Nucl - Phys.B v.583

Nuclear Physics B 583 (2000) 334
www.elsevier.nl/locate/npe
Renormalization and running of quark mass and

field in the regularization invariant and MS schemes
at three and four loops
K.G. Chetyrkin 1 , A. Rtey
Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
Received 12 November 1999; revised 17 May 2000; accepted 23 May 2000
Abstract
We derive explicit transformation formulae relating the renormalized quark mass and field as
defined in the MS-scheme with the corresponding quantities defined in any other scheme. By
analytically computing the three-loop quark propagator in the high-energy limit (that is keeping
only massless terms and terms of first order in the quark mass) we find the NNNLO conversion
factors transforming the MS quark mass and the renormalized quark field to those defined in a
Regularization Invariant (RI) scheme which is more suitable for lattice QCD calculations. The
NNNLO contribution in the mass conversion factor turns out to be large and comparable to the
previous NNLO contribution at a scale of 2 GeV the typical normalization scale employed in
lattice simulations. Thus, in order to get a precise prediction for the MS masses of the light quarks
from lattice calculations the latter should use a somewhat higher scale of around, say, 3 GeV where
the (apparent) convergence of the perturbative series for the mass conversion factor is better.
We also compute two more terms in the high-energy expansion of the MS renormalized quark
propagator. The result is then used to discuss the uncertainty caused by the use of the high energy
limit in determining the MS mass of the charmed quark. As a by-product of our calculations we
determine the four-loop anomalous dimensions of the quark mass and field in the Regularization
Invariant scheme. Finally, we discuss some physical reasons lying behind the striking absence of
(4) in these computed anomalous dimensions. 2000 Elsevier Science B.V. All rights reserved.
PACS: 12.38.B; 12.38.C; 12.38.G; 12.15.F
Keywords: Quantum chromodynamics; Lattice QCD calculations; Perturbation theory; Quark masses;
Anomalous dimensions; Scheme dependence
ar@particle.physik.uni-karlsruhe.de
1 Permanent address: Institute for Nuclear Research, Russian Academy of Sciences, 60th October Anniversary
Prospect 7a, Moscow 117312, Russia.
0550-3213/00/$ see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 3 1 - X
K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334
1. Introduction
Quark masses are fundamental parameters of the QCD Lagrangian. Nevertheless,
their relation to measurable physical quantities is not direct: the masses depend on the
renormalization scheme and, within a given one, on the renormalization scale .
In the realm of pQCD the definition which is most often used is based on the MS-scheme
[1,2] which leads to the so-called short-distance MS mass. Such a definition is of great
convenience for dealing with mass-dependent inclusive physical observables dominated
by short distances (for a review see [3]). Unfortunately it is usually difficult to get precise
information about the quark masses from predictions from these considerations, as their
mass dependence is relatively weak.
To determine the absolute values of quark masses, one necessarily has to rely on the
methods which incorporate the features of nonperturbative QCD. So far, the only two
methods which are based on QCD from first principles are QCD sum rules and Lattice
QCD (for recent discussions see, e.g., [413]. Rather accurate determinations of the ratios
of various quarks masses can be obtained within Chiral Perturbation Theory [14].
Lattice QCD provides a direct way to determine quark masses from first principles.
Unlike QCD sum rules it does not require model assumptions. It is possible to carry out a
systematic improvement of Lattice QCD so that all the discretization errors proportional to
the lattice spacing are eliminated (a comprehensible review is given in [15]). The resulting
quark mass is the (short distance) bare lattice quark mass. The matching of the lattice
quark masses to those defined in a continuum perturbative scheme requires the calculation
of the corresponding multiplicative renormalization constants. In the RI scheme [16] the
renormalization conditions are applied to amputated Green functions in Landau gauge,
setting them equal to their tree-level values. This allows the non-perturbative calculation
of the renormalization constants. An alternative to the RI approach is the Schroedinger
functional scheme (SF) which was used in [17,18].
An impressive number of various lattice determinations of quark masses recently has
been performed (see Refs. [1931]).
Once the RI quark masses are determined from lattice calculations they can be related to
the MS mass by a corresponding conversion factor. By necessity this factor can be defined
and, hence, computed only perturbatively. The conversion factor is presently known at
next-to-next-to-leading order (NNLO) from [32]. The NNLO contribution happens to be
numerically significant. This makes mandatory to know the NNNLO O(s3 ) term in the
conversion factor.
In the present article we describe the calculation of this term. It turns out that the size of
the newly computed term is comparable to the previous one at a renormalization scale
of 2 GeV the typical scale currently used in lattice calculations of the light quark
masses. This means that perturbation theory can not be used for a precise conversion of
the presently available RI quark masses to the MS ones. A simple analysis shows that the
convergence gets much better if the scale is increased to, say, 3 GeV. Thus, once the lattice
calculations produce the RI quark masses at this scale our formulas will allow an accurate
conversion to the MS masses at the same scale.
The article is organized as follows. In Section 2 we discuss the scheme dependence of

the quark field and mass and present a general procedure to find corresponding conversion
factors from one scheme to another. The general technique is illustrated by constructing
the conversion factors between the MS and the RI scheme. In Section 3, we present the
first few terms of the small mass expansion of the three-loop quark-propagator in the MS
scheme. In Section 4 we first present the conversion functions for the quark mass and
field, and then investigate the validity of the massless approximation for these functions.
Then we use these results to calculate the anomalous dimensions for the quark mass and
field in the RI scheme, the so-called RG invariant mass m
b. The final section is devoted to
conclusions.
In Appendices A and B we display our results for the small mass expansion of the
fermion propagator and the various conversion factors with their full dependence on
the group theoretical factors CF , CA and T . In Appendix C the four loop anomalous
dimensions of the quark mass and field are listed for the case of a SU(N) gauge group.
Our main results are also available as ASCII input for the programs FORM and
M ATHEMATICA at the following internet address: http://www-ttp.physik.uni-karlsruhe.de/
Progdata/ttp99/ttp99-43/
2. Scheme dependence of quark mass and field

2.1. Generalities
We start by considering the bare quark propagator (for simplicity we stick to the Landau
gauge in this section and, thus, do not explicitly display the gauge dependence)
Z

1
,
(1)
S0 q, s0 , m0 = i dx eiqx T 0 (x) 0 (0) =
m0 q/ 0
with the quark mass operator 0 being conveniently decomposed into Lorentz invariant
structures according to
0 = q/ V0 + m0 S0 ,
(2)
where m0 and 0 are the bare quark mass and field, respectively. Additionally we are using
the common shortcuts for the coupling constant throughout this article:
g2
s0
= 02 ,
4
g0 being the bare QCD gauge coupling. To be precise we assume that (1) is dimensionally
regulated by going to non-integer values of the spacetime dimension D = 4 2 [3336].
The MS renormalized counterpart of the Green function (1) reads
Z

1
=
S(q, s , m, ) = i dx eiqx T (x)(0)
m q/

1
0

(3)
= Z2 S0 q, s , m0 m =Zm m, 0 = Z s ,
as0
where the renormalized quark field is

1/2
= Z2
and the t Hooft mass parameter is the scale at which the renormalized quark mass is
defined. The renormalization constants Z2 , Z and Zm are series of the generic form
X (i) 1
X (i,j ) s j
Z? i ,
Z?(i) =
Z?
.
(4)
Z? = 1 +

j >i
i>0
A general theorem (first rigorously proven for minimal subtraction in [37,38]) states
that there is a unique choice of renormalization constants of the form (4) which makes the
propagator 2 finite in the limit of D 4.
The independence of the bare coupling constant, mass and quark field on leads in
the standard way to the following renormalization group equations for their renormalized
counterparts

X s i+2 X

2 d s ()
=
(
)
i
=
(i)Z(1,i),
(5)
d2 g0 ,m0
i>0
i>0

i+1
X

d
(i) s

m()
=
m()
(
)
m
s
m

d2
g0 ,m0
2
i>0
= m()as
and
(1)
Zm
(6)
as

X (i) s i+1

d

()
= ()2 (s ) ()
2
2
d2
0 ,g0 ,m0
2
i>0
= ()as
Z2(1)
.
(7)
as
Now let us consider the quark propagator renormalized according to a different
subtraction procedure. Marking with a prime parameters of the second scheme one can
write
Z

1
0
0
0
S q, s , m , = i dx eiqx T 0 (x) 0 (0) = 0
m q/ 0

1

= Z20
S0 q, s0 , m0
,
(8)
0
0
0
m0 =Zm m, s = Z s
where without essential loss of generality we have set 0 = . The finiteness of the
renormalized fields and parameters in both schemes implies that, within the framework
of perturbation theory, the relation between them can uniquely be described as follows
m = Cm m0 ,
2 In fact all Green functions if proper renormalization constants for gluon and ghost fields are introduced.
(9)
C2 0 ,
with the conversion functions being themselves finite series in

X (i) 0 i
s
C?
C? 1 +
(10)
s0 ,
i.e.,
(11)
i>0
for ? = m or 2.
Note that in general the coefficients C?i may depend on the ratio m0 /. If such a
dependence is absent then the corresponding subtraction scheme is referred to as a mass
independent one. In what follows we mainly limit ourselves to considering this latter case.
In addition, being only interested in the conversion functions C2 and Cm , we will assume
that the function C has already been determined and, thus, will deal with the following
representation of C2 and Cm in terms of the MS coupling constant s :
X (i) s i
C?
.
(12)
C? 1 +
i>0
The running of m0 and 0 is governed by the corresponding anomalous dimensions

m (as ) and 2 (as ). A direct use of Eqs. (9), (10) gives
ln Cm ,
(13)
m0 = m
as
ln C2 .
(14)
20 = 2
as
At last, from Eq. (1) it is easy to see that
C2
(15)
S(q) = C2 S 0 (q) = 0
0
m (1 S ) q/ (1 + V0 )
or, equivalently,
C2 (1 + V ) = 1 + V0 ,
(16)
C2 Cm (1 S ) = 1 S0 .
(17)
The renormalization conditions for the non-MS scheme should then be used to provide
the necessary information about the right hand side to calculate the conversion factors Cm
and C2 once the MS renormalized V and S are known.
2.2. Regularization invariant scheme versus MS
The MS subtraction scheme is intimately connected to dimensional regularization and,
thus, can not be directly used with other regularizations, including the lattice one. In
addition, the physical meaning of its normalization parameter is not transparent and
leads to the well-known ambiguities when considering the decoupling of heavy particles.
It is well-known that the above shortcomings are absent for a wide class of so-called
momentum subtraction (MOM) schemes. 3 The MOM schemes require the values of
3 In a sense the oldest subtraction scheme the on-shell one for QED can also be considered as an example
of a MOM scheme.
properly chosen Green functions with predefined dependent configurations of external

momenta to be fixed (usually to their tree values) independently on the considered order.
Practical calculations can then be performed with any regulator (or even without it in the
regulator-free approach of [39,40]). A shortcoming of MOM schemes is that they are in
general not mass-independent which leads to a complicated running of coupling constant(s)
and mass(es).
A general analysis of the problem of constructing mass-independent subtraction
schemes was performed long ago in [41]. Following essentially Weinbergs ideas, a specific
example of a mass independent MOM scheme for QCD has recently been considered
in [16] under the name of RI (Regularization Invariant) scheme (see also [42]). The
corresponding renormalization conditions for SRI and VRI read

(/
q (1 + VRI ))
1
Tr
= 1,
lim
m0 48
q
q 2 =2
(18)

1
RI
Tr 1 S q 2 =2 = 1,
lim
m0 12
where the trace is to be taken over Dirac, Lorentz and colour indices. Note that the zero
mass limit in (18) means that both VRI and SRI are effectively massless functions only
depending on the QCD coupling constant, the normalization point and q 2 . This also
implies that it is sufficient to compute the MS functions V and S in massless QCD
when computing the conversion factors from relations (16) and (17).
Application of the renormalization conditions (18) to the conversion formulae (16) and
RI and C RI . All the dependence of
(17) leads to equations that can simply be solved for Cm
2
V on q 2 is of the form of ` = log(q 2 /2 ), which simplifies the trace and derivative
w.r.t. q and leads to

1 V (`) 1
,
(19)
C2RI = 1 + V +
2
`
q 2 =2 , m=0

V (`)
1 + V + 12 `
RI
=
.
(20)
Cm
1 S
q 2 =2 m=0
Another quark field renormalization that is useful in numerical lattice simulations is defined
by the condition

1
0
Tr 1 + VRI q 2 =2 = 1,
12
which results in the even simpler conversion factors
lim
m0
,
C2RI = [1 + V ]1
q 2 =2 , m=0

1 + V
RI0
=
.
Cm
1 S q 2 =2 , m=0
(21)
(22)
(23)
Using these equations all conversion factors can easily be obtained, once the MS
renormalized expressions S and V are known.
It should be noted that in practical lattice calculations the massless limit on the left hand
side of (18) and (21) is implemented by choosing m. On the other hand, should be
much less than the inverse lattice spacing 1/a. Typically is taken around 2 GeV. This
means that lattice determinations do not lead directly to the RI quark mass but rather to the
mass in a different, mass-dependent, scheme. The difference between both schemes can be
numerically non-negligible in the case of the charmed quark.
Thus, for both the RI and the RI0 scheme it is suggestive to introduce their massdependent counterparts MOM and MOM0 as defined by the same Eqs. (18) and (21) but
without taking the m 0 limit. The corresponding conversion factors to the MS scheme
read

1
V 1
,
(24)
C2MOM = 1 + V + q 2 2
2
q q 2 =2
MOM
Cm
1 + +
V
1
2
V
q 2
q 2
1 S
(25)
q 2 =2
and
0
,
C2MOM = [1 + V ]1
q 2 =2

1 + V
MOM0
=
,
Cm
1 S q 2 =2
(26)
(27)
respectively.
3. Three-loop quark propagator in MS-scheme

To find the conversion factors for MOM and MOM0 one needs to compute the functions
V and S including their full mass dependence. A full analytical result at two-loop level
has been obtained only recently in [43]. An extension of this calculation up to three-loops
is out of reach of present calculational technologies. Fortunately for the RI and RI0 schemes
one effectively only needs to compute massless three-loop diagrams a problem which
in principle was solved long ago in [44]. A promising approach to recover the full mass
dependence of the quark propagator seems to be to employ an expansion in (m2 /q 2 ) [45].
Indeed, as has been demonstrated in [4649] small mass expansions can be a very effective
tool for accurate predictions of mass dependences provided one is not too close to the
threshold (q 2 = m2 in our case). The exact two-loop result for the quark propagator can
provide some insight into the accuracy of such an expansion.
We have analytically computed three terms in the small mass expansion of the quark
propagator to order s3 . The calculation has been done with intensive use of computer
algebra programs. In particular, we have used QGRAF [50] for the generation of
diagrams and LMP [51] for the diagrammatic small mass expansion. The small mass
expansion results in products of massless propagators and massive tadpoles. These have
been evaluated with the help of the FORM packages MATAD [53] and MINCER [52].
10
(A detailed description of the status of these algebraic programs can be found in [54].) It
is convenient to write the functions S/V in the following way:
X (i)
S/V asi ().
(28)
S/V = 1 +
i>1
Our results for the separate contributions in the Landau gauge and in the MS scheme
read, where we here only keep the nf dependence, nf being the number of light quark
flavours, with one quark flavour of mass m and nf 1 massless quark flavours 4 (lqm
2 ), lq ln(q 2 /2 ), z = m2 /q 2 , and in Landau gauge V(1) 0):
ln(q 2 /m

359
7
3
67
1
(2)
nf 3
lq +
nf lq
V =
144 48
4
48
12

9
1
79 1
nf 3 lq lqm
+z
24 6
4
2

1
10
209
1
19 2
209
331
nf +
3
lq nf lq
lq
lqm
+ z2
216 24
3
72
6
12
144

1
19
19 2
nf lqm
lq lqm
l
12
12
48 qm

1
109
7 2
109
7
7 2
3 123
+
nf +
lq +
l +
lqm +
lq lqm +
l
+z
32
18
54
36 q 108
36
144 qm

1
5093
317 2
5093
668909
+
nf +
lq
lq +
lqm
+ z4
172800 48
2160
72
4320

317
317 2
lq lqm
l
72
288 qm

1
249353
25553 2
249353
23155073
+
nf +
lq
l +
lqm
+ z5
2592000
90
10800
1080 q
21600

25553
25553 2
lq lqm
l
,
(29)
1080
4320 qm

(3)
V
439543 12361
785 2 8009
55
79
nf +
nf
3 +
nf 3
4
10368
2592
7776
384
72
256
1165
52321
559
13 2
607
5
lq +
nf lq
n lq +
3 lq
+
192
2304
216
216 f 128
1
737 2
133
1 2 2
2
lq
nf lq
n l
nf 3 lq +
+
4
192
288
72 f q

5 2 10025
65
3
5461 667
nf +
nf
3 +
nf 3 + 4
+z
96
144
54
288
36
2
6235
857
185
1 2
123
5
lq +
nf lq
n lq +
3 lq
576
24
72
18 f
8
4 The expressions including the full dependence on the gauge parameter and the group theoretical factors C ,
A
CF and T are given in Appendix A. Note that the result for the three loop massless quark propagator can also be
found in the source code of the updated version of the FORM program MINCER [55].
11
3
9 2
481
1
3
nf 3 lq +
lq
lqm + nf lqm 3 lqm
4
16
96
9
4
51
1
111 2
1
2
lq lqm + nf lq lqm
l +
nf lqm
16
6
64 qm 12

11 2 196475
719
397
389057
nf +
n +
3
nf 3
+ z2
82944
1296
216 f
3456
432
35
8815
19
8303
1697
4
5 +
B4
lq
nf lq
+
36
432
216
384
648
17 2
5177
47
4739 2
n lq
3 lq +
nf 3 lq
l
+
216 f
144
18
288 q
553
1 2 2
973 3
23
115007
2
3
nf lq
n l +
l
nf lq
lqm
+
432
18 f q 288 q 54
6912
3029
5 2
617
3
nf lqm +
nf lqm
3 lqm + nf 3 lqm
2592
108
288
4
4127
35
1 2
157 2
lq lqm +
nf lq lqm
nf lq lqm
l lqm
144
36
36
64 q
3
3923 2
287
1915
2
2
2
lqm +
nf lqm
lq lqm
nf lq
lqm
8
384
1728
384

1
3359 3
11
2
3
nf lq lqm
l +
nf lqm ,
(30)
18
2304 qm 864

4
1
1
1 4
1 5
(1)
z +
z ,
(31)
S = + lq z 2lq z lqm z + z2 + z3 +
3
2
6
12
20

5009 13
47
509
4
15 2
1
2
+
nf +
3 +
lq nf lq
lq +
nf lq
S(2) =
288
18
12
48
9
8
12

5
43
319
7
7 2
1
791
2
+
nf
3
lq +
nf lq + lq nf lq
+z
144 18
12
24
18
2
3

409
5
9
1
2
lqm +
nf lqm lq lqm nf lq lqm 2lqm
48
18
4
6

7
95
1
49
1
1
13
2
nf
lq + nf lq
lq
lqm +
nf lqm
+ z2
12 72
24
4
12
24
12

49
49 2
lq lqm
l
12
48 qm

5
829
1
527 2
295
451
nf +
lq +
nf lq +
lq +
lqm
+ z3
324 54
216
12
108
108

1
527
527 2
nf lqm +
lq lqm +
l
+
36
108
432 qm

53
4831
1
4843 2
211333
nf
lq +
nf lq +
l
+ z4
20736
864
432
24
216 q

551
1
4843
4843 2
lqm +
nf lqm +
lq lqm +
l
108
72
216
864 qm
12

77
349147
1
3539 2
24966073
nf
lq +
nf lq +
l
+ z5
864000
1800
3600
40
40 q

21667
1
3539
3539 2
lqm +
nf lqm +
lq lqm +
lqm ,
450
120
40
160
(3)
S
(32)

1612847 79621
191 2 150265
755
+
nf
n +
3
nf 3
=
6912
3888
729 f
1728
216
1 2
79
5
6455
40329
n 3 +
4 +
nf 4
5 +
lq
54 f
256
12
576
256
6083
73 2
10013
17
nf lq +
n lq
3 lq +
nf 3 lq
432
324 f
384
36

2589 2
119
2 2 2
65 3
7
1 2 3
2
3
lq +
nf lq
nf lq +
lq
nf lq
nf lq
+
64
32
27
16
18
108

25 2 178745
313
303803 2969
+
nf
n
3 +
nf 3
+z
6912
648
324 f
3456
54
1
13
9635
2
172685
4
5 + B4
lq
n2f 3
9
3
1728
9
1152
4703
5 2
7493
17
367 2
nf lq
nf lq +
3 lq +
nf 3 lq +
l
+
432
81
192
36
6 q
193
17 2 2
135 3
14
1 2 3
2
3
nf lq
n l
l +
nf lq
n l
24
108 f q
8 q
9
18 f q
73547
839
25 2
1645
lqm +
nf lqm
n lqm +
3 lqm
768
108
324 f
384
5
431
391
5 2
lq lqm
nf lq lqm +
n lq lqm
+ nf 3 lqm
6
32
144
54 f
167 2
1
1 2 2
2359 2
2
lq lqm + nf lq
nf lq lqm
lqm
lqm
16
3
36
96
17
9 3
1
2
2
3
nf lqm
+
10lq lqm
lqm
+ nf lqm
18
2
9

35 2 34153
91
882619 3151
nf
n
3 +
nf 3 3 4
+ z2
82944
1728
648 f
1728
144
55
131681
3293
13 2
3251
5
lq +
nf lq
nf lq
3 lq
54
6912
432
72
576
1203 2
35
5 2 2
5401 3
49
2
3
l +
nf lq
n l +
l
nf lq
+
32 q 432
72 f q
432 q 54
47095
2285
2 2
3251
lqm +
nf lqm
n lqm
3 lqm
+
13824
864
27 f
1152
1303
887
1 2
773 2
lq lqm +
nf lq lqm +
nf lq lqm
l lqm
24
432
36
288 q
49
2431 2
1271
6947
2
2
2
nf lq
lqm +
nf lqm
lq lqm
lqm
72
192
1728
576

13121 3
49
3
l +
nf lqm .
(33)
3456 qm 864
13
4. RI scheme versus MS and MOM schemes

4.1. Three loop conversion functions
A direct use of Eqs. (19), (20), (22) and (23) leads to the following analytical expressions
for the conversion factors between the MS and RI schemes. The results are shown for QCD
(SU(3)) and Landau gauge as functions of nf (the i are the values (i) of Riemanns Zeta
function):
2

s
5
517
RI
+ 12 3 + nf
C2 = 1 +
4
18
3
3
79
1165
1287283 14197
s
+
3 +
4
5
+
4
648
12
4
3

18014
368
1102 2
nf
3 nf
nf ,
+
(34)
81
9
243

2

s
89
16
s
1990 152
+
3 +
nf
4
3
4
9
3
9
3
2960
236650
6663911 408007
s
+
3
5 +
nf
+
4
648
108
9
243

4936
80
8918 2 32
2
3 nf +
4 nf
n
3 nf ,
27
3
729 f 27
RI
=1+
Cm
0
C2RI
RI0
Cm
(35)

s 2 359
7
+ 12 3 + nf
=1+
4
9
3
3
79
1165
439543 8009
s
+
3 +
4
5
+
4
162
6
4
3

24722
440
1570 2
nf
3 nf
n ,
+
81
9
243 f

2

16
s
s
83
3779 152
+
+
3 +
nf
=1+
4
3
4
18
3
9
3
2960
217390
3115807 195809
s
+
3
5 +
nf
+
4
324
54
9
243

4720
80
7514 2 32
3 nf +
4 nf
nf
3 n2f .
27
3
729
27
(36)
(37)
At a scale of = 2 GeV and nf = 4, the numerical contributions of the leading order to

NNNLO terms are as follows (for simplicity we inserted s / = 0.1):
C2RI = 1.0 + 0.0 0.00476 0.00508,
(38)
RI
= 1.0 0.1333 0.0754 0.0495,
Cm
(39)
14
and
0
C2RI = 1.0 + 0.0 0.0101 0.0095,
(40)
RI0
Cm
(41)
= 1.0 0.1333 0.0701 0.0458.
RI at this scale
One observes that the sizes of the NNLO and NNNLO contributions to Cm
amount to about 7.5% and 5% respectively. This shows that perturbation theory can not be
used for a precise conversion of the RI quark masses to the MS ones at the renormalization
scale = 2 GeV. The convergence can be improved if one increases to, say, 3 GeV.
Indeed, with this choice of the standard three-loop evolution gives s (3 GeV) = 0.262
and Eqs. (38), (39) transform to
C2RI = 1.0 + 0.0 0.00333 0.00296
(42)
RI
= 1.0 0.111 0.0526 0.0289.
Cm
(43)
and
The accuracy of the massless approximation can be tested by computing the ratio
C?RI /C?MOM (? = m, 2) as a series in z = m2q /(2 ). Using the results of the previous
section one obtains
C2RI
C2MOM
=1

+ as2 0.28981 z 0.89236 z2 + 1.5284 z3 3.3649 z4 + 7.6945 z5
+ 0.25 zlz 0.39583 z2lz + 0.45602 z3lz 2.2796 z4lz
+ 23.231 z5lz + 0.024306 z3lz2 + 1.1007 z4lz2 8.8726 z5lz2

+ as3 4.4496 z 20.979 z2 + 2.5931 zlz 3.0558 z2lz

+ 0.70052 zlz2 0.49414 z2lz2 ,
(44)
RI
Cm
=1
MOM
Cm

+ as 1.0 z 0.5 z2 + 0.16667 z3 0.083333 z4 + 0.05 z5 1.0 zlz

+ as2 7.6458 z + 0.86458 z2 3.5129 z3 + 13.913 z4 36.828 z5
6.3264 zlz + 0.10417 z2lz + 2.3866 z3lz + 7.3259 z4lz
71.347 z5lz 2.0 zlz2 + 1.0208 z2lz2 + 1.1956 z3lz2

6.706 z4lz2 + 30.991 z5lz2

+ as3 59.008 z + 48.743 z2 41.464 zlz 5.4235 z2lz

18.829 zlz2 + 9.1024 z2lz2 4.0556 zlz3 + 3.5697 z2lz3 ,
(45)
and we have evaluated the coefficients in the series in z with

where lz =
nf = 4.
To illustrate the quality of these expansions, we have plotted (see Figs. 1 and 2) the ratio
MOM and C RI as functions of 1/z = 2 /m2 in the
of the 1, 2 and 3 loop coefficients of Cm
m
log(m2 /2 )
15
MOM /C RI as functions of
Fig. 1. The ratio of the RI and MOM scheme conversion functions C2,l
2,l
2
2
/m . Shown are the coefficients in the expansion in as / as expansions to order z to z5 (z2
for 3 loops). Note that in Landau gauge C2? = 0, for 2 loops some numeric values for the exact mass
dependence are shown as well.
Landau gauge and for simplicity with nf = 4 for all values of z. The circles in the plots for
1 and 2 loops correspond to the exact results from [43]. The convergence of the small mass
(corresponding to large negative values of 1/z) expansions is good for 1/z < 4, where the
expansions for higher orders of z are almost indistinguishable as well among each other as
well as from the numbers received from the exact 2 loop propagator. On the other hand, due
to the z ln(z)i , i = 1, . . . , l (where l is the number of loops) terms, the MOM coefficients
are approaching the corresponding values in the RI-scheme for increasing 1/|z| only very
slowly. This makes the RI-scheme as an approximation to the MOM-scheme for the c
quark useless.
4.2. Four loop quark anomalous dimensions
We start from the MS scheme. The quark mass anomalous dimension was computed at
four loops quite recently in [56,57] and reads
X
m(i) asi+1 ,
m (as )
i>0
m(0)
= 1,

1 202
20
+ nf
,
m(1) =
16 3
9

1
2216 160
140
1249 + nf
3 + n2f
,
m(2) =
64
27
3
81
m(3)
1
=
256

4603055 135680
+
3 8800 5
162
27
(46)
(47)
(48)
16
MOM /C RI as functions of
Fig. 2. The ratio of the RI and MOM scheme conversion functions Cm,l
m,l
2 /m2 . Shown are the coefficients in the expansion in as / as expansions to order z to z5 (z2 for
3 loops). For 1 and 2 loops some numeric values for the exact mass dependence are shown as well.

18400
91723 34192
3 + 880 4 +
5
+ nf
27
9
9

160
5242 800
332 64
+
3
4 + n3f
+
3 .
+ n2f
243
9
3
243 27
(49)
The result for the quark field anomalous dimension was found by one of the authors in
the course of computing m [56] and read for the QCD case in Landau gauge 5
X (i)
2 (as )
2 asi+1 ;
i>0
5 is gauge dependent and in the Landau gauge (0) = 0; the results for SU(N ) group in general covariant
2
2
gauge are given in Appendix C.

(1)
2 =

1 67
4
+ nf
,
16 3
3

(2)
2
1
=
64
(3)
2
1
=
256
17
(50)

20729 79
550
2 20
3 + nf
+ nf
,
36
2
9
27

2109389 565939
2607
761525
3 +
4
5
162
324
4
1296

79
160
162103 2291
3
4
5
+ nf
81
27
2
3

3853 160
140
+
3 + n3f
.
+ n2f
81
9
243
(51)

(52)
For completeness we also give the field anomalous dimension for the case of QED with
nf different fermion species: 6
1
2QED (0) = [L ],
4

3
1
QED (1)
+ nf [2] ,
=
2
16
2

1 3
QED (2)
2 20
+ nf [3] + nf
=
,
2
64 2
9
QED (3)
2
1
=
256
(53)
(54)
(55)

1027
460
400 3 + 640 5 + nf
64 3
8
3

304
280
32 3 + n3f
.
+ n2f
9
81
(56)
In order to compute the corresponding anomalous dimensions for the RI and RI0
schemes, one just needs to make use of Eqs. (46) to (49) in combination with the three
loop conversion functions (34,37). As a result for the QCD case we get for the RI scheme:
mRI(0) = 1,
(57)

1
52
126 + nf
,
mRI(1) =
16
9
mRI(2) =
mRI(3)
1
64

1
=
256
(58)

20911 3344
18386 128
928
3 + nf
+
3 + n2f
,
3
3
27
9
81

300665987 15000871
6160
3 +
5
648
108
3
(59)
6 For the QED case, all gauge dependence is in (0) , where is defined in Appendix A ( = 0 for Landau
L
L
2
gauge).
18

4160
7535473 627127
+
3 +
5
+ nf
108
54
3

6416
670948
18832
3 + n3f
.
+ n2f
243
27
729
(60)
The corresponding equations for the RI0 scheme are:

0
mRI (0) = 1,
0
mRI (1) =
0
mRI (2)

1
52
126 + nf
,
16
9
1
=
64
mRI (3) =
(61)

1
256

20174 3344
17588 128
2 856
3 + nf
+
3 + nf
,
3
3
27
9
81
(62)
(63)

6160
141825253 7230017
3 +
5
324
54
3

4160
3519059 298241
+
3 +
5
+ nf
54
27
3

611152 5984
16024
3 + n3f
.
+ n2f
243
27
729
(64)
The quark field anomalous dimensions for the RI and RI0 schemes are given in
Appendix C.
Common to all our results for the quark mass and field anomalous dimensions in the RI
and RI0 schemes (see Eqs. (64) and Appendix C) is the absence of 4 even at the four-loop
level. This is in striking contrast to the behavior of their MS counterparts and calls for an
explanation. In fact, a similar absence of the constant 4 at four loops in the gauge beta
functions recently has been neatly explained in [59].
Unfortunately, we have not been able to extend the argument of [59] to the present case.
Nevertheless, we tend to agree with David Broadhurst and the referee in suggesting that
the appearance of 4 in Eqs. (49), (52) is apparently an artifact of the MS-scheme.
To support this idea we demonstrate below that, in a sense, the RI/RI0 schemes are
more physical than the MS one. To illustrate this statement, let us define the q-dependent
b RI
b RI (s , 2 /(q 2 )), with ? = 2 or m, of the conversion functions C
generalization C
?
by the same Eqs. (19), (20) but without the condition q 2 = 2 . Let us in addition define

RI
b? s , 2 /(q 2 )
?RI s , 2 /(q 2) = 2 2 log C
RI
b? s , 2 /(q 2 ) .
= q 2 2 log C
q
Using these definitions, the following relations hold:
Boundary condition:
b?RI (s , 1) = C?RI (s ).
C
(65)
(66)
19
Evolution equations (they follow directly from the evolution equation of the fermion
propagator):
2
d b RI
b?RI ,
C = ? C
d2 ?
d RI
= 0,
d2 ?
or, equivalently,

b RI
b?RI ,
C? = ? C
2 2 +
as

2 2 +
?RI = 0.
as
(67)
(68)
Relations between the ?RI and the ?RI (can be received by combining Eqs. (67), (68)
and (13), (14):
?RI (s ) = ?RI (s , 1).
(69)
In fact, the functions ?RI happen to be both scale and scheme independent. In Eqs. (19)
(20) it is understood that the fermion propagator is defined in the MS-scheme. However,
one can easily check that even if the fermion propagator would be defined in any other
b RI would amount to a
(mass-independent) scheme, then the resulting change of the C
?
rescaling by a q-independent factor which can not, obviously, change the very functions
?RI .
We conclude that the ?RI are physical scheme invariant quantities, at least within the
framework of perturbation theory. Thus, due to the relation (69), the absence of 4 in the
quark mass and field anomalous dimensions in the RI scheme should be considered as a
phenomenon which is no more puzzling than the similar absence 7 of 4 in the four-loop
total cross-section of e+ e annihilation into hadrons calculated in massless QCD [6062].
The same reasoning is fully applicable to the case of the RI0 scheme.
4.3. NNNLO relation for the RI quark mass and the RGI mass m
bq
It is customary to solve the RG equation (6) for the quark running mass m() as follows
c(as ())
m()
=
,
m(0 ) c(as (0 ))
(70)
where (x stands for either s ()/ or s (0 )/ )

)
(Zx
0)
(x
m
dx 0
c(x) = exp
(x 0 )

= (x)0 1 + 1 1 0 x
2

1
2
1 1 0 + 2 + 1 0 1 1 2 0 x 2
2
3 1

1
2
1 1 0 + 1 1 0 2 + 1 0 1 1 2 0
+
6
2
7 We do not count the well-understood 2 -term arising due to the analytical continuation to the physical region
of energies.
20

1
3
2
3 1 0 + 21 2 0 3 0 + 1 1 2 1 1 2 x 3
3
)
+ O(x 4 ) .
(71)
Here i = m(i) /0 , i = i /0 (i = 1, 2, 3) and i are the coefficients of the QCD betafunction as defined in Eq. (5). The four-loop beta-function recently has been computed in
[58] with the result

1
1
2
38
11 nf ,
102 nf ,
1 =
0 =
4
3
16
3

1 2857 5033
325 2
nf +
n ,
2 =
64
2
18
54 f
(72)

1078361 6508
1 149753
+ 3564 (3)
+
(3) nf
3 =
256
6
162
27

50065 6472
1093 3
2
+
(3) nf +
n .
+
162
81
729 f
Eq. (70) directly leads to the RG invariant independent mass m
bq
m
bq =
mq ()
.
c(as ())
(73)
An important property of m
bq is its and scheme independence. The latter follows from
the fact that m
bq could be alternatively defined as follows

s ()
m
bq = lim mq ()
m0 /0
,
(74)
and from the well-known universality of the one loop coefficients of the quark mass
anomalous dimension and the -function.
Evaluating the four loop approximation of the c-function in the RI and RI0 schemes, we
can state our results for the conversion functions as a relation between the RG invariant
0
mass m
b and the masses mRI and mRI . For nf = 3, 4, 5 Eq. (73) assumes the form:
4/9

2

s
722
s
m
b (3)
s
2521517 536
+
3
=
+
1+
mRI
4
81
4
13122
9

3
18640
88484924345 3089567
s
+
3
5 , (75)
+
4
12754584
972
81
m
b (4)
=
mRI
12/25

17606
s
1+
4
1875
2

s
3819632767 952
+
3
+
4
21093750
15
s
+
4
m
b (5)
=
mRI
3
8512503162869851
1423828125000

1035345331
3 304 5 ,
+
337500
12/23

15926
s
1+
4
1587

2
2559841211 4696
s
+
3
+
4
15111414
69
3
4334826270205387
s
+
4
863345304648

26960
3889063057
3
5 .
+
1314036
69
21
(76)
(77)
5. Conclusions
In this paper we have analytically computed the first few terms of the high-energy
expansion of the three-loop quark propagator. These results have been used to find the
NNNLO conversion factors transforming the MS quark mass and the renormalized quark
field to those defined in the RI scheme which is more suitable for lattice QCD calculations.
The newly computed NNNLO corrections are numerically significant and should be taken
into account when transforming the RI quark masses to the MS ones.
We also have presented the four loop results for the quark mass and wave function
anomalous dimensions in the RI and RI0 schemes. Unlike the case of MS-scheme, the
results display a striking absence of the 4 irrational constant even at four loops. This
could be attributed to the fact that the RI/RI0 quark mass and field anomalous dimensions
could be defined in a scheme-invariant way (see subsection (4.2) for more details).
In principle, the knowledge of N4 LO conversion factors would be useful to even
better control the convergence of the perturbation series. Unfortunately, such a calculation
requires the knowledge of the quark propagator at four loops a problem certainly out of
the range of present calculational techniques.
Acknowledgements
We would like to thank Oleg Veretin for his valuable comments on the work [43] as
well as for his help in the numerical evaluation of the two loop fermion propagator. One
of us (K.G.Ch.) is grateful to Vladimir Smirnov and Damir Becirevic for discussions
and advice. A.R. would like to thank Robert Harlander and Thorsten Seidensticker for
help and advice. We thank David Broadhurst and the referee who both have drawn our
attention to the absence of 4 in the quark mass and field anomalous dimensions in the
22
RI and RI0 schemes. This work was supported by DFG under contract Ku 502/8-1 (DFGForschergruppe Quantenfeldtheorie, Computeralgebra und Monte-Carlo-Simulationen)
and the Graduiertenkolleg Elementarteilchenphysik an Beschleunigern.
Note 1
In [63] the NNNLO MS RI conversion relations have been used to transform the
lattice results for the RI light quark masses into those for the MS ones. In [64] the results
for the (QED) fermion mass and field anomalous dimensions (Eqs. (54)(56) and Eqs. (7)
(9) of [56]) have been reproduced within an entirely different approach.
Appendix A. Quark propagators

Below we list the full three loop results for the quark propagator 8 computed in general
covariant gauge with the tree gluon propagator

1
g (1 L )q q /q 2 .
2
q
For SU(N) gauge group colour factors have the values CA = N , CF = (N 2 1)/(2N)
and T = 1/2. The QED case is obtained with substitutions CA 0, CF 1 and T 1.

3
1
1
(A.1)
S(1) = CF 1 L + lq + L lq ,
2
4
4

1
1 2 3
5
1
13 3
(2)
L + lq + L lq + L2 lq
S = CF2 3 L
16 4
2
16
4
8
8

9 2
3
1
2
l
L lq
2 l2
32 q 16
32 L q

5
3
15 2 445
1531 21
+
3 L +
3 L
+
lq
+ CF CA
384
16
8
16
128 L 192

5
5 2
11 2
3
1 2 2
2
L lq +
L lq
lq
L lq
L lq
+
16
64
32
64
64

1 2
13 2
lq + lq ,
(A.2)
+ CF T nf
12 3
8

15
29
21
1 2
3
229 19
(3)
3
3 +
5
L
3 L
L +
3 L2
S = CF
48
32
8
64
32
16
32
1
105
9
37
3
3 L3 +
lq +
3 lq +
L lq +
3 L lq
+
96
64
16
64
16
11 2
1 3
9 2
21
1
2
2
L lq +
L lq
lq
L lq
+
L2 lq
64
64
32
64
8

1 3 2
9 3
9
3 2 3
1 3 3
3
L lq +
lq +
L lq
L lq +
L lq
+
64
128
128
128
384
8 For the exact definition of the s see Eqs. (2) and (28).
23

3
5
1861
43
3005 313
3
4 + 5
L +
3 L
+ CF2 CA
1152
96
32
8
768
64
5
35 2
5 3
1
2467
5 L
L
L
3 L3 +
lq
16
128
256
64
512
37
4511
15
221 2
3 lq +
L lq
3 L lq +
lq
+
64
1536
32
512 L
3
27 3
533 2
793
2
3 L2 lq +
lq
l
L lq
64
512 L
256 q 768
45 2 2
7 3 2
33 3
31
3
l
l +
l +
L lq
256 L q 256 L q 128 q 256

3 2 3
1 3 3
L lq +
l
+
128
256 L q

3
23
1
167
4699 11
3 4 +
L + 3 L
lq
+ CF2 T nf
1152 12
8
48
8
64
1
29
23 2
11
3 3
2
L lq +
lq +
L lq
lq
+ 3 lq
4
48
32
48
32

1
3
L lq
32

69
405
66301
4315565 20305
+
3 +
4
5
L
+ CF CA2
248832
2304
1024
256
36864
193
3
5
1377 2
3 L
4 L
5 L
+
256
512
128
4096 L
3
3
5
191 3
3 L2
4 L2
5 L2
+
32
1024
256
3072 L
1
294793
1301
6509
3 L3 +
lq
3 lq +
L lq
+
192
27648
512
6144
59
467 2
9
43 3
3 L lq +
lq
3 L2 lq +
lq
256
2048 L
512
1024 L
5507 2
715
61 2 2
3 3 2
2
lq
L lq
L lq
l
2304
3072
1024
256 L q

121 3
31
3 2 3
1 3 3
3
lq +
L lq
L lq +
L lq
+
+
576
1536
512
768

3
2773
3
16381 193
3 + 4 +
L
3 L
+ CF CA T nf
1944
144
8
4608
16
5081
1
251
1
lq + 3 lq
L lq +
3 L lq
864
8
768
16

887 2
25
11 3
1
2
3
lq +
L lq
lq
L lq
+
576
384
72
192

1
73
2 2
1 3
191
3 +
lq lq
lq ,
+
(A.3)
+ CF T 2 n2f
243 18
108
9
36

(1)
V
= CF

1
1
L L lq ,
4
4
(A.4)
24

3
1 2
1 2 2
5
(2)
+
lq
L lq +
L lq
V = CF2
128 32
16
32

3
13
3
9 2 25
41
3 +
L
3 L +

lq
+ CF CA
64 16
32
16
128 L 64

7
3 2
3
1 2 2
2
+
L lq
L lq +
L lq
L lq
32
64
64
64

1
7
+ CF T nf + lq ,
32 8

7
1
3
17
73
+
L
3 L3
lq +
L lq
V(3) = CF3
768 512
96
128
512

3
1 3 2
1 3 3
2
L lq
L lq
L lq
+
128
128
384

11
3
5
1
17
997
+
3 +
4
5 +
L
3 L
+ CF2 CA
1536 16
32
16
16
64
5
3 2
1 3
1
121
5 L +
L
L +
3 L3 +
lq
+
16
128
512
64
192
3
33
3
17 2
3 lq
L lq +
3 L lq
lq
16
128
64
128 L
3
11 3
11 2
25
2
3 L2 lq
lq
l +
L lq
+
64
512 L
128 q 256

17 2 2
1 3 2
3 2 3
1 3 3
l +
l
l
l
+
256 L q 64 L q 256 L q 256 L q

1
3
7
11
79
+ 3
L
lq +
L lq
+ CF2 T nf
384 4
128
96
128

1 2
1
2
l
L lq
+
32 q 32

3139
69
165
39799
2 159257
3
4 +
5 +
L
+ CF CA
41472
1536
1024
256
36864
35
3
5
787 2
3 L +
4 L +
5 L +
64
512
128
4096 L
39
3
5
55 3
3 L2 +
4 L2 +
5 L2 +
512
1024
256
1536 L
1
19979
245
4393
3 L3
lq +
3 lq
L lq
192
9216
512
6144
59
295 2
9
27 3
3 L lq
L lq +
3 L2 lq
lq
+
256
2048
512
1024 L
275 2
529
43 2 2
1 3 2
2
l +
L lq
l +
l
+
+
768 q 3072
1024 L q 128 L q
31
3 2 3
1 3 3
3
L lq
L lq
l
1536
512
768 L q
(A.5)

1723
1
11887 13
+
3
L + 3 L
+ CF CA T nf
5184
48
4608
8
191
1
175
1
lq 3 lq +
L lq
3 L lq
+
144
8
768
16
47 2
19
1
2
3
+
l
L lq
L lq
192 q 384
192

13
1 2
785
lq +
lq .
+ CF T 2 n2f
2592 72
24
25
(A.6)
Appendix B. Conversion functions

The full gauge dependent conversion factors read 9

1
RI(1)
C2
= CF L ,
8
C2RI(2)
C2RI(3)

3
19
3
3 2
57
+
3
L +
3 L
= CA CF
128 16
64
16
64 L

3 2
5
1
+
+ CF nf T
,
+ CF2
128 64 L
32
(B.1)
(B.2)

69
165
6655
457217 5543
+
3 +
4
5
L
165888 3072
1024
256
9216
221
3
5
123 2
3 L
4 L
5 L
+
512
512
128
1024 L
69
3
5
139 3
3 L2
4 L2
5 L2
+
1024
256
6144 L
1024
1
3 L3
+
192

3
5
91
25
171 19
3
4 +
5 +
L +
3 L
+ CA CF2
512 32
32
16
512
128

5
15 2
9
25 3
1
5 L +
L
3 L2 +
L
3 L3
16
128
128
1024
64

29
5
1
41
L
3 +
3 L3
+ CF3
384 1024
512 L 96

5
8449
599
3
3 +
L
3 L
+ CA CF nf T
5184 24
2304
32

1
31
15
551
3
L + CF n2f T 2
,
(B.3)
+ CF2 nf T
128 4
256
2592
= CA2 CF
9 Note that these results are expanded in a = / and not in /(4 ) as in Eqs. (34) to (37).
s
s
s
26

3
RI(1)
Cm
= CF 1 L ,
(B.4)
8

21
9 2
85 9
RI(2)
L
Cm = CA CF + 3
24 8
64
128 L

3
3
3 2
89
2 25
3 + L +
L + CF nf T
,
(B.5)
+ CF
128 4
8
32
96

15
4409
7259479 64591
RI(3)
2
+
3
5
L
Cm = CA CF
497664
9216
16
4096

165
885 2
27
81 3
2
3 L
+
3 L
+
512
4096 L 1024
2048 L

5
1771
13
21133 245
3 +
5 +
L
3 L
+ CA CF2
4608
48
16
1024
16

69 2
3
3 3
L
3 L2 +
+
256
64
64 L

15
93
3 2
3
3 3
409 29
3
2
+
3 +
5
L
+
3 L
+ CF
96
32
8
1024
32 L 32
128 L

105701 163
3
175
3
3 + 4 +
L
3 L
+ CA CF nf T
15552
144
8
512
32

263 2
3
95
1
3 4
L + 3 L
+ CF2 nf T
144 3
8
256
8

1
4459
2 2
3 ,
(B.6)
+ CF nf T
7776 18

1
RI0 (1)
= CF L ,
(B.7)
C2
4

3
13
3
9 2
41
RI0 (2)
3
L +
3 L
= CA CF +
C2
64 16
32
16
128 L

1 2
7
5
+
+ CF nf T
,
(B.8)
+ CF2
128 16 L
32

0
11
3
5
33
11
997
3
4 +
5 +
L +
3 L
C2RI (3) = CA CF2
1536 16
32
16
128
64

5
23 2
3
19 3
1
5 L +
L
3 L2 +
L
3 L3
16
128
32
512
64

69
165
159257 3139
+
3 +
4
5
+ CA2 CF
41472
1536
1024
256
39799
35
3
5
L +
3 L
4 L
5 L
36864
64
512
128
787 2
39
3
5
+
3 L2
4 L2
5 L2
4096 L 512
1024
256

55 3
1
L +
3 L3
1536
192

17
1 3
1
73
L
L +
3 L3
+ CF3
768 512
64
96

11887 13
1723
1
3 +
L 3 L
+ CA CF nf T
5184
48
4608
8

1
79
11
785
2
2 2
3
L + CF nf T
,
+ CF nf T
384 4
128
2592
27

1
RI0 (1)
= CF 1 L ,
Cm
4

7
3 2
1285 9
RI0 (2)
+ 3
L
L
= CA CF
Cm
384
8
32
64

3
1
1 2
83
19
3 + L +
L + CF nf T
,
+ CF2
128 4
4
16
96
RI0 (3)
Cm
(B.9)
(B.10)
(B.11)
15
4417
53
3360023 31193
+
3
5
L +
3 L
248832
4608
16
6144
256

295 2
9
27 3
+
3 L2
2048 L 512
1024 L

5
1771
43
18781 481
3 +
5 +
L
3 L
+ CA CF2
4608
96
16
1536
64

23 2
3
1 3

3 L2 +
+
128 L 64
32 L

15
31
3
1 2
3227 29
+
3 +
5
L
3 L
+ CF3
768
32
8
512
32
16 L

3
1 3
3 L2
L
+
32
64

95387 77
3
175
1
3 + 4 +
L
3 L
+ CA CF nf T
15552 72
8
768
16

1109 2
3
95
1
3 4
L + 3 L
+ CF2 nf T
576
3
8
384
8

1
3757
3 .
+ CF n2f T 2
7776 18
= CA2 CF
(B.12)
Appendix C. Anomalous dimensions

The quark mass and field anomalous dimensions for general SU(N) and for the MS, RI
and RI0 schemes are listed below. See Eq. (46) for the conventions. For the MS case also
28
the field anomalous dimension 2 is given for general gauge and SU(N), while for the RI
(0)
and RI0 only the Landau-gauge formulae are given (for which all 2 are 0):

N2 1 3
(0)
,
(C.1)
m =
N
8
m(1) =
N2 1
16N 2
m(2) =
N2 1
64N 3
m(3)
N2 1
=
256N 4

3 203 2
5
N + nf N ,
+
8
24
6
(C.2)

129 129 2 11413 4
N +
N
16
16
216

1177 3
23
N
N 6N 3 6N 3 3
+ nf
4
108

35
+ n2f N 2 ,
54
(C.3)

1261 50047 2 66577 4 460151 6

+
N
N +
N + 21 3
128
384
1152
1152

47 2
1157 6
4
4
6
N 3 110N 5 110N 5
N 3 + 52N 3 +
2
18

10475 3 11908 5 111
37
N+
N
N
N 3
+ nf
6
216
81
2
889 5
N 3 + 33N 3 4 + 33N 5 4
85N 3 3
6

30N 5 + 50N 3 5 + 80N 5 5

899 4
19
N + 10N 2 3 + 10N 4 3 6N 2 4
N2 +
27
324

4
6N 4

83 3 8 3
3
N + N 3 ,
+ nf
162
9
+ n2f
mRI(0) =

N2 1 3
,
2N
4
mRI(1) =
N2 1
16N 2
mRI(2)
N2 1
=
64N 3

(C.5)

3 379 2
13
N + nf N ,
+
8
24
6
126239 4
129
17N 2 22 3N 2 +
N 44 3N 4
16
432

18611 3
75
3
N 2 3 N
N + 2 3 N
+ nf
8
216
(C.4)
(C.6)


+ n2f
mRI(3)
N2 1
=
256N 4

116 2
N
27
29

,
(C.7)
1261 198679 2 778421 4 202580155 6

+
N
N +
N
128
384
1152
31104
149 2
4133 4
59269 6
N 3
N 3
N 3
+ 21 3
4
6
32

165N 2 5 + 275N 4 5

206575 3 7671073 5
3175
N+
N
N
+ nf
48
432
2592
22
7767 5
N 3
59N 3 N 3 3 +
3
16
20N 3 5 + 60N 5 5

5767 2 113747 4 62 2
4
N +
N + N 3 32N 3
108
324
3

2354
N3 ,
+ n3f
243
+ n2f
mRI (0) =
0
mRI (1)
N2 1
=
16N 2
mRI (2) =
0
mRI (3)

N2 1 3
,
2N
4
N2 1
64N 3
N2 1
=
256N 4
(C.9)

3 379 2
13
N + nf N ,
+
8
24
6

121883 4
129 147 2
N 22 3N 2 +
N 44 3N 4
16
8
432

17819 3
77
N 2 3 N
N + 2 3 N 3
+ nf
8
216

107
N2 ,
+ n2f
27

(C.8)
(C.10)

198283 2 149
1261
+ 21 3 +
N
3 N 2 165 5N 2
128
384
4
844829 4 2017
N
3 N 4 + 275 5N 4
1152
3

191436121 6 28551
6
N
3 N
+
31104
16

211669 3 31
199
N 59 3N +
N
3 N 3
+ nf
3
432
3
(C.11)
30

1794277 5 3697
N +
3 N 5 + 60 5N 5
648
8

62
25946
1444
N2 +
3 N 2 +
N 4 30 3N 4
+ n2f
27
3
81

2003
N3 ,
(C.12)
+ n3f
243

N2 1 1
L ,
(C.13)
N
8

N 2 1 3 11 2
1 2 2
1
2
+ N + N L + N L + nf N ,
(C.14)
8
4
8
2
16N 2

137 2 6635 4 263 4
N2 1
39
3
+
N +
N +
N L + N 4 L2
3
16
16
288
64
64
64N

5 4 3
21
3
3 4 2
2
4
4
+ N L 3N 3 N 3 + N L 3 + N L 3
32
16
8
16

547 3 17 3
3
5
N N L + n2f
N2 ,
(C.15)
+ nf N
8
72
16
18

N 2 1 1027 11281 2 86017 4 1785121 6
+
N +
N +
N
256N 4
128
384
1152
10368
5
1644899 6
6467 6 2
N L +
N L
N 4 L +
8
62208
2304
149 6 3
19 6 4
N L +
N L + 25 3 + 31N 2 3
+
192
128
4287 4
1103 6
115 4
N 3
N 3 +
N L 3
64
64
16
2761 6
11
289 6 2
N L 3 + N 4 L2 3 +
N L 3
+
192
32
96
3
19
7
N 4 L3 3 + N 6 L3 3 N 4 L4 3
16
64
64
1
33
231 6
97
N 4 N 6 L 4
N 6 L4 3 + N 4 4 +
64
2
32
64
17
3
N 6 L2 4 N 6 L3 4 40 5 60N 2 5
32
64
1945 4
1375 6
65
N 5
N 5 N 4 L 5
+
64
128
8
95 6
35 4 2
75 6 2
N L 5 N L 5 N L 5
8
32
64
5 4 4
5 6 4
N L 5
+ N L 5 +
64
128

1555 3 35641 5 767 3
307
N
N
N +
N L
+ nf
12
144
432
96
20 5N 3
(0)
2 =
2(1) =
(2)
2 =
(3)
2 =
31
161347 5
109 5 2
N L
N L 4N 3 + 8N 3 3
15552
288
35
11
15
N 5 3 N 3 L 3 N 5 L 3
8
2
4
7 5 2
21 5
3
3
N L 3 3N 4 N 4 N 3 L 4
24
16
2

1 5
1 5 2
3
N L 4 + N L 4 20N 5
2
16

445
269 4
19
N4
N L + 2N 2 3 + 2N 4 3
+ n2f N 2 +
9
72
486

2 4
3 35
3
N
+ N L 3 + nf
,
3
162
RI(1)
N2 1
16N 2
RI(2)
N2 1
64N 3
2RI(3)
RI0 (1)
N2 1
=
256N 4

3 11 2
1
+ N + nf N ,
8
4
2
(C.16)
(C.17)

25
14225 4
197 4
3
+ N2 +
N 3N 2 3
N 3
16
3
288
16

611 3
1
10 2
N + 2N 3 3 + n2f
N
,
+ nf N
3
36
9
(C.18)

N2 1
16N 2
RI0 (2)
2
N2 1
=
64N 3
1027 7673 2 174565 4 3993865 6

+
N +
N +
N
128
384
1152
3456
10975 4
111719 6
N 3
N 3
+ 25 3 + 31N 2 3
64
192

5465 4
20625 6
2
N 5 +
N 5
40 5 60N 5 +
64
128

557 3 172793 5
1307
N+
N
N
+ nf
48
144
288
7861 5
N 3
4N 3 + 2N 3 3 +
48

125 5
N 5
30N 3 5
4

26
521 2 259 4
N +
N + 6N 2 3 N 4 3
+ n2f
72
3
3

86
+ n3f N 3 ,
27

3 11 2
1
+ N + nf N ,
8
4
2
233 2 17129 4
197 4
3
+
N +
N 3N 2 3
N 3
16
24
288
16
(C.19)
(C.20)

32

743 3
7
N + 2N 3 3
+ nf N
12
36

13
N2 ,
+ n2f
9
RI0 (3)
2
N2 1
=
256N 4
(C.21)

1027 8069 2 240973 4 5232091 6

+
N +
N +
N
128
384
1152
3456
12031 4
124721 6
N 3
N 3
+ 25 3 + 31N 2 3
64
192

5465 4
20625 6
N 5 +
N 5
40 5 60N 2 5 +
64
128

1141 3 113839 5
329
N
N
N
+ nf
12
144
144
2245 5
N 3
4N 3 + 5N 3 3 +
12

125 5
N 5
30N 3 5
4

32
515 2 1405 4
N +
N + 6N 2 3 N 4 3
+ n2f
72
12
3

125
N3 .
+ n3f
27
(C.22)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
G. t Hooft, Nucl. Phys. B 61 (1973) 455.

W.A. Bardeen, A.J. Buras, D.W. Duke, T. Muta, Phys. Rev. D 18 (1978) 3998.
K.G. Chetyrkin, J.H. Khn, A. Kwiatkowski, Phys. Rep. 277 (189) (1996).
J. Prades, Nucl. Phys. Proc. Suppl. 64 (1998) 253, hep-ph/9708395.
M. Jamin, Nucl. Phys. Proc. Suppl. 64 (1998) 250, hep-ph/9709484.
C.A. Dominguez, L. Pirovano, K. Schilcher, Nucl. Phys. Proc. Suppl. 74 (1999) 313, hepph/9809338.
S. Narison, hep-ph/9905264.
K.G. Chetyrkin, J.H. Kuhn, A.A. Pivovarov, Nucl. Phys. B 533 (1998) 473, hep-ph/9805335.
J. Prades, A. Pich, Nucl. Phys. Proc. Suppl. 74 (1999) 309, hep-ph/9811263.
V. Lubicz, Nucl. Phys. Proc. Suppl. 74 (1999) 291, hep-ph/9809417.
R.D. Kenway, Nucl. Phys. Proc. Suppl. 73 (1999) 16, hep-lat/9810054.
V. Gimenez, Nucl. Phys. Proc. Suppl. 74 (1999) 296, hep-ph/9810532.
S.R. Sharpe, hep-lat/9811006.
H. Leutwyler, Nucl. Phys. B 337 (1990) 108, and references therein; Phys. Lett. B 378 (1996)
313.
M. Luscher, hep-lat/9802029.
G. Martinelli, C. Pittori, C.T. Sachrajda, M. Testa, A. Vladikas, Nucl. Phys. B 445 (1995) 81,
hep-lat/9411010.
S. Capitani et al., Nucl. Phys. Proc. Suppl. 63 (1997) 153, hep-lat/9709125.
[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]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
33
S. Sint, P. Weisz, ALPHA Collaboration, Nucl. Phys. B 545 (1999) 529, hep-lat/9808013.
R. Gupta, T. Bhattacharya, Phys. Rev. D 55 (1997) 7203, hep-lat/9605039.
B.J. Gough et al., Phys. Rev. Lett. 79 (1997) 1622, hep-ph/9610223.
C.R. Allton, V. Gimenez, L. Giusti, F. Rapuano, Nucl. Phys. B 489 (1997) 427, hep-lat/9611021.
N. Eicker et al., SESAM Collaboration, Phys. Lett. B 407 (1997) 290, hep-lat/9704019; Phys.
Rev. D 59 (1999) 014509, hep-lat/9806027.
A. Cucchieri, M. Masetti, T. Mendes, R. Petronzio, Phys. Lett. B 422 (1998) 212, heplat/9711040.
M. Gockeler, R. Horsley, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, P. Stephenson, Phys.
Rev. D 57 (1998) 5562, hep-lat/9707021.
V. Gimenez, L. Giusti, F. Rapuano, M. Talevi, Nucl. Phys. B 540 (1999) 472, hep-lat/9801028.
D. Becirevic, Ph. Boucaud, J.P. Leroy, V. Lubicz, G. Martinelli, F. Mescia, Phys. Lett. B 444
(1998) 401, hep-lat/9807046.
J. Garden et al., ALPHA and UKQCD Collaboration, DESY-99-075, hep-lat/9906013.
M. Gockeler, R. Horsley, H. Oerlich, D. Petters, D. Pleiter, P. Rakow, G. Schierholz,
P. Stephenson, DESY 99-097, hep-lat/9908005.
S. Aoki et al., JLQCD Collaboration, Phys. Rev. Lett. 82 (1999) 43924395, hep-lat/9901019.
S. Aoki et al., CP-PACS Collaboration, UTCCP-P-65, hep-lat/9904012.
T. Blum, A. Soni, M. Wingate, BNL-HET-99-2, to appear in Phys. Rev. D, hep-lat/9902016.
E. Franco, V. Lubicz, Nucl. Phys. B 531 (1998) 641, hep-ph/9803491.
G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.
C.G. Bollini, J.J. Giambiagi, Phys. Lett. B 40 (1972) 566.
G.M. Cicuta, E. Montaldi, Nuovo Cimento Lett. 4 (1972) 329.
J.F. Ashmore, Nuovo Cimento Lett. 4 (1972) 289.
P. Breitenlohner, D. Maison, Commun. Math. Phys. 52 (1977) 11.
P. Breitenlohner, D. Maison, Commun. Math. Phys. 52 (1977) 55.
J. Lowenstein, W. Zimmermann, M. Weinstein, Phys. Rev. D 10 (1974) 2500.
J. Lowenstein, M. Weinstein, W. Zimmermann, Phys. Rev. D 10 (1974) 1854.
S. Weinberg, Phys. Rev. D 8 (1973) 3497.
M. Gockeler et al., Nucl. Phys. B 544 (1999) 699, hep-lat/9807044.
J. Fleischer, F. Jegerlehner, O.V. Tarasov, O.L. Veretin, Nucl. Phys. B 539 (1998) 671, hepph/9803493.
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
J.H. Kuhn, hep-ph/9901330, A recent review of expansion techniques and physical results
obtained with their help.
K.G. Chetyrkin, J.H. Kuhn, Nucl. Phys. B 432 (1994) 337, hep-ph/9406299.
K.G. Chetyrkin, R. Harlander, J.H. Kuhn, M. Steinhauser, Nucl. Phys. B 503 (1997) 339, hepph/9704222.
R. Harlander, M. Steinhauser, Phys. Rev. D 56 (1997) 3980, hep-ph/9704436.
R. Harlander, M. Steinhauser, Eur. Phys. J. C 2 (1997) 151, hep-ph/9710413.
P. Nogueira, J. Comp. Phys. 105 (1993) 279.
R. Harlander, PhD Thesis, Shaker Verlag, Aachen, 1998, ISBN 3-8265-4545-1.
S.A. Larin, F.V. Tkachov, J.A.M. Vermaseren, NIKHEF-H/91-18, Amsterdam, 1991.
M. Steinhauser, PhD Thesis, Shaker Verlag, Aachen, 1996, ISBN 3-8265-1680-X.
R. Harlander, M. Steinhauser, Progr. Part. Nucl. Phys. 43 (1999) 167, hep-ph/9812357.
J.A. Vermaseren, private communication.
K.G. Chetyrkin, Phys. Lett. B 404 (1997) 161, hep-ph/9703278.
J.A. Vermaseren, S.A. Larin, T. van Ritbergen, Phys. Lett. B 405 (1997) 327, hep-ph/9703284.
T. van Ritbergen, J.A. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379, hep-ph/9701390.
D.J. Broadhurst, hep-th/9909185.
S.G. Gorishny, A.L. Kataev, S.A. Larin, Phys. Lett. B 259 (1991) 144.
34
[61] L.R. Surguladze, M.A. Samuel, Phys. Rev. Lett. 66 (1991) 560; Phys. Rev. Lett. 66 (1991) 2416
(Erratum).
[62] K.G. Chetyrkin, Phys. Lett. B 391 (1997) 402.
[63] D. Becirevic, V. Gimenez, V. Lubicz, G. Martinelli, Phys. Rev. D 61 (2000) 114507, heplat/9909082.
[64] D.J. Broadhurst, Phys. Lett. B 466 (1999) 319, hep-ph/9909336.

Radiative symmetry breaking in brane models

I. Antoniadis a , K. Benakli b, M. Quirs c,d
a Centre de Physique Thorique, Ecole Polytechnique, 91128 Palaiseau, France 1
b CERN Theory Division CH-1211, Genve 23, Switzerland
c Laboratoire de Physique Thorique, ENS, 24 rue Lhomond, F-75231, Paris, France 2
d Instituto de Estructura de la Materia (CSIC), Serrano 123, E-28006 Madrid, Spain
Received 12 April 2000; accepted 25 May 2000
Abstract
We propose a way to generate the electroweak symmetry breaking radiatively in non-supersymmetric type I models with string scale in the TeV region. By identifying the Higgs field with a
tree-level massless open string state, we find that a negative squared mass term can be generated at
one loop. It is finite, computable and typically a loop factor smaller than the string scale, that acts as
an ultraviolet cutoff in the effective field theory. When the Higgs open string has both ends confined
on our world brane, its mass is predicted to be around 120 GeV, i.e., that of the lightest Higgs in the
minimal supersymmetric model for large tan and mA . Moreover, the string scale turns out to be
one to two orders of magnitude higher than the weak scale. We also discuss possible effects of higher
order string threshold corrections that might increase the string scale and the Higgs mass. 2000
Elsevier Science B.V. All rights reserved.
1. Introduction
Following the recent understanding of string theory, the string scale, Ms , is not tied to the
Planck mass but corresponds to an independent arbitrary parameter [112], restricted by
present experimental data to be Ms & 1 TeV [1315]. 3 Therefore a non-supersymmetric
string model with a string scale in the TeV range provides a natural solution, alternative
to supersymmetry, to the gauge hierarchy problem [46]. For such models an important
question is to understand the origin of electroweak symmetry breaking, and explain the
mild hierarchy between the weak and string scales. In string models all tree-level masses
are fixed by the string scale, except for flat directions that give arbitrary masses to the
fields that couple to them. This implies that electroweak symmetry breaking should occur
karim.benakli@cern.ch
1 Unit mixte du CNRS et de lEP, UMR 7644.
2 Unit mixte du CNRS et de lENS, UMR 8549.
3 For a recent analysis, see Ref. [16].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 5 7 - 6
36
I. Antoniadis et al. / Nuclear Physics B 583 (2000) 3548
radiatively in two possible ways: (a) If the Higgs corresponds to a massless field with a
quartic tree-level potential, and a negative squared mass is generated by string one-loop
radiative corrections which are not protected by supersymmetry. (b) If the Higgs vacuum
expectation value (VEV) is classically undetermined by a flat direction which is lifted
radiatively and fixed at a local minimum of the effective potential.
In this letter we study these issues in the context of type I string models possessing nonsupersymmetric brane configurations [1720]. We will first present a one-loop computation
of the effective potential in the presence of a Wilson-line background that corresponds
to a classically flat direction. We will show that the resulting potential has a non-trivial
minimum which fixes the VEV of the Wilson line or, equivalently, the distance between
the branes in the T -dual picture. Although the obtained VEV is of the order of the string
scale, the potential provides a negative squared-mass term when expanded around the
origin. Next we discuss models, obtained by orbifolding the previous example, where the
Wilson line is projected away from the spectrum while keeping charged massless fields
with quartic tree-level terms. These fields acquire one-loop negative squared masses, that
can be computed using the previous calculation. By identifying them with the Higgs field
we can achieve radiative electroweak symmetry breaking, 4 and obtain the mild hierarchy
between the weak and string scales in terms of a loop factor.
This mechanism becomes very predictive in a class of models where the Higgs field
corresponds to a charged massless excitation of an open string with both ends confined on
our world brane (analog to the untwisted states of heterotic orbifolds). In this case, the treelevel potential can be obtained by an appropriate tree-level truncation of a supersymmetric
theory leading to two predictions. On the one hand, the Higgs mass is predicted to be that
of the lightest Higgs in the minimal supersymmetric model (MSSM) for large values of
tan and mA , i.e., 120 GeV [2227]. On the other hand, the string scale is computable
and turns out to be around one to two orders of magnitude higher than the weak scale,
roughly 110 TeV. This mechanism is similar to the ColemanWeinberg idea, except that
there are no logarithms in the computation. Indeed, from the field theory point of view the
string scale provides an ultraviolet cutoff which regulates the quadratic divergence of the
Higgs mass. Finally, we discuss higher order string threshold corrections which can affect
the above results, for instance by large logarithms when there are massless bulk fields that
propagate in two large transverse dimensions [6,2830]. In this case, the string scale and
possibly the Higgs mass could be pushed up to higher values.
The reader who is not familiar with string theory could skip the following rather
technical section and go directly to Eq. (9) and Fig. 2, which provides an estimate of the
generated string one-loop mass term for a tree-level massless scalar on our world brane.
2. One-loop effective potential

Here we will consider a simple non-supersymmetric tachyon-free Z2 orientifold of type
IIB superstring compactified to four dimensions on T 4 /Z2 T 2 [17]. Cancellation of
4 For an earlier attempt to generate a non-trivial minimum of the potential, see Ref. [21].
37
branes. 5
RamondRamond charges requires the presence of 32 D9 and 32 anti-D5 (D5)
The bulk (closed strings) as well as the D9 branes are N = 2 supersymmetric while
The massless closed string
supersymmetry is broken on the world-volume of the D5s.
spectrum contains the graviton-, 19 vector- and 4 hyper-multiplets, while the massless
open string spectrum on the D9 branes contains an N = 2 vector multiplet in the adjoint of
the SO(16) SO(16) gauge group and a hypermultiplet in the (16, 16) representation.
When all D5 branes are put at the origin of T 4 , the non-supersymmetric D5 sector
contains gauge fields and complex scalars in the adjoint representation of USp(16)
USp(16) gauge group, a pair of complex scalars in the (16, 16) representation, and Dirac
fermions in the (120, 1) + (1, 120) + (16, 16) representations. Finally there are 95 strings
giving rise to complex scalars in the (16, 1; 1, 16) + (1, 16; 16, 1) together with Weyl
fermions in the (16, 1; 16, 1) + (1, 16; 1, 16) representations, with respect to SO(16)
SO(16) USp(16) USp(16). Note that the 95 spectrum is supersymmetric when D5
gauge interactions are turned off.
We will restrict ourselves to the effective potential involving the scalars of the D5 branes,
namely the adjoints and bifundamentals of the USp(16) USp(16) gauge group. The
relevant part of the one-loop partition function corresponding to 5 5 open strings is
1
V8 S8
W4 P2
A5 = (d1 + d2 )2
4
8
2
1
2 V4 O4 O4 V4 C4 C4 + S4 S4 2
+ (d1 d2 )
P2 ,
4
8
2
b b
2
b4 V
b4 C
b4 O
b4 O
b4 + C
b4 b
1
V8 + S8
S4 b
V
S4 2b
W
+
P2 ,
M5 = (d1 + d2 )
4
b
4
b
8
b
8
2
(1)
where A and M denote the contributions from the annulus and Mbius strip, respectively.
In the above equation d1 = d2 = 16, while V2n , O2n , C2n and S2n are the SO(2n)
characters,
3n 4n
,
2n
n i n n
C2n = 2 n 1 ,
2
V2n =
3n + 4n
,
2n
n + i n n
S2n = 2 n 1 ,
2
O2n =
where i are the Jacobi theta functions and the Dedekind eta function, depending on
the usual complex variable = it/2, with t being the (real) annulus parameter. In the
product of characters, the first factor stands for the contribution of spacetime and T 2
world-sheet fermions, while the second factor represents the corresponding contribution
from the internal T 4 . The hatted functions are defined by fb f ( + 1/2). Finally, P2 (W4 )
denotes the momentum (winding) lattice sum along the T2 (T4 ) torus; for one dimension,
they read:
X 2i m2 0 /R 2
X
2 2
0
k,
e
W1 ( ) =
e2i n R / ,
(2)
P1 ( ) =
m
5 In general arbitrary numbers of pairs D9+D9 and D5+D5 can also be added [1820].
38
where 0 Ms2 is the Regge slope, and Rk (R ) denotes the radius of the corresponding
dimension parallel (transverse) to the D-brane.
In both the annulus and Mbius amplitudes the first term stands for the untwisted
contribution while the second term accounts for the Z2 orbifold projection which
differentiates T 4 and T 2 contributions. Its presence is due to the non-freely action of Z2 at
the origin of T 4 and thus it depends only on the lattice of T 2 . It is obvious from Eq. (1) that
the Z2 projection acts in a supersymmetric way, and therefore the second terms containing
the twisted contribution vanish identically and will not play any role in our calculation.
In the first terms containing the untwisted contribution, V8 and S8 arise from bosons
and fermions, respectively. Here, supersymmetry is explicitly broken via the orientifold
projection realized by the Mbius amplitude. Indeed, from the change of sign of S8
between A and M, it is manifest that the orientifold projection acts in opposite ways
for bosons and fermions and breaks supersymmetry. More precisely, it symmetrizes the
bosons and antisymmetrizes the fermions in each USp(16) factor.
The tree-level scalar potential can be obtained by a truncation of an N = 2 supersymmetric theory and has flat directions corresponding to the Wilson lines a along the T 2 or
T 4 directions. For longitudinal directions they amount to shifting the momenta m m + a
in Eq. (2), while for transverse directions they shift the windings n n + a and describe
brane separation. It follows that at one-loop level the flat directions are lifted since the
Wilson lines acquire a potential from the Mbius amplitude which breaks supersymmetry.
Without loss of generality we will consider a Wilson line a along one direction of T 2 of
radius R, and treat the other, upon T-duality, on the same footing as the dimensions of T 4
with a common radius r. After transforming the amplitudes (1) in the transverse (closed
string) channel and using the standard -function Riemann identity, the one loop effective
potential for the Wilson line is given by:
1
Veff (a) =
32 4 02
Z
0
Z

24
1 R X 2 mE22 l X 4ina 2n2 R 2 l
r
dl 12 il +
e
e
e
2 r5
4
n
m
E

24
1
1 R X 2 mE22 l
r
dl
e
il
+
32 4 02
412
2 r5
m
E
0

X
2n2 R 2 l
1+2
cos(4na)e
,
(3)
n>0
where the radii R and r are defined in units of 0 .

In this setup, the canonically normalized scalar field h associated to the Wilson line a is
h = a/gR, where g is the gauge coupling, as can be easily seen by dimensional reduction.
Let us first expand the effective potential in powers of h and extract its quadratic (squared
mass) term 2 h2 /2. The result is:
g2
= 2 0
2

24
1 R 3 X 2 mE22 l X 2 2n2 R 2 l
r
dl 12 il +
e
n e
.
4
2 r5
n
m
E
(4)
39
It is easy to see that the integral converges. In fact, in the limit l the integrand falls
off exponentially, while for l 0 one can use the Poisson resummations
X 2 m2 l
r X r 2 p2
r2 =
e
e 2l ,
(5)
2l p
m

X
n2
1 X
1
2 n2
2 2n2 R 2 l
2R l
n e
=
,
(6)
e
2l
4l2
4R
4R
R
2l
n
n
and the identity

4
24
1
1
i
4 2
+
=
(2l)
,
il
+
12
2
12 4l 2
to show that the integrand goes to a constant. Moreover, 2 is negative which implies
that the origin is unstable and h must acquire a non trivial VEV breaking the gauge
symmetry. Note that the negative sign comes from the expansion of cos 4na in Eq. (3)
and is correlated with the positive sign of the contribution from the same states to the
cosmological constant. Although this seems to be a general property in these models, we
do not have a deeper understanding of the correlation between the sign of the mass term
and the (massive) spectrum of the theory.
Even if a is a periodic variable of period 1, Veff is periodic under the shift a a + 1/2,
since its contribution originates from the Mbius amplitude. Moreover, in this particular
example, the one-loop effective potential has a global minimum at a = 1/4. This follows
trivially from its expression (3), whose derivative with respect to a is a sum of terms
proportional to sin 4na, while its second derivative gives

00
Veff
a=1/4
1
=
2 2 02
Z
dl
0

24
1 R X 2 mE22 l X
2 2
r
e
()n+1 n2 e2n R l .
il
+
2 r5
412
n
m
E
(7)
Positivity of the integrand is manifest for all factors with the exception of the last sum
for which a careful analysis is required. This sum can be written as 4 ( )/2i with
= 2 iR 2 l, which can be easily shown to be a positive function.
In the T -dual picture, the VEV a = 1/4 corresponds to separating a brane at a distance
from the origin equal to half the compactification interval R. By turning on all Wilson
P
lines aI , the effective potential becomes a sum I Veff (aI ), with Veff (aI ) given in (3),
which upon minimization fixes all aI at the same value 1/4. Thus, the global minimum
of all Wilson lines corresponds to put all branes at the same point in the middle of the
compactification interval. The USp(16) USp(16) gauge group is then broken down to a
U (8) U (8) or USp(16) subgroup, corresponding to turning on Wilson lines along the
T 2 or T 4 directions, transforming in the adjoint or in the bifundamental representation,
respectively.
In order to make a numerical estimate of the results, we will consider the case of a
4-brane with five large transverse dimensions by taking the limit r and keeping the
radius R (along the 4-brane) as a parameter. To take the limit r , we use Eq. (5) for
40
Fig. 1. Effective potential as a function of a, for 2 6 R 6 3 in units of 104 Ms4 .
each of the five transverse dimensions, and note that only p = 0 contributes in the sum.
In fact, non-vanishing values of p may contribute only in the region l , in which case
the corresponding integrand in Eq. (3) vanishes as l 5/2 . It follows that in the limit r
the potential becomes:
R
Veff (a, R) =
32 4 02
Z
0

24
dl
1 X 4ina 2n2 R 2 l
e
e
.
il +
2 n
(2l)5/2 412
(8)
The effective potential (8) is plotted in Fig. 1 for the range of values of the radius 2 6 R 6 3
as a function of a inside its period (1/2, 1/2). Following our previous analysis, it has a
maximum at the origin and a minimum at a = 1/4 for any value of R. The mass term at
the origin, in the limit r and for arbitrary R, can be equally computed from Eq. (4)
using the Poisson resummation (5). The result is:
2 (R) = 2 (R) g 2 Ms2
(9)
with
1
(R) =
2 2
X

24
dl
1
2 2
n2 e2n R l .
R3
il +
5/2
12
(2l) 4
2
n
(10)
41
Fig. 2. The parameter in (10) as a function of R in 0 units.
The parameter is plotted in Fig. 2 as a function of R in a typical range 1/4 < R < 5. At
the lower end, it has almost reached its asymptotic value for R 0, 6 (0) ' 0.14, and the
effective cutoff for the mass term at the origin is Ms , as can be seen from Eq. (9). At large
R, 2 (R) falls off as 1/R 2 , which is the effective cutoff in the limit R , in agreement
with field theory results in the presence of a compactified extra dimension [11]. 7 In fact,
in the limit R an analytic approximation to (R) can be computed as,
(R) '
,
Ms R
3(5)
' 0.008,
4 4
(11)
which approximately describes Fig. 2 for large values of R.

Notice that the mass term (9) we found for the Wilson line a also applies, by gauge
invariance, to the charged massless fields which belong to the same representation.
3. Electroweak symmetry breaking

In the previous example we obtained a VEV of the order of the string scale, because
we only considered Wilson lines, which correspond to tree-level flat directions in the
Cartan subalgebra of the gauge group, and have put to zero the VEVs of all other
fields. Thus, the total potential to be minimized appeared at the one-loop level. Had
we minimized the effective potential with respect to fields charged under the Cartan
subalgebra, we would have found the same solution (which corresponds to a true minimum
in the multidimensional field space) since the charged fields acquire, from the Wilson lines,
6 This limit corresponds, upon T-duality, to a large transverse dimension of radius 1/R.
7 Actually this effect is at the origin of thermal squared masses, T 2 , in four-dimensional field theory at
finite temperature, T , where the time coordinate is compactified on a circle of inverse radius 1/R T and the
Boltzmann suppression factor generates an effective cutoff at momenta p T .
42
positive tree-level squared masses and have vanishing VEVs. In more realistic models, the
Wilson lines are at least partially projected away by an orbifold projection which also
breaks the gauge group. If the orbifold projection acts in a supersymmetric way, as was the
case of the Z2 in the previous example, the calculation of the squared mass term remains
valid for the left-over charged scalars in the spectrum, up to an overall numerical factor
given by the order of the orbifold group (1/N for a ZN orbifold). Moreover, the charged
scalars have a tree-level potential which can be obtained by an appropriate truncation,
dictated by the orbifold, of a supersymmetric theory. These two facts allow the existence
of a (local) perturbative minimum, around which higher order terms in the expansion of
the one loop potential can be neglected since the charged scalars would acquire a VEV
controlled by the quadratic terms.
We will illustrate these points within the context of the toy model described in the
previous section. The crucial property is that the bosonic sector of the non-supersymmetric
D5 branes is identical to the one of an N = 2 supersymmetric theory obtained by a Z2
orbifold projection from an N = 4 theory based on a fictitious" USp(32) gauge group.
The latter contains six adjoint scalars that can be organized in three N = 1 chiral multiplets
i with i = 1, 2, 3. Notice that in this model supersymmetry is explicitly broken because
the fermions belong to the antisymmetric instead of the adjoint (symmetric) representation
of USp(32). The Z2 projection breaks USp(32) into USp(16) USp(16) and keeps the
adjoint of USp(16) USp(16) from 1 and the (16, 16) components from 2,3 . The treelevel scalar potential can be obtained straightforwardly by a corresponding truncation of
the potential of the N = 4 theory:
!2 !
X
X
2
g2

i , i
i , j +
.
(12)
VN=4 = Tr
2
i,j
The result is identical to the potential of an N = 2 theory with USp(16) USp(16)

gauge group and one hypermultiplet in the (16,
16) representation. In N = 1 notation,
it corresponds to the superpotential W = g/ 2 2 1 3 , where 1 is the adjoint from
1 and 2,3 are the two bifundamental chiral multiplets from 2,3 . The F - and D-term
contributions to the potential come from the first and second term of Eq. (12), respectively.
As we discussed in detail after Eq. (2), the Z2 orbifold projection does not by itself
break all supersymmetries and does not play any role in the computation of the potential.
As a result, the scalar mass terms generated at one loop receive contributions only from
the untwisted sector which treats the adjoint and the (16, 16) scalars in the same way, as
an adjoint of USp(32). Thus, the generated masses of the different scalars can be obtained
from the same functional of the radii through permutations. In particular, this means that
scalars describing displacement of branes in dimensions of the same size acquire equal
masses. For instance, in the isotropic 3-brane limit of six large transverse dimensions,
r and R 0, the result (9) applies for all scalar components.
We would like now to discuss possible phenomenological applications of these results.
Let us assume that there is a sequence of supersymmetric orbifold projections that lead
to the Standard Model living on some non-supersymmetric brane configuration along the
43
line of the toy model presented above. In the minimal case, where there is only one Higgs
doublet h originating from the untwisted sector, the scalar potential would be:
2

(13)
V = h h + 2 h h ,
where arises at tree-level and is given by an appropriate truncation of a supersymmetric
theory. Within the minimal spectrum of the Standard Model, = (g22 + g 02 )/8, with g2
and g 0 the SU(2) and U (1)Y gauge couplings, as in the MSSM. On the other hand, 2 is
generated at one loop and can be estimated by Eqs.
(9) and (10).
The potential (13) has a minimum at hhi = (0, v/ 2), where v is the VEV of the neutral
component of the h doublet, fixed by v 2 = 2 /. Using the relation of v with the Z
gauge boson mass, MZ2 = (g22 + g 02 )v 2 /4, and the fact that the quartic Higgs interaction is
provided by the gauge couplings as in supersymmetric theories, one obtains for the Higgs
mass a prediction which is the MSSM value for tan and mA :
Mh = MZ .
(14)
Furthermore, one can compute Mh in terms of the string scale Ms , as Mh2 = 22 =

2 2 g 2 Ms2 , or equivalently
Mh
.
Ms =
2 g
(15)
The lowest order relations (14) and (15) receive in general two kinds of higher order
corrections. On the one hand, there might be important string corrections that we will
discuss in the next section. On the other hand, from the point of view of the effective
field theory, they are valid at the string scale Ms , and Standard Model radiative corrections
should be taken into account for scales between Ms and MZ . In particular, the tree level
Higgs mass has been shown to receive important radiative corrections from the topquark sector. For present experimental values of the top-quark mass, the Higgs mass in
Eqs. (14) and (15) is raised to values around 120 GeV [2227]. Moreover from Eq. (15),
we can compute the string scale Ms . There is a first ambiguity in the value of the gauge
coupling g at Ms , whichdepends on the details of the model. Here, we use a typical
unification value g ' 1/ 2. A second ambiguity concerns the numerical coefficient
which is in general model dependent. In our calculation, this is partly reflected in its
R-dependence, as seen in Fig. 2. Varying R from 0 to 5, that covers the whole range
of values for a transverse dimension 1 < 1/R < , as well as a reasonable range for a
longitudinal dimension 1 < R . 5, one obtains Ms ' 15 TeV. Note that in the R 1
(large longitudinal dimension) region our theory is effectively cutoff by 1/R and the Higgs
mass is then related to it by,
Mh
1
=
.
R
2 g
(16)
Using now the value for in the present model, Eq. (11), we find 1/R & 1 TeV. A further
model dependence of comes from the order of the orbifold group. As mentioned above,
had we considered a higher order orbifold, e.g.,
Z2N instead of Z2 as required by more
realistic models, would decrease by a factor N . As a result, the radiative electroweak
44
symmetry breaking can be consistent with a string scale as heavy as O(10) TeV, and a
compactification scale 1/R & 2 TeV.
In a more general context, the Higgs sector may be more complicated and the
scalar potential could have classically undetermined flat directions as discussed in the
introduction. For concreteness we will consider the case of two Higgs doublets h1 and
h2 with a tree-level potential, obtained by an appropriate truncation of a supersymmetric
theory, and equal to that of the MSSM. We are also assuming two different one-loop
generated squared mass terms 21 and 22 for the Higgs fields:

2
2
(17)
V = |h1 |2 |h2 |2 + h1 h2 + 21 |h1 |2 + 22 |h2 |2 ,
where = (g22 + g 02 )/8 and = g22 /2. The conditions for having a stable minimum are
21 + 22 > 0 and 21 22 < 0. These conditions are fulfilled provided that one of the masses,
say 22 , is negativeand the other, say 21 , is positive. In this case we get the VEVs hh1 i = 0
and hh2 i = (0, v/ 2), where v 2 = 22 /. Using again the relation of v with MZ , we
obtain the tree-level Higgs mass spectrum:
Mh2 = MZ ,
Mh20 = 21 + 22 ,
1
2
Mh2 = Mh20 + MW
,
1
(18)
where h2 corresponds to the Standard Model Higgs, and h01 , h

1 to the neutral and charged
components of the h1 doublet. Moreover, the string scale is given by
Mh
Ms = 2
2 g2
(19)
with 22 = 22 g 2 Ms2 .
Again, these are tree-level relations which are subject to both string and Standard Model
radiative corrections. In particular, the latter provide important contributions to the mass of
the Standard Model Higgs h2 , which is increased roughly to 120 GeV, and accordingly
to the string scale given in Eq. (19). It is interesting that we obtained the same relations
as in the previous example with a single Higgs field. The difference is that there is also
a left-over scalar doublet whose neutral and charged components acquire masses given in
Eq. (18). As we have pointed out, in this case one needs the one-loop generated squared
masses for the two scalar doublets, 21 , 22 , to be different and opposite in sign. Although
our toy string example allows for different values by introducing different radii, the change
in sign requires more general models, such as those obtained for instance by introducing
additional pairs of branesanti-branes [1820].
4. Discussion on string threshold corrections

We discuss now string threshold corrections to the relations (14) and (15). These are
moduli dependent and may become very important only when some radii become large
compared to the string length. Otherwise, if all radii are of order one in string units,
higher loop corrections are order one numbers multiplied by loop factors which are
suppressed when string theory is weakly coupled. Of course, these (model dependent)
45
corrections are needed for a detailed phenomenological analysis and could be as important
as those of the MSSM that increase the Higgs mass by roughly 10%. An estimate of
these corrections can be done by an explicit computation of the a 4 terms in the expansion
of the potential (8). Notice though that these terms do not determine uniquely the oneloop corrections to the quartic couplings of the charged fields, partly because there are
more than one gauge invariant combinations. An additional subtlety is the existence of an
infrared divergence as l 0, which is due to the low energy running of the couplings
and must be appropriately subtracted to obtain the string threshold corrections in a definite
renormalization scheme [31].
For dimensions longitudinal to our world brane, the large radius limit leads in general
the theory very rapidly to a non perturbative regime, since the (ten-dimensional) string
coupling becomes strong when four-dimensional gauge couplings are of order unity. On
the other hand, for large transverse dimensions, the tree-level string coupling remains
perturbative (of order of the gauge couplings), and therefore their size can in principle
become as large as desired. If this is the case, the decompactification limit exists, and
threshold corrections are again controlled by the string coupling and are suppressed by loop
factors. However, this limit does not exist in general when there are massless bulk fields
that propagate in one or two transverse dimensions, and threshold corrections become very
important [6].
A way to see how these large corrections to the parameters of the effective lagrangian
on the brane arise, is to look at the ultraviolet open string loop diagrams as emission
of massless closed strings in the bulk at the location of distant sources created by other
branes or orientifold planes. This emission leads to corrections that diverge linearly or
logarithmically with the size of transverse space, if there are massless closed string states
propagating in one or two dimensions, respectively. The case of one large transverse
dimension is similar to that of a large longitudinal one, since threshold corrections grow
linearly with the radius and bring rapidly the theory to a non-perturbative regime [32]. In
this case, one can fine-tune the radius to a narrow region near the string scale and the low
energy parameters will be very sensitive to the initial conditions.
In the case of two large transverse dimensions, the logarithmic contributions to the
parameters of the effective action on the brane are similar to those in a renormalizable
theory and can be resummed as in the renormalization group improved MSSM [6]. In this
analogy, the string scale Ms plays the role of the supersymmetry breaking scale, while
the size of the transverse space replaces the ultraviolet cutoff at the Planck mass, MP . For
instance, if the bulk contains n large transverse dimensions of common radius R , while
n
.
the remaining 6 n have string size, one obtains the familiar relation MP2 = Ms2+n R
When there are massless bulk fields propagating in two of them, like e.g. twisted
moduli localized at an (n 2)-dimensional subspace, the logarithmic corrections are
log(R Ms ) = (2/n) log(MP /Ms ).
Concerning the Higgs mass considered here, such large radius dependent contributions
would arise if there are bulk massless fields emitted by the Higgs at zero external
momentum. The vanishing of such tree-level couplings, as for instance with bulk gravitons,
implies the absence of large threshold corrections for the Higgs mass at the one-loop level.
46
This is in agreement with our result (4) which remains finite in the decompactification
limit for any number of large transverse dimensions. However, large corrections can arise at
higher orders, e.g., through gravitons emitted from open string loops. While computation of
such effects is out of the scope of this work, we would like to discuss the general structure
of such corrections and comment on their phenomenological implications.
In the simplest case, the relevant part of the world brane action in the string frame is:

1 + tan2 W 4 2 1 2
H H + FSU(2) + cot2 W FY2
Lbrane = e 2 |DH |2 +
8
4
2 Ms2 4 |H |2,
(20)
where is the string dilaton, the scale factor of the four-dimensional (world brane)
metric, H the Higgs scalar (in the string frame) and D the gauge covariant derivative. The
weak angle at the string scale W must be correctly determined in the string model. Notice
that the last term has no e dependence since it corresponds to a one loop correction. The
bulk fields and are evaluated in the transverse coordinates at the position of the brane.
The physical couplings g2 , and the mass 2 are given by
1 + tan2 W
e ,
2 = 2 e 2 Ms2 ,
8
while Eq. (14) remains unchanged and the relation (15) becomes
g2 = e/2 ,
Ms =
(21)
Mh
.
(22)
2 e/2
The lowest order result (15) corresponds to the (bare) value = 1.
As we discussed above, when the bulk fields and propagate in two large transverse
dimensions, they acquire a logarithmic dependence on these coordinates due to distant
sources. Since the value of at the position of the world brane is fixed by the value of
the gauge coupling in Eq. (21), the relation (14) for the Higgs mass is not affected, while
Eq. (22) for the string scale is corrected by a renormalization of which takes the generic
form:
= 1 + b g22 ln(R Ms ),
(23)
where b is a numerical coefficient. This correction is similar to a usual renormalization

factor in field theory, which here is due to an infrared running in the transverse space.
Depending on the sign of b , it can enhance (b < 0) or decrease (b > 0) the value of the
string scale by the factor 1/. This effect can be important since the involved logarithm is
large, varying between 7 and 35, for R between 1 fm and 1 mm.
In more general models, there are additional bulk fields entering in the expression of
low energy couplings on the brane, such as the twisted moduli localized at the orbifold
fixed points. As a result, every term in the Lagrangian (20) may be multiplied by a
different combination of the bulk fields that acquires an independent correction, similarly
to Eq. (23). Thus, in the generic case, both relations (14) and (15) may be modified by
corresponding renormalization factors that are computable in every specific model. In
particular, the prediction of 120 GeV for the Higgs mass, which coincides with that
47
of the lightest Higgs in the MSSM for large values of tan and mA , can change by this
effect.
A final important question that we have not addressed in this letter is the possible
signatures of Higgs production in brane world models. Previous works done in the context
of the effective field theory suggest that there may be new effects, leading in general to
signatures that are different from those in the Standard Model or the MSSM [3339]. It
will be interesting to study this issue in the framework of the non-supersymmetric type I
string models we discussed here.
Acknowledgements
This work was partly supported by the EU under TMR contracts ERBFMRX-CT960090 and ERBFMRX-CT96-0045, by CICYT (Spain) under contract AEN98-0816 and
by IN2P3-CICYT contract Pth 96-3. K.B. thanks the CPHT of Ecole Polytechnique for
hospitality.
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]
I. Antoniadis, Phys. Lett. B 246 (1990) 377.

