Nucl - Phys.B v.668

Nuclear Physics B 668 (2003) 376
www.elsevier.com/locate/npe
Master integrals with one massive propagator for

the two-loop electroweak form factor
U. Aglietti a , R. Bonciani b,1
a Dipartimento di Fisica Universit di Roma La Sapienza and INFN Sezione di Roma, I-00185 Roma, Italy
b Physikalisches Institut, Albert-Ludwigs-Universitt Freiburg, D-79104 Freiburg, Germany
Received 15 April 2003; accepted 8 July 2003
Abstract
We compute the master integrals containing one massive propagator entering the two-loop
electroweak form factor, i.e., the process f f X, where f f is an on-shell massless fermion pair
and X is a singlet particle under SU(2)L U (1)Y , such as a virtual gluon or an hypothetical Z . The
method used is that of the differential equation in the evolution variable x = s/m2 , where s is the
c.m. energy squared and m is the mass of the W or Z bosons (assumed to be degenerate). The 1/
poles and the finite parts are computed exactly in terms of one-dimensional harmonic polylogarithms
of the variable x, H (w;
x), with = 2 D/2 and D the spacetime dimension. We present largemomentum expansions of the master integrals, i.e., expansions for |s|
m2 , which are relevant for
the study of infrared properties of the Standard Model. We also derive small-momentum expansions
of the master integrals, i.e., expansions in the region |s| m2 , related to the threshold behaviour of
the form factor (soft probe). Comparison with previous results in the literature is performed finding
complete agreement.
2003 Elsevier B.V. All rights reserved.
PACS: 11.15.Bt; 12.15.Lk
Keywords: Feynman diagrams; Multi-loop calculations; Vertex diagrams; Electroweak Sudakov
E-mail addresses: ugo.aglietti@roma1.infn.it (U. Aglietti), roberto.bonciani@physik.uni-freiburg.de

(R. Bonciani).
1 This work was supported by the European Union under contract HPRN-CT-2000-00149.
0550-3213/$ see front matter 2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2003.07.004
U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376
1. Introduction
Present day and planned high-energy experiments often require accurate perturbative
calculations. While in quantum-choromodynamics (QCD) higher order computations are
available for various processes, in the electroweak (EW) case progress is more difficult
because of the presence of the large, nonnegligible, masses of W , Z and Higgs bosons.
A full calculation of two-loop electroweak corrections to a typical process such as e+ e
+ is still outside present-day technology. It involves indeed the computation of a
large number of massive two-loop box diagrams having thresholds at s = 0, m2 , 4m2
and s = 9m2 , assuming m mW mZ mH . To have an explicit estimate of the size
of two-loop electroweak corrections, we consider then the model process
f (p1 ) + f(p2 ) X(q),
(1)
where f (f) is a massless fermion (antifermion) on its mass-shell, p12 = 0 (p22 = 0), and X
is any particle without electromagnetic or weak charge, i.e., a singlet under the electroweak
gauge group SU(2)L U (1)Y , such as a (time-like) gluon or an hypothetical Z . Eq. (1)
also represents the simplest process containing electroweak double logarithms, considered
for the first time, as far as we know, in [1]. Reaction (1) is the electroweak analog of the
QCD processes involving a color-neutral probe, such as DrellYan production of vector

bosons or muon pairs: qq ( ) W, Z, + or Higgs production by gluongluon fusion:
gg H .
At two-loop level, the annihilation in Eq. (1) involves the emission of a pair of virtual
bosons among , W , Z and H s. In the case of double photon emission, no mass is present
in the loops and we recover the known results for the QED or QCD amplitudes. The latter,
for dimensional reasons, are of the form

(massless amp.) = G()
s
2
2

s 4nd = a 4nd
2
a
2
G()x 4nd 2 ,
(2)
where s is the c.m. energy squared, G() is a function having a Laurent expansion in
= 2 D/2 (with D the spacetime dimension), is the mass-scale of dimensional
regularization and nd is the number of scalar propagators. We have defined
x =
s
,
m2
(3)
where m is the mass of the W , Z or Higgs bosons.2 On the other hand, if W , Z or H s are
also emitted, the dependence on x is not factorized in a single power and the amplitude has
the more general form

(massive amp.) = a 4nd
2
a
2
G(x; ).
2 The minus sign in the definition of x is inserted for convenience.
(4)
The amplitudes involving a single massive propagator, i.e., a single W or Z emission,

have thresholds only at s = 0 and s = m2 . As a consequence of this simple structure, these
amplitudes can be exactly computed in terms of one-dimensional harmonic polylogarithms
[2,3] (a generalization of the Nielsens polylogarithms [4,5]) of argument x, H (w,
x),
with weight up to w = 4 included. In this paper we present the evaluation of the
master integrals entering the above-mentioned amplitudes, leaving to a future work the
more complicated amplitudes with two and three massive propagators, involving also the
threshold s = 4m2 . Since external fermions are massless and are taken on their mass-shell,
infrared divergences (soft and collinear) occur in the amplitudes. When an intermediate
vector boson is radiated, its mass m acts as an infrared regulator and mass logarithms
do appear in four dimensions, i.e., terms of the form logk (s/m2 ) with k up to two
included. Infrared divergences related to photon emission instead are not regulated by
any physical parameter. Our computation is done within the dimensional regularization
scheme [6], in which the latter manifest themselves as poles in 1/. These poles are
unphysical and, in the physical cross-section, are canceled by the corresponding ones
appearing in the real photon emission contributions or are factorized into QED structure
functions. Ultraviolet divergencesrelated to short-distance behaviour and leading to
additional 1/ polesare also regulated within dimensional regularization and can be
subtracted with ordinary renormalization prescriptions, such as, for example, the MS
scheme.
The computation of the form factor is naturally divided in two steps.
The first one involves the reduction of the original tensor diagrams, given in Figs. 1
and 2, into a minimal set of scalar amplitudes, the so-called master integrals. This step is
discussed in Section 2 and involves in turn three different operations: (a) the decomposition
of tensor amplitudes into scalar amplitudes by means of projectors on invariant form
factors, see Section 2.1; (b) the transformation to linearly independent scalar amplitudes by
means of scalar product rotations, as discussed in Section 2.2; (c) the reduction of all the
independent scalar amplitudes generated to a small subset of themthe so-called master
integralsby means of the integration-by-parts identities [7] and other symmetry relations,
as discussed in Section 2.3.
The second step involves the explicit evaluation of the master integrals, which we
perform in all the massive cases with the method of differential equations on the external
kinematical invariants [814]. A set of special functions of the polilogarithm kind is
needed to represent the pole and finite parts of the master integrals. We choose the
basis of the (one-dimensional) harmonic polylogarithms [2,3]. This step is described in
Section 3.
In Section 4 we list the results for the pole and finite parts of the master integrals
containing up to five denominators included. In Section 5 we give the expressions of all the
scalar integrals containing six denominators. The planar topologies are reducible while the
crossed ladderstwo amplitudesare master integrals. We decided to list them together
because all the six-denominator amplitudes are interesting by themselves for reference.
We also compare with previous results in the literature. Section 6 is devoted to the large
momentum expansion |s|
m2 of all the six-denominator amplitudes. As already said,
this expansion is relevant for the study of asymptotic properties of multiple electroweak
radiation. Section 7 deals with the small momentum expansioni.e., the expansion for
|s| m2 of all the six-denominator amplitudes. In the conclusions we summarize the

results obtained and we discuss future perspectives. To not bore the reader with too
many technical details or formulaswhich are anyway worth reportingwe added a few
appendices. Appendix A contains the expressions of all the denominators entering the twoloop amplitudes. In Appendix B we give the results of the one-loop amplitudes which enter
the factorized two-loop diagrams, i.e., the diagrams in which the two loops do not have
common propagators. Finally, in Appendix C, we list the results for all the independent
topologies of Figs. 5, 6 and 7 which are not master integrals but can be useful for general
reference.
2. The reduction to master integrals

All the two-loop vertex diagrams containing six denominators with at most one massive
propagator are reported in Figs. 1 and 2.3 The graphical conventions we use are the
following. A massless propagator is represented by a wavy internal line, while a massive
propagator is denoted by a straight segment. An external line carrying the light-cone
momentum p1 or p2 is represented by an external wavy line, while the external particle
X carrying the general momentum q is depicted as a straight external line. This set of
conventions is clearly a generalization of the well-known QED ones. In Fig. 1 we consider
the basic three topologies with the various mass configurations:
(a), (b), (c) and (d) are ladder diagrams, i.e., corrections to the hard vertex of the basic
one-loop amplitude;
(e) and (f) are crossed laddersthe nonplanar topology;
(g), (h) and (i) finally are the vertex-insertion diagrams, containing a correction to the
vertices with momentum p1 or p2 .
The number of internal lines with different momenta entering a diagram is an indicator of
the topology of the diagram. The diagrams in Fig. 1 are real six-denominator amplitudes
and the related topologies are given in Fig. 4.
In Fig. 2 the diagrams containing self-energy corrections to the internal lines of the
one-loop vertices are plotted. Note that, in contrast to Fig. 1, diagrams (a), (b), (c) and
(d) in Fig. 2 involve only five distinct denominatorsone of them being squaredas two
internal lines carry the same momentum. These diagrams then, do not represent real sixdenominator topologies and are more naturally grouped among the five-denominator ones.
The topologies of diagrams (a), (b), (c) and (d) are shown in Fig. 5 in (h), (e), (f) and
(s), respectively. Diagram (e) of Fig. 2 involves two internal lines with the same momenta
but with different masses. It can be expressed as a superposition of integrals belonging to
3 Actually, diagram (f) in Fig. 2 contains only five denominators, but we list it in this figure as it belongs to
the same class of the six-denominator ones.
Fig. 1. Two-loop vertex diagrams with up to one massive propagator. They involve the ladder, crossed-ladder and
vertex-insertion topologies and are real six-denominator topologies (see text).
topologies (s) and (t) of Fig. 5, by means of partial fractioning in the loop momentum k1 :

1 1
1
1
=
.
(5)
k12 (k12 + a) a k12 k12 + a
4 In graphical form,
The scalar product of two vectors is defined as: a b = a0 b0 + a b.
the decomposition reads
1
=
a
4 Note a difference of sign with respect to the more common definition.
(6)
Fig. 2. Two-loop vertex diagrams with up to one massive propagator involving a self-energy correction to the
basic one-loop vertices. The related topologies are five and four denominator topologies (see text).
Finally, diagram (f) is actually a four-denominator topology, indicated in Fig. 6(g).

To conclude, let us note that the topologies related to diagrams of Fig. 2 can be obtained
by properly shrink one internal line (or two internal lines in the case of diagram (f)) among
those in the diagrams of Fig. 4.
2.1. Decomposition of a tensor amplitude into scalar amplitudes
In this section we consider the decomposition of a general tensor amplitude among
those shown in Figs. 1 and 2 into scalar amplitudes. Even though the decomposition is
completely general, let us pick up, for claritys sake, a specific case: the ladder diagram
with a mass on the external boson line (the diagram (b) in Fig. 1). Omitting an overall
constant dependent on the couplings of the theory, this diagram involves a double integral
over the loop momenta k1 and k2 and can be written, in all generality as5

Iij ... (p1 , p2 ) =

Nij ... (p1 , p2 , k1 , k2 ; a)
d D k1 d D k2
,
P1 P2 P3 P4 P5 P6
(7)
5 Depending on the gauge choice for vector particles, some of the denominators P may actually appear to
i
power two instead of one: the reduction scheme is clearly unchanged. We do not consider the noncovariant gauges,
leading to additional denominators and scalar products involving a constant gauge vector c .
where {d D k} is the loop integration measure (see Section 4). According to our routing of
the loop momenta, we have defined the denominators:
P1 = k12 + a,
(8)
P2 = (p1 k1 ) ,
(9)
P3 = (p2 + k1 ) ,
(10)
2
2
P4 = k22 ,
(11)
P5 = (p1 k1 + k2 ) ,
(12)
P6 = (p2 + k1 k2 )2 .
(13)
The numerator Nij ... depends on the spin of the particles entering the diagram and on the
type of interaction; the indices ij . . . denote collectively Lorentz indices, Dirac indices,
etc. By using projectors on invariant form factors, we can always restrict ourselves to
consider scalar numerators (see for example [15]). The latter typically involve a trace over
spinor indices and a sum over the polarizations of the vector particles. Therefore, from now
on, let us consider

D D N (S1 , S2 , S3 , S4 , S5 , S6 , S7 )
d k1 d k2
,
I(p1 p2 ) =
(14)
P1 P2 P3 P4 P5 P6
where we have explicitly indicated the dependence on the seven invariants, dependent on
the loop momenta k1 and k2 :
S1 = k12 ,
(15)
S2 = k22 ,
(16)
S3 = k1 p1 ,
(17)
S4 = k1 p2 ,
(18)
S5 = k2 p1 ,
(19)
S6 = k2 p2 ,
(20)
S7 = k1 k2 .
(21)
The scalar numerator can be expanded in a power series of the invariants as:
l
l l
N (S1 , S2 , S3 , S4 , S5 , S6 , S7 ) =
b(l1 , l2 , . . . , l7 )S1l1 S2l2 S33 S4l4 S55 S66 S7l7 ,
(22)
l1 ,...,l7 0
where the (known) coefficients b(l1 , l2 , . . . , l7 ) also depend on the external invariant s,
the mass squared a = m2 and the spacetime dimension D. With the typical gauge
interactions, we expect the indices li to be restricted to li 3.
2.2. Decomposition into independent scalar amplitudes
The scalar amplitudes in Eq. (14), with N given by the expansion in Eq. (22) are not,
in general, linearly independent on each other; we can indeed express six of the invariants
listed in Eqs. (15)(21) in terms of the denominators Pi . The next step is the reduction to
10
a set of linearly independent scalar amplitudes in order to progress in the evaluation of the
tensor amplitude of Eq. (7).
In this work, we used two different reduction schemes of the scalar amplitudes, using
consistency of the results as a crossed check. The first one is the scheme of the auxiliary
denominator (or auxiliary diagram), which is essentially based on the reduction of all the
amplitudes to scalar ones containing formally only denominators; the second one is the
scheme of the shifts, in which the independent scalar amplitudes are integrals with a subset
of the scalar products at the numerator.
In the following two sections we will recall briefly both schemes, paying more attention
on the first one, not discussed often in the literature, and just sketching the second one, for
which we refer to the bibliography. Our results are presented in Section 4 in the framework
of this second scheme.
2.2.1. Auxiliary-denominator scheme
As already said in the previous section, we aim at expressing the invariants listed in
Eqs. (15)(21) in terms of the denominators Pi appearing in the diagram. Since there are
seven invariants and only six denominators, we introduce a seventh auxiliary denominator,
independent from the others, in such a way to form a basis. The choice is of course, not
unique; let us take for example
P7 = (k1 k2 )2 .
(23)
We then have the linear relations

Si =
Aij (Pj aj )
(i = 1, . . . , 7),
(24)
j =1
where aj = j,1 a and Aij an invertible 7 7 constant matrix. The nonzero matrix elements
are given by
S1 = P1 a,
(25)
S2 = P4 ,
(26)
1
S3 = (P2 + P1 a),
(27)
2
1
S4 = (+P3 P1 + a),
(28)
2
1
S5 = (P7 + P5 P2 + P1 a),
(29)
2
1
S6 = (+P7 P6 + P3 P1 + a),
(30)
2
1
S7 = (P7 + P4 + P1 a).
(31)
2
Let us note that one can construct a fictitious Feynman diagram having all the seven
denominators, called the auxiliary diagram [16]:

{d D k1 }{d D k2 }
.
(aux. diag.) =
(32)
P1 P2 P3 P4 P5 P6 P7
11
Fig. 3. Double box auxiliary diagram for the case of the vertex diagram in Fig. 1(b). The dashed line represents
the auxiliary denominator P7 .
In our example, the auxiliary diagram is a planar double box with final momenta equal to
the initial ones p1 and p2 , i.e., a four-point forward amplitude: see Fig. 3. In general, an
auxiliary diagram has a number of independent denominators equal to the number of the
invariants (seven in our case). The scalar numerator N can now be expanded in powers of
the basis vectors {Pi }i=1,7 as
N=
c(s1 , s2 , . . . , s7 )P1s1 P2s2 P7s7 ,
s1 ,...,s7 0
with the new coefficients c(s1 , s2 , . . . , s7 ) being linear combinations of the old ones
b(l1 , l2 , . . . , l7 ). Inserting the above expansion in the expression (14) for the diagram, we
obtain

{d D k1 }{d D k2 }
I(p1 p2 ) =
c(s1 , . . . , s7 )
.
1s
1s
1s
P11s1 P21s2 P3 3 P41s4 P5 5 P6 6 P7s7
s1 ,...,s7 0
(33)
Eq. (33) is the main result of this section. It says that the general scalar diagram I is a linear
combinationwith known coefficientsof independent scalar amplitudes involving the
original denominators {Pi }i=1,6 and an additional, fictitious, denominator P7 . A compact
notation for the auxiliary scalar amplitudes is the following:

{d D k1 }{d D k2 }
Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ) =
(34)
n1 n2 n3 n4 n5 n6 n7 ,
P1 P2 P3 P4 P5 P6 P7
according to which Eq. (33) is written as
I(p1 , p2 ) =
d(n1 , n2 , . . . , n7 ) Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ).
(35)
n1 ,...,n6 1,n7 0
A few remarks are in order. First, the auxiliary denominator P7 , if it appears, it only does at
the numerator: it cannot clearly be generated at the denominator by tensor decomposition.
In other words, ni = 1, 0, 1, 2, 3, . . . for i = 1, . . . , 6 while n7 = 0, 1, 2, 3, . . ..
The second remark concerns the so-called topologies. As already anticipated, an important
quantity for a scalar amplitude is the number of distinct denominators with positive indices,
12
ni > 0, which we call t:

t
(ni ),
(36)
i=1
with (x) the unit step function defined in such a way that (0) = 0. The scalar amplitude
with n1 = 2, n2 = = n6 = 1 and n7 = 1 for example has t = 6, the maximal value for
two-loop vertex functions. If a scalar amplitude Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ) has ni 0,
the ith denominator eventually appears only at the numerator and the ith internal line is
shrunk to a point: the diagram then has less than six internal lines. Here we give an example
of a decomposition into auxiliary amplitudes with t = 6 and t = 5:

D D
S7
d k1 d k2
P1 P2 P3 P4 P5 P6
a
1
= Topo(1, 1, 1, 1, 1, 1; 1) Topo(1, 1, 1, 1, 1, 1; 0)
2
2
1
1
+ Topo(0, 1, 1, 1, 1, 1; 0) + Topo(1, 1, 1, 0, 1, 1; 0).
(37)
2
2
In general, a diagram I is decomposed into auxiliary amplitudes with decreasing t
starting from t = 6 included, i.e., t = 6, 5, 4, 3.6
2.2.2. Shift scheme
We described above the scheme of the auxiliary diagram to reduce a general scalar
amplitude. Let us now present another scheme of reduction which has also been used in
the past (see [13,14,17]). The main results of this work are presented in Sections 4 and 5
according to this second scheme, so let us briefly overview it.
In essence, it uses shifts on the loop momenta in order to maximally simplify the
structure of the sub-topologies. Let us consider a routing with two internal lines having the
loop momenta k1 and k2 , as for instance P1 and P4 in the discussion above. We choose for
example to keep the invariant S7 and we simplify, i.e., cancel, the factors Si (i = 1, . . . , 6)
appearing at the numerator with the denominators by means of the formulas:
S1 = P1 a,
(38)
S2 = P4 ,
1
S3 = (P2 + P1 + a),
2
1
S4 = (+P3 P1 + a),
2
1
S5 = (+P5 P4 P2 ) + S7 ,
2
1
S6 = (P6 + P4 + P3 ) S7 .
2
(39)
6 Amplitudes with t < 3 vanish, in our case, as they contain a massless tadpole.
(40)
(41)
(42)
(43)
13
The simplification of a scalar product with the selected denominator brings to the
appearance of sub-topologies in the case the denominator appears to power one. In
principle, a diagram of topology number t can have t! different sub-topologies. In our case,
however, some sub-topologies actually vanish because all the propagators, except possibly
one, are massless. Furthermore, eventual discrete symmetries of the diagram, such as an
updown symmetry, further reduce the number of independent sub-topologies.
After the cancellations between the Si and Pj have been done, we end up with a set
of subdiagrams in which the denominators P1 and P4 containing the loop momenta k1
and k2 may be absent. In this case, we perform shifts on the loop momenta to reproduce
denominators containing again k1 and k2 . For example, if P1 is absent but P2 is still present
in the given amplitude, we may do the shift
k1 p1 k1 .
(44)
If P1 and P2 are absent but P3 is still present, we may do the shift k1 k1 p2 , and
so on. In general, we construct a table of shifts on the loop momenta. With this step, new
denominators are generated. Under (44) for example
P3 (p1 + p2 k1 )2 ,
P5 (k1 + k2 )2 ,
P6 (p1 + p2 k1 k2 )2 ,
(45)
denominators which were not initially present. We then extend the set of cancellation rules
including those between the new denominators and the scalar products. The scalar product
k1 k2 cancels against the denominator (k1 +k2 )2 no arbitrarenesswhile (p1 +p2 k1 )2
may be cancelled for example with p2 k1 and (p1 + p2 k1 k2 )2 with p2 k2 . The
remaining scalar products are then p1 k1 and p1 k2 . Depending on the topology number
t of the diagram under consideration, we can re-express t of the 7 scalar products in
terms of denominators and simplify such expressions with a power of the corresponding
denominator. It follows that a diagram with topology number t can have at most t = 7 t
independent scalar products in the numerator. A six-denominator amplitude (t = 6), for
example, has at most one scalar product, while a sunrise diagram (t = 3) has at most four
distinct scalar products in the numerator. After all the second-step cancellations have been
done, we go for a second step of shifts of the loop momenta, and so on. We repeat iteratively
the cancellation-shift procedure until no more simplifications are feasible. The whole set of
required denominators is listed in Appendix A. In our computation we found 9 independent
scalar amplitudes with six denominators, i.e., t = 6 (Fig. 4), 22 independent amplitudes
with five denominators (Fig. 5), 17 independent amplitudes with four denominators (Fig. 6)
and 4 independent amplitudes with 3 denominators (Fig. 7). This makes a total of 44
independent scalar amplitudes.
Within this method, for a given topology, i.e., for a given set of denominators
Di1 Di2 Dit (see Appendix A), the independent scalar integrals are of the form

Topo(n1 , . . . , nt ; s1 , . . . , st ) =
s
t
D D Si11 Sit
d k1 d k2 n1
,
Di1 Dint t
(46)
14
Fig. 4. The set of 9 independent 6-denominator topologies.
where t = 7 t denotes the number of independent scalar products Si1 Sit . Note that
s1 , . . . , st 0 while n1 , . . . , nt 1.
The advantages of the auxiliary diagram method are the simplicity and a shorter code
for its implementation. The main advantage of the shift method is that a given topology
appears in a unique form, so the CPU time needed is less.
Let us conclude this section with a general remark. A straightforward strategy to
evaluate I is that of computing all the independent scalar amplitudes entering its
decomposition. In simple calculations, this is probably the simplest thing to do. In more
complicated cases, however, there may be a large number of such scalar amplitudes and
the individual computation of all of them may become rather laborious. In these cases, it
is more convenient to use a set of relations connecting different scalar amplitudes, and to
compute directly only a subset of them. The latter is our strategy that is described in the
next section.
2.3. Relation between independent scalar amplitudes: the integration by parts identities
As we have shown in the previous section, the task of evaluating a two-loop
vertex diagram with an arbitrary scalar numerator N is shifted to that of computing
a set of independent scalar amplitudes either containing an auxiliary denominator:
Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ), or containing a selected basis of denominators and scalar
products: Topo(n1 , . . . , nt ; s1 , . . . , st ). Let us discuss the simpler case of the auxiliary
diagram first. The allowed range of the indices are: n1 , . . . , n6 1 and n7 0 (see
Eq. (34)), but in practice, we expect the relevant ni s to be restricted to ni 3, which
however gives still a rather large set of scalar amplitudes to compute. Relations among
different scalar amplitudes are obtained by considering the so-called integration-by-parts
(ibp) identities [7]

D D
v
0=
d k1 d k2
(47)
,
n
n
n
ki P1n1 P2n2 P3 3 P4n4 P5 5 P6 6 P7n7
15
with i = 1, 2 and v = k1 , k2 , p1 , p2 . The integral above vanishes because of the following

argument. It is the space integral in D dimensions of the total divergence of the vector
in curly brackets. It can then be transformed into the flux integral of this vector over a
spherical surface with infinite radius, r = . For small enough dimension D, the surface
integral vanishes for r . It can then be defined to identically vanish for any D. By
explicitly performing the derivations in Eq. (47) and expressing the scalar products in terms
16
of the assumed basis {Pi }i=1,7 according to Eqs. (24), we obtain identities of the form
0 = c Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ) +
ni di Topo(n1 , . . . , ni + 1, . . . , n7 )
i=1
1,7
ni eij Topo(n1 , . . . , ni + 1, . . . , nj 1, . . . ; n7 ).
(48)
i=j
For a given set of the indices {ni }i=1,7 there are eight of such identities; in general, however,
they are not all linearly independent. Each identity contains, in general, three kinds of
amplitudes:
17
(1) the amplitude itself: Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 );

(2) amplitudes with one of the indices increased by one, ni ni + 1 (i = 1, 7):
Topo(n1 , . . . , ni + 1, . . . ; n7 ). These terms clearly originate from the derivation of the
factor Pini ni Pini 1 . We pulled a factor ni out of the coefficient di to point out
that these terms are absent for ni = 0. In other words, it is not possible to generate an
absent denominator by differentiation. This implies that the ibp identities for a given
amplitude with topological number t involve only amplitudes with smaller or equal
topological number, t t;
(3) amplitudes with one index increased by one, ni ni + 1 and another index decreased
by one, nj nj 1: Topo(n1 , . . . , ni + 1, . . . , nj 1, . . . ; n7 ). These terms originate
n
from the derivation of the factor Pini and the cancellation of a power of Pj j by
the invariants generated at the numerator. We pulled out a factor ni also out of the
coefficients eij to point out similar properties as those discussed in (2).
Other identities are obtained by considering discrete symmetries of the auxiliary
diagram. The ladder topologies in (a), (b) and (c) of Fig. 4 for example have an updown
symmetry, resulting in the exchange of the fermion lines expressed by the relation
Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ) = Topo(n1 , n3 , n2 , n4 , n6 , n5 ; n7 ).
(49)
The massless crossed ladder also has an updown symmetry, which interchanges also the
boson lines. By solving the above identities, it is possible to express a general scalar
amplitude as a linear combination of a finite set of NF basic integrals, called master
integrals (MIs)
Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ) =
NF
ci (n1 , n2 , n3 , n4 , n5 , n6 ; n7 )Fi .
(50)
i=1
In more formal terms, we may say that the ibp identities induce such a strong linear
dependence between all the infinite scalar amplitudes to project them into a finitedimensional linear space. The Fi s have the same role as the basis vectors in a linear space.
The choice of the MIs is arbitrary, as it is the choice of the basis vectors; only their number
NF is fixed and equals the dimension of the space spanned by any choice of the Fi s.
Changing the set of MIs in the decomposition (50) is analogous to a change of basis. To
give an explicit example, let us report the MI decomposition of the basic six-denominator
amplitude of the ladder with a massive external boson in Fig. 1(b), the example in the
previous section
Topo(1, 1, 1, 1, 1, 1; 0)

1
1
2
7
1
=2 2 +6
+ +
Topo(0, 1, 0, 1, 0, 1; 0)

x2 x 1 x

1
1
1
4
1
1
+
3
1
+
2
+
2
(1 x)2

x 1x

Topo(1, 0, 0, 1, 1, 0; 0) + Topo(1, 0, 0, 0, 1, 1; 0)
18

1
4
2 2 + 4 Topo(1, 1, 0, 1, 0, 1; 0)

1
1
2 Topo(1, 0, 0, 1, 1, 1; 0)

x

1
1
1
2
9
2
11
+
9
+
12
+
2
(1 x)2
2

x 1x

1
1
2
Topo(1, 1, 1, 0, 1, 1; 0)
Topo(1, 0, 0, 1, 1, 1; 1) +

x
2
+ Topo(1, 1, 0, 1, 1, 1; 0),
(51)
x
where we have taken for simplicity a = 1. In this case NF = 8. We have chosen
MIs with a subset of the denominators with unitary indices: ni = 1, the remaining
denominators having vanishing indices: nj = 0. For the topology represented by the
amplitude Topo(1, 0, 0, 1, 1, 1; 0) this was ot sufficient as two master integrals are
involved. We added the amplitude with the auxiliary denominator brought to the numerator,
i.e., with index n7 = 1: Topo(1, 0, 0, 1, 1, 1; 1); we could have taken n2 = 1, n3 =
0, n7 = 0 or n2 = 0, n3 = 1, n7 = 0 as well. Still another possibility would
have been to consider amplitudes with one of the denominators squared, such as for
example Topo(2, 0, 0, 1, 1, 1; 0). In general, the choice of the MIs is dictated by practical
considerations. In some cases, for example, it may be useful to require the absence of
ultraviolet or infrared singularities from the MIs [18]; in other cases, the set of MIs can
be chosen in such a way that the corresponding system of differential equations become
easier to solve (see next section).
Let us now discuss the methods to solve the ibp identities to arrive to the results (50).
The ibp identities constitute a system of linear equations in which the unknowns are the
scalar integrals themselves. The oldest approach, developed in the original article on the
ibps, involves a symbolic solution of the identities, treated as recurrence equations on the
indices [7,18]. One introduces operators raising or lowering one of the indices
i Topo(n1 , . . . , ni , . . . ; n7 ) = Topo(n1 , . . . , ni 1, . . . ; n7 ).
(52)
i+
i+ j .
and
Each ibp equation involves the identity operator 1 and the raising operators
If one can combine the equations in such a way that an amplitude is written in terms
of amplitudes all containing lowering operators, then reduction to simpler topologies has
been achieved. The advantages of the symbolic approach are the elegance and that its
implementation does not require in general a lot of CPU time. The main disadvantage is
that a careful, case-by-case, inspection of the equations is required. A more recent approach
is that of replacing numerical values for the indices and solve the resulting linear system
[13,14]. Let us describe this method in practical terms. An initial linear system can be
obtained by assigning to the indices the values ni = 0, 1 for i = 1, . . . , 6 and n7 = 0. This
is a set of 8 26 = 512 equations.7 If these equations are not sufficient for a complete
7 As already said, not all the equations in the system are linearly independent on each other and some of them
are actually trivial (i.e., 0 = 0), such as for example those with five indices equal to zero.
19
MI reduction, one may solve a larger system in which one of the indices can also take the
value ni = 1 for i = 1, . . . , 7; this is a set of (3 25 6 + 27) 8 = 5632 equations. Let
us stress that all the equations with one index = 1 have to be considered simultaneously,
because the amplitudes involved are coupled by the i+ j operators; for example
7+ 6 :
Topo(1, 1, 1, 1, 1, 0; 1) Topo(1, 1, 1, 1, 1, 1; 0).
(53)
One may also consider a system in which one of the indices takes the value ni = 2 for
i = 1, 6; this amounts to 3 25 6 8 = 4608 equations. If these systems are not
sufficiently large, one may consider equations with indices taking both the values 2 and
1 and so on. The general idea is that, by going to larger systems, the number of equations
grows faster than the number of amplitudes, till a critical mass of equations is reached,
allowing the full reduction to MIs8 [17]. The linear system can be solved with the method
of elimination of the variables. In general, one has to devise a rule about which amplitudes
in the system to solve first. Important quantities for the scalar integrals are the sum of the
positive indices
d=
ni (ni )
(54)
i=1
and minus the sum of the negative indices

s=
ni (ni ).
(55)
i=1
A convenient criterion is to solve for the amplitudes with the greatest t, d and s first. The
advantage of the numerical indices method is that it does not require a hand inspection
of the equationsrequiring patience and often a good mathematical intuitionand is
therefore very general. The main disadvantage is that it may require a large CPU time
and very long intermediate expressions may be generated because of the division by the
coefficients of the unknowns amplitudes.
Let us now discuss the ibp identities in the framework of the second reduction scheme
discussed in the previous section, the one involving independent scalar products at the
numerator and loop momentum shifts. As with the auxiliary diagrams, the scalar integrals
of Eq. (46) can be related by the ibp identities, which now read

s

S s1 Si t
D D
t
i1
= 0,
d k1 d k2
v
ki
Din11 Dint t
(56)
where, as before, i = 1, 2 and v = k1 , k2 , p1 , p2 . For a given set of indices {si , nj } of

the input amplitude, we find eight identities involving integrals with up to an additional
power on a denominator and an additional power of a scalar product at the numerator.
Sub-topologies with t 1, t 2, . . . coming from simplifications between scalar products
and corresponding denominators also appear. As in the auxiliary diagram case, discrete
8 One of us (U.A.) wishes to thank S. Laporta for a discussion on this point.
20
Fig. 8. The set of 22 master integrals. Four of them are the product of one-loop master integrals.
symmetries of various topologies generate additional identities to be combined with the

ibp identities [14].
Concerning our problem, we can reduce the calculation of all the scalar integrals related
to the topologies shown in Figs. 5, 6 and 7and therefore the calculation of the Feynman
diagrams of Figs. 1 and 2to that of the 22 MIs in Fig. 8. Four of them are the product
of two one-loop MIs. The picture of the diagram alone denotes the scalar integralonly
denominatorswhile the picture of the diagram accompanied by a scalar product denotes
the scalar integral with that scalar product at the numerator.
21
The algorithms explained in this section were implemented in a chain of programs

written in the algebraic computer language FORM [19] and, in the case of the auxiliarydenominator scheme, also in Mathematica [20]. For the solution of the linear system of
ibp identities, we used SOLVE [21]. This program uses the language C as a shell to manage
a user-provided basic code written in FORM. External calls to Maple [22] are done in order
to simplify the coefficients coming in intermediate steps.
3. Evaluation of the MIs: the differential equations method

Let us consider a general amplitude Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 ) in the framework of
the auxiliary diagram. As shown in the previous section, it can be decomposed, by means
of the ibp identities, into a superposition of a set of primitive amplitudes called master
integrals (MIs). The latter cannot be computed with the ibp identities, which, in general,
only allow for a reduction of the number of independent amplitudes to be individually
computed. The MIs have been computed with a variety of techniques over the years:
Feynman parameters, dispersion relations, small momentum expansions, large momentum
expansions, just to mention the more common ones. A technique developed over the last
few years which is very efficient in our case is that of the differential equations in the
external kinematical invariants [814]. Let us then review the basics of this method. Once
again, let us consider a specific example: the computation of the MIs in Fig. 8(k) in the
auxiliary diagram scheme. Let us define
F1 (s, a, ) = Topo(1, 0, 0, 1, 1, 1; 0),
(57)
F2 (s, a, ) = Topo(1, 0, 0, 1, 1, 1; 1).
(58)
The above two MIs appear in the decomposition of the six-denominator amplitude
presented in the previous section (Fig. 1(b)). They are functions of s and a, as well as
of the spacetime dimension D = 4 2. Let us take the derivative of Fi with respect to s
at fixed a and with respect to a at fixed s:

Fi (s, a, ) = p1 Fi (s, a, ) = p2 Fi (s, a, ),
s
p1
p2
a Fi (s, a, ).
a
(59)
(60)
The derivatives with respect to the external momentum component p1 or p2 and with
respect to the mass squared a can be taken inside the loop integral, because of the properties
of dimensional regularization. The resulting amplitudes can be reduced to combinations of
MIs according to the methods described in the previous section, so that
s Fi (s, a, ) =
(61)
Cij (s, a, )Fj (s, a, ),
s
j
Fi (s, a, ) =
a
Kij (s, a, )Fj (s, a, ).
(62)
22
Note the double role played by the ibp identities in the calculation: they allow for the
reduction of independent scalar amplitudes into combinations of master integrals and
they also allow for the generation of the differential equations to be solved for the MIs
themselves.
Eqs. (61) and (62) can be simplified changing variables from s and a to x = s/a and a.
This triangular change of variables implies that
Fi (x, a, ) = s Fi (s, a, ),

x
s
di
a Fi (x, a, ) = s Fi (s, a, ) + a Fi (s, a, ) = Fi (x, a, ).

a
s
a
2
The second equation holds because of the dimensional relation
x
Fi (x, a, ) = a di /2 fi (x, ),
7
(63)
(64)
(65)
(i)
k=1 nk .
where di is the mass dimension of the MI: di /2 = D

In the new variables,
the scale a represents an over-all scale of the amplitudes and its evolution equation is
consequently trivial. We therefore concentrate on the x-evolution equation only and we set
a = 1 from now on in this section:
F
Fi (x, ) =
Aij (x, )Fj (x, ).
x
(66)
j =1
In our case, by performing the above steps, one obtains the following linear system of
coupled ordinary differential equations with variable coefficients

1
1
1
dF1
(x, ) = F1 (x, ) (2 3)
+
F2 (x, )
dx
x
x 1x

1
1
(1 )
F3 () + F4 (x, ) ,
+
(67)
x 1x

1
1
dF2
(x, ) = (1 ) + (2 3)
F2 (x, )
dx
x
1x

1
1
1
(1 )
(68)
+
F3 () (1 )
F4 (x, ),
x 1x
1x
where we defined
F3 () = Topo(1, 0, 0, 1, 1, 0; 0),
(69)
F4 (x, ) = Topo(1, 0, 0, 0, 1, 1; 0).
(70)
F3 is represented in Fig. 8(u); it is a sunrise diagram with one massive line, evaluated at
a light-cone external momentum, equivalent to the null momentum. F3 is then effectively
a vacuum amplitude, so we dropped the dependence on x. It is obtained from Fig. 8(k)
by shrinking one of the internal lines having an endpoint in the hard vertex. F4 is
represented in Fig. 8(v) and is the product of a massless bubble and a tadpole; it is
obtained from Fig. 8(k) by shrinking the massless line in the bubble. Note that there are no
other nonvanishing sub-topologies. We consider the sub-topologies F3 and F4 as known
23
amplitudes,9 so that the system of Eqs. (67) and (68) is schematically written as
F
Fi (x, ) =
Aij (x, )Fj (x, ) + i (x; ),
x
(71)
j =1
where the i (x, ) are known functions constituting the non-homogeneous part of the
system. Let us note that the coefficients of the homogeneous terms have a regular expansion
around = 0, involving, in this case, only two terms
(1)
Aij (x, ) = A(0)
ij (x) + Aij (x).
(72)
A great simplification is offered by the fact that the system is triangular in our basis,
because the equation for dF2 /dx depends only on F2 and not on F1 : A21(x, ) = 0. Let us
then consider this equation first. We substitute the explicit expression for the expansion of
2 in powers of and the symbolic expansion for F2 , up to the desired order n:
2 (x, ) =
F2 (x, ) =

(j )
j 2 (x) + O n+1 ,
(73)

(j )
j F2 (x) + O n+1 .
(74)
j =2
n
j =2
The double polethe leading singularityobeys a differential equation of the form

(2)
dF2
dx
(2)
(x) = A(0)
(x) + (2)(x),
22 (x)F2
(75)
where
2
1
.
x 1x
The associated homogeneous equation
(0)
A22 (x) =
d2
(0)
(x) = A22 (x)2 (x)
dx
is solved by separation of variables to give
x

(1 x)2
,
(t)
dt
=
A(0)
2 (x) = exp
22
x
(76)
(77)
(78)
with an overall constant omitted in the last member. The double pole coefficient is then
computed with the method of the variation the constants
(2)
F2 (x) = 2 (x)
x
K2 (t) (2) (t) dt,
(79)
9 Their expressions are given respectively in Eqs. (121)(125) (where we put F = F (4) ) and Eqs. (128)(132)
3
(where F4 = F (5) ).
24
where the primitive involves an arbitrary constant to be fixed imposing initial conditions,
and where the kernel K2 is given by
1
1
1
=
+
.
2 (t)
1t
(1 t)2
Higher orders are computed through the recursive relation in :
K2 (t) =
(80)
(i)
dF2 (x)
(i)
(1)
(i1)
(81)
(x) + (i)(x).
= A(0)
22 (x)F2 (x) + A22 (x)F2
dx
The solution of the above equation is again obtained with the method of variation of
constants
x
(i)
(i) (t) dt,
F2 (x) = 2 (x) K2 (t)
(82)
where
(i) (t) = (i)(t) + A(1)(t)F (i1) (t)
22
2
(83)
is the inhomegeneous term redefined to include the amplitude of previous order in the
expansion. Note that the kernel K2 is the same for each order in .
The full evaluation of F2 requires, as we already said, an initial condition on the x
evolution. The latter can often be derived by studying the behaviour of the MI close to
a threshold or a pseudothreshold. In our case, let us consider the point x = 1: it is a
pseudothreshold as it corresponds to s = m2 < 0 and therefore is not a singular point
for the amplitude, so that
dF2 (x, )
= 0.
(84)
dx
By imposing the above condition to the differential equation, an initial value for the
amplitude in x = 1 is obtained
lim (1 x)
x1
(2 3)F2 (1; ) + (1 )F3 () + (1 )F4 (1; ) = 0.
(85)
Once F2 has been explicitly computed, one replaces the result in the first equation and
solves it for F1 in the same way.
In general, however, the system is not triangular and the above method is not directly
applicable. One can try a transformation to a new set of MIs,
F1 , F2 F1 , F2 ,
(86)
in order to have the system in a triangular form. A triangularization to order is actually

(1)
10
sufficient, i.e., one for which A(0)
21 = 0 but A21 = 0.
10 In more complicated cases, it may happen that a triangular form of the system cannot be reached, and the
coupled system of nF equations must be solved. This is equivalent to the solution of a differential equation of
order nF for anyone of the MIs, for which no general algorithm exists. In our problem, the number of MIs for a
given topology is at most nF = 2, corresponding to a second-order differential equation. Moreover, it turned out
that the solution of the associated homogeneous system was rather simple, so that the complete solution could be
found with the method of the variation of arbitrary constants.
25
The MIs can be expressed in terms of the one-dimensional harmonic polylogarithms

(HPLs), the latter having the following recursive integral definition
H (1; x) = log(1 x),
(87)
H (0, x) = log(x),
(88)
H (1; x) = log(1 + x)
(89)
and
x
f (a; y)H (w;
y) dy,
H (a, w;
x) =
(90)
where the basic weight functions are defined as

f (1; x) =
1
,
1x
Fig. 9. Flowchart of the method used for the reduction to the MIs and their evaluation.
(91)
26
f (0; x) =
1
,
x
(92)
1
.
(93)
1+x
For a complete discussion of HPLs and their properties we refer to [2]; for their
numerical evaluation see [3].
As an empirical rule, we found that it was sufficient to stop the expansion beyond
the order involving weight w = 4 HPLs.
The massless MIs cannot be computed with the differential equation method and a more
traditional method as that of Feynman parameters or dispersion relations has to be used.
As discussed in the introduction, they are indeed of the form
2 2
4nd
G()x 4nd 2 .
(massless amp.) = a
(94)
a
Since the x-dependence is factorized, the x evolution gives only a trivial, dimensional,
information, analogous to the a evolution. The nontrivial quantity to compute is G(). In
other words, all the information is contained in the initial datafor example the amplitude
at x = 1that can be extracted only by direct integration.
For a graphical summary of the whole method discussed in Sections 2 and 3, see the
flowchart in Fig. 9.
f (1; x) =
4. Results for the master integrals

In this section we present the results of our computation of the MIs involving up to
five denominators included, which constitute a necessary input for the calculation of the
two-loop vertex diagrams in Figs. 1 and 2. They are expanded in a Laurent series in
= 2 D/2,
(95)
up to the required order in . The coefficients of the series are expressed in terms of HPLs
[2,3] of the variable x, defined as
q2
,
(96)
a
where q = p1 + p2 , and11 a = m2 . We denote by the mass scale of the dimensional
regularization (DR)the so-called unit of mass. The explicit values of the denominators
Di are given in Appendix A. We work in Minkowski space and we normalize the loop
measure as
D
d (42)k
dDk
= (2)
.
d k =
(97)
D
9(1 + )
i 2 9(3 D2 ) i
x=
This definition makes the expression of the one-loop tadpolethe simplest of all loop
diagramsparticularly simple (see Appendix B).
11 Note that with our definition of the scalar product s = q 2 .
27
4.1. Topology t = 3

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D2 D11

i Fi(1) + O 3 ,
(98)
(99)
i=2
where:
(1)
F2
a
(1)
F1
= 0,
(100)
x
= ,
a
4

(1)
F0
13 1
= x
H (0, x) ,
a
8
2

(1)
F1
x 115
=
2 (2) 13H (0, x) + 4H (0, 0, x) ,
a
4 4

(1)
F2
x 865 13 (2)
115
=
5(3)
2(2) H (0, x)
a
2 16
2
4

+ 13H (0, 0, x) 4H (0, 0, 0, x) .
(101)
(102)
(103)
(104)
The above amplitude is easily computed with Feynman parameters or with an iteration of
the general formula for a massless bubble given in [7].

D D
1
(105)
= 2(4D)
d k1 d k2
D2 D11 D12
2 2
2

(2)
=
i Fi + O 3 ,
(106)
a
i=2

(k1 k2 ) =

=
2(4D)
2
a
D D
d k1 d k2
2
(3)
i Fi
k1 k2
D2 D11 D12

+ O 3 ,
(107)
(108)
i=2
where:
(2)
F2
1
= ,
2
(109)
3 x
= ,
a
2 4
(110)
a
(2)
F1
28

F0(2)
13
1 1
= 3 (2) x
x H (1, x) + H (0, 1, x),
a
8
2 x
(111)

F1(2)
15
x
= 3 (2) + (3)
115 + 8(2)
a
4
16

1 1
x 13H (1, x) 8H (1, 1, x)

4 x

1
1
+ 6 + 2x
H (0, 1, x) 4H (0, 1, 1, x) + H (0, 0, 1, x),
2
x
(112)

(2)
2
F2
21
9 (2)
865 13(2) (3)
=
6 (2)
+ 3(3)
+
x
a
8
5
32
4
2

1
1
1
x H (1, x) + 13
x H (1, 1, x)
115 + 8 (2)
8
x
x

1
13
+ 24 + 8 (2) + 26x
H (0, 1, x)
4
x

1
x 8H (1, 1, 1, x) 3H (1, 0, 1, x)
x

1
H (0, 0, 1, x) 4H (0, 1, 1, x)
+ 3+x
2x
+ 16H (0, 1, 1, 1, x) 6H (0, 1, 0, 1, x) 4H (0, 0, 1, 1, x)
+ H (0, 0, 0, 1, x),
(3)
F2
a2
(3)
F1
a2
(113)
1
= ,
4
(114)
3 x x2
= ,
4 4 24
(115)

(3)
F0
1
2
19 (2) 35x 13x 2
2
3
+
+
6x
+
x
H (1, x)
=
12
2
24
48
12
x
a2
1
+ H (0, 1, x),
(116)
2

F1(3)
95 (2)
5 3 (2) (3)
115 (2) 2
+
+
=
x
a2
2
2
2
16
2
96
12

13
35 37x 13x 2
+
+
H (1, x)
+
24
12
24
12x

7
x2
1
x2
2
+
+x +
H (0, 1, x) 1 + 2x +
H (1, 1, x)
4
6
6x
3
3x
1
(117)
2H (0, 1, 1, x) + H (0, 0, 1, x),
2
29

F2(3)
1
41 19 (2) 9 2(2) 3(3)
2015
=
35(2) 6(3) +
x
a2
24
6
10
2
12
8

865 2
1
13 (2) 2 (3) +
x
24
8

105
115 (2) 2
1 95
+ (2) +
+ 2(2) x +
+
+
x
2 8
4
24
3

115 2 (2) 1
+
H (1, x)
12
3
x

37
13x 2
13
+
+ (2) + 6x +
H (0, 1, x)
8
12
12x

1
26
35 + 74x + 13x 2
H (1, 1, x)
6
x

2
1
3H (1, 0, 1, x) 8H (1, 1, 1, x)
3 + 6x + x 2
6
x

1
2
H (0, 0, 1, x) 4H (0, 1, 1, x)
+
21 + 12x + 2x 2
12
x
1
+ H (0, 0, 0, 1, x) 6H (0, 1, 0, 1, x) 4H (0, 0, 1, 1, x)
2

(118)
+ 16H (0, 1, 1, 1, x) .

=

=
2(4D)
2
a

d D k1 d D k2
2
(4)
i Fi
1
D2 D8 D12

+ O 3 ,
(119)
(120)
i=2
where:
(4)
F2
a
(4)
F1
1
= ,
2
(121)
3
= ,
2
(122)
(4)
F0
7
= (2),
a
2
(123)
(4)
F1
15
= 3 (2) + (3),
a
2
(124)
(4)
F2
31
9 2 (2)
= 7 (2)
+ 3(3).
a
2
5
(125)
30
The above amplitude is easily computed with Feynman parameters.

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D9 D13

i Fi(5) + O 3 ,
(126)
(127)
i=2
where:
(5)
F2
= 1,
(128)
= 3 + H (0, x),
(129)
(5)
F1
a
(5)
F0
= 7 + (2) + 3H (0, x) H (0, 0, x),
a
(130)

F1(5)
= 15 + 3 (2) + 2 (3) + 7 (2) H (0, x) 3H (0, 0, x) + H (0, 0, 0, x),
a
(131)

F2(5)
9
= 31 + 7 (2) + 2 (2) + 6(3) + 15 3(2) 2(3) H (0, x)
a

10
7 (2) H (0, 0, x) + 3H (0, 0, 0, x) H (0, 0, 0, 0, x).
(132)
4.2. Topology t = 4

=

=
2(4D)
2
a

d D k1 d D k2
2
1
D2 D3 D9 D12

i Fi(6) + O 3 ,
(133)
(134)
i=2
where:
1
(6)
F2 = ,
2

5
1
(6)
F1 = 1 +
H (1, x),
2
x

1
19
+ (2) H (0, x) 1 +
F0(6) =
4H (1, x) 2H (1, 1, x)
2
x

H (1, 0, x) + H (0, 1, x) ,
(135)
(136)
(137)

(6)
F1 =

1
65
+ 3 (2) (3) 7H (0, x) 12 1 +
H (1, x) + 2H (0, 0, x)
2
x

1
4H (1, 0, x) 4H (0, 1, x) + 8H (1, 1, x)
+ 1+
x
4H (1, 1, 1, x) 2H (1, 1, 0, x) + 2H (1, 0, 1, x)
2H (1, 0, 0, x) + 2H (0, 1, 1, x) + H (0, 1, 0, x)

H (0, 0, 1, x) ,
F2(6) =
31
(138)

211
+ 5 (2) +
9(3) 33 2(2) H (0, x)
2
5

1
2 16 3 (3) 1 +
H (1, x) + 14H (0, 0, x) 4H (0, 0, 0, x)
x

1
2 6 (2) H (1, 0, x) 12H (0, 1, x) + 24H (1, 1, x)
+ 1+
x
8H (1, 1, 0, x) + 8H (1, 0, 1, x) 8H (1, 0, 0, x)
9 2 (2)
+ 8H (0, 1, 1, x) + 4H (0, 1, 0, x) 4H (0, 0, 1, x)

16H (1, 1, 1, x) + 8H (1, 1, 1, 1, x)
+ 4H (1, 1, 1, 0, x) 4H (1, 1, 0, 1, x)
+ 4H (1, 1, 0, 0, x) 4H (1, 0, 1, 1, x) 2H (1, 0, 1, 0, x)
+ 2H (1, 0, 0, 1, x) + 4H (1, 0, 0, 0, x) 4H (0, 1, 1, 1, x)
2H (0, 1, 1, 0, x) + 2H (0, 1, 0, 1, x) 2H (0, 1, 0, 0, x)

+ 2H (0, 0, 1, 1, x) + H (0, 0, 1, 0, x) H (0, 0, 0, 1, x) .

=

=
2(4D)
2
a

d D k1 d D k2
2
(7)
i Fi
1
D1 D2 D9 D10

+ O 3 ,
(139)
(140)
(141)
i=2
where:
(7)
F2 = 1,
(142)
(7)
F1
(7)
F0
(7)
F1
(143)
= 4 2H (0, x),
(144)
= 12 2 (2) 8H (0, x) + 4H (0, 0, x),

= 32 8 (2) 4 (3) 4 6 (2) H (0, x) + 16H (0, 0, x) 8H (0, 0, 0, x),
(145)

4 2 (2)
(7)
F2 = 80 24 (2)
16(3) 8 8 2(2) (3) H (0, x)

5
+ 8 6 (2) H (0, 0, x) 32H (0, 0, 0, x) + 16H (0, 0, 0, 0, x).
(146)
32

= 2(4D)

=
2
a

d D k1 d D k2
2
(8)
i Fi
1
D2 D10 D9 D12

+ O 3 ,
(147)
(148)
i=2
where:
(8)
F2 = 1,
(8)
F1

1
= 4 H (0, x) 1 +
H (1, x),
x

1
H (1, x) + H (0, 0, x)
F0(8) = 12 (2) 4H (0, x) 4 1 +
x

1
H (1, 0, x) + 2H (1, 1, x) ,
+ 1+
x
(149)
(150)
(151)

F1(8) = 32 4 (2) 2 (3) 12 (2) H (0, x)

1
12 (2) 1 +
H (1, x) + 4H (0, 0, x)
x

1
+ 1+
4H (1, 0, x) + 8H (1, 1, x) H (0, 0, 0, x)
x

1
H (1, 0, 0, x) + 2H (1, 1, 0, x) + 4H (1, 1, 1, x) ,
1+
x
(152)
9 2 (2)
8(3)
10

1
32 4 (2) 2 (3) H (0, x) + 1 +
H (1, x)
x

1
+ 12 (2) H (0, 0, x) + 1 +
2H (1, 0, x) + 2H (1, 1, x)
x

1
8H (1, 1, 0, x) + 4H (1, 0, 0, x)
4H (0, 0, 0, x) 1 +
x
+ 16H (1, 1, 1, x) H (1, 0, 0, 0, x) 4H (1, 1, 1, 0, x)

2H (1, 1, 0, 0, x) 8H (1, 1, 1, 1, x) + H (0, 0, 0, 0, x).
F2(8) = 80 12 (2)
(153)

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D2 D7 D8 D12

i Fi(9) + O 2 ,
33
(154)
(155)
i=2

(p2 k1 ) =

=
2(4D)
2
a

d D k1 d D k2
2
(10)
i Fi
p2 k1
D2 D7 D8 D12

+ O 2 ,
(156)
(157)
i=2
where:
1
(9)
F2
= ,
2
5
(9)
F1
= H (0, x),
2

1
19
(9)
F0 =
2 (2) 5H (0, x) + H (0, 0, x) 1
H (1, 0, x)
2
x
1
+ H (0, 1, 0, x),
x

1
65
(9)
F1 =
10 (2) (3) 19 (2) H (0, x) 3(2) 1
H (1, x)
2
x

1
3 (2)
H (1, 0, x)
H (0, 1, x) 5 1
+ 5H (0, 0, x) +
x
x
2
H (0, 0, 0, x) + H (0, 1, 0, x)
x

1
1
H (0, 1, 0, x) H (1, 0, 0, x) + 3H (1, 1, 0, x)
x

1
+ H (0, 0, 1, 0, x) H (0, 1, 0, 0, x) + 3H (0, 1, 1, 0, x) ,
x
(10)
F2
a
a
(159)
(160)
(161)
x
,
16
(162)

x
9 4H (0, x) ,
32
(163)
(10)
F1
(158)
34

F0(10) 5
x
3
x
= +
63 16 (2) [2 + 3x]H (0, x) + H (0, 0, x)
a
8 64
16
8

1
1
1
H (1, 0, x) + H (0, 1, 0, x),
x
8
x
4x
(164)

F1(10) 57 405x 9
x
1
=
+
(1 + x) (2) (3)
70 + 63x 4(2)x H (0, x)
a
16
128
8
8
32

3 (2)
1
3
x
H (1, x) + [2 + 3x]H (0, 0, x)
8
x
16

1
11
2 + 9x
H (1, 0, x)
16
x
3 (2)
1
1
+
H (0, 1, x) H (0, 0, 0, x) + H (0, 1, 0, x)
4x
8
2x

1
1
H (0, 1, 0, x) H (1, 0, 0, x) + 3H (1, 1, 0, x)
x
8
x

1
(165)
+
H (0, 0, 1, 0, x) H (0, 1, 0, 0, x) + 3H (0, 1, 1, 0, x) .
4x

=

=
2(4D)
2
a

d D k1 d D k2
2
(11)
i Fi
1
D1 D2 D7 D8

+ O 3 ,
(166)
(167)
i=2
where:
1
(11)
F2
= ,
2
5
(11)
F1 = H (0, x),
2
19
(11)
F0 =
5H (0, x) + 2H (0, 0, x),
2
65
(11)
F1 =
4 (3) 19H (0, x) + 10H (0, 0, x) 4H (0, 0, 0, x),
2

211 12 2
F2(11) =
(2) 20(3) 65 8(3) H (0, x) + 38H (0, 0, x)
2
5
20H (0, 0, 0, x) + 8H (0, 0, 0, 0, x).
(168)
(169)
(170)
(171)
(172)
As discussed in Section 3, the above amplitude, being massless, cannot be calculated

by means of the differential equations technique. It is easily computed with Feynman
35
parameters.

= 2(4D)

=
2
a
D D
d k1 d k2
2
1
D2 D4 D8 D12

i Fi(12) + O 3 ,
(173)
(174)
i=2
where:
1
(12)
F2 = ,
2
3
(12)
F1 = ,
2

1
5
(12)
F0 = + (2) + H (0, x) + 1
H (1, 0, x),
2
x

1
1
(12)
F1 = + 5 (2) (3) + 7H (0, x) + 2(2) 1
H (1, x)
2
x

1
4H (1, 0, x) + H (0, 1, 0, x)
2H (0, 0, x) + 1
x

2H (1, 0, 0, x) + 2H (1, 1, 0, x) ,
(12)
F2
(175)
(176)
(177)
(178)

51
+ 19 (2) +
+ (3) + 33 2(2) H (0, x)
2
5

1
+ 4 (3) + 2 (2) 1
H (1, x)
x

1
H (0, 1, x)
14H (0, 0, x) + 2 (2) 1
x

1
1
+ 2 6 (2) 1
H (1, 0, x) + 4(2) 1
H (1, 1, x)
x
x

1
4H (0, 1, 0, x) 8H (1, 0, 0, x)
+ 4H (0, 0, 0, x) + 1
x
+ 8H (1, 1, 0, x) + H (0, 0, 1, 0, x) 2H (0, 1, 0, 0, x)
9 2 (2)
+ 2H (0, 1, 1, 0, x) + 4H (1, 0, 0, 0, x) + 2H (1, 0, 1, 0, x)

4H (1, 1, 0, 0, x) + 4H (1, 1, 1, 0, x) .
(179)
4.3. Topology t = 5

= 2(4D)
D D
d k1 d k2
1
D1 D4 D7 D8 D13
(180)
36

=
2
a
2
(13)
i Fi
+ O(),
(181)
i=1
where:

1
H (0, 0, 1, x) + H (0, 1, 0, x) ,
x
1
aF0(13) = 3 (2)H (0, 1, x) 4H (0, 0, 1, 1, x) + H (0, 0, 0, 1, x)
x

+ H (0, 0, 1, 0, x) H (0, 1, 0, 0, x) + 3H (0, 1, 1, 0, x) .
(13)
=
aF1

= 2(4D)

=
2
a
2
D D
d k1 d k2
(14)
F0
1
D2 D4 D7 D8 D12
+ O(),
(182)
(183)
(184)
(185)
where:
aF0(14) =

1
(2)H (0, 1, x) + H (0, 1, 0, 0, x) + H (0, 1, 1, 0, x) .
x

= 2(4D)

=
2
a
2
D D
d k1 d k2
F0(15) + O(),

(k1 k2 ) = 2(4D)

=
1
D1 D2 D7 D8 D15
2
a
D D
d k1 d k2
2
(186)
(187)
(188)
k1 k2
D1 D2 D7 D8 D15

i Fi(16) + O 2 ,
(189)
(190)
i=2
where:

1
2 (2)H (0, 1, x) H (0, 1, 0, 1, x) + H (0, 1, 0, 0, x) ,
x
1
(16)
aF2 = ,
2
(16)
aF1 = 2 H (0, x),
(15)
aF0
(191)
(192)
(193)

(16)
aF0
(16)
aF1

5
1
= 7 3 (2) H (0, x) 2 (2) 1 +
H (1, x)
2
x

1
1
1
H (0, 1, x)
+ H (0, 0, x) + 3
2
2
x

1
1
H (1, 0, 0, x) H (1, 0, 1, x) ,
+ 1+
2
x
37
(194)

(3) 1
17
7 3(2) H (0, x)
= 24 (2) +
2
2
2

3
1
13 2 (2) + (3) 1 +
H (1, x)
2
x

(2)
1
1+
H (1, 0, x) + 5H (0, 1, x)
2
x

1
7
1+
H (0, 1, x)
+ (2)
2
x

1
1
+ 2 4 (2) 1 +
H (1, 1, x) + H (0, 0, 0, x)
x
2

1
1
8H (0, 1, 1, x) + 2H (0, 0, 1, x) + 1 +
4H (0, 1, 1, x)
2
x
H (0, 0, 1, x) 2H (1, 0, 1, x) + 2H (1, 0, 0, x)
+ 2H (1, 1, 0, 1, x) 2H (1, 1, 0, 0, x) + 4H (1, 0, 1, 1, x)
H (1, 0, 0, 1, x) 3H (1, 0, 0, 0, x) H (0, 1, 0, 1, x)

+ H (0, 1, 0, 0, x) .
(195)
The above topology is the only 5-denominator one having 2 MIs.

=

=
2(4D)
2
a

d D k1 d D k2
2
1
D1 D2 D6 D11 D15
i Fi(17) + O(),
(196)
(197)
i=1
where:
1
H (0, 1, 0, x),
x
1
(17)
aF0 = 2 (2)H (0, 1, x) + H (0, 0, 1, 0, x) 2H (0, 1, 0, 0, x)
x

+ 2H (0, 1, 1, 0, x) .
(17)
aF1 =
(198)
(199)
38
= 2(4D)

=
2
a

d D k1 d D k2
2
(18)
i Fi
1
D2 D4 D5 D10 D12

+ O 2 ,
(200)
(201)
i=1
where:
1
(18)
aF1
(202)
= H (1, 0, x),
x

1
(18)
aF0 = (2)H (1, x) + 2H (1, 0, x) 3H (1, 0, 0, x) + H (1, 1, 0, x) , (203)
x

1
(18)
aF1 = 2 (2) + (3) H (1, x) + (2)H (1, 1, x) + 4 3(2) H (1, 0, x)
x
6H (1, 0, 0, x) + 2H (1, 1, 0, x) + 7H (1, 0, 0, 0, x)

(204)
3H (1, 1, 0, 0, x) + H (1, 1, 1, 0, x) .
The above amplitude has a simple ultraviolet pole coming from the massless bubble.

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D3 D9 D11 D13

i Fi(19) + O 2 ,
(205)
(206)
i=0
where:

2
(2)H (1, x) H (0, 0, 1, x) + H (1, 0, 0, x) 2H (1, 0, 1, x) ,
x
(207)

2
(19)
2 (2) (3) H (1, x) + (2) H (1, 0, x) + H (1, 1, x)
aF1 =
x
2H (0, 0, 1, x) 4H (1, 0, 1, x) + 2H (1, 0, 0, x)
aF0(19) =
+ 4H (0, 0, 1, 1, x) H (0, 0, 0, 1, x) + 8H (1, 0, 1, 1, x)

3H (1, 0, 0, 1, x) 3H (1, 0, 0, 0, x) 2H (1, 1, 0, 1, x)

+ H (1, 1, 0, 0, x) .
(208)
The finite part of the above amplitude is also given in [23].

=

=
2(4D)
2
a

d D k1 d D k2
2
1
i=1
(20)
i Fi
1
D2 D3 D4 D5 D12

+ O 2 ,
(209)
(210)
39
where:
1
(20)
aF1 = H (1, 0, x),
x
1
(20)
aF0 = (2)H (1, x) + H (0, 1, 0, x) + 2H (1, 0, x) + H (1, 1, 0, x)
x

2H (1, 0, 0, x) ,

1
aF1(20) =
2 (2) + 3 (3) H (1, x) + (2)H (0, 1, x) + 4H (1, 0, x)
x
+ (2)H (1, 1, x) + H (0, 0, 1, 0, x) + 2H (0, 1, 0, x)
(211)
(212)
2H (0, 1, 0, 0, x) + H (0, 1, 1, 0, x) 4H (1, 0, 0, x)

+ 4H (1, 0, 0, 0, x) + H (1, 0, 1, 0, x) + 2H (1, 1, 0, x)

2H (1, 1, 0, 0, x) + H (1, 1, 1, 0, x) .
(213)
The above amplitude has a simple ultraviolet pole coming from the nested bubble.
5. Six-denominator diagrams
In this section we present the results of all the six-denominator diagrams with up to
one massive line entering the EW form factor. The ladder and vertex insertion diagrams
the planar topologiesare all reducible, while both the crossed laddersthe nonplanar
topologiesare MIs, i.e., cannot be reduced to simpler topologies by using the ibp
identities. Even though these diagrams have a different diagrammatic status, we present
them together because they are the most important ones and may be used for reference.
Series representations for the pole and finite parts of the ladder and crossed ladder diagrams
have been given in [23,24]. The authors compute a large number of terms in a small
momentum expansion of the diagrams and then fit the results to an assumed basis of
functions and transcendent constants. In the above paper, formulas for the resummation
of the series in terms of Nielsen polylogarithms [4,5] are also reported. By using them,
we could check that the pole parts of our results agree with theirs. Finally, the leading
logarithms of all the six-denominator amplitudes have been computed in [25] by means
of a pole approximation for the vector boson propagators; we are in agreement with them
also.
5.1. Planar topologies

= 2(4D)

=
2
a
D D
d k1 d k2
2
0
i=4
1
D1 D2 D4 D5 D7 D8
i Ai + O(),
(214)
(215)
40
where:
1
,
(216)
4x 2
1
a 2 A3 = 2 H (0, x),
(217)
2x

1
a 2 A2 = 2 (2) + H (0, 0, x) ,
(218)
x

1
a 2 A1 = 2 5 (3) 2 (2)H (0, x) 2H (0, 0, 0, x) ,
(219)
x

1 11 2(2)
a 2 A0 = 2
10 (3)H (0, x) + 4(2)H (0, 0, x) + 4H (0, 0, 0, 0, x) .
5
x
(220)
a 2 A4 =
The above diagram has also been computed in [26] using Feynman parameters and
in [27] using dispersion relations; we computed it by means of the reduction to simpler
(massless) amplitudes, the latter computed with Feynman parameters. We are in complete
agreement with these references. The leading singularity of this diagram is a pole in of
fourth order or, equivalently, a forth power in log(x) in the finite part for x 0. We have
indeed a soft and a collinear singularity for each loop, each one contributing by a simple
pole to the amplitude.

= 2(4D)

=
2
a
D D
d k1 d k2
1
D2 D4 D5 D7 D8 D12
(221)
2
B0 + O(),
(222)
where:
a 2 B0 =
1
6 (3)H (1, x) 2(2)H (1, 1, x) + H (1, 0, 1, 0, x)
x2

2H (1, 1, 0, 0, x) 2H (1, 1, 1, 0, x) .
(223)
As expected on the basis of ones physical intuition, the above diagram is infrared finite.
The massive vector boson cannot indeed propagate for large distances and the same is true
for the photon, which is trapped inside the external loop.

= 2(4D)

=
2
a
D D
d k1 d k2
2
0
i=2
1
D1 D4 D5 D7 D8 D13
i Ci + O(),
(224)
(225)
41
where:
1
(226)
H (1, 0, x),
x2

1
a 2 C1 = 2 (2)H (1, x) + 4H (1, 0, 1, x) + 3H (1, 1, 0, x) 3H (1, 0, 0, x) ,
x
(227)

1
a 2 C0 = 2 2 (3)H (1, x) + 7(2)H (1, 1, x) 3(2)H (1, 0, x)
x
16H (1, 0, 1, 1, x) + 6H (1, 0, 0, 1, x) + 7H (1, 0, 0, 0, x)
a 2 C2 =
+ 3H (1, 0, 1, 0, x) + 4H (1, 1, 0, 1, x) 5H (1, 1, 0, 0, x)

+ 9H (1, 1, 1, 0, x) .
(228)
Unlike Eq. (221), in the above diagram the photon is not trapped by the massive line
and can propagate to asymptotic times. As a consequence, we have a double infrared pole,
having a similar dynamical origin to the one in the massless one-loop vertex.

= 2(4D)

=
2
a
D D
d k1 d k2
2
1
D1 D2 D5 D7 D8 D15
i Ei + O(),
(229)
(230)
i=2
where:

1
1 1
(2) + H (0, 1, x) ,
a E2 =
(231)
2 x (1 + x)

1
1 1
2
(3) 2(2) H (0, x) + H (1, x)
a E1 =
2 x (1 + x)

(232)
2H (1, 0, 1, x) 2H (0, 1, 0, x) + H (0, 0, 1, x) ,
2

1
(2)
1
a 2 E0 =
(3) H (0, x) + H (1, x)

x (1 + x)
10

+ (2) 2H (0, 0, x) + 2H (1, 1, x) + 2H (1, 0, x) 3H (0, 1, x)
2
4H (0, 1, 1, 1, x) + 2H (0, 1, 0, 1, x) + H (0, 1, 0, 0, x)

1
+ H (0, 0, 0, 1, x) H (0, 0, 1, 0, x) + 2H (0, 1, 1, 0, x)
2
+ 2H (1, 0, 1, 0, x) H (1, 0, 0, 1, x)

(233)
+ 2H (1, 1, 0, 1, x) .
42

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D2 D3 D4 D5 D6
i Gi + O(),
(234)
(235)
i=3
where:
3
,
4x 2

3
a 2 G2 = 2 1 + H (0, x) ,
2x

3
2
a G1 = 2 1 (2) + H (0, x) + H (0, 0, x) ,
x

3
a 2 G0 = 2 2 2 (2) + 3(3) + 2 1 (2) H (0, x) + 2H (0, 0, x)
x

+ 2H (0, 0, 0, x) .
a 2 G3 =
(236)
(237)
(238)
(239)
The above diagram has also been computed in [26] using Feynman parameters and in
[27] using dispersion relations; we have computed it by means of the reduction to simpler
(massless) amplitudes, the latter computed with Feynman parameters in Section 3. We are
in complete agreement with the previous results. The massless vertex correction has at
most a triple pole in and is therefore less singular than the massless ladder or crossed
ladder (see later). As well known, the vertex correction is indeed neglected in the double
logarithmic analysis of QED form factors.

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D2 D3 D4 D5 D17
i Ii + O(),
(240)
(241)
i=3
where:
3
,
(242)
4x

1
a 2 I2 = 2 (1 + x)H (1, x) xH (0, x) ,
(243)
2x

1
a 2 I1 = 2 (2)x 2xH (0, 1, x) + 2(1 + x)H (1, 1, x) ,
(244)
x
1
a 2 I0 = 2 (2)(1 + x)H (1, x) (2)xH (0, x) xH (0, 0, 0, x)
x

+ 2x 4H (0, 1, 1, x) H (0, 0, 1, x)

+ (1 + x) H (1, 0, 0, x) + 2H (1, 0, 1, x) 8H (1, 1, 1, x) .
a 2 I3 =
(245)

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D2 D3 D4 D5 D6 D12
i Ji + O(),
43
(246)
(247)
i=2
where:
(2)
,
2x

1
a 2 J1 =
(3) 2 (2) H (0, x) + H (1, x) 2H (0, 1, 0, x)
2x

2H (1, 1, 0, x) ,

1 2 (2)
2
a J0 =
+ (3) H (0, x) + H (1, x)
x
10

(2) 2H (0, 0, x) 3H (0, 1, x) + 2H (1, 0, x) 3H (1, 1, x)
a 2 J2 =
(248)
(249)
H (0, 0, 1, 0, x) + 2H (0, 1, 0, 0, x) 3H (0, 1, 1, 0, x)

H (1, 0, 1, 0, x) + 2H (1, 1, 0, 0, x) 3H (1, 1, 1, 0, x) .

= 2(4D)

=
2
a
D D
d k1 d k2
2
1
D1 D2 D42 D5 D7
i Ki + O(),
(250)
(251)
(252)
i=2
where:
3
,
2x 2

3
a 2 K1 = 2 1 2H (0, x) ,
2x

3
2
a K0 = 2 3 2 (2) 2H (0, x) + 4H (0, 0, x) .
2x
a 2 K2 =

= 2(4D)

=
2
a
D D
d k1 d k2
2
0
i=2
1
D1 D42 D5 D7 D13
i Li + O(),
(253)
(254)
(255)
(256)
(257)
44
where:

4
1
1
,
a L2 =
(258)
2x
x

4
1
1 5
2
+ 1
a 2 L1 =
(259)
H (0, x) 1 + + 2 H (1, x) ,
2x x
x
x x

4
8
1 17
+ 2 (2) 1
+ 3
H (0, x)
a 2 L0 =
4x x
x
x

3
4
4
3 1 + + 2 H (1, x) 2 1 +
H (0, 0, x) 6H (0, 1, x)
x x
x

1
2
+ 8 1 + + 2 H (1, 1, x) .
(260)
x x
2

=

=
2(4D)
2
a
D D
d k1 d k2
2
1
D2 D42 D5 D7 D12
i Mi + O(),
(261)
(262)
i=2
where:
1
(263)
,
2x

1
1
1
a 2 M1 = +
(264)
+
H (0, x),
2x
x (1 x)

1
1
1
2(2) + H (0, x) 2H (0, 0, x)
+
7 2 (2) +
a 2 M0 =
2x
x (1 x)

+ 2H (1, 0, x) .
(265)
a 2 M2 =

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D12 D2 D3 D4 D5
i Ni + O(),
(266)
(267)
i=2
where:
3
,
4x 2
3
a 2 N1 = 2 H (0, x),
2x

3
a 2 N0 = 2 1 + H (0, 0, x) .
x
a 2 N2 =
(268)
(269)
(270)

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D2 D3 D4 D5 D12
i Oi + O(),
45
(271)
(272)
i=3
where:
a 2 O3 =
1
,
4x
1
(273)

1 H (0, x) ,
2x

1
a 2 O1 = 1 H (0, x) + H (0, 0, x) H (1, 0, x) ,
x
1
2
a O0 = 2 2 (3) 2H (0, x) + (2)H (1, x) + 2H (0, 0, x)
x
+ 2H (1, 0, x) 2H (0, 0, 0, x) + H (0, 1, 0, x) 2H (1, 0, 0, x)

+ H (1, 1, 0, x) .
a 2 O2 =

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D12 D4 D5 D13
i Pi + O(),
(274)
(275)
(276)
(277)
(278)
i=2
where:
2
(279)
,
x2
2
aP1 = 2 H (0, x),
(280)
x

2
aP0 = 2 1 (2) + H (0, 0, x) .
(281)
x
The above amplitude is actually a 5 denominator one; we include it here just for practical
convenience.
aP2 =
5.2. Crossed ladders

The massless crossed ladder is a MI, i.e., it cannot be reduced to simpler (massless)
topologies by means of ibp identites. Its direct computation is rather involved so we report
the result as taken from Ref. [26]. We checked Gonsalves result for the crossed ladder with
Eq. (3.4) in [27]. If we include in the latter equation the dimensional factor x 22 (in our
46
notation), a complete agreement between the two computations is found.12

= 2(4D)

=
2
a
D D
d k1 d k2
2
1
D1 D2 D3 D4 D6 D11
i Qi + O(),
(282)
(283)
i=4
where:
1
(284)
,
x2
2
a 2 Q3 = 2 H (0, x),
(285)
x

1
a 2 Q2 = 2 7 (2) 4H (0, 0, x) ,
(286)
x

1
a 2 Q1 = 2 27 (3) 14 (2)H (0, x) + 8H (0, 0, 0, x) ,
(287)
x

1 57 2(2)
2
a Q0 = 2
54(3)H (0, x) + 28(2)H (0, 0, x) 16H (0, 0, 0, 0, x) .
x
5
(288)
a 2 Q4 =
Note that the above diagram is the only one having a fourth order (infrared) pole in
together, as we have seen, with the planar ladder.

= 2(4D)

=
2
a
D D
d k1 d k2
2
1
D1 D2 D3 D6 D11 D15
i Ri + O(),
(289)
(290)
i=2
where:

1
1 1
a R2 =
+
(2) 2H (0, 1, x) ,
2 x (1 x)

1
1 1
2
a R1 =
+
(3) 2(2) H (0, x) 2H (1, x)
2 x (1 x)
+ 8H (0, 1, 1, x) 4H (0, 0, 1, x) + 2H (0, 1, 0, x)

8H (1, 0, 1, x) ,
2
(291)
(292)
12 Note that the definition of in the above cited Van Neerven paper is different from ours:
V N = 2.
47
2

1
(2)
1
+
(3) H (0, x) 2H (1, x)

x (1 x)
10

+ (2) 2H (0, 0, x) + 8H (1, 1, x) 4H (1, 0, x) 6H (0, 1, x)

+ 7H (0, 1, x) 16H (0, 1, 1, 1, x) + 8H (0, 1, 0, 1, x)

a 2 R0 =
2H (0, 1, 0, 0, x) + 8H (0, 0, 1, 1, x) 2H (0, 0, 0, 1, x)

+ H (0, 0, 1, 0, x) 8H (0, 1, 0, 1, x) H (0, 1, 0, 0, x)
+ 3H (0, 1, 1, 0, x) + 16H (1, 0, 1, 1, x) 8H (1, 0, 0, 1, x)

+ 4H (1, 0, 1, 0, x) 16H (1, 1, 0, 1, x) .
(293)
The differential equation for the above amplitude is a rather lengthy expression, as it
involves fourteen subtopologies.
6. Large momentum expansion, |s| m2
In this section we give the asymptotic expansions of the six denominator scalar integrals,
i.e., the expansion in powers of 1/x and L = log(x) up to the order 1/x 4 included. As
already noted in the introduction, these expansions are relevant for the study of the structure
of the infrared logarithms coming from multiple emission. The exact expression of the
massless diagrams involves only an overall power of x and powers of L, so their expansion
coincides with the former and is not duplicated. The coefficients not explicitly given are
zero.

=
2
a
2
4
j =2

1 0
1
B
+
O
,
j
j
x
x5
(294)
where:
37 2(2)
1
4 (3)L + 2(2)L2 + L4 ,
5
24

1
1
2 0
a B3 = 10 2 (2) + 4 (3) 4 1 + (2) L L2 L3 ,
6

2
5
1 2
43 (2)
1
2 0
+ 2 (3)
+ 2(2) L L L3 .
a B4 = +
8
2
2
8
12
0
a 2 B2
=

=
2
a
2
4
j =2
1
xj
i=2

i
Cj
(295)
(296)
(297)

1
+O 5 ,
x
(298)
where:
1
2
= 2 (2) + L2 ,
a 2 C2
2
(299)
48
1
1
a 2 C2
= 3 (3) (2)L L3 ,
3
2 (2)
61
1
0
11 (3)L + 2(2)L2 + L4 ,
a 2 C2
=
10
6
2
= 1 L,
a 2 C3
(300)
(301)
(302)
1
a 2 C3
= (2) L + L2 ,
(303)

2
0
a 2 C3
= 26 + 3 (2) + 11(3) 2 3 + 2(2) L + 3L2 L3 ,
3
1 1
2 2
a C4 = L,
4 2
1
5 (2) 7
1
a 2 C4
L + L2 ,
= +
4
2
4
2

5
13
117 17 (2) 11(3)
1
2 0
+
+
+ 2(2) L + L2 L3 .
a C4 =
8
4
2
2
4
3
(304)
(305)
(306)
(307)
The expansion above is in agreement with that obtained in [28] by using a kinematic region
decomposition of the integrand.

=
2
a
2
4
j =2
1
xj
i=2

i
Ej

+O

1
,
x5
(308)
where:
1
2
a 2 E2
= (2) + L2 ,
4
1
5 (3) 3 (2)
2 1
L L3 ,
a E2 =
2
2
4
2
7(2) 2
7 (2)
7
0
8 (3)L +
L + L4 ,
=
a 2 E2
4
4
48
1
1
2
a 2 E3
= (2) L2 ,
2
4
5 (3) 3(2)
1
1
1
2 1
+
L L2 + L3 ,
a E3 = 2 (2)
2
2
2
2
4
7 2 (2)
17
0
a 2 E3
5(3) 4L + 3(2)L + 8(3)L
= + 2 (2)
2
4
7 (2) 2 1 3
7
L + L L4 ,
4
2
48
1
5
2
= + (2) + L2 ,
a 2 E4
8
4
5(3) 3(2)
3
1
1
1
L + L2 L3 ,
= + 3 (2) +
a 2 E4
16
2
2
4
4
(309)
(310)
(311)
(312)
(313)
(314)
(315)
(316)
0
=
a 2 E4
9(2)
297 (2) 7 2 (2) 15(3) 11
+
+
+ L
L 8(3)L
32
2
4
2
2
2
1
7 (2) 2 3 3
7
+ L2 +
L L + L4 ,
2
4
4
48

=
2
a
2
4
j =1
1
xj

i
i Ij
i=3
49
(317)

1
+O 5 ,
x
(318)
where:
3
3
a 2 I1
= ,
4
1
a 2 I1
= (2),
(319)
(320)
= 4 (3),
1 1
2
a 2 I2
= L,
2 2
1
a 2 I2
= 2 + 2L + L2 ,
0
a 2 I1
(321)
(322)
(323)
5
5
0
= 5 + 3 (2) 4 (3) 5L + 3(2)L L2 L3 ,
a 2 I2
2
6
1
2
a 2 I3
= ,
4
1
2 1
a I3 = + L,
2
5
21 3 (2) 1
2 0
a I3 =
+
+ L L2 ,
8
2
4
4
1
2
a 2 I4
= ,
12
2 1
2 1
a I4 = L,
9 3
5
211 (2) 47
2 0
a I4 =
L + L2 ,
216
2
36
12

=
2
a
2
4
j =1
1
xj
i=2

i
i Jj
(324)
(325)
(326)
(327)
(328)
(329)
(330)

1
+O 5 ,
x
(331)
where:
(2)
,
2
(3)
1
a 2 J1
,
=
2
2
a 2 J1
=
(332)
(333)
50
0
a 2 J1
=
6 2 (2)
,
5
(334)
1
1
a 2 J2
= 1 + (2) + L + L2 ,
2
(335)
2
0
a 2 J2
= 4 + (2) 4 (3) 4L + (2)L 2L2 L3 ,
3
1 2
5 (2) 1
2 1
L+ L ,
a J3 = +
8
2
4
4
9
(2)
3
13 7 (2)
1
2 0
+
2 (3) + L +
L + L2 L3 ,
a J3 =
8
4
4
2
4
3
(2)
7
1
59
1
+
L + L2 ,
=
a 2 J4
108
3
18
6
29 (2) 4(3) 155
(2)
37
67
2
0
+
+
L+
L + L2 L3 ,
a 2 J4
=
648
18
3
108
3
36
9

=
2
a
2
4
j =1
1
xj

Lij

+O
i=2

1
,
x5
(336)
(337)
(338)
(339)
(340)
(341)
where:
1
a 2 L2
1 = 2 ,
a 2 L01 = (2),
(342)
(343)
a 2 L2
2 = 2,
a 2 L1
2
a 2 L02
(344)
= 2 3L,
(345)
= 5 2 (2) 3L + 3L ,
3 1
a 2 L1
3 = 4 2 L,
3
a 2 L03 = 2 + L + L2 ,
4
1
2 1
a L4 = ,
6
1 2
2 0
a L4 = + L,
6 3
2

=
2
a
2
4
j =1
1
xj
i=2
(346)
(347)
(348)
(349)
(350)

i
Mj

1
+O 5 ,
x
(351)
where:
1
2
= ,
a 2 M1
2
(352)
1
1
a 2 M1
= ,
2
3
2 0
a M1 = (2),
2
2 1
a M2 = L,
(353)
(354)
(355)
0
a M2
= 2 (2) L + 2L2 ,
1
= L,
a 2 M3
2 0
a M3 = 2 + 2 (2) 3L + 2L2 ,
1
= L,
a 2 M4
2
0
=
a 2 M4
(356)
(357)
(358)
(359)
5
+ 2 (2) 4L + 2L2 ,
2

=
2
a
51
2
4
j =1
1
xj
(360)

i
Oj
i=3

1
+O 5 ,
x
(361)
where:
1
3
= ,
a 2 O1
4
1 1
2 2
a O1 = L,
2 2
2 1
a O1 = 1 2 (2) L,
(362)
(363)
(364)
0
= 2 4 (2) (3) 2L (2)L,
a O1
1
= 1 + L,
a 2 O2
2
(365)
(366)
3
0
= 1 (2) L L2 ,
a 2 O2
2
1
1
1
a 2 O3
= + L,
4 2
(2) 3
3
7
0
a 2 O3
+ L L2 ,
=
8
2
4
4
1
1
1
a 2 O4 = + L,
9 3
(2) 5
1
25
0
a 2 O4
+ L L2 ,
=
36
3
6
2

=
2
a
2
4
j =2
1
xj
i=2
(367)
(368)
(369)
(370)
(371)

i
Rj

1
+O 5 ,
x
(372)
52
where:
(2) 1 2
+ L ,
(373)
2
2
5
21 (3)
1
+ 3 (2)L L3 ,
=
a 2 R2
(374)
2
6
19(2) 2 13 4
163 2(2)
0
+ 36 (3)L
L + L ,
=
a 2 R2
(375)
10
2
24
(2) 1 2
2
a 2 R3
(376)
+ L ,
= 1 +
2
2
21 (3)
5
1
a 2 R3
(377)
= 8 + 2 (2)
+ 9L + 3(2)L + 2L2 L3 ,
2
6
163 2
0
a 2 R3
= 34 + 15 (2)
(2) 42(3) 2L + 12(2)L + 36(3)L
10
19 (2) 2 10 3 13 4
(378)
13L2
L L + L ,
2
3
24
3 (2) 1 2
2
a 2 R4
(379)
+ L ,
= +
4
2
2
21(3) 37
15
5
1
+ 3 (2)
+ L + 3(2)L + 3L2 L3 ,
a 2 R4
(380)
=
2
2
4
6
31
453 87 (2) 163 2(2)
0
a 2 R4
+
63(3) + L + 18(2)L + 36(3)L

=
8
4
10
2
39 2 19 (2) 2
13 4
3
L 5L + L .
(381)
L
4
2
24
2
a 2 R2
=
7. Small momentum expansions, |s| m2

In this section we present the small momentum expansions of the six-denominator
diagrams, i.e., expansions in powers of x and L = log(x) up to first order in x included.
As in the previous section, we do not consider the massless scalar integrals since their
expansion is trivial. The coefficients not explicitly given are zero.

=
2
a
2

x j Bj0 + O x 2 ,
(382)
j =1
where:
0
= 6 (3),
a 2 B1
1
a 2 B00 = 3 (2) + 3 (3) + 2L L2 ,
2
17
5
1
2 0
a B1 = (2) + 2 (3) + L L2 ,
12
4
2
(383)
(384)
(385)

=
2
a
2

x
j =1

Cji

+ O x2 ,
53
(386)
i=2
where:
2
a 2 C1
= 1 L,
(387)
3
1
= 3 (2) 3L + L2 ,
a 2 C1
2
(388)
7
7
0
a 2 C1
= 7 3 (2) 2 (3) 7L + 3(2)L + L2 L3 ,
2
6
1 1
a 2 C02 = L,
4 2
3
5 (2) 9
2 1
a C0 =
L + L2 ,
8
2
4
4
49
3(2)
17
89 17 (2)
7
2 0
(3) L +
L + L2 L3 ,
a C0 =
16
4
8
2
8
12
1 1
a 2 C12 = L,
9 3
(2) 11
1
31
a 2 C11 =
L + L2 ,
36
3
6
2
59
4709 23 (2) 2(3) 160
7
2 0
L + (2)L + L2 L3 ,
a C1 =
648
6
3
27
36
18

=
2
a
2
1
j =1

x

Eji

+ O x2 ,
(389)
(390)
(391)
(392)
(393)
(394)
(395)
(396)
i=2
where:
(2)
,
2
(3)
1
a 2 E1
=
(2)L,
2
2 (2)
0
(3)L + (2)L2 ,
=
a 2 E1
10
1 (2)
,
a 2 E02 =
2
2
(3)
5
L + (2)L,
a 2 E01 = (2)
2
2
1
13
1
5 (2) + 2 (2) (3) 3L + 2(2)L + (3)L + L2 ,
a 2 E00 =
2
10
2
5 (2)
,
a 2 E12 = +
8
2
2
a 2 E1
=
(397)
(398)
(399)
(400)
(401)
(402)
(403)
54
53 3 (2) (3) 5
+
+
+ L (2)L (2)L2 ,
16
2
2
4
2 (2)
333
29
(2)
3(3) 39
a 2 E10 =
+
+
L 3(2)L (3)L
32
4
10
2
8
5 2
2
L + (2)L ,
8
a 2 E11 =

=
2
a
2

x
j =1

i Iji

+ O x2 ,
(404)
(405)
(406)
i=3
where:
3
3
= ,
a 2 I1
4
1 1
2 2
a I1 = + L,
2 2
1
a 2 I1
= (2),
(407)
(408)
(409)
1
1
0
a 2 I1
= 1 + (2) L (2)L + L2 L3 ,
2
6
1
2 2
a I0 = ,
4
a 2 I01 = 1,
1
1 (2) 3
L + L2 ,
a 2 I00 = +
8
2
4
4
1
2 2
a I1 = ,
12
1
2 1
a I1 = ,
2
5
1
215 (2)
2 0
+ L L2 ,
a I1 =
216
6
36
12

=
2
a
2
1
j =1

x

i Jji

+ O x2 ,
(410)
(411)
(412)
(413)
(414)
(415)
(416)
(417)
i=2
where:
(2)
,
2
(3)
1
(2)L,
=
a 2 J1
2
2 (2)
0
(3)L + (2)L2 ,
a 2 J1
=
10
2
a 2 J1
=
(418)
(419)
(420)
a 2 J01 = 2 (2) L,
(421)
= 9 5 (2) (3) 5L + 2(2)L + L ,

(2) 3
L,
a 2 J11 = 1
2
4
3
81 11 (2) (3) 27
2 0
L + (2)L + L2 ,
a J1 =
16
4
2
8
4
a 2 J00

=
2
a
2
55

x
j =2

Lij

+ O x2 ,
(422)
(423)
(424)
(425)
i=2
where:
a 2 L2
2 = 2,
a 2 L1
2
a 2 L02
(426)
= 2 2L,
(427)
= 2 2 (2) 2L + L
1
a 2 L2
1 = 2 ,
3 1
a 2 L1
1 = 4 + 2 L,
7 (2) 3
+ L
a 2 L01 = +
8
2
4
1
2 1
a L0 = ,
6
1
a 2 L00 = ,
2
1
2 1
a L1 = ,
24
11
2 0
a L1 = ,
48

=
2
a
2
1
j =1
(429)
(430)
1 2
L ,
4

x
(428)
(431)
(432)
(433)
(434)
(435)

Mji

+ O x2 ,
(436)
i=2
where
1
2
a 2 M1
= ,
2
1
1
a 2 M1
= + L,
2
3
0
a 2 M1
= + (2) + L L2 ,
2
(437)
(438)
(439)
56
a 2 M01 = L,
(440)
a M00 = 2 + 2 (2) + 3L L2 ,
a 2 M11 = L,
(441)
(442)
5
a 2 M10 = + 2 (2) + 4L L2 ,
2

=
2
a
2

x
j =1
(443)

Oji

+ O x2 ,
(444)
i=3
where:
1
3
a 2 O1
= ,
4
1
1
2
a 2 O1
= L,
2 2
1
1
a 2 O1
= 1 L + L2 ,
2
(445)
(446)
(447)
1
0
a 2 O1
= 2 2 (3) 2L + L2 L3 ,
3
a 2 O01 = 1 + L,
(448)
(449)
= 6 + (2) + 5L L ,
1 1
a 2 O11 = + L,
4 2
1
7 (2) 9
+ L L2 .
a 2 O10 = +
4
2
4
2
a
O00

=
2
a
2
1
j =1

x
(450)
(451)
(452)

Rji

+ O x2 ,
(453)
i=2
where:
(2)
,
2
(3)
1
(2)L,
=
a 2 R1
2
2(2)
0
(3)L + (2)L2 ,
a 2 R1
=
10
(2)
a 2 R02 = 1 +
,
2
(3)
+ L (2)L,
a 2 R01 = 4 + 2 (2) +
2
2
=
a 2 R1
(454)
(455)
(456)
(457)
(458)
a 2 R00 = 14 + 5 (2)
3
2 (2)
+ 2(3) + 7L 4(2)L (3)L L2 ,
10
2
3 (2)
a 2 R12 = +
,
4
2
(3) 5
+ L (2)L + (2)L2 ,
2
4
2 (2)
53
(2)
77
187
+
+ 3(3) + L 6(2)L (3)L

a 2 R10 =
8
4
10
8
11 2
L + (2)L2 .
8
a 2 R11 = 5 + 3 (2) +
57
(459)
(460)
(461)
(462)
8. Conclusions
We have presented an exact evaluation of the master integrals with at most one massive
propagator entering the electroweak form factor. Due to the rather simple threshold
structure, we have succeeded in obtaining exact representations of all the master integrals in
terms of one-dimensional harmonic polylogarithms of x. By an expansion of the harmonic
polylogarithms for large and small values of the argument, we could directly obtain the
large and small momentum expansions of the master integrals. We checked our results
with previous ones present in the literature finding complete agreement [23,2528]. The
remaining master integrals, containing two or three massive propagators, have thresholds
not only in s = 0 and s = m2 but also in s = 4m2 and their exact evaluation requires
probably the introduction of new classes of special functions.
Acknowledgements
We wish to thank E. Remiddi for discussions and for use of the C program SOLVE to
solve the linear systems generated by the ibp identities.
U.A. wishes to thank T. Gehrmann for discussions about two-loop computations and
D. Comelli, M. Ciafaloni and P. Ciafaloni for discussions about the ew form factor.
R.B. wishes to thank the Fondazione Della Riccia for supporting his permanence at
CERN, the Theory Division of CERN for the hospitality during a large portion of this
work, and the Universit di Roma La Sapienza for hospitality during the final part of this
work.
Appendix A. Propagators
In this appendix we list all the denominators entering the two-loop integrals computed
in this paper (see Section 2.2).
D1 = k12 ,
(A.1)
D2 = k22 ,
(A.2)
58
D3 = (k1 + k2 )2 ,
(A.3)
D4 = (p1 k1 ) ,
(A.4)
D5 = (p2 + k1 ) ,
(A.5)
2
2
D6 = (p2 k2 )2 ,
(A.6)
D7 = (p1 k1 + k2 ) ,
(A.7)
D8 = (p2 + k1 k2 ) ,
(A.8)
D9 = (p1 + p2 k1 ) ,
(A.9)
2
2
D10 = (p1 + p2 k2 ) ,
2
(A.10)
D11 = (p1 + p2 k1 k2 ) ,
(A.11)
D12 = k12
D13 = k22
(A.12)
+ a,
+ a,
(A.13)
D14 = (k1 + k2 ) + a,
(A.14)
D15 = (p1 k1 ) + a,
(A.15)
D16 = (p2 + k1 ) + a,
(A.16)
D17 = (p2 k2 ) + a,
(A.17)
D18 = (p1 k1 + k2 ) + a,
(A.18)
D19 = (p2 + k1 k2 ) + a,
(A.19)
D20 = (p1 + p2 k1 ) + a,
(A.20)
D21 = (p1 + p2 k2 ) + a,
(A.21)
D22 = (p1 + p2 k1 k2 ) + a.
(A.22)
2
2
2
2
2
2
2
Appendix B. One-loop amplitudes

In this appendix we present the results for the one-loop diagrams which are necessary
for the computation of the factorized two-loop amplitudes. We have recomputed them with
the method of differential equations described in the main body of the paper in terms of
HPLs. In some cases, we found that it was necessary to compute the one-loop amplitudes
up to the third order in .
B.1. Tadpole

= (4D)

=
2
a
D
d k

3
i=1
1
k2 + a

i Ai + O 4 ,
(B.1)
59
where:
Ai
= 1.
a
(B.2)
B.2. Bubbles

=

=
(4D)
2
a
D
d k

1
k 2 (p k)2

i Ei + O 4 ,
(B.3)
i=1
where:
E1 = 1,
(B.4)
E0 = 2 H (0, x),
(B.5)
E1 = 4 (2) 2H (0, x) + H (0, 0, x),

(B.6)

E2 = 8 2 (2) + (3) + (2) 4 H (0, x) + 2H (0, 0, x) H (0, 0, 0, x), (B.7)

9
E3 = 16 4 (2) 2 (2) 4(3) 8H (0, x) + 2 (2) + (3) H (0, x)

10
(B.8)
+ 4 (2) H (0, 0, x) 2H (0, 0, 0, x) + H (0, 0, 0, 0, x).
The above amplitudethe simplest massless oneis easily computed with a Feynman
parameter or with the general formula in the appendix of [7].

D
1
(4D)
=
d k 2
k [(p k)2 + a]
2
2

=
(B.9)
i Fi + O 3 ,
a
i=1
where:
F1 = 1,

F0 = 2 1 +

F1 = 4 1 +

F2 = 8 1 +

1
H (1, x),
x

1
2H (1, x) + H (0, 1, x) 2H (1, 1, x) ,
x

1
4H (1, x) + 2H (0, 1, x) 4H (1, 1, x)
x
+ 4H (1, 1, 1, x) 2H (1, 0, 1, x) 2H (0, 1, 1, x)

+ H (0, 0, 1, x) ,
(B.10)
(B.11)
(B.12)
(B.13)
60

1
8H (1, x) + 8H (1, 1, x) 4H (0, 1, x)
F3 = 16 + 1 +
x
8H (1, 1, 1, x) + 4H (1, 0, 1, x) + 4H (0, 1, 1, x)
2H (0, 0, 1, x) + 8H (1, 1, 1, 1, x) 4H (1, 1, 0, 1, x)
4H (1, 0, 1, 1, x) + 2H (1, 0, 0, 1, x) 4H (0, 1, 1, 1, x)

+ 2H (0, 1, 0, 1, x) + 2H (0, 0, 1, 1, x) H (0, 0, 0, 1, x) , (B.14)

= (4D)

=
2
a
D
d k

1
k 2 [k 2 + a]

i Gi + O 4 ,
(B.15)
i=1
where:
Gi = 1.
(B.16)
B.3. Vertices

=
4D
D
d k
1
k 2 [(p1 k)2 + a](p2 + k)2

1
1 2
=
a(1 + x)

=
2
a

i Li + O 3 ,
1
+

(B.17)
(B.18)
i=1
where:
1
aL1 = H (1, x),
(B.19)
x

1
aL0 = 2H (1, 1, x) H (0, 1, x) ,
(B.20)
x
1
aL1 = 2H (1, 0, 1, x) + 2H (0, 1, 1, x) 4H (1, 1, 1, x)
x

(B.21)
H (0, 0, 1, x) ,
1
aL2 = 8H (1, 1, 1, 1, x) 4H (1, 1, 0, 1, x)
x
4H (1, 0, 1, 1, x) + 2H (1, 0, 0, 1, x) 4H (0, 1, 1, 1, x)

+ 2H (0, 1, 0, 1, x) + 2H (0, 0, 1, 1, x) H (0, 0, 0, 1, x) . (B.22)
61
The above amplitude is not a master integral.

=

=
(4D)
2
a
D
d k

[k 2
1
+ a](p1 k)2 (p2 + k)2

i Mi + O 3 ,
(B.23)
i=0
where:
1
aM0 = H (1, 0, x),
(B.24)
x

1
aM1 = (2)H (1, x) + H (0, 1, 0, x) H (1, 0, 0, x) + H (1, 1, 0, x) , (B.25)
x

1
aM2 = 2 (3)H (1, x) (2) H (0, 1, x) H (1, 0, x) + H (1, 1, x)
x
H (0, 0, 1, 0, x) + H (0, 1, 0, 0, x) H (0, 1, 1, 0, x) H (1, 0, 0, 0, x)

H (1, 0, 1, 0, x) + H (1, 1, 0, 0, x) H (1, 1, 1, 0, x) .
(B.26)
The above amplitude does not have infrared 1/ poles because a nonvanishing mass in
the boson line acts as an infrared cutoff on the transverse momenta: kt > m [1].

=
D
d k
k 2 (p1
1
k)2 (p2 + k)2
(1 2) 1

ax
(4D)
2
a

(B.27)

i Ni + O 3 ,
(B.28)
i=2
where:
1
,
x
1
aN1 = H (0, x),
x

1
aN0 = (2) H (0, 0, x) ,
x

1
aN1 = 2 (3) (2)H (0, x) + H (0, 0, 0, x) ,
x

1 9 2
(2) 2 (3)H (0, x) + (2)H (0, 0, x) H (0, 0, 0, 0, x) .
aN2 =
x 10
aN2 =
(B.29)
(B.30)
(B.31)
(B.32)
(B.33)
62
Appendix C. Independent two-loop amplitudes

In this appendix we give the expressions of the reducible diagrams of Figs. 5, 6 and 7.

=

=
2(4D)
2
a

d D k1 d D k2
2
1
D1 D2 D9 D11

i Ri(1) + O 3 ,
(C.1)
(C.2)
i=2
where:
1
(1)
= ,
R2
2
5
(1)
R1
= H (0, x),
2
19
(1)
R0 =
(2) 5H (0, x) + 2H (0, 0, x),
2

65
(1)
5 (2) 5 (3) 19 2(2) H (0, x)
R1 =
2
+ 10H (0, 0, x) 4H (0, 0, 0, x),
R2(1) =
(C.3)
(C.4)
(C.5)
(C.6)

11 2(2)
211
19 (2)
25(3) 65 10(2) 10(3) H (0, x)
2
5

(C.7)
+ 38 4 (2) H (0, 0, x) 20H (0, 0, 0, x) + 8H (0, 0, 0, 0, x),

= 2(4D)

=
2
a

d D k1 d D k2
2
(2)
i Ri
1
D1 D9 D11 D13

+ O 3 ,
(C.8)
(C.9)
i=2
where:
1
(2)
R2
= ,
2
5
(2)
R1
= H (0, x),
2

19
1
(2)
R0 =
2 (2) 3H (0, x) 2 1 +
H (1, x) + H (0, 0, x)
2
x

1
+ 1
H (0, 1, x),
x
(C.10)
(C.11)
(C.12)

(2)
R1 =
R2(2) =

1
65
4 (2) (3) 7 (2) H (0, x) 12 1 +
H (1, x)
2
x

1
5
H (0, 1, x) + 8 1 +
H (1, 1, x)
+ 3H (0, 0, x) 1 +
x
x

1
H (0, 0, 1, x) 4H (0, 1, 1, x) ,
H (0, 0, 0, x) + 1
x
63
(C.13)

27 2 (2)
211
2 (2)
5(3) 15 3(2) 2(3) H (0, x)
2
10

1
50 + 4 (2) 1 +
H (1, x) + 7 (2) H (0, 0, x)
x

1
(19 + 2(2))
H (0, 1, x) + 48 1 +
H (1, 1, x)
17 2 (2) +
x
x

5
H (0, 0, 1, x) 4H (0, 1, 1, x)
3H (0, 0, 0, x) 1 +
x

1
12H (1, 0, 1, x) 32H (1, 1, 1, x)
+ 1+
x

1
H (0, 0, 0, 1, x) 6H (0, 1, 0, 1, x)
+ 1
x

+ 16H (0, 1, 1, 1, x) 4H (0, 0, 1, 1, x) + H (0, 0, 0, 0, x),
(C.14)

=

=
2(4D)
2
a

d D k1 d D k2
2
1
D2 D9 D10 D12

i Ri(3) + O 2 ,
(C.15)
(C.16)
i=2
where:
1
(3)
R2 = ,
2
(C.17)

1
5
(C.18)
H (1, x),
= 1+
2
x

1
19
R0(3) =
+ (2) 1 +
5H (1, x) + H (0, 1, x) 4H (1, 1, x)
2
x
(C.19)
H (0, 1, x),
(3)
R1
64

(3)
R1 =

65
1
+ 5 (2) (3) 19 + 2(2) 1 +
H (1, x)
2
x

1
H (0, 1, x) 2H (1, 1, x) + 5H (0, 1, x)
10 1 +
x

1
H (0, 0, 1, x) 4H (0, 1, 1, x) 6H (1, 0, 1, x)
1+
x

+ 16H (1, 1, 1, x) + 4H (0, 1, 1, x) H (0, 0, 1, x),
(C.20)

=

=
2(4D)
2
a

d D k1 d D k2
2
1
D2 D5 D8 D12

i Ri(4) + O 3 ,
(C.21)
(C.22)
i=2
where:
1
(4)
R2 = ,
2
3
(4)
R1 = ,
2
7
(4)
R0 = + (2),
2
15
R1(4) =
+ 3 (2) (3),
2
9 2 (2)
31
(4)
+ 7 (2) +
3(3),
R2 =
2
5

= 2(4D)

=
2
a

d D k1 d D k2
2
(C.23)
(C.24)
(C.25)
(C.26)
(C.27)
1
D1 D5 D8 D13

i Ri(5) + O 3 ,
(C.28)
(C.29)
i=2
where:
1
(5)
R2 = ,
2
1
(5)
R1 = ,
2
1
(5)
R0 = (2),
2
1
(5)
R1 = (2) + (3),
2
9 2(2)
1
(5)
+ (3),
R2 = (2)
2
5
(C.30)
(C.31)
(C.32)
(C.33)
(C.34)

=
2(4D)

=
2
a

d D k1 d D k2
2
(6)
i Ri
1
D1 D2 D7 D15

+ O 3 ,
65
(C.35)
(C.36)
i=2
where:
1
(6)
R2 = ,
2
3
(6)
R1 = ,
2
7
(6)
R0 = + (2),
2
15
(6)
+ 3 (2) (3),
R1 =
2
9 2 (2)
31
(6)
R2 =
+ 7 (2) +
3(3).
2
5

=

=
2(4D)
2
a
(C.37)
(C.38)
(C.39)
(C.40)
(C.41)

d D k1 d D k2
2
1
D2 D4 D10 D12

i Ri(7) + O 3 ,
(C.42)
(C.43)
i=2
where:
(7)
R2 = 1,
(C.44)
(7)
R1
(7)
R0
(7)
R1
(C.45)
= 3 H (0, x),
= 7 (2) 3H (0, x) + H (0, 0, x),

(C.46)

= 15 3 (2) 2 (3) 7 (2) H (0, x) + 3H (0, 0, x) H (0, 0, 0, x),
(C.47)
(7)
R2 = 31 7 (2)
9 2 (2)

2(3) 15 3(2) 2(3) H (0, x)
10
+ 7H (0, 0, x) 3H (0, 0, 0, x) + H (0, 0, 0, 0, x),

= 2(4D)

=
2
a
D D
d k1 d k2
2
1
i=2
1
D1 D4 D8 D13

(8)
i Ri + O 2 ,
(C.48)
(C.49)
(C.50)
66
where:
1
(8)
R2 = ,
(C.51)
2

1
5
1
(8)
R1 = + 1 +
(C.52)
H (1, x) + H (0, 1, x),
2
x
x

1
19
R0(8) = (2) + 1 +
5H (1, x) 4H (1, 1, x) + 3H (0, 1, x)
2
x

1
(C.53)
H (0, 1, x) 4H (0, 1, 1, x) H (0, 0, 1, x) ,
x

1
65
(8)
R1 = 5 (2) + (3) + 19 + 2(2) 1 +
H (1, x)
2
x

1
1
H (1, 1, x) + 10 + 9 + 2(2)
H (0, 1, x)
20 1 +
x
x

1
+ 1+
3H (0, 0, 1, x) + 16H (1, 1, 1, x) 6H (1, 0, 1, x)
x

12H (0, 1, 1, x) H (0, 0, 1, x) + 4H (0, 1, 1, x)
1
+ 16H (0, 1, 1, 1, x) 6H (0, 1, 0, 1, x)
x

(C.54)
4H (0, 0, 1, 1, x) + H (0, 0, 0, 1, x) .

=

=
2(4D)
2
a
D D
d k1 d k2
2
1
D1 D2 D8 D15

(9)
i Ri + O 3 ,
(C.55)
(C.56)
i=2
where:
1
(9)
R2 = ,
(C.57)
2
3
(9)
R1 = ,
(C.58)
2

1
5
R0(9) = + (2) + 1 +
(C.59)
H (1, x) H (0, 1, x),
2
x

1
1
(9)
7H (1, x) 4H (1, 1, x)
R1 = + 3 (2) (3) + 1 +
2
x

+ H (0, 1, x) 3H (0, 1, x) + 4H (0, 1, 1, x) H (0, 0, 1, x),
(C.60)
67

9 2 (2)
1
51
+ 5 (2) 3 (3) +
+ 33 + 2(2) 1 +
H (1, x)
2
5
x

1
7
28 1 +
H (1, 1, x) + 2 2(2) +
H (0, 1, x)
x
x

1
16H (1, 1, 1, x) 6H (1, 0, 1, x)
+ 1+
x

1
4H (0, 1, 1, x) H (0, 0, 1, x)
+ 2
x
16H (0, 1, 1, 1, x) + 6H (0, 1, 0, 1, x) + 4H (0, 0, 1, 1, x)
R2(9) =
H (0, 0, 0, 1, x).

=

=
2(4D)
2
a
(C.61)
D D
d k1 d k2
2
1
D1 D2 D5 D7

i R (10)i + O 3 ,
(C.62)
(C.63)
i=2
where:
1
R (10) 2 = ,
2
5
R (10) 1 = + H (0, x),
2
19
(10)
R 0 = + (2) + 5H (0, x) 2H (0, 0, x),
2

65
R (10) 1 = + 5 (2) + (3) + 19 2(2) H (0, x) 10H (0, 0, x)
2
+ 4H (0, 0, 0, x),
(C.64)
(C.65)
(C.66)
(C.67)

211
+ 19 (2) +
+ 25(3) + 65 10 (2) + (3) H (0, x)
2
5

38 4 (2) H (0, 0, x) + 20H (0, 0, 0, x) 8H (0, 0, 0, 0, x). (C.68)
11 2(2)
R (10) 2 =

= 2(4D)

=
where:
2
a
D D
d k1 d k2
2
1
i=3
1
D1 D4 D5 D13

i Ri(11) + O 2 ,
(C.69)
(C.70)
68
1
(11)
R3 = ,
(C.71)
x

1
(11)
R2
(C.72)
= 1 H (0, x) ,
x

1
(11)
R1 = 1 (2) H (0, x) + H (0, 0, x) ,
(C.73)
x

1
(11)
R0 = 1 (2) 2 (3) 1 (2) H (0, x) + H (0, 0, x) H (0, 0, 0, x) ,
x
(C.74)

2

1
9 (2)
(11)
1 (2) 2(3) H (0, x)
R1 = 1 (2) 2 (3)
x
10

+ 1 (2) H (0, 0, x) H (0, 0, 0, x) + H (0, 0, 0, 0, x) .
(C.75)

= 2(4D)

=
2
a

d D k1 d D k2
2
(12)
i Ri
1
D1 D2 D3 D9 D11

+ O 2 ,
(C.76)
(C.77)
i=0
where:
6 (3)
,
x

1 18 2
(12)
aR1 =
(2) + 12 (3) 12(3)H (0, x) .
x 5
aR0(12) =

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D2 D3 D9 D11 D12

i Ri(13) + O 2 ,
(C.78)
(C.79)
(C.80)
(C.81)
i=0
where:
(13)
aR0
1
(2)H (1, x) + 2H (1, 0, 1, x) + H (1, 0, 0, x)
x

+ 2H (0, 1, 1, x) H (0, 1, 0, x) ,
(C.82)

(13)
aR1
69

1
2 (2) 2 (3) H (1, x) 2(2)H (0, 1, x) + (2)H (1, 0, x)
x
2 (2)H (1, 1, x) 4H (1, 0, 1, x) + 2H (1, 0, 0, x)
+ 4H (0, 1, 1, x) 2H (0, 1, 0, x) + 4H (1, 1, 0, 1, x)
2H (1, 1, 0, 0, x) + 4H (1, 0, 1, 1, x) + 2H (1, 0, 1, 0, x)
2H (1, 0, 0, 1, x) 3H (1, 0, 0, 0, x) 12H (0, 1, 1, 1, x)
+ 2H (0, 1, 1, 0, x) + 4H (0, 1, 0, 1, x) + 2H (0, 1, 0, 0, x)

+ 2H (0, 0, 1, 1, x) H (0, 0, 1, 0, x) ,
(C.83)

= 2(4D)

=
2
a

d D k1 d D k2
2
(14)
i Ri
1
D1 D2 D3 D5 D17

+ O 2 ,
(C.84)
(C.85)
i=2
where:
1
(14)
aR2 = ,
2
1
(14)
aR1 = ,
2
1
(14)
aR0 = + (2),
2
1
aR1(14) = + (2) (3),
2

= 2(4D)

=
2
a
D D
d k1 d k2
2
(C.86)
(C.87)
(C.88)
(C.89)
1
D1 D2 D5 D7 D8
i Ri(15) + O(),
(C.90)
(C.91)
i=2
where:
(2)
,
2x

1
(15)
aR1 =
(3) 2 (2)H (0, x) ,
2x

1 2 (2)
(15)
aR0 =
+ (3)H (0, x) 2(2)H (0, 0, x) ,
x 10
(15)
=
aR2
(C.92)
(C.93)
(C.94)
70

= 2(4D)

=
2
a
D D
d k1 d k2
2
1
D1 D4 D5 D8 D13

i Ri(16) + O 2 ,
(C.95)
(C.96)
i=3
where:
1
,
x

1
(16)
aR2 = 1 H (0, x) ,
x

1
1
(16)
aR1 =
2 (2) H (0, x) 1 +
H (1, x) + H (0, 0, x)
x
x

+ H (0, 1, x) ,
(16)
aR3 =
(C.97)
(C.98)
(C.99)

1
1
(16)
4 (2) 2 (3) 1 (2) H (0, x) 3 1 +
H (1, x)
aR0 =
x
x

1
H (1, 1, x) H (0, 1, x)
+ H (0, 0, x) + 4 1 +
x

H (0, 0, 0, x) + H (0, 0, 1, x) 4H (0, 1, 1, x) ,
(C.100)
(16)
aR1

1
9 2 (2)
2(3) 1 (2) 2(3) H (0, x)
=
8 + (2)
x
10

1
7 + 2 (2) 1 +
H (1, x) + 1 (2) H (0, 0, x)
x

1
5 2 (2) H (0, 1, x) + 12 1 +
H (1, 1, x) H (0, 0, 0, x)
x
H (0, 0, 1, x) + 4H (0, 1, 1, x)

1
6H (1, 0, 1, x) 16H (1, 1, 1, x) + H (0, 0, 0, 0, x)
+ 1+
x
+ H (0, 0, 0, 1, x) + 16H (0, 1, 1, 1, x) 6H (0, 1, 0, 1, x)

4H (0, 0, 1, 1, x) ,
(C.101)

= 2(4D)
D D
d k1 d k2
1
D2 D4 D5 D8 D12
(C.102)

=
2
a
2
(17)
i Ri

+ O 2 ,
71
(C.103)
i=1
where:
1
(17)
= H (1, 0, x),
aR1
(C.104)
x
1
(17)
aR0 = 2 (2)H (1, x) + 2H (1, 0, x) + H (0, 1, 0, x) 2H (1, 0, 0, x)
x

(C.105)
+ 2H (1, 1, 0, x) ,

1
aR1(17) = 4 (2) + (3) H (1, x) + 2(2)H (0, 1, x) + 2 2 (2) H (1, 0, x)
x
+ 4 (2)H (1, 1, x) + 2H (0, 1, 0, x) 4H (1, 0, 0, x)
+ 4H (1, 1, 0, x) + H (0, 0, 1, 0, x) 2H (0, 1, 0, 0, x)
+ 2H (0, 1, 1, 0, x) + 4H (1, 0, 0, 0, x) + 2H (1, 0, 1, 0, x)

4H (1, 1, 0, 0, x) + 4H (1, 1, 1, 0, x) ,

= 2(4D)

=
2
a
D D
d k1 d k2
2
(18)
i Ri
1
D1 D2 D4 D8 D15
+ O(),
(C.106)
(C.107)
(C.108)
i=2
where:
1
(C.109)
H (1, x),
2x

1
(18)
aR1
(C.110)
=
2H (1, x) 4H (1, 1, x) + H (0, 1, x) ,
2x

1
2 2 + (2) H (1, x) + 2H (0, 1, x) 8H (1, 1, x)
aR0(18) =
2x
+ 16H (1, 1, 1, x) 6H (1, 0, 1, x) 4H (0, 1, 1, x)

+ H (0, 0, 1, x) ,
(C.111)
(18)
aR2 =

=

=
where:
2(4D)
2
a
D D
d k1 d k2
2
1
i=3
(19)
i Ri
1
D1 D2 D4 D5 D8

+ O 2 ,
(C.112)
(C.113)
72

(19)
aR3 =
(19)
aR2 =
(19)
aR1
=
aR0(19) =
(19)
aR1
1
,
(C.114)
2x

1
(C.115)
1 H (0, x) ,
x

1
(C.116)
2 (2) 2H (0, x) + 2H (0, 0, x) ,
x

1
4 2 (2) 5 (3) 2 2 (2) H (0, x) + 4H (0, 0, x)
x

(C.117)
4H (0, 0, 0, x) ,

2

1
11 (2)
8 4 (2)
10(3) 2 4 2(2) 5(3) H (0, x)
x
5

+ 4 2 (2) H (0, 0, x) 8H (0, 0, 0, x) + 8H (0, 0, 0, 0, x) , (C.118)

=

=
2(4D)
2
a
D D
d k1 d k2
2
(20)
i Ri
1
D1 D2 D5 D7 D15

+ O 2 ,
(C.119)
(C.120)
i=2
where:
1
(20)
= H (1, x),
aR2
(C.121)
x

1
(20)
= 2H (1, x) 2H (1, 1, x) H (1, 0, x) + H (0, 1, x) ,
aR1
x
(C.122)

1
(20)
aR0 = 4H (1, x) + 4H (1, 1, x) + 2H (1, 0, x) 2H (0, 1, x)
x
4H (1, 1, 1, x) 2H (1, 1, 0, x) + 2H (1, 0, 1, x)
2H (1, 0, 0, x) + 2H (0, 1, 1, x) + H (0, 1, 0, x)

H (0, 0, 1, x) ,
(20)
aR1
(C.123)

1
8 6 (3) H (1, x) + 4H (0, 1, x) + 2(2)H (1, 0, x)
32x
4H (1, 0, x) 8H (1, 1, x) + 4H (1, 0, 0, x)
4H (1, 0, 1, x) 4H (0, 1, 1, x) + 2H (0, 0, 1, x)
2H (0, 1, 0, x) + 4H (1, 1, 0, x) + 8H (1, 1, 1, x)
8H (1, 1, 1, 1, x) 4H (1, 1, 1, 0, x)
+ 4H (1, 1, 0, 1, x) 4H (1, 1, 0, 0, x) + 4H (1, 0, 1, 1, x)
73
+ 2H (1, 0, 1, 0, x) 2H (1, 0, 0, 1, x) 4H (1, 0, 0, 0, x)

+ 4H (0, 1, 1, 1, x) + 2H (0, 1, 1, 0, x) 2H (0, 1, 0, 1, x)
+ 2H (0, 1, 0, 0, x) 2H (0, 0, 1, 1, x) H (0, 0, 1, 0, x)

+ H (0, 0, 0, 1, x) ,
(C.124)

= 2(4D)

=
2
a

d D R1(21) d D R2(21)
2
(21)
i Ri
1
D1 D2 D3 D6 D15
+ O(),
(C.125)
(C.126)
i=2
where:
1
H (0, 1, x),
(C.127)
2x

1
(21)
aR1
(C.128)
H (0, 0, 1, x) 4H (0, 1, 1, x) ,
=
2x
1
aR0(21) =
2 (2)H (0, 1, x) + H (0, 0, 0, 1, x) 4H (0, 0, 1, 1, x)
2x

(C.129)
6H (0, 1, 0, 1, x) + 16H (0, 1, 1, 1, x) ,
(21)
aR2
=

= 2(4D)

=
2
a

d D k1 d D k2
2
(22)
i Ri
1
D1 D2 D3 D4 D6
+ O(),
(C.130)
(C.131)
i=4
where:
1
,
(C.132)
4x
1
(22)
aR3
(C.133)
= H (0, x),
2x

1
(22)
aR2
(C.134)
=
(2) 2H (0, 0, x) ,
2x

1
(22)
=
aR1
(C.135)
5 (3) 2 (2)H (0, x) + 4H (0, 0, 0, x) ,
2x

1 11 2(2)
(22)
5 (3)H (0, x) + 2(2)H (0, 0, x) 4H (0, 0, 0, 0, x) ,
aR0 =
x
10
(C.136)
(22)
aR4
=
74

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D2 D3 D4 D5

i Ri(23) + O 2 ,
(C.137)
(C.138)
i=3
where:
(23)
=
aR3
(23)
aR2
=
(23)
=
aR1
aR0(23) =
(23)
aR1
1
,
4x

1
1 H (0, x) ,
2x

1
1 H (0, x) + H (0, 0, x) ,
x

2
1 (3) H (0, x) + H (0, 0, x) H (0, 0, 0, x) ,
x

4
3 2 (2)
1 (3) H (0, x) + H (0, 0, x)
1 (3)
x
10

H (0, 0, 0, x) + H (0, 0, 0, 0, x) ,

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D2 D4 D5 D10

i Ri(24) + O 2 ,
(C.139)
(C.140)
(C.141)
(C.142)
(C.143)
(C.144)
(C.145)
i=3
where:
(24)
aR3 =
(24)
aR2
=
(24)
=
aR1
aR0(24) =
aR1(24) =
1
(C.146)
,
x

2
(C.147)
1 H (0, x) ,
x

2
2 (2) 2H (0, x) + 2H (0, 0, x) ,
(C.148)
x

4
2 (2) (3) 2 (2) H (0, x) + 2H (0, 0, x) 2H (0, 0, 0, x) ,
x
(C.149)

2 (2)
4
4 2 (2)
2(3) 2 2 (2) (3) H (0, x)
x
5

+ 2 2 (2) H (0, 0, x) 4H (0, 0, 0, x) + 4H (0, 0, 0, 0, x) , (C.150)

= 2(4D)

=
2
a

d D k1 d D k2
2
1
D1 D2 D4 D5 D15

i Ri(25) + O 2 ,
75
(C.151)
(C.152)
i=2
where:
1
(25)
aR2 = H (1, x),
x

1
(25)
aR1 = 2H (1, x) + H (1, 0, x) 2H (1, 1, x) ,
x

1
(25)
aR0 =
4 (2) H (1, x) 2H (1, 0, x) 4H (1, 1, x)
x

+ 4H (1, 1, 1, x) + 2H (1, 1, 0, x) + H (1, 0, 0, x) ,

1
aR1(25) = 2 4 (2) (3) H (1, x)
x

4 (2) H (1, 0, x) 2 4 (2) H (1, 1, x)
(C.153)
(C.154)
(C.155)
+ 2H (1, 0, 0, x) + 4H (1, 1, 0, x) + 8H (1, 1, 1, x)

H (1, 0, 0, 0, x) 2H (1, 1, 0, 0, x) 4H (1, 1, 1, 0, x)

(C.156)
8H (1, 1, 1, 1, x) ,

= 2(4D)

=
2
a

d D k1 d D k2
2
(26)
i Ri
1
D1 D2 D4 D5 D17

+ O 2 ,
(C.157)
(C.158)
i=3
where
(26)
aR3 =
(26)
aR2
=
(26)
=
aR1
aR0(26) =
(26)
aR1
1
(C.159)
,
x

1
(C.160)
1 H (0, x) ,
x

1
1 (2) H (0, x) + H (0, 0, x) ,
(C.161)
x

1
1 (2) 2 (3) 1 (2) H (0, x) + H (0, 0, x) H (0, 0, 0, x) ,
x
(C.162)

1
9 2(2)
1 (2) 2(3) H (0, x)
1 (2) 2 (3)
x
10

+ 1 (2) H (0, 0, x) H (0, 0, 0, x) + H (0, 0, 0, 0, x) .
(C.163)
76
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
P. Ciafaloni, D. Comelli, Phys. Lett. B 446 (1999) 278, hep-ph/9809321.

E. Remiddi, J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725, hep-ph/9905237.
T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 141 (2001) 296, hep-ph/0107173.
N. Nielsen, Nova Acta Leopoldiana, (Halle) 90 (1909) 123.
K.S. Klbig, J.A. Mignaco, E. Remiddi, BIT 10 (1970) 38.
G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189;
C.G. Bollini, J.J. Giambiagi, Phys. Lett. B 40 (1972) 566;
C.G. Bollini, J.J. Giambiagi, Nuovo Cimento B 12 (1972) 20;
J. Ashmore, Lett. Nuovo Cimento 4 (1972) 289;
G.M. Cicuta, E. Montaldi, Lett. Nuovo Cimento 4 (1972) 329;
R. Gastmans, R. Meuldermans, Nucl. Phys. B 63 (1973) 277.
F.V. Tkachov, Phys. Lett. B 100 (1981) 65;
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
A.V. Kotikov, Phys. Lett. B 254 (1991) 158.
E. Remiddi, Nuovo Cimento A 110 (1997) 1435, hep-th/9711188.
M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Acta Phys. Pol. B 29 (1998) 2627, hep-ph/9807119;
M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Nuovo Cimento A 111 (1998) 365, hep-ph/9805118.
T. Gehrmann, E. Remiddi, Nucl. Phys. B 580 (2000) 485, hep-ph/9912329.
R. Bonciani, P. Mastrolia, E. Remiddi, Nucl. Phys. B 661 (2003) 289, hep-ph/0301170.
R. Barbieri, J.A. Mignaco, E. Remddi, Nuovo Cimento A 11 (1972) 824;
R. Barbieri, J.A. Mignaco, E. Remddi, Nuovo Cimento A 11 (1972) 865.
E.W.N. Glover, M.E. Tejeda-Yeomans, Nucl. Phys. B (Proc. Suppl.) 89 (2000) 196, hep-ph/0010031.
S. Laporta, E. Remiddi, Phys. Lett. B 379 (1996) 283, hep-ph/9602417.
T. van Ritbergen, R.G. Stuart, Nucl. Phys. B 564 (2000) 343, hep-ph/9904240.
J.A.M. Vermaseren, Symbolic Manipulation with FORM, Version 2, CAN, Amsterdam, 1991;
J.A.M. Vermaseren, New features of FORM, math-ph/0010025.
Mathematica 4.2, Copyright 19882002 Wolfram Research, Inc.
SOLVE, by E. Remiddi.
MAPLE 7, Copyright 2001 by Waterloo Maple Software and the University of Waterloo.
J. Fleischer, A.V. Kotikov, O.L. Veretin, Nucl. Phys. B 547 (1999) 343, hep-ph/9808242.
J. Fleischer, A.V. Kotikov, O.L. Veretin, Phys. Lett. B 417 (1998) 163, hep-ph/9707492.
M. Hori, H. Kawamura, J. Kodaira, Phys. Lett. B 491 (2000) 275, hep-ph/0007329.
R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542.
W.L. van Neerven, Nucl. Phys. B 268 (1986) 453.
V.A. Smirnov, Phys. Lett. B 404 (1997) 101, hep-ph/9703357.

Multi-spin string solutions in AdS5 S 5

S. Frolov a,1 , A.A. Tseytlin a,b,2
a Department of Physics, The Ohio State University, Columbus, OH 43210, USA
b Blackett Laboratory, Imperial College, London, SW7 2BZ, UK
Received 4 June 2003; accepted 1 July 2003
Abstract
Motivated by attempts to extend AdS/CFT duality to non-BPS states we consider classical closed
string solutions with several angular momenta in different directions of AdS5 and S 5 . We find a novel
solution describing a circular closed string located at a fixed value of AdS5 radius while rotating
simultaneously in two planes in AdS5 with equal spins S. This solution is a direct generalization of
a two-spin flat-space solution where the string rotates in two orthogonal planes while always lying
on a 3-sphere. Similar solution exists for a string rotating in S 5 : it is parametrized by the angular
momentum J of the center of mass and two equal SO(6) angular momenta J2 = J3 = J in the
two rotation
planes. The remarkably simple case is of J = 0 where the energy depends on J as

E = (2J )2 + ( is the string tension or t Hooft coupling). We discuss interpolation of the
E(J ) formula to weak coupling by identifying the gauge theory operator that should be dual to the
corresponding semiclassical string state and utilizing existing results for its perturbative anomalous
dimension. This opens up a possibility of studying AdS/CFT duality in this new non-BPS sector. We
also investigate small fluctuations and stability of these classical solutions and comment on several
generalizations.
PACS: 11.25.-w
1. Introduction
Generalizing AdS/CFT duality to non-BPS string mode sector can be guided by
semiclassical considerations, as suggested in [1,2]. Identifying classical solitonic solutions
E-mail address: tseytlin@pacific.mps.ohio-state.edu (A.A. Tseytlin).
1 Also at Steklov Mathematical Institute, Moscow.
2 Also at Lebedev Physics Institute, Moscow.
doi:10.1016/S0550-3213(03)00580-7
78
S. Frolov, A.A. Tseytlin / Nuclear Physics B 668 (2003) 77110
of AdS5 S 5 sigma model carrying basic global charges is important in order to understand
the structure of the full string theory spectrum. In general, the states belonging to
representations of the isometry group SO(2, 4) SO(6) are expected to be classified by
6 = 3 + 3 charges corresponding to Cartan subalgebra generators (E, S1 , S2 ; J1 , J2 , J3 ).
Here S1 and S2 are the two spins of the conformal group (labelling representations of SO(4)
isometry of S 3 subspace) and Ji are the three angular momenta of the S 5 isometry group.
One may search for classical string solutions which have minimal energy for given values
of the 5 charges, E = E(S1 , S2 , J1 , J2 , J3 ).3 The importance of such solutions (in contrast
to various other oscillating or pulsating solutions) is that having non-zero global charges
simplifies identification of the corresponding dual CFT operators.
Particular classical string solutions with special combinations of these charges were
discussed in the past. Point-like string solution (geodesic) lying in AdS5 does not carry
intrinsic spin. Geodesic running in S 5 can carry only one component of momentum in S 5
(e.g., J = J1 ), and expansion of AdS5 S 5 string theory near such geodesic was studied
in [1]. Extended string solution describing folded closed string rotating in a plane in AdS5
carries single spin, e.g., S = S1 [2,3]. One can boost the center of mass of the string rotating
in AdS5 along a circle of S 5 getting a solution with two charges (S, J ) [4]. Alternatively,
one can construct a solution describing folded string rotating about a pole of S 5 [2]; while
it carries again only one component (say, J = J2 ) of the SO(6) spin it is not equivalent to a
point-like orbiting solution.4 An interpolating solution with the three charges (S, J, J ) was
constructed in [6]. One may think that while in general there should certainly be extended
string solutions with more spins in either or both AdS5 and S 5 spaces, they may be difficult
to construct explicitly, and also their AdS/CFT interpretation may be unclear. Here we
would like to point out that such more general solutions are actually easy to find in the
special case when the two spins S1 , S2 in AdS5 or the two of the three angular momenta in
S 5 (e.g., J2 and J3 ) are equal. The analytic form of the solution with S1 = S2 S turns out
to be much simpler than in the single spin case of [2,3]: it is a direct generalization of the
flat-space solution describing circular string rotating simultaneously in the two orthogonal
spatial planes with equal angular momenta. Such self-dual string always lies on an S 3
surface in Minkowski space, and thus it can be easily embedded into AdS5 space using
global (or covering) coordinates. The string is positioned at a fixed value of the AdS5 radius
= 0 , being stabilized by rotation. Similar string solution exists for a more general class
of 5d metrics ds 2 = g() dt 2 + d 2 + h() d3 , but not in AdS3 or AdS4 : to stabilize
the circular string at a fixed value of one needs at least two equal spin components.5
We will show that this stationary solution is stable under small perturbations if the spins
S1 = S2 = S are smaller than a critical value. The energy E is an algebraic function of
S. For a small-radius string having small S S 1, i.e., located close to the center of
3 The Casimir operators should be functions of these charges. In general, string theory knows about many
other conserved charges, being integrable. Here we concentrate on most obvious local charges.
4 The corresponding string vertex operators [5] as well as dual gauge theory operators should be different,
with the point-like (BPS) one having minimal energy (dimension) for a given value of the angular momentum
(R-charge).
5 Note that a similar (but non-rotating) winding string configuration near the boundary of AdS is stabilized
3
by the Bmn flux [7].
79
AdS5 , one finds the usual Regge trajectory relation, E = ( 4S)1/2 + . For large string
located close to the boundary of AdS5 we get E = 2S + c1 (S)1/3 + (S
1).6 The
leading E = 2S behaviour is the same as for the single-spin folded string solution [3], but
the subleading correction here is proportional to S 1/3 instead of ln S in [2].
Modulo the problem of instability of the two-spin solution at large S

, it is
natural to conjecture (see Section 5) that the Euclidean gauge theory operator in R 4 that
should be dual to this two-spin string state should have the form tr[M (D1 + iD2 )S (D3 +
iD4 )S M ] + , where M are SYM theory scalars. It would be interesting to find how
its perturbative anomalous dimension depends on large S. An interpolation formula
suggested by the semiclassical analysis is (S) = S + f ()S 1/3 + , S
1, where
f ()1 = a1 + a2 2 + , and f ()
1 = 1/3 [c1 + c2 + ], As we shall see below
(following a similar discussion in the single-spin case [4]), if one formally ignores the
instability, the 1-loop string correction to the energy of the semiclassical 2-spin solution
does scale with spin as S 1/3 . However, it seems implausible that a perturbative anomalous
dimension may depend on the spin as a fractional power. We shall comment on this further
in Section 3.
The AdS/CFT correspondence seems easier to establish in a different sector corresponding to the string solution carrying SO(6) spins, i.e., rotating in S 5 . Indeed, there exists a
similar circular string solution with two equal angular momenta in S 5 : the rotating string
moves on S 3 within S 5 , with the radius of the string or of S 3 being related to the value
of J . In addition, the center of mass of the string may be rotating along another circle
of S 5 , leading to a particular string solution with all the three S 5 charges being non-zero
(J3 = J , J1 = J2 =
J ). In the most transparent case of J = 0 when the string has maximal

size (so
that 2J ) the energy turns out to depend on J in a remarkably simple way:
E = (2J )2 + . While the solution with J = 0 turns out to be unstable, there is always
a non-trivial region of stability
when J = 0, and there are stable solutions with both J and
J being large compared to [8].
We suggest that the corresponding dual CFT operator (having minimal canonical
dimension for given values of R-charges J and J2 = J3 = J ) should be tr[(1 +

i2 )J (3 + i4 )J (5 + i6 )J ] + , where dots stand for appropriate permutations
of factors. For J = 0 the above semiclassical formula E(J ) suggests that for large J the
anomalous dimension of such operator should be = 2J + f4J() + . We conjecture that
f () should start with a -term both at strong and weak coupling, and we propose to check
this against known [9,10] perturbative results. Another interesting direction is to consider
the limit J
J and relate the resulting expression for the energy to the dimensions of
operators in the sector studied in [1].
The paper is organized as follows. In Section 2 we first describe the two-spin closed
string solution in flat space and then generalize it to AdS5 and S 5 spaces viewed as
hypersurfaces in R 2,4 and R 6 , respectively. In Section 3 we rederive the AdS5 solution
and study it in more detail using explicitly the standard set of global coordinates in AdS5
6 Interestingly, the power of that multiplies a power of S in the semiclassical correction to E 2S is the
same as the power of S only when it is equal to 1/3. One may speculate that a weak-coupling interpolation of
this formula is then E 2S = (1 + cS)1/3 .
80
space. In particular, we obtain the action for small fluctuations near thissolution and find
that the solution is stable only if the spin is bounded from above, S a , where a is of
order 1.
In Section 4 we present a similar analysis of the S 5 solution. We also note that there
exists a combined solution where string rotates in both AdS5 and S 5 and thus carries
5 charges: S1 = S2 , J = J1 and J2 = J3 . We observe that there are two branches of the

solutionwith S = 0 and J = 0 with
different dependence of the
energy on J : one for
J 12 and another for J 12 . The solution with J 12 is found to be stable
for J 38 , while the solution with large J and J = 0 turns out to be unstable. Given
that there are more general stable solutions with both J and J being large [8], we shall
assume that the instability
may not preclude one from using the remarkably simple solution
with J = 0, J
in the context of the AdS/CFT duality. For example, there may
exist a more complicated (e.g., pulsating) solution with the same quantum numbers J = 0,
J = 0 whose basic features like energy dependence on J are the same as of our simple
solution. With this motivation in mind we comment on the form of the 1-loop sigma model
correction to the energy, and conjecture about the existence of an interpolation formula for
E(J , ).
In Section 5 we discuss the structure of the (Euclidean) CFT operators which should be
dual to the semiclassical two-spin states. We mention, in particular, that the 1-loop results
of [9,10] may be used to check our conjecture that the anomalous dimension of the scalar
SYM operator with J = 0, J2 = J3 = J (with lowest dimension above the BPS bound)
should scale with J
1 as = 2J + 4J + not only at strong but also at weak
coupling. Section 6 contains some remarks on generalizations and open problems.
In Appendix A we give some details of the stability analysis for both the AdS5 and
S 5 solutions (this will be discussed in more detail in [8]). In Appendix B we derive the
quadratic fermionic part of the AdS5 S 5 GreenSchwarz action that supplements the
bosonic fluctuation actions in Sections 3 and 4. The total action should be the starting
point for a computation of 1-loop corrections to the energies of our solutions following
[4]. We check, in particular, that the fermionic mass matrix contribution to the logarithmic
divergences cancels against the bosonic one, in agreement with the conformal invariance of
the AdS5 S 5 superstring sigma model action. In Appendix C we sketch the derivation of
the bosonic fluctuation actions in the conformal gauge, and check consistency with the static gauge results for the fluctuation actions used in Sections 3 and 4. Appendix D contains
standard facts about relation between Young tableau and Dynkin labels of representations
of SU(4) group which is used to identify the operators on the gauge theory side.
2. Two-spin solution in flat space and its AdS5 or S 5 generalizations
2.1. Flat case
Let us start with closed bosonic string solutions in flat Minkowski space. In orthogonal
gauge, string coordinates are given by solutions of free 2d wave equation, i.e., by
= 0.
combinations of ein( ) , subject to the standard constraints X 2 + X 2 = 0, XX
Let us consider, in particular, a closed string (with its center of mass at rest at the origin of
81
the Cartesian coordinate system) which is rotating in the two orthogonal spatial planes 12
and 34. From the closed string equations on a 2-cylinder (, + 2 ) with Minkowski
signature in both target space and world sheet one finds
X0 = ,
X = X1 + iX2 = r1 ( )ei( ) ,
= n1 ,
= n2 ,
Y = X3 + iX4 = r2 ( )ei( ) ,
r1 = a1 sin(n1 ),

r2 = a2 sin n2 ( + 0 ) .
(2.1)
(2.2)
Here 0 is an arbitrary integration constant, and ni are arbitrary integer numbers. In what
follows we assume that ni are positive. The relation between ai , ni and follows from the
conformal gauge constraint:
2 = n21 a12 + n22 a22 .
(2.3)
The energy and the two spins are

1
E=
2
2
d X 0 = ,
(2.4)
S1 =
i
4
2

n a2
X
X = 1 1 ,
d XX
2
S2 = S1 (X Y ) =
n2 a22
,
2
(2.5)
i.e.,
2
(2.6)
(n1 S1 + n2 S2 ).

