Nucl - Phys.B v.612

Nuclear Physics B 612 (2001) 324
www.elsevier.com/locate/npe
Schrdinger functional at negative flavour number

ALPHA Collaboration
Bernd Gehrmann, Juri Rolf, Stefan Kurth, Ulli Wolff
Institut fr Physik, Humboldt-Universitt zu Berlin, Invalidenstr. 110, 10115 Berlin, Germany
Received 2 July 2001; accepted 19 July 2001
Abstract
The scaling of the Schrdinger functional coupling is studied numerically and perturbatively for
an SU(3) lattice gauge field coupled to an O(a) improved bosonic spinor field. This corresponds to
QCD with minus two light flavours and is used as a numerically less costly test case for real QCD.
A suitable algorithm is developed, and the influence of the matter fields on the continuum limit and
the lattice artefacts are studied in detail. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ha; 12.38.Bx; 11.10.Gh; 11.10.Hi
1. Introduction
The strong coupling constant s of QCD is of particular theoretical interest. On the one
hand, it can be extracted from jet events which are a property of the strong interaction
at large energies. On the other hand, as has been discussed in detail in [1], the running
coupling can be computed in lattice gauge theory. There, the parameters may be fixed
in the non-perturbative hadronic regime, taking as experimental input, for example, the
pion decay constant and the masses of the , K, D and B. The computation of the
running of the coupling up to large energies thus provides a quantitative test of the theory,
which is believed to be fundamental in the hadronic as well as in the high energy regime.
Furthermore, it is interesting to find out at which energies the perturbative behaviour of a
given coupling sets in.
The basic strategy for such a computation has been proposed by Lscher, Weisz and
Wolff [2]. They use a non-perturbative definition of the coupling, which runs with the
spacetime volume. Its evolution is mapped out by a recursive finite size scaling technique
up to large energies where contact with the minimal subtraction scheme is made by
E-mail address: uwolff@physik.hu-berlin.de (U. Wolff).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 6 3 - 7
ALPHA Collaboration / Nuclear Physics B 612 (2001) 324
perturbation theory. The central object in this computation is the step scaling function,
which can be understood as a beta function for finite scale transformations. At each step in
the recursive evolution of the coupling, the continuum limit is taken. Other key ingredients
include O(a) improvement and Schrdinger functional boundary conditions [35].
This strategy applies to any asymptotically free theory, and it has first been tested for the
nonlinear O(3) model in two dimensions [2], pure SU(2) gauge theory [6], and pure SU(3)
gauge theory [7,8], which can also be interpreted as the quenched approximation of QCD.
In [9], the ALPHA Collaboration has just published their first quantitative results for the
evolution of the coupling in QCD with two flavours.
However, since simulations in full QCD are notorious for high computational cost, the
data at Nf = 2 do not yet reach very close to the continuum limit in the individual steps of
the computation. Therefore, we have decided to also study the approach to the continuum
in a simpler model that is more easily accessible to simulation. It differs from the quenched
approximation by depending on the fermionic determinant. In this computation, the focus
is not so much on the running coupling as a function of the energy scale, but rather on the
details of the approach of the continuum limit in the perturbative regime and at slightly
lower energy.
The model we investigate in this paper may be viewed as arising from an analytic
continuation of the flavour number to negative values and in particular to Nf = 2. Since
a negative power of the fermionic determinant may be represented by bosonic spinor
fields with the same indices as fermionic fields, the name bermions was coined for such
theories [10]. The main virtue of these models is that the interaction term becomes local
and thus numerical simulations are considerably cheaper than in full QCD.
In the literature (e.g., [10,11]), they were mainly considered from an algorithmic point of
view with the idea in mind to extrapolate in Nf from negative values to Nf = 2. However,
this is problematical, since fermionic zero modes may be encountered (for example, at
small quark masses or in large physical volume), which dominate the dynamics in the
theories at negative flavour numbers. Thus, in our work, no extrapolation of results from
negative to positive values of Nf is attempted or aimed at.
In [12], two of the authors have published results for the step scaling function for
unimproved Wilson bermions (Nf = 2) in the perturbative regime. Lattice artefacts
turned out to be very large. As for the quenched approximation and for full QCD with two
flavours, we now study the O(a) improved Nf = 2 theory. The inclusion of the clover
term into the bermionic action poses certain algorithmic problems that are dealt with in
this article. We present a detailed study of the performance of the algorithm used for our
Monte Carlo simulations.
An important input for the understanding of our Monte Carlo data comes from lattice
perturbation theory. The cutoff effects of the step scaling function can be computed
perturbatively. They are used in the data analysis. The cutoff effects have already been
estimated in [13] to 2-loop order. However, in that calculation, the continuum value of the
critical mass has been used as a first estimate instead of the finite lattice value, which
was not yet available (see discussion in [13]). The computation of this critical mass
is technically more involved due to extra tadpole diagrams that emerge from the non-
vanishing background field. Here we present a computation that includes all diagrams and
completes the study of [13].
This article is organized in the following way. In the next section, we reflect the most
important definitions that occurred in our previous articles. In Section 3, we discuss the
perturbative expansion of the lattice artefacts of the step scaling function. After that, the
bermion model and the algorithm used in our non-perturbative calculations is discussed. In
the last section, our numerical results are summarized.
2. Lattice theory
Since this work extends our earlier work reported in [3,7,12], we only briefly summarize
the necessary notations. For unexplained conventions, we refer in particular to [14,15].
The theory is set up on a four-dimensional hyper-cubic Euclidean lattice with lattice
spacing a and size T L3 , T = L, L being an integer multiple of a. The gauge field
on this lattice is represented by an SU(3) matrix U (x, ) that is defined on every link
between nearest neighbour sites x and x + a of the lattice (
denotes the unit vector in
the direction = 0, 1, 2, 3). Furthermore, on the lattice sites reside Nf flavours of mass
degenerate fermionic quark fields (x), which also carry Dirac and colour indices. We
do not specify Nf at the moment. Later, we want to consider the theory in which Nf is
continued to negative numbers. This has to be done after the integration over the quark
fields has been performed.
The spatial sub-lattices at fixed times x0 are thought to be wrapped on a torus. The gauge
field thus fulfils periodic boundary conditions in the space directions while the quark fields
obey periodic boundary conditions in these directions up to a phase ei [16]. In the time
direction, we impose Dirichlet boundary conditions. The gauge field at the boundary takes
the form
U (x, k)|x0=0 = exp(aC),
U (x, k)|x0 =T = exp(aC ).
(1)
The constant diagonal fields C and C can be chosen such that a constant colour electric
background field is enforced on the system [3]. The boundary conditions for the quark
fields are discussed in detail in [15]. The boundary quark fields serve as sources for
fermionic correlation functions. They are set to zero after differentiation.
The Schrdinger functional is the partition function of the system,
S[U, ,]
.
Z = e = D[U ]D[]D[]e
(2)
It involves an integration over the fields with fixed boundary values at x0 = 0 and x0 = T .
For the action, we take the sum
] = Sg [U ] + Sf [U, ,
]
S[U, ,
of the O(a) improved plaquette action
(3)
Sg [U ] =

1
w(p) tr 1 U (p)
2
g0 p
and the fermionic action

] =
Sf [U, ,
(x)(D
+ m0 )(x).
(4)
(5)
Here, D is the O(a) improved WilsonDirac operator including the Sheikholeslami

Wohlert term [17] multiplied with the improvement coefficient csw and a boundary
improvement term that goes with ct . Details can be found in [14,15]. As discussed, for
example, in [3], the leading cutoff effects from the gauge action can be cancelled by
adjusting the weights w(p) of the plaquettes at the boundary, i.e., one sets
w(p) = ct (g0 )
(6)
if p is a time-like plaquette attached to a boundary plane. In all other cases, w(p) = 1.

The improvement coefficient csw has been computed to 1-loop order of perturbation
theory with the result [15,18]
csw (g0 ) = 1 + 0.26590(7)g02,
(7)
independent of Nf to this order. For Nf = 0 and Nf = 2, csw has also been computed nonperturbatively [5,19]. The results of these simulations can be represented in the region
0 g0 1 in good approximation by the rational functions

1 0.656g02 0.152g04 0.054g06
,
csw (g0 )N =0 =
f
1 0.922g02

1 0.454g02 0.175g04 + 0.012g06 + 0.045g08
csw (g0 )N =2 =
.
f
1 0.720g02
(8)
The boundary improvement coefficients are only known perturbatively. The 2-loop value
for ct depends (in principle) quadratically on Nf and has the form

ct (g0 ) = 1 + 0.08900(5) + 0.0191410(1)Nf g02

+ 0.0294(3) + 0.002(1)Nf + 0.0000(1)Nf2 g04 .
(9)
ct is known to 1-loop order,
ct (g0 ) = 1 0.0180(1)g02.
(10)
From the Schrdinger functional, a running coupling may be defined by differentiating

with respect to the boundary fields. To obtain a complete definition, the diagonal matrices
C and C and the direction of the differentiation must be specified. Here we differentiate
along a curve parametrized by the dimensionless parameter at the boundary field A of
Ref. [7], which is favoured by practical considerations such as mild cutoff effects. With
this choice, the induced constant colour electric background field can be represented by
V (x, ) = eaB (x),
(11)
with

Bk = x0 C + (T x0 )C /T .
B0 = 0,
(12)
Now, since = log Z/ is a renormalized quantity [7] with the perturbative

expansion = g02 0 + 1 + , a renormalized coupling with the correct normalization
is defined as

0
2
g (L) = .
(13)
=0
This coupling can be computed efficiently in numerical simulations as the expectation
value / = S/.
For Nf = 0, the coupling depends not only on the scale L but also on the mass m1
of the quarks, which we define via the PCAC relation [20]. To this end, the fermionic
boundary fields of the Schrdinger functional are used to transform this operator relation
to an identity that holds up to O(a 2 ) between improved fermionic correlation functions on
the lattice. In Section 3, this will be explained in more detail.
To define the step scaling function (u), we set u = g 2 (L) and tune m1 (L/a) = 0. Then
we change the length scale by a factor 2 and compute the new coupling u = g 2 (2L). The
lattice step scaling function at resolution L/a is defined as

(u, a/L) = g 2 (2L)u=g 2 (L),m (L/a)=0.
(14)
1
g 2
and m1 fix the bare parameters of the theory. The continuum limit
These conditions on
(u) can be found by an extrapolation in a/L. We expect that in the O(a) improved theory
(u, a/L) converges to (u) with a rate roughly proportional (i.e., up to logarithms and
higher orders) to (a/L)2 .
3. Perturbative computation of the cutoff effects

The size of the cutoff effects in the step scaling function can be estimated in perturbation
theory. To this end, the relative deviation of the step scaling function from its continuum
limit is expanded in powers of u,
(u, a/L) (u)
(u)

= [10 + 11 Nf ]u + 20 + 21 Nf + 22 Nf2 u2 + O u3 .
(u, a/L) =
(15)
It turns out to be quite small at 2-loop level. However, it is still necessary to extrapolate
the Monte Carlo data to the continuum limit by simulating a sequence of lattice pairs with
decreasing lattice spacing and fixed coupling u. One may use the perturbative expansion of
(u, a/L) to remove the O(a) cutoff effects up to 2-loop order from the non-perturbative
values of the step scaling function (u, a/L).
The 1-loop coefficients 1j (a/L), first listed in [21], are shown in Table 1. The 2-loop
coefficients 2j (a/L) have been estimated in [13]. However, the parts of 2j involving
Table 1
Perturbative results for (u, a/L) up to 2-loop order
L/a
10
11
20
21
22
4
5
6
7
8
9
10
11
12
0.01033
0.00625
0.00394
0.00268
0.00194
0.00148
0.00117
0.00095
0.00079
0.00002
0.00013
0.00014
0.00014
0.00011
0.00009
0.00007
0.00006
0.00005
0.00159
0.00087
0.00055
0.00038
0.00027
0.00020
0.00015
0.00011
0.00009
0.00069
0.00048
0.00033
0.00021
0.00013
0.00010
0.00007
0.00006
0.00005
0.000724
0.000411
0.000199
0.000102
0.000058
0.000038
0.000026
0.000020
0.000016
contributions from the quarks contain the 1-loop coefficient of the critical bare quark mass
mc at which the renormalized quark mass vanishes. This zero mass condition has to be
specified with the cutoff in place. Thus, in the expansion
4
(1) 2
mc = m(0)
(16)
c + mc g0 + O g0 ,
we get a/L dependent expansion coefficients m(i)
c , which only in the limit a/L 0 go
(0)
(1)
over to their continuum values, which is mc = 0 at tree level, while the 1-loop value mc
can be found in Table 1 of [13]. In [13], these continuum values were used to compute the
2-loop coefficients 2j . So, as the authors state, the results presented there can only give a
first idea of the size of the cutoff effects. To obtain the correct values, we need to compute
(1)
mc at finite a/L.
For the quark mass, we adopt the definition of [14] based on the PCAC relation. We first
introduce the bare correlation functions

1
a9 1 a
A0 (x) (y)5 a (z) ,
fA (x0 ) = 3
(17)
L x,y,z 3
2

1 a
a9 1 a
fP (x0 ) = 3
(18)
P (x) (y)5 (z) ,
L x,y,z 3
2
where Aa and P a denote the axial current and density and (x) is the functional derivative
with respect to the boundary quark fields at x0 = 0. Now we can define the x0 dependent
current quark mass
m(x0 ) =
1
2 (0
+ 0 )fA (x0 ) + cA a0 0 fP (x0 )

,
2fP (x0 )
(19)
where 0 and 0 are the naive forward and backward derivatives on the lattice. As an
unrenormalized quark mass, we will use m in the middle of the lattice, i.e.,

for even T /a,
m T ,
m1 = 1 2 T a
(20)
T +a
+m 2
, for odd T /a.
2 m 2
The O(a) correction of the axial current is proportional to the improvement coefficient cA
which is
cA (g0 ) = 0.00756(1)g02
(21)
to 1-loop order in perturbation theory [15]. Here, we are interested in the critical bare quark
mass mc at which the renormalized quark mass is zero. Since m1 is only renormalized
multiplicatively, it is sufficient to require m1 = 0 in order to make the renormalized quark
mass vanish.
The current quark mass may be expanded in powers of g02 ,

(0)
(1)
m1 = m1 (m0 ) + m1 (m0 )g02 + O g04 ,
(22)
where the expansion coefficients depend on the bare quark mass m0 . To set up perturbation
theory, we consider m0 also as a series,

(0)
(1)
m0 = m0 + m0 g02 + O g04 ,
(23)
and expand m1 further as
(0) (0)
m1 = m1 m0 +

(1) (0)
(1)
(0) (0) 2
m1 m0 + m0
g0 + O g04 .
m m0
m0 1
(1)
(24)
(0)
Therefore, the computation of mc has to be done in two steps. First we compute mc by

requiring
(0)
m(0)
(25)
= 0,
1 mc
then we can determine m(1)
c from
(1)

(0)
(26)
m(0)
+ m(1)
= 0.
m1 m(0)
c
c
c
m0
The first step is easily done numerically, the results are shown in Table 2. The second step
mainly amounts to expanding fA and fP up to order g02 and requires a slightly bigger
effort.
m1
Table 2
Perturbative results for the critical quark mass mc up to 1-loop order
(0)
(1)
(1)
L/a
mc
mc0
mc1
4
5
6
7
8
9
10
11
12
0.0015131
0.0016969
0.0006384
0.0005761
0.0003209
0.0002753
0.0001835
0.0001561
0.0001145
0.26667
0.26782
0.26984
0.26995
0.27004
0.27005
0.27006
0.27006
0.27007
0.0027136
0.0006933
0.0001990
0.0000852
0.0000451
0.0000026
0.0000017
0.0000011
0.0000008
10
The expansion of fA and fP is outlined in [15]. After integrating over the quark fields
and inserting the contractions of the quark and boundary fields, one obtains
fA,P (x0 ) = ct2
a9 1
0)S(z, x)
tr P+ P U (z a 0,
L3 x,y,z 2

0)1
S(x, y)U (y a 0,
G y
0 =z0 =a
(27)
where = 0 5 for fA and = 5 for fP , and the trace is to be taken over the Dirac and
colour indices only. P+ and P are the projectors
1
P = (1 0 ),
(28)
2
and . . .G denotes the gauge field average with a probability density proportional to
det(D + m0 ) exp{SG [U ]}.
(29)
The correlation functions fA and fP in (27) contain two quantities that have to be expanded
in powers of g0 . One is the quark propagator

S(x, y) = S (0) (x, y) + S (1) (x, y)g0 + S (2) (x, y)g02 + O g03 ,
(30)
the other one is the gauge field U , which is expanded around the background field V
U (x, ) = V (x, ) exp{g0 aq (x)}

= V (x, ) 1 + g0 aq (x) + g02 a 2 q (x)2 + O g03 .
(31)
At 1-loop order, fA and fP now get several contributions:

0)
1. contributions from the first and second order terms of the link variables U (z a 0,
0)1 , resulting in diagrams 1a, 1b and 2 of Fig. 1,
and U (y a 0,
2. contributions from contractions between the first order terms of the link variables and
the first order terms of the quark propagators, resulting in diagrams 3a, 3b, 4a and 4b
of Fig. 1,
3. contributions from the second order terms of the quark propagators, resulting in
diagrams 5a, 5b, 6a and 6b of Fig. 1, and
4. one contribution from the contraction of the first order terms of both quark
propagators, resulting in diagram 7 of Fig. 1.
As already mentioned in Section 2, we need a non-vanishing background field in order
to define the running coupling. Due to the presence of this field, we obtain four more
contributions. Taking the average . . .G leads not only to the diagrams in Fig. 1, but also
to contractions of the first order term of the link variables and quark propagators with the
first order term of the total action, including the gauge fixing and ghost terms needed for
perturbation theory (see [3,7]). With zero background field, this first order term vanishes,
but here it has to be taken into account. These contributions result in the diagrams of
Fig. 2. Note that only the diagrams containing closed fermion loops depend on the number
of flavours Nf . So the only Nf dependent contributions to fA and fP at 1-loop order come
from diagrams 8a and 8b, whereas diagrams 9a and 9b are of opposite sign and thus cancel
11
Fig. 1. Diagrams contributing to fA (x0 ) and fP (x0 ) at one-loop order of perturbation theory. The
dotted lines denote the links between x0 = 0 and x0 = a.
(1)
in the sum. Thus mc becomes Nf dependent,

(1)
(1)
m(1)
c = mc0 + mc1 Nf .
(32)
In contrast to the case of vanishing background field, the propagators are not known
analytically, so they have to be computed numerically. Here, we have used the method
described in [22]. Due to these numerical computations, computer time is not negligible.
For example, on a 200 MHz Pentium PC, the computation of fA and fP at L/a = 16 at
12
Fig. 2. Tadpole diagrams contributing to fA (x0 ) and fP (x0 ) at one-loop order of perturbation theory
with non-vanishing background field. The dashed line represents the ghost propagator.
1-loop level took us about 16 hours of CPU time. As one has to sum over three momentum
components and two vertex times, the time needed scales asymptotically with (L/a)5 .
The 1-loop correlation functions fA(1) and fP(1) are found by summing all the diagrams.
We are now able to compute m(1)
c and thus get the deviation (u, a/L) up to 2-loop order.
The results are shown in Tables 1 and 2. As stated before, the 2-loop coefficients 2j are
found to be small.
4. Monte Carlo simulations with bermions

4.1. The bermion model
In the following, we will especially be interested in the theory in which the number of
flavours has been continued to Nf = 2 [1012]. This has to be done after the integration
over the fermion fields has been performed. At Nf = 2 the partition function

N /2
Z = D[U ]eSg det D D f
(33)
can be written as

Z = D[U ]D[ + ]D[]eSg Sb
with a local bosonic action
(34)
Sb [U, ] = a 4

(D)(x)2 .
13
(35)
Note that, in the actual numerical simulation, we have used the hopping parameter
representation M of the Dirac operator, which is related to D by
M = 2(D + m0 ),
= (8 + 2am0)1 .
(36)
In a similar way, we can express expectation values of fermionic observables at Nf = 2

after integration over the quark fields by expectation values in the bermion theory. As
an a priori guess consistent with 2-loop perturbation theory, we chose the improvement
coefficient csw by linearly extrapolating the non-perturbative results at Nf = 2 and Nf = 0,
csw (g0 )|Nf =2 = 2csw (g0 )|Nf =0 csw (g0 )|Nf =2 .
(37)
This choice guarantees that the observables are smooth functions of the bare coupling and
an extrapolation to the continuum limit is feasible. Furthermore, we have also computed
the value for csw for the most critical parameters used in this work along the lines of [23]
and found good agreement with (37), see the next subsection. For the other improvement
coefficients, we take the perturbative results with their explicit respective Nf dependence.
In the bermion model, the occurrence of Dirac operator zero modes is dynamically
enhanced. Thus, in situations in which zero modes (or exceptional configurations) are to be
expected, such as large physical volumes or large values for the bare coupling, these may
render the functional integral ill-defined or at least hamper its Monte Carlo evaluation.
However, we will study this theory not too far from the perturbative regime so that these
problems do not occur. Already in the quenched approximation, exceptional configurations
occur and invalidate measured fermionic correlation functions. To reach larger couplings,
one could also here consider a twisted mass term [24,25], which we shall however not
pursue in this paper.
4.2. The size of csw
In perturbation theory, csw is linear in Nf up to two loops, which motivated us to first do
a linear extrapolation of existing non-perturbative data. To corroborate this further, we have
computed csw also non-perturbatively at Nf = 2 for the bare coupling = 8.99, which
was used in our simulations, compare Table 5. The computation was done along the lines
of Ref. [23], to which we refer for unexplained notation and an explanation of the method.
We have computed the current mass aM and the lattice artefact a9M for various values
of at three trial values of csw . As seen before in [5] and [19], we note that 9M depends
only weakly on the mass M. Therefore, we are satisfied with a current mass M that roughly
vanishes, thereby introducing a negligible error. Our results for the lattice artefact are
summarized in Table 3.
A linear interpolation of these three points to the improvement condition a9M =
0.000277 yields csw = 1.285(7), which is to be compared with the value csw = 1.271815
used in the simulation. The effect of this difference in csw on the step scaling function
can be estimated at 1-loop order of perturbation theory. For all lattice sizes, the change in
14
Table 3
a9M at three trial values of csw at = 8.99
csw
aM
a9M
1.171815
1.271815
1.371815
0.0095(1)
0.0003(2)
0.0006(2)
0.00218(19)
0.00066(24)
0.00127(21)
(u, a/L) is smaller than 4 104 , which is negligible compared to the statistical errors,
compare Table 6.
4.3. Simulation algorithm
Unimproved Wilson bermions can be simulated with a hybrid overrelaxation algorithm,
in which the gauge fields are generated by a combination of heatbath and overrelaxation
steps and the bermion fields are generated by overrelaxation steps only [12]. Since the
improved bosonic action depends quadratically on an individual gauge link U (x, )
through the clover term, we found it practically impossible to generalize finite step size
algorithms to simulations of improved bermions. A local (link by link) hybrid Monte
Carlo algorithm would be feasible and experience from pure gauge theory shows that it
is worthwhile to consider it [26]. However, for the same reason as above, a part of the force
would have to be recomputed at each step on the trajectory. Thus we expect that such an
algorithm would be very expensive. Also a global hybrid Monte Carlo algorithm would be
expensive, of course.
Therefore, we decided to perform global heatbath steps for the bosonic field and local
overrelaxation steps with respect to the unimproved action for the gauge fields. The clover
term is then taken into account in an acceptance step in which the local action difference
with respect to the full improved action is used. The acceptance rate in this step turns
out to be large enough to pursue this algorithm. As the overrelaxation step can be set up
symmetrically, the combined update fulfils detailed balance. Together with the heatbath
step for the bosonic field, our algorithm also satisfies ergodicity.
In principle, boson fields with the correct distribution can be generated by drawing
a random field from a Gaussian distribution and then applying = M 1 . However,
this procedure is expensive because it requires to run a solver with full accuracy for each
update. A method published in [27], which is based on an approximate inversion followed
by an additional acceptance step, allows to reduce the cost of this step.
For the update of the gauge fields, we perform a sweep over the lattice and update the
links sequentially. By an overrelaxation step with respect to the unimproved action we
propose a new configuration U which differs from the old configuration only for the link
variable U (x, ) U (x, ). Thus, for the acceptance step, only the part of the action that
depends on this link variable is needed. For the gauge part of the action, the difference
Sg [U ] Sg [U ] =

Re Tr U (x, ) U (x, ) S (x, )

3
(38)
15
can easily be obtained. Here, S (x, ) is the sum of the staples at (x, ) and = 6/g02 .
The bermion contribution can also be computed as a local difference. At the beginning
of a sweep through the lattice, the Dirac operator is applied on the whole lattice, and the
auxiliary field = M[U ] is stored. In each local step, = M[U ] is computed from
by modifying it at those lattice sites that depend on the link variable U (x, ) either
through the hopping or through the clover term. In order to simplify our notation, we split
M = M1 + M2 into a term M1 which is diagonal in coordinate space and a term M2 which
contains nearest-neighbour contributions. Then we need to compute

9(1)
x (z) = M1 [U ](z) M1 [U ](z),

9(2)
x (z) = M2 [U ](z) M2 [U ](z).
Two lattice sites are affected by the hopping term,

9(2)
x (x) = (1 ) U (x, ) U (x, ) (x + a ),

= (1 + ) U (x, ) U (x, ) (x).
9(2)
x (x + a )
(39)
(40)
Here 0 = 1, k = ei . Fourteen lattice sites are affected by the clover term, namely x,
x + and for all directions = x and x + . At x we get, for example,

i
9(1)
x (x) = csw U (x, ) U (x, )
8
U (x + a ,
)U (x + a , )U (x, ),
(41)
=
and similar terms at the other points. Since the update is local, one has to be careful in
parallelizing the algorithm. In a simulation of a lattice on our SIMD machine with only
two lattice points per node in any direction, neighbouring nodes would modify the field
at a given point simultaneously through the clover term. To avoid this conflict, the local
lattice size per node in each direction has to be larger or equal to three.
4.4. Performance
As a measure for the efficiency of our algorithm, we use the machine dependent quantity
Mcost focusing on the Schrdinger functional coupling g 2 . It is defined as
2

Mcost = (update time in seconds on machine M) error of 1/g 2 (4a/T )(4a/L)3.
(42)
In our case, Mcost refers to the CPU time spent on an 8-node machine with APE100 architecture. This performance measure allows us to compare for example with performance
data obtained in [28] for full QCD.
A further important indicator of the efficiency of the algorithm explained above is the
acceptance rate of the clover term in the gauge field update. This acceptance rate turns out
to depend only weakly on the parameters in the range of couplings considered here. At
g 2 = 0.9793, it is about 76%, while at g 2 = 1.5145, the acceptance is roughly 70%.
16
The cost of our simulations can in principle be optimized by tuning the precision of the
solver in the boson field update. We could however only obtain a total advantage compared
to a full precision solver of roughly 10% on the small lattices, with a rather flat minimum.
This can mainly be attributed to the fact that the time of an update step is dominated by
the gauge field update. Since the optimal size of the residue has to be scaled down when
increasing the lattice size [27], this advantage gets even smaller on larger lattices. Hence,
we expect for our application that tuning the precision parameter on large lattices is more
expensive than running with an ad hoc guess.
The cost at g 2 = 0.9793 for various lattice sizes for the improved and the unimproved
bermion theory is shown in Table 4. Obviously, improvement of bermions (with the
algorithms explained above) leads to a substantial increase in computer time, which we
estimate to be a factor 12. In Section 5, it will become clear however, that it is still
profitable. The data of Table 4 are also shown in Fig. 3. A linear fit in this plot (that
Table 4
Costs for improved bermions in comparison with Wilson bermions at
u = 0.9793. Note that the last two entries for improved bermions are at
u 1.11
L/a
Mcost improved
Mcost unimproved
4
5
6
8
10
12
0.061(2)
0.107(3)
0.212(6)
0.457(11)
0.790(17)
1.30(3)
0.00535(7)
0.00866(13)
0.0155(2)
0.0319(4)
0.0788(12)
Fig. 3. Costs for the measurement of the coupling u = 0.9793 for Wilson bermions and improved
bermions.
17
excludes the smallest lattices of the improved theory) shows that in the improved as well
as in the unimproved theory, Mcost scales as a 2.5 .
Comparing with data from [28], it turns out that improved bermions in our implementation are roughly a factor 10 cheaper than simulations in full QCD with two flavours.
Furthermore, the scaling with a seems to be slightly better for the bermions. On the other
hand, we have estimated the additional cost of unimproved bermions in comparison to pure
gauge theory to be only a factor 3.
5. Technical details and results

5.1. Parameters
We have computed the step scaling function (u, a/L) for the two values of the
coupling u = 0.9793 and u = 1.5145 at lattice sizes L/a = 4, 5, 6, 8. To this end, the
bare parameters and have to be tuned such that these couplings are reached while the
current masses m1 (L/a) vanish. The precision required in the tuning of can be estimated
in perturbation theory [21]. In order to estimate the effect of a slight mismatch in the tuning
of m1 , one defines the derivative of with respect to z = Lm1 ,

2
g (2L)g 2 (L)=u,m (L)=z/L z=0 = 1 (a/L)u2 + .
1
z
(43)
Under the assumption that it suffices to approximate 1 by its universal part (valid for
L/a ), we obtain

Nf

c1,1 (z)
1 (0) =
(44)
= 0.00957Nf.
4 z
z=0
Here, c1,1 (z) as defined and computed in [16] has been used. This means for example that
a tuning of the current mass to zero up to 0.001 on an L/a = 8 lattice leads to an error
in the step scaling function smaller than 0.0002u2 and even less on the smaller lattices.
Table 5 summarizes the results of the tuning procedure. These results allow to neglect the
Table 5
Parameters and results for the coupling and the mass at L
L/a
g 2 (L)
m1 (L/a)
10.3488
10.5617
10.7302
11.0026
0.131024
0.130797
0.130686
0.130489
4
5
6
8
0.9793(19)
0.9795(21)
0.9793(11)
0.9793(14)
0.00000(31)
0.00055(13)
0.00000(5)
0.00000(6)
0.132959
0.132637
0.132433
0.13209
4
5
6
8
1.5145(23)
1.5145(17)
1.5145(33)
1.5145(33)
0.00000(28)
0.00000(7)
0.00000(4)
0.00066(8)
8.3378
8.5453
8.70830
8.99
18
tuning error for the current mass while the error of g 2 (L) is propagated into the step scaling
function by perturbation theory. In those cases where m1 = 0 is displayed in the table, this
is a result of an interpolation in the best tuning runs, which is justified by very small values
of 2 obtained in the interpolation.
5.2. The numerical simulation
Most of our numerical simulations were performed on APE100/Quadrics parallel
computers with SIMD architecture and single precision arithmetic. We have used machines
with up to 256 nodes with an approximate peak performance of 50 MFlops per node.
Roughly half of the statistics for the simulation at L/a = 16 and u = 1.5145 has been
accumulated on one crate (128 nodes) of an APEmille installation in Zeuthen. Since our
program was not yet really optimized for APEmille, the advantage is only a factor 3. In our
simulations, we have made much use of trivial (replica) parallelization.
The coupling and other inexpensive observables have been measured after each update,
which corresponds to a bosonic heatbath step followed by an overrelaxation step for the
gauge fields. The fermionic correlation functions to obtain the current mass m1 have been
measured only rarely, e.g., every 100th update sweep, because the mass does not fluctuate
much. We have done up to 16 31500 full updates and measurements of the coupling.
The statistical errors of the observables have been determined by a direct computation of
the autocorrelation matrix along the lines of Appendix A of [28]. Typical autocorrelation
times for the coupling range from 3 to 10 (in units of updates).
5.3. Discussion of results
For the non-perturbative computation of the step scaling function, we have simulated
pairs of lattices with size L and 2L at the same bare parameters in the bermion theory. The
results of these computations are listed in Table 6. For the propagation of the statistical
error and the mismatch of the tuning results for the coupling, we use a perturbative ansatz.
Then we obtain the lattice step scaling function (u, a/L) that is shown in Fig. 4. We pass
Table 6
Results for the coupling and the mass at 2L at the parameters defined by the given value of g 2 (L)
L/a
g 2 (L)
g 2 (2L)
m1 (2L/a)
4
5
6
8
0.9793(19)
0.9795(21)
0.9793(11)
0.9793(14)
1.1090(28)
1.1079(29)
1.1053(30)
1.1093(40)
0.00300(10)
0.00086(5)
0.00094(4)
0.00025(3)
4
5
6
8
1.5145(23)
1.5145(17)
1.5145(33)
1.5145(33)
1.8734(74)
1.8648(82)
1.8488(86)
1.869(14)
0.00266(12)
0.00094(7)
0.00070(5)
0.00002(5)
to the continuum limit by an extrapolation in a/L with the ansatz

(u, a/L) = (u) 1 + (u)(a/L)2 .
19
(45)
As shown in Fig. 4, this ansatz works perfectly, i.e., within the error bars no linear
dependence of the step scaling function on a/L can be detected. The results for the
continuum step scaling function and the corresponding 2- and 3-loop values are given
in Table 7. The difference between 2- and 3-loop perturbation theory is thus of the same
Fig. 4. Step scaling function for improved bermions for the couplings u = 0.9793 and u = 1.5145
with fits linear in (a/L)2 . Shown is also the extrapolated continuum value and the 2- and 3-loop
results.
Table 7
Extrapolated simulation results and perturbation theory for the step scaling function at Nf = 2
u
(u)
(u)|2-loop
(u)|3-loop
0.9793
1.5145
1.1063(46)
1.871(17)
1.10435
1.85122
1.10691
1.87026
20
Fig. 5. Results for the step scaling function at u = 0.9793, together with a quadratic fit under the
constraint of universality.
size as the error of the extrapolated simulation results. Within the error bars, both values of
the step scaling function (u) are consistent with perturbation theory.
The typical size of the O(a) effects can be estimated by considering the step scaling
function in the unimproved bermion theory as well. This has been done in [12] for
u = 0.9793. Fig. 5 shows these results together with the data after implementing
improvement. Obviously, the linear cutoff effects are quite seizable for this observable,
of the order of a few percent. The data are fitted with a combined fit under the constraint
that their continuum limit agrees, that means assuming universality. This fit is linear plus
quadratic in a/L for the Wilson bermion data and quadratic in a/L in the improved
data. Although the additional input from the Wilson data is included, the joint continuum
limit combined (0.9793) = 1.1059(43) agrees almost completely with the value in Table 7.
A linear plus quadratic fit in a/L of the unimproved data alone would have given the
continuum result unimproved(0.9793) = 1.103(12).
The uncertainties of these extrapolated values show a remarkable success of improvement. Although the total computer time for the improved simulations was only by a factor
1.7 higher than for the Wilson bermions, the error after extrapolation is by a factor 2.7
smaller. Since a large portion of the computational cost for (u) comes from the largest
lattice, this advantage can be attributed to the lattice size needed for a reasonable extrapolation. For Wilson bermions, this was L/a = 24, whereas simulations are limited
to L/a = 16 for the improved case. Of course, our observation is restricted to one value of
the coupling; at different values, lattice artefacts may behave differently. It should also be
noted that in the bermion case, adding the clover term leads to a significant performance
penalty for the numerical simulation. For typical algorithms for the simulation of dynam-
21
ical fermions, like variants of the hybrid Monte Carlo, the inclusion of the clover term
implies a much smaller overhead. Hence, the advantage of improvement should be even
bigger there.
In addition to our determination of the step scaling function by extrapolating (u, L/a)
from Monte Carlo simulations, we have also analysed the approach to the continuum limit
for data which have the 2-loop perturbative lattice artefacts cancelled. To this end, we
replace the lattice step scaling used for the fit by the corrected values
(2) (u, a/L) =
(u, a/L)
1 + 1 (a/L)u + 2 (a/L)u2
(46)
with from (15). The results for these fits are shown in Fig. 6, together with the uncorrected
fits shown before in Fig. 4. Again, we leave out the point at L/a = 4 for the fits. As can
be seen from the plot, this procedure does not visibly reduce remaining lattice artefacts.
For the coupling u = 1.5145, the slope of the fitted line gets slightly smaller, whereas for
Fig. 6. Step scaling function for improved bermions for the couplings u = 0.9793 and u = 1.5145
with fits linear in (a/L)2 . The rectangles represent the data points obtained from our simulations,
whereas the circles represent the data (2) corrected by perturbation theory.
22
Fig. 7. Check of lattice artefacts in the mass at u = 0.9793 and u = 1.5145.
u = 0.9793 it remains roughly equal. However, if we also include the point at L/a = 4, the
lattice artefacts have even larger O(a)2 effects than for the uncorrected data. Nevertheless,
their continuum limit agrees within the error bars with the one obtained by the procedure
used before. It is also consistent with perturbation theory in the continuum limit.
Since the mass m1 is tuned to zero on the small lattices, we expect that L(m1 (2L)
m1 (L)) is a pure lattice artefact that vanishes in the continuum limit with a rate proportional
to (a/L)2 . This expectation is confirmed by Fig. 7 in which this mass difference is shown
as a function of (a/L)2 . While the scaling is perfect for the smaller of the two couplings,
there are small deviations at u = 1.5145, which however can be attributed to statistical
fluctuations.
6. Conclusions
In this paper, we have investigated the step scaling function in the O(a) improved
bermion model by means of extrapolating Monte Carlo data at different lattice resolutions
23
to the continuum limit. The results obtained were compared with renormalized perturbation
theory in the continuum limit and with data obtained from simulation of unimproved
bermions. They also serve as a guide in planning analogous simulations with dynamical
quarks.
It is demonstrated that the implementation of improvement successfully reduces lattice
artefacts and allows a fit of the step scaling function (u, a/L) linearly in (a/L)2 (see
Fig. 5). This raises our confidence that the same extrapolation procedure can be applied
for dynamical fermions. We note however, that to our disappointment, the computer
time required for our algorithm turned out to be only about a factor 10 smaller than for
dynamical fermions. This means that the lattice sizes that can be reached are not much
larger than for fermions, in contrast to the situation without improvement.
Acknowledgements
We would like to thank Peter Weisz for essential checks on our perturbative calculations
and Rainer Sommer for helpful discussions. DESY provided us with the necessary
computing resources and the APE group contributed their permanent assistance. This work
is supported by the Deutsche Forschungsgemeinschaft under Graduiertenkolleg GK 271
and by the European Communitys Human Potential Programme under contract HPRNCT-2000-00145.
References
[1] P. Weisz, Nucl. Phys. Proc. Suppl. 47 (1996) 7183, hep-lat/9511017.
[2] M. Lscher, P. Weisz, U. Wolff, Nucl. Phys. B 359 (1991) 221243.
[3] M. Lscher, R. Narayanan, P. Weisz, U. Wolff, Nucl. Phys. B 384 (1992) 168228, hep-lat/
9207009.
[4] S. Sint, Nucl. Phys. B 421 (1994) 135158, hep-lat/9312079.
[5] M. Lscher, S. Sint, R. Sommer, P. Weisz, U. Wolff, Nucl. Phys. B 491 (1997) 323343, hep-lat/
9609035.
[6] G. de Divitiis, R. Frezzotti, M. Guagnelli, M. Lscher, R. Sommer, P. Weisz, U. Wolff, Nucl.
Phys. B 437 (1995) 447470, hep-lat/9411017.
[7] M. Lscher, R. Sommer, P. Weisz, U. Wolff, Nucl. Phys. B 413 (1994) 481502, hep-lat/
9309005.
[8] J. Garden, J. Heitger, R. Sommer, H. Wittig, Nucl. Phys. B 571 (2000) 237256, hep-lat/
9906013.
[9] A. Bode et al., ALPHA Collaboration, Phys. Lett. B 515 (2001) 49, hep-lat/0105003.
[10] G.M. de Divitiis, R. Frezzotti, M. Guagnelli, M. Masetti, R. Petronzio, Nucl. Phys. B 455 (1995)
274, hep-lat/9507020.
[11] S.J. Anthony, C.H. Llewellyn Smith, J.F. Wheater, Phys. Lett. B 116 (1982) 287.
[12] J. Rolf, U. Wolff, Nucl. Phys. Proc. Suppl. 83 (2000) 899901, hep-lat/9907007.
[13] A. Bode, P. Weisz, U. Wolff, Nucl. Phys. B 576 (2000) 517539, with Erratum, Nucl.
Phys. B 600 (2001) 453, hep-lat/9911018v3 .
[14] M. Lscher, S. Sint, R. Sommer, P. Weisz, Nucl. Phys. B 478 (1996) 365400, hep-lat/9605038.
[15] M. Lscher, P. Weisz, Nucl. Phys. B 479 (1996) 429458, hep-lat/9606016.
24
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
S. Sint, R. Sommer, Nucl. Phys. B 465 (1996) 7198, hep-lat/9508012.

B. Sheikholeslami, R. Wohlert, Nucl. Phys. B 259 (1985) 572.
R. Wohlert, DESY 87/069.
K. Jansen, R. Sommer, Nucl. Phys. Proc. Suppl. 63 (1998) 853855, hep-lat/9709022.
K. Jansen, C. Liu, M. Lscher, H. Simma, S. Sint, R. Sommer, P. Weisz, U. Wolff, Phys.
Lett. B 372 (1996) 275282, hep-lat/9512009.
R. Sommer, The step scaling function with two flavours of massless Wilson quarks, 1998,
unpublished.
R. Narayanan, U. Wolff, Nucl. Phys. B 444 (1995) 425, hep-lat/9502021.
K. Jansen, R. Sommer, Nucl. Phys. B 530 (1998) 185203, hep-lat/9803017.
R. Frezzotti, P.A. Grassi, S. Sint, P. Weisz, Nucl. Phys. Proc. Suppl. 83 (2000) 941946, hep-lat/
9909003.
R. Frezzotti, P.A. Grassi, S. Sint, P. Weisz, hep-lat/0101001.
B. Gehrmann, U. Wolff, Nucl. Phys. Proc. Suppl. 83 (2000) 801803, hep-lat/9908003.
P. de Forcrand, Phys. Rev. E 59 (1999) 36983701, cond-mat/9811025.
R. Frezzotti, M. Hasenbusch, U. Wolff, J. Heitger, K. Jansen, Comput. Phys. Commun. 136
(2001) 113, hep-lat/0009027.

Systematic approach to exclusive B V +, V

decays
M. Beneke, Th. Feldmann, D. Seidel
Institut fr Theoretische Physik E, RWTH Aachen, 52056 Aachen, Germany
Received 8 June 2001; accepted 10 July 2001
Abstract
We show by explicit computation of first-order corrections that the QCD factorization
approach previously applied to hadronic two-body decays and to form factor ratios also allows us
to compute non-factorizable corrections to exclusive, radiative B meson decays in the heavy quark
mass limit. This removes a major part of the theoretical uncertainty in the region of small invariant
mass of the photon. We discuss in particular the decays B K and B K + and complete
the calculation of corrections to the forwardbackward asymmetry zero. The new correction shifts
the asymmetry zero by 30%, but the result confirms our previous conclusion that the asymmetry
zero provides a clean phenomenological determination of the Wilson coefficient C9 . 2001 Elsevier
Science B.V. All rights reserved.
PACS: 13.25.Hw; 12.38.Bx; 12.15.-y
1. Introduction
Radiative B meson decays are interesting because they proceed entirely through loop
effects. The chiral nature of weak interactions implies additional suppression factors thus
enhancing the sensitivity of radiative decays to virtual effects in theories beyond the
standard theory of flavour violation, if these new effects violate chirality. The observation
of b s transitions [1] with a branching ratio of about 104 (in agreement with the
standard theory) has since provided significant constraints on extensions of the standard
model. The semileptonic decay b s+ is expected to occur with a branching fraction
of a few times 106 and will probably be first observed in the near future.
Radiative flavour-changing neutral current transitions can be measured inclusively over
the hadronic final state (containing strangeness in the particular case considered here)
or exclusively by tagging a particular light hadron, typically a kaon. The inclusive
E-mail address: feldmann@physik.rwth-aachen.de (Th. Feldmann).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 6 6 - 2
26
M. Beneke et al. / Nuclear Physics B 612 (2001) 2558
measurement is experimentally more difficult but theoretically simpler to interpret, since

the decay rate is well and systematically approximated by the decay of a free b quark
into light quarks and gluons. In the long run, however, the easier detection of exclusive
transitions requires the development of a systematic theoretical framework for these modes
as well.
The theoretical difficulty with exclusive decay modes is usually phrased as the need to
know the hadronic form factors for the B K () transition, but this is only one aspect
of the problem. Even if the form factors were known with infinite precision, the present
treatment of exclusive, radiative decays would be incomplete, because there exist nonfactorizable strong interaction effects that do not correspond to form factors. They arise
from electromagnetic corrections to the matrix elements of purely hadronic operators in the
weak effective Hamiltonian. We compute these non-factorizable corrections in this paper
and demonstrate that exclusive, radiative decays can be treated in a similarly systematic
manner as their inclusive counterparts. As a result we obtain the branching fractions for
B K and B K + for small invariant mass of the lepton pair to next-toleading logarithmic (NLL) order in renormalization-group-improved perturbation theory.
(In another commonly used terminology that takes into account the 1/s enhancement
of the semileptonic operator O9 the result would be referred to as next-to-next-to-leading
logarithmic. However, in order not to have to distinguish in terminology the photonic and
semileptonic decay, we shall refer to both as NLL when actually NNLL is meant for the
semileptonic decay. Furthermore, since some of the 3-loop anomalous dimensions required
for the semileptonic decay are still not computed, the accuracy is not strictly NNLL.)
Our method draws heavily on methods developed recently for other exclusive B decays.
In [2,3] power counting in the heavy quark mass limit and standard factorization arguments
for hard strong interaction processes have been used to demonstrate that decay amplitudes
for hadronic two-body decays of B mesons can be systematically computed in terms of
form factors, light-cone distribution amplitudes and perturbative hard scattering kernels.
A similar reasoning has been applied by two of us [4] to hadronic form factors for B
decays into light mesons in the kinematic region of large recoil of the light meson. We also
noted that a combination of the work on non-leptonic decays and on form factors could be
used to compute non-factorizable corrections to radiative decay amplitudes. Furthermore,
the ten different form factors that exclusive decays may depend on can be reduced to only
three [5]. More precisely (but still schematically), the amplitude can be represented as

+
a Heff
B = Ca a + B Ta K ,
K
(1)
where a = , refers to a transversely and longitudinally polarized K , respectively.
(An analogous result applies to the decay into a pseudoscalar meson, but we specify our
notation to a vector meson for simplicity.) In this equation a represent universal heavy-tolight form factors [4,5] and light-cone-distribution amplitudes. The factors Ca and Ta
are calculable in renormalization-group improved perturbation theory. Previous treatments
of exclusive radiative decays correspond to evaluating (1) to leading logarithmic accuracy
(up to a weak annihilation contribution as we discuss below). The main result of this paper
is to show that systematic improvement is possible and to extend the accuracy to the next
27
order. In Section 2 we present the result of the calculation. In Section 3 we first discuss
the inputs to our numerical result, such as Wilson coefficients and meson parameters. We
then analyse the stability of the result with respect to the input parameters and a variation
of the renormalization scale. The phenomenological analysis is given in the subsequent
two sections. Section 4 is devoted to B K ; Section 5 to B K + . We focus in
particular on corrections to the forwardbackward asymmetry, which is independent on the
form factors a to first approximation. In the final Section 6 we discuss power corrections
related to photon structure on a qualitative level and summarize our conclusions.
2. Non-factorizable corrections
We generally neglect doubly Cabibbo-suppressed contributions to the decay amplitude.
The weak effective Hamiltonian is then given by
10

GF
Ci ()Oi (),
Heff = Vts Vt b
2
i=1
(2)
and we use the operator basis introduced in [6] for the operators Oi , i = 1, . . . , 6. (More
precisely, due to the normalization of (2), Oi = 4Pi with Pi defined in [6].) We define
b
gem m
s (1 + 5 )bF ,
8 2
b
gs m
si (1 + 5 )TijA bj GA
O8 =
,
8 2
O7 =
O9,10 =
em
()V ,A (s b)VA ,
2
(3)
(4)
2 /4 is the fine-structure constant and m

where em = gem
b () the b quark mass in the MS
scheme. We shall later trade the MS mass for the pole mass mb and the PS mass mb,PS . To
next-to-leading order the pole and MS masses are related by

2
m2b
s CF
m
b () = mb 1 +
(5)
3 ln 2 4 + O s ,
4
with s s (). The sign convention for O7,8 corresponds to a negative C7,8 and
+igs T A , +igem ef for the ordinary quarkgauge-boson vertex (ef = 1 for the lepton
i (for i = 1, . . . , 6),
fields). We will present our result in terms of barred coefficients C
defined as certain linear combinations of the Ci as described in Appendix A. The linear
i coincide at leading logarithmic order with the
combinations are chosen such that the C
Wilson coefficients in the standard basis [7].
As for form factors and non-leptonic two-body decays there exist two distinct classes
of non-factorizable effects. (By non-factorizable we mean all those corrections that are
not contained in the definition of the QCD form factors for heavy-to-light transitions.
For example, the familiar leading-order diagrams shown in (a) and (b) of Fig. 1 are
factorizable.) The first class involves diagrams in which the spectator quark in the B meson
28
|Heff |B
. The circled cross marks the possible insertions of
Fig. 1. Leading contributions to K
the virtual photon line.
participates in the hard scattering. This effect occurs at leading order in an expansion
in the strong coupling constant only through a weak annihilation diagram [Fig. 1(c)].
The relevant diagrams at next-to-leading order are shown as (a) and (b) in Fig. 2 and in
Fig. 3. They contribute at order s0,1 to the functions Ta in (1). Diagrams of this form have
already been considered (for q 2 = 0) in [8]. However, bound state model wave-functions
(rather than light-cone distribution amplitudes) were used and no attempt was made to
systematically expand the hard scattering amplitude in 1/mb . As a consequence, the result
K depends on an infrared cut-off. This difficulty is resolved in the present
of [8] for B
factorization approach. The second class contains all diagrams shown in the second row of
Fig. 2. Here the spectator quark is connected to the hard process represented by the diagram
only through soft interactions. The result is therefore proportional to the form factor a and
the hard-scattering part gives an s -correction to the functions Ca in (1).
In this section we present the results of the calculation of these diagrams. Some of the
results needed for diagrams of the second class can be extracted from work on inclusive
radiative decays [9,10] and we have made use of these results as indicated below. The
conventions for the form factors and light-cone distribution amplitudes for B mesons and
light mesons are those of [4].
2.1. Notation and leading-order result
Since the matrix elements of the semileptonic operators O9,10 can be expressed through
B K form factors, non-factorizable corrections contribute to the decay amplitude only
through the production of a virtual photon, which then decays into the lepton pair. We
therefore introduce

p , Heff
(q, )K
B(p)

igem mb
GF
2T1 q 2 & p p
= Vt s Vt b
2
4
2
2 2

MB m2K q p + p
iT2 q

2

q2

iT3 q q q 2
(6)
p +p
,
MB m2K
where Heff denotes the weak effective Hamiltonian. We define the overall quark mass
factor as the pole mass here. The matrix element decomposition is defined such that the
29
leading order contribution from the electromagnetic dipole operator O7 reads Ti (q 2 ) =

C7 Ti (q 2 ) + , where Ti (q 2 ) denote the tensor form factors. Including also the fourquark operators (but neglecting for the moment annihilation contributions), the leading
logarithmic expressions are [11]

T1 q 2 = C7eff T1 q 2 + Y q 2

q2
V q2 ,
2mb (MB + mK )

q2
A1 q 2 ,
T2 q 2 = C7eff T2 q 2 + Y q 2
2mb (MB mK )

MB mK 2 MB + mK 2
A2 q
A1 q ,
T3 q 2 = C7eff T3 q 2 + Y q 2
2mb
2mb
(7)
(8)
(9)
3 C
5 )/9 (4C
4 C
6 )/3
with C7eff = C7 C3 /3 4C4 /9 20C5/3 80C6 /9 = C7 (4C
and
1 + C
2 + 3C
3 + C
4 + 3C
5 + C
6 )
Y (s) = h(s, mc )(3C
1

1
3 + 3C
3 + C
4 ) + 3C
5 + C
6 h(s, 0)(C
4 )
h(s, mb ) 4(C
2
2
2 2
5 .
4 + 16 C
C3 + 2C
+
9 3
3
(10)
The function

m2q
4
2
h(s, mq ) = ln 2 z
9
arctan
,
z > 1,

z1
4
(2 + z) |z 1|
9
1 + 1 z i
, z1
ln
z
2
(11)
is related to the basic fermion loop. (Here z is defined as 4m2q /s.) Y (s) is given in the NDR
scheme with anticommuting 5 and with respect to the operator basis of [6]. Since C9 is
basis-dependent starting from next-to-leading logarithmic order, the terms not proportional
to h(s, mq ) differ from those given in [7]. The contributions from the four-quark operators
O16 are usually combined with the coefficient C9 into an effective (basis- and schemeindependent) Wilson coefficient C9eff (q 2 ) = C9 + Y (q 2).
The results of this paper are restricted to the kinematic region in which the energy of
the final state meson scales with the heavy quark mass in the heavy quark limit. In practice
we identify this with the region below the charm pair production threshold q 2 < 4m2c
7 GeV2 . The various form factors appearing in (7)(9) are then related by symmetries
[4,5]. Adopting the notation of [4], (7)(9) simplify to

q2
Y q2 ,
T1 q 2 T q 2 = q 2 C7eff 1 +
(12)
2mb MB
2E 2
T q ,
T2 q 2 =
(13)
MB
30

MB 2

MB 2
T2 q T q 2 = q 2 C7eff 2 +
T3 q 2
Y q 3 ,
2E
2mb
(14)
where E = (MB2 q 2 )/(2MB ) refers to the energy of the final state meson and , refer
to the form factors in the heavy quark and high energy limit. The factors i are defined such
that i = 1 + O(s ). The s -corrections have been computed in [4] and will be incorporated
into the next-to-leading order results later on. The appearance of only two independent
structures is a consequence of the chiral weak interactions and helicity conservation, and
hence holds also after including next-to-leading order corrections [4,12]. We therefore
present our results in terms of the invariant amplitudes T, (q 2 ), which refer to the decay
into a transversely and longitudinally polarized vector meson (virtual photon), respectively.
At next-to-leading order we represent these quantities in the form

s CF (1)
Ca
Ta = a Ca(0) +
4
1
d
2 fB fK ,a
B, () du K ,a (u)Ta, (u, ),
+
(15)
a
Nc MB
where CF = 4/3, Nc = 3, 1, m
/E, and Ta, (u, ) is expanded as
s CF (1)
T (u, ).
(16)
4 a,
fK , denotes the usual K decay constant fK . fK , refers to the (scale-dependent)
transverse decay constant defined by the matrix element of the tensor current. The leadingorder coefficient Ca(0) follows by comparison with Eqs. (12) and (14) setting i = 1.
To complete the leading-order result we have to compute the weak annihilation
amplitude of Fig. 1(c), which has no analogue in the inclusive decay and generates the
(0)
(u, ) in (15). To compute this term we perform the projection
hard-scattering term Ta,
of the amplitude on the B meson and K meson distribution amplitude as explained in [4].
The four diagrams in Fig. 1(c) contribute at different powers in the 1/mb expansion. It
turns out that the leading contribution comes from the single diagram with the photon
emitted from the spectator quark in the B meson, because this allows the quark propagator
to be off-shell by an amount of order mb QCD , the off-shellness being of order m2b for
the other three diagrams. With the convention that the K meson momentum is nearly
light-like in the minus light-cone direction, the amplitude for the surviving contribution
depends only on the minus component of the spectator quark momentum. This is in contrast
to the discussion in [4], where, using the same convention for the outgoing meson, the
hard-scattering amplitude depended only on the plus component of the spectator quark
momentum. As a consequence the B meson distribution amplitude B, () now appears
as coefficient of n
/ in the B meson projector. Except for this difference the calculation
follows the rules of [4]. The result reads
(0)
(u, ) +
Ta, (u, ) = Ta,
(0)
(0)
(0)
T,+
(u, ) = T,
(u, ) = T ,+
(u, ) = 0,
(17)
MB
4MB
(0)
4 ).
T , (u, ) = eq
(C3 + 3C
2
MB q i& mb
(18)
31
At leading order weak annihilation occurs only through penguin operators with small
Wilson coefficients. For B decay, there is an additional Cabibbo-suppressed contribution
involving currentcurrent operators, which can be obtained by multiplying the previous
equation with the factor
V C
2
Vus
ub 1 + 3C
,
4

Vt s Vt b C3 + 3C
(19)
whose modulus is about 0.7. Both contributions are numerically small relative to the
dominant leading-order terms. Note that since T (q 2 ) does not contribute to the decay
amplitude in the limit q 2 0, there is no weak annihilation contribution to B K at
leading order in s and at leading order in 1/mb .
We emphasize that our discussion of weak annihilation refers only to the leading
contribution in the heavy quark limit. It is a novel aspect of B K + (with the
K longitudinally polarized) that weak annihilation does not vanish in this limit, if
q 2 mb QCD . On the other hand, the sum of transverse and longitudinal contributions
is dominated by the transverse decay amplitude at small q 2 , and it is probable that the
power-suppressed annihilation contributions to the transverse amplitude T (q 2 ), which we
neglected, are numerically more important. For q 2 = 0 annihilation contributions of this
sort have been studied in [13]. The interplay of the various terms merits consideration for
b d transitions, in which weak annihilation through currentcurrent operators is neither
CKM- nor s -suppressed. Since our main analysis concerns b s decays, we defer this
issue to a later discussion.
2.2. Next-to-leading order spectator scattering
(1)
in (15) contain a factorizable term from expressing
The hard scattering functions Ta,
the full QCD form factors in terms of a , related to the s -correction to the i in
(1)
(f)
(nf)
= Ta,
+ Ta,
. The factorizable correction reads [4]:
Eqs. (12), (14). We write Ta,
2MB
(f)
T,+ (u, ) = C7eff
,
uE

2MB2
q2
(f)
T ,+ (u, ) = C7eff +
Y q2
,
2mb MB
uE
2
(f)
(f)
T,
(u, ) = T ,
(u, ) = 0.
(20)
(21)
(22)
The non-factorizable correction is obtained by computing matrix elements of four-quark

operators and the chromomagnetic dipole operator represented by diagrams (a) and (b)
in Fig. 2. The projection on the meson distribution amplitudes is straightforward. In the
result we keep only the leading term in the heavy quark limit, expanding the amplitude in
powers of the spectator quark momentum whenever this is permitted by power counting. In
practice this means keeping all terms that have one power of the spectator quark momentum
in the denominator. Such terms arise either from the gluon propagator that connects to the
spectator quark line or from the spectator quark propagator, when the photon is emitted
32
|Heff |B
. The circled cross marks the possible
Fig. 2. Non-factorizable contributions to K
insertions of the virtual photon line. Diagrams that follow from (c) and (e) by symmetry are not
shown. Upper line: hard spectator scattering. Lower line: diagrams involving a B K form factor
(the spectator quark line is not drawn for these diagrams).
from the spectator quark line. We then find:

MB
2 + C
4 C
6 )
eu t (u, mc )(C
2mb

3 + C
4 C
6 4mb /MB C
3 , (23)
5 ) + ed t (u, 0)C
+ ed t (u, mb )(C
(nf)
T,+
(u, ) =
4ed C8eff
u + uq
2 /MB2
(nf)
T, (u, ) = 0,
(24)
MB
(nf)
2 + C
4 C
6 ) + ed t (u, mb )(C
3 + C
4 C
6 )
T ,+
(u, ) =
eu t (u, mc )(C
mb

3 ,
+ ed t (u, 0)C
(25)

8C8eff
MB
(nf)
T , (u, ) = eq
MB q 2 i& u + uq 2/MB2

6MB
2 + C
4 + C
6 )
h uM
B2 + uq 2 , mc (C
+
mb

4 + C
6 )
3 + C
+ h uM
B2 + uq 2 , mb (C

2
2
3 + 3C
4 + 3C
6 )
+ h uM
B + uq , 0 (C

8
5 15C
6 ) .
(C
(26)
3C
27
3 C
5 )/3, eu = 2/3, ed = 1/3
Here C8eff = C8 +C3 C4 /6 +20C5 10C6/3 = C8 +(4C
and eq is the electric charge of the spectator quark in the B meson. h(s, mq ) has been
defined above. The functions ta (u, mq ) arise from the two diagrams of Fig. 2(b) in which
the photon attaches to the internal quark loop. They are given by
t (u, mq ) =

2MB
q2
I1 (mq ) + 2 2 B0 uM
B2 + uq 2 , mq B0 q 2 , mq ,
uE
u E
(27)
2MB
I1 (mq )
uE

2

uM
B2 + uq 2
2
2
B
uM
B
q
,
+
uq
,
m
,
m
+
0
q
0
q
B
u 2 E 2
where B0 and I1 are defined as

1
B0 (s, mq ) = 2 4m2q /s 1 arctan
,
4m2q /s 1
33
t (u, mq ) =
I1 (mq ) = 1 +
and
2m2q
u(M
B2
q 2)

L1 (x+ ) + L1 (x ) L1 (y+ ) L1 (y ) ,
1/2
m2q
1
1
x =
,
2
4 uM
B2 + uq 2

2
1 mq 1/2
1
2
y =
,
2
4
q

2
x 1
x
ln(1 x)
+ Li2
L1 (x) = ln
.
x
6
x 1
(28)
(29)
(30)
(31)
(32)
The correct imaginary parts are obtained by interpreting m2q as m2q i&. Closer inspection
shows that contrary to appearance none of the hard-scattering functions is more singular
than 1/u as u 1. It follows that the convolution integrals with the kaon light-cone
distribution are convergent at the endpoints.
The limit q 2 0 (E MB /2) of the transverse amplitude is relevant to the decay

. The corresponding limiting function reads
BK

2m2q
4
t (u, mq )|q 2 =0 =
(33)
[L
(x
)
+
L
(x
)]|
1+
2
1 +
1 q =0 .
u
uM
B2
In the same limit the longitudinal amplitude develops a logarithmic singularity, which
K
+ decay
is of no consequence, because the longitudinal contribution to the B
2
rate is suppressed by a power of q relative to the transverse contribution in this limit.
It does,
however, imply that the longitudinal amplitude itself is sensitive to distances of
order 1/ q 2 and not perturbatively calculable unless q 2 mb QCD . (This long-distance
sensitivity appears already at leading order in (14) through the light-quark contributions
to Y (q 2 ).)
2.3. Next-to-leading order form factor correction
(1)
The next-to-leading order coefficients Ca in (15) contain a factorizable term from

expressing the full QCD form factors in terms of a , related to the s -correction to the
i in Eqs. (12), (14). We write Ca(1) = Ca(f) + Ca(nf) . The factorizable correction reads [4]:

m2b
(f)
eff
C = C7 4 ln 2 4 L ,
(34)

m2
MB 2
(f)
Y q (2 2L)
C = C7eff 4 ln 2b 6 + 4L +
(35)
2mb
34
with
L
m2b q 2
q2
ln
1
.
q2
m2b
(36)
The brackets multiplying C7eff include the term 3 ln(m2b /2 ) 4 from expressing the MS
quark mass in the definition of the operator O7 in terms of the b quark pole mass according
to (5). The non-factorizable correction is obtained by computing matrix elements of fourquark operators and the chromomagnetic dipole operator represented by diagrams (c)
through (e) in Fig. 2.
The matrix elements of four-quark operators require the calculation of two-loop
diagrams with several different mass scales. The result for the currentcurrent operators
O1,2 is presented in [10] as a double expansion in q 2 /m2b and mc /mb . Since we are
only interested in small q 2 , this result is adequate for our purposes. (Note that only the
result corresponding to Fig. 1ae of [10] is needed for our calculation.) The 2-loop matrix
elements of penguin operators have not yet been computed and will hence be neglected.
Due to the small Wilson coefficients of the penguin operators, this should be a very good
approximation. The matrix element of the chromomagnetic dipole operator [Fig. 2(c)] is
also given in [10] in expanded form. The exact result is given in Appendix B. All this
combined, we obtain
(nf)
2 F (7) C eff F (7)
CF C = C
8
2
8

q2
1 (9)
(9)
(9)
eff (9)

C2 F2 + 2C1 F1 + F2
+ C8 F8 ,
2mb MB
6
2 F (7) + C eff F (7)
CF C (nf) = C
8
2
8

MB (9)
1 F (9) + 1 F (9) + C eff F (9) .
+
C2 F2 + 2C
8
1
8
2mb
6 2
(37)
(38)
(7,9)
The quantities F1,2
can be extracted from [10]. 1 The quantities F8(7,9) are given in
Appendix B or can also be extracted from [10] in expanded form. In expressing the result in
1,2 , we have made use of F (7) + F (7)/6 = 0. We also substituted
terms of the coefficients C
1
2
C8 by C8eff , taking into account a subset of penguin contributions.
2.4. Weak annihilation

In addition to the two sets of next-to-leading order corrections discussed up to now,
there exist vertex corrections to the weak annihilation graph, as shown in Fig. 3. The
distinction between weak annihilation and the previous spectator scattering diagrams is
not clear-cut. The diagrams in Fig. 2(b) with the photon attached to any of the external
lines can also pass as a correction to the leading-order weak annihilation graph. In fact, the
scale-dependent logarithm in h(q 2 , mq ) in (26) cancels part of the scale-dependence of the
leading order annihilation amplitude, (17).
1 We thank M. Walker for providing us with additional data points which cover the range 0.25 m /m 0.35
c
b
needed for the subsequent numerical analysis.
35
Fig. 3. Vertex corrections to the weak annihilation diagram in Fig. 1(c). Only the photon coupling to
the spectator quark line contributes at leading order in 1/mb as indicated by the circled cross.
For b s transitions currentcurrent operators with large Wilson coefficients cannot be

contracted to give diagrams of the type shown in Fig. 3. The total contribution is therefore
suppressed by three factors: the strong coupling constant; the small Wilson coefficients of
QCD penguin operators; and the suppression of the longitudinal amplitude in the decay
rate for small q 2 . We shall see subsequently that the leading-order weak annihilation effect
is already small. Hence, although the radiative corrections to this leading-order term are of
some conceptual interest in connection with renormalization of the B meson distribution
amplitude, we shall not include the tiny radiative corrections represented in Fig. 3 into our
numerical analysis.
2.5. Summary
(p , )|Heff |B(p)

We now summarize our main result. The matrix element (q, )K
is not scheme- and scale-independent in the standard operator formalism with an on-shell
basis, unless the photon is on-shell. Therefore T, (q 2 ) are not physical quantities. From
the expressions for the decay rates given below, it will be clear that the following three
quantities are independent of the conventions chosen to renormalize the weak effective
Hamiltonian:
T (0)
= C7eff + ,
C7
(39)
(0)

2mb MB eff
2mb MB T (q 2 )
= C9 + Y q 2 +
C7 + ,
C9, q 2 C9 +
(40)
2
2
q
(q )
q2

2mb eff
2mb T (q 2 )
C9, q 2 C9
= C9 + Y q 2 +
C
2
MB (q )
MB 7

2
4MB
fB fK
MB B, ()
4 )
eq
(C3 + 3C
d
mb
Nc MB (E/mK ) (q 2 )
MB q 2 i&
+ .
(41)
36
The quantity C9, (q 2 ) depends on the charge of the B meson through eq , but this is
suppressed in our notation. The ellipses denote the s -corrections calculated above and
defined through (15). The explicit expressions for the quantities defined in (15) are given
in (34)(38) and (20)(26). We note that C7 , C9, and C9, depend on the conventions
adopted in [4] to define the soft form factors and . In addition C7 depends also on the
b quark mass renormalization scheme (but mb C7 does not). We discuss this further after
(60).
We verified that our results are scale-independent to the required order. More precisely,
we have

d
C7 , C9, q 2 , C9, q 2 = O s2 , s C36 .
d
(42)
The uncancelled terms at order s are proportional only to the small Wilson coefficients
of QCD penguin operators. They are related to the unknown two-loop matrix elements
of penguin operators, and to the incomplete evaluation of weak annihilation effects in the
case of C9, (q 2 ). Numerically, however, the missing contributions are negligible. When the
strong interaction corrections to (39)(41) are included, the next-to-leading logarithmic
expression for C7eff [6] must be used. However, because C9 ln(MW /) 1/s at
leading order, the coefficient C9 is needed to next-to-next-to-leading logarithmic order. The
relevant initial conditions for the renormalization group evolution can be taken from [14]
and we have incorporated them into our analysis. (The necessary renormalization group
formulae are summarized in Appendix C.) However, the three-loop anomalous dimension
matrix needed for the evolution is presently not known. Except for this (and the numerically
K
+ is complete at next-toinsignificant terms discussed above), our result for B
next-to-leading logarithmic order.
3. Numerical analysis of C7 , C9, , C9,

3.1. Specification of input parameters
Wilson coefficients We begin our discussion of input parameters with the Wilson
coefficient C9 , which we need to next-to-next-to-leading logarithmic order. In our
analysis we use the complete next-to-next-to-leading logarithmic expression derived in
Appendix C, in which we set the unknown 3-loop anomalous dimensions (1) and (2) to
zero. To estimate the uncertainty this introduces we note that the neglected terms read

T
[V 1 (2) V ]ij
as (MW ) (1)
(1)
1
C9 =
D1 (, MW )V
V
,
D1 (, MW ) +
20
40 + i(0) j(0)
2
(43)
where all notation is defined in Appendix C and the subscript 2 refers to the component
of the six-component vector in brackets. Under reasonable assumptions on the pattern of
the unknown anomalous dimension matrix (vector) we observe that the largest contribution
37
Table 1
Wilson coefficients at the scale = 4.6 GeV in leading-logarithmic (LL) and next-to-leading(5)
= 0.220 GeV, m
t (m
t ) = 167 GeV, MW =
logarithmic order (NLL). Input parameters are
MS
80.4 GeV, and sin2 W = 0.23. 3-loop running is used for s . We also give C9,10 at NNLL obtained
as described in the text
LL
NLL
LL
NLL
1
C
2
C
3
C
4
C
5
C
6
C
0.257
0.151
1.112
1.059
0.012
0.012
0.026
0.034
0.008
0.010
0.033
0.040
C7eff
C8eff
C9
C10
C9NNLL
NNLL
C10
0.314
0.308
0.149
0.169
2.007
4.154
0
4.261
4.214
4.312
(1)
to the right-hand side comes from 2 . A rough estimate is therefore

C9 1032(1)

s ()
s (MW )
0.74

1 .
(44)
(1)
Allowing |2 | < 100, we conclude that at scales of order of the b quark mass C9 is
(1)
known in the standard model to an accuracy of about 0.1. (If 2 is much smaller than
the upper limit we assume, other terms may dominate C9 , but the uncertainty estimate
should remain valid.) This is to be compared to an error of 0.05 due to the error on the
top quark MS mass, m
t (m
t ) = (167 5) GeV. The value of C9 at the scale = 4.6 GeV
is given in Table 1 together with the remaining Wilson coefficients. The electroweak input
parameters that go into these numbers are summarized along with other input parameters
in Table 2. The strong coupling is always evolved according to the 3-loop formula.
Quark masses Up to now we have assumed that mb is the pole mass. It is usually assumed
that the pole mass of the b quark is less well known than the MS mass or the potentialsubtracted (PS) mass [15]. (The PS mass replaces the pole mass in quantities in which the b
quark is nearly on-shell, but has some large infrared contributions removed.) We therefore
replace mb by the PS mass through the relation
mb = mb,PS (f ) +
4s
f
3
(45)
and use mb,PS (2 GeV) = (4.6 0.1) GeV [16] as an input parameter. An exception to this
replacement is applied to the function Y (s) in which we keep the b quark pole mass
computed according to (45) in the small contributions from b quark loops. Apart from
reinterpreting mb as mb,PS (2 GeV) the only consequence of introducing the PS mass is an
additional term 4f /mb that appears in the round brackets multiplying C7eff in (34), (35).
We use the charm quark pole mass, mc = (1.4 0.2) GeV.
38
Table 2
Summary of input parameters and estimated uncertainties
MW
t)
m
t (m
|Vt s |
em
80.4 GeV
167 5 GeV
0.041 0.003
1/137
QCD
mb,PS (2 GeV)
mc
220 40 MeV
4.6 0.1 GeV
1.4 0.2 GeV
(nf =5)
fB
1
B,+
fK , (1 GeV)
fK ,
),
a1 (K
),
a2 (K
MB (0)/(2mK )
(0)
180 30 MeV
(3 1) GeV1
185 10 MeV
225 30 MeV
0.2 0.1
0.05 0.1
0.47 0.09
0.35 0.07
Meson parameters Many of the relevant meson parameters are not directly known from
experiment. QCD sum rule calculations substitute for this lack of information and we use
[17] as our source of information. A summary of all input parameters together with their
assumed errors is given in Table 2. We now discuss some of the input parameters in more
detail.
By convention [4] the soft form factors (including the one for a pseudoscalar kaon) at
zero momentum transfer are related to the full form factors by
P (0) = F+K (0),
(0) =
MB
V K (0),
MB + mK
MB
(0) = AK
(0).
0
2mK
(46)
to all orders in perturbation theory. For A0 (0) we have taken the QCD sum rule result
from [17] and this gives the soft form factor (0) listed in Table 2. The choice (0) =
0.35 0.07, also given in that table, requires comment since (46) and V K (0) from [17]
would give (0) = 0.39 0.06. The motivation for our choice of input is related to the
fact that the QCD sum rule prediction for T1K (0)/V K (0) deviates from the behaviour
expected in the heavy quark limit [4]. An alternative way of computing (0) uses (0)
0.93T1K (0) (with T1K (0) evaluated at the scale = mb ). Taking T1K (0) from [17] gives
K

(0) 0.35. Using this smaller input has the advantage that our result for the B
decay rate is automatically consistent with an alternative representation of this decay rate
in terms of the full QCD form factor T1K (0). (This will be discussed in more detail later.)
K
at next-to-leading
Furthermore, we shall see that the decay rate predicted for B
order favours form factors smaller than those estimated with QCD sum rules. We may then
K
) and hence also T K (0)
take the point of view that (0) is determined by (B
1
and V K (0) through the relations of [4]. This logic of course implies that we take seriously
the heavy quark limit prediction for the decay rate and form factor relations in the heavy
quark limit. The energy dependence of the form factors is assumed to be

q 2 = (0)
1
1 q 2 /MB2
2
,

= (0)
as predicted by power counting in the heavy quark limit.
1
1 q 2 /MB2
3
,
(47)
39
The hard-spectator scattering (and annihilation) contribution depend on the lightcone wave functions of the B-meson and the light meson in the final state which are
characterized in terms of decay constants, Gegenbauer coefficients etc. The leading
contribution in the heavy quark mass expansion involves the leading-twist distribution
amplitudes of light mesons only. We truncate the expansion of these functions into
(3/2)
Gegenbauer-polynomials Ci
at second order, i.e., (a = , )

(3/2)
(3/2)
C
C
(2u 1) + a2 K
(2u 1) .
K ,a (u) = 6u(1 u) 1 + a1 K
a 1
a 2
(48)
The values for the Gegenbauer coefficients are taken from [17], but for simplicity we
) and ai (K
) . We also enlarge the error,
neglect the small differences between ai (K
see Table 2. In the same table we also give numerical values for the light meson decay
constants [17]. (The Gegenbauer coefficients and the decay constant for a transversely
polarized vector meson, f , are scale-dependent and assumed to be evaluated at the
scale 1 GeV. The variation of Gegenbauer moments has a negligible effect on our
result compared to the overall parameter uncertainties. We therefore neglect the scaledependence of the Gegenbauer moments in the numerical analysis, but we evolve f
according to f () = f (0 )(s ()/s (0 ))4/23 .) The renormalization scale in the hardscattering and annihilation terms is chosen lower than in the form factor contributions
in order to reflect the fact that the typical virtualities in the hard scattering term are of
order mb QCD rather than m2b . If 1 (assumed to be of order mb ) is the scale in the form
factor term, we choose (1 h )1/2 with h = 0.5 GeV in the hard-scattering term. This
convention applies to all scale-dependent quantities in the hard-scattering term including
s and the Wilson coefficients.
The two B meson light-cone distribution amplitudes enter only through the two
moments

B,+ ()
1
B,+ = d
(49)
,
2
1
B, q =

d
0
B, ()
.
q 2 /MB i&
(50)
The moment of B,+ is identical to the moment that appears also in non-leptonic B
decays [2], the decay B l [18] and in the factorization of form factors [4]. The
appearance of B, at leading order in the heavy quark expansion is a new aspect of
K
+ .
the decay B
The light-cone distribution amplitudes B, () appearing in (49), (50) are not wellconstrained at present and provide a major source of uncertainty in our calculation. Some
general properties have been derived in the literature [4,19]. The equations of motions for
the light quark in the B-meson lead to
1
B, () =
0
d
B,+ (/)

B,+ () = B,
().
(51)
40
1
2
We conclude from this that 1
B,+ = B, (0). Furthermore B, (q ) must diverge
logarithmically for q 2 0, provided that 1
B,+ = 0:
lim
q 2 0
2
1
1
B, q = B,+
1
d
.
(52)
2
On the other hand, for q 2 = O(MB QCD ) the moment 1
B, (q ) is finite and of
order 1/QCD . The equations of motion for the heavy quark relate the first moments
B, to the mass difference HQET = MB mb , leading to B,+ = 2B, =
4 HQET/3. For B,+ the upper bound 4 HQET/3 has been derived in [18]. The simple
model functions
1 /0
B, () =
e
B,+ () = 2 e/0 ,
(53)
0
0
with 0 = 2 HQET/3, that have been proposed in [19], are consistent with these general
constraints, leading to B,+ = 2 HQET/3, and
2
1
B, q =
eq
2 /(M )
B 0

Ei q 2 /MB 0 + i ,
(54)
where Ei(z) is the exponential integral function. Numerically, for HQET 500 MeV,
1
with an estimated error of about 1 GeV1 . For the same
one obtains 1
B,+ 3 GeV
1
2
parameter values the absolute value of 1
B, (q ), normalized to B,+ is plotted in Fig. 4.
There exist alternative proposals for B-meson light-cone wave functions, for instance from
the BauerStechWirbel model [20] or variants of it. Since in these models the distinction
between B,+ () and B, () is not made, we refrain from presenting a thorough
comparison of different models and stick to (54) to evaluate the moment of B, ().
This means that we neglect a systematic uncertainty related to the shape of B, (), but
this uncertainty is irrelevant numerically, because the moment in question appears only in
the small annihilation contribution and the small correction to C9, (q 2 ).
1
2
2
Fig. 4. The absolute value of the ratio of B-meson moments |1
B, (q )|/B,+ as a function of q .
1
1
The B-meson distribution amplitudes are taken as in (53) with 0 = (3 1) GeV .
41
All input values from the meson sector together with their estimated uncertainties are
summarized in Table 2. We note that apart from the renormalization scale uncertainty and
the error in our knowledge of s , the most important uncertain parameters can be collected
into a single factor
2 fB fK ,a
Nc MB B,+ a (0)
(55)
that determines the relative magnitude of the hard-scattering versus the form factor term.
Adding all errors in quadrature, this factor is uncertain by about 50%, where the largest
error is currently from 1
B,+ .
3.2. Exclusive effective Wilson coefficients
Having specified our numerical input, we now discuss the three effective Wilson
coefficients C7 , C9, (q 2 ) and C9, (q 2 ).
K
is proportional.
We begin with the quantity |C7 |2 to which the decay rate of B
In Fig. 5 we show the renormalization scale dependence of this quantity at leading and at
next-to-leading order. We also show a curve that corresponds to setting the hard-scattering
(1)
term to zero, i.e., to taking into account only the correction Ca in (15). The reason for
considering this term separately is that it should cancel the sizeable leading-order scale
dependence, while the hard scattering correction is a physically different effect that appears
first at next-to-leading order. Fig. 5 shows that this is indeed correct. The hard scattering
correction reintroduces a mild scale-dependence. The most important effect is however a
large enhancement of |C7 |2 at next-to-leading order. At the scale mb = 4.6 GeV we find

|C7 |2NLO |C7 |2LO 1.78,
(56)
which corresponds to a sizeable, but not unreasonable 33% correction on the amplitude
level. The form factor and hard-scattering correction contribute about equally to this
(1)
enhancement. More precisely, the [non-]factorizable part of C (defined in (15)) is a
8% [+24%] correction to the real part of the amplitude, the [non-]factorizable part
Fig. 5. Renormalization scale-dependence of |C7 |2 at leading (LO) and next-to-leading order (NLO).
The curve NLO1 shows the NLO result without the spectator scattering correction.
42
(1)
of T,+
is a +11% [+5%] correction (all numbers at = mb ). The error on |C7 |2 is
estimated by combining in quadrature the error from all input parameters as specified in the
previous section and from allowing the renormalization scale to vary from mb /2 to 2mb .
(The scale in the hard scattering term is accordingly lower, see the discussion above.) The
result is
|C7 |2NLO = 0.175+0.029

0.026,
(57)
where the largest errors are from scale dependence (0.014), 1

B,+ (0.015), (0)
(0.009), and QCD (0.010). The value given refers to using the PS mass mb,PS (2 GeV),
since only the product mb,PS (2 GeV)|C7 | is convention-independent. The NLO result for
|C7 |2 must be compared to |C7 |2LO = 0.098 with a 30% error from scale dependence
alone. At this place it is also appropriate to note that our result is based on the heavy
quark expansion, and therefore there exist corrections of order QCD /mb at the amplitude
level. At present we have no means of quantifying these corrections in a systematic
way.
The q 2 -dependent coefficients C9, (q 2 ), C9, (q 2 ) defined in (40), (41) are shown in
Fig. 6. The left panels display the reduction of renormalization scale dependence in going
from leading order to next-to-leading order (including the intermediate approximation
with the hard scattering term set to zero). The right panel gives the complete NLO
result with all input parameter uncertainties and scale dependence included in the error
estimate.
The characteristic features of the result can be understood from the discussion of C7
above. Since C9 > 0 and C7eff < 0, the contribution to C9,a (q 2 ) from the virtual photon
matrix element, Ta (q 2 ), is always negative (a = , ). In the case of a transverse virtual
photon, this contribution is enhanced by a factor MB2 /q 2 owing to the real photon pole and
hence dominates C9, at small q 2 . This causes Re(C9, ) to change its sign and results in
the characteristic shape of |C9, |. Due to the large contribution from the photon pole the
next-to-leading order contribution is again substantial and the value of momentum transfer
q02 at which Re(C9, ) changes its sign is increased. Since q02 determines the location of the
forwardbackward asymmetry zero, this important correction will be discussed in some
detail in Section 5.2. The photon pole is absent in the longitudinal coefficient C9, , so
the NLO correction to it is much smaller. The form factor and hard scattering correction
(both small) nearly compensate each other leaving no net modification of the leading order
coefficient. Contrary to the transverse coefficient, the longitudinal one depends on the
charge of the spectator quark in the B meson, i.e., on the charge of the decaying B meson.
This effect is already present at leading order via the annihilation contribution. At NLO
it amounts to a few percent of C9, , the precise magnitude being q 2 -dependent. We return
K
+ decay
to a discussion of this isospin-breaking effect in the context of the B
spectrum.
The peculiar discontinuity in the error band around q 2 = 6 GeV2 in two of the right
panels of Fig. 6 is unphysical and related to the fact that in computing the error we allow
the charm quark mass to be as small as 1.2 GeV. The discontinuity is a consequence
of the cc threshold in the charm loop diagrams and reminds us that the validity of our
43
Fig. 6. Momentum-transfer dependence of |C9, |, |C9, | and Re(C9, ). The left panels show the LO,
NLO and NLO result without spectator scattering (NLO1 ) for fixed input parameters. The width
of the bands demonstrates the renormalization scale dependence estimated by variation from mb /2
to 2mb . The right panel shows the next-to-leading order result including renormalization scale and
input parameter uncertainties (all added in quadrature), and for comparison the leading order result
(dashed). The solid line in the center of the band refers to the NLO result with default parameters.
|C9, | is given for B mesons.
calculation is not only limited by the requirement that the K momentum is large (which
puts an upper bound on q 2 ), but also by the lack of a model-independent treatment of the
resonant charmonium contributions. The physical threshold begins at MJ2/ 9.6 GeV2 ,
but the perturbative approximation is expected to fail earlier. Model-dependent studies of
the charmonium contributions suggest that the perturbative approximation should be valid
up to q 2 (67) GeV2 [21,22]. (This supposes that we discard a charm mass as small as
1.2 GeV in estimating effects related to the charm threshold.)
44
K

4. The decay B
K
in the heavy quark limit is given by
The decay rate for B

G2 |V V |2
m2 3
K
= F t s t b M 3 1 K em m2b,PS (0)2 |C7 |2 ,
B
B
8 3
4
MB2
(58)
with mb,PS mb,PS (2 GeV) the PS mass. We included the kaon mass squared in a phase
space and kinematic correction, but we generally neglect such terms in the dynamical
quantity T (0) = (0)C7 . By helicity conservation the kaon is transversely polarized and
only the transverse soft form factor (0) appears in the decay rate. In numerical form, the
branching fraction is

B
mb,PS 2 (0) 2
K
= 7.9+1.8 105
Br B
1.6
1.6 ps
4.6 GeV
0.35
+3.5
5
= 7.93.0 10
(59)
using |C7 |2 = 0.175+0.029
0.026 and mb,PS = (4.6 0.1) GeV, (0) = 0.35 0.07, |Vt s | =
0.041 0.003, em = 1/137, as in Table 2. (B denotes the B meson lifetime.) For
comparison we note that the leading order prediction is 4.5 105 . The large difference
reflects the enhancement of |C7 |2 at next-to-leading order as discussed in Section 3.2. The
branching ratio predicted for the default parameter set is nearly twice as large as the current
experimental averages [2325]

0
0
K
= (4.54 0.37) 105 ,
Br B
(60)
exp

5

Br B K
(61)
= (3.81 0.68) 10 .
exp
Note that we predict no difference of the neutral and charged B meson decay rates at
leading order in the heavy quark expansion, so that the main effect of isospin breaking on
branching fractions is the different lifetimes of the charged and neutral mesons.
Before speculating on the origin of this discrepancy, we note that there are alternative
ways of representing the result of our calculation. Since the decay amplitude is proportional
to a single form factor, T1K (0), there is no need to introduce the soft form factor (0),
and we can express the decay rate directly in terms of the full QCD form factor. (The need
+ final state, which
to introduce the soft form factors arises when we consider the K
involves many form factors that all relate to the same soft form factors.) Furthermore the
dominant dependence on the b quark mass arises through the factor mb in the definition
b (m
b ) as a prefactor
of the operator O7 , so it also more natural to keep the MS mass m
rather than the PS mass. This alternative representation amounts to a redefinition of C7
in which all factorizable corrections vanish at the scale m

b , if we understand T1K (0) as
renormalized at the scale m
b . We have computed the coefficient C7 also in this alternative
scheme (indicated by the prime) and find that (58) is modified as follows:
m2b,PS (0)2 |C7 |2
m
b (m
b )2 T1K (0)2 |C7 |2
|C7 |2 = 0.175+0.029
0.026
|C7 |2 = 0.165+0.018
0.017.
(62)
45
The error is smaller in the primed scheme, because the hard-scattering correction is much
reduced once the factorizable correction is reabsorbed into the full QCD form factor. This
implies less sensitivity to the uncertain parameters that normalize this correction. The two
values given in (62) are consistent with each other if one relates T1K (0) = 1.08 (0) [4]
b (m
b ) = 4.4 GeV.
and takes into account that mb,PS (2 GeV) = 4.6 GeV corresponds to m
Several facts may explain the difference between the predicted and observed decay rate.
The first and most interesting possibility is a modification of the standard model at short
distances that would result in a smaller value of |C7 |. However, the modification needed
to explain the observed decay rate is far too large not to be ruled out by the agreement
Xs . It appears equally
between experiment and theory for the inclusive decay B
implausible that new interactions would modify only the spectator scattering and hence not
show up in the inclusive rate. We therefore conclude that the explanation must be sought
in our understanding of QCD effects. Most of the NLO enhancement is related to the nonfactorizable form factor type correction which appears in a nearly identical form in the
inclusive decay rate as well. Hence this enhancement cannot be simply dismissed without
putting the agreement for the inclusive decay into question. A second and less interesting
possibility is therefore to invoke a large 1/mb correction that would render our calculation
unreliable. Given the smallness of the non-factorizable hard scattering correction it is not
obvious how such a large dynamical enhancement of 1/mb terms could be explained.
This is in particular so as large 1/mb corrections known to exist for non-leptonic decays
[2,26] such as chirally enhanced corrections and large weak annihilation contributions
are absent for decays into vector mesons.
We therefore consider seriously the possibility that the form factors at q 2 = 0 are
substantially different from what they are assumed to be in the QCD sum rule approach
or in quark models. This possibility is entertained in Fig. 7, where we show how the
experimental data constrains the soft form factor (0) and the normalization of the hard
scattering term. (We take all parameters other than 1
B,+ in the normalizing factor constant
1
and let B,+ be a representative. That is, a variation of 1
B,+ should be considered as a
Fig. 7. Fit to (0) and B,+ . The solid curves give the 1 , 2 and 3 regions that follow from the
K
decay rate. The light solid curves mark the ranges assumed for (0) and B,+
observed B
in Table 2.
46
variation of a combination of parameters of the type (55). All other parameter uncertainties
are approximately accounted for by adding a 20% uncertainty to the measured branching
fraction which we combine with the experimental error in quadrature.) The result of this fit
is that
T1K (0)=m = 0.27 0.04

(63)
[ (0) = 0.24 0.06],
b
where the error may be loosely interpreted as a one standard deviation error. This
determination yields a smaller form factor than a similar determination in [12], where the
NLO correction is not included and the bottom quark mass is handled differently.
. This decay
Our results can be extended in a straightforward way to the decay B
is particularly interesting in searches for extensions of the standard model, because of the
suppression of b d transitions in the standard model and the simultaneous chirality
final
suppression. Except for trivial adjustments the most important difference to the K
state is that the different flavour structure allows weak annihilation to proceed through the
2 . This annihilation contribution
currentcurrent operator with large Wilson coefficient C
is power-suppressed in 1/mb , but the suppression is compensated by the large Wilson
coefficient and the occurrence of annihilation at tree level. The annihilation contribution is
calculable and has to be taken into account in a realistic analysis [13,27]. This will be done
elsewhere [28,29], and we restrict ourselves to b s transitions in this paper.
K
+
5. The decay B
K
+ we obtain the double differential decay spectrum (summed
For the decay B
over final state polarisations, and lepton masses neglected)

G2F |Vts Vt b |2 3 2 2 3 em 2
d 2
=
M
q
,
m
K
B
dq 2 d cos
128 3
4

2

2 2

2q
C9, q 2 2 + C 2
q
1 + cos2
10
MB2

E (q 2 ) 2

2
C9, q 2 2 + C 2 q 2 2
+ 1 cos
10
mK

2
2 2
2
8q
cos 2 q Re C9, q C10 ,
(64)
MB
where

q 2 , m2K =

m4K 1/2
q 2 2 2m2K
q2
1 2
,
1+ 2 + 4
MB
MB2
MB
MB
(65)
and q 2 is the invariant mass of the lepton pair. The angle refers to the angle between the
positively charged lepton and the B meson in the center-of-mass frame of the lepton pair.
We have kept terms of order m2K in an overall factor related to phase space and kinematics
but emphasize that the expression in brackets neglects such terms. The first two terms
47
with angular dependence (1 cos2 ) correspond to the production of transversely and

longitudinally polarized kaons, respectively. The third term generates a forwardbackward
asymmetry with respect to the plane perpendicular to the lepton momentum in the centerof-mass frame of the lepton pair. The factor multiplying the Wilson coefficient C10
in (64) arises from the factorizable corrections to the form-factor A2 and is given by
(cf. Eq. (66) in [4])

2 fB fK 1
s CF
s CF 2q 2
B+
=1+
(2 + 2L)
4
4 E 2 Nc MB (E/mK ) (q 2 )
1
0
du
(u)
u
(66)
with L defined in (36) and E = (MB2 q 2 )/(2MB ). (This factor could be eliminated by
choosing a factorization convention different from (46) for the soft form factor and by
redefining C9, accordingly.)
Without going into much detail, we remark that the result for the decay into pseudoscalar
K
+ , is easily obtained from the corresponding expressions for the decay
mesons, B
into longitudinally polarized vector mesons. The angular distribution is predicted to be
proportional to (1 cos2 ) and for the lepton invariant mass spectrum we obtain

2 2
G2F |Vts Vt b |2 3 2 2 3 em 2 2 2
d
+ C2 ,
C
q
q
(67)
=
M
q
,
m
P
9,P
K
B
10
dq 2
96 3
4
where the quantities analogous to those in (6), (40) are now defined as

2mb TP (q 2 )
C9,P q 2 = C9 +
,
MB P (q 2 )
and
(68)

)Heff
B(p)
(q, )K(p

GF
gem mb TP (q 2 ) 2
= Vts Vt b
(69)
q p + p MB2 m2K q .
2
MB
4
2
The non-factorizable contributions to TP are the same as those to T up to an overall sign
and the replacement fK, K , (u) fK K (u) in (15). The factorizable contributions
follow from the ratio of the tensor form factor fT (q 2 ) and the soft form factor P (q 2 ) as
calculated in [4]. The net result for C9,P is then
C9, (q 2 )
C9,P q 2 =
,
(q 2)
(70)
where the replacements fK, K , (u) fK K (u), (q 2 ) P (q 2 ) are understood to be performed on the right-hand side.
5.1. Lepton invariant mass spectrum
The lepton invariant mass spectrum d Br /dq 2 , obtained by integrating (64) over cos ,
is shown in Fig. 8. (We use the B meson lifetime B = 1.65 ps.) As the decay rate is
48
Fig. 8. Differential decay rate d Br(B K + )/dq 2 at next-to-leading order (solid center
line) and leading order (dashed). The band reflects all theoretical uncertainties from parameters and
scale dependence combined, with most of the uncertainty due to the form factors , (0).
dominated by the contribution of longitudinally polarized K mesons and the contribution

from O10 except for small q 2 , the impact of the next-to-leading order correction is small
for q 2 > 2 GeV2 . (This can be directly inferred from the small correction to |C9, | as seen
in Fig. 6.) The apparently rather large uncertainty of our prediction is mainly due the form
factors with their current large uncertainty and to a lesser extent due to |Vt s | and the top
quark mass. It may be hoped that in the longer term future the form factors could be known
with much greater confidence. The uncertainties can then be reduced by a factor of three
in which case the limiting factor of our prediction is probably 1/mb corrections, all other
parameter and scale variations being negligible in comparison.
Since |C9, | depends on the charge of the decaying B meson, there is a small
amount of isospin breaking. While this effect is q 2 -dependent for |C9, |, the combination
of coefficients that enters the lepton invariant mass spectrum turns out to be almost
independent on q 2 . The effect is, however, very small,
=
0 K
0 + )/dq 2
d (B K + )/dq 2 d (B
1%.
0 K
0 + )/dq 2
d (B K + )/dq 2 + d (B
(71)
For the differential branching fractions the main isospin breaking effect arises from the
lifetime difference of the charged and neutral B mesons.
The impact of the NLO correction on the lepton invariant mass spectrum to the decay in
a pseudoscalar meson follows a qualitatively similar pattern as for the vector meson final
state, in particular as the decay rate involves only a quantity closely related to |C9, | (see
(70)). We therefore do not discuss this decay in further detail here, but note that the lepton
invariant mass spectrum develops a logarithmic singularity for small q 2 . This is due to the
long-distance sensitivity mentioned above, which is now dominant because the photon pole
is absent. The invariant mass spectrum is therefore non-perturbative for q 2 2QCD , but
perturbative for q 2 mb QCD . Furthermore, the non-perturbative contribution is formally
power-suppressed when the lepton invariant mass spectrum is integrated from 0 to some
q 2 of order mb QCD .
49
5.2. Forwardbackward asymmetry

The QCD factorization approach proposed here leads to an almost model-independent
theoretical prediction for the forwardbackward asymmetry [30]. It has been noted in
[31] that the location of the forwardbackward asymmetry zero is nearly independent of
particular form factor models. An explanation of this fact was given in [32], where it has
been noted that the form factor ratios on which the asymmetry zero depends are predicted
free of hadronic uncertainties in the combined heavy quark and large energy limit. In [4] the
effect of the (factorizable) radiative corrections to the form factor ratios has been studied
and has been found to shift the position of the asymmetry zero about 5% towards larger
values. We are now in the position to discuss the effect of both, factorizable and nonfactorizable radiative corrections to next-to-leading order in the strong coupling constant
on the location of the asymmetry-zero, and hence to complete our earlier analysis.
We define the forwardbackward (FB) asymmetry (normalized to the differential decay
rate d (B K + )/dq 2 ) by
1

0
dAFB
1
d 2
d 2
d(cos ) 2
(72)
d(cos ) 2
.
dq 2
d /dq 2
dq d cos
dq d cos
0
Our result for the FB asymmetry is shown in Fig. 9 to LO and NLO accuracy. From (64)
it is obvious that dAFB /dq 2 Re(C9, (q 2 )), and therefore the FB asymmetry vanishes if
Re(C9, (q02 )) = 0. At leading order this translates into the relation

2MB mb eff
C9 + Re Y q02 =
C7 ,
q02
(73)
which, as already mentioned, is free of hadronic uncertainties. This effect can also be
seen in Fig. 9 where the uncertainty for the prediction of the FB asymmetry gets smaller
2
around q02 . Using our default input parameters we obtain q02 = 3.4+0.6
0.5 GeV at LO where
the error is by far dominated by the scale dependence. The numerical effect of NLO
Fig. 9. Forwardbackward asymmetry dAFB (B K + )/dq 2 at next-to-leading order (solid

center line) and leading order (dashed). The band reflects all theoretical uncertainties from parameters
and scale dependence combined.
50
corrections amounts to a substantial reduction of the FB asymmetry for intermediate lepton

invariant mass (q 2 = 25 GeV2 ) and a significant shift of the location of the asymmetry
zero to
2
q02 = 4.39+0.38
0.35 GeV .
(74)
The largest single uncertainty (about 0.25 GeV2 ) continues to be scale dependence
(varying within mb /2 < < 2mb as usual), but also the variation of mb , QCD , and
B,+ gives an effect.
Our LO result for q02 differs somewhat from the value q02 = 2.88+0.44
0.28 quoted in [32].
The reason for this difference is a combination of three effects: (i) slightly different values
for the SM Wilson coefficients, in particular a larger value for C9 in [32], (ii) a different
treatment of the b-quark mass, (iii) the inclusion of (higher-order factorizable) form factor
corrections in [32], since the full QCD form factors estimated from light-cone QCD sum
rules [17] are used. The first two issues are resolved within the present next-to-leading
order treatment but with respect to (iii) we note that the form factor ratios entering C9,
are exactly those where there exists a discrepancy between the QCD sum rule result and
the heavy quark limit [4]. The origin of this discrepancy has not been clarified, but it is
conceivable that it is related to some 1/MB corrections, which are implicitly included in
the sum rule estimate. In order to estimate the impact of using full QCD form factors on
the location of the asymmetry zero, we reabsorb the factorizable s corrections into the
physical form factors and the MS mass. The FB asymmetry then reads (with the tensor
form factors and the MS mass m
b all renormalized at the scale )

G2F |Vts Vt b |2 3 2 2 2 em 2 8q 2
1
dAFB
=
MB q , mK
C10 A1 q 2 V q 2
2
2
3
2
dq
d /dq
128
4
MB

2 s CF (nf,9) 2
C
Re C9 + Y q +
q
4

m
b
T1 (q 2 )
T2 (q 2 )
+ 2 (MB + mK )
+ (MB mK )
q
V (q 2 )
A1 (q 2 )

s CF (nf,7) 2
C
C7eff +
q
4

m
b
1
q2
1
+ 2 (MB + mK )
+ (MB mK ) 1 2
q
V (q 2 )
MB A1 (q 2 )
1
s CF 2 fB fK , B,+
4 Nc
MB
1
(nf)
du K , (u)T,+
(u)

.
(75)
0
(nf)
The non-factorizable contributions from C in (37) are now divided into two parts,
coming with the operator structure of O7 and O9 , respectively,

(nf,9) 2
1 F (9) q 2 + 1 F (9) q 2
2 F (9) q 2 + 2C
CF C
q = C
2
1
6 2

(9)
+ C8eff F8 q 2 ,
(76)

(nf,7) 2
CF C

2 F (7) q 2 + C eff F (7) q 2 .
= C
8
2
8
51
(77)
These are just the terms that also enter the inclusive decay, while the last term in (75) arises
from the hard spectator-scattering and thus only appears in the exclusive decay. Using the
relations T2 = 2ET1 /MB and A1 = 2EMB V /(MB + mK )2 , which are presumed to hold
to all orders in s , and to leading order in 1/mb [4,12], (75) can be rearranged to contain
only V and T1 /V . If furthermore the latter ratio is expanded in s and if some terms of
order m2K /MB2 are neglected, we return to the next-to-leading order result that we use
by default. On the basis of (75) and the form factor estimates from [17], we now obtain
q02 = 3.25 GeV2 at LO and q02 = 3.94 GeV2 at NLO. (To obtain this result we use mb,PS as
input and treat perturbatively the difference with m
b .) The NLO value is slightly more than
one standard deviation away from our default result in (74). The difference between the
two values could be viewed as an estimate for an additional systematic error, but the main
conclusion is that using the soft form factors or the full QCD form factors from light-cone
QCD sum rules does not cause a large ambiguity in our result for q02 . We therefore assume
q02 = (4.2 0.6) GeV2
(78)
to be a conservative estimate for the asymmetry zero.

The Wilson coefficients may also receive contributions from new particles in theories
beyond the standard model. (Of course extensions of the standard model may introduce a
larger set of operators as well.) The function C9, depends dominantly on C9 and C7eff (see
Xs
(40)). The experimental measurement of the decay rate for the inclusive decay B
already constrains |C7eff | to be close to its standard model value. The forwardbackward
asymmetry will establish the sign of C7eff and assuming this to take its standard model
K
+ provides a
value, a measurement of the FB asymmetry zero in the decay B
way to determine C9 . It is therefore instructive to consider our result as a prediction for C9
for a given value of q02 that will be measured in future experiments. In Fig. 10 we show
our NLO prediction for C9 at the scale = 4.6 GeV as a function of q02 together with the
standard model prediction (see Section 3.1). For comparison we also show the LO result
and the modification arising from using (75) instead of (64). We remark that C9 is schemedependent and so the plot refers to the particular renormalization scheme for O9 adopted
in [6].
Fig. 10 is our main phenomenological result. It illustrates that the FB asymmetry zero
can be used to test the standard model and to search for extensions, since the theoretical
uncertainties relevant for the determination of C9 are under control. In this respect the
systematic inclusion of NLO corrections, which is provided in this work, turns out to
be essential, because it reduces the renormalization scale and scheme dependences to a
large extent and corrects the leading-order result by a large amount. We emphasize that
K
+ the dependence on the
K
and B
in contrast to the absolute rates for B
K
form factors is sub-leading for the FB asymmetry zero. Thus the current difficulty
B
K
branching ratio (which may suggest a smaller value of the
to account for the B
form factor T1 (0) than usually estimated from QCD sum rules as discussed above), does not
necessarily cause a problem for our prediction of C9 vs. q02 . We also note that the resulting
52
Fig. 10. The Wilson coefficient C9 (mb ) deduced from a measurement of the asymmetry zero q02 at
next-to-leading order including all parameter uncertainties (light band), but not including the slightly
enlarged error in (78). For comparison we also show (without uncertainties) the leading order result
(dashed) and the result at next-to-leading order using (75) and the full QCD form factors from [17]
(long-dashed). The horizontal band is the standard model value C9 (4.6 GeV) = 4.21 0.12.
values of q02 are sufficiently below the threshold for charmonium resonances (J /, ),
so that a long-distance contribution to the function Y (q 2 ) can safely be neglected. We
therefore conclude that if the asymmetry zero is found at some q02 < 6 GeV2 , the shortdistance coefficient C9 (4.6 GeV) can be determined with a theoretical error of 10%. If
its value turns out to be different from the standard model value 4.21 0.12 by an amount
and if we assume that all Wilson coefficients other than C9,10 remain unmodified at the
scale MW , then (C.13) below implies a new contribution of magnitude to the coefficient
function C9 at the scale MW .
6. Concluding discussion
Building on our previous work on heavy-to-light form factors [4] and on work on
non-leptonic decays [2,3], we demonstrated that also the radiative B decays B V
and B V + can be computed in the heavy-quark limit. This allows us to solve
the problem of non-factorizable strong interaction corrections (i.e., those corrections not
related to form factors), which has so far prevented a systematic discussion of exclusive
radiative decays, as compared to their inclusive analogues. The approach presented here
does not circumvent the need to know the heavy-to-light form factors, but we may hope
that progress in lattice QCD will give us these form factors at smaller q 2 and more reliably
than at present in the longer term future.
In the present paper we have concentrated on the b s transition. We noted that the
next-to-leading order correction yields an 80% enhancement of the B K decay
rate, a large part of which is related to the NLO correction that appears also in the
inclusive decay [9,10]. The enhancement is so large that the predicted decay rate disagrees
with the observed decay rate unless the form factors at q 2 = 0 are much smaller than
what they are believed to be, or unless our theoretical prediction is made unreliable by
53
a large correction to the heavy quark limit. Perhaps the most interesting outcome of
our analysis is that while the NLO correction to the lepton invariant mass spectrum in
B K + is very small, there is a large correction to the predicted location of the
forwardbackward asymmetry zero. The particular interest in this quantity derives from
the fact that all dependence of form factors arises first at next-to-leading order. This implies
that (given the Wilson coefficient C7 ) the Wilson coefficient C9 can be determined from
a measurement of the location of the asymmetry zero with little theoretical uncertainty.
We find that this conclusion holds true in particular after including the NLO correction,
which however, shifts the predicted location of the zero by 1 GeV2 , or about 30%, with a
residual uncertainty estimated to be about (0.40.6) GeV2 . This implies that an accurate
measurement of the asymmetry zero determines the Wilson coefficient C9 at the scale mb
with an error of 10%.
The method developed here can also be applied to b d transitions. These are
particularly interesting in the search for modifications of the standard model. The better
control of standard model effects after accounting for non-factorizable corrections should
help increasing the sensitivity to such non-standard effects. In particular we note that the
approach presented here allows us to compute isospin breaking effects, which may turn
out to be a particularly nice signal of non-standard physics. A detailed discussion of b d
transitions requires, however, a more careful study of weak annihilation effects, and we
plan to return to non-standard physics in the context of such a study [29].
A systematic, but rather difficult to quantify, limitation of the factorization approach
is the poor control of 1/mb suppressed effects. These arise from various sources and we
conclude by mentioning one of them. We may ask to what extent a non-leptonic decay
B K in which the meson is assumed to convert to a photon through vector meson
dominance may affect our calculation of the direct decay into K . More accurately, the
question is to what extent the hadronic structure of the photon matters at small q 2 . It is
not difficult to see that the indirect contribution is suppressed by one power of mb in the
heavy quark limit. For the sake of the argument we return to the vector dominance picture
and note that there is a one-to-one correspondence between diagrams that appear in the
factorization approach to non-leptonic decays [2,3] and the diagrams computed here. For
example, the naive factorization contribution to the former case corresponds to our leading
order weak annihilation term; the spectator scattering contribution to non-leptonic decays
corresponds to the diagram in Fig. 2(b) with the photon attached to the internal quark loop
and so on. Counting powers of mb for all the quantities that appear in these amplitudes we
find that the direct amplitude is always larger by a factor of mb .
Note added
While completing this work, the article [33] appeared, in which the next-to-leading order
hard-scattering correction to B is calculated. The corresponding quantity should be
(nf)
obtained from the function T,+
in (23) in the limit q 2 0. The small QCD penguin
2 and
contributions have been neglected in [33], so that only the terms proportional to C
54
C8eff have to be retained for the comparison. The charm quark mass in t (u, mc ) is set to 0
in [33], presumably because an endpoint divergence is found for mc = 0. We note that our
result does not exhibit this endpoint divergence (which would invalidate the factorization
approach), but setting mc = 0 for the sake of comparison, we obtain
(nf)
T,+ (u, )
2
4ed C8eff
MB 4eu C
+
.
u
2mb u
We then differ from (4.12) of [33] by a sign in the first term and a factor of 4 in the second
term. We also note that [33] includes a factorizable form factor correction from [4] while
retaining the full QCD form factor T1 (0). This implies that the factorizable correction is
included twice.
Acknowledgements
We would like to thank C. Greub and M. Walker for helpful discussions. While this
work has been done, we became aware of related work in [28]. We thank G. Buchalla for
discussing the results of this paper prior to publication.
Appendix A. Operator bases

We use the renormalization conventions of [6], but present our result in terms of the
following linear combinations of Wilson coefficients:
1 = 1 C1 ,
C
2
2 = C2 1 C1 ,
C
6
3 = C3 1 C4 + 16C5 8 C6 ,
C
6
3
1
4 = C4 + 8C6 ,
C
2
5 = C3 1 C4 + 4C5 2 C6 ,
C
6
3
1
6 = C4 + 2C6 .
C
2
(A.1)
These definitions hold to all orders in perturbation theory. The barred coefficients are
related to those defined in [7] by [34]

i = CiBBL + s Tij CjBBL + O s2 ,
C
4
where
(A.2)
7
3
0
T =
0
23
178
27
34
9
164
27
20
9
49
20
3
23
9
23
3
160
27
16
9
146
27
2
9
0
0
0
0
13
9
.
13
3
32
55
(A.3)
9
16
3
(7,9)
Appendix B. Unexpanded form of F8
Using the conventions of [10] we obtain for the 1-loop matrix element of O8 (Fig. 2(c)
together with a set of symmetric diagrams):
8
4 11 16s + 8s 2
32
8 s
ln
ln s i
9
mb 9 1 s
9
9
(1 s )2

1
4
+
9s 5s 2 + 2s 3 B0 (s ) (4 + 2s )C0 (s ) ,
3
9 (1 s )
8 5 2s
16 1
8 4 s
ln s +
F8(9) =
[(1 + s )B0 (s ) 2C0 (s )],

9 1 s
9 (1 s )2 9 (1 s )3
F8(7) =
(B.1)
(B.2)
where s = q 2 /m2b , B0 (s ) = B0 (q 2 , m2b ) is given in (29), and the integral

1
C0 (s ) =
dx
0
x2
1
ln
x(1 s ) + 1 1 x(1 x)s
(B.3)
can be expressed in terms of dilogarithms.
Appendix C. NNLL formulae for C9 and C1,...,6

Because the anomalous dimension of O9 begins at order s0 , the Wilson coefficient C9
is needed to next-to-next-to-leading logarithmic (NNLL) accuracy. This requires also the
coefficient of the four-quark operators to this accuracy. Although not all of the necessary
inputs are currently known, we present here the solution of the renormalization group
equations to NNLL order.
We first restrict ourselves to the sector of four-quark operators. Define as () =
s ()/(4) and (as ) = 0 as2 + with 0 = 11 2nf /3. We further define the matrix
U (, MW ) such that
C() = U (, MW )C(MW ),
(C.1)
where C() is the vector of Wilson coefficients. U (, MW ) satisfies

d
U (, MW ) = T ()U (, MW ),
d ln
(C.2)
56
and the anomalous dimension matrix is expanded as

= (0) as + (1) as2 + .
(C.3)
Let V be the matrix that diagonalizes

(0)
T
V 1 (0) V = i diag ,
(0) T
, so that
(C.4)
and define

U
(0)
(, MW ) = V
as ()
as (MW )
(0) /(20 )
i
V 1 ,
(C.5)
diag
which solves (C.2) to leading order in as (). Then the NNLL expression for the evolution
matrix U (, MW ) reads

U (, MW ) = 1 + as ()J (1) + as ()2 J (2) U (0) (, MW )

2
1 as (MW )J (1) as (MW )2 J (2) J (1) ,
(C.6)
where
J (n) = V H (n) V 1 .
The matrices
H (1) ,
H (2)
(C.7)
have the entries

(1)
Hij(1) = ij i(0)
(2)
(0)
Hij = ij i
Gij
1
,
2
(0)
(0)
20
20 + i j
(C.8)
20 + (0) (0) (1) (1) 1 (1)
2
i
k
+
Hik Hkj Hij j k
(0)
(0)
0
402
4
+
0
k
i
j
G(2)
ij
40 + i(0) j(0)
(C.9)
and we have defined

T
G(n) = V 1 (n) V .
(C.10)
The same equations can also be used to solve the renormalization group equations when
the electromagnetic and chromomagnetic dipole operators are included. Then is an 8 8
matrix, and one needs to keep in mind that the 6 2 block in (n) that mixes the dipole
and four-quark operators is obtained from (n + 2)-loop diagrams, while all other entries
come from (n+1)-loop diagrams. In our numerical analysis we expand U (, MW )C(MW )
in powers of as () and as (MW ) using (C.6) and the expansion of the initial condition
C(MW ) in powers of as (MW ) and keep only terms to the required order in both expansion
parameters.
The renormalization group equation for C9 is particularly simple and can be obtained
directly by integrating the coefficients of four-quark operators. The equation reads
d
C9 () = ()C(),
d ln
(C.11)
57
where C() is the vector of Wilson coefficients of four-quark operators and

= (1) + (0)as +
(C.12)
is the 1 6 matrix that describes the mixing into O9 . The calculation of (n) involves
(n + 2)-loop diagrams. The solution of (C.11) is given by
C9 () = C9 (MW ) + W (, MW )C(MW ),
(C.13)
with the 1 6 matrix

1
W (, MW ) =
2
as ()
das
(as )
U (, MW )
(as )
(C.14)
as (MW )
with the evolution matrix U (, MW ) from the four-quark sector as defined above. The
solution to NNLL accuracy is obtained by inserting U (, MW ) to this accuracy. We
introduce the 6 6 matrices
%
(0)
&

1
as () ni /(20 )
Dn (, MW ) = V
1
V 1 , (C.15)
(0)
n /(20 ) as (MW )
i
diag
in terms of which the solution is given by

(1)
D1 (, MW )
20 as (MW )

1
1
(0) (1) + (1)J (1) D0 (, MW )

20
0

(1)
(1)
D1 (, MW )J
W (, MW ) =
2

1
as (MW )
1
2 (1)
(1) (0) +
20
0
02 0

1
+ (0) (1) J (1) + (1) J (2) D1 (, MW )
0

1 (1)
(0)
(1) (1)

+
J
D0 (, MW )J (1)
0

(2)
(1)
(1) 2
.
D1 (, MW ) J J
(C.16)
In our numerical analysis we expand (C.13) in powers of as (MW ) and keep only terms to
the required order in the expansion parameter. To NNLL we thus need C9 (MW ) to order
as and since the first term in the expansion of this quantity is of order as0 , we recover from
(C.16) the well-known result that the LL solution for C9 is induced only by mixing and is
of order 1/as .
58
References
[1] R. Ammar et al., CLEO Collaboration, Phys. Rev. Lett. 71 (1993) 674.
[2] M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Phys. Rev. Lett. 83 (1999) 1914, hep-ph/
9905312.
[3] M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 591 (2000) 313, hep-ph/
0006124.
[4] M. Beneke, T. Feldmann, Nucl. Phys. B 592 (2001) 3, hep-ph/0008255.
[5] J. Charles, A. Le Yaouanc, L. Oliver, O. Pene, J.C. Raynal, Phys. Rev. D 60 (1999) 014001,
hep-ph/9812358.
[6] K. Chetyrkin, M. Misiak, M. Mnz, Phys. Lett. B 400 (1997) 206, hep-ph/9612313.
[7] G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125, hep-ph/9512380.
[8] H.H. Asatrian, H.M. Asatrian, D. Wyler, Phys. Lett. B 470 (1999) 223, hep-ph/9905412.
[9] C. Greub, T. Hurth, D. Wyler, Phys. Rev. D 54 (1996) 3350, hep-ph/9603404.
[10] H.H. Asatrian, H.M. Asatrian, C. Greub, M. Walker, Phys. Lett. B 507 (2001) 162, hep-ph/
0103087.
[11] B. Grinstein, M.J. Savage, M.B. Wise, Nucl. Phys. B 319 (1989) 271.
[12] G. Burdman, G. Hiller, Phys. Rev. D 63 (2001) 113008, hep-ph/0011266.
[13] B. Grinstein, D. Pirjol, Phys. Rev. D 62 (2000) 093002, hep-ph/0002216.
[14] C. Bobeth, M. Misiak, J. Urban, Nucl. Phys. B 574 (2000) 291, hep-ph/9910220.
[15] M. Beneke, Phys. Lett. B 434 (1998) 115, hep-ph/9804241.
[16] M. Beneke, A. Signer, Phys. Lett. B 471 (1999) 233, hep-ph/9906475.
[17] P. Ball, V.M. Braun, Phys. Rev. D 58 (1998) 094016, hep-ph/9805422.
[18] G.P. Korchemsky, D. Pirjol, T. Yan, Phys. Rev. D 61 (2000) 114510, hep-ph/9911427.
[19] A.G. Grozin, M. Neubert, Phys. Rev. D 55 (1997) 272, hep-ph/9607366.
[20] M. Wirbel, B. Stech, M. Bauer, Z. Phys. C 29 (1985) 637.
[21] C.S. Lim, T. Morozumi, A.I. Sanda, Phys. Lett. B 218 (1989) 343.
[22] F. Krger, L.M. Sehgal, Phys. Lett. B 380 (1996) 199, hep-ph/9603237.
[23] T.E. Coan et al., CLEO Collaboration, Phys. Rev. Lett. 84 (2000) 5283, hep-ex/9912057.
[24] V. Brigljevic, BaBar Collaboration, Talk presented at the XXXVIth Rencontres de Moriond
QCD and High Energy Interactions, March 1724, 2001, Les Arcs, France.
[25] G. Taylor, Belle Collaboration, Talk presented at the XXXVIth Rencontres de Moriond
Electroweak Interactions and Unified Theories, March 1724, 2001, Les Arcs, France.
[26] M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 606 (2001) 245, hep-ph/
0104110.
[27] A. Ali, L.T. Handoko, D. London, hep-ph/0006175.
[28] S. Bosch, G. Buchalla, hep-ph/0106081.
[29] M. Beneke, T. Feldmann, D. Seidel, in preparation.
[30] A. Ali, T. Mannel, T. Morozumi, Phys. Lett. B 273 (1991) 505.
[31] G. Burdman, Phys. Rev. D 57 (1998) 4254, hep-ph/9710550.
[32] A. Ali, P. Ball, L.T. Handoko, G. Hiller, Phys. Rev. D 61 (2000) 074024, hep-ph/9910221.
[33] A. Ali, A.Y. Parkhomenko, hep-ph/0105302.
[34] K. Chetyrkin, M. Misiak, M. Mnz, Nucl. Phys. B 520 (1998) 279, hep-ph/9711280.

Neutrino mass spectrum and future beta decay

experiments
Y. Farzan a , O.L.G. Peres b,c , A.Yu. Smirnov c,d
a Scuola Internazionale superiore di Studi Avanzati, via Beirut 4, I-34014 Trieste, Italy
b Instituto de Fsica Gleb Wataghin, Universidade Estadual de Campinas, UNICAMP,
13083-970 Campinas SP, Brazil

c The Abdus Salam International Centre for Theoretical Physics, I-34100 Trieste, Italy
d Fermi National Accelerator Laboratory, P.O. Box 500, Batavia, IL 60510, USA
Abstract
We study the discovery potential of future beta decay experiments on searches for the neutrino
mass in the sub-eV range, and, in particular, KATRIN experiment with sensitivity m > 0.3 eV. Effects
of neutrino mass and mixing on the beta decay spectrum in the neutrino schemes which explain the
solar and atmospheric neutrino data are discussed. The schemes which lead to observable effects
contain one or two sets of quasi-degenerate states. Future beta decay measurements will allow to
check the three-neutrino scheme with mass degeneracy, moreover, the possibility appears to measure
the CP-violating Majorana phase. Effects in the four-neutrino schemes which can also explain the
LSND data are strongly restricted by the results of Bugey and CHOOZ oscillation experiments: apart
from bending of the spectrum and the shift of the end point one expects appearance of small kink of
(< 2%) size or suppressed tail after bending of the spectrum with rate below 2% of the expected rate
for zero neutrino mass. We consider possible implications of future beta decay experiments for the
neutrino mass spectrum, the determination of the absolute scale of neutrino mass and for establishing
the nature of neutrinos. We show that beta decay measurements in combination with data from the
oscillation and double beta decay experiments will allow to establish the structure of the scheme
(hierarchical or non-hierarchical), the type of the hierarchy or ordering of states (normal or inverted)
and to measure the relative CP-violating phase in the solar pair of states. 2001 Elsevier Science
B.V. All rights reserved.
PACS: 14.60.Pq; 14.60.Lm
Keywords: Neutrino masses and mixing; Beta decay
E-mail addresses: farzan@sissa.it (Y. Farzan), orlando@ifi.unicamp.br (O.L.G. Peres),

smirnov@ictp.trieste.it (A.Yu. Smirnov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 6 1 - 3
60
Y. Farzan et al. / Nuclear Physics B 612 (2001) 5997
1. Introduction
The reconstruction of the neutrino mass spectrum is one of the fundamental problems
of particle physics. The program includes the determination of the number of mass
eigenstates, and of the values of masses, mixing parameters and CP-violating phases.
At present, the evidence for non-zero neutrino mass follows from oscillation experiments
which allow to measure the mixing parameters |Uj |, the mass squared differences and, in
principle, the so-called Dirac CP-violating phases. However, the absolute values of the
neutrino masses cannot be determined. From the oscillation experiments one can only
extract a lower bound on the absolute value of neutrino mass. Obviously, for a given m2 ,
at least one of the mass eigenvalues should satisfy inequality:

mi | m2 |.
Thus, the oscillation interpretation of the atmospheric neutrino data [1] gives the bound:

m3 m2atm 0.040.07 eV.
Clearly, without knowledge of the absolute values of neutrino masses our picture of
Nature at quarklepton level will be incomplete. The knowledge of absolute values of
neutrino masses is crucial for understanding the origin of the fermion masses in general,
the quarklepton symmetry and unification. The determination of the absolute mass scale
of neutrinos is at least as important as the determination of other fundamental parameters
such as the CP-violating phases and the mixing angles. Actually, it may have even more
significant and straightforward implications for the fundamental theory. It is the absolute
mass which determines the scale of new physics.
The absolute values of masses have crucial implications for astrophysics and cosmology,
in particular, for structure formation in the Universe. In fact, the recent analysis of the latest
CMB data (including BOOMERanG, DASI, Maxima and CBI), both alone and jointly
with other cosmological data (e.g., galaxy clustering and the Lyman Alpha Forest) shows
that [2]
m < 2.2 eV,
(1)
for a single neutrino in eV range. Future observations can improve this bound. The Planck
experiment will be sensitive to neutrino masses down to m 1 eV [3]. However, the
cosmological data may not be conclusive. Even if some effects are found, it will be difficult
to identify their origin. Modification of the original spectrum of the density fluctuations
can mimic to some extent the neutrino mass effect. If no distortion is observed in the
spectrum, one can put an upper bound on the neutrino mass assuming, however, that there
is no conspiracy which leads to cancellation of different effects [4]. Therefore, independent
measurements of the neutrino mass are needed and their results will be used in the analysis
of the cosmological data as an input deduced from particle physics.
Several methods have been proposed to determine neutrino masses by using the
supernova neutrino data. One method is based on searches for the energy ordering of events
which has, however, rather low sensitivity [5]. The limits on the mass can be also obtained
61
from observations of sharp time structures in the signals. It was suggested to study the
time distribution of detected neutrino events emitted from supernova which entails to black
hole formation [6]. By this method Super-Kamiokande can measure values of the e mass
down to 1.8 eV and SNO can put an upper bound 20 eV on the and masses [6].
(Clearly, this bound on the and masses is much weaker than bounds implied by
combined analysis of the solar and atmospheric neutrino data and direct measurements of
the e mass.) In this case one can check the still non-excluded possibility in which the
solar neutrino problem is solved by the oscillations to sterile neutrino and the masses of
and are in 20 eV range. (Such neutrinos should be unstable in cosmological time.)
The absolute values of the neutrino masses can be determined in the assumption that the
cosmic rays with energies above the GZK cutoff are produced in annihilation of the ultrahigh energy neutrinos with the cosmological relic neutrinos [79]. The analysis of the
observed energy spectrum of cosmic rays above 1020 eV gives the mass m = 1.53.6 eV,
if the power-like part of the ultra-high energy cosmic rays spectrum is produced in Galactic
halo, and m = 0.120.46 eV, if this part has the extragalactic origin [10].
Neutrinoless double beta decay (20) searches are sensitive to the Majorana mass
of the electron neutrino. However, in the presence of mixing the situation can be rather
complicated: The effective Majorana mass of e relevant for the 20-decay, mee , is a
combination of mass eigenvalues and mixing parameters given by

mee =
(2)
mi Uei2 .
i
From this expression it is easy to find that if the 20-decay is discovered with the
rate which corresponds to mee , at least one of the mass eigenvalues should satisfy the
inequality [11]
mee
,
mi
(3)
n
where n is the number of neutrino mass eigenstates that mix in the electron neutrino. This
bound is based on the assumption that exchange of the light Majorana neutrinos is the only
mechanism of the 20-decay and all other possible contributions are absent or negligible.
Another uncertainty is related to n. We know only the lower bound: n 3.
The best present bound on the 20-decay obtained by HeidelbergMoscow group
gives [12]
mee < 0.34 (0.26) eV,
90% (68%) C.L.
(4)
This bound, however, does not include systematic errors related to nuclear matrix
elements. 1
A series of new experiments is planned with increasing sensitivity to mee : CUORICINO [13], CUORE (mee 0.1 eV) [14], MOON (mee 0.03 eV) [15] and GENIUS
(mee 0.002 eV) [16].
1 In what follows we will use the bound (4) in our estimations for definiteness. At the same time, one should
keep in mind that due to uncertainties of nuclear matrix element the values of mee up to 0.5 eV cannot be
excluded.
62
Although the knowledge of mee provides information on the mass spectrum independent
of m2 s, from mee one cannot infer the absolute values of neutrino masses without
additional assumptions. Since in general the mixing elements are complex there may be
a strong cancellation in the sum (2). Moreover, to induce the 20 decay, e must be a
Majorana particle.
The information about the absolute values of the masses can be extracted from kinematic
studies of reactions in which a neutrino or an anti-neutrino is involved (e.g., beta decays
or lepton capture). The most sensitive method for this purpose is the study of the electron
spectrum in the tritium decay:
3
H 3 He + e + e .
(5)
e
in (5) is described by
In absence of mixing, the energy spectrum of

1/2
dN
(6)
= R(E) (E E0 )2 m2
dE
(see, e.g., [17]), where E is the energy of electron, E0 is the total decay energy and R(E)
is a m -independent function given by
m5e
(7)
cos2 C |M|2 F (Z, E)pE(E0 E).
2 3
Here GF is the Fermi constant, p is the momentum of the electron, C is the Cabibbo
angle and M is the nuclear matrix element. F (Z, E) is a smooth function of energy which
describes the interaction of the produced electron in final state. Both M and F (Z, E)
are independent of m , and the dependence of the spectrum shape on m follows from
the phase volume factor only. The bound on neutrino mass imposed by the shape of the
spectrum is independent of whether neutrino is a Majorana or a Dirac particle.
The best present bound on the electron neutrino mass (obtained in the assumption of no
mixing) is given by Mainz tritium beta decay experiment [18]:
R(E) = G2F
me 2.2 eV
(95% C.L.).
(8)
Analysis of the Troitsk results leads to the conditional (after subtraction of the excess of
events near the end point) bound [19]
me 2.5 eV
(95% C.L.).
(9)
The present spectrometers are unable to improve the bounds (8), (9) substantially.
Further operation of Mainz experiment may allow to reduce the limit down to 2 eV. In this
connection a new experimental project, KATRIN, is under consideration with an estimated
sensitivity limit [20]
me 0.3 eV.
(10)
In the case of negative result from the KATRIN searches one can get after three years of
operation the bound me 0.35 (0.40) eV at 90% (95%) C.L. [20].
Note that with this bound KATRIN experiment can explore the range of neutrino
mass which is relevant for the Z-burst explanation of the cosmic ray with super-GZK
energies [7].
63
The aim of this paper is to study the discovery potential of the next generation tritium
beta decay experiments with sensitivity in the sub-eV range and in particular, KATRIN
experiment. We consider the effects of neutrino mass and mixing on the -decay spectrum
expected for specific neutrino schemes. We describe the three-neutrino schemes which are
elaborated to explain the data on the solar and atmospheric neutrinos as well as the fourneutrino schemes which accommodate also the LSND result. We study the bounds that
the present and forthcoming 20-decay searches, as well as the oscillation experiments
can put on possible tritium decay results. We also consider the implications of future beta
decay measurements for the identification of the neutrino mass spectrum.
The paper is organized as follows. In Section 2 we give a general description of the effect
of massive neutrinos on the beta decay spectrum in the presence of mixing. In Section 3
the three-neutrino schemes are explored. In Section 4 we present a general discussion
of predictions for the beta decay in the four-neutrino schemes which explain the LSND
result. We emphasize the importance of the bounds on the beta decay parameters imposed
by Bugey and CHOOZ experiments. In Section 5 we study the properties of the beta
decay in the hierarchical four-neutrino schemes. In Section 6, the non-hierarchical fourneutrino schemes are considered. In Section 7 we summarize the role that future beta decay
measurements will play in the reconstruction of the neutrino mass spectrum. Conclusions
are given in Section 8.
2. Neutrino mixing and beta-decay. The effects of degenerate states

In presence of mixing, the electron neutrino is a combination of mass eigenstates i with

masses mi : e = i Uei i . So that, instead of (6), the expression for the spectrum is given
by

1/2
dN
= R(E)
|Uei |2 (E0 E)2 mi 2
(E0 E mi ),
dE
(11)
where R(E) is defined in (7). The step function, (E0 E mi ), reflects the fact that a
given neutrino can be produced if the available energy is larger than its mass. According to
Eq. (11) the presence of mixing leads to distortion of the spectrum which consists of 2
(a) the kinks at the electron energy Ee(i) = E E0 mi whose sizes are determined by
|Uei |2 ;
(b) the shift of the end point to Eep = E0 m1 , where m1 is the lightest mass in the
neutrino mass spectrum. The electron energy spectrum bends at E Eep .
So, in general the effect of mixed massive neutrinos on the spectrum cannot be described
by just one parameter. In particular, for the three-neutrino scheme, five independent
parameters are involved: two mixing parameters and three masses.
Substantial simplification, however, occurs in the schemes which explain the solar and
atmospheric neutrino data and have the states with absolute values of masses in the range
2 In what follows we will use the terminology elaborated for the ideal Kurie plot without background.
64
of sensitivity (10). The simplification appears due to existence of sets of quasi-degenerate

states. Indeed, in these schemes there should be eigenstates with mass squared differences
m2 < 2 104 eV2 and m2atm 3 103 eV2 . If the neutrino masses, mi , are larger
than 0.3 eV (10), the mass differences
m
m2
2m
(12)
turn out to be smaller than 5 103 eV. Moreover, m/m m2 /2m2

1, that is,
the states are strongly degenerate. Since the detectors cannot resolve such a small mass
split, different masses will entail just to one visible kink with certain effective mass and
mixing parameter. As a consequence, the number of relevant parameters which describe
the distortion of the beta spectrum is reduced to one or three, depending on the type of the
scheme (see Sections 36).
In the Ref. [21] it has been shown that for energies Ee mi the distortion of the
electron energy spectrum in the -decay due to non-zero neutrino mass and mixing is
determined by the effective mass

2
i mi |Uei |
.
meff =
(13)
2
i |Uei |
However, the highest sensitivity to the mass of i appears in the energy range close to the
end point where E mi and therefore the approximation used in [21] to introduce meff
does not work. In what follows we show that still it is possible to use the mass parameter
(13) for a set of quasi-degenerate states.
In general, the neutrino mass spectrum can have one or more sets of quasi-degenerate
states. Let us consider one such a set which contains n states, j , j = i, i + 1, . . . , i + n 1,
with mj i
mj . We define E as the smallest energy interval that the spectrometer can
resolve. (Note that E may be smaller than the width of resolution function, and the latter
is about 1 eV in KATRIN experiment.) We assume that mij
E.
Let us introduce the coupling of this set of the states with the electron neutrino as

|Uej |2 ,
e
(14)
j
where j runs over the states in the set. We will show that the observable effect of such
a set on the beta spectrum can be described by e and the effective mass m which can
be introduced in the following way. Let us consider the interval E in the region of the
highest sensitivity to the neutrino mass, that is, the interval of the electron energies
(E0 mi E) (E0 mi ),
(15)
where mi is the mass of the lightest state in the set. The number of events in this interval,
n, is given by the integral
E
0 mi
n =
E0 mi E
dN
dE.
dE
(16)
65
We will define the effective mass m in such a way that the number of events calculated
for the approximate spectrum with single mixing parameter e and mass m , n(e , m ),
reproduces, with high precision, the number of events calculated for exact neutrino mass
and mixing spectrum n(Uej , mj ). That is,
R
n(e , m ) n(Uej , mj )
1.
n(e , m )
(17)
Expanding n (see Appendix A) in powers of mj / E

1, where
mj mj m ,
we obtain:
R

j
mj
+O
|Uej |
E
2
(18)

mj
E
2
.
It is easy to see that the first term vanishes if

2
j mj |Uej |
m =
.
e
(19)
(20)
So, for this value of m , R is of the order of ( mj / E)2 . Note that if we set m to be equal
to the value of any mass from the set or the average mass, the difference of the number of
events would be of the order of m/ E. If E is relatively small, this correction may be
significant. The expression (20) is similar to (13), but in (20) j runs over a quasi-degenerate
set (not over all the states). Moreover, provided that m
E, the approximation works
for all energies.
In reality the background should be taken into account. However it is easy to see that
if the change of the background with energy in the interval E is negligible, our analysis
will be valid in the presence of the background, too.
If the scheme contains more than one set of quasi-degenerate states with the correq
q
sponding effective masses m and mixing parameters e , the observable spectrum can
be described by the following expression
q
q 2 1/2
dN
q
(21)
= R(E)
e (E0 E)2 meff
E0 E meff ,
dE
q
where q runs over the sets. Each set of quasi-degenerate states will produce a single kink
q
q
at the electron energy E q E0 m with the size of the kink determined by e . The set
with the lightest masses leads to bending of spectrum and the shift of the end point.
3. Three-neutrino scheme
Let us consider the three-neutrino schemes which explain the solar and atmospheric
neutrino
results. In the case of mass hierarchy, m1
m2
m3 , the largest mass, m3

2
matm = (47) 102 eV, is too small to produce any observable effect in the planned
tritium decay experiments (see Eq. (10)).
66
If m3 is in the sensitivity range of KATRIN experiment (m3 0.3 eV), the mass
spectrum should be quasi-degenerate. Indeed,
m31 m2atm

0.03.
m3
2m23
Moreover, from the unitarity condition we get the coupling parameter

e =
|Uej |2 = 1.
(22)
j =1,2,3
Therefore the effect of non-zero neutrino masses and mixing on the -decay spectrum is
characterized by unique parameter the effective mass

mj |Uej |2 m3 .
m =
(23)
j =1,2,3
Correspondingly, the distortion of the -decay spectrum consists of a bending of the

spectrum and shift of the end point determined by m (E0 E0 m ), as in the case
of e with definite mass and without mixing. Let us consider the bounds on m imposed
by the 20-decay searches and the oscillation experiments. (The 20-decay in schemes
with three degenerate neutrinos has been extensively discussed before [22,23].) Assuming
that neutrinos are Majorana particles we get from (2) and (20) the relation between the
effective masses in the beta decay and the double beta decay:

mee m |Ue1 |2 + ei2 |Ue2 |2 + ei3 |Ue3 |2 ,
(24)
where 2 and 3 are the relative CP-violating phases of the contributions from the second
and the third mass eigenstates.
According to the CHOOZ bound [24] which is confirmed by the slightly weaker bound
obtained in Palo Verde experiment [25], one of the squared mixing elements (let us take
|Ue3 |2 for definiteness) must be smaller than 0.05. The other two elements are basically
determined by the mixing angle responsible for the solution of the solar neutrino
problem, so that the Eq. (24) can be rewritten as

mee = m 1 |Ue3 |2 cos2 + ei2 sin2 + ei3 |Ue3 |2 .
(25)
From this equation we find the following bounds on the beta decay mass (see also [11]):
mee
mee < m <
(26)
,
|| cos 2 |(1 |Ue3 |2 ) |Ue3 |2 |
where the upper bound corresponds to the maximal cancellation of the different terms
in (25).
The bounds (26) are shown in Fig. 1. (See also [22].) The following comments are in
order:
(1) For zero value of Ue3 the weakest bound on m from the double beta decay appears
at maximal mixing: tan2 = 1. For non-zero Ue3 the points of the weakest bound shift to
tan2 1 2|Ue3 |2 . In the vicinity of these points, the upper bound on m is given by
the present beta decay result (see Eq. (8)).
67
Fig. 1. The bounds on the effective -decay mass, m , in the 3 scheme with mass degeneracy.
Shown are the upper bounds from the 20-decay as the functions of mixing angle relevant for the
solution of the solar neutrino problem (see Eq. (26)). The upper solid (dashed) line corresponds to
the present bound mee 0.34 eV and |Ue3 |2 = 0 (|Ue3 |2 = 0.05). The lower solid (dashed) line
corresponds to mee 0.05 eV and |Ue3 |2 = 0 (|Ue3 |2 = 0.05). These lines are drawn in assumption
of strong degeneracy of neutrino masses. The vertical lines mark the 90% C.L. borders of the
LMA solution region. Shown also are the present upper bound on the neutrino mass from structure
formation [41] and the sensitivity limit of KATRIN experiment.
(2) Taking the best fit values of from the various large mixing solutions of the solar
neutrino problem [26] we find from Eq. (26) the following bounds:
0.670.74 eV LMA,
m < 1.62.2 eV LOW,
(27)
1.01.3 eV VAC,
where we have used mee 0.34 eV (4) and the two numbers in each line correspond
to |Ue3 |2 = 0 and 0.05, respectively. Note that already existing data on the 20-decay
give bounds (at the best fit points) which are stronger than the present bound from direct
measurement.
Moreover, for mee < 0.07 eV which can be achieved already by CUORE experiment
[14], the bound on m from 20-decay in the LMA preferable region of tan2 is
below the sensitivity of KATRIN experiment. Therefore, the positive result of KATRIN
68
experiment (and identification of the LMA solution of the solar neutrino problem) will
lead to exclusion of such a 3 scheme.
(3) For the SMA solution of the solar neutrino problem we get m mee , and
consequently, according to the bound (4): m 0.34 eV. So, the expected range of m
only marginally overlaps with the KATRIN sensitivity region. Thus, if the SMA solution
is identified and the LSND result is not confirmed, favoring the three-neutrino scheme, the
chance for observation of the beta spectrum distortion in KATRIN experiment is rather
small.
(4) A positive signal in the 20-decay searches will have important implications for
the tritium decay measurements:
(a) According to (26), it gives a lower bound on m independently of the solution of the
solar neutrino problem: mee m .
(b) If the values of mee , m and |Ue3 |2 are measured, we will be able to determine the
CP-violating phase 2 in (25):

1
mee 2
mee 2 mee
2
= 2
sin2
1
2|Ue3 |2
2
m
m
m
sin 2

2
1
mee
2
1
,
m
sin 2
2 ( in (24)).
where () sign of the last term reflects an uncertainty due to the phase of Ue3
3
(c) If m turns out to be smaller than mee , we will conclude that there are some additional
contributions to the 20-decay unrelated to the Majorana neutrino mass.
If future decay measurements with sensitivity (10) give a negative result, the largest
part of the allowed mass range of the 3 scheme with strong degeneracy will be excluded.
Still a small interval (m 0.10.3 eV) will be uncovered. This will have important
implications for the theory of the neutrino masses.
In Fig. 1 we show also the upper bound on m from data on the large scale structure of
the Universe obtained in [41] for values of total matter contribution to the energy density
of the Universe, m = 0.4 and the reduced Hubble constant, h = 0.8.
Without -decay measurements the absolute value of the neutrino mass can be
determined and the scheme can be identified provided that all the following conditions
are satisfied:
no effect of sterile neutrinos is observed,

the SMA solution is established as the solution of the solar neutrino problem,
the neutrinoless double beta decay gives a positive result with mee close to the present
upper bound.
In this case m = mee . However, not too much room is left for such a possibility keeping
in mind that recent solar neutrino data disfavor the SMA solution. For all large mixing
solutions of the solar neutrino problem, mee gives only the lower bound on the absolute
scale of masses.
Let us stress that, even in the case of the SMA solution one should make the assumption
that there are no additional contributions to the 20-decay apart from the exchange of
69
light Majorana neutrinos. In this scheme there are no test equalities, that is, the relations
between mee and the oscillation parameters, m2 , , which could allow to check this
assumption independently. Furthermore, the determination of mee and therefore m will
be restricted by uncertainties in the nuclear matrix elements. Thus, the study of the beta
spectrum is the only way to measure the absolute scale of neutrino mass without ambiguity.
Clearly, if the LSND result is confirmed the scheme will be excluded.
4. 4 schemes: Bugey, CHOOZ and LSND bounds

Four-neutrino schemes, which explain the LSND result in terms of oscillations, have
two sets of mass eigenstates separated by m2LSND (see Fig. 2). Hereafter, we call them
the light set of states and the heavy set of states.
Let us consider the heavy set. The masses
of states in this set are equal or larger than
m2LSND . The mass differences are equal or
Fig. 2. The four-neutrino schemes of mass and mixing. The boxes correspond to the mass eigenstates.
The position of the box in the vertical axis determines the mass. The shadowed parts of the boxes
indicate the amount of the electron flavor in the corresponding eigenstate i , that is, |Uei |2 . The solar
pair is formed by the two strongly degenerate states with m2 and significant amount of the electron
flavor. In the hierarchical schemes the effective mass of the light set is much smaller than the mass
of the heavy set. In the non-hierarchical schemes these two masses are comparable. For definiteness
we show distribution of the electron flavor in the solar pair which corresponds to the large mixing
solution of the solar neutrino problem.
70
smaller than
m2
atm
2 m2LSND
or/and
m2

.
2 m2LSND
Both splits are much smaller than the energy resolution E as well as masses themselves.
So, the states in the heavy set are quasi-degenerate and their effect on the beta spectrum
can be characterized by mh and eh given in Eqs. (14) and (20).
In the (2 + 2) schemes both the heavy and light sets contain two states, whereas in the
(3 + 1) scheme one set contains 3 states, while the other set consists of only one state (see
Fig. 2).
The e oscillation disappearance experiments, Bugey [27] and CHOOZ [24], impose a
direct and very strong bound on eh , and therefore on the expected effects in -decay in
all 4 schemes. Since Bugey and CHOOZ experiment do not resolve small mass squared
differences, m2atm and m2 , their results can be described by 2-oscillations with a
unique mass squared difference m2 m2LSND and the effective mixing parameter

2
2
2
sin 2eff = 4
|Uei | 1
|Uei | ,
i
where the sum runs over the heavy (or light) set. Using the definition of eh in Eq. (14) we
can rewrite the mixing parameter as

sin2 2eff = 4eh 1 eh .
(28)
Thus, the negative results of the oscillation searches in Bugey and CHOOZ experiments
give immediate bound on eh as a function of m2LSND (see Figs. 37).
For the range of masses relevant for LSND experiment ( m2 > 0.2 eV2 ) two
possibilities follow from (28) and the Bugey or CHOOZ bounds:
(1) Small eh :
eh < 0.027.
(29)
That is, the admixture of e in the heavy set is very small and the electron flavor is
distributed mainly in the light set. This corresponds to the schemes with normal mass
hierarchy (or to normal ordering of states in the non-hierarchical schemes). Let us recall
that in the schemes with normal hierarchy (order of states) the light set contains the pair
of states which are separated by m . This pair is responsible for the conversion of the
solar neutrinos, and for brevity, we will call it solar pair. According
to (29), in this class
of schemes one expects small kink at Ee E0 m , where m m2LSND . Here, the
inequality corresponds to the non-hierarchical case (see Section 6). Also in the case of
non-hierarchical scheme one predicts an observable shift of the end point associated to the
masses of the light set.
(2) Large eh :
1 eh < 0.027.
(30)
71
The admixture of e in the heavy set is close to one: the electron flavor is mainly distributed
in the heavy set. This corresponds to the schemes with inverted mass hierarchy or inverted
ordering of states. The effect in the beta spectrum consists of a large kink with size close
to 1 at Ee E0 m . Above Ee E0 m the spectrum has a tail related to emission
of the neutrinos from the light set. The rate of events in the tail is suppressed by factor
given in Eq. (30). If the tail is unobservable, the whole effect will look like the effect of
the electron neutrino with a unique definite mass m . (And it is similar to the effect in the
scheme with three degenerate neutrinos.)
We will further discuss these possibilities in the Sections 57.
Let us consider implications of the LSND result itself for the -decay searches. Apart
from providing the mass scale in the range of sensitivity of future -decay experiment, it
imposes an important bound on the relevant mixing parameters.
In the 4 schemes under consideration the oscillations in LSND experiment are reduced
to two neutrino oscillations with m2 m2LSND and the effective mixing parameter
2
2

sin2 2LSND = 4
(31)
Uj Uej = 4
Uj Uej ,
j h
j l
where summations run over the states of the heavy set in the first equality and of the light
set in the second equality. Using Schwartz inequality we get

sin2 2LSND 4
(32)
|Uei |2
|Uj |2 = 4eh h ,
j h
ih
or equivalently,
sin 2LSND 4
2

il
where
h
|Uj |2
|Uei |

|Uj |

= 4 1 eh 1 h ,
(33)
j l
(34)
j h
is the coupling of the heavy set with the muon neutrino. The upper bound on h (or upper
bound on 1 h , depending on the scheme) follows from CDHS experiment at high m2
[28], and from the atmospheric neutrino studies at low m2 [29]. From Eq. (32) we get a
lower bound on eh :
eh >
sin2 2LSND
,
4h
(35)
where both sin2 2LSND and h are functions of m2 . Implications of this bound for the
-decay measurements strongly depend on specific scheme, and we will discuss them in
Sections 5 and 6.
The character of the distortion of the spectrum and the sizes of the effects depend on
(1) the structure of the scheme: hierarchical or non-hierarchical;
72
(2) the type of hierarchy (ordering of levels): normal or inverted;

(3) the number of states in the heavy and light sets.
In what follows we consider possible 4 schemes in order.
5. Four-neutrino schemes with mass hierarchy

In the hierarchical schemes, the masses of the states from the light set are much smaller
than the sensitivity
limit 0.3 eV (10). In the (2
+ 2) schemes they are restricted by the

l

l
2
atmospheric m matm or solar m m2 neutrino mass scales. Therefore,
the observable distortion of the spectrum is only due to effect of the heavy set with the
effective mass

m m2LSND.
(36)
As we have mentioned in the previous section, the character of distortion of the beta
decay spectrum depends, first of all, on the type of hierarchy.
5.1. Schemes with normal mass hierarchy
In the schemes with the normal hierarchy (both in the (2 + 2) and (3 + 1) cases) the
electron flavor is in the light set and eh is strongly restricted by the Bugey result (see
h
Eq. (29) and Figs.
3, 4). The beta decay spectrum has only a small kink with e < 0.027
at E = E0 m2LSND .
Additional restrictions on the -decay parameters may appear depending on whether the
scheme is of the (2 + 2) or (3 + 1) type.
(1) The (3 + 1) scheme. In this scheme one gets a substantial lower bound on eh from
the LSND result (see Eqs. (32), (35)). Indeed, in this scheme h is restricted from above
by the CDHS result [28]. Inserting the bound on h into Eq. (35) we get the lower bound

on eh which is close to the upper Bugey bound or, for certain ranges of m2LSND , even
above it. So that only certain ranges of m are allowed (see Fig. 3). This is a manifestation
of the fact that in the (3 + 1) scheme an explanation of the LSND result requires the e
admixture in the isolated state to be at the level of the upper Bugey bound [30].
Let us consider implications of the 20-decay search. The contribution to mee from the
fourth (isolated) state dominates. It can be estimated as:

m(4)
(37)
=
m2LSND |Ue4 |2 (0.0050.05) eV.
ee

In the hierarchical case with m2 = m2 , the contributions from other mass

eigenstates can be estimated as m(3)
m2atm |Ue3 |2 < 3.5 103 eV and m(2)
ee =
ee

2
2
3
m sin < 7 10 eV. Hence
h
mee m(4)
ee = m e ,
(38)
73
Fig. 3. The bounds on m and eh in the (3+1) scheme with normal mass hierarchy. The dashed curve
and solid lines attached to it (from below and above) show the upper bound on eh from Bugey and
CHOOZ experiments, respectively. The LSND lower bound (see Eq. (35)) is shown by dot-dashed
curves. The allowed regions are shadowed. The triplets of solid lines show the upper bounds on m
assuming that future 20-decay searches will give mee 0.01, 0.03 and 0.05 eV. The central line
in each triplet corresponds to the contribution from the heaviest mass eigenstate (Eq. (38)) and the
other two lines show the uncertainty due to the contribution of light states in the modification of the
scheme in which the mass hierarchy of the three light states is inverted (see Eq. (39)).
or m = mee /eh .
A version of the scheme is possible in which the mass hierarchy in the light set is
inverted,
so that the states which contain the electron flavor have masses m2 m3

m2atm and m1
m2 . In this case we have

mee m4 |Ue4 |2 ei + m2atm cos2 ei + sin2 ,
(39)
and the contribution from the light set can be comparable to the contribution from the 4th
state. The corresponding lines are shown in Fig. 3. According to the figure one expects
mee to be substantially below the present bound: We find mee 0.005 eV, 0.015 eV,
0.0150.03 eV and 0.06 eV for the allowed islands of m and eh (from smallest
to largest m ). Clearly, the observation of mee near its present experimental bound will
exclude the scheme.
For the SMA solution of the solar neutrino problem, Eq. (39) reduces to
74
Fig. 4. The bounds on m and eh in the (2+2) scheme with normal mass hierarchy. The dashed curve
and solid lines attached to it show the upper bounds on eh from Bugey and CHOOZ experiments,
respectively. The LSND lower bound (see Eq. (35)) is shown by dash-dotted line. The allowed
regions of parameters are shadowed. The triplets of solid lines show the lower bounds on m from a
positive signal in future 20-decay searches which would correspond to mee = 0.01, 0.03 and 0.05
eV. The central lines correspond to contribution to mee from the heavy set only.

mee m2LSNDeh ei + m2atm .
In principle, using this equation one can determine the relative phase .
For large angle solutions of the solar neutrino problem still significant contributions can
come from the solar pair of states and therefore two different phases ( and ) are involved
in the determination of mee (see (39)).
(2) The (2 + 2) scheme. In the (2 + 2) scheme with normal mass hierarchy the mass
difference of the heavy set is given by m2atm , and is distributed, mostly, in the heavy
set. According to the CHDS bound, h should be close to 1 in this scheme. Hence, the
LSND bound is less restrictive:
eh >
1 2
sin 2LSND,
4
(40)
and eh can be as small as (23) 104 (see Fig. 4). The effect of the kink is unobservable
for such a small eh .
75
Let us consider bounds from the 20-decay searches. The Majorana mass of the
electron neutrino can be written in the following form

m2

2

2
mee m Ue3 + Ue4
(41)
sin2 1 |Ue3 |2 |Ue4 |2 .
+
m

Neglecting the contribution of light neutrinos m2 sin2 < 7 103 eV , we obtain
from (41) the following bound on the effective -decay mass:
mee
mee
.
< m <
(42)
h
2
e
||Ue3 | |Ue4 |2 |
If the neutrinoless double beta decay is discovered, this inequality will put a strong lower
bound on m :
m > 25mee ,
(43)
where we have used the bound on eh from the Bugey experiment. For mee = 0.1 eV we
get m = 2.5 eV which is in the upper allowed region of the LSND experiment. For this
reason, one cannot expect mee > 0.1 eV in this scheme.
5.2. Schemes with inverted mass hierarchy
In the schemes with inverted hierarchy, the electron flavor is distributed mainly in the
heavy set (see Fig. 2). Therefore
the -decay spectrum should have a large kink with
parameters eh 1 and mh
m2LSND . The light set leads to appearance of the tail at
Ee > E0 mh with the suppressed rate determined by (1 eh ) < 0.027. To detect the
signal from the tail one may search for the integral effect above E0 m4 .
Note that for the LSND region with m2LSND = 68 eV2 allowed at 99% C.L. one gets

m > m2LSND 2.5 eV which is already excluded by the Mainz result at 95% C.L. (see
Fig. 5). Hence, for the schemes with inverted hierarchy the preferable range of mass would
be
0.4 eV < m < 1.75 eV.
(44)
(1) The (3 + 1) scheme. The upper Bugey bound and the lower LSND bound on 1 eh
are shown in Fig. 5. They are similar to the bounds on eh in schemes with normal mass
hierarchy.
Let us consider the implications of the 20-decay searches. The contribution of the
light states to mee is negligible. As a consequence, the bounds imposed by the 20-decay
searches are similar to the bounds in the three-neutrino scheme, and the only difference is
that
in the latter scheme, m is a free parameter, whereas in the (3 + 1) scheme it equals to
m2LSND .