E. Witten, Nucl. Phys. B 471 (1996) 135.
J.D. Lykken, Phys. Rev. D 54 (1996) 3693.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263.
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
I. Antoniadis, C. Bachas, Phys. Lett. B 450 (1999) 83.
G. Shiu, S.-H.H. Tye, Phys. Rev. D 58 (1998) 106007.
Z. Kakushadze, S.-H.H. Tye, Nucl. Phys. B 548 (1999) 180.
L.E. Ibez, C. Muoz, S. Rigolin, Nucl. Phys. B 553 (1999) 43.
I. Antoniadis, B. Pioline, Nucl. Phys. B 550 (1999) 41.
K. Benakli, Phys. Rev. D 60 (1999) 104002.
K. Benakli, Y. Oz, Phys. Lett. B 472 (2000) 83.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004.
G.F. Giudice, R. Rattazzi, J.D. Wells, Nucl. Phys. B 544 (1999) 3.
E.A. Mirabelli, M. Perelstein, M.E. Peskin, Phys. Rev. Lett. 82 (1999) 2236.
S. Cullen, M. Perelstein, M.E. Peskin, hep-ph/0001166, references therein.
I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 464 (1999) 38.
G. Aldazabal, A.M. Uranga, JHEP 9910 (1999) 024.
G. Aldazabal, L.E. Ibez, F. Quevedo, JHEP 0001 (2000) 031.
C. Angelantonj, I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, hep-th/9911081.
B. Grzadkowski, J.F. Gunion, Phys. Lett. B 473 (2000) 50.
J.A. Casas, J.R. Espinosa, M. Quirs, A. Riotto, Nucl. Phys. B 436 (1995) 3.
M. Carena, J.R. Espinosa, M. Quirs, C.E.M. Wagner, Phys. Lett. B 355 (1995) 209.
M. Carena, M. Quirs, C.E.M. Wagner, Nucl. Phys. B 461 (1996) 407.
H.E. Haber, R. Hempfling, A.H. Hoang, Z. Phys. C 75 (1997) 539.
M. Carena, H.E. Haber, S. Heinemeyer, W. Hollik, C.E.M. Wagner, G. Weiglein, hepph/0001002.
[27] J.R. Espinosa, R.-J. Zhang, JHEP 0003 (2000) 026, hep/ph/0003246.
48
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
C. Bachas, JHEP 9811 (1998) 23.

I. Antoniadis, C. Bachas, E. Dudas, Nucl. Phys. B 560 (1999) 93.
N. Arkani-Hamed, S. Dimopoulos, J. March-Russell, hep-th/9908146.
I. Antoniadis, M. Quirs, Phys. Lett. B 392 (1997) 61.
J. Polchinski, E. Witten, Nucl. Phys. B 460 (1996) 525.
N. Arkani-Hamed, S. Dimopoulos, hep-ph/9811353.
L. Hall, C. Kolda, Phys. Lett. B 459 (1999) 213.
R. Barbieri, A. Strumia, Phys. Lett. B 462 (1999) 144.
X.-G. He, Phys. Rev. D 61 (2000) 036007.
T.G. Rizzo, J.D. Wells, Phys. Rev. D 61 (2000) 016007.
A. Datta, X. Zhang, Phys. Rev. D 61 (2000) 074033.
G.F. Giudice, R. Rattazzi, J.D. Wells, hep-ph/0002178.

How valuable is polarization at a muon collider?

A test case: determining the CP nature of
a Higgs boson
Bohdan Grzadkowski a,1 , John F. Gunion b,2 , Jacek Pliszka a,3
a Institute of Theoretical Physics, Warsaw University, Warsaw, Poland
b Davis Institute for High Energy Physics, UC Davis, CA, USA
Received 21 March 2000; accepted 12 April 2000
Abstract
We study the use of polarization asymmetries at a muon collider to determine the CP-even and CPodd couplings of a Higgs boson to + . We determine achievable accuracy as a function of beam
polarization and luminosity. The appropriate techniques for dealing with the polarization precession
are outlined. Strategies especially appropriate for a two-Higgs-doublet model (including the MSSM)
are given. Our general conclusion is that polarization will be very useful, especially if the proton
source is such that full luminosity in the storage ring can be retained even after imposing cuts on the
originally accepted muons necessary for P & 0.4 for each beam. 2000 Elsevier Science B.V. All
rights reserved.
1. Introduction
Although the origin of mass has still not been established, it is widely expected that
electroweak symmetry breaking is driven by elementary scalar dynamics, leading to one or
more physical Higgs bosons (for a review and references, see [1]). It is also very possible
that CP violation arises either partially or entirely as result of CP violation in the Higgs
sector [2,3]. Even if not, CP violation in other sectors of the theory can induce CP violation
in the Higgs sector at the loop level [4,5]. Thus, we anticipate that the direct determination
of the CP nature of each observed Higgs boson could be crucial to unraveling the nature of
the full theory.
Even if the minimal one-doublet standard model turns out to be natures choice, we will
certainly want to know that the single observed Higgs boson is entirely CP-even in nature.
1 bohdang@fuw.edu.pl
2 jfgucd@physics.ucdavis.edu
3 pliszka@fuw.edu.pl
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 2 9 - 7
50
B. Grzadkowski et al. / Nuclear Physics B 583 (2000) 4975
Alternatively, if electroweak symmetry breaking turns out to be driven by technicolor-like

dynamics, it would be highly desirable to be able to check that the narrow, light pseudoNambuGoldstone bosons (PGBs), that often arise in such a theory and are rather Higgslike in many respects (see, for example, [6]), do indeed have CP-odd coupling to fermions.
The means for determining the CP nature of an observed narrow resonance are limited.
If it is determined that its W W or ZZ coupling is substantial (small), then we will
know that it has a substantial (small) CP-even component. But, even in the general twoHiggs-doublet model the relation between this coupling and the CP-even and CP-odd
fermionic couplings (denoted a and ib5 , respectively, in f(a + ib5)f 4 ) is modeldependent. However, polarization correlations can allow one to extract the ratio b/a in
a model independent manner. For a light resonance, one possibility is to employ the +
decay of the resonance. If the tt mode is open, then it too can be employed. These finalstate possibilities were examined for a muon collider in [7,8]. One can also look for
certain characteristic angular distributions in associated production, bb + resonance or
tt + resonance, that are sensitive to b/a [911]. However, the most elegant approach to
determining the CP nature of a neutral resonance is to employ initial state polarization
asymmetries. This is possible only for [12] or + [8,13] production of the
resonance. But, the coupling of a neutral resonance is the result of either loop(s) (in the
Higgs case) or an anomaly (in the PGB case), and is not a direct measure of the elementary
fermionic couplings. This leaves + collisions, which are also the only way in which
we will probe 2nd generation fermionic couplings in models where the strength of the
fermionic coupling is proportional to the fermion mass (implying that a resonance with
mass > 2m will always decay to 3rd generation fermions and any CP-odd coupling to
two photons will be dominated
by the top quark loop).
q
In the limit of = 1 4m2 /m2R 1, the cross section for production of a resonance
R with (a + ib5 ) coupling to the muon takes the form

2
2ab
a b2
S ( ) = S0 1 + PL+ PL + PT+ PT 2
cos
sin
a + b2
a 2 + b2

(1)
= S0 1 + PL+ PL + PT+ PT cos(2 + ) ,
where tan1 ab and PT (PL ) is the degree of transverse (longitudinal) polarization 5 of

each of the beams defined as P f + f , with f + being the fraction of muons with
spin in the dominant direction and f the fraction of muons with opposite spin. Here, is
the angle of the + transverse polarization relative to that of the as measured using the
the direction of the s momentum as the z axis. The S subscript denotes signal. Both
the cos and sin dependences allow significant sensitivity to the ratio of interest, b/a,
even though only the sin term is truly CP-violating. The value of S0 , which results from
convoluting the resonance shape with a Gaussian distribution in s, depends upon the
model, the detector, the final state mode and the machine parameters. Ignoring final state
4 Phases can always be chosen so that, for any given f , a and b are real.
5 Note that the cross section for longitudinally polarized muons depends only on a 2 + b2 and cannot be used
to extract information about the CP nature of the resonances couplings to muons.
51
efficiencies and acceptance cuts, and integrating over final state phase space, one finds [13]
a result that can be approximated by 6
40(R + )BR(R F )
(2)
h
2 i1/2 ,
m2R 0Rtot 1 + 8 0 tots
R
where s is the Gaussian resolution in s.

It is expected that the muon collider will first be operated with the relatively small natural
polarization of P 0.2 for the + and bunches, since this will allow maximal machine
luminosity and lead to the largest number of Higgs events. The bunch polarizations will
be oriented in the plane of the final storage ring so that their precession as the muon
bunches circulate (for roughly 1000 turns) can be used to precisely determine the central
energies of the muon bunches and their Gaussian energy spread [15] (see also [16,17]).
(One measures the oscillations of the energies of the electrons from the decaying muons,
which oscillations depend very sensitively upon both the muon energy and the Gaussian
energy spread.) From Eq. (1) and the fact that |PL+ PL | and |PT+ PT | are both 6 0.04, it is
clear that these observations will, to an excellent approximation, provide a measurement of
S0 . (By choosing appropriate relative phases between the + and polarization angles,
the polarization dependent terms in Eq. (1) will average to zero see the next section.)
Thus, a very accurate measurement of S0 in several channels F can be performed and a
reasonably accurate measurement of 0Rtot by a three-point scan will be possible. A modelindependent determination of a 2 + b2 then becomes possible from Eq. (2) by computing
0(R + ) using the measured values of S0 , 0Rtot , and s and a determination of
BR(R F ). To obtain BR(R F ) requires using the missing mass technique in the ZR
final state (in either e+ e or + collisions) to measure (ZR ZX) (X = anything)
and the measurement of (ZR ZF ) to compute BR(R F ) = (ZR ZF )/
(ZR ZX). Once the basic Higgs observations have been performed so that the Higgs
width, branching ratios and a 2 + b2 are well-determined, the next task will be to perform
a model-independent measurement of b/a. This require maximizing the influence of the
polarization-dependent terms in Eq. (1). Determining the best procedure for doing so and
estimating the accuracy with which this measurement can be carried out is the main goal
of this paper.
Any given final state F will have a significant background. In general, the cross
section for the background is nearly independent of s . If we integrate over final
state configurations then the background is independent of PT+ and PT and . In order
to roughly understand the level of sensitivity to b/a that can be achieved, let us for
the moment imagine that we can set PL+ = PL = 0 and choose PT+ = PT = PT .
We then denote the integrated background cross section by B0 . We imagine isolating
(a 2 b2 )/(a 2 + b2 ) and 2ab/(a 2 + b2 ), respectively, via the asymmetries
S0 =
AI
S ( = 0) S ( = )
a 2 b2
= PT2 2
= PT2 cos 2,
S ( = 0) + S ( = )
a + b2
(3)
6 This form, from Ref. [14], has the correct asymptotic limits for 0 / 0 and 0 / and is
R
R
s
s
always within 18% of the exact result.
52
AII
S ( = /2) S ( = /2)
2ab
= PT2 2
= PT2 sin 2.
S ( = /2) + S ( = /2)
a + b2
(4)
Assuming that a 2 + b2 must be measured at the same time as the asymmetries, the error
for the measurement of either of the asymmetries AI or AII is given by

S0 + B0 + A2 B0 S0
2
,
(5)
[A] =
2
L S0
where L is the integrated collision luminosity, assumed distributed equally between the two
measurements and we have temporarily taken PT to be the same for all collisions (which
is not actually the case, as will be discussed shortly). If (a 2 + b2 ) has already been very
accurately determined using initial resonance production measurements and the technique
outlined below Eq. (2), then the expression for [A]2 takes the form
2
1
S0 + B0 A2 S0 S0 + B0
2
.
(6)
[A] =
2
L S0
In either case, if PT < 0.5, and since A2 6 PT4 , the A2 in the numerator of [A]2 can be
neglected, implying that A2 /[A]2 is proportional to PT4 L.
From this discussion, it is apparent that the ideal situation would be to arrange for the
muons to be entirely transversely polarized at the interaction point (IP) and to be able to
adjust the angle between the and + polarizations to be fixed at one of the four values
= 0, /2, , 3/2 for each collision. However, this is not possible if we are interested in
a very narrow resonance, such as a light SM-like Higgs boson which will have a width of
just a few MeV. The reasons follow. First, it is important to note that without intervention
any horizontal (i.e., perpendicular to the magnetic field of the storage ring) polarization will
precess about the magnetic field and therefore rotate relative to the momentum direction.
Consequently, the amounts of transverse and longitudinal polarization at the time the bunch
passes the IP will oscillate. This will be described in more detail in the following section.
The only way to overcome this oscillation would be to have a section of the storage
ring devoted to compensating for this spin precession on a turn-by-turn basis. However,
this conflicts with the requirement that we be able to measure the central energy of each
muon bunch circulating in the storage ring to 1 part in 106 (and the beam energy spread
of the bunch to better than 1 part in 102 ) using measurements of the oscillations of the
energy of the secondary electrons from the muon decays as the bunches circulate and
their spins precess. A large number of turns in which the precession is allowed to occur
without compensation is required. The above precisions are those needed for a light Higgs
resonance with a SM-like width (of a few MeV) in order to be certain of remaining rather
precisely centered on the resonance peak and knowing exactly how that resonance peak is
being sampled. For a resonance with width of a few hundred MeV or larger, knowledge
of the beam energies to 1 part in 104 would be adequate. Since magnetic fields would be
known and be stable at this level, direct measurement of the bunch energies would not
be required and one could consider turn-by-turn spin compensation so as to achieve the
ideal transverse configurations at the IP for each collision. Here, we will assume that the
53
storage ring will not initially be built with the extra magnetic components required for such
compensation.
Thus, regardless of whether the Higgs or other resonance is broad or narrow, it is
necessary to determine the effects of spin precession on the accuracy with which cos 2 and
sin 2 can be determined. In what follows, we develop a procedure whereby the luminosity
needed to achieve a given accuracy is only about 50% larger when the spins are allowed to
precess than if the spins could be taken to be purely transverse for each collision.
2. Polarization precession at a muon collider

Most muon collider designs are such that there are two bunches, each of + s and s
circulating in opposite directions in the storage ring. Typically, these bunches will be stored
for about 1000 turns. Once the bunches enter the storage ring, any polarization in the
horizontal plane will precess in the vertical magnetic field of the storage ring. Further, as
noted in the introduction, for the case of a very narrow resonance this precession is needed
to measure the bunch energies and Gaussian spreads to high accuracy, implying that the
polarizations of the bunches should not be manipulated once they are stored in the ring.
The impact of such precession on extracting physics has not been carefully examined to
date. Thus, we provide in this section a fairly detailed explanation of the considerations
and procedures that must be employed. Even though we shall focus on the case of a narrow
spin-0 resonance, the general features of our discussion will have wider applicability.
The typical configuration will be such that a given bunch enters the ring with
a component PEV (PEH ) of polarization vertical (horizontal) with respect to the plane of the
storage ring. (PEV and PEH are defined in the muon rest frame; see Eqs. (7) and (8) below.)
PEH and PEH+ will rotate in the same directions as the and + themselves (that is in
opposite directions). The rate of rotation of the PEH s as viewed from the laboratory frame
is somewhat different than the rate at which the bunches themselves rotate. The mismatch
means that if, for instance, PEH were longitudinal at the time the bunch first enters the
storage ring, it will not remain so but rather it will precess into the transverse direction and
then back to the opposite longitudinal direction, and so forth.
To be precise, we assume that the storage rings magnetic field points in the b
y direction:
y axis. At any given moment, we
BE = Bb
y . The is then rotating clockwise about the b
x points radially outward. The angle is defined as the angle by
define b
z=p
b . Then, b
y axis in order to get from b
z to
which one must rotate (in the rest frame) about the b
PEH ; is an over-rotation angle in that it is the additional angle of rotation of the spin
as compared to the angle of rotation of pE . After boosting to the laboratory frame, the
complete four-component spin vector for the is then:

x ) sin + PV (0,b
y ).
(7)
s = PH (,b
z ) cos (0,b
The standard result of Ref. [18], assuming that the beam enters the storage ring with
b , is (NT ) = (NT 1/2), where NT is the number of turns during storage,
PbH = p
counted starting with NT = 1 the first time the bunch passes the IP (NT = 1/2 at bunch
54

g 2
g 2
insertion), and = 2 2 , with = E/m and 2 = 1.165924 103 . One finds

that = for a beam energy of 90.6223 GeV/2 (i.e., close to mZ /2), implying that an
initially longitudinal horizontal polarization becomes transverse after the travels
half way around the ring to the interaction point. However, for the somewhat higher
energies of interest for a Higgs factory, the degree of horizontally transverse polarization
(denoted by ) oscillates with NT according to PH (NT )/|PEH | = sin (NT ) as illustrated
in Fig. 1, where we have chosen a convention in which PH is positive when it points
towards the center of the storage ring. We now discuss how to take this oscillation into
account.
First, we note that similar results apply for the + . The + will be traveling in a counterclockwise direction about the b
y axis; at the interaction point 1/2 way around the ring,
z. Once again, we can define an over-rotation angle + (in the + rest frame),
p
b+ = b
in terms of which the + s spin vector (in the laboratory frame) at the interaction point is
written as

x ) sin + + PV+ (0,b
y ).
(8)
z ) cos + (0,b
s+ = PH+ (, b
b+ at the time of insertion, then + (NT ) = (NT 1/2), just as
If we start with PbH+ = p
for (NT ). More generally, we can insert the + beam with any initial angle for PbH+ that
we desire.
We assume that each time a Higgs event is observed we can compute, or will have
measured (prior to the IP, and then extrapolated to the IP), the transverse polarizations
of the bunches. That is, we will know and + for each interaction. In fact, + will
(N )/|PE | at the interaction point as a function of the number of times, N , that

Fig. 1. We plot PH
T
H
T
the beam passes the IP, assuming the muon collider is operating at a total center of mass energy of
k
s = 110 GeV and that the enters the storage ring with longitudinal polarization PH /|PEH | = 1.
This plot is from [19].
55
be completely correlated with , the correlation being determined by the initial spin
configuration with which the bunches are injected into the storage ring. We now give the
expression for the cross section as a function of and + , defining c cos , etc.:
S ( + , )
S0

= 1 + PH+ PH c+ c + cos 2 PV+ PV + PH+ PH s+ s

+ sin 2 PH PV+ s PH+ PV s+ .
(9)
Note that we obtain Eq. (1) using the obvious replacements: PH+ c+ = PL+ , PH c = PL ,
PV+ PV + PH+ PH s+ s = PT+ PT cos , and PH s PV+ PH+ s+ PV = PT+ PT sin . We
also note that if we choose PV+ = PV = 0 and insert the bunches so that + = + /2,
then all polarization dependent terms in Eq. (9) average to zero after many turns. This is the
configuration that would normally be employed for the initial Higgs resonance scans so that
knowledge of b/a would not be needed in order to properly interpret these measurements.
However, to determine b/a we will wish to employ one or more different polarization
configurations and retain maximum information by binning Higgs events according to
( , + ( )). After integrating over (, ) configurations in an f f final state, one finds
(in the limit of mf / s 0):

B , + ef2 e2 1 PH+ PH c+ c 2

+ 2ef e cf c 1 PH+ PH c+ c + d PH c PH+ c+ <(Z )

+ (cf2 + df2 ) c2 + d2 1 PH+ PH c+ c + 2c d PH c PH+ c+ |Z |2 , (10)
where the , Z propagators and couplings are and Z and ief and i (cf + df 5 ),
respectively; see the appendix for more details.
In order to give a simplified discussion, let us assume that S0 (and, thence, a 2 + b2 ) has
been precisely determined by first resonance production measurements that do not focus
on determining the Higgs CP properties. As we have described in the introduction, this is
very likely to be the case. Then, remembering that + can be considered to be a function
of for any given choice of storage ring insertion configuration, we can write the signal
cross section as
C
C
C
C
S ( ) = f0 ( ) + cos 2fc ( ) + sin 2fs ( ),

f0C ( ) = S0 1 + PH+ PH c+ c ,

fcC ( ) = S0 PV+ PV + PH+ PH s+ s ,

fsC ( ) = S0 PH PV+ s PH+ PV s+
PH+ ,
PH ,
PV+ ,
where
(11)
(12)
PV
and + ( ). (Recall that

depend upon the configuration C chosen for
+
( ) is determined by the choice made for the relative angle between the polarization
of the + as compared to that of the at the time of bunch insertion.) The assumption
that a 2 + b2 has been precisely determined already corresponds to assuming that f0C ( )
C
2
is completely known. Then, writing 6 C ( ) C
S ( ) + B ( ), the 1 difference
2
2
between two different Higgs models with the same a + b but different values of will
be given by
56
1 2 =

C
C
C
LC 12c Mcc
+ 21c 1s Mcs
+ 12s Mss
,
(13)
where LC is the luminosity devoted to configuration C,

2

Z
Z
d fcC ( )
d fcC ( )fsC ( )
C
C
,
Mcs =
,
Mcc =
2 6 C ( )
2
6 C ( )

2
Z
d fsC ( )
C
,
Mss =
2 6 C ( )
(14)
and we have defined 1c 1 cos 2 and 1s 1 sin 2 (the differences in these two
quantities between the two models). 7 The integral is over the value of (and the
correlated + ( ) value) at the interaction point. Here, we are approximating the sum
over the discrete values that arise during the course of 1000 turns of the bunches by
an integral over . This is an excellent approximation unless s/90.62 GeV (where
90.62 GeV is the s value such that is constant) is a ratio of small (compared to 1000)
integers.
We will demonstrate below that the configurations can be chosen sufficiently cleverly
that one can neglect the , and, indeed, the entire configuration dependence of 6 C ( )
and approximate 6 C ( ) S0 + B0 , the sum of the unpolarized cross sections. Under
these circumstances, the 1 2 for discriminating between two different models with the
same a 2 + b2 value (i.e., same total unpolarized rate) will then be given by
0 2
X
2
LC SbC2 , where
(15)
1 = 0 S 0
S + B C

2
(16)
SbC2 = 1c PV+ PV + PH+ PH s+ s + 1s PH PV+ s PH+ PV s+
C
is a measure of our sensitivity to cos 2 and sin 2. The averaging is over roughly 1000
turns of the bunches in the storage ring. To an excellent approximation, this average
depends only on the relative values of + and at the time of bunch insertion. We will
now give results for Sb 2 for (a) = + (b) = + + (c) = + + /2. The result
for (d) = + + 3/2 is the same as for (c). We find

2
2

2
Sba2 = 12c 38 PH+ PH + PH+ PH PV+ PV + PV+ PV + 12 12s PH+ PV PH PV+ ,

2
2

2
Sbb2 = 12c 38 PH+ PH PH+ PH PV+ PV + PV+ PV + 12 12s PH+ PV + PH PV+ ,

2
2

2
2
(17)
Sbc2 = 12c 18 PH+ PH + PV+ PV + 12 12s PH+ PV + PH PV+ .
We note that Mcs = 0 when the dependence of 6 C is neglected.
As noted earlier in the previous section, the ideal situation would be to be able to choose
four basic relative orientations of the and + transverse polarizations with respect
to each other: = 0, /2, , 3/2. Constant = 0 or = could be accomplished by
setting PH+ = PH = 0 and PV+ = PV or PV+ = PV , respectively. However, as noted in
7 We note that in the limit of just two bins corresponding to sin = 1 or cos = 1, 1 2 = 1 corresponds
to errors for sin 2 or cos 2, respectively, as given in Eq. (6).
57
the previous section, some degree of horizontal precessing polarization is necessary if we

are to be able to measure the energies of the muon bunches with the accuracy of 1 part in
106 needed for a very narrow Higgs resonance. It is estimated that PH+ = PH 0.050.1
[19] is required for the energy measurement. (We will adopt the optimistic choice of 0.05
in the remainder of this paper.) The rest can be placed in the vertical directions.
The spin precessions make it impossible to maintain = /2 or = 3/2 as the
bunches circulate. The simplest thing that one can do is to inject, say, the bunches
with purely horizontal polarization, and maximize the vertical polarization for the +
bunches subject to the requirement that PH+ = 0.05. Sensitivity to both the magnitude and
sign of sin 2 arises from the variation of the sin 2PH s PV+ term in Eq. (16), which at
various extremes samples = /2 and = 3/2. (We again emphasize the importance of
either calculating, starting from the insertion configuration, or measuring, from the decay
spectrum, the precession angles associated with each observed Higgs event.)
Very specifically, we thus consider the following configurations, keeping in mind that
the net polarization P for the + and bunches will be essentially the same.
I. To approximate the = 0 configuration, we choose PH+ = PH = PH = 0.05,
q
= + , PV+ = PV = P 2 PH2 . The sensitivity is then given by

2
(18)
SbI2 = Sba2 = 12c 3PH4 /8 + PH2 P 2 PH2 + P 2 PH2 .
II. To approximate the = configuration, we choose PH+ = PH = PH = 0.05, =
q
+ + , PV = PV+ = P 2 PH2 . The sensitivity is then given again by Eq. (18):
Sb 2 = Sb 2 .
II
We now justify, for the case of C = I, II, the approximation of taking 6 C S0 + B0 ,

employed in obtaining Eq. (15) from Eqs. (13) and (14). We note that for both
configuration I and configuration II the background, see Eq. (10), will depend only very
weakly on ( , + ) simply because PH and PH+ are both small. Similarly, the PH+ PH c+ c
term in f0C ( ) and the PH+ PH s+ s term in fcC ( ) will be very small. Further, fsC ( )
can be approximately neglected because both PH+ and PH are small. Finally, we will sum
over configurations I and II (with equal luminosity weighting) in order to determine cos 2.
This has the important consequence that (using |PV | ' P , see above) the leading P 4 term
in Eq. (18) is obtained from Eqs. (13) and (14) via the structure

X
0 2
1
1
C
4
+
Mcc P S
B0 + S0 (1 + P 2 cos 2) B0 + S0 (1 P 2 cos 2)
C = I,II

2
2( B0 + S0 )
,
(19)
= P 4 S0
( B0 + S0 )2 P 4 cos2 2( S0 )2
so that even for P as large as 0.5 or so (as we shall later consider) the P 4 correction term
in the denominator can be neglected. 8 Thus, the approximation of taking 6 C S0 + B0
is quite accurate in this case.
8 Note the analogy to Eq. (6) that is apparent when we recall that |A|2 corresponds to 1 2 = 1 and that A2
in the present case is P 4 cos2 2.
58
III. To emphasize the = /2 and = 3/2 configurations over many q

turns of the
+
+
bunches, we choose PH = P (PV = 0), PH = PH = 0.05 and PV = P 2 PH2 .
In addition, we choose = + + /2 (or = + + 3/2) so as to minimize
dependence on 1c . The sensitivity is then given by

(20)
SbIII2 = Sbc2 = 18 12c P 2 PH2 + 12 12s P 2 P 2 PH2 .
Note that if we choose either = + or = + + , then the (undesired)
sensitivity to 1c would be increased. For PH 6 0.05 and P > 0.2 (as will be the
case), the 12c term can be dropped in Eq. (20).
In order to justify neglecting the dependence of 6 C on in case III, we note that the
terms of concern in B ( , + ), Eq. (10), are those proportional to (PH c PH+ c+ )
P c . The term of concern in S ( , + ) is that proportional to fs ( ) S0 P 2 s . We
will return to this latter term in a moment. First, we note that the P c term in B ( , + )
can be approximately eliminated without affecting S ( , + ) by binning together into
0
0
0
bin events with the same sign of s but with c = c
and c = c
. (Even
a single c
0
after 1000 turns, this is only an approximation since the number of turns for which c c
0
can differ significantly from the number of turns for which c c , depending upon the
Higgs boson mass and the number of c bins employed.) Concerning the S ( , + ) term
proportional to P 2 s , we note that in computing Mss , Eq. (14), the leading term in the
numerator is even in s while the P 2 s correction term in the S ( , + ) contribution to
6 C is odd. This means that we will get a cancellation analogous to that in Eq. (19) so that
these corrections to our approximation will be of relative order P 4 and can be neglected.
It is useful to note how these results compare to the ideal where we imagine that can
be held fixed. For = 0, , our sensitivity Sb 2 would be 12c P 4 while for = /2, 3/2
it would be 12s P 4 . For PH = 0.05 we suffer a loss of about 0.469 (0.939) for 1s (1c ) for
P = 0.2 and 0.495 (0.990) for P = 0.5. In the remainder of this paper, we approximate
these PH = 0.05 results by assuming no loss in Sb 2 for 1c and a factor of 1/2 loss in
Sb 2 for 1s . This latter factor must be overcome by increased luminosity to obtain the
same statistical accuracy as in the ideal case. To equalize sensitivity to 1s and 1c , we
will assume that we accumulate twice as much luminosity for the 1s configurations as
for the 1c configurations. Thus, for total integrated luminosity L we accumulate L/6
in configuration I, L/6 in configuration II, and 2L/3 in configuration III. Of the latter
2L/3, as the spins precess L/3 will be accumulated in configurations with sin < 0 and
L/3 with sin > 0. The sensitivity achieved is then equivalent to probing the four fixed
configurations ( = 0, /2, , 3/2) with L/6 each. The results of the following sections
will sometimes be phrased in this latter language.
3. Maximizing sensitivity to CP violation

The number and density of muons in each bunch are limited by space-charge and
muon beam power considerations. Existing designs have proton source intensity such that
these limits are saturated when momentum cuts on the initially accepted muons coming
59
from the target are such that the muon beams have polarization of P 0.2 (or less). To
date, there has been no strong reason for designing the proton source so as to have full
luminosity (fully saturated bunches) for larger P values. This is because the typical design
for the storage ring is such that the polarization is longitudinal at the interaction point
half way around the ring, motivated by the fact that (for longitudinal P ) the signal rate is
proportional to 1 + P 2 , while , Z induced (e.g., bb) backgrounds are proportional to
1 P 2 . In this configuration, the improvement of S/ S + B with increasing P (assuming

the integrated luminosity can be kept fixed) is rather slow, rising by . 20% from P = 0.2 to
P = 0.4. Thus, although higher (longitudinal) P would have some advantage if the proton
intensity were such that the two muon bunches were both fully saturated at P & 0.4 rather
than at P 0.2, it is usually accepted that this advantage is not sufficient to justify the
expense associated with building a more intense proton source.
However, if determining the CP nature of the resonance couplings to + is the goal,
the (approximate) proportionality of the Sb 2 sensitivity measures (see the previous section)
to P 4 implies that retaining maximal luminosity at P & 0.4 has much greater advantages.
The extra proton source intensity required to achieve this is determined by the fraction,
fsurv (P ), of muons that survive after imposing the momentum cuts required to achieve a
certain P (prior to inserting the muon bunches into the storage ring). The variation of fsurv
as a function of P has been estimated by many different groups (see, e.g., [16,17]). Here,
we adopt the post-cooling result given in [16, Fig. 21] which is approximately described
by fsurv (P ) = 1.45 2.51P for 0.2 6 P 6 0.6. The number of muons in each bunch, Nb ,
is proportional to the proton source intensity Lps times fsurv . Using the specified form, one
finds, for example, that fsurv (0.45) : fsurv (0.39) : fsurv (0.2) 1/3 : 1/2 : 1, implying that
Lps at P = 0.45 (0.39) must be 3 (2) times that for P = 0.2 to maintain full bunches.
If the machine is constructed with Lps onlylarge enough to saturate the bunches at
P = 0.2, then one finds that fsurv is such that S/ S + B declines if one increases P above
0.2 (and arranges the polarization to be longitudinal at the interaction point). However,
maximum Sb is rather achieved by increasing P to the point (P 0.39) where each muon
bunch has one-half of the P = 0.2 number Nb , and then merging the two bunches into
one bunch (prior to insertion into the storage ring). Bunch merging effectively doubles the
storage ring luminosity at this P compared to what would result without bunch merging.
(Without bunch merging, one would have luminosity proportional to 2 (Nb /2)2 whereas
with bunch merging the luminosity would be proportional to 1 Nb2 .) As a result,
the
collider luminosity at P = 0.39 is only a factor of 2 lower than at P = 0.2 and P 2 L is
roughly a factor of 2.7 larger. (At the same time, S/ S + B has declined by about a factor
of 1.4.) If CP studies are the goal, one will choose the P = 0.39 option with bunch merging.
If Lps larger than that required to saturate the muon bunches at P = 0.2 is available, bunch
merging may or may not be desirable for CP studies. For Lps such that bunch saturation
is achieved for P < 0.42 (> 0.42), the sensitivity Sb for CP studies will (will not) benefit
by increasing P still further to the point where the bunches can be merged. If the bunchmerging option is appropriate, the integrated collider luminosity L will be one-half the (no
merging) value at P = 0.2. If bunch saturation is the better option, L will be exactly the
same as the P = 0.2 value. In either case, increasing P beyond the bunch saturation or
60
Fig. 2. We plot the maximum achievable relative sensitivity RS of the CP determination as a function
of the relative proton source intensity I , both being relative to values corresponding to two full +
and bunches (each) at P = 0.2. At I = 2.38, one switches from bunch merging to bunch filling.
Also shown is the polarization being employed for a given I .
bunch merging point (whichever is more optimal) causes Sb to decline.

b Sb0 , where the
To be more explicit, we define the relative sensitivity RS S/
0
denominator is that achieved using P = 0.2 and Lps such that the two bunches are saturated
for P = 0.2. Let us also define the relative source intensity I Lps /L0ps . In Fig. 2, we
plot RS obtained either by bunch merging or bunch saturation (whichever gives the larger
RS ) as a function of I . Also shown is the corresponding polarization being employed as
a function of I . The corresponding functional forms are: for bunch merging, P m (I ) =
a b/2I with RSm (I ) = (P m (I )/0.2)2 21/2 ; for simply filling two bunches, P f (I ) =
f
a b/I with RS (I ) = (P f (I )/0.2)2 . Here, a ' 0.577 and b ' 0.377 for the form of fsurv
f
given earlier. Below (above) I = 2.38 we employ P m and RSm (P f and RS ). The results
for RS show clearly that if determining the CP nature of a resonance is the goal, then one
gains by optimizing P and having as large a proton source intensity as is feasible.
Of course, the absolute accuracy with which a measurement of b/a can be made is
dependent upon the resonance model. In the following section, we will consider several
Higgs boson examples and demonstrate that by using optimal polarization, as described
above, a very meaningful measurement of b/a can typically be performed. Overall, we
will conclude that the increase in our ability to determine the CP nature of the muonic
couplings of a resonance would provide substantial motivation for spare proton source
intensity.
61
Before ending this section, we note that there is a very interesting possibility for
increasing the number of muons with high polarization retained after making the necessary
selection cuts. In particular [16,20,21], if the accelerating gradient of the phase-rotation
device that immediately follows the pion capture solenoid is sufficiently high ( 45
MV/m), the correlation between the muon arrival time and average muon polarization
might be significantly enhanced, resulting in a more effective selection of polarized muons,
and a higher final polarization for the same luminosity. Such high-gradient phase-rotation
designs are being actively considered for the neutrino beam facility version of the muon
accelerator and storage ring [21].
4. Two test cases: a light Higgs with SM-like bb rate and a degenerate H 0 A0
MSSM pair
Consider first the case of a Higgs boson with SM-like bb final state rate and total width.
For example, it might be that a Higgs boson is detected and appears to have SM-like
W W/ZZ couplings and branching ratios to accessible final states, in which case we will
wish to determine if its fermionic couplings are indeed CP-even. Here, we assess the
accuracy with which this verification can be performed at a muon collider. As detailed
in earlier studies [13,14], the result of Eq. (2) and the very small width of a light SM-like
Higgs boson means that the muon collider must be operated with the smallest possible
beam energy resolution, even if this results in substantial luminosity sacrifice. For currently
understood designs, the best that can be achieved is 1Ebeam /Ebeam R with R of order
R = 3 105 . 9 For such an R value, the yearly integrated luminosity is anticipated to be
of order L = 0.1 fb1 when the bunches are full. We will examine results achievable for
mh = 110 GeV and mh = 130 GeV by employing only F = bb.
The invariant amplitude squared for Higgs + + Z exchange is given in the appendix
as a function of the relative angle between the transverse polarizations of the and
+ and as a function of both and , the angles describing the orientation of the b and b
momenta in the final state. We first note that the interference term is never of importance
for the cases we explore here. Secondly, we observe that |M|2 from Higgs exchange is
independent of and , whereas that for the + Z background does depend upon both
and . Our analysis proceeds as follows.
= (a, b)/(gm /2mW ) (a a , b b ), denoting
We assume input values for (a,
b)
them by a 0 and b0 . We integrate over , but bin events in cos (bin label j ). As
described, the spin precession means that we must also bin events in (remembering
that + will always have a known correlation with for a given spin-precession
configuration); bins are labelled by i. We devote L/6, L/6 and 2L/3 to the spinprecession configurations C = I, II, III described in the previous section. 10 The number of
events (including both signal and background) for a given configuration choice C, given
9 In this paper, all R values will be quoted in absolute units and not in per cent.
10 In all cases studied, if there are no a priori restrictions on the possible couplings, sensitivity to b/
a is
maximized by employing all the configurations.
62
bin i and given cos bin j is denoted by N(C, i, j ). The statistical error for each such
bin is 1N(C, i, j ) = N(C, i, j ). We then consider a different model characterized by a

and b and compute:
1 2 =
X (N(C, i, j, a,
N(C, i, j, a 0 , b0 ))2
b)
.
1N 2 (C, i, j, a 0, b0 )
(21)
C,i,j
The discussion of Section 2 shows that with appropriate binning and appropriate choices
for our configurations we can effectively drop the C, i, j dependence of 1N(C, i, j ). The
following discussion will implicitly rely on this fact.
The accuracy with which a and b can be p
determined can be assessed by drawing contours
2
1
r a 2 + b 2 parameter space. The n error in

of constant 1 in tan b/a,
a so as
for example (which implicitly assumes that r is always adjusted for a given b/
2
to minimize the 1 ) is given by drawing the vertical line tangent to the 1 2 = n2
contour. We will also include the 1 2 =p6.635 contour corresponding to 99% CL. The
corresponding statistical errors for r a 2 + b 2 are obtained by the horizontal lines
tangent to the contours. However, the true errors for r must incorporate the error coming
from our imprecise knowledge of BR(h bb) and 0htot .
Let us begin with the example of a SM Higgs boson (a 0 = 1, b0 = 0) with mass mh =
110 GeV. To illustrate the importance of polarization, we use as a reference point a total
integrated luminosity at P = 0.2 (with full muon bunches) of L = 0.15 fb1 (per year).
We plot contours in (, r) parameter space for some sample cases:
(i) P = 0.2, I = 1 (i.e., nominal proton source intensity, L0ps );
(ii) P = 0.39, I = 1 (bunch merging at L0ps );
(iii) P = 0.48, I = 2 (bunch merging at Lps = 2L0ps );
(iv) P = 0.45, I = 3 (full bunches at Lps = 3L0ps ).
a = tan is 1 for
The contours are presented in Fig. 3. It is useful to keep in mind that b/
= /4 0.785. For the cases (i)(iv) above, the 1 , 2 , 99% CL, and 3 limits on
are: (i) 0.94, . . .; (ii) 0.30, 0.64, 0.89, 1.14; (iii) 0.20, 0.41, 0.53, 0.64; and (iv) 0.15, 0.32,
a = tan are: (i) 1.36, . . .; (ii)
0.42, 0.50. The corresponding 1 and 99% CL limits on b/
0.31, 1.23; (iii) 0.20, 0.58; (iv) 0.15, 0.45. We see that even I = 1 with bunch merging
a requires I > 2.
gives a reasonable 1 measurement; however, a good 99% CL limit on b/
Of course, as the Higgs mass increases, the event rates decline and the results worsen.
This is illustrated in Fig. 4 which gives the same plots as Fig. 3, but for mh = 130 GeV.
Above mh = 130 GeV, the bb branching ratio begins to decline sharply and one should
see a significant W W final state rate, which in itself would be a signal of a significant
CP-even component for the Higgs boson. A high discrimination check of the CP nature of
the muonic coupling would be possible by including the W W channel in the transverse
polarization analysis. We will not pursue this here.
a measurement
Another perspective on our errors is to ask how well separated the b/
for a SM Higgs boson is from the same measurement for an alternative Higgs boson of
the general two-Higgs-doublet
model (2HDM). To illustrate, consider mh = 110 GeV with
a 0 = b0 = 1/ 2. The 1 2 contours for this input model are given in Fig. 5. The 1
63
Fig. 3. We give contours at 1 2 = 1, 4, 6.635, 9 in the , r parameter space assuming that the
integrated luminosity at polarization P = 0.2 (with full muon bunches) is L = 0.15 fb1 . Four
cases are compared: (i) P = 0.2, L = 0.15 fb1 (which corresponds to I = 1); (ii) maintaining
same proton source intensity, I = 1, but merging the bunches, corresponding P m (I = 1) 0.39,
for which L = 0.075 fb1 ; (iii) increasing the proton source intensity by a factor of two, I = 2,
while merging the bunches, corresponding to P m (I = 2) 0.48, for which L = 0.075 fb1 ; (iv)
I = 3, using just-full bunches, corresponding to P f (I = 3) 0.45, for which L = 0.15 fb1 . We
assume a SM Higgs boson (a 0 = 1, b0 = 0) with mh = 110 GeV and an overall efficiency (including
b-tagging efficiency) of 0.54.
contours are already nonoverlapping with those for a pure CP-even Higgs (Fig. 3) for
P = 0.39 and I = 1. However, the 99% CL contours have a small overlap even at I = 3
(P = 0.45). This again emphasizes the potential importance of larger-than-nominal source
luminosity for such discrimination.
Let us now turn to a situation in which there are two degenerate Higgs bosons. This can
arise in the 2HDM and also in the MSSM. The rates as a function of the relative angle
between the transverse polarizations, , will depend in detail upon the a and b values of
the two Higgs bosons. As an example, suppose thatwe have a degenerate pair at mh =
110 GeV, one of which is pure

CP-even with a = 1/ 2, b = 0 ( = 0) and the other pure
CP-odd with a = 0, b = 1/ 2 ( = /2). Here, we have chosen the normalizations so that
the total unpolarized (i.e., averaged over the four settings) production rate is the same as
for a SM Higgs boson, with each of the two Higgs contributing equally to this rate. (We will
also assume that the two Higgs bosons have the same bb branching ratio as a SM Higgs
64
Fig. 4. Same as Fig. 3, but for mh = 130 GeV.
Fig. 5. Same as Fig. 3, but for a mixed-CP Higgs boson with a 0 = b0 = 1/ 2 and mh = 110 GeV.
65
Table 1
Event number pattern for different Higgs models as a function of , assuming PL = 0 and PT = P ;
see Eq. (1)
(a,
b)
=0
= /2
= 3/2
(1, 0)
(1/ 2, 1/ 2)
(0, 1)
(1/ 2, 0) + (0, 1/ 2)
1 +P2
1
1 P2
1
1
1 P2
1
1
1 P2
1
1 + P2
1
1
1+ P2
1
1
boson.) In this situation, Eq. (1) shows that the production rate will have no dependence on
[cos(0 + ) + cos( + ) = 0], whereas any single Higgs model will exhibit a distinct
pattern as a function of , as illustrated in Table 1. Were we able to accumulate events
at fixed , the pattern of 1 2 for discriminating any two models from one another
(assuming all have the same bb production rate) is apparent in the approximation where
the dependence of the 1N( ) errors is neglected (as approximately appropriate given
the dominance of the background contribution to 1N and the weak dependence of the
background on ); 1 2 would simply be proportional to the squares of the rate differences
between two models summed over the four values. In the case where we compare the
-independent rate predicted in a degenerate Higgs pair model to expectations for any
given single Higgs boson, it is apparent from the table that the 1 2 between the degenerate
model and any single Higgs model is independent of the latter. (This is true for arbitrary
for a single Higgs since cos2 (2)+cos2 (2 +/2)+cos2 (2 +)+cos2 (2 +3/2) = 2.)
However, because of spin precession, we must actually employ Eq. (9). Since sin 2 = 0
for both Higgs bosons while cos 2 = 1 for the CP-even and cos 2 = 1 for the CP-odd
Higgs boson, after summing over both, the cross section will be the same as that obtained
if we set cos 2 = 0 as well as sin 2 = 0. To compare to the simplified discussion of
the previous paragraph, let us recall that for our choices of configurations and binning
we can approximately neglect the C, i, j (i.e., configuration, and ) dependence of
the 1N(C, i, j ) denominators in Eq. (21). Then, for the L/6, L/6, 2L/3 luminosity
weightings for configurations C = I, C = II, C = III, respectively, the effective sensitivity
for discriminating between the [cos 2 = sin 2 = 0] equivalent situation for the
degenerate pair and the results expected for any single Higgs boson characterized by angle
will be proportional to
1 4
1 4
1 b2
2 b2
1 4
2
2
2
2
3 Sa + 3 Sc 3 P 1c + 1s = 3 P cos 2 + sin 2 = 3 P
(see Eqs. (18) and (20)), i.e., again independent of the value for any single Higgs boson
to which one might compare. One finds that this approximation is actually quite good.
Taking L = 0.15 fb1 for I = 1 and P = 0.2 and assuming a SM-like bb production rate,
we find 1 2 = (i) 0.38; (ii) 2.8; (iii) 6.4; (iv) 9.8, for the four (I, P ) scenarios defined
earlier, essentially independent of the type of single Higgs boson exchange to which one
compares. Note that we get > 99% CL exclusion of a single Higgs model only for I > 2
if nature chooses a degenerate pair. In any case, the dependence is key to separating a
66
degenerate CP-even plus CP-odd pair of Higgs bosons from a single Higgs boson of any
given CP nature.
In the MSSM, in the absence of CP violation the most likely situation in which we
will encounter a highly degenerate pair of Higgs bosons that cannot be easily separated
by scanning in s is in the limit of large mA0 and large tan . The increasing degeneracy
with increasing tan is illustrated for mA0 = 400 GeV in Fig. 6, assuming squark masses
of 1 TeV and no squark mixing and a beam energy spread of R = 0.001. Since the total
widths of the H 0 and A0 are substantial (> 1 GeV) for the mA0 and tan values being
considered, it is not guaranteed that we will be able to separate the peaks. The figure shows
that we are able to observe two separate peaks (the A0 peak being at lower mass than the
H 0 peak) for moderate tan . 6. But, for higher tan values the peaks begin to merge;
for tan & 8, |mH 0 mA0 | < 1 GeV and one sees only a single merged peak. The picture
changes if squark mixing is substantial; for instance, for mA0 = 300 GeV, squark masses
1
Fig. 6. We plot bb (solid) and tt (dashed) event rates
for total integrated luminosity of L = 7 pb
coming from + H 0 + A0 as a function of s, assuming mA0 = 400 GeV. Each window is
for the specific tan value noted. These event rates are to be multiplied by a factor of 1000 for the
expected yearly integrated luminosity of 7 fb1 . We employ squark masses of 1 TeV and no squark
mixing. Supersymmetric decay channels are assumed to be closed.
67
of 1 TeV and large squark mixing (At = Ab = 3 TeV), the H 0 and A0 peaks actually cross
at tan 5.
To explore the various possible scenarios, we begin by considering our ability to
discriminate between an exactly degenerate pair of CP-even and CP-odd Higgs bosons
vs. a single Higgs boson (of a given type) as a function of tan . We note that the present
situation is significantly more favorable than that discussed above with a light degenerate
pair with SM-like widths, branching ratios and production rate. Since the widths of the
H 0 and A0 are significant, one can operate the muon collider with the natural beam
energy resolution of order R = 0.001 and still have s < 0, which maximizes the Higgs
production rate. 11 At such R and for s 300 GeV (400 GeV), the nominal yearly
integrated luminosity for full bunches at P = 0.2 is estimated at L 2 fb1 (7 fb1 ), i.e.
more than a factor of ten larger than the R = 3 105 value. Second, the bb branching ratio
for large mA0 and large tan is inevitably of order 88%, the only significant competitor
being + . Even for moderate tan and for Higgs masses above tt threshold, the
bb branching ratio remains substantial. Even more importantly, at large mA0 the +
couplings of the H 0 and A0 are enhanced relative to SM strength by a factor of tan so
that the + branching ratio asymptotes to a constant.
As shown in Fig. 6, the H 0 and A0 need not give exactly the same bb event rate (due
to differences in BR(+ )BR(bb)). This is true even if parameters are chosen so that
they are exactly degenerate in mass at some given tan value. On the other hand, it is
also typically possible to choose parameters so that they do have exactly the same rate. We
consider this last possibility first. (As already discussed, when the H 0 and A0 have exactly
the same bb rate, there is no dependence of the summed bb rate on . As a result, even
after including spin precession, 1 2 will be independent of the CP nature of any single
Higgs boson to which we compare if we sum over the three configurations C = I, II, III
with luminosity weighting L/6, L/6, 2L/3.) To simulate this situation, we choose a value
of tan and a value of the A0 mass. We then compute the H 0 and A0 event rates (assuming
stop masses of 1 TeV and no squark mixing) and reset the H 0 and A0 event rates so that
both are equal to the average of the originally computed H 0 and A0 rates. We consider
mA0 values of 300 GeV and 400 GeV. In the latter case, the tt decay channel is open and
has significant branching ratio at lower tan values. This, along with the decreased +
coupling at lower tan , is included in computing the bb rate. We make no use of the tt
channel; its inclusion would, of course, increase our discrimination power, especially if the
final state correlations that can be probed there are employed. The resulting discrimination
power is generally more than adequate without its inclusion.
Figs. 7 and 8 give the results for mA0 = 300 GeV(400 GeV), respectively. We plot
1 2 obtained for the four polarization/proton-source-intensity options (i)(iv) delineated
earlier, but using the higher nominal luminosities stated above:
11 Of course, one could possibly separate the two Higgs bosons by employing R < 0.0001, but it is unlikely
that the machine would be operated in this way at this higher energy unless a Higgs resonance peak is seen and
there is already some evidence, through the techniques considered here, that there are actually two overlapping
resonances with different CP properties. Even then, there will remain the possibility of very close or even exact
degeneracy.
68
Fig. 7. In the upper left window, we plot the bb event rates in the MSSM for the H 0 and A0 (the
A0 rate is the larger of the two) as a function of tan for mA0 = 300 GeV, assuming squark masses
of 1 TeV, no squark mixing and integrated luminosity of L = 2 fb1 . Also shown is the (relatively
small) background rate. In the remaining windows we plot 1 2 , after including precession and
increasing L to L = 3 fb1 , as a function of tan for three different cases in which we forcibly
lower mH 0 to 300 GeV (for exact degeneracy). (1) We adjust the H 0 and A0 event rates so that each
is exactly equal to the average of the A0 and H 0 rates as predicted by the MSSM, and compute 1 2
for H 0 + A0 vs. a single Higgs resonance (of any type) with the same total event rate, employing
only the bb channel. (2) We use the actual H 0 event rate and compute 1 2 for H 0 + A0 vs. a single
CP-even resonance yielding the same bb event rate. (3) As in (2), but vs. a single CP-odd resonance.
In cases (1)(3), we give results as a function of tan for the four polarizationluminosity situations
(i)(iv) (as labelled on the curves) described in the text.
(i) P = 0.2, and the I = 1 values of L = 3(10.5) fb1 ;

(ii) P = 0.39, and the I = 1 merged-bunch L = 1.5(5.25) fb1 values;
(iii) P = 0.48 and I = 2, yielding merged-bunch L = 1.5(5.25) fb1 ;
(iv) P = 0.45 and I = 3, without bunch merging, yielding L = 3(10.5) fb1 .
We emphasize that options (i) and (ii) require no over-design of the proton source. The
1 2 plots show that good discrimination is obtained even for option (i) once tan > 10.
Option (ii) would be needed for good discrimination if tan 5.
We next consider the discrimination power achieved if we continue to enforce exact
degeneracy by lowering mH 0 to mA0 , but employ the actually predicted A0 and H 0 event
69
Fig. 8. The same as Fig. 7 except for mA0 = mH 0 = 400 GeV, and using L = 7 fb1 for the rate
window increased to L = 10.5 fb1 for the 1 2 windows (in order to account for the inefficiency
associated with spin precession). The upper-left window gives the H 0 and A0 and background tt
event rates as dashed curves. In the 1 2 analysis, only the bb final state (after accounting for the
depletion from H 0 , A0 tt) is employed.
rates. For mA0 = 300 GeV, the H 0 event rate for bb is lower than the A0 rate, and the
worst (best) discrimination is achieved relative to a purely CP-odd (purely CP-even) Higgs
boson (again assumed to have exactly the same total bb event rate). The resulting 1 2
values for options (i)(iv) are given in the two bottom windows of Fig. 7. Even in the worst
case, good discrimination power is achieved for tan > 5 by using option (ii). Exactly
the reverse situation arises for mA0 = 400 GeV. As seen in Fig. 8, the A0 rate in the bb
channel is smaller than for the H 0 . Discrimination against a purely CP-even Higgs will
be the most difficult. In fact, we see that option (ii) will not provide clear discrimination
against a purely CP-even Higgs if tan . 7.
Finally, let us consider the case where the H 0 and A0 are not exactly degenerate, but are
only nearly so. To be specific we adopt the MSSM predictions of Fig. 6. As noted earlier, if
the collider is operated with R = 0.001 beam energy spread so as to maximize luminosity,
for tan & 8 one observes a single broad peak in a scan and we will not know that there are
two Higgs bosons present. The apparent peak of the resonance shape will tend to coincide
with the mass of the Higgs boson which yields the higher bb rate (which Higgs this is
depends on the mass, as we have seen). If we center on this apparent peak, the contribution
70
to the bb rate from the weaker Higgs will be further reduced because s is somewhat
on the wing of its resonance shape. This further decreases the 1 2 discrimination power
relative to a purely CP-odd (CP-even) Higgs boson if the dominant Higgs is CP-odd (CPeven). Even if we see two separate peaks, we will wish to experimentally determine which
is the H 0 and which is the A0 .
To illustrate, let us define
q
b
b2 0 + N
b2 0 and = tan1 NA0 ,
= N
A
H
b
NH 0
bH 0 ) is the number of A0 (H 0 ) events at the chosen s for a possible model

bA0 (N
where N
1/2
divided by (NA2 0 + NH2 0 )MSSM . In Fig. 9, we present 1 2 contours in , parameter
space for the input MSSM model (see Fig. 6) specified by mA0 = 400, tan = 7, squark
masses of 1 TeV and no squark mixing. To determine that there are A0 events under the H 0
peak at 99% CL requires the I = 2 proton-source-intensity option (iii). A global overview
of our ability to discriminate the MSSM input model from a single purely CP-even (the
worst case) or single purely CP-odd (the best case) Higgs boson yielding exactly the same
bb event rate is provided by Fig. 10. There, we plot 1 2 for these two discriminations
as a function of tan for polarization/proton-source-intensity options (i)(iv). We find
Fig. 9. We plot contours of 1 2 = 1, 4, 6.635, 9 in the , parameter space, assuming

mA0 = 400 GeV, tan = 7, squark masses of 1 TeV and no squark mixing. The different windows
give results for the different proton source intensity and bunch merging options described in the text.
71
Fig. 10. We plot 1 2 as a function of tan for discriminating

between the A0 H 0 mixture
predicted by the MSSM model (as specified in the text) at s = mH 0 compared to a single CP-even
or CP-odd Higgs boson yielding exactly the same bb event rate. Results are presented for the
polarization/proton-source-intensity options (i)(iv) described in the text.
that discrimination against a purely CP-odd Higgs boson is excellent even for tan 5,
whereas tan > 9 [tan > 7.3] is required for 99% CL discrimination against a single
CP-even Higgs boson using option (i) [(ii)]. Of course, if squark mixing is large, the degree
of degeneracy between the H 0 and A0 can be such that the peaks will merge even when
tan 5. In this case, good discrimination power would typically require enhanced proton
source intensity.
5. Summary and conclusions

The most natural polarization for the muon bunches at a muon collider is P 0.2.
For this polarization, the collider luminosity will be maximal. In initial exploration of the
Higgs or other narrow resonance, one will choose the polarizations to lie in the horizontal
plane of the storage ring. As the polarizations rotate it will then be possible to perform
a precise measurement of the beam energies and their Gaussian widths and of the degree
of polarization itself. Further, by choosing the + polarization to be 90 degrees out of
phase with the polarization, the effect of the rotating polarization will cancel out
when averaging over the 1000 turns during which the typical bunches are stored; i.e.,
the turn-averaged cross section will be identical to the polarization-averaged cross section.
However, once the basic properties of the Higgs boson or resonance are known (total width,
72
branching ratios, etc.) the next important goal will be to determine its CP properties as
codified in the relative strength of its scalar and pseudoscalar couplings to fermions. The
muon collider provides a perfect opportunity for determining these relative strengths for the
muon itself, but only if one has retained the ability to reconfigure the collider so as to run
with high polarizations for the muon bunches and to have flexibility in the orientation of
these polarizations at the time the bunches are inserted into the storage ring. In most muon
collider designs, high polarization can only be achieved by making strong momentum cuts
on the muons, which, unless the proton source has spare luminosity relative to bunch
saturation limits for P 0.2, will result in some loss of collider luminosity. However, if the
goal is to measure the CP-odd/CP-even coupling ratio, this loss of luminosity is more than
compensated by increased sensitivity. Our goal in this paper has been to develop efficient
techniques and polarization configurations for determining the CP-odd/CP-even coupling
ratio and to quantify the accuracy with which this ratio could be extracted from the data.
We have found that one very effective technique is to accumulate events for three
carefully chosen polarization configurations (as defined by the polarization of the + and
bunches at the time they enter the storage ring). To achieve these three configurations,
one will need appropriate solenoids and/or small rings for manipulating the polarizations
prior to injection into the storage ring. Once the bunches are in the storage ring, further
manipulation would destroy our ability to measure the bunch energies to the 1 part in 106
level needed (at least for a narrow Higgs boson). We have demonstrated that by selecting
high-polarization muons and performing bunch merging, meaningful constraints on the
CP nature of the muonic couplings of a Higgs boson are possible even if it is light and
narrow and even if the proton source has only the nominal luminosity required to saturate
the bunch limits in the storage ring at relatively low polarization. However, we have seen
that extra proton source luminosity (sufficient to saturate the bunch limits of the storage
ring when muon selections leading to large polarization are employed) may be needed to
achieve a high level of certainty regarding the CP nature of the muonic couplings of such
a Higgs boson. We have also shown that distinguishing a highly degenerate pair of CP-even
and CP-odd Higgs bosons from a single Higgs with definite CP nature will generally be
possible, especially if this is the degenerate H 0 A0 pair of the MSSM at high tan and
large mA0 for which event rates are high and high-luminosity-running at R = 0.001 would
suffice.
Overall, while the machine capabilities required to make a good determination of the
CP nature of the muonic couplings of a Higgs boson do not come without a price, it could
well happen that the true nature of an observed Higgs resonance would be obscure without
the measurements considered here. Further, Higgs/resonance CP studies provide but one
example of how unique sensitivity to new physics can result if we can take advantage
of the slow precession of the muon bunch polarizations by manipulating their orientation
and relative phases and recording events on a turn-by-turn basis. Thus, we encourage the
muon collider designers to retain the flexibility needed to insert devices for rotating the
spins of the beams to a variety of initial (i.e., prior to insertion into the storage ring)
configurations of longitudinal and transverse polarization and to consider seriously the
possibility of over-designing the proton source relative to storage ring bunch saturation
73
limits associated with low polarization. Or, perhaps it will prove possible to employ a very
high gradient in the initial phase-rotation stage of pion/muon capture [16,20,21], thereby
maximizing the luminosity available after the selection cuts required for high polarization.
Acknowledgements
We thank S. Geer, R. Raja and R. Rossmanith for helpful conversations on experimental
issues. This work was supported in part by the US Department of Energy, the UC Davis
Institute for High Energy Physics, the State Committee for Scientific Research (Poland)
under grant No. 2 P03B 014 14 and by Maria Sklodowska-Curie Joint Fund II (Poland
USA) under grant No. MEN/NSF-96-252. Two of the authors (BG, JP) are indebted to
the UC Davis Institute for High Energy Physics for the great hospitality extended to them
while this work was being performed.
Appendix
In this appendix, we present the analytic formulae for the signal and background cross
sections in an arbitrary fermionic final state. Fermionic couplings take the form i ef
(photon), i (cf + df 5 ) (Z), and i(af + i5 bf ) (Higgs). Here,

g
g
f
f
cf =
T ,
T3 2Qf sin2 W ,
df =
ef = eQf ,
2 cos W
2 cos W 3
and, for a SM Higgs boson, af = gmf /(2mW ) and bf = 0. In terms of these quantities,
the decay width of the Higgs boson to f f is given by

1
f mh af2 f2 + bf2 ,
0(h f f) =
8
q
where f 1 4m2f /s.
(22)
We employ the center of mass system with s = ECM and define + as the
angle of the + transverse polarization relative to that of the . The explicit expressions
for the 4-momenta and spin vectors of the + and in the laboratory frame are:
s
s
(1, 0, 0, ),
p+ =
(1, 0, 0, ),
p =
2
2
s = PL (, 0, 0, 1) + PT (0, cos , sin , 0),
s+ = PL+ (, 0, 0, 1) + PT+ (0, cos + , sin + , 0),
s
(1, f sin cos , f sin sin , f cos ),
pf =
2
s
(1, f sin cos , f sin sin , f cos ),
(23)
pf =
2
where and are the standard angles defined using polar coordinates for the final
fermion. The forms for s and s+ above are those which make the separation between
74
longitudinal and transverse polarization most evident. The conversion between these forms
and those given earlier in Eqs. (7) and (8) appropriate to making the precession physics
most transparent can be accomplished by the following mappings:
PH cos = PL ,
PH sin = PT cos ,
PV = PT sin .
(24)
We now give expressions for the invariant matrix element squared, |M|2 , for +
f f, summed over final spins and averaged over initial spins. These will be given in terms
of PL , PT and . Eq. (24) can be used to convert to the PH , PV and precession
variables. We divide |M|2 into three pieces: the absolute square of the Higgs diagram, the
interference between the Higgs diagram and the + Z exchange background diagrams,
and the absolute square of the + Z background diagrams. The symbols < and =
denote the real and imaginary parts. We give the results for m / s 0, an excellent

1
we
approximation for any reasonable Higgs mass. Defining X s m2X + i0X mX
have the following.
Absolute square of Higgs exchange:

af2 f2 + bf2 |h |2 s 2 a2 + b2 (1 + PL PL+ )

+ PT PT+ a2 b2 cos 2a b sin

= af2 f2 + bf2 |h |2 s 2 a2 + b2 1 + PL PL+ + PT PT+ cos( + 2) .
(25)
Higgs + Z interference:
1/2
mf f s 3/2
4 sin af a2 + b2

sin( + + )PT+ ef e =( h ) + cf (c + d PL )=(Z h )

sin( )PT ef e =( h ) + cf c d PL+ =(Z h )

cos( + + )PT+ ef e PL <( h ) + cf (c PL + d )<(Z h )

+ cos( )PT ef e PL+ <( h ) + cf c PL+ d <(Z h ) . (26)
Absolute square of + Z :

8df2 c2 + d2 1 PL PL+ + 2c d PL PL+ m2f |Z |2 s
+ 2 sin( 2) sin2 PT PT+ cf d ef e =(Z ) f2 s 2

+ cos( 2) sin2 PT PT+ cf2 + df2 c2 d2 |Z |2

+ ef2 e2 2 + 2ef e cf c <(Z ) f2 s 2

+ 2 f2 sin2 cf2 + df2 c2 + d2 1 PL PL+ + 2c d PL PL+ |Z |2

+ 2cf ef e c 1 PL PL+ + d PL PL+ <(Z )

+ ef2 e2 1 PL PL+ 2 s 2

+ 4 cos df cf 2c d 1 PL PL+ + c2 + d2 PL PL+ |Z |2

+ ef e d 1 PL PL+ + c PL PL+ <(Z ) f s 2 .
(27)
75
In terms of |M|2 , the cross section as a function of s, and is given by

1
d
=
f |M|2.
d cos d 64 2 s
(28)
For example, if we spin average the Higgs portion of |M|2 by combining = 0 and =
(so that the average of cos is zero), integrate over d cos and d, and use Eq. (22), then
we obtain the standard spin-averaged total cross section: (s) = 40(h + )0(h
f f )|h |2 . When this latter form is convoluted with a Gaussian distribution in s, one
obtains the result of Eq. (2).
References
[1] J.F. Gunion, H.E. Haber, G. Kane, S. Dawson, The Higgs Hunters Guide, Addison-Wesley,
Redwood City, CA, 1990.
[2] T.D. Lee, Phys. Rev. D 8 (1973) 1226.
[3] S. Weinberg, Phys. Rev. D 42 (1990) 860.
[4] A. Pilaftsis, Phys. Rev. D 58 (1998) 096010; Phys. Lett. B 435 (1998) 88.
[5] D.A. Demir, Phys. Rev. D 60 (1999) 055006.
[6] R. Casalbuoni, A. Deandrea, S. De Curtis, D. Dominici, R. Gatto, J.F. Gunion, Nucl. Phys.
B 555 (1999) 3.
[7] B. Grzadkowski, J.F. Gunion, Phys. Lett. B 350 (1995) 218.
[8] D. Atwood, A. Soni, Phys. Rev. D 52 (1995) 6271.
[9] J.F. Gunion, B. Grzadkowski, X.-G. He, Phys. Rev. Lett. 77 (1996) 5172; hep-ph/9605326.
[10] J.F. Gunion, J. Pliszka, Phys. Lett. B 444 (1998) 136; hep-ph/9809306.
[11] B. Grzadkowski, J. Pliszka, Phys. Rev. D 60 (1999) 115018; hep-ph/9907206.
[12] B. Grzadkowski, J.F. Gunion, Phys. Lett. B 294 (1992) 361; hep-ph/9206262.
[13] V. Barger, M.S. Berger, J.F. Gunion, T. Han, Phys. Rept. 286 (1997) 1; hep-ph/9602415; Phys.
Rev. Lett. 75 (1995) 1462; hep-ph/9504330.
[14] J.F. Gunion, in: Proc. Workshop Physics at the First Muon Collider and at the Front End of the
Muon Collider Batavia, IL, 69 November, 1997, p. 37; hep-ph/9802258; in: Proc. 5th Internat.
Conf. Physics Beyond the Standard Model, Balholm, Norway, 29 April4 May 1997, p. 234;
hep-ph/9707379.
[15] R. Raja, A. Tollestrup, Phys. Rev. D 58 (1998) 013005.
[16] C.M. Ankenbrandt et al., Phys. Rev. ST Accel. Beams 2 (1999) 081001.
[17] B. Autin, A. Blondel, J. Ellis (Eds.), Prospective Study of Muon Storage Rings at CERN, 1999,
CERN 99-02; ECFA 99-197.
[18] V. Bargmann, L. Michel, V.L. Telegdi, Phys. Rev. Lett. 2 (1959) 435.
[19] R. Raja, A. Tollestrup, private communication.
[20] D.M. Kaplan, Fermilab-Conf-00/019.
[21] We thank S. Geer for providing details regarding this option.

Quantum fluctuations of Wilson loops

from string models
Y. Kinar a,1 , E. Schreiber a,2 , J. Sonnenschein a,3 , N. Weiss b,4
a Raymond and Beverly Sackler Faculty of Exact Sciences, School of Physics and Astronomy,
Tel Aviv University, Ramat Aviv, 69978, Israel

b Department of Physics, University of British Columbia, Vancouver, B.C., V6T2A6, Canada
Received 2 February 2000; revised 10 April 2000; accepted 17 April 2000
Abstract
We discuss the impact of quadratic quantum fluctuations on the Wilson loop extracted from
classical string theory. We show that a large class of models, which includes the near horizon limit
of Dp branes with 16 supersymmetries, admits a Lscher type correction to the classical potential.
For a BPS configuration of a single quark in the AdS5 S 5 model, we show that the bosonic and
the fermionic determinants cancel, in particular confirming the absence of divergences. We find that
for the Wilson loop in that model, however, there is no such cancellation. For string models that
correspond to gauge theories in the confining phase, we show that the correction to the potential is
of a Lscher type and is attractive. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.15.-q; 11.15.Kc; 11.25.-w; 12.38.Aw; 12.60.Jv
Keywords: Wilson loops; Confinement; AdS/CFT correspondence; Semi-classical approximation
1. Introduction
The idea of describing the Wilson loop of QCD in terms of a string partition function
dates back to the early Eighties. In a landmark paper in this direction [1] it was found that
the potential of quarkanti-quark separated at a distance L acquires a c/L correction term
(c is a positive universal constant independent of the coupling) due to quantum fluctuations
of a NambuGoto (NG) like action. This term is commonly called the Lshcer term. An
Work supported in part by the USIsrael Binational Science Foundation, by GIF the GermanIsraeli
Foundation for Scientific Research, and by the Israel Science Foundation.
1 yaronki@post.tau.ac.il
2 schreib@post.tau.ac.il
3 cobi@post.tau.ac.il
4 weiss@physics.ubc.ca
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 3 8 - 8
Y. Kinar et al. / Nuclear Physics B 583 (2000) 76104
77
exact expression of the partition function of the NG action was derived in the large d
limit [2], where d is the spacetime dimension, and the 1-loop and 1/d expansions were
considered for strings and p-branes in various spacetime topologies [3,4]. The large d
result, when translated to the quarkanti-quark potential, takes the form
s
2c
L
E(L) 0 1 2 0 .
L /
Thus, by expanding it for small 2c/(L2 / 0 ), one finds the linear confinement potential as
well as the Lscher term. It was further shown that in this approximation the semiclassical
potential associated with Polyakovs action and NG action are identical [5]. It was later
realized that in fact this expression is identical to the energy of the tachyonic mode of the
bosonic string in flat spacetime with Dirichlet boundary conditions at L/2 [6].
Recently there has been a Renaissance of the idea of a stringy description of the Wilson
loop in the framework of Maldacenas correspondence between large N gauge theories
and string theory [7]. Technically, the main difference between the old calculations and
the modern ones is the fact that the spacetime background is no longer flat but rather an
AdS5 S 5 or certain generalizations of it. Conceptually, the modern gauge/string duality
gave the stringy description a more solid basis. The first modern computation [8,9] was
for the AdS5 S 5 metric which corresponds to the N = 4 supersymmetric theory. To make
contact with non-supersymmetric gauge dynamics one makes use of Wittens idea [10] of
putting the Euclidean time direction on a circle with anti-periodic boundary conditions.
This recipe was utilized to determine the behaviour of the potential for the N = 4 theory at
finite temperature [11,12] as well as 3d pure YM theory [13] which is the limit of the former
at infinite temperature. Later, a similar procedure was invoked to compute Wilson loops of
4d YM theory, t Hooft loops [13,14] and the quarkanti-quark potential in MQCD and
in Polyakovs type 0 model [1517]. A unified scheme for all these models and variety of
others was analyzed in [17]. A theorem that determines the leading and next to leading
behaviour of the classical potential associated with this unified setup was proven and
applied to several models. In particular a corollary of this theorem states the sufficient
conditions for the potential to have a confining nature.
The issue of the quantum fluctuations and the detection of a Lscher term was raised
again in the modern framework in [18]. It was noticed there that a more accurate evaluation
of the classical result [17] did not have the form of a Lscher term. This is, of course,
what one should have anticipated, since after all the origin of the Lscher term [1] is the
quantum fluctuations of the NG like string. The determinant associated with the bosonic
quantum fluctuations of the pure YM setup was addressed in [19]. It was shown there that
the system is approximately described by six operators that correspond to massless bosons
in flat spacetime and two additional massive modes. The fermionic determinant was not
computed in this paper. However, the authors raised the possibility that the latter will be
of the form of massless fermions and hence there might be a violation of the concavity
behaviour of gauge potentials [20,21]. One of the results of our work is that in fact the
fermionic operators are massive ones and thus the bosonic determinant dominates and
there is an attractive interaction after all. The impact of the quantum fluctuations for the
78
case of the AdS5 S 5 case was discussed in [22,23]. Using the GS action [2427] with a
particular symmetry fixing, it was observed that the corresponding quantum Wilson loop
suffers from UV logarithmic divergences. It was argued that by renormalizing the mass of
the quarks one can remove the divergence.
The computations of Wilson loops in 3d and 4d pure YM theory can be confronted
with the results found in lattice calculations. In particular, the main question is whether
the correction to the linear potential in the form of a Lscher term can be detected in
lattice simulations. According to [28] there is some numerical evidence for a Lscher
term associated with a bosonic string, however, the results are not precise enough
to be convincing. Obviously, our ultimate dream is compatibility with heavy meson
phenomenology.
In this paper our main goal has been the quantum corrections of the quarkanti-quark
potential in the class of non-supersymmetric confining theories. On the route to this target
we had to overcome several related obstacles. An important technical problem is the issue
of gauge fixing. Whereas for flat spacetime backgrounds there are several known fixing
procedure which are under full control, it turns out that for the non flat background the
situation is more subtle. Already in fixing the world sheet diffeomorphism we were facing
gauge choices that looked innocent but later were found to be problematic. The situation
with the fixing of the -symmetry is even trickier. We wrote the form of the bosonic action
truncated to quadratic fluctuations for the unifying scheme of [17]. As warm up exercises
we considered the fluctuations of a string in flat spacetime and the fluctuations of a BPS
quark of the AdS5 S 5 model. Whereas in the former case it is quite straightforward
to realize that the bosonic part yields the Lscher term and the fermionic part exactly
cancels it, in the later case the picture is more involved. Eventually, the basic expectation
that there are no divergent corrections to the energy was verified. However, this result
emerged only in a particular class of -symmetry fixing schemes. Without performing an
explicit computation, using a scaling argument we were able to write down the general L
dependence for a large class of models. It turns out that the set of models based on Dp
branes with 16 supersymmetries have a Lshcer type behaviour. Note that unlike the p = 3
case, for p 6= 3 there is no reason from dimensional grounds for the potential to be of the
form 1/L, and indeed the classical result is not of this form.
As for the case of the Wilson loop of the AdS5 S 5 model we found that there is only
partial cancellation between the bosonic and fermionic determinant. The net free energy is
of a Lscher form, but we were unable to determine neither its coefficient nor even its sign.
In spite of the fact that there is no GS action which corresponds to the non flat
background associated with confining gauge theories, we argue that there is a net attractive
Lscher type correction to the linear potential in the pure YM case.
The outline of the paper is as follows. We start with a brief description of the classical
setup. For that purpose we use the NambuGoto action associated with a spacetime
metric which is diagonal and depends on only one coordinate. The AdS5 S 5 is a special
case of this configuration and so are backgrounds that correspond to confining gauge
dynamics. In Section 3 we introduce quantum fluctuations to the string coordinates. We
expand the action to quadratic order in the fluctuations and write down the general form
79
of the operators whose (log) determinant determines the free energy of the system. When
translating to the gauge language the free energy has the interpretation of the quantum
correction of the classical quarkanti-quark potential. We discuss three possible gauge
fixing schemes and argue that only one of them, the normal coordinate fixing is a safe
gauge. As a warm-up exercise of computing and renormalizing the determinant, we derive
in Section 4 the Lscher term in the flat spacetime background. Section 5 is devoted
to an analysis of the dependence of the free energy on the separation distance L based
on a general scaling argument. An application of this result to the case of Dp branes
with 16 supersymmetries [29] shows that the quantum correction for this configuration
must also be of Lscher type. We then use, in Section 6, the general expressions of the
operators for the AdS5 S 5 case. We rederive these results in the framework of Polyakovs
action in Section 7. The next task of incorporating the fermionic fluctuations is discussed
in Section 8, again first in the simplest case of flat spacetime. We then proceed to the
case of a single BPS quark in the AdS5 S 5 background. We discuss the fixing of the
-symmetry and show that there are certain subtle issues involved. Eventually, using a
theorem about Laplace type operators, we show In Section 9 that the BPS free energy is
free from divergences due to cancellations between the bosons and fermions. The quantum
fluctuations of the Wilson loop in the AdS5 S 5 background or its generalization to the
cases of Dp with 16 supersymmetries are analyzed in Section 10. We argue that there is
left over uncancelled Lshcer term but we are unable to determine neither its coefficient
nor its sign. Sections 1113 are devoted to a similar analysis in the setup corresponding
to a confining behaviour. In Section 11 we show that the general case of this type behaves
in a similar manner to the pure gauge configuration, namely, that the Wilson line is
well approximated by a straight string parallel to the horizon and close to it. For this
type of backgrounds, in spite of the fact that we do not have a detailed GreenSchwarz
action we were able to show that the residual interaction is of an attractive nature. The
bosonic operators are considered in Section 12, while the fermionic ones are considered in
Section 13. We end this paper with a summary of our results and a list of open questions.
2. The classical setup

Let us first, before introducing quantum fluctuations, summarize the results derived
from classical string theory. Having in mind spacetime backgrounds associated with the
gravity/gauge dualities and MQCD, we restrict ourselves to diagonal background metrics
with components that depend only on one coordinate. The metric G for the coordinates
{t, x, yi , u, I }, where i = 1, . . . , p 1 and I = p + 2, . . . , 9 is given by

(1)
G = 0 diagonal Gt t (u), Gxx (u), Gyi yi (u), Guu (u), GI I (u) .
The classical (and, as we shall see, the one-loop) bosonic string has two equivalent
descriptions in terms of the NambuGoto action and the Polyakov action. Throughout this
work we incorporate quantum fluctuations mainly in the former picture. Comparison with
the analysis in the Polyakov formulation is presented in Section 7.
80
The NambuGoto action of a string propagating in spacetime with a general

background metric takes the form
Z
p
1
(2)
d d det h ,
S=
0
2
where
h = X X G (X).
(3)
The classical string configurations associated with Wilson loops are characterized by
boundary conditions which take the form of a loop on a boundary plane. The plane is
taken at some large but finite value u = us which later is taken to infinity. The form of the
loop considered in the present work is that of a very long strip along a spatial direction
(L/2 < x < L/2) and a temporal length T with L T . In certain models we will
take the temporal direction to be along another spatial direction.
Since we assume invariance under time translation, we consider classical static string
solutions. These are spanned in the x, u plane by the function u(x) and we set yi = 0,
I = 0.
The NG action is invariant under world-sheet coordinate transformation. It is convenient
to use the static gauge in which we set the world-sheet time to be identical to that of the
target space, = t. Various different fixings of will be discussed in the next section when
quantum fluctuations are incorporated. For the classical configuration we take = 2x.
Defining the two functions
f 2 (u) Gt t (u)Gxx (u),
g 2 (u) Gt t (u)Guu (u),
the action becomes
Z
Z
q
S = T L dx = T dx f 2 (u) + g 2 (u) (x u)2 .
(4)
(5)
Using the fact that the Lagrangian does not depend explicitly on x we write the equation
of motion in terms of the conserved generator of translations along x so that
p
f (ucl ) f 2 (ucl ) f 2 (u0 )
(6)
x ucl =
g(ucl )
f (u0 )
where u0 is the minimal value of u reached by the string.
The classical profile of the Wilson loop is a solution of (6) which satisfies the boundary
conditions stated above. Since, in the language of the corresponding gauge model, the
quark and anti-quark are set at the coordinates x = L/2, the relation between L and u0
is given by
Zus
L=2
u0
f (u0 )
g(u)
p
du
f (u) f 2 (u) f 2 (u0 )
and the action is given by
(7)
Zus
S = 2T
u0
f 2 (u)
g(u)
p
du.
f (u) f 2 (u) f 2 (u0 )
81
(8)
The basic conjecture of the string/gauge duality is that the natural candidate for the
expectation value hW i of the Wilson loop is proportional to the partition function of the
corresponding string action
Z
(9)
hW i Z = DX exp (S),
where the integral is over all surfaces whose boundary is the given loop. Moreover, hW i is
related to the quarkanti-quark potential energy E(L) via

hW i exp T E(L) ,
(10)
so that in fact (8) determines the potential in the classical limit in which the configuration
of least action dominates. However, it is easy to see that the expression (8) is linearly
divergent as us and hence has to be renormalized. The mass of the quarks which
translate in the string language into a straightR line between u = 0 (or u = uh in case there
u
is an horizon at uh [12]), is given by mq = 0 s g(u) du. It is thus physically natural to
determine the quarkanti-quark potential as E = (S/T ) 2mq . For the special case of the
AdS5 S 5 model this definition of the energy matches the Legendre transform of the action
suggested in [30].
By implementing this expression for various different background metrics, Wilson loops
corresponding to certain gauge systems were computed [17]. Moreover, a theorem was
stated in [17] determining the dependence of E as a function of L for the general setup
of (5).
3. Quadratic fluctuations and gauge fixings in the NG action

In order to account for the contributions to the Wilson loop from quantum fluctuations
we expand the coordinates around their classical values
x (, ) = xcl (, ) + (, ),
that is,

t = tcl + t ; x = xcl + x ; u = ucl + u ; yi = yi ; I = I .
(11)
(12)
The next step is to expand the NambuGoto action up to terms quadratic in the fluctuations
so that S = Scl + S(2) , and compute the following path integral
Z
Y
Y
(13)
Dyi
DI exp{S(2)},
Z(2) = Dt Dx Du
where we write
XZ
d d a Oa a
S(2) =
a
(14)
82
with a the various fields. 5

The Gaussian integration over the fluctuations yields (after gauge fixing) a product of
Q
determinants of second order differential operators a det Oa so that the corresponding
free energy is given by
X1
log det Oa .
(15)
FB = log Z(2) =
2
a
As was mentioned above, the NG action is invariant under world-sheet reparameterizations, and in order to compute explicitly (13) one has to introduce a gauge choice. In fact,
without gauge fixing the operators are degenerate due to the reparameterization invariance
[27]. Obviously, the set of operators Oa depends on the gauge. In the following subsections we write down the form of the operators in three different gauges. We set in all the
three gauges = t so there are no fluctuations in the time direction. The gauges differ in
the way we fix and the choice of the fluctuating coordinates. In the first gauge we fix
the u coordinate and compute the fluctuations of x, yi and I while in the second one x is
fixed and u, yi and I fluctuate. In the third gauge the fluctuations are taken to be along the
normal coordinate to ucl (in the ux plane) and along yi and I .
3.1. The fixed u gauge
Imposing the assignment = u (running from u0 to ) one finds that to the second
order in the fluctuations x and yi , the expression h = det h takes the form

X Gy y (u)
Gxx (u)
i i
(t x )2 +
(t yi )2
h(2) = g 2 (u) 1 +
Gt t (u)
Gt t (u)
i

+ f 2 (u) (u xcl )2 + 2(u xcl )(u x ) + (u x )2

X Gy y (u)
Gyi yi (u)
i i
(u yi )2 +
(u xcl )2 (t yi )2 ,
(16)
+ f 2 (u)
Gxx (u)
Gt t (u)
i
where xcl (u) is defined by ucl (xcl (u)) = u.