To get the states on the leading Regge trajectory (having minimal energy for given values
of the spins) one is to choose n1 = n2 = 1.
While for a special solution in (2.2) with 0 = 0 the string is folded, for generic values
of 0 it has the form of an ellipse. Another remarkable special case is when n2 0 = 2 , i.e.,
when |X| sin but |Y | cos , and
E=
n1 = n2 = n,
a1 = a2 = .
2n
(2.7)
Then the string becomes circular and, while rotating, it always lies on S 3 in R 4 space
formed by (X1 , X2 , X3 , X4 ). Indeed, the radius in R 4 then remains constant
|X(, )|2 + |Y (, )|2 = X12 + X22 + X32 + X42 =
2
.
2n2
(2.8)
In this case the two spins are equal

4n
2
S1 = S2 =
(2.9)
S,
E=
S.
4n

Thus, the X Y symmetric circular string rotating in the two orthogonal planes and
corresponding to a state on the leading Regge trajectory (n1 = n2 = 1) is described by
82
the following solution
(2.10)
X1 + iX2 = sin ei ,
X3 + iX4 = cos ei .
2
2
The crucial observation is that such solution can be easily generalised to a solution
describing a string rotating in any homogeneous space containing S 3 , in particular, AdSn
or S n with n 5.
X0 = ,
2.2. AdS5 case

Indeed, let us consider the AdS5 space described as a hypersurface in 6-dimensional
space R 2,4 :
XM XM MN XM XN = X52 X02 + X12 + X22 + X32 + X42 = 1.
(2.11)
The corresponding string sigma model Lagrangian is ( is a Lagrange multiplier field)

R2
I =
(2.12)
d d L, L = a XM a XM + (XM XM + 1).
4
The equations of motion in the orthogonal gauge then are
2 XM + XM = 0,

XM
= 0,
X M X M + XM
XM XM = 1,

X M XM
= 0.
= a XM a XM ,
(2.13)
(2.14)
A special class of solutions of these non-linear equations is characterised by the property

= const. It is natural to organize the six coordinates XM into the three 2-planes or
complex lines,
W X5 + iX0 ,
X X1 + iX2 ,
Y X3 + iX4 ,
|W |2 |X|2 |Y 2 | = 1.
(2.15)
Then it is easy to check that the following configuration is an example of the solution of
(2.13), (2.14) with = 2 = const (cf. (2.10))7
W = cosh 0 ei ,
X = sinh 0 sin ei ,
Y = sinh 0 cos ei ,
(2.16)
where 0 and are related to as follows

1
sinh2 0 = 2 .
(2.17)
2
Notice that the expressions for X and Y look exactly the same as in the flat space solution
(2.10). Indeed, since |W |2 = cosh2 0 = 1 + 2 the AdS5 constraint |W |2 |X|2 |Y 2 | = 1
is automatically satisfied.
This solution describes a circular closed string rotating in the 12 and 34 planes in
the global AdS5 time t = . It will be rederived in the next section using explicitly the
2 = 2 + 1,
7 It would be interesting to find other solutions with = const. It might even be possible to classify all such
solutions.
83
standard set of global AdS5 coordinates (t, , , , ) related to XM as follows:

W = cosh eit ,
X = sinh sin ei ,
Y = sinh cos ei .
(2.18)
It is clear that the string described by this solution has equal angular momenta in the 12
and 34 planes. In this R 2,4 embedding representation it is easy to identify the charges of
the isometry group SO(2, 4) of AdS5 that are non-vanishing on this solution. In general, the
15 rotation generators JMN of SO(2, 4) can be related to the conformal group generators
as follows (see, e.g., [11])
J = M ,
1
J4 = (K P ),
2
1
J5 = (K + P ),
2
J54 = D,
(2.19)
where , = 0, 1, 2, 3. We can identify the standard spin with S1 = M12 = J12 , the second
(conformal) spin with S2 = J34 = 12 (K3 P3 ), and finally the rotation generator in the 05
plane with the global AdS5 energy, E = J05 = 12 (K0 + P0 ).8 In the present case the only
non-vanishing charges are J50 and J12 , J34 , i.e., the energy and the two spins
E=

1 2
1 + ,
2
S S1 = S2 =
1 2 2
+ 1.
4
(2.20)
The three Casimir operators of SO(2, 4) are then expressedin terms of . Note also that
E 2 (2S)2 = ( 34 4 + 2 ) 0. For small we get E 4 S, i.e., the usual Regge

trajectory relation, while for large we have E 2S = S1 + S2 , similar to the single-spin
case in [2].
It may be worth stressing the following point. We consider classical string solutions with
large angular momenta. In quantum theory such a classical solution should correspond to a
vector of an irreducible representation labeled by the charges which would then be (half-)
integer. Since on our classical solutions S1 and S2 (and E) are the only non-vanishing
charges among the relevant generators JMN , we may use them to label representations
of the SO(4) = SU(2) SU(2) subgroup of the conformal group SO(2, 4). Assuming
that S1 S2 , the usual labels j1 and j2 of SU(2) SU(2) can be expressed in terms
of S1 and S2 as 2j1 = S1 + S2 , 2j2 = S1 S2 .9 Moreover, the fact that only S1 and S2
do not vanish means that in quantum theory the corresponding quantum state should be
the highest weight vector of an SO(4) representation. A similar remark will apply to the
case of solutions with angular momentum in S 5 we study below: a similar relation will
exist between the only non-vanishing SO(6) charges Ji = (J12 , J34 , J56 ) which are directly
(up to permutations) related to the Young tableau labels, and the Dynkin labels of SU(4)
representations.
8 After the Euclidean continuation X iX
3
0
0E and mapping to R S it is natural to classify the
representations of the conformal group in terms of maximal compact subgroup SO(4) SO(2), or SU(2)
SU(2) SO(2). Exchanging X0E with X4 exchanges the generator J54 = D with J05 = 12 (P0 + K0 ) = E.
9 The charges S and S are, in fact, directly related to the GelfandZeitlin labels of representations of SO(4).
1
2
84
2.3. S 5 case
Let us now consider an S 5 analogue of the flat-space solution (2.1). Here all is similar
to the AdS5 case, apart from the fact that the decoupled time coordinate t is introduced in
addition to the S 5 directions XA
XA XA = X12 + + X62 = |Z|2 + |X|2 + |Y |2 = 1,
Z = X1 + iX2 ,
X = X3 + iX4 ,
Y = X5 + iX6 .
The relation to the standard 5 angles ( , , 1 , 2 , 3 ) of

Z = cos ei1 ,
X = sin cos ei2 ,
S5
(2.21)
is
Y = sin sin ei3 .
(2.22)
S5
A particular solution of the

sigma model equations (which are the direct analogues of
(2.13), (2.14)) is (cf. (2.10), (2.16))
Y = sin 0 sin eiw ,
(2.23)
2
where = and and are independent parameters while the constants 0 and w are
expressed in terms of them (cf. (2.17))

1
sin2 0 = 2 2 .
w2 = 1 + 2 ,
(2.24)
2
Here the energy of the solution is E = , and in addition we have 3 non-vanishing

components of the SO(6) angular momentum tensor JAB :
t = ,
Z = cos 0 ei ,
X = sin 0 cos eiw ,
J1 = J12 ,
J2 = J34 ,
J3 = J56 ,

1 2
J J1 = 1 2 ,
2

1
J J2 = J3 =
(2.25)
2 2 2 + 1.
4
This solution describes a circular closed string rotating (with equal speeds) in the two
planes in S 3 within S 5 , with its center of mass orbiting along the orthogonal circle of S 5 .
When embedded into AdS5 S 5 it will be located at the origin = 0 of AdS5 . We shall
return to the discussion of this solution in Section 4.1.
Notice that (2.24) implies the bound
2 2 2 + 2.
(2.26)
J ,
One limiting case is = when the string is point-like and has no spin
i.e., moves
along the geodesic discussed in [1] and thus has E = J . The other is 2 = 2 2 so that
2 2.10 Here J = 0 and the string has maximal size (0 = /2), while the energy and
the two equal SO(6) angular momenta J1 , J2 take values
1 2
1
J =
(2.27)
1
,
E = = (2J )2 + 2.
2
2
10 Here one cannot take the flat-space limit in which 0.
85
This expression for E(J ) is very simple and interesting, and we shall return to the
discussion of it in Sections 4.2 and 5.
Another special case with J = 0 is = 0: here 2 2 and w = 1 so that

1 2 1

,
E = 4 J 2.
J =
(2.28)
4
2
Remarkably, here the dependence of the energy on the SO(6) spin is exactly the same as for
the leading Regge trajectory in flat space! This is not too surprising since the corresponding
string solution (2.23) is then the direct embedding of the flat space solution (2.10) into the
S 3 part of S 5 (note that the Lagrange multiplier in the S 5 analog of (2.13) vanishes when
= 0 and so XM satisfy the flat-space equations of motion).
One can also construct more general multi-spin solution which has all 5 charges being
non-vanishingS1 = S2 in AdS5 , and J = J1 and J2 = J3 in S 5 . It will be given by a
direct combination of (2.16) and (2.23) with the parameters , , 0 , 0 related by the
conformal gauge constraint as 2 = 2 + 2 sinh2 0 + 2 sin2 0 . The expressions for the
SO(2, 4) SO(6) charges are essentially the same as in (2.20) and (2.25), i.e., (see also
Section 4.2)
E=
cosh2 0 ,
J = J1 =
cos2 0 ,

1
sinh2 0 2 + 1,
2

1
J = J2 = J3 =
sin2 0 2 + 1.
2
S = S1 = S2 =
(2.29)
(2.30)
The parameters 0 , , 0 and thus can be expressed in terms of S, J and J , giving the
general expression for the energy E = E(S, J, J ).
One can also study other similar multi-charge solutions, like an interpolation between
the J, J solution on S 5 and the single-spin S1 = 0, S2 = 0 solution [2,3] in AdS5 (in
this case the radial coordinate will no longer be constant). Then one will find E =
E(S1 , J, J ) which will be a generalization to J = 0 of the expression obtained in [2,4].
3. Two-spin solution in AdS5 in global coordinates and stability

Here we shall rederive the AdS5 two-spin solution starting from a more general ansatz
with two unequal rotation parameters and then study small fluctuations near the resulting
solution.
The string action in AdS5 written in the conformal gauge in terms of independent global
coordinates x m is

R2
(AdS5 )
I =
(3.1)
(x)a x m a x n ,
.
d 2 Gmn
4
Here a = (, ), + 2 . We shall use the Minkowski signature in both target space
and world sheet, so that in conformal gauge g g ab = ab = diag(1, 1). The equations
of motion following from the action should be supplemented by the conformal gauge
constraints.
86
We shall use the following explicit parametrization of the (unit-radius) AdS5 metric
(related to the embedding coordinates of the previous section by (2.18))
2
(AdS5 )
ds AdS = Gmn
(3.2)
(x) dx m dx n = cosh2 dt 2 + d 2 + sinh2 d3 ,
5
d3 = d 2 + sin2 d 2 + cos2 d 2 .
(3.3)
This metric has translational isometries in t, , so that a general string solution should
possess the following three integrals of motion:
d
cosh2 0 t E,
E = Pt =
2
2
(3.4)
d
sinh2 sin2 0 S1 ,
S1 = P =
2
2
(3.5)
0
2
d
sinh2 cos2 0 S2 .
S2 = P =
2
(3.6)
The first integral is the spacetime energy, and the second and third ones are the spins
associated with rotations in and .
3.1. Solution
Our aim is to look for a solution describing a closed string rotating in both and , thus
generalizing the single-spin solution of [2,3]. A natural ansatz for such a solution is
t = ,
= ,
= ( ) = ( + 2),
= ,
, , = const,
= ( ) = ( + 2),
(3.7)
where and are subject to the corresponding second-order equations (prime denotes
derivative over )

= sinh cosh 2 + 2 2 sin2 2 cos2 ,
(3.8)

1

sinh2 = 2 2 sinh2 sin 2.
(3.9)
2
The first of the conformal gauge constraints

m n
5 ) (x) x
5 ) (x)x
G(AdS
x + x m x n = 0,
m x n = 0,
G(AdS
(3.10)
mn
mn
then implies that ( ) and ( ) must satisfy the following 1st order equation

2 + sinh2 2 = 2 cosh2 sinh2 2 sin2 + 2 cos2 .
(3.11)
Unfortunately, we do not know how to solve the system of non-linear Eqs. (3.8), (3.11) for
generic values of the frequencies and , so in what follows we shall assume that the
87
frequencies are equal:

= = .
(3.12)
Then (3.9) implies

=
c
sinh2
c = const.
(3.13)
The special solution of (3.9) with c = 0, i.e., = const leads us back to the single-spin case
of [2,3]: one can make a global SO(3) rotation (or redefinition of , ) to put the rotating
string in a single plane. If one assumes that = 0, one can show by a detailed analysis
that there exists no solution to (3.8), (3.11) with non-constant , i.e., one must set11
( ) = 0 = const.
(3.14)
Then Eqs. (3.8) and (3.11) take the form

2 = 2 2 ,
2 = coth2 0 2 2 .
(3.15)
The solution to these equations is given by

= n,
2 = 2n2 sinh2 0 ,

2 = n2 1 + 2 sinh2 0 = 2 + n2 .
(3.16)
Here n is an arbitrary integer representing how many times the string winds around the
-circle (cf. (2.18)). The parameter 0 determines the radius (sinh 0 ) of a circular string
rotating in S 3 . It is remarkable that one needs two rotation parameters to be non-zero in
order to stabilize the size of the string at fixed value of the AdS5 radius .
In what follows we shall consider the case of (cf. (2.17))
n = 1, i.e., = 2 sinh 0 ,
(3.17)
2 = 2 + 1.
In the flat space limit ( 0, 0 0) this corresponds to a state on the leading Regge
trajectory, i.e., having minimal energy for a given spin. The dependence on the winding
number n can be easily restored in all the equations below.
The integrals of motion (3.4), (3.5), (3.6) on this solution are given by thesame
expressions as in (2.20) (we consider the values rescaled by the string tension as
defined in (3.4), (3.5), (3.6))

1 2
2
E = cosh 0 = 1 + ,
(3.18)
2
1
1
S1 = S2 S, S = sinh2 0 = 2 2 + 1.
(3.19)
2
4
One can easily solve the cubic equation for 2 as a function of S to find the dependence of
E on S. In the case of a small string with 0 0, 0 (i.e., a string near the center of
11 Using (3.13) we get from (3.11): 2 = V (), V ()
c2 2 cosh2 + 2 sinh2 . From the form of

sinh2
the effective potential V in this equation one finds that one cannot satisfy the closed string periodicity condition
in (3.7) unless is fixed to be at zero of V .
88
AdS5 ) we get

E = 4S 1 + S + O S 2 ,
S 1.
(3.20)
This is the usual Regge trajectory relation in flat space plus the first correction due to the
curvature of AdS5 . In the case of a large string with 0
1,
1 (i.e., a long string close
to the boundary of AdS5 ) we get

3
E = 2S + (4S)1/3 + O S 1/3 ,
(3.21)
S
1.
4
Note that here the first correction to E 2S goes as S 1/3 , which is different from the
ln S behavior in the single-spin (folded rotating closed string) case in [2]. However, as we
explain in the next section, the solution with large S turns out to be unstable.
3.2. Fluctuations, stability and 1-loop correction
To compute the quadratic action for fluctuations near the above solution it is useful to
start with the NambuGoto analog of the action (3.1) and choose the static gauge
t = ,
= .
(3.22)
Let us note that the induced metric on our solution is flat:

ds22 = sinh2 0 d 2 + d 2 .
(3.23)
Shifting the remaining three fields away from their classical values
0 + ,
+ ,
+ ,
(3.24)
and expanding the NambuGoto action up to the second order in the fluctuation fields, we
get the quadratic Lagrangian for the fluctuations ,
and

1
1
L = (a )
2 cos2 2 1 + cos2 1 + 2 (a )
2
2
4

1
1
2 cos2 sin2 2 1 + 2 a
a
sin2 2 1 + sin2 1 + 2 (a )
4
2

+ 2 2 1 + 2 2 + 2 0 cos2 + 0 sin2 + 2 + 2 2 .
(3.25)
This Lagrangian takes simpler form after making the following change of variables
)
(,
(, )12

= a cos2 + sin2 ,
= b sin 2 ( ),
(3.26)
= +
(3.27)
tan ,
=
cot ,
a 2b
a 2b
1
1
a=
(3.28)
2 + 2,
b = .
2
2
12 This transformation has a very simple interpretation in terms of fluctuations of the complex fields X and Y
in (2.18): expanding near their classical values (2.16) in the static gauge where is not fluctuating and ignoring
= i sinh 0 cos ei .

= i sinh 0 sin ei ,
Y
Then , expressed in terms of X
fluctuations of we get: X
take the form (at each given ) of an O(2) rotation with angle .
and Y
89
An apparent singularity of the transformation (3.26) at = 0, /2, , 3/2 is not physical

because it is a reflection of the obvious coordinate singularity of the AdS5 metric (3.2),
(3.3) at = 0, /2, , 3/2.13 In terms of the new fields (3.25) takes the following simple
form

1
1
1
2 (a )2 (a )2 + 2 2(1 + 2 ) 0
L = (a )
2
2
2

2
2 2 + 1 2 1 + 2 2 + 2 + 2 2 .
(3.29)
The Lagrangian (3.29) can be rewritten as (omitting total derivative)
2 1
1
s
Da s Msr s r , Da s = a s + Asr
, ), (3.30)
a r , = (,
2
2

A0 = A0 = 2 1 + 2 ,
A1 = A1 = 2 + 2 ,

Msr s r = 2 2 + 2 2 + 2 + 3 2 2 ,
(3.31)
L=
where the 2d non-Abelian SO(3) gauge field Asr

a has a constant but non-vanishing

2
2
field strength F01 = 2(1 + )(2 + ). It is remarkable that in spite of the and
dependence of our background solution, the fluctuation action is quite simple, having
constant coefficients; in particular, it is simpler than the corresponding action in the onespin case in [4]. The absence of explicit dependence on will allow us to solve the
linearized equations of motion for the fluctuations.
Note that the radial fluctuation has a negative mass term in (3.31), suggesting an
instability (the Hamiltonian corresponding to (3.30) is not positive definite). However,
since it is coupled to a gauge field (i.e., is mixing with other fluctuations) the latter may, in
principle, stabilize the evolution.14 Thus the stability issue needs to be carefully studied.
This is done in Appendix A. It is found there that the solution is stable only for a certain
range of values of , i.e., for not very large values of S (see (3.19))

5 7
5
2
0 , i.e., 0 S
(3.32)
.
2
8 2
Note that for the maximal value of S, i.e., S 1.17 the value of
the spin S = S is still
large since in the semiclassical approximation it is assumed that
1.
It is of interest to compute the 1-loop superstring sigma model correction to the energy
of the two-spin solution. In principle, it can be done following the same approach as was
used in [4] in the single-spin case. It is straightforward to supplement (3.29) with the
GreenSchwarz quadratic fermionic term as in [4,12] (see Appendix B). The fermionic
contribution cancels the 2d
logarithmic divergence coming from the mass term in (3.30)
(which is proportional to r Mrr = 4 2 ). If we ignore the instability of the solution for
13 While to cover S 3 once one is usually assuming 0 /2 with and changing from 0 to 2 , here to
embed the closed string in S 3 at each fixed moment of time (3.7) we need to consider in the interval from 0
to 2 .
14 Examples of similar situations are a charged (inverted) harmonic oscillator in a magnetic field, and a
tachyon mode in AdS space.
90
large and formally consider the limit

1 in (3.29) we will get
1
1
1
2 (a )2 (a )2 + 2 2 0
L
1 (a )
2
2
2
21 2 2 2 + 2 2 .
(3.33)
Since is the only non-trivial parameter in (3.33), the 1-loop correction to the energy on
the 2d cylinder is expected to scale as (see the discussion [4]). That would imply that the
large S expansion of the energy in (3.21) is corrected at the one loop order by (cf. (3.19))
E1 1 S 1/3 , S
1. This may be consistent with the following conjecture for the
general behaviour of E(S, )

1/3
E = 2S + h() + f ()S
(3.34)
+ , S
1,
where

c1
f ()
1 = c0 + + ,
(3.35)
f ()1 = (b0 + b1 + ).
We caution, however, that the instability of the solution for large or large S may preclude
interpolation from strong to weak coupling in the large spin limit. One could still hope that
since the solution with large S should evolve into a solution which will still carry the same
spin, one may still find the classical energy behaving with spin as in (3.21). However,
one may not be able then to compute the sigma model loop corrections to the energy in a
reliable way.
4. Multi-spin string solutions in AdS5 S 5
Let us now find a similar rotating string solution in S 5 and its generalizations having
spins in both AdS5 and S 5 factors. This was already discussed in terms of the embedding
coordinates in Section 2. Here we will rederive these solutions in terms of angles of S 5 and
study some of their properties in more detail.
The bosonic part of the AdS5 S 5 string action is

R2
5
5 ) (x) x m a x n + G(S ) (y) y p a y q ,
I =
. (4.1)
d 2 G(AdS
a
a
mn
pq
4
We shall use the following explicit parametrization of the unit-radius metric on S 5 :

2
(S 5 )
ds S 5 = Gpq
(y) dy p dy q

= d 2 + cos2 d12 + sin2 d 2 + cos2 d22 + sin2 d32 .
(4.2)
This metric has three translational isometries in i , so that in addition to the three AdS5
integrals of motion (3.4), (3.5), (3.6), a general solution should also have the following
three integrals of motion depending on the S 5 part of the action:
d
cos2 0 1 J1 ,
=
2
2
J1 = P1
(4.3)
d
sin2 cos2 0 2 J2 ,
=
2
91
J2 = P2
(4.4)
J3 = P3
2
d
sin2 sin2 0 3 J3 .
=
2
(4.5)
4.1. Circular string rotating in S 5

Let us look for a solution describing a closed string located at the center = 0 of AdS5
and at a fixed value of one of the S 5 angles = 0 = const, rotating within S 3 part of S 5
(with equal frequences as in the AdS5 case), with its center of mass orbiting along a circle
of S 5 . A natural ansatz for such a solution is
t = ,
= 0,
= 0 ,
= ,
(4.6)
where , 0 , , w = const. The equations of motion for the fields and the conformal gauge
constraints then lead to the following relations between 0 , , and w
w2 = 2 + 1,
1 = ,
2 = 3 = w,
2 = 2 + 2 sin2 0 .
(4.7)
Just as in the case of the two-spin solution in AdS5 the induced metric here is flat

ds22 = sin2 0 d 2 + d 2 .
(4.8)
Taking into account that

J J1 = cos2 0 ,
J2 = J3 = J ,
we find the following equation for = (J , J )

2 + 1 = J 2 + 1 + 2J .
J
1 2
sin 0 w,
2
(4.9)
(4.10)
Since the energy E in (3.4) is equal to , we can use Eq. (4.10) to find the dependence of
the energy on the R-charges J and J
4J
(4.11)
, E = E(J , J ).
2 + 1
It is instructive to restore the -dependence in the formulas (4.10) and (4.11), i.e., to
rewrite
terms of the energy E, the R-charges J, J and the auxiliary charge
them in 2
V = = cos J :

V V 2 + = J V 2 + + 2J V,
(4.12)
V ,

4J
J
E2 = V 2 +
(4.13)
= V 2 + 2 1
,
E = E(J, J ).
V
V2 +
A nice feature of this representation is that if V

then the expression for the energy
takes the form of a perturbative expansion in because V can be found from (4.12) as a
E 2 = 2 +
92
series in (J +2J
) . In particular, we get the following expression for the energy at the first
order in
E 2 (J + 2J )2 +
where V J + 2J
2J
,
J + 2J
E J + 2J +
J
,
(J + 2J )2
(4.14)
J
.
(J +2J )2
Note that here there is no restriction on values of J and J
apart from the requirement that J + 2J

.
One may be tempted to conjecture that the formula (4.14) may be valid at small values
of if J + 2J is very large. However, (4.14) was obtained in the strong coupling
1
regime, and we expect it to receive 1 string sigma model corrections. In particular, even
the coefficient in front of J + 2J may get changed by the corrections, and, if so, the oneloop perturbative correction to the dimension of the corresponding CFT operator dual to
J

the string solution will not be of order (J +2J
)2 but of order J + 2J .
If J
J the energy (4.14) takes the form
J
.
(4.15)
J2
This expression
for E is consistent with the string oscillation spectrum in the sector with
large J
[1,13], i.e., with the plane-wave spectrum (similar comparison was done in
[4]). From the plane-wave spectrum point of view, J represents the angular momentum
carried by string oscillations. Since the linear term in J is not renormalized in the BMN
limit, one may conjecture that the same should happen here.
If we set J = 0 in (4.15) we get
E J + 2J +
E 2 = (2J )2 + ,
i.e.,
E 2J +
.
4J
(4.16)
Thus at large J the correction goes as J1 , instead of a constant shift found in the case of
the single-spin folded string rotating in S 5 [2].
We can also consider the case with V (when V J )
E2 4 J + J 2.
(4.17)
Setting J = 0 we reproduce the usual Regge trajectory relation.
It is interesting to note that the limit J 0 depends on the value of the second angular
momentum J (see also the discussion of this case in Sections 2 and 4.3). When J = 12
the dependence of the energy on the angular momentum changes its form, i.e., the system
undergoes a kind of second order phase transition (the
second derivative of the energy
over the orbital momentum has a discontinuity at J = 12 ). This happens because for
= 0 one has J = 12 sin2 0 , so that the value J = 12 is found when the string reaches its
maximal size (0 = 2 ), i.e., when it rotates on the maximal-size 3-sphere within S 5 .
4.2. Circular string rotating in both AdS5 and S 5
As already discussed in Section 2, it is straightforward to combine the two-spin
solution in AdS5 with the three angular momenta solution in S 5 . For completeness, let
93
us summarize the resulting solution depending on 3 different parameters in terms of the

global coordinates of AdS5 and S 5 used in Section 3 and in this section:
= 0 ,
t = ,
= = ,
= ,
(4.18)
= 0 ,
1 = ,
2 = 3 = w,
= .
(4.19)
The equations of motion and the conformal gauge constraint lead to the following relations
2 = 2 + 1,
w2 = 2 + 1,
2 = 2 + 2 sinh2 0 + 2 sin2 0 ,
(4.20)
and the energy and the 5 conserved charges are given by the same relations as in (2.29),
(2.30)

1 2
2
2
2
E = cosh 0 = 1 + 2 sin 0 ,
(4.21)
2

1
S = sinh2 0 = 2 2 sin2 0 2 2 + 1,
(4.22)
2

1
1
J = sin2 0 w = sin2 0 2 + 1.
J = cos2 0 ,
(4.23)
2
2
One can use these equations to analyse the dependence of the energy on the spins and
SO(6) charges. In particular, when J is very large while the string size is small, one
reproduces the corresponding part of the oscillator plane-wave string spectrum (with spin
S and angular momentum J here carried by the semiclassical string instead of being
produced by string oscillations as in [1]).
4.3. Fluctuations and stability of S 5 solution with J = 0, J = 0
Let us now discuss the stability of the simplest S 5 solution with J = 0 and J = 0.
There are two different cases that should be discussed separately.
Case J 12
The solution with J
Then (see (4.17))
1
2
is found by setting = 0 in (4.7), (3.14) (see also (2.28)).
1
J = 2,
E = 4J . (4.24)
4
As was already mentioned in Section 2 (below (2.28)), this solution is essentially the
embedding of the flat space solution into S 5 . However, the fluctuation spectrum will of
course be different from the flat space case.
The computation of the quadratic fluctuation action is a repetition of the one done in the
AdS5 case in Section 3.2. We choose the static gauge
= 0,
w = 1,
t = ,
2 = 2 sin2 0 2,
and expand the NambuGoto action up to the second order in fluctuations. As in the case
of point-like string orbiting in S 5 [4], the Lagrangian for the AdS5 fluctuations will be
94
represented by the 4 massive field contributions15

1
1
LAdS5 = (a k )2 2 k2 ,
2
2
k = 1, 2, 3, 4.
(4.25)
The additional contribution of S 5 fluctuations is (cf. (3.29))

1
1
1
1
LS 5 = (a 1 )2 (a )2 (a )2 (a )2 20
2
2
2
2
2 2 1 + 2 2 2 2 , 2 2 2 .
Here 1 =
1 1
2
(4.26)
and the fields and are defined as in (3.26)

= 2 cos2 + 3 sin2 ,
= sin 2 (2 3 ).
2 2
(4.27)
This Lagrangian is similar to the one (3.29) for fluctuations around the two-spin solution
in AdS5 . Its , , part can be written in the form (3.30) as follows
2
1
1
1
1
L( , , ) = (0 + )2 (1 )2 + (0 )2 1 + 2
2
2
2
2
2
1
1
2
+ (0 ) 1 2
2
2

1 2 2 1
+ + 2 2 2 2 .
(4.28)
2
2
Note that the sum of squares of masses here vanishes, in agreement with the discussion of
fluctuation Lagrangian in conformal gauge in Appendix C.
The Hamiltonian corresponding to (4.28) does not appear to be positive definite, and
so the stability of the solution is a priori in question. It is shown in Appendix A that the
solution is, in fact, stable if
1
2 2,
2
i.e.,
3
0J .
8
Case J 12
The solution with J 12 is found by setting 0 = 2 . Since cos 0 = 0, the value of
in this case is undetermined (cf. (4.2)), and the conformal gauge constraint gives (cf. (4.7),
(3.14), (4.16))
2 = w2 + 1 2,
1
J = w,
2
E = ,
i.e.,
E=
(2J )2 + 1.
(4.29)
15 Expanding near the = 0 point in (3.2) one needs to introduce the 4 Cartesian-type coordinates, e.g., writing
the AdS5 metric as
ds 2 =
(1 + 1/42 )2 2
dk dk
dt +
.
(1 1/42 )2
(1 1/42 )2
95
The coordinates and 1 are not suitable for studying fluctuations around = 2 (which
is a center of the 2-sphere part d 2 + cos2 d12 of the S 5 metric (4.2)). Introducing
instead the Cartesian-type coordinates X1 , X2 as in (2.22) (which have zero values on
the classical solution)
Z = X1 + iX2 = cos ei1 ,
(4.30)
we find that the quadratic fluctuation action (obtained in the static gauge t = , = )
is then given by the sum of the AdS5 part (4.25) and (we ignore total derivative terms)

1
1
LS 5 = |a Z|2 2 2 |Z|2 + L(, ),
(4.31)
2
2

1
1
L(, ) = (a )2 (a )2 21 2 2 1 2
2
2
1
1
1
1
2
= (0 ) (1 + )2 + (0 )2 (1 )2
2
2
2
2
2
1 2 2 1 2
(4.32)
+ 3 4 ,
2
2
where, as in (3.26), (4.27),

1
= sin 2 ( 2 3 ).
= cos2 2 + sin2 3 ,
(4.33)
2
The Lagrangian (4.32) is simpler than (4.26), describing a collection of 2 coupled fields
with a constant Abelian connection; however, the mass matrix is not O(2) invariant, so
after the rotation there will be a remaining -dependence in the mass matrix. Note that
the sum of mass-squared terms for S 5 fluctuations is equal to 2( 2 2) 2 + 3 2
4 = 4( 2 2) in correspondence with the results in the conformal gauge and with the
cancellation of divergences between the bosonic and fermionic sectors (see Appendices B
and C).
The negative mass term for in (4.32) raises again the question about stability. To
analyze the stability of this solution it is sufficient to consider only the , part (4.32) of
the Lagrangian. Following the procedure explained for the AdS5 case in Appendix A, i.e.,
expanding the fluctuations and in Fourier series in , and then looking for solutions in
the form ein , we find the following frequency spectrum

2

2
2
2
n = n + 2 1 2 2 1 + 2 n2 .
(4.34)
It is clear that the n spectrum is real if

2

2

2
n + 2 2 1 4 2 1 + 2 n2 = n2 n2 4 0.
(4.35)
This condition does not depend on and is not satisfied for the mode with n = 1.16 A
possible interpretation of this mode is that in the frame rotating together with the string
16 The stability is obvious for the ein fluctuation modes with n 2. Indeed, (4.32) can be written as
L(, ) = 12 (0 )2 12 (1 + 2)2 + 12 (0 )2 12 (1 )2 + 2 2 , and thus the corresponding Hamiltonian
is non-negative for the modes with n 2. Let us note also that the problem of analysing the small fluctuation
spectrum of this theory is similar to the one of solving string theory in non-diagonal (metric and 2-form field)
plane-wave background (cf. [14]).
96
where string is at rest, it describes the obvious instability of a circular string wound around
large circle of S 3 [8].17
We conclude that the rotating
solution with J = 0 is not stable for any value of the
angular momentum J 12 .
As was already mentioned in the introduction, to get a stable solution with large J
one needs also to switch on a non-zero (and large) value of the angular
momentum J
[8]. If one could ignore the instability, the solution with J = 0, J 12 would be the
most simple and interesting case for the study of the AdS/CFT correspondence in a novel
sector of states. One could try to compute the 1-loop string sigma model correction to
the classical energy in (4.29) by starting with the sum of the bosonic fluctuation action
(4.32) (assuming one could formally omit the unstable mode absent at large J ) and the
fermionic action derived in Appendix B. This will be discussed (for the general stable case
of J, J = 0) in [8]. To estimate this correction at large values of J one may note that for
large all non-zero masses of the 2d fields are equal to , and thus (see [4]) the 1-loop
correction to the energy should be expected to scale as
1
E1 J ,
1
J
1.
2
(4.36)
That seems to suggest the following interpolation formula for the energy (cf. (4.16))
E = 2h()J + f ()
,
4J
a1
h()
1 = 1 + + ,
J
1,
b1
f ()
1 = 1 + + .
(4.37)
(4.38)
In spite of the absence of 2d supersymmetry in the corresponding quadratic part of GS

action, it may actually happen that the coefficients a1 and b1 vanish, i.e., the first two
terms in the energy are not renormalized at the leading order in 1 expansion. That would
support the conjecture, prompted by the appearance of the first power of in the classical
expression for E, that E = 2J + 4J + is actually true also at weak coupling, i.e., is the
exact result for the first two terms in the anomalous dimension of the corresponding dual
gauge theory operator. We shall discuss this conjecture further in the next section.
5. Towards testing AdS/CFT duality in non-supersymmetric multi-spin sectors

Let us now discuss the gauge-theory operators that should correspond to the string states
represented by the classical solutions found above. The eventual aim is to try to compare
their anomalous dimensions as functions of spins and R-charges to the semiclassical
expressions for the energies found above.
17 Such unstable mode would be absent in the S 5 /Z case.
2
97
5.1. AdS5 rotation case

The semiclassical closed string states found in global coordinates in AdS5 S 5 should
be dual to SYM states on R S 3 . Going through the usual argument of Euclidean
continuation and conformal mapping to R 4 (cf. [15]) they should correspond to local
operators in Euclidean 4d space. One can then rotate back to Minkowski space, but
here we prefer not to do that. The SO(4) isometry of S 3 in AdS5 is then becoming the
Lorentz symmetry of R 4 . Thus the Euclidean gauge theory operators will be classified
by its representations, i.e., will be labelled by the values (S1 , S2 ) of the two SO(4) =
SU(2) SU(2) spins. In addition, they will carry also the three quantum numbers of SO(6)
R-symmetry group.
Let us first recall the form of the gauge theory operators that are expected to be dual
to the single-spin string state in AdS5 [2]. If M (M = 1, . . . , 6) are the adjoint scalars
of N = 4 SYM theory and D is the covariant derivative, a representative operator
with canonical dimension 0 = S + 2 is the standard gauge-invariant minimal twist
operator O{1 ...S } = tr(M D{1 DS } M ), where {1 S } denotes symmetrization
and subtraction of traces. This operator will in general mix with similar operators obtained
by replacing the scalars M by the gauge field strength F or by the fermions so to find
its perturbative anomalous dimension one would need to diagonalise the corresponding
anomalous dimension matrix (see, e.g., [1618] and references there).18 The equivalent
form of the above operator is

S

OS = tr(M D1 DS M )n1 nS = tr M n D M ,
(5.1)
where the multiplication by the product of constant null vector n (n n = 0) factors
implements the symmetrization and subtraction of traces. In Minkowski R 1,3 theory one
may choose n = (1, 0, 0, 1), getting OS = tr(M (D+ )S M ), D+ = D0 + D3 . In the
Euclidean version which we use here one is to choose n to be complex, e.g., n =
(1, i, 0, 0), getting

OS = tr M (DX )S M , DX D1 + iD2 .
(5.2)
It is now clear how to generalize this discussion to the case of operators carrying two spins
of SO(4): a representative operator will be

OS1 ,S2 = tr M (DX )S1 (DY )S2 M ,
(5.3)
DX D1 + iD2 ,
DY D3 + iD4 ,
0 = S1 + S2 + 2.
(5.4)
Since the covariant derivatives DX and DY do not commute, this operator will be mixing,
in particular, with various other operators containing permutations of S1 derivatives DX
and S2 derivatives DY and having the same canonical dimension, e.g.,

tr M (DX )k1 (DY )m2 (DX )kl (DY )ml M ,

(5.5)
ki = S1 ,
mi = S2 .
i
18 Two-loop anomalous dimensions for some higher spin currents were found in [19].
98
The eigenvector of anomalous dimension matrix is expected to be a particular combination

of such operators (in addition to others involving gauge field strength and fermions).
An irreducible representation of the rotation group is represented by a particular Young
tableau with two rows with S1 and S2 as numbers of boxes, i.e., should contain additional
antisymmetrizations of DX and DY factors in (5.3).19
The equivalent covariant form of (5.3) is found by introducing two independent null
vectors n and m and generalising (5.1) as follows
OS1 ,S2 = tr(M D1 DS1 D1 DS2 M )n1 nS1 m1 mS2 ,
n n = 0,
m m = 0.
(5.6)
This operator can be readily extended to the Minkowski version of the theory by choosing,
e.g., n = (1, 0, 0, 1) and m = (0, 1, i, 0) with = (1, 1, 1, 1).
As is well known [17], for S1 = S
1, S2 = 0 (or vice versa) the perturbative
anomalous dimension of such operators scales as ln S, which is the same as the scaling
of the energy of the single-spin rotating string solution in AdS5 ; this strongly supports the
existence of an interpolation formula = S + f () ln S between the weak coupling and
the strong coupling regions [2,4].
To compare with the string solution found in the present paper where S1 = S2 = S
1
one needs to know the perturbative (one-loop) anomalous dimension of the operators like

OS,S = tr M (DX )S (DY )S M + ,
(5.7)
where dots stand for appropriate permutations of DX and DY . We are not aware of
computations of anomalous dimension of such operators in the literature, and here can only
speculate about possible interpolation formula for (S) in this case (see also Section 3.2).
Assuming one can trust the expression (3.21) for the energy for large S = S (in spite of
instability of the solution for large S) one would expect to find the order S 1/3 correction in
the anomalous dimension replacing the familiar ln S correction for the single-spin operators
(5.2). However, it seems hard to imagine how such fractional-power term could appear in
the 1-loop SYM computation for the anomalous dimension of (5.7). We suspect that the
interpolation formula (3.10), (3.35) may be a more plausible alternative, implying that at
large S but small (with S 1) one should expect to find the anomalous dimension of
(5.7) going as
= 2k()S + ,
k() = 1 + a1 + a2 2 + .
(5.8)
It would be very interesting to check this by direct perturbative computations on the gauge
theory side.
5.2. S 5 rotation case
The construction of operators that carry several SO(6) spins and thus should be dual to
the string states represented by the solutions in Section 4 describing closed strings rotating
19 Due to these antisymmetrizations the resulting operators may not be superconformal primary operators.
99
in S 5 is somewhat similar. Let us introduce the notation for the three complex scalars of
N = 4 SYM theory (cf. (2.21))
Z = 1 + i2 ,
X = 3 + i4 ,
Y = 5 + i6 .
(5.9)
Among the operators with the minimal canonical dimension for given (J1 , J2 , J3 ) values
of SO(6) charges there are operators with holomorphic dependence of the three complex
scalars:

OJ1 ,J2 ,J3 = tr (Z )J1 (X )J2 (Y )J3 + ,
(5.10)
0 = J1 + J2 + J3 ,
where dots stand for permutations of X , Y , Z factors needed to form an irreducible
representation of SO(6) that is expected to be an eigenvector of the anomalous dimension
matrix. The 1-loop anomalous dimension matrix for generic scalar operators of the form
OM1 Mj = tr(M1 Mj )
(5.11)
was computed in [9,10]. Taking its symmetric traceless part (i.e., multiplying (2.19)
by a null vector, e.g., nM = (1, i, 0, 0, 0, 0)) one finds a chiral primary operator whose
dimension is protected. This case is equivalent to (5.10) with J1 = J and J2 = J3 = 0, i.e.,
OJ,0,0 = tr(Z )J , which is dual to the ground state of the string theory expanded near the
point-like string orbiting in S 5 [1].20
The operator that should be dual to the string solution with J1 = J , J2 = J3 = J found
above should then be

0 = J + 2J .
OJ,J ,J = tr (Z )J (X )J (Y )J + ,
(5.12)
This operator belongs to the irreducible representation of SU(4) with Young tableau labels
(J, J , J ) or with Dynkin labels [0, J J , 2J ] (see Appendix D)21 if J J , and
to the representation (J , J , J ) = [J J, 0, J + J ] if J J , and does not seem to
be a superconformal primary operator. For example, it is known that the operator with
J = 0, J = 2 is a superconformal descendant of the Konishi operator K = tr(M M ).
In the near-BPS limit J
J the operators of the form (5.12) are examples of the BMN
operators [1] with a small number of impurities, and one can, in principle, make detailed
comparison between semiclassical predictions (4.14) and the perturbative results of [1] and
[9,10]. If J is comparable to or much larger than J , we are very far from the BPS operator
tr(Z )J , and the conformal dimensions of the operators cannot be computed from the
plane-wave string spectrum.
The semiclassical results obtained in Section 4 are the only source of non-perturbative
predictions for the dimensions of these operators. Let us stress again that among a large
number of operators in these representations only the one with the lowest conformal
dimension should be dual to the string solution we found. It is also interesting to point
20 We are considering only single-trace operators as seems appropriate for the elementary string stategauge
theory operator correspondence in the large N limit. Note that there exist also multi-trace operators which may
carry the same quantum numbers and may be 1/4 or 1/8 BPS (see, e.g., [20] and references there).
21 We are grateful to M. Staudacher and N. Beisert for correcting a mistake in this identification in the original
version of this paper.
100
out that one and the same formula (4.14) should be giving the conformal dimensions of the
operators from the two different ([0, J J , 2J ] or [J J, 0, J + J ]) representations.
This should be true not only in the large limit but also in the weak-coupling perturbation
theory.
As discussed in Section 4.3, the simplest novel case for a non-trivial check of the
AdS/CFT duality in a non-supersymmetric sector
is when the circular string orbiting in
S 5 has maximal size, i.e., has J = 0, J 12 . The exact expression for its classical

energy given in (2.27), (4.16) is E = (2J )2 + . Provided the instability of this solution
could be ignored (e.g., by embedding it into a class of stable solutions with J = 0) for

large J = J the expression for the energy E(J ) gives the following prediction for the
strong-coupling behaviour of the anomalous dimension of the corresponding operator

O0,J ,J = tr (X )J (Y )J + ,
(5.13)
= 2J + + , J

1.
(5.14)
4J
As was already discussed in Section 4.3, our conjecture is that this expression is actually
valid also at weak coupling, i.e., the first two terms in are not renormalized.
One can check this against the results of [21] applied to the operators transforming in
the [J , 0, J ] representation for J = 4 and J = 5. At J = 4 one finds that the lowest
anomalous dimension of the operators in [4, 0, 4] is = 0.0411 while Eq. (5.14) predicts
the anomalous dimension to be = 0.0625, i.e., the deviation is about 34%.22 However,
at J = 5 one gets the lowest anomalous dimension to be g = 0.042 while Eq. (5.14)
predicts = 0.05, i.e., the deviation in this case is just 16%. It would be useful to
compute one-loop anomalous dimensions of the [J , 0, J ] operators for J = 6 to see if
the agreement with Eq. (5.14) is really getting better at large J .
Since it is sufficient to consider the operators with holomorphic dependence on the
fields X and Y , the Hamiltonian of the integrable SO(6) spin chain considered in [10]
reduces to the Hamiltonian of the simplest XXX1/2 spin model [22]. It would be very
interesting to find the corresponding one-loop anomalous dimension and the explicit form
of the associated operator (5.13) in the large J limit by utilizing this connection to the
XXX1/2 model.
6. Concluding remarks
There are various generalizations of the AdS5 and S 5 solutions we have found. For
example, the special S 5 solution considered in Section 4.3 with maximal size of the string
(0 = 2 ) is also a solution of string theory on Rt S 3 :
ds 2 = dt 2 + d 2 + cos2 d22 + sin2 d32 ,
22 Diagonalizing the matrix of anomalous dimensions from [9,10] one finds also an operator with much closer
value 0.0619, but comparison with the results of [9] shows that it belongs to [2, 4, 2] representation. We are
grateful to G. Arutyunov and J. Minahan for explanations related to this point.
t = ,
= ,
2 = 3 = w,
2 = w2 + 1.
101
(6.1)
It is plausible, therefore, that it can be embedded into various other spaces containing
S 3 factors, in particular into AdS5 T 1,1 space related via AdS/CFT to an N = 2
superconformal theory [23].
Similarly, analogs of AdS5 two-spin solution of Section 3 exist for a more general class
of 5d (or higher-dimensional) metrics with SO(4) isometry, e.g., ds 2 = g() dt 2 + d 2 +
h() d3 . As in the AdS5 case, the two equal rotations in S 3 allow one to stabilize a
circular string at a fixed value of = 0 (stability under small fluctuations will depend on
the explicit form of g() and h() and on the value of the spin). In particular, such solution
will exist for an AdS black hole metric.23
There are several directions in which the present work needs to be completed or
extended. In [8] we shall study the S 5 solution with J = 0, J = 0 in detail, identifying the
range of its stability, and computing the 1-loop superstring sigma model correction to the
classical energy. One should then be able to compare the string results (e.g., for J J )
to the perturbative results for anomalous dimensions of the corresponding gauge theory
operators. It would be very interesting also to compute the leading-order perturbative
contributions to the anomalous dimensions of the operators discussed in Section 5. This
applies, in particular, to the operator (5.13) whose dimension should be possible to find
using the integrable-model connection suggested in [10].
Acknowledgements
We are grateful to G. Arutyunov, N. Beisert, R. Metsaev, J. Minahan, M. Staudacher and
K. Zarembo for useful discussions and comments. This work 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.
Appendix A. Stability analysis

Here we analyze the stability of the AdS5 and S 5 two-spin solutions discussed above
under small perturbations.
A.1. Stability of the two-spin AdS5 solution
The equations of motion for fluctuations that follow from (3.29) are

2 2 + 2 2 2 1 + 2 = 0,

2 2 + 2 + 2 2 1 + 2 = 0,

+ 4 1 + 2 + 2 2 + 2 = 0.
23 Single-spin solutions in this and similar non-conformal cases were discussed in [3,24].
(A.1)
(A.2)
(A.3)
102
To solve the system (A.1)(A.3) of linear differential equations we expand the fluctuations
in Fourier series in

=
(A.4)
n ( )ein ,
=
n ( )ein ,
=
n ( )ein .
n
Then we get the following system of equations for the nth modes of the fluctuations

n + n2 n 2 2 + 2 n 2n 2 1 + 2 = 0,
(A.5)

n + n2 n + 2n 2 1 + 2 2inn 2 + 2 = 0,
(A.6)

n + n2 n + 4 1 + 2 n + 2inn 2 + 2 = 0.
(A.7)
We start the analysis of the system (A.5)(A.7) by looking for solutions of the form
n ( ) = Cn ein ,
n ( ) = Cn ein ,
n ( ) = Cn ein .
(A.8)
The stability of the two-spin string solution requires the frequencies n to be real.
Substituting (A.8) into (A.5)(A.7) we get

2
n 2 2 + 2 n2 Cn 2in 2 1 + 2 Cn = 0,
(A.9)

2
n n2 Cn + 2in 2 1 + 2 Cn 2in 2 + 2 Cn = 0,
(A.10)

2

n + 4 1 + 2 n2 Cn + 2in 2 + 2 Cn = 0.
(A.11)
This is a system of linear equations for the coefficients Cn , Cn , Cn which can be written
in the form Aij C j = 0; it has non-trivial solutions only for such values of n for which the
determinant of the matrix Aij vanishes. This gives the equation for n :
fn (z) = 0,
z n2 ,

fn (z) z3 8 + 10 2 + 3n2 z2 + 16 + 40 2 + 24 4 + 8 2n2 + 3n4 z

n2 n2 4 n2 4 2 2 .
(A.12)
We need to find the values of such that all roots of this cubic equation are positive. It is
not difficult to show that the two extrema z and z+ (z < z+ ) of fn (z) are positive, and
fn (z ) > 0 and fn (z+ ) < 0. Therefore, all roots are positive if

fn (0) = n2 n2 4 n2 4 2 2 0.
(A.13)
We see that f0 (0) = f2 (0) = 0, f1 (0) < 0 and f3 (0) = 45(5 2 2). From the expression
for f3 (0) we conclude that the solution is stable only if
5
2 .
(A.14)
2
We also see that for n = 0 and n = 2 the determinant of the matrix Aij in (A.12) vanishes
at n = 0. That means that for these modes there is a solution of the form
n + Cn ,
n ( ) = C
n + Cn ,
n ( ) = C
n + Cn .
n ( ) = C
(A.15)
103
Substituting (A.15) into the equations of motion (A.5), (A.6) and (A.7), one can easily find
the corresponding solutions for n = 0 and n = 2
0 ( ) = C0 ,
2 ( ) = C2 ,
2 + 2
0 ( ) =
C0 + C0 ,
2
2(1 + )
0 ( ) = 0,
2 ( ) =
C0 + C2 ,
2(1 + 2 )
i
2 ( ) =
2 ( ).
2 + 2
(A.17)
(A.16)
Since the fluctuation corresponding to (A.16), (A.17) does not depend on , such
solutions do not lead to instability of the two-spin string solution. In fact, both solutions
(A.16) and (A.17) have simple interpretation. The n = 0 solution reflects the fact that the
radius 0 (or ) of the two-spin string solution (3.12)(3.16) is a free parameter. This
parameter can be changed, and this leads to the existence of the zero mode in the spectrum
of fluctuations. The n = 2 solution appears because the two-spin string solution was found
only for equal frequencies. We expect that the general solution will depend on the two
independent frequencies, and, therefore, on the two parameters; the existence of the zero
mode fluctuation with n = 2 is related to this (cf. (3.26)).
A.2. Stability of the two-spin S 5 solution
The analysis of stability in the S 5 case follows closely the procedure explained above
in the AdS5 case. Here we shall consider explicitly only the case of the solution with J = 0
and J 12 discussed in Section 4.3. The equations of motion for fluctuations that follow
from (4.26) are
22 + 2 = 0,
2 2 2 = 0,
+ 4 + 2 2 = 0.
(A.18)
(A.19)
(A.20)
Expanding the fluctuations , and in Fourier series in , and then looking for solutions
in the form ein , we find the following equation for the frequency spectrum (2 2 2 )
fn (z) = 0,
z n2 ,

fn (z) z3 4 + 22 + 3n2 z2 + 82 + 3n4 z n2 n2 4 n2 22 . (A.21)
Just as in the AdS5 case, the two extrema z < z+ of fn (z) are positive, with fn (z ) > 0
and fn (z+ ) < 0. Therefore, all roots are positive if

fn (0) = n2 n2 4 n2 22 0.
(A.22)
Since 2 2, the solution is stable only if f1 (0) = 3(1 22 ) 0, i.e., if
1
2 2.
2
104
Appendix B. Quadratic fermionic part of the superstring action

The quadratic part of the AdS5 S 5 GreenSchwarz superstring action [25] expanded
near a particular bosonic string solution can be found as described, e.g., in [12,26] and
used in a similar context in [4]. Assuming the induced metric is flat, the relevant part of the
fermionic action is

A
a XM ,
LF = i ab I J M ab s I J I Na Db J , Na A eaA , eaA = EM
(B.1)
where I, J = 1, 2, s I J = diag(1, 1), and Na are projections of the 10d Dirac matrices.
Here XM are the string coordinates (given functions of and for a particular classical
solution) corresponding to the AdS5 (M = 0, 1, 2, 3, 4) and S 5 (M = 5, 6, 7, 8, 9) factors.
The covariant derivative Da can be put into the following form

i IJ
I
IJ
Da = Da M Na J , i01234, 2 = 1,
(B.2)
2
where
1
AB
Da = a + aAB AB , aAB a XM M
,
4
and the mass term originates from the RR 5-form coupling [25].
(B.3)
B.1. AdS5 case

In the case of the AdS5 two-spin solution (4.18) the 2d projections of -matrices that
enter the fermionic action are (the indices A = 0, 1, 2, 3, 4 here will label the t, , , ,
directions in the tangent space):
N0 = cosh 0 0 + sinh 0 4 ,
3 cos 3 sin 4 ,
N1 = sinh 0 2 ,
N(a Nb) = sinh2 0 ab ,

(B.4)
(B.5)
4 cos 4 + sin 3 .
AB has the following components

The projected Lorentz connection aAB = a XM M
001 = sinh 0 ,
031 = cosh 0 sin ,
041 = cosh 0 cos ,
032 = cos ,
042 = sin ,
121 = cosh 0 .
(B.6)
Then
1
1
D0 = 0 + ( sinh 0 0 + cosh 0 4 )1 + 2 3 ,
2
2
1
D1 = 1 cosh 0 1 2 .
2
After the -dependent rotation of I in the 34-plane,

1
I
I
1 I
=S ,
S = exp 3 4 ,
2
(B.7)
105
we find 3,4 3,4 in (B.4) and (B.7), at the expense of getting an additional constant
3 4 term in D1 . This eliminates the -dependence from the fermionic action. Note that
is invariant under this rotation.
To interpret the resulting fermionic action as a collection of massive 2d fermions
with standard kinetic terms it is useful to make as in [4] a further rotation (Lorentz
boost) in the 04-plane, with S = exp( 12 0 4 ), where cosh = cosh 0 . Then we get
Na = sinh 0 a , a = (0 , 2 ), (a b) = ab . There is the corresponding change in Da while
remains invariant. Fixing the kappa-symmetry gauge by 1 = 2 and rescaling the
fermions by sinh 0 we can interpret the resulting action
a

Da + i
M ,
LF = 2i
(B.8)
1
1
M = sinh 0 M ab a b = imF 134, mF = sinh 0 =
(B.9)
2
2
as describing a collection of 2d massive Majorana fermions on a flat 2d background
coupled to a constant non-Abelian 2d gauge field (represented by constant Lorentz
connection aAB terms). Indeed, in the representation for A where 0 and 2 are 2d Dirac
matrices times a unit 8 8 matrix we get as in [4] 4 + 4 species of 2d Majorana fermions
with masses mF .
We will not go into a detailed analysis of this action here and just mention that, as
expected on the general grounds of conformal invariance of the AdS5 S 5 string action
[25], the fermionic contribution to the divergent part of the 1-loop effective action (which
2
is proportional to the sum of mass-squared terms, i.e., 8 2 ) cancels the logarithmic
divergence coming from the bosonic fluctuation action (3.30) (connection terms in both
the bosonic and fermionic actions do not contribute to logarithmic divergences in 2
dimensions).
B.2. S 5 case
In the case of the S 5 solution (4.6) we shall label the tangent space coordinates by
A = 0, 5, 6, 7, 8, 9 corresponding to the t direction of AdS5 and , 1, , 2 , 3 directions
of S 5 . Then
N0 = 0 + cos 0 6 + w sin 0 8 ,
8 cos 8 + sin 9 ,
N1 = sin 0 7 ,
N(a Nb) = sin2 0 ab ,
9 cos 9 sin 8 .
(B.10)
(B.11)
The projected Lorentz connection has the following non-zero components

065 = sin 0 ,
085 = w cos 0 cos ,
095 = w cos 0 sin ,
087 = w sin ,
097 = w cos ,
175 = cos 0 .
(B.12)
As above, we first do local Lorentz rotation in the 89-plane to eliminate the -dependence;
as a result, 8,9 8,9 . Then (for generic and 0 ) we need to do two rotations
in the 68 and 06 planesto put N0 into the form N0 = sin 0 0 . After the rotation
in the 68-plane (under which in (B.2) is invariant) we get N0 = 0 + a6 , a 2 =
106
2 cos2 0 + w2 sin2 0 = 2 + sin2 0 . Under the boost in the 06-plane S = exp( 12 0 6 ),

where cosh = sin0 , the expression for becomes
= S 1 S = i(cosh 0 sinh 6 )1234.
(B.13)
Then fixing the kappa-symmetry gauge by 1 = 2 and rescaling the fermions by sin 0
we finish with the same action as in (B.8) with a = (0 , 7 ) and
1
sin 0 M ab a b = imF 0712346,
2
1
mF = sin 0 sinh = 2 + 2 .
2
M=
(B.14)
The contribution to the divergences is then proportional to 8 12 ( 2 + 2 ) = 4 2 + 4 2

which is indeed the same as coming from the bosonic sector (see also Appendix C).
The same result is found also in the special cases discussed in Section 4.3. When = 0
(see (4.24)) we have N0 = 0 + sin 0 8 , N1 = sin 0 7 , sin 0 = . Here we need a
2
boost in 08-plane with parameter cosh = 2 to get Na = sin 0 a . Then mF is the same
as in (B.14). When 0 = 2 , w2 = 2 1 (see (4.29))24 we get N0 = 0 + w 8 , N1 = 7
and thus the required 08-boost parameter has cosh = . That gives
0 = : M = imF 07 12348, mF = sinh = 2 1.

(B.15)
2
Note that the fermionic and bosonic masses are different, reflecting the absence of the 2d
supersymmetry. The fermionic contribution to the logarithmic divergence is proportional
to 8 ( 2 1) = 4 2 + 4( 2 2) which indeed cancels the contribution from the bosonic
fluctuations: 4 massive AdS5 fields (4.25) and 4 massive S 5 fields in (4.31).
Appendix C. Bosonic fluctuation action in conformal gauge
In discussing fluctuation actions in the main part of the paper we used the static gauge.
Let us note for completeness that similar conclusions can be reached also if one uses the
conformal gauge (see, e.g., [4,12]). Let us recall that the general form of the quadratic
fluctuation action for a sigma model in the conformal gauge written in terms of tangentA (X) M ) is
space fluctuations (XM XM + M , A = EM

1
IB(2) =
(C.1)
d 2 gg ab Da A Db B AB + MAB zA zB ,
2
A
MAB = g g ab eaC ebD RACBD ,
eaA a XM EM
(X),
gab = eaA ebB AB ,
(C.2)
where Da A = a A + aAB (X) B with the same projected Lorentz connection as in (B.3).
For the AdS5 part the curvature is RACBD = AB CD + AD CB while for the S 5 part
it has the opposite sign, RACBD = AB CD AD CB . If the induced metric is flat (as in
24 Note that for cos = 0 the connection (B.12) simplifies substantially.
0
107
all examples discussed in the present paper), the divergent part of the 1-loop action is determined simply by the trace of MAB and should be the same as found in the static gauge,
cf. [4].
Let list the expression for the mass matrix in (C.2) for the solutions discussed above
(the expressions for the corresponding connections can be found in Appendix B). In the
AdS5 case (4.18) one gets:
e00 = cosh 0 ,
e12 = sinh 0 ,
e02 = sinh 0 cos ,
e03 = sinh 0 sin ,
gab = sinh2 0 ab ,
2
,
so that AB M AB = MAdS
5

2
MAdS
= 4ab eaC ebD CD = 4 2 cosh2 0 + sinh2 0 2 sinh2 0 = 4 2.
5
(C.3)
This gives the same contribution to divergences as coming from the fermions (cf. (B.9)).
For the S 5 solution (4.6) one has both the AdS5 and S 5 fluctuations, and
e00 = ,
e06 = cos 0 ,
e17 = sin 0 ,
e08 = w sin 0 cos ,
e09 = w sin 0 sin ,
gab = sin2 0 ab ,
so that here
2
AB M AB = MAdS
+ MS25 ,
5
2
MAdS
= 4ab eaC ebD CD = 4 2 ,
5

= 4ab eaC ebD CD = 4 sin2 0 2 cos2 0 w2 sin2 0 = 4 2 .
(C.4)
MS25
(C.5)
This is in agreement with the static gauge result (4.26) for = 0 and is the same as the
divergent contribution coming from the fermionic sector with mass matrix (B.14).
In the special case 0 = 2 , w2 = 2 1 we get instead

MS25 = 4 2 2 ,
(C.6)
which is also in agreement with the static gauge result (4.31) and again cancels the
divergences coming from the fermionic sector (4.31).
Appendix D. Dynkin labels and Young tableau labels of SU(4) irreps
In this appendix we recall (see, e.g., [27]) how the Dynkin labels of representations
of the algebra so(6) or, equivalently, of su(4), are expressed in terms of the Young labels
(numbers of boxes in raws of a Young tableau). Recall that the generators J1 = J12 , J2 =
J34 , J3 = J56 form a basis of Cartan generators of so(6). The simple roots can be chosen as
so(6)
= e1 e2 = {1, 1, 0},
3so(6)
= e2 + e3 = {0, 1, 1}.
so(6)
= e2 e3 = {0, 1, 1},
(D.1)
Since 2so(6) 1so(6) = 1so(6) 3so(6) = 1 and (aso(6) )2 = 2 the root system is equivalent
to the su(4) root system with the following identification of the simple roots
1su(4) = 2so(6) ,
2su(4) = 1so(6),
i.e., the two algebras are isomorphic.
3su(4) = 3so(6) ,
(D.2)
108
An irreducible representation of su(4) can be labelled by the eigenvalues of the Cartan

generators Ji on the highest weight vector:
Ji |j1 , j2 , j3 = ji |j1 , j2 , j3 ,
j1 j2 j3 0,
(D.3)
where ji is the number of boxes in the ith row of the Young tableau associated with the
representation which can then be denoted as (j1 , j2 , j3 ). The same representation can be
also labeled by the Dynkin labels da that are related to the Young labels ji as follows:
j=
3

da a ,
j {j1 , j2 , j3 },
(D.4)
a=1
where a are the fundamental weights defined by

2a b
= a b = ab .
b2
(D.5)
By using (D.1) and (D.2), we can easily solve (D.5) to get

1 1 1
1 1 1
1 =
2 = {1, 0, 0},
, , ,
, , .
3 =
2 2 2
2 2 2
(D.6)
Then from (D.4) and (D.6) we find

1
1
1
j1 = (d1 + 2d2 + d3 ),
(D.7)
j2 = (d1 + d3 ),
j3 = (d1 + d3 ).
2
2
2
Solving the system, we get the relation between the Dynkin labels and the Young labels
d 1 = j2 j3 ,
d 2 = j1 j2 ,
d 3 = j2 + j3 .
(D.8)
The Dynkin labels have to be non-negative. The representation associated with the Dynkin
labels di is denoted as [d1 , d2 , d3 ].
Coming back to the string solution (2.23), we see that if J J then the representation is
(J, J , J ) or [0, J J , 2J ]. If J J the representation is (J , J , J ) or [J J, 0, J +
J ]. Note that in the last case we also have to rearrange the Cartan generators in such an
order that j1 j2 j3 0.
References
[1] D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-YangMills,
JHEP 0204 (2002) 013, hep-th/0202021.
[2] 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.
[3] H.J. de Vega, I.L. Egusquiza, Planetoid string solutions in 3 + 1 axisymmetric spacetimes, Phys. Rev. D 54
(1996) 7513, hep-th/9607056.
[4] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S 5 , JHEP 0206 (2002)
007, hep-th/0204226.
[5] A.A. Tseytlin, On semiclassical approximation and spinning string vertex operators in AdS5 S 5 , hepth/0304139.
109
[6] J.G. Russo, Anomalous dimensions in gauge theories from rotating strings in AdS5 S 5 , JHEP 0206 (2002)
038, hep-th/0205244.
[7] J.M. Maldacena, H. Ooguri, Strings in AdS(3) and SL(2,R) WZW model. I, J. Math. Phys. 42 (2001) 2929,
hep-th/0001053;
J.M. Maldacena, J. Michelson, A. Strominger, Anti-de Sitter fragmentation, JHEP 9902 (1999) 011, hepth/9812073;
N. Seiberg, E. Witten, The D1/D5 system and singular CFT, JHEP 9904 (1999) 017, hep-th/9903224.
[8] S. Frolov, A.A. Tseytlin, in press.
[9] 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, BMN operators and superconformal symmetry, hep-th/0211032;
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.
[10] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super-YangMills, JHEP 0303 (2003) 013, hepth/0212208.
[11] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and
gravity, Phys. Rep. 323 (2000) 183, hep-th/9905111.
[12] N. Drukker, D.J. Gross, A.A. Tseytlin, GreenSchwarz string in AdS5 S 5 : semiclassical partition function,
JHEP 0004 (2000) 021, hep-th/0001204.
[13] R.R. Metsaev, Type IIB GreenSchwarz superstring in plane wave RamondRamond background, Nucl.
Phys. B 625 (2002) 70, hep-th/0112044;
R.R. Metsaev, A.A. Tseytlin, Exactly solvable model of superstring in plane wave RamondRamond
background, Phys. Rev. D 65 (2002) 126004, hep-th/0202109.
[14] M. Blau, M. OLoughlin, G. Papadopoulos, A.A. Tseytlin, Solvable models of strings in homogeneous plane
wave backgrounds, hep-th/0304198.
[15] G.T. Horowitz, H. Ooguri, Spectrum of large N gauge theory from supergravity, Phys. Rev. Lett. 80 (1998)
4116, hep-th/9802116;
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150;
T. Banks, M.R. Douglas, G.T. Horowitz, E.J. Martinec, AdS dynamics from conformal field theory, hepth/9808016.
[16] F.A. Dolan, H. Osborn, Superconformal symmetry, correlation functions and the operator product expansion,
Nucl. Phys. B 629 (2002) 3, hep-th/0112251.
[17] 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.
[18] A.V. Belitsky, A.S. Gorsky, G.P. Korchemsky, Gauge/string duality for QCD conformal operators, hepth/0304028.
[19] 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.
[20] E. DHoker, D.Z. Freedman, Supersymmetric gauge theories and the AdS/CFT correspondence, hepth/0201253.
[21] N. Beisert, C. Kristjansen, M. Staudacher, The dilatation operator of N = 4 super-YangMills theory, hepth/0303060.
[22] L.D. Faddeev, How algebraic Bethe ansatz works for integrable model, hep-th/9605187.
[23] I.R. Klebanov, E. Witten, Superconformal field theory on threebranes at a CalabiYau singularity, Nucl.
Phys. B 536 (1998) 199, hep-th/9807080.
[24] A. Armoni, J.L. Barbon, A.C. Petkou, Orbiting strings in AdS black holes and N = 4 SYM at finite
temperature, JHEP 0206 (2002) 058, hep-th/0205280;
A. Armoni, J.L. Barbon, A.C. Petkou, Rotating strings in confining AdS/CFT backgrounds, JHEP 0210
(2002) 069, hep-th/0209224.
[25] R.R. Metsaev, A.A. Tseytlin, Type IIB superstring action in AdS5 S 5 background, Nucl. Phys. B 533
(1998) 109, hep-th/9805028.
[26] R. Kallosh, A.A. Tseytlin, Simplifying superstring action on AdS5 S 5 , JHEP 9810 (1998) 016, hepth/9808088;
110
S. Forste, D. Ghoshal, S. Theisen, Stringy corrections to the Wilson loop in N = 4 super-YangMills theory,
JHEP 9908 (1999) 013, hep-th/9903042.
[27] H.F. Jones, Groups, Representations and Physics, IOP, Bristol, 1998.

Glueball Regge trajectories in (2 + 1)-dimensional

gauge theories
Harvey B. Meyer, Michael J. Teper
Theoretical Physics, University of Oxford, 1 Keble Road, OX1 3NP Oxford, United Kingdom
Abstract
We compute glueball masses for even spins ranging from 0 to 6, in the D = 2 + 1 SU(2) lattice
gauge theory. We do so over a wide range of lattice spacings, and this allows a well-controlled
extrapolation to the continuum limit. When the resulting spectrum is presented in the form of a
ChewFrautschi plot we find that we can draw a straight Regge trajectory going through the lightest
glueballs of spin 0, 2, 4 and 6. The slope of this trajectory is small and turns out to lie between the
predictions of the adjoint-string and flux-tube glueball models. The intercept we find, 0 1, is
much lower than is needed for this leading trajectory to play a Pomeron-like role of the kind it
is often believed to play in D = 3 + 1. We elaborate the Regge theory of high-energy scattering in
2 space dimensions, and we conclude, from the observed low intercept, that high-energy glueball
scattering is not dominated by the leading Regge pole exchange, but rather by a more complex
singularity structure in the region 0 Re 1/2 of the complex angular momentum plane. We
show that these conclusions do not change if we go to larger groups, SU(N > 2), and indeed to
SU(), and we contrast all this with our very preliminary calculations in the D = 3 + 1 SU(3)
gauge theory.
PACS: 11.55.Jy; 12.39.Mk
1. Introduction
On a ChewFrautschi plot (J versus m2 ) the experimentally observed mesons and
baryons appear to lie on (nearly) linear and parallel Regge trajectories, J = 0 + m2 ,
with the exchange of the corresponding Regge poles (normally) dominating any highE-mail addresses: meyer@thphys.ox.ac.uk (H.B. Meyer), teper@thphys.ox.ac.uk (M.J. Teper).
doi:10.1016/j.nuclphysb.2003.07.003
112
H.B. Meyer, M.J. Teper / Nuclear Physics B 668 (2003) 111137
energy scattering that involves the exchange of non-trivial quantum numbers [1]. The
total cross-section, on the other hand, is related by unitarity to forward elastic scattering
and this is dominated by the Pomeron which carries vacuum quantum numbers [13].
The Pomeron trajectory is qualitatively different from other Regge trajectories in that it
appears to be much flatter ( much smaller) and it is not clear what physical particles
correspond to integer values of J . There are long-standing speculations that these might be
glueballs. Thus if one were to calculate the mass spectrum of the SU(3) gauge theory one
could investigate whether the glueball masses fall on linear trajectories and whether these
trajectories have the properties of the Pomeron.
Of course one does not expect the leading glueball trajectory to be exactly like the
Pomeron since in the real world there will be mixing between glueballs and flavour-singlet
q q mesons; just as the usual meson/baryon trajectories cannot be exactly linear because
the higher-J mesons are unstable. It is only in the limit of SU(N ) that one can
expect exactly linear trajectories (no decays) and the leading glueball Regge trajectory to
be the Pomeron (no mixing). However there are good heuristic arguments to believe that
QCDN=3 is close to QCDN= [4], and lattice calculations support this idea to the extent
that they have shown that for pure gauge theories even N = 2 is close to N = [5].
Although it is now easy to calculate the lightest masses of a pure gauge theory using
lattice Monte Carlo simulations, calculating masses of higher-J states is more subtle
because a cubic lattice respects only a small subgroup of the full rotation group and
each representation of this subgroup contains states that correspond to different J in the
continuum limit. This difficulty is compounded by the fact that the higher-J states, being
heavier, are more difficult to calculate accurately. In a recent paper [6] we have developed
techniques for identifying states of arbitrary J on a lattice and we are now in the process
of applying these techniques to determine if glueballs fall on linear Regge trajectories and
if so whether the leading trajectory has the characteristics of the Pomeron.
In this paper we address this question in the context of the D = 2 + 1 SU(2) gauge
theory. At first glance this may seem far removed from the case that is of immediate
physical interest, SU(3) in D = 3 + 1, but the fact that the computations are much faster
in D = 2 + 1 than in D = 3 + 1 means that we can expect to obtain more accurate results,
more quickly, and so test the efficiency of our approach. Moreover, at closer inspection, one
finds that D = 2 + 1 non-Abelian gauge theories resemble those in D = 3 + 1 in a number
of relevant respects. They become free at short distances, the coupling sets the dynamical
length scale, and the (dimensionless) coupling becomes strong at large distances. They
are linearly confining, and the confining flux tube appears to behave like a simple bosonic
string at large distances [7]. The light glueball spectrum more-or-less fits the expectations
of a flux loop model, just as it does in (3 + 1) dimensions [810]. Moreover, as we shall
see below, the same simple models that serve to predict linear glueball Regge trajectories
in D = 3 + 1, predict their existence in the case of D = 2 + 1. We also note that the link
to string theories (at least at large N ) can be made in D = 2 + 1 just as in D = 3 + 1 [13].
For all these reasons we believe that our exercise is of significant theoretical interest.
At a more heuristic level, one is motivated to search for a Pomeron trajectory where one
has high-energy cross-sections that are roughly constant in energy. Although the scattering
of glueballs has not been observed experimentally, ones intuition is that they will behave as
black discs, just like the usual mesons and hadrons, and so it makes sense to speculate that
113
the Pomeron might be the leading (glueball) Regge trajectory in the D = 3 +1 SU(3) gauge
theory. We do not expect this to depend strongly on the number of colours, so it should be
a property of all SU(N) gauge theories. Finally, since we can think of no obvious reason
why going from 3 to 2 spatial dimensions should prevent colliding glueballs from having
roughly constant cross-sections at high energiesalthough as black segments rather than
as black discswe believe it makes sense to search for something like the Pomeron in
D = 2 + 1 SU(N) gauge theories.
Although our primary aim here is to see if we can learn something about the Pomeron
from our lattice glueball calculations, the fact is that our results can also be used both to
test models and to test the accuracy of approximate analytic calculations. An example of
the former is the flux loop model for glueballs [8] which has recently been compared
to the low-J glueball mass spectrum of D = 2 + 1 SU(N) gauge theories [10]. An
example of the latter is transverse lattice light front quantisation [11]. (Which has some
problems similar to ours in determining states of arbitrary J .) Another example is
the conjectured duality between supergravity and large N gauge theories [12] and its
generalisation to the non-supersymmetric case [13]. Extensive quantitative calculations
have been undertaken [14] to compute glueball masses in (2 + 1) and (3 + 1) dimensions
using the AdS/CFT correspondence. The spectra corresponding to the quantum numbers
2++ , 1+ , 1 , 0++ , 0 and 0+ have been computed. It would be interesting to have
a computation of the higher spin states in order to compare them to the new lattice data and
to see if they lie on straight Regge trajectories. That would require the inclusion of stringy
corrections to the low-energy effective action. However, because these higher spin states
derive from string dynamics, one can expect them to lie on straight Regge trajectories.
The contents of this paper are as follows. In the next section we remind the reader of
two simple, but plausible, models for glueballs in (3 + 1) dimensions and show how they
directly carry over into (2 + 1) dimensions where they also predict linear Regge trajectories
with small slopes. In the following section we discuss high-energy scattering. We begin
by showing how Regge theory can be applied in two space dimensions (with the details
relegated to Appendices A and B). We then turn to perturbative Pomeron calculations,
where there has been a great deal of work during the last decade, and investigate what
happens when one moves from 3 to 2 space dimensions. Here we start with the old Low
model, and the more recent dipoledipole scattering approach, then we move on to the
issue of gluon reggeisation and finish with the BFKL Pomeron. Having established the
general theoretical and phenomenological background to the problem, we turn to our lattice
calculation of how glueball masses increase with their spin J . The calculation is novel in
two respects. Firstly one has to surmount the problems posed by the limited rotational
invariance of a square lattice. This we do using the techniques recently developed in [6].
Secondly, the fact that the higher-J glueball masses are considerably heavier makes it
difficult to calculate their masses from Euclidean correlators. Here we use a multi-level
algorithm recently developed in [15], which extends to glueballs earlier work [16] for
Polyakov loops. Using these techniques we obtain quite accurate continuum extrapolations
of the J = 0, 2, 4, 6 glueball masses. We find that the lightest glueballs of even J lie on a
linear trajectory in a ChewFrautschi plot of J versus m2 , and that the slope is small just as
one would expect for a Pomeron pole. However the intercept is much too low to provide a
constant high-energy cross-section, and we discuss the physical implications of this result.
114
Finally we present some results for the leading glueball trajectory in SU(N > 2) gauge
theories, showing that there is no qualitative change as N varies from 2 to .
We conclude with a summary, a discussion of how our calculations should be improved
upon, and a mention of the preliminary results we have obtained in a similar study of the
quenched (3 + 1)-dimensional SU(3) theory.
2. Glueball models and the large-J limit

In the standard valence quark picture, a high-J meson will consist of a q and q rotating
rapidly around their common centre of mass. For large J they will be far apart and the
chromoelectric flux between then will be localised in a flux tube which also rotates rapidly,
and so contributes to J . In a generic model of such a system (see, e.g., [17]) a simple
calculation (that we shall repeat below) shows that the spin and mass are related linearly,
J = 0 + M 2 , and that the slope is related to the tension f of the confining string as
= 1/2f . (The subscript f indicates that the charges and flux are in the fundamental
representation. We will often follow convention and use f instead.) If one uses a
phenomenologically sensible value for f one obtains a value of very similar to that
which is experimentally observed for meson trajectories. This picture might well become
exact in the large-N limit where the fundamental string will not break and all the mesons
are stable.
This picture can be generalised directly to glueballs. We have two rotating gluons
joined by a rotating flux tube that contains flux in the adjoint rather than fundamental
representation. This is the first model we consider below. However for glueballs there is
another possibility that is equally natural: the glueballs may be composed of closed loops
of fundamental flux. This is the second model we consider. The first model is natural in a
valence gluon approach, while the second arises naturally in a string theory. They are not
exclusive; both may contribute to the glueball spectrum. Indeed if there are two classes
of glueball states, each with its own leading Regge trajectory, then this might provide an
explanation for why experimentally there appear to be two distinct Pomeron trajectories,
the hard Pomeron and the soft Pomeron [3]. However such speculations belong to
D = 3 + 1 rather than to the D = 2 + 1 gauge theories that we analyse in this paper. What
is interesting for our purposes is that both models can be motivated as easily in D = 2 + 1
as in D = 3 + 1 and in both cases predict (see below) linear glueball trajectories with some
Pomeron-like properties.
2.1. Adjoint string model
In this model a high-J glueball is imagined to be composed of two gluons joined by
a string in the adjoint representation, with the whole system rotating rapidly. This is
a direct extension to glueballs of the usual model for high-J q q mesons where the q
and q are joined by a string in the fundamental representation. The adjoint string is of
course unstable, once it is long enough (as it will be at high J ), but this is also true of
the fundamental string in QCD. What is important for the model to make sense is that the
decay width should be sufficiently small. (Essentially that the lifetime of the adjoint string
115
should be much longer than the period of rotation.) In SU(N) gauge theories, both the
adjoint and fundamental strings become completely stable as N . So if we are close
to that limit the model should make sense. Since adjoint string breaking in SU(N) occurs
at O(1/N 2 ) while fundamental string breaking in QCDN occurs at O(1/N), one would
expect the instability to be less of a problem in the former case. Moreover there is now
considerable evidence [1820] from lattice calculations that the D = 3 + 1 SU(3) gauge
theory is indeed close to SU(), and that this is also the case for the D = 2 + 1 SU(2)
gauge theory [5].
The calculation of how J varies with the mass M of the glueball is exactly as for the q q
case [17]. That is to say, we consider the string joining the two gluons as a rigid segment
of length 2r0 , rotating with angular momentum J (the contribution of the valence gluons
being negligible at high enough J ). The local velocity at a point along the segment is thus
v(r) = r/r0 (one maximises J at given M if the end-points move with the speed of light),
so that
r0
M =2
= a r0 ,
(1)
a rv(r) dr

= a r02 ,
2
2
1 v (r)
(2)
r0
J =2
0
a dr
1 v 2 (r)
and, eliminating r0 ,
J=
M2
2a
(3)
= 1
we obtain a linear Regge trajectory of slope AS
2a where a is the adjoint string
tension. So this model predicts that the slope of the leading glueball trajectory is smaller
than that of the leading meson trajectory by a factor a /f . For SU(3) we know that
0.88/2.25
a
2.25f [21], so the leading glueball trajectory will have a slope AS
1
2

0.39 GeV if we input the usual Regge slope of about R = 2f
0.88 GeV2 . This
is only a little larger than the actual slope of the Pomeron. Thus to this extent the model
is consistent with the idea that the Pomeron is the leading glueball trajectory, perhaps
modified by mixing with the flavour-singlet meson Regge trajectory. Unfortunately the
model cannot predict the intercept 0 of the trajectory, because it is valid at best at large J .
Since in this model the rotating glueball lies entirely within a plane, the calculation is
identical for D = 2 + 1 and D = 3 + 1, as indeed is the motivation for the applicability of
the model. Thus it is a plausible model for the leading Regge trajectory in the D = 2 + 1
SU(2) gauge theory that we investigate in this paper. The only difference is that in SU(2)
one expects a
8f /3 and hence an even flatter trajectory than in SU(3).
The above simple classical calculation can be made more rigorous in an effective
Hamiltonian approach; see [1] for a review. There the calculation is for D = 3 + 1, but
the main conclusion remains the same in (2 + 1) dimensions.
116
2.2. The flux-tube model

An open string model of the kind described above, is essentially forced upon us if we
wish to describe high-J mesons within the usual valence quark picture. For glueballs,
however, there is no experimental or theoretical support for a valence gluon picture.
A plausible alternative is to see a glueball as being composed of a closed loop of
fundamental flux. Such a picture arises naturally in a string theory approach to gauge
theories. A simple first-quantised model of glueballs as closed flux tubes was formulated
some time ago [8] and has been tested with some success [9,10] against the mass spectrum
of D = 2 + 1 SU(N) gauge theories as obtained on the lattice [5].
In this model the essential component is a circular closed string (flux tube) of radius .
There are phonon-like excitations of this closed string which move around it clockwise
or anticlockwise and contribute to both its energy and its angular momentum. The system
is (first) quantised so that we can calculate, from a Schrdinger-like wave equation, the
amplitude for finding a loop of a particular radius. The phonon excitations are regarded as
fast so that they contribute to the potential energy term of the equation and the phonon
occupation number is a quantum number labelling the wave-function. If we are interested
in the lowest mass at a given J , as we are for the leading trajectory, then we want the
potential energy [10] that corresponds to the minimum number of phonons needed to
provide that J
E() = 2f +
J c
(4)
and we minimise this expression with respect to (in much the same way as in bag models)
to obtain the glueball mass M
M = min E() = (8f J )1/2 .
(5)
This corresponds to a linear Regge trajectory

J=
M2
8f
(6)
= 1 = 1
0.22 GeV2 , if we input the usual value for the Regge
with a slope FT
8f
4 R
slope (as in Section 2.1). This is similar to, but somewhat smaller than, the observed
Pomeron slope. This analysis can be readily transformed into a variational calculation that
minimises the Hamiltonian, without changing the final conclusion (for large enough values
of J ).
The above calculation was carried out for the D = 2 + 1 version of the model. In
D = 3 + 1 there are extra phonons corresponding to fluctuations of the loop that are
orthogonal to the plane of the loop, and in addition the system can acquire angular
momentum through spinning about its axis. This alters the details of the calculation, but
not the qualitative conclusion.
Finally we remark that for SU(N > 3) gauge theories the fundamental string is no
longer the only one that is absolutely stable, and closed loops of these higher representation
strings provide an equally good model for glueballs [7,10]. These extra glueballs will
117
however be heavier and, to the extent that we are only interested in the leading Regge
trajectory, will not be relevant for the discussion of this paper.
3. High-energy scattering in (2 + 1) dimensions

We begin by describing how the familiar Regge theory of 3 spatial dimensions can be
taken over to 2 space dimensions. We confine ourselves to a brief summary here, leaving
details to Appendix A.
The Pomeron has been the focus of a large number of field theoretic studies, many
motivated by the hard Pomeron and low-x physics in deep inelastic scattering. Just as
we asked how glueball models change when we go from (3 + 1) to (2 + 1) dimensions,
it is interesting to know the predictions of these perturbative approaches when we change
dimension. That will be the focus of the remainder of this section.
3.1. Regge theory predictions
The optical theorem relates the total cross section to the dimensionless scattering
amplitude T (s, t) by
1
tot = Im T (s, t = 0).
s
(7)
(In two space dimensions the cross section has dimensions of length.) As is shown in
Appendix A, it is given by the following contributions:
T (s, t) = a0 (s) + background integral

Regge pole contributions s (t ) ,
+
(8)
where (t) describes the Regge trajectory in the ChewFrautschi plot. This equality is
based on the analytic continuation of the partial waves in , the angular momentum, and
on crossing symmetry. There are two differences with respect to the (3 + 1)-dimensional
case: the background
integral gives a constant contribution to the amplitude, rather than
decreasing as 1/ s; and the s-wave exchange is not included in the SommerfeldWatson
transform. In potential scattering, and even more general situations, it can be shown to be
a branch point in the complex plane at threshold (see [22] and Appendix B).
3.2. QCD2 at high energies
We first give the simplest estimates of the color-singlet exchange for high-energy
scattering. We then comment on the failure of gluon reggeisation and review the
results of Li and Tan [23] for color-singlet exchange obtained in the leading logarithm
approximation. In order to develop some intuition for (2 + 1)-dimensional physics, we
finish with a discussion of the momentum dependence of hadronic structure functions.
118
3.2.1. Color singlet exchange in leading order

If we compute the color-singlet part of a two-gluon exchange diagram between two
quarks in (2 + 1) dimensions (the first perturbative Pomeron model due to Low in 1975
[24]), we find

N2 1
dk
(1)
,
A1 = is2 s c 2
(9)
Nc
k 2 (k q)2
implying, by use of the optical theorem in (2 + 1) dimensions (Appendix A),

N2 1
dk
tot (qq qq) = s2 c 2
.
Nc
k4
(10)
The result is completely similar to the (3 + 1) case, except that the IR divergence is
worse by one powertot has units of length. The Pomeron exchange amplitude is finite
once impact factors are introduced for the hadronsgiving the incoming quarks a slight
offshellness and effectively introducing a cutoff in the integral on the right-hand side of
(10).
Another approach consists in computing dipoledipole scattering (for an introduction,
see [25] and references therein). Proceeding as in the (3 + 1)-dimensional case, the leading
order (large N ) dipoledipole cross-section reads
d"

(1 cos "x01)(1 cos "x01
)
"4

x>
2 3
= x< 3
1
(2 + 1)
x<
as compared to [25]

x>
2
dd = 2s2 x<
1 + log
(3 + 1),
x<
dd = 4s2
(11)
(12)
. In both cases, we
where x> (x< ) is the greater (lesser) of the two dipole sizes x01 and x01
find a constant cross-section.
To go beyond the leading contribution, several calculational schemes are available. In
particular, the BFKL Pomeron is obtained by keeping, order by order in g 2 , only the leading
logarithmic contribution in the perturbative expansion. The first step in calculating the
amplitude for Pomeron exchange is to establish gluon reggeisation.
3.2.2. The issue of gluon reggeisation

In the Regge limit s t g 4 , where s, t are the Mandelstam variables, the usual
expansion in g 2 log s yields the 0-, 1- and 2-loop amplitudes for color octet exchange:
(8) s
A(8)
(13)
0 = 8s G0 t ,
s
(8)
(8)
A1 = A0 &G (t) log 2 ,
(14)
k

1 (8)
s 2
A(8)
(15)
A
&
=
(t)
log
,
G
2
2 0
k2
119
where k 2 = O(t) and

s =
g2
,
4
a a
G(8)
0 = ij kl ,
+
&G (t) = Nc s

dk
t
0 t = q 2 .
2
2
2 k (k q)
(16)
Thus, at least formally, the gluon reggeises [26]:

A
(8)
(8)
= A0
s
k2
&G (t )
.
(17)
The infrared divergence in the quantity &G (t) is linear (as opposed to logarithmic in
(3 + 1) dimensions), and it must be so since s carries dimension of mass. The IR gluon
mass cutoff M has to be introduced, in which case &G (t) = Nc s /M. Physically M
can be interpreted as a non-perturbative mass that the gluon acquires at the confining
scale; therefore we expect g 2 /M = O(1). This however shows that, due to the infrared
divergence, the result of the perturbative calculation has a linear sensitivity to physics at
the confinement scale g 2 , where the perturbative expansion breaks down.
In the Verlinde approach [27] to high-energy scattering adopted by Li and Tan [23],
gluon reggeisation fails. However, as the authors remark, this is not necessarily in
contradiction with conventional perturbative calculations, since the really physical quantity
is the color-singlet exchange.
3.2.3. The 2 + 1 perturbative Pomeron
The BFKL equation was solved exactly in the presence of a gluon mass in [26]. However
when this mass is taken to zero, the IR divergence shows up in the fact that the BFKL
exponent 0 runs as s /M; this fact could be guessed on dimensional grounds. Within
perturbation theory, such a mass M can only appear as an IR regulator. The situation is in
radical contrast to the (3 + 1)-dimensional case, where the cancellation of IR divergences
in the BFKL equation makes it self-consistent. In the detailed calculation, the simple
structure of the infinite series is spoiled in the M 0 limit by the re-emergence of a
power dependence on s at each order due to the IR divergences. Thus, in this framework,
a power-like dependence of the cross-section on s in the limit of zero gluon mass is not
possible in (2 + 1) dimensions.
A thorough investigation of QCD2 high-energy scattering was undertaken by Li and Tan
[23]. In their first paper, they used the Verlinde approach [27] to obtain a one-dimensional
action, where they are able to compute the (finite) color-singlet exchange exactly. They
predict a
1/ log s
(18)
dependence of the total cross-section on the center-of-mass energy. In a second publication,

they re-derive this result using the dipole picture ([25] and references therein) of highenergy scattering. In this case all quantities are naturally IR-safe.
120
3.2.4. Deep inelastic scattering in (2 + 1) dimensions

A standard prediction of the BFKL Pomeron in (3 + 1) dimensions is the strong rise of
the deep inelastic structure functions at low x when Q2 is large (for an introduction, see
[2]):

x 0
F x, Q2
,
log 1/x
(19)
where 0 = 4 Nc s log 2 is the BFKL exponent.

On the other hand, the DGLAP equation for the evolution of the moments M(n, Q2 ) of
the parton distributions leads to the behaviour

Q2 An
2
M n, Q = Cn log 2
,
(20)
where the pure number An is an anomalous dimension computed in perturbative QCD.

The Q2 dependence comes from the running of the coupling s ; in (2 + 1) dimensions, the
equation is therefore replaced by

M(n, Q2 )
s
= An M n, Q2
2
log Q
Q
yielding the following large Q2 behaviour:

2An s
s
2
M(n, ) 2An .
M n, Q = M(n, ) exp
Q
Q
(21)
(22)
That is to say, the structure functions tend to finite constants at large Q2 . The physical
reason for this is that at large-energy scales, the theory becomes free very rapidly (the
effective coupling scales as 1/E), which does not allow for an evolution of the structure
functions. Thus above the confinement scale, we qualitatively expect a rapid evolution in
Q2 of the structure function toward its asymptotic value. Once a high Q2 has been reached,
the virtual photon does not see more partons when its resolution is increased, because
the amplitude that they be emitted is suppressed by s /Q.
4. The D = 2 + 1 SU(2) glueball spectrum

In the first part of this section we discuss the technical aspects of the lattice calculation.
In the second part we present the results of our calculations.
4.1. Technical aspects
The identification of the continuum spin of a lattice state requires novel techniques, as
does the accurate calculation of the mass of a heavy, high-spin state, and we discuss these
below. First however we briefly summarise the more standard aspects of the calculation.
121
4.1.1. Masses from lattice simulations

To obtain masses we calculate Euclidean correlation functions of some well-chosen
operators i

Cii (t) i (t)i (0) = i eH t i =

|an |2 eEn t .
(23)
n
Here the En are the energies and an = n|i |vac. If the vacuum has trivial quantum
numbers, then only states with the quantum numbers of i will have an = 0. Suppose
we have i = 1, . . . , n0 operators of the same quantum numbers and we also calculate
the off-diagonal correlators Cij (t). Then an effective way to calculate the lightest few
states with these quantum numbers is to perform a variational calculation using the basis
{i ; i = 1, . . . , n0 }. For a recent exposition of this standard method see [5].
To calculate such Euclidean correlation functions we use lattice Monte Carlo methods.
As usual one needs to ensure that the operators are smooth and extended, so that they
have a good projection onto the lighter physical states, and we use an iterative smearing
technique for that purpose [28]. We use an increasing number of smearing iterations (and
also increase the smearing weights) as we approach the continuum. All our calculations
use the standard Wilson action. The update is a 1 : 3 mixture of heat-bath [29,30] and
over-relaxation [31] sweeps. Because we calculate the values of many operators, the
measurement is time-consuming and we can do a significant number of these compound
sweeps between measurements without significantly increasing the total cost of the
calculation. We typically perform O(105 ) sweeps and collect the data in 100 bins. Errors
are estimated with a standard jackknife analysis.
4.1.2. Continuum spin on a cubic lattice
Consider the eigenstates of the transfer matrix of the lattice field theory. These will
belong to the irreducible representations of the cubic rotation group and will not, in general,
possess the rotational properties that characterise a continuum state of a definite spin.
However as a 0 each of these states will tend to an energy eigenstate of the continuum
theory that possesses some definite spin J . By continuity for sufficiently small a the
rotational properties of this lattice state will be arbitrarily close to those of a continuum
state of spin J . We will therefore refer to such a state not only by its lattice representation
but also by the appropriate continuum spin J . To be able to do this we need to identify, for
each such lattice state, what continuum J it tends to as a 0. Once we know this then we
can perform a standard continuum extrapolation of the calculated lattice masses so as to
obtain the continuum mass of the spin J glueball.
A detailed investigation into how to do this was presented in [6]. Here we shall apply
a systematic version of the method that was presented there under the name Strategy II.
We briefly remind the reader of this method.
The operators we use lie in definite lattice irreducible representations (IRs), and we use
the variational method [32] to extract estimates for the eigenstates (in our operator basis)
and their masses. In this way we calculate the mass of the lightest state and of several
excited states in the given lattice IRtypically the number is one third of the number of
operators we are using. To identify which J each of these states tends to, we do a simple
Fourier analysis of the wave function of the corresponding diagonalised operator. To do this
122
we measure the correlations between it and a probe operator that we are able to rotate to
a good approximation by angles smaller than /2. This provides a measurement of the
wave function. We check the rotational properties of the probe by measuring the vacuum
wave function; typically it is found to be isotropic at the one percent level, which is taken
as evidence that the probe has good rotational properties. We are mainly interested in the
dominant component and the fact that there is an uncontrolled uncertainty of a few percent
in the measurement of the wave function does not impede significantly the procedure of
determining the spin of a state in the continuum limit. This is so because one is usually
discriminating between a constant and a change of sign behaviour of the wave function. In
principle, the continuum extrapolation of these coefficients should determine uniquely the
spin of the glueball. In practice, it usually turns out that there is already a very dominant
coefficient at finite lattice spacingexcept when a crossing of states occursas we shall
see in an example later on.
4.1.3. The multi-level algorithm
A state with high spin will generically have a large mass, and the value of the
corresponding correlation function can be very small. Indeed, even if we have a perfect
operator the value of the correlator will be O(eaMJ nt ) at t = ant and if aMJ 1 then it
will be in danger of being swamped by the statistical errors. In practice it is not possible
to restrict oneself to values of a such that aMJ 1 for all values of J that are of interest.
Especially so because we need a range of a that is large enough for a well-controlled
continuum extrapolation.
To deal with this problem we make use of a recently proposed error reduction algorithm
for glueball correlators [15]. It proved very useful on all but the smallest lattice spacings
(i.e., on = 6, 7.2, 9, 12). For 6 9, we used O(500) sub-sweeps, while we decreased
their number to 50 at = 12. These sub-sweeps are done on sub-lattices which represent
time-blocks of width 4. Our experience is that it is more efficient to do all these sweeps
at one fixed time-block (that is, a 2-level algorithm) rather than splitting up the sweeps
between width 2 and width 4 blocks (that is, a 3-level algorithm). This was noted previously
in [16], where a multi-level algorithm was applied to Polyakov loop measurements, and
suggested in [33].
The choice of the number of sweeps was done on the basis of the thumb-rule nsweeps =
em6t , where in our case 6t = 4a and the algorithm was optimised for the spin 4
glueball, that is m = m4 . Indeed when one is measuring large portions of the spectrum,
a compromise has to be made. For the lighter states, it is more efficient to do few subsweeps, while the heavier states require more of them. That is, for the same computer time,
we could have measured the lightest glueball mass more accurately had we not used a 2level algorithm, but we would then have far less accurate and reliable results for the spin 4
and 6 glueballsthe main goal of the present simulations.
The multi-level algorithm enables us to apply the variational method on the correlation
matrices at 2 and 3 lattice spacings even on the coarsest lattice. Very often, when the
correlators are measured in the traditional way, the noise that dominates the signal of the
heavier states spoils the positivity of the correlation matrices if the method is not applied
between 0 and 1 lattice spacing, thus impeding the variational calculation. In our case
however the very massive eigenstates benefit most from the sub-averaging procedure and
123
the positivity is maintained. Usually the orthogonalised operators have reached their mass
plateaux (within error bars) by three lattice spacings for the data we present in this paper;
this increases our confidence in the reliability of the variational method.
At = 18, the correlation length 1/
12a; that would be the natural width in
the Euclidean time direction over which to do sub-sweeps. However that would mean
extracting the masses at time separations of that order. It turns out that our smeared
operators have sufficiently good overlaps onto the physical states to reach a plateau far
earlier than 12 lattice spacings. For that reason, we did not apply the multi-level algorithm
to that case. Nevertheless we regard this as a consequence of the improved smoothness of
the operators close to the continuum rather than as a defect of the multi-level algorithm.
4.2. Results
In Table 1 we list the values of the masses we calculate on L3 lattices at various values
of = 4/ag 2 . The masses are in lattice units and are labelled both by the lattice IR to
which they belong, and by the spin J of the state to which they tend in the continuum
limit. The latter assignment is achieved as described above, and an explicit example will be
given below. We also have calculated the confining string
tensions as indicated. The string
tension provides a natural dynamical
length
scale
1/
which tells us how small
the
lattice spacing a is, a/ = a , and how large our lattice size, aL, is, aL/ = aL .
Table 1
The lightest (2 + 1) SU(2) glueball masses on L3 lattices at the values of indicated. An asterisk on the string
tension indicates that the value is taken from [5]
=6
L = 16
= 7.2
L = 20
= 7.2
L = 40
=9
L = 24
= 12
L = 32
= 12
L = 42
= 18
L = 50
0.2538(10)
0.2044(5)
0.2072(46)
0.1616(6)
0.1179(5)
0.1179(5)
0.0853(14)
0+
0+
0+
0+
4+
4+
1.198(25)
1.665(43)
2.198(76)
2.27(10)
2.44(27)
0.981(14)
1.396(21)
1.859(25)
2.084(41)
2.07(33)
2.53(13)
0.951(14)
1.394(18)
1.778(34)
2.067(54)
2.146(64)
2.50(14)
0.7652(78)
1.108(23)
1.426(37)
1.522(36)
1.570(39)
1.700(52)
0.570(11)
0.847(18)
0.980(28)
1.226(17)
1.195(47)
1.419(90)
0.577(13)
0.839(40)
1.00(60)
1.16(12)
1.259(98)
1.500(48)
0.3970(78)
0.584(32)
0.717(76)
0.845(37)
0.798(32)
0.963(45)
2.54(12)
2.210(53)
2.270(64)
1.957(48)
2.08(18)
2.34(25)
2.65(29)
2.93(21)
1.584(18)
1.870(37)
2.219(90)
2.451(71)
2.51(19)
1.567(18)
1.891(39)
2.242(77)
2.47(12)
2.64(15)
1.232(38)
1.421(44)
1.660(54)
1.746(56)
1.878(86)
0.933(11)
/
1.152(42)
1.459(29)
1.438(28)
1.035(16)
1.090(19)
1.096(92)
1.385(36)
1.544(60)
0.634(18)
0.667(20)
0.862(14)
1.019(92)
0.906(69)
1.274(37)
1.359(46)
1.629(59)
1.741(59)
1.949(64)
0.931(24)
/
1.222(19)
0.643(19)
0.676(27)
1.411(21)
0.952(61)
IR State
A1
A2 4
A3
2+
2+
2+
2+
6+
2.071(48)
A4 2
2.084(56)
2
2
2
6
124
Since high-J states are expected to be very extended, it is important to check that our
J = 0, 2, 4, 6 mass estimates are not subject to large volume corrections. This is the first
thing we do in this subsection. We then give an explicit example of the Fourier coefficients
which we use to identify the J of the state. Finally we discuss the extrapolation to the
continuum limit.
4.2.1. Finite volume effects
As one can
see from Table 1, the spatial size that we use for most of our calculations
satisfies aL 4. This choice was based on earlier finite volume studies [5] that showed
that it appeared to be large enough for the lightest glueball states. In particular, on such a
volume the lightest state of two periodic flux loops (which can couple to local glueball
operators) will be heavier than the lightest few A1 states and the lightest A3 state. In
this paper, however, we are interested in higher spin states that may be significantly more
extended than these lightest states, so it is important to check for finite volume corrections
by performing at some the same calculations on much larger volumes. We do this at
= 7.2, where the spatial extent of our comparison volume is twice as large. In addition
we perform a more limited comparison at a finer lattice spacing, = 12, on a comparison
lattice that is about 30% larger.
We see from Table 1 that there is in fact no significant
change
in any of the masses
listed when we double the lattice size from aL 4 to aL 8 at = 7.2. In

particular this is true for the J = 4 and J = 6 states where our concern is greatest. We
note also that on the L = 40 lattice a state composed of two periodic flux loops will have
a mass amT 2La 2
3.45 which is much heavier than any of the masses listed and so
it will not be a source of finite volume corrections there. From the comparison we infer
that these torelon states cause no problem on the L = 20 lattice even though their mass
amT 1.7 is light enough for it to mix with the states of interest. This tells us that in
general such mixing will not be important even where it is possible. This is consistent with
the observation [5] that in many respects SU(2) is close to SU(), since in the latter case
the mixing will vanish.
The more limited finite volume check at = 12 also shows no significant volume
variations for the J = 4 and J = 6 states, as well as the J = 0 states. On the other hand
there are some anomalies in the J = 2 part of the A3 spectrum. Since on these two lattices
the torelon state has a mass amT 0.9 and amT 1.2, respectively, it is possible that
it is mixing with some nearly degenerate A3 glueball states on both the lattice sizes, and
that this explains the anomalies. However we also remark that these calculations were
performed at an early stage (unlike those at = 7.2) and a different basis of operators
was used on the two volumes. Thus it is not clear how seriously we should take the
discrepancies observed in the A3 sector.
4.2.2. The Fourier coefficients
As described in Section 4.1.2, we determine the J of a lattice energy eigenstate by a
Fourier decomposition of its wavefunction. For example, for a state in the trivial A1 lattice
IR we have

cn (j = 4n)+ lat with
|cn |2 = 1.
|1 =
(24)
n0
125
Table 2
The Fourier coefficients of the spin J states given in Table 1: |cJ |2 at = 7.2, 9, 12 and 18. When the coefficient
is larger than 0.99, we round it off to 1
State
= 7.2, L = 20
= 9, L = 24
= 12, L = 32
0+
0+
0+
0+
4+
4+
1
1
1
0.59(12)
0.94(9)
/
1
1
1
0.68(62)
0.98(4)
0.67(16)
1
1
1
0.97(13)
0.38(2)
0.59(4)
2+
2+
2+
2+
6+
1
1
1
1
1
1
1
1
0.87(8)
1
1
0.97(11)
1
0.88(4)
= 18, L = 50
1
1
/
1
0.95(2)
0.98(3)
The Fourier decomposition is performed using as a probe a loop O for which we have a
number of other loops O that are (to a good approximation) copies of O rotated by angles
that are not multiples of /2:

cO cn ei4n
vac|O |1 =
(25)
n0
(where cO is independent of n in the limit where O is an exact rotated copy of O).

Note that since the angular resolution is O(a) one can attempt to resolve spins up to
Jmax O(1/a) in this way. In Table 2 we show the Fourier coefficients calculated at
the lattice spacings = 7.2, 9, 12, 18. The table shows the normalised |c|2 coefficients
corresponding to the spin that the state is assigned in the continuum limit. We see that the
states that become 0+ have very isotropic wavefunctions even at the finite lattice spacings
considered. The spin 4 coefficients of the spin-4-to-be states vary a lot more. Let us look
at the fundamental spin 4 glueball in more detail.
The coefficient is very close to one at = 7.2, 9 and 18, but shows a big dip at = 12.
We attribute this to the crossing of the lightest spin 4 state and the 0+ . Indeed looking
at the masses in Table 1, we observe that these two states are always nearly degenerate, the
spin 4 being slightly heavier on the coarse lattices and slightly lighter on the finer lattices,
while they are closest precisely at = 12. As was pointed out in [6], an accidental
degeneracy like this automatically leads to maximal quantum mechanical mixing between
the states, since there is no lattice symmetry to prevent that. Taking this into account, the
observed evolution of the Fourier coefficient is not implausible.
At = 6 we did not perform a Fourier decomposition but rather chose the value of J
using the level ordering already established for the other values of . The reason we did not
perform such a decomposition is that this calculation was originally intended as a test of the
multi-level algorithm rather than as a contribution to the present study. However given its
accuracy it seemed wasteful not to use it. We have a less accurate older study [6] where the
Fourier decomposition was performed and that supports our spin identification. In addition
the mass of the identified 4 state, and the fact that it should be (nearly) degenerate with
the 4+ leaves no doubt about the correctness of the J = 4 identification in this case as well.
126
Fig. 1. The continuum extrapolation of the lightest (2 + 1) SU(2) glueball in the 0+ , 2+ , 4+ and 6+ sectors. The
points at a 2 = 0 represent the result of the continuum extrapolation.
4.2.3. Continuum extrapolation

Our continuum extrapolation of the states in the A1 and A3 representations is entirely
conventional and follows [5,6]. We plot the glueball masses in units of the string tension
as a function of a 2 , and attempt a linear fit
amG (a) mG (a) mG (0)
=
+ ca 2.
a (a)
(a)
(0)
(26)
This is motivated by the fact that we know the leading lattice correction to be O(a 2 ) for
the plaquette action. If such a fit has a bad confidence level, then we remove the coarser
lattice data until a good fit is obtained. We require that at least three points are left.
Fig. 1 shows the actual extrapolation for the lightest state of each spin. We observe as
in [6] that the evolution is weak. Note also that thanks to the error reduction technique
employed, the error bars corresponding to the coarser lattices do not appear very much
larger than those associated with finer lattices. Table 3 gives the continuum spectrum in
units of the string tension, as well as the confidence level and the number of different
lattice spacings included in the fit. For the fundamental states of spin 0, 2, 4 and 6, the
confidence levels are good and include all five lattice spacings.
Not surprisingly, the second and third excited states have less reliable extrapolations.
A slight dependence of the masses obtained for the excited states on the basis used in the
variational method is likely to be the cause of the stronger spread of the lattice data in the
extrapolation plot. Notice, for instance, that all four states shown on Fig. 1 appear slightly
lighter at = 18 than at the other lattice spacings.
Table 3
The lightest (2 + 1) SU(2) glueball masses in the continuum limit
State
m/
Confidence level
127
Number of
0+
0+
0+
0+
4+
4+
4.80(10)
7.22(24)
8.47(30)
11.15(45)
9.75(45)
12.06(88)
78
64
17
14
71
29
5
4
5
3
5
3
2+
2+
2+
2+
6+
7.85(15)
7.90(25)
10.00(33)
13.90(71)
12.09(40)
50
15
32
35
38
5
5
5
3
5
5. SU(N > 2)
As we remarked in Section 1, it is only in the N limit, where all glueballs
become stable, that one can hope to identify the ideal linear Regge trajectory. In principle
all one needs to do is to repeat the above SU(2) calculation for N = 3, 4, 5, . . . . We
know from [5] that the approach to N = is rapid so that the first few values of
N should suffice for a good extrapolation to all values of N . However while such a
calculation is certainly feasible, it is beyond the very limited computational resources
currently available to us. Fortunately we are able to finesse this practical problem using
the fact that it has been shown in [6] that the lightest state in the lattice A2 IR, which
contains J P = 0+ , 4+ , 8+ , is in fact the 4+ rather than the 0+ . This confirmed
earlier suggestions, based on an analysis of the predictions of flux-tube models [10], that
the lightest 0+ should be much more massive than the lightest observed A2 state, while
the latter was consistent with the model prediction for the lightest 4+ state. Due to
parity doubling in D = 2 + 1 this mass is the same as that of the 4++ (in the infinite
volume continuum limit). Thus we can use the lightest states in the A1 , A3 and A2 lattice
representations, as calculated for various SU(N) groups in [5], to provide us with the
lightest J = 0, 2, 4 glueball masses. (Note that this means that the masses labelled in
the tables of [5] as being those of the lightest 0+ should in fact be relabelled as being
the lightest 4+ .) This is more limited than our explicit SU(2) calculation where we also
identify the J = 6 glueball, but it is adequate given the presumption that there will be no
qualitative change as we increase N from N = 2.
The assumption that for SU(N > 2) the lightest A2 state is the 4+ is very reasonable
given that this is so in SU(2) [6] and that it is predicted to be so by generic flux-loop models
[10]. Nonetheless it is an assumption and should be checked. We have therefore performed
such a check in the SU(5) case, at = 64, L = 24, where 1/2
6a. Using a 16-fold
rotated triangular probe operator reveals that the wave function of our best A2 operator,
measured at a Euclidean time separation of one lattice spacing, behaves like sin 4x, with
a normalised coefficient consistent with 1 at the few percent level. This confirms the
correctness of our assumption.
128
6. Physical discussion
We begin by asking what our glueball spectrum tells us about the nature of the leading
glueball Regge trajectory, both for SU(2) and for larger N . We then compare what we find
to the predictions of the simple glueball models in Section 2. Finally we discuss what role
this trajectory will play in high-energy scattering.
6.1. The glueball spectrum in a ChewFrautschi plot
In Fig. 2 we plot our continuum SU(2) glueball spectrum in a ChewFrautschi plot of
m2 / against the spin J . We see that the lightest J = 0, 2, 4, 6 masses appear to lie on a
straight line. If we fit them with a linear function J = (t), where (t) = 0 + t and
t = m2 , then we obtain

2 (m)
= 0.322(16),
(m)
= 1.18(11),
(27)
with a confidence level of 65%. (If we drop the J = 0 state from the fit, the errors become
somewhat larger, but the trajectory is essentially the same.) Thus we reach the remarkable
conclusion that the lightest glueballs of spin J fall on a linear Regge trajectory. This is the
leading trajectory, hence the index (m) standing for mother trajectory.
Although there is more uncertainty in establishing the excited spectrum, particularly in
the spin 2 sector where finite volume effects are not completely understood, we also fit the
0+ , 2 and 4 states to a straight line and find

2 (d)
= 0.265(36),
(d)
0 = 2.20(44),
(28)
with a confidence level of 93%. This daughter trajectory is approximately parallel to the
leading one, and its intercept is down by about one unit.
As we explained in Section 5, we can also say something about the leading Regge
trajectory in SU(N > 2) gauge theories, if we use the masses calculated in [5] and relabel
as 4+ the state that is labelled there as the lightest 0+ . Since the 4+ and 4++ are
degenerate in the (infinite volume) continuum limit, this gives us the 0++ , 2++ and 4++
continuum masses for N = 2, 3, 4, 5. A linear fit, J = (m2 ) = 0 + m2 , works in all
cases and yields:
[5] data
N
N
N
N
=2
=3
=4
=5

2 (m)
0.324(15)
0.384(16)
0.374(18)
0.372(22)
(m)
1.150(75)
1.144(71)
1.068(75)
1.036(88)
Confidence level
89%
54%
71%
86%
We have also included the result for SU(2) and we note that the parameters of the trajectory
are in very good agreement with the data of this paper. It is clear that for all the number of
colors available, the linear fit has a very good confidence level.
129
Fig. 2. The ChewFrautschi plot of the continuum (2 + 1) SU (2) glueball spectrum.
We conclude that all SU(N) gauge theories possess approximately linear Regge
trajectories, with slopes 0.30.4 in units of 2 , and intercepts close to 1, which appear
to be approaching that value as N .
6.2. Comparison to glueball models
As we saw in Section 2, the closed flux-tube model of glueballs predicts a leading Regge
trajectory that is linear, with a slope that is independent of N :
1
N.
(29)
4
The adjoint string model also predicts a linear Regge trajectory but with a slope
= 1 that in general depends on N through the N dependence of / . Lattice
2a AS
a
calculations [21] (and some theoretical arguments [34]) support a dependence that is close
to Casimir scaling,

=
2 FT
CA
a
N2
=
.
=2 2
CF
N 1
(30)
Assuming this, the slope predicted by the adjoint string model becomes
N = 2:
3

2 AS
= ,
8
(31)
N = :
1

2 AS
= 2 (2 FT
)= .
2
(32)
130
Fig. 3. The slope of the leading Regge trajectory in (2 + 1) SU (N ) gauge theory, in units of 1/(2 ), as a
function of 1/N 2 .
Interestingly, for all the N considered, the lattice result for is almost exactly at the
midpoint between the two model predictions. We illustrate this fact in Fig. 3. We might
speculate that even if both models are valid, thus producing two glueball trajectories with
different slopes, at finite N mixing will deform these trajectories from exact linearity and
that such a deformation will be greatest at some lower J where the states of the two
trajectories are closest and also where we perform our calculations.
We observe that the intercept of the leading Regge trajectory that we have obtained is
close to 1, and becomes perhaps even closer at larger N , as we see if we plot
6 (0 + 1),
(33)
in Fig. 4. Now we recall that in D = 3 + 1 [2] the fact that the color-singlet amplitude
is, order by order, down by a factor of Ns with respect to the color octet exchange
amplitude (see Section 3.2) can be interpreted as coming from an expansion of the signature
suppression factor around J = 1, using the fact that in this picture (t = 0) = 1 + O(s ).
A similar argument will work in D = 2 + 1 as long as (t = 0) is near an odd integer. We
assume that (t = 0)
1 is disfavoured since it would lead to a rapidly rising cross-section,
and so the next possibility would be (t = 0)
1, as observed in our calculations. This
line of reasoning relies on the idea that the perturbative calculation remains qualitatively
valid even as t , which is of course not guaranteed.
At finite N Regge trajectories are not expected to rise linearly at arbitrarily large t = m2 .
In particular we should expect that due to mixing between high spin glueballs and multi-
131
Fig. 4. The difference 6 (0 + 1) of the intercept to the value 1, as a function of 1/N 2 in (2 + 1) SU (N )

gauge theory.
glueball scattering states, for which
(t) t,
(34)
the local slope of the trajectory decreases as J increases. This effect is, however,
suppressed by 1/N 2 in the large N limit.
6.3. High-energy scattering prediction
The contribution of the leading glueball trajectory to the total cross-section behaves as
6 s 0 1/2 ,
(35)
which means, given our calculated value 0

1, that it is suppressed as s 3/2 . Thus
the high-energy scattering of glueballs is not dominated by Regge pole exchange in (2 + 1)
dimensions; at least if we believe that the cross-section should be constant at high energies
(up to powers of log s).
Going back to Section 3.1, we note that the other terms contributing to the scattering
amplitude are the isolated s-wave amplitude a0 (s) and the background integral. Because
there is a unitarity bound on each partial wave separately, namely
Im an |an |2 ,
(36)
the contribution of any partial wave amplitude to the total cross-section is bounded by
s 1/2 . Thus this s-wave amplitude will not dominate either at high energies. That,
132
then, only leaves the background integral. If the partial wave amplitude a(, t) were
meromorphic in the region 0 < Re < 1/2, we would simply get additional Regge pole
contributions, which should show up as physical states by analytic continuation. Therefore
there must be a more complicated singularity structure in that region. For instance it is
well known that = 0 is a logarithmic branch point of the partial wave amplitude a(, t)
at low energies ([22]; see also Appendix B). Also, Li and Tan [23] remark in their second
publication that the dipoledipole forward scattering amplitude can be written as a contour
integral in the complex plane around = 0:

2g 2 dd
2g 2 dd 1
d

=
s log ,
A(d, d , s) =
(37)
Nc
log s
Nc
2i
where d, d are the sizes of the scattering dipoles; again, the logarithmic branch point
seems to dominate the scattering process. This intriguing similarity suggests a universal
contribution from the point = 0.
7. Conclusion
In this paper we have carried out a lattice calculation of part of the higher spin mass
spectrum of SU(2) gluodynamics in (2 + 1) dimensions. Such a calculation can tell us what
the leading glueball Regge trajectory looks like and, in particular, whether it resembles the
Pomeron.
To provide some motivation for this question, we showed how simple glueball models
predict linear Regge trajectories, with small slopes, in both (2 + 1) and (3 + 1) dimensions.
We emphasised that, unlike the case of q q mesons, there are two natural models: the
open adjoint string that is the natural extension of the usual Regge model for mesons,
and the closed flux tube which has no analogue for the usual mesons, but which arises
naturally in string theory approaches to SU(N) gauge theories. One may speculate that both
models contribute and that there are two Pomerons (for which there is some experimental
evidence).
Another part of our motivation for a study in D = 2+1 is an intuition that in high-energy
scattering the colliding glueballs should behave like black segments (analogous to the
black discs of D = 3 + 1) so that the cross-section is approximately constant at high s.
Of course in D = 2 + 1 we have no experimental support for such an intuition and we
therefore investigated how various field theoretic approaches to high-energy scattering can
be translated from D = 3 + 1 to D = 2 + 1. The generic change is that infrared divergences
become much more severe so that, for example, one can no longer predict directly from
the BFKL equation [26] a power-like dependence of the cross-section in s. However there
exist alternative analyses [23] done in the framework of leading-logarithmic perturbative
expansion that do indeed obtain cross-sections which are constant (up to logarithms).
The framework for Regge poles is Regge theory and we saw that there are significant
changes when we go from 3 to 2 spatial dimensions. In particular the = 0 partial wave
is not included
in the SommerfeldWatson transform and the background integral is only
down by 1/ s. (In D = 2 + 1 the intercept of a trajectory J = (t) that gives a constant
cross-section is at (0) = 1/2 in contrast to (0) = 1 in D = 3 + 1.) One can imagine that
133
the complicated singularity structure at Re = 0, which is not associated with particles of

the theory, might be promoted to a dominant contribution to the high-energy cross-section.
With the above background in mind, we presented the results of our lattice calculation
of the higher spin glueball spectrum. This is a pioneering calculation and like all such
calculations can be improved upon in many respects. However we are confident in the
robustness of the results that we obtain. In particular, extrapolating our masses to the
continuum limit shows that the leading Regge trajectory in the (mass)2 versus spin plane
is in fact linear (to a good approximation). Moreover it has a small slope that lies roughly
midway between the predictions of the flux-tube and adjoint-string models. The intercept
at t = m2 = 0 is 0
1. We identified a parallel daughter trajectory, lying about one unit
of J lower. We were also able to determine the leading Regge trajectory for other SU(N)
groups and found that the result depends very little on N . In particular it is essentially the
same for the theoretically interesting SU() limit.
The very low intercept of the leading glueball trajectory (0
1) indicates that the
moving Regge pole corresponding to these glueball states gives a negligible contribution
to high-energy scattering in (2 + 1) dimensions. We concluded that there must be a more
complicated singularity structure of the partial wave amplitude in the complex angular
momentum plane . Evidence for a possibly universal branch point at = 0 comes mainly
from low-energy potential scattering (where the result is independent of the potential [22])
and is suggested by the 1/log s scattering amplitude found by Li and Tan in QCD2 highenergy scattering.
These statements are all quite different to what one expects in (3 + 1) dimensions.
There the phenomenological Pomeron is widely thought to be related to the glueball
spectrum of the SU(3) gauge theory, in which case the leading glueball trajectory had
better have an intercept 0 around 1. We are currently performing a similar calculation
in D = 3 + 1 SU(3) gluodynamics. Preliminary results [35] indicate that a straight line
passing through the 2++ and 4++ states in a ChewFrautschi plot does not pass through
the 0++ (in contrast to what we found in D = 2 + 1) and has parameters 0 = 1.03(40) and
2 = 0.27(9); the latter translating to
0.22(8) GeV2 if we use
(0.44 GeV)2 .
These characteristics are broadly compatible with the well-known properties of the soft
Pomeron.
Acknowledgements
One of us (H.M.) thanks the Berrow Trust for financial support. The numerical
calculations were performed on PPARC and EPSRC funded workstations in Oxford
Theoretical Physics.
Appendix A. Regge theory in (2 + 1) dimensions

The theory of the S-matrix can be developed in an entirely analogous way to the (3 + 1)dimensional case, using the usual fundamental postulates [2]:
134
the S-matrix is Lorentz invariant;

the S-matrix is unitary;
the S-matrix is an analytic function of Lorentz invariants, with only those singularities
required by unitarity.
We denote by A the a +b c +d amplitude. In (3 +1) dimensions, this is a dimensionless
quantity, whereas in (2 + 1) dimensions, it has unit of mass. Therefore we define
1
T (s, t) = A(s, t).
s
The partial wave expansion in the s-channel reads

2s
.
T (s, t) = a0 (t) + 2
a (t)C 1 +
t
(A.1)
(A.2)
1
Here
2s
= cos
t
in the s-channel and
1+
(s-channel)
C (cos ) = cos
(A.3)
(A.4)
is a Chebyshev polynomial. The absence of a factor 2 in the first term originates from
the geometric difference between the spin 0 and the other partial waves. If we define a
parity axis along the axis of the collision, then while the left- and right-winding spin
wave functions add up to 2 cos , in the spin 0 sector the negative parity state does not
have a wave function (as is the case in (3 + 1) dimensions); therefore only the 0+ state
contributes as a partial wave. This separation of the spin 0 sector is necessary in order
to carry out the analytic continuation in through the SommerfeldWatson transform. In
(2 + 1) dimensions, the complications due to the signature = 1 also appears since the
wave functions of spin are associated with a phase (1) under a rotation by . Thus we
have to introduce two analytic functions a + (, t) and a (, t), so that

+ ei a ()(, t)
2s
C , 1 +
.
T (s, t) = a0 (t) + i d
(A.5)
2
sin
t
We now want to deform the contour as is done in (3 + 1) dimensions. However because

C (z) z||
(|z| ),
(A.6)
we cannot reduce the background integral by pushing it to Re = 1/2. Therefore, we

integrate along the imaginary axis and arrive at the following expression:

+ ei a ()(, t)
2s
C , 1 +
d
2
sin
t
&+i
T (s, t) = a0 (t) + i
&i

+ ein 2 Resn (t)
2s
C n (t), 1 +
.
2
sin n (t)
t
(A.7)
135
So unless a0 (t) and the background integral vanish, we obtain

),0)
T (s, t) s max((t
,
(A.8)
where (t)
is the pole with the largest real part (leading Regge pole). Using the optical
theorem at high energies
tot =
1
1
Im A(s, t = 0) = Im T (s, t = 0),
s
s
(A.9)
we obtain the prediction, for scattering driven by Regge-pole-exchange,
tot s (0)1/2
.
(A.10)
If all Regge trajectories have negative intercept, the background term prevails at high
energy. In the case of potential scattering, = 0 is an accumulation point of Regge poles
when t 0 [22]. It is for that reason that we kept the background integral along Re = &.
Appendix B. Potential scattering and bound states in the plane
The ansatz

(r) i
(r, ) =
e
r
=
(B.1)
plugged into the Schrdinger equation leads to the following radial equation for :
2 1

( 4 )
(r) +
(B.2)
+
V
(r)
(r) = E (r).
r2
Thus there is a trivial correspondence between scattering in 3 dimensions and 2 dimensions
via the substitution:
1
1
"=
(B.3)
"(" + 1) = 2 .
2
4
This effective shift in the angular momentum has important consequences. Regge
originally showed for a large class of potentials in 3d scattering that the partial wave
amplitudes are meromorphic in " in the region Re " > 1/2; this corresponds to the region
Re > 0 in 2d. It was already known in the sixties that at threshold E 0, there is an
accumulation of an infinite number of Regge poles around = 0.
The point = 0 We momentarily restore the ordinary units of quantum mechanics.
Because of the Heisenberg uncertainty principle pr h /2r, we have
h 2 2 1/4
h 2
+
+ V,
(B.4)
8mr 2 2m
r2
which at = 0 simply reduces to E V (r). Thus this exact cancellation between zeropoint quantum fluctuations implies that any attractive potential, however weak, will create
a bound state at = 0. Indeed, a heuristic calculation can be found in [36] showing that
E
136
the binding energy is a non-perturbative expression in the potential:

1
E
exp
.
V (r)r dr
(B.5)
Low-energy potential scattering It was shown in [22] that under very general conditions,
the s-wave amplitude vanishes logarithmically at threshold. This can be interpreted as a
branch point singularity in the complex plane:
d
=
k log (k 0).
a0
(B.6)
2 log k
2
2i
References
[1] A.B. Kaidalov, hep-ph/0103011.
[2] J.R. Forshaw, D.A. Ross, Quantum Chromodynamics and the Pomeron, in: Cambridge Lecture Notes in
Physics, Cambridge Univ. Press, Cambridge, 1997.
[3] P.V. Landshoff, hep-ph/0108156.
[4] G. t Hooft, Nucl. Phys. B 72 (1974) 461;
E. Witten, Nucl. Phys. B 160 (1979) 57;
S. Coleman, Erice Lectures, 1979;
A. Manohar, Les Houches Lectures, 1997, hep-ph/9802419.
[5] M.J. Teper, Phys. Rev. D 59 (1999) 014512, hep-lat/9804008.
[6] H.B. Meyer, M.J. Teper, Nucl. Phys. B 658 (2003) 113, hep-lat/0212026.
[7] B. Lucini, M. Teper, Phys. Rev. D 64 (2001) 105019, hep-lat/0107007.
[8] N. Isgur, J. Paton, Phys. Rev. D 31 (1985) 2910.
[9] T. Moretto, M. Teper, hep-lat/9312035.
[10] R.W. Johnson, M.J. Teper, Phys. Rev. D 66 (2002) 036006, hep-ph/0012287.
[11] S. Dalley, B. van de Sande, Phys. Rev. D 63 (2001) 076004, hep-lat/0010082.
[12] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, Int. J. Theor. Phys. 38 (1998) 1113, hep-th/9711200.
[13] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 505, hep-th/9803131.
[14] C. Csaki, H. Ooguri, Y. Oz, J. Terning, JHEP 9901 (1999) 017, hep-th/9806021;
N.R. Constable, R.C. Myer, JHEP 9910 (1999) 037, hep-th/9908175;
R.C. Brower, S.D. Mathur, C.-I. Tan, Nucl. Phys. B 587 (2000) 249276, hep-th/0003115.
[15] H.B. Meyer, JHEP 0301 (2003) 048, hep-lat/0209145.
[16] M. Luscher, P. Weisz, JHEP 0207 (2002) 049, hep-lat/0207003.
[17] D. Perkins, Introduction to High Energy Physics, AddisonWesley, Roading, MA, 1972.
[18] B. Lucini, M. Teper, JHEP 0106 (2001) 050, hep-lat/0103027.
[19] B. Lucini, M. Teper, U. Wenger, Phys. Lett. B 545 (2002) 197, hep-lat/0206029.
[20] L. Del Debbio, H. Panagopoulos, E. Vicari, JHEP 0208 (2002) 044, hep-th/0204125.
[21] S. Deldar, Phys. Rev. D 62 (2000) 034509, hep-lat/9911008;
G. Bali, Phys. Rev. D 62 (2000) 114503, hep-lat/0006022.
[22] K. Chadan, N.N. Khuri, A. Martin, T.T. Wu, Phys. Rev. D 58 (1998) 025014, hep-th/9805036.
[23] M. Li, C.I. Tan, Phys. Rev. D 50 (1994) 11401149, hep-th/9401134;
M. Li, C.I. Tan, Phys. Rev. D 51 (1995) 3287, hep-ph/9407299.
[24] F.E. Low, Phys. Rev. D 12 (1975) 163.
[25] A.H. Mueller, hep-ph/9911289.
[26] D.Y. Ivanov, R. Kirschner, E.M. Levin, L.N. Lipatov, L. Szymanowski, M. Wusthoff, Phys. Rev. D 58 (1998)
074010, hep-ph/9804443.
[27] E. Verlinde, H. Verlinde, hep-th/9302104.
[28] M. Albenese, et al., APE Collaboration, Phys. Lett. B 192 (1987) 163.
[29] N. Cabibbo, E. Marinari, Phys. Lett. B 119 (1982) 387.

[30] K. Fabricius, O. Haan, Phys. Lett. B 143 (1984) 459;
A.D. Kennedy, B.J. Pendleton, Phys. Lett. B 156 (1985) 393.
[31] S.L. Adler, Phys. Rev. D 23 (1981) 2901.
[32] B. Berg, A. Billoire, Nucl. Phys. B 221 (1983) 109;
M. Lscher, U. Wolff, Nucl. Phys. B 339 (1990) 222.
[33] M. Lscher, private communication.
[34] V. Shevchenko, Yu. Simonov, hep-ph/0104135.
[35] H.B. Meyer, Talk given at CERN, 24 April, 2003.
[36] L.D. Landau, E.M. Lifshitz, Quantum Mechanics, 3rd Edition, Pergamon, Elsford, NY, 1977.
137

Non-positive fermion determinants in lattice

supersymmetry
Joel Giedt
University of Toronto, 60 St. George St., Toronto, ON M5S 1A7, Canada
Received 8 May 2003; received in revised form 3 July 2003; accepted 15 July 2003
Abstract
We find that fermion determinants are not generally positive in a recent class of constructions with
explicit lattice supersymmetry. These involve an orbifold of supersymmetric matrix models, and have
as their target (continuum) theory (2, 2) 2-dimensional super-YangMills. The fermion determinant
is shown to be identically zero for all boson configurations due to the existence of a zeromode fermion
inherited from the mother theory. Once this eigenvalue is factored out, the fermion determinant
generically has arbitrary complex phase. We discuss the implications of this result for simulation of
the models.
PACS: 11.15.Ha; 11.30.Pb
1. Introductory remarks
Models with exact lattice supersymmetry have been discussed in the literature by
a few groups. For example, latticizations of super-YangMills [1,2], supersymmetric
quantum mechanics [3,4], the 2d WessZumino model [4,5], and direct constructions in
the spirit of the GinspargWilson relationas suggested by Lscher [6] and worked out
in noninteracting examples [7,8]have all been considered.1 In this article we will be
interested in the super-YangMills constructions that lead to a Euclidean lattice theory [1].
The method of building such models is based on deconstruction of extra dimensions [10,
11]. The corresponding interpretation in terms of the world-volume theory of D-branes
has led to the latticizations of 2d, 3d and 4d supersymmetric gauge theories. These lattice
E-mail address: giedt@physics.utoronto.ca (J. Giedt).
1 An approach very similar to [8] has been applied in [9], yielding slightly different expressions.
doi:10.1016/j.nuclphysb.2003.07.006
J. Giedt / Nuclear Physics B 668 (2003) 138150
139
constructions are all arrived at by orbifold projections of supersymmetric matrix models;

i.e., in each case we quotient a matrix model by some discrete symmetry group of the
theory. Degrees of freedom that are not invariant with respect to the combined action of the
orbifold generators are projected out.2 Thus, throughout this paper we will have occasion
to speak of orbifolded matrix models and nonorbifolded matrix models.
A major motivation for efforts to latticize supersymmetric models is that some
nonperturbative aspects of supersymmetric field theories are not accessible by the usual
techniques, such as holomorphy. One hope of a lattice supersymmetry program of
research is that it would lead to, e.g., simulations that would provide further data on
supersymmetric field theories, especially those that include super-YangMills.3 With exact
lattice supersymmetry the target (continuum) theory may be obtained in a more controlled
fashion. Indeed, in some cases it may be obtained without the need for fine-tuning [1,2].
In this enterprise, it is of great practical importance that the fermion determinant,
obtained by integrating over the fermion degrees of freedom in the partition function, be
positive. For let be the lattice bosons in the theory, and , the lattice fermions. We
obtain4

Z = [d d d] exp SB () M()

= [d] det M() exp SB ()

= [d] exp Seff () ,
(1)
Seff () = SB () ln det M().
(2)
A positive5 det M() for all allows us to unambiguously calculate in an equivalent

bosonic theory with action Seff (). Expectation values of operators O() may be obtained
using this action:

[d] O() exp[Seff ()]
O() =
(3)
.
[d] exp[Seff ()]
Fermionic correlators will simply involve M 1 () in the operator of interest. In the case
of positive det M(), the techniques of estimation by Monte Carlo simulation are robust;
otherwise the situation is murkier.
We now summarize the content of our work:
2 For a detailed discussion, we refer the reader to [1].
3 For a recent review of existing work on this broad topic, and a complete list of relevant references, see [12].
4 Here we assume ordinary finite-dimensional Berezin integration for the fermion coordinates; thus we avoid
subtleties of the Weyl determinant present in continuum theories [13,14]. The class of theories studied here result
from a 4d 0d dimensional reduction, where the Pfaffian one obtains has an equivalent determinant form [15].
5 Strictly speaking, we require only positive semi-definite det M(), provided det M() = 0 occurs only for
some subset of the boson configurations; this subset has measure zero. However, if det M() 0, i.e., for all
boson configurations, then Z is not well-defined. This latter situation is what occurs in the theory that we study
here.
140
In this paper we show that the lattice theory with (2, 2) 2d super-YangMills as
its target [1], obtained from orbifolded supersymmetric matrix models, possesses a
problematic fermion determinant. Due to a zeromode fermion, det M() 0, i.e., for
all boson configurations.
We then suggest how the zero eigenvalue can be factored out in a controlled way
in order to exhibit the determinant for the other fermions. For the orbifolded matrix
models, we carry out this factorization (numerically) in the special case of a U (2)
gauge theory, and show that it is robust. As an example, we present our results for a
2 2 lattice.
We further validate our method by applying it to nonorbifolded U (k) supersymmetric
matrix models, which also contain ever-present zeromode fermions, corresponding
to the U (1) gaugino in the decomposition U (k) U (1) SU(k). We correctly
reproduce the fermion determinant for the nonorbifolded SU(k) matrix models by our
factorization method. As an example, we present our results for U (5) U (1) SU(5).
Once the zeromode fermion has been factored out of the orbifolded matrix models
studied here, we find that the remaining product of eigenvalues is generically nonzero
with arbitrary complex phase.
We explain how this is not in conflict with results in nonorbifolded SU(k) supersymmetric matrix models. (For example, in an appendix of [16] it was shown that the
product of eigenvalues for the fermion matrix is positive semi-definite.) It is found that
the orbifold procedure lies at the heart of this matter.
We conclude with a discussion of the implications of our results for lattice simulations
of the latticized (2, 2) 2d super-YangMills theories.
2. Zeromode fermion
Here we will focus on the orbifolded supersymmetric matrix models that have as their
target theory (2, 2) 2d super-YangMills with U (k) gauge group. The mother theory is a
nonorbifolded U (kN 2 ) supersymmetric matrix model. The daughter theory is obtained
by orbifolding the mother theory by a ZN ZN symmetry group, leaving intact (among
2
other things) a U (k)N symmetry group that will become the gauge symmetry of the N N
lattice theory. The lattice theory is obtained by studying the daughter theory about a
particular boson configuration that is a stationary point of the action; i.e., it is a point in the
moduli space of the daughter theory.
The orbifolded matrix model discussed here has been described in detail in [1]; we refer
the reader there for further details. For our purpose it suffices to note that in the U (k) case
the theory contains6 bosons xm , ym that are k k complex matrices. The bosons xm , ym
may written in terms of a Hermitian basis

2
a
1k , T ,
T
(4)
xm = xm T ,
ym = ym T .
(T a ) = T a ,
k
6 Our notation is standard: m labels sites on a 2d square lattice, so that m Z Z, with Z the set of integers.
141
It is always possible to choose the T a such that

Tr(T T ) = 2 .
Furthermore, we define
(5)
2
t
k
and note that (underlining implies all permutations are to be taken):
Tr(T T T ) = 2
(6)
t 0 = .
(7)
One finds that the fermionic part of action is given by

2

n
SF = m , m Mm,n .
n
g2 k
(8)
Here m , m , n , n are the lattice fermions, with upper index corresponding to the basis
T introduced above. The fermion matrix Mm,n is given by (sum over implied in the
entries, , unit vectors):

t y

tm,n xm tm,n x m
m,n m+ + tm,n+ yn
Mm,n =
(9)
.
t x
tm,n ym tm,n ym
m,n m+ tm,n+ xn
Here we have introduced the compact notation
tm,n = m,n t .
(10)
The fermion zero mode is easily established. We consider fermions of the form
0
n

, n.
=
0
n
(11)
Then in this case

0

0
tm,n tm,n n =
tm,n tm,n
xm
xm
n
n

= xm
(m,n
n
By similar arguments

tm,n tm,n n = 0.
ym
m,n ) = 0.
(12)
(13)
Thus (11) is an eigenvector of M with eigenvalue zero. It follows that det M(x, y) 0, for
all configurations of bosons. This clearly poses a difficulty for defining Seff (x, y). We will
shortly address a method to factor out this zero eigenvalue, and study it in some detail for
the U (2) case. However, we first make a few remarks on the existence of the indentically
zero eigenvalue.
142
The zeromode fermion is nothing but the zero-momentum mode of the Fourier
transform of a given 0m :
1 0
00 =
(14)
.
N m m
In the formalism introduced in [1], 0m appears in the superfield

0
0m = 0m xm xm xm xm + ym ym ym ym + idm
.
(15)
0 is an auxiliary boson and is an odd (Grassmann) superspace coordinate. The

Here dm
zero-momentum part of this supermultiplet is just

1 0
0m = + i d,
d
d .
(16)
N m m
m
That is, the zeromode fermion is in a multiplet that contains just itself and an auxiliary
boson. It is a lattice version of a Fermi multiplet [17]. Thus there is no physical zeromode
boson that corresponds to the zeromode fermion.
The existence of the zeromode fermion can be understood in terms of the mother
theory; i.e., the nonorbifolded matrix model. There, the bosons x, y and their conjugates
are understood in terms of a vector boson v:

x y
,
v = vm m = v0 + iv = 2
(17)
y x
where vm are Hermitian matrices that are Lie algebra valued in U (kN 2 ):
(vm )i,j = vm
(T )i,j ,
i, j = 1, . . . , k,
, , , = 1, . . . , N.
(18)
2
Note that the indices of generators have been written in such a way that the U (k)N
subgroup has been manifestly factored out. One should think of one U (k) factor living
at each site. Similarly, the fermions of the mother theory are given by

,
=
i,j = (T )i,j ,
(19)
with a corresponding expression for . The fermion action in the mother theory takes the
form

1
SF = 2 Tr m [vm , ] .
(20)
g
Now note that the diagonal U (1)diag U (kN 2 ) fermions do not appear SF . That is,
(T )i,j 0 i,j
(21)
and 0 disappears because of the commutator. Since 0 is a two component fermion, it

gives two zeromode fermions of the mother theory, independent of the boson configuration
vm . The boson action takes the form

1
SB = 2 Tr [vm , vn ][vm , vn ] .
(22)
4g
143
Here again, the U (1)diag boson disappears from the action because of the commutators. It
is a zeromode for all boson configurations.
Most of the fields of the mother theory are projected out in the orbifold construction.
The projections depend on charges with respect to a U (1)r1 U (1)r2 global symmetry
group. As it turns out, the fermions are U (1)r1 U (1)r2 neutral. It follows that only the
diagonal parts with respect to the site indices (, ) survive in the orbifolded matrix
model (daughter theory):
(T )i,j (T )i,j .
(23)
But this includes the zeromode fermion of the mother theory. On the other hand, none
of the components of v are U (1)r1 U (1)r2 neutral. It follows that, in the orbifolded
matrix model, they are all off-diagonal7 with respect to the indices (, ). Hence the
zeromode bosons of the mother theory are projected out in the orbifolded theory. Similarly,
none of the fermions of the mother theory are U (1)r1 U (1)r2 neutral, so that the
second zeromode fermion 0 of the mother theory disappears in the projection.
3. Deformed U (k)
The zeromode eigenvalue of the orbifolded matrix model can be factored out as follows.
We deform the fermion matrix (9) according to
M M + %1Nf ,
(24)
where Nf is the dimensionality of the fermion matrix and % 1 is a deformation parameter

that we will eventually take to zero. We factor out the zero mode through the definition
M(0)
= lim M(%),
M(%)
= % 1/Nf M + %1Nf ,
%0

det M(0)
= lim % 1 det M + %1Nf .
(25)
%0
If this deformation is added to the action, it explicitly breaks the exact lattice
supersymmetry and gauge invariance. This infrared regulator could be removed in the
continuum limit, say, by taking %a N 1 . Noting that L = Na is the physical size of
the lattice, the equivalent requirement is that % L1 be maintained as a 0, for fixed
L. Thus in the thermodynamic limit (L ), the deformation is removed. The parameter
% is a soft infrared regulating mass, and is quite analogous to the soft mass introduced
by Cohen et al. [1] in their Eq. (1.2) to control the bosonic zeromode of the theory. In the
same way that a finite does not modify the final results of the renormalization analysis
of Section 5 of [1], our % does not modify the result of the quantum continuum limit.
The essence of the argument is that we have introduced a vertex that will be proportional
to the dimensionless quantity g22 %a 3 g22 a 3 /L, where g2 is the 2d coupling constant.
Such contributions to the operator coefficients CO in Eq. (5.2) of [1] vanish in the
7 In the lattice theory, these are interpreted as link fields beginning at the site labeled by (, ) and ending at
the site labeled by ( , ).
144
thermodynamic limit and because the target theory is super-renormalizable, we are assured
that the perturbative power counting arguments are reliable and the correct continuum limit
is obtained.
Of course it remains to study the convergence of det M(%)

det M(0).
Indeed, for
2
N
U (2) lattice theory (i.e., the orbifolded matrix model with U (2) symmetry) we have
performed this analysis numerically for a large number of boson configurations drawn
randomly from a Gaussian distribution, as will be detailed below. We find that the
convergence is rapid and that a reliable estimate for det M(0)

can be obtained in this way.
Furthermore, we find that for % 1, the phase of det M(%) quickly converges to a constant
value, and that it is uniformly distributed throughout the interval (, ] for the random
Gaussian boson variables.
As a check of our method, we have studied also the analogous deformation of the
nonorbifolded U (k) supersymmetric matrix models, where the two zeromode fermions
in 0 are present. Indeed, as will be discussed below, we find that

lim % 2 det MU (k) + %1Nf = det MSU(k) .