For the SMA solution of the solar neutrino problem we have mee m2LSND > 0.4 eV
which is already larger than the present upper bound. So, keeping in mind the uncertainties
of the nuclear matrix element, we can say that this possibility is disfavored.
76
Fig. 5. The bounds on m and eh in the (3 + 1) scheme with inverted mass hierarchy. The dashed
curve and solid lines attached to it (from below and above) show the upper bound on eh deduced
by Bugey and CHOOZ experiments, respectively. The LSND lower bound (see Eq. (35)) is shown
by dot-dashed lines. The allowed regions are shadowed. The horizontal dashed lines are the upper
bounds on m from the 20-decay searches which correspond to mee < 0.34 eV, Ue3 = 0, and
different values of sin2 2 .
In Fig. 5, we show the upper bounds on m (m < mmax

ee / cos 2 ) which correspond to
=
0.34
eV.
The
bounds
are
shown
for
different values of sin2 2 from
|Ue3 | = 0 and mmax
ee
the large mixing solution regions. Note that for the LMA solution the 99% upper bound,
sin2 2 0.95, gives m < 1.5 eV, which is already below the present kinematic bound.
The best fit region sin2 2 (0.60.8) leads to m (0.550.75) eV well in the range of
the KATRIN sensitivity. For the LOW solution the maximal mixing is possible for which
cos 2 = 0, and the whole region of m up to the present kinematic bound (8) is accepted.
(2) The (2 + 2) scheme. Here the effects are similar to those in the (3 + 1) scheme with
two differences:
(i) As it is shown in Fig. 6, the LSND result is less restrictive and the rate in the tail can
be substantially lower.
(ii) The bound on m from the neutrinoless double beta decay coincides with the bound
in the (3 + 1) scheme at |Ue3 | = 0:
mee
mee < m <
(45)
.
cos 2
77
Fig. 6. The bounds on m and eh in the (2 + 2) scheme with inverted mass hierarchy. The dashed
curve and solid lines attached to it (from below and above) show the upper bound on eh deduced
by Bugey and CHOOZ experiments, respectively. The LSND lower bound (see Eq. (35)) is shown
by dot-dashed lines. The allowed regions are shadowed. The horizontal dashed lines are the upper
bounds on m from the 20 searches which correspond to mee < 0.34 eV and different values of
sin2 2 . We assume that solar pair of states gives the dominant contribution (see Eq. (45)). The curve
for the SMA solution (not shown here) practically coincides with the curve of KATRIN sensitivity.
6. Non-hierarchical schemes
Let us consider schemes in which the masses of states from the light set are also in
the range of sensitivity of KATRIN experiment: m1 > 0.3 eV. Clearly, these states are
quasi-degenerate, and their effect on the -decay spectrum can be characterized by the
effective mass, ml , and the total coupling with the electron neutrino, el . From the unitarity
condition we have el + eh = 1.
In the non-hierarchical schemes, the effect of neutrino mass on the -decay spectrum
consists of the kink and the shift of the end point. The kink is at
Eekink = E0 mh ,
where
mh =
m2LSND + (ml )2 ,
(46)
(47)
78
and its size is determined

by eh . In contrast to the hierarchical case, the mass of the heavy
set is not fixed by m2LSND. The tail above the kink, Ee > Eekink , is described by el and
the end point is shifted to
ep
Ee = E0 ml .
(48)
Note that if ml 0.5 eV, from (47) we obtain the effective mass of the heavy set mh
2.5, 1.5 and 0.8 eV for m2LSND = 6, 1.5, and 0.4 eV2 correspondingly. For large values
of m2LSND both the structures (the kink and the bending of the spectrum at the end point)
can in principle be separately detected by KATRIN. For m2LSND at the lower allowed end
the difference of the two effective masses becomes small: mh ml < 0.3 eV, and these
two structures may not be resolved.
With increase of ml the scheme transforms into the scheme with four degenerate
neutrinos. Already at ml = 1 eV, we get the mass of the heavy set mh = 2.7, 1.5, and
1.2 eV, for m2LSND = 6, 1.5, and 0.4 eV2 . For the smallest mass, mh = 1.2 eV, the
difference of masses, mh ml < 0.2 eV, is too small to be resolved. With the increase of
ml the two structures in the spectrum, the kink and the bending, merge.
The parameters of the kink and of the tail depend on specific properties of the scheme.
We will call it the scheme with normal ordering of states when the electron neutrino is
distributed mainly in the light set. And we will refer to the opposite situation, when e is
in the heavy set, as to the scheme with inverted order of states.
In the non-hierarchical schemes the mass spectrum is shifted to larger values of mass.
The oscillation pattern, however, is not changed and the oscillation bounds are the same as
in the hierarchical schemes described above (see Figs. 3, 6). However, the implications of
the 20-decay searches are changed.
6.1. Schemes with normal order of states
The electron neutrino is mainly in the light set. The beta decay spectrum should have a
small kink at E0 mh with the size restricted by Bugey experiment: eh < 0.027, and a
strong bending of spectrum at E0 ml .
In Fig. 7 we show the bounds on the -decay parameters ml and eh for two
representative values of the mass squared difference: m2LSND = 1.75 eV2 , which
corresponds to the weakest bound on eh from Bugey experiment, and m2LSND =
0.227 eV2 from the lowest (in m2 scale) LSND region.
For m2LSND = 1.75 eV2 the allowed region of the -decay parameters is between
the vertical lines at eh = 0.013 (dashed) and eh = 0.026 (solid) for the (3 + 1) scheme
(shadowed), while for the (2 + 2) scheme the valid region stays between eh = 0.004 (dashdotted) and eh = 0.026 (solid).
For m2LSND = 0.227 eV2 the allowed regions are substantially smaller: for the (3 + 1)
scheme the region is between lines at eh = 0.0095 (dashed) and 0.010 (solid). For the
(3 + 1) scheme the region (shadowed) is restricted by lines at 0.0002 (dash-dotted) and
0.010 (solid). All the regions are bounded from above by the Mainz result.
79
Fig. 7. The bounds on the effective mass of the light set ml , and the coupling of the electron neutrino
with heavy set, eh , in the non-hierarchical schemes with normal order of levels. The vertical solid
lines show the upper bounds on eh from Bugey experiment. The dashed and dash-dotted vertical
lines show lower bounds on eh from LSND experiment in the (3 + 1) and (2 + 2) schemes,
respectively (see Eq. (35)). The allowed regions for (3 + 1) scheme are shadowed. The lines with
different values of sin2 2 are the upper bounds from the 20-decay searches which correspond
to mee < 0.34 eV. The line denoted by 3 + 1 shows the upper bound for the (3+1) scheme with
|Ue3 |2 = 0.05 (see Eq. (51)), while the others are valid both for the (2 + 2) scheme and the (3 + 1)
scheme
with |Ue3 |2 = 0 (see Eqs. (53), (51)). The lines marked by (a) and (b) are calculated for

m2LSND = 1.32 eV and 0.477 eV, respectively.
The implications of the 20-decay searches depend on the specific arrangements of

levels.
(1) The (3 + 1) scheme. The contribution to the effective Majorana mass, mee , from
different mass eigenstates can be evaluated in the following way. The solar pair of states
(e is mainly in these states) yields the contribution:

2

l
2
2
m(1+2)
(49)
cos + ei sin2 .
msun
ee
ee = m 1 |Ue3 | |Ue4 |
The contributions from the two other states are
l 2
m(3)
ee = m Ue3 ,
h 2
m(4)
ee = m Ue4 .
(50)
In general these contributions may have arbitrary relative phases and cancel each other in
the sum. The contribution from the third level, which belongs to the light set, is restricted
80
l
by the CHOOZ bound: m(3)
ee < 0.05 eV for m < 1 eV. In turn, the fourth contribution
(from the isolated level) is restricted by the Bugey result: m(4)

ee < 0.05 eV. The contribution
of the solar pair (49) depends on the solution of the solar neutrino problem and, in most
of the cases, dominates over other contributions. Indeed, for the SMA solution we get
l
msun
ee m > 0.3 eV (if masses of the light set are in the range of sensitivity of KATRIN
experiment). For the LMA solution the cancellation of the two terms in (49) may occur but
l
typically, msun
ee (0.2 1)m > 0.1 eV is large enough to be detected in the forthcoming
20-decay experiments. For other solutions with large mixing angle (LOW, VO) the
cancellation can be stronger. The solar pair contribution can be comparable with two others
if the mixing is close to maximal: sin2 2 > 0.98 and the two terms in (49) have opposite
signs.
So, in general we expect the effective mass of the Majorana neutrino to be mee 0.1 eV,
that is, not too far from the present experimental bound. The identification of the solution
of the solar neutrino problem can clarify a situation leading to more definite predictions.
If the solar pair gives the dominant contribution, the measurements of mee , ml and
will allow to determine the relative phase of masses in the solar pair, (see Eq. (49)).
Otherwise, due to the presence of three different phases we cannot determine values of
these phases, and mee will give only a lower bound on the mass scale.
Assuming the maximal cancellation of contributions in mee we find from (49) and (50),
an implicit upper bound on ml as a function of eh :

l
m 1 eh | cos 2 | (ml )2 + m2LSND eh

ml 1 | cos 2 | |Ue3 |2 < mee .
(51)
We show this bound on m for different values of sin2 2 in the Fig. 7. Note that for
sin2 0.95, the effect of non-zero |Ue3 |2 (the last term in the left-hand side of Eq. (51))
by 3 + 1 shows
is non-negligible but it decreases with sin2 2 . In Fig. 7, the line marked
the 20-decay bound for the (3 + 1) scheme at |Ue3 |2 = 0.05 and m2LSND = 0.477 eV.
The lower line in the pair marked by 2 + 2 is the corresponding bound for the (3 + 1)
scheme at |Ue3 |2 = 0. At |Ue3 |2 = 0 the bounds for the (2 + 2) and (3 + 1) schemes
coincide (see below). The pair of lines marked
by 2 + 2 illustrates dependence of bound
on m2LSND . The upper line (a) is for
m2LSND = 1.32 eV while the lower one is for
0.477
eV. For smaller values of sin2 2 , the lines have been calculated at |Ue3 |2 = 0 and

m2LSND = 1.32 eV.
(2) The (2 + 2) scheme. Now the heavy set contains two states, 3 and 4 , and the solar
pair is in the light set. The contribution to the Majorana mass, mee , from the light set is
described by Eq. (49). The heavy set gives

m(3+4)
(52)
= mh |Ue3 |2 + ei |Ue4 |2 < mh eh ,
ee
(3+4)
< 0.04 eV for mh < 0.1 eV. Again,

which is restricted by the Bugey bound: mee
the solar pair gives the dominating contribution unless the solar mixing is very close to
maximal and the phase in (49) is close to .
81
Assuming the maximal cancellation of contributions in mee we find from (52) and (49),
an implicit upper bound on ml as a function of eh :

l
(53)
m 1 eh | cos 2 | (ml )2 + m2LSND eh < mee .
We show this bound on m for different values of sin2 2
in the Fig. 7. The pair of lines
marked by 2 + 2, corresponds to two different values of m2LSND : 1.32 eV (upper line)

and 0.477 eV (lower line). For other values of sin2 2 , m2LSND is taken to be 1.32 eV.
Note that for the (3 + 1) scheme with |Ue3 |2 = 0, the bounds from the 20-decay searches
are the same as for the (2 + 2) scheme.
6.2. Schemes with inverted order of states
The electron neutrino is mainly distributed in the heavy set, so that one expects a large
kink at E0 mh with the size close to 1 and the suppressed tail with the end point at
E0 ml . In the tail the rate is restricted by the Bugey bound: el < 0.027. As for the
case of the spectrum in the schemes with inverted mass hierarchy, we can conclude that
m2LSND < 3 eV2 .
In Fig. 8, we show the bounds on the relevant beta decay parameters: mh , the effective
mass of the heavy set, and (1 eh ) which determines small admixture of the electron
neutrino in the light set. Clearly,

ml = (mh )2 m2LSND
(54)
and
mh
m2LSND.
We show the bounds for two representative values of
(55)
m2LSND : 1.32 and 0.447 eV (as
for the scheme with normal ordering). The allowed ranges for (1 eh ) determined by
the oscillation experiments are the same as the ranges of eh for normal ordering (Fig. 8).
The allowed regions for the (3 + 1) scheme are shadowed. The allowed values of mh are
restricted by the Mainz limit from above and by the LSND result from below (see Eq. (55)).
Let us consider the bounds from the 20-decay searches. The effective Majorana mass
can be immediately obtained from the results for the scheme with normal ordering by the
interchange: mh ml . This leads to a stronger dominance of the solar pair contribution
to the effective Majorana mass.
(1) The (2 + 2) scheme. The contribution from the heavy set is given by

2

h
2
2
cos + ei sin2 ,
mhee msun
(56)
ee = m 1 |Ue3 | |Ue4 |
whereas the contribution from the light states can be written as
2

2
+ ei Ue2
mlee = ml Ue1
< ml el ,
(57)
82
Fig. 8. The bounds on the effective mass of the heavy set, mh , and the coupling of the electron
neutrino with light set, 1 eh , in the non-hierarchical schemes with inverted order of states. The
vertical solid lines show the upper bounds on (1 eh ) from Bugey experiment. The dashed and
dash-dotted vertical lines show lower bounds on (1 eh ) from LSND experiment in the (3 + 1)
and (2 + 2) schemes, respectively (see Eq. (35)). The allowed regions for the (3 + 1) scheme are
shadowed. The lines with different values of sin2 2 are the upper bounds from the 20-decay
searches which correspond to mee < 0.34 eV. The line denoted by 3 + 1 corresponds to the (3 + 1)
scheme with |Ue3 |2 = 0.05 (see Eq. (62)) while others are valid both for the (2 + 2) scheme and the
(3 +1) scheme with |Ue3 |2 = 0 (see Eqs. (53), (59)). The lines marked by (a) and (b) are calculated
for
m2LSND = 1.32 eV and 0.477 eV, respectively.
and according to the Bugey result: el < 0.027. Implications of the 20 searches are
similar to those in the hierarchical case. However, now mh , and consequently mee , can

be even larger than m2LSND .
From Eqs. (56), (57) we find the lower bound on mee :

mee > mh 1 el | cos 2 | ml el .
(58)
Using this inequality and the upper experimental bound on mee , we get an implicit upper
bound on mh as a function of eh :
mh eh | cos 2 |

(mh )2 m2LSND 1 eh mmax
ee = 0.34 eV.
(59)
83
The bounds for different values of the solar mixing parameter are shown in Fig. 8. The
identification of the solution of the solar neutrino problem and measurements of sin2 2
as well as mild improvement of the bound on the Majorana mass will have strong impact on
this scheme. For instance, as follows from the Fig. 8, the possible bounds: sin2 2 < 0.9
and mee < 0.1 eV would exclude whole the region of parameters of the scheme down to
the KATRIN sensitivity limit.
(2) The (3 + 1) scheme. The contributions to mee from the heavy and the light sets are
equal to:

2
mhee = mh 1 |Ue3 |2 |Ue4 |2 cos2 + ei sin2 + Ue3
(60)
and

mlee = ml 1 eh ,
(61)
respectively. In (60) |Ue3 |2 is restricted by the CHOOZ results and mlee is restricted by
the Bugey results: mlee 0.03 eV taking ml 1 eV. The contribution from the heavy
set is similar to the one in the scheme with three degenerate neutrinos. For the largest
part of the allowed parameter space the contribution of the solar pair to mee dominates.
Still significant cancellation is not excluded which can cause the two contributions in (60)
(from the solar pair and the third mass eigenstate) to be comparable. Note that when |Ue3 |2
is smaller than 0.0150.05, mlee can be as large as mh |Ue3 |2 .
For the SMA solution we have mee > 0.6 eV for ml > 0.3 eV, so that such a possibility
is excluded by the present bound (2).
Assuming the maximal cancellation of contributions in mee , we find from (60) and (61)
an implicit upper bound on mh as a function of eh :

mh eh | cos 2 | (mh )2 m2LSND 1 eh

mh 1 | cos 2 | |Ue3 |2 < mee .
(62)
We show this bound on mh for different values of sin2 2 in the Fig. 8. Note that for
sin2 2 = 0.95, the effect of non-zero |Ue3 |2 is non-negligible but for smaller values of
sin2 2 we can neglect |Ue3 |2 . In the Fig. 8, for sin2 2 = 0.95, we have taken |Ue3 |2 =
0.05 and for other values of sin2 2 we have set |Ue3 |2 equal to zero.
6.3. 4 schemes without LSND
Apart from the LSND result, there is a number of other motivations to introduce new
neutrino mass eigenstates. In particular, the sterile neutrino in the eV-range has been
discussed in connection to the supernova nucleosynthesis (r-processes) [31]. The mixing
of the keV-mass sterile neutrino with the active neutrinos can provide a mechanism of the
pulsar kicks [32]. The keV sterile neutrinos may compose the warm dark matter of the
Universe [33]. Small mixing of the sterile neutrino with the active neutrinos can induce the
large mixing among the active neutrinos [34].
84
Light (SU(2) U (1)) singlet fermions (sterile neutrinos) can originate from some
new sectors of the theory beyond the standard model. Neutrinos, due to their neutrality
are unique particles which can mix with these fermions. So, searches for the effects of the
sterile neutrinos are of fundamental importance even if these fermions do not solve directly
any known problem and thus their existence is not explicitly motivated.
In this connection we will consider a general four-neutrino scheme in which three
(dominantly active) neutrinos are light and the fourth neutrino (dominantly sterile) has
a mass in the eVkeV range. The three light neutrinos may form a hierarchical structure
with the heaviest component being below 0.07 eV, so that their masses will not show up in
the planning beta decay experiments. The fourth neutrino has a small mixing with active
neutrinos, and in particular, with the electron neutrino.
The scheme is similar to the (3 + 1) schemes with normal mass hierarchy. The difference
is that now the mass m4 and the mass squared differences m24 m2i (i = 1, 2, 3) are not
restricted by the LSND result so that m4 can be larger or much larger than 2.5 eV.
Let us consider the possible effect of this fourth neutrino in the beta decay. We
concentrate on the range of masses (0.55) eV which satisfy the cosmological bounds.
The fourth state produces the kink in the beta decay spectrum at E = E0 m4 with the
size
e = |Ue4 |2 .
(63)
Let us evaluate the allowed range of e for different values of m4 . At m4 > 1.5 eV
the strongest bound follows from the CHOOZ result: e < 0.027 (we assume that
other neutrinos are much lighter, but generalization to the non-hierarchical case is
straightforward). This bound does not depend on the mass for m > 1.6 eV.
Another direct restriction comes from the 20-decay searches provided that neutrino
is the Majorana particle:
m4 <
mmax
ee
,
e
(64)
where mmax
ee is the upper bound on the effective Majorana mass of the electron neutrino.
Taking the present bound mmax
ee = 0.34 eV and the maximal allowed value of e we find that
m4 < 12.6 eV. Thus, one may see the kink of the 3% size (or smaller) in the energy interval
(013) eV below the end point. The recent cosmological bound (1) shrinks substantially
this interval.
Additional restrictions on the possible effects appear if there is a substantial admixture
of the muon flavor in the fourth state. In this case one predicts the existence of the e
oscillations with the effective mixing parameter
sin2 2e = 4|Ue4 |2 |U4 |2
(65)
and m2 m24 = (1100) eV2 . For m2 > 7 eV2 the stronger bound, sin2 2e < 1.3
103 , is given by KARMEN experiment [35]. Clearly, these bounds are satisfied, if |U4 |2
is small enough. However, if |U4 |2 |Ue4 |2 (which might be rather natural assumption)
the bound from KARMEN experiment becomes important. Taking |U4 |2 = |Ue4 |2 we get
from (65)
e <

1
sin2 2e ,
2
85
(66)
and for m24 > 7 eV2 , it follows from (66) that e < 1.7 102 which is stronger than
the Bugey bound. For heavy neutrinos (in keV range) the neutrinoless double beta decay
gives very strong bound on the size of the kink: e < 3 104 (1 keV/m4 ) which will
be very difficult to observe. This bound does not exist if the keV neutrino is the Dirac (or
pseudo-Dirac) particle.
7. Beta decay measurements and the neutrino mass spectrum

In this section we consider the astrophysical and cosmological bounds on the neutrino
mass. We will discuss possible future developments in the field.
7.1. Beta decay measurements and supernova neutrinos
Studies of the supernova neutrinos open unique possibility to test the schemes of
neutrino mass and mixing [45]. Therefore they may have an important impact on
predictions for future beta decay measurements.
Considering the level-crossing patterns [45] and the adiabaticity conditions in various
resonances it is easy to show that in all the (2 + 2) schemes with inverted mass hierarchy
(or ordering of the states) the originally produced e -flux is almost completely converted to
some combination of , and s -fluxes at high densities in the resonance associated to
m2LSND . The mixing parameter in this resonance, given by sin2 2LSND , is large enough to
guarantee the adiabaticity of the conversion. In this case the e survival probability equals
to P sin2 LSND < 102 . At the same time, the e -flux observed at the Earth appears as
a result of conversion
, e
(67)
at high densities inside the star. Therefore the spectrum at the Earth will practically
coincide with the hard original spectrum of and :
Fe (E) F0 (E),
(68)
and moreover, this result does not depend on the solution of the solar neutrino problem.
Such a hard spectrum of e is strongly disfavored by the SN1987A data [46]. Future
detection of the Galactic supernova can exclude the conversion (67), and consequently
the schemes will be excluded, completely.
In the (3 + 1) scheme with inverted mass hierarchy (ordering) the result of conversion
depends on the solution of the solar neutrino problem [30]. As in the (2 + 2) scheme, the
original e -flux is converted to a combination of and s -fluxes. For the SMA solution
no opposite conversion (that is, and to e ) occurs. Therefore, in this scheme the
transitions lead to practically complete disappearance of the e -flux. The suppression
86
factor, sin2 LSND < 102 , cannot be compensated by the allowed increase of the original
flux. The disappearance of e contradicts the data from SN1987A, so that the scheme is
excluded. Notice that this scheme is also practically excluded by the present bound from
the neutrinoless double beta decay.
In the case of the LMA solution some part of the original - and -fluxes will be
transformed to the e -flux at low densities, so that at the surface of the Earth one expects:
Fe (E) sin2 F0 (E).
(69)
Thus, e will have hard spectrum suppressed by factor 1/31/2. This is again disfavored
by the SN1987A data.
Notice that in all these schemes the electron neutrino flux is not eliminated from the
region proposed for the r-processes [31]. So that the mechanism of production of the heavy
elements will not work.
In contrast, the non-hierarchical 4 schemes with normal ordering are well consistent
with the SN1987A data and they predict an observable effect in the -decay spectrum, as
was discussed in Section 6.
7.2. Beta decay and forthcoming experiments
The results of the forthcoming oscillation as well as non-oscillation experiments can
substantially influence the both predictions of the effects of neutrino mass and mixing
in the beta decay spectrum and the significance of future beta decay measurements. In
particular,
(i) the identification of the solution of the -problem and measurements of relevant
oscillation parameters,
(ii) the MiniBooNE result,
(iii) further searches for the neutrinoless double beta decay will have crucial impact.
Also further improvements of the bound on |Ue3 | will be important. Cosmology can give a
hint for the absolute scale of neutrino mass.
Let us analyze consequences of possible results from these experiments.
(1) The solution of the solar neutrino problem can be identified in the forthcoming
experiments: SNO [36,37], KamLAND [38], BOREXINO [39]. The identification will
not influence the predictions for the -decay immediately. Indeed, from these experiments
we will get specific values (ranges) of m2 and mixing angle , i.e., the distribution
of the electron flavor in the solar pair of states will be determined. But the effects in the
-decay are not sensitive to a particular value of m2 , since for all possible solutions
m2 cannot be resolved in -decay searches. Also the effects are not sensitive to the
distribution of the electron flavor since they are determined by the sum over states in the
solar pair: |Ue1 |2 + |Ue2 |2 1. However, the identification of the solution of the solar
neutrino problem will influence substantially the bounds on the -decay effects from the
20-decay searches. A number of schemes discussed here will be excluded and for other
schemes the possible effects in the -decay spectrum will be strongly restricted.
87
The key issues are whether the correct solution of the solar neutrino problem is the
small mixing solution or the large mixing solution, and if it is the large mixing how large
the deviation from maximal mixing is.
Suppose that the SMA solution will be identified, then the following information can
be obtained from the 20-decay searches in the assumption that the Majorana neutrino
exchange is the only mechanism of the decay:
For the 3 scheme this will imply that m < 0.34 eV and further moderate
improvement of the 20-decay
bound will exclude the scheme.
According to present data: m2LSND > 0.38 eV (at 99% C.L.) [40]. Therefore, in
the
schemes with inverted mass hierarchy or inverted order of states we get mee
m2LSND > 0.38 eV. On the other hand, the bound from the 20-decay is mee <
0.34 eV (at 90% C.L.). Therefore, these schemes are excluded at stronger than 90%
C.L. One should, however, keep in mind the uncertainties of the nuclear matrix
elements. Future double beta decay measurements will be able to confirm and improve
the bound. Similar conclusion can be made for non-hierarchical schemes with normal
ordering of states and ml > 0.34 eV.
Thus, the schemes which will survive after the identification of the SMA solution,
are the schemes with normal hierarchy as well as the non-hierarchical schemes with
normal order of states and ml < 0.34 eV.
If mee turns out to be close to the present bound (e.g., 0.2 eV) and the LSND result
is confirmed, the only possibility will be the non-hierarchical scheme with normal
ordering of states and ml = mee .
Suppose now that one of the large mixing solutions of the solar neutrino problem will
be identified. In this case a possible cancellation between various contributions to mee will
relax the bounds from the 20-decay.

The important point is that the present upper bound on mee is already smaller than
m2LSND :
mmax
ee

<
m2LSND
(although, one should keep

in mind the uncertainties of the nuclear matrix element).
At the same time, mh > m2LSND . Consequently, the schemes with inverted mass
hierarchy or inverted ordering of states require cancellation between the contributions
from the solar pair states, and therefore, there should be CP-violating phase difference
in the corresponding mass eigenvalues. The scheme with the SMA solution of the
solar neutrino problem is practically excluded.
The upper bound on sin2 2 (lower bound on the deviation from maximal mixing)
restricts a possible cancellation of contributions to mee . This, in turn, gives an upper
bound on the mass mh . Therefore, further improvements of the bounds on mee and
sin2 2 can strongly restrict the parameter space of the schemes and even exclude
them.
88
Similar conclusions hold for the non-hierarchical schemes with ml > mmax
ee
0.34 eV.
As we discussed in Section 5, the Majorana mass mee gives the lower bound on the
mass of the solar pair. So, if mee turns out to be close to the present bound (4), we can
exclude the schemes with normal hierarchy (in which mee < ml < 0.07 eV).
If mee is much smaller than the present bound (e.g., 0.01 eV) the scheme should
have the normal mass hierarchy or should lead to a strong cancellation of contributions
to the 20-decay.
(2) MiniBooNE experiment will give strong discrimination among the possibilities.
If MiniBooNE does not confirm the LSND result, a large class of schemes discussed
here will be excluded. There will be no strong motivation to consider the 4 schemes (see,
however, Section 6.3).
Also further improvements of bounds on the involvement of a sterile neutrino in the
solar and in the atmospheric neutrino conversions may give independent confirmation of
the 3 schemes.
We will be left with the three-neutrino scheme with strong degeneracy or with schemes
having more than three mass eigenstates without observable signal in MiniBooNE (see
Section 6).
So, in this case the searches of the neutrino mass in the sub-eV range will basically test
the 3 scheme with strong degeneracy. The observation of the shift of end point to E0 m
in KATRIN experiment will be the proof of the scheme with m1 m2 m3 m .
Further insight can be obtained confronting the results of the -decay and the 20
decay measurements, as we have discussed in Section 3.
If MiniBooNE confirms the LSND result we will be forced to consider the 4 schemes.
In this case we get for the absolute value of the mass of heavy set:

m m2LSND,
(70)
where the equality corresponds to the scheme with the mass hierarchy (mh ml ).
MiniBooNE experiment will not only check the LSND result but also further restrict
the oscillation parameters. Moreover, it may allow to disentangle the (3 + 1) and (2 + 2)
schemes. Further searches of the sterile neutrinos in the solar and atmospheric neutrino
fluxes should discriminate the (2 + 2) and (3 + 1) schemes [30].
(3) Let us consider implications of future cosmological measurements. The present and
future cosmological bounds on the mass scale of neutrinos are summarized in the Table 1.
The bound on m taken from [41] corresponds to the energy density of matter m = 0.4
and the reduced Hubble constant h = 0.8.
Note that by chance the best cosmological bound (1) coincides numerically with the best
laboratory limit (8). However, in contrast to the latter, the cosmological bound is valid for
any flavor including the sterile neutrino, provided that this neutrino had been equilibrated
in the Early Universe.
The bound (1) (obtained for one neutrino in the eV range) applies immediately to the
(3 + 1) scheme with normal hierarchy. In this scheme the heaviest (isolated) state can
89
Table 1
Cosmological bounds on the neutrino mass scale. All the four-neutrino schemes are considered to be
hierarchical
Mass scheme
Present m , eV
Ref.
Future m , eV
Ref.
3 or (3 + 1)
inverted
1.8
2.5
[41]
[44]
0.41
[42]
(2 + 2)
inverted or normal
3.0
3.8
[43]
[44]
0.57
[42]
(3 + 1) normal
5.5
7.6
2.2
[43]
[44]
[2]
0.99
[42]
produce only a small kink in the -decay spectrum whichwill be difficult to detect. So
the cosmological bound being confronted with the value of m2LSND will play important
role in checking the scheme. The bound is even stronger for the non-hierarchical (3 + 1)
schemes.
The bound is also immediately applied to the schemes with additional heavy neutrino
(Section 6.3). Even for very small admixture of the electron neutrino in this state (|Ue4 |2
0.05) this, predominantly sterile, neutrino will have equilibrium concentration in the
Universe.
The bound on mass of two or three heavy degenerate neutrinos will be stronger than (1).
However, the decrease of the limit is weaker than just n1 , where n is the number of
the degenerate neutrinos. We estimate the bounds for two and three neutrinos performing
rescaling of the bound (1) according to results for 1, 2, and 3 neutrinos in [42]. Thus, for
two degenerate neutrinos we get from [43] and [2] m < 1.3 eV. In the (2 + 2) schemes with
normal mass hierarchy or with normal ordering this limit excludes the upper island of
parameters allowed by LSND (see Fig. 4). In the case of inverted hierarchy it confirms the
Mainz result.
For the three degenerate neutrinos we find m < 0.9 eV. This bound excludes significant
parts of the otherwise allowed regions of the 3 scheme and the (3 + 1) schemes with
inverted hierarchy or order of the states.
Forthcoming cosmological data can further substantially improve the bounds. In the last
column of the Table 1 we show the bounds which can be obtained using data from the
Sloan Digital Sky Survey (SDSS) [42] for m h2 < 0.17.
7.3. Beta decay measurements and the neutrino mass spectrum
The identification power of the -decay studies depends on whether future measurements will be able to observe the small kink and the tail after large kink or not.
If the -decay measurements are not able to identify the small kinks and the
suppressed tail, the only expected distortion effect is a shift of the end point and the
corresponding bending of the spectrum.
90
Suppose that the LSND result will not be confirmed and the effects of sterile neutrinos
will not be not found neither in the solar, nor in the atmospheric neutrino fluxes. This will
be the evidence for the three neutrino scenario. (The existence of additional sterile neutrino
which mixes weakly with the block of active neutrinos does not change the conclusion.)
As we have discussed, in this case the -decay parameter, m , will determine the scale of
three degenerate neutrinos.
If the LSND result is confirmed and m2LSND is measured, the important conclusions

will be drawn from the comparison of the values of m and m2LSND .
Let us remind that experiments which are insensitive to the small kinks and tails
will find m = 0 for the schemes with normal mass hierarchy, and m = m2LSND
for the schemes with inverted mass hierarchy. Any type of relation
is possible for the
non-hierarchical scheme with normal ordering of states: (i) m < m2LSND , (ii) m >

m2LSND in the case of relatively small m2LSND or (iii) m m2LSND (which looks

l = (mh )2 (ml )2 . For
as an accidental coincidence since this will imply the equality m
the non-hierarchical schemes with inverted order one has m > m2LSND .
Therefore:
A negative result of the neutrino mass measurements will be the evidence of the
scheme with normal hierarchy or the non-hierarchical scheme with normal order and
ml below the sensitivity limit:
ml < 0.3 eV (see Eq. (10)).
If it is established that m < m2LSND , the non-hierarchical scheme with normal

order of states should be selected.
Moreover, the mass scales will be completely
l
h
determined: m = m , and m = m2 + m2LSND .

The inequality m > m2LSND will testify for the non-hierarchical scheme with

inverted order of states. In this case: mh = m and ml = m2 m2LSND . The
inequality can correspond
also to the non-hierarchical scheme with normal order, so
that m = ml and mh =
m2 + m2LSND .
If m coincides (within the error bars) with

possibilities will be realized :
m2LSND , one of the following

(a) the schemes with inverted mass hierarchy and m mh m2LSND ,
(b) the non-hierarchical scheme with inverted order of levels and relatively small mass
of the light set, ml , so that mh is only slightly (within the error bars) larger than

m2LSND ,
l
(c) non-hierarchical
schemes with normal order of levels and m = m . In this case
the equality ml
m2LSND is accidental.
Recently, possible implications of results from LSND and KATRIN experiments have been
discussed also in [47].
91
If future -decay experiments detect the small kinks and the suppressed tail, we will be
able to measure eh and el and unambiguously discriminate the hierarchical schemes from
the non-hierarchical ones and the inverted scenarios from the normal ones even without
using the LSND result. This will be independent test of the 4 scheme. Indeed:
The spectrum with the small kink at m and the tail without bending, will be the
evidence of the scheme with normal mass hierarchy.
The spectrum with the large kink at m and the tail without bending will testify for
the schemes with inverted mass hierarchy.
The spectrum with the small kink and the bending at higher energies will correspond
to the non-hierarchical scheme with normal order of the states.
The spectrum with the large kink and the suppressed tail with shifted end point
(the latter will probably be impossible to establish) will indicate the non-hierarchical
scheme with inverted order of states.
Notice that all these results are the same for the (3 + 1) and (2 + 2) schemes. The
study of the -decay spectrum cannot distinguish these schemes. The (3 + 1) and (2 + 2)
schemes can be discriminated by studies of effects of the sterile neutrinos in the solar
and atmospheric neutrino oscillations, by MiniBooNE experiment, by searches for the
oscillations in the e and disappearance channels, etc.
The detection of the small kinks will also open a possibility to search for the mixing of
the sterile neutrinos which are not associated with the LSND result.
7.4. Measuring the absolute mass scale
Without direct kinematic measurements, the absolute scale of neutrino mass can be
established only in certain exceptional cases. This will be possible if solar neutrino data
are explained by the SMA solution and the 20-decay is observed. In this case mee will
give the mass of the solar pair: msun = mee . Then using the oscillation results one can
reconstruct the whole spectrum.
In the 3 scheme mee will immediately determine the mass of all the three quasidegenerate neutrinos.
In 4 schemes the mass reconstruction will depend on the type of the scheme:
In the scheme with normal mass hierarchy mee is expected to be small: mee
m2atm < 0.07 eV. So, if a small mee is detected or a strong bound on mee is obtained, the

mass of the heavy set will be given by the LSND result: mh m2LSND . Establishing the
scale of light masses still will be rather problematic. Moreover, if the 20-decay searches
give only the upper bound on mee , we should assume that neutrinos are the Majorana
particles to make conclusion on the mass.
In the non-hierarchical scheme with normal ordering of states, mee can be as large as the
l
present upper bound. Thus, m
ee will determine the mass of the light set: m = mee , and for
the heavy set we find mh = m2ee + m2LSND .
The schemes with inverted mass hierarchy or inverted ordering are almost excluded by
2
2
the fact that already present data indicate inequality (mmax
ee ) < mLSND .
92
For the large mixing solutions of the solar neutrino problem (LMA, LOW, VAC)
the absolute mass scale cannot be restored from mee due to possible cancellation
which depends on unknown CP-violating phase. Inversely, the data from the -decay
measurements can be used to determine this phase. If mee is measured, we will be able
to put both the lower and the upper bounds on the absolute scale of masses: m
mee / cos 2 and m > mee .
Even in those special cases where the determination of the absolute scale is possible
there are two problems:
Uncertainties of the nuclear matrix elements will lead to significant uncertainty in the
determination of the absolute scale.
We should assume that the exchange of the light Majorana neutrinos is the only
mechanism of the 20-decay.
In view of this, and keeping also in mind that the SMA solution is disfavored by the
latest solar neutrino data, we can conclude that developments of the direct methods of
determination of the neutrino mass (and KATRIN may be only the first step) is unavoidable
if we want to reconstruct neutrino mass spectrum completely.
8. Conclusions
(1) We have studied effects of the neutrino mass and mixing on the -decay spectrum in
three neutrino schemes which explain the solar and atmospheric neutrino data as well as in
all possible 4 schemes which explain also the LSND result.
We find that schemes which can produce an observable effect in the planned sub-eV
measurements should contain the sets (one or more) of quasi-degenerate states. The only
exception is the (3 + 1) scheme with normal mass hierarchy. However, it leads to a small
kink which will be difficult to observe. We show that the effects in the -decay spectrum
(q)
are described by the effective masses of the quasi-degenerate sets, m , and their coupling
(q)
with the electron neutrino, e .

(2) At present, a rather wide class of realistic schemes exist which can lead to an
observable effect in the sub-eV studies of the -decay spectrum. We show however that
future oscillation experiments and 20-decay searches can significantly restrict these
possibilities.
(3) The three neutrino schemes which explain the solar and atmospheric neutrino
anomalies in terms of neutrino oscillations can lead to an observable effect in future decay measurements only if all three mass eigenstates are quasi-degenerate. The -decay
measurements give unique possibility to identify these schemes. Even if SMA solution is
established and the neutrinoless double beta decay is discovered, so that mee sets the scale
of the neutrino masses, the question will rise whether the neutrino exchange is the only
mechanism of the neutrinoless double beta decay. Only comparison of the mee with results
of the beta decay measurements will give the answer. In the case of large mixing solutions
of the solar neutrino problem, simultaneous measurements of the mee and m as well as
93
the solar mixing angle open the possibility to determine the relative CP-violating phase of
the mass eigenvalues.
(4) In the four-neutrino schemes which explain also the LSND result, observable effects
in the -decay are strongly restricted by the Bugey and CHOOZ bounds. Four types of
spectrum distortion are expected depending on the type of mass hierarchy (ordering of
levels) in the scheme:
(a) The spectrum with the large kink and suppressed tail above the kink. This type of
distortion realizes in the scheme with inverted mass hierarchy.
(b) The spectrum with small kink. This type of distortion is expected in the scheme with
normal mass hierarchy.
(c) The spectrum with small kink and strong bending at the shifted end point. Such a
distortion corresponds to the non-hierarchical spectrum with normal ordering of states.
(d) Spectrum with large kink and strongly suppressed tail above the kink and a shift
of end point. This type of distortion realizes in the non-hierarchical scheme with inverted
order of states.
The rates in the suppressed tails and sizes of small kinks are determined by eh or (1eh )
and the latter quantities are restricted by the Bugey or CHOOZ bounds: eh 0.027.
The lower bound on eh appears from the LSND result. This bound is close to the Bugey
upper bound in the (3 + 1) schemes and it can be much weaker for the (2 + 2) schemes.
The 4 schemes with inverted mass hierarchy are disfavored by the data from SN1987A,
leading to a hard spectrum of e .
(5) The identification power of the -decay measurements will depend on the possibility
to detect the suppressed tail and the small-size kinks in the spectrum.
Even if small kinks or suppressed tails are unobservable,
the important conclusions
can be drawn from comparison of the values of m and m2LSND . This will allow
one to identify the type of mass hierarchy (in 4 schemes) and also to distinguish the
hierarchical from non-hierarchical schemes. Note that if the small kink or suppressed
tail are unobservable, the effect expected from the presently favored schemes of neutrino
masses is the same as the effect of the electron neutrino with definite mass.
Observations of the small kinks or suppressed tails will allow us to measure the mixing
parameters eh and el and to make the independent identification of the scheme. The decay measurements can also distinguish the three and four-neutrino schemes. However,
this can be done by MiniBooNE and other experiments even before new -decay results
will be available.
(6) Even if the LSND result is not confirmed, there are some motivations to search for
the kinks in the energy interval (110) eV below the end point. The kinks can be due to
mixing of the active neutrinos with the light singlet fermions which originate from some
other sectors of theory beyond the Standard model.
(7) The important conclusions can be drawn from the combined analysis of results of
the -decay measurements, 20-searches and the identification of the solution of the
solar neutrino problem.
(8) Negative results of future -decay experiments will have a number of important
implications. In particular,
94
Large part of the parameter space of the 3 schemes with degenerate mass spectrum
will be excluded;
If
the LSND result is confirmed, and the bound from the beta decay is m <
m2LSND the schemes with inverted hierarchy (order) as well as the schemes with

normal order and ml > m2LSND will be excluded.
If we want eventually to know the whole story about neutrinos we should measure
the absolute values of their masses. From the point of view of implications for the
fundamental theory, for astrophysics and cosmology, the masses are at least as important as
other neutrino parameters such as the mixing angles and CP-violating phases. As follows
from our study, to reconstruct the absolute values of masses unambiguously and without
additional assumptions one needs almost unavoidably to develop and to perform new direct
(kinematic) measurements (or look for some new alternatives). The KATRIN experiment
may be just the first step. There is no reason, why we should devote less time, effort and
resources for determination of the absolute scale of neutrino mass than, e.g., we are going
to devote to measurements of the oscillation parameters and the CP-violating phases.
Acknowledgements
One of us (A.S.) would like to thank J. Beacom, S. Parke and F. Vissani for useful
discussions. A.S. is grateful to G. Drexlin and C. Weinheimer, organizers of The
International Workshop On Neutrino Masses in the sub-eV range where preliminary
results of this paper have been reported [48]. We thank G. Drexlin for discussion of
sensitivity limits of future beta decay experiments, C. Lunardini, H. Pas and A. Ringwald
for remarks concerning the first version of the paper. The work of O.P. was supported
by Fundao de do Estado de So Paulo (FAPESP), by Conselho Nacional de Cincia e
Tecnologia (CNPq) and by the European Union TMR network ERBFMRXCT960090. O.P.
is grateful to GEFAN for valuable discussions and useful comments.
Appendix A. Effective mass of the set of quasi-degenerate states

The effect of the set of quasi-degenerate states on the -decay spectrum can be described
by the effective mass, m , and the coupling e of the set with the electron neutrino. Let us
evaluate the accuracy of such an approximation. The error is maximal in the energy interval
close to E0 mi , where mi is the mass of the lightest state from the quasi-degenerate
set. Let us compare number of events (decays) due to states from this set in the interval:
(E0 mi E) (E0 mi ), using (1) the effective parameters e and m : n(e , m ),
and (2) the precise parameters of states: n(Uei , mi ). Here E is the energy interval
which can be resolved by the detector. Let us calculate
R
n(e , m ) n(Uej , mj )
.
n(e , m )
95
Since in the energy range that we are interested, F (E, Z) and p in (7) are smooth functions
of energy we can factor out them and estimate the relative error as follows

2 E0 mi
|U
|
(E
E)
(E0 E)2 m2j (E0 E mj ) dE
ej
0
j
E0 Emi
R1

E0 E
2
2
E0 Emi e (E0 E) (E0 E) m (E0 E m ) dE

e [(mi + E)2 m2 ]3/2 j |Uej |2 [(mi + E)2 m2j ]3/2
=
,
e [(mi + E)2 m2 ]3/2
where in the sum j runs over the set.
Note that although the derivative of spectrum at E0 mi is divergent, n has finite
derivative. If |mj m |
m one can expand the relative error around m over
m/((mi + E)2 m2 ) as follows

3 j |Uej |2 [2m2 (mi + E)2 ]( mj )2
3 j |Uej |2 m mj
+
,
R=
(A.1)
e [(mi + E)2 m2 ]
e [(mi + E)2 m2 ]2
where mj mj m . Using the inequality m
E, we can rewrite (A.1) as

3 j |Uej |2 m mj
3 j |Uej |2 [m2i 2mi E E 2 ]( mj )2
R=
(A.2)
.
+
e (2mi E + E 2 )
e [2mi E + E 2 ]2
If bending of the energy spectrum is observable, E is of the order of mi or smaller.
Consequently, the first term in (A.1) is of the order of m/ E and the second term is of
order ( m/ E)2 . If we set

2
j mj |Uej |
m =
,
2
j |Uej |
the first term in (A.2) vanishes and R will be of the order ( m/ E)2 .
References
[1] Super-Kamiokande Collaboration, S. Fukuda et al., Phys. Rev. Lett. 85 (2000) 3999;
C. McGrew, Talk presented at International Workshop on Neutrino Telescopes, Venice, Italy,
March 69, 2001.
[2] X. Wang, M. Tegmark, M. Zaldarriaga, hep-ph/0105091.
[3] D.J. Eisenstein, W. Hu, M. Tegmark, Ap. J. 518 (1999) 2.
[4] W.H. Kinney, Phys. Rev. D 63 (2001) 043001.
[5] T. Totani, K. Sato, H.E. Dalhed, J.R. Wilson, Astrophys. J. 496 (1998) 216.
[6] J.F. Beacom, R.N. Boyd, A. Mezzacappa, Phys. Rev. D 63 (2001) 073011, astro-ph/0010398.
[7] D. Fargion, B. Mele, A. Salis, Astrophys. J. 517 (1999) 725;
D. Fargion, B. Mele, astro-ph/9906451.
[8] T.J. Weiler, Astropart. Phys. 11 (1999) 303.
[9] H. Pas, T.J. Weiler, Phys. Rev. D 63 (2001) 113015.
[10] Z. Fodor, S.D. Katz, A. Ringwald, hep-ph/0105064.
[11] H.V. Klapdor-Kleingrothaus, H. Paes, A.Yu. Smirnov, Phys. Rev. D 63 (2001) 073005.
96
[12] H.V. Klapdor-Kleingrothaus, for HeidelbergMoscow Collaboration, hep-ph/0103062;

H.V. Klapdor-Kleingrothaus, for HeidelbergMoscow Collaboration, hep-ph/0103074.
[13] S. Pirro et al., Nucl. Instrum. Methods A 444 (2000) 71.
[14] E. Fiorini et al., Phys. Rep. 307 (1998) 309;
A. Bettini, Nucl. Phys. Proc. Suppl. 100 (2001) 332.
[15] H. Ejiri, J. Engel, R. Hazama, P. Krastev, N. Kudomi, R.G.H. Robertson, Phys. Rev. Lett. 85
(2000) 2917.
[16] GENIUS Collaboration, H.V. Klapdor-Kleingrothaus et al., hep-ph/9910205.
[17] F. Boehm, P. Vogel, Physics of Massive Neutrinos, Cambridge Univ. Press, 1992.
[18] J. Bonn et al., Nucl. Phys. Proc. Suppl. 91 (2001) 273.
[19] V.M. Lobashev et al., Proc. of the Int. Conf. Neutrino 2000, Sudbury, Canada, Nucl. Phys.
Proc. Suppl. 77 (1999) 327;
V.M. Lobashev et al., Nucl. Phys. Proc. Suppl. 91 (2001) 280.
[20] V. Aseev et al., A next generation tritium decay experiment with sub-eV sensitivity for the
electron-neutrino mass, Talks presented at International Workshop on Neutrino Masses in the
sub-eV Range, Bad Liebenzell, Germany, January 1821, 2001, http://www-ik1.fzk.de/tritium/.
[21] C. Weinheimer et al., Phys. Lett. B 460 (1999) 219;
F. Vissani, hep-ph/0102235;
F. Vissani, Nucl. Phys. Proc. Suppl. 100 (2001) 273.
[22] S. Petcov, A.Yu. Smirnov, Phys. Lett. B 322 (1994) 109.
[23] A.S. Joshipura, Z. Phys. C 64 (1994) 31;
H. Minakata, O. Yasuda, Phys. Rev. D 56 (1997) 1692;
F. Vissani, hep-ph/9708483;
J. Hellmig, H.V. Klapdor-Kleingrothaus, Z. Phys. A 359 (1997) 351;
H.V. Klapdor-Kleingrothaus, M. Hirsch, Z. Phys. A 359 (1997) 361;
H.V. Klapdor-Kleingrothaus, J. Hellmig, M. Hirsch, J. Phys. G 24 (1998) 483;
H. Minakata, O. Yasuda, Nucl. Phys. B 523 (1998) 597;
V. Barger, K. Whisnant, Phys. Lett. B 456 (1999) 194;
F. Vissani, hep-ph/9904349, Proc. of the 6th Tropical Seminar on Neutrino and Astroparticle
Physics, May 1721, 1999, San Miniato, Italy;
J. Ellis, S. Lola, Phys. Lett. B 458 (1999) 310;
G.C. Branco, M.M. Rebelo, J.I. Silva-Marcos, Phys. Rev. Lett. 82 (1999) 683;
R. Adhikari, G. Rajasekaran, Phys. Rev. D 61 (2000) 031301;
H. Georgi, S.L. Glashow, Phys. Rev. D 61 (2000) 097301;
W. Rodejohann, Nucl. Phys. B 597 (2001) 110;
S.M. Bilenky, S. Pascoli, S.T. Petcov, hep-ph/0102265.
[24] M. Apollonio et al., Phys. Lett. B 466 (1999) 415.
[25] A. Piepke, for the Palo Verde Collaboration, Nucl. Phys. Proc. Suppl. 91 (2001) 91;
F. Boehm et al., Phys. Rev. D 62 (2000) 072002.
[26] M.C. Gonzalez-Garcia, C. Pea-Garay, Phys. Rev. D 63 (2001) 033005.
[27] Bugey Collaboration, B. Achkar et al., Nucl. Phys. B 434 (1995) 503.
[28] CDHS Collaboration, F. Dydak et al., Phys. Lett. B 134 (1984) 281.
[29] T. Kajita, Nucl. Phys. Proc. Suppl. 85 (2000) 44;
S.M. Bilenky, C. Giunti, W. Grimus, Prog. Part. Nucl. Phys. 43 (1999) 1;
C. Giunti, Nucl. Phys. Proc. Suppl. 100 (2001) 244.
[30] O.L.G. Peres, A.Yu. Smirnov, Nucl. Phys. B 599 (2001) 3.
[31] Y.-Z. Qian et al., Phys. Rev. Lett. 71 (1993) 1965;
G.C. McLaughlin, J.M. Fetter, A.B. Balantekin, G.M. Fuller, Phys. Rev. C 59 (1999) 2873.
[32] A. Kusenko, G. Serge, Phys. Lett. B 369 (1997) 197;
A. Kusenko, astro-ph/9903167;
E. Nardi, J.I. Zuluaga, astro-ph/0006285.
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
97
J. Peltoniemi, hep-ph/9506228.
K.R.S. Balaji, A. Perez-Lorenzana, A.Yu. Smirnov, hep-ph/0101005.
KARMEN Collaboration, K. Eitel, Nucl. Phys. Proc. Suppl. 91 (2000) 191.
SNO Collaboration, J. Boger et al., Nucl. Instrum. Methods A 449 (2000) 172.
N. Bahcall, P.I. Krastev, A.Yu. Smirnov, Phys. Rev. D 63 (2001) 053012;
M.C. Gonzalez-Garcia, C. Pen-Garay, Phys. Rev. D 63 (2001) 073013;
V. Barger, D. Marfatia, K. Whisnant, hep-ph/0104166.
A. Suzuki, Invited Talk at 8th International Workshop on Neutrino Telescopes, February 2326,
1999, Venice, Italy;
H. Murayama, A. Pierce, hep-ph/0012075;
KamLAND Collaboration, F. Boehm et al., Nucl. Phys. Proc. Suppl. 91 (2001) 99;
V. Barger, D. Marfatia, B.P. Wood, Phys. Lett. B 498 (2001) 53.
G. Ranucci et al., Nucl. Phys. Proc. Suppl. 91 (2001) 58;
G.L. Fogli, E. Lisi, D. Montanino, A. Palazzo, Phys. Rev. D 61 (2000) 073009.
LSND Collaboration, A. Aguilar et al., hep-ex/0104049.
M. Fukugita, G.-C. Liu, N. Sugiyama, Phys. Rev. Lett. 84 (2000) 1082.
W. Hu, D.J. Eisenstein, M. Tegmark, Phys. Rev. Lett. 80 (1998) 5255.
R. Croft, W. Hu, R. Dave, Phys. Rev. Lett. 83 (1999) 1092.
M. Tegmark, M. Zaldarriaga, A.J.S. Hamilton, Phys. Rev. D 63 (2001) 043007.
A.Yu. Smirnov, in: Proc. of Int. Symposium New Era in Neutrino Physics, Tokyo, Japan, hepph/9811296;
A.S. Dighe, A.Yu. Smirnov, Phys. Rev. D 62 (2000) 033007.
A.Yu. Smirnov, D.N. Spergel, J.N. Bahcall, Phys. Rev. D 49 (1994) 1389;
B. Jegerlehner, F. Neubig, G. Raffelt, Phys. Rev. D 54 (1996) 1194;
H. Minakata, H. Nunokawa, Phys. Lett. B 504 (2001) 301.
S.M. Bilenky, S. Pascoli, S.T. Petcov, hep-ph/0104218.
A.Yu. Smirnov, Talk presented at International Workshop on Neutrino Masses in the subeV Range, Bad Liebenzell, Germany, January 1821, 2001. Transparencies at http://wwwik1.fzk.de/tritium/.

Alternative N = 2 supergravity in singular five

dimensions with matter/gauge couplings
Hitoshi Nishino a , Subhash Rajpoot b
a Department of Physics, University of Maryland, College Park, MD 20742-4111, USA
b Department of Physics & Astronomy, California State University, Long Beach, CA 90840, USA
Received 17 May 2001; accepted 25 July 2001
Abstract
We present an extended study of our previous work on an alternative five-dimensional N = 2
supergravity theory that has a single antisymmetric tensor and a dilaton as a part of supergravity
multiplet. The new fields are natural NeveuSchwarz massless fields in superstring theory. Our
total matter multiplets include n copies of vector multiplets forming the sigma-model coset
space SO(n, 1)/SO(n), and n copies of hypermultiplets forming the quaternionic Khler manifold
Sp(n , 1)/Sp(n ) Sp(1). We complete the couplings of matter multiplets to supergravity with the
gauged group of the type SO(2) Sp(n ) Sp(1) H [U (1)]np+1 for an arbitrary gauge group
H with p dim H + 1, and the isotropy group Sp(n ) Sp(1) of the coset Sp(n , 1)/Sp(n ) Sp(1)
formed by the hypermultiplets. We also describe the generalization to singular 5D spacetime as in
the conventional formulation. 2001 Published by Elsevier Science B.V.
PACS: 04.50.+h; 04.65.+e; 11.30.Pb; 04.50.+h
Keywords: Supergravity; Five dimensions; Singular spacetime; M-theory
1. Introduction
The importance of supergravity in 5D spacetime manifests itself in many contexts,
such as the supersymmetrization [1,2] of RandallSundrum type brane-world scenario [3],
namely, gauged supergravity in singular 5D spacetime. In Ref. [2], the introduction of
a 4th-rank antisymmetric tensor A made it easier to handle supergravity in such a
singular spacetime with the orbifold-type singularity S 1 /Z2 . Another important aspect of
5D supergravity is related to what is called holographic anti-de Sitter and superconformal
field theory (AdS/SCF) correspondence, namely the conjecture that the large N limit of
SU(N) superconformal field theories in 4D are dual equivalent to supergravity on AdS
This work is supported in part by NSF grant # PHY-98-02551.

E-mail addresses: nishino@nscpmail.physics.umd.edu (H. Nishino), rajpoot@csulb.edu (S. Rajpoot).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 7 2 - 8
H. Nishino, S. Rajpoot / Nuclear Physics B 612 (2001) 98122
99
spacetime in 5D [4,5]. In both of these aspects of 5D supergravity, the presence of the

5D cosmological constant, via the gauging of the N = 2 automorphism group SL(2, R) =
Sp(1) (or its SO(2) subgroup) plays a crucial role.
The conventional on-shell formulation of N = 2 supergravity in 5D was initiated in
[6] in which an arbitrary number of vector multiplets is coupled to supergravity, and
generalized further in [79]. However, in these formulations [69], the dilaton field as
one of the important NS fields does not have manifest dilaton scale invariance. Moreover,
an additional complication is that the tensor fields in [69] appear in symplectic pairs,
obeying the self-duality condition in odd spacetime dimensions, and therefore the single
antisymmetric tensor field B as another important NS field [10] is mixed up with other
tensor fields. In order to overcome these drawbacks in these on-shell formulations [69],
we may try an off-shell formulation as an alternative, but such a formulation lacks the
manifest -model geometry formed by scalars, which is hidden at the off-shell level
before eliminating auxiliary fields. This is similar to the 4D case of Khler manifold
structure in on-shell N = 1 supergravity [11] which is hidden in the off-shell formulation.
In our previous paper [12], we have proposed an alternative on-shell N = 2 supergravity
multiplet in 5D, which has an irreducible field content larger than the conventional one
[6,8,9] including an antisymmetric tensor and a dilaton fields that are NeveuSchwarz
(NS) massless fields in superstring theory [10]. Our supergravity multiplet has the field
content (e m , A , B , A , A , ) with 12 + 12 on-shell degrees of freedom, where the
fnfbein e m , the gravitini A , and the graviphoton A are the same as the conventional
N = 2 supergravity [69], while an antisymmetric tensor B , a dilatino A , and a dilaton
are our new field content. Among these, the antisymmetric tensor B and the dilaton
are natural NS massless fields in superstring theory [10].
In the present paper, we will continue the study of our alternative on-shell N = 2
supergravity [12] coupled to n copies of vector multiplets and n copies of hypermultiplets.
In our formulation, the dilaton and the antisymmetric fields as the important NS fields are
treated separately from other scalars. Our n scalars form the coordinates of the -model
coset SO(n, 1)/SO(n), while the 4n scalars form the coordinates of the -model
coset Sp(n , 1)/Sp(n ) Sp(1) [12]. We present the general gaugings of our system in the
presence of hypermultiplets that were not given in our previous paper [12], and consider our
supergravity in singular 5D spacetime, as the supersymmetrization of RandallSundrum
brane-world scenario [3], following the prescription of [2] for dealing with the orbifoldtype S 1 /Z2 singularity.
This paper is organized as follows: in Section 2, we review our N = 2 alternative
supergravity as a notational preparation before gaugings. In Section 3, we give the general
treatment for the gauging of an arbitrary non-Abelian gauge group that has nothing to do
with the coset Sp(n , 1)/Sp(n ) Sp(1). Section 4 is devoted to our main focus in the
present paper, namely, to show how to gauge the automorphism group Sp(1) = SL(2, R)
of N = 2 supersymmetry, or more generally the whole isotropy group Sp(n ) Sp(1)
of the coset Sp(n , 1)/Sp(n ) Sp(1), in the presence of hypermultiplets, which was not
accomplished in our previous paper [12]. As a by-product, we will give the most general
case of gauging of the total group SO(2) Sp(n ) Sp(1) H [U (1)]np+1 for an
100
arbitrary gauge group H with p dim H + 1, and the isotropy group Sp(n ) Sp(1) of
the coset Sp(n , 1)/Sp(n ) Sp(1) formed by the hypermultiplets. Section 5 is for the
formulation of our alternative supergravity in a singular 5D spacetime, with the orbifoldtype singularity S 1 /Z2 , i.e., the supersymmetrization [1,2] of RandallSundrum braneworld scenario [3]. Section 6 is for our conclusion, while the important notations and
conventions are given in appendix.
2. Coupling of vector multiplets and hypermultiplets to 5D, N = 2 supergravity

We start with reviewing the couplings of 5D, N = 2 supergravity to vector multiplets
and hypermultiplets [12], before general non-Abelian gaugings. The field content of the
multiplet of supergravity is (e m , A , A I , B , A , , aA , , a ) with 12 + 12 onshell degrees of freedom [12]. Here , , . . . are for the curved world indices, while
m, n, . . . are local Lorentz with the metric (mn ) = diag(, +, +, +, +), e m is the
fnfbein, A is the gravitino with A = 1, 2 for 2-representation of the automorphism
group Sp(1) = SL(2, R) for the N = 2 supersymmetry. The raising/lowering of the indices
A, B, . . . is performed by the Sp(1) metric AB , AB , and therefore special attention is
needed for superscript/subscript of these indices, in particular, their inner products. As
in [12], we use here Sp(1) = SL(2, R) notation instead of SU(2) as the automorphism
group, in order to make all the bosonic fields manifestly real, just for simplicity.
The vectors A I (I = 0, 1, 2, . . . , n) form the (n + 1)-representation of SO(n, 1) in
the coset SO(n, 1)/SO(n) [13,14]. The ( = 1, 2, . . . , n) are the n-dimensional
-model coordinates of the coset SO(n, 1)/SO(n), aA (a = 1, 2, . . . , n) are in the
n-representation of SO(n), ( = 1, 2, . . . , 4n ) are the 4n -dimensional coordinates of
the quaternionic Khler manifold Sp(n , 1)/Sp(n ) Sp(1), and a (a = 1, 2, . . . , 2n )
are in the 2n -representation of Sp(n ). As described in [12], this is the combination
of our multiplet of supergravity (e m , A , A , B , A , ), n copies of the vector
multiplets (C , A , ), and 4n copies of the hypermultiplets ( , a ). In particular, the
graviphoton A is identified with the zeroth component A 0 , while the n copies of the
vector field C from the vector multiplets renamed as A 1 , A 2 , . . . , A n , combined
into the unified notation A I (I = 0, 1, 2, . . . , n). Since the indices I, J, . . . are with the
indefinite metric (I J ) = diag(, +, +, . . . , +), we make the raising/lowering of these
indices explicit. Note that our multiplet of supergravity is distinct from the conventional
one (e m , A , A ) [6], in which only the fnfbein, gravitino and the graviphoton form
the irreducible field content.
The geometrical relationships associated with the coset SO(n, 1)/SO(n) are conveniently listed up as [12,13]
[Hab , Hcd ] = bc Had ac Hbd + ad Hbc bd Hac ,
(2.1a)
[Ka , Kb ] = +2Hab ,
[Hab , Kc ] = bc Ka ac Kb ,
1
LA I LI B = A ab (Hab )A B + V a (Ka )A B ,
Ab c = Lb I LI c ,
2
(2.1b)
(2.1c)
(Ka )b(0) = (Ka )(0)b = 2 ab ,
(Hab )c d = ac b d bc a d ,
LI LA = I ,
A
LI LI
L L(0) ,
I
(2.1d)
LA LI = A ,
(0)
101
(2.1e)
LI L = +1,
I
La LI 0,
I
LI L 0,
a
(2.1f)
LI J AB LI LJ = LI LJ + LI LJ a ,
(2.1g)
LI J L = LI ,
(2.1h)
LI J La = +LI a ,
D LI a = 2 LI V a ,
D LI = LI = 2 LI a Va ,

[D , D ]LI a = 2 V a V b V a V b LI b ,

R ab = 2 V a V b V a V b ,
R = 2n(n 1) 0,
J
LI J = 0, (2.1i)
(2.1j)
(2.1k)
D Xa Xa + Aa Xb ,
b
(2.1l)
which are self-explanatory exactly in the same notation as in [12]. The Cartan decomposition of the SO(n, 1) Lie algebra is dictated by the SO(n) generators Hab and the
coset generators Ka , satisfying (2.1). The indices a, b, . . . = (1), (2), . . . , (n) are for
the vectorial representation of SO(n). The indices A, B, . . . = ((0), a), ((0), b), . . . =
(0), (1), (2), . . ., (n) are for the local coordinates on Sp(n , 1)/Sp(n ) Sp(1). 1 In other
words, A, B, . . . = ((0), a), ((0), b), . . . are the (n + 1)-dimensional extension of the original n-dimensional indices a, b, . . .. The indices I, J, . . . = 0, 1, . . . , n are for the curved coordinates, while , , . . . = 1, 2, . . . , n are for the coordinates on Sp(n , 1)/Sp(n ) Sp(1).
The raising/lowering of the indices A, B, . . . is performed by the metric tensor (AB ) =
diag(, +, +, . . . , +). The MaurerCartan form made of LI A decomposes as in (2.1c).
Eq. (2.1d) gives the explicit components of H s and Ks, while (2.1e)(2.1i) are relevant
orthonormality relations. Eq. (2.1i)(2.1l) are for the SO(n) covariant derivative D .
As for the geometry related to the quaternionic Khler manifold Sp(n , 1)/Sp(n )Sp(1)
we start with the representative La , which satisfies the MaurerCartan form for the coset
Sp(n , 1)/Sp(n ) Sp(1) [79,12,15,16]:
L1 L = A i T i + A I TI + V aA KaA ,
aB
= + 12 g A B
i B
T A ,
(2.2a)
g VaA VbB = ab AB ,

(2.2b)
VaA V

1
J i = T i A B VaB V aA VaB V aA = F i ,
(2.2c)
2
where TI (I = 1, 2, . . . , n (n + 1)/2) are the generators of Sp(n ), T i (i = 1, 2, 3) are
the generators of Sp(1), while KaA are the coset generators of Sp(n , 1)/Sp(n ) Sp(1)
[15,16]. All of these equations involve the vielbein V aA for this quaternionic Khler
manifold.
With all other details of geometry skipped, our Lagrangian before gaugings is [12]

1
1
1
e1 L0 = R D e4 G2
4
2
12

1
1
e2 LI a LJ a + LI LJ F I F J a D a
4
2
i
1
2 F
1 The indices A, B, . . . used both for the 2-representations and for these local Lorentz coordinates are not to be
confused each other, as long as we keep track of the context they are used.
102

3

1
1
D
g ( )2
2
4
2

1
1
i
g a D a + V a a
2
2
2

3i
+
+ iV aA a
2

i
e + 2 LI F I
4 2

i
+ e2 G
6 3

1
5
e2 6 G e2
G
24
72

1
1
e a LI a F I e LI F I
2 6
2 2

i
i a
I
LI F + e
a LI F I
e
12 2
4 2
1 2 a
1 2 a

+ e
a G + e
a G
24
24
a

i
1
+ e2
a LI F I e a a LI F I ,
(2.3)
6
4 2
yielding an invariant action S0 under supersymmetry

Q e m = + m ,
i
Q = + ( ),
3

i
Q A = +D A + e 4 A LI F I
6 2

1 2
3 A

+ e

G ,
18
2

i
1
1
Q A I = e LI ( ) + e LI ( ) + e a La I ,
6
2
2

i
Q B = +e2 ( [ ] ) + e2 ( ) 2LI J A[| I Q A|] J ,
3
3i A
1 A
i 2 A
A
I
,
G
Q = e LI F + e
2
2 6
6 3

i
Q = + Va a ,
2
1
i
Q aA = e A LI a F I A V a ,
2 2
2

A a
Q = +iVaA ,
Q a = iV aA A ,
(2.4)
103
up to quartic fermion (or quadratic fermion) terms in the Lagrangian (or transformation
rules). Here we have omitted the Sp(1) indices A, B, . . . in the Sp(1)-invariant products,
e.g., ( m ) ( A m A ).
As in the usual dilaton couplings in supergravity [17], the antisymmetric field B and
the vectors A I are scaled, when the dilaton is shifted by a constant value:
+ c,
B e2c B ,
A I e c A I ,
(2.5)
where c is an arbitrary constant parameter. This global symmetry controls the various
exponential couplings of in the Lagrangian (2.3).
The various covariant derivatives and the field strength G in these equations are given
by

A
D A D () A + A i T i ,

A

D[ ] A D[| ()|] A + [| A i T i |] ,

A
D A D () A + A i T i ,

A

D aA D ()aA + A ab b A + A i T i a ,

a
D a D () a + A I T I ,
G 3[ B] 3LI J F[ I A] J ,
(2.6)
where A ab is the composite SO(n) connection on the coset SO(n, 1)/SO(n), while
A I and A i are, respectively, the composite connections of Sp(n ) and Sp(1) in
Sp(n , 1)/Sp(n ) Sp(1). The action of the generators T i or T I is such as (T i )A
(T i )AB B or (T I )a (T I )ab b . Since we are not concerned with the quadratic
fermionic terms in the transformation rule (2.4), the Lorentz connection rs contains
the usual unholonomy coefficients just made of the fnfbeins.
Compared with the conventional formulations [6], there is a similarity as well as
basic difference. The similarity is that our tensor field B can be dualized into a
vector field B by a duality transformation so that the final field content will be
(e m , A , A , B , A , ). From this viewpoint, our system (2.1) is dual equivalent
to the conventional formulation with only one vector multiplet, in particular the dilaton
field plays the coordinate of SO(1, 1), as usual in superstring theory. However, the caveat
at this stage is that even though such a duality transformation is possible even after
coupling vector multiplets, the resulting -model structure is qualitatively different from
that given in the conventional formulations [69], as has been also explained in our
previous paper [12].
As has been also stressed in [12], the antisymmetric field B and the dilaton are
the natural NS massless fields in superstring [10] or M-theory [18]. Therefore, it is more
natural to have a supergravity with these fields in the point field theory limit. Another
advantage of introducing an antisymmetric tensor B is associated with the recent
development of non-commutative geometry in which the tensor B develops certain nontrivial constant value. We stress the fact that our supergravity multiplet contains the NS
fields B and as irreducible component fields, indicating that our supergravity is a
104
more natural point field theory limit of superstring theory [10] or M-theory [18] than the
conventional one [69].
3. Non-Abelian gauging of subgroup of SO(n, 1)

We next establish general non-Abelian gaugings in the presence of vector multiplets and
hypermultiplets. In this section, we consider the case that the gauged non-Abelian group
G has nothing to do with isotropy groups Sp(n ) Sp(1) in the coset Sp(n , 1)/Sp(n ),
but is just any other independent Lie group, which may be needed for more practical
model building. Since all the n copies of vectors in the vector multiplets together with the
graviphoton in the multiplet of supergravity form the (n + 1)-representation of SO(n, 1) in
the coset SO(n, 1)/SO(n), we need special care for such non-Abelian gaugings. Such nonAbelian gauging has been performed in the conventional formulation [6], as well as recent
works in [79], and in other dimensions such as in 7D [19]. In the formulation below, we
will mainly follow the notation in [19], in which the coset space formed by the scalars in
the vector multiplets is SO(n, 3)/SO(n) SO(3). This is slightly different from our coset
SO(n, 1)/SO(n), but we still can take advantage of the similarity between them.
First of all, the non-Abelian gauge group G should be the subgroup of SO(n, 1), and at
the same time dim G = n+1 should be satisfied, due to the coset structure to be maintained.
Second, the structure constant fI J K should satisfy the relationship [19]
fI J K fI J L LLK = f[I J K] ,
(3.1)
where LI J is the indefinite metric on SO(n, 1)/SO(n) as in Section 2. This condition is

satisfied when this indefinite metric LI J is identified with the CartanKilling metric I J
of G. To be specific, we can have G = SO(2) H [U (1)]np+1 , so that dim H = p 1,
dim G = n + 1, and
(LI J ) (I J ) = diag(1, I J , +1, +1, . . . , +1),
(3.2)
where I , J = 1, 2, . . . , p 1 with 1 p n + 1. Also in (3.2), the first 1 is the Cartan

Killing metric of SO(2) for the 0th direction in the (n + 1) dimensions, I J is that of H ,
while the last (+1, +1, . . ., +1) are the metrics for the Abelian factor groups [U (1)]np+1 .
In the special case of p = n + 1, there is no U (1) factor group. This situation is similar to
that in [19].
For such a gauge group G, we introduce the minimal coupling with the coupling
constant g.
Typically, we have [19]
Pa La I L = 2 Va

Pa La I LI + gf
I J K A J LK La I D LI
2 Va D ,
(3.3a)

b
b
Aa Aa
Aa b Aa b + gf
I J K A I La J LK b ,
(3.3b)
105

A

D aA D ()aA + A ab b A + A i T i a

A

D aA D ()aA + A ab b A + A i T i a ,
(3.3c)

1
I J K A I A J AK .
G G 3 [ B] LI J F[ I A] J + gf
3
(3.3d)
Eq. (3.3a) is none other than the standard minimal non-Abelian coupling for the adjoint
index I . Needless to say, the structure constants fI J K with the indices I, J, K for any of
the U (1) factor groups or SO(2) are supposed to vanish. So effectively, only the indices
I , J , K on fI J K remain. If we rewrite (3.3a) as

P Va P a 2 gA
(3.4)
I I 2 D ,
then its comparison with (3.3a) implies that
1
I = fI J K V a La J LK ,
2
(3.5)
with the Killing vectors I in the directions of the gauged group G. Eq. (3.3c) has a new
term for the non-Abelian coupling. Relevantly, we have
D LI = P a LI a ,
D LI a = P a LI ,
D LI = P a La I ,
D La I = Pa LI .
(3.6)
By defining
Cab fI J K La I Lb J LK = 2 I Vb La I = Cba ,
(3.7)
we have the important relationship

1
I LI b Cab ,
D[ P]a = + gF
2
by the use of another identity
LI b Cab = fI J K La J LK ,
(3.8)
(3.9)
confirmed by (2.1). As has been already mentioned, in expressions like (3.7)(3.9), the
structure constants fI J K in any directions of the U (1)s or SO(2) are supposed to vanish,
not to mention any other mixed directions of different groups. The same is also true for
the index I in the last term in (3.3d), in which any irrelevant component gives the vanishing
of A I . These geometrical structures are parallel to the 7D case in [19].
Note that in this non-Abelian gauging, the gaugini aA are not in the adjoint
representation, as opposed to the usual vector multiplets in higher dimensions [17], such
as that in 10D with the gaugino in the adjoint representation. This is in a sense not
surprising, because the gaugino fields should form the n-representation instead of the
(n + 1)-representation of SO(n, 1), and therefore their range of indices should differ from
that of the vector fields. This situation in 5D is similar to the original work in [6], or also
in [9,19].
106
We next introduce the Killing vectors I for the direction of the gauged group G. In
order to fix an invariant action, we also follow the result in [19], and we can postulate an
additional term needed in the Lagrangian [19]

Cab a b ,
e1 Lg = +ia ge
(3.10)
for the non-Abelian gauging, while putting no explicit 2 g-dependent
terms in the
transformation rules. Now the variation of Lg generates only two sorts of terms, when
fermionic cubic terms are ignored: (i) gF
-terms and (ii) gD-terms.
For the term in (i),

the variation of the D Noether-term is the only counter-contribution, while for the
term in (ii), the kinetic term of is the only contribution to cancel. Both of these two
sectors yield the same condition a = +23/2 consistently.
Armed with these preliminaries, we are ready to give the Lagrangian

1
1
1
e1 L1 = R D e4 G2
4
2
12

1
1
4
2

3

1
1
2
g D D ( )
D
2
4
2

1
1
g a D a
2
2

3i
i

a

+ V a D +
2
2

i
+ iV aA a e + 2 LI F I
4 2
i 2

+ e
G
6 3

1
5
e2 6 G e2
G
24
72

1
1
2 6
2 2

i
i
e LI F I + e a a LI F I
12 2
4 2
1 2 a
1 2 a
+ e
a G + e
a G

24
24
a

i
1
+ e2
a LI F I e a a LI F I
6
4 2

i
+ ge
(3.11)
Cab a b ,
2 2
2 The word explicit here implies any g-dependent
term other than those hidden in the covariant derivatives

such as (3.3).
107
with all of the D and G in (3.3), yielding an invariant action S1 under supersymmetry

Q e m = + m ,
i
Q = + ( ),
3

i
6 2

1 2
3
A G ,
+ e
18
2

i
1
1
Q A I = e LI ( ) + e LI ( ) + e a La I ,
6
2
2

i
Q B = +e2 ( [ ] ) + e2 ( ) 2LI J A[| I Q A|] J ,
3
3i A
1
i 2 A
A
A
I
,
G
Q = e LI F + e
2
2 6
6 3

i
Q = + Va a ,
2
1
i
Q aA = e A LI a F I A V a D ,
2 2
2

A a
Q = +iVaA ,
Q a = iV aA A .
(3.12)
Note that there is no need of any explicitly g-dependent
terms in the transformation rule.

There is no potential term generated in this gauging, which is similar to the conventional
N = 2 theories in 5D [79]. Compared with [19], since our vector fields do not carry extra
Sp(1) indices, no scalar potential term is generated.
Analogous to (2.5), we have the scaling invariance of L1 when the coupling constant g
transforms as
g ec g,
(3.13)
when the fields transform as in (2.5).

4. Sp(n ) Sp(1)-gauging
In our previous paper [12], we studied the gauging of SO(2) which is the subgroup of
Sp(1) in the isotropy groups Sp(n ) Sp(1) in the coset Sp(n , 1)/Sp(n ) Sp(1). Most of
the geometric relationships related to the coset SO(n, 1)/SO(n) are parallel to the SO(2)gauging [12], so we give important relations in such a way that the comparison with [12]
is easy to make.

Our total gauged group in this section is G = SO(2) Sp(n ) Sp(1) H
np+1
, which is a special case of the previous section. In fact, the first SO(2) is
[U (1)]
for the I = 0-direction for the indices I = 0, 1, . . . , n + 1, and the groups Sp(n ) Sp(1)
for the group H in the last
are regarded as a special case of H Sp(n ) Sp(1) H
108
with dim H
= p n (2n + 1) 4, such that the
section, and an arbitrary gauge group H
previous condition dim H = p 1 is maintained. Since the dimension p is still arbitrary,
for a large enough dimension of n.
we have enough freedom for choosing the group H
Accordingly, we arrange our index convention as follows. Among the indices I, J, . . . =
0, 1, 2, . . . , n for the total n + 1 copies of vector fields, we use the indices I , J , . . . =
1, 2, . . . , n (2n + 1) for the adjoint representation of Sp(n ), and combine them with
i, j, . . . = 1, 2, 3 for that of Sp(1), in terms of the combined indices I (I , i), J
, we
(J , j ), . . . for the gauged groups Sp(n ) Sp(1). For the adjoint indices for H
use the barred ones: I, J, . . . = 1, 2, . . . , p 1. As for the remaining product groups
SO(2) [U (1)]np+1 , we do not need particular indices in this section, so we do not
specify the indices for these groups. Compared with the indices I, J, . . . , all these indices
I, J, . . . ; I , J , . . . ; i, j, . . . ; I, J, . . . do not need distinctions of their raising/lowering due
to their positive definite metrics. Therefore their contractions are given as superscripts like
AI B I .
In our previous paper [6], the SO(2)-gauging was performed by introducing the constant
vectors V I , with the coupling constant g. In our present case of Sp(n ) Sp(1)-gauging,
this SO(2) group is enlarged to Sp(n ) Sp(1). In this section, we use the coupling constant
. Accordingly, all the combination of gVI A I in
g for Sp(1), g for Sp(n ), and g for H
[12] will be replaced by gA i i + g A I I , where I and i are the Killing vectors
for the gauged groups Sp(n ) Sp(1) in the coset Sp(n , 1)/Sp(n ) Sp(1).
Accordingly, the covariant derivatives on Sp(1) non-invariant fermions acquire the Sp(1)
minimal couplings in addition to the D s or in Section 2 as

A
A
D A D A D () A + D A i T i + gA i T i ,
(4.1a)
i i
A

A
A
A
D[ ] D[ ] D[| ()|] + D[| A T |]

A
+ gA[| i T i |] ,
(4.1b)

A
A
D A D A D () A + D A i T i + gA i T i ,
(4.1c)
i i a A

aA
aA
aA
ab A
D D D () + A b + D A T

A
+ gA i T i a ,
(4.1d)
D A I I ,
(4.1e)
I I a
I a

I
a
a
a

D D D () + D A T + g A T , (4.1f)
D g A I I gA i i A I I ,
with the generalized Killing vectors

I
I
I g
i
g i
I
(for Sp(n )),

(for Sp(1)),
(4.1g)
(4.2)
(otherwise).

For example, an expression like I LI actually means I LI I LI g I LI +
J
i
i
is understood from the fact that the group

g L . The absence of the component
109
has nothing to do with the coset Sp(n , 1)/Sp(n ) Sp(1). In (4.1), all the terms other
H
than explicit g-terms are just the previous covariant derivatives in (2.6) in which is
replaced by D , and the matrices T I are the anti-hermitian generator of Sp(n ) Sp(1),
as its index I reveals. This structure is similar to the models in [69].
The covariance of the derivatives in (4.1) are confirmed by considering the transfor
mations of these fields under the gauged groups G SO(2) Sp(n ) Sp(1) H
np+1
, such as
[U (1)]
G = + I I ,
G A I = I + fJ K I A J K D I ,

G I = J I J fI J K J K ,

G D = I D I ,
(4.3a)
(4.3b)
(4.3c)
for the local parameters I for the gauged groups in G, and the structure constants

I J K = g f I J K (for Sp(n )),
fIJK
= f
(for Sp(1)),
fij k = g ij k
fI J K
(4.4a)
IJ K
I J K = gf
),
(for
H
0
(otherwise),
L
fI J K fI J LLK = f[I J K] ,
(4.4b)
in the combined notation.
for the respective structure constants for Sp(n ), Sp(1) and H
Since the SO(2) group in the negative metric 0th direction is Abelian, it does not enter
(4.4), and therefore we do not need to distinguish the super/subscripts on the r.h.s. of (4.4a).
The field strength (3.3d) should be also modified by all the non-Abelian couplings:

1
I
J
I
J
K
G 3 [ B] LI J F[ A] + fI J K A A A .
(4.5)
3
The commutator of two covariant derivatives acting on A provides certain important
geometric quantity in our system:

1
[D , D ]A = R mn mn A + D[| D|] F i T i A
4

i I i
F I C
T ,
(4.6)
i I is defined by
where the function C
iJ
iJ

i J
(for Sp(n )),
i J = C g C g A
ij gC ij g A i j ij
C
(for Sp(1)),
iJ
C
0
(otherwise),
(4.7)
i J T i is an analog
which is analogous to the N = 2 case in 6D [16], or our combination C
ij
j
of PI i in the notation in [9]. The component C in (4.7) implies that all the terms with
i I , when we gauge G SO(2) Sp(n )

gT 2 VI in [12] should be replaced by T i C
np+1
[U (1)]
instead of SO(2) in [12]. Some illustrative examples of the
Sp(1) H
110
replacements of the terms in [12] are given by
i I ,
+gT 2 VI T i C
(4.8a)
+g VI + I ,
(4.8b)

i
i
i I LI ,
+ ge T2 VI LI e Ti C
(4.8c)
2 2
2 2
1
i
i I J
1
g 2 e2 VI VJ LI J C
(4.8d)
IC JL .
8
8
Needless to say, when the gauged group is truncated from G SO(2) Sp(n ) Sp(1)
[U (1)]np+1 back into SO(2) with the VI s as in [12], then all the r.h.s. in (4.8) go
H
back to their l.h.s. This can provide a good confirmation at various stages of computations,
in particular the invariance check of total action under supersymmetry. Due to the indefinite
metric involved, special care is needed for the contraction of the I -indices here, while the
ups/downs of the index i does not matter. Relevantly, we can define the covariant derivative
i I as
on C
i J C
i J + ij k A j C
k J ,
D C
(4.9)
so that

i I C
i I + g ij k A j C
k I + fIJK
A J C
i K
+ ij k D A j C
k I
D C

i I .
= D D C
(4.10)
To confirm the last equality, we need the relationship

i J

iK

k J
I C
= fI J K C
+ fI ik C
,
(4.11)
derived from the Lie derivatives
L I J I J J I = fI J K K ,
L I A i I A i I A i = fI ij A j .
(4.12)
Note that as in the 6D case in [16], there is no term with fJ K L A K needed in (4.9), in
order to be consistent with supersymmetry of the action. Other important corollaries with
are such as
these Cs
i I = I F i ,
D C
i I 0,
I D C
(4.13)
which have parallel structures as in the 6D case [16].
iJ in our system is the constraint required by

The most crucial relationship involving C
the supersymmetric invariance of the total action, needed for the consistency between the
two cosets SO(n, 1)/SO(n) and Sp(n , 1)/Sp(n ) Sp(1):
j I C
k J fI J K C
i K = 0.
F i I J ij k C
(4.14)
This constraint is required by the cancellation of -linear terms with the structure ( T i )
with one T i -generator sandwiched. This constraint is also analogous to Eq.(3.15) in [9],
111
or to Eqs. (2.21)(2.24) in [8]. The necessity of such a constraint is natural from the fact
that the vector fields A I in our system are both in the (n + 1)-representation of SO(n, 1)
and the adjoint representations of the gauged groups in G at the same time. And therefore
their mutual consistency, in particular, under supersymmetry requires such a constraint.
It is taken for granted that in (4.14), there are many trivially vanishing components for
each terms depending on the combination of the adjoint indices. For example, according
to (4.4), the structure constants fI J K vanishes identically for any U (1)-directions, or for
any mixed directions of different gauge groups. However, note that the first term in (4.14)
does not automatically vanish for such mixed directions. Our previous SO(2) gauging in
[12] also satisfies (4.14) trivially, because the last two terms vanish, while I g VI
makes the first term vanish, too.
The tensor Cab in (3.7) is also redefined in terms of fI J K by
ba .
ab fI J K La I Lb J LK = C
C
(4.15)
With these preliminaries, we now give our Lagrangian 3

1
1
1
e1 L2 = R D e4 G2
4
2
12

1
1
4
2

3

1
1
D
g D D ( )2
2
4
2

1 a

1
g D D D a
2
2

i
3i
a

+ V a D +
2
2

i
+ iV aA a D e + 2 LI F I
4 2

i 2
+ e
G
6 3

1
5
e2 6 G e2
G
24
72

1
1
2 2
2 6

i
i
e LI F I + e a a LI F I
12 2
4 2
1 2 a
1 2 a

+ e
a G + e
a G
24
24
a

i
1
+ e2
a LI F I e a a LI F I
6
4 2
3 We mention the errors in signatures of terms in our previous paper [12]. The sign errors in the g-linear
Lagrangian (4.2) and g-linear transformation rule (4.3) in [12] are now corrected in (4.16) and (4.17).
112

i I

i
i
I L + 1 e T i a C
I La I
+ e T i C
2 2
2

i I
i
2i i a
i
IL +
C I La I
+ e a T i a C
e T
2 2
6
i i I

i I
i
I L + 1 e T i C
IL
C
e T
6 2
6

1
+ 2ie a aA VaA I La I + e A a VaA I LI
2

2i A a
i
+ e VaA I LI + e a b Va B VbB D I LI
6
4 2

1
i
a b
1 2
iI
iJ
+ e Cab
+ e C C LI J e2 I J LI LJ ,
8
4
2 2
(4.16)
where the penultimate pair of the square brackets [ ] is for the terms at O(g),
while the
last pair is for the terms at O(g 2 ), where g is any minimal coupling constant for gaugings
iI , C
ab and I are all at O(g).

This is because C
Our Lagrangian (4.16)
among g, g or g.
yields an action S2 invariant under supersymmetry

Q e m = + m ,
i
Q = + ( ),
3

i
6 2

A i I
1 2
3 A
i

IL ,
+ e

G e T i C
18
2
3 2

i
1
1
Q A I = e LI ( ) + e LI ( ) + e a La I ,
6
2
2

i
Q B = +e2 ( [ ] ) + e2 ( ) 2LI J A[| I Q A|] J ,
3
1
i
3i A

Q A = e A LI F I + e2 A G
2
2 6
6 3

A i I
1
IL ,
+ e T i C
6

i
Q = + Va a ,
2
1
i
Q aA = e A LI a F I A V a
2 2
2
1 i A
i aI
+ e T C IL ,
2

Q = +iVaA A a ,
1
Q a = iV aA A D e B V aB I LI ,
(4.17)
2
113
Similarly to the SO(2)-gauging [12], the potential term is positive definite, except for
the term with LI LI (0) in LI J = LaI LJ a LI LJ [69,12]:
1
iI C
iJ LI J + 1 e2 I J LI LJ .
Vpot = + e2 C
(4.18)
8
4
As in Section 3, our Lagrangian L2 has the scaling invariance when g, g and g transform
like
g ec g,
iI ec C
iI ,
C
g ec g ,
I ec I ,
g ec g,
Cab e Cab ,
(4.19)
in addition to (2.5).
Similarly to the 6D case with the Sp(n ) Sp(1)-gauging [16], any subgroup of these
gauge groups can be also gauged consistently with supersymmetry, even though the details
of its process are skipped here. In such a case, the indices I , J , . . . and i, j, k are to be
replaced by the corresponding indices of such gauged subgroups. In particular, for the
SO(2) subgroup gauging out of the Sp(1) above, only the second direction 2 out of the
original indices i, j, k is relevant, so that we can use the notation such as T2 as in [12].
As the SO(2)-gauging described in [12] or in [69], we can combine the products of U (1)
groups, by introducing the constant couplings VI (I, J, . . . = 0, 1, 2, . . . , n), as a slight
generalization of the single SO(2) subgroup gauging. In any of these cases, our results
above are formally valid, and only the interpretation or the range of indices are changed.
5. Alternative N = 2 supergravity in singular 5D spacetime
As has been developed in [1,2] for the conventional N = 2 supergravity [6,7,9],
we can generalize our alternative N = 2 supergravity into singular 5D spacetime, as
supersymmetrization of RandallSundrum brane-world scenario [3].
As in our previous paper [12], we follow the prescription in [2] designed for the case
of Abelian SO(2) gauging for the singular 5D spacetime with the orbifold-type singularity
of S 1 /Z2 . However, since our present total gauged group is non-Abelian: G = SO(2)
[U (1)]np+1 which is much bigger than just SO(2), we need special
Sp(n ) Sp(1) H
care when applying the method in [2].
We start with fixing the bulk 5D spacetime action Sbulk before considering the
singularity. Mimicking the Abelian case [2], we first replace the original Sp(1)-gauging
coupling constant g everywhere in L2 by a spacetime-dependent real scalar field G(x),
and then introduce a fourth-rank antisymmetric tensor potential A , with a new term in
the Lagrangian [2]

1
5
5
SAG d x LAG d x
(5.1)
A G .

24
The reason we replace only g by G(x) is that this coupling is for the Sp(1) group that can
contain the SO(2) subgroup in our previous case [12] which is analogous to the Abelian
group in [2]. The scalar field G(x) has inherited the scale transformation property from
114
the coupling constant g ec g under the scaling transformation (2.5). Accordingly, for
the action SAG to be also invariant under this scale transformation, A should be also
rescaled as
G ec G,
A ec A
(5.2)
when other fields and constants are transforming like (2.5) and (4.19) except for g now
replaced by G.

The total 5D bulk action is now Sbulk S2 + SAG d 5 x (L2 + LAG ). Here S2
is no longer invariant under supersymmetry, but has terms proportional to G, which
is supposed to be cancelled by the variation of SAG [1,2]. There are eight sectors
contributing to such G-dependent terms out of L2 after the replacement g G(x):
(i) gravitino kinetic term, (ii) ( )D-Noether term, (iii) ( )G -Noether term,
(vi) ( )C-term,
(vii) ( )C(iv) ( )G -Noether term, (v) ( )C-term,
The terms (i)(iv) contribute, when a derivative D hits either

term, (viii) ( a )C-term.
some covariant derivatives or field strengths, after a partial integration of the contribution
Q D , while (v)(viii) terms contribute, when the derivative hits hatted quantities
iJ or I . All of these terms containing the derivative G, are therefore cancelled by the
C
appropriate supersymmetry transformation Q A in Q LAG . If we restrict ourselves
to the case of linear order supersymmetry transformation of Q A , 4 then we easily
see that only (v)(viii) terms contribute. For example, (iii) term contributes the cubic
combination ij k A i A j A k coming from the ChernSimons term in G , and such
terms are omitted from now on. After these considerations, we get

1
[| T i | ] C ij Lj
+2 2 e
Q (L2 |gG ) =
24

i
i
+ e T i a C ij La j + e T i C ij Lj
2
6

i
a
i i
+ e VaA L G,
(5.3)
2
up to cubic or higher order terms. Note that there are no hats on C ij and i here, and
Li is the I = i component of LI . Eq. (5.3) is supposed to be cancelled by the new
supersymmetry transformation rule Q A in Q LAG : 5

i
Q A = 2 2 e [| T i | ] C ij Lj e T i a C ij La j
2

ij j
i
i
i
e T C L e a VaA i Li
6
2
+ (quadratic terms).
(5.4)
4 The word linear here does not include the quantities C iJ or . This is because in the reduced case of
I
Abelian SO(2)-gauging, C ij is reduced to be a constant, while since C ij has the part A i j , and therefore it is
more convenient to regard the s as the same order as the Cs itself.

5 The previously-mentioned correction of sign errors in (4.2) in [12] leads to the corrected signs for the first
and third terms both in (5.3) and (5.4).
115
The fact that the Cs and s here have no hats is consistent with the scaling property
(5.2). As the standard first step of this prescription [2], we require Q G = 0, so that there
is no other contribution from Q LAG . Our previous result [12] can be recovered easily by
truncating a 0 and reducing C 22 = 1, C ij (otherwise) = 0. Now our action Sbulk is
invariant under (5.4), Q G = 0 and (4.17) with g G.
Since we are now dealing with the prescription in [2] originally designed for
Abelian gauging without hypermultiplets, applied to our non-Abelian gauging also with
hypermultiplets, it is better to confirm the closure of supersymmetry on the field A by
the commutator [Q (1 ), Q (2 )] = P (2 m 1 ) acting on A , where P (m ) implies
the translation operator. In what follows, we confirm this closure up to quadratic field level.
The linear terms in this commutator are composed of six sectors: F , G , , ,
2 - or 2 -terms. Here, the first four sectors work with no problem,
linear sectors, and C
while the -linear sector needs special care. To be more specific, we get

1
i i
1

(1 2 )L D
[Q (1 ), Q (2 )]A = e e
2

2ie 2 [| T i 1 F i j Lj D| ] ,
(5.5)
where the last term can be interpreted just as the usual desirable gauge transformation
of the type [ ] up to quadratic terms, while the first term needs special care. This
term is actually interpreted as an Sp(1) gauge transformation of A . Even though this
seems slightly bizarre at first sight, it can be easily understood, once we notice that the
-kinetic term is no longer Sp(1) invariant after the replacement g G(x). In fact, after
this replacement, (4.3c) is to be modified as

G D = I D I + i i G,
(5.6)
with the new effect of G, while all other equations in (4.3) are formally intact.
This results in the non-trivial contribution of the -kinetic term under the gauge
transformation G :

1
1
G eg g D D = e1 i i D G,
2
24
(5.7)
It is now clear that this contribution can be cancelled by an extra transformation G A
via LAG , such that
G A = e1 i i D .
(5.8)
In other words, when we identify i (1/2)(2 1 )Li in (5.5), then the first term in (5.5)
is absorbed into the Sp(1) gauge transformation.
Even though the result that the tensor potential field A is transforming under the
gauge group Sp(1) seems unnatural at first glance, this is nothing new in supergravity. In
fact, in [2] it was pointed out that the original action S2 is no longer R-invariant, i.e., in
our case Sp(1) non-invariant producing a quadratic terms in fermions after the replacement
116
g G(x). 6 Analogous situation can be found in GreenSchwarz mechanism in anomaly

cancellation in the usual formulation [10,20] or in the dual formulation [21], in which the
tensor field B or M1 6 transforms under Lorentz as well as gauge transformation, as
the zero-slope limit effect of superstring theory.
2 - and 2 -terms
Going back to our closure question, the only left over sector is the C
which turn out to be

1
i K Lj K + 2 i J Li LJ , (5.9)
[Q (1 ), Q (2 )]A C
2 = e2 C ij C
4
where (2 1 ). If this system is analogous to the Abelian case [2,12], these two
terms are supposed to be proportional to F , upon the use of G-field equation,
for the field strength of the potential A : F 5[ A ] . In fact, the G-field
equation is easily obtained as

1
i K Lj K + 2 i J Li LJ ,
F = e1 e2 C ij C
(5.10)
4
up to quadratic terms. After simple algebra, it is easy to show that (5.9) is desirably
proportional to F which is equivalent to the combination of the usual translation
P ( )A accompanied by a gauge transformation. This conclude the linear-order
closure of supersymmetry on A , which provides a non-trivial consistency check of
our system with Sbulk , in particular with non-Abelian gauged groups.
We next consider a possible brane action Sbrane to be added. To this end, and for the
reason to be clarified later, we truncate the hypermultiplets ( , a ), and we restrict the
gauged group to be SO(2) out of the Sp(1) isotropy group in the coset Sp(n , 1)/Sp(n )
Sp(1). We do not have to restrict other gauged groups in G, but it is only SO(2) out of the
Sp(1) group to be gauged.
We next assume that the branes are located at y x 5 = 0 and y = b > 0 in the 5th
dimension, requiring all the fields to obey the usual periodic boundary condition f (b) =
f (0) = f (b). Subsequently, we assign the parities under y y on the branes on all the
fields in our system, following [1,2]:

e m = e5 (5) = A5 I = ( ) = = (A )
e5
= ( ) = () = +1,

= e (5) = A I = (G) = (5 ) = a
(5.11a)
= () = 1,
(5.11b)
where e m denotes the 4D part of the fnfbein. The parity for an arbitrary bosonic () or
fermionic ( ) field is defined by [1,2]
(y) = ()(y),
(y) = i( )5 T 2 (y).
(5.12)
Here the real constant = 1 reflects the signature ambiguity, but its sign should be
common to all the fermions [1,2].
6 These terms did not matter in our treatment, because we are looking into only linear terms in the
transformation rule of Q A .
We now consider the brane action

1
1 5
5
I
Sbrane 2h d x [(y) (y b)] a ee VI L +
A
24
2

d 5 x Lbrane ,
117
(5.13)
where h, a are real constants with |a| = 1, and e is the 4D part of the determinant of the
fnfbein, while 5 is the = 5 component of . The exponential factor e is
needed for invariance of Sbrane under supersymmetry, as will be seen. 7 In order for the
action Sbrane to be invariant under the scaling transformation (5.2), the constant h should
also be rescaled as
h hc .
(5.14)
when other fields and constants are transforming like (2.5) and (4.17), except for g replaced
by G.
The action Sbrane modifies the A -field equation from the original one G = 0 into
5 G(x) = 2h[(y) (y b)],
G = 0 (for = 5).
The solution for this field equation is [1,2]

+h (for 0 < y < +b),
G(x) = G(y) = h(y) =
h (for b < y < 0).
(5.15)
(5.16)
This gives the desirable kinks for the SO(2) coupling constant G(x) [1,2].
We now take the variation of Q Sbrane under supersymmetry (4.15) and (5.4) for
Q A :

1
Q Lbrane = 2h[(y) (y b)] a ee ia 1 5 T 2 VI LI
2

i
2h[(y) (y b)] a e i a a 1 5 T 2 a VI La I
2

i
2h[(y) (y b)] a e i( ) a 1 5 T 2 VI LI . (5.17)
6
Comparing these three lines with (5.12), we see that if
a = = 1,
(5.18)
then (5.17) is simplified to be

Q Lbrane
1
= 2h[(y) (y b)] a e(y)e (y) (y) m [m (y) m (y)]VI LI (y)
2

i
2h[(y) (y b)] a e(y)e (y) (y) a (y) a (y) VI La I (y)
2
7 This factor was not considered in [12].
118
i
2h[(y) (y b)] a e(y)e (y) (y)[(y) (y)]VI LI (y),
(5.19)
6

where each line vanishes after the dy-integration, under the periodic boundary condition
f (b) = f (0) = f (b) for an arbitrary field f (y) in (5.19). This concludes the proof of the
invariance
Q Sbrane = 0,
(5.20)
and therefore that of the total action Stotal S2 |gG + SAG + Sbrane for the SO(2)-gauging
in the absence of hypermultiplets.
Let us briefly comment on the difficulty of the brane mechanism with the gauging Sp(1)
or with hypermultiplets. The difficulty with the Sp(1)-gauging is that we do not have a
good analog of VI LI we can use as the first term in Sbrane as an invariant quantity. This is
because Q A in (5.4) with general T i matrices with general index i cannot cancel the
) + considering the parity (5.12). As for the inclusion of the
variation Q e = e(
hypermultiplets, the a -dependent term in Q A yields

1
1

A = e 5 a VaA 2 L2 ,
Q
(5.21)
24
2
a
via Sbrane . The trouble here is that there seems to be no invariant Lagrangian that will
cancel (5.21). For example, a quantity like 2 L2 is not appropriate, because of its lack
of gauge invariance under (4.3). From these viewpoints, we seem to have no generalization
of the brane action Sbrane (5.13), when Sp(1) is gauged and/or the hypermultiplets are
included.
The brane action Sbrane we gave here is supposed to be the simplest one based on [2],
among other potentially possible Lagrangians invariant under local supersymmetry in
singular 5D spacetime [1]. However, we stress that our alternative N = 2 supergravity
in 5D is equally applicable to these formulations, as well.
6. Concluding remarks
In this paper, we have completed the non-Abelian gauging of our alternative N = 2
supergravity to n copies of vector multiplets in 5D, and n copies of hypermultiplets, with
a simpler coupling structure compared with the conventional supergravity [69], up to
quartic fermion terms in the action. Our result is the combination of considerable works on
supergravity couplings in the past, such as vector multiplet couplings in 9D case with
the scalars forming the coset SO(n, 1)/SO(n) [13,14], together with the scalars in the
hypermultiplets forming the quaternionic Khler manifold in N = 2 supergravity in 4D
[15] and in 5D [69] as well as in 6D [16].
As in 9D [13], the scalars in the vector multiplets form the coordinates of the -model
for the non-Jordan family scalar coset H n SO(n, 1)/SO(n), and the vector fields with
the total number n + 1 form the (n + 1)-representation of SO(n, 1), while the gaugini a
form the n-representation of SO(n). The scalars in the hypermultiplets form the -model
on the quaternionic Khler manifold Sp(n , 1)/Sp(n ) Sp(1).
119
Our result is valid for any arbitrary gauge groups of the type G SO(2) Sp(n )
[U (1)]np+1 for any arbitrary group H
with dim H
= p n (2n +1)4, and
Sp(1) H

Sp(n )Sp(1) are the isotropy groups of the coset Sp(n , 1)/Sp(n )Sp(1), with a peculiar
potential term in the Lagrangian. Accordingly, we have obtained a crucial constraint
(4.14) required by consistency between the two different cosets under supersymmetry.
iJ -functions, whenever a nonThis constraint relates the Sp(1) curvature F i and the C
Abelian group in the isotropy groups Sp(n ) Sp(1) is gauged. Moreover, the isotropy
groups Sp(n ) Sp(1) can be also reduced into their subgroups, e.g., Sp(1) SO(2),
where Sp(1) Sp(n ), SO(2) Sp(1). This a generalization of our previous paper [12],
in which we gauged only the SO(2) subgroup of Sp(1) Sp(n ) Sp(1). Therefore,
by adjusting the parameters n, n and p for dimensions appropriately, our results in
this paper are considerably general, and cover a wide range of combinations of gauged
groups.
Since there are two non-trivial coset structures SO(n, 1)/SO(n) and Sp(n , 1)/Sp(n )
Sp(1) present in our system, our vector fields A I for gaugings are both in the (n + 1)representation of SO(n, 1) and in the adjoint representations of the gauged groups in G
at the same time. The mutual consistency of these two structures under supersymmetry
requires the constraint (4.14) which corresponds to analogous equations in [8,9].
Even though we did not perform explicitly in this paper, we can also combine any
Abelian factor groups [U (1)]np+1 , by introducing constant vectors VI , as has been done
in [69,12]. This provides another freedom for practical applications for phenomenological
model building.
We have also generalized this result to the case of singular 5D spacetime for the case
of SO(2)-gauging as a supersymmetric RandallSundrum brane-world scenario [3] similar
to the conventional 5D supergravity [1,2]. We have applied the prescription in Ref. [2] in
order to confirm the supersymmetry of our brane action with the singularity of the type
S 1 /Z2 . We have also seen some difficulty, when the gauged group is Sp(1) larger than its
subgroup SO(2), or when the hypermultiplets are present. As far as the formulation for
singular spacetime is concerned, there seems to be no fundamental difference between our
alternative supergravity and the conventional one [69].
For some readers who wonder why our larger supergravity multiplet [12] has never
been studied as a special case in the conventional and exhaustively studied formulation
[69], we repeat the following points already given in [12]: the original result in [6] was
presented before the discovery of the importance of superstring in 1984 [10], so that there
was no strong motivation to include the dilaton or antisymmetric tensor fields, which
are important NS fields in terms of superstring language. To put it differently, it is only
superstring [10] or M-theory [18] that motivates the peculiar couplings of dilaton and
antisymmetric tensor to supergravity, as we have done in the present paper.
Even though we have stressed the difference of our formulation from other general
matter couplings in [69], it is fair to point out some similarities. For example, we expect it
possible to generalize the number of the additional tensor fields B in addition to the one
in the supergravity multiplet with a -model structure similar to that presented in [79].
However, we also emphasize that our antisymmetric tensor B is still to be distinguished
120
from these additional ones which always appear in pairs [79]. This situation is analogous
to our dilaton as another NS field separated from other scalar fields.
We believe that our result in this paper will be of great help for the study of the Randall
Sundrum brane-world scenario [3] associated with superstring theory and M-theory.
Acknowledgements
We are grateful to J. Bagger, A. Ceresole, G. DallAgata, S.J. Gates Jr., M. Gnaydin,
R. Kallosh, E. Sezgin, P.K. Townsend and M. Zucker for helpful discussions.
Appendix A. Notations and conventions

In this appendix, we deal with important notations and conventions in our paper. Typically, our indices , , . . . = 0, 1, . . . , 4 are for the curved world indices, and m, n, . . . =
(0), (1), . . . , (4) are local Lorentz indices with the metric (mn ) = diag(, +, +, +, +),
and 012345 = +1. Most frequently used relevant relations are such as
m1 m5n r1 rn r1 rn n1 n5n = (n!)[(5 n)!][m1 [n1 m5n ] n5n ] ,

m1 m5n n1 nn
n1 nn = i(1)
n(n1)/2
(n!)
m1 m5n
(A.1a)
(A.1b)
As the most important notational preparation, we first deal with the fermions in our
5D with the signature (, +, +, +, +). In this paper, we refer the reader to the general
description of arbitrary fermions in diverse spacetime dimensions [17]. Using the notation
in [17], we start with the number of space and time dimensions together with the parameters
and that control the properties of fermions [17]:
s = 4,
t = 1,
= 1,
= 1,
(A.2)
We next define the complex conjugations by [17]

A = AB BB ,
B = B 1 BA A ,
(A.3)
with the Sp(1) metric AB = BA [17,22]. Accordingly [17],

m = Am A1 ,
A (0),
m = +(0)m (0) = +Am A,

B T = B,
mT = Cm C 1 ,
A = AAA1 = A,
((0) )2 = I,
= A = (0),
C T = C,
B B = I,
C = BA.
(A.4)
Relevantly, we have the fermionic quadratic combinations [17]:

A A

= A A = A A A = A A A = A A A

= A A = AB ( A B ) = AB ( B A ) = A A .
(A.5)
121
Note that we have taken into account a sign error 8 in Eq. (6) in [17] that affects the first
hand side. In a more general case for 0 n 5, the hermitian conjugation works like [17]
A

m1 mn A = A (m1 mn ) A

= A A[mn | A1 A|mn1 | A1 A|m1 ] A1 A A

= (1)n+n(n1)/2 A Am1 mn A1 A A

= (1)n+n(n1)/2 A m1 mn A
= (1)n(n+1)/2 AB ( B m1 mn A )
2 /2+n/2+n2 /2n/2
= (1)n
AB ( A m1 mn B )

2
= (1)n +1 A m1 mn A .
(A.6)
Hence, omitting the explicit A-indices for contractions as in Section 2, we get

(
m1 mn ) = (1)n+1 (
m1 mn ).
(A.7)
Typical examples are

()
= (),
(
m ) = +(
m ),
(
mn ) = (
mn ),
etc. (A.8)
In other words, any combination (

m1 mn ) with an even number of gammas need a pure
imaginary unit i in front to be an hermitian expression, while a combination with an odd
number of gammas is already hermitian.
In (A.6), we have used the flipping property for 0 n 5
A m mn

1
B = (1)n+1 (1)(1n)(2n)/2 B m1 mn A

= (1)n(n1)/2 B m1 mn A .
(A.9)
Therefore, we have

1 m1 mn 2 1A m1 mn 2A

2A m1 mn 1A
=

+ 2A m1 mn 1A
(for n = 0, 1, 4, 5),
(for n = 2, 3).
(A.10)
References
[1] R. Altendorfer, J. Bagger, D. Nemeschansky, Supersymmetric RandallSundrum scenario, hepth/0003117;
T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141, hep-ph/0003129;
N. Alonso-Alberca, P. Meessen, T. Ortin, Phys. Lett. B 482 (2000) 400, hep-th/0003248;
A. Falkowski, Z. Lalak, S. Pokorski, Phys. Lett. B 491 (2000) 172, hep-th/0004093;
A. Falkowski, Z. Lalak, S. Pokorski, Four-dimensional supergravities from five-dimensional
brane worlds, hep-th/0102145;
M. Zucker, Supersymmetric brane world scenarios from off-shell supergravity, hep-th/0009083;
S. Nojiri, S.D. Odintsov, Supersymmetric new brane-world, hep-th/0102032.
8 We acknowledge E. Sezgin for informing about this error.
122
[2] E. Bergshoeff, R. Kallosh, A. Van Proyen, JHEP 0010 (2000) 033, hep-th/0007044.
[3] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690;
J. Lykken, L. Randall, JHEP 0006 (2000) 014.
[4] G. t Hooft, Dimensional reduction in quantum gravity, in: A. Aly, J. Ellis, S. Randjbar-Daemi
(Eds.), Salam Festschrift, World Scientific, Singapore, 1993, gr-qc/9310026;
L. Susskind, J. Math. Phys. 36 (1995) 6377;
L. Susskind, E. Witten, The holographic bound in anti-de Sitter space, hep-th/9805114.
[5] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253;
S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105.
[6] M. Gnaydin, G. Sierra, P.K. Townsend, Phys. Lett. B 133 (1983) 72;
M. Gnaydin, G. Sierra, P.K. Townsend, Phys. Lett. B 144 (1984) 41;
M. Gnaydin, G. Sierra, P.K. Townsend, Phys. Rev. Lett. 53 (1984) 332;
M. Gnaydin, G. Sierra, P.K. Townsend, Nucl. Phys. B 242 (1984) 244;
M. Gnaydin, G. Sierra, P.K. Townsend, Nucl. Phys. B 253 (1985) 573.
[7] M. Gnaydin, M. Zagermann, Phys. Rev. D 62 (2000) 044028, hep-th/0002228;
M. Gnaydin, M. Zagermann, Nucl. Phys. B 572 (2000) 131, hep-th/9912027.
[8] M. Gnaydin, M. Zagermann, Gauging the full R-symmetry group in five-dimensional N = 2
YangMills/Einstein/tensor supergravity, hep-th/0004117.
[9] A. Ceresole, G. DallAgata, Nucl. Phys. B 585 (2000) 143, hep-th/0004111.
[10] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Vols. I and II, Cambridge Univ.
Press, 1987.
[11] E. Cremmer, B. Julia, J. Scherk, S. Ferrara, L. Girardello, P. van Nieuwenhuizen, Nucl. Phys.
B 147 (1979) 105;
E. Cremmer, S. Ferrara, L. Girardello, A. van Proyen, Nucl. Phys. B 212 (1983) 413.
[12] H. Nishino, S. Rajpoot, Phys. Lett. B 502 (2001) 246.
[13] S.J. Gates Jr., H. Nishino, E. Sezgin, Class. Quant. Grav. 3 (1986) 21.
[14] M. Awada, P.K. Townsend, M. Gnaydin, G. Sierra, Class. Quant. Grav. 2 (1985) 801.
[15] J. Bagger, E. Witten, Nucl. Phys. B 222 (1983) 1.
[16] H. Nishino, E. Sezgin, Phys. Lett. B 144 (1984) 187;
H. Nishino, E. Sezgin, Nucl. Phys. B 278 (1986) 353;
H. Nishino, E. Sezgin, Nucl. Phys. B 505 (1997) 497.
[17] A. Salam, E. Sezgin (Eds.), Supergravity in Diverse Dimensions, North-Holland/World
Scientific.
[18] For reviews of M-theory, see, e.g., W. Taylor IV, The M(atrix) model of M-theory, Lectures for
NATO school Quantum Geometry, Iceland 1999, hep-th/0002016;
T. Banks, TASI lecture note on Matrix theory, hep-th/9911068, and references therein.
[19] E. Bergshoeff, I.G. Koh, E. Sezgin, Phys. Rev. D 32 (1985) 1353.
[20] M. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117.
[21] S.J. Gates Jr., H. Nishino, Phys. Lett. B 157 (1985) 157.
[22] T. Kugo, P.K. Townsend, Nucl. Phys. B 211 (1983) 157.

General matter coupled N = 4 gauged supergravity

in five dimensions
G. DallAgata a , C. Herrmann b , M. Zagermann b
a Institut fr Physik, Humboldt-Universitt Berlin, Invalidenstrasse 110, D-10115 Berlin, Germany
b Fachbereich Physik, Martin-Luther-Universitt Halle-Wittenberg, Friedemann-Bach-Platz 6,
D-06099 Halle, Germany

Received 21 March 2001; accepted 19 July 2001
Abstract
We construct the general form of matter coupled N = 4 gauged supergravity in five dimensions.
Depending on the structure of the gauge group, these theories are found to involve vector and/or
tensor multiplets. When self-dual tensor fields are present, they must be charged under a onedimensional Abelian group and cannot transform non-trivially under any other part of the gauge
group. A short analysis of the possible ground states of the different types of theories is given. It
is found that AdS ground states are only possible when the gauge group is a direct product of a
one-dimensional Abelian group and a semi-simple group. In the purely Abelian, as well as in the
purely semi-simple gauging, at most Minkowski ground states are possible. The existence of such
Minkowski ground states could be proven in the compact Abelian case. 2001 Elsevier Science B.V.
All rights reserved.
1. Introduction
The last few years have witnessed a renewed intense interest in five-dimensional gauged
supergravity theories. This interest was largely driven by the study of the AdS/CFT correspondence [14], but also by recent attempts to construct a supersymmetric version of the
RandallSundrum (RS) scenario [514].
One especially fruitful direction in the study of the AdS/CFT correspondence was its
generalization to include certain four-dimensional quantum field theories with non-trivial
renormalization group (RG) flows. The best-studied examples of such RG-flows arise from
relevant perturbations of the d = 4, N = 4 super YangMills (SYM) theory, and were
mapped to domain wall solutions of d = 5, N = 8 gauged supergravity (see [1517] for
the first explicit examples).
E-mail addresses: dallagat@physik.hu-berlin.de (G. DallAgata), herrmann@physik.uni-halle.de
(C. Herrmann), zagermann@physik.uni-halle.de (M. Zagermann).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 6 7 - 4
124
G. DallAgata et al. / Nuclear Physics B 612 (2001) 123150
In [17], an RG-flow interpolating between two supersymmetric conformal fixed points

of a mass deformed version of d = 4, N = 4 SYM was studied. The corresponding domain
wall solution was constructed after a truncation of the d = 5, N = 8 supergravity theory to
a particular N = 4 subsector. This subsector describes N = 4 supergravity coupled to two
N = 4 tensor multiplets with scalar manifold M = SO(1, 1) SO(5, 2)/[SO(5) SO(2)]
and gauge group SU(2) U (1). In the same paper this particular N = 4 theory was
also conjectured to be the holographic dual of the common sector of all d = 4, N = 2
superconformal gauge theories based on quiver diagrams [18,19]. The holographic duals
of these quiver gauge theories were identified in Refs. [19,20] as IIB string theory on
AdS5 (S 5 / ), where is a discrete subgroup of SU(2) SU(4) of the ADE type.
The KaluzaKlein spectrum of IIB string theory on AdS5 (S 5 / ) was studied in
Ref. [21], where the IIB supergravity modes were fit into SU(2, 2|2) multiplets, and in
Ref. [22], which also considered the = Zn -twisted string states. These twisted states live
on AdS5 S 1 , and were found to correspond to five-dimensional, N = 4 self-dual tensor
multiplets of SU(2, 2|2) [22].
It has therefore been suggested [17,23] that five-dimensional, N = 4 gauged supergravity coupled to vector and/or tensor multiplets might encode some non-trivial information
on the N = 2 quiver theories even though a ten-dimensional supergravity description of
the corresponding orbifold theory is not available. In particular, it has been conjectured
in [23] that certain aspects of the flow from the S 5 /Z2 to the T 1,1 compactification of
IIB string theory [24] might be captured by a 1/4 BPS domain wall solution of a suitable
N = 4 gauged supergravity theory with tensor multiplets. The tensor multiplets are crucial
for such a flow because they host the supergravity states dual to the twisted gauge theory
operators that drive this flow [23].
Unfortunately, too little was known about five-dimensional, N = 4 gauged supergravity
coupled to an arbitrary number of matter (i.e., vector- and tensor-) multiplets in order to
further investigate the corresponding gravity description. In fact, only a SU(2) U (1)
gauging of pure N = 4 supergravity [25] and a peculiar SU(2) gauging of N = 4
supergravity coupled to vector multiplets [26] were available so far.
It is the purpose of this paper to close this gap in the literature and to construct the
general matter-coupled five-dimensional, N = 4 gauged supergravity.
This, with [2730] for the N = 2 case and [31,32] for the N = 8 case, completes also
the description of all standard gauged supergravity theories in d = 5.
These theories should also help to clarify whether the different no-go theorems that have
been put forward against a smooth supersymmetrization of the RS scenario [68,14,33] can
be extended to the N = 4 sector as well. As for the (more successful) supersymmetrizations
based on singular brane sources [913], the theories constructed in this paper allow for the
possibility to confine N = 2 supergravity theories on the brane, as it is the self-dual tensor
fields in the bulk that were shown to circumvent the no photons on the brane theorems
[34,35].
The paper is organized as follows. All the gauged N = 4 supergravity theories we
will construct can be derived from the ungauged Maxwell/Einstein supergravity theories
(MESGT) studied in [26]. Section 2.1, therefore, first recalls the relevant properties of these
125
ungauged theories. In Section 2.2, then, we will take a closer look at the global symmetries
of these theories and analyze to what extent these global symmetries can be turned into
local gauge symmetries. This discussion will reveal some interesting differences to the
N = 2 and the N = 8 cases and also results in a rather natural way to organize the rest
of the paper: Section 3 discusses the general gauging of an Abelian group, which turns
out to require the introduction of tensor multiplets. In Section 4, we will then construct
the combined gauging of a semi-simple group and an Abelian group. The resulting scalar
potentials are then briefly analyzed in Section 5, which also contains the reductions to
various relevant special cases previously considered in the literature. In the last section, we
draw some conclusions and list a few open problems.
2. Ungauged Maxwell/Einstein supergravity

2.1. General setup
The starting point of our construction is the ungauged MESGT of Ref. [26] which
describes the coupling of n vector multiplets to N = 4 supergravity. 1 Our spacetime
conventions coincide with those of Ref. [26] and are further explained in Appendix A.
The field content of the N = 4 supergravity multiplet is
m i ij

e , , A , a , i , ,
(2.1)
ij
i.e., it contains one graviton e m , four gravitini i , six vector fields (A , a ), four
spin 1/2 fermions i and one real scalar field . Here, /m are the five-dimensional
Einstein/Lorentz indices and the indices i, j = 1, . . . , 4 correspond to the fundamental
representation of the R-symmetry group USp(4) of the underlying N = 4 Poincar
ij
superalgebra. The vector field a is USp(4) inert, whereas the vector fields A transform
in the 5 of USp(4), i.e.,
ji
Aij
= A ,
Aij
ij = 0,
(2.2)
with ij being the symplectic metric of USp(4). In the following, we will sometimes
make use of the local isomorphism SO(5)
= USp(4) to denote SO(5) representations using
USp(4) indices.
An N = 4 vector multiplet is given by

A , i , ij ,
(2.3)
where A is a vector field, i denotes four spin 1/2 fields, and the ij are scalar fields
transforming in the 5 of USp(4).
Coupling n vector multiplets to supergravity, the field content of the theory can be
summarized as follows
m i I

e , , A , a , i , ia , , x .
(2.4)
1 For ungauged five-dimensional supergravity theories, vector and tensor fields are Poincar dual, and we
therefore do not have to distinguish between vector and tensor multiplets at this level.
126
Here, a = 1, . . . , n counts the number of vector multiplets whereas I = 1, . . . , (5 + n)

ij
collectively denotes the A and the vector fields of the vector multiplets. Similarly,
x = 1, . . . , 5n is a collective index for the scalar fields in the vector multiplets. We will
further adopt the following convention to raise and lower USp(4) indices:
T i = ij Tj ,
Ti = T j j i ,
(2.5)
whereas a, b are raised and lowered with ab .

As was shown in [26], the manifold spanned by the (5n + 1) scalar fields is
M=
SO(5, n)
SO(1, 1),
SO(5) SO(n)
(2.6)
where the SO(1, 1) part corresponds to the USp(4)-singlet of the supergravity multiplet.
The theory has therefore a global symmetry group SO(5, n) SO(1, 1) and a local
composite SO(5) SO(n) invariance.
The coset part of the scalar manifold M can be described in two ways: one can either,
as in (2.4), choose a parameterization in terms of coordinates x , where the metric gxy on
the coset part of M is given in terms of SO(5) SO(n) vielbeins by (cf. Appendix B):
1 ij a a
,
gxy = fx fyij
4
(2.7)
such that the kinetic term for the scalar fields takes the typical form 12 gxy x y of a
non-linear sigma-model. This way of describing M is particularly useful for discussing
geometrical properties of the theory.
As an alternative description, one can use coset representatives LI A where I denotes a
G = SO(5, n) index, and A = (ij, a) is a H = SO(5)SO(n) index. Denoting the inverse of
LI A by LA I (i.e., LI A LB I = BA ), the vielbeins on G/H and the composite H-connections

are determined from the 1-form:
L1 dL = Qab Tab + Qij Tij + P aij Taij ,
(2.8)
where (Tab , Tij ) are the generators of the Lie algebra h of H, and Taij denotes the
generators of the coset part of the Lie algebra g of G. More precisely,
Qab = LI a dLI b
and Qij = LI ik dLIk j
(2.9)
are the composite SO(n) and USp(4) connections, respectively, and

1 aij
P aij = LI a dLI ij = fx d x
2
(2.10)
describes the spacetime pull-back of the G/H vielbein. Note that Qab
is antisymmetric in
ij
the SO(n) indices, whereas Q is symmetric in i and j . This second way of parameterizing
the scalar manifold is particularly useful to exhibit the action of the different invariance
groups as clearly as possible. We will make this choice in what follows for the construction
of the gauged theories. Eqs. (2.10) and (B.8)(B.11) can be used to easily switch between
both formalisms. Appendix B contains more details on the geometry of G/H.
127
The Lagrangian of the ungauged MESGT reads [26]:

1
1
1
1
I
F J 4 G G
e1 L = R i D i 2 aIJ F
2
2
4
4
1
1
1
1
( )2 iD
/ i iaD
/ ai Paij Paij
2
2
2
2
3
i
ij I
L F
i i + i ia j Pij a +
i j
I
2
6
1
1
I ai
i 2 i i G
LaI F
4
2 6
5i
i
ij I
+ 2 i i G L F
i j
I
12
24 2
i
i
I ia
i 2 G ia ai
LaI F
2 3
8 2

i
i
ij I a a
ij I
+ L F i j L F
i j + 2 i j
I
I
4
4

i
2 G i i + 2 i i
8 2
2 1
I
J
e CIJ / F
+
(2.11)
F
a + e1 L4f ,
8
and the supersymmetry transformation laws are given by 2
1
em = i m i ,
2

i
I
4 j
i = D i + LIij F
6

i
+ 2 G 4 i + 3-fermion terms,
12 2
i
1
3
I
LIij F
i = / i +
j 2 G i ,
2
6
2 6
1
I
a
ai = iPij
j LaI F
i + 3-fermion terms,
4
AI = I ,
1
i
a = 2 i i 2 i i ,
6
2 2
i i
= i ,
2

1
ij
j]
L = iLaI k[i l ij kl k la ,
I
4
LaI = iLIij i j a
2 We use the following (anti)symmetrization symbols: (ij ) 1 (ij + j i), [ij ] 1 (ij j i).
2
2
(2.12)
128
with
1
1
I 1 LIij i j i 1 LIij i j + LIa 1 i ai .

2
3
(2.13)

I
F
= AI AI ,
(2.14)
Here,
G = ( a a ),
are the Abelian field strengths of the vector fields, whereas the scalar field in the
supergravity multiplet is parameterized by
= e/
(2.15)
Moreover, the covariant derivative, D , that acts on the fermions is given by

b
D ai = ai + Qi j aj + Qab
i ,
(2.16)
with being the Lorentz- and spacetime covariant derivative.

Supersymmetry imposes
ij
I
aIJ = L LJij + LaI LaJ ,
ij
I
CIJ = L LJij LaI LaJ ,
(2.17)
where aIJ acts as a metric on the I, J indices:
LA
= aIJ LJ A .
I
(2.18)
The symmetric tensor CIJ turns out to be constant, and in fact is nothing but the SO(5, n)
metric.
Supersymmetry also imposes a number of additional algebraic and differential relations
on the LA , which are listed in Appendix B. Actually, the requirement of invariance under
I
supersymmetry heavily constrains the form of the possible couplings and, as we show
in Appendix C using the superspace language, if one does not consider non-minimal
couplings, those presented here already use all the freedom allowed by the N = 4
superalgebra.
2.2. The global symmetries and their possible gaugings
The above ungauged Maxwell/Einstein supergravity theories are subject to several
global and local invariances. Apart from supersymmetry, the local invariances are 3
a local composite USp(4) SO(n) symmetry
3 In large parts of the literature on gauged supergravity, none of these two local invariances is referred to
as a gauge symmetry: the local composite Usp(4) SO(n) symmetry is not based on physical vector fields,
and the Maxwell-type invariance is, in this context, more viewed as a special case of a more general invariance
under C (p) C (p) + d(p1) for arbitrary p-form fields C (p) and (p 1)-forms (p1) . The term gauged
supergravity, by contrast, is used when some physical vector fields of a (usually N -extended) supergravity theory
couple to other matter fields via gauge covariant derivatives. In most cases, the gauge symmetry of such a theory
reduces to a global symmetry of an ungauged supergravity theory when the gauge coupling is turned off.
129
a Maxwell-type invariance of the form 4
AI AI + I ,
(2.19)
a a + .
(2.20)
In addition, the ungauged theories of Section 2.1 are invariant under global SO(1, 1)
SO(5, n) transformations, which exclusively
act on the vector fields, the coset representa
tives LIA , and the scalar field = e/ 3 according to
I
AJ ,
SO(5,n) AI = r M(r)
J
I
LJ ,
SO(5,n) LIA = r M(r)
J A
SO(5,n) = 0,
SO(5,n) a = 0
(2.21)
and
SO(1,1)AI = AI ,
SO(1,1)LIA = 0,
SO(1,1) = ,
SO(1,1)a = 2a ,
(2.22)
where M I so(5, n), and r (r = 1, . . . , dim(SO(5, n))) and are some infinitesimal
(r)J
parameters.
We will now analyze to what extent this global SO(1, 1) SO(5, n) invariance can
be turned into a local (i.e., YangMills-type) gauge symmetry by introducing minimal
couplings of some of the vector fields, a process commonly referred to as gauging.
The first thing to notice is that the SO(1, 1) factor cannot be gauged, as all the vector
fields transform non-trivially under it (i.e., none of the vector fields in the theory could
serve as the corresponding (neutral) Abelian gauge field). Thus, we can restrict ourselves
to gaugings of subgroups K SO(5, n).
The vector fields AI and the coset representatives LIA transform in the (5 + n) of
SO(5, n); all other fields are SO(5, n)-inert (cf. Eqs. (2.21)). Hence, any gauge group
K SO(5, n) has to act non-trivially on at least some of the vector fields AI (as well
as on some of the coset representatives LIA ).