Simplifying the last two terms (using (6)) and integrating by parts, we find that the set
of operators Oa includes two types of quadratic operators:
2
q

1
g(u) Gxx (u) 2
(f (u) f 2 (u0 ))3/2
2
2
u + f (u) f (u0 )
,
Ox = u
2
f (u)g(u)
f (u) Gt t (u) t

q
Gyi yi (u)
1
f (u)
u
f 2 (u) f 2 (u0 )
Oyi = u
2
Gxx (u)
g(u)

Gy y (u)
g(u)f (u)
p
t2 ,
(17)
+ i i
2
Gt t (u)
f (u) f 2 (u0 )
and similarly for OI with GI I (u) replacing Gyi yi (u). The boundary conditions we
impose on the eigenfunctions are (u, 0) = (u, T ) = 0, (, t) = 0 and (u0 , t) = 0
5 Our treatment differs from that of [22] in that the authors of that article explicitly include in the integrals of
(0)
(14) the measure (det h )1/2 corresponding to the classical solution, and change the operators accordingly.
83
or 0 (u0 , t) = 0 (the conditions at u0 imply that the function should be symmetric or

antisymmetric around x = 0).
For a p + 1 dimensional spacetime (t, x, yi ), the bosonic free energy is given by:
1
p1
(8 p)
log det Oy
log det O .
(18)
FB = log det Ox
2
2
2
We can change variables (i.e., change world-sheet coordinates) to = xcl and obtain the
quadratic operators:

f (u0 )
f 2 (u0 )
Gxx (ucl ) f 2 (ucl )
b
x 1 2
x +
1 t2 ,
Ox =
2
Gt t (ucl ) f 2 (u0 )
f (ucl )

2
byi = f (u0 ) x Gyi yi (ucl ) x + Gyi yi (ucl ) f (ucl ) t2 ,
(19)
O
2
Gxx (ucl )
Gt t (ucl ) f 2 (u0 )
bI where the boundary conditions are (L/2, t) = (L/2, t) = 0. The
and similarly for O
free energy is still given by (18).
3.2. The normal coordinate gauge
In this gauge we take the fluctuations to be everywhere normal to the classical curve (see
Fig. 1). With this choice, both ucl and xcl are changed by the fluctuation to the final values
u and x. We further choose the gauge = t (as before), and = ucl . The components
of the tangent vector to the classical curve obey tu /tx = x ucl which is given by (6), and
therefore the components of the unit normal vector are the solutions of
nx (ucl ) tx (ucl ) Gxx (ucl ) + nu (ucl ) tu (ucl ) Guu (ucl ) = 0,
nx (ucl )2 Gxx (ucl ) + nu (ucl )2 Guu (ucl ) = 1.
(20)
We denote the magnitude of the fluctuation along the normal direction by the
dimensionless field n (ucl , t). The coordinate vector in the (x, u) plane is thus
x = xcl (ucl ) + nx (ucl ) n (ucl , t),
u = ucl + nu (ucl ) n (ucl , t),
(21)
and both ucl and xcl change after the fluctuation to the final values u and x.
Unlike the fixed u gauge, the fluctuations result also in a change in the metric. In our
gauge the off-diagonal elements are zero and the diagonal elements, to second order, are
given by:
G (X) = G (ucl ) + u G (ucl ) nu (ucl ) n (ucl , t)
2
1
(22)
+ u2 G (ucl ) nu (ucl ) n (ucl , t)
2
(with no summation over indices). As the action is an integral of L essentially from u =
to u = u0 and back again to u = , and we have the boundary conditions n (, t) = 0,
the part of L linear in n does not have to vanish, but rather should be a total derivative of
the form u (F(1) n (ucl , t)), in order to ensure that the classical solution is an extremum.
We find this to be indeed the case. Inspection of h readily reveals that there are in h no
quadratic terms of the types t n n or t n u n (and also no linear term of the type
84
Fig. 1. The fluctuations in the normal gauge.
t n ). By adding a total derivative of the type u (F(2) n (ucl , t)2 ) we can also get rid of the
u n n term. Finally, we get the quadratic expression
L(2) = Lcl + C (ucl )n2 + Ct t (ucl )(t n )2 + Cuu (ucl )(u n )2 .
Thus the operator On takes the form

On = u Cuu (ucl )u Ct t (ucl )t2 + C (ucl ).
(23)
(24)
Since the explicit expressions of the Cs are cumbersome for the general case we do not
find it useful to write them down. In Sections 6, 12 we derive them for certain special cases.
3.3. The fixed x gauge
Here we impose = x and fix the gauge by x = 0. Expanding now the square root of
h to quadratic order yields yet another operator for the longitudinal fluctuations. We write
it this time in the form of the action.

Z
f (u0 )
Guu (u)
g(u)2 f (u0 )2
2
2
2
2
(
(
)
+
)
+
(
)
d d
S(2) =
u
u
u
2
f (u)4
Gt t (u)

Gyi yi (u)Gxx (u)
Gyi yi (u)
( i )2 +
( i )2 . (25)
+
Gxx (u)
f (u0 )2
The mass term 2 ( ) will be written explicitly for specific metrics. The fact that there
is a mass term present in this gauge, and in the normal coordinate gauge, and there is no
85
such term in the fixed u gauge is not surprising since such a term necessarily arises when
G (u) is expanded about ucl . In the gauge = u there is no fluctuation in u so no such
term arises. The significance of this term will be discussed further when we study special
cases such as AdS and pure YangMills.
3.4. Comparing the different gauges
One expects that physical, gauge invariant quantities should be identical when using
different gauges. This general belief is justified provided that one considers legitimate
gauges. The differential operators derived in the three different gauges are manifestly not
identical. It may happen that their determinant is nevertheless identical. Formally, one can
pass from one gauge to another. For instance changing the integration variable from = x
to = u together with the rescaling of u
u = x x u(x)
(26)
maps between the fixed x gauge to the fixed u one. However, this rescaling itself may be
singular as indeed occurs at u0 .
The computation of the operator Ox in the fixed u gauge corresponds to the assumption
that the longitudinal fluctuation of the string is in the x direction (gauge fixing the
coordinates u, t as , , respectively). This assumption leads to two problems in the
physical interpretation. First, near u = u0 , even a small fluctuation can lead to multiple
valuedness of x as a function of u. Second, the extremal point can itself fluctuate in the u
direction, and no longer correspond to u = u0 (see Fig. 2). The problem is that at u = u0 the
x direction is tangential to the classical curve, and that choice of the direction of the field of
fluctuations becomes singular. As the metrics depend only on u, this choice of fluctuation
involves no change in metric. It is easy to understand, therefore, why the operator Ox
involves only derivatives of the fluctuating field.
By choosing to be x, and the direction of fluctuations to be along u, the limit u =
becomes singular in the aforementioned sense. However, this singularity exists also in the
fluctuations of the bare quarks, and after the subtraction of the quark masses from (8) it
might cancel.
The safest choice, however, seems to be taking the fluctuations to be anywhere normal
to the classical curve, as previously explained. 6
Fig. 2. A problematic fluctuation in the fixed u gauge.

6 In [22] the computations were performed in the Riemann normal coordinate approach. This method is
similar to the normal coordinate we are using.
86
4. Bosonic string in flat spacetime

We can now easily derive the free energy of a bosonic string in flat spacetime (Lscher
term). The metric is given by Gt t = 1; Gxx = Gyi yi = 1, where i = 1, . . . , D 2. Note
that u is now one of the yi coordinates. The classical solution in this case is just a straight
line along the x direction from L/2 to L/2. Obviously, for this case the three different
gauges are identical. We use (19) and regard only the transverse fluctuations operator
byi ). This operator is the ordinary Minkowski Laplacian. Considering now Euclidean
(O
space time and demanding the eigenfunctions to vanish on the boundary, the eigenvalues
of the Laplacian are (minus)
2

n
m 2
+
(27)
En,m =
L
T
and the free energy is given by
Y
X n 2 m 2
2
FB = log det 1 = log
En,m =
log
+
D2
L
T
n,m
n,m
XT Z
n
(28)

X T n
n 2
+ o(L)
d log 1 +
+ o(L) =
L
L
n
X
n + o(L),
=T
L n
(29)
where we assumed that T is large.

Regulating this equation using the Riemann zeta function and discarding (infinite) terms
that do not depend on L, we find
T
.
(30)
24 L
This expression also gives a quantum correction to the linear quarkanti-quark potential
in a bosonic QCD-string model in D-dimensional flat space. Using the Wilson loop to
calculate this potential we have
FB = (D 2)
1
1
1V (L) = FB = (D 2)
T
24 L
which is called the Lscher term.
(31)
5. A general scaling result

The argument of the previous section can be generalized to operators which do not
necessarily correspond to the flat case and will lead to a general scaling result that will
prove useful in our work.
Let us define the operators
O[A, B] = A2 Ft (v)t2 + B 2 Ov
(32)
87
with t the time, v a coordinate whose range is independent of L. Ov is a general operator of

v alone, independent of L, and A, B are constants which may depend on L. Then, V [A, B],
defined as the correction to the potential arising from O[A, B], is proportional to B/A. In
particular, the potential is independent of any overall factor multiplying the operator O.
The idea of the proof is that by redefining t we can scale the operator O[A, B] to O[1, 1],
and the factors by which the eigenvalues get multiplied in this operation become irrelevant
in the large T limit. We proceed to the proof itself.
Separation of variables shows that the eigenfunctions of O[A, B] are (v, t) = eit (v).
en (0 ) by
We define the eigenvalues E

en (0 ) (v)
(33)
=E
Ft (v) 02 + Ov (v)
and En (0 ) by

A2 Ft (v) 02 + B 2 Ov (v) = En (0 ) (v)
or
(34)

Ft (v)(A0 /B)2 + Ov ) (v) = B 2 En (0 ) (v)
(35)
so that
en (A0 /B).
En (0 ) = B 2 E
(36)
The boundary conditions for the time coordinate lead to = m/T for integer m. Hence
we have for V [A, B] (when o(T ) designates a term which does not depend on T , or grows
sublinearly with it),
1
log det O[A, B]
2

1 XX
m
log En
=
2 n m
T

1 XX
en A m + 2 log B
log E
=
2 n m
B T

1 XX
en m
+ o(T )
log E
=
2 n m
BT /A
T V [A, B] + o(T ) =
BT
V [1, 1] + o(T ),
(37)
A
and so indeed V [A, B] = (B/A) V [1, 1].
We can make a simple consistency check of our result. Let us look at the flat Laplacian
1 = t2 + x2 . The range of v = x/L is 0 6 v 6 1 independent of L, and 1 = t2 + L2 v2 ,
so by the above result we get V (L) L1 as we have indeed seen in Section 4.
=
6. Bosonic string in the AdS 5 S 5 background

In AdS5 S 5 background the metric can be written in the following form
Gt t = Gxx = Gyy =
u2
;
R2
Guu =
R2
;
u2
G = R 2 ,
(38)
88
where y is a transverse coordinate of the AdS5 , while is a transverse coordinate in S 5 .

We therefore get
u2
,
R2
g(u) = 1.
f (u) =
(39)
(40)
Note that as we deal with small fluctuations and big R, the angular coordinates of S 5 can
be treated as if they were not compactified.
Unlike the flat case, for the AdS5 S 5 metric the different gauge fixings yield different
operators.
In the fixed u gauge one finds using (19) the following quadratic operators

4
u2
u4
bx = 1 0 x 1 0 x u 1 t2
(41)
O
2 R2
u4
u40
and

4
u2
by = 1 0 x2 u t2 ,
O
2 R2
u40
4
4 2
u2
b = 1 0 x R x R u t2 .
O
2 R2
u2
u40
(42)
In the normal fluctuations gauge, the fluctuations along the directions y, which are
perpendicular to the (x, u) plane are the same as (42). However, for On we have
q

2u8 5u4 u40 + 3u80
R4
1
0
2
4
4
q
q
t u
u u0 u .
(43)
On =
2
2u6 u4 u40
2 u4 u40
In the dimensionless variable v = u/u0 , it is easy to see that this operator is of the form
2
(32), with A2 = R 4 u2
0 , B = 1. We conclude, according to the result of Section 5, that
the correction to the potential is proportional to R 2 u0 , that is [8], inversely proportional
to L. Moreover, there is no further dependence on R when we eliminate the variable u0
in favour
of L. The classical and quadratic quantum quarkanti-quark potential add up
2 N R 4 the t Hooft coupling constant, and c 0 a
to V /L + c0 /L with gYM
universal constant, independent of . We see that the quantum expansion in h is equivalent,
for large , to an expansion in R 2 or equivalently in 1/2 . The quantum expansion does
not spoil, of course, the conformal nature of the field theory. The other operators for the
normal gauge can be seen, by similar reasoning, to give rise also to a potential correction
of the same characteristics.
Note also that if we approximate the classical curve to be flat over most of its range and
thus approximate u = u0 throughout, the mass term drops out and the operator simplifies
to the flat Laplacian,
b 0x 1 t2 1 x2 ,
(44)
O
2
2
and for the longitudinal direction we get exactly the same Lscher term as in the flat case.
The same happens also to the transverse directions. (Note that On0 cannot be approximated
89
in this way since u is constant and it is thus not integrated over). The shape of the string,
however, does not change with L but only scales. It is thus not a good approximation, in our
case, to assume that the string is flat. In Section 11 we will give quite a stringent criterion
for flatness, and prove that strings in metrics corresponding to confining field theories are
indeed flat.
When we revert to a differential operator involving the coordinate x instead of u (with
the integration of the Lagrangian in the t, x coordinates), the operator (43) is translated to
2 2
2u8 5u4 u40 + 3u80 R 2 u2 2
R u0
b 0n =
x .
(45)
O
x
t
2
2
2u2
2R 2 u4 u0
2u0
In the fixed x gauge we have again the same transverse operators, while the longitudinal
operator becomes
6 2
5u60 2u4 u20 u20 R 2 2
u R
0
b
+
+ x 0 8 x .
(46)
Ox =
u6 R 2
2u4 t
2u
Note that we can use the same scaling arguments as for the normal gauge to show that
the correction is inversely proportional to L. Another remark is in order. The mass term
is more properly looked at as a shift in the levels of the massless modes rather than as a
real mass term, since it behaves like u20 /R 4 which is proportional to 1/L2 . Thus the mass
gap in this case will be of the same order as the energy of a massless mode in a box of
length L.
7. The bosonic string in the AdS 5 S 5 background Polyakovs action
The analysis of the quadratic quantum fluctuations can be performed, as was discussed
at the end of Section 2, also in the framework of Polyakovs action. Here we present a
rederivation of the bosonic determinant of the AdS5 S 5 Wilson loop using the Polyakov
action. Recall that the action (2) is given by
Z
1
(47)
d d h h X X G (X)
S=
2
with the metric given in (38). This action is minimized by the classical configuration
t (, ) = ;
u(, ) = ucl ( );
x(, ) = ;
y (, ) = 0;
i
I (, ) = 0.
(48)
The function ucl ( ) satisfies the equation [8]

uZ
cl /u0
y2
u0
dy
p
= 2 ,
4
R
y 1
(49)
where u0 = u( = 0), the minimal value of ucl , is determined by the boundary conditions
at = L/2 where u , to be
2 2 3/2 R 2 1
L .
(50)
u0 =
( 14 )2
90
The classical value of the worldsheet metric is determined by the equations of motion
derived by variation of the action with respect to h ,
h = X X G (X)
(51)
which take the following form

(h )cl =
u6cl
u40 R 2
(h )cl =
u2cl
.
R2
(52)
The quantum fluctuations of the string coordinates are introduced as in (12), and those of
the world sheet metric are as follows
h = (h )cl (1 + ),
p
h = (h )cl (h )cl ,
h = (h )cl (1 + ),
(53)
where , , parameterize the metric fluctuations.

We now use the reparameterization invariance of the action to choose the gauge in
which
u (, ) = t (, ) = 0,
(54)
so that
u = ucl ( ),
x t = ,
x1 = + 1,
x2 = 2,
x3 = 3.
(55)
Note that we do not consider here the fluctuations in the directions.

Next we expand the action (47) to quadratic order in the s and the s. The first order
correction of the classical action S(1) is expected to vanish and indeed
Z
u2
(56)
S(1) = 02 d d 1
R
vanishes by the boundary conditions on 1 . The quadratic term is given by:
4

Z

ucl
1
1
2
2
i 2
d d
+ ( ) +
S(2) =
2
4
R 2 u20

2u2
2
u2
+ 02 i + 1 ( ) 2cl 1 .
R
R
(57)
Our next step is to perform the Gaussian integrals over the s. Recall that our goal is to
compute
Z
3
Y
D i exp{S(2)},
(58)
Z(2) = Dh
i=1
where we integrate over the fluctuations of h . This can be translated to integrations of

the various s. The result of the integral is:
Z(2) =
3
Y
i=1
with
D i exp{Seff }
(59)
1
Seff =
2

4

u40
u20
ucl
1 2
1 2
1
+ 1 4
d d
R2
u40
ucl

4

u20
ucl
2
2
R2
R 2 u20
91
(60)
where the notation refers to 2 and 3 . It is thus clear that in the fixed u gauge the
bosonic operators derived from Polyakovs action are identical to those found from the NG
action (42).
8. Introducing fermionic fluctuations

So far we have considered only quantum fluctuations of the bosonic degrees of freedom.
Recall, however, that gauge/gravity duality was originally [7] proposed in the context
of superstring theories which include worldsheet fermions. This, together with the hope
that supersymmetry will lead to the cancellation of divergences, raises the question of
the contribution of fermionic fluctuations to the free energy. It might seem an easy task
to truncate the NSR action and collect terms quadratic in the fermionic fluctuations.
This is indeed the case for a flat background with vanishing RamondRamond form.
However, there is so far no satisfactory formulation of the world-sheet supersymmetric
NSR action for a background with non-vanishing RR forms. Recall that in the AdS5 S 5
model, the self dual 5-form plays an essential role. Hence we use here the manifestly
spacetime supersymmetric GS approach. For the AdS5 S 5 background this action has
been constructed in [24] as a supersymmetric sigma model with the coset supermanifold
SO(2, 2|4)/(SO(1, 4) SO(5)) as its target space. The GS action is invariant under local
-symmetry. By gauge fixing this symmetry one reduces the number of fermionic degrees
of freedom by a factor of two so that it matches the number of bosonic degrees of freedom.
In [27] a 2d scalarscalar duality transformation was invoked to transform the gauge fixed
action into an action which is quadratic in the fermions.
In this section we first consider the fermionic contribution of the supersymmetric theory
in flat spacetime. We then use the gauge fixed action [27] to deduce the contribution of
the fermionic quantum correction in the AdS5 S 5 background.
8.1. Fermions in flat spacetime
For flat Minkowski spacetime we choose the super-Poincar group as our target space.
When we explicitly consider a flat classical string, the fermionic part of the simplified
gauge fixed GS-action in the flat spacetime is simply
Z
(61)
SFflat = 2i d d i i ,
where is a WeylMajorana spinor and i are the SO(1, 9) gamma matrices. The
fermionic operator is, hence
bF = DF = i i
O
(62)
92
and squaring it we get 1 = x2 t2 . The total free energy of the supersymmetric case
is therefore (using the fact that for D = 10, we have 8 transverse coordinates and 8
components of a WeylMajorana spinor),

1
(63)
F = 8 log det 1 + log det DF = 0,
2
and there is no Lshcer term.
8.2. Fermions in AdS5 S 5
The target space for the GS action in the AdS5 S 5 background is the coset superspace
SU(2, 2|4)/(SO(1, 4) SO(5)) [24]. The action is found to be in the form of the usual
supersymmetric string action (but with the induced non-flat metric), with a change of the
derivative acting on the spinor variables 1 , 2 . If eia is the induced worldsheet vielbein,
and iab the induced worldsheet spin connection, then
i I Di I Di I
i IJ a a J
ei
2
(64)
with Di = i + 14 iab ab the usual induced worldsheet covariant derivative. The interpretation is that the metric influences the action through Di while the RamondRamond flux is
responsible for the second term in (64). Using the gauge fixing suggested in [2527], the
fermionic part of the action for the classical solution can be computed and it leads to the
operator
q
!
4
4
u20 1
u4cl 0 u2cl ucl u0 2
bF =
x +
+ 4
t ,
(65)
O
R2
R
u20 R 2
u20
where we use gamma matrices of SO(1, 4) which is the AdS5 tangent space. Squaring this
operator, we find

2
u4cl 2 R 2
R b 2
2
by .
O
=
t = 2 O
(66)
F
x
u20
u40
u0
We find therefore, that the transverse fluctuations partially cancel the fermionic fluctuations
and we are left with six fermionic degrees of freedom and the longitudinal and angular
bosonic fluctuations.
9. Bare quark in AdS 5 S 5

In this section we investigate the single bare quark, which is a flat string in AdS5 S 5
space stretching along the AdS radial coordinate. In this case the string is a BPS state and
we expect neither its charge nor its mass to be modified by quantum fluctuations. This
problem is related to issues associated with certain BPS soliton solutions [3133].
The AdS5 S 5 metric is

ds = GMN dx dx =
2
2

R2
u
2
du +
dx dx + R 2 d52 ,
2
u
R2
93
(67)
where we take {x0 , x1 , x2 , x3 } with Minkowskian signature.

The classical solution of the bare quark is = u, = t, xi = 0 and the S 5 angles I = 0.
It is obvious that there is a separation of the quadratic action, and the fluctuations of the
fields can be taken one at a time. We shall take the static gauge of the action, so that the
fluctuating fields are the three xi and the five I .
For the five angular fields, since we are working in the large R regime, the curvature of
S 5 is not felt, and we can approximate S 5 locally by the flat space R 5 . We take one field
(, ) along one angle , i.e., study the worldsheet with (u, t, ) = (, , (, )). We
have

(u/R)2 + R 2 2
R 2 0
M
N
(68)
h = GMN x x =
(R/u)2 + R 2 02
R 2 0
so that

(69)
h = det h = 1 + R 4 /u2 2 u2 02 .
The NambuGoto action (2) is therefore S h = 1 + S(2) + O( 4 ) with the quadratic

action
S(2) (R/u)2 2 (u/R)2 02

= (R/u)2 t2 + (u/R)2 u2 1/R 2 ,
(70)
where we have used integration by parts and the Dirichlet boundary conditions. We
therefore see that for the five angular bosonic fluctuations, the quadratic operator is O =
(R/u)2 t2 + (u/R)2 u2 1/R 2 .
For the three spatial coordinates, we want to keep the fluctuating field dimensionless, so
we take (u, t, x) = (, , (, )/u). A similar calculation yields in this case the operator
O = (R/u)2 t2 + (u/R)2 u2 3/R 2 .
Regarding the fermionic fluctuations around the classical bare quark configuration, one
can see that the -symmetry fixing suggested at [25,27] is not applicable, and produces a
degenerate operator. We should therefore consider only the quadratic part of the action and
try another way to fix the gauge. The quadratic part of the GS action is, in this case,

u
u
R
R
(2)
LF = 0 i 0 D0 + i 4 D1 i 0 D1 + i 4 D0 0
u
R
R
u

u
u
R
R
+ 1 i 0 D0 + i 4 D1 + i 0 D1 i 4 D0 1
u
R
R
u

1

0
0
0
4
1
0
4
+ 1 + 1 .
(71)
Fixing the -symmetry by setting 0 = 1 and writing the covariant derivative explicitly,
D0 = 0 + (u/2) 0 4 , D1 = 1 , we have
R
u
1 4
i
+ 0 4.
OF = 0 t + 4 u +
u
R
2R
R
(72)
94
Squaring this operator, integrating by parts and ignoring total derivatives, we find the
operator
7/4
(73)
OF2 = (R/u)2 t2 + (u/R)2 u2 2 .
R
One should note, however, that taking another gauge fixing, which is a-priori as good as
ours (e.g., 1 = i 3 0 ) leads to a different result. We prefer our gauge fixing as it leads
to the expected cancelation of the divergences.
All the quadratic operators of the dynamical degrees of freedom are seen to be of the
form Oj = (R/u)2 t2 + (u/R)2 u2 cj /R 2 . There are eight bosonic ones, corresponding
to the eight coordinates left after fixing the worldsheet reparameterizations, and eight
fermionic ones. The fermionic operators were squared, and are taken with a negative sign,
as the fields are complex Grassmannian. Therefore, the quadratic quantum contribution to
the mass is
!
X
X
1
log det Oj
log det Oj .
(74)
1m =
2T
bosons
fermions
We were unable to find explicitly the eigenvalues of those operators and multiply them
using some regularization scheme. However, it is well known that the logarithm of the
determinant of such second order operators has in general quadratic, linear and logarithmic
divergences in the cutoff scale . Moreover, the corresponding coefficients are known (see,
for example, [5]). The sheer fact that the number of fermionic and bosonic operators is
the same, and that they differ only in the mass term cj /R 2 is sufficient to ensure that
the quadratic and linear divergences cancel out in (74). The logarithmic divergence also
cancels out provided that
X
X
cj
cj = 0.
(75)
bosons
fermions
Indeed, we have 3 3 + 5 1 8 7/4 = 0. We therefore see that there is no infinite

part, needing renormalization, to the aforementioned contribution to the bare quark mass.
We were unable to show, however, that the finite part also vanishes, as expected from
supersymmetry.
10. Fluctuations of the Dp backgrounds

In this section we show that the quadratic contribution of the quantum fluctuations to
the potential is proportional to 1/L in the general Dp background (p 6 4). In the D3 case,
where the result is dictated by conformality, this was shown in Section 6.
For metrics relevant to Dp-branes, Gt t = Gxx = Gyy and
f (u) = auk ,
(76)
g(u) = bu ,
(77)
with arbitrary a, b 6= 0, k, j (for those specific metrics, a R (7p)/2 ). The power-like

behaviour of f, g is essential in order to get the desired result.
95
The operator for transverse fluctuations is given by (17). Inserting f, g to that equation
and changing the variable u to v = u/u0 (which has the range 1 6 v < regardless of L),
we get

p

1
a 2 2kj 2
vj
j
2
2kj
2k
t + u0
v v
1v
v ,
(78)
Oy = bu0
2
b
1 v 2k
which is of the form (32) with
j
A2 = bu0 ,
B2 =
a 2 2kj 2
u
.
b 0
kj 1
Therefore, by the result of Section 5, the potential is proportional to B/A = ab u0

b j +1k
,
a u0
. As it
L1 ,
we finally get that V

that is, the correction to the
is known [17] that L
potential resulting from transversal bosonic fluctuations (and from all fermionic ones) is
inversely proportional to L. Unfortunately, we do not know the constant of proportionality,
and not even its sign. This constant of proportionality, however, cannot depend on R since
the constants a, b cancel out. Indeed, for p 6= 3, the t Hooft coupling constant R 7p is
dimensionful, and therefore can not enter the aforementioned constant of proportionality.
The classical and quantum quadratic potentials add up to V 1/(5p)/L2/(5p) + c0 /L
and the quantum expansion is in 1/(5p) .
One can easily see that the scaling for the longitudinal bosonic operator Ox in the fixed
u gauge is essentially identical, as it can be casted to the form (32) with the same A, B as
in the transverse case. Therefore, we conclude that the correction arising from this gauge
fixing also obeys V L1 . The picture in the normal gauge, which we argued is safer, is
more involved. We have computed it explicitly and shown the L1 behaviour only in the
p = 3 case.
11. Flatness of the string in the confining case

The string in the confining pure YangMills case was shown [19] to be very flat for
large L. This result will be needed for the next section. In this section we shall generalize
this result in a precise manner for all metrics giving confinement. We shall pick a constant
value w (independent of L) and inquire at what distance d from the ends of the string
(at x = L/2) the string reaches u = w. We shall find that we cannot show that d is
independent of L but rather we find that it can grow with L. This growth is however mild
and the ratio d/L tends to zero as L grows (see Fig. 3).
The analysis requires the Taylor expansion of f (u) and g(u) (we use the conventions of
[17]),

(79)
f (u) = f (0) + ak uk + O uk+1 ,

j
j +1
(80)
g(u) = bj u + O u
with f (0) 6= 0 due to confinement. We pick w to be sufficiently small so that all our
subsequent approximations are valid.
96
(a)
(b)
Fig. 3. (a) The classical solution is rescaled as L in the AdS5 S 5 background while (b) it
becomes flatter in the confining scenario.
For small u, substitution of (79), (80) in (6) gives

q
2f (0)ak (uk uk0 )
,
u0 x u
b j uj
(81)
or, with v u/u0 ,
2f (0)ak k/2j 1 v k 1
vk 1
k/2j 1
0
u0
cu0
.
v
j
bj
v
vj
(82)
The critical case k = 2(j + 1) is the generic one (k = 2, j = 0) and also corresponds
to the pure YangMills case (k = 1, j = 1/2, where the horizon u = uT should be
substituted for u = 0). In this case
vk 1
0
6 cv k/2j = cv,
(83)
v c
vj
so that (log v)0 = v 0 /v 6 c, or log v 6 log v0 + cx. Therefore, v 6 v0 ecx , or finally u 6
u0 ecx .
If we demand u0 ecx 6 w, we get
x6
1
log(w/u0 )
log u0 .
c
c
(84)
On the other hand, it is known [17] that in the critical case

1
1
log u0 = L + O(log L),
c
2
(85)
so if x 6 12 L O(log L) it is guaranteed that u 6 w. In other words, d 6 O(log L).

In the non critical case k > 2(j + 1), the minimum of f (u) is more flat, and we expect
the results to be somewhat weaker. Using arguments similar to those in the previous case
(but this time not approximating v k 1 v k ), and using results from [17], one can show
k/2j1
1
k/2j
that in this case d 6 O(L k/2j ). d can grow with L, but the ratio d/L 6 O(L
tends indeed to zero as L grows.
97
12. Bosonic fluctuations in the pure YangMills setup

When one of the spatial coordinates of the AdS5 S 5 metric is compactified on a circle
of radius u1
T , with the proper boundary condition [13], the modes in the compact
dimension become KaluzaKlein modes with masses of the order of uT , and the fermions
and scalars of the conformal four-dimensional theory also acquire masses of the same order
and decouple in the low energy regime of the theory. Therefore, the N = 4 conformal
theory in four dimensions is deformed to a theory similar to non supersymmetric Yang
Mills in three dimensions.
In the brane language, the metric corresponds to non extremal D3 branes in the near
horizon limit, and uT is related to the energy density on the D3 brane.
Let us call the compact direction z. With the conventions introduced earlier, working
Euclideanly, and still neglecting factors of 2 , the metric is
+ Gt t = Gxx = Gyy = u2 /R 2 ,

4 2
uT
u
,
Gzz = 1
u
R2

4 2 1
uT
u
,
Guu = 1
u
R2
and so
(86)
(87)
(88)
(89)
f (u) = u2 /R 2 ,

4 1/2
uT
.
(90)
g(u) = 1
u
The exact classical solution for the Wilson loop in this setup is given for this case by
[17,18]
u2T
L 2 + O(eL )
(91)
2R 2
with some constant and = 2uT /R 2 .
In order to compute the quantum correction to the energy, we should compute the
appropriate operators. From (19) we still have that

4
u2
by = 0 x2 + u t2 ,
(92)
O
2R 2
u40
4
4 2
u2
b = 0 x R x + R u t2 ,
(93)
O
2R 2
u2
u40
E=
but now
bz =
O

4

4
u20
uT
u4
uT
x 1
x + 4 1
t2 .
u
u
2R 2
u0
(94)
A calculation for the longitudinal part using normal fluctuations, similar to that performed
for the conformal case in Section 6, can be carried out, and the result turns out to be similar
to the longitudinal result in that case, only with a different mass term.
98
In the large L limit, as we have seen in Section 11, the string is, for most of its length,
very flat, and has u u0 . Following [19], we assume that we can trust this analysis even
after substituting u u0 in the operators. In that limit we find

u20 2
+ t2 ,
2R 2 x
4
4
u2
b 0 R x2 + R t2 ,
O
2R 2 u20
u20

4

2
u2
2
bz 0 1 uT
+
O
x
t
u0
2R 2
4

2uT
1 2 1 2
b
+ x + t .
Ox 2
2
u0 R 4 2
by
O
As L grows, u0 uT , and the tendency is exponential [17],

u 0 uT
eL .
uT
Inserting this to the operators we find for large L that

u2T 2
x + t2 ,
2
2R
2

b R x2 + t2 ,
O
2

u2T 2uT L 2
bz
e
x + t2 ,
O
2
2R
2

4uT
1 2 1 2
b
+ + .
Ox
2R 4 2 x 2 t
by
O
(95)
(96)
(97)
(98)
(99)
(100)
(101)
(102)
(103)
b , O
bz , turn out to be simply
by , O
We see that the operators for transverse fluctuations, O
the Laplacian in flat spacetime, multiplied by overall factors, which are, as we have
seen, irrelevant. Therefore, the transverse fluctuations yield the standard Lscher term [1]
proportional to 1/L.
bx corresponding to a 1 + 1
The longitudinal normal fluctuations give rise to an operator O
2
dimensional
q scalar field with mass 2uT /R = . Such a field contributes a Yukawa-like
2L
e
to the potential.
term L
We see that the under the assumption that we can approximate the classical string
configuration by a flat one, we get that one of the bosonic degrees of freedom becomes
massive and its 1/L contribution to the potential is reduced to an exponentially small
contribution (similar to, but even weaker than, the exponential part of the classical
correction (91)). This is in contrast to the results of [19], where two bosonic degrees
of freedom become massive. The source of this discrepancy is that in [19], the string is
approximated to lie on the horizon, u uT . Our scheme, in which the approximation is
u u0 , is more accurate.
In the calculations of [19], one must take into account the force exerted on the (almost)
flat string by the non flat segments near x = L/2. The addition of the corresponding
99
potential causes the flat string to be in equilibrium at u = u0 . Therefore (in the notations
of [19]), we get Z 2 + T 2 = u0 uT 6= 0, and there is a spontaneous breaking of the
rotational symmetry. The resulting Goldstone boson is exactly the degree of freedom which
is massless in our calculation but massive in that of [19].
13. Fermionic fluctuations in the pure YangMills setup

The derivative Di in (64) is argued in [24] to originate from the derivative
1
(104)
D = + ab ab + c 1 5 e F1 5
4
which appears also in the Killing spinor equation of type IIB supergravity. Here, is the
dilaton and F1 5 is the RamondRamond field strength. In the AdS5 S 5 case, there
is no dilaton background ( = 0), and F1 5 u3 1 5 . In the actual computation, a
factor of ( det G )1/2 u3 cancels the u3 from F , and the last term of (104) indeed
reduces to the last term of (64), with the Minkowski-space 01234 replaced by I J .
When moving to the pure YangMills metric (87), upon compactification of the
coordinate z, the dilaton and RamondRamond flux remain the same, and det G does
not change either. Assuming the same form (104) of the derivative, we therefore again
arrive at (64).
Based on the results of Section 11, we now assume, as in the bosonic computation of the
previous section, that the classical string is flat. That is, 0 = t, 1 = x 1 , with u = u0 and
all other coordinates zero (or constant). The combined fluctuations of the string around its
classical solution and of the fermions around zero give terms of high order; the quadratic
terms of the fermions are coupled only to the bosonic classical solution.
In this section, we find it more convenient to work in Euclidean coordinates. It is
straightforward to see that the induced vielbein eia obeys e00 = e11 = u0 /R, and all the other
components vanish. The induced metric is therefore proportional to the identity matrix, as
it clearly should, and the Lagrangian reduces to
L + 1 0 D0 1 + 2 0 D0 2 + 1 1 D1 1 + 2 1 D1 2
1 0 D1 1 + 2 0 D1 2 + 1 1 D0 1 2 1 D0 2 ,
(105)
with Di the covariant derivative containing the effects of gravity and of the Ramond
Ramond flux.
The relevant components of the spin-connection induced on the classical worldsheet are,
by (99),
q
u0
2uT
2
04
14
(106)
2 0 = 1 = 2 1 (uT /u0 )4 2 e(uT /R )L .
R
R
By the previous discussion, the term arising from the RamondRamond flux in Di is
c i I J . In order to agree with (64), c = u0 /2R 2 . In conclusion,

(107)
D0 I = 0 + 0 4 I J + c 0 I J J ,

I
1 4 IJ
1 IJ J
(108)
D1 = 1 + + c .
100
Inserting (107), (108) into (105) we get 24 terms which reduce to

L 1 0 0 + 1 1 0 1 + 1 0 + 2 4 2 0 1 4 1

+ 2 0 0 + 1 1 + 0 1 1 0 + 2 4 + 2 0 1 4 2

+ 1 +2c 2c 0 1 2 + 2 2c 2c 0 1 1 .
(109)
Now we should impose a -symmetry gauge. Choosing the gauge 2 = i 4 1 , as

advocated by [27] for the AdS5 S 5 case, we get

(110)
L 1 0 0 + 1 1 + 2i(c i) 0 1 4 1 1 OF 1 .
When we square the fermionic operator, and remove, using integration by parts, the terms
with a single derivative, we finally get
OF2 = 02 + 12 + 4(c i)2 .
(111)
We thus see that all the eight fermionic modes become massive, with mass m = 2(c i).
The mass of the fermions has an imaginary part, but as by (106), is vanishingly small
for large distances L, we see that m 2c = uT /R 2 is half the mass of the only massive
bosonic mode.
Choosing the similar -symmetry gauge 2 = i 3 1 we get a similar result

(112)
L 1 0 0 + 1 1 + 2 4 + 2i c 0 1 3 1 .
Now
OF2 = 02 + 12 + 4(c2 + 2 )
(113)
and we have m2 = 4(c2 + 2 ) 4c2 , so again m 2c in the large L limit.

In conclusion, we see that in the pure YM case there are seven massless bosonic
modes, and no fermionic ones. The contribution of the Lscher term to the quarkantiquark potential is (7 0) (/24) (1/L), which is attractive and concave in agreement
with the general result in quantum field theory [20,21].
14. Summary and discussion

Wilson loops are some of the most important gauge invariant physical quantities associated with non-Abelian gauge theories. They play an essential role in the understanding
of the underlying structure of YM theory. In particular, the rectangular loop has a simple
interpretation, being related to the static potential between a quark and an anti-quark. Wilson loops have thus attracted a tremendous amount of work throughout the years. Among
the various approaches invoked to investigate the Wilson loop, the description in terms
of a string model is the most promising. This approach is based on the fact that stringy
Wilson loops obey the loop equation, and on the phenomenological picture of a flux tube
connecting the quark pair. Recently the duality conjecture of Maldacena [7] has given this
approach new impetus from a new perspective.
It is easy to realize that a classical string in flat spacetime results in an area law
behaviour which is expected for a gauge theory in the confining phase. An interesting issue
101
that had been discussed in the early 80s is the quantum corrections to the linear potential.
The purpose of the present work is to shed additional light on this issue, using the gravity
description of large N gauge theories. En route to this goal we have also addressed the
issues of the quantum correction to the quarkanti-quark potential in N = 4 SYM via
quadratic fluctuations of the corresponding string in the AdS5 S 5 background, as well as
the contribution of these fluctuations to the mass of the BPS quark in that approach.
Determining the quantum corrections involves four steps: (i) Writing down the
corresponding action, (ii) Choosing an appropriate gauge fixing, (iii) Expanding the gauge
fixed action to quadratic order and (iv) Computing, using some regularization scheme, the
functional determinant and the resulting free energy. Whereas the bosonic part of the action
is well known for any background, the incorporation of fermionic degrees of freedom
is more involved, due to the presence of a non trivial RamondRamond background
in the models we discuss. So far no complete NSR action for such systems has been
formulated. A GreenSchwarz type of action was formulated for the AdS5 S 5 model
[24], but not for the more general backgrounds that correspond to confining scenarios. In
attempt to generalize the fermionic action we used the fact that the AdS5 S 5 action is a
covariantization of the flat GS action, and, assuming the same behaviour, wrote down the
fermionic operators for the confining scenario.
It turned out that the issue of choosing a gauge slice is more tricky. Already in fixing
the 2d reparameterization we found that different gauge choices lead to different operators.
In principle, different operators could yield the same regularized determinant. Moreover,
since we were not able to compute explicitly the determinants, we cannot really claim that
they are indeed different. However even the argument that the fixed u gauge, the fixed x
gauge and the normal gauge lead to the same free energy was plagued with possible
singularities. We also argued that the latter gauge is safer than the others. The projection
of the fermionic fields into 8 Grassmannian degrees of freedom (a single WeylMajorana
spinor) through fixing of the -symmetry can also be performed in various ways, and again,
different gauge slices yield different operators. For the BPS single quark configuration of
the AdS5 S 5 , the 4 gauge fixing, introduced in [27], was found to be inapplicable (as
in general, fixing the -symmetry using a i projection cannot be used when the classical
configuration spans along xi ) and of the other projections we tried one lead to a cancelation
of the logarithmic divergences while another did not. Further investigation is required in
order to gain better understanding of the issue of choosing a legitimate and effective symmetry gauge fixing.
As for step (iv), we were not able to compute explicitly the determinants, apart from
the known case of flat spacetime. Regarding the confining setup, we used similar
considerations to those of [19] in treating the bosonic degrees of freedom and found that
(only) one mode becomes massive. We were also able to find some evidence that the
fermionic modes are massive as well, thus leading to a quantum correction to the linear
potential which is attractive and of Lscher type. This provides an evidence in favour of
the gauge/gravity duality, as it was shown in [20], based on general arguments, that the
overall quantum potential should be concave.
102
The ultimate test of the results concerning the confining setups is comparison with
experimental observations. There were several attempts to deduce the corrections to the
linear potential from the spectrum of heavy mesons but it is not clear to us whether one
can reach a coherent picture from this analysis. On the way to comparing with experiment
one should also consider lattice computations of Wilson loops in 3- and 4-dimensional
pure YM theory. In particular, one should address the question whether a Lscher term
correction to the linear potential can be seen in lattice simulations. According to [28] there
are some numerical evidence for a Lscher term associated with a bosonic string. However
the results are not precise enough to be convincing.
It is the quantum fluctuations of the AdS5 S 5 case that leave us with several unresolved
puzzles. In [22] it was found that there is a logarithmic divergence in the determinant and
it was suggested that the infinity should be canceled by renormalizing the mass of the
bare quarks. However, the latter are BPS states and thus should not receive any quantum
corrections. Indeed, we have shown that, at least in a certain -symmetry gauge fixing there
are no logarithmic divergences. On the other hand we could not find any gauge fixing in
which the quantum corrections to the Wilson loop are free of logarithmic divergences.
Hence, the renormalization and gauge fixing procedures for this case deserve further
investigation. As for the question of whether the AdS5 S 5 case admits a Lscher term
and of what sign, we again do not have a conclusive answer. There are no hints from the
explicit calculation that the bosonic and fermionic determinants cancel. As discussed in
the introduction, the coefficient of the Lscher term is in fact related to the intercept, the
normal ordering constant in the Virasoro L0 generator. We do not have any evidence that
this should vanish in the AdS case.
There is, however, a naive argument why the Lscher term should not exist. In the limit
of R one may speculate that locally one sees a flat metric and hence there should
be a vanishing coefficient. Now, since the basic property of the Lscher term is that it is
independent on R, the result for R should hold for any R. As we could not verify
this prediction we may suspect that the argument is too naive and one cannot extrapolate
smoothly to the Wilson loop in flat space time. It is thus clear that the explicit evaluation of
the various determinants in the AdS5 S 5 case is still an open and challenging question.
Acknowledgements
We have greatly benefitted from discussions with O. Aharony, R.L. Jaffe, S. Theisen,
A.A. Tseytlin and S. Yankielowicz.
Note added
After the completion of this paper, the manuscript [34], which has some overlaps with
it, also appeared in the electronic archives.
103
References
[1] M. Lscher, K. Symanzik, P. Weisz, Anomalies of the free loop wave equation in the WKB
approximation, Nucl. Phys. B 173 (1980) 365.
[2] O. Alvarez, Phys. Rev. D 24 (1981) 440.
[3] S.D. Odintsov, D.L. Wiltshire, The static potential for bosonic p-branes, Class. Quant. Grav. 7
(1990) 14991510.
[4] A.A. Bytsenko, S.D. Odintsov, The Casimir energy for p-branes compactified on constant
curvature space, Class. Quant. Grav. 9 (1992) 391403.
[5] E.S. Fradkin, A.A. Tseytlin, Ann. Phys. 143 (1982) 413447.
[6] J.F. Arvis, Phys. Lett. B 127 (1983) 106.
[7] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231; hep-th/9711200.
[8] J. Maldacena, Wilson loops in large N field theories, Phys. Rev. Lett. 80 (1998) 48594862;
hep-th/9803002.
[9] S.J. Rey, J.T. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de
Sitter supergravity, hep-th/9803001.
[10] E. Witten, Anti-de Sitter space, thermal phase transition, and confinement in gauge theories,
hep-th/9803131.
[11] S.J. Rey, S. Theisen, J.T. Yee, WilsonPolyakov loop at finite temperature in large N gauge
theory and anti-de Sitter supergravity, hep-th/9803135.
[12] A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, Wilson loops in the large N limit
at finite temperature, hep-th/9803137.
[13] A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, Wilson loops, confinement, and
phase transitions in large N gauge theories from supergravity, hep-th/9803263.
[14] D.J. Gross, H. Ooguri, A spects of large N gauge theory dynamics as seen by string theory,
hep-th/9805129.
[15] Y. Kinar, E. Schreiber, J. Sonnenschein, Precision measurements of the QQ potential in
MQCD, hep-th/9809133.
[16] A. Polyakov, The wall of the cave, hep-th/9809057.
[17] Y. Kinar, E. Schreiber, J. Sonnenschein, QQ potential from strings in curved spacetime
classical results, hep-th/9811192.
[18] J. Greensite, P. Olesen, Remarks on the heavy quark potential in the supergravity approach,
hep-th/9806235.
[19] J. Greensite, P. Olesen, Worldsheet fluctuations and the heavy quark potential in the AdS/CFT
approach, hep-th/9901057.
[20] C. Bachas, Phys. Rev. D 33 (1986) 2723.
[21] H. Dorn, V.D. Pershin, Concavity of the QQ potential in N = 4 super YangMills gauge theory
and AdS/CFT duality, hep-th/9906073.
[22] S. Forste, D. Ghoshal, S. Theisen, Stringy corrections to the Wilson loop in N = 4 super Yang
Mills theory, hep-th/9903042.
[23] S. Naik, Improved heavy quark potential at finite temperature from anti-de Sitter supergravity,
hep-th/9904147.
[24] R.R. Metsaev, A.A. Tseytlin, Type IIB superstring action in AdS5 S 5 background, hepth/9805028.
[25] I. Pesando, A gauge fixed type IIB superstring action on AdS5 S 5 , hep-th/9808020.
[26] R. Kalosh, J. Rahmfeld, The GS string action on AdS5 S 5 , hep-th/9808038.
[27] R. Kallosh, A.A. Tseytlin, Simplifying superstring action on AdS5 S 5 , hep-th/9808088.
[28] M. Teper, Glueball masses and other physical properties of SU(N) gauge theories in D = 3 + 1,
hep-th/9812187.
104
[29] N. Itzhaki, J. Maldacena, J. Sonnenschein, S. Yankielowicz, Supergravity and the large N limit
of theories with sixteen supercharges, hep-th/9802042.
[30] N. Drukker, D. Gross, H. Ooguri, Wilson loops and minimal surfaces, hep-th/9904191.
[31] N. Graham, R.L. Jaffe, Phys. Lett. B 435 (1998) 141145; hep-th/9805150; Nucl. Phys. B 544
(1999) 432447; hep-th/9808140.
[32] M. Shifman, A. Vainshtein, M. Voloshin, Phys. Rev. D 59 (1999) 045016, hep-th/9810068.
[33] H.M. Stephanov, P. van Nieuwenhuizen, A. Rebhan, Nucl. Phys. B 542 (1999) 471514; hepth/9802074.
[34] N. Drukker, D. Gross, A.A. Tseytlin, GreenSchwarz string in AdS5 S 5 : semiclassical
partition function, hep-th/0001204.

Level truncation and the tachyon in open bosonic

string field theory
Nicolas Moeller 1 , Washington Taylor 2
Center for Theoretical Physics, MIT, Bldg. 6-306, Cambridge, MA 02139, USA
Received 30 March 2000; accepted 16 May 2000
Abstract
The tachyonic instability of the open bosonic string is analyzed using the level truncation approach
to string field theory. We have calculated all terms in the cubic action of the string field theory
describing zero-momentum interactions of up to level 20 between scalars of level 10 or less. These
results are used to study the tachyon effective potential and the nonperturbative stable vacuum. We
find that the energy gap between the unstable and stable vacua converges much more quickly than
the coefficients of the effective tachyon potential. By including fields up to level 10, 99.91% of the
energy from the bosonic D-brane tension is cancelled in the nonperturbative stable vacuum. It appears
that the perturbative expansion of the effective tachyon potential around the unstable vacuum has a
small but finite radius of convergence. We find evidence for a critical point in the tachyon effective
potential at a small negative value of the tachyon field corresponding to this radius of convergence.
We study the branch structure of the effective potential in the vicinity of this point and speculate
that the tachyon effective potential is globally non-negative. 2000 Elsevier Science B.V. All rights
reserved.
1. Introduction
The appearance of a tachyon in the spectrum of both open and closed bosonic strings
has appeared to be a fundamental obstacle to a physical interpretation of these theories
since the early days of the dual resonance model. Early work on the subject indicated the
possible existence of a more stable nonperturbative vacuum which can be reached by a
condensation of the tachyon and other string fields in the open bosonic string theory [1,2].
At that time, however, the connection between open strings and Dirichlet branes [3] had
not yet been realized, so that the significance of the nonperturbative stable vacuum was not
widely appreciated.
1 moeller@pierre.mit.edu
2 wati@mit.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 9 3 - 5
106
N. Moeller, W. Taylor / Nuclear Physics B 583 (2000) 105144
It was recently pointed out by Sen [4] that the condensation of the tachyon in open
bosonic string theory should correspond to the process of annihilation of an unstable
D25-brane. The nonperturbative stable vacuum of the open string should simply be the
vacuum corresponding to empty space, and the energy difference between the stable and
unstable vacua should therefore be given by the mass of the D25-brane. Sen suggested
that it should be possible to precisely calculate this energy gap using open string field
theory. In fact, it was found over a decade ago by Kostelecky and Samuel that truncating
open bosonic string field theory at low mass levels gives a systematic approximation
scheme which seems to converge to a finite value for the energy difference between the
unstable and stable vacua [2,5]. Sen and Zwiebach have carried out a calculation of this
type and have shown that including fields up to mass level 4 and interactions up to mass
level 8 gives a mass gap of 98.6% of the D25-brane energy [13]. Similar calculations of
tachyon condensation have recently been performed in open superstring field theory [6,7].
The level-truncation approach to string field theory has also been used to study lowerdimensional Dp-branes as solitonic lumps [7,8]. In all these calculations, truncation of
string field theory to the first few levels seems to give a sequence of successively better
approximations to nonperturbative physical quantities.
In this paper we extend the level truncation approach to open bosonic string field theory
to include scalar fields of mass levels > 4, following earlier work in [9]. We use the leveltruncated theory as a tool for exploring various features of the tachyon and its condensation
into a stable vacuum. We compute all terms in the string field theory action involving fields
of level less than or equal to 10 and interactions of up to mass level 20. We perform a
nonperturbative calculation of the energy gap between the stable and unstable vacua and
find that 99.91% of the D25-brane tension is produced in the level (10, 20) truncated theory.
We study the structure of the tachyon effective potential in some detail using the
various level-truncated theories. We compute the first 60 coefficients cn of n in the
effective tachyon potential in several successive level truncations up to level (10, 20).
We find that the effective tachyon potential has a radius of convergence which decreases
to an apparently finite asymptotic value as the level number increases. This radius of
convergence is substantially smaller than the tachyon value corresponding to the stable
vacuum, although the associated singularity is at a negative value of while the stable
vacuum arises at > 0. We investigate the branch structure of the tachyon effective
potential near the singular point, and use our numerical results to speculate about the
behavior of the effective potential beyond this point.
In Section 2 we review the level truncation approach to string field theory and outline our
calculation of the action up to level (10, 20). In Section 3 we use our results to analyze the
perturbative expansion of the tachyon effective potential, and in Section 4 we discuss the
nonperturbative stable vacuum and the branch structure of the effective tachyon potential.
107
2. Level truncation of string field theory

In this section we briefly review the level truncation approach to string field theory
and describe our calculation of interactions between scalar string fields of level 6 10.
Throughout this paper we follow the conventions and notation of [2,9]. For more detailed
reviews of open string field theory see [10,11].
2.1. The scalar potential in open bosonic string field theory
String field theory is described in terms of a string field which contains a component
field for every state in the first-quantized string Fock space. For the open bosonic string,
a particularly simple string field theory was suggested by Witten [12] in which the action
takes the cubic form
Z
Z
g
1
? ? ,
(1)
S = 0 ? Q +
2
3!
where Q is the BRST operator and ? is the string field theory star product.
In FeynmanSiegel gauge, the string field can be expanded in the form

1

+ b1 c1 + |0i,
= + A 1 + B 1 1
2
(2)
where |0i = c1 |i is the state in the string Hilbert space associated with the tachyon
field. The expansion (2) contains an infinite series of fields. The level of each field in the
expansion is defined to be the sum of the level numbers of the creation operators which act
on |0i to produce the associated state. Thus, the tachyon is the unique field of level 0, the
gauge field A is the unique field of level 1, etc. The states associated with these fields are
all in the subspace H of the full string Hilbert space containing states of ghost number 1.
In this paper we are interested in the behavior of the tachyon field . In particular, we
wish to study questions related to the appearance of a Lorentz-invariant stable vacuum in
the theory when the tachyon and other scalar fields acquire nonzero condensates. Because
all the questions we will address here involve Lorentz-invariant phenomena, we can restrict
attention to scalar fields in the string field expansion. We write the string field expansion
in terms of scalar fields as
=
i |si i,
i=1
where |si i are all the scalar states in H. The scalar states at levels 0 and 2 are
|s1 i = |0i,
|s2 i = 1 1 |0i,
|s3 i = b1 c1 |0i.
The associated scalar fields can be related to the fields in the expansion (2) through
= 1,
(3)
108
B =
2 2 ,
= 3.
We will often refer to the tachyon field 1 simply as . (Note that a superscript on
will always indicate a field index while a superscript on indicates an exponent.) For the
calculations of interest here, contributions from scalar fields of odd level cancel due to a
twist symmetry [2,13]. Thus, we need only consider scalar fields of even level number. All
scalar states at levels 0, 2, 4 and 6 are listed in Appendix A.
An explicit algorithm for computing the terms in the string field theory action (1) using
the oscillator modes of the matter fields and ghost fields of the bosonic conformal field
theory was given in [1416]. The aspects of this formalism needed for the computations in
this paper are reviewed in [9]. The quadratic term evaluated on a state |si is simply
Z

(4)
|si ? Q|si = hs|c0 0 p2 + 12 M 2 |si,
where 12 M 2 is the level of s minus 1, p is the momentum of the state s, hs| is the BPZ dual
state to |si produced by acting with the conformal transformation z 1/z, and the dual
vacuum satisfies h0|c0 |0i = 1. The cubic interaction terms in the string field action can be
described in terms of Wittens vertex operator V through
Z
? ? = hV |( ),
(5)
where V is a state in the tensor product space H H H , given in terms of oscillator
modes by

hV | = (p(1) + p(2) + p(3) ) h0|c0(1) h0|c0(2) h0|c0(3)

rs
(s)
rs (s)
m
+ cn(r)Xnm
bm .
(6)
exp 12 n(r) Nnm
rs , X rs can be calculated from formulae in [1416]; explicit tables of
The coefficients Nnm
nm
these coefficients for n, m 6 8 are given in [9].
Using Eqs. (4), (5) for the quadratic and cubic terms in the string field theory action, the
potential for the zero momentum scalar fields in H can be written as
X
X
dij i j + g
tij k i j k ,
(7)
V=
i,j
i,j,k
where g is the string coupling constant and

=
37/2
27
has been chosen so that t111 = 1. Throughout the paper we set 0 = 1. The coefficients
dij , tij k can be explicitly computed for any given values of the indices. The action can be
truncated by only including fields up to a fixed level L and interactions including terms
whose total level does not exceed another fixed value I . In [2], the complete action with
(L, I ) = (2, 6) was calculated and shown to be
109
1
1
V = 2 + 26(2 )2 (3 )2
2
2

11
5 26
3
2
2 3
+ g +
9
9

4
5
11 13
19
4 7 13 83
2
2
(
)
+
+
(
)
+
2
2 3
3
35
35
34
22 7 13 11 83
23 3 7 13 41 73
3
(
)
(2 )2 3
2
8
39
3
2 5 13 19
1
2 (3 )2 4 (3 )3 .
7
3
3
(Note that we have lowered indices on the fields 2 , 3 for clarity.) Results using the
complete action with (L, I ) = (4, 12) were described in [5]. In [13] calculations were
performed at level (L, I ) = (4, 8).
With the help of the symbolic manipulation program M ATHEMATICA, we have
calculated all terms in the action up to levels (L, I ) = (10, 20). There are 252 fields at
levels 6 10 and 138202 distinct cubic interaction terms between fields whose total level is
6 20, so it is clearly impractical to reproduce the full action here. It is worth mentioning,
however, some points which help to simplify the calculation.
One significant simplification arises from the fact that the matter and ghost fields almost
completely decouple in the action. In fact, it is clear from (6) that this decoupling is
complete in the cubic terms, and from (4) that the only coupling in the quadratic terms
arises from the appearance of the total level of the scalar field in question, including the
levels of both the matter and ghost oscillators needed to produce the state. Because of this
decoupling, we find that it is convenient to decompose the Hilbert space into the matter
and ghost Hilbert spaces
H = Hmat Hgh
(8)
and to separately enumerate the scalar fields in the matter and ghost sectors |m i
Hm , |g i Hg . The first few states in each of these component Hilbert spaces are
|1 i = |1 i = |0i,
|2 i = 1 1 |0i,
|2 i = b1 c1 |0i.
Given the decomposition (8) of the Hilbert space we can write each of the scalar fields |si i
in H as a tensor product
|si i = |m(i) i |g(i) i.
(9)
Thus, for example, m(1) = 1 and g(1) = 1. A list of matter and ghost scalars up to level 6
is given in Appendix B, and the decomposition of the scalars in H into matter and ghost
factors is given in Appendix A for the scalars of level 6 6.
From the decomposition into matter and ghost components, we can now write the
quadratic and cubic coefficients (4), (5) as
110
dij =
2(level(i) 1)
gh
d mat
d
,
(level(m(i)) 1) (level(g(i)) 1) m(i)m(j ) g(i)g(j )
(10)
gh
mat
tij k = tm(i)m(j
)m(k) tg(i)g(j )g(k),
gh
gh
mat , d , t mat and t

where dmn
gh mnp
ghj are the quadratic and cubic coefficients in the matter and
ghost sectors, and where level(i) = level(g(i))+ level(m(i)) indicates the level of the
state |si i. Quadratic and cubic coefficients of the matter and ghost scalars appearing in
scalar fields up to level 6 are tabulated in Appendices B and C. From these coefficients and
(10) one can reproduce the entire string field theory action for zero momentum scalars up
to level 6 fields and level 16 interactions. We have carried out the calculation of all matter
and ghost interactions up to fields of level 10 and interactions of level 20. While we do not
reproduce here the complete results of this calculation, we will use these results to study
questions of physical interest in the remainder of the paper.
We end this section with a brief discussion of the approach used in [13], in which a
reduced Hilbert space of background-independent fields was used. It was pointed out in
[4] that it is possible to truncate the Hilbert space H in a consistent fashion by considering
the subspace H1 obtained by considering only those states produced by acting with the
ghost fields b, c and the Virasoro generators Ln , n > 2, associated with the stress tensor
of the matter fields. In the stable vacuum of the theory, only scalar fields in H1 acquire
nonzero expectation values. At level 4 and above, the number of scalars in H1 is less than
the number of scalars in H, so that particularly when the level number becomes very large
the action for the level-truncated theory is significantly simplified by restricting attention
to the truncated Hilbert space. To compare the number of scalars in H at a fixed level n,
which we denote hn , with the number of scalars in H1 , which we denote h1n , we can write
generating functions
f (x, y) =
f1 (x, y) =
Y
p=2
Y
p=2
X
Y
1
(1 + x q y)(1 + x q /y) =
hn,m x n y m ,
p
bp/2c
(1 x )
n,m
(11)
X
Y
1
(1 + x q y)(1 + x q /y) =
h1n,m x n y m
p
(1 x )
n,m
(12)
q=1
q=1
in terms of which
hn = hn,0 ,
h1n = h1n,0 .
The number of scalars in H and H1 at each even level up to n = 20 is tabulated in Table 1.
Up to the level 10 fields we consider in this paper, the difference between H and H1 is
less than a factor of 2, so that there is no extraordinary advantage to be gained by using
the smaller Hilbert space. Clearly, however, as the level becomes much larger than 10,
the reduced size of the truncated Hilbert space would make explicit calculations much
easier, assuming that calculations in H1 could be done just as efficiently as corresponding
calculations in H.
111
Table 1
The number of scalars hn and h1n in H and H1 at level n
n
10
12
14
16
18
20
hn
h1n
1
1
2
2
7
6
21
17
60
43
161
102
415
231
1021
496
2432
1027
5620
2060
12639
4010
In our calculations we have worked with the complete oscillator Hilbert space H and
not with H1 . The main reason for this is that at this point no method has been developed
for computing cubic interactions between fields in H1 which is as systematic and efficient
as the oscillator method described above for computing cubic interactions between fields
in H. It is possible to calculate cubic interactions in H1 by computing the relevant
correlation functions in the open bosonic string theory [10], but this method is somewhat
more complicated than the straightforward oscillator approach. Of course, the interactions
in H1 can always be computed by rewriting each state in terms of the oscillator basis
and then using (6), but this requires just as much work as the computation directly in the
oscillator basis. In any case, the results in Appendix C for fields up to level 6 are given in
the oscillator basis of H; these interactions can easily be translated into an action for fields
in H1 by explicitly writing the Virasoro generators Ln in terms of the oscillator basis.
3. Tachyon effective potential: perturbation expansion
Given the (rational) coefficients in the scalar potential (7) truncated at level (L, I ), we
can perform a number of interesting calculations. In particular, we can study the effective
potential of the tachyon field and identify the stable vacuum or vacua of the theory.
An effective potential for the tachyon field can be determined by starting with the
complete set of terms in the cubic potential (7) truncated at level (L, I ), fixing a value of
= 1 , solving for all fields i , i > 1, and plugging back into the potential to rewrite it
as a function of . If there are N scalar fields involved, this means that we need to solve a
system of N 1 simultaneous quadratic equations. In principle there are many solutions
of this system of equations, but if we are interested in the branch of the solution on which
all fields vanish when = 0, it is easy to determine which branch to choose for each of the
quadratic equations (an explicit example of this choice is described below). Clearly when
N becomes large it is impractical to find an exact analytic form for the effective tachyon
potential. Various numerical methods can be used to approximate the effective potential
arising from integrating out a large number of fields; for example, a numerical analysis
of the effective potential was performed in [2] using the level 2 action with the two fields
2 , 3 integrated out. In Section 4.2 we extend these results by analyzing the structure of
the effective potential at levels 4 and 6 using numerical methods.
Another approach to studying the effective potential, which we consider in this section,
is to determine the terms in a power series expansion of the potential around the unstable
vacuum = 0. In Section 3.1 we describe an algebraic method for efficiently summing all
112
diagrams contributing to the n term in the effective potential. In Section 3.2 we summarize
the results of our calculation of these coefficients up to n = 60 at level L 6 10, and discuss
the implications of these results. In particular, we find that the power series expansion of
the tachyon effective potential seems to have a radius of convergence which approaches a
finite but nonzero value as the level number is increased. As we will discuss in the next
section, the stable vacuum of the theory lies outside the radius of convergence of this power
series.
3.1. Summing planar diagrams
In the vicinity of the unstable vacuum i = 0 we can perform a power series expansion
of the effective tachyon potential
V () =
X
n=2
1
cn (g)n2 n = 2 + g 3 + .
2
(13)
There are a number of ways in which we might try to calculate the coefficients cn in
this expansion. Given a cubic potential (7) for N fields, one approach to determining the
coefficients cn would be to write the quadratic equations of motion for each of the N 1
fields which we wish to integrate out, expand each field i , i > 1, as a formal power series
in , and solve a linear system of equations at each order in , plugging the final results
back into the cubic potential. We will not directly use this method, although it is technically
equivalent to the graphical approach we will use.
An alternative approach to calculating the coefficients in the effective tachyon potential
is to treat (7) as the action for a 0-dimensional field theory and to use Feynman diagrams to
sum all terms contributing to a given coefficient cn . This method was used by Kostelecky
and Samuel in [2] to calculate the level 2 and level 4 approximations to the quartic
coefficient c4 , which they had calculated exactly in [17]. They found that at level 2, 72% of
the exact coefficient is generated, and at level 4 this increases to 84%. This calculation was
extended in [9], where contributions from all fields up to level 20 were included, generating
9697% of the exact term c4 .
We can use the exact cubic terms we have found up to level (10, 20) as cubic interaction
vertices in Feynman diagrams which we can then sum to find the contributions to
each of the coefficients cn . Because the combinatorics of the Feynman diagrams grows
exponentially, we will find it useful to use an algebraic simplification to expedite the
summation over graphs. The essence of the simplification we will use is that because we
are summing over planar tree graphs, we can write a recursion relation relating the set of
graphs Gk with k external edges and single external i edge to the sets of graphs Gk 0
with k 0 < k. This relationship is indicated schematically by
113
i

@
AA

(n s)
s l
kH

H
H

P
m, k, l

@
@
AA
AA

(m s)
(n m s)
Algebraically, we can define an N -dimensional vector vni for each n, representing the
summation over all graphs with n external edges and a single external i (including
the propagator for ). We then have
v1i = 1i ,
vni =
n1
3 X ij
k l
d tj kl vm
vnm ,
2
(14)
m=1
where d ij is the inverse matrix to dij and

0, i = 1 and n > 1,
i
vn = i
vn , otherwise
(15)
has been defined to project out internal edges. In terms of the vectors vni , the coefficients
cn are given by
1 1
.
cn = vn1
n
(16)
In fact, the recursion relations (14) precisely encode the relations we would find between
the coefficients of the fields i expanded in powers of if we used the power series
approach to solving the N 1 quadratic equations term by term in as described in the
first paragraph of this subsection. The terms vni are precisely the coefficients of n in the
expansion of the field i .
In any case, using this recursive formalism, the computation of cn becomes a
polynomial-time algorithm taking time of order O(N 3 n2 ), instead of an exponentially
hard algorithm such as we would encounter if we tried to directly sum over all Feynman
diagrams. It may be helpful to illustrate this approach with a simple example. Consider
a single massive field which couples to only through a term of the form 2 in the
potential

1 2
3
1 2
3
2
+ .
(17)
V = + g + + g
2
2
6
In this case we can explicitly solve for
q

1
1 + 1 2g 2 2 2 .
=
g
(18)
114
As mentioned above, we have chosen the branch of the square root which gives = 0
when = 0. Substituting (18) into (17) gives us the effective potential for on the branch
containing the unstable vacuum

1
1
V = 2 + g 3 + 2 2 2 1 3g 2 2 2 (1 2g 2 2 2 )3/2 .
2
3g
(19)
The coefficients in a power series expansion of this potential are

c2n =
(2n 5)!! n n2

n!
(20)
for 2n > 4.
To derive these coefficients using the recursive formalism (14) we need the coefficients
1
t112 = ,
d11 = ,
2
3
1
t222 = ,
d22 = ,
2
6
which give the recursion relations
v11 = 1,
vn1 = 2
n1
X
1 2
2
vm
vnm = 2vn1
,
n > 1,
m=1
v12 = 0,
v22 = v11 v11 = ,
vn2 =
n1
X 2 2
vm vnm ,
2
n > 2.
m=1
We begin by solving for vn2 . These terms clearly vanish unless n is even. Defining
2
wm = v2m
(21)
we have
w1 = ,
wm =
(22)
m1
X
wk wmk .
k=1
Defining a generating function

f (x) =
wm x 2m .
(23)
m=1
Eqs. (22) are equivalent to the quadratic equation
f = f 2 + x 2
2
(24)
115
with solution
q
X
(2m 3)!! m m1 2m
1

x .
1 1 2 x 2 =
f=
m!
(25)
m=1
So
1
(2n 5)!! n n2
(2)wn1 =

2n
n!
in agreement with (20).
c2n =
(26)
3.2. Coefficients of effective tachyon potential

We have used the method described in the previous subsection to calculate the
coefficients cn in the perturbative expansion of the effective tachyon potential around the
unstable vacuum for n 6 60 in each of the level-truncated theories up to (10, 20). In Table 2
we have tabulated the successive approximations to cn for a representative set of values of
n at various levels. There are several observations we can make based on these results. In
[9] successive approximations to the coefficient c4 were computed using fields at up to
level 20. It was found that while the level truncation method gives a monotonic sequence
of approximations which seem to converge to the known exact value for this coefficient
c4 1.75, the convergence to the asymptotic value is fairly slow. For c4 , the contribution
from the set of graphs including at least one field of level k but no fields of level k + 1
decreases as k increases. The same is true for successive approximations to cn for small n,
but as n increases we find that at n = 8, the contribution from graphs with some fields of
level 4 exceeds that of graphs with only fields of level 2. At n = 15, the contribution from
graphs with some fields of level 6 exceeds that of graphs with fields of level at most 4. The
corresponding thresholds where fields of levels 8 and 10 dominate the contributions from
lower levels are n = 26 and n = 43, respectively. The fact that higher level fields become
Table 2
Level truncation approximations to coefficients cn in effective tachyon potential
n
(2, 6)
(4, 12)
(6, 18)
(8, 20)
(10, 20)
3
4
5
6
7
8
9
10
20
30
40
50
1
1.2592
5.9478
27.909
143.67
786.81
4513.9
26845.
4.0710 1012
1.3415 1021
6.0173 1029
3.1889 1038
1
1.4724
7.8370
43.529
269.25
1777.5
12323.
88690.
9.2769 1013
2.1308 1023
6.6758 1032
2.4732 1042
1
1.5562
8.6398
50.768
333.83
2346.7
17345.
133179.
2.6957 1014
1.2050 1024
7.3563 1033
5.3126 1043
1
1.6004
9.0756
54.816
371.36
2691.9
20529.
162693.
4.5477 1014
2.8154 1024
2.3817 1034
2.3840 1044
1
1.6276
9.3479
57.353
395.12
2913.4
22606.
182299.
6.0780 1014
4.4920 1024
4.5372 1034
5.4231 1044
116
Fig. 1. (ln |cn |)/n for coefficients cn in effective tachyon potential in different level truncations.
more relevant for higher order terms in the effective potential is a natural consequence of
the tradeoff between the exponential growth in the number of diagrams contributing to cn
and the suppression of higher order diagrams by (g)n2 . From this behavior, however, we
see that for large n it will be necessary to include fields of increasingly high level in order
to have a good approximation to the coefficients cn .
One interesting question which we can explore using our results for the coefficients cn
is the radius of convergence of the power series expansion of the effective potential after
truncating at a fixed level. Because of the square root branch cuts which arise from the
quadratic equations of motion for the fields, at some finite value of the effective potential
becomes singular for any given level truncation. For example, the effective potential arising
from (17) has a radius of convergence rc = 1/(g 2 ), which is manifest in (19) and
which can be seen from the asymptotic form of the coefficients c2n n ( 2 )2n where
, are numerical constants. It was found in [2] that after truncation at level 2, there
is a singularity at = rc 0.35/g. This radius of convergence can be seen from the
coefficients of the effective potential in the level 2 truncated theory, which for large n go as
1
.
(27)
|cn | n
(grc )n
A graph of (ln |cn |)/n as a function of n is shown for each level truncation in Fig. 1.
The asymptotic value of (ln |cn |)/n gives the (logarithm of the inverse of the) radius
of convergence of the effective potential at each level. It is clear from the graph that this
radius of convergence is decreasing and seems to be approaching a nonzero limiting value.
By matching the cn trajectories to a 3-parameter family of functions of n of the form (27)
we can find a close approximation to the radius of convergence at each of the levels we
have computed. Table 3 shows the approximate radius of convergence at each level; we see
that in the limiting theory the radius of convergence approaches something like
rc 0.25/g.
(28)
117
Table 3
Approximate radius of convergence of effective potential in different level truncations
Level
grc
(2, 6)
0.345
(4, 12)
0.283
(6, 18)
0.265
(8, 20)
0.256
(10, 20)
0.252
Because the signs of the coefficients cn are alternating, this corresponds to a singularity in
the effective potential at g 0.25. We study the behavior of the effective potential in
the vicinity of this singularity in more detail in Section 4.2.
From this analysis of the perturbative effective potential using string field theory we
have found several things. First, we find that we can in principle determine any coefficient
in the perturbative expansion of the effective tachyon potential to an arbitrary degree of
accuracy with a finite calculation in string field theory, but that the size of the calculation
needed to determine the coefficients cn increases significantly as n increases. Second, we
have found that the resulting effective potential has a radius of convergence on the order of
(28). As we will discuss in the next section, the stable vacuum of the theory lies well
outside this radius, near 1.1/g. It is tempting to conclude from this that, even in
principle, the existence and structure of the stable vacuum of the theory are fundamentally
nonperturbative phenomena.
4. The true vacuum and other nonperturbative features

The appearance of a stable vacuum in the open bosonic string field theory was first
shown by Kostelecky and Samuel in [2]. They found the vacuum in the (2, 6) leveltruncated theory. In later work, Kostelecky and Potting extended this analysis to the (4, 12)
level-truncated theory and reported that the energy density and the values of the scalar
fields in the new vacuum both seemed to be rapidly converging as the level number was
increased. In [4], Sen argued that the energy difference between the false and true vacua
should precisely correspond to the tension of the bosonic D25-brane. Strong evidence for
this conclusion was given by Sen and Zwiebach in [13], where they showed that at level
(4, 8) the energy gap between the stable and unstable vacua was 98.64% of the D-brane
energy.
In this section we use our results for the level-truncated string field action to compute
some nonperturbative features of the open bosonic string. In Section (4.1) we determine the
values of the scalar fields and the total energy of the system in the stable vacuum including
fields of up to level 10. At level (10, 20) we find that the energy gap between the stable and
unstable vacua corresponds to 99.91% of the D25-brane energy. In Section 4.2 we study
the effective potential of the tachyon, focusing on the branch structure of the effective
potential in the region < 0.
118
4.1. The stable vacuum

In order to find the stable vacuum of the level-truncated string field theory action in the
zero momentum sector it is necessary to solve a system of N coupled quadratic equations,
where N is the number of scalar string fields involved. In general, as N becomes large it is
difficult to rapidly solve such a system of equations to a high degree of precision. Because
we know which branch of each quadratic equation the physical solution lies on, however,
we can use an iterative approximation algorithm to rapidly converge on the true vacuum.
As discussed in Section 3, the branch of the solution for each of the N quadratic
equations arising from the level-truncated string field theory action is determined by the
condition that the stable vacuum lies on the same branch as the unstable vacuum which has
all fields vanishing. Thus, the choice of branch for each field is dictated by the sign of the
quadratic term in the string field action. For most fields i , therefore, the value of the field
can be expressed in terms of the remaining fields j , j 6= i, through
b + sign(dii ) b2 4ac
i
,
(29)
=
2a
where
a = 3tiii ,
X
6tiij j + 2dii ,
b=
j 6=i
c=
3tij k j k +
j,k6=i
(30)
2dij j .
j 6=i
There are some exceptions to this general rule, however. For the tachyon = we choose
the + branch of the square root even though the kinetic term is negative. In addition,
for most fields with ghost excitations the quadratic term in the action is off-diagonal. For
example, the level 4 fields 8 , 10 are coupled through the quadratic term d8 10 = 3/2.
For fields such as this we use the equation of motion for field i to solve for the dual field
i which has ghost and antighost modes exchanged through cn bn .
We have used (29), (30) to solve iteratively for the stable vacuum. We begin by solving
(29) for all fields i , using zero for all fields appearing on the RHS. We then insert the firstround solutions for the fields back on the RHS to solve again for all the i , and repeat this
process many times. This algorithm is numerically very stable near the nonperturbative
vacuum and converges quite rapidly to a simultaneous solution of all N equations. At
level 6, for example, including all interactions up to level 18, after 30 rounds of this
procedure the energy stabilizes to 10 digits, and after 50 rounds the values of all fields
stabilize to 10 digits.
In the units we are using here, the tension of the bosonic D25-brane is [4]
T25 =
1
.
2 2 g 2
(31)
In Table 4 we have tabulated the results of our calculation of the exact vacuum in the theory
truncated at various levels. The value of the tachyon is given in units of 1/g (with 0 = 1),
119
and the vacuum energy difference from the unstable vacuum is given as a proportion of the
D25-brane tension. The results at levels (2, 4) and (2, 6) agree with those of [2], and the
results at level (4, 8) agree with [13].
We see from Table 4 that at level (10, 20) the energy gap between the unstable and stable
vacua is 99.91% of the D25-brane tension. This seems to affirm the prediction of Sen in [4]
beyond any reasonable doubt. It is interesting to note that the error in the vacuum energy
(1 + V /T25) is multiplied by approximately 1/3 as each new level is added. It would
be nice to have a theoretical explanation for this rate of convergence.
In [2] it was found that the vacuum energy and tachyon expectation values change very
little between the (2, 4) and (2, 6) level truncations. In [13] a calculation was performed at
level (4, 8) giving 98.6% of the D-brane tension. The improvement on this result given by
including level 10 and 12 interactions between level 4 fields is fairly small. We have found
that this pattern persists at higher level. The results at level (6, 18) are not very different
from those found at level (6, 12), and the results at level (8, 20) are very close to those
at level (8, 16). If, however, we drop cubic interactions at the same level as the quadratic
interactions at that level, for example by truncating all cubic interactions to level (6, 6), the
energy of the stable vacuum varies much more wildly and can even decrease below T25 .
For this reason, we do not think that it would be useful to consider including higher level
fields unless the level of cubic interactions calculated could be increased to at least twice
the level number. Using our rather inefficient program it would take a significant amount of
computer time to go to level (12, 24) on a standard desktop machine. We discuss in the last
section the prospects for continuing these numerical studies to higher levels of truncation
Table 4
Tachyon VEV and vacuum energy in stable vacua of level-truncated theory
Level
ghi
V /T25
(0, 0)
0.91236
0.68462
(2, 4)
1.08318
0.94855
(2, 6)
1.08841
0.95938
(4, 8)
1.09633
0.98640
(4, 12)
1.09680
0.98782
(6, 12)
1.09602
0.99514
(6, 18)
1.09586
0.99518
(8, 16)
1.09424
0.99777
(8, 20)
1.09412
0.99793
(10, 20)
1.09259
0.99912
120
of the full theory.