(26)
%0
That is, we obtain the determinant of the nonorbifolded SU(k) supersymmetric matrix
model, which has the zeromode fermions factored out. With appropriate conventions,
det MSU(k) is positive semi-definite.8
4. Deformed U (2)
2
Here we specialize to the orbifolded matrix model with U (2)N symmetry, which
becomes the U (2) gauge invariance of an N N lattice theory. In this case we take

0
a a
0
a a
xm = xm
T 12 , a ,
(27)
+ xm
,
ym = ym
+ ym
.
Then the fermion matrix is given in (9), where in the present case
t 000 = 1,
t a00 = 0,
t ab0 = ab ,
t abc = i% abc .
(28)
The lattice theory is obtained by expansion about a point in moduli space:

1
0
xm
= + ,
a 2
1
0
ym
= + ,
a 2
(29)
where represent the quantum fluctuations. For this reason, in our study of
det M(%)
we
0
0
scan over a Gaussian distribution where xm , ym have a nonzero mean 1/a 2 1. The
remainder of the bosons are drawn with mean zero. All bosons are taken from distributions
with unit variance.
In Figs. 1 and 2 we display ln | det M(%)|

and arg det M(%)
versus % for the case of
N = 2, the smallest lattice possible, of size 2 2. Each line corresponds to a different
8 For some conventions on the SU(k) generators and the overall coefficient in S of Eq. (20) it is possible that
F
a constant phase (i.e., independent of boson configurations) may be present. However, this factors out of (3) and
is of no concern.
145
Fig. 1. ln | det M(%)|

versus % for a sequence of random draws. These results are for the U (2) lattice theory, with
2 2 lattice.
Fig. 2. = arg det M(%)

versus % for a sequence of random draws. Note that crossing over the boundary of the
domain (, ] to an equivalent point within that domain is indicated by the nearly vertical lines. These results
are for the U (2) lattice theory, with 2 2 lattice.
random draw. It can be seen that the convergence is rapid and reliable. As a check, we have
computed the eigenvalues of the undeformed matrix M, using the math package M APLE,
for the same set of random boson configurations. We find that in each case the product
of nonzero eigenvalues agrees with det M(0)

in magnitude and phase to within at least 5
significant digits. This conclusively demonstrates that the complex determinant is not an
artifact of the deformation, but is a property of the nonzero eigenvalues of the undeformed
matrix M.
146
Fig. 3. Frequency distribution F () for = arg det M(0),

for 105 random (Gaussian) draws, binned into intervals
of /100. The distribution of is seen to be nearly uniform. These results are for the U (2) lattice theory, with
2 2 lattice.
For a set of 105 draws on the bosons of this 2 2 lattice, we have extrapolated to % 0
and binned arg det M(%)

over its range, with bins of size /100. In Fig. 3 we show the
frequency for each bin, as a fraction of the total number of draws. In the extrapolation, we
decreased % by powers of 10 until the change in both ln | det M(%)|

and arg det M(%)
were
4
both less than 10 per decade. This was done for each draw to get a reliable estimate for
det M(0).
In each of the 105 draws the 104 per decade criterion was reached well before
% = 1010 .
To summarize, once the zeromode eigenvalue is factored out, the product of the nonzero
eigenvalues has arbitrary phase. Consequently an ambiguity exists in defining Seff (x, y) for
the orbifolded matrix model theory. We are presently exploring whether or not this can be
overcome for the purposes of simulation. At present, however, all we can say is that this
difficulty is rather troubling.
5. Comparison to nonorbifolded U (k) and SU(k) matrix models

As mentioned above, the nonorbifolded supersymmetric U (k) matrix models contain
two ever-present zeromode fermions in the 0 that appears in (21). Then det MU (k) 0.
However, it is easy enough to just work with the nonorbifolded SU(k) matrix model,
so that 0 , 0 are never in the theory to begin with. Then with appropriate conventions
det MSU(k) 0. A proof of this result has been given in an appendix of [16].
As a test of our method, we have verified (26) numerically, for a sequence of random
that appear in (20). In Fig. 4 we show the quantity
Gaussian draws on the bosons vm

2

(%) = ln % det MU (k) + %1Nf lndet MSU(k)
(30)
as a function of % for the case of k = 5. Indeed it can be seen that the convergence as % 0
is quite rapid. We find that |(103 )| 104 and |(104 )| 109 as a rule. In Fig. 5
147
Fig. 4. (%) versus % for a sequence of random draws in the nonorbifolded U (5) matrix model.
Fig. 5. versus % for sequence of random draws in the nonorbifolded U (5) matrix model.
we show the quantity

(%) = arg % 2 det MU (k) + %1Nf
(31)
as a function of %, again for k = 5. The value for SU(5) in our conventions is = 0. It can
be seen that a rapid convergence to this value is obtained.
One might wonder how the generic phases of Fig. 3 are obtained in the orbifolded
matrix models, given the positivity of the fermion determinant in the nonorbifolded SU(k)
matrix models. Firstly, one should note that the proof given in [16] shows only that for
each eigenvalue e of MSU(k) there exists also an eigenvalue e , and that these always come
148
in pairs. The proof relies essentially on the relation

2 v 2 = 2 v0 + iv 2 = (v ) .
(32)
On the other hand, the ZN ZN orbifold action on the mother theory bosons is generated
by
v (T )i,j ei
v (T )i,j ei
3 /N
3 /N
v ei
v ei
3 /N
3 /N
(T )i,j 1
,
(T )i,j 1
,
(33)
where = diag(, 2 , . . . , N ), = exp(2i/N). Thus because of the exp(i 3 /N)

factors, the orbifold projection does not commute with the operations of the proof given in
[16]; i.e., Eq. (32). The fermion matrix of the projected theoryi.e., the orbifolded matrix
modellacks many of the eigenvalues of the mother theory; it should come as no surprise
that not all eigenvalues are removed in pairs (e, e ). After all, we already know that only
one of the zero eigenvalues is removed from the U (kN 2 ) mother theory. For this reason it
is not at all contradictory that the projected theory has a product of nonzero eigenvalues
that is not positive, nor real.
6. Discussion
The existence of a zeromode fermion in the orbifolded matrix models has been reliably
handled by our deformation method, and the approach is easily implemented numerically.
Analytic methods in nonorbifolded supersymmetric matrix models have also involved
deformations of the theory to render quantities of interest well-defined [18]; however,
these involve the introduction of auxiliary fields, which are expensive to implement in
a simulation. For this reason, we prefer the method described here. In spite of the complex
fermion determinant in the orbifolded matrix models, our approach allows for a detailed
study of the partition function by Monte Carlo methods, analogous to what has been
performed in [15] for nonorbifolded supersymmetric matrix models. It would be interesting
to see how expectations based on continuum results might be realized in the present
context. For example, in [14] it was shown that 4d N = 1 pure super-YangMills has a
positive fermion determinant. Since the (2, 2) 2d target theory studied here is a dimensional
reduction of this model, it is rather surprising that we find arbitrary complex phase.
However, it remains to be seen whether or not the lattice action described here possesses
reflection positivity, or whether this is only obtained in the continuum limit. Study of this
issue is in progress.
We suspect that the difficulties faced here may be, broadly speaking, related to those
faced in defining the phase of the fermion measure in the attempts to realize chiral gauge
theories on the lattice with an exact chiral gauge symmetry [1921]. In the present context,
we have an exact chiral fermionic symmetry: lattice supersymmetry acts on the m , m but
not on the m , m that appear in (8). A resolution of one problem may lead to answers for
the other.
In our opinion, the deconstruction approach to lattice supersymmetry remains an
exciting topic, whatever difficulties may face attempts to simulate the theory. For example,
149
the relatively simple systems described here provide another context in which the
difficulties9 that plague the complex Langevin approach [24,25] and other complex action
techniques might be studied.
We are presently exploring whether or not the complex phase can be overcome for the
purposes of simulation. A typical approach would be to compute averages of an operator
O from the reweighting identity:
O =
Oei p.q.
ei p.q.
(34)
Here, = arg det M(0),

as above, and p.q. indicates phase-quenching: expectation values
are computed with the replacement det M(0)

| det M(0)|.
Thus the effective bosonic
action

Seff
(35)
= SB lndet M(0)
is used to generate the phase-quenched ensemble by standard Monte Carlo techniques.
However, it is well known that this suffers efficiency problems: the number of configurations required to get an accurate estimate for, say, exp(i)p.q. , grows like exp(2F Nf2 ).
Here, 2F is the difference in free energy densities between the full ensemble and the
phase-quenched ensemble. Recall that Nf is the dimensionality of the fermion matrix. On
the other hand, if the phase-quenched distribution is highly concentrated near one value of
, the phase essentially factors out of the partition function and efficient simulations can
be done using the phase-quenched ensemble. Thus it is of interest to study the distribution of in the phase-quenched ensemble rather than the Gaussian distribution used here.
Research in this direction is currently in progress and we hope to report on it soon.
Acknowledgements
The author would like to thank Erich Poppitz for numerous helpful discussions and suggestions. Thanks are also owed to Wolfgang Bietenholz for enlightening communications,
and to the referee who made a number of useful comments. This work was supported by
the National Science and Engineering Research Council of Canada.
References
[1] A.G. Cohen, D.B. Kaplan, E. Katz, M. Unsal, Supersymmetry on a Euclidean spacetime lattice. I: A target
theory with four supercharges, hep-lat/0302017.
[2] D.B. Kaplan, E. Katz, M. Unsal, Supersymmetry on a spatial lattice, hep-lat/0206019.
[3] S. Catterall, E. Gregory, A lattice path integral for supersymmetric quantum mechanics, Phys. Lett. B 487
(2000) 349, hep-lat/0006013.
[4] S. Catterall, S. Karamov, A two-dimensional lattice model with exact supersymmetry, Nucl. Phys. B (Proc.
Suppl.) 106 (2002) 935, hep-lat/0110071.
9 See, for example, [22,23] and references therein.
150
[5] S. Catterall, S. Karamov, Exact lattice supersymmetry: the two-dimensional N = 2 WessZumino model,
Phys. Rev. D 65 (2002) 094501, hep-lat/0108024.
[6] M. Lscher, Exact chiral symmetry on the lattice and the GinspargWilson relation, Phys. Lett. B 428 (1998)
342, hep-lat/9802011.
[7] T. Aoyama, Y. Kikukawa, Overlap formula for the chiral multiplet, Phys. Rev. D 59 (1999) 054507, heplat/9803016.
[8] W. Bietenholz, Exact supersymmetry on the lattice, Mod. Phys. Lett. A 14 (1999) 51, hep-lat/9807010.
[9] H. So, N. Ukita, GinspargWilson relation and lattice supersymmetry, Phys. Lett. B 457 (1999) 314, heplat/9812002.
[10] N. Arkani-Hamed, A.G. Cohen, H. Georgi, (De)constructing dimensions, Phys. Rev. Lett. 86 (2001) 4757,
hep-th/0104005.
[11] C.T. Hill, S. Pokorski, J. Wang, Gauge invariant effective Lagrangian for KaluzaKlein modes, Phys. Rev.
D 64 (2001) 105005, hep-th/0104035.
[12] A. Feo, Supersymmetry on the lattice, hep-lat/0210015.
[13] L. Alvarez-Gaume, E. Witten, Gravitational anomalies, Nucl. Phys. B 234 (1984) 269.
[14] S.D. Hsu, Gaugino determinant in supersymmetric YangMills theory, Mod. Phys. Lett. A 13 (1998) 673,
hep-th/9704149.
[15] W. Krauth, H. Nicolai, M. Staudacher, Monte Carlo approach to M-theory, Phys. Lett. B 431 (1998) 31,
hep-th/9803117.
[16] J. Ambjorn, K.N. Anagnostopoulos, W. Bietenholz, T. Hotta, J. Nishimura, Large N dynamics of
dimensionally reduced 4D SU(N ) super-YangMills theory, JHEP 0007 (2000) 013, hep-th/0003208.
[17] E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hep-th/9301042.
[18] G.W. Moore, N. Nekrasov, S. Shatashvili, D-particle bound states and generalized instantons, Commun.
Math. Phys. 209 (2000) 77, hep-th/9803265.
[19] M. Lscher, Abelian chiral gauge theories on the lattice with exact gauge invariance, Nucl. Phys. B 549
(1999) 295, hep-lat/9811032.
[20] M. Lscher, Weyl fermions on the lattice and the non-abelian gauge anomaly, Nucl. Phys. B 568 (2000)
162, hep-lat/9904009.
[21] M. Lscher, Chiral gauge theories revisited, hep-th/0102028.
[22] K. Fujimura, K. Okano, L. Schulke, K. Yamagishi, B. Zheng, On the segregation phenomenon in complex
Langevin simulation, Nucl. Phys. B 424 (1994) 675, hep-th/9311174.
[23] H. Gausterer, S. Lee, The mechanism of complex Langevin simulations, hep-lat/9211050.
[24] G. Parisi, On complex probabilities, Phys. Lett. B 131 (1983) 393.
[25] F. Karsch, H.W. Wyld, Complex Langevin simulation of the SU(3) spin model with nonzero chemical
potential, Phys. Rev. Lett. 55 (1985) 2242.

Supersymmetric models for gauge inflation

R. Hofmann, F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze 1
Institut fr Theoretische Physik, Universitt Heidelberg, Philosophenweg 16, 69120 Heidelberg, Germany
Received 2 June 2003; accepted 27 June 2003
Abstract
We present possible realizations of gauge inflation arising from a 5D N = 1 supersymmetric
U (1) model, where the extra dimension is compactified on a circle. A one-loop inflaton effective 4D
potential is generated, with the inflaton being a Wilson-line field. It relies on a SUSY breaking. We
first consider SUSY breaking to occur spontaneously within a no-scale model by a non-zero F -term
of the radion superfield which transmits SUSY breaking into the visible sector. As an alternative,
we study D-term SUSY breaking originating directly from the 5D gauge supermultiplet. Together
with the usual KK resummation method, we present a calculation with the world-line formalism.
The latter allows one to get the resulting effective potential directly as a sum over all winding modes.
For both presented scenarios, the generated effective potentials have suitable forms for realizing
successful inflation, i.e., are flat enough and give the needed number of e-foldings. In addition, there
is a natural way to get strongly suppressed values for the potentials, which then could be associated
with dark energy/quintessence.
2003 Published by Elsevier B.V.
PACS: 11.15.-q; 12.60.Jv
1. Introduction
Inflation is an efficient way to solve the cosmological flatness, horizon, and monopole
problem [1]. It explains naturally why the visible part of the universe appears to be isotropic
on large scales. Field theoretic real-time models of inflation can explain the seeding of
structure formation. The experimental data on the spectral properties of primordial density
E-mail addresses: r.hofmann@thphys.uni-heidelberg.de (R. Hofmann),
f.paccetti@thphys.uni-heidelberg.de (F. Paccetti Correia), m.g.schmidt@thphys.uni-heidelberg.de
(M.G. Schmidt), z.tavartkiladze@thphys.uni-heidelberg.de (Z. Tavartkiladze).
1 On leave of absence from Institute of Physics, GAS, Tbilisi 380077, Georgia.
0550-3213/$ see front matter 2003 Published by Elsevier B.V.
doi:10.1016/S0550-3213(03)00574-1
152
R. Hofmann et al. / Nuclear Physics B 668 (2003) 151166
perturbations do not lead to a unique inflation model in field theory. If in a given field theory
model the scale of inflation is close to the 4D Planck scale MP the slow-roll requirements
linked to the models tree-level potential generically are violated by the appearance of gravitationally induced operators. Even for low-scale inflation the coupling of the inflaton to
other fields induces radiative corrections to the inflaton potential which are hard to control.
This problem is much milder if a symmetry protects the potential. For example, spontaneously broken, rigid supersymmetry (SUSY) yields flat potentials [2] for inflaton fields if
the scale of SUSY breaking is much lower than MP . Another possibility for the generation
of radiatively stable, flat potentials is the spontaneous breakdown of a global symmetry at
scale f . Very recently, this possibility was used to construct Little Higgs theories [3]. Furthermore, spontaneously broken global symmetries may underly the curvaton scenario for
the non adiabatic generation of curvature perturbations after inflation [4]. Coming back to
inflation itself, it is well known that a small amount of explicit symmetry breaking on top
of the spontaneous breakdown opens up the possibility that (quasi) de Sitter-cosmology
can be driven by the associated pseudo NambuGoldstone (PNG) fields [5]. Slow roll of
these fields is ensured if MP /f 1. However, a symmetry breaking above MP is probably
beyond a field theoretical treatment. This significantly reduces the appeal of the 4D PNG
model. This problem was recently pointed out in [6], and a resolution in terms of gauge
symmetry combined with the assumption of extra, compactified dimensions was proposed.
In this context the possible generation of a varying spectral index of the primordial, adiabatic scalar perturbations was investigated in Ref. [7]. WMAP data [8] indicate a tilted
spectrum.
A 4D effective potential for the Wilson-line phase
2R
dx 5 A5
(1)
(R denotes the compactification radius) arises by means of the Hosotani mechanism [9]
if charged states are present in a theory. The potential is protected from local quantum
gravity corrections due to 5D gauge invariance. Moreover, the scale f , which enters the
slow-roll condition MP /f 1, is naturally larger than MP if the effective 4D gauge
theory is weakly coupled. A variant of this model uses a mass M R 1 for the charged
bulk fluctuations which exponentially suppresses the value of the potential [6] making it a
suitable candidate [10] for dark energy, i.e., quintessence [11]. To derive the effective 4D
potential for one usually appeals to KaluzaKlein (KK) regularization, i.e., a Poisson
resummation of the KK spectrum [12]. This recipe avoids the use of a cut-off which would
destroy some of the symmetries of the underlying higher-dimensional theory. Recently,
KK regularization was intensively disputed, and it was pointed out that SUSY and gauge
symmetry in field theory help to make KK regularization consistent but that ultimately it
is string theory that provides a physical KK completion in the ultraviolet [13]. It may be
helpful in this context to point out that in a world-line formulation of quantum field theory
[14] in 5D spacetime no Poisson resummation of the KK tower is needed, since directly a
153
sum over winding modes is introduced in the action [23]

4
d x dx5
0
T

dT
(xM )2
5
2
+ iq x 5A5 + M
,
D y exp d
T
4
k=
with xM ( ) = 2Rk M5 + yM ( ) + xM ,
T
T
d yM ( ) = 0,
(2)
(see Appendix B for details). In the non-SUSY case this approach exhibits a 5D UV
divergence from the zero-winding sector in the one-loop effective potential which is cut-off
by restricting the proper-time integration to T 2 .
In a SUSY setting, inflation can only occur if SUSY is broken somehow. The purpose
of the present paper is to investigate how in the simple situations of 5D N = 1 SUSY U (1)
gauge dynamics SUSY breaking translate into a potential for . We assume the extradimensional coordinate y to describe a circle. Our results can be used as guidelines for the
more realistic cases of non-Abelian gauge symmetry and/or a larger number of (orbifolded)
extra dimensions. In Section 2 we introduce our model. We use the formulation of [15] with
a chiral radion field T and consider spontaneous SUSY breaking by a no-scale sector such
that FT = 0. We also consider D-term SUSY breaking by assuming that the neutral scalar
component of the chiral superfield in the 5D gauge supermultiplet obtains a profile along
the extra dimension. The effect on the spectrum of the KK modes is investigated for both
cases. In Sections 3 and 4 we compute the effective one-loop potentials for . In Section 5
we present cosmological applications of the models. Section 6 gives a discussion and an
outlook. Appendix A is devoted to an analysis of the KK spectra. It also contains technical
details concerning the computation of the effective potential. In Appendix B a calculation
of the effective potential is presented which uses the world-line method.
2. The model
2.1. Set-up of 5D dynamics
Consider 5D N = 1 SUSY U (1) gauge theory formulation [16] with the fifth spatial and
flat dimension y compactified on a circle, 0 y 2R. We stick to the usual convention
that 4D coordinates are labeled by Greek indices ( = 0, . . . , 3). 5D N = 1 supersymmetry
is equivalent to 4D N = 2 SUSY. In terms of N = 1 supermultiplets, the 5D gauge
supermultiplet is VN =2 = (V, ), where
1
+ i 2 1 i 2 1 + 2 2 D
V = A
2
is a 4D vector superfield, while is a chiral superfield
1
= ( + iA5 ) + 2 2 + 2 F .
2
(3)
(4)
154
The 5D Lagrangian LV of the pure gauge sector reads [15,16]

1
2 5 V +
2 5 V (5 V)2
LV = 2 d 4
g

1
d 2 W W + h.c. ,
+ 2
4g
(5)
where

W = i1 + D i A + 2 ,

1
= ,
A = A A ,
4

W W = D 2 + .
(6)
In addition, there are two chiral superfields
= + 2 + 2 F ,
= + 2 + 2 F
(7)
with respective charges q and q. These superfields constitute the 5D hypermultiplet

The 5D Lagrangian L for the matter reads
N =2 = (, ).

2qV +
L = d 4 + e2qV + e

2
+
d M + 5 2 q + h.c. ,
(8)
where M denotes a real SUSY bulk mass. Each of the Lagrangians (5), (8) is invariant
under 5D U (1) gauge transformations

1
1
V V + + + ,
+ 5 ,
2
2
q
q
e ,
(9)
e
.
Also there is invariance under two 4D N = 1 SUSY transformations. These two
supersymmetries are related by an SU(2)R symmetry. The bosonic part of LV + L ,
involving 4D scalar and auxiliary components, reads
(LV + L )B

1
1
1
= 2 D 2 2 5 D + 2 |F |2 + |F |2 + |F |2 qD ||2 ||
2
2g
g
g

+ F M + 5 2 q( + iA5 ) + M + 5 q( + iA5) F
2 q F
+ h.c. .
(10)
Eliminating F - and D-terms from (10), the part V in (10), which does not contain 4D
derivatives, reads
V=
2
1
g2 2 2
2
q || ||
(
)
+
2 + 2g 2 q 2 ||2 ||
2
5
2
2g
2
2
2

+ (M q)2 ||2 + ||
2 + (5 iqA5 ) + (5 + iqA5 ) .
(11)
155
If we assume the VEV A5 to be constant in y, the Wilson-line phase generated by a line

integration along the entire extra dimension reads
= 2RA5 .
(12)
In order to generate a 4D effective one-loop potential for the field by the Hosotani
mechanism [9] SUSY must be broken in some way. We distinguish two cases of SUSY
breaking in the following: (i) spontaneous breaking by the dynamics of a no-scale sector,
(ii) D-term breaking by a non-vanishing (initial) value of the VEV of s scalar component
in (4).
We will discuss the physics being potentially responsible for situation (ii) below. As
for case (i) some technical remarks are in order. A superspace formulation of the theory
LV + L involving a chiral radion superfield
T = R + iB5 + R5 + 2 FT ,
(13)
+T
)/(2R) for
has been proposed in Ref. [15]. The field T appears in terms of a factor (T
the first term in (8), and in terms of a factor 2R(T + T )1 and a factor T /R for the first
and second term, respectively, in (5). Upon and integration, the elimination of the auxiliary fields by their equations of motion this formulation yields the usual component-field
5D Lagrangian LV +L . The radion field T is of the no-scale type [15] and has a flat potential in the T direction in string-derived [18] supergravity models [17]. The superfunction
G = 3 ln K + ln |g|2 , K denoting the Khler potential and g the superpotential, in such
no-scale [17] (non-minimal [19] or minimal SU(n, 1) [20]) models can be written in
the form G = 3 ln(T + T |i |2 ) + ln |g3 (i , T )|2 where i denote chiral fields, g3 is
a superpotential, and all fields are in Planck units. G is invariant under particular Khler
transformations being related to dilatations. On free-level scale models have R = R as
a sliding scale in a flat potential, and the scale of SUSY breaking FT FT is also not
fixed. We assume R to be stabilized by some additional dynamics, see, for example, [21].
With Ref. [15] we will discuss in the next section how FT = 0 affects the spectrum of the
charged KK modes whose fluctuations generate a 4D one-loop effective potential for .2
2.2. KK decomposition and SUSY breaking
Since we compactify on a circle the usual consideration of reflection parity for orbifold
compactifications does not apply. The charged scalar fields and have the following KK
decompositions
+

ny
1
(x, y) =
(n) (x)ei R ,
2R n=
+

ny
1
(n) (x)ei R ,
(x,
y) =
2R n=
(14)
2 A fixed VEV F = 0, which arises due to a radiative stabilization in 5D supergravity [25] was mimicked by
T
a flat-space no-scale sector in [15]. In this model a scale W enters the superpotential for a conformal compensator
chiral superfield and determines the VEV FT , stabilization is therefore assumed in this effective description.
156
while the decomposition of the real field is

(x, y) =
(n)
(x) sin
n=1
ny (n)
ny
+ (x) cos .
+
R
R
(15)
n=0
A similar decomposition as in (14), (15) holds for the charged fermions , and the
field A5 , respectively. In case (i) we consider spontaneous SUSY breaking. To assume
(n)
(n)
in (15) (x) 0 and + (x) = n0 f (x) with some slowly varying function f would
(n)
simply shift the SUSY mass M by a finite amount. We set + (x) 0 when considering
case (i). We are interested in the generation of a 4D effective one-loop potential for .
Since the gauginos 1,2 do not couple to A5 they are irrelevant for the Hosotani mechanism.
As in (12) we assume A5 to be constant in yonly its lowest KK mode contributes. The
,
shift of the masses mn of the charged, bosonic KK modes can be calculated after a
diagonalization of the (2 2)-mass matrix for n and n at the nth KK-level [15]:
2
1/2
n
2
mn = mn =
+M
R

1/2
1
FT 2
,,
2
mn
=
+M
( < n < ).
(16)
n
2
R2
The KK masses of charged fermions,
2
1/2
n
2
mn = mn =
+M
,
R
(17)
are not shifted. This breaking of SUSY occurs through the non-zero FT of the radion
supermultiplet (13). Therefore, the SUSY transformation for the R5 state (the fifth
component of right-handed gravitino) is R5 FT ( is a Grassmann variable of the
SUSY transformation). Thus, R5 emerges as a goldstino, becomes the longitudinal mode
of 4D gravitino and gives mass to it, see [25].
Let us now consider case (ii). Here SUSY breaking proceeds by a non-vanishing
scalar VEV of 5 and therefore, the latter appears as an order parameter for SUSY
breaking. One can see from (10) that, D = 5 (since =
= 0) and 5 = 0
indicates complete SUSY breaking through the D-term. Recalling the 4D N = 1 SUSY
transformation for the gaugino
1 = iD + A ,
(18)
we see that in this case the 1 gaugino appears as a goldstino. For simplicity we assume
in (15)
(n)
(x) n1 V ,
(n)
+ (x) 0.
(19)
This induces next and next-to-next neighbor interactions in the fluctuating KK tower,
see Appendix A. Correspondingly, higher KK modes of would cause non-vanishing
elements of the KK mass matrix further away from the diagonal. The field expectation
157
value V is not dynamical.3 For the one-loop potential V eff () to be relevant for inflation,
we assume here that some external dynamics keeps V (x) constant and away from zero
and that the overall vacuum energy vanishes at tree-level. This would happen, for example,
if a SUSY-breaking potential P (V ) was added to the effective 4D dynamics such that
the sum of the tree-level potential R 2 V 2 and P (V ) has a nontrivial minimum at zero
energy. Since only very particular potential couplings are allowed by 5D SUSY, one can
induce P (V ) on the 4D level. This would require an orbifold scenario with appropriate
brane potential couplings. The calculation of the effective potential for a Wilson-line phase
within an orbifold construction is beyond the scope of this paper [22].
3. V eff () from F -term SUSY breaking
At tree-level the Wilson-line phase would not have a potential in 4D due to 5D
gauge invariance. For a compactified extra dimension this is no longer true if the quantum
fluctuations of charged bulk matter are integrated out [9]. We assume that w.r.t. 4D-x A5
varies on scales larger than the compactification scale R 1 such that the lowest order in a
derivative expansion of the quantum effective potential in 4D is a good approximation.
When calculating the one-loop effective potential in 4D for the presence of the A5
coupling to (n) , (n) and , in (10) and the fact that FT = 0 effectively modify the
KK masses as follows

2
1/2
1
q 1
,,
2
FT
mn
=
+M
,
n+
R2
2
2

1/2

q 2
1
2
mn = mn =
(20)
n
+
+
M
.
2
R2
After a Schwinger parametrization of tr log we have
1
V () =
STr
(4)2
eff
dt
exp(Mt)
t3
1
=
(4)2 n=

dt
exp M 2 t
3
t

q FT 2
t
q FT 2
t
+
+ exp 2 n +
exp 2 n +
R
2
2
R
2
2

2
t
q
2 exp 2 n +
(21)
,
R
2

3 On the one hand, slow-roll for V due to 4D tree-level dynamics (V 2 potential) would require V M which
P
is precisely what we would like to avoid. On the other hand, loop effects, which qualitatively change the tree-level
potential for V such that V is slowly rolling at V < MP , are not under control in a perturbative treatment of the
supertrace appearing in the computation of the Hosotani potential for .
158
where STr and M denote the supertrace and the (diagonal) supermass matrix for the
fluctuating bosonic and fermionic fields, respectively. As usual,4 we perform a Poisson
resummation,
+
2 (n+ )2
+
22k2 2ik
e e
,
n=
(22)
k=
and afterwards carry out the t integral in (21). The result of the integration is 2M 5 K5/2
(2RMk) where K5/2 denotes a modified Bessel function. Collecting everything, we
finally obtain
3 1
V () =
exp(2kRM) 1 cos(kFT )
16 6 R 4
k=1

1
cos(qk)
2
1 + 2kRM + (2kRM) ,
3
k5
eff
(23)
in complete agreement with a world-line calculation (Appendix B). Notice that the
divergent bosonic and fermionic contributions at k = 0 cancel exactly because of SUSY
the usual fine-tuning of the cosmological constant is not needed. For FT = 0 and small
the effective potential is positive, and thus it drives de Sitter expansion if is rolling
sufficiently slowly. In the limit of unbroken SUSY, FT 0, the effective one-loop
potential for vanishes.
4. V eff () from D-term SUSY breaking

Let us now investigate case (ii), namely the effect of D-term SUSY breaking according
to (19). The one-loop effective potential in this case reads
V
eff
1
=
STr
16 2
dt (D+B)t
e
,
t3
(24)
where STr also indicates the trace in KK space. The matrices D and B denote the
diagonal and the off-diagonal parts of the KK mass-squared matrices. We will compute
l
l
l
V eff in quadratic order in B. The treatment of the term (1)
l! t STr(D + B) (expansion of
the exponential in (24)) is presented in Appendix A. Under consideration of (A.11), the
quadratic order in B reads

+ +
l1
(D+B)t
qV 2 (1)l l Dnl1 Dn1
t
STr e
= 2
2R n=
(l 1)! Dn Dn1
l=2
4 See, however, Appendix B where we sketch a calculation of V eff () using the world-line formalism.

= 2
qV
2R
qV
= 2
2R
2
t
+

eDn t eDn1 t
Dn Dn1
n=
2
+ 1

n=
dy eDn (1y)t Dn1yt .
159
(25)
We now perform a Poisson resummation (22) in (25). Recalling that

q 2
1
q 2V 2
,
Dn = 2 n
+ M2 +
2
2
R
(26)
we obtain
V
eff
1

+

qV 2
1
= + 5/2
R dy
eikq2iky Ik
2R
8
0
(27)
k=
with

Ik =
dt
t 3/2

q 2V 2
2 R2 2
2
2
+
R
k
.
exp M +
y(1 y) t
2
t
(28)
Performing the t-integration for |k| 1, we get

1/2

1
2
2 2
2
exp 2R|k| M + q V /2 + R y(1 y)
Ik =
.
R|k|
(29)
The case k = 0 leads to a -independent, linear UV divergence which renormalizes the

4D gauge coupling (a discussion of this is presented at the end of the paper). We omit this
independent part when studying inflation. Thus the final expression for V eff reads
V eff = +
+

1 qV 2 cos(qk)
Ck ,
k
4 2 2R
(30)
k=1
where
1
Ck =

1/2
.
dy cos(2ky) exp 2Rk M 2 + q 2 V 2 /2 + R 2 y(1 y)
(31)
The same result is obtained by a (very fast!) world-line method calculation in Appendix B.
Notice the exponential suppression of higher winding modes in the coefficient Ck . If there
is a hierarchy MR 1 the effective potential is strongly suppressed.
The effective potentials in (23), (30) both are invariant under transformations
+ 2
q m (m integer). This is an expected result. Due to the form of KK mass spectrum
in (20), (A.3), (A.4), the Lagrangian is not changing under this shift because there is
summation over an infinite number of KK states. This manifests the 5D gauge invariance.
160
5. Cosmological applications
5.1. Slow roll and e-foldings during inflation
To decide under what conditions can drive inflation we have to look at its 4D kinetic
term as it follows from the 5D gauge curvature by integrating over y [6]. This may seem
strange at first sight since we used a derivative expansion in zeroth-order approximation
when calculating V eff (). However, in the slow-roll regime derivative terms arising from
the expansion of V eff (, ) are strongly suppressed as compared to the ones coming
from the 5D gauge curvature, see below. We have
L4D
=
1
V eff ().
2 (2R)2 g42
(32)
So the canonically normalized field is defined as

f
,
2Rg4
(33)
(5) 2
where (g (4) )2 (g2R) and MP is the 4D reduced Planck mass. The slow-roll conditions
for the field read
eff

(MP )2 (V eff ) 2
2 (V )
1,
|| = (MP ) eff 1,
5=
(34)
2
V eff
V
where (V eff ) , (V eff ) denote the potentials derivatives in respect of . Let us first discuss
case (i). We only consider the contribution at k = 1 in (23) since terms with k > 1 are
exponentially and strongly power suppressed. Substituting the k = 1 part of (23) or (30)
into (34), we obtain
1
5 = (2qg4 )2 (MP R)2 tan2 [2qg4 R ],
2
|| = (2qg4 )2 (MP R)2 .
(35)
R 1
According to (35) a hierarchy between

and MP can be compensated by a small 4D
gauge coupling g4 1. Moreover, a small g4 keeps the argument of the tangent small in 5
and thus yields an additional suppression. Let us now look at the amount of e-foldings we
may expect to be produced by the slowly rolling field .
The number of e-foldings for canonically normalized field (of ) is given by the
following formula
t f
N =
ti
f

V eff
8

H dt 2
d
.

MP (V eff )
(36)
Using (23), (30), (33), for both scenarios considered above, we obtain
N =
,
(MP R)2 g42
f
2
i O(1).
ln
sin(2qg
R
)
4
q2
(37)
161
Taking in (37) MP R = 10100, g4 102 103 , we easily get N 55, which evades
the horizon problem and guarantee flatness of the Universe up to the needed accuracy.
5.2. Application for quintessence
As was pointed out in [10], for a hierarchy 1/R M (recall that M is a SUSY
bulk mass for matter), the generated effective potentials are getting strongly suppressed.
From (23), (30) one sees that the potentials in both scenarios (i) and (ii) aquire a factor
exp(2MR), which for MR = 3050 provides suppression such that V eff (3
103 eV)4 . This amount can be associated with dark energy, making the model a candidate
for quintessence [11].
The presented effective potentials may be relevant for both inflation and quintessence.
Inflation would start at MR 1, and at the end of inflation or after reheating the radion
field would relax to the minimum of a stabilizing potential corresponding to MR 1.
This, of course, requires a dynamical model for the radion field, for its foundation within
5D supergravity see [25,26].
6. Discussion and outlook
In this paper we have realized the idea of gauge inflation within a 5D supersymmetric
setting exploiting U (1) gauge symmetry. For this, two possibilities of SUSY breaking were
considered. The no scale one-loop potential has the advantage that the tree-level contribution vanishes. In both scenarios, the obtained effective potentials make the Wilson-line
phase a candidate for the inflaton. Namely, for a given hierarchy between MP and the
compactification scale 1/R a proper choice of the 4D gauge coupling g4 guarantees slow
roll conditions and the required number of e-foldings. Due to SUSY the fine-tuning of the
4D cosmological constant after inflation is with respect to R 4 as opposed to the nonSUSY
value R(MP(5) )5 where MP(5) MP denotes the 5D Planck mass. Also, the specific forms
of the effective potentials open up the possibility to describe quintessence.
As a byproduct we have demonstrated that the winding mode representation can be
written down directly using the world-line method (Appendix B).
For quantitative estimates (in Eqs. (34)(37)), one should use a perturbatively small,
renormalized value of g4 . As it is known the KK excitations above 1/R induce a power
law running of g4 (linear in case of 5D). This effect was observed in our calculation of
V eff () in the case of D-term SUSY breaking. The emerging linear UV divergence, coming from the zero winding mode, should be absorbed into the bare 1/g42 (the corresponding
operator is 12 (5 )2 ). This is nothing but a renormalization of the gauge coupling. Ang4
other question, which should be understood (for both scenarios), is the origin of the small
value for g4 and the mechanism which guarantees its smallness. For example, an embedding of the U (1) in a non-Abelian gauge symmetry, which is unbroken at high energies,
prevents g4 to reach the Landau pole. For such a scenario, one could have in mind some
specific GUT, which naturally unifies U (1) with the SM interactions. Another way for an
ultraviolet completion is the possibility of embedding the 5D SUSY U (1) gauge theory in
a superstring model.
162
Acknowledgements
Research of F.P.C. is supported by Fundao para a Cincia e a Tecnologia (Grant
SFRH/BD/4973/2001).
Appendix A. KK spectrum and STr[e(D+B)t ]
Here we quote the mass matrices for the fluctuating KK modes as they arise from the
D-term SUSY breaking according to (19). After KK decomposition of the fields appearing
in (11) and subsequent integration over y the bilinear terms in the effective 4D bosonic KK
modes are
+
V (2) =
(n) 2 (m)

2
m,n
Mm,n
+ (n) M
(m) .
(A.1)
n,m=
The bilinear expressions for the superpartners have the form

L(2)
=
+
(n) (m) .
(n)
Mn,m
(m)
+ Mn,m
(A.2)
n,m=
The matrix elements appearing in (A.1) and (A.2) read as follows

q 2
n
q2
2
2

Mn,n = Mn,n =
+ M 2 + V 2,
R 2R
2
2
2
2
2

Mn,n+1 = Mn,n1 = Mn,n+1 = Mn,n1 = iqMV ,
2
q
2
2
2
2
n,n+2
n,n2
= Mn,n2
=M
=M
= V 2,
Mn,n+2
4
all other elements are zero. The matrix elements of fermionic states are

q
n
iq
Mn,n
+ M,
Mn,n1
=i
= Mn,n+1
= V,
R 2R
2
(A.3)
(A.4)
while all other elements are zero. For the calculation of the effective potential the matrix
M M + (M 2 ) is relevant where M + denotes complex conjugation of M and
transposition in the KK indices. The elements of (M 2 ) are

2
q 2
n
q2
M n,n =
+ M 2 + V 2,
R 2R
2
2

q2
M n,n+2 = M 2 n+2,n = V 2 ,
4
2
2
qV
qV
,
M n+1,n = iqMV
.
M n,n+1 = iqMV
(A.5)
2R
2R
All other elements are zero.
163
Let us now evaluate the expression

STr e(D+B)t = STr
+

(1)l
l!
l=0
t l (D + B)l .
(A.6)
We have
(D + B) = D +
l
l1
l1p
l2 l2p

BD +
p
p=0
D l2pq BD p BD q + . (A.7)
p=0 q=0
The supertrace of the first term in (A.7) vanishes because the elements of D coincide for
scalars and fermions, see (A.3) and (A.5). The trace of the second term in (A.7) is zero
because B is an off-diagonal matrix. Therefore the leading contributions are due to the
third term in (A.7). We have

Tr D l2pq BD p BD q = Tr D l2p BD p B
=
+

l2p
Dn
l2p
Dn1 |Bn,n1 |2 + Dn

p
Dn2 |Bn,n2 |2 ,
(A.8)
n=

+
l2pq

qV 2 l2p p
p
p
q
STr D
Dn1 + Dn+1 .
BD BD = 2
Dn
2R n=
(A.9)
According to (A.7), (A.9) we evaluate the following sum

+
l2 l2p

l2p
Dn
Dn1 + Dn+1
n= p=0 q=0
+
l2
l2p
(l 1 p)Dn
Dn1 + Dn+1
n= p=0
+
l2
l2p
(l 1 p)Dn
Dn1 +
n= p=0
+
l2
+
l2
l2p
(1 + p)Dn
Dn1
n= p=0
l2p
lDn
+
Dn1 =
n= p=0
n=
l1
Dnl1 Dn1
Dn Dn1
(A.10)
We have obtained the second term in the second line of (A.10) by substituting p
l 2 p and shifting n n 1. Finally, we have

STr(D + B)l = 2

= 2
qV
2R
2
l2
+
Dn1
n= p=0
+
qV 2
2R
l2p
lDn
n=
l1
Dnl1 Dn1
Dn Dn1
(A.11)
164
Appendix B. Calculation of V eff () in world-line formalism

Here we briefly sketch how an effective one-loop potential for arises in the worldline formalism. First, we consider bosonic fluctuations. Like in thermal physics [23]
the effective one-loop potential in the world-line formalism is expressed in terms of an
integration over the proper length of all possible closed-path trajectories. We have

d 4 x V eff ()

=

d 5x
T

2
dT
)
(
x
M
+ iq x 5 A5 + M 2 ,
D5 y exp d
T
4
k=
(B.1)
where
xM ( ) = 2Rk M5 + yM ( ) + xM ,
T
T
d yM ( ) = 0.
(B.2)
In (B.2) a constant part x5 and a non-constant, periodic part y( ) have been separated from
the topological part of the trajectory. Performing the integration over all trajectories and
using = 2RA5 , we arrive at

d 4 x V eff ()

=
d x
exp[ikq]
k=

1
dT
( Rk)2
2
M
exp
T
T (4T )5/2
T
3 1
= d 4x
16 6 R 4

1
cos(kq)
2
(2kRM)
1
+
2kRM
+
exp(2nRM)
3
k5
k=1

5
d x

1
dT
exp M 2 T .
T (4T )5/2
(B.3)
1/2
The last term in (B.3) is a contribution from k = 0 and is quintically UV divergent. It

expresses the usual cosmological-constant problem in 5D. Whether a cut-off must be
introduced or not in the KK tower needs not be addressed in the world-line formulation.
In particular in the non-SUSY case we do, however, need a 5D cut-off 2 T for the
zero-winding contribution.
In case (i) (spontaneous, no-scale SUSY breaking) the coupling to the gauge field of the
two complex, fluctuating scalars and the two Dirac fermions is in the world-line formalism
described by

i
qA5
2R FT
,
iq x5 A5 ,
ix5
(B.4)
i
2R
FT
qA5
165
respectively. Using these interactions to calculate the effective potential in analogy to (B.1)
and taking a trace, we again arrive at expression (23) which was obtained from a Poisson
resummation of the KK spectrum.
In case (ii) (D-term SUSY breaking by = V sin Ry ) the world-line treatment is more
involved since now the outer field is not constant [24]. Let us briefly sketch how
things work here. For charged scalars we only have to shift M M q2 in (B.1). For
Dirac fermions interacting with the path integral in (B.1) is enlarged by a world-line
Grassmann integration
T

1 M
1
D6 exp d M 6 6 qi6 5 5 .
2
2
(B.5)
The additional component 6 is needed if the coupling of the spinning particle to the scalar
background is of the Yukawa-type in field theory [24]. Since the part in (B.1), which
depends on x5 is universal for fermionic and bosonic fluctuations we only need to consider
the term qi6 5 5 in (B.5). This term breaks the supersymmetric cancellation. We write
= V /(2i)(exp[ix5 ( )/R] exp[ix5( )/R]), substitute this into (B.5) and expand the
exponential in (B.5) up to second order in V . Performing the Grassmann-integration and
exploiting translational invariance on the circle, the part of (B.5) due to qi6 5 5 turns
into
V2
2T 2
R
T

d G2F ( ) exp GB ( )/R 2 cos 2k

,
T
(B.6)
where GF = sign and GB = (1 )/T denote the fermionic and bosonic world-line
Greens functions, respectively. Substituting = yT in (B.6), performing the T - and the
x 5 -integrations, and neglecting the dependent contributions from the bosonic sector and
the linearly UV divergent k = 0 contribution, we arrive at the following effective potential

1/2
(qV )2 cos(kq)
.
dy cos(2ky) exp 2k M 2 R 2 + y(1 y)
2
2
k
R
1
k=1
(B.7)
This is the formula one would obtain from (30) by isolating the quadratic order in V .
References
[1] A.H. Guth, Phys. Rev. D 23 (1981) 347;
A.D. Linde, Phys. Lett. B 108 (1982) 389.
[2] E. Copeland, et al., Phys. Rev. D 49 (1994) 6410;
G. Dvali, Q. Shafi, R. Schaefer, Phys. Rev. Lett. 73 (1994) 1886.
[3] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Rev. Lett. 86 (2001) 4757, hep-th/0104005;
C.T. Hill, S. Pokorski, J. Wang, Phys. Rev. D 64 (2001) 105005, hep-th/0104035;
N. Arkani-Hamed, A.G. Cohen, E. Katz, A.E. Nelson, T. Gregoire, J.G. Wacker, JHEP 0208 (2002) 021;
I. Low, W. Skiba, D. Smith, Phys. Rev. D 66 (2002) 072001, hep-ph/0207243;
D.E. Kaplan, M. Schmaltz, hep-ph/0302049.
166
[4] S. Mollerach, Phys. Rev. D 42 (1990) 313;

D.H. Lyth, D. Wands, Phys. Lett. B 524 (2002) 5, hep-ph/0110002;
T. Moroi, T. Takahashi, Phys. Lett. B 522 (2001) 215, hep-ph/0110096;
K. Dimopoulos, D.H. Lyth, A. Notari, A. Riotto, hep-ph/0304050.
[5] K. Freese, J.A. Friemann, A.V. Olinto, Phys. Rev. Lett. 65 (1990) 3233;
F.C. Adams, J.R. Bond, K. Freese, J.A. Friemann, A.V. Olinto, Phys. Rev. D 47 (1993) 426, hep-ph/9207245.
[6] N. Arkani-Hamed, H.-C. Cheng, P. Creminelli, L. Randall, hep-th/0302034, hep-th/0301218;
D. Kaplan, N. Weiner, hep-ph/0302014.
[7] B. Feng, M. Li, R.-J. Zhang, X. Zhang, astro-ph/0302479.
[8] C.L. Bennett, et al., astro-ph/0302207.
[9] Y. Hosotani, Phys. Lett. B 126 (1983) 309, hep-ph/0303066.
[10] L. Pilo, D.A.J. Rayner, A. Riotto, hep-ph/0302087.
[11] C. Wetterich, Nucl. Phys. B 302 (1988) 668;
B. Ratra, P.J. Peebles, Phys. Rev. D 37 (1988) 3406;
R.R. Caldwell, R. Dave, P.J. Steinhardt, Phys. Rev. Lett. 80 (1998) 1582.
[12] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311;
N. Arkani-Hamed, L.J. Hall, Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81, hepph/0102090;
I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20.120.24, hep-th/0108005;
G. von Gersdorff, N. Irges, M. Quiros, Nucl. Phys. B 635 (2002) 127, hep-th/0204223, hep-th/0206029.
[13] D.M. Ghilencea, H.-P. Nilles, Phys. Lett. B 507 (2001) 327, hep-ph/0103151;
D.M. Ghilencea, H.-P. Nilles, S. Stieberger, New J. Phys. 4 (2002) 15, hep-th/0108183;
D.M. Ghilencea, S. Groot Nibbelink, H.-P. Nilles, Nucl. Phys. B 619 (2001) 385, hep-th/0108184.
[14] M.J. Strassler, Nucl. Phys. B 385 (1992) 145;
M.G. Schmidt, C. Schubert, Phys. Lett. B 318 (1997) 438;
M.G. Schmidt, C. Schubert, Phys. Lett. B 331 (1994) 69;
M.G. Schmidt, C. Schubert, Phys. Rev. D 53 (1996) 2150;
M. Reuter, M.G. Schmidt, C. Schubert, Ann. Phys. 259 (1997) 313.
[15] D. Marti, A. Pomarol, Phys. Rev. D 64 (2001) 105025, hep-th/0106256.
[16] N. Arkani-Hamed, T. Gregoire, J. Wacker, JHEP 0203 (2002) 055, hep-th/0101233;
A. Hebecker, Nucl. Phys. B 632 (2002) 101, hep-ph/0112230.
[17] E. Cremmer, S. Ferrara, C. Kounnas, D.V. Nanopoulos, Phys. Lett. B 133 (1983) 61.
[18] E. Witten, Phys. Lett. B 155 (1985) 151;
U. Ellwanger, M.G. Schmidt, Nucl. Phys. B 294 (1987) 445;
S. Ferrara, C. Kounnas, M. Porrati, Phys. Lett. B 181 (1986) 263;
S. Ferrara, C. Kounnas, F. Zwirner, Nucl. Phys. B 429 (1994) 589;
E. Dudas, C. Grojean, Nucl. Phys. B 507 (1997) 553.
[19] N. Chang, S. Ouvry, X. Wu, Phys. Rev. Lett. 51 (1983) 327.
[20] U. Ellwanger, N. Dragon, M.G. Schmidt, Phys. Lett. B 145 (1984) 192;
U. Ellwanger, N. Dragon, M.G. Schmidt, Nucl. Phys. B 255 (1984) 544;
U. Ellwanger, N. Dragon, M.G. Schmidt, Prog. Part. Nucl. Phys. 18 (1987) 1.
[21] M. Faibarn, L. Lopez-Honorez, M.H.G. Tytgat, hep-ph/0302160.
[22] R. Hofmann, F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, in preparation.
[23] D.G.C. McKeon, A. Rebhan, Phys. Rev. D 47 (1993) 5487;
M. Haack, M.G. Schmidt, Eur. Phys. J. C 7 (1999) 149.
[24] M. Mondragon, L. Nellen, M.G. Schmidt, Ch. Schubert, Phys. Lett. B 351 (1995) 200, hep-th/9502125;
M. Mondragon, L. Nellen, M.G. Schmidt, Ch. Schubert, Phys. Lett. B 366 (1996) 212, hep-th/9510036.
[25] G.v. Gersdorff, M. Quiros, Phys. Rev. D 65 (2002) 064016, hep-th/0110132;
G.v. Gersdorff, M. Quiros, A. Riotto, Nucl. Phys. B 634 (2002) 90, hep-th/0204041.
[26] For a recent formulation of 5D SUGRA see:
M. Zucker, Nucl. Phys. B 570 (2000) 267, hep-th/9907082;
M. Zucker, Phys. Rev. D 64 (2001) 024024, hep-th/0009083;
T. Kugo, K. Ohashi, Prog. Theor. Phys. 108 (2002) 203, hep-th/0203276.

ChernSimons vs. YangMills gaugings

in three dimensions
Hermann Nicolai a , Henning Samtleben b
a Max-Planck-Institut fr Gravitationsphysik, Albert-Einstein-Institut, Mhlenberg 1, D-14476 Golm, Germany
b Institute for Theoretical Physics & Spinoza Institute, Utrecht University,
Postbus 80.195, 3508 TD Utrecht, The Netherlands

Received 28 March 2003; accepted 27 June 2003
Abstract
Recently, gauged supergravities in three dimensions with YangMills and ChernSimons type
interactions have been constructed. In this article, we demonstrate that any gauging of YangMills
type with semisimple gauge group G0 , possibly including extra couplings to massive ChernSimons
vectors, is equivalent on-shell to a pure ChernSimons type gauging with non-semisimple gauge
group G0 T G, where T is a certain translation group, and where G is the maximal global
symmetry group of the ungauged theory. We discuss several examples.
PACS: 04.65.+e; 11.15.-q
1. Introduction
As borne out by recent work, gauged supergravities in three spacetime dimensions
come in more guises than the corresponding models in dimensions D 4. This variety
of theories is not least due to the fact that in three dimensions vector and scalar fields are
related by duality (see, e.g., [1] for a general discussion of such dualities). The on-shell
equivalence of scalars and vectors in three dimensions is not only reflected in a much
larger choice of gauge groups but also in the (co-)existence of both ChernSimons (CS)
type gaugings and YangMills (YM) type gaugings, depending on which fields carry the
propagating bosonic degrees of freedom.
E-mail addresses: nicolai@aei.mpg.de (H. Nicolai), h.samtleben@phys.uu.nl (H. Samtleben).

doi:10.1016/S0550-3213(03)00569-8
168
H. Nicolai, H. Samtleben / Nuclear Physics B 668 (2003) 167178
YM type gaugings can be obtained alternatively by direct construction, by torus

reduction of gauged supergravities in higher dimensions to three dimensions, or by
KaluzaKlein reduction on non-flat internal manifolds to three spacetime dimensions
(see [24], for examples, of such constructions). Examples of CS type gaugings were first
obtained for N = 2 supergravity with an Abelian gauge group [5]. Maximal (N = 16)
and half maximal (N = 8) gauged supergravities of CS type with various compact and
non-compact semisimple gauge groups were constructed in [6] and [7], respectively (the
relevant gauge groups are subgroups of E8(8) for N = 16 and SO(8, n) for N = 8).
The CS type supergravities are based on those versions of the ungauged theories, in
which all propagating bosonic degrees of freedom reside in the scalar fields, and which
therefore exhibit the largest global symmetry group G; the scalar fields parametrizing
the coset space manifold G/H. By contrast, YM type gaugings are deformations of a
dualized version of the ungauged theory where some of the propagating bosonic degrees
of freedom are carried by Abelian vector fields. The actual construction of the CS type
gauged Lagrangians is greatly facilitated by exploiting the group structure; a simple and
universal group-theoretical consistency condition determines all the admissible gauge
groups for an arbitrary number of local supersymmetries, as first shown for the N = 16 and
N = 8 theories [6,7], and subsequently for all other theories with N < 16 [8]. The direct
construction of the YM type theories, on the other hand, seems more involved, because
the global symmetry group G is broken to a smaller group G G, and furthermore the
remaining scalars cannot be assigned to a single coset space any more in general. The
relative simplicity of the CS type formulation in comparison with the YM type formulation
is evident from inspection of the resulting on-shell equivalent Lagrangians (2.1) and (3.8)
below.
In this paper we establish, independently of the number N of local supersymmetries,
that YM gaugings in three dimensions are in fact equivalent on-shell to CS type gaugings,
in the following sense. The equations of motion of any gauged supergravity of YM type
with (semisimple) gauge group G0 coincide with the equations of motion obtained from a
corresponding CS type gauged supergravity with non-semisimple gauge group G0 T ,
where T is a group of = dim G0 (Abelian) translations transforming non-trivially
under G0 . Our second main result is the extension of this construction to include couplings
of massive CS vector fields to the YM type Lagrangian. On the CS side these correspond
to additional nilpotent directions in the gauge group with a particular algebra structure,
see (3.1) below.
Generally, the scalar fields of the CS gauged theory (whose number we denote by d)
parametrize a coset space G/H with H the maximal compact subgroup of G. Here G is
the maximal global symmetry of the ungauged theory that can be achieved by dualizing all
propagating bosonic degrees of freedom into scalar fields. Obviously, the group we wish
to gauge must then satisfy G0 T G. Given a gauge group of this kind, we show that
scalar and vector fields may be eliminated together, whereupon the theory turns into a
YM gauged theory with (d) scalars and propagating vector fields gauging the group
G0 . It is important that the scalar potential is independent of the scalar fields in question
and is therefore not affected by this elimination procedure. After the elimination, the (d)
scalars of the YM gauged theory in general can no longer be uniformly described as a coset
space; only part of them can be assigned to a (smaller) coset space G /H with G G,
Ungauged theory
d scalars, no vectors
dualization
gauging
Gauged theory (2.1)

d scalars, 2 CS-vectors
gauge group: G0 T
169
Ungauged theory
d scalars, Abelian vectors
gauging
elimination
by means of (2.19)
Gauged theory (2.22)

d scalars, YM-vectors
gauge group: G0
Fig. 1. CS and YM gauged supergravity in three dimensions.
H H (of course, the YM gauge group G0 must be contained in G ). Allowing also for
some bosonic degrees of freedom to be realized as massive CS vectors (see Section 3), the
following general matching condition for the bosonic degrees of freedom must evidently
be satisfied
d = dim G/H = #(scalars) + #(YM vectors) + #(massive CS vectors).
(1.1)
The procedure relating the two types of theories is schematically represented in Fig. 1.
The necessity of flat directions in the CS gauge group, and hence of non-semisimple
gauge groups, for the transmutation of a CS type theory into a YM type theory may be
understood by noting that only those scalar fields on which the scalar potential of the
gauged theory does not depend can be dualized away and replaced by YM vector fields.
Hence the associated translations along these directions on the target space manifold must
be among the local symmetries of the CS type Lagrangian. There exist numerous results
on non-semisimple gaugings in D 4 [913], but non-semisimple CS gauge groups in
three dimensions have been considered only very recently [14], so we here only note that
there are two methods to search for them. The first is to directly solve the group-theoretical
consistency conditions of [68]. For the second, one performs an infinite boost on a
known admissible semisimple gauge group with a suitable non-compact element of the
global symmetry group G; this boost must be accompanied by a singular rescaling of the
coupling constants, which is adjusted in such a way that the embedding tensor has a
finite limit.
In summary, the set of CS gauged supergravities in three dimensions with nonsemisimple gauge groups contains all the known types of YM gauged supergravities.
Because there exist numerous admissible semisimple gauge groups, that cannot be related
to YM type gaugings, it follows that the CS gauged supergravities encompass a much
larger class of models than those of YM type. Combining the classification of ungauged
three-dimensional theories [15] with the group-theoretical results on the existence of CS
gaugings [68] thus provides a straightforward route to constructing gauged supergravity
theories in three dimensions for a given field content, gauge group and number of
supersymmetries.
The paper is organized as follows. In Section 2 we provide details of the correspondence
outlined in Fig. 1, exhibiting the YM type gaugings as a particular subclass of CS gauged
170
theories. Section 3 describes the generalization to coupling additional massive vector fields.
We close by briefly discussing several examples, including the compactifications of type
I, IIA, IIB supergravity in D = 10 on AdS3 S 7 and the six-dimensional supergravity on
AdS3 S 3 .
2. ChernSimons vs. YangMills gauging
The bosonic Lagrangian of a CS gauged supergravity theory in three dimensions is
always of the form (see [6,7] for our conventions and notations; we use the metric (+ ))
1
1
e1 L = R + e1 gLCS + g PA PA W + .
(2.1)
4
4
The dots stand for fermionic terms which we will here ignore because they are not relevant
for the argument we are going to present; see however [68] for further details concerning
the fermionic Lagrangian and the supersymmetry variations. The first term in (2.1) is just
the usual Einstein term, while the second is the CS Lagrangian

1
1
LCS = BM MN BN + gf N P L PK BK BL ,
(2.2)
4
3
with the constant symmetric embedding tensor MN , which characterizes the CS gauge
group, see (2.3) below. Note that LCS comes with a factor g, the gauge coupling constant.
g 0 describes the (smooth) limit to the ungauged theory.
To explain the remaining two terms in (2.1) we recall that the d scalar fields parametrize
a coset space G/H where H is the maximal compact subgroup of G. Of course, the choice
of possible coset spaces depends on the number N of local supersymmetries and becomes
more and more restricted with increasing N [15]. Explicitly, the scalar fields are described
by a group valued matrix S G such that the current

Q + P S 1 + gMN BM t N S,
(2.3)
takes values in the associated Lie algebra g, with a suitably normalized basis {t M }, where
M, N = 1, . . . , dim G. The quantity P = PA t A appearing in the kinetic term of the scalar
fields in the Lagrangian (2.1) is the projection of this current onto the non-compact part of
g, which is spanned by the generators {t A } with labels A, B, . . . . The compact part Q , on
the other hand, serves as a (composite) connection for the maximal compact subgroup H
and governs the scalarfermion couplings of the ungauged theory. The embedding tensor
MN describes the coupling of the vector fields to the generators of the action of the
symmetry group, hence the embedding of the CS gauge group into G.
The potential W is likewise a function of the scalar fields S. More specifically, it is a
quadratic polynomial in the entries of the T -tensor TAB which in turn is given in terms of
the matrix V M A representing the group element S in the adjoint representation:
TAB MN V M A V N B ,
V M A t A S 1 t M S.
(2.4)
The exact dependence of W on TAB as well as the possible gauge groups and their
embedding matrices MN can be found in [68] and is completely determined by
supersymmetry.
171
The above Lagrangian is invariant under the (infinitesimal) gauge transformations

S = gMN M t N S,
BM = M + gf MP L PK BK L .
(2.5)
Variation of the Lagrangian (2.1) with respect to the vector fields gives rise to the first
order duality equations (again omitting fermionic contributions):

N
N
MN B
MN 2[ B]
+ gKP f N K Q BP BQ
= e MN V N A P A .
(2.6)
To proceed we now assume a very particular type of gauge group, namely, a

non-semisimple group of the form G0 T G, where T is a set of = dim G0
translations transforming in the adjoint representation of G0 . A systematic discussion and
representative examples of such gaugings will be given in [14]. Denoting the generators
of g0 Lie G0 by {J m t m } and those of t Lie T by {T m t m }, respectively, where
both m and m range over 1, . . . , dim G0 , we have the commutation relations
m n
m n
m n
J , J = f mn k J k ,
(2.7)
J , T = f mn k T k ,
T , T = 0,
where f mn k denote the structure constants of G0 , so the translation generators transform
in the adjoint of G0 . It can now be shown that for this particular choice of gauge group,
a consistent gauging is possible only if the embedding tensor MN is of the form
gmn = gmn = g1 mn ,
gmn = g2 mn ,
(2.8)
with all remaining components equal to zero. Here mn is the CartanKilling form on g0 .
The ansatz (2.8) is uniquely fixed by demanding invariance of MN under the gauge
group (2.7). In particular, this invariance requires mn = 0; happily, this is also the
condition needed for our elimination procedure to work. The real constants g1 , g2 in
general cannot be freely chosen, but are determined as a function of the one free gauge
coupling constant g by how G0 T is embedded in G, and by the fact that the embedding
tensor (2.8) must satisfy the group-theoretical identities of [68] to ensure compatibility
with supersymmetry. After the elimination procedure we are about to describe, the coupling
g1 will play the role of the YM gauge coupling constant while g2 corresponds to an
inequivalent deformation of the theory by an additional CS term.
m
We denote the vector fields associated with G0 and T by Cm Bm and Am
B ,
respectively. Their transformation properties follow from (2.5):
m
m
k l
Am
= + g1 f kl A ,

Cm = m + g1 f m kl Ck l + f m kl Ak g1 l + g2 l ,
(2.9)
where f m kl nk f mn l . The associated field strengths can be read off from (2.6); they are
m
m
m
m
k l
Am
B = A A + g1 f kl A A ,
m
m
k
B
= Cm Cm + 2g1 f m kl C[
Al] + g2 f m kl Ak Al .
C
(2.10)
Next we split off the translation part from the scalar field matrix S
),
em T m S(
S(, )
(2.11)
172
in terms of scalars m associated with the translation generators {T m }, and the remaining
scalar fields some of which coordinatize the YM coset manifold G /H . What is
important is that the matrix S no longer depends on the translational degrees of freedom
m . Defining the modified field strength
m
m
C
f mn k n Ak ,
C
(2.12)
we have the transformation properties

m
k
l
Am
= g1 f kl A ,
m
k
= g1 f m kl C
l ,
C

S= g1 mn m J n S,

m = g1 mn n g2 mn + g1 f l mn l n ,
(2.13)
from (2.5). Note that the fields m and Cm are the only ones to transform with the
translation parameters m , and that m is shifted under such transformation and hence
could be gauged away altogether. Accordingly, we define the quantities

M A t A S1 t M S,
V

1
n
+ P
S + g1 mn Am
Q
J S,
(2.14)
which do not depend on the m either. It is then easy to check that, for M = m, m
m A = V m A ,
V

V m A = V m A f mn k n V k A .
(2.15)
As expected, the T -tensor (2.4) does not depend on m . This follows from the fact that
by construction it is gauge invariant, and more specifically invariant under the (local) m
translations on m , see (2.13), but it is also easy to verify directly that
V MA
VN B,
TAB = MN V M A V N B = MN
(2.16)
with MN from (2.8). Consequently, the scalar potential W in (2.1) does not depend
on m . After a little algebra, the current Q + P can be rewritten as

m A

+ P
+ m + mn g1 Cn + g2 An + g1 f k mn An k
V At
Q + P = Q
+ P
+ D m
V mA t A,
Q
(2.17)
with the definition of the covariant derivative in accordance with (2.13). The first order
duality equations (2.6) for the gauge group (2.7), (2.8), take the form

m A + V
n A D n ,
Am
= e V A P
A

m
m A P
+
C
(2.18)
= e V
V n A D n ,
m from (2.12). Hence they may be formulated exclusively
with the modified field strength C
in terms of objects that are invariant under m and transform covariantly under m .
Our aim is now to eliminate all m dependence from the equations of motion. To this
end, assume the matrix M mn
V mA
V n A to be invertible with inverse Mmn . This allows to
m
solve Eq. (2.18) for D m and C
A ,
V nAP
e D m = Mmn An e Mmn

m
m B
m A
A + V
V k B Mkl
VlA P
V k A Mkn An .
= e
C
V mA V
(2.19)
173
These equations can now be used to eliminate both m and Cm from the theory. Solubility
of the first equation in (2.19) implies an integrability condition on the r.h.s. which is
straightforwardly computed using

n
[D , D ]m = mn g1 C
(2.20)
+ g2 An ,
and leads to the following second order field equation for the vector fields Am

D Mmn An

B
A + g1 mn

= e D Mmn
V k A Mkl
V nAP
V n A AB
VlB P

1
l A Mln An .
+ e g2 mn + g1 mk
(2.21)
V k AV
2
Hence this field equation is equivalent to the set of first order equations (2.18) while the
fields m and Cm are completely decoupled and may be restored from (2.19). (Integrability
of the second equation in (2.19) in addition requires also part of the scalar field equations.)
Eq. (2.21) may be derived from the Lagrangian
= 1 R e1 g2 L
CS (A) 1 Mmn Am An + 1 g GAB P
A P
B W
e1 L
4
8
4
1
n A Am
A
+ e1 Mmn V
(2.22)
P ,
4
with
m A Mmn
GAB AB V
V nB ,

n
CS (A) = 1 Am
L
mn A +
4
m n 1
A ,
Mmn
V AV

1
g1 f n kl Ak Al .
3
(2.23)
It requires a little more work to show that the scalar field equations derived from (2.22)
reproduce those descending from (2.1) upon eliminating D m by means of (2.19). Note
that the metric GAB on the scalar target space is degenerate along the directions
V mA .
This just means that the elimination of the scalar fields m has effectively reduced the
dimension of the scalar manifold in (2.22) by . The ChernSimons term in (2.22) collects
only part of the terms from the corresponding term in (2.1). The scalar potentials W in
(2.1) and (2.22) coincide. The resulting YM type theory thus has d scalar fields and
propagating YangMills vectors. The residual gauge group G0 acts canonically as

S = g1 mn m J n S,
m
m
k l
Am
= + g1 f kl A .
(2.24)
The fermionic part of the Lagrangian (2.22) as well as the supersymmetry transformation
m by
rules may be directly obtained from those of (2.1) upon eliminating D m and C
means of (2.19).
In the (smooth) limit g1 0, g2 0, the Lagrangian (2.22) reduces to the ungauged
theory with d scalar fields and Abelian vectors. The metrics Mmn and GAB in the
kinetic terms remain unchanged in this limit. As anticipated above, the two constants
g1 and g2 from (2.8) in (2.22) correspond to deformations of the ungauged theory of
two different types: The constant g1 arises as gauge coupling constant of the gauge
group G0 while g2 appears as proportionality factor of the ChernSimons term which
174
may be viewed as another deformation of the ungauged YM theory. In addition, both

constants appear in the T -tensor and thereby in the scalar potential W which is a quadratic
polynomial in g1 , g2 . Let us however stress once more that in general g1 and g2 are not free
parameters in (2.8) but related by some consistency relation implied by supersymmetry. In
the degenerate case g1 = 0, the dualized theory (2.22) appears with an Abelian gauge group
U (1) which does not act on the scalar fields, and is deformed only by the presence of the
ChernSimons term and a scalar potential. This has been worked out in [16] in the context
of the N = 2 theories describing the CalabiYau fourfold compactifications of M-theory
with flux [17].
3. Coupling massive vector fields

The above elimination procedure can be extended in a straightforward fashion to include
couplings to massive vector fields in the framework of pure CS gaugings (2.1). As we will
now explain these massive vector fields correspond to additional nilpotent directions in the
CS gauge group.
To this aim we consider an extension of the CS gauge group G0 T by a set
Tp
of p nilpotent generators transforming in some representation of G0 and closing into T .
Accordingly, the Lie algebra relations (2.7) are extended by
m

m
J , T = t m T ,
(3.1)
T , T = t m T m ,
T , T = 0,
while the embedding tensor MN has the additional components
g = g1 ,
(3.2)
with the structure constants in (3.1) and the symmetric tensor being related by
mn t n = t m in order to have MN invariant under the gauge group. We denote the
group corresponding to (2.7), (3.1) by G0 (
Tp , T ). In addition to the vector fields (2.9)
there are now also vector fields B corresponding to the nilpotent generators T . Similar
to (2.11), we may also split off the scalars associated with the generators T from the
scalar field matrix S
).
em T m e T S(
S(, )
(3.3)
Explicitly, the individual parts of (3.3) transform under gauge transformations as

S = g1 mn m J n S,
= g1 + g1 tm m ,
1
m = mn g1 n g1 tm
2

1
l
g2 mn + g1 f mn l g1 t m t n n ,
2
(3.4)
and correspondingly this defines their covariant derivatives. The elimination procedure
described in the last section may now straightforwardly be generalized to this setting. We
175
refrain from giving details of the computation and just note that (2.17) generalizes to

1
A
A
m

P = P + V A D + V A D m t m D ,
(3.5)
2
defined as in (2.14). The first order duality equations (2.6) then allow to express
with
V, P
m in terms of the remaining fields, thereby eliminating them
D m and the field strength C
from the theory. As above, integrability of these equations implies a second order field
equation for the vector fields Am
that generalizes (2.21). In addition, we remain with a first
order equation for the vector fields B associated with the nilpotent generators T

B
D B D B t m Am
B

+V
B D ,
V m A Mmn An + e+
V A GAB P
=
V A
(3.6)
with GAB from (2.23), and where left- and right-hand side are separately invariant under
gauge transformations with parameters , m . Finally, the entire set of field equations
may be derived from the Lagrangian
A
B
1
n
A D P
= 1 R + 1 GAB P
+ V
+
e1 L
V B D Mmn Am
A
4
4
8
A
1 1
1
n A Am
D

B
+ e1 Mmn V
P + V A D e
4
8
CS (A) W,
+ e1 g2 L
(3.7)
which generalizes (2.22) by including a coupling to the additional vector fields B which
arise with the first order field equation (3.6). The Lagrangian (3.7) has the additional gauge
symmetry exclusively acting on B and shifting . In particular, this symmetry may
be gauge fixed by imposing = 0 which leads to the Lagrangian
1
n
= 1 R + 1 GAB P
A P
B 1 Mmn Am
n A Am
A
e1 L
+ e1 Mmn V
A
P
4
4
8
4
1
1
A B Am
+ g12
VA GAB
VB B B + g1 e1 Mmn
V nAV
4
4
1
A B 1 g1 e1 B B
VB P
+ g1 GAB
2
8
CS (A) W.
+ e1 g2 L
(3.8)
together with
This is a theory describing (dp) scalar fields combined in the matrix S,
m
p massive CS vector fields B , and YM vector fields A gauging the group G0 with the
gauge symmetry acting as
m
m
k l
Am
= + g1 f kl A ,

S = g1 mn m J n S,
B = g1 t m B m .
(3.9)
Lagrangians of the type (3.8) typically arise in dimensional reduction on non-trivial

internal manifolds [3,4], see the following section. Note that in contrast to (2.1), (3.7),
this Lagrangian no longer has a smooth limit g1 0 because some propagating degrees of
freedom decouple with the B . It is worthwhile to emphasize the simplicity of the original
176
Lagrangian (2.1) in comparison with (3.8), which is due to the fact that all the different
scalar tensors which describe the couplings of the various fields in (3.8) are encoded in
the original G/H coset structure. Moreover, the gauge deformation of (2.1) is uniformly
described in terms of the embedding tensor which transforms covariantly under the
maximal global symmetry G underlying the CS description of the gauged theory.
4. Examples
Having shown that the CS gauged supergravities (2.1) contain the YM type theories
(2.22) and (3.8) as special cases, we now have the means to construct any three-dimensional
supergravity given the number of supersymmetries, gauge group and field content. To do
so, one first identifies that version of the underlying ungauged supergravity for which all
propagating bosonic degrees of freedom appear as scalar fields and are uniformly described
by a maximal coset space G/H. By contrast, the YM type theory is based on a description
where only part of the bosonic degrees of freedom correspond to scalar fields, such that
(1.1) holds.
Together with the precise representation content under the gauge group and for
sufficiently large number N of supersymmetries, this is already sufficient to identify
the corresponding theory in the list of [15]. Next, the gauge group must be chosen
as a subgroup of G such as to reproduce the correct representation content while
its non-semisimple part determines the nature of the vector couplings as explained in
Sections 2 and 3. In addition, the embedding tensor MN of this group is constrained
by the group-theoretical consistency condition of [68] where the complete form of the
CS gauged theory is then found. Finally, one may apply the constructions presented in this
paper to cast the vector couplings into the desired form (2.1), (2.22) or (3.8).
We conclude with some examples that reproduce the mass spectra and symmetries of
known AdS3 compactifications. Most of these theories have not been constructed before.
Recall that in dimensional reduction one typically encounters YM gauged theories, i.e.,
the Lagrangian obtained directly by compactification will take the form (2.22), (3.8)
rather than the equivalent simpler form (2.1). As pointed out in [6] the gauged CS type
theories with semisimple gauge groups have no obvious higher dimensional ancestor. It is
therefore remarkable that the CS type models which can be linked to higher dimensional
supergravities, all have non-semisimple gauge groups. Could there be a higher-dimensional
theory that gives rise to these models in a singular limit akin to the boost limit producing
non-semisimple gaugings from semisimple ones? We should also stress that at this stage we
restrict attention to the (unique) lower-dimensional theories with the correct field content,
and are not concerned with the consistency of the truncations from the higher-dimensional
point of view.
One of the main examples of the AdS/CFT correspondence is the duality between
type IIB string theory on AdS3 S 3 M 4 and certain two-dimensional conformal field
theories [18]. The spectrum of N = (2, 0) supergravity on AdS3 S 3 has been computed
in [1921], in three dimensions this is a half-maximal, i.e., N = 8 theory. In [7] we have
shown that the lowest multiplets of this spectrum together with the expected global and
local symmetries are reproduced by a three-dimensional theory (2.1) with coset space
177
SO(8, n)/(SO(8) SO(n)), and CS gauge group SO(4), where n denotes the number
of tensor multiplets in six dimensions. An outstanding question has been the coupling
of this theory to the YM vector multiplet containing additional 26 scalars. In view of the
above results, one may now verify that this larger theory may be described by a coset
space SO(8, 4 + n)/(SO(8) SO(4 + n)) and CS gauge group SO(4) T6 . This SO(4) is
embedded as a certain diagonal of two factors in the SO(8) and the SO(4 + n), respectively,
such that the scalar spectrum decomposes as (8, (n + 4)) n 4 + 4n 1 + 1 + 9 + 4
4 + 3+ + 3 . Eliminating the six Abelian translations as described in Section 2 leads to a
YM SO(4) gauged theory (2.22) coupling 26 + 8n scalar fields. Details are given in [22].
Surprisingly, using the results of Section 3 and a particular coset space, it is even possible
to describe the coupling of multiplets from arbitrary (!) levels of the massive spin-1 KK
towers.
The near horizon limit of the so-called double D1D5 system describes an
AdS3 S 3 S 3 geometry. The supergravity spectrum on this background has been
computed in [23]. The three-dimensional theory describing the lowest mass multiplets
again is organized by a coset space SO(8, n)/(SO(8) SO(n)), now with gauge group
SO(4) SO(4) [7]. Similar to the construction given above, one may further couple the
two YM multiplets by a proper enlargement of the coset space to SO(8, 8 + n)/(SO(8)
SO(8 + n)) and embedding a gauge group SO(4)diag SO(4)diag together with the
corresponding nilpotent directions.
Recently, the reduction of six-dimensional N = (1, 0) supergravity on AdS3 SU(2),
has been performed in [4]. The field content of the three-dimensional N = 4 theory
comprises three YM gauge fields together with three massive vector fields and six scalars
parametrizing the coset space GL(3)/SO(3). This spectrum suggests that the CS version
(2.1) of this theory is governed by the larger coset space SO(4, 3)/(SO(4) SO(3)) with
gauge group SO(3)diag (
T3 , T3 ). The group SO(4, 3) indeed has a unique subgroup of
this type, whose algebra generators satisfy relations of the type (2.7), (3.1). Its semi-simple
part is embedded as the diagonal of the three SO(3) factors in the compact SO(4) SO(3),
such that the 12 scalars decompose as 12 1 + 3 + 3 + 5. Eliminating the three Abelian
translations and gauge-fixing the other three nilpotent directions as described in Section 3
leads to a theory (3.8) with the desired spectrum. Indeed, the couplings appearing in (3.8)
for this case are of the form found in [4].
The reduction of six-dimensional N = (1, 0) supergravity on AdS3 S 3 leads to
a three-dimensional N = 4 theory, whose spectrum has been computed in [19,21]. The
resulting theory will be described by a coset space SO(4, 4)/(SO(4) SO(4)) and CS
gauge group SO(4)diag T6 embedded such that the scalar spectrum correctly decomposes
as 16 1 + 3+ + 3 + 9, of which the 3+ + 3 transforming in the adjoint representation
of SO(4)diag are eliminated according to Section 3.
The compactification of simple five-dimensional supergravity on S 2 whose KK
spectrum has been analyzed in [21,24,25] should be related to proper gaugings of the
N = 4 theory with coset space G2(2)/SO(4). The CS gauge group is a SO(3)diag T3
under whose semisimple part the scalar spectrum decomposes as 8 3 + 5.
178
The most interesting example is the (warped) compactification of ten-dimensional

supergravity on S 7 . For the type I theory, the reduction has been performed explicitly
in [3]. We may recover that theory by starting from an N = 8 theory (2.1) with coset
space SO(8, 8)/(SO(8) SO(8)), and CS gauge group SO(8)diag T28 upon eliminating
the 28 translations, leading to an SO(8) gauged YM theory (2.22) with 36 scalar fields.
Using the above results, it is then straightforward to extend this construction to the
maximally supersymmetric theories which are supposed to describe the compactifications
of type IIA/IIB theory on S 7 , and whose spectra have been given in [26]. These two
different compactifications correspond to two inequivalent embeddings of the gauge group
SO(8)diag T28 into E8(8) [14].
Acknowledgements
It is a pleasure to thank M. Berg, B. de Wit, T. Fischbacher, and M. Haack for
discussions. This work is partly supported by EU contract HPRN-CT-2000-00122.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
E. Cremmer, B. Julia, H. Lu, C.N. Pope, Nucl. Phys. B 523 (1998) 73144, hep-th/9710119.
H. Lu, C.N. Pope, P.K. Townsend, Phys. Lett. B 391 (1997) 3946, hep-th/9607164.
M. Cvetic, H. Lu, C.N. Pope, Phys. Rev. D 62 (2000) 064028, hep-th/0003286.
H. Lu, C.N. Pope, E. Sezgin, SU(2) reduction of six-dimensional (1, 0) supergravity, hep-th/0212323.
N.S. Deger, A. Kaya, E. Sezgin, P. Sundell, Nucl. Phys. B 573 (2000) 275290, hep-th/9908089.
H. Nicolai, H. Samtleben, Phys. Rev. Lett. 86 (2001) 16861689, hep-th/0010076;
H. Nicolai, H. Samtleben, JHEP 0104 (2001) 022, hep-th/0103032.
H. Nicolai, H. Samtleben, Phys. Lett. B 514 (2001) 165172, hep-th/0106153.
B. de Wit, I. Herger, H. Samtleben, Gauged locally supersymmetric D = 3 non-linear sigma models, hepth/0307006.
C.M. Hull, Phys. Lett. B 142 (1984) 3941;
C.M. Hull, Phys. Lett. B 148 (1984) 297300;
C.M. Hull, Phys. Rev. D 30 (1984) 760.
L. Andrianopoli, F. Cordaro, P. Fr, L. Gualtieri, Class. Quantum Grav. 18 (2001) 395413, hep-th/0009048.
L. Andrianopoli, R. DAuria, S. Ferrara, M.A. Lled, JHEP 07 (2002) 010, hep-th/0203206.
C.M. Hull, New gauged N = 8, D = 4 supergravities, hep-th/0204156.
B. de Wit, H. Samtleben, M. Trigiante, Nucl. Phys. B 655 (2003) 93126, hep-th/0212239.
T. Fischbacher, H. Nicolai, H. Samtleben, Non-semisimple and complex gaugings of N = 16 supergravity,
hep-th/0306276.
B. de Wit, A.K. Tollstn, H. Nicolai, Nucl. Phys. B 392 (1993) 338, hep-th/9208074.
M. Berg, M. Haack, H. Samtleben, CalabiYau fourfolds with flux and supersymmetry breaking, JHEP 04
(2003) 046, hep-th/0212255.
M. Haack, J. Louis, Phys. Lett. B 507 (2001) 296304, hep-th/0103068.
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231252, hep-th/9711200.
S. Deger, A. Kaya, E. Sezgin, P. Sundell, Nucl. Phys. B 536 (1998) 110140, hep-th/9804166.
F. Larsen, Nucl. Phys. B 536 (1998) 258278, hep-th/9805208.
J. de Boer, Nucl. Phys. B 548 (1999) 139166, hep-th/9806104.
H. Nicolai, H. Samtleben, KaluzaKlein supergravity on AdS3 S 3 , hep-th/0306202.
J. de Boer, A. Pasquinucci, K. Skenderis, Adv. Theor. Math. Phys. 3 (1999) 577614, hep-th/9904073.
A. Fujii, R. Kemmoku, S. Mizoguchi, Nucl. Phys. B 574 (2000) 691718, hep-th/9811147.
Y. Sugawara, JHEP 06 (1999) 035, hep-th/9903120.
J.F. Morales, H. Samtleben, JHEP 08 (2002) 042, hep-th/0206247.

Membrane and non-commutativity

Rabin Banerjee a,1 , Biswajit Chakraborty b , Kuldeep Kumar b
a Institute of Particle and Nuclear Studies, High Energy Accelerator Research Organisation (KEK),
Tsukuba 305-0801, Japan

b S.N. Bose National Centre for Basic Sciences, Block JD, Sector III, Salt Lake, Kolkata 700 098, India
Abstract
We analyse the dynamics of an open membrane, both for the free case and when it is coupled to a
background three-form, whose boundary is attached to p-branes. The role of boundary conditions and
constraints in the NambuGoto and Polyakov formulations is studied. The low-energy approximation
that effectively reduces the membrane to an open string is examined in detail. Non-commutative
features of the boundary string coordinates, where the cylindrical membrane is attached to the Dpbranes, are revealed by algebraic consistency arguments and not by treating boundary conditions as
primary constraints, as is usually done. The exact form of the non-commutative algebra is obtained
in the low-energy limit.
PACS: 11.25.-w; 11.15.-q; 11.10.Ef
Keywords: Constrained Hamiltonian analysis; Membranes; Non-commutativity
1. Introduction
Over the last decade string theory has been gradually replaced by M-theory as the most
natural candidate for a fundamental description of nature. While a complete definition of
M-theory is yet to be given, it is believed that the five perturbatively consistent string
theories are different phases of this theory. With the replacement of string theory by
M-theory, the string itself has lost its position as the main candidate for the fundamental
degree of freedom. Instead, higher-dimensional extended objects like membranes are being
E-mail addresses: rabin@post.kek.jp, rabin@bose.res.in (R. Banerjee), biswajit@bose.res.in
(B. Chakraborty), kuldeep@bose.res.in (K. Kumar).
1 On leave of absence from S.N. Bose National Centre for Basic Sciences, Kolkata, India.
doi:10.1016/j.nuclphysb.2003.07.009
180
R. Banerjee et al. / Nuclear Physics B 668 (2003) 179206
considered [1]. Indeed it is known that membrane and five-brane occur naturally in elevendimensional supergravity, which is argued to be the low-energy limit of M-theory. Also,
string theory is effectively described by the low-energy dynamics of a system of branes.
For instance, the membrane of M-theory may be wrapped around the compact direction
of radius R to become the fundamental string of type-IIA string theory, in the limit of
vanishing radius.
An intriguing connection between string theory, non-commutative geometry, noncommutative (as well as ordinary) YangMills theory was revealed in [2]. With the shift
in focus from string theory to M-theory, there has been a flurry of activity in analysing
non-commutativity in membranes, specifically when an open membrane that couples to a
three-form, ends on a D-brane [38]. The motivation of the present paper is to further this
investigation, but with a new perspective and methodology, as explained below.
The study of non-commutative properties in membranes is more involved than the
analogous study in the string case since the equations to be solved are non-linear.
Naturally, in contrast to the string situation, the results could be obtained only under some
approximations. It is useful to recapitulate how non-commutativity is derived in either
the string coupled to the two-form or the membrane coupled to the three-form. There are
non-trivial boundary conditions which are incompatible with the basic Poisson brackets
of the theory. These boundary conditions are considered as primary constraints in the
algorithm of Diracs constrained Hamiltonian dynamics [610]. The primary constraints
lead to secondary constraints. Non-commutativity is manifested through the occurrence of
non-trivial Dirac brackets. The brackets are found to be gauge dependent, but there is no
gauge where it can be made to vanish.
Recently, an alternative approach to deal with non-commutativity in strings was
advocated in a paper [11] involving two of us. Contrary to other approaches, the boundary
conditions are not interpreted as primary constraints. The non-commutative algebra
emerges from a set of consistency requirements. It is rather similar in spirit to the original
analysis of [12] where a modified algebra, involving the periodic delta function instead of
the usual one, was found for the coordinates and their conjugate momenta, in the example
of the free NambuGoto (NG) string.
In this paper we adopt our previous strategy for strings to the membrane model.
We discuss both the NG and Polyakov forms of action, although non-commutativity is
explicitly considered only in the latter formulation. The similarities or otherwise in the
analysis of the two actions are illuminated. Analogous to the set of orthonormal gauge
fixing conditions given for the free NambuGoto string [11,12], we derive a set of quasiorthonormal gauge conditions for the free NG membrane. Just as the orthonormal gauge
in the NG string corresponds to the conformal gauge in the Polyakov string, we find out
the analogue of the quasi-orthonormal gauge in the Polyakov membrane. It corresponds
to a choice of the metric that leads to equations of motion that can be explicitly solved in
the light-front coordinates [13]. The structure and implications of the boundary conditions
in the two formulations have been elaborated. In the NG case, the conditions involve the
velocities that cannot be inverted so that a phase-space formulation is problematic. Only
by fixing a gauge is it possible to get hold of a phase-space description. In the Polyakov
type, on the other hand, the boundary condition is expressible in phase-space variables
without the need of any gauge choice. This is because the metric itself is regarded as an
181
independent field. In this sense, therefore, there is no qualitative difference between string
and membrane boundary conditions, since even in the NG string, a gauge fixing is required
for writing the boundary conditions in terms of phase-space variables. We thus differ from
[7] where it is claimed that it is imperative in the membrane case, as opposed to the string
case, to gauge fix in order to express the boundary conditions in phase-space coordinates
as a first step in the Hamiltonian formalism.
The mandatory gauge fixing in the NG membrane, as we shall show, converts the
reparametrization invariant (first-class) system into a second-class one, necessitating the
use of Dirac brackets. This involves the inversion of highly non-linear expressions, so that
approximations become essential to make any progress. Hence, we avoid this formulation
in favour of the Polyakov version, where gauge fixing is not mandatory.
A detailed constrained Hamiltonian analysis of the free bosonic Polyakov membrane
naturally leads to three restrictions on the world-volume metric. These are found to be
identical to those obtained by counting the independent degrees of freedom. Unlike the
case of the classical string where there are three components of the metric and three
continuous symmetries (two diffeomorphism symmetries and one scale symmetry), leading
to a complete specification of the metric by gauge fixing, for the membrane there are six
independent metric components and only three diffeomorphism symmetries. Thus only
three restrictions on the metric can be imposed. Interestingly, the restrictions usually put in
by hand [13] to perform calculations in the light-front coordinates are obtained directly in
our Hamiltonian formalism. This gauge fixing is only partial in the sense that the non-trivial
gauge generating first-class constraints remain unaffected. Effectively, therefore, it is a
gauge-independent Hamiltonian formalism. We show that the boundary string coordinates
corresponding to the membraneDp-brane system (i.e., when the boundary of the open
membrane is attached to p-branes) satisfy the usual Poisson algebra without any noncommutativity. By imposing further gauge conditions, it is possible to simulate a situation
where the cylindrical membrane is wrapped around a circle of vanishing radius so that the
open membrane passes over to an open string. The boundary conditions of the membrane
reduce to the well-known Neumann boundary conditions of the string in the conformal
gauge, just as the membrane metric reduces to the conformal metric of the Polyakov string.
Next, the interacting membrane in the presence of a constant three-form tensor potential
is discussed. Proceeding in a gauge-independent manner, it is shown that, contrary to the
free theory, the boundary string coordinates must be non-commutative. This is shown from
certain algebraic conditions. However, in contrast to the string case where it was possible
to solve these equations [11], here an explicit solution is prevented from the non-linear
structure. Nevertheless, by passing to the low-energy limit (wrapping the membrane on a
circle of vanishingly small radius), the explicit form of the non-commutativity in an open
string, whose end points are attached to a D-brane, are reproduced.
The paper is organised as follows: in Section 2 the free NG membrane is discussed
and the form of the quasi-orthonormal gauge conditions, which act as the analogue of the
orthonormal gauge conditions in the NG string [12], is derived. The role of the boundary
conditions in maintaining stability of the membrane is discussed. The free Polyakov
membrane is considered in Section 3, where its detailed constrained Hamiltonian account
is given. The complete form of the energymomentum tensor is derived. All components
of this tensor are written as a linear combination of the constraints. This is a generalization
182
of the string case since even though Weyl symmetry is absent in the membrane, the energy
momentum tensor has a (weakly) vanishing trace; namely, it vanishes only on the constraint
shell. The brackets for the free theory with a cylindrical topology for the membrane,
computed in Section 4, yield the expected Poisson algebra without any non-commutativity.
The low-energy limit where the membrane is approximated by the string, is discussed
in Section 5. Section 6 gives an analysis of the interacting theory. General algebraic
requirements enforce a non-commutativity of the boundary coordinates of the membrane,
which are attached to the p-branes. No gauge fixing or approximation is needed to reveal
this non-commutativity. The explicit structure of the algebra is once again computed in the
low-energy approximation, when the result agrees with the conformal-gauge expression
for the non-commutativity among the coordinates of the end points of the string attached
to D-branes. Concluding remarks are given in Section 7. Appendix A, summarizing the
basic results of our earlier paper [11] on strings, has been included for easy comparison
with the membrane analysis.
2. The free NambuGoto membrane

A dynamical membrane moving in D 1 spatial dimensions sweeps out a threedimensional world-volume in D-dimensional spacetime. We use a metric with signature
(, +, +, . . . , +) in the target space whose indices are , = 0, 1, 2, . . . , (D 1). We
can locally choose a set of three coordinates i , i = 0, 1, 2, on the world-volume to
parameterize it. We shall sometime use the notation = 0 and the indices a, b, . . . to
describe spatial coordinates a , a = 1, 2, on the membrane world-volume. In such a
coordinate system, the motion of the membrane through spacetime is described by a set of
D functions X ( 0 , 1 , 2 ) which are the membrane coordinates in the target space.
Although we are going to study the non-commutativity through the Polyakov action,
we find it convenient to briefly discuss the NG action also. The NG analysis will be just an
extension of the string case, considered in [12]. The NG action for a membrane moving in
flat spacetime is given by the integrated proper volume swept out by the membrane:

3
d h d 3 LNG X , i X ,
SNG = T
(1)
where T is a constant which can be interpreted as the membrane tension and h = det hij
with
hij = i X j X
(2)
being the induced metric on (2 + 1)-dimensional world-volume, which is nothing but

the pullback of the flat spacetime metric on this three-dimensional sub-manifold. This
induced metric, however, does not have the status of an independent field in the worldvolume; it is rather
determined through the embedding fields X . The Lagrangian density
is LNG = T h. The EulerLagrange equation is given by

i h hij j X = 0,
(3)
while the boundary conditions are given by

Pa = T h a X = 0,
183
(4)
where
Pi =
LNG
= T h i X
(i X )
(5)
and represents the boundary. The components P0 are the canonical momenta
conjugate to X . Using this, the EulerLagrange equation (3) can be rewritten as
0 + a P a = 0.
(6)
It can be seen easily that the theory admits the following primary constraints:
2 + T 2 h 0,
a a X 0,
(7)
a = 1, 2,
(8)
where 2 and h = det hab = h11 h22 (h12 )2 . These constraints are first-class
since the brackets between them vanish weakly:

(, ), (, )

= 4T 2 h22 (, )1 (
) h12 (, )2 (
) 1 (, )

) h12 (, )1 (
) 2 (, )
+ h11 (, )2 (

h22 (, )1 (
) h12 (, )2 (
) 1 (, )

h11 (, )2 (
) h12 (, )1 (
) 2 (, ) 0,

) a (, )b (
) 0,
a (, ), b (, ) = b (, )a (

) + a (
) 0.
(, ), a (, ) = 2(, )a (
(9)
The canonical world-volume energymomentum tensor density2 can be obtained
through Noether theorem:
[C ]i j =
LNG
j X i j LNG .
(i X )
(10)
In particular, [C ]0 0 = 0, [C ]0 a = a 0, [C ]a 0 = 0 and [C ]a b = 0. We notice that

the canonical Hamiltonian density, HC = [C ]0 0 , obtained by Legendre transformation,
vanishes strongly. Since the canonical energymomentum tensor density is first-class, we
may add to it a linear combination of first-class constraints with tensor-valued coefficients
to write down the total energymomentum tensor density as
i j = U i j + V ai j a 0.
The generators of and

HT = d 2 0 0 ,
translations are

Ha = d 2 0 a .
2 Note that L
NG transforms as a scalar density under diffeomorphism.
(11)
(12)
184
As one can easily see, there are no secondary constraints. The Hamiltons equation
X = {X , HT } gives
0 X = 2U 0 0 + V a0 0 a X ,
which reproduces the definition of momenta for the following choice of U 0 0 and V a0 0 :
h
hh0a
0
U 0=
(13)
,
V a0 0 =
= h ab h0b ,
2T h
h
where h ab (= hab , which is obtained by chopping off first row and first column from
hij , the inverse of hij ) is the inverse of hab in the two-dimensional subspace. The
other equation, = { , HT }, reproduces the EulerLagrange equation (3) whereas
a X = {X , Ha } gives
a X = 2U 0 a + V b0 a b X ,
which is satisfied for
U 0 a = 0,
V b0 a = b a .
(14)
Coming to the conserved Poincar generators in the target space, the translational
generator is given by

P = d 2 ,
and the angular momentum generator is given by

M = d 2 X X .
As can be easily checked, these generators generate appropriate Poincar transformations.
The above analysis can be generalized in a straightforward manner to an arbitrary p-brane.
There is an interesting implication of the boundary conditions (4). For a cylindrical
membrane with 1 [0, ], 2 [0, 2), 2 representing the compact direction, the
boundary condition is written as

P1 1 =0, = T h 1 X 1 =0, = 0.
Squaring the above equation, we get

hh11 1 =0, = h00 h22 (h02 )2 1 =0, = 0,
which implies
h00 | 1 =0,

(h02 )2
=
.
h22 1 =0,
(15)
(16)
However, h22 is strictly positive and cannot vanish at the boundary in order to prevent it
2
from collapsing to a point as the length of the boundary is given by 0
h22 d 2 . This
indicates that

X 2 1 =0, = h00 | 1 =0, 0
185
so that the points on the boundary move along either a space-like or light-like trajectory.
If we now demand that the speed of these boundary points should not exceed the speed of
light then we must have h02 | 1 =0, = 0 in Eq. (16) so that

X 2 1 =0, = h00 | 1 =0, = 0.
Therefore the boundary points move with the speed of light which is a direct generalization
of the string case where a similar result holds. For a square membrane with 1 , 2 [0, ],
the boundary conditions (4) are written as

P1 1 =0, = T h 1 X 1 =0, = 0,

P2 2 =0, = T h 2 X 2 =0, = 0.
Therefore, in addition to Eq. (15), we also have

hh22 2 =0, = h00 h11 (h01 )2 2 =0, = 0.
Proceeding just as in the case of cylindrical membrane, we find that we must have
h02 | 1 =0, = 0 and h01 | 2 =0, = 0 so that

= 0 = X 2 2
,
X 2 1
=0,
=0,
which shows that the boundary points move with the speed of light. Also, since h0a 0 at
the boundary, for both the cylindrical or square topology, it implies that the vector 0 X
is not only null, but also orthogonal to all directions tangent to the membrane worldvolume. Hence, the boundary points move with the speed of light, perpendicularly to the
membrane. This peculiar motion is exactly reminiscent of the string case. The tension in
the free membrane would cause it to collapse. This is prevented by the angular momentum
generated by the boundary motion, just as the collapse of the free string is thwarted by a
similar motion of the string end points [14].
2.1. Quasi-orthonormal gauge fixing conditions
As we shall see now, the membrane case, or any p-brane with p > 1 for that matter,
involves some subtle issues. The first step is to provide a set of complete gauge fixing
conditions. Taking a cue from the previous analysis we would like to generalize the
condition h0a 0, so that it holds everywhere, instead of just at the boundary. This is
also quite similar in spirit to what is done for implementing the orthonormal gauge in the
string case. Indeed, following the string analysis of [12], we first impose the following
gauge fixing conditions:

0,
X (, )
(17)
TA

P
0,
(, )
(18)
A
where = ( 1 , 2 ) and is an arbitrary constant vector and A is taken to be the
parametric area of the membrane. For example, if the membrane is of square topology
186
with 1 , 2 [0, ], it will be 2 and for cylindrical topology with 1 [0, ], 2

[0, 2) (membrane periodic along 2 -direction), it will be 2 2 . Clearly, this parametric
area is not an invariant quantity under two-dimensional diffeomorphism. One can think
of the square or cylindrical membrane to be flat at one instant to admit a Cartesian-like
coordinate system on the membrane surface which will provide a coordinate chart for it
during its future time evolution.
Differentiating Eq. (17) with respect to and using Eq. (18), we get
.
X
TA
T
Differentiating Eq. (17) with respect to a , and Eq. (18) with respect to we get
(19)
a X 0,
(20)
0 ( ) 0.
(21)
Using Eq. (21), it follows from the form (6) of EulerLagrange equation that

a P a 0.
(22)
Upon contraction with , the boundary conditions (4) give

P a = 0.
(23)
Now we impose an additional gauge fixing condition3

ab a ( Pb ) 0.
(24)
Thus, we have from Eqs. (22) and (24) both the divergence and curl vanishing for the vector
field ( P a ) in the 2-dimensional membrane, which is also subjected to the boundary
conditions (23). We thus have
Pa = 0
a .
(25)
In view of Eq. (20), we have
T h h00 ( 0 X),
3 One can generalize this gauge fixing condition (24) for higher-dimensional hyper-membranes. Any
n-dimensional divergenceless vector field Aa , subjected to the boundary condition Aa | = 0 (just like P a
in (22) and (23)) can be expressed as Aa = abc1 cn2 b Bc1 cn2 , where Bc1 ...cn2 are the components of
an (n 2)-form. Like the KalbRamond gauge fields, these Bs have a hierarchy of gauge symmetries given

= B(n3) + dB(n4) , . . . , so on and so forth, where B(p) is a
by B B = B + dB(n3) , B(n3) B(n3)
p-form. One can, therefore, easily see that the demand Aa = 0 entails (n 1) additional constraints as there
are (n 1) independent components of B(n2) . With two gauge fixing conditions of type (17) and (18), this
gives rise to (n + 1) number of independent constraints, which exactly matches with the number of first-class
constraints of the type (7) and (8) of the theory. For the special case of n = 2, Aa = ab b B, where B is now a
pseudo-scalar. Clearly the demand Aa = 0 is equivalent to the gauge fixing condition (24). For the case n = 3,
Aa = abc b Bc so that 3-vector is expressed as a curl of another 3-vector, in a standard manner, having only two
transverse degrees of freedom; the longitudinal one having been eliminated through the above mentioned gauge
transformation.
which, using Eq. (19) gives
h00 h 1.
187
(26)
Using Eqs. (19) and (20), Eq. (25) gives h0a 0 which implies
1
.
h00
From Eqs. (26) and (27) it follows that
h0a 0,
h00
h00 + h 0.
(27)
(28)
Observe that the term quasi-orthonormality in this case means that the time-like vector 0
is orthogonal to the space-like vectors a , which follows from Eq. (27). However, the two
space-like directions 1 and 2 need not be orthogonal to each other. Also note that by
replacing , a constant number, in Eq. (17), the normalization condition (28) will
change to h00 + 2 h 0.
Using the quasi-orthonormal conditions (27) and (28), the Lagrangian density becomes
T
(h00 h).
2
The effective action thus becomes

T
eff
S =
d 3 h00 h11 h22 + (h12 )2 ,
2
LNG T h T h00
(29)
which gives the equation of motion:

0 0 X + 1 (h12 2 X h22 1 X ) + 2 (h12 1 X h11 2 X ) = 0.
(30)
Note that the quasi-orthonormal conditions (27) and (28) do not correspond to any
gauge conditions themselves as they contain time derivatives. Actually they follow as a
consequence of the conditions (17), (18) and (24) which are to be regarded as gauge fixing
conditions. These gauge conditions, when imposed, render the first-class constraints (7)
and (8) of the theory into second-class as can be seen from their non-vanishing Poissonbracket structure. Therefore, NG formalism requires the evaluation of Dirac brackets where
these constraints are implemented strongly. As we shall see subsequently, in the Polyakov
formulation the constraints (7) and (8) are not rendered into second-class and we can avoid
the detailed calculation of Dirac brackets.
It is possible to draw a parallel between the quasi-orthonormal gauge discussed here
and the usual orthonormal gauge in NG string, which is the analogue of the conformal
gauge in the Polyakov string. In the latter case the equations of motion linearize reducing
to the DAlembert equations. This is possible because the gauge choice induces a net of
coordinates that form a locally orthonormal system [15]. For the membrane, the invariances
are insufficient to make such a choice and the best that we could do was to provide a
quasi-orthonormal system. It is however amusing to note that if we forced an orthonormal
choice, so that h0a 0 is supplemented with h12 0 and h11 = h22 1, then the equation
of motion (30) indeed simplifies to the DAlembert equation. This provides an alternative
way of looking at the quasi-orthonormality.
188
If we do not impose quasi-orthonormality, it is highly non-trivial, if not totally

impossible, to express the boundary conditions (4) in terms of phase-space variables
because the canonical momentum = P0 (5), which can be re-expressed as

T h
=
a X h ab b X 0 X
h
involves a projection operator given by the expression within the parentheses in the above
equation.
The velocity terms appear both in the right of the projection operator and in
h appearing in the denominator. This makes the inversion of the above equation to
write the velocities in terms of momenta highly non-trivial. Nevertheless, all this simplifies
drastically in the quasi-orthonormal gauge to enable us to simplify the above expression to
= T 0 X
(31)
so that the boundary condition (4) is now expressible in terms of phase-space variables as

(h22 1 X h12 2 X ) 2 1 =0, = 0.
Finally we notice that the parameters U 0 0 and V a0 0 given by Eq. (13) simplify in this
gauge to
U 00 =
1
,
2T
V a0 0 = 0
(32)
while U 0 a and V b0 a given by Eq. (14) remain unchanged. Now the generators of and a
translations (12) become

1
H a = d 2 a .
HT =
(33)
d 2 ,
2T
It is straightforward to reproduce the action (29) by performing an inverse Legendre
transformation. Computing the Poisson bracket of X (, ) with the above HT , the
Hamiltons equation 0 X = {X , HT } gives Eq. (31), the definition of momenta in this
gauge. Then,

S eff = d 3 0 X d HT
just yields (29). The other equation, 0 = { , HT }, reproduces Eq. (30), which is the
EulerLagrange equation following from the effective action (29).
Notice that the values of U 0 0 and V a0 0 are gauge dependent. The particular values
given by Eq. (32) correspond to our quasi-orthonormal gauge. Had we chosen a different
gauge, we would have obtained different values for these parameters. On the contrary, the
parameters U 0 a and V b0 a are gauge independent. This is consistent with the symmetries of
the problem. There are three reparametrization invariances, so that three parameters among
these U s and V s must be gauge dependent, manifesting these symmetries. Since the
reparametrization invariances govern the time evolution of the system, the gauge dependent
parameters are given by U 0 0 and V a0 0 , while the others are gauge independent.
189
3. The free Polyakov membrane

The Polyakov action for the bosonic membrane is [13]

T
SP =
d 3 g g ij i X j X 1 ,
2
(34)
where an auxiliary metric gij on the membrane world-volume has been introduced and will
be given the status of an independent field variable in the enlarged configuration space. The
final term (1) inside the parentheses does not appear in the analogous string theory action.
A consistent set of equations can be obtained only by taking the cosmological constant to
be (1). Indeed, the equations of motion following from the action (34) but with arbitrary
cosmological constant are

i g g ij j X = 0,
(35)

1 kl
hij = gij g hkl +
(36)
2
while the boundary conditions are

a X = 0.
(37)
Eq. (36) can now be satisfied if and only if we identify gij with hij :
gij = hij i X j X ,
(38)
for the case = 1 so that the action (34) reduces to the NG action (1). The canonical
momenta corresponding to the fields X and gij are
L
= T g 0 X ,
X
L
ij
=
= 0.
g ij
(39)
(40)
Clearly, ij 0 represent primary constraints of the theory. The canonical Hamiltonian

density is
HC = 0 X L
g 2 gg 0a
T g
=

a X +
(g22 h11 + g11 h22 2g12 h12 g).
(41)
2T g
g
2g
Therefore, the total Hamiltonian is written as

HT = d 2 HC + ij ij ,
(42)
where ij are arbitrary Lagrange multipliers. Conserving the constraint 00 0 with time:

00 = 00 , HT 0,
190
we get
0.
1 2 + T 2 (g22 h11 + g11 h22 2g12 h12 g)
Similarly, conserving other primary constraints with time, we get

g
2g22 gg
2
11 2 + T 2 (g22 h11 + g11 h22 2g12 h12 )
2
4T g

g02
T g
gg22 0a
2 X
2h22 gg
11 0,
2 g a X
g
g
4g

g
3
2g11 gg
22 2 + T 2 (g22 h11 + g11 h22 2g12 h12 )
2
4T g

g01
T g
gg11 0a
1 X
2h11 gg
22 0,
2 g a X
g
g
4g

g
4
2g12 + gg
12 2 + T 2 (g22 h11 + g11 h22 2g12 h12 )
2
2T g
g02
g01
2gg12 0a
1 X +
2 X
g a X +
+
2
g
g
g

T g
+
12 0,
2h12 + gg
2g

gg 01 2
+ T 2 (g22 h11 + g11 h22 2g12 h12 g)
5
2T
g22 1 X + g12 2 X 0,

gg 02 2
6
+ T 2 (g22 h11 + g11 h22 2g12 h12 g)

2T
g11 2 X + g12 1 X 0.
(43)
(44)
(45)
(46)
(47)
(48)
The above constraints appear to have a complicated form. Also, their connection with
the constraints obtained in the NG formalism, is not particularly transparent. To bring the
constraints into a more tractable form and to illuminate this connection, it is desirable to
express them by the following combinations:
1 = T 2 0,

gg22 0a
g02
T g
g
11
2 +
22 0,
2g22 gg
1 2 g a
2 =
4T g 2
g
g
2g

g
gg11
g01
T g
1 +
11 0,
2g11 gg
22 1 2 g 0a a
3 =
2
g
2g
4T g
g

2gg12 0a
g02
g01
g
1 +
2
2g12 + gg
4 =
12 1 +
g a +
2
2
2T g
g
g
g
T g
12 0,
(49)
(50)
(51)
(52)
gg 01
1 g22 1 + g12 2 0,
2T
gg 02
1 g11 2 + g12 1 0,
6 =
2T
where
5 =
191
(53)
(54)
2 + T 2 h 0,
(55)
a a X 0,
(56)
ab gab hab 0
(57)
and = 11 22 (12 )2 . As all the constraints s appearing in Eqs. (49)(54) are

combinations of , a and ab in Eqs. (55)(57), we can treat these , a and ab
as an alternative set of secondary constraints. These constraints along with the primary
constraints ij 0 (40) constitute the complete set of constraints of the theory. This is
because the canonical Hamiltonian density (41) can be expressed as a combination of
constraints in the following manner:
gg 0a
g
T g
a
0
HC =
(58)
2T g
g
2g
and the non-vanishing Poisson brackets between the constraints of the theory are

(, ), ab (, ) 2 a b X + b a X (
),

) + hac (, )b (
)
a (, ), bc (, ) = hab (, )c (

+ b X c a X + c X b a X (
),
ab

1
(, ), cd (, ) = ca db + da cb (
),
2
(59)
where a a , while the weakly vanishing brackets are the same as given by (9). As
far as the rest of the brackets are concerned, it is trivial to see that they vanish strongly.
Thus, as it appears, none of the constraints except 0i in the set is first-class. But we
have not yet extracted the maximal number of first-class constraints from the set (40),
(55)(57) by constructing appropriate linear combinations of the constraints. However, it
is highly non-trivial to find such a linear combination in the present case as one can see
from the complicated structure of the Poisson brackets given above in (59). Nevertheless,
one can bypass such an elaborate procedure to extract the first-class constraints from the
given set by noting that the complete set of constraints can be split into two sectors.
In one sector we retain , a and 0i , which are first-class among themselves, while
the other sector contains the canonically conjugate pairs ab and ab . This allows an
iterative computation of the Dirac brackets [16]; namely, it is possible to eliminate this set
completely by calculating the Dirac brackets within this sector. The brackets of the other
constraints are now computed with respect to these Dirac brackets. Obviously , a , will
have vanishing brackets with ab , cd . Moreover, the original first-class algebra among
and a will be retained. This follows from the fact that the Dirac constraint matrix
involving ab and cd has entries only in the off-diagonal pieces, while and a have
non-vanishing contributions coming just from the bracket with one of them, i.e., cd (see
192
(59)). The Dirac brackets of and a are thus identical to their Poisson brackets, satisfying
the same algebra as in the NG case.
We are therefore left with the first-class constraints 0, a 0 and 0i 0. At this
stage, we note that the constraints 0i 0 are analogous to 0 0 in free Maxwell theory,
where 0 is canonical conjugate to A0 . Consequently, the time evolution of g0i is arbitrary
as follows from the Hamiltonian (42). Therefore, we can set
g0a = 0,
g00 = h,
(60)
as new gauge fixing conditions.4 With that (g0a , 0a ) and (g00 , 00 ) are discarded from
the phase-space. This is again analogous to the arbitrary time evolution of A0 in Maxwell
theory, where we can set A0 = 0 as a gauge fixing condition and discard the pair (A0, 0 )
from the phase-space altogether.
These gauge fixing conditions (60) are the counterpart of the quasi-orthonormal
conditions (27) and (28) in the NG case. However, unlike the NG case, these secondclass constraints (60) do not render the residual first-class constraints of the theory, viz.
0 and a 0 into second-class constraints. Therefore, they represent partial gauge
fixing conditions. This stems from the fact that g0i were still regarded as independent field
variables in the configuration space whereas gab have already been strongly identified with
hab (57). We therefore note that the calculation of the Dirac brackets is not necessary
in Polyakov formulation. This motivates us to study the non-commutativity vis--vis the
modified brackets {X , X } in the simpler Polyakov version. For that we shall first consider
the free theory in the next section.
Before we conclude this section, let us make some pertinent observations about the
structure of the symmetric form of energymomentum tensor, which is obtained by
functionally differentiating the action with respect to the metric. The various components
of this tensor are given by:
2g g 0a
g00
1
+
g
g
T00 =
a
00
2T g
T g
g 2
g 00
2
T g00
2
T g 2
2 g 01 h11 + g 02 h22 + 2g 01 g 02 h12
(61)
,
2g
g

g01
T
g
1 +
T01 =
(g02 h11 + g01 h12 )12
2T g
g
g

g01
(62)
,
g02 h12 11 g01 h11 22
2

g
g02
T
T02 =
2 +
(g01 h22 + g02 h12 )12
2T g
g
g

g02
(63)
,
g01 h12 22 g02 h22 11
2
T gab
gab
T gab
+ T ab +

(g22 11 + g11 22 2g12 12 ).
Tab =
(64)
2T g
2g
g
4 We cannot set g = 0 as it will make the metric singular. We therefore set g = h to make it match with
00
00
the corresponding condition (28) in NG case.
193
Note that unlike the case of string [11], the component T00 cannot be written in terms of
constraints of the theory. However, the other components can be expressed in terms of these
constraints, of which ab are second-class and have already been put strongly to zero by
using Dirac brackets, so that the form of T0a and Tab simplifies to
g0a
T0a =
2T g
gab
Tab =
.
2T g
g
a ,
g
However, for T00 we have to make use of the gauge conditions (60), which hold strongly
as was discussed earlier, to enable us to write
T00 =
1
.
2T
Let us now compare it with NG case. First we notice that i j appearing in Eq. (11) is
1
i j . In quasinot a tensor itself but it is a tensor density. The corresponding tensor is g
orthonormal gauge, we have
1
,
g T 0 0 = 0 0 =
2T
which reproduces the canonical Hamiltonian density (58) in this gauge. Also, in this gauge,
we have
g T 0 a = a ,
which matches with 0 a in quasi-orthonormal gauge. This also provides a direct
generalization of the string case [11]. Although, unlike the string case, the Weyl symmetry
is absent in the membrane case, we still have a vanishing trace, albeit weakly, of the
energymomentum tensor:
T ii =
1
0.
2T h
4. The brackets for a free theory

Here we consider a cylindrical topology for the membrane which is taken to be periodic
along 2 -direction, i.e., 2 [0, 2) and 1 [0, ]. Following the example of string case
[11], we write down the first version of the brackets as:

X (, ), (, ) = + 1 , 1 P 2 2 ,
(65)
194
and the other brackets vanishing.5 Here

P ( ) =
1 in( )
e
2
(66)
nZ
is the periodic delta function of period 2 which satisfies

+
d P ( )f ( ) = f ( )
(67)
for any periodic function f ( ) = f ( + 2) defined in the interval [, +]; and if, in
addition, f ( ) is taken to be an even function in the interval [, +], then the above
integral (67) reduces to
d + (, )f ( ) = f ( ),
(68)
where
+ (, ) = P ( ) + P ( + ) =
1
1
+
cos(n ) cos(n ).
(69)
n=0
This structure of the brackets is, however, consistent only with Neumann boundary
conditions along 1 -direction. On the other hand, we have a mixed boundary condition
(37) which can be expressed in terms of phase-space variables as

g22 T 1 X + g g 01 g12 T 2 X 1 =0, = 0.

(70)
We notice that in NG formulation it was necessary to fix gauge in order to express the
boundary condition in terms of phase-space variables. However, this is not the case with
Polyakov formulation since gij are taken to be independent fields. Using the strongly valid
equations (57) and the gauge fixing conditions (60), this simplifies further to

2 X 2 X 1 X 1 X 2 X 2 X 1 =0, = 0.
(71)
Although we are using the gauge (60), the non-trivial gauge generating first-class
constraints (55) and (56) will be retained in the gauge-independent analysis both here and
in the interacting case. As we see, the above boundary condition is non-trivial in nature and
involves both the 1 and 2 derivatives. But, since the coordinates and momenta are not
related at the boundary, we do not require to postulate a non-vanishing {X , X } bracket
as in the case of free string in conformal gauge [11]. Therefore, the free membrane theory,
like its string counterpart, does not exhibit non-commutativity in the boundary coordinates.
5 Note that the {X , } brackets are not affected as we implemented the second-class constraints and the
gauge fixing conditions strongly in the preceding section. They are the only surviving phase-space variables as
gij have lost their independent status.
195
5. The low-energy limit

In this section, we would like to see how the results in the free membrane theory go
over to those of free string theory in the limit of small radius for the cylindrical membrane.
The cylindrical membrane is usually taken to propagate in an 11-dimensional compactified target space R9p M p S 1 I , where M p is a p-dimensional flat Minkowski
spacetime and I is an interval with finite length. There exist at the boundaries of I two
p-branes on which an open membrane can end. And the topology of the p-branes is given
by M p S 1 . Also, the cylindrical membrane is assumed to wrap around this S 1 . The radius
of this circle is supposed to be very small so that in the low-energy limit the target space
effectively goes over to 10-dimensional R9p M p I and the cylindrical membrane
goes over to the open string.
At this stage, we choose further gauge fixing conditions:
X0 = ,
X2 = 2 R,
(72)
where we have introduced R to indicate the radius of the cylindrical membrane and X2
represents the compact dimension S 1 .6 Before choosing the gauge conditions (72), the
1
and a translations were generated by the constraints 2T
and a , respectively, just as in
the NG case (33). Now we have

(, ), X0 (, ) = 2 0 (, )+ 1 , 1 P 2 , 2 ,

2 (, ), X2 (, ) 2 R = 2 X2 (, )+ 1 , 1 P 2 , 2

R+ 1 , 1 P 2 , 2 ,
whereas

1 (, ), X0 (, ) = 1 X0 (, )+ 1 , 1 P 2 , 2 0,

1 (, ), X2 (, ) 2 R = 1 X2 (, )+ 1 , 1 P 2 , 2 0.
Thus, the (partial) gauge fixing conditions (72) take care of the world-volume diffeomorphism generated by and 2 in the sense that these constraints are rendered into secondclass while the diffeomorphism generated by 1 is still there.
Coming back to the low-energy limit, we would like to show that the 2 dependence
of all the fields except X2 itself drops out effectively in the gauge (72). To motivate it, let
us consider the case of a free massless scalar field defined on a space with one compact
dimension of ignorable size. Let the space be M p S 1 , where M p is a p-dimensional
Minkowski spacetime taken to be flat for simplicity and S 1 is a circle of radius R which
is very small. We take [0, 2) to be the angle coordinate corresponding to this circle

so that the metric is given by ds 2 = dx dx = dx dx + R 2 d 2 with ,
6 In [6], another gauge fixing condition X1 = 1 ( being the length of the cylindrical membrane) has been
used. But we notice that imposition of this gauge fixing condition would be inconsistent with the boundary
condition (71) since, for = 1, it yields a topology changing condition (cylinder sphere), R 2 | 1 =0, = 0,
which is clearly unacceptable. Therefore, the choice (72) does not allow us to choose X1 = 1 as well, which is
not needed either for our purpose.
196
ranging from 0 to p and , from 0 to (p 1). The action is

1
S =
d p x d .
2
Separating the index corresponding to the compact dimension, we rewrite it as

1
1

S =
d p x d + 2 .
2
R
Substituting the Fourier expansion
1
(x, ) =
(n) (x)ein ,
2 nZ
(n) = (n)
in the action and integrating out the compact dimension, we get

1
n2

p
(n) (n) + 2 (n) (n) .

S S =
d x
2
R
nZ
Thus the Fourier coefficients represent a whole tower of effective massive complex scalar
fields of mass n/R in a lower-dimensional non-compact spacetime. These masses are
usually of the Planck order if R is of the order of Planck length and are therefore ignored in
the low-energy regime. Equivalently, one ignores the dependence of the field . This can
also be understood from physical considerations. In the low-energy limit, the associated
wavelengths are very large as compared to R so that variation of the field along the circle
is ignorable.
Now the membrane goes over to string in the low-energy regime when the circle S 1
effectively disappears in the limit R 0. So the field theory living in the membrane
world-volume is expected to correspond to the field theory living on string world-sheet. To
verify this, let us substitute the Fourier expansion of the world-volume fields X (, 1 , 2 )
around 2 :

1
2

X(n) , 1 ein ,
X(n) = X(n)
X , 1 , 2 =
(73)
2 nZ
in the Poisson bracket (65) to find that the Fourier coefficients X(n) (, 1 ) satisfy

X(n) , 1 , (m) , 1 = nm + 1 , 1 .
(74)
As in the case of free scalar field discussed above, the Fourier coefficients X(0) (, 1 ) will
represent the effective (real) fields in the string world-sheet satisfying

X(0) , 1 , (0) , 1 = + 1 , 1 ,
(75)
which reproduces the Poisson bracket for string (see Appendix A). The sub/superscript
(0) will be dropped now onwards for convenience. Using 2 X = R2 , the boundary
condition (71) gives

1 X 1 =0, = 0
(76)
197
so that we recover the boundary condition for free string in conformal gauge (see
Appendix A).7
Now we would like to show how the gauge fixed world-volume membrane metric (60)
reduces to the world-sheet string metric in conformal gauge. For that first note that the
components of the metric tensor in a matrix form can be written as

g00 0
0
g00 g01 g02
{gij } = g01 g11 g12 = 0 h11 0 ,
0
0 R2
g02 g12 g22
where we have made use of the first gauge fixing condition in (60) and by now the strongly
valid equations (57). Clearly, this matrix becomes singular in the limit R 0 taken in
a proper mathematical sense. It must therefore correspond to a two-dimensional surface
embedded in three-dimensional world-volume. The metric corresponding to it can be easily
obtained by chopping off the last row and last column in the above three-dimensional

metric to get g000 h0 . Now, we make use of the second gauge fixing condition in (60) to
11
However, this h can be simplified further using the gauge (72) to get
replace g00 by (h).
2
2
R h11 so that the above 2 2 matrix becomes h11 R 0 and the diagonal elements get
0 1
identified up to a scale factor. It can now be put in the standard form, diag(1, 1), upto an
overall Weyl factor, by replacing the second condition in (60) by g00 = 2 h and choosing
suitably. We also notice that using h0a = 0, the NG action for the membrane becomes

SNG = T d 3 h00 h,
which using the gauge conditions (72) and integrating out 2 , reduces to the NG action for
string in orthonormal gauge:

SNG SNG
= 2RT d 2 h00 h11 .
This also shows that the string tension is T R if the original membrane tension is given
by T . Actually one takes the limit R 0 together with the membrane tension T in
such a way that their product (T R) is finite. Such a limit was earlier discussed, from other
considerations, in [17].
6. The interacting membrane

The Polyakov action for a membrane moving in the presence of a constant antisymmetric background field A is

e ij k
ij
T
3
d
g g i X j X 1 + i X j X k X A , (77)
SP =
2
3
7 Actually, we do not get (76) directly, rather it is accompanied by a pre-factor R 2 . However, this equation is
not satisfied trivially if R 0, as this limit should not be taken literally in a mathematical sense. This just means
that R should be taken to have a very small non-zero value and presumably should be of the order of Planck
length, as we have mentioned earlier.
198
where we have introduced a coupling constant e.8 The equations of motion are
e ij k
ij
i
g g j X + j X k X A = 0,
2
gij = hij i X j X .
(78)
(79)
Note that the second equation does not change from the free case (e = 0) despite the
presence of interaction term as this term is topological in nature and does not involve the
metric gij . The canonical momenta are

L
e
= T
g 0 X + ab a X b X A ,
=
(80)
2
X
L
ij =
(81)
= 0.
g ij
For convenience, we define
+ eT ab a X b X A = T g 0 X .
(82)
2
Proceeding just as in the free case, the structure of the Hamiltonian density HC and the set
, so that we are finally left with the
of constraints is obtained just by replacing
following first-class constraints:
2 + T 2 h 0,

a X 0
a
(83)
(84)
and, as argued in the free case, we adopt the same gauge fixing conditions (60).
For a cylindrical membrane periodic along 2 -direction with 1 [0, ], 2 [0, 2),
the boundary condition is given by

(85)
g 1 X + e2 X 0 X A 1 =0, = 0,
which when expressed in terms of phase-space variables looks as

g22 T 1 X g12 T 2 X + g g 01

+ e + eT 1 X 2 X A 2 X A 1 =0, = 0.
(86)
As in the free case, here also we use the strongly valid equations (57) and the gauge fixing
conditions (60) so that the above boundary condition simplifies to

T 2 X 2 X 1 X T 1 X 2 X 2 X

+ e + eT 1 X 2 X A 2 X A 1 =0, = 0.
(87)
8 As it stands, the interaction term involving the three-form field A
in (77) is not gauge invariant under the
transformation A
A + d, where is a two-form field. One can, however, make it gauge invariant by adding a
surface term 2e B, where B is a two-form undergoing the compensating gauge transformation
B B .
But, using Stokes theorem, this gets combined to a single integral over the world-volume as (A + dB) so
that (A + dB) is gauge invariant as a whole. In the action (77), A is taken to correspond to this gauge invariant
quantity by absorbing dB in A.
199
Here we notice that the above boundary condition involves both phase-space coordinates
(X , ). Using the brackets of the free theory to compute the Poisson bracket of the lefthand side of above equation with X (, ), we find that it does not vanish. The boundary
condition is therefore not compatible with the brackets of the free theory. Thus, we have to
postulate a non-vanishing {X , X } bracket.9 For that we make an ansatz:

, ) = C 1 , 1 P 2 2
X (, ), X (, ) = C (
(88)
with

C 1 , 1 = C 1 , 1
(89)
{X , }
and the
bracket is taken to be the same as in the free case, Eq. (65). At this
stage, we note that the boundary condition (87), if bracketted with X (, ), yields at the
boundary

, )
T 2 X 2 X T 2 X 2 X + e2 T 2 X 2 X A A 1 C (

+ 2T 1 X 2 X T 1 X 2 X T 2 X 1 X + e A

+ e2 T 1 X 2 X A A + A A 2 C (
, )

= e2 X A + 1 , 1 P 2 2 ,
(90)
, ) = 0.
which involves both 1 C and 2 C and leads to a contradiction if we put C (
This is another way of seeing that there must be a non-commutativity in the membrane
coordinates. However, there is no contradiction with C (
, ) = 0 provided A = 0,
thereby implying that there is no non-commutativity in the free theory.
Because of the non-linearity in the above equation, it is problematic to find an exact
solution. It should however be stressed that the above relation has been derived in a general
(gauge-independent) manner. At this point there does not seem to be any compelling reason
to choose a particular gauge to simplify this equation further to enable an exact solution.
Non-linearity would, in all probability, prevent this. This is in contrast to the string case
where the analysis naturally leads to a class of light-cone gauges where the corresponding
equation was solvable [11]. However, by taking recourse to the low-energy approximation,
we show that the results for the string case are recovered. To this end, we substitute the
expansion (73) in (88) to get

, 1 = n,m C 1 , 1 .
X(n) , 1 , X(m)
(91)
But again, as in the free case, we retain only the real fields X(0) (, 1 ) X (, 1 ) when
we consider the low-energy regime. Using the gauge fixing conditions (72), the boundary
condition (87) reduces to

(T R)1 X e A2 e2 (T R)1 X A 2 A2 1 =0, = 0.
(92)
9 In the case of free Polyakov string also, the incompatibility of the boundary condition with the basic Poisson
brackets forces us to postulate a non-vanishing {X , X }. However, in contrast to the interacting string, this
bracket vanishes in a particular gaugethe conformal gauge. (See Appendix A.)
200
(0)
Here, X and can be taken to correspond to X(0) (, 1 ) and (, 1 ), respectively.

Thus we recover the boundary condition of the string theory in conformal gauge with the
correspondence T R TS and A2 B , where TS is the effective (string) tension and
B is the 2-form background field appearing in the string theory [11]. Now taking the
Poisson bracket of the boundary condition (92) with X (, 1 ), the low-energy effective
real fields, one gets for = 2 the following differential condition satisfied by C at the
boundary

TS e2 A2 A 2 1 C 1 , 1 1 =0, = eA 2 + 1 , 1 1 =0, , (93)
which just reproduces the corresponding equation in string theory (see Appendix A, in
particular Eq. (A.30)). We, therefore, obtain the non-commutativity:

1

C 1 , 1 = NM 1 () 1 , 1 1 , 1
2

1
+ NM 1 [] 1 , 1 + 1 , 1 1 ,
2
where

N = eA 2 ,
M = TS e2 A2 A 2
(94)
with (NM 1 )() the symmetric and (NM 1 )[] the antisymmetric part of (NM 1 )
while
1

1 1 1 1
+
sin n cos n
1 , 1 =
(95)
n
n=0
being the generalized step function which satisfies

1 1 , 1 = + 1 , 1 .
It has the property

1 , 1 = 1

1 , 1 = 0
(96)
for 1 > 1 ,
for 1 < 1 .
We have analysed an open membrane, with square and cylindrical topology, ending on
p-branes. Both the free case as well as the theory where the membrane is coupled to a
background three-form potential were considered.
For the free theory, the world-volume action was taken to be either the NG type or
the Polyakov type. For the NG action, a gauge independent formulation, similar to that
adopted in [12] for the string theory, was presented. The reparametrization invariances
were manifested by the freedom in the choice of the multipliers enforcing the constraints
of the theory. The implications of the boundary conditions in preserving the stability of the
free membrane were discussed, highlighting the parallel with the string treatment. A set of
201
quasi-orthonormal gauge fixing conditions was systematically obtained, which simplified

the structure of the Hamiltonian.
A constrained analysis of the Polyakov action, contrary to the NG action, led to the
presence of second-class constraints. However, by an iterative prescription of computing
Dirac brackets, the first-class sector was identified. The Dirac brackets of this sector were
identical to the Poisson brackets and exactly matched with the involutive algebra found
in the NG case. The analogue of the quasi-orthonormal gauge was also discussed in
the Polyakov formulation. It naturally led to the choice of the metric which is used to
perform calculations in the light-front variables [13]. Moreover, in this gauge, the energy
momentum tensor was expressed as a combination of the constraints. On the constraint
shell, this tensor was seen to have a vanishing trace.
A fundamental difference of the quasi-orthonormal gauge fixing in the two cases was
pointed out. In the Polyakov case, gauge fixing entailed certain restrictions on the metric.
Since the metric is regarded as an independent field, the gauge fixing does not affect the
constraints of the theory which generate the reparametrization invariances. The discussion
was thus confined to the Poisson algebra only. A similar gauge fixing in the NG case
obviously restricts the target space coordinates. The first-class constraints get converted
into second-class ones, thereby necessitating the use of Dirac brackets. Their evaluation is
quite complicated due to non-linear terms.
Since Dirac brackets were avoided in the Polyakov formulation, we proceeded to
discuss non-commutativity only in this formulation. Also, cylindrical topology of the
membrane was considered. We stress that, contrary to standard approaches [610],
boundary conditions were not treated as primary constraints of the theory. Our approach
was in line with the previous treatment for string theory [11], which has been summarized
in the appendix. Thus, non-commutative algebra, if any, would be a manifestation of the
Poisson brackets and not Dirac brackets. The non-commutative algebra was required to
establish algebraic consistency of the boundary conditions with the basic Poisson brackets.
For the free theory it was found that there was no clash between the boundary conditions
and the Poisson brackets; hence there was no non-commutativity.
For the membrane interacting with a three-form potential a non-trivial algebraic relation
was found that revealed the occurrence of non-commutativity, independent of any gauge
choice or any approximations. Since this equation could not be solved, we passed on to its
low-energy limit. Now this limit, which takes a membrane to a string, has been known for
quite some time [18] and has been studied or exploited in several circumstances [17,19,
20]. The cylindrical membrane is assumed to wrap around a circle, whose radius is taken
to be vanishingly small. This enforces a double dimensional reduction with the elevendimensional compactified target space passing over to the ten-dimensional space while the
membrane effectively reduces to an open string. We have studied this limit in some detail
and showed how the membrane boundary conditions, action and the world-volume metric
were transformed into the corresponding expressions for the string. The equation governing
non-commutativity in the membrane was likewise shown to reduce to the string example.
Since every point in D-brane can correspond to the end points of the cylindrical membrane,
we get non-commutativity in D-brane coordinates alsoalbeit in this low-energy limit. Of
course, this feature of non-commutativity will persist even if this limit is not considered,
otherwise the basic equation (90) becomes inconsistent.
202
It might be worthwhile to pursue the connection of the low-energy limit with the noncommutativity. Indeed, instead of taking a vanishingly small radius, it is possible to retain
terms up to some orders of the radius [17]. This would presumably illuminate the nature of
non-commutativity in the membrane (where it cannot be computed exactly) vis--vis the
string.
Acknowledgements
The authors thank Pradip Mukherjee for discussions. B.C. and K.K. gratefully
acknowledge help from a stimulating discussion with Koushik Ray. R.B. thanks the Japan
Society for Promotion of Science (JSPS) for support and the members of the theory group,
KEK, for their gracious hospitality. K.K. thanks the Council of Scientific and Industrial
Research (CSIR), Government of India, for financial support.
Appendix A. Non-commutativity in open string
For an easy comparison of our results of open membrane with those of open string, we
summarize here the essential results of [11].
A.1. The free Polyakov string
The free Polyakov string action is
TS
d d g g ij i X j X , i, j = 0, 1,
SP =
(A.1)
2
where TS stands for string tension and gij , up to a Weyl factor, is the induced metric
hij = i X j X on the world-sheet. The canonical momenta are
= TS g 0 X , ij = 0.
(A.2)
It is clear that while are genuine momenta, ij 0 are the primary constraints of
the theory. To determine the secondary constraints, one can either follow the traditional
Diracs Hamiltonian approach, or just read it off from the equation obtained by varying gij
since this is basically a Lagrange multiplier. This imposes the vanishing of the symmetric
energymomentum tensor:
2 SP
TS
= TS i X j X + gij g kl k X l X 0.
Tij =
g g ij
2
Because of the Weyl invariance, the energymomentum tensor is traceless:
T i i = g ij Tij = 0
(A.3)
(A.4)
so that only two components of Tij are independent. These components, which are the
constraints of the theory, are given by
1 2TS gT 00 = 2TS T11 = 2 + TS2 h11 0,
2 g T 0 1 = 1 X 0.
(A.5)
(A.6)
203
The canonical Hamiltonian density obtained from (A.1) by a Legendre transformation is

given by
g
g01
HC = g T 0 0 =
(A.7)
1 +
2 ,
2TS g11
g11
which, as expected, turns out to be a linear combination of the constraints. The boundary
condition written in terms of phase-space variables is given by

TS 1 X + g g 01 =0, = 0,
(A.8)
where the string parameters are in the region +, 0 . This boundary
condition is incompatible with the first of the basic Poisson brackets:

X (, ), (, ) = ( ),
(A.9)

1
gij (, ), kl (, ) = ik jl + il jk ( ).
(A.10)
2
From the basic Poisson brackets, it is easy to generate a first-class (involutive) algebra:

1 ( ), 1 ( ) = 4TS2 2 ( ) + 2 ( ) 1 ( ),

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

2 ( ), 2 ( ) = 2 ( ) + 2 ( ) 1 ( ).
(A.11)
The constraints 1 and 2 generate the diffeomorphism transformations.
The boundary condition (A.8) is not a constraint in the Dirac sense, since it is applicable
only at the boundary. Thus, there has to be an appropriate modification in the Poisson
brackets, to incorporate this condition. This is not unexpected and occurs, for instance, in
the example of a free scalar field (x) in 1 + 1 dimensions, subjected to periodic boundary
condition of period, say, 2 . There the Poisson bracket between the field (t, x) and its
conjugate momentum (t, x) is given by

(t, x), (t, y) = P (x y),
(A.12)
which is obtained automatically if one starts with the canonical harmonic-oscillator algebra
for each mode in the Fourier space.
Before discussing the mixed condition (A.8), consider the simpler Neumann-type
condition. Since the string coordinates X (, ) transform as a world-sheet scalar under
its reparametrization, it is more convenient to get back to scalar field (t, x) defined
on (1 + 1)-dimensional spacetime, but with the periodic boundary condition replaced by
Neumann boundary condition,
x | =0, = 0,
(A.13)
at the end points of a 1-dimensional box of compact size, i.e., of length . Correspondingly,
the P appearing there in the Poisson bracket (A.12)consistent with periodic boundary
conditionhas to be replaced now with a suitable delta function incorporating Neumann
boundary condition, rather than periodic boundary condition.
204
Note that the following usual property of delta function is also satisfied by P (x x ):
+
dx P (x x )f (x ) = f (x)
(A.14)
for any periodic function f (x) = f (x + 2) defined in the interval [, +]. Restricting
to the case of even and odd functions, f (x) = f (x), the above integral reduces to
dx (x, x )f (x ) = f (x).
(A.15)
Since any function (x) defined in the interval [0, ] can be regarded as a part of an even
or odd function f (x) defined in the interval [, ], both (x, x ) act as delta functions
defined in half of the interval at the right, i.e., [0, ] as follows from Eq. (A.15). It is still
not clear which of these (x, x ) functions should replace P (x x ) in the Poisson
bracket relation. Now consider the Fourier decomposition of an arbitrary function f (x)
satisfying periodic boundary condition f (x) = f (x + 2):

fn einx .
f (x) =
(A.16)
nZ
Clearly,
f (0) = i

n>0
n(fn fn ),
f () = i

(1)n n(fn fn ).
n>0
Now for even and odd functions, the Fourier coefficients are related as fn = fn so
that Neumann boundary condition f (0) = f () = 0 is satisfied if and only if f (x) is
even. Therefore, one has to regard the scalar field (x) defined in the interval [0, ] and
subjected to Neumann boundary condition (A.13) as a part of an even periodic function
f+ (x) defined in the extended interval [, +]. It thus follows that the appropriate
Poisson bracket for the scalar theory is given by

(t, x), (t, x ) = + (x, x ).
It is straightforward to generalize it to the string case:

X (, ), (, ) = + (, ),
(A.17)
the Lorentz indices playing the role of isospin indices, as viewed from the world-sheet.
Observe also that the other brackets

X (, ), X (, ) = 0,
(A.18)

(, ), (, ) = 0
(A.19)
are consistent with the boundary conditions and hence remain unchanged.
The mixed condition (A.8) is compatible with the modified brackets (A.17) and (A.19),
but not with (A.18). Therefore, make an ansatz:

X (, ), X (, ) = C (, ),
(A.20)
205
where
C (, ) = C ( , ).
(A.21)
Imposing the boundary condition (A.8) on this algebra, one gets

1 C (, ) =0, = 1 C (, ) =0, = g g 01 + (, ).
(A.22)
For a restricted class of metrics that satisfy 1 gij = 0 it is possible to give a quick solution
of this equation as

C (, ) = g g 01 (, ) ( , ) .
(A.23)
This non-commutativity can be made to vanish in gauges like conformal gauge, where
g 01 = 0, thereby restoring the usual commutative structure. The essential structure of the
involutive algebra (A.11) is still preserved, only that ( ) has to be replaced by
+ (, ).
A.2. The interacting Polyakov string
The Polyakov action for a bosonic string moving in the presence of a constant
background NeveuSchwarz two-form field B is given by

TS
SP =
(A.24)
g g ij i X j X + eA ij B i X j X ,
d d
2
where a coupling constant e has been introduced. A usual canonical analysis leads to the
following set of primary first-class constraints:

1
+ eB 1 X + eB 1 X + TS2 h11 0,
gT 00 =
2
g T 0 1 = .1 X 0,
(A.25)
(A.26)
where
= TS

g 0 X + eB 1 X
(A.27)
is the momentum conjugate to X . The boundary condition written in terms of phase-space

variables is

1 X + NM 1 =0, = 0,
(A.28)
where

M = TS e2 B B ,
g01
N = + eB .
g
(A.29)
The {X , } Poissson bracket is the same as that of the free string whereas considering
the general structure (A.20) and exploiting the above boundary condition, one obtains

1 C (, )| =0, = NM 1 + (, ) =0, .
(A.30)
206
As in the free case, restricting to the class of metrics satisfying 1 gij = 0, the above
equation has a solution

1
NM 1 () (, ) ( , )
2

1
+ NM 1 [] (, ) + ( , ) 1 .
(A.31)
2
The modified algebra is gauge dependent; depends on the choice of the metric. However,
there is no choice for which the non-commutativity vanishes. To show this, note that the
origin of the non-commutativity is the presence of non-vanishing N in the boundary
condition (A.28). Vanishing N would make B and proportional which obviously
cannot happen, as the former is an antisymmetric and the latter is a symmetric tensor.
Hence, non-commutativity will persist for any choice of world-sheet metric gij . Specially
interesting are the expressions for non-commutativity at the boundaries:
C (, ) =

1
NM 1 [] ,
2

1
C (0, ) = C (, 0) = NM 1 () .
2
C (0, 0) = C (, ) =
(A.32)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
For a review of the theory of membranes, see: J. Hoppe, hep-th/0206192, and [13].
N. Seiberg, E. Witten, JHEP 9909 (1999) 032, hep-th/9908142.
E. Bergshoeff, D.S. Berman, J.P. van der Schaar, P. Sundell, Nucl. Phys. B 590 (2000) 173, hep-th/0005026.
F. Ardalan, H. Arfei, M.M. Sheikh-Jabbari, JHEP 9902 (1999) 016, hep-th/9810072.
F. Ardalan, H. Arfei, M.M. Sheikh-Jabbari, Nucl. Phys. B 576 (2000) 578, hep-th/9906161.
S. Kawamoto, N. Sasakura, JHEP 0007 (2000) 014, hep-th/0005123.
A. Das, J. Maharana, A. Melikyan, JHEP 0104 (2001) 016, hep-th/0103229.
K.-I. Tezuka, Eur. Phys. J. C 25 (2002) 465, hep-th/0201171.
C.-S. Chu, P.-M. Ho, Nucl. Phys. B 550 (1999) 151, hep-th/9812219.
J.M. Romero, J.D. Vergara, hep-th/0212035.
R. Banerjee, B. Chakraborty, S. Ghosh, Phys. Lett. B 537 (2002) 340, hep-th/0203199.
A. Hanson, T. Regge, C. Teitelboim, Constrained Hamiltonian Systems, Accademia Nazionale Dei Lincei,
Roma, 1976.
W. Taylor, Rev. Mod. Phys. 73 (2001) 419, hep-th/0101126.
S. Mandelstam, Phys. Rep. 13 (1974) 259.
C. Rebbi, Phys. Rep. 12 (1974) 1.
D.M. Gitman, I.V. Tyutin, Quantization of Fields with Constraints, Springer-Verlag, Berlin, 1990.
U. Lindstrom, Phys. Lett. B 218 (1989) 315.
M.J. Duff, P.S. Howe, T. Inami, K.S. Stelle, Phys. Lett. B 191 (1987) 70.
E. Bergshoeff, E. Sezgin, P.K. Townsend, Ann. Phys. 185 (1988) 330.
L. Smolin, Phys. Rev. D 57 (1998) 6216, hep-th/9710191.