Having physical applications in mind, we will restrict our subsequent discussion to
gauge groups K that are either
(i) Abelian or
(ii) semi-simple or
(iii) a direct product of a semi-simple and an Abelian group.
4 Note that it is essential for this symmetry to hold that the C
be constant.
IJ
130
2.2.1. K is Abelian
Let us first assume that K is Abelian. As we mentioned above, K has to act non-trivially
on at least some of the vector fields AI . In addition to such non-singlets of K, there
might also be vector fields among the AI that are K-inert. Thus, in general, we have a
decomposition of the form
(5 + n)SO(5,n) singlets(K) non-singlets(K).
(2.23)
Let us split the vector fields AI accordingly:

AI = AI , AM
,
where the indices I, J, . . . label the K-singlets, and M, N, . . . denote the non-singlets.
As was first pointed out in [31,32] for the N = 8 theory, the presence of such nonsinglet vector fields poses a problem for the supersymmetric gauging of K and requires the
dualization of the charged vectors into self-dual [36] tensor fields.
As we will explain in Section 3, this conversion of the AM
to self-dual tensor fields
M is achieved in practice by simply replacing all field strengths F M by B M and by
B
adding a kinetic term of the form LBdB = B dB to the Lagrangian. The derivative in
LBdB then turns out to be automatically K-covariantized by the B B a-term in the
Lagrangian, which originates from the F F a term in the ungauged Lagrangian (2.11)
M B M . This has an important consequence: only the SO(5, n)
upon the replacement F
M for this kind of gauging, and, consinglet vector field a can couple to the tensor fields B
sequently, any Abelian gauge group K can be at most one-dimensional, i.e., it can be either
SO(2) or SO(1, 1) with gauge vector a . Note that the converse is also true: if tensor fields
have to be introduced in order to gauge a (not necessarily Abelian) subgroup K SO(5, n),
these tensor fields can only be charged with respect to a one-dimensional Abelian subgroup
of K. This is an interesting difference to the N = 8 and the N = 2 theories, where the tensor fields can also transform in non-trivial representations of non-Abelian gauge groups K
[28,31,32].
We will discuss the gauging of an Abelian group K in Section 3.
2.2.2. K is semi-simple
Let us now come to the case when the gauge group K SO(5, n) is semi-simple. In that
case, some of the vector fields AI of the ungauged theory have to transform in the adjoint
representation of K so that they can be promoted to the corresponding YangMills-type
gauge fields. Put another way, the (5 + n) of SO(5, n) has to contain the adjoint of K as a
sub-representation. A priori, one would therefore expect the decomposition
(5 + n)SO(5,n) adjoint(K) singlets(K) non-singlets(K).
(2.24)
Just as for the Abelian case, any non-singlet vector fields outside the adjoint of K would
M . On the other hand, we already found
have to be converted to self-dual tensor fields B
that, due to the peculiar structure of the ChernSimons term in the ungauged theory (2.11),
131
only the vector field a could possibly couple to such tensor fields. As one vector field
alone can never gauge a semi-simple group, we are led to the conclusion that the gauging
of a semi-simple group K cannot introduce any tensor fields, and we can restrict ourselves
to the case
(5 + n)SO(5,n) adjoint(K) singlets(K).
(2.25)
The vector fields in the adjoint will then serve as the K gauge fields, whereas the singlets
are mere spectator vector fields.
Let us conclude this subsection with a rough classification of the possible compact semisimple gauge groups K. Obviously, a compact gauge group K has to be a subgroup of the
maximal compact subgroup, SO(5) SO(n) SO(5, n). Furthermore, K can be either a
subgroup of the SO(5) factor or a subgroup of the SO(n) factor or a direct product of both
of these.
If K SO(5), it can only be the standard SO(3)
= SU(2) subgroup of SO(5), because
this is the only semi-simple subgroup of SO(5) with the property that the 5 of SO(5)
decomposes according to (2.25).
If K SO(n), the fundamental representation of SO(n) has to contain the adjoint of
K (as well as possibly some K-singlets). However, the adjoint of any compact semisimple group K can be embedded into the fundamental representation of SO(n) as long
as dim(K) n (One simply has to take the positive definite CartanKilling metric as the
SO(dim(K)) metric.) Thus, any compact semi-simple group K can be gauged along the
above lines as long as dim(K) n.
An obvious combination of the previous two paragraphs finally covers the case when K
is a direct product of an SO(5)- and an SO(n)-subgroup.
We will not write down the gauging of a semi-simple group separately, as it can be easily
obtained as a special case of the combined Abelian and semi-simple gauging, which we
discuss now.
2.2.3. K = KAbelian Ksemi-simple
If K is a direct product of a semi-simple and an Abelian group, one simply has to
combine Sections 2.2.1 and 2.2.2:
Let KS denote the semi-simple and KA the Abelian part of K. Then the gauging of KA
implies
(5 + n)SO(5,n) singlets(KA ) non-singlets(KA ).
(2.26)
As discussed in Section 2.2.1, the non-singlet vector fields of KA have to be converted to

self-dual tensor fields. These tensor fields cannot be charged under the semi-simple part
KS (see the discussion of Section 2.2.2). Hence, KS can only act on the singlets of KA .
Furthermore, the action of KS on these KA singlets cannot introduce additional tensor
fields, i.e., there can be no non-singlets of KS beyond the adjoint of KS :
singlets(KA ) adjoint(KS ) singlets(KS ).
(2.27)
132
As described in Section 2.2.1, the ChernSimons term of the ungauged theory (2.11)
requires KA to be one-dimensional, i.e., we can have either K = U (1) KS or
K = SO(1, 1) KS .
We will take a closer look at this general gauging in Section 4.
3. The gauging of an Abelian group KA SO(5, n)

The goal of this paper is to construct the most general gauging of N = 4 supergravity
coupled to an arbitrary number of vector and tensor multiplets. As was discussed in
the previous section, this corresponds to the simultaneous gauging of a one-dimensional
Abelian group KA and a semi-simple group KS . However, it is the Abelian gauging
that introduces the tensor fields, so it is worth treating this technically more subtle case
separately:
The gauging of an Abelian group KA SO(5, n) proceeds in three steps (see also the
N = 2 [28] and N = 8 [31,32] cases):
Step 1. As discussed in Section 2.2.1, the most general decomposition of the (5 + n) of
SO(5, n) with respect to an Abelian gauge group KA is
(5 + n)SO(5,n) singlets(KA ) non-singlets(KA ),
(3.1)
We will again use I, J, . . . for the KA singlets and M, N, . . . for the non-singlets of KA .
Note that the (rigid) KA invariance of the ChernSimons term in (2.11) already implies
that we have CI M = 0: if CI M = 0, we would need M
N CI M = 0 for the invariance of
CI M F I F M a, where M
denotes
the
K
transformation
matrix of the non-singlet
A
N
M
vector fields A . But then the KA representation space spanned by the AM
would have
to have one non-trivial null-eigenvector, i.e., there would be at least one singlet among
the AM
, contrary to our assumption. This same condition CI M = 0 also follows by
requiring the closure of the supersymmetry algebra on the vector fields AI , which in the
superspace analysis amounts to the requirement of the closure of the Bianchi identities for
the supercurvatures F I .
In order to gauge KA , the non-singlet vector fields AM
now have to be converted to
M , whereas the singlet vector fields AI will play the rle of
self-dual tensor fields B
spectator vector fields. Using the Nther procedure, the tensor field dualization is done
M in the ungauged Lagrangian (2.11) and the ungauged
by first literally replacing all F
M . This is more than a mere relabeling,
transformation laws (2.12) by tensor fields B
M are no longer assumed to be the curls of vector fields AM . Because of
as the B
M , i.e., in general, we
this, we no longer have a Bianchi identity for the tensor fields B
now have dB M = 0. This, however, already breaks supersymmetry, because some of the
supersymmetry variations in the ungauged theory only vanished due to dF M = 0. As a
remedy, one therefore adds the extra term
LBdB =
1
M
/
MN B
BN
4gA
(3.2)
133
M to be
to the Lagrangian and requires the supersymmetry transformation rule of the B
M
M
N
NM i
B
[ ] j
= 2[ ]
+ 2gA a[ M
N ] gA LNij
i
i
+ gA LaN NM ia i gA LNij NM i j .
(3.3)
4
2 3
Here, gA denotes the KA coupling constant, MN is a constant and antisymmetric metric
with inverse MN :
P
MN NP = M
,
MN = NM ,
M into L
and M is as defined in Eq. (2.13). It is not too hard to show that inserting B
BdB
M
exactly cancels the supersymmetry breaking terms that arise due to dB = 0.
Step 2. Closer inspection of LBdB reveals that it combines with
2
M N
CMN / B
LBBa =
B a
8
(which stems from the former ChernSimons term in (2.11)), to automatically form a
KA -covariant derivative:
LBdB + LBBa =
1
M
/
MN B
D BN ,
4gA
where
M
M
N
D B
B
+ gA a M
N B
(3.4)
with
1 MP
CP N
M
(3.5)
N =
2
being the KA -transformation matrix of the tensor fields. The same is true for the first
M (Eq. (3.3)), which naturally combine to 2D M . However, apart
two terms in B
[ ]
from these two derivatives, none of the various other derivatives in the Lagrangian or the
transformation laws are covariantized with respect to KA . On the other hand, the only
M
derivatives affected by this are derivatives acting on the coset representatives LA
M , LA ,
a
j
b
which exclusively appear inside the composite connections Qi , Qa and the Pij
(cf. Eqs. (2.9), (2.10)). Introducing KA -covariant derivatives
M
M N
D LM
A LA + gA a N LA
(3.6)
a is tantamount to making the replacements

also inside Qi j , Qa b and Pij
Nkj
Qi j Qi j = Qi j gA a M
,
N LMik L
(3.7)
Nb
Qa Qa = Qa + gA a M
N LMa L ,
a
a
a
Na
Pij
Pij
= Pij
gA a M
N LMij L
(3.8)
(3.9)
everywhere in the Lagrangian and the transformation laws.

Step 3. After Steps 1 and 2, the theory is now KA gauge invariant to all orders in
gA and supersymmetric to order (gA )0 . However, at order (gA )1 the theory fails to be
134
supersymmetric due to the numerous new gA -dependent terms we introduced. As a third

step, one therefore has to restore supersymmetry by adding a few gA -dependent (but gauge
invariant) terms to the covariantized Lagrangian and transformation rules.
After all these modifications, the final Lagrangian is given by (up to 4-fermion terms):
1
1
1
1
I
H J 4 G G
e1 L = R i D i 2 aIJ H
2
2
4
4
1
1 i
1 ia a 1 aij
2
( ) D
/ i D
/ i P Paij
2
2
2
2
3
i i
ij I
ia
j
a
L H
i + i Pij +
i j
I
2
6
1
1
I ai
LaI H
i 2 i i G
4
2 6
5i
i
ij I
2 i
+ i G L H
i j
I
12
24 2
i
i
I ia
i 2 G ia ai
LaI H
2 3
8 2

i
i
ij I a a
ij I
i j + 2 i j
i j L H
+ L H
I
I
4
4
i

i
2
i + 2 i i
G
8 2
2 1
e1
I
J
M
e CI J / F
+
F
a +
/
MN B
D BN
8
4gA
3i
5i
+ gA Uij i j + igA Nij ab ia j b gA Uij i j
2
2
gA Vija i j a 2 3 gA Uij i j
4i
2 (A)
gA Vija i j a gA
V ,
3
(3.10)
and the supersymmetry transformation laws are given by (up to 3-fermion terms)
1
em = i m i ,
2

i
I
4 j
i = D i + LIij H
6

i
+ 2 G 4 i + igA Uij j ,
12 2
3
i
1
I
LIij H
i = / i +
j 2 G i 2 3 gA Uij j ,
2
6
2 6
1
I
a
ai = iPij
j LaI H
i + gA Vija j ,
4
AI = I ,
M
M
i
= 2D[ ]
gA LNij NM [
] j
B
135
i
i
+ gA LaN NM ia i gA LNij NM i j ,
4
2 3
1 2 i
i
a = i 2 i i ,
6
2 2
i i
= i ,
2

1
ij
j]
L = iLaI k[i l ij kl k la ,
I
4
LaI = iLIij i j a .
In the above expressions, we have introduced the tensorial quantity
I

I
M
H
F
, B
(3.11)
(3.12)
as well as the new USp(4) SO(n) covariant derivatives

b
i
(3.13)
with the new, gA -dependent connections (3.7) and (3.8).

The new scalar field dependent quantities Uij , Vija , Nij ab , as well as the scalar potential
V (A) are fixed by supersymmetry:
2 2 N
M LN(i|k| LMk j ) ,
Uij =
(3.14)
6
1
Ma
,
Vija = 2 N
(3.15)
M LNij L
2
1
3
a Mb
Nij ab = 2 N
(3.16)
ij + Uij ab ,
M LN L
2
2 2
1
V (A) = Vija V aij .
(3.17)
2
In particular, the Uij and Vija shifts in the supersymmetry transformations of the
fermionic fields are all expressed in terms of the so-called boosted structure constants,
which are just the representation matrices of the vector fields under the gauged group
multiplied by the coset representatives. This is a common feature of all gauged
supergravities which have a scalar manifold given by an ordinary homogeneous space [37].
4. The simultaneous gauging of KS KA
The starting point for the simultaneous gauging of KS KA is the KA gauged
Lagrangian (3.10) with the corresponding supersymmetry variations (3.11). In this case,
fields carrying an SO(5, n) index I decompose according to (2.26) and (2.27) into
adjoint(KS ), singlets(KS ) and non-singlets(KA ) fields. However, to avoid the introduction
of a third type of SO(5, n) index beside I and M, we will collectively denote the
adjoint(KS ) and singlet(KS ) fields by I, J, K, while non-singlet(KA) fields will carry
indices M, N . The actual distinction between the adjoint(KS ) and the singlet(KS ) fields
136
will be made implicitly by the KS structure constant fIKJ which will be taken to vanish
whenever one index denotes a KS singlet field. In order for the ChernSimons term to be
globally KS invariant, these structure constants have to satisfy the following identity
I
I
+ CI L fKJ
= 0.
CI J fKL
(4.1)
The additional gauging of KS essentially proceeds in two steps:

Step 1. All derivatives acting on KS charged fields must be KS covariantized. This
modifies the definition of the field strengths of the gauge fields AI in the standard way:
I
I
I
F
= F
+ gS AJ fJIK AK
F

I
I
I
M
H H = F , B .
thus
(4.2)
In order to simplify the notation, we use the same derivative symbols as in the previous
section, but now also with gS dependent contributions:
I
I J I
N
K
a M
D LIA = LIA + gA M
N LA + gS I A fJ K LA .
(4.3)
a ,
This in turns modifies the USp(4) and SO(n) connections, as well as the vielbein Pij
as these quantities contain derivatives of the coset representatives. Again, we use the
same symbols as in the previous section, but now also include the new gS dependent
contributions, i.e.,
kj
Nkj
I
+ gS AJ LK
Qi j = Qi j gA a M
N LMik L
ik fJ K LI ,
(4.4)
Nb
I
b
Qa = Qa + gA a M
gS AJ LK
a fJ K LI ,
N LMa L
a
a
Na
I
a
= Pij
gA a M
gS AJ LK
Pij
N LMij L
ij fJ K LI .
(4.5)
(4.6)
These new objects appear in the derivatives of the fermions, which means that
b
i
(4.7)
should now be understood as containing also the gS dependent terms inside the composite
connections (4.4), (4.5).
After these modifications in the supersymmetry transformations rules (3.11) and the
Lagrangian (3.10), the latter is supersymmetric up to a small number of gS dependent
terms. These uncanceled terms take the form
1
i
J
J
a ia j
fJIK LI ik LKkj i j + gS F
fJIK LK
e1 (L) = gS F
ij LI
2
2

a
+ gS fJKI LI ij LaK Pij AJ .
(4.8)
Step 2. Just as in the Abelian case, the remaining terms can be compensated by adding
fermionic mass terms as well as potential terms to the Lagrangian, and by suitable
modifications of the fermionic transformation rules.
The additional mass and potential terms needed are
Lmass =
3i
i
gS Sij i j + igS Iij ab ia j b + gS Sij i j + gS Tija i j a
2
2
2i
i
j
a
i
j
a
+ 3 gS Sij gS Tij gS gA V (AS) gS2 V (S),
(4.9)
3
137
whereas the additional pieces in the transformation rules of the fermions take the form
new i = igS Sij j ,
new ai
= gS Tija j ,
new i = gS 3 Sij j .
(4.10)
(4.11)
(4.12)
The scalar field dependent functions in the above expressions are given by
2
I
Sij = 1 LJ(i|k| fJKI Lkl
K L|l|j ) ,
9
k I
Tija = 1 LJ a LK
(i fJ K LI |k|j ) ,
3
I
b
Iij ab = Sij ab 1 LJ a LK
ij fJ K LI ,
2
whereas the potential terms read
(4.13)
(4.14)
(4.15)
9
1
(4.16)
V (S) = Sij S ij + Tija T aij .
2
2
In addition, supersymmetry implies a series of derivative relations on the scalar
quantities introduced above, which will be very useful for the study of the vacua of the
theory:
V (AS) = 18Uij S ij ,
2
2 a
Pak j ) ,
D Uij = Uij V(i|k|
3
3
2
D Vija = 2N[i k ab Pb |k|j ] 3U[i|k| Pak j ] + Vija ,
3
2 a
1
D Sij = T(i|k| Pak j ) Sij ,
3
3
1
D Tija = Tija + 3S(i|k| Pak j ) 2I(i|k|ab Pbk j ) ,
3
(4.17)
(4.18)
(4.19)
(4.20)
where the derivatives should be understood as being fully SO(n) and USp(4) covariant
derivatives based on the new composite connections and vielbeins (4.4)(4.6). Using
the numerous constraints satisfied by the coset representatives LA , one can show that
I
Eqs. (4.17)(4.20) follow automatically from the definitions of the fermionic shifts.
Just as for the gauging of the Abelian factor, we point out that the shifts in the
supersymmetry laws of the fermionic fields are given by the boosted structure constants
of the gauged semi-simple group.
5. The scalar potential V
The full scalar potential obtained from the (KS KA ) gauging is
2 (A)
V = gA
V + gA gS V (AS) + gS2 V (S)

1 2 a aij
= gA
Vij V 36gA gS Uij S ij + gS2 Tija T aij 9Sij S ij .
2
(5.1)
138
This potential could have been derived directly from the expression of the shifts in
the supersymmetry laws of the fermionic fields and the expression of their kinetic terms.
Indeed, as proved for the first time in four dimensions [38], in all supersymmetric theories
the potential is given by squaring the shifts of the fermions using the metric defined by the
kinetic terms:
1
1
1
1
1 i 2 i 1 i 2 i 1 ia 2 ai = 1i 2i V, (5.2)
2
2
2
4
provided one satisfies the generalized Ward-identity:

k
1 ak a

1
1 2 ak a
ak a
k
ij V = gA Vi Vkj + gA gS 9 Si Ukj + Ui Skj + Vi Tkj Ti Vkj
4
2
2

1 2 ak a
gS Ti Tkj 9Si k Skj .
(5.3)
2
This can indeed be verified by using the expression of these objects in terms of the coset
representatives and then using the properties of the latter.
It is instructive to rewrite the scalar potential, as well as the other scalar field dependent
quantities in terms of the Killing vectors characterizing the gauged isometries of the scalar
manifold M. To this end, we switch back to the parameterization in terms of the x
(cf. Section 2.1). On the x , KS KA transformations act as isometries:
x = K x + I KIx ,
(5.4)
where and I are, respectively, the infinitesimal KA and KS transformation parameters,

whereas K x and KIx denote the corresponding Killing vectors, which can be expressed as:
1
1
Na
K x = f xij a M
KIx = f xij a LJij fIKJ LaK .
N LMij L ,
2
2
As in the N = 2 case [30,39], one has Killing prepotentials defined as:
Dx Pij(A) = K y Rxyij ,
Dx PI(S)
ij = KI Rxyij ,
(5.5)
(5.6)
where Rxyij is the USp(4) curvature and the derivatives Dx contain the original composite
connections of the ungauged theory (see Eq. (2.9)):
Ia
b
Qab
x = L x LI
ij
and Qx = LI ik x LIk j .
(5.7)
These prepotentials are found to be:

1
1
(A)
(S)
k
N
Pij = M
PI ij = fIKJ LJi k LKkj
N Lik LM j ,
2
2
and are exactly the objects that appear in the shift of the composite connection:
(S) j
i .
Qi j = Qi j 2gA a P (A) i j 2gS AI PI
(5.8)
(5.9)
Moreover, the prepotentials have to satisfy the algebra of the KS isometries [39]
y
K (S)
KIx KJ Rxyij + 4P[I(S)i k PJ(S)
]kj + fI J PKij = 0.
(5.10)
However, unlike in the N = 2 case [30,39], this does not give any additional constraint on
the prepotentials. Indeed, using Eqs. (5.5), (5.8) and (B.6), one can show that the relation
is identically satisfied.
139
Now, using the above relations, the shifts in the gaugini transformation rules can be
expressed as:
1
a
Vija = 2 fxij
Kx,
2 2
1
a
Tija = 1 LIi k fxkj
KIx ,
2
(5.11)
whereas the shifts in the gravitini transformation rules are given in terms of the
prepotentials:
2 2 (A)
4
Pij ,
Uij =
(5.12)
Sij = 1 LIi k PI(S)
kj .
3
9
A general study of the vacua and domain-wall solutions of this theory is beyond the
scope of this paper, and will be left for a future publication [40]. Instead, we conclude with
a list of special cases, some of which were already studied in the literature.
gS = 0. Turning off the KS gauging leads us back to the Abelian theories studied in
Section 3. In this case, the potential reduces to 5
1 2 a aij 1 2 4 a aij
2 (A)
V = gA
V = gA
Vij V = gA Vij V ,
(5.13)
2
2
where, in the following, a hat always denotes the independent part of the scalar field
dependent quantities. It is then easy to see that minimizing the potential with respect to
requires the potential to vanish at the extrema:
V|c = 0
V|c = 0.
(5.14)
Hence, if this potential admits critical points, they have to correspond to Minkowski ground
states of the theory. On the other hand, such critical points do not necessarily have to exist.
In the case when KA is compact (i.e., KA = U (1)), however, one can prove that the theory
indeed has at least one Minkowski ground state: expressing the potential in terms of the
Killing vectors, we obtain, using (5.11):
1 2 4
gxy K x K y .
V = gA
(5.15)
4
Being compact, the U (1) gauge symmetry is a subgroup of the maximal compact subgroup
H = SO(5) SO(n) of G = SO(5, n), i.e., it is a subgroup of the isotropy group of the
scalar manifold. This ensures that there exists at least one point o M that is invariant
under the action of KA , i.e., at this point, the U (1) Killing vector and, consequently, the
potential vanish:
K x
o = 0 V|o = 0.
(5.16)
Hence, there is a least one Minkowski ground state for the U (1) gauged theory.
gA = 0. At a first look, a naive limit gA 0 seems to be ill-defined due to the presence
of the 1/gA term in the Lagrangian (3.10). As was shown in [41], however, there is a
5 Note the similarity with the corresponding N = 2 theories [28], where the part of the scalar potential that is
due to the presence of tensor multiplets is also non-negative.
140
perfectly well-defined procedure to take this limit, based on a field redefinition of the form
M
M
M
gA C
+ F
,
B
M
F
(5.17)
AM
.
where
is the curl of some vector field
After such a limit has been taken, the
KS KA theory reduces to a theory of the type discussed in Section 2.2.2, in which only
a semi-simple group KS is gauged.
The scalar potential of such a theory is then given by:

1
V = gS2 V (S) = gS2 Tija T aij 9Sij S ij .
(5.18)
2
Again, one can factor out the -dependent part, and write the potential as
a aij

1
ij T
9
V = gS2 2 T
(5.19)
Sij
S ij .
2
The form of the -dependence shows that, as in the purely Abelian case, at most
Minkowski ground states can exist:
V|c = 0
V|c = 0.
(5.20)
However, in contrast to the Abelian case, we cannot prove the existence of such a vacuum
state, even if we assume KS to be compact. Indeed, whereas the Tija are expressed in terms
of the Killing vectors as in Eq. (5.11), the Sij are proportional to the Killing prepotentials,
which need not vanish at the invariant point o M where the Killing vectors are zero
for a compact KS . Eq. (5.6) merely implies that the prepotentials should be covariantly
constant at that point.
The particular choice KS = SU(2) SO(5) corresponds to the case studied in [26].
Indeed, our potential (5.18) has precisely the same form as the potential of Ref. [26].
However, in our case, we have
x V = 0,
(5.21)
due to the fact that we allowed for a general mass term Iij ab for the gaugini, that also
contains a part antisymmetric in a, b, whereas in [26], one has Iij ab = 32 Sij ab . Thus, it
seems that the SU(2) gauging considered in [26] is not the most general one.
n = 0. In this case, only the N = 4 supergravity multiplet is present. This means that
the global symmetry group SO(1, 1) SO(5, n) of the ungauged theory (2.11) reduces
to SO(1, 1) SO(5), where SO(5) is just the R-symmetry group of the N = 4 Poincar
superalgebra. As discussed in Section 2.2.2, KS can then only be the standard SO(3)
=
SU(2) subgroup of SO(5), and we therefore expect to recover the SU(2) U (1) gauged
theory of [25] when this SU(2) is gauged together with the obvious additional SO(2)
=
U (1) subgroup of SO(5).
Indeed, in the absence of exterior vector multiplets, we have Vija = Tija = 0, whereas
ij = 1 (45 )ij and
ij and
Sij become constant matrices (U
the independent quantities U
6
2
2
6

Sij = 9 (123 )ij = 9 (45 )ij ) such that the potential (5.1) reduces to

V = gS gA + gS 2 ,
(5.22)
6 The gamma matrices introduced here refer to Euclidean gamma matrices of SO(5).
141
where we have discarded some numerical factors. This potential indeed coincides with the
potential of [25].
n = 2, K = SU(2) U (1). In the ungauged theory with scalar manifold
M=
SO(5, 2)
SO(1, 1),
SO(5) SO(2)
(5.23)
one can gauge an SU(2) U (1) subgroup such that four tensor fields appear, two in the
supergravity multiplet and two coming from the former vector multiplets. To this end, the
SU(2) factor has to be the standard SO(3)
= SU(2) SO(5), whereas the U (1) factor has
to be a diagonal subgroup of the two SO(2)s in the obvious subgroup SO(2) SO(3)
SO(2) SO(5, 2). This corresponds to the N = 4 theory obtained in [17] by truncating
the N = 8 theory to an SU(2)I SU(4) invariant subsector. The relevance of this sector
for the study of certain RG-flows in the AdS/CFT correspondence was stressed in the
introduction.
6. Conclusions
In this paper, we have studied the possible gaugings of matter coupled N = 4
supergravity in five dimensions. All these theories can be obtained from the ungauged
MESGTs of Ref. [26] by gauging appropriate subgroups of the global symmetry group
SO(5, n) SO(1, 1), where n is the number of vector plus tensor multiplets. A more careful
analysis of the possible gauge groups revealed that the SO(1, 1) factor cannot be gauged,
whereas the possible gaugeable subgroups K of SO(5, n) naturally fall into three different
categories:
If K is Abelian, its gauging requires the dualization of the K charged vector fields into
self-dual antisymmetric tensor fields. For consistency with the structure of the ungauged
theory, such an Abelian gauge group K has to be one-dimensional, i.e., it can be either
U (1) or SO(1, 1). In the case K = U (1), we could prove the existence of at least
one Minkowski ground state. Such an existence proof could not be given for the case
K = SO(1, 1). 7 In any case, however, no AdS ground states are possible if K is Abelian.
The gauging of a semi-simple group K SO(5, n), by contrast, does not introduce any
tensor fields. Conversely, self-dual tensor fields can only be charged with respect to a onedimensional Abelian group. This is an interesting difference to the N = 2 and N = 8 cases
[28,31,32], where the tensor fields could also be charged under a non-Abelian group. As
for the critical points of the scalar potential, we found that they can at most correspond
to Minkowski spacetimes. However, we were not able to give an existence proof similar
to the U (1) case. To sum up, neither the pure Abelian gauging nor the gauging of a
semi-simple group alone allow for the existence of AdS ground states. As for N = 4
supersymmetric AdS ground states, this is to be expected from the corresponding AdS
superalgebra SU(2, 2|2), which has the R-symmetry group SU(2) U (1).
7 In fact, in [42] a counter-example was found for a particular non-compact gauging of the N = 2 analog.
142
This situation changes when K = KS KA is the direct product of a semi-simple and

an Abelian group. In this case, the Abelian group KA again has to be one-dimensional, and
tensor fields are required for its gauging. The resulting scalar potential for the simultaneous
gauging of KS and KA consists of the sum of the potentials that already arise in the
separate gaugings of KS and KA , but also contains an interference term that depends
on both coupling constants. Again, this is an interesting difference to the analogous
N = 2 gaugings [28], where no such interference was found. Due to all these different
contributions to the scalar potential, it is now no longer excluded that AdS ground states
exist. In fact, this is to be expected from various special cases that have appeared in the
literature [17,25]. On the other hand, the more complicated form of this potential also
makes a thorough analysis of its critical points along the lines of, e.g., [43] more difficult.
At this point, we should mention that the conclusions of our analysis on the possible
gaugings might be modified if one were to consider even more general gaugings based on
non semi-simple algebras as was done, e.g., in the N = 8 case in [44].
Having obtained the general form of matter coupled N = 4 gauged supergravity, one
might now try to clarify several open problems. First, it would be interesting to see whether
some of these theories can indeed be used to extract information on the N = 2 quiver gauge
theories discussed in [1722], and to recover the gravity duals of various RG-flows that
have been constructed [17] or were conjectured to exist [24]. Furthermore, one now has
the tools to elucidate some of the questions raised in [45] concerning possible alternative
gaugings of certain N = 2 matter coupled theories.
Another interesting issue would be to clarify whether the gaugings presented here can be
reproduced by dimensional reduction of the heterotic string on a torus with internal fluxes,
as was done in [46]. In fact, the dimensional reduction does not introduce any additional
tensor fields (beside the NSNS two-form). As these were shown to play a essential rle
for the Abelian gauging, one might expect some differences concerning this part of the
gauging between the two approaches.
Finally, it should be extremely interesting to use this new theory to obtain new insight
on the existence of smooth supersymmetric brane-world solutions la RandallSundrum.
Even if, so far, the study of the N = 2 theory has not yet provided a final answer [47],
it has anyhow restricted the analysis to models involving vector- and hypermultiplets.
Unfortunately, there are many possible scalar manifolds and gaugings that one can build
for such models. In any case, the models that can be obtained by reduction of some N = 4
realization must be contained in the general setup presented here. Moreover, since the
N = 4 scalar manifold has the unique form presented in (2.6), the analysis of the possible
gaugings and flows, should be simpler and we hope that it should lead to more stringent
results.
Acknowledgements
The work of G.D. is supported by the European Commission under TMR project
HPRN-CT-2000-00131. The work of C.H. is supported by the GermanIsraeli Foundation
143
for Scientific Research (GIF). The work of M.Z. is supported by the German Science
Foundation (DFG) within the Schwerpunktprogramm 1096 Stringtheorie.
Appendix A. Spacetime and gamma matrix conventions

This appendix summarizes our spacetime and gamma matrix conventions. The spacetime metric g and the fnfbein e m are related by
g = e m e n mn
with mn = diag( + + + +). The indices , , . . . and m, n, . . . are curved and flat
indices, respectively, and run from 0 to 4. Our conventions regarding the Riemann tensor
and its contractions are

Rmn () = mn + mp p n ( ) ,
(A.1)
R(, e) = em en Rmn ,
(A.2)
where mn (e) is the spin connection defined via the usual constraint
[ e]m + [m n e]n = 0.
The gamma matrices m in five dimensions are constant, complex-valued (4 4)-matrices
satisfying
{m , n } = 2mn .
Given a set {0 , 1 , 2 , 3 } of gamma matrices in four dimensions, the definition
4 := i01 2 3
(A.3)
yields a representation of the corresponding Clifford algebra in five dimensions for either
choice of the sign. We will always choose this sign such that Eq. (A.4) below holds. The
gamma matrices with a curved index are simply defined by := e m m .
Antisymmetrized products of gamma matrices are denoted by
1 2 p := [1 2 p ] ,
where the square brackets again denote antisymmetrization with strength one. Using this
definition, we have
i
= / .
e
The charge conjugation matrix C in five dimensions satisfies
T
C C 1 = .
(A.4)
(A.5)
It can be chosen such that

C T = C = C 1 .
The antisymmetry of C and the defining property (A.5) imply that the matrix (C1 p )
is symmetric for p = 2, 3 and antisymmetric for p = 0, 1, 4, 5.
144
All spinors we consider are anticommuting symplectic Majorana spinors

i := (i ) 0 = ij jT C.
As a consequence of the symmetry properties of (C1 p ), these (anticommuting)
spinors satisfy the following useful identities:

j
i
i 1 p j = + j 1 p i (p = 0, 1, 4, 5),
1 p (p = 2, 3)
which implies

i
i 1 p i = i 1 ...p i
+ 1 ...p i
(p = 0, 1, 4, 5),
(p = 2, 3).
Appendix B. The geometry of the scalar manifold

The scalars of the theories studied in this paper parameterize the manifold:
M=
SO(5, n)
SO(1, 1),
SO(5) SO(n)
(B.1)
where the SO(1, 1) factor is spanned by the scalar of the supergravity multiplet and the
rest by the scalars x of the matter multiplets.
Having introduced the coset representatives LI A , one can determine the metric and
curvatures of the manifold (B.1). Indeed the vielbeins are obtained from
aij
fx
ij
I
= 2LI a x L ,
(B.2)
whereas the metric is given by

ij a
a
fx fyij
= 4gxy .
(B.3)
The inverse of the vielbeins are defined via

1
ij a
j]
fx fklxb = 4 k[i l ij kl ab .
4
(B.4)
As usual, the vielbeins are covariantly constant with respect to the full covariant derivative:
ij a
x fy
ij a
xy z fz
ij b
Qxb a fy
kj a
Qxk i fy
Qxk j fyika = 0,
(B.5)
z denotes the Christoffel symbols on M.

where xy
The USp(4) and SO(n) curvatures fulfill the following identities:
1 a
akj
fy] ,
Rxyi j = f[xik
4
1 ij
Rxya b = f[xa fy]ij b ,
4
(B.6)
(B.7)
145
which can also be found as the solutions to the integrability conditions coming from the
differential equations satisfied by the coset representatives
1
a
Dx LIij = LaI fxij
,
2
1 a
Dx LIij = LI a fxij
,
2
1 a ij
Dx LaI = fxij
L ,
I
2
1
ij a
Dx LI a = fx LIij ,
2
(B.8)
(B.9)
(B.10)
(B.11)
where Dx denotes the USp(4) and SO(n) covariant derivative. Let us finally display the
useful identity:
1 j
L(Iik LJ) j k = i LI kl LJkl ,
4
(B.12)
which is nothing but the defining relation of the SO(5) Clifford algebra used to convert the
SO(5) index of the LI A into the corresponding composite USp(4) index with the property
that
LI ij = LI j i ,
LI ij ij = 0.
(B.13)
Appendix C. Superspace analysis

In this appendix we want to propose a brief analysis of the constraints needed to
reproduce the theory built in the main text in superspace. This will help us understand
to which extent the couplings proposed here (besides non-minimal ones) are general.
In the superspace formalism, the dynamics of the fields is determined by the constraints
one imposes on the supercurvatures. To classify the possible consistent sets of constraints,
one adopts a general strategy which is based on group-theoretical analysis [48]. The
superspace formalism gives quite stringent restrictions on the possible couplings of the
various multiplets of the theory and the analysis of the lowest-dimensional components of
the various superfields (supertorsion and supercurvatures) reveals what freedom one has in
coupling for instance the gravity multiplet to the matter ones.
Moreover, we report some additional equalities, derived by the solution of the superspace
Bianchi identities for the various supercurvatures, which are useful to derive some of
the properties of the shifts in the transformation laws of the fermionic fields. The same
relations could be derived at the component level by closing the supersymmetry algebra on
the various fields.
We denote a generic superform as
=
1 A1
e eAn An A1 ,
n!
(C.1)
146
where A = (, i) collectively denote vector () and spinor (i) indices, and eA are the
supervielbeins (which means that the projection of ei on ordinary spacetime dx i
contains the gravitino field).
The (super-)torsion and (super-)curvatures are defined as
DeA = T A ,
DT A = eB RB A ,
1
F I = dAI + gS fJIK AJ AK ,
2
N
H M = dB M + gA M
N aB ,
(C.2)
G = da,
(C.3)
(C.4)
and they satisfy the Bianchi identities:

DT A = eB RB A ,
DF = 0,
I
DRB A = 0,
dG = 0,
N
DH M = gA M
N GB .
(C.5)
(C.6)
(C.7)
At this level, one tries to impose constraints on such superfields which are compatible with
their Bianchi identities (BI) such as to remove all the auxiliary fields that appear in their
component expansion.
In the language of superspace, the basic constraint one has to impose to close N = 4
supersymmetry in a linear way is given by
1
m
m
= ij
.
Tij
(C.8)
2
The other basic constraints on the torsion field are given by the definition of the Tim n
and Tij k components. To analyze the freedom we have, and therefore the possibility of
coupling gravity with matter, we make a group theoretical analysis of these structures.
The Tij k tensor contains the following SO(5)USp(4)R representations 5 44 +
3 416 + 420 + 3 164 + 2 1616 + 1620 + 204 + 2016 + 2020 , whereas Tim n contains
2 44 + 2 164 + 204 + 404 .
Using the connection redefinition im n im n + Xi[mp] pn we are free to reabsorb
many of the components of Tim n which then remains with only the 44 + 164 + 404
representations. Moreover, using the gravitino redefinition i em Hm i we are left with
only the 404 . This latter is then set to zero by the lowest dimensional T -BI, which is the
equation R(ij k) m = 0.
At the same time the Tij k tensor remains with the irreps 2 44 + 416 + 420 which
allows only for structures of the kind i j k and Qi(j k) . Indeed the first kind of term
is needed to close the BI of the graviphoton, whereas the second is used to couple matter
(it has the same tensor structures as an USp(4) connection, or more precisely its pull-back
on superspace). This means that, even if one has not yet defined the scalar manifold, the
type of interaction between the gravity sector and the matter must be mediated by the
appearance of some type of USp(4) connection field.
Once the connection-type term has been reabsorbed in the definition of a new
supercovariant derivative and the F -BI (with coupling to matter) have been solved,
one finds that the definition of the torsion constraint which is compatible with the
147
supersymmetry transformations presented in this paper is given by

i
k
k
k
Tij k = i j + j i + 2 k C ij + 4[i jk] C j i
2 3

k
i j + k ij k ij ,
(C.9)
which translates in components as the following three-Fermi term in the transformation

rule of the gravitino:
i
k = + i i k i i k + 2 i i k 2 k i i
2 3

+ 2 i k i + k i i k i i i k i i k i ,
(C.10)
where in the dots are the components presented in the main text and the three-Fermi terms
due to the pull-back of the USp(4) connection.
For completeness, we list here also the other fundamental constraints:
i
Di = i ,
2
i
Di x = ja fijx a ,
2
i
Gij = 2 ij C ,
2 2
I
= i 1 LIij C ,
Fij
M
Hij
k = 0.
(C.11)
(C.12)
(C.13)
(C.14)
(C.15)
As promised, we also show the equalities that one derives between the various shifts in
the supersymmetry laws of the Fermi fields by solving the superspace Bianchi identities
(BI). From the H -BI we derive
a
M
2 1
M
N
k M
(4U + 2S)[i k LM
j ]k V[ij ] La + 2(S + U )[i Lj ]k = N Lij ,
2
(C.16)
whereas from the F -BI we get

a
M
k M
(4U + 2S)[i k LM
j ]k + V[ij ] La + 2(S + U )[i Lj ]k = 0.
(C.17)
Moreover, the closure of the -BI implies that

1
Vija = 2 N M LNij LMa ,
2
(C.18)
Tija = 1 fJIK LI (i m LJ|m|j ) LKa .
(C.19)
Using the solution of these quantities in terms of the coset representatives, the (C.17)
equation becomes just C I M = 0 (for the Abelian part, an identity for the non-Abelian).
148
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231, hep-th/9711200.
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/
9802150.
[4] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, Phys. Rep. 323 (2000) 183, hep-th/9905111.
[5] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690,
hep-th/9906064;
L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev.
Lett. 83 (1999) 3370, hep-ph/9905221.
[6] R. Kallosh, A. Linde, Supersymmetry and the brane world, JHEP 02 (2000) 005, hep-th/
0001071.
[7] K. Behrndt, M. Cvetic, Anti-de Sitter vacua of gauged supergravities with 8 supercharges, Phys.
Rev. D 61 (2000) 101901, hep-th/0001159.
[8] G.W. Gibbons, N.D. Lambert, Domain walls and solitons in odd dimensions, Phys. Lett. B 488
(2000) 90, hep-th/0003197.
[9] R. Altendorfer, J. Bagger, D. Nemeschansky, Supersymmetric RandallSundrum scenario, hepth/0003117.
[10] T. Gherghetta, A. Pomarol, Bulk fields and supersymmetry in a slice of AdS, Nucl. Phys. B 586
(2000) 141, hep-ph/0003129.
[11] A. Falkowski, Z. Lalak, S. Pokorski, Supersymmetrizing branes with bulk in five-dimensional
supergravity, Phys. Lett. B 491 (2000) 172, hep-th/0004093.
[12] A. Falkowski, Z. Lalak, S. Pokorski, Five-dimensional gauged supergravities with universal
hypermultiplet and warped brane worlds, hep-th/0009167.
[13] E. Bergshoeff, R. Kallosh, A. Van Proeyen, Supersymmetry in singular spaces, JHEP 0010
(2000) 033, hep-th/0007044.
[14] M.J. Duff, J.T. Liu, K.S. Stelle, A supersymmetric type IIB RandallSundrum realization, hepth/0007120.
[15] L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, Novel local CFT and exact results on
perturbation of N = 4 super YangMills dynamics from AdS dynamics, JHEP 12 (1998) 022,
hep-th/9810126.
[16] J. Distler, F. Zamora, Nonsupersymmetric conformal field theories from stable anti-de Sitter
spaces, Adv. Theor. Math. Phys. 2 (1999) 1405, hep-th/9810206.
[17] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from
holography-supersymmetry and a c-theorem, hep-th/9904017.
[18] M.R. Douglas, G. Moore, D-branes, quivers, and ALE instantons, hep-th/9603167.
[19] A.E. Lawrence, N. Nekrasov, C. Vafa, On conformal field theories in four dimensions, Nucl.
Phys. B 533 (1998) 199, hep-th/9803015.
[20] S. Kachru, E. Silverstein, 4d conformal theories and strings on orbifolds, Phys. Rev. Lett. 80
(1998) 4855, hep-th/9802183.
[21] Y. Oz, J. Terning, Orbifolds of AdS(5)S(5) and 4d conformal field theories, Nucl. Phys. B 532
(1998) 163, hep-th/9803167.
[22] S. Gukov, Comments on N = 2 AdS orbifolds, Phys. Lett. B 439 (1998) 23, hep-th/9806180.
[23] S. Gubser, N. Nekrasov, S. Shatashvili, Generalized conifolds and four-dimensional N = 1
superconformal theories, JHEP 9905 (1999) 003, hep-th/9811230.
[24] I.R. Klebanov, E. Witten, Superconformal field theory on threebranes at a CalabiYau
singularity, Nucl. Phys. B 536 (1998) 199, hep-th/9807080.
149
[25] L.J. Romans, Gauged N = 4 supergravities in five dimensions and their magnetovac backgrounds, Nucl. Phys. B 267 (1986) 433.
[26] M. Awada, P.K. Townsend, N = 4 MaxwellEinstein supergravity in five dimensions and its
SU(2) gauging, Nucl. Phys. B 255 (1985) 617.
[27] M. Gnaydin, G. Sierra, P.K. Townsend, Gauging the D = 5 MaxwellEinstein supergravity
theories: more on Jordan algebras, Nucl. Phys. B 253 (1985) 573.
[28] M. Gnaydin, M. Zagermann, The gauging of five-dimensional, N = 2 MaxwellEinstein
supergravity theories coupled to tensor multiplets, Nucl. Phys. B 572 (2000) 131, hep-th/
9912027.
[29] M. Gnaydin, M. Zagermann, Gauging the full R-symmetry group in five-dimensional, N = 2
YangMills/Einstein/tensor supergravity, Phys. Rev. D 63 (2001) 064023, hep-th/0004117.
[30] A. Ceresole, G. DallAgata, General matter coupled N = 2, D = 5 gauged supergravity, Nucl.
Phys. B 585 (2000) 143, hep-th/0004111.
[31] M. Gnaydin, L.J. Romans, N.P. Warner, Gauged N = 8 supergravity in five dimensions, Phys.
Lett. B 154 (1985) 268;
M. Gnaydin, L.J. Romans, N.P. Warner, Compact and noncompact gauged supergravity
theories in five dimensions, Nucl. Phys. B 272 (1986) 598.
[32] M. Pernici, K. Pilch, P. van Nieuwenhuizen, Gauged N = 8 D = 5 supergravity, Nucl.
Phys. B 259 (1985) 460.
[33] J. Maldacena, C. Nuez, Supergravity description of field theories on curved manifolds and a
no go theorem, hep-th/0007018.
[34] M.J. Duff, J.T. Liu, W.A. Sabra, Localization of supergravity on the brane, hep-th/0009212.
[35] M. Cvetic, H. L, C.N. Pope, Brane-world KaluzaKlein reductions and branes on the brane,
hep-th/0009183.
[36] P.K. Townsend, K. Pilch, P. van Nieuwenhuizen, Selfduality in odd dimensions, Phys.
Lett. B 136 (1984) 38.
[37] L. Castellani, A. Ceresole, S. Ferrara, R. DAuria, P. Fr, E. Maina, The complete N = 3 matter
coupled supergravity, Nucl. Phys. B 268 (1986) 317;
R. DAuria, S. Ferrara, S. Vaula, Matter coupled F(4) supergravity and the AdS(6)/CFT(5)
correspondence, JHEP 0010 (2000) 013, hep-th/0006107.
[38] S. Cecotti, L. Girardello, M. Porrati, Constraints on partial super-Higgs, Nucl. Phys. B 268
(1986) 295.
[39] L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fr, General matter
coupled N = 2 supergravity, Nucl. Phys. B 476 (1996) 397417, hep-th/9603004;
L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fr, T. Magri, N = 2
supergravity and N = 2 super-YangMills theory on general scalar manifolds: symplectic
covariance, gaugings and the momentum map, J. Geom. Phys. 23 (1997) 111, hep-th/9605032.
[40] G. DallAgata, C. Herrmann, M. Zagermann, in preparation.
[41] P.M. Cowdall, On gauged maximal supergravity in six dimensions, JHEP 9906 (1999) 018,
hep-th/9810041.
[42] M. Gnaydin, M. Zagermann, The vacua of 5d, N = 2 gauged YangMills/Einstein/tensor
supergravity: Abelian case, Phys. Rev. D 62 (2000) 044028, hep-th/0002228.
[43] A. Khavaev, K. Pilch, N.P. Warner, New vacua of gauged N = 8 supergravity in five
dimensions, Phys. Lett. B 487 (2000) 14, hep-th/9812035.
[44] L. Andrianopoli, F. Cordaro, P. Fr, L. Gualtieri, Non-semisimple gaugings of D = 5 N = 8
supergravity and FDAs, Class. Quant. Grav. 18 (2001) 395, hep-th/0009048.
[45] K. Behrndt, C. Herrmann, J. Louis, S. Thomas, Domain walls in five-dimensional supergravity
with non-trivial hypermultiplets, JHEP 0101 (2001) 011, hep-th/0008112.
[46] N. Kaloper, R.C. Myers, The O(dd) story of massive supergravity, JHEP 9905 (1999) 010,
hep-th/9901045.
150
[47] See, for instance, M. Cvetic, Domain wall world(s), hep-th/0012105;

A. Ceresole, G. DallAgata, Brane-worlds in 5D supergravity, hep-th/0101214.
[48] L. Bonora, M. Bregola, K. Lechner, P. Pasti, M. Tonin, A discussion of the constraints in N = 1
SUGRA-SYM in 10 D, Int. J. Mod. Phys. A 5 (1990) 461;
A. Candiello, K. Lechner, Duality in supergravity theories, Nucl. Phys. B 412 (1993) 479.

The large N limit of N = 2 super YangMills,

fractional instantons and infrared divergences
Frank Ferrari 1
Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544, USA
Abstract
We investigate the large N limit of pure N = 2 supersymmetric gauge theory with gauge group
SU(N) by using the exact low energy effective action. Typical one-complex dimensional sections of
the moduli space parametrized by a global complex mass scale v display three qualitatively different
regions depending on the ratio between |v| and the dynamically generated scale . At large |v|/,
instantons are exponentially suppressed as N . When |v| , singularities due to massless
dyons occur. They are densely distributed in rings of calculable thicknesses in the v-plane. At small
|v|/, instantons disintegrate into fractional instantons of charge 1/(2N). These fractional instantons
give non-trivial contributions to all orders of 1/N, unlike a planar diagrams expansion which
generates a series in 1/N 2 , implying the presence of open strings. We have explicitly calculated the
fractional instantons series in two representative examples, including the 1/N and 1/N 2 corrections.
Our most interesting finding is that the 1/N expansion breaks down at singularities on the moduli
space due to severe infrared divergencies, a fact that has remarkable consequences. 2001 Published
by Elsevier Science B.V.
1. Introduction
Quantum SU(N) gauge theory has a single free parameter, the number of colours N ,
or equivalently the number of interacting gluons N 2 1. Our best chance to understand
the strongly coupled dynamics of gluons is probably to perform an analysis of the theory
at large N [1]. In the present work we will use exact results available for theories with
eight supercharges (N = 2 supersymmetry) [2,3] in order to try to better understand the
large N limit of four-dimensional gauge theories. Some basic qualitative features were first
discovered by the author while working on simple two-dimensional toy models, which can
E-mail address: fferrari@feynman.princeton.edu (F. Ferrari).
1 On leave of absence from Centre National de la Recherche Scientifique, Laboratoire de Physique Thorique
de lcole Normale Suprieure, Paris, France.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 7 0 - 4
152
F. Ferrari / Nuclear Physics B 612 (2001) 151170
be supersymmetric [4], but also non supersymmetric [5,6]. In four dimensions, we can
unfortunately provide explicit calculations for N = 2 theories only.
The paper is organized as follows. In Section 2, we give a general qualitative discussion
of some of our main results, without entering into technical details. In Sections 3 and 4,
we discuss in turn the structure of the moduli space at large N , the nature and properties of
the large N expansion of the low energy observables, and two full large N calculations
of representative SeibergWitten period integrals. In Section 5 we discuss some open
problems.
2. General discussion
Large N in SU(N) gauge theories and planar diagrams From the point of view of
perturbation theory, the large N expansion is a reordering of Feynman diagrams with
respect to their topology, with diagrams of genus h being proportional to 1/N 2h . The
original perturbative series has zero radius of convergence, but the large N reordering
can produce convergent series in the gauge coupling constant at each order in 1/N 2 , thus
providing a non-perturbative treatment. The series in 1/N 2 is reminiscent of the genus
expansion of an oriented closed string theory [7] of coupling constant 1/N , and
one may hope that a full non-perturbative equivalence between string theories and gauge
theories could exist. This is supported by the empirical facts that hadrons are found on a
Regge trajectory, and that quarks seem to be confined by the collimation of Faraday flux
lines in string-like structures. In a supersymmetric context, one can make this idea more
precise by using the fact that D-branes, on which gauge theories live, can be viewed as
solitons in closed string theories [8].
Large N and instantons A general amplitude in gauge theory can pick up contributions
of different types. In addition to the terms having a Feynman diagram representation that
are discussed in the context of the t Hooft expansion, non-trivial field configurations
invisible in Feynman diagrams can in principle contribute to the path integral. The most
popular configurations of this type are instantons [9,10]. Instantons are responsible for
important semi-classical effects like tunneling. In the case of N = 2 supersymmetric gauge
theories, on which we will focus in the present paper, a non-renormalization theorem
[11] implies that the low energy effective action Seff up to two derivatives terms picks
up only a trivial one-loop term from Feynman diagrams, but has an infinite series of
instanton contributions. From the point of view of the large N limit, however, instantons
are exponentially suppressed, and thus do not seem to play an interesting rle [12]. The
2 1/N . For N = 2
instanton action is indeed proportional to N in the t Hooft scaling gYM
super YangMills, this seems to imply that Seff at large N is simply given by the one-loop
formula. One would have to look at higher derivative terms in order to obtain a non-trivial
large N expansion.
More precisely, and as was emphasized in [12], the effects of instantons of size 1/v are
proportional in the one-loop approximation to
2
8 2 /gYM

=
153
0
,
(1)
where gYM is the gauge coupling constant at scale v, the dynamically generated scale
of the theory, and 0 N is given by the one-loop function. The one-loop formula (1)
is exact for N = 2 super YangMills, with 0 = 2N . It suggests that the only smooth
limit of instanton contributions when N is zero [12]. Large instantons (small v), if
relevant, would produce catastrophic exponentially large contributions, and if one is willing
to assume that the large N limit makes sense the only physically sensible conclusion
is that instantons are irrelevant variables. In real-world QCD, instantons of all sizes can
potentially contribute, and this led Witten to argue that the instanton gas must vanish [12].
In Higgs theories, like N = 2 super YangMills, the Higgs vevs introduce a natural cutoff
v on the size of instantons, and for v large enough (weak coupling) the instanton gas can
exist but is just negligible at large N . At small v (strong coupling), we run into the same
difficulties as in QCD, and instantons must somehow disappear.
Large N and fractional instantons An interesting puzzle is to try to understand the nature
of the physics at strong coupling in the large N limit. Unlike in non-supersymmetric
theories where one may argue that the physics is dominated by planar diagrams, in N = 2
gauge theories we have already pointed out that some important observables do not have
any non-trivial contribution from Feynman diagrams. In their derivation of the low energy
effective action Seff , Seiberg and Witten [2] made use of the electric/magnetic duality
transformations of the low energy abelian theory. These transformations are very useful
to understand the physics near singularities where magnetically charged particles become
massless, but they say nothing about the large N limit at strong coupling. The purpose of
the present paper is to elucidate this problem. We will see that instantons actually do not
literally vanish, but disintegrate into fractional instantons of topological charge 1/(2N) (or
a multiple of 1/(2N)). Such fractional instantons give contributions of order
0 /2N
= ,
(2)
v
v
and can obviously survive at large N . The fact that fractional instantons could play a
rle in gauge theories has been suspected for some time, independently of the large N
approximation. In the case of N = 1 supersymmetric gauge theories, chiral symmetry is
broken by a gluino condensate = 3 . Since 0 = 3N in this context, we see that
the gluino condensate can be interpreted as coming from a fractional instanton of charge
1/N (a recent discussion of the gluino condensate can be found in [13]; see also [14] for
a discussion of the closely related problem of N = 1 superpotentials in confining vacua).
In the following we will see that fractional instantons do play a fundamental rle in N = 2
gauge theories as well, when the theory is analysed at large N .
Fractional instantons is a new type of contribution in the large N limit. Though it is
known that Feynman diagrams generate a series in 1/N 2 and that instantons must be
exponentially suppressed, it is not obvious what is the analogous statement for fractional
instantons. We will explicitly demonstrate below that in N = 2 gauge theories, fractional
154
instantons generate a series in 1/N . The leading contribution can be of the same order of
magnitude as the one coming from planar graphs. The first correction is then of order
1/N , and would dominate any subleading 1/N 2 corrections from Feynman diagrams.
This proves directly from the field theory point of view that the string theory dual must
contain open strings in addition to the familiar closed strings. This fact is actually consistent
with our present knowledge on string duals to N = 2 gauge theories. In the supergravity
approximation, the closed string backgrounds dual to such theories usually have unphysical
naked singularities called repulsons [15]. String theory does actually make sense on such
backgrounds thanks to the enhanon mechanism [16]: the constituent branes expand and
form a finite shell excising the singularity. The open strings responsible for the fractional
instantons contributions must be open strings attached to these branes. Such open strings,
and the fact that they would produce unusual 1/N corrections on the field theory side, do
not seem to have been discussed in the literature. It would be extremely interesting if they
could be used to match with non-trivial results of the type we are going to derive on the
gauge theory side.
Large N and singularities Possibly the most interesting result derived in this paper
concerns the behaviour of the large N expansion at singularities on the moduli space.
These results were anticipated in our study of two-dimensional models [46].
On the moduli space of vacua, the gauge group SU(N) is broken down to U(1)N1 ,
and generically the low energy theory is a simple pure N = 2 abelian gauge theory. For
some particular, critical, values of the moduli, however, some additional hypermultiplets
can become massless [2]. In the pure SU(N) case on which we focus, such critical points
can only be observed at strong coupling (the W bosons have then a mass of order ) and
correspond to massless magnetically charged particles called dyons. Since those couple
non-locally to the original photons, it is a priori not obvious what the low energy theory
can be, but we know from [2] that the dyons couple locally to dual photons and generally
produce a trivial free abelian theory in the IR. When the light particles are not mutually
local with respect to each other, which can be achieved by adjusting several moduli to
special values (higher critical point), a genuinely non-trivial CFT develops in the IR [17].
In gauge theories with quark hypermultiplets, very general types of critical behaviour can
be obtained, see, e.g., [18].
Commonly, trying to find a good approximation scheme to describe a non-trivial critical
point can be subtle in field theory. A typical example is 4 theory in dimension D. The
theory has two parameters, the bare mass m (or temperature) and the bare coupling
constant g. By adjusting the temperature, we can go to a point where we have massless
degrees of freedom, and a non-trivial Ising CFT in the IR. The difficulty is that the
renormalized fixed point coupling g is large, and thus ordinary perturbation theory in
g fails. It is meaningless to try to calculate universal quantities like critical exponents
as power series in g, since those are g-independent. Either the tree-level, g-independent
contributions are exact and the corrections vanish (this occurs above the critical dimension,
which is Dc = 4 for the Ising model, and we have a trivial fixed point g = 0 well-described
by mean field theory), or the expansion parameter corresponds to a relevant operator and
155
corrections to mean field theory are plagued by untamable IR divergencies. The way out
of this problem is to use expansion parameters on which the low energy CFT depends
(corresponding in some sense to marginal or nearly marginal operators): we can use an
expansion by going to 4 dimensions, or a large N expansion by considering O(N)
vector models and O(N) invariant critical points.
The critical points on the moduli space of N = 2 super YangMills are very similar
to the Ising critical point below the critical dimension, and 1/N is very similar to the
4 coupling g. On any generic finite-dimensional submanifold of the moduli space,
one finds critical points, characterized by a set of critical exponents [1719], that are
independent of the number of colours N of the gauge theory in which the CFT is embedded.
These critical exponents cannot consistently be calculated in a 1/N expansion. Even the
simple monopole critical points that are known to be trivial are not described consistently
by the N limit of the original gauge theory, because electric-magnetic duality
is not implemented naturally in this approximation scheme. We will indeed explicitly
demonstrate in Section 4 that, though the leading large N approximation for the critical
points is smooth (mean field theory), the 1/N corrections are IR divergent.
Note that by adjusting a large number N of moduli, one can also study N -dependent
critical points. For these, the large N expansion can certainly be consistent. The N special
vacua where a maximum number N 1 of mutually local dyons become massless and
which where studied in [20] because of their relationship with N = 1 vacua are of this
type.
3. The structure of moduli space at large N

3.1. Brief review of N = 2 super YangMills
The fields of pure SU(N), N = 2 supersymmetric gauge theory all transform in the
adjoint representation of the gauge group and make up an N = 2 vector multiplet which
contains the gluons, gluinos and a complex scalar field with scalar potential
V=
2
1
tr , .
2
g
(3)
The D-flatness conditions [, ] = 0 are solved by

= diag(a1 , . . . , aN )
(4)
with
N

ai = 0.
(5)
i=1
N1 , with W bosons of masses
Classically,
the gauge group is generically broken to U(1)
mij = 2 |ai aj |. When some of the ai s coincide, the gauge symmetry is enhanced.
Quantum mechanically, the moduli space is parametrized by the vevs of the gauge invariant
156
operators

ui = tr i ,
2 i N,
(6)
or equivalently by the i s defined implicitly by the relations

uk =
N

ik ,
i=1
N

i = 0.
(7)
i=1
The ai s are non-trivial functions of the i s, with ai i only at weak coupling. In

particular, it turns out to be impossible to have ai = aj , which implies that a non-abelian
gauge group is never restored in the quantum theory.
The low energy effective action is then generically a pure abelian U(1)N1 theory. The
leading terms in a derivative expansion can be written in N = 1 superspace as

1
1
4
4
i
2
i j
Im d x
d i F (A)A +
d i j F (A)W W ,
Seff =
(8)
4
2

where the (Ai , Wi )s are the massless abelian N = 2 vector multiplets satisfying i Ai = 0
and F (A) is a holomorphic function called the prepotential.
According to Gausss law, the central charge Z of the supersymmetry algebra, and the
masses
MBPS = 2 |Z|
(9)
of the BPS states, only depend on the values of the fields at large distances, and thus they
can be expressed in terms of F only. Introducing the dual variables
aDi =
F
ai
(10)
we have [2]
Z=
N

(qi ai + hi aDi ),
(11)
i=1

where the integers qi , i qi = 0, and hi , i hi = 0, are the electric and magnetic quantum
numbers, respectively. Low energy electric/magnetic duality is manifest on the formulas
(8) and (11) which are Sp(2(N 1), Z) invariant, and it turns out that the variables a and
aD are most naturally interpreted as sections of an Sp(2(N 1), Z) bundle.
If one introduces the genus N 1 hyperelliptic curve C defined by
C :Y2 =
N

(X i )2 2N = P (X)2 2N
(12)
i=1
equipped with a canonical basis of homology cycles (i , i ), and the differential form
SW =
X dP
,
2iY
(13)
then a and aD are simply given by [2,3]

ai = SW ,
aDi = SW .
157
(14)
It is natural to rewrite the equation for C as

Y 2 = P (X) N P (X) + N = P+ (X)P (X)
N

= (X Xi,+ )(X Xi, )
(15)
i=1
and to view the curve C as two copies of the complex plane with cuts joining the branching
points Xi,+ and Xi, . The electric contour i defining ai encircles the cut from Xi,+
to Xi, . At weak coupling (small ), one then recovers ai i . For our purposes,
however, it is not particularly useful to try to make a distinction between electric and
magnetic variables. At strong coupling, the two notions are mixed by the non-trivial
monodromies [2]. The important point is that there is a one-to-one correspondence between
possible BPS states and the integer homology of the curve C, in such a way that to any cycle
H1 (C, Z) we can associate the central charge

Z( ) = SW .
(16)
Singularities on the moduli space are obtained when H1 (C, Z) degenerates, or equivalently
when two of the branching points coincide. Since the polynomials P+ and P obviously
have no common roots, this can happen only when the discriminants of P+ or P
independently vanish,

(P+ ) = (Xi,+ Xj,+ )2 = 0 or (P ) = (Xi, Xj, )2 = 0.
(17)
i<j
i<j
3.2. The large N limit

3.2.1. Generalities
At large N , it is convenient to parametrize the (N 1)-dimensional moduli space by the
density function
N () =
N
1 (2)
( i ),
N
(18)
i=1
which satisfies

N () d 2 = 1,

N () d 2 = 0.
(19)
At weak coupling, N () simply gives the density of complex eigenvalues of the Higgs
field. The differences i j are related to the masses of the W bosons. If the lightest
W has a mass of order mW , then the heaviest W will generically have a mass of order
158
MW N mW . Particles having a mass growing with N are problematic in the large N

expansion. This is due to the fact that the amplitude for the propagation of a state of mass
M for a time t involves the factor exp(iMt) (this argument was used long ago by Witten
for baryons [21]). A consistent large N limit, for which propagators of W bosons have a
smooth expansion, can thus be achieved only if the heaviest Wshave a mass of order N 0 ,
and thus the lightest Ws have generically a mass of order 1/ N . For one-dimensional
distributions corresponding to the case where all the i s are aligned, the lightest Ws have
a mass of order 1/N . This correct large N scaling was already considered in [20]. Under
the above conditions, the distribution N () will typically have an N limit ()
which is the sum of a smooth function with compact support plus function terms. We
will consider examples below.
A good strategy to study the large N , (N 1)-dimensional, moduli space, is to consider
well-chosen finite-dimensional subspaces. The simplest subspaces M are one-complex
dimensional and are parametrized by a complex mass scale v. One picks up a particular
(v)
fixed distribution i characterized by () and considers i = vi or equivalently

1
.
v () = 2
(20)
v
v
When () is a smooth function, then the limit v is a weak coupling limit since
the lightest Ws have a mass proportional to v. If () has some function singularities,
which means that a large number (of order N ) of the i s coincide, then we are at strong
coupling for any v. The study of the simple subspaces M can reveal a great deal of
physics, including the transition from weak coupling to strong coupling and the presence
of the simplest singularities. To obtain more complicated singularities (kth order critical
points), one has to consider k-dimensional subspaces, or introduce quark flavors and
quark mass parameters. For the purposes of the present paper, it is enough to stick to
the one-dimensional subspaces M defined by (20). It is then natural to introduce the
dimensionless ratio
v
and the polynomials

r=
p(x) =
(21)
N

(x i ),
p (x) = p(x) 1/r N ,
(22)
i=1
and to rewrite the fundamental equations (12), (13) and (16) in terms of dimensionless
variables,
C : y 2 = p(x)2 1/r 2N = p+ (x)p (x),
= SW /v = x dp/(2iy),

z( ) = Z( )/v = .
(23)
(24)
(25)
159
3.2.2. Singularities
The locus of singularities S corresponds to the points where the discriminants (17)
vanish, or equivalently when one of the polynomials p+ or p has a multiple root.
The discriminants are themselves polynomials of degree N 1 in r N , and thus we will
have generically 2N(N 1) singularities where a dyon becomes massless on M . These
singularities are arranged on 2(N 1) concentric circles. To better understand the structure
of S, it is useful to study first the large N distribution of the roots of the polynomials p .
The equation p (x) = 0 can be rewritten

lnr N p(x) + i arg r N p(x) = 0.
(26)
The real part of the equation is of order N while the imaginary part is of order N 0 . In the
leading large N approximation, both p+ = 0 and p = 0 thus reduce to
def
EN (x) = |p(x)|1/N = 1/|r|.

The N limit of the envelope EN is easily evaluated,

E (x) = exp d 2 () ln | x|.
(27)
(28)
When x is in the support of , the above approximate formula only gives an average value
of EN , which in fact has very sharp wells around the i s, EN (i ) = 0. In the large N
expansion, the wells are approximated by infinitely thin lines joining the smooth envelope
E (see Fig. 1, for an example). Any correction to this picture would be exponentially
small at large N , an instanton effect. At large |x|,

E (x) = |x| 1 + O(1/|x|) .
(29)
Eq. (27) is now easily solved. When 1/|r| < minxC E (x), the plane z = 1/|r| can only
intersect z = EN along the thin wells: the roots of p+ and p coincide with the classical
roots i . This corresponds to the instanton dominated region, and those are suppressed at
Fig. 1. Plot of the function EN=25 (x) corresponding to the distribution (32) with i = 2(i 1)/N 1
(1 i N ) (thin line) together with the envelope E (thick line).
160
large N . When |r| decreases and 1/|r| > minxC E (x), z = 1/|r| and z = EN continues
to meet along some of the thin wells, but the roots nearest to the minimum of E are now
arranged along an inflating curve. This curve gives the general shape of the enhanon
discussed in [16]. Note that the enhanon exists for all r if the distribution has some
function singularities. It has in general several connected components associated with
the different connected components of the support of . At large 1/|r|, all the roots are
located on the enhanon whose various connected components eventually merge to form
asymptotically a circle of radius 1/|r| according to (29).
For example, let us consider the simple uniform distribution

1/, if || < 1,
0 () =
(30)
0,
if || > 1.
It yields an envelope

(|x|2 1)/2 , if |x| < 1,
E,0 (x) = e
(31)
|x|,
if |x| > 1.
For |r| > e, allthe roots are classical. At |r| = e, the enhanon begins to inflate. It is a
circle of radius 1 ln |r|2 . At |r| = 1, the enhanon has eaten up all the classical roots,
and for |r| < 1 it is simply a circle of radius 1/|r|. For the purpose of illustration, we have
depicted in Fig. 1 the case of a unidimensional uniform distribution

(Im())/2, if || < 1,
1 () =
(32)
0,
if || > 1,
for which
(1 x)(1x)/2(1 x)(1+x)/2/e, if x < 1,

E,1 (x R) = (1 + x)(1+x)/2(1 x)(1x)/2/e,
(33)
if 1 < x < 1,
(1 + x)(1+x)/2(1 + x)(1x)/2/e, if x > 1.

Now that we understand the location of the roots of p , it is straightforward to deduce
the singularity locus. Generically, roots are either equal to the classical values, or are
smoothly distributed along the enhanon. Singularities can then occur only when the
classical roots are eaten up by the inflating enhanon, or when two connected components
of the enhanon merge. This is particularly transparent on Fig. 1, where real roots join the
enhanon by pairs and then become complex. We thus deduce that the singularities are
densely distributed on rings in the r-plane corresponding to the melting of classical roots
with the enhanon (singularities of the first class) or with the melting of several connected
components of the enhanon with each other (singularities of the second class). In Fig. 2 we
have represented the space M0 with the associated singularity locus S (only singularities
of the first class are present in this example). We will see in Section 4 that the large N
expansion is afflicted with power-like divergences near singularities of the second class,
while logarithmic divergencies also mix up near singularities of the first class.
3.2.3. Special cases

We have discussed above how the singularity locus S can be deduced for a given
density . It is amusing to study the inverse question: can we find a density that would
161
Fig. 2. The one-complex dimensional subspace M0 corresponding to the uniform distribution (30)
with the associated ring of singularity S.
correspond to an a priori given locus S? This amounts to inversing the functional relation
(28) between E and . If we limit ourselves to the consideration of unidimensional
distributions (by constraining all the i s to be real), the mathematical problem is very
similar to the one encountered in the study of the leading N limit of simple matrix
integrals [22]. If we introduce

()
,
(x) = d
(34)
x
we have

i
( + i) ( i) ,
2
and (28) yields
() =
(35)

1 dE 1
= ( + i) + ( i) .
E dx
2
(36)
As an application, let us try to find a simple unidimensional smooth density with a

connected compact support [1, 1] for which all the singularities are distributed on a single
circle (a singularity ring of zero thickness). From the discussion of Section 3.2.2 it is clear
that this can only happen when dE /dx = 0 on the support of . The only solution to (36)
consistent with (x) 1/x at infinity is then
1
,
(x) =
(x 1)(x + 1)
(37)
and (35) implies

() =
1 2
(38)
This is the density distribution of the roots of the Chebyshev polynomials TN , for which
i = cos((1/2 + i)/N). This very special distribution was studied in [20]. Eq. (28) then
gives the radius of the singularity circle |r| = 2. Clearly, many generalisations could be
studied.
162
3.2.4. Corrections to N =
The singularity rings of the first class separate an outer purely instanton-dominated
region from an inner strongly coupled region which cannot be described by semi-classical
physics (the outer region only exists when does not have any function singularities).
In the outer region, corrections at large N are exponentially suppressed. To the contrary,
we can a priori expect non-trivial corrections in the inner, strongly coupled, regions of
moduli space. We will elucidate the nature and the main properties of these corrections in
Section 4, but we can have a first instructive look at them by investigating the envelope
E or equivalently the shape of the enhanon on a simple example. Let us choose i =
2(i 1)/N 1 as in Fig. 1, and let us calculate the corrections to (33) for x > 1. By
applying Eulers formula

n1

ba
(b a)(f (b) f (a))
ba
= f
+ O 1/n2
f a+k
n
n
2n
b
k=0
(39)
we obtain

x1
1
2
EN,1 (x) = E,1 (x) 1
ln
+ O 1/N
,
2N x + 1
for x > 1.
(40)
This formula displays two of the main features of the corrections to N = : they are
of order 1/N (not of order 1/N 2 ) and they diverge near the singularity locus which
corresponds here to |r| (1/2)e+ or equivalently x 1+ (see (27) and (33)). A third
feature, that the corrections are given by series of fractional instantons, is also to be
expected if one consider that (27) depends on 1/r.
4. Calculation of the 1/N and 1/N 2 corrections
This section is devoted to a detailed study of the corrections to the leading N
approximation in two particular, but generic, examples, encompassing the cases of both
types of singularities (first class and second class). Since we want to consider the purely
strongly coupled region, where the physics cannot be described by instantons, we can put
most of the singularities at infinity. Yet, we will keep one singular circle at finite distance,
so that we can discuss the physics associated with singularities.
4.1. An example with a first class singularity
We choose the distribution N to be
N 1
1
( + 1/N) + ( + 1/N 1).
(41)
N
N
The associated function EN and envelope E are depicted in Fig. 3. The parameter r
will be real and positive, without loss of generality (the final formulas can be trivially
analytically continued). We have a singularity of the first class corresponding to the melting
of the classical root xcl = 1 1/N with the enhanon.
N () =
163
Fig. 3. Plot of the function EN=25 (x) corresponding to the distribution (41) (thin line) together
with the envelope E (thick line). The roots x and x1 of the polynomial p discussed in the text
(Eqs. (44) and (45)) are also indicated for a particular value of 1/r.
4.1.1. Infrared divergences and fractional instantons

A very simple, yet very instructive, calculation one can do is the critical value rc of r for
which the singularity occurs. The leading N approximation rc = 1 can be read off
immediately from Fig. 3. The exact value can be found by noting that it corresponds to the
point where the two positive real roots of the polynomial p defined by (22) coincide,
rc =
N
.
(N 1)11/N
(42)
The large N expansion of the exact formula (42) has strange looking logarithmic terms,
ln N + 1
(43)
+ o(1/N).
N
A standard, well-behaved, large N expansion cannot produce such (ln N)/N corrections.
Remarkably, the same kind of (ln N)/N corrections to the critical parameter were found
recently by the author in a study of a two-dimensional QFT that was argued to be very
similar to gauge theories with Higgs fields [5,6]. These (ln N)/N corrections were the
consequence of the breakdown of the large N expansion near the critical point due to
infrared divergences (a full calculation can be found in Appendix C of [6]). We will see
below that the qualitative physics in four dimensions is strictly similar to the physics
previously discussed in two dimensions [46].
The next simple calculation one can obviously do is the large N expansion of the roots of
the polynomials p+ and p . These roots also enter explicitly in the SeibergWitten period
integrals (25). We will focus on the two positive real roots of the polynomial p which
exists for r > rc and coincide at r = rc . One of these two roots that we call x is simply

x = 1 1/N 1/r N + O 1/r 2N .
(44)
rc = 1 +
The non-trivial corrections to the classical root xcl = 1 1/N are exponentially small.
The other root x1 has a non-trivial and very interesting large N expansion that can be
164
straightforwardly deduced from (26),

2 2r ln(r 1)

1
1 r + ln(r 1)
+
ln(r 1)
x1 =
+ O 1/N 3 . (45)
2
r
Nr
2N r
r 1
This very simple formula displays all the features that were advertised in Section 2.

In particular, by writing ln(r 1) = ln r + ln(1 1/r) = ln r k1 1/(kr k ), we see
that each order in 1/N is given by a series of fractional instantons of fractional charge
1/(2N), mixed with the one-loop diagram contribution in ln r. At r = 1, the corrections
are blowing up like the logarithm of the mass of the lightest degrees of freedom, an infrared
divergence. Using (44) and (45) to solve x = x1 , one can recover the expansion (43) with
the (ln N)/N correction.
Of course, roots like x1 are not directly observables. The physics is encoded in the
SeibergWitten periods (25), and though they depend on the roots, one might argue that
cancellations could occur and that the final result could be smooth. It is also conceivable
that only corrections in 1/N 2 could show up in the final results. To answer rigorously
these questions, we will compute in the next subsection the SeibergWitten period integral
z corresponding to the cycle which vanishes at r = rc . It turns out that the only potential
divergences come from (45) in this case, and that the qualitative features of the expansion
for x1 and for z are the same.
4.1.2. The calculation of the period
We thus proceed to compute the large N expansion of

z=
1
=
i
x
x1
xp (x)
dx
,
p(x)2 1/r 2N
(46)
with
p(x) = (x + 1/N)N1 (x 1 + 1/N).
(47)
It is convenient to trade the variable x for u = 1/(r(x + 1/N)), in terms of which

N
z=
ir
u1
u
du (1 ru/N)(1 (1 1/N)ru)

,
u2
(1 ru)2 u2N
(48)
with

1
1 + 1/r N + O 1/r 2N ,
r

2

1
ln(r 1)
2r
+
ln(r
1)
+ O 1/N 3 .
ln(r
1)
+
u1 = 1 +
2
N
r 1
2N
u =
(49)
(50)
At large N , it is tempting to neglect the term u2N in (48), which is exponentially small
on most of the integration region. This is correct except for u u1 1, when u2N is not
negligible, and for u u 1/r, when (ru 1)2 can be smaller than u2N . We will thus
165
distinguish three different contributions to z,

z = z0 + z< + z> ,
(51)
with
N
z0 =
ir
u1
u
N
z> =
ir
u1
U
N
z< =
ir
U
u

du
ru
(N 1)ru
1
,
1
u2
N
N
ru 1
(52)

du
1
ru
(N 1)ru
1

, (53)
u2
N
N
ru 1
(ru 1)2 u2N

(N 1)ru
1
ru
du
1

1
, (54)
1
u2
N
N
ru 1
(ru 1)2 u2N
where U ]u , u1 [ is an arbitrary constant, for example, U = (u + u1 )/2. The integral

z0 can be explicitly evaluated, and its expansion is given by

iz0
r 1
1
r 1
(ln(r 1))2
+ O 1/N 3 .
=
+ 1
ln r +
ln(r 1) +
2
N
r
N
Nr
2N r
(55)
Finding the asymptotic expansion of z> at large N is a bit more subtle. The integral would
be exponentially small if not for the integration region near u1 . The term u2N cannot be
neglected in a region of size 1/N around u1 . The idea is to use a new variable w defined
by
u = u1 w/N.
(56)
The large N expansion of the integrand can then be straightforwardly deduced, and one
is left with several definite integrals on the variable w [0, N(u1 U )]. The integration
region can be extended to [0, [ at the expense of neglecting exponentially suppressed
terms. The calculation of the 1/N and 1/N 2 corrections involve five non-trivial definite
integrals that can be explicitly evaluated. Instead of writing down all these complicated
formulas, we have illustrated the procedure on a simpler, but qualitatively similar, example
in the appendix. Fortunately, the final result is simple-looking,

ln 2
1
2
iz>
1
=
+ 2
.
(57)
ln(r 1) + r + ln 2 ln 2
N
Nr
2
24
N r
The evaluation of z< proceeds along the same lines, except that now the relevant integration
region around u = u is exponentially small. The asymptotic expansion to all orders in
1/N can then be obtained in this case by changing the variable to w defined by u = u +
w/r N+1 . We get

iz< N 1
=
ln 2 + O 1/r N .
N
N2
(58)
166
Putting all the contributions together, we finally end up with

r 1
1
r 1
(r 1) ln(r 1)
iz
=
+ ln r +
ln r +
ln 2 +
N
r
N
r
r

1
1
2
1
2
+ 2
+ O 1/N 3 .
ln(r 1) + (ln 2) ln(r 1) + (ln 2)2
2
24
N r 2
(59)
This final formula displays all the remarkable properties of the large N expansion of
N = 2, SU(N) super YangMills theory in four dimensions that we have already discussed
at lenght in the previous sections. We note that genuine divergencies when r 1 show up
only at order 1/N 2 , but non-analytic terms are already present at order 1/N .
4.2. An example with a second class singularity
Let us now pick N to be even for convenience and consider the case of the distribution

1
N () = ( 1) + ( + 1) ,
(60)
2
which corresponds to the polynomial p(x) = (x 2 1)N/2 (see (22)). The enhanon has two
components at large |r| which merge at a second class singularity for |r| = 1. The shape of
the enhanon, together with the roots of p(x)2 1/r 2N are depicted in Fig. 4.
If we take r to be real for concreteness, then the critical value of r is exactly given by
rc = 1
and corresponds to the merging of the branching points

x = 1 1/r 2.
(61)
(62)
Fig. 4. The enhanon for the distribution (60) at r = 1.02, r = 1, r = 0.98 and r = 0.5. The dots
correspond to the actual roots of the polynomial p(x)2 1/r 2N for N = 30. At r = rc = 1, two
roots coincide, the two components of the enhanon merge, and we have a second class singularity.
At small r the enhanon is a circle of radius 1/r.
167
The formulas (61) and (62) are to be compared with (42)(45). They exhibit one important
difference between first class and second class singularities: the large N expansion of
the roots (or the shape of the enhanon) is perfectly smooth in the case of second
class singularities. However, for the same general reasons as in the case of first class
singularities, the large N expansion does break down as well. To demonstrate this, we
have to consider the SeibergWitten periods which are the physical observables,

x+
1
du 1 1/(ur 2 )
xp (x)
N
1
dx
=
,
z=
(63)
i
i
u
1 uN
p(x)2 1/r 2N
x
1/r 2
where u = 1/(r 2 (1 x 2 )). A straightforward calculation along the lines of the preceding
subsection or of the appendix then yields the large N expansion

2 1 1/r 2 ln 2
iz
2
2
= 2 1 1/r 2 ln 1 + 1 1/r 2 ln r
N
N

2 /6 2(ln 2)2

+ O 1/N 3 .
+
(64)
2N 2 r 2 1 1/r 2
We get non-analytic terms at leading order, and genuine divergences at order 1/N 2 .
5. Open problems
We have seen that, despite the fact that the low energy effective action of N = 2 super
YangMills is, in some sense, generated by instantons only, it has a surprisingly rich and
interesting large N expansion.
Low-dimensional sections of the moduli space can be easily analysed. For example,
Fig. 2 is very reminiscent of the structure of the moduli space for N = 2. This suggests
that the spectrum of BPS states and the curves of marginal stability could have a simple
description at large N , and that the methods of [23] may be useful.
We have elucidated the fate of large instantons at strong coupling and large N in N = 2
super YangMills: they disintegrate into fractional instantons. These fractional instantons
are responsible for a new class of contributions in the large N expansion, generating a
series in 1/N in addition to the standard series in 1/N 2 from Feynman diagrams. These
new contributions were brought to the fore particularly clearly because we have studied
observables which pick up a single one-loop term from Feynman diagrams. More general
observables will probably have non-trivial contributions from both types of terms. Note
that the leading corrections to N = are dominated by fractional instantons. It would
be very interesting to find direct general arguments explaining why fractional instantons
contribute a well-behaved asymptotic series in 1/N . The fate of instantons in real world
QCD remains of course an open problem, but it is fascinating to contemplate the possibility
that phenomena qualitatively similar to the one studied in the present paper could occur.
This is suggested by the fact that the basic difficulty with instantons IR problems due to
the Landau pole for large instantons are independent of supersymmetry.
168
The standard t Hooft analysis of perturbation theory at large N in SU(N) gauge theories
[1] is consistent with a dual description in terms of closed oriented strings. Our analysis
of the fractional instanton contributions shows that open strings must also be present.
Though adding open strings to a closed string theory is a non-trivial step, it is satisfying
that fractional instanton terms can be interpreted in a string picture at all.
The breakdown of the large N expansion at singularities may come as a surprise (it was
anticipated in the recent studies of two-dimensional models by the author [46]), but we
have argued that the physics from the field theory point of view is the same as the one that
produces IR divergencies in corrections to mean field theory below the critical dimension.
On the other hand, the interpretation from the string theory point of view is much less
clear. A naive application of the UV/IR relation seems to indicate that the string dual is
not UV renormalizable, but this would be a rather strange property for a string theory.
Another possibility, which was pointed out to me by Klebanov, is that the divergencies are
due to tensionless strings at the singularities. This interpretation requires the appearance
of divergences (or at least non-analytic contributions) at leading order, which is not the
case in (59). This might nevertheless be made consistent by correctly identifying the terms
generated by closed strings or open strings diagrams. This point clearly deserves further
investigation.
Another fascinating line of research, suggested by the divergencies of the 1/N expansion
near singularities, is that it might be possible to extract finite universal string amplitudes
by taking N and r rc in a correlated way, along the lines of [24]. A preliminary
investigation will be presented in a forthcoming paper [25].
Acknowledgements
I would like to acknowledge useful discussions with Igor R. Klebanov and Kostas
Skenderis.
Appendix A. A simple toy integral

Let us consider the integral
1
IN =
0
dx
1 xN

1 .
(A.1)
By changing the variable to y = x N , IN is straightforwardly expressed in terms of Euler

beta function,
?(1 + 1/N)
1
1.
IN = B(1/N, 1/2) 1 =
(A.2)
N
?(1/2 + 1/N)
The large N expansion can then be readily found,
IN =

2 ln 2 2(ln 2)2 2 /6
+
+ O 1/N 3 .
N
N2
(A.3)
169
The difficulty one encounters to understand the large N asymptotic expansion directly on
the integral formula (A.1) is the same as for the integral (53) studied in the main text: the
integrand is exponentially small (an instanton effect) except for a region of size 1/N near
x = 1. One then use the variable y defined by x = 1 y/N , and expand the integrand,
which yields
1
IN =
N
N
0
dy
1 ey
1
1
4N 2
N
dy
0

y 2 ey
+ O 1/N 3 .
y
3/2
(1 e )
(A.4)
The integration region can be extended to y [0, [ by neglecting terms of order eN .

The integral at order 1/N is elementary and gives the (2 ln 2)/N correction. The integral
at order 1/N 2 is ofthe type of the integrals encountered in the main text. By going to
the variable y = 1/ 1 ey and integrating by part, it can be related to the dilogarithm
function Li2 and thus explicitly evaluated.
References
[1] G. t Hooft, Nucl. Phys. B 72 (1974) 461.
[2] N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 485, Erratum;
N. Seiberg, E. Witten, Nucl. Phys. B 431 (1994) 484.
[3] P.C. Argyres, A.E. Faraggi, Phys. Rev. Lett. 74 (1995) 3931;
A. Klemm, W. Lerche, S. Yankielowicz, S. Theisen, Phys. Lett. B 344 (1995) 169.
[4] F. Ferrari, The large N expansion, fractional instantons and double scaling limits in two
dimensions, PUPT-1997, LPTENS-01/11, to appear.
[5] F. Ferrari, Phys. Lett. B 496 (2000) 212.
[6] F. Ferrari, JHEP 6 (2001) 57.
[7] A.M. Polyakov, Phys. Lett. B 103 (1981) 207.
[8] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253.
[9] A.M. Polyakov, Phys. Lett. B 59 (1975) 82;
A.A. Belavin, A.M. Polyakov, A. Schwartz, Y. Tyupkin, Phys. Lett. B 59 (1975) 85.
[10] G. t Hooft, Phys. Rev. D 14 (1976) 3432;
G. t Hooft, Phys. Rev. D 18 (1978) 2199, Erratum.
[11] N. Seiberg, Phys. Lett. B 206 (1988) 75.
[12] E. Witten, Nucl. Phys. B 149 (1979) 285.
[13] T.J. Hollowood, V.V. Khoze, W. Lee, M.P. Mattis, Nucl. Phys. B 570 (2000) 241.
[14] N. Dorey, S.P. Kumar, JHEP 2 (2000) 6.
[15] K. Behrndt, Nucl. Phys. B 455 (1995) 188;
R. Kallosh, A. Linde, Phys. Rev. D 52 (1995) 7137;
M. Cvetic, D. Youm, Phys. Lett. B 359 (1995) 87.
[16] C.V. Johnson, A.W. Peet, J. Polchinski, Phys. Rev. D 61 (2000) 86001.
[17] P.C. Argyres, M.R. Douglas, Nucl. Phys. B 448 (1995) 93.
[18] T. Eguchi, K. Hori, K. Ito, S.-K. Yang, Nucl. Phys. B 471 (1996) 430.
[19] P.C. Argyres, M.R. Plesser, N. Seiberg, E. Witten, Nucl. Phys. B 461 (1996) 71.
[20] M.R. Douglas, S.H. Shenker, Nucl. Phys. B 447 (1995) 271.
170
[21] E. Witten, Nucl. Phys. B 160 (1979) 57.

[22] . Brzin, C. Itzykson, G. Parisi, J.B. Zuber, Commun. Math. Phys. 59 (1978) 35.
[23] F. Ferrari, A. Bilal, Nucl. Phys. B 469 (1996) 387;
A. Bilal, F. Ferrari, Nucl. Phys. B 480 (1996) 589;
F. Ferrari, Nucl. Phys. B 501 (1997) 53;
A. Bilal, F. Ferrari, Nucl. Phys. B 516 (1998) 175.
[24] . Brzin, V.A. Kazakov, Phys. Lett. B 236 (1990) 144;
M.R. Douglas, S. Shenker, Nucl. Phys. B 355 (1990) 635;
D.J. Gross, A.A. Migdal, Phys. Rev. Lett. 64 (1990) 127.
[25] F. Ferrari, Exact amplitudes in four-dimensional non-critical string theories, PUPT-1993,
LPTENS-01/10, hep-th/0107096.

How a quarkgluon plasma phase modifies

the bounds on extra dimensions from SN1987a
Daniel Arndt
Department of Physics, University of Washington, Seattle, WA 98195-1560, USA
Abstract
The shape of the neutrino pulse from the supernova SN1987a provides one of the most stringent
constraints on the size of large, compact, gravity-only extra dimensions. Previously, calculations
have been carried out for a newly-born proto-neutron star with a temperature of about 50 MeV at nuclear matter density. It is arguable that, due to the extreme conditions in the interior of the star, matter
might be a quarkgluon plasma, where the relevant degrees of freedom are quarks and gluons rather
than nucleons. We consider an energy-loss scenario where seconds after rebounce the core of the star
consists of a hot and dense quarkgluon plasma. Adopting a simplified model of the plasma we derive
the necessary energy-loss formulae in the soft-radiation limit. The emissivity is found to be comparable to the one for nuclear matter and bounds on the radius of extra dimensions are similar to those
found previously from nuclear matter calculations. 2001 Elsevier Science B.V. All rights reserved.
PACS: 04.50; 97.60.B; 12.38.M
1. Introduction
Ideas that we live in a world consisting of d compact extra dimensions in addition to
the usual four infinite dimensions are not new and date back to the 1930s. They imply
that there are KaluzaKlein (KK)-modes corresponding to excitations of ordinary standard
model particles in the extra dimensions. Such modes, however, have not been seen in any
collider experiment so far [1].
Recently, a variation of this concept has been revitalized by Arkani-Hamed et al. [24]
who considered an alternative picture in which standard model fields are confined to a
four-dimensional brane while gravity, constituting the dynamics of spacetime itself, is
allowed to propagate into the whole (4 + d)-dimensional bulk. This picture caused some
excitement because it provides a natural solution to the hierarchy problem. As detailed
E-mail address: arndt@u.washington.edu (D. Arndt).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 5 7 - 1
172
D. Arndt / Nuclear Physics B 612 (2001) 171190
in [3], the extension of the four-dimensional space for gravitons leads to a dilution of
gravity into the extra dimensions and therefore to the smallness of Newtons gravitational
constant GN . Matching the Planck mass MPl(4) in the four-dimensional world to the one
in the (4 + d)-dimensional world, MPl(4+d) , via the relation
2+d
2
MPl(4)
R d MPl(4+d)
,
(1.1)
one can bring the Planck mass down from 1019 TeV to the weak scale of 1 TeV by choosing
an appropriate size R of the extra dimensions. 1 For one extra dimension, R 1011 m, in
this case gravity specifically Newtons inverse-square law would be modified on the
scale of astronomical distances, a case clearly ruled out. But already for d = 2, R 1 mm,
which is just out of reach to present experimental measurements of the gravitational force
law [6].
Another way of testing these models is to search for beyond the standard model
physics at the scale MPl(4+d) . Even though one would naively expect a violation of the
standard model at this scale to be of only gravitational strength, the large number of
excited KK-modes at such high energies compensates for the weak coupling between
these KK-modes and the ordinary matter fields. Such effects could alter standard model
predictions due to virtual KK-mode contributions or manifest themselves in the form of
massive graviton production [79]. It might be possible that such effects can be seen in
the form of missing energy at future colliders operating at the TeV scale like the CERN
LHC [10,11].
The most stringent constraints on the size of extra dimensions come from astrophysical
considerations. Recently it has been pointed out that an overproduction of massive KK
modes in the early universe might result in early matter domination and therefore a lower
age of the universe [12].
Another way to obtain bounds on the size of extra dimensions from astrophysics
comes from supernova SN1987a. Our current theory of supernovae predicts the shape and
duration of SN1987as neutrino pulse very well. Hence, if some new channel transports
too much energy from the interior of the supernova then the current understanding of
SN1987as neutrino signal gets invalidated. Raffelt [13] has quantified the maximum
possible emissivity for a new energy loss process which does not conflict with our current
understanding of the neutrino signal as 1019 ergs/g/s. Using this criteria there have been
several calculations [1416] to obtain bounds on the size of the extra dimensions for a
newly born proto-neutron star assuming that the interior of the star consists of nuclear
matter at a temperature of 50 MeV. A more rigorous approach has been undertaken
recently by Hanhart et al. [17] who performed detailed simulations of the effect of exotic
radiation on the neutrino signal.
Although the stars inner core might well consist of nuclear matter, because of the
extreme conditions seconds after core bounce with a density in some regions possibly up
1 A different way of solving the hierarchy problem has been proposed by Randall and Sundrum [5]. In their
model the metric is non-factorizable but rather the four-dimensional metric is multiplied by an exponential warp
factor. We do not consider this model here.
173
to 15 times the nuclear matter density n0 = 0.17 fm3 and a temperature of maybe up
to 100 MeV, matter is near a phase transition to a deconfined QCD plasma.
It is therefore important to investigate how bounds on the size of extra dimensions
change if a scenario is considered where, in contrast to taking nucleons as the relevant
degrees of freedom, the calculation is carried out using quarks and gluons as degrees of
freedom. However, in the regime we are considering here matter is not in a state where
the density and temperature are so high that a first order perturbative calculation would
suffice; it is rather in a regime where deconfinement just sets in and the QCD coupling
constant is still large (s 1). For this reason we will carry out our calculations for a
(unphysical) regime of very high density (up to 107n0 ). There, the coupling is weak and
perturbative calculations are feasible. From there we extrapolate down to get the emissivity
in the physical region.
Note that if the temperature falls below a critical temperature Tc of 50 MeV (for a
density of a few times nuclear density) this very dense plasma most likely undergoes a
phase-transition to a color superconducting phase (for a recent review see [18]). Since we
do not consider such a scenario here, we assume the temperature to be above Tc .
The paper is organized as follows: in Section 2 we calculate the amplitude for the
emission of soft gravi- and dilastrahlung into KK-modes from quarkquark (qq) scattering
in a degenerate QCD plasma. This calculation is fairly general, it applies to soft radiation of
KK-modes from light relativistic fermions. Then, in Section 3, we consider a simple model
of the plasma consisting of three light quark flavors to estimate the size of the qq scattering
cross sections. To treat the occurring divergences for colinear scattering we incorporate the
effects of the surrounding plasma into the gluon propagator by introducing a cutoff mass.
We do not consider any other many-body effects; our goal is to obtain an estimate for the qq
scattering cross section, an exact calculation is presently not feasible. In Section 4, we carry
out the phase-space integration to get a formula for numerical calculation of the emissivity
of a gas of weakly interacting quarks. We also provide an approximation formula for the
degenerate limit. Finally, in Section 5, we calculate the emissivity and the bounds on the
size of the d extra dimensions for the case of d = 2 and d = 3 and compare them to the
result from previous nuclear matter calculations.
2. 4D-graviton, KK-graviton, and KK-dilaton radiation from quarkquark

scattering
We assume the d extra dimensions to form a d-torus with radius R so that the
KK-graviton and KK-dilation mode expansions are given by

j y
,j
and
h
(x) exp i
h (x, y) =
R
j
ab (x, y) =

j

j y
j
hab (x) exp i
,
R
(2.1)
174
respectively. In addition to these particles there emerge massive spin-1 particles from the
KK reduction of the FierzPauli Lagrangian [7] which decouple from matter and need not
be considered here. 2

The coupling of the 4D-graviton g h, KK-graviton hj , and KK-dilaton j to the
energymomentum tensor T of the matter field is governed by the Lagrange density

2

j
L = h T +
(2.2)
h,j T +
,
aa T
2
3(d + 2)
j
where is defined as = 16GN with GN being Newtons gravitational constant. Note

that in Eq. (2.2) the massless graviton h being the zero-mode of the h,j modes
has been written separately for clarity. The energymomentum tensor T for a free quark
is given by

i
i
T = g
(2.3)
m
.
2
2
As we will discuss in the next section, the leading contribution from gravitational
emission for the degenerate quarkgluon plasma of our model comes from the qq
scattering processes qq qqX, where a low-momentum 4D-graviton (X = g), KKgraviton (X = h), or KK-dilaton (X = ) is emitted.
2.1. Amplitude for qq qqX in the soft limit
For a degenerate plasma the participating quarks are near the Fermi surface and hence
the momentum k of the emitted graviton is small compared to the quark momenta. In the
soft limit the leading diagrams for the process qq qqX are those where the graviton
radiates off one of the external legs as shown in Fig. 1 (bremsstrahlung process). For small
k these diagrams are O(k 1 ) while diagrams where the graviton couples to some internal
line are O(1) and therefore suppressed [2022].
The amplitude M of this process then factorizes into the amplitude A (which includes
a sum over spins and colors of the incoming and outgoing quarks) for on-shell qq qq
M can be
scattering and the conserved gravitational current T . To lowest order in k,
written as [16]
1

A,
j pj pj
2
pj k
(2.4)
1,
if j = 1, 2,
1, if j = 3, 4.
(2.5)
M =
j =1
where
j =
2 Note that while the toroidal compactification is conceptually simple, it might not be realistic since it does
not take into account curvature (warping) caused by fields in the bulk and on the brane. As has been recently
investigated by Fox [19], weak warping might increase the emissivity by a factor 2 and therefore strengthen
the bounds.
175
Fig. 1. The leading order diagrams for the process qq qqX. In the soft limit these diagrams are
dominant over those where the graviton couples to some internal line. A is the on-shell amplitude
for quarkquark scattering.
Here, the pj denote the four-momenta of the incoming (j = 1, 2) and outgoing (j = 3, 4)

is the gravitons four-momentum. The result Eq. (2.4) has been
quarks while k = (, k)
calculated by Weinberg [23] using an entirely classical derivation. This is not unexpected
since to lowest order in k graphs where the radiation comes from an internal line are
suppressed and so the radiation from the external legs is the dominant one, corresponding
to classical bremsstrahlung from the gravitational charges accelerated during the collision.
2.2. Energy-loss formulae
The energy loss into a KK-mode j and into the momentum space volume element d 3 k
due to 4D-graviton, KK-graviton, and KK-dilation emission is given by

d 3 k
) M ' (k,
)
'
(
k,
M
(2)3 2

d 3k
1 2
=
M M M ,
2
(2)3 2
d'g =
d 3 k j j
M ' (k, ) M ' (k, ),
(2)3 2
(2.6)
j
d'h =
(2.7)
=1
and
j
d' =
d 3k
2
M M
3(d + 2) (2)3 2
2(d 1)
d 3k
,
M M
=
3
2(2)
3(d + 2)
n(n1)/2
j

) ej (k,
)
eii (k,
jj
=1
(2.8)
176

j
j
respectively. Here, ' , ' , and e are the polarization tensors of the 4D-graviton, KKgraviton, and KK-dilaton, respectively [7].
Utilizing the conservation of the energymomentum tensor, k M = 0, M can be
expressed in terms of the purely spacelike components as
kj
Mj i ,
or, upon defining

M0i =
M00 =
Wij = i j i 0
ki kj
Mj i
2
kj
ki kj
ki
0 j + 0 0 2 ,
(2.9)
(2.10)
be written as
M = Wij M ij .
(2.11)
Using this notation the energy loss formulae in Eqs. (2.6)(2.8) can be written solely in
terms of the spacespace components of the tensor M as
j
d'X =
dk k 2 d k
ij kl
Mij X Mkl ,
4 2 4
(2.12)
ij kl
where the X are given by
1
1
1
= ik j l ij kl 2 4k j k l ik k k k l ij k i k j kl + 4 k i k j k k k l ,
2
2k
2k
(2.13)
ij kl
ij
kl
h = W
B W

2
4
= ik j l + il j k ij kl + 4 k i k j k k k l
3
3

1
2
2 ik k j k l + j k k i k l + il k j k k + j l k i k k k i k j kl + k k k l ij ,
3
(2.14)
ij kl
and
ij kl
2(d 1)
= W ij W kl
3(d + 2)

2(d 1)
1
1
.
= ij kl + 2 ij k k k l + kl k i k j + 4 k i k j k k k l
3(d + 2)
(2.15)
The tensor B , defined by the spin sum of the KK-graviton polarization tensors
5
j

j
j
) = B

'
(k, ) '
(k,
(k),
(2.16)
=1
is given in [7,8].
We will now choose the center-of-momentum (COM) frame as our reference frame
and (p0 , p)
and
where the 4-momenta of the two incoming quarks are given by (p0 , p)
177
those of the outgoing quarks by (p 0 , p ) and (p 0 , p ). The COM scattering angle cm
is defined by p p = pp cos cm . We also assume the quarks to have equal mass.
If the momenta of the scattering particles are much smaller than their rest mass M as
is the case in soft radiation from nucleonnucleon scattering it can be neglected in the
denominator in Eq. (2.4) and M is simply given by [16]
Mij =
(pi pj pi pj )A.

M
(2.17)
In our case, however, this approximation is inappropriate since the quark mass is very small
compared to the quark Fermi momenta, we therefore write Eq. (2.4) as

pi pj
pi pj
pi pj
pi pj
A. (2.18)
Mij =
+
2 p0 p k p0 + p k p0 p k p0 + p k

The d k/4
integration in Eq. (2.12) can then be carried out analytically for the three
cases. For soft graviton radiation the quark momenta are assumed to lie on the Fermi
surface and it is justified to assume |p|
|p |. We can then approximate the momenta

p and p by
p 2 =
p2 + p 2
.
2
Replacing p and p with p the formulae simplify considerably and the

can be written as

d k
2 p 2
ij kl
Mij X Mkl =
gX (x, y, cm )|A|2
4
k
(2.19)

d k/4
integration
(2.20)
in terms of A and the dimensionless functions gXwhich are given in the appendix. Here
we have also defined x = mj/ and y = p 0 /p = p 2 + M 2 /p.
Note that for massless quarks one of the denominators in Eq. (2.18) becomes zero if
the momentum of that quark is parallel to k and one might think that this term becomes
infinite. However, as already noted in Ref. [23] for the case of the massless 4D-graviton,
ij kl
this singularity is only spurious and gets canceled by X in the numerator so that gX is
finite not only for X = g, but also for X = h and X = .
We can then employ the energymomentum relation for the KK-mode j and write the
formula for the energy loss per unit frequency interval into the given KK-mode of mass
m2 = 2 k 2 as
j
j
d'X
k
=
d
4 2
2 p 2
d k
ij kl
Mij X Mkl =
gX (x, y, cm )|A|2 .
4
4 2
(2.21)
This result is general as it applies to soft radiation from any light relativistic fermions with
equal mass, not just to the quarks we are about to consider.
178
3. A simple plasma model

After bounce the homologous core of the star undergoing a supernova explosion gets
compressed to a very high density. Previous calculations [1416] assumed a density of
13 times nuclear matter density n0 = 0.17 fm and a temperature of T 3080 MeV
corresponding to a (non-relativistic) nucleon chemical potential of N 50100 MeV.
Under such extreme conditions the stars core not only consists of free neutrons and
protons but these degrees of freedom are highly excited and near a phase transition
to a deconfined quarkgluon plasma. It is therefore worthwhile to investigate how the
graviton emissivity and hence the bounds on extra dimensions change if one carries out
the calculations in this regime. Our intention here is to investigate graviton radiation into
extra dimension from such a QCD plasma.
The physics of a QCD plasma under such violent conditions is particularly rich. In
addition to quarks of different flavors, a variety of other particles are present and being
produced constantly, such as electrons, positrons, neutrinos, and collective excitations
like plasmons. The properties of the quarkgluon plasma are far from being completely
understood so that we do not even attempt to take the whole richness of matter under such
extreme conditions into account here. Instead, we assume a rather simply model for the
plasma (not unlike the one used in Ref. [24]) where, in addition to electrons and neutrinos,
only u, d, and s quarks are assumed to be present. The possible weak reactions in the
plasma are then
d u + e + e
and s u + e + e .
(3.1)
At thermal equilibrium the chemical potentials have to add up to zero. For a low
temperature T this requirement reads
d = u + e +
and s = u + e + .
(3.2)
In addition, we want to satisfy charge neutrality, thereby imposing the constraint

2
1
1
0 = Q = nu nd ns ne .
3
3
3
These three conditions can be simultaneously satisfied by choosing
(3.3)
nq
= nB and ne = n = 0,
(3.4)
3
where nq is the quark number density and nB denotes the baryon number density. We do
not consider dynamical production of these or any heavier quark flavors. As an additional
simplification we take the mass of the u, d, and s quarks to be zero.
The main contribution to graviton radiation from this plasma comes from the bremsstrahlung reaction qq qqX (X = g, h, ) where a graviton is radiated from an external
quark leg. This reaction is of order O(s ) where s is the QCD coupling constant. As
we will see below, s is rather large and O(1). Moreover, we get contributions from
processes involving dynamical produced electrons (e) and positrons (p), like ee eeX
or ep epX. These reactions are suppressed compared to qq qqX by a factor
nu = nd = ns =
179
of QED /s , where QED = 1/137 is the fine-structure constant. More contributions

come from graviton-production reactions like q qXX. These, however, are O( 2 ) and
therefore highly suppressed. We also assume that because of T q the number density
of gluons is much smaller than that of quarks and therefore we do not consider the reaction
qg qX. Note that the inclusion of any KK-modes generated by channels we neglect will
only increase the emissivity and strengthen the bounds obtained below.
The baryon number density nB in the central core region of a stable neutron star is
115 times the nuclear matter density n0 = 0.17 fm3 , depending on the nuclear matter
equation of state used (compare Ref. [24] and values cited therein). Assuming quark matter,
the (relativistic) quark chemical potential q can be related to the quark density nq using
the formula

1
nq = 6 2 d' ' 2 Nq ('),
(3.5)
2
0
where Nq (') = (exp[(' q )/T ] + 1)1 is the mean occupation number and ' is the
(relativistic) quark energy. Choosing nq and T we can then determine q ; for T = 50 MeV
and nq = 1020n0 we find q 500600 MeV, which is in good agreement with other
calculations [24]. Therefore quark matter has q /T 1 which confirms our assumption
that it is degenerate.
Furthermore, using q and T we can calculate the average momentum square p2
carried by the quarks using
3 2
2
d p p Nq (')
p = 3
(3.6)
.
d p Nq (')
For typical parameters for the plasma (nq 1020n0 , T 50 MeV) we calculate the
average momentum carried by the quarks to be p 420520 MeV which is just about
the order of energy at which the strong interaction becomes strong and the QCD coupling
constant s is expected to be rather large. Of course, this is eventually a sign that matter
under these conditions is at the phase transition to the QCD phase. Therefore this regime
is particularly difficult to treat because on the one hand nucleonic degrees of freedom
are no longer the proper degrees of freedom while on the other hand perturbative QCD
calculations are notoriously difficult.
For this reason we will evaluate the emissivity for unphysically high densities of
nq 100107 n0 and extrapolate from this region down to the physical regime. At such
high densities the momentum carried by the quarks is p 140 GeV. This being in
the perturbative regime, we can estimate the strength of the QCD coupling constant s
employing the leading-log formula for the running coupling [derived from the -function
to O(s2 )]
s =
12
g2
=
,
4
(33 2NF ) ln(p2 /2 )
(3.7)
which gives s 0.480.13 small enough for perturbative calculations. To check the
effects of higher order correction we also use the formula
180

412
21 ln u
g2
1 2 2 0 5
4
1 2
s =
+
ln u
=
+ 4 2
4
0 u
2
4
0 u
812
0 ln u
(3.8)
derived from the -function to 4th order in s . Here

2
19
1 = 51 NF ,
0 = 11 NF ,
3
3
5033
325 2
p2
NF +
NF and u = ln 2 ,
2 = 2857
(3.9)
9
27
where 200 MeV is taken as the QCD breakdown scale and NF = 3 is the number of
flavors. The formula in Eq. (3.8) changes s by about 1837% compared to Eq. (3.7).
It is instructive to compare the number of nucleons NN and quarks Nq participating in
scattering for the two cases of nucleonnucleon scattering (for simplicity we assume here
neutron matter) and qq scattering in the degenerate limit. For low temperature the baryon
number density nB is approximately proportional to pF3 . For nucleons nB pF3 /(3 2 ),
while for quarks nB pF3 / 2 . The number of particles NN [Nq ] participating in scattering
is roughly proportional to the momentum-space volume of a shell of the Fermi sphere
with radius pF and thickness dE T leading to NN 8pF2 dp [Nq 72pF2 dp] in
the nucleon [quark] case. Making use of the energymomentum relations p dp = M dE
1/3
2/3
[p dp = p dE] this becomes NN (3 5 )1/3 nB MT [Nq = 72 7/3nB T ]. Using M
940 MeV the ratio of NN and Nq for a density of nB = 10n0 is
NN
0.3
Nq
(3.10)
and hence, despite the fact that the quarks have a higher number of degrees of freedom, the
number of particles participating in scattering is not much different in the two cases.
3.1. Quarkquark scattering cross sections
The matrix elements for the scattering of quarks to first order in s can be calculated
from the one-gluon exchange diagrams and are given in [25]. 3 For quarks of different
flavor the squared matrix element can be written as
s 2 + u2
t2
and for quarks of the same flavor as
2

2
s2 + t 2 2 s2
2
2 2 s +u
+
|Aqq | = 256 s
t2
u2
3 ut
|Aqq |2 = 256 2s2
(3.11)
(3.12)
(see Fig. 2). Here s, t, and u are the usual Mandelstam variables given in the COM frame
for zero quark mass by s = 4p2 = 4 2 , t = 2p2 (1 cos cm ), and u = 2p2 (1 + cos cm ),
where cm is the COM scattering angle.
3 These matrix elements, however, include an averaging [summing] over initial [final] spin and color states.
Since for the calculation of the emissivity we need to sum over spin and color of both the initial and final states,
we have to include an additional factor of 36 in the squared matrix elements.
181
Fig. 2. The lowest order diagrams for scattering of quarks of the same flavor.
Note that similar diagrams exchanging a photon instead of a gluon are suppressed by a
factor of QED /s and need not be considered.
The matrix elements in Eqs. (3.11) and (3.12) can become highly singular because of
the t 2 and u2 in the denominators. The momentum q transferred by the exchanged gluon
vanishes as cm 0 or cm , which corresponds to colinear scattering. On the other
hand, the energymomentum tensor M also vanishes for colinear scattering and one
might think that this would render the singularity spurious. However, this is not the case.
2
An analysis of the functions gX shows that for cm 0 [cm ] gX behaves like cm
4 [ 1/( )4 ].
[ ( cm )2 ] while |Aqq |2 and |Aqq |2 behave like 1/cm
cm
3.2. Curing the divergences
In the case of electronelectron scattering in an electron gas the divergences that one
encounters are of a similar origin and are usually taken care of by including effects which
screen the bare particles arising from the interaction of the electrons and photons with
the surrounding matter. We will employ, without justification, a similar procedure in the
present case of a QCD plasma.
The bare gluon propagator D is related to the gluon self-energy by Dysons
equation

1
= g 2 q 2 + .
D
Following Ref. [26], the matrix element for scattering of like quarks for small momentum
transfer is equal to the one for unlike quarks and can be written as

1
(1 x 2 ) cos 2
2
2
2

,
|Aqq | = 128 s
(3.13)
q 2 + l2 q 2 2 + t
where l and t are the longitudinal and transverse gluon polarization functions given for
q q by

2
x(1 x 2 ) x + 1
x x+1
2
2 x
l = qD 1 ln
(3.14)
and t = qD
+
ln
.
2 x1
2
4
x 1
Here x = /q and qD is the Debye wave number for cold quark matter of three flavors
2 = 6 2 / 2 . Note that for quarks of equal mass in the COM frame = 0
given by qD
s q
and therefore t = 0; the transverse part of the propagator is still unscreened, at least for
t calculated to zeroth order in x. Since T the participating quarks momenta are
near the Fermi surface and hence 0, independent of the chosen reference frame. The
problem of an unscreened transverse gluon propagator can be solved in some cases by
retaining O(x) terms in Eq. (3.14) and is related to frequency-dependent screening [27].
182
Because of these complications we adopt here a much simpler method to incorporate

the complicated screening effects. We will render the amplitudes in Eqs. (3.11) and (3.12)
finite by including a Debye screening mass mD , by substituting
t t m2D
and u u m2D
(3.15)
for u and t in the denominator. For a small coupling constant the Debye mass to lowest
order is given by

1 2
2
2 3 2
mD = g
(3.16)
T +
f
2
2 2
f
which makes it about 6020% of a typical momentum transfer for densities nq =

100107 n0 . To consider higher order corrections to mD and to account for other manybody effects in the QCD plasma at these high densities we carry out the calculation for
the emissivity not only using the value mD from Eq. (3.16) but also the values mD /3 and
3mD , thereby spanning about an order of magnitude in the cutoff mass. As an example we
show in Fig. 3 the dimensionless function B(cm ) defined by

|Aqq |2
B(cm ) = |Aqq |2 +
sin2 cm ,
4
which enters the formula for the emissivity for a varying cutoff mass mD . The factor
1
sin2 cm in B(cm ) is an approximation for 0 dx x d1gX (x, 1, cm) as explained in the
next section. B(cm ) has been calculated for a plasma at the density nq = 105 n0 where
a typical COM momentum is about 8400 MeV and mD 2250 MeV. The variation of
Fig. 3. The function B(cm ) for nq = 105 n0 and for different cutoff masses mD /3 MeV (solid line),
mD MeV (dashed line), and 3mD MeV (dot-dashed line) with mD = 2250 MeV. The typical COM
momentum is 8400 MeV and s is set to 1.
183
mD of about one order of magnitude causes a variation in B(cm ) of about 12 orders of

magnitude.
Other possible ways to regulate these divergences could be to introduce a cutoff of order
mD for the gluon-momentum integration or to cutoff the integration over cm at some small
angle 1/ when calculating observables. The latter technique was used by Weinberg [23]
to calculate the power produced in gravitational radiation from the hydrogen plasma in the
sun.
4. Gravitational emissivity of a gas of relativistic fermions

In order to calculate the emissivity due to graviton radiation from the two-body
scattering reaction qq qqX in a gas of relativistic quarks we have to integrate over
the phase space of incoming (p1 , p2 ) and outgoing (p1 , p2 ) quarks as well as that of the
The formula for the emissivity is then given by
emitted gravitational particle (k).

d 3 pi
d 3 pi
dEX
f1 f2 (1 f1 )(1 f2 )
=
d
dt
(2)3 2Ei (2)3 2Ei
j
i=1,...,2
j
d'X
.
(4.1)
d
Here, again, the X stands for the type of gravitational radiation, massless 4D-gravitons
(X = g), massive KK-gravitons (X = h), or massive KK-dilatons (X = ). The functions
f are Pauli blocking factors given by fi = (exp[(Ei )/T ]+1)1 , furthermore Ei = |pi |
for massless quarks. The summation over the KK-modes j does not apply to the massless
4D-graviton, obviously.
This formula applies for scattering of differently flavored quarks. A symmetry factor of
1/4 would have to be included for scattering of quarks having the same flavor to account
for identical particles in the initial and final states. We will consider this symmetry factor
at the beginning of the next section.
In the soft-radiation limit we can neglect the graviton momentum k in the momentumconserving delta function. Furthermore, by introducing the total momentum P = p1 +
p2 = p1 + p2 and relative initial and final momenta p = (p1 p2 )/2 and p = (p1 p2 )/2
we can reduce the number of integrations by exploiting spherical symmetry and momentum
conservation to get, after inserting Eq. (2.21) for the massless quark case (y = 1),
(2)4 4 (p1 + p2 p1 p2 k)

dEX
2

2
2
2
= 11 8
dP P
dp p
dp p
d
d d cos d cos
dt
2
1
j
1

f1 f2 (1 f1 )(1 f2 )
1
(E1 + E2 E1 E2 )p 2 gX (x, 1, cm )|A(p,
cm )|2 .
E1 E2 E1 E2
(4.2)
Here we have also used the COM frame for the amplitude A. The angles and are
defined by p P = pP cos and p P = p P cos , while is the angle between the
184
projections of p and p on the P-plane, that is

cos cm = cos sin sin + cos cos .
(4.3)
If the mass splitting of the KK-modes becomes comparable to the experimental energy
resolution which is true for about d 6 the summation over the modes j can be
approximated by an integration:

R d
d
j
1
dm m
d1
= R d
d
dx x d1 ,
(4.4)
where d is the number of extra dimensions and d is the surface area of the d-dimensional
sphere given by d = 2 d/2 /=(d/2).
To further simplify the integration we make another change of variables to the initial and
final COM total energies ET and ET given by

2
P
P2
ET = E1 + E2 = M 2 +
+ p2 + Pp cos + M 2 +
+ p2 Pp cos
4
4
(4.5)

and a similar expression for ET . The Jacobian J of the variable transformation (p, p )
(ET , ET ) is given by

ET 4 P 2 (2ET 2 4M 2 P 2 ) cos2
p p
=
J=
ET ET
2 ET 2 4M 2 P 2 (ET2 P 2 cos2 )3/2

(ET ET ), ( ) .
(4.6)
After interchanging the ET and ET integrations and shifting the integration variables the
final result for KK-gravitons and KK-dilatons becomes
GN d
dEX
=
R d
dt
64 7

d
f1 f2 (1 f1 )(1 f2 )

dP

d
p 2
E1 E2 E1 E2
d sin sin P 2 d J p2 p 2
2
d |A(p,
cm )|2
0
1
dx x d1 gX (x, 1, cm ),
(4.7)
where we have defined

= ET 2 M 2 +
P2
4
and = = ET ET .
(4.8)
Note that only the function gX depends on x = mj/. Also, the integration over is
hidden via cm in the function gX and therefore nontrivial.
185
4.1. Approximation formula for degenerate quark matter

Quark matter with a baryon number density nB 10n0 has at a temperature T
50 MeV a quark chemical potential q 500 MeV and is highly degenerate, q /T 1.
At higher densities matter is even more degenerate. Therefore one can assume that radiation
only arises from scattering involving quarks near the Fermi surface in the initial and final
states. Assuming soft radiation the energy of the quarks in Eq. (4.1) can be set to pF
causing a decoupling of the energy and angular integrations:

p4
dEX
= 12F 8 d dE1 dE2 dE1 dE2 f1 f2 (1 f1 )(1 f2 )
dt
2
(E1 + E2 E1 E2 )
j

d'
d1 d2 d1 d2 (p1 + p2 p1 p2 ) X .

d
(4.9)
j
The energy integral is well-known [28,29] and given by

dE1 dE2 dE1 dE2 f1 f2 (1 f1 )(1 f2 )(E1 + E2 E1 E2 )
=
(2 + 4 2 T 2 )
.
6(e/T 1)
(4.10)
Replacing the sum over mj by an integration [see Eq. (4.4)] we can carry out the
-integration to get
p3
dEX
= F 7 GN d R d Yd
dt
32

d sin3
0
1
dcm |A| (cm , p)
dx x d1 gX (x, 1, cm ),
(4.11)
where p = pF sin . The function Yd is given by

Yd =

T 4+d 2
4 =(2 + d) (2 + d) + =(4 + d)(4 + d) ,
6
(4.12)
where (x) is the Riemann zeta function. This reduces the number of integrations to be
carried out numerically to three.
The temperature dependence of the emissivity in Eq. (4.11) can be easily understood.
The degenerate quarks must have energies that lie within T of the Fermi surface and are
contributing one power of T each. Because of the energy conserving Delta function the
energy of the graviton is of order T as well. Each extra dimension contributes one more
power of T from the KK mode. Because only four out of the five energies are independent,
the power of T gets reduced by one and we end up with the T 4+d dependence.
186
Furthermore, for energy loss due to KK-graviton (X = h) radiation the dominant

mode we can replace the x-integration by

1
dx x
d1
gh (x, 1, cm ) =
(2 2 log 2) sin2 cm ,
0.38 sin2 cm ,
for d = 2,
for d = 3,
(4.13)
which introduces an error of about 10% but reduces the number of numerical integrations
to two.
5. Results and comparison to nucleonnucleon scattering

For quark matter with equal number fractions for the three flavors u, d, and s, Xu =
Xd = Xs = 1/3, the total emissivity E can be written as a sum over possible combinations
of two out of the three flavors in the plasma:
E =
1
Eqq Xq2 + Eud Xu Xd + Eus Xu Xs + Eds Xd Xs ,
4
(5.1)
q=u,d,s
where the 1/4 is the symmetry factor for the scattering of like quarks. Then the formula
for the emissivity [Eq. (4.11)] can be used to calculate the total emissivity by simply
substituting

|Aqq |2
1
2
2
|A|
(5.2)
|Aqq | +
.
3
4
Note that for massless quarks the trace over the energymomentum tensor T is zero
and so the KK-dilatons decouple and need not be considered here. Furthermore, the
contribution of the massless 4D-graviton can safely be neglected since it is several orders
of magnitude smaller than that for the massive KK-gravitons.
In Fig. 4 we show the graviton emission rate from quark matter for a density range of
100107n0 for the case of d = 2 extra dimensions and a temperature of T = 50 MeV,
assuming that the size of the extra dimensions is R = 1 mm. The solid line shows the
result using the cutoff mass mD calculated from Eq. (3.16) valid for small s and using the
formula Eq. (3.8) for the QCD coupling constant derived from the -function to O(s4 ).
To estimate the error introduced by calculating only first order diagrams we give also the
result for the emissivity using the (larger) s [see Eq. (3.7)] calculated from the -function
to O(s2 ) (dashed line). This results in an increase in the cross section which gets partially
compensated by a larger cutoff mass. Therefore the emissivities for both values of s are
very similar.
To account for uncertainties in the cutoff mass (higher order corrections, unaccounted
many-body effects) mD we also show the emissivity using mD /3 (upper dotted line) and
3mD (lower dotted line). With mD spanning about one order of magnitude in the cutoff
mass the emissivity varies by about 12 orders of magnitude. We also present the emission
rates from nucleonnucleon scattering [16] for lower densities.
187
Fig. 4. Emissivity from quark matter for high densities of 100107 n0 . The solid [dashed] line shows
the result using the formula for the QCD coupling constant derived from the -function to O(s4 )
[O(s2 )]. The cutoff mass mD is calculated using Eq. (3.16). The upper and lower dotted lines
correspond to a cutoff mass of mD /3 and 3mD , respectively. The results for nuclear matter at lower
densities from Ref. [16] are also included, using the approximations for degenerate (dot-dashed line)
and non-degenerate matter (dot-dot-dashed line). All calculations are for the case of d = 2 extra
dimensions and a temperature of T = 50 MeV; the size of the extra dimensions is taken to be 1 mm.
The gray area denotes the region of realistic density of nq 1020n0 .
Note that it is not sensible to compare the emission rates for quark matter and nuclear
matter for an equal density, since for densities much higher than 5n0 nucleons are certainly
no longer the appropriate degrees of freedom. Similarly, for densities smaller than about
100n0 the QCD coupling constant becomes too large so that perturbative QCD calculations
will no longer be applicable. It is therefore necessary to extrapolate from the two regions
where reliable calculations are possible to the interesting transition region with nq =
1020n0 (gray area in Fig. 4).
Determining the emissivity for the physical density region from Fig. 4 for d = 2 and
carrying out a similar analysis for d = 3, we can now use the Raffelt criterion, limiting the
energy lost into any non-standard physics channel to 1019 ergs/g/s, to calculate bounds on
the size of new extra dimensions. The radius R enters the calculation through the prefactors
in Eqs. (4.7) and (4.11). Our bounds are calculated to be
R < 2.9 104 mm,
R < 3.9 10
mm,
for n = 2
for n = 3
and
(5.3)
(5.4)
which are quite similar to the ones determined for nuclear matter [16].
It should be kept in mind that these bounds become somewhat weaker [stronger] as
mD is taken to larger [smaller] values thereby decreasing [increasing] the emissivity. Our
conservative estimates are therefore weaker than those found from previous nuclear matter
calculations.
188
6. Conclusion
Supernovas like SN1987a provide one of the most stringent constraints on the size of
gravity-only extra dimensions. Bounds on these sizes have been calculated before [1417]
assuming a density of 13n0 and a typical temperature of T 50 MeV where matter is
nucleonic. The relevant degrees of freedom for this regime are protons and neutrons.
In view of the uncertainties with which the parameters governing the condition of matter
at the stars inner core, such as density and temperature, are given, and in view of the
fact that, as can be found from various nuclear matter equations of state, matter in the
stars core lies very close to a phase transition to either a deconfined QCD plasma or,
for lower temperatures, a color superconducting phase, it is worthwhile and important to
investigate how emission rates and bounds on the radius of the extra dimensions change in
these regimes.
In this paper we have calculated emissivities and bounds on the size of gravity-only
extra dimensions from a deconfined quarkgluon plasma in the core of a star seconds
after the supernova core bounce. We have not considered the case where the temperature
falls below the critical temperature Tc defining the transition of quark matter to a color
superconducting state.
By extrapolating the emissivity from a quarkgluon plasma at very high density (nq =
100107 n0 ) down to the physically interesting region of nq 1020n0 we find that the
KK-graviton emissivity is comparable to those found from nuclear matter calculations. We
have examined the error introduced into our calculation due to the fact that we are working
only to first order in the QCD coupling constant. We estimate this error to be at the most
100%. To consider the rather uncertain value of the cutoff mass we have calculated the
emissivity using a cutoff mass which varies about one order of magnitude about the central
value from the lowest order perturbative calculation [Eq. (3.16)]. We find that this causes
an uncertainty in the emissivity of about two orders of magnitude.
We have not examined any other many-body effects which could modify the vacuum
rates for KK-graviton emission, such as the LandauPomeranchukMigdal effect, which
would decrease the emissivity because of multiple quark scattering before the graviton gets
emitted.
Note that the source of uncertainties is not the soft radiation theorem, which is well
suited for the degenerate case, but rather our lack of knowledge about the nuclear physics
nature of QCD. This deficit of understanding to only impacts our calculation of the quark
quark scattering cross section and the treatment of the QCD plasma, but also forces us to
access the interesting transition region only with the aid of extrapolation.
Despite the significant uncertainties in our calculation for quark matter, we believe to
have ruled out the possibility of a significantly larger emissivity compared to the nuclear
matter case. It is up to further investigation to provide better information on the conditions
which govern the state of the matter at the stars core during and shortly after a supernova
event. It is unlikely that the whole core will consist of dense quark matter, more likely is a
scenario where at the cores center matter is deconfined and undergoes a phase transition
to nuclear matter as one goes towards the surface. Furthermore, a better understanding of
189
QCD in the non-perturbative regime and at finite density would enable a reduction of the
uncertainties.
Acknowledgements
I would like to thank Daniel Phillips, Sanjay Reddy, and Martin Savage for very helpful
discussions and for useful comments on the manuscript. I also thank Zacharia Chacko,
Jason Cooke, Patrick Fox, Christoph Hanhart, and Michael Strickland for interesting
conversations during various stages of this project. This work is supported in part by the
US Department of Energy under Grant No. DE-FG03-97ER4014.
Appendix A. Results for the angular integration

The functions gX (x, y, cm) defined in Eq. (2.20) are calculated for the three cases of
4D-graviton (g), KK-graviton (h), and KK-dilaton () to be

(1 + 6y 2 + y 4 ) log y+1
2y 2 + y 4 + 4y 2 cos cm + cos 2cm
1
y1
2
gg =
1 + y +
4
2y
W(x, y, cos cm )

2
2
2
2
2
(y 1) + 2(y + cos cm ) + 2(y + cos cm )W(x, y, cos cm )
log
(y 2 1)2
1
+ {cos cm cos cm },
(A.1)
4
2

2(y 4 + 4y 2 + 1) log y+1x 2
2
2

4(y
1)
1
y
1x
gh =
1 x2 2
+
6
x + y2 1
1 x 2y
12y 2 cos cm + (2y 2 + 1)2 + 3 cos 2cm
(x 2 1)W(x, y, cos cm )

2

(x 2 + y 2 1)2
y2
x2 1
+
cos
+
2
log
cm
(x 2 + y 2 1)2
(x 2 1)
1 x2

y2
+2
+ cos cm W(x, y, cos cm )
1 x2
1
+
1 x 2 {cos cm cos cm },
6
+
and
g =

d 1
1 x2 y2 1
3(d + 2)
+
y2 1
(x 2 1)W(x, y, cos cm )
2 1) log y+1x 2
(y
y2 1
y 1x 2
+
2
2
x +y 1
2 1 x2y
(A.2)
190

2

(x 2 + y 2 1)2
(x 2 1)2
y2
log 2
+2
+ cos cm
(x + y 2 1)2
(x 2 1)2
1 x2

y2
W(x,
y,
cos
+2
+
cos
)
cm
cm
1 x2

d 1
+
1 x 2 y 2 1 {cos cm cos cm }.
3(d + 2)
The function W(x, y, h), introduced for convenience, is defined as

(h + 1)[x 2 2y 2 + (x 2 1)h + 1]
.
W(x, y, h) =
x2 1
(A.3)
(A.4)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.

N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004, hep-ph/9807344.
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257, hepph/9804398.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
C.D. Hoyle et al., hep-ph/0011014.
T. Han, J.D. Lykken, R.-J. Zhang, Phys. Rev. D 59 (1999) 105006, hep-ph/9811350.
G.F. Giudice, R. Rattazzi, J.D. Wells, Nucl. Phys. B 544 (1999) 3, hep-ph/9811291.
J.L. Hewett, Phys. Rev. Lett. 82 (1999) 4765, hep-ph/9811356.
L. Vacavant, I. Hinchliffe, hep-ex/0005033.
I. Antoniadis, K. Benakli, hep-ph/0004240.
M. Fairbairn, hep-ph/0101131.
G.G. Raffelt, Stars as Laboratories for Fundamental Physics, Univ. Chicago Press, 1996.
V. Barger, T. Han, C. Kao, R.J. Zhang, Phys. Lett. B 461 (1999) 34, hep-ph/9905474.
S. Cullen, M. Perelstein, Phys. Rev. Lett. 83 (1999) 268, hep-ph/9903422.
C. Hanhart, D.R. Phillips, S. Reddy, M.J. Savage, nucl-th/0007016.
C. Hanhart, J.A. Pons, D.R. Phillips, S. Reddy, astro-ph/0102063.
K. Rajagopal, F. Wilczek, hep-ph/0011333.
P.J. Fox, hep-ph/0012165.
F.E. Low, Phys. Rev. 110 (1958) 974.
A.L. Adler, Y. Dothan, Phys. Rev. 151 (1966) 1267.
C. Hanhart, D.R. Phillips, S. Reddy, astro-ph/0003445.
S. Weinberg, Gravitation and Cosmology: Principles and Applications of the General Theory
of Relativity, Wiley, 1972.
N. Iwamoto, Ann. Phys. 141 (1982) 1.
B.L. Combridge, J. Kripfganz, J. Ranft, Phys. Lett. B 70 (1977) 234.
H. Heiselberg, C.J. Pethick, Phys. Rev. D 48 (1993) 2916.
M.L. Bellac, Thermal Field Theory, Cambridge Univ. Press, 1996.
P. Morel, P. Nozires, Phys. Rev. 126 (1962) 1909.
G. Baym, C. Pethick, in: K.H. Bennemann, J.B. Ketterson (Eds.), The Physics of Liquid and
Solid Helium, Wiley, 1978, Part II.

Dynamical emergence of extra dimensions and

warped geometries
Konstadinos Sfetsos
Institut de Physique, Universit de Neuchtel, Breguet 1, CH-2000 Neuchtel, Switzerland
Abstract
We present four-dimensional gauge theories in Minkowski spacetime which effectively generate in
certain energy regimes five-dimensional warped geometries whereas, in general, the fifth dimension
is latticized. After discussing in detail several general aspects in such theories we present a number
of exactly solvable examples. We also point out how a particular case, defined in an N-sided polygon
and having a ZN symmetry, has a similar realization in an appropriate supersymmetric setting with
D3-branes. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.10.-g
1. Introduction
A quite attractive idea that has received a lot of attention is that our spacetime is higher
than four-dimensional. This idea resonates well with string theory considerations, as well
as with the more recent realization, in connection with a geometrical reformulation of
the hierarchy problem, that such extra dimensions can be even macroscopic in size since
this is allowed by the present day experiment evidence [1]. In most of the literature extra
dimensions is an assumption that one makes. Their number, size and other properties are
constrained from theoretical consistency and experimental results. In such approaches
the important point is that, certain phenomena in our four-dimensional world are just
projections of physics in higher dimensions, where the fundamental theories are defined.
From a philosophical point of view it is quite different to demand that extra dimensions
emerge dynamically and that the resulting theories are only effective descriptions of
fundamental four-dimensional theories in appropriate energy regimes. Generically, such
extra dimensions should emerge between certain energy scales, but the corresponding
theories should look four-dimensional at very high energies in the ultraviolet (UV) and
E-mail address: sfetsos@mail.cern.ch (K. Sfetsos).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 7 1 - 6
192
K. Sfetsos / Nuclear Physics B 612 (2001) 191214
very low energies in the infrared (IR). That guarantees the four-dimensional nature
of interactions at a macroscopic as well as at a microscopic level, provided the fourdimensional theories are well behaved, i.e., are renormalizable, asymptotically free etc.
In studies of gauge theories at strong coupling using the AdS/CFT correspondence,
we have constructed a model which shares the desired features we just mentioned [2].
In particular, we constructed a D3-brane solution describing part of the Coulomb branch
of N = 4 SYM theory with gauge group SU(Nk), in which vacuum expectation values
(vevs) to two of the six scalars transforming in the adjoint of the gauge group were given.
The scalar vevs were taken to lie in an N -sided polygon and in this way the original SO(6)
global invariance of the theory breaks to an SO(4) ZN . This theory turns out to look
four-dimensional for very large and very low energies and effectively five-dimensional for
a wide range of energies in between. This conclusion was also verified by examining the
potential energy for a quarkantiquark pair in [3]. This potential, as we will also see in this
paper, has the usual Coulombic behaviour 1/r for very large and very low distances, but
it behaves as 1/r 2 in between, hence confirming the effective emergence of a fifth spatial
dimension.
More recently a similar idea was proposed in [4] and in [5] directly in a conventional
field theory which need not be supersymmetric. A four-dimensional field theory was
defined in an N -sided polygon with gauge group SU(N) SU(k). Fermions provide
the necessary link and transform bilinearly under nearest neighbor pairs of gauge
transformations. If the dynamical energy scale for the SU(N) gauge theory N is much
larger than that for the SU(k) theory k , then for energies comparable to N the fermions
condense in pairs since at these scales each of the SU(N) gauge groups becomes strong,
whereas the gauge groups SU(k) can still be treated perturbatively. The authors of [4]
and [5] argued that the appropriate theoretical description is in terms of a latticized gauge
theory for SU(k) interacting with a -model that provides the link between the different
sites with nearest neighbor type of interaction. The gauge group is broken down to the
diagonal subgroup SU(k) with the familiar Higgs mechanism. The vevs v were chosen
to be the same at each site and are inversely proportional to the lattice spacing. It was
subsequently shown that the theory looks four-dimensional for distances smaller than the
lattice spacing 1/v and for distances larger than the size of the polygon N/v. However,
in between, for 1/v r N/v the theory looks five-dimensional with the fifth dimension
built up by the KaluzaKlein states of the large circle of radius N/v. The situation is the
same as in [2] in the sense that in the latter case the -model is provided by the six scalars
in the N = 4 SYM theory that transform in the adjoint of the gauge group. Also in both
cases the scalars assume vevs precisely the same way. This will be shown explicitly in
Section 4.
We note that a similar idea of a latticized array of gauge theories has been proposed in the
past in an attempt to explain the hierarchy of strengths between the electroweak and strong
interactions [6]. Also, there is a close relation to lattice gauge theories with anisotropic
couplings in five or higher dimensions, which, for certain values of the couplings, exhibit
four-dimensional layer-type phases as it was proposed in [7] and further elaborated on,
in [8,9]. We also note recent related work that has appeared, with emphasis into more
193
phenomenological applications in [1014] and some connection with non-commutative

geometry in [15,16].
The purpose of this paper is to first provide in Section 2, a systematic discussion of the
general case where the latticized array is not a polygon and moreover the vevs depend on
the site location (the case of varying gauge coupling in treated in the appendix). This will
give rise, in certain energy regimes, to a five-dimensional description in a curved warped
geometry instead of just the usual KaluzaKlein case, where the fifth-dimensional is flat.
In Section 3 we work out some examples completely in the lattice formulation, whereas for
some others we resort to the continuum approximation. We believe that these results can be
further used towards constructions that are more phenomenologically oriented. Finally, in
Section 4, we review and refine certain aspects of the model introduced in [2] in connection
with the more recent developments. We also suggest a natural mechanism for the resolution
of singularities in solutions that appear in studies of the Coulomb branch of the N = 4
SYM theory. We have also written an appendix where we treat the case of varying gauge
coupling according to the lattice position (but constant vev). We show that this is related to
the case of varying vevs and constant gauge coupling of Section 2, by a generalization of
the formulation of the usual supersymmetric quantum mechanics, but now on the discrete
lattice.
2. The general formalism

As in [4,5] we start with an effective four-dimensional low energy action which is the
sum of YM actions for SU(k) as well as for a Higgs field that provides the necessary
linking
S =
1
4(N + 1)g42

d 4x
N

j 2 N1
tr F
+
tr (D j ) D j ,
j =0
= 0, 1, 2, 3.
j =0
(1)
Each gauge term is labeled by an integer j , whereas the Higgs field j links the different
gauge terms. In transforms in the (Nj ,
Nj +1 ) representation of the gauge group, where
we note that group generators corresponding to different sites commute. Accordingly the
j
j +1
covariant derivative is defined as D j = j iA j +ij A . The gauge coupling
(N + 1)g42 is associated with each factor SU(k) separately. On the other hand the coupling
g42 is associated with the diagonal SU(k) that is left unbroken and dictates the physics at
very low energies. We assume that there is a potential for the Higgs field j that allows
this field to develop a vev vj = vfj , where we find it convenient to separate an energy
scale v from a set of numbers fj that characterize the strength of the vev at each site j .
This is our main difference from [4,5], which however will have important consequences.
A similar action to (1) has appeared before in Section 2 of [8], where in a five-dimensional
U (1) gauge theory on an anisotropic lattice, the continuum limit in the four-dimensional
layer was taken, but the fifth dimension was left latticized.
194
We digress, to note that, the case with varying gauge coupling, and constant vevs
are completely related to the case of constant gauge coupling and varying vevs that we
examine here. This will be shown in the appendix.
Turning back to (1) we see that the relevant part of the action corresponding to the gauge
field fluctuations is

N
N1

2
j 2

1
4
2
S =
(2)
x
tr
F
+
v
fj2 tr Aj+1 Aj .
d
2
4(N + 1)g4
j =0
j =0
The mass term above can be written in terms of an (N + 1) (N + 1) matrix T defined by
writing it in the form At T A . Finding the mass spectrum amounts to simply diagonalizing
the matrix T . However, this can be a cumbersome procedure except for a small number of
sites. In order to develop a systematic approach we prefer to write this mass term as
N
Aj Tj Aj ,
(3)
j =0
where the operator Tj is defined as

Tj = v 2 edj 1 fj2 edj 1 = 4v 2 sinh(dj /2)fj2 1 sinh(dj /2),
(4)
with dj being the usual differential operator satisfying [dj , j ] = 1. The rewriting (3) of
the mass term is correct up to a term which is taken care of by the boundary conditions as
discussed below.
In principle, in order to find the spectrum of the gauge field fluctuations we need to
diagonalize the operator Tj . This can be done in some cases explicitly, but of course not
in general. Before we come to specific examples it is important to mention some generic
properties and features. Consider the eigenvalue equation
T j j = M 2 j .
(5)
By construction the operator Tj is Hermitian and therefore it has real eigenvalues. In

addition, since it can be written in the form Tj = qj qj , where qj = vfj (edj 1), its
spectrum is positive semi-definite, i.e., M 2 0. The eigenvalue equation (5) should be
supplemented with some boundary condition. In order to find what consistent boundary
condition can be imposed we first realize that the equation of motion for AN
as it follows
with the mass term (3) is different than that following with a mass term as in (2), by the
2 (AN+1 AN ). Translating this into a condition for the eigenvectors we
extra term vN
j
have
boundary condition I: N+1 = N .
(6)
This boundary condition preserves the zero mode.