As we can see from Table 4, the expectation value of the tachyon in the stable vacuum
converges fairly quickly as the level number is increased. The same is true of all the scalar
fields; in Appendix A the values of the scalar fields at levels 6 6 are given for the level
(4, 12), (6, 18) and (10, 20) truncations. The tachyon field itself converges to a value near
ghi 1.09.
(32)
In [13] Sen and Zwiebach observed that the level 2 fields with canonical normalization
v = 2 13 2
u = 3 ,
take almost identical expectation values in the level (4, 8) truncation. This led these authors
to conjecture that in the exact theory these fields would take identical expectation values,
possibly due to some hidden symmetry in the theory. Checking this relationship in the
vacua of the higher level truncations, we find that this conjecture is not supported. While
at level (4, 8) these fields differ by about 0.1%, at level (10, 20) the difference has grown
to over 5%. The values of these fields in the different level-truncated vacua are graphed
in Fig. 2, using the value g = 2 to conform with the conventions of [13]. The failure of
this equality to hold in the asymptotic theory indicates that it may be more difficult than
previously thought to implement the suggestion made in [13] of finding an exact solution
for the nonperturbative vacuum in the full theory using a hidden symmetry relating fields
of this type.
As a check on our results, we can verify that the vacuum expectation values of the
fields i are such that the nonperturbative vacuum lies in the truncated Hilbert space H1
described in [4]. For example, the level 4 fields should obey the linear relation
h 4 i 2h 5 i 4h 6 i = 0
Fig. 2. Expectation values of fields u = 3 and v =
(33)
52 2 for g = 2 in different level truncations.
121
in the vacuum. We have checked this relation and other analogous relations for higher level
fields and find that they are satisfied up to the level of accuracy (10 significant digits) to
which we have calculated the vacuum expectation values.
Now that we have identified the stable vacuum of the theory after truncating at levels
up to (10, 20), a natural next step is to investigate the structure of the theory around this
stable vacuum. The stable vacuum should correspond to the empty vacuum of the closed
bosonic string theory, with modes corresponding to the open string fields on the D25-brane
decoupling from the theory. Of particular interest in this regard is the fate of the U(1) vector
field A on the brane [1820]. Some progress in understanding the structure of the theory
around the stable vacuum was made in [2]. It would be interesting to study this question
further and to investigate the spectrum of modes around the stable vacuum using the more
detailed picture we have given of the expectation values of the scalar string fields in this
vacuum. We leave these questions to further work.
4.2. Effective tachyon potential
In Section 3.2 we studied the effective potential of the tachyon through its power series
expansion around the unstable vacuum i = 0. We found that the perturbation series had a
finite radius of convergence in each of the level-truncated theories, indicating a possible
singularity near 0.25/g. In this section we study the nonperturbative effective
potential, and investigate the branch structure of the potential near the singularity.
For a fixed value of the tachyon field, there are in general many solutions of the equations
for the remaining fields i , i > 1, which correspond to different branches of the effective
potential. The perturbative expansion describes only one branch of the effective potential,
the one which is connected to the unstable vacuum. We will refer to this as branch 1. For
levels higher than (2, 4), we cannot exactly integrate out all the non-tachyonic fields, so we
are forced to use numerical methods to study the effective potential outside its radius of
convergence or on other branches. We have used Newtons method to find the zeros of the
partial derivatives of the potential. The choice of initial values for the fields determines on
which branch the algorithm will converge. At mass level 2, we can find all the solutions,
and thus determine all the branches. We can then use the field values from these branches
as initial values for the algorithm at higher levels in order to stay on the same branch.
In Fig. 3, we have plotted branch 1 of the effective potential at levels (2, 6), (4, 12) and
(6, 18). At level (4, 12), our algorithm stops converging after g 1.8. At level (6, 18), the
algorithm becomes unstable after g 1.6. This may indicate either that the branch ends
at that point or that it meets one or more other branches which play the role of attractors,
making it difficult to converge to the chosen branch. A similar breakdown of convergence
also happens at level (2, 6) around g 6, where a new pair of branches appear [2].
A particularly interesting physical question is to determine the structure of the tachyon
effective potential for negative values of . In the level 0 truncated theory, this potential
decreases to as . If this behavior is not modified by higher level corrections,
it poses the natural question of why the D25-brane would choose to condense to the stable
vacuum rather than the runaway solution at negative .
122
Fig. 3. Branch 1 of the effective tachyon potential at different truncation levels.
When we attempt to continue branch 1 of the effective potential to include arbitrary

negative values of the tachyon field, we find that at each level our algorithm stops
converging very near the radius of convergence given in Table 3, which we calculated using
the perturbative expansion coefficients of the effective potential. In [2], Kostelecky and
Samuel studied the effective potential at mass level 2 near this point. They found that the
branch connected to the unstable and stable vacua (branch 1) ends at a singularity at g
0.35 and, at that point, meets another branch (branch 2) which continues upward for
increasing values of . This branch then meets a third branch (branch 3) which continues
downward for decreasing values of . We have studied the branch structure of the effective
potential in level truncations (4, 12) and (6, 18) and we find precisely the same branch
structure, although the shape of the branches changes noticeably at each level. In Fig. 4,
a graph is given of the branch structure of the effective potential at each of these levels. We
observe that branch 3 becomes less steep as we go at higher levels, and that it begins to
approach the singular point where branch 1 meets branch 2. While branch 3 does not seem
to approach the singular point of branch 1 exponentially quickly, its rate of approach seems
compatible with the rate of convergence of the coefficients in the perturbative expansion of
the tachyon effective potential. Although we only have limited evidence for this suggestion
at this time, we conjecture that in the limit of the full string field theory, branch 3 precisely
intersects the singular point connecting branches 1 and 2. The physical effect of this would
be to create a critical point near g 0.25 at which there would be a higher-order phase
transition in the tachyon effective potential. Unlike the effective potential on branch 1, the
structure of the effective potential on branch 3 changes dramatically with each additional
level of string fields which are included. Thus, we cannot trust our low-level truncation
of the theory to accurately describe the effective potential beyond the critical point. It is
natural to speculate that the effective potential becomes non-negative (relative to the energy
of the stable vacuum) for all values of in the full string field theory.
123
Fig. 4. Structure of branches 1, 2, and 3 at different levels.
It would be very nice to have some better understanding of the physics of open bosonic
string theory beyond the critical point; there must also be a natural physical explanation
for the existence of the critical point, which might be understood from the point of view
of bosonic D-branes. We leave these questions for further research, but we conclude
this section by pointing out one possible amusing scenario: it is quite possible that the
numerical difficulties we have encountered in extending branch 1 of the effective potential
near g 1.6 indicate a second critical point at approximately the same energy as that we
have found at g 0.25. If this is indeed the case, it may be that these two critical points
should be physically identified in the full theory, so that the tachyon effective potential will
essentially become (in some coordinates) periodic, and perhaps even smooth. This would
provide a satisfying resolution to the question of what happens when the tachyon rolls in
the negative direction: it would simply bring the system to the same stable vacuum as if
it rolled in the positive direction. More evidence is needed before this possibility can be
taken seriously, but it would provide a nice scenario in which the effective potential for the
tachyon has a unique global minimum corresponding to the stable vacuum, just as occurs
for the superstring [7]. One potentially serious drawback to this scenario, however, is that
it seems to predict the existence of a stable 24-brane kink for whose existence there is no
evidence in the bosonic string theory.
5. Conclusions
In this paper we have used the level-truncation approach to string field theory to perform
a detailed study of the tachyonic instability of the open bosonic string. We have calculated
the cubic string field theory potential up to terms of mass level 20, including fields up
to level 10. We have used these results to analyze the effective tachyon potential and the
124
nonperturbative stable vacuum of the system. Our results for the perturbative form of the
effective potential indicate that the radius of convergence of this potential decreases to an
apparently finite value as higher level fields are included, indicating a critical point in the
effective potential near 0.25/g. The stable vacuum lies well outside this radius of
convergence, although the critical point and the stable vacuum lie on opposite sides of the
unstable vacuum so there is no phase transition between the two vacua. We have found
that while the coefficients in the effective potential converge relatively slowly using the
level truncation method, the energy of the stable vacuum solution and the values of the
fields in this vacuum converge much more quickly. We found that the discrepancy between
the exact D-brane tension and the vacuum energy calculated in the truncated string field
theory decreases by approximately a factor of 1/3 when string fields at each additional
mass level are included, although we do not have a theoretical explanation for this rate of
convergence.
It is perhaps somewhat surprising that this method converges more rapidly for the
vacuum energy calculation, which is a truly nonperturbative feature of the system, than
for the coefficients of the effective potential, which are in principle computable using
perturbative methods. This finding seems to indicate that string field theory has the
potential to be an extremely useful tool in studying detailed aspects of nonperturbative
string physics. Even if it is not possible to find an exact analytic solution of string field
theory describing the nonperturbative vacuum, it would be of great interest to carry out a
more detailed analysis of the asymptotic properties of the level truncation approach.
The methods used in this paper give us a fairly complete picture of the behavior of the
effective tachyon potential when 0.25 < g < 1.5. We have found that a critical point
appears in the tachyon potential near g 0.25, beyond which it is difficult to precisely
determine the physics. We have speculated that the effective potential stays above the stable
vacuum energy for negative beyond the critical point, but we do not yet have conclusive
evidence for this conclusion. It would be very nice to have a better understanding of the
behavior of the theory in the regime where the tachyon is large and negative. We have
outlined one possible scenario, in which the tachyon effective potential becomes periodic
and the stable vacuum appears at the unique minimum of this potential. If this scenario is
realized it would provide a simple answer to the question of how the theory behaves if the
tachyon chooses to roll in the negative direction. This would also provide a picture in which
unstable bosonic D-branes could naturally be interpreted as sphalerons, as advocated in the
supersymmetric case in [21].
There are many directions in which it would be interesting to extend the work in this
paper. One obvious question is to ask how far it is possible to extend the level truncation
method in the open bosonic string field theory. For the results in this paper where we
included fields at level 10 we have computed 138202 distinct cubic vertices. Because
many of these vertices involve hundreds of possible index contraction combinations, this
computation is already rather time-consuming using a higher-level symbolic manipulation
program such as M ATHEMATICA on a desktop PC. To continue to level (12, 24), over
a million vertices (1381097) would be needed, each involving hundreds or thousands of
index contractions. With a more efficient program and a powerful computer, it might be
125
possible to push as far as level (16, 32) or higher in the forseeable future if there are
physical questions of sufficient interest to motivate further development in this direction
(at this level there are over 108 cubic vertices). Although the number of vertices grows
as the cube of the number of fields at the level at which the theory is truncated, both the
number of fields and the complexity of computing each vertex grows exponentially, so that
unless a better theoretical understanding of the structure of the theory is developed it will
probably never be feasible to perform calculations in this theory beyond interactions of
total level 40 or so.
There are a number of other questions of significant interest which could be investigated
using truncations of bosonic string field theory at levels considered in this paper. As
discussed in Section 4.1 it is clearly important to attain a better understanding of the
structure of the theory around the nonperturbative vacuum. While some progress was made
in this direction in [2], we do not yet have a very clear picture of how the open string
fields such as the U(1) gauge field decouple from the theory in the stable vacuum. Another
question of interest which can be addressed in the bosonic theory is the existence of vacua
which break Lorentz symmetry. Such vacua have been shown to exist and to converge up
to level (6, 18) truncations of the theory [2,5]. It would be nice to have a better theoretical
understanding of the significance of these vacua.
A related question which can be addressed using open bosonic string field theory is that
of describing the structure of Dp-branes with p < 25 as unstable configurations of open
string fields. Solutions of this type have been found in conformal field theory [2225]. The
possibility of realizing a bosonic D(p 1)-brane as a tachyonic lump on a Dp-brane was
discussed in [26]. The use of the level truncation approach to string field theory in this
context was suggested in [13], and was implemented by Harvey and Kraus [8], who found
numerical evidence for Dp-branes with p > 18 as lumps on the D25-brane after including
effects of some level 2 fields. It would be very interesting to extend this work to higher level
truncations and to see whether the physics of all Dp-branes is indeed accurately described
in level-truncated string field theory.
The open bosonic string is an interesting model in which string field theory is particularly
simple. Studying this theory further may allow us to refine our understanding of certain
qualitative aspects of D-brane physics and tachyonic instabilities. Nonetheless, if string
theory is ever to make concrete contact with observable phenomenology it will almost
certainly be in the context of a supersymmetric theory. Given that the open string field
theory seems to be able to access interesting nonperturbative structure in string theory, it is
clearly of substantial interest to develop an equally systematic approach to performing
nonperturbative calculations using superstring field theory. While Wittens formulation
of open string field theory can be extended to the supersymmetric theory [27,28], this
formalism may be problematic due to the necessity of considering higher order contact
interactions [29,30]. Several alternative formulations of open superstring field theory have
been suggested [3133]. Recently Berkovits has used his alternative formulation of the
supersymmetric open string field theory [33] to carry out a first approximation of the
energy gap in a tachyonic brane system, and has found that 60% of the brane energy is
reproduced even in the first truncation of the theory [6]. This formalism has been more
126
explicitly developed and the calculation of the energy gap has been extended by Berkovits,
Sen and Zwiebach [7], who showed that including the three scalar fields at the next relevant
level gives 85% of the brane energy. If explicit calculations in open superstring field theory
can indeed be systematically carried out in high-level truncations as we have done for the
bosonic theory, this promises to provide an exciting new tool for investigating in detail
nonperturbative issues in string theory such as the description of non-BPS D-branes as
braneantibrane bound states [3436] and the associated connection between D-brane
charges and K-theory [19,37].
Acknowledgements
We would particularly like to thank Barton Zwiebach for helpful discussions and
explanations in the course of this work. Thanks also to Amihay Hanany, Yang-Hui He,
Andreas Karch, Joe Minahan, Leonardo Rastelli and Ashoke Sen for helpful conversations.
The work of NM was supported in part by the DOE through contract #DE-FC0294ER40818. The work of WT was supported in part by the A.P. Sloan Foundation and
in part by the DOE through contract #DE-FC02-94ER40818.
Appendix A. Table of scalar states at levels 6 6
The following Table A.1 describes the scalar states at levels 0, 2, 4 and 6. The SZ column
relates these fields to the background-independent fields used in [13]. m and g are the
indices of the matter and ghost states into which each scalar decomposes through (9).
gh i iL denote expectation values of the scalar fields in level truncations (4, 12), (6, 18)
and (10, 20).
Appendix B. Table of matter and ghost states at levels 6 6
The following Tables B.1, B.2 list the matter and ghost states which contribute to scalar
fields at levels 6 6. The quadratic terms involving each state are listed in the last column.
Appendix C. Cubic interactions at levels (6, 16)

In the following pages we give Tables C.1, C.2 of all cubic interactions between the pure
matter and ghost scalar fields listed in Appendix B which contribute to cubic interactions of
total dimension 6 16 between the scalar fields listed in Appendix A. All cubic interactions
of level less or equal to (6, 16) between the scalar fields listed in Appendix A can be
reproduced using these tables and (10).
127
Table A.1
g
gh i i4
gh i i6
gh i i10
1.09680
1.09586
1.09259
(1 1 ) |0i
b1 c1 |0i
2
1
1 0.05692 0.05714 0.05723

2 0.41135 0.42363 0.43372
(1 3 ) |0i
(2 2 ) |0i
(1 1 )(1 1 ) |0i
(1 1 ) b1 c1 |0i
b1 c3 |0i
b2 c2 |0i
b3 c1 |0i
4
5
6
2
1
1
1
1 0.01142 0.01146 0.01148

1 0.00512 0.00511 0.00509
1 0.00029 0.00031 0.00032
2 0.00686 0.00740 0.00786
5 0.11242 0.11478 0.11654
6 0.06621 0.06813 0.06990
7 0.03747 0.03826 0.03885
(1 5 ) |0i
(2 4 ) |0i
(3 3 ) |0i
(1 3 )(1 1 ) |0i
(2 2 )(1 1 ) |0i
(1 2 )(1 2 ) |0i
(1 1 )(1 1 )(1 1 ) |0i
(1 3 ) b1 c1 |0i
(2 2 ) b1 c1 |0i
(1 1 )(1 1 ) b1 c1 |0i
(1 2 ) b1 c2 |0i
(1 2 ) b2 c1 |0i
(1 1 ) b1 c3 |0i
(1 1 ) b2 c2 |0i
(1 1 ) b3 c1 |0i
b1 c5 |0i
b2 c4 |0i
b3 c3 |0i
b4 c2 |0i
b5 c1 |0i
b2 b1 c2 c1 |0i
10
11
12
13
14
15
16
4
5
6
3
3
2
2
2
1
1
1
1
1
1
SZ
|0i
2 v/ 52
3
u
4 A + B
5
A/2
6
B/4
7 F /2
8
C
9
E
10
D
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
State m
1
1
1
1
1
1
1
2
2
2
3
4
5
6
7
12
13
14
15
16
17
0.00349
0.00293
0.00144
0.00028
0.00015
0.00001
0.00000
0.00263
0.00116
0.00008
0.00013
0.00006
0.00336
0.00283
0.00112
0.06003
0.03750
0.02283
0.01875
0.01201
0.01547
0.00350
0.00291
0.00143
0.00028
0.00015
0.00001
0.00000
0.00280
0.00121
0.00009
0.00015
0.00007
0.00351
0.00295
0.00117
0.06094
0.03816
0.02291
0.01908
0.01219
0.01612
128
Table B.1 Matter states

|m i
State
|1 i
|0i
|2 i
(1 1 ) |0i
d2mat
2 = 2 13
|3 i
(1 2 ) |0i
2
d3mat
3 = 2 13
|4 i
(1 3 ) |0i
2
d4mat
4 = 3 13
|5 i
(2 2 ) |0i
3
d5mat
5 = 2 3 13
|6 i
(1 1 )(1 1 ) |0i
|10 i
(1 5 ) |0i
mat = 52 13
d10
10
|11 i
(2 4 ) |0i
mat = 23 5 13
d11
11
|12 i
(3 3 ) |0i
mat = 2 32 5 13
d12
12
|13 i
(1 3 )(1 1 ) |0i
mat = 23 3 5 7 13
d13
13
|14 i
(2 2 )(1 1 ) |0i
|15 i
(1 2 )(1 2 ) |0i
mat = 25 5 132 , d mat = 24 5 13

d14
14
14 15
mat = 24 5 13, d mat = 23 33 5 13
d15
14
15 15
|16 i
(1 1 )(1 1 )(1 1 ) |0i
Quadratic terms
1
d1mat
1 = 2
5
d6mat
6 = 2 3 7 13
mat = 27 32 52 7 13
d16
16
Table B.2 Ghost states

|g i
State
|1 i
|0i
d1 1 = 21
|2 i
b1 c1 |0i
d2 2 = 21
|3 i
b1 c2 |0i
d3 4 = 1
|4 i
b2 c1 |0i
d4 3 = 1
|5 i
b1 c3 |0i
d5 7 = 32
|6 i
b2 c2 |0i
d6 6 = 32
|7 i
b3 c1 |0i
d7 5 = 32
|12 i
b1 c5 |0i
d12 16 = 52
|13 i
b2 c4 |0i
d13 15 = 52
|14 i
b3 c3 |0i
d14 14 = 52
|15 i
b4 c2 |0i
d15 13 = 52
|16 i
b5 c1 |0i
d16 12 = 52
|17 i
b2 b1 c2 c1 |0i
Quadratic terms
gh
gh
gh
gh
gh
gh
gh
gh
gh
gh
gh
gh
gh
d17 17 = 52
Table C.1
Cubic matter field coefficients tijmat
k
i
k
2513
3
2513
3
10
11
12
13
14
15
16
0
26 13
35
22 132
35
3
2 52 713
36
6 5132
2
39
12 13
2
9
3
2131947
38
8 5713
2
38
3 5133
2
38
2
2
2 513
38
4 4
2 5 713
38
3
22 71383
36
12 13 3
2
8
3
7
2 1373
38
3 13487
2
39
4
2 571331
36
27 5132 73
312
213 1341
312
2
2 13103571
311
29 71323
38
24 11132 1129
312
3
2 11131129
312
25 53 7131093
311
212 13 3
38
15
2
13
39
212 713 3
310
216 13 3
10
3
15 5713 3
2
11
3
12
3
2 51343
314
18
3
2 513
314
12
3
2 513
311
14
3
2 713
312
14
2
2 1310 3
3
13
3
2 13
10
3
215 53 713 3
312
26 13
35
7 1373
2
38
12 713 3
2
10
3
210 13479
311
28 13163
39
9
2 5713
37
10 513251
2
313
16 13
2
14
3
7
2
11131523
313
12 7134753
2
314
9 132 2549
2
313
8
2 132549
313
10 53 71389
2
313
22 132
35
3 13487
2
39
16 13 3
2
10
3
28 13163
39
24 13171093
310
5
2 13223
2 57 12
3
8
2 51331139
314
14 13233
2
314
3
2 71344269
313
10 71317311
2
314
5 7132 36013
2
314
4
2 71336013
314
26 53 713401
312
23 52 713
36
4 571331
2
36
215 5713 3
11
3
29 5713
37
5 572 13223
2
312
6
2 713100537
311
9 52 7132
2
311
15 571317
2
315
4
2 571359581
314
11 71319599
2
313
26 713257981
315
5
2 713678779
314
7 52 713522283
2
314
129
130
Table C.1 continued

i
k
10
11
12
13
14
15
16
22 71383
36
212 13 3
8
3
7
2 1373
38
3 13487
2
39
4 571331
2
36
27 5132 73
312
213 1341
312
22 13103571
311
29 71323
38
24 11132 1129
312
23 11131129
312
25 53 7131093
311
15
2
13
39
3
2 7134173
8
0
28 713397
310
4 1379157
2
311
25 713613119
311
8 57132 397
2
314
14 513733
2
315
3 7133153593
2
314
10 71323623
2
313
25 72 1319243
314
4 7135996691
2
315
6 52 713958361
2
314
216 1319
311
216 1319
311
13 13743 3
2
313
14
3
2 1323101
314
215 71331 3
311
213 5137307 3
317
219 13773 3
317
214 136883 3
16
3
215 72 13151 3
315
218 1327043 3
317
17 1371359 3
2
17
3
17 2
3
2 5 7131093
316
28 713397
310
213 13743 3
13
3
11
2 5135683
314
9 13113257
2
313
10 71321247
2
313
211 51358321
317
17 713233
2
317
28 131109113
316
213 571317909
316
210 139293673
316
9 1367137587
2
317
211 52 713241373
316
19 1317
2
313
4
2 1379157
11
25 713613119
311
15 71331 3
2
11
3
10
2 71321247
313
26 72 132 17351
315
7 71313224697
2
314
210 57132 21247
317
216 7132 283
317
25 7138116779
315
212 7131536989
316
7 713199109
2
312
26 711132794133
317
28 57131314192389
316
18 713167
2
314
214 1323101 3
14
3
9
2 13113257
313
5 135315137
2
313
26 72 132 17351
315
29 52 7136047
317
15 1113109
2
317
4 1380863921
2
317
211 72 1328387
316
26 136191542007
318
5 1313114473187
2
318
7 52 71314832711
2
317
17 5713
2
313
Table C.1 continued

i
k
12
3
2 713
10
16
3
2 13
10
10
11
12
13
14
15
16
215 5713 3
11
3
212 51343 3
314
218 513 3
314
212 513 3
11
3
214 713 3
312
214 132 3
310
213 13 3
10
3
15 3
3
2 5 713
312
210 13479
311
28 13163
39
213 13743 3
13
3
214 1323101 3
314
215 71331 3
11
3
13
3
2 5137307
317
19
3
2 13773
317
14
3
2 136883
316
15 2
3
2 7 13151
315
18
3
2 1327043
317
217 1371359 3
317
217 52 7131093 3
316
11
2 5135683
314
9 13113257
2
313
19
2 1317
13
17
2 5713
13
14
3
2 131889
15
3
18 713167
2
314
219 5132 17
317
225 1317
317
216 138513
317
21
2
713
317
18 1373757
2
318
217 1372101
317
219 52 72 13139
317
0
214 131889
315
215 7131889 3
15
3
21 3 13 3
2 5 18
3
224 13 3
316
13 13144847 3
2
18
3
24
3
2 571341
319
15
3
2 51319419
319
214 513139739 3
318
16 52 71347173 3
2
318
217 132357
315
212 13463781
317
214 131889
315
217 7131889 3
16
3
14
3
2 513149233
319
20
3
2 513599
319
14
3
2 1350707
318
16
3
2 713472081
319
16
3
2 13915379
319
215 132 311419 3

319
17 52 713773 3
2
317
212 13463781
317
210 13653661
315
15
3
2 7131889
15
17
3
2 7131889
16
215 572 1331101 3

319
221 7131523 3
319
215 713432897 3
19
3
217 5711132383 3
318
217 7132119221 3
320
16 71113471163 3
2
19
3
18
3
2 5713522283
319
213 72 1341151
316
211 713361687
317
131
132
Table C.1 continued

i
k
10
11
12
13
14
15
16
10
1
29 5713
37
24 13171093
310
5 572 13223
2
312
26 713100537
311
10 513251
2
313
16 13
2
14
3
7 11131523
2
313
12 7134753
2
314
9 132 2549
2
313
8 132549
2
313
10 53 71389
2
313
8 51331139
2
314
10
2 71321247
13
15
3
2 7131889
15
3
3
5
2 135315137
313
26 72 132 17351
315
7 71313224697
2
314
211 51358321
317
17 713233
2
317
28 131109113
316
213 571317909
316
210 139293673
316
9 1367137587
2
317
211 52 713241373
316
29 52 7136047
317
217 7131889
316
0
221 53 13
318
24
3
2 13
16
213 13144847 3
318
224 571341 3
319
215 51319419 3
19
3
14
3
2 513139739
318
216 52 71347173 3
318
214 513149233 3
19
3
213 72 1341151
316
210 13653661
315
211 713361687
317
212 7131516037
316
17 51342323
2
320
26 111319
2
320
11 1313727823
2
319
19 71389137
2
318
13 11134761441
2
320
12 1318654169
2
320
14 52 72 133 101
2
319
12 52 11137727
2
319
211 713361687
317
26 1113197151
314
27 71333179283
318
8 571311099833
2
317
12 52 11137727
2
319
18 1314249
2
319
9 132 278269
2
318
14 7131928961
2
319
11 131002523
2
315
10 111316213121
2
319
12 52 711137727
2
318
10 5134394717
2
319
212 7131516037
316
27 71333179283
318
8 571311099833
2
317
29 7111399710181
315
13 57134261
2
316
219 713673
317
10 7131886231
2
318
15 57132336217
2
319
12 7111314913171
2
319
211 713293313
314
13 572 137872477
2
318
11 572 133761109
2
321
Table C.1 continued

i
j
k
11
12
13
14
15
16
10
11
12
13
14
15
16
14
2 13233
14
3
23 71344269
313
10 71317311
2
314
5 7132 36013
2
314
4 71336013
2
314
26 53 713401
312
9 52 7132
2
311
15 571317
2
315
4 571359581
2
314
11 71319599
2
313
26 713257981
315
25 713678779
314
7 52 713522283
2
314
15
18
220 513599 3
2 1113109
2 1314249
317
319
319
14 1350707 3
4 1380863921
9 132 278269
2
2
2
317
318
318
14 7131928961
216 713472081 3
211 72 1328387
2
316
319
319
16
6
11 131002523
2 13915379 3
2 136191542007
2
318
319
315
15
2
5
10 111316213121
2
13
311419
3
2
1313114473187
2
318
319
319
17 52 713773 3
7 52 71314832711
12 52 711137727
2
2
2
317
317
318
15
2
10
2
13 57134261
2
57
1331101
3
2 5713 21247
2
317
319
316
21
16
2
19 713673
2
7131523
3
2 713 283
2
317
319
317
15 713432897 3
10 7131886231
2
25 7138116779
2
315
319
318
17
12
15 57132336217
2 7131536989
2
5711132383
3
2
316
318
319
17
7
12 7111314913171
2
7132119221
3
2
713199109
2
312
320
319
216 71113471163 3
26 711132794133
211 713293313
317
319
314
13 572 137872477
218 5713522283 3
28 57131314192389
2
316
319
318
5
16
2 1382267
18
217 713232389
3
320
25 1358008371
26 7131597319129
318
320
12 71325715359
13 713195501
2
2
319
315
7
8 7138867199049
2 1352387049
2
318
320
6
7 7138422919627
2
1319491192547
2
319
321
8 52 72 112 139749
9 571319113681761
2
2
319
320
11 572 133761109
12 57132 1516037
2
2
321
320
17
18 72 138467
2 713232389
2
320
320
7 73 13471932671
26 7131597319129
2
320
319
13
14 71312167927
2
713195501
2
315
317
8
9 52 72 13473411
2
7138867199049
2
320
316
8 52 7132925394441
27 7138422919627
2
321
320
9
2 571319113681761 210 571319185072623
320
318
133
134
Table C.1 continued

i
k
10
10
10
11
10
12
10
13
10
14
10
15
10
16
11
11
11
12
11
13
11
14
11
15
11
16
12
12
210 513109839
317
216 51329
314
27 572 134767
317
212 571337113
317
29 57132 3271
317
28 57133271
317
210 54 7132 89
317
222 13151
317
213 13173373
317
18 72 132
2
317
215 132 853
317
214 13853
317
16 53 72 13
2
317
22 13173367057
316
11
2 51325159107
21
217 52 71317033
322
227 513599
323
211 52 13773063
321
219 57132792
322
213 51343252291
323
212 51343671567
321
214 53 713291553
320
231 71331
322
217 1337227311
322
229 71383
323
219 513188767
323
218 513262153
322
20 52 71329723
2
322
28 133713732591
320
212 51316686323
322
218 513768437
323
9 51370937567
2
322
214 57135795159
323
211 513843766543
323
210 513517123747
322
12 53 713431213
2
322
224 11132887
323
215 1312201337
322
220 57131993
322
217 138037049
322
216 1337143319
323
18 52 71383857
2
322
4 13540822211
2
321
213 52 7133142743
322
219 57133128927
324
210 571347149143
321
215 57139732327
321
212 5713376016133
323
11 5713173763033
2
322
213 52 72 132 7872477
322
225 71353647
323
216 71320818581
323
21 713529531
2
322
18 71331735527
2
324
17 713191131449
2
323
19 571113625367
2
323
25 71337789649033
322
3
17
2 513168013
321
8 51311307557
2
320
13 5713192463
2
319
10 52 1397101557
2
320
29 52 1113575251
321
11 53 7132 241373
2
320
23 72 13167
2
320
14 513777317
2
320
219 7132 461
319
16 1310313469
2
321
215 131817927
320
217 52 7138539
320
3 1322571334751
2
319
Table C.1 continued

i
k
12
13
12
14
12
15
12
16
13
13
13
14
13
15
13
16
14
14
14
15
14
16
15
15
15
16
16
16
29 71384533
316
4 132 2336607
2
316
3 132336607
2
316
25 53 713292081
316
214 7132 3943
316
211 713723103
317
210 713232313
316
212 52 713232113
316
26 133974876247
318
25 13770448241
318
7 52 7132692939
2
316
24 13622014119
317
6 52 71315113757
2
316
28 57132367101317
315
10 2
2 7 13469801
18
213 72 132 257857

322
210 111320838609
321
29 132 46317911
321
211 52 71323251483
321
221 571349669
320
215 713194108801
322
214 71331225183
322
216 5713349215801
322
212 13167191856721
323
211 13192932519347
323
213 572 11131977479
321
210 133 343313883
322
12 57131931099547
2
321
214 52 713153516109
320
211 71379304153
322
6 131628880601
2
322
5
2 133113722910321
322
7 52 71323995369
2
319
216 713118767577
323
213 713831272397
322
212 713683836203
323
214 5713182363391
322
8 513194015474007
2
323
27 51380743100293
323
29 57133361619981
321
26 51325493228103
322
8 572 133598876653
2
322
10 5713230727914457
2
321
212 713167962661
321
7 713593126649637
2
323
6 711131175551529
2
322
28 5713833407109579
322
217 713893588601
322
214 7131971443659
321
13 7111314718779
2
321
215 52 71373609741
320
29 71113312441317421
324
8 7133796581201997
2
324
10 5713221979487743
2
323
27 71389161993441
320
29 57111329333163613
322
211 52 713121821760531
321
3
25 1321221396061
320
24 1315641122557
320
6 52 72 1319539257
2
319
15 72 136112433
2
319
12 72 136151607
2
319
11 711133971759
2
319
13 5713791571091
2
318
7 53 13823116673
2
321
6 52 132 90630013
2
321
28 57139743261631
320
5 52 131303790523
2
320
27 571317353265541
320
9 571389497316369
2
318
135
136
Table C.2
gh
Cubic ghost field coefficients tij k
i
k
113
113
24 5
36
12
13
14
15
16
17
19
35
24 5
35
4
2
71
39
5 52
2
9
3
2099
39
6 52
2
9
3
4 571
2
9
3
1119
38
19
35
3
19
35
26 5 3
8
3
27 5 3
38
4 2
2 5
38
4389
9
3
4 2
2 35
7
24 571
311
25 5211
312
18289
312
26 5211
312
4
2 52 71
311
38449
312
14
3
3
26 52
39
27 7
38
26 52 3
311
7 57 3
2
10
3
6
2 199 3
3
26 571 3
314
27 713 3
12
3
6
3
2 519
312
8 3
2 514 3
3
26 543 3
12
3
26 52 3
310
7
2 58 3
3
27 7
38
28 52
39
27 19 3
310
8 57 3
2
10
3
7 2
2 510 3
3
27 43 3
312
28 53 3
14
3
7
3
2 519
312
29 713 3
312
7
2
2 5 71 3
14
3
27 52 3
310
6
2 58 3
24 5
36
4 2
2
5
38
26 52 3
11
3
27 19 3
310
26 52
311
4 5719
2
311
6 1747
2
312
6 571
2
14
3
27 1723
315
4 52 433
2
314
8 52 269
2
315
6
2 52089
315
24 51973
314
24 5
38
19
35
4389
9
3
27 57 3
10
3
28 57 3
310
4
2 5719
311
31131
9
3
4
2 5719
310
24 37103
314
5 54
2 13
3
2
17 1983
314
6 54
2
13
3
24 537103
314
1033313
314
196571
312
24 5
35
4 2
2
35
7
26 19 3
9
3
27 52 3
310
6
2 1747
312
4 5719
2
310
26 52
39
6 2089
2
14
3
7 52 269
2
314
4 52 433
2
313
28 1723
314
6
2
2 51371
3
24 51973
313
24 5
37
Table C.2 continued

i
j
6
2 58 3
7
2 58 3
12
13
14
15
16
17
3
4 52
2
8
3
4389
39
4 52
2
37
24 571
311
25 5211
312
18289
312
26 5211
312
24 52 71
311
38449
312
26 52
39
27 7
38
28 52
39
26 52 3
11
3
24 5
38
196571
312
24 5
37
4
2
71
311
5
2 519109
314
107283
14
3
6 519109
2
314
4 571
2
11
3
6571
11
3
6 2
2
5
39
27 137
312
8 2
2 5
39
6 52 3
2
11
3
6 2
2 5
9
3
27 137
312
26 52 3
11
3
26 52 713 3
314
26 257 3
312
6 571 3
2
14
3
7
3
2 43227
317
7
3
2 51729
317
28 52 109 3
316
6 57461 3
2
316
27 5137 3
314
27 137
312
8 2
2 5
39
27 257 3
13
3
27 52 713 3
314
27 52 3
310
7 7461 3
2
16
3
28 52 109 3
316
8
3
2 51729
317
9
3
2 43227
317
7 2
3
2 5 71
314
28 5137 3
314
213 52
314
6
2 511 3
6 2
3
2 5 713
14
6
3
2 257
12
7
3
2 257
13
7 2
3
2 5 713
14
7 2
2 510 3
3
6 52
2
11
3
24 57887
315
6 71113
2
314
26 571
314
7 4173
2
316
24 512491
317
28 52 13137
317
26 51032
317
24 51331
315
0
213 52
314
17
2 155
3
3
24 57887
315
1313331
314
24 57887
314
4 531773
2
318
25 572 1667
318
475361199
318
26 572 1667
318
4 52 31773
2
318
463116167
318
26 52
39
7 72 97
2
14
3
28 52
39
6 52 397 3
2
16
3
3
6 71113
2
314
24 57887
314
6 2
2 5
39
26 1032
316
27 52 13137
316
24 512491
316
8 4173
2
315
26 52 71
313
24 51331
314
15
2 145
3
213 52
313
0
214 523
317
137
138
Table C.2 continued

i
j
k
6
12
13
14
15
16
17
12
13
14
15
27 57 3
10
3
26 19 3
39
27 19 3
10
3
28 57 3
310
27 52 3
310
6 571 3
2
14
3
7
3
2 713
312
26 519 3
312
28 53 3
14
3
6
3
2 543
312
6 2
2 510 3
3
7 43 3
2
12
3
28 53 3
314
27 519 3
312
9 713 3
2
12
3
6 2
3
2 5 713
14
3
6
3
2 257
312
7
3
2 257
313
7 2
3
2 5 713
314
7 2
2 510 3
3
26 571 3
14
3
27 43227 3
317
27 51729 3
317
8 52 109 3
2
16
3
26 57461 3
316
7
3
2 5137
314
27 7461 3
316
8 52 109 3
2
316
28 51729 3
317
29 43227 3
17
3
3
26 52
39
15
2
145
3
213 52
314
7 72 97
2
14
3
213 52
313
0
15
2 559
18
3
6 53 967
2
318
0
215 52
318
6 2
2 5
39
13 89
2
16
3
214 57
316
27 111409
317
215 57
316
4
7 2
2 71497
3
213 52
313
17
2 155
3
28 52
39
0
13 89
2 16
3
214 57
316
27 111409
317
215 57
316
13 589
2
16
3
7 797137
2
318
217 5
318
0
8 3
2 5 18967
3
18
2 559
18
26 52 397 3
16
3
14
3
2 523
317
27 72 13 3
314
215 523 3
317
6 52 863 3
2
319
214 53 11 3
19
3
14 3
2 5 207 3
3
6 2
3
2 5 397
316
216 11 3
318
14 5211 3
2
320
27 1911491 3
319
15 52 71 3
2
19
3
26 72 13
313
6 2
3
2 7 13
13
7 2
2 7 13
314
7 2
3
2 5 397
15
26 55641
13
3
2 5349
20
318
27 717331 3
318
6
3
2 5197
319
28 53 3
314
26 5711489 3
319
6
3
2 5797
314
27 711489 3
319
28 53 3
314
7
3
2 5197
319
29 717331 3
318
0
15
2 523
317
27 52 397
315
0
214 373
319
13 2 71 3
2 5 18
26 1911491 3
318
14 5211 3
2
19
3
215 511 3
317
6
3
2 663797
319
15 2
2 5 197 3
3
214 53 11 3
318
7 2
3
2 5 863
318
15
3
2 5349
319
Table C.2 continued

i
j
16
17
12
13
14
15
16
17
12
7 2
3
2 5 71
14
3
7 2
2 510 3
3
26 52
311
4 5719
2
311
6 1747
2
312
31131
9
3
4
2 5719
310
26 52
39
6 571
2
14
3
27 1723
315
4 52 433
2
314
8 52 269
2
315
6 52089
2
315
24 51973
314
24 37103
314
27 52 71 3
314
8 5137 3
2
314
6 52
2
11
3
24 57887
315
6 71113
2
314
1313331
314
24 57887
314
6 2
2 5
39
26 571
314
7 4173
2
316
24 512491
317
28 52 13137
317
26 51032
317
24 51331
315
4 531773
2
318
13
2 589
16
3
7
2 797137
318
6 2
3
2 5 397
316
214 523 3
317
0
26 72 13
313
0
0
213 5349 3
320
26 52 863 3
319
14 53 11 3
2
19
3
214 53 7 3
320
26 52 397 3
316
6 55641 3
2
18
3
215 5373 3
320
7 663797 3
2
320
0
8 2
2 5
39
6 2
2 5 1929
317
20
2
175
3
4
2 5131397
316
26 72 5641
318
20
2 165
3
7 2
3
2 7 13
314
15
3
2 523
317
27 52 397
315
0
16
2 11
318
214 5211 3
320
7 1911491 3
2
319
215 52 71 3
19
3
15
3
2 5373
320
27 663797 3
320
7 711489 3
2
19
3
0
214 5711
320
6 52
2 11
3
15 3
2 52011
3
20 2
2 205
3
26 52 1929
317
26 589313
320
27 52 5641 3
318
7 5797 3
2
314
6 52 1929
2
317
4 5131397
2
316
26 72 5641
318
312
38
4 5131397
2
315
6 52 1929
2
315
26 589313
320
27 487599
320
24 519227307
320
8 52 74657
2
320
26 5773491
320
24 52 3971499
320
24 1315641
318
0
27 52 397
3
315
20
2 175
3
26 72 5641
318
20
2 165
3
4
2 5131397
315
6 52 1929
2
315
0
222 7
320
14
2 51019
21
3
6 3760769
2
320
15 51019
2
321
222 57
320
6 1130627
2
321
26 773491
319
139
140
Table C.2 continued

i
k
13
14
15
16
17
12
13
14
15
16
17
1
5 54
2 13
3
2
17 1983
314
6 54
2
13
3
24 537103
314
1033313
314
6 2089
2
14
3
7 52 269
2
314
4 52 433
2
313
28 1723
314
6 52 71
2
13
3
24 51973
313
25 572 1667
318
475361199
318
26 572 1667
318
4 52 31773
2
318
463116167
318
26 1032
316
27 52 13137
316
24 512491
316
8 4173
2
315
26 52 71
313
24 51331
314
7
3
2 717331
18
3
26 5197 3
319
8 53 3
2
314
6 5711489 3
2
19
3
26 5797 3
314
14
3
2 373
319
213 52 71 3
318
6 1911491 3
2
318
214 5211 3
319
15 511 3
2
17
3
26 663797 3
319
8 3
2 514 3
3
27 5197 3
319
9
3
2 717331
318
7 2
3
2 5 5641
318
27 5797 3
314
215 52 7 3
319
14 53 11 3
2
18
3
27 52 863 3
318
215 5349 3
319
0
27 52 397
315
27 487599
320
24 519227307
320
8 52 74657
2
320
26 5773491
320
24 52 3971499
320
222 7
320
14 51019
2
321
6 3760769
2
320
15 51019
2
321
222 57
320
6 1130627
2
321
25 52 31
313
1113119289
318
26 52 31
313
24 51315641
318
313331
313
26 773491
319
7 52 74657
2
319
24 519227307
319
28 487599
319
26 52 89313
319
24 52 3971499
319
7 2
2 5 74657
19
3
24 519227307
319
28 487599
319
26 52 89313
319
24 52 3971499
319
20
2 185
3
14 3
2 51811
3
6 2
2 5
39
215 5711
318
0
26 52 1929
315
Table C.2 continued

i
k
12
12
12
13
12
14
12
15
12
16
12
17
13
13
13
14
13
15
13
16
13
17
14
14
14
15
14
16
26 712
317
7 7993
2
18
3
24 1119521
318
8 5277
2
314
6 6711083
2
318
4 52 487
2
316
28 54
317
5 5712251
2
318
9 1114887
2
318
7 52 277
2
314
25 517139
316
239710891
317
6 5712251
2
318
24 51119521
318
6
2
2 71
17
6
2 715531
22
220 71
322
214 52 563
323
26 72 1317109
322
15 1213
2
321
22 5713
2
322
6 72 317621
2
323
215 54 11
322
27 52 277903
321
16 533427
2
323
14 53 1123
2
323
27 52 96911
321
4 52397
2
313
8 29338839
2
322
26 52 6710343
322
3
27 7723
319
4 5106921
2
320
28 5409831
321
26 72 532
320
4 79301
2
318
8 53 263
2
320
25 518110159
321
9 193601
2
321
27 52 409831
321
5 54 47109
2
320
72 132397173
320
26 518110159
321
4 52 106921
2
320
14
2 31
2 11
23
3
26 56710343
323
15 52 1123
2
324
22 5713
2
323
6 589313
2
320
15 52 711
2
323
7 29338839
2
323
16 533427
2
324
14 51213
2
322
7 61367781
2
324
4 52397
2
314
28 52 277903
322
26 572 1317109
323
3
7
2 189187
321
24 25908643
323
8 54174887
2
323
26 1322929453
323
4 514995641
2
322
28 54
317
5 56828197
2
322
29 18129207
322
7 52 4174887
2
323
5 5713243
2
318
194726310891
322
6 56828197
2
322
24 525908643
323
141
142
i
j
14
17
413928843
21
4
2 5111949877
23
15
15
16
15
17
16
16
6139025327
323
210 54
317
8 5189187
2
321
6 5713243
2
318
6 52 715531
2
322
4 52 14995641
2
322
1123127191
317
4
2 5111949877
22
15
19298933
317
210 54
317
8 57993
2
318
26 517139
316
26 52 712
317
4 53 487
2
316
1129151191
318
16
17
17
17
3
10 53 263
2
320
28 57723
319
6 54 47109
2
320
6 52 712
2
317
4 579301
2
318
2329127151
317
3
217 54 11
323
215 53 563
324
28 52 96911
322
220 52 71
323
6 572 317621
2
324
4 529151397
2
320
3
17 52 711
2
322
15 5112 31
2
322
8 61367781
2
323
0
6 2
2 5 89313
319
4 529151397
2
319
Table C.2 continued
143
References
[1] K. Bardakci, Spontaneous symmetry breaking in the standard dual string model, Nucl. Phys.
B 133 (978) 297.
[2] V.A. Kostelecky, S. Samuel, On a nonperturbative vacuum for the open bosonic string, Nucl.
Phys. B 336 (1990) 263296.
[3] J. Polchinski, Dirichlet-branes and Ramond-Ramond charges, Phys. Rev. Lett. 75 (1995) 4724,
hep-th/9510017.
[4] A. Sen, Universality of the tachyon potential, hep-th/9911116.
[5] V.A. Kostelecky, R. Potting, Expectation values, Lorentz invariance, and CTP in the open
bosonic string, Phys. Lett. B 381 (1996) 89, hep-th/9605088.
[6] N. Berkovits, The tachyon potential in open NeveuSchwarz string field theory, hep-th/
0001084.
[7] N. Berkovits, A. Sen, B. Zwiebach, Tachyon condensation in superstring field theory, hep-th/
0002211.
[8] J.A. Harvey, P. Kraus, D-branes as unstable lumps in bosonic open string field theory, hep-th/
0002117.
[9] W. Taylor, D-brane effective field theory from string field theory, hep-th/0001201.
[10] A. Leclair, M.E. Peskin, C.R. Preitschopf, String field theory on the conformal plane (I), Nucl.
Phys. B 317 (1989) 411463.
[11] M.R. Gaberdiel, B. Zwiebach, Tensor constructions of open string theories 1., 2., Nucl. Phys.
B 505 (1997) 569, hep-th/9705038; Phys. Lett. B 410 (1997) 151, hep-th/9707051.
[12] E. Witten, Non-commutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253.
[13] A. Sen, B. Zwiebach, Tachyon condensation in string field theory, hep-th/9912249.
[14] D.J. Gross, A. Jevicki, Operator formulation of interacting string field theory (I), (II), Nucl.
Phys. B 283 (1987) 1; Nucl. Phys. B 287 (1987) 225.
[15] E. Cremmer, A. Schwimmer, C. Thorn, The vertex function in Wittens formulation of string
field theory, Phys. Lett. B 179 (1986) 57.
[16] S. Samuel, The physical and ghost vertices in Wittens string field theory, Phys. Lett. B 181
(1986) 255.
[17] V.A. Kostelecky, S. Samuel, The static tachyon potential in the open bosonic string, Phys. Lett.
B 207 (1988) 169.
[18] M. Srednicki, IIB or not IIB, JHEP 08 (1998) 005, hep-th/9807138.
[19] E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th/9810188.
[20] P. Yi, Membranes from five-branes and fundamental strings from Dp-branes, Nucl. Phys. B 550
(1999) 214, hep-th/9901159.
[21] J.A. Harvey, P. Horava, P. Kraus, D-sphalerons and the topology of string configuration space,
hep-th/0001143.
[22] C. Callan, I.R. Klebanov, A.W. Ludwig, J. Maldacena, Exact solution of a boundary conformal
field theory, Nucl. Phys. B 422 (1994) 417, hep-th/9402113.
[23] J. Polchinski, L. Thorlacius, Free fermion representative of a boundary conformal field theory,
Phys. Rev. D 50 (1994) 622, hep-th/9404008.
[24] P. Fendley, H. Saleur, N.P. Warner, Exact solution of a massless scalar field with a relevant
boundary interaction, Nucl. Phys. B 430 (1994) 577, hep-th/9406125.
[25] A. Recknagel, V. Schomerus, Boundary deformation theory and moduli spaces of D-branes,
Phys. Lett. B 545 (1999) 233, hep-th/9811237.
[26] A. Sen, Descent relations among bosonic D-branes, hep-th/9902105.
[27] E. Witten, Interacting field theory of open superstrings, Nucl. Phys. B 276 (1986) 291.
[28] D.J. Gross, A. Jevicki, Operator formulation of interacting string field theory (III), Nucl. Phys.
B 293 (1988) 29.
144
[29] C. Wendt, Scattering amplitudes and contact interactions in Wittens superstring field theory,
Nucl. Phys. B 314 (1989) 209.
[30] J. Greensite, F.R. Klinkhamer, Superstring amplitudes and contact interactions, Phys. Lett.
B 304 (1988) 108.
[31] I.Ya. Arefeva, P.B. Medvedev, A.P. Zubarev, Background formalism for superstring field
theory, Phys. Lett. B 240 (1990) 356362.
[32] C.R. Preitschopf, C.B. Thorn, S.A. Yost, Superstring field theory, Nucl. Phys. B 337 (1990)
363.
[33] N. Berkovits, Super-Poincar invariant superstring field theory, Nucl. Phys. B 450 (1995) 90,
hep-th/9503099.
[34] A. Sen, Stable non-BPS bound states of BPS D-branes, JHEP 9808 (1998) 010, hep-th/9805019.
[35] A. Sen, Tachyon condensation on the brane antibrane system, JHEP 9808 (1998) 012, hepth/9805170.
[36] O. Bergman, M.R. Gaberdiel, Type I D-particle and interactions, Phys. Lett. B 441 (1998) 133,
hep-th/9806155.
[37] P. Horava, Type IIA D-branes, K-theory, and matrix theory, Adv. Theor. Math. Phys. 2 (1999)
1373, hep-th/9812135.

Diagrams of noncommutative 3 theory

from string theory
Oleg Andreev 1 , Harald Dorn 2
Institut fr Physik, Humboldt-Universitt zu Berlin, Invalidenstrae 110, D-10115 Berlin , Germany
Received 29 March 2000; revised 17 May 2000; accepted 23 May 2000
Abstract
Starting from tree and one-loop tachyon amplitudes of open string theory in the presence of
a constant B-field, we explore two problems. First we show that in the noncommutative field theory
limit the amplitudes reduce to tree and one-loop diagrams of the noncommutative 3 theory. Next,
we check factorization of the one-loop amplitudes in the long cylinder limit. 2000 Elsevier Science
B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Db
Keywords: Strings; Noncommutative field theory
1. Introduction
The fact that string theory can not only be used as a candidate for unified theories but
as a powerful tool for computing perturbative field theory amplitudes is known since the
early seventies. In the simplest case of a scalar field Scherk showed how to derive tree and
one-loop diagrams of the 3 theory from the dual model (prestring theory) [1,2]. Later
this approach was extensively used in the framework of string theory by many people (see,
e.g., [38], and [9,10] for a review). Recently, the idea that the spacetime coordinates do
not commute has drawn much attention (see [11] and a list of references therein). On the
one hand, scalar noncommutative field theories were studied in [1220]. On the other hand,
it was realized that noncommutative geometry naturally appears in the framework of open
string theory in the presence of a constant B-field. The purpose of this paper is to show
that tree and one-loop diagrams of the noncommutative 3 theory can be also derived from
string amplitudes.
1 Permanent address: Landau Institute, Moscow, Russia. E-mail: andreev@physik.hu-berlin.de
2 E-mail: dorn@physik.hu-berlin.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 2 6 - 6
146
O. Andreev, H. Dorn / Nuclear Physics B 583 (2000) 145158
Before starting our discussion of one-loop diagrams of the noncommutative 3 theory,

we will make a detour and discuss open strings in the presence of a constant B-field at the
tree level (see, e.g., [11] and references therein).
In this case the world-sheet action is given by
Z

1
d2 z gij a Xi a Xj 2i 0 Bij ab a Xi b Xj + ,
(1.1)
S=
0
4
D
where D means the string world-sheet, namely a disk, gij , Bij , are the constant metric,
antisymmetric tensor and dilaton fields, respectively, Xi map the world-sheet to the target
space (Dp-brane) and i, j = 1, . . . , d = p + 1. The world-sheet indices are denoted by
a, b. The disk can be, of course, conformally mapped to the upper-half plane, i.e., the
region Im w > 0 of the complex-plane points w.
To analyze open string theory defined by the world-sheet action (1.1), one first has to
determine the propagator. To do so, it is necessary to define the boundary conditions. They
are: 3
g( )G(w,
w0 ) + 2 0 B( + )G(w,
w0 ) = 0 for Im w = 0.
(1.2)
Here Gij (w, w0 ) = hXi (w)Xj (w0 )i and = /w, = /w.

Evaluated at boundary points, the propagator with these boundary conditions is [11,21,
22]:
i
G(s, s 0 ) = 2 0 G1 ln |s s 0 | + (s s 0 ),
2
where
(1.3)
G = (g 2 0 B)g 1 (g + 2 0 B),
= (2 0 )2 (g + 2 0 B)1 B(g 2 0 B)1 ,
(1.4)
s = Re w and (s) is the step function that is 1 or 1 for positive or negative s. Following
[11], we will refer to g, B as closed string parameters (variables) and G, as open string
parameters. There is an interesting point that we should mention about the tree level. The
propagator between boundary points depends only on the open string parameters.
Let us now define the open string tachyonic vertex operator 4
Z
V (k) = ds eikX ,
(1.5)
where A B Ai B i . A simple analysis shows that the vertex operator (its integrand) is a
primary field of conformal dimension one as long as 0 kG1 k = 1.
For M open string tachyons, the tree amplitude is given by

1
AM = A(k1 , . . . , kM ) = N 0 0 GM
s Tr(1 . . . M ) Vd V (k1 ) . . . V (kM )
+ noncyclic permutations,
3 For the sake of simplicity, we use the matrix notations here and below.
4 We will give some motivations for such a definition in the next section.
(1.6)
147
R
d
where h . . . i = DX0 eS . . . , = d2
4 M 2 . We split the integral into the integral over
the zero mode X and the integral over nonzero modes. We use the following measure for
the integral over the zero mode: 5
Z
Vd = d p+1 X G.
(1.7)
This assumes that the dilaton field is redefined as [11]

= + 12 ln det G(g + 2 0 B)1 .
(1.8)
However, we have defined the open string coupling as G2s = e rather than Gs = e as it
was done in [11].
Moreover, by dividing the invariant measure of the Mbius group in 1.6, three vertex
operators are fixed to arbitrary positions on the boundary. Each vertex operator is related
to a factor Gs together with the ChanPaton degrees of freedom. Thus the amplitude has
the appropriate trace factor. N0 is a normalization constant (see, e.g., [8]).
In fact, there are two possibilities in taking the limit 0 0 [11]. The first is to do so
while keeping the closed string parameters g and B fixed. As a result, in this case one
expects the ordinary 3 theory. This is the standard field theory limit. The second one is
to keep the open string theory parameters G and fixed. 6 The expected result now is
the noncommutative 3 theory. This is the noncommutative field theory limit or Seiberg
Witten limit. Since we are interested in the noncommutative field theory we will mainly
discuss the noncommutative field theory limit.
The tree-tachyon amplitudes already show that the noncommutative field theory limit
of string amplitudes corresponds to tree-diagrams of the noncommutative 3 theory with
colour indices:
Z

b
(1.9)
S = Tr dp+1 X det G 12 i i + 12 m2 2 + 16 g ,
where the coupling constant g depends on the open string coupling and string parameter as
Gs ( 0 )(d6)/4. The -product is defined by
i ij

(1.10)
f (X) (X) = e 2 y i zj f (X + y)(X + z)y=z=0 .
2. Open string one-loop amplitudes in the presence of constant B-field
2.1. General analysis
The world-sheet action is now given by
Z

1
d2 w gij a Xi a Xj 2i 0 Bij ab a Xi b Xj ,
S=
0
4
C2
5 The zero mode integration includes the X-dependence in h i and gives a factor (P k ).
i
6 Both limits assume that the corresponding tachyon mass is treated as a free, but fixed parameter.
(2.1)
148
where C2 denotes the string world-sheet for the one-loop orientable open string, i.e.,
a cylinder (annulus). Note that there is no a constant dilaton field as the Euler characteristic
of the cylinder is zero. We describe C2 as the region
0 6 Re w 6 1,
w w + 2i
on the complex plane whose metric is ds 2 = dw dw. A flat annulus with inner radius a and
outer radius b can be obtained from the cylinder by
z = a exp(w ln q),
where the modular parameter of the annulus is given by q = a/b = exp( ).
To analyze open string theory defined by the world-sheet action (2.1), we take a slight
modification of the propagator found in [21,22]. So, we define the boundary conditions as
( 0
for x = 0,
(2.2)
g G(w, w0 ) 2 0 iB G(w, w0 ) = 2 0
x
y
2 for x = 1.
Here w = x + iy. The propagator with these boundary conditions is then
1 0
1
0
G(w, w0 ) = 0 g 1 lnq 2 (w w) q 2 (ww )
2 0 G1
0 g 1
X

0
0
ln 1 q 2n2+w+w 1 q 2nww
n=1

0
0
ln 1 q 2n+w w 1 q 2n+ww
n=1

0 1
0 1
1 q 2n2+w+w 1 q 2nww
1 X
0
0
ln 1 q 2n2+w+w 1 q 2nww
2
n=1
1 q 2n2+w+w
1
1 q 2nww
1
(2.3)
Since we are interested in open string vertex operators that are inserted on the boundaries
of the cylinder, we need to restrict the above propagator to the boundaries to get the relevant
propagator. Evaluated at boundary points, it is

1
1
|y y 0 | i
,
/D( )
for x 6= x 0 , (2.4)
G(y, y 0 ) = 0 g 1 ln q 2 0 G1 ln q 4 4
2
2

1
|y y 0 | i
,
/D( )
for x = x 0 , (2.5)
G(y, y 0 ) = i (y y 0 ) 2 0 G1 ln 1
2
2
where correspond to x = 1 and x = 0, respectively, 1 and 4 are the Jacobi theta

functions. We have also added constants to the propagators. They will be set to a convenient
149
value in the next section. The function excludes the zero mode contribution. Explicitly,
it is given by

1 q iy
i
y
(2.6)
= (y) .
(y) = ln
1 q iy
In fact, what we have found above considerably differs from what we had in Section 1
where the theory (its open string sector) is completely described in terms of the open string
parameters. In our present discussion, we have obtained that the closed string metric g no
longer decouples. This fact has important consequences that we will discuss in the next
section.
One basic question we should ask now is how to determine the modular measure [d ]B
of open bosonic string with a constant B-field in arbitrary dimension. For this it is crucial
that the measure factorises as f (B)[d ]0 , where the B-dependent factor is given by

f (B) = det 1 + 2 0 g 1 B .
It was found by direct calculation in [21]. From Eq. (1.4) we get det gf (B) = det G
(see also [23]). Moreover, it was suggested in [24] that the original theory whose action is
given by (1.1) or (2.1) can be also described by a simpler action
Z
1
d2 z Gij a Xi a Xj ,
(2.7)
S=
4 0
C2
while correlation functions of the vertex operators include the build in star products
( -dependence). It is easy to see that it indeed works at the tree-level where the ansatz
is simply
hV1 V2 VN i.
(2.8)
). 7 8
After
In fact, it follows because the corresponding ChanPaton factor is Tr(1 . . . N
this is understood, it becomes clear that the -dependence is more involved on higher loop
levels where there are products of traces. For example, in the case of interest the Chan
Paton factor is Tr(1 . . . N ) Tr(N+1 . . . M ). For N = 0 or M N = 0 that corresponds
to planar diagrams, the ansatz reduces to the above one. This also follows from the
corresponding propagator (2.5). It is clear that for N 6= 0, M N 6= 0 that corresponds
to nonplanar diagrams, the ansatz (2.8) has to be modified. In this case the propagators
(2.4)(2.5) say us what to do. However, the measure is completely defined by the action
(2.7). 9
Thus the problem reduced to the old one namely, how to extend the modular measure of
open string to arbitrary dimension. This has been much studied to compute perturbative
7 A motivation for this analogy is due to matrix (reduced) models where a map from a matrix to a function
A a(x) leads to AB a(x) b(x) (see, e.g., [25] and references therein).
8 It is well known that string theory imposes restrictions on possible gauge groups introduced via the Chan
Paton method (see, e.g., [28] and references therein). In the problem at hand the B-field may lead to a potential
clash with the ChanPaton method. So the question arises whether such a method is still consistent. We will show
in the appendix that this is the case. However there exists the only allowed gauge group, namely, U (N ).
9 In a general case one also has to add a constant dilaton field as ,
here is the Euler characteristic.
150
Fig. 1. The ordering of open string vertex operators on the cylinder.
field theory amplitudes via string theory in the 0 0 limit (see, e.g., [9,10]). In
making our further analysis, we will adopt the proposal in [26] according to which the
interpretation of each factor is transparent. The measure is thus
Z
Z
[d ]0 =
2d
d d
2 (i )
,
(2.9)
where is the Dedekind eta function.

The question that immediately arises is whether this measure could be obtained directly
from a calculation, for instance, along the lines of [26]. At the tree-level, the commutation
relations for the modes of Xi that follow from the action (2.7) are as usual, namely:
i j
i j
0 ij
X , p = 2i G ,
n , m = 2 0 nn+m.0 Gij ,
(2.10)
where pi 0i . Moreover, it turns out that the Virasoro generators do not depend on the
zero modes X. So, one can formally repeat the standard analysis to build the physical states
via states in the Fock space. Moreover, it also means that the tachyonic vertex operator is
given by (1.5). d 2 in (2.9) assumes that the reparametrization ghosts are included.
The scattering amplitudes can be defined in two ways. One can modify the ansatz (2.8)
to do it consistent with the dependence that follows from the propagators or it can
be done by
AN.M = A(k1 , . . . , kN ; kN+1 , . . . , kM )
= N1 ( 0 ) GM
s Tr(1 . . . N ) Tr(N+1 . . . M ) Vd V (k1 ) . . . V (kM )
Z
Z
+ noncyclic permutations, h . . . i = [d ]0 DX0 eS .
(2.11)
Here N vertex operators are attached to one boundary and M N vertex operators to the
other one as in Fig. 1. N1 is a proper normalization constant (see [8]). The correlation
functions of exponential operators are simply computed by using the explicit form of the
propagator.
So, the amplitudes are:
AN.M = N1 (2)d ( 0 ) GM
s Tr(1 . . . N ) Tr(N+1 . . . M )

M
X
! Z
ki
i=1
M
Y
i
N
Y
dyi
0
2d 1 0 kg 1 k
d d
2 (i )
q2
yi1
Z
12 i (yij )ki kj
"
e
1
4
i=1 j =N+1
i<j
M
Y
"
M
N
Y
Y
"
e
151
1
2 i (yij )ki kj
0
1
|yij | i #2 ki G kj
2 ,
D( )
0
1
|yij | i #2 ki G kj
2 ,
D( )
1
N+1
i<j
0
1
|yij | i #2 ki G kj
2 ,
D( )
(2.12)
PN
where yij = yi yj and k = i=1 ki .
To complete the story, perhaps we should point out that the P-dependence of planar
1
diagrams is very simple. It reduces to an overall phase factor e 2 i

as for planar diagrams of noncommutative field theories [12].
N
i<j ki kj
, i.e., the same
2.2. Noncommutative field theory limit

As discussed in the previous section, the noncommutative field theory limit is defined by
0 0 at fixed m, G and (i.e. g 1 = G/(2 0 )2 via Eq. (1.4)). On the other hand,
it is well known that only the neighbourhood of q = 1 or, equivalently , contributes
to the field theory limit (see, e.g., [9,10]). The restriction to the edge of the moduli
space remains true also in the noncommutative field theory limit since finite values of
0 1
are even more suppressed due to the factor q kg k/2 in Eq. (2.12).
The noncommutative field theory limit is also simply carried out by sending both and
y to infinity while keeping variables 10
t = 2 0 ,
i = yi /2
(2.13)
finite.
In this limit, the propagators become: 11

1
G + G1 t ( 0 )2 | 0 | for x 6= x 0 ,
(2.14)
4t

1
(2.15)
G(, 0 ) = i ( 0 ) + G1 t ( 0 )2 | 0 | for x = x 0 .
2
At this point a couple of comments are in order.
(y) = 2 (y) to the delta

(i) The function defined by Eq. (2.6) is related via y
function that excludes the zero mode, i.e., (y) = (y) 1/2 . Naively, the
G(, 0 ) =
10 We disregard the pinching configurations that result in the one-particle reducible diagrams.
11 We have set D( ) = 1 [(i/ )]3 .
152
difference between these functions disappears in the limit. In fact, this

is not the case because the last term survives as far as the rescaling (2.13) is taken
into account. Thus
() = () 2.
(2.16)
(ii) The difference between and indeed disappears for all planar amplitudes as well
as tree level amplitudes. In this case the zero mode contribution to the amplitudes
P
P
pi = 0.
i<j pi pj ij vanishes due to the momentum conservation
Thus, in the field theory limit the amplitudes (2.12) become:
AN.M = N10 gM Tr(1 . . . N ) Tr(N+1 . . . M )
M
X
!
ki
i=1
M Zi1
Y
i
N
Y
2i ki kj
i<j
di
M
Y
M
Y
N+1
i<j
dt M d m2 t kk/t
2 e
t
t
et (|ij |ij )ki G
i<j
i
2 ki kj
1 k
j
N
Y
eiij ki kj
i<j
M
Y
eiij ki kj
N+1
i<j
(2.17)
where k k = 14 k G k. To get this form we introduced the mass for the open string
tachyon as m2 = (2 d)/24 0 and used the relation between the open string coupling
constant and the 3 theory coupling constant.
We will conclude this subsection with a couple of examples to illustrate the use of
the string amplitudes in practical calculations of one loop Feynman diagrams of the
noncommutative 3 theory.
Example 1. As a warmup, let us consider the simplest nonplanar amplitude and, as byproduct, reproduce a result of [17,18]. The amplitude (modulo a normalization constant) is
given by
Z
A1.2 = g Tr(1 ) Tr(2 ) (k1 + k2 )
2
dt t 1 2 em
d
2 t k k /t
1 1
Z1
d1 et (1 1 )k1 G
2
1 k
1
(2.18)
Here we fix the translational invariance on the cylinder by setting the second vertex
operator at the origin, i.e., 2 = 0.
The next step in finding the correspondence with the field theory diagram is to introduce
new integration variables which correspond to the Schwinger parameters
1 = t1 ,
2 = t (1 1 ).
(2.19)
153
Fig. 2. The simplest nonplanar diagram of the noncommutative 3 theory.
In terms of these variables, the amplitude is written as

A1.2 = g2 Tr(1 ) Tr(2 ) (k1 + k2 )
ZZ
k1 k1
1 2
1
d1 d2
m2 (1 +2 ) (1 +2 ) (1 +2 ) k1 G k1
e
e
e
.
d
(1 + 2 ) 2
(2.20)
0 0
It is now clear that what we have found is exactly the simplest nonplanar Feynman
diagram of the noncommutative 3 theory as shown in Fig. 2.
Before we give the next example, we first want to make one important remark. The
point is that the second exponent, in fact, acts as an effective regulator of the otherwise UV
divergent diagram. This interesting observation was made in [1618] by original analysis
of perturbation expansions of noncommutative theories. In [17,18] it was also suggested
that the effect is due to the closed string metric and a stringy interpretation was given. It is
based on the known fact that for B = 0 the UV limit ( 0) of the open string channel
corresponds to the IR limit for the closed string channel. What we found starting directly
from open string theory with the B-field is a little bit different. Indeed, the effect is due to
the closed string metric but it has nothing common with the 0 limit for open string.
As we have seen the crucial point is that the closed string metric g does not decouple
from the propagator between boundary points in the noncommutative field theory limit,
i.e., ! So, in the noncommutative field theory limit there is effectively a signal of
closed string sector.
Example 2. The next example we would like to consider is the A2.4 string amplitude. Here
we fix the translational invariance on the cylinder by setting the fourth vertex operator at the
origin, i.e., 4 = 0. We now can replace the multiple integral by a sum of ordered multiple
integrals. This is clear just by substituting the expansion of unity that for the problem at
hand (for ordered 1 and 2 ) is 1 = H (23) + H (13 )H (32 ) + H (31 ). 12 Now we can
proceed along the lines of [9,10], i.e., we rewrite the ordered integrals via the standard
integrals over the Schwinger parameters. Let us explicitly illustrate how it works for the
first term where the ordering is 1 > 2 > 3 . In this case the corresponding contribution
(modulo a normalization constant) is given by
12 H means the Heaviside step function.
154
Fig. 3. Three nonplanar field theory diagrams that correspond to the A2.4 string amplitude.
A2.4 = g Tr(1 2 ) Tr(3 4 )

4
4
X
!
ki e
2i k1 k2 + 2i k3 k4
i=1
Z1
Z1
d1
3
Y
Z2
d2
dt t 3 2 em
d
2 t kk/t
d3 ei12 k1 k2 i3 k3 k4
et (i i )ki G
2
1 k
i
i=1
3
Y
e2t (i j j )ki G
1 k
j
(2.21)
i<j
The Schwinger parameters can be introduced as

1 = t12 ,
2 = t23 ,
3 = t3 ,
4 = t (1 1 ).
(2.22)
So, the expression (2.21) becomes

A2.4 = g Tr(1 2 ) Tr(3 4 )
4
4
X
!
ki e 2 k1 k2 + 2 k3 k4
i
i=1
4 Z
Y
di 2 em e kk e (1 k1 k2 3 k3 k4 )
2
i=1 0
1
e (1 4 k1 G
e (1 3 (k2 +k3 )G
1
P4
1 k k G1 k k G1 k k G1 k )
1
1 2 2
2
2 3 3
3
3 4 4
4
1 (k +k ) kG1 k)
2
3
2 4
(2.23)
where = i=1 i . It is straightforward to get the Schwinger representation for the other
terms. As a result, the amplitude A2.4 reduces to a sum of three nonplanar Feynman
diagrams of the noncommutative 3 theory as shown in Fig. 3.
2.3. 0 limit
Now we consider another limit: 0 while keeping all other parameters fixed. It is
well known that for B = 0 this is the open string UV limit. Moreover, it is interpreted as a
long-distance effect because the leading asymptotics are given by the closed string tachyon
as well as the lightest closed string states. So, our purpose is to analyze what happens in
the presence of the B-field.
155
In this limit, the propagators (2.4)(2.5) become:

0 1
g 2 0 G1 ln for x 6= x 0 ,
2
G(y, y 0 ) = 12 i (y y 0 ) + 2 0 G1 P (y y 0 ) 2 0 G1 ln
G(y, y 0 ) =
(2.24)
for x = x 0 .
(2.25)
Here P (y) = n=1 n1 cos n

y. As in the case that we discussed first, it is useful
to define new variables
t=
0
,
2
i = yi /2.
(2.26)
Then we get the amplitudes (modulo a normalization constant)

!"
M
X
0 1
k i GN
AN.M = ( )
s Tr(1 . . .N )
i=1
1
kg 1 k + m2
N1
Y
i1
Z
di
i=1 0
"
N
Y
#
e
12 i (ij )ki kj 2 0 ki G1 kj P (ij )
N =0
i<j
Tr(N+1 . . .M )
GMN
s
M1
Y
#
i1
Z
M
Y
1
0 k G1 k P ( )
i
(
)k
k
2
i
j
ij
di
e 2 ij i j e
i=N+1 0
+ noncyclic permutations.
N+1
i<j
M =0
(2.27)
The translational invariance is now fixed by setting N = M = 0. We also introduce the

mass for the closed string tachyon as m2 = (2 d)/6 0 .
What we see from the above is that in the 0 limit for B 6= 0 the amplitudes factorize
as in the case B = 0. There is no new effect here. So, the interpretation is the standard one
as a long-distance effect with the asymptotics due to the closed string states.
3. Concluding comments
What we have learned is that in the noncommutative field theory limit of open string
theory with the B-field at the one loop level there is a signal of closed string sector.
To be more precise, we found that appears not only via the -product as it does at
the tree level but via 14 G that corresponds to the closed string metric g. A recent
analysis of noncommutative field theories assumes that some closed string modes already
appear there [20]. We do not see this explicitly in our analysis of the noncommutative
field theory limit where we found only the closed string parameters rather than the modes.
156
However, to be cautious, we should mention that we did not discuss singularities and their
regularization.
1 0
1
It is important to emphasize that the factor q 2 kg k , or equivalently the g-dependence
of amplitudes, which is crucial for the noncommutative field theory limit of nonplanar
diagrams is universal. It does not depend on the kind of vertex operators. This is clear
because the effect is due to the first term in the propagator (2.4). It is a constant, so
contributions come only from exponents.
From the physical point of view it is more interesting to consider the noncommutative
gauge theory. In fact, at the one-loop level it can be done along the lines of the present
paper by considering the vertex operator for a gauge field
Z
(3.1)
V (, k) = ds s X eikX
instead of the tachyon operator (1.5). Then, the corresponding amplitudes are computed by
using the propagators (2.4)(2.5) within the point splitting renormalization scheme [11].
We hope to return to this important problem in the near future [27].
Acknowledgements
We would like to thank A.A. Tseytlin for a collaboration at an initial stage, and also for
useful comments and reading the manuscript. The work of O.A. is supported in part by the
Alexander von Humboldt Foundation and by Russian Basic Research Foundation under
Grant No. 9901-01169. The work of H.D. is supported in part by DFG.
Appendix A
The purpose of this appendix is to show how restrictions on gauge groups arise within
open string theory in the presence of the B-field. In general, there are a variety of ways to do
this. Here we will focus on the classical recipe [28] which is based on simple factorization
properties of string amplitudes. So, let us take the tree amplitude (1.6) and look at its
factorization AM AP AMP . It is well known that the only novelty due to the B-field is
a phase factor
P1M =
M
Y
e 2 ki kj
i
(A.1)
1
i<j
that appears in the expression for the amplitude. The crucial fact for what follows is that
P1M due to momentum conservation obeys the factorization relation
P1M = P1P PPM+1 .
(A.2)
With above formula the generalization of the classical analysis is straightforward. As a

result, Eq. (6.1.11) of [28] becomes

1
Tr 1 . . . P P1P ()P P . . . 1 P1P
P +1 . . . M PPM+1 ()MP M . . . P +1 PPM+1
X
1
Tr 1 . . . P P1P ()P P . . . 1 P1P
157
1
1

.
Tr T P +1 . . . M PPM+1 ()MP M . . . P +1 PPM+1
This equation is satisfied if a matrix
= 1 . . . P P1P ()P P . . . 1 P1P
1
(A.3)
(A.4)
belongs to the algebra of the matrices i . It is known that in the case of = 0 the allowed
gauge groups are U (n), SO(n) and USp(2n). To see what survives for nonzero let us
specialize to P = 2. Then the following direct algebra:

i
i
(A.5)
= 1 2 e 2 k1 k2 2 1 e 2 k1 k2 = for i u(n),

i
i
(A.6)
T = 1 2 e 2 k1 k2 2 1 e 2 k1 k2 6= for i so(n),

i
i
T = s 1 2 1 e 2 k1 k2 1 2 e 2 k1 k2 s 6= s 1 s for i usp(n),
where Ti = s 1 i s,
(A.7)
shows that the only surviver is U (n). The latter is in harmony with the result obtained
within noncommutative gauge theory claiming that the noncommutative gauge transformation is consistent only for the unitary group U (n) (see, e.g., [29]).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
J. Scherk, Nucl. Phys. B 31 (1971) 222.

V. Alessandrini, D. Amati, M. Le Bellac, D. Olive, Phys. Rep. 1 (1971) 269.
R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 298 (1988) 109.
J.A. Minahan, Nucl. Phys. B 298 (1988) 36.
T.R. Taylor, G. Veneziano, Phys. Lett. B 212 (1988) 147.
Z. Bern, D.A. Kosover, Phys. Rev. D 38 (1988) 1888; Nucl. Phys. B 321 (1989) 605.
V.S. Kaplunovsky, Nucl. Phys. B 307 (1988) 145; Nucl. Phys. B 382 (1992) 436.
P. Di Vecchia, L. Magnea, A. Lerda, R. Marotta, R. Russo, Phys. Lett. B 388 (1996) 65.
P. Di Vecchia, L. Magnea, A. Lerda, R. Russo, R. Marotta, Renormalization constants from
string theory, in: Erice, 1995, hep-th/9602055.
P. Di Vecchia, L. Magnea, A. Lerda, R. Marotta, R. Russo, The field theory limit of multiloop
string amplitudes, in: Workshop on Gauge Theories, Applied Supersymmetry and Quantum
Gravity, London, 1996, hep-th/9611123.
N. Seiberg, E. Witten, JHEP 9909 (1999) 032.
T. Filk, Phys. Lett. B 376 (1996) 53.
J.C. Varilly, J.M. Gracia-Bondia, Int. J. Mod. Phys. A 14 (1999) 1305.
T. Krajewski, R. Wulkenhaar, Perturbative quantum gauge fields on the noncommutative torus,
Report No. CPT-99-P-3794, hep-th/9903187.
S. Cho, R. Hinterding, J. Madore, H. Steinacker, Finite field theory on noncommutative
geometries, Report No. LMU-TPW-99-06, hep-th/9903239.
I. Chepelev, R. Roiban, Renormalization of quantum field theories on noncommutative R d ,
I. Scalars, Report No. ITP-SB-99-61, hep-th/9911098.
158
[17] S. Minwalla, M. Van Raamsdonk, N. Seiberg, JHEP 02 (2000) 020.

[18] N.Seiberg, The IR/UV connection, in: String Theory at the Millennium, Caltech, 2000, http://
www.theory.caltech.edu/conf2000.
[19] I.Y. Arefeva, D.M. Belov, A.S. Koshelev, A note on UV/IR for noncommutative complex scalar
field, Report No. SMI-5-00, hep-th/0001215.
[20] M. Van Raamsdonk, N. Seiberg, Comments on noncommutative perturbative dynamics, Report
No. PUPT-1920, hep-th/0002186.
[21] A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 (1987) 599.
[22] C.G. Callan, C. Lovelace, C.N. Nappi, S.A. Yost, Nucl. Phys. B 288 (1987) 525.
[23] O. Andreev, Phys. Lett. B 481 (2000) 125.
[24] O. Andreev, H. Dorn, Phys. Lett. B 476 (2000) 402.
[25] N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Wilson loops in noncommutative YangMills,
Report No. KEK-TH-649, hep-th/9910004.
[26] J. Polchinski, String Theory, Vol. I, Cambridge University Press, 1998, p. 223.
[27] O. Andreev, H. Dorn, work in progress.
[28] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory: Introduction, Cambridge University
Press, 1988, p. 297.
[29] J. Madore, S. Schraml, P. Schupp, J. Wess, Gauge theory on noncommutative spaces, Report
No. LMU-TPW 2000-05, hep-th/0001203.

String scale in noncommutative YangMills

Nobuyuki Ishibashi a,1 , Satoshi Iso a,2 , Hikaru Kawai b,3 ,
Yoshihisa Kitazawa a,4
a Laboratory for Particle and Nuclear Physics, High Energy Accelerator Research Organization (KEK),
Tsukuba, Ibaraki 305-0801, Japan

b Department of Physics, Kyoto University, Kyoto 606-8502, Japan
Received 14 April 2000; revised 12 May 2000; accepted 19 May 2000
Abstract
We identify the effective string scale of noncommutative YangMills theory (NCYM) with the
noncommutativity scale through its dual supergravity description. We argue that Newtons force law
may be obtained with 4 dimensional NCYM with maximal SUSY. It provides a nonperturbative
compactification mechanism of IIB matrix model. We can associate NCYM with the von Neumann
lattice by the bi-local representation. We argue that it is superstring theory on the von Neumann
lattice. We show that our identification of its effective string scale is consistent with exact T-duality
(Morita equivalence) of NCYM. 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
In recent studies of nonperturbative aspects of superstring theory, type IIB superstring is
found to provide the simplest setting [14]. However, it is difficult to obtain a realistic
unified theory in IIB superstring at least perturbatively. Therefore we may expect that
an entirely new type of nonperturbative compactification mechanism of IIB superstring
exits [5,6]. On the other hand, a new compactification mechanism which involves branes
has been proposed [7]. Since branes naturally appear in superstring theory [8], such a
mechanism is expected to apply for IIB superstring theory.
Noncommutative YangMills theory (NCYM) has been obtained by compactifying IIB
matrix model on noncommutative tori [9]. We can simply obtain d dimensional NCYM by
expanding IIB matrix model around d dimensional noncommuting backgrounds [10]. In
IIB matrix model, the dynamical variables are the Hermitian matrices which are interpreted
1 ishibash@post.kek.jp
2 satoshi.iso@kek.jp
3 hkawai@gauge.scphys.kyoto-u.ac.jp
4 kitazawa@post.kek.jp
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 8 - 4
160
N. Ishibashi et al. / Nuclear Physics B 583 (2000) 159181
as the spacetime coordinates. A d dimensional noncommuting background corresponds

to a d dimensional noncommutative spacetime. The simplest idea for compactification in
IIB matrix model is to postulate that the compactification down to d dimensions is realized
by expanding the model around d dimensional backgrounds.
In this paper we point out that Newtons force law may be obtained with four
dimensional NCYM with maximal SUSY (NCYM 4 ). Our argument is based on its dual
supergravity description [11,12]. We argue that there exists a massless bound state in the
effective Hamiltonian of supergravity which gives rise to Newtons force law a la Randall
and Sundrum. Therefore NCYM 4 may be regarded as a four dimensional compactification
of IIB superstring. The remarkable feature is that it compactifies ten dimensional
superstring straight down to four dimensions. The compactification of matrix models has
been the outstanding problem [13,14]. We argue that we can obtain four dimensional gauge
theory and gravitation with four dimensional noncommutative backgrounds in IIB matrix
model. In this sense, we have identified the most satisfactory compactification mechanism
of matrix models. It is also possible to obtain d + 1 dimensional NCYM theories by
expanding BFSS matrix model around d dimensional noncommutative backgrounds in
an analogous way. It may be interesting to investigate such theories through supergravity
approach. However, it is beyond the scope of this paper.
In the large N expansion of gauge theory, Feynman diagrams can be classified with their
world sheet topology. This is a generic feature of matrix valued field theory. On the other
hand, string theory is perturbatively defined in terms of field theory on the world sheet.
String theory may be nonperturbatively formulated in the large N limit of matrix models
in view of these remarkable correspondences. IIB matrix model is such a proposal which
is a large N reduced model of maximally supersymmetric gauge theory [1]:

1
1
1
(1.1)
S = 2 Tr [A , A ][A , A ] + [A , ] .
g
4
2
Here is a ten dimensional MajoranaWeyl spinor field, and A and are N N
Hermitian matrices.
With vanishing fermionic backgrounds, the equations of motion are:

(1.2)
A , [A , A ] = 0.
The following solutions correspond to BPS-saturated backgrounds:
[A , A ] = c-number C .
(1.3)
Since we interpret A as spacetime coordinates due to N = 2 SUSY, we expect to obtain

d dimensional spacetime with d dimensional solutions of this type. We further expect to
obtain d dimensional gauge theory. Since matrices form noncommutative but associative
algebra, we expect a deep connection to noncommutative geometry [15]. In fact we have
obtained NCYM of 16 supercharges with these backgrounds [10]. Ordinary gauge theory
appears as the low energy effective theory. Since short open strings correspond to gauge
particles, we indeed find another evidence that IIB matrix model can describe infinite
numbers of fundamental strings. We have further pointed out that NCYM contains nonlocal
degrees of freedom which may be interpreted as long open strings [16,17]. We have indeed
161
shown that they give rise to gravitational interactions at the one loop level as it is expected
in superstring theory [10,17].
Since NCYM seems to contain the both gauge theory and gravitation, it is very likely
that it is equivalent to superstring theory in a particular background. The major issue here
is the renormalizability of NCYM [38,39]. We have shown that the high energy behavior of
NCYM is equivalent to large N gauge theory by using the bi-local field representation [17].
Although it also exhibits long range interactions which we interpret as gravitation, it is very
likely that NCYM exits at least for d 6 4.
NCYM is often argued to be the low energy limit of string theory with constant b field
[1823]. However, long range interactions are found due to the presence of long open
strings which might signal the presence of closed strings [10,17,3943]. These issues
are currently under active investigations [4953]. In this paper we propose that NCYM
0 is set
is superstring theory on the von Neumann lattice whose effective string scale eff
by C .
The organization of this paper is as follows. In Section 2, we argue that Newtons
force law is obtained with NCYM 4 . Since it is a nonperturbative problem, we study
its dual supergravity description. In Section 3, we briefly summarize our formulation
of NCYM as twisted reduced models. In Section 4, we estimate the string tension of
NCYM using the formalism of Section 3. We find that our estimate is consistent with
string theoretic expectations in Section 2. In Section 5, we investigate the graviton
exchange process by the one loop perturbation theory. We conclude in Section 6 with
discussions.
2. NCYM4 as a unified theory

In this section, we argue that NCYM 4 contains four dimensional gauge theory
and gravitation. It is clear that NCYM contains ordinary gauge theory since the
noncommutative phases become ineffective at tree level in the low energy limit. The
remarkable possibility is that it may also contain gravitation. We first observe the long
range interaction at the one loop level which is specific in NCYM. It can be interpreted as
gravitational interaction in IIB superstring as it is explained in Section 5. In string theory,
closed string exchanges should be visible at open string one loop level. Therefore this
phenomenon is another stringy feature of NCYM. Although we can see the glimpse of
closed strings at the one loop level, we need to understand the quantum effects to all orders
to investigate the gravitational sector of NCYM 4 .
In order to study such a problem, we recall the supergravity solution of m coincident
D3-branes with the constant NS B field strength b [12]:
(1 + g m 02 (1 + 02 b2 )/r 4 )
,
(1 + g m 02 /r 4 )

1 2
1
g m 02 (1 + 02 b2 ) 1/2
dE
x2
2
2
2
ds = 0 1 +
+ dr + r d5 ,
0
r4
1 + g m 02 /r 4
e = g
162
b
b
dx dy +
dz d,
02
4
(1 + g m /r )
(1 + g m 02 /r 4 )
1
B2 ,
C2 = i
g (1 + 02 b2 )
1
b2
,
C0 = i
02
2
g (1 + b ) (1 + g m(1 + 02 b2 )/r 4 )
1
m
.
F0123r = 4i
(1 + g m 02 /r 4 )2 r 5
B2 =
(2.1)
Here g is the dilaton expectation value at r = . xE denotes four dimensional spacetime

coordinates in this section.
These background fields appear in the Euclidean IIB supergravity action:
SIIB = SNS + SR + SCS ,

Z
1
1 2
10 2
d x ge
SNS =
R + 4 H3 ,
2
2

Z
1
1
e32 + F
e2 ,
d10x g F12 + F
SR =
4
2 5
Z
1
C4 H3 F3 ,
SCS =
4
(2.2)
where
e3 = F3 C0 H3 ,
F
e5 = F5 1 C2 H3 + 1 B2 F3 .
(2.3)
F
2
2
We identify the r dependent metric g in Eq. (2.1) as the four dimensional metric for
fundamental strings:

g m 02 (1 + 02 b2) 1/2
1
(2.4)
.
g = 1 +
r4
1 + g m 02 /r 4
We postulate that D3-branes are located at the maximum of g , namely, at the boundary
r = (g m 02 )1/4 . Since open strings live on the D3-branes, we identify the open string
metric G with g at the boundary as
1/2
.
(2.5)
G 1 + 02 b2
Eq. (2.1) indicates that fundamental string metric grows at smaller r. This phenomenon
may be interpreted that closed strings become dynamical due to the quantum effects in
NCYM. The graviton exchanges we find at the one loop level in Section 5 support such an
interpretation. We consider the case that the noncommutativity scale lNC is much smaller
than the string scale ls . Let us focus on the physics at the noncommutativity scale by letting
0 but keeping b, r O(1). Although it might appear to be a strange limit, it is
equivalent to consider the standard 0 1/2 , x 1/2 x , b 1 limit. The remarkable
point is that open string metric is given by Eq. (2.5) as 0 b in this limit.
The Polyakov action for fundamental strings becomes in such a limit as
1
0
163
+ d2 z b 0 x 1 x +
d2 z G x x
Z
Z

b d2 z 0 x 0 x 1 x 1 x + d2 z b 0 x 1 x + .
So the Hamiltonian for open strings behaves like

X
knk ,
cp2 +
(2.6)
(2.7)
k6=0
where c = 1/b, and nk denote the number operators of the oscillator modes. Here we find
that the noncommutativity scale now acts as the effective string scale!
Supergravity description of U (m) NCYM 4 may be obtained by considering large 0 and
small g limit while keeping = g 02 b2 fixed [12]:
2
m
1
,
e =
4
r
(1 + m/r 4 )

1 2
m 1/2
dE
x2
2
2
2
ds =
+ dr + r d5 ,
0
r4
1 + m/r 4
1
1
dx dy +
dz d,
B2 =
(1 + m/r 4 )
(1 + m/r 4 )
1
r4
C2 = i B2 ,
C0 = i 2 ,
m
1
m
.
(2.8)
F0123r = 4i
4
2
(1 + m/r ) r 5
Here we have also put b = 1 which implies that the noncommutativity scale lNC is O(1).
Since we are looking at the vicinity of the D3-branes in this limit, we expect to find
massless open strings. However we also find oscillator modes since the effective string
scale is set by lNC as in Eq. (2.7).
We recall that there is a crossover at the noncommutativity scale lNC in NCYM.
When we consider the Wilson loops, we find that the planar diagrams dominate at larger
momentum scale than 1/ lNC and the diagrams of all topology contribute in the opposite
limit [31]. It may be interpreted that the string coupling (dilaton expectation value) is scale
dependent. It is because in our IIB matrix model conjecture, the tree level string theory is
considered to be obtained by summing planar diagrams and string perturbation theory is
identified with the topological expansion of the matrix model. With this interpretation, the
string coupling grows as the relevant momentum scale is decreased while it vanishes in the
opposite limit. In the D brane interpretation, the small momentum region corresponds to
the vicinity of the brane, while the large momentum region corresponds to the region far
from the brane since the Higgs expectation value plays the same role with the momentum
scale. In this sense e in Eq. (2.8) behaves just like the NCYM 4 [16,39].
The small r behavior of Eq. (2.8) is identical with ordinary AdS/CFT correspondence if
we identify as the coupling of ordinary YangMills theory [24]. This result is reasonable
since the low energy limit of NCYM 4 contains ordinary gauge theory with precisely the
164
same relation between the coupling constants. It is because in string theory the coupling of
NCYM 4 is given by = g 0 b2 when 0 is large.
We may now resort to the standard argument to justify the supergravity description as
follows. Since m sets the radius of AdS5 and S5 , supergravity description is valid in the
strong t Hooft coupling limit of NCYM 4 . The mass scale for the KaluzaKlein modes can
be estimated to be of order 1/(m)1/4 lNC . We need to consider large m limit also in order
to keep the dilaton expectation value to be small. As we have argued, the mass scale of the
oscillator modes is set by the effective string scale as 1/ lNC .
In order to investigate the gravitational interaction, we introduce external energy
momentum tensor T . As an explicit example, we may consider the photonphoton
scattering on the brane as in Section 5. Having such a case in mind, we assume that
the indices of the nonvanishing components of T run over four dimensional spacetime
coordinates. We also assume that it is traceless in the four dimensional subspace. There
is an ambiguity concerning its dilaton dependence. It may be natural to assume that it
contains the factor e from string theory point of view. However such an ambiguity does
not change the main conclusions in this section.
We may adopt the coordinate system where the five dimensional subspace (E
x , ) is
conformally flat

1 2
ds = (m)1/2 A()(dE
x 2 + d 2 ) + d52 .
(2.9)
0
Since
r
Z
m
(2.10)
= dr 1 + 4 ,
r
we find that
A() 1/ 2 ,
(2.11)
= (m)1/4 ).
It is illustrated in Fig. 1. Our strategy

It has the unique maximum at = 0 (r
is to expand the metric and equations of motion around the classical solution to the first
order of the fluctuation. In the following investigation, we use the formalism developed in
[24]. 5 We parametrize the metric as g (eh ) where g is the background metric and
h = 0 (traceless). The tensor indices are raised and lowered by the background metric.
This formalism explicitly separates h from the conformal mode of the metric.
The equation of the motion with respect to h is
1
(2.12)
R + 2 H3 H3 + = 2 T ,
2
where = e . We have suppressed the contributions from the R-R sector. The advantage
to consider the four dimensional traceless energy momentum tensor T is that all other
equations of motion are satisfied to the first order of h . In this sense it minimally excites
gravitons.
5 We use the sign conventions for the curvature tensors of Misner, Thorn and Wheeler in this paper while those
of t Hooft and Veltman [25] have been used in [24]. Although the signs of the Riemann tensor are the same,
those of the Ricci tensor and scalar curvature are the opposite in these conventions.
165
Fig. 1. Conformal factor A() is plotted as a function of . m is set to 1.
Eq. (2.12) is expanded to the first order of h as

1
1
1
h h , h,
2
2
2
1
1
h , H H h H H h + = 2 T .
2
4
(2.13)
In the coordinate system of Eq. (2.9), it is consistent to assume that the tensor indices of the
nonvanishing h resides in the four dimensional spacetime since the tensor indices of
T are also four dimensional. It is also consistent to assume that h = 0 since T = 0.
We also adopt the h , = 0 gauge.
Our strategy is to first study the following free equation of motion for h in the ten
dimensional curved spacetime:
1
h = 2 T .
2
(2.14)
Eq. (2.14) can be rewritten as:

H h = 2A 2T ,
A
H = g g .
g
(2.15)
The Hamiltonian is

2
3 A0
1
2
2
E
+ AL ,
H =
+ 2+
(m)1/2
2 A
(2.16)
E 2 and L 2 denote the Laplacians on R 4 and S 5 ,

where A0 = A/. The symbols
respectively. We can further simplify Eq. (2.15) by the similarity transformation as
166
eA3/4 h = 2 2 A7/4 T ,
H
e = A3/4 H A3/4
H

3 A02
1
2
3 A00
2
2
E
+
=
+ AL .
+ 2
(m)1/2
4 A
16 A2
(2.17)
e is
We concentrate on the S wave on S 5 in what follows. The eigenfunction of H
found to be exp(ikE xE)l with the eigenvalue (m)1/2 (kE2 + El ). Note that El acts as the
four dimensional mass of the various modes. It can be obtained by solving the following
quantum mechanics problem

3 A00
3 A02
2
= E .
(2.18)
2+
4 A
16 A2
Let us introduce a super-charge

Q=
3 A0
1 +
2 .
i
4A
(2.19)
Since the relevant Hamiltonian can be embedded in Q2 , we only need to solve Q = 0 to

find the zeromodes. The solution is
= e 4
d(A0 /A)
(2.20)
We conclude that there is a single zero energy bound state with the conformal factor A of
our type as follows
0 = A3/4 .
Such a zero mode corresponds to a massless field in four dimensions.
The propagator in this basis is
X
1
hx|ni hn|yi,
G(x, y) =
En
n
(2.21)
(2.22)
e with the eigenvalue En . We may adopt the vacuum satwhere |ni is the eigenstate of H
uration type approximation by only considering hx|ni = exp(ikE xE) 0 as the intermediate
states. In this way we obtain the propagator of massless fields in four dimensions:
Z
1
x yE ) 2 0 () 0 ( 0 )
G(x, y) d4 k exp ikE (E
k
1
0 () 0 ( 0 ).
(2.23)
(E
x yE)2
As the final result of these investigations, we find the following gravitational interaction:
Z
1
x , )h (E
x , )
d10 x A5/2 ()T (E
2
Z
Z
x , ) 0 ()
= d10 x d10 y A7/4()T (E
1
y , 0 )A7/4( 0 )
0 ( 0 ) 2 ( 0 )T (E
(E
x yE)2
1
.
(E
x yE )2
We have introduced the four dimensional energy momentum tensor
Z
x ) = d d5 T (E
x , )A5/2 ().
Te (E
2
d4 x
d4 y Te (E
x )Te (E
y)
167
(2.24)
(2.25)
The interaction between them is of the four dimensional graviton exchange type with the
gravitational coupling = (0).
Here we remark on the Hermiticity of the Hamiltonians in Eqs. (2.16) and (2.18). The
latter is Hermitian with respect to the trivial norm since

Z
3 A0
3 A0
i
+i
d
i
4A
i
4A

Z
3 A0
3 A0
i
+i
.
(2.26)
= d
i
4A
i
4A
It is translated into the Hermiticity condition on the former after the similarity transformation as

Z
Z
3/2
,
(2.27)
d g g
d A
i
i

Therefore our Hamiltonian is positive definite with respect to the
where = A3/4 .
natural norm defined by the string frame metric g . In this sense our important physical
input is our identification of the physical metric with string frame metric g . We have
checked that our zero mode remains the exact zero mode after taking account of other
terms in Eq. (2.13).
Therefore we argue that we can obtain four dimensional gravity with NCYM 4 a la
RandallSundrum [7]. Since not only the metric but also the dilaton expectation value
(string coupling) rapidly decay in the large r region, we expect that there is essentially
nothing outside the noncommutativity scale transverse to the brane. In fact we have
postulated this kind of compactification mechanism in the matrix models [5,6,54]. We
have expected that four dimensional gravitation is obtained if the eigenvalue distribution
of the matrices are four dimensional. It is because the matrices represent spacetime
coordinates in our proposal. In our interpretation, there is simply nothing outside the
support of the eigenvalue distributions, not even spacetime.
We observe that this Euclidean solution can be analytically continued into Minkowski
spacetime only in the small r region. One possible interpretation of such a solution
is to maintain that Minkowski spacetime appears from NCYM 4 as its low energy
approximation. We may identify the noncommutativity scale with Planck scale if we
apply this model to our spacetime. Although the Lorentz invariance is broken at
the noncommutativity scale in this model, such a possibility is not excluded by the
experiments. Therefore NCYM 4 is a candidate of the unified theory of interactions. We
explain in the subsequent sections that IIB matrix model naturally provides us with such
a theory. We still need to solve many problems such as breaking SUSY and finding chiral
fermions to construct a realistic unified theory. We hope that these problems can be solved
by further investigations in IIB matrix model.
168
3. Noncommutative field theories as twisted reduced models

In this section we briefly recapitulate our formulation of NCYM through large N
reduced models. We have pointed out that well-known twisted reduced models [2633] 6
are equivalent to NCYM. This connection is further studied in [36,37,47]. We consider
d-dimensional U (n) gauge theory coupled to adjoint matter as an example:

Z
1
1
1
d
(3.1)
S = d x 2 Tr [D , D ][D , D ] + [D , ] ,
4
2
g
where is a Majorana spinor field. The corresponding reduced model is

1
1
1
(3.2)
S = 2 Tr [A , A ][A , A ] + [A , ] .
4
2
g
Now A and are n n Hermitian matrices and each component of is d-dimensional
Majorana-spinor.
We expand A = p + a around the following classical solution

(3.3)
p , p = iB ,
where B are c-numbers. We assume the rank of B to be d and define its inverse C in
d dimensional subspace. The directions orthogonal to the subspace is called the transverse
directions. p satisfy the canonical commutation relations and they span the d dimensional
phase space. The
semiclassical correspondence shows that the volume of the phase space
is Vp = n(2)d/2 det B.
We Fourier decompose a and fields as
X

a(k)
exp iC k p ,
a =
k
(k)
exp iC k p ,
(3.4)
where exp(iC k p ) is the eigenstate of adjoint P = [p , ] with the eigenvalue k . The
and (k) = (k).