Exceptional confinement in G(2) gauge theory

K. Holland a , P. Minkowski b , M. Pepe b , U.-J. Wiese b,1
a Department of Physics, University of California at San Diego, La Jolla, CA 92093, USA
b Institute for Theoretical Physics, Bern University, Sidlerstrasse 5, CH-3012 Bern, Switzerland
Received 25 April 2003; received in revised form 18 June 2003; accepted 27 June 2003
Abstract
We study theories with the exceptional gauge group G(2). The 14 adjoint gluons of a G(2)
and {8} under the subgroup SU(3), and hence have the color
gauge theory transform as {3}, {3}
quantum numbers of ordinary quarks, anti-quarks and gluons in QCD. Since G(2) has a trivial
center, a quark in the {7} representation of G(2) can be screened by gluons. As a result, in
G(2) YangMills theory the string between a pair of static quarks can break. In G(2) QCD there is
a hybrid consisting of one quark and three gluons. In supersymmetric G(2) YangMills theory
with a {14} Majorana gluino the chiral symmetry is Z(4) . Chiral symmetry breaking gives rise to
distinct confined phases separated by confinedconfined domain walls. A scalar Higgs field in the {7}
representation breaks G(2) to SU(3) and allows us to interpolate between theories with exceptional
and ordinary confinement. We also present strong coupling lattice calculations that reveal basic
features of G(2) confinement. Just as in QCD, where dynamical quarks break the Z(3) symmetry
explicitly, G(2) gauge theories confine even without a center. However, there is not necessarily a
deconfinement phase transition at finite temperature.
PACS: 11.15.-q; 11.15.Ha
1. Introduction
Understanding confinement and the dynamical mechanism behind it is a big challenge in
strong interaction physics. In SU(3) YangMills theory confinement is associated with the
Z(3) center of the gauge group. Since the center symmetry is unbroken at low temperatures,
an unbreakable string confines static quarks in the fundamental {3} representation to
E-mail address: pepe@itp.unibe.ch (M. Pepe).
1 On leave from MIT.
doi:10.1016/S0550-3213(03)00571-6
208
K. Holland et al. / Nuclear Physics B 668 (2003) 207236
representation. In the high-temperature deconfined phase the

static anti-quarks in the {3}
Polyakov loop [1,2] gets a non-zero expectation value and the Z(3) symmetry breaks
spontaneously. As a result, there are three distinct deconfined phases. Potential universal
behavior at the deconfinement phase transition is described by an effective 3d scalar field
theory for the Polyakov loop [3]. The center symmetry and its spontaneous breakdown
were recently reviewed in [4,5]. In full QCD the Z(3) symmetry is explicitly broken
because quarks transform non-trivially under the center. As a result, the string connecting
a quark and an anti-quark can break via the creation of dynamical quarkanti-quark
pairs. Still, color remains confined and non-Abelian charged stateslike single quarks
or gluonscannot exist.
In this article we ask how confinement arises in a theory whose gauge group has a
trivial center. The simplest group with this property is SO(3) = SU(2)/Z(2). While SO(3)
has a trivial center, its universal covering group SU(2) has the non-trivial center Z(2).
Similarly, SU(Nc )/Z(Nc ) has a trivial center and the corresponding universal covering
group SU(Nc ) has the non-trivial center Z(Nc ). When one formulates YangMills theories
on the lattice, one usually works with Wilson parallel transporters in the universal covering
group SU(Nc ). However, one can also work with parallel transporters taking values in the
group SU(Nc )/Z(Nc ). In that case, it is impossible to probe the gluon theory with static
test quarks represented by Polyakov loops in the fundamental representation of the gauge
group. Instead one is limited to purely gluonic observables. In fact, SO(3) = SU(2)/Z(2)
gauge theories have been studied in detail on the lattice [612]. One finds that lattice
artifactsnamely, center monopolesmake it difficult to approach the continuum limit
in this formulation. There is a phase transition in which the lattice theory sheds off these
artifacts, and one then expects it to be equivalent to the standard SU(2) YangMills theory
in the continuum limit. This suggests that it is best to formulate lattice gauge theories using
the universal covering group, e.g., SU(Nc ) rather than SU(Nc )/Z(Nc ), in order to avoid
these lattice artifacts. The universal covering group of SO(N) is Spin(N) which also has a
non-trivial center: Z(2) for odd N , Z(2) Z(2) for N = 4k, and Z(4) for N = 4k + 2. The
center of the group Sp(N) is Z(2) for all N . Hence, the universal covering groups of all
main sequence Lie groups have a non-trivial center. What about the exceptional groups?
Interestingly, the groups G(2), F (4), and E(8) have a trivial center and are their own
universal covering groups. The groups E(6) and E(7), on the other hand, have the nontrivial centers Z(3) and Z(2), respectively. The exceptional Lie group G(2) is the simplest
group whose universal covering group has a trivial center.
The triviality of the center has profound consequences for the way in which confinement
is realized. In particular, a static quark in the fundamental {7} representation of G(2)
can be screened by three G(2) gluons in the adjoint {14} representation. As a result,
in G(2) YangMills theory the color flux string connecting two static G(2) quarks can
break due to the creation of dynamical gluons. This phenomenon is reminiscent of full
QCD (with an SU(3) color gauge group) in which the string connecting a static quark
and anti-quark can break due to the pair creation of light dynamical quarks. Indeed, 6 of
under the SU(3) subgroup of G(2) and thus
the 14 G(2) gluons transform as {3} and {3}
qualitatively behave like dynamical quarks and anti-quarks. In particular, they explicitly
break the Z(3) center symmetry of the SU(3) subgroup down to the trivial center of G(2).
The remaining 14 6 = 8 G(2) gluons transform as {8} under the SU(3) subgroup and
209
hence resemble the ordinary gluons familiar from QCD. It should be pointed out that
despite the broken stringjust like full QCD, G(2) YangMills theory is still expected to
confine color. In particular, one does not expect colored states of single G(2) gluons in
the physical spectrum. The triviality of the center of G(2) YangMills theory also affects
the physics at high temperatures. In particular, there is not necessarily a deconfinement
phase transition, and we expect merely a crossover between a low-temperature glueball
regime and a high-temperature G(2) gluon plasma. Due to the triviality of the center,
unlike, e.g., for SU(Nc ) YangMills theory, there is no qualitative difference between the
low- and the high-temperature regimes because the Polyakov loop is no longer a good order
parameter.
It is often being asked which degrees of freedom are responsible for confinement.
Popular candidates are dense instantons, merons, Abelian monopoles and center vortices.
Center vortices (and t Hooft twist sectors) are absent in G(2) gauge theories, while the
other topological objects potentially exist, although their identification is a very subtle issue
that often involves unsatisfactory gauge fixing procedures. At strong coupling G(2) lattice
gauge theories still confine without a center. Hence, center vortices should not be necessary
to explain the absence of colored states in the physical spectrum [13]. Still, the center
plays an important role for the finite temperature deconfinement phase transition in SU(Nc )
YangMills theory, and center vortices may well be relevant for this physics. If G(2) Yang
Mills theory, indeed, has no finite temperature deconfinement phase transition, one might
argue that this is due to the absence of center vortices and twist sectors. Assuming that
they can be properly defined, Abelian monopoles are potentially present in G(2) gauge
theory, and might be responsible for the absence of colored states. On the other hand, if
despite of the existence of Abelian monopolesa deconfinement phase transition does not
exist in G(2) YangMills theory, monopoles might not be responsible for the physics of
deconfinement. In any case, quantifying these issues in a concrete way is a very difficult
task.
The exceptional confinement in G(2) gauge theory can be smoothly connected with the
usual SU(3) confinement by exploiting the Higgs mechanism. When a scalar field in the
fundamental {7} representation of G(2) picks up a vacuum expectation value, the gauge
symmetry breaks down to SU(3), and the 6 additional G(2) gluons become massive. By
progressively increasing the vacuum expectation value of the Higgs field, one can decouple
those particles, thus smoothly interpolating between G(2) and SU(3) gauge theories. In
this way, we use G(2) gauge theories as a theoretical laboratory in which the SU(3)
theories we are most interested in are embedded in an unusual environment. This provides
theoretical insight not only into the exceptional G(2) confinement, but also into the SU(3)
confinement that occurs in the Nature.
The rest of the paper is organized as follows. In Section 2 we review the center
symmetry, the construction of the Polyakov loop, and some subtle issues related to the
physics of non-Abelian gauge fields in a finite volume. Some details of periodic and Cperiodic boundary conditions are discussed in an appendix. In Section 3 we present the
basic features of the exceptional group G(2). Section 4 contains the discussion of various
field theories with gauge group G(2). As a starting point, we consider G(2) YangMills
theory, which we then break to the SU(3) subgroup using the Higgs mechanism. We then
add fermion fields in both the fundamental and the adjoint representation, thus arriving at
210
G(2) QCD and supersymmetric G(2) YangMills theory, respectively. In both cases, we
concentrate on the chiral symmetries and we discuss how they are realized at low and at
high temperature. In Section 5 we substantiate the qualitative pictures painted in Section 4
by performing strong coupling calculations in G(2) lattice gauge theory. In particular, we
show that the theory confines although there is no string tension. Finally, Section 6 contains
our conclusions.
2. Center symmetry, Polyakov loop and gauge fields in a finite volume

When defined properly, the Polyakov loop is a useful order parameter in YangMills
gauge theories with a non-trivial center symmetry, which distinguishes confinement at low
temperatures from deconfinement at high temperatures. In particular, the Polyakov loop
varies under non-trivial transformations in the center of the gauge group and it thus signals
the spontaneous breakdown of the center symmetry at high temperatures. The expectation
value of the Polyakov loop = exp(F ) measures the free energy F of an external
static test quark. In a confined phase with unbroken non-trivial center symmetry the free
energy of a static quark is infinite. Hence, = 0 and the center symmetry is unbroken.
In a deconfined phase, on the other hand, F is finite,
= 0, and the center symmetry is
spontaneously broken.
The Polyakov loop is a rather subtle observable whose definition needs special care.
In particular, it is sensitive to spatial and temporal boundary conditions. For example,
for a system of SU(3) YangMills gluons on a finite torus, the expectation value of the
Polyakov loop is always zero even in the deconfined phase [14]. This is a consequence of
the Z(3) Gauss law: a single static test quark cannot exist in a periodic volume because
its color flux cannot go to infinity and must thus end in an anti-quark. Due to the Gauss
law, a torus is always neutral. Since it always vanishes, on a finite torus the expectation
value of the Polyakov loop does not contain any useful information about confinement
or deconfinement. Still, using the Polyakov loop, one can, for example, define its finite
volume constraint effective potential, which does indeed allow one to distinguish confined
from deconfined phases.
Let us consider a non-Abelian YangMills theory with gauge group G and antiHermitean vector potential A (x). The physics is invariant under non-Abelian gauge
transformations

A (x) = (x) A (x) + (x),
(2.1)
where (x) G. We now put the system in a finite 4-dimensional rectangular spacetime
volume of size L1 L2 L3 L4 . Here Li is the extent in the spatial i-direction and L4 =
= 1/T is the extent of periodic Euclidean time which determines the inverse temperature
= 1/T . We consider periodic boundary conditions in both space and Euclidean time,
such that our 4-dimensional spacetime volume is a hyper-torus. This means that gaugeinvariant physical quantitiesbut not the gauge-dependent vector potentials themselves
are periodic functions of spacetime. The gauge fields themselves must be periodic only
up to gauge transformations, i.e.,

A (x + L e ) = (x) A (x) + (x) .
(2.2)
211
Here e is the unit-vector in the -direction and (x) is a gauge transformation that
relates the gauge field A (x + L e ), shifted by a distance L in the -direction, to the
unshifted gauge field A (x). Mathematically speaking, the (x) define a universal fiber
bundle of transition functions which glue the torus together at the boundaries. As explained
in Appendix A, the transition functions must obey the cocycle condition
(x + L e ) (x) = (x + L e ) (x)z .
(2.3)
This consistency condition contains the twist-tensor z which takes values in the center
of the gauge group.
It should be noted that the transition functions (x) are physical degrees of freedom of
the non-Abelian gauge field, just like the vector potentials A (x) themselves. In particular,
in the path integral one must also integrate over the transition functions, otherwise gaugevariant unphysical quantities like A (x) itself might also become periodic. Under general
(not necessarily periodic) gauge transformations (x) the transition functions transform
as
(x) = (x + L e ) (x)(x).
(2.4)
In lattice gauge theory the transition functions are nothing but the Wilson parallel
transporters on the links that connect two opposite sides of the periodic volume.
The twist-tensor is gauge-invariant. Hence, as was first pointed out by t Hooft [15],
non-Abelian gauge fields on a torus fall into gauge equivalence classes characterized by
the twist-tensor, which provides a gauge-invariant characterization of distinct topological
sectors. A non-trivial twist-tensor z
= 0 induces background electric or magnetic fluxes
that wrap around the torus in various directions. Interestingly, in a non-Abelian YangMills
theory there is a symmetry transformation
(x) = (x)z ,
(2.5)
which leaves the boundary condition as well as the twist-tensorand hence the topological
sectorinvariant. Here z is an element of the center of the gauge group G. This center
symmetry exists only if all fields in the theory are center-blind. This is automatically the
case for the gauge fields which transform in the adjoint representation. However, if there
are fields that transform non-trivially under the center, the center symmetry is explicitly
broken. For example, a matter field that transforms as
(x) = (x) (x),
(2.6)
under gauge transformations, obeys the boundary condition

(x + L e ) = (x) (x),
(2.7)
which is gauge-covariant, but not invariant under the center symmetry of Eq. (2.5).
Based on the previous discussion, following van Baal [16,17], we are finally ready to
attempt a first definition of the Polyakov loop

(
x ) = Tr 4 (
x , 0)P exp

dt A4 (
x , t)
(2.8)
212
which is invariant under the transformations of Eq. (2.4) only because the transition
function 4 (
x , 0) is included in its definition. Then one obtains

x , 0) P exp
(
x ) = Tr 4 (

dt A4 (
x , t)

= Tr (
x , )4 (
x , 0)(
x , 0) (
x , 0)P exp

dt A4 (
x , t) (
x , )
= (
x ).
(2.9)
Under the center symmetry transformation of Eq. (2.5) the Polyakov loop transforms as
(
x ) = (
x )z4 ,
(2.10)
and thus it provides us with an order parameter for the spontaneous breakdown of the
center symmetry. However, on a torus the expectation value of the Polyakov loop
always vanishes, simply because spontaneous symmetry breakingin the sense of a nonvanishing order parameterdoes not occur in a finite volume. Alternatively, one may say
that the expectation value of the Polyakov loop always vanishes because the presence of a
single static quark is incompatible with the Gauss law on a torus. In any case, since it is
always zero, on a finite torus the expectation value of the Polyakov loop does not contain
any information about confinement or deconfinement, or about how the center symmetry
is dynamically realized.
Still, even on a finite torus the Polyakov loop can be used to define related quantities
that indeed contain useful information about confinement versus deconfinement, and about
the realization of the center symmetry. For example, one can define the finite volume
(V = L1 L2 L3 L4 ) constraint effective potential V() of the Polyakov loop as

exp V V() =
1
DA
V

exp S[A] ,

x , 0)P exp
d x Tr 4 (

dt A4 (
x , t)
(2.11)
where S[A] is the Euclidean YangMills action. In the confined phase, the constraint
effective potential V() has its minimum at = 0, while in the deconfined phase it
has degenerate minima at
= 0 which are related to one another by center symmetry
transformations.
As an alternative to periodic boundary conditions, we now consider a spacetime
volume with C-periodic boundary conditions in the spatial directions. Thermodynamics
dictates that the boundary conditions in the Euclidean time direction remain periodic. The
expectation value of the Polyakov loop itself becomes a useful observable when C-periodic
boundary conditions are used in SU(3) YangMills theory. In that case, a spatial shift by a
distance Li is accompanied by a charge-conjugation transformation [18,19]. A single static
quark can exist in a C-periodic box because its color flux can end in a mirror anti-quark
on the other side of the boundary. As a consequence, the expectation value of the Polyakov
213
loop no longer vanishes automatically [20]. In fact, it now vanishes only if one takes the
infinite volume limit in the confined phase, while it remains non-zero in the deconfined
phase.
In a C-periodic volume the physics is periodic up to a charge-conjugation twist, i.e.,
all physical quantities are replaced by their charge-conjugates when shifted by a distance
Li in a spatial direction. Of course, the gauge fields themselves are C-periodic only up to
gauge transformations, i.e.,

A (x + Li ei ) = i (x) A (x) + i (x) .
(2.12)
Here A (x) is the charge-conjugate of the gauge field A (x). In the Euclidean time
direction we keep periodic boundary conditions, i.e.,

A (x + ei ) = 4 (x) A (x) + 4 (x) .
(2.13)
As shown in the appendix, the cocycle conditions for C-periodic boundary conditions
are given by
i (x + Lj ej )j (x) = j (x + Li ei )i (x) zij ,
i (x + e4 )4 (x) = 4 (x + Li ei )i (x)zi4 ,
(2.14)
and hence they differ from those for periodic boundary conditions. With C-periodic
boundary conditions the transition functions transform under gauge transformations as
i (x) = (x + Li ei )i (x)(x)T,
4 (x) = (x + e4 )4 (x)(x),
(2.15)
where T denotes the transpose. As we work out in the appendix, unlike for periodic
boundary conditions, there are constraints on the twist-tensor itself. First,
2 2 2
zj k zki = 1,
zij
(2.16)
and second,
2
zi4
= zj24 .
(2.17)
Interestingly, with C-periodic boundary conditions the twist-tensor is no longer invariant

against the center symmetry transformations of Eq. (2.5). One finds

= zij zi2 zj 2 ,
zij

zi4
= zi4 z4 2 .
(2.18)
These relations can be used to relate gauge-equivalent twist-tensors to one another.

First, let us assume that the center of the gauge group G is Z(Nc ) with odd Nc . This is
the case for SU(Nc ) groups with odd Nc as well as for E(6) which has the center Z(3). Of
course, the physical color gauge group SU(3) with its center Z(3) also falls in this class.
In that case, in a C-periodic volume all twist-sectors are gauge-equivalent. In particular,
=1
using Eq. (2.17) and putting the transformation parameter to z42 = zi4 one obtains zi4
2
for all i. Next, we put the transformation parameter z1 = 1 and we choose z2 = z12 such
= 1, and z2 = z

that z12
13 such that z13 = 1. Using the consistency condition Eq. (2.16)
3
2

2

2
= 1. Hence, in this case the entire
one finally obtains z23 = z21 z13 = 1 such that z23
214
= 1 is trivial. Consequently, for gauge groups with the center Z(N ) with
twist-tensor z
c
odd Nc there is only one C-periodic boundary condition. In other words, for C-periodic
boundary conditions no analog of t Hoofts electric and magnetic flux sectors existsall
twist-sectors are gauge-equivalent to the trivial one. In these cases, according to Eq. (2.17)
the twist-tensor element zi4 is independent of the spatial direction i. Furthermore, since
Nc is odd, its square-root in the center zi4 Z(Nc ) is uniquely defined (without any
sign-ambiguity). According to Eq. (2.18) it transforms as

= z z ,
zi4
(2.19)
i4 4
under center transformations. This finally allows us to write down a completely gaugeinvariant definition of the Polyakov loop in a C-periodic volume

(
x ) = Tr zi4 4 (
x , 0)P exp
dt A4 (
x , t) .
(2.20)
Unlike for periodic boundary conditions, the center transformation of the transition
function 4 (
x , 0) = 4 (
x , 0)z4 is now compensated by the variation of the square-root of
the twist-tensor from Eq. (2.19) and one obtains (
x ) = (
x ). Defined in this completely
gauge-invariant way, the expectation value of the Polyakov loop = exp(F ) indeed
determines the free energy F of a single static quark. In contrast to the periodic torus, a Cperiodic volume can contain a single static quark, because the color flux string emanating
from it can end in a charge-conjugate anti-quark on the other side of the boundary.
The groups SU(Nc ) with even Nc have a sign-ambiguity in the definition of the squareroot of a center element. In that case, the expectation value of the Polyakov loop vanishes
even in a C-periodic volume. The same is true for Spin(N)the universal covering group
of SO(N)which has the center Z(2) for odd N , Z(2) Z(2) for N = 4k, and Z(4) for
N = 4k + 2, as well as for the symplectic groups Sp(N) and the exceptional group E(7)
which both have the center Z(2). In those cases, one is limited to the construction of the
finite volume constraint effective potential. The exceptional groups G(2), F (4) and E(8)
have a trivial center. Then the Polyakov loop is not an order parameter, but it can at least
be defined without any problems even in a simple periodic volume.
Keeping in mind the subtleties discussed above, when we refer to the Polyakov loop
in the rest of this paper, strictly speaking, we mean the location of a minimum of its
constraint effective potential on the torus, or its expectation value in a C-periodic box.
Both are identical in the infinite volume limit.
3. The exceptional group G(2)

In this section we discuss some basic properties of the Lie group G(2)the simplest
among the exceptional groups G(2), F (4), E(6), E(7) and E(8)which do not fit into
the main sequences SO(N) Spin(N), SU(N) and Sp(N). While there is only one nonAbelian compact Lie algebra of rank 1namely, the one of SO(3) SU(2) = Sp(1)
there are four of rank 2. These rank-2 algebras generate the groups G(2), SO(5) Sp(2),
215
SU(3) and SO(4) SU(2) SU(2), which have 14, 10, 8 and 6 generators, respectively.
For us the group G(2) is of particular interest because it has a trivial center and is its own
universal covering group. As we will see later, this has interesting consequences for the
confinement mechanism.
It is natural to construct G(2) as a subgroup of SO(7) which has rank 3 and 21
generators. The 7 7 real orthogonal matrices of the group SO(7) have determinant
1 and obey the constraint
ab ac = bc .
(3.1)
The G(2) subgroup contains those matrices that, in addition, satisfy the cubic constraint
Tabc = Tdef da eb f c .
(3.2)
Here T is a totally anti-symmetric tensor whose non-zero elements follow by antisymmetrization from
T127 = T154 = T163 = T235 = T264 = T374 = T576 = 1.
(3.3)
The tensor T also defines the multiplication rules for octonions [21]. Eq. (3.3) implies
that Eq. (3.2) represents 7 non-trivial constraints which reduce the 21 degrees of freedom
of SO(7) to the 14 parameters of G(2). It should be noted that G(2) inherits the reality
properties of SO(7): all its representations are real.
We make the following choice for the first 8 generators of G(2) in the 7-dimensional
fundamental representation [21]

0
0
a
1
0 a 0 .
a =
(3.4)
2 0
0
0
Here a (with a {1, 2, . . . , 8}) are the usual 3 3 Gell-Mann generators of SU(3) which
indeed is a subgroup of G(2). We have chosen the standard normalization Tr a b =
Tr a b = 2ab . The representation we have chosen involves complex numbers. However,
it is unitarily equivalent to a representation that is entirely real. In the chosen basis of the
generators it is manifest that under SU(3) subgroup transformations the 7-dimensional
representation decomposes into
{1}.
{7} = {3} {3}
(3.5)
Since G(2) has rank 2, only two generators can be diagonalized simultaneously. In our
choice of basis these are the SU(3) subgroup generators 3 and 8 . Consequently, just as
for SU(3), the weight diagrams of G(2) representations can be drawn in a 2-dimensional
plane. For example, the weight diagram of the fundamental representation is shown in
Fig. 1. One notes that it is indeed a superposition of the weight diagrams of a {3},
and {1} in SU(3). Since all G(2) representation are real, the {7} representation is
{3}
equivalent to its complex conjugate. As a consequence, G(2) quarks and anti-quarks
are indistinguishable. In particular, a G(2) quark {7} consists of an SU(3) quark {3},
and an SU(3) singlet {1}. It should be noted that the {3} {3}
an SU(3) anti-quark {3}

contained in the {7} of G(2) corresponds to a real reducible 6-dimensional representation
of SU(3).
216
Fig. 1. The weight diagram of the 7-dimensional fundamental representation of G(2) (rescaled by a factor
2).
As usual,
1
T + = (1 + i2 ) = |1 2| |5 4|,
2
1
T = (1 i2 ) = |2 1| |4 5|,
2
1
U + = (4 + i5 ) = |2 3| |6 5|,
2
1
U = (4 i5 ) = |3 2| |5 6|,
2
1
V + = (4 + i6 ) = |1 3| |6 4|,
2
1
V = (6 i4 ) = |3 1| |4 6|,
2
(3.6)
define SU(3) shift operations between the different states |1 , |2 , . . . , |7 in the fundamental representation. The remaining 6 generators of G(2) also define shifts
1
X+ = (9 + i10 ) = |2 4| |1 5| 2 |7 3| 2 |6 7|,
2
1
X = (9 i10 ) = |4 2| |5 1| 2 |3 7| 2 |7 6|,
2
1
+
Y = (11 + i12 ) = |6 1| |4 3| 2 |2 7| 2 |7 5|,
2
Y = (11 i12 ) = |1 6| |3 4| 2 |7 2| 2 |5 7|,

2
1
Z + = (13 + i14 ) = |3 5| |2 6| 2 |7 1| 2 |4 7|,
2
Fig. 2. The weight diagram of the 14-dimensional adjoint representation of G(2) (rescaled by a factor
217
2).
1
Z = (13 i14 ) = |5 3| |6 2| 2 |1 7| 2 |7 4|.
(3.7)
2
The generators themselves transform under the 14-dimensional adjoint representation of
G(2) whose weight diagram is shown in Fig. 2. From this diagram one sees that under an
SU(3) subgroup transformation the adjoint representation of G(2) decomposes into
{14} = {8} {3} {3}.
(3.8)
This implies that G(2) gluons {14} consist of the usual SU(3) gluons {8} as well as of 6
additional gluons with SU(3) quark and anti-quark color quantum numbers {3} and {3}.
Let us now discuss the center of G(2). It is interesting to note that the maximal Abelian
(Cartan) subgroup of both G(2) and SU(3) is U (1)2 which must contain the center in both
cases. Since G(2) contains SU(3) as a subgroup its center cannot be bigger than Z(3)
(the center of SU(3)) because the potential center elements of G(2) must commute with
all G(2) matrices (not just with the elements of the SU(3) subgroup). In the fundamental
representation of G(2) the center elements of the SU(3) subgroup are given by

z1 0 0
Z = 0 z 1 0 .
(3.9)
0
0 1
where 1 is the 3 3 unit matrix and z {1, exp(2i/3} is an element of Z(3). By
construction, the three 7 7 matrices Z commute with the 8 generators of the SU(3)
subgroup of G(2). However, an explicit calculation shows that this is not the case for
the remaining 6 generators. Consequently, the center of G(2) is trivial and contains only
the identity. The above argument applies to any representation of G(2). In other words,
the universal covering group of G(2) is G(2) itself and still it has a trivial center. As we
will see, this has drastic consequences for confinement. In particular, the string between
static G(2) quarks can break already in the pure gauge theory through the creation of
dynamical gluons.
In SU(3) the non-trivial center Z(3) gives rise to the concept of triality. For example,
the trivial representation {1} and the adjoint representation {8} of SU(3) have trivial triality,
218
have non-trivial opposite trialities.

while the fundamental {3} and anti-fundamental {3}
Since its center is trivial, the concept of triality does not extend to G(2). In particular, as
one can see from Eqs. (3.5), and (3.8), G(2) representations decompose into mixtures of
SU(3) representations with different trialities. This has interesting consequences for the
results of G(2) tensor decompositions [22]. For example, in contrast to the SU(3) case, the
product of two fundamental representations
{7} {7} = {1} {7} {14} {27},
(3.10)
contains both the trivial and the adjoint representation. The {1} and {27} representations are
symmetric under the exchange of the two {7} representations, while {7} and {14} are antisymmetric. As a result of Eq. (3.10), already two G(2) quarks can form a color-singlet.
Just as for SU(3), three G(2) quarks can form a color-singlet baryon because
{7} {7} {7} = {1} 4{7} 2{14} 3{27} 2{64} {77}.
(3.11)
Another interesting example is

{14} {14} {14}
= {1} {7} 5{14} 3{27} 2{64} 4{77} 3{77}
{182} 3{189} {273} 2{448}.
(3.12)
Note that there are two inequivalent 77-dimensional G(2) representations: {77} and {77 }.
As a consequence of the absence of triality, the decomposition of the tensor product of
three adjoint representations contains the fundamental representation. This means that
three G(2) gluons G can screen a single G(2) quark q, and thus a color-singlet
hybrid qGGG can be formed. Later we will also need the results for further tensor product
decompositions, two of which are listed here
{7} {14} = {7} {27} {64},
{14} {14} = {1} {14} {27} {77} {77 }.
(3.13)
It is also interesting to consider the homotopy groups related to G(2) because this tells
us what kind of topological excitations can arise. As for SU(3), the third homotopy group
of G(2) is

3 G(2) = Z.
(3.14)
Hence, there are G(2) instantons of any additive integer topological charge and,
consequently, also a -vacuum angle. Another homotopy group of interest is

2 G(2)/U (1)2 = 1 U (1)2 = Z2 .

(3.15)
Again, this is just like for SU(3). Physically, this means that t HooftPolyakov monopoles
with two kinds of magnetic charge show up when G(2) is broken to its maximal Abelian
(Cartan) subgroup U (1)2 . For SU(3) with center Z(3) the homotopy

1 SU(3)/Z(3) = 0 Z(3) = Z(3)

(3.16)
219
implies that the pure gauge theory has non-trivial twist-sectors. Interestingly, in contrast to
SU(3), for G(2) which has a trivial center I = {1} the first homotopy group

1 G(2)/I = 0 [I ] = {0}
(3.17)
is trivial. Hence, even in the pure gauge theory non-trivial twist-sectors do not exit.
4. G(2) gauge theories

In this section we discuss various theories with gauge group G(2). We start with pure
YangMills theory and then add charged matter fields in various representations. For
example, we consider a scalar Higgs field in the fundamental representation which breaks
G(2) down to SU(3). By varying the vacuum expectation value of the Higgs field one can
interpolate between a G(2) and an SU(3) gauge theory. We also add Majorana quarks
first in the fundamental {7} representation and then also in the adjoint {14} representation.
The former theory is closely related to SU(3) QCD, while the latter corresponds to N = 1
supersymmetric G(2) YangMills theory.
4.1. G(2) YangMills theory
Let us first consider the simplest G(2) gauge theoryG(2) YangMills theory. Since
G(2) has 14 generators there are 14 gluons. Under the subgroup SU(3) 8 of them
transform as ordinary gluons, i.e., as an {8} of SU(3). The remaining 6 G(2) gauge bosons
i.e., they have the color quantum numbers of ordinary quarks
break-up into {3} and {3},
and anti-quarks. Of course, in contrast to real quarks, these objects are bosons with spin 1.
Still, these additional 6 gluons have somewhat similar effects as quarks in full QCD. In
particular, they explicitly break the Z(3) center symmetry of SU(3) and make the center
symmetry of G(2) YangMills theory trivial. The Lagrangian for G(2) YangMills theory
takes the standard form
LYM [A] =
1
Tr F F ,
2g 2
(4.1)
where the field strength

F = A A + [A , A ],
(4.2)
is derived from the vector potential

1
A (x) = igAa (x) a .
2
The Lagrangian is invariant under non-Abelian gauge transformations
A = (A + ) ,
(4.3)
(4.4)
where (x) G(2). Like all non-Abelian pure gauge theories, G(2) YangMills theory is
asymptotically free. Complementary to this, at low energies one expects confinement.
220
However, in contrast to SU(3) YangMills theory, the triviality of the G(2) center has far
reaching consequences for how confinement is realized. In particular, as we have already
seen in Eq. (3.12), an external static quark in the fundamental {7} representation can
be screened by at least three gluons. Hence, via creation of dynamical gluons the
confining string connecting two static G(2) quarks can break and the potential flattens
off. Hence, the string tensionas the ultimate slope of the heavy quark potential at
distance R vanishes. Thus, in contrast to SU(3) YangMills theory where gluons
cannot screen quarks, there is a more subtle form of confinement in G(2) YangMills
theory, very much like the confinement in SU(3) QCD. In the QCD case screening arises
due to dynamical quarkanti-quark pair creation, and again the confining string breaks.
Thus, G(2) pure YangMills theory provides us with a suitable theoretical laboratory
to investigate confinement without facing additional complications due to dynamical
fermions. In the next section the issue of G(2) string breaking will be studied in the strong
coupling limit of lattice gauge theory.
Of course, unless one proves confinement analytically, one cannot be sure that QCD,
G(2) YangMills theory, or any other gauge theory is indeed in the confined phase in the
continuum limit. Based on general wisdom, one would certainly expect that G(2) gauge
theory confines color in the same way as QCD. In particular, we do not expect it to be in
a massless non-Abelian Coulomb phase. An order parameter that distinguishes between a
confined phase (without a string tension, however, with color screening) and a Coulomb
phase has been constructed by Fredenhagen and Marcu [23]. In the next section we will
show that G(2) lattice YangMills theory is indeed in the confined phase in the strong
coupling limit.
Due to the triviality of the center one also expects unusual behavior of G(2) Yang
Mills theory at finite temperature. In SU(Nc ) YangMills theory there is a deconfinement
phase transition at finite temperature where the Z(Nc ) center symmetry gets spontaneously
broken. For two colors (Nc = 2) the deconfinement phase transition is second order [24
28] and belongs to the universality class of the 3d Ising model [29,30]. For Nc = 3, on the
other hand, the phase transition is first order [3137] and the bulk physics is not universal.
This is consistent with what one expects based on the behavior of the 3d 3-state Potts
model [3842]. The high-temperature deconfined phase of SU(Nc ) YangMills theory is
characterized by a non-zero value of the Polyakov loop order parameter and by a vanishing
string tension. On the other hand, in the low-temperature confined phase, the Polyakov loop
vanishes and the string tension is non-zero. As we have seen before, already in the confined
phase of G(2) YangMills theory the string tension is zero. Since for G(2) the center is
trivial, the Polyakov loop no longer vanishes in the confined phase and it is hence no
longer an order parameter for deconfinement. As a result, for G(2) there is no compelling
argument for a phase transition at finite temperature. In particular, a second order phase
transition is practically excluded due to the absence of a symmetry that could break
spontaneously. Even without an underlying symmetry, a second order phase transition
can occur as an endpoint of a line of first order transitions. However, these particular
cases require fine-tuning of some parameter and can thus be practically excluded in G(2)
YangMills theory. On the other hand, one cannot rule out a first order phase transition
because this does not require spontaneous symmetry breaking. Since the deconfinement
phase transition in SU(3) YangMills theory is already rather weakly first order, we expect
221
the G(2) YangMills theory to have only a crossover from a low- to a high-temperature
regime. In QCD dynamical quarks also break the Z(3) center symmetry explicitly. As
the quark masses are decreased starting from infinity, the first order phase transition of
the pure gauge theory persists until it terminates at a critical point and then turns into a
crossover [43,44]. Of course, in the chiral limit there is an exact chiral symmetry that is
spontaneously broken at low and restored at high temperatures. Hence, one expects a finite
temperature chiral phase transition which should be second order for two and first order for
three massless flavors [45]. A second order chiral phase transition will be washed out to
a crossover once non-zero quark masses are included. In G(2) YangMills theory there is
no chiral symmetry that could provide us with an order parameter for a finite temperature
phase transition.
4.2. G(2) gauge-Higgs model
In the next step we add a Higgs field in the fundamental {7} representation in order
to break G(2) spontaneously down to SU(3). Then 6 of the 14 G(2) gluons pick up a
mass proportional to the vacuum value v of the Higgs field, while the remaining 8 SU(3)
gluons are unaffected by the Higgs mechanism and are confined inside glueballs. For large
v the theory thus reduces to SU(3) YangMills theory. For small v (on the order of QCD ),
on the other hand, the additional G(2) gluons are light and participate in the dynamics.
Finally, for v = 0 the Higgs mechanism disappears and we arrive at G(2) YangMills
theory. Hence, by varying v one can interpolate smoothly between G(2) and SU(3) Yang
Mills theory and connect the exceptional G(2) confinement with the usual confinement in
SU(3).
The Lagrangian of the G(2) gauge-Higgs model is given by
1
LGH [A, ] = LYM [A] + D D + V ().
2
(4.5)
Here (x) = ( 1 (x), 2 (x), . . . , 7 (x)) is the real-valued Higgs field,

D = ( + A ),
is the covariant derivative and
2

V () = 2 v 2
(4.6)
(4.7)
is the scalar potential. We have seen in Eq. (3.11) that the tensor product {7} {7} {7}
contains a singlet. Hence, one might also expect a cubic term Tabc a b c in the
Lagrangian. However, due to the anti-symmetry of the tensor T such a term vanishes.
The product {7} {7} {7} {7} contains four singlets. One corresponds to v 2 2 and
one to 4 . The other two again vanish due to anti-symmetry. Hence, the scalar potential
from above is the most general one consistent with G(2) symmetry and perturbative
renormalizability.
Let us first consider the ungauged Higgs model with the Lagrangian
1
LH [] = + V ().
2
(4.8)
222
This theory has even an enlarged global SO(7) symmetry which is spontaneously broken
to SO(6). Due to Goldstones theorem there are 21 15 = 6 massless bosons and one
Higgs particle of mass squared MH2 = 8v 2 . When we now gauge only the G(2) subgroup
of SO(7) we break the global SO(7) symmetry explicitly. As a result, the previously
intact global SO(6) SU(4) symmetry turns into a local SU(3) symmetry. Hence, a
Higgs in the {7} representation of G(2) breaks the gauge symmetry down to SU(3).
The 6 massless Goldstone bosons are eaten and become the longitudinal components
of G(2) gluons which pick up a mass MG = gv. The remaining 8 gluons are those
familiar from SU(3) YangMills theory. Choosing the vacuum value of the Higgs field as
(x) = (0, 0, 0, 0, 0, 0, v), the unbroken SU(3) invariance can be explicitly verified using
Eqs. (3.4) and (3.7).
It is interesting to compare this situation with what happens in the standard model.
Before gauge interactions are switched on, the standard model Higgs field can be viewed as
a vector in the {4} representation of SO(4) SU(2)L SU(2)R . When it picks up a vacuum
expectation value this global symmetry is spontaneously broken to SO(3) SU(2)L=R and
there are 6 3 = 3 massless Goldstone bosons. When one gauges only the SU(2)L U (1)Y
subgroup of SU(2)L SU(2)R one again breaks the global SO(4) symmetry explicitly. As
a result, the previously intact global SO(3) SU(2)L=R symmetry turns into the local
U (1)L=R = U (1)em symmetry of electromagnetism. In this case, the 3 Goldstone bosons,
of course, become the longitudinal components of the massive bosons W and Z 0 .
With the Higgs mechanism in place, we can think of the G(2) model from above
as an SU(3) gauge theory with 6 additional vector bosons of mass MG in the {3} and
representation and a scalar Higgs boson with mass MH as a {1} of SU(3). Just like
{3}
representation explicitly
dynamical quarks in QCD, the massive gluons in the {3} and {3}
break the center of SU(3). As a result, the confining string connecting a static quarkantiquark pair can break by the creation of massive G(2) gluons. As the mass of these
gluons increases with v, the distance at which the string breaks becomes larger. In
the limit v the additional gluons are removed from the theory, the Z(3) center
symmetry is restored, and the unbreakable string of SU(3) YangMills theory emerges.
Using the Higgs mechanism to interpolate between SU(3) and G(2) YangMills theory,
we again consider the issue of the deconfinement phase transition. In the SU(3) theory this
transition is weakly first order. As the mass of the 6 additional G(2) gluons is decreased, the
Z(3) center symmetry of SU(3) is explicitly broken and the phase transition is weakened.
Qualitatively, we expect the heavy gluons to play a similar role as heavy quarks in SU(3)
QCD. Hence, we expect the first order deconfinement phase transition line to terminate at
a critical endpoint before the additional G(2) gluons have become massless [43,44]. In
that case, the pure G(2) YangMills theory should have no deconfinement phase transition,
but merely a crossover.
4.3. G(2) QCD
Let us now consider G(2) gauge theory with Nf flavors of fermions. As before, we will
use the Higgs mechanism induced by a scalar field in the {7} representation to interpolate
between G(2) and SU(3) QCD. We introduce G(2) quarks as Majorana fermions in
the fundamental representation. Since all G(2) representations are real, a Dirac fermion
223
simply represents a pair of Majorana fermions. Hence, it is most natural to work with
Majorana quarks as fundamental objects. Under SU(3) subgroup transformations a {7}
of G(2) decomposes into {3}{3}{1}.

Hence, when G(2) is broken down to SU(3), a {7}
as
Majorana quark of G(2) turns into an ordinary Dirac quark {3} and its anti-quark {3}
well as a color singlet Majorana fermion that does not participate in the strong interactions.
The G(2) Majorana quark spinor can be written as

T

,
= C
(4.9)
where is an SU(3) Dirac quark spinor, is the color singlet Majorana fermion, and
C is the charge-conjugation matrix. The anti-quark spinor is related to by chargeconjugation

, T C 1 , T C 1 .
=
(4.10)
The Lagrangian of G(2) QCD takes the form
] = LYM [A] + 1
( + A ).
LQCD [A, ,
(4.11)
2
In G(2) gauge theory, quark masses arise from Yukawa couplings to the scalar field as
well as from Majorana mass terms. For simplicity, in what follows we consider massless
quarks only.
4.3.1. The Nf = 1 case
As a first step we consider a single flavorsay, the u-quark. Ordinary Nf = 1 SU(3)
QCD has a U (1)B symmetryjust baryon number which is unbroken. In particular, there
are no massless Goldstone bosons and the theory has a mass-gap. Color singlet states
include uu mesons with a valance quark and anti-quark as well as a uuu baryon ++
with three valance quarks. The lightest particle in this theory is presumably a vector-meson
similar to the physical -meson. Like in ordinary QCD, the pseudo-scalar meson gets
its mass via the anomaly from topological charge fluctuations. Only in the large Nc limit
it becomes a Goldstone boson. For Nc = 3 it may or may not be lighter than the vectormeson.
As we have seen before, in G(2) gauge theory quarks and anti-quarks are
indistinguishable. Consequently, the U (1)L=R = U (1)B baryon number symmetry of
SU(3) QCD is reduced to a Z(2)B symmetry. One can only distinguish between states
with an even and odd number of quark constituents. In particular, Eq. (3.12) implies that
one can construct a colorless state uGGG with one G(2) quark screened by three G(2)
gluons. This state mixes with other states containing an odd number of quarkse.g.,
with the usual uuu statesto form the G(2) baryon. In contrast to SU(3) QCD, two
G(2) baryons (which are odd under Z(2)B ) can annihilate into mesons. When one
uses the Higgs mechanism to break G(2) to SU(3), one can remove the 6 additional G(2)
gluons by increasing the vacuum value v. As a consequence, the states uGGG become
heavy and can no longer mix with uuu. As a result, the standard U (1)B baryon number
symmetry of SU(3) QCD emerges as an approximate symmetry. As long as v remains
224
finite, the heavy G(2) gluons mediate weak baryon number violating processes. Only in
the limit v U (1)B becomes an exact symmetry.
4.3.2. The Nf 2 case
Let us first remind ourselves of standard two flavor QCD with SU(3) color gauge group.
The chiral symmetry then is SU(2)L SU(2)R U (1)B which is spontaneously broken
to SU(2)L=R U (1)B . Consequently, there are 7 4 = 3 massless Goldstone pions + ,
0 and . For a G(2) Majorana quark left- and right-handed components cannot be
rotated independently by unitary transformations L SU(2)L and R SU(2)R . In fact,
the Majorana condition requires L = R . Hence, the chiral symmetry of Nf = 2 G(2)
QCD is SU(2)L=R Z(2)B . Note that in the same way U (1)B = U (1)L=R is reduced
to U (1)L=R =R = Z(2)B . The reduced chiral symmetry of G(2) QCD is expected to
still break spontaneously to the maximal vector subgroup which is now SU(2)L=R =R
Z(2)B = SO(2)L=R Z(2)B . In this case, there are only 3 1 = 2 Goldstone bosons. They
can be identified as 0 and the linear combination of + and that is even under charge
conjugation. The mixing between + and is induced by the exchange of one of the 6
G(2) gluons that do not belong to SU(3). The linear combination of + and that is
odd under charge-conjugation has a non-zero mass and is not a Goldstone boson of G(2)
QCD. When we remove the 6 additional gluons via the Higgs mechanism, the mixing
of + and becomes weaker as v increases. Consequently, the mass splitting between
the charge-conjugation even and odd states also decreases until it ultimately vanishes at
v = . In this limit the larger chiral symmetry of SU(3) QCD emerges and we are left
with three massless pions.
It is straightforward to generalize the above discussion to arbitrary Nf 2. For SU(3)
QCD with general Nf the chiral symmetry is SU(Nf )L SU(Nf )R U (1)B which is
spontaneously broken to SU(Nf )L=R U (1)B , and there are Nf2 1 massless Goldstone
bosons. As before, the Majorana condition requires L = R . Hence, the chiral symmetry
of G(2) QCD with Nf massless Majorana quarks is SU(Nf )L=R Z(2)B , which is
expected to break spontaneously to SU(Nf )L=R =R Z(2)B = SO(Nf )L=R Z(2)B .
Then there are only Nf (Nf + 1)/2 1 Goldstone bosons. These consist of Nf 1 neutral
Goldstone bosons ( 0 and for Nf = 3) and Nf (Nf 1)/2 charge-conjugation even
0 for Nf = 3).
combinations of charged states ( + + , K + + K , K 0 + K
4.4. Supersymmetric G(2) YangMills theory
In this section we add a single flavor adjoint Majorana gluino to the pure gluon Yang
Mills theory and thus turn it into N = 1 supersymmetric YangMills theory. First, we
compare the SU(3) to the G(2) case.
Let us first describe the situation in SU(3) supersymmetric YangMills theory. Then,
in addition to the gluons, there is a color octet of Majorana gluinos. The chiral symmetry
of this theory is Z(3) Z(2)B where Z(2)B is the fermion number symmetry of the
Majorana fermion, and Z(3) is a remnant of the axial U (1)R symmetry which is broken
by the anomaly. At low temperatures the Z(3) symmetry is spontaneously broken through
the dynamical generation of a gluino condensate . Since both gluons and gluinos are
in the adjoint representation, the Z(3) center of the color gauge group is an exact symmetry
225
of supersymmetric SU(3) YangMills theory. At low temperature this discrete symmetry

is unbroken, just as in the non-supersymmetric case. As a result, external static quarks and
anti-quarks are confined to one another through an unbreakable color flux string.
As a consequence of the spontaneous breakdown of the discrete Z(3) symmetry, there
are, in fact, three different low-temperature confined phases which are distinguished by
the Z(3) phase of the gluino condensate. When such phases coexist with one another,
they are separated by confinedconfined domain walls with a non-zero interface tension.
These walls are topological defects which are characterized by the zeroth homotopy group
0 [Z(3) ] = Z(3). Based on M-theory, Witten has argued that the walls behave like Dbranes and the confining string (which behaves like a fundamental string) can end on
the walls [46]. In a field theoretical context this phenomenon has been explained in [47].
Just like other topological excitations, such as monopoles, cosmic strings and vortices, a
supersymmetric domain wall has the unbroken symmetry phase in its core. Consequently,
inside the wall (as well as inside the string) the chiral Z(3) symmetry is restored.
Interestingly, the restoration of Z(3) induces the spontaneous breakdown of the center
symmetry Z(3). Hence, the core of a supersymmetric domain wall is in the deconfined
phase. When a confining string enters the wall the color flux that it carries spreads out and
the string ends.
As one increases the temperature, a transition to a deconfined phase with restored
chiral symmetry occurs. As usual, in the deconfined phase the Z(3) center symmetry is
spontaneously broken. When a confinedconfined domain wall is heated up to the phase
transition, the deconfined phase in its core expands and forms a complete wetting layer
whose width diverges at the phase transition [47]. Due to the broken Z(3) center symmetry
there are also three distinct deconfined phases. When those coexist, they are separated by
deconfineddeconfined domain walls. As the phase transition is approached from above,
similarly, a deconfineddeconfined domain wall splits into a pair of confineddeconfined
interfaces and its core turns into a complete wetting layer of confined phase [48,49].
Let us now ask how the situation is modified for G(2) supersymmetric YangMills
theory for which the center is trivial. Interestingly, the remnant chiral symmetry is
enhanced to Z(4) Z(2)B which breaks spontaneously to Z(2)B . As a result, similar
to the SU(3) case, there are now four different low-temperature chirally broken phases
which are characterized by the Z(4) phase 1, i of the gluino condensate. Due to
the triviality of the center, we have the exceptional confinement with a breakable string
that we have already discussed in the non-supersymmetric G(2) YangMills theory. When
two distinct chirally broken phases coexist, they are again separated by a domain wall. In
contrast to the SU(3) case, G(2) strings cannot only end on such walls because they can
break and thus end anywhere.
When G(2) gluons and gluinos are heated up, their chiral symmetry is restored
in a finite temperature phase transition. In contrast to G(2) pure YangMills theory, a
phase transition must exist because there is now an exact spontaneously broken Z(4)
chiral symmetry for which the gluino condensate provides us with an order parameter.
As another consequence of the triviality of the center, there is only one high-temperature
chirally symmetric phase. In particular, deconfineddeconfined domain walls do not exist.
However, there are now two types of domain walls in the low-temperature phase. A wall
of type I separates chirally broken phases whose gluino condensates are related by a
226
Z(4) transformation 1, while for a wall of type II the phases are related by a irotation. The chiral phase transition may be first or second order. In the latter case the
low- and high-temperature phases do not coexist at the phase transition and complete
wetting does not arise. However, we find it more natural to expect a first order phase
transition. For example, the deconfinement phase transition of an ordinary SU(4) Yang
Mills theory, which has a Z(4) center symmetry, is first order [5054]. If the chiral phase
transition of G(2) supersymmetric YangMills theory is first order as well, the low- and
high-temperature phases coexist and complete wetting may arise. When a wall of type I
is heated up to the phase transition, we expect it to split into a pair of interfaces with a
complete wetting layer of chirally symmetric phase in between. It is less clear if complete
wetting would also occur for domain walls of type II. We do not enter this discussion here.
In any case, complete wetting is no longer needed for strings to end on the walls.
5. G(2) lattice gauge theory at strong coupling

In order to substantiate some of the claims made in the previous sections we now
formulate G(2) YangMills theory on the lattice and derive some analytic results in the
strong coupling limit. As usual, such results do not directly apply to the continuum limit
and should ultimately be extended by Monte Carlo simulations into the weak coupling
continuum regime. Still, assuming that there is no phase transition separating the strong
from the weak coupling regime, the strong coupling results provide insight into dynamical
behavior such as confinementwhich persists in the continuum limit. For example, for
G(2)in agreement with the expectationsthe lattice strong coupling expansion confirms
that the color flux string can break by dynamical gluon creation. In this sense, the
string tension is zero and the Wilson loop is no longer a good order parameter. In order
to characterize the phase of the theory, in particular, in order to distinguish between a
massive confinement phase like in full QCD and a massless Coulomb phase, one can use
the FredenhagenMarcu order parameter [23] which we calculate analytically at strong
coupling. Indeed, this confirms that G(2) lattice YangMills theory confines color in the
same way as SU(3) QCD.
The construction of G(2) YangMills theory on the lattice follows the standard
procedure. The link matrices Ux, G(2) are group elements in the fundamental {7}
representation, i.e., they can be chosen entirely real. The standard Wilson plaquette action
takes the usual form
1
1
Tr U = 2
Tr Ux, Ux+,
S[U ] = 2
(5.1)
Ux+ , Ux, ,
g
g x,<
where g is the bare gauge coupling. The partition function is given by

Z = DU exp S[U ] ,
(5.2)
where the measure of the path integral

DU =
dUx, ,
(5.3)
x,
G(2)
227
is a product of local Haar measures of the group G(2) for each link. By construction, both
the action and the measure are explicitly invariant under gauge transformations

Ux,
= x Ux, x+
,
with x G(2). The Wilson loop

Ux,
WC = Tr P
(5.4)
(5.5)
(x,)C
is the trace of a path ordered product of link variables along the closed path C, and its
expectation value is given by

1
WC =
(5.6)
DU WC exp S[U ] .
Z
For a rectangular path C of extent R in the spatial and T in the temporal direction the
Wilson loop

lim WC = exp V (R)T
(5.7)
T
determines the potential V (R) between static color sources at distance R. In a phase with
a linearly rising confining potential V (R) R, where is the string tension, the Wilson
loop obeys an area law. If the potential levels off at large distances, the Wilson loop obeys
a perimeter law. In the strong coupling limit of SU(3) YangMills theory the Wilson loop
indeed follows an area law. In G(2) YangMills theory, on the other hand, we expect static
quarks to be screened by gluons and hence the string to break. As a result, the static
quark potential levels off and large Wilson loops obey a perimeter law.
In the strong coupling regime g 2 1, and 1/g 2 can be used as a small expansion
parameter. The first step of the strong coupling expansion is the character expansion of the
Boltzmann factor for an individual plaquette

1
1
(U ),
c
exp 2 Tr U =
(5.8)
g
g2
where is a generic representation of the gauge group and the corresponding character
(U ) is the trace of the matrix U in that representation. The coefficients c (1/g 2 )
enter the strong coupling expansion as power-series in 1/g 2 . For example, for G(2) we
have

1
1
1
1
1
7
c{1} 2 = 1 + 4 + 6 + 8 +
+
+ ,
10
g
2g
6g
6g
12g
144g 12

1
1
1
2
5
7
1
c{7} 2 = 2 + 4 + 6 +
+
+ 12 + ,
8
10
g
g
2g
3g
12g
24g
6g

1
1
1
3
1
1
c{14} 2 = 4 + 6 + 8 + 10 + 12 + .
(5.9)
g
2g
3g
8g
4g
6g
Let us now compute the expectation value of a rectangular Wilson loop of size R T
in the strong coupling limit. In SU(Nc ) lattice YangMills theory the leading contribution
228
in the strong coupling expansion results from tiling the rectangular surface enclosed by
the Wilson loop with R T elementary plaquettes in the fundamental representation.
This gives rise to the strong coupling area law. All higher order contributions amount to
deformations of this minimal surface of plaquettes. Such contributions are also present
for G(2). However, in the G(2) case there are additional terms arising from a tube of
plaquettes along the perimeter of the Wilson loop. In fact, these contributions dominate at
large R and give rise to a strong coupling perimeter law. For small R, on the other hand,
the surface term dominates and yields a linearly rising potential at short distances. It should
be noted that tube contributions arise even in SU(Nc ) YangMills theory for Wilson loops
of adjoint charges. In that case, there is again no linearly rising confinement potential. Due
to the triviality of the center, for G(2) the perimeter law arises already for fundamental
charges.
Here we consider only the leading contribution to the strong coupling expansion. For
small R Rc the surface term dominates and gives

1 RT T
WC = 7
(5.10)
exp V (R)T ,
2
7g
which yields a linear potential

1
V (R) = log
R.
7g 2
(5.11)
In an SU(Nc ) YangMills theory the linear potential would extend to arbitrary distances
and log(1/7g 2 ) would play the role of the string tension. In the G(2) case, however, the
large R Rc behavior is dominated by the perimeter term

1 8(R+T 2)
WC = 4
(5.12)
,
7g 2
which gives rise to a flat (R-independent) potential

1
V (R) = 8 log
.
7g 2
(5.13)
At distances larger than Rc = 8 the perimeter term overwhelms the surface term and the
string breaks. Hence, there is no confinement in the sense of a non-vanishing string tension
characterizing the slope of the potential at asymptotic distances. When one pulls apart a
G(2) quark pair beyond the distance Rc , gluons pop up from the vacuum and screen
the fundamental color charges of the quarks. This is possible only because G(2) has a
trivial center. From the tensor product decomposition of Eq. (3.12) one infers that at least
three gluons (which are in the {14} of G(2)) are needed to screen a single quark (in
the {7} of G(2)).
Since its string can break, pure G(2) YangMills theory resembles full QCD. In that
case, dynamical quarkanti-quark pairs materialize from the vacuum to screen an external
static quarkanti-quark pair at large separation. Hence, also in full QCD the static quark
anti-quark potential flattens off and ultimately there is no string tension. Of course, this
does not mean that QCD does not confine. In particular, there should be no single quark
or gluon states in the physical spectrum, i.e., QCD should not be realized in a non-Abelian
229
Coulomb phase. An order parameter that distinguishes between Coulomb and confinement
phases (even if there is no string tension) has been constructed by Fredenhagen and Marcu
[23]. This order parameter can also be adapted to G(2) pure YangMills theory and can,
in fact, be evaluated in the strong coupling limit.
The FredenhagenMarcu order parameter is a ratio of two expectation values. The
numerator consists of parallel transporters along an open staple-shaped path connecting
a source and a sink of color flux and the denominator is the square root of a closed Wilson
loop
(R, T ) =

1/2
The open path symbolic object in the numerator stands for

= Tr(Ux a ) Tr a UCxy b UCxy Tr U y b ,
where
UCxy = P
Uz,
(5.14)
(5.15)
(5.16)
(z,)Cxy
is a path-ordered product of parallel transporters along the open path Cxy . This stapleshaped path of time-extent T connects the source and sink points x and y that are spatially
separated by a distance R. At these points dynamical G(2) gluons are created by
plaquette operators Ux and Uy . The factors a and b reflect the fact that gluons
transform in the adjoint representation. The closed path symbolic object in the denominator
is a Wilson loop of size R 2T in the adjoint representation.
The FredenhagenMarcu order parameter describes the creation of a pair of adjoint
dynamical charges that propagate for a time T and measures their overlap with the vacuum
in the limit R, T . In a Coulomb phase charged states exist in the physical spectrum
and are orthogonal to the vacuum. Consequently, the FredenhagenMarcu vacuum overlap
order parameter then vanishes. In a confined phase, on the other hand, the dynamical
charges are screened and (R, T ) goes to a non-zero constant for large R and T .
In the strong coupling limit the leading contribution to the numerator of the vacuum
overlap order parameter is a tube of plaquettes emanating from the source plaquette x ,
following the staple-shaped path, and ending at the sink plaquette y . This leads to a
perimeter law in the numerator. Just like the Wilson loop in the fundamental representation
that was calculated before, the adjoint Wilson loop in the denominator of the order
parameter also obeys a perimeter law. Due to the square root and the doubled temporal
extent, the perimeter behavior in the numerator and the denominator cancel exactly and
one is left with

1 8
112(1/7g 2)4(2T +R)
=
56
.
(R, T ) =
(5.17)
7g 2
2(1/7g 2 )4(2T +R2)
Here the plane of the plaquettes x and y is dual to the plane of the staple-shaped path
Cxy . Eq. (5.17) shows that we are indeed in a confined phase (without a string tension,
however, with color charge screening) and not in a non-Abelian Coulomb phase. Of course,
230
this strong coupling result does not guarantee that G(2) YangMills theory confines also
in the continuum limit, as one would naturally expect. It would be interesting to investigate
this issue in numerical simulations.
One might argue that in a pure gluon theory quarks are simply not present and can
hence not even be used as external static sources. If one wants to study confinement of
gluons without using static quarks, one can use the vacuum overlap operator also in an
SU(3) YangMills theory. In the strong coupling limit one then finds again that gluons are
in a confined phase with color screening by dynamical gluon creationand are not in a
Coulomb phase.
The FredenhagenMarcu operator of Eq. (5.15) in the adjoint representation describes
the creation of a pair of dynamical gluons and can be considered for any gauge group.
However, for G(2) there is an additional FredenhagenMarcu operator which has no analog
for gauge groups with a non-trivial center2 and which is given by

ef
= Uabx Tabc UCcdxy Tdef Uy .
(5.18)
Much like the original operator constructed by Fredenhagen and Marcu for QCD, this
operator describes the creation of a pair of fundamental charges. The group G(2) has the
particular feature that a fundamental {7} charge can be formed just out of {14} gluons.
In this case, the Wilson loop in the denominator of Eq. (5.14) is taken in the fundamental
representation. At strong coupling the corresponding order parameter is given by

168(1/7g 2)4(2T +R)
1 8
(R, T ) =
(5.19)
= 84
.
7g 2
2(1/7g 2 )4(2T +R2)
This again confirms that G(2) YangMills theory indeed confines, at least in the strong
coupling limit.
It should be noted that the FredenhagenMarcu order parameter makes sense only
at zero temperature, because it requires to take the limit T . Since G(2) has a
trivial center (and thus a vanishing string tension) there is no need for the standard finite
temperature deconfinement phase transition. In particular, there is no center symmetry that
could break spontaneously at high temperatures. Of course, this argument does not exclude
the existence of a first order phase transition at finite temperature. We find it more natural
to expect just a crossover. Again, this is an interesting point for numerical investigation. In
any case, analytic strong coupling calculations cannot answer this question.
6. Conclusions
We have compared qualitative non-perturbative features such as confinement and chiral
symmetry breaking in theories with G(2) and SU(3) gauge groups. In particular, we have
exploited the Higgs mechanism (induced by a scalar field in the {7} representation of G(2))
in order to interpolate smoothly between these two cases. We have focused on effects which
2 We thank U. Heller for drawing our attention to this operator.
231
are intimately related to the center of the gauge group, and which hence are qualitatively
different for SU(3) with center Z(3) and G(2) with a trivial center.
When all dynamical fields in an SU(3) gauge theory are center-blind (such as gluons
or gluinos which have trivial triality) the Z(3) center is an exact symmetry. Infinitely
heavy quarks with non-trivial triality can be used as external probes of the gluon dynamics
that provide information about how the Z(3) symmetry is realized. In a confined phase
with intact center symmetry the confining string connecting a static quarkanti-quark
pair is absolutely unbreakable and has a non-zero string tension that characterizes the
interaction at arbitrarily large distances. The string tension can vanish only when the center
gets spontaneously broken, which is indeed unavoidable at high temperatures. Then the
Euclidean time extent is short and the gauge field configuration becomes almost static. As
a consequence, the Polyakov loop order parameter becomes non-zero. Hence, the exact
center symmetry provides us with an argument for the existence of a deconfinement phase
transition. If the transition is second order, universality arguments suggest that it is in the
universality class of a 3d center-symmetric scalar field theory for the Polyakov loop [3].
For example, for Nc = 2 it is second order [2428] and falls in the universality class of the
3d Ising model [29,30].
Since it has a trivial center, the concept of triality does not extend to G(2).
Consequently, any infinitely heavy external source can be screened by dynamical gluons
and thus the string always breaks at large distances through the creation of dynamical
gluons. As a result, the string tension ultimately vanishes. However, a strong coupling
lattice study of the FredenhagenMarcu vacuum overlap order parameter shows that the
theory is still confiningi.e., no colored states exist in the spectrum. Confinement without
a (fundamental) string tension is indeed exceptional for a pure gauge theory. It only arises
for the exceptional Lie groups G(2), F (4) and E(8). As another consequence of the trivial
center, the G(2) Polyakov loop is no longer an order parameter. Hence, in contrast to
SU(Nc ) YangMills theory, for G(2) there is no compelling reason for a finite temperature
deconfinement phase transition. We cannot exclude a first order phase transition but we
expect only a crossover. Clearly, the triviality of the center implies less predictive power
about a possible phase transition.
Once dynamical fields with non-trivial triality (such as light quarks) are included in
an SU(3) gauge theory, they break the center symmetry explicitly. As a result, the string
connecting a static quarkanti-quark pair can now break through pair creation of dynamical
quarks and the string tension ultimately vanishes. Again, the FredenhagenMarcu order
parameter still signals confinement. In addition, the Polyakov loop is no longer a good
order parameter. From this point of view, the confinement in G(2) YangMills theory
resembles the one of SU(3) QCD, and hence, it is not so exceptional after all.
Again, by using the Higgs mechanism, we have also interpolated between G(2) and
SU(3) gauge theories with massless dynamical fermions, both in the fundamental and in
the adjoint representation. In many of these cases, there is a non-trivial chiral symmetry that
breaks spontaneously at low temperatures. Since the G(2) representations are real, we have
considered Nf flavors of Majorana fermions. The chiral symmetry then is SU(Nf )L=R
Z(2)B which breaks spontaneously to SO(Nf )L=R Z(2)B . It is interesting how this
pattern of symmetry breaking turns into the breaking of SU(Nf )L SU(Nf )R U (1)B
to SU(Nf )L=R U (1)B that occurs in QCD.
232
As we have seen, there are many interesting non-perturbative phenomena that arise
in G(2) gauge theories. Despite the fact that Nature chose not to use G(2) (at least
at presently accessible energies) it may be of theoretical interest to study G(2) gauge
theories more quantitatively. Lattice gauge theory provides us with a powerful tool for
such investigations. For example, it would be interesting to check if the strong coupling
confined phase extends to the continuum limit, by measuring the FredenhagenMarcu
order parameter in a numerical simulation. With lattice methods one can also decide if
the low- and high-temperature regimes in G(2) YangMills theory are separated by a first
order phase transition or just by a crossover.
It is also interesting to investigate YangMills theories with other gauge groups such as
Sp(N), which have a Z(2) center symmetry. If they have a second order deconfinement
phase transition, one expects it to be in the universality class of the 3d Ising model.
A numerical study of Sp(2) and Sp(3) gauge theory is presently in progress [55]. The
group Sp(2) with 10 generators is the fourth of the rank 2 Lie groups besides SO(4)
SU(2) SU(2), SU(3) and G(2). Based on its rank and its number of generators one
might expect that it should behave more like SU(3) than like SU(2) = Sp(1). Indeed, we
find that both Sp(2) and Sp(3) YangMills theory have a first order deconfinement phase
transition.
In conclusion, we have used G(2) gauge theories as a theoretical laboratory to study
SU(3) theories in an unusual environment. In particular, the embedding of SU(3) in G(2)
with its trivial center forces us to think about confinement without the luxury of the Z(3)
symmetry. As one would expect, confinement itself works perfectly well without the center
symmetry. However, in its absence we loose predictive power about a possible phase
transition at finite temperature.
Acknowledgements
We like to thank S. Caracciolo, P. de Forcrand, R. Ferrari, U. Heller, O. Jahn, J. Kuti,
A. Smilga, and P. van Baal for interesting discussions. This work is supported by the DOE
under the grant DOE-FG03-97ER40546, by the Schweizerischer Nationalfond, as well as
by the European Communitys Human Potential Program under the grant HPRN-CT-200000145 Hadrons/Lattice QCD, BBW Nr. 99.0143.
Appendix A. Transition functions, twist-tensor and consistency conditions

In this appendix we derive some relations for periodic and C-periodic boundary
conditions that are used in Section 2. First, we consider periodic boundary conditions.
Shifting the gauge field in two orthogonal directions e and e , on the one hand, one
obtains
A (x + L e + L e )

= (x + L e ) A (x + L e ) + (x + L e )

= (x + L e ) (x) A (x) + (x) (x + L e ) .
(A.1)
233
On the other hand, by performing the two shifts in the opposite order, one finds
A (x + L e + L e )

= (x + L e ) A (x + L e ) + (x + L e )

= (x + L e ) (x) A (x) + (x) (x + L e ) .
(A.2)
The two results are consistent only if the transition functions obey the cocycle condition
Eq. (2.3).
Eq. (2.4) guarantees gauge-covariance of the boundary condition, i.e.,

A (x + L e ) = (x + L e ) A (x + L e ) + (x + L e )

= (x + L e ) (x) A (x) + (x) (x + L e )

= (x) A (x) + (x) .
(A.3)
Interestingly, the gauge transformed cocycle condition takes the form

(x + L e ) (x) = (x + L e ) (x) z
(x + L e + L e ) (x + L e )(x + L e ) (x + L e ) (x)(x)
= (x + L e + L e ) (x + L e )(x + L e )

(x + L e ) (x)(x)z

(x + L e ) (x) = (x + L e ) (x)z
.
(A.4)
= z , i.e., the twist-tensor is gauge invariant.

Consequently, z
Next, we consider C-periodic boundary conditions. As before, we shift the gauge field
in two orthogonal directions. First, we pick two different spatial directions i and j , and we
obtain
A (x + Li ei + Lj ej )

= i (x + Lj ej ) A (x + Lj ej ) + i (x + Lj ej )

= i (x + Lj ej )j (x) A (x) + j (x)T i (x + Lj ej ) .
(A.5)
Performing the two shifts in the opposite order, one now finds
A (x + Lj ej + Li ei )

= j (x + Li ei ) A (x + Li ei ) + j (x + Li ei )

= j (x + Li ei )i (x) A (x) + i (x)T j (x + Li ei ) .
(A.6)
The two results are consistent only if the transition functions obey the first cocycle
condition of Eq. (2.14). Next, we pick the spatial i-direction and the Euclidean time
direction, such that
A (x + Li ei + e4 )

= i (x + e4 ) A (x + e4 ) + i (x + e4 )

= i (x + e4 )4 (x) A (x) + 4 (x)T i (x + e4 ) .
(A.7)
234
Again, performing the two shifts in the opposite order we obtain

A (x + e4 + Li ei )

= 4 (x + Li ei ) A (x + Li ei ) + 4 (x + Li ei )

= 4 (x + Li ei )i (x) A (x) + i (x) 4 (x + Li ei ) .
(A.8)
In this case, the resulting cocycle condition is the second one of Eq. (2.14).
Eq. (2.15) ensures the gauge-covariance of C-periodic boundary condition, i.e.,

A (x + Li ei ) = (x + Li ei ) A (x + Li ei ) + (x + Li ei )

= (x + Li ei )i (x) Ai (x) + i (x) (x + Li ei )

= i (x) A (x) + i (x) .
(A.9)
Let us consider the gauge transformed cocycle condition

i (x + Lj ej ) j (x) = j (x + Li ei ) i (x) zij
(x + Lj ej + Li ei )i (x + Lj ej )(x + Lj ej )T
(x + Lj ej ) j (x) (x)
= (x + Li ei + Lj ej )j (x + Li ei )(x + Li ei )T

(x + Li ei ) i (x) (x) zij

i (x + Lj ej )j (x) = j (x + Li ei )i (x) zij
.
(A.10)
= z , i.e., the twist-tensor is invariant under the transformations of Eq. (2.15).

Hence, zij
ij
Similarly, we obtain

i (x + e4 ) 4 (x) = 4 (x + Li ei ) i (x) zi4
(x + e4 + Li ei )i (x + e4 )(x + e4 )T (x + e4 ) 4 (x) (x)T

= (x + Li ei + e4 )4 (x + Li ei )(x + Li ei ) (x + Li ei )

i (x)(x)Tzij

i (x + e4 )4 (x) = 4 (x + Li ei )i (x)zi4
,
(A.11)
=z .
such that zi4
i4
Interestingly, with C-periodic boundary conditions there are further consistency
conditions besides the cocycle condition Eq. (2.14). For example, on the one hand, one
has
i (x + Lj ej + Lk ek )j (x + Lk ek ) k (x)
= j (x + Li ei + Lk ek )i (x + Lk ek ) k (x)zij
= j (x + Li ei + Lk ek )k (x + Li ei ) i (x)zij zki
= k (x + Li ei + Lj ej )j (x + Li ei ) i (x)zij zki zj k ,
(A.12)
235
while, on the other hand,

i (x + Lj ej + Lk ek )j (x + Lk ek ) k (x)
= i (x + Lj ej + Lk ek )k (x + Lj ej ) j (x)zkj
= k (x + Lj ej + Li ei )i (x + Lj ej ) j (x)zkj zik
= k (x + Lj ej + Li ei )j (x + Li ei ) i (x)zkj zik zj i .
(A.13)
Hence, unlike for periodic boundary conditions, there is the constraint of Eq. (2.16),
2 z2 z2 = 1, on the twist-tensor itself. Similarly, if one shifts in two spatial directions
zij
j k ki
as well as in the Euclidean time direction, on the one hand, one finds
i (x + Lj ej + e4 )j (x + e4 ) 4 (x)
= j (x + Li ei + e4 )i (x + e4 ) 4 (x)zij
= j (x + Li ei + e4 )4 (x + Li ei ) i (x) zij zi4
= 4 (x + Li ei + Lj ej )j (x + Li ei )i (x) zij zi4

zj 4 ,
(A.14)
while, on the other hand,

i (x + Lj ej + e4 )j (x + e4 ) 4 (x)
= i (x + Lj ej + e4 )4 (x + Lj ej ) j (x) zj4
= 4 (x + Lj ej + Li ei )i (x + Lj ej )j (x) zj4 zi4
= 4 (x + Lj ej + Li ei )j (x + Li ei )i (x) zj4 zi4 zij .
(A.15)
2 = z2 .
Consequently, one also obtains Eq. (2.17), zi4
j4
References
[1] A.M. Polyakov, Phys. Lett. B 72 (1978) 477.
[2] L. Susskind, Phys. Rev. D 20 (1979) 2610.
[3] B. Svetitsky, L.G. Yaffe, Nucl. Phys. B 210 (1982) 423;
B. Svetitsky, L.G. Yaffe, Phys. Rev. D 26 (1982) 963.
[4] K. Holland, U.-J. Wiese, in: M. Shifman (Ed.), At the Frontier of Particle PhysicsHandbook of QCD,
World Scientific, Singapore, 2001, hep-ph/0011193.
[5] J. Greensite, hep-lat/0301023.
[6] J. Greensite, B. Lautrup, Phys. Rev. Lett. 47 (1981) 9.
[7] G. Bhanot, M. Creutz, Phys. Rev. D 24 (1981) 3212.
[8] I.G. Halliday, A. Schwimmer, Phys. Lett. B 101 (1981) 327;
I.G. Halliday, A. Schwimmer, Phys. Lett. B 102 (1981) 337.
[9] L. Caneschi, I.G. Halliday, A. Schwimmer, Nucl. Phys. B 200 (1982) 409.
[10] S. Cheluvaraja, H.S. Sharatchandra, hep-lat/9611001.
[11] S. Datta, R.V. Gavai, Phys. Rev. D 57 (1998) 6618, hep-lat/9708026.
[12] P. de Forcrand, O. Jahn, Nucl. Phys. B 651 (2003) 125, hep-lat/0211004.
[13] H. Reinhardt, T. Tok, Phys. Lett. B 500 (2001) 173, hep-th/0011068.
[14] E. Hilf, L. Polley, Phys. Lett. B 131 (1983) 412.
[15] G. t Hooft, Nucl. Phys. B 138 (1978) 1;
G. t Hooft, Nucl. Phys. B 153 (1979) 141.
236
[16] P. van Baal, Commun. Math. Phys. 85 (1982) 529.

[17] P. van Baal, in: M. Shifman (Ed.), At the Frontier of Particle PhysicsHandbook of QCD, World Scientific,
Singapore, 2001, hep-ph/0008206.
[18] L. Polley, U.-J. Wiese, Nucl. Phys. B 356 (1991) 629.
[19] A.S. Kronfeld, U.-J. Wiese, Nucl. Phys. B 357 (1991) 521;
A.S. Kronfeld, U.-J. Wiese, Nucl. Phys. B 401 (1993) 190, hep-lat/9210008.
[20] U.-J. Wiese, Nucl. Phys. B 375 (1992) 45.
[21] M. Gunaydin, F. Gursey, J. Math. Phys. 14 (1973) 1651.
[22] W.G. McKay, J. Patera, D.W. Rand, Exceptional Simple Lie Algebras, in: Tables of Representations of
Simple Lie Algebras, Vol. I, Les Publications CRM, Montral, 1990.
[23] K. Fredenhagen, M. Marcu, Phys. Rev. Lett. 56 (1986) 223.
[24] L.D. McLerran, B. Svetitsky, Phys. Rev. D 24 (1981) 450;
L.D. McLerran, B. Svetitsky, Phys. Lett. B 98 (1981) 195.
[25] J. Kuti, J. Polonyi, K. Szlachanyi, Phys. Lett. B 98 (1981) 199.
[26] J. Engels, F. Karsch, H. Satz, I. Montvay, Phys. Lett. B 101 (1981) 89.
[27] R.V. Gavai, Nucl. Phys. B 215 (1983) 458.
[28] R.V. Gavai, F. Karsch, H. Satz, Nucl. Phys. B 220 (1983) 223.
[29] J. Engels, J. Fingberg, M. Weber, Nucl. Phys. B 332 (1990) 737.
[30] J. Engels, J. Fingberg, D.E. Miller, Nucl. Phys. B 387 (1992) 501.
[31] T. Celik, J. Engels, H. Satz, Phys. Lett. B 125 (1983) 411.
[32] J.B. Kogut, M. Stone, H.W. Wyld, W.R. Gibbs, J. Shigemitsu, S.H. Shenker, D.K. Sinclair, Phys. Rev.
Lett. 50 (1983) 393.
[33] S.A. Gottlieb, J. Kuti, D. Toussaint, A.D. Kennedy, S. Meyer, B.J. Pendleton, R.L. Sugar, Phys. Rev. Lett. 55
(1985) 1958.
[34] F.R. Brown, N.H. Christ, Y.F. Deng, M.S. Gao, T.J. Woch, Phys. Rev. Lett. 61 (1988) 2058.
[35] R.V. Gavai, F. Karsch, B. Petersson, Nucl. Phys. B 322 (1989) 738.
[36] M. Fukugita, M. Okawa, A. Ukawa, Phys. Rev. Lett. 63 (1989) 1768.
[37] N.A. Alves, B.A. Berg, S. Sanielevici, Phys. Rev. Lett. 64 (1990) 3107.
[38] S.J. Knak Jensen, O.G. Mouritsen, Phys. Rev. Lett. 43 (1979) 1736.
[39] H.W.J. Blte, R.H. Swendsen, Phys. Rev. Lett. 43 (1979) 799.
[40] H.J. Herrmann, Z. Phys. B 35 (1979) 171.
[41] F.Y. Wu, Rev. Mod. Phys. 54 (1982) 235.
[42] M. Fukugita, M. Okawa, Phys. Rev. Lett. 63 (1989) 13.
[43] P. Hasenfratz, F. Karsch, I.O. Stamatescu, Phys. Lett. B 133 (1983) 221.
[44] F. Karsch, S. Stickan, Phys. Lett. B 488 (2000) 319, hep-lat/0007019.
[45] R.D. Pisarski, F. Wilczek, Phys. Rev. D 29 (1984) 338.
[46] E. Witten, Nucl. Phys. B 507 (1997) 658, hep-th/9706109.
[47] A. Campos, K. Holland, U.-J. Wiese, Phys. Rev. Lett. 81 (1998) 2420, hep-th/9805086;
A. Campos, K. Holland, U.-J. Wiese, Phys. Lett. B 443 (1998) 338, hep-th/9807207.
[48] Z. Frei, A. Patkos, Phys. Lett. B 229 (1989) 102.
[49] T. Trappenberg, U.-J. Wiese, Nucl. Phys. B 372 (1992) 703.
[50] M. Gross, J.F. Wheater, Nucl. Phys. B 240 (1984) 253.
[51] A. Gocksch, M. Okawa, Phys. Rev. Lett. 52 (1984) 1751.
[52] G.G. Batrouni, B. Svetitsky, Phys. Rev. Lett. 52 (1984) 2205.
[53] M. Wingate, S. Ohta, Phys. Rev. D 63 (2001) 094502, hep-lat/0006016.
[54] B. Lucini, M. Teper, U. Wenger, Phys. Lett. B 545 (2002) 197, hep-lat/0206029.
[55] K. Holland, M. Pepe, U.-J. Wiese, in preparation.

SU(2) reduction of six-dimensional (1, 0)

supergravity
H. L 1 , C.N. Pope 1 , E. Sezgin 2
George P. and Cynthia W. Mitchell Institute for Fundamental Physics,
Texas A&M University, College Station, TX 77843-4242, USA
Received 23 May 2003; accepted 16 June 2003
Abstract
We obtain a gauged supergravity theory in three dimensions with eight real supersymmetries
by means of a ScherkSchwarz reduction of pure N = (1, 0) supergravity in six dimensions on
the SU(2) group manifold. The SU(2) YangMills fields in the model propagate, since they have
an ordinary kinetic term in addition to ChernSimons couplings. The other propagating degrees
of freedom consist of a dilaton, five scalars which parameterise the coset SL(3, R)/SO(3), three
vector fields in the adjoint of SU(2), and twelve spin- 12 fermions. The model admits an AdS3
vacuum solution. We also show how a charged black hole solution can be obtained, by performing
a dimensional reduction of the rotating self-dual string of six-dimensional (1, 0) supergravity.
2003 Published by Elsevier B.V.
PACS: 04.65.+e; 11.27.+d
1. Introduction
The importance of supergravities in diverse dimensions in the context of string theory
has been widely appreciated for a long time. However, the role of gauged supergravities
has been appreciated only relatively recent, and especially it has become more evident with
developments in the AdS/CFT correspondence. Supergravities are referred to as gauged
when their R-symmetry group, or any subgroup thereof, is gauged. Those which have
played role in the AdS/CFT correspondence admit AdS vacuum solutions, but not all
E-mail address: sezgin@physics.tamu.edu (E. Sezgin).
1 Research supported in part by DOE grant DE-FG03-95ER40917.
2 Research supported in part by NSF Grant PHY-0070964.
0550-3213/$ see front matter 2003 Published by Elsevier B.V.

doi:10.1016/S0550-3213(03)00534-0
238
H. L et al. / Nuclear Physics B 668 (2003) 237257
gauged supergravities admit AdS vacua (see, for example, [1]), and not all supergravities
which admit AdS vacua are gauged (see, for example, [2]). We shall refer to those gauged
supergravities that do admit AdS vacua as AdS supergravities.
In this paper, we shall fill a gap in our knowledge of AdS supergravities by constructing
one in three dimensions with eight real supersymmetries. We shall do this by means of a
ScherkSchwarz reduction [3] of the chiral N = (1, 0) supergravity in D = 6. The main
motivation for our work is to make progress in finding the still elusive supergravity theory
that is expected to arise in the AdS3 S3 compactification of (2, 0) supergravity theory [4,
5], which in turn, emerges from type IIB string theory reduced on K3 [6,7]. As a first
step in this direction, here we construct the AdS3 supergravity that is pertinent to the
AdS3 S 3 compactification of pure N = (1, 0) supergravity in D = 6 [8], which can
also be embedded in heterotic string compactifying on K3. We do so by employing a
ScherkSchwarz reduction on the SU(2) group manifold. Before commenting further on
our results, let us review some facts about gauged and AdS supergravity theories that can be
obtained from consistent KaluzaKlein reductions, which will clarify the rationale behind
our having chosen the ScherkSchwarz reduction scheme here.
Many examples of gauged supergravities can be obtained by consistent KaluzaKlein
sphere reduction of higher-dimensional ungauged supergravities.3 Notable cases include
the S 5 reduction of type IIB supergravity and the S 7 [9] and S 4 [10,11] reductions of
eleven-dimensional supergravity; these give rise to the maximal gauged supergravities in
D = 5, D = 7 and D = 4 respectively.4 The reductions to maximal supergravities are
extremely complicated, especially in the S 5 and S 7 cases. Several examples giving gauged
supergravities with lesser supersymmetry have been worked out in complete detail (at
least in their bosonic sectors). These include the half-maximal supersymmetric gauged
supergravities in D = 7, 6, 5 and 4 dimensions [1215].
The fact that the above KaluzaKlein sphere reductions are consistent is quite
remarkable, and there is no general understanding of why they work. The consistency
depends crucially on conspiracies between contributions from the metric and the other
fields in the higher-dimensional theories. One might suspect from the above examples that
supersymmetry plays a crucial role, but in fact this is somewhat misleading. There are
examples where supersymmetric theories cannot be consistently reduced on spheres, and
there are examples where non-supersymmetric theories can be consistently reduced on
spheres.
3 A consistent KaluzaKlein reduction is defined here as one where all the gauge bosons of the isometry
group G of the compactifying manifold are retained in a truncation keeping only a finite number of the lowerdimensional fields, with the essential requirement that setting the truncated fields to zero must be consistent with
their own equations of motion. Put another way, the reduction ansatz is consistent if all the higher-dimensional
equations of motion are satisfied as a consequence of the equations of motion for the retained lower-dimensional
fields. It is only in very exceptional cases that such consistent KaluzaKlein reductions on compactifying spaces
other than tori are possible.
4 To be precise, the consistency of the S 5 reduction has never been proven, and the reduction ansatz for the
7
S example has not been fully explicitly exhibited. The explicit reduction ansatz for the S 4 case has been given,
and its consistency has been proven, modulo the assumption that the inclusion of quartic fermion terms will not
upset results established at lower order in fermions. In all cases, compelling circumstantial evidence implying the
consistency of the reductions has been found.
239
In fact, a more universal characterisation of when a theory admits a consistent sphere

reduction can be given by first studying the global symmetry of the theory when it is instead
KaluzaKlein reduced on a torus of the same dimension as the sphere. This was discussed
in depth in [16]. The key point is that a generic theory reduced on T n has a GL(n, R) global
symmetry, and so its maximal compact subgroup is SO(n). By contrast, the theory that one
would obtain by instead reducing on S n would have an SO(n + 1) gauge group, and so by
sending the gauge-coupling to zero there would have to be at least an SO(n + 1) compact
subgroup in the resulting global symmetry group, contradicting the previous observation
that generically the maximal compact subgroup is only SO(n). Put another way, if there
were a consistent S n reduction then one would have to be able to gauge an SO(n + 1)
subgroup of the global symmetry group of the T n reduction, and for a generic theory the
toroidal reduction does not yield a large enough global symmetry group.
The only way in which a reduction on S n could be consistent is, therefore, if there
is actually an enhanced global symmetry group in the reduction on T n . This occurs
only if there is some conspiracy between the contributions from the T n reduction of
metric and the other fields in the theory. Such conspiracies are indeed sometimes seen
in supergravity reductions (including the toroidal reductions of type IIB and elevendimensional supergravity), and it is precisely these conspiracies that also allow the
consistent sphere reductions to work. However, as was shown in [16], there exist also
examples of purely bosonic theories that also admit consistent sphere reductions; a notable
set of cases is provided by the S 3 or S D3 reductions of the bosonic string effective action
in any dimension D.
Since there exist AdS3 S 3 supersymmetric vacua of the six-dimensional ungauged
supergravities, it is natural to inquire whether there might exist associated consistent S 3
reductions. An extension of the arguments presented in [16] suggests that such consistent
reductions are not possible.5 On the other hand, one would expect that there should exist
an AdS3 gauged supergravity, with a higher-dimensional origin, which would play an
important role in the AdS3 /CFT2 correspondence. This would then be analogous to the
examples in 4 D 8.
Indeed, there exists an alternative way of performing a consistent KaluzaKlein
reduction on S 3 , exploiting the fact that it is isomorphic to the group manifold SU(2).
This reduction, known as ScherkSchwarz reduction [3], has the merit that it is guaranteed
to be consistent when applied to any theory at all. It does, however, give rise only to the
gauge fields of SU(2), rather than the SO(4) SU(2) SU(2) that would have arisen had
there existed a consistent sphere reduction of the kind we were describing above. In this
paper, we shall implement the ScherkSchwarz procedure for the case of an S 3 reduction
of the N = (1, 0) chiral supergravity in D = 6. This gives rise to a gauged supergravity in
three dimensions with an SU(2) gauge group that admits an AdS3 vacuum solution.
The gauged supergravity obtained here bears similarities to maximal AdS supergravities
in D = 5 [17,18], D = 6 [19] and D = 7 [20]. After we present our results, we shall
5 It is possible, however, to perform a different consistent S 3 reduction that results in a gauged threedimensional supergravity which does not admit an AdS3 vacuum solution, but instead a domain wall [16]. From
the six-dimensional point of view, this solution is the near-horizon limit of a purely electric or purely magnetic
string.
240
comment on these similarities in the concluding section. It is worth pointing out here,
however, that our AdS3 supergravity differs from other gauged supergravities that have
been constructed so far in three dimensions [2126], in which the YangMills fields are
non-propagating since they belong to the supergravity multiplet and are described solely
by a ChernSimons term. Our model corresponds to a fusion of non-propagating Poincar
supergravity [27,28] with propagating fields originating from the reduction of the graviton
and 2-form field of N = (1, 0) supergravity in D = 6. In particular, the bosonic field
content consists of a dilaton; five scalars which parameterise the coset SL(3, R)/SO(3);
the gauge fields of SU(2) SO(3), all of which originate from the six-dimensional metric;
and three vector fields which originate from the six-dimensional 2-form potential. Thus
altogether there are twelve bosonic degrees of freedom, and by supersymmetry, twelve
fermionic ones. This is the same count of degrees of freedom that one finds in a toroidal
compactification of pure N = (1, 0) supergravity in D = 6. We shall compare this with
the massless KaluzaKlein spectra of the AdS3 S 3 compactified (1, 0) and N = (2, 0)
supergravities in the concluding section.
In this paper we shall also elaborate on the structure of the scalar potential that arises in
our model, describe the U (1) truncation of the model, and we shall show how a charged
black hole solution can be obtained by performing a dimensional reduction of the rotating
self-dual string in the six-dimensional (1, 0) supergravity.
The organisation of the paper is as follows. In Section 2 we begin with a review of
the ScherkSchwarz S 3 reduction of a D-dimensional metric, deriving expressions for the
Ricci tensor in (D 3) dimensions. We then specialise to the case of six-dimensional
N = (1, 0) supergravity, deriving the expressions for the reduction of the self-dual 3form field, and hence for the complete reduction of the bosonic sector of the theory. We
show how the potential for the GL(3, R)/SO(3) scalars can be expressed in terms of a
superpotential, and we derive a consistent truncation in which the SU(2) YangMills fields
are reduced to U (1).
In Section 3 we extend the construction of the three-dimensional gauged supergravity
by obtaining the supersymmetry transformation rules for the fermionic fields. In Section 4
we consider some specific solutions of the three-dimensional supergravity, and their sixdimensional interpretation in terms of rotating self-dual strings. The paper closes with
conclusions and speculations in Section 5.
2. The bosonic sector

The bosonic sector of the (1, 0) six-dimensional supergravity theory comprises the
metric tensor gMN and a 2-form potential B (2) whose field strength H (3) = d B (2) is selfdual. The six-dimensional bosonic equations of motion are
R MN = H MP Q H N P Q ,
d H (3) = 0,
H (3) = H (3).
(1)
We shall reduce the six-dimensional theory to three dimensions by compactifying on a

3-sphere, viewed as the group manifold SU(2). The KaluzaKlein reduction scheme that
we employ will be the group manifold reduction of Scherk and Schwarz [3], in which a
truncation to the set of all fields invariant under the left action GL of the total isometry
241
group GL GR acting on the group manifold G. This truncation is guaranteed to be a

consistent one, since on group-theoretic grounds non-linear products of the retained GL singlet fields cannot act as sources for the discarded GL -non-singlet fields.
2.1. Reduction of the metric
It is convenient to introduce the left-invariant SU(2) 1-forms , which satisfy the
MaurerCartan algebra
1
d = f ,
(2)
2
where f are the SU(2) structure constants. For now we shall consider the case where
we reduce on SU(2) from (n + 3) to n dimensions. The KaluzaKlein metric reduction
ansatz will then be given by6
d s2 = e2 ds 2 +
4 2
e
h ,
g2
(3)
where is the breathing-mode scalar, h denotes the remaining n-dimensional scalar

fields (with the symmetric tensor h being unimodular), and is given by
g A .
(4)
denotes the SU(2) YangMills potentials corresponding to the right-acting SU(2)

Here
isometry of the 3-sphere. The constants and in (3) will be determined later.
It will prove convenient to work in a vielbein basis, which we take to be
e a = e ea ,
e i = 2g 1 e Li .
ea
Here is a vielbein basis for the n-dimensional metric

h , and so

h = Li Li ,
det Li = 1.
(5)
ds 2 ,
and
Li
is a square root of
(6)
Defining the YangMills field strengths F = dA + 12 gf A A , we have:

DF dF + gf A F = 0,
1
D d + gf A = gF f .
(7)
2
It is also useful to define the YangMills covariant exterior derivative acting on the scalars
Li :
DLi dLi gf A Li .
(8)
6 In this paper we are using supergravity conventions, in which the bosonic Lagrangian is written with
i )2 + ). For the convenience of readers who prefer

normalisations of the form g( 14 R 12 ()2 14 (F(2)
1
1
i
2
2
the customary g(R 2 () 4 (F(2) ) + ) convention, we include a hidden appendix which repeats
Sections 2 and 4 in this notation. It can be accessed by deleting the \end{document} in the Latex file at the end
of the references.
242
Using these expressions, we find from (5) that

d e a = e b e a e b a b e b ,

i a
d e i = e L1 j Da Li e a e j + e a e a e i e(2) Fab
e e b
1
ge T i( )j k( e j e k ,
4
where we have defined
i
Fab
Li F ,
T ij Li Lj .
(9)
(10)
(Note that T ij is SU(2)-covariant, despite superficial appearances, since is an invariant

tensor in SO(3) SU(2).) From (9), we calculate the torsion-free spin connection A B ,
defined by d e A = A B eB and AB = BA , finding

i i
e ,
ab = ab + e b ac e c a bc e c + e(2) Fab
i b
ai = e Paij e j e a e i + e(2) Fab
e ,

1
ij = e Qaij e a + ge T k( )ij ( + T j ( )ik( T i( )j k( e k .
(11)
4
Here we have defined

1 1
L i Da Lj + L1 j Da Li ,
Paij
2

1 1
Qaij
(12)
L i Da Lj L1 j Da Li .
2
The next step is to calculate the Ricci tensor R AB = R C ACB , which can in principle be
done by first calculating the curvature 2-forms
1
A B = d A B + A C C B = R A BCD eC e D .
2
This is quite an involved calculation. In practice, a simpler way to find the Ricci tensor is
to use an observation that was made in [3], which is that the dimensional reduction of the
EinsteinHilbert Lagrangian L = eR is given, up to a total derivative, by7

L = e ABC CAB + A A ,
(13)
where
AB CAB e C ,
A BC BCA .
(14)
It is convenient at this point to make the following choices for the constants and :
2 =
6
,
(n 2)(n + 1)
1
= (n 2).
3
(15)
7 It is crucial, in order to apply this argument, that the reduction be a consistent one, meaning that the equations
of motion derived from the dimensionally-reduced action coincide with those that follow from the dimensional
reduction of the higher-dimensional equations of motion.
243
The second condition ensures that the lower-dimensional metric is also in the Einstein
frame, and the first condition simply sets the scale for so that it has a canonicallynormalised kinetic term. After making these choices, we find after a simple calculation

from (11) and (13) that the higher-dimensional EinsteinHilbert Lagrangian L = R g
reduces to give

g R =

i 2
2
g R 2()2 (Paij )2 e 3 (n+1) Fab

1 2 2 (n+1) ij ij 1 2
T T T
+ total derivative,
g e3
4
2
(16)
where T T ii .
From (16), we can easily obtain the lower-dimensional equations of motion for the
metric, the YangMills potentials, and the scalar fields. These equations in fact match up
with the R ab , R ai and R ij vielbein components of the higher-dimensional Ricci tensor.
The only subtlety is that there are overall scalings to be determined, involving certain
specific powers of e , and that the components R ab are actually formed from a linear
combination of the lower-dimensional Einstein equation and ab times a multiple of the
trace of the scalar field equations. These scalings and combinations are easily determined
by considering special cases. By this means, we therefore arrive at the expressions for
the components of the higher-dimensional Ricci tensor with much less labour than by the
direct approach via the curvature 2-forms. We find

2
i
i cd
R ab = e2 Rab 2a b Paij Pbij ab 2e 3 (n+1) Fac
Fbd
,

1
1
(n5)
b 23 (n+1) i
23 (n+1) j
b
k(
3
D e
Rai = e
Fab + e
Fab P ij g)ij k T Paj ( ,
2

2
1
2
j
i
R ij = e2 Da P a ij (n 2) ij 2e 3 (n+1) Fab
Fcd ac bd
2
3

1 2 2 (n+1) k( k( 1 2
1
2 23 (n+1)
ik j k
ij
3
T T TT
+ g e
T T T ij ,
g e
2
2
2
(17)
where we have now defined a derivative Da that is covariant not only with respect to general
coordinate and YangMills transformations, but also it involves the composite connection
Qaij :
j
Da Li Da Li + Qaij L ,
(18)
(and similarly when acting on Paij and F i ). Note that using Da , we have

L1 i Da Lj = Paij .
(19)
244
2.2. Reduction of the 3-form

(3) + G
(3) . We find
The self-dual 3-form in six dimensions can be written as H (3) = G
that the appropriate reduction ansatz is given by taking
(3) = 8m (3) + 2 ) B ,
G
g3
g2
(20)
where m is a constant and 1 2 3 . Dualising (20) in the metric (3), we find that
4
(3) = me4 )(3) 2 e 3 h B ,
G
g
(21)
where h Li Li . (Here we have used equation (15), which for n = 3 implies = 13

and 2 = 38 .) Noting that d(3) = 12 g) F , we find that the Bianchi identity
d H (3) = 0 implies the equation
1 4
(DB )Li 2mF i + ge 3 T ij Bj = 0.
2
(22)
We find that the vielbein components of the self-dual field strength H (3) are given by
H abc = me )abc ,
H ij k = me )ij k ,
H abi = e 3 )ab c Bci ,

1
H aij = e 3 )ij k Bak ,

1
(23)
where we have defined B i Li B .

2.3. Three-dimensional bosonic equations of motion and Lagrangian
From (23), together with our expressions (17) for the Ricci tensor, we find that the
six-dimensional bosonic equations (1) imply the following three-dimensional bosonic
equations of motion. First, from the R ij components of the Einstein equation, we obtain
the scalar equations

2
i 2
4
8
1 2 8 ij ij 1 2
2 4
3
= 6m e
+ g e
,
T T T + 2e 3 Bai e 3 Fab
4
2

8
1
1
1
Da P a ij = g 2 e 3 T ik Tj k T T ij
T k( T k( T 2 ij
2
3
2

4
1 2
j
+ 8e 3 Bai Bb ab Bak ij
3

8
1 k 2
j
i
+ 2e 3 Fab
Fcd ac bd Fab
ij .
(24)
3
From the R ai components of the Einstein equation, we obtain the YangMills equation
8

8
4
1
j
i
= e 3 Fab P b ij + g 2 )ij k T k( Paj ( 4me 3 Bai
Db e 3 Fab
4
j
+ 2)ij k )a bc Bb Bck .
(25)
245
From the R ab components of the Einstein equation we obtain, after using the equation
above to replace a term,

4
1 i 2
i i
38
i
i cd
3
Ba Bb + 2e
Fac Fbd (Fcd ) ab
Rab = 2a b + Paij Pbij + 4e
2

8
1
1
+ 4m2 e4 ab + g 2 e 3 T ij T ij T 2 ab .
(26)
4
2
Finally, there is the equation (22) that came from d H (3) = 0.
We find that these equations of motions can be derived from the 3-dimensional
Lagrangian
4
1 8
1
1
1
L = R 1 d d Pij Pij e 3 B i B i e 3 F i F i
4
2
4
2

1 2 8 ij ij 1 2
2 4
m e
1 g e3
(27)
T T T 1 + LCS ,
16
2
where the ChernSimons contribution LCS is given by

2
8m
8m2
LCS = DB B +
F B
(3) ,
g
g
g
(28)
where
1
(3) A dA + ) A A A
(29)
3
is the usual ChernSimons 3-form for the SU(2) YangMills fields, satisfying d(3) =
F F . Note that the Lagrangian is invariant under m m, together with B B .
On the other hand, it is invariant under g g and A A only if one also performs
a parity or time-reversal transformation.
Although one cannot directly take the g 0 limit in the Lagrangian, it can clearly
be done at the level of the equations of motion. This is analogous to the case of sevendimensional gauged supergravity, where the limit was discussed in detail in [29]. The
g 0 limit corresponds to a flattening of the reduction 3-sphere, which in the truncation
to the massless sector can be replaced by a torus. The theory also admits a different limit, in
which one instead sends m to zero. As can be seen from (20), this corresponds to setting the
3-form flux on S 3 to zero. The maximally-symmetric vacuum solution would then be sixdimensional Minkowski spacetime d s62 = dx dx + dr 2 + r 2 d32 , instead of AdS3 S 3 .
It is interesting to note that in the ScherkSchwarz reduction of ten-dimensional
N = 1 supergravity on S 3 , one can consistently truncate out the scalar fields that are
parameterised by Li , whilst retaining the SU(2) YangMills fields. (This was proved in
[30] for ScherkSchwarz reductions of the low-energy effective action of the bosonic string
in any dimension, reduced on any group manifold. The truncation of the scalars is possible
provided that the vectors coming from the reduction of the 2-form potential are equated to
the YangMills potentials.) By contrast, in our present case we cannot consistently truncate
the scalars Li without also truncating the SU(2) YangMills fields, as can be seen from
the scalar equations for Da Pija in (24). The key difference is that in the present case the
3-form field strength in six dimensions is subject to a self-duality condition.
246
2.4. Superpotential for the three-dimensional theory

The scalar potential in (27) can be expressed in terms of a superpotential W . To do so,
it is useful first to absorb the dilaton into Li , by defining
L i e 3 Li .
2
(30)
The scalar sector of the Lagrangian (27) can then be written as

1
Lscal = Pij Pij V 1,
4
where the scalar potential is given by

1 2 ij ij 1 2
2
ij
V m det T + g T T T ,
16
2
(31)
(32)
and T ij L i L .
Introducing coordinates I on the GL(3, R)/SO(3) scalar coset manifold, a vielbein on
the coset can be written as

ij
VI = L 1 i D I L j ,
(33)
j
where
L i
ij
+ Q I L j ,
D I L i
I

j
1 1 L 1 L i
ij
L i
QI
L j
.
2
I
I
ij
(34)
ij
In terms of the coset metric GI J VI VJ , the scalar Lagrangian (31) can be written as
Lscal = 14 GI J d I d J V 1. The potential V can be expressed in terms of a
superpotential W as
V = GI J
W W
W 2,
I J
where we find that W is given by

1
W = 2 m det T ij + g T ,
4 2
4
1
= 2 me2 + ge 3 T .
4 2
(35)
(36)
If m has the same sign as g, (36) has an extremum at T ij = ij , corresponding to the

pure AdS3 supersymmetric vacuum solution. More generally, there are solutions that break
half the supersymmetry, which can be lifted back to six dimensions where they correspond
to the standard self-dual string. If, on the other hand, m has the opposite sign to g, then
AdS3 is not included among the solutions of the associated first-order equations coming
from W . There will instead be a half-supersymmetric domain-wall solution, which can be
lifted back into six dimensions where it acquires an interpretation as a disjoint interior
branch of a negative-mass self-dual string. (This issue was discussed extensively in [31].)
247
2.5. A consistent U (1) truncation

It is straightforward to see that we can perform a consistent truncation of the threedimensional theory obtained in the previous section, in which we set to zero two of the
three YangMills potentials A and two of the B 1-forms (for = 1 and 2), and at the
same time we truncate the 5 scalars in the unimodular matrix Li to a single diagonal scalar:

e
0
0
i
0
,
L =
(37)
0 e
0
0 e2
where 1/ 3. After establishing that the truncation is consistent (by looking at the
previous field equations), we can then simply impose the truncation in the Lagrangian (27).
This gives
1
1
1
L = R 1 d d d d
4
2
2
4
8
1
e 3 4 B B e 3 4 F F
2

1 2 8 8
2 4
e
1 g e3
4e2 1 + LCS ,
m e
32
where the ChernSimons contribution LCS is given by
8m
8m2
2
F B
dA A.
LCS = dB B +
g
g
g
(38)
(39)
Note that the Lagrangian is invariant under m m, together with B B . On the

other hand, it is invariant under g g and A A only if one also performs a parity
or time-reversal transformation.
The three-dimensional equations of motion following from this truncated Lagrangian
are:

8
8
1
= 6m2 e4 + g 2 e 3 e8 4e2 e 3 4 F 2
8
4
1
+ g 2 e 3 4 B 2 ,
2
1 8
4
1
1 2 8 8
e
e2 e 3 4 F 2 g 2 e 3 4 B 2 ,
= g e3
12
3
3
8 4

4
4
d e 3
F = 4me 3
B,
4
e 3 4 B =
2
4m
F dB,
g
g

4
8
1
1
2
Rab = a b + 6a b + 4e 3 4 Ba Bb + 2e 3 4 Fab
F 2 ab
2
2

8
1
(40)
+ 4m2 e4 ab + g 2 e 3 e8 4e2 ab .
8
248
The corresponding truncation in the SU(2) reduction ansatz is given by

4 2 2 2
e
1 + 22 + e4 (3 gA)2 ,
e 3
g2
(3) = 8m 1 2 (3 gA) + 4 B 1 2 .
G
g3
g2
d s62 = e2 ds32 +
(41)
(3) is given by
The dual of G
4
(3) = me4 )(3) 2 e 3 4 B (3 gA).
G
g
(42)
It is also of interest to look for superpotentials from which the scalar potential

1 2 8 8
g e3
e
4e2
32
can be derived. In this case, V will be expressed as

1 W 2 1 W 2
+
W 2.
V=
4
4
V = m2 e4 +
(43)
We find that the following choices for W are possible:

4
1
W = 2 me2 + ge 3 e4 + 2e2 ,
4 2

4
1
W = 2 me2 + ge 3 e4 + 4e .
4 2
(44)
(45)
The choice in (44) correspond to the superpotential given in (36), specialised to the U (1)
truncation given in (37). The choices in (45) do not arise from the specialisation of (36).
A particular solution, if the minus sign is chosen, is the same AdS3 vacuum. But in
general, the solutions of the first-order equations associated with (45) are disjoint from
those associated with (44).
3. The fermionic supersymmetry transformations

The D = 6, (1, 0) supergravity multiplet consists of the vielbein, 2-form potential
with self-dual field strength and a gravitino which is symplectic MajoranaWeyl spinor
in doublet representation of the R-symmetry group Sp(1). As is well known, a manifestly
covariant action containing these fields alone cannot be written down due to the self-duality
condition. However, the coupling of this multiplet to a tensor multiplet consisting of a
two-form potential with anti-self dual field strength, a dilaton and anti-chiral symplecticMajorana spinor does admit a Lagrangian formulation. Indeed, the complete Lagrangian,
field equations and supersymmetry transformation rules for the coupled system have been
constructed in [34]. Starting from these field equations and transformation laws, we can
obtain the corresponding ones for the pure supergravity theory by setting the dilaton and
the tensor-multiplet spinor to zero, and imposing the exact, supersymmetric self-duality
249
condition
1
+ D [D ABC E] E = 0.
HABC
(46)
8
One can show that the supersymmetric variation vanishes modulo the gravitino field
equation. To implement this self-duality condition, we need to substitute
1
+
HABC = HABC
(47)
D [D ABC E] E ,
8
everywhere HABC occurs in the equations of motion and the transformation laws.
In the present paper, we shall just derive the fermionic supersymmetry transformation
rules for the ScherkSchwarz reduced three-dimensional theory; these suffice for testing
the supersymmetry of three-dimensional bosonic solutions. In a later paper, we shall
present the entire fermionic three-dimensional Lagrangian and transformation rules.
The supersymmetry transformations obtained from [34] by the truncation of the tensor
multiplet are given up to leading order in fermions by
eM A = ) A M ,
(48)
BMN = ) [M N] ,

1 + AB
M = M + HM
AB ),
4
(49)
(50)
where M = M + 14 M AB AB . The chiral truncation leading to the a transformation rules

have also been obtained in [8].
To perform the reduction of the fermionic transformation rules, we begin by making
an ansatz for the reduction of the gravitino field. In doing so, we shall make use of the
original treatment of this problem in [3], and [35] where it has been studied further in the
context of SU(2) reduction of D = 11 supergravity. One technical aspect of the reduction
is the diagonalisation of the lower-dimensional gravitino and spinor kinetic terms. It is
convenient to treat the diagonalisation problem after performing the SU(2) reduction, at
the level of determining the field redefinitions in the lower-dimensional Lagrangian that
will yield diagonal fermionic kinetic terms. Thus, we begin with the ansatz
i (x, y) = e i (x),
a (x, y) = e a (x),
(51)
where is a constant. To ensure that the gravitino kinetic term is canonical, with no dilaton
prefactor, we must set

1
= (n 1) + 3
(52)
2
when reducing from n + 3 to n dimensions. In our case, with n = 3, we therefore have
1
1
= ,
= .
3
2
We also specify our conventions as follows:
a = a 1 1,
i = 1 i 2,
C(6) = (i ) (i ) ,
2
abc
=)
abc
(53)
ij k
= i)
ij k
(54)
7 = 1 1 ,
{a , b } = 2ab ,
(55)
{i , j } = 2ij .
(56)
250
We use the metrics AB = ( + + + + +) and ab = ( + +). Furthermore, ) 012 =

1 = ) 345 . In our convention, ) and ) are constant, and thus, e) and e1 ) are
tensors.
Although we shall not derive the complete fermionic sector of the three-dimensional
theory in the present paper, it is nonetheless useful to examine the structure of the fermionic
kinetic terms, in order to see how the diagonalisation of three-dimensional fermion fields
should be performed. To do this we can write down the Lagrangian of the six-dimensional
tensor multiplet coupled supergravity Lagrangian in which the dilaton field and the tensor
multiplet spinor are set to zero, namely
1
1
1
e1 L = R HABC H ABC A ABC BC
4
12
2
1
D [D ABC E] E HABC .
(57)
24
After variation, one then needs to impose the self-duality condition (46). Of course for
our present purposes, only the gravitino kinetic term is relevant. Substituting (51) into the
Lagrangian, we obtain the three-dimensional kinetic terms
1
1
e1 Lkin = D i i i D + i ij D j ,
2
2
where have defined
= ea a ,
= ea a .
(58)
(59)
As expected, the gravitino and spinor kinetic terms are mixed. We have verified that
they can be simultaneously be diagonalised for any dimension n. In particular, the field
redefinitions which do the job for n = 3 are given by
i
1
i = k i k .
= k k ,
2
2
It follows that the inverse transformation is
= + i k k ,
i = ij j .
(60)
(61)
The kinetic terms become diagonal in terms of the primed fields, and in particular the
spinor kinetic term becomes 12 i D i .
The supersymmetry transformations of the vielbein and gravitino in pure (1, 0)
supergravity in six dimensions are
B A ,
eA M eMB = )
(62)
1 + CD
A = A ) + HA
(63)
CD ) .
4
As for the potential B MN , it is more convenient to work with its field strength since its 3D
field B it gives rise to is related directly to the field strength H abi and H aij as in (23).
From (49) we have

1

3
+
= [A ) B C] )ABCDEF D ) E F .
H ABC
2
4
(64)
251
To perform the SU(2) reduction of the above transformation laws, we begin by making
an ansatz for the supersymmetry parameter ) (x, y). Obtaining a (x) = a )(x) +
requires the ansatz
) (x, y) = e/2 )(x),
(65)
where we have used (53). This allows us to carry out the reduction of the gravitino
transformation rule (63); to leading order in fermions, we find that this gives
i

1
= D ) ie4/3 F
2e2 Bi i )
4
1
1
(66)
+ ) me2 ),
2
2

i
1
1

i = j Pij ij ) + e4/3 Fi ee2 i k ) Bk )
2
3
4

i 4/3
1
i
+ ge
(67)
Tij ij T j ) + me2 i ).
2
2
2
Note that an expected T -dependent term in the gravitino transformation rule will emerge
after performing the redefinition (60).
After performing the redefinitions given in (59), we find that the supersymmetry
transformation rules for the diagonalised fermionic fields can be expressed as

2
i 4 i
i
2 W ) ie 3 Bi i ) + e 3 F
+ F
i ),
4

W
i
ij
i = VI j I 2 GI J J )
2
2
1 4
j
ij ),
(68)
e 3 Bk i k ) + e 3 F
4
= D ) +
where W , GI J and I are the superpotential, scalar sigma-model metric and target-space
coordinates introduced in Section 2.4.
4. Charged AdS3 black hole

In this section, we shall show how a charged black hole solution in the three-dimensional
theory of Section 2.5 can be obtained, by performing a dimensional reduction of a rotating
self-dual string in the six-dimensional (1, 0) supergravity. A rotating dyonic string solution
of six-dimensional (1, 1) supergravity was constructed in [32]. This involved two angular
momentum parameters (1 and (2 , associated with two commuting U (1) factors in the
SO(4) rotation group acting on 3-spheres in the four-dimensional transverse space of
the string. There were also two parameters 1 and 2 that characterised the electric and
magnetic charges of the string. The configuration can be viewed as a solution in the
(1, 0) supergravity if one sets 1 = 2 = , since then the electric and magnetic charges
become equal and consequently the 3-form field strength becomes self-dual and the dilaton
decouples. If the two angular momentum parameters are also set equal, (1 = (2 = (, the
252
rotating string solution then fits within our SU(2) reduction ansatz, and in fact it fits within
the U (1) truncation of Section 2.5. The metric of this self-dual rotating string solution can
be read off from [32]:

2k
dt 2 + H 1 dx 2
d s62 = H 1 1 2
r + (2
2
H h4 (2 (r 2 + (2 ) 2
H r 2 (r 2 + (2 )
2
c
dt
+
s
dx
+
dr 2
(1 + h2 (2 )
(r 2 + (2 )2 2kr 2

1
2 ,
+ H r 2 + (2 12 + 22 + 1 + h2 (2 (3 A)
4
(69)
where
H =1+
A =
2ks 2
,
r 2 + (2
h2 =
2k
,
H 2 (r 2 + (2 )2
2h2 (
(c2 dt + s 2 dx),
1 + h2 (2
and we have defined c cosh , s sinh .

We find that the self-dual 3-form is given by

ksc
2(
H(3) = 1 2 (3 A) + A
(dt + dx) 1 2
H (r 2 + ()2
2
4r(
2 2
(dt + dx) dr (3 A)
H (r + (2 )2

8r
+ 2
dt dx dr .
H (1 + h2 (2 )(r 2 + (2 )2
(70)
(71)
Comparing the metric (69) and the field strength (71) with the reduction ansatz in
Section 2.5, we obtain a three-dimensional AdS charged black hole solution, given by

2k
g6 3 2
2
2 3
2 2
1
ds3 = H r + (
1 2
dt 2 + H 1 dx 2
1 + h ( H
64
r + (2

2

H h4 (2 (r 2 + (2 ) 2
H r 2 (r 2 + (2 )
2
2
,
dt
+
s
dx
+
dr
c
(1 + h2 (2 )
(r 2 + (2 )2 2kr 2
2

2h2 (
c dt + s 2 dx ,
2
2
g(1 + h ( )

2(
B = 2 coth gA +
(dt
+
dx)
,
H (r 2 + (2 )
A=
3

g6 3 2
(72)
H r + (2 1 + h2 (2 ,
e6 = 1 + h2 (2 ,
64

where m = 2 g coth and g = 2/(ks 2 ). One can indeed directly verify that this
configuration satisfies the three-dimensional equations of motion (40).
e2 =
253
In the extremal limit, which is obtained by sending k to zero and to infinity, keeping
the charge parameter Q = k sinh 2 fixed, the solution becomes

Q2 (2
W r 2 + (2
W r 2 dr 2
2
2
2
2
dt + dx
(dt + dx) + 2
ds3 = 3
Q
W
W3
(r + (2 )2

2((r 2 + (2 + 12 Q)(dt + dx)
Q3 ((dt + dx)
A= 2
,
B=
,
2
2
(r + ( + Q)
(r 2 + (2 + Q)2

2
= 0,
e 3 = Q1 r 2 + (2 + Q ,
(73)
where W = r 2 + (2 + Q.
The charged AdS3 black hole (72) also admits a decoupling limit, namely,
Q
,
k) 2
with ) taken to be small. In this limit, the solution becomes
2

1
(2
2
2
2 2
2
2 dx
ds3 = ) N dt + 2 dr + r 2 dt
,
N
Q
Qr
)(
A = (dt + dx), B = 0, = 0, = 0,
Q
2
2 2
g= ,
m= ,
) Q
) Q
r = ) r ,
( = ) (,
k = ) 4 k,
sinh 2 =
(74)
(75)
where
r 2 2((2 k)
(4
+
+ 2 .
Q
Q
r Q
After performing the rescaling of three-dimensional fields and coupling constants
N2 =
g = ) 2 g ,
A = ) A ,
B = ) B ,
g = ) 1 g,
m = ) 1 m,
(76)
(77)
which leaves the equations of motion invariant, we obtain in the ) 0 limit the threedimensional BTZ black hole metric [33], with the mass M and angular momentum J given
by
M=
2(k (2 )
,
Q
2(2
J= .
Q
(78)
It is worth remarking that in our solution, the mass M can have either sign according
to the relative values of k and (. It becomes negative when the non-extremality parameter
k goes to zero.The BTZ black hole, which has a horizon, requires that M be positive,
and M |J |/ Q. In our parameterisation it is then necessary that 2(2 k . It is
rather surprising that the extremal BTZ black hole arises from the decoupling limit of the
non-extremal rotating black self-dual string with the non-extremality parameter k = 2(2 .
If on the other hand, we let ( i(, the mass and angular momentum becomes
M=
2(k + (2 )
,
Q
2(2
J= .
Q
(79)
254
The mass then satisfies the bound M |J |/ Q for all k 0, with the extremal limit
corresponding to k = 0.
5. Conclusion and discussion

In this paper we have carried out the ScherkSchwarz reduction of pure N = (1, 0)
chiral six-dimensional supergravity. Although the reduction ansatz is guaranteed to be
consistent, there is a subtlety that the reduction can only be performed at the level of
the equations of motion, because the six-dimensional chiral theory cannot be described
in terms of a Lagrangian. The resulting three-dimensional theory, however, can be derived
from a Lagrangian, and we have constructed this explicitly in the bosonic sector.
The three-dimensional theory that we have constructed comprises the metric, the Yang
Mills fields of SU(2), three massive vectors in the adjoint of SU(2), six scalar fields in the
coset GL(3, R)/SO(3), and their fermionic partners. The theory admits and AdS3 vacuum
solution. Although there exist many AdS3 gauged supergravities, ours is the only known
example that has a higher-dimensional string origin. One feature of the ScherkSchwarz
reduction that the breathing mode is part of the lower-dimensional massless supermultiplet.
This contrasts with the situation in the exceptional cases where a consistent KaluzaKlein
supergravity reduction on a sphere such as S 4 , S 5 or S 7 can be performed; in these cases
the breathing mode would be massive, but must be excluded in the consistent reduction. In
particular, the inclusion of the breathing mode in our ScherkSchwarz reduction implies
that we can reduce the full six-dimensional self-dual string solution, and not merely its
AdS3 S 3 near-horizon limit, to give rise to a domain wall in D = 3. Indeed, we obtained
such a solution, with rotation as well, by reducing the six-dimensional self-dual rotating
string.
It is interesting to compare the spectrum of our model with that arising in the linearised
analysis of the AdS3 S 3 compactification of the N = (1, 0) supergravity, obtained in
[4]. The massless bosonic sector of that reduction comprises a dilaton, nine scalars and the
gauge fields of SO(4), all of which originate from the metric, together with an additional six
vectors that come from the six-dimensional 2-form potential. This suggests the existence
of a gauged supergravity theory with SO(4) gauge symmetry and a scalar sector describing
a GL(4, R)/SO(4) coset, and admitting an AdS3 vacuum. Note, however, that the total
bosonic degrees of freedom in this case will be 16, in contrast to the 12 of our model
obtained by SU(2) reduction, and thus it differs also from the count of 12 obtained from
toroidal compactification. The model with SO(4) gauge fields is therefore not expected
to arise from a consistent KaluzaKlein reduction of the pure N = (1, 0) supergravity in
D = 6. Moreover, it must differ from the SO(4) gauged supergravity implied by the results
of [16], which admits a domain wall, but not AdS3 , as a solution.
Of course, an AdS3 supergravity with propagating SO(4) gauge sector may exist in
its own right, notwithstanding the fact that it is not expected to arise from a consistent
KaluzaKlein ansatz. This latter statement would mean that caution is necessary when
utilizing such a theory in an AdS3 /CFT2 correspondence, since the massive KaluzaKlein
modes may have to be taken into account in this case. In fact, the situation is similar to that
encountered in the T 1,1 compactification of the type IIB string, for which the linearised
255
analysis yields a minimal gauged supergravity coupled to matter in D = 6. A supergravity

theory with the same massless spectrum does exist in its own right, but it cannot arise from
a consistent truncation of the massive KaluzaKlein modes in the T 1,1 reduction of type
IIB supergravity, since there can be no such consistent reduction ansatz [37].
A model of great interest in the AdS3 /CFT2 context is the AdS3 S 3 compactification
of the N = (2, 0) supergravity in D = 6. In this case, the theory has 16 real supersymmetries, both in D = 6 and D = 3, and it was determined in [4] that the propagating
massless KaluzaKlein spectrum consists of 21 hyper-multiplets and a special 32 + 32
vector multiplet which consists of SO(4) YangMills fields, 6 additional vector fields and
26 scalar fields, which include a dilaton and other scalars in various representations of
the SO(4)local SO(4)global groups involved. The presence of this multiplet has not been
emphasized in the literature so far, but it is clearly of great relevance to the ultimate construction of the AdS supergravity theory that describes the full massless spectrum. The
model presented in [25] describes the coupling of arbitrary number of hyper-multiplets
to an AdS3 supergravity with 16 real supersymmetries, but it lacks the coupling of the
propagating vector multiplet mentioned above, and as such the problem of constructing
the desired supergravity theory still remains open. It is tempting to conjecture that the 26
scalars in the model parameterise an E6 /F4 coset. This problem is currently under investigation [36].
There are other interesting aspects of gauged or AdS supergravities in D = 3, having to
do, for example, with the understanding of the relation between the models constructed in
[22] and [2426]. Ultimately, progress in this area ought to lead to a better understanding
of various interesting AdS3 /CFT2 issues as well. Of course, the supersymmetric field
theories in D = 3 may have a number of other applications too, including some aspects
of brane dynamics. It is interesting that after years of progress in more complicated higherdimensional supergravities, it is relatively recently that the interest in D = 3 supergravities
has increased. It is becoming increasingly clear that this is a very rich area with a wealth
of phenomena still waiting to be uncovered.
Acknowledgements
We are grateful to Eric Bergshoeff and Per Sundell for useful discussions.
References
[1] A. Salam, E. Sezgin, Chiral compactification on Minkowski S 2 of N = 2 EinsteinMaxwell supergravity
in six dimensions, Phys. Lett. B 147 (1984) 47.
[2] P.K. Townsend, Cosmological constant in supergravity, Phys. Rev. D 15 (1977) 2802.
[3] J. Scherk, J.H. Schwarz, How to get masses from extra dimensions, Nucl. Phys. B 153 (1979) 61.
[4] S. Deger, A. Kaya, E. Sezgin, P. Sundell, Spectrum of D = 6, N = 4b supergravity on AdS3 S 3 , Nucl.
Phys. B 536 (1998) 110, hep-th/9804166.
[5] J. de Boer, Six-dimensional supergravity on S 3 AdS3 and 2d conformal field theory, Nucl. Phys. B 548
(1999) 139, hep-th/9806104.
[6] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, hep-th/9711200.
256
[7] J. Maldacena, A. Strominger, AdS3 black holes and as stringy exclusion principle, JHEP 9812 (1998) 005,
hep-th/9804085.
[8] L.J. Romans, Self-duality for interacting fields: Covariant field equations for six-dimensional chiral
supergravities, Nucl. Phys. B 276 (1986) 71.
[9] B. de Wit, H. Nicolai, The consistency of the S 7 truncation in D = 11 supergravity, Nucl. Phys. B 281
(1987) 211.
[10] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistent nonlinear KK reduction of 11D supergravity on
AdS7 S 4 and selfduality in odd dimensions, Phys. Lett. B 469 (1999) 96, hep-th/9905075.
[11] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistency of the AdS7 S4 reduction and the origin of
self-duality in odd dimensions, Nucl. Phys. B 581 (2000) 179, hep-th/9911238.
[12] H. L, C.N. Pope, Exact embedding of N = 1, D = 7 gauged supergravity in D = 11, Phys. Lett. B 467
(1999) 67, hep-th/9906168.
[13] M. Cvetic, H. L, C.N. Pope, Gauged six-dimensional supergravity from massive type IIA, Phys. Rev.
Lett. 83 (1999) 5226, hep-th/9906221.
[14] H. L, C.N. Pope, T.A. Tran, Five-dimensional N = 4, SU(2) U (1) gauged supergravity from type IIB,
Phys. Lett. B 475 (2000) 261, hep-th/9909203.
[15] M. Cvetic, H. L, C.N. Pope, Four-dimensional N = 4, SO(4) gauged supergravity from D = 11, Nucl.
Phys. B 574 (2000) 761, hep-th/9910252.
[16] M. Cvetic, H. L, C.N. Pope, Consistent KaluzaKlein sphere reductions, Phys. Rev. D 62 (2000) 064028,
hep-th/0003286.
[17] M. Gunaydin, L.J. Romans, N. Warner, Gauged N = 8 supergravity in five dimensions, Phys. Lett. B 154
(1985) 268.
[18] M. Pernici, K. Pilch, P. van Nieuwenhuizen, Gauged N = 8, d = 5 supergravity, Nucl. Phys. B 259 (1985)
460.
[19] L.J. Romans, The F (4) gauged supergravity in six dimensions, Nucl. Phys. B 276 (1986) 691.
[20] M. Pernici, K. Pilch, P. van Nieuwenhuizen, Gauged maximally supersymmetric extended supergravity in
seven dimensions, Phys. Lett. B 143 (1984) 103.
[21] A. Achucarro, P.K. Townsend, A ChernSimons action for three-dimensional AdS supergravity theories,
Phys. Lett. B 180 (1986) 89.
[22] J.M. Izquierdo, P.K. Townsend, Supersymmetric spacetimes in (2 + 1) AdS-supergravity models, Class.
Quantum Grav. 12 (1995) 895, gr-qc/9501018.
[23] P.S. Howe, J.M. Izquierdo, G. Papadopoulos, P.K. Townsend, New supergravities with central charges and
Killing spinors in 2 + 1 dimensions, Nucl. Phys. B 467 (1996) 183, hep-th/9505032.
[24] N.S. Deger, A. Kaya, E. Sezgin, P. Sundell, Matter coupled AdS3 supergravities and their black strings,
Nucl. Phys. B 573 (2000) 275, hep-th/9908089.
[25] H. Nicolai, H. Samtleben, N = 8 matter coupled AdS3 supergravities, Phys. Lett. B 514 (2001) 165, hepth/0106153.
[26] H. Nicolai, H. Samtleben, Compact and noncompact gauged maximal supergravities in three dimensions,
JHEP 0104 (2001) 022, hep-th/0103032.
[27] N. Marcus, J.H. Schwarz, Three-dimensional supergravity theories, Nucl. Phys. B 228 (1983) 145.
[28] B. de Wit, A.K. Tollsten, H. Nicolai, Locally supersymmetric D = 3 nonlinear sigma models, Nucl. Phys.
B 392 (1993) 3, hep-th/9208074.
[29] M. Cvetic, H. L, C.N. Pope, A. Sadrzadeh, T.A. Tran, S 3 and S 4 reductions of type IIA supergravity, Nucl.
Phys. B 590 (2000) 233, hep-th/0005137.
[30] M.J. Duff, B.E.W. Nilsson, C.N. Pope, KaluzaKlein approach to the heterotic string, Phys. Lett. B 163
(1985) 343.
[31] M. Cvetic, H. L, C.N. Pope, Localized gravity in the singular domain wall background?, hep-th/0002054.
[32] M. Cvetic, F. Larsen, Near horizon geometry of rotating black holes in five dimensions, Nucl. Phys. B 531
(1998) 239, hep-th/9805097.
[33] M. Banados, C. Teitelboim, J. Zanelli, The black hole in three-dimensional spacetime, Phys. Rev. Lett. 69
(1992) 1849, hep-th/9204099.
[34] H. Nishino, E. Sezgin, The complete N = 2, D = 6 supergravity with matter and YangMills couplings,
Nucl. Phys. B 278 (1986) 353.
[35] A. Salam, E. Sezgin, D = 8 supergravity, Nucl. Phys. B 258 (1985) 284.
257
[36] E. Bergshoeff, E. Sezgin, P. Sundell, in preparation.

[37] P. Hoxha, R.R. Martinez-Acosta, C.N. Pope, KaluzaKlein consistency, Killing vectors, and Khler spaces,
Class. Quantum Grav. 17 (2000) 4207, hep-th/0005172.