An alternative boundary condition which does not preserve the zero mode is
boundary condition II: 0 = 0
and (or) N = 0.
(7)
The physical motivation and consistency for such a boundary condition needs some further
explanation. Let us relax the assumption that the gauge coupling is the same at all sites
195
and introduce a defect by assuming that one of gauge couplings at the site j = N is
much smaller than the couplings at the other sites which are taken to be equal to g42 , i.e.,
2 g 2 . Hence, in this limit, after we scale AN AN g
g4,N
4,N , we see that the gauge field

4
drops
out
of
the
mass
term
in
(2).
In
terms
of
the
corresponding eigenfunction N
AN
this implies the condition N = 0 in (7). A similar comment applies if we take the gauge
coupling at the site j = 0 very small, which implies then the boundary condition 0 = 0.
Imposing the boundary condition (6) we find that there is one massless eigenvalue
corresponding to the zero model, as well as N massive ones. We emphasize that this
procedure of finding the eigenvalues is completely equivalent to simply diagonalizing the
(N + 1) (N + 1) matrix T defined before by writing the mass term in (2) in the form
At T A . We can always construct a set of N + 1 eigenvectors {j,n , n = 0, 1, . . . , N}
of the operator Tj that obey the eigenvalue equation (5) as well as the orthogonality and
completeness relations
N
N
j,n j,m
= n,m ,
j =0
j,n k,n
= j,k .
(8)
n=0
The normalized wavefunction corresponding to the zero mode is j,0 = (N + 1)1/2 with
M0 = 0.
On the other hand, imposing one of the boundary condition in (7) we find that there are
N massive eigenvalues, but the zero mode is projected out of the spectrum. Then, in the
finite sums in (8) we exclude the corresponding terms.
Let us introduce a source term corresponding to a unit charge at the origin of the
coordinate system and at the kth site. Then, the differential equation that determines the
potential energy at site j due to a charge located at site k is given by
32 Vj,k Tj Vj,k = 4(N + 1)g42 j,k (3) (x).
(9)
We may expand Vj is terms of the basis eigenvectors of the differential operator Tj as

n,k (r)j,n . Using (8) we eventually arrive at
Vj,k (r) = n V
Vj,k (r) =
N
(N + 1)g42
Mn r
j,n k,n
e
.
r
(10)
n=0
We may discuss quite generally, 1 several features for the behaviour of the potential in
1 For completeness, we note that, when the eigenvectors
j,n form a complete but not orthonormal set, we may
follow the procedure of GramSchmidt in order to construct such a set. Otherwise, (8) is replaced by
N

j =0
=A ,
j,n j,m
mn
N
j,n A1
nm k,m = j,k .
(11)
n,m=0
where Amn is some (N + 1) (N + 1) Hermitian matrix. Consequently, (10) is replaced by

Vj,k (r) =
N
(N + 1)g42
n,m=0
Mn r .
j,n A1
nm k,m e
(12)
196
(10). For instance, at distances much larger that the typical size of the array, i.e., r N/v,
obviously only the zero mode, if it exists, contributes to the sum in (10). Hence we have
that
g42
(13)
, for r N/v.
r
Therefore, at low energies we obtain the usual Coulombic behaviour with coupling
constant g42 (we ignore possible slow variations due to the running of couplings). This is the
expected behaviour for a four-dimensional gauge theory for SU(k) which is the unbroken
diagonal subgroup of the original N + 1 copies of SU(k). This behaviour is universal for
all sites due to the fact that the wavefunction for the zero mode has the same value at all
sites. For the cases that the zero mode does not exist the typical behaviour is
zero mode exists: Vj,k (r)
g42 j,1 k,1
eM1 r , for r N/v,

(14)
r
where the inverse of the lowest eigenvalue, which is expected to be comparable to the
typical size of the array, i.e., M11 N/v, sets the range of the Yukawa interaction. Its
effective overall strength depends on the sites. 2 In contrast, for distances r 1/v all
modes in the sum in (10) contribute equally. After using the completeness relation in (8)
we obtain
zero mode does not exist: Vj,k (r)
(N + 1)g42
j,k as r 1/v.
(15)
r
This is the expected behaviour for a four-dimensional theory at high energies where we
have N + 1 independent copies of gauge theories for SU(k), each with gauge coupling
(N + 1)g42 . Since, the various copies are independent the potential is non-vanishing only
for charges charged with the same SU(k) factor, i.e., j = k. For charges charged with
respect to different SU(k) factors where j = k, the potential has an expansion in powers
of r with leading term of order g42 vN for site-separation of order 1, and g42 v/N for siteseparation of order N . Hence, the magnitude of the force between charges located at the
same site is much larger than that between charges located at different sites. In the latter
cases, higher loop diagrams that contribute may be considered as, depending on the value
of the coupling constant, they may give comparable contributions (a similar comment has
been made for a specific example in [4]).
We conclude our general discussion with further remarks on the general behaviour of the
potential as a function of r. First, notice that, as it follows from (13), we have implicitly
chosen opposite charges for the two charged particles with respect to the diagonal SU(k)
subgroup that prevails at low energies. The question is whether it is possible in some
models to have a repulsive instead of attractive force for some vr 1, although of course
eventually the force has to turn attractive according to (13). Certainly, the possibility of an
repulsive force for some range of r is excluded for charges at the same site since all terms
Vj,k (r)
2 In (14) the strength of the Yukawa interaction is constant. This is due to the discreteness of the spectrum.
In cases where the continuum limit is considered this strength is r-dependent, giving rise to an overall factor for
eM1 r which is proportional to 1/r 3/2 instead of 1/r.
197
in (10) come with the same sign when j = k. However, for j = k not all terms have the
same sign and the possibility of a repulsive force is not excluded for some range of the
values for r. Nevertheless, in all of our examples in this paper the force remains attractive
for all r, in agreement with physical expectations.
2.1. Continuum limit and a fifth dimension
A quite interesting behaviour occurs for intermediate distances, which are much larger
that the lattice spacing, but also much smaller than its typical size N/v, namely for
1/v r N/v. Here we assume that N 1. Then we may approximate the lattice
by a continuum with fifth space-variable x5 = j/v. Then the differential operator Tj is
well approximated by T (x5 ) = 5 f (x5 )2 5 , where f (x5 ) is the continuum limit of the
sequence of numbers {fj } which is assumed to exist. Then, the continuous limit of the
action (2) is written as

2
1
g52 = Ng42 /v.
+ f (x5 )2 (5 A )2 ,
S = 2 d 4 x dx5 tr F
(16)
4g5

2 ), where
This can be reinterpreted as the five-dimensional action 1 2 d 5 x G tr(FMN
4g5
M = (, 5) and in the gauge A5 = 0, defined in the background with a metric given by the
line element
ds 2 = f (x5 )2 dx dx + dx52.
(17)
Hence, what we have derived is that there is an energy regime where we may safely
use a five-dimensional gauge theory defined in a curved, in general, warped spacetime,
in order to compute physical processes in a gauge theory in the usual four-dimensional
Minkowski spacetime. The four-dimensional gauge theory is renormalizable whereas the
five-dimensional one is not and only emerges as an effective description in the appropriate
energy regime.
We make the important note that the regime where 1 vr N is a strict continuum
limit where the operator Tj is well approximated by the second order differential operator
T (x5 ) = 5 f (x5 )2 5 . Higher terms that appear in the derivative expansion of Tj are
neglected. However, there is a less restrictive continuum limit that retains all such higher
derivative terms. As it will also be confirmed in the examples below, this limit is attained
for vr N and therefore is valid for distances vr 1, where the approximation that leads
to (16) fails.
For future convenience it is better to use the variable z instead of x5 , with change of
variables as dx5 = eA dz. Also we will rename f 2 by e2A . The equation determining
the potential between two charges is still given by (9) with Tj replaced by T (x5 ). The
corresponding eigenvalue equation for this operator becomes
T n = eA z eA z n (z) = Mn2 n (z),
(18)
where we have used the variable z instead of x5 . The orthogonality and completeness
relations read
198
dz eA n (z)m
(z) = n,m ,
n (z)n (z ) = eA (z z ).
(19)
n=0
As in the discrete case, we may distinguish the following two cases: if the spectrum of the
operator T has a mass gap then the behaviour of the potential between charges is, for large
distances, of the Yukawa-type, whereas if there is no such mass-gap we have a power-law
type of behaviour.
We note that, even though for notational purposes we have used the discrete set of
number {n Z} in order to label our states, the spectrum can be discrete of continuous
depending on the function A(z). In order to check that it is convenient to cast the
differential equation in (18) as a Schrdinger differential equation. Indeed, after defining
n = eA/2 n we obtain
1
1
(20)
VSch = (z A)2 + z2 A.
4
2
This potential has the form of the potentials appearing in supersymmetric quantum
mechanics [17] with corresponding superpotential W = 12 z A. Hence, the spectrum is
positive semi-definite as already expected from the fact that it originates from the spectrum
of the operator Tj which we have already argued that it is positive semi-definite.
If the Schrndinger potential in (20) can be neglected for some wide range of values
of the variable
z, then in this range the wavefunctions are approximated by plane-waves
as k 1/ 2 eikz . Then, using that k = eA/2 k = f (x5 )1/2 k we obtain that the
potential between charges at positions z and z in the fifth dimension, is given by
z2 n + VSch n = Mn2 n ,
g 2 e 2 (A(z)+A(z )
.
Vz,z (r) 5 2
r + (z z )2
1
(21)
This behaviour resembles the usual KaluzaKlein case, but with an effective coupling that
depends on the fifth dimension. This is due to the fact that the effective fifth dimension that
emerges is not flat. Such behaviour will be encountered in some of the explicit examples
below.
3. Solvable examples
The examples that we specifically work out in this section belong to two categories:
Those where we work directly in the lattice formulation and those where the continuous
approximation based on (16) is employed from the very beginning. We first consider the
case of a periodic array on a polygon that has been considered before in this context in [4,5]
and its supersymmetric variant already within the context of the AdS/CFT correspondence
in [2]. In our second example the periodicity condition is relaxed, resulting into a model
that has a discrete mass spectrum, but no zero mode. Both cases share the feature that the
vevs are the same at all sites. Our third example has vevs that vary with the site position,
but even so it turns out that is is completely solvable with the help of Laguerre polynomials.
Finally, within the continuous approximation we consider two classes of examples with
varying vevs. Both have a continuous spectrum but one of them has a mass gap.
199
3.1. Periodic array on a polygon

In this example we give the same vev at all sites, i.e., vj = v. Moreover, because of the
ZN+1 symmetry of the array, we require periodicity of the wave function as j +N+1 = j .
Since we have that fj = 1 the operator in (4) is simply Tj = 4v 2 sinh2 (dj /2). This
commutes with the differential operator dj and therefore they have the same set of
eigenfunctions which when properly normalized read
j,n =
jn
e2i N+1 ,
(22)
j = 0, 1, . . . , N, n = 0, 1, . . . , N.
N +1
The N + 1 mass eigenvalues are given by

n
, n = 0, 1, . . . , N,
Mn = 2v sin
(23)
N +1
where we also observe that a zero mode exists. Since |j,n | = 1/ N + 1 the potential (10)
is the same for all sites, as expected due to the ZN+1 symmetry of the array. Even in this
simple case, it does not seem possible to perform the sum in (10) explicitly. 3 Nevertheless,
the behaviour for r N/v and r 1/v is that predicted on general grounds by (13)
and (15), respectively.
The continuum limit holds for vr N and j k N . Then the original discrete
periodic array is replaced by a continuum periodic array, but we retain all terms in the
derivative expansion of the operator Tj . The expression for the potential turns out to be
given by the following integral
g2N
Vj,k (r) = 2 4
r
/2

d cos 2(j k) e2vr sin .
(24)
We may evaluate this in terms of generalized hypergeometric functions, but we will refrain
from presenting the result since the resulting expressions are quite complicated.
If we further restrict to distances 1 vr N we find the behaviour
Vj,k (r)
g52
1
,
2
r + (x5 x5 )2
for 1/v r N/v,
(25)
where x5 = j/v and x5 = k/v. This is the regime where the description in terms of a
five-dimensional theory in flat space (since f (x5 ) = 1) becomes appropriate. Then the
continuum periodic array is replaced by a an infinite continuum
array. In that case the
properly normalized wavefunctions are k (x5 ) = eikx5 / 2, with z taking values in the
whole real line,
the mass spectrum M 2 = k 2 is continuous. Then using that
whereas
2
|k|r
V = g5 /r dk e
/(2) we exactly reproduce (25). This is the typical Kaluza
Klein behaviour corresponding to the five-dimensional effective theory (16) and was
already found in [4,5].
3 However, we may easily prove that for r 1/v, the behaviour of the effective potential is V
j,k
j,k g42 (N + 1)/r vg42 /(N + 1)(sin2 ((j k)/(N + 1)) ( 21 /(N + 1))2 )1 + . This is in accordance
with the general behaviour that we have stated after Eq. (15).
200
There is in addition the scale regime vr 1 where the original discrete periodic array
is replaced by a infinite, but nevertheless discrete array. Then the potential is given by an
infinite sum which can be computed exactly
Vj,k (r) =
=

g42 2i(j k)n/N 2vr|n|/N
1
e
e
= g52
2
r n=
r + (j k Nn)2 /v 2
n=
g42
sinh(2rv/N)
.
r cosh(2rv/N) cos(2(j k)/N)
(26)
This expression interpolates between those in (25) for 1 vr N and (13) for vr N .
In the former case the corrections to (25) are in powers of vr/N , whereas in the latter case
the corrections to (13) are Yukawa-like, corresponding to the KaluzaKlein modes of a
reduction to a circle of radius N/v.
3.2. Relaxing the periodicity condition
As in the previous example we give the same vev to all sites, but we no longer impose
the periodicity condition. Instead, we impose the boundary condition (7) that the wave
function j vanish at the endpoints, namely that 0 = N = 0. The properly normalized
wavefunctions that form a complete set are
2
sin(j n/N), j = 0, 1, . . . , N, n = 1, 2, . . . , N 1.
j,n =
(27)
N
The mass spectrum is given by

n
Mn = 2v sin
, n = 1, 2, . . . , N 1,
2N
(28)
where the zero mode is projected out. With these eigenfunctions and eigenvalues the sum
in (10) cannot be computed exactly. Nevertheless, we may show that the behaviour for
r N/v and r 1/v is that predicted on general grounds by (13) and (15), respectively.
In the limit vr 1 and for N 1 we have that (we set L = N/v)
Vj,k (r)
2g 2
= 4
sin(j n/N) sin(kn/N)evnr/N
r
n=1
g2
sin(j/N) sin(k/N) sinh(r/L)
.
= 4
r [cosh(vr/N) cos((j + k)/N)][cosh(vr/N) cos((j k)/N)]
(29)
For r x5 x5 L, where x5 = j/v and x5 = k/v, we obtain the same behaviour as in
(25). Also
Vx5 ,x
5
4g42 Lx5 x5

,
r 4
x5 , x5 r L,
(30)
201
which is a behaviour similar to that of a seven-dimensional theory. Finally, for distances

much larger than the length L we have
2g42
(31)
sin(j/N) sin(k/N)er/L , for r L,
r
in accordance with the general formulae (13).
There is also a continuum limit for distances rv N , that goes beyond the two
derivative approximation (14)
Vj,k
g2N
Vj,k (r) = 2 4
r
/2
d sin(2j ) sin(2k)e2vr sin .
(32)
Again this is expressible in terms of generalized hypergeometric functions.

3.3. Array with varying vevs
Cases with vevs varying according to the site position are in general more difficult to
handle for increasing site number N . However, let us consider the case of vj = v j ,
j 1. 4 Though not obvious, it is possible to completely determine in general the
eigenvalues and eigenfunctions of the operator Tj which now equals

Tj = v 2 edj 1 j edj 1 .
(33)
As in some quantum mechanical problems it is convenient to pass to the momentum
representation where the operator dj is just a number denoted by ip and the site position
j becomes the differential operator ip . Then the eigenfunctions in the momentum
representation satisfy the differential equation

iv 2 eip 1 p eip 1 p = M 2 p ,
(34)
with properly normalized solution 5
2

1
i eim /2 cot(p/2)
2
= eim /2 cot(p/2) 1 + i cot(p/2) ,
p =
ip
e 1
2 i
0 p 2, m = M/v.
(35)
In order to compute the potential between charges we need the wavefunction in the position
representation j . This is given by
j =
2
dp sin(jp)p .
(36)
0
4 For convenience, we have shifted the range of j by one. Hence, j starts from j = 1 and not from j = 0 as in
the general formulation before. We take this into account in the rest of this subsection.
5 We note that the case of an array with v = vj , j 1 is also solvable by applying the same idea. Indeed, in
j

this case after we substitute p = (1 q) F (q), where q = cos2 (p/2) and = 1/4 + i/2 m2 1/4, we obtain
that F (q) satisfies a hypergeometric equation with a = b = and c = 1/2 in the standard notation. We also note
that, normalizability of p requires that the mass spectrum has a gap, Mgap = v/2.
202
After changing variables as x = cot(p/2) we have to compute the integral in (36) which
takes the form

1
(x + i)j 1 (x i)j 1 im2 x/2
j =
(37)
e
dx
(1 + ix).
2
(x i)j +1 (x + i)j +1
The integrand has poles at x = i in the complex plane and we may compute it using
standard methods in complex analysis. Since m2 > 0 we may choose the contour of
integration to close in the upper half plane and to include x = i. Then, only the first term
in the integrand in (37) contributes. The result is
j =

j 1 m2 x/2
1
e
(x + 1)j 1 x=1
j
1
(j 1)! x
=e
m2 /2
j 1

(j 1)!(m2 )n
.
(j 1 n)!(n!)2
(38)
n=0
One recognizes the last sum as an expression for the Laguerre polynomials Lj 1 (m2 ).
Therefore

2
j = em /2 Lj 1 m2 , j = 1, 2, . . . .
(39)
Having obtained this solution in a systematic way, it is important to verify directly that
(39) indeed satisfies the eigenvalue equation Tj j = M 2 j , where Tj is given by (33).
This is quite straightforward, using the recursion relations for the Laguerre polynomials,
Lj 1 (m2 ).
3.3.1. The limit of infinite array
In the case that the array is large, so that N 1, boundary effects can be neglected for
vr N . Then the mass spectrum is continuous with no mass gap (M 2 0). Using the
orthogonality and completeness properties of the Laguerre polynomials for m2 (0, )
we see that (8) are satisfied. The potential between charges at sites j and k is then
g2N
Vj,k (r) = 2 4
r
dm mem
2 mvr

Lj 1 m2 Lk1 m2 ,
for vr N.
(40)
For specific values for j and k we may give the result in terms of the error function
erf(vr/2) and powers of vr, but we will not present any results explicitly.
It is easy to see that for vr 1 we obtain (15) whereas for vr 1 and small values for
j and k we obtain an 1/r 3 fall-off for the potential as
g42 N
(41)
, for j k vr N.
v2 r 3
This is characteristic of the behaviour of a six-dimensional theory.
The limit of site position j 1 and m2 1 is also of particular interest. Then using
properties of the Laguerre polynomials we find that

j J0 2m j , as j 1, m2 1 and j m2 = finite,
(42)
Vj,k (r) 2
203
where J0 is the zeroth order Bessel function. This is the result that one would have obtained
from the five-dimensional two-derivative effective theory based on (16), as we will shortly
verify (see footnote 6).
3.3.2. Finite arrays
In the case that the array is finite with j = 1, 2, . . . , N + 1 we have to consider boundary
effects. There are two choices, one projecting out the zero mode and one preserving it.
Projecting out the zero mode We impose the boundary condition (7) that the wavefunction vanishes at the end, namely that N+1 = 0. This condition determines the allowed
values for m2n as the N zeros of the corresponding Laguerre polynomial LN

LN m2n = 0, n = 1, 2, . . . , N.
(43)
The resulting N eigenfunctions j,n = emn /2 Lj 1 (m2n ), j = 1, 2, . . . , N with
n = 1, 2, . . . , N form a complete orthogonal set, but there are no longer properly normalized to 1. Using the DarbouxChristoffel formula for the Laguerre polynomials we find
that the correctly normalized orthogonal complete set of eigenvectors is given by
2
j,n =
mn
N|LN1 (m2n )|

Lj 1 m2n ,
j = 1, 2, . . . , N, n = 1, 2, . . . , N.
(44)
Then the potential between charges located at the lattice-sites j and k, and separated by a
distance r is

g42 (N + 1)
m2n
Lj 1 m2n Lk1 m2n emn vr .
2
2
2
r
N LN1 (mn )
N
Vj,k (r) =
(45)
n=1
As an example consider the simplest

non-trivial case with N = 2. The two roots of
L2 (m2 ) = 0 are given by m2 = 2 2. Using the fact that L0 (x) = 1 and L1 (x) = 1 x
we find that the two eigenfunctions forming a complete orthonormal set according to (8),
are
1, =
m
,
2
2, =
m
.
2
(46)
The potential between charges contains of course only Yukawa-type terms. We have
Vij (r) =

g42 m vr
a e
+ aij+ em+ vr ,
r ij
where the 2 2 matrices a are given by

3 2 2 2
.
a =
2 2 2
4
i, j = 1, 2,
(47)
(48)
Keeping the zero mode Let us now impose the boundary condition (6), that is N+2 =

= 0. That determines N
N+1 , or using the properties of the Laguerre polynomials, N+2
204
nonzero mass eigenvalues m2n , n = 1, 2, . . . , N from the condition

LN+1 (m2n ) = 0 LN+1 m2n = LN m2n = LN m2n ,
n = 1, 2, . . . , N.
(49)
However, there is also the zero mode which is preserved by the boundary condition. As
before, using the DarbouxChristoffel formula for the Laguerre polynomials we find that
the correctly normalized orthogonal complete set of eigenvectors is given by

1
Lj 1 m2n , j = 1, 2, . . . , N + 1, n = 1, 2, . . . , N,
j,n =
2
N + 1 |LN (mn )|
1
j,0 =
(50)
, j = 1, 2, . . . , N + 1.
N +1
Then the potential between charges at sites j and k has a Coulombic as well as Yukawa
terms
Vj,k =
N

g42 g42
1
Lj 1 m2n Lk1 m2n emn vr .

2
2
r
r
L (mn )
n=1 N
(51)
As an example consider again the case of N = 2. Now there are three mass eigenvalues
2

2
2
given
by m0 = 0 and the two zeroes of the algebraic equation L3 (m ) = 0 which are m =
3 3. The corresponding eigenfunctions are
31
1 3
1
1
, 3, = , j,0 = .
1, = , 2, =
(52)
2 3
2 3
3
3
The potential between charges is
Vj,k (r) =

g42 g42 m vr
bj k e
+ bj+k em+ vr ,
r
r
j, k = 1, 2, 3,
where the 3 3 matrices b are given by

1
1 3
2 3
1
b = 1
2
3 3 1.
2
1 3 3 1
2
(53)
(54)
Finally, we note that when N becomes very large, the spectrum in both cases
that we have considered becomes continuous and the above expressions approach the
corresponding ones in (39) and (40). This can be seen using the property of the Laguerre
polynomials

ex/2
cos(2 nx /4) + O n3/4 ,
Ln (x) =
(nx)1/4
for n 1.
(55)
3.4. Example within the continuous approximation

We consider now cases where the lattice is already treated in the continuum approximation where a description in terms of a gauge theory in a five-dimensional warped spacetime
(16) is valid, namely when 1 vr N . The distribution of vevs might be such that the
205
mass parameter that sets the scale at these distances is not v itself, but some other mass
parameter M0 . This is assumed to be such that v/N M0 v which implies that the
corresponding length scale 1/M0 is within the allowed distances for the continuum approximation to be valid.
First, let us consider the case with f (x5 ) = 2M0 x5 , x5 0. Applying the general
formalism we easily find that the variable x5 and the variable z that enters in (18) are related
by 2M0 x5 = e2M0 z , with < z < and that the function A(z) = 2M0 z. Therefore the
Schrdinger potential in (20) is just a constant, i.e., VSch = M02 . That defines a mass gap
since in order to have normalizable wavefunctions we must require that M 2 M02 . We
obtain
M (z) =
1/4 i(M 2 M 2 )1/2 z

eM0 z 2
0
e
.
M M02
2
(56)
Hence, the potential between charges at positions z and z in the fifth dimension is

g52 M0 eM0 (z+z )
K1 (M0 R), R = r 2 + (z z )2 ,
(57)
R
where K1 is the modified Bessel function of order one. If we consider distances much
smaller than the characteristic scale of the mass gap 1/M0 , then we obtain indeed the
behaviour (21). Instead, in the opposite limit of distances much larger than 1/M0 we find
that
Vz,z =
g2
M02

Vz,z (r) 5 eM0 (z+z )
eM0 R ,
(M0 R)3/2
2
for 1/M0 R N/v.
(58)
Naturally, the range of Yukawa interaction is set by the mass gap, whereas the strength has
the 1/r 3/2 behaviour as noted in footnote 2.
Also, the value of the mass gap M0 is in agreement with what was noted in footnote 5
for the exact lattice treatment of the vev distribution vj = vj , after we set M0 = v/2.
3.5. Another example within the continuous approximation
A generalization of the previous case which however does not lead to a mass gap for the
2+1
spectrum, is to take f (x5 ) = (2( + 1)M0 x5 ) 2(+1) with x5 0 and where the exponent of
x5 is introduced as such for later notational convenience. We find that the variables x5 and
z are related as 2( + 1)M0 x5 = (M0 z)2(+1) and also that eA = (M0 z)2+1 . Hence, the
Scrhdinger potential takes the form
VSch =
2 1/4
,
z2
(59)
from which we conclude that the mass spectrum is continuous without a mass gap. The
eigenfunctions are given in terms of Bessel functions as
M (z) =
1
(M0 z) J (Mz).
2M0
(60)
206
It turns out that 6 the potential between charges at the same position z in the fifth dimension
is computed in terms of a hypergeometric function as (for simplicity we do not deal with
the case of charges at different positions in the fifth dimension)

g52 /M02+1
28( + 3/2)
4z2
Vz,z (r) =
,
F + 1/2, 1/2, 2 + 1, 2
8( + 1) r 2 (r 2 + 4z2 )+1/2
r + 4z2
for 1/v r N/v.
(61)
We also note the limiting case
28( + 3/2) g52 /M02+1
Vz,z (r) =
,
8( + 1) r 2+3
for 1/v z r
(62)
and that for r z we confirm the behaviour (21).
4. Strong coupling considerations

We would like to study similar effects at strong coupling using the AdS/CFT
correspondence [18]. We will see that our findings and the emergence of an effective fifth
dimension are in precise agreement with the gauge theory considerations so far.
The clearest discussion for our purposes can be done for the Coulomb branch of
the N = 4 SYM theory with gauge group SU(Nk). This theory has six scalars XI ,
I = 1, 2, . . . , 6 in the adjoint of the gauge group. We will reanalyze and refine the model
introduced in [2] and we will study the potential between a quark and an antiquark.
We give vevs to two of the six scalars in the theory in a way that the original SO(6)
global invariance of the theory that rotates the six scalars, breaks to SO(4) ZN . We
may represent these scalars as Nk Nk traceless matrices. In general, since the scalar

potential in the SYM action is proportional to I <J tr[XI , XJ ]2 we can choose the vevs
along the diagonal so that the potential is zero. That preserves sixteen supercharges and
breaks the conformal invariance. We arrange the Nk Nk matrices XI s in the form of
block-diagonal matrices with N diagonal k k blocks, where in each block the vevs

are the same. The N = 4 SYM theory action contains the term 6I =1 tr(D I )2 which
couples the gauge fields and the six scalars. It is convenient to choose the standard real
1
ij INkNk , where the matrix elements
basis for the SU(Nk) generators Jij = eij Nk
of the matrices eij are (eij )mn = im j n . Hence, we give vevs to the scalars represented
i , where h = J are the diagonal Cartan
vev
as = hi X
by the six-dimensional vector ,
i
ii

generators. The masses of the gauge fields arise from the term Lmass 6I =1 tr[ I , A ]2 .
ij
After we define A = A Jij we have
Lmass k
N1
i
ij ,j i
j
i
i
vev X
vev
vev
vev
X
A A
X
X
.
(63)
i,j =0
6 Note that, for M = v/2 and = 0 we have that v = v j . Then indeed the wavefunction (60) agrees with
j
0
(42).
207
This basic formula can now be applied to specific cases. As in [2] we choose for the
vevs in the j th block the values

j
vev
X
= 0, 0, 0, 0, r0 cos(2j/N), r0 sin(2j/N) ,
j = 0, 1, . . . , N 1.
(64)
These form an N -sided polygon enclosed by a circle in the x5 x6 plane of radius r0 . Then
we find that (63) becomes
Lmass r02 k
N1

,j i
sin2 (j i)/N Aij
.
A
(65)
i,j =0
From this we read off the masses as
Mn r0 k sin(n/N), n = 0, 1, . . . , N 1,
(66)
which are the same as the ones we found in (23). They have a degeneracy dn which

2
is: d0 = N 1 and dn = 2(N n) for n 1 (note that N1
n=0 dn = N 1). Using
the fact that Mn = MNn it is easily seen that the expression for the potential is given
g2
Mn r 1 ), which for large enough N is the same as the potential
by V r4 ( N1
n=0 e
N
corresponding to the mass formula in (23). That makes apparent that the model in [2] and
the construction of [4,5] (reviewed in Section 3.1) describe essentially the same physics.
The advantage of having a D3-brane realization in a model with an emerging latticed
fifth dimension, is that we may consider the theory at strong coupling using the AdS/CFT
correspondence. In the type-IIB theory we consider a large number Nk of D3-branes that
are distributed in the transverse to the branes six-dimensional space according to (64).
Namely, the gravitational metric that we will consider is

ds 2 = H 1/2 dy dy + H 1/2 dx12 + + dx62 ,
(67)
where y , = 0, 1, 2, 3 are the parallel to the brane directions. The harmonic function H
in the six-dimensional space spanned by xi , i = 1, 2, . . . , 6 is given by
H=
N1

j =0
R44
,
j 4
vev
|
xX
|
(68)
where R44 = 4g42 k is the t Hooft coupling associated with the SU(k) gauge theory. We see
that the vevs in the field theory (64) become the centers in the harmonic function above.
We emphasize that the supergravity approximation is valid even close to each one of the
centers if we choose that R44 1 and also small values for g42 which implies that there are
many branes located at each center, i.e., k 1. The number of such centers N need not be
large.
The harmonic function H has been computed in [2]. It takes the form
H = R4
U 2 + r02
((U 2 + r02 )2 4r02 u2 )3/2
N ,
(69)
208
where R 4 = 4g42 kN . Also U 2 = x12 + + x62 , u2 = x52 + x62 and

((U 2 + r02 )2 4r02 u2 )1/2 cosh Nx cos N 1
sinh Nx
+N
,
cosh Nx cos N
(cosh Nx cos N)2
U 2 + r02
(70)
where is the polar angle in the x5 x6 plane and

2
2 + r2
U + r02 2
U
0
x
+
1.
e =
(71)
2r0 u
2r0 u
N =
Notice that H explicitly processes the symmetry SO(4) in the four-dimensional subspace
spanned by x1 , x2 , x3 and x4 as well as the ZN symmetry under the shift + 2/N .
The potential between a pair of heavy quark and antiquark separated by a distance L is
found 7 by minimizing the NambuGoto action of a fundamental string in the supergravity
background (67). We will not give details as the necessary techniques we first developed
for the conformal case in [19]. For backgrounds of the type that we have here we will
use Eqs. (5)(7) in [3]. The trajectory that minimizes the action lies on the x5 x6 plane
where u = U . However, not every angle is allowed, but only those with sin(N) = 0.
All angles satisfying this condition are equivalent because of the ZN symmetry. For
concreteness we may choose the trajectory along the x5 -axis (for = 0 and u = U ). Lets
define the function f = R 4 /H . Then for the trajectory that we are interested in we have
f (U ) =
=
(U 2 r02 )3 1
,
U 2 + r02
(U/r0 )N + 1
2N
(U/r0 )2 1
+
,
N
N
N
(U/r0 ) 1 (U/r0 ) + (r0 /U ) 2 (U/r0 )2 + 1
(72)
where we have used the fact that ex = U/r0 , when u = U and = 0. Then the integrals
for the length L and the energy V are given by

f (U0 )
2
,
L = 2R
(73)
dU
f (U )(f (U ) f (U0 ))
U0
and
1
V=

dU
U0

1
f (U )
1 (U r0 ),
f (U ) f (U0 )
(74)
where U0 denotes the lowest value of U that can be reached by the string geodesic. In
principle, we have to evaluate the integrals and try to express V as a function of L. This
is impossible in general though it is easy to deduce that V is a monotonously increasing
functions of L from to 0 in the interval L (0, ).
7 We denote the distance between the charges by L in order to confirm with standard notation in the AdS/CFT
literature. This is nothing but the distance denoted by r in the previous part of the paper.
209
4.1. Examining various energy regimes

As usual, in the AdS/CFT correspondence large (small) distances in the five-dimensional
geometry, corresponds to small (large) distances in the gauge theory side [20]. The variable
U represents a distance in the AdS side and it is interpreted as an energy variable in the
gauge theory side. It is instructive to find the various limits of the metric (67) and of the
quarkantiquark potential, by examining the behaviour of the harmonic function H as we
go from the UV to the IR. The UV corresponds to energies large enough so that the vevs
can be neglected. The deep IR corresponds to energies very close to the vevs.
4.1.1. Behaviour for U r0
In the UV for U r0 since the harmonic takes the form
H
R4
,
U4
for U r0 ,
(75)
the metric becomes that for AdS5 S 5 with each factor having radius R in string units,
Then the original SO(6) global symmetry of the metric is restored. That corresponds to
the fact that in the deep UV the values of the vevs can be neglected and we should be
describing an SU(Nk) SYM theory. In this regime the integrals can be evaluated and give
for the potential energy between a quark and antiquark the results of [19]. Hence we have

R42 N
8(3/4) 2 R42 N
, L
.
V 2
(76)
8(1/4)
L
r0
This behaviour is similar to the one found from pure gauge theory considerations in (15)
for the interaction of charges at the same site j = k.
4.1.2. Behaviour for U r0 r0 /N
For distances from the location of the centers much larger than the distance between the
different centers we may approximate the harmonic function by [2]
H = R4
U 2 + r02
((U 2 + r02 )2 4r02 u2 )3/2
for U r0 r0 /N,
(77)
since then N 1. Then we see that the discrete part ZN of the symmetry group SO(4)
ZN becomes a continuous U (1). Then the resulting integrals for the length and the energy
are quite complicated (see Eqs. (A.6) and (A.7) of [3]) and it does not seem possible to
compute them explicitly, let alone to explicitly obtain the energy as a function of the
separation length. This approximation is valid for L R42 N/r0 . The situation here is
similar to the one described from a gauge theory point of view by Eq. (24) (for j = k).
As a separate remark we note that the metric (67) with the above harmonic (77) has a
ring singularity defined by the equations U 2 = u2 = x52 + x62 = r02 . It is obvious that this is
an artifact of the continuous approximation which is not valid near the singularity, where
the harmonic which should be used is given by (80) and the metric then is completely
non-singular. The same resolution of singularities applies to solutions that have been used
for studies of the Coulomb branch of the N = 4 SYM gauge theory in, for instance,
210
[3,21,22]. These solutions, take the form (67), but unlike (69), the distribution of D3-branes
is continuous instead of discrete. It is obvious that the solutions will become non-singular if
we replace the continuous by a discrete distribution of D3-branes with the given continuous
limit. An example of that was already given in the appendix of [2] representing the discrete
version of a continuous distribution of D3-branes on a disc.
As another side remark we note that the solution with harmonic (69) as well as its
continuous approximation (77), do not have a five-dimensional gauged supergravity origin.
4.1.3. Behaviour for r0 /N U r0 r0
For distances from the centers much larger than the distance between the different
centers, but also much smaller than the radius of the circle, we may approximated the
harmonic function by
H
R4
,
4r0 U 3
for r0 /N U r0 r0 .
(78)
In this case we have a smeared D3-brane solution along one transverse direction which
is T-dual to a D4-brane solution. Also the global symmetry has been enhanced from an
SO(4) U (1) into an SO(5). Since a D4-brane solution corresponds to a five-dimensional
theory we expect that there will be a 1/L2 behaviour for the quarkantiquark potential.
Indeed we obtain the result [3]

8(2/3) 3 R44 N
R42 N
R42 N
,
for

L

.
V 4
(79)
8(1/6) r0 L2
r0
r0
The behaviour is similar to (25) from the usual gauge theory point of view.
4.1.4. Behaviour for U r0 r0 /N
In the deep IR for energy scales very close to the vevs
H
R42
,
U4
for U r0 r0 /N,
(80)
where R44 = 4g4 k is the t Hooft coupling for an SU(k) gauge theory. Then we have

V 2
8(3/4)
8(1/4)
R42
,
L
for L
R42 N
.
r0
(81)
This is a Coulombic behaviour for a gauge group SU(k) and is similar to that in (13) from
the usual gauge theory point of view.
4.1.5. Discrete infinite array
Finally, we may also find a regime where the original circular discrete array is
approximated by a discrete infinite array on a straight line. This is possible for distances
from the centers much smaller that the radius of the circle. From the gauge theory point
of view at weak coupling the analogous potential is described by (26). The solution has a
211
metric given by (67) with harmonic
R44
(U 2 + (x6 2vj )2 )2
j =

R44
cosh(U/v) cos(x6 /v) 1
v sinh(U/v)
+U
=
,
4U 3 v 2 cosh(U/v) cos(x6 /v)
(cosh(U/v) cos(x6 /v))2
H=
(82)
distances for the

where U 2 = x12 + x22 + + x52 and v = r0 /N . Translated into separation
quarkantiquark pair, this approximation is valid when L R42 N /r0 . The trajectory that
we will use to compute the potential amounts to setting x6 = 0. Then we utilize the general
formulae (73) and (74) with R replaced by R4 and the function f being given by
1

U/(2v)
f (U ) = 4vU 3 coth(U/(2v)) +
(83)
.
sinh2 (U/(2v))
As we move from U v to U v this case interpolates between those described before
by (79) and (81). It is not possible to compute the energy as a function of the separation
explicitly, but we may use the above expression in order to find the first correction to the
1/L2 behaviour in (79). Expanding (83) for U v and keeping the two leading terms we
find that

4
8(2/3) 3 R44
R46 c2 2R4 2
v
L
V = 4
(84)
1
+
c
e
+
,
1 3 3
8(1/6) vL2
v L
R4
where the c1 and c2 are two constants of oder 1. 8 Since 1 v 2 L4 2 N the corrections
to the leading 1/L2 behaviour are small. However, they do not obey a power-like law as
the corrections to (25) (noted after (10)). Similarly, the corrections to (81) are found by
expanding (83) for U v. It is easy to see that these corrections are organized in powers
R2 N
of r04L 1, in contrast to the corrections to (13) which are of the Yukawa type (noted
again below (10)). Hence, it seems that the mechanisms for developing an effective fifth
dimension are different at weak and strong coupling.
In will be interesting to further extend the results of this paper in order to incorporate
fermions and to investigate the renormalization of the gauge coupling constant. Also an
extension to non-commutative gauge theories could be of some interest.
Acknowledgements
I would like to thank A. Kehagias for a discussion. This research was supported by the
European Union under contracts TMR-ERBFMRX-CT96-0045 and -0090, by the Swiss
Office for Education and Science, by the Swiss National Foundation and by the contract
HPRN-CT-2000-00122.
8 Precisely, c =
1
8 1)(8(2/3)/ 8(1/6))2 0.083 and c = 4(8(2/3)/ 8(1/6))2 0.744.

8 3 (
2
3 6
212
Appendix A. Varying gauge coupling

In this appendix we first study differential operators defined on the lattice of the type
Tj = qj qj ,
j = 0, 1, . . . , N,
(A.1)
for some operator qj and which have energy eigenvalues that are positive semi-definite,
i.e., M 2 0. In the case of the operator (4) discussed in the main text we have that
qj = vfj (edj 1), as already noted. If a state j,0 satisfies the equation qj j,0 = 0, then
automatically we have a solution of Tj j,0 = 0 corresponding to a zero mass eigenvalue.
Define now the lattice differential operator Tj obtained by reversing the order of qj and qj
in (A.1)
Tj = qj qj ,
j = 0, 1, . . . , N.
(A.2)
The discussion next is completely parallel to that in supersymmetric quantum mechanics

n2 of Tj and T
j
(SQM) [17]. The spectra and eigenfunctions of Mn2 and M
2
n2 = Mn+1
,
M
j,n =
M02 = 0,
qj j,n+1 ,
Mn+1
1
j,n+1 =
q .
n j j,n
M
(A.3)
Hence, the spectra of the operators Tj and Tj are identical except for the zero mode. As
we have seen, boundary conditions may project it out of the spectrum in which case the
spectra are identical. Moreover, acting with qj (qj ) we may convert an eigenvector of Tj
j ) into an eigenvector of Tj (Tj ).
(T
Let us define the matrices

Tj 0
0 0
0 qj
,
Q
,
Q
.
=
=
Hj =
(A.4)
j
j
0 0
qj 0
0 Tj
It is trivial that these obey the closed superalgbebra

[Hj , Qj ] = Hj , Qj = {Qj , Qj } = Qj , Qj = 0,

Qj , Qj = Hj .
(A.5)
In terms of this superalgebra, the fact that Qj and Qj commute with Hj is responsible for
the degeneracy of the spectra.
As an application of the above general results, let us consider again the low energy
2 and Higgs
effective action in the lattice (1), but now with varying gauge coupling g4,j
values vj . The relevant term derived from (1) corresponding to the gauge field fluctuations
is
1
S =
4(N + 1)

d 4x
N

2
j 2 N1

tr F
+
vj2 tr g4,j +1 Aj+1 g4,j Aj .
j =0
j =0
(A.6)
213
Consider first the case of constant vevs vj = v. The mass term above can be written as
N
Aj Tj Aj ,
(A.7)
j =0
where the operator Tj is defined as

Tj = v 2 g4,j edj 1 edj 1 g4,j = 4v 2 g4,j sinh2 (dj /2)g4,j .
(A.8)
j has precisely the form (A.2) with qj = g4,j (edj 1). Using the
Hence, the operator T
formalism that we developed we may map this problem into a problem for the superpartner
operator Tj in (A.1). Comparing with the case of constant gauge coupling and varying
vevs considered in Section 2, we find that they are identical after we identify v 2 g 2 1 by
v2 f 2
4,j + 2
1.
j 2
As a specific application consider the case of couplings g4,j = g4 j , j 1. After using

the results of Section 3.3 we find that the set of eigenfunctions of the operator Tj are

j Lj m2 ,
m=
M
.
g4 v
Using (A.3) we find that the eigenfunctions of the operator (A.8) are

j j Lj 1 m2 Lj m2 .
(A.9)
(A.10)
We may easily verify, using recursion relations for the Laguerre polynomials, that this
solution satisfies the eigenvalue equation for the operator Tj in (A.8). We have left
undetermined the overall normalization constant, since this would depend on the boundary
conditions that we will impose. Note that a choice of a boundary condition for either j or
j implies the boundary condition for the other, if we want that the mass spectra are also
related according to (A.3). Of course we may still impose independent boundary conditions
in which case the two spectra are not related.
Finally, we note that in the most general case of varying vevs and gauge couplings, the
operator Tj still takes the form (A.1) with qj = vj (edj 1)g4,j .
References
[1] I. Antoniadis, S. Dimopoulos, G. Dvali, Nucl. Phys. B 516 (1998) 70, hep-ph/9710204;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
For related experiments see, for instance, J.C. Long, H.W. Chan, J.C. Price, Nucl. Phys. B 539
(1999) 23, hep-ph/9805217.
[2] K. Sfetsos, JHEP 01 (1999) 015, hep-th/9811167.
[3] A. Brandhuber, K. Sfetsos, Adv. Theor. Math. Phys. 3 (1999) 851, hep-th/9906201.
[4] N. Arkani-Hamed, A.G. Cohen, H. Georgi, (De)constructing dimensions, Phys. Rev. Lett. 86
(2001) 4757, hep-th/0104005.
[5] C.T. Hill, S. Pokorski, J. Wang, Gauge invariant effective Lagrangian for KaluzaKlein modes,
hep-th/0104035.
[6] M.B. Halpern, W. Siegel, Phys. Rev. D 11 (1975) 2967.
214
[7] Y.K. Fu, H.B. Nielsen, Nucl. Phys. B 236 (1984) 167.
[8] Y.K. Fu, H.B. Nielsen, Nucl. Phys. B 254 (1985) 127.
[9] A. Hulsebos, C.P. Korthals-Altes, S. Nicolis, Nucl. Phys. B 450 (1995) 437, hep-th/9406003;
P. Dimopoulos, K. Farakos, C.P. Korthals-Altes, G. Koutsoumbas, S. Nicolis, JHEP 0102 (2001)
005, hep-lat/0012028.
[10] H. Cheng, C.T. Hill, S. Pokorski, J. Wang, The standard model in the latticized bulk, hep-th/
0104179.
[11] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Electroweak symmetry breaking from dimensional
deconstruction, hep-ph/0105239.
[12] H. Cheng, C.T. Hill, J. Wang, Dynamical electroweak breaking and latticized extra dimensions,
hep-ph/0105323.
[13] C. Csaki, J. Erlich, C. Grojean, G. Kribs, 4D constructions of supersymmetric extra dimensions
and gaugino mediation, hep-ph/0106044.
[14] H.D. Kim, To be (finite) or not to be, that is the question KaluzaKlein contribution to the Higgs
mass, hep-ph/0106072.
[15] M. Alishahiha, (De)constructing dimensions and non-commutative geometry, hep-th/0105153.
[16] J. Dai, X. Song, Spontaneous symmetry broken condition in (de)constructing dimensions from
noncommutative geometry, hep-ph/0105280.
[17] E. Witten, Nucl. Phys. B 188 (1981) 513;
For a review see, F. Cooper, A. Khare, U. Sukhatme, Phys. Rep. 251 (1995) 26, hep-th/9405029.
[18] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[19] J. Maldacena, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002;
S. Rey, J. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de Sitter
supergravity, hep-th/9803001.
[20] L. Susskind, E. Witten, The holographic bound in anti-de Sitter space, hep-th/9805114.
[21] I. Bakas, K. Sfetsos, Nucl. Phys. B 573 (2000) 768, hep-th/9909041.
[22] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, JHEP 0007 (2000) 038, hep-th/9906194.

Constructing doubly self-dual chiral p-form actions

in D = 2(p + 1) spacetime dimensions
Yan-Gang Miao a,b , Harald J.W. Mller-Kirsten a , Dae Kil Park c,d
a Department of Physics, University of Kaiserslautern, P.O. Box 3049, D-67653 Kaiserslautern, Germany
b Department of Physics, Xiamen University, Xiamen 361005, Peoples Republic of China
c Department of Physics, Kyungnam University, Masan 631-701, South Korea
d Michigan Center for Theoretical Physics, Randall Laboratory, Department of Physics, University of Michigan,
Ann Arbor, MI 48109-1120, USA

Abstract
A Siegel-type chiral p-form action is proposed in D = 2(p + 1) spacetime dimensions. The
approach we adopt is to realize the symmetric second-rank Lagrange-multiplier field, introduced
in Siegels action, in terms of a normalized multiplication of two (q + 1)-form fields with q indices
of each field contracted in the even p case, or of two pairs of (q + 1)-form fields with q indices
of each pair of fields contracted in the odd p case, where the (q + 1)-form fields are of external
derivatives of one auxiliary q-form field for the former, or of a pair of auxiliary q-form fields for
the latter. Using this action, it is straightforward to deduce the recently constructed PST action for
q equal to zero. It is found that the Siegel-type chiral p-form action with a fixed p (even or odd) is
doubly self-dual in D = 2(p + 1) spacetime dimensions when the auxiliary field(s) is/are also chosen
to be of p-form. This result includes PSTs as a special case where only the chiral 0-form action is
doubly self-dual in D = 2 dimensions. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10.-z; 11.15.-q; 11.30.-j
1. Introduction
Chiral p-forms exist in the D = 2(p + 1) dimensional spacetime since their external
derivatives, i.e., the field strengths, are (p + 1)-forms and the Hodge duals as well, which
requires the spacetime dimensions to be twice big the form number of the field strengths.
The chirality usually means that the field strengths satisfy a self-duality condition. In
spacetime with Lorentzian metric signature, self-duality requires the chiral p-forms to be
complex if p is odd. In this case one may equivalently introduce [1] a pair of real p-forms,
E-mail addresses: miao@physik.uni-kl.de (Y.-G. Miao), mueller1@physik.uni-kl.de
(H.J.W. Mller-Kirsten), dkpark@hep.kyungnam.ac.kr (D.K. Park).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 6 2 - 5
216
Y.-G. Miao et al. / Nuclear Physics B 612 (2001) 215225
Aa(p)(x), a = 1, 2, and impose upon the corresponding field strengths, F a(p+1)(A) =

dAa(p)(x), the duality condition, 1
F a(p+1)(A) ab F b(p+1) (A) F a(p+1)(A) = 0,
F a(p+1)(A) = ab F b(p+1) (A),
where 12 = 21 = 1, a, b = 1, 2, and F a(p+1)(A) stand for the Hodge duals of
F a(p+1)(A). However, the self-duality requires the chiral p-forms to be real if p is even.
For this case one p-form potential, A(p) (x), is enough, 2 and its field strength, F (p+1) (A) =
dA(p)(x), of course, satisfies the usual self-duality condition,
F (p+1) (A) F (p+1) (A) F (p+1) (A) = 0,
F (p+1) (A) = F (p+1)(A).
For the sake of conciseness, we combine these two (self-)duality conditions into one
general form 3 for both even p and odd p cases,
F a(p+1)(A) ab F b(p+1)(A) F a(p+1)(A) = 0,
F a(p+1)(A) = (1)p+1 ab F b(p+1)(A).
(1)
For the main reason why chiral p-forms have received so much attention, we may
summarize that they appear in various theoretical models that relate to superstring theories,
and reflect especially the existence of a variety of important dualities that connect these
theories among one another.
One has to envisage the two basic problems in a Lagrangian description of chiral
p-forms: one is the consistent quantization and the other is the harmonic combination of
manifest duality and spacetime covariance, since the equation of motion of a chiral p-form,
i.e., the self-duality condition, is first order with respect to the derivatives of space and
time. In order to solve these problems, some non-manifestly spacetime covariant [1,38]
and manifestly spacetime covariant [912] models have been proposed. It is remarkable
that these chiral p-form models have close relationships among one another, especially
various dualities that have been demonstrated in detail from the points of view of both
configuration [1215] and momentum [16,17] spaces.
The recently constructed PST action [12], motivated by the SS action [8] together
with the idea [18] of spacetime covariance in it, has become a successful model in
putting both the manifest spacetime and duality symmetries into a chiral p-form action.
The characteristic of this model is to introduce a Lagrangian-multiplier term in a nonpolynomial way, or more precisely, in terms of a normalized multiplication of two vector
(1-form) fields that are external derivatives of one auxiliary scalar (0-form) field. Though
there is the merit of connecting the Zwanziger [1] with SS actions or with others through
various dualities, it is quite unsatisfactory that the model lacks of [13] double self-duality
1 If Aa(p) (x) are treated as an O(2) doublet and an extended dualization is defined, the extended duals of
F a(p+1) (A), if we choose ab F b(p+1) (A), satisfy a self-duality condition. See, for example, Ref. [2] for
details.
2 We do not introduce two p-form potentials that constitute a doublet adopted in Ref. [2] because that treatment
is not economical for the even p case.
3 For the definition of ab , see Eq. (10).
217
in higher (than two) dimensions. The double self-duality means that the action of chiral
p-forms is self-dual with respect not only to dualization of chiral fields, but also to
dualization of auxiliary fields. We may ask whether the double self-duality does or does
not appear in higher (than two) dimensions.
Analysing in detail Siegels action 4 [9] and its generalization [17] in D = 2(p + 1)
(p even) dimensions, we immediately find the relation between the Siegel and the PST
actions. That is, the former becomes the latter if the symmetric second-rank Lagrangianmultiplier is realized in terms of the approach mentioned above. In fact, this discovery
relates closely to our question. As we have proven, if the Lagrangian-multiplier is realized
in terms of higher than zero form auxiliary fields, the double self-duality can occur in
higher (than two) dimensions. In this way, we generalize PSTs result deduced by the
assumption of introducing one and the same auxiliary scalar field in any even dimensions.
It is the aim of this paper to construct a doubly self-dual chiral p-form action in D =
2(p + 1) spacetime dimensions. The paper is arranged as follows. In the next section a
Siegel-type chiral p-form action is proposed in D = 2(p + 1) dimensions. The method we
utilize is to realize the symmetric second-rank Lagrange-multiplier field of Siegels action
in terms of a normalized multiplication of two (q + 1)-form fields with q indices of each
field contracted for the even p case, 5 or of two pairs of (q + 1)-form fields with q indices
of each pair of fields contracted for the odd p case, where the (q + 1)-form fields are of
external derivatives of one auxiliary q-form field for the former, or of one pair of auxiliary
q-form fields for the latter. In Section 3 we investigate duality properties of this action
with respect to both the dualization of chiral p-form fields and the dualization of auxiliary
q-form fields, and make search for the condition when the double self-duality appears.
Finally Section 4 is devoted to a conclusion.
The metric notation we use throughout this paper is
g00 = g11 = = gD1,D1 = 1,
012D1 = 1.
(2)
Greek letters stand for spacetime indices (, , , , . . . = 0, 1, 2, . . . , D 1).

2. Siegel-type chiral p-form action in D = 2(p + 1) dimensions
We begin with Siegels action in D = 2(p + 1) (p even) dimensions [17],

1
Ss = d D x
F (A)F 1 p+1 (A)
2(p + 1)! 1 p+1

1
+ F1 p (A)F 1 p (A) ,
2
(3)
4 There is a recent paper [19] where a so-called generalized gauged Siegel model is developed. This model and
its related discussions on interference phenomena and Lorentz invariance are limited only in D = 2 dimensions.
However, our attention here focuses on beyond D = 2 dimensions.
5 Zero is included in the even p case throughout this paper.
218
where A1 p (x) is a real p-form field, F1 p+1 (A) its field strength, and F1 p+1 (A),
which is called the self-dual tensor in Refs. [12,13], is the difference of the field strength
and its Hodge dual. Note that the Lagrange-multiplier, (x), is in general a symmetric
second-rank tensor field. We may introduce a real auxiliary q-form field, Y1 q (x), and
realize (x) in some sense in terms of a normalized multiplication of two (q + 1)-form
fields with q indices of each field contracted, where the (q + 1)-form field, T1 q+1 (x),
is defined as the external derivative of the q-form field, i.e.,
=
T 1 q T1 q
T2
(4)
where
T1 q+1 [1 Y2 q+1 ] ,
T 2 T1 q+1 T 1 q+1 .
(5)
Here we mention that q takes numbers 0, 1, . . . , D 2 = 2p for a fixed p (even or odd)

because the dual of a q-form is a (D q 2)-form, i.e., a (2p q)-form in D = 2(p + 1)
spacetime dimensions. For details, we refer to the next section. Substituting Eq. (4) into
Eq. (3), one obtains a Siegel-type action for chiral p-forms in D = 2(p + 1) (p even)
dimensions,

1
D
F (A)F 1 p+1 (A)
Sst = d x
2(p + 1)! 1 p+1

T 1 q T1 q
1 p
+
(6)
F1 p (A)F
(A) .
2T 2
When p is odd, we may introduce a pair of real p-form fields, Aa1 p (x), a = 1, 2, as
explained in Section 1. Correspondingly, we also introduce a pair of real auxiliary q-form
fields, Yb1 q (x), b = 1, 2. Therefore, in accordance with Eq. (6), we conclude that the
Siegel-type action that combines both even p and odd p cases takes the form,

1
Fa
(A)F a1 p+1 (A)
SST = d D x
2(p + 1)! 1 p+1

b
T b1 q T
1 q
a
a1 p
+
(7)
F
(A)F
(A)
,
1 p
2T 2
where
Fa1 p+1 (A) [1 Aa2 p+1 ] ,
Fa 1 p+1 (A) ab Fb1 p+1 (A)
1
F a1 p+1 (A),
(p + 1)! 1 p+1 1 p+1
(8)
Ta1 q+1 [1 Ya2 q+1 ] ,

T 2 Ta1 q+1 T a1 q+1 ,
(9)
and

=
ab
219
1, for even p,
2, for odd p,
for even p,
ab = 1, a = b = 1,
ab , a, b = 1, 2, and 12 = 21 = 1, for odd p.
(10)
In addition, one obtains from Eq. (8) that for the both cases the field strength differences,
or the so-called self-dual tensors, satisfy the relations
Fa 1 p+1 (A) =
(1)p+1 ab
1 p+1 1 p+1 F b1 p+1 (A),
(p + 1)!
(11)
which will be useful in calculations in the next section. These relations are the
generalization of the special forms discussed in D = 2, 4, 6 dimensions [12,13].
We point out that in the even p case the Siegel-type action, Eq. (7) or Eq. (6), contains
the PST action in D = 2 and D = 6 dimensions when q equals zero, and in the odd p
case Eq. (7) includes the PST action in D = 4 dimensions if only one auxiliary 0-form
field is introduced. Consequently, the PST action can be treated as a special example of
the Siegel-type action. Furthermore, due to the fact that the PST action originates from
the SS action [8] which is dual to the Zwanziger action [1], our discovery establishes the
relationship between the Siegel-type and the SS or Zwanziger actions through the PST
action as a bridge.
3. Duality properties of the chiral p-form action in D = 2(p + 1) dimensions
In the following we investigate duality properties of the Siegel-type chiral p-form action,
SST , with respect to both the dualization of the chiral p-form fields, Aa1 p (x), and the
dualization of the auxiliary q-form fields, Yb1 q (x), and make search for the condition
under which double self-duality occurs. For the dualization of the chiral fields we shall
follow the usual line of Refs. [1315], but for the dualization of the auxiliary fields we
shall proceed in our own way with the intention to avoid extensive calculations and to
arrive directly at the desired result.
3.1. Dualization of the chiral p-form fields
By introducing two (p even) or two pairs of (p odd) independent (p + 1)-form fields,
Fa1 p+1 and Gb1 p+1 , we construct a new action to replace SST ,

1
D
SST = d x
Fa
F a1 p+1
2(p + 1)! 1 p+1
b
T b1 q T
1 q
a
F
F a1 p
1 p
2T 2

1
Ga1 p+1 Fa1 p+1 [1 Aa2 p+1 ] ,
+
(p + 1)!
(12)
220
where F a1 p+1 has the same definition as that of F a1 p+1 (A) in Eq. (8), but here it is
with respect to Ga1 p+1

not dealt with as a functional of Aa1 p (x). Variation of SST
gives
Fa1 p+1 = [1 Aa2 p+1 ] ,
(13)
, yields the classical equivalence between the two actions,

which, when substituted into SST
with respect to Fa1 p+1 leads to the

SST and SST . Furthermore, variation of SST
a
expression of G 1 p+1 in terms of F 1 p+1 and other fields,

1
a1 p+1
a1 p+1
=F
2 ab T c1 q [1 F 2 p+1 ]q+1 b Tc1 q+1
G
T

1
1 p+1 1 p+1 b
a
b1 q+1

+
T1 q [1 F2 p+1 ]q+1 T
. (14)
(p + 1)!
If one defines another (p even) or another pair of (p odd) field strength difference(s) that
is/are relevant to Ga1 p+1 ,
G a1 p+1 ab Gb1 p+1
1
1 p+1 1 p+1 Ga1 p+1 ,
(p + 1)!
(15)
and substitutes Eq. (14) into Eq. (15), one then establishes the relation(s) between the two
(p even) or two pairs of (p odd) field strength differences,
F a1 p+1 = G a1 p+1 .
(16)
One may note that such a relationship exists in various chiral p-form actions, which has
been pointed out in both configuration [14] and momentum [17] spaces. With the aid of
Eq. (16), one can easily obtain from Eq. (14) F a1 p+1 expressed in terms of Ga1 p+1
and other fields,

1
F a1 p+1 = Ga1 p+1 + 2 ab T c1 q [1 G 2 p+1 ]q+1 b Tc1 q+1
T

1
1 p+1 1 p+1 b
a
b1 q+1

+
T1 q [1 G2 p+1 ]q+1 T
. (17)
(p + 1)!
We can check from Eq. (14) that when the self-duality condition is satisfied, i.e.,
F a1 p+1 = 0, which is also called an on mass shell condition, F a1 p+1 and
Gb1 p+1 relate with a duality,
F a1 p+1 =
(1)p ab 1 p+1 1 p+1 b

G1 p+1 .
(p + 1)!
(18)
Substituting Eq. (17) into Eq. (12) and dropping a total derivative term, we obtain the dual
action of SST ,

1
dual
Ga
SST
= dDx
Ga1 p+1
2(p + 1)! 1 p+1

b
T b1 q T
1 a
1 q
a
a1 p
a1 p
A
+
G
G
+
G
. (19)
1 p
2T 2
1 p
221
Variation of Eq. (19) with respect to Aa1 p (x) gives Ga1 p = 0, whose solution
has to be
(1)p ab 1 p+1 1 p+1

[1 Bb2 p+1 ]
(p + 1)!
(1)p ab 1 p+1 1 p+1 b

F1 p+1 (B),
(p + 1)!
Ga1 p+1 =
(20)
where Ba 1 p (x) is one (p even) or are a pair of (p odd) arbitrary p-form field(s) we
introduce. Substituting Eq. (20) into Eq. (19), one finally obtains the dual action in terms
of Ba 1 p (x) and other fields,

1
dual
SST
Fa
= dDx
(B)F a1 p+1 (B)
2(p + 1)! 1 p+1

b
T b1 q T
1 q
a
a1 p
+
(21)
F
(B)F
(B)
.
1 p
2T 2
This action, Eq. (21), has the same form as the original one, Eq. (7), only with
the replacement of Aa1 p (x) by Ba 1 p (x). As analysed above, Aa1 p (x) and
Bb 1 p (x) coincide with each other up to a constant when the self-duality condition is
imposed. Consequently, the Siegel-type chiral p-form action is self-dual with respect to
Aa1 p (x) Bb 1 p (x) dualization given by Eqs. (13), (18) and (20).
3.2. Dualization of the auxiliary q-form fields
For the sake of convenience in the following discussion, we rewrite Eq. (7) to be

1
D
Fa
SST = d x
(A)F a1 p+1 (A)
2(p + 1)! 1 p+1

[ Y 1 q ]b [ Yb1 q ]
1
a
a1 p
F
(A)F
(A) .
+
2 [ Yc ] [1 Y 2 q+1 ]c 1 p
1
2
q+1
(22)
We introduce two (for even p) or two pairs of (for odd p) (q + 1)-form fields,
Ua1 q+1 (x) and Vb1 q+1 (x), and replace Eq. (22) by the action,

1
D
Fa
(A)F a1 p+1 (A)
SST = d x
2(p + 1)! 1 p+1
+
+
b
b
1 U 1 q U1 q a
F
(A)F a1 p (A)
2 Uc1 q+1 U c1 q+1 1 p

1
V a1 q+1 Ua1 q+1 [1 Ya2 q+1 ] ,
(q + 1)!
(23)
where Ua1 q+1 (x) and Vb1 q+1 (x) act, at present, as independent auxiliary fields.
Variation of Eq. (23) with respect to V a1 q+1 (x) gives
Ua1 q+1 = [1 Ya2 q+1 ] ,
(24)
222
which yields the equivalence between the actions, Eq. (22) and Eq. (23). On the other hand,
variation of Eq. (23) with respect to Ua1 q+1 (x) leads to the expression of V a1 q+1 (x)
in terms of U b1 q+1 (x) and F c1 p+1 (A),
V a1 q+1 = (q + 1)!U a1 q+1
b
U b1 q U
1 q
(Ud1 q+1 U d1 q+1 )2
1
(Uc1 q+1 U c1 q+1 )
c
F
(A)F c1 p (A)
1 p
Fb1 p [1 (A)Up+1 2 q+1 ]a F b1 p+1 (A).

(25)
Multiplying both sides of Eq. (25) by Ua1 q+1 (x), one obtains
Ua1 q+1 V a1 q+1 = 0.
(26)
Ua1 q+1 (x)
V a1 q+1 (x)
and
generalizes the case
This orthogonality relation between
of two vectors (1-forms) which are external derivatives of one and the same auxiliary
scalar (0-form) adopted in Ref. [13]. Thus, the third term of Eq. (23) reduces to

1
d D x Ya1 q V a1 q ,
(27)
where a total derivative contribution has been dropped. Note that in the dual action of
Eq. (23) that is expressed in terms of Fa1 p+1 (A), Yb1 q (x) and Vc1 q+1 (x), the
first term of Eq. (23) remains unchanged and the second retains no relation with the
auxiliary q-form fields, Ya1 q (x), because no such fields appear in Eq. (25). Therefore,
the variation of the dual of Eq. (23) with respect to Ya1 q (x) is the same as the variation
of Eq. (23) itself with respect to Ya1 q (x). This gives V a1 q = 0, whose solution
has to be
V a1 q+1 =
(1)p
ab 1 q+1 1 2pq+1 [1 Zb2 2pq+1 ] ,
(2p q + 1)!
(28)
where Za 1 2pq (x) is one (p even) or are a pair of (p odd) auxiliary (2p q)-form
field(s) that is/are the dual(s) of the q-form(s), Yb1 q (x), in D = 2(p + 1) dimensions.
In order for the dual of Eq. (23) to possess self-duality, or putting it differently, in order
for Eq. (22) to be self-dual with respect to dualization of the auxiliary fields, Ya1 q (x)
and Zb 1 2pq (x) must have the same form number, q = 2p q, i.e.,
q = p.
(29)
That is, the auxiliary fields have to be p-form. This is the condition that the double selfduality happens.
With Eq. (29) one obtains from Eq. (26) the dual relation between V a1 p+1 (x) and
Ub1 p+1 (x),
V a1 p+1 ab 1 p+1 1 p+1 Ub1 p+1 ,
(30)
where the proportionality coefficient, which is not explicitly shown, is in general a

functional of Ua1 p+1 (x) and Fb 1 p+1 (A), but with no indices left free. By using
223
Eq. (30), one can prove the crucial relation,

b
U b1 p U
1 p
Uc1 p+1 U c1 p+1

=
a
F
(A)F a1 p (A)
1 p
b
V b1 p V
1 p
Vc1 p+1 V c1 p+1
a
F
(A)F a1p (A),
1 p
(31)
where the identity,

Fa 1 p+1 (A)F a1 p+1 (A) = 0,
(32)
which can be deduced from Eq. (11), was used. We note that it is a hard job to determine the
coefficient in Eq. (30) even for the special cases of D = 2, 4, 6 dimensions as can be seen
in Ref. [13]. However, it is interesting that we are not obliged to determine the concrete
form of the coefficient by following our line of analyses shown from Eq. (25) to Eq. (32).
Substituting Eqs. (26), (29), (31) and the special case of Eq. (28) with q = p into Eq. (23)
and using Eq. (32) again, one finally obtains the dual action in terms of the dual fields,
Zb 1 p (x), and Fa1 p+1 (A),

1
dual
D
Fa
SST = d x
(A)F a1 p+1 (A)
2(p + 1)! 1 p+1

[ Z 1 p ]b [ Zb1 p ]
1
a
a1 p
F
(A)F
(A) .
+
(33)
2 [ Zc ] [1 Z 2 p+1 ]c 1 p
1
2
p+1
As expected, the Siegel-type action with auxiliary p-form fields, Ya1 p (x), possesses
self-duality with respect to dualization of the auxiliary fields given by Eqs. (24), (28),
(29) and (30). Note that the dualization of the auxiliary fields is off mass shell, which
is different from the dualization of the chiral p-form fields. This property is obviously
shown in Eq. (25). We mention that the Siegel-type action is not self-dual with respect to
dualization of auxiliary fields if the condition, Eq. (29), is not satisfied.
4. Conclusion
We have constructed the Siegel-type chiral p-form action by realizing the symmetric
second-rank Lagrange-multiplier field in terms of a normalized multiplication of two
(q + 1)-form fields with q indices of each field contracted in the even p case, or of two
pairs of (q + 1)-form fields with q indices of each pair of fields contracted in the odd p
case, where the (q + 1)-form fields are of external derivatives of one auxiliary q-form field
of the former, or of a pair of auxiliary q-form fields of the latter. From this action, one
can deduce the PST action in D = 2 (p = 0) and D = 6 (p = 2) dimensions simply by
letting q equal to zero, or the PST action in D = 4 (p = 1) dimensions by introducing
only one auxiliary 0-form field. This means that the PST action can be treated as a special
example of the Siegel-type action. It is known [1215] that the PST action originates from
the SS action that is dual to the Zwanziger action and that it has close relations with others.
224
For instance, on the one hand, it reduces to the non-manifestly covariant FJ action [5,6]
provided appropriate gauge fixing conditions are chosen, on the other hand, it becomes the
MWY action [11] if one gets rid of the PST actions non-polynomiality and eliminates its
scalar auxiliary field at the price of introducing polynomially auxiliary (p + 1)-forms, or,
vice versa, if one consistently truncates the MWY actions infinite tail and puts at its end
the auxiliary scalar field. Consequently, our result shows that Siegels action is in some
sense the source of chiral p-form actions.
We have shown that the Siegel-type chiral p-form action is self-dual with respect to
dualization of chiral p-form fields when q assumes any of the numbers 0, 1, . . . , 2p, but it
is self-dual with respect to dualization of auxiliary q-form fields only when q equals p, for
a fixed p in D = 2(p + 1) spacetime dimensions. As a consequence, q = p is the condition
under which the double self-duality appears. This result includes PSTs as a special case
where only the chiral 0-form action is doubly self-dual in D = 2 dimensions.
Finally, comment on the odd p case.
(i) The Siegel-type action is still self-dual with respect to dualization of chiral p-form
fields even if only one auxiliary q-form field is introduced. In fact, this self-duality exists
without any relation to concrete realization of the Lagrange-multiplier field.
(ii) Regarding the self-duality with respect to dualization of auxiliary fields, we
emphasize that it is not enough to merely introduce one auxiliary p-form field. The reason
is that no duality relation between V1 p+1 (x) and U 1 p+1 (x), like Eq. (30), could be
derived from Eq. (26) with q = p if there were only one auxiliary p-form field.
Acknowledgements
Y.-G. Miao acknowledges supports by an Alexander von Humboldt fellowship, by the
National Natural Science Foundation of China under grant No. 19705007, and by the
Ministry of Education of China under the special project for scholars returned from abroad.
He would like to thank D. Sorokin for correspondence. D.K. Park acknowledges support
from the Basic Research Program of the Korea Science and Engineering Foundation (Grant
No. 2001-1-11200-001-2).
References
[1] D. Zwanziger, Phys. Rev. D 3 (1971) 880.
[2] R. Banerjee, C. Wotzasek, Phys. Rev. D 63 (2001) 045005.
[3] S. Deser, C. Teitelboim, Phys. Rev. D 13 (1976) 1592;
S. Deser, J. Phys. A 15 (1982) 1053.
[4] N. Marcus, J.H. Schwarz, Phys. Lett. B 115 (1982) 111.
[5] R. Floreanini, R. Jackiw, Phys. Rev. Lett. 59 (1987) 1873.
[6] M. Henneaux, C. Teitelboim, Phys. Lett. B 206 (1988) 650.
[7] A. Tseytlin, Phys. Lett. B 242 (1990) 163;
A. Tseytlin, Nucl. Phys. B 350 (1991) 395.
[8] J.H. Schwarz, A. Sen, Nucl. Phys. B 411 (1994) 35.
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
225
W. Siegel, Nucl. Phys. B 238 (1984) 307.

P.P. Srivastava, Phys. Rev. Lett. 63 (1989) 2791.
B. McClain, Y.S. Wu, F. Yu, Nucl. Phys. B 343 (1990) 689.
P. Pasti, D. Sorokin, M. Tonin, Phys. Lett. B 352 (1995) 59;
P. Pasti, D. Sorokin, M. Tonin, Phys. Rev. D 52 (1995) R4277;
P. Pasti, D. Sorokin, M. Tonin, Phys. Rev. D 55 (1997) 6292.
A. Maznytsia, C.R. Preitschopf, D. Sorokin, Nucl. Phys. B 539 (1999) 438;
A. Maznytsia, C.R. Preitschopf, D. Sorokin, Dual actions for chiral bosons, hep-th/9808049.
Y.-G. Miao, H.J.W. Mller-Kirsten, Phys. Rev. D 62 (2000) 045014.
Y.-G. Miao, R. Manvelyan, H.J.W. Mller-Kirsten, Phys. Lett. B 482 (2000) 264.
R. Banerjee, B. Chakraborty, J. Phys. A 32 (1999) 4441.
Y.-G. Miao, H.J.W. Mller-Kirsten, D.K. Park, Duality symmetry in momentum frame, hep-th/
0105029.
A. Khoudeir, N. Pantoja, Phys. Rev. D 53 (1996) 5974.
E.M.C. Abreu, A. de Souza Dutra, Interference phenomena, chiral bosons and Lorentz
invariance, hep-th/0009157.
Nuclear Physics B 612 [FS] (2001) 229290

Quantum groups and non-Abelian braiding in

quantum Hall systems
J.K. Slingerland, F.A. Bais
Institute for Theoretical Physics, University of Amsterdam, Valckenierstraat 65, 1018 XE Amsterdam,
The Netherlands
Received 10 April 2001; accepted 22 June 2001
Abstract
Wave functions describing quasiholes and electrons in non-Abelian quantum Hall states are well
known to correspond to conformal blocks of certain coset conformal field theories. In this paper
we explicitly analyse the algebraic structure underlying the braiding properties of these conformal
blocks. We treat the electrons and the quasihole excitations as localised particles carrying charges
related to a quantum group that is determined explicitly for the cases of interest. The quantum group
description naturally allows one to analyse the braid group representations carried by the multiparticle wave functions. As an application, we construct the non-Abelian braid group representations
which govern the exchange of quasiholes in the fractional quantum Hall effect states that have been
proposed by N. Read and E. Rezayi [Phys. Rev. B 59 (1999) 8084], recovering the results of C. Nayak
and F. Wilczek [Nucl. Phys. B 479 (1996) 529] for the Pfaffian state as a special case. 2001
Published by Elsevier Science B.V.
PACS: 73.40.Hm; 11.25.H
1. Introduction
In a (2 + 1)-dimensional setting, quantum mechanics leaves room for particles with
exchange properties other than those of bosons and fermions and the exchanges of n such
particles are governed by a representation of the braid group Bn . These representations may
be Abelian, or, more excitingly, non-Abelian. Quasihole excitations of fractional quantum
Hall plateaus have already provided us with examples of the former possibility and may
possibly reveal the latter as well. Several (series of) candidate non-Abelian states have
been proposed in the literature. Examples are the HR-state [3], the Pfaffian state [4], the
spin singlet states of Ardonne and Schoutens [5] and the parafermionic generalisations of
the Pfaffian proposed by Read and Rezayi [1]. It is the last series of states that we will
E-mail addresses: slinger@science.uva.nl (J.K. Slingerland), bais@science.uva.nl (F.A. Bais).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 0 8 - X
230
J.K. Slingerland, F.A. Bais / Nuclear Physics B 612 [FS] (2001) 229290
focus on in this paper, although the methods we use will also be applicable to the other
cases. It has been suggested that the ReadRezayi states should give a good description
of quantum Hall plateaus which occur at several filling fractions [1,6,7]. In particular, the
Pfaffian is thought to describe the plateau observed [8] at filling fraction = 52 . Numerical
support for these claims has been provided in [1,9,10], where it was shown that some of
the RR-states (among which the Pfaffian state) have large overlaps with the exact ground
states for electrons with Coulomb interactions at the same filling fractions. Many aspects of
the ReadRezayi states have already been well-studied. For example, one may show (see
[1,4,11]) that they are exact ground states of certain ultra-local Hamiltonians with (k + 1)body interactions, which gives hope that they will indeed represent new universality classes
of two-dimensional physical systems. Also, their zero modes have been counted and a
suitable basis for the description of the quasiholes has been obtained [11,12]. Finally, there
is recent work which explains how the RR-states may be obtained as projections of Abelian
theories [1315]. Still, the braiding of the quasiholes has been described explicitly only for
the Pfaffian state [2].
One of the general aims of this paper is to analyse some of the physical properties of Hall
systems, not by studying the explicit form of the wave functions but rather by exploiting the
underlying algebraic structure, which in turn derives from the associated conformal field
theories. This allows us, for example, to give an explicit description of the braid group
representations that govern the exchange properties of the quasiholes for all of the RRstates. In order to do this, we first describe the electrons and quasiholes of the RR-states as
particles that carry a representation of a certain quantum group. That such a description is
possible is a logical consequence of the well known relation between quantum groups and
conformal field theories and in fact, we expect that a similar description is possible for all
the non-Abelian quantum Hall states that have been proposed. We believe that the quantum
group description of quantum Hall states will prove a useful complement to the existing
conformal field theory and wave function methods, both technically, because it makes
braiding calculations much easier, and conceptually. The reason that braiding calculations
are so much simplified, is that the quantum group picture allows one to deal with quasiholes
and electrons without dealing with their exact spatial coordinates. Exchanging two particles
becomes a purely algebraic operation, simple enough to be carried out explicitly for large
numbers of particles.
One of the useful features of the present paper is that it does bring together a number of
sophisticated physical and mathematical ingredients, everybody knows to exist in principle,
and integrates them into a rather self-contained toolbox to analyse the problems at hand.
The outline of the paper is as follows. In Section 2, we review the usual description of
the ReadRezayi states in terms of conformal blocks of parafermionic conformal field
theories. We also count the number of independent states with a fixed number of quasiholes
in fixed positions. In Section 3, we give the motivation for the use of quantum groups in the
description of particles with non-Abelian braiding and provide the necessary background.
In particular we describe the braid group representations that describe the exchanges in
a system of localised particles with a hidden quantum group symmetry. In Section 4,
we recall the connection between quantum groups and conformal field theories and in
231
particular, we obtain the quantum groups which can be used to describe the braiding of
the parafermion CFTs which are important for the ReadRezayi states. In Section 5 we
describe the RR-states as systems of point particles with a hidden quantum group symmetry
and give the explicit form of the associated braid group representations. We also check that
the results of Nayak and Wilczek for the case of the Pfaffian are recovered. A discussion
of the results, including questions for future research can be found in Section 6.
2. The CFT description of the ReadRezayi states

In this section, we will review the conformal field theoretic description of the Read
Rezayi states. The definition of these states can be found in Subsection 2.2, after the
required preliminaries on parafermionic conformal field theories in Subsection 2.1. In
Subsection 2.3, we describe the fusion of the ReadRezayi quasiholes in terms of paths
on a Bratteli diagram and in the following subsection, we use this description to calculate
the dimensions of the braid group representations which govern the exchanges of these
quasiholes. The final subsection is devoted to a description of the results of Nayak and
Wilczek on the braiding of the quasiholes of the Pfaffian or MooreRead state, which is
the simplest state in the ReadRezayi series.
2.1. The parafermion CFT
Like many other (candidate) quantum Hall states, the RR-states can be described using
conformal blocks of a rational conformal field theory. Here, the theory in question is the Zk
parafermionic theory of Zamolodchikov and Fateev [16,17]. Before we write down any
explicit expression for the RR-states, we recall some well known facts about this CFT.
The Zk parafermionic CFT may be described completely in terms of a chiral algebra
generated by the modes of k parafermionic currents (see [16,17] and also [18] for some
more recent work in this vein), but it also has two different coset descriptions. The cosets

1 sl(k)
1 /sl(k)
2 . The first of these descriptions was
k /U
(1)k and sl(k)
involved are sl(2)
used extensively already in [16,17], to determine fusion rules, characters and partition
functions for the parafermions. The treatment of the parafermions in most of the literature
on the RR-states has been influenced by this description. The second coset was introduced
by Bais et al. in [19,20] and used in [21] to construct a Coulomb gas representation of the
theory which led to alternative character formulae [22]. This coset description has recently
also been used in the work of Cappelli, Georgiev and Todorov on the RR-states [13]. In the
rest of this section, we will give a quick description of both pictures and indicate how they
are connected.

k /U
(1)k
2.1.1. The coset sl(2)
k /U

We start with the coset sl(2)
(1)k . For this coset, we have more information about
the coset primaries than usual. In particular, one can decompose certain fields of the parent
WZW-theory as a product of a coset primary field and a U (1) primary field (see
sl(2)
formula (6) below). In order to describe this decomposition, it is convenient to start with a
232
theory before moving on to the

short description of the fields and fusion rules of the sl(2)
parafermions (for much more detail on WZW-theories, see, for example, [23]). Note that
when we speak of primary fields in the sequel, we will always mean chiral primary fields.
k model is the affine Lie
Recall that the spectrum generating algebra of the sl(2)
at level k. The Virasoro algebra is embedded in the enveloping algebra of the
algebra sl(2)
affine algebra through the Sugawara construction. When discussing primary fields of the
k model, we need to distinguish between primary fields of the affine algebra (affine
sl(2)
primaries) and primary fields of the Virasoro algebra (Virasoro primaries). Each affine
primary field is necessarily also a Virasoro primary, but not vice versa. In fact, one can
always find infinitely many Virasoro primaries among the affine descendants of an affine
primary.
Let us be more explicit. If is the highest root of a simple Lie algebra g, then the affine
primaries of the gk model are labelled by the dominant integral weights of g for which
(, ) k. For g = sl(2) this just means 0 k. Let us call these fields G . The
conformal dimension h of G is given by
h =
( + 2)
.
4(k + 2)
(1)
The fusion rules of the G are

G G
,2k }
min{+

G .
(2)
=| |
module with
There is an affine descendant field of G for each of the states in the sl(2)
highest weight . Among these descendants, there are infinitely many Virasoro primaries,
which we may name G

. The field G is by definition the field of lowest conformal
dimension among the affine descendants of G which carry sl(2)-weight . Naturally,
we have G = G
. Also, we have to demand that ( ) = 0 (mod 2), otherwise the
weight will not appear in the representation with highest weight . One may check (see,
for instance, [23]) that all the G
defined this way are indeed Virasoro primary. Their
conformal weights are given by
( + 2)
+ n
(3)
,
4(k + 2)
where n, is the lowest grade at which the weight appears in the affine Lie algebra
representation of highest weight . If is a weight in the (ordinary) Lie algebra
representation of highest weight , then n, will be zero and we will have h = h
.
The fusion rules of the G are easily obtained from (2) and the sum rule for weights in
operator products. They are
h
=

G

,2k }
min{+

G
+ .
(4)
=| |
k /U

Now let us turn to the Zk parafermion theory, as described by the coset sl(2)
(1)k .
As usual for cosets, the Virasoro primary fields of the parafermion CFT may be labelled
233
by a highest weight of the horizontal algebra of the parent theory (sl(2)) and by a
similar weight P of the embedded theory (U (1)), which is obtained by a projection
matrix P from a weight of the parent theory. These weights moreover have to satisfy
a branching condition, which ensures that the representation P of the embedded algebra
can occur as a summand in the decomposition of the representation of the parent algebra
into representations of the embedded algebra. If we denote by M the root lattice of the
horizontal algebra of the parent algebra, then this branching condition is
P P PM.
(5)
k /U

In the case of sl(2)
(1)k , the projection matrix is trivial and the branching rule just says
that the difference of the weights and has to be an element of the root lattice of sl(2),
i.e., the difference of and has to be an even number. Thus, the parafermion theory has
Virasoro primaries labelled by a highest weight 0 k of sl(2) and a weight of
sl(2) for which we have = 0 (mod 2).
Since the parafermion fields are now labelled in the same way as the Virasoro

primary fields G
of the sl(2)k theory, one might hope that there is a simple relation
between these fields. In fact, it was pointed out already in [16] that each of the fields
G
may be written as the product of a field from the parafermion theory and a

vertex operator
of the U
(1)k theory, which is just the theory of a free boson on a circle
of radius 2k. This was further clarified in [17], using the results of [24]. One has
i/
G
= e
2k
(6)
From this relation, one immediately reads off that the field must have conformal weight
(h )
given by

2 ( + 2) 2
h = h
=
+ n

.
4k
4(k + 2)
4k
(7)
As in other coset theories, the labelling of the fields as we introduced it above is

redundant. First of all, the U (1) label is usually taken to be defined modulo 2k, since

and +2k , that correspond to the vertex operators
the (extended)
U
(1)k characters
i/
2k
i(+2k)/
2k
e
and e
, are equal (see, for example, [23]). Because of this and because
of the fusion rules (9) below, the label is called the Z2k charge of the field . 1 Also, in
order to get proper behaviour of the fields characters under modular transformations, one
has to identify fields whose labels are sent onto each other by an external automorphism of
the parent algebra [25]. In the case at hand, this means that we have to identify with
k
k
. Collecting, we get the field identifications
+2k
,
k
k
.
(8)
Using these identifications, we can choose a labelling of the primaries such that is
a weight in the representation of sl(2), i.e., . In fact, we may require
1 Note that in the original parafermion theory of [16], there was a Z
Zk symmetry. The Zk
Zk charge
k
of the field (z) (z) was given by l = 1 ( + ),
l = 1 ( ),
so that clearly in this theory, one
(l, l)
2
2
needed + to be even. Here, we will not require this and thus allow chiral fields like 11 .
234
< and if we do this then every set of labels corresponds uniquely to a Virasoro
primary Thus, the number of Virasoro primaries is 12 k(k + 1) (note: there are only k
).
primaries of the full parafermion algebra: the fields
One may check that the conformal weights given in (7) are equal for identified fields.
Also, note that the grade n
in (7) is zero if the labels (, ) are in the range chosen above.
Using the factorisation (6) and the field identifications, we may now also write down the
fusion rules for the parafermion fields. They are

,2k }
min{+

+
.
(9)
=| |
In other words, they are the same as the fusion rules for the G
, except that the labels
on the right hand side have to be brought back into the set chosen above, using the field
identifications (8).
1 /sl(k)
2
1 sl(k)
2.1.2. The coset sl(k)
1 sl(k)
1 /sl(k)
2 . This coset is a special case of the general
We now turn to the coset sl(k)
class considered in [19,20]. Its current algebra is denoted as a W -algebra and much is
known about such algebras. In the quantum Hall application however, the parafermion
analyses is more directly relevant and applicable as is described in some detail in [13].
Nevertheless one should keep in mind that the discussion of the braid group representations
that feature in these models and which will be extensively analysed later on, readily extend
to the W representation theory. The Virasoro primaries of the coset may be labelled by an
1 weight (or, equivalently, two sl(k)
1 weights) and an sl(k)
2 weight. Let us
1 sl(k)
sl(k)
2 weight , then we can write 1 ,2 . The
1 weights 1 and 2 and the sl(k)
call the sl(k)
weights 1 , 2 and once again have to satisfy the branching condition (5). In this case,
the projection P maps (1 , 2 ) onto 1 + 2 and it maps the root lattice of sl(k) sl(k)
onto the root lattice of sl(k). Hence we have the following requirement
1 + 2 Msl(k) ,
(10)
where Msl(k) is the root lattice of sl(k). In other words, the weights 1 + 2 and should
be in the same conjugacy class (for details on this concept see, for example, [23,26]). In
terms of the Dynkin labels of the weights, this means that one has
k1

(j )

(j )
j 1 + 2 (j ) = 0 (mod k).
(11)
j =1
(j )
(j )
Now denote by ei the sl(k) weight whose Dynkin labels ei are given by ei = ij . (These
correspond to the fundamental representations of sl(k).) Then 1 is either zero or equal to
one of the ei , since it is a level one weight. The same goes for 2 . For the level two
weight , there are three possibilities: it can be zero, equal to one of the ei or equal to the
sum of two of the ei (which may be the same). If we define e0 = 0, then we may simplify
this description and say that 1 and 2 will equal one of the ei and will equal the sum
of two of the ei (where i {0, . . . , k 1}). The branching rule above then states that only
235
triples (1 , 2 , 3 ) of the form (el , em+nl (mod k) , em + en ) are admissible. This leaves
1 2
2 k (k + 1) admissible triples. However, there are also field identifications, induced by the
1 sl(k)
1 . These identifications take the form
external automorphisms of sl(k)
e ,e
,e
i+s j+s
eli+ejm el+s
+em+s ,
(12)
for s {1, . . . , k 1}. The sums in the indices on the right hand side have to be taken
modulo k. Using these identifications, we can choose to set either 1 or 2 to zero. Say we
set 1 to zero. Then we are left with the triples (0, em+n (mod k) , em + en ). Clearly, 2 is
2 weight
now uniquely determined by and we may choose to label the fields by the sl(k)

only: . Every sl(k)2 weight is admissible and we are left with as many Virasoro primary
2 weights: 1 k(k + 1). This is just a reduction of the number of fields
fields as there are sl(k)
2
before identification by a factor of k, as should be expected. Also, we get the same number
of fields that we got in the other coset description of the parafermions.
,
The fractional part of the conformal weight of the field 1 2 can be calculated
directly from the coset description; it is the same as the fractional part of the difference
between the conformal weight of field with labels (1 , 2 ) in the parent theory and the
conformal weight of the field with label in the embedded theory. One may show that this
recipe always yields the same fractional part, independently of the labels 1 , 2 , that
are chosen to represent a certain field (i.e., labels that are identified through (12) yield
the same fractional part). Let us look at the field em +en , with m n. A particularly
convenient choice of labels for this field, made in [13], is (ekn , em , ek+mn ). The
conformal dimension of the WZW-field labelled by the weight em is given by
(em , em + 2) m(k m)(k + 1)
=
,
(13)
2(p + k)
2k(k + p)
where is the Weyl-vector of sl(k) and p is the level (here, we have p = 1 or p = 2). From
this, we find
p (em ) =
1 (ekn ) + 1 (em ) 2 (ek+mn )

m(k n) (n m)(k + m n)
+
=
k
2k(k + 2)
(k + m n)(k + m n + 2) (m + n k)2
.
=
(14)
4(k + 2)
4k
The middle expression is the one given in [13] and from the last expression, we see that
it is equal to the weight of the field with = k + m n and = m + n k (cf.
formula (7)). Thus, we have the correspondence
k+mn
em +en
m+nk
e + +e 2k+ ,
2
(15)
which is further supported by the fact that these fields have the same fusion rules. 2 In
2
fact, the fusion rules are the same as the fusion rules for the corresponding sl(k)
k+mn
2 Note that we could also identify the field
em +en with the field kmn , which is the conjugate of the field
k+mn
m+nk . It is impossible to decide between these identifications on the level of conformal weights and fusion
rules.
236
k /U

representations and these are the same as the fusion rules of the sl(2)
(1)k coset as a
consequence of level-rank duality (see [23] and references therein). One may also find the
equality of the fusion rules directly by looking at the fusion rules of the field 11 e1
with an arbitrary field. These fusion rules are easily seen to be the same and since the
field 11 generates all the fields in the theory by repeated fusion, it follows that the fusion
rules of all the fields that are identified through (15) are the same in both cosets.
2.2. Definition of the RR-states
As we mentioned before, the wave functions for the RR-states can be described (or even
defined) as conformal blocks in a certain conformal field theory. The CFT in question is
the tensor product of the parafermion theory with the theory of a chiral boson.
From the parafermion theory, one needs the chiral vertex operators := 11 = e1 and
:= 20 = 2e1 defined above and from the bosonic theory one uses the usual vertex
operators ei , of weight 12 2 . Here, is the bosonic field (which should not be confused
with the bosonic field in the factorisation formula (6)). Electrons and quasiholes are
represented by products of these bosonic and parafermionic
vertex operators. Specifically,

the electron is represented by the product e
kM+2
k
i
k(kM+2)
, where M Z and the quasihole is
represented by the product e

. These combinations of bosonic and parafermionic
fields are not arbitrary; if the parafermion factor is given, then the bosonic factor for the
electron can be fixed by requiring that electrons are mutually local (that is, the OPE of two
electron operators does not have a branch cut). This requirement is needed to make sure
that the wave functions defined below are single valued in the electrons coordinates. If
the electron is to have half-integer spin, then one should also require that M be odd. The
k
by
exponent of the bosonic factor for the quasihole is fixed up to integer times kM+2
the requirement that the quasihole and the electron are mutually local.
k which have N electrons with coordinates z , . . . , z
The linear space of RR-states N,n
1
N
and n quasiholes located at positions w1 , . . . , wn is now generated by the conformal blocks
of a correlator of N electron fields and n quasihole fields inserted at these positions and
supplemented by a homogeneously spread positive background charge, which ensures
overall charge neutrality [1,4]. This correlator may be factorized into parafermion and
boson correlators, the latter of which may be evaluated explicitly (paying due attention
to the background charge [4]), after which one obtains
k
(z1 , . . . , zN , w1 , . . . , wn )
N,n
N
= (w1 ) (wn )(z1 ) (zN )

(zi zj )M+2/ k
(zi wj )1/ k
i<j
(wi wj )
1
k(kM+2)
Fg (z1 , . . . , zN , w1 , . . . , wn ).
i=1 j =1
(16)
i<j
Here, the zi and wi are complex coordinates which parametrise the sample. Fg is a factor
which depends on the geometry of the sample. If the sample is a disc then this factor just
237
implements the usual Gaussian factors which confine the electrons and quasiholes to the
disc. If the sample is a sphere of radius R, then the zi and wi are related to the usual
coordinates on the sphere through stereographic projection (see [1] for details) and one has
Fg (z1 , . . . , zN , w1 , . . . , wn )

N /21

N /21
1 + |zi |2 /4R 2
1 + |wj |2 /4(kM + 2)R 2
=
,
i
(17)
where N = N/ M 2 + n/k is the number of flux quanta that go through the surface
k
is the filling fraction.
of the sphere and = kM+2
The construction of the Pfaffian and RR-states through conformal field theory is
analogous to the construction of the Laughlin state and the Pfaffian state through conformal
field theory as given in [4]. Indeed, the k = 2 case of the above wave function just describes
the Pfaffian state of [4] with N electrons and n quasiholes.
2.3. Fusion of quasiholes and the Bratteli diagram
It is interesting to know the number of independent states which the formula (16)
encodes, i.e., the number of independent states with N electrons that have n quasiholes
at fixed positions w1 , . . . , wn . This interest is twofold. First of all, we want to know which
combinations (N, n) are allowed. Second, the number of independent states is also the
dimension of the braid group representation that governs the exchanges of electrons and
quasiholes. Hence a necessary condition for non-Abelian braiding is that it be larger than
one. A basis for the space of states that we are looking for is given by the states we obtain
if we replace the parafermion correlator in (16) by its respective conformal blocks. The
number of such blocks is equal to to the number of fusion channels that make the correlator
in (16) non-vanishing. Hence, the number we are looking for is just the number of ways in
which N electron fields and n quasihole fields may fuse into the vacuum.
Now the fusion of the fields is very simple; it just corresponds to addition of the Z2k
0 =
charges. Hence the N electron fields fuse into the 2N
2eN sector. The fusion rules
of the sigma fields, as given in Eq. (9), are a bit more complicated, but they have a nice
graphical description in terms of a Bratteli diagram (see Figs. 1, 2).
These diagrams must be read as follows: each starting point or end point of an arrow has
coordinates (, ) and represents the sector of the parafermion CFT. Note that this
means that coordinates related by the identifications (8) represent the same sector. In Fig. 2,
one may see this explicitly for k = 3. Here, we have at each node of the diagram inserted the
2 weight of the field which resides there. The correspondence
Young diagram for the sl(3)
between the fields of the parafermionic theory and such weights or diagrams is one to one
and we see that the same diagram appears in different places. The fusion rules of the sigma
field are encoded in the arrows; we start in the lower left corner, i.e., the 00 sector which
is the vacuum sector of the theory. Then we take the operator product expansion with the
field = 11 , which naturally, following the arrow, lands us in the 11 sector. Once more
taking the OPE with , we end up, following the arrows, in the 22 or in the 20 sector.
238
Fig. 1. Fusion diagram for the field . The diagram must be thought extended indefinitely in the
-direction and up to = k in the -direction (the case k = 3 is as drawn here). On each line, we
have drawn the Young diagram of the sl(2) representation that resides on that line.
Fig. 2. The same diagram as in Fig. 1, but this time each site in the diagram is labelled by the

Young diagram for the sl(k
= 3)2 weight of the field that resides there. The dot represents the empty
diagram. Again, generalisation to arbitrary k is straightforward. Note that in this picture, the weights
label the fields unambiguously, whereas in Fig. 1, one still has to take the field identifications (8) into
account.
In this way, each path of length n through the diagram represents a fusion channel for n
-fields.
To make the correlator in the wave function non-vanishing, all the quasihole and electron
fields need to fuse into the vacuum sector. Now since the electron fields (z1 ), . . . , (zN )
0 , the quasihole fields (w ), . . . , (z ) have to fuse to
in the correlator above fuse to 2N
1
n
0
k
the field 2N = k2N . The number of ways to do this is just the number of paths of
length n through the diagram of Fig. 1 which end up at a point whose coordinates (, )
satisfy either (, ) = (0, 2N (mod 2k)) or (, ) = (k, k 2N (mod 2k)). Clearly,
for fixed N, such paths occur only for values of n which are a multiple of k apart, so
quasiholes can only be created in multiples of k at a time (maybe with the exception of
the first few quasiholes if N is not a multiple of k). Note that, although the same fields (or
sectors) occur in different heights in the diagram, the same field never occurs more than
239
once at given and hence different paths are never identified by the field identifications.
Thus, the number of fusion channels for the parafermion CFTs is the same as that for the
corresponding WZW-theories.
2.4. Counting the independent n-quasihole states
Let us denote the number of paths through the Bratteli diagram which end up at the point
(, n) by D(, n). Also, let us define D(, n) = 0 if there is no point with coordinates
(, n). The number of independent n-quasihole states encoded by (16) is then D(0, n) in
case 2N + n = 0 (mod 2k), D(k, n) in case 2N + n = k (mod 2k), and zero otherwise.
It should be obvious from looking at the Bratteli diagram that the D(, n) satisfy the
following recursion relation:
D(, n) = D( 1, n 1) + D( + 1, n 1).
(18)
Using this relation and the fact that D(1, 1) equals one, D(, n) can be easily calculated
in each particular case. At least for low k, the recursion relation can also be used to prove
simple closed expressions for the D(, n). In particular, we find for k = 2, k = 3 and k = 4
D2 (0, 2n) = D2 (1, 2n 1) = 2n1 ,
D3 (0, 2n) = D3 (1, 2n 1) = Fib(2n 2),
D3 (2, 2n) = D3 (3, 2n + 1) = Fib(2n 1),
3n1 + 1
,
D4 (0, 2n) = D4 (1, 2n 1) =
2
D4 (2, 2n) = 3n1 ,
3n1 1
.
D4 (3, 2n + 1) = D4 (4, 2n + 2) =
(19)
2
In these equations, we have written Dk instead of D for clarity and we have used the
notation Fib(n) to denote the nth Fibonacci number, defined by
Fib(0) = Fib(1) = 1,
Fib(n + 1) = Fib(n) + Fib(n 1).
It is also not that difficult to find and prove a closed formula for infinite k. We have

+1 n+1
D (, n) =
(n + = 0 (mod 2)).
n + 1 n
2
(20)
(21)
Of course this formula is valid for all k as long as n + 2k.

To get formulae for other values of k it is more convenient to rewrite the recursion
relation (18) in matrix form. We consider the D(, n) at a fixed n together as a k-vector
and write the step from n to n + 1 as multiplication with a (k + 1) (k + 1) matrix Mk .
That is, we have
D(0, n)
D(0, n + 1)
..
..
,
= Mk
(22)
.
.
D(k, n)
D(k, n + 1)
240
where Mk is given by
(Mk )ij = i,j +1 + i+1,j .
(23)
The asymptotic behaviour of the D(, n) for large n will be related to the largest
eigenvalue of the matrix Mk . The eigenvalues of the Mi are just the zeros of their
characteristic polynomials Pk . For these, we can easily deduce a recursion relation and
initial conditions:
P2 () = 2 1,
P3 () = 3 2,
Pi+1 () = Pi () Pi1 (),
(24)
but these are just the defining relations for the Chebyshev polynomials, whose zeros are
given by (see, for example, [27])

(m + 1)
.
k,m = 2 cos
(25)
k+2
Since we know all the eigenvalues of Mk , we can now in principle solve for the
eigenvectors and using the solution, give explicit formulae for the Dk (, n) for any k.
We will however content ourselves with giving the asymptotic behaviour of the Dk (, n)
at large n. The largest eigenvalues (in absolute value) of the matrix Mk are clearly 0 and
k = 0 . Hence, the asymptotic behaviour of the Dk (, n) is given by
n

( + n even),
Dk (, n) 2 cos
k+2
Dk (, n) = 0 ( + n odd).
(26)
This conforms with the closed formulae we gave for k = 2, 3, 4.
2.5. Braiding for k = 2
In the previous section, we have calculated the dimensions of the braid group
representations that govern the exchanges of the electrons and the quasiholes of the RRstates. We have seen that these dimensions increase with the number of quasiholes, which is
an indication for non-Abelian braiding. However, this indication is not conclusive evidence.
To be sure, one needs to calculate the actual matrices that describe the braiding of the
-fields in the conformal block in formula (16) above. Nayak and Wilczek [2] have done
this calculation for the case k = 2 (the Pfaffian). The method they used was basically
to compute the conformal block for four quasihole fields explicitly and then to extend
the resulting braid group representation to a braid group representation for any even
number of quasiholes. 3 For general k, it is quite difficult to calculate conformal blocks
for four, let alone for arbitrary numbers of quasiholes. Fortunately it turns out that we
can circumvent this problem by using the known duality between conformal field theory
3 Note that the four point blocks in the case k = 2 are just the four point blocks for the chiral Ising model, which
have, within a different context, been known for a long time (see, for instance, [28] for explicit expressions). The
same is true for the corresponding braid group representations. However, the embedding of the resulting braid
group representation into a rotation group, as given by Nayak and Wilczek (see below) seems to be new.
241
and quantum groups and using this, we will give a nice description of the braiding for
arbitrary k. However, we will first briefly recall the results of Nayak and Wilczek for
k = 2, for later reference.
The braid group representation for n = 2m quasiholes has dimension 2m1 (cf. (19)).
Nayak and Wilczek describe this space as a subspace of a tensor product of m twodimensional spaces. Each of the two-dimensional spaces has basis vectors {|+, |} and
the physical subspace of the tensor product is the space generated by the vectors whose
overall sign is positive (so for m = 2, | is physical, but |+ is not). On the tensor
product space, there is a spinor representation of SO(2m) U (1). The U (1) acts as a
multiplicative factor, while the generators ij of the SO(2m) may be written in terms of
the Pauli matrices i . We have
1
ij = i[i , j ],
4
(27)
with
1 = 1 3 3 ,
2 = 2 3 3 ,
3 = 1 1 3 3 ,
4 = 1 2 3 3 ,
..
.
2m = 1 1 2 .
(28)
Here the states |+ and | are the spin up and spin down states for the Pauli matrices.
Now recall that the braid group Bn is generated by n1 elementary exchanges 1 , . . . , n
subject to the relations
i j = j i
(|i j | 2),
i i+1 i = i+1 i i+1 .
(29)
Here, i represents an exchange of quasihole i and quasihole i + 1. The action of the braid
group on the n-quasihole space is embedded in the action of SO(2n) U (1), as follows:
i ei 4 ei 2 i,i+1 .
(30)
The SO(2m) generators i,i+1 which appear in this equation are given by
1
3 1 1,
2
1
2,3 = 2 2 1 1,
2
1
3,4 = 1 3 1 1,
2
1
4,5 = 1 2 2 1 1,
2
1,2 =
etc.
(31)
242
So we see that, for odd i, i acts only on the ith tensor factor, whereas for even i, i acts
only on the (i 1)th and ith tensor factors. Moreover, the 2 2-matrix by which describes
the action for even i and the 4 4-matrix which describes it for odd i do not vary with i.
Explicitly, they are given by
1+i
0
0
1 + i

1 0
1 0
1+i 1i
0
.
2i+1
(32)
,
2i
0 i
1i 1+i
0
2 0
1 + i
0
0
1+i
3. The quantum group picture
In this extensive section, we give a description of the braiding for a system of n particles
with a hidden quantum group symmetry. We expect that the braiding properties of a
quantum Hall state with n quasiholes is conveniently described in terms of such a system.
In the first subsection, we motivate such a description and mention some general features.
In the remaining subsections, we give a fairly detailed description to the quantum group
Uq (sl(2)) and its representation theory for q a root of unity, which culminates in an explicit
description of the braid group representations that are associated to this quantum group. We
are well aware of the fact that most of the material treated in this section is not new, except
possibly for the identities (92) for the 6j -symbols, however the input came from quite a
variety of sources and putting it together in comprehensive way seemed a nontrivial and
even useful thing to do. It should serve as a quick introduction to braiding in systems with
hidden quantum group symmetry, which is why we have tried to keep the treatment as
self-contained as possible.
3.1. Using a quantum group rather than the full CFT
One may always choose to describe a quantum system in terms of its explicit wave
functions but we know that it can be extremely profitable to exploit its operator algebra, in
particular its symmetries. These allow one to extract many of the physical features without
reference to the explicit realisation in terms of wave functions. Quite similarly one could
in the present context remark that there is an aspect of the description of the ReadRezayi
states that is less than satisfactory: one has to use the full machinery of a (conformal) field
theory to calculate wave functions or even just braiding properties for a finite number of
quasiholes and electrons. There are many questions one may want to answer for which this
seems like overkill: for example, one would hope to be able to describe the braiding of
finitely many particles by means of a theory with only finitely many degrees of freedom.
Indeed, there is such an alternative description and we pursue it here. It is well known
that conformal field theories possess a hidden quantum group symmetry (see Section 4 for
details and references). What we propose is to describe the electrons and quasiholes of a
quantum Hall state that would usually be described by a certain CFT as localised particles
that carry representations of the quantum group that is associated with this CFT. Such a
description has several advantages.
243
It avoids the introduction of a field theory to describe a system with only a finite
number of particle degrees of freedom.
It provides a conceptual understanding of a phenomenon which emerges in the
usual CFT description. This is the fact that, while a state with a low number of
indistinguishable quasiholes can be described with a one component wave function,
a system with a higher number of these quasiholes may need a several component
wave function. Clearly, it should only be possible to distinguish between the
components of the wave function by making a measurement that involves several
holes (otherwise the holes would not be indistinguishable). Hence, there should
be operators in the many hole Hilbert space that distinguish states that cannot be
distinguished by operators that act only on the state of one of the particles. The
quantum group picture provides these in a natural way. They are the operators
that correspond to the global (i.e., invariant) quantum group charges of groups of
quasiholes. Even though all individual quasiholes have the same quantum group
charge, i.e., belong to the same representation, a group of n such holes can occur
in different representations leading to distinguishable n-hole states. As a simple
example, suppose that the quasiholes carried the two-dimensional representation of
SU(2) (or equivalently, of the quantum group U (sl(2))). In that case a two quasihole
state could be either in the singlet or in the triplet representation and the singlet states
could be distinguished from the triplet states by measuring the global charge.
The quantum group picture allows for a very elegant description of the braiding
properties of the n-quasihole states; all braiding properties are encoded into a single
algebraic object: the quantum groups R-matrix (cf. Section 3.2.6). Starting from the
R-matrix, braiding calculations can be done in a purely algebraic way and often
a detailed picture of the braid group representation that governs the exchanges of
particles can be constructed. In a CFT description, the information contained in the
R-matrix of the quantum group would be much less manifest. In fact, to extract it
from this description of the system, one would have to calculate the braiding and
fusion matrices starting from the conformal blocks of the CFT, which is usually quite
hard.
Of course the description we propose also has its disadvantages when compared to the
CFT description. For instance, it seems much harder to describe dynamical aspects of
the quantum Hall states in this picture. Still, we like to emphasise that the quantum group
picture we propose is a useful complementary way of thinking about non-Abelian quantum
Hall states.
3.2. Uq (sl(2)) for pedestrians
In this section we have collected some basic information about quantum groups in
general and the quantum group Uq (sl(2)) in particular. For those well versed in math there
are excellent text books, for example, [29], covering the full story.
244
3.2.1. The algebra and its unitary representations

Quantum groups are actually not groups, but algebras. To be a quantum group, an
algebra has to have several structures associated with it. As an algebra, it is already a
vector space with a multiplication, but next to this, there should also be a comultiplication,
counit, antipode and -structure (which makes the algebra a Hopf--algebra), an R-matrix
(which makes it a quasitriangular Hopf--algebra) and optionally a coassociator (which
would make it a quasitriangular quasi-Hopf--algebra or something with a longer name).
We briefly describe these structures and their use in our context in the following sections,
always using the quantum group Uq (sl(2)) as an example. In this section, we will describe
the algebra Uq (sl(2)) and its irreducible representations.
Uq (sl(2)) can be defined as the algebra generated by a unit 1 and the three elements
H, L+ and L . These satisfy the relations
q H/2 q H/2
(33)
L+ , L = 1/2
,
q q 1/2
where q may be set to any non-zero complex value. One may check that these relations
reduce to those of the Lie algebra sl(2) when q goes to one. Hence, U1 (sl(2)) is just the
universal enveloping algebra U (sl(2)) of sl(2) and we say that Uq (sl(2)) is a q-deformation
of U (sl(2)).
If q is taken as a formal variable and also for all q C, q not a root of unity, the
representation theory of Uq (sl(2)) is very similar to that of U (sl(2)). For each non-negative
j 12 Z there is an irreducible highest weight representation of dimension 2j + 1. We
will denote this representation by , where = d 1 = 2j is the highest weight. The
modules V of these representations have an orthonormal basis that consists of kets |j, m,
with m = j, j + 1, . . . , j , and the generating elements H, L+ and L act on this basis
as follows

H, L = 2L ,
H |j, m = 2m|j, m,

L |j, m = j mq j m + 1q |j, m 1.
(34)
Here, the q-number mq is defined as

q m/2 q m/2
.
(35)
q 1/2 q 1/2
These q-numbers enter the formulae for the representations through the commutation
relation of L+ and L , the right hand side of which can be written H q . The q-number
mq approaches m when q goes to one and hence we see that the representations given
above reduce to the usual U (sl(2)) representations for q = 1. Furthermore, when q R or
|q| = 1, the q-number xq is a real number when x is a real number.
We will call a representation of Uq (sl(2)) unitary when (L+ ) = L , H = H. More
formally, we may define an antilinear algebra antihomomorphism on Uq (sl(2)) by

L = L ,
(36)
(H ) = H,
mq =
and a representation of Uq (sl(2)) is then called unitary, or a -representation, exactly if

x Uq sl(2) : (x) = (x) .
(37)
245
Fig. 3. Diagram of an indecomposable representation as defined by (34). The dots represent the basis
states |j, m, in particular, we have written |h for the highest weight state and |l for the lowest
weight state. The arrows and indicate the action of L+ and L , respectively.
In the representations defined above, we always have (H ) = (H ) and (L ) =

(L ), where the bar denotes complex conjugation of the elements. Hence, these
representations are unitary when the matrix elements of L are real, i.e., if the square
root in (34) is real for all admissible values of m. This will certainly be the case if q is real
and positive and also if q = ei with R, || 2j2+1 . Thus we see that, for real q, all
the representations above are -representations, while for q = ei , the representations 2j
with || 2j2+1 are -representations.
The q-numbers satisfy many identities which are useful in representation theoretic
calculations. Two examples of such identities, which hold for all q C are
q n/2 mq + q m/2 nq = m + nq ,
n + mq n mq = n2q m2q .
(38)
3.2.2. Representations at q = e2i/(k+2)

When q is set equal to a root of unity, the properties of most of the representations
defined by (34) change quite drastically. Specifically, at q = e2i/(k+2), the representations
2j with j > k+1
2 will no longer be irreducible. This can be traced back to the fact that for
q = e2i/(k+2), the q-numbers satisfy the extra identity
m + k + 2q = mq ,
(39)
and more specifically

k + 2q = 0.
(40)
Because of this, one has (L+ )k+2 = (L )k+2 = 0 in all the representations defined
by (34). Of course, in the representations with j < k+2
2 , this was already the case and
for these representations, nothing essential changes. In particular, they are still irreducible.
However, in the representations with j k+2
2 , there will now be extra highest and lowest
weight states, which are annihilated by L+ , respectively, L . For example, the state
(L+ )k+1 |j, j in the module V 2j of the representation 2j (j k+2
2 ) will now be an
extra highest weight state, since (L+ )k+2 = 0 in this representation. The descendants of
this highest weight state (that is, the states which can be obtained from it by applying
powers of L ) now span an invariant subspace W of V 2j , so 2j is no longer irreducible.
Fig. 3 illustrates this situation in a simple case.
Although the module 2j is now reducible, it cannot be written as a direct sum of
irreducibles. One says that it is indecomposable. This indecomposability is directly related
to the fact that 2j is not a -representation. For a -representation, the orthogonal
complement of an invariant submodule of the representation module is itself invariant and
246
this guarantees that any finite dimensional representation has an orthogonal decomposition
into irreducibles. The fact that 2j does not have a decomposition into irreducibles shows
that it is not just non-unitary, but even non-unitarisable. That is, it is impossible to choose
an inner product such that 2j is unitary with respect to it.
Summarising, for q = e2i/(k+2) , we are left with only k + 2 irreducibles out of the
infinitude that we would usually get from (34). 4 These are the unitary representations 2j
with j < k+2
2 . The other representations defined by (34) are no longer irreducible. They
have become indecomposable, and therefore they are non-unitarisable.
3.2.3. Tensor products of representations; the coproduct
The Hilbert space of a system of N particles that carry quantum group representations
1 , . . . , N is just the tensor product V 1 , . . . , V N of the modules of these representations. Thus, we want to have representations of the quantum group on such tensor product
spaces. The decomposition of these tensor product representations into irreducibles will
then give the fusion rules for the particles in the theory. Clearly, it is desirable that the
tensor product decomposition should be orthogonal, since otherwise the quantum group
charges of fusion products would not be well-defined. Orthogonality of the decomposition
can certainly be realised if the tensor product representation is unitarisable. Therefore, we
will spend some time determining which tensor product representations of Uq (sl(2)) are
unitarisable.
For any quantum group A, tensor product representations are formed with the help of
a so called coproduct. This coproduct is an algebra homomorphism from A to A A.
The coproduct of Uq (sl(2)) is given on the generators and the unit 1 by
(H ) = 1 H + H 1,

L = L q H/4 + q H/4 L ,
(1) = 1 1.
(41)

Given two representations , of the quantum group, we may form the tensor product
representation by the formula

x ! (x) .
(42)
Here, x is an arbitrary element of A and one may check that, for A = Uq (sl(2)) and
q = 1, we recover the ordinary tensor product of U (sl(2)) representations. The fact that
is an algebra homomorphism ensures that the tensor product we have just defined is
indeed a representation of the algebra. Moreover, this tensor product of representations
is associative, because is coassociative, that is, satisfies
(1 ) = ( 1).
(43)
4 Note that these irreps are by no means all the irreps at q = ei2/(k+2) ; in fact, there are more irreps of
dimensions 1, . . . , k + 1 (how many more depends on the precise definition of Uq (sl(2)), see, e.g., [29,30])
and there is a family of inequivalent representations of dimension k + 2, parametrised by a complex number z.
However, these representations will not concern us here.
247
A tensor product of -representations will be itself a -representation with respect to the

standard inner product on the tensor product space if star and coproduct commute, that is,
if one has

( )(x) = (x) .
(44)
If this does not hold, it may still be possible to choose an inner product other than
the standard inner product with respect to which the tensor product representation is a
-representation. For Uq (sl(2)), star and coproduct commute only when q is real and
positive. Thus, for q R+ , the tensor product of two unitary representations is unitary
and decomposes orthogonally with respect to the standard inner product. When q is not
real, star and coproduct do not commute and hence, the tensor product of two unitary
representations is not necessarily unitary with respect to the standard inner product. Hence,
tensor product decompositions should be expected not to be orthogonal with respect to this
inner product, but they may still be orthogonal with respect to a different inner product (see
further on). 5
Tensor product decompositions and even ClebschGordan coefficients for tensor
products of unitary representations of Uq (sl(2)) may be calculated similarly as for
U (sl(2)). The highest weight state |jj of each the irreducible representations in the tensor
product may be found by solving the equations L+ |jj = 0 and (L )2j |jj = 0. The other
states are produced by the action of L on the highest weight states. In the calculations,
the following formula for the coproduct of (L )n is a great help:
n

n
n
n
(L )m q (nm)H/4 (L )nm q mH/4 .
(L ) = (L ) =
(45)
m q
m=0
The
n
q-binomial m q
n
m
:=
q
nq ! :=
in this formula is defined by
nq !
,
mq !n mq !
(46)
mq .
(47)
m=1

When q is not a root of unity, the tensor product representation has the same
decomposition into irreps as for q = 1, i.e.,

+

(48)
=| |
where increases in steps of 2.

5 Note that we do have ( )(x) = ((x)) for |q| = 1. Here, is the operator which flips the factors
of the tensor product. Thus, one might be tempted to take ( ) as the -structure on A A when |q| = 1.
This is done, for example, in [31]. In our context, this is not the right course to pursue, since tensor product
representations will still not be -representations with the new on A A.
248
Explicit ClebschGordan coefficients may be calculated for any tensor product of

irreducibles, using, for example, (45). One writes
j1 j2 j
|j, m =
(49)
|j1 , m1 |j2 , m2 ,
m1 m2 m q
m ,m
1
for the vector with H -eigenvalue 2m in the irrep 2j in the decomposition of the tensor
product 2j1 2j2 . The above formula is only meant to introduce the notation for the
q-ClebschGordan coefficients. Several general formulae for these coefficients are proved
in [32,33] and collected in [34]. We will not give these explicit (complicated) formulae
here, but we do give the coefficients for the case j2 = 12 , as an illustration and because this
case is of interest to us later. For j > 0, one has

j + 1 , j + 1 p = q p/4 2j + 1 pq |j, j p 1 , 1
2
2
2 2
2j + 1q

pq
(p2j 1)/4
+q
|j, j p + 1 12 , 12 ,
2j + 1q

j 1 , j 1 p = q (p2j )/4 p + 1q |j, j p 1 1 , 1
2
2
2
2
2j + 1q

2j pq
(50)
q (p+1)/4
|j, j p 12 , 12 .
2j + 1q
The decomposition for j = 0 should be obvious (it is the same as for U (sl(2))). In making
a decomposition such as the one above, one has the freedom to multiply all the states in
each summand irrep by a constant phase factor. Here, the phases are chosen in such a way
that, when q goes to one, the coefficients reduce to the usual ClebschGordan coefficients
for U (sl(2)). One may check (for example, using (38)) that, when q is a real number, the
tensor product vectors on the right hand side are orthonormal, as they should be, because
the tensor product representation is unitary in this case. However, as expected, if q is not
real, then the vectors above are no longer normalised or orthogonal with respect to the
standard inner product on the tensor product space. This can be remedied by choosing
the inner product on the tensor product space which is defined by the fact that it makes the
states above orthonormal. With respect to this q-deformed inner product, the tensor product
representation is a unitary representation of Uq (sl(2)) whenever all the irreps that appear
in its decomposition are unitary. This happens when q R (in which case the new inner
product coincides with the old one) and also when |q| = ei with R, || 2j2+2 . For
general tensor products of two irreps, we have a similar situation. Here, too, we may define
an inner product by requiring that the different states given by (49) are orthonormal and
this inner product will make the tensor product representation into a -representation when
the irreps into which it decomposes are -representations. From this point, one may go
on and define inner products on N -fold tensor products of irreducibles by requiring that a
complete set of states obtained by iterative use of (49) is orthonormal. The set of states that
is declared orthonormal now depends on the order in which one takes the tensor product
249
of the irreducibles, so (V 1 V 2 ) V 3 and V 1 (V 2 V 3 ) may in principle

end up with a different inner product. We will say more about this issue in Sections 3.2.7
and 3.2.8.
The orthogonality of the tensor product decomposition which holds when q is real is
reflected in the following identity for the q-ClebschGordan coefficients:
j1 j2 j j1 j2 j
(51)
= j,j m,m
m1 m2 m q m1 m2 m q
m ,m
1
Note that, although tensor product decomposition is orthogonal with respect to the standard
inner product only when q is real and positive, this equation holds by analytic continuation
for all q where the summands are not singular.
Another useful identity (taken from [34]), which relates the coefficients for the tensor
product 2j1 2j2 with those for the opposite tensor product is:

j1 j2 j3
j1 j3
j1 +j2 j j2
= (1)
.
(52)
m1 m2 m3 q
m2 m1 m3 q 1
In particular, this allows one to write down the ClebschGordan coefficients for 1 2j
using (50).
3.2.4. The root of unity case: truncated tensor products
In the previous section, we described the tensor product of representations of Uq (sl(2))
for the case that q is not a root of unity. In that case, many of the usual properties of
tensor products at q = 1 could be recovered. For example, the tensor product of two irreps
could be decomposed into a direct sum of irreps (see (48)). Also, this decomposition is
orthogonal with respect to a suitable deformation of the standard inner product.
When q is a root of unity, say q = ei2/(k+2), the situation is quite different. In this
case, tensor products of two irreps will not split into a direct sum of irreps, but will contain
indecomposable summands. This is not in itself surprising, because the representations
with > k + 1 that would occur in the usual decomposition (48) become indecomposable
for q = ei2/(k+2) . However, what really happens is a bit more complicated. As an example,
let us look at the decomposition of the tensor product of the spin 12 and the spin k+1
2
module. As usual, the tensor product space may be decomposed into eigenspaces of the
operator H . These eigenspaces will be one-dimensional for the extremal eigenvalues H =
k + 1 and H = (k + 1) and two-dimensional for the other eigenvalues. If q were not a
root of unity, then we would have two highest weight states in the tensor product module.
k+1 1 1
k1 k1
The H = k + 1 state | k+1
2 , 2 | 2 , 2 and the H = k 1 state | 2 , 2 given in (50).
i2/(k+2)
, the coefficients of this second state diverge, but if we multiply the state
At q = e
by k + 2q , then this no longer happens and we still have two good highest weight states.
However, we have a third candidate highest weight state, which is the H = k 1 state one
k+1 1
1
gets when one lets (L+ )k+1 act on the lowest weight state | k+1
2 , 2 | 2 , 2 (remember
(L+ )k+2 gives zero for this value of q). This new highest weight state is just proportional
k1
to the state | k+1
2 , 2 given in (50). Comparing this state with the other highest weight
state at H = k 1, we see that although they would be linearly independent for any
250
Fig. 4. Diagram of an indecomposable representation as it would occur in the tensor product of two
Uq (sl(2))-irreps at q = e2i/(k+2) . The dots represent the basis states in the module, the arrows
and indicate the action of L+ and L , respectively. The split arrows are meant to indicate that
the descendants of the state | are mapped onto linear combinations of descendants of | and
(L+ )k+1 |l.
arbitrary q, they are actually proportional to each other for q = e2i/(k+2) . It follows that
the irreducible spin k1
2 -module has become a submodule of the module generated by the
highest weight state at H = k + 1. Also, since we have only one highest weight state in
the H = k 1 eigenspace and since this space is two-dimensional, there must also be a
non-highest weight state in this module. The two-dimensional H -eigenspaces of the tensor
product module will then be spanned by a descendant of the highest weight state at H =
k 1 and a descendant of the non-highest weight state at H = k 1. We see thus that,
at q = e2i/(k+2), the modules k+2 and k have disappeared from the decomposition of
1 k+1 and in stead there is one indecomposable module, which has the module k as
an irreducible submodule.
This general picture extends to all tensor products of irreps; in general, all the
modules with > k and all the corresponding modules 2k will disappear from
the decomposition (48) and in stead, there will be indecomposable modules with the
modules 2k as irreducible submodule. The structure of these indecomposable modules
is analogous to the structure of the module we described above and is illustrated in Fig. 4.
For more detail on tensor product decomposition when q is a root of unity, one can consult
for example [29,30,35].
Clearly, the indecomposable representations which occur in the tensor products are nonunitarisable; this follows from the indecomposability, but one can also see easily that any
inner product that would make these representations unitary would give the states in the
irreducible submodule zero norm. In fact, one can see this happen explicitly for the limit
of q e2i/(k+2) of the q-deformed inner product we defined at the end of the previous
section.
In the present context, the states in the indecomposable representations are considered
which is the old
non-physical. Thus, we need to define a new tensor product
tensor product with the indecomposable modules projected out. However, the tensor
product defined this way would not be associative, since we would have, for example,
k)
( k
k+1 ) = 0 for odd k. (For even k, we get
k+1 = 2 k+1 and 1
( 1
similar problems.) Also, the fusion rule k+1 = 0 ( even) is clearly unphysical;
after adding a particle in the representation k+1 we would be left without a Hilbert space!
These problems can be solved both at once by projecting out not just the indecomposable
modules, but also any modules of type k+1 that may occur. The resulting tensor product
is called the truncated tensor product. Similarly as in Section 3.2.3, one may define an
251
inner product on any truncated tensor product of irreducible Uq (sl(2))-modules, which

makes the tensor product decomposition orthogonal. Moreover, since the irreps into which
a truncated tensor product factors are all -representations, this inner product makes
the truncated tensor product representation into a -representation. The truncated tensor
product decomposition at q = ei2/(k+2) is given by the following formula, which is
k chiral primaries:
identical to the formula (2) for the fusion rules of sl(2)
,2k }
min{+

(53)
=| |
From this formula, one may check easily that the truncated tensor product is indeed as 3 and 1
2 )
( 2
3 )
sociative, that is, the tensor product modules ( 1
are isomorphic. Note however that these two modules are different subspaces of the ordinary tensor product, so we might say that the truncated tensor product is associative at
the level of Uq (sl(2))-modules, but not associative at the level of states. This might seem
like a problem at first sight, because we want to have a unique three-particle Hilbert space,
but this problem disappears if we can find a canonical Uq (sl(2))-isomorphism between the
two three-particle spaces which preserves the inner product. We will say more about this
in Sections (3.2.7) and (3.2.8).
As an illustration, let us take a closer look at the truncated tensor product of the
2-dimensional irrep 1 with the unitary irreps 0 , 1 , . . . , k . For this case, the truncated
tensor product decomposition is given by
1 = 1,
0
1 = +1 1

=

k
k1
( {1, . . . , k 1}),
(54)
As one can see, the only difference with the ordinary tensor product occurs in the last line.
The decomposition on the level of states can be read off from (50). Using this formula, we
can also give an example of the non-associativity at the level of states that we were talking
1 and
1)
about: At k = 1 (or q = e2i/3 ), the truncated tensor products V1 = ( 1
1
1
1
1
(
) are both isomorphic to as Uq (sl(2))-modules, but any state in V1
V2 =
may be written as

1 1

1 1/4 1 1 1 1
1/4 1 1 1
1
q
1 2 , 2 + 2 12 , 12 ,
2,2 2, 2 q
2, 2 2,2
2q
(55)
while any state in V2 may be written as

1 1 1
1 2 , 2 + 2 12 , 12 q 1/4 12 , 12 12 , 12 q 1/4 12 , 12 12 , 12 .
2q
(56)
From this, we see that a vector in V1 can only equal a vector in V2 if it is zero. Hence, V1
and V2 are different subspaces of 1 1 1 .
Before ending this section, let us write down two useful identities for truncated tensor
k that we discussed
decomposition which are related to the external automorphism of sl(2)
252
in relation to the field identifications (8). If we define

:= k ,
(57)
then we have
=

,2k }
min{+
,
=| |
,2k }
min{+
.
(58)
=| |
Here, we have written in stead of to avoid notational overload. These identities tell
us that the truncated fusion rules of Uq (sl(2)) do not allow us to make a distinction between
a particle that carries the representation and a particle that carries the representation .
3.2.5. Counit, antipode and quantum trace
Every quantum group A is required to have a one-dimensional representation 9, called
the counit, which represents the vacuum (or A-neutral) sector of the theory. The counit has
to satisfy
(9 id) = (id 9)() = id,
(59)
where id is the identity map on A. It follows that 9 = 9 = for any

representation of the quantum group, so the vacuum (or an A-neutral particle) has the
fusion properties that one would expect. For Uq (sl(2)), we have already seen the counit; it
is just the representation 0 . Explicitly, we have

9(1) = 1,
9(H ) = 0,
9 L = 0.
(60)
The counit of a quantum group is an analogue of the trivial representation of a group and
hence, we will say that a state |s in a representation of the quantum group A transforms
trivially if (a)|s = 9(a)|s for all a A.
Next to the counit, A also required to have a linear algebra antihomomorphism
S : A A which satisfies
(S id)(a) = (id S)()(a) = 9(a)1,
(61)
where, : A A A is just the multiplication of A. S is called the antipode and S(a)

should be thought of as a quantum group analogue of the inverse of a. If we are given a
representation of A, then the antipode makes it possible to define the representation
conjugate to by the formula

t
(a)
= S(a) .
(62)
The fact that S is an antihomomorphism then makes sure that is a representation of A,
while the properties (61) ensure that the tensor product representations and
will contain the trivial representation 9 in their decomposition. Thus, a particle which
253
carries the representation and its antiparticle, which carries the representation may
indeed annihilate.
The antipode also allows us to define the action of A on the space of linear operators on
an A-module V by

|v = a k OS
a k |v.
aO
(63)
1
2
Here, |v is any state in the module V and the a1k and a2k are the left and right components
of the coproduct of a, i.e.,

(a) =
(64)
a1k a2k .
k
Again using the fact that S is an antihomomorphism, one can see that (63) defines a
representation of A, while using the property (61), one can see that A acts trivially on
operators that commute with the action of A on V .
For Uq (sl(2)), the antipode is given by

S(1) = 1,
S(H ) = H,
S L = q 1/4 L .
(65)
The representations of Uq (sl(2)) are clearly all isomorphic to their conjugates (one may
see this, for example, from the decomposition (48) without even using the explicit form of
is given by
the antipode) and the action of Uq (sl(2)) on an operator O

= H, O
,
= L Oq
H/4 q (H 1)/4OL
,
H O
L O
(66)
which reduces to the usual commutator for q 1.
One can define a kind of trace on operators, which has the property that it transforms
trivially under Uq (sl(2)) when the operator is transformed. For q = 1, the ordinary trace
= 0 = 9(a)Tr(O)
for all a sl(2) and for arbitrary O.

has this property, since Tr([a, O])
However, for q %= 1, we have to use a modified trace to get this property. This trace is
usually called the quantum trace and we will denote it Trq . Of course, the quantum trace is
supposed to preserve some nice properties of the ordinary trace. Most importantly, the trace
of a tensor product of operators should be the product of the traces of the tensor factors.
A quantum trace with this property can be defined for a large class of quantum groups (see
cf. [29]). For Uq (sl(2)), it is given by

= Tr q H/2 O
.
Trq O
(67)
) = 9(a) Trq (O).
The fact that Trq ( O
1 O
2 ) =
One may verify readily that Trq (a O
H/2
H/2
H/2

)=q
q
.
Trq (O1 ) Trq ( O2 ) follows from the comultiplication (q
Using the quantum trace, one may now define the quantum dimension dimq () of
a representation of the quantum group as the quantum trace of the unit operator in
this representation. For the representations of Uq (sl(2)), this yields dimq ( ) =
+ 1q = dim( )q . In particular, the quantum dimension of k+1 is zero. The
quantum dimensions of all the indecomposable modules of dimension 2k + 4 that appeared
in the (untruncated) tensor products of the are also zero, since these modules were a
(non-direct) sum of two modules of dimensions k + 2 d and k + 2 + d and we have
254
k + 2 dq + k + 2 + dq = dq + dq = 0. Since the quantum dimensions of the

modules 1 , . . . , k , are non-zero, we see that we might also have defined the truncated
tensor product of two modules in this set as the ordinary tensor product with the modules
of quantum dimension zero projected out. With this definition, the truncated tensor product
is automatically associative and the module k+1 does not need separate treatment.
Quantum traces may also be used to construct knot invariants (see, for example, [29,36]
and references therein). For Uq (sl(2)), one of the knot invariants which can be constructed
this way is the famous Jones polynomial.
3.2.6. Braiding: the R-matrix
Suppose that we have two particles that carry the Uq (sl(2)) representation . The total
internal state of the system can then be represented by a state |s in the (possibly truncated)
. Now if we exchange the particles, how does the state of the
tensor product
system change? In the q = 1 case, we can describe the exchange simply by exchanging
the tensor factors in the state |s. Let us call this exchange of the factors in the tensor
product . An exchange of two adjacent particles in a system of N identical particles may
then be described by the action of on the corresponding factors of the N -fold tensor
product that describes the system. For example, in a 4-particle system, the exchange of
the second and third particles is affected by the operation 1 1. Such exchanges
generate a representation of the permutation group SN . Also, they commute with the
action of the algebra at q = 1. This can be seen from the formulae for the coproduct at
q = 1; the coproduct of any element of the algebra is invariant under the exchange of the
tensor factors. We say that the coproduct is cocommutative. Because the algebra action
and the permutation group action commute, the N -particle state space of the system can
be decomposed into representations of U (sl(2)) SN .
When q %= 1, the situation is different. The coproduct is now no longer cocommutative
and hence, exchanging the order in the tensor product no longer commutes with the
Uq (sl(2)) action. However, this can be remedied with the use of a universal R-matrix.
This is an invertible element of Uq (sl(2)) Uq (sl(2)) which acts before in the
appropriate representation. For example, in a system of three particles, all of which carry
the representation of Uq (sl(2)), the exchange of the first and second particles will now
be affected by ( (R)) 1. Note that, in any tensor product of two representations,
the universal R-matrix does indeed act as a matrix, but the matrix in question depends on
the representations. The R-matrix is required to have the following properties:
op R = R,
( 1)R = R13 R23 ,
(1 )R = R13 R12 .
(68)
Here, op is the comultiplication, followed by an exchange of the tensor factors in A A

and Rij is an abbreviation for the action of R on the factors i and j of A3 , so, for example,
R12 = R 1. The first of the properties above ensures that the exchanges affected by R
in the tensor product representations commute with the action of the quantum group, as we
wanted. The second and third property make sure that braiding of two particles around a
third one and then fusing them together gives the same result as fusing the two particles
first and then braiding the result around the third one. Using the above properties of the
255
universal R-matrix, one may also prove the following equality, which is called the quantum
YangBaxter equation
R12 R13 R23 = R23 R13 R12 .
(69)
This equation implies that in any representation, we have

( R 1)(1 R)( R 1) = (1 R)( R 1)(1 R).
(70)
It follows that, for a system of N identical particles that carry a representation of the
quantum group, the exchanges of adjacent particles, as performed using R, satisfy the
relations (29). Hence, since R is invertible, they generate a representation of the braid
group on N strands, BN . Since the exchanges commute with the action of the quantum
group A, the system thus carries a representation of A BN .
When the particles do not all carry the same quantum group representation (and are
hence not identical), then the R-matrix no longer gives us a representation of the braid
group on the Hilbert space of the system, simply because the exchanges now act between
different vector spaces; the flip operator sends V 1 V 2 into V 2 V 1 . This is
not a problem, because exchanges of non-identical particles are not symmetries of the
system. What we do still get from the R-matrix is a representation of a so-called coloured
braid group, which consists of the braids for which the final position of any particle is the
original position of a particle of the same kind (or colour). These coloured braidings
will still commute with the quantum group action. One should note that the colouring
restriction leaves plenty of room for non-trivial and even non-Abelian monodromies
between distinguishable particles. All the braiding transformations in coloured braid
groups may still be generated by elementary exchanges of adjacent particles, although
some of these will no longer have any physical meaning and should be called halfmonodromies rather than braidings. We will be a bit sloppy about this in the rest of this
article, but we assume that this will not cause confusion.
Before going on, let us note that one could multiply the R-matrix by a constant
phase factor. The matrices ( R)12 , ( R)23 , . . . , ( R)n1,n , which model the exchange
of adjacent particles would then still generate a representation of the braid group and
commute with the quantum group. The second and third condition in (68), which express
compatibility of fusion and braiding, may also be satisfied if we give the coproduct the
same phase factor as the R-matrix. The tensor product representation defined in (42)
will then in general no longer be a representation of the quantum group, but it will still be a
projective representation, which is really what we should expect in a quantum mechanical
setting. Thus, we may say that the universal R-matrix determines the braiding of the
particles only up to a constant phase factor in each (clockwise) exchange (obviously, the
counterclockwise exchanges get the inverse phase factor).
The universal R-matrix which fulfills the requirements (68) for Uq (sl(2)) is given by
R=q
H H
4

(1 q 1 )n
n=0
nq !

n nH/4 n
q n(1n)/4 q nH/4 L+
q
L
.
(71)
256
We see that, when q approaches one, only the n = 0 term in (71) contributes and we get
R = 1 1, as expected.
The action of the universal R-matrix on the module V 1 V of the tensor product
representation 1 is given by
R|j1 , m1 |j2 , m2

j1 m1 j2 + m2 j1 + m1 + nq !j2 m2 + nq ! n(1n)
q 4
=
n
n
j1 + m1 q !j2 m2 q !
q
q
n0

n
1
q 2 (m2 nm1 n+2m1 m2 ) 1 q 1 |j1 , m1 + n|j2 , m2 n,
(72)
where the sum extends over all n for which the kets on the right hand side are well defined.
Using this formula, one may easily find the exchange matrix R in any tensor product
module. For example, in the tensor product 1 1 of two two-dimensional modules, we
have
1/2
0
0
0
q
0
0
1
0
.
R 1,1 := q 1/4
(73)
1/2
1/2
0
0
1 q q
0
q 1/2
Note that, if q %= 1, this is not a unitary matrix, which is not good if it is supposed
to represent a symmetry transformation on a physical system. Still, we could hope to
make R unitary by choosing a suitable inner product on the module V 1 V 1 . This will
certainly not succeed unless |q| = 1. To see this, note that the eigenvalues of R 1,1 are q 1/4
(with multiplicity 3) and q 3/4 and these will only have norm 1 (as required for the
eigenvalues of a unitary transformation) if q does. However, if |q| = 1, then R 1,1 is indeed
unitary with respect to the q-deformed inner product which we defined below formula (50)
and which makes the tensor product decomposition orthogonal. Of course, the reason that
this works is the fact that the action of R commutes with the quantum group action, which
means that the eigenspaces of R are submodules of the tensor product module. We will say
more about inner products and unitarity in Section 3.2.7.
In the above, we have always implicitly assumed that q was not a root of unity. If
it is, then the whole story is more complicated, because of the truncation of the tensor
product which has to be taken into account for these values of q. The R-matrix (71)
still describes the braiding of two particles, 6 but if we go to three or more particles,
then we can get problems. For example, three particles in the representation may
V and
V )
be described by a state in the truncated tensor product space (V
)(R) 1, which
we can exchange the two leftmost particles by means of (
V )
V , as it should. However, if we want to exchange the
gives us a state in (V
V if we just apply
V )
two rightmost particles, then we can leave the space (V
6 Note that this R-matrix is not well defined if q = ei2/(k+2) , since the q-factorial n ! which appears in the
q
((L+ )n ) ((L )n ) term becomes zero for n k + 2. This problem is usually resolved by adding the relations
(L+ )k+2 = (L )k+2 = 0 to the algebra for this value of q. To us, this subtlety is not very important, since these
relations already hold in the unitary representations we are interested in.
257
)(R). One may see this explicitly in the example we gave in formula (55);
1 (
exchanging the last two particles in this state, we get a state which can clearly not be
1)
V 1 . If we use the other
written in the same form and hence does not belong to (V 1 V
(V
V )), then we can exchange
bracketing of the truncated tensor product (i.e., V
the last two particles in the expected way, but then the problem occurs in the exchange of
the first two. In this way, we can always expect problems when we try to exchange two
particles over a bracket. Thus, we will not get a representation of the braid group on the
truncated tensor product, unless we modify the way in which we exchange particles. We
will explain the modification that is needed in some detail in Section 3.2.8.
3.2.7. q-6j -symbols and their properties
If we take a tensor product of three Uq (sl(2)) modules 1 , 2 and 3 , then there
are two different ways to decompose this tensor product into irreducibles. We may either

first decompose the product 1 2 and then the resulting modules 3 , or we

may first decompose the product 2 3 and then the resulting modules 1 .
These two procedures yield two different natural bases for the vector space V 1 V 2
V 3 . In each case, the basis vectors are labelled by their H -eigenvalue, the label of their
overall fusion channel and the label of their intermediate fusion channel (which is the
representation into which 1 and 2 fuse in the first case and the representation into
j1 ,j2 ,j3
which 2 and 3 fuse in the second case). Let us call the vectors in the first basis ej12
,j,m
j ,j ,j
and the vectors in the second basis fj231 ,j2 ,m3 . Here, j1 , j2 and j3 correspond to 1 , 2 and
3 , m and m give the H -eigenvalues, j and j give the overall fusion channels and j12
j1 ,j2 ,j3
j1 ,j2 ,j3
and j23 represent the intermediate fusion channels. The vectors ej12
,j,m and fj23 ,j ,m may
be written in terms of the standard (product) basis for the tensor product by means of the
ClebschGordan coefficients. We have
j1 j2 j12 j12 j3 j
j1 ,j2 ,j3
=
ej12
,j,m
m1 m2 m12 q m12 m3 m q
m ,m ,m
1
j ,j ,j
2 3
=
fj231 ,j,m
m1 ,m2 ,m3
|j1 , m1 |j2 , m2 |j3 , m3 ,

j2 j3 j23
j1 j23
m2 m3 m23 q m1 m23
j
m
(74)
|j1 , m1 |j2 , m2 |j3 , m3 ,

where m12 = m1 + m2 and m23 = m2 + m3 . The vectors in the e-basis may also be
expressed in terms of the f -basis vectors and this expression takes the following form:

j1 j2 j12
j1 ,j2 ,j3
j ,j ,j

ej12
(75)
=
fj231 ,j2 ,m3 .

jj mm
,j,m
j
j
j
3
23
j ,j ,m
23
The coefficients represented by the curly brackets are now called the 6j -symbols of
Uq (sl(2)). By definition, these q-6j -symbols must be equal to the 6j -symbols for SU(2)
when q equals one. The 6j -symbol in the formula above will clearly be zero unless
the representation j12 occurs in the tensor product of the representations j1 and j2 ,
the representation j occurs in the tensor product of the representations j12 and j3 , etc.
258
It follows that the 6j -symbol will be zero unless its arguments satisfy the following
requirements:
|j1 j2 | j12 j1 + j2 ,
j1 + j2 + j12 Z,
|j2 j3 | j23 j2 + j3 ,
j2 + j3 + j23 Z,
|j12 j3 | j j12 + j3 ,
j12 + j3 + j Z,
|j1 j23 | j j1 + j23 ,
j1 + j23 + j Z.
(76)
If these requirements are met, then the 6j -symbol may be written in terms of Clebsch
Gordan coefficients; using (74) and the relations (51), one easily finds that

j1 j2 j12
j3
j j23
j1
m1
m ,m
2
j2
m2
j12
m12

q
j12
m12
j1
m1
j3
m3
j23
m23
j
m
j
m

q
j2
m1
j3
m2
j23
m23

q
(77)
From this formula, one may obtain explicit formulae for the 6j -symbols. We will not do
the (long) computations here, but just give one of the possible explicit answers, as given
in [34] (see also [37]).

j1 j2 j12
j3
=
j j23

2j12 + 1q 2j23 + 1q (j1 , j2 , j12 )(j12, j3 , j )(j2 , j3 , j23 )(j1 , j23 , j )

z
(1)z z + 1q !
z j1 j2 j12 q !z j12 j3 j q !z j2 j3 j23 q !z j1 j23 j q !
1
, (78)
j1 + j2 + j3 + j zq !j1 + j12 + j3 + j23 zq !j2 + j12 + j + j23 zq !
where
(a, b, c) :=
a + b + cq !a b + cq !a + b cq !