Hermiticity requires that a (k) = a(k)
We can construct a map from a matrix to a function as
X
a(k)
exp(ik x),
(3.5)
a a(x) =
where k x = k
x . By
this construction, we obtain the ? product
a b a(x) ? b(x),

2
C
(3.6)
a(x + )b(x + ) ==0 .
a(x) ? b(x) exp
2i
The operation Tr over matrices can be exactly mapped onto the integration over functions
as
d/2
Z
(3.7)
dd x a(x).
Tr[a]
= det B
2
6 The relevance of reduced models and string theory was first recognized in [34,35].
169
The twisted reduced model can be shown to be equivalent to NCYM by the the following
map from matrices onto functions
a a(x),
a b a(x) ? b(x),
d/2
Z
dd x.
Tr det B
2
(3.8)
The following commutator is mapped to the covariant derivative:

1
(3.9)
p + a , o o(x) + a (x) ? o(x) o(x) ? a (x) D , o(x) .
i
We may interpret the newly emerged coordinate space as the semiclassical limit of x =
C p . Therefore we can interpret A as momenta as well in IIB matrix model with
noncommutative backgrounds since x and p are linearly related. It is the reflection of
the remarkable T-duality property of the theory. The spacetime translation is realized by
the following unitary operator:
exp(ip d) x exp(ip d) = x + d .
(3.10)
Applying the rule Eq. (3.8), the bosonic action becomes

1
Tr A , A A , A
2
4g
d/2

Z
2
1
1 1
dnB
d
[D , D ][D , D ]
det
B
x
d
=
4g 2
2
g2 4

1
1
+ [D , ][D , ] + [ , ][ , ] .
2
4
?
(3.11)
In this expression, the indices , run over d dimensional world volume directions and
, over the transverse directions. We have replaced a in the transverse directions.
Inside ( )? , the products should be understood as ? products and hence commutators do not
vanish. The fermionic action becomes
1
[A , ]
Tr
g2
d/2
Z

1
1
[ , ] .
(3.12)
dd x 2
= det B
[D , ] +
?
2
g
We therefore find noncommutative U(1) gauge theory. In order to obtain NCYM with U (m)
gauge group, we need to consider new classical solutions which are obtained by replacing
each element of p by the m m unit matrix:
p p 1m .
(3.13)
The Hermitian models are invariant under the unitary transformation: A U A U ,

U U . As we shall see, the gauge symmetry can be embedded in the U (N)
symmetry. We expand U = exp(i ) and parameterize
170
(k) exp(ik x).
(3.14)
Under the infinitesimal gauge transformation, we find the fluctuations around the fixed
background transform as

a a + i p , i a , ,

.
i ,
(3.15)
We can map these transformations onto the gauge transformation in NCYM by our rule
Eq. (3.8):
(x) ia (x) ? (x) + i(x) ? a (x),

x
(x) (x) i (x) ? (x) + i(x) ? (x),
a (x) a (x) +
(x) (x) i(x) ? (x) + i(x) ? (x).
(3.16)
We have introduced another representation of matrices [17]. For simplicity we consider

the two dimensional case first:
[x,
y]
= iC.
(3.17)
This commutation relation is realized by the guiding center coordinates of the two
dimensional system of electrons in magnetic field. We recall that we have n quanta with
n dimensional matrices. Each quantum occupies the spacetime volume of 2C. We may
2 = 2C.
consider a square von Neumann lattice with the lattice spacing lNC where lNC
This spacing lNC gives the noncommutative scale. Let us denote the most localized state
We construct
centered at the origin by |0i. It is annihilated by the operator x = x iy.
states localized around each lattice site by utilizing translation operators |xi i = exp(ixi
p)|0i.
They are the coherent states on a von Neumann lattice xi = lNC (ni ex + mi ey ), where
n, m Z. The generalizations to arbitrary even d dimensions are straightforward.
We evaluate the following matrix elements

(xi xj )2
i
.
(3.18)
ij hxi |xj i = exp B xi xj exp
2
4C
Although |xi i are non-orthogonal, hxi |xj i exponentially vanishes when (xi xj )2 gets
large. We also find

(xi + xj )
i

+ B xi xj
xj = exp ik
xi exp(ik x)
2
2

(xi xj d)2
,
(3.19)
exp
4C
where d = C k . This matrix element sharply peaks at xi xj = d. It supports our
interpretation that the eigenstate exp(ik x)
with |k | > 2/ lNC can be interpreted as string
like extended objects whose length is |C k |. When |k | < 2/ lNC , on the other hand,
this matrix becomes close to diagonal whose matrix elements go like

(3.20)
xj exp(ik xi )hxi |xj i.
xi exp(ik x)
171
Fig. 2. Noncommutative fields are represented as bi-local fields on the von Neumann lattice with
the lattice spacing lNC . Small momentum modes (|k| < 2/ lNC ) represent ordinary (commutative)
fields. Large momentum modes (|k| > 2/ lNC ) represent bi-local open strings which are highly
nonlocal. Only a fraction of the full momentum (|k c | < 2/ lNC ) can be interpreted as the momentum
associated with the center of mass motion of an open string on the von Neumann lattice. In the
figure, d = C kc .
It again supports our interpretation that exp(ik x)

correspond to the ordinary plane waves
when |k | < 2/ lNC . They are represented by the matrices which are close to diagonal.
We may expand matrices in the twisted reduced model by the following bi-local basis
as follows:
X
(xi , xj )|xi ihxj |,
(3.21)
=
i,j
where the Hermiticity of implies (xj , xi ) = (xi , xj ). The matrices represent a

or in the super YangMills case but the setting here is more generally applied to an
arbitrary noncommutative field theory. The bi-local basis spans the whole n2 degrees of
freedom of matrices. 7
Here we work out the translation rule between the momentum eigenstate representation
P
exp(ik x)
and the bi-local field representation of Eq. (3.21):
= k (k)
1X
1
=
hxi | exp(ik x)|x
j i(xj , xi )
(k)
= Tr exp(ik x)
n
n
=
(xc , d) =
1X
n
(xc , d) exp(ik xc ),
xc
X
xr
i,j

i
(xr d)2

(xj , xi ),
exp B xr xc exp
2
4C
(3.22)
where xc = (xi + xj )/2 and xr = xi xj . From Eq. (3.22), we observe that the slowly
varying field with the momentum smaller than 2/ lNC consists of the almost diagonal
7 Bi-local fields have also appeared in c = 1 string theory [55,56].
172
components. Hence close to diagonal components of the bi-local field are identified with
the ordinary slowly varying field (xc ). On the other hand, rapidly oscillating fields are
mapped to the off-diagonal open string states. A large momentum in the th direction
|k | > 2/ lNC corresponds to a large distance in the th direction |d | = |C k | > lNC .
We can decompose d as d = d0 + d where d0 is a vector which connects two points on
the von Neumann lattice and |d | < lNC . This decomposition is illustrated in Fig. 2. Then
the summation over xr in (3.22) is dominated at xr = d0 . In this way the large momentum
degrees of freedom are more naturally interpreted as extended open string-like fields. They
are denoted by open strings in this paper. In this representation, we make contact with
the quenched reduced models [2630] in the large momentum region.
Here we remark the important property concerning the infra-red and ultra-violet cutoffs of NCYM constructed with n dimensional matrices. Since the unit lattice size of the
von Neumann lattice is lNC , the total lattice size is lNC n1/d . It implies that the maximum
1/
d
momentum is 2n / lNC by using the relation p = B x . It in turn implies that the
minimum momentum of the system is 2/n1/d lNC since we have n2 momentum modes.
The matrix model construction of NCYM implies very natural infra-red and ultra-violet
cut-offs which disappear in the large n limit.
4. Estimations of the string scale

0 in IIB matrix model with noncommutaIn this section, we estimate the string scale eff
tive backgrounds. We have explained that the von Neumann lattice naturally appears in the
preceding section. We argue that NCYM is superstring theory on the von Neumann lattice.
We first give the arguments based on the tree level propagators. We then explain that our
claim is supported by T-duality arguments. We give another evidence for it by investigating
graviton exchange processes in the next section.
As we have shown in the preceding section, the momentum which can be associated with
the center of mass motion of an open string is not full k but rather kc = B d . There
we have decomposed d = C k as d = d0 + d where d0 is a vector which connects
of
two points on the von Neumann lattice and |d | < lNC . We can indeed represent (k)
Eq. (3.22) as follows:
X
c , d) = 1
(xc , d) exp(ik c xc ).
(4.1)
(k
n x
c
It is because

exp iB d0 xc = 1.
(4.2)
c , d) as the creation-annihilation operator for the open string with

We interpret (k
momentum k c and length d.
We consider the following tree level propagator:
X
(4.3)
(k)
(k)
exp(ik0 ) exp(m ),
k0
173
where k0 and kE denote time-like and spatial momenta respectively in this section. The mass
term is conventionally identified with the zero spatial momentum limit of the correlator
Eq. (4.3). In order to relate it to the mass of an open string, we consider a state with
k1q< 2/ lNC < k2 , k3 . Such a state is extended in (x 2 , x 3 ) plane with the length l =
C k22 + k32 . As we have argued, the momentum which can be associated with the center
of mass motion of an open string is not full kE but rather kEc . We find m Bl from the
zero momentum limit (kEc 0) of the correlator Eq. (4.3). From these considerations, we
propose to identify the mass of the state with the length l lNC as Bl. We recall that an
0 = C with
open string with the length l has the mass of l/(2 0 ). Therefore we find 2eff
such an identification.
We remark that our estimate is consistent with the string theory arguments. From
0 c in Section 2. A hint for such an
Eq. (2.7), we have indeed found that the eff
identification has been found in a finite temperature investigation in [42]. Our estimate
is also consistent with the spacetime uncertainty principle of Yoneya [44].
It is certainly true that we obtain NCYM in the low energy limit of string theory with
b backgrounds. In such a limit, we retain only those degrees of freedom whose masses
are smaller than the string scale. However in IIB matrix model with noncommutative
backgrounds, we also find very high energy modes which may be interpreted as long
open strings due to the relation x = C p . As can be seen in Eq. (2.7) they are as
massive as oscillator modes if their lengths exceed lNC . We therefore emphasize here that
our formulation is not a low energy limit of string theory. The graviton exchanges are
observed only because we have very long open strings in the matrix model. To put it
differently, infrared singular behaviors are observed in noncommutative field theory only
if we consider the ultraviolet cut-off which is much larger than the noncommutativity scale.
It is the reason why closed strings do not decouple in such a formulation.
In our picture, we interpret the bi-local fields as the zero modes of open strings. We
classify the zero modes as momentum modes and winding modes as follows. We recall
the von Neumann lattice |xi i which is constructed by the generators of the translation
We can compactify the theory by imposing the
operators U = exp(ilNC e p).
following conditions for fluctuations
.
= U U
(4.4)
In this T-dual picture, the von Neumann lattice can be identified with the lattice spanned by
the winding modes. We thus classify those modes as winding modes. We have explained
that kc can be interpreted as momentum modes which can be associated with the center
of mass motion of open strings on the von Neumann lattice. We thus classify them as
momentum modes.
In string theory, the introduction of constant b background is known to interpolate
the Neumann and Dirichlet boundary conditions. In the large and small b limit, we
find Dirichlet and Neumann boundary conditions respectively. We find only winding
and momentum modes in these limits. If we expand IIB matrix model around the
commutative backgrounds, we only find winding modes. In string theory we also find
174
only winding modes if we consider strings which connect D instantons. The advantage of
noncommutative backgrounds is that we find the both momentum and winding modes.
Since we have found the both momentum and winding modes, it is no surprise
that the theory possesses T-duality. The remarkable property of NCYM is the existence of
Morita equivalent pairs [23,45,46]. We propose that two Morita equivalent theories can be
related by the exchange of the momentum and winding modes.
We have argued that the winding modes of NCYM span the von Neumann lattice
whose lattice unit size is lNC . The total lattice size is lNC n1/d . We may reinterpret it as the
maximum momentum 2n1/d / lNC of the dual lattice. The dual lattice possesses the unit
1/
d
lattice size of lNC /n . We consider a twisted large N reduced model on such a lattice.
In this way we find a pair of theories with the compactification radii R = lNC n1/d /2
0 /R with
and R 0 = lNC /(n1/d 2). They are related by the duality transformation R 0 = eff
0
2eff = C. Next we recall that the momentum modes of NCYM are quantized in the
unit of lNC /n1/d C. We can naturally reinterpret them as the winding modes of the dual
lattice. These winding modes can be obtained by introducing the unit magnetic flux in
U (n) gauge theory by imposing twisted boundary conditions. In this sense NCYM and
twisted reduced models are Morita equivalent [47].
We remark that the T-duality transformation we have discussed is expressed by the
following open string metric transformation in string theory [23]

(4.5)
G G T
where = C /(2R 2 ) 1/n2/d . Our interpretation is that the two metrics which
are related by the T-duality transformation in Eq. (4.5) describe two tori we have just
0 = C is
constructed. We conclude that our estimate of the inverse string tension 2eff
also supported by the T-duality arguments. We argue that this is the exact result since it is
obtained from the exact T-duality property of the theory.
5. Graviton exchange processes

In this section, we study gravitational interactions in IIB matrix model with noncommutative backgrounds. To be specific, we consider photonphoton scattering via exchange of
a graviton. This process can be studied by considering blockblock interactions. Namely,
we consider the backgrounds of the following type:

p + a
0
,
(5.1)
A =
0
p + a0
where a and a0 denote the backgrounds which represent two colliding photons.
The one-loop effective action of IIB matrix model is

1
i
1 + 11
1
Re W = T r log P2 2iF T r log P2 + F
2
4
2
2

2
T r log P .
(5.2)
175
Here P and F are operators acting on the space of matrices as

P X = [p , X],
F X = [f , X],
(5.3)
where f = i[p , p ]. Now we expand the general expression of the one-loop effective
action (5.2) with respect to the inverse powers of the relative distance between the two
blocks. We quote the general expression in what follows [1]:

1
1
1
1
F
F
F
F
W = T r
2 2 2
P2
P
P
P

1
1
1
1
F
F
F
F
2T r
2 2 2
P2
P
P
P

1
1
1
1
1
F
F
F
F
+ Tr
2 2 2
2
P2
P
P
P

1
1
1
1
1
5
F
F
F
F
(5.4)
+ Tr
2 2 2 + O (F ) .
2
4
P
P
P
P
Since P and F act on the (i, j ) blocks independently, the one-loop effective action W
is expressed as the sum of contributions of the (i, j ) blocks W (i,j ) . Therefore we may
consider W (i,j ) as the interaction between the ith and j th blocks.
p
Using (5.4) we can easily evaluate W (i,j ) to the leading order of 1/ (d (i) d (j ) )2 as

1
W (i,j ) = 8 T r (i,j ) (F F F F ) 2 T r (i,j ) (F F F F )
r

1
1
+ T r (i,j ) (F F F F ) + T r (i,j ) (F F F F )
2
4

3
= 8 nj b8 f (i) ni b8 f (j )
2r

(i) (i)
(j ) (j )
(i) (i)
(j ) (j )
f Tr f
f + Tr f
f Tr f
f ,
(5.5)
8 Tr f
where

2
Tr(f f f f ) + 2 Tr(f f f f )
b8 (f ) =
3

1
1
(5.6)
Tr(f f f f ) Tr(f f f f ) .
2
4
So the photonphoton scattering amplitude which corresponds to nonplanar diagrams in
noncommutative gauge theory is

3
(i) (i)
(j ) (j )
(i) (i)
(j ) (j )
f
f
f
f
Tr
f
+
Tr
f
Tr
f
8
Tr
f

2r 8
d
Z
Z
1
3 1
B d8 dd x dd y
=
2 2
(x y)8
n

8 tr f (x)f (x) tr f (y)f (y)

o
(5.7)
+ tr f (x)f (x) tr f (y)f (y) ,
where we have used our mapping rule Eq. (3.8).
176
We consider the scattering of two incident photons with the wave functions
e(p) exp(ip x),
where
p2
= q2
e(q) exp(iq x),
(5.8)
= 0 and p e(p) = q e(q) = 0. In this case
f (p) = p e(p) p e(p) ,
f (p)f (q) = p q e(p)e(q) q p e(q)e(p)

+ p q e(p) e(q) + e(p) e(q) p q.
(5.9)
If we consider the forward scattering limit k = p q 0:

f (p)f (q) p p ,
we find only graviton exchange in this limit.
d
Z
Z
1
1
B d8 dd x dd y
12
2
(x y)8

tr f (x)f (x) tr f (y)f (y) .
(5.10)
(5.11)
This expression reminds us the one photon exchange amplitude between two conserved
currents j and j in QED4 :
Z
1
d4 k
j (k) 2
j (k)
4
(2)
k + i
Z
Z
1
i
j (y).
(5.12)
d4 x d4 y j (x)
=
4
(x y)2 i
We decompose currents into positive and negative frequency parts
j (x) = j+ (x) + j (x),
Z
+
j (x) = d exp(ix 0 )j+ (, xE ),
0
j (x) =
d exp(ix 0 )j (, xE).
(5.13)
We rewrite Eq. (5.12) as follows

Z
Z

1
1
4
j+ x 0 |E
x yE|, yE
d x dyE j (x 0 , xE )
4
|E
x yE|
Z
Z
1
1
x j+ y 0 |E
x yE|, xE
d4 y dE
j (y 0 , yE).
+
4
|E
x yE|
(5.14)
In this way we find retarded LienardWiechert type interactions. The point we would like
to make here is that covariance implies causality. Since ten dimensional covariance is
manifest in IIB matrix model, it naturally leads to ten dimensional causality in Minkowski
spacetime. On the other hand, ensuring ten dimensional causality is highly nontrivial in
AdS/CFT correspondence [48].
We recall the relevant part of the NCYM action as follows

d/2
Z

d
1
1
4
2
dd x tr f f ? .
B
2
2
4g
The energymomentum tensor can be read off from it in the low energy limit as
d/2

d
1
1
1
f
f
f
.
B 2 4
2
4
g2
So we can rewrite the gravitational interaction Eq. (5.11) as follows
Z
Z
1
d
4
T (x)T (y).
d x dd y
12g
(x y)8
We recall the d dimensional propagators
Z
((d 2)/2) 1
dd p 1
exp(ip x) =
.
p2
d
(2)
4 d/2
x d2
For d = 10, we obtain
3 1
.
G10 (x) =
2 5 x 8
The gravitational coupling is found to be
2 = 16 5 g 4 .
177
(5.15)
(5.16)
(5.17)
(5.18)
(5.19)
(5.20)
We also read off the d dimensional YangMills coupling from Eq. (5.15) as
2
= C (d8)/2g 2 (2)d/2 .
gYM
(5.21)
Here we quote string theory predictions:

1
2
gYM
= (2)d3 gs 0(d4)/2 .
(5.22)
2 = (2)7 gs2 04 ,
2
0 = C. We
Eq. (5.22) agrees with Eq. (5.20) and Eq. (5.21) with our identification 2eff
0
also find that the IIB matrix model coupling can be expressed by eff and gs as
0
,
g 2 = (2)gs eff
2
(5.23)
which is consistent with our previous estimate through D-strings [1]. What these
investigations indicate is that NCYM is superstring theory with the above identified string
scale and string coupling. We have argued that it is superstring theory on the von Neumann
lattice. Since the lattice spacing lNC is not visible in the low energy limit, it may be expected
that it behaves like ordinary superstring theory in the low energy limit.
We find that the gravitational interaction Eq. (5.17) exhibits the identical power law
It appears as
behavior 1/(x y)8 irrespective of the dimensionality of the backgrounds d.
if these background represent D-branes in flat ten dimensional spacetime. However we
argue that such an interpretation is premature since we have only considered the one loop
effect here. We argue that a more reliable picture is obtained through supergravity approach
which allows us to investigate nonperturbative effects.
178
As it is explained in Section 2, we find Newtons force law with these backgrounds. This
is due to the existence of a massless bound state a la Randall and Sundrum. Such an effect
is not visible in the perturbative calculations in this section. Therefore the supergravity
description of NCYM 4 suggests a nonperturbative compactification mechanism in IIB
superstring and matrix model.
In the matrix model construction, the longitudinal size of the system is bounded by
lNC n1/4 . It also implies that the transversal size r is bounded by lNC n1/4 since it is identified
2 ). In Eq. (2.8), the dilaton
with the maximum energy scale of the system (multiplied by lNC
2
expectation value is O(1) at the noncommutativity scale r 1. We then find g is
O(1/n) since the dilaton decays like 1/r 4 beyond the noncommutativity scale. We have
fixed = g 02 b2 to be O(1) which implies that 0 b n. We conclude that lNC

ls /n1/4 and r never exceeds O(ls ) where ls is the string scale. Therefore there is simply no
region with r > ls in the matrix model. We have taken the noncommutativity scale to be
O(1) and the cuff-off scale of r becomes O(n1/4 ). The cut-off can be removed in the large
n limit of the matrix model construction. In this way we can realize the entire spacetime
which is described by Eq. (2.8).
6. Conclusions and discussions

In this paper we have argued that we can obtain Newtons force law with NCYM 4 .
Since it contains four dimensional gauge theory and gravitation, it is a candidate of the
unified theory. It can be regarded as a compactification of ten dimensional IIB superstring
straight down to four dimensions. It is naturally obtained in IIB matrix model with
noncommutative backgrounds. Therefore it provides a nonperturbative compactification
mechanism of matrix models.
We have identified the bi-local fields as the zero modes of open strings. They can be
interpreted as the momemtum and winding modes on the von Neumann lattice. Our
identification of the effective string scale with the noncommutativity scale is consistent
with the exact T-duality which interchanges the momentum and winding modes.
Although we have identified the zero modes of open strings, we have not constructed
oscillator modes. We expect to find them in higher order diagrams. Let us consider a
propagator (ribbon diagram). We associate it with a bi-local field since the double lines
of the ribbon are mapped to two distinct spacetime points. We need to draw many loops
inside the ribbon at higher orders in perturbation theory. We can assign a spacetime point
to each loop. Our conjecture is that such an object can be interpreted as the propagator of
oscillation modes. These arguments are illustrated in Figs. 3 and 4.
It is important to recall here that we identify the string coupling with the topological
expansion parameter of the Feynman diagrams of NCYM. It is not equal to the NCYM
coupling although the both are related in the low energy limit as suggested by supergravity
solutions. The tree level string propagator is obtained by summing all planar diagrams. Our
proposal that NCYM with maximal SUSY may be interpreted as superstring theory on the
von Neumann lattice should be understood in this context.
179
Fig. 3. Propagators of bi-local fields. The two end-points are mapped onto spacetime coordinates.
Fig. 4. Higher order corrections to propagators of bi-local fields may render the bi-local fields stringy.
As we have pointed out, NCYM is obtained with a particular classical background

in IIB matrix model. IIB matrix model is postulated as a nonperturbative formulation
of superstring theory. In our proposal, the matrices A are to be interpreted as space
time coordinates. If so, d dimensional distributions of eigenvalues of matrices represent d
dimensional spacetime. It is then expected that we find d dimensional gauge theory and
gravitation with such a background. In this paper we have argued that it is indeed the case
with maximally supersymmetric backgrounds. From the findings in this paper, we draw the
conclusion that NCYM provides a strong support for our basic premises of our IIB matrix
model conjecture.
Acknowledgements
This work is supported in part by the Grant-in-Aid for Scientific Research from the
Ministry of Education, Science and Culture of Japan.
References
[1] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, A large-N reduced model as superstring,
Nucl. Phys. B 498 (1997) 467, hep-th/9612115.
[2] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[3] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[4] E. Witten, Adv. Math. Phys. 2 (1998) 253, hep-th/9802150.
[5] H. Aoki, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Spacetime structures from IIB matrix model,
Prog. Theor. Phys. 99 (1998) 713, hep-th/9802085.
[6] H. Aoki, S. Iso, H. Kawai, Y. Kitazawa, A. Tsuchiya, T. Tada, IIB matrix model, Prog. Theor.
Phys. Suppl. 134 (1999) 47, hep-th/9908038.
[7] L. Randall, R. Sundrum, An alternative to compactification, hep-th/9906064.
[8] J. Polchinski, Phys. Rev. Lett. 75 (1995) 4724.
[9] A. Connes, M. Douglas, A. Schwarz, JHEP 9802 (1998) 003, hep-th/9711162.
[10] H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Noncommutative YangMills in
IIB matrix model, Nucl. Phys. 565 (2000) 176, hep-th/9908141.
[11] A. Hashimoto, N. Itzhaki, Noncommutative YangMills and the AdS/CFT correspondence,
hep-th/9907166.
180
[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]
J. Maldacena, J.G. Russo, Large N limit of noncommutative gauge theory, hep-th/9908134.

T. Banks, W. Fischler, S.H. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hep-th/9610043.
W. Taylor, Phys. Lett. B 394 (1997) 283, hepth/9611042.
A. Connes, Commun. Math. Phys. 182 (1996) 155, hep-th/9603053.
N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Wilson loops in noncommutative YangMills, Nucl.
Phys. B 573 (2000) 573, hep-th/9910004.
S. Iso, H. Kawai, Y. Kitazawa, Bi-local fields in noncommutative field theory, Nucl. Phys. B 576
(2000) 375, hep-th/0001027.
Y.E. Cheung, M. Krogh, Noncommutative geometry from 0-branes in a background B-field,
Nucl. Phys. B 528 (1998) 185, hep-th/9803031.
F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Noncommutative geometry from strings and
branes, JHEP 9902 (1999) 016, hep-th/9810072.
C. Chu, P. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hepth/9812219.
C. Chu, P. Ho, Constrained quantization of open string in background B field and noncommutative D-brane, Nucl. Phys. B 568 (2000) 447, hep-th/9906192.
V. Schomerus, JHEP 9906 (1999) 030, hep-th/9903205.
N. Seiberg, E. Witten, String theory and noncommutative geometry, hep-th/9908142.
H. Kawai, Y. Kitazawa, M. Ninomiya, Nucl. Phys. B 393 (1993) 280.
G. t Hooft, M. Veltman, Ann. Inst. Henri Poincare 20 (1974) 69.
T. Eguchi, H. Kawai, Phys. Rev. Lett. 48 (1982) 1063.
G. Parisi, Phys. Lett. B 112 (1982) 463.
D. Gross, Y. Kitazawa, Nucl. Phys. B 206 (1982) 440.
G. Bhanot, U. Heller, H. Neuberger, Phys. Lett. B 113 (1982) 47.
S. Das, S. Wadia, Phys. Lett. B 117 (1982) 228.
A. Gonzalez-Arroyo, M. Okawa, Phys. Lett. B 120 (1983) 174; Phys. Rev. D 27 (1983) 2397.
T. Eguchi, R. Nakayama, Phys. Lett. B 126 (1983) 89.
T. Filk, Phys. Lett. B 376 (1996) 53.
D.B. Fairlie, P. Fletcher, C.Z. Zachos, J. Math. Phys. 31 (1990) 1088.
I. Bars, Phys. Lett. B 245 (1990) 35.
J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, Finite N matrix models of noncommutative gauge theory, hep-th/9911041.
I. Bars, D. Minic, Noncommutative geometry on discrete periodic lattice and gauge theory,
hep-th/9910091.
I. Chepelev, R. Roiban, Renormalization of quantum field theories on noncommutative R d ,
I. Scalars, hep-th/9911098.
S. Minwalla, M.V. Raamsdonk, N. Seiberg, Noncommutative perturbative dynamics, hepth/9912072.
M. Hayakawa, Perturbative analysis on infrared aspects of noncommutative QED on R 4 , hepth/9912094.
M. Hayakawa, Perturbative analysis on infrared and ultraviolet aspects of noncommutative QED
on R 4 , hep-th/9912167.
W. Fischler, E. Gorbatov, A. Kashani-Poor, S. Paban, P. Pouliot, J. Gomis, Evidence for winding
states in noncommutative quantum field theory, hep-th/0002067.
M.V. Raamsdonk, N. Seiberg, Comments on noncommutative perturbative dynamics, hepth/0002186.
T. Yoneya, Prog. Theor. Phys. 97 (1997) 949, hep-th/9703078.
B. Pioline, A. Schwarz, Morita equivalence and T duality, hep-th/9908019.
A. Hashimoto, N. Itzhaki, On the hierarchy between non-commutative and ordinary supersymmetric YangMills, hep-th/9911057.
181
[47] J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, Nonperturbative dynamics of noncommutative gauge theory, hep-th/0002158.
[48] D. Bak, S.J. Rey, Holographic view of causality and locality via branes in AdS/CFT
correspondence, hep-th/9902101.
[49] O. Andreev, H. Dorn, Diagrams of noncommutative phi-three theory from string theory, hepth/0003113.
[50] Y. Kiem, S. Lee, UV/IR mixing in noncommutative field theory via open string loops, hepth/0003145.
[51] A. Bilal, C.S. Chu, R. Russo, String theory and noncommutative field theores at one loop, hepth/0003180.
[52] J. Gomis, M. Kleban, T. Mehen, M. Rangamani, S. Shenker, Noncommutative gauge dynamics
from string worldsheet, hep-th/0003215.
[53] H. Liu, J. Michelson, Stretched strings in noncummutative field theory, hep-th/0004013.
[54] J. Nishimura, G. Vernizzi, Spontaneous breakdown of Lorentz invariance in IIB matrix model,
hep-th/0003223.
[55] A. Dhar, G. Mandal, S. Wadia, Mod. Phys. Lett. A 7 (1992) 3129, hep-th/9207011.
[56] S. Iso, D. Karabali, B. Sakita, Phys. Lett. B 296 (1992) 143, hep-th/9209003.

Two-body decays of the lightest stop in minimal

supergravity with and without R-parity
Marco A. Daz a,c , Diego A. Restrepo b , Jos W.F. Valle b,
a Department of Physics, Florida State University, Tallahassee, FL 32306, USA
b Departamento de Fsica Terica, IFIC-CSIC, Universidad de Valencia Burjassot, Valencia 46100, Spain
c Departamento de Fsica, Universidad Catlica de Chile, Av. Vicua Mackenna 4860, Santiago 6904411,
Chile
Received 25 October 1999; revised 15 March 2000; accepted 23 May 2000
Abstract
We study the decays of the lightest top squark in supergravity models with and without R-parity.
Using the simplest model with an effective explicit bilinear breaking of R-parity and radiative
electroweak symmetry breaking we show that, below the threshold for decays into charginos t1
b 1+ , the lightest stop decays mainly into third generation fermions, t1 b instead of the R-parity
conserving mode t1 c 10 , even for tiny tau-neutrino mass values. Moreover we show that, even
above the threshold for decays into charginos, the decay t1 b may be dominant. We study the role
played by the universality of the boundary conditions on the soft supersymmetry breaking terms. This
new decay mode t1 b as well as the cascades originated by the conventional t1 c 10 decay
followed by the R-parity violating neutralino decays can provide new signatures for stop production
at LEP and the Tevatron. 2000 Elsevier Science B.V. All rights reserved.
PACS: 14.80.Ly; 11.30.Pb; 11.30.Fs; 12.10.Dm; 12.10.Kt; 12.60.Jv
1. Introduction
The Minimal Supersymmetric Standard Model (MSSM) [1] or its minimal supergravity
(SUGRA) version [2] are by far the most well studied realizations of supersymmetry.
However, neither gauge invariance nor supersymmetry requires the conservation of Rparity. Indeed, there is considerable theoretical and phenomenological interest in studying
possible implications of alternative scenarios (for a recent review, see [3,4]) in which
R-parity is broken [59]. This is especially so considering the fact that it provides an
appealing joint explanation of the solar and atmospheric neutrino anomalies which has,
in addition, the virtue of being testable at future accelerators like the LHC [10,11]. The
valle@neutrinos.uv.es
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 4 4 - 8
M.A. Daz et al. / Nuclear Physics B 583 (2000) 182210
183
violation of R-parity could arise explicitly [1218] as a residual effect of some larger
unified theory [6], or spontaneously, through nonzero vacuum expectation values (vevs)
for scalar neutrinos [5,79]. In realistic spontaneous R-parity breaking models there is an
SU (2) U (1) singlet sneutrino vev characterizing the scale of R-parity violation [1931]
which is set by the supersymmetry breaking scale.
There are two generic cases of spontaneous R-parity breaking models to consider. In
the absence of any additional gauge symmetry, these models lead to the existence of
a physical massless NambuGoldstone boson, called majoron (J) which is the lightest
SUSY particle, massless and therefore stable. It plays an important role in making these
models fully consistent with astrophysics and cosmology [3,4] if one wishes to contemplate
the case of large breaking scales and heavy tau neutrino. If the majoron acquires a
small mass due to explicit breaking effects at the Planck scale then it may decay into
electron and muon neutrinos or photons, on very large time scales of cosmological
interest, playing a possible role as unstable dark matter [32,33]. Alternatively, if lepton
number is part of the gauge symmetry and R-parity is spontaneously broken then there
is an additional gauge boson which gets mass via the Higgs mechanism, and there
is no physical Goldstone boson [31]. As in the standard case in R-parity breaking
models the lightest SUSY particle (LSP) is in general a neutralino. However, it now
decays mostly into visible states, therefore diluting the missing momentum signal and
bringing in increased multiplicity events which arise mainly from three-body decays such
as
10 f f,
(1)
where f denotes a charged fermion. The neutralino also has the invisible decay
mode
10 3.
(2)
as well as
10 J,
(3)
in the case the breaking of R-parity is spontaneous [1924]. This last decay conserves
R-parity since the majoron has a large R-odd singlet sneutrino component.
If R-parity is broken then supersymmetric (SUSY) particles need not be produced only
in pairs, and the lightest of them could decay. The effects of R-parity violation can be large
enough to be experimentally observable.
In this paper we focus on the decay modes of the lightest top squark in supergravity
models where supersymmetry is realized with R-parity violation. In such models the
lightest stop could even be the lightest supersymmetric particle and be produced at LEP.
Neither e+ e collider data [3438] nor pp data from the Tevatron [3942] preclude this
possibility. In contrast with Ref. [43] here we focus on an effective model where the
breaking of R-parity is introduced through an explicit bilinear term in the superpotential.
This is substantially simpler than the full majoron version of the model considered
previously. In fact, this bilinear model is not only especially simple theoretically, also its
184
phenomenological implications in collider physics can be studied in a very systematic way.

The bilinear model constitutes the simplest R-parity breaking model [44] consistent with
radiative electroweak symmetry breaking, very much the same way as the minimal Rparity conserving supergravity models with universal soft SUSY breaking terms [45,46],
MSUGRA, for short. As mentioned it also provides an attractive joint explanation of the
present neutrino anomalies [10,11].
In order to discuss stop decays we also refine the work presented in Ref. [4755] by
giving, for the first time, an exact numerical calculation for the FCNC process t c 10 .
We also compare the results obtained this way with those one gets by adopting the usual
one-step or leading logarithm approximation in the RGEs. In contrast with the R-parity
conserving model such an approximation would be rather poor for our purposes, since
we will be interested in comparing FCNC with R-parity violating stop decay modes
(see Section 5). Moreover, in contrast to Ref. [43], where the magnitude of the stop
charmneutralino coupling was a phenomenological parameter, here we assume a minimal
supergravity scheme with universality of soft terms at the unification scale in which this
coupling is induced radiatively. As we will see this has important phenomenological
implications as for the behaviour of the R-parity violating stop decays with respect to
tan . We calculate its magnitude using a set of RGEs in which the running of the
Yukawa couplings and soft breaking terms is taken into account. Here we also provide
the analysis of the relationship of the R-parity violating stop decays with the magnitude
of the tau neutrino mass. Motivated by the simplest oscillation interpretation of the
Super-Kamiokande atmospheric neutrino data, we also generalize the treatment of the Rparity violating decays by explicitly considering the case of light tau-neutrino masses, not
previously discussed.
For definiteness and simplicity we focus on supersymmetric models where the breaking
of R-parity is parametrized explicitly through a bilinear superpotential term of the type
`Hu [56,57]. The stop can have new decay modes such as
t1 b
(4)
due to mixing between charged leptons and charginos. We show that this decay may be
dominant or at least comparable to the ordinary R-parity conserving mode
t1 c 10 ,
(5)
where 10 denotes the lightest neutralino.

The paper is organized as follows. The model and an analytical analysis of the tree-level
tau-neutrino in terms of SUGRA parameters is described in Section 2. The mass matrices
are given in Section 3 while in Section 4 we present the top squark decay widths in the
minimal supergravity model with universal soft SUSY breaking terms [45,46], MSUGRA,
for short. The relevant Feynman rules and the squark decay widths and branching ratios
are calculated in Appendix A. They are studied numerically in Section 5 and we present
our conclusions in Section 6.
185
2. The model
The supersymmetric Lagrangian is specified by the superpotential W given by
ij a

ij bb b ba
ba bb
bi U
bj H
bb + hij Q
bb b ba
W = ab hU Q
2
D i Dj H1 + hE Li Rj H1 H1 H2
bai H
bb ,
+ ab i L
2
(6)
where i, j = 1, 2, 3 are generation indices, a, b = 1, 2 are SU(2) indices, and is a

completely antisymmetric 2 2 matrix, with 12 = 1. The symbol hat over each letter
b1 , and H
b2 being SU(2) doublets with hypercharges
bi , b
Li , H
indicates a superfield, with Q
1
b
b
b
3 , 1, 1, and 1 respectively, and U , D, and R being SU(2) singlets with hypercharges
4 2
3 , 3 , and 2, respectively. The couplings hU , hD and hE are 3 3 Yukawa matrices, and
and i are parameters with units of mass.
Supersymmetry breaking is parametrized by the standard set of soft supersymmetry
breaking terms
ij 2 ea ea
ij 2 e e
ij 2 e e
ij 2 ea ea
ij 2 e e
Vsoft = MQ Q
i Qj + MU Ui Uj + MD Di Dj + ML Li Lj + MR Ri Rj
+ m2H1 H1aH1a + m2H2 H2a H2a

12 M3 3 3 + 12 M2 2 + 12 M 0 1 1 + h.c.
ij ij a
ij ij eb e
a
ei U
ej H b + Aij hij Q
eb e a
+ ab AU hU Q
2
D D i Dj H1 + AE hE Li Rj H1

La H b .
BH a H b + Bi i e
1
(7)
For definiteness and simplicity we assume only R-parity Violation (RPV) in the
third generation, neglecting the effects of RPV on the two first families, adopting the
superpotential [5872]
b3 U
b3 H
b2 + hb Q
b3 D
b3 H
b1 + h b
b3 H
b1 H
b2 + 3 L
b3 H
b2
b1 H
L3 R
W = ht Q
(8)
to describe the R-parity violating violating decay mode t1 b . In this case we will omit
the labels i, j in the soft breaking terms given above. Note that the bilinear term 3 can
not be rotated away, since the rotation that eliminates it reintroduces an R-parity violating
trilinear term, as well as a sneutrino vacuum expectation value. Notice that, in contrast with
Ref. [10,11] where the doublet sneutrino vev in the bilinear model is much more loosely
constrained, in this case it is not subject to constraints from astrophysics.
Note, in contrast, that in order to describe Flavour Changing Neutral Current (FCNC)
effects such as the R-parity conserving process t1 c 10 we need the three generations of
quarks.
The above model can be described in various equivalent bases, for example:
1. one in which bilinear term and sneutrino vev are non-zero, 3I 6= 0 and v3I 6= 0
[3,4,44],
II
2. one in which trilinear term 1 and sneutrino vev are non-zero, II
3 6= 0 and v3 6= 0
[73],
1 In the one generation case there is only one trilinear RPV term in the superpotential written in our notation as
b3 D
b3 b
L3 .
3 Q
186
III
3. the vev-less basis in which 3III and III
3 are non-zero but v3 = 0 [74,75],
where the R-parity violating parameters can be expressed in terms of dimension-less basisindependent alignment parameters sin , sin 0 and sin 00 [71,72,7678] (X = I, II or III)
as follows:
sin =
v3II 3III
3X vdX + X v3X
=
= 0,
0 vd0
vd0
sin 0 =
X X
X X
II
I
3 + 3 hb
= 30 = 30 ,
0
0
hb
hb
(10)
sin 00 =
X X
vdX X
v3I
III
3 + v3 hb
3
=
=
,
vd0 h0b
vd0
h0b
(11)
where
h0b =
vd0 =
q
q
X
hX
b + 3 ,
vdX + v3X ,
0 =
(9)
X + 3X ,
X = I, II, or III.
(12)
Note that, in the notation of Eqs. (9)(11), the parameters 3 and appearing in Eq. (8)
should bear the superscript I.
Of these parameters only two are independent because they satisfy
sin 00 = cos 0 sin sin 0 cos .
(13)
In the limit when the R-parity violating parameters vanish one recovers the MSSM.
From now on we will work in the I3 = 0 basis, unless otherwise stated. As a result we will
omit the label I in all the parameters associated with this basis. We also will drop out the
prime in hb . One of the advantages in working in this basis is that the RGEs evolution
does not induce the trilinear R-parity violating terms neither in the superpotential nor in
the scalar potential if at the beginning we impose universality [71,72].
It is convenient to introduce the following notation in spherical coordinates for the
vacuum expectation values (vev):
vd = v sin cos ,
vu = v sin sin ,
v3 = v cos
(14)
which preserves the standard MSSM definition tan = vu /vd . In the MSSM limit, where
3 = v3 = 0, the angle is equal to /2. This makes sense in the I3 = 0 basis where the
usual MSSM relation
1
(15)
mb = hb vd
2
holds.
In this model the presence of RPV induces a mass for the tau-neutrino at the tree level
[79], as well as radiative masses to the the e and . As already mentioned it is sufficient
for our present discussion of stop decays to keep only the tau-neutrino.
187
In order to study the mass it is convenient to have an analytical expression for m in

this limit. The tree level tau-neutrino mass may be expressed as [6,1218,44,7188]
m
(g 2 M 0 + g 02 M)0 2
4MM 0 0 2
2(g 2 M 0
vd0 sin2
2
+ g 02 M)0 vu vd0
cos
(16)
in terms of basis-independent parameters 0 , vd0 and sin defined in Eqs. (12) and (9).
The second term in the denominator may be neglected if M, >
mZ , as often happens in
minimal supergravity models with universal soft SUSY breaking terms (see, for example,
[45,46]). Thus one may obtain an estimate of the neutrino mass by keeping only the first
term in the denominator.
m
g2 0 2 2
v sin ,
2M d
(17)
where we have used M 0 = Mg 02 /g 2 . For sin 1 one can easily check that m could be
as large as the experimental upper bound of 18 Mev [89]. However in MSUGRA models
one may obtain naturally small sin values, calculable from the RGE evolution from the
unification scale down to the weak scale. Indeed, using the minimization equations sin
can be written in terms 1m2 = m2H1 m2L3 and 1B = B3 B [90] as
!
1m2 v2 0 1B
0
0
.
(18)
sin = cos sin cos 2 + 0
vd m0 20
m0 0

One may give a simplified approximate analytical expression for the tau-neutrino mass in
this model by solving the renormalization group equations for the soft mass parameters
m2H1 , m2L3 , B, and B3 in the one-step approximation. This gives [90]
" 2
#

2 + M 2 + A2
mH1 + MQ

MU
3
D
D
0
0
2

ln
sin 1m2 cos sin cos hb
mt
8 2
m0 2 0

MU
3
ln
(19)
cos 0 sin 0 cos h2b
2
mt
8
and
"
0
sin |1B cos sin
tan 0 h2b
0 A D
m0 2 0
#

MU
3
ln
,
8 2
mt
(20)
where we have denoted by the symbols sin |1m2 and sin |1B the two terms contributing
to sin in Eq. (18). Using these expressions and assuming no strong cancellation between
these terms one finds that the minimum neutrino mass is controlled by the sin |1m2 . As a
result one finds,

3
g 2 m2b
MU 2
2 0 2

sin hb
ln
.
(21)
m min
M
8 2
mt
The above approximate analytical form of the tau-neutrino mass is useful, as we will
see later (e.g., Eq. (42)) in order to display explicitly the degree of correlation between the
R-parity violating decays, such as t1 b , with the tau-neutrino mass.
188
Table 1
Estimated magnitude of R-parity violating parameters required for a tau-neutrino mass in the eV
range, without requiring cancellation in sin in the three bases defined before
Basis I: I3 = 0
Basis-independent
sin 0
sin
105
104
102
3 (GeV)
104
102
v3 (GeV)
1
103
Basis II: 3II = 0
Basis III: v3III = 0
II
3
v3II (GeV)
III
3
104
103
104
3III (GeV)
103
102
The minimum value for sin 0 hb is determined by the value sin 0 and that of tan . For
sin 0 1 and relatively small tan so that ht is perturbative, one has
(22)
m >
10 keV
for M 1 TeV. In order to get smaller masses one needs to suppress sin2 0 additionally,
for example, to reach 1 eV the required R-parity violating parameters are given in Table 1.
These order-of-magnitude estimates are given in terms of the basis-independent angles
and 0 , and in the relevant parameters for the three bases defined before.
Note that whenever the parameter has two values, the first correspond to tan = 2 (the
lower perturbativity limit) and the second to tan = 35. In Table 1, sin was estimated
from Eq. (17) and sin 0 from Eq. (21).
Note that the RGE-induced suppression depends basically in the h2b factor in Eq. (19)
which is 103 ( 1) for small (large) tan . As a result the bigger the value of tan ,
the smaller sin 0 will have to be for a fixed tau neutrino mass. The RPV parameters in the
several bases were estimated from Eqs. (9)(11) and (13).
In Eq. (21) we have neglected 1B contribution with respect to the one coming from
1m2 . It is possible, however, that the 1B term may be sizeable. In the large 1B case
then it may cancel the 1m2 contribution in sin , leading to an additionally suppressed
neutrino mass. As we will see, however, in SUGRA models with universal soft terms at the
unification scale (SUGRA for short) we do not need any substantial cancellation in order
to obtain masses below the electronvolt scale.
3. Squark mass matrices

The up and down-type squark mass matrices of our model have already been given
previously in Ref. [44]. Here we generalize those to the three-generation case. The mass
matrix of the up squark sector follows from the quadratic terms in the scalar potential

2 uL
eU
+
(23)
Vquadratic = u L uR M
uR
given by
eU2 =
M
2
MQ
+ 12 vu2 hU hU + U L
1 vu Ah
U
2
1 vu Ah 1 (vd 3 v3 )hU
U
2
2
1 (vd 3 v3 )hU
MU2 + 12 vu2 hU hU + U R
2
, (24)
189

where U L = 18 (g 2 13 g 0 2 ) vd2 vu2 + v32 1 and U R = 16 g 0 2 (vd2 vu2 + v32 )1 are the
splitting in the squark mass spectrum produced by electro-weak symmetry breaking, and
ij ij
e 2 are
AhU ij AU hU . The eigenvalues of M
U
"
#

2 U L
eU
.
(25)
diag {mu 1 , mu 2 , . . . , mu 6 } = U L U R M
U R
This way the six weak-eigenstate fields u iL and u iR (i = 1, 2, 3) combine into six up-type
k = Uki
k , u iR = Uik
k = Uki
k.
mass eigenstate squarks u k as follows: u iL = Uik
Lu
Ru
Lu
Ru
For completeness, we also give the mass matrix of the down squark sector. The quadratic
scalar potential includes
"
#
2 d L
e
+
(26)
Vquadratic = dL dR MD
dR
given by
2
eD
=
M
2
MQ
+ 12 vd2 hD hD + DL
1 vd Ah 1 vu hD
D
2
2
2
1
1
vu hD
MD + 2 vd2 hD hD + DR
2
,
(27)

1 02 2
g (vd vu2 +v32 )1, and AhD ij
where DL = 18 g 2 + 13 g 0 2 vd2 vu2 +v32 1, DR = 12
ij ij
e2 are
AD hD . The eigenvalues of M
D
" #

2 DL
eD
diag {md1 , md2 , . . . , md6 } = DL DR M
.
(28)
DR
1 vd Ah
D
2
One is left with six mass-eigenstate down squarks fields dk related to diL and diR fields as
ik
ki d , d = ik d = ki d .
follows: diL = DL
dk = DL
k iR
UR k
DR k
For the Higgs-slepton part of the quadratic scalar potential in the one generation case of
the Bilinear R-parity Violating (BRpV) model, see Refs. [91] and [56,57].
Of particular interest to us is the chargino/tau mass matrix. For our present purposes it
is sufficient to have the form of this matrix for one generation, which is given by
1 gv2
M
0
2
1
1 h v3
(29)
MC =
2 gvd
.
2
1 gv3
1 h vd
3
2
This form is common to all models with spontaneous breaking of R-parity, as well as in
the simplest truncation of these models provided by the BRpV model considered here. We
note that the chargino sector decouples from the tau sector in the limit 3 = v3 = 0. As in
the MSSM, the chargino mass matrix is diagonalized by two rotation matrices U and V
m
0
0
1
(30)
U MC V 1 = 0
m 0 .
2
The lightest eigenstate of this mass matrix must be the tau lepton ( ) and so the mass is
constrained to be 1.77705+0.29
0.26 GeV. To obtain this the tau Yukawa coupling becomes a
190
function of the parameters in the mass matrix, and the full expression is given in [91]. The
composition of the tau is given by
R+ = V3j j+ ,
L = U3j j ,
(31)
e1 , 0+ ) and T = (i , H
e2 , 0 ). The two-component Weyl
where +T = (i+ , H
2 R
1 L
0
0+
+
spinors R and L are weak eigenstates, while R and L are the mass eigenstates. It
follows easily from Eq. (30) that the matrix MC MCT is diagonalized by U and the matrix
MCT MC is diagonalized by V .
The soft SUSY breaking parameters at the electroweak scale needed for the evaluation
of the mass matrices and couplings are calculated by solving the renormalization group
equations (RGEs) of the MSSM and imposing the radiative electroweak symmetry
breaking condition. From the measured quark masses, CKM matrix elements and tan
we first solve one-loop RGEs for the gauge and Yukawa couplings to calculate their
corresponding values at the unification scale. Assuming now universal soft supersymmetry
breaking boundary conditions, we evolve downwards the RGEs for all MSSM parameters,
including full three-generation mixing in the RGEs for Yukawa coupling constants, as well
as soft SUSY breaking parameters. Next, we evaluate the Higgs potential at the mt scale
including the one-loop corrections induced by the Yukawa coupling constants of the third
generation. The radiative electroweak symmetry breaking requirement fixes the magnitude
of the SUSY Higgs mass parameter and the soft SUSY breaking parameters B and B3 .
Notice that due to the third minimization condition one can solve for v3 as a function of 3 .
At this point, all RPV parameters at the electroweak scale are determined as functions of
the input parameters (tan , m0 , A0 , m1/2 , sign(), 3 ). Iteration is required because and
3 are inputs to evaluate the loop-corrected minimum. Having determined all parameters at
the electroweak scale, we obtain the masses and the mixings of all the SUSY particles by
diagonalizing the corresponding mass matrices. At this stage we also choose 3 in order to
get a sufficiently light tau-neutrino.
We scan the soft SUSY breaking parameter space in the range
m0 6 700 GeV,
100 GeV < m1/2 6 400 GeV,
|A0 | 6 1000 GeV,
(32)
|3 | < 200 GeV,

1.8 < tan < 60
the previous range on tan guarantee that both ht and hb will be perturbative. For the CKM
matrix, we use the Particle Data Group convention [92], taking Vus = 0.2205, Vcb = 0.041,
|Vub /Vcb | = 0.08 and neglecting CP violation, i.e., = 0. Notice that here we scan over a
much larger range for epsilon than used in Ref. [43].
The resulting region of lightest stop and chargino masses is displayed in Fig. 1.
Neglecting the three-body decays, we find that in Region I of the mt1 m + plane, BR(t1
1
c 10 ) + BR(t1 b ) 1. In Region II BR(t1 b ) + BR(t1 b i+ ) 1 (i = 1, 2).
In Region III BR(t1 b ) + BR(t1 b + i+ ) + BR(t1 t ) 1 (i = 1, 2), while in
191
Fig. 1. Kinematical regions in the mt1 m + plane. From left to right: Region I mt1 < m + + mb ;
1
1
Region II m + + mb < mt1 < mt ; Region III mt < mt1 < m 0 + mt ; and region IV mt1 > m 0 + mt .
1
b i+ )
region IV BR(t1 b ) + BR(t1

+ BR(t1 t ) + BR(t1
1 (j =
1, . . . , 4). Note that in each region the exact equality to 1 is reached when the FCNC
processes are fully included.
In Appendix A we give the Feynman rules for all vertices involving squarks, quarks,
charginos and neutralinos, as well as the two-body squark decay-widths, for squarks of all
three generations. These equations reduce to the expressions found in Ref. [93] provided
one identifies U33L = cos t and U33R = sin t. They also generalize the results for the
BRpV model to the three-generation case.
t j0 )
4. Lightest stop two-body decays in MSUGRA

In an R-parity conserving supergravity theory the main t1 decay channel expected
in region I of Fig. 1 is the loop-induced and flavour-changing t1 c 10 [4754].
As is well-known, the FCNC processes in the MSSM in general involve a very large
number of input parameters. For this reason, following common practice, we prefer to
perform the phenomenological study of flavour changing processes in the framework of a
supergravity theory with universal supersymmetry breaking. The simplest description of
FCNC processes in R-parity conserving minimal SUGRA models uses the so-called onestep approximation. Here we start by reproducing the standard calculation for t1 c 10 as
in [47]. To do this consider only the effect of the third generation Yukawa coupling. From
our general Eq. (A.10) we have for t1 = u l

2
0
g2
N12
1 2 2
2
0
(U L13)2 sin W N11
sin W
t1 c 10
+
8
3
2 3
cos W
!2
2
m 0
mt1 1 21
(33)
mt
1
192
with
U L13 =
L cos t R sin t
.
mc2L m2t
(34)
In the one-step approximation L , R are given by

tU
2
2
2
eU2
MQ
Kcb Kt b h2b MQ
+ MD
+ m2H1 + A2b ,
L = M
23
23
2
16

tU
1
2
h
2
e
Kcb Kt b hb mt Ab + At
R = MU 26 AU 23
16 2
2
(35)
(36)
with tU = ln(MG /mt ). So, in the one-step approximation we have

2

2
0
4
2
(tan W N11 + 3N12 )
t1 c 1 F hb (m2 cos t A sin t)
0
6
!2
2
m 0
mt1 1 21 ,
mt
(37)
where the pre-factor F =

by
m2 =
0
g2
tU
2
16 ( 16 2 Kcb Kt b )
2 + M 2 + m2 + A 2
MQ
D
H1
b
mc2L m2t
6 107 and the parameter m2 is given

0
(38)
is basically independent of the initial conditions due to the m0 dependence both in the
numerator as in the denominator and
A =
mt (Ab + 12 At )
mc2L m2t
(39)
Note however, that the one-step approximation includes only the third generation
Yukawa couplings and neglects the running of the soft breaking terms [4755]. Such an
approximation is rather poor for our purposes, since we will be interested in comparing
with R-parity violating decay modes (see the next section). In order to have an accurate
calculation of the respective branching ratios we need to go beyond the one-step
approximation. We therefore use a exact numerical calculation for the FCNC process t
c 10 in which the running of the Yukawa couplings and soft breaking terms is taken into
account. First we have checked that indeed the effect of the Yukawas from the two first
generations is negligible. However the same is not true for the running of the soft breaking
terms. As can be seen from Fig. 2 the range of variation that we obtain from the numerical
solution is

(40)
t1 c 10 1016106 GeV
depending on the assumed value of m1/2 and tan . In this figure we have compared the
decay width obtained from Eq. (A.10) with the approximate formulae in Eq. (33) for two
fixed values of m1/2 , tan and taking A0 = 0. The approximate formulae only reproduce
well the numerical result for the academic case of no SUSY breaking gaugino mass,
193
Fig. 2. Comparison between the exact numerical calculation (ordinate) and the one-step approximation (abscissa) for the t1 c 10 decay width for various values of tan and m1/2 with A0 = 0 and
m0 varying in the indicated range. The dotted left diagonal line would signify the equality between
the estimates, while the right diagonal line would indicate one order of magnitude difference. Results
of both estimates indicated in the lower right legend. More details are found in the text.
m1/2 = 0. For the more realistic case m1/2 > 100 GeV, the exact solution is usually one
decade smaller than the approximate one. In the one-step approximation (t1 c 10 )
can be arbitrarily small if the two terms m2 cos t and A sin t in Eq. (37) cancel. This
0
behaviour can be illustrated in Fig. 2 by the dashed line labelled 358, which corresponds
to m0 = 358 GeV. One sees clearly that while the approximate solution goes to zero, the
numerical one reaches a minimum value around 1011 GeV. The wrong behaviour of the
approximate solution indicates that the A depends strongly on the scale. For example, the
RGE for Ab is very sensitive on m1/2 and tan and in the one-step approximation there
is no explicit dependence on m1/2 , which is crucial. Both solutions increase with tan ,
as expected by the bottom Yukawa dependence explicit in Eq. (37) and remain practically
constant for large m1/2 values.
5. Two-body decays of the lightest stop: the R-parity violating case

In contrast to the case of an R-parity conserving supergravity theory, in our broken Rparity model one can have a competing R-parity violating stop decay mode in region I of
Fig. 1. From Eq. (A.11) with = F3+ one can easily compute the R-parity violating stop
decay width t1 b ,
(t1 b ) =
g 2 1/2 (m2t , m2b , m2 )

1
16m3t
1
hb ct V32
ht st V31
ct mb m
4U32

2

2 2 2
h t st V31
hb ct m2t m2b m2
+ V32
ct + U32
1
(41)
194
which coincides with the result found in Ref. [43]. In [94] it was shown that, except for U32
which determines the SU(2)-conserving mixing of the Higgsino with the left-handed , all
other mixing matrix elements V3i and U3i are proportional to v3II and therefore to the tauneutrino mass. Neglecting these terms we have from Eq. (41)
2 1/2
2
2
2
g (mt1 , mb , m ) 2 0 2 2 2
sin h b ct mt m2b m2
t1 b
3
1
16mt
(42)
noting that, to a good approximation,

3
|U32 | 0 = | sin 0 |,
(43)
where 3 corresponds to the bilinear mass parameter in basis I. The lesson here is that the
R-parity violating decay rate (t1 b ) is proportional to 3 or, equivalently, to sin2 0 ,
instead of sin2 , and thus not necessarily small, since it is not directly controlled by the
neutrino mass. In other words, there can be cancellations in the latter but not in the R-parity
violating branching ratio.
The meaning of the factor sin2 0 h2b may also be seen in basis II, where 3II = 0. In this
case v3II is proportional to the tau-neutrino mass so that, as already mentioned, in this basis
all the elements U3i and V3i are small [71,72]. Neglecting these terms, (t1 b ) may
be written directly from the interaction term tL bR L , which is induced by the trilinear term
in the 3II = 0 basis given in Eq. (10) as

0
0
(44)
II
3 = 3 / hb = hb sin
which is the factor in Eq. (42). Note, however, that in our numerical calculation to be
described in the next section we have used for (t1 b ) the full expression given in
Eq. (A.10) of the appendix.
In the next section we will determine the conditions under which the R-parity violating
decay width (t1 b ) can be dominant over the R-parity conserving ones, (t1
c 10 ) and (t1 b 1+ ).
5.1. Region I
Using the one-step approximation for (t1 c 10 ) one finds
!
m2 0 2

1
0
6
4
.
t1 c 1 10 hb mt1 1 2
mt
(45)
Using the Eq. (42) and neglecting charm, tau and bottom masses we get
!
m2 0 2
2
(t1 c 10 )
1
5 hb
10
.
1 2
(t1 b )
m
sin2 0
(46)
t1
Therefore (t1 c 10 ) will start to compete with (t1 b ) from sin 0 <
5
103 (104 ) for tan large (small). In Fig. 3 we compare BR(t1 c 10 ) [calculated
195
Fig. 3. Regions where the t1 b decay branching ratio exceeds 90% in the mt1 m 0 plane for
1
different m values. The MSSM parameters are randomly varied as indicated in the text under the
restriction mt1 < m + mb . The upper-left triangular region corresponds to mt1 < m 0 + mc
1
so that only the t1 b decay channel is open. The lower-right unshaded region corresponds to
mt1 > m + + mb .
1
numerically from their exact formula in (A.10)] with BR(t1 b ) within the restricted
region of the mt1 m 0 plane where only those two decay modes are open. We consider
1
different m values (these correspond to relatively small values of the R-parity parameters
|3 |, |v3 | <
1 GeV). We vary the MSSM parameters randomly obeying the condition mt1 <
m + mb and depict the corresponding region in light grey. The upper-left triangular
1
region is defined by kinematics and corresponds to mt1 < m 0 + mc , so that BR(t1
1
b ) = 100%. The lower-right grey corresponds to mt1 > m 0 + mc when the sampling is
1
done over the region defined by Eq. (32). One notices from Fig. 3 that in the central region
the dominant stop decay mode is t1 b with branching ratio BR(t1 b ) > 0.9. The
dotted lines in the light grey region indicate maximum mass values obtained in the scan.
In the calculation of the mass, we have allowed only up to one order of magnitude of
cancellation between the two terms which contribute to sin . Therefore if the lightest stop
only decays into the two modes considered here, the processes t1 b , will be important
even for the case of very light tau-neutrino masses.
We note however that we can use the limits obtained from leptoquark searches [95]
in order to derive limits on the top-squark for our R-parity violating case. In particular,
if BR(t1 b ) = 1 stop masses less than 99 GeV are excluded at 95% of CL, under
the assumption that the three-body decays of the stops are negligible. Therefore, the dark
region in Fig. 3 would be ruled out. In Ref. [96] we have determined the corresponding
restrictions on the SUGRA parameter space.
The dependence on the tau-neutrino mass may be seen in Fig. 4 where the role played
by tan is manifest. In this figure we have shown BR(t1 b ) as function of the lighter
196
Fig. 4. BR(t1 b ) as function of the lighter stop mass for tau-neutrino mass in the sub-eV range
and two different values of tan and 3 /. This prediction is natural in the sense that we have
allowed only up to one order of magnitude of cancellation between the two terms that contribute to
sin .
stop mass for tau-neutrino mass in the sub-eV range, indicated by the simplest oscillation
interpretation of the Super-Kamiokande atmospheric neutrino data. We have obtained such
tau-neutrino mass values numerically, allowing only one decade of cancellation between
the two terms that contribute to sin in Eq. (18). The degree of suppression for 3 /
obtained numerically agrees very well with the expectations from the approximate formula
for the minimal tau-neutrino mass in Eq. (21). In contrast with Ref. [43], in our case
BR(t1 b ) decreases with tan . The reason for this difference is that here we take into
account the fact that the mixing parameter U L13 obtained from the RGE depends on h2b
in Eq. (33), while in Ref. [43] was simply regarded as a phenomenological input parameter
(called there).
The message from this subsection is that in our SUGRA R-parity violating model the
R-parity violating decay mode t1 b can very easily dominate the R-parity conserving
decay mode t1 c 10 , even for very small neutrino masses.
5.2. Region II
In region II the R-parity conserving decay mode t1 b 1+ is open (but not t1 t),
and competes with the R-parity violating mode t1 b . Replacing the subindex 3 by 1 on
the diagonalization matrices U and V in Eq. (41) we get the corresponding expression
for (t1 b 1+ ). In order to get an approximate expression for the ratio of the two
main decay rates in this region, we note that in MSUGRA the lightest chargino is usually
2 1. In addition, the lightest stop is usually right-handed,
gaugino-like, implying that V11
2
2
hence sin t >
cos t. This way we find
197
Fig. 5. Contours of BR(t1 b ) > BR(t1 b1+ ) in the mt1 m + plane for |v3 | < 10 GeV. Three
1
different maximum values for |3 | are considered: |3 | < 40 GeV (dot-dash), |3 | < 60 GeV (dots),
and |3 | < 80 GeV (dashes). The region where mt1 < mb + me+ corresponds to the previously
1
studied Region I.
sin2 0 h 2b cos2 t
(t1 b )
( )
K, (47)
+
+
cos V h
2 2
2
2
( 1 ) (t1 b 1 ) [(V11
t
12 t sin t) + U12 hb cos t]
where K is a kinematical factor depending on the lightest stop and chargino masses, and
here we have defined h t,b ht,b /g. The presence of the bottom quark Yukawa coupling
indicates that large values of tan are necessary to have large R-parity violating branching
ratios in this region. In fact, we have checked numerically with the exact expressions that
in Region II (RII) ( )/ ( 1+ ) >
1 only for large tan as we will see in the next figures.
In Fig. 5 we show the regions in the m mt1 plane where BR(t1 b ) dominates
1
1+ ). In the upper-left region the decay mode t1 be
1+ is not allowed and
over BR(t1 be
corresponds to Region I. Below and to the right of this zone, and above and to the left of
three rising lines, lies region RII where ( )/ ( 1+ ) > 1. The three lines correspond
to |3 | < 80 GeV (dashed), |3 | < 60 GeV (dotted), and |3 | < 40 GeV (dot-dashed),
respectively. The proximity to the upper-left zone indicates that the RPV decay dominates
only close to the threshold where there is a high kinematical suppression of the R-parityconserving one, through the factor K. Unlike the case of region I this requires large values
of the RPV parameters. Note, moreover, that if the stops have a small mixing (cos t 0),
then ( )/ ( 1+ ) 1 in RII.
A simpler expression for the ratio of decay rates in Eq. (47) is obtained if we take V11 1
and assume no kinematical suppression in Eq. (47) through the factor K:
( )
sin2 0 h 2b .
( 1+ )
(48)
Note that the presence of the parameter sin 0 = 3 /0 indicates that the R-parity violating
decay mode is not strictly proportional to the neutrino mass, but proportional to the BRpV
parameter 32 .
198
Fig. 6. Regions for (t1 b )/ (t1 b 1+ ) as a function of the tau-neutrino mass with the
universality condition B = B3 at the unification scale imposed at the 0.1% level as indicated. Its
effect is to alter the maximum attainable tau-neutrino mass. The dot-dashed line corresponds to the
case where 1B = 0 at the weak scale.
However generically we expect some correlation with the mass, especially in the
case where the boundary conditions in the RGE are universal and there are no strong
cancellations between two terms that contribute to sin as shown in Fig. 6. In this figure we
plot the ratio ( )/ ( 1+ ) in RII as a function of the tau-neutrino mass. Both decay rates
have been calculated numerically from the exact formulas. In this figure we have imposed
both m2H1 = ML2 and B = B3 within 0.1% at the GUT scale. Cancellation between the 1m2
and 1B terms in the neutrino mass formula of Eq. (18) are accepted only within 1 decade.
As a reference we have drawn the line corresponding to 1B = 0 and 1m2 = 1m2min (1m2
is negative and its magnitude is bounded from below by 1m2min ) at the weak scale, which
gives an idea of the value of the neutrino mass when there is no cancellation between the
1B and 1m2 terms.
We have imposed an upper bound on m at the collider experimental limit of the tauneutrino mass, and have chosen fixed values of 3 / = 1, 0.1, and 0.01. The allowed region
for 3 / = 1 is above the dashed line. In the case of 3 / = 0.1 (0.01) the allowed region
lies enclosed between the solid (dotted) lines. The effect of tan is to increase the ratio
( )/ ( 1+ ): the minimum value of the ratio is obtained for tan 2 and the maximum
corresponds to tan 60. The extreme values of tan are dictated by perturbativity.
A number of statistically less significant points appear outside the drawn regions in Fig. 6
and are not depicted. They correspond to points with mt1 mb m < 10 GeV which
1
appear above the diagonal line, and points with cos t < 0.1 which appear below the horizontal line corresponding to the lowest values of tan . In the last case, our approximation
in Eq. (48) does not work any more. On the other hand, Eq. (48) predicts very well the be+
0
havior
of ( )/ ( 1 ) if cos t > 0.1. For example for 3 / = 1, or equivalently sin =
1/ 2, we expect from Eq. (48) a maximum value of order 1 for large tan (hb 1) and
199
Fig. 7. Regions for (t1 b )/ (t1 b 1+ ) as a function of the tau-neutrino mass for different
levels of cancellation between the two terms that contribute to the neutrino mass. We impose the
2 at the unification scale, but B and B are not universal. We take
universality condition m2H = ML
3
1
|3 /| = 1 (inside the dashed lines), |3 /| = 0.1 (solid lines), and |3 /| = 0.01 (dotted lines).
a minimum value of order 103 for small tan (h2b 103 ), and this is confirmed by
Fig. 6. High values of the R-parity violating branching ratio for large 3 values are highly
restricted for large tan . This can be understood as follows. In the case of 3 / = 1 and
tan = 60 acceptable neutrino masses are obtained only if sin 1. On the other hand, in
this regime we find from Eq. (18) that the 1B term is large because of the high value of
tan , and that the 1m2 term is large because m2H1 becomes negative and 1m2 = m2H1
ML2 grows in magnitude. This way, acceptable neutrino masses are achieved only with cancellation within more than one decade. In any case, we think that Fig. 6 is very conservative
considering that in MSSM-SUGRA with unification of top-bottom-tau Yukawa couplings,
the large value of tan implies that a cancellation of four decades among vevs is needed.
The width of the band in Fig. 6 reflects the degree of correlation between the ratio
(t1 b )/ (t1 b 1+ ) and the neutrino mass under the mentioned conditions.
Note that one would have an indirect measurement of the neutrino mass if this ratio
were determined independently. The band will open to the left if one allows a stronger
cancellation between the terms in 1B and 1m2 . On the other hand it will open to the right
if the universality between B and B3 is relaxed. This is shown in Fig. 7 where we plot
the ratio ( )/ ( 1+ ) in RII as a function of the tau-neutrino mass, but without imposing
universality between B3 and B. If we accept cancellation within one decade between the
1B and 1m2 terms , then the allowed region is at the right and below the corresponding
dashed tilted line. If a larger degree of cancellation is accepted, the left boundary of the
allowed region moves to the left as indicated in the figure, enhancing the R-parity violating
channel. In addition if we accept only a decade of cancellation between the two terms
that contribute to the tau-neutrino mass, then our approximate formula which predicts the
minimum tau-neutrino mass in Eq. (21) works very well.
200
In summary, in this subsection we have shown that even in region II, where the Rparity conserving decay mode t1 b 1+ is also open, the R-parity violating decay mode
t1 b can be comparable to t1 b 1+ for large tan and 3 , and relatively close to
the chargino production threshold. In general, this implies a large neutrino mass unless a
cancellation is accepted between the two terms contributing to the tree level neutrino mass.
In addition, the non-universality of the B and B3 terms at the GUT scale does not increase
appreciably the allowed parameter space, except at large tan . The main consequence
of this non-universality is to restrict the allowed values of A0 at large tan . In the next
subsection we study the effects introduced by the non-universality of m2H1 and m2L3 .
5.3. Effects of non-universality
We now study the effect of possible non-universality of soft-breaking SUSY parameters
on our previous results. In particular, the non-universality between m2H1 and m2L3 at
the GUT scale. The Minimal SUGRA model, while highly predictive, rests upon a
number of simplifying assumptions which do not necessarily hold in specific models
due to the possible evolution of the physical parameters in the range from MPlanck to
MGUT . Specifically, there are several models in the literature with non-universal soft
SUSY breaking mass parameters at high scales. A recent survey can be found in [97],
where several models such as based on string theory, M-theory, and anomaly mediated
supersymmetry are analyzed. For this reason we find interesting to explore here the effects
of non-universal soft terms.
The SUGRA spectra are typically found for given values of m1/2 , m0 , A0 , tan and
Sgn(). In our case we have in addition sin 0 (or equivalently, 3 ). The value of v3 is
determined by the previous parameters through the minimization conditions. In addition,
a relation between A0 and the ratio B3 /B at the GUT scale (which indicates the degree
of universality) emerges. This relation can be seen in Fig. 8 for 3 / = 1 and the values
tan = 3, 40, and 60, for m2H1 = ML2 . The relation becomes more restrictive as tan is
increased, starting from 1000 < A0 < 1000 GeV allowed for tan = 3, to a single A0
value compatible with unification for tan = 60.
Another way to enhance the R-parity violating channel, enlarging the band towards the
left in Fig. 6, is by relaxing the universality between m2H1 and ML2 at the GUT scale. In
Fig. 9 we plot the ratio m2H1 /ML2 at the weak scale as a function of the same ratio at the
unification scale MGUT for tan = 3. The shaded region is allowed, implying a maximum
value for the ratio m2H1 /ML2 at the weak scale for a given value of the ratio at the GUT
scale. We see from Fig. 9 that a relaxation of universality of 0.5% or more is enough to
make (m2H1 /ML2 )weak = 1 possible, meaning that smaller neutrino masses are attainable
without having to rely on a cancellation between the 1m2 and 1B terms or small values
of (t1 b ). However as we increase tan the maximum value of m2H1 /ML2 decreases,
and thus, the required non-universality between mH1 and ML at unification scale grows
drastically. In Fig. 10 we show the ratio m2H1 /ML2 at the weak scale as a function of tan .
We appreciate clearly the growing of |1m2 |min with tan . We remind the reader that this
kind of non-universality in the soft terms is not uncommon in string models [98], or GUT
201
Fig. 8. Universality condition B = B3 at the unification scale as a function of A0 . As tan increases,

the allowed values of A0 are more constrained.
2 at the weak and the unification scales for tan = 3.

Fig. 9. Comparison between the ratio m2H /ML
1
2
2
Universality at the unification scale, mH /ML = 1, implies a maximum value for this ratio at the
1
weak scale.
models based on SU(5) [99] or SO(10) [100,101] for example. There are in fact some
SO(10) models for non-universality of the GUT scale scalar masses which naturally favour
light neutrino mass [100,101].
The effect of non-universality it is also explored in Fig. 11 where it is shown the relation
between the neutrino mass and the parameter sin for 3 / = 1. Two different bands are
202
2 ) evaluated at the weak scale as a function of tan . This ratio is always less than
Fig. 10. (m2H /ML
1
one and decreases with tan .
Fig. 11. Minimum value of the tau-neutrino mass as a function of sin for different values of
mH1 /ML at the GUT scale and two values of tan . The ratio 3 / is fixed to the indicated value,
leading to a nearly constant value for (t1 b )/ (t1 b 1+ ). Here we assume that the two
terms contributing to the tau-neutrino mass cancel to within an order of magnitude.
shown: one for tan = 3 and ( )/ ( 1+ ) = 2 103 , and a second one for tan = 46
and ( )/ ( 1+ ) = 0.4 0.2. The required degree of universality at the GUT scale is
indicated inside the bands. For example, in order to have neutrino masses of the order of
eV for tan = 3, m2H1 needs to be at least 0.2% larger than ML2 . Similarly, for tan = 46
we need a m2H1 twice as large as ML2 at the GUT scale in order to have neutrino masses of
203
1 eV. We stress the fact that for Fig. 11 we have conservatively accepted cancellation at the
level of one order-of-magnitude only.
In summary, the lesson to learn here is that non-universal soft SUSY breaking terms
at the GUT scale have the potential of making it easier to reconcile sizeable R-parity
violating effects in the stop sector with very small neutrino masses, without resorting to
cancellations.
6. Conclusions
We have studied the decays of the lightest top squark in SUGRA models with
and without R-parity. We have improved the calculation for the decay t1 c 0 by
numerically solving the renormalization group equations (RGEs) of the MSSM including
full generation mixing in the RGEs for Yukawa couplings as well as soft SUSY breaking
parameters. The decay-width is in general one order of magnitude smaller than the one
obtained in the usual one-step approximation. This result will therefore enlarge the regions
of parameter space where the four-body decays of lightest stop dominate over the decay
into a charm quark and the lightest neutralino. As a result it will affect the present
experimental lower bound on the t1 mass even in the R-parity conserving case [55]. If
R-parity breaks of course new decay modes appear and, as we have shown, they can
be sizeable. In fact we have shown that the lightest stop can be the LSP, decaying with
100% rate into a bottom quark and a tau lepton. We have shown that the decay mode
t1 b dominates over t1 c 0 even for neutrino masses in the range suggested by the
simplest oscillation interpretation of the Super-Kamiokande atmospheric neutrino data.
This result would have a strong impact on the top squark search strategies at LEP [102]
and TEVATRON [103], where it is usually assumed that the t1 c 0 decay mode is the
main channel. In addition to the signal of two jets and two taus present when the two
produced stops decays through the R-parity violating channel, one expects a plethora of
exotic high-multiplicity fermion events arising from neutralino decay, since such decay
can happen inside the detector even for the small neutrino masses in the range suggested
by the to oscillation interpretation of the atmospheric neutrino anomaly [10,11].
We have also compared the decay t1 b with the R-parity and flavor conserving mode
t1 b + and shown that the rate of the former can be comparable or even bigger than
the latter if the tau-neutrino mass and tan are large. However one may have a sizeable
branching of t1 b in the case of suppressed tree level neutrino mass as a result of
strong cancellations between the two terms that contribute to sin , or in some regions of
parameter space of non-universal SUGRA models with (m2H1 /m2L3 )GUT 6= 1. A detailed
analysis of the detectability prospects of such related signatures at present and future
accelerators lies outside of the scope of the present paper and it will be taken up elsewhere.
Acknowledgements
We thank J. Ferrandis, O.J.P. Eboli and W. Porod for useful discussions. This work
was supported by DGICYT under grants PB95-1077 and by the TMR network grant
204
ERBFMRXCT960090 of the European Union. M.A.D. was supported by a DGICYT

postdoctoral grant, by the U.S. Department of Energy under contract number DE-FG0297ER41022, and by CONICYT grant 1000539. D.R. was supported by Colombian
COLCIENCIAS fellowship.
Appendix A. Feynman rules

In this appendix we derive the Feynman rules Fj0 qi qk (involving a neutralino/tauneutrino, a quark, and a squark) and Fj qi qk0 (involving a chargino/tau, a quark, and a
squark of different electric charge) in the case of three generations and RPV in the third
generation. This is a generalization of the Feynman rules contained in [104], which are
done for the R-parity conserving MSSM and for one generation of quarks and squarks.
0
, uL = Ku0L ,
Following [105] we work in a quark interaction basis where dL,R = dL,R
and uR = u0R (we denote q and q 0 the mass and current eigenstates respectively), as
opposed to Ref. [106] where a more general basis is used. In addition, we implement the
0
notation qL,R qL,R
for the interaction basis.
The starting point is the following piece of the Lagrangian

2 sin W eU N 0 J 1 +

1
eU sin2 W N 0 J 2 u iL
2

0U
mij
N 0 J 4 u j R PR FJ0
+
2 mW sin sin

1
+ g u 0i
2 sin W eU N 0 J 1 +
eU sin2 W N 0 J 2 u iR
cos W

0U
mij
N 0 J 4 u j L PL FJ0 + h.c.
2 mW sin sin
0i
LuuF
0 = g u
1
cos W
(A.1)
written in the quark interaction basis. The 5 5 matrix N 0 diagonalizes the neu H 0 , H 0 , ) basis as defined in [91], with the
tralino/neutrino mass matrix in the ( , Z,
1
2
index J = 1, . . . , 5. The 3 3 up-type quark mass matrix m0U is not diagonal, with the
indexes i, j = 1, 2, 3.
In order to write the above Lagrangian with mass eigenstates we use the basic relations
mentioned before, in particular, u0iL = (K )ij uj L , which implies that u 0iL = u j L K j i . We
need the following relations:
ki
u k ,
u 0iL u iL = u iL UL (K )

0
0U
U ki
u k ,
u iL mij u j R = u iL U R m
k,
u iR u iR = u iR Uki
Ru
j L = u iR U L K mU
u iR m0U
ij u
ki
u k
(A.2)
205
where i, j = 1, 2, 3 label the quark flavours, k = 1, . . . , 6 labels the squarks, and mU

diag {mu , mc , mt } is the diagonal up-type quark mass matrix. In this way, the Lagrangian
in Eq. (A.1) can be written as
j ki
j ki

j ki
j ki
i
2 G0U L + H0U R PR 2 G0U R H0U L PL Fj0 u k
LuuF
0 = g u
+ h.c.,
(A.3)
where the different couplings are

ki
1
1
j ki
0
2
0
eU sin W N j 2 U L K ,
G0U L = sin W eU N j 1 +
cos W 2

0
1
j ki
0
2
(A.4)
eU sin W N j 2 UkiR ,
G0U R = sin W eU N j 1 +
cos W
ki
j ki
H0U L = N 0 j 4 U L K h U ,
ki
j ki
H0U R = N 0 j 4 U R h U
and h U diag(mu , mc , mt )/( 2 mW sin sin ). Graphically, the Fj0 ui u k Feynman rules
are given by
The analogous Feynman rules in the MSSM are obtained by replacing Fi0 i0 , by
interpreting the matrix N as the usual 4 4 neutralino mixing matrix, and by setting =
/2 in the formula for the Yukawa couplings.
in Eq. (A.1) and starting from
Similarly, replacing all u(u)
by d(d)

0U
mij
VJ 2 u j R VJ 1 u iL PR FJc
Lq q 0 F + = g di
2 mW sin sin
mD
ij
UJ2 u j L PL FJc
+ g di
2 mW cos sin

mD
ij
UJ 2 dj R UJ 1 diL PR FJ+
+ g u 0i
2mW cos sin
m0U
ij
0
+ g u i
(A.5)
VJ2 bj L PL FJ+ + h.c.
2mW cos sin
206
we can obtain the complete Feynman rules for the neutralino/tau-neutrino and chargino/tau
with quarks and squarks. The results, that complements the obtained in [105], are:
The mixing matrices G0D and H0D are defined as

0
1
j ki
0
2
ki
,
T3D eD sin W N j 2 DL
G0DL = sin W eD N j 1 +
cos W

1
j ki
ki
,
eD sin2 W N 0 j 2 DR
G0DR = sin W eD N 0 j 1 +
cos W
ki
j ki
H0DL = N 0 j 3 DL h D ,
ki
j ki
H0DR = N 0 j 3 DR h D .
(A.6)
Chargino/tau(d)quark(u)squark
where C is the charge conjugation matrix (in spinor space) and the mixing matrices GU
and HU are defined as
ki
j ki
j ki
HU L Uj2 U L h D ,
GU L Vj1 UkiL ,
j ki
HU R Vj2 (U R h U K)ki .
Chargino/tau(u)quark(d)squark
(A.7)
207
where the mixing matrices GD and HD are defined as

ki
ki
j ki
j ki
HDL Vj2 DL K h U ,
GDL Uj1 DL K ,
ki
j ki
HDR Uj2 DR h D K ,
(A.8)
In order to derive the decays widths we write, for example Eq. (A.3) as

j ki
j ki
0i fU PR + hU PL Fj0 u k + h.c.
LuuF
0 = gu
(A.9)
The result is
qk qi + Fj0
g 2 1/2 (m2qk , m2qi , m2 0 )

Fj
qk qi0

+ Fj =
16m3qk
j ki
j ki
(hQ )2 + (fQ )2
j ki

m2qk m2qi m2F 0 ,
j ki
j ki
4lQ HQL mqi0 mF
j ki

2

m2qk m2q 0 m2F ,
where Q = U, D refers to q and

j ki
j ki
j ki
fQ = 2 G0QL + H0QR ,
j ki
j ki
j ki
hQ = 2 G0QR H0QL,
j ki
(A.10)
+ (lQ )2 + (HQL )
j ki
j ki
g 2 1/2 (m2qk , m2q 0 , m2F )

16m3qk
j ki
4hQ fQ mqi mF 0
j ki
lQ = HQR GQL
(A.11)
(A.12)
(A.13)
(A.14)
with the G and H couplings defined earlier in this appendix.

References
[1] H. Haber, G. Kane, Phys. Rep. 117 (1985) 75.
[2] H.P. Nilles, Phys. Rep. 110 (1984) 1.
[3] J.W.F. Valle, Supergravity unification with bilinear R parity violation, in: Proceedings of
PASCOS98, P. Nath (Ed.), World Scientific, hep-ph 9808292.
208
[4] J.W.F. Valle, in: Physics Beyond the Standard Model, lectures given at the VIII Jorge Andre
Swieca Summer School, Rio de Janeiro, February 1995 and at V Taller Latinoamericano
de Fenomenologia de las Interacciones Fundamentales, Puebla, Mexico, October 1995, hepph/9603307.
[5] C.S. Aulakh, R.N. Mohapatra, Phys. Lett. B 119 (1982) 136.
[6] L. Hall, M. Suzuki, Nucl. Phys. B 231 (1984) 419.
[7] G.G. Ross, J.W.F. Valle, Phys. Lett. 151 B (1985) 375.
[8] J. Ellis, G. Gelmini, C. Jarlskog, G.G. Ross, J.W.F. Valle, Phys. Lett. 150 B (1985) 142.
[9] A. Santamaria, J.W.F. Valle, Phys. Lett. 195B (1987) 423; Phys. Rev. D 39 (1989) 1780; Phys.
Rev. Lett. 60 (1988) 397.
[10] J.C. Romao, M.A. Diaz, M. Hirsch, W. Porod, J.W. Valle, Phys. Rev. D 61 (2000) 071703,
hep-ph/9907499.
[11] M. Hirsch, M.A. Diaz, W. Porod, J.C. Romao, J.W.F. Valle, hep-ph/0004115.
[12] S. Dimopoulos, L.J. Hall, Phys. Lett. 207B (1988) 210.
[13] E. Ma, D. Ng, Phys. Rev. D 41 (1990) 1005.
[14] V. Barger, G.F. Giudice, T. Han, Phys. Rev. D 40 (1989) 2987.
[15] T. Banks, Y. Grossman, E. Nardi, Y. Nir, Phys. Rev. D 52 (1995) 5319.
[16] M. Nowakowski, A. Pilaftsis, Nucl. Phys. B 461 (1996) 19.
[17] G. Bhattacharyya, D. Choudhury, K. Sridhar, Phys. Lett. B 349 (1995) 118.
[18] B. de Carlos, P.L. White, Phys. Rev. D 54 (1996) 3427.
[19] A Masiero, J.W.F. Valle, Phys. Lett. B 251 (1990) 273.
[20] J.C. Romo, C.A. Santos, J.W.F. Valle, Phys. Lett. B 288 (1992) 311.
[21] J.C. Romao, A. Ioannisian, J.W.F. Valle, Phys. Rev. D 55 (1997) 427, hep-ph/9607401.
[22] G. Giudice, A. Masiero, M. Pietroni, A. Riotto, Nucl. Phys. B 396 (1993) 243.
[23] M. Shiraishi, I. Umemura, K. Yamamoto, Phys. Lett. B 313 (1993) 89.
[24] I. Umemura, K. Yamamoto, Nucl. Phys. B 423 (1994) 405.
[25] P. Nogueira, J.C. Romo, J.W.F. Valle, Phys. Lett. B 251 (1990) 142.
[26] R. Barbieri, L. Hall, Phys. Lett. B 238 (1990) 86.
[27] M.C. Gonzalez-Garca, J.W.F. Valle, Nucl. Phys. B 355 (1991) 330.
[28] J. Romo, J. Rosiek, J.W.F. Valle, Phys. Lett. B 351 (1995) 497.
[29] J.C. Romo, N. Rius, J.W.F. Valle, Nucl. Phys. B 363 (1991) 369.
[30] J.C. Romo, J.W.F. Valle, Phys. Lett. B 272 (1991) 436; Nucl. Phys. B 381 (1992) 87.
[31] J.W.F. Valle, Phys. Lett. B 196 (1987) 157.
[32] V. Berezinskii, J.W.F. Valle, Phys. Lett. B 318 (1993) 360.
[33] A.D. Dolgov, S. Pastor, J.W.F. Valle, astro-ph/9506011.
[34] P. Abreu et al., DELPHI Collaboration, CERN-EP-99-049.
[35] J.-F. Grivaz, Rapporteur Talk, International Europhysics Conference on High Energy Physics,
Brussels, 1995.
[36] ALEPH Collaboration, Phys. Lett. B 373 (1996) 246260.
[37] H. Nowak, A. Sopczak, L3 Note 1887, Jan. 1996.
[38] S. Asai, S. Komamiya, OPAL Physics Note PN-205, Feb. 1996.
[39] B. Abbott et al., D0 Collaboration, hep-ex/9902013.
[40] D0 Collaboration, Phys. Rev. Lett. 75 (1995) 618, and references therein.
[41] CDF Collaboration, Phys. Rev. Lett. 69 (1992) 3439.
[42] UA2 Collaboration, Phys. Lett. B 235 (1990) 363.
[43] A. Bartl, W. Porod, M.A. Garcia-Jareno, M.B. Magro, J.W.F. Valle, W. Majerotto, Phys. Lett.
B 384 (1996) 151.
[44] M.A. Daz, J.C. Romao, J.W. Valle, Nucl. Phys. B 524 (1998) 23.
[45] S.P. Martin, in: G.L. Kane (Ed.), Perspectives on Supersymmetry, World Scientific, 1998, 88
pp., hep-ph/9709356.
209
[46] A. Bartl et al., in: D.G.Cassel, L.Trindle Gennari, R.H. Siemann (Eds.), Proceedings of the 1996
DPF/DPB Summer Study on New Directions For High-Energy Physics, 1997, pp. 693705; See
also http://www.slac.stanford.edu/pubs/snowmass96/.
[47] K.I. Hikasa, M. Kobayashi, Phys. Rev. D 36 (1987) 724.
[48] H. Baer, M. Drees, R. Godbole, J.F. Gunion, X. Tata, Phys. Rev. D 44 (1991).
[49] T. Kon, T. Nonaka, Phys. Lett. B 319 (1993) 355.
[50] T. Kon, T. Kobayashi, S. Kitamura, K. Nakamura, S. Adachi, Zeit. Physik C 61 (1994) 239.
[51] T. Kon, T. Nonaka, ITP-SU-94-02, hep-ph/9404230.
[52] W. Porod, T. Wohrmann, Phys. Rev. D 55 (1997) 2907.
[53] W. Porod, Phys. Rev. D 59 (1999) 095009.
[54] W. Porod, Ph.D. thesis, hep-ph/9804208, and references therein.
[55] C. Boehm, A. Djouadi, Y. Mambrini, hep-ph/9907428.
[56] F. de Campos, M.A. Garca-Jareo, A. Joshipura, J. Rosiek, J.W.F. Valle, Nucl. Phys. B 451
(1995) 315.
[57] H. Hempfling, Nucl. Phys. B 478 (1996) 3.
[58] M.A. Daz, J. Ferrandis, J.C. Romo, J.W.F. Valle, Phys. Lett B 453 (1999) 263.
[59] M.A. Daz, E. Torrente-Lujan, J.W.F. Valle, Nucl. Phys. B 551 (1999) 78.
[60] L. Navarro, W. Porod, J.W.F. Valle, Phys. Lett. B 459 (1999) 615, hep-ph/9903474.
[61] S. Roy, B. Mukhopadhyaya, Phys. Rev. D 55 (1997) 7020, hep-ph/9903418.
[62] A. Datta, B. Mukhopadhyaya, S. Roy, Phys. Rev. D 61 (2000) 055006, hep-ph/9905549.
[63] T. Feng, hep-ph/980650; hep-ph/9808379.
[64] C. Chang, T. Feng, Eur. Phys. J. C 12 (2000) 137, hep-ph/9901260.
[65] I.-H. Lee, Phys. Lett. 138B (1984) 121.
[66] A.S. Joshipura, M. Nowakowski, Phys. Rev. D 51 (1995) 21.
[67] S. Roy, B. Mukhopadhyaya, Phys. Rev. D 55 (1997) 7020.
[68] F.M. Borzumati, Y. Grossman, E. Nardi, Y. Nir, Phys. Lett. B 384 (1996) 123.
[69] E. Nardi, Phys. Rev. D 55 (1997) 5772.
[70] H.P. Nilles, N. Polonsky, Nucl. Phys. B 484 (1997) 33.
[71] J. Ferrandis, Phys. Rev. D 60 (1999) 095012, hep-ph/9810371.
[72] M.A. Diaz, J. Ferrandis, J.C. Romao, J.W.F. Valle, hep-ph/9906343.
[73] F. Borzumati, J. Kneur, N. Polonsky, Phys. Rev. D 60 (1999) 115011, hep-ph/9905443.
[74] Y. Grossman, H.E. Haber, Phys. Rev. D 59 (1999) 093008; Phys. Rev. Lett. 78 (1997) 3438.
[75] M. Bisset, O.C. Kong, C. Macesanu, L.H. Orr, Phys. Lett. B 430 (1998) 274; hep-ph/9811498;
hep-ph/9907359.
[76] S. Davidson, J. Ellis, Phys. Lett. B 390 (1997) 210.
[77] S. Davidson, J. Ellis, Phys. Rev. D 56 (1997) 4182.
[78] S. Davidson, Phys. Lett. B 439 (1998) 63.
[79] B. Mukhopadhyaya, S. Roy, F. Vissani, Phys. Lett. B 443 (1998) 191.
[80] V. Bednyakov, A. Faessler, S. Kovalenko, Phys. Lett. B 442 (1998) 203.
[81] O.C. Kong, Mod. Phys. Lett. A 14 (1999) 903.
[82] E.J. Chun et al., Nucl. Phys. B 544 (1999) 89.
[83] S.Y. Choi, E.J. Chun, S.K. Kang, J.S. Lee, Phys. Rev. D 60 (1999) 075002, hep-ph/9903465.
[84] A.S. Joshipura, S.K. Vempati, Phys. Rev. D 60 (1999) 111303, hep-ph/9903435; Phys. Rev.
D 60 (1999) 095009hep-ph/9808232.
[85] R. Adhikari, G. Omanovic, hep-ph/9802390.
[86] D.E. Kaplan, A.E. Nelson, JHEP 0001 (2000) 033, hep-ph/9901254.
[87] Y. Grossman, H.E. Haber, hep-ph/9906310.
[88] S. Rakshit, G. Bhattacharyya, A. Raychaudhuri, Phys. Rev. D 59 (1999) 091701.
[89] ALEPH Collaboration, Eur. Phys. J. C 2 (1998) 395.
[90] M.A. Daz, in: I. Antoniadis, L.E. Ibanez, J.W.F. Valle (Eds.), Proceedings of Valencia 97,
Beyond the Standard Model: From Theory to Experiment, World Scientific, p. 188, hepph/9802407.
210
[91] A. Akeroyd, M.A. Daz, J. Ferrandis, M.A. Garcia-Jareo, J.W.F. Valle, Nucl. Phys. B 529
(1998) 3.
[92] C. Caso et al., Eur. Phys. J. C 3 (1998) 1.
[93] A. Bartl, W. Majerotto, W. Porod, Zeit. Physik C 64 (1994) 499.
[94] A.G. Akeroyd, M.A. Daz, J.W.F. Valle, Phys. Lett. B 441 (1998) 224.
[95] F. Abe et al., CDF Collaboration, Phys. Rev. Lett. 78 (1997) 2906.
[96] F. de Campos, M.A. Diaz, O.J.P. Eboli, M.B. Magro, L. Navarro, W. Porod, D.A. Restrepo,
J.W.F. Valle, in: Proceedings Physics at Run II: QCD and Weak Boson Physics Workshop
Batavia, IL, 46 March 1999, hep-ph/9903245.
[97] H. Baer, M.A. Diaz, P. Quintana, X. Tata, JHEP 0004 (2000) 016, hep-ph/0002245.
[98] A. Brignole, L.E. Ibaez, C. Muoz, hep-ph/9707209.
[99] N. Polonsky, A. Pomarol, Phys. Rev. Lett. 73 (1994) 2292.
[100] H. Murayama, M. Olechowski, S. Pokorski, Phys. Lett. B 371 (1996) 57.
[101] H. Baer, M.A. Daz, J. Ferrandis, hep-ph/9907211.
[102] G. Abbiendi et al., OPAL Collaboration, Phys. Lett. B 456 (1999) 95.
[103] C. Holck, CDF Collaboration, Talk given at American Physical Society (APS) Meeting of
the Division of Particles and Fields (DPF 99), Los Angeles, CA, 59 January 1999, hepex/9903060.
[104] J.F. Gunion, H.E. Haber, Nucl. Phys. B 272 (1986) 1.
[105] S. Bertolini, F. Borzumati, A. Masiero, G. Ridolfi, Nucl. Phys. B 353 (1991) 591.
[106] J. Rosiek, Phys. Rev. D 41 (1990) 3464.