Model for neutrino mixing based on SO(10)

Noriyuki Oshimo
Institute of Humanities and Sciences and Department of Physics, Ochanomizu University,
Tokyo 112-8610, Japan
Abstract
Assuming grand unified theory (GUT) and supersymmetry, we propose a simple model which can
consistently accommodate the masses and mixings for quarks and leptons. The grand unified group
is SO(10), and 10, 120, and 126 representations are introduced for the Higgs superfields which give
masses to the quarks and leptons. The differences of masses and mixings between the quarks and the
leptons are attributed to the Higgs boson structure. Below the GUT energy scale, the model is the
same as the minimal supersymmetric standard model except its inclusion of dimension-5 operators
for the small neutrino masses. The renormalization group equations of the independent parameters
for the Higgs couplings with the quarks and leptons are given explicitly to connect the two energy
scales of GUT and electroweak theory.
PACS: 12.10.Dm; 12.15.Ff; 12.60.Jv; 14.60.Pq
Keywords: Neutrinos; SO(10); Higgs
1. Introduction
Implications of non-vanishing neutrino masses are accumulating from the experiments
for solar [1] and atmospheric [2] neutrinos which show neutrino oscillations. This is the
first experimental evidence that suggests physics beyond the standard model (SM). The
possibility of oscillation has also been examined by the neutrinos from nuclear reactors.
The negative result of CHOOZ [3] gives constraints on the masses and mixing allowed for
the neutrinos. The result of KamLAND [4] confirms the oscillations of the solar neutrinos.
It is a task for the extension of the SM to accommodate these experimental results. In
E-mail address: oshimo@phys.ocha.ac.jp (N. Oshimo).
doi:10.1016/S0550-3213(03)00554-6
N. Oshimo / Nuclear Physics B 668 (2003) 258272
259
particular, the extreme smallness of the masses and the largeness of the generation mixing
angles, both of which are observed in the experiments, should be accounted for naturally.
From a theoretical point of view, it is reasonable for the SM to be extended under
grand unified theory (GUT). Furthermore, supersymmetry may have to be introduced in
order that the SM be consistently embedded in GUT models. On the other hand, some of
the GUT models, such as those based on SO(10) [5], automatically contain right-handed
neutrinos. If they have large Majorana masses, the lightness of the neutrinos could be
naturally understood. The plausibility of supersymmetric GUT models as the extension of
the SM are then further strengthened.
One tough obstacle confronts GUT models, concerning the generation mixing of the
leptons. The lepton mixing is described by the MakiNakagawaSakata (MNS) matrix as
the CabibboKobayashiMaskawa (CKM) matrix does the quark mixing. Contrary to the
CKM matrix, the observed neutrino oscillations show that the off-diagonal elements of the
MNS matrix are not suppressed. However, since the leptons and the quarks are contained
in the same representations of the grand unified group, similar magnitudes are deduced, by
simple consideration, for corresponding elements of the two matrices. A successful GUT
model is required to naturally implement the difference of mixing between the quarks and
the leptons. Although there are various solutions advocated for the problem [6], rather
contrived schemes have been invoked.
Aiming at a natural explanation for the generation mixings of the leptons and the
quarks, we present a model base on SO(10) and supersymmetry in which the difference
is simply attributed to the Higgs boson structure [7]. This model includes 10, 120,
and 126 representations of the SO(10) group for the Higgs bosons which give masses
to the quarks and leptons. The 120 representation becomes the origin of the different
generation mixings. Large Majorana masses for the right-handed neutrinos are generated
by a vacuum expectation value (VEV) of the 126 representation. Below the GUT energy
scale, this model is described as the minimal supersymmetric standard model (MSSM)
with dimension-5 operators composed of the SU(2) doublet superfields for the left-handed
leptons and the Higgs bosons. The coefficients of the Higgs couplings with the quarks
and leptons evolve between the energy scales of GUT and electroweak theory. We obtain
the renormalization group equations for the eigenvalues of the coefficient matrices and
the independent arguments of the MNS and CKM matrices. Quantitative analyses are then
made on the neutrino oscillations. It is shown that the lepton mixing observed is compatible
with the quark mixing in certain ranges of the model parameters.
In SO(10) GUT models the Higgs bosons for the quark and lepton masses must be
contained in 10, 120, or 126 representations. The direct product for one of these three
representations with two spinor 16 representations for the quarks and leptons becomes
an SO(10) singlet. The model with two Higgs fields of 10 and 126 has been studied
extensively [8]. However, all the experimental results for the neutrino oscillations are only
marginally accommodated. The 10 and 126 representations make the Higgs couplings
symmetric on the generation indices, and their SU(2)-doublet components couple to
both the quarks and the leptons. On the other hand, the 120 representation leads to
antisymmetric Higgs couplings, and the SU(2)-doublet components which couple to the
quarks may be different from those to the leptons. By including 120, the correlation of the
masses and mixings between the quarks and the leptons becomes more flexible.
260
In Section 2 we summarize the interpretation of neutrino oscillations in the framework

of the model with three generations for leptons. In Section 3 the renormalization group
equations for the MSSM with the dimension-5 operators are obtained explicitly in terms
of independent parameters for the coefficients of Higgs couplings with quarks and leptons.
In Section 4 the GUT based on SO(10) and supersymmetry is discussed. After deriving in
general the relations of Higgs couplings for the quarks and the leptons, a plausible model
is specified. In Section 5 we make numerical analyses of the masses and mixing of the
neutrinos. Conclusions are given in Section 6.
2. Neutrino oscillations
The deficit of neutrinos from the sun and the anomaly for neutrinos from the atmosphere
can be explained by neutrino oscillations. The atmospheric neutrino observation gives that
the mass-squared difference and the mixing angle between the -neutrino and a certain
neutrino are given by
m2atm = (18) 103 eV2 ,
sin2 2atm > 0.85.
(1)
There are several oscillation scenarios for the solar neutrino problem. However, a most
likely solution is given by a large mixing angle between the e-neutrino and a certain
neutrino under the MikheyevSmirnovWolfenstein effect, suggesting the ranges
m2sol = (210) 105 eV2 ,
sin2 2sol = 0.50.9.
(2)
The measurements by KamLAND are consistent with these results. On the other hand, by
the experiment of CHOOZ the oscillation has not been observed for the e-neutrino, which
excludes the combined region
m2chooz > 1 103 eV2 ,
sin2 2chooz > 0.2.
(3)
These experimental measurements are taken into consideration in constructing our model.
We interpret the above results in the model with three generations of massive neutrinos.
In this framework the strength of the W -boson interaction for a neutrino and a charged
lepton depends on the generations. The mass eigenstate i for the neutrino of the ith
generation is related to the eigenstate j of interaction with the charged lepton of the j th
generation by
i = (VMNS )ij j ,
(4)
where VMNS denotes the MNS matrix. The survival probability for the interaction
eigenstate i after run of distance L with energy E is given by
3
2

2
2
m

P (i i ) =
(5)
(VMNS )ki exp i k L ,

2E
k=1
where mk represents the mass eigenvalue for the neutrino of the kth generation. The masssquared difference is hereafter written as
m2ij = m2i m2j .
(6)
261
We assume that the neutrino masses are hierarchical, m1 < m2 < m3 , and not
degenerated, similarly to the quarks and charged leptons.
In the CHOOZ experiment, the phases in Eq. (5) are found to satisfy the relations
(m221/2E)L (m231 /2E)L 1. The survival probability of the e-neutrino is expressed
by
2
2

m231
L.
P (e e ) = 1 4(VMNS )31 1 (VMNS )31 sin2
4E
Thus, the measured quantities are translated as

2
2
m2chooz = m231,
sin2 2chooz = 4(VMNS )31 1 (VMNS )31 .
(7)
(8)
For the atmospheric neutrino oscillation, the phases in Eq. (5) satisfy the relations
(m221/2E)L (m232 /2E)L and (m221 /2E)L 1. The survival probability of the
-neutrino is given by

2
2
m232
L.
P ( ) = 1 4(VMNS )32 1 (VMNS )32 sin2
4E
The experimental results are expressed by

2
2
m2atm = m232,
sin2 2atm = 4(VMNS )32 1 (VMNS )32 .
(9)
(10)
For the solar neutrino oscillation, the relations |(VMNS )31 |2 |(VMNS )21 |2 and
|(VMNS )31 |2 |(VMNS )11 |2 could hold from Eqs. (1), (3), (8), and (10). These constraints
make it possible to evaluate the survival probability of the e-neutrino by

2
2
m221
L.
P (e e ) = 1 4(VMNS )21 1 (VMNS )21 sin2
4E
Therefore, we obtain the equations

2
2
m2sol = m221,
sin2 2sol = 4(VMNS )21 1 (VMNS )21 .
(11)
(12)
These evaluations are used to discuss the predictions of the model.
3. Energy evolution
We assume that physics below the GUT energy scale is described by the MSSM
with the dimension-5 operators composed of the superfields for left-handed leptons and
Higgs bosons. This assumption is fulfilled in the model which is presented afterward. The
superpotential relevant to the quark and lepton masses is given by
W = W1 + W2 ,
ij
ij
ij
W1 = d H1 Qi D cj + u H2 Qi U cj + e H1 Li E cj + H.c.,
1
W2 = ij H2 Li H2 Lj + H.c.,
2
(13)
(14)
262
where H1 and H2 stand for the superfields for the Higgs bosons with hypercharges 1/2
and 1/2, respectively. Superfields are denoted by Qi , U ci , and D ci for the quarks and Li
and E ci for the leptons, in a self-explanatory notation, with i being the generation index.
The SU(3) group indices are understood. The dimension-5 superpotential W2 yields the
operators
+ i c + j
H2
L
1 ij H2
+ H.c.,
c
L
L=
(15)
j
0
0
i
2
H2
H2
eL
eL
+
0 represent the scalar components of the superfield H . At the
where H
and H
2
2
2
electroweak energy scale where the SU(2)-doublet Higgs bosons have non-vanishing
VEVs, these dimension-5 operators become tiny Majorana mass terms for the left-handed
neutrinos. The other operators arising from W2 cause negligible effects.
The coefficient matrices d , u , e , and are diagonalized by unitary matrices UL s
and UR s. The CKM matrix for the quarks and the MNS matrix for the leptons are defined
by
VCKM = ULu ULd ,
ULuT u URu = uD ,
VMNS = UL ULe ,
ULT UL
= ,
D
ULdT d URd = dD ,
ULeT e URe
= eD ,
(16)
(17)
where dD , uD , eD , and D stand for diagonalized matrices. Taking the VEVs of

electroweak symmetry breaking for positive, the diagonal elements of dD , eD , and D
should be positive, while those of uD should be negative. The quarks and leptons then
have positive masses. Those diagonal elements are expressed by di , ui , ei , and i .
A 3 3 unitary matrix has nine independent parameters. For the explicit expression of
VCKM or VMNS , we adopt the following parametrization [9]:
V = P+ V0 P ,
s1 s2 c3 + ei c1 c2
V0 = s c c ei c s
c1 s2 c3 ei s1 c2
c1 c2 c3 + ei s1 s2
s1 s3
c1 s3

P+ = diag exp(i1 ), exp(i2 ), 1 ,

P = diag exp(i3 ), exp(i4 ), exp(i5 ) ,
1 2 3
1 2
s2 s3
c2 s3 ,
c3
(18)
where ci = cos i and si = sin i (i = 1, 2, 3). Without loss of generality, the angles 1 ,
2 , and 3 can be taken to lie in the first quadrant. At the electroweak energy scale, the
numbers of the physical parameters for the CKM matrix and the MNS matrix are four
and six, respectively. However, we need the general form of parametrization in Eq. (18) to
discuss the energy dependencies of the CKM and MNS matrices.
The values of dD , uD , eD , D , VCKM , and VMNS evolve depending on the energy
scale. The renormalization group equations for these parameters and the gauge coupling
constants of SU(3) SU(2) U(1) close on themselves at the one-loop level. We give the
2 2 , 2 2 ,
equations of the independent parameters, where the inequalities da
ua
d3
u3
2
2
and ea e3 (a = 1, 2) are taken into account.
263
3.1. The gauge coupling constants
g4
dg32
= 3 32 ,
d
8
g4
dg22
= 22 ,
d
8
dg12 33 g14
=
.
d
5 8 2
(19)
3.2. The diagonalized Higgs coupling coefficients

2
2
2 2

ddi
di
16 2
7 2
2
2
2
2

= 2
g + 3g2 + g1 3d3 3di e3 (VCKM )3i u3 ,
d
3 3
15
8
(20)

2
2

ui
dui
16 2
13
2
2
2
2
= 2
g + 3g22 + g12 3u3
,
(21)
3ui
(VCKM )i3 d3
d
3 3
15
8

2
d2
9
2
2
2
ei = ei2 3g22 + g12 e3
(22)
3ei
3d3
,
d
8
5

2 2
i
di
3
2
= 2 3g22 + g12 3u3
(23)
(VMNS )i3 e3
.
d
8
5
3.3. The CKM matrix
2
dq1
= u32 s1 c1 s32 ,
d
16
(24)
2
dq2
= d32 s2 c2 s32 ,
d
16
(25)
2 + 2
dq3
= u3 2 d3 s3 c3 ,
d
16
dq
= 0,
d
dqa
= 0 (a = 15),
(26)
(27)
(28)
where qi and q stand for the arguments in V0 for the CKM matrix, with ci = cos qi and
si = sin qi , and qa represents an additional phase parameter in P+ and P .
3.4. The MNS matrix

2
dl1
= e3 2 s2 c2 c3 (BR + CR )c (BI CI )s ,
d
16

2
dl2
= e3 2 s2 c2 AR s32 + (BR + CR )c32 ,
d
16
(29)
(30)
264

2
dl3
= e3 2 s3 c3 BR c22 + CR s22 ,
d
16
2
e3
c12 s12
dl
2
=
s
(B
+
C
)s
s2 c2 c3
A
I
R
R
3
d
16 2
s1 c1
2

c s12
BI c 1
s2 c2 c3 + s22 c32 c22
s1 c1
2

c1 s12
2 2
2
s2 c2 c3 c2 c3 + s2 ,
+ CI c
s1 c1

2
dl1
= e3 2 (AI BI )c22 s32 + CI c32 s22 s32 ,
d
16

2
dl2
= e3 2 (AI CI )s22 s32 + BI c32 c22 s32 ,
d
16

2
e3
c1
c1
dl3
2
=
c3 (BR + CR )s s2 c2 + BI c s2 c2 c2 c3
d
16 2
s1
s1

c1
CI c s2 c2 + s22 c3 ,
s1

2
dl4
s1
s1
= e3 2 c3 (BR + CR )s s2 c2 BI c s2 c2 + c22 c3
d
16
c1
c1

s1
+ CI c s2 c2 s22 c3 ,
c1

2
dl5
= e3 2 s32 BI c22 + CI s22 ,
d
16
12 + 22
21 2
cos
2(
)
+
,
AR = 2
1
2
1 22
12 22
21 2
AI = 2
sin 2(1 2 ),
1 22
(31)
(32)
(33)
(34)
(35)
(36)
(37)
22 + 32
22 3
cos
2
+
,
2
22 32
22 32
22 3
BI = 2
sin 22 ,
2 32
BR =
32 + 12
23 1
cos
2
+
,
1
32 12
32 12
23 1
CI = 2
sin 21 ,
3 12
CR =
where li and l stand for the arguments in V0 for the MNS matrix, with ci = cos li , si =
sin li , c = cos l , and s = sin l , and la represents an additional phase parameter in P+
and P . The mutual dependencies of the independent parameters in energy evolution are
265
manifestly seen, thanks to the explicit expressions in terms of the parameters themselves.
For instance, the MNS matrix receives large quantum corrections if some of the neutrino
mass coefficients i are roughly degenerated [10].
4. Model
The grand unified group of our model is SO(10). Its spinor 16 representation contains
all the superfields for quarks and leptons of one generation, for both left-handed and righthanded components. The right-handed neutrinos are therefore naturally incorporated. The
decomposition of the direct product for two 16s is given by 16 16 = 10 + 120 + 126.
The Higgs superfields which give masses to the quarks and leptons must be assigned to 10,
120, or 126 representations. We introduce one superfield for each representation.
The superpotential relevant to the quark and lepton masses are given, in the framework
of SU(3) SU(2) U(1), by
5

5
Qi D cj + Li E cj + H10
Qi U cj + Li N cj
W = ij H10

1 45 i cj
ij
5
+ / H120
Qi D cj + Li E cj + H120
Q D 3Li E cj
3

2
5
i cj
45
i cj
+ 2H120 L N + H120 Q U
3

45 i cj

ij
5
+ H126 Q D 3Li E cj + H126
Qi U cj 3Li N cj

15 j
1
+ 6 LiT H126
(38)
L + 6 H126
N ci N cj + H.c.,
where superfields N ci for the right-handed neutrinos appear in addition to the superfields
Qi , U ci , D ci , Li , and E ci for the quarks and leptons. Higgs superfields are denoted by H s
with upper and lower indices showing transformation properties under SU(5) and SO(10),
45
5
5
5
, H120
, H120
, and H 45 are SU(2) doublets with hypercharge 1/2; H10
,
respectively: H10
126
5 , H 45 , and H 5 are SU(2) doublets with hypercharge 1/2; H 15 is an SU(2) triplet;

H120
120
126
126
and H 1 is an SU(2) U(1) singlet. Each superfield has been normalized. Owing to the
126
5 Qi U cj does not appear. The coupling

antisymmetric property of 120, the coupling H120
ij
ij
constants and are symmetric for the generation indices, while / ij are antisymmetric.
We can see from Eq. (38) the characteristic of the 120 representation. Any SU(2) doublet
in 10 or 126 couples both the quark and the lepton superfields. As a result, in Eqs. (13) and
(14), the Higgs coupling coefficients for the leptons become related to those for the quarks.
5 + H 45 )/2, (H 5 3 H 45 )/2,
On the other hand, for 120, four SU(2) doublets ( 3 H120
120
120
120
5 , and H 45 couple to Qi D cj , Li E cj , Li N cj , and Qi U cj , respectively. The Higgs
H120
120
coupling coefficients could be less constrained.
The SU(2)-doublet Higgs superfields for electroweak symmetry breaking are composed
of the superfields in 10, 120, 126, and some other representations. Among the possible
linear combinations of SU(2) doublets with the same hypercharge, only one doublet should
be kept light to satisfy the unification of the gauge coupling constants for SU(3) SU(2)
266
U(1). The two doublets with hypercharges 1/2 and 1/2 assume the role of the Higgs
superfields H1 and H2 in Eqs. (13) and (14). The other linear combinations must have
large masses and decouple from theory below the GUT energy scale. We express the Higgs
superfields by

45
45
5
5
+ C1 12 H120
+ C1 13 H120
+ C1 14 H126
+ ...,
H1 = C1 11 H10
(39)

5
5
45
5
+ C2 12 H120
+ C2 13 H120
+ C2 14 H126
+ ...,
H2 = C2 11 H10
(40)
where C1 and C2 represent unitary matrices. Some components of H1 and H2 belong to
the representations different from 10, 120, and 126, which are denoted by the ellipses. For
instance, one superfield of 126 is included in the model. Its SU(5)-singlet scalar component
has a large VEV to cancel the VEV of H 1 , and the VEVs of the auxiliary D fields
126
for SO(10) are kept small. This 126 representation contains SU(2) doublets, which may
become the components of H1 and H2 .
The matrices C1 and C2 should be determined by the Higgs potential at the GUT energy
scale. However, we take them as independent parameters in this paper, assuming that an
appropriate Higgs potential could be constructed. Since the Higgs potential contains also
the fields which are to break SO(10) correctly down to SU(3) SU(2) U(1), it is very
complicated to analyze the whole potential. The Higgs potential is also supposed to induce
the well-known split between the light SU(2) doublets and the heavy SU(3) triplets, as
well as the split between the light SU(2) doublets, H1 and H2 , and the other heavy SU(2)
doublets.
The superpotential in Eq. (13) is now determined. The coefficient matrices are given by

1
d = (C1 )11 + / (C1 )21 + (C1 )31 + (C1 )41 ,
(41)
3
2
u = (C2 )11 + /(C2 )31 + (C2 )41 ,
(42)
3

e = (C1 )11 + / (C1 )21 3 (C1 )31 3(C1 )41 .
(43)
The dimension-5 superpotential in Eq. (14) is induced by the interactions of N ci in
Eq. (38),
1
N ci N cj + H.c.,
W = ij H2 Li N cj + 6 ij H126
(44)
= (C2 )11 + 2/(C2 )21 3(C2 )41 .
(45)
If the scalar component of H 1

126
has a large VEV, the right-handed neutrinos and sneutrinos

receive large masses. The mass matrix of the neutrinos is given by
MR = 2 3 vS ,
(46)
1
with H = vS / 2. The mass-squared matrix of the sneutrinos is expressed as MR MR .
126
Exchanges of these particles lead to an effective superpotential given in Eq. (14),
= (MR )1 T .
The coupling
LiT H 15 Lj
126
(47)
in Eq. (38) could give Majorana masses to the left-handed
neutrinos, if the neutral scalar component of H 15 has a non-vanishing VEV. However,

126
267
this VEV has to be as small as the neutrino masses measured in experiments. Then, an
extreme fine-tuning of the Higgs potential would become inevitable. We therefore discard
this possibility, assuming that H 15 is heavy enough not to develop a non-vanishing VEV.
126
The coefficient matrices d , u , and e in Eq. (13) and in Eq. (14) at the GUT
energy scale are related to each other through Eqs. (41)(43), and (47). For independent
parameters we can take u + uT , d + dT , /, C1 , and C2 . Then the symmetric parts of the
coefficient matrices for the leptons are given by

3r1 + r4
4
d + dT +
u + uT ,
r1 r4
r1 r4

r
r
+ 3r4
4r
1
4
1
d + dT +
u + uT ,
+ T =
r1 r4
r1 r4

1
3 vS
r1
d + dT
u + uT ,
MR =
(C1 )41 r1 r4
r1 r4
(C2 )11
(C2 )41
,
r4 =
.
r1 =
(C1 )11
(C1 )41
e + eT =
(48)
(49)
(50)
The mass matrix MR is symmetric. The antisymmetric parts of the coefficient matrices
are given by

1
T
d d = 2 (C1 )21 + (C1 )31 /,
(51)
3
4
u uT = (C2 )31 /,
(52)
3

e eT = 2 (C1 )21 3 (C1 )31 /,
(53)
T = 4(C2 )21 /.
(54)
Depending on the structures of Eqs. (39) and (40), the 120 representation could contribute
exclusively to any one of d , u , e , and .
5 + H 45 )/2 and H 45 components in

We now make an assumption that the ( 3 H120
120
120
the Higgs superfields
H1 and H2 , respectively, can be neglected, taking the equations
(C1 )21 + (1/ 3 )(C1 )31 = (C2 )31 = 0. Then, the coefficient matrices for the quarks
become symmetric, d = dT , u = uT . Adopting a generation basis in which the coefficient
matrix for the up-type quarks is diagonal, we obtain the equations
d = VCKM
dD VCKM
,
u = uD .
(55)
The matrices d and u are determined by their eigenvalues and the CKM matrix. On the
other hand, the coefficient matrices for the leptons are given by
3r1 + r4
4
d +
u + 4/ ,
r1 r4
r1 r4
4r1 r4
r1 + 3r4
d +
u + 2r2 / ,
=
r1 r4
r1 r4
e =
(56)
(57)
268

r1
2 3 vS
1
MR =
d
u ,
(C1 )41 r1 r4
r1 r4
(C2 )21
r2 =
.
/ = (C1 )21 /,
(C1 )21
(58)
The matrices e , , and MR are expressed as linear combinations of d , u , and / . With

u being diagonal, the matrix d is roughly diagonal simultaneously. However, for the
matrix /,
only off-diagonal elements have non-vanishing values. The contribution of 120
may make the off-diagonal elements of e and/or non-negligible, which could enhance
the generation mixing for the leptons.
The independent model parameters at the GUT energy scale are given by the diagonal
matrices dD and uD , the CKM matrix VCKM , the ratio r1 , r2 , and r4 , the antisymmetric
matrix / , and the right-handed neutrino mass scale vS /(C1 )41 . At the electroweak energy
scale, the eigenvalues of d , u , and e are known experimentally, if the ratio tan of
the VEVs for H1 and H2 is given. The CKM matrix has been measured. The quantities
obtained experimentally for the neutrinos are the mass-squared differences and the mixing
angles. These observed quantities have to be accommodated by suitable values of the model
parameters at the GUT energy scale.
5. Numerical analyses
We show numerically that the observed neutrino oscillations can be described by our
model. Since the number of the model parameters at the GUT energy scale is large, a
systematical analysis in the whole parameter space is complicated. Instead, we demonstrate
the viability of the model by presenting two numerical examples within the ranges of real
values for r1 , r2 , r4 , and /.
In Figs. 1 and 2 the mixing parameters sin2 2atm , sin2 2sol , sin2 2chooz , and the ratio
of mass-squared differences m2sol/m2atm at the electroweak energy scale are shown as
functions of the model parameter r2 . The values of tan , r1 , r4 , and / are listed in Table 1,
with the sets (A) and (B) corresponding to Figs. 1 and 2, respectively. The other model
parameters uD , dD , and VCKM at the GUT energy scale are tuned to give the quark and
charged lepton masses and the CKM matrix at the electroweak energy scale shown in
Table 2. The CKM matrix can be expressed by the standard parametrization [11] and
the CP -violating phase of this parametrization is denoted by SP . These masses at the
electroweak energy scale for the quarks and charged leptons, as well as the magnitudes
of the CKM matrix elements, are consistent with the experiments [12]. The CP -violating
phase SP lies in the range allowed by all the CP violation phenomena observed in the
0 and B 0 B
0 systems [13]. The outcomes in Table 2 do not vary with r2 . The value
K 0 K
of vS /(C1 )41 determines the scale of and does not affect the presented four observables
for the neutrino oscillations.
We can see from Figs. 1 and 2 that the atmospheric and solar neutrino oscillations,
under the constraints from the CHOOZ experiment, are realized simultaneously for certain
parameter values. In Table 3 the resultant values of sin2 2atm , sin2 2sol , sin2 2chooz , and
m2sol/m2atm are explicitly given for the set (A) with r2 = 12.5 and for the set (B) with
269
Fig. 1. The mixing parameters and the ratio of mass-squared differences for the neutrinos at the electroweak
energy scale for the parameter set (A): (i) sin2 2atm , (ii) sin2 2sol , (iii) sin2 2chooz , (iv) m2sol /m2atm .
Table 1
The parameter sets (A) and (B) for Figs. 1 and 2, respectively
(A)
tan
r1
r4
/12
/13
/23
20
3.9
7.3
4.0 104
3.6 103
9.2 103
(B)
30
1.9
5.0
9.0 104
4.9 103
7.6 103
r2 = 7.5. If the mass scales of the right-handed neutrinos are put at vS /(C1 )41 = 1.0 1015
GeV in the set (A) and at vS /(C1 )41 = 6.0 1014 GeV in the set (B), the mass-squared
differences are given by m2sol = 2.5 105 GeV2 , m2atm = 5.5 103 GeV2 and by
m2sol = 8.7 105 GeV2 , m2atm = 1.6 103 GeV2 , respectively. Note that, for the
case (A), the mixing angle sin2 2chooz and the ratio m2sol/m2atm are predicted to be
around present experimental bounds. In fact, the value of the mass-squared ratio is outside
the allowed range of the recent experiment [4].
270
Fig. 2. The mixing parameters and the ratio of mass-squared differences for the neutrinos at the electroweak
energy scale for the parameter set (B): (i) sin2 2atm , (ii) sin2 2sol , (iii) sin2 2chooz , (iv) m2sol /m2atm .
Table 2
The masses of the quarks and charged leptons (in unit of GeV) and the CKM matrix (its elements and CP violating phase in the standard parametrization) at the electroweak energy scale

(VCKM )12
0.22
mt
1.8 102

(VCKM )13
mc
6.8 101
0.0036

(VCKM )23
2 103
0.041
mu
mb
ms
md
m
m
me
SP
3.0
9.3 102
5 103
1.8
1.0 101
5 104
1.3
Table 3
The mixing parameters and the ratio of mass-squared differences for the neutrinos at the electroweak energy scale
with r2 = 12.5 and r2 = 7.5 for the sets (A) and (B), respectively
(A)
sin2 2atm
sin2 2sol
sin2 2chooz
m2sol /m2atm
(B)
0.95
0.59
0.16
0.94
0.86
0.0026
0.0045
0.055
271
6. Conclusions
We have discussed the masses and mixings of quarks and leptons within the framework
of GUT coupled to supersymmetry. In GUT models the Higgs couplings for quarks and
leptons are closely related to each other. The large generation mixing for the leptons, which
is observed experimentally through the neutrino oscillations, cannot coexist trivially with
the small generation mixing for the quarks. Some natural explanation for the difference of
mixing between the quarks and the leptons is sought.
To solve the problem of generation mixing, we have proposed a model with SO(10)
and supersymmetry. This model is a simple extension of the minimal SO(10) model, and
includes a Higgs superfield of 120 representation, as well as two Higgs superfields of
10 and 126 representations. The energy evolution of the Higgs couplings with quarks
and leptons have also been taken into account between the energy scales of GUT and
electroweak theory.
By the simple enlargement of the Higgs sector, the masses and mixings of the quarks
and leptons are consistently accommodated without invoking much contrived schemes. The
120 representation makes the Higgs couplings different between the quarks and the leptons.
The small neutrino masses are traced back to large masses for the right-handed neutrinos
and sneutrinos generated by 126. The model parameters are constrained to correctly give
the quark and charged lepton masses and the CKM matrix. In spite of these constraints, the
experimental results for the neutrino masses and the MNS matrix can be described well in
certain regions of the parameter space. In SO(10) string GUT models, high-dimensional
representations, such as 120 and 126, are not easily realized [14], suggesting that our model
is not compatible with string theory.
Acknowledgements
This work is supported in part by the Grant-in-Aid for Scientific Research on Priority
Areas (No. 14039204) from the Ministry of Education, Science and Culture, Japan.
References
[1] B.T. Cleveland, et al., Astrophys. J. 496 (1998) 505;
GALLEX Collaboration, W. Hampel, et al., Phys. Lett. B 447 (1999) 127;
SAGE Collaboration, J.N. Abdurashitov, et al., Phys. Rev. C 60 (1999) 055801;
GNO Collaboration, M. Altmann, et al., Phys. Lett. B 490 (2000) 16;
Super-Kamiokande Collaboration, S. Fukuda, et al., Phys. Rev. Lett. 86 (2001) 5656;
SNO Collaboration, Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011301.
[2] Super-Kamiokande Collaboration, S. Fukuda, et al., Phys. Rev. Lett. 85 (2000) 3999.
[3] M. Apollonio, et al., Phys. Lett. B 466 (1999) 415.
[4] KamLAND Collaboration, K. Eguchi, et al., Phys. Rev. Lett. 90 (2003) 021802.
[5] G. Anderson, S. Dimopoulos, L.J. Hall, S. Raby, G.D. Starkman, Phys. Rev. D 49 (1994) 3660;
D.-G. Lee, R.N. Mohapatra, Phys. Rev. D 51 (1995) 1353;
L.J. Hall, S. Raby, Phys. Rev. D 51 (1995) 6524;
Y. Achiman, T. Greiner, Nucl. Phys. B 443 (1995) 3;
272
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
K.S. Babu, S.M. Barr, Phys. Rev. Lett. 75 (1995) 2088;

C.H. Albright, S. Nandi, Phys. Rev. D 53 (1996) 2699;
Z. Berezhiani, Z. Tavartkiladze, Phys. Lett. B 409 (1997) 220;
C.H. Albright, K.S. Babu, S.M. Barr, Phys. Rev. Lett. 81 (1998) 1167;
K. Matsuda, Y. Koide, T. Fukuyama, Phys. Rev. D 64 (2001) 053015;
M. Bando, M. Obara, Prog. Theor. Phys. 109 (2003) 995.
I. Dorsner, S.M. Barr, Nucl. Phys. B 617 (2001) 493, and references therein.
N. Oshimo, Phys. Rev. D 66 (2002) 095010.
K.S. Babu, R.N. Mohapatra, Phys. Rev. Lett. 70 (1993) 2845;
B. Brahmachari, R.N. Mohapatra, Phys. Rev. D 58 (1998) 015001;
T. Fukuyama, N. Okada, JHEP 11 (2002) 011.
S.G. Naculich, Phys. Rev. D 48 (1993) 5293.
N. Haba, N. Okamura, Eur. Phys. J. C 14 (2000) 347;
J.A. Casas, J.R. Espinosa, A. Ibarra, I. Navarro, Nucl. Phys. B 573 (2000) 652.
K. Hagiwara, et al., Phys. Rev. D 66 (2002) 010001.
See, e.g., H. Fusaoka, Y. Koide, Phys. Rev. D 57 (1998) 3986.
G.C. Branco, G.C. Cho, Y. Kizukuri, N. Oshimo, Nucl. Phys. B 449 (1995) 483.
K.R. Dienes, Nucl. Phys. B 488 (1997) 141.

The Higgs mass as a function of

the compactification scale
Riccardo Barbieri, Guido Marandella, Michele Papucci
Scuola Normale Superiore and INFN, Piazza dei Cavalieri 7, I-56126 Pisa, Italy
Abstract
We calculate to a few percent precision the Higgs potential in a model with supersymmetry broken
by boundary conditions on an extra-dimension, compactified to a segment of length L, and a top
quark quasi-localized on one of the two boundaries. 1/L alone, in the range 24 TeV, determines the
Higgs mass, in the range 110125 GeV, and the spectrum of gauginos, higgsinos and of the thirdgeneration squarks. Lower values of 1/L cannot be excluded, with a progressive delocalization of
the top quark.
PACS: 12.60.Jv; 11.10.Kk; 11.15.Ex
1. Introduction
1/2
In the Standard Model (SM) the sensitivity of the Fermi scale GF

to the cut-off
1/2
strongly suggests the presence of new physics close to GF
itself. Since long time
the possibility is contemplated that such new physics be represented by supersymmetric
1/2
particles with masses of order GF . Relatively more recently, the scale of a compactified
extra dimension (one or more) has also been put forward to play the same role [1]. In
this paper we study, as precisely as possible, the consequences for electroweak symmetry
breaking (EWSB) from the merging of these two ideas in a definite scheme, as defined
in [2].1
This contamination has a main motivation: it allows to describe supersymmetry
breaking in terms of a single parameter, L, the length of the segment to which the
E-mail address: guido.marandella@sns.it (G. Marandella).
1 See also [3,4].
doi:10.1016/S0550-3213(03)00573-X
274
R. Barbieri et al. / Nuclear Physics B 668 (2003) 273292
extra-dimension is compactified. Although L is the limit where supersymmetry

is recovered, the compactification scale is not the scale above which a supersymmetric
spectrum approximately appears, due to the presence of the KaluzaKlein (KK) modes.
There is no such scale. In fact, the KK modes play a crucial role in rendering more
precise the supersymmetric softening of the ultraviolet divergences. The introduction of
the usual soft supersymmetry-breaking parameters is avoided. Nor there is any need of
introducing a supersymmetric -parameter, since only one Higgs field gets a non-vanishing
vacuum expectation value consistently with all phenomenological requirements and with
the supersymmetry constraints [5].
For a theory that is designed to render the Fermi scale insensitive to the cut-off, the
1/2
connection between GF
and the scale of new physics should be determined as neatly as
possible. This is highly desirable not only theoretically but also from a phenomenological
and pragmatic point of view, since the possibility to test the theory at the Tevatron or at
the LHC crucially depends on this property. This motivates a careful study of the Higgs
potential. A peculiar feature that emerges from this study is the gap that can result between
1/2
1/L and GF
or, even more so, the Higgs mass. Although anticipated in [2], the precise
assessment of this property as the equally precise determination of the Higgs mass as a
function of 1/Lour main goals in this paperrequire the two loop calculation of the
Higgs potential described below.
The paper is organized as follows. The model is defined in Section 2. In Section 3 the
Higgs potential is calculated to two loop accuracy in s = gs2 /(4) and t = yt2 /(4)
with the top supermultiplet exactly localized at y = 0. In Section 4 a quasi-localized
top is introduced and both the Higgs mass and the determination of the Fermi scale
are studied as function of 1/L = 24 TeV. The issue of the uncertainties is explicitly
addressed in Section 4.3. In Section 5 the possibility is considered of a lower value of
the compactification scale, allowing for the presence of a common mass for the Higgs
hypermultiplets. The spectrum of the model is given in Section 6. The summary and the
conclusions are drawn in Section 7.
2. The model
We consider a 5D, SU(3) SU(2) U (1) invariant theory with every multiplet of the
SM, gauge, matter or Higgs, promoted to a N = 1, 5D supermultiplet. The fifth dimension
is meant to be compactified on a segment parameterized by y [0, L].2
As advocated long ago by Scherk and Schwarz [6], we break supersymmetry by
boundary conditions at y = 0, L that distinguish the different fields in every supermultiplet.
Given the 5D gauge theory, there is a single choice of these boundary conditions, consistent
with the symmetries of the theory and with supergravity, that leads to a massless spectrum
identical to the SM one [5]. This non-trivial property is at variance with what happens when
global supersymmetry is spontaneously broken in a 4D theory, where a hidden sector and
a mediation mechanism of supersymmetry breaking have to be introduced. With these
2 In Refs. [2,4,5] R = 2L/ has been used.
275
Table 1
c , c )
Boundary conditions for gauge (A , , c , ), matter (M , M , M
M
c
c
and Higgs (H , H , H , H ) multiplets at y = 0 and y = L
M , H , A
M , H ,
c , c , c
M
H
c , c ,
M
H
(+, +)
(+, )
(, +)
(, )
boundary conditions, summarized in Table 1, every field acquires, in the effective 4D

picture, a definite KK tower. The first KK states of the SM particles are at /L, whereas
the extra particles implied by supersymmetry and/or Poincar invariance are at /(2L) or
higher.
A theory thus defined has a divergent FayetIliopoulos term induced at one loop [7],
which can be canceled by introducing a second Higgs-like hypermultiplet, Hc . In this
case one obtains, at tree level, two (massless) Higgs-like scalars as in the Minimal
Supersymmetric Standard Model (MSSM). At variance with the MSSM, however, the
second Higgs-like scalar does not need to get a vacuum expectation value (VEV) and plays
no role in EWSB.
Without affecting the massless spectrum, it is consistent to introduce suitable supersymmetric mass terms M for the matter hypermultiplets [8]. They deform the massive spectrum
as well as the wave function in y of every state [4]. The wave function of the massless states
(the matter fermions) becomes an exponential, 0 (y) |M|1/2 eMy+(M|M|)L/2 , so that
in the large ML limit, |ML| 1, the massless fermion gets localized on one of the two
boundaries, y = 0 or y = L, depending on the sign of M. At the same time a chiral N = 1,
4D supermultiplet is recovered with one scalar becoming also massless and localized. All
the other states get a heavy mass, increasing like M. These hypermultiplet masses, in general different for every matter multiplet, are non-renormalized parameters and could be
fixed in a more fundamental theory. Not to introduce again a FayetIliopoulos term they
have to satisfy simple conditions [8]: e.g., a common mass for the quark hypermultiplets of
a given generation, MQi , for the lepton hypermultiplets, MLi , and for the Higgs-like hypermultiplets MH . For concreteness we stick to this configuration in the following, although
EWSB is essentially independent from this choice.
The top Yukawa coupling, necessarily localized at one of the two boundaries, say y = 0,
is the source of EWSB, which is therefore influenced by the mass terms of the top quark
hypermultiplets Q and U , MQ3 M. We shall take them quite closely localized at y = 0,
with ML 1 for most of the paper. For this reason, it is a significant approximation to
consider first what happens when Q and U are exactly localized.
3. The Higgs potential with a top localized on one boundary

Symbolically, the reference Lagrangian is
N=1
L = LN=1
5D (gauge Higgs) + L4D (Q U )(y),
(1)
276
where the 4D, N = 1 supersymmetric Lagrangian for the Q, U chiral supermultiplets

includes the top Yukawa coupling

N=1
4
V
4
V
2
d t H QU + h.c. .
L4D (Q U ) = d Q e Q + d U e U +
(2)
We ignore for the time being possible kinetic terms for the gauge and Higgs multiplets also
localized at the boundaries (see Section 4.3).
With the boundary conditions in Table 1, after the integration over y, the tree level
potential for the real part of the zero mode of the Higgs field is
g2 + g 2 4
h .
V tree h2 =
32
(3)
We are interested in the effective potential V (h2 ) which, expanded around h = v, gives

V h2 2vV v 2 h + V v 2 + 2v 2 V v 2 h2 .
(4)
Hence

V v2 = 0
(5)
1/2
is the equation which determines v or the Fermi scale GF

m2h = 4v 2 V v 2
and
is the physical Higgs squared mass. We aim to an accuracy of a few percent in

The 1-loop electroweak contribution to V has been computed in [9]
2
7 (3)(3g 2 + g 2 )
102

v Vew
(0) =
=
0.93
Vew
128 2L2
L2
(6)
V
and V .
(7)
up to corrections of relative order (gvL)2 .

The one-loop (g 2 + g 2 )t correction to V , in localized approximation for the top,
coincides, at logarithmic level, with the same correction in the MSSM for appropriate
values of the stop masses m2Q , m2U (see below). Its contribution to the Higgs squared mass
is [10]
m2Q m2U

3
GF m2t MZ2 log
.
m2h g 2 + g 2 t =
m4t
4 2 2
(8)
Our task is then to compute to the relevant order of approximation the (t , s )dependent contributions to Eqs. (5) and (6).
3.1. (t , s )-corrections. General expressions
The corrections of interest contain logs of the fine structure constants, s and t , which
arise from the infrared behavior of the integrals. This is due to the masslessness of the
and U
at tree level, which become massive only at one loop. To deal properly
squarks Q
(a)
277
(b)
Fig. 1. The diagrams that contribute to the Higgs potential at order s t and t2 in superfield notation.
with this situation we introduce, as infrared regulators, the squark masses m20,Q and m20,U
for the two multiplets, also localized at y = 03 . These masses will be sent to zero at the end
of the calculation. With these masses the potential of interest, Vtop (v 2 ), has a one-loop
contribution

d 4p 2
1-loop
log p + m20,Q + m20,t + log p2 + m20,U + m20,t
Vtop = 3
(2)4

2 log p2 + m20,t ,
(9)
where m0,t = yt v/ 2 is the unrenormalized top quark mass, and a two-loop contribution
2-loop
which arises from the diagrams in Fig. 1, in superfield notation, and is explicitly
Vtop
given in Appendix A.
The propagators for all the components of the Q, U supermultiplets are in the
background of the field h. Since Q and U propagate in ordinary Minkowsky space, at
y = 0, the only components of the Higgs and gauge supermultiplets, H and V , that
contribute in Fig. 1 are those with (+) boundary conditions at y = 0. Up to trivial kinematic
factors, after the Wick rotation, their propagators are proportional to

S +,+ (k) coth k 2 L,

S +, (k) tanh k 2 L.
(10)
From
1-loop
Vtop = Vtop
2-loop
+ Vtop
(11)
the contributions to v 2 V (v 2 ) and v 4 V (v 2 ) are finite after reexpressing m0,t , m20,Q ,

m20,U in (9) in terms of the physical masses4

m2t = m20,t 1 + BU + BQ + 2Zyt ,
(12a)
3 This also helps in keeping right track of the order in the loop expansion.
4 m is the mass of the stop-left. For the mass of the sbottom-left, which only enters in the two-loop diagrams,
Q
we use m2B = m2Q m2t .
278

m2Q = m20,Q 1 + BQ + m20,t 1 + BQ + BFU + m2Q ,
(12b)

Q
m2U = m20,U 1 + BU + m20,t 1 + BU + BF + m2U
(12c)
and after making an expansion in the one-loop quantities: m2U,Q , the one-loop corrections
to the squared masses, Zyt , the Higgs-toptop vertex correction, and the various B-factors,
the corrections to the wave functions of the various fields. To the precision of interest all
these quantities are computed at zero external momenta, except for those involved in (12a)
where we use for mt the running mass at p2 = m2t . The explicit expressions for all these
factors are given in Appendix B.
Finally, as anticipated, the fictitious bare masses m20,Q , m20,U are set to zero. Note that
they only appear in m2Q , m2U in Eqs. (12a)(12c). To leading order in t and s one
has [11]
m2U = m2t +

7 (3) 8s + 6t
+ O t2 , t s ,
2
24
L
(13a)
m2Q = m2t +

7 (3) 8s + 3t
+ O t2 , t s .
24
L2
(13b)
3.2. (t , s )-corrections. Results

To calculate explicitly the corrections of interest we make a systematic expansion of
(v 2 ) and v 4 V (v 2 ) in , and m2 , m2 , m2 , all formally treated as quantities of
v 2 Vtop
t
s
top
t
Q
U
the same order. In so doing, care must be taken in avoiding spurious infrared divergences.
(v 2 ) we keep terms quadratic in these quantities, whereas in v 4 V , which
In v 2 Vtop
top
starts quadratic in m2t , we keep those cubic terms which also include at least a factor of
(v 2 ) the 5D propagators in Eqs. (10) are
log m2Q L2 , log m2U L2 or log m2t L2 . In v 2 Vtop
(v 2 ), which is more convergent in the
crucial in giving a finite result, whereas in v 4 Vtop
ultraviolet (but less convergent in the infrared), the 5D propagators can be approximated
with their low momentum expansion. We find
2

v 2 Vtop
v ; mQ , mU
=

3m2t
2
mQ 2 log(mQ L) c + m2U 2 log(mU L) c
2
16

2m2t 2 log(mt L) c ,
2

v ; mQ , mU
v 4 Vtop

3m4t
mQ mU
=
log
8 2
m2t
(14)
279

mQ
3m4t m2t GF
2 mQ
+
log(m
2
log
L)
log
t
mt
mU
16 3 2

mU
+ log2 mU mQ L2 + log2
mt

mQ
8s
log2
4 log2 (mt L)
3
mt

mQ mU
2 mU
4 log(mt L) log
+
log
mt
m2t

m2t GF
10 log(mQ L) + 6 log(mt L) + 12 log(mU L)
4 2

16s
mQ mU
3 log(mt L) ,
(1 6 log 2) log
3
m2t
(15)
where c = 4 2 (12 log 2)/7 + 2 (3)/(3) 1.33.

(v 2 ) coincides with the result given in Ref. [2]
In view of Eqs. (13), the result for v 2 Vtop
pole
in logarithmic approximation. Numerically, for mt
= 174.3 5.1 GeV, we find
2
102

Vtop
(16)
v = (0.73 0.05) 2
L
at 1/L = 3 TeV, with a negligible residual dependence on 1/L of the coefficient in the
range 1/L = 24 TeV due to (mt L)2 terms. Note the near cancellation in V (v 2 ), at
the 20% relative level, between the electroweak term, Eq. (7), and the two loop (t , s )contribution, Eq. (16), with a predominance of the first positive term. To the extent that this
calculation is reliable (see below), the Higgs potential has a positive curvature at h = 0, so
that EWSB does not take place with exactly localized Q, U multiplets.
For V (v 2 ) no simple connection can be established between this theory and a suitably
defined MSSM, because of the difference in the way higgsinos get a mass: by a -term
in the MSSM, by pairing with conjugate states here. Since V (v 2 ) is ultraviolet sensitive
in the MSSM, this makes an essential difference. On the contrary, the stronger ultraviolet
convergence of the momentum integrals in V (v 2 ), relative to V (v 2 ), makes it closer
to the analogous quantity in a suitably defined MSSM. The leading m4t -term coincides.
The same is also true for the next order terms m4t GF m2t and m4t s , in leading log2
approximation, if one compares Eq. (15) at 1/L = mQ = mU MS with the MSSM result
for tan = , At = 0 and all superpartners at MS [12].
4. The case of a quasi-localized top

As anticipated in Section 2, with the Q, U hypermultiplets not exactly localized, all their
components acquire a KK tower of states with M-dependent masses. Most importantly, for
finite ML, this tree level spectrum is not supersymmetric. As a consequence, already at
one loop the Higgs potential receives a non vanishing contribution from Q, U exchanges,
1-loop
Vtop (v 2 ; ML), calculated in [4]. For ML 2.5 the slope of this potential, of negative
280
Fig. 2. The mass of the left-handed stop as a function of the localization parameter ML, compared with the same
mass in the ML limit.
sign, dominates over the electroweak contribution in Eq. (7) and triggers EWSB. For
1-loop
ML 1.5, however, Vtop
is not a sufficiently accurate description of the top-stop
contribution to the Higgs potential, as we shall see explicitly: in localized approximation,
1-loop
vanishes, whereas the top-stop contribution at two loop does not, as
ML = , Vtop
seen in the previous section.
The most important effect of a finite ML, compared to ML = , is on the tree
level mass of the lightest squarks in the corresponding KK tower. Although this mass
converges exponentially to mt for the stops, or to zero for the sbottom left, its effect is still
significant at ML 23. In Fig. 2 we compare mQ (ML) with the corresponding quantity,
mQ () = mQ , Eq. (13b), in localized approximation. The radiative one-loop contribution
is only weakly sensitive to ML and dominates over the tree level mass. Nevertheless the
deviation of mQ (ML) from mQ in the region of interest is sizable. A similar situation
holds for mU (ML). To account for this effect in the potential, we consider the first and
second derivatives of Vtop in Eqs. (14) and (15) with mQ and mU replaced by mQ (ML)
and mU (ML) respectively, so that

Vtop v 2 , ML Vtop v 2 ; mQ (ML), mU (ML) .
(17)
A better approximation of Vtop (of its derivatives) is in fact the following

1-loop 2
2-loop 2
v , ML + Vtop
v , ML ,
Vtop v 2 , ML Vtop
(18)
where
2-loop 2

v , ML

tree
= Vtop v 2 ; mQ (ML), mU (ML) Vtop v 2 ; mtree
Q (ML), mU (ML)
Vtop
(19)
properly subtracted to avoid double counting with the one-loop term. For ML 2,
however, the difference between (17) and (18) is negligible.
281
Fig. 3. Slope of the full radiative Higgs potential, as discussed in the text, versus the top localization parameter
pole
ML, compared to the one-loop approximation, for mt = 174.3 5.1 GeV.
4.1. Determination of the Fermi scale

With the inclusion of the tree level contribution from Eq. (3), the minimum equation (5)
reads

MZ2
= V v 2
(20)
4
which must be viewed as a relation between the compactification scale 1/L and the
localization parameter ML. Fig. 3 shows

2
2

L2 V v 2 = L2 Vew
(21)
v L2 Vtop
v ; ML
with the electroweak contribution given in Eq. (7) and the top contribution from Eq. (17)
or (18). After rescaling by 1/L2 , V (v 2 ) has no significant residual dependence on 1/L in
the region of interest, for 1/L 2 TeV. For these values of 1/L, it is (MZ L)2 103 , so
is clearly inadequate. The flattening of V (v 2 ) at
that the one loop approximation to Vtop
and V , is important in reducing
ML 3, due to the partial cancellation between Vtop
ew
the tuning between ML and 1/L. Also in view of the uncertainties to be discussed below,
this same flattening of V (v 2 ) makes the precise relation between ML and 1/L uncertain.
This has little influence, however, on the relation between the Higgs mass and 1/L, as we
discuss shortly.
4.2. The Higgs mass as a function of 1/L
With the inclusion of the correction in Eq. (8) and of the top contribution from Eqs. (17),
(6) reads

2
mQ (ML)m2U (ML)
3
2
2
2
GF mt log
mh = MZ 1
m4t
4 2 2

+ 4 2 GF v 4 Vtop
(22)
v 2 ; ML .
By means of the relation between ML and 1/L as determined from the minimum equation,
mh is plotted in Fig. 4 as function of 1/L only, for three different values of the pole
282
pole
Fig. 4. Higgs mass as function of 1/L for mt
= 174.3 5.1 GeV.
pole
Fig. 5. Higgs mass for mt

= 174.3 GeV at different fixed values of the top localization parameter. The result
with the full Higgs potential is compared with the one only including the standard top correction at one loop.
pole
top mass, mt = 174.3 5.1 GeV. In Fig. 5 we show the band of values that would
be obtained if ML were not related to 1/L by the minimum equation (20), but kept fixed at
values between 2 and 4. This shows that the precise relation between ML and 1/L is almost
irrelevant in order to determine the connection between mh and the compactification scale.
In the same Fig. 5 we also compare the Higgs mass, calculated on the basis of Eq. (22),
with the one that would be obtained from a minimally improved lowest order formula
2

mQ (ML)m2U (ML)
3 2
2
2
4
mh (naive) = MZ +
(23)
GF mt log
4
m4t
and ML = 24. This comparison makes clear that the improved two-loop potential is
essential for a better determination of mh .
283
4.3. Uncertainties
In this way one is naturally lead to the issue of the overall uncertainties on mh , in
particular on Fig. 4. We believe that the deviation from the localized approximation is
properly accounted for by Eqs. (17), (18). The real question, then, is the validity of
Eqs. (14), (15) with an exactly localized top. To this end, it is necessary to discuss
the possible role of other operators than those in Eq. (2). Among those operators, the
potentially most important ones are the kinetic terms of the Higgs and the gauge multiplets
localized on the boundaries. In the notation of Eq. (1), the constants zH , za in

L = (y) zH d 4 H eV H +
(24)
za d 2 W(a) W(a) + h.c.
a
with a = SU(3), SU(2), U (1) and similar terms localized at y = L, have to be treated as
additional parameters. For them we need an estimate or a natural assumption for their size.
Since H and W(a) are 5D fields, their z-factors in Eq. (24) have dimension of an inverse
mass. Pure dimensional analysis leads to an estimate z 1/, where could either be
the scale np at which some of the couplings become non-perturbative or a cut-off scale
cutoff np below which our 5D theory represents the low energy effective description of
some more fundamental theory. By studying the evolution of the Yukawa couplings of the
top and the bottom, localized at y = 0 and y = L respectively, one finds that np L = 610
at ML 3 [4,5]. For equal masses, (equal localization), of the quark hypermultiplets of
the third generation, M, it is in fact b that becomes non-perturbative first. To produce the
physical top and bottom masses, both t and b depend on ML and become comparable
for ML = 22.5, whereas for higher ML b becomes progressively bigger.5
Based on these considerations/assumptions, we take for the dimensionless Z-factors
z
Z (1015)%.
(25)
L
On the other hand the introduction of these z-factors affects the radiative Higgs potential
or the radiative squark masses m2Q , m2U only at quadratic order in the dimensionless
Zs. On this basis we expect that the calculations of the Higgs and squark masses are
correct within a few %. A more critical quantity is V , since the electroweak and the top
contributions cancel quite accurately against each other and are renormalized by different
factors (1 + O(Z 2 )). As already mentioned in Section 4.1, although making uncertain the
relation between ML and 1/L, this has little influence, however, on the relation between
mh and 1/L.
5. Lower values of the compactification scale

An interesting question is what happens for 1/L lower than 2 TeV. Below this value
the Higgs mass gets nominally lower than the experimental bound. Had we drawn the
5 The influence of on the Higgs potential is negligible because the integrals in V are dominated by low
b
momenta [2].
284
Fig. 6. Expected range of the Higgs mass with the inclusion of a hypermultiplet Higgs mass and a maximum fine
pole
tuning in the slope of the potential at 10% level. The darker (lighter) area is for mt = 174.3 (174.3 5.1).
same figure as Fig. 4 for lower 1/L, mh would have reached values as low as 105 GeV
at 1/L 600 GeV to grow again up to 130 GeV at 1/L 300 GeV. At the same time,
ML progressively decreases from about 2 at 1/L 2 TeV, to zero at the lowest value
of 1/L 300 GeV, where one makes contact with the Constrained Standard Model of
Ref. [5]. We do not show this plot because in the intermediate region of 1/L 1 TeV or
1-loop
ML 1, our calculation is not fully reliable: Vtop
does not clearly dominate over the
2-loop contribution, which, on the other hand, is only to be trusted for sufficiently large
ML.
To make sense of the model at these lower values of 1/L, one has also to make sure
that the potential with two Higgs doublets, h and hc , does not get destabilized, given the
absence of a bilinear term hhc . This is possible, without introducing a FI term, by adding
a small common mass, |MH L| 0.1 for the Higgs hypermultiplets.
A non-zero MH does not affect the physical Higgs mass, through V , but only the
determination of the Fermi scale, via V . With an extra term (MH L), present in the
right-hand side of Eq. (21), 1/L is not tied anymore to ML, which can in turn vary in
a range consistent with a moderate amount of fine tuning. The result of this is shown in
Fig. 6. Different values of ML are used, but always in such a way that no fine tuning
1/2
occurs stronger than 10% in the determination of GF . The rise in mh is due to the
1-loop
stronger influence, for low ML, of Vtop , a fact which has no correspondence in the
MSSM [4,5].
Taking into account the uncertainties mentioned above, a value of mh marginally
consistent with the experimental lower bound of 114 GeV cannot be excluded in the
entire region of 1/L. The existence of independent lower bounds on 1/L becomes then
of relevance. This in turn crucially depends on the masses MQi , MLi for the quark and
lepton hypermultiplets of the different generations, i = 1, 2, 3 [2]. Depending on these
masses, modifications of the Fermi constant [13] and/or flavor changing neutral currents
effects [14] are possible by tree level exchanges of KK states of the gauge bosons. These
285
effects, however, are minimized (made to vanish at tree level) by the choice MLi = 0,
MQi = M. It is in fact possible that the most important constraint on 1/L comes from
the electroweak parameter. Simple dimensional analysis shows that the correction to the
parameter from the towers of top and bottom states scales as = cv 2 /(L2 4 ) with a
dimensionless coefficient that could perhaps be as large as 30 [5]. Although this correction,
at fixed L = 610, vanishes as L2 for small L, a value of 1/L as large as 1 TeV might be
needed to suppress this correction below the uncertainty of the observed value.
6. Spectrum
The most peculiar feature of the spectrum is the heaviness of gauginoshiggsinos with
a mass /(2L). The lightest superpartner is therefore a squark or a slepton, most likely
charged, which can be stable or practically stable if a small U (1)R -breaking coupling is
present. The first KK states corresponding to the SM particles are at /L. The masses of
the sfermions of charge Q and hypercharge Y are given by
m2 = m2tree + m2rad + Y m2Z Qm2W ,
(26)
where mtree is the tree level mass, including the Yukawa contribution, and mrad is the oneloop contribution, as in Eqs. (13). As seen in Section 4, both mtree and mrad depend upon
the corresponding localization parameter ML. For ML = 0, mtree = /(2L) dominates
and the sfermion becomes degenerate with the gauginos and the higgsinos. For ML 1,
mrad dominates and rapidly approaches the localized limit where
1
,
L TeV
) = (382 GeV) 1 ,
mrad (U
L TeV
= (310 GeV) 1 ,
mrad (D)
L TeV
1
,
mrad (
L) = (137 GeV)
L TeV
= (79 GeV) 1 .
mrad (E)
L TeV
= (370 GeV)
mrad (Q)
(27a)
(27b)
(27c)
(27d)
(27e)
Up to the D-term effects in Eq. (26), these are lower values for the sfermion masses.
As we have seen, the localization parameter of the third generation squarks is correlated
with the compactification scale by EWSB. For 1/L 2 TeV, Eq. (27a) gives the mass
2 + m2t )1/2 and sbottom mrad (Q),
split by mtree (t) = mt ,
of the left-handed stop (mrad (Q)
2
2
1/2

whereas (mrad (U ) + mt ) is the mass of the right-handed stop.
For lower values of 1/L, if a tree level mass of the Higgs hypermultiplets is present
(see Section 5), the third generation squarks can be progressively delocalized. In this case
Eqs. (27) give a lower bound, with the overall masses that can go up to 800 GeV even for
1/L below 1 TeV.
286
Eqs. (26) and (27d) apply also to the masses of the scalars in the second Higgs
hypermultiplet Hc , without a vev and with gauge couplings identical to those of a lefthanded slepton multiplet. Not to undo EWSB, the Higgs hypermultiplets are always
almost fully delocalized, |MH L| 0.1. Eq. (27d) is strictly valid for 1/L 2 TeV and
) 270 GeV. For smaller values of 1/L a tree level Higgs mass can play a role and
mrad (H
can raise the total mass of the charged and neutral scalars in Hc , relative to Eq. (27d), by
about 100 GeV at 1/L 1 TeV.
7. Summary and conclusions
We have made an accurate calculation of the Higgs potential in a theory of EWSB
triggered by the top Yukawa coupling, where supersymmetry is broken by boundary
conditions on a fifth dimension. The calculability of the Higgs potential rests on the fact
that supersymmetry breaking is described in terms of a single parameter, the length L of the
compactified extra dimension. This, in turn, is possible because the zeroth order spectrum
has all the extra particles implied by supersymmetry and/or 5D Poincar invariance at
/(2L) or higher.
To define the phenomenology of this proposal, a central issue is the determination of
the range of the compactification scale 1/L. Although not with certainty, the electroweak
precision tests most likely want a value of 1/L above 1 TeV. This is a manifestation of the
usual little hierarchy problem: the apparent need of a gap between the scale of physics
1/2
or the Higgs mass, which is low according to the same
that triggers EWSB and GF
1/2
is influenced by the level
EWPT. In the present case, the relation between 1/L and GF
of localization of the top quark and, to a lesser extent, by a hypermultiplet Higgs mass.
These are both unrenormalized supersymmetric parameters.
The two-loop calculation of the Higgs potential shows that 1/L can go up to 4 TeV
without a significant amount of fine tuning and with the third generation Yukawa coupling
maintaining perturbativity up to = (610)/L. This is made possible by the localization
of the top near the boundary of the 5th dimension where its Yukawa coupling is present. In
this way the otherwise dominant one-loop top contribution to the curvature of the potential
V (v 2 ) gets exponentially suppressed. Furthermore the two-loop diagrams involving the
top Yukawa couplings in exactly localized approximation contribute to V (v 2 ) with a
negative term that almost cancels the positive electroweak contribution, so that L2 V (v 2 ) =
1/2
O(103 ) for ML 2.53.5. This is the basis for allowing values of 1/L larger than GF
by about one order of magnitude.
To make sure that this is at all consistent with the phenomenological constraints, it is
crucial to compare the expected Higgs mass with the current lower bound. To this purpose,
with a reasonable assumption on the uncertainties induced by boundary Z-factors, it has
been necessary to extend to two loop also the calculation of V (v 2 ). For large values of
1/L = 24 TeV, the Higgs mass is reduced with respect to the naive expectation and is in
the 110125 GeV range. For lower values of 1/L the Higgs mass is more uncertain but can
be above the experimental bound up to the lowest possible values of 1/L.
The spectrum implied by this picture of supersymmetry and electroweak symmetry
breaking has a few characteristic features. The first KK states corresponding to the SM
287
particles are at /L. Also all gauginos and higgsinos are heavy, with a mass at /(2L).
Therefore the lightest superpartner is a sfermion, most likely charged, which can be stable
or practically stable for collider experiments [2,5]. Finally one also expects relatively light
scalars, one charged and one neutral, belonging to the Hc hypermultiplet, introduced to
cancel the FI term. They are coupled to the gauge bosons as a lepton doublet but have
unknown couplings to quarks and leptons.
Acknowledgements
We thank P. Slavich for useful comments concerning the two-loop corrections to the
Higgs mass in the MSSM. This work has been partially supported by MIUR and by the EU
under TMR contract HPRN-CT-2000-00148.
Appendix A. The 2-loop contribution to the Higgs potential

In this appendix we give some details of the 2-loop contribution to the Higgs potential
2-loop
Vtop .
At the 2-loop level there are two contributions. The first one comes from the expansion,
in Eq. (9), of the one-loop corrections in Eqs. (12) after expressing m0,t , m0,Q and m0,U in
terms of the renormalized masses mt , mQ and mU , respectively. The second contribution
is a pure 2-loop correction and corresponds to the diagrams of Fig. 1 in terms of the top
superfields U, Q, the Higgs superfield H and the SU(3) vector superfield V . In localized
approximation for U and Q, only V and H are 5-dimensional superfields.
Defining
C(x) = x coth(x),
T (x) = x tanh(x)
(A.1)
the pure 2-loop gauge correction (arising from Fig. 1(b)) is given by the following
expression
V2-loop,gauge

d 4 p d 4 q T (qL)
2
= 4gs
(2)4 (2)4 q 2

2p2
2q 2
((p q)2 + m2Q )(p2 + m2t ) ((p q)2 + m2Q )(p2 + m2t )
+
2(p q)2
((p q)
+ m2Q )(p2
2p2
+ m2t )
(p2 + m2t )((p q)2 + m2U )

d 4 p d 4 q C(qL)
+ 4gs2
(2)4 (2)4 q 2
2q 2
(p2 + m2t )((p q)2
2
+ m2U )
2(p q)
(p2 + m2t )((p q)2 + m2U )
288
4
((p q)2 + m2Q )
q2
(p2 + m2Q )((p q)2 + m2Q )
p2
(p2 + m2Q )((p q)2 + m2Q )
2q 2
(p2 + m2t )((p q)2 + m2t )

2(p q)2
(p2 + m2t )((p q)2 + m2t )
(p q)2
(p2 + m2Q )((p q)2 + m2Q )
2p2
(p2 + m2t )((p q)2 + m2t )

8m2t
(p2 + m2t )((p q)2 + m2t )
q2
(p2
4
+
+ m2U ) (p2 + m2U )((p q)2 + m2U )
p2
(p2 + m2U )((p q)2 + m2U )
(p q)2
(p2 + m2U )((p q)2 + m2U )

.
Analogously the pure 2-loop Yukawa correction (arising from Fig. 1(a)) is given by the
following expression
V2-loop,Yuk.

= 3yt 2
d 4 p d 4 q T (qL)
(2)4 (2)4 q 2
(p q)2
((p q)2 + m2Q )(p2 + m2t )
q2
((p q)2 + m2Q )(p2 + m2t )
p2
((p q)2 + m2Q )(p2 + m2t )
p2
((p q)2 + m2Q m2t )(p2 + m2t )
(p q)2
((p q)2 + m2Q m2t )(p2 + m2t )
q2
((p q)2 + m2Q m2t )(p2 + m2t )
q2
p2
1
+
p2 + m2U
(p q)2 (p2 + m2U ) (p q)2 (p2 + m2U )
p2
((p q)2 + m2t )(p2 + m2U )
q2
((p q)2 + m2t )(p2 + m2U )

d 4 p d 4 q C(qL)
+ 3yt 2
(2)4 (2)4 q 2
(p q)2
((p q)2 + m2t )(p2 + m2U )
p2
(p2 + m2Q )((p q)2 + m2Q )
+
+
+
+
289
m2Q m2t
(p2 + m2Q )((p q)2 + m2Q )
p2
(p2 + m2Q )((p q)2 + m2Q m2t )
m2Q m2t
(p2 + m2Q )((p q)2 + m2Q m2t )
q2
p2
1
+
p2 + m2t
(p q)2 (p2 + m2t ) (p q)2 (p2 + m2t )
p2
(p2 + m2t )((p q)2
q2
+ m2t )
+ m2t )((p q)2

q2
+ m2t )
(p2
((p q)2 + m2Q )(p2 + m2U )
(p q)2
(p2 + m2t )((p q)2
p2
+ m2t )
1
+ m2U
q2
((p q)2 + m2Q m2t )(p2 + m2U )

m2t + m2U
(p q)2
+
+
.
(p2 + m2U )((p q)2 + m2U ) (p2 + m2U )((p q)2 + m2U )
In the 2-loop potential, we have used the physical (renormalized) quantities m2t , m2U , m2Q
because the corrections are of higher order. When one takes the derivatives of the potential
one has to remember that

d
v 2 2 = m2t
(A.2)
+
+
dv
m2t
m2U
m2Q
because all the 3 masses depend on the VEV v.
The integrals in p can be performed analytically. Then, to get the leading logarithmic
contributions as L 0, one can use the asymptotic behavior of the q-integrand functions.
In this way one gets the results given in Eqs. (14), (15).
Appendix B. 1-loop renormalization functions at order t , s
In order to compute the physical masses in Eqs. (12a)(12c), the one loop corrections
O(t , s ) to the propagators of the U , Q-multiplets and to the Yukawa vertex are needed.
At order t the propagators of the top, stop and auxiliary field get corrected from the
exchange of the Higgs supermultiplet H and a U (or Q) quark supermultiplet.
These corrections can be parameterized as usual

= i p2 BU,Q p2 m2U,Q ,
290

= i/
pBU,Q p2 ,
U,Q 2
= iBF
p ,
where we have used a superfield notation in the loop and an Euclidean external momentum p. The Yukawa vertex receives no direct correction at order t because of the nonrenormalization properties of the superpotential.
All the quantities defined above, to the order yt2 , are given by

1
1
d 4q
(B.1a)
C(qL)
+
,
(2)4
q 2 + m2Q + m2t
q 2 + m2Q

1
d 4q
,
BFQ (0) = yt2
(B.1b)
C(qL)
(2)4
q 2 + m2Q + m2t

m2t (m2t + m2Q )
q 2 (m2t + m2U )
d 4q 1
BQ (0) = yt2
C(qL)
(2)4 q 2
(q 2 + m2t + m2U )3 (q 2 + m2t + m2Q )3
4

q + 3m2t q 2
(B.1c)
T (qL)
,
(q 2 + m2t )3

d 4q 1
BU (0) = yt2
(2)4 q 2

2 2
q (mt + m2Q )
q 2 m2Q
m2t (m2t + m2U )
C(qL)
+
(q 2 + m2t + m2Q )3 (q 2 + m2t + m2U )3 (q 2 + m2Q )3

4
1
q + 3m2t q 2
,
+
T (qL)
(B.1d)
q2
(q 2 + m2t )3

y2
q2
q 2 + 2m2t
d 4q 1
Q
B (0) = t
+
T
(qL)
C(qL)
,
2
(2)4 q 2
(q 2 + m2t )2
(q 2 + m2U + m2t )2
BFU (0) = yt2
BU (0) =
yt2
2
(B.1e)

2

2
1
1
q + 2mt
+
C(qL)
(2)4 q 2
(q 2 + m2t )2 q 2

q2
q2
+
T (qL)
,
(q 2 + m2Q + m2t )2 (q 2 + m2Q )2
(B.1f)
d 4q

m2Q
= yt2

q 2 + m2Q
q2
d 4q 1
+
C(qL)
(2)4 q 2
q 2 + m2Q + m2t
q 2 + m20,u + m2t

2q 2
T (qL)
,
q 2 + m2t
d 4q 1
(2)4 q 2

q 2 + m2U
q2
q2
C(qL)
+1+
+
q 2 + m2U + m2t
q 2 + m2q + m2t
q 2 + m2Q

2q 2
T (qL)
+2 ,
q 2 + m2t
291
(B.1g)
m2U = yt2
(B.1h)
where the functions C(x) and T (x) are given in (A.1). The integration over the momentum
q has to be performed on Euclidean space.
At order s only the propagators of the top and the stops, but not of their auxiliary
fields, get corrected from the exchange of the SU(3) gauge supermultiplet V and of a
quark supermultiplet. Performing the calculation in the WessZumino gauge there is also
a direct correction to the Yukawa interaction, so that

= i p2 BU,Q p2 m2U,Q ,

= i/
pBU,Q p2 ,

= iyt Zyt p2 .
Parameterizing these gs2 -corrections as in the yt2 case, one has

q 2 + 2(m2i + m2t )
8
q 4 + 3q 2m2t
d 4q 1
T
(qL)
C(qL)
,
Bi (0) = gs2
3
(2)4 q 2
(q 2 + m2i + m2t )2
(q 2 + m2t )3
(B.2a)

4q 1
2 + 2m2
2
4
q
d
q
t
+ T (qL)
Bi (0) = gs2
C(qL)
,
3
(2)4 q 2
(q 2 + m2t )2
(q 2 + m2i + m2t )2
(B.2b)
292

q2
d 4q 1
C(qL)
T
(qL)
,
(2)4 q 2
q 2 + m2t

16
q2
d 4q 1
Zyt (0) = gs2
C(qL)
,
3
(2)4 q 2
(q 2 + m2t )2
m2i =
16 2
g
3 s
(B.2c)
(B.2d)
where i = U, Q.
All the quantities defined above are regular in the IR except BU . Because we are
interested only in the logarithmic contributions to V (v 2 ), we can evaluate the B and BF
functions at vanishing external momentum. Instead, the functions B and Zyt , involved in
Eq. (12a), have to be evaluated at p2 = m2t . Given their expressions at p2 = 0 one has to
add

t
s
+
,
BQ m2t BQ (0) =
(B.3)
6
16

t
s
d 4q
BU m2t BU (0) =
+
+ 4t
6 16
(2)4

2q
4 log
2
2
1
1
mt + 4q +mt
(B.4)
,
3/2
2q 4
q 2 (4q 2 + m2t )
mt (4q 2 + m2t )

4s
(2 3 log 2).
Zyt m2t Zyt (0) =
(B.5)
3
Note that the IR divergence in (B.4) cancels the one in BU (0).
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
I. Antoniadis, C. Munoz, M. Quiros, Nucl. Phys. B 397 (1993) 515, hep-ph/9211309;
A. Delgado, A. Pomarol, M. Quiros, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
[2] R. Barbieri, L.J. Hall, G. Marandella, Y. Nomura, T. Okui, S.J. Oliver, M. Papucci, hep-ph/0208153.
[3] D. Marti, A. Pomarol, Phys. Rev. D 66 (2002) 125005, hep-ph/0205034.
[4] R. Barbieri, G. Marandella, M. Papucci, Phys. Rev. D 66 (2002) 095003, hep-ph/0205280.
[5] R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.
[6] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
[7] D.M. Ghilencea, S. Groot Nibbelink, H.P. Nilles, Nucl. Phys. B 619 (2001) 385, hep-th/0108184.
[8] R. Barbieri, R. Contino, P. Creminelli, R. Rattazzi, C.A. Scrucca, Phys. Rev. D 66 (2002) 024025, hepth/0203039.
[9] I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410.
[10] A. Brignole, Phys. Lett. B 281 (1992) 284.
[11] A. Delgado, A. Pomarol, M. Quiros, Ref. [1]
[12] R. Hempfling, A.H. Hoang, Phys. Lett. B 331 (1994) 99, hep-ph/9401219.
[13] A. Strumia, Phys. Lett. B 466 (1999) 107, hep-ph/9906266.
[14] A. Delgado, A. Pomarol, M. Quiros, JHEP 0001 (2000) 030, hep-ph/9911252.

General properties of non-commutative field theories

Luis lvarez-Gaum, Miguel A. Vzquez-Mozo 1
Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Received 16 May 2003; received in revised form 27 June 2003; accepted 1 July 2003
Abstract
In this paper we study general properties of non-commutative field theories obtained from the
SeibergWitten limit of string theories in the presence of an external B-field. We analyze the
extension of the Wightman axioms to this context and explore their consequences, in particular, we
present a proof of the CPT theorem for theories with spacespace non-commutativity. We analyze as
well questions associated to the spin-statistics connections, and show that non-commutative N = 4,
U(1) gauge theory can be softly broken to N = 0 satisfying the axioms and providing an example
where the Wilsonian low energy effective action can be constructed without UV/IR problems, after a
judicious choice of soft breaking parameters is made. We also assess the phenomenological prospects
of such a theory, which are in fact rather negative.
PACS: 11.10.Nx; 03.70.+k; 11.15.-q
1. Introduction and motivation

Non-commutative geometry has had a profound influence in Mathematics (cf. [1]). Also
in Physics it has had an effect in a number of subjects ranging from applications to the
integer and fractional quantum Hall effects [2,3] in condensed matter physics [4], to the
non-commutative formulation of the standard model by Connes and Lott [5] (see also [6]
for a review).
In string theory the use of non-commutative geometry was pioneered by Witten [7]
in his formulation of open string field theory. Compactifications of string and M-theory
on non-commutative tori were studied in [8]. Shortly after this, Seiberg and Witten [9]
E-mail addresses: luis.alvarez-gaume@cern.ch (L. lvarez-Gaum), miguel.vazquez-mozo@cern.ch
(M.A. Vzquez-Mozo).
1 On leave from Fsica Terica, Universidad de Salamanca, Salamanca, Spain.
doi:10.1016/S0550-3213(03)00582-0
294
L. lvarez-Gaum, M.A. Vzquez-Mozo / Nuclear Physics B 668 (2003) 293321
realized that a certain class of quantum field theories on non-commutative Minkowski

spacetimes can be obtained as a particular low energy limit of open strings in the presence
of a constant NSNS B field (see also [10]). This result generated a flurry of activity
in the study of quantum field theories in non-commutative spaces (see [11] for reviews).
Part of the interest has been aimed at getting new insights into the regularization and
renormalization of quantum field theories in this novel framework. In fact, many features
of ordinary (commutative) field theories have found rich analogues in the non-commutative
context [12].
The non-commutative field theories obtained from string theory via the SeibergWitten
limit are neither local nor Lorentz invariant, since the fields in the action appear multiplied
with Moyal products. Locality, together with Lorentz invariance, have been traditionally
considered to be two of the holy principles in a quantum field theory. With few exceptions,
there has been little motivation to try to extend the principles of quantum field theory to
non-local (or non-Lorentz invariant) theories. Although in general allowing for non-locality
in quantum field theory creates havoc, the SeibergWitten limit yields a very specific class
of non-local theories where, together with some unusual features, some of the desirable
properties of local theories are preserved. Therefore it has been a challenge to understand
the consequences, both theoretical and phenomenological, of these theories [13].
The first unexpected property of non-commutative field theories was pointed out by
Minwalla et al. [14]. These authors realized that quantum theories on non-commutative
spaces are afflicted from an endemic mixing of ultraviolet and infrared divergences. Even in
massive theories the existence of ultraviolet divergences induce infrared problems [15]. As
a consequence the Wilsonian approach to field theory seems to break down: integrating out
high-energy degrees of freedom produces unexpected low-energy divergences, inducing in
the infrared operators of negative dimension. This lack of decoupling of high energy modes
seem to doom these theories from any phenomenological perspective (see Section 5).
In ordinary Quantum Field Theory there are important consequences that follow from
the general principles of relativistic invariance and locality, which do not necessarily extend
to the non-commutative case. Results like the CPT theorem [1618] and the spin-statistic
connection [19] will not necessarily hold. Similarly, questions concerning the existence of
an S-matrix, its unitarity and the notion of asymptotic completeness are in need of drastic
revisions. In Refs. [2023] the CPT invariance of non-commutative field theories was
studied, and it was concluded that the CPT theorem holds, both in the case of spacespace
and timespace non-commutativity. However, this analysis deals with the tree-level action
and it is not sensitive to possible problems arising from quantum corrections. In the case of
non-commutative field theories problems might appear in the form of unitarity violations
or UV/IR mixing. In particular, the mixing of scales may jeopardize the tempered nature
of the Wightman functions as distributions.2 These issues might demand a revision of the
proof of the CPT theorem presented in [23].
In the present paper we will study in detail some general properties of non-commutative
quantum field theories, such as the CPT theorem and the spin-statistics connection, as well
2 A rigorous definition of the class of functions on which a non-commutative field theory should be built has
been given in [24].
295
as the possibility of constructing theories that are well-defined in the infrared. We will
propose an axiomatic formulation leading to a proof of the CPT theorem along the lines
of the one given by Jost for ordinary theories [17,18]. In this axiomatic formulation the
vacuum expectation value of the Heisenberg fields should define tempered distributions.
This mathematical condition imposes non-trivial restrictions on the correlation functions
both in the ultraviolet and the infrared. In order to give meaning to non-commutative
gauge theories in the infrared we present a detailed analysis of non-commutative U(1)
gauge theory with N = 4 supersymmetry softly broken to N = 0 in a way that preserves
finiteness in the ultraviolet and that leads to a well-defined theory at low energies. The
resulting theory provides correlation functions that behave like tempered distributions,
while in the infrared we recover the free Maxwell theory. We show that the theory defined
in this way satisfies the proposed axioms and assess its phenomenological viability.
This paper is organized as follows: in Section 2 we briefly review the SeibergWitten
limit, and analyze heuristically when CPT-invariance and unitarity of the S-matrix are
expected to hold. In Section 3 we adapt the Wightman axioms to the non-commutative
context. In particular, we analyze the issue of microscopic causality [17]. In Section 4 we
show that with the modified axioms it is still possible to prove the CPT theorem but, in
the general case, the connection between spin and statistics is not guaranteed. In Section 5
we analyze the softly broken N = 4 non-commutative U(1) gauge theory and show that
it is possible to formulate it in such a way that the adapted axioms are satisfied at least in
perturbation theory. We also make some general remarks on the possible phenomenological
perspectives of non-commutative field theories. Finally, our conclusions are presented in
Section 6.
A final remark is in order before closing this introduction. We are going to study
the class of non-commutative field theories obtained from the SeibergWitten limit of
string theories in the presence of a constant B-field. These theories are defined quantummechanically in perturbation theory in terms of their Feynman rules which follow from the
parent string theory. For reasons to be explained in the next section, we consider here only
the case of spacespace non-commutativity. However, from a purely field-theoretical point
of view, other procedures can be envisaged to quantize them. In particular, the authors
of Ref. [25], motivated by the work of [26], proposed a different way to look at noncommutative theories that leads to a unitary S-matrix and, if uncertainty relations for the
spacetime coordinates are implemented, they claim to obtain not only a unitary theory but
also one that is ultraviolet-finite. In [27] the authors start with Dysons formula to define
the Green functions, and by a careful analysis of the passage from time-ordered products
to Wick products, they conclude that the Feynman rules change when there is timespace
non-commutativity in a way that preserves perturbative unitarity. For spacespace noncommutativity the Feynman rules are the same as those obtained from string theory via the
SeibergWitten limit.
2. Heuristic considerations
Non-commutative field theories, i.e., theories with ordinary products replaced by Moyal
products, are from a quantum point of view theories of dipoles [28]. Since the elementary
296
excitations are extended objects, the resulting theory is non-local and the scale of nonlocality is set by the length of these dipoles.
This fact is easy to visualize in the cases where non-commutative theories arise as
effective description of the dynamics in a certain limit. In Ref. [9] it was shown how noncommutative field theories are obtained as a particular low-energy limit of open string
theory on D-brane backgrounds in the presence of constant NSNS B-field. In this case,
the endpoints of the open strings behave as electric charges in the presence of an external
magnetic field B resulting in a polarization of the open strings. Labelling by i = 1, . . . , p
the D-brane directions, the difference between the zero modes of the string endpoints is
given by (B0i = 0) [10]
Xi = Xi (, 0) Xi (, ) = (2 )2 g ij Bj k pk ,
(2.1)
where g is the closed string ( -model) metric and p is the momentum of the string. In
the ordinary low-energy limit ( 0, g , B fixed) the distance |
X| goes to zero and
the effective dynamics is described by a theory of particles, i.e., by a commutative quantum
field theory.
There are, however, other possibilities of decoupling the massive modes without
collapsing at the same time the length of the open strings. Seiberg and Witten proposed
to consider a low energy limit 0 where both Bij and the open string metric

Gij = (2 )2 Bg 1 B ij
(2.2)
are kept fixed. Introducing the notation ij = (B 1 )ij , the separation between the string
endpoints can be expressed as
Xi = ij Gj k pk ,
(2.3)
fixed in the low energy limit. The resulting low-energy effective theory is a noncommutative field theory with non-commutative parameter3 ij .
The SeibergWitten limit can be viewed as a stringy analog of the projection onto the
lowest Landau level in a system of electrons in a magnetic field. Indeed, the condition
fixing the open string metric G in the low-energy limit is equivalent to considering a
large B-field limit in which the interaction of the external field with the end-points of
the string dominates the worldsheet dynamics. In more physical term, Eq. (2.1) can be
interpreted as a competition between the Lorentz force exerted on the endpoints of the
string and the tension that tends to collapse the string to zero size. The SeibergWitten limit
corresponds to making the string rigid by taking the tension to infinity (thus decoupling the
excited states) while at the same time keeping the B-field large in string units, in such a
way that the length of the string is kept constant.
The previous analysis was confined to situations in which the B0i components are
set to zero. This results in a non-commutative field theory with only spacespace noncommutativity. From a purely field-theoretical point of view it is possible to consider also
3 Starting with type IIA/IIB string theory in the presence of N coincident Dp-branes, the resulting theory
after the SeibergWitten limit is a (p + 1)-dimensional U(N ) non-commutative super-YangMills theory with
16 supercharges.
297
non-commutative theories where the time coordinate does not commute with the spatial
ones, i.e., i0 = 0. In this case, however, non-locality is accompanied by a breakdown of
unitarity reflected in the fact that the optical theorem is not satisfied [29,30]. In addition
there is no well-defined Hamiltonian formalism (see, however, [25]).
Non-commutative field theories with timespace non-commutativity cannot be obtained
from string theory in a limit in which strings are completely decoupled. One can try
to obtain these kind of theories by looking for a decoupling limit of string theory in
D-brane backgrounds in the presence of B-fields with B0i = 0. That the existence of such
a decoupling limit is not immediate can be envisaged by noticing that having B0i = 0 is
equivalent to a non-vanishing electric field on the brane, and that in this case the electric
field has to be smaller that the critical value [31]
E < Ecrit =
1
.
2
(2.4)
For fields larger than Ecrit the background decays via the formation of pairs of open
strings in a stringy version of the Schwinger mechanism. A partial decoupling limit can
be achieved in the limit E Ecrit while keeping and the string tension fixed. In this
limit closed strings decouple from open strings while the latter ones remain in the spectrum
[32]. Actually, the optical theorem can be formally restored by taking into account the
interchange of undecoupled string states in intermediate channels, although they are both
tachyonic and have negative norm [30].
In order to avoid these problems, in the following we restrict ourselves to theories with
spacespace non-commutativity. Then the existence of a parent string theory provides a
good guiding principle to study their properties. The parent string theory also provides a
ultraviolet completion where non-locality is realized by string fuzziness.
The CPT theorem in string theory has been investigated by several groups [33]. If the
parent string theory satisfies the CPT theorem in perturbation theory, since the constant
background B-field is CPT-even, it is reasonable to expect that the non-commutative
quantum field theory obtained in the SeibergWitten limit should also preserve CPT. At
the level of the non-commutative field theory it is also expected to have CPT-invariance
for theories with 0i = 0. As discussed above, these kind of non-commutative theories
preserve perturbative unitarity. In ordinary quantum field theory there is an intimate
connection between unitarity and CPT-invariance. Indeed, if the condition of asymptotic
completeness holds, it can be seen using either the YangFeldman relation [34] or the
HaagRuelle scattering theory [35] that the S-matrix can be written in terms of the CPT
operator of the complete theory and the corresponding one for the asymptotic theory 0
(see also [36])
S = 1 0 .
(2.5)
The unitarity of the S-matrix follows then from the antiunitarity of and 0 . Therefore,
a theory with CPT invariance is likely to be unitary.
Motivated by these heuristic arguments we proceed to construct the CPT theorem along
the lines of the proof given in [18]. As a first step we attempt an axiomatic formulation of
non-commutative field theories as a modification of the Wightman axioms [3739].
298
3. Modified constructions
Compared to ordinary theories, two basic features of non-commutative theories are
their non-locality and the breaking of Lorentz invariance due to the presence of the
antisymmetric tensor . In the following we will take it of the form
0
0
0
e
0
0
0

= e
(3.1)
,
0
0
0
m
0
0 m 0
with e , m R. The Wightman axioms have to be modified in order to accommodate
non-commutativity and reduced Poincar invariance. Eventually we will be interested in
theories with spacespace non-commutativity e = 0. Nevertheless, for most of this section
we will keep e explicitly.
3.1. Symmetries of non-commutative field theories
The commutation relations for the coordinates [x , x ] = i are not preserved
by the Lorentz group O(1, 3) although they are indeed left invariant by the group of
rigid translations x x + a with a R. Using4 (3.1) it is straightforward to
find that the largest subgroup of O(1, 3) leaving invariant the commutation relations is
SO(1, 1) SO(2), where the SO(1, 1) factor acts on the electrical coordinates xe =
(x 0 , x 1 ) whereas SO(2) rotates the magnetic ones xm = (x 2 , x 3 ). In the case of space
space non-commutativity (e = 0) this group is enhanced to O(1, 1) SO(2). Since we
will be mostly concerned about this latter case, we will take the symmetry group of the
non-commutative theory to be P = [O(1, 1) SO(2)] T4 ,where T4 is the group of
translations.
The representation theory of O(1, 1) SO(2) shares many common features with that
of the Lorentz group O(1, 3). Representations of SO(2) are parametrized by the angle
[0, 2]. The O(1, 1) factor has a richer structure. As in the standard case, it has four
sheets
L : det = 1;
L :
00 1,
det = 1;
(3.2)
00 1.
(3.3)
L+
The elements of the four sheets can be expressed as a product of an element of

and one
of the three discrete transformations inverting one or the two electric coordinates
L+ = L+ Ist ,
where
Ist =
L = L+ Is ,
1 0
,
0 1

Is =
L = L+ It ,
1 0
,
0 1
(3.4)

It =
1 0
.
0 1
(3.5)
4 For a general the unbroken subgroup of O(1, 3) is the one preserving both 0i and " j k . Hence, in
ij k
general, the Lorentz group is completely broken unless it has the form of Eq. (3.1).
299
Each element of L+ is parametrized by a single number R, the rapidity of the boosts

along x 1

cosh sinh
=
.
(3.6)
sinh cosh
The transformations of xe are specially simple introducing null-coordinates x =
1 (x 0
2
x 1 ). These transform according to
L+ : x e x
(3.7)
together with
Ist x = x ,
Is x = x ,
It x = x .
(3.8)
Similar to four dimensions, the complexification of O(1, 1) connects continuously
different sheets. In particular L+ and L+ can be connected by a family of transformations
with complex rapidity + i where 0 . Denoting L+ = L+ L+ SO(1, 3) we

find
L+ (C) = C C {0},
(3.9)
where the action of L+ (C) on the coordinates is given by x z1 x , z C . In

particular, this result implies that the transformation inverting xe can be continuously
connected with the identity using transformations of L+ (C). At the same time xm can
also be inverted by a 180 degrees SO(2)-rotation. This result is central in the proof of the
CPT theorem (see Section 4).
Representations of SO(1, 1) are labelled by a helicity R and those of SO(2) by
(j )
the angular momentum j (2j Z). Any quantity in the (, j ) representation of
O(1, 1) SO(2) transforms as
(j )
(j )
e ej .
In P we can build two Casimir operators with P
2 2
2 2
s P0 P1 ,
t P2 + P3 .
(3.10)
(3.11)
In perturbative calculations in non-commutative field theories these two invariants appear

usually in the combinations
p2 = s t,
p p = e2 s + m2 t.
(3.12)
Since by definition t 0, the representations of P can be classified according to the

sign of s into massless (s = 0), massive (s > 0) and tachyonic (s < 0). Since
P P = O(1, 3) T4, it is important to notice that the type of the representations of P do
not necessarily coincide with those of P . In particular, from (3.12), we see that massive
and massless representations of P can be tachyonic when interpreted as representations
of P . On the other hand, any tachyonic representation of P will be tachyonic with
respect to P , but not the other way around.
300
3.2. Microcausality
The second issue that has to be studied in our attempt to find an axiomatic formulation
of non-commutative field theories is the notion of microcausality. In ordinary theories
causality is implemented by demanding that fields should commute or anticommute outside
their relative light cone. For non-commutative theories there is no structure preserving the
four-dimensional Lorentz group O(1, 3) and the concept of a light cone is lost.
From our discussion of the symmetries of non-commutative field theories it follows
that causality, if preserved at all, must be defined with respect to the O(1, 1) factor of the
symmetry group. This means that the light cone x 2 = 0 is actually replaced by the light
wedge V+ = {x R1,3 |xe2 = 0} (see Fig. 1). This suggests a change in the concept of
microcausality by replacing the light cone by the light wedge, so fields will either commute
or anticommute whenever their relative coordinate xy satisfies (xe ye )2 < 0.
In the following, we study a weaker version of this condition, namely under which
conditions the vacuum expectation value of the commutator of two non-commutative
scalar fields vanishes outside the light wedge. To achieve this we construct a Klln
Lehmann representation for |[(x), (y)]| (with | the unique vacuum state of the
theory) using invariance under the group P. Therefore, we introduce the [O(1, 1)SO(2)]invariant measure:
d(p) =

2

d 4p 0
p (2) pe2 2 (2) pm
2 .
4
(2)
(3.13)
Using a basis of the Hilbert space {|j, pe , pm }, where j is a collective index denoting any
other set of quantum numbers needed to specify the state, we write the closure relation
Fig. 1. The causal wedge V+ .
1 = || +
d(pj )|j, pe , pm j, pe , pm |.
301
(3.14)
Starting with |(x)(y)| and inserting the identity as in (3.14) one arrives, after
some algebra, at

| (x), (y) |

2

0
d 2
d 2 2
0
3 , Im GF xe ye , 2
= i sign x y
2
2
0

J0 2 (
(3.15)
xm ym )2 .
Here J0 (x) is a Bessel function of the first kind, and GF (x, m2 ) is the two-dimensional
Feynman propagator for a free scalar field of mass m,
2

d pe
i
GF x e y e , m 2 =
(3.16)
eipe (xe ye ) .
(2)2 pe2 m2 + i"
Finally, the spectral function 3( 2 , 2 ) is given by
2

(2) 2 j2 (2) 2 j2 |(0)|j, pe , pm .
3 2 , 2 =
(3.17)
Using the invariance of the theory under P it is straightforward to show that the overlaps
||(0)|j, pe , pm |2 only depend on j2 and j2 , the eigenvalues of the Casimir operators
Pe2 and Pm2 corresponding to the state |j, pe , pm .
Up to this point we have only used the symmetry properties of the theory, without any
reference to the fact that we might be dealing with a non-commutative theory. Actually,
from Eqs. (3.15) and (3.17) it is possible to extract some general consequences. Taking
into account the property of the two-dimensional Feynman propagator

Im GF xe ye , m2 = 0, for (xe ye )2 < 0 and m2 0,
(3.18)
we conclude that the vacuum expectation value of the commutator vanishes outside
the light wedge whenever the theory does not contain in its Hilbert space tachyonic
representations of O(1, 1).
If there are O(1, 1)-tachyonic states in the spectrum, i.e., if the spectral function
3( 2 , 2 ) has support for 2 < 0, the imaginary part of the Feynman propagator does not
vanish and in general |[(x), (y)]| will not be zero outside the light wedge. In the
latter case, the modification of microcausality proposed here will not be satisfied (cf. [30]).
Particularizing this result to the case of non-commutative field theories we conclude
that, in general, the adapted notion of microcausality is not preserved by theories with
timespace non-commutativity. For them poles in the two-point function at p p < 0
induce a support of 3( 2 , 2 ) for 2 < 0. This also occurs for theories with spacespace
non-commutativity but containing O(1, 1)-tachyons, as it is the case for non-commutative
QED [40]. In Section 5 we will study how to construct theories in which this tachyonic
instabilities are absent and look like QED at low energies.
302
3.3. The adapted axioms

After discussing the consequences of Lorentz symmetry breaking in non-commutative
theories, we come to the modification of Wightman axioms needed to accommodate noncommutative theories. Here we will not discuss in details the whole set of axioms [18] but
only comment on the modifications required.
Concerning the space of states of the theory, we take it to be given by a (separable)
Hilbert space carrying a unitary representation of the group5 P = [O(1, 1) SO(2)] T4 .
In addition we will assume the existence of a unique vacuum state | invariant under P.
The spectrum of the momentum operator P is in the forward light wedge

2 2
Spec(P ) = p0 p1 0, p0 0 .
(3.19)
Fields (x) are operator-valued distributions in the non-commutative space R1,3
transforming under P as
U(, a)(x)U(, a)1 = U (x + a),
(3.20)
where U is a matrix acting on the indices of the field (x). Wightman functions, i.e.,
vacuum expectation values of fields,
W (x1 , . . . , xn ) |(x1 ) (xn )|
(3.21)
S[(R1,3 )n ]
define tempered distributions on the Schwartz space

of smooth test functions
which decrease, together with all their derivatives, faster than any power at infinity [42].
As we will see in Section 5, in non-commutative theories, the tempered character of the
distributions can be spoiled by the appearance of hard infrared singularities induced by
quadratic divergences due to UV/IR mixing.
Finally, as argued above, in order to adapt the postulate of microcausality one has to
relax the condition that field (anti)commutators vanish outside the light cone. This is done
by replacing the light cone by the light wedge, namely,
2
2

(x), (y) = 0 if x 0 y 0 x 1 y 1 < 0.
(3.22)
The rest of the axioms, such as hermiticity and completeness are taken over without
modification.
4. The CPT theorem in non-commutative quantum field theory

In what follows we proceed to prove the CPT theorem for non-commutative theories
satisfying the adapted axioms of Section 3.3. In order to keep things simple, we restrict our
analysis in the case of a real scalar field. The generalization to fields transforming in other
representations of O(1, 1) SO(2) is not difficult.
5 When gauge fields are present, we have to generalize this condition to admit an indefinite Hilbert space.
This permits a rigorous implementation of the GuptaBleuler procedure [41]. For simplicity, however, we do not
consider this more general situation.
303
In ordinary theories satisfying the Wightman axioms, the CPT theorem can be proved
directly on the Wightman functions by taking advantage of their analyticity properties
[17]. Because of translational invariance, the n-point Wightman function Wn (x1 , . . . , xn ) is
actually a function of the (n1) coordinate differences 1 = x1 x2, . . . , n1 = xn1 xn
Wn (x1 , . . . , xn ) = Wn (x1 x2 , . . . , xn1 xn ) Wn (1 , . . . , n1 ).
(4.1)
The functions Wn (1 , . . . , n1 ) can be analytically continued into the tube Tn1 =

{j ij , (j )2 > 0, j0 > 0}, and further into the extended tube Tn1
containing all the
4(n1)
points in C
that can be reached from Tn1 by a complex Lorentz transformation.
The CPT theorem is proved by noticing that the analytic continuation of the Wightman
function Wn (1 , . . . , n1 ) and its CPT-transformed Wn (n1 , . . . , 1 ) coincide on a real
, so by the edge of the wedge theorem they have to coincide on
neighborhood of Tn1
their whole domain of analyticity. A key ingredient in the proof is that at spatially separated
real points the following identity holds
Wn (n1 , . . . , 1 ) = Wn (1 , . . . , n1 ),

and j2 < 0.
for real (1 , . . . , n1 ) Tn1
(4.2)
This property, weak local commutativity, follows from the postulate of microcausality.
4.1. Preliminaries
For non-commutative theories, we follow the strategy outlined above. Two important
ingredients in the proof are: the possibility of implementing the PT-transformation on
the coordinates by a transformation of the complexified symmetry group L+ (C), and a
weaker version of local commutativity (3.22). Since the causal structure of the theory
is fully determined by the O(1, 1) symmetry, we perform analytic continuation of the
Wightman function only with respect to the electrical coordinates xe = (x 0 , x 1 ), while
the magnetic ones xm = (x 2 , x 3 ) are left as spectators, keeping in mind that the inversion
xm
xm is implemented by a 180 degrees SO(2) rotation.
We consider the n-point Wightman function for a non-commutative scalar field theory6
Wn (1 , . . . , n ). Using arguments paralleling the standard case (see, for example, [39]) the
Wightman functions can be analytically continued in the electric coordinates to the tube

Tn1 = j ij (j )2e > 0, j0 > 0, (
(4.3)
j )m = 0 .
By definition, the set Tn1 does not contain any real point. The original Wightman
function Wn (1 , . . . , n ) is the boundary value of the analytic function defined on this
tube. However, similarly to the commutative case, the Wightman function can be further
analytically continued to the so-called extended tube Tn1 formed by all the points
reachable from Tn1 by a transformation in L+ (C). It is important to notice that the
transformations of L+ (C) leave the magnetic coordinates invariant, so we never perform
6 Notice that, since the non-commutative theory is invariant under the group of translations, the Wightman
function only depends on the (n 1) coordinate differences as in the commutative case.
304
any analytical continuation on them. This is one of the main difference with respect to the
commutative case where the continuation is made in all four coordinates.
We can use the very definition of the extended tube Tn1 to analytically continue
the Wightman functions on it. If (1 , . . . , n1 ) Tn1 , by definition, there exists
a transformation L+ (C) such that (1 , . . . , n1 ) Tn1 . The value of the
Wightman function in (1 , . . . , n1 ) is given by7
Wn (1 , . . . , n1 ) Wn (1 , . . . , n1 ).
(4.4)
to Tn1 requires
(1 , . . . , n1 )
The condition to have a unique analytical continuation

to verify that
the definition (4.4) is actually unique: if the same point
Tn1 can be
reached from two different points (1 , . . . , n1 ), (1 , . . . , n1 ) Tn1 by means of the
transformations , L+ (C), Eq. (4.4) should yield the same value for the Wigthman
function Wn (1 , . . . , n1 ). In our case this is guaranteed by the condition that for all
(1 , . . . , n1 ) Tn1 and all L+ (C) such that (1 , . . . , n1 ) is also a point
of the tube Tn1 , it holds that
Wn (1 , . . . , n1 ) = Wn (1 , . . . , n1 ).
(4.5)
If is a real transformation, the condition (4.5) obviously holds because of the covariance
of the Wightman function. The way to prove that the condition is satisfied also for arbitrary
complex transformations is to perform analytic continuation of (4.5) along curves in the
complex-rapidity plane of O(1, 1) boosts. This works if there exists a one-parameter family
of complex transformations (t) L+ (C), 0 t 1 interpolating between the identity
and the transformations such that all the points ((t)1 , . . . , (t)n1 ) lie in the tube
Tn1 .
Because of the Abelian character of the symmetry group the proof of this result is much
simpler than in the commutative case. We start with T1 and choose null coordinates .
The condition that this point belongs to the tube implies Im < 0. Under a transformation
of L+ (C) with complex rapidity + i, e(+i) . This point belong to T1 if
Im e(+i) < 0. With this setup it is easy to prove that there is always a one-parameter
family of complex transformations interpolating between the identity and a given complex
transformations and such that the whole family lies in the tube. As shown in Fig. 2 the
effect of a complex transformation is to rotate the null coordinates by an angle in
opposite directions (at the same time than rescaling them by e ). Therefore, if the initial
and the final points lie in T1 , i.e., in the lower half plane, there is always a family of
transformations which interpolate between the two and such that the transformed points
always have negative imaginary parts. The extension to Tn1 with n > 2 is straightforward.
This proves that the Wightman functions can be analytically continued to the extended
tube Tn1 . Unlike the tube Tn1 , the extended tube does contain real points. By analogy
with the standard case, we call these real points Jost points. Actually, the analog of a
theorem proved by Jost [17] holds in this case, namely, that a point (1 , . . . , n1 ) Tn1
7 In the general case where fields in arbitrary representations of O(1, 1) SO(2) are considered, the Wightman
functions transform covariantly with respect to the transformations in L+ (C). In this case, there would be nontrivial factors on the right-hand side of Eq. (4.4) depending on the helicities of the fields involved.
305
Fig. 2. Under a complex O(1, 1) boost with rapidity + i the null coordinates are rotated in opposite
directions. A one-parameter family of transformations can be constructed by considering two monotonous
functions ((t), (t)) with 0 t 1 interpolating between (0, 0) and (, ) such that all the transformed points
lie in the lower half-plane and therefore belong to T1 .
(a)
(b)
Fig. 3. Set of Jost points: (a) for T1 ; (b) for Tn1 with n > 2 the set of Jost points is the product of n 1 copies
of the wedge w+ .
is real if and only if, any combination of the form

n1

j =1
j j ,
with j 0 and
n1

j =1
2j > 0
(4.6)
306
lies outside the light wedge,

n1
2 n1
2

0
1
j j
j j
< 0.
j =1
(4.7)
j =1
Although the condition (4.7) is weaker than the ordinary one, the Jost points still form
a real neighborhood in the extended tube on which the edge of the wedge theorem can
be applied. For T1 the set of Jost points is the shadowed region depicted in Fig. 3(a). They
are formed by the wedge (e )2 < 0. On the other hand, for Tn1 with n > 2 the set of Jost
n1
points is given by w+
, where w+ is the wedge defined by (e )2 < 0 and 1 > 0, as shown
in Fig. 3(b).
4.2. The proof of the theorem
We are ready now to proceed to demonstrate the CPT theorem. The main difference with
the standard proof lies in the fact analytic continuation is performed only in the electric
components of the coordinates and only invariance under O(1, 1) SO(2) is assumed.
However, the existence of this unbroken subgroup of the Lorentz group is enough to ensure
that the PT transformation can be connected to the identity by a family of complex boosts.
In terms of the Wightman functions, the CPT theorem for a neutral scalar field (x)
states that
|(x1 ) (xn )| = |(xn ) (x1 )|.
(4.8)
This identity is equivalent to the following one for the analytically continued Wightman
function into the extended tube Tn1 (cf. [17,18])
Wn (1 , . . . , n1 ) = Wn (n1 , . . . , 1 ),
(1 , . . . , n1 ) Tn1 .
(4.9)
We prove now that Eq. (4.8), and therefore the CPT theorem, is equivalent to the condition
of adapted weak local commutativity
|(x1 ) (xn )| = |(xn ) (x1 )|
(4.10)
with (x1 x2 , . . . , xn1 xn ) a Jost point. Since Jost points are real and lie outside the
light wedge, the condition (4.10) follows from the adapted postulate of microcausality
introduced in Section 3.3.
We begin by rewriting the condition (4.10) as an identity between Wightman functions
at Jost points
Wn (1 , . . . , n1 ) = Wn (n1 , . . . , 1 ).
(4.11)
Since this identity holds at Jost points, and these form a real neighborhood of the domain
of analyticity of the Wightman functions, one concludes, using the edge of the wedge
theorem [18], that Eq. (4.10) is valid in the whole extended tube Tn1 .
As it was discussed in Section 2, the inversion of the four spacetime coordinates is
the product of the transformation Ist L+ (C) (see Eq. (3.5)) and a SO(2)-rotation of .
Therefore, the transformation Ist is a real transformation belonging to O(1, 1) SO(2).
307
Applying the covariance of the Wightman functions under L+ (C) SO(2) we can write8
Wn (1 , . . . , n1 ) = Wn (Ist n1 , . . . , Ist 1 ) = Wn (n1 , . . . , 1 ).
(4.12)
This identity is true for all points in Tn1 and therefore it also holds in the tube Tn1 . Since
the Wightman function |(x1 ) (xn )| are the boundary values of the analytic
function Wn (1 , . . . , n1 ) defined in the tube, this proves the CPT theorem (4.8) for a
neutral scalar field
The reverse is also easily proved. Indeed, if Eq. (4.9) holds in the extended tube using
again the covariance of the Wightman functions under L+ (C) SO(2) we conclude that
Eq. (4.11) holds also at all points in the extended tube. Therefore the relation is satisfied in
particular at the Jost points and we recover the adapted weak local commutativity condition
(4.10).
We have proved that the CPT theorem holds in non-commutative quantum field theories
satisfying the adapted axioms, in particular the postulate of weak local commutativity. As
discussed in Section 3.2 the vacuum expectation value of the commutator of two scalar
fields does not vanish for those non-commutative theories containing states in tachyonic
representations of P. This implies that for this type of theories the adapted postulate of
microcausality does not hold in general. This is, for example, the case of theories with
timespace non-commutativity (e = 0) [30]. In Refs. [20,23] it was argued that the CPT
theorem is also satisfied in these theories, based on the transformation properties of the
different terms in the classical action. What we see is that even if CPT is a symmetry at
tree level, the full quantum theory is not necessarily invariant.
4.3. Remarks on spin-statistics
If we consider representations of SO(1, 1) SO(2) induced from string theory, they are
reductions of representations of SO(1, 3). For them it is easy to extend the above proof of
the CPT theorem
|1 (x1 ) n (xn )| = (1)J i F |n (xn ) 1 (x1 )|,
(4.13)
where F is the number of fermionic fields (which has to be even) and J is the number
of undotted indices that results when the SO(1, 1) SO(2) representations are written in
terms of representations of SO(1, 3).
If instead one wants to consider representations of SO(1, 1) SO(2) that do not follow
from restrictions of those of SO(1, 3), the structure of the phases is more delicate. The
kinetic term for the fields will generically be very different from the standard fourdimensional case, and although the proof of CPT can be adapted to this case, the spinstatistics connection generically will fail.
For fully relativistic field theories, the spin-statistics theorem follow from the phases
appearing in the CPT theorem Eq. (4.13). In the non-commutative case there are conceptual
issues indicating that any spin-statistics theorem will be more difficult to come by. In
8 The invariance of the Wightman function in the extended tube under I follows from Eq. (4.4).
st
308
the standard case, and for massive theories, the little group contains SO(3), a nonAbelian group, which guarantees that the internal spin s of the physical states will be
quantized according to s Z for bosons and s Z + 1/2 for fermions. Thus, proving
the spin-statistics connection requires among other things to show that their interpolating
fields satisfy microcausality with commutators or anticommutators, respectively.9 In the
general non-commutative case these arguments fail, states with arbitrary helicity can be
constructed, and it would not be surprising to find anyonic (or more exotic) behavior. For
theories whose field content descent from string theory via the SeibergWitten limit one
expects that the standard arguments for the spin-statistic theorem can be adapted to the
non-commutative case. The general case, however, remains to be elucidated.
5. Additional issues
In theories with spacespace non-commutativity, violations of CPT invariance can still
appear for two reasons. The first one is the appearance of tachyonic states in the spectrum
which would spoil adapted weak local commutativity. The second thing that can go wrong
with the proof is that the Wightman functions do not define tempered distributions on the
space S[(R1,3 )n ]. This is the case, for example, if hard infrared singularities appear in the
correlation functions induced by ultraviolet quadratic divergences via UV/IR mixing [14].
If this happens, the analytic continuation of the Wightman functions into the (extended)
tube might not be possible.
The two problems appear in the case of pure non-commutative QED with spacespace
non-commutativity (e = 0, m .), where the one-loop corrected dispersion relation
presents a tachyonic instability at low energies [40]
(p )2 = p2
2g 2
.
2p p
(5.1)
At the same time, the correlation functions are afflicted with infrared singularities derived
from uncancelled quadratic ultraviolet divergences. A regularization of these divergences
by considering N = 1, U(1) gauge theory softly broken to N = 0 does not eliminate
the problem of the tachyonic states, since a negative, -independent squared mass for
the photon is generated m2photon = g 2 M 2 /(2 2 ), with M the soft breaking mass of the
gaugino [43].
To overcome these problems, we study U(1) , N = 4 non-commutative gauge theory
softly broken to N = 0 by mass terms for the fermions and the scalars [44,45]. This softbreaking of supersymmetry introduces at most logarithmic divergences in higher order
amplitudes that do not destroy the tempered nature of the correlation functions [46].
Before entering into the details of how this theory renders a non-tachyonic dispersion
relation for the photon at low energies, it is convenient to briefly discuss some aspects of
the regularization of non-commutative gauge theories. In the usual case, the most popular
9 We are oversimplifying here the argument because there are subtle Klein factors that have to be taken into
account in general (see [18]).
309
regularization procedure is dimensional regularization, which straightforwardly extends to

the non-commutative case. However, in studying the Wilsonian low-energy effective action
of non-commutative gauge theories, it would be convenient to introduce a sharp cutoff
in momentum space . In this case the origin of the difficulties with UV/IR mixing are
relatively easy to understand: since the elementary objects are not particles but rigid dipoles
of length | p |, a ultraviolet cutoff in momenta induces an infrared cutoff 1/( ), the
inverse of the maximal dipole length. Since at distances much larger than the length of the
maximal dipole the elementary objects behave again like particles, the commutative theory
has to be recovered at momenta |p|
1/( ) and at the same time Lorentz invariance is
restored.
In the case of gauge theories, however, there is no obvious sharp momentum cutoff
(apart from the lattice) compatible with gauge invariance and the previous picture is not
fully realized. Using dimensional regularization, for example, the whole region 0 |p|

1/( ) disappears. On the other hand, using a cutoff in the Schwinger parameter leads to
violations of gauge invariance. This contrast with the case, for example, of 3 -theory in
six dimensions, where the picture outlined above is realized.
In order to avoid these problems we could use an intermediate cutoff: we use
dimensional regularization to regularize the integrals along the directions orthogonal to the
space-like vector p p (with p the external momentum) while for the computation
of the last integral along the direction of p we use a Schwinger cutoff. This procedure
works for generic external momenta, i.e., p = 0, p = 0. Its advantage lies in the fact that it
exposes some of the features of the sharp cutoff explained above while preserving gauge
invariance.
We proceed now to compute the one-loop effective action (A)eff for gauge field for a
N = 4, U(1) non-commutative gauge theory softly broken to N = 0 by mass terms for the
fermions and scalars in the theory
nf
1
ns
log det f
log det s
(A)eff = log det gauge + log det ghosts +
2
2
2

1
d 4p
A (p) (p)A (p) + O A3 ,

(5.2)
4
2 (2)
where nf and ns denote, respectively, the number of Weyl fermions and real scalars in the
theory. For N = 1 non-commutative U(1) supersymmetric YangMills we have nf = 1,
ns = 0, whereas for N = 2 one has nf = 2, ns = 2 and nf = 4, ns = 6 for N = 4.
Following [45], we use the background field method and work in Euclidean space. The
photon self-energy can be written as the sum of the planar and the non-planar contribution
as [47]
(p) = (p, F = 0) (p, F = p)

with
(5.3)

2
(2k + p) (2k + p)
d 4k
d(j )
j
(p, F) = 2
(2)4
(k 2 + m2j )[(k + p)2 + m2j ] k 2 + m2j
j

p2 p p
+ 4C(j )
(5.4)
eikF .
(k 2 + m2j )[(k + p)2 + m2j ]

310
Table 1
j
Ghost
j
d(j )
C(j )
mj
1
1
0
0
Real scalar
Weyl fermion
Gauge field
21
1
2
1
0
Ms
12
4
2
0
1
2
Mf
The sum is over all states running in the loop (gauge fields, ghosts, fermions and scalars).
The constant (j ), d(j ) and C(j ), as well as the masses for the different fields, are given
in Table 1.
In theories with unbroken (or softly broken) supersymmetry the condition Str 1 = 0
translates into

ns
1=0
(5.5)
j d(j ) = nf
2
j
whereas in theories with N = 4 (or N = 2 with two hypermultiplets) we have the

additional identity

nf
(5.6)
1 = 0.
j C(j )
4
j
For supersymmetric theories (5.5) guarantees the vanishing of the terms proportional
to p p in the vacuum polarization, and the dispersion relation of the photon is not
modified. In the particular case of non-commutative gauge theories with N = 4 unbroken
supersymmetry the identity (5.6) further implies that (p) = 0. If soft breaking masses
are included these cancellations are not complete but, as we will see below, they are enough
to tame the problems arising from UV/IR mixing.
In evaluating (5.4) we come back to the issue of regularization in more detail. There
are various ways to regulate the planar diagram contribution (F = 0), but few for the nonplanar part (F = p)
preserving gauge invariance. Physically, when working with the lowenergy Wilsonian effective action it is convenient to introduce some kind of sharp cutoff
which eliminates the physics at scales E > . In theories without gauge symmetry, and
at one loop, this can be achieved by exponentiation of the propagators using Schwinger
parameters and modifying their integration measure. For instance, for two propagators the
procedure amounts to the prescription
1
(p12
+ m21 )(p22

+ m22 )

d1 d2 exp 1 p12 + m21 1 p22 + m22
=
0

d1 d2 exp 1 p12 + m21 1 p22 + m22

1
.
42 (1 + 2 )
(5.7)
311
Proceeding along these lines, Eqs. (5.3) and (5.4) become after some computation

1 2
p p p
(p) =
j
2
4
j

dx 4C(j ) (1 2x)2d(j )

j
j
K0
K0
eff

1
p p 2eff
j d(j )
+
2
(4)
j
1
dx j K2
j
eff
+ [gauge non-invariant term],
(5.8)
where j = m2j + x(1 x)p2 and the effective cutoff is given by [14]
1
1
= 2 + p 2
2
eff
(5.9)
and we have expressed the integrals over Schwinger parameters in terms of modified Bessel
functions of the second kind
K (z) =
2
dt
t +1
e 2 (t +
2
t )
(5.10)
Ignoring momentarily the gauge non-invariant term, the answer (5.8) is satisfactory in
several ways. For fixed , in the momentum region |p|
< 1/( ) we recover Lorentz
invariance and the standard dispersion relation for the photon. Also there is no ambiguity
in taking the limit p 0. The big drawback, of course, is the lack of gauge invariance, due
to the term proportional to . This term can only be subtracted by local counterterms for
the planar diagram.
Had we used dimensional regularization, the answer would be given by the leading term
of (5.8) in the limit , at fixed p,
j
j
1
1
K0
(5.11)
log
;
eff
2
2
|p|
4
and with the gauge non-invariant term absent (cf. [48]). In this limit the region 0 < |p|
<
1/( ) disappears.
It is possible to imagine an intermediate regulator where the loop integrals are split
into a three-dimensional part, which is evaluated using dimensional regularization in 3 "
dimensions, and a fourth integral which is regularized using a Schwinger cutoff. In order to
implement this regularization we consider for generic p (i.e., p = 0, p = 0) an orthonormal
= 1, . . . , 4)
frame ea
(a
e1
=
p
,
|p|
e2
=
p
|p|
(5.12)
312
and e3
, e4
chosen so that ea
eb
= a b . In this case Eq. (5.4) can be split into an
integral over k1 , which is regulated using a Schwinger cutoff (physically, this corresponds
to regulating the length of the dipoles running in the loop), and the remaining threedimensional integral, which can be dealt with using dimensional regularization.
Implementing this mixed regulator we find a result for the polarization tensor which is
not identical to (5.8), but in which the gauge non-invariant term is absent and the effective
cutoff eff appearing in the third line of this equation is replaced by (2p 2 2eff 1)/p 2 . If
we take the limit with p fixed, we recover the result of dimensional regularization,
again with the identification 2 42 . The main problem with this procedure is that it
is not clear whether it can be systematically extended to higher loops.
5.1. Softly broken N = 4, U(1) gauge theory
Let us now focus on the case of N = 4, non-commutative U(1) gauge theory. Using
dimensional regularization, the polarization tensor can be written as

p p
(p) = 1 (p) p2 p p + 2 (p) 2 ,
(5.13)
p
where

1

v
1
2 1
log
1 (p) =
dx
4
(1
2x)
|
k|
+
K
0
v
4 2
2
42
0

1

f
2
log
1 (1 2x)
+ K0 f |k|
2
42
f

1

s
1
log
(1 2x)2
(5.14)
|
k|
+
K
0
v
2
2
42
s
and
1
2 (p) = 2

dx v K2 v |p|

f K2 f |p|
1
2
s K2

s |p|
.
(5.15)
The subindices v, f and s indicate respectively the contributions of the vectorghost

system, fermions and scalars, and
v = x(1 x)p2 ,
f = m2f + x(1 x)p2 ,
s = m2s + x(1 x)p2 .
(5.16)
The Wilsonian effective coupling constant at momentum p is determined by 1 (p),

namely,
1
1
= 1 (p).
g(p)2 g02
(5.17)
313
For values of the momenta larger that the non-commutative scale, p ! 1/ , the Bessel
functions decay exponentially and only the logarithms in (5.14) contribute, reproducing
the standard -function of the theory

p
1
1 11 2
1
n
n
1 (p) "
(5.18)
,
|p| ! .
f
s log
3
6
8 2 3
42
If we have a field content corresponding to softly broken N = 4 U(1) theory, a N = 1

vector multiplet and three chiral multiplets in the adjoint representation, the -function
vanishes. In fact the theory is believed to be ultraviolet finite [49].
Before considering the modifications that arise from the introduction of the soft
breaking masses, we consider the supersymmetric case where Mf = Ms = 0. In this
case there is an interesting phenomenon-associated
with UV/IR mixing. If we consider
the region of small momenta p 1/ the Bessel function can be approximated by its
leading logarithm behavior and therefore we have

1
1
" log 42 p 2 .
log
(5.19)
K0 |p|
2
2
4
2
Therefore, in this limit we find for 1 (p)

11 2
1
1
2 2
+
n
n
+
1 (p) "
f
s log 4 p ,
2
3
3
6
8
1
|p| .
(5.20)
Comparing Eq. (5.18) with (5.20) we find that the running of the coupling constant in
the infrared is completely similar to the one in the ultraviolet, except for a change in
the sign. The different sign indicates that, for theories with a negative -function, the
theory becomes weakly coupled again at low energies. This kind of duality in momenta,
p 1/p,
is a reflection of the mixing of high and low energy scales characteristic of
non-commutative field theories.
We next proceed to include the soft breaking masses for the fermions and the scalars.
In
particular, we will consider the N = 4 theory in the region of momenta |p| mj < 1/ .
In this case, 1 (p) can be written as

2 2
11
1
log Meff
p + exponentially vanishing terms,

1 (p) =
(5.21)
2
3
8
where the effective mass is given by
1/11
4
6

1/2
Meff =
Mf2
Ms
,
f =1
(5.22)
s=1
again leading to a weakly coupled theory in the infrared. Hence from this point of view
the Wilsonian effective action is well defined. At low-energies we recover asymptotically
an effective theory of a photon whose coupling constant vanishes in the infrared. These
conclusions hold as long as there are no tachyons in the spectrum, which among other
things destroy the possibility of having a weak version of local commutativity needed in
the proof of the CPT theorem, as already discussed in Section 4.
314
As we mentioned, pure N = 1, U(1) with a mass term for the photino leads to a lowenergy tachyon pole for the photon [43]. In our case, since we can also play with the scalar
masses, it is possible to choose values of the soft breaking masses for the fermions and
the scalar in such a way that we avoid this problem. To understand the photon dispersion
relation we need to compute the full propagator following from (5.13).
In order to find the poles of the full propagator we should resum the two-point 1PI
diagrams. The structure of (5.13) suggest the introduction of the non-orthogonal projectors

p p
p p
P =
(5.23)
,
Q =
2
p
p 2
which satisfy (in matrix notation)
Pn = P,
Qn = Q,
PQ = QP = Q
(5.24)
so the polarization tensor in Eq. (5.13) can be written

(p) = p2 1 (p)P + 2 (p)Q.
The full propagator can then be written as

2
2

g 2
ig 2
g0
2
G (p) = 20 1 + 20 (p) +
(p)
+
.
p
p
p2
In order to resum the series, we use Eq. (5.24) to write

n

n
n
(p)n = p2 1 (p) P + p2 1 (p) + 2 (p) p2 1 (p) Q
(5.25)
(5.26)
(5.27)
so the full propagator is given by

ig02 p p
ig02
p p

+
G (p) =
p2
p2
p2 [1 + g02 1 (p)]

ig02
ig02
p p
+
.
2
2
2
2
2
p [1 + g0 1 (p)] + g0 2 (p) p [1 + g0 1 (p)] p 2
Therefore, from Eq. (5.28) we find that the pole conditions are

1
p2 2 + 1 (p) = 0
g0
and

p
1
g02
(5.28)
(5.29)

+ 1 (p) + 2 (p) = 0.
(5.30)
To obtain the dispersion relations for physical particles we have to rotate to Minkowski
space by taking p2 p2 and p 2 p p. Eq. (5.29) implies that one of the polarizations
of the photon corresponds to a massless degree of freedom. On the other hand, Eq. (5.30)
is associated to a photon whose polarization is proportional to p . To determine the pole
in this case we need the low energy behavior of 2 (p). From (5.15), and using the small
argument expansion of K2 (z), we obtain after simple manipulations

2
1
ns
p2
2 (p) = 2 1 + nf
p 2 12

1 2
1
2
M
s + .
f
2 2
2 s
315
(5.31)
In N = 4 non-commutative super-YangMills, we have 1 + nf + ns /2 = 0. Hence to

avoid tachyons we need to satisfy

1 2
(5.32)
Mf2
Ms 0
2 s
f
which implies Str M 2 0. This condition guarantees, to this order in perturbation theory,
that the spectrum is free of tachyons. Although when the equality in (5.32) is satisfied we
recover formally a massless photon at zero momentum, the function 2 (p) is negative in
a neighborhood of p = 0. Therefore, in order to avoid problems the soft-breaking masses
has to be tuned so the photon has a positive mass squared at zero momentum.
In Fig. 4 we have plotted the dispersion relation by solving (5.30) as an implicit equation
for p2 s t and t p p/ 2 , when the strict inequality in (5.32) is satisfied. The positive
intercept in the curve implies that the photons with this polarization become massive,
something that phenomenologically is a disaster. The current bound on the photon mass
is m < 2 1016 eV [50], hence to satisfy it one would have to do a rather non-trivial
fine tuning in the expansion of 2 (p). However, even if this is achieved, the dispersion
relation is likely to produce birefringence, i.e., the speed of light would be different for
polarizations along the commutative and non-commutative directions. This effect would
come from the terms in 2 (p) quadratic in p and p . Again in this case the bounds are
Fig. 4. Dispersion relation p 2 vs. t.
316
very restrictive. Another very strict bound can also be extracted by looking at the departures
from the black body radiation that a massive polarization for the photon produces.
It is clear that the phenomenological prospects for this theory are rather slim. The main
purpose to study it was to show that it is possible to construct a non-commutative field
theory satisfying the adapted axioms introduced in Section 3.3. At least in perturbation,
theory the correlation functions calculated are tempered distributions, the theory is well
defined both in the ultraviolet and the infrared, and the renormalization procedure and
the computation of the Wilsonian effective action will not be afflicted by hard infrared
divergences that would lead to Wightman functions with non-tempered singularities. This
is, for instance, the reason why massless scalar fields in two-dimensions do not exist [51].
Before closing this section, we make some remarks about the resulting dispersion
relations using other regularizations procedures mentioned above involving a sharp cutoff.
Modulo the problems already pointed out, in this case we find that for the photons with
polarizations along p the usual dispersion relation p2 = 0 is recovered for both
low and
high non-commutative momenta with respect to the non-commutative scale 1/ . Around
this non-commutative scale, however, one finds a region where the group velocity of
waves packets becomes superluminal. This situation is highly reminiscent of the situation
described in [52] for non-commutative field theories at finite temperature. This is however
not so surprising if one keeps in mind that in thermal non-commutative field theory the
temperature plays the role of a sharp cutoff. Indeed, at fixed T a non-commutative
U(1) gauge theory has a smooth infrared (and commutative) limit [53] due to the fact
that the Boltzmann factors in the temperature-dependent part of the loop integrals cut off
any physics above the scale T , thus regularizing the UV/IR mixing of non-commutative
gauge theories.
Although we have concentrated only on the one-loop vacuum polarization for the
photon, we expect the same conclusions to apply to higher orders, and also to higher point
functions. The theory considered is believed to be finite in the ultraviolet, and once the
possibility of tachyon poles is allayed, the correlation functions for the theory should be
described in terms of tempered distributions, and the low-energy Wilsonian effective action
should also be well defined. Nevertheless, the constraint between the soft-breaking masses
(5.32) is very likely to be modified by the inclusion of higher loop corrections.
In this paper we have studied some general properties of non-commutative quantum
field theories. An axiomatic formulation can be achieved by modifying the standard
Wightman axioms using as guiding principles (i) the breaking of Lorentz symmetry down
to the subgroup O(1, 1) SO(2), and (ii) the relaxation of local commutativity to make
it compatible with the causal structure of the theory, given by the light-wedge associated
with the O(1, 1) factor of the kinematical symmetry group.
These axioms are enough to demonstrate the CPT theorem. Indeed, this theorem holds
for these theories for reasons not very different from the ones behind the CPT theorem
for commutative theories. The transformation PT is in the connected component of the
identity in the group that result from the complexification of the O(1, 1) subgroup of the
317
symmetry group. This, together with the tempered character of the distributions defined
by the Wightman functions, allows to use results like the edge of the wedge theorem to
show that the Wightman functions and their CPT-transforms coincide.
The main source of difficulties in formulating non-commutative field theories which
satisfy the adapted axioms is UV/IR mixing. The existence of hard infrared singularities
in the non-planar sector of the theory, induced by uncancelled quadratic ultraviolet
divergences, can result in two kinds of problems: they can destroy the tempered nature
of the Wightman functions and/or they can introduce tachyonic states in the spectrum, so
the modified postulate of local commutativity is not preserved.
In this sense, we have also shown how it is possible to construct non-commutative field
theories which, at least in perturbation theory, are compatible with the adapted axioms.
Non-commutative QED is an example of a theory which does not satisfy the axioms
due to the emergence of tachyons due to UV/IR mixing. This mixing of scales can
be tamed by completing non-commutative QED to N = 4 U(1) super-YangMills and
breaking supersymmetry softly by introducing masses for the scalar and fermion fields.
This eliminate the quadratic divergences and, if the soft-breaking masses are tuned, remove
tachyonic instabilities at low momentum. Similar arguments will apply to higher orders in
perturbation theory and higher point functions since the underlying theory is finite N = 4
non-commutative QED. We have also shown that in this context the U(1) theory obtained
is a phenomenological disaster. There are other approaches to phenomenology in noncommutative field theories (see, for instance, [54]), where one first expands the action
up to a certain order in and then quantizes the theory obtained. We know, however, that
generally the two procedures do not necessarily generate the same results due to the UV/IR
mixing.
Acknowledgements
We would like to thank Jos L.F. Barbn, Jos M. Gracia-Bonda, Kerstin E. Kunze,
Dieter Lst, Raymond Stora and Julius Wess for useful discussions. We would like to
thank specially Masud Chaichian for discussions about non-commutative theories and the
CPT theorem which prompted our interest in the subject. Both authors wish to thank
the Humboldt Universitt zu Berlin, and in particular Dieter Lst, for hospitality during
the completion of this work. M.A.V.-M. acknowledges the support of Spanish Science
Ministry Grants AEN99-0315 and FPA2002-02037.
References
[1] A. Connes, Non-Commutative Geometry, Academic Press, 1994;
J. Madore, An Introduction to Non-Commutative Differential Geometry and its Applications, Cambridge
Univ. Press, Cambridge, 1995;
G. Landi, An Introduction to Non-Commutative Spaces and their Geometries, Springer, 1997, hepth/9701078;
J.M. Gracia-Bonda, J.C. Varilly, H. Figueroa, Elements of Non-Commutative Geometry, Birkhuser, 2001.
318
[2] K. von Klitzing, G. Dorda, M. Pepper, New method for high-accuracy determination of the fine-structure
constant based on quantized Hall resistance, Phys. Rev. Lett. 45 (1980) 494;
R.B. Laughlin, Anomalous quantum Hall effect: an incompressible quantum fluid with fractionally charged
excitations, Phys. Rev. Lett. 50 (1983) 1395;
R.B. Laughlin, Quantized motion of three two-dimensional electrons in a strong magnetic field, Phys. Rev.
B 27 (1983) 3383.
[3] Z.F. Ezawa, Quantum Hall Effects. Field Theoretical Approach and Related Topics, World Scientific,
Singapore, 2000.
[4] J. Bellissard, D.R. Grempel, F. Martinelli, E. Scoppola, Localization of electrons with spinorbit or magnetic
interactions in a two-dimensional disordered crystal, Phys. Rev. B 33 (1986) 641;
J. Bellissard, A. van Elst, H. Schulz-Baldes, The non-commutative geometry of the quantum Hall effect,
cond-mat/9411052;
L. Susskind, The quantum Hall fluid and non-commutative ChernSimons theory, hep-th/0101029;
A.P. Polychronakos, Quantum Hall states as matrix ChernSimons theory, J. High Energy Phys. 04 (2001)
011, hep-th/0103013;
J.L.F. Barbn, A. Paredes, Non-commutative field theory and the dynamics of quantum Hall fluids, Int. J.
Mod. Phys. A 17 (2002) 3589, hep-th/0112185.
[5] A. Connes, J. Lott, Particle models and non-commutative geometry, Nucl. Phys. B (Proc. Suppl.) 18 (1991)
29.
[6] C.P. Martn, J.M. Gracia-Bonda, J.C. Varilly, The standard model as a non-commutative geometry: the
low-energy regime, Phys. Rep. 294 (1998) 363, hep-th/9605001.
[7] E. Witten, Non-commutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253.
[8] A. Connes, M.R. Douglas, A. Schwarz, Non-commutative geometry and matrix theory: compactification on
tori, J. High Energy Phys. 02 (1998) 003, hep-th/9711162;
M.R. Douglas, C.M. Hull, D-branes and the non-commutative torus, J. High Energy Phys. 02 (1998) 008,
hep-th/9711165.
[9] N. Seiberg, E. Witten, String theory and non-commutative geometry, J. High Energy Phys. 09 (1999) 032,
hep-th/9908142.
[10] M.M. Sheikh-Jabbari, Open strings in a B-field background as electric dipoles, Phys. Lett. B 455 (1999)
129, hep-th/9901080.
[11] M.R. Douglas, N.A. Nekrasov, Non-commutative field theory, Rev. Mod. Phys. 73 (2001) 977, hepth/0106048;
R.J. Szabo, Quantum field theory on non-commutative spaces, hep-th/0109162.
[12] J.M. Gracia-Bonda, C.P. Martn, Chiral gauge anomalies on non-commutative R4 , Phys. Lett. B 479 (2000)
321, hep-th/0002171;
R. Gopakumar, S. Minwalla, A. Strominger, Non-commutative solitons, J. High Energy Phys. 05 (2000)
020, hep-th/0003160;
C.P. Martn, The UV and IR origin of non-Abelian chiral gauge anomalies on non-commutative Minkowski
spacetime, J. Phys. A 34 (2001) 9037, hep-th/0008126;
D.J. Gross, N.A. Nekrasov, Solitons in non-commutative gauge theory, J. High Energy Phys. 03 (2001) 044,
hep-th/0010090;
J.A. Harvey, G.W. Moore, Non-commutative tachyons and K-theory, J. Math. Phys. 42 (2001) 2765, hepth/0009030;
N. Nekrasov, A. Schwarz, Instantons on non-commutative R4 and (2, 0) superconformal six-dimensional
theory, Commun. Math. Phys. 198 (1998) 689, hep-th/9802068;
C.S. Chu, V.V. Khoze, G. Travaglini, Notes on non-commutative instantons, Nucl. Phys. B 621 (2002) 101,
hep-th/0108007;
J.A. Harvey, Komaba lectures on non-commutative solitons and D-branes, hep-th/0102076;
A. Armoni, E. Lpez, S. Theisen, Non-planar anomalies in non-commutative theories and the Green
Schwarz mechanism, J. High Energy Phys. 06 (2002) 050, hep-th/0203165.
[13] J. Madore, S. Schraml, P. Schupp, J. Wess, Gauge theory on non-commutative spaces, Eur. Phys. J. C 16
(2000) 161, hep-th/0001203;
B. Jurco, L. Mller, S. Schraml, P. Schupp, J. Wess, Construction of non-Abelian gauge theories on noncommutative spaces, Eur. Phys. J. C 21 (2001) 383, hep-th/0104153;
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
319
H. Grosse, Y. Liao, Anomalous C-violating three photon decay of the neutral pion in non-commutative
quantum electrodynamics, Phys. Lett. B 520 (2001) 63, hep-ph/0104260;
X. Calmet, B. Jurco, P. Schupp, J. Wess, M. Wohlgenannt, The standard model on non-commutative space
time, Eur. Phys. J. C 23 (2002) 363, hep-ph/0111115.
S. Minwalla, M. Van Raamsdonk, N. Seiberg, Non-commutative perturbative dynamics, J. High Energy
Phys. 02 (2000) 020, hep-th/9912072;
A. Matusis, L. Susskind, N. Toumbas, The IR/UV connection in the non-commutative gauge theories, J. High
Energy Phys. 12 (2000) 002, hep-th/0002075.
M. Van Raamsdonk, N. Seiberg, Comments on non-commutative perturbative dynamics, J. High Energy
Phys. 03 (2000) 035, hep-th/0002186;
G. Arcioni, J.L. Barbn, J. Gomis, M.A. Vzquez-Mozo, On the stringy nature of winding modes in noncommutative thermal field theories, J. High Energy Phys. 06 (2000) 038, hep-th/0004080;
M. Van Raamsdonk, The meaning of infrared singularities in non-commutative gauge theories, J. High
Energy Phys. 11 (2001) 006, hep-th/0110093;
A. Armoni, E. Lpez, UV/IR mixing via closed strings and tachyonic instabilities, Nucl. Phys. B 632 (2002)
240, hep-th/0110113;
A. Armoni, E. Lpez, A.M. Uranga, Closed strings tachyons and non-commutative instabilities, J. High
G. Lders, On the equivalence of invariance under time reversal and under particleantiparticle conjugation
for relativistic field theories, Dansk. Mat. Fys. Medd. 28 (1954) 5;
G. Lders, Proof of the CPT theorem, Ann. Phys. 2 (1957) 1;
W. Pauli, Exclusion principle, Lorentz group and reflection of spacetime and charge, in: Niels Bohr and the
Development of Physics, Pergamon, 1955;
H. Epstein, CTP invariance of the S-matrix in a theory of local observables, J. Math. Phys. 8 (1967) 750.
R. Jost, Eine Bemerkung zum CTP Theorem, Helv. Phys. Acta 30 (1957) 409.
R.F. Streater, A.S. Wightman, PCT, Spin and Statistics, and All That, Benjamin, 1964.
W. Pauli, The connection between spin and statistics, Phys. Rev. 58 (1940) 716;
G. Lders, B. Zumino, Connection between spin and statistics, Phys. Rev. 110 (1958) 1450.
M.M. Sheikh-Jabbari, Discrete symmetries (C,P,T) in non-commutative field theories, Phys. Rev. Lett. 84
(2000) 5265, hep-th/0001167.
S.M. Carroll, J.A. Harvey, V.A. Kosteleck, C.D. Lane, T. Okamoto, Non-commutative field theory and
Lorentz violation, Phys. Rev. Lett. 87 (2001) 141601, hep-th/0105082.
P. Aschieri, B. Jurco, P. Schupp, J. Wess, Non-commutative GUTs, standard model and C, P, T, Nucl. Phys.
B 651 (2003) 45, hep-th/0205214.
M. Chaichian, K. Nishijima, A. Tureanu, Spin-statistics and CPT theorems in non-commutative field theory,
hep-th/0209008.
A. Schwarz, Gauge theories on non-commutative Euclidean spaces, hep-th/0111174.
D. Bahns, S. Doplicher, K. Fredenhagen, G. Piacitelli, On the unitarity problem in space/time noncommutative theories, Phys. Lett. B 533 (2002) 178, hep-th/0201222;
D. Bahns, S. Doplicher, K. Fredenhagen, G. Piacitelli, Ultraviolet finite quantum field theory on quantum
spacetime, hep-th/0301100.
S. Doplicher, K. Fredenhagen, J.E. Roberts, The quantum structure of spacetime at the Planck scale and
quantum fields, Commun. Math. Phys. 172 (1995) 187, hep-th/0303037.
Y. Liao, K. Sibold, Time-ordered perturbation theory on non-commutative spacetime: Basic rules, Eur.
Phys. J. C 25 (2002) 469, hep-th/0205269;
Y. Liao, K. Sibold, Time-ordered perturbation theory on non-commutative spacetime. II. Unitarity, Eur.
Phys. J. C 25 (2002) 479, hep-th/0206011;
Y. Liao, K. Sibold, Spectral representation and dispersion relations in field theory on non-commutative
space, Phys. Lett. B 549 (2002) 352, hep-th/0209221.
D. Bigatti, L. Susskind, Magnetic fields, branes and non-commutative geometry, Phys. Rev. D 62 (2000)
066004, hep-th/9908056.
J. Gomis, T. Mehen, Spacetime non-commutative field theories and unitarity, Nucl. Phys. B 591 (2000)
265, hep-th/0005129;
320
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]

O. Aharony, J. Gomis, T. Mehen, On theories with light-like non-commutativity, J. High Energy Phys. 09
(2000) 023, hep-th/0006236.
L. lvarez-Gaum, J.L.F. Barbn, R. Zwicky, Remarks on timespace non-commutative field theories,
J. High Energy Phys. 05 (2001) 057, hep-th/0103069.
C. Bachas, M. Porrati, Pair creation of open strings in an electric field, Phys. Lett. B 296 (1992) 77, hepth/9209032.
N. Seiberg, L. Susskind, N. Toumbas, Strings in background electric field, spacetime non-commutativity
and a new non-critical string theory, J. High Energy Phys. 06 (2000) 021, hep-th/0005040;
R. Gopakumar, J.M. Maldacena, S. Minwalla, A. Strominger, S-duality and non-commutative gauge theory,
J. High Energy Phys. 06 (2000) 036, hep-th/0005048;
J.L.F. Barbn, E. Rabinovici, Stringy fuzziness as the custodian of timespace non-commutativity, Phys.
Lett. B 486 (2000) 202, hep-th/0005073.
H. Sonoda, Hermiticity and CPT in string theory, Nucl. Phys. B 326 (1989) 135;
V.A. Kosteleck, R. Potting, CPT and strings, Nucl. Phys. B 359 (1991) 545;
A. Pasquinucci, K. Roland, CPT invariance of string models in a Minkowski background, Nucl. Phys. B 473
(1996) 31, hep-th/9602026.
C.N. Yang, D. Feldman, The S-matrix in the Heisenberg representation, Phys. Rev. 79 (1950) 972.
R. Haag, Quantum field theories with composite particles and asymptotic conditions, Phys. Rev. 112 (1958)
669;
D. Ruelle, On the asymptotic condition in quantum field theory, Helv. Phys. Acta 35 (1962) 147.
N.N. Bogolubov, A.A. Logunov, I.T. Todorov, Introduction to Axiomatic Quantum Field Theory, Benjamin,
1975;
N.N. Bogolubov, A.A. Logunov, A.I. Oksak, I.T. Todorov, General Principles of Quantum Field Theory,
Kluwer, 1990.
A.S. Wightman, Quantum field theories in terms of vacuum expectation values, Phys. Rev. 101 (1956) 860.
R. Jost, The General Theory of Quantized Fields, American Mathematical Society, 1965.
R. Haag, Local Quantum Physics: Fields, Particles, Algebras, Springer, 1992.
F. Ruiz Ruiz, Gauge-fixing independence of IR divergences in non-commutative U(1), perturbative
tachyonic instabilities and supersymmetry, Phys. Lett. B 502 (2001) 274, hep-th/0012171.
F. Strocchi, A.S. Wightman, Proof of the charge superselection rule in local relativistic quantum field theory,
J. Math. Phys. 15 (1974) 2198;
F. Strocchi, A.S. Wightman, J. Math. Phys. 17 (1976) 1930, Erratum.
L. Schwartz, Thorie des Distributions, Hermann, 1966.
C.E. Carlson, C.D. Carone, R.F. Lebed, Supersymmetric non-commutative QED and Lorentz violation,
Phys. Lett. B 549 (2002) 337, hep-ph/0209077.
J.G. Taylor, Soft breaking of N = 4 super-YangMills theory, Phys. Lett. B 121 (1983) 386;
J.J. van der Bij, Y.P. Yao, Soft breaking of N = 4 supersymmetry, Phys. Lett. B 125 (1983) 171.
V.V. Khoze, G. Travaglini, Wilsonian effective actions and the IR/UV mixing in non-commutative gauge
theories, J. High Energy Phys. 01 (2001) 026, hep-th/0011218.
L. Girardello, M.T. Grisaru, Soft breaking of supersymmetry, Nucl. Phys. B 194 (1982) 65;
S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Superspace, or One Thousand and One Lessons in
Supersymmetry, AddisonWesley, 1983, hep-th/0108200.
A. Gonzlez-Arroyo, C.P. Korthals Altes, Reduced model for large N continuum field theories, Phys. Lett.
B 131 (1983) 396;
T. Filk, Divergencies in a field theory on quantum space, Phys. Lett. B 376 (1996) 53.
C.P. Martn, F. Ruiz Ruiz, Paramagnetic dominance, the sign of the beta function and UV/IR mixing in
non-commutative U(1), Nucl. Phys. B 597 (2001) 197, hep-th/0007131.
I. Jack, D.R. Jones, Ultra-violet finite non-commutative theories, Phys. Lett. B 514 (2001) 401, hepth/0105221.
K. Hagiwara, et al., Particle Data Group, Review of particle physics, Phys. Rev. D 66 (2002) 010001,
http://pdg.lbl.gov.
A.S. Wightman, Introduction to some aspects of the relativistic dynamics of quantized fields, in: High Energy
Electromagnetic Interactions and Field Theory, Cargse Lect. in Theoretical Physics, Gordon and Breach,
1967.
321
[52] K. Landsteiner, E. Lpez, M.H. Tytgat, Excitations in hot non-commutative theories, J. High Energy
Phys. 0009 (2000) 027, hep-th/0006210;
K. Landsteiner, E. Lpez, M.H. Tytgat, Instability of non-commutative SYM theories at finite temperature,
J. High Energy Phys. 0106 (2001) 055, hep-th/0104133.
[53] G. Arcioni, M.A. Vzquez-Mozo, Thermal effects in perturbative non-commutative gauge theories, J. High
[54] P. Schupp, J. Trampetic, J. Wess, G. Raffelt, The photon neutrino interaction in non-commutative field theory
and astrophysical bounds, hep-ph/0212292;
W. Behr, N.G. Deshpande, G. Duplancic, P. Schupp, J. Trampetic, J. Wess, The Z gg decays in the
non-commutative standard model, hep-ph/0202121.

Casimir energies due to matter fields in T 2 and

T 2 /Z2 compactifications
Masato Ito
Department of Physics, Nagoya University, Nagoya 464-8602, Japan
Received 5 February 2003; accepted 17 June 2003
Abstract
We calculate the Casimir energies due to matter fields with various boundary conditions along
two compact directions in T 2 compactification. We discuss whether the Casimir energies generate
attractive or repulsive forces. On the theories with extra dimensions, the Casimir energy plays
a crucial role in the mechanism for stabilizing the size of extra dimensions. Finally we argue a
procedure of the application to Z2 orbifold.
1. Introduction
Motivated by Kaluza [1] and Klein [2], it is expected that the phenomenology of low
energy physics should be explained by extra spatial dimensions. Based on the Kaluza
Klein theory it is natural to consider that the radii of the compactified extra dimensions
should be Planck size. After the KaluzaKlein theory, there had been many works in the
framework of theories with extra dimensions [36]. Relying on the assumption that the
gravity only feels large extra compact dimensions at sub-millimeter range, a remarkable
model of Ref. [7] provided a breakthrough of hierarchy problem.
Recent developments are based on the idea that ordinary matter fields could be confined
to a three-brane world embedded in the higher-dimensional world. The assumption of
such models is that the three-brane can be identified with the fixed plane via orbifold
of extra-dimensional manifold. At the beginning of warped braneworld scenario, Randall
and Sundrum had proposed a new suggestion to the hierarchy problem by using separated
three-branes on S 1 /Z2 orbifold embedded in AdS5 [8]. Furthermore, it was shown that
E-mail address: mito@eken.phys.nagoya-u.ac.jp (M. Ito).
doi:10.1016/S0550-3213(03)00537-6
M. Ito / Nuclear Physics B 668 (2003) 322334
323
the localization of the gravity occurs on the positive tension three-brane and usual fourdimensional Newton law can be recovered at distance which is much larger than radius
of AdS5 [9]. Thus the orbifold of compact spaces may be very fascinating idea in the
framework of braneworld scenarios.
Interestingly, an operation of the orbifold is very useful to construct phenomenological
models because unwanted matter fields can be projected out, and the operation enables
for the models to be chiral. In order that the model can be phenomenologically viable,
it is possibility that wanted matter fields can be assigned to appropriate charges under
orbifold symmetry. Furthermore it is widely discussed that the orbifold brings on the
supersymmetry breaking or the breaking of gauge group [10,11]. The idea of the orbifold
had been developed by string theory in non-trivial background. Recent phenomenological
successful models via the orbifold are not necessary based on string theory. However, it
is expected that these remarkable models can provide several solutions to tripletdoublet
splitting, proton decay, neutrino models and so on.
On the theories with extra dimensions, an important issue arises. The stabilization
mechanism for the size of the extra spatial dimensions has not been discovered yet. As
a clue to the issue, the Casimir energy plays a crucial role in radius stabilization of models
with compactified extra dimensions [12,13]. The fate of compactified extra dimensions
depends on the Casimir force generated by the Casimir energy. Attractive force leads to
shrink of the compactified manifold, on the other hand, repulsive force leads to inflation
of the compactified manifold. Accordingly it is considered that radius stabilization can be
realized by balance between the attractive and the repulsive force.
In this paper, we calculate the Casimir energy V for a massless scalar field in M 4 T 2 ,
where M 4 is ordinary four-dimensional Minkowski space and T 2 is two-torus. It is
assumed that the fifth and the sixth dimensions are compactified. Note that the Casimir
energy depends on the topology and boundary conditions of compactified manifold, and we
consider that the boundary condition is periodic or anti-periodic along each compactified
direction. We focus on the Casimir energies with respect to the following four cases,
V (P ,P ) , V (P ,A) , V (A,P ) and V (A,A) , where P denotes periodic boundary condition and
A does anti-periodic boundary condition. In the round bracket of V , the former letter
and the latter letter represents the boundary condition for the fifth direction and for the
sixth direction, respectively. The evaluation of V (P ,P ) has been performed in Ref. [13],
accordingly, we will evaluate the remaining three cases. By examining the sign of the
Casimir energy, we will investigate whether the Casimir force is attractive or repulsive.
Finally we discuss an application to Z2 orbifold. Since we are interested in the Casimir
energies due to the matter fields, in the present paper we neglect the contributions of the
various localized terms on the orbifold fixed planes in addition to a bulk cosmological
constant. We are going to provide the formulas of the Casimir energies due to the matters
fields when considering the radius stabilization of a certain model.
The Casimir energy due to a scalar field with the KaluzaKlein modes can be given by
324
1
2 m,n
2

d 4k
2
log
k
+
M
m,n
(2)4

s
1
d 4k 2
=
k + M2m,n

4
2 m,n s s=0 (2)

s
1
1
M2m,n ,
=

2
32 s s=2 s(s + 1) m,n
V=
(1)
where m, n Z. Note that M2m,n corresponds to the KaluzaKlein spectrum which

is determined by boundary conditions. The evaluation of the double sum in (1) is
explicitly performed by using the techniques of -function regularization [1418]. When
we calculate the Casimir energy of matter field with various boundary conditions on T 2 ,
we use the above formula. Applying our results to a certain phenomenological models, we
multiply (1) by the number of degrees of freedom of scalar field in the model, while for
fermion we need to multiply by 1.
The paper is organized as follows. In Section 2 we describe the detailed calculations
of the Casimir energies due to a massless scalar field with boundary conditions along two
directions of T 2 . The Casimir energies can be written in terms of the area of the torus and
the ratio of two radii, and we study the Casimir forces. Finally we shall briefly discuss
the case of Z2 orbifold. In Section 3 we mention a summary with respect to the results
obtained in the paper. In Appendix A we show the detailed evaluations of the double sums
including in calculations of the Casimir energies. In Appendix B we derive the relation
between (2n) and (2n + 1) by calculating the Casimir energy on S 1 compactification.
Because we encounter the first derivative of zeta function when evaluating the Casimir
energy. In final part, a procedure of the evaluation of (0) is described.
2. Calculation of the Casimir energies

We consider T 2 compactification in the six-dimensional theory, where the two compact
directions are mutually perpendicular. It is assumed that the compact radius of the fifth
dimension is R1 and the sixth dimension is R2 . We begin to calculate the Casimir energy
of a massless scalar field with periodic boundary conditions along two directions in T 2 .
The double sum of the corresponding KaluzaKlein spectrum in (1) can be expressed as

1
1 2 s
2
n + 2m
2
R2
m,n R1

(s 12 ) R1 12s
(2s 1)
= R12s 2 (2s) + 2
(s)
R2

1
1
8 s R1 2 s n s 2
R1
+
Ks 1 2 mn ,
(2)
2
(s) R2
m
R2
m=1 n=1
where the prime denotes the exclusion of a zero mode. Here Ks (x) is the modified Bessel
function and (x) is the Riemann zeta function. In Appendix A the evaluation of the double
sum is explicitly performed. Plugging into (1) leads to the Casimir energy

1
8 R1 5
3
(5)
+
V (P ,P ) =
945 R2
64 2 R14 4
5

52

m
R1
16 R1 2
+ 2
K 5 2 mn ,
2
n
R2
R2
325
(3)
m=1 n=1
where (P , P ) denotes periodic boundary conditions along two compact directions in T 2 ,

the former letter corresponds to the fifth direction and the latter letter does the sixth
direction. Performing the calculation of the derivative with respect to s in (1), we used
the following particular values:

8
3
(2)
5
=
(5),
=
2,

,
(4) =
4
2
2
15
4
(2)
15
1
.
(5) = 6 (6) =
(4)
252
4
In Appendix B we represented the evaluation of (2n), where n is non-negative integer.
By using the following form of the modified Bessel function:
3
5
1
z 2 + 3z 2 + 3z 2 ez ,
K 5 (z) =
(5)
2
2
we can obtain
V
(P ,P )

3
1
8 R1 5 8L3 ( ) R1 2
=
(5) +
+
945 R2
R2
2
64 2 R14 4

12L4 ( ) R1 6L5 ( )
+
+
,
3 R2
4
(6)
where we defined
L3 ( )
L4 ( )
L5 ( )
q e

m=1

m=1
m=1
2
m2 Li3 (q m ) =
1 coth n
,
4
n3 sinh2 n
(7)
n=1
1
1
m Li4 (q m ) =
,
4
4
n sinh2 n
(8)
n=1
Li5 (q m ) =

n=1
1
n5 (e2 n
1)
(9)
(10)
is the polylogarithm function. In (7), (8) and

where = R1 /R2 and Lik (x) = n=1
(9), the infinite sums of geometric sequence over m are performed.
The toroidal compactification has modular symmetry, namely, it is always possible
to redefine the values of R1 and R2 . Therefore the area of the torus A and the ratio of
two radii have physical meaning. The ordinary modular symmetry can be specified by
x n /nk
326
the three parameters, two radii R1 , R2 and relative angle between two directions of
compactification. In terminology of string theory, the area A corresponds to the Kahler
moduli and the modular parameter (R1 /R2 )ei corresponds to the complex moduli. The
modular invariance of the Casimir energy can be demonstrated by the Poisson resummation
technique described in [19]. In Ref. [13], it was shown that the Casimir energy V (P ,P )
has extreme points at the two self-dual points ( = /2, 2/3 and R1 = R2 ) of modular
symmetry. In the present paper it is assumed that = /2. From (6), the Casimir energy
can be written in terms of A = 4 2 R1 R2 and = R1 /R2 :

1 8 3 3
12L4 ( ) 1 3(5) + 6L5 ( ) 1
+ 8L3 ( ) +
+
V (P ,P ) = 2
(11)
.
4A 945
2
2
Note that since the Casimir energy V (P ,P ) is negative for arbitrary , the Casimir force
due to a scalar field with periodic boundary conditions along two directions is attractive.
For instance, adopting = 1 (R1 = R2 ) to be maximal symmetry, the value of V (P ,P ) is
approximately given by
V (P ,P ) 0.1502385/A2 ,
(12)
where the area is fixed. Thus it turns out to be attractive force.