.
a + b + c + 1q !
(79)
The sum in (78) is taken over all z for which all the q-factorials in the summands are
well-defined.
The q-6j -symbols are invariant under many symmetries (described in [34,37]) which
are analogues of the symmetries of the 6j -symbols of SU(2) (see [38]). For us, the most
important of these are the so-called classical symmetries. These symmetries can be treated
slightly more elegantly if one works with the q-Racah coefficients in stead of the q-6j symbols. The Racah coefficients are just the 6j -symbols with a different normalisation;
they are given by the formula for the 6j -symbols above with the first square root factor left
out. Invariance under the classical symmetries means that the Racah coefficients remain
unchanged under permutations of the columns and under exchanging the upper and lower
entry in two columns simultaneously. In effect, this means that we have the following
259
identities for the 6j -symbols

j1
j2
j12
j3
j23
j1
j2
j12
j3
j23

=

=
j2
j1
j12
j3
j23
j1
j23
j3
j2
j12

=
,
2j12 + 1q 2j23 + 1q

2j2 + 1q 2j + 1q
j1
j12
j2
j3
j23
,
(80)
and all the identities generated by these. The other symmetries of the 6j -symbols are
analogues of the Regge and reflection symmetries.
When q R+ , the bases for the three-fold tensor product given in (74) are orthonormal
and hence the basis transformation between these bases is unitary. As a consequence, the
6j -symbols satisfy the following orthogonality relation (see cf. [34])
j1 j2 j12 j1 j2 j12
= j23 j .
(81)

23
j3 j4 j23
j3 j4 j23
j
12
Here, we have used the fact that the 6j -symbols are real for q R+ . The equation above
indeed tells us that the matrix for the coordinate transformation between the e and f
bases for the charge j subspace of V j1 V j2 V j3 is real-orthogonal. When q is not
a positive real number, the above relation for the 6j -symbols remains valid by analytic
continuation, as long as the summands are not singular, but it does not tell us that the
matrix for the coordinate transformation we mentioned is orthogonal unless all the 6j symbols that appear are real. For |q| = 1, these 6j -symbols will certainly be real as long
as | arg(q)| is small enough to make sure that all the q-numbers that appear in these 6j symbols are positive. This will be the case (cf. formula (78)) when

2
2
arg(q) < min
,
,
j12 j1 + j2 + j12 + 1 j2 + j3 + j23 + 1
2
2
,
,
(82)
j12 + j3 + j4 + 1 j1 + j23 + j4 + 1
where the minimum is over all j12 that appear in (81). Hence we see that also for |q| = 1,
| arg(q)| small enough, the matrix of the coordinate transformation from the e to the f
basis of the charge j subspace of the space V 2j1 V 2j2 V 2j3 is real-orthogonal.
It follows by iterative use of this fact that, for |q| = 1, the q-deformed inner product
on a given N -fold tensor product of irreducible Uq (sl(2))-modules that we defined in
Section 3.2.3 can be made independent of the choice of the order in which the tensoring
of the irreducibles is performed by taking | arg(q)| small enough. For the case N = 3,
the different sets of vectors that are declared orthonormal are just the e and the f -vectors
and the fact that the matrix which relates these is unitary shows that declaring one set to
be orthonormal is equivalent to declaring the other set to be orthonormal. On the other
hand, given any fixed value of | arg(q)|, it will not be difficult to construct tensor product
representations in which the inner product does depend on the order of the tensoring. In
fact, we can expect this to happen as soon as the decomposition of the tensor product
module contains non-unitary irreps.
260
3.2.8. Truncated 6j -symbols; the coassociator

When q is a root of unity (q = ei/(k+2) ), we can define truncated 6j -symbols, related
to the truncated tensor product. For these to be non-zero, the conditions (76) have to be
changed in such a way that they require that j12 be not just in the tensor product, but
even in the truncated tensor product of j1 and j2 , etc. This means that the upper bounds
j1 + j2 , . . . , j1 + j23 , in (76) are sharpened to min{j1 + j2 , k j1 j2 }, . . . , min{j1 +
j23 , k j1 j23 }. When the arguments satisfy these sharpened conditions, the truncated
6j -symbols are still given by the formula (78). The truncated 6j -symbols defined in
this way give a canonical isomorphism between the truncated tensor product modules
V 2j2 )
(V 2j2
V 2j3 ). Moreover, this isomorphism is preserves
V 2j3 and V 2j1
(V 2j1
the q-deformed inner products on the truncated tensor product spaces. To see this,
note first of all that the truncated 6j -symbols are real (this follows easily from (78)).
Also, it is known that the truncated 6j -symbols satisfy an analogue of the orthogonality
relations (81). We have
j1 j2 j12 j1 j2 j12
= j j ,
(83)

23 23
j3 j4 j23
j3 j4 j23
j
12
where the sum is now restricted to the j12 that are allowed by the truncated tensor
V 2j2 )
V 2j3 and
product. It follows that the matrix of the mapping between (V 2j1
2j
2j
2j
1
2
3
(V
V ) is real-orthogonal and hence that the mapping preserves the inner
V
product. The proof of the relations (83) uses the usual orthogonality relations and the
formula (90) in Section 3.2.9.
Using the isomorphism given by the truncated 6j -symbols, we can identify the
V 2j2 )
(V 2j2
V 2j3 ), so that we have a well-defined
V 2j3 and V 2j1
spaces (V 2j1
three-particle Hilbert space. The isomorphism may also be used to define braiding
transformations on truncated tensor products. Recall from the end of Section 3.2.6 that we
could use the R-matrix to define braiding of two particles, but that there were difficulties if
we wanted to braid particles over a single bracket in a multi-particle Hilbert space. These
difficulties can now be resolved using the mappings given by the truncated 6j -symbols.
For example, if we want to exchange the two rightmost particles in the representation
2j , then we can first use the 6j -symbols to map the representation space
2j )
( 2j
( 2j
2j ), then use the R-matrix to exchange the particles and finally
onto that for 2j
use the inverse of the mapping given by the 6j -symbols to get back to the representation
2j )
2j . Similarly, any braiding in a multiple truncated tensor product
space of ( 2j
may now be achieved by using the 6j -symbols to move the brackets around before and
after the actual braiding.
The whole procedure of truncating the tensor product so that it is no longer associative
and then defining braiding by identification mappings may be elegantly formalised and
brought to the level of the algebra, at the cost of making the connection with non-truncated
Uq (sl(2)) somewhat less apparent. This has been done in [31] and the resulting structure
is called a weak quasitriangular quasi-Hopf algebra, or a weak quasi-quantum group. Let
us call this A. Some important features of the resulting picture are the following. The
coproduct is modified in such a way that it has the truncation built in. As a result, one no
261
Fig. 5. This diagram shows two ways of going from one bracketing of a fourfold tensor product to
another. The arrows denote the canonical isomorphisms given by the coassociator (or the truncated
6j -symbols). The diagram will commute if the condition (85) on the coassociator is satisfied.
longer has (1) = 1 1 and one also loses coassociativity. A so-called coassociator is
!
introduced to compensate for this loss. This coassociator is an element = k k1 k2
k3 of A3 which is not invertible in A, but which has a quasi-inverse called 1 which is
the inverse in all the representations that one is interested in and which has the following
important property for all a A:
( 1)(a) = (1 )(a).
(84)
2j3 and 2j1

2j2 )
( 2j2
2j3 ) are
This ensures that the representations ( 2j1
2j
2j
2j
1
2
3

(). Of course, this
isomorphic, with the isomorphism given by
isomorphism is just the one given by the truncated 6j -symbols. Clearly, one would like
to be able to go from one bracketing of a multiple tensor product to another, using , in
such a way that it does not matter which individual steps are taken on the way. This will be
the case if the diagram in Fig. 5 commutes.
To make this diagram commute, we need to impose the following condition on the
coassociator [39]:
(1 1 )()( 1 1)() = (1 )(1 1)()( 1).
(85)
In terms of 6j -symbols, this condition becomes

j1 j2 j12
j12 j3 j123
j4
=
j j34
j34 j j234
j1 j2 j12 j1 j23
j23
j3
j123
j23
j4
j123
j234
j2
j3
j23
j4
j234
j34
(86)
This condition will clearly be satisfied for non-truncated 6j -symbols, since the sides of the
equation just correspond to two ways of doing the same coordinate transformation in that
case. For non-truncated 6j -symbols the coordinate transformations change to mappings
that really do something, but the equation above still holds.
When there is a non-trivial coassociator, the last two conditions in (68), which
guaranteed the compatibility of fusion and braiding, change to
1
R23 123 ,
( 1)(R) = 312 R13 132
1
1
(1 )(R) = 231
R13 231R12 123
, (87)
262
and these in turn imply the following quasi-YangBaxter equation [39], which is the
analogue of (70):
1
1
R12 132R13 132
R23 123 = 321 R23 231
R13 213 R12 .
(88)
This relation ensures that the recipe that we gave for performing braidings does indeed give
a representation of the braid group.
3.2.9. Symmetries of the truncated 6j -symbols
The truncated 6j -symbols still satisfy the classical, Regge and reflection symmetries of
the 6j -symbols, but as it turns out, we also have two sets of extra symmetries, which do
not have an analogue at q = 1.
The first of these two sets was mentioned already in [34] and it is important in the
proof of the truncated orthogonality relations (83). If we define j := k + 1 j, then the
symmetries in this set can all be generated from the untruncated symmetries (such as the
classical symmetries (80)) and the identity

j1 j2 j12
j2 +j23 j j12 +2j1 +1 j1 j2 j12
(89)
= (1)
.
j3 j j23
j3 j j23
In particular, we get from this that

j1
j1 j2 j12
j2 +j3 j1 j +2j12 +1
= i(1)
j3 j j23
j3
j2
j12
j23
(90)
To prove the truncated orthogonality relations (83), we may now start from the untruncated
orthogonality relations (81). We may split the sum in (81) into three parts as in
min{j1 +j
2 ,j3 +j4 }

min{j1 +j2 ,j3 +j4

,kj1 j2 ,kj3 j4 }
j12 =max{|j1 j2 |,|j3 j4 |}
j12 =max{|j1 j2 |,|j3 j4 |}
kj
1 j2

min{j1 +j2 ,j3 +j4 ,kj1 j2 ,kj3 j4 }+1
min{j1 +j
2 ,j3 +j4 }

kj1 j2 +1
(91)
Now if min{j1 + j2 , j3 + j4 , k j1 j2 , k j3 j4 } equals j1 + j2 or j3 + j4, then all the
6j -symbols in the last two summations are zero, because their arguments do not satisfy the
conditions (76). If min{j1 + j2 , j3 + j4 , k j1 j2 , k j3 j4 } equals k j1 j2 , then
the second summation on the right hand side is empty and the third is zero because the j12
and j12 terms cancel each other using (90) (if there is a middle term in the summation then
this also vanishes using (90)). Finally, if min{j1 + j2 , j3 + j4 , k j1 j2 , k j3 j4 }
equals k j3 j4 , then one can use the explicit formula (78) for the 6j -symbols to show
that all the terms of the middle summation vanish, while (90) still makes sure that the last
summation vanishes because of pairwise cancellation of terms. In any case, the summation
on the left, which is the summation in (81), equals the first summation on the right, which is
the summation in (83) and this shows the validity of the truncated orthogonality relations.
263
We will now present the second set of symmetries of the truncated 6j -symbols. These
symmetries are related to the identities (58) for the truncated fusion rules and seem to have
gone unnoticed until now. They are generated by the following identities

j1 j2 j12
j1 j2 j12
= (1)k+j1 +j3 +j12 +j23
j3 j j23
j3 j j23

j1 j2 j12
= (1)k+j12 +j3 +j
(92)
.
j3 j j23
Of course
Here, we have defined j := k2 j, in accordance with the definition of (57) of .
all the identities related to these by the classical, Regge and reflection symmetries are also
symmetries. The above identities may be proved in the following way. First notice that the
replacements of spins are made in a way that is consistent with the truncated tensor product
decomposition. Hence, the arguments of the 6j -symbol on the left satisfy the truncated
version of the conditions (76) exactly if the arguments in the other two 6j -symbols do.
This means we can fill in formula (78) in all three cases. To show that the results are equal,
one needs an identity which holds for q-factorials at q = e2i/(k+2) . We have
k + 1 aq ! =
k + 1q !
.
aq !
(93)
Using this identity, it is easy to show that the -factors are equal for all three 6j -symbols
in (92). For the middle 6j -symbol in (92), we can now see that the sum over z in (78) is
equal to that for the untransformed 6j -symbol by making the substitution z z+k (j1 +
j3 + j12 + j23 ) and using the q-factorial identity above twice. The proof for the rightmost
6j -symbol in (92) is similar, but uses the substitution z z + k + j3 + j + j12 .
3.2.10. Braiding and 6j -symbols
In this section, we will give a systematic description of the braid group representations
that are associated with (truncated) tensor products of Uq (sl(2)) representations. First of
all, let us look at the braiding in a tensor product of two irreps 1 and 2 . We can
decompose this tensor product into irreps as in Eqs. (48) or (53). From these formulae,
we see that any irrep can occur at most once in this decomposition; we say that the
tensor product decomposition is multiplicity-free. It follows from this, using Schurs
lemma, that any map from the tensor product module V 1 V 2 to the tensor product
module V 2 V 1 that commutes with the quantum group action on these modules, is a
constant on each of the irreducible summands of V 1 V 2 . The exchange matrix R
( 1 2 )(R) is such a map. Hence, we can choose bases for V 1 V 2 and
V 2 V 1 such that the action of R is described by a diagonal matrix with respect
to these bases. Of course, the basis vectors in each case are just the basis vectors | 2 , m of
each irreducible summand and the action of R on these will depend on 1 , 2 and
and not on m. Explicitly, one has
1 2 1

1 2 (R)|V = (1) 2 + 2 2 q 2 (c c1 c2 ) ,
(94)
where ci = 2i ( 2i + 1) is the value of the undeformed Casimir for the representation i .
This can be derived from the formula (72) for the elements of the R-matrix, using the
264
formulae for the ClebschGordan coefficients given in [34]. For the case 2 = 2, one may
also check it from (50), using (52). Note that the action of R given above is effected by
a unitary matrix exactly if |q| = 1. Therefore, if we use the q-deformed inner product on
the tensor product space that makes the basis described above orthonormal, then R is a
unitary operation, as we already remarked in a special case below formula (73).
Now let us look at a tensor product of n quantum group irreps 1 , . . . , n . In such a
tensor product, there are a number of natural bases which reflect the structure of the tensor
product. More precisely, there is on such basis for each way in which the tensor product can
be built up by adding subsequent factors. We have already described the situation for three
tensor factors in Section 3.2.7. In this case, there were two of these natural bases and the
transformation that related these was described by the 6j -symbols. In the case of n factors,
we will choose to work with the natural basis one gets by adding subsequent tensor factors
on the right, i.e., the basis induced by the following bracketing of the tensor product:

1 2 n = 1 2 3 n1 n . (95)
The elements of this basis can be labelled by their overall H eigenvalue m, their overall
fusion channel jn and and n 1 intermediate fusion channels j1 , . . . , jn1 . We may thus
,...,Jn
write these basis elements as ejJ11,...,j
, where we have defined Ji = 2i . Here, we have
n ,m
Ji for i > 1. If there is no cause for
j1 = J1 and ji is one of the summands in ji1
confusion, we will suppress the upper indices and write ej1 ,...,jn ,m . It is easy to show that the
set of ej1 ,...,jn ,m for which all the j s are held fixed forms a basis for an irrep of Uq (sl(2)) of
type 2jn , i.e., the action on this set corresponds to the action given in formula (34). Hence,
it follows that the tensor product representation becomes a -representation if we take the
inner product which makes the ej1 ,...,jn ,m orthonormal and if each of the possible 2jn is
itself a -representation. Note that if we are working with a truncated tensor product, then
there will be a different truncated tensor product space for each bracketing, because of the
non-associativity of this tensor product. The bases we have described here then provide
canonical bases for the different subspaces of the ordinary tensor product that one gets
from the different bracketings.
The basis of ej1 ,...,jn ,m is very suited to a description of the braiding. Suppose we want
to exchange particles i and i + 1, i.e., we want to calculate the action of the exchange i
on ej1 ,...,jn ,m . We can do this in three steps:
1. Move particle i completely to the left, using right-over-left exchanges. Since the
representations 1 , . . . , i1 fuse together to the representation 2ji1 and since
the fusion of this 2ji1 with i gives 2ji , this operation gives us just a constant
factor. We have
J ,J ,...,J
,...,Jn
n
i
1
(1)ji Ji ji1 q 2 (cji1 +cJi cji ) fj1 ,...,j
.
ejJ11,...,j
n ,m
n ,m
1
(96)
Here we have defined cj = j (j + 1), in accordance with the definition of c above.

i ,J1 ,...,Jn
Also, the vector fjJ1 ,...,j
is an element of the natural basis for the tensor product
n ,m
that one gets by first tensoring together 1 , . . . , i1 , adding successive factors on
the right, then tensoring on i from the left and finally tensoring on the remaining
265
factors from the right. To get the result (96), one uses (94) and, repeatedly, (68) or,
for truncated tensor products, (87).
2. Now change the bracketing, using the 6j -symbols, so that we end up in a basis
in which the representations i and 1 , . . . , i1 , no longer fuse to a fixed
representation, but the representations 1 , . . . , i1 , and i+1 do. The new basis
is the natural basis for the tensor product which one gets by first tensoring together
1 , . . . , i1 , adding successive factors on the right, then tensoring on i+1 from
the right, then tensoring on i from the left and finally tensoring on the remaining
factors from the right. In the new basis, the label ji (which gave the overall quantum
group charge of particles 1 to i) is replaced by a new label j , which gives the overall
quantum group charge of particles 1, 2, . . . , i 1, i + 1. All the other labels are as
J ,J ,...,J
before. If we denote the elements of the new basis by gj1i,...,j1 ,...,jnn ,m , then the f -basis
can be written in terms of the gs as
Ji
ji1 ji
,J1 ,...,Jn
i ,J1 ,...,Jn
=
gjJ1i,...,j
fjJ1 ,...,j
(97)
,...,j ,m ,
i ,...,jn ,m

n
J
j
j
i+1
i+1
j
where we have used the fact that the representations carried by the particles 1, . . . ,
i 1 fuse to 2ji1 and that these particles can thus be treated as one particle that
carries the representation 2ji1 .
3. Now we move particle i to the right, using left-over-right exchanges, until it has
reached the position to the right of particle i 1. At the end of this process, we have
effectively only produced a left-over-right exchange of the particles i and i + 1, as we
wanted. In the g basis, the process of exchanging particle i past particles 1, . . . , i 1,
and i + 1 is described once again by a simple phase factor (compare the first step of

the calculation), since the representations on particles 1, . . . , i 1, i + 1, fuse to 2j
and this fuses with 2Ji into the fixed fusion channel 2ji+1 . We get

J ,J ,...,J
J ,...,J
,J ,...,Jn
i+1 i
gj1i,...,j1 ,...,jnn ,m (1)j +Ji ji+1 q 2 (cji+1 cj cJi ) ej11,...,j ,...,j
n ,m
J ,...,J
,J ,...,Jn
i+1 i
where ej11,...,j ,...,j
n ,m
(98)
is an element of the basis we started with.
We may now write down the action of the elementary exchange i on the e-basis as the
cumulative effect of these three steps. We have

1

J ,...,Ji ,Ji+1 ,...,Jn
i ej11,...,ji ,...,j
=
(1)ji ji1 +j ji+1 q 2 (cji1 cji +cji+1 cj )
,m
n
j
Ji
ji1
ji
Ji+1
ji+1
j
J ,...,J
,J ,...,Jn
i+1 i
ej11,...,j ,...,j
n ,m
(99)
Using Eq. (83), one may check easily that the matrix that describes this transformation is
unitary if q is a root of unity, which is the case we are interested in. Hence, if we take the
inner product which makes the ej1 ,...,jn ,m orthonormal, then the braid group representation
which governs the exchanges of particles with Uq (sl(2))-charges is unitary, as it should be.
If either Ji = Ji+1 or ji1 = ji+1 , then it follows from the classical symmetries (80) that
the matrix for i is also symmetric.
266
3.2.11. Hidden quantum group symmetry

We will say that a quantum mechanical system has a hidden quantum group symmetry
if there is an action of a quantum group A on the Hilbert space of the theory which has the
property that it commutes with all the observables of the theory. For a system of particles
which carry Uq (sl(2))-representations, this means in particular that the H -eigenvalues
associated to the particles will not be observable, while on the other hand, one can allow
observables which make it possible to determine the Uq (sl(2))-representation associated
to each of the particles. In other words, the total quantum spin of each particle would be
measurable, but the components of this quantum spin would not be measurable. The above
definition of hidden quantum group symmetry is just what we have distilled from various
sources in the literature that mention hidden quantum group symmetries (see Section 4 for
references). Note however that there does not seem to be a completely standard definition
of this concept. Let us say more about what the above definition means within our context.
Suppose we have a system of n particles that carry representations 1 , . . . , n , of a
quantum group A. In that case the whole system will be in a state in the tensor product
space V 1 V n . If this tensor product may be decomposed into irreducibles then
the decomposition will take the form

U1 ,...,n V .
V 1 V n =
(100)
U1 ,...,n
is a vector space whose dimension equals the multiplicity of the irrep V

Here,
of A in the tensor product. When no confusion seems possible, we will just write U . If the
A-symmetry of this system is a hidden symmetry, then it follows that all the observable

operators act only on the spaces U without mixing these. That is, every observable O
should take the form

=
IV ,
O
(101)
O
is an operator acting on U and IV is the identity operator on V . Since

where each O
all the observables have this structure, the state of the system can be uniquely characterised
by a list of vectors, one for each of the spaces U . Usually, the overall quantum group
charge(s) of the system will be well-defined. In other words, the state of the system will
be described by a vector in one of the summands in the decomposition (100). In fact, there
may be superselection rules which prevent superposition of states from different summands
in (100). If the system as a whole is in the quantum group representation , then the
state of the system may be described by a vector in the space U . Now note that the
braid group representation on the tensor product V 1 V n which comes from the
action of the R-matrix of A induces an action of the braid group on each of the U . This
follows from the fact that the action of the braid group elements, like the action of any
observable, commutes with the action of A. Any operator that represents a braid group
element will thus be of the general form given above for observables. Thus, if one wants
to describe only the monodromy or braid group representation that governs the statistics
of the system at a fixed number of particles with given overall quantum group charge ,
267
one can restrict oneself to the space U . It should be clear from the previous section what
form such a representation would take for a system of n particles with a hidden Uq (sl(2))
,...,Jn
symmetry. In this case we have the canonical basis of the ejJ11,...,j
for the tensor product
n ,m
of the n representations of Uq (sl(2)). Of these, we need only retain the ones whose overall
charge jn is equal to the fixed total charge of the system, say jn = j. The remaining vectors
may then be written as tensor product vectors:
,...,Jn
,...,Jn
= ejJ11,...,j
|j, m,
ejJ11,...,j,m
(102)
J ,...,J
where ej11,...,j n now denotes a vector in the space U j . The braid group representation
on U j may now be read off immediately from the formula (99) which gave the braiding
for the full tensor product of Uq (sl(2))-representations. The matrix elements between the
,...,Jn
ejJ11,...,j
which are given in this formula can be used in unchanged form for the vectors
n ,m
,...,Jn
, since they already did not depend on m and did not mix different jn . A similar
ejJ11,...,j
treatment of braid group representations for systems with hidden quantum group symmetry
is possible in any situation in which 6j -symbols may be defined for the quantum group
representations involved. This is the case if the tensor products of these representations
have a multiplicity free decomposition into irreducibles.
3.2.12. Braiding of identical particles and fusion diagrams

In the previous subsections, we have described the braiding for a system of n particles
with a hidden Uq (sl(2))-symmetry and we have indicated that a similar description should
hold for many systems that involve other quantum groups. Let us now look at the special
case in which the particles are identical. This case is of interest for the description of the
braiding of identical quasiholes in the RR-states. When the particles are identical, they
all carry the same quantum group representation 2J and hence the upper indices on the
,...,Jn
of the canonical basis for the space U j are all equal to J . Fixing J , we
elements ejJ11,...,j
may thus forget about the upper indices and write just ej1 ,...,jn . As in the previous section,
we also fix jn = j . Now the n-tuple (j1 , . . . , jn ) may be seen as a path of length n through
the space of representation labels of Uq (sl(2)), which starts at the trivial representation
j0 = 0 and ends at j . Of course, not all paths through the space of representation labels
will correspond to an element of the canonical basis. A path will represent a basis vector
precisely if the representation at position m of the path may always be found in the
tensor product of the representation at position m 1 with the representation 2J . This
in turn means precisely that the path lies on the fusion (or Bratteli) diagram for the
representation 2J . Thus, the paths of length n on the fusion diagram of the representation
2J may be taken as a basis for the braid group representation that describes exchanges in
a system of n particles that carry the quantum group representation 2J . From Eq. (99),
one may now easily read off that the braid group generator m will only mix paths that are
identical everywhere except at position m.
As an example let us look at the case of n particles in the 2-dimensional representation
of Uq (sl(2)). The fusion diagram for this representation is just the diagram drawn in Fig. 1.
(n)
Let p = (p(1) , . . . , p(n) ) = (((1)
p , 1), . . . , (p , n)) be a path on this diagram which starts
268
at (0, 0), then goes through p(1) , p(2) , etc. and which ends at the point p(n) = ((n)
p , n) =
(, n). Then there is either no path which differs from p only at its mth vertex or there is
exactly one such path. If there is such a path, we will call it m (p). Let us write down the
action of the exchange m on a path p. We start with the cases in which p does not have a
partner path. Using Eq. (99), we see that such paths just get a phase factor. There are four
cases:
(m+1)
= 2j < (m)
(m1)
p
p < p
m p = q 1/4 p,
(m+1)
= 2j > (m)
(m1)
p
p > p
m p = q 1/4 p,
= (m+1)
= 0,
(m1)
p
p
(m)
p =1
m p = q 3/4 p,
= (m+1)
= k,
(m1)
p
p
(m)
p =k1
m p = q 3/4 p.
(103)
These equations may be summarised by saying that the path p gets a factor of q 1/4 if
it does not change direction at its mth vertex, whereas it gets a factor q 3/4 if it does
change direction (which can only happen at the boundary of the diagram). In obtaining the
equations, we used the following values for the 6j -symbols involved:
1
1
1
1
j
j + 12
j
j 12
2
2
2 0 2
=
=
= 1,
1
1
1
1
1
1
2 j +1 j + 2
2 j 1 j 2
2 0 2
1 k k1
2
2
2
(104)
= 1.
k
k1
1
2
We are now left with the case in which p does have a partner path (p). In this case, we
(m1)
(m+1)
(m1)
(m+1)
= p
= (p) = (p) = 2j and, exchanging p with
will certainly have p
(m)
(m)
(m)
(p) if needed, we can also make sure that p > (p) , so that p = 2j + 1, (p) =
2j 1. The relevant 6j -symbols for this case are given by
1
1
1
2 j j + 2
,
=
1
1
2j
+
1q
j
j
+
2
2
1
1
1
2 j j 2
,
=
1
1
2j + 1q
2 j j 2

1
1
1
1
2j + 2q 2jq
2 j j + 2
2 j j 2
,
(105)
= 1
=
1
1
1
2j + 1q
2 j j 2
2 j j + 2
and combining this with the phase factors in (99), we see that, in the linear space with basis
{p, (p)}, the exchange m is represented by the matrix

"
#
q d/2
d + 1q d 1q
q 1/4
,
m
(106)

dq d + 1q d 1q
q d/2
where we have defined d := 2j + 1. This matrix for m is obviously symmetric. It is also
unitary, as can be easily seen, using the fact that d + 1q d 1q equals d2q 1. We
will denote the braid group representation on the paths which start from (0, 0) and end
at (, n) by n and the corresponding modules by Un . An induction argument taken
269
from [40] shows that the n are all irreducible and that they are non-isomorphic for different : the representation 11 of the trivial group B1 is irreducible because it is onedimensional and = 1 is the only possibility at n = 1. Now suppose that, for all and all
n < m, all n are irreducible and non-isomorphic for different . Then the representations
are irreducible and mutually non-isomorphic for all . To see this look at U and supm
m
pose for convenience that does not equal 0 or k. Um has a unique decomposition into
1
+1
Um1
(just forget the
Bm1 -invariant submodules which is clearly given by Um = Um1
last step in the paths). Because the m1 are non-isomorphic for different values of (by
are also non-isomorphic for
the induction hypothesis), it follows immediately that the m
different values of ; their modules have different decompositions into irreducible Bm1 1
+1
and m1
are irreducible and non-isomorphic, it follows
modules. Moreover, since m1
1
+1
that the only possible proper Bm1 -invariant submodules of Um are Um1
and Um1
.
However, these will clearly be mixed by the exchange m1 , so that Um has no proper
is irreducible. Of course if equals 0 or k then the
Bm -invariant subspaces. Hence m
argument becomes even simpler and we need not repeat it.
4. Conformal field theory and quantum groups
In this section, we review the correspondence between conformal field theory and
quantum groups. In Section 4.1, we give a short general description of this correspondence,
k WZW-theory. In the next
illustrated with the example of Uq (sl(2)) versus the sl(2)
k
section, we go on to describe the quantum group Uq (sl(m)) and its relation to the sl(m)
WZW-theory. In Section 4.3, we describe representations of the braid group Bn which
factor over the Hecke algebra Hn,q . These are important in the description of the braiding
of a system of n particles with hidden Uq (sl(m))-symmetry. In Section 4.4, we describe a
quantum group for the chiral boson. Finally, in Section 4.5, we indicate quantum groups
which correspond to the parafermion theory that is used in the description of the Read
Rezayi states.
4.1. The CFTQG relation
The relation between quantum groups and conformal field theories has been much
studied over the years and it is believed that every conformal field theory has associated to
it some quantum group (or generalisation thereof) with the following properties:
Each chiral primary field of the CFT (or equivalently: each irreducible representation
of the chiral algebra) corresponds to an irreducible representation of the quantum
group.
The fusion algebra of the CFT is identical to the representation ring of the quantum
group, i.e., fusion of chiral primaries corresponds to taking the tensor product of
quantum group irreps.
The braiding of the chiral primary fields in conformal blocks corresponds to the
braiding in the tensor product of quantum group representations, as described by
means of an R-matrix and, if needed, a coassociator.
270
k WZW-theory, whose associated

The points above can be illustrated by the case of the sl(2)
2 i
k+2
quantum group is Uq (sl(2)) at q = e . For this value of q, the unitary irreducible
representations of Uq (sl(2)) that have positive quantum dimension are indeed in one
to one correspondence with the affine primary fields G of the WZW-theory. Moreover,
comparing Eqs. (2) and (53), we see that the fusion rules for the WZW-fields are the same
as the decomposition rules for tensor products of Uq (sl(2))-representations. We described
the braid group representations associated to the fundamental representation of Uq (sl(2))
in Section 3.2.12. The braiding of the corresponding conformal blocks of the WZW-theory
was calculated by Tsuchiya and Kanie [41,42] and this braiding is indeed the same as that
described in Section 3.2.12, up to a renormalisation of the blocks. In connection with this,
k conformal field
the q-6j -symbols may be identified with the fusion matrix of the sl(2)
theory as defined by Moore and Seiberg [43,44]. The pentagon equation for this fusion
matrix then corresponds to the Eq. (86) (see also Fig. 5) and the hexagon equation is just
the quasi-YangBaxter equation (88), written in terms of 6j -symbols by means of (99).
Note that it is essential in the above, that the truncated tensor product of Uq (sl(2))representations is used, rather than the ordinary one. In other words, we may say that it
is essential that one uses a weak quasi-quantum group rather than an ordinary quantum
group. This is not a very special situation; the fusion rules of many CFTs cannot be
reproduced by those of ordinary quantum groups (or quantum groups with an ordinary
tensor product). On the other hand, there is mathematical work [45,46] in which it is shown
that, given a CFT, one may always find weak quasi-quantum groups that will reproduce its
fusion and braiding properties. This does not mean that it is known for all conformal field
theories how the quantum group generators can be represented in terms of operators in
the conformal field theory. In fact, no general construction for these operators seems to
be known, although several proposals have been made for CFTs that have a Coulomb gas
description [4749]. Through this work, much is known about the quantum groups for the
WZW-models. In particular, it is well known that for any semisimple Lie algebra g, the
are related in the way we
gk WZW-model and the quantum group Uq (g) at q = e2i/(k+g)
have described above (here g is the dual Coxeter number of g). In the following, we shall
be especially interested in the case g = sl(m), because of the close relation between the
2 WZW-theories.
k and sl(k)
parafermion theory that describes the RR-states and the sl(2)
Before we go on, let us cite a few general references on the relation between CFTs and
quantum groups. Books that include information on this are, for example, [36,50] and a
review article is [28]. An early description of the correspondence between Uq (sl(2)) and
k WZW-theory can be found in [51].
the sl(2)
k WZW-theory
4.2. Uq (sl(m)) and the sl(m)
In this subsection, we recall some facts about the quantum group Uq (sl(m)) that is
k WZW-theory. Uq (sl(m)) is a q-deformation of the universal
associated with the sl(m)
enveloping algebra U (sl(m)) of sl(m). Such a q-deformation can be constructed for any
simple Lie algebra g. If we denote the simple roots of g by i , then we can associate to
each of these three generators Hi , L+

i , Li and these will generate Uq (g) as an algebra,
subject to the relations

+

Hi , L
Li , Lj = ij Hi q ,
Hi , Hj = 0,
j = Aj i Lj ,
0, if Aij = 0 or i = j,

1Aij

s (i ,i )s(s+Aij 1)/2 1 Aij
(1) q
Li , Lj =
s
s=0

1Aij s
s
Li
Lj Li , otherwise.
271
(107)
Here, A is the Cartan matrix of g. When q = 1, these relations reduce to the relations for
the ChevalleySerre basis of U (g) and when g = sl(2), they reduce to the relations we
gave in Section 3.2. If q is not a root of unity, the irreducible representations of Uq (g) are
labelled by dominant integral weights of g and one may give formulae for the action of the
generators which are similar to those given in (34). When q is a root of unity, one finds
again that all these representations remain well-defined, but many are no longer irreducible
and in particular there are indecomposable representations.
The coproduct , counit 9 and antipode S are given by
(Hi ) = 1 Hi + Hi 1,

Hi /4
+ q Hi /4 L
L
i = Li q
i ,
9(1) = 1,
S(Hi ) = Hi ,
9(L
i ) = 9(Hi ) = 0,
/2 /2
S(L
Li q
.
i ) = q
(108)
Here, is the Weyl-vector of g, which is equal to half the sum of the positive roots,
or equivalently, to the sum of the fundamental weights. One may check easily that this
comultiplication, counit and antipode satisfy the conditions given in Section 3.2. As usual,
one can define the tensor product of representations through the formula (42) and as in
the case of Uq (sl(2)), this tensor product will usually not be fully decomposable if q is
a root of unity. However, it is once more possible to define a truncated tensor product
which involves only a finite set of unitary irreducible representations and which is fully
, where g is the dual Coxeter number of g, then the
decomposable. If q = e2i/(k+g)
irreducible representations involved are each labelled by a dominant integral weight
such that (, ) k, where is the highest root of g. Hence, they are in one to one
correspondence with the affine primary fields of the g k WZW-theory. Moreover, as in
the case of sl(2), the decomposition rules of the truncated tensor product are identical
to the fusion rules of the WZW-primaries. One may also define a quantum trace and a
corresponding quantum dimension and one may then go from the ordinary to the truncated
tensor product by projecting out modules of zero quantum dimension.
An R-matrix that satisfies the requirements (68) with the given comultiplication is also
known (see, for example, chapter eight of [29] for details and references), but it is in general
not so easy to obtain the exchange matrices in any given tensor product of representations
from this R-matrix. The reason for this is that, to calculate the action of the R-matrix on
a tensor product of representations, one needs formulae for the action of the elements of
Uq (g) associated to the roots of g on both representations in the tensor product. Although
272
it is quite easy to obtain formulae similar to (34) for the action of the raising and lowering
operators L
i associated with the simple roots i , the same does not go for the action of
the raising and lowering operators that correspond to non-simple roots. Nevertheless, the
exchange matrices have been calculated in special cases, one of which is important to us.
This is the case of the tensor product of the fundamental m-dimensional representation of
Uq (sl(m)) with itself (see [29] for a detailed calculation). Let us denote this fundamental
representation by e1 , where e1 denotes the highest weight of the representation as in
Section 2.1.2. We may then write
"
m

1
(e1 e1 )(R) = q 2m q 1/2
Eii Eii +
Eij Ej i
i=1
i%=j
#

1/2
1/2
+ q q
Ejj Eii ,
(109)
i<j
where Eij denotes the matrix whose (i, j ) entry is one and whose other entries are zero.
One may check easily that this formula gives back the matrix (73) in the case of sl(2).
In the following, we want to describe the braid group representation that is associated
with an n-fold truncated tensor product en

. Since tensor products that involve e1 are
1
multiplicity free, we can apply the methods described in Section 3.2.10 and onwards. That
is, we may define 6j -symbols for the tensor products involved and describe the braiding
by formula (99) and finally graphically, in terms of paths on the fusion diagram of the
representation e1 . In fact, we can already say quite a lot about the braiding just from the
fusion diagram, without a detailed knowledge of the 6j -symbols. So let us describe this
, each vertex of the fusion diagram may be labelled the
fusion diagram. For q = e2i/(k+g)
number of fundamental representations that have been tensored up to that point, together
with a dominant integral weight of sl(m) which satisfies the requirement (, ) k.
We may equivalently represent this weight by its Young diagram and if we do this, then
the requirement that (, ) k translates to the restriction that the diagrams should not
have more than k columns. Fusion diagrams of this kind have already been drawn in
Figs. 1 and 2. Instead of using the particle number and the Young diagram for the overall
Uq (sl(m))-charge, one may also use just a Young diagram to represent each vertex. This
Young diagram is then the diagram which reduces to the Young diagram for Uq (sl(m))charge if columns of m boxes are removed and whose number of boxes is equal to the
number of representations tensored up to that vertex. As an example, we show a diagram
for Uq (sl(3)) in Fig. 6.
The connections between the different vertices are of course determined by the fusion
rules for the fundamental representation. These can be elegantly described in terms of
Young diagrams. The truncated tensor product of the fundamental representation with the
representation that has Young diagram Y decomposes into the sum of the representations
whose Young diagrams have at most m columns and may be formed by adding one box
to Y and removing any columns of length m that result. If one keeps the columns of length
m then one obtains the Young diagrams which label the vertices of the Bratteli diagram.
273
Fig. 6. The Bratteli diagram for the fundamental representation of Uq (sl(3)) at q = e2i/5 . This is
in fact the same diagram as that shown in Fig. 2, but this time each site in the diagram is uniquely
labelled by a Young diagram only. The diagrams in Fig. 2 may be recovered by removing columns
of 3 boxes.
The restriction on the number of columns then applies only to the number of columns of
length less than m.
The representation of Bn that describes exchanges for a system of n particles with
Uq (sl(m)) symmetry may now be described in terms of these Bratteli diagrams. In fact,
given the overall Uq (sl(m))-charge of the system, we may find the Young diagram Y with n
boxes which gives this overall charge and then the braid group representation space is just
the space of paths on the diagram which start at the empty diagram and end at Y . Moreover,
each of the exchanges i will only mix paths that are the same everywhere except possibly
at the ith vertex. Note that such paths occur at most in pairs, since there are no more than
two orders in which one can place the two boxes that are added in going from vertex i 1
to vertex i + 1 (if the boxes are added in the same row, for example, then there is only
one admissible order and thus only one path). The paths which are mixed transform into
each other by means of unitary matrices and since the diagrams all become periodic after
a while, one needs only to find a finite number of such matrices (see also Section 5.2 for
this). To find these matrices exactly, one should calculate the 6j -symbols of Uq (sl(m)).
However, we will not do this here, but in stead take a short cut by using the fact that
all the braid group representations we need are related to representations of the Hecke
algebra Hn,q , whose representation theory has been well studied.
4.3. The Hecke algebra Hn,q
In this paragraph, we give a short description of an algebra which plays an important role
in our understanding of the braiding of Uq (sl(m))-representations: the Hecke algebra Hn,q .
We will also describe the irreps of this algebra that are relevant to us.
Hn,q may be defined as the complex algebra with generators 1, g1 , g2 , . . . , gn1 , subject
to the relations
gi gj = gj gi
(|i j | 2),
gi gi+1 gi = gi+1 gi gi+1 ,

gi2 = (q 1)g i + q.
(110)
274
From these relations, we see that Hn,q is a q-deformation of the group algebra CSn of
the symmetric group, to which it reduces at q = 1. The reason that the Hecke algebra
comes into play in the braiding of Uq (sl(2)) representations is that the exchange matrix for
the fundamental representation of Uq (sl(m)), given in (109), satisfies the following extra
relation next to the braiding relation given in (70):
e ,e 2 m1
m1
R 1 1 = q 2m q 2m R e1 ,e1 + q 1/m (1 1).
(111)
As a consequence of this relation, the braid group representation that can be constructed
from the R-matrix also gives a representation of the Hecke algebra Hn,q . This representation is given by the prescription

m+1
gi ! q 2m R e1 ,e1 i,i+1 ,
(112)
and one may easily verify that (68) and (111) guarantee that the relations (110) for the gi
are satisfied.
The representations of the Hecke algebra which are induced by Uq (sl(m)) in this way all
factor over a quotient of the Hecke algebra, the so-called m-row quotient. This is because
the exchange matrix (109) satisfies even further relations apart from the ones already
given. For example, the 2-row quotient of the Hecke algebra (which is also called the
TemperleyLiebJones algebra) can be defined by adding the following relations to those
given in (110):
1 + gi + gi+1 + gi gi+1 + gi+1 gi + gi gi+1 gi = 0,
(113)
and one may check that the matrix q 3/4R e1 ,e1 for Uq (sl(2)) satisfies the corresponding
equation. Note that the situation we are describing already occurs in the q = 1 case. In that
case, we are describing representations of CSn in which the (m + 1)-row antisymmetriser
vanishes (as it should do for exchanges in a tensor product of m-dimensional spaces). The
Young tableaus for these representations thus have at most m rows. The equation above
indeed just says that the 3-row antisymmetriser vanishes, if one identifies the generators i
of the permutation group with the elements gi of the Hecke algebra at q = 1. We
have kept this minus sign for better compatibility with the literature on Hecke algebra
representations (for example, [40]). The relations that need to be added to the Hecke
algebra relations to obtain the general m-row quotient may be written similarly as above;
they are just the equations that say that the (m + 1)-row antisymmetrisers vanish, with
each i replaced by gi .
The representation theory of the Hecke algebra is analogous to that of the group algebra
of the symmetric group as long as q is not a root of unity, but when q is a root of unity
(and q %= 1), there are complications, similar to those that arise in the representation theory
of Uq (sl(m)). In particular, the representation ring of Hn,q is no longer semisimple for
these values of q. The same is true for the representation ring of the m-row quotients.
However, one may restrict oneself to representations that factor over a certain subquotient
of the m-row quotient and these representations do form a semisimple ring. We will now
give a quick description of the irreducible representations of Hn,q at q = e2i/(k+2) , that
factor over this semisimple quotient. These representations (among many others) have
275
been constructed by Wenzl [40] and, independently, by Ocneanu [52]. Another relevant
early reference is [53]. They are q-deformations of Youngs orthogonal representations
of the symmetric group (see, for example, [54]). Each one of the representations we are
interested in is characterised by a Young diagram Y that has at most m rows and at most k
columns of length less than m. The module of the representation characterised by Y is the
module generated by all paths on the Bratteli diagram of the fundamental representation of
Uq (sl(m)) that start at the empty diagram and end at Y . Thus, we do indeed get the same
representation modules that we described in Section 4.2. Also, the action of the elementary
exchanges is of the kind we described in Section 4.2; gi does not mix paths that differ in
any place other than at their ith vertex. Using the work of Wenzl and Ocneanu, we may
now write down the matrices through which gi acts on the spaces of paths that do differ
(i)
only at this vertex. Let us denote the Young diagram at the ith vertex of the path p as Yp ,
(1)
(2)
(n)
so that p = (Yp , Yp , . . . , Yp = Y ). Then two paths p and p can only be mapped into
each other by gi if one has Yp(n) = Yp(n)
for all n %= i. As we have already remarked, the
spaces of paths that are mapped into each other by gi are at most two-dimensional, since
the last two boxes that are added in going from the Young diagram at vertex i 1 to that
at vertex i + 1 can be added in at most two different orders. Suppose that there are indeed
two admissible orders. These two orders then correspond to two paths p and p that form a
basis for the space that we are interested in. One may define the distance between the two
boxes involved as the number of hops from box to box that one has to make if one walks
(i+1)
.
from the first to the second box along the right hand side of the Young diagram Yp
One may also define a signed version of this distance, which we will denote dp,i . Suppose
(i)
that the first box is added in row r1 and column c1 of the Young diagram Yp and that the
second is added in row r2 and column c2 of the Young diagram Yp(i+1) , then this signed
distance is given by
dp,i = r2 r1 + c1 c2 .
(114)
Clearly, this is equal to the ordinary distance if the first box is located higher up and more to
the right than the second. In the opposite case the formula above gives minus the distance.
Using this signed distance, we may now write the action of the exchange gi on the path p
as

dp,i + 1q dp,i 1q
q (1dp,i )/2
p + sign(dp,i )
p.
gi p =
(115)
dp,i q
dp,i q
Hence if dp,i > 0 (which may be achieved by exchanging p and p if necessary), then the
matrix for the action of the exchange gi on the basis {p, p } is given by

"
#
q dp,i /2
dp,i + 1q dp,i 1q
q 1/2
,
gi
(116)

dp,i q
dp,i + 1q dp,i 1q
q dp,i /2
and one may check easily that this matrix is symmetric and unitary.
The action of the exchange gi on a path p that does not have a partner path, i.e., for which
(n)
(n)
there is no other path p %= p such that Yp = Yp for all n %= i, is just multiplication by a
phase factor. This phase factor equals q if dp,i = 1, which means that the two boxes were
276
added in the same row, and it equals 1 if dp,i = 1, which means the boxes were added to
the same column, or if dp,i = k + 1, which happens at the edges of the fusion diagrams.
Clearly, the representations we have just described are the right ones for the description
of the braiding of particles with a hidden Uq (sl(m))-symmetry. For a system of n particles
with overall quantum group charge , we need the representation labelled by the Young
diagram that consists of n boxes and which reduces to the Young diagram of the charge
on removal of all columns with m boxes. With this correspondence, one may indeed
check that the formulae given in this section are the same as the ones we gave for
Uq (sl(2)) in Section 3.2.12, up to the phase factor in (112). Using the explicit form of
the exchange matrices, it is not difficult to prove some nice mathematical properties of
the representations above. For example, Wenzl has proved [40] that they are irreducible
and that representations labelled by different Young diagrams are non-isomorphic. The
argument used in the proof is essentially the same as the one we described at the end of
Section 3.2.12 for the case of Uq (sl(2)).
4.4. A quantum group for the chiral boson
In this section, we will indicate how a quantum group may reproduce the fusion and
braiding of the CFT that describes a chiral boson on a circle of radius m, wherem Z.
This CFT has m chiral primary fields, which are the vertex operators l = eil/ m , for
l2
m {0, . . . , k 1}. The vertex operator l has conformal weight 2m
and the fusion is very
simple, one has
l1 l2 = l1 +l2 (mod
m) .
(117)
All conformal blocks of primaries may be calculated explicitly, giving

l1 (z1 ), . . . , ln (zn ) = (zi zj )li lj /m .
(118)
i<j
It follows that the braiding of these blocks is Abelian; the half-monodromy which takes a
l1 around a l2 gives the block a factor of eil1 l2 /m .
We would like to reproduce the above data through a quantum group, that is we would
like to find a quantum group which has m irreducible representations whose fusion rules
and braiding are identical to those of the boson vertex operators. It turns out that this cannot
be achieved by a quantum group whose coassociator is trivial. This can be most easily seen
in the case m = 2. In this case, we would need a quantum group with two representations 0
and 1 satisfying
0 0 = 0 ,
0 1 = 1 ,
1 1 = 0 .
(119)
Now if we look at the threefold tensor product 1 1 1 , then we know on the one
hand that braiding the left 1 over the other two must give two factors of ei/2 yielding
a total factor of ei = 1. On the other hand, the braiding factor bf is also given by the
following formula (which is only valid if the coassociator is trivial)

bf = 1 1 1 (1 )(R) = 1 0 (R).
(120)
277
Here we have used the information that the two rightmost representations must fuse to 0 .
We are thus really just exchanging the left 1 over a 0 and this should give a factor
of e0i/m = 1, which yields a contradiction. To describe the chiral boson, we should
thus either use a quantum group with a non-trivial coassociator or relax the demands
on the correspondence between quantum group and CFT. A good candidate for a noncoassociative quantum group would be Uq (sl(m)) at q = e2i/3 (or k = 1). This weak
quasi-quantum group does have m representations with the right fusion rules and the
braiding for the fundamental representation does reproduce that for the vertex operator 1 ,
up to a trivial scalar factor in every exchange. However, checking the correctness of the
braiding for all the other representations of Uq (sl(m)) seems to be rather complicated and
therefore we will choose a different approach.
If we relax the demands on our quantum group such that more than one quantum group
representation may correspond to the same charge sector in the CFT, then we can reproduce
the braiding of the chiral boson CFT by a very simple finite-dimensional coassociative
quantum group. As a Hopf algebra this quantum group is the group algebra of the cyclic
group Z2m . A convenient basis for this algebra is given by the primitive idempotents
e0 , e1 , . . . , e2m1 , which project on the isotypical components of the representations of Z2m
in the group algebra. In formulae: the ei are elements that satisfy
ei ej = ij ei ,
(121)
and the full set 0 , . . . , 2m1 , of irreps of Z2m is given by

i (ej ) = ij .
(122)
The coproduct reads

(ej ) =
ei ej i (mod
2m) ,
(123)
and one may easily check that this leads to the fusion rules
i j = i+j
(mod 2m) .
(124)
Counit and antipode are given by

9(ei ) = 0 (ei ) = 0i ,
S(ei ) = e2mi .
(125)
So far, we have just described the group algebra of Z2m in terms of the ei . Now let
us introduce an R-matrix. One may easily check that the most general R-matrix which
satisfies the requirements (68) is of the form

q ij ei ej ,
R=
(126)
i,j
where q is a 2m-th root of unity. If we take q = 1, then this is just the identity on
CZ2m CZ2m and braiding is trivial. On the other hand, if q %= 1 then we have non-trivial
braiding and the braiding factor we get when taking a i around a j will be q ij .
Let us call the quantum group we have just described CZ2m,q . The correspondence
between this simple quantum group and the chiral boson CFT can now be made as
278
follows: the chiral sector corresponding to l is represented by the two quantum group
representations l and l+m . Of these, the l with 0 l < m represent the primary
fields and their even conformal descendants, while the l with m l < 2m represent
all the odd descendants. The fusion rules of the quantum group are then consistent with
those of the CFT. In particular, it is impossible to distinguish particles represented by the
representations l and l+m by means of the tensor product decomposition rules for these
representations, just as it is impossible to distinguish the primary field l from one of its
descendants by means of the CFTs fusion rules. The braiding is also correct, if we choose
q = ei/m . The braiding factors we get for two different representatives of the same CFTsector may now differ by a minus sign, but this is in fact just what we want. To clarify this,
let us look once more at the example we gave for the case m = 2. For this case, the vacuum
sector will now be represented by the representation 0 and also by the representation 2 ,
but if we exchange a 1 with a 0 then we get a factor of 1, while if we exchange a 1
with a 2 , we get factor of 1. We already know that if we have three 1 fields and we
exchange the first over the last two, we will get a factor of 1. This is due to the fact that
the correlator (118) will have a single zero at any point where two 1 -fields are brought
together. If one would just place a 0 at this point there would be no such zero and hence
also no braiding factor. Hence we can think of 0 as representing the vacuum sector, while
we can think of 2 as representing a charge-neutral bound state of two 1 fields.
4.5. Quantum groups for the parafermions
In this section, we shall describe quantum groups for the Zk -parafermion conformal
field theory that is used in the description of the RR-states. Since there are two different
coset descriptions of this CFT (cf. Sections 2.1.2 and 2.1.1), one can also expect to get two
different quantum group theoretic descriptions.

k /U
(1)k
4.5.1. The quantum group for sl(2)
k /U

Let us start with the coset sl(2)
(1)k . For this coset, we have the factorisation
k theory as a product
formula (6), which describes a Virasoro primary field of the sl(2)
of a parafermion
field and a vertex operator for a chiral boson that lives on a circle of
radius 2k. We already used this formula to explain the conformal weights and fusion
rules of the parafermions and clearly, it can also be used to calculate braidings. To see this,
look at the following equality of correlators which follows from (9):
1
i

1
n
n
1 (z ), . . . , e n (z ) . (127)
G1 (z1 ), . . . , G
1
n
n (zn ) = 1 (z1 ), . . . , n (zn ) e
k -fields and the
The braiding on the left hand side of this equation is just a braiding of sl(2)
matrices which describe this are known to be the same matrices that describe the braiding of
Uq (sl(2)) representations at q = e2i/(k+2) (the labels i do not play a role in the braiding).
The braiding on the right hand side will be described by matrices which are products of
a matrix for braiding the parafermions and a known scalar factor for braiding the bosons
vertex operators. Thus, we may obtain the braiding matrices for the parafermion fields by
just bringing the scalar factor obtained from the bosonic correlator to the left.
279
The braiding matrices which are obtained from this recipe are the same braiding
matrices that one gets for the quantum group Aq1 ,q2 := Uq1 (sl(2)) CZ4k,q2 , where q1 =
e2i/(k+2) and q2 = ei/2k . The irreducible representations of this quantum group are
tensor products of Uq1 (sl(2))-irreps and CZ4k,q2 -irreps and hence they are labelled by
an sl(2)-weight 0 k and an integer 0 < 4k. We will write these representations
as . The representation will represent the (mod 2k) -sector of the parafermion CFT.
As in the case of the chiral boson we thus have more than one quantum group representation
that corresponds to the same sector of the CFT. In fact, we now have four quantum group
representations for every sector of the CFT, since not only have we doubled the period
of the label (as we did for the boson), but we have also not taken the second of the
field identifications (8) into account. Looking at this identification, we see that the labels
(, ) and (k , k), that should be identified, usually stand for quantum group
representations of different dimensions ( + 1 and k + 1, respectively), although
their quantum dimensions are equal ( + 1q = k + 2 ( + 1)q = k + 1q ).
Nevertheless, we believe that the quantum group Aq1 ,q2 will give a good description of the
CFTs braiding properties. To motivate this statement, let us first look at the tensor product
decomposition of Aq1 ,q2 . This is given by

=

,2k }
min{+

+
,
(128)
=| |
which is the same as the fusion rules (9) for the parafermions, except that the
identifications (8) are not incorporated. Nevertheless, using the formulae (58) for the
truncated tensor product of the Uq1 (sl(2))-representations, one can see that it is impossible
to distinguish particles in representations that correspond to the same parafermion sector
by means of these fusion rules alone. In other words, it is consistent with these fusion rules
k
and k
are indistinguishable,
to declare that particles in the representations , +2k
just as it is consistent with the fusion rules of a CFT to declare all descendants of a field
indistinguishable.
Now let us look at the braiding of Aq1 ,q2 -representations. To describe this braiding,
we can use the bases that we introduced for Uq1 (sl(2)) in Section 3.2.10, because the
representations of CZ4k,q2 are one-dimensional. The matrices that describe the braiding
w.r.t. these bases will be the product of the matrices for Uq1 (sl(2)) that we gave in (99)
with the powers of q2 that we get from the R-matrix (126) for CZ4k,q2 . Using the
symmetries (92) of the truncated 6j -symbols, one may then check that, when one changes
the representations which represent CFT-sectors in a way which is consistent with the
quantum groups fusion rules, the elements of the braiding matrices will at most get minus
signs. Again, this situation is similar to the situation for the chiral boson that we discussed
in Section 4.4.
We are now left with the difficulty of choosing the quantum group representations which
should represent the fields 20 and 11 , which are important for the description of electrons
and quasiholes in the RR-states. We will use the representations 20 and 11 for this (rather
k
1 ). This choice keeps comparison to the CFT-picture
and k+1
than, for example, k2
280
easy and it gives good results. Also, it gives results which are consistent with those of
1 sl(k)
1 /sl(k)
2 , for which there is a one-to-one
the quantum group for the coset sl(k)
correspondence between quantum group representations and CFT-sectors.
1 sl(k)
1 /sl(k)
2
4.5.2. The quantum group for sl(k)
1 sl(k)
1 /sl(k)
2 , we do not have a factorisation formula like (9) and
For the coset sl(k)
therefore we cannot find the braiding matrices for this coset by the method we used for
k /U

sl(2)
(1)k in the previous section (cf. formula (127)). Still, the results of the previous
1 sl(k)
1 /sl(k)
2 suggest a
section and also the fusion rules and modular properties 7 of sl(k)
natural candidate for a quantum group related to this coset: the quantum group Uq3 (sl(k))
Uq3 (sl(k)) Uq4 (sl(k)) with q3 = ei/(k+1) and q4 = ei/(k+2) = (q1 )1 . The irreducible
representations of this quantum group are tensor products of those of the factors and hence
2 -weight, just like the fields 1 ,2
1 -weights and an sl(k)
they are labelled by two sl(k)
,
of Section 2.1.2. Let us thus denote the quantum group irreps as 1 2 . We are now in

k /U
(1)k ; we have
the same situation that we encountered in the case of the coset sl(2)
several quantum group representations per CFT-sector, because the quantum group does
not take the identifications (12) into account. However, in this case, there is a subset of
representations of the quantum group which closes under fusion and which contains exactly
,
one representation for each CFT-sector. In fact, there are two such subsets: the set of 1 2
,
with 1 = 0 and the set of 1 2 with 2 = 0. If we restrict to one of these sets (clearly,
it does not matter which of the two we use), then we are effectively forgetting about one of
the Uq3 (sl(k))-factors of the quantum group and hence we may say that the quantum group
1 /sl(k)
2 is Bq ,q := Uq (sl(k)) Uq (sl(k)). The irreducible
1 sl(k)
for the coset sl(k)
3 4
3
4
2 1 -weight 1 and an sl(k)
representations of this quantum group are labelled by an sl(k)
1weight . For the irreps that are relevant to the description of the coset CFT, the sl(k)
2 -weight through the branching rule (10), so that
weight is uniquely determined by the sl(k)
2 -weight . We may write
we may choose to label the relevant irreps by just a single sl(k)
these representations and they are in one-to-one correspondence with the fields we
defined in Section 2.1.2. Clearly, the fusion rules of the are the same as those of the ;
2 -fields or Uq2 (sl(k))they are identical to the fusion rules for the corresponding sl(k)
representations. We have not checked if all the braiding representations we get from the
quantum group Uq3 (sl(k)) Uq4 (sl(k)) are equivalent to those one gets from the quantum
group Aq1 ,q2 of the previous section, but we do know this for the representations that are
related to the quasiholes of the RR-states. In Section 4.5.1, the quasihole was represented
by the irrep 11 of Aq1 ,q2 , whereas here, it must clearly be represented by the irrep e1
of Bq3 ,q4 (for the notation e1 , see Section 2.1.2). Explicit calculation of braiding matrices,
using formulae (116), (112) and (126) shows that the braid group representations related
to these irreps are indeed equivalent. Rather than writing out all these calculations in detail
here, we will make some remarks which make this result very plausible. First of all, the
have the same
braid group representations we get from the tensor products (11 )n and en
1
7 For modular properties of cosets, one can consult for example [23,25]. The relationship between modular and
braiding properties of CFTs and quantum groups is clarified in [28].
281
fusion diagram associated to them. This guarantees, for example, that the representations
will have a similar structure (see the discussion at the end of Section 4.2) and in particular
that their dimensions are equal. Second, the eigenvalues of the matrices that represent the
fundamental exchanges may be easily found if we note that the representation matrices of
the canonical Hecke algebra generators always have eigenvalues 1 and q (this follows
directly from the last relation in (110)). Thus, if we denote the eigenvalues of the braiding
for 11 by 1 and 2 , then we have, using (112) and (126):
i
1 = (q1 )3/4 q1 q2 = e k(k+2) ,
2 = (q1 )3/4 (1)q2 = e
i(2k+1)
k(k+2)
(129)
On the other hand, the eigenvalues 1 and 2 for the braiding associated with e1 can be
found using (112) and this yields
1 = (q4 )
2 = (q4 )
k+1
2k
k+1
2k
(1)(q2 )
(1)(q2 )
k+1
2k
k+1
2k
q2 = e
i(2k+1)
k(k+2)
(1) = e
i
k(k+2)
= 2 ,
= 1 ,
(130)
so that the eigenvalues of the braidings are equal, as they should be.
5. Quantum group picture and braiding for the ReadRezayi states

In this section, we will describe the ReadRezayi states as systems of point particles
with a hidden quantum group symmetry. We also give an explicit description of braiding
representations that are associated with these states. All of this will be done in Section 5.1.
In Section 5.2, we will give an alternative description of the resulting braid group
representations, which does not make any explicit use of quantum groups. In this
description, it is also somewhat easier to change the number of quasiholes in the system.
Finally, in Section 5.3, we show how the results a of Nayak and Wilczek [2] for the Pfaffian
state arise as a special case.
5.1. RR-states and hidden quantum group symmetry
In Section 2.2, we described the RR-states as conformal blocks in a CFT which was
the tensor product of the parafermion CFT and a CFT for a chiral boson. In Section 4.5,
we derived a relation between the parafermion CFT and the quantum groups Aq1 ,q2 =
Uq1 (sl(2)) CZ4k,q2 and Bq3 ,q4 = Uq3 (sl(k)) Uq4 (sl(k)). In Section 4.4, we gave a
quantum group for the chiral boson. Clearly, we can thus make a quantum group which
will describe fusion and braiding for the ReadRezayi states by tensoring the parafermion
and boson quantum groups. However, since the extra boson factor does not affect the fusion
of the relevant representations and only adds some scalar factors to the braiding matrices,
we will choose to work with Aq1 ,q2 and Bq3 ,q4 and to add the scalar factors by hand.
Thus, we see the following picture of the RR-states emerge. The RR-system of electrons
and quasiholes can be seen as a system of point particles with hidden quantum group
symmetry (cf. Section 3.2.11). The electrons, which were represented by the operator =
20 = 2e1 in the CFT-picture, are now point particles which carry the representation 20
282
of Aq1 ,q2 or the representation 2e1 of Bq3 ,q4 . Similarly, the quasiholes, which used to be
represented by the field = 11 = e1 , now carry the representation 11 of Aq1 ,q2 or the
representation e1 of Bq3 ,q4 . The state of an RR-system with N electrons and n quasiholes
may then be described as a vector in the tensor product of N 20 (or 2e1 ) modules and n
11 (or e1 ) modules. However, not all of the vectors in this tensor product correspond to
physical states. First of all, we have to restrict to the states in a truncated tensor product with
a given bracketing, as explained in Sections 3.2.4 and 3.2.8. Second, there is a restriction
that comes from the fact that the conformal block in (16) vanishes unless all the fields that
appear in it fuse into the vacuum sector. This now means that the system as a whole is
in one of the Aq1 ,q2 -representations 00 , kk or in the Bq3 ,q4 -representation 0 . Thus, the
physical states in the tensor product are those that lie in a truncated tensor product and
are in a global quantum group representation that corresponds to the CFTs vacuum sector.
The second condition has the same consequence as the corresponding condition for the
CFT; one has to have N + n equal to zero modulo k, because otherwise there are no states
that fulfill this condition. The condition can be interpreted as saying that the incorporation
of more electrons in an RR-system and the creation of quasiholes in such a system are
A-charge preserving processes. It follows as in the CFT-picture that quasiholes may only
be created in multiples of k at a time (if the number of electrons is kept fixed).
Now let us look at the braiding of electrons and quasiholes. For convenience, we will do
this in terms of Aq1 ,q2 , but the treatment in terms of Bq3 ,q4 will give equivalent results (see
the discussion at the end of Section 4.5.2). Since the representation 20 is one-dimensional,
the braiding between electrons is Abelian. This means any exchange of electrons will just
give a phase factor. To find this factor, one may use the formulae (71) and (34) for the
universal R-matrix and for the representations of Uq1 (sl(2)), the analogous formulae (126)
and (122) for the universal R-matrix of CZ4k,q2 , and the explicit factors from the boson
vertex operators which appear in the expression (16) for the wave function. All this together
just gives a factor of 1, as is appropriate for fermions. Similarly, one may show that there
is no non-Abelian braiding between electrons and quasiholes and that the braiding factor
for electronquasihole exchanges is equal to one. Hence, as far as the braiding is concerned,
the electrons and quasiholes can be treated separately. Since only the quasiholes have nonAbelian braiding, we will from now on focus on these.
The braiding associated to a system of identical particles with hidden quantum group
symmetry can be elegantly described in terms of a basis that is labelled by the paths on the
fusion diagram of the quantum group representation carried by the particles (we described
this in detail in Sections 3.2.10 to 3.2.12). The quasiholes of the RR-states carry the Aq1 ,q2 representation 11 and the fusion diagram for this representation is the same as that for
the parafermionic field 11 , which is in turn the same as the fusion diagram for the spin1
2 -representation of Uq1 (sl(2)). The braiding representations associated to this Uq (sl(2))representation were described in detail in Section 3.2.12 (and they were also included in the
material of Section 4.3). The only difference between the braid representations described
there and the braiding for the RR-quasiholes lie in a scalar factor for every exchange,
which comes from the CZ4k,q2 part of Aq1 ,q2 and from the explicit factors in the wave
function (16). Thus, a basis for the space of states with n quasiholes in fixed positions is
283
labelled by the paths on the fusion diagram of Fig. 1 which start at the point (0, 0) and
which end at the point (n, ), where = N (mod k), so that the Aq1 ,q2 -charge of the
whole system corresponds to the vacuum sector of the CFT. We will call the braid group
representation on this space n . In this representation, the braid group generator m will
only mix paths which differ from each other only at the mth node. Given any path p, there
will be at most one path p which differs from p only at the mth node and we will call this
the partner path of p at node m. If a path does not have a partner path at node m, then the
action of m on this path will be multiplication by a scalar factor. To give this factor, let us
first take q = e2i/(k+2), so that we have
q2 = q 2 k .
1
q1 = q,
(131)
1M
The path will then get a factor of q 1+ 2(kM+2) if it changes direction at the mth node and
1M
2(kM+2)
a factor of q
otherwise. When M takes its lowest physical value of 1, these factors
1
reduce to q and 1, respectively. If a path does have a partner path at its mth node, then
the path and its partner path will have the same representations at nodes m 1 and m + 1
and these representations will have the same dimension. Let us call this dimension d. The
exchange m will then act on the vector space generated by the path and its partner path
through the matrix
q 2 + 2(kM+2)
dq
1
n (m )
1M
q d/2
d + 1q d 1q
d + 1q d 1q
q d/2
(132)
Here we have ordered the basis so that the first of the basis vectors corresponds
to the path with the highest representation at node m of the diagram. The matrix
above is symmetric and unitary and should be compared with (106). The braid group
representations n are irreducible by the same arguments as those we used for the braid
group representations associated with Uq (sl(2)). Information on their dimensions has been
gathered in Section 2.4.
5.2. Tensor product description of the braiding
We will now set up an alternative description for the braid group representations n of
the previous section. In this description, the representation spaces are seen as subspaces
of n-fold tensor product spaces. This makes it somewhat closer in spirit to the description
Nayak and Wilczek have given of the braiding for the Pfaffian state [2], a fact we will utilise
in Section 5.3. We also feel that the description of this section is useful in itself, because
it shows very clearly how braidings in systems with an arbitrary number of quasiholes can
be performed by a recipe that depends very little on this number.
Let us start by defining Vk,l to be the k-dimensional vector space spanned by
orthonormal vectors which represent the possible lth steps in a path on the fusion diagram
of Fig. 1. We write:
284
Span{v0,1 , v2,3 , . . . , vk2,k1 , v2,1 , v4,3 , . . . , vk,k1 }
Span{v , v , . . . , v
1,2 3,4
k1,k , v1,0 , v3,2 , . . . , vk1,k2 }
Vk,l =
Span{v0,1 , v2,3 , . . . , vk1,k , v2,1 , v4,3 , . . . , vk1,k2 }
Span{v1,2 , v3,4 , . . . , vk2,k1 , v1,0 , v3,2 , . . . , vk,k1 }
(k even, l odd),
(k even, l even),
(k odd, l odd),
(k odd, l even).
(133)
Here the indices on each basis vector represent the weights at the starting points and end
points of the piece of path represented by the vector. In order to simplify the description,
we have also, for l k, included some vectors which do not actually correspond to bits
of path in the fusion diagram of Fig. 1 (for example the vector v2,1 at l = 1). Clearly, any
continuous path of length n through the fusion diagram may be represented by a canonical
basis vector of the domino form v1 ,2 v2 ,3 v3 ,4 . . . vn1 ,n in the tensor
product space Vk,1 Vk,2 Vk,n . The paths in the representation space of n can be
isolated by requiring 0 = 0 and n = .
We can now define a matrix representation k,n of the relations (29) on the given
tensor product space which has a very simple form and which reduces to the braid
group representation n when one restricts to the subspace of the tensor product which
corresponds to the paths in n . The action of the i is defined as follows. k,n (i ) is
always of the form 1 1 Rk,i 1 1, where the matrix Rk,i acts only on
Vk,i Vk,i+1 . On the basis vectors which are of domino form in the ith and (i + 1)th factor,
we take
Rk,i vi ,i +1 vi +1,i +2 = vi ,i +1 vi +1,i +2 ,
Rk,i vi ,i 1 vi 1,i 2 = vi ,i 1 vi 1,i 2 ,
Rk,i vi ,i +1 vi +1,i
i
q 2 1
=
v , +1 vi +1,i
i + 1q i i
1
q 2 i + 2q i q
vi ,i +1 vi +1,i
i + 1q
(1 i k 1),
Rk,i vi ,i 1 vi 1,i
i
q 2
=
v , 1 vi 1,i
i + 1q i i
1
q 2 i + 2q i q
vi ,i +1 vi +1,i
i + 1q
Rk,i vi ,i +1 vi +1,i = q 1 vi ,i +1 vi +1,i
Rk,i vi ,i 1 vi 1,i = q
vi ,i 1 vi 1,i
(1 i k 1),
(i = 0),
(i = k),
(134)
where we have defined

1M
= q 2(kM+2) .
(135)
This factor of course reduces to 1 when M = 1. For basis vectors v which are not of the
domino form in Vk,i Vk,i+1 , we define Rk,i v = 0. With this definition, the matrices in
285
Fig. 7. Fusion diagram for the quasiholes at k = 2. The diagram must be thought extended indefinitely
in the -direction.
k,n satisfy the relations (29), but they are not invertible (of course they are invertible if
we restrict to the space generated by vectors of the domino form). Alternatively, one may
set Rk,i v = v. In that case, the matrices in k,n are invertible, but they no longer satisfy
the second relation in (29) (of course, they still do satisfy this relation on the domino
state space). It is not difficult to see that, on the set of vectors which corresponds to the
paths of n , the representation defined here indeed reduces to n . The matrices Rk,i , for
given k, depend only on the parity of i, which means that knowledge of Rk,1 and Rk,2 is
enough to determine the representation k,n and hence also all the n completely. This is
quite useful, because it gives us an easy way to go from a description of n particles to a
description of n + 1 particles; we just add another tensor factor and use the same matrices
Rk,1 and Rk,2 as before to implement particle exchanges. We hope that this explicit recipe
can be a small first step towards a second quantised description of particles with nonAbelian statistics.
5.3. Reproducing the results for the Pfaffian state
Now let us check that our results reproduce those of Nayak and Wilczek [2] for k = 2,
n = 2m even, N even and M = 1. In this case, the relevant paths on the fusion diagram
have to end at the coordinates (0, 2m) in case m is even and at the coordinates (2, 2m) in
case m is odd. The fusion diagram for k = 2 is given in Fig. 7.
Each of these paths can be uniquely characterised by stating whether or not it changes
direction at each of its odd numbered vertices. If m is even, then the paths have to change
direction an even number of times in order to end up at the point (0, 2m). If m is odd then
the number of changes of direction also has to be odd in order for the path to end up at
the point (2, 2m). Thus, for m even, we may represent any path of length 2m by a ket
|s1 , s2 , . . . , sm , where each of the si is a sign, a plus sign denoting a change of direction
and a minus sign no change. The physically relevant paths are then the paths for which the
product of all these signs is a plus sign. For m odd, we may do the same, but now with a
minus sign denoting a change of direction and a plus sign denoting no change. The relevant
states are then once more the ones whose overall sign is positive. Both for m odd and for m
even, we thus describe a 2m1 -dimensional space whose basis vectors are labelled in the
same way as those of Nayak and Wilczek. Just as Nayak and Wilczek have done, we will
interpret this space as a subspace of an m-fold tensor product of two-dimensional spaces,
each of which has basis {|+, |}. Now let us check that the action of the braiding matrices
on these states is also the same as in [2].
286
For k = 2, the tensor product Vk,1 Vk,2 contains four states with the domino property:
the states v0,1 v1,0 , v0,1 v1,2 , v2,1 v1,2 and v2,1 v1,0 . Using these as an ordered
basis of relevant states, the matrix Rk,1 can now be found by filling in (134). It is given by
i 0 0 0
0 1 0 0
Rk,1 =
(136)
0 0 i 0.
0 0 0 1
From this, we may read off that the action of the braid group generator 2l+1 on the sign
states |s1 , . . . , sm is given by

1|s1 , . . . , sm (s2l+1 = m + 1 (mod 2),
2l+1 |s1 , . . . , sm =
(137)
i|s1 , . . . , sm (s2l+1 = m (mod 2).
In this equation, we let the value + of the symbol s2l+1 correspond to 0 (mod 2) and we
let the value correspond to 1 (mod 2). We see that 2l+1 acts only on the (2l + 1)th
factor of the tensor product of sign spaces and on this factor it is given by the following
2 2 matrix:

1 0
(m = 1 (mod 2)),
0 i
2l+1
(138)

i 0
(m = 0 (mod 2)).
0 1
In Nayak and Wilceks work, the action of l+1 also corresponds to the action of a diagonal
matrix in the (l + 1)th tensor product factor. In this case, the matrix does not depend on m
and it is given by (cf. (32))

i 0
NW
i 4 i
3
4
=
.
2l+1 e e
(139)
0 1
Here, 3 denotes the third Pauli matrix. We see that the NayakWilczek matrix is the same
as ours, up to a change in the order of the basis when m is odd.
The tensor product Vk,2 Vk,3 contains only two states with the domino property: the
states v1,0 v0,1 and v0,1 v1,0 . Using this as an ordered basis for the relevant states, the
matrix Rk,2 can again be found from (134) and is given by
1+i
1+i
2
2
.
Rk,2 = 1+i
(140)
1+i
2
From this, we may read off the action of the braid group generator 2l on the states
|s1 , . . . , sm . This generator acts only on the (2l)th and (2l + 1)th tensor factors of the
sign space and on those it is given by the matrix:
1+i
1+i
0
0
2
2
1+i
1+i
0
0
2
2
2l
(141)
,
1+i
1+i
0
0
2
2
1+i
2
1+i
2
287
where the basis on which this matrix acts, is {|++, |+, |+, |}. The matrix above
is identical to Nayak and Wilczeks matrix for 2l , which we gave in (32). Hence, we have
reproduced Nayak and Wilczeks result. Note that the change of order in the basis which
was needed for 2l+1 when m is odd has no effect on the matrix for 2l , which is why the
result does not depend on the parity of m this time.
6. Discussion and outlook

In this paper, we have shown how quantum groups may be used to give an algebraic
description of the braiding and fusion properties of the excitations of non-Abelian quantum
Hall systems. Due to the well-established relationship between conformal field theory and
quantum groups, it is in principle be possible to find such a description for any quantum
Hall state that has a CFT-description. As an application, we obtained the explicit braiding
matrices for the quasihole excitations over the ReadRezayi series of states. In a special
case, these reduce to the matrices given by Nayak and Wilczek in [2], as they should.
The obvious question to ask is now whether one can somehow make predictions about
physical quantities from the results we have derived. The answer to this depends very much
on what quantities one considers as physical. For example, one may fairly easily calculate
amplitudes for AharonovBohm scattering of quasiholes from the braiding matrices we
have given, but it seems unlikely that the control over quasiholes that one needs to test these
will soon be reached in experiments. One would probably have better chances of making
contact with experiment if one could find effects of the non-Abelian braiding in some
transport properties of the quantum Hall state. To be able to make predictions about such
quantities, one would most probably need to have a better understanding of the relation
between the overcomplete set of states with localised quasiholes which we deal with in
this article and a basis of the Hilbert space of the quantum Hall state. Suitable bases of the
Hilbert space for the ReadRezayi-states are constructed in [11,12] and a logical next step
in the program of understanding the consequences of non-Abelian braiding in quantum
Hall states would thus be to express the states we deal with in this article in terms of these
bases and vice versa. Another important question is whether there is some intuitive way
of understanding which features of the underlying theory cause the quantum symmetry
exhibited by the effective theories at the plateaus. If such an intuitive picture could be
found it would probably be very helpful in extracting physics from the effective theories.
There are also some questions of a more mathematical nature which arise naturally from
our work. For example, one would like to generalise the way we associated quantum
groups to coset CFTs to more general cosets than the ones we considered. A first
generalisation would be to look at the generalisations of the parafermions that were defined
by Gepner [55]. We expect that most arguments we gave for the parafermions will go
through unchanged for these theories. In connection with this, there should be identities
like (92) for the 6j -symbols of quantum universal enveloping algebras more general
than Uq (sl(2)); one identity for each external automorphism of the corresponding affine
Lie algebra. Another interesting mathematical issue is whether the groups generated by
288
the braiding matrices we have found are finite and/or can be characterised in a nice way.
For the case studied by Nayak and Wilczek, this is certainly so: the group generated by
the fundamental exchange matrices is just a double cover of the permutation group in this
case. However, for the general case we have not yet obtained interesting results.
Acknowledgements
The authors wish to thank Eddy Ardonne, Robbert Dijkgraaf, Jrgen Fuchs, Kareljan
Schoutens and Christoph Schweigert for illuminating discussions and comments. We thank
Tom Koornwinder for useful remarks on some of the quantum group theory we used.
References
[1] N. Read, E. Rezayi, Beyond paired quantum Hall states: parafermions and incompressible states
in the first excited Landau level, Phys. Rev. B 59 (12) (1999) 80848092, cond-mat/9809384.
[2] C. Nayak, F. Wilczek, 2n quasiholes realize 2n1 -dimensional spinor braiding statistics in
paired quantum Hall states, Nucl. Phys. B 479 (1996) 529553, cond-mat/9605145.
[3] F. Haldane, E. Rezayi, Spin-singlet wave-function for the half-integral quantum Hall effect,
Phys. Rev. Lett. 60 (1988) 956.
[4] G. Moore, N. Read, Nonabelions in the fractional quantum Hall effect, Nucl. Phys. B 360 (1991)
362396.
[5] E. Ardonne, K. Schoutens, A new class of non-Abelian spin-singlet quantum Hall states, Phys.
Rev. Lett. 82 (25) (1999) 50965099, cond-mat/9811352.
[6] M. Greiter, X.G. Wen, F. Wilczek, Paired Hall state at half filling, Phys. Rev. Lett. 66 (24)
(1991) 32053208.
[7] M. Greiter, X.G. Wen, F. Wilczek, On paired Hall states, Nucl. Phys. B 374 (1992) 567.
[8] R. Willet, J. Eisenstein, H.L. Strmer, D.C. Tsui, A.C. Gossard, J.H. English, Observation of an
even-denominator quantum number in the fractional quantum Hall effect, Phys. Rev. Lett. 59
(1987) 1776.
[9] R. Morf, Transition from quantum Hall to compressible states in the second Landau level: new
light on the = 52 enigma, Phys. Rev. Lett. 80 (1998) 1505, cond-mat/9809024.
[10] F. Haldane, E.H. Rezayi, Incompressible paired Hall state, stripe order, and the composite
fermion liquid phase in half-filled Landau levels, Phys. Rev. Lett. 84 (20) (2000) 46854688,
cond-mat/9906137.
[11] V. Gurarie, E. Rezayi, Parafermion statistics and quasihole excitations for the generalizations
of the paired quantum Hall states, Phys. Rev. B 61 (8) (2000) 54735482, cond-mat/9812288.
[12] E. Ardonne, N. Read, E.H. Rezayi, K. Schoutens, Non-Abelian spin-singlet quantum Hall
states: wave functions and quasi-hole state counting, cond-mat/0104250.
[13] A. Cappelli, L. Georgiev, I. Todorov, Parafermion Hall states from coset projections of Abelian
conformal field theories, hep-th/0009229.
[14] T.L. Ho, Broken symmetry of two-component = 12 quantum Hall states, Phys. Rev. Lett. 75
(1995) 1186.
[15] D.C. Cabra, A. Lopez, G.L. Rossini, Transition from Abelian to non-Abelian quantum Hall
states, Eur. Phys. J. B 19 (2001) 21, cond-mat/0006328.
[16] A. Zamolodchikov, V. Fateev, Nonlocal (parafermion) currents in two-dimensional conformal
quantum field theory and self-dual critical points in Zn -symmetric statistical systems, Sov.
Phys. JETP 62 (2) (1985) 215225.
289
[17] D. Gepner, Z. Qiu, Modular invariant partition functions for parafermionic field theories, Nucl.
Phys. B 285 (1987) 423453.
[18] P. Jacob, P. Mathieu, Parafermionic character formulae, hep-th/0006233.
[19] F.A. Bais, P. Bouwknegt, M. Surridge, K. Schoutens, Extensions of the Virasoro algebra
constructed from KacMoody algebras using higher order Casimir invariants, Nucl. Phys. B 304
(1988) 348370.
[20] F.A. Bais, P. Bouwknegt, M. Surridge, K. Schoutens, Coset construction for extended Virasoro
algebras, Nucl. Phys. B 304 (1988) 371391.
[21] P. Griffin, D. Nemeschansky, Bosonization of Zn parafermions, Nucl. Phys. B 323 (1989) 545
571.
[22] D. Nemeschansky, FeiginFuchs representation of string functions, Nucl. Phys. B 363 (1991)
665678.
[23] F. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, 1997.
[24] J. Fuchs, D. Gepner, On the connection between WZW and free field theories, Nucl. Phys.
B 294 (1987) 3042.
[25] D. Gepner, Field identification in coset conformal field theories, Phys. Lett. B 222 (2) (1989)
207212.
[26] J. Fuchs, C. Schweigert, Symmetries, Lie Algebras and Representations, a Graduate Course for
Physicists, Cambridge Monographs on Mathematical Physics, Cambridge Univ. Press, 1997.
[27] M. Abramowitz, I. Stegun (Eds.), Handbook of Mathematical Functions, 5th edn., Dover
Publications, 1968.
[28] L. Alvarez-Gaum, C. Gmez, G. Sierra, Topics in conformal field theory, in: L. Brink,
D. Friedan, A.M. Polyakov (Eds.), The Physics and Mathematics of Strings, Memorial Volume
for Vadim Knizhnik, World Scientific, Singapore, 1990, pp. 16184.
[29] V. Chari, A. Pressley, A Guide to Quantum Groups, Cambridge Univ. Press, 1994.
[30] G. Keller, Fusion rules of Uq (sl(2, C)), q m = 1, Lett. Math. Phys. 21 (1990) 273286.
[31] G. Mack, V. Schomerus, Quasi-Hopf quantum symmetry in quantum theory, Nucl. Phys. B 370
(1992) 185.
[32] A.N. Kirillov, ClebschGordan quantum coefficients, Zap. Nauchn. Sem. LOMI 9 (1988) 67
84 [J. Sov. Math. 53 (3) (1991) 264276].
[33] L. Vaksman, q-analogues of ClebschGordan coefficients, and the algebra of functions on
the quantum group su(2), Dokl. Akad. Nauk. SSSR 306 (2) (1989) 269271 [Sov. Math.
Dokl. 39 (3) (1989) 467470].
[34] A.N. Kirillov, N.Y. Reshetikhin, Representations of the algebra Uq (sl(2)), q-orthogonal
polynomials and invariants of links, in: V.G. Kac (Ed.), Infinite Dimensional Lie Algebras and
Groups, Proceedings of the Conference held at CIRM, Luminy, Marseille, World Scientific,
Singapore, 1988, pp. 285339.
[35] V. Pasquier, H. Saleur, Common structures between finite systems and conformal field theories
through quantum groups, Nucl. Phys. B 330 (1990) 523556.
[36] C. Gomez, M. Ruiz-Altaba, G. Sierra, Quantum Groups in Two-Dimensional Physics,
Cambridge Monographs on Mathematical Physics, Cambridge Univ. Press, 1996.
[37] I.I. Kachurik, A.U. Klimyk, On Racah coefficients of the quantum algebra Uq (su(2)), J. Phys.
A: Math. Gen. 23 (1990) 27172728.
[38] D.A. Varshalovich, A.N. Moskalev, V.K. Khershonskii, Quantum Theory of Angular Momentum, World Scientific, 1988.
[39] V.G. Drinfeld, Quasi-Hopf algebras and KnizhnikZamolodchikov equations, in: Problems of
Modern Quantum Field Theory, Springer, Berlin, 1989, pp. 113.
[40] H. Wenzl, Hecke algebras of type An and subfactors, Invent. Math. 92 (1988).
[41] A. Tsuchiya, Y. Kanie, Vertex operators in conformal field theory on P1 and monodromy
representations of the braid group, Lett. Math. Phys. 13 (1987) 303.
290
[42] A. Tsuchiya, Y. Kanie, Vertex operators in conformal field theory on P1 and monodromy
representations of braid group, in: M. Jimbo, T. Miwa, A. Tsuchiya (Eds.), Conformal
Field Theory and Solvable Lattice Models, Advanced Studies in Pure Mathematics, Vol. 16,
Academic Press, Boston, 1988, pp. 297372;
A. Tsuchiya, Y. Kanie, in: Integrable Systems in Quantum Field Theory and Statistical
Mechanics, Advanced Studies in Pure Mathematics, Vol. 19, Academic Press, Boston, 1989,
pp. 675682, Erratum.
[43] G. Moore, N. Seiberg, Polynomial equations for rational conformal field theories, Phys. Lett.
B 212 (1988) 451460.
[44] G. Moore, N. Seiberg, Classical and quantum conformal field theory, Commun. Math. Phys. 123
(1989) 177254.
[45] S. Majid, Reconstruction theorems and rational conformal field theories, Int. J. Mod. Phys. A 6
(1991) 4359.
[46] V. Schomerus, Construction of field algebras with quantum symmetry from local observables,
Commun. Math. Phys. 169 (1995) 193.
[47] C. Gomez, G. Sierra, Quantum group meaning of the Coulomb gas, Phys. Lett. B 240 (1990)
149.
[48] C. Gomez, G. Sierra, The quantum group symmetry of rational conformal field theories, Nucl.
Phys. B 352 (1991) 791828.
[49] P. Bouwknegt, J. McCarthy, K. Pilch, Quantum group structure in the Fock space resolutions
of sl(n)
representations, Commun. Math. Phys. 131 (1991) 125.
[50] J. Fuchs, Affine Lie Algebras and Quantum Groups, Cambridge Monographs on Mathematical
Physics, Cambridge Univ. Press, 1992.
[51] L. Alvarez-Gaum, C. Gmez, G. Sierra, Quantum group interpretation of some conformal field
theories, Phys. Lett. B 220 (1989) 142152.
[52] P. Freyd, D. Yetter, J. Hoste, W. Lickorish, K. Millett, A. Ocneanu, A new polynomial invariant
of knots and links, Bull. Amer. Math. Soc. 12 (1985) 239246.
[53] P. Hoefsmit, Representations of Hecke algebras of finite groups with BN pairs of classical type,
Ph.D. Thesis, University of British Columbia, 1974.
[54] H. Boerner, Representations of Groups With Special Consideration for the Needs of Modern
Physics, 2nd edn., North-Holland, 1969.
[55] D. Gepner, New conformal field theories associated with Lie algebras and their partition
functions, Nucl. Phys. B 290 (1987) 1024.

Large N spin quantum Hall effect

D. Bernard 1 , N. Regnault, D. Serban
Service de Physique Thorique de Saclay, Laboratoire de la Direction des Sciences de la Matire du
Commisariat lEnergie Atomique, URA 2306 du CNRS, F-91191 Gif-sur-Yvette, France
Abstract
We introduce a large N version of the spin quantum Hall transition problem. It is formulated as a
problem of Dirac fermions coupled to disorder, whose Hamiltonian belong to the symmetry class C.
The fermions carry spin degrees of freedom valued in the algebra sp(2N), the spin quantum Hall
effect corresponding to N = 1. Arguments based on renormalization group transformations as well
as on a sigma model formulation, valid in the large N limit, indicate the existence of a crossover as
N varies. Contrary to the N = 1 case, the large N models are shown to lead to localized states at zero
energy. We also present a sigma model analysis for the system of Dirac fermions coupled to only
sp(2N) random gauge potentials, which reproduces known exact results. 2001 Elsevier Science
PACS: 73.43.Cd
1. Introduction
The landscape of delocalization transitions is wider in two dimensions than in higher
dimensions, paralleling the classification of the new random ensembles of Ref. [1]. Many
of these transitions may be modeled by Dirac fermions coupled to various disordered
variables. When formulated as field theories these transitions are usually mapped into
difficult strong coupled systems requiring the identification of nontrivial infrared fixed
points. In this respect, the recently introduced su(2) spin quantum Hall transition [2],
which corresponds to random Hamiltonians belonging to class C of Ref. [1], appears as
an exception. Indeed, using a network formulation, it has been argued [3] that the critical
properties of the latter model are potentially described by percolation in two dimensions.
This opened the possibility of exact computations of characteristic critical exponents [3] or
of the mean conductance [4]. See also Ref. [5] for a spin chain formulation of this model.
E-mail address: regnault@spht.saclay.cea.fr (N. Regnault).
1 Member of the C.N.R.S.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 5 2 - 2
292
D. Bernard et al. / Nuclear Physics B 612 [FS] (2001) 291312
However, these results still resist to a field theory derivation, see, e.g., [6]. Such alternative
approach would be useful for the related, but yet unsolved, problem of the quantum Hall
transition.
The aim of this paper is to study a large N version of the spin quantum Hall transition.
We shall formulate it as Dirac fermions with sp(2N) spin degrees of freedom whose
Hamiltonians belong to the class C. This requirement, which forces us to choose the
algebra sp(2N), is compatible with the introduction of four types of disordered potentials.
These potentials are all generated under renormalization group transformations. Some of
the results valid in the N = 1 spin quantum Hall effect extend to the large N generalization.
This holds true for instance for the spin-charge separation for particular fine tuned versions
of the models. However, the following analysis indicates that there exists a crossover as N
is increased. Namely, we will find that, contrary to the N = 1 case, zero energy states are
localized at large N .
We analyse these models in two steps. First, renormalization group transformations
allow us to identify the field theory describing the universality class of such models.
It corresponds to disordered variables isotropically distributed among the four possible
types of disordered potentials. We then use supersymmetric sigma model techniques to
analyse the universal model, which for class C Dirac fermions is a sigma model on
the Riemannian symmetric superspace OSp(2|2)/GL(1|1), of type DIII|CI according
to Ref. [7]. This sigma model, characteristic for the symmetry class C, also appeared
in the context of disordered superconductors with spin rotation symmetry and no time
reversal invariance [8,9]. The resulting sigma model turns out to be a massive theory.
As a consequence the low energy limit is noncritical. This is in contrast with class D
Dirac fermions which were described in Refs. [10,11] by a massless sigma model of type
CI|DIII. In both cases, the derivation of the sigma model is justified only at large N .
For completeness, we have also included a sigma model description of Dirac fermions
coupled to sp(2N) random gauge potentials. This corresponds to a particular fine tuning of
the previously discussed models enlarging the symmetry of the Hamiltonians which then
belongs to class CI. Surprisingly, large N sigma model analysis reproduces the exact result
[6] that the critical theory is an osp(2|2)k=2N WZW model. The fixed point of class CI,
including the WZW term, was also found in [12,13] using replica and in [14] by direct
means.
The paper is organized as follows. The models as well as their supersymmetric
formulations are introduced in Sections 2 and 3. There, we also introduce the appropriate
orthosymplectic transformations and the effective actions. Symmetries of the pure systems
are identified in Section 4. The beta functions computed in Section 5 show that the
model with isotropic randomness is universal and attractive at large N . The sigma model
formulations of the latter, which is valid in the large N limit, is analysed in Section 6. We
show that, at large scale, this model is driven to a strongly coupled massive phase, due
to the absence of nontrivial topological -term. This leads to our main result concerning
localization of zero energy eigenstates. Spin-charge separation in the pure system and
sigma model formalism for perturbations by random gauge potentials, corresponding to
random Hamiltonians in class CI, are presented in Sections 7 and 8.
293
2. The models
We consider 2N species of Dirac fermions, coupled by disorder in class C [1]. The Dirac
Hamiltonian for generic disorder in this class is

+ M + m 2i + A
H=
,
(1)
2i + A M m
where = (x iy )/2, = (x + iy )/2, A = Ax iAy , A = Ax + iAy . Here, A =
a

a
sp(2N) gauge potentials, = a a (x, y) a is a random
a A (x, y) are random

spin potential, and M = i Mi (x, y)T i and m = m(x, y)1 are random mass terms. We
consider generators of sp(2N) defined by the relation
a = ( a )T 1 ,
where the superscript T denotes usual transposition, and is the symplectic unit

0
1N
.
=
1N 0
(2)
The generators T i , with the property T i = (T i )T 1 , form the complement of sp(2N)

with respect to sl(2N). The generators a and T i , together with the identity 1, span the
algebra gl(2N). Remark that for N = 1 there is no generator T i , so that sp(2) su(2).
The Hamiltonian (1) enjoys the following particle-hole symmetry defining symmetry
class C (spin rotation invariance and no time reversal symmetry) in the classification of [1]
H = CH T C 1 ,
(3)
with C = 1 an antisymmetric matrix. This relation implies that the eigenvalues occur
in pairs with opposite signs, and it relates the advanced and retarded Green functions
GR (E) = CGA (E)T C 1 .
We consider centered Gaussian distributions for the four types of disorder, with strengths
gA , g , gM and gm (positive real numbers for real disorder). The disorder variances are:

gM (2)

gm (2)
(x y),
(x y),
M(x)M(y) =
m(x)m(y) =
N
N
a
g
a
2gA ab (2)
(x y).
(x) b (y) = ab (2) (x y),
A (x)A b (y) =
N
N
The reason to consider the four types of disorder is that they are generated by
renormalization of the effective action. As we will see in the following sections, there
are particular cases where the symmetry of the Hamiltonian (1) is greater than (3) and the
renormalization group flow closes on a subset of the disorder coupling constants; this is the
case, for example, for the random vector potential alone, when the symmetry class changes
to CI.
The single particle Green functions are defined by the functional integral

Z 1 D D exp(S)
294
with Z the partition function and

2
d x %
(x)i(H E) (x),
S=
2
where E = E + i. For = 0+ , this defines the retarded Green function

1
GR (x, x ; E) = lim x|
|x = lim i (x) % (x ) .
+
+
H (E + i)
0
0
Letting
=
+
+

,
% = ( , ),
(4)
(5)
(6)
i
, i = 1, . . . , 2N , and similarly for , one finds
where + is a 2N -component fermion +
2
d x
+ + (2 + iA) +
S=
(2 + i A)
2
+ i( + + + ) + im( + + )

+ i( M + M+ ) iEE ,
(7)
where E = + + + .
3. Supersymmetric effective action
Since the disorder is Gaussian and the deterministic part of the Hamiltonian is free,
we can use the supersymmetric method to compute disorder averages of Green functions.
i , i , i = 1, . . . , 2N , so that the
For each fermion field, we introduce bosonic partners,
inverse of the partition function for fermions become a partition function for bosons

Z(, m, M, A)1 = D eS() .
(8)
In order to simplify the notation and to take advantage of the symmetries of the problem, we
are going to introduce supermultiplets, each containing 4N fermion fields and 4N boson
fields

+
+
T
T

,
=
+ ,
+
T
T

with defined in Eq. (2). Here the index i = 1, . . . , 2N , was omitted, and the superscript
T represents the usual transposition. + is a column vector and a row vector. We define
also orthosymplectic transposes t of and t of by

T 1
T 1
T 1
T 1
t , +
, , +
, , +
,
t , +
. (9)
The inner product associated to this transposition is skew
t = t .
295
Similarly,
t a = t a ,
t T i = t T i .
(10)
One can define a transpose of 8N 8N supermatrices A by (A)t t At . Its explicit

relation to the usual supertranspose (we use same symbol T to denote the transposition for
the usual matrices and the supertransposition for the supermatrices), is given by

0 1
0
1
t
T 1
EFF
EBB ,
A =A ,
(11)
=
0
EFF and EBB being the projectors on the fermionfermion space and bosonboson space,
respectively. The present definition of the supermultiplets and t , as well as of the
orthosymplectic transpose is different from the one used in [11], and it is adapted to the
present type of disorder. To be able to compare the symmetry properties of supervectors
and supermatrices in the two cases it useful to retain the following relations

1 0
1susyclass D ,
class C =
0

1
0
t
t
=
class
(12)
C
class D 0 1 1susy,
and similarly for and t .
The free theory of fermions plus the bosonic ghosts is conformal, with Virasoro central
charge c = 0 and with action
2

d x
Scft = 2
z + + z + + z + + z +
2
2

d x t
z + t z .
(13)
2
The short distance singularity of the holomorphic fields in the supermultiplet can be
written as
1
(z) t (w)
(14)
zw
and similarly for the supermultiplet .

After performing the integrals over the disorder, we obtain the following effective action
2

1
d x
g O + gm Om + gM OM + gA OA ,
Seff = Scft +
(15)
2N
2
with the following operators perturbing away from the conformal field theory

1 t t
O = t a t a ,
Om =
,
2N

OM = t T i t T i ,
(16)
OA = t a t a .
It will turn out to be useful to know alternative ways of writing these operators. Using the
antisymmetry properties (10) as well as the cyclicity of the supertrace, one has:

O = STr t a t a = STr t a t a ,
296

1
1
STr t t =
STr t t ,
2N
2N

OM = STr t T i t T i = STr t T i t T i ,
t a t a
.
OA = STr
Om =
(17)
Finally, one can consider the energy term, which can be written in the present notations
as

E = t 3 ,
(18)
where 3 = 12N 3 1susy . In order for the integrals over the bosonic field to converge,
one has to relate the bosonic fields to each other by complex conjugation
= + ,
+
= .
(19)
Then, a positive imaginary part of the energy insures convergence of the bosonic functional
integrals.
4. Symmetries
Before perturbation by the disorder operators (16), the action (13) possesses conformal
invariance
with current algebra symmetry [23]. Bilinears in the chiral supermultiplets
t
generate an osp(4N|4N) algebra.
Two natural subalgebras of osp(4N|4N) are sp(2N) and osp(2|2). The sp(2N) algebra,
which we refer to as the spin algebra, is generated by the currents,

J a (z) = t a (z).
(20)
They satisfy
J a (z)J b (0)
fcab c
J (0) + reg.,
z
with fcab the sp(2N) structure constants, showing that they form a representation of the
affine algebra sp(2N) at level zero.
The osp(2|2) algebra, which we refer to as the charge algebra, is generated by the sp(2N)
scalar

K(z) = Trgl(2N) t (z) = Trgl(2N) EI t (z) EI ,
(21)
where the trace Trgl(2N) is over the spin indices and EI is any orthonormalized basis of
gl(2N). A convenient basis is be made of a , T i and the identity 1. The currents K may

t , with components:
be viewed as a 4 4 supermatrix, K = i i i
j
i i
,
J = +
i
J = ij
,
i i
,
H = +
i i
S =
,
i j

S = ij
.
The even part of K is made of two blocks KFF and KBB with:

J
J+
H
0
.
,
KBB =
KFF =
0 H
J J
(22)
297
The currents (22) generate an osp(2|2)k current algebra at level k = 2N , cf Ref. [6]. In
particular, the KBB block generates a so(2) algebra and the KFF block generates a sp(2)
algebra.
The osp(2|2) generators commute with the sp(2N) currents:
a
J , K = 0.
The supermultiplets and t transform as affine primary fields with value in the
tensor product of the defining vector representation of sp(2N) by the 4-dimensional
representation of osp(2|2). One may alternatively [15] view the field as a rectangular
supermatrix with components i , i = 1, . . . , 2N , = 1, . . . , 4, on which the algebra
sp(2N) acts by multiplication on the left, while osp(2|2) acts by multiplication on the
right. These actions of course commute.
5. One loop beta functions

The first step in studying the effect of the disorder perturbation in (15) is to analyse
the renormalisation group flow for the coupling constants. Even at one loop, beta functions
code for important properties of the system. They can be deduced from the operator product
expansions (OPE) of marginal perturbing operators Oi , where i stands for one of the
indices of the disorder operators (16). Using the following notation for OPE
1 k
C Ok (0) + reg.,
zz ij
the one loop beta functions are given by [16,17]
1 k
Cij gi gj .
k ll gk =
2N
Oi (z, z )Oj (0)
i,j
After some careful calculations, we obtain the following results

(2N + 1)
(N 1)
1 2
gm gM 2 gm
m =
,
2gA (gm + g ) gm g
N
N
N
2(N + 1) 2
1
1 2
1
2
g gM
gA + 2 g (gm gM ),
A = (gM + g )2 +
+
2
2N
N
N
1
1
= 2gA (g + gM ) + g (gM g ) + 2 (g + 2gA )(gm gM ),
N
N
2(N + 1)
1
1
gA (g + gM ) + gM (gM g ) + 2 gM (gM gm ).
M =
(23)
N
N
N
Readers who are interested in details can find some hints in Appendices AC.
From now on, we shall be interested in the large N limit. The beta functions (23) then
simplify dramatically
1
2
A = (g + gM )2 + 2gA
,
2
= 2gA (g + gM ).
m = 2gm (g + gM ) + 4gA (g + gm ),
= 2gA (g + gM ),
Of course the large N limit only applies for g N .
(24)
298
After a large number of RG iterations, RG trajectories are asymptotic to directions which

are preserved by the RG flow and which pass through the origin. One can show that there
exist six stable directions. But only one is attractive in the region where coupling constants
are positive. This direction, which we call the isotropic direction, corresponds to the
case where all coupling constants are equal: gm = g = gM = gA = g. The associated beta
function is
g = 4g 2 .
(25)
To show the isotropic direction is attractive, we use the same method as in [6]. We project
RG flow on the sphere and parameterize coupling constants with a radial coordinate and
three angle coordinates i . We rewrite the RG equations as
i = i (1 , 2 , 3 )
and = 2 (1 , 2 , 3 ).
We develop the i around the isotropic direction, and compute the corresponding
eigenvalues. These eigenvalues are all negative, proving that the direction is attractive.
Furthermore, it can be shown that > 0. These results are corroborated by the all order
computations of Appendix C.
Thus, the isotropic direction describes the universality class in the region where coupling
constants are all positive. It corresponds to a strong coupled system that we will study in
details in the following section.
6. The sigma model approach

When the disorder couplings are all positive (and N is large), the renormalization group
flow is attracted towards the line g = gm = gM = gA = g (isotropic disorder). On
this line, preserved by the flow, the low energy physics can be described by a nonlinear
sigma model with a topological term. In this section, we derive the effective action for this
sigma model. Its target space is the Riemannian symmetric superspace OSp(2|2)/GL(1|1),
denoted DIII|CI in [7] (note that there exists another OSp(2|2)/GL(1|1) symmetric
superspace, denoted CI|DIII, appearing in the context of the symmetry class D [11]).
When all the couplings are equal, the perturbation term in the Lagrangian, quartic in
the Dirac fields can be decoupled by a HubbardStratonovich transformation involving an
unique supermatrix field Q. Using the fact that ( t ) = ( t T i ) = 0 and denoting
t + t
B =
the perturbation term becomes

g t a t a
1 t t
Lpert =
+

2N
2N

+ t T i t T i + t a t a

1
g
STr B a B a +
B + T i BT i .
=
4N
2N
(26)
299
The supermatrix B obeys the relation B = B t , relation which defines an element of the
algebra osp(4N|4N). Conjugation by the generators of the subalgebra gl(2N) projects B
on the identity on this subalgebra

1
B + T i BT i = Trgl(2N) B 12N ,
2N
so that the perturbation term reads

t
2
g
+ t .
STr Trgl(2N)
Lpert =
4N
This interaction can be decoupled using a supermatrix belonging to the osp(2|2) algebra,
t + t ). The resulting effective Lagrangian is
Q Trgl(2N) (

N
+ STr Q Trgl(2N) t +
t STr Q2
Leff = t + t
g

t t Q
N
STr Q2 .
=
(27)
Q
g
a B a +
The next step is to perform the Gaussian integrals over the Dirac fields, resulting in the
following effective action

2
N
d x
3 Q
STr Q2 + N STr ln
S[Q] =
(28)
,
Q3
g
2
where STr combines the operations of taking the supertrace STr over the matrix indices
and integrating over position space. The factors 3 under the logarithm appear after
the transformation 3 and t t 3 , correcting for the fact that the complex
conjugate of is not t but t 3 , see Eq. (19). The number N appears now as a factor
in the action, suggesting to treat the integral in the saddle point approximation. The saddle
point equation for the action (28) is given by

2Q(x)
Q
1
= Trgl(2) x|D |x , with D =
.
(29)
Q
g
We look for a spatially homogeneous solution of the form Q(x) = 3 . The saddle-point
equation then reduces to
2
1
d k 2
+ k 2 /4 .
g 1 =
2
Cutting off the integral in the ultraviolet by |k| < 1/=0 yields the equation

1/2g = ln 1 + (2=0 )2 2 ln(2=0 ),
(30)
and by inversion,

= (2=0 )1 / e1/2g 1 (2=0 )1 e1/4g ,
(31)
As the dynamically generated mass is a renormalization group invariant, Eq. (31)

corresponds to a beta function g = 4g 2 in agreement with Eq. (25).
300
6.1. Symmetry of Q and the saddle point manifold

t + t belongs to the complex algebra osp(4N|4N)
The bilinear in Dirac fields B =
defined by relation B = B t . The even part of this algebra is sp(4N) so(4N), with
so(4N) in the fermionfermion (FF) block and sp(4N) in the bosonboson (BB) block
(this is due to the fact that the matrix in (11) is symmetric in the FF sector and
antisymmetric in the BB sector). Taking the trace over gl(2N) in B leaves us with an
object belonging to the osp(2|2) algebra, but now with the so(2) part in the BB sector and
the sp(2) part in the FF sector. To see the way it happens, let us look at the FF part in
a osp(4N|4N) matrix, defined by M t = M. It is easily verified that this block has the
structure

M11
M12
MFF =
(32)
T 1 ,
M21 1 M11
T = M
where M11 , M12 and M21 are 2N 2N ordinary matrices obeying M12
12 and
T
M21 = M21 . This implies that MFF belongs to so(4N). Let us now take the trace over the
colour indices

11 M
12
M

MFF Trgl(2N) MFF =
(33)
T .
M21 M
11
Using the cyclicity of the trace and the antisymmetry of the matrix , we can show that
12 and M
T = M
21 , which means that M
FF belongs to the sp(2) algebra. Similarly,
T = M
M
12
21
BB can be shown to belong to so(2). The matrix Q inherits this symmetry from the object
M
= Trgl(2N) B, therefore, it belongs to osp(2|2).
to which it couples, B
When decoupling the interaction part with the help of the supermatrix Q, one of
the question that has to be addressed is the choice of the contour of integration. In
particular, solving this question allows to choose the acceptable solutions for the saddle
point equation, that is the ones which lie on the contour of integration or which can be
attained from it by analytical continuation. These questions have been addressed in detail
in [7] for the class C. There, it was shown that the dominant diagonal saddle point is of the
form
Q0 = 3 ,
with 3 = 3 1susy being an element of osp(2|2). Due to the global OSp(2|2) symmetry
of the effective action, this saddle point extends to a saddle point manifold
Q0 T 3 T 1 ,
where T is a constant element of OSp(2|2). Since the stabilizer of 3 is GL(1|1), the saddle
point manifold is the coset space OSp(2|2)/GL(1|1). This coset space can be parameterized
by T = exp X, with {X, 3 } = 0.
The convergence conditions for the integrals over Q restrict the saddle point manifold to
a real submanifold of the complex space OSp(2|2)/GL(1|1). In the FF sector, convergence
of the integrals over Q can be insured by choosing QFF = QFF . At the level of the
saddle point, this translates into XFF

= XFF . Therefore, the fermionfermion sector of
301
the saddle point manifold is isomorphic to the compact symmetric space Sp(2)/U (1).
The convergence conditions on Q in the bosonboson sector are more involved [7]; on
the saddle point manifold they can be reduced to XBB

= XBB , showing that the bosonic
part of the saddle point manifold is noncompact. When averages of n Green functions
are considered, it is isomorphic to the noncompact symmetric space SO (2n)/U (n), with
SO (2n) some real form of SO(2n). For n = 1, the bosonboson sector is empty. The two
symmetric spaces form the base manifold of a Riemannian symmetric superspace of type
DIII|CI [7].
6.2. Gradient expansion
The next step in the derivation of an effective action is to perform a gradient expansion
of the action (28). The low energy configurations are given by the slowly varying field
q(x) Q(x)/ = T (x)3 T (x)1 ,
(34)
where T (x) is a (slowly varying) element of OSp(2|2). Note that q(x) satisfies the
nonlinear constraint q(x)2 = 1. The degrees of freedom q(x) correspond to the Goldstone
modes of the broken symmetry OSp(2|2) GL(1|1). Fluctuations transverse to the saddle
point manifold are massive and can be neglected at this stage.
The effective action for the Goldstone modes is a nonlinear sigma model on the
symmetric superspace OSp(2|2)/GL(1|1) described previously. This sigma model may
support a topological term, since 2 (Sp(2)/U (1)) = Z. The easiest way to extract
the coupling constants of the kinetic and topological term is by using the non-Abelian
bosonisation [18]. In the supersymmetric setting, this method was used and explained in
detail for class D in [11] and it can be applied with minimal changes to the present case.
We want to evaluate the action (28) on configurations of the type (34). Due to the nonlinear
constraint q(x)2 = 1, the first term in (28) vanishes. The second term can be written, by
undoing the integral over the Dirac fields,

S[q]
+ t q + t q .
e
(35)
= D D exp d 2 x t + t
Here q has to be understood as acting like the identity on the spin indices i = 1, . . . , 2N .
The free Dirac theory plus the bosonic ghosts is equivalent to a WZW model with action

2 i [M]
1
,
Wosp(4N|4N) [M] =
(36)
d 2 x STr M 1 M +
16
24
S
where the matrix M takes values in a subspace of the complex supergroup OSp(4N|4N),
of M to a 3-ball B
and the topological term is expressed by assuming some extension M
that has position space S for its boundary (B = S)

1 M
M
1 M
M
1 M.

[M] = d 3 x F STr M
(37)
B
The rules of bosonisation for the last two terms in the exponent in (35) are to replace
t by =1 M resp. =1 M 1 , where the factor =1 is a large
the bilinears t 3 and 3
302
mass scale, of the order of =1

0 , which enters for dimensional reasons. Up to a conjugation
with the matrix diag(1, ), these are the same bosonisation rules as in [11]. The term
STr(M3 q + q3 M 1 ) can be viewed as a kind of mass term. At large /= it forces
the field M to follow q3 . This approximation is valid at momentum scale k (/=)1/2 .
Neglecting the fluctuations we can set M3 q = 1, which yields

S[q]q=T T 1 = 2NW [q3 ].
3
where the factor 2N appears from taking the trace over the spin indices. Here W [q3 ] is
the WZW action on OSp(2|2). Recall that 3 osp(2|2) so that i3 = exp(i3 /2)
and q3 belong to OSp(2|2). However, q3 does not explore all this group as q is only
a function on the coset space OSp(2|2)/GL(1|1). Evaluating the topological term for this
configuration can be done by making a smooth extension of M = q3 to the ball B with
radial coordinate 0 s 1, for example
s) = T (x) exp(is3 /2)T (x)1 (i3 ).
M(x,
1) = q(x)3 , while for s = 0 we get M(x,
0) = i3 ,
At s = 1 we have M(x,
independent of x. Inserting this extension into the expression (37) for [M], and
converting the integral over B into an integral over S = B, we find a theta term

i
1
[q3 ] =
d 2 x F STr q q q Stop (q).
24
32
Since the value of the WZW topological term does not depend on the extension, the two
opposite expressions for i [q3 ]/24 are equivalent and Stop (q) iZ. Gathering the
kinetic and topological term, we obtain the following effective action

2N
2N
S[q]q=T T 1 =
d 2 x STr q q
d 2 x F STr q q q. (38)
3
16
32
The angle of the theta term is = 2N . It contributes trivially to the path integral as the
topological action is multiplied by 2N so that 2NStop 2iZ. The effective action is thus:

2N
Seff [q] =
(39)
d 2 x STr q q.
16
The natural ultraviolet cut-off for this effective action is 1 2=0 e1/4g since in deriving
it we neglected transverse modes of effective mass .
Action (39) is conjectured to be a massive theory as it is a sigma model on a symmetric
space with positive curvature. Recall [19] the one loop renormalization group equations
for sigma model metrics Gab :
ll Gab = Rab
with Rab the Ricci curvature. For symmetric spaces the Ricci tensor is proportional to
the metric. We need to compute this proportionality coefficient in our case. Since q =
T 3 T 1 , the tangent space at the point q = 3 is spanned by elements of the form [X, 3 ]
with X osp(2|2)/gl(1|1). By construction we can choose X such that {X, 3 } = 0. The
303
metric is then:

2
1
1
STr [X, 3 ] =
STr X2 ,
8
2
where the supertrace is understood in the defining 4-dimensional representation, and =
1/2N . Similarly, the Ricci tensor at q = 3 is defined by [20]:
G(X, X) =
R(X, X) = STr(ad X)2 .

Here the supertrace is in the adjoint representation. Recall that Ricci tensor are invariant
under metric dilatations. To compute the proportionality coefficient we pick a particular
element of osp(2|2) anticommuting with 3 , e.g., X = EFF 1 . We have STr(X2 ) =
2. Diagonalizing the adjoint action, we get a set of eigenvalues zero, two bosonic
nondegenerate eigenvalues 2 and two fermionic eigenvalues 1 with multiplicity two.
Hence, STr(ad X)2 = 2(22 2) = 4. Thus G = R/4. The RG equation then becomes:
ll = +42 .
At large distance, the model is driven to strong coupling, and it presumably becomes
massive because there is no contribution from the topological term.
The generated mass scale is of order mN eN/2 as the coupling constant
is equal to N at the ultraviolet cut-off . This is the energy scale at which the coupling
constant becomes of order one.
As a consequence, the infrared fixed point is trivial and the zero energy states are
localized with localization length of order 1/mN . In the localized regime, the behavior
of the density of states is expected to be governed by the class C matrix ensemble [21].
7. Spin-charge separation
The conformal field theory with action (13) admits a spin-charge separation. Its stress
tensor can be decomposed into the sum of the Sugawara stress tensors associated to the
spin and charge current algebras:
Tcft = Tsp(2N)0 + Tosp(2|2)2N .
(40)
Both Sugawara stress tensors have Virasoro central charge zero and are bilinear in the
corresponding currents:
1
lim J a (z)J a (w),
8(N + 1) wz

1
Tosp(2|2)2N
lim STr K(z)K(w) .
wz
8(N + 1)
Tsp(2N)0
The normalization of Tosp(2|2)2N may be found in Ref. [22]. Eq. (40) is proved in
Appendix B.
This may be checked by computing the dimension of the supermultiplet . In the free
theory, its conformal dimension is 1/2. The dimensions in the spin sector are sp(2N)0 =
304
Cas
2(N+1)
with Cas the Casimir of the corresponding representation of sp(2N), cf. Ref. [23].
2N+1
For the vector representation this gives sp(2N)0 = 4(N+1)
. In the charge sector, regular
representations of osp(2|2) are labeled by two integers j, b and their conformal dimensions
2 b2 )
are osp(2|2)2N = 2(j
2(N+1) , cf. Ref. [22]. For the 4-dimensional representation with j =
1
1/2, b = 0 this gives osp(2|2)2N = 4(N+1)
. As it should, the spin and charge conformal
dimensions add up to 1/2.
For N = 1, it was shown in Ref. [24] that the four point correlation function of the
supermultiplet may be factorized as the product of correlation functions in the sp(2N)0
and osp(2|2)2N conformal theories. However, this spin-charge factorization possesses
peculiar properties inherited from indecomposability properties of representations of
osp(2|2). In particular, osp(4N|4N) decomposes as:
sp(2N) + [8] [R]

osp(4N|4N) = osp(2|2) [1] + [8]
with [R] a (2N + 1)(N 1)-dimensional representation of sp(2N) and [8] isomorphic

to the adjoint representation of osp(2|2). The spin currents J a = t a belong to [8]
sp(2N) with [8] an eight-dimensional indecomposable representation of osp(2|2). Thus,

although these currents commute with the osp(2|2) charge generators they do not belong
to a trivial representation of the charge algebra.
This separation between spin and charge degree of freedoms still holds in the perturbed
theory provided one fine tunes the coupling constants such that g + gm = g + gM = 0.
In this case:
Lcharge g (O Om OM )

1
= g STr a t a +
t + T i t T i t ,
2N
where we used again the antisymmetry property (10). The sum in the above r.h.s. projects
out the sp(2N) colour indices leaving only the osp(2|2)2N currents (21). Thus

Lcharge = g STr(K K).
(41)
The remaining perturbing operator,

Lspin gA OA = gA J a Ja
(42)
describes a current interaction involving only the sp(2N)0 generators. Hence the total
perturbation,
Lpert = Lspin + Lcharge
is then the sum of the two commuting currentcurrent interactions.
The spin-charge separation can also be seen on the beta functions. When fine tuning
coupling constants such that g + gm = g + gM = 0, the beta functions (23) decouple:
2 and = = = 2 g 2 . In the fine tuned regime, disorder in the
gA
A = 2(N+1)
m
M
N
N
gauge potential A is marginally relevant while disorder in the spin potential is marginally
irrelevant.
305
8. Sigma model approach for the spin-charge separated system

The line g = gm = gM = 0, gA = g > 0 is stable for the renormalization group flow. It
is attractive in the fine tuned regime g +gm = g +gM = 0, g > 0, and along it the flow is
towards strong coupling, gA . This model was formulated and analysed in [12] using
replica and in [14] by direct means or with supersymmetry. Comparison of the various
methods was done in [25]. Based on the spin-charge separation, it can be deduced [6] that
the low energy physics on this line is given by an osp(2|2)k=2N theory. Let us derive the
same result by using the sigma model approach. We shall obtain that the effective action
is a sigma model on the supergroup OSp(2|2), endowed with a WZW term, the coupling
constants being such that k = 2N . Strictly speaking [11], the resulting WZW model is
defined on a submanifold of the complex supergroup which is a Riemannian symmetric
superspace of type D|C (meaning that the bosons have an orthogonal structure and the
fermions are symplectic). Let us remind that by bosonising the free Dirac fermions/bosons,
one obtains a WZW model on a Riemannian symmetric superspace of type C|D, at level
k = 1. Since the metric changes sign when passing from a space of type C|D to one of
type D|C, the WZW model on the space of type D|C is well defined for negative values
of the level k.
On the fixed line g = gm = gM = 0, gA = 0 the symmetry is not that of class C any
more but that of class CI. The reason is that the Hamiltonian has now an extra symmetry
H = T H T T 1
with
T = i2 ,
which can be interpreted as a time reversal symmetry.

The disorder perturbation is now simply the sp(2N)0 currentcurrent perturbation.
Adding formally terms which are zero, we obtain

gA t a t a gA
t EI t EI
STr
=
Lspin =
2N
2N

gA
t Trgl(2N) t ,
STr Trgl(2N)
=
(43)
2N
t )t = t , we can decouple
with EI the (orthonormal) generators of gl(2N). Since (
this interaction by introducing a supermatrix Q Trgl(2N) t . We can define an
orthosymplectic transposition for Q by Qt Trgl(2N) (1 Q)t . Remark that in contrast
to the preceding section Q has no specific symmetry properties. In particular Q = Qt ,
so the supermatrix Q belongs to a space larger than osp(2|2). After decoupling of the
interaction term, the effective Lagrangian becomes
t 2N

+ 2 STr Q Trgl(2N)
+
Leff = t + t
STr QQt
gA

2

N
0
Q
= t t
STr
.
Qt 0
Qt
gA
(44)
Here, we have embedded Q in the osp(4|4) algebra represented by matrices of the form

A
B
A =
, with At = A, D t = D.
B t D
306
It is easy to see that the diagonal blocks of A span two commuting osp(2|2) algebras,
so we conclude that the space to which Q belongs is isomorphic to the complement of
osp(2|2) osp(2|2) in osp(4|4).
In the absence of the energy term, the Lagrangian (44) is invariant under the
holomorphic/antiholomorphic transformations
gL (z),
gR (z),
t t gL1 (z),
t t gR1 (z),
Q gL (z)QgR1 (z),
Qt gR (z)Qt gL1 (z),
(45)
group: gLt = gL1
gRt = gR1 .
with gL (z) and gR (z) elements of the OSp(2|2)

and
It defines
an action of the group G OSp(2|2) OSp(2|2) on Q. We, therefore, expect conformal
invariance of the osp(2|2) (or charge) part of the theory. This is not surprising, in view of
the spin-charge separation, since the charge sector is unperturbed by the disorder.
Integrating out the Dirac fields, we obtain the effective action

2
2
N
d x
0
Q
Q3
STr
+
N
STr
ln
.
Seff =
(46)
Qt 0
3 Qt
gA
2
Again, for N large, the integral over the Q field can be treated in the saddle point
approximation. The diagonal saddle point solution is also given by a constant matrix of
the form Q = A 3 . The saddle point equation fixes the value of A ,
A (2=0 )1 e1/2gA .
Remark that the factor in the exponent has changed compared to (31). Due to the fact that,
2 , is also invariant under the RG flow and it has the properties of a
at large N , A = 2gA
A
dynamically generated mass.
As usually, the saddle point solution extends to a saddle point manifold, due to the
invariance of the effective action. This time, the saddle point manifold is generated by
Q = A gL (z)3 gR1 (z),
which is a much larger manifold than that of constant matrices. Q satisfies the non linear
constraint QQt = 1. Remember that 3t = 3 . The stabilizer H of 3 is constituted
of elements diag(gL , gR ) obeying
gL 3 gR1 = 3
"
gL = 3 gR 3 .
This is compatible with the multiplication law as gL hL = 3 gR hR 3 for any pairs

(gL , gR ) and (hL , hR ). This means that H OSp(2|2). The fluctuations around the
saddle point are described by a sigma model on the coset space G/H = OSp(2|2)
OSp(2|2)/OSp(2|2). Elements of G/H are pairs (gL , gR ) with the identification (gL , gR )
(gL 3 h3 , gR h) with h OSp(2|2). Convergence of the integrals on the saddle point
manifold is insured, in the FF sector, by QFF = QtFF . In the BB sector, convergence
1
= 3 gR,B
3 , or QBB = gL,B gL,B
3 . This is equivalent to demand that
requires gL,B
the hermiticity condition (19) on the bosonic component, = t 3 , is preserved by
B
307
the transformation (45). The even part (or the base) of the saddle point manifold is then
equivalent to the product of two symmetric spaces, the fermionic sector being compact and
the bosonic one being noncompact. The full saddle point manifold is a subspace of the
complex group OSp(2|2), being a Riemannian symmetric superspace of type D|C.
In order to perform the gradient expansion, we are making use again of the non-Abelian
bosonisation. First, we redefine the supermatrix field Q(x)
Q(x) = A g(x)3 ,
with
g = gL 3 gR 3 .
The nonlinear constraint on the new field is g t (x) = g(x)1 which is nothing else that
the defining relation for an element of the (complex) supergroup OSp(2|2). This shows
that the coset space G/H is diffeomorphic to OSp(2|2). When inserting this expression
in the effective Lagrangian (44) the last term vanishes. The other terms can be bosonized
using the same rules as in the preceding section. The free Dirac part gives the WZW action
t by =1 M
(36), while the mass term can be bosonised replacing again t 3 and 3
1
1
1
t
resp. = M = = M . The effective action becomes
S[Q] = Wosp(4N|4N) [M] +
A
=

d 2x
STr gM 1 + Mg 1 .
2
The mass term forces M to follow g. Neglecting the fluctuations of M around the minimum
M = g, we obtain
S[Q] = 2NWosp(2|2)[g],
(47)
where the factor 2N appears from taking the trace over the spin indices, on which g acts
as the identity. The topological WZW term survives to this reduction from OSp(4N|4N)
to OSp(2|2). The reality conditions discussed above mean that the restrictions gF and gB
of g in the FF and BB sectors satisfy gF1 = gF and gB = gB , respectively. This ensures the
stability of the action (47).
As already mentioned, Wosp(4N|4N) [M] is defined on a Riemannian symmetric space of
type C|D, while Wosp(2|2) [g] is defined on a Riemannian symmetric space of type D|C.
The quadratic form defining the metric changes sign between the two types of spaces, since
STr TrBB TrFF = TrSp TrSO in the first case and STr = TrSO TrSp in the second
case. In order to have a well defined functional integral, the level k of the WZW action
has to change sign between the two cases. We conclude that the action (47) corresponds
to an osp(2|2)k=2N theory. It is surprising that the saddle point approximation is able to
reproduce this exact result.
Acknowledgements
It is a pleasure to thank M. Zirnbauer for exchanges on related matter, for pointing to
us that spin-charge separation should extend to any N and for furnishing us Ref. [15], and
M. Bauer and A. LeClair for numerous discussions and related collaborations.
308
Appendix A. Algebraic coefficients

The aim of this appendix is to define the algebraic conventions used for the calculations.
We will also gather various algebraic formulas helpful for reproducing our results for the
beta functions.
The ensemble of sl(2N) generators are denoted by T A , and they can be divided in
generators of the subalgebra sp(2N), a , and generators of the complement of sp(2N)
in sl(2N), T i . The normalization we choose is

Tr T i T j = ij ,
Tr a b = ab .
Tr T A T B = AB ,
The various structure constants are defined by
a b
i j
, = if abc c ,
T , T = if ij k T k ,
a b

, = ib abi T i + x ab I,
T i , T j = if ij a a ,
a i
a i
, T = if aij T j ,
, T = ibaib b ,
A B
T , T = if ABC T C ,
where [. , .] stands for commutators and {. , .} anticommutators.
When evaluating beta-functions, a certain number of algebraic identities are needed to
simplify expressions. Here is a list of algebraic coefficients that are needed:
Coefficient
C
C
C
xT T T
x
xT T
x T
xT T T
DT T
D
DT T
D T
B
B T
x
Identity
T aT A = C
aa = C
T i T i = C
T A T B T A = xT T T T B
a b a = x b
T i a T i = xT T a
a T i a = x T T i
T i T j T i = xT T T T j
CDA
f
f CDB = DT T AB
cda
f
f cdb = D ab
ij
a
f f ij b = DT T ab
f aki f akj = D T ij
babi babj = B ij
bcia bcib = B T ab
x ab x ab = x
Value
(4N 2 1)/2N
(2N + 1)/2
(2N + 1)(N 1)/2N
1/2N
1/2
(N 1)/2N
1/2
(N + 1)/2N
4N
2(N + 1)
2(N 1)
2N
2(N + 1)
2(N 2 1)/N
(2N + 1)/N
Appendix B. Stress tensor factorization

The decomposition (40) may be verified directly by computing the operator products
defining the stress tensors using Wick theorem with the normalization (14). For the sp(2N)
sector we get:
309

t a a
1
2
1
1
lim
(z) (w) + t a (z)
(N + 1) wz 2(z w)
8

1
2N + 1
2
t +
t a ,
=
4(N + 1)
8(N + 1)
Tsp(2N)0 =
where we used t = 0, by antisymmetry, and the value of the sp(2N) Casimir in the
defining representation, a a = (2N + 1)/2. Similarly for the osp(2|2) sector we first
introduce the appropriate basis of gl(2N) to decompose Tosp(2|2)2N as follows:

1
lim STr t (z) EI t (w) EI
Tosp(2|2)2N =
8(N + 1) wz

t a
t a

1
=
(w) (w)
(z)
lim (z)
8(N + 1) wz

t
t i
t

1 t
+
(z) (w) (w)
(z) + (z)
T (w) (w)
T i (z) .
2N
Because of the antisymmetry property, t = t T i = 0, Eq. (10), the two last terms
are proportional to ( t ). The first term gives a contribution proportional to ( t ) but
also to ( t a )2 . Gathering these contributions and using again the value of the sp(2N)
Casimir as well as T i T i = (2N + 1)(N 1)/2N , we get:
Tosp(2|2)2N =
t
t a 2
1
1

.
4(N + 1)
8(N + 1)
Hence:
1 t
Tcft
2
which proves the claimed factorization.
Tsp(2N)0 + Tosp(2|2)2N =
Appendix C. All order beta functions

In this appendix we derive expressions for the beta functions using formula suggested
in Ref. [26]. These formula apply to currentcurrent perturbations of WZW models of the
form:

dx 2 K
O , OK = daKa J a J a ,
hK
S = Swzw +
2
K
Ja
J a
where
and
are the left and right conserved currents of the WZW models generating
an affine (super) algebra at some level k. In the case of a superalgebra the currents
J a possess a bosonic or fermionic character depending whether their degree |a| are
zero or one. In our case, the underlying algebra is the affine osp(4N|4N) at level k =
1. The currents are bilinear in the supermultiplet J a = t Xa , with Xa generators of
osp(4N|4N).
310
For the theory to be perturbatively renormalizable, one needs to choose the tensors daKa
such that:
K L ac bd
()|b||c| dab
dcd fi fj = CIKL dijI ,
K L
I
ij dai
dbj = DIKL dba
,
ja
L
,
dijK fk fbik = RLK ac dcb
where ab is the Killing invariant bilinear form of the superalgebra and fcab its structure
constants. These conditions are satisfied by the four operators O , OA , Om and OM
defined in Eq. (16).
With an appropriate renormalization prescription, the proposed beta functions [26] read
for k = 1,

h = C(h , h ) 1 + D(h)2 + 2C h D(h), h D(h) 2h D(h)RD(h),
(C.1)
KI
where D(h) is the matrix D(h)K
L = DL hI , C(x, y) is the row vector with components
LI x y , and finally
C(x, y)K = CK
L I
1

h = h 1 D(h)2 .
We shall assume that these formula are correct and capture perturbative contributions to all
orders.
With the normalization of Eq. (15), hK = gK /2N .
One has to compute all the tensors CIKL , DIKL and RLK . The first one codes the operator
product expansion of the perturbing operators:
1 KL I
CI O (0) + reg.
OK (z)OL (0) 2
|z|
C
CIKL
The coefficients
have been determined when computing the one-loop beta functions,
see Eq. (23).
The tensor DIKL may be computed using a variation on the free field representation of
the currents. Namely, introduce auxiliary copies , = 0, 1 or 2, of the supermultiplets
with two point function t = 1. Here denotes the expectation value in the
auxiliary Fock space associated the the auxiliary field . Let

K
O
daKa t Xa t Xa .
Then one has:
K L
I
O01 O02 0 = DIKL O12
.
Hence, DIKL is computable simply using Wick theorem.
The last tensor RLK is computable by looking at the following operator product
expansion:
T K (z)OL (0)
I
1 KL
K LN
2D
O (0) + reg.,
+
R
D
I
N
I
z2
K J a (z)J b (z).
with T K (z) = dab
311
The result is summarized in the following:

mm
Am
Mm
= Dm
= Dm = DM
= 1/N,
Dm
A
= 1,
DA
DM
= D = N + 1,
Dm
= 2N + 1,
DMM = N + 1,
MA
DA
= (N 1)/N,
M
DM
= N,
MM
Dm
= (2N + 1)(N 1)/N,
MM
DM
= (N + 1)(N 2)/N,

DM = N 2 1 /N.
K D LN are:
The nonvanishing FIKL = 2DIKL + RN
I
A
AM
= 2FM
= 2FM
= 8(N + 1),
FAAA = 2FM
FmAm = FmA = Fm = 2Fmm = 4(2N + 1),

2M
FAAm = FAm /2 = FAm = Fm /2 = FAmA = Fmmm = Fm = FM
= 2/N,
FAA = FA = 2,
FA = 4N,
MM
FAMA = FmMm = FM = FM
= 2(2N + 1)(N 1)/N,

AM
M
FA = FA = 2(N 1)/N,
FAM = FM = 4 N 2 1 /N,
FAA = 2(2N 1),
F = 2(4N + 3),
M
FM
= 2(4N + 1).
For arbitrary N , the beta functions are then derived from Eq. (C.1). In the N = limit,
they reduce to:
M = = 2(g + gM )gA ,
2
+ (g + gM )2 /2,
A = 2gA
m = 4gA (g + gm ) 2gm (g + gM )
(g + gM )(g gM )2
.
(g gM 2)
(C.2)
These are easy to integrate. In particular, (g gM ) is a RG invariant in the large N limit

while the beta functions for 2gA (g + gM ) are separated and quadratic.
It is then simple to verify that the isotropic line g = gm = gM = gA g is stable
and attractive to all orders and g = 4g 2 . The fact that it is quadratic is in agreement
with Eq. (30) and it provides a tiny check of the all order beta functions. The isotropic
coupling grows indefinitely with the scale. Of course, the large N approximation remains
valid only for g N . It is possible to verify that all RG trajectories starting in the domain
of positive couplings sufficiently close to the origin are asymptotic to the isotropic line at
large distances. This confirms the one loop computation.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
A. Altland, M. Zirnbauer, Phys. Rev. B 55 (1997) 1142.

V. Kagalovsky, B. Horovitz, Y. Avishai, J.T. Chalker, Phys. Rev. Lett. 82 (1999) 3516.
I.A. Gruzberg, A. Ludwig, N. Read, Phys. Rev. Lett. 82 (1999) 4524.
J. Cardy, Phys. Rev. Lett. 84 (2000) 3507.
T. Senthil, J.B. Marston, M.P.A. Fisher, Phys. Rev. B 60 (1999) 4245.
D. Bernard, A. LeClair, cond-mat/0003075.
M. Zirnbauer, J. Math. Phys. 37 (1996) 4986.
312
[8] R. Bundschuh, C. Cassanello, D. Serban, M.R. Zirnbauer, Nucl. Phys. B 532 (1998) 689;
T. Senthil, M.P.A. Fisher, L. Balents, C. Nayak, Phys. Rev. Lett. 81 (1998) 4704;
N. Read, D. Green, Phys. Rev. B 61 (2000) 10267.
[9] A. Altland, B.D. Simons, M.R. Zirnbauer, cond-mat/0006362.
[10] T. Senthil, M.P.A. Fisher, Phys. Rev. B 61 (1999) 6893.
[11] M. Bocquet, D. Serban, M.R. Zirnbauer, Nucl. Phys. B 578 (2000) 628.
[12] A.A. Nersesyan, A.M. Tsvelik, F. Wegner, Phys. Rev. Lett. 72 (1994) 2628.
[13] P. Fendley, R.M. Konik, Phys. Rev. B 62 (2000) 9359, cond-mat/0003436.
[14] D. Bernard, (Perturbed) conformal field theory applied to 2d disordered system: an introduction,
in: L. Baulieu, V. Kazakov, M. Picco, P. Windey (Eds.), Low-Dimensional Applications of
Quantum Field Theory, NATO Science Series: Physics B, Vol. 362, 1997, hep-th/9506290.
[15] R. Howe, Lectures in Applied Mathematics, Vol. 21, 1985, p. 179.
[16] A. Zamolodchikov, Sov. J. Nucl. Phys. 46 (1987) 1090.
[17] J. Cardy, in: E. Brzin, J. Zinn-Justin (Eds.), Les Houches, North-Holland, 1998.
[18] E. Witten, Commun. Math. Phys. 92 (1984) 455.
[19] D.H. Friedan, Ann. Phys. 163 (1985) 318.
[20] S. Helgason, Differential Geometry, Lie Groups and Symmetric Spaces, Academic Press, New
York, 1978.
[21] T. Senthil, M.P.A. Fisher, Phys. Rev. B 60 (1999) 6893, cond-mat/9810238.
[22] Z. Maassarani, D. Serban, Nucl. Phys. B 489 (1997) 603625.
[23] V. Knizhnik, A. Zamolodchikov, Nucl. Phys B 247 (1984) 83.
[24] M.J. Bhaseen, cond-mat/0011229.
[25] M.J. Bhaseen, J.-S. Caux, I.I. Kogan, A.M. Tsvelik, cond-mat/0012240.
[26] B. Gerganov, A. LeClair, M. Moriconi, hep-th/0011189.

Supersymmetry transformation of quantum fields

Christian Rupp a , Rainer Scharf b , Klaus Sibold b
a Institut fr Theoretische Physik, Universitt Bern, Sidlerstrasse 5, CH-3012 Bern, Switzerland
b Institut fr Theoretische Physik, Universitt Leipzig, Augustusplatz 10/11, D-04109 Leipzig, Germany
Received 23 January 2001; accepted 27 June 2001
Abstract
If supersymmetry is realized non-linearly as is the case when auxiliary fields are eliminated and/or
one works in the WessZumino gauge it is usually incorporated in terms of BRS transformations
and SlavnovTaylor identities. On the vertex functional susy transformations act even non-locally.
Furthermore, the gauge fixing term breaks supersymmetry. In the present paper we clarify in which
sense supersymmetry is still a symmetry of the system and how it is realized on the level of quantum
fields. We treat the WessZumino model as an example for chiral models, SQED and massless SYM
as prototypes of gauge theories. 2001 Elsevier Science B.V. All rights reserved.
PACS: 03.70.+k; 11.15.Bt; 11.30.Pb
Keywords: Quantum field theory; Supersymmetry; WessZumino gauge; Anomalous dimension
1. Introduction
Electroweak processes in todays particle physics are very successfully described
by the standard model [1]. The precision of the LEP II experiments will require
two loop corrections [2], which in turn will be calculable because the algebraic
method of renormalization has been extended to the standard model [3]. This method
permits an unambiguous construction of the model via the SlavnovTaylor (ST) identity
and a rigid and a local gauge Ward identity (WI) in a way independent from the
applied renormalization scheme. At present the only viable alternative to the SM is a
supersymmetric extension thereof, e.g., a minimal one (MSSM). Here the status of loop
corrections is much less satisfactory because the effect of the renormalized supersymmetry
WI has not yet been systematically taken into account. A first step into this direction has
been undertaken in [4] for the susy extension of QED. It applies techniques developed
in [6,7,17,19,20] and gives clear prescriptions on how to establish and use the susy WI.
This is non-trivial because in the chosen WessZumino gauge [8] the susy transformations
E-mail addresses: rupp@itp.unibe.ch (C. Rupp), rainer.scharf@itp.uni-leipzig.de (R. Scharf),
klaus.sibold@itp.uni-leipzig.de (K. Sibold).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 2 9 - 7
314
C. Rupp et al. / Nuclear Physics B 612 [FS] (2001) 313339
are non-linear and as is shown in [4] also non-local. The question then arises in
which sense susy is still a symmetry of the system. Clearly, it is realized on the Green
functions but is it realized on the local fields, on the state space? Does there exist a susy
charge which commutes with the S-matrix? In the present paper we wish to answer these
questions for the WessZumino model and then for SQED and SYM, basing our analysis
on the results of [4,14].
Another motivation for the considerations to follow originates from the use of the
superconformal algebra which is made in the context of string theory and in Seiberg
Witten theory. Here the algebra is taken for granted and used on the operator level for
abstract reasoning. But most of our knowledge also for higher N models comes from
perturbation theory which is the only constructive approach which one can really handle
and control. In particular the emergence of anomalies which refer to the deep ground
of quantum field theory can effectively be realized only in this context. Hence it is an
important problem to construct the generators of the superconformal algebra as operators
in perturbation theory. We prepared the basis for doing so in the papers [5] by deriving the
Green functions of the theory with multiple insertions of the supercurrent. From there to
the operators one has to pass with the help of the LSZ reduction formalism. Obviously the
first step in this program is the study of the transformations of field operators which is the
aim in the present paper.
2. The WessZumino model without auxiliary fields

The model comprises a complex scalar and a (Weyl) spinor field together with their
conjugates, forming the classical action

i
1
WZ = d4 x m2 + + m( + )
2
4

1
g
1
g
+ )
.
g 2 2 2 + + mg (
(2.1)
16
8
8
4
It has one common mass for all fields and only one coupling. This is due to
supersymmetry and parity which leave invariant this action. Without auxiliary fields the
susy transformations are non-linear and their algebra closes only on the mass shell
a problem for renormalization. It has been solved in [9] by introducing external fields
coupled to the non-linear variations and formulating the Ward identity (WI) for susy as a
-bilinear equation. Crucial is the fact that in eff appears an expression bilinear in the
external fields. Meanwhile this method has been greatly refined [6,7] and systematized,
hence we shall present the relevant results in this form.
2.1. BRS transformations, ST identity
The systematic procedure for treating susy without auxiliary fields consists in promoting
the parameters of the transformations to constant ghosts [6,7,17]. Those carry a Grassmann
number which is always opposite to its statistics: the ghosts of susy ( , ) commute, the
315
translations ( ) anticommute. In order to have nilpotent transformations the ghosts too

have to transform. With the assignment
s = i ,
s = g 2 + 2 m 2i i ,
2
s = i ,

s = g2 2
2 m 2i
(2.2)
(2.3)
(2.4)
i ,
(2.5)
s = 0 = s ,
(2.6)
s = 2
(2.7)
one finds
s2 = 0 = s2 ,
WZ
s2 = 4
,

WZ
s2 = 4
.