On the nonlinear KK reductions on spheres of

supergravity theories
Horatiu Nastase 1 , Diana Vaman 2
C.N. Yang Institute for Theoretical Physics, SUNY Stony Brook, NY 11794-3840, USA
Received 9 March 2000; accepted 4 April 2000
Abstract
We address some issues related to the construction of general KaluzaKlein (KK) anstze for the
compactification of a supergravity (sugra) theory on a sphere Sm . We first reproduce various anstze
for compactification to 7D from the ansatz for the full nonlinear KK reduction of 11D sugra on
AdS7 S4 . As a side result, we obtain a lagrangian formulation of 7D N = 2 gauged sugra, which
so far had only a on-shell formulation, through field equations and constraints. The AdS7 S4 ansatz
generalizes therefore all previous sphere compactifications to 7D. Then we consider the case when
the scalars in the lower dimensional theory are in a coset Sl(m + 1)/SO(m + 1), and we keep the
maximal gauge group SO(m + 1). The 11-dimensional sugra truncated on S4 fits precisely the case
under consideration, and serves as a model for our construction. We find that the metric ansatz has
a universal expression, with the internal space deformed by the scalar fluctuations to a conformally
rescaled ellipsoid. We also find the ansatz for the dependence of the antisymmetric tensor on the
scalars. We comment on the fermionic ansatz, which will contain a matrix U interpolating between
the spinorial SO(m + 1) indices of the spherical harmonics and the R-symmetry indices of the
fermionic fields in the lower dimensional sugra theory. We derive general conditions which the
matrix U has to satisfy and we give a formula for the vielbein in terms of U . As an application
of our methods we obtain the full ansatz for the metric and vielbein for 10D sugra on AdS5 S5
(with no restriction on any fields). 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.50.+h; 04.65.+e; 11.10.Kk; 11.25.Mj
Keywords: Supergravity; KK compactification; Nonlinear ansatz
1. Introduction
The problem of KaluzaKlein (KK) reduction of sugras on spheres is a very difficult
one. In general, one obtains a gauged sugra after KK reduction, and finding the nonlinear
Research supported by National Science Foundation Grant Phy 9722101.
1 hnastase@insti.physics.sunysb.edu
2 dvaman@insti.physics.sunysb.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 1 4 - 5
212
H. Nastase, D. Vaman / Nuclear Physics B 583 (2000) 211236
ansatz which realizes the embedding of this gauged sugra into the original model is quite
nontrivial. In the case of the reduction of 11D sugra on AdS4 S7 , parts of the ansatz where
found in [14]. In particular, the ansatz for the antisymmetric tensor was not explicitly
obtained (only in a highly implicit form, not suitable for calculations). The nonlinear ansatz
for the AdS7 S4 reduction of 11D sugra was found by the present authors together with
Peter van Nieuwenhuizen in [5,6], and was the first to give a complete solution for the
nonlinear KK ansatz for a sphere compactification. The ansatz for the compactification of
10D IIB sugra on AdS5 S5 is the most notable absence in this respect. Another problem
which has to be addressed is the consistency of the truncation (KK reduction). In [6],
the consistency of the KK reduction of an SU(8)-invariant version of the 11D sugra was
given. In [6], we gave a proof of consistency of the reduction of the usual 11D sugra on
AdS7 S4 . In this paper, we will not have anything to say about the consistency of the
anstze we will propose, except when we take various further truncations of our AdS7 S4
ansatz to compare to other work in the literature.
The recent interest in sphere compactifications came from the AdSCFT correspondence [720]. A lot of work has been done in finding ansatze for further truncations of
the models after the KK reduction on spheres. In [21,22] it was considered the further
truncation of the KK reduction of 11D and 10D IIB sugras on AdS4 S7 , AdS7 S4 and
AdS5 S5 to subsets of U (1) gauge fields and scalars. In [23] was studied the truncation
of the KK reduction of 11D sugra on AdS7 S4 to the bosonic sector of an N = 2 (SU(2))
gauged sugra. A set of only gravity and scalar fields was considered in [24], whereas [25]
analyzed various compactifications giving rise to gravity plus scalar fields in a coset manifold. In [26] it was found an explicit ansatz for the embedding of the N = 4 (SO(4))
gauged sugra in 11D (via S7 compactification). In [27,28] the KK reduction of 10D sugra
on S3 S3 was analyzed, and in [29] the embedding of N = 4 gauged sugra into 10D sugra.
In this paper we want to generalize some of the results obtained in [6], most notably
for application to the AdS4 S7 and AdS5 S5 cases, and also to relate our results to the
other works in the literature. We will explicitly show that the other anstze for embedding
of subsets of fields of the 7D N = 4 gauged sugra into 11 dimensions can be obtained as
particular cases of the ansatz in [6]. More precisely, we will recover: the ansatz involving
the graviton and 4 scalars in [24], the ansatz involving the graviton, an SU(2) gauge field,
a scalar and an antisymmetric tensor in [23], the ansatz for the S3 compactification of
10D sugra of [30] (which leads to d = 7, N = 2 gauged sugra without topological mass
term) and the ansatz involving a graviton, two abelian gauge fields and two scalars in [21].
The correspondence between our ansatz truncated to the fields of d = 7, N = 2 gauged
sugra [41] and the ansatz of L and Pope [23] is discussed in detail. To obtain agreement
between the two ansatze (and between the field equations derived from the truncated action
of the N = 4 model and the field equations of N = 2 gauged sugra), we need to make
a redefinition of the three index potential A(3) (in the N = 2 model) as a linear combination
of the three index potential S(3) (in the N = 4 model) and the ChernSimons form (3) .
We noted that there is a lagrangian formulation of d = 7, N = 2 gauged sugra, obtained
from the maximal 7D gauged sugra Lagrangian by truncation, which yields the set of field
equations and constraints which so far had been used to describe the N = 2 model. By
213
taking a singular limit of the ansatz which yields N = 2, d = 7 gauged sugra from d = 11
sugra with topological mass term [6,23], we obtain an ansatz which yields N = 2, d = 7
gauged sugra without topological mass term. Thus we confirm the consistency of the ansatz
proposed by Chamseddine and Sabra [30] for the S3 compactification of type I 10D sugra to
7D, who derived their ansatz only from the requirement to reproduce the action of gauged
N = 2, d = 7 sugra (which does not necessarily imply consistency). Since we obtain it as a
singular limit from a consistent ansatz we implicitly prove the consistency of the truncation
in [22].
The generalizations we are interested in are related to the ansatz for the embedding
of the lower (d-)dimensional scalar fields into the higher dimension (D). We will see
that when one compactifies on a sphere Sn , one has a scalar coset which is at least
Sl(n + 1)/SO(n + 1). We will mostly work with this coset, but for the AdS4 S7 and
AdS5 S5 compactifications we will analyze also the case of the full coset. One would
think that only the cases (d, n) = (4, 7), (7, 4) and (5, 5) are of interest, and the metric
and antisymmetric tensor ansatz for this case were found in [24], but actually the case of
general d and n has its own interest. That is for instance because it was found useful
in problems related to the RandallSundrum model [31,32], for instance see [33]. For
RandallSundrum type scenarios, one needs models of gravity interacting with scalar
fields. Since finding a consistent realization of this scenario inside a susy theory is an (yet)
unsolved problem (see [3436]), it is worth studying general gravity + scalar systems.
Moreover, [24] considered only n scalars, which is OK as long as no gauge fields couple to
them. But we find the metric and antisymmetric tensor ansatz and show how to couple the
gauge fields to the scalars. A completely new result is that we obtain the full metric ansatz
for the KK reduction of 10D IIB sugra on AdS5 S5 , which means that if one has a solution
of 5D gauged sugra one can find the metric of the full 10D solution, and sometimes that is
all one needs (for instance for the applications for the RandallSundrum type scenarios).
Yet another extension of the ansatz in [6] is related to the fermions. We see in general that
there will be a scalar field-dependent matrix U which relates the fermions to the Killing
spinors, as was done in [1] for the AdS4 S7 case and in [5,6] for the AdS7 S4 case (the
matrix U in the 4D case was introduced also in [37,38]). We find an ansatz for the vielbein
which depends on the matrix U and an equation which U has to satisfy. We analyze
separately the AdS4 S7 and AdS5 S5 cases, and we find there the vielbein and the
U equation for the theory with the full scalar coset (no restriction to Sl(n + 1)/SO(n + 1)).
The paper is organized as follows. In Section 2 we derive from the ansatz [6] for the
consistent KK reduction of d = 11 sugra to maximal d = 7 gauged sugra the other anstze
in the literature [2124,30] for the KK reduction to smaller bosonic sectors of the maximal
gauged sugra in d = 7. In Section 3 we write the metric ansatz for the Sn compactification
involving a scalar coset Sl(n + 1)/SO(n + 1), and the dependence of the antisymmetric
tensor field on the scalars. For the AdS4 S7 and the AdS5 S5 cases, we get the full metric
ansatz (with no restrictions). In Section 4 we say a few words about the fermionic ansatz,
introduce the matrix U , and derive the vielbein and U equation. Again, in the AdS4 S7
and AdS5 S5 cases, we impose no restriction. We finish with discussions and conclusions.
In Appendix A we give some useful things about the Killing spinors on spheres.
214
2. From d = 11 sugra to d = 7 sugra via S4 compactification

The consistent embedding of 7-dimensional supergravity theories in higher dimensions
was thoroughly studied in the recent literature. We will first describe the case of S4
truncation of the original d = 11 supergravity [5,6], and then obtain other embeddings
as particular cases. When compactifying on S4 we obtain N = 4 maximal gauged sugra
in seven dimensions [39]. The scalars 5A i (y) parameterize an Sl(5, R)/SO(5)c coset
manifold. The gauge fields BAB (y) are in the adjoint of SO(5)g , and the gravitini and spin1/2 fields transform in the spinorial representation of the R-symmetry group SO(5)c and in
the vector-spinor representation, respectively. As we shall see, this model admits consistent
truncations to smaller field subsets: N = 2 gauged sugra, 3 or purely bosonic sectors such
as the graviton and the whole set of scalars, or the graviton, two abelian gauged fields
(in the Cartan subalgebra of SO(5)) and two scalars. Our ansatz for the metric factorizes
into a rescaled 7-dimensional metric and a gauge invariant two form which depends on the
A
B
scalar fields through the composite tensor T AB (y) = 51 i 51 i :

1
1
2
2/5
A TAB (y)
B
(DY ) , (2.1)
(x, y) g (y) dy dy + 2 (DY )
ds11 (x, y) = 1
m
Y T (y) Y
DY A = dY A + 2mB AB (y)YB .
(2.2)
The scale factor is

16/5 (x, y) = Y A (x)TAB (y)Y B (x).
(2.3)
1
Y B describes a conformally rescaled
The full internal metric g = m2 14/5 Y A TAB
ellipsoid: g = m2 14/5 Z A ZA where the Z A (x)s are constrained to lie on the
ellipsoid. Therefore the overall effect of all scalar fluctuations on the internal metric
is to deform the background sphere g = m2 Y A Y B AB , Y A (x)Y B (x)AB = 1
into a conformally rescaled ellipsoid. Other types of fluctuations would correspond to
massive modes.
The 4-form field strength of the three index photon of the standard d = 11 sugra is
given by

2
(T Y )A5
1
F(4),11 = A1 ...A5 3 (DY )A1 . . . (DY )A4
3
Y T Y
3m

(T Y )A4
4
A1
A2
A3
Y A5
(DY
)
(DY
)
(DY
)
D
+
Y T Y
3m3

A5
1 A1 A2 A3 A4 A5
2 A1 A2
A3
A4 (T Y )
+ F(2) F(2) Y
+ 2 F(2) (DY ) (DY )
m
Y T Y
m

B
(2.4)
+ d S(3)B Y ,
AB = 2(dB
AB and
where we define the YangMills field strength to be F(2)
(1) + (B(1) B(1) )
8i
the independent fluctuation form S(3)B = S ,B is real. (We also used the implicit
3
convention F(4),11 = F dx dx dx dx .)
3 We are thankful to C.N. Pope, H. L and A. Sadrzadeh for pointing to us this fact.
215
Finally, in order to produce a first-order field equation for the antisymmetric tensor field
S (self-duality in odd dimensions [40]) as well as for the consistency of the fermionic
susy transformation laws we had to introduce in the d = 11 sugra model an auxiliary
field BMNP Q . The ansatz for the KK truncation of the 4-index auxiliary field has a single
nonvanishing component
B
i
S7
=
Y A,
(2.5)
g11
S,A
2 3
where S7 is the action of the 7-dimensional gauged sugra. The other models studied in the
literature (in d = 7) are special cases of this general ansatz.
2.1. Graviton and scalars in Sl(5)/SO(5) [24]
1 2/3 X 1 2
e
1
Xi di ,
(2.6)
g2
i
X
X

e i (7) 1
Xi1 ? dXi d 2i ,
2Xi2 2i 1X
(2.7)
F(7),11 = g
2g
i
i
Q
where the scalar fields satisfy the constraint 5i=1 Xi = 1, and spherical harmonics i are
P
P5
e is defined as 1
e = 5i=1 Xi 2 . We easily
such that i=1 2i = 1. The conformal factor 1
i
recognize that the two sets of spherical harmonics i and our Y A are the same. Moreover,
the scalar fields Xi are embedded in TAB as
2
e1/3 ds72 +
=1
ds11
TAB = diag(X1 , X2 , . . . , X5 )
(2.8)
e and the metric ansatz given in (2.6) coincides with the ansatz (2.1)
Therefore, 16/5 = 1,
if we make the identification (2.8), and we set the gauge fields to zero. However, it is not
immediately obvious that the ansatz for the antisymmetric tensor in (2.7) also coincides
with (2.4), upon truncation. To check agreement between the two ansatze we compute the
dual of F(7),11 :
F(4),11 = ? F(7),11
X
g7
g11 1 ...7
2 2
e

... dx dx dx dx
=
g
1X
(2.9)
2X
i
i i
4!
7! 1 7
i
X
g7
g11 1 ...6
1
2

... dy 7 dx dx dx ,
X
7
i
i
i
4!
6!2g 1 6
i
where the notation ? means Hodge dual in the higher-dimension space.

Using that the inverse metric is
e 1/3
,
(2.10)
g11 = g7 1

X
X
X

1
e
e 1/3
(i )Xi (i ) ,
g11 = g 2 1
2i Xi
2j Xj 1
4
i
where = g and g is the inverse background metric on S4 , and that g11 =

e7/6 e1/3 p
g4 g7 1 = 1
g7 g4 g 4 , we get:
216
p

g4 2 3 X
e g
eXi dx dx dx dx
1
F(4) =
2Xi2 2i 1
4!
i
p
e 2 1
e 1/3 0 X 1

g4 1
+ 0
g
Xi Xi 2i dy dx dx dx . (2.11)
3
4!
2g
i
Using further that i i = g and that [6]

p
ABCDE dY A . . . dY D = g4 Y E dx dx dx
(2.12)
we can show that (2.11) is a particular case of (2.4).

2.2. Graviton, SU(2) gauge fields Ai(1) , a three form A(3),7 and one scalar X
The ansatz for the consistent bosonic truncation of d = 11 sugra to N = 2, d = 7 gauged
sugra [41] was given by H. L and C.N. Pope [23]:
2
e1/3 ds72 +
=1
ds11
X
2
2 3 1/3 2
1 2/3 1
e d +
e
1
X 1
X cos2
i gAi(1) , (2.13)
2
2
g
2g
i

1
X8 sin2 2X2 cos2 + 3X3 cos2 4X3
F(4),11 =
2 2g 3
e2 X4 sin cos4 dX (3) + sin F(4),7
e2 cos3 d (3) 5 1
1
2 2g 3
1
2
i
cos X4 ? F(4),7 d
+
cos F(2)
d hi
g
2g 2
1
e1 sin cos2 F i hk hk ij k ,

X4 1
(2.14)
(2)
4 2g 2
where the 4-sphere is described as a foliation of 3-spheres, parameterized by the Euler
angles (, , ) with latitude coordinate ; i are left invariant one forms on S3 : d s42 =
i = dAi + g/2 Aj Ak are the YangMills field strengths
d 2 + 1/4 cos2 i i ; F(2)
ij k (1) (1)
(1)
with g the coupling constant and hi = i gAi(1) are the gauge invariant one forms; (3)
i F i . As discussed in [6] the maximal gauge field
is a ChernSimons form, d(3) = F(2)
(2)
group inherited from the 4-sphere, SO(5), needs to be broken first to SO(4) by first setting
B A5 = 0 and further, from SO(4) to one of its SU(2) subgroups, by imposing an anti-selfduality condition. Thus the embedding of the gauge fields is
B A5 = 0,

B = 12
B ,
= 1, 4,
B i4 =
Ai .
2 2 (1)
(2.15)
The scalar field of N = 2 gauged sugra is embedded in the Sl(5)/SO(5) coset of maximal
N = 4 sugra as

(2.16)
TAB = diag X, X, X, X, X4 .
A is also truncated to a singlet under
The three index form of maximal d = 7 sugra S(3)
SU(2):

5
S(3) S(3)
,
0 = S(3) ,
= 1, 4.
217
(2.17)
The correspondence between the Cartesian spherical harmonics Y A and the new parameterization of the 4-sphere is
Y 5 = sin ,
b ,
Y = cos Y
= 1, 4,
(2.18)
b are
b are constrained to lie on a 3-sphere. Finally, the Cartesian coordinates Y
where Y
related to the Euler angles (, , ) by

b2 + i Y
b 3 = cos exp i( + ) ,
b1 = sin exp i( ) , (2.19)
b4 + i Y
Y
Y
2
2
2
2
and by organizing them into an SU(2) matrix

4
b3 Y
b1
b2 + iY
b + iY
Y
k bk
b4
G=
b4 i Y
b2 + iY
b1 Y
b 3 = Y 1 + i Y ,
Y
(2.20)
we get the invariant SU(2) one forms i in terms of the Cartesian coordinates from

1
bj d Y
b4 d Y
bi Y
bi d Y
b4 + ij k Y
bk .
tr i G 1 dG = 2 Y
i
Also, from (2.21) we get

bi d Y
b4 + 1 ij k d Y
bj d Y
bk .
d i = 4 d Y
2
i =
(2.21)
(2.22)
Thus, the equivalence between the two metric ansatze is straightforward [6]. In (2.1) we
substitute the truncated fields and the redefined spherical harmonics and we get

1
1
2
2/5
A TAB (y)
B
(DY )
(x, y) g (y) dy dy + 2 (DY )
ds11 = 1
Y T (y) Y
m
e1/3 X3 d 2
e1/3 ds72 + m2 1
=1
b + 2m(B Y
b) d Y
b
b d Y
e2/3 cos2 X1 d Y
+ m2 1

b) (B Y
b) .
+ m2 (B Y
(2.23)
Using the relation (2.21) and the anti-self-duality condition on the gauge fields, the 11dimensional invariant line element becomes:
i

2
i
e 1/3 ds72 + 2 1
e 1/3X3 d 2 + 1 cos2 X1 i gAi
=1
(2.24)
ds11
(1) gA(1)
g2
2g 2
and coincides with (2.13), provided that the YangMills constant is g = 2m.
The ansatz for the 11-dimensional 3-index tensor A given in [6] differs from the
one of L and Pope. However, since it is defined only up to a gauge transformation (namely
up to [ ] ) we will compare the ansatz for its field strength F(4),11 = dA(4),11, i.e.,
(2.4) vs. (2.14). The two anstze will turn out to be the same if we assume that the threeform A(3) is a linear combination of S(3) and (3) .
To show the equivalence of the two anstze, we need to make use of the following
identities involving the spherical harmonics:
218
b D Y
b D Y
b D Y
b = 0,
D Y
b
b b
D Y D Y D Y Y = 18 ij k hi hj hk 3!8 (3) ,

1
i
b D Y
b F =
ij k F(2)
hj hk ,
D Y
(2)
2 2

b Y = 1 F i hi .
F(2) D Y
2 (2)
(2.25)
(2.26)
(2.27)
(2.28)
To prove these identities, one needs to use the form of i and d i in (2.21), (2.22) and the

embedding of the gauge fields in (2.15) (remember that F(2) = 2(dB(1) + 2mB(1) B(1) )).
Then, one finds the various terms in (2.4) in terms of the left-hand side of these identities:
3!
T Y A5
= cos3 d (3) ,
Y T Y
2

A
4
T Y
DY A3
Y A5
Y T Y
A1 ...A5 DY A1 . . . DY A4
A1 ...A5 DY A1 DY A2
=
3!
e2 cos3
cos3 (3) d 1
2

e 2 sin2 cos2 X8 2X3 + X2 + X2 cos2
3X1

+ X3 sin4 cos3 + cos7 + X8 sin2 cos5 ,
(2.29)
(2.30)
T Y A5
A1 A2
(2.31)
DY A3 DY A4
A1 ...A5 F(2)
Y T Y
2

cos sin X4
b
b
b
b
F(2) D Y D Y + 2 cos d F(2) D Y Y ,
= 2
e
1
and one recovers most of (2.14). 4
In the N = 2 model of [41] the Lagrangian is quadratic in dA(3) , and thus to reduce the
on-shell number of degrees of freedom of A(3) one needs to supplement the field equations
with the self-duality in odd-dimensions constraint [23,41]:
X4 ? F(4),7 = 1 gA(3),7 + 12 (3) .
(2.32)
In fact, we should stress that the N = 4 model yields a first order field equation for S(3) ,
which is the square root of the quadratic field equation of A(3),7. We note that the field
equations obtained from the truncated N = 4 action correspond to the field equations and
constraints of the N = 2 model. This implies that one should use the following Lagrangian
for describing the bosonic sector of N = 2, d = 7 gauged sugra with topological mass
term: 5
q

g71 L7D, N =2 = R + g 2 2X3 + 2X2 X8
+ 5 X1 X +
g 2 4
1
i
X S S X2 F
F i
2
4
4 To get the same overall coefficient as in (2.14) we have to take into account that the three-form potential of
11D sugra used in [23] is related to the one used in [5,42] by a rescaling with 6 2, and that in the conventions
of [23] F equals 4[ A ] , while for the authors of [5,42] F = 24[ A ] . Also note that
in [5] the form F(4) was not normalized: F(4) = F dx dx dx dx .
5 Note that the limit g 0 is singular, as it was in the N = 4 gauged sugra model.
219

q
+ g71
g
S F
(2.33)
24 2

i
1
i
i
F
.
+ + S F
48 2g
8 3
The field equation for S reads

i
2 4
2gS(3) + 3
g X S(3) = ?d
2 3
(2.34)
and coincides with the self-duality constraint (2.32) if

i
(3)
.
S(3) = A(3),7
2 3
2g
(2.35)
On the other hand, in F(4) (2.4), the remaining terms which at the first sight seem to differ
from the ones in (2.14) are

3 8i
i
i
F(2)
d(S(3) sin ) + 2g sin F(2)

3
2

i
8i
3 8i
2S(3) + (3) + S(3) cos d
(2.36)
=
sin d
6
2 3
3
2

8
3 8
dA(3),7 sin + cos X4 ? F(4),7 d ,
(2.37)
=
2 6
3g 2
where to go from (2.36) to (2.37) we made use of the self-duality constraint (2.32) which
is part of the truncation procedure for [23]. Thus we were able to completely recover the
ansatz (2.14) from our ansatz (2.4). This concludes that indeed, there exists a consistent
truncation of the maximal gauged 7D sugra to N = 2 gauged (SU(2)) 7D sugra with
topological mass term.
Finally, by taking the singular limit S4 S3 R in the same way as in [22], we recover
the ansatz of Chamseddine and Sabra [30]. They proposed an ansatz for obtaining the
N = 2, d = 7 gauged sugra, with no topological mass term (m = 0) for the three form
potential A(3),7, and checked that the action of lower-dimension theory is reproduced
from the 11-dimensional sugra action, when compactifying on S3 S1 . This, however,
does not generally guarantee consistency of the KK truncation. Since we get the ansatz of
Chamseddine and Sabra as a limiting case of an ansatz whose consistency was rigorously
proven, we conclude that their ansatz is also consistent.
We begin with S4 written as before as a foliation of S3 with latitude coordinate . Then,
we redefine
= a 0
(2.38)
and take the limit a 0+ which will yield an unrestricted 0 , taking values on the real
e X, and taking the limit a 0 in (2.13), we get
line. Then 1
X
2
2
2
= X1/3 ds72 + 2a 2g 2 X10/3 d 0 + 12 g 2 X5/3
(2.39)
i gAi(1) .
ds11
i
220
We also redefine the fields and the coupling constant:

X = X0 a 2/5,
g = g 0 a 2/5,
A(1) = A0(1)a 2/5,
(2.40)
2 :
in order to produce only an overall dependence on a in ds11
2
ds11
=a
2/15

2
X01/3 ds72 + 2g 02 X010/3 d 0
+
1 02 05/3
X
2g
2
gAi(1)

.
(2.41)
At this moment, noticing that the metric has a Killing vector 0 we decide to reinterpret
0 as an angular coordinate on S1 . With a similar analysis for F(4),11 given in (2.14), and
with the redefinition
0
a 4/5
F(4),7 = F(4),7
(2.42)
we get
F(4),11 = a
1/5
g 03 0
0
d 0
d (3) + 2g 01 X04 ? F(4),7
2

1 02 0i
0
i
g F(2) d h
2

g 01
0
0i
g 01 F(2)
hi .
= a 1/5 d 0 g 02 (3) 2X04 F(4),7
2
(2.43)
To get rid of the a dependence in the metric and F(4),11 ansatze we exploit the scaling
symmetry of the 11-dimensional sugra equations of motion [22,43]:
2
ds11 k 2 ,
ds11
(2.44)
F(4),11 F(4),11 k .
3
(2.45)
The relation between the scalar field X0 and the dilaton of [30] is X0 = exp 4/5.
Substituting it into (2.41) and (2.43) and we are led to the ansatz of Chamseddine and
Sabra:
2
2
= e4/15 ds72 + 14 e4/3 i 2Ai(1) + e8/3 d 2 ,
ds11

j
F(4),11 = d dB(2) Ai(1) dAi(1) + 13 ij k Ai(1) A(1) Ak(1)
1 dAi i + 1 ij k Ai j
(1)
(1)
4
2

i
j
k
1
ij k ,
12 2
k
(2.46)
where the two form B(2) is related to the three form A0(3),7 through a duality transformation,
and to reach the conventions of Chamseddine and Sabra, we set g 0 = 2.
221
2.3. Graviton, two abelian gauge fields Ai(1) and two scalars Xi , i = 1, 2
This model was discussed in [21] and allows to interpret 2-charge AdS black holes as
the decoupling limit of rotating M2-branes. The ansatz for a consistent bosonic truncation
of d = 11 sugra to this subset of bosonic fields reads:
!
2
X
2
1
1
1/3
2
e 2/3
2
2
2
i
e
,
Xi di + i d + gA(1)
X0 d0 +
ds11 = 1 ds7 + g 21
i=1
? F(4),11 = 2g
2
X
2
X

e i (7) + g 1X
e 0 (7) + 1
Xi1 ? dXi d 2i
Xi2 2i 1X
2g
i=0
i=0
1
2g 2
2
X

i
Xi2 d 2i d + gAi(1) ?F(2)
,
(2.47)
i=1
where X0 = (X1 X2 )2 , i , i = 0, 1, 2, and i , i = 1, 2, parameterize the 4-sphere, and

P
e = 2i=0 Xi 2 . The relationship between the Cartesian coordinates Y A , A = 1, . . . , 5,
1
i
and the new variables i and i is given by
Y1 = 1 sin 1 ,
Y2 = 1 cos 1 ,
Y3 = 2 sin 2 ,
Y4 = 2 cos 2 ,
Y5 = 0 ,
20
+ 21
(2.48)
= 1 translates into
= 1. The embedding of the
where the constraint
scalar fields in the symmetric matrix TAB of the maximal d = 7 gauged sugra is
Y AY A
+ 22
TAB = diag(X1 , X1 , X2 , X2 , X0 ).
Ai(1) ,
Finally, the abelian gauge fields

1 A1 = B 12 + B 34 ,
(1)
(1)
(1)
2
(2.49)
i = 1, 2, are in the Cartan subalgebra of SO(5)g :

1 A2
2 (1)
12
34
= B(1)
B(1)
.
(2.50)
3. The metric ansatz

We will now analyze the case when we reduce a certain sugra in D dimensions on
a sphere Sn = SO(n + 1)/SO(n) and see what we can learn. In order for the sphere to be
a background solution for the sugra of the spontaneous compactification type, we need an
antisymmetric tensor F(n) , such that this background is of the FreundRubin type, i.e.:
p
1 ...n .
(3.1)
F1 ...n = g
We note that for the compactification of 11D sugra on AdS4 S7 , one has a F(4) , but we
would take the dual of this, namely, an F(7) . So, we have to start with an action
Z
2
+ ,
(3.2)
S (D) = d D x g R (D) + F(n)
where contain other fields which we put to zero. A first observation is that this starting
point is not the most general. If one would add a dilaton coupling
Z
2
+ b(D)()2 + ,
(3.3)
S (D) = d D x g R (D) + ea(D,n) F(n)
222
one would get the most general starting point for a supergravity action admitting n 2
brane solutions. Then the near-horizon of that p-brane will be an AdSDn Sn background
solution of the theory, but the dilaton would not be constant in that case. We leave the
generalization to this (more interesting) case to future work. But the action in (3.2) has
been used already in [33], so it is not without interest to consider it.
We first ask what is the compactification ansatz if in the lower dimension d = D n
we keep only scalars. We know that the d-dimensional sugra which is obtained by
compactification is a gauged sugra, because the sphere Sn has isometry group SO(n + 1),
which becomes the gauge group SO(n + 1)g of the d-dimensional sugra. From the known
examples of gauged 4D sugra, obtained as 11D sugra on AdS4 S7 , 7D gauged sugra as
11D sugra on AdS7 S4 , and the less understood case of 5D gauged sugra as 10D IIB
sugra on AdS5 S5 , we know that the scalar fields are in a coset G/H , where G is a global
isometry of the ungauged model which is broken down to SO(n + 1)g by the gauging, and
H is a local composite symmetry. The gravitinos of the gauged sugra are in a fundamental
spinorial representation of this group H .
Moreover, we know that one way to determine which coset we have is to count the
number of scalar degrees of freedom and the number of gravitinos and to try to match
a coset G/H which does the job. The number of scalar degrees of freedom is always the
same as for the torus compactification (going from the torus to the sphere is the same as
going from the ungauged to the gauged model). So the minimum number of scalar degrees
of freedom is given by the following. The metric will have n(n + 1)/2 scalar d.o.f. from the
compact space metric. The antisymmetric tensor will have at least n scalar d.o.f., coming
from the component A1 ...n1 , with i compact space indices. It could have more, if
one of the components with spacetime indices can be dualized to a scalar. That does not
happen if A1 ...n1 is the d-dimensional dual of a form Ak with k > 0, i.e., if d n1 > 0.
This is the case of the AdS7 S4 background, for instance. In such a case, the coset will
be an (n(n + 1)/2 + n = n(n + 3)/2)-dimensional manifold. This is exactly the dimension
of Sl(n + 1)/SO(n + 1)c ((n + 1)2 1 n(n + 1)/2). Also, we know that the gravitinos
multiply Killing spinors in the linearized KK reduction ansatz (we have something of the
type I I , as we will see later). That means that it is safe to assume in that case that
the fermions are in a spinor representation of SO(n + 1)c (because I is a spinor index for
SO(n + 1)g ).
So, we can say that the scalar coset in the case d > n + 1 is determined, namely
Sl(n + 1)/SO(n). In the other cases, the scalar manifold will be bigger, but Sl(n + 1)/SO(n)
will still be a submanifold, so we will restrict to it in the general case. By a straightforward
extension of the AdS7 S4 case, where the coset is Sl(5)/SO(5), a coset element will be a
matrix A i , where A is an SO(n + 1)g index and i is an SO(n + 1)c index. The n(n + 3)/2
A
physical scalars can be described by the matrix T AB = i1 i1 , which is symmetric

and has unit determinant. The d-dimensional gauged sugra will have gravity + scalar terms:
Z
S (d) =
dd x

2
g R (d) + Pij
V (T ) ,
(3.4)
223
B
k
where P ij is the symmetric part of ( 1 )A
i A B kj (where the gauge field was
put to zero), and so the kinetic term in (3.4) is 1/8 Tr( T 1 T ). The scalar potential is
2 ). We will find an ansatz for the embedding of (3.4) into (3.2), and
V (T ) = g 2 /4(T 2 2TAB
see how to extend it for the case that in (3.4) we keep also the gauge fields of SO(n + 1)g .
We note here that the problem of dimensional reduction to gravity + scalars was analyzed
also in [24]. But the authors of [16] looked only at the cases AdS4 S7 , AdS7 S4 and
AdS5 S5 , moreover they restricted themselves to a diagonal T matrix. That is OK as long
as gauge fields do not couple to scalars, but we want to explore how to generalize to the
case where gauge fields are also nonzero.
Guided by the ansatz in [5,6], we expect that the metric g on the compact space has
the form
1
,
g = 1 Y A Y B TAB
p
where 1 is det g / det g , with the inverse metric given by

1
(T Y )A (T Y )B
AB
YA YB T
.
g =
Y T Y
Y T Y
(3.5)
(3.6)
But
0
1 A B
Y Y 1 ...n 1 ...n g1 0 . . . g n 0n
det g det TAB
1
= 1 ...n 1 Y A1 . . . n Y An 1 ...n 1 Y B1 . . . n Y Bn TA1

. . . TA1
n Bn
1 B1
= g
A1 ...An+1 B1 ...Bn+1 YAn+1 YBn+1 TA1
. . . TA1
n Bn
1 B1
= gT
An+1 Bn+1 YAn+1 YBn+1 ,
(3.7)
where to get from the second to the third line we have used a simple generalization of the
relations in [6] and the fact that det T = 1. Now we can see that
12n = T AB YA YB .
(3.8)
We take a standard ansatz for the vielbein, keeping the gauge fields, of the type
Ea = 1 d2 ea (y),
1
Ea = 0,
Em =
Em = B Em ,
1 m
e (x, y),
g
(3.9)
(3.10)
with B = BAB VAB , where BAB is the SO(n + 1)g gauge field and VAB is the Killing
vector. Then the D-dimensional line element becomes
1
2
= 1 dsd2 + g 2 1 TAB
DY A DY B
dsD
1
= 1 d2 dsd2 + g 2 1 TAB
DY A DY B ,
2
(3.11)
where DY A dY A + gB AB YB is a covariant derivative. We note that the factor 1 d2

is necessary to obtain the correct Einstein action in d dimensions. Next we will determine
from the requirement
that we recover the correct scalar potential V (TR) after integrating
R
2 .
the Einstein action d D xR (D) and the kinetic term of the n-form field dx D F(n)
2
224
First, we compute the contribution coming from the integration of the Einstein action.
In what follows we set (for simplicity reasons) the YangMills coupling constant to g = 1.
Using the metric ansatz given in (3.11) we get:
Z q
q
g (d) g (n) 1d/2+1 d D x R (D)
(3.12)

Z q
q
d d 3 2
+
g (d) g (n) d D x 1 R (n) +
12 1 1 g ,
=
2
2
where we already used that d/2 + 1 = . Moreover, if we write

g = 1 g
(3.13)
we can use the following formula for the relation between the Ricci scalars of two metrics
related by a conformal rescaling as in (3.13):

e(n) 1 1 (n 1)(n 2)g ln 1 ln 1
R (n) = R
4
(n 1)g D ln 1 .
(3.14)
For our particular metric g = Y T Y , the Ricci scalar has the following expression
(independent of the dimensionality of the compact space):
3 AB
2 AB
AB
2
2
e(n) = 2YA YB (T ) + 2T YA YB (T ) + YA YB T (Tr(T ) T ) .
(3.15)
R
AB
2
(YA YB T )
And then we obtain finally:
Z q
Z q
q
q
g (d) g (n) 1 d D x R (D) =
g (d) g (n) d D x 1

4
d d 3 2 2 (n 1)(n 2) 2
e(n)
+
1)(n
1)
R
4
(2 n)2 2

2
2
3
Y T Y
(Y T Y )
(3.16)
(Y T Y )3
(Y T Y )2
and we can already see that we need to have
1 = Y T Y
(3.17)
in order to obtain terms of the type T 2 in d dimensions. That gives a solution for as
2 d 1
.
(3.18)
=
n1d 2
And now we can write the full line element as
2
= (Y T Y )
dsD
2
(d2)(2n)
1
ds72 + (Y T Y ) 2n TAB
DY A DY B
(3.19)
with found before. We can compare with the ansatz given in [24] for the case of no
gauge fields and a diagonal matrix TAB = XA AB , valid for the AdS4 S7 , AdS7 S4 and
AdS5 S5 compactifications, namely:
2
d3 1 2
2
= XA YA2 d1 dsd2 + XA YA2 d1 XA
dYA .
(3.20)
dsD
By comparing the two results, we see that we have the same result if d(n 3) = 3n 5, an
equation which has as the only solutions (d = 4, n = 7), (d = 7, n = 4) and (d = n = 5). So
225
we have the curious fact that the formula derived in [16], although is written in a general
form (with a general d), is only valid in the cases for which it was derived, not for general d.
For general d and n, one should take the appropriate truncation of our formula (3.19).
Finally, we will complete the calculation of the scalar potential. We can compute the
integrals on the sphere using the formulas:
Z
1
Y AY B
=
T 1 ,
(3.21)
Y T Y
n + 1 AB
Z A B C D
1
Y Y Y Y
1 1
1 1
1 1
TAB
=
TCD + TAC
TBD + TAD
TBC ,
(3.22)
2
(Y T Y )
(n + 1)(n + 3)
etc. From that we derive, for instance,

Z
Y T2Y 2
T2
1
Y T3Y
2
TAB
.
(3.23)
=
Y T Y
Y T Y
n+3
n+1
In conclusion, the Einstein term in the D-dimensional action gives the following
contribution to the d-dimensional scalar potential:

Z q

1
1
d 1
T2
2
2
TAB
+
g (d) d d x
T 2 . (3.24)
TAB
(n 1)
d +n2n+3
n+1
n+1
We will now turn to the ansatz for the antisymmetric tensor field. We will again
generalize the result in [6], where the ansatz for the 3-index tensor in the presence of
only scalar fields is
A5

1
T Y
A1
A2
A3 A4
.
(3.25)
A(3) = A1 ...A5 dY dY dY Y
Y T Y
6 2
We write, similarly, an ansatz for the (n 1)-form:
An+1

T Y
A1
An1 An
Y
A(n1) = aA1 ...An+1 dY . . . dY
Y T Y
p
1 D n
1 ...n g ,
(3.26)
n 2
where we kept only one factor of T Y/(Y T Y ) because we want to have an
d-dimensional action involving only T 2 terms (and no Tr T 1 terms, for example),
completely analog to the case considered in [6]. Here a is a constant to be fixed. However,
if we look then at the terms with only scalars and no derivative in F(n) , we have:
p

F(n) |no T = g 1 ...n dx 1 . . . dx n

Y T2 Y
T
a
(n + 1) 2
1
,
(3.27)
1+
n Y T Y
(Y T Y )2
and then the term in the action becomes:

2
Y T2Y
a(n 1) a
T
2
2
+
2
. (3.28)
F(n) 1 |no T = 1 n! 1
n
n Y T Y
(Y T Y )2
But in order to get only T 2 terms in the action, we need that the the power of 1 is 2(2n),
so that we obtain a factor (Y T Y )2 multiplying the bracket in (3.28). That gives again
226
the equation d(n 3) = 3n 5, which is solved only by (d = 4, n = 7), (d = 7, n = 4) and

(d = n = 5).
But we can easily see how we can modify the ansatz. Just multiply F(n)1 n by
3n 2
. Then A(n1) will be
1
An+1

T Y
+
A(n1) = 13n 2 aA1 ...An+1 dY A1 . . . dY An1 Y An
Y T Y
p

1 D n
1 ...n g 13n 2 .
(3.29)
n 2
D
Notice that we could have put 12n 2 outside 2n , but that would generate an extra term
in the scalar potential. Now the term in the action becomes:

Z
Y T2 Y
a(n 1) a
T
2
+
2
F(n) 1 |no T = n! (Y T Y ) 1
n
n Y T Y
(Y T Y )2

2

Y T2Y
2
(3.30)
1
+ 2a 1
2(2 n)
(Y T Y )2
2 and T 2 terms, with a free parameter a.
and by integration we get a combination of TAB
Then a is fixed by requiring that in the sum of (3.30) and (3.24) the relative coefficient of
2 and T 2 is 2 (the overall coefficient depends on the coupling constant g). It is not
TAB
D
clear how one gets the correct kinetic terms for T , since A(n1) will contain 2n
2 kinetic
terms. When integrated on the sphere, they should give contributions to the Pij
terms, too.
In the remainder of this section we will look at applications of our metric ansatz (3.19).
Let us see that it can be applied for the AdS7 S4 case. The metric deduced by de Wit and
Nicolai [2] is
0 0
11 g = 18 V I J V KL wi 0 j 0 I J wi j
wij
IJ
= uij
IJ
+ vij I J ,
ij
KL ,
KL
= uij KL + v ij KL ,
(3.31)
where I, J, . . . are spinorial SO(8)g indices, and i 0 , j 0 , . . . are SU(8)c indices, both [i 0 j 0 ]
and [I J ] are antisymmetrized, and u and v together form a representation of E7 . Then one
can define:
I J
(3.32)
V AB = AB VI J ,

AB
AB
IJ
1
= 64
w
,
(3.33)
wij
I J ij
where A, B, . . . are vector SO(8)g indices, and VAB = Y A Y B and we used that
( AB )I J (AB )KL = 16IKL
J . Then if one restricts the scalar coset from E7 /SU(8)c to
Sl(8)/SO(8)c , with SO(8)g embedded in Sl(8), the matrix w will reduce to
A
wij AB = [i1 j1
]
wi 0 j
AB
0
i0j 0
and
[A B]
CD = T[C TD] ,
and then we recover our general formula (3.6).
(3.34)
(3.35)
227
As a small application, we note that now we can write full line element as
2
= 11 ds72 + 11 MAB DY A DY B ,
ds11
eAB = wi 0 j 0 AC w
where MAB is the inverse of M
(3.36)
i0 j 0
BD YC Y
D,
i.e.,
eAC MBC = BA + (YB terms).

M
(3.37)
e has no inverse in the usual sense, because it has an eigenvector with zero
Note that M
eigenvalue, namely YA , but we only need (3.37) to get (3.36).
A new application for our method is an ansatz for the metric in the AdS5 S5 case. We
guess it by the inverse procedure to the one described for the AdS4 S7 case. Namely, we
A
B
by the projection of the E6 /U Sp(8)c scalar coset vielbein onto the
replace [i1 j1
]
15 representation of SO(6)g :
AB

A
B
1 (15) ab 0 AB P (15) 1 ab .
(3.38)
[i1 j1
]
In this way we arrive at the ansatz
1 (15)
12/3 g = VAB VCD
AB
ab
1 (15)
CD
(3.39)
ab
as a natural extension of our particular ansatz
TAC TBD .
12/3 g = VAB VCD
(3.40)
We will give another argument for (3.39) in the following section, by first computing the
vielbein.
We note that we can again put the 10D line element in a form similar to (3.36), namely:
2
2
= 1 d2 ds52 + 12/3NAB DY A DY B ,
ds10
eAB = ( 1 (15))ab
where NAB is the inverse of N
eAC NBC = BA + (YB terms).
N
(3.41)
AC
YC ( 1 (15))ab
BD
YD , i.e.,
(3.42)
4. The fermion fields and vielbein ansatz

In this section we will try to give a general discussion of the fermionic fields and the
vielbein in the case of compactifications on spheres.
For the vielbein, we already gave a general ansatz, namely:
1
Ea = 1 d2 ea ,
Ea
= 0,
Em
Em = B Em ,
(4.1)
(4.2)
with B = BAB VAB , the only thing that remains to give is an ansatz for the compact space
vielbein, Em . For that, we need first an ansatz for the fermions.
In a supergravity theory, we always have a gravitino field. By dimensional reduction, we
obtain a set of gravitinos in d dimensions, transforming in the fundamental representation
under the composite symmetry of the gauged sugra. For example, in the AdS7 S4 case, we
228
have gravitinos in the 4 of SO(5)c = USp(4)c . For the AdS4 S7 case we have gravitinos
in the 8 of SU(8)c , and in the AdS5 S5 case, the gravitinos are in the 8 of USp(8)c .
Throughout most of the discussion we will restrict ourselves to a SO(n + 1)c composite
symmetry group (SO(5)c , SO(8)c SU(8)c , SO(6)c USp(8)c ).
However, the usual (linearized) ansatz for the gravitinos involves a Killing spinor
(9 (y, x) = I (y)I (x)), which has an SO(n + 1)g spinor index of the gauge field. That
0
tells us that there should exits a matrix U I I , with I 0 a composite group index, interpolating
between the gravitinos and the Killing spinor.
The result we obtained in 7D in [5,6] is
9a = 11/10(5 )1/2 a 15 a 5 m 11/10(5 )1/2 m ,
9m = 1
1/10
(5 )
1/10
1/2
m ,
(4.4)
1/10
,
= 1
(5 ) ,
0
I
I
(y, x) = I 0 (y)UI (y, x) (x), where a = ea (y) (y, x),
0
0
0
J KL
(x),
m (y, x) = J 0 K 0 L0 (y)UJJ (y, x)UKK (y, x)ULL (y, x)m
0 I
I
(y, x) = I 0 (y)UI (x).
=1
(5 )
1/2
(4.3)
1/2
(4.5)
(4.6)
(4.7)
We see that we needed a certain rotation to diagonalize the 7-dimensional gravitino and
spin-1/2 kinetic operators. In general, we expect that the rotation is more complicated,
such that both 9a and 9m depend on both gravitino and spin-1/2 fields in d dimensions.
Moreover, in general, the D-dimensional sugra itself has spin-1/2 fields, so the rotation
involved is probably more general than the one considered here. We will not have more to
0
say about that, we will just note that the advertised matrix U I I appears in the ansatz for
all the fermions.
0
One possible interpretation is that U I I gives a (field- and spherical harmonicdependent) composite rotation. Indeed, U is a matrix in the coset Hc /SO(n + 1)g (for
instance because it has one index transforming under Hc and one transforming under
SO(n + 1)g ), so it can be interpreted as a field dependent Hc rotation. We note that if
we break Hc to SO(n + 1)c (we will do it for most of this section), then U becomes
simply an SO(n + 1) matrix in the spinor representation. Since the vielbein couples to the
fermions, we should expect that Em depends on U also, and we shall find exactly that. We
note here that restricting U to be and SO(n + 1) matrix in the spinor representation means
also that the fermions are restricted to lie in a spinor representation of SO(n + 1). For
a concrete example, for the AdS4 S7 case that would mean that the fermions, which are
in a representation of SU(8) are restricted to transform only under an SO(8) representation.
We can however still say the following about the fermions. We take a = a , as the
dimensional reduction of the gamma matrices from D to d dimensions (M to a ), where
is one for odd-dimensional spheres and 2n+1 for even-dimensional spheres. If both d
and n are odd, the gamma matrix split involves an extra matrix, e.g., in AdS5 S5 we
have a = a 1 1 (and m = 1 m (2 )). We will examine the AdS5 S5 case
separately. From the requirement that the D-dimensional gravitino action reduces to the
d-dimensional one, more precisely:

0
1 d2 9a ab 9b = aI 0 ab bI +
229
(4.8)
(on the left-hand-side indices are flattened with the D-dimensional vielbein and on the
right-hand-side with the d-dimensional one), we get the ansatz
0
9a (y, x) = 1 2(d2) ( )p aI 0 U I I (y, x)I (x) + spin-1/2 terms,
(4.9)
where again 9a is flattened with the D-dimensional vielbein and a with the ddimensional one. Moreover, we can derive the ansatz for the susy parameter:
1
(y, x) = 1 2(d2) ( )p I 0 U I I I .
(4.10)
Using this fermionic anstze, we will derive the vielbein ansatz and then the relation that
the matrix U has to satisfy.
The vielbein ansatz is easily derived from the susy law of the gauge fields. We have in
general a susy law of the type
AB I 0 J 0
+ spin-1/2 terms.
(4.11)
BAB = 1 I 0 J 0
That is so because a boson transforms into times a fermion, and by matching indices, it
is easy to see that we need a bosonic matrix in a coset involving the gauge group (via the
A, B indices) and the composite group (via the I 0 , J 0 indices). The only one available is
AB
which is either the scalar coset vielbein, or some product of such
a matrix ( 1 )I 0 J 0
vielbeins. We will treat separately the cases of AdS4 S7 , AdS7 S4 and AdS5 S5 , but in
the case that we restrict the scalar manifold to the coset Sl(n + 1)/SO(n + 1) (in particular,
for AdS7 S4 there is no restriction made), we know that the susy law takes the form:

0
0
A
B
(4.12)
BAB = ci1 j1 ij I 0 J 0 I J + spin-1/2 terms,
where c is a normalization constant. On the other hand, we can obtain the variation of B
starting from D dimensions, in a manner entirely analog to the one used in [6]. Namely, by
assuming only that in D dimensions we have the following susy law for the vielbein:
M
= 12 M 9
E
(4.13)
and using the fact that in the gauge Ea = 0 we have a compensating off-diagonal Lorentz
a 9m , we get the variation of B :

rotation m a = a m = 12 0

+ Em Ea a m + Ea a m Em
B = susy Em Em

(4.14)
= 12 Em Ea a 9 m + m 9a .
Now we can substitute the gravitino ansatz and (in the hypothesis that there is no ddimensional gravitino contribution in 9m ), we get

s 1 m
1 d2 Em + spin-1/2 terms ,
(4.15)
2
where s is a phase coming from the commutation relation of ( )p with m . By comparing
with the d-dimensional result, we get
I 0 J 0
1
2c 1 A 1 B
0
m,I J I 0
V
U I UJ J =
VAB ij
,
(4.16)
1 d2 Em
i
j
s
230
and therefore we obtain an ansatz for the vielbein:

1
2ce 1 A 1 B
=
VAB Tr ij U Vn U 1 .
1 d2 Em
i
j
s
(4.17)
Here e is defined by (U V m U T )I J (U Vn U T )I J = enm . Squaring the result for the vielbein

we get back the metric in (3.6), provided 4c2 = s 2 e.
We also get a self-consistency condition on the matrix U , by plugging (4.17) back into
(4.16):
A
B

0
0
e 1 i 1 j VAB Tr ij U Vm U 1 V m,I J U I I U J J
A
B
I 0 J 0
= 1 i 1 j VAB ij
.
(4.18)
This equation is the most general condition on the matrix U . In [6], we found the equation
(in the AdS7 S4 case)
A
(4.19)
U Y/ = v/U, vi = 13/5 1 i YA ,
and showed that it implies (4.18). But the reverse is also true: (4.18) implies (4.19). If one
multiplies (4.18) by (kl )I 0 J , one gets on the left-hand side:

A
B
(4.20)
Tr U Y/ U 1 kl U Y/ U 1 i j 1 i YA 1 j CB .
But U Y/ U 1 will be equal to v/ , where v is a unit vector, just because U is a SO(n + 1)
matrix in the spinor representation. To determine v, we impose that we recover the righthand side and obtain
A
(4.21)
vi = 13/5 1 i YA ,
as desired. So we can say that (4.18) is the most general equation on U , but it can be
brought to the nicer form (4.19).
Although we look here at the case AdS7 S4 , this is valid for all (d, n), the only thing
we used here was that Tr MA Tr NA Tr MN (A.8), under the assumption Tr M =
Tr N = 0, and that U Y/ U 1 = v/ for some v. Then we get, in general,
A
(4.22)
vi = 1n/21 1 i YA .
We will now finally turn to the specific cases of the AdS4 S7 and the AdS5 S5
compactifications.
For the AdS4 S7 compactification, the full vielbein ansatz (with no restriction to the
Sl(8)/SO(8) scalar coset) was implicitly derived in [2], by the equation

2 1/2
i1
Em U V m U T ij = wij I J VI J ,
(4.23)
8
and we can derive the vielbein and the equation for U from it. The vielbein is

1
2 1/2
i1
Em = wij I J VI J U V m U T ij
8
16
(4.24)
231
n ) and the equation for U (the self(we used here that (U V m U 1 )ij (U V n U 1 )ij = 16m
consistency equation) is

1
wkl I J VI J U V m U T kl U V m U T ij = wij I J VI J .
(4.25)
16
We note that now we can not find a nicer form for the U equation, because the vielbein w
does not factorize into two, as it does if we restrict to the Sl(8)/SO(8) coset. We also note
that from the vielbein (4.23) one can rederive the metric (3.31).
For the AdS5 S5 case, the ansatz for the vielbein can be obtained easily via the general
procedure outlined before. From the 10D transformation law, we get

m
(4.26)
+ spin-1/2 terms ,
B 21 11/3 Em
whereas the 5D susy law is

BAB = 2i 1 (15)
AB
ab

a b + spin-1/2 terms
(4.27)
implying an equation for the vielbein:

m,I J a
V
U I U b J = 4i 1 (15)
11/3 Em

ab
AB
VAB .
(4.28)
Here U a I is a matrix in the coset U Sp(8)/SO(6)g . Using (U V m U 1 )ab (U Vn U 1 )ab =

n we can square (4.28) to get back (3.39), as promised. The vielbein then is
4m
AB
= i 1 (15) ab VAB V m,I J U a I U b J

(4.29)
11/3 Em
and the self-consistency equation (the equation for the matrix U ) is
AB
14 1 (15) ab VAB V m,I J U a I U b J V m,I J U a I U b J
AB
= 1 (15) ab VAB .
(4.30)
A comment is in order for the fermionic ansatz for the AdS5 S5 case: in 10D we
already have two gravitinos, which one can take to form a complex gravitino. Then the KK
reduction will have the same form as the general one, but now the Killing spinors are in the
spinorial representation of SO(6)g , or the fundamental representation of SU(4), which is
a 4D complex representation. The 5D gravitinos are in the fundamental of U Sp(8)c , which
is a 8D real representation, so one has a choice of writing the U matrix as an 8 by 8 real
matrix, or a 4 by 4 complex one.
5. Conclusions and discussion

We have shown that the ansatz in [6] can be used for a variety of purposes. By further
truncating the number of fields, we were able to reproduce the anstze of [2124,30]. That
is to be expected since the KK truncation of 11D sugra on AdS7 S4 in [6] is a consistent
one (i.e., satisfies the higher-dimensional equations of motion and susy laws), as are the
truncations found in [2124,30]. As a byproduct of our analysis, we find a lagrangian
formulation of the N = 2 model, which was missing in the literature. This N = 2 model
232
was described until now only through equations of motion, supplemented with the selfduality constraint. The Lagrangian which describes the bosonic sector of the N = 2 model
reads:
q

g2
g71 L7D, N = 2 = R + g 2 2X3 + 2X2 X8 + 5 X1 X + X4 S S
2

q
1 2 i i
g
S F
X F F
+ g71
4
24 2

1
i
i
i
+ + S F F ,
(5.1)
8 3
48 2g
where the three-form A(3) used so far in the N = 2 model is a linear combination of the
three-form S(3) of the maximal gauged sugra model and the ChernSimons three-form (3)
(2.35).
Then we have found that if we make the KK reduction on a sphere Sn of a theory
involving gravity and an antisymmetric tensor F(n) to a theory involving gravity and
scalars, the scalar coset has a submanifold Sl(n + 1)/SO(n + 1)c , and the ansatz for the
line element is
2
= (Y T Y )
dsD
2
(d2)(2n)
1
ds72 + (Y T Y ) 2n TAB
DY A DY B ,
(5.2)
and DY A = dY A + gB AB YB is a covariant derivative. For the

where =
antisymmetric tensor, the ansatz for the dependence on the coset scalars is
An+1

T Y
3n 2
A1
An1 An
aA1 ...An+1 dY . . . dY
Y
A(n1) = 1
Y T Y
p

1 D n
1 ...n g 13n 2 ,
(5.3)
n 2
p
1 ...n dx 1 . . . dx n
F(n) |no T = 13n 2 g

T
Y T2 Y
a
(n + 1) 2
1
1+
n Y T Y
(Y T Y )2

Y T2Y
2
1
,
(5.4)
+ 2a 1
2(2 n)
(Y T Y )2
2 d1
n1 d2
and we see that by making derivatives covariant we can couple to gauge fields as in the
metric. But there are other terms involving the YangMills field strength in the ansatz for
the antisymmetric tensor, as we can easily see in the AdS7 S4 example. By imposing
Bianchi identity on F(n) we can presumably generate these terms. However, it is possible
to miss a separately closed form, gauge-invariant, and which vanishes when the scalar
fields are set to zero. So, the ansatz for the antisymmetric tensor field strength generated
by this method should be checked in the action too, to verify its completeness.
Moreover, we note that the T terms in F(n) contain 1/2x terms, and so we were not
able to reproduce the scalar field kinetic terms in the d-dimensional action, except in the
cases of AdS4 S7 , AdS7 S4 and AdS5 S5 , when these 1/2x terms disappear already
in F(n) .
In the AdS5 S5 case, we got a line element:

2
= 12/3 ds52 + NAB DY A DY B ,
ds10
233
(5.5)
eAB = ( 1 (15))ab AC YC ( 1 (15))ab BD YD , i.e.,

where NAB is the inverse of N
A
eAC NBC = D
+ (YD terms).
N
(5.6)
One can now use this to lift any solution of gauged sugra to 11 dimensions, at least for the
metric.
We were able to obtain the vielbein too, by first fixing a small part of the fermion
ansatz, namely the relation between the D-dimensional and the d-dimensional gravitinos.
0
We found that also in the general case we need a matrix U I I interpolating between the
composite symmetry group indices of the gravitino and the gauge group spinor indices of
the Killing spinors. This matrix appears in the ansatz for the vielbein, which for the case
of the Sl(n + 1)/SO(n + 1) coset reads

1
2ce 1 A 1 B
=
VAB Tr ij U Vn U 1 .
(5.7)
1 d2 Em
i
j
s
In this case, the matrix U obeys the nice equation:
A
(5.8)
U Y/ = v/U, vi = 1n/21 1 i YA ,
which probably has the most general solution very similar to the one found in [6] for the
AdS7 S4 case, but we did not investigate that further.
We have also found the full ansatz for the vielbein in the AdS5 S5 case, as well as
the U equation. While these probably have less practical applications, they will be useful
if one wants to check the consistency of the AdS5 S5 KK reduction, which is the one
thing notably missing in the business of consistent truncations (it is the case most used for
the AdSCFT correspondence). We note that if one does not restrict to the Sl(6)/SO(6)
scalar coset, the equation for the matrix U does not take a nice form, but instead is found
in (4.30).
Note added. After the first version of the paper was posted, we became aware of the
paper [45], where the form of the 5D metric (5.5) was conjectured.
Acknowledgements
We are grateful to Peter van Nieuwenhuizen for discussions and for the the collaboration
to an earlier paper [6], from where most of the work reported here originated. We are also
grateful to Chris Pope for a discussion of his work, which led us to the analysis in Section 2,
and to Radu Roiban for discussions of our results.
Appendix A. Killing spinors on spheres
A Killing spinor (x) on a sphere Sn parameterized by the coordinates x is defined by:
D (x) = cm (x),
(A.1)
234
where we introduced a dimensionful constant m, and c is a another dimensionless constant

which will be fixed soon. Spheres are Einstein spaces, and moreover, are maximally
symmetric:

(A.2)
R mn = m2 em (x) en (x) en (x) em (x) .
Then, the integrability condition reads:
[D , D ] = 14 R mn mn =
m2
= 2c2 m2 ,
2
(A.3)
and we derive that c = i/2. 6 Corresponding to the sign of c we define two sets of Killing
spinors, D = 1/2m . Spheres are the only manifold where we can define both
sets . An explicit representation can be given in terms of stereographic coordinates
(see [44]):
em (z) =
mn (z) =
,
2
1+z
1 im m z
(0).
(z) =
1 + z2
We define further
2m zn + 2n zm
1 + z2
(A.4)
(A.5)
T C,
(A.6)
where C is the charge conjugation matrix: T = C C 1 and C T = C.

Sn
I J
Gauge group
Number of + s
n=2
n=3
n=4
n=5
n=6
n=7
n=8
n=9
+
+
+
I J
I J
IJ
IJ
I J
I J
I J
I J
SO(3) ' SU(2)

SO(4) ' SU(2) SU(2)
SO(5) ' U Sp(4)
SO(6) ' SU(4)
SO(7)
SO(8)
SO(9)
SO(10)
2
4
4
8
8
8
8
16
The labeling index I is in the spinor representation of the gauge group.

I (z)J (z) = J (0)I (0) and the Killing
For example, if T = C C 1 , then
spinors are normalized to either I J or I J (if Cn+1 = C() or C(+) , respectively)

depending on the sign of : I J = ( I J )T = I J . In this case, we can express the
Cartesian coordinates Y A parameterizing the sphere (with Y A in the vector representation
of the gauge group) as:
I J
,
Y A = 0IAJ +
(A.7)
6 In general, using the integrability condition we derive that Killing spinors exist only on Einstein spaces,
namely R = 4c2 (n 1)g .
235
where in even dimensions 0 A = {im 5 , 5 }, and in odd dimensions we use the 0 A in the
chiral representation of SO(n + 1). Substituting (A.5) into (A.7) we can check that with an
appropriate choice of + (0) and (0) we indeed find Y A Y A = 1.
A similar analysis can be done for the case = +1.
Another useful relation is the completeness relation of the Clifford algebra matrices:
= 2[n/2]1 (C C + C C ) + A( + 1)C C ,
(A.8)
where 2A = n 2[n/2]1 (1 + 2[n/2] ).

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
B. de Wit, H. Nicolai, Nucl. Phys. B 243 (1984) 91.

B. de Wit, H. Nicolai, N.P. Warner, Nucl. Phys. B 255 (1985) 29.
H. Nastase, D. Vaman, P. van Nieuwenhuizen, hep-th/9905075.
H. Nastase, D. Vaman, P. van Nieuwenhuizen, hep-th/9911238.
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
W. Muck, K.S. Viswanathan, Phys. Rev. D 58 (1998) 041901.
D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96.
H. Liu, A.A. Tseytlin, Nucl. Phys. B 533 (1998) 88.
G. Chalmers, H. Nastase, K. Schalm, R. Siebelink, Nucl. Phys. B 540 (1999) 247.
G. Arutyunov, S. Frolov, Phys. Rev D 60 (1999) 026004.
H. Liu, A.A. Tseytlin, Phys. Rev D (1999) 086002.
D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Phys. Lett. B 452 (1999) 61.
G. Chalmers, K. Schalm, Nucl. Phys. B 554 (1999) 215, hep-th/9901144.
E. DHoker, D.Z. Freedman, Nucl. Phys. B 544 (1999) 612, Nucl.Phys. B 550 (1999) 261.
E. DHoker, D. Freedman, S. Mathur, A. Matusis, L. Rastelli, hep-th/9903196.
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, hep-th/9905111.
M. Cvetic, M.J. Duff, P. Hoxha, J.T. Liu, H. L, J.X. Lu, R. Martinez-Acosta, C.N. Pope, H. Sati,
T.A. Tran, hep-th/9903214.
M. Cvetic, J.T. Liu, H. L, C.N. Pope, hep-th/9905096.
H. L, C.N. Pope, hep-th/9906168.
M. Cvetic, S.S. Gubser, H. L, C.N. Pope, hep-th/9909121.
E. Cremmer, B. Julia, H. L, C.N. Pope, hep-th/9909099.
M. Cvetic, H. L, C.N. Pope, hep-th/9910252.
M. Volkov, hep-th/9910116.
A.H. Chamseddine, M.S. Volkov, Phys. Rev. D 57 (1998) 6242.
H. L, C.N. Pope, T.A. Tran, hep-th/9909203.
A.H. Chamseddine, W.A. Sabra, hep-th/9911180.
L. Randall, R. Sundrum, hep-th/9905221.
L. Randall, R. Sundrum, hep-th/9906064.
I. Bakas, A. Brandhuber, K. Sfetsos, hep-th/9912132.
M. Wijnholt, S. Zhukov, hep-th/9912002.
R. Kallosh, A. Linde, hep-th/0001071.
K. Behrndt, M. Cvetic, hep-th/9909058, hep-th/0001159.
236
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
M.A. Awada, B.E.W. Nilsson, C.N. Pope, Phys. Rev. D 29 (1984) 334.
B.E.W. Nilsson, Phys. Lett. B 155 (1985) 54.
M. Pernici, K. Pilch, P. van Nieuwenhuizen, Phys. Lett. B 143 (1984) 103.
P.K. Townsend, K. Pilch, P. van Nieuwenhuizen, Phys. Lett. B 136 (1984) 38.
P.K. Townsend, P. van Nieuwenhuizen, Phys. Lett. 125 (41) 1983.
E. Cremmer, B. Julia, Nucl. Phys. B 141 (1979) 141.
E. Cremmer, H. L, C.N. Pope, K.S. Stelle, hep-th/9707207.
P. van Nieuwenhuizen, in: Proc. Trieste Spring School, 1984, p. 239.
A. Khavaev, K. Pilch, N. Warner, hep-th/9812035.

Strongly coupled gravity and duality

C.M. Hull
Physics Department, Queen Mary and Westfield College, Mile End Road, London E1 4NS, UK
Received 1 May 2000; accepted 22 May 2000
Abstract
A strong coupling limit of theories whose low-energy effective field theory is 5-dimensional N = 8
supergravity is proposed in which the gravitational coupling becomes large. It is argued that, if
this limit exists, it should be a 6-dimensional theory with (4, 0) supersymmetry compactified on
a circle whose radius gives the 5-dimensional Planck length. The sector corresponding to the D = 5
supergravity multiplet is a (4, 0) D = 6 superconformal field theory based on the (4, 0) multiplet with
27 self-dual 2-forms, 42 scalars and, instead of a graviton, a fourth-rank tensor gauge field satisfying
a self-duality constraint. The superconformal field theory has 32 supersymmetries and 32 conformal
supersymmetries and its dimensional reduction gives the maximal supergravity in five dimensions.
Electromagnetic duality generalises to a gravitational triality. 2000 Elsevier Science B.V. All rights
reserved.
1. Introduction
The concept of a spacetime manifold as the arena for physics may have to be modified
at some very small distance scale, L, but at larger scales it should give a good description.
There are a number of scenarios in which gravity could become strong at some much larger
length scale, l L. For example, in M-theory L might be the 11-dimensional Planck
length while l could be the length scale l = 2/D2 giving the gravitational coupling
constant for some D-dimensional gravity theory arising from a compactification or
brane-world scenario. At energy scales E such that 1/ l E 1/L, gravity is strongly
coupled but one might still have some sort of conventional spacetime description, and
in particular there could be a dual field theory describing this strong coupling phase.
Alternatively, it could be that there is some dual theory describing the gravitational theory
at all energies above some lower bound E > 1/ l without any cut-off at a higher scale.
In M-theory, there would be dual phases of M-theory, one of which provided a good
description at low energies and the other in a certain high energy regime, and each would
have its own field theory limit.
The aim is then to find a gravitational analogue of some of the weakstrong coupling
dualities that have been found in field theories and string theories. At first sight, it would
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 2 3 - 0
238
C.M. Hull / Nuclear Physics B 583 (2000) 237259
seem that such dual theories of gravity would be unlikely to exist, but the close relations
between gravity and gauge theories and the implications such a dual description could have
for quantum gravity suggest that it could be worthwhile to investigate this possibility.
A useful example to try and generalise is 5-dimensional maximally supersymmetric
YangMills theory. It is non-renormalisable and so new physics should emerge at short
distances. Its strong coupling limit is a 6-dimensional (2, 0) supersymmetric theory,
compactified on a circle, in which the gauge field is replaced by a 2-form gauge field
with self-dual field strength, and the 5 scalar fields are all promoted to scalar fields in
6 dimensions [13]. The D = 6 theory is believed to be a nontrivial superconformally
invariant quantum theory [3] and the D = 5 gauge coupling gYM arises as the radius of the
2 = R. The relationship between the D = 5 and D = 6 theories is straightforward
circle, gYM
to establish for the free case in which the YangMills gauge group is abelian, but in
the interacting theory the 6-dimensional origin of the D = 5 non-abelian interactions is
mysterious; there are certainly no local covariant interactions that can be written down that
give YangMills interactions when dimensionally reduced [4]. Nonetheless, the fact that
these D = 5 and D = 6 theories arise as the world volume theories of D4 and M5 branes
respectively [5,6] gives strong support for the existence of such a 6-dimensional origin for
the gauge interactions. The (2, 0) conformal theory also arises in the compactification of
the IIB string on K3 [7] and as the zero-slope limit of a little string theory [810].
Interactions of the mysterious type that are believed to occur in the (2, 0) conformal
field theory will be referred to here as M-interactions. Given that such interactions with
no conventional field theory formulation arise in the M5-brane world-volume theory, it is
natural to ask whether similar M-interactions could arise elsewhere; given that they arise in
one corner of M-theory, it seems reasonable to expect that there are other corners in which
such phenomena occur.
The theory that will be considered here is five-dimensional N = 8 supergravity
(ungauged), which has a global E6 symmetry and a local Sp(4) = USp(8) symmetry [11].
It is non-renormalisable, and will be regarded as arising as a massless sector of some
consistent theory, such as M-theory compactified on a 6-torus, in which the global E6
symmetry is broken to a discrete U-duality subgroup [12]. The massless bosonic fields
consist of a graviton, 27 abelian vector fields and 42 scalars. The action is

Z
1
1 2
5
(1.1)
S = d x g 3 R F + ,
l
4l
where l = 2/3 is the 5-dimensional Planck length. If this does have a dual at strong
gravitational coupling, i.e. a limit as l , then it is natural to look first for a theory
with 32 supersymmetries and with E6 Sp(4) symmetry, as the simplest possibility would
be if these symmetries survived at strong coupling. The arguments used in this paper will
be similar to those used in [1214] for extrapolating to large values of dimensionless string
couplings and moduli, and it will be assumed that all symmetries are preserved and BPS
states are protected and survive as the coupling l is increased. The symmetries then impose
stringent constraints on what can happen, and one can then seek checks on the predicted
strong-coupling dual.
239
Comparison with the D = 5 gauge theory case suggests seeking a 6-dimensional theory
compactified on a circle in which the 27 D = 5 vector fields are replaced with 27 self-dual
2-form gauge fields in D = 6. Indeed, the decomposition of N = 8 supergravity into N = 4
2 = l, and the
multiplets includes five N = 4 vector multiplets with coupling constant gYM
strong coupling limit of each should give a (2, 0) 6-dimensional tensor multiplet, as will be
argued in Section 2. However, the 27 vector fields of the N = 8 theory fit into an irreducible
representation of E6 , and so if E6 symmetry is to be maintained and if some of the vector
fields become self-dual 2-form gauge fields in 6 dimensions, then all of them should.
Furthermore, all the fields fit into an irreducible multiplet of the N = 8 supersymmetry
algebra, so that if the 32 supersymmetries survive at strong coupling and some of the fields
become 6-dimensional, then the whole theory should become 6-dimensional. As in the
gauge theory case, all the scalar fields should survive at strong coupling and so should lift
to 42 scalars in 6 dimensions. If the D = 6 theory were conformally invariant, then the
D = 5 gravitational coupling l could arise from the radius of the circle R.
This then suggests that the strong coupling dual field theory should be a superconformal
theory in six dimensions with 32 ordinary supersymmetries (and a further 32 conformal
supersymmetries), and should have 42 scalars and 27 self-dual 2-form gauge fields.
The unique supergravity theory in 6 dimensions with 32 supersymmetries fails on every
one of these requirements with the obvious exception of having 32 supersymmetries.
Remarkably, such a conformal supermultiplet in six dimensions with 32 supersymmetries,
Sp(4) R-symmetry, 42 scalars and 27 self-dual 2-form gauge fields does exist and so is
an immediate candidate for a strong coupling dual. However, instead of a graviton, it has
an exotic fourth-rank tensor gauge field satisfying a self-duality constraint. The relation
between linearised 5-dimensional N = 8 supergravity and this free 6-dimensional theory
will be established here, but there are no covariant local interactions in D = 6 that could
give rise to the D = 5 supergravity interactions, so that their 6-dimensional origin would
have to be some form of M-interactions. In this case there is no analogue of the M5-brane
argument to support this, although the M5-brane case does set a suggestive precedent.
The purpose of this paper is to investigate the possibility that the strong coupling limit of
the D = 5 theory is this exotic 6-dimensional theory. In the context in which the D = 5
supergravity is the massless sector of M-theory on T 6 , then the D = 6 superconformal
theory would be a field theory sector of a new 6-dimensional phase of M-theory.
2. D = 5 super-YangMills theory and the D = 6 (2, 0) tensor theory

The N = 4 YangMills theory in five dimensions has a vector field A, 5 scalars I
and four 4-component fermions a satisfying a symplectic Majorana reality condition,
all taking values in the adjoint of the gauge group. There is a global Spin(5) = Sp(2)
R-symmetry under which the scalars transform as a 5 and the fermions as a spinor 4. The
action is
Z

1
(2.1)
d5 x 14 F 2 + 12 (D)2 + .
S = 2
gYM
240
2
The coupling constant gYM is dimensionful and l = gYM
is a length scale. The theory
is non-renormalisable, and so new physics is expected at energies E > 1/ l. The
5-dimensional gauge theory is then a low-energy effective theory and it will be supposed
that the D = 5 super-YangMills multiplet arises as part of a consistent theory that can be
extrapolated to scales E > 1/ l. For example, it could arise as part of the 5-dimensional
heterotic string, or on a stack of D4-branes in the IIA string, and in both these cases there
would be in addition supergravity fields and an infinite number of massive fields in the full
theory.
Consider the behaviour of the theory as the coupling gYM is increased. The simplest
possibility would be for the theory to continue to be invariant under 16 supersymmetries
and the Spin(5) R-symmetry, for all 5 scalar fields to remain so that the theory has the
same moduli space, and for the BPS spectrum to remain the same, although their masses
will depend on gYM .
The theory has 1/2-supersymmetric BPS solitons or 0-branes given by lifting selfdual YangMills instantons in 4 Euclidean dimensions to soliton world-lines in 4 + 1
dimensions. These solitons have mass
|n|
,
2
gYM
(2.2)
where n is the instanton number, so that these become light as gYM becomes large.
A supersymmetric theory in 6 dimensions compactified on a circle of radius R gives a
tower of 1/2 supersymmetric multiplets of BPS massive states with momentum n/R in the
circular dimension for integers n, with masses
|n|
.
(2.3)
R
In [1], it was suggested that the light solitons in D = 5 YangMills should be reinterpreted
2 . Theories
as KaluzaKlein modes for a 6-dimensional theory on a circle of radius R gYM
in six dimensions with 16 supersymmetries can have either (1, 1) supersymmetry or (2, 0)
supersymmetry, 1 but only (2, 0) supersymmetry allows Spin(5) R-symmetry and keeping
the same number of scalars, and this gives a unique candidate for a strong coupling limit
that is consistent with the assumptions, which is the superconformal (2, 0) theory. In the
abelian case, this is the free field theory of the (2, 0) tensor multiplet in 6 dimensions. This
multiplet consists of 5 scalars I , four 4-component symplectic MajoranaWeyl fermions,
and a 2-form gauge field BMN satisfying a self-duality condition H = H where H is the
field strength H = dB. This theory has no dimensional coupling constants (after rescaling
the fields) and is superconformal invariant with superconformal group OSp (8/4). This
has bosonic subgroup SO (8) USp(4) = SO(6, 2) Sp(2) which is the product of
the conformal group and the R-symmetry group, together with 32 fermionic generators.
Dimensional reduction on a circle of radius R gives the D = 5 abelian super-YangMills
2 = R.
theory with action (2.1) and coupling constant gYM
M
1 (p, q) supersymmetry in 6 dimensions has p right-handed symplectic MajoranaWeyl supercharges and q

left-handed ones.
241
This leads to the conjecture that the strong coupling limit of the D = 5 superYangMills theory is an interacting 6-dimensional superconformal theory with (2, 0)
supersymmetry and OSp (8/4) superconformal symmetry compactified on a circle of
2 [13]. In the abelian case, it is easily verified that dimensional reduction
radius R = gYM
of the (2, 0) theory gives the D = 5 theory, but there are no local covariant interactions
that can be written for the tensor multiplet that give the non-abelian D = 5 interactions
on dimensional reduction. Nonetheless, there are arguments that such an interacting theory
in D = 6 should exist, based on the assumption that M-theory is consistent. For example,
it arises as part of the world-volume theory for a stack of M5-branes [5,6] and there is a
decoupling limit leaving precisely the (2, 0) interacting superconformal theory [13]. The
consistency of this limit and the existence of a nontrivial quantum theory in six dimensions
depends crucially on the conformal invariance of the (2, 0) theory. An important check on
this interpretation is that when the original 4 spatial dimensions are compactified on T 4 ,
the theory has an SL(5, Z) invariance [1,2], while for the uncompactified theory in 4 + 1
dimensions there is evidence that the theory has 5 + 1 dimensional Lorentz invariance in
the limit gYM . The M5-brane realisation leads to a matrix theory construction of the
interacting theory in discrete light-cone gauge [15].
The interacting theory then has some M-interactions which give rise to the 5-dimensional non-abelian interactions but do not have a conventional field theory formulation in
6 dimensions. It is not clear whether the tensor multiplet provides the correct degrees of
freedom for the interacting theory, or whether something more exotic such as string fields
might be needed [3]. It was originally suggested that the interacting (2, 0) theory might be
a theory of tensionless noncritical strings [7], but later it was proposed that it could be a
superconformal quantum field theory of a more conventional type (see, e.g. [16]).
The scalars in the action (2.1) have dimension one, but in 6-dimensional superconformal field theory the canonical dimension for scalars is two. Then defining the dimension-2
scalars
1
(2.4)
I = 2 I ,
gYM
the 6-dimensional action
Z
1
d6 x ()2 +
(2.5)
has no dimensional parameters but gives the scalar part of (2.1) on reducing on a circle of
2 . The limit leading to the (2, 0) theory is a strong coupling limit in which
radius R = gYM
2
2 fixed, so that the scalars are all scaled with g 2 . The
gYM keeping I = I /gYM
YM
(2, 0) theory has no dimensionful parameters and scales are set by the expectation values
of the scalars .
The BPS states of the D = 5 theory include the instantonic 0-branes, electrically
charged 0-branes and magnetically charged strings. The electric 0-branes are the W-boson
multiplets with mass M where is the magnitude of the scalar expectation value,
2 .
= |h I i|, and the BPS strings have tension = /gYM
The five-dimensional N = 2n superalgebra with automorphism group Sp(n) is
242

Qa , Qb = ab C P + ab C K

ab
,
+ C Zab + C Z ab + 12 C Z
(2.6)
where , = 0, 1, . . . , 4 are spacetime indices, = 1, . . . , 4 are spinor indices, a =

1, . . . , N are Sp(n) indices, C is the charge conjugation matrix and ab is the symplectic
invariant of Sp(n). The supercharges Qa are symplectic Majorana spinors satisfying
(Q)a = C ab Qb .
(2.7)
For the N = 4 super-YangMills theory, the 5 central charges

Z ab = Z ba with
R
ab
I
ab Z = 0 are proportional to the five electric charges q tr(F I ) and are carried
by massive vector multiplets. There is no vector field coupling to the singlet central charge
K, which is a topological charge carried by the instantonic 0-branes and is proportional to
the instanton number. The spatial components Ziab (i, j = 1, . . . , 4) of Zab are the magnetic
charges carried by the magnetically charged strings. The other charges on the right-handside of (2.6) are 2, 3 and 4-brane charges [17]; for example, a vortex solution in 2 + 1
dimensions lifts to a 2-brane in 4 + 1 dimensions with charge Zij .
There are two distinct types of 1/2 supersymmetric BPS 0-brane representations [18].
The first has central charges K = 0 and Z ab = iq 3 2 for some charge q, breaking the
R-symmetry from Spin(5) to Spin(4), and is a massive vector multiplet with mass M = |q|.
The second has Z = 0 and K 6= 0 and is a massive self-dual tensor multiplet with mass
M = |K| [18] and preserves the full Spin(5) R-symmetry. The W-bosons fit into massive
vector multiplets and the instantonic 0-branes fit into massive self-dual tensor multiplets.
The limit in which the tower of instantonic 0-branes becomes massless is necessarily a
(2, 0) theory in 6 dimensions; this is because the KaluzaKlein modes for a (2, 0) tensor
theory fit into massive tensor multiplets while those for a (1, 1) theory in 6 dimensions
would fit into massive vector multiplets. The central charge K becomes the 6th component
of momentum, P M = (P , K), while the central charges Z ab fit together with the D = 5
ab
= (Zab , Z ab ).
string charges Zab to give the D = 6 string charges ZM
The six-dimensional (2, 0) theory has BPS strings coupling to the self-dual antisymmetric tensor gauge fields of tension where is the the magnitude of the expectation
value of I . On dimensional reduction, the unwrapped strings give the magnetic strings in
2 while the wrapped strings give the W-bosons with
five dimensions of tension = /gYM
mass M R = . Thus the BPS spectra match, and the strong coupling limit is the one
in which gYM while the tensions of the BPS strings are kept fixed.
At the origin of moduli space at which the scalar expectation values all vanish, all
W-bosons become massless and the full gauge symmetry of the D = 5 YangMills theory
is restored and, as 0, the tensions of the magnetically charged strings approach zero.
In the six dimensional theory, the tensions of the BPS strings also approach zero as 0.
The nature of the theory at such points is unclear. For example, one guess might be that it
could be described by some kind of string field theory, but it was argued in [3] that such
a description would overcount the degrees of freedom.
These arguments should apply to 5-dimensional YangMills theories independently
of how they are embedded in other theories. The (2, 0) superconformal field theory is
243
believed to be a consistent quantum theory in its own right, and so can provide a complete
formulation of the theory described at weak coupling by an effective 5-dimensional Yang
Mills theory. Conversely, any theory whose effective description is given by 5-dimensional
N = 4 YangMills theory plus other N = 4 multiplets will, in the limit in which the Yang
Mills coupling becomes strong while the magnetic string tensions remain constant, be the
six-dimensional (2, 0) conformal field theory plus other supermultiplets. For example, the
dynamics of a stack of m parallel D4-brane is described by D = 5, N = 4 YangMills
theory with SU(m) gauge symmetry plus a tower of massive D = 5, N = 4 multiplets,
coupled to a 10-dimensional bulk theory consisting of supergravity plus an infinite number
of massive multiplets. At strong coupling, the D4-branes become M5-branes wrapped on a
2
whose dynamics are described by the D = 6 (2, 0) conformal
circle of radius R = gYM
field theory theory, plus a tower of massive D = 6 (2, 0) supermultiplets, coupled to
an 11-dimensional bulk theory whose massless sector is 11-dimensional supergravity
compactified on a circle of radius R. As another example, the heterotic string compactified
on T 5 is described by D = 5, N = 4 YangMills theory plus D = 5, N = 4 supergravity
plus a tower of massive D = 5, N = 4 multiplets. These include massive string modes
and KaluzaKlein modes. At strong coupling this becomes the six-dimensional (2, 0)
supersymmetric string theory obtained by compactifying type IIB string theory on K3 [7],
2 . The 6-dimensional
further compactified to five dimensions on a circle of radius R = gYM
(2, 0) theory at generic points in moduli space is (2, 0) supergravity coupled to (2, 0) tensor
multiplets, together with an infinite number of massive (2, 0) multiplets, with tensionless
strings appearing at special points in the moduli space [7].
In both cases, each massless vector multiplet in D = 5 becomes a D = 6 (2, 0) tensor
multiplet at strong coupling at generic points in moduli space, with tensionless strings
occurring at special points. It is natural to expect the same to happen for N = 4 vector
multiplets in any N > 4 supersymmetric theory in five dimensions, and in particular that
the abelian N = 4 vector multiplets occurring in D = 5, N = 8 supergravity theory become
six-dimensional (2, 0) tensor multiplets.
3. N = 8 supergravity in five dimensions

Maximal supergravity theories have a local symmetry group H and a global symmetry
G [19], broken to a discrete U-duality subgroup in the quantum theory [12]. The scalars
take values in the coset space G/H and, if the local H symmetry is fixed by choosing
the symmetric gauge, then a diagonal R-symmetry subgroup H of G H is manifest.
The N = 8 supergravity theory in five dimensions [11] has local H = Sp(4) = USp(8)
symmetry and a global G = E6 symmetry. In symmetric gauge, the N = 8 supergravity
multiplet decomposes into the following representations of the Spin(3) Sp(4) little group:
28 = (5, 1) + (3, 27) + (1, 42) + (4, 8) + (2, 48).
(3.1)
The vector fields transform as the 27 of E6 while the 42 scalars take values in E6 /Sp(4).
The Einstein-frame action takes the form (1.1), with l the 5-dimensional Planck length.
244
If the theory is obtained from 11 dimensional supergravity compactified on a 6-torus of

volume L6 , then l is given in terms of L and the 11-dimensional Planck length l11 by
3 /L2 .
l = l11
The N = 8 superalgebra has automorphism group Sp(4) and is given by (2.6) with
a, b = 1, . . . , 8. The algebra with scalar central charges only is

a b
(3.2)
Q , Q = ab C P + C Z ab + ab K .
The 27 central charges Z ab satisfy Z ab = Z ba , ab Z ab = 0, and are the electric charges
for the 27 vector fields (dressed with scalars). There are 28 central charges but only
27 vector fields, and K is not the conserved charge for any gauge field. However, on
dimensional reduction from 5 to 4 dimensions, K becomes one of the 28 magnetic charges
of D = 4, N = 8 supergravity, and is the one coupling to the gravi-photon field (i.e. the
electromagnetic field from the dimensional reduction of the metric). U-duality requires
that 1/2-supersymmetric states with M = |K| occur in the D = 4 BPS spectrum, and K
is quantized [12]. However, these D = 4 states must arise from 1/2-supersymmetric states
in D = 5, carrying the central charge that gives the KaluzaKlein magnetic charge in four
dimensions. They must then be 1/2-supersymmetric D = 5 states with M = |K|, and K
should be quantized, so that K n/ l where n is an integer.
Thus BPS states carrying the charge K should be present in the D = 5 spectrum.
A definition of the charge K was given in [17], where it was shown that it indeed arises in
the superalgebra (3.2). Examples of 1/2-supersymmetric supergravity solutions carrying
the charge K are metrics of the form N R where N is a self-dual gravitational instanton
[20] in 4 Euclidean dimensions and R is time, with all other fields trivial [17,21]. If N is
the multi-Taub NUT instanton, the solution has an interpretation as a multi KaluzaKlein
monopole space [22,23] and K is the KaluzaKlein monopole charge. If N is an ALE
space, the solution can be interpreted as representing a set of 0-branes or ALE-branes [17].
The BPS representations in D = 5 were classified in [18]. A nonzero value for Z breaks
the E6 to a subgroup and BPS states carrying the charge Z will have masses dependent on
the 42 scalars, through standard BPS formulae, as in, e.g. [24]. However, if Z is zero and
K is nonzero, the E6 is unbroken (K is an E6 singlet) and BPS states with mass M = |K|
preserve half the supersymmetry and have a mass that is independent of all the scalar fields
(in Einstein frame). Such states have Einstein-frame mass
M=
c|n|
l
(3.3)
for some numerical constant c and integer n. Comparing with the D = 4 BPS spectrum
shows that states should occur for all values of the integer n. For the gravitational instanton
solutions, n is a topological number (essentially the NUT charge or instanton number) [17].
The mass in D = 5 string frame is
Ms =
c|n|
lg 2/3
with an unusual power of the string coupling g.
(3.4)
245
The BPS spectrum is then expected to have M = |K| BPS states with (3.3) for all n
which become light as l and the examples of [1,13] suggest identifying them with
KaluzaKlein modes for a 6-dimensional theory compactified on a circle of radius R l.
Then the charge K would become the 6th component of the momentum and these states
becoming light would not break any of the E6 Sp(4) duality symmetry. Of course, this
suggestion requires many checks. However, if this limit does give a 6-dimensional theory,
it can be expected to give a theory with 32 supersymmetries and E6 Sp(4) symmetry.
These BPS states fit into massive D = 5 multiplets containing 42 massive scalars and 27
massive self-dual 2-form fields B, satisfying dB = m B, in the 27 of E6 [18] (see also
Section 8). This is precisely what is needed for the KaluzaKlein modes of a massless
6-dimensional theory with 27 self-dual 2-forms and 42 scalars. Then if these instantonic
0-branes are to be interpreted as KaluzaKlein modes of a 6-dimensional theory with 32
supersymmetries, then the D = 6 theory will have 27 self-dual 2-forms and 42 scalars, as
demanded in the introduction.
BPS states in M-theory or string theory have associated supergravity solutions which
in some cases are singular and in others are nonsingular. The 1/2-supersymmetric
supergravity solutions carrying the K-charge with Z = 0 include the multi-instanton
solutions N R. The multi-instanton metric of [20] has Dirac (or Misner) string
singularities, although the spin-connection is well-defined. The space is asymptotic to
flat R4 (the curvature is finite and falls off asymptotically, even though the metric is illdefined). A nonsingular metric is obtained if the charges of the sources are all equal, and
suitable identifications are made to give an ALE space which is like R4 / asymptotically,
for a discrete group (which is Zn for the explicit solutions of [20]). Without the discrete
identifications, the solutions with various K charges are asymptotic to R4 but with singular
metrics. However, as will be discussed in Section 6, these singularities can be avoided by
using dual variables. It will be assumed that the BPS states with M = |K| can be viewed
as 0-branes and can be identified with KaluzaKlein modes.
4. Theories with 32 supersymmetries in six dimensions

The considerations of previous sections have suggested looking for a conformal theory
in six dimensions with 32 supersymmetries, E6 Sp(4) symmetry, 27 self-dual 2-forms
and 42 scalars. In this section the various D = 6 multiplets with 32 supersymmetries
will be discussed. In six dimensions, the (p, q) superalgebra has p right-handed
symplectic MajoranaWeyl supercharges and q left-handed ones, with automorphism
group Sp(p) Sp(q). The massless representations decompose into representations of
the little group SU(2) SU(2) Sp(p) Sp(q), and a list of physically acceptable
representations (reducing to D = 4 representations with spin not greater than 2) is given
in [25]. For 32 supersymmetries, there are three possibilities, (2, 2), (3, 1) or (4, 0)
supersymmetry (together of course with (1, 3) and (0, 4)). 2 The (2, 2) multiplet has the
SU(2) SU(2) Sp(2) Sp(2) content
2 The possibility of (3, 1) and (4, 0) multiplets in D = 6 was noted in [26].
246
(3, 3; 1, 1) + (1, 3; 5, 1) + (3, 1; 1, 5) + (1, 1; 5, 5) + (2, 2; 4, 4)

+ (2, 3; 4, 1) + (3, 2; 1, 4) + (2, 1; 4, 5) + (1, 2; 5, 4)
(4.1)
and is the maximal supergravity multiplet in 6 dimensions, obtained by reducing 11dimensional supergravity on a 5-torus. The supergravity has duality symmetries G =
SO(5, 5) and H = SO(5) SO(5) Sp(2) Sp(2), and the graviton is in the (3, 3; 1, 1)
representation.
The (3, 1) multiplet has the SU(2) SU(2) Sp(3) Sp(1) content [25]
(4, 2; 1, 1) + (3, 1; 6, 2) + (1, 1; 140, 2) + (2, 2; 14, 1)
+ (3, 2; 6, 1) + (2, 1; 14, 2) + (1, 2; 140, 1).
(4.2)
The (3, 1) multiplet has 28 scalars in the (140 , 2) representation of Sp(3) Sp(1), and the
unique coset structure that this is consistent with is
F4
Sp(3) Sp(1)
(4.3)
suggesting that G = F4 and H = Sp(3) Sp(1) in this case; it is remarkable that the
exceptional group F4 plays a role here (in the noncompact form with maximal compact
subgroup Sp(3) Sp(1)). There are 14 vector fields and 12 self-dual 2-forms.
The (4, 0) multiplet has the SU(2) SU(2) Sp(4) content [25]
(5, 1; 1) + (3, 1; 27) + (1, 1; 42) + (4, 1; 8) + (2, 1; 48).
(4.4)
The (4, 0) multiplet is remarkably simple, and has all the properties demanded in the
introduction: there are 27 self-dual 2-forms and 42 scalars, it has an Sp(4) R-symmetry and
the spectrum is consistent with rigid G = E6 invariance (with the 2-forms transforming in
the 27). Moreover, the free theory based on this multiplet is a superconformally invariant
theory, with conformal supergroup OSp (8/8). This has bosonic subgroup USp(8)
SO (8) = Sp(4) SO(6, 2) and 64 fermionic generators. Thus this multiplet satisfies all
the conditions required for a strong-coupling limit of the D = 5 supergravity, and a theory
of this multiplet with M-interactions is the candidate dual theory.
However, both the (3, 1) and (4, 0) multiplets have no graviton (which would be in the
(3, 3) of SU(2) SU(2)) but instead have exotic tensor gauge fields. The (4, 0) multiplet
has a gauge field whose physical degrees of freedom are in the (5, 1) representation of
SU(2) SU(2) while the (3, 1) multiplet has a field in the (4, 2) representation. These
will be discussed in the next section, and the covariant gauge fields with these degrees of
freedom will be given.
5. Exotic tensor gauge fields

In discussing the representations under the SU(2) SU(2) little group for massless
particles in 6 dimensions, it is useful to introduce indices A, B = 1, 2 for the first SU(2)
and indices A0 , B 0 = 1, 2 for the second, together with SO(4) vector indices i, j = 1, . . . , 4,
raised and lowered with the metric ij . The translation between the two types of index uses
247
the sigma matrices iAB , and the indices A, B 0 can be combined into a 4-component spinor
index = (A, A0 ).
The (4, 0) supermultiplet has fields in the SU(2) SU(2) representations (s, 1) for s =
1, 2, 3, 4, 5 corresponding to rank s 1 totally symmetric tensors A1 ...As1 = (A1 ...As1 ) .
The (1, 1) is a scalar and the (2, 1) is a chiral spinor. The (3, 1) representation is
a symmetric bi-spinor BAB = BBA , corresponding to a self-dual 2-form Bij = Bj i ,
Bij = 12 ij kl B kl ,
(5.1)
while the (1, 3) representation BA0 B 0 is an anti-self-dual 2-form.