Next we calculate the Casimir energy of a massless scalar field with periodic boundary
condition in the fifth direction and anti-periodic boundary condition in the sixth direction.
The double sum of the corresponding KaluzaKlein spectrum in (1) can be expressed as

1
1
1 2 s
2
n + 2 m+
2
2
R2
m,n R1

(s 12 ) R1 12s
1
2s
2s 1,
= R1 2
(s)
R2
2
1
1

1 s+ 2 s 1
8 s R1 2 s
m+
n 2
+
(s) R2
2
m=0 n=1

R1
1
Ks 1 2
(13)
m+
n ,
2
R2
2
where (s, ) is the Hurwitzs zeta function. In Appendix A we evaluated the above double
sum which corresponds to the case of 0 in (A.5). From (1), we can obtain

1
31 R1 5
V (P ,A) =
3780 R2
64 2R14
5
5
16 R1 2
1 2 5
n 2
+ 2
m+
R2
2
m=0 n=1

R1
1
K 5 2
m+
n , (14)
2
R2
2
327
where we used (s, 1/2) = (2s 1)(s) and (P , A) denotes periodic boundary condition
in the fifth direction and anti-periodic boundary condition in the sixth direction. Using (5),
the Casimir energy can be expressed as
V
(P ,A)

1
31 R1 5 8L 3 ( ) R1 2
=
+
3780 R2
2
R2
64 2R14

12L 4 ( ) R1 6L 5 ( )
+
+
,
3 R2
4
(15)
where we defined
L 3 ( )
L 4 ( )
L 5 ( )

1
1 cosh2 n + 1
1 2
Li3 q m+ 2 =
,
m+
2
8
n3 sinh3 n
m=0

m=0

m=0
(16)
n=1

1
1
1 coth n
Li4 q m+ 2 =
m+
,
2
4
n4 sinh n
(17)
n=1
Li5 q

m+ 1
2
1
n=1
1
.
n5 sinh n
(18)
Using the area of torus A and the ratio of two radii , (15) is given by
V (P ,A) =
1
31 3 3
3 ( ) + 12L4 ( ) 1 + 6L5 ( ) 1 .
+
8
L
4A2
3780
2 2
(19)
In the bracket of the above equation, the first term contributes to the repulsive force and the
remaining terms have the contributions of the attractive forces. The value of the Casimir
energy for = 1, when the area is fixed, is approximately given by
V (P ,A) +0.0139727/A2 ,
(20)
consequently, repulsive force arises. In the case of 1, since the last term in (19)
is dominant, the attractive force is generated. By choosing the appropriate value of ,
vanishing V (P ,A) can be realized by balance between attractive and repulsive.
Successively we calculate the Casimir energy of a massless scalar field with antiperiodic boundary condition in the fifth direction and periodic boundary condition in the
sixth direction. By making an exchange of 1/ (R1 R2 ) in (19), we can obtain
V (A,P ) =

1
1
1
31 3 1
3 ( 1 ) + 12L4 ( ) + 6L5 ( ) 2 .
+
8
L
4A2
3780 3
(21)
As a matter of course, V (P ,A) is equal to V (A,P ) for = 1. In the case of 1, since the
first term is dominant, it is repulsive force.
Finally, we will calculate the Casimir energy of a massless scalar field with anti-periodic
boundary conditions in two directions. The double sum of the corresponding KaluzaKlein
modes in (1) can be given by
328

1
1 2
1 2 s
1
n+
+ 2 m+
2
2
2
R2
m,n R1

(s 12 ) R1 12s
1
2s 1,
= R12s 2
(s)
R2
2
1
1

1 s+ 2 s 1
8 s R1 2 s
m+
n 2 (1)n
+
(s) R2
2
m=0 n=1

R1
1
n .
m+
Ks 1 2
2
R2
2
(22)
The evaluation of the above double sum corresponds to the case of = = 1/2 in
Appendix A. Plugging into (1), we have

31 R1 5 8L 3 ( ) R1 2 12L 4 ( ) R1
1
V (A,A) =
+
+
3780 R2
2
R2
3 R2
64 2 R14

6L 5 ( )
,
+
(23)
4
where we defined

(1)n
m+
n3
n=1 m=0

(1)n
m+
L 4 ( )
n4
L 3 ( )
L 5 ( )
n=1 m=0

n=1 m=0
1
2
2
1
e2 n(m+1/2) = L 3 ( ) + L 3 (2 ),
4

1 2 n(m+1/2)
1
e
= L 4 ( ) + L 4 (2 ),
2
8
(1)n 2 n(m+1/2)
1
e
= L 5 ( ) + L 5 (2 ).
5
n
16
(24)
(25)
(26)
Here we rewrote the above sums by using (16), (17) and (18). The Casimir energy with
anti-periodic boundary conditions in two directions can be rewritten in terms of A and
as follows

31 3 3
1
12L 4 ( ) 1 6L 5 ( ) 1
+ 8L 3 ( ) +
+
.
V (A,A) = 2
(27)
3780
4A
2 2
Since L 3 , L 4 and L 5 are decreasing for , L 3 , L 4 and L 5 are always negative. Namely
V (A,A) is positive for arbitrary . Therefore, it generates repulsive force. For instance, in
the case of = 1 when A is fixed, the value of V (A,A) is approximately
V (A,A) +0.1126133/A2 ,
(28)
thus it is repulsive force.

Consequently, V (P ,P ) and V (A,A) can generate the attractive force and repulsive force,
respectively. Whether V (P ,A) and V (A,P ) generate attractive or repulsive forces depends
on the value of . We calculated the Casimir energies of a single massless scalar field in
329
the present paper, while for fermion field we need to multiply by 1. Taking into account
of the number of degrees of freedom in matter fields with various boundary conditions, the
contributions of the matter fields are given by the sum of (11), (19), (21) and (27). Thus the
total Casimir energy due to the matter fields depends on model-building.
Moreover
we need

to add the contribution of a bulk cosmological constant: d 6 x G = d 4 x A, where
is a bulk cosmological constant which corresponds to the vacuum energy. Accordingly
it is possible for the total Casimir energy to have minimum for A and . However, the
stability of the system depends on the sign of which leads to the familiar cosmological
constant problem. In the present paper we do not mention the problem. Setting to a certain
supersymmetric model, the preservation of supersymmetry via toroidal compactification
guarantees vanishing Casimir energy [13].
3. Application to Z 2 orbifold
We shall discuss a procedure of the application to T 2 /Z2 orbifold. The matter fields
with boundary conditions of (P , P ), (P , A), (A, P ) and (A, A) can be decomposed by Z2
orbifold. The wave function of each matter field can be separated into the cosine function
and the sine function by performing the operation of the orbifold. Equivalently, this implies
Z2 projections of the following three transformations
r1 :
(x5 , x6 ) (x5 + 2R1 , x6 ),
(29)
r2 :
(x5 , x6 ) (x5 , x6 + 2R2 ),
(30)
r3 :
(x5 , x6 ) (x5 , x6 ),
(31)
where x5 and x6 denotes the fifth coordinate and the sixth coordinate, respectively. The
Z2 parity of matter fields via these transformations can be specified by (r1 , r2 , r3 ), where
r1 , r2 , r3 = 1 under the orbifold symmetry. Namely, this means that +1 corresponds to
the even state and 1 odd state. The illustration of the decompositions via Z2 projection
can be represented as follows
x

cos R51 m + Rx62 n ,
x6
x5

(P , P ): exp i m + i n
(32)
sin Rx51 m + Rx62 n ,
R1
R2
x

cos R51 m + Rx62 n + 12 ,
x5
1
x6

x5
(P , A): exp i m + i
(33)
n+
R1
R2
2
sin R1 m + Rx62 n + 12 ,
x

cos R51 m + 12 + Rx62 n ,
x5
1
x6

(A, P ): exp i
(34)
m+
+i n
R1
2
R2
sin Rx51 m + 12 + Rx62 n ,

x6
1
1
x5
+i
m+
n+
(A, A): exp i
R1
2
R2
2
x
x6

1
5
cos R1 m + 2 + R2 n + 12 ,

(35)
sin Rx51 m + 12 + Rx62 n + 12 ,
330
where the symbol denotes the operation of Z2 orbifold. Here we omitted the
normalization factors and the four-dimensional parts in matter fields. Under the orbifold symmetry, note that (+1, +1, +1) and (+1, +1, 1) states result from (P , P ),
(+1, 1, +1) and (+1, 1, 1) from (P , A), (1, +1, +1) and (1, +1, 1) from
(A, P ), (1, 1, +1) and (1, 1, 1) from (A, A). Thus the Z2 orbifold can produce the eight states. Furthermore there are four orbifold fixed planes on T 2 /Z2 at
(x5 , x6 ) = (0, 0), (R1 , 0), (0, R2 ), (R1 , R2 ). There exist extra contributions of the
localized terms on these fixed planes, for example, kinetic terms, mass terms and interaction terms between bulk and brane as well as brane tension. Since the KaluzaKlein
spectrum is modified by these brane-localized terms, these effects must be considered. For
example, in five-dimensional S 1 /Z2 model, the Casimir energy including brane-localized
kinetic terms had been calculated [13]. Consequently the total Casimir energy consists of
the matter fields, a bulk cosmological constant as well as brane-localized terms. It is considered that the Casimir energy including all contributions is very complicated form, and
minimum problem of the Casimir energy is closely related to the cosmological constant
problem. We are going to investigate the points elsewhere.
Adopting a certain model on T 2 /Z2 orbifold, the Casimir energies due to the matter
fields will be calculated by taking account of the number of the degrees of freedom and
(r1 , r2 , r3 ) parity assignments given in matter content of the model when neglecting the
effects of branes. When performing the calculation, we need to make the replacements
R1 R1 /2 and R2 R2 /2, simultaneously, multiply 1/2 1/2 factor via half modes
over m, n. Therefore we can obtain the following forms

1
R1 R2
V (++) = V (P ,P )
,
,
4
2 2

R1 R2
1
,
,
V (+) = V (P ,A)
4
2 2

1 (A,P ) R1 R2
(+)
V
,
,
= V
4
2 2

R1 R2
1
V () = V (A,A)
(36)
,
.
4
2 2
When considering the radius stabilization of the concrete model on T 2 /Z2 orbifold, it is
important to use results obtained here. The concrete model is beyond the scope of this
paper and we do not mention it here.
4. Summary and discussion

We calculated the Casimir energies due to a massless scalar field with various boundary
conditions on T 2 compactification, assuming that two compact directions are perpendicular
each other. The Casimir energies can be explicitly represented in terms of the area of torus
and the ratio of two radii. Consequently, it was shown that the case of (P , P ) boundary
condition is attractive force and the case of (A, A) boundary condition is repulsive force.
331
For (P , A) and (A, P ), whether attractive or repulsive depends on the ratio of two radii.
For fermion field, opposite force works.
When calculating the Casimir energies of the matter fields on a certain T 2 /Z2 orbifold
model, we need to multiply our results to the number of degrees of freedom assigning
eight kinds of Z2 parity (1, 1, 1) of matter content. Thus the contributions of the
matter fields in radius stabilization of the model will be explicitly evaluated. Furthermore
there are contributions of a bulk cosmological constant and brane-localized terms (kinetic
term, mass term, interaction terms and brane tension) on the orbifold fixed planes. Since
KaluzaKlein spectrum is modified by these brane-localized terms, it can be considered
that the total Casimir energy is very complicated form. That the total Casimir energy
has a minimum point for the area of the torus and the ratio of two radii is related to the
cosmological constant problem. We are going to describe it elsewhere. In the present paper
we could provide the formulas of the Casimir energies with various boundary conditions
when considering the radius stabilization in the model.
Appendix A. Evaluation of double sums

Calculating the Casimir energy for field with various boundary conditions on toroidal
compactification, we encounter the double summation of infinite series [14].
When we compute the Casimir energy of a massless scalar field with periodic boundary
conditions along two compact directions in T 2 , we must evaluate the following double sum

s
I (a; s) =
(A.1)
n2 + a 2 m2 ,
m,n
where the prime denotes (m, n) = (0, 0). The double sum can be decomposed as follows

s
n2 + a 2 m2 .
n2s +
(A.2)
n
The first term can be written in terms of the Riemann zeta function. The second term can be
rewritten by using the gamma function what is called Mellin transformation. Consequently,
we obtain [13]
1
2
2 2
I (a; s) = 2 (2s) +
dt t s1 e(n +a m )t
(s)
m
n

2
s 32 a 2 m2 t
t n2
1+2
= 2 (2s) +
dt t
e
e
(s)
m
0
n=1
(s 12 ) 12s
|a|
(2s 1)
= 2 (2s) + 2
(s)

s 12

1
8 s
n
s
2
+
|a|
Ks 1 2|a|mn ,
2
(s)
m
m=1 n=1
(A.3)
332
where Ks (x) is the modified Bessel function, we used the Poisson re-summation formula
in the second line.
Next we shall evaluate the double sum with non-periodic boundary conditions as follows

s
I (a; , ; s) =
(A.4)
(n + )2 + a 2 (m + )2 ,
m,n
where 0 < , < 1. Following the similar procedures in (A.3), we obtain

I (a; , ; s)
1
2
2
2
=
dt t s1 e((n+) +a (m+) )t
(s)
m,n

2

s 32 a 2 (m+)2 t
t n2
1+2
=
t
e
cos(2n)e
(s) m=

1
n=1

s2
|a|12s (2s 1, ) + (2s 1, 1 )
(s)

s

1
1
4
|a| 2 s
ns 2 cos(2n)
+
(s)
m=0 n=1

1
(m + )s+ 2 Ks 1 2|a|(m + )n
2

1
+ (m + 1 )s+ 2 Ks 1 2|a|(m + 1 )n ,
2
(A.5)
where
(s, ) =

n=0
1
(n + )s
(A.6)
is the Hurwitzs zeta function.

Appendix B. Evaluation of (2n)
We derive the relation between (2n) and (2n+1) by calculating the Casimir energy
for R2n S 1 , where n is non-negative integer. We consider a massless scalar field with
periodic boundary condition on the S 1 compactification with radius L. The Casimir energy
E is given by
E=

m2
d 2n k
1
2
log
k
+
2 m= (2)2n
L2
333

1
m2 s
d 2n k
2
+
k
2 s s=0 m= (2)2n
L2

2

1
d 2n k 1
(k 2 + m2 )t s1
L
=
dt
e
t
2 s s=0 m= (2)2n (s)
0

n
(s n) 2s
=
L (2s 2n)
(2)2n L2n s s=0 (s)
=
2(1)n n
(2n).
(2)2n (n + 1)L2n
(B.1)
Next we can rewrite the first line of (B.1) as follows

1
1 (k 2 + m22 )s
d 2n k
L
e
E=
ds
2 m= (2)2n
s
n
=
2(2)2n
=
n
2(2)2n

ds
0
s n+1

1
s n+1
m2 s
L
m=

ds
L2 2 L2 m2
e s
s m=

1

n
+
(2n + 1),
1
2
22n 3n+ 2 L2n
1
(B.2)
where we used the Poisson re-summation formula in the third line. Since (B.1) is equal to
(B.2), we obtain

(1)n
1

(2n) =
(n + 1) n +
(2n + 1)
1
2
2 2n+ 2
(1)n
(2n + 1)(2n + 1).
=
(B.3)
2(2)2n
Below we tabulate the particular values of (2n), for instance, for n = 1, 2, 3, 4.
1
(3) 0.03048,
4 2
3
(4) =
(5) 0.0080,
4 4
45
(6) = 6 (7) 0.00592,
8
315
(8) =
(9) 0.00835.
4 8
(2) =
(B.4)
334
Next, for your information, we can obtain a specific value of n = 0 by using the
following identity [20]
z
(z) (z).
(1 z) = 21z z cos
(B.5)
2
Performing the logarithmic derivative with respect to z, we get
(1 z)
z
(z)
= log(2) + tan
(z)
,
(1 z)
2
2
(z)
where (z) is the polygamma function. Taking the limit of z 1, we have

(0)
z
(z)
= log(2) (1) + lim
+ tan
= log(2).
z1
(0)
(z)
2
2
Here we used (1) = and (z) = 1/(z 1) + + O(|z 1|) for z 1, where
Euler constant. Therefore, we have
1
(0) = (0) log(2) = log(2).
2
(B.6)
(B.7)
is the
(B.8)
References
[1] Th. Kaluza, Sitzungsber. Preuss. Akad. Wiss. Phys. Math. K 1 (1921) 966.
[2] O. Klein, Nature 118 (1926) 516;
O. Klein, Z. Phys. 37 (1926) 895.
[3] T. Appelquist, A. Chodos, P.G. Freund, Modern KaluzaKlein Theories, AddisonWessley, USA, 1987.
[4] T. Appelquist, A. Chodos, The Quantum Dynamics of KaluzaKlein Theories, Phys. Rev. D 28 (1983) 772.
[5] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, New dimensions at a millimeter to a Fermi
and superstrings at a TeV, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[6] I. Antoniadis, A possible new dimension at a few TeV, Phys. Lett. B 246 (1990) 377.
[7] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, The hierarchy problem and new dimensions at a millimeter,
Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[8] L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev. Lett. 83 (1999)
3370, hep-ph/9905221.
[9] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690, hepth/9906064.
[10] Y. Kawamura, Gauge symmetry reduction from the extra space S(1)/Z(2), Prog. Theor. Phys. 103 (2000)
613, hep-ph/9902423.
[11] R. Barbieri, L.J. Hall, Y. Nomura, A constrained standard model from a compact extra dimension, Phys.
Rev. D 63 (2001) 105007, hep-ph/0011311.
[12] R. Hofmann, P. Kanti, M. Pospelov, (De-)stabilization of an extra dimension due to a Casimir force, Phys.
Rev. D 63 (2001) 124020, hep-ph/0012213.
[13] E. Ponton, E. Poppitz, Casimir energy and radius stabilization in five- and six-dimensional orbifolds,
JHEP 0106 (2001) 019, hep-ph/0105021.
[14] E. Elizalde, S.D. Odintsov, A. Romeo, A.A. Bytsenko, S. Zerbini, Zeta Regularization Technique with
Applications, World Scientific, Singapore, 1994.
[15] S.W. Hawking, Commun. Math. Phys. 55 (1977) 133.
[16] E. Elizalde, Multiple zeta functions with arbitrary exponents, J. Phys. A 22 (1989) 931.
[17] C. Itzykson, J.M. Luck, Arithmetical degeneracies in simple quantum systems, J. Phys. A 19 (1986) 211.
[18] T. Kubota, Elementary Theory of Eisenstein Series, Kodanha and Halsted Press, 1973.
[19] R. Rohm, Spontaneous supersymmetry breaking in supersymmetric string theories, Nucl. Phys. B 237 (1984)
553.
[20] E.T. Whittaker, G.N. Watson, Course of Modern Analysis, Cambridge Univ. Press, Cambridge, 1996.

On the low-energy limit of the QED

N-photon amplitudes
Louise C. Martin a , Christian Schubert a,b ,
Victor M. Villanueva Sandoval a
a Instituto de Fsica y Matemticas, Universidad Michoacana de San Nicols de Hidalgo,
Edificio C-3, Apdo. Postal 2-82, C.P. 58040 Morelia, Michoacn, Mexico
b California Institute for Physics and Astrophysics, 366 Cambridge Ave., Palo Alto, CA 94306, USA
Received 30 December 2002; received in revised form 3 June 2003; accepted 1 July 2003
Abstract
We derive an explicit formula for the low energy limits of the one-loop on-shell massive N-photon
amplitudes, for arbitrary N and all helicity assignments, in scalar and spinor QED. The two-loop
corrections to the same amplitudes are obtained for up to the ten point case. All photon amplitudes
with an odd number of + helicities are shown to vanish in this limit to all loop orders.
1. Introduction: on-shell QED photon amplitudes

In recent years substantial progress has been made in the computation of on-shell oneloop amplitudes. This has been due to the development of new techniques [1,2] which
provide alternatives to the standard Feynman diagrammatic approach, as well as to progress
in the calculation of the basic integrals [310]. Much of this work has been concerned
with massless amplitudes, which are computationally the most accessible ones. It led
to a number of unexpectedly simple results for certain special helicity configurations of
photon or gluon amplitudes (see [2] for a review). A particularly striking result is Mahlons
observation that the massless one-loop QED N -photon amplitudes with all helicities equal
vanish on-shell for all N 6 [11].
For the corresponding amplitudes involving massive loops little is known beyond the
four-point case [1215]. In the present paper, we will investigate the QED N photon
E-mail address: schubert@panam.edu (C. Schubert).
doi:10.1016/S0550-3213(03)00578-9
336
L.C. Martin et al. / Nuclear Physics B 668 (2003) 335344
amplitudes in the limit of low photon energies, i.e., with photon momenta such that all
kinematic invariants ki kj are small compared to m2 (see [16] for a discussion of this
approximation for the photon scattering case). As is well known (see, e.g., [17]), in this
limit the photon amplitudes are directly related to the QED effective Lagrangian L(F )
for a background field with a constant field strength tensor F . Namely, to obtain the
amplitude with photon momenta k1 , . . . , kN and polarization vectors 1 , . . . , N , define for
every leg the field strength tensor
Fi
ki i ki i
(1.1)
and
Ftot
N
Fi .
(1.2)
i=1
The corresponding amplitude is then obtained by inserting Ftot into the effective
Lagrangian, and extracting the terms involving each F1 , . . . , FN precisely once:
(EH) [k1 , 1 ; . . . ; kN , N ] = L(iFtot )|F1 FN .
(1.3)
At one loop, the QED effective Lagrangian for the constant field strength case is just the
well known EulerHeisenberg Lagrangian [18,19] whose weak field expansion is known in
closed form. Nevertheless, it appears that (1.3) was previously applied only to the textbook
case of photonphoton scattering.
In [20,21] Dunne and one of the authors had considered the special case of constant selfdual background fields, and derived closed-form expressions for the corresponding twoloop EulerHeisenberg Lagrangian and its scalar QED analogue. Due to the well-known
correspondence between self-dual fields and helicity eigenstates [2226] this Lagrangian
still contains the full information on the low energy limit of the all + component of the
N -photon amplitudes in the helicity decomposition. The relation (1.3) could thus be used
to derive simple closed-form expressions for these amplitudes not only at one but also at
two loops, for arbitrary N , in scalar and spinor QED.
In the present paper we extend the same approach to the case of arbitrary helicity
assignments. Applying (1.3) to the EulerHeisenberg Lagrangian and its scalar QED
analogue will allow us to obtain closed-form expressions for the low energy limits of the
one-loop N -photon amplitudes with arbitrary helicity assignments. The standard spinor
helicity technique [2729] will turn out highly useful in working out the algebra of the
field strength tensors Fi .
Although various integral representations have been derived for the corresponding
two-loop effective Lagrangians [3036], for the case of a general constant field none of
them is sufficiently explicit to obtain corresponding all N formulas at the two-loop level.
Nevertheless, we will use the formulas given in [3032] to obtain the weak-field expansions
of these two-loop effective Lagrangians up to the order (F 10 ), which will allow us to
compute the corresponding photon amplitudes up to the ten point case.
337
2. One-loop spinor QED

Let us begin with spinor QED at the one-loop level. We will use the standard integral
representation of the EulerHeisenberg Lagrangian [18],
L(1)
spin
1
= 2
8

e2 2
e2 ab
dT m2 T
1
2
e
a b 2 .
T
tanh(eaT ) tan(ebT )
3
T
(2.1)
Here T denotes the proper time of the loop fermion, and a, b are related to the two
invariants of the Maxwell field by a 2 b 2 = B 2 E 2 , ab = E B. The charge e will
often be set to unity in the following. The subtraction of the terms of zeroth and second
order in a, b corresponds to on-shell renormalization. These terms are not relevant for our
purposes and will be omitted in the following. The invariants a, b can be related to the field
= 1 F ,
strength tensor F and its dual1 F
2

2

1
2 + 1 F F ,
F F + F F
a2 =
4
4

1
1
2
2
F F .
b2 =
(2.2)
F F + F F
4
4
We wish to use this Lagrangian to obtain the low energy limit of the on-shell N -photon
amplitude for arbitrary N 4 and with an arbitrary helicity assignment. Due to Furrys
theorem we can, of course, assume that N is even. Since in the Abelian case the ordering
of the legs does not matter we shall further assume that legs 1, . . . , K carry the helicity
+ and the remaining ones helicity . Also, by CP invariance flipping all helicities is
equivalent to changing all momenta from outgoing to ingoing. It is therefore sufficient
to consider the case K N K. To construct suitable polarization vectors we use the
standard spinor helicity formalism. In this formalism, a polarization vector with circular
polarization for a photon with momentum k is written as
q | |k
=
.
2q |k
(2.3)
Here q |k = u (q)u (k), etc. are basic spinor products, and q is a reference
momentum (see [37] for details and conventions). Changes of the reference momentum
amount to gauge transformations. As usual we will use the notation

ij ki
kj+ ,
(2.4)
[ij ] ki+
kj .
So, let us use (2.3) with some arbitrary choice of reference momenta qi to define
+
polarization vectors 1+ , . . . , K
, K+1 , . . . , N
. In the corresponding field strength tensor
for leg i
ki i
i
ki
Fi
1 We work in Minkowski space with = diag(1, 1, 1, 1) and

0123 = 1.
(2.5)
338
the dependence on qi already drops out, as is easily verified. Using standard manipulations
(see, e.g., [37]) the following identities are found to hold

1
= [ij ]2 ,
2

1
Fi , Fj
= ij 2 ,
2
+
Fi , Fj = 0,

tr Fi+ Fj = 0.
Fi+ , Fj+
(2.6)
(2.7)
(2.8)
(2.9)
Moreover, as expected on general grounds [2226] one finds the self duality properties
= iF .
F
i
i
(2.10)
With the help of these relations it is easy to compute the two Maxwell invariants for the
case of F = Ftot :
1
Ftot Ftot = + + ,
4
1
tot
= i(+ ),
Ftot F
4
where we have introduced
1
1
+
[ij ]2,

2
2
1i<j N
Using (2.11) in (2.2) yields
a = + + ,
b = i
(2.11)
ij 2 .
(2.12)
1i<j N

+ .
(2.13)
The choice of sign for a, b does not matter since a and b appear only squared in the
Lagrangian (2.1). Similarly, there is no need to introduce a sign convention for .

Using (2.13) in (2.1) we get (omitting the subtraction terms)
L(1)
spin (iFtot )
1
= 2
8

0
( + + )( + )
dT m2 T
e
.
T
tan(( + + )T ) tan(( + )T )
(2.14)
As we explained in the introduction, the right-hand side constitutes a generating functional

for the one-loop (on-shell) photon amplitudes. The N -photon amplitude will be obtained
by a double truncation of this formal expression: first, it must be expanded in powers of
N kept from this series. Then, from the result those
+ , , and only the part of order Ftot
terms should be extracted involving each individual Fi just once.
Using the Taylor series,

22n B2n 2n
x
=
x
(1)n
tan x
(2n)!
n=0
(2.15)
339
(the B2n are Bernoulli numbers) the first step yields

(1)
Lspin (iFtot ) =

K NK

N
m4 2e N (1) K N K
,
c
+2 2 ,
spin
8 2
m2
2
2
N=4
where
(1)
cspin
K N K
,
2
2
(2.16)
K=0
K even
= (1)N/2 (N 3)!
K NK

k=0 l=0
(1)NKl
Bk+l BNkl
.
k!l!(K k)!(N K l)!
(2.17)
Here we have omitted the irrelevant terms of order 0 , 1 . According to the above, the
amplitude with K + and N K helicities is obtained from the corresponding term
in the sum (2.16) by picking out the terms multilinear in the Fi s. It is immediately seen
that such terms exist only if K is an even number. Thus all amplitudes with an odd number
of + helicities do, in fact, vanish in the low energy limit. For K even, let us define
+
(+ )K/2
all different
K

2

(K/2)!
= K/2 [12]2[34]2 (K 1)K + all permutations ,
2
NK
NK ( ) 2
all different
=

2
2
2
( NK
2 )!
(K + 1)(K + 2) (K + 3)(K + 4) (N 1)N
NK
2 2

+ all permutations .
The final result for the amplitude can then be written as

(1)(EH) +
+

spin
1 ; . . . ; K
; K+1; . . . ; N

m4 2e N (1) K N K
+
,
= 2
cspin
NK
K
8 m2
2
2
(2.18)
(2.19)
(here and in the following we omit the momenta k1 , . . . , kN in the argument of amplitudes).
We remark that the introduction of the variables is not essential in this calculation.
An alternative, though less elegant, way of arriving at the same result would be to expand
L(iFtot ) directly in powers of Ftot , perform the truncation to the multilinear part of the
N terms, and only after this use the spinor helicity identities (2.6), (2.7).
order Ftot
3. One-loop scalar QED

The scalar QED case is completely analogous, and we will write down only the main
formulas. The analogue of the EulerHeisenberg Lagrangian (2.1) for the scalar QED case
340
was given by Schwinger [38]:

(1)
Lscal
1
=
16 2

e2 2
1
e2 ab
dT m2 T
2
e
+
a b 2 .
T
sinh(eaT ) sin(ebT )
6
T
(3.1)
Using the Taylor expansion

x
(22n 2)B2n 2n
=
x
(1)n
sin x
(2n)!
(3.2)
n=0
we can use the same procedure as in the spinor QED case. Again the result vanishes for
odd K, and for even K one obtains a formula analogous to (2.19):
(1)(EH) +
+

1 ; . . . ; K
; K+1; . . . ; N
scal

N

2e
m4
(1) K N K
+
K
,
=
(3.3)
cscal
NK ,
2
2
16 2 m2
where now

K NK

(1) K N K
N/2
= (1) (N 3)!
,
cscal
(1)NKl
2
2
k=0 l=0
(1 21kl))(1 21N+k+l )Bk+l BNkl

.
k!l!(K k)!(N K l)!
(3.4)
4. Two-loop scalar and spinor QED

The following integral representation was obtained in [30] for the two-loop generalization of the EulerHeisenberg Lagrangian (2.1):

T
K0 (T )
(2)

Lspin =
dT
dT
)
K(T
,
T
16 3
T
0
0

2
5
+ K0 (T ) ln m T +
(4.1)
,
6
where

a 2 b2 2
4m (SS + P P )I0 + I
PP

2
1
4m2 +
T T (T + T )
T +T

2

5T T
a 2 b2
2
2

2m 2T + 2T T T
+
3
T +T

ab
1
a 2 b2
2
2
,
K0 (T ) = em T 4m2
(4.2)
T
tanh(aT ) tan(bT ) T
3

K(T , T ) = e
m2 T
I0 =

B
1
ln
,
B A
A
I=
341
q/B p/A
q p
I0
,
B A
B A
a 2 cos(b(T T ))
b 2 cosh(a(T T ))
,
q =2
,

sinh(aT ) sinh(aT )
sin(bT ) sin(bT )
P = sinh(aT ) sin(bT ),
S = cosh(aT ) cos(bT ),

A = a coth(aT ) + coth(aT ) ,
B = b cot(bT ) + cot(bT ) .
p=2
(4.3)
Here is the EulerMascheroni constant. The charge e has been set to unity.
In contrast to the one-loop formula (2.1) it is not known how to obtain from this integral
representation a closed-form expression for the coefficients of the weak field expansion.
Therefore, at two loops we contend ourselves with a calculation of this expansion to a
certain order. Using M ATHEMATICA we have found it straightforward to compute this
expansion up to the order (F 10 ). As in the one-loop case, from the resulting polynomial in
a, b we can directly read off the helicity amplitudes for N = 4, 6, 8, 10. To obtain a nonvanishing result, again we have to assume that not only N but also K are even. Its form is
analogous to (2.19):
(2)(EH) +
+

; K+1 ; . . . ; N
1 ; . . . K
spin

m4 2e N (2) K N K
+
K
,
cspin
NK ,
=
(4.4)
2
2
8 2 m2
(2) K NK
( 2 , 2 ) are given in Table 1.
where the coefficients cspin
For the scalar QED case, we use the similar representation [31]

T
0 (T )
K
(2)

Lscal =
dT
dT
)
K(T
,
T
T
32 3
0
0

2
7
0 (T ) ln m T +
+K
,
6
(4.5)
where now
2 2

, T ) = em2 (T +T ) a b m2 I0 I
K(T
PP
2

1
1
m2
T T (T + T )
T +T

a 2 b2
11T T
m2 (T + T )2 m2 T T
,
6
(T + T )

1
1
a 2 b2
ab
m2 T
2

K0 (T ) = e
(4.6)
+
m +
,
2 T
sinh(aT ) sin(bT ) T 2
6
q/B
p/A
q p
I0
,
I =
B A
B A

q = 2b2 cot(bT ) cot(bT ) + 3

p = 2a 2 coth(aT ) coth(aT ) 3 ,
(4.7)
342
Table 1
Two-loop coefficients

NK
c(2) K
2, 2
c(2) (5, 0)
c(2) (4, 1)
c(2) (3, 2)
c(2) (4, 0)
c(2) (3, 1)
c(2) (2, 2)
c(2) (3, 0)
c(2) (2, 1)
c(2) (2, 0)
c(2) (1, 1)
Scalar QED
Spinor QED
611
80640 2
349609
3628800 2
688637
2332800 2
317
40320 2
8707
1814400 2
3190547
8164800 2
67
12800 2
273619
6350400 2
2055163
25401600 2
2221
403200 2
151379
6350400 2
37763
282240 2
13
1920 2
8563
259200 2
7
960 2
5821
129600 2
3
128 2
307
5184 2
5
192 2
391
2592 2
I0 , P , A, B are as in (4.3). Computation of the weak field expansion to the same order
(F 10 ) yields
(2)(EH) +
+

scal
1 ; . . . K
; K+1 ; . . . ; N

m4 2e N (2) K N K
+
K
,
=
(4.8)
cscal
NK
2
2
16 2 m2
with coefficients also given in Table 1.
5. Conclusions
To summarize, we have shown here that the use of the effective action, when combined
with spinor helicity techniques, provides a simple and elegant way to obtain information
on the low energy limit of the QED N -photon amplitudes. This has allowed us to
derive an explicit formula for the one-loop N point amplitudes, as well as for the
two-loop amplitudes up to the ten-point case. In particular, it has turned out that all
amplitudes with an odd number of + helicities vanish in the low energy limit. From
the approach presented here it is clear that this property follows directly from the fact
that the constant field effective action can be written as a function of the two Maxwell
invariants. We therefore conclude that this vanishing must persist to all loop orders. Since
these amplitudes are not forbidden by any known symmetries, and indeed, the one-loop
four-point (+ + + ) amplitude is known to be non-vanishing with full momentum [13],
this comes rather unexpected (for the special case of the amplitudes with all helicities but
one positive this vanishing had been noted already in [21]).
343
Obviously, the self-duality relations fulfilled by field strength tensors with definite
helicities, Eq. (2.10) play an important part in these simplifications. We expect that these
relations, as well as the variables , will also have a useful role to play for the photon
amplitudes at full momentum. One indication for this is the appearance of factors of traces
of products of field strength tensors in the parameter integrals for the N -photon amplitudes
generated by the BernKosower formalism [1,3943]. Work in this direction is in progress.
Acknowledgements
We would like to thank L. Dixon for information concerning [37]. L. Martin thanks
the IFM, UMSNH, for their hospitality and the Secretaria de Relaciones Exteriores for
funding. We are also indebted to the referee for several important remarks.
References
[1] Z. Bern, D.A. Kosower, Phys. Rev. Lett. 66 (1991) 1669;
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
[2] Z. Bern, L.J. Dixon, D.A. Kosower, Annu. Rev. Nucl. Part. Sci. 46 (1996) 109, hep-ph/9602280.
[3] A.I. Davydychev, Phys. Lett. B 263 (1991) 107.
[4] Z. Bern, L.J. Dixon, D.A. Kosower, Nucl. Phys. B 412 (1994) 751, hep-ph/9306240.
[5] J.M. Campbell, E.W.N. Glover, D.J. Miller, Nucl. Phys. B 498 (1997) 397, hep-ph/9612413.
[6] R. Pittau, Comput. Phys. Commun. 104 (1997) 23, hep-ph/9607309.
[7] S. Weinzierl, Phys. Lett. B 450 (1999) 234, hep-ph/9811365.
[8] J. Fleischer, F. Jegerlehner, O.V. Tarasov, Nucl. Phys. B 566 (2000) 423, hep-ph/9907327.
[9] T. Binoth, J.P. Guillet, G. Heinrich, Nucl. Phys. B 572 (2000) 361, hep-ph/9911342.
[10] T. Binoth, J.P. Guillet, G. Heinrich, C. Schubert, Nucl. Phys. B 615 (2001) 385, hep-ph/0106243.
[11] G. Mahlon, Phys. Rev. D 49 (1994) 2197, hep-ph/9311213.
[12] R. Karplus, M. Neuman, Phys. Rev. 80 (1950) 380.
[13] V. Costantini, D. De Tollis, G. Pistoni, Nuovo Cimento 2A (1971) 733.
[14] G. Passarino, M. Veltman, Nucl. Phys. B 160 (1979) 151.
[15] A. Denner, Fortschr. Phys. 41 (1993) 307.
[16] D.A. Dicus, C. Kao, W.W. Repko, Phys. Rev. D 57 (1998) 2443, hep-ph/9709415.
[17] C. Itzykson, J. Zuber, Quantum Field Theory, McGrawHill, 1985.
[18] W. Heisenberg, H. Euler, Z. Phys. 98 (1936) 714.
[19] V. Weisskopf, Kong. Dans. Vid. Selsk. Math. Fys. Medd. XIV No. 6 (1936), reprinted, in: J. Schwinger
(Ed.), Quantum Electrodynamics, Dover, New York, 1958.
[20] G.V. Dunne, C. Schubert, Phys. Lett. B 526 (2002) 55, hep-th/0111134.
[21] G.V. Dunne, C. Schubert, JHEP 0208 053 hep-th/0205004.
[22] M.J. Duff, C.J. Isham, Phys. Lett. B 86 (1979) 157;
M.J. Duff, C.J. Isham, Nucl. Phys. B 162 (1980) 271.
[23] W.A. Bardeen, preprint FERMILAB-CONF-95-379-T.
[24] A.A. Rosly, K.G. Selivanov, Phys. Lett. B 399 (1997) 135, hep-th/9611101.
[25] D. Cangemi, Nucl. Phys. B 484 (1997) 521, hep-th/9605208;
D. Cangemi, Int. J. Mod. Phys. A 12 (1997) 1215, hep-th/9610021.
[26] G. Chalmers, W. Siegel, Phys. Rev. D 54 (1996) 7628, hep-th/9606061.
[27] F.A. Berends, R. Kleiss, P. De Causmaecker, R. Gastmans, T.T. Wu, Phys. Lett. B 103 (1981) 124.
[28] R. Kleiss, W.J. Stirling, Nucl. Phys. B 262 (1985) 235.
[29] Z. Xu, D.-H. Zhang, L. Chang, Nucl. Phys. B 291 (1987) 392.
[30] V.I. Ritus, Zh. Eksp. Teor. Fiz. 69 (1975) 1517, Sov. Phys. JETP 42 (1975) 774 (in English).
344
[31] V.I. Ritus, Zh. Eksp. Teor. Fiz. 73 (1977) 807, Sov. Phys. JETP 46 (1977) 423 (in English).
[32] V.I. Ritus, The Lagrangian function of an intense electromagnetic field, in: V.I. Ginzburg (Ed.), Proc.
Lebedev Phys. Inst., Issues in Intense-Field Quantum Electrodynamics, Vol. 168, Nova Science, New York,
1987.
[33] W. Dittrich, M. Reuter, Effective Lagrangians in Quantum Electrodynamics, Springer, 1985.
[34] M. Reuter, M.G. Schmidt, C. Schubert, Ann. Phys. (N.Y.) 259 (1997) 313, hep-th/9610191.
[35] D. Fliegner, M. Reuter, M.G. Schmidt, C. Schubert, Teor. Mat. Fiz. 113 (1997) 289, Theor. Math. Phys. 113
(1997) 1442 (in English), hep-th/9704194.
[36] B. Krs, M.G. Schmidt, Eur. Phys. J. C 6 (1999) 175, hep-th/9803144.
[37] L. Dixon, TASI Lectures, Boulder TASI 95 539, hep-ph/9601359.
[38] J. Schwinger, Phys. Rev. 82 (1951) 664.
[39] Z. Bern, D.C. Dunbar, Nucl. Phys. B 379 (1992) 562.
[40] M.J. Strassler, Nucl. Phys. B 385 (1992) 145.
[41] M.J. Strassler, preprint SLAC-PUB-5978 (1992), unpublished.
[42] M.G. Schmidt, C. Schubert, Phys. Lett. B 318 (1993) 438, hep-th/9309055.
[43] C. Schubert, Eur. Phys. J. C 5 (1998) 693, hep-th/9710067.

On solutions of the BalitskyKovchegov equation

with impact parameter
K. Golec-Biernat a , A.M. Stasto a,b
a Institute of Nuclear Physics, Radzikowskiego 152, Krakw, Poland
b DESY, Theory Division, Notkestrasse 85, 22603 Hamburg, Germany
Abstract
We numerically analyze the BalitskyKovchegov equation with the full dependence on impact
parameter b. We show that due to a particular b-dependence of the initial condition the amplitude
decreases for large dipole sizes r. Thus the region of saturation has a finite extension in the dipole
size r, and its width increases with rapidity. We also calculate the b-dependent saturation scale and
discuss limitations on geometric scaling. We demonstrate the instant emergence of the power-like tail
in impact parameter, which is due to the long range contributions. Thus the resulting cross section
violates the Froissart bound despite the presence of a nonlinear term responsible for saturation.
PACS: 12.38.-t
1. Introduction
The high energy limit of QCD is one of the most intriguing aspects of hadronic
physics. With the advent of new generation accelerators like HERA, Tevatron, RHIC
and in a near future LHC, the basic problems of strong interactions are experimentally
studied and confronted with theoretical predictions of high energy QCD. The important
discovery of the rise of the proton structure functions at HERA [1] at small values of
the Bjorken variable x (which is equivalent to the high energy limit) is an example of
such a confrontation. This rise was predicted by high energy QCD [2] and is related to
the increase of the gluon density. Ultimately the rise has to be damped by the presence
of the saturation effects which enter via nonlinear modification to the QCD evolution as
E-mail address: stasto@mail.desy.de (A.M. Stasto).
doi:10.1016/j.nuclphysb.2003.07.011
346
K. Golec-Biernat, A.M. Stasto / Nuclear Physics B 668 (2003) 345363
first proposed in the pioneering work [3]. Thus QCD at high energy is the theory of high
density systems of colored particles. In such systems hard scales appear which allow to
apply perturbative techniques although effective interactions in the dense partonic system
are of nonperturbative origin. The interplay between hard and soft (perturbative and nonperturbative) aspects of QCD, which also touches in the operational way the issue of
confinement, is the most exciting element of high energy QCD.
The effective theory which describes high energy scattering in QCD is color glass
condensate [4]. The basic equations of this theory [5] are equivalent [6] to the hierarchy of
equations derived by Balitsky [7] and later on reformulated in a compact form by Weigert
[8]. These equations contain the BFKL evolution and also the triple Pomeron vertex [9]. In
this work we present the results of numerical studies of these basic equations. To be precise
we study the simplified version of the hierarchy of Balitskys equations which reduces
to one equation in the limit of large number of colors. This is the equation obtained by
Kovchegov [10] in the dipole approach [11] to high energy scattering in QCD. With this
equation the deep inelastic leptonnucleus scattering can be described and also information
on small-x hadronic wave function be obtained.
We study this equation in the full form, including the impact parameter dependence b.
Previous analytical [12,13] and numerical [1416] studies were done under a simplified
assumption of infinitely large and uniform nucleus, i.e., neglecting impact parameter b.
Recently, an approximate solution to this equation in semiclassical approach was
considered [17]. As we will show, the solutions of the full form of the BalitskyKovchegov
(BK) equation possess important new features in comparison to the uniform case, e.g.,
restricted scaling properties with impact parameter dependent saturation scale. The detailed
analysis of the b-dependence of the solution allows to study the high energy behaviour of
the N cross section. We show that the Froissart bound [18] is violated due to the long
range contributions in the kernel [19] which brings the issue of the lack of confinement
effects in the BK equation. The problem of the Froissart bound was also extensively
discussed in the same context in [19,20] based on analytical considerations.
The paper is organised in the following way. In the next section we briefly present
the BK equation and its symmetries. In Section 3 we describe the numerical methods of
finding the solution and discuss the initial condition. In Section 4 we present the resulting
amplitude N as a function of dipole size r and extract the b-dependent saturation scale.
In Section 5 we discuss the form of the impact parameter profile which emerges in the
evolution, and in particular we concentrate on the emergence of the power tails in b in the
amplitude. We also present the estimate of the cross section of the black disc radius and its
dependence on the rapidity. Finally, in Section 6 we state our conclusions.
2. The BalitskyKovchegov equation

The deep inelastic scattering of a lepton on a nucleus at high energy in the dipole picture
[11,21] is viewed in the nucleus rest frame as the splitting of an exchanged virtual photon
into a q q dipole and the subsequent interaction of the dipole with the nucleus. The latter
process is described by the dipole-nucleus scattering amplitude N(x, y), where x, y are
two-dimensional vectors of the transverse position of the dipole ends. Alternatively, one
347
can introduce the dipole vector r = x y, and the impact parameter b = (x + y)/2. Thus
in general, the amplitude depends on the four transverse degrees of freedom and rapidity,
Y = ln(1/x), playing the role of the evolution parameter
N(x, y, Y ) Nxy (Y ).
(1)
From now on, for shortness of the notation, we assume the Y -dependence implicit.
In the leading logarithmic approximation, the dipole scattering amplitude obeys a
nonlinear evolution equation derived by Balitsky and Kovchegov [7,10]
Nxy
= s
Y
d 2z
(x y)2
{Nxz + Nyz Nxy Nxz Nyz },
2 (x z)2 (y z)2
(2)
where s = s Nc / . In addition, one has to specify an initial condition at Y = Y0 :

Nxy = N 0 (r, b). The amplitude N(x, y) in (2) is given by the following correlator
N(x, y) =

1
Tr 1 U (x)U (y) ,
Nc
(3)
where the trace is done in the colour space, and the eikonal factor U is defined as the path
ordered exponential with the SU(N) gauge fields (in the gauge A
a = 0)

U (x) = P exp i dx T a A+
(4)
a (x , x) .
The averaging in (3) is performed over an ensemble of classical gauge fields. In
general, an infinite hierarchy of equations is found for correlators of the U factors [7]. In
the large Nc limit, however, the closed form (2) can be found for the two point amplitude
Nxy [10]. The linear part of (2) corresponds to the dipole version [11] of the BFKL equation
[22] at nonzero impact parameter and its solution has been studied in the Monte Carlo
simulation of oniumonium scattering [23,24]. The additional quadratic term emerges due
to the summation of multiple interactions of more than one dipole with the nucleus at the
same time.
The r.h.s. of Eq. (2) has a nice geometrical interpretation: the parent dipole (x, y) splits
into two new dipoles (x, z) and (y, z), and the summation is taken over all new dipoles.
On the other hand, the nonlinear term describes the recombination of the two dipoles (x, z)
and (y, z) into one (x, y). The three dipoles form a triangle, shown in Fig. 1. The first part
of the integral kernel in (2) only depends on the triangle sides: |x y|, |x z| and |y z|.
A nontrivial dependence on the position of the triangle in the plane, i.e., the dependence
on the impact parameter vectors bxy , bzx , byz , is introduced through the arguments of the
amplitudes N . In particular, it is interesting to study how the b-dependence introduced
by an initial condition N 0 (r, b) propagates with increasing Y . Let us also note that the
singularities at z = x, y in Eq. (2) are integrable provided
lim Nxy |x y| ,
xy
> 0.
Thus the dipole which shrinks to a point does not scatter.
(5)
348
Fig. 1. The triangle geometry in Eq. (2). The points x, y, z are the dipole ends in the transverse space, and the
vectors bxz , bxy , bzy are the impact parameters of the three dipoles.
2.1. Symmetries of the BK equation

The BK equation (2) has a rich symmetry structure. Introducing the complex number
notation for the transverse vectors, e.g., x = x1 + ix2 and x = x1 ix2 for x = (x1 , x2 ), it
can be easily shown that the measure in Eq. (2),
s
(x y)2
d 2 z,
2 (x z)2 (y z)2
(6)
is invariant under the Mbius transformation1

a x + b
,
(7)
cx + d
where the parameters a, b, c, d C and ad bc = 0. Identical transformations are also
applied to y (y)
and z (z). Thus the BK equation is covariant with respect to the
Mbius transformation. In particular, the following elementary transformations from (7)
are relevant for our discussion
x
ax + b
,
cx + d
global two-dimensional translations by vectors b : x x + b;

global two-dimensional rotations by angles : x O()x;
scale transformations with a real, positive parameter : x x,
inversion (in complex notation): x 1/x.
Notice, that if an initial condition N 0 is invariant under any of the discussed

transformations, the solution of the BK equation Nxy (Y ) preserves the corresponding
symmetry. In particular, if an initial condition is invariant under translations and rotations,
N(x, y) = N 0 (|x y|), the solution at any rapidity Y has the same property, i.e., it only
depends on the dipole size r = |x y| but not on the impact parameter b = (x + y)/2. The
1 Provided
s is kept constant.
349
Fig. 2. The dipole degrees of freedom: the dipole vector r and impact parameter b.
problem of finding solution simplifies enormously in this case since only one degree of
freedom is relevant, namely the dipole size r.
Physically, this approximation (called local approximation) corresponds to an infinitely
large and uniform nucleus. Previous analytical [12,13] and numerical [1416] studies of
the BK equation were based upon this assumption. In this approximation, the solution
shows saturation, N(r) 1, with the characteristic scale Qs (Y ). For dipoles smaller than
the inverse of the saturation scale, r < 1/Qs (Y ), the solution is governed mainly by the
linear term of Eq. (2) and shows the exponential rise in rapidity, N exp(P Y ) where
P = 4 ln 2 s is the intercept of the BFKL kernel. On the other hand, in the region where
dipoles are large, r > 1/Qs (Y ), the nonlinear term slows down the rise and eventually
the amplitude saturates to 1. The saturation scale Qs (Y ) depends on the rapidity in the
following way [1214,16,25]
Qs (Y ) = Q0 exp( s Y ),
(8)
2. The solution in the local approximation also exhibits a

where the
property of the geometric scaling [26], namely for r > 1/Qs (Y )

N(r, Y ) N rQs (Y ) ,
(9)
coefficient2
which means that the amplitude N in the saturated region only depends on one combined
variable rQs (Y ) instead of r and Y separately.3 Thus the diffusion into infrared, typical
for the linear BFKL equation, is damped by the emergence of the saturation scale Qs (Y )
[16].
Since in our analysis we want to study the impact parameter dependence, we obviously
have to abandon the assumption about the translational invariance. However, in order to
simplify the problem, we adopt more physical assumption that our nucleus is cylindrically
symmetric, i.e., N(x, y) is invariant under the global rotation. It means that in the
parameterisation of the dipole position, see Fig. 2,
(x, y) = (r, b, , ),
(10)
2 To be precise in Ref. [12] the coefficient was found to be the same as the Pomeron intercept = /
P s =
4 ln 2, but this value was not confirmed by subsequent analytical and numerical studies.
3 This is also a feature of the saturation model [27].
350
we drop the dependence on the azimuthal angle . Note that we keep the dependence on
which is the angle between the vectors r and b
rb
(11)
.
rb
The assumption about cylindrical symmetry reduces the number of parameters in the
amplitude N to four: three degrees of freedom for a dipole and the evolution variable Y .
cos =
3. The numerical method of finding solution

The numerical method for the solution of the BK equation is very similar to the one used
for the solutions of the linear equations, see [28]. We discretise the amplitude N(r, b, cos )
in all 3 variables (ln r, ln b, cos ). A simple linear interpolation has been chosen which is
the fastest in this case. To find the solution as a function of rapidity we take in the first step
N 0 , the initial condition at Y = Y0 , and evaluate Eq. (2) with the step $Y . This gives the
first approximation for the solution at Y = $Y

0

(1)
0
0
0
0 0
Nxz + Nyz
= Nxy
+ $Y
Nxy
Nxz
Nyz ,
Nxy
(12)
z
where we symbolically denoted the integration over z with the measure (6). If the relative
(1)
0 )/N 0 | < (some accuracy) at each point of the grid, we finish our
difference |(Nxy Nxy
xy
procedure, otherwise solution (12) is used to find the second approximation from:

$Y
0
(2)
0
0
0
0 0
Nxz + Nyz
= Nxy
+
Nxy
Nxz
Nyz
Nxy
2
z

$Y
(1)
(1)
(1)
(1) (1)
Nxz + Nyz
Nxy
Nxz
Nyz
+
2
z

(1)

$Y
(1)
0
0
+
(13)
Nxz Nxz
Nyz Nyz
.
6
z
The r.h.s. of Eq. (13) was found after integrating over Y from 0 to $Y , assuming a linear
interpolation in Y between N 0 and N (1) ($Y ). This is justified for small enough value of
$Y . We iterate in this way, using Eq. (13), until the desired accuracy is achieved.
Usually only couple of iterations are needed to find the right answer, in fact we fix the
maximal number of iterations to 5. In order to get satisfactory results one has to work with
a grid which is at least (100r 100b 20c ). Each step in rapidity produces about 1.5 MB
of output data and takes about 300 min. when run on PC machine with 2.5 GHz processor
and 2.0 GB RAM memory.
One of the important issues while studying the BK equation is the choice of the initial
condition at rapidity Y = 0. As mentioned above, because of the cylindrical symmetry our
initial amplitude N 0 should only depend on the three variables (r, b, ). Since we have
nearly no information on the angle between r and b, we do not assume any dependence
on it in the initial conditions. We shall see that the nontrivial dependence is nevertheless
351
generated through the evolution. As far as the r and b dependence is concerned we have
chosen the initial distribution in the GlauberMueller form [29,30]

N 0 (r, b) = 1 exp r 2 S(b) ,
(14)
with S(b) being a steeply falling profile in b, e.g., S(b) exp(b2). The GlauberMueller
formula resums the multiple scatterings of a single dipole on a nuclear target and it has
been advocated to be a natural choice for the starting distribution of the BK equation [10].
Let us note that the above distribution has a property that for any fixed value of impact
parameter it saturates to 1 for sufficiently large values of the dipole size r. More detailed
analysis with other forms of the initial conditions will be presented elsewhere [31].
4. The dependence of the solution on a dipole size r

We have performed the numerical evolution of the BK equation starting from formula
(14) as the initial condition with the following impact parameter profile

S(b) = 10 exp b2/2 .
(15)
Throughout this work we keep the coupling constant in (2) fixed, s = 0.2. The running of
the coupling, although physically more justified, would be an additional complication to
the solution which we would like to avoid at this stage.4 Nevertheless, one should stress that
a running coupling is an important NLL effect [34], which should be taken into account in
future investigations. We are going to study now how the initial condition (14) is modified
by the evolution in rapidity. We concentrate first on the dependence of the solution on a
dipole size r.
The results of our numerical analysis are shown in Fig. 3 for two values of the impact
parameter: b = 0.2 and b = 5. We have also fixed the orientation of the dipole such that
r b that is cos = 0, see Eq. (11). For the values of r which are small (here r 1) the
amplitude rises with increasing values of rapidity Y . In this region, the linear part of the
BK equation dominates. With increasing r, for fixed rapidity, the amplitude finally reaches
the saturation value Nsat = 1, were the nonlinearity of the BK equation is crucial. This
is mostly visible for small value of impact parameter, Fig. 3 (left). For a larger value of
impact parameter, b = 5, in the plot to the right, the saturation has almost been reached
only for the highest value of rapidity shown Y = 11. It is interesting that the amplitude has
a maximum for the dipole size which is twice its impact parameter r = 2b, that is r = 10
in this case.
At large r, the amplitude first decreases until rapidity Y 4 and then starts to increase
for larger values of rapidities. This should be contrasted to the small r region just discussed
where the amplitude always increases with increasing Y . This behaviour is a consequence
of the initial condition which we have chosen to start with. What is also remarkable is
the fact that this behaviour of the amplitude is quite different from the case with the
4 In the local approximation, running coupling changes rapidity dependence of the saturation scale: Qrun (Y )
s
exp(c Y ), see [16,32,33].
352
Fig. 3. The amplitude N (r, b, ; Y ) as a function of dipole size r for the indicated values of rapidity and two
values of the fixed impact parameter: b = 0.2 (left) and b = 5 (right). The orientation of the dipole cos = 0. The
dashed line is the input distribution (14) with the profile (15).
translational invariance (no b-dependence). The amplitude decreases with increasing r

even if the initial condition is saturated (N 0 = 1) in this region. The reason for this
behaviour can be easily understood analysing the solution after the first iteration (12)

0

s
(x y)2
(1)
0
0
0
0 0
Nzy + Nxz
Nxy
Nxz
Nyz , (16)
Nxy = Nxy + $Y
d 2z
2
2
2
(x z) (y z)
with the GlauberMueller initial condition

N 0 (r, b) = 1 exp cr 2 exp b 2 d .
(17)
Saturation means that for a large enough r = |x y| (and fixed b) Nxy 1, i.e., a large
dipole is totally absorbed. This is indeed the case for the initial condition (17). After the
(1)
first evolution step, however, Nxy becomes less than 1 for large r due to the large and
negative integrand in the curly brackets in (16), which leads to the effect shown in Fig. 3
at large r.
0 1 in Eq. (16), the integrand is always negative or equal zero
Indeed, if Nxy

0
0
0 0
0
0
1 Nxz
0.
+ Nyz
1 Nyz
Nxz = 1 Nzy
Nxz
(18)
In Fig. 1 we show an example of typical configurations of the (x, z) and (y, z) dipoles
which give dominant contribution to the integral in Eq. (16). For such a contribution
rxz ryz 2bxz 2byz r (for bxy 0), and for the initial condition (17) the
0 N 0 0 for the sufficiently large value of r. Thus the integral picks
amplitudes Nxz
yz
up significant negative contribution from this configuration. This should be contrasted
to the translationally invariant case with no b-dependence in the initial condition, e.g.,
N 0 (r) = 1 exp{r 2 }. For such initial condition the configuration from Fig. 1 gives
0 N 0 1, and the expression in Eq. (18) is negligible. One can prove that any other
Nxz
yz
configuration, for example, when |x z| |x y| or |x z| |x y|, leads to the same
result. Thus, in the case of infinitely large and uniform nucleus, the amplitude N stays
always saturated for large dipole sizes.

(1)
353
(1)
In the next evolution step, Nxy plays the role of the initial condition. Since Nxy is no
longer equal to 1 for large dipole sizes, the analysis becomes more complicated. At some
rapidity, the discussed integral becomes positive leading to the effect observed in Fig. 3
where N starts to grow again, reaching unity at large r for large values of rapidity.
In more physical terms the fall-off of the amplitude at large r can be explained by
realizing that the end points x and y of a sufficiently large dipole are in the region where
there is no gauge field. In this case U (x) = U (y) = 1, and the correlator (3) vanishes. It
simply means that the dipole is so large that it misses the localised target given by S(b).
On the other hand, the nonvanishing value of the amplitude for very large values of r in
the previous studies in the local approximation [1216] was a consequence of the fact that
the field was uniform and present everywhere, thus it did not matter how large the dipole
was, it always scattered. The appearance of the gauge field, signalled by the increase of N
with rising rapidity, corresponds to expansion of a black (or grey) region of a nucleus, see
Section 5.3.
4.1. Saturation scale and geometric scaling
The saturation scale Qs is defined from the condition

N(r = 1/Qs , b, , Y ) = ,
(19)
where the constant is of the order of unity ( = 1/2 in our numerical analysis) and
we have taken an average over the orientation of the dipole with respect to the vector of
impact parameter. Notice that after solving the above equation, the saturation scale depends
on impact parameter in addition to rapidity.5 It is evident that for the form of N , shown
in Fig. 3, there are two solutions of Eq. (19) for sufficiently large rapidities. One solution,
which corresponds to small dipole size, is the saturation scale Qs (b, Y ). And the other
one, for large dipole size which we call RH (b, Y ). Thus saturation (defined here as N > )
occurs over a finite region of dipole sizes
1/Qs (b, Y ) < r < RH (b, Y ).
(20)
The scale Qs (b, Y ) is the b-dependent saturation momentum introduced in [30]. The
b-independent saturation scale found in the previous analyses [1316] can be regarded as
an average over all area of interaction. The second scale RH (b, Y ) appears because we
have introduced exponential impact parameter profile, as has been explained in detail in
the previous section, and reflects the boundary of the nucleus. In Fig. 4 we show the form
of the impact parameter profile of the saturation scale Qs (b, Y ) for two different rapidities
Y = 5 and Y = 11. The saturation scale at Y = 0, shown by the dashed line for comparison,
is computed from the initial condition (14) with (15)

Q2s (b, 0) = Q20 exp b2 /2 ,
(21)
where we have fixed the normalisation Q0 to match the Qs (b, Y ) at small values of b = 0.1.
Thus the dashed lines in Fig. 4, show the profile of Qs (b, Y ) which we would get if the
5 Of course, Q depends also on but this dependence is irrelevant for our discussion.
s
354
Fig. 4. The dependence of the saturation scale Qs (b, Y ) on impact parameter for two fixed values of rapidity Y .
For the comparison the saturation scale at initial rapidity (21) is shown by the dashed line, normalised to the value
of Qs (b, Y ) at b = 0.1.
exponential impact parameter dependence set in the initial conditions would be preserved
through the evolution. We clearly see that for higher impact parameters, the exponential
fall-off set by initial condition is replaced by the power-like tail Qs (b, Y = 11) 1/b
with 1.62.0 for b > 7.
As it is evident from Fig. 3, both the saturation scale Qs (b, Y ) and RH (b, Y ) rise with
increasing rapidity, thus leading to broadening of the region in r for which saturation
occurs. The energy dependence of Qs (b, Y ) has been estimated to be proportional to
exp(s s Y ) with s 1.72.0 at the highest values of rapidity Y = 911. We also checked
that this value of s is not very sensitive to impact parameter. This would mean that (at
high rapidities) the saturation scale has a factorised form Qs (Y, b) = exp(s s Y )g(b). We
expect exponential dependence on rapidity of RH (b, Y ) exp(H s Y ) too, and we have
found H 2.9.
Since there are two scales in the problem, Qs and RH , geometric scaling of the form (9)
is not present anymore for arbitrarily large dipoles. However, when r RH (b, Y ) there
is still an approximate geometric scaling in the combined variable rQs (b, Y ) for fixed b.
This is shown in Fig. 5, obtained from Fig. 3 after rescaling r by exp(1.7 s Y ), i.e., by the
rapidity dependence of the saturation scale Qs (b, Y ), extracted in the region 7 Y 11.
As discussed, there is no geometric scaling for large dipole sizes. The problem of scaling in
the b-dependent case has already been addressed in [20] and in recent phenomenological
studies [35], and still needs deeper analysis.
Let us finally note that for our solution there is a region in b in which the saturation
scale does not exist at all. It is evident from the right-hand side plot in Fig. 3, where up to
the rapidities Y 6 the amplitude N < for all values of the dipole sizes r. In this case
Eq. (19) does not have a real solution.
355
Fig. 5. Geometric scaling for b = 0.2 obtained after rescaling r by the rapidity dependence of the saturation scale
Qs (b, Y ). The dotted, dashed and solid lines are for Y = 11, 9, 7 respectively. The orientation of the dipole has
been fixed such that cos = 0.
5. The impact parameter dependence of the solution

We will discuss now the impact parameter dependence of the solution to the BK
equation with the GlauberMueller input distribution (14), (15).
In Fig. 6 we show the amplitude N(r, b, ; Y ) as a function of impact parameter b for
fixed and small dipole size r = 0.1 and various values of rapidity. We have also fixed the
orientation of the dipole such that cos = 0. The first striking feature of the solution is
the fact that the steeply falling exponential dependence in b, given by the initial profile
S(b), is washed out by the evolution and instead clear power behaviour is generated for
large values of b. Initially, at lowest rapidity $Y = 0.1 the amplitude N(b) 1/b3.6 ,
whereas later on it has a milder dependence: N(b) 1/b with between 2 and 3 for
Y 5. The growth of the amplitude as a function of rapidity at large impact parameters
is exponential, N(Y ) exp(Y ) with 2.7 s for b = 10. This strong dependence is
clearly governed by the linear BFKL part of the equation since the amplitude is small
enough for the nonlinear term to be safely neglected and it is in a very good agreement
with the hard Pomeron intercept P = 4 ln 2 s = 2.77 s .
On the other hand, at small values of impact parameters, b < 1, where the amplitude
is large, we clearly observe that the growth of amplitude is strongly damped due to the
nonlinear term. This is the region of b where the saturation sets in first.
356
Fig. 6. The amplitude N (r, b, Y ) as a function of impact parameter b for different values of rapidity Y . The dipole
size and orientation are fixed, r = 0.1 and cos = 0. The dashed line is the input distribution (14) with the profile
(15).
5.1. The origin of the power-like tail

It has been argued in [19] that the power behaviour in impact parameter N 1/b
originates from the configurations with very large dipoles. Following [19] let us take
small dipole with size r = |x y| which is located far away from the target, at large
impact parameter b, in the area where the colour field is very weak. In that case one has
U (x) U (y) 1 and thus N(x, y) 0.
The nonvanishing contribution to the r.h.s. of Eq. (2) comes from configurations of large
dipoles with one end-point situated at x, and the other at z, close to the center of the target
where the field is strong. The phase in eikonal (4) oscillates strongly and thus U (z) 0.
Therefore N(x, z) N(y, z) 1, and this configuration gives

(x y)2
r2
r2
2
2
(N
+
N
N
N
)

z

R02 (Y ),
d
d z
xz
yz
xy
xz
yz
(x z)2 (y z)2
b4
b4
(22)
where R02 (Y ) is the area of strong field in the target over which we integrate. Strictly
speaking the statement that N(x, z) = 1 for the configuration considered above is valid
only at very high rapidities. Due to the initial conditions (14) at intermediate rapidities there
will be always such b, large enough, for which these configurations will have N(x, z) < 1,
however always N(x, z) N(x, y).
In order to check this statement numerically we perform only one iteration of the BK
equation with very small step in rapidity $Y = 0.1, Eq. (12), and divide the integration
region into two parts: |z b| < r0 and |z b| > r0 with the cutoff r0 = 1 for the choice of
357
Fig. 7. The r.h.s. of the BK equation (2) computed for the initial condition (14), (15) (solid line) and the
contributions from the short and long dipoles (dashed lines), see (23), plotted as a function of b for the fixed
dipole size r = 0.1 and orientation cos = 0.
small dipole r = 0.1. To be precise we evaluate two integrals:

short
long

r0 |z b| + |z b| r0

d 2 z (x y)2 0
0
0
0 0
N
+
N
N
N
xz
yz
xy
xz
yz
(x z)2 (y z)2
(23)
with N 0 being our initial condition. The results are presented in Fig. 7, where the short
range contribution dominates at small values of impact parameters and the long range one
appears to be responsible for the power behaviour in b at large values. The point at which
the long range contribution starts to dominate over the short range one is determined by
the form of the initial condition.
5.2. The angular dependence for large dipole sizes
It is interesting to study also the angular dependence of the solution N . For sufficiently
small dipole sizes r RH , the angular dependence is negligible. However, when r RH
the difference in the amplitude due to the dipole orientation is quite substantial, especially
at large impact parameters b r.
In Fig. 8 we plot N(r, b, ; Y ) as a function of b for a large dipole with r = 20.
Two rapidities were considered. In both cases we have compared calculation with r b
358
Fig. 8. The amplitude N (r, b, ; Y ) as a function of impact parameter b for two rapidities: Y = 2 (left) and Y = 8
(right). The dipole size is fixed to r = 20. The solid line corresponds to the case r b, and for the dashed line
r b. The dotted line is the input distribution (14) with (15).
(solid line) and r b (dashed line) together with the input distribution (dotted line).
The calculation with parallel orientation shows a characteristic peak at b = r/2. This
corresponds to the situation when one end of the dipole, x, is situated in the center of
the target where the field is strong U (x) 0 and the other end, y, is located in the area
where the field is very weak U (y) 1. Thus the amplitude (3) N 1. On the other hand
the large dipole perpendicular to the impact parameter axis has both ends in the region
where the field is weak and therefore N(x, y) < N(x, y) .
5.3. Black disc radius and unitarity bound
The black disc radius RBD (r, Y ) defines the region in the impact parameter space where
the amplitude N saturates (for a given dipole size r and rapidity Y ). It is defined as the
solution of the equation

N(r, b = RBD , , Y ) = ,
(24)
with respect to the impact parameter b. As before we choose = 1/2 and average over the
angle . Thus the black disc radius is a function of the dipole size and rapidity.
The expansion of the black disc area with increasing Y is very important for the
behaviour of the total dipolenucleus cross section with energy s eY ,

(r, Y ) = 2 d 2 b N(r, b, Y ).
(25)
This cross section is bounded from below by the cross section integrated over the black
disc area

2
BD (r, Y ) = 2 d 2 b N(r, b, Y )(N ) 2RBD
(26)
(r, Y ).
Thus the problem of the Froissart bound [18], (r, Y ) Y 2 for asymptotically high
energies, can be explicitly studied by looking at the rapidity dependence of the black disc
radius.
359
Fig. 9. The dipole cross section BD (r, Y ) computed from Eq. (26), plotted as a function of rapidity Y for three
different dipole sizes r. The dashed line corresponds to Test = cY .
In Fig. 9 we plot BD from (26) as a function of rapidity. For the comparison we

illustrate the Test = cY behaviour which would be present if the initial profile in impact
parameter S(b) exp(b 2 ) was preserved. Clearly the behaviour of the black disc cross
section is much faster than the linear dependence in rapidity. We have found that
0
exp(BD s Y ).
RBD = RBD
(27)
The extracted value of BD 0.60.7 for r = 0.31 and BD 0.91 for r = 330 at the
highest rapidity Y = 11.
The value of BD increases rapidly for dipole sizes smaller than the given range
above, mostly due to the pre-asymptotic effects. The value of BD is smaller than BD =
0.87P/(2 s ) 1.2 which was quoted in Ref. [19]. The origin of this discrepancy can be
roughly explained by noting that the power given in [19] most probably overestimates the
growth of the black disc radius because it was derived from the analysis of the saddle point
solution to the linear equation. On the other hand in our numerical simulations the value
of BD = 0.91.0 for large r is probably closer to the asymptotic one. It is due to the fact
that for large r, the amplitude has a clear peak for b = r/2 which is generated entirely
through the evolution. In other words the amplitude in this region is less sensitive to the
initial profile in b. It is also interesting that this value BD is consistent with the relation
BD = 1/2s , based on the argument of conformal invariance of the equation [36].
Let us recall the saddle point solution to the dipole version of the BFKL equation in the
transverse space [23,37]

1
rr0
16b2
2 rr0 1
,
ln
exp
ln
NBFKL (r0 , r, b, Y )
(28)
P
rr0
16b2 (kY )3/2
16b2 kY
360
0
where r0 is the size of the initial dipole. Eq. (28) holds if 1 ln rr
kY (k = 14(3) s ).
b2
2
The amplitude is a function of rr0 /b only, instead of r and b separately, which is the
consequence of conformal invariance. This should also be true for the nonlinear equation,
i.e., NBK (rr0 /b2 , Y ) if r, r0 < b. Under that assumption the conditions for the saturation
scale (19) and black disc radius (24) mean that (roughly)6
r
2 (r, Y )
RBD
1
b2 Qs (b, Y )
2 (r, Y )
RBD
= rb2 .
Qs (b, Y )
(29)
Since in the last equality the right-hand side does not depend on the rapidity so should
not the left-hand side too. This means that if we have RBD exp(BD s Y ) and Qs
exp(s s Y ) then
s = 2BD .
(30)
Relation (29) means also that Qs 1/b2 , which we have already checked to be
2 r.
approximately true (at least for large values of b > 7 see Section 4.1) and also RBD
2
a
We have verified that RBD (r, Y = 11) r with a 0.7 for r = 0.11.0 and a 1.01.1
for r = 5.010.0.
Clearly the onset of the exponential behaviour in rapidity proportional to exp(BD s Y ),
visible in Fig. 9, signals violation of the Froissart bound. This behaviour is a consequence
of the power tails in impact parameter N b , in the same way as the steep exponential
profile exp(b2 ) leads to the linear dependence Y of the cross section, compare
the argument of Heisenberg in [38]. The exponential increase of the cross section with
rapidity has been also observed in the Monte Carlo study of the amplitude for onium
onium scattering [23], despite the presence of multiple interactions.
Since the power-like tail is generated by the long-range contribution, the violation of the
Froissart bound is caused by the long-range Coulomb-like interactions [19], which in the
reality should be suppressed due to confinement. Thus a modification of the BK equation
by confinement effects is necessary.
6. Conclusions
The solution to the BK equation with the b-dependence presented in this paper differs
substantially from the one with translational invariance. The most important result is
the fact that the exponential profile in b in the initial condition is not preserved by the
evolution. Instead, the power behaviour is generated for large bs whose origin comes from
the structure of the BFKL kernel. Such behaviour leads to the violation of the Froissart
unitarity bound despite the presence of unitarity at fixed impact parameter. The violation of
this bound is caused by nonsuppressed long range contribution, i.e., the lack of confinement
in the BK equation. This feature of the b-dependent amplitude is consistent with numerical
6 In our case r is some average size which is given by the input. Nevertheless all the statements here are
0
valid. In particular we have checked that in fact NBK (r/b2 , Y ) provided r/b2 1.
361
studies of oniumonium scattering with multiple scattering of dipoles [23], and with the
qualitative analysis of the BK equation [19].
Thus in order to satisfy the Froissart bound through the evolution one would have to
modify the evolution kernel in the region of long range contribution in order to incorporate
confinement. In this case the conformal symmetry of the BK equation has to be broken and
additional scale has to appear. This aspect will be studied in our future analysis [31]. For
example, it is evident from Fig. 7, that the naive cut-offdropping the second term in the
square brackets in Eq. (23)would do the job. More physically, one could supplement the
kernel by the exponential decrease at large distances.
Another interesting feature is the dependence of the solution on the dipole size, which
shows that the amplitude is saturated in the limited range of scales: 1/Qs (b, Y ) < r <
RH (b, Y ). Of course, the true behaviour for large dipole sizes should also be modified by
the long-distance, confinement physics.
The studied solution was obtained for one particular initial condition, it would be
interesting to study also the behaviour of the amplitude for other forms of the input. In
particular, one should look at the solution starting from one dipole, i.e., the input localised
both in r and b. We leave this problem for future investigation.
Acknowledgements
We would like to thank Jochen Bartels, Edmond Iancu, Alex Kovner, Jan Kwiecinski,
Misha Lublinsky, Leszek Motyka, Al Mueller, Misha Ryskin, Gavin Salam and Urs Wiedemann for interesting discussions. K.G.-B. is grateful to Deutsche Forschungsgemeinschaft
for a fellowship. This research was supported by the Polish Committee for Scientific Research grants Nos. 2P03B 051 19, 5P03B 144 20.
References
[1] ZEUS Collaboration, M. Derrick, et al., Phys. Lett. B 316 (1993) 412;
ZEUS Collaboration, M. Derrick, et al., Z. Phys. C 65 (1995) 379;
ZEUS Collaboration, M. Derrick, et al., Z. Phys. C 72 (1996) 399;
ZEUS Collaboration, M. Derrick, et al., Eur. Phys. J. C 21 (2001) 443;
H1 Collaboration, I. Abt, et al., Nucl. Phys. B 407 (1993) 515;
H1 Collaboration, I. Abt, et al., Nucl. Phys. B 470 (1996) 3;
H1 Collaboration, C. Adloff, et al., Nucl. Phys. B 497 (1997) 3;
H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 13 (2000) 609;
H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 21 (2001) 33.
[2] L.N. Lipatov, Sov. J. Nucl. Phys. 23 (1976) 338;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 (1977) 199;
I.I. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 338.
[3] L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 (1983) 1.
[4] L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233;
L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 3352;
L. McLerran, R. Venugopalan, Phys. Rev. D 50 (1994) 2225;
J. Jalilian-Marian, A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 55 (1997) 5414;
R. Venugopalan, Acta Phys. Pol. B 30 (1999) 3731;
362
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 692 (2001) 583;

E. Ferreiro, E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 701 (2002) 489;
E. Iancu, R. Venugopalan, hep-ph/0303204.
J. Jalilian-Marian, A. Kovner, A. Leonidov, H. Weigert, Nucl. Phys. B 504 (1997) 415;
J. Jalilian-Marian, A. Kovner, A. Leonidov, H. Weigert, Phys. Rev. D 59 (1999) 014014;
J. Jalilian-Marian, A. Kovner, H. Weigert, Phys. Rev. D 59 (1999) 014015.
A.H. Mueller, Phys. Lett. B 523 (2001) 243.
I.I. Balitsky, Nucl. Phys. B 463 (1996) 99;
I.I. Balitsky, Phys. Rev. Lett. 81 (1998) 2024;
I.I. Balitsky, Phys. Rev. D 60 (1999) 014020;
I.I. Balitsky, Phys. Lett. B 518 (2001) 235;
I.I. Balitsky, A.V. Belitsky, Nucl. Phys. B 629 (2002) 290.
H. Weigert, Nucl. Phys. A 703 (2002) 823.
J. Bartels, M. Wsthoff, Z. Phys. C 66 (1995) 157.
Yu.V. Kovchegov, Phys. Rev. D 60 (1999) 034008.
A.H. Mueller, Nucl. Phys. B 415 (1994) 373;
A.H. Mueller, B. Patel, Nucl. Phys. B 425 (1994) 471.
Yu.V. Kovchegov, Phys. Rev. D 61 (2000) 074018.
E.M. Levin, K. Tuchin, Nucl. Phys. B 573 (2000) 833;
E.M. Levin, K. Tuchin, Nucl. Phys. A 691 (2001) 779;
E.M. Levin, K. Tuchin, Nucl. Phys. A 693 (2001) 787.
M.A. Braun, Eur. Phys. J. C 16 (2000) 337, hep-ph/0010041;
N. Armesto, M.A. Braun, Eur. Phys. J. C 20 (2001) 517.
E. Gotsman, E.M. Levin, M. Lublinsky, U. Maor, Nucl. Phys. A 696 (2001) 851;
M. Lublinsky, Eur. Phys. J. C 21 (2001) 513.
K. Golec-Biernat, L. Motyka, A.M. Stasto, Phys. Rev. D 65 (2002) 074037.
S. Bondarenko, M. Kozlov, E. Levin, hep-th/0305150.
M. Froissart, Phys. Rev. 123 (1961) 1053;
A. Martin, Phys. Rev. 129 (1963) 1432.
A. Kovner, U.A. Wiedemann, Phys. Rev. D 66 (2002) 051502;
A. Kovner, U.A. Wiedemann, Phys. Rev. D 66 (2002) 034031;
A. Kovner, U.A. Wiedemann, Phys. Lett. B 551 (2003) 311.
E. Ferreiro, E. Iancu, K. Itakura, L. McLerran, Nucl. Phys. A 710 (2002) 373.
N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 49 (1991) 607;
N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 53 (1992) 331.
L.N. Lipatov, Sov. Phys. JETP 63 (1986) 904.
G.P. Salam, Nucl. Phys. B 461 (1996) 512;
G.P. Salam, Nucl. Phys. B 449 (1995) 589.
G.P. Salam, Comput. Phys. Commun. 105 (1997) 62.
J. Bartels, E. Levin, Nucl. Phys. B 387 (1992) 617.
A.M. Stasto, K. Golec-Biernat, J. Kwiecinski, Phys. Rev. Lett. 86 (2001) 596.
K. Golec-Biernat, M. Wsthoff, Phys. Rev. D 59 (1999) 014017;
K. Golec-Biernat, M. Wsthoff, Phys. Rev. D 60 (1999) 114023;
K. Golec-Biernat, M. Wsthoff, Eur. Phys. J. C 20 (2001) 313.
G. Bottazzi, G. Marchesini, G.P. Salam, M. Scorletti, Nucl. Phys. B 505 (1997) 366.
R.J. Glauber, Phys. Rev. 100 (1955) 242.
A.H. Mueller, Nucl. Phys. B 335 (1990) 115.
K. Golec-Biernat, A.M. Stasto, in preparation.
E. Iancu, K. Itakura, L. McLerran, Nucl. Phys. A 708 (2002) 327.
A.H. Mueller, D.N. Triantafyllopoulos, Nucl. Phys. B 640 (2002) 331.
V.S. Fadin, M.I. Kotsky, R. Fiore, Phys. Lett. B 359 (1995) 181;
V.S. Fadin, M.I. Kotsky, L.N. Lipatov, BUDKERINP-96-92, hep-ph/9704267;
V.S. Fadin, R. Fiore, A. Flachi, M.I. Kotsky, Phys. Lett. B 422 (1998) 287;
V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 127;
[35]
[36]
[37]
[38]
G. Camici, M. Ciafaloni, Phys. Lett. B 386 (1996) 341;

G. Camici, M. Ciafaloni, Phys. Lett. B 412 (1997) 396;
G. Camici, M. Ciafaloni, Phys. Lett. B 417 (1997) 390, Erratum;
G. Camici, M. Ciafaloni, Phys. Lett. B 430 (1998) 349.
S. Munier, S. Wallon, hep-ph/0303211.
M.G. Ryskin, Talk on DIS2003 conference;
J. Bartels, private communication.
A.H. Mueller, Nucl. Phys. B 437 (1995) 107.
W. Heisenberg, Z. Phys. 133 (1952) 65.
363

Tau polarization in tau-neutrino nucleon scattering

K. Hagiwara a , K. Mawatari b , H. Yokoya c,d
a Theory Group, KEK, Tsukuba 305-0801, Japan
b Graduate School of Science and Technology, Kobe University, Nada, Kobe 657-8501, Japan
c Department of Physics, Hiroshima University, Higashi-Hiroshima 739-8526, Japan
d Radiation Laboratory, RIKEN, Wako 351-0198, Japan
Abstract
We investigate the spin polarization of leptons produced in and nucleon scattering
via charged currents. Quasi-elastic scattering, resonance production and deep inelastic scattering
processes are studied. The polarization information is essential for measuring the appearance rate
in long baseline neutrino oscillation experiments, because the decay particle distributions depend
crucially on the spin. In this article, we calculate the spin density matrix of each process and
estimate the spin polarization vector in medium and high neutrino energy interactions. We find that
the produced s have high degree of polarization, and their spin direction depends non-trivially on
the energy and the scattering angle of in the laboratory frame.
PACS: 14.60.Fg; 13.88.+e; 13.15.+g
1. Introduction
Recent studies from neutrino oscillation experiments are revealing the amazing nature
of the neutrino sector, with their non-zero masses and large flavor mixings. Especially,
reports from Super-Kamiokande (SK) Collaboration [1] strongly suggest that nearly
maximal oscillation from into is occurring in the atmospheric neutrino flux.
To demonstrate this oscillation, it is important to detect appearance in oscillation
experiments. Several long-baseline neutrino oscillation experiments, such as ICARUS [2],
MINOS [3], OPERA [4] are proposed, and they are expected to detect the appearance by
E-mail addresses: mawatari@radix.h.kobe-u.ac.jp (K. Mawatari), yokoya@theo.phys.sci.hiroshima-u.ac.jp
(H. Yokoya).
doi:10.1016/S0550-3213(03)00575-3
K. Hagiwara et al. / Nuclear Physics B 668 (2003) 364384
charged current (CC) reactions off a nucleon target

( ) + N + + X
365
(1)
with N = p, n. Because production by a nucleon target has a threshold for neutrino

energy at E 3.5 GeV, these experiments should provide high energy neutrino flux. It
has also been pointed out by Hall and Murayama [5], that SK may be able to detect the
appearance events with more than several years of running.
The produced decays into several particles, always including a neutrino ( ).
Therefore the appearance signal should be obtained from decay particle distributions.
Because the decay distributions depend significantly on it is spin polarization [6], the
polarization information is essential for us to identify the production signal.
polarization should also be studied in order to estimate background events for
the e appearance reaction, which will be searched for in neutrino oscillation
experiments, such as those using high intensity neutrino beams from J-PARC [7]. Because
the oscillation amplitude of is larger than that of e [8], and because the
branching ratio of e + X is relatively large, the e production via the
e chain can be significant [9]. Since the e energy and angular distribution depends
crucially on the polarization, its information is necessary to estimate the background.
So far, several authors have calculated the production cross section for nucleon targets
[5,10,11], but to our knowledge, no estimation of the polarization of produced s is
available. In this paper, we study the spin polarization of produced by scattering off
a nucleon target. We consider the quasi-elastic scattering (QE), resonance production
(RES), and deep inelastic scattering (DIS) processes, which are known to give dominant
contributions in the medium and high neutrino energy region [10]. The spin polarization
vector is obtained from the spin density matrix which is calculated for each process.
The article is organized as follows. We give the general kinematics of production
in neutrinonucleon interaction and the relation between the spin density matrix and the
spin polarization vector in Section 2. Then we present the details of the spin density
matrix calculation of QE, RES, and DIS processes, in Sections 3, 4, and 5, respectively.
In Section 6, the differential cross section and the spin polarization vector of produced
are estimated for medium and high neutrino energies. Section 7 gives discussions and our
conclusions.
2. Kinematics and formalism

In this section, we show the physical regions of kinematical variables and give the
relation of the spin polarization vector and the spin density matrix of the charged current
(CC) production process. Firstly, we define the four-momenta of incoming neutrino (k),
target nucleon (p) and produced lepton (k ) in the laboratory frame
k = (E , 0, 0, E ),
(2)
p = (M, 0, 0, 0),
(3)
= (E , p sin , 0, p cos ).
(4)
366
Here, E and E are the incoming neutrino and outgoing

energies, respectively, in the
laboratory frame, M is the nucleon mass, and p = E2 m2 with the lepton mass
m = 1.78 GeV. We also define some Lorentz invariant variables
q = k k ,
Q2 = q 2,
W = (p + q) .
2
(5)
(6)
Q2 is the magnitude of the momentum transfer and W is the hadronic invariant mass. The
physical regions of these variables are given by
M W s m ,
(7)
and
Q2 (W ) Q2 Q2+ (W ),
(8)
where s = (k + p)2 and

Q2 (W ) =

2

s M2
1 W 2 + m2 M W 2 m2
(1 )
2
2
s
(9)
with = 1/2 (1, m2 /s, W 2 /s) and (a, b, c) = a 2 + b2 + c2 2(ab + bc + ca).

The scaling variables are defined as usual:
Q2
Q2
= 2
,
2p q
W + Q2 M 2
W 2 + Q2 M 2
pq
E
=
y=
=1
.
2
pk
E
s M
x=
(10)
(11)
Here, x is the Bjorken variable and y is the inelasticy. The physical regions for x and y are
obtained by Albright and Jarlskog [11,12]:
m2
x1
2M(E m )
(12)
A B y A + B,
(13)
and
where

m2
1
xM
m2
A=
1+
,
1
2
2ME x 2E2
2E

2

xM
m2
1
m2 1/2
1+
1
.
B=
2
2
2ME x
E
2E
(14)
(15)
The above regions agree with those determined by Eqs. (7) and (8).
We label the relevant subprocesses by using the hadronic invariant mass W and the
momentum transfer Q2 . We label QE (quasi-elastic scattering) when W = M, RES
(resonance production)
when M + m < W < Wcut , and IS (inelastic scattering) when
Wcut < W < s m . Wcut is an artificial boundary between RES and IS processes, to
367
Fig. 1. Physical region at E = 10 GeV in the xy plane (left), and in the p cos p sin plane (right). Open
circles denote QE (quasi-elastic scattering), open triangles denote RES ( resonance production), and the cross
symbols give the DIS (deep inelastic scattering) region with Q2 1 GeV2 . The region marked by the star symbol
() gives inelastic scattering (IS), at W 1.4 GeV and Q2 < 1 GeV2 .
avoid double counting. The Wcut value is taken in the region 1.41.6 GeV. Within the IS
region, the region where Q2 1 GeV2 may be labeled as DIS, where the use of the parton
model can be justified.
Fig. 1 shows the kinematical regions of each QE, RES and IS process on the xy
plane (left) and the p cos p sin plane (right) at E = 10 GeV. The QE region is
shown by open circles, the RES region by open triangles, and the DIS region is shown
by the cross symbols. The region shown by the star symbol () gives the IS process at
low Q2 (Q2 < 1 GeV2 ). In this region the parton model is not reliable and we must use
the experimental data to reduce errors. In this report, however, we use the parton model
throughout the IS region. Studies on uncertainties in this region will be reported elsewhere.
Produced will be partially polarized. We define the spin polarization vector,
parameterized as
P
(sin P cos P , sin P sin P , cos P ),
(16)
2
in the rest frame in which the z-axis is taken along its momentum direction in the
laboratory frame. In Eq. (16), P and P are the polar and azimuthal angle of the spin
vector in the rest frame, respectively, and P denotes the degree of polarization. P = 1
gives the fully polarized , and P = 0 gives unpolarized . The azimuthal angle is
measured from the scattering plane where P = /2 is along the p p direction in
the laboratory frame. The degree of polarization (P ) and the spin directions (P , P ) are
functions of E and cos . This spin polarization vector is related with the spin density
matrix R , by the following relation:

1 1 + P cos P P sin P eiP
dsum
dR
=
.
(17)
i
P
1 P cos P
dE d cos
2 P sin P e
dE d cos

The density matrix is calculated as R M M , where M is the helicity amplitude
with the helicity /2 defined in the laboratory frame, and dsum = dR++ + dR is the
s = (sx , sy , sz ) =
368
usual spin summed cross section. The summation symbol implies the summation over final
states, and the spins of the target and final-state particles.
The spin density matrix of production is obtained by using the leptonic and hadronic
tensor as
G2 2 p
dR
= F
L W ,
dE d cos
4 ME
(18)
2 /(Q2 + M 2 ) is the propagator factor with the

where GF is Fermi constant and = MW
W
W -boson mass MW = 80.4 GeV. For production, the leptonic tensor L is expressed
as
L = j j ,
where the leptonic weak current
(19)
is
1 5
j = u (k , )
u (k)
2

( = +),
2E (E p ) sin 2 , cos 2 , i cos 2 , sin 2
=

( = ),
2E (E + p ) cos 2 , sin 2 , i sin 2 , cos 2
(20)
in the laboratory frame. For + production, we must replace the leptonic tensor L into
L defined as
L = j j ,
where
(21)
is
1 5
v (k , )
j = v (k)
2

2E (E + p ) cos 2 , sin 2 , i sin 2 , cos 2
=

2E (E p ) sin 2 , cos 2 , i cos 2 , sin 2
( = +),
( = ),
(22)
which is related with j by j = j , in the phase convention of Ref. [13]. In the

following sections, we will abbreviate the overline of the leptonic tensor and currents for
+ production process.
The hadronic tensor is expressed in general as
W (p, q)

p p

p q
2
p
q,
Q
i2
= g W1 p q, Q2 +
W
W3 p q, Q2
2
2
2
M
2M

p q + q p

q q
2
2
W4 p q, Q +
W5 p q, Q ,
+
(23)
M2
2M 2
where the totally antisymmetric tensor 2 is defined as 20123 = 1, and the structure
functions Wi=1,...,5 (p q, Q2 ) can be estimated for each subprocess. Since q j is
proportional to m , the structure functions W4 and W5 appear only in the heavy lepton
production case [12].
369
Inserting these equations into Eqs. (18) and (17), we find

dsum
dE d cos

G2 2 p
m2
= F
2W1 + 2 W4 (E p cos ) + W2 (E + p cos )
2 M
M

m2
W3
E E + p2 (E + E )p cos W5
M
M
G2F 2 p
F,
2 M
and the spin polarization vector takes

m sin
E
m2
E
F,
W3 2 W4 +
W5
2W1 W2
sx =
2
M
M
M
sy = 0,

1
m2
2W1 2 W4 (p E cos ) + W2 (p + E cos )
sz =
2
M

m2

W3
F,
(E + E )p E E + p2 cos W5 cos
M
M
for productions. The degree of polarization is given by

P = 2 sx2 + sy2 + sz2 .
(24)
(25a)
(25b)
(25c)
(26)
The above results Eqs. (24)(26) agree with Ref. [12]. From the above equations we find
(i) P takes either 0 or for any momentum, which means that the polarization vector
lies always in the scattering plane, and (ii) if m = 0, then s could take only (0, 0, 12 ),
which means fully left-handed or right-handed + .
3. Quasi-elastic scattering
In this section, we give the spin density matrix calculation for the QE scattering
processes
+ n + p,
+
+ p + n.
(27)
(28)
Following Llewellyn Smith [14], the hadronic tensor is written by using the weak transition
current J() as follows:
QE
W
=

cos2 c () () 2
J J W M 2 ,
4
spins
(29)
370

(+)
()
where c is the Cabbibo angle. The weak transition currents J and J for the and
scattering, respectively, are defined as

J(+) = p(p )J(+) n(p) = u p (p ) un (p),
(30)

J() = n(p )J() p(p) = u n (p ) up (p) = p(p)J(+) n(p ) ,

(31)
V
where is written in terms of the six weak form factors of the nucleon, F1,2,3
, FA , F3A
and Fp , as
i q V 2 q V 2
F2 q +
F q
= F1V q 2 +
2M
M 3

(p + p ) A 2 q 2
+ FA q 2 +
F3 q +
Fp q 5 .
M
M
(32)
For the scattering, the vertex is obtained by = 0 0 . We can drop two

form factors, F3V and F3A , because of time reversal invariance and isospin symmetry (or
equivalently no second-class currents). Moreover, the vector form factor F1V and F2V are
related to the electromagnetic form factors of nucleons under the conserved vector current
(CVC) hypothesis:

2
2
q 2 V 2
q 2 1
V
V
F1 q = GE q
(33)
G q
,
1
4M 2 M
4M 2

q 2 1
F2V q 2 = GVM q 2 GVE q 2
(34)
,
1
4M 2
where

GVE q 2 =
1
,
(1 q 2 /MV2 )2

GVM q 2 =
1+
,
(1 q 2 /MV2 )2
(35)
with a vector mass MV = 0.84 GeV and = p n = 3.706. p and n are the
anomalous magnetic moments of proton and neutron, respectively. For the axial vector
form factor FA , we adopt the following parameterization:

FA q 2 =
FA (0)
(1 q 2 /MA2 )2
(36)
with an axial-vector mass MA = 1.0 GeV and FA (0) = 1.23 [14]. For the pseudo-scalar
form factor Fp , we adopt the parametrization of Ref. [14]

FA (q 2 )
Fp q 2 = 2M 2 2
m q 2
(37)
with the pion mass m = 0.14 GeV. The normalization of Fp (0) is fixed by the partially
conserved axial vector current (PCAC) hypothesis. It should be stressed here that, the
form factor Fp (q 2 ) has not been measured experimentally because its contribution is
proportional to the lepton mass. The production cross section and the polarization of
are sensitive to Fp (q 2 ) because of the large mass and the spin-flip nature of the form
factor.
371
Fig. 2. The neutrino energy dependence of the total cross sections of the QE (quasi-elastic) processes, l n l p
(solid lines) and l p l + n (dashed lines). The thick lines are for l = , while the thin lines are for l = .
In Fig. 2, we show the total cross sections of the QE process versus the incoming
neutrino energy. We plot not only the -neutrino interaction process, but also the neutrino interaction process for comparison. Solid curves are the and scattering cross
sections and dashed curves are the and scattering cross sections. Our results agree
well with those of Hall and Murayama [5].
4. Resonance production
In this section, we present the spin density matrix calculation for the production
processes

+ n(p) + + ++ ,
(38)

+ p(n) + + 0 .
(39)
We neglect N and the other higher resonance states, which are known to give small
contributions [10,15,16]. For the resonance production, we calculate the hadronic tensor
by using the nucleon- weak transition current J as follows:
RES
W
=
cos2 c
1
W (W )
.
J J
2 )2 + W 2 2 (W )
4
(W 2 M
spins
(40)
Here, M is the resonance mass, M = 1.232 GeV, and (W ) is its running width
estimated by assuming the dominance of S-wave N + decay:
(W ) = (M )
1/2 (W 2 , M 2 , m )
2 , M 2, m )
1/2 (M
with (M ) = 0.12 GeV and (a, b, c) = a 2 + b2 + c2 2(ab + bc + ca).
(41)
372
The current J for the process + n + + is defined by

J = + (p )J n(p) = (p ) un (p),
(42)
where is the spin-3/2 particle wave function and the vertex is expressed in terms
V ,A
of the eight weak form factors Ci=3,4,5,6
[14,1719] as
C3V (q 2 )
C V (q 2 )
(g q/ q ) + 4 2 (g p q p q )
M
M

V 2
C6V (q 2 )
C5 (q )
(g p q p q ) +
q q 5
+
M2
M2
C3A (q 2 )
C A (q 2 )
(g q/ q ) + 4 2 (g p q p q )
M
M
A (q 2 )

C
+ C5A q 2 g + 6 2 q q .
(43)
M
By using the isospin invariance and the WignerEckart theorem, we obtain the other
nucleon weak transition currents as
++ + 0
J p = 3 J n = 3 J p = J n .
(44)
+
V
From the CVC hypothesis, C6V = 0 and the other vector form factors Ci=3,4,5
are related
to the electromagnetic form factors. We adopt the following parametrizations:

C3V q 2 =
C3V (0)
(1 q 2 /MV2 )2

C5V q 2 = 0,

M V 2
C4V q 2 =
C q ,
M 3
(45)
with C3V (0) = 2.05 and a vector mass MV2 = 0.54 GeV2 . For the axial vector form factors
A
Ci=3,4,5
, we use the modified dipole form factors [17,19]
A
Ci=3,4,5

q 2 2
ai q 2
1 2
= Ci (0) 1
bi q 2
MA
(46)
with C3 (0) = 0, C4 (0) = 0.3, C5 (0) = 1.2, a4 = a5 = 1.21, b4 = b5 = 2.0 GeV2 and
MA = 1.0 GeV. And for C6A , we use the following relation [20]:

C6A q 2 = C5A q 2
M2
,
m2 q 2
(47)
which agrees with the off-diagonal GoldbergerTreiman relation in the limit of m2 0

and q 2 0 [17]. The pseudo-scalar form factor C6A (q 2 ) has not been measured because
its contribution vanishes for massless leptons. As in the case of the Fp (q 2 ) form factor of
the QE process, C6A (q 2 ) has significant effects on the production cross section and the
polarization.
373
Fig. 3. Total cross sections of the production (RES) processes, plotted against the incoming (anti)neutrino
energy. The solid, long-dashed, dashed, and dot-dashed lines show ++ , + , 0 , and production cross
sections, respectively. The thick lines are for and scatterings and the thin lines are for and scatterings.
In Eq. (40), summation over the hadronic spins is done by using a spin projection
operator of the spin-3/2 particle wave function which is given by
P =
(p ) (p )
spin

2 p p
1 p p 1
= (/
p + M ) g
.
2
3 M
3
M
3
(48)
The hadronic tensor is now calculated by
RES
W
=
1
cos2 c
W (W )
Tr P (/
.
p + M)
2 )2 + W 2 2 (W )
2
4
(W M
(49)
By integrating over E and cos within the kinematical region of M + m < W <
1.4 GeV, we estimate the total cross section of the production (RES) processes. In Fig. 3,
we show the total cross section vs. the incoming neutrino energy. We also plot the total
production cross sections for and scattering processes, in order to examine the
lepton mass dependence. The production cross sections grow sharply at low E , while
the production cross section grow mildly from around E = 4 GeV. The cross sections
of ++ and production processes are larger than those of + and 0 productions.
This feature is expected from the ClebshGordan coefficients of the transition currents, in
Eq. (44). Our results agree approximately with those of Paschos and Yu [10], which include
the contributions from N (S11 , P11 ) resonance productions.
374
5. Deep inelastic scattering

In this section, we present the spin density matrix calculation for the deep inelastic
scattering (DIS) processes
+ N + X,
(50)
+ N + X.
(51)
In the DIS region, the hadronic tensor is estimated by using the quarkparton model:
d

(q,q)
DIS
fq,q , Q2 K
W (p, q) =
(52)
(pq , q).
q,q
Here, pq
= p is the four-momentum of the scattering quark, is its momentum fraction,

and fq and fq are the parton distribution function (PDF)s inside a nucleon. By taking the
spin average of initial quark and by summing over the final quark spins, we find the quark
tensor

(q,q)
K
(pq , q) = 2pq q Q2 m2q

2 g (pq q) + 2pq pq

i2 pq q + (pq q + q pq ) .
(53)
The upper sign should be taken for quarks and the lower for antiquarks. We retain the final
quark mass, mq , for the charm quark as mc = 1.5 GeV, but otherwise we set mq = 0. We
neglect charm and heavier-quark distributions in the nucleon, as well as bottom and top
production cross sections.
By neglecting the nucleon mass and the initial quark masses consistently, we find the
following relations:

W1 p q, Q2 = F1 x, Q2 ,

M2
Fi=2,...,5 x, Q2 .
Wi=2,...,5 p q, Q2 =
pq
(54)
Here,
F1 =

fq,q , Q2 ,
q,q
F2 = 2

q,q
F3 = 2

fq,q , Q2 ,
(55b)

fq , Q2 2
fq , Q2 ,
(55c)
F4 = 0,

F5 = 2
fq,q , Q2 ,
q,q
(55a)
(55d)
(55e)
375
Fig. 4. Total cross sections of the DIS processes divided by the neutrino energy are plotted against the neutrino
energy. Solid, long-dashed, dashed, and dot-dashed lines show l p l X, l n l X, l p l + X, and
l n l + X processes, respectively. The thick lines are for l = and the thin lines are for l = .
where the momentum fraction is = x for massless final quarks (mq = 0), and = x/
with = Q2 /(Q2 + m2q ) for q = c. In the mc 0 limit, the CallanGross relation
F2 = 2xF1 and the AlbrightJarlskog relations F4 = 0, 2xF5 = F2 hold.
However, the differential cross section (Eq. (24)) does not satisfy the positivity condition
near the threshold with this naive replacement. We find that the following modification of
the W1 structure function suffices to ensure the positivity constraints:1

M2
F1 ,
W1 = 1 +
(56)
pq
and find that the positivity is maintained when the charm quark mass is introduced by using
the rescaling variable = x/.
There is further uncertainty in our parton model predictions for the inelastic scattering
processes where the hadronic final state is heavy, W 1.4 GeV, but the momentum transfer
is small, Q2 1 GeV2 . This is the region of the phase-space depicted by the star symbol
() in the xy plane and the p cos p sin plane of Fig. 1. In order to estimate the
cross section and the spin polarization vector in this region, we use naive extrapolation
of the parton model calculation, by using the parton distribution at the minimum Q2
(Q20 = 1.25 GeV2 for the parametrization of Martin et al. [21]) even when Q2 < Q20 .
In Fig. 4, we plot the total cross sections of the DIS (IS) process for N and N
scatterings by thick lines. Those of the N and N scattering processes are shown
by thin lines for comparison. Those curves are obtained by using the parton distribution
function (PDF)s of Martin et al. [21]. The results are similar to the RES case, production
cross sections grow rapidly from low E , and the production cross sections grow mildly
from around E = 5 GeV. These results are consistent with the calculations of Kretzer and
Reno [11], which include the NLO corrections.
1 Slightly more complicated rescaling low has been examined by Albright and Jarlskog [12].
376
Fig. 5. The neutrino energy dependence of the total cross section of (left) and + (right) productions off the
isoscalar target, normalized by incoming neutrino energy. The contributions from QE, RES and DIS processes
are shown by dot-dashed, dashed and long-dashed lines, respectively, and their sums are shown by thick solid
lines.
Uncertainties in the total cross section due to the modification of the structure function
W1 (Eq. (56)) and in the contribution from the Q2 < 1 GeV2 region are found to be rather
small. A more quantitative study of the uncertainty in the theoretical predictions will be
reported elsewhere.
In Fig. 5, we show the total cross section of all the production process for the
isoscalar target. The cross sections normalized to the neutrino energy are plotted against
the neutrino energy. The left figure is for production and the right figure is for +
production. We find that at medium neutrino energies, the QE contribution dominates
the total cross section near the threshold, and the sum of the QE and RES cross sections
are significant throughout the energy range of the future neutrino oscillation experiments.
Significance of the QE and RES contribution is more pronounced for the N + X
reaction shown in the right-hand figure, where the DIS contribution starts dominating the
total cross section only above E = 10 GeV. Those trends agree with the earlier results of
Paschos and Yu [10].
6. Polarization of the produced
In this section, we show the spin polarization vector of the produced lepton as a
function of its energy E and the scattering angle in the laboratory frame. We show
our results for two arbitrarily fixed neutrino and antineutrino energies, E = 10 GeV and
20 GeV, for isoscalar targets.
Fig. 6 summarizes our results for the N X process at E = 10 GeV. The top
three figures show the double differential cross section, Eq. (24), as a function of E , at
= 0 (left figures), 5 (center figures) and 10 (right figures). The DIS (IS) contributions
are shown by dashed lines, and the RES and QE contributions are shown by the solid
lines. The area of the histogram for the QE process is normalized to the cross section. A
set of three middle figures give the degree of polarization, P of Eq. (16), as functions of
377
Fig. 6. Production cross section and the polarization of the process N X at E = 10 GeV. E
dependence of the differential cross section (top), the degree of polarization P (middle) and the polar component
of the normalized polarization vector cos P (bottom) are shown along the laboratory frame scattering angle
= 0 (left), 5 (center) and 10 (right), respectively. The Histograms in the top figures and the circles in the
middle and the bottom figures represent QE process, solid lines show RES process, and the dashed lines are for
DIS process. The spin in the rest frame is s = P2 ( sin P , 0, cos P ).
E . In the bottom three figures, we show the E dependence of the polarization direction,
Eq. (16), by using cos P = sz /(P /2). This suffices to determine the polarization direction
because sx turns out to be always negative, sx = P2 sin P , and sy = 0 (P = ). It should
be noted that all the 9 figures have common horizontal scale. The overall phase-space of
the E = 10 GeV experiment in the laboratory frame has been shown in Fig. 1 (right).
The differential cross sections are obtained from Eq. (24). According to the phase-space
plot of Fig. 1 (right), along a fixed laboratory scattering angle , there are two E s at
which QE and RES reactions can take place. The top figures of Fig. 6 show us that the
cross sections in the lower E sides are negligible. This is because of the form factor
suppression which is significant already at E = 10 GeV. The QE and RES cross sections
are large at forward scattering angles, and the DIS contribution become more significant
at large scattering angles, though the cross section gets smaller. In order to examine the
transition between the resonance production (RES) process and the DIS process, we
show our predictions for RES up to W < 1.6 GeV and those for DIS from W > 1.4 GeV,
allowing for the overlap. Although there is no strong reason to expect smooth transition,
we find the tendency that our predictions for the production are relatively smoothly
changing in the transition region.
The degree of polarization P and the polar angle P are defined in Eq. (16). The
produced is almost fully polarized except at the very small scattering angle. As for the
angle of the polarization vector, the high energy is almost left-handed (cos P = 1).
378
Fig. 7. The same as Fig. 6, but for the process N + X at E = 10 GeV. The + spin in the + rest frame
is s = P2 (sin P , 0, cos P ).
On the other hand, the spin of low energy turns around. The azimuthal angle P takes
at all energies, which means that the spin vector points to the direction of the initial
neutrino momentum axis.
In order to understand the above features, it is useful to consider the polarization of
in the center of mass (CM) frame of the scattering particles. Let us consider the DIS process
in the q CM frame, since the q scattering is dominant in the N X process. In this
frame, produced is fully left-handed polarized at all scattering angles. This is because
the initial and q (d or s quarks) are both left-handed, and hence angular momentum
along the initial momentum direction is zero, while in the final state, the produced u quark
is left-handed and hence only the left-handed is allowed by the angular momentum
conservation. This selection rule is violated slightly when a charm quark is produced in the
final state and because of gluon radiation at higher orders of QCD perturbation theory. The
polarization in the laboratory frame is then obtained by the Lorentz boost. In the QE
and RES processes, situations are almost the same as in the DIS process. In the CM frame
of N collisions, the lepton produced
by the QE or RES process is almost left-handed
at all angles, for the CM energy of 2ME + M 2 4.4 GeV for E = 10 GeV, for our
parametrizations of the transition form factors. High energy s in the laboratory frame
have left-handed polarization because those s have forward scattering angles also in the
CM frame. However, lower energy s in the laboratory frame tends to have right-handed
polarization because they are produced at backward angles in the CM frame. At the zero
scattering angle = 0 of the laboratory frame, the change in the momentum direction
occures suddenly, and hence the transition from the left-handed at high energies to
the right-handed at low energies is discontinuous. Since the degree of polarization P
vanishes at this point, the polarization vector, or the density matrices are continuous.
379
Fig. 8. The differential cross section of the DIS process at E = 10 GeV and at = 0 in the laboratory frame
for the processes N X (a) and N + X (b), where N is an isoscalar nuclei. The dashed and
dot-dashed lines represent, respectively, the contributions of the neutrinoquark and neutrinoantiquark scattering
subprocesses. The solid lines show their sum.
Fig. 7 shows the N + X case at E = 10 GeV. The predictions of the QE and

RES processes are quite similar to those of the N X process in Fig. 6, except that
the + polarization is almost right-handed. In the DIS process, however, the polarization
vector of + is predicted to be quite different from the case in a non-trivial manner.
This is because the q scattering contribution is not so small as compared to the q
scattering contribution. In case of the + production process, the azimuthal angle of the
+ polarization vector takes P = 0 at all energies, which gives sx = P2 sin P and sy = 0.
Therefore the + spin vector points away from the initial neutrino beam axis, contrary to
the spin case.
In order to understand the difference between the + and spin polarization
predictions in Figs. 6 and 7, we show in Fig. 8 the E dependence of the differential cross
section at = 0 for the N X process (a) and for the N + X process (b).
The contributions from the q or q scattering process are shown by dashed lines, those
from the q or q scattering process are shown by dash-dotted lines, and their sum
by solid lines. It is clear that the q scattering contribution dominates the N X
process, whereas for the N + X process, both q or q scattering contribution are
significant.
Let us consider the q
and q scattering in the parton collision CM frame. As for the
q scattering, the amplitude of right-handed + production is proportional to (1 + cos )
and that of left-handed + production is proportional to sin , where is the scattering
angle in the CM frame. On the other hand, for the q scattering the produced + has fully
right-handed polarization, and the angular distribution is flat in the CM frame. Next, let
us consider the = 0 case in the laboratory frame, which correspond to = 0 or 180
in the CM frame. Because of the (1 + cos )2 and sin 2 distributions in the CM frame,
the + from q scattering has fully right-handed polarization and is produced only in the
forward direction ( = 0 ) in the CM frame. Hence, all + s from q
scattering are righthanded along = 0 in the laboratory frame. On the other hand, the + s from the q
scattering are purely right-handed both at = 0 and 180 in the CM frame, and hence
380
Fig. 9. The same as Fig. 6, but for the process N X at E = 20 GeV.
in the laboratory frame, the high energy + s have right-handed polarization, and the low
energy + s have left-handed polarization. By comparing the cross section of q and q
scattering in Fig. 8(b), we find that high energy + s are mostly produced by q
scattering,
and hence they are almost right-handed. But as the energy decreases the contributions from
q scattering increases and the degree of polarization becomes lower by cancellation. In
the QE and RES process of + production, the mechanism is the same as the production
case but for the sign of the polarization. In the CM frame of N
scattering, + has almost
right-handed polarization for both QE and RES processes at all angles. Therefore in the
laboratory frame, high energy + s are right-handed, while low energy + s are left-handed
because of the helicity flip by boost.
Let us now show our predictions at higher neutrino energies. In Figs. 9 and 10,
we show our predictions for N X and N + X processes, respectively, at
E = 20 GeV. The energy dependence of the differential cross sections shows clearly the
dominance of the DIS contribution at higher energies except at = 0 . Fig. 10 shows
that the QE and RES contributions are more significant in the N scattering, as compared
to the N scattering case shown in Fig. 9. The relative importance of the QE or RES
contribution to the N
scattering persists at high energies as can be seen from Fig. 5.
The degree of the polarization remains high in Fig. 9, except for the special angle of
= 0 , and its polarization direction are essentially understood by the boost effect, as for
the E = 10 GeV case. In Fig. 10, the degree of the + polarization decreases at lower +
energy in the laboratory frame for the N + X process. This is understood as a result
of the cancellation between the q
and q scattering contributions as in the E = 10 GeV
case.
It is notable that the produced has almost 100% polarization (P 1) at all energies
except at around = 0 while its polarization direction deviates from the pure left-handed
381
Fig. 10. The same as Fig. 7, but for the process N + X at E = 20 GeV.
direction (cos P = 1) even at relatively high energies in the laboratory frame. On the
other hand, the + polarization deviates from 100% at relatively high E while its direction
is along the right-handed direction (cos P = +1) down to half the maximum energy. Those
qualitative difference between the polarization in the N X process and the +
polarization in the N + X process is understood as a consequence of the significance
of the antiquark contribution to the DIS process in N
scattering.
7. Discussion and conclusion

The information on the polarization of produced through the N and N
scattering is essential to identify the production signal since the decay particle
distributions depend crucially on the spin. It is needed in long baseline neutrino
oscillation experiments which should verify the large oscillation, and is also
needed for the background estimation of e appearance experiments which should
measure the small mixing angle of e oscillation.
In this paper we studied in detail the spin polarization of produced in and
nucleon scattering via charged currents. Quasi-elastic scattering (QE), resonance
production (RES) and deep inelastic scattering (DIS) processes have been studied. The
three subprocesses are distinguished by the hadronic invariant mass W . W = M(= mN )
gives QE, M + m < W < Wcut gives RES and W > Wcut gives DIS. In this article, we set
the kinematical boundary of RES and DIS process at Wcut = 1.4 GeV.
The spin density matrix of production has been defined and the spin polarization
vector has been defined and parametrized in the rest frame whose polar-axis is taken
along the momentum direction of in the laboratory frame. The spin density matrix
382
Fig. 11. The contour map of the DIS cross section in the plane of p cos and p sin for the N X
process at E = 10 GeV in the laboratory frame. The kinematical boundary is shown by the thick grey curve. The
QE process contributes along the boundary, and the RES process contributes just inside of the boundary. The
polarization are shown by the arrows. The length of the arrows give the degree of polarization, and the direction
of arrows give that of the spin in the rest frame. The size of the 100% polarization (P = 1) arrow is shown
as a reference. The arrows are shown along the laboratory scattering angles, = 0 , 2.5 , 5 , 7.5 , and 10 , as
well as along the kinematical boundary.
has been calculated for each subprocess by using the form factors for the QE and RES
processes, and by using the parton distribution functions of Ref. [21] for the DIS process.
We have shown the spin polarizations of as function of the energy and the scattering
angle in the laboratory frame for N X and N + X processes at E = 10 GeV
and 20 GeV. We find that the produced have high degree of polarization, but their spin
directions deviate significantly from the massless limit predictions at low and moderate
energies. Qualitative feature of the predictions have been understood by considering the
helicity amplitudes in the CM frame of the scattering particles and the effects of Lorentz
boost from the CM frame to the laboratory frame.
Finally, we summarize our findings in Figs. 11 and 12. In Fig. 11, we show the
polarization vector s of for the N X process at E = 10 GeV on the p cos
p sin plane, where p and are the produced momentum and the scattering angle in
the laboratory frame. The length of each arrow gives the degree of polarization (0 P 1)
at each phase-space point and its orientation gives the spin direction in the rest frame.
The differential cross section is described as a contour map, where only the DIS cross
section is plotted to avoid too much complexity. The outer line gives the kinematical
boundary, along which the QE process occures. Fig. 11 is a more visual version of the
information given in Fig. 6. Fig. 12 gives the + polarization for the N + X process
at E = 10 GeV, compiling the cross section and the polarization information of Fig. 7.
Before closing our discussion, we point out some uncertainties in our calculation. One is
the uncertainty at small Q2 region (Q2 1 GeV2 ) in our DIS calculation. In this paper, we
used an extrapolation of the parton model calculation in this region by freezing the PDFs
below their validity region. Because the parton model must break down in this region,
and because our estimation of the cross section in this region is not small, a more careful
treatment, e.g. by using the structure function data is needed. Another is the uncertainty
383
Fig. 12. The same as Fig. 11, but for N + X case.
in the pseudo-scalar form factors, Fp (q 2 ) for the QE and C6A (q 2 ) for the RES processes,
which have not been measured in the past. Because of the large mass and because of the
spin-flip nature of those form factors, they can affect the predictions of the polarization
significantly. QCD higher-order corrections should affect the polarization in the DIS
region. We plan to study those uncertainties elsewhere.
We hope that this work will be useful in detecting the appearance signal in long
baseline neutrino oscillation experiments, and that it will also be useful in understanding
the l background for the e appearance experiments.
Acknowledgements
We are grateful to E.A. Paschos for useful comments and discussions. K.M. and H.Y.
thank KEK theory group for the hospitality, where parts of this work were performed.
K.M. would like to thank T. Morii and S. Oyama for discussions. H.Y. would like to thank
M. Hirata, J. Kodaira for discussions, and RIKEN BNL Research Center for the hospitality
where this work was finalized.
References
[1] Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1988) 1562;
Super-Kamiokande Collaboration, Phys. Rev. Lett. 85 (2000) 3999.
[2] ICARUS Collaboration, hep-ex/103008;
See also ICARUS Collaborations home page, http://www.aquila.infn.it/icarus/.
[3] MINOS Collaboration home page, http://www-numi.fnal.gov:8875/.
[4] A. Rubbia, Nucl. Phys. B (Proc. Suppl.) 91 (2000) 223;
See also OPERA Collaboration home page, http://operaweb.web.cern.ch/operaweb/index.shtml.
[5] L.J. Hall, H. Murayama, Phys. Lett. B 463 (1999) 241.
[6] S. Jadach, Z. Was, R. Decker, J.H. Kuhn, Comput. Phys. Commun. 76 (1993) 361.
[7] J-PARC home page, http://j-parc.jp/index.html.
[8] M. Apollonio, et al., Phys. Lett. B 420 (1998) 397;
384
M. Apollonio, et al., Phys. Lett. B 466 (1999) 415.

[9] M. Aoki, et al., hep-ph/0112338, Phys. Rev. D, in press;
M. Aoki, K. Hagiwara, N. Okamura, Phys. Lett. B 554 (2003) 121.
[10] E.A. Paschos, J.Y. Yu, Phys. Rev. D 65 (2002) 033002.
[11] S. Kretzer, M.H. Reno, Phys. Rev. D 66 (2002) 113007.
[12] C.H. Albright, C. Jarlskog, Nucl. Phys. B 84 (1975) 467.
[13] K. Hagiwara, D. Zeppenfeld, Nucl. Phys. B 224 (1986) 1;
H. Murayama, K. Hagiwara, I. Watanabe, HELAS, KEK Report 91-11, 1992.
[14] C.H. Llewellyn Smith, Phys. Rep. 3 (1972) 261.
[15] D. Rein, L.M. Sehgal, Ann. Phys. 133 (1981) 79.
[16] E.A. Paschos, L. Pasquali, J.Y. Yu, Nucl. Phys. B 588 (2000) 263.
[17] P.A. Schreiner, F. Von Hippel, Nucl. Phys. B 58 (1973) 333.
[18] G.L. Fogli, G. Nardulli, Nucl. Phys. B 160 (1979) 116.
[19] L. Alvarez-Ruso, S.K. Singh, M.J. Vicente Vacas, Phys. Rev. C 57 (1998) 2693.
[20] S.K. Singh, Nucl. Phys. B (Proc. Suppl.) 112 (2002) 77.
[21] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 23 (2002) 73.

Correlated decay widths of a pair for the Michel

parameter measurement
K. Tamai
High Energy Accelerator Research Organization, KEK, Ibaraki 305-0801, Japan
Abstract
We express the correlated decay widths of a + pair produced by an e e+ collision in terms
of the observed variables in order to measure the Michel parameters. The decay widths are derived
from theoretical predictions by integrating over the unobserved direction. The introduction of a
coordinate system for the direction makes it easy to formulate the integral. Based on the new
formulation, we extract the Michel parameters from typical leptonic decays, + X and
+ X , which are generated by a Monte Carlo simulation.
PACS: 13.35.Dx; 14.60.Fg
Keywords: Michel parameter; V A current; Jacobian
1. Introduction
Leptonic decays of lepton are especially well suited to investigate structure of charged
weak interactions, which can be determined without any specific model assumptions [1].
The structure is featured by Michel parameters , , and , which govern the energy
spectrum of the charged lepton decayed from the parent lepton. The standard model, V A
weak current, is characterized by = 3/4, = 0, = 1 and = 3/4. One can associate the
deviation from those values with the contribution of other types of currents as a scalar (S),
tensor (T), V + A and so on. Measurements of the muon decay indicated no inconsistency
with the standard model [2].
E-mail address: kunio.tamai@kek.jp (K. Tamai).

doi:10.1016/S0550-3213(03)00576-5
386
K. Tamai / Nuclear Physics B 668 (2003) 385402
An extension from the muon to leptonic decay can be expected naturally based on the
assumption of lepton universality. However, the heaviest lepton mass might accelerate to
induce a new interaction [6] and distort the energy spectrum. The lepton with two decays,
and e e , also provides an opportunity to study the e universality
in the charged leptonic weak interaction. Recent results [1522] of the decay indicate
V A dominance with less precision than the muon. Most of the measurements were
carried out around (4S) and the Z 0 mass by e e+ colliders. One reason of the lower
precision is that the direction cannot be measured due to the short lifetime. The direction
is necessary to transform all of the measured momenta into the rest frame to calculate
the polarization as well as the decayed lepton energy.
Before considering this problem, let us survey the typical decay set of pairs
accompanying single leptonic decay + ( )(h
+ ), and double leptonic decay,
+
+
k ), where (k ) and h represent a charged lepton and a

( )(k
hadron, respectively. h+ may be a resonance as + , a1+ or K + . We designate the combined
decay of pair as correlated decay, since and + decay coherently with each other
through their spins in pair production, e e+ + .
All particles, except neutrinos, are assumed to be observed completely. The measurement restricts the direction kinematically. The correlation between the observed particles
provides us with information on the spin polarization, which enables the measurement
of parameters and [35]. Based on Tsai [3], the formulations are reviewed in Section 2.
Fetscher [7] proposed a method to extract the Michel parameters from the correlations
between and h+ (or k + ), which are integrated over the angles of production and of
the observed particles, and h+ (or k + ). However, the equation cannot be applied to
quasi 2-body semileptonic decay, such as h+ = + , K + or a1+ . Actually, the method
was not used to measure the Michel parameters in the (4S) region [1517]. At the
Z 0 peak, where the terms with (M /ECM )2 can be neglected, each term of the energy
spectrum coupled to the Michel parameters are formulated by 3rd-order polynomials of the
scaled laboratory energy [2,8]. The energyenergy correlation and angle correlation of the
charged particles have been predicted [8,9]. Using the equations and the large polarization
of the lepton from Z 0 decay, the Michel parameters were measured at LEP [1821] and
SLD [22]. In the low-energy region of (4S), integration over the direction involving
the unobserved radiative emission has been performed sophisticatedly by Monte Carlo
simulation [1517].
A new formulation for the Michel parameter measurement is the predicted decay widths
expressed in terms of the measured variables by integrating over the unmeasurable
direction. No measured information is lost in the integration. A change in the variables
from the rest frame to the measured Lorentz frame accompanies a Jacobian, which
can be easily calculated by introducing a special coordinate system. The formulation
with single leptonic decays, ( )(h+ ), ( )( + ), and a double leptonic
decay, ( )(k + k , ), are derived in Sections 3 and 4, respectively, after the pairproduction cross section and the decay widths are reviewed in Section 2. Based on the
formulations, the Michel parameters are actually extracted in Section 5 using the simulated
events to the decays, + ( )( + ) and + ( )(+ , ).
A realistic detector acceptance and performance are taken into account in the extraction.
387
2. Theoretical predictions
Before deriving the correlated decay widths of a + pair in terms of the measured
momenta and angles, we review the theoretical predictions, which are expressed by the
variables in the rest frame and in the center-of-mass system (CMS) of e e+ .
2.1. Angular distribution of -pair production
We give the spin-dependent cross section of -pair production, e e+ ( s ) + ( s
),
by an unpolarized e beam, since the and + decay coherently through their spins.
The cross section is a convolution of the production angle, = ( , cos ), spin
direction ( s ) and + spin direction ( s
):
d ( s , s
)
2
=
(D0 + Dij si sj
),
d
16E2
1
sin2 ,
2

0
1 + 1/2 sin2
2
(Dij ) =
0
sin2
D0 = 1 + cos2 +
sin 2 /
sin 2 /
0
1 + cos2
sin
(1)
/2
where is the fine-structure constant and E , p , and are the energy, momentum,
velocity (p /E ) and E /M of in the CMS, respectively. The spin direction, s1,2,3 and
, are equivalent to sx,y,z and sx,y,z

, which are defined in the and + rest frame,
s1,2,3
respectively. The coordinate system is defined by the direction (
n ) and the incident e
direction (
ne ) in the CMS, such that the z-axis is n and the y-axis is parallel to n n e [3].
Fig. 1 shows the coordinate system.
Fig. 1. Coordinate system of and + rest frame and notation of the direction vectors, which are defined in the
CMS.
388
2.2. Lepton energy spectrum

The lepton energy spectrum of the decay is obtained from the differential
decay width,

d ( )
= W3 FE E w cos w GE E ,
d cos w dE
(2)
where w and w are the polarization of the and the angle between the polarization
vector and the direction of , respectively. The lepton energy (momentum), E (p ) and
w are defined in the parent rest frame. The sign before w coincides with the charge of
the parent .
The isotropic part (FE ) and the polarization dependent part (GE ) of the energy spectrum
are decomposed into terms with the Michel parameters, ( , , ), [1,10]:

fr (x),
FE E = f1 E + f2 E + f3 E +
4

3p E

2p E
m2
W E ,
f2 E =
f1 E =
4E 3W ,
E
W3
3W3
3p m

W E ,
f3 E =
(3)
W3

GE E = g1 E + fg2 E +
gr (x),
4

p2

2p2
m2
4E
.
3W
g2 E =
g1 E = 3 W E ,

M
W
3W3
(4)
Here, the Michel parameters are distinguished by the suffix to test the e universality;
W = (m2 + M2 )/(2M ) is the maximum allowed energy of ; fr and gr are
internal radiative corrections [11] to the intrinsic energy spectrum, FE (E ) and GE (E ),
respectively, and are functions of the scaled energy, x = E /W . Complicated corrections
for external and internal bremsstrahlung are not taken into account throughout the
following formulation. Ignoring the lepton mass (m ), Eqs. (3) and (4) can be rewritten
as familiar expressions of x for the energy spectrum. The equation is approximated very
well in the decay e e , where electron mass can be neglected compared with the
muon mass; however, the muon mass effect in the decay is not negligible.
Significant differences between the exact and approximated equation are seen near the
threshold, x = m /W , in Fig. 2, where the spectrum functions, FE and GE , are calculated
based on the V A prediction, = 3/4, = 0 and = 1, = 3/4.
2.3. Correlated decay widths
We define the coordinates and the corresponding variables to describe the correlated
decay width through -pair production. The variables with an asterisk () are defined
in the rest frame, whose coordinate system is described in Section 2.1, and shown in
Fig. 1. The variables without an asterisk are defined in the CMS, where the z-axis is the
389
Fig. 2. Muon () energy distributions as a function of the scaled energy, x = E /W , by the maximum energy,
W : (a) isotropic part and (b) polarization dependent part, predicted by V A theory, = = 3/4, = 0
and = 1. The dashed lines are the familiar equations, ignoring the muon mass, and the filled lines are the exact
equations.
incident e direction (
ne ) and the y-axis is on a plane consisting of n e and the y-axis of the
detector. Let us designate the coordinate system in the CMS as CMS(xyz). The CMS(xyz)
coincides with a detector coordinate system whose z-axis is the e direction in an e e+
symmetric collision. Suffices (, , h, k . . . ) are used to identify their variables. The vector
components(nx , ny , nz ) of n are equivalent to (n1 , n2 , n3 ).
The correlated decay widths were predicted by Tsai [3] under the V A assumption
before the discovery of the lepton. The predictions are generalized to a polarized e e+
collision by Pi and Sanda [4] and applied to KORALB/TAUOLA [13,14,23], which is a
general simulation program of pair production/decay in the (4S) energy region. The
typical decay widths are written as
2.3.1. e e+ + ( )(h )
Here, h indicates a stable particle, and is either or K , accordingly,

F , h
d ( , h )
,
d dE d dh
= h

2
D0 FE E + h Dij ni nhj GE E ,
2
16E
(5)
where and h do not contain any kinematical variables.

The new parameter, h , is a helicity-dependent parameter in the semileptonic decay,
h , and is related to AV [8,12] as
h = AV =
2gA gV
,
2 + g2
gA
V
(6)
where gV and gA are the vector and axial vector coupling constants of the leptonic charged
current, u (gV + gA 5 ) u , respectively. The V A weak current with lefthanded
390
neutrino fixes the coupling constants as gV = +1 and gA = +1, which lead to h = +1.
The equation says that h and cannot be extracted separately from a measurement of the
decay width. The measurable information is the product, h .
Integrating the decay width over , h , and , the width is written as

1 + 2

d ( , h )
E
+
E
(7)
F
cos
cos
G
.
E
h

E

h

dE d cos d cos h
3 2
The equation means that the contribution of the polarization term varies from 1/3 ( = 0)
to one ( = 1) with an increase of the collision energy. Increasing the CMS energy
to measure the Michel parameters, we can enhance the polarization effect than in the
threshold experiment by a factor of three. This statement is also applicable to other decays.
2.3.2. e e+ + ( )( ) and 0
This is the case that h decays into two stable particles. Let us assign 4-momenta of
, 0 , and to p1 , p2 , and p = p1 + p2 , respectively, to write the decay width. We

have E = M (E1 + E2 ), P = (p 1 + p 2 ) or E = M E , P = p in the
rest frame. According to Refs. [3,23], the width is written as

F ,
d ( , )
,
1
d dE d d dm212 d
=

2
D0 H0 FE E + Dij ni Hj GE E BPS,
2
16E

H0 = 2 E1 E2 P Q E Q2 ,
2
2

Q + 2P Q + p2j
Q 2P Q ,
Hj = p1j
(8)
1 is a solid angle of along the direction (

n )
where Q = p1 p2 , m212 = p2 and
in the rest frame. is a constant without any kinematic variables.