(2.8)
(2.9)
(2.10)
In order to obtain off-shell closure one introduces external fields coupled to the non-linear
variations and a term bilinear in these external fields

s + 4 Y Y
,
Y s + Y
ext =
(2.11)
cl = WZ + ext .
The symmetry of the model can now be expressed as a ST identity

+ s
+ s = 0.
+
+
S( )
s
(2.12)
(2.13)
It holds for = cl . One observes furthermore that the linearized operator

+ s
s
+
+
+
S
+

Y Y
Y
Y
+ s
(2.14)
satisfies
S2 = 0,
(2.15)
S s = 0,
(2.16)
S S( ) = 0
(2.17)
-term in (2.11).
for = cl . Crucial for the validity of (S2 = 0) is the presence of the Y Y
It clearly contributes those equations of motion terms which guarantee on the functional
level nilpotency as opposed to (2.9), (2.10) where on the level of elementary fields
it does not hold. Eq. (2.16) will serve as a consistency condition for constraining potential
non-symmetric higher order corrections to the ST identity (2.13).
316
The renormalization is now straightforward [9] and yields the ST identity (2.13) if one
imposes in addition
cl
=

which is possible, and recursively
(S s) = 0.
(2.18)
(2.19)
(2.18) and (2.19) imply

,
(2.20)
=
i
Y + Y
= i + 2i .
(2.21)
Y
In order to prepare for the subsequent discussion we go via Legendre transformation over
to the connected Green functions
Zc
j =
(2.22)
,
= ,
=
,
Y
Y

Zc = +
(2.23)
j + j + +
and then to the general Green functions
Z = eiZc .
(2.24)
The validity of the ST identity (2.13) entails the existence of a conserved current for a local
generalized BRS transformation

Sloc (y)Z = J (y) Z,
(2.25)
where
j ij
j i i

j

Y
j
Y

2 Y + Y
.
Sloc ij
(2.26)
2.2. Supersymmetry transformations of functionals

The presence of constant ghosts in the theory permits the subdivision of functionals
and sometimes also of their transformations into sectors of fixed ghost number this is
commonly called a filtration [6,7]. We expand
= (0) + (1) +
(2.27)
according to the number of constant ghosts. Inserting this expansion into the ST identity
the latter can first be rewritten as

S( ) = S (0) + S (0) (1) + O 2 ,
(2.28)
where
S(
(0)

(0) (0) (0)
+
)=
s
+ c.c.
(2.28) is verified by observing that (0) /Y = 0. Using

(1)

(0)
(1)
S (0) =
+ c.c. + s + O 2
Y
one can write

S( ) = (S )(1) (0) + O 2 ,
317
(2.29)
(2.30)
(2.31)
hence, since the ST identity holds order by order in the ghosts

S(1) (0) = 0.
(2.32)
We decompose
(2.33)
such that functional differential operators are defined by differentiation w.r.t. the ghosts

(1)
,
W = S
(2.34)

==0

= S (1)
W
(2.35)
,

==0

(1)
S
.
W =
(2.36)
==0
+ W 2
S = W + W
(1)
(1)
The filtration of S contains besides S
(0)
+ c.c.,
Y
(2)
(2)
S =
+ c.c.
Y
Next we exploit the fact that S is nilpotent on functionals F for which
(0)
S =
F
= i + 2i
Y
holds. This yields
(0) (0)
S S = 0,
(0) (1)
S , S = 0,
(0) (2)
(1) (1)
S S + S , S = 0.
(2.37)
(2.38)
(2.39)
(2.40)
(2.41)
(2.42)
(1) (1)
Projecting out of S S the -part one finds

(1) (1)
} 2 W .
S S = {W , W

(2.43)
(2.42) with (2.43) shows that even on Y -independent functionals F , the SUSY algebra
closes only up to equation of motions terms F /.
318
2.3. Symmetry transformations on quantum fields

Vertex functions can essentially be understood as matrix elements of operators but
are certainly not the ideal tool for revealing the underlying properties of the respective
operators, hence we study general Green functions.
As the simplest example we identify the energymomentum operator and its action on
field operators. We therefore insert (2.21) into (2.13) and permit , and to be local,
= (x) etc.,

+ K Z.
Sloc Z = J + K + K
(2.44)
Differentiating w.r.t. (z) and integrating over z yields the local WI

w | = =0, i
+
+
+

= =0

=0
Y =Y
=0
Y =Y

= T = =0 ,
=0
Y =Y
(2.45)
(2.46)
guarantees that no other contributions

with T = J . ((2.20) taken at Y = 0 = Y
depending on show up.) Translated onto Z we obtain

j
(2.47)
+ + c.c. Z
= i T Z
j
Y =0
a local WI for the translations, with T being the energymomentum tensor. By
differentiating with respect to a suitable combination of sources we generate on the right
hand side the Green function T(T X) , where X stands for an arbitrary number
of elementary fields; on the left-hand side there will always occur Green functions with
one field argument less and a -function instead. Multiplying with inverse propagators and
going on mass shell we obtain zero on the l.h.s. (as a consequence of the -functions) and
Op
on the r.h.s. the respective matrix elements of the operator T . Hence
Op
T = 0,
(2.48)
the energymomentum tensor is conserved.

Defining the energymomentum operator by

Op
Op
P = d3 x T0 ,
(2.49)
Op
(2.48) means that P is time independent. This detailed presentation served as an

illustration of the LSZ reduction technique which permits one to generate operator relations
out of Green functions [10].
In order to obtain the transformation law of field operators we deduce from (2.47)

y (y x)(y)X + = i y T T (y)(x)X ,
(2.50)
which yields

Op
y (y x) Op (y) = i y T T (y)(x) .
(2.51)
Integration over
d3 y and
x 0 +
x 0
Op (x) = i[P , (x)]Op,
319
dy 0 leads to
,
for = , ,
(2.52)
the well-known transformation law of a field operator under translations.

We learn from this result that the local version of the ST identity on Z (2.44)
should also be most useful for deriving the susy current, charge and field transformation.
Differentiation of (2.44) w.r.t. yields

eff
eff
2 eff
Z
+ i
Z
Z
+
i(y z)j (z)

(z)

Y
(z) Y (y)

eff
2 eff
eff
+ i
Z
+
Z

(y)

Y
(z) Y

eff
J (y)
Z + i J (y)
Z + y (y z)[K (y)] Z.
= y
(z)
(z)
(2.53)
Since has ghost number one, it may occur only in combination with Y , hence no
-dependent term survives at Y = 0, and the double insertions in (2.53) are absent. One
can safely integrate over z which yields
ij (y)

Z
eff
eff

(y)
+ i (y)
= J (y) Z,

(y)
Y (y)
Y (y)
(2.54)
where J J represents the susy current. From here on one may just perform all
steps as before for the translations and obtain
Op
J = 0,
(2.55)
the current conservation. With the charge

Q d3 x J0 (x)
(2.56)
the field transformations follow,

Op
i[Q , ]Op = ,
Op

Op
i{Q , }Op =
.
eff
Y
(2.57)
(2.58)
Analogously for the conjugate quantities.

As a consequence of (2.57), (2.58) and the translations one derives the validity of the
algebra
}, ] = {iQ , [iQ
, ]} + {iQ
, [iQ , ]}
[{iQ , iQ
, } =
= {iQ
eff = 2i
Y
= 2 [P , ].
(2.59)
320
Similarly,

, eff
}, ] = iQ , eff + iQ
[{iQ , iQ
Y
Y
= 2i ,
where the second equality is established by differentiating (2.44) w.r.t.

performing the LSZ reduction and using (2.21). Thus we have
(2.60)
,
} = 2 P .
{Q , Q

Y,
(2.61)
act as symmetry operators on the Hilbert

The current conservation implies that Q , Q
space of the theory, with the transformation of field variables given by (2.57), (2.58). This
field transformation law is non-linear (for ), but it is local! And in this sense one can say
that the transformations as given on the functionals (2.43) resemble the operator law here
and hint indeed correctly to a local symmetry. We obtain a closed algebra here because the
field operators obey the equations of motion.
This concludes our discussion of the WessZumino model.
3. SQED in the WessZumino gauge

The susy and gauge invariant action of SQED in the WessZumino gauge reads as
follows

1
1
SQED =
F F +
i
4
2

2
(i )D
+ |D L |2 + D +
R

PL + PL R PR
PR L
2 eQL
R
L

2
1
eQL |L |2 + eQR |R |2
2

2

m m
|L |2 + |R |2 ,
(3.1)
where
D + ieQA ,
(3.2)
F A A
(3.3)
and (A , , ) denote the photon and Weyl-photino, respectively, whereas (L , L ) and

(R , R ) refer to two chiral multiplets with charges QL = 1, QR = +1. The electron
Dirac spinor and the photino Majorana spinor are given by

L
i
,
=
.
=
i
R
(3.4)
In this gauge model susy transformations are non-linear not only because the auxiliary
fields have been eliminated but also because longitudinal components of the vector
321
superfield are being transformed away. This causes an additional problem: every susy
transformation has to be followed by a field dependent gauge transformation such that one
stays in the WessZumino gauge. Hence only gauge invariant terms can at the same time
be supersymmetric; a gauge fixing term can never be susy invariant. The task is therefore
to construct the model and then physical (i.e., gauge invariant) quantities modulo gauge
fixing.
3.1. BRS transformations, ST identity
The problems that susy transformations are non-linear, close only on-shell and that
gauge fixing is not supersymmetric have all been overcome by going to a BRS formulation
of all transformations and introducing suitable external fields, in particular also terms in
eff which are bilinear in the latter [4,6,7,19]. These generalized BRS transformations have
the form
sA = c + i i i A ,

s = 2i F i eQL |L |2 |R |2 i ,

s = 2i F i eQL |L |2 |R |2 i ,
sL = ieQL c L + 2 L i L ,
sL = ieQL c L + 2 L i L ,

sL = ieQL c L 2 mR 2 i D L i L ,

s L = ieQL c L + 2 mR + 2 i (D L ) i L ,
sc = 2i A i c,
s = 0,
(3.5b)
(3.5c)
(3.5d)
(3.5e)
(3.5f)
(3.5g)
(3.5h)
(3.5i)
s = 0,
(3.5j)
s = 2 ,
(3.5a)
(3.5k)
sc = B i c,
(3.5l)
sB = 2i c i B.
A suitable form of gauge fixing turns out to be

+ cB
g.f. = s cA
2

2
=
BA + B cc
c
i i + i cc .
2
The non-linear transformations will be defined via their coupling to external fields

ext =
Y s + Y s + YL sL

+ Y sL + YL sL + Y L s L + (L R)
L
complemented by well specified correction terms in higher orders.
(3.5m)
(3.6)
(3.7)
(3.8)
322
The contributions which are bilinear in the external fields have the form

bil =
(Y )(Y ) + 2(YL )(Y L ) + 2(YR )(Y R ) .
(3.9)
The classical action

cl = SQED + g.f. + ext + bil
(3.10)
satisfies then a SlavnovTaylor identity which can be extended [4] to all orders in the loop
expansion for the vertex functional ,

4
+ sc
+ sB
+
+
S( ) d x sA + sc
A
c
c
B
Y Y

+
+
+
+
+
(L
R)
YL L Y
YL L Y L
L
L
L
+ s
+ s + s

si +
i
Yi i
=0
(3.11)
(3.12)
(3.13)
( : all linearly transforming field and the ghosts; : all non-linearly transforming fields).
Part of the hypotheses is the gauge fixing as above (this can be established to all orders)
and the ghost dependence (see [4] for details).
An important calculational tool is given by the linearized ST operator

si +
+
S
(3.14)
l
Yi i
i Yi
which satisfies
S S( ) = 0
(3.15)
provided
S2 A = 0.
(3.16)
(The latter relation is true for the final vertex functional.)

As a consequence of gauge fixing, ghost equations and (3.13) it has been shown in [4]
that the following WIs hold
= iewem B + O(),
A

YL
+ L
YL
wem = QL L
L
YL
L
YL

+ (L R)
L + Y
L
Y L
L Y
Y L
L
L
(3.17)
(3.18)
and

i + i
= 0.
+ Yi
i
i
Yi
323
(3.19)
(3.17) expresses the local gauge invariance of , whereas (3.19) says that is translation
invariant.
3.2. Susy on the vertex like functionals
Like in the WessZumino model (Section 2.2) we follow [6,7] and introduce with the
help of the operator
+
+
a filtration. S will be expanded according to the number of constant ghosts:

(n)
S =
S ,
N
(3.20)
(3.21)
n0
where

N , S(n) = nS(n) .
(3.22)
We have

S( ) = S (0) + S (0) (1) + O 2 .
Like above the nilpotency (3.15) of S leads to the consequence

(0) 2
= 0,
S
(0) (1)
S S
S(0) S(2) + S(1)S(1)
(1) (0)
+ S S
+ S(2) S(0)
(3.23)
(3.24a)
= 0,
(3.24b)
= 0.
(3.24c)
But it is to be noted that the sector with ghost number 0 is now the one of ordinary BRS
invariance. Hence S(0) is the ordinary BRS variation of functionals and (3.24a) expresses
its nilpotency.
Decomposing S(1) again according to
(1)
+ W +
S = W + W
and inserting into (3.24c) one finds
} 2 W ,
{W , W

, W
},
{W , W } 0 {W
(3.25)
(3.26)
(3.27)
i.e., the supersymmetry algebra on functionals F () which obey

S(0) F = 0
(3.28)
(0)
and where means up to S -variations. On functionals which are not invariant under
ordinary BRS transformations the well-known susy algebra is realized only modulo BRS
324
as giving the supersymmetry variation of

variations. Hence one can interpret W and W
an arbitrary functional.
The simplest ones are the fields themselves and in [4] several examples have been
calculated in the one-loop approximation with the outcome that, e.g., (x) transforms
non-locally under W . The closer analysis of this result is the subject of the subsequent
analysis in terms of general Green functions and then of operators.
3.3. Symmetries on quantum fields
For the case of spacetime translations there is no difference to the WessZumino model.
The WI (3.19) can be turned into a local one, formulated on the general Green functions Z
and translated into operator relations analogous to (2.52).
Similarly we rewrite the gauge WI (3.17) as an equation on Z
Z
+ o(),
i Jem Z = ewem (j )Z + i
jB

Z,
+ ji
+ Yi
0=
ji
ji
ji
Yi
(3.29)
(3.30)
and define a conserved electromagnetic current insertion:

ewem (j )Z| Y =0, = i
=0
Z
+ J Z = i jem Z.
jB
We find first of all that the corresponding operator is conserved

Op
0 = jem .
(3.31)
(3.32)
Differentiating once w.r.t. the source jL and performing LSZ reduction we obtain

Op
Op
ieQL (y x)L (y) = i T jem (y)L (x) .
(3.33)
Integration yields the transformation for the field operator
Op
ieQL Op (x) = i Qem , L (x)
where
Qem
(3.34)

d3 y j0em (y),
similarly for all other field operators. In particular

Op
= 0,
i Qem , A (x)
(3.35)
(3.36)
the field A is not charged [11].

Aiming now at the BRS charge we derive the first consequence from the ST identity
directly. (3.13) is rewritten on Z

Z
Z
Z
4
+ i
i
S(Z) d x ijA
jc

jA

Z
Z
Z
Z
ijc 2i i
i
ijc
jA
jc
jB
jc

Z
Z
Z
Z
+ ijB 2i
i i
i

jc
jB
Y
Y
Z
Z
Z
Z
+ ijL
+ ijL
i
i L
+ (L R)
L

YL
YL
YL
Y
L
325
Z
(3.37)
= 0.
If a local ST identity is desired we have to permit the ghosts , , to become local. The
local ST identity will then look as follows

Z
Z
Z
Sloc Z ijA
+ i
i
jc

jA

Z
Z
Z
Z
ijc 2i i
i
ijc
jA
jc
jB
jc

Z
Z
Z
Z
+ ijB 2i
i i
i

jc
jB
Y
Y
2i
+ ijL
Z
Z
Z
Z
+ ijL
i
i L
+ (L R)
L

YL
YL
YL
Y
Z
2i
+ K Z.
= J + K + K
(3.38)
and for constant ghosts integration will indeed

Here J depends on the ghosts , ,
yield zero on the r.h.s. At = = = 0 we are obviously dealing with a local version of
the ordinary ST identity, i.e., J |===0 will correspond to the divergence of the BRS
current. Reduction leads to the operator equation
Op
Op

0 = J ===0 JBRS .
(3.39)
The conserved BRS current leads as usual to a time independent charge via the definition

QBRS = d3 x J0BRS .
(3.40)
Since (3.39) holds on the complete Fock space QBRS commutes with the S-operator on
Fock space [10].
The transformation law for a field operator of type follows from the operator equation
(y x)
Op

Op
eff
= i T J (y)(x) = =0
Y (y)
(3.41)
by integrating over space, d3 y, and time in the interval (x0 , x0 + ) with 0 and
leads to

Op
eff Op
= i QBRS , (x) .
(3.42)

Y (x)
===0
326
(eff /Y (x))Op starts with the non-linear terms given by the tree approximation and
listed in (3.5); higher order corrections appear as they contribute to eff .
For a linearly transforming field, type , the operator law is the one of the classical
approximation
Op
Op
s (x) = i QBRS , (x) .
(3.43)
The charge QBRS is nilpotent.
As is clear from the discussion in Section 2.3 information on supersymmetry follows
from (3.38) by differentiating once with respect to a local susy ghost ( (z) or analogously
(z)). Doing so and performing LSZ reduction we obtain the operator equation

eff Op
Op
Op
+ x (x z)K (x).
0 = x J (x) + i T J (x)
(3.44)
(z)
(z)
Integration over z yields
Op
0 = x J (x) + i

eff Op
d4 z T J (x)
.
(z)
(3.45)
Op
Hence there is a candidate for a susy current, J (x), but it is not conserved. Defining
a charge by

Op
Q (t) d3 x J0 (x)
(3.46)
it will depend on t. Integrating (3.45) over all of x-space leads to
Op

.
0 = dt t Q (t) i QBRS , eff
(3.47)
Taking the time integral for asymptotic times t = and identifying there the charges
we can write

Op
out
in
BRS
.
Q Q = i Q , eff
(3.48)
in
Since Qout
develops out of Q via the time evolution operator S, the scattering operator,
in
Qout
= SQ S ,
(3.48) implies
in

Q , S = i QBRS , eff S .
(3.49)
(3.50)
Here we have used that QBRS and S commute. The interpretation of this result is clear:
the charge Qin
which may be taken to be the generator of susy transformations on the free
in-states does not commute with the S-operator, the reason being the -dependence of eff .
Looking at (3.7) this arises from the gauge fixing term whose -dependent part is as a
consequence of the ghost equation not renormalized. Matrix elements between physical
states however yield a vanishing r.h.s. in (3.50), hence there Qin
is a conserved charge.
327
In the derivation of analogous results for field transformations we concentrate on nonlinearly transforming ones. Differentiating (3.38) w.r.t. and a source for we obtain after
LSZ reduction the operator relation 1

eff eff
2 eff
i
(y
x)T
(y x)
Y (y) (z)
Y (y) (z)

eff
(x)
= iT
J
(y)(x)
J
(y)
(z)
(z)

+ i y (y z)T K (y)(x) .
(3.51)
Unfortunately one cannot straightforwardly integrate (3.51) and conclude how a field
transforms under susy because in the T-product T(J eff ) distributional singularities
at coinciding points may arise. It is clear that every renormalization scheme leads to a
well-defined T-product which is integrable in the sense of distributions but of which type
this distribution is has to be found out. We present this analysis in the appendix and quote
here only its result. One finds that contributions with (y x) arise which cancel with
the operator product T(eff /Y (y)eff / (z)) on the l.h.s., and in addition a term with
a double delta function (x y)(y z) which contributes to the susy variation of the
photino field only. Altogether we arrive at
Y (x)
Op
Op
eff = i Q (x 0 ), (x) ,
Y (x)
Op
Op
eff + B Op (x) = i Q (x 0 ), (x) .
if = ,
(3.52)
(3.53)
Hence the time-dependent susy charge defined in (3.46) (t x0 ) generates a local, nonlinear transformation on all fields. The fact that this transformation does not correspond to
a symmetry on the Fock space, but only on the Hilbert space of the theory is completely
encoded in the time dependence of the charge Q (t) and in the additional B term in
the susy transformation of which vanishes between physical states.
Comparing this result with the explicit one-loop calculation in [4] of S (x) which
yielded a non-local expression we can trace the origin of this non-locality: it comes
from the operator product eff /Y (y)eff / (z) in the l.h.s. of (3.51) which could
be separated in the above analysis and seen to cancel against the (y x) terms on
the r.h.s. By going to the operator level one could isolate the local transformation. The
transformation on vertex type functionals given by S(1) does not distinguish between these
different contributions since it fixes in a sense susy transformations only up to BRS
variations.
The somewhat surprising modification (3.53) of the transformation of may be
understood as follows. We have defined a time-dependent susy charge Q (x 0 ) which
generates susy transformations [Q (x 0 ), (x 0 , x)] at time x 0 . The time dependence of
all these operators is determined by the same unitary time evolution operator in Fock
space, i.e., our susy transformation is compatible with this time dependence. However,
1 The explicit indication Op has been suppressed.
328
the equation of motion for contains contributions from the gauge fixing term which are
not supersymmetric and therefore the time dependence of cannot be compatible with
ordinary susy. This shows that Q (x 0 ) cannot generate the standard susy transformation
of . More explicitly, the equation of motion for has the form
+ i c,
i = F
(3.54)
contains all interaction terms of the classical action (3.3). This equation is
where F
invariant under the combined gauge + susy BRS transformations (3.5), which means to
first order in :

+ i sgauge c ,
i ssusy = ssusy F
(3.55)

where
sgauge = s|===0 ,

susy

s
s
=
+
.
===0
(3.56)
(3.57)
Like in the filtration (3.23), supersymmetry holds only up to gauge-BRS transformations,

in particular we have

.
i ssusy = ssusy F
(3.58)
Since s gauge c = B, (3.55) may be rewritten as

.
i ssusy B = ssusy F
(3.59)
This shows that the modification of the susy transformation law of ensures compatibility
with time evolution.
4. Super YangMills theory

Now we turn to the super YangMills case, assuming a simple gauge group and matter
multiplets in some representation. In order to make the paper self-contained we write down
explicitly the field transformations in the classical approximation
sA = c ig[c, A ] + i i i A ,

s = ig c, + 2i F + i D i ,

s = ig{c, } 2i F + i D i ,
si = igc i + 2 i i i ,

si = +ig c i + 2 i i i ,

si = igc i + 2 Fi 2 i D i i i ,

s i = ig( c)i 2 Fi + 2 i (D i ) i i ,
sc = igc + 2i A i c.
2
(4.1)
(4.2)
(4.3)
(4.4)
(4.5)
(4.6)
(4.7)
(4.8)
329
The transformations of , , c and B are the same as in SQED, (3.5i)(3.5m). The auxiliary
fields D and Fi have to be replaced by their equations of motion.
The classical action
cl = inv + g.f. ,

1
inv = tr F F + iD
/ + matter,
4

+ cA
g.f. = s tr cB
2
(4.9)
(4.10)
(4.11)
is invariant under the generalized BRS transformations. In order to deal with the nonlinearity of the transformations and the fact that they close only on-shell one introduces
external fields and lets the action depend upon them even bilinearly:
= cl + g.f. + ext.f. + bil,
(4.12)

tr YA sA + Y s + Y s + Yc sc + (Y s + Y s + c.c.) , (4.13)
ext.f. =

bil = 12 tr(Y + Y )2 + (Y + Y )2 .
(4.14)
The classical vertex functional (4.12) satisfies the ST identity

+
S( )
(s ) = 0
Y
(4.15)
(here runs over all non-linearly transforming fields, over all linearly transforming
ones).
It has been shown in [4,6,12] that the gauge condition
= B + A,
B
the rigid Ward identity

W =
= 0,
i f
f
(4.16)
(4.17)
all fieldsf
the translational ghost equation
ext
=
(4.18)
and the ST identity (4.15) determine the vertex functional uniquely, if the AdlerBardeen
anomaly is absent and one has supplied suitable normalization conditions.
The aim is now to study the transformation law of the quantum fields under the above
transformations.
330
4.1. Transformations of the quantum fields

Translations are realized as in the WessZumino model and in SQED as
Op
0 = T ,
Op = i[P , ]Op .
(4.19)
The BRS charge

BRS
= d3 x J0BRS
Q
(4.20)
is conserved and the fields transform as
Op
i QBRS , c
= B Op ,
Op eff
=
i QBRS ,
Y
(4.21a)
Op
(4.21b)
B, c, , ,, ). In order to derive the susy transformations we again

( = A , , ,
assume all parameters , , to be local:

Z
Z
Z
Z
Sloc Z ijA
+ jA
ijc 2i i
YA
jA
jA
jc

Z
Z
Z
Z
ijc
+ ijB 2i
i
i
jB
jc
jc
jB
Z
Z
Z
Z
i i
iji
iji

Y
Yi
Y
Y
i
Z
Z
Z
i
i i
2i
i

Yi
Yi

BRS

+ K Z
= J + K + K
(4.22)
i , i ).
( , , i , i : sources for , ,
J depends on the ghosts , , and for constant ghosts integration yields zero on the
r.h.s.
For quite a few steps the calculation runs completely parallel to the abelian case, hence
we can be very brief. The susy current is not conserved, and the time-dependent susy charge
shows the asymptotic behaviour
Op
BRS
in

in
, eff ,
Q , S = i QBRS , eff S .
Qout
(4.23)
Q = i Q
The interpretation of this result is clear: the charge Qin
which may be taken to be the
generator of susy transformations on the free in-states does not commute with the Soperator, the reason being the -dependence of eff . Looking at (4.11) this arises from
the gauge fixing term. Matrix elements between physical states however yield a vanishing
r.h.s. in (3.50), hence there Qin
is a conserved charge.
The transformation law for (non-linearly transforming) fields can be found by
differentiating (4.22) w.r.t. and . After LSZ reduction one obtains the operator relation

eff eff
2 eff
i(y
x)T
(y x)
Y (y) (z)
Y (y) (z)

eff
(x)
J
(y)(x)
J
(y)
(z)
(z)

+ iy (y x)T K (y)(x) .
331
= iT
(4.24)
Since in the T-product T(J eff ) distributional singularities may arise for coinciding
points one cannot straightforwardly integrate (4.24) but has to determine these singularities. As a consequence of the renormalization scheme they are well-defined but have to
be identified. This analysis parallels first the abelian case: contributions proportional to
(y x) arise and are cancelled with the operator product T(eff /Y (y) eff / (z))
on the l.h.s. But then occurs a change: the contributions which lead to a double delta function (x y)(y z) are represented by diagrams
(4.25)
which differ from SQED: the vertex function c receives loop corrections because the
ghost equation of motion is non-trivial. This implies that the local contribution aB to
the transformation s becomes order dependent through a coefficient a. A way out of this
difficulty is to impose as normalization condition
(p c )|p=0,s=1 = 1.
As a consequence of the ghost equation

+ + i
+ 2i c i B = 0,
c
B
YA
(4.26)
(4.27)
one finds

YA

eff
eff
.
=
[ eff ]
YA
YA
c =
(4.28)
(4.29)
The first term on the r.h.s. consists of purely local contributions whereas the second
(a double insertion) is non-local. Hence one can rewrite (4.26) as

eff
eff
= 1,
a2 +
(4.30)
YA
p=0,s=1

i.e., one fixes indeed the coefficient a2 of the counterterm c(i
i ) by this
condition:
a2 = 1 + o(h ).
(4.31)
From tree approximation and with the help of the symmetric differential operators (see
next section) one reads off that it belongs to the same invariant as i/
, the kinetic
332
term. Imposing this normalization condition one can attribute the double delta function
contribution of diagram (4.25) to a local transformation
s = B +
(4.32)
(analogously for ).
With this result one arrives at the same conclusions as in the abelian situation. The term
B in the transformation law of removes the mismatch between the transformation as
given by the above Q (which has been constructed without contribution of the gauge
fixing term) and the time evolution of as given by its equation of motion to which the
gauge fixing term, of course, contributes. Hence we have local and evolution compatible
transformation laws for all fields. The breaking of supersymmetry caused by the gauge
fixing is entirely cast into the time dependence of the susy charge.
4.2. The anomalous dimensions within the vector multiplet
We shall now show that the normalization condition (4.26) has an interesting consequence for the anomalous dimensions within the vector multiplet.
The renormalization group (RG) or CallanSymanzik (CS) equation which are identical
in the massless case follow as usual [13] as differential equations symmetric with respect
to the ST identity and compatible with the gauge condition. A basis of the respective
differential operators is provided by
YA
c + 2 ,
B
NA A
(4.33)
A
YA
B
c
N Y
(4.34)
+ c.c.,
Y
Nc c + Yc
(4.35)
,
c
Yc
Y
N
(4.36)
+ c.c.,
Y
Y
+ c.c.
N
(4.37)
Y
The CS operator can be expanded in this basis and the CS equation then reads

N = 0
C mm + g g
(4.38)
(m comprises all mass parameters, g all couplings of the theory, runs over A, , c, ,
). This implies that a priori the anomalous dimension A of the vector field is not related
to , the anomalous dimension of , .
Testing (4.38) w.r.t. and c and we first find
(mm + g g )c = (A + )c = 0.
(4.39)
333
Hence, acting with (mm + g g ) on (4.26) we obtain

= A .
(4.40)
The normalization condition (4.26) leads to coinciding anomalous dimensions for the
gaugino and the gauge vector field.
The result (4.40) fits to the asymptotic supersymmetry which also followed as a
consequence of it as shown in the last section. It is not in contradiction to previous
calculations [15] because there other normalization conditions have been used and
anomalous dimensions depend quite generally on normalization conditions and gauge
fixing.
Without further inquiries nothing can be said about the anomalous dimensions within
the matter multiplets. It may however very well turn out that a suitable non-linear gauge
fixing could render them equal as well. That would be of importance in models with N = 2
symmetry. For the N = 4 theory, too, the role of gauge fixing and its effect on anomalous
dimensions seems not to have been discussed thoroughly.
5. Discussion and conclusions

From the point of view of algebraic renormalization, super YangMills theories are
gauge theories with additional non-linear, rigid symmetries. This class of models is treated
by introducing constant ghosts for the additional rigid symmetries, coupling all nonlinear field transformations to external fields, and formulating a common SlavnovTaylor
identity for both gauge and rigid invariance [6,7,1720]. The gauge fixing term need
not be (and is not in our case) invariant under the rigid transformations, but it has a
BRS-invariant extension in which the constant ghosts appear explicitly [6,7,12,18]. In
this way, the covariance of the breaking terms can be controlled. By solving the BRS
cohomology, a SlavnovTaylor-invariant vertex functional may be constructed order by
order in perturbation theory. It is not surprising that such an elegant and general formalism
also holds some disadvantage: at first sight, the transformations encoded in the ST operator
look non-local, and it is not clear how they are realized on quantum operators. The BRS
formulation of supersymmetry and gauge invariance intertwines these two symmetries and
obscures the meaning of the separate transformations. The rigid symmetries are explicitly
broken by the gauge fixing term which is nicely incorporated in the functional equations,
but it is not obvious how the breaking terms enter relations on the operator level. The
entangling of gauge and rigid symmetries may be elegantly demonstrated on the level of
functionals by performing a filtration w.r.t. the number of constant ghosts, but this involves
Ward operators for truly non-local transformations, and the implications of this filtration
for operators and states are sketchy.
The main objective of this paper was to examine in detail the implementation of
symmetries on the level of operators as it follows from the BRS construction of the
generating functional. In particular, we have investigated how a supersymmetry charge can
be defined, how it acts on interacting field operators and how the asymptotic charges are
334
related. We have discussed the operator relations on Fock space (i.e., including unphysical
states) as well as on physical Hilbert space (i.e., excluding negative-norm states).
Comparing the results of our analysis for the WessZumino model and SQED and SYM
in the WessZumino gauge we find
Closely analogous ones: translations and gauge transformations obey simple WIs and
can easily be interpreted in terms of operators; susy requires the use of the generalized
ST identity expressing the generalized BRS invariance; its non-linear character does
not really cause harm.
Clearly different ones: in the WessZumino model susy is not broken although nonlinearly realized; its charge exists and is time independent, the transformation is
local; in SQED and SYM susy is broken by the gauge fixing term and the BRS
type formulation only enables one to carry that breaking along in a fashion which
permits renormalization and consistent treatment to all orders (the gauge fixing term
is a BRS variation); the susy charge is time dependent, but in such a way that the
dependence disappears between physical states; remarkably enough there exists still
a local operator expression for the field transformations. The susy transformation of
the vectorino field is modified by the term B which vanishes between physical
states.
In SYM, the modified transformation laws for , can be understood as proper,
local transformations B, B if one imposes a suitable normalization condition (4.26)
which determines the relevant part of gauge fixing. For the fields c,
c one
then has local susy transformations such that asymptotically susy is restored. This
normalization condition also implies that the anomalous dimension of equals that
of the vector field A .
The comparison with the linear realization of supersymmetry which is possible in these
examples yields qualitative agreement. In the WessZumino model the respective charges
) exist and operate on a Hilbert space which is essentially the same as the
(P , Q , Q
one generated without auxiliary fields. The latter can be understood as interpolating fields

which are useful but not necessary. In the massive SQED/SYM the charges P , Q , Q
also exist, likewise the electric charges because there exist conserved currents which are
gauge invariant. All transformations are linear, given by WIs hence the implementation on
the Fock space is straightforward. The transformation laws for operators are local. Here,
of course, the Fock spaces differ tremendously, but one expects (although a detailed proof
seems to be missing) that the Hilbert spaces are equivalent and on them the theories should
and the (true) electric charge Q.
coincide, in particular the charges P , Q , Q
A first comment is in order as far as our rather careless treatment of operators in SQED
and SYM with massless photon and photino is concerned. We have written all formulae
for a completely massless theory. Then, of course, there is no S-matrix and one cannot go
on-shell. Hence we may add mass terms for all fields and thus maintain in general only
translational invariance but neither ordinary BRS nor susy. Our equations should then be
read modulo soft breaking terms which is sufficient for clarifying the role of gauge
fixing and normalization condition (4.26). For models where susy is maintained but the
335
gauge invariance is completely broken down such that all vector fields (and the vectorinos)
get mass our results should be strictly true. Alternatively one could have derived within the
massless theory the desired operator equations by performing the reduction not with respect
to the asymptotic states but with suitable other combinations. With an eye on the linear
realization we did not do so: our main interest was to clarify the role of the gauge fixing
term and the understanding of the breaking of susy which its presence causes in the Wess
Zumino gauge formulation of supersymmetry. The infraredwise existence of operators was
not our concern in this respect our work is formal.
A second comment is appropriate as far as the breaking of susy is concerned. Since
it takes place only in Fock space but not in Hilbert space we have the impression that the
collapse of supersymmetry as described in the paper by Buchholz, Ojima et al. (see [16])
is still avoided for the physical part of the theory.
A third comment finally refers to the type of operator equations which we have derived.
Obviously we worked in perturbation theory and therefore in a very specific representation
of the (anti-)commutation relations. Hence these equations cannot be considered as abstract
ones, valid for any arbitrary representation. 2 In fact, it seems to be an open problem in
which sense they could be lifted to have representation independent meaning. Or stated
differently: it is not known which equations valid in perturbation theory hold also beyond
perturbation theory. On the other hand it is also obvious that the abstract non-perturbative
approach to quantum field theory is not yet able to handle the type of charges dealt with in
this paper.
Acknowledgement
We thank Harald Grosse for helpful discussions.
Appendix A. Singularity structure of some T-products

The starting point for the present analysis is the local ST identity

Sloc Z = J + K + K + K Z
(A.1)
with with Sloc given by (3.38). Since we are mainly interested in the properties of a
non-linearly transforming field, say (x), under supersymmetry we differentiate w.r.t. the
source for and (z). This gives for a general Green function

eff eff
2 eff
X
+
X
(y x)
(z)Y (x)
(z) Y (x)

eff eff
2 eff
(x)Xk +
X
(y xk )
+
(z)Yk (xk )
(z) Yk (xk ) k
k
2 We are very much indebted to H. Grosse for clarifying discussions of this point.
336

(sk (y))
eff

(x)X
(y)
+
s
(x),
X
k
k
k
(z)
(z)
k

J (y)
+
(y
z)
J
(y)
(x)X
=
(z) =0

eff
(x)X .
+ J (y)
(A.2)
(z)

=0

(y xk )
Here denotes X a general string of fields, Xk indicates that the field with index k is missing,
the same for Xk . Fields transform linearly. J |=0 = JBRS is the conserved BRS
current.
The problem is now to identify within

eff eff
2 eff
+T
(y x)
(z)Y (y)
(z) Y (y)

J (y)
(x) + y (y z)T K (y)(x)
=T
(z)

eff
(x)
+ T JBRS (y)
(A.3)
(z)
the singularities which form (y z) or (z x) or both. These singularities determine the
possible contributions from the T-product when it is integrated over y and z. In particular,
we would like to integrate in the order
x0 +
dy0
d z
d3 y
(A.4)
x0
and take the limit 0. Clearly, we can get a contribution only from terms containing a
factor (x y). In the triple insertion

eff
JBRS (y) (x)X ,
(A.5)
(z)
-functions may arise when a suitable number of partial derivatives form a wave equation
operator and act on the corresponding propagator, e.g.,
c c (y
z) = i(y z),
B A (y u) = i (y u),

(y z) = (y z).
(A.6)
(A.7)
(A.8)
In terms of diagrams, the propagator shrinks then to a point and is replaced by a -function.
We first consider the case that JBRS (y) and (x) are directly connected by a line,
leading to a contribution with (x y). Comparing with the WI for ordinary BRS
transformations into which one inserts a vertex eff / by hand it is clear that such
contributions cancel against the double insertion

eff eff
(y x)
(A.9)
X
Y
337
on the l.h.s. of (A.2), including all possible double delta functions (y x)(y z).
However, there is a second possibility how (x y) may be generated: if the vertices
are connected in the order
JBRS (y)
eff
(x)
(z)
we may obtain a double function (x z)(z y) which is equivalent to (x y)(z y)

but is not present in the double insertion (A.9). This can only happen for = (or ),
the relevant contributions being
JBRS = Bc + ,
(A.10)
eff
= ic
+ .
(A.11)
(y z) (x z)
(A.12)
Here, the lines

and
represent propagators c c ,
respectively, A B .
According to (A.6), (A.8), the diagram on the l.h.s. of (A.12) corresponds to a
contribution
(x y)(y z) B(x)X .
(A.13)
We will now further investigate this contribution in order to clarify for which field
configurations X it is really non-vanishing. From the Ward identity (3.17), it follows
immediately that the operator B Op is a free field,
B Op = 0.
(A.14)
Therefore there is only one contribution to the diagram on the right-hand side of (A.12),
namely
+ contact terms.
(A.15)
Reduction w.r.t. A yields zero, but reduction w.r.t. B is non-zero. Thus we see that
the insertion of B (A.13) is indeed nonvanishing, but it is made up entirely of trivial
contributions.
338
Taking into account the above discussions, (A.3) reads

J
2 eff
+ (y z)K (x) + less singular,

(y x)
=T
Y
if = ,
2 eff
=T
(y x)
Y

(A.16)

J
+ (y z)K (x)
+ (y x)(y z) B(x) + less singular.
(A.17)
Here, less singular stands for terms containing no singularity of the type (x y).
Integration in the order (A.4) yields a local susy variation for all fields,
Y (x)
Op
Op
eff = i Q x 0 , (x) ,
Y (x)

Op
Op
eff + B(x) = i Q x 0 , (x) .
if = ,
(A.18)
(A.19)
We still note that these expressions are covariant, i.e., renormalization scheme independent, because they are unique: after the double insertion contributions cancelled each other
the covariance is guaranteed.
References
[1] S.L. Glashow, Partial symmetries of weak interactions, Nucl. Phys. 22 (1961) 579;
S. Weinberg, A model of leptons, Phys. Rev. Lett. 19 (1967) 1264;
A. Salam, in: N. Svartholm, Almquist, Wiksell (Eds.), Proc. of the 8th Nobel Symposium,
Stockholm, 1968, p. 367.
[2] W. Hollik, G. Duckeck, Electroweak Precision Tests at LEP, Berlin, Springer, 2000, p. 161;
M.W. Grunewald, Experimental tests of the electroweak standard model at high energies, Phys.
Rep. 322 (1999) 125.
[3] E. Kraus, Renormalization of the electroweak standard model to all orders, Ann. Phys. 262
(1998) 155, hep-th/9709154.
[4] W. Hollik, E. Kraus, D. Stckinger, Renormalization and symmetry conditions in supersymmetric QED, Eur. Phys. J. C 11 (1999) 365, hep-ph/9907393.
[5] J. Erdmenger, C. Rupp, K. Sibold, Conformal transformation properties of the supercurrent in
four-dimensional supersymmetric theories, Nucl. Phys. B 530 (1998) 501, hep-th/9804053;
J. Erdmenger, C. Rupp, Geometrical superconformal anomalies, hep-th/9809090;
J. Erdmenger, C. Rupp, Superconformal Ward identities for Green functions with multiple
supercurrent insertions, Ann. Phys. 276 (1999) 152187, hep-th/9811209;
J. Erdmenger, C. Rupp, K. Sibold, Superconformal transformation properties of the supercurrent II: Abelian gauge theories, Nucl. Phys. B 565 (2000) 363, hep-th/9907169;
C. Rupp, Conformal transformation properties of the B-type supercurrent, hep-th/9910153.
[6] P.L. White, An analysis of the cohomology structure of super YangMills coupled to matter,
Class. Quant. Grav. 9 (1992) 1663;
P.L. White, Analysis of the superconformal cohomology structure of N = 4 super YangMills,
Class. Quant. Grav. 9 (1992) 413.
339
[7] N. Maggiore, O. Piguet, S. Wolf, Algebraic renormalization of N = 1 supersymmetric gauge

theories, Nucl. Phys. B 458 (1996) 403, hep-th/9507045.
[8] J. Wess, B. Zumino, Supergauge invariant extension of quantum electrodynamics, Nucl. Phys.
B 78 (1974) 1.
[9] O. Piguet, K. Sibold, Renormalizing supersymmetry without auxiliary fields, Nucl. Phys. B 253
(1985) 269.
[10] T. Kugo, Eichtheorie, Springer-Verlag, Berlin, 1997.
[11] W. Zimmermann, Lectures on Electrodynamics, 1978, unpublished.
[12] N. Maggiore, O. Piguet, S. Wolf, Algebraic renormalization of N = 1 supersymmetric gauge
theories with supersymmetry breaking masses, Nucl. Phys. B 476 (1996) 329, hep-th/9604002.
[13] O. Piguet, K. Sibold, Renormalized Supersymmetry. The Perturbation Theory of N = 1
Supersymmetric Theories in Flat SpaceTime, Birkhuser, Boston, 1986.
[14] W. Hollik, E. Kraus, D. Stckinger, Renormalization of supersymmetric YangMill theories
with soft supersymmetry breaking, hep-ph/0007134.
[15] L.V. Avdeev, O.V. Tarasov, A.A. Vladimirov, Vanishing of the three loop charge renormalization function in a supersymmetric gauge theory, Phys. Lett. B 96 (1980) 94.
[16] D. Buchholz, I. Ojima, Spontaneous collapse of supersymmetry, Nucl. Phys. B 498 (1997) 228,
hep-th/9701005.
[17] J.A. Dixon, BRS cohomology of the chiral superfield, Commun. Math. Phys. 140 (1991) 169.
[18] A. Blasi, R. Collina, Renormalization a la B.R.S. of the nonlinear sigma model, Nucl.
Phys. B 285 (1987) 204;
C. Becchi, A. Blasi, G. Bonneau, R. Collina, F. Delduc, F. Delduc, Renormalizability and
infrared finiteness of nonlinear sigma models: a regularization independent analysis for compact
coset spaces, Commun. Math. Phys. 120 (1988) 121.
[19] I.A. Batalin, G.A. Vilkovisky, Relativistic S matrix of dynamical systems with boson and
fermion constraints, Phys. Lett. B 69 (1977) 309;
I.A. Batalin, G.A. Vilkovisky, Quantization of gauge theories with linearly dependent
generators, Phys. Rev. D 28 (1983) 2567;
I.A. Batalin, G.A. Vilkovisky, Phys. Rev. D 30 (1983) 508, Erratum;
I.A. Batalin, G.A. Vilkovisky, Gauge algebra and quantization, Phys. Lett. B 102 (1981) 27.
[20] C. Becchi, A. Rouet, R. Stora, Renormalizable theories with symmetry breaking, in:
E. Tirapegui (Ed.), Field Theory, Quantization and Statistical Physics, Reidel, Dordrecht, 1981;
C. Becchi, A. Rouet, R. Stora, Renormalizable models with broken symmetries, in: G. Velo,
A.S. Wightman (Eds.), Renormalization Theory, Reidel, Dordrecht, 1976, p. 299.

Two-loop renormalization-group analysis of

critical behavior at m-axial Lifshitz points
M. Shpot, H.W. Diehl
Fachbereich Physik, Universitt Essen, D-45117 Essen, Federal Republic of Germany
Received 31 May 2001; accepted 22 June 2001
Abstract
We investigate the critical behavior that d-dimensional systems with short-range forces and
an n-component order parameter exhibit at Lifshitz points whose wave-vector instability occurs
in an m-dimensional isotropic subspace of Rd . Utilizing dimensional regularization and minimal
subtraction of poles in d = 4 + m
2 dimensions, we carry out a two-loop renormalization-group
(RG) analysis of the field-theory models representing the corresponding universality classes. This
gives the beta function u (u) to third order, and the required renormalization factors as well as
the associated RG exponent functions to second order, in u. The coefficients of these series are
reduced to m-dependent expressions involving single integrals, which for general (not necessarily
integer) values of m (0, 8) can be computed numerically, and for special values of m analytically.
The expansions of the critical exponents l2 , l4 , l2 , l4 , the wave-vector exponent q , and the
correction-to-scaling exponent are obtained to order 2 . These are used to estimate their values for
d = 3. The obtained series expansions are shown to encompass both isotropic limits m = 0 and
m = d. 2001 Elsevier Science B.V. All rights reserved.
PACS: 05.20.-y; 11.10.Kk; 64.60.Ak; 64.60.Fr
Keywords: Field theory; Critical behavior; Anisotropic scale invariance; Lifshitz point
1. Introduction
The modern theory of critical phenomena [13] has taught us that the standard ||4
models with an n-component order-parameter field = (i , i = 1, . . . , n) and O(n)
symmetric action have significance which extends far beyond the models themselves: they
describe the long-distance physics of whole classes of microscopically distinct systems
near their critical points. In fact, they are the simplest continuum models representing the
O(n) universality classes of d-dimensional systems with short-range interactions whose
dimensions d exceed the lower critical dimension d (= 1 or 2) for the appearance of a
E-mail addresses: shpot@icmp.lviv.ua (M. Shpot), phy300@theo-phys.uni-essen.de (H.W. Diehl).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 0 9 - 1
M. Shpot, H.W. Diehl / Nuclear Physics B 612 [FS] (2001) 340372
341
transition to a phase with long-range order, and are less or equal than the upper critical
dimension d = 4 (above which Landau theory yields the correct asymptotic critical
behavior). By investigating these models via sophisticated field theoretical methods [4],
impressively accurate results have been obtained for universal quantities such as critical
exponents and universal amplitude ratios.
A well-known crucial feature of these models is their scale (and conformal) invariance at
criticality: the order-parameter density behaves under scale transformations asymptotically
as
(x) x (x)
(1)
in the infrared limit 0, where x = (d 2 + )/2 is the scaling dimension of . This

scale invariance is isotropic inasmuch as all d coordinates of the position vector x are
rescaled in the same fashion.
There exists, however, a wealth of phenomena that exhibit scale invariance of a more
general, anisotropic nature. Roughly speaking, one can identify four different categories:
(i) static critical behavior in anisotropic equilibrium systems such as dipolar-coupled
uniaxial ferromagnets [5] or systems with Lifshitz points [6,7], (ii) anisotropic critical
behavior in stationary states of nonequilibrium systems (like those of driven diffusive
systems [8] or encountered in stochastic surface growth [9]), (iii) dynamic critical
phenomena of systems near thermal equilibrium [10], and (iv) dynamic critical phenomena
in nonequilibrium systems [8].
In the cases of the first two categories, the coordinates x can be divided into two (or
more) groups that scale in a different fashion. Writing x = (x , x ), we call these parallel
and perpendicular, respectively. Instead of Eq. (1) one then has

x , x x (x , x ) ,
(2)
where , the anisotropy exponent, differs from one. Categories (iii) and (iv) involve genuine
time-dependent phenomena for which time typically scales with a nontrivial power of the
length rescaling factor . For phenomena of category (ii), one cannot normally avoid to deal
also with the time evolution. This is because a fluctuation-dissipation theorem generically
does not hold for such nonequilibrium systems; their stationary-state distributions are not
fixed by given Hamiltonians of equilibrium systems and hence have to be determined from
the long-time limit of their time-dependent distributions in general.
Category (i) provides very basic examples of systems exhibiting anisotropic scale invariance whose advantage is that they can be investigated entirely within the framework
of equilibrium statistical mechanics. The particular example we shall be concerned with
in this paper is the familiar continuum model for an m-axial Lifshitz point, defined by the
Hamiltonian

0
0
1
0
u0
( )2 + ( )2 + ( )2 + 2 + ||4 .
H = dd x
(3)
2
2
2
2
4!
Here (x) = (i (x))ni=1 is an n-component order-parameter field where x = (x , x )
Rm Rdm . The operators , , and denote the d = m and d = d m dimensional parallel and perpendicular components of the gradient operator and the associated
342
Laplacian = 2 , respectively. The parameters 0 and u0 are assumed to be positive.

At zero-loop order (Landau theory), the Lifshitz point is located at 0 = 0 = 0.
We recall that a Lifshitz point is a critical point where a disordered phase, a spatially
uniform ordered phase, and a spatially modulated ordered phase meet. For further
background and extensive lists of references, the reader is referred to review articles by
Hornreich [6] and Selke [7] and to a number of more recent papers [1118].
An attractive feature of the model (3) is that the parameter m can be varied. It was studied
many years ago [1922] by means of an expansion about the upper critical dimension
d (m) = 4 + m/2,
m 8.
(4)
The order- results for the correlation-length exponents l2 and l4 , first derived by Hornreich et al. [19], are generally accepted. Yet long-standing controversies existed on the 2
terms of the correlation exponents l2 and l4 and the wave-vector exponent q : Mukamel
[20] gave results for all m with 1 m 8. These agreed with what Hornreich and Bruce
[21] found in the uniaxial case m = 1 via an independent calculation, but were at variance
with Sak and Grests [22] for m = 2 and m = 6 (who investigated only these special cases).
More recently, Mergulho and Carneiro [13,14] presented a reanalysis of the problem
based on renormalized field theory and dimensional regularization. Treating explicitly only
the cases m = 2 and m = 6, they recovered Sak and Grests results for l2 and l4 , but did
not compute q . They analytically continued in d rather than in d and, fixing the latter at
d = 4 m/2 while taking the former as d = m , with m = 2 or 6, they also derived
the expansions of the correlation-length exponents l2 and l4 to order 2 . The -expansion
results they gave in Ref. [14] for l2 and l4 with m = 2 and m = 6 differ from ours to be
given below because of two minor computational errors (see Section 4.2 and Note added
in proof on p. 362).
The purpose of the present paper is to give a full two-loop renormalization group (RG)
analysis of the model (3) for general, not necessarily integer values of m (0, 8) in d =
d (m) dimensions. 1 As a result we obtain the expansions of all critical exponents
l2 , l4 , q , l2 , and l4 to order 2 . In a previous paper [18], hereafter referred to as I, we
have shown how to overcome the severe technical difficulties that had hindered analytical
progress in this field and prevented a resolution of the above-mentioned controversy for
so long. Working directly in position space and exploiting the scale invariance of the free
propagator at the Lifshitz point, we were able to compute the two-loop graphs
d of the two(2)
,
its
analog
with
an
insertion
of
d x( )2 /2.
point vertex function (2) and (
2
)
Together with one-loop results, these suffice for determining the exponents l2 , l4 , and q
1 Another two-loop calculation was recently attempted by de Albuquerque and Leite [23,24]. In their evaluation
of two-loop graphs e.g., of the last graph of (4) shown in Eq. (B.7) they replaced the integrand of
the double momentum integral by its value on a line. We fail to see why such a procedure, by which more
or less arbitrary numbers can be produced, should give meaningful results. Let us also emphasize that using
such approximations leads to the following problem: unless the corresponding approximations are made for
higher-loop graphs involving this two-loop graph as a subgraph, pole terms that cannot be absorbed by local
counterterms are expected to remain because they will not be canceled automatically through the subtractions
provided by counterterms of lower order.
343
to order 2 . In order to obtain the correlation-length exponents l2 and l4 to this order in

(2)
we must compute the two-loop graphs of the four-point vertex function (4) and of 2 .
Our results are of importance to recent work on the generalization of conformal
invariance to anisotropic scale invariant systems [2527]. Some time ago Henkel [25]
proposed a new set of infinitesimal transformations generalizing scale invariance for
systems of this kind with an anisotropy exponent = 2/, = 1, 2, . . . . He pointed
out that the case = 4, = 1/2, is realized for the Lifshitz point of a spherical (n
) analog of the ANNNI model [28,29], and that the same m-independent value of in
Ref. [28] was found to persist to first order in for the Lifshitz point of the model (3).
However, as can be seen from Eq. (66) [and Eq. (84) of I], deviates from 1/2 at order
2 . This shows that the Lie algebra discussed in Ref. [25] cannot strictly apply below the
upper critical dimension ( > 0) if n is finite, except in the trivial Gaussian case u0 = 0.
The remainder of this paper is organized as follows. In the next section we recapitulate
the scaling form of the free propagator for 0 = 0 = 0. We give the explicit form of its
scaling function as well as those of similar quantities, and discuss their asymptotic behavior
for large values of their argument. These informations are required in the sequel since the
expansion coefficients of our results for the renormalization factors and critical exponents
can be expressed in terms of single integrals involving these functions.
In Section 3 we specify our renormalization procedure and present our two-loop results
for the renormalization factors. Our -expansion results for the critical, correction-toscaling, and crossover exponents are described in Section 4. Utilizing these we determine
numerical estimates for the values of these exponents in d = 3 dimensions, which
we compare with available results from Monte Carlo calculations and other sources.
Section 5 contains a brief summary and concluding remarks. In Appendixes AE various
calculational details are described.
2. Scaling functions of the free theory and their asymptotic behavior

2.1. The free propagator and its scaling function
Following the strategy utilized in I, we employ in our perturbative renormalization
scheme the free propagator with 0 = 0 = 0. In position space, it is given by

eiqx
G(x) = G(x , x ) =
(5)
.
2 + q4
q
0
q
Here x = |x | and x = |x | are the Euclidean lengths of the parallel and perpendicular
components of x, and we have introduced the notation

m

d q
ddm q
with
and
(6)
(2)m
(2)dm
q
q q
Rm
Rdm
for integrals over momenta q = (q , q ) Rm Rdm . Whenever necessary, these

integrals are dimensionally regularized.
344
1/4
Rescaling the momenta as q 0
x q and q x q yields the scaling form
[18] (cf. Refs. [25,29])
1/2
2+ m/4 1/4
0
0
x x
G(x , x ) = x
(7)
with the scaling function

() (; m, d) =
q q
eiq e eiq
,
2 + q4
q

(8)
where e is an arbitrary unit d -vector while stands for the dimensionless d -vector
1/4
= 0
1/2
x x
(9)
In I the following representation of () in terms of generalized hypergeometric functions

was obtained:

() = 22m 2 (),

1 2
1 1 m 4

() = 1 m 1F2 1 ; , + ;
2 2 2
4 64
2+ 4
3

2
m 4
22
3 3

;
,
1
+
;
F
,
1
2
4 1 + m4
2 2 2
4 64
d1
(10)
(11)
with = 4 + m2 d. Upon expanding the hypergeometric functions in powers of 4 and

resumming, one arrives at the Taylor expansion

2 k
1 1 2 + k2

()
=
(12)
.
1 m k
k! 2 + 4 + 2
4
k=0
can be written in the form

The result tells us that

1
2
1 m 1

+ ,
() =1 1 1 ,
(13)
;
;
,
2 2
2
4 2
4
where 1 1 is a particular one of the FoxWright functions (or Wright functions) p q
[3034], further generalizations of the generalized hypergeometric functions p Fq whose
series representations are given by
p

1 i=1 (ai + Ai k) k
(14)
x .

p q {(ai , Ai )}, {(bj , Bj }; x =
k! qi=1 (bj + Bj k)
k=0
In the sequel, we shall need the asymptotic behavior of () as . This may be
inferred from theorems due to Wright [31,32] about the asymptotic expansions of the
functions p q . We discuss this matter in Appendix A, where we show that the asymptotic
expansion these theorems predict for nonexceptional values of m and ,
4+2
4k
(1)k 2 (2 + 2k)

( )
(15)
,
2
k! ( m4 12 + 2 k) 2
k=0
follows from the integral representation (8) in an equally straightforward manner as the
Taylor expansion (12). Nonexceptional values of m and are characterized by the property
that none of the poles which the nominator of the coefficient

1 2 + k2
f (k) = 1 m k
2+ 4 +2
345
(16)
of the power series (12) has at

k = kl 2l,
l = 1, 2, . . . ,
(17)
gets canceled by a pole of the denominator. If = 0, the only values among m = 1, 2, . . . , 7

for which such cancellations occur are m = 2 and m = 6. More generally, this happens for
d = m + 1 and d = m + 3 where the expansion (15) terminates after the first (k = 0) term
and vanishes identically, respectively. In accordance with Wrights theorems, corrections
to these truncated expansions are exponentially small. In fact, in these two cases ()
reduces to the much simpler expressions

m 2 2
u2m
,
,
d = m + 1,
() =
(18)
8 m/2
2
4
and
() = (4)
m+2
2
2 /4
d = m + 3,
(19)
where (a, x) is the incomplete gamma function. These equations comprise two cases
where d becomes the upper critical dimension (4), namely (d, m) = (6, 7) and (d, m) =
(2, 5). In the former, Eq. (18) simplifies to

2 2 /4
1 1
e
1 1+
,
m = 6, d = d = 7.
() =
(20)
4
(2)3 4
This result as well as Eq. (19) with m = 2 were employed in I, where we also derived the
leading term (k = 0) of the asymptotic series (15).
For general values of m and d = d , the scaling function (; m, d) () can be
written as
1 2+m 4

m 2
2 1 4
1
1F2 1; 2 , 4 ; 64
I m4
(; m, d ) =

2+m
6+m
8
4
4
22+m 4
2 1 m4 2

2

1
1
m
m
=
I4
,
L4
2+m +
6+m
8
4
4
2+m
2
4 4
(21)
where L (z) and I (z) are modified Struve and Bessel functions, respectively, [35].
The second form is in conformity with the one given by Frachebourg and Henkel
[29] for the case m = 1. These authors encountered this (and similar) scaling functions
when studying Lifshitz points of order L 1 of spherical models. They also analyzed the
large- behavior of these functions, verifying explicitly the asymptotic forms predicted
by Wrights theorems. If we let a = 1 and set x = 2 /4, their scaling function denoted
346

(a, x) corresponds precisely to our (),
and the asymptotic expansion they found is
consistent with ours in Eq. (15) and the large- form (22) presented below. 2
Finally, let us explicitly give the large- forms of (, m, d ) as implied by Eq. (15):

m2
1
24(m 6)
1m 6+m
4
(; m, d ) 2
m+2 4
8
4

16
960(m 10)(m 6)
+
+
O
(22)
.
12
In accordance with our previous considerations, all terms or all but the first one of this
series vanish when m = 2 or m = 6, respectively.
2.2. Other required scaling functions of the free theory
Besides , our results to be given below involve two other scaling functions. One is the
function (; m, d) of I. This is defined through
1/2
1+ (m+2)/4 1/4
( G G)(x) = x
(23)
,
0
0
x x
d
where the asterisk indicates a convolution, i.e., (f g)(x) d x f (x x )g(x ). The
explicit form of (; m, d ) may be gleaned from Eq. (A.5) of I, where it was given in
terms of Bessel and hypergeometric functions. This can be written more compactly as

2
2
m
2 2
I m4
(; m, d ) =
(24)
L m4
.
m
4+m
4
4
4
4
26+ 4 4
The asymptotic expansion of the difference of functions in the square brackets of this
equations follows from Eqs. (12.2.6) and (9.7.1) of Ref. [35], implying

m 2 2
6 m 42
2m 6+m
4
1+
(; m, d ) 2
m+2
2 4
4

12
6 m 3(10 m) 44
+
+O
(25)
.
2
2
8
The leading term 2 was already given in I. Note also that again all terms or all but
the first one of this series vanish when m = 2 or m = 6, respectively, as is borne out by the
explicit forms
e /4
1
(; 2, 5) = (; 2, 5) =
2
32 2
2
(26)
and
1 e /4
.
(4)3 2
2
(; 6, 7) =
(27)
2 Note that the Hamiltonians of the spherical models considered in Ref. [29] involve instead of ( )2 a

2 2
m
derivative term of the form m
i=1 (i ) , which breaks the rotational invariance in the parallel subspace R .
Comparisons with our results for the free theory are therefore only possible for L = 2 and m = 1.
The third scaling function we shall need is defined through the Taylor expansion

(k)
2 k
.
(; m)
1 2m k
4
k! 2 + 4 + 2
347
(28)
k=1
It can be expressed in terms of generalized hypergeometric functions p Fq by summing the

contributions with even and odd k separately. One finds

1
3 m 4
3
4
(; m) =

2F3 1, 1; , 2, + ;
32 32 + m4
2
2
4 64

2
m 4
1 3
(29)
; ,1 + ;
.
1F2
4 (1 + m4 )
2 2
4 64
Details of how this function arises in the computation of the Laurent expansion of the fourmay be found in Appendix C. In Appendix D we show that (, m)
point graph
behaves as

m+2
1
4
m2
+
(; m)
CE 8 4
ln

2+m
16
4
4

(m 2)(m 6)
+ 96
(30)
+ O 12
8
in the large- limit, where CE 0.577216 is Eulers constant and while (x) denotes the
digamma function.
For the special values m = 2 and m = 6, the function (; m) reduces to a sum of
elementary functions and the exponential integral function E1 (x). As can be easily deduced
from the series expansion (28), one has
2

2
+ E1
(, 2) = 2 CE + ln
(31)
4
4
and
2
2
2
8e /4
4
e /4 1
+
2E1
+
32
16
4
2
4
6, 7),
= 1 + (; 2) (;
(, 6) = 1 2CE ln
(32)
respectively.
3. Renormalization
3.1. Reparametrizations
From I we know that the ultraviolet singularities of the N -point correlation functions

N
=1 i (x ) of the Hamiltonian (3) can be absorbed via reparametrizations of the form
= Z 1/2 ren ,
(33)
348
0 0c = 2 Z ,
(34)
0 = Z ,
(35)
u0 0 m/4 Fm, = Zu u,
(36)
and
1/2
(0 0c )0
= Z .
(37)
Here is an arbitrary momentum scale. The critical values of the Lifshitz point, 0c and
0c , vanish in our perturbative RG scheme based on dimensional regularization and the
expansion. The factor Fm, serves to choose a convenient normalization of u. A useful
as
choice is to write the following one-loop integral for

Fm,
1
.
=
(38)
4
2
4
2

(q + q )[q + (q + e ) ]
q
This integral is evaluated in Appendix B. The result, given in Eq. (B.13), yields

1 + 2 2 1 2 m4
8+m2
4

Fm, = (4)
(2 ) m2

m

2
(4)2 4 m4

=
1 + (2 CE + ln 4) + O .
2
m2
(39)
3.2. Renormalization factors

With this different choice of normalization of the coupling constant, our results of I for
Z , Z , and Z translate into

n + 2 j (m) u2
+ O u3 ,
3 12(8 m)

n + 2 j (m) u2
Z Z = 1 +
+ O u3 ,
3 96m(m + 2)
Z = 1
(40)
(41)
and
Z Z Z1/2 = 1 +

n + 2 j (m) u2
+ O u3
3
8m
(42)
with

j (m) Bm
d m1 3 (; m, d ),
(43)

j (m) Bm
0
d m+3 3 (; m, d ),
(44)
349
and

j (m) Bm
d m+1 2 (; m, d ) (; m, d ).
(45)
Except for the factor Bm , which is

Bm
S4 m2 Sm
2
Fm,0

3m
210+m 6+ 4 m2
=
2 ,
2 m4 m4
(46)
the coefficients j (m), j (m), and j (m) are precisely the integrals denoted, respectively,
as J0,3 (m, d ), J4,3 (m, d ), and I1 (m, d ), in I. The quantity Sd = 2 d/2/ (d/2) in
Eq. (46) (with d = m, e.g.) means the surface area of a d-dimensional unit sphere.
Our two-loop results for the remaining Z-factors, Zu and Z , can be written as

n+2 n+5 1
n+2 u
Ju (m) u2
+
+ O u3
Z Z = 1 +
(47)
3 2
3
6 2
2
2
and
Zu Z2 Zm/4 = 1 +
n + 8 3u
+
9 2
n+8 3
9 2
2
3

5n + 22 Ju (m) 2
u + O u3 ,
27
2
(48)
with

CE + 2
2
where ju (m) means the integral
Ju (m) = 1
Bm
ju (m) = 4+m (6+m)/4
2
m
4

+ ju (m),
d m1 2 (; m, d )(; m).
(49)
(50)
For the values m = 2 and m = 6, the above integrals j , j , j , and ju can be computed
analytically (cf. I and Appendices B and C). This gives

4
4
8
j (6) = 1 3 ln ,
j (2) = ,
(51)
3
3
3
128
448
j (2) =
(52)
,
j (6) =
,
27
9

8
8
4
j (2) = ,
(53)
j (6) = 1 + 6 ln ,
9
3
3

128
3
1
ju (6) =
+ ln
ju (2) = ln ,
(54)
,
2
6
27
and
4
4
5
Ju (6) = 3 ln .
Ju (2) = ln ,
(55)
3
6
3
For other values of m we determined these integrals by numerical integration in the manner
explained in Appendix E. The results are listed in Table 1.
350
Table 1
Numerical values of the integrals j (m), . . . , Ju (m)
m
j
j
j
ju
Ju
1.642(9)
1.339(4)
0.190(6)
0.203(7)
0.383(8)
1.33333
4.74074
0.88889
0.40547
0.28768
1.055(6)
10.804(3)
1.999(9)
0.624(2)
0.200(8)
0.803(7)
20.067(7)
3.464(1)
0.880(1)
0.119(8)
0.57(4)
32.95(4)
5.23(4)
1.21(1)
0.04(3)
0.36521
49.77778
7.26958
1.72286
0.02971
0.17(4)
70.74(7)
9.53(6)
2.92(4)
0.09(9)
3.3. Beta function and fixed-point value u

From the above results for the renormalization factors the beta function
u (u) |0 u
(56)
and the exponent functions

|0 ln Z ,
= , , , , u,
(57)
can be calculated in a straightforward manner. Here |0 denotes a derivative at fixed bare

values 0 , 0 , 0 , and u0 . Since we employed minimal subtraction of poles, the exponent
functions satisfy the following simple relationship to the residua (Res) of the Z factors:
(u) = uu Res=0 [Z (u)],
= , , , , u.
(58)
For the beta function, which is related to u via u (u) = u[ + u (u)], we obtain

n+83 2
5n + 22
u 3
Ju (m)
u (u) = u +
9 2
27

1 n + 2 j (m)
j (m) u3 + O u4 .
+
24 3
8(m + 2)
(59)
Upon solving for the nontrivial zero of u , we see that the infrared-stable fixed point is
located at

3
8 2
2 9
5n + 22
9
+
Ju (m)
u =
3
3 n+8
27 n + 8
27

1 n + 2 j (m)
j (m) + O 3 .
+
24 3
8(m + 2)
(60)
We refrain from giving the resulting lengthy expressions for the exponent functions here.
The values (u ) of these functions at the infrared-stable fixed point are presented
in Eqs. (61)(66).
351
4. Critical exponents
4.1. Analytic -expansion results
The fixed-point value (60) can now be substituted into the exponent functions (58) that
are implied by our results (40)(42) and (47) for the renormalization factors to obtain the
universal quantities = (u ), = , , , and . Recalling how these are related to the
critical exponents (cf. I), one arrives at the expansions

n + 2 2j (m) 2
+ O 3 ,
(n + 8)2 8 m
+

j (m)
n+2
=
l4 = 4
2 + O 3 ,
2 +
(n + 8)2 2m(m + 2)

n+2
1
n+2
n + 2 j (m)
7n + 20

Ju (m) +
2 = =
4
2
l2
n+8
2(n + 8)
n+8
n + 8 8(m + 2)

m(n + 2) + 4(4 n)
j (m) 2 + O 3 ,
+
(n + 8)(8 m)

1 n+2
1 n+2
2 +
7n + 20
= +
+
Ju (m)
n+2+4
l4 =
4(2 + ) 4 n + 8 8 16 (n + 8)2
n+8

j (m)
mn+2
n+2
2 + O 3 ,
j (m) 1
n+8
4 n + 8 2m(m + 2)
l2 = =
and

3j (m) j (m) 2
n+2
j (m)
1 = =
+ O 3 .
l2
(n + 8)2 8m(m + 2)
m
8m
The anisotropy exponent = l4 /l2 of Eq. (2) is given by

1
n+2
j (m) 2
2 +
j (m)
=
+
=
+ O 3 ,
2
4
2 2(n + 8) 8m(m + 2) 8 m
and for the correction-to-scaling exponent, we obtain

36 2
5n + 22

Ju (m)
l2 u (u ) =
3
2
(n + 8)
27

1 n + 2 j (m)
+
j (m) + O 3 .
24 3
8(m + 2)
(61)
(62)
(63)
(64)
(65)
(66)
(67)
Our rationale for denoting the latter analog of the usual Wegner exponent as l2 is the
following: it governs those corrections to scaling that are weaker by a factor of l2
/
| |l2 l2 than the leading infrared singularities. Since l2 l2 it is natural to

introduce also the related correction-to-scaling exponent
l2
l4
(68)
.
In the case of an isotropic Lifshitz point (cf. Section 4.5), in which only the correlation
length is left, this exponent retains its significance and becomes the sole remaining
analog of Wegners exponent.
352
4.2. The special cases m = 2 and m = 6

For these two special cases, a two-loop calculation was performed in Ref. [14]. In order
to compare its results with ours, we must recall that these authors took d = m and
d = 4 m/2, with m = 2 and m = 6. Accordingly, our must be identified with = 2 .
If we substitute the analytic values (51)(55) of the integrals j , . . . , Ju into our results
(61)(64), the latter reduce to
3
4 n+2 2
(69)

+
O
,
9 (n + 8)2

8 n+2 2
l4 (m = 2) =
(70)
+ O 3 ,
2
27 (n + 8)

n+2
2(n + 2) n2 131n 35 7n + 20 4 2
1
l2 (m = 2) = +
+
+
+
+
ln

2 4(n + 8)
216
27
4
3
(n + 8)3 16
l2 (m = 2) =
+ O( 3 ),
(71)

2
n+2 n
115n 19 7n + 20 4 2
1 n+2
+
+
+
+
ln
l4 (m = 2) = +
4 n + 8 8 (n + 8)3 16
216
27
4
3
+ O( 3 ),
and
(72)

4 n+2 2
1
ln
(73)
+ O 3 ,
2
3
3 (n + 8)

14 n + 2 2
+ O 3 ,
l4 (m = 6) =
(74)
2
27 (n + 8)

n+2
2(n + 2) n2 331n 547 23n + 88 4 2
1

+
+
+
ln
l2 (m = 6) = +
2 4(n + 8)
(n + 8)3 16
144
72
4
3

+ O 3 ,
(75)

2
1 n+2
n+2
835n 1009
4 2
n
l4 (m = 6) = +
+
+
+
(19n
+
56)
ln

4 n + 8 8 4(n + 8)3 4
108
54
3

+ O 3 ,
(76)

l2 (m = 6) = 8
respectively. Our results (69), (70), (73), and (74) for the correlation exponents are
consistent with Eqs. (57), (56), (61), and (60) of Ref. [14]. However, the 2 terms of the
correlation-length exponents given in its Eqs. (58), (59), (62), and (63) are incompatible
with ours. These discrepancies have two causes. There is a sign error in Mergulho and
Carneiros [14] definition (50): The term in its second line should be replaced by its
negative; only then are their general formulas (54) and (55) for the correlation-length
exponents l ( our l4 ) and l ( our l2 ) correct. With this correction, these formulas
yield results in conformity with ours if m = 6. However, in the case m = 2, another
correction must be made: we believe that in their Eq. (C.7) for the integral I3 the prefactor
of the multiple integral on the right-hand side is too small by a factor of 2 . This entails that
the logarithm term ln(16/3) of the 1 pole in their Eq. (C.10) gets modified to ln(64/3).
353
With this additional correction, the 2 terms following for m = 2 from their corrected
Eqs. (54) and (55) turn out to be consistent with ours.
4.3. Numerical values of the second-order expansion coefficients
In order to analyze further the above results, let us denote the coefficients of the 2 terms
()
of the exponents = l2 , l4 , . . . , as C2 (n, m), so that, for example,
l2 =

1 1n+2
( )
+
+ C2 l2 (n, m) 2 + O 3 .
2 4n+8
(77)
The numerical values of these coefficients for m = 1, . . . , 6 are listed in Table 2 for the
case n = 1.
In deriving the coefficients of the critical exponents l , l , and l , we utilized the
familiar hyperscaling and scaling relations [7,28]
l = 2 (d m)l2 ml4 ,
l2
l4
(d m 2 + l2 ) +
m,
l =
2
2
(78)
(79)
and
l = l2 (2 l2 ) = l4 (4 l4 ),
(80)
respectively. Specifically, the first one, Eq. (78), in conjunction with Eqs. (63) and (64)
yields
l =

4n
( )
+ C2 l (n, m) 2 + O 3 ,
n+82
with
( )
C2 l (n, m) =
(81)

m
1n+2
( )
( )
4
C2 l2 (n, m) mC2 l4 (n, m).
4n+8
2
(82)
Table 2
()
Numerical values of the second-order expansion coefficients C2 (n = 1, m)
m
(l2 )
C2
( )
C2 l4
( )
C2 l
( )
C2 l
( )
C2 l
()
C2
( )
C2 l
0.043
0.037113
0.032158
0.027744
0.023677
0.01987
0.01626
0.090
0.006
0.015867
0.013335
0.011083
0.009010
0.00707
0.00524
0.062429
0.039810
0.019277
0.000059
0.01817
0.03564
0.00155
0.008137
0.014196
0.019926
0.02541
0.03070
0.077
0.065533
0.056085
0.047668
0.039911
0.03265
0.02576
0.023210
0.004723
0.011535
0.026221
0.03965
0.05203
0.482459
0.361628
0.253233
0.152736
0.05818
0.03190
0.630
354
()
Fig. 1. Second-order coefficients C2 (n, m) of the exponents = l2 (), l (), l (), and l
() for n = 1 and m = 0, . . . , 7.
Note that, to first order in , the expansions of the critical exponents are independent
of m. This means that one can set m = 0. Hence the expansions to first order in of
all critical exponents of the Lifshitz point that have well-defined analogs for the usual
isotropic (m = 0) critical theory coincide with those of the latter, which can be looked up
in textbooks [36]. Examples of such critical exponents are l2 , l , l , and l .
The source of this m independence is the following. The operator product expansion
(OPE) of the theory considered here, for > 0, is a straightforward extension of the
familiar one of the isotropic (m = 0) 4 theory. Proceeding by analogy with Chapter 5.5 of
Ref. [37], one can convince oneself that the O() corrections to the critical exponents
are given by simple ratios of OPE expansion coefficients. These do not require the
explicit computation of Feynman graphs but follow essentially from combinatorics. For
critical exponents with an m = 0 analog, this has the above-mentioned consequence. The
difference between the cases of a Lifshitz point and of a critical point manifests itself in
the O() expressions of these exponents only through the modified, m-dependent value of
= 4 d + m/2, a difference that disappears for m = 0.
In Fig. 1 the coefficients C2() of the exponents = l2 , l , l , and l for the case n = 1
are displayed as functions of m. The results indicate that these coefficients depend in a
smooth and monotonic fashion on m, approaching the familiar isotropic m = 0 values
linearly in m. Owing to the above-mentioned independence of the O() expressions of
these critical exponents, this behavior carries over to the numerical estimates one gets for
the critical exponents in three dimensions by extrapolation of our O( 2 ) results. We shall
see this explicitly shortly (see Section 4.6). However, before turning to this matter, let us
briefly convince ourselves that our analytic two-loop series expressions for those critical
355
exponents, renormalization factors, etc. that remain meaningful for m = 0 go over into the
corresponding well-known results of the usual isotropic 4 theory for a critical point.
4.4. The limit m 0
In this limit, d perpendicular coordinates, but no parallel ones remain. Hence x can
be identified with x , and the free propagator (7) becomes
1
G(x) = x 2+ ( = 0; m = 0, d = 4 ) = d/2 x 2+ .
(83)
4
The = 0 value of () can be read off from the k = 0 term of the Taylor series (12).
The expression on the far right of Eq. (83) is indeed the familiar massless free propagator
[]1 (x).
We must now determine the m 0 limits of the integrals j , j , j , and Ju (i.e., ju ),
in terms of which our analytic results given in Sections 3.2, 3.3, and 4.1 are expressed. We
assert that the correct limiting values are
1
Ju (0) = .
(84)
2
m

To see this, note that m-dimensional integrals d fm () = Sm 0 d fm () m1
should approach their zero-dimensional analogues, namely f0 ( = 0), as m 0. In the
2 = 2. The respective values f (0) for j
case of j , we have f0 (0) = S4 3 (0; 0, 4)/F0,0
0
and j vanish because of the explicit factors of 4 and 2 appearing in their integrands.
On the other hand, the vanishing of ju (0) is due to the factor () of its integrand and
the fact that (0) = 0 according to the Taylor series (28). Finally, the value of Ju (0) given
above follows from ju (0) = 0 via Eq. (49).
If one sets m = 0 in the series expansions of quantities whose analogs retain their significance in the case of the usual critical-point theory, e.g., in Eqs. (40), (47), (48), (59), (60),
m/4
(63), and (67) for Z , Z Z , Zu Z2 Z , u , u , , and l2 , utilizing the above values
(84) of the integrals, one recovers the familiar two-loop results for the standard 4 theory.
Let us also mention that (because of the factor Sm m in Bm ) the integrals j , j , and
1/2
ju vanish linearly in m as m 0. Therefore, quantities like Z Z , Z Z Z , l4 , l4 ,
, , etc. that involve ratios such as j (m)/m have a finite m 0 limit. It is conceivable
that the m 0 limits of these quantities might turn out to have significance for appropriate
problems. However, we shall not further consider this issue here.
j (0) = 2,
j (0) = j (0) = ju (0) = 0,
4.5. The case of the isotropic Lifshitz point

In the case of an isotropic Lifshitz point one has d = d and d = 0. In a conventional
expansion one would expand about d (8) = 8, setting d = d = 8 . Results to order
2 that have been obtained in this fashion for the correlation exponents l2 and l4 and the
correlation-length exponents l2 and l4 can be found in Ref. [19]. 3
3 We have checked these results by means of an independent calculation, using dimensional regularization and
minimal subtraction of poles.
356
Since the constraint d = d implies that both d and d vary as is varied, it may not be
immediately clear that results for this case can be extracted from our expansion at fixed
d = m. Let us choose a fixed m = 8 and utilize the expansion. This yields

(n, m, d) = (0) + (1) (n) + (2) (n, m) 2 + O 3
(85)
with = 8 d /2, where means any of the critical exponents considered in
Section 4.1 that remain meaningful in the case of an isotropic Lifshitz point, such as l4 ,
l4 , , l , q , and l4 . As indicated in Eq. (85), the coefficients of the terms of orders
0 and do not depend on m. We now set d = m, which implies that = /2. The
upshot is the following: in order to obtain from our -expansion results the dimensionality
expansions of the critical exponents of the isotropic Lifshitz point about d = 8 to second
order, we must simply replace the second-order coefficients (2) (n, m) by their limiting
values (2) (n, 8 ) and identify with /2 = (8 d)/2.
The limiting values of the integrals j , j , j , and Ju are
1
Ju (8 ) = .
(86)
6
To see this, note that the factor Bm appearing in j , . . . , ju varies (8 m) near m = 8. In
the case of j , the integral that multiplies Bm has a finite m 8 limit, so j (8 ) vanishes.
By contrast, the corresponding integrals pertaining to j and j have a pole of first order
at m = 8. We have in the case of j ,
j (8 ) = 0,

d
m+3
j (8 ) = 96,

; m, d (m) =
j (8 ) = 12,

dw w7m f (w)[1 + O(8 d)]

f (0)
+ O (8 m)0 ,
8m
(87)
where f (w) is the function f (w) [w4 (1/w; 8, 8)]3 whose value f (0) = (2)12
follows from Eq. (22). Using this together with B (8) = 3 217 12 yields the above
result for j (8 ). The value of j (8 ) follows in a completely analogous manner.
The computation of Ju (8 ) is somewhat more involved because the integral giving
ju /Bm has poles of second and first order. The second-order pole, due to the appearance
of the term ln in the large- form (30) of the scaling function (), produces a firstorder pole in ju , which cancels the pole resulting from the contribution 12 (2 m4 ) to Ju
[cf. Eq. (49)]. The first-order pole of the integral results in the finite value (86) of Ju (8 ).
Upon substituting the values (86) into the respective -expansion results of Section 4.1
and setting 2 = = 8 d, we recover indeed the results of Ref. [19] for the correlation
exponent l4 and the correlation-length exponent l4 :
l4 (m = d) =

3 n+2 2
+ O 3 ,
20 (n + 8)2
8 d,
(88)
and
l4 (m = d) =

n+2
1
(n + 2)(15n2 + 89n + 4) 2
+
+
+ O 3 .
3
4 16(n + 8)
960(n + 8)
(89)
357
Furthermore, we can infer the previously unknown series expansions of the remaining
exponents of the isotropic Lifshitz point. Specifically for the wave-vector exponent, we
find that

2 +
1 21 n + 2 2
q (m = d) =
(90)
+ O 3 .
= +
2
4(1 + ) 2 40 (n + 8)
Another significant exponent is the crossover exponent . Its 8 d expansion follows from
Eq. (90) via the scaling law = l4 /q . The one of the correction-to-scaling exponent (68)
becomes

202 + 41n 2
+ O 3 .
l4 (m = d) = +
(91)
2
30(n + 8)
4.6. Series estimates of the critical exponents for d = 3 dimensions
We now wish to exploit our -expansion results of the foregoing subsections to obtain
numerical values of the critical exponents in d = 3 dimensions. We shall mainly consider
the cases of uniaxial Lifshitz points (m = 1) for order-parameter dimensions n = 1, 2, 3 and
of biaxial (m = 2) Lifshitz points for n = 1. Of particular interest is the case m = n = 1,
which is realized by the Lifshitz point of the ANNNI model [7,38] and is encountered in
many experimental systems.
Cases with m 2 are of limited interest whenever n 2, for the following reason. If
a Lifshitz point exists, then low-temperature spin-wave-type excitations whose frequen2 as q 0 must occur. By analogy with the MerminWagner
cies vary as q = 0 q 4 + q
theorem [39] one concludes that such excitations would destabilize an ordered phase in
dimensions d d (m) = 2 + m/2, ruling out the possibility of a spontaneous breaking of
the O(n) symmetry at temperatures T > 0 for such values of d. (This conclusion is in complete accordance with Grest and Saks work [40] based on nonlinear sigma models.) Hence
in three dimensions one is left with the case m = 1 of a uniaxial Lifshitz point if n 2.
In Table 3 we list numerical estimates of the critical exponents for d = 3, n = 1, and
m = 1, 2, . . . , 6. For comparison, we also included the m = 0 values of those critical and
correction-to-scaling exponents that go over into their standard counterparts , , , , and
for a critical point. 4 As is explained in the caption, these estimates were either obtained
by setting = d (m) 3 in the O( 2 ) expressions of the exponents or else via [1/1] Pad
approximants.
According to Table 2, the coefficients of most of these series with n = 1 do not alternate
in sign. Exceptions are the ones of l and l2 for small values of m, that of l for m = 0
(i.e., of the usual critical index ), and the one of for larger values of m. For d = 3,
the second-order contributions grow very rapidly as m increases because of the factor
2 = (1 + m/2)2 . Therefore the numerical estimates become less reliable for large m.
This effect is more pronounced for [1/1] estimates from non-alternating series than for the
corresponding direct evaluations at d = 3 (marked by superscripts ()). The better-behaved
4 Numerical estimates of the correlation exponents , , and are not included in Table 3, as they can be
q
l2 l4
found in I.
358
Table 3
Numerical estimates for the critical exponents with n = 1 and d = 3. The values marked by
superscripts () were obtained by setting = 1 + m/2 in the expansions to order 2 of the
exponents; those marked by superscripts [1/1] were determined from [1/1] Pad approximants
whose parameters were fixed by the requirement that the respective expansions to second order in
are reproduced
m
()
l2
[1/1]
l2
()
l4
[1/1]
l4
()
l
[1/1]
l
()
l
[1/1]
l
()
l
[1/1]
l
()
[1/1]
()
l2
[1/1]
l2
0.627
0.709
0.795
0.882
0.963
1.035
1.093
0.673
0.877
1.230
1.742
2.19
2.26
2.02
0.348
0.387
0.423
0.456
0.482
0.500
0.396
0.482
0.561
0.606
0.609
0.585
1.244
1.397
1.558
1.715
1.859
1.983
2.08
1.310
1.609
2.02
2.46
2.78
2.86
2.75
0.077
0.110
0.174
0.296
0.499
0.806
1.24
0.108
0.160
0.226
0.323
0.499
0.94
4.6
0.340
0.247
0.134
0.005
0.18
0.39
0.7
0.28
0.75
2.0
0.339
0.246
0.131
0.029
0.677
0.686
0.636
0.514
0.306
0.001
0.715
0.688
0.654
0.628
0.609
0.595
0.370
0.414
0.553
0.240
0.255
0.930
1.786
0.614
0.870
1.161
1.313
1.439
1.545
1.635
expansions yield smaller differences between these two kinds of estimates. In unfavorable
cases with rather large we reject the [1/1] estimates for the non-alternating series, which
tend to overestimate the values of the corresponding exponents. Instead we prefer the direct
evaluations at d = 3.
A reversed situation occurs for l and l2 with m = 0, 1, 2. The respective series
are alternating; they have negative O( 2 ) corrections, which tend to underestimate the
values of the exponents for d = 3 in direct evaluations of the O( 2 ) expressions. On
the other hand, the [1/1] approximants for these series 5 seem to do a better job,
suppressing the influence of the second-order corrections in a correct way. We believe
[1/1]
that l (m = n = 1) 0.160 belongs to our best numerical d = 3 estimates that are
obtainable from the individual expansions.
In the case of l2 , the second-order correction is much larger. While therefore less
accurate numerical estimates must be expected, the structure of the expansion for l2
suggests nevertheless that this correction-to-scaling exponent should have a larger value
5 The expansions of and start at order . We add unity to these series, construct the [1/1] approximants,
l
l2
and subsequently subtract unity from the resulting numerical values of the approximants.
359
than its m = 0 counterpart for the critical point. (Recall that the latter has a value close
to 0.8 [41].) Our best estimate is the [1/1] value
l2 0.870.
(92)
In order to obtain improved estimates we proceed as follows. We choose the best d = 3

estimates we can get from the apparently best-behaved expansions of certain exponents,
express the remaining critical indices in terms of the former, and compute their implied
values. Thus we select from Table 3 the numbers
()
[1/1]
l4 (d = 3; m = 1, n = 1) 0.348 and l
(3; 1, 1) 0.160,
(93)
which we complement by our estimate

l4 (3; 1, 1) 0.019
(94)
from I. Substituting these into the second one of the scaling relations (80) for l and the
hyperscaling relation (78) for l yields
l (3; 1, 1) 1.399 and l2 (3; 1, 1) 0.746,
(95)
respectively, from which in turn the values

l2 (3; 1, 1) 0.124 and l (3; 1, 1) 0.220
(96)
follow via the scaling relations l2 = 2 l /l2 and l = (2 l l )/2.

Likewise, the choices
()
(3; 1, 1) () = 0.677 and l4 (3; 1, 1) l4 = 0.348
(97)
give for the wave-vector exponent 6

q = l4 /
(98)
the estimate
q (3; 1, 1) 0.514,
(99)
which is fairly close to the value q 0.519 of I. We consider the values (92)(97) and
(99) as our best estimates for these 10 critical exponents.
Table 4 presents an overview of our numerical findings for m = 1. For convenience, the
mean-field results are included along with the values the expansions to first and second
order take at = 3/2. The case m = 1 with n = 2 corresponds to the Lifshitz point of the
axial next-nearest-neighbor XY (ANNNXY) model, which Selke studied many years ago
by means of Monte Carlo simulations [42].
In Table 5 we have gathered the available experimental results for critical exponents
together with estimates obtained from Monte Carlo calculations and high-temperature
series analyses. As one sees, our field-theory estimates are in good agreement with the
6 Note that Eq. (77) of I, which recalls the conventional definition of , contains a misprint: the variable
q
should be replaced by q.
360
Table 4
Critical exponents for m = 1, n = 1, 2, 3, and d = 3. The row marked MF gives the mean-field
values; rows marked O() and O( 2 ) list the values obtained by setting = 3/2 in the expansions to
first and second order in , respectively. The remaining rows contain the estimates from [1/1] Pad
approximants and our best estimates (Scal) obtained via scaling relations in the manner explained
in the main text
MF
O()
l2
l4
l
l
l
q
l2
l4
l2
1
2
1
4
0
1
2
1
1
2
1
2
n=2
Scal
O()

O 2
0.757
n=3
O()

O 2
0.67
0.798
0.625
0.709
0.877
0.746
0.65
0.313
0.25
0.25
1.25
0.625
0.348
0.110
0.247
1.397
0.677
0.396
0.160
0.246
1.609
0.715
0.348
0.160
0.220
1.399
0.677
0.325
0.15
0.275
1.3
0.65
0.372
0.047
0.276
1.495
0.725
0.335
0.068
0.295
1.34
0.67
0.392
0.178
0.301
1.576
0.765
0.5
0
0
1.5
0.521
0.042
0.020
0.466
0.5
0
0
1.5
0.521
0.044
0.021
0.517
0.5
0
0
1.5
0
0
0
n=1
2
O
[1/1]
0.519
0.039
0.019
0.414
0.514
0.124
0.019
0.870
Table 5
Values of critical exponents for uniaxial Lifshitz points (m = 1). Exp means summarized
experimental results, taken from Ref. [43]; HT denotes the high-temperature series estimates of
Ref. [44]. The rest are the Monte Carlo results of Refs. [45] (MC1), [46] (MC2), [27] (MC3), and
[42] (MC), respectively
l4
0.600.64
0.440.49
0.5
n=1
Exp
HT
MC1
MC2
MC3
0.41 0.03
0.33 0.03
0.4 0.5
0.20 0.15
0.2
0.18 0.03
0.21 0.03
0.19 0.02
0.235 0.005
1.62 0.12
1.36 0.005
1.4 0.06
1.36 0.03
n=2
MC
0.1 0.14
0.20 0.02
1.5 0.1
Monte Carlo results. The experimental value for l (3; 1, 1) deviates appreciably both from
all theoretical estimates (including ours) as well as from the Monte Carlo results, and
is probably not very accurate. On the other hand, the very good agreement of our fieldtheory estimates with the most recent Monte Carlo estimates by Pleimling and Henkel [27]
(which we expect to be the most accurate ones) is quite encouraging. Certainly, renewed
experimental efforts for determining the values of the critical exponents in a more complete
and more precise way would be most welcome.
361
The field-theory models (3) were introduced more than 25 years ago to describe the
universal critical behavior at m-axial Lifshitz points [19]. While some field-theoretic
studies based on the expansion about the upper critical dimension d (m) emerged soon
afterwards, these were limited to first order in , or restricted to special values of m or to
a subset of critical exponents, or challenged by discrepant results (see the references cited
in the introduction). Two-loop calculations for general values of m appeared to be hardly
feasible because of the severe calculational difficulties that must be overcome.
Complementing our previous work in I, we have presented here a full two-loop RG
calculation for the models (3) in d = d (m) dimensions, for general values of m
(0, 8). This enabled us to compute the expansions of all critical indices of the considered
m-axial Lifshitz points to second order in . We employed these results in turn to determine
field-theory estimates for the values of these critical exponents in three dimensions.
Although the accuracy of these estimates clearly is not competitive with the impressive
precision that has been achieved by the best field-theory estimates for critical exponents of
conventional critical points (based on perturbation expansions to much higher orders and
powerful resummation techniques [4,41]), they are in very good agreement with recent
Monte Carlo results for the uniaxial scalar case m = n = 1. We hope that our present work
will stimulate new efforts, both by experimentalists and theorists, to investigate the critical
behavior at Lifshitz points.
There is a number of promising directions in which our work could be extended. For
example, building on it, one could compute other universal quantities, such as amplitude
ratios and scaling functions, via the expansion.
A particular interesting and challenging question is whether the generalized invariance
found by Henkel [25] for systems whose anisotropy exponents take the rational values =
2/, = 1, 2, 3 . . . can be generalized to other, irrational values. That such an extension
exists, is not at all clear since the condition = 2/ is utilized in Henkels work to ensure
that the algebra closes. But if such an extension can be found, then the invariance under this
larger group of transformations should manifest itself through properties of the theories
scaling functions in d < d (m) dimensions, which could be checked by means of the
expansions. Furthermore, even if an extension cannot be found, one should be able to
benefit from the invariance properties of the free theory (with = 1/2) when computing
the expansion of anomalous dimensions of composite operators in a similar extensive
fashion as in the case of the standard critical-point 4 theory [4749].
An important issue awaiting clarification arises when m 2. In the class of models (3)
studied here, the quadratic fourth-order derivative term was taken to be isotropic in the
subspace Rm . However, in general further fourth-order derivatives cannot be excluded.
That is, the term (0 /2)( )2 should be generalized to
(0 /2)wa Tij(a)
kl (i j )k l ,
(100)
(a)
where the summation over a comprises all totally symmetric fourth-rank tensors Tij kl
compatible with the symmetry of the considered microscopic (or mesoscopic) model. The
362
isotropic fourth-order derivative term corresponds to Tij(a=1)

(ij kl + ik j l + il j k )/3
kl
with w1 1.
In order to give a simple example of a system involving a further quadratic fourthorder derivative term, let us consider an m-axial modification of the familiar uniaxial
ANNNI model [7,38] that has competing nearest-neighbor (nn) and next-nearest-neighbor
interactions along m equivalent of the d hypercubic axes (and only the usual nn bonds
along the remaining d m ones). Owing to the Hamiltonians hypercubic (rather than
isotropic) symmetry in the m-dimensional subspace, just one other fourth-order derivative
(2)
term, corresponding to the tensor Tij kl = ij kl li , must generically occur besides the
isotropic one, in a coarse-grained description. The associated interaction constant w2 is
dimensionless and hence marginal at the Gaussian fixed point. To find out whether the
nontrivial (u > 0, w2 = 0) fixed point considered throughout this work remains infrared
stable, one must compute the anomalous dimension of the additional scaling operator
that can be formed from the above two fourth-order derivative terms. This issue will be
taken up in a forthcoming joint paper with R.K.P. Zia [50], where we shall show that
the associated crossover exponent, to order 2 , is indeed positive. Hence, deviations from
w2 = 0 correspond to a relevant perturbation at the w2 = 0, u > 0 fixed point, which
should destabilize it unless m = 1.
Finally, let us mention that the potential of the -expansion results presented in this
paper certainly has not fully been exploited here. When estimating the values of the critical
exponents for d = 3 dimensions, we utilized only their expansions for a fixed integer
number of the parameter m. However, our results hold also for noninteger values of m.
Making use of this fact, one should be able to extrapolate to points (d = 3, m = integer) of
interest in a more flexible fashion, starting from any point on the critical curve d = d (m)
and going along directions not perpendicular to the m axis. By exploiting this flexibility
one should be able to improve the accuracy of the estimates.
Last but not least, let us briefly mention where the interesting reader can find information
about experimental results. Earlier experimental work is discussed in Hornreichs and
Selkes review articles [6,7]. A more recent summary of experimental results for the critical
exponents and other universal quantities of the Lifshitz point in MnP and Mn0.9 Co0.1 P can
be found in Ref. [43] and its references. (These results were partly quoted in Table 5.)
However, the variety of experimental systems having (or believed to have) Lifshitz points
is very rich, ranging from ferromagnetic and ferroelectric systems to polymer mixtures.
A complete survey of the published experimental results on Lifshitz points is beyond the
scope of the present article.
Note added in proof
After submission of this paper, C.E.I. Carneiro kindly rechecked his calculations with
C. Mergulho, Jr., and informed us that he now agrees with our -expansion results for l2
and l4 in the cases m = 2 and m = 6. The sources of our discrepancies with the incorrect
results of L.C. de Albuquerque and M.M. Leite [23,24] are discussed in detail in our recent
comment (preprint cond-mat/0106502).
363
Acknowledgements
It is our pleasure to thank Malte Henkel and Michel Pleimling for informing us about
their work [27] prior to publication as well as for discussions and correspondence. We
gratefully acknowledge support by the Deutsche Forschungsgemeinschaft (DFG) via the
Leibniz program Di 378/2-1.
Appendix A. Series representation and asymptotic expansion of ()

From the integral representation (8) of the scaling function () we find
i 2 q e iq e
iq e iq
e
e
e
e
4+2
=
,
() =
2
4
2
q + q
q + q 4
q q
(A.1)
q q
where e = / is an arbitrary unit m-vector. The second form follows via rescaling of the
momenta; it lends itself for studying the large- behavior. The first one is appropriate for
deriving the small- expansion.
Upon utilizing the Schwinger representation
1
=
2
q + q 4
ds es(q +q )
2
(A.2)
for the momentum-space propagator in Eq. (A.1), we can perform the integration over q
to obtain

4
4 1
d /2 4+2
() = (4)
(A.3)
ds s d /2 esq (4s ) eiq e .

q
Doing the angular integrations gives

e
sq 4 iq e
= (2)
m/2
dq q m/2 J m2 (q)esq .
4
(A.4)
We insert this into Eq. (A.3), expand the exponential e1/(4s

integrate the resulting series termwise over s. This yields
24
4k
(1)k
2m d1
2
k
() = 2
2
k!
2
4)
in powers of 4 , and
(A.5)
k=0
with
k =
m
4
1k

2

26m+6k3
dq q 4(1+k) 2 2 J m2 (q)
2
2 (2 + 2k )

,
m4 12 + 2 k
(A.6)
364
which is the asymptotic expansion (15). The Taylor series (12) can be derived along similar
lines, starting from the first form of the integral representation (A.1).
Appendix B. Laurent expansion of required vertex functions

In this appendix we gather our results on the Laurent expansions of those vertex
functions whose pole terms determine the required renormalization factors. It is understood
that 0 and 0 are set to their critical values c = c = 0. For notational simplicity, we
introduce the dimensionless bare coupling constant
m/4
u 0 0
u0 = Z u u
(B.1)
and specialize to the scalar case n = 1. The generalization to the n-component case
involves the usual tensorial factors and contractions of the standard ||4 theory and should
be obvious.
We use the notation ({q j }) for the Fourier transforms of vertex functions ({x j })
(with the momentum-conserving function taken out):

N
d
({q j })(2)
q j ei j=1 q j x j .
({x j }) =
(B.2)
j
q 1 ,...,q N
(2)
B.1. Two-point vertex functions (2) and (
)2

From our results obtained in I we find 7

2
(2) (q) = 0 q 4 + q

+ O u30
(B.3)
with
=

2

u 20 j (m)0 q 4
j (m)q
+ O 0
6 16m(m + 2) 2(8 m)
(B.4)
and
(2)
( )2 (q, Q = 0) = q 2

+ O u30
(B.5)
with
=

u 20 j (m) 2
q + O 0 ,
2 4m
(B.6)
where Q = 0 is the momentum of the inserted operator ( )2 /2; i.e., the insertion
considered is dd x ( )2 /2.
7 We suppress diagrams involving the one-loop (sub)graph
since the latter vanishes for = 0 if
dimensional regularization is employed, as we do throughout this paper.
365
B.2. Four-point vertex function (4)

The four-point vertex function was computed only to one-loop order in I. To the order
of two loops it reads 7

+ 2P
+ 2P
(4) (q 1 , . . . , q 4 ) = u0

(B.7)
+ 5P + O u40

u 2
u 0
I2 (q ij ) 0 I22 (q ij )
= u0 1
2
4
(ij )=(12),(23),(24)

u 2
+ 0 I4 (q 12 , q 3 ) + 5P .
(B.8)
2
Here P means permutations (of the external legs). The momenta marked by a check ()
are dimensionless ones defined via

1/4
q = (q , q ) 0 1/2 q , 1 q ,
(B.9)
and q ij q i + q j . The integrals I2 and I4 are given by

1
1
I2 (Q)
2
4
2
q + q (q + Q) + (q + Q)4
(B.10)
and

I4 (Q, K)
q
1
2
q
+ q 4
1
(q
+ Q)2
+ (q + Q)4
I2 (q K).
(B.11)
The pole term of I2 (Q) can be read off from Eqs. (24) and (89) of I. However, in our
two-loop calculation I2 also occurs as a divergent subintegral. To check that the associated
pole terms are canceled by contributions involving one-loop counterterms, we also need
the finite part of I2 . The calculation simplifies considerably if the momentum Q is chosen
to have a perpendicular component only, so that Q = Qe (which is sufficient for our
purposes). 8 For such values of Q, the integral I2 (Q) can be analytically calculated in a
straightforward fashion, either by going back to Eqs. (16) and (19) of I and computing the
Fourier transform of these distributions, or directly in momentum space, as we prefer to do
here. For dimensional reasons, we have
I2 (Qe ) = Q I2 (e ).
(B.12)
The integral on the right-hand side is precisely the one written as Fm, / in Eq. (38).
Utilizing a familiar method due to Feynman for folding two denominators into one
8 Previously e denoted a fixed arbitrary unit d m vector. For convenience, we use here and below the same
symbol for the associated d vector whose projection onto the perpendicular subspace yields the former while its
m parallel components vanish.
366
(Eq. (A8-1) of Ref. [36]), one is led to

1
I2 (e ) =
ds
0
=2
2
q 4 + q
+ 2sq e + se2
2
q
4+ m
2 +

2+ m
4 +2
m + 2
1

ds
m 2

d/2 4 2 1 2
= (4)
,
2 (2 ) m2
m+2
4
4
q + s(1 s)
(B.13)
from which the result (39) for Fm, follows at once.

The calculation of I4 (Q, K) is more involved; it is described in Appendix C, giving

2 1
2 Q
+ Ju (m) + O() ,
I4 (Qe , K) = Fm,
(B.14)
2
where Ju (m) is the quantity defined in Eq. (49).
(2)
B.3. Vertex function 2
1 2
Next, we turn to the vertex function (2)
2 with an insertion of 2 ( )Q =
1
2
dd x 2 (x)
eiQx . To two-loop order it is given by 7

(2)
2 (q; Q) = 1

+ O u30

u 0
1
2 1 2
= 1 I2 (Q) + u 0 I2 (Q) + I4 (Q, q)

+ O u30 ,
2
4
2
(B.15)
where the checked momenta are again dimensionless ones, defined by analogy with
Eq. (B.9).
Appendix C. Laurent expansion of the two-loop integral I4

As can be seen from Eq. (B.11), the integral I4 (q 12 , q 3 ) associated with the graph
involves the divergent subintegral I2 (q q 3 ). The latter has a momentum-independent pole
term 1 [cf. Eqs. (B.13) and (39)]. Furthermore, the graph that results upon contraction
of this subgraph to a point [which itself is proportional to I2 (q 12 )] has contributions of
order 0 that depend on q12 . Taken together, these observations tell us that the pole term
1 of I4 (q 12 , q 3 ) depends on q12 but not on q3 . 9 Since the 2 pole of I4 (q 12 , q 3 ) is
momentum independent, we can set q3 = 0 when calculating the pole part of this integral.
9 It is precisely this q -dependent pole term that gets canceled by subtracting from the divergent subgraph its
12
pole part.
367
To further simplify the calculation, we can choose q 12 to have vanishing parallel

component again, setting q 12 = Qe . The integral to be calculated thus becomes

d
I4 (Qe ; 0) = d x dd y G(y)G(x y)eiQe y G2 (x),
(C.1)
where G(y) now means the free propagator (5) with 0 = 1. Let us substitute the free
propagators
of the factor G2 (x) by their scaling form (7) and rewrite the Fourier integral
d
d y . . . as a momentum-space integral, employing the Schwinger representation (A.2)
for both of the two free propagators in momentum space. Making the change of variables
1/2
x = x x , we arrive at
I4 (Qe , 0) =
d ()
2+ m 4
x x 2

ds
dm
2 (s+t )+q (ix 2t Qe )t Q2

q

e
dt
q 4 (s+t )+iq x
(C.2)

Now the momentum integrations q and q are decoupled and can be performed in a

straightforward fashion. That the latter integral takes such a simple form is due to our
choice of Q with Q = 0. Performing the angular integrations yields
4
eq (s+t )+iq x
q
= (2)
m
2
m

2m

4
1/2
dq q 2 eq (s+t ) 2 x 4 J m2 q x .
2
(C.3)
We replace the Bessel function in Eq. (C.3) by its familiar Taylor expansion

(1)k w2k
w
,
J (w) =
2
22k k! ( + k + 1)
(C.4)
k=0
integrate term by term over q , employing

dq q m1+2k eq
4 (s+t )

m+2k
m + 2k
1
(s + t) 4 ,
=
4
4
(C.5)
and simplify the resulting ratio of -functions by means of the well-known duplication
formula (6.1.18) of Ref. [35]. This gives
4
1m
m+2k
( 2 x )k
eq (s+t )+iq x = 2
(C.6)
2+m+4k (8 s + t ) 2 .
k!
4
k=0
q
The integration over q in Eq. (C.2) is Gaussian. Upon substituting the result together with
the above equations into (C.2), we get
368
I4 (Qe , 0) = 2
4m
24m
4

m
d ()
k

Ck (Q; s, t)
2
2+m+2k
8 s+t
k!
4
k=0
with

Ck (Q; s, t)
dm
k4+2+ m
2
x x
dt (s + t)
ds
4
2
e s+t Q
st
(C.7)
2

x + 4itQe x
exp
.
4(s + t)
(C.8)
We first perform the angular integrations and subsequently the radial integration of the
latter integral, obtaining

Ck (Q; s, t) = (2)
dr r
r
4(s+t)
k2+ m+6
4
tQ
s+t
1m
Jm 1
tQr
s+t

k+ k+
k+
t 2 Q2
k+
m 2
2
(s + t) 2 1F1
; m ;
=
,
(m )
2
s+t
(C.9)
where we have introduced

m
d m
=2 .
m
(C.10)
2
4
2
Next we insert this result into expression (C.7) for I4 , and make a change of variable s
z = s/t. The t-integration then becomes straightforward (see Eq. (2.22.3.1) of Ref. [51]),
and we find that
1m
I4 (Qe , 0) = Q2
224m 2 ()

8m2
4
m
d (; m, d)
k

Ak (m, ) 2 ,
(C.11)
k=0
with
Ak (m, ) =
k+
2

k! 22k 2+m+2k
4
0
dz z
(z + 1)
22

k+
1
; m ;
.
2F1 ,
2
z
(C.12)
Owing to the overall factor () and the additional factor (/2) of the coefficient
A0 (m, ), the k = 0 term of the above series contributes poles of second and first order
in to I4 . The remaining terms with k 1 yield poles of first order in . Consider first
the k = 0 term. The value of the integral over may be gleaned from I [cf. its Eqs. (3.16)
and (4.47)]:

m

25 2 4 m4 2 m4 2 1 2
m
2

d (; m, d) =
(C.13)
.
m2 (2 )
The integral over z in A0 (m, ) can be evaluated explicitly by means of M ATHEMATICA
[52]. Alternatively, one can change to the integration variable = 1/z and look up the
369
transformed integral in Eq. (2.21.1.15) of the integral tables [51]. The result has a simple
expansion to order , giving
A0 (m, ) =

(/2)
2+m 1 + 2 + O( 2 )
4
(C.14)
upon substitution into Eq. (C.12).

Turning to the contributions with k 1, we note that both the scaling function
(; m, d) and the coefficients Ak (m, ) may be taken at d = d (i.e., = 0). Then

the integral 0 dz . . . reduces to one and the series

k=1 becomes the function (; m)
introduced in Eq. (28). It follows that
k ju (m)

1 + O() ,
dm 2 (; m, d)
(C.15)
Ak (m, ) 2 =
Bm
k=1
where ju (m) and Bm are the integral (50) and the coefficient (46), respectively. Combining
the above results and expanding the prefactors of the integral in Eq. (C.11), we finally
obtain the result stated in Eq. (B.14).
Appendix D. Asymptotic behavior of ()

Upon differentiating the series (28) of (; m) termwise and comparing with the Taylor
one sees that the following relation holds:
expansion (12) of the scaling function ,

(; m) 4
m, d ) (0;
m, d ) .
= (;
(D.1)
m, d ) = 1/ [(m + 2)/4]. Let us substitute

From Eq. (12) we can read off the value (0;
m, d ) into this equation and integrate. This
the asymptotic expansion Eq. (15) of (;
yields

4 ln + C (m)
2(1)k (2k) 4k
+
(; m)
(D.2)
.
m+2
m+24k
2
4
k!
4
k=1
The terms of orders 4 and 8 agree with those of the asymptotic form (30) of (; m).
Hence it remains to show that the integration constant C is given by

m+2
C (m) =
(D.3)
CE + ln 16.
4
To this end an integral representation of (; m) is helpful. Consider the integral

iq +iq e
2 k
k
6+m2
e
2 ( )
J (; m, ) 24+m 4
=
,

2
2+m+2k
2k
k!2
(q 4 + q 2 )
4
k=0
q
(D.4)
in terms of which (; m) can be written as

J (; m, ) 2
(; m) = lim

0
2+m
4
(D.5)
370
and whose large- form

J (; m, ) 24+m
6+m2
4

2
q
eiq e
2)
(q 4 + q

1 + O 4

212 () 2
2+m 1 + O 4
4
(D.6)
is easily derived. Insertion of the latter result into Eq. (D.5) gives the value (D.3) of C .
Appendix E. Numerical integration
The quantities j (m), j (m), j (m), and ju (m) in terms of which we expressed the
series expansion coefficients
of the renormalization factors and the critical exponents are

integrals of the form 0 d f () [cf. Eqs. (43)(45) and (50)]. Their integrands, f , while
integrable and decaying to zero as , in general involve differences of generalized
hypergeometric functions, i.e., differences of functions that grow exponentially as .
Therefore standard numerical integration procedures run into problems when the upper
integration limit becomes large.
To overcome this difficulty, we proceed in a similar manner as in I. From our knowledge
of the asymptotic expansions of the functions (; m, d ), (; m, d ), and (; m) we
can determine that of the integrand. Let fas(M) () be the asymptotic expansion of f () to
order M . Then we have
(k)
f () fas(M) ()
(E.1)
Cf k .
k=M+1
We split the integrand as

0
d f () =
0
d f ()
d fas(M) () + Rf(M) (0 ),
(E.2)
where
Rf(M) (0 )

d f () fas(M) () .
(E.3)
Then we choose 0 as large as possible,

but small enough so that M ATHEMATICA [52]
is still able to evaluate the integral 0 0 f () d by numerical integration, determine the
second term on the right-hand side of Eq. (E.2) by analytical integration, and neglect the
third one. The asymptotic expansion of the latter is easily deduced from Eq. (E.1). It reads
(M)
Rf
(0 )
C (k)
f
k=M
0k .
(E.4)
Since the expansion (E.1) is only asymptotic, the value of M must not be chosen too
large. In practice, we utilized the asymptotic expansions of , , and up to the orders
371
12 , 10 , and 8 explicitly shown in the respective Eqs. (22), (25), and (30), and then
truncated the resulting expression of the integrand f () consistently at the largest possible
order. As upper integration limit 0 of the numerical integration we chose values between
9 and 10.
As a consequence of the fact that all integrands f () have an explicit factor of m , the
precision of our results decreases as m increases. Furthermore, the accuracy is greatest for
j (m), whose integrands asymptotic expansion starts with (13m) , a particularly high
power of 1 . The precision is lower for j (m) and ju (m) because their integrands involve
either four more powers of than that of j or else the function () as a factor, whose
asymptotic expansion starts with a term ln .
As a test of our procedure we can compare the numerical values of the integrals it
produces for m = 2 and m = 6 with the analytically known exact results (51)(54).
The agreement one finds is very impressive: nine decimal digits of the exact results
are reproduced (even for m = 6) when the numerical integration is done by means
of the M ATHEMATICA [52] routine NIntegrate with the option WorkingPrecision =
40. However, we must not forget that the cases m = 2 and m = 6 are special in that
the asymptotic expansions of the functions , , and and hence those of the
integrands vanish or truncate after the first term. Hence it would be too optimistic to
expect such extremely accurate results for other values of m. In the worst cases (e.g., that
of j (7) and ju (7)), the fourth decimal digit typically changes if is varied in the range
911. Therefore we are confident that the first two decimal digits of the m = 7 values given
in Table 1 are correct. For smaller values of m the precision is greater. 10
References
[1] M.E. Fisher, Rev. Mod. Phys. 46 (1974) 596597.
[2] C. Domb, M.S. Green (Eds.), Phase Transitions and Critical Phenomena, Vol. 6, Academic,
London, 1976.
[3] M.E. Fisher, in: F.J.W. Hahne (Ed.), Critical Phenomena, Lecture Notes in Phys., Vol. 186,
Springer-Verlag, Berlin, 1983, pp. 1139.
[4] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Int. Ser. Monographs Phys., 3rd
edn., Clarendon Press, Oxford, 1996.
[5] A. Aharony, in: C. Domb, J.L. Lebowitz (Eds.), Phase Transitions and Critical Phenomena, Vol.
6, Academic, London, 1976, pp. 358424.
[6] R.M. Hornreich, J. Magn. Magn. Mater. 1518 (1980) 387392.
[7] W. Selke, in: C. Domb, J.L. Lebowitz (Eds.), Phase Transitions and Critical Phenomena,
Vol. 15, Academic, London, 1992, pp. 172.
[8] B. Schmittmann, R.K.P. Zia, Statistical Mechanics of Driven Diffusive Systems, Phase
Transitions and Critical Phenomena, Vol. 17, Academic, London, 1995.
[9] J. Krug, Adv. Phys. 46 (1997) 139282.
[10] B.I. Halperin, P.C. Hohenberg, Rev. Mod. Phys. 49 (1977) 435479.
10 For example, in the cases of j (4), j (4), and j (4) changes of [9, 11] affect only the respective
0
last decimal digits (in parentheses) of the numerical values j (4) 20.0677(4), ju (4) 0.80378728(5), and
j (4) 0.80378728(5).
372
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
I. Nasser, R. Folk, Phys. Rev. B 52 (1995) 1579915806.

I. Nasser, A. Abdel-Hady, R. Folk, Phys. Rev. B 56 (1997) 154160.
C. Mergulho Jr., C.E.I. Carneiro, Phys. Rev. B 58 (1998) 60476056.
C. Mergulho Jr., C.E.I. Carneiro, Phys. Rev. B 59 (1999) 1395413964.
K. Binder, H.L. Frisch, Eur. Phys. J. B 10 (1999) 7190.
H.L. Frisch, J.C. Kimball, K. Binder, J. Phys. Condens. Matter 12 (2000) 2942.
M.M. Leite, Phys. Rev. B 61 (2000) 1469114693.
H.W. Diehl, M. Shpot, Phys. Rev. B 62 (2000) 1233812349, cond-mat/0006007.
R.M. Hornreich, M. Luban, S. Shtrikman, Phys. Rev. Lett. 35 (1975) 16781681.
D. Mukamel, J. Phys. A 10 (1977) L249L252.
R.M. Hornreich, A.D. Bruce, J. Phys. A 11 (1978) 595601.
J. Sak, G.S. Grest, Phys. Rev. B 17 (1978) 36023606.
L.C. de Albuquerque, M.M. Leite, cond-mat/0006462 v3.
L.C. de Albuquerque, M.M. Leite, J. Phys. A 34 (2001) L327L332.
M. Henkel, Phys. Rev. Lett. 78 (1997) 19401943.
M. Henkel, Conformal invariance and critical phenomena, Texts Monographs Phys., Springer,
Berlin, 1999.
M. Pleimling, M. Henkel, hep-th/0103194.
R.M. Hornreich, M. Luban, S. Shtrikman, Phys. Lett. 55A (1975) 269270.
L. Frachebourg, M. Henkel, Physica A 195 (1993) 577602.
C. Fox, Proc. London Math. Soc. 27 (1928) 389400.
E.M. Wright, J. London Math. Soc. 10 (1935) 286293.
E.M. Wright, Proc. London Math. Soc. 46 (1940) 389408.
A.R. Miller, I.S. Moskowitz, Comput. Math. Appl. 30 (1995) 7382.
A.A. Inayat-Hussain, J. Phys. A 20 (1987) 41094117.
Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables,
M. Abramowitz, I.A. Stegun (Eds.), Applied Math. Ser., 10th edn., National Bureau of
Standards, Washington, DC, 1972.
D.J. Amit, Field Theory, the Renormalization Group, and Critical Phenomena, 2nd edn., World
Scientific, Singapore, 1984.
J. Cardy, Scaling and Renormalization in Statistical Physics, Cambridge Lecture Notes in Phys.,
Cambridge Univ. Press, Cambridge, UK, 1996.
M.E. Fisher, W. Selke, Phys. Rev. Lett. 44 (1980) 15021505.
N.D. Mermin, H. Wagner, Phys. Rev. Lett. 17 (1966) 11331136.
G.S. Grest, J. Sak, Phys. Rev. B 17 (1978) 36073610.
R. Guida, J. Zinn-Justin, J. Phys. A 31 (1998) 81038121.
W. Selke, J. Phys. C 13 (1980) L261L263.
A. Zieba, M. Slota, M. Kucharczyk, Phys. Rev. B 61 (2000) 34353449.
Z. Mo, M. Ferer, Phys. Rev. B 43 (1991) 1089010905.
W. Selke, Z. Phys. 29 (1978) 133137.
K. Kaski, W. Selke, Phys. Rev. B 31 (1985) 31283130.
S.K. Kehrein, F.J. Wegner, Y.M. Pismak, Nucl. Phys. B 402 (1993) 669692.
S.K. Kehrein, F. Wegner, Nucl. Phys. B 424 (1994) 521546.
S.K. Kehrein, Nucl. Phys. B 453 (1995) 777806.
H.W. Diehl, M. Shpot, R.K.P. Zia, 2001.
A. Prudnikov, Yu.A. Brychkov, O.I. Marichev, Integrals and Series, Vol. 2, Gordon & Breach,
New York, 1986.
Mathematica, version 3.0, a product of Wolfram Research.

Expectation values of descendent fields in the

BulloughDodd model and related perturbed
conformal field theories
P. Baseilhac a , M. Stanishkov b,1
a Department of Mathematics, University of York, Heslington, York YO105DD, UK
b Asia Pacific Center for Theoretical Physics, Seoul, 130-012, South Korea
Received 2 May 2001; accepted 12 June 2001
Abstract
2 ea in
The exact vacuum expectation values of the second level descendent fields ()2 ()
the BulloughDodd model are calculated. By performing quantum group restrictions, we obtain
L2 L 2 lk in the 12 , 21 and 15 perturbed minimal CFTs. In particular, the exact expectation
value T T is found to be proportional to the square of the bulk free energy. 2001 Elsevier Science
PACS: 11.10.-z; 11.25.Hf
1. Introduction
In a 2-D integrable quantum field theory (QFT) which can be realized as a conformal
field theory (CFT) perturbed by some relevant operator, it is well-known that any
correlation function of local fields Oa (x) in the short-distance limit can be reduced down to
one-point functions Oa (x) by successive application of the operator product expansion
(OPE) [1,2]. These vacuum expectation values (VEV)s contain important information
about the IR environment. Together with the structure constants characterizing the UV
limit of the QFT, they provide the UV behaviour of the correlation functions whereas the
so-called form-factors characterize their IR behaviour. Since three years important progress
has been made concerning the evaluation of some VEVs in different integrable QFTs. In
Ref. [3], an explicit expression for the VEVs of the exponential fields in the sinh-Gordon
and sine-Gordon models was proposed. In Ref. [4,5] it was shown that this result can be
obtained using the reflection amplitude [6] of the Liouville field theory. This method
E-mail addresses: pb18@york.ac.uk (P. Baseilhac), marian@mail.apctp.org (M. Stanishkov).
1 On leave of absence from INRNE, Sofia, Bulgaria.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 8 7 - 5
374
P. Baseilhac, M. Stanishkov / Nuclear Physics B 612 [FS] (2001) 373390
was also applied in the so-called BulloughDodd model with real and imaginary coupling.
In QFT involving more fields, the VEVs for a two-parameter family of integrable QFTs
introduced and studied in [7] gave rise to the VEVs of local operators in parafermionic
sine-Gordon models and in integrable perturbed SU(2) coset CFT [8]. Also, the VEVs
in simply-laced affine Toda field theories are known for a long time [9] and the case of
nonsimply laced dual pairs was recently studied in [10,11].
However, the higher-order corrections to the short-distance expansion of two-point
correlation functions involve the VEVs of the descendent fields. This question was
2 ea in the sinhaddressed in [12]. There, the VEV of the descendent field ()2 ()
Gordon (ShG) and sine-Gordon (SG) with a i models was calculated as well
as in its related perturbed CFT, i.e., 13 perturbation of minimal models. From this result,
the next-order correction of the two-point function in the scaling LeeYang model was
computed [2,5].
The purpose of this paper is to calculate the VEV of the simplest nontrivial descendent
field in the BulloughDodd (BD) model which is generally described by the following
action in the Euclidean space:

1
2
2
b
2b
( ) + e + e
.
ABD = d x
(1.1)
16
Here, the parameters and are introduced, as the two operators do not renormalize in
the same way, on the contrary to any simply-laced affine Toda field theory. This model
has attracted over the years a certain interest, in particular in connection with perturbed
minimal models: c < 1 minimal CFT perturbed by the operators 12 , 21 or 15 can be
obtained by a quantum group (QG) restriction of imaginary BulloughDodd model [5,
1316] with special values of the coupling. We will use this property to deduce the VEV
L2 L 2 lk in the following perturbed minimal models:

A = Mp/p + d 2 x 12 ,
(1.2)

A = Mp/p + d 2 x 21 , or
(1.3)

A = Mp/p + d 2 x 15 ,
(1.4)
where we denote, respectively, 12 , 21 and 15 as specific primary operators of the
unperturbed minimal model Mp/p and the parameters , and characterize the strength
of the perturbation.
This paper is organized as follows. In the next section we introduce the notations and
write the short-distance expansion of the two-point correlation function which involves
2 ea in the BD model associated with the
the VEV of the descendent field ()2 ()
action (1.1). Using the method based on the reflection relations [6] we find a conjecture
for this last quantity in Section 3. Whereas it exists an ambiguity for the solution of
these functional equations, we choose the minimal one which is compatible with the
resonance conditions (see Ref. [12] for details). In Section 4 we compare the semiclassical limit of the short distance expansion of the two-point function with the semi-
375
classical approximation based on the action (1.1). In Section 5 we deduce the VEV
L2 L 2 lk in the models (1.2), (1.3) and (1.4). Concluding remarks follow in the last
section.
2. Short-distance expansion of the two-point function in the BD model

Similarly to the ShG model [12], the BD model can be regarded as a relevant
perturbation of a Gaussian CFT. In this free field theory, the field is normalized such that:
(z, z )(0, 0)Gauss = 2 log(zz)
(2.1)
and we have the classical equation of motion:

= 0.

(2.2)
Instead of considering the action (1.1) we turn directly to the case of an imaginary coupling
constant which is the most interesting for our purpose in Section 5. The perturbation is then
relevant if 0 < 2 < 1. Although the model (1.1) for real coupling is very different from
the one with imaginary coupling in its physical content (this latter model contains solitons
and breathers), there are good reasons to believe that the expectation values obtained in
the real coupling case provide also the expectation values for the imaginary coupling. The
calculation of the VEVs in both cases (b real or imaginary) within the standard perturbation
theory agree through the identification b = i [5]. With this substitution in (1.1), the
general short distance OPE for two arbitrary primary fields ei1 (x) and ei2 (y) takes
the form:
ei1 (x)ei2 (y)

n,0
n,0
n
i(+n)
(y) + +
=
C1 2 (r)e
C 1 2 (r)ei( 2 ) (y) +
n=0
n=1

n,0
1
+
D1 2 (r)ei(+(n 2 )) (y) + ,
(2.3)
n=1
where = 1 + 2 , r = |x y| and the dots in each term stand for the contributions of the
descendents of each field. The different coefficients in Eq. (2.3) are computable within the
conformal perturbation theory (CPT) [2,17]. We obtain:

2
2 2
2
Cn,0
(r) = n r 41 2 +4n(1 +2 )+2n(1 )+2n fn,0
( )2 r 63 ,
1 2
1 2
2
n2 2
2
n,0
n
n,0
C 1 2 (r) = r 41 2 2n(1 +2 )+2n(1 4 )+ 2 f 1 2 ( )2 r 63 ,

1
2
2 2
2
(r) = n r 41 2 +4(n 2 )(1 +2 )+2n(12 )+2+2n gn,0
Dn,0
( )2 r 63 ,
1 2
1 2
(2.4)
where any function h {f, f , g} admits a power series expansion:
hn,0
1 2 (t) =

k=0
k
hn,0
k (1 , 2 )t .
(2.5)
376
Each coefficient in (2.4) is expressed in terms of Coulomb type integrals. The corresponding leading terms are, respectively, given by:

f0n,0 (1 , 2 ) = jn 1 , 2 , 2 for n = 0,

2 2
1
n,0
,
,
,
f 0 (1 , 2 ) = jn
2
2
4

g0n,0 (1 , 2 ) = Fn,1 1 , 2 , 2 ,
(2.6)
where we introduced the integrals:

n
n
n

1
jn (a, b, ) =
d 2 xk
|xk |4a |1 xk |4b
|xk xp |4 ,
n!
k=1
k=1
k<p

n
n
m
n

1
d 2 xk
d 2 yl
|xk |4a |1 xk |4b
|xk xp |4
Fn,m (a, b, ) =
n!m!
k=1
l=1
|yl |2a |1 yl |2b
l=1
k<p
k=1
|yl yq |
l<q
n,m

|xk yl |2 .
(2.7)
k,l
Notice that f00,0 (1 , 2 ) = 1 and that the first subleading term of the coefficient C0,0
is:
1 2

f10,0 (1 , 2 ) = F1,2 1 , 2 , 2 .
(2.8)
The integrals jn (a, b, ) have been evaluated explicitly in [17] with the result:
n

n
jn (a, b, ) = ()
(k)
n
k=1
n1

k=0
(1 + 2a + k) (1 + 2b + k)

1 2a 2b (n 1 + k)
(2.9)
where the usual notation (x) = (x)/ (1 x) is used.

As we already said in the introduction, the next subleading terms in (2.3) involve the
descendent fields. There are four independent second-level descendent fields in BD:
2 ei ,
()2 ()
2 ei ,
( 2 )()
()2 ( 2 )ei ,
( 2 )( 2 )ei .
(2.10)
Similarly to the SG (or ShG) case, using (2.2) it is easy to show that linear combinations of
these descendent fields can be written in terms of total derivatives of local fields (we refer
the reader to [12] for details about these relations). As a result, the VEVs of the composite
fields (2.10) can all be expressed in terms of a single VEV, say:

2 ei .
()2 ()
(2.11)
BD
Let us make an important observation. The second subleading terms in the OPE (2.3)
appear to be the third order descendents of the primary fields. Analogously to the previous
377
discussion linear combinations of them can be expressed in terms of total derivatives

of some local fields. As before, all the corresponding VEVs can be expressed through
3 ei . Unlike the SG case, it is nonvanishing due to the absence of a conserved
()3 ()
charge of spin 3 in the BD model. We will not enter in details about this VEV since its
computation is not our purpose in this paper.
One can now write the short-distance expansion for the two-point function:

G1 2 (r) = ei1 (x)ei2 (y) BD with r = |x y|
(2.12)
by taking the expectation value of the r.h.s. of the OPE (2.3) in the BD model with
imaginary coupling. Due to the previous discussion, the first nonvanishing contribution
of the VEVs of lowest descendent fields in the r.h.s. of the VEV of (2.3) correspond to the
following terms:

2 ei(+n) ,
(r) ()2 ()
Cn,2
BD
1 2

n,2
2 2 i( n
)
2
C 1 2 (r) () () e
,
BD

1
2 ei(+(n 2 )) ,
(r) ()2 ()
Dn,2
(2.13)
BD
1 2
respectively. These coefficients also admit expansion similar to Eqs. (2.4), (2.5) and (2.6).
In particular we have:

(1 2 )2 41 2 +4
2
r
1 + O ( )2 r 63 .
(2.14)
4
Finally, the short-distance (r 0) expansion of the two-point correlation function in the
BD model with imaginary coupling writes:
(r) =
C0,2
1 ,2
G1 2 (r) = G1 +2 r 41 2

(1 2 )2
2
H(1 + 2 )r 4
1 + F1,2 1 , 2 , 2 ( )2 r 63 +
4

2 2 (1 2 )2
2
K(1 + 2 )r 6 + O 2 ( )4 r 126
1 2
144

2
2 2
+
n r 41 2 +4n(1 +2 )+2n(1 )+2n jn 1 , 2 , 2
n=1

2
G1 +2 +n 1 + O ( )2 r 63
r 41 2 2n(1 +2 )+2n(1
n
n=1
n
1 +2 2

1 + O ( )2 r
1

2 2
1
,
,
jn
2
2
4

63 2
2
n2 2
4 )+ 2
n r 41 2 +4(n 2 )(1 +2 )+2n(12
2 )+2+2n2 2

Fn,1 1 , 2 , 2
n=1

2
G1 +2 +(n 1 ) 1 + O ( )2 r 63 ,
2
(2.15)
378
where we defined H() and K() by the ratios:

2 ei BD
()2 ()
,
ei BD
3 ei BD
()3 ()
K() =
ei BD
H() =
(2.16)
(2.17)
and G = ei BD is the VEV of the exponential field in the BD model. A closed analytic
expression for this latter VEV has been proposed in Ref. [5]:
2 /2
2
2
(1 + 2 ) (1 2 /4) 3
ei BD =
(1 2 ) (1 + 2 /4)
m 1 2 2 +2 2
63 2
63 2
22/3 3 (1/3)
+

sinh((2 2 )t) (t, )
dt
2 2t
exp
(2.18)
2 e
t sinh(3(2 2 )t) sinh(2t) sinh( 2 t)
0
where

(t, ) = sinh(2t) sinh (4 2 2)t sinh (2 2 2 + 2)t

+ sinh (2 2 2)t sinh (2 2 + 2)t

sinh (2 + 2 2)t .
Its integral representation is well defined if:
1
1
< Re() <
2
(2.19)
and obtained by analytic continuation outside this domain.

It is then straightforward to obtain the result associated with the action (1.1), i.e., for
real values of the coupling constant b which follows from the obvious substitutions:
ib,
,
1 ia1 ,

2 ia2,
(2.20)
In the (Gaussian) free field theory, the composite fields (2.10) are spinless with scale
dimension:
D + = 2 2 + 4.
(2.21)
For generic value of the coupling some divergences arise in the VEVs of the fields (2.10)
due to the perturbation in (1.1) with imaginary coupling. They are generally cancelled
if we add specific counterterms which contain spinless local fields with cutt-off dependent
coefficients. For 0 < 2 < 1 the perturbation becomes relevant and a finite number of lower
scale dimension couterterms are then sufficient. However, this procedure is regularization
scheme dependent, i.e., one can always add finite counterterms. For generic values of
379
this ambiguity in the definition of the renormalized expression for the fields (2.10) can be
eliminated by fixing their scale dimensions to be (2.21). It exists however a set of values of
for which the ambiguity still remains, but here we will not consider these isolated cases.
In the BD model with imaginary coupling, this situation arises if two fields, say O and
O , satisfy the resonance condition:

2
D = D + 2n 1 2 + 2m 1
(2.22)
with (n, m) N
4
associated with the ambiguity:
O O + n O .
m
(2.23)
In this specific case one says that the renormalized field O has an (n | m)th resonance 2 [12] with the field O . Due to the condition (2.19) and using (2.21) we find im 2 ei and the
mediatly that a resonance can appear between the descendent field ()2 ()
following primary fields:
(i)
(ii)
(iii)
(iv)
1
,
2
ei(+2) , i.e., (n | m) = (2 | 0) for = ,

2
ei() , i.e., (n | m) = (0 | 2) for = ,

4
ei(+) ,
ei(+ 2 ) ,
i.e., (n | m) = (1 | 0) for =
i.e., (n | m) = (1 | 1) for = .
(2.24)
If we now look at the expression (2.15), we notice that the contribution (2.16), brought
by the second level descendent field, and that of any of the exponential fields in (i), (ii),
(iii) and (iv), have the same power behaviour in r (r 41 2 +4 ) at short-distance for the
corresponding values of in (2.24). The integrals which appear in these contributions
and their corresponding poles are, respectively:

j1 1 , 2 , 2

j2 1 , 2 , 2

2 2
1
,
,
j2
2
2
4

F1,1 1 , 2 , 2
1
,
2
with the pole = ,

2
with the pole = ,

4
with the pole =
with the pole = .
(2.25)
By analogy with the SG (or ShG) model, one expects that the VEV (2.16) (and similarly
for the real coupling case) exhibits, at least, the same poles in order that the divergent
contributions compensate each other. This last requirement leads for instance to the
2 The same situation arise in any more general affine Toda theory.
380
relations:
(i )
Res= 1 H() = 8 3
2

G+
,
G = 1
2
(ii )
(iii )
(iv )
G+2
(2 2 )

1 2
Res= H() = 32 2 3 2
,
2
2
( )
G =

2
G

2 3 2 ( /2)
2

Res= H() = 4
,
1 /4
4
G =
( 2 /4)
4
G
4
+ 2
Res= H() =
G =
(1 2 )2

Res= F1,1 1 , 2 , 2 .
(2.26)
These last conditions will be used in the next section to fix the normalization of the
VEV (2.16). Let us now turn to the evaluation of (2.16) which plays an important role
in the two-point function (2.15).
3. Reflection relations and descendent fields

The BD model (1.1) can be regarded as two different perturbations of the Liouville field
theory [5]. First, one can consider the Liouville action:

1
2
2
b
(
.
=
d
x
)
+
e
A(1)
(3.1)
L
16
b
The perturbation is then identified with e 2 . Alternatively, we can take:

b
1
(2)
( )2 + e 2
AL = d 2 x
16
(3.2)
as the initial action and consider eb as a perturbation. Using the first picture, the
holomorphic stress-energy tensor:
Q
1
T (z) = ()2 + 2 ,
(3.3)
4
2
ensures the local conformal invariance of the Liouville field theory (3.1) and similarly
for the anti-holomorphic part. The exponential fields ea are spinless primary fields with
conformal dimension:
= a(Q a).
(3.4)
The property of reflection relations which relates operators with the same quantum
numbers is a characteristic of the CFT. Using the CPT framework, one expects that similar
relations are also satisfied in the perturbed case (1.1). With the change b b/2 in (3.3)
and using the second picture (3.2), one assumes that the VEV of the exponential field
ea BD satisfies simultaneously the following two functional equations [5]:

ea BD = R(a) e(Qa) BD ,

ea BD = R (a) e(Q +a) BD
381
(3.5)
with
Q=
1
+b
b
and Q =
2 b
+ .
b 2
(3.6)
The functions R(a), R (a) are called reflection amplitudes. An exact expression for
R(a) was presented in [6]. R (a) is obtained from R(a) by the substitutions b b/2
and . Under certain assumptions about the analytic properties of the VEV, the
system (3.5) was solved and the VEV for these exponential fields was derived in [5].
Let us denote the descendent fields:
L[n] L [m] ea Ln1 Ln1 L m1 L mK ea
(3.7)
where [n] = [n1 , . . . , nN ] and [m] = [m1 , . . . , mK ] are arbitrary strings and Ln , L n
are the standard Virasoro generators:

L n z n2 .
Ln zn2 and T(z) =
T (z) =
(3.8)
nZ
nZ
The descendent fields (3.7) and the ones obtained after the reflection a Q a possess
the same quantum numbers. Consequently, using the arguments of [5,12] based on the
CPT framework, one also expects that their VEVs in the perturbed theory (1.1) satisfy the
following reflection relation:

L[n] L [m] ea BD = R(a) L[n] L [m] e(Qa) BD .
(3.9)
However, it is more convenient to use the basis:
n n m m a
1 N 1 K e .
(3.10)
The main reason is that in (3.9) the components Ln , L n of the modified stress-tensor
depend on a. Using (2.2) one can always express (3.7) in the basis (3.10). For our purpose
we will need the relation [6]:

1
Q
1 2
Q
a
2
2
2
+ a () +
+ a ea .
L2 L2 e = () +
4
2
4
2
(3.11)
Furthermore, using (2.2) it implies:
2

1
2 ea
1 + 2a(Q + 2a) ()2 ()
BD
16
which leads to the following reflection relation:

2

2 ea
1 + 2a(Q + 2a) ()2 ()
BD

2

2 e(Qa) .
= 1 + 2(Q a)(3Q 2a) ()2 ()
BD

L2 L 2 ea
BD
(3.12)
(3.13)
382
One can also consider the second picture (3.2) where the Liouville theory has coupling
b/2 instead of b and is perturbed by eb . If we define the analytic continuation of (2.16):
H (a) =
2 ea BD
()2 ()
,
ea BD
then the two different pictures provide us the following two functional relations:

(2b + 3/b 2a)(3b + 2/b 2a) 2
H (a) =
H (Q a),
(b + 2a)(1/b + 2a)

(b + 6/b 2a)(3b/2 + 4/b 2a) 2
H (a) =
H (Q + a).
(b/2 + 2a)(2/b + 2a)
(3.14)
(3.15)
Notice that these equations are invariant with respect to the symmetry b 2/b with
a a in agreement with the well-known self-duality of the BD-model.
As was shown in the previous section, the solution of these functional equations should
exhibits, at least, the poles (2.25) through the identification b = i and a = i. Since the
solution of (3.15) is defined up to a multiplication constant, we naturally choose to fix
it by imposing Eq. (2.26). We find that the minimal solution which follows from these
constraints is:
4

2 (1/3)
2ba + 4
m (b2 / h) (2/ h)
H (a) =
h
(1/3) 3 22/3+3/2(Q + Q )2 (2b2 / h) (4/ h)

2
2
2ba b
2ba + 3 + b
2ba 1
2ba + 2b2
h
h
h
h

2ba 2
2ba + 2 + 3b2/2
2ba b2/2
(3.16)
,
h
h
h
where h = 6 + 3b2 is the deformed Coxeter number [18,19]. Here we have used the exact
relation between the parameters and in the action (1.1) and the mass of the particle
m [5]:

2/ h

1/ h
2 3 (1/3)
m=
(3.17)
.
1 + b 2
2 1 + b2 /4
2
(1 + b / h) (2/ h)
It is easy to see that for b = i and a = i, H() possess poles located at:

3 3
0 + q
, q Z with
2

3
1
1 3 2
1
, , ,
, , +
,
.
0 ,
2 2 4
2
2
4
But as long as we consider that satisfy (2.19) there remain 3 the expected poles (2.25):

1
, , .
0 ,
(3.18)
2 2 4
3 Notice that for 2 < 2 < 1 one also has the pole 1 + 3 . However, this one is nothing but obtained
3
4
0
from the reflection 0 = ( 1 ) 4 .
383
It is well-known that the BulloughDodd model at the specific value of the coupling
b2 = 2 and the sinh-Gordon model at b2 = 1/2 give an equivalent Lagrangian representation of the same QFT. Then, as expected, one can check that (3.16) evaluated at b2 = 2
coincides exactly with the same quantity in the ShG model evaluated at b2 = 1/2.
Accepting the conjecture (3.16) and using Eq. (3.12) for a = 0 one can easily deduce for
instance:

2
T T BD L2 L 2 I BD = 2 fBD
(3.19)
where
fBD =
m2
16 3 sin(b2 / h) sin(2/ h)
(3.20)
is the bulk free energy of the BD model [5].
4. Comparison with the semi-classical results

As we saw previously, the OPE proposed in Eq. (2.15) plays a crucial role in the
determination of the prefactor of the -dependent part of H(), using Eq. (2.26). It
is therefore, important to check this expression, using for instance the semi-classical
expansion. In what follows, we will compare (2.15) with the semi-classical calculations
based on the action (1.1).
Let us consider (2.15) for 1 = /, 2 = in the classical limit 0. Then the
saddle-point evaluation of the functional integral based on the action (1.1) leads to the field

, where (t) is a solution of the classical
)), t = mr
configuration cl = i2 ((t) + 13 log( 2
2 3
BulloughDodd equation:

t2 + t 1 t = 4 e2 e
(4.1)
with the following asymptotic conditions:
(t) A log t log B + o(1) as t 0, and

4 3
(1 + )
sin
(t)
as t .
sin
K0 2 3t
3
3
(4.2)
Here we denoted:
A 2,
32
B 1 22 ,
3 3
(4.3)
and K0 (x) is the MacDonald function. Such a solution was considered in [20].
Taking into account the above considerations, the two-point function takes the following
form in the semi-classical limit:
i

i
2
e
(x)e (y) BD

(4.4)
=
e
.

i
ei BD e BD 2 0
384
Following [20] it is not difficult to obtain the first few terms in the t 0 expansion:

2
8B
40B 2 4+2A
2(t )
2 2A
22A
2+A
1 2
e
=B t
t
+
t
+
t
B (A 1)2
(A + 2)2
(A + 2)4
8(5A2 4A 28)
1
t 4A +
t 44A
2
2
2
(A 1) (A + 2) (A 4) B
(A 1)4 B 4
6+3A 6
+O t
,t
.
(4.5)
We would like now to compare these results with the corresponding limit of (2.15). First,
using the result for the exact VEV in the BulloughDodd model (2.18) proposed in [5] we
obtain the following ratios:
1
2
( 3 ) ( 22
G/+
3 )
= m4
,
G/ G
24 36

22 2+2n
m4 +4n +2n ( 1
3 ) ( 3 )
n G/++n
=
,
G/ G
2n n
24 +4n +2n 36 +6n +n

22 2n
m4 2n +2n ( 1
3 ) ( 3 )
n G/+n/2
=
,
G/ G
2n n
24 2n +n 36 3n +n
G/++(n1/2)
n
G/ G
1
2+2n1
4
( 3 ) ( 22
m +4n 2 +2n+2
3 )
=
.
(4.6)
2n+2 n+1
24 +(4n2) +2n+136 +(6n3) +n+1
Furthermore, the mass- relation (3.17) proposed in [5] gives:

m 6 4
2

3 6
2 0 2 3
(4.7)
whereas, using Eq. (2.9) and F1,1 (1 , 2 , 2 ) which can be deduced from the results
of [21], we have:

jn , 2 , 2
n 2n
(2 + n)
,
2n
2
(2)
0 n! (1 + 2 )

jn /2, 2/2, 2 /4
2 0
n 2n
n!
22n (1 )2n
(4 + n)
,
(4)

F1,1 , 2 , 2
2 4
2 0 (1 + 2 )2 (1 )2 (2 + )2

2 (2 + 7) + (8 + 10) + 8 + 1 ,

H(/ + ) O 4 .
2 0
(4.8)
If we now use the same notations as above (4.3), the semi-classical limit of the
expression (2.15) takes the following form:
385
G/,
t 2A B 2
G/ G 2 0

8B 2+A
(2 + 1) 44A
2
t 22A
t
+ 4
t
1+ 2
B (1 A)2
(2 + A)2
B (1 A)4
8(4 1)B 2 4+2A
t
(2 + A)4

2
8 A (2 + 7) A(16 + 20) + 32 + 4 4A
t
B(1 A)2 (2 + A)2 (4 A)2

+ O t6 .
+
(4.9)
It is straightforward to check that this result agrees perfectly with (4.4) through (4.5).
5. Expectation values of the descendent fields in 12 , 21 and 15 perturbed

minimal models
For imaginary value of the coupling b = i, and the action of the
BD model (1.1) becomes complex. Whereas it is not clear if it can be defined as a QFT, this
model is known to be integrable and its S-matrix was
constructed
in [13]. It is known that
(2)

this model possess a quantum group symmetry Uq A2 with deformation parameter q =

i
e 2 [13,14]. An important role is played by one of its subalgebras Uq (sl2 ) Uq A(2)
2 .
Following [13] (see also [5]), we can restrict the Hilbert space of states of the complex BD
model at special values of the coupling constant, more precisely when q is a root of unity,
i.e., for:
2 =
p
p
or 2 =
p
p
with 1 < p < p
(5.1)
relative prime integers, in which case the complex BD is identified with the perturbed
minimal models (1.2) or (1.3), respectively. In the following, lk will denote a primary
field of the minimal model Mp/p .
In the first case, the exact relation between the parameters in (1.2) and the mass of the
fundamental kink M can be found in [5] with the result:
1

2 +2
3+6
+4
1
2
+1
+1
4 +4
2
+1
3 +6
2 = +5
(5.2)
.