The (5, 1) representation is a symmetric 4th rank SU(2) tensor CABCD = C(ABCD)
which corresponds to a 4th rank tensor Cij kl of SO(4) with the same algebraic properties
as the Weyl tensor:
Cij kl = Cj i kl = Cij lk = Ckl ij ,
C[ij k]l = 0, C k j kl = 0.
(5.2)
In addition it is self-dual:
Cij mn = 12 ij kl C kl mn = 12 Cij mn klmn
(5.3)
or C = C = C.
The (4, 1) representation gives a spinor field ABC which translates to a spinor-valued
2-form ij A which is self-dual:
ij A = 12 ij kl Akl .
(5.4)
This is a spinor-valued self-dual 2-form field ij ( = ) satisfying the chirality
condition 0 ij = ij , or equivalently ( j ) ij = 0.
The (3, 1) multiplet has a field DABCD 0 = D(ABC)D 0 in the (4, 2) representation. This
corresponds to a 3rd rank tensor Dij k satisfying
Dij k = Dj i k ,
Dij = 0,
D[ij k] = 0.
(5.5)
In addition it satisfies the self-duality constraint

Dij k = 12 ij lm D lm k .
(5.6)
It is straightforward to find the free covariant fields in six dimensions corresponding to

these degrees of freedom. The (3, 1) representation arises from a 2-form gauge field BMN
(where M, N = 0, 1, . . . , 5) with the gauge invariance
BMN = [M N]
(5.7)
and invariant field strength 3-form H = dB satisfying the self-duality constraint:

HMNP = 16 MNP QRS H QRS .
(5.8)
The (5, 1) representation arises from a gauge field CMN P Q with the algebraic properties
of the Riemann tensor:
CMN P Q = CNM P Q = CMN QP = CP Q MN ,
and the gauge symmetry:
C[MNP ]Q = 0,
(5.9)
248
CMN P Q = [M N]P Q + [P Q]MN 2[M NP Q]
(5.10)
with parameter MP Q = MQP . The invariant field strength is

GMNP QRS =
1
36 (M S CNP RS
+ ) = [M CNP ] [QR,S] ,
(5.11)
so that
GMNP QRS = G[MNP ] [QRS] = GQRS MNP
(5.12)
and satisfies the self-duality constraint:

GMNP QRS = 16 MNP T U V GT U V QRS
(5.13)
or G = G = G.
The (4, 2) representation arises from a gauge field DMN P satisfying
DMN P = D[MN] P ,
D[MN P ] = 0
(5.14)
with the gauge symmetry

DMN P = [M N]P [M NP ]
(5.15)
with parameter NP . The invariant field strength is

SMNP QR = [M DNP ] [Q,R]
(5.16)
and satisfies the self-duality constraint

T UV
.
SMNP QR = 16 MNP T U V SQR
(5.17)
Similarly, the covariant description of the spinor-valued 2-form ij in the (4, 1)

representation of SU(2) SU(2) is a fermionic 2-form gauge field which is also a Weyl
= [MN]
where is now a 4-component D = 6 Weyl spinor index.
spinor, satisfying MN
The gauge symmetry is
MN = [M N]
(5.18)
. The field strength

with parameter a spinor-vector N
= [P MN]
MNP
(5.19)
satisfies the self-duality constraint
= 16 MNP T U V T U V .
MNP
(5.20)
The field in the (4, 0) multiplet is in the (4, 1; 8) of SU(2) SU(2) Sp(4) and is a field
a carrying an Sp(4) index a = 1, . . . , 8, and satisfying a symplectic MajoranaWeyl
MN
reality condition.
The other representations of SU(2) SU(2) occurring in the (p, q) supermultiplets have
more conventional covariant representations. The (3, 3) comes from a graviton hMN , the
, the (2, 1) from a
(2, 2) from a vector gauge field, the (3, 2) from a chiral gravitino M
chiral spinor and the (1, 1) from a scalar.
249
6. Electromagnetic duality
In D dimensions, an abelian 1-form gauge field A can be dualised to an n-form
e1 2 ...n where
A
n=D3
(6.1)
e = F where F = dA, F
e = d A.
e Electric charges lead to a well-defined
with field strength F
e
potential A but A has Dirac string singularities. Magnetically charged n 1 branes give
e is well-defined but A has Dirac string singularities. Magnetic charges
fields for which A
can be accommodated by allowing A to be a connection for a nontrivial bundle, or by
e or by allowing string singularities in A. The
switching to a description in terms of A,
generalisation of this duality to non-abelian gauge theory is problematic, and in particular
e if n > 1.
one cannot write down covariant local non-abelian interactions for A
This picture can be generalised to gravity; the results will be summarised here and
developed in more detail elsewhere. Consider the free theory given by taking the D
dimensional metric
g = + h
(6.2)
and linearising Einsteins equations in the fluctuation h about the Minkowski metric .
Similarly to the gauge theory case, in the free theory one can dualise one of the indices on
h to obtain a field
D1 2 ...n = D[1 2 ...n ]
(6.3)
or both indices to obtain a gauge field

C1 2 ...n 1 2 ...n = C[1 2 ...n ][1 2 ...n ] ,
(6.4)
where again n = D 3. For example, the gauge field C has a field strength
G1 2 ...n+1 1 2 ...n+1
(6.5)
defined by a formula similar to (5.11), related to the linearised Riemann curvature

R (h) by
G1 2 ...n+1 1 2 ...n+1 = 14 1 2 ...n+1 1 2 ...n+1 R
(6.6)
or G = R. As in the gauge theory case, the generalisation to the interacting theory is

problematic, and one cannot write covariant local gravitational interactions for the dual
fields C or D.
These dualities are easily derived in physical gauge. Introducing transverse coordinates
i, j = 1, . . . , D 2, a vector field has physical light-cone gauge degrees of freedom Ai in
the vector representation of the little group SO(D 2) and can be dualised to an n = D 3
form
ej1 ...jn = j1 ...jn i Ai
A
(6.7)
which can then be identified with the physical degrees of freedom of an n-form gauge
e = d. The physical degrees of freedom are in
e1 ...n , with gauge invariance A
field A
250
e give equivalent field theory

equivalent representations of the little group and so A and A
representations of the physical degrees of freedom.
A physical gauge graviton is a transverse traceless tensor hij satisfying
hij = hj i ,
hii = 0.
(6.8)
One or both of the indices on hij can then be replaced by n antisymmetric indices to give
a dual form. Choosing D = 5, n = 3 for simplicity, one can define
Dij k = ij l hl k
(6.9)
so that (6.8) implies the conditions

Dij k = Dj i k ,
Dij = 0,
D[ij k] = 0 or
Cij kl = ij m kln hmn
(6.10)
(6.11)
which then satisfies

Cij kl = Cj ikl = Cij lk = Cklij ,
C[ij k]l = 0, C k j kl = 0.
(6.12)
These are precisely the conditions (5.2) and (5.5) found previously in another context,
and covariant gauge fields C and D which reduce to Cij kl or Dij k can be found
as in the last section. These then give three equivalent field representations of the same
degrees of freedom, and the duality of gauge theory generalises to a triality of (linearised)
gravitational theories.
e with
In gauge theory, electric charges couple naturally to A but give a dual potential A
e
Dirac string singularities while magnetic sources couple to A but give an A with string
singularities. This generalises to linearised gravity, in which there are three types of fields,
h, C and D, and if a source leads to a nonsingular configuration for one of the three fields
h, C, D, then that configuration can in general have string singularities when expressed in
terms of one of the other two fields. It seems natural to suppose that sources for all three
fields could arise in the full theory. This would mean that, when formulated in terms of h
alone, some string singularities in h should be anticipated.
Consider in particular the BPS states carrying the charge K that were discussed in
Section 3. Reducing to D = 4 gives magnetic monopoles which have vector potentials A
with Dirac string singularities and, as A appears explicitly in the KaluzaKlein ansatz for
the D = 5 metric, the D = 5 metric will also have string singularities. In D = 4, the string
e and this appears in the
singularities can be avoided by using the dual vector potential A,
e
KaluzaKlein ansatz for the D = 5 dual field D , as Dm4 4 Am (, = 0, 1, . . . , 4 and
m, n = 0, 1, 2, 3). This suggests that in a dual formulation of gravity using D instead
of h , states carrying the charge K could correspond to field configurations without Dirac
string singularities.
7. Dimensional reduction
The dimensional reduction of a 2-form gauge field BMN from 6 to 5 dimensions
gives a vector field A = B5 and a 2-form b (where , = 0, 1, . . . , 4 and
251
e with d A
e=
M, N = 0, 1, . . . , 5). In 5 dimensions, a vector field is dual to a 2-form field A,
e
dA. If B is self-dual in D = 6, dB = dB, then in 5 dimensions db = dA, so that b = A
is the 2-form gauge field dual to the photon. Then A and b are not independent and the
theory can be written in terms of A alone. Thus a self-dual BMN reduces to a vector field
A in D = 5. Moreover, the self-duality condition implies that A satisfies the source-free
Maxwell equations.
A 4th rank gauge field CMN P Q with the algebraic properties (5.9) gives on reduction
the fields
h = C5 5 ,
d = C 5 ,
c = C .
(7.1)
However, it was seen in the last section that in the linearised D = 5 theory a linearised
graviton h has dual representations in terms of fields D or C . The 4th gauge
field CMN P Q occurring in the D = 6 (4, 0) multiplet satisfies the self-duality condition
(5.13) which implies that the fields d , c from dimensional reduction should be
identified with the duals D , C of the D = 5 linearised graviton h , so that the
reduced theory can be written in terms of h alone, and the reduction of the exotic D = 6
tensor gauge field CMN P Q gives a conventional D = 5 graviton field h . The relationship
R = 16 S
(7.2)
between the linearised Riemann curvature R and the field strength S for D and the identity
S[ ] = 0
(7.3)
together imply that h satisfies the linearised Einstein equation, R = 0.

The reduction of a D = 6 field DMN P satisfying (5.14) gives
h = D5() ,
d = D ,
A = D55 ,
b = D5 ,
(7.4)
and if it satisfies the self-duality constraint (5.17), then the 2-form gauge field b is dual to
the vector field A and the field d is dual to h , so that the reduction of DMN P gives
a graviton plus a vector field in D = 5. Thus whereas the reduction of a D = 6 graviton
hMN gives a graviton, a vector and a scalar, the reduction of DMN P gives a graviton and
a vector but no scalar, and the reduction of CMN P Q gives a graviton only.
a satisfying the self-duality
Similarly, the reduction of the D = 6 fermionic field MN
constraint (5.20) gives the D = 5 gravitini a of D = 5, N = 8 supergravity, satisfying
the linearised D = 5 RaritaSchwinger equation.
The reduction of the other fields in the (2, 2), (3, 1) and (4, 0) D = 6 multiplets is
straightforward, and the dimensional reduction of all three D = 6 multiplets give the same
D = 5 multiplet, the N = 8 supergravity multiplet. Thus the dimensional reduction of the
free D = 6 field theories of the (2, 2), (3, 1) and (4, 0) multiplets all give the same theory,
the linearised D = 5, N = 8 supergravity. This was of course to be expected; although
there are three multiplets with 32 supersymmetries in D = 6, there is only one in D = 5
(with spins not greater than two).
If the interacting gauge or gravity theories can be expressed as interacting theories of
e C or D introduced in the last section, then the interactions could not
the dual variables A,
252
be of a standard type, but they could be M-interactions. Indeed, the reduction of the (2, 0)
theory to 5-dimensions is naturally expressed in terms of a 2-form potential B (together
with its dual vector potential) and the D = 6 M-interactions should give an M-interacting
theory of a 2-form gauge field in D = 5. However, this is dual to a theory of A alone with
the standard YangMills interactions. In the same way, reducing the M-interacting (4, 0)
theory (if such a theory exists) would give an M-interacting theory in D = 5 in terms of
gauge fields C, D and h, which can be dualised to a theory of h alone with the standard
local gravitational interactions. Thus M-interactions should play a central role in the dual
formulations of gravity or gauge theory.
8. BPS states in five dimensions

The full N = 2n superalgebra in 4 + 1 dimensions is given by (2.6) with a, b = 1, . . . , N .
For N = 8, the charges and the BPS states carrying them are as follows [17]. The 27
central charges Z ab are the electric charges for the 27 abelian vector fields and are carried
by electrically charged 0-branes, the 27 Ziab with spatial indices i, j = 1, . . . , 4 are the
corresponding magnetic charges carried by magnetically charged strings, and the charge
K is the gravitational charge of [17], carried by BPS instantonic 0-branes. There are also
2-branes coupling to axionic scalars with charge Zijab , domain wall 3-branes with charge
bab = ij kl Z ab .
bab = ij kl Z 0l ab and space-filling 4-branes with charge Z
Z
ij k
ij kl
0
The massive particle representations with central charges Z ab , K were classified in [18].
The supercharges transform as a (2, 1; N) + (1, 2; N) under the rest-frame little group
SU(2) SU(2) Sp(n), giving the chiral supercharges Qa with Qa = 0 Qa . For
BPS representations with mass saturating a bound in terms of the central charges Z ab , K
which are invariant under the action of 2r+ of the 2-component supercharges Q+ and 2r
of the 2-component supercharges Q , so that the fraction of supersymmetry preserved
is = (r+ + r )/2n, the states will fit into a representation of the supersymmetry algebra
generated by the remaining 4p positive chirality supersymmetries and 4q negative chirality
ones, with p = n r+ , q = n r . This is the (p, q) massive D = 5 supermultiplet
of [18], and the dimensional reduction on a circle of a massless representation of (p, q)
supersymmetry in 5 + 1 dimensions gives a KaluzaKlein tower of massive multiplets in
4 + 1 dimensions, each of which is a (p, q) massive D = 5 supermultiplet with the same
p and q. The central charge Z breaks the Sp(n) to a subgroup and the representation can
be decomposed into representations of SU(2) SU(2) Sp(p) Sp(q), which is also the
little group for massless representations of the (p, q) superalgebra in 5 + 1 dimensions.
Of particular interest are those BPS representations preserving 1/2 the supersymmetry.
For N = 4, there are two distinct such massive representations, the (2, 0) (or (0, 2)) multiplet and the (1, 1) multiplet. The massive vector multiplet is a (1, 1) multiplet while the
massive tensor multiplet is a (2, 0) multiplet. For N = 8, there are three massive representations preserving 1/2 the supersymmetry, the (2, 2), (3, 1) and (4, 0) representations, with
little group decompositions given by (4.1), (4.2) and (4.4), respectively [18].
253
Free field theory representations can be found from the massive KaluzaKlein modes
of the circle compactifications of the massless (p, q) representations in 6 dimensions
discussed in previous sections. For example, the massive (3, 1) multiplet in 5 dimensions
has a (4, 2; 1, 1) of SU(2) SU(2) Sp(3) Sp(1), which is carried by a field D =
D[] with D[ ] = 0 satisfying the massive D = 5 self-duality constraint
[ D] = m D
(8.1)
with mass m.
BPS 0-brane states preserving 1/2 the supersymmetry in (2, 2), (3, 1) and (4, 0)
multiplets all occur in the D = 5 supergravity theory. Those in (2, 2) multiplets all have
K = 0 and Z 6= 0 and the supergravity solutions are electrically charged black holes, while
those in (3, 1) multiplets have both K and Z nonzero. Those in (4, 0) multiplets have
Z = 0 and K 6= 0 and the supergravity solutions are gravitational instantons lifted to 4 + 1
dimensions (self-dual instantons are in (4, 0) multiplets while anti-self-dual ones are in
(0, 4) multiplets).
Compactification of a (p, q) supersymmetric theory in 5 + 1 dimensions on a circle of
radius R gives a KaluzaKlein tower of massive (p, q) D = 5 supermultiplets of masses
n/R for all positive integers n. Conversely, a limit of a five-dimensional N = 8 theory
in which a mass M 0 and a tower of massive (p, q) multiplets (with p + q = 4) of
masses nM, n = 1, 2, 3, . . . , becomes light could be a decompactification limit in which
an extra dimension opens up to give a 6-dimensional theory with (p, q) supersymmetry. In
particular, a 0-brane with mass M = |K| n/ l will fit into a D = 5 (4, 0) multiplet. The
(4, 0) multiplet has a free-field representation with 42 massive scalars, 27 massive self-dual
2-form fields B satisfying
dB = m B,
(8.2)
and a massive field C satisfying

[ C] = m C
(8.3)
If these 0-branes are to be interpreted as KaluzaKlein modes for a massless 6-dimensional

theory compactified on a circle of radius R = l, then that 6-dimensional theory must be
(4, 0) supersymmetric and have 42 scalars, 27 self-dual 2-forms and a self-dual CMN P Q
gauge field, as before.
9. BPS branes in six dimensions

The (4, 0) theory has 27 self-dual 2-forms and these couple to 27 self-dual BPS strings,
or more precisely, to strings whose charges take values in a 27-dimensional lattice. In
addition, there are 42 scalars and these can couple to BPS 3-branes in 6 dimensions.
The (4, 0) superalgebra in six dimensions with central charges is

a b
ab
Q , Q = ab + M C PM + + M C ZM

ab
+ 16 + MNP C ZMNP
,
(9.1)
254
where + is the chiral projector:

= 12 1 7 .
(9.2)
ab
ba
ab
ab
ba
= ZM
, ZM
ab = 0, while ZMNP
= ZMNP
and is
The 27 1-form charges satisfy ZM
a self-dual 3-form:
ab
= 16 MNP QRS Z abQRS .
ZMNP
(9.3)
The 27 charges Ziab with spatial indices i, j = 1, . . . , 5 are the string charges, and Zijabk are
the 36 3-brane charges, while Z0ab are the charges for space-filling 5-branes. (Note that the
self-duality condition (9.3) implies that Z0ij are not independent charges.)
On dimensional reduction to 5 dimensions, the charges decompose as

ab
Zab , Z5ab = Z ab ,
P M P , P 5 = K ,
ZM
ab
ab
ab
Z5
= Z
.
ZMNP
(9.4)
The extra component of momentum becomes the K-charge, so that the 5-dimensional
instantonic 0-brane is, from the D = 6 viewpoint, a wave carrying momentum P 5 in the
extra circular dimension. The electric 0-branes in D = 5 arise from strings winding around
the 6th dimension while the magnetic strings arise from unwrapped D = 6 strings. The
2-branes and 3-branes in D = 5 arise from D = 6 3-branes, while the space-filling D = 5
4-branes come from wrapped D = 6 5-branes. As in the (2, 0) case reviewed in Section 2,
the strong coupling limit l should be one in which the tensions of the magnetic
strings are kept finite if the BPS spectrums are to match, as will be discussed in the next
section.
10. The strong coupling limit of D = 5, N = 8 supergravity
In the previous sections, evidence has been accumulated to support the conjecture that
D = 5, N = 8 supergravity (embedded in some consistent theory) with 5-dimensional
Planck length l is a D = 6 (4, 0) superconformal field theory compactified on a circle of
radius R = l, so that the strong coupling limit l is one in which an extra dimension
decompactifies. From the five-dimensional perspective, an infinite tower of BPS 0-branes
in massive (4, 0) multiplets carrying the central charge K become massless in this limit,
signalling the opening up of an extra dimension. For M-theory compactified on T 6 , this
gives a six-dimensional strong-coupling phase which is (4, 0) supersymmetric and which
contains the (4, 0) superconformal field theory.
For the free theories, it has been shown that the linearised D = 5 supergravity with
Planck length l is given by the dimensional reduction of the (4, 0) tensor theory, formulated
as a theory of free fields propagating in a fixed background spacetime which is a product
of 5-dimensional Minkowski space and a circle of radius R = l. For the interacting D = 5
supergravity, the conjecture requires that there be an interacting form of the (4, 0) tensor
theory which is also superconformal and whose dimensional reduction gives the full
supergravity, although there are no local covariant interactions for the (4, 0) multiplet
255
with this property. This requires that the 6-dimensional theory have some magical form
of M-interactions which give the nonpolynomial supergravity interactions on reduction.
It could be that these are some nonlocal or noncovariant self-interactions of the (4, 0)
multiplet, or it could be that other degrees of freedom might be needed; one candidate
might be some form of string field theory. However, if a strong coupling limit of the theory
does exist that meets the requirements assumed here, then the limit must be a (4, 0) theory
in six dimensions, and this would predict the existence of M-interactions arising from the
strong coupling limit of the supergravity interactions. Although it has not been possible to
prove the existence of such a limit, it is remarkable that there is such a simple candidate
theory for the limit with so many properties in common with the (2, 0) limit of the D = 5
gauge theory. Conversely, if there is a 6-dimensional phase of M-theory which has (4, 0)
supersymmetry, then its circle reduction to D = 5 must give an N = 8 supersymmetric
theory and the scenario described here should apply.
Linearised gravity has dual descriptions in terms of exotic tensor fields propagating
in a fixed background, and these play an important role in the construction used here.
The triality between the formulations of gravity in terms of h, C or D, together with the
duality of abelian gauge theories, could have an extension to the interacting nonlinear
theories in which M-interactions play an important part. Such matters would need to
be understood before questions of back-reaction and background independence can be
addressed. However, it is intriguing that M-theory seems to be pointing to a more
general framework than Riemannian geometry, and one which would be better suited to
accommodate magnetic masses in gravity.
Both the (2, 0) and (4, 0) theories are chiral and so are liable to anomalies. In both
cases there are potential gravitational and supersymmetry anomalies, so that the coupling
to gravity or supergravity is problematic. For the M5-brane world-volume theory, the
coupling to gravity is such that these anomalies are cancelled against bulk effects through
anomaly inflow [2729], while for the IIB theory compactified on K3, the anomalies from
the (2, 0) tensor multiplets are cancelled by those from the (2, 0) supergravity multiplet.
There is presumably some non-abelian M-symmetry in the M5-brane world-volume
theory that reduces to the non-abelian gauge symmetry in five dimensions and in
six dimensions reduces to the 2-form gauge invariance B = d in the abelian case.
A conventional gauge symmetry in a chiral six-dimensional theory would typically be
anomalous, but in the case of the M-symmetry of a theory with M-interactions, the
supposed consistency of the theory suggests that such problems are absent. For the (4, 0)
theory, the inability to couple to gravity is not a problem: it is already a theory of gravity of
sorts (at least it dimensionally reduces to supergravity) and so there is no need to couple to
another gravity; if one did, there would be two gravitons on reduction to five dimensions.
Indeed, there is no (4, 0) supergravity to attempt to couple to. There are presumably
M-symmetries in the (4, 0) theory giving rise to the diffeomorphism symmetry and local
supersymmetry of the D = 5 supergravity and which reduces to the symmetries (5.10)
and (5.18) in the free (4, 0) theory. It seems plausible that, as for the (2, 0) theory, the Msymmetry could be free from anomalies, and the consistency of the limit considered here
would seem to require this.
256
The electrically charged 0-branes of N = 4 gauge theory and N = 8 supergravity in

D = 5 are lifted to wrapped strings of the (2, 0) and (4, 0) theories, respectively. In both
cases there are points in the D = 5 moduli spaces or their boundaries at which 0-branes
become massless and as these points are approached, the tensions of certain D = 6 strings
tend to zero. The (2, 0) and (4, 0) superconformal theories are very similar and both might
be regarded as theories of tensionless strings, as in [7]. However, it is possible that both are
in fact quantum field theories, with the superconformal invariance playing a crucial role in
their consistency. That this should be the case for the (2, 0) theory has been suggested in,
e.g. [16]. The possibility that the (4, 0) theory could exist as an interacting superconformal
field theory is remarkable, as this would be a quantum theory whose compactification
would give N = 8 supergravity in 4 or 5 dimensions, together with massive fields.
In six-dimensional conformal field theory, the natural dimension for scalars is 2, and so
it is useful to define the 42 dimension-2 scalars
= / l 2
(10.1)
in terms of the 42 dimensionless scalars of the D = 5 supergravity theory. Writing the

6-dimensional theory without coupling constants requires using instead of , and, just
as in the (2, 0) theory, the appropriate limit in the linearised supergravity theory is l
while is kept fixed. Thus the limit involves scaling all the scalars in the same way,
and can be viewed as a conformal transformation on the D = 5 moduli space, taking one
towards the boundary of moduli space in every possible direction simultaneously. Such
limits were not included in the analysis of [13,14].
The dimension of makes it difficult to write down covariant local interactions in
six dimensions without any explicit dimensionful parameters but which reduce to the
interactions in 5-dimensions that are non-polynomial in . This has particular relevance
for the tensions of branes in the theory. Choosing any particular scalar (which could be
a string theory dilaton, or could be given by the volume of the M-theory 6-torus in Planck
units e = L/ l11 , for example), then the tension of a given p-brane in the 5-dimensional
theory will depend on the asymptotic value 0 of in general and is typically given by a
formula of the form (in Einstein frame and suppressing a constant of proportionality and
dependance on the other scalars)
Tp = ep 0
1
l p+1
(10.2)
for some p , which will depend on the choice of and on the choice of brane. For example,
the part of the D = 5 supergravity action involving and a particular vector field A will
be of the form

Z
1
1
1
(10.3)
S = d5 x g 3 R e2a F 2 3 ()2 +
4l
l
2l
for some a, and A will couple to an electric 0-brane solution with 0 = a and a magnetic
e dual to A has field strength H = e2a F ,
string with 1 = a. The 2-form gauge field A
e must satisfy a nonlinear
so that the D = 6 2-form gauge field B which reduces to A, A
self-duality constraint whose dependence is given by
H = e2a H.
257
(10.4)
To lowest order in , (10.2) can be written in terms of the asymptotic value 0 of as

1
1
(10.5)
Tp = 2 + p 0 p1
l
l
dropping nonlinear terms. For a magnetically charged D = 5 string with p = 1, this has
a well-defined limit 1 0 as l , provided is kept fixed (i.e. , l keeping / l 2
fixed). The D = 5 string then lifts to a BPS string in D = 6 dimensions with tension 1 0 .
Compactifying on a circle of radius R, a wrapped string of tension 1 + 1/R 2 gives
a 0-brane of mass proportional to R1 + 1/R, which agrees with (10.5) for p = 0
(taking into account the dependence of (10.4) with a = 1 , so that 0 = 1 ). Thus to
lowest order in p the 6-dimensional strings give strings and 0-branes in D = 5 with the
appropriate tensions.
It is interesting to ask how this could generalise to a nonlinear theory with M-interactions. The tension of the BPS strings in six dimensions should depend on but not on
any dimensionful parameters. One possibility is that the tension could be of the form

(10.6)
T1 = 0 + 12 2 ( )0 + O 3 ,
where is some M-product, which must be of dimension 2, even though the simple
product 2 has naive dimension 4, and which tends asymptotically to ( )0 . On
dimensional reduction on a circle of radius R, this must give the 5-dimensional string
tension, which requires that ( )0 reduces to 02 R 2 = 02 R 2 l 4 in five dimensions
(perhaps with small corrections). This would then give the correct result if R = l, and
furthermore wrapped strings then give the correct 0-brane mass. This suggests that the
D = 5 tension formula (10.2) should be modified to an expression involving a product,
with the property that the standard supergravity results are recovered at low energies but
the limit l can be taken in such a way that both and the string tension remain
finite.
The strings in 6-dimensions then have tensions which depend on the scalars in some
mysterious way, but on compactifying on a circle of finite radius R = l, they must give
strings and 0-branes whose tensions depend on in the standard way (10.2) at low energies.
In particular, at points in the moduli space of the 5-dimensional theory in which the masses
of some 0-branes or tensions of some strings become zero, the tensions of corresponding
strings in 6-dimensions should approach zero or infinity. The various strong-coupling limits
of the 5-dimensional string theory (obtained by compactifying M-theory on T 6 ) that were
analysed in [13,14] in which 0-branes become massless all correspond to limits of the (4, 0)
theory in which some string tensions become infinite and others approach zero.
For example, consider the limit in which the volume in Planck units of the T 6 becomes
large, L/ l11 . In that limit the KaluzaKlein modes, viewed as 0-branes with
charges in a 6-dimensional lattice, all become massless, signalling a decompactification
to 11 dimensions. In addition, there are six 0-branes whose masses become infinite
(from wrapped M5-branes), six strings that become tensionless (partially wrapped
M2-branes) and six strings whose tensions become infinite (from wrapped KaluzaKlein
258
monopoles). Despite the presence of strings whose tensions approach zero, this limit of
the 5-dimensional theory is a decompactification limit to M-theory in 11-dimensional
Minkowski space [14]. However, if one first takes the strong gravitational coupling limit
to obtain the 6-dimensional (4, 0) theory, then instead of 6 types of 0-branes becoming
massless, there are 6 types of strings (with charges in a 6-lattice) which are becoming
tensionless, and a further 6 types of string whose tensions are becoming infinite. The
interpretation is not clear here, and in particular whether it makes sense to think of this
in terms of a further 6 dimensions. This limit is closely related to certain limits discussed
in [14], where a 12-dimensional interpretation was considered.
The maximal supergravity theory in D 6 11 dimensions together with its BPS states
and their excitations contain much of nonperturbative string theory. For example, one of
the BPS strings in the D = 5, N = 8 theory is the fundamental string for type II string
theory compactified on T 5 , and the perturbative string states arise from fluctuations of this
string. In [12], the possibility that the full nonperturbative string theory might in some
way be equivalent to an effective supergravity theory plus its solitons was suggested. An
obvious drawback with this is that the supergravity theory does not appear to give a welldefined quantum theory. However, if the (4, 0) theory is a consistent superconformal field
theory, then this (with its BPS states) could give a formulation of M-theory, with the string
modes and KaluzaKlein modes all carried by strings. One of the BPS strings will give the
perturbative type II string on compactification to D = 5, while others will give Kaluza
Klein modes. Compactifying to D = 5 and then going to various boundaries of the moduli
space will give decompactifications to other vacua of M-theory, including 11-dimensional
Minkowski space.
Alternatively, it could be that the (4, 0) superconformal field theory is a sector of a larger
D = 6 (4, 0) supersymmetric theory arising from the strong coupling limit of M-theory
compactified on T 6 . If the (4, 0) superconformal field theory is consistent in its own right,
it might then arise as a decoupling limit of the full theory. However, many of the known
degrees of freedom of M-theory are already contained in the (4, 0) superconformal field
theory as BPS branes and their excitations. The issue is perhaps whether the BPS branes
(and their excitations) can be treated properly, possibly as solitons of some sort, in the (4, 0)
superconformal field theory, or whether they should be regarded instead as independent
degrees of freedom (e.g. some sort of string fields) that need to be introduced in addition.
Similar issue arise in the (2, 0) tensor theory and even in N = 4 super-YangMills theory
in four dimensions, where an infinite number of BPS states that are usually treated as
solitons the magnetic monopoles and dyons become massless at particular points in
the moduli space [24].
Limits similar to the one considered here in other dimensions and other theories will be
considered in a separate publication. It seems that there could be many phases of M-theory
which do not have a conventional field theory description but which involve M-interactions,
and understanding these should be of considerable importance in unravelling the secrets of
M-theory.
259
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]
M. Rozali, Phys. Lett. B 400 (1997) 260, hep-th/9702136.

M. Berkooz, M. Rozali, N. Seiberg, Phys. Lett. B 408 (1997) 105, hep-th/9704089.
N. Seiberg, hep-th/9705117.
X. Bekaert, M. Henneaux, A. Sevrin, Phys. Lett. B 468 (1999) 228, hep-th/9909094.
A. Strominger, Phys. Lett. B 383 (1996) 44, hep-th/9512059.
P.K. Townsend, Phys. Lett. B 373 (1996) 68, hep-th/9512062.
E. Witten, in: Proc. Strings 95, 1995, hep-th/9507121.
N. Seiberg, Phys. Lett. B 408 (1997) 98, hep-th/9705221.
R. Dijkgraaf, E. Verlinde, H. Verlinde, hep-th/9603126, hep-th/9604055.
O. Aharony, hep-th/9911147.
E. Cremmer, in: S.W. Hawking, M. Rocek (Eds.), Supergravity and Superspace, Cambridge
University Press, Cambridge, 1981.
C.M. Hull, P.K. Townsend, Nucl. Phys. B 438 (1995) 109, hep-th/9410167.
E. Witten, Nucl. Phys. B 443 (1995) 85, hep-th/9503124.
C.M. Hull, Nucl. Phys. B 468 (1996) 113, hep-th/9512181.
O. Aharony, M. Berkooz, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 119, hep-th/97012117.
N. Seiberg, Phys. Lett. B 390 (1997) 169, hep-th/9609161.
C.M. Hull, Nucl. Phys. B 509 (1997) 252, hep-th/9705162.
C.M. Hull, hep-th/0004086.
B. Julia, in: S.W. Hawking, M. Rocek (Eds.), Supergravity and Superspace, Cambridge
University Press, Cambridge, 1981.
G.W. Gibbons, S.W. Hawking, Phys. Lett. B 78 (1978) 430.
G.W. Gibbons, M.J. Perry, Nucl. Phys. B 248 (1984) 629.
R. Sorkin, Phys. Rev. Lett. 51 (1983) 87.
D. Gross, M. Perry, Nucl. Phys. B 226 (1983) 29.
J. Strathdee, Int. Jour. Mod. Phys. A 2 (1987) 273.
P.K. Townsend, Phys. Lett. B 139 (1984) 283.
E. Witten, J. Geom. Phys. 22 (1997) 103, hep-th/9610234.
L. Bonora, C.S. Chu, M. Rinaldi, JHEP 9712 (1997) 007, hep-th/9710063.
D. Freed, J.A. Harvey, R. Minasian, G. Moore, Adv. Theor. Math. Phys. 2 (1998) 601, hepth/9803205.

The minimum width condition for neutrino

conversion in matter
C. Lunardini a,b , A.Yu. Smirnov c,d,
a SISSA-ISAS, via Beirut 2-4, 34100 Trieste, Italy
b INFN, sezione di Trieste, via Valerio 2, 34127 Trieste, Italy
c The Abdus Salam ICTP, Strada Costiera 11, 34100 Trieste, Italy
d Institute for Nuclear Research, RAS, Moscow, Russia
Received 22 February 2000; revised 16 May 2000; accepted 25 May 2000
Abstract
2 ) the width of
We find that for small vacuum mixing angle and low energies (s MZ
matter, d1/2 , needed to have conversion probability P > 1/2 should be larger than dmin =
/(2 2 GF tan 2): d1/2 > dmin . Here GF is the Fermi constant, s is the total energy squared in the
center of mass and MZ is the mass of the Z boson. The absolute minimum d1/2 = dmin is realized
for oscillations in a uniform medium with resonance density. For realistic density distributions
(monotonically varying density, castle wall profile, etc.) the required width d1/2 is larger than dmin .
2 we get that d
The width dmin depends on s, and for Z-resonance channels at s MZ
min (s) is 20
times smaller than the low energy value. We apply the minimum width condition, d > dmin , to high
energy neutrinos in matter as well as in neutrino background. Using this condition, we conclude that
the matter effect is negligible for neutrinos propagating in AGN and GRBs environments. Significant
conversion can be expected for neutrinos crossing dark matter halos of clusters of galaxies and
for neutrinos produced by cosmologically distant sources and propagating in the universe. 2000
PACS: 14.60.Pq; 14.60.St

Keywords: Neutrinos; Matter effects; Universe
1. Introduction
Since the paper by Wolfenstein [1], the neutrino transformations in matter became one
of the most important phenomena in neutrino physics. Neutrinos propagating in matter
undergo coherent forward scattering (refraction) described at low energies by the potential
(1)
V = 2 GF n,
smirnov@ictp.trieste.it
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 4 1 - 2
C. Lunardini, A.Yu. Smirnov / Nuclear Physics B 583 (2000) 260290
261
where GF is the Fermi constant, and n is a function of the density and chemical
composition of the medium. For the case of e and e conversion in matter n
coincides with the electron number density, ne .
Refraction can lead to an enhancement of oscillations in media with constant density,
and to resonant conversion in the varying density case (MSW effect) (for a review, see
Ref. [3]). For periodic, or quasi-periodic density profiles, various parametric effects can
occur [46].
The MSW effect has been applied to solar neutrinos [2], and to neutrinos from
supernovae (for a review, see Ref. [7]). Oscillations of neutrinos of various origins (solar,
atmospheric, supernovae neutrinos, etc.) in the matter of the Earth have been extensively
studied. Apart from resonance enhancement of oscillations, parametric effects are expected
for neutrinos crossing both the mantle and the core of the Earth [811]. The oscillations and
conversion of active neutrinos into a sterile species can be important in the Early Universe
[12]. Recently, matter effects on high energy neutrino fluxes from Active Galactic Nuclei
(AGN) and Gamma Ray Bursters (GRBs) have been estimated [13]. Propagation of ultrahigh energy neutrinos in halos of galaxies has been considered [14].
It is intuitively clear that to have a significant matter effect a sufficiently large amount
of matter is needed. Let us define the width of the medium as the integrated density along
the path travelled by the neutrino in the matter:
Z
(2)
d = ne (L) dL.
This quantity is frequently named column density in astrophysical context. We will show
that there exists a minimum value dmin for the width below which it is not possible to have
significant neutrino conversion. This lower bound is independent of the density profile and
of the neutrino energy and mass. That allows us to make conclusions on the relevance of
matter effect in various situations without knowledge of the density distribution.
The paper is organized as follows: in Section 2 we derive the minimum width condition
for the conversion in matter between two active neutrino flavours, and check it for different
density profiles. In Section 3 we discuss the generalizations of the condition to the activesterile case and to conversion induced by flavour changing neutrinomatter interactions.
We also study the matter effect in the small width limits. Section 4 presents a study of
the minimum width condition for high energy neutrinos both in matter and in neutrino
background. Section 5 is devoted to applications of our results to neutrino propagation in
AGN and GRBs environments, in dark matter halos and in the Early Universe. Conclusions
and discussion follow in Section 6.
2. The minimum width condition

In this section we consider various mechanisms of matter enhancement of neutrino
flavour conversion. For each of them we work out the minimum width of the medium
needed to have significant conversion probability, showing that a lower bound for the width
exists and is realized in the case of uniform medium with resonance density.
262
2.1. The absolute minimum width

Let us consider a system of two mixed flavour states 1 , e and ( ), characterized by
vacuum mixing angle and mass squared difference 1m2 . In a uniform medium the states
oscillate and the transition probability, as a function of the distance L, is given by:

L
,
(3)
Pe (L) = sin2 2m sin2
lm
where m and lm are the mixing angle and the oscillation length in the medium:
sin 2m =
lm =
sin 2
[(2EV /1m2 cos 2 )2
+ sin2 2 ]1/2
l
[(2EV /1m2 cos 2 )2
+ sin2 2 ]1/2
,
.
(4)
Here l = 4E/1m2 is the vacuum oscillation length and E is the neutrino energy.
We assume that the vacuum mixing is small, so that vacuum oscillations effects are
1) and a strong transition in medium, i.e., Pe = O(1), is
negligible (Pvac
e
essentially due to matter effect. For definitness, we choose the condition of significant
conversion to be
1
(5)
Pe > .
2
Let us consider a uniform medium with resonance density [2]:
1m2
=
cos 2.
nres
e
2 2 EGF
In this case the oscillation amplitude is sin 2m = 1, and the oscillation length equals
(6)
4E
l
=
.
(7)
sin 2
1m2 sin 2
According to Eq. (3) the condition (5) starts to be satisfied for L = lm /4, and the
corresponding width is:
l res =
1
l res .
dmin = nres
4 e
res given in (6) and (7), we get:
Inserting the expressions of nres
e and l
d0
=
,
dmin =
tan
2
2 2 GF tan 2
where
1.11
'
.
d0 =
GF
2 2 GF
(8)
(9)
(10)
We will call d0 the refraction width. Numerically,

d0 = 2.45 1032 cm2 = 4.08 108 A cm2 ,
1 The arguments remain the same for three neutrinos.
(11)
263
where A = 6 1023 is the Avogadro number. 2

The widths dmin and d0 have a simple physical interpretation. The refraction width d0
is a universal quantity: it is determined only by the Fermi coupling constant, and does not
depend on the neutrino parameters at all. Using the definition of refraction length
2
2
(12)
=
V
2 ne GF
we can write:
d0 l0
(13)
= .
ne
4
It appears that d0 corresponds to the distance at which the matter-induced phase difference
between the flavour states equals /2. This can be considered as the definition of refraction
width, which by Eq. (12) can be written in the general form:
nm
,
(14)
d0
2 Vm
where Vm is the neutrino-medium potential and nm is the concentration of the relevant
scatterers in the medium.
The minimum width, dmin , is inversely proportional to tan 2 , which represents
properties (the mixing) of the neutrino system itself. The smaller the mixing , the larger
is the width dmin needed for strong transition.
The condition (5) can be generalized. It corresponds to the case of initial state coinciding
with a pure flavour state. In general one can require that the change of the probability to
detect a given flavour is larger than 1/2:
l0
1
(15)
1P Pf ( ) Pi ( ) > ,
2
where Pi and Pf are the initial and final probabilities. The condition (5) corresponds to
Pi ( ) = 0, so that Pf ( ) = Pe . Taking Pi = 1/4 and Pf = 3/4, we get in a similar
way:
2
.
(16)
d1/2 = dmin =
3
3 2 GF tan 2
This d1/2 is the extreme value, however for most practical situations the condition (5) is
more relevant, and from here on we will use the the width dmin determined in (9).
In what follows we will show that for all the other density profiles the width d1/2 required
by the condition (5) is larger than dmin .
2 It can be checked that different choices of the condition (5) lead to analogous results. For instance, taking
Pe > 34 we find
3/4
d0
4
2
= 5.41 108 A cm2 ,
d =
3 0 3 2 GF
and, for Pe > 14 :

2
1/4
d0 = d0 =
= 2.7 108 A cm2 .
3
3 2 GF
264
2.2. Uniform medium with density out of resonance

1
2
For ne 6= nres
e the inequality (5) can be satisfied only if sin 2m > 2 , which means that
res
the density is required to be in the resonance interval: nres
e (1 tan 2 ) 6 ne 6 ne (1 +
tan 2 ). At the edges of the interval we get the width

1
1 ' 2 dmin ,
(17)
d1/2 =
2GF tan 2
which is largerthan dmin . For other values of the density in the resonance interval we have
dmin < d1/2 < 2 dmin . 3
2.3. Medium with varying density
In general the neutrino propagation has a character of interplay of resonance conversion
and oscillations. Two conditions are needed for strong transition:
(1) Resonance condition: the neutrinos should cross the layer with resonance density.
(2) Adiabaticity condition: the density should vary slowly enough. This condition can
be written in terms of the adiabaticity parameter at resonance:
1,

2E cos 2 1 dne
.
1m2 sin2 2 ne dL
(18)
(19)
Notice that both the conditions (1) and (2) are local, and can be fulfilled for arbitrarily small
widths of the medium. Clearly, they are not sufficient to assure a significant conversion, and
a third condition of large enough matter width is needed.
Let us consider a linear density profile with length 2L and average density equal to the
res
resonance one, so that nmax = nres
e + 1n and nmin = ne 1n. Denoting 1m and 2m the
mixings in the initial and final points, we find that in the first order of adiabatic perturbation
theory the conversion probability is given by:
Pe (L) =
1 1
cos 21m cos 22m
2 2

1
1
sin 21m sin 22m cos f (x)
2

1
f (x) ,
2 sin(21m 22m )(x) cos
2
(20)
where
x = 2
L
,
l resp
f (x) = ln(x +
p
1 + x 2) + x 1 + x 2,
3 It can be checked that the width d

1/2 in Eq. (17) is larger than dmin for small mixing: sin 2 . 0.3. We will
use this condition as criterion of smallness of the mixing.
Zx
(x) =
0

1
dy
f (y) .
cos
1 + y2
2
For 1 we get from Eqs. (20) and (21):

2 .
d1/2 = dmin 1 + 1
8
265
(21)
(22)
This expression shows that for the adiabatic case d1/2 ' dmin and for weak violation of
adiabaticity the minimum width increases quadratically with . We remark that in this case
the effect is dominated by oscillations with large (close to maximal) depth. The change of
density gives only small corrections.
Let us consider now a situation in which the resonance adiabatic conversion is the main
mechanism of flavour transition. A pure conversion effect is realized if the initial neutrino
state that enters the medium coincides with one of the eigenstates of the Hamiltonian in
matter, and the propagation in matter is adiabatic. In this case no phase effect, and therefore
no oscillations occur. Let us denote ni and nf the initial and final densities of the medium,
and suppose the initial state is i = 2m = sin m e + cos m . The probability to find a
in this state is Pi ( ) = cos2 m (ni ). The state evolves following the change of density,
so that it remains an eigenstate of the Hamiltonian, and the probability to find in the
final state is Pf ( ) = cos2 m (nf ). Since the initial state i does not coincide with a pure
flavour state we will use the condition (15) as criterion of strong matter effect. Inserting Pi
and Pf in (15), we get the condition for d1/2 :
cos 2m (nf ) cos 2m (ni ) = 1.
(23)
res
Taking the initial and final values of the density as ni = nres
e + 1n and nf = ne 1n
(1n > 0), and using the definition (4) we find that the equality (23) leads to
1
1n = nres
e tan 2.
3
(24)
Clearly, for a given 1n the size of the layer, and therefore its width, depend on the gradient
of the density which can be expressed in terms of the adiabaticity parameter , Eq. (19).
We get:
ne (L) dL =
2E cos 2 1
dne ,
1m2 sin2 2
(25)
and then integrating this equation we obtain:

d=
4E cos 2 1
1n.
1m2 sin2 2
(26)
Finally, inserting the expressions (24) and (6) in Eq. (26) we find:
d1/2 =
4 1
dmin.
(27)
Let us comment on this result. As far as the adiabaticity condition is satisfied, the change
of probability does not depend on the density distribution; it is a function of the initial and
266
final densities only. If 1n is fixed, the decrease of the width means the decrease of the
length L of the layer, and therefore increase of the gradient of the density. This will lead
eventually to violation of the adiabaticity condition. Thus, the minimal width corresponds
to the maximal for which the adiabaticity is not broken substantially.
For strong adiabaticity violation an increase of d1/2 is expected, due to the increase
of the minimum 1n required by the condition (15), and therefore of the corresponding
length. This can be seen if we consider the previous argument taking into account the effect
of the adiabaticity breaking from the beginning. Using the LandauZener level crossing
probability PLZ = exp(/2 ), which describes the transition between two eigenstates,
we get, instead of (23):

(28)
(1 2PLZ ) cos 2m (nf ) cos 2m (ni ) = 1,
where we have averaged out the interference terms. Then instead of Eq. (24) we get
1n = nres
e q
1
2
16PLZ
tan 2.
(29)
16PLZ + 3
Finally, the condition for d1/2 can be written as:

q
d1/2 =
4
2
16PLZ
16PLZ + 3
dmin .
(30)
For 0 Eq. (30) gives d1/2 , according to the fact that the density changes very
slowly and therefore the width needed to have significant conversion increases. With the
increase of the width d1/2 decreases and has a minimum at ' 0.7, for which we find
d1/2 ' 1.5 dmin . With further increase of ( & 0.7) the width d1/2 increases rapidly.
According to (30) it diverges for P 1/4, when /(4 ln 2) ' 1.13. This value
corresponds to the case in which the adiabaticity violation is so strong that even an infinite
amount of matter is not enough to satisfy the condition (15). Thus, we have found that also
in this case d1/2 > dmin .
2.4. Step-like profile
As an extreme case of strong adiabaticity violation, let us consider the profile consisting
res
of two layers of matter, having densities n1 = nres
e + 1n and n2 = ne 1n (1n > 0),
and equal lengths L1 = L2 = L. At the border between the layers the density has a jump
of size 21n. We fix L = l res /8, so that d = dmin . The result for the conversion probability
can be computed exactly:

2
step
,
(31)
+ s 2 c2 1 cos
Pe = s 2 sin2
4s
4s
where we denote the mixing parameters in the two layers as sin 22m = sin 21m s,
step
cos 22m = cos 21m c. In absence of the step (1n = 0), Pe equals 1/2,
267
recovering the case ne = nres

e = const. The probability (31) decreases monotonically as
2
4
1n increases. Expanding in = (1n/nres
e tan 2) we get:

1
step
(32)
Pe '
21
' 0.02.
2
8
2
step
step
According to (32), for d = d0 we have Pe < 1/2. This implies that, to have Pe =
1/2 one needs d1/2 > dmin .
2.5. Castle-wall profile
The profile consists of a periodical sequence of alternate layers of matter, having two
different densities n1 and n2 . We denote the corresponding mixing angles as 1m and 2m .
In this case, a strong transformation requires certain conditions on the oscillation phases
acquired by neutrinos in the layers [11]; therefore the transformation is a consequence of
the specific density profile, rather than of an enhancement of the mixing. Suppose n1
nres
e and n2 = 0, and take the width of each layer to be equal to half oscillation length, so
that the oscillation phase acquired in each layer is . It can be shown [6] that for small
this is the condition under which the conversion probability increases most rapidly with the
distance. As a function of the number N of periods (a period corresponds to two layers),
the probability is given by [5,6]:
Pe (N) = sin2 (2N1 ),
(33)
where 1 1m 2m = 1m . Using the approximation 21m 2 ' sin 21m sin 2 ,

and expanding sin 21m in n1 , we get:
2
1
' dmin .
d1/2 =
2 2 GF sin 2
(34)
Again, we find that d1/2 > dmin .

Thus, for all the known mechanisms of matter enhancement of flavour transition
(resonant oscillations, adiabatic conversion, parametric effects), we have found that the
width d1/2 is larger than dmin , which is realized for the case of uniform medium with
resonance density. In fact the constant profile with resonance density could be expected
from the beginning to represent an extreme case: this profile is singled out, since it is the
simplest distribution with the density fixed at the unique value nres
e .
It is worthwhile to introduce also the total nucleon width. Let us consider a medium
made of electrons, protons and neutrons with number densities ne , np and nn . Defining
the number of electrons per nucleon as Ye ne /(nn + np ), we can write the total nucleon
width that corresponds to d0 as:
d0N
d0
.
Ye
(35)
4 This approximation proves to be very good (relative error 6 0.5%) for 0 6 6 1, i.e., for n and n in the
1
2
resonance interval.
268
We can also introduce the total mass width d :

d mN d0N =
mN d 0
.
Ye
(36)
For electrically and isotopically neutral medium (ne = nn = np ), Eq. (35) gives:
d0N = 2d0 ,
(37)
and numerically:
d0N = 4.9 1032 cm2 ,
d = 8.16 108 g cm2 .
(38)
3. Generalizations
In this section we generalize the previous results to activesterile transition and to the
case of flavour-changing induced conversion. We also discuss the small width limits.
3.1. Activesterile conversion
In this case the scattering both on electrons and on nucleons contributes to the
conversion, and the effective potential for an electron neutrino in an electrically neutral
medium equals

nn
.
(39)
V = 2 GF ne 1
2np
Thus, the results for e s transition can be obtained from those for e by the
replacement ne ne (1 nn /2np ). For the refraction width we get immediately:

2np
.
d0 (e s ) = d0
(40)
2np nn
In particular, for an isotopically neutral medium Eq. (40) gives
d0 (e s ) = 2d0.
(41)
Notice that, for highly neutronized media (nn np ) the width d0 (e s ) gets
significantly smaller than d0 . In this case, however, the physical situation is more properly
described by the total nucleon width d0N defined in Eq. (35), since the effect is mainly due
to the scattering on neutrons. We find:

np + nn

,
d0N (e s ) = 2d0
(42)
2np nn
which gives in the limit nn :
d0N (e s ) = 2d0 ,
similarly to Eq. (41).
(43)
269
For the s case, the potential, and consequently the width, can be obtained by
replacing ne ne (nn /2np ), which gives
d0 ( s ) = d0
2np
.
nn
(44)
For isotopically neutral medium, Eq. (44) reduces to Eq. (41). For highly neutronized
media, the argument is analogous to the one for the e s case, and we get the same result
as in Eq. (43).
3.2. Oscillations induced by flavour changing (FC) neutrinomatter interactions
In this case the neutrino masses can be zero, or negligible, and the flavour transition is a
pure matter effect. The Hamiltonian of the system has the following form [1]:

0
nf
,
(45)
H = 2 GF
nf 0 nf
where nf is the effective number density of the scatterers, and and 0 are parameters of
the interaction.
As follows from Eq. (45), in a uniform medium the neutrinos oscillate with transition
probability:

4 2
L
2
sin
,
(46)
Pe (L) = 2
4 + 02
l
1
2
.
(47)
l=
2
02
G
nf
F
4 +
We assume 0 < , which is needed to have a significant oscillation amplitude. Using
Eq. (8), we get:
1
ne
FC
=
d1/2
2 2 GF 4 2 + 02 nf
d0
ne
=
.
2
02
4 + nf
(48)
FC differs from the width d

The quantity d1/2
min of Eq. (9) by a factor (ne /nf ) tan 2/
FC
2
02
4 + . This implies that d1/2 can be significantly smaller than dmin , and the oscillation
effect can be observed in media of smaller width. For a FC neutrino interaction with up (or
down) quarks and isotopically neutral medium we have ne /nf = 1/3, and therefore:
1
d0 .
(49)
6
Notice that there are two sources of decrease of the width: the factor 2 is given by the
presence of two off diagonal terms in the Hamiltonian (45) and the factor 3 is due to the
larger number of scatterers. Taking = 1, we find the value:
FC
'
d1/2
FC
' 1.36 108 g cm2 .
d1/2
(50)
270
For a density n = 4 g cm3 (Earths crust), this corresponds to the distance L = 337 km,
which is comparable to the length of the present long base-line neutrino experiments: K2K
(base-line 250 km) and ICANOE (740 km).
3.3. The small width limits
In a number of situations (see Section 5) the width of the medium is smaller, or
much smaller, than dmin . We consider, then, the matter effect on oscillations in the limit
d/dmin 1. Introducing the two variables:
L
l res /4
ne
,
nres
e
(51)
we can write the small width condition as:

d/dmin = 1.
(52)
We focus on two important realizations of this inequality:

(1) Small size of the layer and density close to resonance. As we have shown in
Section 2, a strong transition requires ne ' nres
e . In case of small width this implies
small size of the layer. Therefore we have 1 and 1. In this case d/dmin ' .
We expand the oscillation probability (3) in at the lowest (nonzero) order, and find
that the matter effect vanishes quadratically with d/dmin :
2

2
d 2
2

.
(53)
P (, ) '
4
4
dmin
(2) Small density of the medium and length close to the minimum value l res /4. Another
condition of strong conversion is to have the size of the layer of the order of the
oscillation length. According to Eq. (52), this means that the density is small. Thus,
we have 1 and 1, and therefore d/dmin ' . In order to give a phaseindependent description of the matter effect, we perform an expansion in of the
oscillation amplitude:
sin2 2m sin2 2 = 2 sin2 2 cos 2 2
d
dmin
sin2 2 cos 2.
(54)
Unlike the previous case, the relative matter effect is linear in d/dmin .
4. Refraction of high energy neutrinos

In this section we examine the refraction of high energy neutrinos (s & MZ2 ), both in
matter and in neutrino background.
4.1. High energy neutrinos in matter
Let us consider the propagation of high energy neutrinos in medium composed of
protons, neutrons and electrons. The expressions (1) and (10) refer to the low energy range,
271
2 , where s is the center of mass energy squared of the incoming neutrino and the
s MW
target electron, and MW is the mass of the W boson. The general formulas, valid for high
energies too, can be obtained by restoring the effect of the complete propagator of the W
boson in the expression of the potential (an analogous argument holds for the Z boson).
Let us consider e conversion. Since the refraction effects are determined by the real
part of the propagator, the potential (1) is generalized as:

2
,
(55)
V = 2 GF ne f qW

2
f qW
2
1 qW
2 )2 + 2
(1 qW
W
(56)
2 q 2 /M 2 and /M ; q and are the four momentum and the width

where qW
W
W
W
W
W
of the W boson.
The only contribution to the potential (55) is given by the forward charged current
scattering on electrons (u-channel exchange of W ), for which q 2 ' s. Therefore,
2 , we have q 2 ' s . From the potential (55) we can find the
introducing sW s/MW
W
W
refraction width d0 using the definition (14). For the nucleon refraction width (35) we find:
d0N (sW ) =
d0
1
d0
= (1 + sW ),
Ye f (sW ) Ye
(57)
where we have neglected the width W . Eq. (57) shows that d0N (sW ), and therefore
dmin (sW ), increase linearly with sW above the threshold of the W boson production.
For the active-sterile conversion, one has to take into account also the neutral current
interaction channel (t-channel exchange of Z), for which q 2 = 0, so that the low energy
formulas (1)(10) are still valid. For s only neutral current processes are involved,
thus the low energy result, Eq. (44), holds at high energies too. In contrast, for the e s
case both charged and neutral current interactions contribute, and for an electrically neutral
medium the high energy potential can be written as:

nn
1
.
(58)
V = 2 GF ne
1 + sW
2np
The second term in Eq. (58) does not depend on sW , thus coinciding with the corresponding
term in the low energy expression (39). The potential (58) gives the nucleon refraction
width:

1 + sW
.
(59)
d0N (sW ) = 2d0
(3Ye 1) sW (1 Ye )
For isotopically neutral medium (Ye = 1/2) we get:

1 + sW
.
d0N (sW ) = 4d0
1 sW
(60)
The width d0N (sW ) diverges for sW 1 (see Fig. 1).

At high energies inelastic interactions and absorption become important: at sW 1 the
imaginary part of the interaction amplitude is comparable with the real part. In Fig. 1 we
show the refraction width d = mN d0N for e and e s conversion and the absorption
272
Fig. 1. The width d = mN d0N for e (dashed line) and for e s (dotted line) channels, and the
absorption width, dabs , for the electron neutrino (solid line), as functions of the neutrino energy. We
have considered isotopically neutral medium, Ye = 0.5. The data for dabs are taken from Ref. [15].
width dabs [15,16] as functions of the neutrino energy E in the rest frame of the matter.
We have considered isotopically neutral medium, Ye = 1/2. The absorption width dabs is
dominated by the contribution of neutrinonucleon scattering; it decreases monotonically
with the energy E. In contrast, d starts to increase at sW 1, which corresponds to E =
106 107 GeV, according to Eq. (57). For E ' 106 GeV absorption and refraction become
comparable; at higher energies, the former effect dominates: dabs . d0 . This means that for
a e with energy E > 106 GeV the conversion in matter is damped by inelastic interactions
and absorption [1719], therefore a strong conversion effect cannot be expected.
Notice that for small mixing angle the minimum width dmin is significantly larger than
the refraction width d0 , therefore the absorption starts to be important at lower energies.
Taking, for instance, sin 2 = 0.3 we have dmin ' d0 / sin 2 ' 3.3d0 , and find that dabs .
dmin already for E & 5 105 GeV.
Let us consider now the matter effect for conversion of antineutrinos. For e channel
the only contributing interaction is the e e scattering with W exchanged in the s-channel.
In this case q 2 = s, and using Eq. (56) we get:
d0N (sW ) =
1
d0
.
Ye |f (sW )|
(61)
This function has a pole at sW = 1, i.e., in the W -boson resonance. The pole appears
because the amplitude becomes purely imaginary in the resonance, so that the potential is
zero. The width d0N (sW ) diverges for sW , due to the 1/sW decrease of the amplitude.
The function (61) has two minima:

d0
min
min
= 2W d0N (sW = 0) at sW
= 1 W .
d0N sW
= 2W
Ye
(62)
273
Fig. 2. The width d = mN d0N for e (dashed line) and for e s (dotted line) channels, and
the absorption width, dabs , for the electron antineutrino (solid line), as functions of the neutrino
energy. We have considered isotopically neutral medium, Ye = 0.5. The data for dabs are taken from
Ref. [16].
min ) = 0.05 d (s
Numerically d0N (sW
0N W = 0), which shows that refraction effects are
enhanced close to the W resonance. However, in this region inelastic interactions become
already important.
For e s channel the contribution of neutrinonucleon scattering should be included,
and for electrically neutral medium we find:

(1 sW )2 + W2
.

(63)
d0N (sW ) = 2d0
2
2
(3Ye 1) + 2sW (1 2Ye ) sW (1 Ye ) W (1 Ye )
In the case of isotopically neutral matter Eq. (63) gives:

(1 sW )2 + W2

,
d0N (sW ) = 4d0
2 2
1 sW
W
(64)
which has the value 4d0 in the limits sW 1 and sW 1, and a pole at sW ' 1. Similarly
to Eq. (62) we find the minima:
min
min
) = 4W d0 = W d0N (sW = 0) at sW
= 1 W .
d0N (sW
(65)
In Fig. 2 we show the refraction width d = mN d0N for e and e s channels and
the absorption width for the electron antineutrino, dabs [16], as functions of the neutrino
energy. We have considered isotopically neutral medium. For energies outside the W boson
resonance interval the main contribution to dabs is given by the neutrinonucleon scattering;
at sW ' 1 the effect of the resonant e e scattering dominates, providing the narrow
peak. It appears that absorption prevails over refraction (dabs < d0 ) for E & 6 106 GeV,
corresponding to d ' 6 107 g cm2 , for both e and e s cases.
The effect of absorption on neutrino conversion starts to be important at lower energies:
for sin 2 = 0.3 we find that dabs . dmin at E & 6 105 GeV.
274
4.2. High energy neutrinos in neutrino environment

Let us consider a beam of neutrinos which propagates in a background made of neutrinos
of very low energies. 5 This could be the case of beams of low energy neutrinos from
supernovae, or high energy neutrinos from AGN and GRBs, or neutrinos produced by the
annihilation of superheavy relics, etc. We assume that the background consists of neutrinos
and antineutrinos of various flavours, with number densities ni (i = e , e , , etc.). In the
case of relativistic neutrino background we assume its isotropy.
The potential for a neutrino ( = e, , ) due to neutrinoneutrino interaction can be
written as:

X
(ni ni ) ,
(66)
V (sZ ) = 2 GF n f (sZ ) n f (sZ ) +
i=e , ,
where the propagator function f (sZ ) has been defined in Eq. (56). Here sZ s/MZ2 and
Z Z /MZ ; MZ and Z are the mass and width of the Z-boson. The first term in eq.
(66) is due to scattering with Z-boson exchange in u-channel, and the second term
is the contribution from annihilation.
4.3. Neutrino conversion in CP-asymmetric neutrino background
As a first case we consider a strongly CP-asymmetric neutrino background, and suppose
ni ni , so that we can neglect the contributions of antineutrinos in (66). For simplicity,
we assume equal concentrations for the neutrino species: ne = n = n . In terms of the
total number density of neutrinos, n ne + n + n , the potential (66) reduces to:

1
(67)
V (sZ ) = 2 GF n 1 + f (sZ ) .
3
The potential for the antineutrino is given by V (sZ ) = V (sZ ).
Let us now find the refraction width d0 and dmin for various channels.
1. For the activesterile conversion, s , the potential (67) coincides with the
difference of the potentials for the two species, and therefore, by Eq. (14), it gives
immediately the refraction width of neutrinos:
1

1

(68)
d0 (sZ ) = d0 1 + f (sZ ) .
3
The width d0 (sZ ) is constant for sZ 1 and sZ 1: d0 (sZ 1) ' 3d0 /4 = 1.84
1032 cm2 , and d0 (sZ 1) ' d0 (see Fig. 3).
Let us now compare the refraction and absorption effects. The main contribution to
the absorption width dabs [20] is given by the and ( 6= ) scatterings.
The width dabs decreases monotonically with sZ and at sZ 1 it takes the value
dabs (sZ 1) ' /(2G2F MZ2 ) ' 3.6 1033 cm2 . Due to its non-resonant behaviour, dabs
5 We will not consider the conversion of neutrinos in the background itself, which can significantly affect the
flavour content of the background.
275
Fig. 3. The dependence of the refraction width d0 for s (dashed line) and s (solid line)
channels on sZ in strongly CP-asymmetric background, ni ni . Equal concentrations are assumed
for the various flavours.
is largerthan d0 for any energy of the neutrinos: at sZ 1 we find that d0 /dabs '
GF MZ2 / 2 = W /(2 cos2 W ) ' 0.1, where W and W = g 2 /4 are the weak mixing
angle and coupling constant. Therefore, dabs is also larger than the minimum width, dmin ,
for sin 2 & d0 /dabs ' 0.1.
2. For the conversion of an active antineutrino into a sterile species, s , we get the
width:
1

1

(69)
d0 (sZ ) = d0 1 + f (sZ ) .
3
This function (see Fig. 3) has a resonant behaviour with minima at sZ ' 1Z : d0 (1Z )
' (1/6Z + 1)1 d0 ' d0 /7 and d0 (1 + Z ) ' (1/6Z 1)1 d0 ' d0 /5. Outside the
Z-boson resonance d0 (sZ ) is constant: d0 (sZ 1) ' 3d0 /4 and d0 (sZ 1) ' d0 . In
the range sZ 1 inelastic scattering and absorption become important. We evaluate the
absorption width dabs for antineutrino in neutrino background using the plots in Ref. [20].
For sZ < 1, the width dabs decreases with the increasing sZ ; at sZ ' 1 it shows the
characteristic peak due to the resonant scattering. For sZ & 1 the absorption width
increases with sZ up to the limit dabs (sZ 1) 3 1033 cm2 , due to the contributions
of scatterings ( 6= ) and interaction in the t-channel. We find that dabs &
d0 (sZ ) for sZ . 0.8 (s . 7 103 GeV2 ) and for sZ & 2 (s & 1.7 104 GeV2 ). Furthermore,
d0 (sZ = 0.8) ' 1032 cm2 and d0 (sZ = 2) ' 4 1032 cm2 . Taking sin 2 = 0.3 we
get that dabs & dmin for sZ . 0.7 and sZ & 2.4; dmin (sZ = 0.7) ' 3 1032 cm2 and
dmin (sZ = 2.4) ' 1033 cm2 .
Notice that, in contrast with the conversion in matter (see Figs. 1 and 2), for neutrinos
and antineutrinos in neutrino environment we can have dabs & dmin (sZ ) even in the high
energy range, sZ & 1: in particular, we find that dabs(sZ 1) & dmin (sZ 1) for sin 2 &
0.1.
276
3. Let us now consider the activeactive conversion, . Assuming equal concentrations of neutrinos of different flavours, ne = n = n , we find from Eq. (67) that the
difference of the potentials of the two species equals:

1

2 GF n f sZ f sZ ,
(70)
1V, = V V =
3
where sZi s i /MZ2 (i = , ), and s i is the center of mass energy squared of the incoming
and the background neutrino of the same type i. If the species and in the background
have different energies we get that sZ 6= sZ , and therefore 1V, 6= 0, leading to matter
induced neutrino conversion even if and have equal concentrations. This situation
is realized if the background neutrinos and have different masses, e.g., m > m ,
and are non-relativistic. Denoting by E the energy of the neutrino beam, in the rest frame
of the background we have s i = 2mi E, and thus sZ /sZ = m /m > 1. The condition
sZ 6= sZ is achieved also if one of the neutrino species is relativistic and the other is not:
m E m , where E is the energy of in the background. Assuming the isotropy
of the neutrino gas, we have that s ' 2E E.
Using (70) and (14), we get the refraction width:

(1 + sZ )(1 + sZ )

1
i

.

' 3d0
(71)
d0 sZ = 3d0 f sZ f sZ

(sZ sZ )
The function (71) diverges for sZi and sZi 0. In particular, in the low energy
limit, sZ 1, it reduces to d0 (sZi 1) ' 3d0 /(2E1m), where 1m m m . For
the realistic case sZ & 1 and sZ 1, Eq. (71) can be written as:

1 + sZ

d0 sZ ' 3d0 ,
sZ
(72)
which approaches the minimum value 3d0 when sZ 1 (see Fig. 4). Taking the maximal
realistic values for the mass and energy of the neutrino, m = 5 eV and E = 1022 eV we
get sZ ' 12 at most, so that d0 (sZi ) ' 3.5d0 .
4. For the channel the effective potential equals:

1

2 GF n f sZ f sZ ,
(73)
1V, =
3
and therefore we get the width:

1
(74)
d0 sZi = 3d0 f sZ f sZ .
Due to the resonant character of the function f (sZ ), the width d0 (sZi ) has the following
features (see Fig. 4):
(i) It reaches the local minimum d0 (sZi ) ' 6Z d0 d0 /6 when one of the sZi s is at
resonance and the other is outside the resonance: sZ ' 1 and sZ 6= 1 (or vice versa).
(ii) The absolute minimum d0 (sZi ) ' 3Z d0 is achieved when sZ ' 1 + Z and sZ '
1 Z (or vice versa). These conditions can be satisfied for certain relations
between the masses of the background neutrinos. For non-relativistic background:
m /m = (1 + Z )/(1 Z ).
277
Fig. 4. The dependence of the refraction width d0 for (dashed line) and (solid line)
channels on sZ in strongly CP-asymmetric background, ni ni . Equal concentrations are assumed
s 0.
for the various flavours. We have considered sZ
Z
Notice that for sZi discussed in (i) and (ii) the effects of inelastic scattering and
absorption can be important.
(iii) If one of the sZi s is far below the resonance and the other is far above (e.g., sZ 1
and sZ 1) then d0 (sZi ) 3d0 .

i
(iv) d0 (sZ ) d0 if both the sZi s are far below or far above the resonance.
Obviously, for strong CP-asymmetric background with ni ni the results for and
channels should be interchanged.
4.4. Neutrino conversion in CP-symmetric neutrino background
Let us now consider the neutrino conversion in a CP-symmetric neutrino background,
ni = ni , with ne = n = n . In this case the potential (66) can be written as:

(75)
V = 2 GF n f (sZ ) f (sZ ) .
It vanishes in the low energy limit sZ 0, but it is unsuppressed at high energies, sZ & 1,
leading to significant matter effect. We will consider the conversion of neutrinos; due to
the CP-symmetry antineutrinos undergo analogous effects.
1. For s conversion from the potential (75) we find the refraction width for the
neutrinos of flavour :
1

d0 (sZ ) = d0 f (sZ ) f (sZ )

(1 + sZ2 + Z2 )2 4sZ2
.

(76)
= d0
2sZ [(1 sZ2 ) + Z2 ]
For sZ 1 it behaves as d0 (sZ ) ' d0 /(2sZ ) and for sZ 1 we have d0 (sZ ) ' d0 sZ /2 (see
Fig. 5). The width (76) has two minima:
278
Fig. 5. The dependence of the refraction width d0 for s and s channels on sZ in

CP-symmetric neutrino background.
d0 (sZmin ) ' 2Z d0
at sZmin = 1 Z .
(77)
Numerically, d0 (sZmin ) = 0.055 d0 ' 1.35 1031 cm2 .

The absorption width dabs is dominated by annihilation, with a resonance peak
at sZ 1. Using the results of Ref. [20] we find that dabs is larger than d0 (sZ ) outside the
Z-boson resonance, and the two quantities are comparable at sZ 1 or at sZ 1. For
sin 2 = 0.3 we find that the minimum width dmin is larger than dabs in the range 0.7 .
sZ . 1.6, corresponding to 6 103 GeV2 . s . 1.3 104 GeV2 . At the edges of this
interval the width dmin takes the value dmin(sZ = 0.7) ' dmin (sZ = 1.6) ' 3 1032 cm2 .
2. For the channel, Eq. (75) gives the difference of potentials:
1V, = V V

= 2 GF n f sZ f sZ f sZ f sZ .
(78)
The corresponding refraction width equals:

1
d0 sZi = d0 f sZ f sZ f sZ f sZ .
(79)
We find that d0 (sZi ) & d0 when both sZ and sZ are outside the Z-boson resonance and
d0 (sZi ) takes its minimum values when either sZ or sZ is close to the Z-resonance:
If sZ 1 + Z and sZ 1 Z (or vice versa) d0 (sZi ) has the absolute minimum
d0 (sZi ) ' Z d0 . For sZ 1 and sZ 6= 1 (or vice versa) we have the local minimum d0 (sZi ) '
2Z d0 . Notice that in the realistic case sZ sZ 0 the width (79) reduces essentially
to the one in Eq. (76). At resonance, where d0 (sZi ) has minima, the effects of inelastic
collisions and absorption are important.
279
5. Applications
The results derived in the previous sections are now applied to some physical situations
of interest. After a brief discussion of well known cases, like the Earth, the Sun and
supernovae, we present results for neutrinos in some new astrophysical environments. We
find that significant matter induced conversion can be expected for neutrinos crossing the
dark matter halos of clusters of galaxies and for neutrinos from cosmologically distant
sources.
5.1. Minimum width condition and bounds on the mixing
As follows from the analysis in Section 2, a significant neutrino conversion in matter
requires the fulfilment of the minimum width condition: 6
d0
.
(80)
tan 2
This condition is independent of the density distribution, and therefore of the specific
matter effect involved. Thus the knowledge of the width d allows one to conclude about
the significance of the matter effect even if the density profile is unknown. This is the case
of some astrophysical objects for which estimates or bounds on d can be obtained directly
by observational data with no assumption on their internal structure.
In the Table 1 we show the parameters of interest of some objects, together with the
values of the ratio
d
(81)
r .
d0
For r < 1, and small mixing angle, the condition (80) can not be satisfied, thus no
significant neutrino conversion is expected. Conversely, for r & 1, Eq. (80) can be fulfilled
and gives the bound on the mixing:
d > dmin =
1 d0
= .
(82)
r
d
Notice that our analysis holds for small mixings: sin 2 1. For applications we assume
sin 2 . 0.3, for which we find from Eq. (82) that the minimum width condition is satisfied
for r & 3.
The inequality (82) can be considered as the sensitivity limit for the mixing angle that
can be achieved by studies of neutrino conversion in a layer of given width d. The real
sensitivity can be however much lower than the absolute limit given by the condition (82).
This is related to the fact that in the case of varying density only part of the total amount of
matter effectively contributes to the conversion. Introducing the corresponding width d conv
we have the condition:
d0
(83)
sin 2 & conv ,
d
sin 2 &
6 This condition refers to the requirement of conversion probability larger than 1/2, Eq. (5). In some
circumstances, however, even a small effect, with conversion probability P 1/2 can be important.
280
Table 1
The density, the size and the matter width in units of refraction width, r = d/d0 , for various physical
objects. The values given for the densities are averaged along the trajectories of the neutrinos. We
quote the number density of electrons for objects made of usual matter, and the concentration of
the neutrino background for the halos and the universe. For the Earth the results are given for two
trajectories with different zenith angle z . The results for the universe correspond to redshift z = 5
for the cases of s and s in CP-symmetric and strongly CP-asymmetric neutrino background
with ' 1
Object
Earth:
cos z = 1
cos z = 0.81
Sun
Moon
Supernova
Universe
(n = n )
Universe
(n n )
Galactic halo
Cluster halo
AGN
GRB
Density (cm3 )
Size (cm)
r = d/d0
2.6 1024
1.5 1024
1.26 109
109
13.6
6.4
7 1024
1024
3 1033
6.96 1010
3.48 108
107
2600
1.4
109
1.5 104
1027
3 102
105
1027
0.3
. 2 106
3 1023
3 1024
22
d ' 10 1023 cm2
< 5 1015
5 102
10
1010 109
< 105
. 5 107
1010 1012
instead of the (82).