BPS denotes the BreitWigner phase space [23],

2 2p

2p 1
.
BPS = BW m212
(9)
m
m12
The BreitWigner function BW(m212 ) for + + 0 is written in a form of the
coherent sum of two resonances, and
[24]:

BW (m212 ) + BW
(m212 )
BW m212 =
,
1+

m2
BW m212 = 2
, = ,
,
m m212 m (m212 )

3
2
m
p 1
m12 = 0
,
m212 p (m )
(10)
where is a coherence coefficient, p1 and p (m ) denote the momentum in the + 0

rest frame with masses m12 and m , respectively. Table 1 gives the shape parameters of the
391
Table 1
Shape parameters of , m (MeV) and 0 (MeV)
CLEO
774.3 0.7 0.9
149.1 1.4 0.8
1370 10 6
386 39 18
0.091 0.008 0.004
m
0
m
K&S
773
144
1320
390
0.103
BreitWigner function for the + 0 invariant mass (m12 ) distribution. Based on the CVC
hypothesis, K&S is derived from low-energy experiments of e e+ + , while CLEO
is obtained from 0 CLEO [25]. The KORALB/TAUOLA generator refers to
K&S.
2.3.3. e e+ + ( )(k + )
The double leptonic decay contains two independent sets of Michel parameters,
( , , , ) and (k , k , k , k ), corresponding to and + k + :

F , k +
d ( , k + )
d dE d dEk dk

2
D0 FE E FE Ek k Dij ni nkj GE E GE Ek .
2
16E
(11)
The measured information on and k is the product k .
= k
3. e e+ + ( )(h+ ) with measured variables

3.1. Missing mass constraints
Taking into account of the kinematical constraints and geometrical configuration of
the observed particles, h+ and , we introduce a coordinate system of (X, Y, Z) in the
nh ) in the
CMS in order to reconstruct a direction. The Z-axis is the h+ -direction (
CMS(xyz) and the Y -axis is parallel to n e n h . The coordinate system is denoted by
CMS(XY Z) to distinguish CMS(xyz). The quantities with capital character are defined in
the CMS(XY Z), while the variables with small characters are in the CMS(xyz).
The direction (N ) is not uniquely determined, but limited by the decays, +
+
h and . The missing mass = 0 constraint in the decay + h+ is
written as (p ph )2 = 0, which gives the + production angle ( ) with respect to n h ,
cos =

1
2E Eh M2 m2h .
2p ph
(12)
This relation says that the + direction N ( + ) is on a circumference of circle Ch with

radius = sin , which is drawn on a unit sphere, as shown in Fig. 3(a). If azimuthal angle
392
( ) is given on the circumference, N ( + ) is determined. The direction N is opposite

to the + , N = N ( + ).
Similarly, the invariant mass M > 0 of the two s system in the decay
gives a constraint to through the relations
cos
cos ,

1
cos =
2E E M2 m2 ,
2p p
(13)
where
is the angle along the direction. The equation cos
= cos draws a
circle C on the surface of a unit sphere. The inequality confines the N inside C . If the
nominal value of cos is < 1 the inequality is valid for all ; that is, no additional
constraint exists on . This arises in decay with a very small lepton energy. The trivial
case is not considered from now on.
Replacing the direction (N ) with N , we reconstruct the leptonic decay in the
same hemisphere as the + decay, because C and Ch are apt to be located in an away
position in the CMS. The equations of circles Ch and C are written, respectively, as
X2 + Y 2 = sin2
and Z = cos ,
(14)
NX X + NY Y + NZ Z = cos .
(15)
The equations determine the configuration of the two circles. When two circles cross
at point A and B, the allowed region of is A B , as shown in Fig. 3(a).
The corresponding configuration is displayed schematically in Fig. 3(b). The kinematically
allowed region, A B , expands with an increase of , and becomes 0 2 at last
when C covers Ch , as C
in Fig. 3(b). When is above /2, the allowed region moves
from the inside to the outside of C . If two circles cross at A and B under the configuration
> /2, the allowed region is B A , where on Ch is taken counterclockwise around the Z-axis.
(a)
(b)
Fig. 3. (a) Configuration of the two circles Ch and C on a unit sphere, which are determined from the decay,
+ h+ and , respectively. The kinematically allowed region for Ch is on the circumference
and the region for C is either inside or outside of the circle, depending on cos > 0 or < 0. (b) Topological
configuration on the surface of the unit sphere.
393
3.2. Transformations of coordinate systems

We introduce an intermediate coordinate system CMS(x
y
z
) between CMS(xyz) and
the rest frame, where the z
-axis is the direction (
n ) and the y
-axis is parallel
to n n e , as shown in Fig. 1. The 4-momentum in the CMS(x
y
z
) is written as
p
=t (px
, py
, pz
, E), which is transformed into rest frame through a simple Lorentz
transformation, Lz ( ), along the z
-axis, where
1 0 0
0
0
0 1 0
Lz () =
(16)
.
0 0
0 0
Thus, a 4-dimensional vector is transformed into another coordinate system by either a
rotation or a Lorentz transformation, as shown in Fig. 4. The rotation Rh from CMS(xyz)
to CMS(XY Z) and the rotation R from CMS(xyz) to CMS(x
y
z
) are determined by
the angles of reference vectors n h (or n k , n ) and n , respectively, and are represented in
4-dimensional space by
Rh = R(h , h ),
(17)
R = R( + , ),
(18)
where
cos cos
sin
R(, ) =
cos sin
0
sin cos
cos
sin sin
0
sin
0
cos
0
0
0
.
0
1
(19)
The new z-axis by the rotation R(, ) coincides with a reference vector, n =t (cos sin ,
sin sin , cos ). The reverse rotations are Rh1 =t Rh and R1 =t R , where the upper
suffix t denotes the transposition of a matrix or a vector.
Giving in A B on the circumference of Ch , the direction (N ) is
written as
t
N =
( cos sin , sin sin , cos ).
(20)
Fig. 4. Definition of the momentum in various coordinate systems and the transformations from a measured
coordinate system whose z-axis is the incident e direction.
394
The direction n in the CMS(xyz) is obtained by the reverse rotation,
n = Rh1 N =t
Rh N ,
(21)
and n are regarded as 4-dimensional vectors with a

where the 3-dimensional vector
dummy 4th component. The equation yields the relations between , and , ,
sin h cos h cos sin + cos h sin sin + sin h sin h cos
,
cos h cos h cos sin sin h sin sin + cos h sin h cos
(22)
cos = sin h cos sin cos h cos .
(23)
tan =
Once n is determined, R can be calculated. The momentum (p ) is given by a Lorentz

transformation (Lz ( )) after the rotation (R ) of p from CMS(xyz) to CMS(x
y
z
) in
4-dimensional space,
p = Lz ( )p
= Lz ( )R p,
(24)
where p denotes either ph or p . The equation leads to the following relations:

E = (E pz
),

n n = p cos( ) sin sin + cos cos ,

pz
= p
sin( ) sin
,
cos( ) cos sin sin cos
E
p
cos = cos( ) sin sin + cos cos
.
p
p
tan =
(25)
(26)
(27)
3.3. Decay width with measured variables

The total decay width, ( , h ), is the integration of the differential width,
given by Eq. (5) over all variables. The integration can be rewritten by using
measured variables and the newly introduced azimuthal angle ( ) as

, h+ = F , h+ d dE d dh ,
F ( , h+ ),
B
=

(E , , h , )
d dp d dph dh ,
, h
(p , , p , , )
(28)
which leads to the differential correlated decay width with measured variables,
d ( , h+ )
=
dp d dph dh
B

(E , , h , )
d .
F , h+
(p , , ph , h , )
(29)
The Jacobian is the determinant of a 7 7 matrix. The elements are calculated using the
relations obtained from Eqs. (21) and (24) which relate (E , , h , ) to the measured
395
variables (p , , ph , h ) and . Since p h is determined by (p h , ) and is independent

of p , the Jacobian can be reduced to a product of two degraded Jacobians,

(E , , h , ) (E , ) (h , )
=

(30)
(p , , p , , ) (p , ) (p , , ) .

h
h

h
h
The first Jacobian originates in the leptonic decay and the second one is the
semileptonic decay + h+ .
After some manipulation, we obtain the following Jacobians:

2

(E , )
cos
p
2

= p
sin ,
(p , )
E p
A2

2

(h , ) M
cos h
ph
2
=

sin
,
(p , , ) p
Eh ph
A2h
h
h
Ai = cos(i ) sin i cos cos i sin
(31)
(32)
(33)
where i = or h for Ai , cos = n n and ph = (M2 m2h )/(2M ). The details on the
calculations can be seen in Ref. [28].
Apart from the factors outside of { }, there is no essential differences between Eqs. (31)
and (32), because is the angle of n h with respect to n , while is the angle of
n . The same expressions are also obtained in the charge conjugated decay, +
(h )(+ ).
3.4. e e+ + ( )( + ( + 0 ) )
This is the case that h+ mentioned in the previous subsection decays into stable particles
via the strong interaction, and therefore a similarity is expected in the derivation. We
select p = p1 + p2 and p1 to express the decay width instead of the directly measured
momenta (p1 and p2 ) themselves, since the predicted width (F ( , + )) is given by the
corresponding variable . The invariant mass (m12 ) belongs to a variable in the + rest
1 + E
2 .
frame because of m12 = E
The transformation of p1 into the rest frame can be performed by Lorentz
transformation, Lz ( ), through a rotation (R( , )), as shown on the right half of
Fig. 4, and is given by p1 = Lz ( )R( , )p1 , where = p /E . In the same way
1 ) and as
as Eq. (29), the total width is written by (p , , p , , m212 ,

, +

1 ,
= F , + d dE d d dm212 d
B
=
A

+ (E , , , )
dp d dp d d dm2 d

1 .
,
12
(p , , p , , )
(34)
396
Thus, the differential decay width with the measured variables is written as
d ( , + )
=
1
dp d dp d dm212 d
B

(E , , , )
d , (35)
F , +
(p , , p , , )
in which the Jacobian has the same form as Eq. (29). The corresponding Jacobians are
obtained from Eqs. (31) and (32), by replacing h with .
One can see that the Jacobian in the semileptonic decay part has a universal form in a
quasi two-body decay. Thus, the Jacobian (32) is applicable to other quasi two body decay
as + K + and + a1+ by replacing the index h with K and a1 , respectively.
4. e e+ + ( )(k + k ) with measured variables
We take the positively charged lepton k + as a reference particle to define the
generation coordinate system, CMS(XY Z). The rotation from CMS(xyz) to CMS(XY Z)
nk ), and is given by R(k , k ).
is determined by the k + direction (
In addition to Eq. (13) for the decay , a similar constraint on is obtained
from + k + k ,
cos cos k ,

1
cos k =
(36)
2E Ek m2k M2 .
2p pk
This is a looser constraint than the semileptonic decay with one missing. The boundary
equation, cos = cos k , draws a circle (Ck ) on a unit sphere and the inequality confines
the N inside of the Ck for cos k > 0 or outside of Ck for cos k < 0 as mentioned
in Section 3.1. Thus, the allowed region S for the direction (= N ) is a common
part of the areas given by Eqs. (13) and (36), whose boundaries draw circles, C and
Ck , correspondingly. Typical configurations for cos > 0 and cos k > 0 are shown in
Fig. 5(a) and (b). S equals Ck for a larger circle (C
). The combinations of cos k < 0, > 0,
cos < 0, > 0 and crossing or no crossing of the circles yield topologically various
shapes of the allowed region (S).
Giving and on S, we have the direction through the rotation n =t
R(k , k )N , which is similar to the derivation of Eq. (21). The total correlated decay
width, ( , k + ), is the integration of Eq. (11) over all variables, and can be rewritten
by measured variables and additional variables ( , ). Similarly to the derivation of
Eq. (29), we have the differential correlated decay width with the measured variables,

(E , , Ek , k , )
d ( , k + )
+
d d cos .
= F , k
dp d dpk dk
(p , , pk , k , , cos )
S
(37)
The Jacobian is decomposed into the product of two Jacobians by the same reason as
mentioned in Section 3.3, and is written as

(E , , Ek , k , ) (E , )
(Ek , k , )
=

(38)
(p , , p , , , cos ) (p , ) (p , , , cos ) .
(a)
397
(b)
Fig. 5. (a) Configuration of the two circles Ck and C on the surface of unit sphere where Ck and C are the
boundaries kinematically allowed in the decay + k + and , respectively. (b) Topological
configuration on a unit sphere. The + direction N passes point T in the common area of circles Ck and C .
The first Jacobian originating in the decay is given by Eq. (31). After some
manipulation for the second Jacobian, we have

2

(Ek , k , )
cos k
pk
2

= pk
(39)
sin
,
(p , , , cos )
Ek pk
A2k
k
k
where Ak is obtained from Eq. (33) by substituting the suffix k for i. The details on the
calculations can be seen in Ref. [28]. Noting that is the angle k between n k and n ,
Eqs. (31) and (39) are expressed by the same function.
5. Extraction of the Michel parameters

We extract the Michel parameters using the correlated decay widths for +
( )(+ ), which are derived in Sections 3.4 and 4,
respectively. Examples of the typical decays can be applied analogously to other decays
of h = , K, K , a1 or double leptonic decay with different daughter leptons ( = k).
Since the decay width is a function of many measured variables, an unbinned maximum
likelihood method is applied to the extraction. The used events are generated by the
MC program, KORALB/TAUOLA, in which the angle correlation between the particles
decayed from and + are implemented [13,14,23]. No radiative corrections are taken
into consideration in the event generation.
( ) ( + ) and +
5.1. Likelihood function

Let us formulate the probability density function (PDF) for + ( )( + ).
The decay width (F ( , + )) given by Eq. (35) can be decomposed into terms with
coefficients (j ) = (1, , , ) [16],
4

F , + =
j Ij (m),
j =1
(40)
398
1 ). For an event with

where m represents the measured variable set (p , , p , , m212 ,
the measured set (mi ) after a cut, Cut (m), the PDF is written as
4
j =1 j Cut (mi )Ij (mi )
,
P(a; mi ) =
(41)
N (a)

4

N (a) =
(42)
j Cut (m)Ij (m) dm,
j =1
where Cut (mi ) = 1 and a represents the parameter set to be extracted.

N (a) is the

normalization factor to insure that the total probability of PDF is P(a; m) dm = 1 for
the various values of the a. The natural choice of a is ( , , ), which coincides
with 2,3,4 . Another choice is the Michel parameter, itself ( , , ), but a larger error
correlation between and is expected. The f3 term for in Eq. (3) is neglected
because of the small factor me /We 0.0006 for = e. We do not include the parameter
in the fitting to the decay = , too [15,17].
The actual calculation of Eq. (42) is not so easy compared with that of theoretical
prediction,

4

j Cut (m)Ij m dm ,
N (a) =
(43)
j =1
where Ij (m ) is the corresponding equation to Ij (m) derived from Eq. (8) and m =
1 ). We can avoid the difficulty of the kinematically allowed
( , E , , , m212 ,
region on in the integration which is described in Section 3.1. The value of m for
Cut (m) is uniquely determined from m through the reverse transformations of Eq. (24).
The actual algorithm for the integration is as follows:
(1)
(2)
(3)
(4)
generation of + ( )( + ) in the uniform phase space of m ;

event cut by Cut (m) after transformation, m m;
calculation of Ij (m );
summation of Ij (m ) over all values of m by repeating (1), (2) and (3).
The likelihood function (L(a)) for Nev events selected by Cut (m) is written logarithmically as
1
ln P(a; mi ).
Nev
Nev
ln L(a) =
(44)
= A(a)
1
The error (ai ) of the fitting parameter (ai ) is evaluated by the matrix C(a)
[26], where the ij element of A(a)
is written as

2 ln L(a)
Aij =
(45)
ai aj a=a
and a is the most probable value of a. The variance of a is written by the diagonal element
of C as 2 (ai ) = Cii .
399
Similarly, the PDF of + ( )(k + k ) is decomposed into 8 terms of

(j ) = (1, , k , 2 , k2 , k , k , k k , k k ). The one choice of a is a =
( , k , k , , k ). In case of decay with = k, the fitting parameters and are reduced
by two, respectively:

a = , 2 , and = 1, , 2 , 2 , 2 , 2 2 .
The Michel parameter extraction was carried out using double leptonic decay with = e
and = . If = k, a considerably hard load of CPU is required to search for the most
probable values of 5 parameters (ai ), simultaneously.
5.2. Michel parameters obtained from MC events
The KORALB/TAUOLA generated MC events are simulated to a KEKB asymmetric
collision of 8 GeV(e ) 3.5 GeV(e+ ). The cut for event selection, Cut (m) is similar
to the Belle detector acceptance and lepton identification performance [27], which are
summarized in Table 2, where m denotes the variables in the asymmetric laboratory frame,
except for Psum . After Cut (m), m represents the variable in the CMS(xyz). The selected
events are listed in Table 3. No m-dependent detection efficiency (L(m)) or backgrounds
are taken into account. Further, the measurement error on m is regarded to be zero.
Using the simulated events, the Michel parameters are estimated by the maximum
likelihood method. The likelihood function is given by Eq. (44). It is difficult to show the
behavior of the likelihood function in the three-dimensional space of the fitting parameters
(ai ). We display the likelihood function using a figure of ln Lmax (a1 ) and the contour
on the (a2 , a3 ) plane with a1 = a 1 where ln Lmax (a1 ) is the maximum of ln L(a) on the
(a2 , a3 ) plane with a given value of a1 . The special treatment of a1 (= ) is based on the
fact that a1 governs the likelihood function and the effects of a2 or a3 through the angle
correlation are subsidiary. Fig. 6(a)(d) show the behavior of the likelihood functions for
+ (e e )( + ) and + ( )(+ ), respectively. As expected
from the selection of the fitting parameter (ai ), a small correlation is seen in Fig. 6(b),
while Fig. 6(d) shows a large correlation. The results are summarized in Table 3. The
errors are estimated by Eq. (45).
Table 2
Event selection criteria
+ ( )( + )
Charged Pt
Pe
P
P
P 0
cos
+ ,
Psum in CMS
> 0.1 GeV/c

> 0.5 GeV/c
> 1.2 GeV/c
> 1.0 GeV/c
> 0.2 GeV/c with E > 0.1 GeV
0.60 0.83
17 150
P + P + < 9 GeV/c
All variables except the last column are measured in the laboratory frame.
+ ( )(+ )
> 0.1 GeV/c
> 0.5 GeV/c
> 1.2 GeV/c
0.60 0.83
P + P+ < 9 GeV/c
400
Table 3
Michel parameters by maximum likelihood fit
(e )( + )
MC event
Selected event
N
(2 ) N
( ) N
500 000
156 899
0.752 0.0052
0.98 0.0220
0.74 0.0121
( )( + )
(e )(e+ )
( )(+ )
500 000
134 465
0.748 0.0059
1.01 0.0252
0.75 0.0133
1 000 000
407 157
0.748 0.0026
0.98 0.0406
0.75 0.020
1 000 000
246 849
0.753 0.0034
1.01 0.0471
0.76 0.024
The fitted parameters for the double leptonic decay are 2 and .
Fig. 6. Likelihood function for the decay, + (e e )( + ); (a) Likelihood values, ln Lmax (e ) as
function of e . ln Lmax (e ) is the maximum values on the ( e , e e ) plane where the offset are adjusted
appropriately. The smooth curve is a second-order polynomial fitted to asterisks which are the probed points
of the likelihood function. (b) Contour of likelihood function on the ( e , e e ) plane with e = 0.75.
The cross bar indicates the region of one for the fitted values. (c), (d) Likelihood function for the decay,
+ ( )(+ ).
6. Summary
Using the measurable variables, we formulate the theoretically predicted decay widths
for leptonic decays correlated with , or leptonic decay
via -pair production. The expression is the integral of the predicted width over the
unobserved direction. Introducing a special coordinate for the direction, the Jacobian
for the integration is given by a simple expression. The integration cannot be calculated
analytically, but the numerical calculation is not so difficult. In order to extract the Michel
401
parameters by the maximum likelihood method, the probability density function (PDF) is
derived from the expression. The Michel parameters extracted from KORALB/TAUOLA
MC-events reproduce the input values based on the V A assumption.
The approximated equation of the lepton energy spectrum, which has been widely
applied to the decay e e , is distorted near to the threshold of the decay
. We should return to the original equation without ignoring the muon mass.
The integration over the direction and the azimuthal angles of the charged particles
leaves a factor (1 + 2 )/(3 2 ) in the polarization term of the lepton energy spectrum.
The factor implies that high-energy experiments yield better measurements of the Michel
parameters ( , ) related to the polarization than the threshold, since the value is 3 at
= 1 and 1 at = 0.
Acknowledgements
The author wishes to express his deep thanks to Prof. T. Ohshima of Nagoya university,
who gave a motive to initiate the study as well as encouragement throughout. The author
also wishes to thank Prof. H. Hayashii of Nara Womens university for a fruitful discussion.
Finally, sincere thanks are given to Profs. S. Iwata and F. Takasaki for their support.
References
[1] L. Michel, Proc. Phys. Soc. London A 63 (1950) 514;
C. Bouchiat, L. Michel, Phys. Rev. 106 (1957) 170.
[2] Particle Data Groupe, Phys. Rev. D 66 (2002) 355, The laboratory lepton energy spectrum are written by
A. Stahl in p. 375.
[3] Y.-S. Tsai, Phys. Rev. D 4 (1971) 2821.
[4] S.-Y. Pi, A.I. Sanda, Ann. Phys. (N.Y.) 106 (1997) 171.
[5] M. Davier, L. Duflot, F. Le Diberder, A. Rouge, Phys. Lett. B 306 (1993) 411.
[6] H.E. Harber, SLAC-report-343 (1989);
W. Hollik, T. Sack, Phys. Lett. B 284 (1992) 427;
B. McWilliams, L.-F. Li, Nucl. Phys. B 179 (1981) 62;
A. Stahl, Phys. Lett. B 324 (1994) 121;
A. Stahl, H. Voss, Z. Phys. C 74 (1997) 73.
[7] W. Fetscher, Phys. Rev. D 42 (1990) 1544.
[8] C.A. Nelson, Phys. Rev. D 40 (1989) 123.
[9] R. Alemany, et al., Nucl. Phys. B 379 (1992) 3.
[10] R.E. Marshak, Riazuddin, C.P. Ruan, Theory of Weak Interactions in Particle Physics, WileyInterscience,
New York, 1968, p. 228.
[11] T. Kinoshita, A. Sirlin, Phys. Rev. 107 (1957) 593;
T. Kinoshita, A. Sirlin, Phys. Rev. 107 (1957) 638;
T. Kinoshita, A. Sirlin, Phys. Rev. 113 (1959) 1652.
[12] H. Kuhn, F. Wagner, Nucl. Phys. B 326 (1984) 16.
[13] S. Jadach, Z. Was, Comput. Phys. Commun. 36 (1985) 191.
[14] M. Jezabek, Z. Was, S. Jadach, J.H. Kuhn, Comput. Phys. Commun. 70 (1992) 69;
S. Jadach, Z. Was, R. Decker, J.H. Kuhn, Comput. Phys. Commun. 76 (1993) 361.
[15] ARGUS Collaboration, Phys. Lett. B 246 (1990) 278;
ARGUS Collaboration, Phys. Lett. B 341 (1995) 441;
ARGUS Collaboration, Phys. Lett. B 349 (1995) 576.
402
[16] ARGUS Collaboration, DESY 97-194;

ARGUS Collaboration, Phys. Lett. B 431 (1998) 179.
[17] CLEO Collaboration, Phys. Rev. Lett. 25 (1997) 4686;
CLEO Collaboration, Phys. Rev. D 56 (1997) 5320;
CLEO Collaboration, Phys. Rev. D 61-012002.
[18] ALEPH Collaboration, Phys. Lett. B 346 (1995) 379.
[19] L3 Collaboration, Phys. Lett. B 377 (1996) 313;
L3 Collaboration, Phys. Lett. B 438 (1998) 405.
[20] OPAL Collaboration, Eur. Phys. J. C 8 (1999) 3.
[21] DELPHI Collaboration, Eur. Phys. J. C 16 (2000) 229.
[22] SLD Collaboration, Phys. Rev. Lett. 25 (1997) 4691.
[23] S. Jadach, J.H. Kuhn, Z. Was, Comput. Phys. Commun. 64 (1991) 275.
[24] J.H. Kuhn, A. Santamaria, Z. Phys. C 48 (1990) 445.
[25] J. Urheim, Nucl. Phys. B (Proc. Suppl.) C 55 (1997) 359.
[26] S. Brandt, Statistical and Computational Methods in Data Analysis, North-Holland, Amsterdam, 1976.
[27] Belle Collaboration, Phys. Lett. B 551 (2003) 16.
[28] K. Tamai, KEK preprint 2003-14.

On the thermodynamical limit of self-gravitating

systems
Victor Laliena
Departamento de Fsica Terica, Universidad de Zaragoza, Cl. Pedro Cerbuna 12, E-50009 Zaragoza, Spain
Received 28 March 2003; accepted 9 July 2003
Abstract
It is shown that the diluted thermodynamical limit of a self-gravitating system proposed by de Vega
and Snchez suffers from the same problems as the usual thermodynamical limit and leads to
divergent thermodynamical functions. This question is also discussed from the point of view of mean
field theory.
PACS: 05.20.-y; 05.90.+m; 05.70.-a
Recently, de Vega and Snchez have pointed out that a kind of thermodynamical
limit of a self-gravitating system can be defined if one considers what they call the
diluted limit: send the number of particles, N , and the volume, V , to infinity, keeping
constant the ratio N/V 1/3 instead of the density, N/V [1]. This is a rather surprising
and very interestingsuggestion, since it is well known that the usual thermodynamical
limit of self-gravitating systems leads to singular thermodynamical functions (due to
the gravitational instability) [2,3]. However, we will show in this paper that the diluted
regime leads to divergent thermodynamical functions in a way similar to the usual
thermodynamical limit. Before demonstrating this statement, we will remember the
differences, in what concerns the thermodynamical limit, between a thermodynamical
stable system [4] and a self-gravitating system, and present the arguments that support the
idea that the diluted limit gives well defined thermodynamical functions for self-gravitating
systems. Afterwards, we shall proof that this is not possible, and we will point out the
loophole in the arguments that lead to the wrong conclusion. We will end the paper with a
brief discussion on mean field theory.
E-mail address: laliena@posta.unizar.es (V. Laliena).
doi:10.1016/j.nuclphysb.2003.07.005
404
V. Laliena / Nuclear Physics B 668 (2003) 403411
Consider a system of N classical particles, endowed with a hard core, the interactions
of which are described by a potential energy which is a sum of two body contributions:

N (r1 , . . . , rN ) =
(1)
(ri rj ).
i<j
Using the notation ij (ri rj ), the microcanonical and canonical partition functions,
ZMC and ZC , respectively, can be written as

3N/21

N

1
3
ZMC =
(2)
d ri E
ij
,
N!(3N/2)
VN
ZC =
1 3N/2
N!

N
VN
i<j
i=1

d 3 ri exp
i=1
ij ,
(3)
i<j
where E is the energy, V the volume, the inverse temperature, [x]+ = x if x > 0 and
[x]+ = 0 if x 0, and we have ignored some irrelevant factors involving powers of the
particle mass. Introducing the family of functions

N
1
()
3
GV (u) =
(4)
d ri u
ij ,
N
VN
i=1
i<j
the partition functions can be obviously written as

1
3N/21
ZMC =
,
du G()
V (u) E N u +
N!(3N/2)

1 3N/2
()
ZC =
du GV (u) exp N u .
N!
(5)
(6)
Assuming that, as N , keeping the density = N/V fixed, the following

asymptotic behavior holds
Ng(u,)
G(1)
,
V (u) e
(7)
we can get the partition functions with the aid of the saddle point method. For the
microcanonical ensemble (MC) we have the equation
g(u, ) 3
1
=
.
u
2 E/N u
(8)
The solution is, obviously, a function of = E/N and = N/V , u = u m (, ), and the
entropy scales with N . The corresponding saddle point equation for the canonical ensemble
(CE) is
g(u, )
= ,
(9)
u
the solution of which is a function of and , u = u c (, ), and the corresponding
thermodynamical potential (usually called the Helmholtz free energy) is extensive. It is
405
not difficult to realize that the saddle point equations imply the equivalence between the
MC and the CE.
Let us formally see that the scaling (7) holds if the two body potential is short ranged.
We will use the notation

1
d 3 ri f (r1 , . . . , rN ).
f (r1 , . . . , rN ) = N
(10)
V
VN
Using the Fourier representation of the Dirac delta we can write

()
GV (u) = V N
d iu
()
e GV (),
2
(11)
where, with our definition (10),

N
1

()
3
GV () exp i
= N
ij
d ri exp i
ij .
N
V
N
i<j
VN
We can now apply the cumulant expansion to the above expression

(i)n
exp i
= exp
ij
...
i1 j1 in jn c ,
n!N n
i<j
n=0
i1 <j1
(12)
i<j
i=1
(13)
in <jn
where i1 j1 in jn c is the connected correlation function. It is well known that only

those sequences of coordinates {i1 j1 . . . in jn } for which the integral of Eq. (14) below
cannot be split into the product of two (or more) integrals contribute to the connected
correlation function. We will call such sequences connected sequences. For large N and
n N , the number of connected sequences asymptotically is cn N n+1 , where cn is a
number independent of N . Let us define l = ril rjl , for l = 1, . . . , n. The number of
connected sequences for which the l s are linearly dependent scales with a lower power
of N , and, therefore, they do not contribute as N . Only connected sequences with
the l s linearly independent contribute to the leading term in N of the nth order of the
cumulant expansion. For such sequences, if the potential decays sufficiently fast at long
distances, the integral of i1 j1 in jn over the l s gives a number, i1 j1 ...in jn , that is
independent of the volume if V is large. The integral over the remaining variables gives a
power of the volume, V Nn . Therefore, we have:

N
1
VN
VN
d 3 ri i1 j1 in jn
i=1
i1 j1 ...in jn
.
Vn
(14)
The connected correlation function is a sum of products of integrals of the above type and
has the same behavior. Collecting all the results, we have

cn n
(15)
...
i1 j1 in jn c N n+1 n ,
V
i1 <j1
in <jn
406
where n is the average of i1 j1 ...in jn over the connected sequences. This behavior is
broken down for n N , but the contribution of these terms to the cumulant expansion
is negligible. Then, the nth term is
n N n
n cn n
(i)
(16)
.
n! N n1 V
Performing the change of variables N in the integral of Eq. (11), we get for = 1
G(1)
V (u)
VN
N
d N[iu+(i)]
e
,
2
(17)
where
(x) =
(1)n
n=0
cn n n
x .
n!
(18)
The integral of Eq. (17) can be evaluated by the saddle point method, and gives
(1)
GV (u)
VN
exp[Ng(u/)],
N
(19)
where
g(u/) = (u/)u/
+ [(u/)],
(20)
and (u/)
is the solution of
()
= u/.
(21)
Therefore, if the potential is short ranged, G(1) (u) verifies the scaling of Eq. (7) and the
thermodynamical functions are well behaved in the thermodynamical limit.
What about a self-gravitating system? Since the potential decays as the inverse distance,
we have that, in the same conditions as before
1
VN

N
VN
k=1
d 3 rk i1 j1 in jn
i1 j1 ...in jn
,
Rn
(22)
where R is the linear size of the system (V = R 3 ). The same arguments of the above
paragraphs lead to the following expression for the nth term of the cumulant expansion:
n
n
n n
2n/3 N
V
.
(i)
(23)
n! N n1
V
The above expression shows that, as expected, the asymptotic behavior (7) does not hold.
However, we see that for = 5/3 the nth order term of the cumulant expansion is
n N n/3
n n
.
(i)
(24)
n! N n1 V
407
Hence, changing the variable N in the integral of Eq. (11) and using the saddle
point method, leads to

VN
exp Ng u/ 1/3
N
for a self-gravitating system.
Now, the saddle point equation that gives the MC partition function is
(5/3)
GV
(u)
1
g(u/ 1/3 ) 3
=
.
u
2 E/N 5/3 u
(25)
(26)
The solution is a function of E/N 5/3 and , u = 1/3 u m [E/( 1/3N 5/3 )], and therefore,
the entropy is not extensive.1 The MC temperature, TMC = ( ln ZMC /E)1 , is
2E 2 2/3 1/3 5/3 1/3
N u m E/ N
.
TMC =
(27)
3N
3
We see that the MC temperature diverges as the number of particles increases. This is a
manifestation of the so-called gravothermal catastrophe [5].
Let us analyze the canonical ensemble. If the temperature is fixed, the contribution
(5/3)
of GV (u) is negligible compared with the contribution of the Boltzmann weight,
exp[N 5/3u], and the saddle point solution is given by u = u c = u0 , where u0 is the
minimum possible value of u, that is, the minimum potential energy. Hence, the system is
completely collapsed. The only way to avoid the complete collapse of the system as the
number of particles increases is to increase the temperature by a factor N 2/3 , in agreement
with the MC analysis.
The thermodynamical potentials for self-gravitating systems are non-extensive and
the system will be inhomogeneous. The thermodynamical limit is singular and gives ill
behaved thermodynamical functions.
Up to here, everything is known. There is, however, another interesting possibility:
as de Vega and Snchez pointed out [1], it seems that the function G(1)
V (u) will have
the asymptotic behavior as exp[Ng(u, N/R)] if we keep = N/R constant (instead of
= N/V ) as N . They called this the diluted limit. Indeed, plugging V = R 3 in
(1)() reads
Eq. (23), the nth term of the cumulant expansion of G
V
n
n
N
n
(i)n
(28)
.
n! N n1 R
Then, G(1)
V (u) exp[Ng(u, )], and the thermodynamical potentials will scale with N . In
particular, the canonical partition function will scale as
V N Nf (, )
e
.
(29)
N
The system will not be extensive, however, and will develop inhomogeneities, but
the thermodynamical functions will be well behaved in the thermodynamical (diluted)
limit [1].
ZC
1 It is not an homogeneous function of N , E, and V .
408
The above proof of the existence of the diluted thermodynamical limit is formal, since it
relies on a series expansion, the convergence of which has not been demonstrated. Indeed,
we are going to show that the diluted limit cannot give well behaved thermodynamical
functions. Let us consider the canonical partition function on a volume V = R 3 in the
diluted regime, so that N R, and let us take a portion of such volume of linear size
R0 < R, such that N R03 . Then, we obviously have

N

V0N
1 3N/2 3
exp
ZC
d ri exp
ij
ij V0 , (30)
N!
N!
V0N
i=1
i<j
i<j
where
ij V0 =
1
V0N

N
V0N
d 3 rk ij ,
(31)
k=1
and the last inequality follows from a well-known property of the exponential function:
exp(y) exp(y ). Since ij V0 = /R0 , where > 0 is a geometrical number
independent of R0 if R0 is large, we have
V0N N(N1)/R0
e
(32)
.
N!
Recalling that R0 L0 N 1/3 , where L0 is a fixed number with dimensions of length, we
have

NN
exp N 5/3 /L0 .
ZC
(33)
N!
Therefore, the canonical partition function cannot behave as exp[Nf (, )] in the diluted
thermodynamical limit. Rather, it behaves as in the usual thermodynamical limit. The
reason is clear: the partition function is dominated by collapsed configurations even in
the diluted regime. The gain in entropy provided by the dilution, which is of the order
N ln V , cannot compete with the energy gain due to collapse, which of the order N 5/3 .
If the canonical partition function does not scale as exp[Nf (, )], it is impossible
that G(1)
V (u) exp[Ng(u, )]. This is in conflict with the result of the cumulant expansion.
The inequality (32) that originates this conflict is rigorous, so that the fallacy must be found
in the cumulant expansion. The solution of the paradox is that the cumulant expansion
(1)
for GV (u) is dominated by terms of the order of N . It should converge for any finite
(1) (N) is the

N , but the radius of convergence shrinks to zero as N . Indeed, G
V
analytical continuation of the ZC () to the imaginary axis. If the radius of convergence of
the cumulant expansion were finite in the diluted thermodynamical limit, it would imply
[cf. Eq. (17)]
ZC
(1) (N) eN(i ) .

G
V
(34)
The asymptotic behavior of the above equation is not compatible with Eq. (33). The
cumulant expansion for (x) given by Eq. (18) must have a vanishing convergence radius
in the diluted thermodynamical limit of a self-gravitating system.
409
There is a clear explanation of the failure of the cumulant expansion. For imaginary ,
(1) (N) is a canonical partition function at temperature 1/ Im(). The cumulant

G
V
expansion
computed with a flat
relies on connected correlation functions
measure,
V N i d 3 ri , instead of the Boltzmann weight, ZC1 i d 3 ri exp[ Im() ij ]. The
cumulant expansion, then, will be valid when the two measures are similar, i.e., in a gas
phase. In a self-gravitating system, Eq. (33) indicates that the canonical partition function
is completely dominated by the potential energy, and therefore that collapse takes place,
at temperatures smaller or of the order of N 2/3 . Hence, for imaginary , the cumulant
expansion will be only valid for Im() < N 2/3 . Thus, the convergence radius of the
cumulant expansion shrinks to zero as N 2/3 in the diluted thermodynamical limit.
The collapse in the thermodynamical limit can be avoided by rescaling properly the
radius of the hard core, a.2 Indeed, it is known that in mean field theory the collapse phase
covers the whole phase diagram in the limit in which the filling parameter, Na 3 /R 3 ,
tends to zero [6]. The filling parameter remains constant in the usual thermodynamical
limit. To keep it constant in the diluted limit, the hard core must be scaled as a N 2/3 .
Then, previous argument does not apply, since such big particles do not fit in a volume of
linear size R0 N 1/3 . The minimum size of a region able to enclose the N particles must
have a linear size scaling as N . Hence, the diluted thermodynamical limit can exist. This
is, however, a rather trivial and ad hoc way of avoiding collapse: particles are forced to
remain far away one from another, and it is physically difficult to justify the scaling of the
hard core. Moreover, this way of preventing collapse in the thermodynamical limit cannot
apply to other types of short distance regularizations, such as softened potentials [7]. Our
proof of the non-existence of the diluted thermodynamical limit is also valid for softened
potentials.
It is generally believed that mean field theory is exact for self-gravitating systems.
Since in mean field theory any thermodynamical function depends on the thermodynamical
variables only through the dimensionless combinations = ER/(Gm2 N 2 ) (MC) or
= Gm2 N/R (CE), the thermodynamical limit must be taken keeping either or
finite. Then, it could be argued that the dependence of or on N can be absorbed in any
of the quantities entering or . For instance, one could take R N and E N , as in the
diluted limit. We will show in the following that this is not correct.
Consider a softened gravitational potential, for instance, Gm2 (r)/R,

with (r)
=
R/ r 2 + s 2 , where s is the softening scale. Dividing the finite region that encloses the
system in W cells of linear size w (w3 = R 3 /W ), and denoting by ni the occupation
number of the ith cell, the partition functions (2) and (3) can be approximated by [2]:
3N/2

N
N

( l nl N)
1

ZMC =
...
ni nj ij
,
+
(3N/2)
l nl !
i,j
n1 =0
nW =0
+

N
N

( nl N)
1
l

exp
...
ni nj ij ,
ZC =
(3N/2)
l nl !
n1 =0
nW =0
i,j
2 I am grateful to P.-H. Chavanis and O. Fliegans for pointing this out to me.
(35)
(36)
410
i rj ) and ri is the position of the center of the ith cell (we have again
where ij = (r
ignored constant factors). The above equalities hold rigorously in the limit W . For
large N , the factorials can be approximated by their asymptotic form. Defining the density
by i = ni /(Nw3 ), so that i [0, 1/w3 ] becomes a continuum variable as N , we
have, ignoring constant factors

W

3
dk
w l 1
ZMC =
l
k=1

3
6
(i ln i i ) + ln +
w i j ij
,
exp N w
2

ZC =

W
k=1

dk

i,j
(37)
w3 l 1

(i ln i i ) +
w6 i j ij
exp N w
3

.
(38)
i,j
Mean field theory is obtained by taking N before W . In such case, the

integrals in the above equations are saturated by the maximum of the integrands, provided
that the number of cells, W , and or , respectively, are kept constant. Obviously, the
maximum of the integrand is given by the maximum of the entropy per particle,

S = d 3 r [(r) ln (r) (r)]

3
r ) ,
+ ln + d 3 r d 3 r (r)(r )(r
(39)
2
in the MC, and with the minimum of the free energy per particle,

3
r ),
F = d r [(r) ln (r) (r)] + d 3 r d 3 r (r)(r )(r
(40)

in the CE, with the constraint d 3 r (r) = 1, if the grid is fine enough. The thermodynamical functions depend, therefore, on two dimensionless parameters: s/R and or .
However, to derive mean field theory rigorously, one has to take the limit N in
(37) and (38), keeping the number of cells constant. Otherwise, if W grows with N , the
integrals of Eqs. (37) and (38) cannot be evaluated by the saddle point. This implies that R
must be kept constant. Therefore, the factors N entering and cannot be absorbed in R.
They can be absorbed in G or in m2 . Hence, mean field theory does not support the diluted
limit either.
In conclusion, the thermodynamical limit of a self-gravitating system does not exist,
either in the usual form or in the diluted regime of Ref. [1]. The meaning of non-existence
of the thermodynamical limit is that the thermodynamical potentials do not scale properly
with N and thus thermodynamical functions, such as temperature, diverge. Nevertheless,
it is possible to take the usual thermodynamical limit and, consequently, to use safely
411
the usual thermodynamical tools by first regularizing the long distance behavior of the
gravitational potential, introducing a very large screening length. The system is then
thermodynamically stable and the thermodynamical limit does exist. Afterwards, one can
study the limit in which the screening length tends to infinity [8].
Acknowledgements
I thank P.-H. Chavanis and O. Fliegans for some interesting discussions. This work has
been carried out with the financial support of a Ramn y Cajal contract of Ministerio de
Ciencia y Tecnologa of Spain.
References
[1] H.J. de Vega, N. Snchez, Phys. Lett. B 490 (2000) 180;
H.J. de Vega, N. Snchez, Nucl. Phys. B 625 (2002) 409;
H.J. de Vega, N. Snchez, Nucl. Phys. B 625 (2002) 460.
[2] T. Padmanabhan, Phys. Rep. 188 (1990) 285.
[3] M. Kiessling, J. Stat. Phys. 55 (1989) 203.
[4] D. Ruelle, Helv. Phys. Acta 36 (1963) 183;
M.E. Fisher, Arch. Rational Mech. Anal. 17 (1964) 377;
J. van der Linden, Physica 32 (1966) 642;
J. van der Linden, Physica 38 (1968) 173;
J. van der Linden, P. Mazur, Physica 36 (1967) 491.
[5] V.A. Antonov, Vestnik Leningrad. Gos. Univ. 7 (1962) 135;
D. Lynden-Bell, R. Wood, Mon. Not. R. Astron. Soc. 138 (1968) 495;
W. Thirring, Z. Phys. 235 (1970) 339.
[6] P.-H. Chavanis, Phys. Rev. E 65 (2002) 056123;
P.-H. Chavanis, I. Ispolatov, Phys. Rev. E 66 (2002) 036109.
[7] E. Follana, V. Laliena, Phys. Rev. E 61 (2000) 6270;
I. Hernquist, N. Katz, Astrophys. J. Suppl. 70 (1989) 419;
J. Sommer-Larsen, H. Vedel, U. Hellsten, Astrophys. J. 500 (1998) 610.
[8] V. Laliena, astro-ph/0202448.
Nuclear Physics B 668 [FS] (2003) 415446

Jost solutions and quantum conserved quantities of

an integrable derivative nonlinear Schrdinger model
B. Basu-Mallick, Tanaya Bhattacharyya
Theory Group, Saha Institute of Nuclear Physics, 1/AF Bidhan Nagar, Kolkata 700 064, India
Received 14 April 2003; received in revised form 25 June 2003; accepted 27 June 2003
Abstract
We study differential and integral relations for the quantum Jost solutions associated with an
integrable derivative nonlinear Schrdinger (DNLS) model. By using commutation relations between
such Jost solutions and the basic field operators of DNLS model, we explicitly construct first few
quantum conserved quantities of this system including its Hamiltonian. It turns out that this quantum
Hamiltonian has a new kind of coupling constant which is quite different from the classical one. This
modified coupling constant plays a crucial role in our comparison between the results of algebraic
and coordinate Bethe ansatz for the case of DNLS model. We also find out the range of modified
coupling constant for which the quantum N-soliton state of DNLS model has a positive binding
energy.
PACS: 11.10.Lm; 11.30.-j; 02.30.Ik; 03.65.Fd
Keywords: Derivative nonlinear Schrdinger model; Jost solution; YangBaxter equation; Algebraic Bethe
ansatz
1. Introduction
Conserved quantities associated with quantum integrable models in low dimensions
have recently found interesting applications in many topics of physics like exact
calculations of transport properties in mesoscopic electronic devices and distribution of
energy level spacing in quantum chaotic systems [1,2]. In the framework of quantum
inverse scattering method (QISM), one can formally generate such conserved quantities
E-mail addresses: biru@theory.saha.ernet.in (B. Basu-Mallick), tanaya@theory.saha.ernet.in
(T. Bhattacharyya).
doi:10.1016/S0550-3213(03)00570-4
416
B. Basu-Mallick, T. Bhattacharyya / Nuclear Physics B 668 [FS] (2003) 415446
by expanding the trace of monodromy matrix in a power series of spectral parameter

[36]. The Lax operator associated with monodromy matrix satisfies quantum Yang
Baxter equation (QYBE). As a result, all of these quantum conserved quantities commute
among themselves. Thus, for constructing a quantum integrable field model or spin chain,
it is natural to start with a suitable quantum Lax operator which satisfies QYBE and find
out corresponding conserved quantities including the Hamiltonian.
However, explicit construction of these conserved quantities in terms of basic quantum
field or spin operators often turns out to be a challenging task which has inspired the
application of several ingenious techniques. For example, in the case of one-dimensional
quantum integrable spin chains like Heisenberg model, supersymmetric tJ model and
Hubbard model, one can explicitly construct the conserved quantities in a recursive way
by using appropriate ladder operators [710]. While dealing with (1 + 1)-dimensional
classically integrable field theoretical systems like nonlinear Schrdinger (NLS) model,
it is again possible to explicitly construct the conserved quantities in a recursive way by
solving corresponding Riccati equations. However, this recursive method of finding the
conserved quantities does not usually work for quantum integrable field models, where the
presence of normal ordering might lead to nonuniformness in the asymptotic expansion of
monodromy matrix in powers of spectral parameter. As a result, it may not be possible
to obtain all quantum conserved quantities simply as normal ordered versions of the
corresponding classical conserved quantities [1113]. Fortunately, however, this problem
does not occur for the case of some lower conserved quantities of quantum NLS model,
which are generated by first few terms in the asymptotic expansion of monodromy matrix
[14]. Consequently, conserved quantities associated with number of particles, momentum
as well as Hamiltonian of the quantum NLS model can be obtained just as normal ordered
versions of the corresponding classical conserved quantities. The Hamiltonian of quantum
integrable sine-Gordon model can also be obtained in a similar way from the corresponding
classical Hamiltonian [3,5].
Even though the Hamiltonians of quantum integrable field models usually coincide
with the normal ordered versions of the corresponding classical Hamiltonians, there is no
guarantee that this thumb rule will always be obeyed. The main purpose of the present
article is to construct the quantum Hamiltonian of a derivative nonlinear Schrdinger
(DNLS) model through the corresponding Lax operator and explore how this quantum
Hamiltonian is related to its classical counterpart. In this context it may be noted that there
exist two variants of classically integrable DNLS model in 1 + 1 dimension [15,16], which
have found applications in physical systems like circularly polarized nonlinear Alfven
waves in a plasma [17,18]. However, only one among these variants of DNLS model is
known to be associated with an ultralocal Poisson bracket (PB) structure which is very
suitable for quantization through QISM [19,20]. The equation of motion for such classical
DNLS model is given by [16]
it (x, t) + xx (x, t) 4i (x, t)(x, t)x (x, t) = 0,
(1.1)
where t /t, x /x, xx 2 /x 2 and is a real parameter representing the

strength of the nonlinear interaction term. The Lax operator related to this DNLS model
may be written in the form [19,21]

(x)(x) 2 /4
U (x, ) = i
(x)

(x)
,
(x)(x) + 2 /4
417
(1.2)
where denotes the spectral parameter and (x), (x) represent field variables at
some fixed time (which is suppressed here and all along in the following). By solving
the Riccati equation associated with Lax operator (1.2), one can explicitly construct the
conserved quantities for this DNLS model in a recursive way. The first few among such
infinite number of classical conserved quantities, representing the mass, momentum and
Hamiltonian of the DNLS system, respectively, are given by [19]
+
(x)(x) dx,
N=
(1.3a)
P = i
+
(x)x (x) dx,
(1.3b)
H=
+

(x)xx (x) + i 2 (x)x 2 (x) dx.
(1.3c)
The field variables appearing in the Lax operator (1.2) obey the following equal time
PB structure: {(x), (y)} = { (x), (y)} = 0, {(x), (y)} = i(x y). With the
help of this ultralocal PB structure, it can be shown that the Lax operator (1.2) satisfies
classical YangBaxter equation. As a result, infinite number of conserved quantities
associated with DNLS model (1.1) yield vanishing PB relations among themselves [19].
This fact establishes the classical integrability of DNLS model (1.1) in the Liouville sense.
It is remarkable that the integrability property of the above mentioned classical DNLS
model can be preserved even after quantization. In this quantized version of DNLS model,
the basic field operators satisfy equal time commutation relations given by

(x), (y) = h(x
y),
(x), (y) = (x), (y) = 0,
(1.4)
h being the Plancks constant. The corresponding vacuum state is defined through the
relation: (x)|0 = 0. The most natural way of constructing such quantum integrable
DNLS model, possessing infinite number of mutually commuting conserved quantities,
is to first find out the quantum analogue of classical Lax operator (1.2) which would
satisfy the QYBE. However, it can be easily shown that QYBE is not satisfied if the
normal ordered version of classical Lax operator (1.2) is directly chosen as the quantum
Lax operator of DNLS model. The correct form of this quantum Lax operator, satisfying
QYBE in continuum, is given by [21]

f (x)(x) 2 /4
(x)
,
Uq (x, ) = i
(1.5)
(x)
g (x)(x) + 2 /4
418

i/2
i/2
e
e
where f = cos
/2 , g = cos /2 and a real parameter (/2 < /2) which is
uniquely determined through the relation
sin = h .
(1.6)
Thus the quantum Lax operator (1.5) depends not only on the parameter , but also on the
Plancks constant h . It is clear from Eq. (1.6) that, for any fixed value of , 0 limit is
essentially equivalent to h 0 limit. Since f and g at 0 limit, the quantum
Lax operator (1.5) reproduces the classical Lax operator (1.2) at h 0 limit. With the help
of Lax operator (1.5) or its lattice version [19,20], one can easily construct the monodromy
matrix of quantum DNLS model in continuum. Quantum conserved quantities can be
defined formally through the diagonal elements of this monodromy matrix, by expanding
them in the power series of spectral parameter. By applying algebraic Bethe ansatz to such
formally defined quantum conserved quantities, one can derive their exact eigenfunctions
and eigenvalues for scattering as well as bound soliton states [19,21]. Moreover, one can
also construct the reflection operators for the DNLS model satisfying the Zamolodchikov
Faddeev algebra and find out the S-matrix for the two body scattering [21].
In spite of these studies on quantum DNLS model, the problem of explicitly
constructing its conserved quantities in terms of basic field operators like (x) and
(x) has not been addressed so far. In particular, it is not known whether, in analogy
with the quantum NLS model and sine-Gordon model, the Hamiltonian of quantum
DNLS model can also be obtained as the normal ordered version of the corresponding
classical Hamiltonian (1.3c). The explicit form of such quantum Hamiltonian would clearly
play a central role in interpreting various properties of this field model in the language
of associated quantum mechanical many-particle system. In this context it should be
observed that, if the normal ordered version of classical Hamiltonian (1.3c) is projected
on an N -particle Hilbert space, that would yield an N -particle bosonic system interacting
through the derivative -function potential [22,23], where represents the strength of the
interaction. Eq. (1.6) however imposes a restriction on the value of this coupling constant as
| |1/h . Thus it is evident that, if the normal ordered version of the classical Hamiltonian
(1.3c) represents the quantum Hamiltonian of DNLS model, the corresponding N -particle
bosonic system cannot be solved through QISM for | | > 1/h . On the other hand, it is
known that this N -particle bosonic system with derivative -function interaction can be
solved exactly for any value of its coupling constant through the coordinate Bethe ansatz
[2224]. Thus one faces a rather curious limitation about the applicability of algebraic
Bethe ansatz to the case of quantum DNLS model.
It is clear that, some direct method of finding the explicit form of quantum Hamiltonian
associated with the Lax operator (1.5) of DNLS model may help us to resolve the
above mentioned problem. In this context, we recall a work by Case [11] where first
few conserved quantities of the quantum NLS model are explicitly constructed and their
spectra are also derived in the following way. At first, Jost solutions associated with the
Lax operator of quantum NLS system are considered. The scattering data, i.e., elements
of monodromy matrix, are identified with the Wronskians corresponding to these Jost
solutions. Subsequently it is proposed that the commutators between quantum conserved
quantities of the NLS model and Wronskians obey the so-called fundamental relation.
This relation can generate the spectra of all quantum conserved quantities in an algebraic
419
way. The explicit form of the first few quantum conserved quantities of NLS model are
obtained from the requirement of satisfying this fundamental relation.
The above mentioned way of constructing quantum conserved quantities and finding
their spectra is clearly different from the usual algebraic Bethe ansatz in QISM. However,
in complete analogy with QISM, finding an appropriate quantum Lax operator is the
starting point of Cases approach. So this approach gives us valuable insight about the
explicit form of quantum conserved quantities which can be obtained from the trace of
monodromy matrix in QISM. In this article we shall study quantum DNLS model through
this approach which is complimentary to QISM. In Section 2, we briefly recapitulate the
construction of quantum Lax operator of DNLS model through a variant of QISM which is
directly applicable to field theoretical systems and also discuss how the related conserved
quantities can be diagonalised through algebraic Bethe ansatz [21]. In Section 3 we use
the quantum Lax operator and monodromy matrix, obtained through QISM, for defining
the Jost solutions of DNLS model. It is surprisingly found that, in contrast to the case
of NLS model, differential equations satisfied by Jost solutions associated with boundary
conditions at x and x do not coincide with each other. Using the Wronskians
and some other bilinear functions of these Jost solutions, in Section 4 we propose the
fundamental relation for the DNLS model and derive the spectra for all conserved
quantities which would satisfy this relation. Here we also discuss how the conserved
quantities satisfying the above relation are related to the conserved quantities which are
formally defined in the framework of QISM. In Section 5, we discuss about the necessary
tools for finding out the explicit form of conserved quantities satisfying the fundamental
relation. In particular, we derive the commutation relations between the Wronskians and
basic field operators of the system. In Section 6, we construct the explicit form of first few
conserved quantities of the quantum DNLS model including its Hamiltonian. Interestingly,
it is found that the interaction part of this quantum Hamiltonian has a new kind of coupling
constant which is quite different from the classical one. Here we also derive the condition
on this coupling constant for which the quantum N -soliton state of DNLS model has a
positive binding energy. Section 7 is the concluding section.
2. Application of QISM to DNLS model

As mentioned earlier, the monodromy matrix plays a key role in formally generating
the quantum conserved quantities of DNLS model and in diagonalising those conserved
quantities through QISM. With the help of Lax operator (1.5), one can define the quantum
monodromy matrix of DNLS model on a finite interval as
x2
Txx12 () = :P exp
Uq (x, ) dx:,
(2.1)
x1
where P denotes the path ordering and the symbol : : denotes the normal ordering of
operators. It is evident that this monodromy matrix satisfies differential equations of the
420
form
x2
T () = :Uq (x2 , )Txx12 ():,
x2 x1
(2.2a)
x2
T () = :Txx12 ()Uq (x1 , ):.
x1 x1
(2.2b)
By using these differential equations and canonical commutation relations (1.4), it can be
shown that the direct product of two such quantum monodromy matrices satisfies QYBE
given by [21]
R(, )Txx12 () Txx12 () = Txx12 () Txx12 ()R(, ).
Here R(, ) is a (4 4) matrix with c-number elements like
1
0
0
0
0 s(, ) t (, ) 0
R(, ) =
,
0 t (, ) s(, ) 0
0
0
0
1
(2.3)
(2.4)
(qq )

i . It is mentioned earlier that
with t (, ) = 2 q
2 q 1 , s(, ) = 2 q2 q 1 and q = e
the real parameter , which is present both in Lax operator (1.5) and R-matrix (2.4), is
fixed through the relation (1.6). Consequently, QISM is applicable for quantum DNLS
model when the parameter satisfies a restriction given by | |1/h .
Next, by using the expression of Txx12 () in (2.1), we define the quantum monodromy
matrix on an infinite interval limit as
2
T () =
lim
x2 +
x1
where e(x, ) = e
T+ (x, ) =
T (x, ) =
e(x2 , )Txx12 ()e(x1 , ) = T+ (x, )T (x, ),
i2 x
4 3
(2.5)
and
lim e(x2 , )Txx2 (),
x2 +
lim T x ()e(x1 , ).
x1 x1
(2.6a)
(2.6b)
Taking into account that the quantum Lax operator (1.5) obeys certain symmetry properties
[21] and assuming to be a real parameter, one can express the quantum monodromy
matrix (2.5) in a symmetric form given by

A() B ()
T () =
(2.7)
,
B()
A ()
and find that these operator valued elements satisfy relations like A() = A(), B() =
B(). Moreover, it is easy to show that these elements act on the vacuum state as:
A()|0 = |0, B()|0 = 0. With the help of Eqs. (2.3) and (2.5), one may now obtain
QYBE for the quantum monodromy matrix on an infinite interval [21]. Expressing this
QYBE in elementwise form, we finally obtain

A(), A() = 0,
421
(2.8a)

A(), A () = 0,
(2.8b)

B(), B() = 0,
(2.8c)
A()B () =
B()A() =
2 q 2 q 1
B ()A(),
2 2 i%
2 q 2 q 1
A()B(),
2 2 i%

B()B () = (, )B ()B() + 4 h 2 2 A ()A(),
(2.8d)
(2.8e)
(2.8f)
where

8h 2 2 2 2
4h 2 2 2 2
(, ) = 1 +
2
.
2
( 2 i%)(2 2 + i%)
(2 2 )
Due to Eq. (2.8a) it follows that all operator valued coefficients occurring in the
expansion of ln A() in powers of must commute among themselves. Consequently,
ln A() may be treated as the generator of conserved quantities for the quantum integrable
DNLS model. For the purpose of diagonalising these quantum conserved quantities, we
first notice that the commutation relation (2.8f) contains product of singular functions
(2 2 i%)1 (2 2 + i%)1 , which does not make sense at the limit . As
a result, actions of operators B (), B() are not well defined on the Hilbert space [4,25]
and generate states which are not normalised on the -function. However, it is well known
that, one can avoid this type of problem in the case of NLS model by considering the
quantum analogue of classical reflection operators [3,26]. So, for the case of DNLS model
also we consider a reflection operator given by

1
R () = B () A ()
(2.9)
and its adjoint R(). By using Eqs. (2.8a)(2.8f), we find that such reflection operators
satisfy well defined commutation relations like [21]
R ()R () = S 1 (, )R ()R (),
R()R() = S 1 (, )R()R(),

R ()R() = S(, )R()R () + 4 h 2 2 2 ,
(2.10)
where
S(, ) =
2 q 2 q 1
.
2 q 1 2 q
(2.11)
422
It is evident that these commutation relations are encoded in a form of Zamolodchikov

Faddeev algebra [3,27] and S(, ) (2.11) represents the nontrivial S-matrix element of
two-body scattering between the related quasi-particles. It is easy to check that this S(, )
satisfies the relations
S 1 (, ) = S(, ) = S (, ),
(2.12)
and remains nonsingular at the limit . As a result, the action of operators like
R () on the vacuum would produce well defined states which can be normalised on the
-function.
The commutation relation between A() and R () may be derived by using Eqs. (2.8b)
and (2.8d) as
A()R () =
2 q 2 q 1
R ()A().
2 2 i%
(2.13)
By applying the above commutation relation and also using A()|0 = |0, it can be shown
that

N 2

r q 2 q 1
A()|1 , 2 , . . . , N =
(2.14)
|1 , 2 , . . . , N ,
2r 2 i%
r=1
where |1 , 2 , . . . , N
1 )R (2 ) R (N )|0 and j s are all distinct real or
pure imaginary numbers. Thus the states |1 , 2 , . . . , N diagonalise the generator of
conserved quantities for the quantum DNLS model. However, by using Eq. (2.14), one
finds that the eigenvalues corresponding to different expansion coefficients of ln A()
would be complex quantities in general. To make the eigenvalues real, we define another
as
operator A()
through the relation: A()
A(ei/2 ) and expand ln A()
R (
ln A()
=

iCn
n=0
2n
(2.15)
With the help of Eqs. (2.14) and (2.15), one can easily find out the real eigenvalues
associated with all Cn s:
C0 |1 , 2 , . . . , N = N|1 , 2 , . . . , N ,

N

2
2n
j |1 , 2 , . . . , N ,
Cn |1 , 2 , . . . , N = sin(n)
n
(2.16a)
(2.16b)
j =1
where n 1. Till now it is assumed that j s are some real or pure imaginary parameters,
for which |1 , 2 , . . . , N represents a scattering state. We can also construct the quantum
soliton states or bound states for DNLS model by choosing complex values of j given by
[19,21]

N +1
j ,
j = exp i
(2.17)
2
423
where is a real or pure imaginary parameter and j [1, 2, . . . , N]. Similar to the case
of scattering states, one can find out the real eigenvalues corresponding to all Cn s for these
quantum soliton states of DNLS model.
Thus, by applying QISM, it is possible to obtain the exact eigenvalues as well as
eigenstates for the quantum conserved quantities of DNLS model which are defined
formally through the expansion (2.15). However, the important problem of expressing
these conserved quantities through basic field operators like (x) and (x) has not been
explored so far. In analogy with the classical case, C0 , C1 and C2 should be related to the
number operator, momentum operator and the Hamiltonian of the quantum DNLS model,
respectively. So it should be particularly interesting to find out the explicit form of these
first three conserved quantities. To this end, we shall study quantum Jost solutions of the
DNLS model.
3. Jost solutions of quantum DNLS model

It may be recalled that the differential equations satisfied by the Jost solutions of
quantum NLS model are defined through the corresponding Lax operator [11]. As a result
all Jost solutions of NLS model, defined through boundary conditions at x + or
x , satisfy exactly the same form of coupled differential equations. At present,
however, we shall not directly use the Lax operator (1.5) for obtaining the differential
equations associated with Jost solutions of quantum DNLS model. Instead, we shall
identify appropriate elements of the matrices T+ (x, ) (2.6a) and T (x, ) (2.6b) as Jost
solutions corresponding to boundary conditions at x + and x , respectively.
The differential equations satisfied by T+ (x, ) and T (x, ) will give us in a natural
way the differential equations for Jost solutions corresponding to boundary conditions at
x + and x , respectively. It will turn out that, contrary to the case of NLS
model, quantum Jost solutions of DNLS model associated with boundary conditions at
x + and x satisfy different types of coupled differential equations. Due to
Eq. (2.5), the elements of monodromy matrix (2.7) can be expressed as Wronskians of such
Jost solutions.
To proceed in the above mentioned way, let us express T (x, ) (2.6b) in elementwise
form as

1 (x, ) 1 (x, )
T (x, ) =
(3.1)
,
2 (x, ) 2 (x, )

) 1 (x,) are two Jost solutions corresponding
and
(x,
where (x, ) 1 (x,)
2 (x,)
2 (x,)
to boundary conditions at x . Due to Eq. (2.2a), T (x, ) satisfies a differential
equation given by
x T (x, ) = :Uq (x, )T (x, ):.
(3.2)
Substituting the explicit form of T (x, ) (3.1) to (3.2), we find that the components of
) satisfy exactly the same form of coupled differential equations given
(x, ) and (x,
424
by
i2
1 (x, ) + if (x)1 (x, )(x) + i (x)2 (x, ),
4
i2
x 2 (x, ) =
(3.3)
2 (x, ) ig (x)2 (x, )(x) + i1 (x, )(x),
4

(x,)
(x,)
(x,)
. Thus (x, )
where 1 (x,) may be chosen either as 1 (x,) or as 1
2
2 (x,)

2
1 (x,)
represents the general form of Jost solutions defined through boundary conditions
2 (x,)
at x . Next, by taking the x limit of T (x, ) (2.6b), we obtain
x 1 (x, ) =
T (x, ) e
i2 x
4 3
(3.4)
Substituting the matrix form of T (x, ) (3.1) to the relation (3.4), we obtain the boundary
) as
conditions associated with Jost solutions (x, ) and (x,

i2 x
1 (x, ) x 10 e 4
,
(3.5)
i2 x
2 (x, )
20 e 4
).
where 10 = 1, 20 = 0 for (x, ) = (x, ) and 10 = 0, 20 = 1 for (x, ) = (x,
Using the boundary conditions (3.5), we can convert the differential equations (3.3) to
their integral forms as
i2 x
1 (x, ) = 10 e 4
x

i2
+i
dz e 4 (zx) f (z)1 (z, )(z) + (z)2 (z, ) ,
(3.6a)
2 (x, ) = 20 e
i2 x
4
x
+i
dz e
i2
4 (xz)

g (z)2 (z, )(z) + 1 (z, )(z) .
(3.6b)
With the help of these integral relations it is easy to show that, for the case of real , the
) are related as
components of Jost solutions (x, ) and (x,
1 (x, ) = 2 (x, ),
2 (x, ) = 1 (x, ).
(3.7)
Next we try to find out the differential equations for the Jost solutions corresponding to
boundary conditions at x +. To this end, we express T+ (x, ) (2.6a) in elementwise
form as

2 (x, ) 1 (x, )
T+ (x, ) =
(3.8)
,
2 (x, ) 1 (x, )

(x,)
(x,)
) 1 (x,) represent two Jost solutions correwhere (x, ) 1 (x,) and (x,
2
2
sponding to boundary conditions at x +. Due to relation (2.2b), T+ (x, ) satisfies
425
a differential equation given by

x T+ (x, ) = :T+ (x, )Uq (x, ):.
(3.9)
Substituting the elementwise form of T+ (x, ) (3.8) to (3.9), it is easy to see that the
components of (x, ) and (x,
) satisfy exactly the same form of coupled differential

equations given by
i2
1 (x, ) + ig (x)1 (x, )(x) + i (x)2 (x, ),
4
i2
x 2 (x, ) =
2 (x, ) if (x)2 (x, )(x) + i1 (x, )(x),
4
x 1 (x, ) =
(3.10)

(x,)
(x,)
where 1 (x,)
may be chosen as either 1 (x,) or 1 (x,) . Thus (x, ) 1 (x,)
(x,)
(x,)
2
2
2
2
represents the general form of Jost solutions defined through boundary conditions at
x +. Next, by taking the x + limit of T+ (x, ) (2.6a), we obtain
x+
T+ (x, ) e
i2 x
4 3
(3.11)
Substituting the explicit form of T+ (x, ) (3.8) to the above relation, it is easy to find out
the boundary conditions associated with Jost solutions (x, ) and (x,
) as

2

i x
1 (x, ) x+ 10 e 4
,
(3.12)
i2 x
2 (x, )
20 e 4
where 10 = 0, 20 = 1 for (x, ) = (x, ) and 10 = 1, 20 = 0 for (x, ) = (x,
).
Using the boundary conditions (3.12), we can convert the differential equations (3.10) to
their integral forms as
i2 x
1 (x, ) = 10 e 4

i2
i dz e 4 (zx) g (z)1 (z, )(z) + (z)2 (z, ) ,
(3.13a)
2 (x, ) = 20 e
i2 x
4
dz e
i2
4 (xz)

f (z)2 (z, )(z) + 1 (z, )(z) .
(3.13b)
By using these integral relations it is easy to show that, for the case of real , the
components of Jost solutions (x, ) and (x,
) are related as
1 (x, ) = 2 (x, ),
1
2 (x, ) = 1 (x, ).
(3.14)
Comparing Eqs. (3.10) and (3.3), we notice that quantum Jost solutions of DNLS model,
associated with boundary conditions at x + and x , satisfy two different sets
426
of coupled differential equations. These two sets of differential equations are related to
each other through an interchange of f and g. However, since both f and g coincide with
the coupling constant at h 0 limit, Eqs. (3.3) and (3.10) have an identical form at this
classical limit. It may also be observed that, due to vanishing boundary condition on the
basic field variables, Eqs. (3.3) and (3.10) have the same asymptotic form at |x| limit.
Now we want to express the elements of quantum monodromy matrix (2.7) in terms
of Jost solutions as obtained above. To this end, we substitute the elementwise form of
T (x, ) (3.1) and T+ (x, ) (3.8) to Eq. (2.5) and compare it with (2.7). In this way, we
obtain
A() = 2 (x, )1 (x, ) 1 (x, )2 (x, ),
(3.15a)
A () = 2 (x, ) 1 (x, ) 1 (x, ) 2 (x, ),
(3.15b)
B() = 2 (x, )1 (x, ) 1 (x, )2 (x, ),
(3.15c)
1
1
B () = 2 (x, ) 1 (x, ) + 1 (x, ) 2 (x, ).
(3.15d)
Since the l.h.s. of Eqs. (3.15a)(3.15d) do not depend at all on the variable x, the r.h.s.
of these equations should also be independent of this variable (in spite of its explicit
appearance). By taking x + or x limit in the r.h.s. of Eqs. (3.15a)(3.15d)
and using boundary conditions (3.5) or (3.12), respectively, we obtain
A() = lim e
i2
4
i2
4 x
i2
4 x
B () =
i2
4 x
x+
A () = lim e
B() = lim e
2 (x, ) = lim e
1 (x, ) = lim e
1 (x, ),
(3.16a)
i2
4
(3.16b)
x+
2 (x, ) = lim e
x+
i2
4 x
2 (x, ),
2 (x, ),
i2
i2
1
1
lim e 4 x 1 (x, ) =
lim e 4 x 1 (x, ).
x
x+
(3.16c)
(3.16d)
Next, let us define the quantum Wronskian associated with the general form of Jost
solutions (x, ) and (x, ) as
, (x, ) = 2 (x, )1 (x, ) 1 (x, )2 (x, ).
(3.17)
Comparing Eqs. (3.15) and (3.17) for all possible choice of (x, ) and (x, ), we find
that
A() = , (x, ),
A () = ,
(x, ),
B() = , (x, ),
1
B () = ,
(x, ).
(3.18)
427
Thus the quantum Wronskian (3.17) represents all elements of the monodromy matrix (2.7)
in a general form.
Since the elements of monodromy matrix (2.7) do not depend on the variable x, the
quantum Wronskian (3.17) must also be independent of this variable. However, we should
be able to demonstrate this fact in a direct way by showing that , (x, ) has a vanishing
derivative with respect to the variable x. To this end, we consider a general type of
quantum integrable field model whose Jost solutions (x, ) and (x, ) satisfy differential
equations given by
x (x, ) = :L (x, )(x, ):,
x (x, ) = :L+ (x, ) (x, ):,
(3.19)
L (x, ) being some (2 2)-matrices with elements L

ij (x, ). As before, the quantum
Wronskian associated with this general case may be defined through Eq. (3.17). For the
sake of convenience, let us ignore at present the effect of normal ordering in Eq. (3.19) and
treat all quantum variables as commuting classical variables. In this way it can be easily
shown that, the derivative of Wronskian (3.17) with respect to the variable x will vanish if
the elements of L+ (x, ) and L (x, ) are related as
+
L
11 (x, ) = L22 (x, ),
+
L
ij (x, ) = Lij (x, ),
+
L
22 (x, ) = L11 (x, ),
(3.20)
where i = j . Thus it follows that, L+ (x, ) would coincide with L (x, ) when it satisfies
the traceless condition. Since the Lax operators of quantum NLS model and almost all
other integrable systems satisfy this traceless condition, L+ (x, ) and L (x, ) coincide
for these cases. However, the quantum Lax operator (1.5) of DNLS model does not satisfy
this condition. Consequently, the corresponding L+ (x, ) and L (x, ) matrices should
not coincide with each other. Expressing Eqs. (3.3) and (3.10) in matrix form, we find that
L (x, ) matrix of DNLS model is same as Uq (x, ) (1.5) and L+ (x, ) may be obtained
from Uq (x, ) by interchanging f and g. Since these matrices satisfy the relation (3.20),
we may conclude that the Wronskian (3.17) of DNLS model has a vanishing derivative
with respect to the variable x. A more rigorous proof about the coordinate independence
of this Wronskian, taking into account the noncommutative nature of quantum operators,
will be given in Section 5 of this article.
4. Spectrum generating algebra for DNLS model

By following the approach of Ref. [11], here we shall propose the fundamental relation
for the DNLS model and explore its connection with the spectrum generating algebra. In
analogy with the quantum Wronskian (3.17), let us define another operator associated with
the Jost solutions of DNLS model as
, (x, ) = 2 (x, )1 (x, ) + 1 (x, )2 (x, ).
(4.1)
This , (x, ) and quantum Wronskian (3.17) are two basic ingredients which are needed
for defining the fundamental relation of DNLS model. Now we propose that, the quantum
conserved quantities (In ) of DNLS model would annihilate the vacuum state and obey the
428
fundamental relation given by

h 2n
In , , () = n+1
2
+
y , (y, ) dy
2n
h
, (+, ) , (, ) ,
(4.2)
2n+1
where n is any non-negative integer. Since , (x, ) (3.17) does not depend on the
coordinate x, we have suppressed this variable in the l.h.s. of above relation.
Next, we shall discuss how the fundamental relation (4.2) leads to the spectrum
generating algebra for all quantum conserved quantities of DNLS model. To this end, it
is needed to find out the x limit of , (x, ). For all possible choices of and ,
, (x, ) (4.1) may be explicitly written as
=
, (x, ) = 2 (x, )1 (x, ) + 1 (x, )2 (x, ),
(4.3a)
,
(x, ) = 2 (x, ) 1 (x, ) + 1 (x, ) 2 (x, ),
(4.3b)
, (x, ) = 2 (x, )1 (x, ) + 1 (x, )2 (x, ),
(4.3c)
,
(x, ) = 2 (x, ) 1 (x, ) + 1 (x, ) 2 (x, ).
(4.3d)
Substituting the asymptotic forms of Jost solutions (3.5), (3.12) to the x limits of
relations (4.3a)(4.3d) and subsequently using (3.16a)(3.16d), we find that
, (, ) = A(),
, (, ) = B(),
,
(, ) = A (),
,
(, ) = B ().
Inserting (4.4) to the fundamental relation (4.2) and also using (3.18), we get

In , A() = 0,
(4.4)
(4.5a)

In , A () = 0,
(4.5b)

h 2n
In , B() = n B(),
2
(4.5c)
h
2n
In , B () = n B ().
(4.5d)
2
With the help of Eqs. (4.5b) and (4.5d), we can find out the commutation relation between
the quantum conserved quantities and reflection operators (2.9) as

h
2n
In , R () = n R ().
2
(4.6)
429
By using the above commutation relation and assuming that In s annihilate the vacuum
state, it is easy to show that these conserved quantities satisfy eigenvalue equations given
by

N
h 2n
j |1 , 2 , . . . , N ,
In |1 , 2 , . . . , N =
(4.7)
2n
j =1
where |1 , 2 , . . . , N R (1 )R (2 ) R (N )|0. Consequently, the commutation

relation (4.6) may be treated as the spectrum generating algebra for the quantum conserved
quantities of DNLS model.
It should be noted that, eigenstates of In are same as Bethe states which we have already
used in the framework of QISM to diagonalise the quantum conserved quantities appearing
in the expansion (2.15). Thus, it is natural to expect a connection between these In s and
the conserved quantities which are formally defined through the expansion (2.15). For
establishing this connection, let us assume that the Bethe states |1 , 2 , . . . , N represent
a complete set of states in the corresponding Hilbert space. Thus two operators would
coincide if they can be simultaneously diagonalised through these complete set of states
and their eigenvalues always match with each other. Comparing Eq. (4.7) with (2.16a) and
(2.16b), it is easy to find that
C0 =
I0 ,
h
Cn =
2n+1
sin(n)In .
nh
(4.8)
Substituting (4.8) to (2.15), we obtain the expansion of ln A()

in terms of In s as
ln A()
=
i 2n+1
i
I0 +
sin(n)In .
h
h
n2n
n=1
(4.9)
We can also define conserved quantities for DNLS model through reflection operators
as
In

2n1 R ()R() d.
2n+1
(4.10)
By using the commutation relations between reflection operators (2.10), which are derived
in the framework of QISM, we obtain

In , Im = 0,
(4.11a)

h
2n
In , R () = n R ().
(4.11b)
2
With the help of (4.11b), one can easily show that |1 , 2 , . . . , N are eigenfunctions of
In with exactly the same eigenvalues as found in the case of In and conclude that In = In .
Consequently, Eq. (4.10) yields an expression of In through the reflection operators of
DNLS model.
Next our aim is to explicitly construct first few quantum conserved quantities of
DNLS model which would satisfy the fundamental relation (4.2). Necessary tools for such
construction will be discussed in the following section.
430
5. Commutation relations between the quantum Wronskian and basic field

operators
Since the quantum Wronskian (3.17) of DNLS model is expressed as a bilinear function
of Jost solutions, at first we consider the commutation relations between these Jost
solutions and basic field operators of the system. In analogy with the case of NLS model
[11], one may take the arguments of Jost solutions and field operators at exactly the same
space point and try to evaluate their commutation relations (e.g., commutators of the
form [i (x, ), (x)]). By using the integral relations (3.6) and canonical commutation
relations (1.4), it can be easily checked
x that the commutators [i (x, ), (x)] lead to
indeterminant integrals of the form (x z)F (z) dz, where F (z) is some function
of z. Such indeterminant integrals, which
also appear in the case of NLS model, may be
x
fixed through a convention given by (x z)F (z) dz = 12 F (x) [11]. However, as will
be explained shortly, the above mentioned convention of fixing indeterminant integrals
would lead to the violation of Jacobi identity in the case of DNLS model. So, instead of
trying to calculate commutators of the form [i (x, ), (x)], at present we shall study
commutators like [i (y, ), (x)] in the limit y x.
To begin with, let us consider the commutators [i (y, ), (x)] and [i (y, ), (x)]
in the region y < x. For this case, all fields (z), (z) appearing in the integral
relations (3.6a), (3.6b) would commute with (x), (x). Consequently, we obtain
[i (y, ), (x)] = [i (y, ), (x)] = 0 in the region y < x. The y x limit of these
commutation relations may be expressed in the form

i (x , ), (x) = i (x , ), (x) = 0,
(5.1)
where the notation i (x , ) lim%0+ i (x %, ) is introduced and % 0+ limit is
taken after evaluating all commutators.
Next, we consider the commutators [i (y, ), (x)] and [i (y, ), (x)] in the region
y > x. For this case, however, Eqs. (3.6a), (3.6b) lead to rather complicated integral
relations which are difficult to solve in a closed form for arbitrary values of x and y.
So, for the sake of convenience, we shall try to evaluate such commutators only at
the limit y x. In analogy with the previous case, we introduce a notation given
by i (x , ) lim%0+ i (x + %; ). We are interested in calculating commutators like
[i (x , ), (x)] lim%0+ [i (x + %, ), (x)], where % 0+ limit should be taken at
the final stage after evaluating all commutation relations. By using integral relations (3.6a),
(3.6b) and canonical commutation relations (1.4) we obtain

1 (x , ), (x) = i h f1 (x , )(x) i h 2 (x , ),
(5.2a)

1 (x , ), (x) = i h f (x)1 (x , ),
(5.2b)

2 (x , ), (x) = i h g2 (x , )(x),
(5.2c)

2 (x , ), (x) = i h g (x)2 (x , ) + i h 1 (x , ).
(5.2d)
The details of derivation for one of the above commutation relations is given in
Appendix A. It is clear from relations (5.1) and (5.2) that the commutators [i (y, ), (x)]
431
and [i (y, ), (x)] are discontinuous at the point y = x. By repeatedly applying the
commutation relations (5.2), we easily obtain

1 (x , ), 2 (x)

= h f (h f 2i)1 (x , ) 2 (x) i h 2 + i h (f g) 2 (x , )(x),
(5.3a)

2

1 (x , ), 2 (x) = hf
(2i hf
) (x)1 (x , ),
(5.3b)

2 (x , ), 2 (x) = h g(2i + h g)2 (x , ) 2 (x),
(5.3c)

2 (x , ), 2 (x) = h g(2i + h g) 2 (x)2 (x , )

+ i h 2 + i h (f g) (x)1 (x , ).
(5.3d)
We would like to make a comment at this point. Since the integral relations of
i (x %, ) and i (x + %, ) coincide with each other at the limit % 0+, one may
say that the operators i (x , ) and i (x , ) are same in the weak sense. However,
we have already observed that the commutators [i (y, ), (x)] and [i (y, ), (x)]
are discontinuous at the point y = x. As a result, operators of the form i (x, )
i (x , ) i (x , ) yield nontrivial commutation relations with (x) and (x). Thus,
borrowing a terminology from the theory of constrained Hamiltonian systems [28], we
may say that the operators i (x , ) and i (x , ) differ from each other in the strong
sense. While deriving commutation relations like (5.2) in Appendix A, we have neglected
some operators which become trivial in the weak sense at % 0 limit. This procedure
does not affect the validity of relations (5.2) in the weak sense. However, it is reasonable
to ask whether the relations (5.2) are also valid in the strong sense. To investigate this
point, one may try to evaluate commutators like [i (x , ), 2 (x)] and [i (x , ), 2 (x)]
from the first principles. This can be achieved with the help of integral relations (3.6a),
(3.6b) and canonical commutation relations (1.4), by evaluating at first the commutators
[i (z, ), (y)(x)] and [i (z, ), (y) (x)] in the region z > y > x and taking
y, z x limit at the final stage. One can verify that such a procedure will exactly reproduce
the relations (5.3), which are obtained through repeated applications of the commutation
relations (5.2). This fact suggests that the commutation relations (5.2) are valid not only in
the weak sense, but also in the strong sense.
Next, we try to evaluate commutation relations between basic field operators and
Jost solutions defined through boundary conditions at x +. At first, we consider
the commutators [i (y, ), (x)] and [i (y, ), (x)] in the region y > x. For this
case, all fields (z), (z) contained in the integral relations (3.13a), (3.13b) would
commute with (x), (x). As a result, we get trivial relations like [i (y, ), (x)] =
[i (y, ), (x)] = 0 in the region y > x. The y x limit of these commutation relations
may be expressed in the form

i (x , ), (x) = i (x , ), (x) = 0,
(5.4)
where i (x , ) lim%0+ i (x + %, ). Next, we consider the commutators [i (y, ),
(x)] and [i (y, ), (x)] in the region y < x. However, it is difficult to find out these
432
commutators in a closed form for arbitrary values of x and y. So, we shall evaluate such
commutators only at the limit y x. By using integral relations (3.13a), (3.13b) and
canonical commutation relations (1.4), we obtain

1 (x , ), (x) = i h g1 (x , )(x) + i h 2 (x , ),
(5.5a)

1 (x , ), (x) = i h g (x)1 (x , ),
(5.5b)

2 (x , ), (x) = i h f 2 (x , )(x),
(5.5c)

2 (x , ), (x) = i h f (x)2 (x , ) i h 1 (x , ),
(5.5d)
(x , ) lim
where i
%0+ i (x %, ). It is clear from the relations (5.4) and (5.5) that the
commutators [i (y, ), (x)] and [i (y, ), (x)] are discontinuous at the point y = x.
By repeatedly applying the commutation relations (5.5), we also get

1 (x , ), 2 (x)

2

= h g(2i + hg)
(5.6a)
1 (x , ) (x) + i h 2 + i h (f g) 2 (x , )(x),

1 (x , ), 2 (x) = h g(2i + h g) 2 (x)1 (x , ),
(5.6b)

2 (x , ), 2 (x) = h f (h f 2i)2 (x , ) 2 (x),
(5.6c)
2 (x , ), 2 (x)

2

= h f (2i hf
) (x)2 (x , ) i h 2 + i h (f g) (x)1 (x , ).
(5.6d)
Till now we have derived all possible commutation relations between Jost solutions
and basic field operators, which will be needed for our calculation of quantum conserved
quantities. Next, we consider commutation relations between two Jost solutions associated
with different boundary conditions, i.e., commutators of the type [i (y, ), j (x, )] at
the limit y x. By using the integral relations (3.6), (3.13) and canonical commutation
relations (1.4), it can be shown that [i (x , ), j (x, )] = [i (x , ), j (x, )] = 0. Thus,
unlike the previous cases, the commutator [i (y, ), j (x, )] is continuous at the limit
y x. Consequently, by following the method of extension [4], one may define the
commutator [i (x, ), j (x, )] either as [i (x , ), j (x, )] or as [i (x , ), j (x, )].
For both of these cases, one obtains the trivial result given by

i (x, ), j (x, ) = 0.
(5.7)
Thus it is evident that, we can freely interchange the ordering of i (x, ) and j (x, ) in
the expressions of quantum Wronskian (3.17) and , (x, ) operator (4.1).
Next, we want to calculate the derivatives for bilinears of Jost solutions, i.e., quantities
like x (i (x, )j (x, )). By using Eqs. (3.3) and (3.10), it is easy to see that such a
derivative is given by the sum of few terms, each of which is a product of Jost solutions and
basic field operators with arguments corresponding to exactly the same space point. It is a
433
standard practice [4,11] to express these terms in a form so that the operator (x) ((x))
is always placed at the extreme left (right), while the ordering of the remaining factors
remains completely unchanged. For example, if the term i (x, )(x) (x)j (x, )
appears in a differential equation, it should be transformed to (x)i (x, )j (x, )(x).
For the purpose of expressing all terms in the above mentioned fashion, it is needed to
use the commutation relations between basic fields and Jost solutions associated with
exactly same space point, i.e., commutators of the form [i (x, ), (x)], [i (x, ), (x)],
[i (x, ), (x)] and [i (x, ), (x)]. We have commented earlier that, evaluation of
these commutators
x through integral relations (3.6) and (3.13) would lead to indeterminant
integrals like (x z)F (z) dz, where F (z) is some function of z. Similar to the case of
NLS model
x [11], one may now try to fix these indeterminant integrals through a convention
given by (x z)F (z) dz = 12 F (x). It can be easily checked that the above mentioned
way of fixing indeterminant integrals and calculating [i (x, ), (x)], [i (x, ), (x)] is
essentially same as defining these commutators as
1

i (x, ), (x) i (x , ) + i (x , ), (x) ,
2
1

i (x, ), (x) i (x , ) + i (x , ), (x) ,
(5.8)
2
evaluating them through the relations (5.1) and (5.2) and substituting the argument x in
place of x and x at the final stage. Similarly, one can calculate [i (x, ), (x)] and
[i (x, ), (x)], by defining them exactly like (5.8) and using the relations (5.4) and (5.5).
Explicit results for all of these commutation relations are given in Appendix B. However,
we find in Appendix B that, unlike the case of NLS model, these commutation relations
violate the Jacobi identity. Consequently, for the case of present DNLS model, it is not
meaningful to define commutation relations between Jost solutions and field operators with
arguments at exactly same space point through the prescription (5.8).
The above mentioned problem, which arises in the computation of x (i (x, )j (x, )),
can be bypassed through the method of extension [4]. According to this method, the
argument of one Jost solution is shifted by a small amount and 0 limit is taken after
evaluating all relevant commutation relations. The final result obtained in this way must be
independent of the sign of . By applying this method of extension, and using differential
equations (3.3), (3.10) as well as commutators (5.1), (5.2), (5.4), (5.5), we obtain

x 1 (x, )2 (x, )

= i (x)2 (x, )2 (x, ) + 1 (x, )1 (x, )(x) ,
(5.9a)

x 2 (x, )1 (x, )

= i (x)2 (x, )2 (x, ) + 1 (x, )1 (x, )(x) ,
(5.9b)

x 1 (x, )1 (x, )
i2
1 (x, )1 (x, ) + i(f + g) (x)1 (x, )1 (x, )(x)
2
+ i (x), (x, ),
(5.9c)
434

x 2 (x, )2 (x, )
=
i2
2 (x, )2 (x, ) i(f + g) (x)2 (x, )2 (x, )(x)
2
+ i, (x, )(x).
(5.9d)
Details of derivation for one of the above differential equations is given in Appendix C.
Using Eqs. (3.17) and (5.9a), (5.9b), we find that
x , (x, ) = 0.
(5.10)
Thus we are able to explicitly show that, the quantum Wronskian (3.17) remains
independent of the variable x even if the noncommutative nature of related operators are
taken into account. Taking advantage of this fact, we often use the notation , (), instead
of , (x, ), to denote the quantum Wronskian. We are also interested in computing the
derivative of , (x, ) operator (4.1), since it appears in the r.h.s. of the fundamental
relation (4.2). With the help of Eqs. (5.9a), (5.9b), we easily obtain

x , (x, ) = 2i (x)2 (x, )2 (x, ) + 1 (x, )1 (x, )(x) .
(5.11)
By using Eqs. (5.11) and (5.9c), (5.9d), one can further show that
(f + g)
x , (x, )
(x)x , (x, )(x) = , (x, ),
4
2
(5.12)

, (x, ) = (x)x 2 (x, )2 (x, ) x 1 (x, )1 (x, ) (x).
(5.13)
where
Finally, we try to find out commutation relations between the quantum Wronskian and
basic field operators. Since , (y, ) is shown to be independent of y, commutators
like [, (y, ), (x)] and [, (y, ), (x)] should not depend on the choice of
argument y. For the case of NLS model, such commutators are calculated for the
choice y = x [11]. However, we have already seen in Appendix B that this choice
leads to the violation of Jacobi identity for the case of DNLS model. So, instead of
choosing y = x, at present we shall calculate the commutators [, (y, ), (x)] and
[, (y, ), (x)] at the limit y x. For this purpose, we introduce quantities like
, (x , ) lim%0+ , (x + %, ) and , (x , ) lim%0+ , (x %, ). Using
Eqs. (5.1), (5.2), (5.4) and (5.5), we find that the commutators [, (x , ), (x)] and
[, (x , ), (x)] yield the same result which may be expressed as

, (), (x) = i h f1 (x, )2 (x, )(x)
i h g2 (x, )1 (x, )(x) i h 2 (x, )2 (x, ).
(5.14a)
In Appendix D we present the details for deriving the above relation. Similarly, the
commutators [, (x , ), (x)] and [, (x , ), (x)] yield the same result given
by

, (), (x)
= i h f (x)1 (x, )2 (x, ) + i h g (x)2 (x, )1 (x, )
i h 1 (x, )1 (x, ).
(5.14b)
435
Using Eqs. (5.14a), (5.14b) and (5.9), one can also find out the derivatives of [, (),
(x)] and [, (), (x)] as

x , (), (x)
h 3
2 (x, )2 (x, )
= h (f + g)1 (x, )1 (x, ) 2 (x) +
2

2
+ h
, (x, )(x) i h f1 (x, )2 (x, )

+ g2 (x, )1 (x, ) x (x),
(5.15a)
and

x , (), (x)
= h (f + g) 2 (x)2 (x, )2 (x, )
h
3
1 (x, )1 (x, ) + h 2 (x), (x, )
2

+ i h x (x) f1 (x, )2 (x, ) + g2 (x, )1 (x, ) .
(5.15b)
We are further interested in evaluating commutation relations between , (y, ) and the
square of basic field operators. Proceeding as before, the commutators [, (x , ), 2 (x)]
and [, (x , ), 2 (x)] yield the same result given by

, (), 2 (x)
2
= h f (h f 2i)1 (x, )2 (x, ) 2 (x) h g(2i + hg)
2 (x, )1 (x, ) (x)

i h 2 + i h (f g) 2 (x, )2 (x, )(x).
(5.16a)
Similarly, the commutators [, (x , ), 2 (x)] and [, (x , ), 2 (x)] yield

, (), 2 (x)
= h f (h f 2i) 2 (x)1 (x, )2 (x, ) + h g(2i + h g) 2 (x)2 (x, )1 (x, )

i h 2 + i h (f g) (x)1 (x, )1 (x, ).
(5.16b)
All of these relations will be extensively used in our calculation of quantum conserved
quantities for the DNLS model.
6. Explicit construction of the quantum Hamiltonian and its spectrum

Here we try to find out the explicit form of the first few quantum conserved quantities
of DNLS model, which would satisfy the fundamental relation (4.2). Analogous to the
classical case (1.3a), we take the first quantum conserved quantity to be
+
I0 =
(x)(x) dx.
(6.1)
436
Using (5.14a), (5.14b), we find that

, (), I0 =
+

, (), (x) (x) + (x) , (), (x) dx
= i h
+

1 1 (x) + (x)2 2 dx.
(6.2)
Note that, in the above relation and in the rest of this section, we omit the arguments of
Jost solutions i (x, ) and i (x, ) for the sake of convenience. With the help of (5.11),
Eq. (6.2) can be simplified as

, (), I0
h
=
2
+
x , (x, ) dx

h
= , (+, ) , (, ) .
(6.3)
2
So one concludes that for n = 0, the fundamental relation (4.2) is satisfied by I0 .
By imitating its classical counterpart (1.3b), the second quantum conserved quantity
may be taken as
+
(x)x (x) dx.
I1 = i
(6.4)
Neglecting some integrals of total derivatives which lead to vanishing surface terms, one
can write the commutation relation between , () and I1 (6.4) as

, (), I1
+

=i
x , (), (x) (x) (x)x , (), (x) dx.
Applying further (5.15a), (5.15b) and neglecting some integrals of total derivatives, we find
that
+

i3
(x)2 2 + 1 1 (x)
, (), I1 = h
2

+ (x) f x (1 2 ) + gx (2 1 ) (x)

i(f + g) (x) (x)2 2 + 1 1 (x) (x) .
Using (5.9a), (5.9b) and (5.11) to simplify the r.h.s. of above relation, we readily obtain

, (), I1
h 2
=
4
+
, (x, )
dx
x

h 2
, (+, ) , (, ) .
4
437
(6.5)
Thus I1 satisfies the fundamental relation (4.2) for n = 1.

Finally, we try to calculate the quantum Hamiltonian of DNLS model. In analogy with
its classical counterpart (1.3c), we propose that this quantum Hamiltonian can be written
in the form
I2 = I2(1) + iq I2(2) ,
(6.6)
where
(1)
I2
+
=
(x)xx (x) dx,
(6.7a)
I2(2)
+
=
2 (x)x 2 (x) dx,
(6.7b)
and q is some yet undetermined coupling constant. Neglecting some integrals of total
derivatives which lead to vanishing surface terms, one can write the commutation relation
(1)
between , () and I2 (6.7a) as

, (), I2(1)
+

=
x , (), (x) x (x) + x (x)x , (), (x) dx.
Using (5.15a), (5.15b) to evaluate the commutators appearing in the r.h.s. of above relation
and neglecting again integrals of some total derivatives, we obtain

(1)
, (), I2
= h
+

2 (x)x , (x, )(x) + (f + g) (x), (x, )(x)
3
, (x, )
2

+ 2(f + g) (x) x (x)2 2 1 1 x (x) (x) dx,
(6.8)
where , (x, ) is given by (5.13). Using the identity (5.12) and substituting explicit
values of f and g (i.e., f = ei/2 /(cos /2), g = ei/2 /(cos /2)), Eq. (6.8) can be
438
written in the form

(1)
, (), I2
+ 4
= h
x , (x, ) 2 2 2 (x)x , (x, ) 2 (x)
8

+ 4 (x) x (x)2 2 1 1 x (x) (x) dx.
(6.9)
Next, we consider the commutation relation between , () and I2(2) (6.7b).

Neglecting the integral of a total derivative, we can write this commutator as

, (), I2(2)
+

=
, (), 2 (x) x 2 (x) x 2 (x) , (), 2 (x) dx.
Applying (5.16a), (5.16b), neglecting again integrals of some total derivatives, and also
using relations like x (1 2 ) = x (2 1 ) = 12 x , (x, ), the above equation can be
brought in the form

(2)
, (), I2
h
=
2
+

2
h f h g 2 2if 2ig 2 (x)x , (x, ) 2 (x)

+ 4i 2 + i h (f g) (x) x (x)2 2 1 1 x (x) (x) dx.
(6.10)
Using Eqs. (6.9), (6.10) (with explicit values of f, g) and (1.6), we find that the quantum
Hamiltonian (6.6) would satisfy the fundamental relation given by

, (), I2
h 4
=
8
+
, (x, )
dx
x

h 4
=
, (+, ) , (, ) ,
8
(6.11)
provided the parameter q is chosen as

q =
.
(6.12)
1 h 2 2
By substituting (6.12) in (6.6), we get an explicit expression for the quantum Hamiltonian
of DNLS model as

+
i
I2 =
(6.13)
2 (x)x 2 (x) dx.
(x)xx (x) +
1 h 2 2
439
Thus, it is established that the above quantum Hamiltonian satisfies the fundamental
relation (4.2) for n = 2. Even though this quantum Hamiltonian (6.13) is not manifestly
Hermitian, we can easily make it Hermitian by adding some integrals of total derivatives
which lead to vanishing surface terms. Comparing (6.13) with (1.3c) we surprisingly find
that, due to quantum effect, the coupling constant of the system is modified. Consequently,
unlike most other integrable systems, the quantum Hamiltonian of DNLS model cannot
be obtained from its classical counterpart by simply applying the normal ordering
prescription. It is interesting to note that, Eq. (6.12) is somewhat similar to the relation
between rest mass and dynamical mass of a relativistic particle given by: m = m02 2 ,
1v /c
where m0 , m and v/c play the role of , q and h , respectively. The v/c 0 limit is like
h 0 limit (for a fixed ) in our case. Just as the dynamical mass of a relativistic particle
coincide with its rest mass in the nonrelativistic limit, the quantum coupling constant q
(6.12) coincides with the bare coupling constant at h 0 limit. On the other hand, the
v/c 1 limit is analogous to | |1/h limit in our case. Just as the dynamical mass of
a particle goes to infinity at ultrarelativistic limit, q (6.12) goes to infinity at | |1/h
limit. Consequently, even though QYBE restricts the value of as | |1/h , there exists
no such restriction on the value of corresponding quantum coupling constant q (6.12).
Thus the apparent limitation about the applicability of QISM in solving quantum DNLS
Hamiltonian for the full range of its coupling constant is resolved in a very nice way.
It is evident that I0 (6.1) and I1 (6.4) represent the number operator and momentum
operator respectively for the quantum DNLS system. Substituting n = 0, 1 and 2 in
Eq. (4.7), one can explicitly write down the eigenvalue relations for I0 , I1 and I2 as
I0 |1 , 2 , . . . , N = h N|1 , 2 , . . . , N ,

I1 |1 , 2 , . . . , N =

N
h 2
j |1 , 2 , . . . , N ,
2
(6.14a)
(6.14b)
j =1

I2 |1 , 2 , . . . , N =

N
h 4
j |1 , 2 , . . . , N .
4
(6.14c)
j =1
Let us now compare these eigenvalue relations with those obtained through the technique
of coordinate Bethe ansatz. Projecting the bosonic Hamiltonian (6.13) on an N -particle
Hilbert space [22], we get

N

2
2
+ 2i h q
(xl xm )
+
HN = h
(6.15)
.
xl xm
xj2
j =1
l<m
The eigenvalues for this Hamiltonian with derivative -function interaction and corresponding momentum operator can be derived through the method of coordinate Bethe
ansatz [22,23]. It is easy to check that such eigenvalues completely match with our result
in Eqs. (6.14b), (6.14c) when we identify the momentum parameters (kj ) of coordinate
Bethe ansatz with the spectral parameters (j ) of present approach through the relation:
kj 2j /2. The eigenfunctions of the Hamiltonian (6.15) can also be constructed through
440
coordinate Bethe ansatz. If, for the simplest N = 2 case, such eigenfunction is chosen in
x1 < x2 region as f (x1 , x2 ) = ei(k1 x1 +k2 x2 ) , then its form in x1 > x2 region would be given
by [22,24]
f (x1 , x2 ) = A(k1, k2 )ei(k1 x1 +k2 x2 ) + B(k1 , k2 )ei(k2 x1 +k1 x2 ) ,
k k +i h (k +k )
where A(k1 , k2 ) = 1 2 k1kq2 1 2 and B(k1 , k2 ) = 1 A(k1 , k2 ) are the so-called

matching coefficients. With the help of these matching coefficients, one can easily find
out the S-matrix for two-body scattering as [24]
S(k1 , k2 ) = A(k1 , k2 )A(k2 , k1 )1 =
k1 k2 + i h q (k1 + k2 )
.
k1 k2 i h q (k1 + k2 )
(6.16)
Using Eqs. (6.12) and (1.6), we can express q as: q = 1/h tan . Putting this form of
q in Eq. (6.16), and identifying momentum parameters with spectral parameters through
relations like k1 2 /2, k2 2 /2, we find that this S-matrix (6.16) exactly matches with
our earlier result (2.11) which is derived in the framework of QISM. The fact that the
renormalized coupling constant q appears in the projected DNLS Hamiltonian (6.15),
instead of its classical counterpart , plays a crucial role in this comparison between the
results of coordinate and algebraic Bethe ansatz.
It is also interesting to compare the results of coordinate and algebraic Bethe ansatz
for the soliton sector of quantum DNLS model. By applying QISM it is found that, the
distribution of complex spectral parameters for such quantum N -soliton state is given by
the relation (2.17). Taking the square of both sides of this relation and substituting kj in
place of 2j /2, we obtain

2
exp i(N + 1 2j ) ,
(6.17)
2
where j [1, 2, . . . , N]. This equation coincides with the momentum distribution in
coordinate Bethe ansatz corresponding to the quantum N -soliton states of DNLS
Hamiltonian (6.13) [22,23]. Again, the fact that the modified coupling constant appears in
the Hamiltonian (6.13) allows us to exactly match the results of coordinate and algebraic
Bethe ansatz.
By using the eigenvalue relations (6.14b), (6.14c), we can also calculate the binding
energy for the above mentioned quantum N -soliton states. Substituting the values of
complex j (2.17) to (6.14b), we obtain the momentum eigenvalue corresponding to these
N -soliton states as
kj =
P=
N
h
2 i(N+12j ) h 2 sin N
.
e
=
2
2 sin
(6.18)
j =1
Similarly, by substituting j (2.17) to (6.14c), we obtain the energy eigenvalue corresponding to these states as
E=
N
h
4 2i(N+12j ) h
4 sin(2N)
.
e
=
4
4 sin(2)
j =1
(6.19)
441
To calculate binding energy, we assume that the momentum P (6.18) of the N -soliton
state is equally distributed among N number of single-particle scattering states. The real
(pure imaginary) spectral parameter associated with each of these single-particle states is
denoted by 0 . With the help of Eqs. (6.14b) and (6.18), we obtain
2 sin(N)
.
(6.20)
N sin
Using Eq. (6.14c), one can easily calculate the total energy for N number of such singleparticle scattering states as
20 =
2
h N 4 h
4 sin N
0 =
.
(6.21)
4
4N sin2
Subtracting E (6.19) from E (6.21), we obtain the binding energy of quantum N -soliton
state as

h 4 sin N sin N
cos N

.
EB = E E =
(6.22)
4 sin
N sin
cos
E =
Substituting N = 2 to the above relation, we obtain EB =
h 4
2
sin2 . Thus we get EB > 0

4
for any nonzero value of . For N = 3, (6.22) takes the form EB = 2h3 sin2 (3
4 sin2 ). Here we get EB > 0 only if || < /3. Applying the method of induction, we
find that the condition EB > 0 is in fact valid within the range || < /N for all values of
N [29]. Thus to obtain quantum N -soliton states with positive binding energy, the coupling
constant of DNLS model should be restricted within the region |q | < 1/h tan(/N).
In analogy with the fundamental relation of NLS model [11], in this article we
propose the fundamental relation (4.2) for DNLS model. This fundamental relation plays
a key role in our construction of quantum conserved quantities of DNLS model and their
spectra. However, from the technical point of view, our construction of quantum conserved
quantities is much more complicated than the case of NLS model due to the following
reasons. Quantum Jost solutions and their commutation relations with basic field operators
are extensively used to obtain the quantum conserved quantities of DNLS model. It turns
out that, in contrast to the case of NLS model, differential equations satisfied by these Jost
solutions corresponding to boundary conditions at x and x do not coincide
with each other. This salient feature of DNLS model is connected with the fact that its
quantum Lax operator (1.5) has a nonvanishing trace. We also find that, unlike the case of
NLS model, the commutation relation between Jost solutions of DNLS model and basic
field operators with arguments at exactly the same space point lead to the violation of
Jacobi identity. So we are compelled to use commutation relations between Jost solutions
and basic field operators with slightly shifted arguments in our calculation of quantum
conserved quantities.
Proceeding in the above mentioned way, we are able to explicitly construct the quantum
Hamiltonian and few other conserved quantities of DNLS model through basic field
442
operators of this system. Surprisingly we find that, unlike the cases of most other integrable
systems, this quantum Hamiltonian (6.13) cannot be obtained as normal ordered version
of the corresponding classical Hamiltonian (1.3c). This is due to the fact that a new
kind of coupling constant (q ), quite different from the classical one ( ), appears in the
quantum Hamiltonian of the DNLS model. Thus we obtain the explicit form of the quantum
Hamiltonian of DNLS model, which has been defined earlier in the framework of QISM
in a formal way. Interestingly, the relation (6.12) between and q is rather similar to
the relation between rest mass and dynamical mass of a relativistic particle. Just as the
dynamical mass of a relativistic particle coincides with its rest mass in the nonrelativistic
limit, q coincides with at h 0 limit. In the ultrarelativistic limit, the dynamical mass
of a particle tends towards infinity. In a similar way, q can take arbitrary large value at
| |1/h limit. Consequently, we can apply QISM to the quantum DNLS model for the
full range of its coupling constant, even though QYBE restricts the value of as | |1/h .
Due to the presence of modified coupling constant in the quantum Hamiltonian (6.13), we
are also able to consistently match various results of algebraic and coordinate Bethe ansatz
in the case of DNLS model. The S-matrix for two particle scattering and the distribution of
single-particle momentum for quantum N -soliton states are two such examples where the
results of algebraic and coordinate Bethe ansatz match with each other. We also calculate
the binding energy for the quantum N -soliton state of DNLS model and find out the range
of coupling constant for which this binding energy has a positive value.
As a future study, it might be interesting to find out the higher quantum conserved
quantities of DNLS model by using its fundamental relation and investigate whether the
coupling constants appearing in such higher conserved quantities also differ from their
classical counterparts. It is well known that, higher quantum conserved quantities of NLS
model cannot be expressed in normal ordered form as the integral of a one-dimensional
density [1113]. A similar situation might also arise for the case of higher quantum
conserved quantities of the DNLS model. Finally, it may be noted that, the present approach
of using fundamental relation for the construction of quantum conserved quantities and
their spectra might be applicable to discrete integrable systems like Heisenberg spin-1/2
chain, supersymmetric tJ model and Hubbard model.
Appendix A
Here we give a detailed derivation of the commutation relation (5.2a). At first, we shall
evaluate the commutator [1 (x + %, ), (x)] for the case % > 0 and take % 0 limit at
the final stage. Using the integral relation of 1 (x, ) (3.6a) and canonical commutation
relations (1.4), we find that

x+%
1 (x + %, ), (x) = if
dz e
i2
4 (zx%)

(z), (x) 1 (z, )(z)
x+%
+ if
dz e
i2
4 (zx%)

(z) 1 (z, ), (x) (z)

x+%
+ i
dz e
i2
4
(zx%)
dz e
i2
4
(zx%)
443

(z), (x) 2 (z, )
x+%
+ i

(z) 2 (z, ), (x)
= i h f e
i2
4 (%)
1 (x, )(x) i h e
i2
4
(%)
2 (x, ) + ,
(A.1)
where
x+%

i2
= if
dz e 4 (zx%) (z) 1 (z, ), (x) (z)
x
x+%

i2
+ i
dz e 4 (zx%) (z) 2 (z, ), (x) .
(A.2)
The lower limits of integrals appearing in the r.h.s. of Eq. (A.2) are fixed by using the fact
that the commutator [i (z, ), (x)] becomes trivial for the case z < x. Next, we rewrite
Eq. (A.1) as

1 (x + %, ), (x)
= i h f e
i2
4 (%)
1 (x + %, )(x) i h e
i2
4
(%)
2 (x + %, ) + + , (A.3)
where

1 (x, ) 1 (x + %, ) (x)

i2
i h e 4 (%) 2 (x, ) 2 (x + %, ) .
= i h f e
i2
4 (%)
It is clear that the above expression of vanishes at % 0 limit. Let us now assume
that commutators like [i (z, ), (x)] do not produce any singular term at the limit
z x. Due to this assumption, the operator (A.2) would also vanish at % 0 limit.
Consequently, by taking % 0 limit of (A.3), we obtain the commutation relation (5.2a).
Other commutation relations appearing in (5.2) can also be derived in a similar fashion.
It should be noted that, the forms of finally derived equations (5.2) justify in a selfconsistent way our assumption about the absence of singular terms in commutators like
[i (z, ), (x)] at z x limit.
Appendix B
Here we derive the commutation relations between Jost solutions and field operators
associated with the same space point through the prescription (5.8) and show that these
commutation relations violate the Jacobi identity. Inserting the commutators (5.1) and (5.2)
444
to the expression (5.8), and substituting the arguments x and x by x at the final stage, we
find that

i h f
i h
1 (x, ), (x) =
1 (x, )(x)
2 (x, ),
2
2
(B.1a)
i h f
(x)1 (x, ),
1 (x, ), (x) =
2
(B.1b)
i h g
2 (x, )(x),
2 (x, ), (x) =
2
(B.1c)

i h g
i h
(x)2 (x, ) +
1 (x, ).
2 (x, ), (x) =
2
2
(B.1d)
Similarly, one can calculate the commutators [i (x, ), (x)] and [i (x, ), (x)], by
defining them exactly like (5.8) and using the relations (5.4) as well as (5.5). In this way,
we obtain

i h g
i h
1 (x, ), (x) =
1 (x, )(x) +
2 (x, ),
2
2
(B.2a)

i h g
(x)1 (x, ),
1 (x, ), (x) =
2
(B.2b)

i h f
2 (x, ), (x) =
2 (x, )(x),
2
(B.2c)
i h f
i h
(x)2 (x, )
1 (x, ).
2 (x, ), (x) =
2
2
Due to Eq. (5.7), Jost solutions i (x, ) and j (x, ) commute with each other.
By using the above commutation relations, it is easy to check that

2 (x, ), 1 (x, ), (x) + 1 (x, ), (x), 2 (x, )

(B.2d)

h 2 f
2 (x, )2 (x, ).
+ (x), 2 (x, ), 1 (x, ) =
(B.3)
4
Thus it is evident that the set of commutation relations (B.1), (B.2) and (5.7) violate the
Jacobi identity.
Appendix C
For deriving the relation (5.9a) through the method of extension, one should shift the
argument of 1 (x, ) by a very small amount and find out x (1 (x + , )2 (x, )) for
both positive and negative . For both cases, 0 limit should be taken at the final stage.
445
Let us first take a positive . Using Eqs. (3.3) and (3.10) we get

)2 (x, )
x 1 (x,
= x 1 (x,
)2 (x, ) + 1 (x,
)x 2 (x, )

1 (x,
)(x)
+ i (x)
2 (x,
) 2 (x, )
= if (x)

) if (x)2 (x, )(x) + i1 (x, )(x) ,
+ 1 (x,
(C.1)
where x x + . The r.h.s. of (C.1) should be written in a way such that an operator like
(x) ((x)) is always placed at the extreme left (right) of each term. The 0 limit
should be taken after rewriting the r.h.s. of (C.1) in the above mentioned way with the help
), (x)]. Thus we have to ultimately use
of commutators [(x),
2 (x, )] and [1 (x,
), (x)], which are given
the 0 limit of commutators [(x),
2 (x, )] and [1 (x,
by Eqs. (5.5c) and (5.2b), respectively. By using these equations and dropping terms which
vanish at 0 limit, we obtain

)2 (x, )
x 1 (x,

= if (1 + i h f ) (x)
1 (x,
)2 (x, )(x)
(x)1 (x,
)2 (x, )(x)
2 (x,
)2 (x, ) + i1 (x,
)1 (x, )(x)
+ i (x)
= i (x)2 (x, )2 (x, ) + i1 (x, )1 (x, )(x).
(C.2)
Similarly we can calculate x (1 (x,

)2 (x, )) for the case < 0 and obtain the same
result for 0 limit. Hence, we conclude that x (1 (x, )2 (x, )) is given by Eq. (5.9a)
in a regularisation independent way.
Appendix D
For the purpose of deriving the relation (5.14a), we write down (x , ) and (x , )
explicitly as
(x , ) = 1 (x , )2 (x , ) 2 (x , )1 (x , ),

(D.1)

(x , ) = 1 (x , )2 (x , ) 2 (x , )1 (x , ).
(D.2)
Using (D.1), (5.2a), (5.2c) and (5.4), we find that

(x , ), (x) = 1 (x , ), (x) 2 (x , ) 2 (x , ), (x) 1 (x , )
= i h f1 (x , )2 (x , )(x) i h g2 (x , )1 (x , )(x)
i h 2 (x , )2 (x , ).
(D.3)
Similarly, using (D.2), (5.1), (5.5a) and (5.5c), we get

(x , ), (x) = 1 (x , ) 2 (x , ), (x) 2 (x , ) 1 (x , ), (x)
= i h f1 (x , )2 (x , )(x) i h g2 (x , )1 (x , )(x)
i h 2 (x , )2 (x , ).
that [(x , ), (x)]
Comparing (D.3) and (D.4), we find

same result (in the weak sense) given by Eq. (5.14a).
(D.4)
and
[(x , ), (x)]
lead to the
446
References
[1] P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. Lett. 74 (1995) 3005;
P. Fendley, H. Saleur, N.P. Warner, Nucl. Phys. B 430 (1994) 577;
P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. B 52 (1995) 8934.
[2] G. Montambaux, D. Poliblanc, J. Bellisard, C. Sire, Phys. Rev. Lett. 70 (1993) 497;
D. Poliblanc, T. Ziman, J. Bellisard, F. Mila, G. Montambaux, Europhys. Lett. 22 (1993) 537.
[3] L.D. Faddeev, Sov. Sci. Rev. C 1 (1980) 107;
L.D. Faddeev, in: J.B. Zuber, R. Stora (Eds.), Recent Advances in Field Theory and Statistical Mechanics,
North-Holland, Amsterdam, 1984, p. 561.
[4] E.K. Skylanin, in: M. Jimbo (Ed.), YangBaxter Equation in Integrable Systems, in: Advanced Series in
Mathematical Physics, Vol. 10, World Scientific, Singapore, 1990, p. 121.
[5] E.K. Skylanin, L.A. Takhtajan, L.D. Faddeev, Theor. Math. Phys. 40 (1980) 688.
[6] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation
Functions, Cambridge Univ. Press, Cambridge, 1993, and references therein.
[7] H.B. Thacker, Physica D 18 (1986) 348.
[8] K. Sogo, M. Wadati, Prog. Theor. Phys. 69 (1983) 431.
[9] F.H.L. Essler, V.E. Korepin, Phys. Rev. B 46 (1992) 9147.
[10] J. Links, H.Q. Zhou, R.H. McKenzie, M.D. Gould, Phys. Rev. Lett. 86 (2001) 5096.
[11] K.M. Case, J. Math. Phys. 25 (1984) 2306.
[12] M. Omote, M. Sakagami, R. Sasaki, I. Yamanaka, Phys. Rev. D 35 (1987) 2423.
[13] B. Davies, Physica A 167 (1990) 433.
[14] M. Wadati, A. Kuniba, J. Phys. Soc. Jpn. 55 (1986) 76.
[15] D.J. Kaup, A.C. Newell, J. Math. Phys. 19 (1978) 798.
[16] H.H. Chen, Y.C. Lee, C.S. Liu, Phys. Scr. 20 (1979) 490.
[17] M. Wadati, H. Sanuki, K. Konno, Y.H. Ichikawa, Rocky Mountain J. Math. 8 (1978) 323;
Y.H. Ichikawa, S. Watanabe, J. Physique 38 (1977) C6-15;
Y.H. Ichikawa, K. Konno, M. Wadati, H. Sanuki, J. Phys. Soc. Jpn. 48 (1980) 279.
[18] P.A. Clarkson, Nonlinearity 5 (1992) 453.
[19] A. Kundu, B. Basu-Mallick, J. Math. Phys. 34 (1993) 1052.
[20] B. Basu-Mallick, A. Kundu, Phys. Lett. B 287 (1992) 149.
[21] B. Basu-Mallick, T. Bhattacharyya, Nucl. Phys. B 634 (2002) 611.
[22] A.G. Shnirman, B.A. Malomed, E.B. Jacob, Phys. Rev. A 50 (1994) 3453.
[23] D. Sen, Quantization of the derivative nonlinear Schrdinger equation, cond-mat/9612077.
[24] E. Gutkin, Ann. Phys. 176 (1987) 22.
[25] E. Gutkin, Phys. Rep. 167 (1988) 1.
[26] H.B. Thacker, D. Wilkinson, Phys. Rev. D 19 (1979) 3660;
D.B. Creamer, H.B. Thacker, D. Wilkinson, Phys. Rev. D 21 (1980) 1523.
[27] A.B. Zamolodchikov, A.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[28] P.A.M. Dirac, Lectures on Quantum Mechanics, in: Belfer Graduate School Monograph Series, Vol. 2,
Yeshiva University, 1964.
[29] B. Basu-Mallick, T. Bhattacharyya, D. Sen, Bound and antibound soliton states for a quantum integrable
derivative nonlinear Schrdinger model, hep-th/0305252.

Fused integrable lattice models with quantum

impurities and open boundaries
Anastasia Doikou
Theoretical Physics Laboratory of Annecy-Le-Vieux, LAPTH, B.P. 110, F-74941 Annecy-Le-Vieux, France
Received 18 March 2003; accepted 2 July 2003
Abstract
The alternating integrable spin chain and the RSOS(q1 , q2 ; p) model in the presence of a quantum
impurity are investigated. The boundary free energy due to the impurity is derived, the ratios of the
corresponding g functions at low and high temperature are specified and their relevance to boundary
flows in unitary minimal and generalized coset models is discussed. Finally, the alternating spin chain
with diagonal and non-diagonal integrable boundaries is studied, and the corresponding boundary
free energy and g functions are derived.
PACS: 05.20.-y; 11.10.-z
1. Introduction
Two-dimensional exactly solvable models with boundaries have attracted a great deal
of research interest recently from the point of view of boundary conformal field theory [1],
and critical behavior [2], but also because of the rich variety of physical phenomena they
display, which in principle can be exactly investigated (see, e.g., [37]). There has been also
much interest on problems related to quantum impurities, mainly because of the important
role they play in low-dimensional physics [8], but also because of their relevance to
boundary conformal field theory [1,2,9,10]. There are numerous studies related to quantum
impurities yielding a great number of interesting and useful results (see, e.g., [1113]). In
this article we focus basically on the thermodynamic analysis of lattice integrable systems
in the presence of quantum impurities and integrable open boundaries.
E-mail address: doikou@lapp.in2p3.fr (A. Doikou).
doi:10.1016/j.nuclphysb.2003.07.001
448
A. Doikou / Nuclear Physics B 668 [FS] (2003) 447468
For both integrable lattice models and relativistic integrable field theories, in the bulk,
the corresponding free energy has been derived and the conformal properties have been
extensively studied [1424] by means of the thermodynamic Bethe ansatz. It is however of
great interest to extend these studies for integrable models with boundaries. In analogy
to the bulk case, when boundaries are added, the corresponding boundary free energy
and the so-called g function, which characterizes the ground state degeneracy due to the
boundaries, [2,2529] can be specified by means of the thermodynamic Bethe ansatz. On
the other hand statistical systems at the critical point it is known to display conformal
invariance [30,31], therefore, they can be associated with certain conformal field theories.
The low temperature behavior of the free energy per unit length of such system, in the bulk,
is described by [32,33]
c 2
T + , T 1,
f (T ) = f0
(1.1)
6u
where c is the central charge of the effective conformal field theory. Furthermore, when
boundaries are added the obtained free energy is modified up to an 1/L contribution (L
denotes the size of the system), namely, [2]
T
ln g,
(1.2)
L
where the ground state degeneracy g is expected to be related with the boundaries of the
system. One of the main aims, when studying such systems is to employ proper techniques
in order to specify the central charge and the ground state degeneracywhen boundaries
are present.
As already mentioned from the integrable systems point of view, the central charge and
the ground state degeneracy g can be identified by employing thermodynamic Bethe ansatz
techniques (see, e.g., [9,10,1427,29]). In this study in particular the alternating integrable
spin chain [34] and the RSOS(q1 , q2 ; p) model [35] in the presence of a quantum impurity
(Kondo type boundaries see, e.g., [11,12]) are investigated via the thermodynamic Bethe
ansatz, and the corresponding free energy is derived at low and high temperature. The
relevance of the results for the RSOS(q1 , q2 ; p) model to the boundary flows in minimal
[9] and generalized coset models [10] is discussed. Finally the alternating open spin chain
with diagonal and non-diagonal integrable boundaries is considered and the corresponding
boundary free energy and the g functions for the left and right boundaries are determined
by first principle calculations at low and high temperature.
f (T ) = f0 (T )
2. Quantum impurity
2.1. The alternating spin chain
Let us first focus on the alternating spin chain [34] in the presence of a quantum
impurity. For what follows it is necessary to introduce the basic constructing element of
the model, namely the R matrix, which is a solution of the YangBaxter equation [36,37]
R12 (1 2 )R13 (1 )R23 (2 ) = R23 (2 )R13 (1 )R12 (1 2 ).
(2.1)
We consider the R matrix obtained in [38], namely

1
1,s
3
3
R0k = sinh + i
+ S
+ sinh + S + S + ,
2
449
(2.2)
where S 3 , S act in general, on a (2S + 1)-dimensional space V = C 2S+1 , and they satisfy
the following commutation relations

sinh 2iS 3
S+, S =
,
sinh i
S 3 |0
= S|0
,
3
S , S = S ,
S + |0
= 0.
(2.3)
We can now define the transfer matrix of the chain1

t = tr0 T0 (),
(2.4)
where
q
1
2
1
2
T0 () = R02N+1 ( )R02N
()R02N1
() R02
()R01
(),
(2.5)
and R i is related to the spin Si = qi /2 (i = 1, 2) representation (R q is related to the spin

q/2 representation) (2.2). Following the standard Bethe ansatz method described in, e.g.,
[34,3941], the Bethe equations are obtained
eq1 ( )N eq2 ( )N eq ( ) =
M

e2 ( ),
(2.6)
=1
where
en (; ) =
sinh ( +
sinh (
in
2)
,
in
2)
(2.7)
Notice that the main difference between the usual bulk case [34] (without impurities) and
(2.6) is the appearance of the eq term in the left-hand side of (2.6) due to the presence of the
spin q/2 impurity. Moreover we should mention that q1 , q2 q1 play the role of flavors
in accordance with the picture in [12], where the spin f/2 (f flavor) chain is studied with
impurity of spin s.
In the thermodynamic limit, N , the string hypothesis is valid [16,17,39], namely
the solutions of Eq. (2.6) can be grouped into strings of length n with the same real part
and equidistant imaginary parts
i
)
(n,j
= n + (n + 1 2j ),
0
,
0,s
= + i
2
j = 1, 2, . . . , n,
(2.8)
1 If we derive the transfer matrix of the model with proper inhomogeneities we obtain massless relativistic
dispersion relations for the particle-like excitations of the model [41].
450
where n is real, and 0,s is the negative parity string and 0 is real. In order to formulate
the thermodynamic Bethe ansatz we consider all the strings with length n = 1, . . . , 1
plus the negative parity string. The Bethe ansatz equations (2.6) can be written after we
apply the string hypothesis [16,17]
M0
Mm
1
2

N

Xnq n
Xnqj n = ()n
Enm n m
Gn1 n 0 ,
j =1

gq 0
2

m=1 =1
N
gqj 0 =
j =1
Mm
1

G1m 0 m
m=1 =1
=1
M0

e2 0 0 ,
(2.9)
(2.10)
=1
where

cosh ( +
i
gn (; ) = en
=
2
cosh (
in
2)
,
in
2)
(2.11)
and Xnm , Enm , and Gnm are given in Appendix A Eq. (A.1).
We introduce the densities of the holes n and pseudo-particles n , and once we take
the logarithm and the derivative of the Bethe ansatz equations (2.9), (2.10) we conclude
n () =
1
1
Znq1 () + Znq2 () + Znq ( )
2
L
1

Anm m () B1n 0 (),

m=1
1
1

0 () + 0 () = bq1 () + bq2 () + bq ( )
2
L
1

B1m m () a2 0 (),
(2.12)
m=1
where 0 is the density of the negative parity string, and L = 2N is the length of the spin
chain,2 and the quantities Znm , Anm , Bnm , and bn are given in Appendix A Eqs. (A.5),
(A.6), (A.7), and (A.4). It is also convenient to solve n () in terms of n (), therefore we
consider the convolution of the first of Eqs. (2.12) with the inverse of Anm
A 1
nm = nm s ()(nm+1 + nm1 ),
where s () =
1
2 cosh 2
, s() =
1
2 cosh .
A1
nm Zmqi () = s()nqi ,
(2.13)
Having in mind the identities
A1
nm B1m () = s()n2
(2.14)
2 We treat the impurity, which sits at the 2N + 1 site of the chain, separately, therefore, L = 2N is the length
of the bulk part.
451
we obtain the following expressions

1
n () = s()(nq1 + nq2 )
A1
nm m () n2 s 0 ()
2
1
m=1
+ n2 s s 0 () +
0 () = 0 () + s 2 ()
1
s( )nq ,
L
(2.15)
note the extra 1/L contribution term in the first of the two above equations, which is due to
the impurity. Let us now derive the free energy of the system f = e T s, which is given
by
1
f =eTs =
2
1

d Znq1 () + Znq2 () n ()
n=1
2
T

d bq1 () + bq2 () 0 ()
1

n=1

n ()
n ()
d n () ln 1 +
+ n () ln 1 +
n ()
n ()

0 ()
0 ()
+ 0 () ln 1 +
.
d 0 () ln 1 +
0 ()
0 ()
(2.16)
The thermodynamic Bethe ansatz equations are obtained by minimizing the free energy
(f = 0) and by virtue of (2.12), we conclude that they coincide with the ones of the model
without impurities [42] and they are given by
1

1
1
T ln 1 + n () = Znq1 () + Znq2 () + T
Anm ln 1 + m
()
2
m=1

1
T B1n ln 1 + 0 () ,

1

1
1 + 0 ()
1
= bq1 () + bq2 () + T
B1m ln 1 + m
()
T ln
1
2
1 + 0 ()
m=1

T a2 ln 1 + 01 () ,
(2.17)
where n = n /n . Alternatively we can write the thermodynamic Bethe ansatz equations

1n ()
by virtue of (2.15) in the following form, n () = e T ,

1
1n () = s() T ln 1 + n+1 () 1 + n1 () s()(nq1 + nq2 )
2

1
+ n2 s() T ln 1 + 0 () ,
452

11 () = s() T ln 1 + 2 () ,

10 () = s() T ln 1 + 2 () .
(2.18)
Although the thermodynamic Bethe ansatz equations remain the same as in the bulk case,
the free energy of the model is modified because of the presence of the quantum impurity.
In particular, there is a non-trivial contribution to the free energy which is given by the
following expression, after we apply (2.15) and (2.17), (2.18) to (2.16)

1
,
f = e0 + f0 + fb + O
(2.19)
L
where f0 is the bulk free energy given by

T
f0 =
2

d s() ln 1 + q1 () 1 + q2 () .
(2.20)
e0 is the energy of the state with the seas of strings with length q1 , q2 filled

2
2
1
1
d Zqiqj ()s()
d Zqqj ( )s(),
e0 =
4
2L
i,j =1
(2.21)
j =1
where the non-trivial 1/L contribution to the bulk ground state energy is due to the
impurity. Finally, fb is the free energy contribution of the impurity (after we make the
shift 1 ln T )
T
fb =
L

1
d s ln T ln 1 + q () .
(2.22)
We can accurately evaluate differences of boundary free energies, or ratios of g functions

(1.2), for different temperatures and not specific values at each temperature. This happens
basically because an overall 1/L contribution, which cannot be explicitly evaluated, may
survive in the bulk calculation (see (2.19)) of the free energy as well (see also [2527]).
The boundary free energy, and the corresponding g-function (1.2) can be computed at any
temperature by employing numerical methods, however it is possible to make analytical
calculations at T weak-coupling and T 0 strong-coupling point (see, e.g.,
[2,12]).
In particular, for T the main contribution to the integral in (2.22) comes from the
behavior. The corresponding behavior of q () is given by (see also [19,42])
1
1
, n = 1, . . . , 2,
(n + 1)2

1

1
1
1
1 + 0
1 + 1
(2.23)
= ,
=1 ,
which obviously does not depend on q1 , q2 . The boundary free energy contribution is for
q < 1 weak-coupling point (see, e.g., [2,12]),
1 + n
fb =
T
ln(q + 1).
L
(2.24)
453
In the special value where q = 1 the boundary energy becomes following (2.23)
T
(2.25)
ln(),
2L
which is the half of the expected value. In the isotropic case , q can take any value
from 1 to and the impurity contribution is given by (2.24).
When T 0 the main contribution to the integral in (2.22) comes from the
behaviorrecall that we considered the shift 1 ln T . The quantity 1 + n for
and T 0 is given by [19,42]

sin2 q1+2

0 1
=
1 + q01 = 1,
n = 1, . . . , q1 1,
1 + n

,
2 (n+1)
sin q1 +2

sin2 q2 q1 +2

0 1
1 + n
=
1 + q02 = 1,

, n = q1 + 1, . . . , q2 1,
2 (nq1 +1)
sin q2 q1 +2
fb =
1
1
,
n = q2 + 1, . . . , 2,
(n q2 + 1)2

1 1
1

1
0
1 + 00
= ,
=1 ,
1 + 1
(2.26)
where = q2 . The situation is more complicated now, because the behavior of q ()

depends clearly on the flavors (see (2.26)). In particular, for q = qi the boundary entropy
is fb T 2 , and this is the completely screened case (see, e.g., [12]). For q < q1

sin (q+1)
T
q
+2
1
fb0 = ln
(2.27)
L
sin q1+2
1 + n0
this is the behavior of the non-trivial strong-coupling point (see also [2,12]) with flavor
q2 and impurity spin q. For q1 < q < q2

1 +1)
sin (qq
T
q2 q1 +2
0
,

fb = ln
(2.28)
L
sin q2 q1 +2
this case also corresponds to a strong-coupling point, with flavor q1 q2 and a reduced
spin impurity q q2 . For q > q2
T
ln(q q2 1)
(2.29)
L
this is the partially screened case with reduced impurity spin q q1 . Finally, for q = 1
(2.26)
fb0 =
T
ln( q2 ).
(2.30)
2L
We should emphasize again that the boundary free energy is calculated up to an overall
1/L contribution which we are not able to derive explicitly. Therefore, we consider only
differences of the free energy for different temperatures, and from that via relation (1.2)
fb0 =
454
we deduce the ratio of the g function,

ln
g 1 (1 + q )
= ln
,
g0
2 (1 + q0 )
(2.31)
where ln(1 + q,0 ) is derived in (2.23), (2.26). The model under study is related to WZW q
and WZW q2 model, as already concluded in [42,44]. Indeed the central charge conjectured
in [44], and found in [42] is given by
c=
3q
3q1
+
q1 + 2 q + 2
(2.32)
and it is expressed as the sum of the central charges of the SU(2) WZW models at level q2
and q (q = q2 q1 ). Therefore, we expect that our results (2.24)(2.30), (2.31) should
be related to boundary flows in WZW k models in analogy to the bulk case. Finally, we
should note that our results for the free energy (2.24)(2.30) for q1 = q2 coincide with the
ones found in [12] for the spin f/2 = q1 /2 chain with spin s = q/2 impurity.
In general the impurity sitting at the 2N + 1 site of the chain can be thought as an
immobile particle with constant rapidity . Therefore, the particle-like excitations of the
chain can interact (scatter) with the impurity giving rise to specific scattering amplitudes
see, e.g., [45], which we are going to study in detail elsewhere. The scattering should
involve in addition to the usual XXX part (in the isotropic limit of our model), RSOS type
scattering as well [45]. This is expected since our chain is related to WZW k model, and it
is known (see, e.g., [46]) that the S-matrix that describes WZW k models has apart from the
SU(2) invariant part an RSOS part as well, namely [46]
(k)
SWZW k = SSU(2) SRSOS .
(2.33)
(k)
SRSOS
SSU(2) is the usual XXX S matrix and

restriction parameter k + 2 see also [47].
is the S matrix for the RSOS model with
2.2. The generalized RSOS(q1 , q2 ; p) model

It is known that the effective conformal field theory for the critical RSOS(1, 1) model
is the unitary minimal model M , whereas the critical RSOS(q1 ; p) model corresponds to
SU(2)q1 SU(2)q1 2
[43]. It has
the generalized SU(2) coset model M(q1 , q1 2)
SU(2)2
been also recently shown [35] that the effective conformal field theory for the generalized
critical RSOS(q1 , q2 ; p) model (q2 > q1 ) consists of two copies of SU(2) coset models,
namely M(q1 , q1 2) M(q1 , q). Our aim is to study the boundary behavior of
the generalized RSOS(q1 , q2 ; p) model, with Kondo type boundaries. In particular, the
corresponding boundary free energy and the g-function will be derived, and their relevance
to boundary flows of conformal field theories [9,10] will be discussed.
To describe the model, an orthogonal lattice of 2N + 1 horizontal and M vertical sites
is considered. The Boltzmann weights associated with every site are defined as

ln lm
w(li , lj , lm , ln |)
(2.34)
.
li lj
455
With every face i of the lattice an integer li is associated, and every pair of adjacent integers
satisfies the following restriction conditions [36,48]
0 li+1 li + P 2P ,
(2.35a)
P li+1 + li 2 P ,
(2.35b)
where P = q1 for i odd, P = q2 for i even (let q2 > q1 ), for i = 1, . . . , 2N and P = q for
i = 2N + 1 for the horizontal pairs, while P = p for the vertical pairs (array type II [49]).
The fussed Boltzmann weights have been derived by Date et al. in [50] and they are
given by
wqi ,1 (a1 , aqi +1 , bqi +1 , b1 |) =
qi

w1,1 ak , ak+1 , bk+1 , bk | + i(k qi ) ,
a2 ...aqi k=1
(2.36)
where b2 , . . . , bqi are arbitrary numbers satisfying |bi bi+1 | = 1.

are the Boltzmann
weights for the SOS(1, 1) model [36], they are non-vanishing as long as the condition
(2.35) is satisfied, and for P = 1 they are given by the following expressions
w1,1
w(l, l 1, l, l 1|) = h(i ),

hl+1
,
w(l 1, l, l 1, l|) = h()
hl
h1
w(l 1, l, l 1, l|) = h(wl ) ,
hl
(2.37)
where
h() = ()H ()
(2.38)
H () and () are Jacobi theta functions and,

hl = h(wl ),
wl = w0 + il.
(2.39)
We are interested in the critical case where h() becomes a simple hyperbolic function,
i.e.,
h() =
sinh
,
sin
(2.40)
w0 , and are arbitrary constants. Furthermore,

wqi ,p (a1 , b1 , bq+1 , aq+1 ) =
p2
i 1
q

1
h i(k j ) +
k=0 j =0
p

wqi ,1 ak , bk , bk+1 , ak+1 | + i(k 1) ,
a2 ...aq k=1
(2.41)
again b2 , . . . , bqi are arbitrary numbers satisfying |bi bi+1 | = 1, and the pairs a1 , aq+1
and b1 , bq+1 satisfy (2.35), for P = q. The fussed weights satisfy the YangBaxter
456
equation in the following form

wpq (a, b, g, f |)wps (f, g, d, e| + )wqs (g, b, c, d|)
g
wqs (f, a, g, e|)wps (a, b, c, g| + )wpq (g, c, d, e|).
(2.42)
Here we only need the explicit expressions for wqi ,1 which are

h(ib )
q 1
,
wqi ,1 l + 1, l + 1, l , l|) = hqii 1 ()ha
hl

h( + ia)
q 1
,
wqi ,1 l + 1, l 1, l , l|) = hqii 1 ()hb
hl

h(id )
q 1
wqi ,1 l 1, l + 1, l , l|) = hqii 1 ()hc
,
hl

h(ic )
q 1
wqi ,1 l 1, l 1, l , l|) = hqii 1 ()hd
,
hl
(2.43)
where
a=
l + l qi
,
2
b=
l l + qi
,
2
c=
l l + qi
,
2
d=
l + l + qi
,
2
(2.44)
and
q
hk () =
q1

h + i(k j ) .
(2.45)
j =0
It is obvious that wqi ,1 (a, b, c, d|) are periodic functions, because they involve only
simple hyperbolic functions (2.43), (2.40) h( + i) = h(), = , i.e.,
wqi ,1 (a, b, c, d| + i) = ()qi wqi ,1 (a, b, c, d|).
(2.46)
Now we can define the transfer matrix of the RSOS(q1 , q2 ; p) model

q ,q ,q;p{b ...b
}
T{a11 ...a2 2N+1 } 1 2N+1
2N1

wq1 ,p (aj , aj +1 , bj +1 , bj |)
j =1
wq2 ,p (aj +1 , aj +2 , bj +2 , bj +1 |)
wq,p (a2N+1 , a2N+2 , b2N+2 , b2N+1 |),
(2.47)
where we impose periodic boundary conditions, i.e., a2N+2 = a1 and b2N+2 = b1 . By

finding the eigenvalues of the transfer matrix we end up with the Bethe ansatz equations of
the model (see also [35,43])
2 eq ( )eq1 ( )N eq2 ( )N =
M

=1
e2 ( ).
(2.48)
457
It is important to mention that the eigenstates of the model are states which satisfy (see
also [35,43,49])
q
1
M = (q1 + q2 )L + .
(2.49)
4
2
To formulate the thermodynamic Bethe ansatz for the RSOS(q1 , q2 ; p) model we consider
all the strings (2.8) with length n = 1, . . . , 2 [35,43]. The densities of the corresponding
holes and pseudo-particles are derived from the Bethe ansatz equations (2.48), after we
insert all the allowed strings, and they satisfy

1 ()
1 ()
()
n () = Znq
Z
()
+
Z
()
+
(
A()
nq
nq
nm m (),
1
2
2
L
2
(2.50)
m=1
where L = 2N . From the constraint (2.49) it follows that 2 = 0, then the density of the
2 string can be written in terms of the remaining densities
2 () = ()
0
3

(2)
a2m
m (),
(2.51)
m=1
where an2 is given in Appendix A Eq. (A.4) with 2, and

0 () =
sinh(q1 2 ) + sinh(q2 2 )
sinh(q 2 )
1

.

+
L 2 cosh( 2 ) sinh ( 2) 2
4 cosh( 2 ) sinh ( 2) 2
(2.52)
By means of the relation (2.51) Eq. (2.50) can be rewritten in the following form

1 (2)
1 (2)
(2)
Z
()
+
Z
()
+
(
A(2)
m (),
n () = Znq
nq
nq
nm
1
2
2
L
m=1
(2.53)
3
(2)
Znm
, A(2)
are given by (A.5), (A.6), with 2. Following the standard procedure
nm
of minimizing the free energy of the system (f = 0, f is derived by (2.16)) we obtain the
thermodynamic Bethe ansatz equations which are the same as in the bulk [35]
3

1 (2)
(2)
1
()
+
Z
()
+
A(2)
T ln 1 + m
() ,
T ln 1 + n () = Znq
nq2
nm
1
2
m=1
(2.54)
alternatively we can write

1
1n () = s() T ln 1 + n+1 () 1 + n1 () s()(nq1 + nq2 ).
2
The corresponding free energy is

1
f (T ) = e0 + f0 + fb + O
,
L
(2.55)
(2.56)
where e0 is given by (2.21) and f0 is the bulk part of the free energy given by (2.20).
We wish to compute the non-trivial boundary part of the free energy, which is given as
458
in the previous casefor the alternating spin chainby (2.22). Again we can evaluate
differences of free energies at T 0 and T . For this purpose we need the solutions
of 1 + q (2.55), for T and T 0. For T , the main contribution in (2.22)
comes for [35,43] the solution is
1 + n
1
sin2 ( )

,
sin2 (n+1)
n = 1, . . . , 3
(2.57)
and for T 0 the main contribution to the free energy (2.22) comes for , and
the corresponding solution is [35]

sin2 q1+2

1
=
1 + q01 = 1,
1 + n0
n = 1, . . . , q1 1,
(n+1) ,
2
sin q1 +2

sin2 q2 q1 +2

0 1
1 + n
=
1 + q02 = 1,

, n = q1 + 1, . . . , q2 1,
2 (nq1 +1)
sin q2 q1 +2

sin2 q

2
0 1
1 + n
(2.58)
=

, n = q2 + 1, . . . , 3.
2 +1)
sin2 (nq
q2
We observe that the solution at T does not depend on qi , while at T 0 the solution
clearly depends on the values of qi . Having in mind the above solutions we are ready to
derive ratios of g functions, in particular
ln
g 1 (1 + q )
,
= ln
g0
2 (1 + q0 )
(2.59)
where 1 + q,0 are given by (2.57), (2.58).