3 +4 1

1
2
3 13 3+6
2 +1 4 +4 2 + +1
Here we denote
p
=
.
p p
(5.3)
For unitary minimal models > 1 which, for Im() = 0, corresponds to the massive
phase [5]. Using the particle-breather identification:

m = 2M sin
(5.4)
3 + 6
386
and Eqs. (3.12), (3.16) for imaginary coupling b = i and parameter a = i( l1

2
it is straightforward to get the VEV in the model associated with the action (1.2):
k1
2 ),
0s |L2 L 2 lk |0s

0s |lk |0s
4

3 ( + 2)M 1 + 2+2

2 (1/3)
3 +6
=
1

2 4+4 W12 ( + 1)l k
3 22/3+1/2 3 +6
3 +6 3 +6
(5.5)
with
W12 () =
2 (
1
w(; 5 + 4, 4 + 2, 1 2, 1 + /2; 3 + 6)
+ 1)2
where we introduce the useful notation:

4

ai
ai +
.
w(; a1 , a2 , a3 , a4 ; g) =
g
g
i=1
Here |0s is one of the degenerate ground states of the QFT (1.2) (see [5] for a detailed
discussion of the vacuum structure of the model). Taking lk in (5.5) to be the identity
operator, it is easy to get:

2 M 4 sin2
3
+6
T T =
(5.6)

.
48 sin2 (2 +2)
3 +6
A simple check consists to consider the scaling LeeYang model which corresponds to
p = 2, p = 5, i.e., = 2/3 in (1.2). As 12 13 for these values, we must obtain the
result of [12]. Using (5.4) the lightest mass in (1.2) is:

m = 2M sin
.
12
Replacing this expression in (5.5) for l = 1, k = 3 and (5.6) it is easy to see that the results
are in perfect agreement with the ones of [12].
In the second restriction 2 = p /p, which leads to the action (1.3). The exact relation
between the parameter and the mass of the fundamental kink M is in this case [5]:
2
+1 1 1
33

2+
2 4
M 323
.
3 1 1 1 1 +1
2 4 2
3 3 3 3
(5.7)
Along the same line as for the 12 perturbation we obtain the following expression for the
VEV in the model associated with the action (1.3):
0s |L2 L 2 lk |0s
0s |lk |0s

2 4
3 (1 )M 1 33

2 (1/3)
=
W ( + 1)l k (5.8)

+1
2 +2 4 21
1 2/3+1/2
( 3 )2
33
33 33
387
with
1
w(; 1 4, 2 2, 1 + 2, 1/2 /2; 3 3 )
+ 1)2
where |0s is one of the degenerate ground states [5] of the QFT (1.3). The analog of the
formula (5.6) is now:

2 M 4 sin2 (1+ )

33
T T =
(5.9)

.
48 sin2 2
33
(2)
Another subalgebra of Uq A2 is the subalgebra Uq 4 (sl2 ). One can again restrict the
phase space of the complex BD with respect to this subalgebra for a special value of the
coupling:
W21 () =
2 (
4p
(5.10)
with 2p < p
p
relative prime integers. Then, for this value of the coupling, the BD model is identified
with the perturbed minimal model with the action (1.4). The exact relation between and
the mass 4 m is [5]:
4
1 5
2
1+ 1+
(1
+
)
1+
2 =
3 14
2(1 4 )(1 2 )
1+ 1+
1+
6(1)

1+ 1+
m 35
33 33
(5.11)
.

2 3 13
2 =
Here we keep the definition (5.3). In particular, the massive phase corresponds to:
1
Im = 0,
0< < ,
4
1
3
< < ,
Re = 0.
4
5
As for the two previous cases one would like to obtain the expectation values of the
descendent fields for any primary operator lk . For l > 1 these fields are not invariant
with respect to the subalgebra Uq 4 (sl2 ) on the contrary to 1k . However, one expects that
they only differ by a c-number coefficient characterizing the degenerate structure of the
vacua |0s . Taking the ratio of the VEV of the descendent field of lk associated with the
action (1.4) and the VEV of the primary field itself, one obtains:
0s |L2 L 2 lk |0s
0s |lk |0s

1+
2 4
33
m 1 + 33

2 (1/3)
=
W ( + 1)l k
1

4 2+2 15
2/3+1/2
3 32
33 33
(5.12)
4 The general vacuum structure in the model (1.4) is not clearly understood. However, it is expected that it
possesses particles and kinks similarly to the other models (see for instance Refs. [16] for details). The physical
mass scale m is then associated with one of its particles [5].
388
with
W15 () =
1
w(; + 5, 4 4, 1 5, 1 ; 6 6 ).
2 ( + 1)2
In particular, we have:

2
T T =

768 sin2
m4

.
2
2 (1+ )
33 sin
33
(5.13)
In conclusion, we proposed in this paper an exact expression for the VEV of the
2 ei in the BD model. The
second level descendent of the exponential field ()2 ()
calculation is based on the so-called reflection relations which lead to a system of
functional equations for this VEV. While the solution is not unique, we chose the minimal
solution obeying some residue conditions. By performing a quantum group restriction in
the case of complex BD model we found also the VEVs of the descendents of the primary
fields in the perturbed minimal CFT models (1.2), (1.3), (1.4).
It is rather interesting to notice that in Eq. (3.19), the exact VEV T TBD is simply
related to the VEV of the trace of the energymomentum tensor:
1
= T = (1 pert )pert
4
where 1 pert = h/4 and pert = e 2 as follows:

T T BD = 2BD .
b
(6.1)
This was already noticed in [12] for the ShG case. We expect this property to be general,
i.e., to be confirmed for other integrable theories. However, we have no proof yet of this
phenomena.
We would like to notice two important differences between ShG and BD models. First,
in the 2 expansion of the two-point function the quantity H() (2.16) comes with a
coefficient of order 2 . Therefore, it cannot be checked directly in the semi-classical
approximation, although the later is in agreement with the short distance expansion of
the two-point function, thus giving a strong support to our conjecture (3.16). For a direct
check one has to go beyond the classical limit and consider the first order in the perturbation
theory based on the action (1.1).
Another difference is the appearing of the third level descendents in the OPE of the
exponential fields. As a consequence, the following quantity appear in the short distance
expansion of the two-point function:
K() =
3 ei BD
()3 ()
.
i
e BD
(6.2)
In contrast with H() it is sensitive to the semi-classical expansion it combines with

the integral F1,2 in (2.15) in order to match the corresponding term coming from the
389
semiclassical calculation based on the action (1.1). The function (6.2) can also be obtained
using the reflection relations approach.
Using our results, one can deduce easily the next-to leading contributions in the short
distance behaviour of the two-point functions of primary operators for different perturbed
minimal models: for instance the Ising model in a magnetic field [22], the tricritical Ising
model perturbed by its leading energy density operator (with conformal dimension 12 =
1/10) [21] or perturbed by its subleading magnetic operator (with conformal dimension
21 = 7/16), and so on.
Several models can be worked out along the same line, using the known results for the
three-point functions of the CFT. For instance, the super ShG model or, more generally,
the parafermionic ShG model [8,23].
We intend to discuss these various questions in a forthcoming publication [24].
Acknowledgements
We are grateful to Al.B. Zamolodchikov and particularly V.A. Fateev for valuable
discussions and interest in this work. We thanks for the hospitality of LPM (Montpellier)
where part of this work was done. M.S. acknowledges the Physics Department of Bologna
University and APCTP, Seoul, for hospitality and financial support. PBs work is supported
in part by the EU under contract ERBFMRX CT960012 and Marie Curie fellowship
HPMF-CT-1999-00094. The work of M.S. is supported in part by KOSEF grant 19992-112-001-05.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
M. Shifman, A. Vainstein, V. Zakharov, Nucl. Phys. B 147 (1979) 385.

Al.B. Zamolodchikov, Nucl. Phys. B 348 (1991) 619.
S. Lukyanov, A.B. Zamolodchikov, Nucl. Phys. B 493 (1997) 571.
V.A. Fateev, S. Lukyanov, A.B. Zamolodchikov, Al.B. Zamolodchikov, Phys. Lett. B 406
(1997) 83.
V.A. Fateev, S. Lukyanov, A.B. Zamolodchikov, Al.B. Zamolodchikov, Nucl. Phys. B 516
(1998) 652.
A.B. Zamolodchikov, Al.B. Zamolodchikov, Nucl. Phys. B 477 (1996) 577.
V.A. Fateev, Nucl. Phys. B 473 (1996) 509.
P. Baseilhac, V.A. Fateev, Nucl. Phys. B 532 (1998) 567.
V.A. Fateev, Mod. Phys. Lett. A 15 (2000) 259.
C. Ahn, P. Baseilhac, V.A. Fateev, C. Kim, C. Rim, Phys. Lett. B 481 (2000) 114.
C. Ahn, P. Baseilhac, C. Kim, C. Rim, hep-th/0102024.
V.A. Fateev, D. Fradkin, S. Lukyanov, A.B. Zamolodchikov, Al.B. Zamolodchikov, Nucl. Phys.
B 540 (1999) 587.
F. Smirnov, Int. J. Mod. Phys. A 6 (1991) 1407.
C.J. Efthimiou, Nucl. Phys. B 398 (1993) 697.
M.J. Martins, Phys. Lett. B 262 (1991) 39;
A. Koubek, M.J. Martins, G. Mussardo, Nucl. Phys. B 368 (1992) 591.
390
[16] G. Takacs, Nucl. Phys. B 489 (1997) 532;

H.G. Kausch, G. Takacs, G.M.T. Watts, Nucl. Phys. B 489 (1997) 557;
G. Takacs, G.M.T. Watts, Nucl. Phys. B 547 (1999) 538.
[17] V.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312;
V.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 251 (1985) 691.
[18] G.W. Delius, M.T. Grisaru, D. Zanon, Nucl. Phys. B 382 (1992) 365.
[19] E. Corrigan, P.E. Dorey, R. Sasaki, Nucl. Phys. B 408 (1993) 579.
[20] A.V. Kitaev, J. Sov. Math. 46 (1989) 2077;
C. Tracy, H. Widom, Physica A 244 (1997) 402;
C. Tracy, H. Widom, Commun. Math. Phys. 190 (1998) 697.
[21] R. Guida, N. Magnoli, Int. J. Mod. Phys. A 13 (1998) 1145.
[22] R. Guida, N. Magnoli, Nucl. Phys. B 483 (1997) 563.
[23] C. Ahn, D. Bernard, A. LeClair, Nucl. Phys. B 346 (1990) 409.
[24] P. Baseilhac, M. Stanishkov, work in progress.

Finite temperature expectation values of local fields

in the sinh-Gordon model
Sergei Lukyanov a,b
a Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855-0849, USA
b L.D. Landau Institute for Theoretical Physics, Kosygina 2, Moscow, Russia
Abstract
Sklyanins method of separation of variables is employed in a calculation of finite temperature
expectation values. An essential element of the approach is Baxters Q-function. We propose its
explicit form corresponding to the ground state of the sinh-Gordon theory. With the method of
separation of variables we calculate the finite temperature expectation values of the exponential fields
to one-loop order of the semi-classical expansion. 2001 Published by Elsevier Science B.V.
PACS: 11.10.Kk
Keywords: Integrable quantum field theory; Sinh-Gordon model; Finite temperature expectation values; Method
of separation of variables
1. Introduction
One-point functions have numerous applications in statistical mechanics and condensed
matter physics [1,2]. They determine generalized susceptibilities, i.e., the linear response
of a system to external fields. In a path integral formulation the one-point function of a local
field O is represented by a Euclidean path integral of the form

1
D[]OeA .
O = Z
(1)
Recently some progress has been achieved in the calculation of one-point functions in
integrable quantum field theory (QFT) defined on a two-dimensional Euclidean plane [3,
4]. In this case, the integral (1) can also be viewed as a vacuum expectation value (VEV) in
(1 + 1)-dimensional QFT associated with the action A. For many applications, especially
in condensed matter physics, it is important to generalize the results of Refs. [3,4] to the
case of Euclidean path integrals defined on an infinite cylinder. In the Matsubara imaginary
E-mail address: sergei@physics.rutgers.edu (S. Lukyanov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 6 5 - 0
392
S. Lukyanov / Nuclear Physics B 612 [FS] (2001) 391412
time formalism such path integrals are interpreted as thermal averages

Tr[eRH O]
(2)
Tr[eRH ],
where H is the Hamiltonian of the corresponding QFT and the temperature coincides with
the inverse circumference of the cylinder.
The path integral defined on a cylinder also allows another interpretation. It is an
expectation value for the ground state |vacR of the (1 + 1)-dimensional theory in the finite
geometry where the spatial coordinate is compactified on a circle. Hence, the VEVs contain
important information about renormalization group flow controlled by the parameter R.
An exact calculation of the finite volume (finite R) VEVs is a challenge even in
integrable QFT. Recent progress made in papers [5,6] should be mentioned here. In [6]
Leclair and Mussardo proposed an integral representation which makes it possible to
generate a low-temperature (R ) expansion for the VEVs in terms of infinite volume
form-factors of local fields and some thermodynamical data. Their conjecture works for
theories with trivial S-matrices such as Ising and free Dirac fermion models [7], but its
validity remains questionable for models with non-trivial scattering amplitudes (see, e.g.,
[8]). Another line of research was proposed in the work [5]. Smirnov applied the method of
separation of variables [9,10] to the semi-classical study of finite volume matrix elements
in the quantum KdV equation. The model does not constitute a relativistic field theory.
Nevertheless, it is of prime importance for the sinh-Gordon QFT since both of the equations
are in the same integrable hierarchy.
In this paper we will implement the method of separation of variables in the case of the
quantum sinh-Gordon theory. The problem is defined by a Euclidean action,
OR =

AshG =
R
dx1
dx2

1
( )2 + 2 cosh(b) ,
16
(3)
where is a scalar field with periodic boundary condition along the x2 -coordinate. We are
focusing on the VEVs of the exponential fields,
O = ea .
For our purposes it will be useful to rewrite (2) in the form,

a
e R = Z 1 D[]02 []ea .
(4)
(5)
Here 0 [] is an integral taken over field configurations on the half-cylinder, x1 < 0,

satisfying the boundary condition,
(x1 , x2 )|x1 =0 = (x2 ),
i.e., it is a wave functional corresponding to the ground state |vacR . The method of
separation of variables [5,9,10] allows one to introduce a change of integration variables
in (5),
(x2 ) {k }
k= ,
393
from the function (x2 ) to the infinite discrete set of k . A notable advantage of the new
variables is that the wave functional in the -representation has a factorizable form,

0 [{k }]
Q[k ].
k
Notice that the integration measure D[{k }] does not factorize in the variables k . At this
moment, we are not able to elaborate on all steps of the changing of variables on a rigorous
basis. Therefore, we suggest the deduced integral representation for ea R as a conjecture
rather than a well established result. To test the validity of this integral representation, we
carry out a semi-classical expansion of the VEV.
More explicitly, the parameter b2 in the action (3) can readily be identified with
the Planck constant. Then, for finite = ab and b2 0, the functional integral (1)
is dominated by a non-trivial saddle-point configuration and admits the semi-classical
expansion,

a
2
e R = eS/b D 1 + O b2 .
(6)
Here S is a Euclidean action on the cylinder evaluated in the saddle-point configuration
and the pre-exponential factor D is the result of evaluating the functional integral (1)
in the Gaussian approximation around the classical solution. With the proposed integral
representation we calculate the functions S and D and find complete agreement with the
expected high- and low-temperature behavior of the VEVs. In particular, our result matches
well with the LeclairMussardo conjecture.
2. Integral representation for VEVs

2.1. FlaschkaMcLaughlin variables
In the paper [11] Flaschka and McLaughlin found remarkable canonically conjugate
variables in the phase spaces of the classical Toda chain and KdV equations. Their
approach can be straightforwardly adapted to the classical sinh-Gordon equation. Here
we give a brief review of the FlaschkaMcLaughlin variables for this model. For more
information and proofs, the reader is referred to Refs. [11,12].
The sinh-Gordon equation admits a zero curvature formulation: there exists a sl(2, R)valued connection 1-form, depending on an auxiliary parameter , such that the condition
of vanishing curvature is equivalent to the equation of motion. Dealing with the theory on
cylinder, one can integrate this 1-form along some cycle, say,
x1 = 0,
0 x2 < R,
and obtain the so-called monodromy matrix,

A()
B()
SL(2, R)
M() =
1 C() D()
(7)
(m = 0).
(8)
This matrix satisfies important analytical conditions which are readily obtained from an
explicit form of the connection. In particular, the matrix elements in (8) are real analytical
394
functions of the variable 2 with two essential singularities at the points 2 = 0, . Zeroes
of B(),
2k : B(k ) = 0,
(9)
are of prime importance in the construction. It is possible to show that all zeroes (9) are
simple, real, positive and accumulate towards the essential singularities. Thus we can order
them,
0 2N < 2N+1 < < 20 < < 2N1 < 2N +
and define two infinite sets:
2
{k }
k= : k = log k ,
{k }
k= : k = 4 log |A(k )|.
(10)
The mapping of the canonical Poisson data, and x1 , to the variables (10) is found to
be a canonical transformation, i.e.,
{k , m } = km ,
{k , m } = {k , m } = 0.
(11)
Hence (10) are canonically conjugate variables

in the phase space of the sinh-Gordon
model. At the same time, we can treat k k= as coordinates in the corresponding
configuration space.
2.2. -representation
The FlaschkaMcLaughlin variables proved to be useful in quantum theory as was
demonstrated in the seminal work [9] on the example of the Toda chain equation (see also
[13,14]). We refer to a quantization in these variables as a quantization in -representation.
Recently the -representation was employed to quantize real KdV theory [5]. 1 In fact,
Smirnov presented heuristic, but convincing, model-independent arguments which can
also be applied to the sinh-Gordon equation. Following these arguments we introduce the
integral,
+
N

dk
IN (R, a) =
b
k=N

Nk>mN
sinh(k m )
N

Q2 [k ]e2k /b
2 (ab+k)
. (12)
k=N
We shall also use slightly different form of IN : With the identity,

k m
exp 2kj
=
Det
2 sinh
,

2
2
b
b
Nj,kN
Nk>mN
1 The monodromy matrix in the real KdV model has the form (8), whereas the monodromy matrix of the
imaginary equation belongs to the group SU(2)(m = 0). The imaginary KdV model is related to the sineGordon theory and (perturbed) CFT with the central charge c < 1 (see, e.g., [15]). A sensible -representation
for the imaginary equation has, to our knowledge, not been found.
395
the integral (12) can be rewritten in the form

+
N

dk
1
IN (R, a) =
(2N + 1)!
b
k=N
N

k m 2
2 sinh(k m ) sinh
Q [k ]e2k a/b .
b2
k=N
Nk>mN
(13)
The function Q[ ] appearing in Eqs. (12), (13) is the so-called Baxters Q-function. It is a
non-singular function for real with leading asymptotic behavior

Q[ ] e2C0 cosh(bq )
as .
(14)
Here and below we use the notation,

q = b + b1 .
The positive constant C0 in (14) reads explicitly,
C0 =
mR
,
4 sin(b/q)
and m is a mass of the sinh-Gordon particle. Therefore, IN (13) is a convergent integral

for any finite N . Notice that the configuration space of the model under consideration
is an infinite-dimensional space. In writing the (2N + 1)-fold integral, we truncate it to
(2N + 1)-dimensional space with coordinates {k }N
k=N . As well as in the KdV theory [5],
we can treat
N [{k }] =
N

Q[k ]
k=N
as a wave functional in the -representation. For N , it corresponds to the ground

state |vacR of the sinh-Gordon theory with periodic boundary conditions. Furthermore,
the double product in (13) is an integration measure and we shall denote it DN [{ }]. It was
calculated in the semi-classical approximation in [5]. The semi-classical analysis suggests
also that the product
ON =
N

e2k a/b
k=N
represents (in the limit N ) the exponential field ea located at the point (R/2, R/2)
on the cylinder. 2 Therefore, the integral (13) has the form of a quantum mechanical
diagonal matrix element,

IN = DN [{ }]N ON N .
(15)
2 The position of the insertion is determined by choosing of the integration contour for the monodromy matrix.
Here we assume that the contour is given by (7).
396
We shall consider the vacuum expectation values only. In this case, Q is an eigenvalue of
the Baxter Q-operator corresponding to the ground state.
2.3. Baxters Q-function in sinh-Gordon theory
We now turn to the most delicate point of our construction: an explicit form of the
function Q. To the best of our knowledge a rigorous derivation of Q is not currently
available. Here we formulate a conjecture for the sinh-Gordon Q-function based on the
following heuristic arguments. First, we note that the substitution b i transforms AshG
(3) to the action of the sine-Gordon model. Naively we could try to obtain Q by means
of analytical continuation from 1 < b2 < 0. In this coupling constant domain, the Baxter
Q-operator is relatively well studied [15]. Unfortunately, the sine-Gordon Q-function has
an essential singularity at b2 = 0 the analytical structure of which is unknown. This makes
the analytical continuation to the domain of the sinh-Gordon model a highly questionable
procedure. One can guess an explicit form for Q by examining its asymptotic behavior.
The Q-function in the sine-Gordon model admits the following asymptotic expansion for
+ [15], 3
log QsinG C0 e
Cn I2n1 e(2n1) +
n=1
n e2n(b2 +1)
G

1 < b2 < 0 .
n=1
Here we introduce a new notation,

bq.
(16)
The leading term of this expansion has already appeared in our consideration (see (14))
n are vacuum eigenvalues of the so-called local and dual unlocal
while I2n1 and G
integrals of motion (IM), respectively. The constants Cn depend on the normalization of
the local IM. In Appendix A (see Eq. (A.2)) we present their form for the normalization
adopted in [15]. A similar asymptotic form holds for . The eigenvalues I2n1 are
regular functions of b2 , and can be continued to the domain of the sinh-Gordon model
n , are
without problems. Contrary to I2n1 , the eigenvalues of the dual unlocal IM, G
2
highly singular functions at b = 0. One can expect that the appearance of these IM is
a consequence of the existence of the soliton sector of the sine-Gordon QFT. This sector is
absent for b2 > 0. All these observations suggest the following large asymptotic behavior
in the sinh-Gordon model,
log Q C0 e
Cn I2n1 e(2n1)

b2 > 0 .
(17)
n=1
In Appendix A we give numerical evidence that the values of local IM can be expressed in
terms of a solution of the thermodynamic Bethe ansatz (TBA) equation:
3 In this work we use a convention for the Q-function which differs from the one of [15] by an overall shift of
the argument.

Cn I2n1 = C0 n1 + (1)

d (2n1)
e
log 1 + e+() .
397
(18)
Here the function +( ) solves the TBA equation [1618]

+( ) mR cosh( ) +

d

( ) log 1 + e+( ) = 0,
2
(19)
with the kernel

4 sin(b/q) cosh( )
(20)
.
cosh(2 ) cos(2b/q)
The series (17) is an asymptotic expansion. In fact, it is a divergent geometrical series
which can easily be summed up and one can guess an explicit form of Q:
( ) =
mR
log Q( ) =
cosh( ) +
2 sin(b/q)
d log(1 + e+( ) )
.
2 cosh( )
(21)
We leave an examination of the properties of this function for future publications.

One more aspect of the -representation deserves a comment. The sinh-Gordon model
manifests an important non-perturbative symmetry. The couplings b and b1 correspond
to physically indistinguishable theories. Q in (21) is a self-dual function in a sense that it
is invariant under the substitution b b1 . Furthermore, it is easy to see that
IN |b = IN |b1 .
This supports our choice of the measure in (13). Strictly speaking, the measure was
obtained in [5] in the semi-classical approximation. The exact invariance of the semiclassical measure suggests its applicability for an arbitrary value of the coupling constant
b2 .
2.4. Large N limit
As was noted above, in writing the (2N + 1)-fold integral (13), we truncate the
configuration space of the theory to (2N + 1)-dimensional one. In fact, the truncation
amounts to an ultraviolet regularization with a momentum cutoff given by
2N
.
R
Now we let N . From (15), it is clear that the ratio
N =
IN (R, a)

IN (R, a) =
(22)
IN (R, 0)
represents the VEV ea R in the large N limit. More explicitly, dimensional analysis
suggests that

2
N 2a

,
IN (R, a)
m
398
thus the correct relation has the form
2

a
4N 2a
e R = a lim
IN (R, a).
N mR
(23)
Here a is an arbitrary R-independent constant. To eliminate this ambiguity one has

to impose some normalization condition on the fields. For example, the so-called
conformal normalization stipulates that the exponential fields with sufficiently small |a|
are normalized in accordance with the short distance behavior of the two-point function

a
2
e (x)ea (y) |x y|4a as |x y| 0.
The key result of Refs. [3,4] is a calculation of the limit,

lim ea R = Ga
(24)
with this normalization. Explicitly,

m0( 1 )0(1 + b ) 2a 2
2bq
2q
Ga =
4

dt
sinh2 (2abt)
2 2t
.
+ 2a e
exp
t
2 sinh(b2 t) sinh(t) cosh(qbt)
(25)
Once we adopt the conformal normalization for the exponential fields, the constant a is
uniquely determined by the condition (24). In particular, the semi-classical consideration
(see below) leads to the relation,

a = Ga 1 + O b2 .
3. Semi-classical expansion
In this section we will study the VEVs in the semi-classical approximation. The VEV
(23) can be represented by the Euclidean path integral (1) on a cylinder with A and O
given by (3), (4). For fixed ,
= ab
(26)
and b2 0, the path integral is dominated by a saddle-point configuration = b and the

VEV has the form (6), where S coincides with the regularized Euclidean classical action
on the cylinder:

d 2 x ( )2
ds
2
2
2
+ m sinh
+
2 log .
S = lim
0
8
2
2
2
|xy|>
|xy|=
(27)
Here the function is a solution of the classical equation of motion,
2 = m2 sinh(),
(28)
399
such that,
4 log |x y| + O(1) as |x y| 0,
(29)
and
0
as |x y| ,
(x1 , x2 + R) = (x1 , x2 ).
(30)
The field configuration develops a singularity at the point y where the exponential field is
inserted. Therefore, in the definition (27) we cut the small disc of radius around this point
and add the boundary term to the action to ensure (29). We also add a field independent
term such that the action is finite at 0. The pre-exponential factor D (6) is a result of
evaluating the path integral (1) in the Gaussian approximation around the classical solution
defined above,
2
2

+ m2 cosh() 1/2
m E 2
e
Det
,
D=
(31)
2
2 + m2
where E = 0.577216 . . . is the Euler constant. The first factor in (31) appears as a result
of the mass renormalization.
3.1. Main semi-classical order
We now proceed to the semi-classical calculation of the VEV ea using the
representation (23). In order to apply the saddle-point machinery, it is convenient to begin
with the form (12) for the integral IN . To the lowest semi-classical order,
Q[ ] er cosh( )/(2b ) .
2
(32)
Here and below the notation

r = mR
(33)
is used. The corresponding saddle-point equations have the form,

r sinh(k ) = 2(k + ),
k = 0, 1, . . ., N.
(34)
In writing (34) we have assumed that b2 0 and = ab is fixed. Thus we find,

log IN =

N
1 r
cosh(
)
2
(k
+
)
+ O(1).
k
k
b2
k=N
The sum can be evaluated with the result,

N

r
cosh(k ) 2k (k + )
k=N

2
r
4Ne
=T
log
+ 2 2 log N
2
r
(35)
400
1
4
+ 2N(N + 1) + + 2 2 log
3
r

N

1
k log(k/e) 4 log AG + O
+4
,
N
(36)
k=1
where AG = 1.282427 . . . is the Glaisher constant and the function T = T (r, ) reads
explicitly,

T =r

d
r cosh( )+2i
.
sinh(
)
e
log
1
e
2
(37)
Now, combining Eqs. (23), (35), (36) one obtains

a
2
e R eS/b ,
with
S = S0 () T (r, ) + T (r, 0),
(38)

dt sinh2 (2t)
m
2 2t
S0 = 2 log
2 e
.
+
4
t
t sinh(2t)
(39)
and
2
Thus the function (38) coincides with the regularized Euclidean action (27). 4
3.2. Semi-classical expansion of Q-function
To compute the VEV to one-loop order, we have to find the next term in the semiclassical expansion of Q. It can be obtained by means of the TBA equation (19). The
kernel (20) allows an expansion in b2 ,
( ) = 2( ) + 2b2 P.V.
cosh( )
2
sinh ( )

+ O b4 ,
(40)
thus to lowest order the solution of the TBA equation has the form,

e+ = er cosh() 1 + O b2 .
With this equation and the definition (21) we calculate
r cosh( ) r cosh( ) r sinh( )
+
2
2
2b2

d log(1 er cosh( ) )
+ O b2 .
2
cosh( )
log Q[ ] =
4 This result was obtained by a different method in [7].
(41)
401
In order to see an analytical structure of this function it is instructive to represent Q in the

form of an infinite product,
E
r cosh( )
r cosh( )
2
1
re
2
= e 2 b
Q[ ]
4

2

r
r cosh( )
2r cosh
1+
+
2
2n
2n
n=1

r cosh( )
e 2 n 1 + O b2 .
(42)
Here E is the Euler constant.

3.3. One-loop order
The saddle-point approximation allows one to find the one-loop order in the semiclassical expansion. To this order,

IN = W
sinh(k m )
N

Q2 [k ]e2k (+k) 1 + O b2 .
(43)
k=N
Nk>mN
Here the function W is a result of the Gaussian integrations in (12) around the saddle points
k = k (34),
N

N+1/2
W = 2 2
k=N
1
.
r cosh(k )
The main steps in the calculation of (8) are given in Appendix B. Our final result has the
form (6) with the function S given by (38) and
log D = log D0 + T (r, ) T (r, 0) T (r, )

2 2
1 2
dr
2

r (r T )2
.
T
=0
8
2 2 =0
(44)
Here
1
log D0 = 2 log 2 +
2
dt
sinh2 (2t)
cosh2 (t)
Recall that (44) should coincide with the functional determinant (31).
4. High-temperature behavior
Now that we have computed (6), let us check the result for some limiting cases. Here we
argue for R 0 behavior of the VEVs.
402
Due to the scaling properties of the interaction operator in (3) one can rescale the
problem to a circle of circumference 2 . Thus, the Hamiltonian of the model under
consideration takes the form
2
HshG =
R
2

2bq

b

1
R
2
2
b
( ) +
e +e
d 4 +
,
16
2
(45)
is the momentum conjugate to = |x1 =0 . The mass of the sinh-Gordon

where =
particle is related to the parameter by [19],

1

b 2bq
0 1 + 2q
m0 2bq
0(b 2 )
.
=
(46)
0(1 + b2 )
4
1
i
For r 0 and a > 0 the main contribution in the path integral (5) comes from a region of
the configuration space corresponding to
2q log(r) 1.
(47)
In this region we can neglect the term eb in the Hamiltonian (1), and approximate
the ground state wave functional 0 [] by a proper wave functional from the Liouville
conformal field theory (CFT).
More explicitly, the Hilbert space of the Liouville CFT contains a continuous set of
primary states |p parameterized by p > 0 with the conformal dimension [20],
p = p 2 +
q2
.
4
(48)
We will assume that these states are canonically normalized,

p |p = 2(p p ).
Let p [] be a normalized wave functional corresponding to the state |p. As was
discussed in [17,18], the following relation holds
0 [] Lp p []
in the region (47). Here p = p(r) solves the equation,

1
1
r0 2bq 0 1 + 2bq
p(r): 2pq log
=
8 3/2(b2 )1/(bq)

+ m log 0(1 + 2ip/b)0(1 + 2ipb) .

2
(49)
(50)
We emphasize that Lp is a unique coefficient provided a normalization of 0 is chosen.

Therefore, one can expect the following relation for r 1:
ea

R
Lp Lp
p|ea |pLiouv
.
R vac|vacR
(51)
403
The matrix element p|ea |pLiouv was found in [18,21]. It reads explicitly,

R
p|e |pLiouv =
2
2a(qa)
0(b2 )b22b
0(1 b2 )
a/b
0 (2a) (2ip) (2ip)

.
2 (a) (a + ip) (a ip)
(52)
Here we use the notations [18]
2

q
a log(2b)
log (a) =
2
2

q
dt
sinh2 ((qb 2ab)t)
2t
a e
+
,
t
2
sinh(2t) sinh(2tb2 )
0
and
0 = a (a)|a=0.
It is easy to see that the function p = p(r) (50) satisfies the condition,
lim p(r) = 0.
r0
Using Eqs. (4.7)(4.9), we can derive

2aq
1

2a(aq)
b
0 1 + 2q
m0 2bq
a
(2a)03
R
2
e R N
,
b2aq
2
4 (a)
4
(53)
(54)
where
N2=
limp0 {4p2 Lp Lp }
R vac|vacR
(55)
does not depend on a. In writing (54) we also used the relation (46). Unfortunately, the
function N = N (r, b) is not known in closed form for an arbitrary value of the coupling
constant. One can obtain its limiting value as b 2 0. For b2 r 1, it is sufficient to
consider the dynamics of the zero-mode X [17,18]:
2
X=
d
( ).
2
(56)
In this approximation, known as the mini-superspace approach [22], the Hamiltonian (1)
is substituted by,

2
2
r
2
Hms =
cosh(bX) .
2X +
R
4b
The Schrdinger equation
Hms 0 (X) = Ems 0 (X),
404
coincides with the modified Mathieu equation and the wave functional 0 is represented
by its lowest eigenfunction. We will use the common normalization condition

dX 02 (X) = 1.
(57)
The Liouville wave functionals p in the mini-superspace approximation have the form,
2ip/b

2
r
r
bX/2
K2ip/b
e
p (X) =
.
0(2ip/b) 8b2
4b2
Here K (z) is the MacDonald function. Now it is clear that the mini-superspace
approximation for N (55) can be obtained from the large X behavior of the normalized
Mathieu function 0 (X) (38),

r
2 Nms
exp
0 (X)
(58)
cosh(bX/2)
as X .
r cosh(bX/4)
2b2
Of concern to us is the behavior of Nms in the domain b2 r 1. In this case, we
replace 0 (X) by its WKB asymptotic (58) and readily obtain,
3/2
2r
2
lim N =
(59)
,
r 1.
2
(2)3
b 0
Having arrived at Eq. (59), we can straightforwardly expand (54) in a power series of b2 ,

2

a
b r 1 ,
e R F 1 + O b2
(60)
with

1
4S0 (1/2 ) S0 (1/2 2)
2
exp
2b2

3
9
r 2 2 0(1 2) 0 2 ()
2+ 12 32 b2
,
2 2b m 4b AG
4
0(2) 0 2 (1 )
R
F=
2
2(1)/b2
2/b2
where S0 is given by Eq. (39) and AG is the Glaisher constant. It is possible to show that the
high-temperature expansion of (6) exactly matches (60) (see Appendix B for some details).
5. Low-temperature expansion
We have mentioned in the introduction that the finite volume VEVs can be understood
as thermal averages (2). Hence, ea R admits the low-temperature (R ) expansion
in the form,

Gk (r).
log ea R Ga = 1 +
k=1
(61)
405
Here Gk represents k-particle contributions in the infinite-volume channel and

Gk (r) ekr .
Recently Leclair and Mussardo [6] proposed an integral representation which is sufficient
to generate Gk (r) systematically in terms of form-factors of the field ea at R = and
the solution of the TBA equation (19). Taking into account contributions of one- and twoparticle states to the trace (2), they obtained

log ea R Ga = 4[a]
d
f ( )
2
+ [2a]
d1 d2 (1 2 )
f (1 )f (2 ) + ,
2 2 cosh(1 2 )
(62)
where the notations

1
f ( ) =
,
1 + e+()
and
sin2 (a/q)
sin(b/q)
are used. The function is the kernel in the TBA equation (19). With (62) and the TBA
equation one can calculate the first two terms in the low-temperature expansion (61):
[a] =
4[a]
K0 (r),

d1 d2
G2 = 4[a]
(1 2 ) 2(1 2 ) er cosh 1 +r cosh 2
2 2
G1 =

+ [2a]
d1 d2 (1 2 ) r cosh 1 +r cosh 2
e
,
2 2 cosh(1 2 )
(63)
where Kn (r) is the MacDonald function. We now expand (63) as a power series in b2 ,

2
2
s ()
2
+ s () s(2) + O b K0 (r),
G1 = 4
(64)
b2
where
sin()
.
s() =
To expand G2 one needs to use Eq. (40),

2

s (2)
2
+
s
(2)
2s(4)
K0 (2r)
G2 =
b2

4s 2 ()r 2 K12 (r) K02 (r) s 2 (2)r 2 K2 (r)K0 (r) K12 (r) + O b2 .
(65)
406
It is quite straightforward to verify that the low-temperature expansion of (6) exactly

reproduces (64) and (65).
6. Conclusion
The proposed -representation (23) is the main result of this paper. Its rigorous
derivation has not yet been achieved. Although (23) are conjectures, the evidence presented
in this paper appears to make it reasonable to take them as the starting point for further
investigation.
One can expect that similar representations exist for non-minimal CFT, say, the Liouville
theory and SL(2, R)/U (1) non-compact -model. It may cast new light on many unsolved
problems of 2D quantum gravity. In this connection an intriguing similarity between the
integrals appeared in the -representation and matrix models of 2D quantum gravity [23
25] can be mentioned.
Note added
After finishing this paper it was drawn to my attention that Al.B. Zamolodchikov
independently introduced and studied the function (21) in Ref. [26]. I am grateful to him
for the communication of that paper, and sharing insights.
Acknowledgements
I am grateful to A.B. Zamolodchikov for interesting discussions. The research is
supported in part by the DOE grant #DE-FG05-90ER40559.
Appendix A
The QFT defined by (3) possesses infinitely many local IM I2n1 whose vacuum
eigenvalues have appeared in Eq. (17). They can be represented in the form,
I2n1 =
R

dx2
T2n (x2 + ix1, x2 ix1) + 2n2 (x2 + ix1 , x2 ix1 ) ,
2
where the local fields T2n and 2n2 satisfy the continuity equations,
z T2n (z, z ) = z 2n2 (z, z ).
Although a general expression for the densities T2n , 2n2 is not known, they are
determined up to normalization by the commutativity conditions,
[I2n1 , I2m1 ] = 0.
407
In Refs. [15,27] the following normalization was adopted,

T2n = 22n (z )2n + ,
(A.1)
where omitted terms contain higher derivatives of and exponential fields. Notice that the
condition (A.1) does not depend on the regularization scheme defining the composite field
(z )2n . With normalization (A.1) the constants Cn (17) were found in Refs. [15,28],

b 1 12n
2n1
0 (2n1)b
m0 2q
0 2bq
0 2qb
2q
Cn =
(A.2)
.
2 n! q
8q
The local IM I2n1 are certain deformations of the local IM of the Liouville CFT. Let
I2n1 (p) be an eigenvalue of the Liouville local IM corresponding to the state |p (48),
while I2n1 is the sinh-Gordon ground state eigenvalues of I2n1 . Eq. (49) suggests the
following relation for r 1:

I2n1 = I2n1 p(r) + O r 4qb , r 4q/b ,
(A.3)
where p(r) solves (50). 5 The power corrections in r (A.3) can be obtained by means of
conformal perturbation theory. Explicit forms of the functions I2n1 (p) (for n = 1, . . . , 8)
are given in Appendix B of Ref. [27]. Here we present only the first two of them,

2
1
2
p
,
I1 (p) =
R
24

3
2
p2 4b4 + 17b2 + 4
I3 (p) =
+
p4
.
R
4
960b2
In Table 1 we list numerical values of the local IM I2n1 for some 0.01 r 1 and
b2 = 0.81 which were obtained by means of numerical solution of the TBA equation (19)
with use of Eqs. (18) and (A.2). These data are compared against values of the Liouville
local IM I2n1 (p(r)). We consider the content of Table 1 to be an impressive evidence in
support of the relation (18).
Appendix B
Here we proceed with calculation of the products in (8) and give some technical hints
on the study of their high-temperature behavior. First, let us consider the product,

sinh(k m ).
eM1 = W
Nk>j N
Using the relation

1
r k = tanh(k ),
r
5 For n = 1 this relation was discussed in Ref. [18].
408
Table 1
Comparison of the LHS and RHS of Eq. (A.3) (b2 = 0.81)
r
I1
I1 (p(r))
1.0
0.8
0.6
0.4
0.2
0.1
0.01
0.0059897196942029
0.0123637695005731
0.0183955019094376
0.0242191017573388
0.0301582132435655
0.0335250692109914
0.0381898149656469
0.0059891933248581
0.0123636397906571
0.0183954816200409
0.0242191003738075
0.0301582132312211
0.0335250692108919
0.0381898149656469
I3
I3 (p(r))
0.01857760504756
0.01975951264163
0.02095100471765
0.02216988492643
0.02348269733500
0.02425825249985
0.01858088002375
0.01976027692024
0.02095111804696
0.02216989225147
0.02348269739676
0.02425825250033
1.0
0.8
0.6
0.4
0.2
0.1
r
I5
I5 (p(r))
I7
I7 (p(r))
1.0
0.8
0.6
0.4
0.2
0.02830178173
0.02936366562
0.03040878256
0.03145618592
0.03256438579
0.02830149724
0.02936359281
0.03040877074
0.03145618508
0.03256438578
0.0731035360
0.0750097817
0.0768644292
0.0787032289
0.0806285556
0.0731032652
0.0750097140
0.0768644184
0.0787032281
0.0806285556
I9
I9 (p(r))
I11
I11 (p(r))
0.29532264
0.30116225
0.30680503
0.31236318
0.31814537
0.29532199
0.30116209
0.30680500
0.31236317
0.31814537
1.73235
1.75983
1.78627
1.81220
1.83907
1.73235
1.75983
1.78627
1.81220
1.83907
1.0
0.8
0.6
0.4
0.2
which follows from the saddle-point equations (34), one obtains

r M1 =
1
2r
N

k,m=N
cosh(k m )
.
cosh(k ) cosh(m )
This sum can be rewritten in the form,

N
2

1
(2N + 1)2
r M1 =
+
tanh(k ) .
2r
2r
k=N
Notice that the sum in (B.1) converges for N ,
(B.1)

N

lim
k=N
409
r
tanh(k ) = r T + 2.
2
To derive this formula we used (36) and the saddle-point equations (34). Thus we obtain,

r
(2N + 1)2
log
M1 = MN
2
4

4N
dr
2
2 log
(B.2)
r(r T )2 + O N 1 .
T (r, )
r
8
r
The constant MN here does not depend on r. To find how it depends on , let us consider
M1 . A similar calculation leads to the equation,

1
M1 = (rr T 4) 2 T + 4 log(4N/r)
4
N
tanh(0 )
2
+
r
cosh(0 ) + cosh(m )
m=N
N
N
2
tanh(k ())
r
cosh(k ()) + cosh(m )
m=N k=1

tanh(k ())
.
cosh(k ()) + cosh(m )
(B.3)
It follows immediately from the last equation that

4N
M1 |r = 4 log
.
r
Therefore, we conclude that the constant MN in (B.2) does not depend on . Notice
that Eq. (B.3) is very convenient for studying the high-temperature limit r 0. It is
straightforward to show that for > 0,

2N(N+1)

4
2
N 2 eMN
eM1 r0
r

2 2 / k
0 2 k+1
20() 2 2
22
2 e
e
1

k+1
.
0(1 )
0 2 +
0 k+1
2 0 2 +
k=1
To finish the calculation of (8) one needs to evaluate the product,

eM2 =
N

Q2 [k ]e2k (+k).
(B.4)
k=N
The first two terms of the semi-classical expansion for Q are given by (41). With the
saddle-point equation (34) M2 in (B.4) can be written as,
M2 = M2 + M2 ,
(B.5)
410
where

M2
= b

N

r
cosh(k ) 2k (k + )
+1
(B.6)
k=N
and
M2 = 2
The sum
M2
r M2
N

d
1
log 1 er cosh( )
.
2
cosh(k )
(B.7)
k=N
is evaluated by means of Eq. (36). To calculate M2 we note that

N

= 2
k=N
1
cosh(k )
m
d
.
r
cosh(
)1
2 e
With the relations

N

k=N

N
r
1
1
r
4N
2
T + 4 log
+O
=
k =
cosh(k ) 2
4
r
N
k=N
and
d
er cosh( )

1
= 2 T =0 ,
1 4
one obtains,
M2

=

dr
r2 T =0 2 T +
2
16

dr
4N
2
.
r T
log
r
4 2 =0
(B.8)
We specify the integration constant here using the condition,

M2 r 0,
which follows from the definition (B.7). The function T (37) satisfies Laplaces equation,
1 2
T = 0.
4 2
This allows one to calculate the second integral in (B.8). Thus we find,
r 1 r (rr T ) +
M2

=

dr
4N
2
2
r T
T + T |=0 + rr T |=0 log
.
16 2 =0
r
Finally we note that the most efficient way to study the r 0 limit of M2 (B.4) is based
on the representation (42). It shows that for r 0 with r cosh( ) fixed,

r cosh( )
)

0 1 + r cosh(
r 2
2
Q[ ] r0
1 + O b2 .
2r cosh( /2) 4
Hence, examination of the product (B.4) at this limit creates no difficulties at all.
411
References
[1] A.Z. Patashinskii, V.N. Pokrovskii, Fluctuation Theory of Phase Transition, Oxford, Pergamon
Press, 1979.
[2] A.M. Tsvelik, Quantum Field Theory and Condensed Matter Physics, Cambridge Univ. Press,
1995.
[3] S. Lukyanov, A. Zamolodchikov, Exact expectation values of local fields in quantum sineGordon model, Nucl. Phys. B 493 (1997) 571587.
[4] V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Expectation values of local
fields in BulloughDodd model and integrable perturbed conformal field theories, Nucl. Phys.
B 516 (1998) 652674.
[5] F.A. Smirnov, Quasi-classical study of form-factors in finite volume, Paris Univ. VIVII,
LPTHE-98-10, February 1998, p. 21, hep-th/9802132.
[6] A. Leclair, G. Mussardo, Finite temperature correlation functions in integrable QFT, Nucl.
Phys. B 552 (1999) 624642.
[7] S. Lukyanov, A. Zamolodchikov, unpublished.
[8] H. Saleur, A comment on finite temperature correlations in integrable QFT, Nucl. Phys. B 567
(2000) 602610.
[9] E.K. Sklyanin, The quantum Toda chain, Lecture Notes in Phys. 226 (1985) 196233;
E.K. Sklyanin, The quantum Toda chain, J. Sov. Math. 31 (1985) 3417.
[10] E.K. Sklyanin, Separation of variables. New Trends, preprint UTMS 95-9, solv-int/9504001.
[11] H. Flaschka, D. McLaughlin, Canonically conjugate variables for the Kortewegde Vries
equation and the Toda lattice with periodic boundary conditions, Progress Theor. Phys. 55
(1976) 438456.
[12] L.D. Faddeev, L.A. Takhtajan, Hamiltonian Method in the Theory of Solitons, Springer,
New York, 1987.
[13] F.A. Smirnov, Structure of matrix elements in quantum Toda chain, J. Phys. A.: Math. Gen. 31
(1998) 89538971.
[14] D. Lebedev, S. Krachev, Integral representation for the eigenfunctions of quantum periodic
Toda chain, preprint ITEP-TH-54/99, hep-th/9911016.
[15] V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Integrable structure of conformal field
theory, I. Quantum KdV theory and thermodynamic Bethe ansatz, Commun. Math. Phys. 177
(1996) 381398;
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Integrable structure of conformal field
theory, II. Q-operator and DDV equation, Commun. Math. Phys. 190 (1997) 247278;
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Integrable structure of conformal field
theory, III. The YangBaxter relation, Commun. Math. Phys. 200 (1999) 297324.
[16] Al.B. Zamolodchikov, Thermodynamic Bethe ansatz in relativistic models: scaling 3-state Potts
and LeeYang models, Nucl. Phys. B 342 (1990) 695720.
[17] Al.B. Zamolodchikov, Resonance factorized scattering and roaming trajectories, ENS-LPS-335,
1991 p. 27.
[18] A.B. Zamolodchikov, Al.B. Zamolodchikov, Structure constants and conformal bootstrap in
Liouville field theory, Nucl. Phys. B 477 (1996) 577605.
[19] Al.B. Zamolodchikov, Mass scale in the sine-Gordon model and its reductions, Int. J. Mod.
Phys. A 10 (1995) 11251150.
[20] T. Curtright, C. Thorn, Conformally invariant quantization of the Liouville theory, Phys. Rev.
Lett. 48 (1982) 13091313;
E. Braaten, T. Curtright, C. Thorn, An exact operator solution of the quantum Liouville field
theory, Ann. Phys. 147 (1983) 365416.
[21] H. Dorn, H.-J. Otto, On correlation functions for noncritical strings with c 1, D 1, Phys.
Lett. B 291 (1992) 3943;
412
[22]
[23]
[24]
[25]
[26]
[27]
[28]
H. Dorn, H.-J. Otto, Two and three point functions in Liouville theory, Nucl. Phys. B 429 (1994)
375388.
N. Seiberg, Notes on quantum Liouville theory and quantum gravity, Prog. Theor. Phys.
Supp. 102 (1990) 319349.
E. Brezin, V. Kazakov, Exactly solvable field theories of closed strings, Phys. Lett. B 236 (1990)
144150.
M. Douglas, S. Shenker, Strings in less than one dimension, Nucl. Phys. B 335 (1990) 635654.
D. Gross, A. Migdal, Nonperturbative two-dimensional quantum gravity, Phys. Rev. Lett. 64
(1990) 127130.
Al.B. Zamolodchikov, On the thermodynamic Bethe ansatz equation in sinh-Gordon model, to
appear.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Quantum field theories in finite volume:
excited state energies, Nucl. Phys. B 489 [FS] (1997) 487531.
Al.B. Zamolodchikov, unpublished.

Analytic calculation of the 1-loop effective action

for the O(N + 1)-symmetric 2-dimensional
nonlinear -model
M. Bartels a , G. Mack a , G. Palma b
a II. Institut fr Theoretische Physik der Universitt Hamburg, D-22761 Hamburg,
Luruper Chaussee 149, Germany

b Departamento Fisica, Universidad de Santiago de Chile, Casilla 307, Correo 2, Santiago, Chile
Received 22 December 1999; accepted 16 July 2001
Abstract
Starting from the 2-dimensional nonlinear -model living on a lattice of lattice spacing a with
action S[] = 12 z , (z) S N we compute a manifestly covariant closed form expression
for the Wilson effective action Seff [] on a lattice of lattice spacing a in a 1-loop approximation
for a Gaussian choice of blockspin, where C(x) C(x)/|C(x)| fluctuates around (x). C is
averaging of (z) over a block x. The limiting case of a -function is also considered.
The result extends Polyakov which had furnished those contributions to the effective action which
are of order ln a/a.
The additional terms which remain finite as a

0 include corrections other
than coupling constant renormalization: a currentcurrent interaction and a contribution from an
augmented Jacobian which has a field dependence of a different kind than S has.
Particular attention is paid to Seff s domain of validity in field space. It turns out that Hasenfratz
and Niedermayers choice of a low value of the parameter which governs the width of the Gaussian
is optimal also in this respect. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ha; 11.10.Kk; 11.10.Hi
1. Introduction and summary of results

In this paper we present an analytic computation of the effective action Seff [] of
the O(N + 1)-symmetric 2-dimensional nonlinear -model in 1-loop approximation for
Gaussian block spins. A short summary was presented in Ref. [18]. Our results extend
Polyakov [22]. He only computed the UV-divergent contributions. Our desires are
Work supported by Deutsche Forschungsgemeinschaft.

E-mail address: gpalma@lauca.usach.cl (G. Palma).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 5 0 - 9
414
M. Bartels et al. / Nuclear Physics B 612 [FS] (2001) 413445
to perform the calculation analytically,

to obtain an explicit and manifestly rotation covariant result in closed form,
the domain of validity of the result for Seff [] should extend to rough configurations
as one expects to encounter in nonperturbative uses of Seff when the effective
coupling is strong.
One cannot expect a local expression for Seff [] which is valid for all . This is known
as the large field problem [13] and has been studied in the rigorous approach to the
renormalization group. 1 Here we take it for granted that the exceptional configurations
where locality gets lost have negligibly small probability. An example of an exceptional
configuration will be mentioned below.
This aside, the above demands can be met, at a prize, by choosing suitable values of
the parameter which governs the width of the Gaussian in the block spin procedure.
should be as small as locality permits, i.e., the width is fairly large [6]. This choice was
proposed by Hasenfratz and Niedermayer [15] because the resulting effective actions have
considerably better locality properties than for = . It turns out to be also the choice
which permits to avoid restrictions to sufficiently smooth .
On the other hand = would in principle allow the most stringent tests of the
accuracy of the result because one may in this case directly compare expectation values
in the original and the effective theory for observables which depend on the original
field through its block averages only [21]. Moreover, this is conceptually the clearest case
because every field in the original functional integral decomposes uniquely into a low
frequency field which is determined by the argument of Seff , and a high frequency
field which is integrated out.
We will derive analytic expressions for Seff for arbitrary , including = . It is
somewhat sobering to see how complicated the 1-loop result is in general. It can be brought
to manageable form if
either is small enough (of order a 2 ),
or is smooth enough so that the background field (z), which smoothly
interpolates , stays close to (x) for z x (see below).
It is an important lesson that a 1-loop effective action can be a complicated expression and
that its domain of validity in field space is a subtle issue. The computation of effective
actions for perturbative purposes can be much easier because it requires a smaller domain
of validity in field space. The considerations in Ref. [14] illustrate this.
Before comparing with earlier results, we need to fix the notation.
The model lives on a quadratic lattice of lattice spacing a with points typically denoted
z, w, . . . . The field (z) S N is a (N + 1)-dimensional unit vector. The action of the
model is 2
1 A device to overcome this difficulty was proposed by Benfatto et al. [5], and subsequently implemented in
the work of Kupiainen and Gawedzki [12] and of Balaban [4]. It involves proofs that large fields in the above
sense are very improbable. Fermi fields have no large field problem [9].

2 We use lattice conventions where is the finite difference derivative, and (. . .) = a 2
z (. . .).
z
S[] =

[ (z)]2 =
415

(z).
(1)
A block lattice of lattice spacing a = sa is superimposed (s positive integer). Its points

are typically denoted x, y, . . . . They are identified with squares of side length a in .
There are s 2 points z x.
We begin by defining a block average (x) which lives on the block lattice. It is a
function (x) = C(x) of the fundamental field. (x) is also a (N + 1)-unit vector;
therefore the operator C is necessarily nonlinear. We choose
C(x)
,
|C(x)|
where C is a linear operator which averages over blocks. We take

C(x) = avzx (z) a 2
(z).
(x) = C(x)
(2)
(3)
zx
The Wilson effective action with Gaussian block spin is defined by

C(x), (x) eS[] ,
eSeff [] = D
D =
d(z),
(4)
where d is the uniform measure on the sphere S N , and is a (N + 1)-dimensional

Gaussian which approximates the -function on the sphere. The explicit formula is in
Eq. (8).
Similarly as in the work of Hasenfratz and Niedermayer [15] and in the earlier rigorous
renormalization group approach of Balaban [3] we will use a background field = []
which is determined by but lives on the original lattice. For = it extremizes the
original action S() subject to the blockspin constraint C = . It follows that
Seff [] = Scl [] + loop corrections
(5)
and Scl [] = S( []) when = . The general formula is found in Eq. (12). Hasenfratz
and Niedermayer showed how to compute for given numerically. In Section 6 of this
paper we will describe an alternative analytic procedure.
Long ago Polyakov [22] computed those contributions to the effective action in
1-loop approximation which become UV-divergent when a
. These contributions are
independent of details of the blockspin choice ( including ). He showed the asymptotic
freedom of the model. In the above terminology, the 1-loop result in this approximation
differs from the tree approximation Scl only by a coupling constant renormalization. Here
we compute also the corrections which remain finite when a
0. These depend on the
details of the block spin procedure, and we find several kinds of 1-loop corrections other
than coupling constant renormalization, including a currentcurrent
interaction and terms
from a Jacobian which depend on the modulus |C (x)| = |a 2 zx (z)| of the block
average of the background field.
416
Recently, Hasenfratz and Niedermayer outlined a procedure for computing the full
effective action in 1-loop approximation term by term in a power series expansion in
-fields [16,17]. parametrizes the deviation of the field from a constant. Our result
sums the whole infinite series. This extends the domain of validity in field space because no
assumption is needed that is small. We use the exact lattice Leibniz rule throughout our
calculation. This is important because lattice artifacts contribute a finite coupling constant
renormalization.
Hasenfratz and Niedermayer used their effective actions in numerical simulations. While
other past applications of real space renormalization group technology had sometimes
given disappointing results, they found very good accuracy and a great saving in computer
time. We hope that our considerations shed additional light on this success and illustrate
some subtleties in the use of effective actions.
It would be feasible but tedious to extend our calculation to gauge theories. The
principles of analytic block spin procedures for gauge theories had been outlined long
ago in [2] and were extensively used in Balabans work [3,4]. To obtain explicit formulae,
one needs the high frequency (or fluctuation field) propagator. It can be found in Ref. [19].
1.1. Gaussian block spin and background field
Before proceeding, we give the Gaussian approximation to the -function which
entered in the definition of Seff .
Let (z) be the component of the unit vector (z) perpendicular to (x), for z x.
We will see later that the blockspin constraint C(x) = (x) is equivalent to C = 0.
Define the linear averaging operator C [] which depends parametrically on by

C [](x) = C(x) (x) (x)C(x) = C (x)
(6)
and let
= a 2.
(7)

2
C(x), (x) = J0 C(x) e 2 C [](x) .
(8)
The Jacobian J0 is determined by the requirement that

d C(x), = 1
(9)
Then
SN
for arbitrary . Explicitly (see Appendix A)

dN
2
2
e 2 | | ||
J0 ( )1 =
2
1 ||
N /2

1
= const | |2 +
.
(10)
(11)
The last formula is valid for large .

Note that the dimensionless quantity is large when
= O(a 2 ) and is large.
417
The tree approximation to the effective action is also called the classical action Scl . It
is the term of order h 1 while the 1-loop approximation is the term of order h 0 in an
expansion of the effective action in powers of h . Explicitly
1
Scl [] = S( ) +
C []
,
(12)
2
x
with the background field determined as a function of by the extremum condition

1
S( ) +
C []
= min.
(13)
2
x
In the limit
, C (x) behaves like 1/. Therefore the second term vanishes, is
the field which extremizes S( ) subject to the blockspin constraint C (x) = (x), and
Scl [] = S( ).
Finally we give the example of a large field configuration for N > 1. It is not large
in a naive sense. Divide the block lattice in black and white squares in a checkerboard
which takes the same value on all black squares
fashion and consider the configuration
and the opposite value on white squares. This configuration is left invariant under a
O(N ) subgroup of the rotation group. An associated background field configuration must
interpolate between opposite values and cannot be O(N )-invariant. Therefore there is a
As a result the action of
continuum of degenerate background field configurations for .
the fluctuation field [which is to be integrated out] has a zero mode, the fluctuation field
propagator does not have exponential decay [1], and one cannot expect a local effective
But note that these are energetically the most
action for block spins very close to .
unfavorable block spin configurations of all; they are near maxima of the effective action.
1.2. Summary of results
The exact 1-loop result for Seff will be derived in Section 5. It involves a high frequency
propagator Q . Q is an (N + 1) (N + 1) matrix propagator,

TC C Q
1
Q = lim + Q
(14)
with a residual field dependence which enters through the matrix

= 1 (z) T (z) + (x) T (z)[1 + cos (z)] T (x) ,
Q(z)
(15)
where T indicates the transpose, and

cos (z) = (z)(x)
(x z).
(16)
In the rest of this section we assume that either is small or is sufficiently smooth so
that (z) stays reasonably close to (x) for z x. In this case the field dependence is
1 and
manageable. In particular if then Q(z)
1),
Q = KG 1 + O(Q
(17)
where KG is the KupiainenGawedzki high frequency propagator for scalar fields. In the
analytic calculation of the background field we will also need the KupiainenGawedzki
418
Fig. 1. This figure shows cross sections through A-kernels for (8 8)-blocks with values of 100
(top), 1/8 (center) and 1/32 (bottom) in units a = 1. Note that the kernel oscillates for high values.
interpolation operator AKG (z, x) = KG C . The Fourier components of these quantities

are recorded in Appendix D.
Their general properties for the special choice (3) of the averaging operator C are well
known [11,14,23]. In particular, they have exponential falloff with decay length of order
one block lattice spacing a if a 2 4 2 . The precise locality properties depend on . The
coordinate space expressions can be evaluated by fast Fourier transformation. Software to
do the computation and visualize the results has been provided by Griessl and Wrthner
and can be downloaded from [24], together with some screen-shots. A sample is shown in
Fig. 1.
Under the stated assumption, Seff simplifies to

1
1
Seff =
ln J0 C (x) Tr ln Q +
eff (z)| (z)|2
2
2
x
z
+ renormalized 1-loop diagram + lattice artifacts

1
[ ](z).
+ (z)
2 cos (z)
(18)
with (z) = N KG (z, z) and (z) the part of (x) perpendicular to (z), z x. We
will now explain the various terms in this result.
Contrary to appearances, the last term is not UV-divergent as a
and is expected to
be negligible except possibly for very rough .
419
Only the eff -term contributes to order ln a/a,
and we recover Polyakovs result when

corrections that remain finite as a
0 are dropped.
The sum of the first two terms will be called the augmented Jacobian. The renormalized
1-loop diagram describes a currentcurrent interaction as indicated below.
The effective coupling constant is
eff (z) = (N 1)KG (z, z).
(19)
KG (z, z) is very nearly constant except near block boundaries. Therefore we expect that
the deviation of eff (z) from its block average can be neglected. The currentcurrent
interaction and the lattice artifact involve
j (z) = (z) T (z) (z) T (z + )
+ T T (z + ),

3
| |2 (z) + O(a) (20)
lattice artifacts = tr j (z) KG (z, w)|w=z+ =
8
z
as a
0. The lattice artifacts would vanish for a
0 if T were zero as is true
in the continuum. On the lattice it is O(a), but nevertheless there remains a contribution
when a
0 because of the singularity of KG at coinciding points. It amounts to a finite
subtraction from the bare coupling constant.
The renormalized 1-loop diagram in (18) describes a currentcurrent interaction and is
(2)
, except that KG is substituted for Q .
given by the following expression Seff
(2)
Seff
1
=
2

tr Q (z, w) jT (w)Q (w + ,
z + )j
(z)
z w
+ Q (z, w + )j (w) Q (w, z + )j

(z)

T
(z w)j (z)Q (w + , z + )j
(z) .
(21)
The (z w)-term subtracts the part which diverges in the limit a

0. The last term
in the definition (20) of j is a lattice artifact and can be dropped inside Eq. (21) because
its contribution is actually of higher order in .
It remains to discuss the 12 Tr ln Q term. Under the stated assumptions it can be
evaluated. For small the result is particularly simple,

T

1
1

(z)Q(w)
Tr ln Q = Tr ln KG 1 +
avzx avwx tr Q
1 KG (w, z)
2
2
2
x
(22)
as above. The first term is a constant, av stands for the block average, and
with Q
gives
superscript T indicates the transpose. Explicitly, the above formula for Q

2

T

(z)Q(w)
1 = (z) (w) 1 + (z) (w) 1 cos (z) cos (w)
tr Q

+ cos2 (z) 1 + cos2 (w) 1 .
(23)
420
1.3. Result for large

When is large or , the above formula for 12 Tr ln Q cannot be used because it
contains an overall factor .
But suppose the blockspin is reasonably smooth, so that (z) (x) may be regarded
as a small quantity, of order &, and x (x) is also small, o(& 0 ).
1 is of order & and 1 Tr ln Q can be expanded in powers of &. To order & 2
Then Q
2
it comes out as
1
Tr ln Q
2

=
AKG (z, x)C(x, z) cos2 (z) + cos (z) 2
x

z,w

+
T

(w) 1 Q(z)
1 KG (z, w)AKG (w, x)C(x, z)
tr Q

1 AKG (z, x)C(x, w)AKG(w, y)C(y, z)
tr Q(w)
1 Q(z)

(24)
x,y
with C (w, x) = C(x, w) = a 2 if w x and 0 otherwise. The Jacobian proper and

12 Tr ln Q are similar but they have a different N -dependence.
In these formulas, AKG and KG are the KupiainenGawedzki interpolation operator
and high frequency propagator for scalar fields We give their definition for finite ; the
limit
can be taken at the end of the calculation. Let vCb = ()1 . Then
1
AKG = vCb C u1 , u = C vCb C + ,

1
KG = + C C
= (1 AKG C)vCb = vCb 1 C AKG .
(25)
They have the following properties and limiting behavior as

1
A
0,
KG
1
CAKG = 1 u1
1.
CKG =
(26)
(27)
2. Linearization of the constraint

A perturbative calculation of the functional integral (4) for the effective action is not
straightforward because the argument of the approximate -function is a nonlinear function
of the field.
To solve this problem, we find a parameterization of an arbitrary field on in terms
of the background field = [] and a fluctuation field such that we end up with an
421
approximate -function whose argument is linear in .

(z) = [, ](z).
(28)
The background field was introduced in the introduction. It is a smooth field. It

represents the low frequency part of , while adds the high frequency contributions.
takes its values in a linear space. It has N components, and we choose it so that
(z) (x)
for z x.
(29)
Later, a further linear transformation to variables (z) (z) is performed.

When the definition of is inserted, the defining equation (4) of the dependent
effective action becomes

2
Seff []
J0 C(x) e 2 C [](x) eS[] ,
e
(30)
= D
x
where J0 is a -dependent
Jacobian (11).
After introducing the background field (z) which extremizes the exponent as in
Eq. (13), one parameterizes the field in terms of a fluctuation field as follows. 3
Decompose (z) into components || (z) and (z) parallel and perpendicular to (x)
for z x.
We parameterize
(z) = (z) + (z),

1/2
|| (z) = 1 (z)2
(x).
(31)
It follows that

C (x)
2 =
C (x)
2 +
2 C C (z) + C C(z) .
a2
x
This may be inserted into Eq. (30). By the saddle point condition which determines ,
the term in the exponent which is linear in must vanish. Inserting the previous definition
of the classical action we end up with

d (z) (C )J (, )eS([, ])
eSeff [] =
(32)
z
which now involves an approximate -function on a linear space,

exp |C (x)|2 ,
(C ) = N
(33)
x
3 Balaban [3] has shown how to find a suitable parameterization in the case of lattice gauge fields. His method
is not applicable for the nonlinear -model for general N , because it makes essential use of the fact that the field
takes values in a group, and right and left multiplication of group elements commute, (gL g)gR = gL (ggR ). But
the suitable parameterization can be written down explicitly as in the main text.
422
where N is an unimportant normalization factor, and a Jacobian J ,

2 1/2
J (, ) = J 0 (C)
1 +
,
J 0 (C) =
(34)
zx

N
(x)C(x) .
J0 C(x) =
(35)
In principle, is determined by and , but in a 1-loop calculation, J0 (C(x)) is

approximated by J0 (C (x)). The last factor in J will be canceled when we transform
to the -variables.
3. The 1-loop approximation

The 1-loop approximation yields the effective action to order 0 . It is obtained by
expanding the action to second order and the Jacobian to zeroth order in the fluctuation
field. This approximates expression (32) by a Gaussian integral. The resulting Tr log
formula is not particularly useful, though.
It is possible to obtain a first simplification by exploiting the fact that the background
field is smooth. This is always true, whether the block spin is smooth or not, provided
the blocks are chosen large enough. A basic reason for this is that there are no domain
walls in a 2-dimensional ferromagnet with continuous symmetry, because the free energy
of such domain walls would decrease by making them wider. This is an old argument by
Fisher [10] which was made mathematically precise by Dobrushin and Shlosmans in their
version of the proof of the absence of spontaneous breaking of continuous symmetries in
2 dimensions [8].
Because of the smoothness of one can neglect terms of higher order than second in
. Note however that this smoothness argument cannot be used to argue that cos (z) =
T (z)(z) must always be close to 1. Only for sufficiently smooth block spin field (x)
will it be true that the component (z) of (z) which is perpendicular to (x) is small.
The action S involves derivatives of which contribute derivatives in . Because of the
constraint (z)(x) = 0 for z x, will have jumps at block boundaries which contribute
to the derivatives. In order to avoid this complication, it is convenient to make a further
linear transformation from (z) (x) to variables (z) (z),
(z) = Q1 (z) (z),
(36)
(z) (z) = 0.
(37)
Q(z) depends on (z) and (x). It is a linear transformation between different tangent
spaces of the sphere
Q(z) : T (z) S N
T(x) S N
for z x.
(38)
The tangent spaces are N -dimensional. We introduce the abbreviation

(z) = 1 (z) T (z).
(39)
423
In covariant form, Q is as follows

Q = + T cos T ,
(40)
1
T .
Q1 = 1
(41)
cos
An expression in a particular basis will be given in Appendix A.1. It shows that the modulus
of the determinant of the resulting N N matrix Q(z) is

2 1/2
|detQ(z)| = cos (z) = 1
(42)
.
: RN +1
RN +1 .
Later on we shall introduce an extension of Q to a map Q
The expansion of the field (z) in powers of comes out as
(z) = (z) + (z)
1
(z)2 (x) + .
2 cos (z)
(43)
The action S() can now be expanded up to second orders in ,

1
S() = S( ) + S ( ) + linear in + .
(44)
2
Now we are ready to consider the effective Boltzmann factor. In one loop approximation,
i.e., to order 0 the Jacobian factor gets expanded to 0th order in the fluctuation field.
Furthermore
d (z) = |det Q(z)| d(z)
and det Q is as given above in Eq. (42). Therefore the factor multiplying J 0 in the Jacobian
in (35) cancels out and we get to 1-loop order

d(z) (CQ )e z 2 ( S ( ) )
eSeff [] = eS( ) J 0 (C )
(45)
z
with Gaussian as in Eq. (33). There is no linear term in in the exponent because of the
saddle point condition on .
The integration of the variable (z) is over the N -dimensional tangent space T (z) S N ,
i.e., subject to the constraint
(z)(z) = 0.
(46)
We have to evaluate a Gaussian integral. As a result, one obtains the effective action as
a sum of the classical action (tree perfect action) S( ), the Jacobian term ln J0 and
a Tr ln -term. The propagator is the covariance of the above mentioned Gaussian
measure.
This formula is not particularly illuminating because the full propagator has a
complicated dependence on the field . It comes from three sources: the constraint (46)
on , the -dependence of Q, and finally the -dependence of S .
A simplification is possible because the smoothness of the background field (on length
scale a = lattice spacing of the fine lattice) can be exploited. In the approximation which
exploits the smoothness of the background field it is not necessary to consider terms of
424
higher than second order in . S contains field dependent terms of first and second
order in . They can be treated as perturbations which are treated by second and first
order perturbation theory, respectively. This extracts the field dependence of S from the
propagator.
The field dependence in Q reflects the detailed choice of the block spin. Its contributions
are not of order ln a/a
and are therefore not included in Polyakovs result. The derivation
of explicit formulae depends on the assumption that the block spin field on the block
lattice is smooth enough, or, more precisely, on sufficient smoothness of on the length
scale of the lattice spacing a of the block lattice. There exists an extension of Q to an
When the assumption holds, Q
is close to 1, and one
(N + 1) (N + 1) matrix Q.
1. We will later
can derive a power series expansion of the propagator in powers of Q
compute this expansion.
There remains the constraint = 0 on the integration variables (z). There are two
ways to handle this
(1) Polyakovs method One expands (z) in a basis e1 (z), . . . , eN for the tangent
space T (z) S N . In differential geometry, such a basis is called a moving frame.
S becomes a N N matrix in this basis.
Polyakovs method has the advantage that the origin of the characteristic factor
N 1 in the formula for the running coupling constant emerges in a very transparent
fashion from the form of S . Therefore we show the details in the next section. The
result agrees with Polyakovs, to order ln a/a.
The disadvantage of Polyakovs method is that the expansion of the propagator in

1 would be very thorny.
powers of Q
(2) (N + 1)-dimensional integration Here one inserts 1 in the form of a Gaussian
integral over an additional integration variable 0 (z) R. This is combined with the
N -dimensional integration over (z) to an (N + 1)-dimensional integration over
(z) = (z) + 0 (z) RN +1 .
1 is straightforward, but
In this formulation the power series expansion in Q
Polyakovs result must be extracted by evaluating the singular part of a 1-loop graph.
It is convenient to write the action in a gauge covariant form by introducing an arbitrary
z-dependent basis. This yields results which can be used in both methods. The basis
consists of an orthonormal set of vectors e (z), = 0, . . . , N , for every site z which
span RN +1 , so that
e e (z) = .
(47)
The field can be expanded in the basis

(z) =
(z)e (z)
(48)
=0
and similarly for and . We assemble the expansion coefficients in (N + 1)-dimensional

column vectors , , and .
425
One introduces matrices A (z) by

A (z) = e (z + )
e (z).
(49)
On the lattice, the Leibniz rule takes the form

g(z).
f (z)g(z) = f (z) g(z) + f (z + )
(50)
Using this one finds the following substitute for antisymmetry in indices , ,
A (z) = A (z + ).
(51)
The action takes the covariant form

| |2 = |( + A )|2 .
(52)
In a constant basis, one has A = 0.

The expansion (43) carries over to the column vectors. Using it one computes with the
help of the lattice Leibniz rule
||2 (z) = | |2 (z) + | |2 (z) +
1
(x) (z) 2 (z)
cos (z)
(53)
with cos (z) = (z)(x) as usual, z x. A total divergence has been omitted which
arises from partial integration of a 2 -term.
Because of the smoothness of , is of order | |2 . To order | |2 we find

( + A )
2 (z) +
(x) (z) 2 (z) .
S =
(54)
cos (z)
z
4. Polyakovs method
In Polyakovs method one uses a basis with
e0 (z) = (z).
(55)
The basis vectors e1 (z), . . . , eN (z) span the tangent space T (z) S N and the -field has no
0-component.
There is a remaining arbitrariness in the choice of basis. The O(N )-group of those local
rotations which leave (z) invariant form a symmetry group of gauge transformations.
The N N matrices
a (z) = (Aij (z), i, j = 1, . . . , N )
transform like gauge fields under these gauge transformations, while

i
A0i (z) = (z)
transform like N -vector fields.
(56)
(57)
426
We compute the field strength tensor 4 for the vector potential a ,

Fij (z) = aij (z) aij (z)
+ aik (z + )akj (z) aik (z + )akj (z).
(58)
Using the completeness relation for the basis in the form

N

ei (z)ei (z)T = 1 T
i=1
one computes the component of the field strength tensor as

Fij = ei (z + + )
(z + )ej (z) (z) ( ).
(59)
We see that the field strength tensor is of order ( )2 . It follows that the vector potential
a in Lorentz gauge,
a = 0,
is also of order ( )2 . The [a , a ] term in expression (58) is negligible and the vector
potential to leading order could be recovered as

a (z) = vCb (z w)F (w).
w
Although the Coulomb potential vCb in 2 dimensions does not exist, its derivative is well
defined.
Separating the terms which involve a and A0i and using the antisymmetry Eq. (51) of
A and (z + )
= (z) we obtain
|( + A ) |2 = |( + a ) |2 + | |2 .
(60)
The last term involves the components [ ]i of with respect to the moving frame,
not i .
In conclusion

S = |( + a ) |2 (z) + SI
(61)
z
with
SI =

[ ]2 (z) + 2 (z)[ ](z) +

1 2
(z) (z) .
cos
(62)
The -term was split into two terms in order to single out the last term in SI . We
will see later that this last term is very small for smooth enough block spin fields. This
4 This is the field strength tensor which one gets by use of non-commutative differential calculus. It was shown
by Dimakis, Mller-Hoissen and Striker [7] that the conventional lattice gauge theory formalism is equivalent to
a non-commutative differential geometry. In this formulation, the lattice Leibniz rule Eq. (50) above takes the
standard form d(f g) = (df )g + f dg, and all the familiar formula of continuum gauge field theory remain valid
on the lattice.
427
is a consequence of the extremizing property of . The term is multiplied with an

expression of order ( )3 , and turns out not to contribute at all to order ln a/a.
The effective Boltzmann factor becomes

2

d(z) (CQ )J 0 (C )e z 2 ( +a ) + SI ( ) . (63)
eSeff [] = eS( )
z
The -function is a Gaussian, and we have to evaluate a Gaussian integral.

Let us write [z] for the block x which contains z. Let us remember that Q(z) is
a map (38) from T (z) S N to T(x) S N , and C defines a map of functions on the fine lattice
with values in T([z]) S N to functions on the block lattice with values in T(x) S N .
Therefore the operator QT C CQ maps functions with values in T (z) S N into functions of
the same kind. The Polyakov basis elements ei (z), i = 1, . . . , N , are a basis for T (z) S N .
We denote by P (z, w) = (Pij (z, w), ij = 1, . . . , N ) the matrix of the kernel of QT C CQ
with respect to this Polyakov basis, viz.

Pij (z, w) j (w) = ei (z) QT C CQ (z).
(64)
w
P (z, w) is only nonzero when z and w belong to the same block x. The -function
becomes the following Gaussian
(CQ ) = lim N e

z w
i (z)P
ij (z,w)
j (w)
(65)
Define the high frequency propagator (= propagator of the -field) in the Polyakov basis

1
e = 1 [ + a ]2 + P
(66)
.
Now we can evaluate expression (63) for Seff with volume element
d(z) = d 1 (z) d N (z).
The result is
Seff [] = ln J 0 (C ) + S( ) +
1
1
tr SI ( )e Tr ln e + const
2
2
(67)
in the limit
. Note that P depends on because Q depends on . 5 Therefore the
propagator e also has a residual dependence. It is small when the block spin field is
of Q is in this case close to 1. Unfortunately it would be
smooth, because the extension Q
1 in this formalism, because the formula for P
difficult to find the first order term in Q
contains the moving frame, and because there could be a term which is first order both in
1.
a and in Q
4.1. Recovery of Polyakovs result
Polyakov determined the contributions to the effective action which are of order ln a/a.
They do not depend on the detailed form of the blockspin which fixes the infrared cutoff
5 In addition there is an implicit -dependence through the moving frame and through a .
428
in the auxiliary theory with fields . The term P in the high frequency propagator has the
effect of an infrared cutoff. This has been discussed in detail in the work of Kupiainen and
Gawedzki [11]. To get the result modulo details of the choice of infrared cutoff, we may
therefore replace P by a mass term M 2 , where M = o(a 1).
The propagator also has a dependence on the O(N )-gauge field a . We show that this
can be neglected, by exploiting the smoothness of the background field . We need only
consider terms up to order O([ ]2 ). The result is gauge invariant. a in Lorentz gauge
is O([ ]2 ) as we saw. A perturbation expansion in a shows that Tr ln e = O(a2 ).
Therefore the a dependence of this term can be neglected. SI is already O([ ]2 ),
therefore the a dependence in the propagator multiplying it can also be neglected. The
high frequency propagator matrix can therefore be replaced by
e (z, w)ij 1 ij vM (z w),
where vM is the Yukawa potential in 2 dimensions with mass M of order
1

vM = + M 2 .
(68)
a 1 ,
viz.
The Tr ln-term has become a constant. The Jacobian is not ultraviolet divergent and is
therefore a feature of the details of the infrared cutoff. Inserting SI we get the result in the
desired approximation

2
1
Seff [] =
(69)
(N 1)vM (0) (z) + SisZero ,
2
z
with
1
SisZero = N vM (0)
2
[cos (z)]1 (z).
(70)
Here as everywhere
cos (z) = (x) (z),

(z) = (x) (z) (x) (z)
(71)
(72)
for z x. is the component of the blockspin which is perpendicular to the background

field.
Except for the term SisZero this is Polyakovs result. We show in Section 6.2 that SisZero
is actually zero as a consequence of the extremality condition on the background field .
Thus, Polyakovs result has been recovered.
4.2. A note on high frequency propagators
We record here a formula for the full high frequency propagator which would figure
in the not very illuminating formula
1
Tr ln
(73)
2
as mentioned earlier. It is obtained by inspection of the exponent in the integral
representation (63), the alternative formula (74) is obtained from the alternative treatment
Seff [] = ln J 0 (C ) + S( )
429
using the constant basis in Section 5 in the same way.

(z, w) =

eiT (z) [ + a ][ + a ]
i=1

1
+ T + + (cos )1 1 + P
ei (w)

T
T
T
= (z) + 2 + + j + jT

TC C Q
1 (w),
+ (cos )1 + Q
(74)
with the understanding that eiT are the basis vectors in the dual space, and
(j )(z) = j (z)(z + ),
(75)
i.e., j contains a shift operator. (Remember the footnote on non-commutative differential

calculus.) Here as everywhere, (z) = 1 T projects on T (z) S N . We see from the
second formulae that
= Q[ ] + O( ).
(76)
There is a correction term of first order in because j is of first order in , see

Eq. (20). This explains why Tr ln produces among others a 1-loop graph (21) which
involves T (z) at two different sites.
5. (N + 1)-dimensional integration
We present now the alternative method for evaluating the Gaussian integral (63) for the
effective action. This will prepare the ground for the expansion of the result in powers of
1.
Q
We insert extra integration variables 0 (z) R by insertion of

0 2
0
0
1 = N
(77)
d 0 e 2 z | | + C C .
N is a constant which is not field dependent, and C is the block average similarly as before.
We will combine the integration variables 0 and to
(z) = (z) + 0 (z) (z)
(78)
so that
0 = T ,
= ,
(z) = 1 (z) T (z).
(79)
Here and in the everywhere we write superscripts T for the transpose. The transpose T of
a column vector is a row vector.
(z) is now considered as an element of RN +1 . It satisfies the constraint T = 0.
The symbol will stand for the finite difference derivative of this RN +1 -valued field.
430
In other words, we expand now in a constant basis e0 , . . . , eN , viz. the natural basis for
RN +1 . In this way we can use the result Eq. (54) with A = 0, and we can write in place
of , etc.
Adding the | 0 |2 term to the action, we obtain an extended action

2
Sext ( |) =
+ S()
2

0
2
1
2
2
= S( ) +
| | + +
(x) (z)
2
cos (z)
z

+ .
S( ) +
(80)
| |2 (z) + T Sext,I
2
z
In Appendix B the sum of the first terms is computed. As a result

T Sext,I
= T (z + )
2 T + (z + )
T T (z + )
(z + )

T T
(x) (z) + T j (z + )
+ transpose .
cos (z)
Repeated indices are summed over. The exponent of the Gaussian , Eq. (33) combines
with 0 C C 0 /2 according to
= 0 C C 0 + T QT C CQ
T C C Q
TQ
(81)
as follows.
with Q
The definition (40) of Q extends to a map RN +1
RN +1 which has the property
that it annihilates (z) and maps T (z) S N to T(x) S N RN +1 . We add to this the
operator (x) T (z) which annihilates T (z) S N and maps the ray through (z) into the
ray through (x). This gives

= 1 (z) T (z) + (x) T (z)[1 + cos (z)] T (x) .
Q(z)
(82)
Using the indicated ranges of the various maps and Eqs. (79), it is readily verified that
formula (81) holds true.
Now we are ready to evaluate the Gaussian integral which defines the effective action

T T
2
T
d N +1 (z)e 2 Q C C Q+| | + Sext,I .
eSeff [] = eS( ) J 0 (C )
z
(83)
We define the new high frequency propagator

TC C Q
1 .
Q = + Q
(84)
When we want to make this dependence explicit, we write

Q depends on through Q.
Q[ ] .
Q (z, w) is a map RN +1
RN +1 , i.e., an (N + 1) (N + 1) matrix. Its only
In Section 5.2 we will show how to expand in a power series
-dependence is in Q.
1. In zeroth order, agrees with the KupiainenGawedzki high frequency
in Q
431
propagator [11],
1).
(Q ) = KG + O(Q
(85)
Using this propagator, the effective action can be computed by perturbation theory.
Because of the smoothness of , we are only interested in terms up to order | |2 . But
j is of first order in . Therefore the j -term must be treated to second order, while all
the other terms need only included to first order in the perturbation expansion. As a result

1
1 T
Tr ln Q[ ] +
Sext,I ( )
2
2
2 c
1 T
Sext,I ( )
(86)
,
8
where is the expectation value in a free field theory with propagator Q of , and the
truncated expectation value f 2 c = f 2 f 2 .
The expectation values can be evaluated. The correction term of second order in j yields
(after a change of summation variables z, w, , )

c
1
1
tr Q (z, w) jT (w)Q (w, z)j (z)

[ ]2 =
8
2
z w

+ Q (z, w)j (w) Q (w, z)j (z) + .
(87)
Seff [] = S( ) ln J0 ( )
The term is logarithmically divergent as a

0.
The first order correction is

1
1
[. . .] =
tr 2 T + (z + )
T T (z + )
2
2
z
z + )
Q (z + ,
+ tr j (z) Q (z, w)
w=z+

+ tr Q (z, z)
1
T ([z]) ;
(88)
2 cos
unwritten arguments are z. The Tr ln Q -term is needed in Seff because of the
1.
dependence of Q ; it becomes constant in zeroth order in Q
From this we obtain the final result result, Eq. (89), by adding to the second order
correction the (z w)-term in expression (21), and subtracting it from the first order
term. We show in Appendix C that this is the appropriate subtraction which renders the
2-vertex diagram convergent in the limit a
0. When j is inserted, the subtraction from
the first order term leads to a partial cancellation. The last term in the definition (20) of
j can be dropped in Eq. (87) and in the subtraction because its contributions will be of
higher order in by Eq. (94).
Seff = Scl

1
ln J0 C (x) Tr ln Q
2
432

([z])T () (z)
T
1
2
(z)eff (z) (z) + eff (z)
cos (z)
(2)
+ Seff + tr j (z) Q (z, w)
1
+
2
(89)
w=z+
(2)
with current j and renormalized 1-loop diagram Seff
as in Eqs. (20), (21) of the

introduction. If is smooth enough, Q 1 is small and we approximate Q by the field
(2)
independent KG 1 inside Seff
. The same approximation is justified when is small, as we
shall see in Section 7.
5.1. Evaluation of a lattice correction term

1, we need to also evaluate
In order to get the simplified result in zeroth order in Q
the lattice artifacts which come from the following term in Eq. (89)

tr j (z) KG (z, w)|w=z+ = 3
T (z) + O a 2 KG (z, w)|w=z+ .
z
This is a lattice artifact; in the continuum limit

Nevertheless it cannot be neglected because
(90)
T
= 0. On the lattice it is of order a.
w)|w=z+
KG (z, w)|w=z+ = KG (z + ,
a
1
+ O(1).
= KG (z + ,
w)w=z+ =
4
4a
(91)
(92)
This holds true because the singular part of KG is translation invariant and because

KG(z, w)|w=z is independent of by lattice symmetry, while f = af .
On the other hand (unwritten arguments are z)

+ T
0 = T (z) = T (z + )
(93)
hence
a
T = | |2 no sum.
2
Therefore the lattice artifacts are as stated in the introduction,
3
tr j (z) KG (z, w)|w=z+ = | |2 (z) + O(a).
8
(94)
(95)
5.2. Field dependence of high frequency propagator for arbitrary

Here we consider the expansion of the high frequency propagator (84) in powers of

Q 1. This expansion is needed when is not small, and results in an expression for
12 Tr ln Q which does not have a factor in front. But it assumes that is smooth
is close to 1.
enough so that Q
433
Consider a propagator of the following form which depends on a real parameter

1

= + C C
(96)
,
where C is a (matrix valued) -dependent block averaging operator. A limit

should be taken in the end, if desired. In our application
1).
C = 1C + C(Q
(97)
We will use a formula which gives the derivative of with respect to .

Let v = 1 . It is known from the work of Kupiainen and Gawedzki, that admits
the following representation
= (1 A C )v,
(98)
where A is an interpolation operator which maps functions on the coarse lattice into
smooth functions on the fine lattice, and which obeys
C A = 1
1 1
u .

(99)
For finite ,
A = vC u1
,
u = C vC +
1
.
(100)

1 1
We denote differentiation with respect to by a prime. Since (u1
) = u u u , one
obtains by straightforward differentiation

= C A + A C ,
(101)

A = C u1
A C A .
(102)
1) independent of , and A=0 is the KupiainenGawedzki

In our application, C = C(Q
interpolation operator AKG multiplied by the (N + 1) (N + 1) unit matrix 1. Therefore
1 reads
the expansion of Tr ln Q to second order in Q

1
Tr ln Q = Tr ln KG + Tr 1
=0 + Tr 1
=0 +
2
(103)

1) + h.c.
Tr 1 = Tr A C(Q
(104)
with
and

Tr 1 = Tr A C + h.c.

= Tr C u1
C A C A C + h.c. =0 .
The first term in Eq. (103) is a field independent constant.
(105)
(106)
434
The kernels A=0 and C are proportional to the unit matrix, therefore the first order term
involves
1) = cos2 + cos 2.
tr(Q
(107)
As a result

1
Tr 1 =
AKG (z, x)C(x, z) cos2 (z) + cos (z) 2 + const.
2
x
(108)
It can now be inserted into the result for the effective action.
1). Using also the fact
The second order term is evaluated by noting that C = C(Q
that C u = AKG we obtain

1
Tr 1 =0
4

T

1
(w) 1 Q(z)
1 KG (z, w)AKG (w, x)C(x, z)
=
tr Q
2
z,w x

1 AKG (z, x)C(x, w)AKG(w, y)C(y, z) . (109)
+ tr Q(w)
1 Q(z)
x,y
These results are equally valid for finite and for = .

Summing the two terms we obtain the result Eq. (24) for the augmentation of the
Jacobian.
6. The background field

Given the block spin field on the coarse lattice, we seek the field = on the fine
lattice which satisfies the extremality condition (13), viz.
1
S( ) +
C []
= min.
(110)
2
x
In the case = , this means that extremizes S subject to the block spin constraint
C = or equivalently, C = 0.
The extremality condition leads to a nonlinear equation for . It is nonlinear because
(z) must have length 1.
To first order in , = + . Inserting Eq. (32) and making the transition to -variables
= Q1 the saddle point condition becomes
S ( ) = QT C C ,
(111)
Our strategy is to start with an approximation (0) which has the expected smoothness
properties of except for discontinuities of the normal derivatives at block boundaries
which are small if is reasonably smooth, which reduces to the exact extremum when
is constant, an which satisfies the block spin condition exactly in case is large.
435
Starting from (0) we can derive improved approximations (k) , k = 1, 2, . . . , by

iteration. We will see that the smoothness of can again be exploited to argue that a
single iteration with result (1) is enough is is reasonably smooth. The formula for (1)
will involve the high frequency propagator Q[ (0) ] which was encountered before. This
propagator contains a dependence on (0) through Q.
The formula for (0) will be derived in Section 6.1.
We proceed to the iteration step. Given any approximate extremum (0) , we parametrize
an arbitrary field with the desired block spin with -variables (0) similarly as
before in Eq. (43), except that (0) is substituted for . To first order in ,
= (0) + .
(112)
Power series expansion to first order around

S ( ) = S (0) + S (0) + .
0)
gives
Using Eq. (112) one computes S ( (0) (z)) = (0) (z). But this is only valid as a linear
form on the tangent space T(0)S N , i.e., when contracted wit arbitrary (z) (0) (z). In
order to remember this fact it is better to write the formula as

S (0) (z) = (0)(z) (0) (z),
(113)
with the projector (0) on the tangent space. Inserting everything into Eq. (111) we get a
linear equation for ,

(0)
+ QT C CQ = S (0) QT C C (0) .
S
(114)
We recall the fact, recorded in Section 3, Eq. (76) that the full high frequency propagator in
a background field agrees with Q[ ] to zeroth order in . To the desired accuracy
one can therefore replace = S 1 by (0) Q . In this way we obtain the approximate
solution
=
(0) + (0) Q[ (0) ] (0) [ (0) QT QC (0) ]

(modulus)

+ O 2 ,
(115)
where Q is the high frequency propagator (84) with finite .
Again there are two choices of to be distinguished. If is not small, the assumption
of sufficient smoothness is in force and we can use the results of Section 5.2 for Q . For
small we can use the perturbation formula for Q , Eq. (136).
It is derived in the same way as Eq. (22).
6.1. Smooth interpolation of blockspin fields
Here we seek a field (0)(z) on the continuum which has a given block spin
C (x)
avzx (z)
,
(116)
(modulus)
(modulus)
which is continuous and smooth except for (small) discontinuities of the normal derivative
on block boundaries, and which is close to (x) for z x if is smooth. The average av
is over the lattice points inside the square.
(x) =
436
The lattice field (0) is obtained by restriction to points z in the lattice.

We assume that the block spin is reasonably smooth so that (x)(y) > 0 when
x, y are nearest or next nearest neighbors. This restriction removes some sign arbitrariness
which could otherwise lead to discontinuities.
The continuum is divided into squares x of side length a;
the lattice points inside form
a block.
We proceed in several steps.
(1) We determine the field at the corners zc of the squares,

x (x)
(0) (zc ) =
(117)
,
(modulus)
where the sum goes over the four squares x with corner zc .
(2) We consider the interpolations of the values of the function at the corners to functions
on the sides between two adjacent corners. In this way, (0) is defined on the whole
boundary of every square x, and is close to (x) when is smooth.
Consider the side with endpoints z0 and z1 which separates squares x and y. Let the
4 squares with joint corner z0 be x, y, x0 , y0 and the squares with joint corner z1 be
x, y, x1, y1 . If the side is parameterized by t = 0, . . . , 1, with z0 = z(0), z1 = z(1), the
interpolation is as follows.
(0) (t) =
(x) + (y) + (1 t)[(x0 ) + (y0 )] + t[(x1 ) + (y1 )]

.
(modulus)
(118)
(3) We consider one square x at a time and construct a preliminary interpolation

(0)

which interpolates (0) from the boundary to the inside, such that it is smooth
inside and takes the prescribed values on the boundary. The resulting function on
the whole continuum is smooth except for discontinuities of the normal derivatives
across the boundaries of the squares. The interpolation is as follows. Introduce the
notation

x (z) = (0)(z) (x) (0) (z)(x) ,
etc. Given x (z), the field (0) (z) for z x can be recovered by Eq. (122). 6
Let the points z in the closed square x be parameterized by (t1 , t2 ), 0 ti 1. The four
sides of the square have t1 = 0 or t1 = 1 or t2 = 0 or t2 = 1, respectively. Regard x ,
of the boundary
etc. as a function of (t1 , t2 ). We consider first the linear interpolation
values of to the inside of the square,

1
1

(t1 , t2 ) = (1 t1 ) (0, t2 ) (1 t2 ) (0, 0) t2 (0, 1)
2
2

1
1
+ t1 (1, t2 ) (1 t2 ) (1, 0) t2 (1, 1)

2
2
6 In general, (z) = (z) for z on the side separating squares x and y, because of the jump of . This is
x
y
why a linear interpolation of could not be used on the sides. Note however that there are no lattice points on
the sides; every lattice point belongs to a unique square.

1
1
+ (1 t2 ) (t1 , 0) (1 t1 ) (0, 0) t1 (1, 0)
2
2

1
1
+ t2 (t1 , 1) (1 t1 ) (0, 1) t1 (1, 1) .
2
2
437
(119)
(4) Adjust the value of the block spin, if desired, while retaining the values of (0)
at the boundaries and maintaining the smoothness. Again this is done separately for the
squares x, using local coordinates (t1 , t2 ) as above.
x (t1 , t2 ) k(t1 , t2 ),
x (t1 , t2 ) =
k(t1 , t2 ) = sin(t1 ) sin(t2 ).
(120)
If is large or , we may choose so that the block spin condition Cx (x) = 0 is exactly
satisfied

x (x),
= s 2 sin2
(121)
C
2s
when there are s 2 lattice points per square. C takes the average over the lattice points
inside the square x similarly as before. The real function k vanishes at the boundaries of
the square. Its block average is [s sin(/2s)]2 . Therefore Cx (x) = 0 as desired.
When is small, one may either choose = 0 or adjust it such that expression (13)
assumes the smallest value.
The field (0) is determined from x ,

2 1/2
(0) (z) = x (z) + 1
x (z)
(122)
(x).
The positive square root is understood. The result satisfies all the requirements. Since the
0 where z0 are the coordinates of the lower left corner of
z-coordinates are z = ast + z
square x, the discontinuities in (0) across boundaries are of order s 1 if the blocks are
large.
Let us note the locality properties of the construction. For z x, (0) (z) depends only
on the value of the blockspin at x and at the 8 nearest and next nearest neighbors of x.
(0) is an explicitly given function of these 9 values by virtue of the formulas above. It
is a non-polynomial function of () because of the factor 1/(modulus) in Eq. (118) and
the factor with the square root in Eq. (122). But if is sufficiently smooth, these factors
could be expanded to obtain a polynomial approximation.
6.2. Vanishing of the correction term to Polyakovs result in order ln(a /a)
Our result for the effective action appeared not to agree exactly with Polyakovs result
to order ln(a /a). There is to this order a correction term

1
vM (0)
(123)
cos (z) (z).
z
Here we wish to show that this term is actually 0 as a consequence of the extremality
condition on the background field.
438
Remark. There is a very small remainder in the exact result because vM (0) gets replaced
1. However, because is also
by a matrix Q (z, z) which is not diagonal in order Q

small, this term is negligible in first order in Q 1.
To derive the result, we need the equation for the background field in a form which
was not used before.
In this subsection we use the parameterization of fields in terms of the -field which
obeys (z) (x).
For notational simplicity introduce
= (x)
(z)
for z x.
(124)
To first order in the deviation of from ,

1

( ).

Rewritten in terms of
1
S ( ) = S()
= QT S
=
= +
(125)
(126)
the saddle point condition (111) takes the form

S ( ) = C
(127)
with a Lagrange multipliers which is a field on the coarse lattice.

Working out the derivative of S one finds the nonlinear equation

= C .

(128)
is constant on blocks, it follows from Eq. (128) that

Since
= (x)(x) = 0.
C
(129)
= 0. Therefore
By definition

= +

C .
=
In the second equation, Eq. (128) was used.
Inserting the definition of , we compute

1
1
) (z)(C )(z)
(
cos (z) (z) =
cos (z)
z
z

1

(x)(x).
= C
cos
(130)
(131)
(132)
We will show that the integrand in expression (132) is zero. This shows that the correction
term is zero.
439
C (x) = (x)(x) by the block spin definition, with some real (x). Moreover,
is constant on blocks, it follows that
because

1
1
C
(x) = (x) avzx
(z).
cos
cos
(133)
This is again a multiple of the vector (x). But according to Eq. (129), (x)(x) = 0.
Therefore the integrand in expression (132) vanishes, and the result is proven.
7. Tr ln Q for small
Instead of inspecting Q we may start from Eq. (83) of Section 5 again. This has the
advantage that we get another simplification for free which had to be justified by smallness
1 if is not small. The exponent involves Q
in the term proportional . We can
of Q
split this term into
T

C C .
C CQ
T C C + T Q
(134)
If is not too large, the second term can be treated as a perturbation. Assuming small
close enough to 1 [which would be
enough thus serves the same purpose as assuming Q
assured by sufficient smoothness of ].
In 1-loop approximation this perturbation is taken into account to first order, like all
other -dependent terms in the exponent of Eq. (83). Evaluation of the Gaussian integral
gives

1
1 T
Tr ln KG 1 +
Sext,I ( )
2
2

1 T
2 T
TC C Q
C C
Sext,I ( )
+ T Q
8
2
Seff [] = S( ) ln J0 ( )
(135)
with the understanding that the expectation values . . . > are those of a free field theory
with propagator KG 1.
This differs from expression (86) in two ways. The expectation values involving S are
evaluated with KG 1 in place of Q , and 12 Tr ln Q is replaced by 12 Tr ln KG 1 + the
last expectation value. But the first approximation is made in any case, whether is
1 is small. The result is therefore again as stated in the introduction, with
small or Q
expression (22) for 12 Tr ln Q .
For use in the analytic computation of the background field we indicate also the
perturbation formula for Q itself. Splitting the -term in the definition (14) of Q as
in (134), we expand in powers of . The zero order term is KG 1 according to (17) and we
obtain
T

C C KG + .
C CQ
Q = KG 1 KG Q
(136)
440
Acknowledgement
Work supported in part by Deutsche Forschungsgemeinschaft. G.P. was partially
supported through projects FONDECYT, Nr. 1980608, and DICYT, Nr. 049631PA. He
would like to thank the II. Institute for Theoretical Physics of the University of Hamburg
for the kind hospitality.
Appendix A. The Jacobian

Consider the Jacobian J0 for finite as defined by the requirement (9).
C [](x) depends on only through (x). Let us write in place of the variable
(x) in the following. We must compute

1
2
J0 C(x)
= d e 2 ||C [](x)|| .
Let = C(x). Then

C [](x)2 = | |2 | |2 = | |2 | 2 |
if we choose a basis so that points in 0-direction, and write

=
1 ||2 , .
Using the standard representation of the uniform measure d on the sphere in terms of
coordinates , we get expression (11).
A.1. det Q
We also need the Jacobian of the transformation from -variables to -variables. The
integration variables i are the coefficients of in an orthonormal basis (e1 , . . . , eN ) for
the tangent space T (z) S N . Such a basis comes from an orthonormal basis for RN +1 with
e0 = (z). Similarly, the integration variables i are the coefficients of in an orthonormal
basis (f1 , . . . , fN ) for the tangent space T(x) S N , x z. Such a basis comes from an
orthonormal basis for RN +1 with f0 = (x). Since Q = ,
N

d k =
det Qij
d k
if Qej =
fi Qij .
(A.1)
i=1
Unwritten arguments are z.

The modulus of the determinant is independent of the choice of orthonormal bases since
det O = 1 for orthogonal transformations O. Therefore we may choose convenient bases
as follows.
e1 =

1
(z)
=
(x) (z)( (z)(x) ,
| (z)| sin (z)
(A.2)
441
and e2 , . . . , eN an arbitrary completion to an orthonormal basis. Similarly we choose

1
(z)
=
(z) (x)( (z)(x) ,
| (z)| sin (z)
fk = ek for k = 2, . . . , N .
f1 =
(A.3)
Basis vectors f0 , f1 are linear combinations of e0 , e1 . Therefore ek , k = 2, . . . , N , are

orthogonal to them and the basis vectors fi are indeed orthonormal. Using Eq. (40) we
compute
Qe1 = cos (z)f1 ,
Qek = fk
for k = 2, . . . , N .
(A.4)
Thus, the matrix (Qij ) is diagonal with a single eigenvalue cos (z) which is distinct from 1.
Therefore
i
det Q
= cos (z).
j
(A.5)
Appendix B. The kinetic term

Let us write
2
| |2 +
0
= | |2 + Lkin .
(B.1)
Our task is to evaluate Lkin . It turns out to be of order | |2 . Therefore it will later be
treated as a perturbation which needs to be taken into account to first order only.
Conventions. Arguments not written are z. To save brackets, we agree that derivatives
act only on the first factor behind them.
We use the exact lattice Leibniz rule (50) throughout. It turns out that this is essential.
By definition, = and 0 = T , while = 1 T .
The use of the lattice Leibniz rule is slightly subtle, because there are always two ways
to use it which differ by the assignment of which factor is f and which is g. Choices have
and factors in T also
to match, so that factors in T have the same argument, z + ,
have the same argument, z. Apart from this the calculation is straightforward and gives

2 T + (z + )
T T (z + )
(z + )
Lkin = T (z + )

T
+ j (z + + transpose).
(B.2)
Because of the smoothness of , we are only interested in terms up to order | |2 . This
(z) = 1 + negligible when multiplied with two
has been used to approximate T (z + )
factors .
442
Appendix C. The second order term

We wish to evaluate the quantity

T
I = T (z)j (z)(z) T (w)j (w)(w) .
(C.1)
There are two possible contractions and we obtain

I (z, w) = I1 + I2 ,

I1 = tr Q (z, w) j (w)Q (w, z)jT (z) ,

I2 = tr j (z)Q (z, w) j (w)Q (w, z) .
(C.2)
(C.3)
(C.4)
We wish to extract the singular part which is proportional to ln a/a

since this part will
T
( jT (z)) we see by partial
contribute to the Polyakov result. Using j (z) = j (z + )
integration that the singular part of I1 and I2 are equal.
We need the singular part of the 1-loop Feynman graph
,&
(z, w) = Q (z, w) Q (w, z)& .
G
(C.5)
The singular part does not depend on Q nor on details of the cutoff. As in Section 4.1 we
may therefore replace the propagators by Yukawa potentials vM (z w)1 with mass M of
order a 1 .
By power counting and rotational invariance the singular part must be of the form
G & (z w)
with an unknown coefficient G. The coefficient can be computed by considering G .
Since = 1 + nonsingular it follows that G = 1/2.
Inserting this yields

I (z, w) = vM (0) tr j (z)T j (z) + nonsingular.
(C.6)
w
z + )
are equal, Eq. (C.6) shows that
Since the singular parts of vM (0) and KG (z + ,
(2)
is indeed finite in the limit a
0.
the renormalized Feynman diagram Seff
Appendix D. Fourier transform of KupiainenGawedzki kernels

Given a lattice of lattice spacing a with points z, w, . . . and a block lattice of lattice
spacing a = sa, (s a positive integer) with sites x, y, . . . , we characterize the points by real
coordinates z , respectively, x , etc. The conjugate variables k and p take their values
in the duals and ,
< k ,
a
a
< p .
a
a
(D.1)
(D.2)
443
If the lattices are infinitely extended, p and k are real variables. If the lattice has
extension La = L a instead,
2
Z.
La

We use the notation k (. . .) = d 2 k if is infinitely extended, and
2

2
(. . .) =
(. . .)
La
p , k
otherwise. The same formulas is used for

different according to Eq. (D.2). Let

Z, < l
D = l
l
.
a
a
a
p,
(D.3)
(D.4)
only the boundaries of the integration are
(D.5)
Then every k admits a unique decomposition

k = p + l,
p , l D.
(D.6)
The Fourier transform of the massless lattice propagator v(z w) is

1
2
v(k)
=
[1 cos k a]
.
a2
(D.7)
=1,2
Because of invariance under translations by lattice vectors of the block lattice, the
averaging kernel C, interpolation kernel A AKG , block propagator u = uKG and high
frequency propagator KG admit Fourier expansions [20]

ik (x z )

C(x, z) = (2)2
(D.8)
,
C(k)e
k
A(z, x) = (2)
ik (x z )
A(k)e
,
(D.9)
u(x y) = (2)2
ip(xy)
u(p)e
(D.10)
(z, w) = (2)2

ll (p)ei(l zlw) eip(zw) .
(D.11)
l,l D
The averaging kernel C(x, z) = a 2 for z x and 0 otherwise. Assuming z x iff a/2 <
z x a/2 one obtains

2
=
C(k)
(D.12)
sin(k a/2) .
ak

=1,2
Notational convention. When variables k, p appear together in one formula, they are
related by the unique decomposition (D.6).
444
From Eqs. (26) one computes

+ l)|2 ,
v(p
+ l)|C(p
u(p)
(D.13)
lD
= v(k)

A(k)
C(k)
u(p)
1 ,
+ l )C(p
+ l)v(p
+ l) A(p
+ l).
ll (p) = ll v(p
(D.14)
(D.15)
The variables l, l assume s 2 values each.

In d dimensions, the formulas remain valid, except that = 1, . . . , d, and factors (2)d
have to be substituted for (2)2 . The variables l, l now assume s d values.
References
[1] T. Balaban, Regularity and decay properties of lattice Greens functions, Commun. Math.
Phys. 89 (1983) 571.
[2] T. Balaban, J. Imbrie, A. Jaffe, Exact renormalization group for gauge theories, in: G. t Hooft et
al. (Eds.), Progress in Gauge Field Theories, 1983 Cargse lectures, Plenum, New York, 1984,
pp. 79104;
T. Balaban, J. Imbrie, A. Jaffe, Renormalization of the Higgs model: minimizers, propagators
and the stability of mean field theory, Commun. Math. Phys. 97 (1985) 299.
[3] T. Balaban, Renormalization group approach to lattice gauge field theories I, Commun. Math.
Phys. 109 (1987) 249, especially Eq. (2.10);
T. Balaban, Spaces of regular gauge field configurations on a lattice and gauge fixing conditions,
Commun. Math. Phys. 99 (1985) 75, Section E;
T. Balaban, The variational problem and background fields in renormalization group method
for lattice gauge theories, Commun. Math. Phys. 102 (1985) 277, Section C.
[4] T. Balaban, The large field renormalization operation for classical N -vector models, Commun.
Math. Phys. 198 (1998) 493, and earlier work.
[5] G. Benfatto, M. Cassandro, G. Gallavotti, F. Nicolo, E. Olivieri, E. Presutti, E. Scacciatelli,
Some probabilistic techniques in field theory, Commun. Math. Phys. 59 (1978) 143.
[6] W. Bietenholz, T. Struckmann, Perfect lattice perturbation theory: a study of the anharmonic
oscillator, hep-lat/9711054, to appear in Int. Mod. Phys. C.
[7] A. Dimakis, F. Mller-Hoissen, T. Striker, Noncommutative differential calculus and lattice
gauge theory, J. Phys. A 26 (1993) 1927;
A. Dimakis, F. Mller-Hoissen, T. Striker, From continuum to lattice theory via deformation of
the differential calculus, Phys. Lett. B 300 (1993) 141.
[8] R.L. Dobrushin, S.B. Shlosman, Absence of breakdown of continuous symmetry in twodimensional models of statistical physics, Commun. Math. Phys. 42 (1975) 31.
[9] J. Feldmann, J. Magnen, V. Rivasseau, R. Seneor, A renormalizable field theory: the massive
GrossNeveu model in two dimensions, Commun. Math. Phys. 103 (1986) 67.
[10] M. Fischer, Magnetic critical point exponents their interrelations and meaning, J. Appl.
Phys. 38 (1967) 981.
[11] K. Gawedzki, A. Kupiainen, A rigorous block spin approach to massless lattice theories,
Commun. Math. Phys. 77 (1980) 3164.
[12] K. Gawedzki, A. Kupiainen, Massless lattice 44 theory: rigorous control of a renormalizable
asymptotically free model, Commun. Math. Phys. 99 (1985) 197.
[13] M. Griel, Self-Consistent Calculation of Real Space Renormalization Group Flows and
Effective Potentials, Dissertation, Hamburg, 1997.
445
[14] M. Griel, G. Mack, Y. Xylander, G. Palma, Self-Consistent Calculation of Real Space

Renormalization Group Flows and Effective Potentials, Nucl. Phys. B 477 (1996) 878.
[15] P. Hasenfratz, F. Niedermayer, Perfect lattice action for asymptotically free theories, Nucl. Phys.
B 414 (1994) 785;
T. DeGrand, A. Hasenfratz, P. Hasenfratz, F. Niedermayer, Nonperturbative tests of the fixed
point action for SU(3) lattice gauge theory, Nucl. Phys. B 454 (1995) 615;
T. DeGrand, A. Hasenfratz, P. Hasenfratz, F. Niedermayer, The classically perfect fixed point
action for SU(3) gauge theory, Nucl. Phys. B 454 (1995) 587.
[16] P. Hasenfratz, Prospects for perfect actions, Nucl. Phys. B (Proc. Suppl.) 63 (1998) 53.
[17] P. Hasenfratz, F. Niedermayer, Fixed point actions in one loop perturbation theory, Nucl. Phys.
B 507 (1998) 399.
[18] M. Bartels, G. Mack, G. Palma, One loop improved lattice action for the nonlinear sigma-model,
Nucl. Phys. B (Proc. Suppl.) 83 (2000) 881883.
[19] U. Kerres, Blockspin transformations for finite temperature field theories with gauge fields,
Dissertation, Univ. Hamburg, July 96, report DESY 96-164.
[20] U. Kerres, G. Mack, G. Palma, Perfect three-dimensional lattice actions for four-dimensional
quantum field theories at finite temperature, Nucl. Phys. B 467 (1996) 510.
[21] G. Mack, Multigrid methods in quantum field theory, in: G. t Hooft et al. (Eds.), Nonperturbative Quantum Field Theory, NATO ASI series B: Physics, Vol. 185, Plenum, New York, 1988.
[22] A.M. Polyakov, Interaction of Goldstone particles in two dimensions, Phys. Lett. 59B (1975)
79.
[23] Y. Xylander, Lattice renormalization group studies of the two-dimensional O(N ) symmetric
non-linear model, Dissertation, Univ. Hamburg, 1997, DESY 97-017.
[24] M. Griessl, J. Wrthner, Software package for calculation of RG interpolation kernels and high
frequency propagators, http://lienhard.desy.de/mackag/cs/RG_KGkernels.

(2)
Dn+1
reflection K-matrices
A. Lima-Santos
Universidade Federal de So Carlos, Departamento de Fsica
Caixa Postal 676, CEP 13569-905 So Carlos, Brazil
Abstract
We investigate the possible regular solutions of the boundary YangBaxter equation for the vertex
(2)
models associated with the Dn+1 affine Lie algebra. We have classified them in terms of three types
of K-matrices. The first one has n + 2 free parameters and all the matrix elements are non-zero. The
second solution is given by a block diagonal matrix with just one free parameter. It turns out that for
n even there exists a third class of K-matrices without any free parameter. 2001 Elsevier Science
PACS: 75.10.Jm; 05.90.+m
1. Introduction
Recently there have been many efforts to introduce boundaries into integrable systems
for possible applications in condensed matter physics and statistical systems with nonperiodic boundary conditions. The bulk Boltzmann weights of an exactly solvable lattice
system are usually the non-zero matrix elements of a R-matrix R(u) which satisfies the
YangBaxter equation. The boundaries entail new physical quantities called reflection
matrices which depend on the boundary properties.
By considering systems on a finite interval with independent boundary conditions at
each end, we have to introduce reflection matrices to describe such boundary conditions.
Integrable models with boundaries can be constructed out of a pair of reflection K-matrices
K (u) in addition to the R-matrix. K+ (u) and K (u) describe the effects of the presence
of boundaries at the left and the right ends, respectively.
Integrability of open chains in the framework of the quantum inverse scattering method
was pioneered by Sklyanin. In Ref. [1], Sklyanin has used his formalism to solve, via
algebraic Bethe ansatz, the open spin-1/2 chain with diagonal boundary terms. This model
had already been solved via coordinate Bethe ansatz by Alcaraz et al. [2].
E-mail address: dals@df.ufscar.br (A. Lima-Santos).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 4 5 - 5
A. Lima-Santos / Nuclear Physics B 612 [FS] (2001) 446460
447
The Sklyanin original formalism was extended to more general systems by Mezincescu
and Nepomechie in [3], where it is assumed that for a regular R-matrix satisfying the
properties
PT-symmetry:
P12 R12 (u)P12 = R21 (u),
unitarity:
R12 (u)R21 (u) = 1,
t2
(u )(U 1)1 ,
crossing unitarity: R12 (u) = (U 1)R12
(1.1)
one can derive an integrable open chain Hamiltonian

H=
N1
Hk,k+1 +
k=1
0

1 dK (u)
tr0 K + (0)HN,0
,
1
+
2
du u=0
tr K+ (0)
(1.2)
where R12 = R 1, R23 = 1 R, etc. P = R(0) is the permutation matrix and the two-site
bulk Hamiltonian Hk,k+1 is given by Hk,k+1 = Pk,k+1 [dRk,k+1 (u)/du]u=0 .
The matrix K (u) satisfies the right boundary YangBaxter equation, also known as the
reflection equation (RE),
1
R12 (u v) K (u)R21 (u + v) K (v)

2
=K (v)R12 (u + v) K (u)R21 (u v),
(1.3)
which governs the integrability at boundary for a given bulk theory. A similar equation
should also hold for the matrix K+ (u) at the opposite boundary. However, for the case of
models whose matrix R(u) satisfies (1.1), one can show that the corresponding quantity
t
(u )M,
K+ (u) = K
M = U tU = M t,
(1.4)
satisfies the left RE. Here is a crossing parameter and U a crossing matrix, both being
specific to each model [4,5]. ti stands for the transposition taken in the ith space and tr0 is
the trace taken in the auxiliary space.
Due to the significance of the RE in the construction of integrable models with open
boundaries, a lot of work has been directed to the study [69] and classification [1012] of
K-matrices.
In spite of all these papers, there is an interesting vertex model based on the non(2)
exceptional Dn+1 Lie algebra for which, until recently, little was known about the solution
of the corresponding RE [13].
While the investigation of particular solutions has been made to a number of lattice
models, Batchelor et al. [10] have derived diagonal solutions of the RE for face and vertex
models associated with several affine Lie algebras, the classification of all possible Kmatrices has been a harder problem. However, recently we have proposed a method which
allowed to classify all possible K-matrices solutions for 19-vertex models [11].
In this paper we demonstrate that this technique is indeed much more general and can
be used, in principle, to study all vertex models based on the Lie algebras [4,5]. In order
(2)
vertex model: it is seen as the most difficult among the
to show that, we choose the Dn+1
non-exceptional cases, as argued recently by Martins and Guan [13].
448
(2)
We have organized this paper as follows. In Section 2 we choose the Dn+1
reflection
equations and in Section 3 their solutions are derived and classified in three types. The last
section is reserved for the conclusion. The D2(2) type-I solution is presented in Appendix A.
(2)
2. The Dn+1 reflection equations

(2)
The R-matrix for the vertex models associated to the Dn+1 affine Lie algebra as
presented by Jimbo in [14] has the form

R=
aij Eij Ei j + a1
Eii Eii
i,j =n+1,n+2
i=n+1,n+2
+ a2
Eii Ejj
i=j,j
i or j =n+1,n+2
+ a3
Eij Ej i + a4
i<j,i=j
i,j =n+1,n+2
+ a5
Eij Ej i
i>j,i=j
i,j =n+1,n+2
(Eij Ej i + Ej i Ei j )
i<n+1
j =n+1,n+2
+ a6
i>n+2
j =n+1,n+2
+ a7
i<n+1
j =n+1,n+2
+ a8
i>n+2
j =n+1,n+2
i=n+1,n+2
j =n+1,n+2
i=n+1,n+2
bi+ [Eij Ei j + Ej i Ej i ] + bi [Eij Ei j + Ej i Ej i ]

c+ Eii Ei i + c Eii Eii + d + Eii Ei i + d Eii Eii ,
(2.1)
where Eii denotes the elementary 2n + 2 by 2n + 2 matrices (ia ib ) and the Boltzmann
weights with functional dependence on the spectral parameter u are given by

a1 = e2u q 2 e2u q 2n ,
a2 = q e2u 1 e2u q 2n ,

a4 = e2u a3 ,
a3 = q 2 1 e2u q 2n ,

1
1
a 5 = eu + 1 a 3 ,
a 6 = eu + 1 eu a 3 ,
2
2

1 u
1 u
a7 = e 1 a3 ,
(2.2)
a 8 = e 1 eu a 3 ,
2
2
449
and for i, j = n + 1, n + 2

2 2u
(i = j ),
q e q 2n e2u 1

(i < j ),
q 2 1 q 2n+ij e2u 1 ij e2u q 2n
aij =

j 2u
q 2 1 e2u q i
e 1 ij e2u q 2n
(i > j ),

bi
(2.3)

(i < n + 1),
q i1/2 q 2 1 e2u 1 eu q n

q in5/2 q 2 1 e2u 1 eu eu q n (i > n + 2),
(2.4)

1 2
q 1 q n + 1 eu eu 1 eu q n + q e2u 1 e2u q 2n ,
2
(2.5)
c =

1 2
q 1 q n 1 eu eu 1 eu q n ,
2
with special attention to the notation
(i < n + 1),
i + 1
i = n + 3/2 (i = n + 1, n + 2) and i = 2n + 3 i,
i 1
(i > n + 2).
d =
(2.6)
(2.7)
Here q = e2 denotes an arbitrary parameter.

For each value of n the matrix R(u) is regular, R(0) = f (0)P , and satisfies PTsymmetry and unitarity,
R21 (u) = P12 R12 (u)P12 ,
R12 (u)R21 (u) = f (u)f (u)1,
(2.8)
where

f (u) = e2u q 2 e2u q 2n .
(2.9)
After we normalize the Boltzmann weights by a factor q n+1 e2u , the crossing-unitarity
symmetry
t2
(u )(U 1)1 ,
R12 (u) = (U 1)R12
(2.10)
holds with the crossing matrices U and crossing parameters , respectively, given by

Uij = i j q (ij )/2 ,
= n ln q = 2n.
(2.11)
Regular solutions of the RE mean that the matrix K (u) in the form
K (u) =
2n+2
kij (u)Eij
(2.12)
i,j =1
satisfies the condition

kij (0) = ij ,
i, j = 1, 2, . . . , 2n + 2.
(2.13)
Substituting (2.1) and (2.12) into (1.3), we have in both 16(n + 1)4 functional equations
for the kij elements, many of which are dependent. In order to solve them, we shall proceed
450
in the following way. First we consider the (i, j ) component of the matrix equation (1.3).
By differentiating it with respect to v and taking v = 0, we get algebraic equations
involving the single variable u and 4(n + 1)2 parameters

dkij (v)
ij =
(2.14)
, i, j = 1, 2, . . . , 2n + 2.
dv v=0
Second, these algebraic equations are denoted by E[i, j ] = 0 and collected into blocks
B[i, j ], i = 1, . . . , 2(n + 1)2 and j = i, i + 1, . . . , (2n + 2)2 i, defined by
E[j, i] = 0,
E[i,

j ] = 0,2
2
B[i, j ] = E 4(n + 1) + 1 i, 4(n + 1) + 1 j = 0,
(2.15)

2
2
E 4(n + 1) + 1 j, 4(n + 1) + 1 i = 0.
For a given block B[i, j ], the equation E[4(n + 1)2 + 1 i, 4(n + 1)2 + 1 j ] = 0 can be
obtained from the equation E[i, j ] = 0 by interchanging
kij ki j ,
ij i j ,
bi bi ,
a3 a4 ,
a6 a6 ,
a7 a8 .
aij ai j ,
(2.16)
and the equation E[j, i] = 0 is obtained from the equation E[i, j ] = 0 by the interchanging
kij kj i ,
ij j i ,
aij aj i .
(2.17)
In this way, we can control all equations and a particular solution is simultaneously
connected with at least four equations.
Since the R-matrix (2.1) satisfies unitarity, P and T invariances and crossing symmetry,
the matrix K+ (u) is obtained using (1.4) with the following M-matrix
Mij = ij q 2n+32i ,
i, j = 1, 2, . . . , 2n + 2.
(2.18)
(2)
3. The Dn+1 K-matrix solutions

(2)
Analyzing the Dn+1 RE one can see that they possess a special structure. Several
equations exist involving only the elements out of a block diagonal structure which consist
of the diagonal elements kii plus the central elements of the secondary diagonal, kn+1,n+2
and kn+2,n+1 . This block diagonal structure commutes with n distinct U (1) symmetries,
the minimal symmetry of the R-matrix [13]. Through this structure we can identify two
types of solutions: the type-I solution which is a general K-matrix solution with all matrix
elements non-zero and the type-II solution for which the K-matrix has this block diagonal
structure. Moreover, for the models with an even number of U (1) conserved charges, we
can find a third type of solution which manifests the U (1) U (1) symmetries. Therefore
for n even we can find the type-III solution which is a diagonal K-matrix.
Having identified these possibilities we may proceed in order to find explicitly the
regular solutions. We start looking for the type-I solution.
451
3.1. The type-I solution

The simplest RE are those involving only the elements of the secondary diagonal. We
chose to express their solutions in terms of the element k1,2n+2 :
kii =
ii
k1,2n+2 ,
1,2n+2
i = {n + 1, n + 2}.
(3.1)
From the collections {B[i, j ]}, i = 1, 2, . . . , n 1 one can see that the RE of the last blocks
of each collection are simple and we can easily solve them expressing the elements kij with
j = i in function of k1,2n+2

Fij (ij a3 aii j i a2 aij )k1,2n+2 (i < j ),
kij =
(3.2)
Fij (ij a4 aii j i a2 aij )k1,2n+2 (i > j ),
where
Fij =
a1 aii a22
1,2n+2 aii2 a3 a4 a22 aij aj i
.
(3.3)
Moreover, for j = n + 1, n + 2 with i = j, j we have

i aii (ij a5 + ij a7 ) a2 (j i bi+ + j i bi ) k1,2n+2
kij =

i aii (ij a6 + ij a8 ) a2 (j i bi+ + j i bi ) k1,2n+2
and for i = n + 1, n + 2 with j = i, i we get

j ajj (ij a5 + i j a7 ) a2 (j i bj+ + j i bj ) k1,2n+2
kij =

j ajj (ij a6 + i j a8 ) a2 (j i bj+ + j i bj ) k1,2n+2
(i > j ),
(i < j ),
(3.4)
(i > j ),
(i < j ),
(3.5)
where
l =
a1 all a22
1,2n+2 all2 (a6 + a8 )(a5 + a7 ) a22(bl+ + bl )(bl+ + bl )
.
(3.6)
(2)
Here we observe that for the D2 model, Fij = 0 and l = 00 . However, through an
appropriate choice of l we can include the case n = 1 in our discussion (see Appendix A).
We substitute these expressions in the remaining RE (in fact, it is enough to consider the
equations of the collections {B[1, j ]} and {B[2, j ]}), and look at the equations of the type
G(u)k1,2n+2 (u) = 0,
(3.7)

where G(u) = k fk ({ij })eku . The constraint equations fk ({ij }) 0, k, can be solved
in terms of the 2n + 2 parameters which allow us to find all kij elements out of the blockdiagonal structure in terms of the k1,2n+2 .
Of course, the expressions for kij will depend on our choice of these parameters.
After some attempts we concluded the choice 12 , 13 , . . . , 1,2n+2 and 21 as the most
appropriate for our purpose.
Taking into account the fixed parameters and the Boltzmann weights defined above we
can rewrite these kij matrix elements for n = 1 in the following way.
452
The elements in the secondary diagonal of K (u) are given by

n1
2
q
+ 1 2 1,i
i2n

ki,i (u) = q
k1,2n+2 (u) (i = 1, 2n + 2),
q +1
1,2n+2
and

k2n+2,1 (u) = q
2n3
21
(3.8)
2
k1,2n+2 (u).
(3.9)
1,2n+1
The elements of the first row and the elements of the first column are, respectively, given
by
k1,j (u) =
q n1 + 1 1,j (u)
k1,2n+2 (u) (j = 2n + 2),
e2u + q n1 1,2n+2
q n1 + 1 1,i (u) 21
k1,2n+2 (u)
e2u + q n1 1,2n+2 1,2n+1
while for the last column and for the last row, we have
ki,1 (u) = q i3
ki,2n+2 (u) = q in2
(i = 2n + 2),
q n1 + 1 1,i (u) 2u
e k1,2n+2 (u) (i = 1),
e2u + q n1 1,2n+2
q n1 + 1
21 1,j (u) 2u
e k1,2n+2 (u)
e2u + q n1 1,2n+1 1,2n+2
The remaining matrix elements are given by
k2n+2,j (u) = q n2
kij (u) = q in1
(3.10)
(j = 1).
q n1 + 1 q n1 + 1 1,i (u) 1,j (u)

k1,2n+2 (u)
q + 1 e2u + q n1 1,2n+2 1,2n+2
(3.11)
(3.12)
(3.13)
(3.14)
for i > j , and
kij (u) = q i2n1
q n1 + 1 q n1 + 1 1,i (u) 1,j (u) 2u

e k1,2n+2(u)
q + 1 e2u + q n1 1,2n+2 1,2n+2
(3.15)
for i < j .
In these expressions we are using a compact notation defined by
1,a
(a = n + 1, n + 2),

1 u
(a = n + 1),
e + +
1,a (u) = 2
(3.16)
u

1
e + + (a = n + 2),
2
and
1,a
(a = n + 1, n + 2),

1 n u
(a = n + 1),
q e + +
1,a (u) = 2
(3.17)

1
q n eu + + , (a = n + 2),
2
where = 1,n+1 1,n+2 . To include the case n = 1 we will make some modifications
in these expressions (see Appendix A).
453
At this point we find the 2n(2n + 3) matrix elements in terms of the 2n + 2 parameters.
However, we still need to find the 2n + 4 matrix elements that belong to the block diagonal.
The block diagonal structure has the form
diag(k11 , k22 , . . . , knn , B, kn+3,n+3 , . . . , k2n+2,2n+2 ),
where B contains the central elements,

kn+1,n+1 kn+1,n+2
,
B=
kn+2,n+1 kn+2,n+2
(3.18)
(3.19)
Here the situation is a little different. It is very cumbersome to write these matrix elements
in terms of the Boltzmann weights. But, after some algebraic manipulations, it is possible
to see that the diagonal elements satisfy two distinct recurrence relations, namely,
q n1 + 1 11 ii
k1,2n+2
(i < n + 1),
11 e2u + q n1
1,2n+2
kii =
(3.20)
q n1 + 1 n+3,n+3 ii 2u
e
k
(i
>
n
+
2).
k
n+3,n+3
1,2n+2
1,2n+2
e2u + q n1
Substituting (3.20) into the RE we can find k11 and kn+3,n+3 and consequently, all
kii
/ B will be known after finding the 2n parameters ii .
The solution of this problem depends on the parity of n. Besides, in this stage, all
remaining parameters ij , including those associated with the central elements, are fixed
in terms of n + 3 parameters.
For n odd the solution is

n1
2 2 ) + 2(e2u q n )(q n 2 + 2 )
+ 1 (q + 1)(e2u + 1)(q n
q
+
+
k11 (u) =
2
q +1
81,2n+2 q n1/2 (e2u + 1)

n1
2q(q n1 1) + (q 1)(e2u + 1)
+1
q
+
k1,2n+2 (u), (3.21)
1,2n+2 (q n 1)(e2u 1)
e2u + q n1
kn+3,n+3 (u)
n1

2 2 ) 2(e2u q n )(q n 2 + 2 )
q
+ 1 (q + 1)(e2u + 1)(q n
+
+
=
2
n1/2
2u
q +1
81,2n+2 q
(e + 1)

n1
n1
2u
2q(q
1) + (q 1)(e + 1)
+1
q
+
e2u k1,2n+2 (u). (3.22)
1,2n+2 (q n 1)(e2u 1)
e2u + q n1
The parameters ii , i = n + 1, n + 2 are fixed by the following recurrence relations

ii + (q)i1 odd
(i < n),
i+1,i+1 =
(3.23)
ii + (q)in3 odd (i > n + 2),
with

n+3,n+3 = 11 + 2 +

2 + 2 )
q n1 + 1 (q n 1)(q n
+
q +1
4q n1/21,2n+2
(3.24)
454
and
odd =
2 qn 2 )
(q + 1)2 (q + 1)(q n1 + 1)(
+
.
n
n1/2
q 1
8q
1,2n+2
(3.25)
Finally, we can solve the last RE to find the central elements. The solution is
n1

2 2

q
+ 1 q n
+ 2u
kn+1,n+1 (u) =
e + qn
n1/2
q +1
8q

n1
2u
(e + 1)(e2u q n )
+ 1 k1,2n+2(u)
q
,
+ 1,2n+2 2u
2
(e 1)(q n 1)
e2u + q n1
1.2n+2
kn+2,n+2 (u) = kn+1,n+1 (u),
(3.26)

2 + 2 )eu 2q n (e2u + 1)
(q n1 + 1)2 (q n + 1)(q n
+
+
kn+1,n+2 (u) =
4q n1/2(q + 1)(e2u + 1)(e2u + q n1 )
eu k1,2n+2 (u)
(3.27)
,
2
1,2n+2

2 + 2 )eu + 2q n (e2u + 1)
(q n1 + 1)2 (q n + 1)(q n
+
+
kn+2,n+1 (u) =
4q n1/2(q + 1)(e2u + 1)(e2u + q n1 )
u
e k1,2n+2 (u)
.
(3.28)
2
1,2n+2
Moreover, there are n parameters which have been fixed by the RE

n1
q
+ 1 2 1,n 1,n+3 1,2n+1
1
(n = 1),
21 = 2n3
2
q +1
q
1,2n+2
(3.29)

2 2 ) + 8q n1/2 (q + 1)
(q n 1)(q n1 + 1)(q n
1,2n+2 (q + 1)
+
8 q(q n 1)(q n1 + 1)1,n+3

(n = 1),
(3.30)
1,n =
1j = ()j 1
1,n 1,n+3
,
1,2n+3j
j = 2, 3, . . . , n 1.
(3.31)
The final result is a general solution with n + 3 free parameters 11 , 1,n+1 , 1,n+2 , . . . ,
1,2n+2 .
By the choice
1
k1,2n+2 (u) = 1,2n+2 (e2u 1),
(3.32)
2
one can, for instance, fix the parameter 11 using the regular condition (2.13) and in that
(2)
models with n odd.
way end with a (n + 2)-parameter solution for Dn+1
The corresponding matrix K+ (u) is obtained used (1.4) and in this case we have
Tr(K+ (0)) = 0. Therefore, Eq. (1.2) gives the corresponding integrable open chain
Hamiltonians, where Hk,k+1 are the tensors sk,k+1 (q) derived by Jimbo [14].
455
Now we turn to the description of the even solution. For n even we find the following
expressions for k11 and kn+3,n+3 , respectively,
k11 (u) =
k1,2n+2 (u)
2
n1/2
2q
1,2n+2 (e4u 1)(e2u + q n1 )

2 2u
2
(q + 1) e2u q n q n
e +

+ 2 q 1,n 1,n+3 e2u 1 2 e2u q n (q

+ 1) e2u + 1 ,
(3.33)
e2u k1,2n+2 (u)
kn+3,n+3 (u) =
2
2q n1/21,2n+2
(e4u 1)(e2u + q n1 )

2u

2
2 2u
(q + 1) e q n q n
+
e

2 q 1,n 1,n+3 e2u 1 2 e2u q n (q + 1) e2u + 1 .

(3.34)
The central elements are given by
kn+1,n+1 (u) = kn+2,n+2 (u) = 2
e2u
q n1 + 1 k1,2n+2 (u)
,
e2u 1 e2u + q n1 1,2n+2
kn+1,n+2 (u)

q n1 + 1 2
=
2
q +1
4q n1/2 1,2n+2
(e2u + q n1 )(e2u + 1)

2
2 2
(q n
+ +
) (q + 1) q n + 1 e2u 2 + (q + 1)q n eu e2u + 1

4 q 1,n 1,n+3 e2u q n e2u 1 ,

k1,2n+2(u)
(3.35)
(3.36)
kn+2,n+1 (u)

n1
q
+1 2
k1,2n+2(u)
=
2
q +1
4q n1/2 1,2n+2
(e2u + q n1 )(e2u + 1)

2
2 2
q n
+ +
(q + 1) q n + 1 e2u + 2 + (q + 1)q n eu e2u + 1

4 q 1,n 1,n+3 e2u q n e2u 1 .
The ii parameters are fixed by the following recurrence relations

(i < n),
ii + (q)i1 even
i+1,i+1 =
ii (q)in3 even (i > n + 2),
with
n+3,n+3 = 11 + 2
(3.37)
(3.38)
2 + 2 ) + 8
3/2 (q n1 + 1)
2(q n 1)(q + 1)(q n
1,n 1,n+3 q
+
2 2 )
(q + 1)(q n 1)(q n
+
(3.39)
and
q (q + 1) 1,n 1,n+3
even = 8
.
2 qn 2
q n 1
+
(3.40)
456
Now the n fixed parameters are

21 = q 32n
q n1 + 1
1,n 1,n+3
,
1,2n+1 2
q +1
1,2n+2

1 q n 1 q n1 + 1 n 2
2
q +
,
n1/2
8q
q +1
1,n 1,n+3
= ()j 1
, j = 2, 3, . . . , n 1,
1,2n+3j
(3.41)
1,2n+2 =
(3.42)
1,j
(3.43)
and the n + 3 free parameters are 11 ,1,n , . . . , 1,2n+1 .

