefka, J.
Pomarol, A.
B671 (2003) 325

B671 (2003) 359
B671 (2003) 148
Ramond, P.
Restuccia, A.
Rey, S.-J.
Roy, P.
Russo, J.
B671 (2003) 383

B671 (2003) 343
B671 (2003) 95
B671 (2003) 483
B671 (2003) 3
Samtleben, H.
Sato, M.
Shifman, M.
B671 (2003) 175

B671 (2003) 293
B671 (2003) 67
Toharia, M.
Tseytlin, A.A.
Tsuchiya, A.
B671 (2003) 243

B671 (2003) 3
B671 (2003) 293
B671 (2003) 217
Umetsu, H.
B671 (2003) 217
B671 (2003) 359

B671 (2003) 359
B671 (2003) 103
B671 (2003) 51
Vaidya, S.
B671 (2003) 401
Ydri, B.
Yee, J.-T.
B671 (2003) 401

B671 (2003) 95
0550-3213/2003 Published by Elsevier B.V.

doi:10.1016/S0550-3213(03)00828-9

Nucl - Phys.B v.671

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.671

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 671 (2003) 350

Spinning strings in AdS5 S 5 and integrable systems

D-14476 Golm, Germany

Instituci Catalana de Recerca i Estudis Avanats (ICREA), Spain

E-mail addresses: agleb@aei-potsdam.mpg.de (G. Arutyunov), frolov@mps.ohio-state.edu (S. Frolov),

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

1. Introduction and summary

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

not satisfied for at least

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

2. Reduction of O(6) sigma-model to the Neumann system

Here XM , M = 1, . . . , 6 and YM , M = 0, . . . , 5 are the embedding coordinates of R 6

Y1 + iY2 = sinh sin ei ,

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

As discussed in [1], to have a consistent semiclassical string state interpretation of these

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

It describes an (n = 3)-dimensional harmonic oscillator constrained to remain on a unit

The canonical momenta conjugate to xi are

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

i.e., we are interested in periodic version of the Neumann model.

Assuming w1 < w2 < w3 we can define the range of a (a = 1, 2) as

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

and the potential U is very simple

Here the function P ( ) is a 5th-order polynomial

In terms of variables bi the Hamiltonian (2.15) reads as

In this case (2.30) implies that

The spins Ji = Ji in (2.10) expressed in terms of 1 , 2 satisfy the following

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

3. Folded string solutions

xi (0) = xi () = 0,

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

the vanishing of the following integral (cf. (2.22))

d sin cos sin .

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

Here we introduced the parameters

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

3.2. System of equations for the leading correction to the energy

2I21 = 32 I10 + I11 ,

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

Making the change of integration variable z w22 (w22 b1 )1 z for I1 , and z

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

Then Eq. (3.36) can be written as

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

In the two-spin case with J1 = J2 J , J3 = 0 we know that [4]

Using these values of the parameters we find for the energy

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

zz2 (z)(z2 (z) z) dz

11 This coefficient has a simple origin: f

= (1/ 2 )(2t 1)[K(t)]2 , where t = 0.826115 is the (unique)

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350

found in the expansion in powers of J3 /Jtot are, respectively, E 2

E = 0.94. They agree with the results in Table 1.

4. Three-spin string solutions of circular type

As was discussed in Section 3, to describe a folded string we should consider b1 lying in

w12 < b2 < w22 ,

w12 < b1 < w22 ,

G. Arutyunov et al. / Nuclear Physics B 671 (2003) 350