It is worth making some remarks concerning (2.57), (2.58), and (2.59). In the special
case where q1 = q2 = 1 we recover the results of [9] for boundary flows in minimal models
M=m+1 (a = q 1), whereas for q1 = q2 > 1 we recover the results of [10] for boundary
flows in generalized SU(2) coset models M(q1 , q1 2) M(k, l). The corresponding
S matrix for the generalized coset M(k, l) model conjectured in [46] is given by
(k)
(l)
S = SRSOS
SRSOS
,
(2.60)
and in the limit l it reduces to SWZW k given in (2.33). Again from (2.57), (2.58) and
(2.59) the structure M(q1, q1 2) M(q1, q) of the effective conformal field theory
is manifest (compare with the boundary flow in SU(2) coset models [10]).
3. Open boundaries
The main aim of this section is the investigation of the thermodynamics of the
alternating spin chain in the presence of integrable boundaries. To construct the spin chain
with boundaries in addition to the R matrix another constructing element, the K matrix, is
459
needed. The K matrix is a solution of the reflection (boundary YangBaxter) equation [3],
R12 (1 2 )K1 (1 )R21 (1 + 2 )K2 (2 )
= K2 (2 )R12 (1 + 2 )K1 (1 )R21 (1 2 ).
(3.1)
In what follows we are going to use Sklyanins formalism [4] in order to construct the
model with boundaries. The corresponding transfer matrix t () for the open alternating
chain of N sites and S1 = q1 /2, S2 = q2 /2 spins is (see also, e.g., [4,34]),
t () = tr0 K0+ ()T0 ()K0 ()T
0 (),
(3.2)
where3
1
2
1
2
T0 () = R0N
()R0N1
() R02
()R01
(),
2
1
2
1
T
0 () = R10
()R20
() RN10
()RN0
(),
(3.3)
and K + (, + , + ) = K ( i, , )t where , are arbitrary boundary

parameters for the left and right boundaries, and K is the matrix [6,7]

sinh ( + i )
sinh 2
K () =
(3.4)
.
sinh 2
sinh ( + i )
It is interesting to point out that the diagonal boundaries for the critical XXZ spin chain
(S1 = S2 = 1/2) correspond to Dirichlet boundary conditions for the sine-Gordon model
[25,26]. Presumably the purely anti-diagonal K matrixwhich completely breaks the
U (1) symmetryshould correspond to Neumann boundary conditions for the sine-Gordon
model.
3.1. The diagonal K matrix
We consider first the case in which the K matrix is diagonal, namely = 0. The
corresponding Bethe ansatz for the model, are known [41] and they are given by
ex + ( )1 ex ( )e1 ( )g1 ( )eq1 ( )N eq2 ( )N
=
M

e2 ( )e2 ( + ),
(3.5)
=1
where x = 2 1 are boundary parameters for the left and right boundaries of the
chain and they are related to some external magnetic field acting on the boundaries. Here,
for simplicity we consider x to be integers less or equal to 1. The corresponding
densities of pseudo-particles and holes, in analogy to (2.12) satisfy (again we consider as
3 We could have considered N = odd and put an impurity at the N site of the chain. Then we would have an
impurity contribution to the free energy similar to the one in Section 2.
460
in the bulk all the strings n = 1, . . . , 1 and the negative parity string (2.8))

1
Kn
Anm m () B1n 0 ()
L
1
n () = Znq1 () + Znq2 () +
m=1
1

1
B1m m () a2 0 (),
0 () + 0 () = bq1 + bq2 + K0
L
m=1
(3.6)
where L = N is the length of the chain and,
n () = a n () + bn () Z
nx + () + Z
nx () 1,
K
0 () = a 1 () + b1 () bx + () + bx () + 1,
K
(3.7)
nm , bn , and a n are given in Appendix A (A.5), and (A.4). The unit that appears in the
Z
n , K
0 is a result of the subtraction of a () term from the densities (see

expressions for K
also [25,27]). This subtraction seems necessary for the accurate derivation of the density,
because in the boundary case s can take values from 0 to , as opposed to the bulk case
where s take all the values from to .
The free energy for the model with open boundaries is given by (2.16) up to a 1/2 factor
in front of the expression. The appearance of the factor 1/2 in front of all the integrals
comes from the fact that we originally derived the integrals from zero to infinity. After we
minimize the free energy, we obtain the same thermodynamic Bethe ansatz equations as
in (2.17), (2.18). Finally, by virtue of (2.17), (2.18) the following expression for the free
energy is obtained

1
,
f = f0 + fb + O
(3.8)
L
where f0 is the bulk free energy given by
T
f0 = e0
2

d s() ln 1 + q1 () 1 + q2 () ,
(3.9)
e0 is the bulk part of the energy of the state with the seas of strings q1 , q2 filled,

2
1
e0 =
d Zqi qj ()s()
4
(3.10)
i,j =1
and the boundary contribution to the free energy is

T
fb =
2L
1
n=1

T
d Kn () ln 1 + n1 () +
2L

d K0 () ln 1 + 01 () .
(3.11)
From the thermodynamic Bethe ansatz equations (2.17) we consider the convolution of
s() with the equation of (2.17) for n = 1 and n = x (with a () sign in front of the
461
equation for n = x + ), we also consider the convolution of s() with the equation for the
negative parity string, recall also relations (2.14). Moreover, we consider the equations
(2.18) for all n = 1, . . . , 1 and the negative parity string, with a () sign infront. We
add all the above equations and we end up with the following expression for the free energy
(at T 0, )
T
T
T
ln(1 + 1 ) +
s ln(1 + x + )
s ln(1 + x ),
L
2L
2L
(3.12)
where eb is the boundary energy contribution
fb = eb + fb = eb +
1
eb =
4L
1
=
2L

d s() Kq1 () + Kq2 ()

2

1
aqi () + bqi () Zqi x + + Zqi x () .
d s()
2

(3.13)
i=1
The boundary contribution to the free energy for each boundary then is
fb =
T
T
ln(1 + 1 )
ln(1 + x ),
2L
4L
(3.14)
and the corresponding ratios of the g + , g (g = g + g ) functions for the left and right
boundaries (1.2) are
ln
1 (1 + x )
1 (1 + 1
g
.
=
ln
ln
0 )
0 )
2 (1 + 1
4 (1 + x
g0
(3.15)
The last term of the above equation, for x integer less than 1, corresponds to the
quantum impurity contribution (up to a 1/2 factor), which has been already computed
explicitly for T , 0, (2.23), (2.26). We focus on the first term of (3.15), which we
compute again for T , 0. It is easy to conclude from the solution for T , 0
(2.23), (2.26)
)
1 (1 + 1
1
.
ln
= ln
0
2 (1 + 1
2 q2
)
(3.16)
The result (3.16) is also valid for the spin S = S1 = S2 chain. In the isotropic limit the
above ratio (3.16) becomes unit, i.e., there is no difference in the boundary free energy at
low and high temperature. In the special case where S1 = S2 = 1/2 we recover the result
found in [26] for the sine-Gordon model with boundaries,4 and also the result of [29] for
the XXZ spin chain with open boundaries at zero external magnetic field. Note that for
x there is no boundary parameter dependence.
4 In the repulsive regime, i.e., when no bound states breathers exist, we can make the following identification
= + 1 or 2 = 8( ) [26,52].
1
462
3.2. The non-diagonal K matrix

We consider, for the first time in a spin chain model, the more general case of the open
alternating spin chain with non-diagonal boundaries. The corresponding Bethe ansatz for
the XXZ model at roots of unity have been recently derived in [51]. The novelty of the
method described in [51] is basically that it does not rely on the existence of a reference
state pseudo-vacuum. It was shown then in [51], that in the case where = p+1
( = p + 1) the problem of finding the eigenvalues of the transfer matrix (3.2) reduces
to a set of functional relations which can be written in the following compact form, (see
also [35,43,51])

det M q1 ,q2 () = 0,
where

M q1 ,q2 ()
q1 ,q2
0
f q1 ,q2
1
=
..
0
q1 ,q2
fp1
q ,q
f11 2
q ,q2
11
(3.17)
0
q ,q
f 1 2
0
..
.
0
..
.
q ,q
f2 1 2
0
q ,q
21 2
0
q ,q
f 1 2
1
..
.
0
..
.
0
..
.
0
...
q1 ,q2
fp1
q ,q
1 2
p1
q ,q2
fp 1
q ,q2
f0 1
,
..
q
,q
1
2
fp2
0
q ,q2
p1
(3.18)
where

i
i sinh (2 + 2i)
f q1 ,q2 () = sinhN + iS1 +
sinhN + iS2 +
2
2 sinh (2 + i)

2
2
sinh ( + i ) sinh ( i ) + sinh 2
(3.19)
and
fq1 ,q2 () = f q1 ,q2 ( 2i),
q ,q2
fk 1
() = f ( + ik).
(3.20)
Let now (Q0 (), . . . , Qp ()) be the null vector of the matrix (3.18) with Qk () =
Q( + ik) and
Q() =
M

sinh ( j ) sinh ( + j + i)
(3.21)
j =1
then the eigenvalues are given by the following expression

q ,q2
q1 ,q2 () = f0 1
( i)
Q( + i)
Q( i)
q ,q
+ f0 1 2 ()
.
Q()
Q()
(3.22)
463
Finally, from the analyticity of the eigenvalues we obtain the Bethe ansatz equations

sinh + 2i (2 1) sinh 2i (2 + 1) + 2 sinh2 (2 i)

sinh 2i (2 1) sinh + 2i (2 + 1) + 2 sinh2 (2 + i)
g1 ( )e1 ( )eq1 ( )N eq2 ( )N =
M

e2 ( )e2 ( + ).
(3.23)
=1
Note that the boundary parameters at the left and right boundaries have been considered to
be the same. In fact, the boundary parameters have to be tuned properly (i.e., = + and
= + ) so that the method described in [51] can be applied. Note also that in this case
the Bethe ansatz states have to satisfy the following constraint,
1
N
.
(3.24)
2
2
The above constraint follows from the asymptotic behavior of the transfer matrix for
, in particular from (3.2) it can be deduced
M = (S1 + S2 )
t ( ) 2e(2N+4)+2i+iN
(3.25)
by comparing the later equation with the eigenvalues (3.22) we obtain the constraint (3.24).
Let us first consider the case with purely anti-diagonal K matrices, namely, becomes
very big. Then it is obvious that the above Bethe ansatz equations (3.23) take the form
g1 ( )2 e1 ( )2 g1 ( )e1 ( )eq1 ( )N eq2 ( )N
=
M

e2 ( )e2 ( + ).
(3.26)
=1
Relation (3.24) does not impose in this case any extra restriction on the densities as opposed
to the case of the RSOS model for which a similar relation (2.49) holds true, and it imposes
restrictions on the hole densities (i.e., 2 = 0). Therefore, the thermodynamic Bethe
ansatz equations are the same as in the bulk case (2.17), and by following exactly the same
procedure as in the diagonal case, but now with
n () = bn () a n () 1,
K
0 () = b1 () a 1 () + 1
K
(3.27)
we find that the non-trivial boundary part of the free energy becomes (at T 0, )
T
ln(1 + 1 ).
4L
Moreover the boundary contribution to the ground state energy is given by
2

1
1
eb =
aqi () bqi () () .
d s()
2L
2
fb =
(3.28)
(3.29)
i=1
It is worth noticing that the state with the seas of strings with length q1 , q2 filled, is an
allowed state for the case with anti-diagonal boundaries, because the constraint (3.24) is
464
satisfied. The ratios of the g + , g functions for the left and right boundary (1.2) are given
by
ln
g
1 (1 + 1 )
,
= 4 ln
g0
(1 + 10 )
(3.30)
where the function (1 + 10, )1 is given, for T and T 0, by (2.23), (2.26). Notice
that for qi = 1 (at T 0) fb T 2 and this isas in the presence of an impurity at the
end of the chainthe completely screened case.
We are not going to treat the general case in full detail here, since this is mostly
a problem of mathematical manipulations, he hope though to report on this in detail
elsewhere. Let us however present the general concept of such computation for the special
simple case where = 0, the so-called free boundary conditions [6], then the Bethe
ansatz equations become
e2 +1 ( )1 e2 1 ( )e12 ( )e1 ( )g1 ( )eq1 ( )N eq2 ( )N
=
M

e2 ( )e2 ( + ).
(3.31)
=1
It is evident that the later Bethe ansatz equations have a similar structure with (3.5)up
to the e12 term in the LHS of (3.31)with + = = and cosh2 i = 41 2 .5 The
boundary part of the free energy is given again by (3.11) and the quantities Kn and K0 are
given by,
nx + () + Z
nx () 1,
n () = a n () + bn () Z
K
0 () = a 1 () b1 () bx + () + bx () + 1
K
(3.32)
with x = 2 1. It remains to treat expression (3.11) at high and low temperature, but
this is relatively easy to do because the solutions for n , 0 are known at T and
T 0 (2.23), (2.26). Basically we have to focus on the part of expression (3.11) that
involves the terms Znx , bx for x non-integer. The remaining x independent part
of the boundary free energy (3.11), denoted as fb(1) , can be easily computed by repeating
the steps of the previous section, and it is given by (at T 0, )
(1)
fb
1
=
2L
+

1
aqi () + bqi () ()
d s()
2
2
i=1
T
T
ln(1 + 1 ) +
ln(1 + 1 ).
L
2L
(3.33)
The x dependent part of the boundary free energy, denoted as fb(2) (with terms that
involve Znx , bx ), which needs to be computed explicitly for x non-integers, has
5 Following [6] we have to chose for the free boundary conditions = 1 2 , which is not an integer.
2 1
465
the form
T
=
2L
1
fb(2)

d Znx () Znx + () ln 1 + n1 ()
n=1
T
+
2L

d bx () bx + () ln 1 + 01 () .
(3.34)
Again once we make the shift 1 ln T we can compute fb(2) analytically, at

T 0, having in mind the expressions for Znm , bn (A.2), (A.3), and n , 0 at high
and low temperatures (2.23), (2.26). More specifically, expression (3.34) can be written as
fb(2) = fb(2) (x ) + fb(2) (x + ) where at T 0, ,
fb(2) (x ) =
1

T
T
Znx (0) ln 1 + n1
bx (0) ln 1 + 01 .
2L
2L
(3.35)
n=1
Naturally, the case is similar when diagonal boundaries with x = 2 1 non-integer, are
considered, i.e., the x dependent part of the free energy (3.11) is given again by (3.35).
Eventually the computation of the expression (3.35) reduces to a simpler problem, which
is the derivation of the exact Fourier transforms of Znx and bx when x is non-integer. In
nx
particular, expression (A.4) for bx is valid as long as x < , while expression (A.5) for Z
is also valid as long as x > n.
Let us point out that the results of this section are novel not only for the fussed model
under study, but also for the case q1 = q2 , which corresponds to the XXZ model (sineGordon model).
4. Discussion
In this investigation we focused on the boundary properties of the alternating spin
chain and the RSOS(q1 , q2 ; p) model, therefore we considered both models with certain
integrable boundaries. A quantum impurity was added at the last site of the chain
(Kondo type boundary see also [11,12]), and the immediate result was a non-trivial
contribution to the ground state energy of the system, as well as in the free energy. We were
able to explicitly evaluate the non-trivial contribution to the boundary free energy at low
and high temperature. A similar investigation was realized for the RSOS(q1 , q2 ; p) model
(see also [35]) and the obtained results, for special values of q1 , q2 , compared with the
known boundary flows in unitary minimal models [9] and generalized SU(2) coset theories
[10]. For general values of q1 , q2 our results are rather novel for both the alternating spin
chain and the RSOS(q1 , q2 ; p) model.
Furthermore, the alternating chain with diagonal and non-diagonal boundaries was
investigated, and again the presence of the boundaries resulted in a non-trivial contribution
to the free energy. This contribution gave rise to the g function (ground state degeneracy)
along the lines described in [2]. We were able to derive ratios of the g function for the left
466
and right boundaries, at low and high temperatures. In the diagonal case for S1 = S2 = 1/2
the boundary parameter independent part of the ratio of the g functions coincides with the
one found in [26] and [29] for zero external magnetic field. We believe the results of this
section are also novel, in particular for the anti-diagonal case our results are rather new
even for q1 = q2 = 1.
For completeness the thermodynamics of the RSOS(q1 , q2 ; p) model with open
boundaries [5355] should be also investigated. It would be also of great interest to
extend the thermodynamic analysis for spin chains and RSOS models related to higher
rank algebras. Furthermore, the derivation of the Bethe ansatz equations for non-diagonal
boundaries [51], is an essential step towards the investigation of exact non-diagonal
boundary S matrices in the spin chain framework. Let us finally note that a new nontrivial dynamical solution of the reflection equation has been recently found [56], and it
has been realized in the context of the sine-Gordon model. It is a challenging problem to
apply this dynamical solution to the XXZ spin chain at roots of unity, and investigate the
thermodynamics and the scattering process in this framework. We hope to address these
questions soon in a future work [57].
Acknowledgements
I am grateful to F. Ravanini for helpful discussions. This work was supported by the
TMR Network EUCLID; Integrable models and applications: from strings to condensed
matter, contract number HPRN-CT-2002-00325.
Appendix A
We give the explicit expressions of the functions that appear in the Bethe asnatz
equations once we apply the string hypothesis, namely
Xnm () = e|nm+1| ()e|nm+3| () e(n+m3) ()e(n+m1) (),
2
2
Enm () = e|nm| ()e|nm+2|
() e(n+m2)
()e(n+m) (),
2
2
Gnm () = g|nm| ()g|nm+2|
() g(n+m2)
()g(n+m) ().
(A.1)
Moreover,
i d
ln gn (),
2 d

i d
ln Xnm (), Enm (), Gnm () .
Znm (), Anm (), Bnm () =
2 d
We finally give the following useful Fourier transforms
an () =
a n () =
i d
ln en (),
2 d
sinh(( n) 2 )
,
sinh(
2 )
bn () =
n < 2,
bn () =
sinh( n
2 )
,
sinh( 2 )
n < ,
(A.2)
(A.3)
(A.4)
sinh(( max(n, m)) 2 ) sinh((min(n, m)) 2 )

,
sinh(
2 ) sinh( 2 )
2 coth( 2 ) sinh(( max(n, m)) 2 ) sinh((min(n, m)) 2 )
A nm () =
,
sinh(
2 )
nm () =
Z
nm () =
B
m
2 coth( 2 ) sinh( n
2 ) sinh( 2 )
.
sinh(
2 )
467
(A.5)
(A.6)
(A.7)
References
[1] J.L. Cardy, Nucl. Phys. B 234 (1989) 581.
[2] I. Affleck, A.W.W. Ludvig, Phys. Rev. Lett. 67 (1991) 161;
I. Affleck, M. Oshikawa, H. Saleur, cond-mat/9804117.
[3] I.V. Cherednik, Theor. Math. Phys. 61 (1984) 977.
[4] E.K. Sklyanin, J. Phys. A 21 (1988) 2375;
P.P. Kulish, E.K. Sklyanin, J. Phys. A 24 (1991) L435.
[5] A. Fring, L. Kberle, Nucl. Phys. B 421 (1994) 159;
A. Fring, L. Kberle, Nucl. Phys. B 419 (1994) 647.
[6] S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841;
S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 4353.
[7] H.J. de Vega, A. Gonzlez-Ruiz, J. Phys. A 26 (1993) L519.
[8] C.L. Kane, M.P.A. Fisher, Phys. Rev. Lett. 68 (1992) 1220;
E. Sorensen, S. Eggert, I. Affleck, J. Phys. A 26 (1993) 6757.
[9] F. Lesage, H. Saleur, P. Simonetti, Phys. Lett. B 427 (1998) 85.
[10] C. Ahn, C. Rim, J. Phys. A 32 (1999) 2509.
[11] V.M. Filyov, A.M. Tsvelik, P.B. Wiegmann, Phys. Lett. A 81 (1981) 175;
A.M. Tsvelick, P.B. Wiegmann, Adv. Phys. 32 (1983) 453.
[12] N. Andrei, C. Destri, Phys. Rev. Lett. 52 (1984) 364.
[13] Y. Wang, Phys. Rev. B 60 (1999) 9236;
Y. Wang, P. Schlottmann, Phys. Rev. B 62 (2000) 3845;
A.P. Tonel, A. Foerster, X.-W. Guan, J. Links, cond-mat/0112115.
[14] C.N. Yang, C.P. Yang, Phys. Rev. 150 (1966) 327;
C.N. Yang, C.P. Yang, J. Math. Phys. 10 (1969) 1115.
[15] C.P. Yang, Phys. Rev. A 2 (1970) 154.
[16] M. Gaudin, Phys. Rev. Lett. 26 (1971) 1301.
[17] M. Takahashi, Prog. Theor. Phys. 46 (1971) 401;
M. Takahashi, M. Suzuki, Prog. Theor. Phys. 48 (1972) 2187;
M. Takahashi, Thermodynamics of One-Dimensional Solvable Models, Cambridge Univ. Press, Cambridge,
1999.
[18] J.D. Johnson, B.M. McCoy, Phys. Rev. A 6 (1972) 1613.
[19] H. Babujian, Nucl. Phys. B 215 (1983) 317;
H. Babujian, A. Tsvelik, Nucl. Phys. B 265 (1986) 24.
[20] L. Mezincescu, R.I. Nepomechie, UMTG-170 (1992).
[21] Al.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695;
Al.B. Zamolodchikov, Phys. Lett. B 253 (1991) 391.
[22] A.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 497;
A.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 524;
A.B. Zamolodchikov, Nucl. Phys. B 366 (1991) 122.
[23] T.R. Klassen, E. Melzer, Nucl. Phys. B 338 (1990) 485.
[24] C. Destri, H.J. de Vega, Nucl. Phys. B 438 (1995).
468
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
A. LeClair, G. Mussardo, H. Saleur, S. Skorik, Nucl. Phys. B 453 (1995) 581.

P. Fendley, H. Saleur, N.P. Warner, Nucl. Phys. B 430 (1994) 577.
P. Dorey, I. Runkel, R. Tateo, G. Watts, Nucl. Phys. B 578 (2000) 85.
F.C. Alcaraz, M.N. Barber, M.T. Batchelor, R.J. Baxter, G.R.W. Quispel, J. Phys. A 20 (1987) 6397.
P. de Sa, A. Tsvelik, cond-mat/9503031.
A.M. Polyakov, JETP Lett. 12 (1970) 381.
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, J. Stat. Phys. 34 (1984) 763;
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
H.W.J. Blte, J.L. Cardy, M.P. Nightingale, Phys. Rev. Lett. 56 (1986) 742;
J.L. Cardy, Nucl. Phys. B 270 (1986) 186.
I. Affleck, Phys. Rev. Lett. 56 (1986) 746.
H.J. de Vega, F. Woyanorovich, J. Phys. A 25 (1992) 4499.
A. Doikou, J. Phys. A 36 (2003) 329.
R.J. Baxter, Ann. Phys. 70 (1972) 193;
R.J. Baxter, J. Stat. Phys. 8 (1973) 25;
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, San Diego, 1982.
V.E. Korepin, Theor. Math. Phys. 76 (1980) 165;
V.E. Korepin, G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method, Correlation Functions and
Algebraic Bethe Ansatz, Cambridge Univ. Press, Cambridge, 1993.
A. Kirillov, N.Yu. Reshetikhin, J. Sov. Math 35 (1986) 2621;
A. Kirillov, N.Yu. Reshetikhin, J. Phys. A 20 (1987) 1565.
L.D. Faddeev, L.A. Takhtajan, Russ. Math. Surveys 34 (1979) 11;
L.D. Faddeev, L.A. Takhtajan, J. Sov. Math. 24 (1984) 241.
L.A. Takhtajan, Phys. Lett. A 87 (1982) 479.
A. Doikou, A. Babichenko, Phys. Lett B 515 (2001) 220;
A. Doikou, Nucl. Phys. B 634 (2002) 591.
A. Bytsko, A. Doikou, in preparation.
V.V. Bazhanov, N.Yu. Reshetikhin, Int. J. Mod. Phys. A 4 (1989) 115.
S.R. Aladim, M.J. Martins, J. Phys. A 26 (1993) 7287.
P. Fendley, cond-mat/9304031.
D. Bernard, Phys. Lett. B 279 (1992) 78.
N.Yu. Reshetikhin, J. Phys. A 24 (1991) 3299.
G.E. Andrews, R.J. Baxter, P.J. Forrester, J. Stat. Phys. 35 (1984) 193.
N.Yu. Reshetikhin, H. Saleur, Nucl. Phys. B 419 (1994) 507.
E. Date, M. Jimbo, T. Miwa, M. Okado, Lett. Math. Phys. 12 (1986) 209.
R.I. Nepomechie, hep-th/0211001.
A. Doikou, R.I. Nepomechie, J. Phys. A 32 (1999) 3663.
C. Ahn, W.M. Koo, J. Phys. A 29 (1996) 5845, hep-th/9708080.
R.E. Behrend, P.A. Pearce, D.L. OBrien, J. Stat. Phys. 84 (1996) 1.
M.T. Batchelor, V. Fridkin, A. Kuniba, Y.K. Zhou, Phys. Lett. B 735 (1996) 266.
P. Baseilhac, K. Koizumi, Nucl. Phys. B 649 (2003) 491.
P. Baseilhac, A. Doikou, K. Koizumi, in preparation.

Classification of reflection matrices related

to (super-)Yangians and application to open spin
chain models
D. Arnaudon a , J. Avan b,1 , N. Cramp a , A. Doikou a ,
L. Frappat a,2 , E. Ragoucy a
a Laboratoire dAnnecy-le-Vieux de Physique Thorique, LAPTH, CNRS, UMR 5108, Universit de Savoie,
B.P. 110, F-74941 Annecy-le-Vieux cedex, France

b Laboratoire de Physique Thorique et Modlisation, Universit de Cergy, 5 mail Gay-Lussac,
Neuville-sur-Oise, F-95031 Cergy-Pontoise cedex, France

Received 23 April 2003; received in revised form 22 May 2003; accepted 10 June 2003
Abstract
We present a classification of diagonal, antidiagonal and mixed reflection matrices related to
Yangian and super-Yangian R matrices associated to the infinite series so(m), sp(n) and osp(m|n).
We formulate the analytical Bethe ansatz resolution for the so(m) and sp(n) open spin chains with
boundary conditions described by the diagonal solutions.
MSC: 81R50; 17B37
PACS: 02.20.Uw; 02.30.Ik; 75.10.Pq
1. Introduction
Quantum R matrices, solutions of the YangBaxter equation, are interpreted as diffusion amplitudes for two-body interactions of particle type eigenstates in integrable twodimensional field theories. The YangBaxter equation then ensures consistent factorisability of the three-body amplitudes in terms of two-body ones.
E-mail address: arnaudon@lapp.in2p3.fr (D. Arnaudon).
1 On leave of absence from LPTHE, CNRS, UMR 7589, Universits Paris VI/VII.
2 Member of Institut Universitaire de France.
doi:10.1016/S0550-3213(03)00503-0
470
D. Arnaudon et al. / Nuclear Physics B 668 [FS] (2003) 469505
When considering integrable field theories with nontrivial boundary effects such as
theories on a half-line, one needs to introduce a new object describing reflection processes
on the boundary. Integrability is preserved provided that the two-body exchange R matrix
and the one-body reflection K matrix obey a quartic consistency condition [14]
R12 (u v)K1 (u)R12 (u + v)K2 (v) = K2 (v)R12 (u + v)K1 (u)R12 (u v).
(1.1)
Let us remark that the reflection equation also appears when considering generalisation
of ZF algebras (i.e., ZamolodchikovFaddeev algebras [5]) allowing the presence of a
boundary. These generalised ZF algebras ensure the total scattering matrix of the model
to be unitary. In this approach, the commutation relations of ZF generators implement
an operator b(u) which obeys the reflection equation. The K matrix would then be a
representation of this boundary operator b [6].
Another point of view has recently been presented using a universal construction of
reflection algebras as twists of quantum algebras [7,8]. The exchange equation (1.1) would
then describe a trivial representation of the quantum generators, although it is not clear to
us that this interpretation is valid for spectral parameter dependent solutions. Furthermore,
solutions with quantum degrees of freedom on the boundary have also been obtained in [9].
These reflection equations, or boundary YangBaxter equations, have recently drawn
attention and systematic ways of computing some solutions for given R matrices were
derived, e.g., in [1012], recovering and extending previous results derived for instance
(1)
in [1316]. The considered cases were related to A1 trigonometric R matrices for all
(2)
(1)
spins [11], and to A2 [17] and An vector representations [12], yielding a wealth of new
solutions.
In this paper, we will consider the case of R matrices corresponding to vector
representations of Yangians and super-Yangians, i.e., rational R matrices constructed in
[18,19]. The complete classification for gl(n) in vector representation was given in [20]:
it appears that generically any solution is conjugated (by a constant matrix) to a diagonal
solution. [21] then derived a series of solutions for so(m) and sp(n) based upon an ansatz
proposed by Cherednik [1]. We present here, for the Lie (super)algebra series so(m),
sp(n) and osp(m|n), a classification of purely diagonal, purely antidiagonal and mixed
diagonal/antidiagonal solutions by directly solving the boundary equation (1.1) for a onedimensional boundary quantum space. Once the diagonal case is exhausted (corresponding
to flavour-preserving reflection matrices), the most natural extension to look for indeed
consists of reflection matrices which preserve pairs of conjugate states (according to
(2.2)). Particular solutions have already been derived for the so(m) algebra [21,22] and
the sp(n) algebra [21]. They can all be identified with particular diagonal solutions of our
classification.
We must point out that there exists in fact another notion of reflection equation, which is
related to the definition of twisted Yangians [23,24], and arises also in the theory of coideal
algebras described in [25,26]. This equation reads
t1
t1
(u v)K2 (v) = K2 (v)R12
(u v)K1 (u)R12 (u v), (1.2)
R12 (u v)K1 (u)R12
where the transposition t is defined below (see Definition 2.2). As pointed out in [19], this
equation is actually the same as (1.1) when R(u) is the R matrix of Yangians of type so(n),
sp(n) or osp(m|n).
471
One essential purpose in establishing such a classification of reflection matrices is to

use them in order to construct and eventually to solve spin chain models with a variety of
boundary conditions. We present here explicit resolutions, using the analytical Bethe ansatz
method, of the so(n) and sp(n) open spin chains with boundary conditions determined by
the diagonal solutions of the reflection equations. In particular, we derive explicit formulae
for the eigenvalues of the low-lying excitations of hole type. We obtain in each case the
full scattering matrix without ambiguities (e.g., CDD factors) including bulk and boundary
interactions. This result has in fact a very general character, not limited to the particular
models considered here. This actually provides us with bulk and boundary S matrices
relevant for integrable field theories with nontrivial boundary conditions, for which they
represent universal S matrices. We expect, in analogy with the bulk case, that such
boundary S matrices should correspond to sp(n) or so(m) GrossNeveu model with certain
boundaries. For this particular model, the study of boundary conditions associated with the
different K matrices deserves further attention but goes beyond the scope of the present
paper.
2. Generalities
Let gl(m|n) be the Z2 -graded algebra of (m + n) (m + n) matrices Xij and 0 = 1.
The Z2 -gradation is defined by (1)[i] = 0 if 1 i m and (1)[i] = 0 if m + 1
i m + n. In the following, we will always assume that n is even.
Definition 2.1. For each index i, we introduce a sign i

+1 for 1 i m + n2 ,
i =
1 for m + n2 + 1 i m + n
and a conjugate index

m+1i
=
2m + n + 1 i
for 1 i m,
for m + 1 i m + n.
(2.1)
(2.2)
In particular, i = 0 (1)[i] .

Definition 2.2. For A = ij Aij Eij , we define the transposition t by

t
Aij Eij .
At =
(1)[i][j ]+[j ] i j Aij E =
ij
(2.3)
ij
It satisfies (At )t = A and, for C-valued matrices, (AB)t = B t At .

As usual Eij denotes the elementary matrix with entry 1 in row i and column j and zero
elsewhere.
We shall use a graded tensor product, i.e., such that, for a, b, c and d with definite
gradings, (a b)(c d) = (1)[b][c] ac bd.
472
Definition 2.3. Let P be the (super)permutation operator (i.e., X21 P X12 P )

P=
m+n
(1)[j ] Eij Ej i
(2.4)
(1)[i][j ] i j E Ej i = P t1 .
(2.5)
i,j =1
and let
Q=
m+n

i,j =1
We define the R matrix

P
Q
R(u) = I +
u
u+
with 2 = (m n 2)0 .
(2.6)
The R matrix (2.6) satisfies the super-YangBaxter equation

R12 (u)R13 (u + v)R23 (v) = R23 (v)R13 (u + v)R12 (u),
(2.7)
where the graded tensor product is understood.

The operators P and Q satisfy
P 2 = I,
P Q = QP = 0 Q and Q2 = 0 (m n)Q.
(2.8)
The R matrix (2.6) is known to yield the osp(m|n) Yangian [19], and leads to the
nonsuperorthogonal (taking n = 0, 0 = 1) and symplectic (m = 0, 0 = 1) Yangians.
For obvious reasons, we will call labels as orthogonal (respectively symplectic), indices i
which satisfy 1 i m (respectively m + 1 i m + n).
Although we will restrict ourselves to the diagonal, antidiagonal and mixed cases for
the reflection matrices K, the following lemma can be used to get more general solutions.
Lemma 2.4. Let K(u) be a solution of the reflection equation (1.1) and U such
that U U t = 1 be a (constant) matrix of the orthogonal, symplectic or orthosymplectic
(super)group (depending upon the choice of R). Then K t (u), U K(u)U t and U K t (u)U t
are also solutions of (1.1).
The proof is straightforward, using the invariance of the R matrix under conjugation by U
t1 t2
= R12 .
and R12
3. Diagonal solutions of the reflection equation
In this section we consider invertible diagonal solutions for K(u), i.e.,

K(u) = diag k1 (u), . . . , km (u); km+1(u), . . . , km+n (u) ,
(3.1)
where the semicolon emphasises the splitting between orthogonal and symplectic indices.
Here the ki (u) are supposed to be analytic C-functions of u, i.e., the boundary quantum
space is one-dimensional.
473
Proposition 3.1. There are three families of generic diagonal solutions and two particular
cases:
D1. Solutions of sl(n) type, with one free parameter, for m even

m
n
1 + cu
i 1, . . . , ; m + 1, . . . , m +
k (u) =
ki (u) = 1,
.
1 cu
2
2
(3.2)
This solution has no extension to odd m.

This solution is obviously invariant under the action of SL( m2 | n2 ).
D2. Solutions with three different values of kl (u), depending on one free parameter
1 + c1 u
1 + cm u
,
km (u) =
,
1 c1 u
1 cm u
kj (u) = 1 j = 1, m,
k1 (u) =
where ( 0 )c1 cm + c1 + cm = 0.
(3.3)
This solution does not hold for m = 0, 1.

The moduli space of this solution is invariant under the action of OSP(m 2|n)
SO(2).
D3. Solutions without any free continuous parameter (but with two integer parameters)
ki (u) = k (u) = 1 i {1, . . . , m1 ; m + 1, . . . , m + n1 },
1 + cu
ki (u) = k (u) =
1 cu
i {m1 + 1, . . . , m m1 ; m + n1 + 1, . . . , m + n n1 },
2
.
where c =
0 (2m1 2n1 1)
(3.4)
This solution is invariant under the action of OSP(2m1 |2n1 )OSP(m2m1 |n2n1 ).
D4. In the particular case of so(4), the solution takes the more general form

1 + c2 u 1 + c3 u 1 + c2 u 1 + c3 u
,
,
.
K(u) = diag 1,
(3.5)
1 c2 u 1 c3 u 1 c2 u 1 c3 u
This solution contains the three generic solutions D1 (c2 c3 = 0), D2 (c2 + c3 = 0) and
D3 (c2 = c3 = ).
D5. In the particular case of so(2), any function-valued diagonal matrix is solution.
In each case D1D4, infinite value of the parameter c is allowed. It corresponds to the
constant value 1+cu
1cu = 1.
These solutions are given up to a global normalisation and a relabelling of the indices,
provided it preserves the orthogonal/symplectic splitting and all the sets of conjugate
indices {i, }.
Let us remind that the solutions for so(m) and sp(n) algebras are obtained by setting
n = 0 or m = 0, respectively.
474
Solutions found in [22] for so(m) algebras (m > 2) are identified after a suitable change
of basis with the sets D1 and D2. Solutions found in [21] for so(m) and sp(n) algebras are
identified with the set D3 and limiting cases of D1. The solution D3 with the special value
m1 = (m 1)/2 (with n = 0) corresponds to the only nontrivial rational limit of the set of
trigonometric solutions given in [28].
Remark that the generic solutions D1, D2, D3 are associated with symmetric superspaces (as they were introduced in [29], in a different context) based on OSP(m|n), namely
OSP(m|n)/SL( m2 | n2 ), OSP(m|n)/OSP(m 2|n) SO(2), and OSP(m|n)/OSP(m 2m1 |
n 2n1 ) OSP(2m1 |2n1 ), respectively. In the case of Lie algebras, we recover the usual
symmetric spaces (based on orthogonal and symplectic algebras).
Proof. The projection of the reflection equation (1.1) on Eij Ej i for i = j, reads
1
1
1
1
ki (u)ki (v) +
kj (u)ki (v) =
kj (u)kj (v) +
kj (v)ki (u), (3.6)
u+v
uv
u+v
uv
the solution of which is given by
1 + cij u
ki (u)
=
,
kj (u) 1 cij u
i = j, .
(3.7)
The cases cij = 0, correspond to the constant ratios
ki (u)
kj (u)
= 1.
ki (u)
kj (u)
(well-defined since K(u) is supposed

For convenience, we will introduce Fij (u)
to be invertible for generic u). When i = j, , Fij is then given by (3.7) for some cij and
obviously cj i = cij . When defined, since Fij (u)Fj k (u) = Fik (u), the parameters c must
also satisfy

cij + cj k + cki = 0,
for i = j, ,
j = k, k and k = i, .
(3.8)
cij cj k cki = 0
To any solution for K(u), we associate a partition of {1, . . . , m + n} where the classes are
defined by i j ki = kj ( cij = 0). The constraints (3.8) are sufficient to conclude
that the partition associated to any solution has at most three different classes, except when
m + n = 4, where it can in principle have four classes. Note that in the case of sl(m|n)
where the constraints (3.8) hold without any restriction on the indices, K(u) is always
built with at most two different functions.
More precisely, if m + n = 4, (3.8) implies that the partition of indices is constituted
either of two subsets, or of three subsets. In the last case, it can only be of the form
(D2) {i}, { } and {1, . . . , m + n} \ {i, },
where i is an orthogonal index, i.e., i {1, . . . , m}.
Projecting the reflection equation (1.1) on E Eij , one gets (for i = j, ) after taking
uv

Fi (u) Fj (u) + 0 Fj i (u) Fij (u) + (2u + ) Fi (u) Fj (u)

=
(3.9)
(1)[l] Fli (u) Flj (u) .
l
475
This equation show in particular that Fij = 1 F = 1. A recursion allows us to write

the two-subset partitions as either
/ I or
(D1) I = {i1 , i2 , . . .} and I = {1 , 2 , . . .}, i.e., i I
(D3) I = {i1 , 1 , i2 , 2 , . . .} and J = {j1 , 1 , j2 , 2 , . . .}, i.e., i I I .
Knowing the possible forms of the partition, one can evaluate the sum in (3.9). This
equation involves at most two different functions Fmn . It provides the constraints on the
parameters cmn . A global check ensures that all the remaining projections do not lead to
new constraints.
The cases m + n = 4 are solved by direct computation. Only the so(4) case eventually
exhibits a more general solution (D4).
Finally, the case of so(2) is special. In particular, R(u) appears to be diagonal. A direct
computation shows that all function-valued diagonal matrix is solution.
Note that when = 0 (as in the case of so(2)), the spectral parameter can be rescaled
before taking the limit 0, and the corresponding R matrix does not involve the identity
anymore [30]. This is the R matrix used in [22] for so(2).
4. Antidiagonal solutions of the reflection equation

We now look for invertible solutions of the antidiagonal form
K(u) =
m+n
%i (u)Ei .
(4.1)
i=1
Proposition 4.1. Solutions of the reflection equation with antidiagonal terms exist only
in the pure so(2m) or sp(2n) cases. They are constant and only restricted by the set of
constraints
%i % = 1,
i.
(4.2)
In the special case of so(2), any function-valued antidiagonal matrix is solution.

The proof of this proposition will be given in the next section since the antidiagonal
solutions appear to be particular cases of the mixed solutions discussed below.
5. Mixed solutions of the reflection equation

We now look for invertible solutions with terms both in the diagonal and in the
antidiagonal part
K(u) =
m+n

i=1
ki (u)Eii + %i (u)Ei
(5.1)
476
with at least one nonzero %i .

Lemma 5.1. The case of so(2) being excluded, the solutions of the reflection equation of
the form (5.1) are all constant (up to a global normalisation function).
In the special case of so(2), the set of solutions of the reflection equation is the union of
function-valued diagonal matrices and function-valued antidiagonal matrices.
Proof. Projecting the reflection equations on the elementary matrices Epq Ers , one gets
on Eij Ej :
on Eij Ei :
on Eij Ej i :
on Eij E :
on Eij Eij :
on Eij E :

(1)j
ki (u)%i (v) + k (v)%i (u)
u+v

(1)j
+
kj (u)%i (v) kj (v)%i (u) = 0,
(5.2)
uv

(1)j
kj (u)% (v) + k (v)% (u)

u+v

(1)j
ki (u)% (v) ki (v)% (u) = 0,

(5.3)
uv

(1)j
ki (u)ki (v) + %i (u)% (v) kj (u)kj (v) % (u)%j (v)
u+v

(1)j
ki (u)kj (v) kj (u)ki (v) = 0,
(5.4)
uv

(1)j
%i (u)% (v) %i (v)% (u) = 0,
(5.5)
uv

(1)ij +i+j
i j 0 (1)i %i (u)% (v) (1)j % (u)%i (v)
u+v+

1
(5.6)
%i (u)% (v) % (u)%i (v) = 0,
+
uv

1
ki (u)k (v) kj (u)k (v)
(1)ij +i+j i j
u+v+

1
+
ki (u)k (v) kj (u)k (v) + (1)i %i (u)% (v)
uv+

(1)j % (u)%j (v)

0
1
k (u)k (v) k (u)k (v)

u+v+ uv

Str K(u)
k (v) k (v)
+
uv+
0
+
(u + v)(u v + )

k (u)k (v) k (u)k (v) + %i (u)% (v) % (u)%j (v) = 0,
(5.7)
on Eij Ej :
(1)ij +i+j i j
on Eij Ei :
on 1 Eii :
on 1 Ei :
1
uv+
477

1
ki (u)% (v) (1)j 0 % (u)k (v)
u+v+

kj (u)% (v) + (1)j 0 % (u)kj (v)

0
+
k (u)% (v) + % (u)kj (v)
u+v

1
0
k (u)% (v) % (u)k (v)

u+v+ uv

Str K(u)
% (v)
= 0,
(5.8)
uv+

1
ij +i+j
kj (u)%i (v) (1)i 0 %i (u)k (v)
(1)
i j
u+v+

1
+
ki (u)%i (v) + (1)i 0 %i (u)ki (v)
uv+

0
k (u)%i (v) + %i (u)ki (v)
+
u+v

0
1
%i (u)k (v) k (u)%i (v)
u+v+ uv

Str K(u)
+
(5.9)
%i (v)
= 0,
uv+

1
%i (u)% (v) % (u)%i (v) = 0,
(5.10)
u2 v 2

1
ki (u)%i (v) + %i (u)k (v) k (u)%i (v) %i (u)ki (v) = 0.
2
2
u v
(5.11)
Of course, when several indices i, j, , coincide, the corresponding equations merge into
a single one.
Consider now a couple (i, j ) of indices such that i, j, , are all different. Then Eq. (5.5)
implies
%i (u)% (v) = % (u)%i (v),
(5.12)
the solution of which is given by %i (u) = %i (0)%(u) where %(u) is an arbitrary function,
which can be factorised by a change of normalisation of the diagonal elements. Hence
one can restrict K in (5.1) to have only constant antidiagonal elements without loss of
generality.
Suppose now that %i = 0 for some i (purely diagonal solutions are not considered in this
section). The general solution to the system formed by Eqs. (5.2) and (5.11) is then given
478
by3
ki (u) = u + ki (0),
k (u) = u ki (0),
kj (u) = u + kj (0) (j = i, ),
(5.13)
where is a constant. The inspection of the E Ei coefficient leads to = 0.

It follows that all diagonal terms are constant. Hence the mixed solutions for the
reflection matrix are necessarily constant.
The case of so(2) is solved by direct calculation.
We now specify the exact form of these solutions. We establish (the case of so(2) being
excluded):
Proposition 5.2. Mixed solutions of the reflection equation for osp(m|n) exist only when
m is even. They fall into two classes:
C1. The so(m) block is diagonal.
The solutions are parametrised by n complex parameters. The matrix K is given by

m
ki = 1, k = 1,
for i 1, . . . ,
:
%i = 0, % = 0,
2

i + = 0,
ki = sin(i )
for i {m + 1, . . . , m + n}:
(5.14)
with
%i = ei cos(i )
i + = 0.
C2. The sp(n) block is diagonal.
The solutions are parametrised by a couple of positive or null integers m1 m2 such
that
m
n
n
m1 m2 ,
m1 m2 [mod 2]
m1 + m2 1,
2
2
2
and by m 2(m1 + m2 ) complex parameters. Setting n1 = (n + 2m1 2m2 )/4, the
matrix K is given by

ki = k = 1,
for i {1, . . . , m1 }:
%
= % = 0,
ki = k = 1,
for i {m1 + 1, . . . , m1 + m2 }:
so(m) part:
(5.15)
%
= % = 0,
i

ki = k = 0,
m
:
for i m1 + m2 + 1, . . . ,
2
% = %1
i , %i C,
ki = k = 1,
for i {m + 1, . . . , m + n1 }:
%i = % = 0,

sp(n) part:
(5.16)
ki = k = 1,
for
i
m
+
n
+
1,
.
.
.
,
m
+
:
1
%i = % = 0.
2
3 Actually, we have restricted ourselves to meromorphic functions on C.
479
For so(m) (respectively sp(n)), the mixed solutions are of the form C2 with n = 0 and
m2 = m1 (respectively C1 with m = 0).
Note that this classification is given up to a global normalisation function and up to a
relabelling of the indices, as noticed in Proposition 3.1.
Proof. We have to consider only constant reflection matrices K. Then, extracting the
residues at the poles in u, Eqs. (5.2) to (5.11) reduce to (for i = , j, and j = )
%i (ki + k ) = 0,
(5.17)
+ %i % %j % = 0,

ki k kj k + 0 (1)i %i % (1)j %j % = 0,

(1)j (1)i %i %j = 0,
(5.18)
ki k kj k = 0,

k (1)j 0 ki %j = 0,

1 + (1)i 0 ki %i = 0,

1 + (1)i 0 ki % = 0,
(5.21)
ki2
kj2
(5.19)
(5.20)
(5.22)
(5.23)
(5.24)
(ki kj ) Str K = 0,
(5.25)
%j Str K = 0.
(5.26)
In the case of so(m) algebras, Eq. (5.22) implies ki = k for all 1 i m, from
which (5.21) follows. Eq. (5.17) is a consequence of (5.23) and (5.24), Eq. (5.25) follows
from (5.26) and (5.20) from (5.19). Finally the mixed solutions in the so(m) case are
characterised by the following equations:
ki = k ,
ki2
ki %i = 0,
+ %i % = kj2
+ %j %
Tr K = 0,
for all i, j.
(5.27)
(5.28)
These constraints exclude the existence of mixed or antidiagonal solutions with odd m,
since a matrix cannot be traceless if its diagonal has an odd number of nonvanishing
elements with equal squares.
In the case of sp(n) algebras, Eqs. (5.23) and (5.24) vanish, while Eq. (5.22) implies
ki = k for all 1 i n, from which (5.17), (5.21), (5.25) and (5.26) follow. Moreover,
Eqs. (5.19) and (5.20) are equal. Finally the mixed solutions in the sp(n) case are
characterised by the following equations:
ki = k ,
ki2
+ %i % = kj2
Tr K = 0,
(5.29)
+ %j %
(5.30)
for all i, j.
In the case of osp(1|n), one has to add the following equations (with 2 i, 1 + n)
k12 ki2 %i % = 0,
k12
ki k %i % = 0.
(5.31)
(5.32)
480
It follows from (5.31), (5.32) and (5.24) that ki2 = 0 for all i = 2, . . . , 1 + n. Since
%i Str K = 0, one has also k1 = 0. Therefore, there does not exist any mixed or antidiagonal
(invertible) solution for osp(1|n).
In the case of osp(m|n), (m > 1), Eq. (5.20) shows that at least one antidiagonal part
(orthogonal or symplectic) is zero: %i = 0 for i [1, m] or %i = 0 for i [m + 1, m + n].
This means that no pure antidiagonal solution exists for osp(m|n) superalgebras. We thus
have to consider two cases.
Case 1. The so(m) block is diagonal.
Eq. (5.22) with index j [m + 1, m + n] implies ki = k for all i, which excludes the
case of osp(m|n) with odd m. The remaining equations then lead to
ki2 + %i % = kj2 + %j %
kj = k
for i, j [1, m + n],
for j [1, m + n].
(5.33)
(5.34)
Case 2. The sp(n) block is diagonal.

Eq. (5.22) with index j [1, m] implies ki = k for all i. The remaining equations lead
then to
ki2 + %i % = kj2 + %j %
kj = k
for i, j [1, m + n],
for j [1, m + n],
ki %i = ki % = 0 for i [1, m],

m
m+n

ki = 0.
Str K =
i=1
(5.35)
(5.36)
(5.37)
(5.38)
i=m+1
Again, these constraints exclude the existence of mixed or antidiagonal solutions with
odd m, since one cannot have a traceless matrix if its diagonal has an odd number of
nonvanishing elements with equal squares.
For example, one finds the two following solutions for osp(4|2):
1
1
0
0 %2
1
1
%2 0
and
0
1
k 5 %5
1
1
%6 k5
where k52 + %5 %6 = 1. For osp(2|4) the two solutions take the form
0 %1
1
%1 0
k3
%3
and
1
,
k 4 %4
%5 k4
1
%6
k3
1
481
where k32 + %3 %6 = 1 and k42 + %4 %5 = 1.
6. Analytical Bethe ansatz for the so(n) and sp(n) open spin chains
The main aim of this section is the derivation of the Bethe ansatz equations for the
so(n) and sp(n) N -site open spin chains with nontrivial reflection conditions by means of
the analytical Bethe ansatz method (see, e.g., [3136]). To construct the open chain transfer
matrix we need to introduce the R matrix [37,38], which is a solution of the YangBaxter
equation (2.7). Normalisation of R matrix will be modified in order to connect it with the
physical XXX type Hamiltonians considered hereafter; in addition, we set = iu. We
focus on the so(n) or sp(n) invariant R matrix given by
R() = ( + i)I + i( + i)P iQ.
The R matrix (6.1) satisfies also crossing and unitarity, namely

t1
R12 ()R12 () = 2 + 2 2 + 1 I,
R12 () = R12
( i),
(6.1)
(6.2)
where t is the transposition defined as in (2.3).

The open chain transfer matrix is defined by [2]
t () = Tr0 K0+ ()T0 ()K0 ()T0 (),
(6.3)
where K0 () is any solution of Eq. (1.1) and K0+ () is a solution of a closely related
reflection equation defined to be (T denotes the usual transposition):
R12 (v u)K1T1 (u)R12 (u v 2i)K2T2 (v)
= K2T2 (v)R12 (u v 2i)K1T1 (u)R12 (v u),
(6.4)
Tr0 denotes trace over the auxiliary space 0, and

T0 () = R0N ()R0N1 () R02 ()R01 (),
T0 () = R10 ()R20 () RN10 ()RN0 ().
(6.5)
It is clear that any solution K (, ) of (1.1) where are arbitrary boundary

parameters, give rise to a solution K + () of (6.4) defined by K + () = K ( i)T ,
also depending on arbitrary boundary parameters + = .
To determine the eigenvalues of the transfer matrix and the corresponding Bethe ansatz
equations, we employ the analytical Bethe ansatz method [31,32]. Namely, we impose
certain constraints on the eigenvalues by exploiting the crossing symmetry of the model,
the symmetry of the transfer matrix, the analyticity of the eigenvalues, and the fusion
procedure for open spin chains. This then allows us to determine the eigenvalues by solving
a set of coupled nonlinear consistency equations or Bethe ansatz equations.
We first focus on the case with trivial boundaries, namely K () = K + () = 1. We will
in a second step derive the Bethe ansatz equations for all diagonal solutions found in the
previous sections, namely D1, D2, D3, D4 and solve them in the thermodynamical limit.
482
To obtain the necessary constraints, we recall that the fusion procedure for the open spin
chain [36,39,40] yields the fused transfer matrix

2N
t() = (2 + 2i)t ()t ( + i) ( + i) q(2 + i)q(2 3i), (6.6)
where we define
() = ( + i)( + i)( i)( i),

( i)( i)
for so(n),
q() =
( + i)( + i) for sp(n).
(6.7)
(6.8)
In addition, from the crossing symmetry of the R matrix (6.2) it follows that (when
K = K + = 1)
t () = t ( i).
(6.9)
K
K+
The transfer matrix with

=
= 1 is obviously so(n) (respectively sp(n)) invariant
since the corresponding R matrix (6.1) is so(n) (respectively sp(n)) invariant. The
symmetry of the transfer matrix makes the computation of its asymptotic behaviour
( ) a relatively easy task. Finally, to implement the analyticity of the eigenvalues,
we require that all poles must vanish. These constraints shall uniquely fix the eigenvalues.
This is the basic outline of the analytical Bethe ansatz method.
6.1. Eigenvalue of the pseudo-vacuum
In order to compute the general eigenvalues we need to first define a reference state or
pseudo-vacuum. After finding its pseudo-energy eigenvalue, we will be able, with the
help of the above discussed constraints, to derive the general eigenvalues of the transfer
matrix. The pseudo-vacuum, which is an exact eigenstate of the transfer matrix, is the state
with all spins up, i.e.,
|+ =
N

|+i ,
(6.10)
i=1
where |+ is the n-dimensional column vector

1
0
|+ =
... .
(6.11)
0
The action of the R matrix on the state |+ from the left gives rise to upper triangular
matrices, whereas the action from the right on +| = |+ gives lower triangular matrices.
It is obvious then that the corresponding action of the T (respectively T ) on |+
(respectively + |) will also give rise to upper (respectively lower) triangular matrices. It is
relatively easy, after some tedious algebra (see [36] for a detailed example), to determine
the action of the transfer matrix, t ()|+ = 0 ()|+ , where 0 () is given by the
following expression
0 () = a()2N g0 () + b()2N
n2

l=1
gl () + c()2N gn1 ()
(6.12)
483
with
a() = ( + i)( + i),
b() = ( + i),
c() = ( + i i) = a( i),
(6.13)

+ i2 2i ( + i)
g0 () =

,
+ i2 + 2i

+ i2 2i
gn1 () =

= g0 ( i).
+ i2 + i 2i
(6.14)
In g0 and gn1 the upper sign corresponds to so(n) and the lower sign to sp(n). Moreover
(when n = 2k),

+ i2 2i ( + i)
gl () =

, 0 < l < k,
+ i2 + il2 + i(l+1)
2
gl () = gnl1 ( i), k l < n 1.
(6.15)
This expression is valid for so(2k) with the (+) sign (in the ) and for sp(2k) with the ()
sign. For the so(2k + 1) chain the expressions for gl , l = k, are the same as in the so(2k)
chain, except for
gk () =
+
( + i)
= gk ( i).

i(k1)
ik
2 +
2
(6.16)
6.2. Dressing functions

Now that we have the expression for the pseudo-vacuum eigenvalue, in accordance with
the general analytical Bethe ansatz procedure, we make the following assumption for the
structure of the general eigenvalues:
() = a()2N g0 ()A0 () + b()2N
n2
gi ()Ai () + c()2N gn1 ()An1 (),
i=1
(6.17)
where Ai (), the so-called dressing functions, will now be determined.
We immediately get from the crossing symmetry of the transfer matrix (6.9):
Al () = Anl1 ( i),
0 l n 1.
(6.18)
Moreover, from the fusion relation (6.6), we obtain the following identity by a comparison
of the forms (6.17) for the initial and fused auxiliary spaces:
A0 ( + i)An1 () = 1.
(6.19)
Gathering the above two Eqs. (6.18), (6.19) we conclude

A0 ()A0 () = 1.
(6.20)
484
Additional constraints are obtained on the dressing functions from analyticity properties.
Studying carefully the common poles of successive gl s, we deduce from the form of the
gi functions (6.14)(6.16) that gl and gl1 have common poles at = il2 , therefore from
analyticity requirements

il
il
Al
(6.21)
= Al1
, l = 1, . . . , k 1.
2
2
The last relation is valid for so(n) and sp(n) as well. However, there are some extra
constraints specific to each case, namely

ik
ik
= Ak1
for so(2k + 1), sp(2k)
Ak
2
2
(6.22)
and

i(k 1)
i(k 1)
Ak
= Ak1
,
2
2

i(k 1)
i(k 1)
= Ak2
for so(2k).
Ak
2
2
(6.23)
Having deduced the necessary constraints for the dressing functions, we now determine
them explicitly. The dressing functions Al are essentially characterised by a set of
(l)
parameters {j | j = 1, . . . , M (l) }, where the integer numbers M (l) are related to the
eigenvalues of diagonal generators of so(n), sp(n). Defining these generators as
S (l) =
N
si(l) ,
s (l) = Ell El l.
(6.24)
i=1
The precise identification of M follows from the symmetry of the transfer matrix (see also
[31]), in particular
S (l) = M (l1) M (l) ,
l = 1, . . . , k 2
(6.25)
which is valid for so(2k + 1), so(2k) and sp(2k). The two remaining quantum numbers are
given by
For so(2k + 1): S (k1) = M (k2) M (k1),
S (k) = M (k1) M (k) ;
For so(2k): S (k1) = M (k2) M (+) M () ,

For sp(2k): S (k1) = M (k2) M (k1) ,
S (k) = M (+) M () ;
S (k) = M (k1) 2M (k) .
(6.26)
We have also M (0) = N (valid for both sp(n) and so(n) algebras).
Let us point out that, away from the boundaries, the behavior of the chain is considered
to be as in the bulk. Therefore, the above quantum numbers describe accurately enough the
states of the system. The dressing functions are given by the following expressions.
485
6.2.1. so(2k + 1)
A0 () =
Al () =
(1)
M

+ j
j =1
+ (1)
j +
(l)
M

+ (l)
j +i +
(1)
(l)
+ j
j =1
(l+1)
M
+ j
j =1
+ j
(1)
i
2
i
2
(1)
j +
il
2
i
2
i
2
(l)
j +
(l)
j
il
2
(l+1)
(l+1)
il
2
il
2
i
2
i
2
il
2
+i
il
2
(l+1)
(l+1)
j
j
il
2
il
2
i
2
i
2
l = 1, . . . , k 1,
Ak () =
(k)
M

+ (k)
j +
(k)
+ j +
j =1
(k)
j +
(k)
j +
i (k)
j +
ik
2
ik
2
ik
2
(k)
j +
ik
2
i + (k)
j +
ik
2
(k)
+ j +
ik
2
i
2
i
2
ik
2
ik
2
i
2
i
2
(6.27)
together with Al () = Anl1 ( i) for l > k. We recall that for so(2k + 1), = k 12 .
6.2.2. so(2k)
The dressing functions are the same as in the so(2k + 1) case for l = 0, . . . , k 3, while
Ak2 () =
(k2)
M
j =1
Ak1 () =
(k2)
+ j
+ (k2)
+
j
ik
2
+ (+)
j +
ik
3i
2 2
(+)
ik
i
j =1 + j + 2 2
(+)
j +
()
M

+ j
()
ik
3i
2 2
()
ik
i
j =1 + j + 2 2
+ (+)
j +
ik
3i
2 2
(+)
ik
i
j =1 + j + 2 2
()
M

+ j
j =1
+ j
()
()
ik
2
ik
2
i (k2)
+
j
(+)
M

(+)
M

(k2)
ik
2
ik
2
ik
2
ik
3i
2 2
(+)
j + ik2 2i
()
ik
3i
2 2
ik
i
()
j + 2 2
(+)
j +
ik
3i
2 2
(+)
j + ik2 2i
i
2
i
2
()
()
j
j
ik
2
ik
2
i
2
i
2
(6.28)
with still Al () = Anl1 ( i) for l > k 1, with = k 1.

6.2.3. sp(2k)
Similarly, the dressing functions are the same as in the so(2k + 1) case for l =
1, . . . , k 3, and
486
Ak2 () =
(k2)
M
(k2)
+ j
j =1
+ (k2)
+
j
(k1)
M
j =1
Ak1 () =
(k1)
M
(k2)
i j
ik
3i
2 2
(k1)
+ j
+ ik2 2i
+ (k1)
+
j
(k1)
(k)
M

ik
2
+ (k1)
+
j
+ j
j =1
+
(k2)
j
ik
2
ik
2
ik
2
+ (k)
j +
i
2
i
2
ik
3i
2 2
(k)
ik
i
j =1 + j + 2 + 2
ik
2
ik
2
(k1)
+
j
ik
3i
2 2
(k1)
j
+ ik2 2i
(k1)
+
j
(k1)
ik
2
ik
2
(k)
j +
ik
3i
2 2
ik
i
(k)
j + 2 + 2
i
2
i
2
(6.29)
In addition Al () = Anl1 ( i) for l > k 1; note that now = k + 1.

6.3. Bethe ansatz equations for K = 1
The above dressing functions Al satisfy all the imposed constraints and they are
unambiguously defined.
Requiring now the analyticity of the eigenvalues, we deduce the Bethe ansatz equations.
More specifically, successive Al s have common poles, which must disappear. Hence, the
sum of corresponding residues of Al and Al+1 in the eigenvalue expression (6.17) must be
zero. The Bethe ansatz equations immediately follow from this condition.
Let us define
ex () =
ix
2
ix
2
(6.30)
for any x. Then, the Bethe ansatz equations read.

6.3.1. so(2k + 1)
(1) 2N
e1 i
=
(1)
M

(1)
(1) (1)
(1)
e2 i j e2 i + j
j =1, j =i
(2)
M

(1)
(2)
(2)
e1 (1)
i j e1 i + j ,
j =1
M (l)
1=

(l) (l)
(l)
e2 (l)
i j e2 i + j
j =1, j =i
(l+)
M
=1 j =1
(l)
(l)
(l+)
(l+)
e1 i + j
,
e1 i j
l = 2, . . . , k 1,
1=
(k)
M

487

(k) (k)
(k)
e1 (k)
i j e1 i + j
j =1, j =i
(k1)
M
(k)
(k)
(k1)
(k1)
e1 i + j
.
e1 i j
(6.31)
j =1
6.3.2. so(2k)
The first k 3 equations are the same as in so(2k + 1), see Eq. (6.31), but the last three
equations are modified, namely
1=
(k2)
M

(k2)

e2 (k2)
(k2)
+ (k2)
e2 i
i
j
j
j =1, j =i
(k3)
M
(k2)
(k2)
(k3)
(k3)
e1 i
e1 i
j
+ j
j =1
(+)
M

(k2)

e1 (k2)
(+)
+ (+)
e1 i
i
j
j
j =1
()
M

(k2)
(k2)
()
()
e1 i
j e1 i
+ j ,
j =1
1=
( )
M

)
( ) ( )
( )
e2 (
i j e2 i + j
j =1, j =i
(k2)
M
)
)
(k2)
(k2)
e1 (
,
e1 (
i j
i + j
= .
(6.32)
j =1
6.3.3. sp(2k)
The first k 2 equations are the same as in the so(2k + 1), while the last two equations
are given by
1=
(k1)
M
(k1)
(k1) (k1)
(k1)
e2 i
e2 i
j
+ j
j =1, j =i
(k2)
M

(k1)

e1 i
e1 (k1)
(k2)
+ (k2)
i
j
j
j =1
(k)
M

j =1
(k1)
(k1)
(k)
(k)
e2 i
j e2 i
+ j ,
488
1=
(k)
M

(k) (k)
(k)
e4 (k)
i j e4 i + j
j =1, j =i
(k1)
M
(k)
(k)
(k1)
(k1)
e2 i + j
.
e2 i j
(6.33)
j =1
For all these cases, we recover the rational limits of the equations given in [34,35].
6.4. Eigenvalues and Bethe ansatz equations for diagonal K
Until now, we have considered the case of trivial boundary effects K + = K = 1.
We here come to the main point of our derivation and insert nontrivial boundary effects.
We shall then rederive modified Bethe ansatz equations. We choose K to be one of
the diagonal solutions D1, D2, D3, D4. We consider, for simplicity but without loss of
generality, K + = 1. Remark that the pseudo-vacuum remains an exact eigenstate after this
modification.
Let us rewrite the solutions D1, D2, D3 and D4 in a slightly modified notation, which
we are going to use from now on.
D1. The solution D1 can be written in the following form
K() = diag(, . . . , , , . . . , ).
(6.34)
The number of s is equal to the number of s, so that this solution exists only for the
so(2k) and sp(2k) cases as stated in Proposition 3.1, and
() = + i,
() = + i,
(6.35)
where = is the free boundary parameter.

D2. Solution D2 can be written as
1
c
K() = diag(, , . . . , , )
(6.36)
with
() =
+ i1
,
+ i1
() = 1,
() =
+ in
,
+ in
(6.37)
where 1 = c11 , n = c1n are the boundary parameters which satisfy the constraint
1 + n = 1.
(6.38)
We remind that this solution exists for so(n) algebras, but not for sp(n).
D3. Solution D3 has the form
K() = diag(, . . . , , , . . . , , , . . . , ).
(6.39)
The number of s is 2m and the number of s is n 2m, where the free integer parameter
m obeys 1 m k 1 for so(2k), sp(2k) and 1 m k for so(2k + 1). Moreover
() = + i,
() = + i,
(6.40)
489
where = n4 m has a fixed value, unlike the su(n) case where the corresponding solution
has a free boundary parameter (see [4,41]).
D4. Finally solution D4, which holds only for the so(4) case, can be written in the
following form
K() = diag(, , , ),
(6.41)
where
() = ( + i )( + i+ ),
() = ( + i )( + i+ ),
() = ( + i )( + i+ ),
() = ( + i )( + i+ ),
(6.42)
= c12 , + = c13 are both free parameters.

We now come to the explicit expression of the eigenvalues when K is one of the
above mentioned solutions. We should point out that the dressing functions are related to
the bulk behavior of the chain and thus they are form-invariant under changes of boundary
conditions. Indeed what is modified in the expression of the eigenvalues (6.17) are the gl
functions which characterise the boundary effects. We call the new gl functions gl .
D1. As already mentioned, the solution D1 can only be applied in so(2k) and sp(2k).
In this case we have
gl () = ( + i )gl (),
l = 0, . . . , k 1,
gl () = ( + i + i)gl (),
l = k, . . . , 2k 1,
(6.43)
where g() are given by (6.14)(6.16). The system with such boundaries has a residual
symmetry sl(k) in both cases, which immediately follows from the structure of the
corresponding K matrix.
D2. We have
( + i1 )
g0 (),
g0 () =
( + i1 )
( + i1 + i) ( + i1 + i)
gn1 () =
gn1 (),
( + i1 ) ( i + i1 + i)
( + i1 + i)
gl () =
(6.44)
gl (), l = 1, . . . , n 2.
( + i1 )
Again the gl () are given by (6.14)(6.16). Similarly, from the structure of the K matrix
we conclude that the residual symmetry is so(n 2) so(2).
D3. For the D3 solution we find the following modified g functions
gl () = ( + i )gl (), l = 0, . . . , m 1,

i
i
gl () = +
+
gl (), l = m, . . . , n m 1,
2
2

i2 2i
gl () = ( i i )
gl (), l = n m, . . . , n 1.
+ i2 2i
(6.45)
The symmetry of the transfer matrix for this K matrix is so(n 2m) so(2m) (respectively
sp(n 2m) sp(2m)).
490
D4. Finally for solution D4 the modified g functions are given by

g0 () = ( + i )( + i+ )g0 (),
g1 () = ( + i + i)( + i+ )g1 (),
g2 () = ( + i+ + i)( + i )g2 (),
g3 () = ( + i + i)( + i+ + i)g3 ().
(6.46)
We now formulate the Bethe ansatz equations for the general diagonal solutions. The only
modifications induced on Eqs. (6.31)(6.33) are the following for each solution:
1
D1. The factor e2
+ () appears in the LHS of the kth Bethe equation.
1
D2. The factor e2
() appears in the LHS of the first Bethe equation.
1 +1
1
D3. The factor e2
+m () appears in the LHS of the mth Bethe equation with m =
1, . . . , k 1 for so(2k + 1), sp(2k) and m = 1, . . . , k 2 for so(2k). For so(2k), when
m = k 1, the factor e11 () appears in the LHS of the (k 1)th and kth Bethe
ansatz equations.
We treat solution D4 separately in the next section.
7. Ground state and excitations

The next step is to determine the ground state and the low-lying excitations of the model.
One of the main aims of this work is indeed the computation of the scattering of the lowlying excitations off the boundaries. The bulk scattering for these models has been already
studied in [27], nevertheless we are going to rederive these results as a check in the next
section.
We recall that the quantum numbers that describe a state are given by (6.24), and the
d
t ()|=0 . It is given by
energy is derived via the relation H = d
(1)
M
1
1
.
E=
(1) 2
2
1
j =1 (j ) + 4
(7.1)
In what follows we write the Bethe ansatz equations for the ground state and the low-lying
excitations (holes) of the models under study. Bethe ansatz equations may in general only
be solved in the thermodynamic limit N . In this limit, it is assumed that a state is
described in particular by the density functions l () of the parameters (l)
i .
We here make the hypothesis that nontrivial boundary effects do not modify the nature
of the ground state and excited states but only the values of the Bethe parameters.
7.1. so(2k + 1)
Let us first consider the so(2k + 1) case, for which the ground state consists of k 1
filled Dirac seas, whereas the kth sea is filled with two-strings of the form 0 4i . Remark
491
the shift 4i , instead of 2i usually expected, because the length of the kth root in the
so(2k + 1) case is half the length of the other roots (see also [27]). This leads us to rewrite
the Bethe ansatz equations for the ground state and the low-lying excitations (holes) in the
following form, inserting the two-strings contribution in the kth set:
(1)
M
(1)

2N+1
(1)
(1)
e1 (1)
=
e2 i (1)
i
j e2 i + j
j =1
(2)
M

(1)
(1)
(2)
(2)
e1 i j e1 i + j ,
j =1
(l)
M
(l)

(l)
(l)
e1 (l)
e2 i (l)
=
i
j e2 i + j
j =1
(l+ )
M
(l)
(l)
(l+ )
(l+ )
e1 i + j
,
e1 i j
l = 2, . . . , k 2,
=1 j =1
(k1)
M

(k1)

e1 (k1)
e2 (k1)
(k1)
+ (k1)
=
e2 i
i
i
j
j
j =1
(k2)
M
(k1)
(k1)
(k2)
(k2)
e1 i
e1 i
j
+ j
j =1
(k)
M

j =1

(k1)

e 1 (k1)
(k)
+ (k)
i
j e 1 i
j
2
(k1)
(k1)
(k)
(k)
j e 3 i
+ j ,
e 3 i
2
M (k)
(k)

2 (k)
2 (k)
(k)
(k)
e1 (k)
=
e1 i (k)
e1 i + (k)
e2 i (k)
i
j
j
j e2 i + j
j =1
(k1)
M
j =1
(k)
(k)
(k1)
(k1)
e 1 i + j
e 1 i j
2
(k)
(k)
(k1)
(k1)
e 3 i j
e 3 i + j
.
2
(7.2)
For the general diagonal solutions of the reflection equation we have to multiply:
1
for D2: the LHS of the 1st Bethe ansatz equation with e2
;
1 +1
1
for D3: we multiply the LHS of the mth Bethe ansatz equation with e2
+m , 1 m
k 1.
492
We are interested in the low-lying excitations, which are holes in the filled sea and are
highest weight representations of so(2k + 1). We restrict ourselves here to the states with
(l) holes in the l sea, which correspond to the vector representations of so(2k + 1), and
also to holes in the k sea, which corresponds to the 2k -dimensional spinor representation
(see also [31]). We convert the sums into integrals by employing the following approximate
relation

M
1 (l)
1
1 (l)
1
l
,
=
f i
d f () ()
f i
f (0) + O
N
N
2N
N2
i=1
(l)
(7.3)
i=1
where the correction terms take into account the (l) holes located at values (l)
i and the
+
contribution at 0 . We shall denote by f () the Fourier transform of any function f ().

Once we take the logarithm and the derivative of (7.2), we extract the densities from the
equation
1
1
() = a()
K()
+ F
() + G(,
),
N
N
where ax () =
a() =
(7.4)
ln ex () and a x () = ex/2 . We have introduced
i d
2 d
1 ()
..
.
() = l () .
.
..
k ()
2a1 ()
0
,
..
.
0
(7.5)
F (), G(, ) are k component vectors as well with

(l)

(l)
(l)
F () = a1 ()j 1 a1 () + a2 () +
a2 j + a2 + j lj
j
j =1
(l)

a1 (l)
+ a1 + (l)
(j,l+1 + j,l1 ),
j
j
j = 1, . . . , k 2,
j =1
k1
(l)

(l)
(l)
a1 j + a1 + j l,k2
() = a2 () a 1 () + a 3 ()
j =1
(l)

a2 (l)
+ a2 + (l)
l,k1
j
j
j =1
(l)

j =1

(l)
(l)
(a 1 + a 3 ) j + (a 1 + a 3 ) + j kl ,
2
493

F k () = 3a1 () + a2 () a 1 () + a 3 ()
2
(l)
j =1
(l)

(a 1 + a 3 ) (l)
+ (a 1 + a 3 ) + (l)
k1,l
j
j
2

(l)
(l)
(2a1 + a2 ) j + (2a1 + a2 ) + j kl
(7.6)
j =1
and the dependent part is

for D2: Gj (, ) = a21+1 ()j 1 ,
for D3: Gj (, ) = a2 +m ()j m ,
m = 1, . . . , k 1.
are
Finally, the nonvanishing entries of K

ij () = 1 + a 2 () ij a 1 ()(i,j +1 + i,j 1 ),
K

k,k1 () = a 1 () + a 3 () ,
k1,k () = K
K
kk () = 1 + 2a 1 () + a 2 ().
K
(7.7)
i, j = 1, . . . , k 1,
(7.8)
The solution of (7.4) has the form

1
1
0 () + 1 (, ),
N
N
are k component vectors with
() = 2C() +
where C and 0,1
(7.9)
i1 ()a 1 (),

C i () = R
0i () =
k
ij ()F
j (),
R
1i (, ) =
j =1
k
ij ()G
j (, ).
R
(7.10)
j =1
=K
1 and C j is the energy of a hole in the j sea which can be written in terms of
R
hyperbolic functions

cosh k 12 j 2
j
C () =
, j = 1, . . . , k 1,

cosh k 12 2
1
C k () =
(7.11)
,

2 cosh k 12 2
ij () = e 2
R

sinh min(i, j ) 2 cosh k 12 max(i, j ) 2
,

cosh k 12 2 sinh 2
i, j = 1, . . . , k 1,
2
sinh j2
kj () = e
j k () = R
,
R

2 cosh k 12 2 sinh 2
j = 1, . . . , k 1,
2
sinh k
kk () = e
R
.
2
2 2 cosh 4 cosh k 12 2 sinh 2
(7.12)
494
7.2. so(2k)
In this case, the ground state consists of k filled Dirac seas of real strings. Therefore the
Bethe ansatz equations have exactly the same form as in (6.32). For the general diagonal
solutions of the reflection equation, we have to multiply:
1
for D1: the LHS of the kth Bethe ansatz equation with e2
+ ,
1
,
for D2: the LHS of the 1st equation with e2
1 +1
1
for D3: the LHS of the mth equation with e2
+m for m = 1, . . . , k 2, while the
LHS of the (k 1)th and kth equations for m = k 1 are multiplied by e11 .
In this case as well, we restrict ourselves to states with (l) holes in the l sea. Note that now
the spinor representation splits into two spinor representations of dimension 2k1 (see also
[31]), and the holes in the +, sea correspond exactly to these two spinor representations.
The densities satisfy the same Eq. (7.4) as in the so(2k + 1) case with and a given by
(7.5) and
C j () =
cosh(k 1 j ) 2
,
cosh(k 1) 2
C () =
1
,
2 cosh(k 1) 2
j = 1, . . . , k 2,
(7.13)
(l)

(l)
(l)
a2 j + a2 + j lj
F () = a1 ()j 1 a1 () + a2 () +
j
j =1
(l)

(l)
(l)
a1 j + a1 + j (j,l+1 + j,l1 ),
j = 1, . . . , k 3,
j =1
(l)
k2

(l)
(l)
a2 j + a2 + j l,k2
() = a2 () 2a1 () +
j =1
(l)

a1 (l)
+ a1 + (l)
(l,k3 + l+ + l ),
j
j
j =1
(l)

a2 (l)
+ a2 + (l)
l
F () = a2 () +
j
j
j =1
(l)

(l)
(l)
a1 j + a1 + j l,k2 ,
j =1
(7.14)
495
for D1: Gj (, ) = a2 + ()j k ,

for D2: Gj (, ) = a(21+1) ()j 1 ,
j
G (, ) = a(2 +m)()j m ,
for D3:
Gi () = a1 ()(i,k1 + i,k ),
m = 1, . . . , k 2,
m = k 1.
(7.15)
are
Again, the nonvanishing entries of K

ij () = 1 + a 2 () ij a 1 ()(i,j +1 + i,j 1 ),
K
i, j = 1, . . . , k 2,
k2, () = K
,k2 () = a 1 (),
K
++ () = 1 + a 2 (),
() = K
K
+ () = K
+ () = 0.
K
(7.16)
We solve Eq. (7.4) and find the densities i which describe a Bethe ansatz state. The
solution of (7.4) has the same form as in (7.9) with
ij () = e 2
R
sinh min(i, j ) 2 cosh(k 1 max(i, j )) 2

,
cosh(k 1) 2 sinh 2
i, j = 1, . . . , k 2,
2
sinh j2
j () = e
j () = R
,
R
2 cosh(k 1) 2 sinh 2
j = 1, . . . , k 2,
2
sinh k
2
() = e
++ () = R
,
R
2 2 cosh 2 cosh(k 1) 2 sinh 2
2
sinh(k 2) 2
+ () = e
+ () = R
.
R
2 2 cosh 2 cosh(k 1) 2 sinh 2
(7.17)
Let us now consider the particular solution D4 for the so(4) case. The corresponding
Bethe ansatz equations, as in the bulk, are basically two copies of the XXX spin chain
equations (so(4) = su(2) su(2)), namely
) 1 ( ) 2N+1
e2 +1 (
e1 i
i
( )
M

)
( ) ( )
( )
e2 (
i j e2 i + j ,
(7.18)
j =1
where = . It is obvious that the only representations that remain are the two twodimensional spinor representations. The Bethe ansatz equations are then two decoupled
equations, as it is also evident from (7.12): one has R+ () = R+ () = 0; moreover
R++ (), R () are given by (7.17) for k = 2, and
( )

)
)
F () = a1 () + a2 () +
a2 (
+ a2 + (
,
j
j
j =1
G (, ) = a2 +1 ().
(7.19)
496
7.3. sp(2k)
(j )
The ground state in this case consists of k 1 filled Dirac seas of two-strings (0 2i ),
and the k sea is filled with real strings. The Bethe ansatz equations take the form
(1)
M
(1)

2N+1
2 (1)
(1) (1)
(1) (1)
(1)
=
e22 i (1)
e2 (1)
i
j e2 i + j e4 i j e4 i + j
j =1
(2)
M

(1)
(1)
(2)
(2)
e1 i j e1 i + j
j =1
(1)
(1)
(2)
(2)
e3 i j e3 i + j ,
(l)
M
(l)
(l)
(l) (l)
(l) (l)
(l) (l)
(l)
e22 i j e22 i + j e4 i j e4 i + j
e2 i =
ij =1
(l+ )
M

(l+ )
(l+ )
e1 (l)
e1 (l)
i j
i + j
=1 j =1
(l)
(l)
(l+ )
(l+ )
e3 i + j
,
e3 i j
l = 2, . . . , k 1,
(k)
M
(k)
(k)
(k) (k)
(k)
e4 i j e4 i + j
e2 i =
j =1
(k1)
M

(k1)
(k1)
e1 (k)
e1 (k)
i j
i + j
j =1

(k1)
(k1)
e3 (k)
e3 (k)
,
i j
i + j
(7.20)
where for the general diagonal solutions of the reflection equation we have to multiply:
1
for D1: the LHS of the kth equation with e2
+ ,
1
1
for D3: the LHS of the mth equations with e2
+m+1 e2 +m1 , m = 1, . . . , k 1.
Again the densities for the state with (l) holes in the l sea satisfy the same Eqs. (7.4) as in
the so(2k + 1) case with given by (7.5), a j () = 2a2 ()j 1 and
C j () =
cosh(k + 1 j ) 2
,
2 cosh 2 cosh(k + 1) 2
j = 1, . . . , k,
(7.21)
497

F j () = 3a2 () + a4 () 2 a1 () + a3 () + a1 () + a3 () j 1
(l)

(l)
(l)
(2a2 + a4 ) j + (2a2 + a4 ) + j j l
+
j =1
(l)

(a1 + a3 ) (l)
+ (a1 + a3 ) + (l)
(j,l+1 + j,l1 ),
j
j
j =1
(j = 1, . . . , k 1)

F k () = a2 () + a4 () a1 () + a3 ()
(l)

(l)
(l)
+
a4 j + a4 + j lk
j =1
(l)

(l)
(l)
(a1 + a3 ) j + (a1 + a3 ) + j l,k1 ,
(7.22)
j =1
for D1: Gj (, ) = a2 + ()j k ,

for D3: Gj (, ) = a2 +m+1 () + a2 +m1 () j m ,
m = 1, . . . , k 1.
(7.23)
are
The nonvanishing entries of K

2

ij () = 1 + a 2 () ij a 1 () + a 3 () (i,j +1 + i,j 1 ),
K
i, j = 1, . . . , k 1,

k,k1 () = a 1 () + a 3 () ,
k1,k () = K
K
kk () = 1 + a 4 ().
K
(7.24)
As in the previous cases the solution of (7.4) has the form (7.9), (7.10) with
ij () =
R
e sinh min(i, j ) 2 cosh(k + 1 max(i, j )) 2

,
2 cosh 2
cosh(k + 1) 2 sinh 2
i, j = 1, . . . , k.
(7.25)
8. Scattering
Having obtained the excitations of the model, we are ready to compute the complete
boundary S matrix. For this purpose we follow the formulation developed by Korepin, and
later by Andrei and Destri [42,43]. First we have to implement the so-called quantisation
condition,

2iNpl
S 1 li = 0,
e
(8.1)
where pl is the momentum of the particle (in our case, the hole) with rapidity l1 . For the
case of (even) holes in l sea we insert the integrated density (7.9) into the quantisation
498
condition (8.1). We use the dispersion relation

1 d l
p ()
(8.2)
2 d

and the sum rule N 0 i d () Z+ . The boundary S matrix for the right boundary is
taken to be one of the diagonal solutions D1, D2, D3, D4, whereas the boundary S matrix
for the left boundary is proportional to unit. We end up with the following expression for
the boundary scattering amplitudes:
C l () =

+l l
1
= exp 2N

d l () 2C l ()

(8.3)
with
l (, ) = k0 ()k1 (, ),
+l () = k0 (),
(8.4)
where l is the first element of the diagonal boundary S matrix. It is obvious that +l has
no dependence and realises just the overall factor in front of the unit matrix at the left
boundary (recall that K + = 1). Moreover,
l1

l
k0 1 = exp i d 0l () ,
0

k1 ( l1 , ) = exp
1
d 1l
2i
l1 ,

(8.5)
0
l
with 0,1
given by (7.10). We finally restrict ourselves to l = 1 in the first sea and we write
1 (7.10),
the latter expression in term of the Fourier transform of 0,1
1
k0 () = exp
2

k1 (, ) = exp

d 1
i
,
()e
0

d 1
(, )ei .
(8.6)
In what follows we express the scattering amplitudes in terms of E-functions.

8.1. so(n)
Before we write down the explicit expressions for the boundary S matrices, let us
recall the form of the exact bulk S matrix. It is easy to compute the scattering amplitude
499
between two holes (vectors) in the first sea. The bulk scattering amplitude comes from the
contribution of the terms of 01 given by Eqs. (7.6), (7.10), (7.12), (7.14), with argument
j . After some algebra and using the following identity
1
2

0

E +1
d e 2
4
= ln +3
cosh 2
E
(8.7)
we conclude that the holehole scattering amplitude is given by the expression

i i 1 i+1 i+1 1

E n2 + 2
tan i1
n2 E n2 E n2 + 2 E n2
S0 () =
i+1 i+1 1 .
i+1 i i
1
tan n2 E n2 E n2 + 2 E n2 E n2 + 2
(8.8)
The explicit bulk S matrix then has the following structure4

S() =

S0 ()
i(i + )I + (i + )P iQ .
(i + )(i + 1)
(8.9)
Now, we give the expressions for the boundary S matrix, which follow from (8.5), (8.6),
and the duplication formula for the E function

1
1
E(x) = 2 E(2x).
22x1E x +
(8.10)
2
The -independent part of the overall factor k0 , Eq. (8.5), is given by
i i 3 i+1/2 3 i+1/2 1
E
+ 4 E n2 + 4 E n2 + 2
E n2
k0 () = Y0 () i n2

i+1/2 1 ,
i
3
E n2 E n2 + 34 E i+1/2
n2 + 4 E n2 + 2
where
Y0 () =
sin
i+1/2
n2
i1/2
sin n2
1
4

1
4
sin
i1/2
n2
i+1/2
sin n2
1
2

1
2
i
n2
i
sin n2
sin
(8.11)
1
4
.
1
4
(8.12)
Note that our solution includes the necessary CDD factors both for the bulk and boundary
matrices (see also [21,27]). The expression for the -dependent part k1 depends on which
solutions D1, D2, D3, D4 we consider.
D1:

i+

1
E i+
n2 + 2 E n2 + 1

k1 (, ) = i+
(8.13)
+i
,
E n2 + 12 E n2
+1
where = 12 is the renormalised boundary parameter (see also [41]). We also compute
the element of the K matrix (6.34) by employing the duality transformation ,
and the symmetry of the K matrix and of the transfer matrix (see also [41]). In particular,
we obtain a set of Bethe ansatz equations for , which allows to determine the
4 We do not compute all the eigenvalues of the bulk S matrix via the Bethe ansatz. Such a general computation,
in the bulk, is beyond the scope of this work.
500
difference
11 (, )
11 (, ) = e(2 1) 2
(8.14)
and consequently
(, )
= e2 ().
(, )
(8.15)
This provides us with a consistency check for our procedure since one obtains independently the exact ratio between different elements of the reflection matrix. Thus, we have
completely determined the K matrix that corresponds to the solution D1.
D2:
i+

i i+
tan n2n E n21 + 12 E n2 1 + 1

k1 (, ) =
i+

i+1

1
1
tan n2n E i+
n2 + 2 E n2 + 1
i+
i+
E n2n + 12 E n2 n
i+
(8.16)
i+ ,
E n2 n + 12 E n2n
where 1 = 1 12 and n = n + 12 are the renormalised boundary parameters. Remark
that the constraint 1 + n = 1 is also true for the renormalised boundary parameters,
namely 1 + n = 1. Similarly we employ the duality transformation which, for this
solution, reads 1 n . We then determine the difference
11 (, n , 1 )
11 (, 1 , n ) = e(21 1) 2 + e(2n+1) 2
(8.17)
and the ratio

(, 1 , n )
= e2 ()e2n ().
1
(, 1 , n )
(8.18)
Again we have here an independent consistency check.

(, , )
1 n
. We have not been
Of course, we need to determine one further ratio, namely (,
1 ,n )
able to explicitly extract this information from the Bethe ansatz formulation. Nevertheless,
since the K matrix is a solution of the reflection equation, the only choice we have for this
ratio is
(, 1 , n )
= e21 ().
(, 1 , n )
(8.19)
With this, we have completed the derivation of the K matrix that corresponds to the
solution D2.
D3: the -dependent part of the K matrix overall factor that corresponds to D3 is
i+
i+1/2 3 i+1/2 1

1
E i+
n2 + 2 E n2 + 1 E
n2 + 4 E n2 + 4

k1 (, ) = i+
i+1/2 3 i+1/2 1 .
i+

1
E n2 + 2 E n2 + 1 E n2 + 4 E n2 + 4
(8.20)
501
The total overall factor for this solution is

i i 3 i+1/2 1 i+1/2 1
E
+ 4 E n2 + 4 E n2 + 2
E n2

k0 ()k1 (, ) = Y1 () i n2

i+1/2 1
i
1
E n2 E n2 + 34 E i+1/2
n2 + 4 E n2 + 2

i+

1
E i+
n2 + 2 E n2 + 1
i+
(8.21)

,

E n2 + 12 E i+
n2 + 1
where
Y1 () =
sin
i+1/2
n2

sin i1/2
n2
1
4

1
4
sin
i1/2
n2

sin i+1/2
n2
1
2

1
2
i
n2
i
sin n2
sin
1
4
.
1
4
(8.22)
The latter expression agrees with the one found in [21], although we find slightly different
CDD factors. Again = 12 is the renormalised boundary parameter which now must
have the fixed value = n4 m, to ensure the resulting K matrix satisfies the reflection
equation. Unfortunately, in this case, we cannot employ a duality transformation to derive
the ratio (6.39). However, since the K matrix is a solution of the reflection equation the
ratio must be
()
= e2 ().
(8.23)
()
D4: for this solution the -independent part is given by (8.11) with n = 4, while the
-dependent part is given by ( = ):
i+

i+
E 2 + 14 E
+ 34

2
k1 (, ) = i+
(8.24)
i+
.
E
+ 14 E 2 + 34
2
We exploit the duality transformation one more time, , to calculate the ratio
()
= e2 ()
()
(8.25)
with the renormalised boundary parameter = 12 . We recall that the two spinor
representations are two-dimensional, and the corresponding K matrices are of course twodimensional with

K (, ) = diag (, ), (, ) .
(8.26)
In other words, we have obtained two copies of the XXX boundary S matrix, with two
free boundary parameters .
8.2. sp(n)
The corresponding bulk scattering amplitude for the sp(n) is given by the following
expression
i i 1 i+1 i+1 1

E n+2 + 2
tan i1
n+2 E n+2 E n+2 + 2 E n+2
S0 () =
i+1 i+1 1 Sb (), (8.27)
i+1 i i
1
tan n+2 E n+2 E n+2 + 2 E n+2 E n+2 + 2
502
where

Sb () = exp
i
cosh k
d
2 e
.
2 cosh cosh (k+1)
2
(8.28)
Again the structure of the S matrix is given by (8.9) with S0 given above. k0 () is again
given by the integral representation (8.6) using (7.10), (7.22) and (7.25). The corresponding
-dependent parts for each solution are given respectively by
D1:
i+

1
E i+
n+2 + 2 E n+2 + 1

k1 (, ) = i+
(8.29)

,

E n+2 + 12 E i+
n+2 + 1
where = 1 is the renormalised boundary parameter. We also compute the element
of the K matrix by employing the duality transformation ( ) and we obtain the
difference
11 (, ) = e(2 2) 2
11 (, )
(8.30)
and consequently
(, )
= e2 ().
(, )
D3:

k1 (, ) =
i+ 1 i+

1
n+2 4 E n+2 + 2 E n+2 + 1

i+ 1 i+

1
tan i+1/2
n+2 + 4 E n+2 + 2 E n+2 + 1
i+1/2 3

1
E i+1/2
n+2 + 4 E n+2 + 4
i+1/2 1 i+1/2 3 .
E n+2 + 4 E n+2 + 4
tan
(8.31)
i1/2
(8.32)
= 1 is the renormalised boundary parameter which must have the fixed value
= n4 m, in order for the resulting K matrix to satisfy the reflection equation. This result
also agrees with the -dependent part of the corresponding result of [21]. Unfortunately we
cannot employ a duality symmetry in this case to derive the ratio (6.39), but again, since
the K matrix is a solution of the reflection equation, the ratio must be
()
= e2 ().
()
(8.33)
9. Conclusion
We have established here the classification of diagonal, antidiagonal and mixed solutions to the reflection equations based on the Yangian R matrices for Lie (super)algebras
so(m), sp(n) and osp(m|n). The next step would then be to obtain the complete classification of K matrices, still associated with one-dimensional quantum boundary space, which
503
is technically more complicated. The cases of higher dimensional bulk representations, as

well as operator valued K-matrices remain to be classified.
Some of the solutions we obtained have then been used to compute the spectrum and
scattering datas for integrable spin chain systems with nontrivial boundary conditions.
Only the Lie algebra case has been considered here; the case of super-Lie algebras
osp(m|n) can now be envisioned: indeed, a suitable redefinition of indices allows us to
consider a similar exact ferromagnetic pseudo-vacuum (see [44]), providing us with the
starting point for the analytical Bethe ansatz.
Remark also that the analytical Bethe ansatz construction proposed here remains valid
for upper triangular reflection matrices. Such matrices can for instance be obtained by
conjugating our diagonal solutions by triangular matrices (following Lemma 2.4).
In addition, the use of the analytical Bethe ansatz method precluded applications to
nondiagonal reflection matrices, for which no exact pseudo-vacuum state is available.
However, this situation can be approached using the methods developed, e.g., in [45,46].
For an alternative approach, see [47].
Acknowledgements
We warmfully thank E. Sokatchev for discussions on symmetric spaces. This work
was supported by the TMR Network EUCLID. Integrable models and applications: from
strings to condensed matter, contract number HPRN-CT-2002-00325. J.A. wishes to
thank LAPTH for kind hospitality.
References
[1] I.V. Cherednik, Factorizing particles on a half line and root systems, Theor. Math. Phys. 61 (1984) 977.
[2] E.K. Sklyanin, Boundary conditions for integrable quantum systems, J. Phys. A 21 (1988) 2375.
[3] S. Ghoshal, A.B. Zamolodchikov, Boundary S matrix and boundary state in two-dimensional integrable
quantum field theory, Int. J. Mod. Phys. A 9 (1994) 3841, hep-th/9306002.
[4] H.J. de Vega, A. Gonzlez-Ruiz, Boundary K matrices for the six vertex and the N (2N 1), A(N 1)
vertex models, J. Phys. A 26 (1993) L519, hep-th/9211114.
[5] A.B. Zamolodchikov, A.B. Zamolodchikov, Factorized S-matrices in two dimensions as the exact solutions
of certain relativistic quantum field theory models, Ann. Phys. 120 (1979) 253;
L.D. Faddeev, Quantum completely integrable models in field theory, Sov. Sci. Rev. C 1 (1980) 107.
[6] A. Liguori, M. Mintchev, L. Zhao, Boundary exchange algebras and scattering on the half-line, Commun.
Math. Phys. 194 (1998) 569, hep-th/9607085.
[7] J. Donin, A.I. Mudrov, Reflection equation, twist, and equivariant quantization, math.QA/0204295.
[8] J. Donin, P.P. Kulish, A.I. Mudrov, On universal solution to reflection equation, math.QA/0210242.
[9] P. Baseilhac, K. Koizumi, Sine-Gordon quantum field theory on the half line with quantum boundary degrees
of freedom, Nucl. Phys. B 649 (2003) 491, hep-th/0208005.
[10] G.W. Delius, N.J. MacKay, Quantum group symmetry in sine-Gordon and affine Toda field theories on the
half-line, Commun. Math. Phys. 233 (2003) 173, hep-th/0112023.
[11] G.W. Delius, R.I. Nepomechie, Solutions of the boundary YangBaxter equation for arbitrary spin, J. Phys.
A 35 (2002) L341, hep-th/0204076.
[12] R.I. Nepomechie, Boundary quantum group generators of type A, Lett. Math. Phys. 62 (2002) 83, hepth/0204181.
504
[13] J. Abad, M. Rios, Nondiagonal solutions to reflection equations in SU(N ) spin chains, Phys. Lett. B 352
(1995) 92, hep-th/9502129.
[14] L. Mezincescu, R.I. Nepomechie, Fractional-spin integrals of motion for the boundary sine-Gordon model
at the free fermion point, Int. J. Mod. Phys. A 13 (1998) 2747, hep-th/9709078.
[15] A. Lima-Santos, Reflection K matrices for 19-vertex models, Nucl. Phys. B 558 (1999) 637, solvint/9906003.
(1)
[16] G.M. Gandenberger, New nondiagonal solutions to the an boundary YangBaxter equation, hepth/9911178.
[17] J.D. Kim, Boundary K matrix for the quantum MikhailovShabat model, hep-th/9412192.
[18] V.G. Drinfeld, Hopf algebras and the quantum YangBaxter equation, Sov. Math. Dokl. 32 (1985) 254;
V.G. Drinfeld, A new realization of Yangians and quantized affine algebras, Sov. Math. Dokl. 36 (1988)
212.
[19] D. Arnaudon, J. Avan, N. Cramp, L. Frappat, . Ragoucy, R matrix presentation for (super-)Yangians Y (g),
J. Math. Phys. 44 (2003) 302, math.QA/0111325.
[20] M. Mintchev, . Ragoucy, P. Sorba, Spontaneous symmetry breaking in the gl(N )NLS hierarchy on the
half line, J. Phys. A 34 (2001) 8345, hep-th/0104079.
[21] N.J. MacKay, B.J. Short, Boundary scattering, symmetric spaces and the principal chiral model on the halfline, Commun. Math. Phys. 233 (2003) 313, hep-th/0104212.
[22] M. Moriconi, Integrable boundary conditions and reflection matrices for the O(N ) nonlinear sigma model,
Nucl. Phys. B 619 (2001) 396, hep-th/0108039.
[23] G.I. Olshanski, Twisted Yangians and infinite-dimensional Lie algebras, in: P. Kulish (Ed.), Quantum
Groups, in: Lecture Notes in Mathematics, Vol. 1510, 1992, p. 104.
[24] A. Molev, M. Nazarov, G. Olshanski, Yangians and classical Lie algebras, Russ. Math. Surveys 51 (1996)
205, hep-th/9409025.
[25] M. Noumi, Macdonalds symmetric polynomials as zonal spherical functions on some quantum homogeneous spaces, Adv. Math. 123 (1996) 16.
[26] A. Molev, . Ragoucy, P. Sorba, Coideal subalgebras in quantum affine algebras, math.QA/0208140.
[27] E. Ogievetsky, N.Yu. Reshetikhin, P. Wiegmann, The principal chiral field in two-dimensions on classical
Lie algebras: the Bethe ansatz solution and factorized theory of scattering, Nucl. Phys. B 280 (1987) 45.
[28] M.T. Batchelor, V. Fridkin, A. Kuniba, Y.K. Zhou, Solutions of the reflection equation for face and vertex
(1)
(1)
(1)
(1)
(2)
models associated with An , Bn , Cn , Dn and An , Phys. Lett. B 376 (1996) 266, hep-th/9601051.
[29] M. Zirnbauer, Riemannian symmetric superspaces and their origin in random-matrix theory, J. Math.
Phys. 37 (1996) 4986, math-ph/9808012.
[30] P.P. Kulish, E.K. Sklyanin, Solutions of the YangBaxter equation, Zap. Nauchn. Sem. LOMI 95 (1980)
129;
P.P. Kulish, E.K. Sklyanin, J. Sov. Math. 19 (1982) 1596.
[31] V.I. Vichirko, N.Yu. Reshetikhin, Excitation spectrum of the anisotropic generalization of an SU 3 magnet,
Theor. Math. Phys. 56 (1983) 805;
N.Yu. Reshetikhin, A method of functional equations in the theory of exactly solvable quantum systems,
Lett. Math. Phys. 7 (1983) 205;
N.Yu. Reshetikhin, Integrable models of quantum one-dimensional magnets with O(n) and Sp(2k)
symmetry, Theor. Math. Phys. 63 (1985) 555;
N.Yu. Reshetikhin, The spectrum of the transfer matrices connected with KacMoody algebras, Lett. Math.
Phys. 14 (1987) 235.
[32] L. Mezincescu, R.I. Nepomechie, Analytical Bethe ansatz for quantum algebra invariant spin chains, Nucl.
Phys. B 372 (1992) 597.
[33] A. Kuniba, J. Suzuki, Analytic Bethe ansatz for fundamental representations of Yangians, Commun. Math.
Phys. 173 (1995) 225, hep-th/9406180.
(2)
[34] S. Artz, L. Mezincescu, R.I. Nepomechie, Spectrum of transfer matrix for Uq (Bn ) invariant A2n open spin
chains, Int. J. Mod. Phys. A 10 (1995) 1937, hep-th/9409130.
(2)
(1)
(1)
(1)
[35] S. Artz, L. Mezincescu, R.I. Nepomechie, Analytical Bethe ansatz for A2n1 , Bn , Cn , Dn quantum
algebra invariant open spin chains, J. Phys. A 28 (1995) 5131, hep-th/9504085.
(1)
[36] A. Doikou, Fusion and analytical Bethe ansatz for the An1 open spin chain, J. Phys. A 33 (2000) 4755;
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
505
A. Doikou, Quantum spin chain with soliton nonpreserving boundary conditions, J. Phys. A 33 (2000)
8797.
R.J. Baxter, Partition function of the eight-vertex lattice model, Ann. Phys. 70 (1972) 193;
R.J. Baxter, J. Stat. Phys. 8 (1973) 25;
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, 1982.
V.E. Korepin, New effects in the massive Thirring model: repulsive case, Commun. Math. Phys. 76 (1980)
165;
V.E. Korepin, G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method, Correlation Functions and
Algebraic Bethe Ansatz, Cambridge Univ. Press, 1993.
L. Mezincescu, R.I. Nepomechie, Fusion procedure for open chains, J. Phys. A 25 (1992) 2533.
Y.-K. Zhou, Row transfer matrix functional relations for Baxters eight-vertex and six-vertex models with
open boundaries via more general reflection matrices, Nucl. Phys. B 458 (1996) 504, hep-th/9510095.
A. Doikou, R.I. Nepomechie, Bulk and boundary S matrices for the su(N ) chain, Nucl. Phys. B 521 (1998)
547, hep-th/9803118;
(1)
A. Doikou, R.I. Nepomechie, Duality and quantum algebra symmetry of the An1 open spin chain with
diagonal boundary fields, Nucl. Phys. B 530 (1998) 641, hep-th/9807065.
V. Korepin, Direct calculation of the S matrix in the massive Thirring model, Theor. Math. Phys. 41 (1979)
953.
N. Andrei, C. Destri, Dynamical symmetry breaking and fractionization in a new integrable model, Nucl.
Phys. B 231 (1984) 445.
M.J. Martins, Bethe ansatz solution of the Osp(1|2n) invariant spin chain, Phys. Lett. B 359 (1995) 334.
V.V. Bazhanov, N.Yu. Reshetikhin, Critical RSOS models and conformal field theory, Int. J. Mod. Phys. A 4
(1989) 115.
R.I. Nepomechie, Functional relations and Bethe ansatz for the XXZ chain, hep-th/0211001.
J.-P. Cao, H.-Q. Lin, K.-J. Shi, Y. Wang, Exact solutions and elementary excitations in the XXZ spin chain
with unparallel boundary fields, cond-mat/0212163.

Test of asymptotic freedom and scaling hypothesis

in the 2d O(3) sigma model
J. Balog a , P. Weisz b
a Research Institute for Particle and Nuclear Physics, H-1525 Budapest 114 Pf. 49, Hungary
b Max-Planck-Institut fr Physik, D-80805 Munich, Germany
Abstract
The 7-particle form factors of the fundamental spin field of the O(3) non-linear -model are
constructed. We calculate the corresponding contribution to the spinspin correlation function, and
compare with predictions from the spectral density scaling hypothesis. The resulting approximation
to the spinspin correlation function agrees well with that computed in renormalized (asymptotically
free) perturbation theory in the expected energy range. Further we observe simple lower and upper
bounds for the sum of the absolute square of the form factors which may be of use for analytic
estimates.
PACS: 11.10.Jj; 11.10.Kk
1. Introduction
(1 + 1)-dimensional integrable models are theoretical laboratories where one can study
selected problems in field theory. These include the question whether the continuum limit
of the lattice regularized version of a field theory coincides with the theory constructed
by continuum techniques. One can study here structural questions related to asymptotic
freedom, topological excitations etc., and, in the case that the theories agree, the nature of
lattice artifacts.
The form factor bootstrap method [1,2] of constructive field theory has been used
to construct many integrable models. The method consists of three main steps. The
starting point is the exact scattering matrix that is calculated (sometimes conjectured)
E-mail address: balog@rmki.kfki.hu (J. Balog).
doi:10.1016/S0550-3213(03)00572-8
J. Balog, P. Weisz / Nuclear Physics B 668 [FS] (2003) 506516
507
by the bootstrap method [3]. Next the form factors (matrix elements of local operators
between scattering states) are calculated by solving the form factor (FF) axioms [2]. Finally
correlation functions are constructed by inserting a complete set of intermediate states
between the local operators.
The program has so far only been carried out completely for the Ising model [4]. The Smatrix is known for many integrable models and if there are no internal degrees of freedom
also the many-particle form factors are easily constructed as a multiple product of the basic
two-particle form factor (which is known again for most models). On the other hand, for
models with internal degrees of freedom the solution of the FF axioms, matrix functional
difference equations in this case, is not known in general. Nevertheless, the form factors
of some models belonging to this class, the sine-Gordon model, the chiral GrossNeveu
model, and the O(3) and O(4) non-linear -models, are available [2,5]. Finally (with the
exception of the Ising model) the r-particle contributions to the correlation functions can
only be computed numerically and therefore this sum has to be truncated after the first few
terms.
Among models with internal (isospin) degrees of freedom, it is the O(3) model, where
one can go furthest. This is because in this special case the many-particle form factors
(after removing a simple factor) are polynomials in the particle rapidities which can be
determined recursively [6]. Fortunately it is also a model which is of particular interest
since it has many properties akin to YangMills theory in 4 dimensions in that it exhibits
asymptotic freedom and has instanton solutions.
The question mentioned above concerning the equivalence between the FF construction
and that from the lattice regularization has been addressed in Ref. [7]. The two
constructions are both non-perturbative and in both cases it is the low energy properties
that are most easily accessible. However, in the bootstrap approach it is simpler to obtain
reliable results at higher energies. For example, for the Fourier transform of the 2-point
correlation function of Noether currents, the contributions of higher r-particle states
become significant at typical energies rapidly increasing with r. Of course, despite the
fact that in practice renormalized perturbation theoretic computations often seem to be
quite accurate down to unexpectedly low energies, the property of asymptotic freedom is
a statement pertaining to asymptotically high energies, and to establish this property in
the bootstrap framework it is necessary to have rigorous knowledge on the contributions
of r-particle states with arbitrarily large r. Although the latter has not yet been achieved,
based on the exact form factors up to 6 particles, Max Niedermaier and one of the present
authors (J.B.) presented convincing evidence for remarkable scaling properties of the
higher intermediate particle contributions [8,9], which are very powerful because they
make it possible to include dominant contributions from all states in the form factor
expansion of correlation functions. Furthermore the scaling hypothesis is completely
compatible with asymptotic freedom.
Although the lattice regularization is an important non-perturbative method (and
practically the only one available in 4d QCD), there are many physical phenomena
which are inaccessible in this approach, for example, nuclear structure functions at small
Bjorken x. In the framework of the bootstrap approach in 2d however, we hope that
properties of the structure functions at small x can be extracted, an insight which would
hopefully also be relevant for QCD. This hope would be realized if we could find evidence
508
for scaling properties of the structure functions similar to those already observed for the
2-point function spectral densities.
With this goal we have initiated a project to compute the structure functions in the
O(3) sigma model, and with the known form factors we have computed contributions from
intermediate states up to 5 particles. To establish their scaling properties however it is
helpful to extend the list of known form factors to higher particle number. In this paper we
discuss the calculation of the 7-particle form factors, which although in principle trivial, is
technically challenging because for r = 7 one has to deal with quite large polynomials. We
will report on the structure functions in a future publication. Here we restrict attention to
the contribution of the 7-particle intermediate states to the spinspin correlation function,
which yields an additional test of the scaling hypothesis. We also exhibit simple bounds
on the square of the known form factor polynomials, which if they could be generalized to
an arbitrary number of particles might be useful for analytic estimates to establish general
properties.
2. Some basic definitions

The central object of attention in this paper is the 7-particle form factor of the spin
operator of the O(3) model. This is the r = 7 case of
2
0|S a (0)|a1, 1 ; . . . ; ar , r = faa1 ...ar (1 , . . . , r ),
(1)
where the particles are labelled by their isospin indices and rapidities
and the spin operator
S a also carries an isospin index. The normalization factor 2/ is introduced here for
later convenience. The spin operator itself is normalized by
0|S a (0)|a1, 1 = aa1 ,
b, |a, = 4 ab ( ),
which means that the Minkowski propagator has unit residue at

physical mass of the O(3) particle.
Introducing the squared form factors

faa1 ...ar (1 , . . . , r ) fab1 ...ar (1 , . . . , r )
ab F (r)(u) =
p2
(2)
= M 2,
where M is the
(3)
a1 ...ar
the spectral density is given by
spin
4 (2k+1)
() =
(),
(4)
k=0
where
(r)
() =

du1 dur1 (r)
F (u) M (r) (u) .
r1
(4)
(5)
Here uj = j j +1 are rapidity differences and M (r) (u) is the r-particle invariant mass.
The Fourier transform of the correlation function is represented as the Stieltjes transform
509
of the spectral density:

I
spin
2
p =

d
0
spin ()
.
p 2 + 2
(6)
We will compare (the truncation of) (6) to the results of perturbation theory. The 2-loop
order perturbative result is [10]

2
1
2 spin 2
p I
(7)
+ (2 + 0 ) + (2 + 0 )(p) + O (p) .
p = 1
(p)
Here the running coupling function (p) is the solution of
p
1
+ ln (p) = ln
(p)
M
(8)
and the parameter 0 gives the connection between the perturbative mass parameter MS
and the exact mass gap M. In the O(3) model their ratio is known exactly [11]:
0 = ln
M
= ln 8 1 1.07944.
MS
(9)
The overall constant 1 cannot be calculated in perturbation theory, but has been
determined exactly [8] using the scaling hypothesis (see also Section 5):
4
.
(10)
3 2
Instead of the physical form factor (1) it is convenient to consider the reduced form
factors gaa1 ...ar defined by
1 =
faa1 ...ar (1 , . . . , r ) =
Here
(1 , . . . , r ) =
3r
2 1
(1 , . . . , r )gaa1 ...ar (1 , . . . , r ).
(i j ),
(11)
(12)
i<j
with
( ) =
tanh2 .
(2i )
2
(13)
We note that (13) is the basic 2-particle form factor and the product (12) would be the
complete r-particle form factor if there were no isospin degrees of freedom.
The advantage of using the reduced form factors is that they are polynomials and can
(in principle) be recursively calculated [6]. The simplest non-trivial case is r = 3:
gaa1 a2 a3 ( ) = aa3 a1 a2 (2 1 ) + aa2 a1 a3 (1 3 2i) + aa1 a2 a3 (3 2 ).
(14)
510
The higher reduced form factors very rapidly become complicated because of the large
number of isospin components and rapidity variables.
The quantity entering the spectral densities and two-point functions is the absolute
square of the form factors, summed over the internal symmetry indices. For the reduced
form factors the corresponding quantities are
G(r) (1 , . . . , r ) =
2
1 a
ga1 ...ar (1 , . . . , r ) .
3 aa ...a
1
(15)
For the r = 3 example we get

G(3) (1 , 2 , 3 ) = 2 (1 2 )2 + (1 3 )2 + (2 3 )2 + 12 2 .
(16)
3. The 7-particle form factors

The 7-particle reduced form factors
gaa1 ...a7 (1 , . . . , 7 )
(17)
are polynomials of degree 15 in the rapidity differences. To count the independent isospin
components we can obviously fix the operator index a = 3 since the rest of the components
are related by simple isospin transformations. Then we get 274 non-vanishing components
not counting terms related by the 1 2 isospin symmetry twice. These components can
be calculated from the known 5-particle form factors using the inhomogeneous FF axioms
[6]. The resulting polynomials contain typically 104 terms. Actually it is not necessary
to do this calculation for all the 274 components since we can find 4 components such that
the rest can be obtained from them by using the homogeneous FF axioms (corresponding
to permutations of the rapidity variables). After having computed all terms we checked
once more that all the FF axioms are satisfied by (17) and also that the coefficients of the
leading terms in the last variable (those proportional to 75 ) are proportional to the 6-particle
reduced form factors as they should [8].
After the calculation of the form factors the next step is to compute the sum of
the absolute square of the components since this is needed in the calculation of the 7particle contribution to the spinspin correlation function and other physical quantities.
This calculation is rather challenging because at intermediate steps one has to handle quite
large polynomials (of degree 30 in 7 variables consisting of about 4 105 terms). The
final result can be simplified enormously because, being a polynomial symmetric under
permutations of the rapidity variables, it can be expanded in terms of the basic symmetric
polynomials 1 , 2 , . . . , 7 (defined in Eq. (A.3) of Ref. [8]). Written in this way, the final
result consists of only 3214 terms and is reduced to a manageable size.1
To be sure that we have obtained the correct result for the square G(7)(1 , . . . , 7 )
we have performed a number of checks. First we checked that the leading terms
in the last variable (the coefficient of 710 ) are twice the analogous 6-particle square
1 Which is however still too large to usefully reproduce here.
511
G(6) (1 , . . . , 6 ) [8]. Next we checked that the overall leading terms (the terms of total
degree 30) agree with those of a simple symmetric polynomial P (7) (t) [8]. This is defined
by (for r particles)
(r)
P0 (t),
P (r) (t) =
(18)
perms
where
P0(r) (t) =
(i j )2 + t 2 .
(19)
j i>1
In (18) the sum extends over the r! permutations of the rapidity variables 1 , . . . , r
and in (19) the correction terms depending on the parameter t do not affect the overall
leading terms, they are included here for later convenience. Finally, we verified that
G(7) (1 , . . . , 7 ) vanishes at the points [8]
1 2 = 2 3 = i,
j = arbitrary
j > 3.
(20)
G(7)
passed all these non-trivial tests gives us some confidence in the

The fact that
correctness of our results.
4. The spinspin correlation function

Having calculated the square of the 7-particle form factors the corresponding contributions to the spinspin correlation function and the spectral density can now be computed
straightforwardly. The only difficulty is that to get correct numerical results one has to use
high precision arithmetic in order to avoid huge rounding errors in calculating G(7) . As
explained above, to put it into a manageable form, we expressed it in terms of the basic
symmetric polynomials 1 , . . . , 7 . Although it is a sum of absolute squares and thus it is
obviously positive, it is not manifestly positive in this form. Actually we experienced large
cancellations between positive and negative contributions. After rescaling all rapidities by
all coefficients of the polynomial become integers and we can illustrate the large degree
of cancellation by considering the following integer rapidities:
1 = 12,
2 = 11,
3 = 10,
4 = 9,
5 = 8,
6 = 7,
7 = 0.
(21)
We can exactly calculate the sum of positive and negative contributions in this case and we
get
positive part:
26512984286926120356867283491794696305689824,
negative part:
26512984286926120356855853186993819121689824,
total:
11430304800877184000000
showing a 21-digit cancellation here!

We have computed the 7-particle contribution to the spinspin correlation function
using the VEGAS integration routine, and a subroutine computing G(7) invoking quartic
512
Fig. 1. Plot of p 2 I spin (p 2 ) against ln(p/M), showing a comparison of the form factor approach to 2-loop
perturbation theory. The solid curve is the FF approximation to p 2 I spin (p 2 ) obtained by including all
intermediate states with 7 particles. The lower curves show the analogous approximations with 5, 3, 1
particles respectively. The normalization of the (dashed) PT curve is fixed according to (10). Finally the top curve
is the PT curve multiplied by a factor 1.05.
precision (32 digit) arithmetic. The results, compared with the prediction of two-loop
perturbation theory, are shown in Fig. 1. Note that here we have no free parameters at
our disposal in the perturbative calculation as would be the case in most other models. As
discussed above, we know the exact relation between the perturbative parameter and the
particle mass M and also the absolute normalization of the perturbative curve is available
in this model. The form factor results are in very good agreement with perturbation theory
in the expected (high) energy range. The small deviation for energies p/M 104 can be
accounted for by the contribution of r > 7 intermediate particles. To illustrate the fact that
this good agreement is quite non-trivial, in Fig. 1 we also show the perturbative result with
the overall factor changed arbitrarily by 5%.
5. Scaling
The 7-particle results also corroborate the scaling hypothesis [8] for the spectral
densities. To study this aspect we introduce the modified r-particle spectral density
depending on the logarithmic variable x by the definition
(r) () = R (r) (x),
= Mex .
(22)
513
Fig. 2. Illustration of the self-similarity property of the rescaled spectral densities. The plots show Y (r) (z)
(dashed) compared with Y (r+1) (z) (solid) for r = 3, 4, 5, 6.
These are defined for all r 2, where the cases with r even refer to the spectral densities
of the two-point function of the isospin Noether current. The graph of this function is a
bell-shaped curve starting as zero at x = ln r, reaching its maximum M (r) at x = (r) and
then slowly decreasing for larger x. Let us introduce the rescaled spectral density Y (r) by

1
(23)
R (r) (r) z .
(r)
M
It has been found (based on the study of up to 6 particles) that the shape of the rescaled
spectral density and the parameters (r) , M (r) satisfy self-similarity
Y (r) (z) =
lim Y (r) (z) Y (z)
(24)
with universal shape function Y (z) and asymptotic scaling,

M (r) M r ,
(r) r 1+
(25)
for large r, with some coefficients M , and exponents , . Whereas the properties of
the form factors are consistent with = 1 only, the other exponent can only be determined
numerically with result = 0.27. For those readers unfamiliar with this model we point
out the amusing fact that the self-similarity holds for all r 2 and thus interrelates the
spectral functions of two different isovector operators, the spin field and the conserved
vector current. Similarly (25) is equally valid for even and odd r values, provided we use
the normalization introduced in (1).
Self-similarity continues to be satisfied very well as demonstrated in Fig. 2. To test
asymptotic scaling, we used the fitted numerical values based on results for up to 6 particles
514
to predict for the r = 7 case

(7)
(7)
Msc
= 0.03188 and sc
= 17.73,
(26)
and compared it to the values directly determined from our 7-particle results
M (7) = 0.03189 and (7) = 17.77.
(27)
Similarly for the integrals

c
(r)

d
(r)
() =

dxR
(r)
(x),
(r)
dx (r)
R (x)
x
(28)
we predict
(7)
csc
= 1.464 and h(7)
sc = 0.04879,
(29)
and actually get

c(7) = 1.46(1) and h(7) = 0.0488(1).
(30)
6. Ultra-positivity
Because of the large cancellation between positive and negative terms in the representation of G(r) in terms of symmetric polynomials, it is natural to ask if there exists an
alternative representation that is manifestly positive. It is indeed possible to find such representations. If we arrange the rapidities in decreasing order (which is always possible
since the polynomial is symmetric under permutations) then all uj = j j +1 rapidity
differences are positive and we found (for all available particle numbers 3 r 7) that if
we expand G(r) in terms of these differences then all coefficients are positive. Moreover,
it is possible to find upper and lower bounds both of which are of the simple form (18)
such that, expanded in terms of the rapidity differences uj , each and every term in the
expansion of G(r) has smaller coefficient than the corresponding one of the upper bound
(r)
P (r) (tu ) and similarly larger than the coefficient of the corresponding term of the lower
(r)
bound P (r) (tl ).
The value of the parameter characterizing the lower bound is uniformly tl(r) = 2 for all
cases 3 r 7, whereas the tu(r) giving the lowest upper bound are
tu(3) = 2,
tu(4) = 9/4,
tu(5) = 2.434,
tu(6) = 2.576,
tu(7) = 2.688.
(31)
The last three values are numerically approximate2 and come from the requirement that
(r)
the constant term of P (r) (tu ) be larger than that of G(r) . This is clearly necessary, but at
least in the available cases for r 5 it is also sufficient.3
2 E.g., the exact value for the case r = 5 is t (5) = 2081/6 .
u
3 The case r = 4 is an exception since the requirement in this case yields a value (34/3)1/3 = 2.246, which is
not quite sufficient.
515
Fig. 3. Values of ln c(r) (circles) and of the product rc(r) (2)/(4 ) (squares). The error on ln c(7) is approximately
the radius of the circle.
The integrals over the spin and current spectral densities are given as sums of the
integrals defined in (28):

C
spin
d
M
spin
4 (2k+1)
() =
c
,
k=0

C
curr
d curr() =
M
c(2k+2).
k=0
(32)
An outstanding question is whether these are finite or infinite in the FF construction.

Certainly renormalized perturbation theory predicts that C spin = = C curr . Also the
validity of the scaling hypothesis requires this because in this scenario c(r) grows as r ;
the numerical evidence thereof is shown in Fig. 3.
If the bounds we found above for 3 r 7 continue to be true for all r, then the simple
structure of (18) allows to study the structure of the correlation function analytically. The
existence of an upper limit of simple form may help proving the existence of the correlation
function whereas the existence of the lower bound may facilitate the construction of a proof
that C spin = = C curr independent of the validity of the scaling hypothesis.
We are not going to discuss these questions further in this paper, but to illustrate the
usefulness of the existence of the simple lower bound we have calculated the integrals
c(r)(2). These are defined analogously to the ones in (28) using P (r) (2) instead of G(r) .
516
Table 1
Numerical values of the integrals c(r) , c(r) (2). Numerical errors are estimated as 1 on the last digit quoted
r
c(r) 0.785
(r)
c (2)
10
11
1.009
1.140
1.140
1.242
1.206
1.327
1.229
1.400
1.225
1.46
1.200
1.164
1.1.118
1.067
1.013
The simplicity of the integrand allows us to calculate these integrals numerically quite
effortlessly4 up to r = 11. The results of the numerical integration are given in Table 1.
Although the integrals c(r)(2) are (apparently) decreasing with r, it is possible that they
are decreasing slow enough to make the series (32) diverge. Indeed, as seen in Fig. 3,
rc(r)(2) is monotonically increasing for all r evaluated up to now.
Acknowledgements
J.B. wishes to thank the Max-Planck-Institut fr Physik in Munich where most of
this work has been done, for hospitality. This investigation was supported in part by the
Hungarian National Science Fund OTKA (under T030099 and T034299).
References
[1] M. Karowski, P. Weisz, Nucl. Phys. B 139 (1978) 455.
[2] F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, World Scientific,
Singapore, 1992.
[3] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[4] V. Yurov, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 6 (1991) 3419;
O. Babelon, D. Bernard, Phys. Lett. B 288 (1992) 113.
[5] F.A. Smirnov, Int. J. Mod. Phys. A 9 (1994) 5121.
[6] J. Balog, T. Hauer, Phys. Lett. B 337 (1994) 115.
[7] J. Balog, M. Niedermaier, F. Niedermayer, A. Patrascioiu, E. Seiler, P. Weisz, Phys. Rev. D 60 (1999)
094508.
[8] J. Balog, M. Niedermaier, Nucl. Phys. B 500 (1997) 421.
[9] J. Balog, M. Niedermaier, Phys. Rev. Lett. 78 (1997) 4151.
[10] U. Wolff, Nucl. Phys. B 334 (1990) 581.
[11] P. Hasenfratz, M. Maggiore, F. Niedermayer, Phys. Lett. B 245 (1990) 522;
P. Hasenfratz, F. Niedermayer, Phys. Lett. B 245 (1990) 529.
4 This is because one can represent c(r) (2) as an integral with the integrand just involving the polynomial
(r)
P0 (2) in Eq. (19) (instead all its permutations), at the cost of extending the limits of integration over the ui
from to +.

CUMULATIVE AUTHOR INDEX B661B668
Abel, S.A.
Acquaviva, V.
Adams, D.H.
Aglietti, U.
Akbar, M.M.
Akemann, G.
Alexandrov, S.Yu.
Alishahiha, M.
ALPHA Collaboration
lvarez, E.
lvarez-Gaum, L.
Antoniadis, I.
Armoni, A.
Armoni, A.
Arnaudon, D.
Arutyunov, G.
Arutyunov, G.
Astefanesei, D.
Avan, J.
Axenides, M.
B663 (2003) 197

B667 (2003) 119
B662 (2003) 220
B668 (2003) 3
B663 (2003) 215
B664 (2003) 457
B667 (2003) 90
B661 (2003) 174
B663 (2003) 3
B663 (2003) 365
B668 (2003) 293
B662 (2003) 40
B664 (2003) 233
B667 (2003) 170
B668 (2003) 469
B663 (2003) 163
B665 (2003) 273
B665 (2003) 594
B668 (2003) 469
B662 (2003) 170
Bais, F.A.
Balog, J.
Banerjee, R.
Barbieri, R.
Barbieri, R.
Bardakci, K.
Bartolo, N.
Basu-Mallick, B.
Beccaria, M.
Becchi, C.
Becker, K.
Becker, M.
Beenakker, W.
Beisert, N.
Belitsky, A.V.
Bellucci, S.
Bellucci, S.
Bena, I.
B666 (2003) 243

B668 (2003) 506
B668 (2003) 179
B663 (2003) 141
B668 (2003) 273
B661 (2003) 235
B667 (2003) 119
B668 (2003) 415
B663 (2003) 394
B664 (2003) 371
B666 (2003) 144
B666 (2003) 144
B667 (2003) 359
B664 (2003) 131
B667 (2003) 3
B663 (2003) 605
B665 (2003) 402
B664 (2003) 45
0550-3213/2003 Published by Elsevier B.V.

doi:10.1016/S0550-3213(03)00711-9
Benakli, K.
Benson, D.
Berezin, V.
Bhattacharyya, T.
Bigi, I.I.
Bilal, A.
Blaek, T.
Blumenhagen, R.
Boer, D.
Bonciani, R.
Bonciani, R.
Bourbonnais, C.
Bouttier, J.
Brandenburg, A.
Brax, P.
Brignole, A.
Brower, R.C.
Brower, R.C.
Bruckmann, F.
Buchbinder, I.L.
Buchmller, W.
B662 (2003) 40
B665 (2003) 367
B661 (2003) 409
B668 (2003) 415
B665 (2003) 367
B663 (2003) 343
B662 (2003) 359
B663 (2003) 319
B667 (2003) 201
B661 (2003) 289
B668 (2003) 3
B663 (2003) 568
B663 (2003) 535
B667 (2003) 394
B667 (2003) 149
B666 (2003) 105
B661 (2003) 344
B662 (2003) 393
B666 (2003) 197
B665 (2003) 402
B665 (2003) 445
Cacciari, M.
Campanario, F.
Cao, J.
Casas, J.A.
Chakraborty, B.
Chandrasekharan, S.
Chapovsky, A.P.
Chetyrkin, K.G.
Chitov, G.Y.
Choi, K.-S.
Conde, J.
Cramp, N.
Cvetic, M.
B664 (2003) 299

B663 (2003) 280
B663 (2003) 487
B666 (2003) 105
B668 (2003) 179
B662 (2003) 220
B667 (2003) 359
B666 (2003) 289
B663 (2003) 568
B662 (2003) 476
B663 (2003) 365
B668 (2003) 469
B662 (2003) 89
Daleo, A.
Darriulat, P.
Dasgupta, K.
de Azcrraga, J.A.
B662 (2003) 334

B661 (2003) 3
B666 (2003) 144
B662 (2003) 185
518
de Boer, J.
de Haro, S.
DElia, M.
Demasure, Y.
Denner, A.
Derkachov, S..
Deser, S.
Di Bari, P.
Di Francesco, P.
Dinh, P.N.
Dobashi, S.
Doikou, A.
Doikou, A.
Dolan, F.A.
Dorey, P.
Dorey, P.
Dorogovtsev, S.N.
Dotsenko, V.S.
Dubovsky, S.L.
Dung, N.T.
B665 (2003) 545

B664 (2003) 45
B661 (2003) 139
B661 (2003) 153
B662 (2003) 299
B661 (2003) 533
B662 (2003) 379
B665 (2003) 445
B663 (2003) 535
B661 (2003) 3
B665 (2003) 94
B668 (2003) 447
B668 (2003) 469
B665 (2003) 273
B661 (2003) 425
B661 (2003) 464
B666 (2003) 396
B664 (2003) 477
B664 (2003) 407
B661 (2003) 3
Emmanuel-Costa, D.
Engquist, J.
Espinosa, J.R.
Etesi, G.
Evlampiev, K.
B661 (2003) 62
B664 (2003) 439
B666 (2003) 105
B662 (2003) 511
B662 (2003) 120
Faisst, M.
Falkowski, A.
Farhi, E.
Feng, B.
Feverati, G.
Floratos, E.
Forger, M.
Frappat, L.
Freidel, L.
Freitas, A.
Frolov, S.
Fursaev, D.V.
Fyodorov, Y.V.
B665 (2003) 649

B667 (2003) 149
B665 (2003) 623
B661 (2003) 113
B663 (2003) 409
B662 (2003) 170
B667 (2003) 435
B668 (2003) 469
B662 (2003) 279
B666 (2003) 305
B668 (2003) 77
B664 (2003) 403
B664 (2003) 457
Ganjali, M.A.
Garca Canal, C.A.
Gardi, E.
Geyer, B.
Ghodsi, A.
Giedt, J.
Giudice, G.F.
Giusto, S.
Golec-Biernat, K.
Gomis, J.
Gonzlez, J.
Gorsky, A.S.
Gracey, J.A.
B661 (2003) 174

B662 (2003) 334
B664 (2003) 299
B662 (2003) 531
B661 (2003) 174
B668 (2003) 138
B663 (2003) 377
B664 (2003) 371
B668 (2003) 345
B665 (2003) 49
B663 (2003) 605
B667 (2003) 3
B662 (2003) 247
Gracey, J.A.
Graham, N.
Groot Nibbelink, S.
Groot Nibbelink, S.
Grozin, A.G.
Grozin, A.G.
Guitter, E.
Gustavsson, A.
B667 (2003) 242

B665 (2003) 623
B663 (2003) 60
B665 (2003) 236
B663 (2003) 280
B666 (2003) 289
B663 (2003) 535
B667 (2003) 111
Hagiwara, K.
Hall, L.J.
Harmark, T.
Herdeiro, C.A.R.
Hernndez, L.
Hieu, B.D.
Hiller, J.R.
Hofmann, R.
Holland, K.
Hollik, W.
Honecker, G.
Hou, B.Y.
Huber, P.
Huber, S.J.
Hwang, K.
B668 (2003) 364

B663 (2003) 141
B662 (2003) 3
B665 (2003) 189
B663 (2003) 365
B661 (2003) 3
B661 (2003) 99
B668 (2003) 151
B668 (2003) 207
B666 (2003) 305
B666 (2003) 175
B663 (2003) 467
B665 (2003) 487
B666 (2003) 269
B662 (2003) 476
Ichinose, I.
Imai, T.
Imbimbo, C.
Intriligator, K.
Isaev, A.P.
Ito, M.
Ivanov, N.Ya.
Izquierdo, J.M.
B663 (2003) 520

B665 (2003) 520
B664 (2003) 371
B667 (2003) 183
B662 (2003) 461
B668 (2003) 322
B666 (2003) 88
B662 (2003) 185
Jack, I.
Jacobsen, J.L.
Jaffe, R.L.
Janik, R.A.
Jones, D.R.T.
B662 (2003) 63
B664 (2003) 477
B665 (2003) 623
B661 (2003) 153
B662 (2003) 63
Kamimura, K.
Kaminsky, K.
Kanaki, A.
Kawai, H.
Kazakov, V.A.
Kehagias, A.
Khater, W.
Khemani, V.
Kim, J.E.
Kimura, Y.
King, S.F.
Kitazawa, Y.
Klebanov, I.R.
Kleinert, H.
B662 (2003) 491

B663 (2003) 33
B667 (2003) 359
B664 (2003) 185
B667 (2003) 90
B662 (2003) 170
B661 (2003) 209
B665 (2003) 623
B662 (2003) 476
B664 (2003) 512
B662 (2003) 359
B665 (2003) 520
B664 (2003) 3
B666 (2003) 361
519
Knechtli, F.
Korchemsky, G.P.
Korchemsky, G.P.
Kostov, I.K.
Kotikov, A.V.
Kraus, E.
Kristjansen, C.
Krykhtin, V.A.
Khn, J.H.
Kumar, K.
Kuroki, T.
Kurylov, A.
Kutasov, D.
B663 (2003) 3
B661 (2003) 533
B667 (2003) 3
B667 (2003) 90
B661 (2003) 19
B661 (2003) 83
B664 (2003) 131
B665 (2003) 402
B665 (2003) 649
B668 (2003) 179
B664 (2003) 185
B667 (2003) 321
B666 (2003) 56
Meggiolaro, E.
Melles, M.
Melnikov, K.
Mendes, J.F.F.
Metzger, S.
Meyer, H.B.
Minkowski, P.
Moore, J.E.
Morita, T.
Moriyama, S.
Morozov, A.
Mulders, P.J.
Mlsch, D.
B665 (2003) 425

B662 (2003) 299
B662 (2003) 409
B666 (2003) 396
B663 (2003) 343
B668 (2003) 111
B668 (2003) 207
B661 (2003) 514
B664 (2003) 185
B665 (2003) 49
B666 (2003) 311
B667 (2003) 201
B662 (2003) 531
Lalak, Z.
Laliena, V.
Larosa, M.
Laugier, A.
Lee, H.K.
Lee, K.
Lehners, J.-L.
Leonhardt, T.
Li, Y.Q.
Lin, H.-Q.
Lin, H.Q.
Lindner, M.
Lindstrm, U.
Lipatov, L.N.
Livine, E.R.
Louapre, D.
Lowe, D.A.
L, H.
L, H.
Ludwig, A.W.W.
Lst, D.
Lynker, M.
B667 (2003) 149

B668 (2003) 403
B667 (2003) 261
B662 (2003) 40
B665 (2003) 153
B665 (2003) 179
B661 (2003) 273
B667 (2003) 413
B666 (2003) 337
B663 (2003) 487
B666 (2003) 337
B665 (2003) 487
B662 (2003) 147
B661 (2003) 19
B663 (2003) 231
B662 (2003) 279
B667 (2003) 55
B662 (2003) 89
B668 (2003) 237
B661 (2003) 577
B663 (2003) 319
B667 (2003) 484
Nadolsky, P.M.
Nadolsky, P.M.
Nan, C.M.
Nason, P.
Nastase, H.
Navarro, I.
Niarchos, V.
Nicolai, H.
Niemi, A.J.
Nilles, H.P.
Ngrdi, D.
Nogueira, F.S.
Nomura, Y.
Nyawelo, T.S.
B666 (2003) 3
B666 (2003) 31
B663 (2003) 591
B667 (2003) 394
B667 (2003) 55
B666 (2003) 105
B666 (2003) 56
B668 (2003) 167
B666 (2003) 311
B665 (2003) 236
B666 (2003) 197
B666 (2003) 361
B663 (2003) 141
B663 (2003) 60
Maillard, T.
Majumdar, P.
Manashov, A.N.
Mannel, T.
Mannel, Th.
Manvelyan, R.
Marandella, G.
Marandella, G.
Marchi, M.
Martin, L.C.
Martucci, L.
Masina, I.
Mastrolia, P.
Mastrolia, P.
Matarrese, S.
Mathur, S.D.
Mawatari, K.
B662 (2003) 40
B664 (2003) 213
B661 (2003) 533
B663 (2003) 280
B665 (2003) 367
B667 (2003) 413
B663 (2003) 141
B668 (2003) 273
B665 (2003) 425
B668 (2003) 335
B666 (2003) 230
B661 (2003) 365
B661 (2003) 289
B664 (2003) 341
B667 (2003) 119
B661 (2003) 344
B668 (2003) 364
Okawa, Y.
Okui, T.
Oleari, C.
Olechowski, M.
Oliver, S.J.
Onorato, P.
Ooguri, H.
Oriti, D.
Osborn, H.
Oshimo, N.
Osland, P.
Owen, A.W.
B663 (2003) 33
B663 (2003) 141
B667 (2003) 394
B665 (2003) 236
B663 (2003) 141
B663 (2003) 605
B663 (2003) 33
B663 (2003) 231
B665 (2003) 273
B668 (2003) 258
B661 (2003) 209
B663 (2003) 197
Paccetti Correia, F.
Papadopoulos, C.G.
Papucci, M.
Papucci, M.
Park, J.
Parvizi, S.
Pearce, P.A.
Pepe, M.
Periwal, V.
Phuong, P.T.
Picn, M.
Pijlman, F.
B668 (2003) 151

B667 (2003) 359
B663 (2003) 141
B668 (2003) 273
B665 (2003) 49
B661 (2003) 174
B663 (2003) 409
B668 (2003) 207
B667 (2003) 484
B661 (2003) 3
B662 (2003) 185
B667 (2003) 201
520
Pinsky, S.S.
Pittau, R.
Plmacher, M.
Pocklington, A.
Pocklington, A.
Pons, J.M.
Pope, C.N.
Pope, C.N.
Pozzorini, S.
Pradisi, G.
Prokushkin, S.
B661 (2003) 99
B667 (2003) 359
B665 (2003) 445
B661 (2003) 425
B661 (2003) 464
B665 (2003) 129
B662 (2003) 89
B668 (2003) 237
B662 (2003) 299
B667 (2003) 261
B666 (2003) 144
Radu, E.
Ragoucy, E.
Ramgoolam, S.
Ramsey-Musolf, M.J.
Ravindran, V.
Remiddi, E.
Remiddi, E.
Renard, F.M.
Riccioni, F.
Riotto, A.
Rocek, M.
Roiban, R.
Roiban, R.
Rhl, W.
Rupp, C.
B665 (2003) 594

B668 (2003) 469
B667 (2003) 55
B667 (2003) 321
B665 (2003) 325
B661 (2003) 289
B664 (2003) 341
B663 (2003) 394
B663 (2003) 60
B667 (2003) 119
B662 (2003) 147
B664 (2003) 45
B665 (2003) 211
B667 (2003) 413
B661 (2003) 83
Sakaguchi, M.
Saleur, H.
Salvay, M.J.
Samtleben, H.
Samukhin, A.N.
Santachiara, R.
Sasaki, R.
Sassot, R.
Savoy, C.A.
Schimmrigk, R.
Schmidt, M.G.
Schubert, C.
Schwetz, T.
Segal, A.Y.
Seidensticker, T.
Seki, S.
Serbo, V.G.
Sezgin, E.
Sezgin, E.
Shafi, Q.
Sharpe, E.
Shi, K.-J.
Shifman, M.
Shifman, M.
Shik, H.Y.
Shimada, H.
B662 (2003) 491

B663 (2003) 443
B663 (2003) 591
B668 (2003) 167
B666 (2003) 396
B664 (2003) 477
B663 (2003) 467
B662 (2003) 334
B661 (2003) 365
B667 (2003) 484
B668 (2003) 151
B668 (2003) 335
B665 (2003) 487
B664 (2003) 59
B665 (2003) 649
B661 (2003) 257
B662 (2003) 409
B664 (2003) 439
B668 (2003) 237
B665 (2003) 469
B664 (2003) 21
B663 (2003) 487
B664 (2003) 233
B667 (2003) 170
B666 (2003) 337
B665 (2003) 94
Sibiryakov, S.M.
Sibold, K.
Siegel, W.
Silva, P.J.
Sin, S.-J.
Singh, H.
Skenderis, K.
Smith, J.
Sokatchev, E.
Sokatchev, E.
Solodukhin, S.N.
Stasto, A.M.
Staudacher, M.
Stefanski Jr., B.
Stelle, K.S.
Stelle, K.S.
Striet, J.
Strumia, A.
Su, S.
Sudb, A.
Sugawara, Y.
Sundell, P.
B664 (2003) 407

B661 (2003) 83
B665 (2003) 179
B666 (2003) 230
B667 (2003) 310
B661 (2003) 394
B665 (2003) 3
B665 (2003) 325
B663 (2003) 163
B665 (2003) 273
B665 (2003) 545
B668 (2003) 345
B664 (2003) 131
B666 (2003) 71
B661 (2003) 273
B662 (2003) 89
B666 (2003) 243
B663 (2003) 377
B667 (2003) 321
B666 (2003) 361
B661 (2003) 191
B664 (2003) 439
Takayama, Y.
Takayanagi, T.
Takeda, K.
Talavera, P.
Tamai, K.
Tan, C.-I.
Tan, C.-I.
Tanaka, T.
Tatar, R.
Tateo, R.
Tateo, R.
Tavartkiladze, Z.
Tavartkiladze, Z.
Taylor, M.
Taylor, T.R.
Teper, M.J.
Thao, N.T.
Thieu, D.Q.
Thorn, C.B.
Thuan, V.V.
Tobe, K.
Tomino, D.
Trittmann, U.
Tseytlin, A.A.
Tseytlin, A.A.
Tsutsui, I.
B665 (2003) 520

B662 (2003) 3
B663 (2003) 520
B665 (2003) 129
B668 (2003) 385
B661 (2003) 344
B662 (2003) 393
B662 (2003) 413
B665 (2003) 211
B661 (2003) 425
B661 (2003) 464
B665 (2003) 469
B668 (2003) 151
B665 (2003) 3
B663 (2003) 319
B668 (2003) 111
B661 (2003) 3
B661 (2003) 3
B661 (2003) 235
B661 (2003) 3
B663 (2003) 123
B665 (2003) 520
B661 (2003) 99
B664 (2003) 247
B668 (2003) 77
B662 (2003) 447
Uchino, T.
Uraltsev, N.
B662 (2003) 447

B665 (2003) 367
van Baal, P.
van Holten, J.W.
B666 (2003) 197

B663 (2003) 60
van Neerven, W.L.

van Nieuwenhuizen, P.
Varela, O.
Vzquez-Mozo, M.A.
Veneziano, G.
Veretin, O.
Verzegnassi, C.
Villanueva Sandoval, V.M.
B665 (2003) 325

B662 (2003) 147
B662 (2003) 185
B668 (2003) 293
B667 (2003) 170
B665 (2003) 649
B663 (2003) 394
B668 (2003) 335
Walcher, J.
Waldron, A.
Walter, M.G.A.
Walter, W.
Wang, W.
Wang, Y.
Wecht, B.
Wehefritz-Kaufmann, B.
Weigel, H.
Weiglein, G.
Weisz, P.
Wells, J.D.
Wiese, K.J.
Wiese, U.-J.
Wiesenfeldt, S.
Winter, W.
B665 (2003) 211

B662 (2003) 379
B665 (2003) 236
B666 (2003) 305
B667 (2003) 349
B663 (2003) 487
B667 (2003) 183
B663 (2003) 443
B665 (2003) 623
B666 (2003) 305
B668 (2003) 506
B663 (2003) 123
B661 (2003) 577
B668 (2003) 207
B661 (2003) 62
B665 (2003) 487
521
Winterhalder, A.
Witten, E.
Wolff, U.
Wu, J.-B.
Wu, J.-B.
Wu, X.
B667 (2003) 435

B664 (2003) 3
B663 (2003) 3
B663 (2003) 79
B663 (2003) 95
B665 (2003) 153
Yang, W.-L.
Ylmaz, N.T.
Yokoya, H.
Yoneya, T.
Yuan, C.-P.
Yuan, C.-P.
Yue, C.
Yung, A.
B663 (2003) 467

B664 (2003) 357
B668 (2003) 364
B665 (2003) 94
B666 (2003) 3
B666 (2003) 31
B667 (2003) 349
B662 (2003) 120
ZeuthenRome (ZeRo) Collaboration

Zheng, Z.-J.
Zheng, Z.-J.
Zhu, C.-J.
Zhu, C.-J.
Znojil, M.
Zong, H.
B664 (2003) 276

B663 (2003) 79
B663 (2003) 95
B663 (2003) 79
B663 (2003) 95
B662 (2003) 554
B667 (2003) 349

Nucl - Phys.B v.668

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Nucl - Phys.B v.668

Transféré par

Droits d'auteur :

Formats disponibles

Nuclear Physics B 668 (2003) 376

Master integrals with one massive propagator for

Received 15 April 2003; accepted 8 July 2003

E-mail addresses: ugo.aglietti@roma1.infn.it (U. Aglietti), roberto.bonciani@physik.uni-freiburg.de

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

2 The minus sign in the definition of x is inserted for convenience.

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

The amplitudes involving a single massive propagator, i.e., a single W or Z emission,

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

|s| m2 of all the six-denominator amplitudes. In the conclusions we summarize the

2. The reduction to master integrals

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

4 Note a difference of sign with respect to the more common definition.

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

Finally, diagram (f) is actually a four-denominator topology, indicated in Fig. 6(g).

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

We then have the linear relations

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

ni > 0, which we call t:

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

Fig. 4. The set of 9 independent 6-denominator topologies.

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

Fig. 5. The set of 22 independent 5-denominator topologies.

with i = 1, 2 and v = k1 , k2 , p1 , p2 . The integral above vanishes because of the following

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

Fig. 6. The set of 17 independent 4-denominator topologies.

Fig. 7. The set of 4 independent 3-denominator topologies.

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

(1) the amplitude itself: Topo(n1 , n2 , n3 , n4 , n5 , n6 ; n7 );

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

Topo(1, 1, 1, 1, 1, 0; 1) Topo(1, 1, 1, 1, 1, 1; 0).

and minus the sum of the negative indices

where, as before, i = 1, 2 and v = k1 , k2 , p1 , p2 . For a given set of indices {si , nj } of

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

symmetries of various topologies generate additional identities to be combined with the

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

The algorithms explained in this section were implemented in a chain of programs

3. Evaluation of the MIs: the differential equations method

F2 (s, a, ) = Topo(1, 0, 0, 1, 1, 1; 1).

Kij (s, a, )Fj (s, a, ).

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

Fi (x, a, ) = s Fi (s, a, ),

a Fi (x, a, ) = s Fi (s, a, ) + a Fi (s, a, ) = Fi (x, a, ).

Fi (x, a, ) = a di /2 fi (x, ),

where di is the mass dimension of the MI: di /2 = D

F4 (x, ) = Topo(1, 0, 0, 0, 1, 1; 0).

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

The double polethe leading singularityobeys a differential equation of the form

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

(2 3)F2 (1; ) + (1 )F3 () + (1 )F4 (1; ) = 0.

in order to have the system in a triangular form. A triangularization to order  is actually

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

The MIs can be expressed in terms of the one-dimensional harmonic polylogarithms

where the basic weight functions are defined as

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

4. Results for the master integrals

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

U. Aglietti, R. Bonciani / Nuclear Physics B 668 (2003) 376

x 13H (1, x) 8H (1, 1, x)

F2 (s, a, ) = Topo(1, 0, 0, 1, 1, 1; 1).

Kij (s, a, )Fj (s, a, ).

Fi (x, a, ) = s Fi (s, a, ),

a Fi (x, a, ) = s Fi (s, a, ) + a Fi (s, a, ) = Fi (x, a, ).

Fi (x, a, ) = a di /2 fi (x, ),

F4 (x, ) = Topo(1, 0, 0, 0, 1, 1; 0).

(2 3)F2 (1; ) + (1 )F3 () + (1 )F4 (1; ) = 0.

in order to have the system in a triangular form. A triangularization to order is actually