Again, one can fix 11 from the regular condition to get a general solution with n + 2
free parameters. Here we also have Tr(K+ (0)) = 0, n, and (1.2) gives the corresponding
integrable open chain Hamiltonians.
There are many particular solutions which are obtained when some of these free
parameters are zero. Despite the existence of n + 2 free parameters in a type-I solution it
seems impossible to obtain a reduction when certain parameters vanish, in order to obtain
a type-II or type-III solution.
3.2. The type-II solution
(2)
models have n distinct U (1) conserved charges,
As was already mentioned, the Dn+1
and the K-matrix ansatz compatible with these symmetries is the block diagonal structure
presented in the previous section.
Looking for the general solution of the corresponding RE we find that the only possible
solution is obtained when the two recurrence relations (3.20) are degenerated into k11 and
into kn+3,n+3 , respectively,
kn,n (u) = kn1,n1 (u) = = k22 (u) = k11 (u),
(3.44)
k2n+2,2n+2 (u) = k2n+1,2n+1 (u) = = kn+4,n+4 (u) = kn+3,n+3 (u).
(3.45)
and
The type-II solution can be obtained by the same procedure described before and in what
follows we only quote the final results.
We have found two solutions for any value of n. The first solution is given by
k11 (u) =
1 (e2u + q n )(q n 1)(e2u + 1) n+1,n+2 (q n + 1)(e2u 1)

,
2
e2u (q 2n 1)
(3.46)
1 (e2u + q n )(q n 1)(e2u + 1) + n+1,n+2 (q n + 1)(e2u 1)

,
2
(q 2n 1)
(3.47)
with central elements
kn+3,n+3 (u) =

1
kn+1,n+2 (u) = kn+2,n+1 (u) = n+1,n+2 e2u 1 ,
2
(3.48)
457

1 2u
(e2u 1)
kn+1,n+1 (u) = e + 1 1 + u 2n
,
2
e (q 1)
(3.49)

1 2u
(e2u 1)
kn+2,n+2 (u) = e + 1 1 + u 2n
,
2
e (q 1)
(3.50)
where
=

1
2
2q n (q n + 1)2 n+1,n+2
(q n 1)2

(q n + 1)2 n+1,n+2 + (q n 1)2

2
q n (q n + 1)2 n+1,n+2
(q n 1)2
(3.51)
and

= (q n + 1)2 n+1,n+2 + (q n 1)2

2
(q n 1)2 .
2 q n (q n + 1)2 n+1,n+2
(3.52)
The signs () and () are indicating the existence of two conjugated solutions. Here
we notice that these solutions degenerated into two complex diagonal solutions when
n+1,n+2 = 0.
For the second solution we have

1 (e2u q n )
2u
n
n 2
1)
(q
+
1)
2
q
1
(e
k11 (u) =
n+1,n+2
n+1,n+2
2 eu (q n 1)2

2u
n
(3.53)
(e + 1)(q 1) ,
kn+3,n+3 (u)
=

1 eu (e2u q n )
2u
n
n 2
(e
1)
(q
+
1)
2
q
1
n+1,n+2
n+1,n+2
2 (q n 1)2

+ (e2u + 1)(q n 1) ,
(3.54)
with the following central elements

1
kn+1,n+1 (u) = kn+2,n+2 (u) = eu (e2u + 1),
2

1 (e2u 1)
kn+1,n+2 (u) =
n+1,n+2 eu (q n + 1)2 2q n (e2u + 1)
2 (q n 1)2

2
(q n + 1) q n n+1,n+2
1 (eu 1)2 ,
(3.55)
(3.56)
kn+2,n+1 (u) =
u n

1 (e2u 1)
2
n 2u
e
(q
+
1)
+
2q
(e
+
1)
n+1,n+2
2 (q n 1)2

2
1 (eu + 1)2 .
(q n + 1) q n n+1,n+2
(3.57)
458
In contrast to the first solution there does not exist a way of deriving a diagonal solution
starting from these conjugated solutions. Moreover, when n = 1, we have Tr(K+ (0)) = 0.
Therefore, in this case the corresponding boundary term in the integrable open chain
Hamiltonian is not more given by (1.2) but taking into account the second order expansion
of the transfer matrix in the spectral parameter u [15].
3.3. The type-III solution
For the models with an even number of U (1) conserved charge, we look for diagonal
K-matrix solutions which manifest the U (1) U (1) symmetries. Here there is only one
solution, namely, the almost unity solution [13]
k11 (u) = e2u ,
k22 (u) = k33 (u) = = k2n+1,2n+1 (u) = 1,
k2n+2,2n+2 (u) = e2u .
(3.58)
In that point we finished noticing the absence of the trivial solution K (u) = 1 for this
system.
The type-II and type-III solutions were already obtained by Martins and Guan [13].
4. Conclusion
We have investigated the regular solutions of the reflection equations for the vertex
(2)
models associated to the Dn+1 affine Lie algebra. After a systematic study of the functional
equations we find that there are three types of solutions. We call of type-I the K-matrix
solutions with n + 2 free parameters. The type-II are block diagonal matrices with one free
parameter. Finally, the type-III are diagonal matrices without free parameters which only
exist for n even.
The absence of an algebraic method to classify the reflection equation solutions leads
us to believe that our results can be extended in order to pick up all non-exceptional Lie
algebras [4,14], including their supersymmetric partners [5].
Acknowledgement
It is a pleasure to thank M.J. Martins for invaluable discussions. This work was supported
in part by Fundao de Amparo Pesquisa do Estado de So Paulo-FAPESP-Brasil and by
Conselho Nacional de Desenvolvimento-CNPq-Brasil.
Appendix A. The D (2)
2 reflection K-matrices
In this appendix we will present the type-I for the D2(2) model. The matrix K (u) has
the form
k12
k22
k32
k42
k11
k21
K =
k31
k41
k13
k23
k33
k43
459
k14
k24
.
k34
k44
(A.1)
The elements k11 , k44 and k22 = k33 , k23 , k32 are read directly from the n odd solution
taking n = 1 into (3.21), (3.22) and (3.26), (3.27), (3.28), respectively:

2 + 2 ) + (q + 1)(e2u + 1)(q 2 2 ) k (u)
1 2(e2u q)(q
14
+
+
k11 (u) =
2
2u
2
2
14 q(q + 1)(e + 1)
+2
k44 (u) =
k14 (u)
,
14(e2u 1)
(A.2)

2 + 2 ) + (q + 1)(e2u + 1)(q 2 2 ) e2u k (u)
1 2(e2u q)(q
14
+
+
2 q(q + 1)(e2u + 1)2
2
14
+2
e2u k14 (u)

,
14(e2u 1)
(A.3)
2

2 2u
+
414(e2u q) k14 (u)
1 q
e +q
+
,
k22 (u) = k33 (u) =
2
2
q(q + 1) e2u + 1 (q 1)(e2u 1)
14

2 + 2
q
2 q + k14 (u)
eu
+
u
k23 (u) = 2u
e
,
2u
2
e +1
q(e + 1)
q +1
14

2 + 2
q
2 q + k14 (u)
eu
+
u
k32 (u) = 2u
+
,
e
2u
2
q +1
e +1
q(e + 1)
14
(A.4)
(A.5)
(A.6)
where = 12 13 .
Due to the indetermination of l (3.6) when n = 1, we can replace l into (3.4) and
(3.5) by l defined by
l l =
e2u
1
q2 1
.
q 2 e2u 1 + e2u 13 (b1+ + b1 )(b4+ + b4 )
(A.7)
(2)
This replacement allows that the equations (3.9)(3.13) hold for the D2 model up to a
q-factor. The result is
k41 (u) =
2
21
2
13
k14 (u),
(A.8)
k12 (u) =
eu + +
k14(u),
14 (e2u + 1)
k13 (u) =
eu + +
k14 (u),
14 (e2u + 1)
k21 (u) =
21
k13 (u),
13
k31 (u) =
21
k12 (u),
13
(A.10)
k42 (u) =
21
k34 (u),
13
k43 (u) =
21
k24 (u),
13
(A.11)
(A.9)
460
1 qeu + + 2u
k24 (u) =
e k14(u),
q 14 (e2u + 1)
1 qeu + + 2u
e k14 (u),
k34 (u) =
q 14 (e2u + 1)
(A.12)
where 21 is given by
21 =
2 2 ) + 4q(q + 1)
1 (q 1)(q
14
+
13 .
2
2
(q 2 1)14
(A.13)
By the choice

1
k14 (u) = 14 e2u 1
2
we can find 11
2 q 12 13
11 =
q + 1 14
(A.14)
(A.15)
in order to get a regular solution with three free parameters 12 , 13 and 14 .
References
[1] E.K. Sklyanin, J. Phys. A: Math. Gen. 21 (1988) 2375.
[2] F.C. Alcaraz, M.N. Barber, M.T. Batchelor, R.J. Baxter, G.R.W. Quispel, J. Math. Phys. A:
Math. Gen. 20 (1987) 6397.
[3] L. Mezincescu, R.I. Nepomechie, J. Phys. A: Math. Gen. 24 (1991) L17.
[4] V.V. Bazhanov, Phys. Lett. B 159 (1985) 321;
V.V. Bazhanov, Commun. Math. Phys. 113 (1987) 471.
[5] V.V. Bazhanov, A. Schadrikov, Theor. Math. Phys. 73 (1987) 402.
[6] I.V. Cherednik, Theor. Math. Phys. 61 (1984) 977;
I.V. Cherednik, Int. J. Mod. Phys. A 7 (Suppl. 1B) (1992) 707.
[7] H.J. de Vega, A. Gonzlez-Ruiz, J. Phys. A: Math. Gen. 26 (1993) L519;
H.J. de Vega, A. Gonzlez-Ruiz, J. Phys. A: Math. Gen. 27 (1994) 6129.
[8] L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. A 6 (1991) 5231.
[9] T. Inami, S. Odake, Y.-Z. Zhang, Nucl. Phys. B 470 (1996) 419.
[10] M.T. Batchelor, V. Fridkin, A. Kuniba, Y.K. Zhou, Phys. Lett. B 376 (1996) 266.
[11] A. Lima-Santos, Nucl. Phys. B 558 (1999) 637.
[12] C.-x. Liu, G.-x. Ju, S.-k. Wang, K. Wu, J. Phys. A 32 (1999) 3505.
[13] M.J. Martins, X.-W. Guan, Nucl. Phys. B 583 (2000) 721.
[14] M. Jimbo, Commun. Math. Phys. 102 (1986) 537.
[15] R. Cuerno, A. Gonzles-Ruiz, J. Phys. A: Math. Gen. 26 (1993) L605.

Integrable open boundary conditions for the Bariev

model of three coupled XY spin chains
A. Foerster a , M.D. Gould b , X.-W. Guan a , I. Roditi c,d , H.-Q. Zhou b
a Instituto de Fisica da UFRGS, Av. Bento Goncalves, 9500, Porto Alegre, 91501-970, Brazil
b Department of Mathematics, University of Queensland, Brisbane, Qld 4072, Australia
c Centro Brasileiro de Pesquisas Fisicas, Rua Dr. Xavier Sigaud 150, 22290-180, Rio de Janeiro-RJ, Brazil
d C.N. Yang Institute for Theoretical Physics, State University of New York at Stony Brook,
Stony Brook, NY 11794-3840, USA

Abstract
The integrable open-boundary conditions for the Bariev model of three coupled one-dimensional
XY spin chains are studied in the framework of the boundary quantum inverse scattering method.
Three kinds of diagonal boundary K-matrices leading to nine classes of possible choices of boundary
fields are found and the corresponding integrable boundary terms are presented explicitly. The
boundary Hamiltonian is solved by using the coordinate Bethe ansatz technique and the Bethe ansatz
equations are derived. 2001 Elsevier Science B.V. All rights reserved.
PACS: 71.10.-w; 71.10.Fd; 75.10.Jm
Keywords: Integrable spin chains; Algebraic Bethe ansatz; YangBaxter algebra; Reflection equations
1. Introduction
Since the discovery of high temperature superconductivity in cuprates [1], a tremendous
effort has been made to uncover a theoretical framework capable of explaining this amazing
phenomenon. It is a general belief that the properties of strongly correlated electron
systems showing a non-Fermi liquid behaviour are closely related to those materials
showing high Tc superconductivity. This has caused an increasing interest in strongly
correlated electron models [25]. Integrable models that have been widely studied include
the 1D Hubbard model solved by Lieb and Wu [6] and the supersymmetric t-J model [7,8].
Another model with special features relevant for high Tc superconductivity is the 1D Bariev
[9,10], as it exhibits the existence of hole pairs of Cooper type.
E-mail address: guan@if.ufrgs.br (X.-W. Guan).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 5 4 - 6
462
A. Foerster et al. / Nuclear Physics B 612 [FS] (2001) 461478
On the other hand, important progress in the realm of completely integrable systems
is the generalization of the usual quantum inverse scattering method (QISM) [1113] to
incorporate open boundary conditions that preserve integrability [1417]. The presence of
the integrable boundaries leading to a pure back-scattering on each end of a quantum chain
results in rich physical phenomena [1823]. It has been clarified that 1D quantum systems
with boundary fields are closely related to impurity problems. Also due to their connection
to the Kondo problem and boundary conformal field theory in low-dimensional quantum
many-body systems, the integrable boundaries could be a useful nonperturbative way to
investigate the impurity effects in the condensed matter physics. The prototypical 1D
Hubbard model and t-J model have been thoroughly investigated with boundary impurities
and boundary fields [1825]. Subsequently, other strongly correlated electron systems,
such as the 1D Bariev model of two coupled XY spin chains have been partially studied and
the reflections equations (RE) solved [26]. In that case, a new class of boundary reflection
K-matrices for the model lead to pure magnetic boundary fields in the Hamiltonian which
may have a feasible realization by applying boundary external fields in experiments on
quantum wires. An interesting aspect here is that after JordanWigner transformation,
the coupled XY spin chains can be presented as correlated electron systems where the
hopping terms depend on the occupation numbers at sublattices. This is thought to be
useful in studying conductivity properties in such non-Fermi liquids. Along this line, the
integrability and boundary conditions for a Bariev model of three coupled XY chains has
been studied in the frame work of the QISM by Zhou and coworkers [27,28] recently.
However, we have good reasons to expect that the model would permit other kinds of
integrable boundary terms associated with some new solutions to the RE. We revisit the
model seeking a complete understanding of the open boundary conditions for the model.
As noticed in [28] the quantum R-matrix of the model we study does not possess the
crossing-unitarity. This causes a violation of the isomorphism between K+ and K matrix
which satisfy two reflection equations (RE) separately. The integrability at left and right
boundaries demand rather complicated RE which in turn expose bulk symmetries on the
left and right boundary terms. In order to maintain the integrability of models where
the R-matrix does not possess the crossing-unitarity property some new objects have
to be introduced. It is found that the three coupled XY chains permits three kinds of
K-matrices to each RE which lead to nine classes of integrable boundary terms containing
the ones for the two coupled XY spin chain as special cases. These integrable boundary
terms containing different on-site Coulomb interactions and chemical potentials reveal the
symmetry of exchanging the sublattices. We derive the Hamiltonian of the model with
nine classes of boundary terms from the expansion of the boundary transfer matrix around
the zero spectral-parameter point up to the fourth, third and second orders, respectively.
Furthermore, we solve the boundary model by means of the coordinate Bethe ansatz
method and derive the Bethe ansatz equations.
The paper is organized as follows. In Section 2 we construct the Bariev model of
three coupled XY spin chains with nine classes of boundary fields by means of the
QISM adapted to special boundary conditions. The basic quantities, e.g., the R-matrix,
the matrices K defining the boundary terms, the monodromy matrices and the transfer
463
matrices are defined. In particular, we solve two RE separately and obtain three
independent classes of solutions to each REs using the variable-separation method. The
relation between the transfer matrices and the Bariev Hamiltonians with different boundary
fields is established. Section 3 is devoted to the solution of the models through coordinate
Bethe ansatz method. Section 4 presents our conclusions.
2. Boundary K-matrices for the coupled spin chains

We consider a spin chain model defined by the following Hamiltonian
H=
L1
Hj,j +1 + B1(m) + BL(l) ,
(2.1)
j =1
where Hj,j +1 denotes the bulk Hamiltonian density of three XY spin chains coupled to
each other [27]
Hj,j +1 =

j+() j+1() + j() j++1()

exp

=

j++( )( ) j+( )( )
(2.2)

y
y
where j() = 12 jx() ij () with jx(), j () , jz() being the usual Pauli spin
operators at site j corresponding to the th ( = 1, 2, 3) XY spin chain, ( ) is a step
function of ( ) and is a coupling constant; B1 and BL are left and right boundary
terms of the form
B1(m) =

3
1
z
2
cosh
1()
2c
exp(2)
=1
z
z
+ sinh cosh
1()
1()
,=1
=

z
z
z
,
+ sinh2 1(1)
1(2)
1(3)

exp()
z
z
z
cosh
1() + sinh 1(2)1(3) ,
2c
=2
1 z
2c 1(3)
for m = 1,
for m = 2,
for m = 3,
(2.3)
464
BL(l) =

3
1
z
2
L()
cosh
2c
=1
z
z
+
sinh
cosh
L()
L()
,=1
=

z
z
z
,
+ sinh2 L(1)
L(2)
L(3)
z
z
z
cosh
+
sinh
L()
L(1) L(2) ,
2c+ exp()
=1
1 z
2c+ L(1)
for l = 1,
(2.4)
for l = 2,
for l = 3,
where c are parameters describing boundary effects. It is worth mentioning that the
boundary terms containing three- and two-spin interactions and chemical potentials at
the left and right ends are consistent with the bulk symmetry which is a combination of
the inversion j L j + 1 and the exchange among the sublattices. With the different
choices of the pair (m, l) m, l = 1, 2, 3, it appears that there exist nine classes of integrable
boundary terms compatible with the integrability of the model. After a generalized Jordan
Wigner transformation, Hamiltonian (2.1) becomes a strongly correlated electronic system
with boundary interactions. Quite remarkably, if one restricts the Hilbert space to the one
which only consists of, say, (1) , (2) , then Hamiltonian (2.1) reduces to that of two coupled
XY open chains with special boundary interactions, which has been considered in [26].
We now establish the quantum integrability for the system defined by Hamiltonian (2.1),
by using the general formalism described in the paper [17]. To fix notation, let us briefly
recall some basic quantities for the bulk model (2.2) with periodic boundary conditions.
As was shown in [27], the bulk model Hamiltonian commutes with a one-parameter family
of bulk transfer matrix (u) of a two-dimensional lattice statistical mechanics model. This
transfer matrix is the trace of a monodromy matrix T (u), which is defined , as usual, by
T (u) = L0N (u) L01 (u)
(2.5)
with L0j (u) of the form,

(1)
(2)
(3)
L0j (u) = L0j (u)L0j (u)L0j (u),

where
()
L0j (u) =
(2.6)
3

1
1
z
z
z
z
+
1 + j () 0() + u 1 j () 0() exp

0( ) 0( )
2
2

=1
=

3

+
+
+ j ()0() + j () 0() 1 + exp 2

0( ) 0( ) u2 .

=1
=
(2.7)
465
The commutativity of the bulk transfer matrices (u) for different values of the spectral
parameter u follows from the fact that the monodromy matrix T (u) satisfies the Yang
Baxter algebra
1
R12 (u, v) T (u) T (v) = T (v) T (u)R12 (u, v).
(2.8)
The explicit form of the corresponding R-matrix R12 (u1 , u2 ) can be found in [27]. Here
we only emphasize that the local monodromy matrix as well as the quantum R-matrix does
not possess the crossing symmetry. It satisfies the nonadditive YangBaxter equation
R12 (u, v)R13 (u, w)R23 (v, w) = R23 (v, w)R13 (u, w)R12 (u, v).
(2.9)
Following the QISM adapted to the case of special boundary conditions, we define the
doubled monodromy matrix as
U (u) = T (u)K (u)T 1 (u),
(2.10)
such that the boundary transfer matrix is given by

(u) = T r0 K+ (u)U (u),
(2.11)
where T 1 is the inverse of the monodromy T and K are the matrices defining the
boundaries. The requirement that they obey the RE [26,28]
1
R12 (u, v) K (u)R21 (v, u) K (v)

2
= K (v)R12 (u, v) K (u)R21 (v, u),

1
(2.12)
t1 t2
12 (u, v) K t+2 (v)
R21
(v, u) K t+1 (u)R
2
21 (v, u) K t+1 (u)R t1 t2 (u, v),

= K t+2 (v)R
12
(2.13)
together with the YangBaxter algebra and the following properties

R12 (u, v)R21 (v, u) = 1,
t
1 (v, u)R t2 (u, v) = 1,
R
21
12
t2 (u, v)R t1 (v, u) = 1.
R
12
21
(2.14)
(2.15)
(2.16)
assure that the transfer matrix commutes for different spectral parameters, proving the
integrability of the model.
Therefore, the transfer matrix (2.11) may be considered as the generating function
of infinitely many integrals of motion for the system. We emphasize that there is no
isomorphism between the matrices K+ (u) and K (u), due to the absence of the crossing
symmetry for the R-matrix. Therefore, we have to solve two REs separately in order to fix
the boundaries. In Appendix A, this calculation is presented in detail using the variableseparation prescription. We find three different classes of boundary K matrices consistent
466
with the integrability of the Hamiltonian (2.1). Let us first list those related with K -matrix
A (u) 0
0
0
0
0
0
0
0 B (u) 0
0
0
0
0
0
0
0
0
0
0
0
0 C (u)
1
0
0
0
0
0 D (u) 0
0
(m)
K (u) =
,
0
0
0
0
0
E (u) 0
0
0
0
0
0
0
0
0 F (u)
0
0
0
0
0
0
0 G (u)
0
0
0
0
0
0
0
H (u)
(2.17)
where for m = 1:

A (u) = (c + u) e2 c + u e4 c + u ,

B (u) = (c u) e2 c + u e4 c + u ,

C (u) = (c u) e2 c + u e4 c + u ,

D (u) = (c u) e2 c u e4 c + u ,

E (u) = (c u) e2 c + u e4 c + u ,

F (u) = (c u) e2 c u e4 c + u ,

G (u) = (c u) e2 c u e4 c + u ,

H (u) = (c u) e2 c u e4 c u ,
=
1
3
e6 c
for m = 2:

A (u) = E (u) = (c + u) c + e2 u ,

B (u) = C (u) = F (u) = G (u) = (c + u) c e2 u ,

D (u) = H (u) = (c u) c e2 u ,
1
,
2
c
for m = 3:
A (u) = C (u) = E (u) = G (u) = (c + u),
B (u) = D (u) = F (u) = H (u) = (c u),
1
.
=
c
However, it is much more tedious to find the boundary K-matrix K+ (u), since not
12 (u, v) is more
only the corresponding RE is more involved but also the new object R
467
complicated. We list the final result here and some details can be found in Appendix A,
A+ (u) 0
0
0
0
0
0
0
0 B+ (u) 0
0
0
0
0
0
0
0
0
0
0
0
0 C+ (u)
0
0
0
0
0 D+ (u) 0
0
(l)
K+ (u) =
, (2.18)
0
0
0
0
0
0
E+ (u) 0
0
0
0
0
0
0
0 F+ (u)
0
0
0
0
0
0
0 G+ (u)
0
0
0
0
0
0
0
H+ (u)
for l = 1:

A+ (u) = e6 c+ u 1 e4 c+ u 1 e2 c+ u 1 ,

B+ (u) = e4 e2 c+ u + 1 e4 c+ u 1 e2 c+ u 1 ,

C+ (u) = e2 e2 c+ u + 1 e4 c+ u 1 e2 c+ u 1 ,

D+ (u) = e6 e2 c+ u + 1 (c+ u + 1) e2 c+ u 1 ,

E+ (u) = e2 c+ u + 1 e4 c+ u 1 e2 c+ u 1 ,

F+ (u) = e4 (c+ u + 1) e2 c+ u + 1 e2 c+ u 1 ,

G+ (u) = e2 (c+ u + 1) e2 c+ u + 1 e2 c+ u 1 ,

H+ (u) = e4 (c+ u + e2 )(c+ u + 1) e2 c+ u + 1 .
for l = 2:

A+ (u) = B+ (u) = e6 c+ u 1 e4 c+ u 1 ,

C+ (u) = D+ (u) = e2 e2 c+ u + 1 e4 c+ u 1 ,

E+ (u) = F+ (u) = e2 c+ u + 1 e4 c+ u 1 ,

G+ (u) = H+ (u) = e2 (c+ u + 1) e2 c+ u + 1 ,
for l = 3:

A+ (u) = B+ (u) = e2 e4 c+ u 1 ,

C+ (u) = D+ (u) = e4 c+ u 1 ,
E+ (u) = F+ (u) = e2 (c+ u + 1),
G+ (u) = H+ (u) = (c+ u + 1).
The above explicit formulae for K (u), derived by solving the two REs directly, clearly
show that no automorphism between K+ (u) and K (u) exists and K+ (u) can not be
obtained from K (u). These three classes of boundary K -matrices provide nine possible
(m)

(l)
choices of BC, according to the combination of the boundary pairs K (u), K+ (u) ,
(m)
(l)
m, l = 1, 2, 3, which originate the pair-boundary terms B1 , BL , respectively.
(1)
(1)
(0) = 0, Tr K (1)
Taking into account the fact that Tr K+
+ (0) = 0, Tr K+ (0) = 0 and
468
(2)
(3)
Tr K+
(0) = 0, Tr K (2)
+ (0) = 0, as well as Tr K+ (0) = 0, we can show that the

(m)
(l)
(u) is related to the
Hamiltonian (2.1) with the boundary terms pairs K (u), K+
transfer matrix (2.11) in the following way
(u) = C1 u3 + C2 (H + const) u4 + ,
for l = 1,
(2.19)
(u) = C3 u + C4 (H + const) u + ,
for l = 2,
(2.20)
(u) = C5 u + C6 (H + const) u + ,
for l = 3.
(2.21)
Above Ci , i = 1, . . . , 6, are some scalar functions of the boundary parameters. The

symbols, either bullet or prime denote the derivative with respect to the spectral
parameter. For l = 1, the Hamiltonian (2.1) is related to the fourth derivative of the
boundary transfer matrix (u) which can be derived by the simplified formula [28]
H
(0)
(1)
8 Tr K+
(0)

1
3
(m)
(1)

,
Tr
K
Hjj +1 + P01 K (0)P01 +
(0)L
(0)P
0L
+
0L
(1)
2
Tr K+
(0)
j =1
(2.22)
1
1
Hj,j +1 = P0,j +1 (0)L0j (0)P0j
(2.23)
(0)P0,j
(0)
+1
=
L1
with m = 1, 2, 3.
(l)
However, to obtain new boundary terms corresponding to the K+ -matrices when
l = 2, 3, we have to develop new formulae regarding to the relations (2.20) and (2.21).
What follows are the general formulae of the Hamiltonian related to Eqs. (2.20) and (2.21)
given by
H
=
t (0)
f
L1
1
(m)
Hjj +1 + P01 K
(0)P01
2
j =1

(2)

1 (2)

Tr K+ (0)L
+
0L (0)P0L + Tr K+ (0)P0L L0L (0)
f
(2)

(2)

+ 6 Tr K+ (0)L0L (0)P0L + 3 Tr K+ (0)L0L (0)P0L

(2)

(2)
(0)P0L L 0L (0) + 3 Tr K+
(0)L0L(0)L 0L (0)
+ 3 Tr K+

(2)

(2)
(0)L0L (0)L 0L (0) + 6 Tr K+
(0)L0L (0)L 0L (0) , (2.24)
+ 3 Tr K+
and
H
=
t (2) (0)
(3)
Tr K+
(0)
L1

j =1
1
(m)
Hjj +1 + P01 K (0)P01
2
1
(3)
Tr K+
(0)
469
(2)

(2)

Tr K+ (0)L0L (0)P0L + Tr K+ (0)P0L L 0L (0)

(2)
(2)
+ 2 Tr K+ (0)L0L (0)L 0L (0) + 2 Tr K+ (0)L0L (0)P0L ,
(2.25)
respectively. Above

d 1

L (u)
= P0L L (0)P0L ,
L0L (0)
du 0L
u=0

2

d
1

= P0L L (0)P0L ,
L 0L (0) 2 L0L (u)
du
u=0

3

d
1

L 0L (0) 3 L0L (u)
= P0L L (0)P0L ,
du
u=0

(2)

(2)
(2)

f = 6 Tr K+ (0)L0L (0)P0L + Tr K+ (0)PL0 L 0L (0) + Tr K+ (0) .
(2.26)
Using above formulae and after lengthy and tough calculation, one can derive the boundary
terms presented by Eqs. (2.3) and (2.4). It was seen that the Hamiltonian of the model can
be derived from expansion of the boundary transfer matrix (2.11) at u = 0 up to the fourth,
third and second orders, respectively. This contrasts to the case of two coupled XY spin
chain where the Hamiltonian with four classes of the boundary terms coming from the
third and second derivatives of the transfer matrix [26]. In the next sections, we shall be
focusing on the solution of the eigenvalue problem of the Hamiltonian (2.1).
3. The Bethe ansatz equations

Having established the quantum integrability of the model, let us now solve it by
using the coordinate space Bethe ansatz method. Following [19], we assume that the
eigenfunction of the Hamiltonian (2.1) takes the form

1 ,...,N x+1 1 . . . x+N N |0,
| =
{(xj ,j )}
1 ,...,N (x1 , . . . , xN )
=
.P AQ1 ,...,QN (kP Q1 , . . . , kP QN ) exp i
N

k Pj x j ,
(3.1)
j =1
where the summation is taken over all permutations and negations of k1 , . . . , kN , and Q is
the permutation of the N particles such that 1 xQ1 xQN L. The symbol .P is
a sign factor 1 and changes its sign under each mutation. Substituting the wavefunction
into the eigenvalue equation H | = E| , one gets
A...,j ,i ,... (. . . , kj , ki , . . .) = Sij (ki , kj )A...,i ,j ,... (. . . , ki , kj , . . .),
Ai ,... (kj , . . .) = s L (kj ; p1i )Ai ,... (kj , . . .),
A...,i (. . . , kj ) = s R (kj ; pLi )A...,i (. . . , kj ),
(3.2)
470
with Sij (ki , kj ) Sij ( 12 (ki kj )) being the two-particle scattering matrix,
sin k
,
sin(k i)
sinh
S (k) = ieisgn()k
, = ,
sin(k i)
S
(k) = 1,
S (k) =
(3.3)
and s L (kj ; p1i ) and s R (kj ; pLi ) the boundary scattering matrices,
s L (kj ; p1i ) =
1 p1i eikj
,
1 p1i eikj
s R (kj ; pLi ) =
1 pLi eikj 2ikj (L+1)

e
, (3.4)
1 pLi eikj
where i = 1, 2, 3 and p1i and pLi are given by the following formulae, corresponding
(m) (l)
to the nine cases with respect to the boundary pair B1 , BL , respectively,
Case (i):
Case (ii):
1
,
c exp(4)
1
;
pL1 = pL2 = pL3 pL =
c+ exp(2)
p11 = p12 = p13 p1 =
p11 = 0,
p12 = p13 p1+ =
pL1 = pL2 pL+ =

Case (iii):
p11 = p12 = 0,
pL1 =
Case (iv):
1
,
c+
Case (v):
Case (vi):
p11 = 0,
pL1 =
Case (vii):
1
,
c+
1
,
c+
p11 = 0,
(3.6)
1
,
c
(3.7)
1
,
c exp(4)
1
,
c+ exp(2)
p11 = p12 = p13 p1 =

pL1 =
pL3 = 0;
pL2 = pL3 = 0;
p11 = p12 = p13 p1 =

pL1 = pL2 =
1
,
c
1
,
c+ exp(2)
p13 =
(3.5)
pL3 = 0;
(3.8)
1
,
c exp(4)
pL2 = pL3 = 0;
p12 = p13 =
1
,
c
pL2 = pL3 = 0;
p12 = p13 =
(3.9)
(3.10)
1
,
c
1
;
c+ exp(2)
(3.11)
Case (viii): p11 = p12 = 0,
Case (ix):
p13 =
1
,
c
1
;
c+ exp(2)
p11 = p12 = 0,
1
,
c
pL1 = pL2 =
p13 =
471
1
,
c+ exp(2)
(3.12)
pL3 = 0.
(3.13)
As is seen above, the two-particle S-matrix (3.4) is nothing but the R matrix of the A2
XXZ Heisenberg model (in the homogeneous gauge) and thus satisfies the quantum Yang
Baxter equation, and the boundary scattering matrices s L and s R obey the corresponding
reflection equations (for details, see, [29]). This is seen as follows. One introduces the
notation
s(k; p) =
1 peik
.
1 peik
(3.14)
Then the boundary scattering matrices s L (kj ; p1i ), s R (kj ; pLi ) can be written as,
corresponding to the nine cases, respectively,
Case (i):
s L (kj ; p1i ) = s(kj ; p1 )I,
Case (ii):
s R (kj ; pLi ) = eikj 2(L+1)s(kj ; pL )I ;
1
0
0 eikj sin(i +kj /2)
L
s (kj ; p1 ) = s(kj ; p1+ )
sin(i kj /2)
0
0
0
s (kj ; pLi ) = e
ikj 2(L+1)
sin(i +k /2)
1 0
s(kj ; pL1 ) 0 1
sin(i +k /2)
0 0
Case (iii):
0
1
0
0
sin(i +k /2)
eikj sin(i kjj /2)
0
Case (iv):
eikj sin(i++ kjj /2)
1
s L (kj ; p1i ) = s(kj ; p11 ) 0
s R (kj ; pLi ) = eikj 2(L+1)s(kj ; pL1 )
1
0
0 eikj sin(i+ +kj /2)
sin(i+ kj /2)
s L (kj ; p1i ) = s(kj ; p1 )I,
eikj sin(i kjj /2)
0
;
(3.16)
0
(3.15)
(3.17)
sin(i +k /2)
1 0
s R (kj ; pLi ) = eikj 2(L+1)s(kj ; pL1 ) 0 1

0 0
sin(i +k /2)
(3.18)
472
Case (v):
s L (kj ; p1i ) = s(kj ; p1 )I,

1
0
sin(i+ kj /2)
0
Case (vi):
1
0
L
s (kj ; p1i ) = s(kj ; p1+ )
Case (vii):
eikj sin(i kjj /2)
sin(i +k /2)
(3.20)
eikj sin(i kjj /2)
eikj sin(i kjj /2)
0
sin(i +k /2)
1 0
L
Case (viii): s (kj ; p1i ) = s(kj ; p11 ) 0 1

0
1 0
L
s (kj ; p1i ) = s(kj ; p11 ) 0 1

0
sin(i +k /2)
Case (ix):
sin(i +k /2)
(3.19)
eikj sin(i kjj /2)
s L (kj ; p1i ) = s(kj ; p1+ ) 0
;
0
0
1
0
sin(i+ kj /2)
sin(i +k /2)
sin(i +k /2)
(3.21)
0
sin(i +k /2)
eikj sin(i kjj /2)
(3.22)
0
0
sin(i +k /2)
eikj sin(i kjj /2)
1 0
R
ikj 2(L+1)
s(kj ; pL1 ) 0 1
s (kj ; pLi ) = e
0 0
0
0
(3.23)
sin(i +k /2)
Here I stands for 3 3 identity matrix and p1+ , pL+ are the ones given in (3.6); ,
are parameters defined by
e2 = c ,
e2+ = c+ e2 ,
e2 = c .
(3.24)
We immediately see that (3.15) are the trivial solutions of the reflection equations,
whereas (3.16) and (3.17) are the diagonal solutions for the A2 XXZ model R-matrix.
The boundary scattering matrices given in (3.15), (3.16) and (3.17) also constitute nine
classes of possible choices of boundary conditions for the spin degrees of freedom of the
model.
473
Then, the diagonalization of Hamiltonian (2.1) reduces to solving the following matrix
eigenvalue equation
Tj t = t,
j = 1, . . . , N,
(3.25)
where t denotes an eigenvector on the space of the spin variables and Tj takes the form
Tj = Sj (kj )s L (kj ; p1j )Rj (kj )Rj+ (kj )s R (kj ; pLj )Sj+ (kj )
(3.26)
with
Sj+ (kj ) = Sj,N (kj , kN ) . . . Sj,j +1 (kj , kj +1 ),
Sj (kj ) = Sj,j 1 (kj , kj 1 ) . . . Sj,1 (kj , k1 ),
Rj (kj ) = S1,j (k1 , kj ) . . . Sj 1,j (kj 1 , kj ),
Rj+ (kj ) = Sj +1,j (kj +1 , kj ) . . . SN,j (kN , kj ).
(3.27)
This problem may be solved using the algebraic Bethe ansatz method. The Bethe ansatz
equations are
ikj 2(L+1)
M1

sin
1
2 (kj
F (kj ; p11 , pL1 ) =
1
=1 sin 2 (kj
(1)
) +
(1)
)
i
2
i
2
sin
sin
1
(1)
2 (kj
+ ) +
2 (kj
+ (1)
)
1

i
sin 12 ((1)
+ k + 2

(1)
k )
sin 12 ( + k ) i
2
1 (1)
1 (1)
(1)
(1)
M
1 sin 2 ( ) + i sin 2 ( + ) + i
(1)
= G ; , +

1 (1)
(1)
(1)
sin 12 ((1)
) i sin 2 ( + ) i
=1
N

sin
1
(1)
2 (
1 (1)
=1 sin 2 (
k ) +
i
2
i
2
i
2
i
2
=
M2

sin
1
(1)
2 (
1 (1)
=1 sin 2 (
=1
=
1
(2)
2 (

(2)
sin 12 (
M2 sin

M1

sin
1
(2)
)
(2)
) +
(2)
i
2
i
2
(2)
2 (
1 (2)
=1 sin 2 (
(1)
) +
(1)
sin
1
(1)
(2)
2 ( + )
(1)
(2)
2 ( + ) +
1
1
(2)
2 (

(2)
sin 12 (
) + i sin
(2)
) i
sin
i
2
i
2
sin
sin
1
(2)
+ ) + i
(2)
+ ) i
(2)
2 (
(2)
2 (
1
(1)
+ ) +
(1)
+ )
where
F (kj ; p1+ , pL+ ) = s(kj ; p1+ )s(kj ; pL+ )
(for all cases),
i
2
i
2

= K (2)
; , +
i
2
i
2
(3.28)
474
1,
case (i),
(1) /2+/2)
(1)
sin(i
i
ei ,
case (ii),
sin(i +i(1) /2+/2)
sin(i+ i(1) /2+/2) i(1)
,
case (iii),
sin(i+ +i(1) /2+/2) e
1,
case (iv),
sin(i+ i(1) /2+/2) i(1)

(1)
case (v),
G ; , + = sin(i+ +i(1) /2+/2) e ,
(1)
(1)
sin(i
sin(i i /2+/2)
+ i /2+/2)
, case (vi),
(1)
(1)
sin(i +i /2+/2) sin(i+ +i /2+/2)
(1)
(1)
sin(i i(1) /2+/2) ei ,

case (vii),
sin(i +i /2+/2)
1,
case (viii),
1,
case (ix),
1,
case (i),
sin(i+ i(2) /2+) i(2)
sin(i +i(2) /2+) e

,
case (ii),
sin(i i(2) /2+) (2)
i ,
case (iii),
sin(i +i(2) /2+) e
(2)
(2)
sin(i+ i /2+) ei ,
case (iv),
(2)
sin(i+ +i(2) /2+)
K ; , + = 1,
case (v),
1,
case (vi),
1,
case (vii),
(2) /2+)
(2)
sin(i
i
case (viii),
sin(i +i(2) /2+) e ,
sin(i i(2) /2+) sin(i+ i(2) /2+) , case (ix).

sin(i +i(2) /2+) sin(i +i(2) /2+)
The energy eigenvalue E of the model is given by E = 2

unimportant additive constant).
!N
j =1 cos kj
(3.29)
(modular an
4. Conclusion
In this paper, we have studied integrable open-boundary conditions for the three coupled
XY spin chain model. The quantum integrability of the boundary system has been
established by the fact that the corresponding Hamiltonian may be embedded into a
one-parameter family of commuting transfer matrices. A desirable way to handle open
boundary conditions for the models associated with a general class of quantum R-matrices
(with or without crossing-unitarity) has been developed. Moreover, the Bethe ansatz
equations are derived by means of the coordinate Bethe ansatz approach. This provides
a basis for computing the finite size corrections to the low-lying energies in the system,
which in turn could be used together with the boundary conformal field theory technique
to study the critical properties of the boundaries.
Lastly, it is interesting to formulate a graded version of the quantum YangBaxter
algebra and reflection equation algebra for the fermionic Bariev model. The algebraic
Bethe ansatz for the three coupled XY model with both periodic and open boundary
475
conditions would be very significant in understanding of the symmetry structure of the

model. Those will be addressed in a future publication.
Acknowledgements
A.F., I.R. and X.W.G. thank CNPq (Conselho Nacional de Desenvolvimento Cientfico
e Tecnolgico) and FAPERGS (Fundao de Amparo Pesquisa do Estado do Rio Grande
do Sul) for financial support, I.R. also thanks PRONEX. M.D.G. and H.Q.Z. acknowledges
the support from the Australian Research Council. We thank R. Mckenzie and Jon Links
for helpful comments on the manuscript.
Appendix A. Derivation of the boundary K -matrices

The R-matrix is a 64 64 matrix with 216 nonzero elements which are given in [27].
We now are looking for diagonal solutions K (u) of the REs. We parametrize K (u) as
z1 (u)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
z2 (u)
0
0
z 3(u)
0
0
0
0
0
0
0
0
0
0
0
z4 (u)
0
K (u) =
(A.1)
.
0
0
0
0
0
0
0
z5 (u)
0
0
0
0
0
0
0
z6 (u)
0
0
0
0
0
0
0
z7 (u)
0
0
0
0
0
0
0
z8 (u)
Using some simpler functional equations from the first RE wich are given by (A.1)(A.8)
in the Appendix A of the paper [28], one can present an ansatz
z2 (u)
z1 (u)
z5 (u)
z1 (u)
z6 (u)
z2 (u)
z8 (u)
z4 (u)
c1 c8 u
,
c1 + c8 u
c3 c10 u
=
,
c3 + c10 u
c5 c12 u
=
,
c5 + c12 u
c7 c14 u
=
,
c7 + c14 u
=
z3 (u) c2 c9 u
=
,
z1 (u) c2 + c9 u
z4 (u) c4 c11 u
=
,
z2 (u) c4 + c11 u
z7 (u) c6 c13 u
=
,
z3 (u) c6 + c13 u
(A.2)
with minimal coefficients ci to be determined. Running the RE again with above ansatz, it
can be found that only one coefficient is free and three classes of boundary K -matrices
can be immediately chosen as the forms presented in (2.17).
To solve the second RE is rather cumbersome and sophisticated. For clarifying the
functional equations arising from the second RE, we denote the Boltzmann weights
associated with the R-matrix as w1 (u, v), . . . , w58 (u, v) in accordance with the orders
listed in the paper [27]. The convenient notations z = z+ (v) and z = z+ (u) will be implied
476
hereafter. Similarly, after substituting K+ -matrix (A.1) into the RE (2.13), we may pick up
some simpler functional equations such as
z 2
z 1
z 3
z 1
z 5
z 1
z 8
z 6
z 6
z 2
z 4
z 2
z 7
z 3
=
=
=
=
=
=
=
w2 (v, u)1 (v, u)z2 + w3 (v, u)2 (v, u)z1

,
w3 (u, v)2 (u, v)e4 z 2 + w2 (u, v)1 (u, v)z1
w2 (v, u)1 (v, u)e2 z 3 + w3 (v, u)2 (v, u)z1
,
w3 (u, v)2 (u, v)z3 + w2 (u, v)1 (u, v)e2 z 1
w2 (v, u)1 (v, u)z5 + w3 (v, u)2 (v, u)e4 z 1
,
w3 (u, v)2 (u, v)z5 + w2 (u, v)1 (u, v)z1
w47 (u, v)3 (u, v)e2 z 6 + w46 (u, v)4 (u, v)z8
,
w46 (v, u)4 (v, u)z6 + w47 (v, u)3 (v, u)e2 z 8
w19 (u, v)6 (u, v)e2 z 2 + w20 (u, v)5 (u, v)e4 z 6
,
w19 (v, u)6 (v, u)e2 z 6 + w20 (v, u)5 (v, u)z2
w20 (u, v)5 (u, v)z2 + w19 (u, v)7 (u, v)z4
,
w19 (v, u)7 (v, u)z2 + w20 (v, u)5 (v, u)z4
w19 (u, v)7 (u, v)e2 z 7 + w20 (u, v)5 (u, v)z3
.
w19 (v, u)7 (v, u)e2 z 3 + w20 (v, u)5 (v, u)e4 z 7
(A.3)
(A.4)
(A.5)
(A.6)
(A.7)
(A.8)
(A.9)
t1 t2
t1 t2
(u, v), R12
(u, v),
Before further going on, we first need to work out the matrices R21
12 (u, v) and R
21 (u, v) according to their definitions in Section 2. For our convenience,
R
12 involving in above equations as
we prefer to present below some entries of R
(1 + e6 uv)(1 + e4 uv)2 (1 + e2 uv)(1 + uv)

,
(A.10)
(u v)2 (e2 u v)(e2 v u)(e4 u v)e4
"
(1 + e6 uv)(1 + e4 uv)2 (1 + e2 uv)(1 + uv) (1 + e4 u2 )(1 + e4 v 2 )
2 (u, v) =
,
(u v)2 (e2 u v)(e2 v u)(e4 u v)(e4 v u)e2
(A.11)
"
4
2
2
2
2
2
(1 + e uv)(1 + e uv)(uv + e )(1 + uv) (1 + u )(1 + v )
3 (u, v) =
,
(u v)2 (e2 u v)(e2 v u)(e4 u v)(e4 v u)
(A.12)
(1 + e4 uv)(1 + e2 uv)(uv + e2 )(1 + uv)2
4 (u, v) =
(A.13)
,
(u v)2 (e2 u v)(e2 v u)(e4 v u)
"
(1 + e4 uv)2 (1 + e2 uv)(1 + uv)2 (1 + e2 u2 )(1 + e2 v 2 )
,
5 (u, v) =
(A.14)
(u v)2 (e2 u v)(e2 v u)(e4 u v)(e4 v u)
(1 + e4 uv)2 (1 + e2 uv)(1 + uv)2
6 (u, v) =
(A.15)
,
(u v)2 (e2 u v)(e2 v u)(e4 u v)e
(1 + e4 uv)2 (1 + e2 uv)(1 + uv)2
7 (u, v) =
(A.16)
.
(u v)2 (e2 u v)(e2 v u)(e4 v u)e
1 (u, v) =
Then by analyzing the structure of the Eqs. (A.3)(A.9), we can get the following relations:
z2+ (u) c1 u + e4 c8
=
,
z1+ (u)
c1 u c8
z3+ (u) c2 u + e2 c9
=
,
z1+ (u) e2 c2 u c9
c3 u + c10
z5+ (u)
=
,
z1+ (u) e4 c3 u c10
z6+ (u)
c5 u + c12
,
=
z2+ (u) e4 c5 u c12
z8+ (u) c4 u + e2 c14
=
.
z6+ (u) e2 c4 u c14
477
z4+ (u) c6 u + e2 c11

=
,
z2+ (u) e2 c6 u c11
z7+ (u)
c7 u + c13
,
=
z3+ (u) e4 c7 u c13
(A.17)
Running second RE (2.13) with above relations again and again, the solutions (2.18) would
be fixed definitely.
References
[1] J.B. Bednorz, K.A. Mller, Z. Phys. B 64 (1986) 189.
[2] V.E. Korepin, F.H.L. Essler, Exactly Solvable Models of Strongly Correlated Electrons, World
Scientific, Singapore, 1994.
[3] P. Schlottmann, Phys. Rev. Lett. 68 (1992) 1916;
S. Sarkar, J. Phys. A: Math. Gen. 23 (1990) L409;
P.A. Bares, G. Blatter, M. Ogata, Phys. Rev. Lett. 73 (1991) 11340;
I.N. Karnaukhov, Phys. Rev. Lett. 73 (1994) 11340.
[4] F.H.L. Essler, V.E. Korepin, K. Schoutens, Phys. Rev. Lett. 68 (1992) 2960;
F.H.L. Essler, V.E. Korepin, Phys. Rev. B 46 (1992) 9147.
[5] A.J. Bracken, M.D. Gould, J.R. Links, Y.Z. Zhang, Phys. Rev. Lett. 74 (1995) 2768.
[6] E.H. Lieb, F.Y. Wu, Phys. Rev. Lett. 20 (1968) 1445.
[7] P.W. Anderson, Science 235 (1987) 1196;
F.C. Zhang, T.M. Rice, Phys. Rev. B 37 (1988) 3759.
[8] B. Sutherland, Phys. Rev. B 12 (1975) 3795;
P. Schlottmann, Phys. Rev. B 36 (1987) 5177;
P. Wiegmann, Phys. Rev. Lett. 60 (1988) 821;
F.H.L. Essler, V.E. Korepin, Phys. Rev. B 46 (1992) 9147;
A. Foerster, M. Karowski, Phys. Rev. B 46 (1992) 9234;
A. Foerster, M. Karowski, Nucl. Phys. B 396 (1993) 611.
[9] R.Z. Bariev, J. Phys. A: Math. Gen. 24 (1991) L549;
R.Z. Bariev, J. Phys. A: Math. Gen. 24 (1991) L919.
[10] R.Z. Bariev, A. Klmper, A. Schadschneider, J. Zittartz, J. Phys. A: Math. Gen. 26 (1993) 4663;
R.Z. Bariev, A. Klmper, A. Schadschneider, J. Zittartz, J. Phys. A: Math. Gen. 26 (1993) 1249.
[11] E.K. Sklyanin, L.D. Faddeev, Sov. Phys. Dokl. 23 (1978) 902;
E.K. Sklyanin, J. Sov. Math. 19 (1982) 1546.
[12] P.P. Kulish, E.K. Sklyanin, Lecture Notes in Physics, Vol. 151, Springer-Verlag, Berlin, 1982,
p. 61.
[13] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and
Correlation Function, Cambridge Univ. Press, 1993.
[14] F.C. Alcaraz, M.N. Barber, M.T. Batchelor, R.J. Baxter, G.R.W. Quispel, J. Phys. A: Math.
Gen. 20 (1987) 6397;
I.V. Cherednik, Theor. Math. Phys. 61 (1984) 911.
[15] E.K. Sklyanin, J. Phys. A: Math. Gen. 21 (1988) 2375.
[16] L. Mezincescu, R.I. Nepomechi, J. Phys. A: Math. Gen. 24 (1991) L17;
L. Mezincescu, R.I. Nepomechi, Int. J. Mod. Phys. A 7 (1991) 5231;
L. Mezincescu, R.I. Nepomechi, Int. J. Mod. Phys. A 7 (1992) 5657.
478
[17] H.Q. Zhou, X.Y. Ge, J.R. Links, M.D. Gould, Nucl. Phys. B 546 (1999) 779.
[18] G. Bedrftig, H. Frahm, Physica E 4 (1999) 246;
G. Bedrftig, H. Frahm, J. Phys. A: Math. Gen. 32 (1999) 4585;
G. Bedrftig, H. Frahm, J. Phys. A: Math. Gen. 30 (1997) 4139;
G. Bedrftig , B. Brendel, H. Frahm, R.M. Noack, Phys. Rev. B 58 (1998) 10225.
[19] H. Asakawa, M. Suzuki, Physica A 236 (1997) 376;
H. Asakawa, Physica A 256 (1998) 229.
[20] H. Asakawa, M. Suzuki, J. Phys. A: Math. Gen. 29 (1996) 225;
M. Shiroishi, M. Wadati, J. Phys. Soc. Jpn. 66 (1997) 1.
[21] S. Skorik, A. Kapustin, J. Phys. A: Math. Gen. 29 (1996) 1629;
S. Skorik, H. Saleur, J. Phys. A: Math. Gen. 28 (1995) 6605.
[22] A.A. Zvyagin, Phys. Rev. B 60 (1999) 15266;
A.A. Zvyagin, P. Schlottmann, Phys. Rev. B 56 (1997) 300;
A.A. Zvyagin, H. Johannesson, Phys. Rev. Lett. 81 (1998) 2751.
[23] A. Foerster, M. Karowski, Nucl. Phys. B 408 (1993) 512;
A.J. Bracken, X.Y. Ge, Y.Z. Zhang, H.Q. Zhou, Nucl. Phys. B 516 (1998) 588;
X.-W. Guan, J. Phys. A: Math. Gen. 33 (2000) 5391.
[24] H.Q. Zhou, Phys. Rev. B 54 (1996) 41;
X.-W. Guan, M.-S. Wang, S.-D. Yang, Nucl. Phys. B 485 (1997) 685;
M. Shiroishi, M. Wadati, J. Phys. Soc. Jpn. 66 (1997) 2288;
T. Deguchi, R. Yue, K. Kusakabe, J. Phys. A: Math. Gen. 31 (1998) 7315.
[25] Z.-N. Hu, F.-C. Pu, Nucl. Phys. B 546 (1999) 691;
H. Fan, M. Wadati, X.-M. Wang, Phys. Rev. B 61 (2000) 3450;
A. Lima-Santos, Nucl. Phys. B 558 (1999) 637.
[26] H.-Q. Zhou, Phys. Rev. B 53 (1996) 5098;
A. Foerster, X.-W. Guan, J. Links, I. Roditi, H.-Q. Zhou, Nucl. Phys. B 596 (2001) 525.
[27] H.-Q. Zhou, D.-M. Tong, Phys. Lett. A 232 (1997) 377.
[28] A.J. Bracken, X.-Y. Ge, Y.-Z. Zhang, H.-Q. Zhou, Nucl. Phys. B 516 (1998) 603.
[29] H.J. de Vega, A. Gonzlez-Ruiz, J. Phys. A: Math. Gen. 26 (1993) L519;
Mod. Phys. Lett. A 9 (1994) 2207;
J. Abad, M. Rios, Phys. Lett. B 352 (1995) 92.

A mechanism for ferrimagnetism and

incommensurability in one-dimensional systems
A.M. Tsvelik
Department of Physics, University of Oxford, 1 Keble Road, Oxford OX1 3NP, UK
Abstract
In this paper I discuss a mechanism for ferrimagnetism in 1 + 1 dimensions. The mechanism is
related to a special class of interactions described by operators with non-zero Lorentz spin. Such
operators are present in such problems as the problem of tunneling between Luttinger liquids and
the problem of frustrated spin ladder. Exact solutions are presented for a representative class of
models possessing a continuous isotopic symmetry. It is shown that the interactions (i) dynamically
generate static oscillations with the wave vector dependent on the coupling constant, (ii) give rise
to spontaneous breaking of this symmetry at T = 0 accompanied by generation of the magnetic
moment and appearance of gapless modes with a non-relativistic (ferromagnetic) dispersion E k 2 ,
(iii) generate massive (roton) modes. 2001 Published by Elsevier Science B.V.
PACS: 71.10.Pm; 75.10.Jm
1. Introduction
Though general stability of critical points is determined by scaling dimensions of the
perturbing operators, to determine the ultimate destination of the renormalization group
(RG) flow is much more difficult problem. Availability of non-perturbative methods makes
this task easier in 1 + 1 and two dimensions.
In 1 + 1 or two dimensions critical points possess conformal symmetry with the
Lorentz (O(2)) symmetry being its part. 1 If the relevant perturbation preserves the Lorentz
symmetry, the most it can do is to open a gap in a part of the spectrum. That is exactly what
happens in all known solvable examples (for instance, in the sine-Gordon model).
E-mail address: a.tsvelik@physics.ox.ac.uk (A.M. Tsvelik).
1 Stricktly speaking, the Lorentz symmetry is violated in the models with complex isotopic symmetry, where
different sectors of the spectrum may have different velocities. However, this does not influence the argument
made in the text.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 3 4 - 0
480
A.M. Tsvelik / Nuclear Physics B 612 [FS] (2001) 479491
It may happen, however, that the perturbation violates the Lorentz symmetry. Such
perturbations certainly make sense in condensed matter physics where preservation of
the Lorentz symmetry is not required. There are quite a few models of condensed matter
physics which can be treated as critical models perturbed by relevant Lorentz-symmetrybreaking operators. These operators have non-zero Lorentz spin. Since in 1 + 1 or two
dimensions the Lorentz symmetry at criticality is extended to conformal symmetry, Lorentz
spin is also called conformal spin. At criticality where all two-point correlation functions
z ) is characterized by its scaling dimension
follow power law behaviour, an operator O(z,
d and conformal (or Lorentz) spin S defined as

1 , z 1 )O(z
2 , z 2 ) = Az(d+S)z (dS),
O(z
(1)
12
12
where z, z are holomorphic coordinates defined as z = + ix, z = ix (in 1 + 1
dimensions is Matsubara time). Since only bosonic operators can appear as perturbations,
possible Lorentz spins are integer. Furthermore, since in unitary theories conformal
dimensions are positive (d S 0) and d 2 for relevant operators, this leaves us with
only one choice: S = 1.
The simplest example of S = 1 operator is a conserved charge. At criticality such
perturbation generates incommensurability. As an illustration one can consider a change of
the chemical potential for massless Dirac fermions:

R + R + L+ L .
A = d dx R + R + L+ L
(2)
This perturbation is removed by a simple transformation of the fields
R eix R,
L eix L,
(3)
and thus leads to a shift in the corresponding Fermi momentum and generates static
oscillations with the new wave vector whose magnitude depends on the value of . The
perturbed theory remains critical in the infrared (IR).
The situation becomes less clear when the perturbation is not a constant of motion. Such
non-trivial perturbation corresponds to a non-holomorphic operator, that is, in the notations
of Eq. (1), an operator with d = S.
A well studied example of S = 1 perturbation is the the so-called ZN chiral clock
model of statistical physics (see [1] for review). In fact, this model is not too good
for statistical physics purposes since its Boltzmann weights are not positively defined.
However, the (1 + 1)-dimensional version of this model makes perfect sense and, as was
demonstrated in [2], describes a tunneling between two spinless Luttinger liquids. It was
established that for the chiral clock model the relevant S = 1 perturbations generate
incommensurability.
An immediate generalization of the problem of tunneling includes spin. The tunneling
between two spin-1/2 Luttinger liquids is described by the Hamiltonian

(1)
(2)
H = HLL + HLL + dx (T+ + T ),
(4)
+

+

T = t L1, L2, + H.c. ,
T+ = t R1, R2, + H.c. ,
(5)
481
where R1,2 and L1,2 are the right- and the left-moving fermions on chains 1 and 2; HLL
describe interacting electrons on the individual chains. Due to the intra-chain interaction
the fermionic operators acquire non-trivial scaling dimensions such that

T (, x)T (0, 0)
1
1
(6)
,
( ix)2 ( 2 + x 2 )
where depends on the interaction (here I do not discuss the effects related to difference
between the charge vc and the spin vs velocities of the excitations and set vc = vs = 1).
According to definition (1) operators T carry conformal (Lorentz) spin S = 1.
Another problem much discussed in the literature is the problem of frustrated spin-1/2
two-leg Heisenberg ladder (alias the zig-zag ladder). In the decoupled limit, two S = 1/2
chains represent an SU(2)1 SU(2)1 WZNW theory. Each SU(2)1 WZNW model has its
matrix field gi (x) (i = 1, 2) = i (x) + ini (n1,2 are staggered magnetizations of chains
1 and 2 and 1,2 (x) are the staggered energy density operators). For the frustrated ladder
the interchain interaction contains is dominated by the so-called twist term [3]. The most
general SU(2)-invariant form of this term is
Otwist = An1 x n2 + B1 x 2 + [1 2].
(7)
In the Heisenberg zig-zag ladder the bare value of B is zero but it is generated in the course
of RG (due to the fusion of the A-operator with scalar marginal ones), where the leading
interchain interaction gives the following contribution to the action density [3]: for a single
chain the two-point function of the operators n, decays as

1/2
n(, x)n(0, 0) = (, x)(0, 0) x 2 + 2
(8)
,
which means that
1

2
( ix)2 + ( + ix)2 .
O(, x)O(0, 0)
= x + 2
(9)
According to definition (1), this operator is a sum of two operators with d = 2, S = 1.

Using the procedure suggested in [8], one can reformulate the zig-zag ladder model as a
model of four Majorana fermions with the Lagrangian density given by
L = L0 + Ltwist,

3

1
1
a ( ix )a + a ( + ix ) a ,
L0 =
2
2
a=0

Ltwist = g1 1 2 3 0 + g2 0 1 2 3 + 3 0 1 2 + 2 3 0 1

+ ( )
,
(10)
where g1 , g2 are linear combinations of A, B. In the limit A = B we get the perfect O(4)
symmetry g1 = g2 , with a suitable choice of A, B one can get g2 = 0 which corresponds
to the model I consider later in the paper.
Operators with non-zero conformal spin may also appear in models describing tunneling
between edges of incompressible Fractional Quantum Hall (FQH) states. 2
2 I am grateful to S.L. Sondhi for bringing this possibility to my attention.
482
Since operators with non-zero conformal spin break Lorentz symmetry, it lifts the
restrictions on the low energy spectrum. With the Lorentz symmetry being preserved
one has two choices: either to have a linear spectrum or to have a spectral gap. Without
Lorentz symmetry one can imagine a quadratic gapless spectrum which would be naturally
associated with a ferromagnetic Goldstone mode. Thus, if a model in question has
a continuous isotopic symmetry, the presence of S = 1 perturbation may lead to a
spontaneously breaking of this symmetry at T = 0 with a formation of a net magnetic
moment and the above-mentioned quadratic spectrum.
In all examples discussed above (except the one involving the edge states) the
perturbations with S = 1 appear together. This introduces an additional complification
because a fusion of two operators with opposite Lorentz spins generates under RG a
relevant operator with zero spin. However, in a certain range of parameters of model (5) this
secondary flow does not catch up with the flow of the original coupling constant (for model
(7) this is always the case) [46] (see also [7], where this topic is extensively discussed).
This gives room for an energy scale on which the effect of interaction between the operators
with S = 1 and S = 1 can be neglected. On this energy scale one can drop one operator
and consider a perturbation with one sign of conformal spin, for example S = 1 [9]. In
FQH systems, where excitations are chiral, appearance of a single S = 1 perturbation is
more likely.
2. A solvable model
In the subsequent sections I discuss a series of exactly solvable models which realize the
symmetry breaking scenario outlined above. These models are related to WessZumino
NovikovWitten (WZNW) model. The latter model together with the related lattice models
has been used extensively to address the problems in magnetism (see, for example, [10
13]). Some of these lattice models are rather similar to the zig-zag spin ladder (10).
The particular model I discuss is described by the following action:

A = Wk + ha d2 x : J a (x)a,a (x) : ,
(11)
where Wk is the action of the critical WessZuminoNovikovWitten (WZNW) model
on the SU(2) group with level k 2, a,a are the primary field from the adjoint
representation, Ja are the left KacMoody currents and ha is a constant vector. This
model can be obtained as a low-energy limit of the general (integrable) WZNW model
in a magnetic field (see below). The restriction on the value of k is related to the fact that
for k = 1 field a a does not exist. The suggested solution can be easily generalized for
other symmetry groups and coset models.
To make the discussion self-contained, I recall several basic facts about WZNW models.
A WZNW model describes a matrix field g(x) defined on a group G whose dynamics is
governed by the action

1
W (g) =
d2 x Tr g 1 g + k [g],
16
i
[g] =
24

d

d2 x Tr g 1 gg 1 gg 1 g .
483
(12)
The action contains two parameters the coupling constant and integer number k. The
model has a global GR GL symmetry being invariant under the transformations
g(x) V g(x)U.
(13)
Hence, there are two conserved charges associated with left and right global shifts of the
matrix field g. In the perturbed model (11) the right symmetry is broken, but the left
symmetry is not. As we shall see below, this symmetry is broken spontaneously at zero
temperature.
Model (12) has a stable critical point; the critical value of the coupling constant is
= k 1 . In what follows I shall distinguish between WZNW model as such and critical
WZNW model described by action (12) with = k 1 .
At the critical point WZNW models possesses a higher symmetry: its right (left) currents
become holomorphic (antiholomorphic):
k 1
J =
J = 0,
g g,
2
k
= 0,
J = gg 1 ,
(14)
J
2
where = z , = z , z = + ix, z = ix, and satisfy the KacMoody algebra:
a

ik ab
(x y) + if abc J c (y)(x y).
J (x), J b (y) =
(15)
4
The critical WZNW model can be conveniently described using the Hamiltonian
formalism:
D

2

H=
dx : J a (x)J a (x) : + : Ja (x)Ja (x) : ,
k + cv
a=1
where cv is the quadratic Casimir in the adjoint representation. For the SU(2) group cv = 2.
At criticality any field which is local in g can be represented as a linear combination
of mutually local operators composing a basis in the operator space. This basis of fields
contains the primary fields (j ) and their descendants. The primary fields transform
under GR GL as tensors belonging to irreducible representations of the group; the
descendants are generated from the primary fields by fusion with the KacMoody currents.
Primary fields from higher representations can be generated by fusion of the fields from
the fundamental representation. However, this process terminates after certain number of
fusions depending on the value of k.
To make the discussion more concrete, I concentrate on the SU(2) group. In this case the
primary fields are characterized by the value of spin j = 1/2, 1, 3/2, . . . . The field with the
smallest spin j = 1/2 is the g-matrix itself. A fusion of two g-operators generates a field
with j = 1. For the SU(2) group this representation is isomorphic to the adjoint one. The
maximal spin one can achieve by fusing g-matrices is equal to k/2. Therefore for k = 1
model there is no primary field with j = 1 and model (11) is not well defined.
484
3. Bethe ansatz solution

As I have stated above, model (11) can be considered as a low-energy limit of the general
WZNW model in a filed. For the latter problem I use the Bethe ansatz solution obtained
in [15]. This solution describes not just the critical point, but the entire RG flow towards it
which allows to consider model (11). As was shown in [16], in the vicinity of the critical
point the leading irrelevant operator responsible for this flow is
V = : Ja Ja (x)a,a (x) : .
(16)
In the presence of the right magnetic field HR this operator generates perturbation (11).
Indeed, this field is coupled only to the currents of right chirality Ja . In a finite magnetic
field these currents acquire a finite expectation value

Jb = ab Ha(R)/2
(17)
and therefore in the leading order in H the irrelevant operator (16) is transformed into a
relevant perturbation (11) with

ha = Ha(R)/2 .
(18)
Thermodynamic Bethe ansatz (TBA) equations for SU(2) WZNW model of level k in
magnetic field have the following form [15]:

n (v) = T s ln 1 + en+1 (v)/T 1 + en1 (v)/T mn,0 ev/2 mn,k ev/2 ,
n
n
= HR ,
= HL ,
lim
lim
(19)
n n
n n
where n = , . . . , . HR and HL are right and left magnetic fields corresponding to
the two conserved charges. In accordance with the above discussion, I will keep only the
leading contribution in HR = H and keep HL infinitezimal.
The free energy is given by

F
= mT dv ev/2 n,0 + n,k ev/2 ln 1 + en (v)/T
(20)
L
(L is the systems size) and

s f (v) =
du
f (u)
.
4 cosh[(v u)/2]
We first study these equations at T = 0. In this case all n with negative n are of order of
nH , 0 (v) is positive at v and changes its sign at B ln(m/H ), n with positive
n are also positive except of k (v) which changes its sign from negative to positive at some
value Q > 0. The ground state energy can be expressed in terms of 0 and k which satisfy
the following integral equations:

du D(v u)k (u) = me
Q
v/2
B
+
du K(v u)0 (u),
(21)
B
485

du D(v u)0 (u) = me
v/2
+H +
du K(v u)k (u),
(22)
where the Fourier transforms of the kernels are

tanh
tanh ek||
,
K() =
.
(23)
2 sinh k
2 sinh k
The limits B and Q are determined by the conditions 0 (B) = 0, k (Q) = 0.
To determine the spectrum, we also need to know the distribution densities 0 , k which
satisfy similar equations:
D() =

du D(v u)k (u) =
m v/2
+
e
4
B
du K(v u)0 (u),
(24)
B
du D(v u)0 (u) =
m v/2
e
+
4

du K(v u)k (u).
(25)
These distribution densities determine the conserved charges (right and left magnetic
moments):
1
QL =
2

du k (u),
1
QR =
2
B
du 0 (u).
(26)
Since we are interested only in the leading asymptotics in H , we can neglect k in

Eq. (22) (and, respectively, k in Eq. (25)) and solve these equations as WienerHopf
ones. Eqs. (21), (24) in this case also become WienerHopf equations. The case k = 2
is somewhat special and will be treated separately. For k > 2 the solutions at large Q, B
(small H /m) are given by

0 (v) =
(+)
0 () =
H
,
i( + 2i)G(+) ()G(+) (0)

k (v) =
k(+) () =
d i(v+B) (+)
e
0 (),
2
(27)
d i(vQ) (+)
e
k (),
2
(k 2)meQ/2
,
( 2i)( ik)G(+) ()G() (i/2)
k(+) () = k(+) ()
( ik)
,
2(k 2)
(28)
(29)
486
where
m exp(B/2) = H
G(+) (i/2)
,
G(+) (0)
(30)
(k+2)
exp(Q/2) = k (H /m) (k2) ,

k/(k2)
(+)

(k+2)
G (i/2) (k2)
k tan(/k)
k =
,
2(k + 2)[G(+)(i/k)]2
G(+) (0)
and

G
()
() =
ik + 0
e
ik

1 i
1 + ik
2 +

2k 1 + i
(31)
(32)
with G() () = G(+) () and D() = G() ()G(+) ().

To obtain the modified TBA equations describing model (11) I simply replace 0 in
Eq. (19) by its zero temperature approximate value (27). The result is

n (v) = T s ln 1 + en+1 (v)/T 1 + en1 (v)/T + n,1 s 0 (v) mn,k ev/2 ,
n = 1, . . . ,
n
= HL .
lim
(33)
n n
The free energy is then given by

F
(34)
= mT dv ev/2 ln 1 + ek (v)/T .
L
Let us return to T = 0 solution. From Eqs. (28), (29) we can extract the following
information. First of all, I observe that k approaches zero at two points: at v = Q and
v . The latter point I identify with zero momentum; then point v = Q corresponds to
the wave vector

(k+2)
P = 2 dv k (v) = 2k(+) ( = 0) = mk (H /m) (k2) ,
(35)
Q
where
k =
k
.
G(+) (i/2)G(+) (0)
This incommensurate wave vector scales with H exactly as one expects it to scale (that
is P H 1/(2d)), taking into account that the scaling dimension of the perturbation in
Eq. (11) is d = 1 + 4/(k + 2). The excitation spectrum in the vicinity of v = Q is linear;
the velocity is given by

1
(k 2)
k (v)
VR =
(36)
=
VL ,

2k (Q) v v=Q
k
where VL is the velocity of the left-moving particles (I have been working in the system of
units where VL = 1).
487
The fact that k approaches zero also at v means that there are soft modes at zero
momentum. Since according to Eqs. (28), (29) k (), k () behave as || at small , their
real space asymptotics at v are k k v 2 . Since the momentum is given by

P (v) = 2
dv k (v)
(37)
this means that the spectrum is quadratic. Using this formula and Eq. (29) we find the
dispersion law:
(k 2) P 2
, P P .
k P
The maximum of the energy is reached at P P and is of order of P .
For k = 2 the formulae for the velocity and the dispersion law should be modified:

1
H G(+)(i/2) 2
2
+ ln(4g),
g=
,
Q=
g
m
VR
1
= 2 g H 2,
VL

2
m
m
P =
exp
.
H [G(+) (i/2)]2
H G(+) (i/2)
(P ) =
(38)
(39)
(40)
(41)
Taking T 0 limit in Eqs. (33) I obtain the following expressions for the energies:
sinh n iQ (+)
sinh(k n) iB (+)
e k () +
e
0 (),
sinh k
sinh k
n+k () = en|| k(+) ()eiQ .
n () =
n < k,
(42)
(43)
Let us study asymptotics of these energy functions. In real space we have

1
n+k (v) =

du
Q
n
(+)
(u).
(v u)2 + n2 k
(44)
From this expression it follows that n+k (v) decay as v 2 on both infinities. On the other
hand, n (v) with n < k decay as v 2 only at v Q. At v < Q they have roton-like minima.
For Q v 1 I obtain from Eq. (42):

n (v) k 1 sin(n/k) k+ (i/k)e(vQ)/ k + 0+ (i/k)e(v+B)/ k

1/2
= 2k 1 sin(n/k) k+ (i/k)0+ (i/k)

e(Q+B)/2k cosh (v v0 )/k
(45)
which corresponds to the relativistic spectrum with spectral gaps for k = 2 (n = 1, . . . ,
k 1) given by

2(k 2) cot(/k) 1/2
M0 =
P .
Mn = M0 sin(n/k),
(46)
k(k + 2)
488
For k = 2 the gap is exponential in H 2 . The energy minimum for n with n = 1, . . . , k 1

occurs at

(k 2) tan(/k)
k
v0 = Q
(47)
ln
.
2
4(k + 2)[G(+) (i/k)]2
This means two things. The first one is that the massive modes are centered at
incommensurate wave vector. The second is that in the rapidity space they are very close
to Q.
4. Bethe ansatz derivation of the low energy effective action

Thus we have the following regions in rapidity space where low-energy excitations are
located:
(i)
(ii)
(iii)
(iv)
at v Q there are gapless modes n (n = 1, . . .) with the spectrum v 2 ;

at v Q there are gapless modes n (n = k + 1, . . .) with the spectrum v 2 ;
at v = Q there is a gapless mode k with the spectrum (v Q);
at v Q there are massive modes n (n = 1, . . . , k 1). The massive modes can
be treated as low energy excitations only for k 1 when their masses are much
smaller than P .
Let us choose v as point P = 0. Then in the momentum space the spectrum in

sectors (i), (ii) is chiral quadratic (P ) P 2 (see Figs. 1, 2), the spectrum in sector (iii)
is linear (P ) |P P (Q)| and the spectrum in sector (iv) is massive relativistic (see
Fig. 3).
Let us first consider the case of moderate k when one does not need to consider roton
modes. Then the low energy sector is described by the truly gapless modes (i)(iii). Since
their positions in rapidity space are well separated from each other and the integration
kernels in Eqs. (33) decay exponentially, one can derive separate sets of TBA equations for
each mode. Comparing these equations with TBA equations for known integrable models
one can deduce the effective action for the low energy sector.
Let us derive TBA equations for sector (i). In this sector all energies are positive.
Therefore it is convenient to have TBA equations in such a form where the integral kernels
act on functions ln(1 + en /T ). Such equations would provide a ready T = 0 limit. To get
such form of TBA. The resulting equations read

T ln 1 + en (v)/T Anm T ln 1 + em (v)/T
= A1
k,k An,k k (v) + nHL ,
(0)
n = 1, . . . ,
v Q,
(48)
where k(0)(v) is the solution at T = 0 and is given by Eq. (28). The kernels have the
following standard Fourier transforms:

Anm () = coth || exp |n m||| exp |n + m||| .
(49)
(0)
These equations are valid for temperatures T P the maximum value of k (v). In
this region these equations coincide with the right chiral sector of the spin-(k/2) integrable
Fig. 1. Dispersion n (P ) (n = k + 1, . . .).
Fig. 2. Dispersion n (P ) (n = 1, . . . , k 1).
Fig. 3. Dispersion k (P ).
489
490
ferromagnet. The latter ones can be extracted from TBA equations obtained in [17]. In a
similar fashion I obtain TBA for sector (ii):

T ln 1 + en+k (v)/T Anm T ln 1 + em+k (v)/T
= an k(0) + nHL ,
n = 1, . . . ,
v Q.
(50)
These equations describe the left chiral sector of the spin-1/2 integrable ferromagnet where
the dispersion law is (k) k 2 (k). Therefore, the parity in the ferromagnetic sector is
broken.
Another soft mode in the right chiral sector has the linear spectrum with velocity (36)
and is centered at the the momentum P . Using the standard manipulations with TBA
equations one can find its contribution to the specific heat and establish that this mode
carries central charge 1. Therefore it is described by the non-chiral Gaussian model.
5. Connection to the zig-zag chain at k = 2
For k = 2 model (11) simplifies considerably. In this case the critical WZNW action is
equivalent to the theory of three massless Majorana fermions a , a (a = 1, 2, 3) and the
adjoint operator and the currents are given by [14]
i
J a = abc b c .
(51)
2
Therefore, for k = 2 one can rewrite the perturbed action (11) solely in terms of Majorana
fermions:

1
1
2
A = d x a ( ix )a + a ( + ix ) a + ha 1 2 3 a .
(52)
2
2
ab = a b ,
In this form the model closely resembles the model for the zig-zag ladder (10) (see also
[9]) and at g2 = 0 it even coincides with model (11) exactly solvable by the Bethe ansatz.
Acknowledgements
I am grateful to Ph. Lecheminant for inspiration, to A.A. Nersesyan, D. Maslov and
F.H.L. Essler for illuminating discussions and interest to the work and E. Papa and S. Carr
for help with preparation of the manuscript.
References
[1] M. den Nijs, in: C. Domb, J. Lebowitz (Eds.), Phase Transitions and Critical Phenomena,
Vol. 12, Academic Press, 1988.
[2] P. Fendley, C. Nayak, cond-mat/0008157.
[3] A.A. Nersesyan, A.O. Gogolin, F.H.L. Essler, Phys. Rev. Lett. 81 (1998) 910.
[4] F.V. Kusmartsev, A. Luther, A.A. Nersesyan, JETP Lett. 55 (1992) 692;
F.V. Kusmartsev, A. Luther, A.A. Nersesyan, Phys. Lett. A 176 (1993) 363.
491
[5] V.M. Yakovenko, JETP Lett. 56 (1992) 5101.

[6] D.V. Khveshchenko, T.M. Rice, Phys. Rev. B 50 (1993) 252;
D.V. Khveshchenko, Phys. Rev. B 50 (1993) 380.
[7] For references and discussion see: A.O. Gogolin, A.A. Nersesyan, A.M. Tsvelik, Bosonization
and Strongly Correlated Systems, Cambridge University Press, 1998, Chapters 8, 20.
[8] D.G. Shelton, A.A. Nersesyan, A.M. Tsvelik, Phys. Rev. B (1996).
[9] M. Zarea, M. Fabrizio, A.A. Nersesyan, in preparation.
[10] L.D. Faddeev, N.Yu. Reshetikhin, Ann. Phys. N.Y. 167 (1986) 227.
[11] H.J. de Vega, F. Woynarovich, J. Phys. A 25 (1992) 4499.
[12] P. Schlottmann, Phys. Rev. B 49 (1994) 9202;
A.A. Zvyagin, P. Schlottmann, Phys. Rev. B 52 (1995) 6569.
[13] A.A. Zvyagin, Low Temp. Phys. 26 (2000) 134, see also: cond-mat/0002356.
[14] A.B. Zamolodchikov, V.A. Fateev, Yad. Fiz. 43 (1986) 1043.
[15] A.M. Polyakov, P.B. Wiegmann, Phys. Lett. B 131 (1983) 121;
A.M. Polyakov, P.B. Wiegmann, Phys. Lett. B 141 (1984) 223.
[16] V.G. Knizhnik, A.B. Zamolodchikov, Nucl. Phys. B 247 (1984) 83.
[17] H.M. Babujan, Nucl. Phys. B 215 (1983) 317.

The magnetized electron gas in terms of Hurwitz

zeta functions
Claudio O. Dib, Olivier Espinosa
Departamento de Fsica, Universidad Tcnica Federico Santa Mara, Casilla 110-V, Valparaso, Chile
Abstract
We obtain explicit expressions for thermodynamic quantities of a relativistic degenerate free
electron gas in a magnetic field in terms of Hurwitz zeta functions. The formulation allows
for systematic expansion in all regimes. Three energy scales appear naturally in the degenerate
relativistic gas: the Fermi energy EF , the temperature T and an energy related to the magnetic field or
Landau level spacing, eB/EF . We study the cold and warm scenarios, T eB/EF and eB/EF T ,
respectively. We reproduce the oscillations of the magnetization as a function of the field in the cold
regime and the dilution of them in the warm regime. 2001 Elsevier Science B.V. All rights reserved.
PACS: 05.30.Fk; 71.10.Ca; 02.30.Gp; 97.60.Jd
Keywords: Electron gas; Landau levels; Magnetism; Hurwitz zeta function
1. Introduction
The study of the thermodynamic properties of degenerate relativistic electron gases in
strong magnetic fields, as those found in compact astrophysical objects, was started long
ago [1]. Although in that work the problem was stated in quite general terms, that is,
arbitrary temperature and relativistic electrons described by Diracs equation, approximate
analytical results were obtained only in the limits of low temperatures and non-relativistic
electrons. While the non-relativistic approximation has some validity in white dwarfs, it is
inappropriate for the electron gas existing inside neutron stars. In fact, in a typical neutron
star the electron density is of the order of 102 fm3 , which implies a Fermi kinetic energy
TF EF m of the order of 100 MeV. The temperature, on the other hand, is at most a
few MeV, which makes the ratio T /TF a small parameter and the gas quite degenerate. The
relativistic case was partially considered later on in Refs. [2,3] with regard to the problem
* Corresponding author. Phone: +56(32)654-506. Fax: +56(32)797-656.
E-mail addresses: cdib@fis.utfsm.cl (C.O. Dib), espinosa@fis.utfsm.cl (O. Espinosa).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 6 0 - 1
C.O. Dib, O. Espinosa / Nuclear Physics B 612 [FS] (2001) 492518
493
of self-magnetization, and in Ref. [4] in connection with the magnetic susceptibility. In

all these works only the dominant contributions relevant for each case were kept in the
analysis.
More recently, a renewed interest in the relativistic magnetized electron gas, both at zero
and finite temperature, has appeared from the point of view of quantum field theory [59].
In these studies most of the emphasis is put on the formal aspects of the problem. For
instance, in Ref. [7] an exact analytic expression is obtained for the effective action (also
called the grand potential in statistical mechanics), which corrects an incomplete result
given earlier [5] that had missed the de Haasvan Alphen oscillations that are present at low
temperatures, exactly as in the non-relativistic case. In Ref. [8] the quantum field theory
results of Ref. [7] are recast, for the case of low temperatures, in terms of sums over filled
Landau levels, which is actually the starting point of the quantum statistical mechanical
calculation developed in the earlier works [14].
In this work we revisit the usual elementary statistical mechanical approach, originally
carried out by Landau [10,11] (see also Ref. [12]) for a non-relativistic gas, to study a
highly degenerate relativistic free electron gas, in the presence of a uniform magnetic field.
Starting from a closed analytical expression for the density of states for this system, we
derive explicit, simple analytical expressions for the various quantities of thermodynamic
interest, such as the density and magnetization. The object of central interest will be the
grand potential, from which all relevant thermodynamic quantities can be computed.
It is well known that for a relativistic electron gas the Coulomb interactions among
electrons and between the electrons and any background of positive charge that should
exist in a neutral system are small (order ) corrections to the kinetic energy, so we will
neglect them. We will therefore work consistently to zeroth order in QED corrections. In
particular, this means that we will take the electron gyromagnetic ratio exactly equal to
two, g = 2. Additionally, since we shall be solely concerned with the degenerate regime,
we neglect the contribution of positrons altogether.
We will show that the grand potential at T = 0 has the integral representation
V
0 (EF , B) = 2 (2eB)5/2
4
2
(EF2 m
)/2eB
H1/2 (q)

dq,
m2 + 2eBq
(1)
where m and e are the electron mass and the fundamental charge, respectively (we use
natural units, so that h = c = kB = 1). Consequently, the electron density at T = 0 is

2
E F m2
1
3/2
.
n0 (EF , B) =
(2eB) H1/2
2eB
2 2
(2)
The function H1/2 (q) appearing in both (1) and (2), is a combination of Hurwitz zeta
functions, H1/2 (q) ( 12 , {q}) ( 12 , q + 1) 12 q, and plays a pervading role in

our study of the magnetized free electron gas. Since the pressure is given by P = /V ,
the expressions (1) and (2) furnish a parametric representation of the equation of state at
T = 0, P0 = P0 (EF , B), n0 = n0 (EF , B).
494
The expressions given above correspond to zero temperature, but, as we show later in the
paper, for a system of non-interacting fermions the full finite temperature grand potential
can be obtained from its zero temperature counterpart. For instance, in the degenerate
regime finite temperature corrections are usually obtained systematically from the ground
state quantities in the form of a Taylor series in powers of T /, where is the chemical
potential. Take, for example, (T , ). As is well known, the leading correction to the zerotemperature result generally goes as (T /EF )2 . This will actually be the case, provided the
grand potential at T = 0, 0 (EF ), does not vary greatly when EF is changed by an amount
of order T . However, as we will see, in our case 0 (EF ) has contributions that oscillate
rapidly with EF due to the filling of discrete Landau levels. Therefore the approximation
above is not valid, except for very low temperatures (T eB/EF , as we will see later).
We will compute with our formalism the correct finite temperature behavior of (T , )
and reproduce the previously known result that the oscillations are smoothed out as the
temperature increases.
The novelty of our approach is the use of the Hurwitz zeta function to deal with the
discrete sums over Landau levels that accommodate the electrons in the presence of a
uniform magnetic field. The traditional approach [12] makes use of the so-called Poisson
summation formula which is of limited use if one needs to expand the resulting expressions
in different regimes, such as for small fields. The Hurwitz zeta function approach is
much more powerful in this case, as it leads to closed analytical expressions, which can
eventually be evaluated numerically with ease, and is suitable for expansion in any desired
regime. Although this is not the first appearance of the Hurwitz zeta function in connection
with the system being studied [6,9], this seems to be the first time that its analytic properties
are fully put into use to unravel the thermodynamics of the relativistic magnetized free
electron gas.
The Hurwitz zeta function (z, q) is defined as the analytic extension to the whole
complex z plane of the series
(z, q) =

n=0
1
(n + q)z
(3)
valid for Re z > 1 and q = 0, 1, 2, . . . . The resulting function is analytic everywhere

except at z = 1, where it has a simple pole with unit residue (see Appendix A for details).
In Section 2 we formulate the density of states in its most general form in terms
of Hurwitz zeta functions. In Section 3 we classify the physical regimes we want to
study. The core of the thermodynamics is presented in Section 4 and particular studies
of the magnetization are shown in Section 5. Details of the calculations and mathematical
formulas are given in the appendices.
2. The density of states in terms of Hurwitz functions

The stationary states of a Dirac electron that moves in a uniform magnetic field B (which
we take to point along the z-direction), are specified in terms of four quantum numbers. In
495
the gauge where A(x) = xB y , these quantum numbers are pz (momentum component
along B), py (momentum component along A), a non-negative integer n = 0, 1, 2, . . .
that specifies the Landau level, and an integer = 1 which denotes the spin parallel
or antiparallel to B. The energy levels are independent of the quantum number py and are
given by 1 [13]

E(pz , j ) = pz2 + m2 + (2eB)j ,
(4)
where j is a non-negative integer defined by j = n + ( + 1)/2. The energy levels (4)
are highly degenerate, due to their independence on py and their dependence on n and
only through the single combination j . The number of states g(pz , j ) dpz for given j and
momentum along B between pz and pz + dpz , in a system of electrons confined to a finite
cubic box of volume V , is given by
eB
V dpz ,
(5)
4 2
with gj = 1 for j = 0 and gj = 2 for j = 0. Physically, this degeneracy embodies the
fact that the levels with j = 0 accommodate only electrons with spin pointing down,
whereas all the others can have electrons spinning in either direction. As an immediate
consequence, the levels with j = 0 alone are responsible for the paramagnetic contribution
to the magnetization. See [14] for details. The density of states in energy space is thus
given by
g(pz , j ) dpz = gj
g(E) dE = 2
jE

g |pz |, j dpz ,
(6)
j =0
where the factor of 2 in front of the sum takes into account both possible signs of pz , and
the integer jE corresponds to the highest Landau level j that starts at an energy less than E.
From Eq. (4) it is clear that jE is the integer part of the quantity
E 2 m2
(7)
,
2eB
denoted usually as jE = qE . The degeneracy of the levels [13] can be expressed explicitly
by
qE =

V
1
g(E) =
2eBE
gj
2
4 2
E m2 2eBj
jE
j =0
qE

1
V
.
(2eB)1/2E
gj
2
4
qE j
(8)
j =0
Considering now the degeneracy gj in Eq. (5) and formula (A.3) with z = 1/2 we find

q
E

1
1
(9)
gj
,
= 2 12 , qE qE 12 , qE + 1
2 qE
qE j
j =0
1 For g = 2, i.e., neglecting QED corrections to the electrons magnetic moment.
496
where the r.h.s. of this equation defines the following family of functions that appear
regularly in the thermodynamic expressions for this system:

1
Hz (q) z, {q} (z, q + 1) q z ,
2
(10)
and where {q} = q q is the fractional part of q.

In what follows, we will denote energies in units of the electron mass, = E/m, and the
magnetic force eB in terms of m2 :
b
2eB
.
m2
(11)
The variable b measures the magnetic field strength in units of the natural strength B0 =
1 2 4
c 2.2 1013 Gauss. 2
2 m c /e h
With this notation and the use of Eq. (9), the density of states for energy E has a simple
form:

2
1
m2 1/2
.
g(E) = V 2 b H1/2
(12)
b
2
The total number of states up to a given energy E is the integral over the density of states:
E
G(E) =
g(E ) dE .
(13)
m

(q) (where the
Using Eq. (A.4) of Appendix A, it is easy to show that H1/2 (q) = 2H1/2
prime denotes differentiation with respect to q), so that we can do this integral in a closed
form, obtaining:

2
1
m3 3/2
.
G(E) = V
b H1/2
b
2 2
(14)
Clearly, G(EF )/V is the electron density in the system, where EF denotes the Fermi
energy. We will consistently use dimensionless variables in the expressions:

EF
,
pF F2 1.
F
(15)
m
While F is the Fermi energy, pF is the Fermi momentum of the j = 0 Landau level, in
units of m. The expression pF2 /b gives a measure of the occupation of the Landau levels in
the ground state: its integer part, pF2 /b, equals the last Landau level that is occupied, and
its fractional part {pF2 /b}, or more precisely 1 {pF2 /b}, gives a measure of the distance
to the next empty one.
2 Notice that B is in our case half of the scale used by other authors [1,4] (m2 c3 /eh 4.4 1013 G), because
0
the natural dimensionless variable associated to the magnetic field in this problem is b = B/B0 = 2eB/m2 , as
seen in Eqs. (1), (2).
497
3. Limiting cases and regimes

In the non-relativistic case, the energy scale associated with the magnetic field is the
spacing between Landau levels, eB/m, which is uniform and equal to B , the cyclotron
frequency. For typical values of magnetic field strengths found in pulsars, say B = 1012
Gauss, one finds B 10 keV.
As is well known, a fermion gas becomes degenerate when the occupation number falls
abruptly for states with energy above a given value EF , the Fermi energy. This happens if
the temperature is much smaller than the kinetic range TF EF m (i.e., the energy range
between the lowest orbital and the Fermi energy). For electrons in neutron stars, typical
values might be T < 1 MeV and TF 102 MeV, so that the gas is not only degenerate but
also highly relativistic. In the relativistic regime however, the level spacing is not uniform,
but decreases with energy, so that close to the Fermi surface it is eB/EF .
While the Fermi energy is invariably larger than the temperature in a degenerate gas, the
number of occupied Landau levels may or may not be large. This number is given by the
integer part of (EF2 m2 )/(2eB), in view of Eq. (7) and its preceding paragraph, and so
it is large if 2eB/EF TF . This is indeed the case for typical magnetic fields in neutron
stars (B < 1014 G), where eB/EF < 10 keV.
We can still classify the gas in a uniform magnetic field as cold of warm, depending on
whether the temperature is smaller or larger than the energy spacing between Landau levels
close to the Fermi surface, the latter being eB/EF . Consequently, in a cold degenerate gas,
the following condition is satisfied:
T
eB
EF
and T TF .
(16)
If, in addition, the magnetic field is small so that there is a large number of occupied Landau
levels, the above hierarchy becomes:
T
eB
TF .
EF
(17)
On the other hand, in the warm regime, the temperature is comparable or larger than the
Landau spacing:
eB
< T TF .
EF
(18)
These three cases are the regimes of interest here.
4. The grand potential and Hurwitz functions

The general expression for the grand potential for an ideal Fermi gas in a uniform
background magnetic field B is given by

ln 1 + e(E )/T ,
(T , B, V , ) = T
(19)
498
where is the chemical potential and the sum is over all 1-particle orbitals . The B
dependence is implicit in the orbital energies E and the density of states. Thermodynamic
quantities can be obtained directly from . In particular, M, the magnetization per unit
volume is obtained as
M=
1
.
V B
(20)
In terms of the density of states g(E) the grand potential (per unit volume) is
= T
V

g(E)
ln 1 + e(E)/T dE.
V
(21)
This expression, together with the explicit form (12) for g(E), can be used to study the
thermodynamics at arbitrary temperature T and magnetic field B.
4.1. The T = 0 limit
As we show in Appendix B, for a system of non-interacting fermions any finite
temperature quantity can be obtained from its zero-temperature counterpart. Therefore,
in the rest of this section we will just concentrate on the grand potential density at zero
temperature, (T = 0) 0 , keeping in mind the standard definition for the Fermi energy
EF (T = 0). The T = 0 limit is simply formulated from

E EF for E < EF ,
lim T ln 1 + e(E)/T =
(22)
0
for E > EF ,
T 0
so that
EF
0 (EF , B) =
g(E)
N
E dE EF .
V
V
(23)
The integral in Eq. (23), which is the ground state energy density ( u0 ), can be done
using the explicit representation for the density of states given in Eq. (12). After standard
manipulation and integration by parts we get:
EF
u0
g(E)
E dE
V
m
2
p

2
F /b
pF
H1/2 (q)
m4
3/2
5/2
=
b
2b F H1/2
dq .
4 2
b
1 + bq
(24)
Notice that the first term above is exactly EF N/V , according to Eq. (14), so it cancels
with the second term in Eq. (23), leaving a compact form for the grand potential density
499
0 0 /V :
m4
0 (F , b) = 2 b5/2
4
2
p
F /b
H1/2 (q)
dq,
1 + bq
(25)
where

1
q.
H1/2 (q) = 12 , {q} 12 , q + 1
(26)
2
Formally, this is the main result of this work.
Expression (25) allows us to recover the well known non-relativistic result [12] in a
straightforward fashion: in the non-relativistic limit pF 1, so that the square root in the
integrand of (25) can simply be replaced by unity. Using then the relation H1/2 (q) =
2
3 H3/2 (q) we find
0(n.r.) (F , b) =

m4 5/2
b H3/2 pF2 /b ,
2
6
(27)
where

1
H3/2 (q) 32 , {q} 32 , q + 1 q 3/2 .
(28)
2
The result (27) is exactly the known non-relativistic result, after identifying pF2 = 2n.r. /m,
where n.r. is the standard (non-relativistic) chemical potential. In particular, notice that
H3/2 (q) splits naturally into a monotonic term and an oscillatory term, H3/2 (q) =
(mon)
(osc)
H3/2 (q) + H3/2 (q), as:

1
(mon)
H3/2 (q) = 32 , q + 1 q 3/2,
2

(osc)
H3/2 (q) = 32 , {q} .
(29)
(30)
(osc)
The oscillating character of H3/2 (q) arises from the fractional part {q} in the definition
above. This is the term that accounts for the de Haasvan Alphen oscillations in the
non-relativistic case, and leads to Landaus result at finite temperature, as it is shown in
Section 5.
Likewise, the relativistic grand potential density (25) separates into two terms, 0 =
(mon)
(osc)
+ 0 , where
0
(mon)
(F , b) =
m4
4
2
p
F /b
b5/2
2
0
m4
0(osc) (F , b) = 2 b5/2
4
( 12 , q + 1) +
1 + bq
2
p
F /b
1
2 q
dq,
(31)
( 12 , {q})
dq.
1 + bq
(32)
(osc)
Here we have kept the notation consistent, however one must be aware that 0 , the
integral of an oscillating function, is not purely oscillatory, as we will see. The integral
500
expressions above are exact and finite for any values of the magnetic field and Fermi
momentum.
Although the integrals in Eqs. (31) and (32) cannot be solved in a closed form, one can
nevertheless use the analytic properties of the Hurwitz zeta function to obtain an expansion
of 0 for small b, or similarly for large pF2 /b (i.e., large number of occupied Landau levels).
(mon)
. A nave binomial expansion of the square root in the integrand
Let us first consider 0
will not work because of the diverging behavior of the integral as pF2 /b . A procedure
to obtain the correct expansion for small b consists in first extracting out a sufficient number
of leading terms in the asymptotic expansion of the zeta function for large q, such that the
remainder is integrable in the limit b 0 [three terms will do for (1/2, q)], and then
integrating the subtracted terms explicitly. Based on Eq. (A.13) of Appendix A we write
(z, q) =
1
z
1
q z+1 + q z + q z1 + 3 (z, q),
z1
2
12
(33)
implicitly defining the subtracted zeta function 3 (z, q). Now, the integrals corresponding
to the exhibited terms in (33) along with the term q/2 in (31) can be done exactly for
finite b, leaving only an integral containing the subtracted function 3 (1/2, q):
0(mon) (F , b)

+ b + p2

4
F
1/2
1
1
1
m
F
1 b + b2 ln
b + pF2
= 2
F b1/2
6
2
4 2
1+ b

3/2

1
1
b + pF2
F b3/2 + b cosh1 (F ) pF F
+
3
2
2 /b
p

F
3 ( 12 , q + 1)
b5/2
(34)
dq .
1 + bq
0
The expression above is still exact, but now the integral containing 3 (z, q) can be
expanded in powers up to order b without entering into trouble, because 3 (1/2, q)
vanishes as q 5/2 for q (see Eq. (A.15) of Appendix A). It suffices to take the upper
limit to infinity and expand the denominator:
2
p
F /b

3 ( 12 , q + 1)
1
2
b0
dq 32 +
3
60
1 + bq

2
1
+ b 52 +
+ O b3/2 .
15
1260
(35)
This integral thus loses the dependence on the chemical potential and becomes a spurious
contribution that cancels in 0 when we put all pieces together (see below). A word of
caution is due: to go to higher orders in b one cannot continue this expansion in the same
fashion. Instead, one should follow the same procedure from the beginning, but extracting
more explicit terms from the asymptotic expansion (33), integrating them directly, and then
501
expanding the n-residual integral containing n (1/2, q) up to the last convergent term,
that goes as b2n5 .
(osc)
Now, let us turn to the oscillatory contribution 0 , which can be written in the
following form (see Appendix C):
m4
0(osc) = 2 b2
4
{pF2 /b}

0
1
+
( 12 , q)
dq
q + 1/b + pF2 /b

1 1

1
2
2 , q 2 , q + 1/b 2 , q + 1/b + pF /b dq ,
(36)
whose expansion for small b is straightforward (however tedious):

m4 2 5/2 1 3 2
b0
b
2 , pF /b 32
0(osc) 2
F
4 3

2
1
+ b7/2 3 52 , pF2 /b 52
+ O b9/2 .
15
F
(37)
Finally, we can put all the terms together and find the expansion of the grand potential
density 0 for small magnetic field (see Appendix D):

m4 1
1
1
b2
cosh1 (F ) + F pF3 F pF +
cosh1 (F )
0 = 2
4 2
3
2
12

4
2 b 5/2 3 2
2 b 7/2 5 2
+
2 , pF /b +
2 , pF /b + O b .
3 F
15 F3
(38)
We note that expansion (38) is in agreement, apart from an overall sign, with the small
field result for the effective action obtained in Ref. [9] (Eq. (74) in Ref. [9]), which in our
notation reads

bm4
F pF ln(F + pF )
(S d=3 T =0;=0 =
2
8
2
4

(b/F2 )l+1/2
b m
l 32 , pF2 /b l 32 , pF2 /b .
+
2
2
(2l + 1)(2l + 3)
(39)
l=0
Although the non-analytic oscillatory terms are immediate to compare, it is quite nontrivial to obtain the analytic terms from (39), since each term of the type bl+1/2(l
3
2
2 , pF /b) in the sum in (39) contributes to all orders in b as b 0, in view of the
asymptotic expansion (A.14) for the Hurwitz zeta function. Using the first three terms of
this asymptotic expansion, one can perform each of the resulting infinite sums in a closed
way to indeed obtain the terms shown in (38), with a vanishing coefficient for the term
linear in b. In retrospect, we realize that the result (39) can be formally obtained directly
from the closed expression (25) by doing a binomial expansion of the square root and then
integrating term by term. However, the latter expansion is valid over the whole integration
502
Fig. 1. The functions (z, {q}), for z = 1/2, 3/2, 5/2 (large, medium and small amplitude,
respectively).
range only if pF2 < 1, i.e., basically in the non-relativistic limit. From a numerical point of
view it is clearly advantageous to use expansion (38) instead of (39), since then one can
compute to any desired precision at small b by keeping only a finite number of terms in the
expansion.
All thermodynamic quantities at T = 0 can be obtained from 0 as derivatives. One
should notice that, since (z, 0) = (z, 1) for z < 0, the function (z, {q}) is continuous in
q for all z < 0 and, in view of Eq. (A.4), has continuous derivative for all z < 1 (Fig. 1).
For example, it is straightforward to derive the expansion for the density at T = 0 using
the thermodynamic identity
1 0
.
(40)
m F
We thus find the the density at T = 0, consistent with Eq. (14), and the corresponding
expansion for small b:
n0 =

m3 3/2
b H1/2 pF2 /b
2
2

1 2
4
m3 2 3
b2
3/2
p + b 2 , pF /b +
+ O b .
24pF
2 2 3 F
n0 (F , b) =
(41)
In the b = 0 limit we recover the free electron gas result, n0 = m3 pF3 /3 2 . As seen in
Eq. (41), the leading correction for finite b and T = 0 is a term that oscillates with pF2 /b,
corresponding to the de Haasvan Alphen effect for the magnetization in metals.
Fig. 2 shows n0 vs. F for two values of the field. The step-like behavior is due to the
filling of consecutive Landau levels as F increases. The density of states in a given Landau
level [see Eq. (5)] goes as b dpz b dE/pz , for pz starting from zero. The higher density
at the bottom of each level causes the step in n0 . On the other hand, as b decreases the steps
503
Fig. 2. The particle density at T = 0 as a function of the Fermi energy, for fixed magnetic field
b = 0.1 (smooth curve) and b = 0.5 (bumpy curve).
Fig. 3. The particle density at T = 0 as a function of the magnetic field, for fixed Fermi energy
F = 1.3 (lower curve) and F = 1.5 (upper curve).
gradually disappear until the smooth b = 0 limit is reached. Alternatively, Fig. 3 shows n0
as a function of the field b, for fixed Fermi energy. Here n0 oscillates as b changes. Imagine
we start from a large value of b. As b decreases, the Landau levels move down in energy,
crossing one by one the threshold defined by F . Since the levels are denser at the bottom,
n0 first grows as each level becomes accessible, but then decreases again, because of the
b factor in the density of states. The oscillation amplitude becomes smaller and smaller as
the Landau levels get closer and closer, until we reach the b = 0 limit.
We note that the expansion in Eq. (41) is actually a large pF2 /b expansion, appropriate
when a large number of Landau levels are occupied at T = 0. So it will remain valid even
for large values of the magnetic field, say b 100, provided the ratio pF2 /b stays much
504
larger than one. This will be the case for electron densities in typical neutron stars, for
which pF2 104 .
We can also obtain the expansion for u0 , the energy density at T = 0 whose exact
integral form is given in Eq. (24) using the expansions in Eqs. (38) and (41) and the
thermodynamic relation u0 = 0 + mF n0 :

m4
1
1
u0 =
F pF3 + F pF cosh1 (F ) + 2b3/2F 12 , pF2 /b
2
2
2
4
2
2
b F
b
2 b5/2 3 2
1
cosh
+
(
)
2 , pF /b
F
12 pF2 12
3 F

4
2 b 7/2 5 2
.
2 , pF /b + O b
(42)
15 F3
Unlike Eq. (41) for the particle density, the expansions for the energy density given
above and for the grand potential given in Eq. (38) are valid for small b only. More useful
in the case of neutron star conditions is an expansion for large pF2 /b, that is, many Landau
levels occupied, but regardless of b. In that case, for instance, the small b expansion of
the denominator in Eq. (35) is not valid. The correct expansion of 0 for large pF2 /b is the
following, where some pF -independent integrals are left to be done numerically:

+ b + p2
4
F
1
1
m
F
1 b + b 2 ln
0(mon) (F , b) = 2
6
4 2
1+ b
1/2
1
3/2

1
F b1/2 +
F b3/2
b + pF2
b + pF2
2
3

1
4

(
,
q
+
1)
b
1
1
3
+ , (43)
+ b cosh1 (F ) pF F b5/2
dq +
2
2
3840 F4
1 + bq
(osc)
0

m4 2 b5/2 3 2
2 b7/2 5 2
= 2
2 , pF /b +
2 , pF /b +
4 3 F
15 F3

1
1 1

2
2 , q 2 , q + 1/b dq .
+b
(44)
4.2. The finite temperature case

As shown in Appendix B, the grand potential at finite temperature can be obtained from
its expression at T = 0, 0 (F ), as:

(T , ) =
1
T
0 ( + T x)
ex
dx,
(ex + 1)2
(45)
505
where the temperature T and the chemical potential are given in units of the electron
mass. If 0 ( + T x) is a slowly varying function of x over a range (x 1, then the
integral in Eq. (45) can be easily expanded in powers of T :
(T , ) = 0 () +
2 2
7 4 4 (4)
T 0 () +
T 0 () + .
6
360
(46)
In our case of interest, this approximation is valid for the non-oscillatory terms 0(mon) , but
not for the oscillatory terms, which vary considerably over the range (x 1 (unless the
unlikely condition T /b 1 is met).
The treatment of the oscillatory terms, which are of the form (z, {(2 1)/b}), for z
a negative semi-integer, can be done as follows. Consider placing the leading oscillatory
term,

m4 2 b5/2 3 2
(osc)
0 2
(47)
2 , pF /b + ,
4 3 F
into the integral of Eq. (45). Accordingly, we must evaluate this function at F = + T x
and pF 2 F 2 1 = pF2 +2T x +T 2 x 2 (where T is the temperature in units of the electron
mass) and weight it with the hump function h(x) = ex /(ex + 1)2 . In the degenerate regime,
T , so the term O(T 2 ) in the expression for pF 2 can be safely neglected. Since the
prefactor of the zeta function in (47) is a slowly varying function of x, we can approximate
(osc)
m4 b5/2
(T , , b) 2
6

32 , pF2 /b + 2xT /b h(x) dx.
(48)
The number of oscillations that fall under the hump will clearly be proportional to the
factor 2T /b. We expect the amplitude of the oscillatory magnetization to be more or
less constant for T /b < 1 and rapidly decreasing for T /b > 1. To obtain an explicit
expression we use Hurwitzs Fourier expansion of (z, q) shown in Eq. (A.18) (valid for
z < 0 and 0 q 1) and integrate term by term using the formula:

eiax h(x) dx =
a
.
sinh(a)
(49)
Defining the integral in Eq. (48) in a generic form as:

Iz (, ) =

z, { + x} h(x) dx,
(50)
we find it to be
Iz (, ) = (2)z+1 /(1 z)
z

n sin(2n + z/2)
sinh(2 2 n)
n=1
(51)
Now, using this result for z = 3/2, the oscillatory part of the grand potential becomes:

cos(2npF2 /b /4)
m4
(osc)
3/2 T
(52)
+ .
(T , , b) + 2 b
4
n3/2 sinh(4 2 nT /b)
2
n=1
506
5. The magnetization
5.1. The T = 0 limit
The magnetization (per unit volume) of the electron gas at T = 0 is given by
2
p

F /b
H
(q)
2e 0 (F , b) em2
0 (F , B)
1/2
= 2
=
b5/2
M0 =
dq .
B
m
b
2 2 b
1 + bq
(53)
Taking the derivative we find the formal expression for the magnetization:
2
p

F /b
e m2 3/2
2
1
M0 (F , b) =
b
dq
H1/2 (q)
+
2 2
(1 + bq)1/2 2(1 + bq)3/2
0

2
2
1/2 pF
b
H1/2 pF /b .
F
(54)
Just as 0 was separated into two terms according to the monotonic and the oscillating
(mon)
(F , b) +
parts of H1/2 , so can we separate the magnetization as M0 (F , b) = M0
(osc)
M0 (F , b). The small b expansion for M0 can be obtained from Eq. (54), or directly
taking the derivative of the expansion for 0 given in Eq. (38):

2

e m2
1/2 pF
12 , pF2 /b + 16 b cosh1 (F )
M0 (F , b) =
b
2
2
F
3/2
b (4F2 + 1) 3 2
2 , pF /b
+
3
F3

3
b5/2 (4F2 + 3) 5 2
+
(55)
,
p
/b
+
O
b
F
2
15
F5
The magnetization oscillates as a function of b (see Fig. 4), just like the well known
de Haasvan Alphen effect of non-relativistic electrons in metals. In this expression, the
oscillation appears in terms of Hurwitz functions of the fractional part of pF2 /b. Notice that,
for b 1, the oscillatory part has an amplitude considerably larger than the monotonic
part, and even larger the larger F is (see envelope curves in Fig. 4). As such, it could
be possible to have spontaneous magnetization for sufficiently dense systems at low
temperature. A discussion about the thermodynamic stability of such state was given in
Ref. [2]. However, at temperatures above some threshold, the oscillation amplitude dies
out, and with it the possibility of spontaneous magnetization [3,4].
5.2. The finite temperature case
We will be concerned in particular with the physically relevant case where pF2 /b 1,
i.e., many Landau levels occupied. In this case, the dominant contribution at T = 0 is the
507
Fig. 4. The magnetization at T = 0 which oscillates as a function of the magnetic field b, for Fermi
energy F = 4. Also shown are the upper and lower envelopes of the curve, and the upper envelopes
for larger Fermi energies (F = 8 and 16).
leading oscillatory term followed by the leading monotonic term, i.e., the second and first
terms in Eq. (55), respectively.
At finite temperature, the monotonic term becomes:

e m2 b
2 2
cosh1 ()
T
M(mon) (T , , b)
2
6 pF3
2 6

7 4 (2 + 32 ) 4
T
+
,
(56)
60
pF7
where pF2 2 1. This contribution is small compared to the oscillatory part for
temperatures smaller than the Landau level splittings, i.e., T < b/. However, for higher
T the amplitude of the oscillations gets smoothed out and it is the monotonic part shown
above what dominates the magnetization. This behavior is shown in Fig. 5.
On the other hand, the leading oscillatory term at finite T is obtained following the same
procedure as in the previous section, or directly taking the derivative of the grand potential
with respect to b:
M(osc) (T , , b) =
e m2 pF2 T sin(2npF2 /b /4)
+ .
n sinh(4 2 nT /b)
2b
(57)
n=1
This expression coincides with the results in Ref. [4]. We must also remark the strong
similitude with Landaus result for the non-relativistic case [12], which is expressed in our
notation as:
M(osc)
n.r. (T , , b) =
em2 pF2 T sin(2npF2 /b /4)
n sinh(4 2 nT /b)
2b
(58)
n=1
This is precisely the non-relativistic limit of our leading oscillating term, shown in Eq. (57):
basically, the relativistic chemical potential reduces to 1 + n.r. 1, in units of
508
Fig. 5. The upper envelope of the magnetization as a function of the temperature, for fixed
chemical potential = 100, and magnetic field b = 0.1, 1 and 10 (lower, medium and upper curve,
respectively).
Fig. 6. The upper envelope of the magnetization as a function of the magnetic field b, for two
temperatures T = 105 and 107 (lower and upper curve, respectively). The chemical potential
3 B, for a uniformly
is fixed at = 100. The dashed line represents the macroscopic relation M = 8
magnetized sphere.
electron mass, while pF 2 1 2n.r. denotes a non-trivial quantity in all

regimes.
A final important point is the possibility of ferromagnetic behavior, that is, a
magnetization that is sustained with the magnetic field generated by the same system. For a
given geometry, the field and magnetization satisfy a relation of the form M = B, where
in the case of, e.g., a sphere, = 3/8 . A self-consistent solution is obtained from this
relation and the thermodynamic relation M = M(T , , B). This problem has been studied
in the past, and our conclusions agree with those results (Fig. 6). For the self-consistent
solution to exist, the oscillations are necessary, because the monotonic part of M vs. b is
509
simply too small to reach a solution with M = B. Moreover, for the oscillations not to be
thermally damped out, the temperature must be below some threshold value [3]. Finally,
there is the question of thermodynamic stability of the self-consistent solution, which is at
most metastable according to OConnell and Roussel [2].
Conclusions
We have developed a closed analytical approach to solve the thermodynamics of a
free gas of electrons immersed in an uniform magnetic field of arbitrary magnitude. The
method is completely relativistic and particularly useful in the case of a degenerate gas, a
likely situation to be met in the interior of white dwarfs, neutron stars, and magnetars. A
central role is played by one of the least known of the special functions of mathematical
physics, the Hurwitz zeta function. Its appearance comes about due to a sum over energy
levels of the form ( + n)z (Landau levels for the system treated in this paper), labeled
by a non-negative integer n. As such, the method should be applicable to study the
thermodynamics of other systems. The grand potential (and therefore all thermodynamic
quantities) can be expressed as a one-dimensional definite integral, which can be explicitly
evaluated in several interesting regimes due to the various analytic properties of the
Hurwitz zeta function. Hence, our work provides a unified derivation of several of the
results found scattered in the literature. We reproduce the de Haasvan Alphen behavior of
the magnetization in the relativistic gas and the dilution of it at high temperatures.
We have not included quantum electrodynamics corrections in this treatment. Work to
extend our results in this direction, in a fully relativistic fashion, is in progress.
Acknowledgements
This work was supported by CONICYT under Grant Fondecyt PLC-8000017.
Appendix A. The Hurwitz zeta function

We collect here some of the properties of the Hurwitz zeta function and present a
(presumably original) derivation of its asymptotic behavior for large q. For a detailed
account consult, for instance, Refs. [15] or [16].
The Hurwitz zeta function (z, q) is defined as the analytic extension to the whole
complex z plane of the series
(z, q) =

n=0
1
(n + q)z
(A.1)
valid for Re z > 1 and q = 0, 1, 2, . . . . The resulting function is analytic everywhere

except at z = 1, where it has a simple pole with unit residue.
510
For q = 0 one has

(z, q + 1) = (z, q)
1
,
qz
(A.2)
which, iterated N times, leads to

N

n=0
1
= (z, q) (z, q + N + 1).
(n + q)z
(A.3)
This finite sum is what appears in the grand potential as a sum over Landau levels,
and its expression in terms of Hurwitz zeta functions is what allows us to expand the
thermodynamic quantities in different limiting scenarios.
The derivative of (z, q) with respect to q is again a Hurwitz zeta function:
(z, q) = z (z + 1, q).
q
(A.4)
Hermites integral representation [15]

1
1
(z, q) =
q z+1 + q z + 2 q z+1
z1
2

0
sin(z tan1 t) dt
,
(1 + t 2 )z/2 (e2t q 1)
(A.5)
valid for all z = 1 and q > 0, can be used to study the large q behavior of (z, q). For large
q, the leading contribution to the integral in Eq. (A.5) comes from the region of small t.
To isolate this contribution we split the integration range into [0, a] and [a, ), where a
is a fixed number less than one (for instance, a = 1/2). In order to approximate the first
integral we shall use the remarkable series expansion:

1 + t2
z/2

(z)2k+1 2k+1
t
sin z tan1 (t) =
(1)k
,
(2k + 1)!
(A.6)
k=0
which converges uniformly for |t| < 1. In (A.6) (z)n is the Pochhammer symbol, or shifted
factorial, defined by
(z)n =
/(z + n)
= z(z + 1) (z + n 1).
/(z)
(A.7)
Thus, for a given z and 0 < t a < 1 we have the uniform approximation

N
sin(z tan1 t)

k (z)2k+1 2k
t
(1)

< ,
t (1 + t 2 )z/2
(2k + 1)!
(A.8)
k=0
where can be an arbitrarily small positive number and N = N(, z) is a sufficiently big
number. So we can write
a
0

sin(z tan1 t)
(z)2k+1
dt =
(1)k
2
z/2
2qt
(1 + t ) (e
1)
(2k + 1)!
N
k=0
a
t 2k+1
dt + R(),
1
e2qt
0
(A.9)
511
with
a
|R()| <
0
t
e2qt 1
dt <
0
t
dt =
.
e2qt 1
24q 2
(A.10)
Also, up to a correction that vanishes exponentially as q , we can approximate

a
0
t 2k+1
dt
2qt
e
1
t 2k+1
B2k+2
dt = (1)k
,
1
4(k + 1)q 2k+2
e2qt
0
(A.11)
where we have used formula 3.411.2. of Ref. [17]. The Bn in (A.11) are the Bernoulli
numbers of even index, B0 = 1, B2 = 1/6, B4 = 1/30, etc. Since the integral over [a, )
in (A.5) also gives a exponentially small contribution for large q, we finally have the result
(z, q) =
N

(z)2k+1 B2k+2
1
1
q z+1 + q z + q z1
(1)k
z1
2
(2k + 2)!q 2k
k=0

+ O() + O eq .
(A.12)
Letting 0 (and consequently N ) we obtain the following expansion for (z, q):
(z, q) =
B2k+2
1
1
1
q 1z + q z +
(z)2k+1 z+2k+1
z1
2
(2k + 2)!
q
(A.13)
k=0
Bk /(k + z 1)
1
(1)k
,
/(z)
k!
q k+z1
(A.14)
k=0
where (A.14) follows because the only non-vanishing Bernoulli number of odd index is
B1 = 1/2.
The cases z = 1/2 and z = 1/2 are of special relevance for the work of this paper and
we give the corresponding asymptotic expansions explicitly:

2
1
1
12 , q = q 3/2 + q 1/2 q 1/2 + O q 5/2 ,
(A.15)
3
2
24

1
1
12 , q = 2q 1/2 + q 1/2 + q 3/2 + O q 7/2 .
(A.16)
2
24
For general z, the expansion (A.13) is only asymptotic. It is easy to check that its radius
of convergence, as a series in the variable 1/q, is zero. However, as an aside we should
mention that for z = m, where m is a non-negative integer, the series (A.13) terminates,
in view of (m)n = 0 for n > m. Hence the series becomes a finite polynomial in q,
actually a Bernoulli polynomial, up to a multiplicative constant:
(m, q) =
1
Bm+1 (q),
m+1
m = 0, 1, 2, . . . .
(A.17)
The Hurwitz zeta function for z < 0 admits the following Fourier expansion in the range
0 < q < 1:
512

2/(1 z)
z
z cos(2qn)
sin(2qn)
(z, q) =
+ cos
sin
.
(2)1z
2
n1z
2
n1z
n=1
n=1
(A.18)
Appendix B. Finite temperature from T = 0
In this appendix we show that the full finite temperature grand potential shown in
Eq. (21) can be obtained from its zero temperature limit as

(T , ) =
0 ( + T x)h(x) dx.
(B.1)
m
T
where h(x) is the FermiDirac hump,

ex
1
(B.2)
.
=
2
+ 1)
4 cosh2 (x/2)
Actually, result (B.1) is a special case of the fact that any physical quantity of the form
h(x) =
(ex

Q(T , ) =
q(E)
1
e(E)/T
+1
dE
(B.3)
can be computed in terms of its zero temperature limit Q0 () as

Q0 ( + T x)h(x) dx.
Q(T , ) =
(B.4)
m
T
To prove (B.4) we express the function q(E) in Eq. (B.3) as the derivative of a function
F (E) defined as:
E
F (E)
q(E ) dE ,
(B.5)
and integrate by parts. After the change of variable E = + T x we obtain

Q(T , ) =
F ( + T x)
m
T
(ex
ex
dx.
+ 1)2
(B.6)
But from (B.3) it is seen that the function F () is precisely the value of Q(T , ) in the
limit T 0:

Q0 () = lim Q(T , ) = q(E) dE F ().
(B.7)
T 0
This establishes (B.4). Now, to prove Eq. (B.1) for the grand potential, we merely notice
that the standard expression for it shown in Eq. (21) can be turned into the form (B.3) by
513
expressing the density of states g(E) as the derivative of the function G(E) which is
the total number of single particle states up to energy E:
E
G(E) =
g(E ) dE ,
(B.8)
and then integrating by parts to obtain

(T , ) =
G(E)
m
1
dE.
e(E)/T + 1
(B.9)
This expression for the grand potential is precisely of the form shown in Eq. (B.3), with
q(E) = G(E).
In the degenerate regime, T m, the lower limit of integration in (B.1) can be
replaced by with negligible error. Additionally, if the function 0 ( + T x) varies
slowly under the hump (this may not be the case for the oscillatory terms; see Section 5),
then it can be expanded in a Taylor series around and the resulting terms integrated one
by one. Using now the results

h(x) dx = 1,
x 2 h(x) dx =
x 4 h(x) dx =
7 4
,
15
2
,
3

x 2n h(x) dx = 2n (22n 2)B2n ,
(B.10)
we obtain the small temperature expansion,

(T , ) = 0 () +
2 2
7 4 4 (4)
T 0 () +
T 0 () .
6
360
(B.11)
Appendix C. Reduction of the oscillatory contribution

Let p(q) be a function in the unit interval [0, 1] periodically extended over the real axis,
and f (q) an arbitrary function. Then, the integral of the product of these functions over
an arbitrary interval can be separated into a sum of integrals over integer intervals, plus a
residual integral, as follows:
Q
1
p(q)f (q) dq =
p(q)
0
Q1

{Q}

f (q + k) dq +
p(q)f q + Q dq.
k=0
We now specialize this result for p(q) = (z, {q}) as the periodic function, f (q) = (1 +
bq)s and Q = pF2 /b, and use Eq. (A.3) to express the sum in terms of Hurwitz functions:
pF2 /b1

k=0
1
1
= s
s
[1 + b(q + k)]
b
pF2 /b1

k=0
1
[(q + 1/b) + k]s
514

1
s, q + 1/b s, q + 1/b + pF2 /b .
s
b
We thus find the integral expression
=
2
p
F /b
(z, {q})
1
dq = s
s
(1 + bq)
b
{pF2 /b}
0
1
+
(C.1)
(z, q)
dq
(q + 1/b + pF2 /b)s

2
(z, q) s, q + 1/b s, q + 1/b + pF /b dq .
(C.2)
Using this generic result one derives the expression given in Eq. (36) for the oscillatory
piece of the grand potential.
Appendix D. Small b expansions

In this appendix we show that
2
p
F /b

3 ( 12 , q + 1)
2 5
1
1
2
+ O b3/2
+b
2
dq = 32 +
3
60
15
1260
1 + bq
(D.1)
and
2
p
F /b
( 12 , {q})
2
2
2
dq = 32 52 b 72 b2
3
15
35
1 + bq
2 1 3 2
2 b 5 2
2 , pF /b +
2 , pF /b
3 F
15 F3

2 b 2 7 2
2 , pF /b + O b3 .
+
5
35 F
(D.2)
The first expansion is needed in order to find the small b behavior of the non-oscillatory
piece of the grand potential, Eq. (34), and the second expansion is required for the
oscillatory piece, which is exhibited in Eq. (36) already making use of the formula (C.2).
To prove Eq. (D.1), we first define the integral
2
p
F /b
R(b)
0
3 ( 12 , q + 1)
dq,
1 + bq
and then try to expand it in powers of b. The first two terms of the expansion are easily
found, using the results:

1
1
2 3
2

R(b)
,
= 52
R(b) b=0 = 2 + ,
3
60
b
15
1260
b=0
515
which follow from the facts that 3 ( 12 , q + 1) decreases like q 5/2 when q , and that
the antiderivative of 3 ( 12 , q), which is 23 3 ( 32 , q), vanishes like q 3/2 when q .
Indeed, we can directly calculate at b = 0:

R(b)

b=0

2
1
2
3 12 , q dq = 3 32 , 1 = 32 + ,
3
3
60
(D.3)
and also

pF2

R(b)
= lim 3 12 , 1 + pF2 /b
b0
b
F b 2
b=0
1
+
2
2
p
F /b
1
2
q
3 12 , q + 1
dq
(1 + bq)3/2

q 3 12 , q + 1 dq.
(D.4)
In this last expression, the b 0 limit of the first term inside the square bracket vanishes
due to the large-q behavior of 3 ( 12 , q), while the last integral can be done by parts:
1
2

1
2
q 3 12 , q + 1 dq =
3 32 , q + 1 dq
2
3
2
12 5
1
3 2 , 1 = 52
.
=
35
15
1260
We should point out that the next term in the expansion (D.1) is not O(b2), as it would be in
a regular Taylor series, because R (0) does not exist. To get this singular term one should
go back to the original definition of R(b) and subtract the leading term in the asymptotic
expansion of 3 ( 12 , q):

3 12 , q =

1
+ 4 12 , q ,
5/2
1920q
(D.5)
where 4 ( 12 , q) is of order O(q 9/2). Then, the resulting term in the integrand,
(1 + q)5/2 (1 + bq)1/2 , can be integrated explicitly, and its small b expansion reads
2
p
F /b
1
(1 + q)5/2

2 F 2
2 2
2pF 1 b3/2 + O b2 .
dq = b +
3
3 3
3 pF
1 + bq
(D.6)
Therefore, the O(b3/2) term in the expansion of R(b) is actually

1 F 2
2pF 1 b3/2.
3
2880 pF
With this result we complete the proof of Eq. (D.1).
(D.7)
516
We now proceed to prove Eq. (D.2), which is relevant for the small b expansion of the
oscillatory piece of the grand potential. First we must realize that the integral in Eq. (D.2)
Eq. (C.1) is of the type studied in Appendix C, and in particular the result shown in
Eq. (C.1) applies. We will then proceed to expand that result for small b. To expand the first
integral in Eq. (C.1), we just need to use the binomial expansion of (q + b1 + pF2 /b)1/2 ,
where q 1/b + pF2 /b, and then use the integrals:
Q

2
2
12 , q dq = 32 , Q 32 ,
3
3
(D.8)

4
4 2
q 12 , q dq = 52 , Q + 52 + Q 32 , Q .
15
15
3
(D.9)
Q
0
We thus find the expansion for the first integral of Eq. (C.1):
2
{p
F /b}

12 , q
(q + 1/b + pF2 /b)1/2
dq
1
2 3 2 2 3
2 , pF /b 2
=
3
(1 + bpF2 /b)1/2 3

2 5
b
2 5 2
2
,
p
/b
+
F
2
15
15
(1 + bpF2 /b)3/2

1 2 3 2
+ pF /b 2 , pF /b
+ O b2 .
3
(D.10)
Notice that the terms of the form (1 + bpF2 /b)n/2 ( n+2

2 ) will cancel in the full
oscillatory piece, Eq. (36), when expansions (D.10) and (D.14) (see below) are combined.
For the remaining terms containing (1 + bpF2 /b) in the denominator we write 1 +
bpF2 /b = (1 + pF2 ) b{pF2 /b} and perform a binomial expansion, which leads to a further
cancellation of all the terms of the type {pF2 /b}k (z, {pF2 /b}) with k 1.
Now we need to expand the second integral in Eq. (C.1). In this case we must expand
the functions inside the squared brackets for large values of their argument. We thus use
the large-q expansions of ( 12 , q) shown in Eqs. (A.15) and (A.16), together with the
definite integrals [18]
1

12 , q dq = 0,
(D.11)

2
q 12 , q dq = 32 ,
3
(D.12)
1
0
1

8
2
q 2 12 , q dq = 32 52 .
3
15
517
(D.13)
We thus get the expansion:

1
1

12 , q 12 , q + 1/b 12 , q + 1/b + pF2 /b dq

2
2 5
1
1
= 32 1
b
1
2
3
15
(1 + bpF2 /b)1/2
(1 + bpF2 /b)3/2

+ O b2
(D.14)
which completes the proof of Eq. (D.2) up to order b. The O(b2 ) terms are obtained in a
similar fashion.
Finally, and for completeness, we also present here the small b expansion of the explicit
terms in (34), which reads

+ b + p2

F
1/2
1
1
1
F
1 b + b 2 ln
b + pF2
F b1/2
2
6
2
1+ b
1

1
3/2
b + pF2
F b3/2 + b cosh1 (F ) pF F
+
3
2
1
1
1
1
= cosh1 (F ) pF F + pF3 F + b2 cosh1 (F )
2
2
3
12
9/2
1 7/2
1 5/2
b +O b
.
b +
(D.15)
60
1260
One should notice that all the terms that are independent of the Fermi energy (i.e., pF or F )
are spurious, cancelling between the different expansions and thus leading to result (38).
References
[1] V. Canuto, H. Chiu, Phys. Rev. 173 (1968) 1210;
V. Canuto, H. Chiu, Phys. Rev. 173 (1968) 1220;
V. Canuto, H. Chiu, Phys. Rev. 173 (1968) 1229;
H.J. Lee, V. Canuto, H. Chiu, C. Chiuderi, Phys. Rev. Lett. 23 (1969) 390;
V. Canuto, H. Chiu, C. Chiuderi, Nature 225 (1970) 47.
[2] R.F. OConnell, K.M. Roussel, Astron. Astrophys. 18 (1972) 198.
[3] J. Schmid-Burgk, Astron. Astrophys. 26 (1973) 335.
[4] R.D. Blanford, L. Hernquist, J. Phys. C 15 (1982) 6233.
[5] A. Chodos, K. Everding, D. Owen, Phys. Rev. D 42 (1990) 2881.
[6] S.K. Blau, M. Visser, A. Wipf, Int. J. Mod. Phys. A 6 (1991) 5409.
[7] P. Elmfors, D. Persson, B.-S. Skagerstam, Phys. Rev. Lett. 71 (1993) 480;
P. Elmfors, D. Persson, B.-S. Skagerstam, Astropart. Phys. 2 (1994) 299.
[8] V. Zeitlin, hep-ph/9412204;
D. Persson, V. Zeitlin, Phys. Rev. D 51 (1995) 2026;
V. Zeitlin, J. Exp. Theor. Phys. 82 (1996) 79.
518
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
D. Cangemi, G. Dunne, Ann. Phys. 249 (1996) 582.

L.D. Landau, Z. Phys. 64 (1930) 629.
L.D. Landau, Proc. R. Soc. London A 170 (1939) 363.
See, for example, L.D. Landau, E.M. Lifshitz, Satistical Physics, Part I, 3rd edn., Pergamon
Press, 1980.
M.H. Johnson, B.A. Lippmann, Phys. Rev. 76 (1949) 828;
H. Robl, Acta Phys. Austriaca 6 (1952) 105.
L.D. Landau, E.M. Lifshitz, Quantum Mechanics (Non-relativistic Theory), 3rd edn., Course
of Theoretical Physics, Vol. 3, Pergamon Press, 1977.
E. Whittaker, G. Watson, A course of Modern Analysis, 4th edn., Cambridge Univ. Press, 1963.
J. Spanier, K.B. Oldham, An Atlas of Functions, Hemisphere Publishing, 1987.
I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products, 5th edn., Academic Press,
1994.
O. Espinosa, V. Moll, On some definite integrals involving the Hurwitz zeta function, April
2000, to appear in The Ramanujan Journal.

CUMULATIVE AUTHOR INDEX B611B612
Abel, S.A.
ALPHA Collaboration
Arndt, D.
Astier, P.
Autiero, D.
B611 (2001) 43
B612 (2001) 3
B612 (2001) 171
B611 (2001) 3
B611 (2001) 3
Bais, F.A.
Baldisseri, A.
Baldo-Ceolin, M.
Banner, M.
Bartels, M.
Baseilhac, P.
Bassompierre, G.
Becher, T.
Beneke, M.
Benslama, K.
Bernard, D.
Besson, N.
Bird, I.
Blumenfeld, B.
Bobisut, F.
Bouchez, J.
Boyd, S.
Brandhuber, A.
Bruzzo, U.
Bueno, A.
Bunyatov, S.
Buras, A.J.
B612 (2001) 229

B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 413
B612 (2001) 373
B611 (2001) 3
B611 (2001) 367
B612 (2001) 25
B611 (2001) 3
B612 (2001) 291
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 179
B611 (2001) 205
B611 (2001) 3
B611 (2001) 3
B611 (2001) 488
Caffo, M.
Camilleri, L.
Cardini, A.
Cattaneo, P.W.
Cavasinni, V.
Cervera-Villanueva, A.
Chaichian, M.
Chukanov, A.
Collazuol, G.
Conforto, G.
Conta, C.
Contalbrigo, M.
Cousins, R.
B611 (2001) 503

B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 383
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
0550-3213/2001 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 4 3 0 - 8
Czarnecki, A.
Czyz, H.
B611 (2001) 488

B611 (2001) 503
DallAgata, G.
Daniels, D.
Degaudenzi, H.
Degrassi, G.
Del Prete, T.
Demichev, A.
De Santo, A.
Dib, C.O.
Diehl, H.W.
Dienes, K.R.
Dignan, T.
Di Lella, L.
Do Couto e Silva, E.
Dumarchez, J.
Drr, S.
B612 (2001) 123

B611 (2001) 3
B611 (2001) 3
B611 (2001) 403
B611 (2001) 3
B611 (2001) 383
B611 (2001) 3
B612 (2001) 492
B612 (2001) 340
B611 (2001) 146
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 281
Ellis, M.
Espinosa, O.
B611 (2001) 3
B612 (2001) 492
Farzan, Y.
Feldman, G.J.
Feldmann, Th.
Ferrari, F.
Ferrari, R.
Ferrre, D.
Flaminio, V.
Foerster, A.
Fraternali, M.
Fucito, F.
B612 (2001) 59
B611 (2001) 3
B612 (2001) 25
B612 (2001) 151
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 461
B611 (2001) 3
B611 (2001) 205
Gaillard, J.-M.
Gambino, P.
Gangler, E.
Garousi, M.R.
Gehrmann, B.
Geiser, A.
Geppert, D.
Gibin, D.
Gninenko, S.
B611 (2001) 3
B611 (2001) 338
B611 (2001) 3
B611 (2001) 467
B612 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
520
Godley, A.
Gomez-Cadenas, J.-J.
Gomis, J.
Gosset, J.
Gling, C.
Gouanre, M.
Gould, M.D.
Grant, A.
Graziani, G.
Guan, X.-W.
Gubser, S.S.
Guglielmi, A.
Gukov, S.
B611 (2001) 3
B611 (2001) 3
B611 (2001) 179
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 461
B611 (2001) 3
B611 (2001) 3
B612 (2001) 461
B611 (2001) 179
B611 (2001) 3
B611 (2001) 179
Hagner, C.
Hatsuda, M.
Hernando, J.
Herrmann, C.
Hubbard, D.
Hurst, P.
Hyett, N.
B611 (2001) 3
B611 (2001) 77
B611 (2001) 3
B612 (2001) 123
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
Iacopini, E.
B611 (2001)
Joseph, C.
Joshipura, A.S.
Juget, F.
B611 (2001) 3
B611 (2001) 227
B611 (2001) 3
Kamimura, K.
Khorsand, P.
Kirsanov, M.
Klimov, O.
Kokkonen, J.
Kovzelev, A.
Krasnoperov, A.
Kurth, S.
Kustov, D.
Kuznetsov, V.E.
B611 (2001) 77
B611 (2001) 239
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 3
B611 (2001) 3
B611 (2001) 3
Lacaprara, S.
Lachaud, C.
Lakic, B.
Lanza, A.
La Rotonda, L.
Laveder, M.
Letessier-Selvon, A.
Levy, J.-M.
Lima-Santos, A.
Linssen, L.
Ljubic, A.
Long, J.
Lukyanov, S.
Lupi, A.
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 446
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 391
B611 (2001) 3
Ma, J.P.
Mack, G.
B611 (2001) 523

B612 (2001) 413
Marchionni, A.
Martelli, F.
Martinelli, G.
Maxwell, C.J.
Mchain, X.
Mendiburu, J.-P.
Meyer, J.-P.
Mezzetto, M.
Miao, Y.-G.
Mirjalili, A.
Mishra, S.R.
Misiak, M.
Misiak, M.
Mizoguchi, S.
Moorhead, G.F.
Mller-Kirsten, H.J.W.
B611 (2001) 3
B611 (2001) 3
B611 (2001) 311
B611 (2001) 423
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 215
B611 (2001) 423
B611 (2001) 3
B611 (2001) 338
B611 (2001) 488
B611 (2001) 253
B611 (2001) 3
B612 (2001) 215
Naumov, D.
Ndlec, P.
Nefedov, Yu.
Neubert, M.
Nguyen-Mau, C.
Nishino, H.
NOMAD Collaboration
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 367
B611 (2001) 3
B612 (2001) 98
B611 (2001) 3
Olum, K.D.
Orestano, D.
B611 (2001) 125

B611 (2001) 3
Palma, G.
Park, D.K.
Paschos, E.A.
Pastore, F.
Peak, L.S.
Pennacchio, E.
Peres, O.L.G.
Pessard, H.
Petrov, A.A.
Petti, R.
Placci, A.
Polesello, G.
Pollmann, D.
Polyarush, A.
Popov, B.
Poulsen, C.
Prenajder, P.
B612 (2001) 413

B612 (2001) 215
B611 (2001) 227
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 59
B611 (2001) 3
B611 (2001) 367
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 383
Rajpoot, S.
Regnault, N.
Remiddi, E.
Rico, J.
Riemann, P.
Roda, C.
Rodejohann, W.
Roditi, I.
Rolf, J.
Rossi, G.C.
B612 (2001) 98
B612 (2001) 291
B611 (2001) 503
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 227
B612 (2001) 461
B612 (2001) 3
B611 (2001) 311

Rubbia, A.
Rupp, C.
Russo, J.G.
B611 (2001) 3
B612 (2001) 313
B611 (2001) 93
Sachrajda, C.T.
Salvatore, F.
Santiago, J.
Schahmaneche, K.
Scharf, R.
Schmidt, B.
Schmidt, T.
Sconza, A.
Seidel, D.
Serban, D.
Servant, G.
Sevior, M.
Sfetsos, K.
Sharpe, S.
Sheikh-Jabbari, M.M.
Shpot, M.
Sibold, K.
Siemens, X.
Sillou, D.
Slavich, P.
Slingerland, J.K.
Smirnov, A.Yu.
Soler, F.J.P.
Sozzi, G.
Stanishkov, M.
Steele, D.
Stiegler, U.
Stipc, M.
Stolarczyk, Th.
B611 (2001) 311

B611 (2001) 3
B611 (2001) 447
B611 (2001) 3
B612 (2001) 313
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 25
B612 (2001) 291
B611 (2001) 43
B611 (2001) 3
B612 (2001) 191
B611 (2001) 311
B611 (2001) 383
B612 (2001) 340
B612 (2001) 313
B611 (2001) 125
B611 (2001) 3
B611 (2001) 403
B612 (2001) 229
B612 (2001) 59
B611 (2001) 3
B611 (2001) 3
B612 (2001) 373
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
Talevi, M.
Tani, T.
Tanzini, A.
Tareb-Reyes, M.
Taylor, G.N.
Taylor, T.R.
Tereshchenko, V.
B611 (2001) 311

B611 (2001) 253
B611 (2001) 205
B611 (2001) 3
B611 (2001) 3
B611 (2001) 239
B611 (2001) 3
521
Testa, M.
Toropin, A.
Touchard, A.-M.
Tovey, S.N.
Tran, M.-T.
Travaglini, G.
Tsesmelis, E.
Tseytlin, A.A.
Tsvelik, A.M.
Tureanu, A.
B611 (2001) 311

B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 205
B611 (2001) 3
B611 (2001) 93
B612 (2001) 479
B611 (2001) 383
Ulrichs, J.
Urban, J.
B611 (2001) 3
B611 (2001) 488
Vacavant, L.
Valdata-Nappi, M.
Valuev, V.
Vannucci, F.
Varvell, K.E.
Veltri, M.
Vercesi, V.
Vidal-Sitjes, G.
Vieira, J.-M.
Vinogradova, T.
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
3
3
3
3
3
3
3
3
3
3
Weber, F.V.
Weisse, T.
Wilson, F.F.
Winton, L.J.
Wolff, U.
B611 (2001)
B611 (2001)
B611 (2001)
B611 (2001)
B612 (2001)
3
3
3
3
3
Yabsley, B.D.
Yndurin, F.J.
B611 (2001) 3
B611 (2001) 447
Zaccone, H.
Zagermann, M.
Zhou, H.-Q.
Zuber, K.
Zuccon, P.
Zwirner, F.
B611 (2001) 3
B612 (2001) 123
B612 (2001) 461
B611 (2001) 3
B611 (2001) 3
B611 (2001) 403

Nucl - Phys.B v.612

Uploaded by

Copyright:

Available Formats

You might also like

Nucl - Phys.B v.612

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.612

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 612 (2001) 324

Schrdinger functional at negative flavour number

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

and the fermionic action

Here, D is the O(a) improved WilsonDirac operator including the Sheikholeslami

if p is a time-like plaquette attached to a boundary plane. In all other cases, w(p) = 1.

From the Schrdinger functional, a running coupling may be defined by differentiating

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

Now, since  = log Z/ is a renormalized quantity [7] with the perturbative

3. Perturbative computation of the cutoff effects

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

+ 0 )fA (x0 ) + cA a0 0 fP (x0 )

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

Therefore, the computation of mc has to be done in two steps. First we compute mc by

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

At 1-loop order, fA and fP now get several contributions:

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

in the sum. Thus mc becomes Nf dependent,

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

4. Monte Carlo simulations with bermions

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

In a similar way, we can express expectation values of fermionic observables at Nf = 2

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

Re Tr U  (x, ) U (x, ) S (x, )

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

Two lattice sites are affected by the hopping term,

x (x) = (1 ) U (x, ) U (x, ) (x + a ),

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

5. Technical details and results

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

to the continuum limit by an extrapolation in a/L with the ansatz

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

Fig. 7. Check of lattice artefacts in the mass at u = 0.9793 and u = 1.5145.

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

ALPHA Collaboration / Nuclear Physics B 612 (2001) 324

S. Sint, R. Sommer, Nucl. Phys. B 465 (1996) 7198, hep-lat/9508012.

Nuclear Physics B 612 (2001) 2558

Systematic approach to exclusive B V +, V

M. Beneke et al. / Nuclear Physics B 612 (2001) 2558

measurement is experimentally more difficult but theoretically simpler to interpret, since

M. Beneke et al. / Nuclear Physics B 612 (2001) 2558

2 /4 is the fine-structure constant and m

M. Beneke et al. / Nuclear Physics B 612 (2001) 2558

M. Beneke et al. / Nuclear Physics B 612 (2001) 2558

leading order contribution from the electromagnetic dipole operator O7 reads Ti (q 2 ) =

M. Beneke et al. / Nuclear Physics B 612 (2001) 2558

/E, and Ta, (u, ) is expanded as

M. Beneke et al. / Nuclear Physics B 612 (2001) 2558

The non-factorizable correction is obtained by computing matrix elements of four-quark

M. Beneke et al. / Nuclear Physics B 612 (2001) 2558

from the spectator quark line. We then find:

Now, since = log Z/ is a renormalized quantity [7] with the perturbative

Re Tr U (x, ) U (x, ) S (x, )

20 + (0) (0) (1) (1) 1 (1)