Let us find the expression of d conv for a medium with monotonically varying density.
As discussed in Section 2.3, the transition occurs mainly in the resonance layer. Using the
result (24) we get:
d conv = nres
e
2
dL
21n = nres
e ln tan 2,
dn
3
(84)
where ln |(dn/dL)1 |res nres

e . Inserting the expression (84) in the condition (83), we find:
3 d0
(85)
sin2 2 & res .
2ne ln
Clearly, d conv could be much smaller than the total width d of the object, so that the
condition (85) on the mixing could be much stronger than (82). Notice that the bound
(85) is quadratic in sin 2 . Using the definition
(19) of the adiabaticity parameter, , the
condition (85) can be written as 6 4/( 3), which corresponds to the adiabaticity
condition close to its limit of validity.
Another important issue is that the maximal sensitivity for the mixing can be achieved
for particular values of 1m2 /E, which depend on the specific density profile. As follows
from (85), for constant (or slowly varying with the distance) ln the smallest sin2 2
281
2
corresponds to the largest nres
e , and therefore to the largest values of 1m /E. This is the
k
case of exponential density profile. For power-law profile, ne L , we get |ln | = L/k,
so that sin2 2 Lk1 . Taking k > 1, fulfilled by practically all the realistic profiles, we
find that the smallest is achieved for the smallest L, and consequently the highest values
of ne and 1m2 /E.
Notice that d conv is a local property which depends on the derivative in ln . Of course, the
description given by d conv is not correct when the density profile is close to the constant
one, so that ln . In this case d conv can be even larger than the total width d. Thus, the
correct condition on the mixing can be written as:
sin 2 &
d0
.
min[d, d conv ]
(86)
5.2. The Sun, the Earth, the Moon and supernovae

For neutrinos crossing the Earth we consider two types of trajectories, corresponding to
different values of the zenith angle z . For cos z = 1 neutrinos travel along the diameter
of the Earth, crossing the core and the two layers of the mantle. We get r = 13.6, and
therefore according to (82) we could expect significant matter conversion for sin2 2 &
5 103 . However this maximal sensitivity, which would be achieved for uniform density
distribution, is not realized for the Earth profile. For small mixing, the difference between
the densities in the core and in the mantle is larger than the resonance interval. As a result,
the oscillations are resonantly enhanced either in the mantle or in the core, and only one of
the two parts effectively contributes to the effect. At the same time, for certain ranges of
1m2 /E, different from both the resonance values in the core and in the mantle, parametric
enhancement of oscillations occurs. Numerical calculations [3] give sin2 2 & 2 102 as
best sensitivity.
For cos z = 0.81 the trajectory is tangential to the core, and therefore it represents the
path of maximal length in the mantle. In this case we find r ' 6.4 and the sensitivity limit
sin2 2 & 2.5 102 . Since this case realizes approximatively the optimal condition of
uniform medium, we have good agreement with the results of exact calculations.
In the case of the Moon, r =1.4, and therefore a large mixing is required: sin2 2 &0.5.
A numerical integration of the density profile of the Sun [21] gives d ' 1.5
1012 g cm2 . Dividing this result by d = mN d0N , with Ye = 0.7, we find r ' 2600. From
the condition (82) we get then sin2 2 & 1.5 107 . This bound is remarkably weaker than
3 and l ' 0.3R , we get
the one obtained from the condition (85): taking nres
n

e ' 50 A cm
2
4
sin 2 ' 2.4 10 , in good agreement with the results of exact computations [2,3].
For supernovae the total width of the matter above the neutrinosphere gives r ' 109 , for
which the condition (82) would lead to sin2 2 & 1018 . Using the density profile ne =
n0e (R0 /R)3 [22], with R0 = 107 cm and n0e ' 1034 cm3 , from (85) we find sin2 & 108 ,
which agrees well with the results of numerical calculations [3].
As shown in the previous examples, the maximal sensitivity for sin2 2 , given by the
total width d, can be achieved in the case of uniform medium at 1m2/E corresponding
to the resonance density. Such a situation is realized for neutrinos crossing the mantle of
282
the Earth. In the case of substantial deviations from the constant density, like in the Sun
or in supernovae, the sensitivity is much lower. The stronger the deviation from constant
density, the smaller d conv , and therefore the lower is the sensitivity.
5.3. AGN and GRBs
Let us now turn to high energy neutrinos from Active Galactic Nuclei (AGN) and
Gamma Ray Bursters (GRBs) (for a review, see [2326]).
In AGN, neutrinos are considered to be produced by the interaction of accelerated
protons with a photon or proton background [2729]. There is a hope that neutrinos from
AGN with energies & 106 GeV could be detected by large scale underwater (ice) and EAS
detectors (for a review see [30,31]).
The width of matter crossed by neutrinos in an AGN can be estimated on the basis of the
existing data on the X-ray emission of these objects. The variability of the spectra suggests
that the X-radiation is emitted very close to the AGN core [24]. The proton acceleration
and therefore the neutrino production are supposed to happen in the same region. For this
reason the width of matter crossed by neutrinos equals approximatively the one crossed
by the X-radiation. For the later the experimental data [3234] give the value dAGN '
(102 101 ) A cm2 , therefore significant neutrino conversion in AGN is excluded 7
(for a short discussion, see also Ref. [35]).
A rather successful description of the origin of GRB is provided by the fireball model
[36], in which neutrino production is predicted to happen in an analogous way as in
AGN [13,37]. A fireball can emit protons, detected as high-energy cosmic rays on Earth,
accompanied by a flux of neutrinos. The requirement that the fireball should be transparent
to protons gives an estimate of the width of the object: dGRB 6 dabs , where dabs = 10
100 A cm2 is the total absorption width for the protons. It is possible to evaluate the width
in a different way. An estimate of the electron number density in the fireball is given in Ref.
[13]: nGRB ' (1010 1012 ) cm3 . Using this value, and taking the fireball mass in the
range of star-like objects, M = (1 10)M , we can get the radius of the object, RGRB =
5 (1014 1015 ) cm, and then the width: dGRB = 10 104 A cm2 . In agreement with
the first argument, we see, then, that also in GRBs the matter effect on neutrino conversion
is negligible.
5.4. Dark matter halos
According to models [38], part of the dark matter in halos of galaxies and clusters of
galaxies should consist of neutrinos. 8 Therefore neutrinos of extragalactic origin crossing
the halo on the way to the Earth undergo refraction on the neutrino background. It was
7 In the present discussion we have considered radial propagation of neutrinos from the inner to the external
regions of the object. We have not considered neutrinos travelling through the core of the AGN. In this case a
significant matter-induced conversion could occur, however neutrinos crossing the core are supposed to be a small
fraction of the total neutrino flux produced.
8 In what follows, we will not consider the heavy particles present in the halos, because their number density is
much smaller than the one of neutrinos, although they provide the largest part of the mass of the halo. Furthermore,
283
suggested in Ref. [14] that, due to non-uniform neutrino density distribution in the galactic
halo, ultrahigh energy neutrinos can be resonantly converted into active and sterile species.
Following Ref. [14] we consider a galactic halo composed of non-relativistic neutrinos
and antineutrinos of the two species and . The electron neutrino is assumed to be
lighter, and therefore less clustered, than and : ne ni , i = , . We assume
CP-symmetry of the background: ni = ni . We take the density profile [14]:
n (r) = n0
1
,
1 + (r/a)2
(87)
where n n + n and a is the core radius of the halo. For galactic halos this radius
is estimated to be a ' 10 100 kpc. An upper bound for n0 is given by the Tremaine
Gunn condition [39]: for identical fermions of mass m, the maximum number density nmax
allowed by the Pauli principle equals
1
(mv esc )3 ,
6 2
where v esc is the escape velocity of the particle:

1/2
1/2
2GM 1/2
M
R
540
km/s.
v esc =
R
1012M
30 kpc
nmax =
(88)
(89)
Here M and R are the total mass and radius of the galaxy. In what follows we will take
R ' 3a, corresponding to the radius at which the density reduces at one tenth of its core
value, n (3a) = n0 /10. From (88) and (89) we get:

3/2

3/2
m 3
M
a
cm3 .
(90)
nmax = 1.7 106
5 eV
10 kpc
1012 M
The integration of the profile (87) gives the matter width:
(91)
d = n0 a.
2
Inserting the expression (90) for nmax in (91), we find the upper bound for the width:

1/2

3/2
m 3
M
a
28
cm2 .
(92)
d . 8 10
5 eV
1012 M
10 kpc
According to (92) the largest values of d are achieved for objects with big mass M and
small radius a; so that compact halos represent the most favourable case.
Let us now check the minimum width condition for s and e conversion in
galactic halos. We use the refraction width d0 (sZmin ) = 0.055d0 ' 1.35 1031 cm2 given
in Eq. (77). This value was obtained for s conversion in CP-symmetric neutrino
background, and is the absolute minimum of d0 (sZ ) (Eq. (76)), realized at the Z-boson
resonance, sZ 1. Notice that, under the assumption ne ni , the result d0 (sZmin ) holds
also for e conversion: due to the absence of electron neutrinos in the background,
the amplitude of the forward scattering of neutrinos on neutrinos and on the heavy particles of dark matter are
comparable, or the former is even larger.
284
ne ' 0, the potential (75) for e is negligible, and therefore the electron neutrino behaves
as a sterile species.
From Eq. (92) we see that, for typical values of M and a of a galaxy, like for instance
the Milky Way (M ' 1012 M and a ' 10 kpc), the minimum width condition is not
satisfied: d/d0 (sZmin ) . 5 103 . For the galaxy M87 (M ' 1013 M and a ' 100 kpc)
we find d/d0 (sZmin ) . 5 102 . Taking an hypothetical very massive and compact object,
with M ' 1013M and a ' 10 kpc we get d/d0 (sZmin ) . 0.2. Thus, a significant neutrino
conversion effect in the galactic halo is excluded, in contrast with the result in Ref. [14].
Conversely, significant matter-induced conversion can be realized in halos surrounding
a cluster of galaxies. Taking the mass of a cluster as M = (1013 1015) M and the size
a ' 1 Mpc, we obtain from Eq. (92)

(93)
d . 1029 3 1032 cm2 ' (102 20)d0 sZmin .
For the maximal value , d/d0 (sZmin ) ' 20, from the condition (82) we get the sensitivity to
the mixing: sin2 2 & 3 103. With this value we find that the adiabaticity condition (18)
is fulfilled for 1m2 & 4 106 eV2 .
The maximal sensitivity is achieved in the energy range of the Z-resonance, where also
inelastic scattering and absorption are important. Indeed, in Section 4.4, taking sin 2 =
0.3, we have found that dabs dmin at s ' 6103 GeV2 , where dmin ' 31032 cm2 . This
value coincides with the upper edge of the interval (93). For smaller sin 2 dmin is larger,
and the absorption effect on oscillations becomes even more important. Notice that d0 (sZ )
takes its minimum value at the Z-resonance: for neutrino energies outside the resonance
d0 (sZ ) is larger, so that d/d0(sZ ) < 1 and the minimum width condition is not satisfied.
Thus, we have found that in the halos of clusters of galaxies a significant matter-induced
conversion can be achieved in the narrow interval of energies of the Z-boson resonance,
where, however, the absorption and the effects of inelastic interactions are important.
5.5. Early universe
In this section we consider neutrinos produced by cosmologically distant sources and
propagating in the universe. The refraction occurs due to the interaction of the neutrinos
with the particle background of the universe made of neutrinos electrons and nucleons.
The number densities of baryons, nb , and electrons, ne , are given by nb = ne =
b n , where b = 1010 109 is the baryon asymmetry of the universe and n is the
concentration of photons. At present time n = n0 ' 400 cm3 . We will describe the
neutrino background by the total number density n + n and the CP-asymmetry
(n n )/n . The value of is unknown. A natural assumption would be ' b : in
this almost CP-symmetric case the total concentration of neutrinos equals n + n =
4n /11 [40]. However strong asymmetry, 1, is not excluded. The upper bound
10 for muon and tau neutrinos follows from the Big Bang Nucleosynthesis [41,42].
For 1 the contribution of the neutrino background to refraction dominates and the
interaction of neutrinos with electrons and nucleons can be neglected. It can be checked
that the contributions of neutrinoelectron and neutrinonucleon scattering to the refraction
width are smaller than d0 at any time after the neutrino decoupling epoch, tdec ' 1 s.
285
In the framework of the standard Big Bang cosmology, the number density of neutrinos
in the universe decreases with the increasing time as [40]:

t0 2
,
t > teq ,
n
t
(94)
n(t) =
3/2
t
eq
, t < teq .
neq
t
Here t0 ' 1018 s is the age of the universe, and teq ' 1012 s is the time at which the energy
densities of radiation and of matter in the universe were approximatively equal. We denote
by n0 and neq the neutrino concentrations at t = t0 and t = teq , respectively.
The matter width d(t) crossed by the neutrinos from the time t of their production to the
present one is given by the integration of the concentration (94):

t0
d
1 , t > teq
Zt0
U t
(95)
d(t) = n( ) d =
1/2
teq
t
deq
, t < teq ,
t
where dU t0 n0 is the present width of the universe and deq = dU [t0 /teq 1] is the width
at t = teq .
In what follows we will focus on the case of matter domination epoch, t > teq , for which
the width (95) can be expressed in terms of the redshift, z (t0 /t)2/3 1, as:

d(z) = dU (z + 1)3/2 1 = di 1 (z + 1)3/2 ,
(96)
where di = tn = dU (z + 1)3/2 is the width of the universe at the time t of production
of the neutrino beam; n is the concentration of the neutrino background at t. According
to Eq. (96), for large enough z the width at the production time, di , gives the dominant
contribution.
Another important feature of the propagation of neutrinos from cosmological sources
is the redshift of energy. The refraction width d0 depends on the center of mass energy
squared s of the incoming and the background neutrinos. As a consequence of the redshift,
sZ = sZ (z), the width d0 changes with time (with z) during the neutrino propagation: d0 =
d0 (sZ (z)). Thus, the width of matter d should be compared with some effective (properly
averaged) refraction width d0 which in fact depends on the channel of transition. For nonrelativistic background neutrinos of mass m we have that s increases with z as: s '
2m E (1 + z), where E is the energy of the neutrino beam. For relativistic background
with energy Eb one gets s ' 2Eb E (1 + z)2 .
Let us consider neutrino propagation in a strongly CP-asymmetric background, ni ni ,
with 1. For simplicity we assume also flavour symmetry: ne = n = n .
For the s channel the refraction width (68) increases smoothly from its low energy
value, d0 (sZ 1) = 3d0 /4, to the high energy one, d0 (sZ 1) = d0 . For neutrinos
produced with energy E . 1021 eV and mass of the background neutrino m ' 2 eV
we get sZ . 0.5, which undergoes redshift during the neutrino propagation. Therefore we
286
can use the low energy value of d0 as the effective refraction width. From (96) we get the
ratio r d/d0 (sZ ) in terms of the redshift z and the asymmetry :
r(z)
4d(z)
' 2.2 102 (z + 1)3/2 .
3d0
(97)
Taking the maximal allowed asymmetry, 10, we find that r = 3 is reached at z ' 5,
which corresponds to rather recent epoch. Possible sources of high energy neutrinos, the
quasars, have been observed at such values of the redshift. With smaller asymmetries,
. 1, the minimum width condition requires much earlier epochs of neutrino production,
z & 27.
Notice that in general the minimum width condition, d > d0 / sin 2 , is not sufficient
to ensure a significant transition effect: as discussed in section 2, the width d1/2 required
to have conversion probability P > 1/2 depends on the specific effect involved. For the
adiabatic conversion in varying density (Section 2.3) we have found the result d1/2 '
1.5dmin (see Eq. (30)). Therefore, the condition P > 1/2 requires larger values of r(z),
r & 4.5. From Eq. (97), with 10, we get r 4.5 for z ' 7.5.
For s channel the refraction width d0 , Eq. (69), has a resonance character with
absolute minimum d0 (sZmin ) ' d0 /7 at sZmin 1. Outside the resonance it takes the values
d0 (sZ 1) = 3d0 /4 and d0 (sZ 1) = d0 . For neutrino energy E . 1021 eV at production
and mass of the background neutrino m . 1 eV we get sZ < 1, so that we can use
the low energy value of the refraction width and the result for the ratio r coincides with
that in Eq. (97). For neutrinos produced at time t = ti with extremely high energies, E
1021 1022 eV and mass of the background neutrino m ' 1 3 eV we get sZ > 1 at the
production time, so that, due to redshift, sZ (z) will cross the resonance interval, in which
d0 has minimum. This, however, does not lead to larger values of the ratio r(z), since
the time interval 1t during which sZ remains in the resonance range is short: 1t/ti '
1.5Z ' 0.04. The matter width collected in the interval 1t is dres ' d1t/ti ' 1.5Z d,
and the ratio r(z) for the resonance epoch equals r(z) = dres /d0 (sZmin ) ' d/4d0 . That is
even smaller than r outside the resonance, Eq. (97); therefore the result (97) holds also in
this case.
Thus, in the extreme condition of very large asymmetry and production epoch at
z & 5 the matter width crossed by neutrinos satisfies the minimum width condition. Let us
comment on the character of the matter-induced neutrino conversion in this case. After its
production, the neutrino beam experiences a monotonically decreasing density. Moreover,
the energy of neutrinos decreases due to redshift, which also influences the mixing. Taking
into account the decrease with time of both the neutrino energy and concentration, the
adiabaticity condition (18)(19) can be generalized as:
t
= 0 (z + 1)3/2 1,
t0
8
,
0 '
3V0 t0 tan2 2
(t) = 0
(98)
(99)
where V0 is the present value of the neutrino-medium potential. For the present epoch we
get 0 ' 102 / tan2 2 , so that the adiabaticity is strongly broken. For tan2 2 . 0.1 the
287
adiabaticity can be realized at t/t0 < 103 , or z > 102 . Thus, for z < 102 the conversion
takes a character of oscillations in the production epoch. A detailed study of the dynamics
of neutrino conversion in the universe will be given elsewhere [43].
For the flavour channel the refraction effect appears at high energies, sZ 1, as
a consequence of different energies of the background neutrinos and . The absolute
minimum of the refraction width (71), d0min (sZi ) = 3d0 , gives a value of r(z) which is
4 times smaller than that in Eq. (97). Correspondingly, even for the extreme conditions
of ' 10 and z = 5 we get r ' 0.8. That is, significant matter effect is excluded for
neutrinos from the oldest observed sources. The value r & 3 can be reached for z & 14.
Notice that, according to Eq. (71), the refraction disappears for low energies, sZi 1, and
the redshift spoils the conditions of absolute minimum of d0 , even if it is realized in certain
epoch. Therefore, larger z are required to have significant conversion effect.
Similar conclusions can be obtained for channel, where the refraction width (74)
has the local minimum d0 (sZi ) ' 6Z d0 due to Z-resonance. This minimum is realized
however during the resonance epoch 1t. Taking the corresponding matter width dres '
1.5Z d, we get r(z) = d(z)/4d0 , which is even smaller than in the case.
Let us consider CP-symmetric neutrino background, ni = ni , with the assumption of
flavour symmetry: ne = n = n . As discussed in Section 4.4, unsuppressed refraction
effect appears for activesterile neutrino channels at high energies, sZ 1, where the
propagator corrections become important.
For s (and similarly for s , due to CP-symmetry) the refraction width (76) has
the absolute minimum d0 (sZmin ) ' 2Z d0 , realized at sZmin = 1 Z , Eq. (77). Assuming
that the neutrinos are produced just before the resonance epoch, we find that the width
collected during the interval 1t equals dres ' d1t/ti ' 1.5Z d, where d is given by
Eq. (96), with dU = 2n0 t0 /11 ' 7 1029 cm2 . For the ratio r(z) we find:
r(z) =
3d(z)
= 2 103 (z + 1)3/2,
4d0
(100)
which is significantly smaller than r(z) for s channel in CP-asymmetric background,

Eq. (97). With the maximal redshift z ' 5 Eq. (100) gives r ' 0.03, which excludes matter
effect for neutrinos from the most distant observable sources. The result (100) holds also
for the ( ) channel, since the refraction width (79) has analogous behaviour
to the one for the s case, Eq. (76), with the same local minimum d0 (sZi ) ' 2Z d0 at
resonance.
In conclusion, we have found that the matter effect for neutrinos crossing the universe is
mainly due to the neutrino background. For neutrinos from observable sources (z ' 5),
significant conversion effect can be achieved in the s and s channels, if the
background has strong CP-asymmetry, close to the maximum value, ' 10. The matter
effect for the other conversion channels and for the CP-symmetric case is suppressed as a
consequence of the redshift.
288
6. Conclusions
1. Matter effects can lead to strong flavour transition even for small vacuum mixing
angle: 1. This however requires a sufficiently large amount (width) of matter crossed
by neutrinos:
the minimum width condition, d > dmin , should be satisfied, where dmin =
/(2 2 GF tan 2 ) = d0 / tan 2 , for low neutrino energies, s MZ2 , and conversion
probability P > 1/2. The absolute minimum dmin is realized for uniform medium with
resonance density nres
e .
2. We have shown that for all the other realistic situations the required width, d1/2 ,
is larger than dmin . In particular, we have found that d1/2/dmin = 1 + (1 /8) 2
for oscillations in medium with slowly varying density ( 1); d1/2/dmin > 1.5 for
conversion in medium with varying density; d1/2/dmin = for castle wall profile.
2
3. We discussed the minimum width condition for high energy neutrinos. For s & MW
the minimum width dmin becomes function of s, due to the propagator effect: dmin =
dmin (s). The function dmin (s) depends on the channel of interactions: in the case of W
(or Z) exchange in the s-channel dmin(s) decreases in the resonance region by a factor 20
2 ) 20. In this region, however, the
with respect to the low energy value: dmin(0)/dmin (MW
inelastic interactions become important, damping the flavour conversion.
4. As a case of special interest we have studied the refraction of high energy neutrinos
in neutrino background, which can be important for propagation of cosmic neutrinos in
galaxies and intergalactic space. Again we find that the annihilation channel gives
enhancement of refraction at s ' MZ2 , so that dmin can be 1/2Z 20 times smaller than
that at low energies. In the case of flavour channels the refraction can appear as the result
of the difference of masses of the background neutrinos, even if the concentrations of the
various flavours are equal.
5. The minimum width condition allows one to conclude on the relevance of the matter
effect without knowledge of the density profile, once the width d is known. In some
astrophysical situations the total width on the way of neutrinos can be estimated rather
precisely (e.g., by spectroscopical methods) although the density distribution is unknown.
Significant matter effect in excluded if d < d0 , or d < dmin if the mixing angle is known.
6. From practical point of view, a study of the matter effects should start with the check of
the minimum width condition, d > dmin . This condition is necessary but not sufficient for
strong conversion effect. If it is fulfilled the ratio d/d0 allows one to estimate the minimal
mixing angle for which significant transition is possible: sin 2 > d0 /d. This condition
gives an absolute lower bound on , which can be achieved for the case of uniform medium
with resonance density. In other words, given the width d of the medium, the highest
sensitivity to the mixing angle is achieved if the matter is distributed uniformly and the
density coincides with the resonance value for a given neutrino energy. For media with
non-uniform matter distribution the sensitivity to is lower. The stronger the deviation
from the constant density, the lower the sensitivity.
7. We applied the minimum width condition to neutrinos in AGN and GRBs
environment. For AGN the width d can be estimated by the experimental data on the
X-ray spectrum, without assumptions on the the density profile. We got d/d0 . 1010
289
for radially moving neutrinos, strongly excluding matter effects. In the case of GRBs the
width d can be evaluated under the assumption that the object is transparent to protons. We
found d/d0 . 105 . Therefore, no significant conversion is expected either.
8. For neutrinos crossing the halos of galaxies and clusters of galaxies the matter
effect is given by the interaction of neutrinos with the neutrino component of the halo.
We have found that for galactic halos the minimum width condition is not satisfied: the
result d(halo)/d0 6 0.1 excludes any significant conversion effect. For halos of clusters of
galaxies we got d(halo)/d0 & 10, and the minimum width condition can be satisfied for
large enough mixing: sin 2 & 0.1.
9. We have considered the refraction of neutrinos from cosmologically distant sources,
interacting with the neutrino background of the universe. Significant activesterile
conversion can be expected in case of large asymmetry. We have found that for
= O(1) the condition d(universe)/d0 & 1 can be achieved for neutrinos from sources,
galaxies or quasars, with redshift z & 5. The effect on detected neutrino fluxes from these
sources could be a distortion of the energy spectrum.
Acknowledgement
The authors wish to thank O.L.G. Peres for useful comments.
References
[1] L. Wolfenstein, Phys. Rev. D 17 (1978) 2369.
[2] S.P. Mikheev, A.Y. Smirnov, Nuovo Cimento C 9 (1986) 17; Yad. Fiz. 42 (1985) 1441; Sov. J.
Nucl. Phys. 42 (1986) 913.
[3] S.P. Mikheev, A.Y. Smirnov, Resonant neutrino oscillations in matter, Prog. Part. Nucl. Phys. 23
(1989) 41.
[4] V.K. Ermilova, V.A. Tsarev, V.A. Chechin, Bull. Lebedev Phys. Inst. 351 (1988).
[5] E.K. Akhmedov, Yad. Fiz. 47 (1988) 475.
[6] P.I. Krastev, A.Y. Smirnov, Phys. Lett. B 226 (1989) 341.
[7] H. Suzuki, Supernova neutrinos, in: M. Fukugita, A. Suzuki (Eds.), Physics and Astrophysics
of Neutrinos, Springer-Verlag, 194.
[8] Q.Y. Liu, A.Y. Smirnov, Nucl. Phys. B 524 (1998) 505, hep-ph/9712493.
[9] Q.Y. Liu, S.P. Mikheyev, A.Y. Smirnov, Phys. Lett. B 440 (1998) 319, hep-ph/9803415.
[10] S.T. Petcov, Phys. Lett. B 434 (1998) 321, hep-ph/9805262.
[11] E.K. Akhmedov, Nucl. Phys. B 538 (1999) 25, hep-ph/9805272.
[12] A.D. Dolgov, S.H. Hansen, S. Pastor, D.V. Semikoz, hep-ph/9910444, and references therein.
[13] E. Waxman, J. Bahcall, Phys. Rev. Lett. 78 (1997) 2292, astro-ph/9701231.
[14] R. Horvat, Phys. Rev. D 59 (1999) 123003, hep-ph/9812228.
[15] R. Gandhi, C. Quigg, M.H. Reno, I. Sarcevic, Phys. Rev. D 58 (1998) 093009, hep-ph/9807264.
[16] R. Gandhi, C. Quigg, M.H. Reno, I. Sarcevic, Astropart. Phys. 5 (1996) 81, hep-ph/9512364.
[17] L. Stodolsky, Phys. Rev. D 36 (1987) 2273.
[18] G. Raffelt, G. Sigl, L. Stodolsky, Phys. Rev. Lett. 70 (1993) 2363, hep-ph/9209276.
[19] G. Sigl, G. Raffelt, Nucl. Phys. B 406 (1993) 423.
[20] E. Roulet, Phys. Rev. D 47 (1993) 5247.
[21] J.N. Bahcall, Neutrino Astrophysics, Cambridge University Press, 1989.
290
[22] D. Notzold, Phys. Lett. B 196 (1987) 315.

[23] R.J. Protheroe, Nucl. Phys. Proc. Suppl. 77 (1999) 465, astro-ph/9809144, and references
therein.
[24] V.L. Ginzburg, V.A. Dogiel, V.S. Berezinsky, S.V. Bulanov, V.S. Ptuskin, Astrophysics of
Cosmic Rays, North-Holland, 1990, 534 p.
[25] S. Blinnikov, astro-ph/9911138, and references therein.
[26] T. Piran, astro-ph/9907392, and references therein.
[27] F.W. Stecker, C. Done, M.H. Salamon, P. Sommers, Phys. Rev. Lett. 66 (1991) 2697.
[28] A.P. Szabo, R.J. Protheroe, Astropart. Phys. 2 (1994) 375, astro-ph/9405020.
[29] L. Nellen, K. Mannheim, P.L. Biermann, Phys. Rev. D 47 (1993) 5270, hep-ph/9211257.
[30] F. Halzen, astro-ph/9810368.
[31] T.K. Gaisser, F. Halzen, T. Stanev, Phys. Rep. 258 (1995) 173, hep-ph/9410384.
[32] R.F. Mushotzky, Astrophys. J. 256 (1982) 92.
[33] T.J. Turner, K.A. Pounds, Mon. Not. Roy. Astron. Soc. 240 (1989) 833.
[34] R.M. Sambruna, M. Eracleous, R.F. Mushotzky, astro-ph/9905365.
[35] H. Minakata, A.Y. Smirnov, Phys. Rev. D 54 (1996) 3698, hep-ph/9601311.
[36] P. Meszaros, M. Rees, Mon. Not. Roy. Astron. Soc. 269 (1994) 41P.
[37] E. Waxman, J. Bahcall, hep-ph/9909286.
[38] M. Davis, F.J. Summers, D. Schlegel, Nature 359 (1992) 393, and references therein.
[39] S. Tremaine, J.E. Gunn, Phys. Rev. Lett. 42 (1979) 407.
[40] E.W. Kolb, M.S. Turner, The Early Universe, Addison Wesley, 1990.
[41] H. Kang, G. Steigman, Nucl. Phys. B 372 (1992) 494.
[42] J. Lesgourgues, S. Pastor, S. Prunet, hep-ph/9912363.
[43] C. Lunardini, A.Yu. Smirnov, in preparation.

Horizon states for AdS black holes

T.R. Govindarajan a , V. Suneeta a , S. Vaidya b
a Institute of Mathematical Sciences, Chennai, 600 113, India
b Tata Institute of Fundamental Research, Colaba, Mumbai, 400 005, India
Received 14 February 2000; accepted 23 May 2000
Abstract
We study the time-independent modes of a massless scalar field in various black-hole backgrounds,
and show that for these black holes, the time-independent mode is localized at the horizon. A similar
analysis is done for time-independent, equilibrium modes of the five-dimensional plane AdS black
hole. A self-adjointness analysis of this problem reveals that in addition to the modes corresponding
to the usual glueball states, there is a discrete infinity of other equilibrium modes with imaginary
mass for the glueball. We suggest these modes may be related to a SavvidyNielsenOlesen-like
vacuum instability in QCD. 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.62.+v; 04.70.-s
Keywords: Scalar fields; Black holes; Zero modes; AdS/CFT; Glueball spectrum
1. Introduction
A study of various kinds of matter fields propagating in black-hole backgrounds yields
information about diverse classical and quantum aspects of black-hole physics. Detailed
analysis of modes of the scalar, spinor and gauge fields in black-hole backgrounds can be
found for example, in [1]. In particular, for scalar fields, the energies of these modes are
given by the square root of the eigenvalues of the spatial part of the KleinGordon operator
in that background. For static spacetimes with null singularities, it has been argued [2,3]
that the spatial part of the KleinGordon operator is essentially self-adjoint. Further, since
it is positive and symmetric, one can choose a positive self-adjoint extension such that the
eigenvalues are all positive and hence the energies real. However, as we show in this paper,
in the case of many black-hole spacetimes, near the null singularity at the horizon, the
zero (time-independent) mode of the scalar field has to be handled separately. In particular,
the boundary conditions imposed on the zero mode both at the horizon and at infinity are
E-mail addresses: trg@imsc.ernet.in (T.R. Govindarajan), suneeta@imsc.ernet.in (V. Suneeta), sachin@
theory.tifr.res.in (S. Vaidya).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 3 6 - 9
292
T.R. Govindarajan et al. / Nuclear Physics B 583 (2000) 291303
different from those on the other modes with real energies. In fact, we will show that there
are an infinite number of boundary conditions, labeled by a U (1) parameter, that lead to
one zero mode solution. This solution could be thought of as a horizon state as it is
localized at the horizon.
An application of this analysis to the time-independent solutions of the infinite mass
limit of the AdSSchwarzschild black hole [4] leads to results that could be interesting in
light of the AdS/CFT correspondence. As proposed by Witten [5], the AdS/CFT duality
relating supergravity on anti-de Sitter space to a supersymmetric YangMills theory on the
boundary can be extended to non-supersymmetric QCD. The AdS background is replaced
by an AdSSchwarzschild black-hole background. It has been shown that gravity on this
background gives many of the features of strong coupling limit of QCD, like the area
law behavior of Wilson loops, confinement, and the glueball mass spectrum with a mass
gap [58].
The glueball mass spectrum is reproduced by certain time-independent and normalizable
modes obtained by solving the dilaton wave equation in the black hole geometry. These
modes were numerically computed first in [6,7]. These modes are equilibrium modes for
the black hole, i.e. the current vanishes at the horizon, which has been recognised in [8] as
the correct boundary condition to be used at the horizon.
It has been argued that nonequilibrium modes of the same black hole (with ingoing
boundary conditions at the horizon) give the time scale of approach to thermal equilibrium
of the boundary YangMills theory. These modes, i.e. the quasinormal modes of the black
hole, have been computed recently [9,10].
In this paper, we study the scalar wave equation in the AdSSchwarzschild background,
and show, that written as a hamiltonian problem, it is not self-adjoint. Self-adjointness
and completeness requires inclusion of modes ignored in [68]. These modes are also
equilibrium modes of the black hole but are irregular at the horizon. 1 They are also
tachyonic. We suggest that these modes are dual in the AdS/CFT sense to modes in QCD3
signaling the onset of a SavvidyNielsenOlesen-like instability of the vacuum [1114].
The organization of the paper is as follows. In Section 2, we briefly describe two kinds
of modes that are commonly discussed in related literature, namely the normalizable
equilibrium modes, and the nonnormalizable quasinormal modes, to emphasize the
differences between them. We also show that for a massless scalar field propagating in
Schwarzschild or ReissnerNordstrom black-hole background, the KleinGordon operator
is self-adjoint. In Section 3, we focus on the zero energy mode of the scalar field in these
backgrounds, and in the background of the (1 + 1)D black hole [15] as well as the BTZ
black hole [16]. The equation obeyed by the zero mode has unusual properties, which
we analyze in Section 4. In particular, we show that this state localized at the horizon. In
Section 5, we apply the results of Section 4 to study the zero mode of the massless scalar
field in the background of the infinite mass limit of the AdSSchwarzschild black hole, and
argue that the horizon states are necessary for completeness. In Section 7, we speculate
1 In [7,8], the existence of irregular modes is mentioned. However, they are not considered.
293
on the interpretation of these irregular modes in the boundary theory, and suggest that they
may be related to a SavvidyNielsenOlesen-like instability.
2. Modes of the scalar field in black-hole background

As mentioned before, the energies of normalizable modes of a scalar field in the exterior
of a black-hole spacetime (i.e. in the region from the outer horizon to infinity) have real
energies.
This can be verified for the Schwarzschild or ReissnerNordstrom black hole in the
exterior. There are no normalizable mode solutions with complex (or pure imaginary)
energies. However, this is not true in a region of the black-hole spacetime near a timelike singularity. For the ReissnerNordstrom spacetime, in the region between the timelike
singularity and the inner horizon, the spatial part of the KleinGordon operator is not
self-adjoint, as also observed by [17], but can be made self-adjoint by a suitable choice
of boundary conditions. There exist boundary conditions for which there is a negative
eigenvalue for this operator, leading to a mode solution with imaginary energy. However,
this solution is not extendible to the physical region of interest between the outer horizon
and infinity.
Other modes of importance in the context of black holes are the quasinormal modes (see,
for example, [18,19]). For the case of asymptotically flat black holes, these are defined to
be purely ingoing near the horizon and outgoing at infinity. These are not normalizable,
but are of interest as their energies are the characteristic frequencies associated with the
perturbation of the black hole. These are in general, complex, and decay with time. In the
Schwarzschild and ReissnerNordstrom cases, there are an infinite number of such modes
(see [18] for references) which include purely imaginary modes [20].
Recently, quasinormal modes for the AdSSchwarzschild black hole have also been
studied [10]. These are different from the quasinormal modes for asymptotically flat black
holes, in that they are not outgoing at infinity, but vanish. This is due to the fact that the
AdS potential diverges at infinity. Numerical results of [10] suggest that these modes are
complex. However, they are still nonnormalizable due to their behavior at the horizon.
An analysis of the spatial part of the KleinGordon operator for the AdSSchwarzschild
black hole shows that as expected in [3], the operator is self-adjoint, and all the
normalizable modes have real energies. The AdSSchwarzschild black hole has a metric
ds 2 = F (r) d 2 + F 1 (r) dr 2 + r 2 d2 ,

F (r) = 1 + r 2 /b2 r02 /r 2 .
where
(2.1)
(2.2)
Here b is the radius of curvature of the anti-de Sitter space and r0 is related to the blackhole mass,
M=
3A3 r02
16G5
and A3 is the area of a unit 3-sphere.
(2.3)
294
Let us look at a massless scalar field in this background geometry. One can in principle
consider a complex scalar field with charge q and mass m, but for simplicity we shall
consider only the massless and uncharged field in the black-hole background. The action
for such a field 8 is
Z
1 p
|g|g ij (i 8)(j 8) d5x,
S=
2
2

Z
Z Z
2
1
3 8
02
2
+ F 8 + 1/r 8L 8 .
dr dt d r
(2.4)
=
2
F
r+
The KleinGordon equation for the field 8 can be obtained from above. On making the
ansatz
f (r)
Y (angles) exp(it),
(2.5)
r 3/2
the wave functions are defined on the measure dr/F . The KleinGordon equation can then
be written in terms of the tortoise coordinate r , which is defined by dr = dr/H . It takes
the form
8=
d2
f + V (r )f = 2 f,
dr2
(2.6)
with the measure now being dr . The potential is positive, vanishes at the horizon r =
and diverges at r = . This corresponds to a finite r and therefore the solutions have to
vanish there. Multiplying (2.6) by the complex conjugate of f and integrating over the
spacetime from the horizon to infinity, it can be seen that there can be no normalizable
solutions that correspond to 2 negative or complex. This is in conformity with the fact
that the KleinGordon operator is self-adjoint.
3. Time independent mode in black-hole solutions

The positive energy solutions to the KleinGordon equation can be analyzed for most
black-hole solutions by going to the tortoise coordinate r mentioned in the previous
section. The solutions behave as f exp i(t r ) near the horizon and near infinity.
The horizon is at r = , while the infinity of the Schwarzschild radial coordinate is
either at r = or at a finite r , depending on the black hole considered. The solutions
are plane-wave normalizable, and have infinitely oscillating phases at the horizon.
The near-horizon analysis of black-hole solutions reveals, however, that the time
independent ( = 0) mode of the scalar field has to be handled carefully.
The metric for an asymptotically flat, spherically symmetric, static black hole in 4D is
of the form
ds 2 = F (r) dt 2 + F 1 (r) dr 2 + r 2 d2 gij dx i dx j .
For a ReissnerNordstrom black hole,
(3.1)
(r r+ )(r r )
,
r2
2
r = QlP + ElP 2QElP3 + E 2 lP4 .
F (r) =
295
(3.2)
(3.3)
Here, lP is the Planck length and E = M Q/ lP is the energy above extremality. For
a Schwarzschild black hole, F (r) = (1 2M/r).
Let us look at a massless scalar field in this background geometry. The action for such
a field is
Z
1 p
|g|g ij i j .
(3.4)
S =
2
If we restrict out attention to spherically symmetric configurations, the action looks like
Z
dr
1 2
dt.
(3.5)
() + F 2 (r)( 0 )2
S =
2
F (r)
This immediately allows us to identify the Lagrangian:
Z

dr 2
1
F (r)2 ( 0 )2 .
()
L=
2 F (r)
(3.6)
The modes of the scalar field are obtained from the ansatz that the time dependence of is
exp(it). We are interested in the time-independent solutions, so we take = 0. Then
the KleinGordon equation for this case is obtained simply by considering the second term
in the Lagrangian, and is

d
1 d
F
= 0,
(3.7)
H =
F dr
dr
where wave functions are defined on L2 [(0, ), r 2F dr]. It is more convenient to work
with the measure dr rather than r 2 F (r) dr,
p so we make a unitary transformation from
2
+
2
+
L [R , F (r) dr] to L [R , dr] via U = r 2 F (r) = . In this new basis, H reads:
2 00 2 0 2
(r F )
(r F )
d2
= 0.
(3.8)
H = 2 +
2F
2F
dr
On putting the value of F for the black hole in (3.8) and taking the near-horizon limit,
we find that both for the nonextremal black holes, (3.8) in the near-horizon limit is

1
d2
(3.9)
2 2 = 0,
dx
4x
where x = (r r+ ) is the near-horizon coordinate, r+ is the horizon. For the extremal
ReissnerNordstrom solution, however, (3.8) reduces near the horizon to
d2
= 0.
(3.10)
dx 2
Another situation where we see a similar equation is the near-horizon geometry of the
one-dimensional black hole discovered by Witten [15]. The metric for this black hole is of
the form
ds 2 = tanh2 (r/R) dt 2 + dr 2 .
(3.11)
296
The action for a scalar field propagating in this background is

Z p
|g|g ij i j dr dt.
S = 1/2
The Lagrangian is

Z
L = 1/2 tanh(r/R)
(3.12)

dr.
(3.13)
tanh2 (r/R)
The KleinGordon equation for the zero mode can be calculated from the 2nd term, the
functions being defined on L2 [(0, ), tanh(r/R) dr]:

d
d
1
tanh(r/R)
= 0.
(3.14)
tanh(r/R) dr
dr
Again, we can make a unitary transformation from L2 [R+ , tanh(r/R) dr] to L2 [R+ , dr]
via U = tanh(r/R) = , the equation now is

3
1
1
1/4
d2
2
(3.15)
+ tanh (r/R) = 0.
2 + 2
2
dr
R tanh2 (r/R) 4
For small r, the equation is approximately

1
1
d2
+
2
= 0.
(3.16)
dr
4r 2 2R 2
Another black hole that exhibits the same behavior is the BTZ black hole in (2 + 1)D
gravity [16]. For simplicity, we take J = 0. It has a metric given by
ds 2 = N 2 dt 2 + 1/N 2 dr 2 + r 2 d 2 ,
(3.17)
where N 2 = (r 2 / l 2 M), 1/ l 2 is the curvature of AdS space and M is the black-hole

mass. Here, again, the near horizon KleinGordon equation is
d2
2 = 0,
(3.18)
dx 2
4x
where (r l M) = x is the near-horizon coordinate. In the case of the Schwarzschild,

nonextremal and ReissnerNordstrom equations, the next-order correction is of order
1/(r r+ ).
Thus, in all these cases barring the extremal RN black hole, (3.18) is the near-horizon
equation for the zero-mode solution. The solutions, both to (3.18) and the extremal case
are discussed in the next section. The eigenvalues of the Hamiltonian which is just the l.h.s
of (3.18) and of the operator which is the l.h.s of (3.10) are obtained. The solutions of
interest are the zero eigenvalue solutions for that Hamiltonian problem. We will see that
the self-adjointness analysis of the Hamiltonian H , i.e., the l.h.s operator in (3.18) will
help us find these solutions.
4. Self-adjointness of the operator H

As is well known (see, for example, [21,22]), discussion of self-adjointness (or
hermiticity) for an unbounded operator O first requires us to define the domain D(O)
297
of O. We will only be interested in operators that are defined on domains that are dense in
the Hilbert space. This allows us to define O , the adjoint of O, and D(O ). By definition,
O is self-adjoint if and only if D(O) = D(O ). A better way of saying this is by looking at
deficiency indices, which are defined as follows. Let K = Ker(i O ), where Ker(X)
is the kernel of the operator X. The integers n dim K are the deficiency indices of the
operator. If n = 0, then O is essentially self-adjoint. If n+ = n = n 6= 0, the O is not
self-adjoint but has self-adjoint extensions. Different self-adjoint extensions of the operator
are in one-one correspondence with unitary maps from K+ to K , that is, they are labeled
by a U (n) matrix. Finally, if n+ 6= n , then O cannot be made self-adjoint.
The Hamiltonian H is a special case of a more general Hamiltonian studied extensively
in the literature. It is defined on a domain L2 [R+ , dx] and is of the form
d
+ 2.
(4.1)
2
dx
x
Classically, the system described by this Hamiltonian is scale invariant ( is a dimensionless constant). However, the quantum analysis of this operator is much more subtle. As was
shown by [23,24], H is essentially self-adjoint only for > 3/4. For > 3/4, the domain
of the Hamiltonian is

(4.2)
D0 = L2 (dx), (0) = 0 (0) = 0 .
H =
For 6 3/4, this operator is not essentially self-adjoint (and therefore cannot play the
role of a Hamiltonian) and so has to be extended to another operator. For this case,
the deficiency indices are h1, 1i, and so the self-adjoint extensions are labeled by a U (1)
parameter eiz , which labels the domains Dz of the Hamiltonian Hz . The set Dz contains all
the vectors in D0 , and vectors of the form + + eiz , where

(4.3)
+ = x 1/2H(1) xei/4 ,

1/2 (2)
i/4
,
(4.4)
= x H xe
where = 1/4 + , and H(1,2)s are the Hankel functions J iN . The small x behavior
of + + eiz is
3i/4
ix 1/2
x e
eiz+3i/4
iz
+ + e
sin() 2
0(1 + )

iz+i/4
e
ei/4
x
.
(4.5)
+
2
0(1 )
We can now solve the eigenvalue equation for bound states:
(4.6)
00 + 2 = E.
x
For > 3/4, there are no bound states. More precisely, there are no normalizable solutions
to the Schrdinger equation with negative energy. However, for 1/4 6 < 3/4 there is
exactly one bound state of energy Eb , where

sin(z/2 + 3/4) 1/
,
(4.7)
Eb = E(, z) =
sin(z/2 + /4)
298
and the corresponding eigenfunction is

p
p
p
1/2

J i Eb x ei J i Eb x .
= N Eb x
(4.8)
The existence of bound states seems to be in contradiction with scale invariance, since
scale invariance implies that there is no length scale in the problem, whereas the existence
of the bound state provides a scale. This tension can be resolved by looking at how scaling
is implemented in the quantum theory. The scaling operator is
xp + px
,
(4.9)
2
where p = id/dx. It is easily seen that 3 is symmetric on the domain D of H , and that
for > 3/4, 3 leaves invariant the domain of the Hamiltonian. For 6 3/4,

0
(4.10)
3 = x 3/2 + + eiz .
3=
The small x behavior of the function 3 is of the form

e3i/4 eiz+3i/4
ix 1/2 x i/4
1
2e
3 '
sin
2
0(1 + )

iz+i/4
i/4
e
e
x
+ ,
+
2
0(1 )
(4.11)
where z = z + /2. So 3 clearly does not leave the domain of the Hamiltonian invariant.
Scale invariance is thus anomalously broken, and this breaking occurs precisely when
the Hamiltonian admits nontrivial self-adjoint extensions. This also explains the quantum
mechanical emergence of a length scale, namely the bound state energy.
We must remark here that there do exist self-adjoint extensions that preserve scale
invariance. For example, if z = (/2), then there is no bound state. From the point
of view of the domains, the operator 3 leaves this domain invariant, implying that scaling
can be consistently implemented in the quantum theory.
Now that we know about the subtleties about quantum mechanical evolution in 1/x 2
potential, we can apply these ideas to our case. The potential near the horizon is like
1/4x 2 for the problem of interest.
For the 1/4x 2 potential, there are infinite number of bound states for a given fixed
self-adjoint extension z. These are given by
p
(4.12)
En (x) = Nn xK0 En x , n Z,
h
zi
(4.13)
En = exp (1 8n) cot , n Z.
2
2
These are found by solving (4.6) for = 1/4 and carefully comparing the behavior of
the eigenfunctions with the analog of (4.5) which is

(4.14)
= eiz/2 x 1/2 + ix 1/2 ln x + eiz/2 x 1/2 ix 1/2 ln x .
Returning to the original problem of finding the zero mode solutions, i.e., the solutions
to (3.18), we see that demanding self-adjointness of the Hamiltonian gives rise to an
infinite number of bound states labeled by an integer n. The zero mode solution is obtained
299
Fig. 1. Absolute value of zero mode solution for a Schwarzschild black hole with horizon radius
r+ = 50.
from (4.13) in the n limit. In particular, the wave function for the solution to (3.18)
near the horizon is
p

(4.15)
= Nn x 1/2 1 + ln En x ,
where En is given by (4.13) and Nn is an appropriate normalization factor. Then one takes
the limit n . This leads to a solution that is nonzero only at the horizon, where it
peaks, and can be thought of as a horizon state. En depends on the self-adjointness
parameter z, which also corresponds to the boundary condition at the horizon. However, in
the limit n , all boundary conditions lead to the same solution of (3.18). Since (3.18)
is the time-independent zero angular momentum mode for the scalar field in all the
aforementioned black-hole backgrounds, the above discussion applies to all those cases.
The behavior of the zero mode found by this method matches that of the numerical zero
mode solution for the Schwarzschild black hole in Fig. 1 (where the horizon is at r = 50)
apart from minor errors in the numerical interpolation.
For the one exception, the extremal ReissnerNordstrom black hole, the equation (3.10)
is easily solved. The corresponding Hamiltonian problem for which the solutions to (3.10)
are the zero eigenvalue solutions was considered in the section above. However, it does
not lead to the kind of nontrivial boundary conditions for the zero eigenvalue solution as
in the other cases. This is because the self-adjointness analysis of that operator yields only
one bound state. The bound state vanishes for a particular value of the self-adjointness
parameter, as discussed. Therefore, there seems to be no nontrivial zero mode for the
extremal black hole.
5. Time-independent modes in the plane AdS black hole

Another black-hole solution which can be obtained in the infinite mass limit from the
AdSSchwarzschild solution, the plane AdS solution, was discussed in [5]. The metric for
the euclidean AdSSchwarzschild black hole in the infinite mass limit is of the form
ds 2 = F (r) d 2 + F 1 (r) dr 2 + r 2
3
X
i=1
dxi2 ,
where
(5.1)
300

F (r) = r 2 /b2 b2 /r 2 .
(5.2)
Let us look at a massless scalar field in this background geometry. One can in principle
consider a complex scalar field with charge q and mass m, but for simplicity we shall
consider only the massless and uncharged field in the black-hole background. The action
for such a field 8 is
Z
1 p
|g|g ij (i 8)(j 8) d5x,
S=
2
2

Z Z
Z
X
8
1
2
+ F (8)0 + 1/r 2
dr d
d3 x r 3
(xi 8)2 .
(5.3)
=
2
F
b
This action, where the scalar field is the type IIB dilaton field, has been discussed in [5
7]. Modes for the field which are independent are considered, where 8(r, x) = f (r)
exp(ik x). Then the equation of motion for f (r) is

(5.4)
r 1 d/dr r 3 r 2 1/r 2 (df/dr) + k 2 f = 0,
where b = 1 is taken for simplicity. On demanding normalizability of f (r) w.r.t. the
measure r 3 dr and regularity of the solution at r = 1, a discrete negative spectrum for
k 2 was obtained. It was identified with the glueball spectrum in the boundary theory.
We show below that one can consider (5.4) as an eigenvalue problem for k 2 and examine
the operator in this equation for self-adjointness. As is well known, a self-adjoint operator
has only real eigenvalues, and any wave function in the domain of the operator can be
written in terms of its eigenfunctions. We therefore wish to find the complete set of k
modes such that any function of compact support in the domain can be expanded in terms
of the mode functions. It is seen that the operator is not self-adjoint, but can be extended to
a self-adjoint operator. However, more general boundary conditions are required at r = 1.
Then, the spectrum of k is also enlarged to include a discrete infinity of positive k 2 states,
and some negative k 2 states as well.
The operator of interest is

(5.5)
T = r 1 d/dr r 3 r 2 1/r 2 d/dr ,
where wave functions are defined on a measure r 3 dr.
We can therefore check the operator T for self-adjointness. We first check if it is
symmetric, i.e. if (, T ) = (T , ), where D(T ), and D(T ).
If the operator T is symmetric, it is self-adjoint if (T i) = 0 has no solutions in
D(T ).
But with this measure, we see that the operator is not even symmetric. We therefore
consider the measure r dr which from the action (5.3) is the natural measure to consider
if one is interested in looking for the eigenvalue problem for the operator (5.5). However,
this measure is not enough to guarantee finiteness of the second term in (5.3). Therefore,
we take the domain of functions D(T ) to consist of C , square integrable functions with
respect to the measure r dr which fall off at least as 1/r 3 (or faster than that) that are
301
of compact support. (Actually, it is enough if they fall of as 1/r 2+ where > 0. For
convenience, we take = 1, and it does not affect any of the analysis.)
The self-adjointness question is easier to address after a change in coordinates,
following [7]. On making the transformations
r 2 = cosh x,
p
A(x) = sinh(2x) f (x),
(5.6)
(5.7)
the measure becomes dx/ cosh x, and (5.4) becomes

4 cosh x d2 /dx 2 A(x) + 4 cosh xA(x) 4 cosh xA(x)/ sinh(2x)2 = k 2 A(x).
(5.8)
In these coordinates, the horizon is at x = 0. Here, one can define the domain of interest
D(T ) to consist of C , square integrable functions A(x) with respect to the measure
dx/ cosh x and which fall off asymptotically at least as A(x) exp(3x/2). Also, they
are of compact support, so A(x = 0) = A0 (x = 0) = 0. Then it can be shown that the
operator on the l.h.s of (5.8) is symmetric, however, the domain of the adjoint T is now
any normalizable function. Thus, D(T ) 6= D(T ). The operator is not self-adjoint. Also,
(T + i) = 0 and (T i) = 0 each have one normalizable solution, as can be verified
numerically (see Fig. 2). If for each eigenvalue i, there is exactly one normalizable
solution, then the deficiency indices of this operator are (1, 1) and it is possible to find
self-adjoint extensions for it. Therefore, one can look for the self-adjoint extension of this
operator. Since a self-adjoint extension involves only a change of boundary condition at
x = 0, we deal with the near-horizon form of (5.8) for simplicity.
On using the near-horizon (x small) approximation, (5.8) becomes

(k 2 + 1)A(x)
A(x)
.
=
d2 /dx 2 A(x)
2
4x
4
(5.9)
This looks like a Hamiltonian problem for a potential 1/(4x 2) (which was discussed
extensively in the previous section) with the eigenvalue (k 2 + 1)/4.
The results of the previous section can be applied to the case of (5.9) to find the
additional states that arise due to the changed boundary condition. They are given by (4.13)
with En = (k 2 + 1)/4. Thus, there are eigenvalues k 2 for each n, and n is any integer. The
eigenvalues also depend on the self-adjoint parameter z. There are positive k 2 eigenvalues.
There is a possibility of finding some values of k 2 with k 2 negative too, for which k 2 < 1.
Fig. 2. Absolute value of solution for k 2 = i as a function of r.
302
What has been done above is a near-horizon analysis of (5.8). It is not clear if all of
these states are solutions to the complete equation (5.8). However, numerically, there seem
to exist normalizable solutions to the complete equation for any positive k 2 , provided one
also accepts the nonregular solutions that have not been considered by [57]. These are
seen to be irregular only at the horizon, exactly like the solution in Fig. 2. Imposing
a particular boundary condition at the horizon demanded by self-adjointness picks out
a discrete infinity of normalizable positive k 2 states as above.
A feature of these modes that is immediately noticeable is that they are irregular at the
black hole horizon. However, from considerations of self-adjointness, they are necessary
for expressing any arbitrary, regular field configuration in the bulk in terms of a complete
set of mode solutions. In fact, the difference of any two of these irregular solutions is
regular. This is because the irregular solutions are irregular only at the horizon, where they
behave as fk (r) ln(k(r 1)) where r = 1 is the horizon. Taking the difference of two
solutions fk1 (r) and fk2 (r), we see that the resultant solution is regular at the horizon.
Therefore, any arbitrary regular field configuration in the bulk can be constructed with
regular mode solutions and an even number of irregular mode solutions.
It may seem that the irregular solutions can be gotten rid of by shifting the domain of
interest a small distance away from the horizon, where > 0 and repeating the selfadjointness analysis for this new domain. However, letting 0, the irregular solutions
reappear. Further, the one parameter ambiguity in boundary conditions is not resolved.
Letting 0 does not pick any particular boundary condition at the horizon [23].
6. Discussion
We find that on examining scalar field theory in the background of the infinite mass
limit of the AdSSchwarzschild black hole, there are more time-independent, equilibrium
modes than previously obtained [6,7]. These are however positive k 2 modes. There is a
parameter labeling the boundary conditions at the horizon (for the self-adjoint extension)
on which these modes depend.
We analyzed the time-independent, L = 0 solutions of the (3 + 1)D Schwarzschild and
ReissnerNordstrom black holes, the (1 + 1)D dilatonic black hole and the BTZ black
hole. There are several features in these backgrounds that are similar to the case of the
plane AdSSchwarzschild black hole. In particular, there is again a one parameter family of
boundary conditions labeled by the self-adjoint parameter z as before. However, now they
lead to the same solution. The solution is a horizon state, i.e. it is localized at the horizon.
There seems to be no such nontrivial zero mode for the extremal ReissnerNordstrom black
hole.
Lastly, we would like to speculate on the possible interpretation of these irregular modes
in the boundary theory. As first observed by [5], the modes with negative k 2 correspond to
glueballs with mass k 2 . This correspondence, when applied to the irregular states, seem to
imply the existence of tachyonic glueball states. Actually, such a scenario is not as exotic
as it may appear to be at first sight. It was pointed out a long time ago by Savvidy [11], and
303
also by Nielsen and collaborators [1214] that the perturbative vacuum of QCD is unstable.
Considering a translation invariant background for SU(2) gauge fields, they obtained the
effective one-loop potential. This has the structure of a double well potential along with
an imaginary term signaling the onset of instability. This persists in SU(N) theories and
at finite temperature [25]. Our scenario resembles this phenomenon, which seems to be
indicated by the appearance of these modes.
Acknowledgements
We would like to thank A.P. Balachandran, S. Kalyana Rama, R. Parthasarathy,
B. Sathiapalan and A. Sen for useful discussions.
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]
S. Chandrasekhar, The Mathematical Theory of Black Holes, Clarendon Press, Oxford, 1983.
R. Wald, J. Math. Phys. 21 (1980) 28022805.
G.T. Horowitz, D. Marolf, Phys. Rev. D 52 (1995) 56705675; gr-qc/9504028.
S. Hawking, D. Page, Commun. Math. Phys. 87 (1983) 577.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 505532; hep-th/9803131.
C. Cski, H. Ooguri, Y. Oz, J. Terning, JHEP 9901 (1999) 017; hep-th/9806021.
R. Koch, A. Jevicki, M. Mihailescu, J.P. Nunes, Phys. Rev. D 58 (1998) 105009; hepth/9806125.
R.C. Brower, S.D. Mathur, C. Tan, hep-th/9908196.
S. Kalyana Rama, B. Sathiapalan, hep-th/9905219.
G.T. Horowitz, V. Hubeny, hep-th/9909056.
G.K. Savvidy, Phys. Lett. B 71 (1977) 133134.
H.B. Nielsen, M. Ninomiya, Nucl. Phys. B 156 (1979) 128.
N.K. Nielsen, P. Olesen, Nucl. Phys. B 144 (1978) 376396.
J. Ambjrn, N.K. Nielsen, P. Olesen, Nucl. Phys. B 152 (1979) 7596.
E. Witten, Phys. Rev. D 44 (1991) 314324.
M. Banados, C. Teitelboim, J. Zanelli, Phys. Rev. Lett. 69 (1992) 1849.
T. Jacobson, Phys. Rev. D 57 (1998) 48904898; hep-th/9705017.
K.D. Kokkotas, B. Schmidt, Living Reviews in Relativity, Vol. 2, 1999; www.livingreviews.org/
Articles/Volume2/1999-2kokkotas.
H. Nollert, Class. Quantum Grav. 16 (1999) R159R216.
S. Chandrasekhar, Proc. Roy. Soc. London Ser. A 392 (1984) 113.
M. Reed, B. Simon, Functional Analysis, Methods of Modern Mathematical Physics, Vol. I,
Academic Press, New York, 1972.
M. Reed, B. Simon, Fourier Analysis, Self-Adjointness, Methods of Modern Mathematical
Physics, Vol. II, Academic Press, New York, 1975.
K. Meetz, Nuovo Cimento 34 (1964) 690.
H. Narnhofer, Acta Phys. Austriaca 40 (1974) 306322.
D. Kay, Phys. Rev. D 29 (1984) 24092411.

Axial anomaly effects in pion and Z 0 radiative

decays
Luca Trentadue 1 , Michela Verbeni
Dipartimento di Fisica, Universit di Parma, INFN Gruppo Collegato di Parma, 43100 Parma, Italy
Abstract
We discuss a connection between axial anomaly and polarized radiative processes. By comparison
with the corresponding unpolarized cases, we consider some physical outputs for the + and Z 0
polarized radiative decays. We analyse in detail the pattern of mass singularity cancellation. 2000
PACS: 11.30.Rd; 13.20.Cz; 11.15.Bt
1. Introduction
Undoubtedly the axial anomaly represents a fundamental issue for understanding the
basic aspects of quantum field theory. This issue has been analysed deeply over the years.
The anomaly problem has been treated by means of renormalization procedure, giving
the interpretation of its origin in terms of ultraviolet divergences [1,2]. A more formal
analysis of the axial anomaly can be made by using the path integral formalism [3].
Dolgov and Zakharov [4] have shown an alternative approach to the axial anomaly,
by studying the VVA triangular diagram through dispersion relations. From this approach
follows the interpretation of the axial anomaly as an infrared phenomenon. It appears as due
to a singularity present in the chiral limit in the absorbitive part of the triangular diagram.
The infrared aspect of the axial anomaly, raised in this paper, is complementary to
the more familiar ultraviolet one, which emerges from the renormalization procedure.
A particularly interesting feature of this approach is that it allows to shed light upon the
physical meaning of the anomalous chiral symmetry breaking, which is connected to a non
conservation of helicity.
The connection between the anomaly and the breaking of a given symmetry has received
a lot of attention in the literature and this subject has been discussed and developed in
1 trentad@fis.unipr.it
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 1 9 - 9
308
L. Trentadue, M. Verbeni / Nuclear Physics B 583 (2000) 307328
several papers. Gribov [5], in a seminal work, has described the source of the anomalies
as a collective motion of particles with arbitrarily large momenta in the vacuum. Related
to this work, in Ref. [6], Mueller has discussed the manifestation of the axial anomaly
as a flow of Landau levels. In the papers of Refs. [79] the origin of the axial anomaly
has been studied in two dimensions and, again, as a level crossing phenomenon. The
infrared interpretation of the axial anomaly, according to the Dolgov and Zakharov
approach, has been also considered in [11]. The leading terms in the chiral limit have been
correctly evaluated in [10,12,13]. Furthermore the dispersive analysis of the triangular VVA
diagram is fundamental in the formulation of the t Hooft consistency condition [14]. The
axial anomaly plays an essential role also in the interpretation of spin dependent parton
distribution (see [1518]).
In this work we attempt to relate the Dolgov and Zakharov approach to the axial anomaly
to some effects in the dynamics of physical reactions, as the radiative decays of + and Z 0 .
We will try to show that the axial anomaly can be related to polarized radiative decays, as
in the usual ultraviolet interpretation it is connected to the 0 decay. We calculate
the corresponding decay rates for the cases where the outgoing leptons are in a definite
helicity state and we examine in some detail the structure of the mass singularities and
their cancellation. We study how the KinoshitaLeeNauenberg (KLN) theorem [19,20]
applies and we consider the analogies and the differences with respect to the corresponding
unpolarized decay rates.
The paper is organized as follows. In Section 2 we briefly reconsider the dispersive
approach to the axial anomaly. In particular we concentrate on the physical origin of the
axial anomaly.
In Section 3 we extend the Dolgov and Zakharov approach to the study of the radiative
pion decay. We calculate the differential decay rate for the process, where the outgoing
lepton undergoes an helicity flip and we interpret its behaviour in the chiral limit, as a
manifestation of the axial anomaly.
In Section 4 we study the behaviour of the Inner Bremsstrahlung contribution to the pion
decay rate in the collinear and infrared limits. We consider separately the unpolarized decay
rate and both the cases of right-handed and of left-handed outgoing lepton. We find that the
mass singularities cancellation mechanism occurs in different ways, according to the polarization of the outgoing lepton. We discuss the various realizations of the KLN theorem.
We also consider a more general process, i.e., the radiative Z 0 decay in a lepton
antilepton pair, with a right-handed polarized lepton. We examine how the mass
singularities cancel in this case and discuss the differences with respect to the pion decay.
Finally, in the conclusions, we summarize our arguments.
A short discussion of the main results obtained has been given in [21].
2. The dispersive approach to axial anomaly

The Dolgov and Zakharov approach [4] to derive the axial anomaly is based on the
dispersion relation method. In this framework, the triangular diagram with two vector and
309
Fig. 1. Lowest order contribution to the process axial source 2 photons.
one axial vertices is seen as the lowest order contribution to the process:
axial-vector source 2 real photons,
as described by the diagrams of Fig. 1. In the physical region, the triangular diagram
5
represents the
possesses a branch cut along the real axis, from 4m2 to infinity. T
corresponding amplitude. We express it by means of a dispersion relation in the variable
s = q 2 = (k1 + k2 )2 , where k1 and k2 are the outgoing photon momenta.
5
satisfies the vector
By requiring parity, Lorentz invariance, Bose symmetry and that T
WardTakahashi identity,
5
5
(k1 , k2 ) = k1 T
(k1 , k2 ) = 0,
k1 T
we can write it in terms of an invariant scalar function g1 (q 2 ) as:

2
5
g1 (q 2 ) k1 k2 q .
(k1 , k2 ) =
(1)
T
Similarly, we can express the contribution of the triangular diagram with the vertex 5
substituted by 5 as:
m
5
g2 (q 2 ) k1 k2 ,
(k1 , k2 ) =
(2)
T
where g2 (q 2 ) is another invariant scalar function.

In terms of dispersion relations, the functions gi (q 2 ), i = 1, 2, can be expressed as
follows:
Z+
1
Im gi (s)
2
ds
, i = 1, 2.
(3)
gi (q ) =
s q2
4m2
We can use unsubtracted dispersion relations, because the integrals contained in gi (q 2 ),

i = 1, 2, are convergent, since these functions are multiplied by three and two powers of
momentum, respectively.
310
Fig. 2. Cut diagram.
The imaginary part of the invariant scalar functions can be derived from the absorbitive
part of the triangular diagram, calculated by cutting the diagram as shown in Fig. 2 and by
using the Cutkosky rules or the perturbative unitarity relation. We obtain [4]:
p

1 + 1 4m2 /q 2
2
2
2 1
p
,
(4)
Im g2 (q ) = 2 (q 4m ) 2 ln
q
1 1 4m2 /q 2
p

2m2
1 + 1 4m2 /q 2
2
2
2
p
.
(5)
Im g1 (q ) = 4 (q 4m ) ln
q
1 1 4m2 /q 2
Using Eqs. (1)(3) and the above expressions, we derive the complete triangular
diagram contribution:
5
(k1 , k2 ) =
T
2m
2 1
5
T
k1 k2 q
(k1 , k2 )q +
+ i
q 2 + i
q2
(6)
and the anomalous axial-vector WardTakahashi identity:

5
(k1 , k2 ) =
q T
2m 5
2
k1 k2 .
T (k1 , k2 ) +
q 2
(7)
Thus, in the dispersive approach we find the same result obtained by using the
renormalization procedure [1,2].
The method above allows a more direct treatment, since we avoid evaluating divergent
integrals and introducing regularization schemes.
The fact that the axial anomaly can be derived without using the renormalization
procedure, suggests that this should not be considered as the only origin of the anomalous
breaking of the chiral symmetry. Moreover, by studying the axial anomaly with the
renormalization procedure, that is by considering its ultraviolet interpretation, an important
aspect of this phenomenon remains obscure and we are bound by a formal derivation only.
311
Fig. 3. Process (b).
As stressed in [4] and [22], the dispersive approach shows that the anomaly is related to
the chiral limit and therefore, it can be interpreted as an infrared effect. In this work we are
close to this infrared interpretation, which as we will see, can help to shed light on some
aspects of the physics connected to the axial anomaly.
Let us briefly discuss the physics involved in the amplitudes contributing to the
absorbitive part of the triangular diagram. The two Born diagrams, obtained from the cut
of the triangular diagram (see Fig. 2), describe the following two processes:
(a) Production of a fermionantifermion pair (for example, e+ e ) by an axial-vector
source;
(b) Subsequent annihilation of the pair into two real photons.
In both these processes there occur helicity flips, thus the chirality is not conserved in
the zero mass limit.
Let us go to the center of mass frame of the two final photons. In the first process the
axial-vector source produces an e+ e pair of total spin zero, since a spin 1 state cannot
annihilate into a two real photons state. Thus e+ and e must have the same helicity and
hence opposite chirality in the massless limit. 2 In the process (b) the e+ e pair annihilates
into two real photons by going through an intermediate virtual state. There are four possible
virtual states [23]: one is drawn in Fig. 3 (p1 and p2 are the e and e+ linear momenta),
a second one is obtained by reversing the virtual state helicity; the remaining two are
obtained replacing the outgoing virtual fermion with an incoming virtual antifermion. Let
us study the case shown in Fig. 3 and assume that e+ and e are both right-handed. In
the vertex B the chirality is conserved in the massless limit, while in the vertex A there is
an helicity flip, thus the chiral symmetry is broken. As can be easily checked, for all the
remaining virtual states we always have an allowed vertex and a forbidden one.
At the Born level, these reactions are described by the classical QED Lagrangian, which,
in the m 0 limit, is invariant under chiral transformations. Thus it seems at first sight
2 Equivalently, considering the outgoing antifermion as an incoming fermion, one can say that the latter makes
an helicity flip interacting with the axial-vector source.
312
that the absorbitive part of the triangular diagram, being proportional to the product of
amplitudes relative to processes forbidden by chiral invariance, vanishes. On the contrary,
one sees that taking the limit m 0 in (5) gives a finite result [4]:
Im g1 (q 2 ) (q 2 )
as m 0.
(8)
Therefore, by studying the absorbitive part of the triangular diagram, one establishes that
a non-conservation of helicity occurs, becoming, in the massless limit, a non-conservation
of chirality. We interpret this as related to the presence of the anomalous term in the
divergence of the axial-vector current. Thus the axial anomaly can be derived by studying
the properties of the amplitude in the infrared region.
Even if in this work we will analyse the cancellation of mass singularities in polarized
processes, we will not discuss the physical implication of the zero fermion mass limit.
As stated in Refs. [4,10,22], the result in Eq. (8) indicates that there occurs a cancellation
of the suppression factor m2 , due to the terms coming from the vertices with helicity flip. In
(5) the logarithmic factor conspires to give a finite result. This logarithm is a collinear one;
we shall discuss about this kind of logarithms in Section 4. Its presence is a manifestation
of the singularity occurring in the fermion propagator as m 0, which exactly cancels the
suppression factor m2 in the numerator.
We observe that the behaviour of the absorbitive part of the triangular diagram given in
Eq. (8) shows that the m 0 limit is not smooth; if we evaluate the amplitudes with the
massless theory, they identically vanish. If, on the contrary, we take the chiral limit after
summing over the intermediate states, we obtain a result different from zero.
3. Helicity changing processes and polarized pion decay

3.1. Helicity changing processes
In the process (b) contributing to the absorbitive part of the triangular diagram a charged
polarized fermion changes helicity by emitting a photon. Thanks to the collinear singularity
within the propagator relative to this fermion, the chiral suppression factor m2 is cancelled
and the absorbitive part of the triangular diagram doesnt vanish in the chiral limit.
Let us now extend these remarks to physical processes having the same features as the
ones characterizing the absorbitive part of the triangular diagram. This means that we want
to investigate about a manifestation of the axial anomaly in reactions where a fermion
changes helicity by emitting a photon. We call these processes helicity changing processes.
They have been considered by Lee and Nauenberg [19], who observed that states with
opposite helicity dont decouple in the massless limit, provided that this limit is taken after
having summed the transition probability over the final phase space. In the process (b) the
incoming e+ and e are in a definite helicity state, as discussed above. In the reactions we
want to study, we calculate the probability that an outgoing particle assumes an helicity
opposite to the one required by the interaction before the emission of the photon. In other
words, we evaluate the probability that in the fermionphoton vertex occurs an helicity
flip. According to the Dolgov and Zakharov analysis, we interpret the presence of a term
313
independent of the fermion mass in the corresponding cross sections as a manifestation of

the axial anomaly. Due to this term, the probability for a process with helicity flip doesnt
vanish in the chiral limit.
3.2. Polarized radiative pion decay
We first examine the non radiative pion decay
+ l + + l ,
where l + is an antilepton (e+ or + ) and l is the associated neutrino. At the Born level
the total decay rate is given by:
G2 f2
2
m2
|Vud |2 3l m2 m2l ,
0 + l + l =
8
m
(9)
where G is the Fermi coupling constant, f is the pion decay constant, Vud is the CKM
matrix element, ml and m are the lepton and pion masses, respectively. The decay rate
(9) is proportional to m2l , since, due to angular momentum conservation, the pion produces
a left-handed lepton, while the structure of the weak coupling requires the l + to be righthanded for ml = 0.
This situation is confirmed by the expression of the total decay rate for the process in
which the antilepton is polarized:

2
G2 f2

pl s l
pol
+
+
2 ml
2
2 2
|Vud | 3 m ml
,
1
0 l l =
16
m
|pl |
(10)
where p l and s l are the linear momentum and the spin vector of the lepton, respectively,
giving p l s l /|pl | = 1, for right-handed and left-handed lepton, respectively. The Eq. (10)
indicates that the lepton is mandatory left-handed, as required by angular momentum
conservation.
We now consider the radiative correction, due to the emission of a real photon, to the
pion decay,
+ l + + l + ,
when the outgoing antilepton is polarized, i.e., when it is in a definite helicity state. As
we have seen in the non-radiative case, due to angular momentum conservation, the + is
coupled to a left-handed lepton. We calculate the probability that the lepton flips its helicity
and becomes right-handed, by emitting a real photon.
The amplitude describing the radiative pion decay can be divided into two parts, the
Inner Bremsstrahlung and the Structure-Dependent amplitudes [24]:

(11)
M + l + l = MIB + MSD .
The Inner Bremsstrahlung amplitude, where the photon is radiated from the external
charged particles, can be calculated using the rules of QED, with a point like pion coupling;
the Structure-Dependent amplitude is governed by the strong interactions.
314
Fig. 4. Inner Bremsstrahlung diagrams.
Clearly, the relevant part for the problem we are considering is the Inner Bremsstrahlung
contribution described by the diagrams of Fig. 4. The term associated to the IB3 diagram
is the so called contact term and it is introduced to ensure gauge invariance (see, for
example, [24]).
We consider only tree level diagrams, since, to reveal the effects of the axial anomaly,
it is sufficient to take into account the contribution of the vertex with the helicity flip. For
the moment, we may neglect the corrections due to the emission of virtual photons; these
will be discussed in Section 4. Finally, at this order in perturbation theory, we retain terms
of all powers in the lepton mass. We will argue that the manifestation of the axial anomaly
is strictly connected to these terms.
The Inner Bremsstrahlung amplitude is given by [24]:
MIB =
3
X
IBi
i=1

G
p k// + 2pl
v(pl , slR ),
= ie f Vud ml u(p
)(1 + 5 )
pk
2pl k
2
(12)
where u and v are the Dirac spinors for the neutrino and the lepton, respectively, p is the
pion momentum, pl and slR are the momentum and the polarization vector of the lepton, k
and are the momentum and the polarization vector of the photon.
We see that MIB is proportional to the lepton mass ml , thus the decay rate is proportional
to m2l . As we have said above, this factor is a consequence of the structure of the weak
coupling. Thus, if we remove it by normalizing the radiative decay rate with respect to
the non radiative one, we can emphasize the mass dependence of the radiation emission
process.
The differential Inner Bremsstrahlung contribution for right-handed lepton is:

R

1
1
1 dIB
=
2A 1 + y 2 2A + r(2A + r 6)
2
0 dy
4 (1 r) A(1 y + r)
y +A

+ (A + 2r)(1 + r)2 + Ay(y 4r) + 2ry(1 + y) y(1 + y 2 + 5r 2 ) ln
y A
+ (1 y + r)2 (y 2r A) ln

y+A2
.
yA2
315
(13)
The dimensionless variable y is defined as

y=
2El
m
r=
m2l
,
m2
and
A=
q
y 2 4r.
The physical region for y is:
2 r 6 y 6 1 + r.
(14)
We see that, in Eq. (13), there is a term independent of the lepton mass, the one in the
first square brackets. Owing to this term, the differential decay rate doesnt vanish in the
limit ml 0. We have:
R
1 dIB
(1 y) as r 0.
0 dy
2
(15)
This indicates that there occurs an helicity flip and thus a chirality non conservation in
the limit of zero lepton mass. According to the interpretation given above, this term can be
interpreted as connected to the axial anomaly. It corresponds to the anomalous term present
in the divergence of the axial current.
Since the polarized radiative process with the right-handed l + is forbidden, in the limit
ml 0, by the chiral invariance of the massless QED Lagrangian, the appearance of a
term different from zero, in this limit, indicates the action of a cancellation mechanism,
analogous to the one acting in the absorbitive part of the triangular diagram.
To see how this mechanism acts, let us examine the decay rate differential with respect
to the lepton energy and to the emission angle:

R
1
1
4r
1 dIB
=
(1 + y 2 y A)
0 dy d cos
4 (1 r)2 (1 y + r) (y A cos )2
4r 2
(A + r y 2) + y(1 + r) A(1 r) 4r
(y A cos )2

+ (A + 2r)(1 + r)2 + Ay(y 4r) + 2ry(1 + y) y(1 + y 2 + 5r 2 )

1
1
2
+ (1 y + r) (y A 2r)
.
(y A cos )
(y A cos 2)
+
(16)
The first term is proportional to the lepton propagator squared. Thanks to this term, when
we carry out the final integration over the emission angle, we obtain, besides logarithmic
collinear divergences, also power collinear divergences. These power terms are essential
for the convergent behaviour of the distribution. The cancellation of the chiral suppression
would not take place were these terms absent. The term related to the axial anomaly (see
Eq. (15)) originates from this cancellation. To obtain the differential decay rate, we have
to evaluate integrals of the form:
316
Z+1
2
ml d cos
1
1
E 2 m2l cos
2 = q
m2l
q
2 E 2 m2l
E 2 m2l
m2l
(17)
where the propagator has a power mass singularity that exactly cancels the factor m2l ,
coming from the leptonphoton vertex.
The integration of the terms in equation (16) containing the lepton propagator gives rise
to the collinear logarithms:
q
E + E 2 m2l
ml
q
(18)
' ln .
ln
E
E E 2 m2
l
In the differential decay rate (13) there are also terms proportional to the logarithm
q
E + E 2 m2l m
q
.
ln
E E 2 m2l m
(19)
This one is another collinear logarithm; it diverges in the limit m 0 and ml 0 and
corresponds to the possibility that the photon is emitted parallel to the pion.
Finally, we point out that the chiral limit (ml 0) is not smooth. In fact we get
different results depending upon whether we describe an helicity changing process using
the massless theory or we take the ml 0 limit, after carrying out the integration over
the final phase space. The radiative pion decay is not a good process to see this, since,
owing to the angular momentum conservation in the pion vertex, this process cannot take
place within a massless theory, given the structure of the V A coupling of the electroweak
theory. To see that the chiral limit is not a smooth one, we consider the scattering of a
polarized electron with a proton (treated as a point like particle) of initial and final momenta
q and q 0 respectively, accompanied by the emission of a real photon. We consider a lefthanded incoming electron with momentum p and spin sL : we calculate the probability
that the electron makes an helicity flip, emitting a real photon with momentum k and
polarization and thus becoming right-handed electron with momentum p0 and spin sR :
e(p, sL ) + p(q) e(p0 , sR ) + p(q 0 ) + (k).
We study this process with the massless QED. The left-handed and right-handed spinors
are given respectively by:
1 5
1 + 5
u(p),
u(p0 , sR ) =
u(p0 ).
2
2
The corresponding transition amplitude identically vanishes:
u(p, sL ) =
e3
[(p0 + k)2 + i](l 2 + i)
1 5
1 5
/(/
p 0 + k/)
u(p)u(q
0 ) u(q) = 0,
u(p
0)
2
2
since it contains the product of different chirality projectors.
M(el p eR p ) =
(20)
317
We now calculate the cross sections for processes with helicity flip using the massive
QED and then we take the massless limit after having summed the transition probability
over the final phase space. An example can be found in [11], where the cross section for
the process
0
, 0 ) + (k, ) + B(qi )
e (p , ) + A(p) e (p
(21)
is calculated. Here A is the target (for example another fermion), is a bremsstrahlung

photon, assumed almost collinear with respect to the direction of the incident electron and
B is a set of particle produced in the reaction.
The helicity flip cross section is given by:

dhf
= 0 s(1 x)
x,
(22)
dx
2
where x = k0 /E, E is the energy of the incoming electron and 0 is the cross section for
the Born process
0
, 0 ) + B(qi ).
e (p k, ) + A(p) e (p
(23)
We see that the expression (22) does not vanish in the massless limit.
The result in Eq. (22) coincides with the one in Eq. (15) at leading order. Indeed, in [11]
the subdominant terms are not accounted for. As it will become apparent in the following
section, these terms are essential for the cancellation of mass singularities.
In Ref. [25], the authors give a different interpretation of the helicity changing processes
and hence come to a different conclusion about the smoothness of the zero mass limit.
4. Mass singularities
As it is well know, there are two types of divergences occurring in a theory when the
mass of a particle goes to zero, which will be comprehensively call in the following mass
singularities.
The first type of divergences appears when we reach the phase space region, where
the momentum of the massless particles vanishes: these are called infrared divergences.
They occur for example in QED when the energy of the photon goes to zero. The Block
Nordsieck theorem [26] assures that the infrared divergences cancel out in any inclusive
cross section. The other type of mass singularities occurs in theories with massless coupled
particles, like in QED when the photon couples to a fermion, in the limit of zero fermion
mass. The origin is purely kinematical: when two massless particles, say with momenta k
and k 0 , move parallel to each other, they have combined invariant mass equal to zero:
q 2 = (k + k 0 )2 = 2EE 0 (1 cos ) 0
k0
as 0,
(24)
even though neither k nor are soft. These divergences are called collinear singularities.
If we keep the fermion mass finite and integrate over the photon emission angle, the
collinear divergence doesnt occur, but the possibility of a divergence in the limit m 0
results in the presence of the collinear logarithms, that is logarithms of the form ln (E/m),
diverging for m 0.
318
In the case of collinear singularities, the theorem by Kinoshita, Lee and Nauenberg
[19,20] guarantees that these divergences cancel out if we sum the transition probability
over the set of degenerate states, order by order in perturbation theory. This cancellation
mechanism is analogous to that of the infrared divergences, as stated by the Block
Nordsieck theorem. Both types of mass singularities arise because the states of a theory
with massless particles are highly degenerate. The infrared divergences can be interpreted
as a consequences of the fact that a state with a single charged particle is degenerate with
a state made of the same particle plus a number of soft photons; this correspond to the
impossibility of distinguishing experimentally a charged particle from one accompanied
by soft photons, owing to the finite resolution of the measurement apparatus. The situation
of the collinear singularities is analogous: the state with a massless charged particle is
degenerate with the states containing the same particle and a number of collinear photons.
This corresponds to the fact that, as a consequence of the finite angular resolution, we
cannot establish if a massless charged particle is accompanied by collinear photons.
Let us now discuss the structure of mass singularities and their cancellation in the decay
rates for the radiative pion and Z 0 decays and how the KLN theorem applies to this case.
4.1. The pion case
It is useful to separate the cases of unpolarized, right-handed and left-handed outgoing
lepton.
Let us consider first the mass singularity cancellation mechanism for the familiar
case of unpolarized radiative + decay to the first order in . The differential Inner
Bremsstrahlung contribution is given by:

y+A
1
1
1 dIB
=
4A(r 1) + (1 + r)2 + y(y 4r) ln
2
0 dy
4 (1 r) (1 y + r)
yA

y
+
A
2
.
(25)
(1 y + r)2 ln
y A2
One can easily see that Eq. (25) is divergent both in the collinear and in the infrared
limits. The coefficients of the collinear logarithms dont go to zero in the limit r 0.
There are also infrared divergences, because if we let y reach its kinematical limit y MAX =
1 + r, corresponding to the photon energy going to zero, the expression (25) diverges.
The decay rate is made free from mass singularities in the ordinary way: the divergences
cancellation occurs in the total inclusive decay rate, when we add all the first order
contributions to the perturbative expansion, i.e., those relative to real and virtual photon
emission.
The diagrams describing the real photon emission contribution were already given in
Fig. 4; the diagram for the virtual correction is drawn in Fig. 5. The expression for the
lepton energy spectrum, including the Inner Bremsstrahlung contribution and the virtual
photon one, calculated to the leading order in ml /m , is given by [27]:

1 d
= D(y, r) 1 + Kl (y) .
(26)
0 dy
319
Fig. 5. Virtual diagram.
D(y, r) is the lepton distribution function, given, to the first order in , by:

2
(L 1) + O( ) P (1) (y),
D(y, r) = (1 y) +
(27)
where L is the logarithm

L = ln
m
,
ml
(28)
diverging in the collinear limit; if the lepton is an electron, L ' 5.6. P (1) (y) is the Gribov
LipatovAltarelliParisi kernel [2830], which can be expressed in the form:
P
(1)
1 + y2
(1 y)
(y) =
1y
Z1
dz
0
1 + z2
.
1z
(29)
Kl (y) is a finite term, free from infrared and collinear singularities, which has the
expression:
1 + y2
1
ln y.
Kl (y) = 1 y (1 y) ln (1 y) +
2
1y
(30)
The differential decay rate to order therefore becomes:
1 d
= (1 y) + (L 1)P (1) (y) + Kl (y).
0 dy
(31)
To calculate the inclusive decay rate, we have to integrate the expression in Eq. (31) over
y; to the leading order in the lepton mass, the physical region for y is 0 6 y 6 1. The kernel
P (n) has the property that:
Z1
dyP (n) (y) = 0;
(32)
thus, when we calculate the inclusive decay rate, the coefficient of the collinear logarithm
vanishes and the resulting expression is finite in the zero mass limit.
320
Carrying out the integration over y, we obtain the well known inclusive decay rate to
order

15 2
=1+
.
(33)
0
8
3
As expected, the expression (33) is finite in the collinear limit and is also free from
infrared divergences, because, as usual, the infrared divergences present in the soft photon
contribution and in the virtual photon contribution have cancelled each other.
Let us now discuss the mass singularities in the case of the right-handed Inner
R /dy is finite
Bremsstrahlung contribution, given in Eq. (13). It is easy to see that dIB
in the limit r 0, i.e., it is free from collinear singularities. In this limit the coefficients
of both the collinear logarithms vanish. Indeed, as we have observed in Section 3.2, only
the term related to the axial anomaly survives in the zero mass limit, that is the part of the
decay rate independent of the lepton mass.
We observe that the right-handed Inner Bremsstrahlung contribution is free from infrared
divergences as well. If we make the lepton energy y reach its kinematical limit y MAX , we
obtain a finite result:
R
1 dIB
0 as y y MAX.
0 dy
(34)
The result (34) shows that the soft photon emission does not contribute to the radiative
+ decay with a right-handed lepton. This is a consequence of the fact that the soft photon
contribution factorizes with respect to the Born decay rate, but this vanishes in the case
of right-handed l + (see Eq. (10)). Physically, Eq. (34) is due to the fact that soft photons
dont carry spin, thus they cannot contribute to the angular momentum balance; therefore
the process with the right-handed lepton emitting a soft photon is forbidden by angular
momentum conservation.
For the same reason of angular momentum conservation, in the right-handed case also
the virtual contribution identically vanishes. The virtual photon diagram (see Fig. 5)
interferes with the Born one; the corresponding correction to the total decay rate for
ll was calculated long ago by Kinoshita [31] and is given by:

1
r
3
ln r + 1 0 ,
b(r) 4 ln
ln r + 3 +
(35)
3 ln
v =
2
m
2
m
1r
where
1+r
ln r + 2.
1r
3 is the ultraviolet cutoff and is the infrared one.
As usual, the virtual correction is factorized with respect to the Born decay rate, but, as
we have already seen, if the lepton is right-handed, this is identically zero.
In the right-handed case, the mass singularities cancellation occurs trough a mechanism
different from the one working in the unpolarized decay rate. The infrared and the collinear
limits give separately a finite result. In particular, the coefficient of the collinear logarithms
is the lepton mass, instead of the usual correction factor coming from the soft and the virtual
b(r) =
321
photon contributions, as in Eq. (31). In this sense, since the soft and collinear radiation
factorizes with respect to the Born helicity changing decay rate, the double logarithm
Sudakov term can be equally factorized. It could be useful to investigate the impact of
the higher order terms on the radiative correction to the Born amplitude. It is an open
question if say hard collinear photons can be factorized and resummed.
The particular mass cancellation mechanism occurring in the right-handed radiative
decay is the consequence of the combination of two constraints: the angular momentum
conservation in the pion vertex and the helicity flip in the photonlepton vertex.
The situation is completely different if we consider the radiative process with the
outgoing left-handed lepton, i.e., the process without helicity flip. The differential Inner
Bremsstrahlung contribution for left-handed outgoing lepton is given by:

L

1
1
1 dIB
=
2A r(2A r + 6) y 2 1 2A
2
0 dy
4 (1 r) A(1 y + r)
y +A

+ (A 2r)(1 + r)2 + Ay(y 4r) 2ry(1 + y) + y(5r 2 + y 2 + 1) ln
y A

y+A2
.
(36)
+ (1 y + r)2 (2r A y) ln
yA2
The expression (36) contains collinear singularities, since the coefficients of the collinear
logarithms dont vanish in the limit r 0, as one can see from Eq. (36). The decay rate
(36) is also infrared divergent, as one can verify by taking the limit y y MAX . In this case
we dont have the constraint constituted by the helicity flip in the photonlepton vertex and
in the pion vertex the angular momentum is conserved for soft and virtual photon emission.
Thus, in the left-handed case, the mass singularity cancellation occurs in the ordinary way,
as in the unpolarized case, i.e., in the total inclusive decay rate, obtained by adding all the
order contributions.
Let us show how the cancellation takes place. As we have already seen (Eq. (10)), in the
Born + decay the outgoing lepton is left-handed, due to angular momentum conservation.
Thus the unpolarized and the left-handed Born decay rate coincide:
0L = 0 .
(37)
Because of the factorization with respect to the Born decay rate, also the unpolarized and
the left-handed virtual contributions are equal:
vL = v .
(38)
Expressing the left-handed Inner Bremsstrahlung contribution in terms of the unpolarized

and the right-handed ones, the total contribution to order to the left-handed process is
given by:
L
R
= (0 + v + IB ) IB
.
TOT
(39)
The expression (39) is finite both in the infrared and in the collinear limit, because the
mass singularities present in the terms between brackets cancel each other, as we have seen
R is free from mass singularities.
(see Eq. (33)) and IB
322
Let us now discuss the origin of these different cancellation mechanisms. The
presence of mass singularities is a consequence of the fact that the states of a theory
containing massless particles are highly degenerate. The KLN theorem states that the
mass singularities disappear from the transition probability when we average it over the
ensemble of degenerate states. This theorem contains the BlockNordsieck theorem as a
special case, when we consider only the cancellation of the infrared divergences.
We can define two degeneration ensembles, one relative to the infrared divergences
and one relative to the collinear singularities. We call them the infrared and the collinear
ensembles, respectively. If we sum the transition probability over the states contained in the
former, the infrared divergences cancel out and if we average it over the latter, we obtain
a quantity free from both infrared and collinear singularities. The infrared ensemble is the
one prescribed by the BlockNordsieck theorem, while the second is the one prescribed by
the KLN theorem and contains the first as a subset.
Let us now examine how the infrared and collinear ensemble are composed for the
radiative pion decay in the cases of unpolarized, left-handed and right-handed outgoing
lepton. This discussion concerns the issue of the degeneration of states already addressed
for the unpolarized case [19,20]. This issue in the case of helicity changing processes
presents peculiar features.
We have seen in Section 4.1 that in the unpolarized and left-handed Inner Bremsstrahlung
contributions there are mass singularities, indicating that we have not summed the transition probability over the entire ensemble of degenerate states. In these cases the infrared
ensemble contains, to order , the state with a pion, a charged lepton and a neutrino and
all the other states differing from this for the presence of a soft virtual or real photon, i.e.,
a photon with an energy E < , where is an infrared cut off, tipically the measurament
apparatus resolution. The collinear ensemble is constituted by all the states of the infrared
ensemble plus the states with a hard photon moving parallel to the pion or the lepton.
Clearly the degeneration arises in the limit ml 0.
R /dy is finite both in the infrared and in
According to the KLN theorem the fact that dIB
the collinear limits means that in the right-handed case, calculating the differential decay
rate (i.e., summing over the photon polarization and integrating over the photon energy
and emission angle), we have already averaged over the set of degenerate states relative
to this process. Let us now consider how the degeneration ensemble for the right-handed
R /dy we have not averaged over the infrared
radiative decay is composed. To obtain dIB
subspace of the collinear ensemble, but this is enough to render the transition probability
free from mass singularities. Indeed in this case the infrared ensemble is empty, owing to
the constraints imposed both by the angular momentum conservation in the pion vertex and
by the helicity flip in the photonlepton vertex. Thus in this case the degeneration ensemble
contains only the states with the outgoing lepton accompanied by hard collinear photons.
We conclude that imposing to the outgoing lepton a polarization opposed to the
one prescribed by the vertex preceding the photon vertex, implies a reduction of the
degeneration subspace. This fact has two consequences: the first is that both the infrared
and the collinear limits are finite and the second is that these limits are disconnected, since
the collinear degeneration subspace is constituted only by the states with the pion and
323
the outgoing lepton accompanied by hard collinear photons. Thus in this case we have a
particular application of the KLN theorem.
4.2. The Z 0 case
In the radiative pion decay, due to the angular momentum conservation in the pion
vertex, there is no room for a right-handed lepton. For such a channel, soft and virtual
photon contributions are zero. This result is valid independently of the lepton mass.
Let us now consider a more general case, by loosing the value of the angular momentum
of the decaying state. As an example, we study the radiative Z 0 decay in a lepton
antilepton (l l + ) pair, in which the lepton is in a definite helicity state.
The Z 0 -leptons vertex is:

MZ G 1/2
(gv ga 5 ),
i
2
2
with
gv = 1 4 sin W 2 ,
ga = 1,
where W is the Weinberg angle and MZ is the Z 0 mass.

We have chosen this process, since we can control the structure of the Z 0 -leptons
coupling; thus it is possible to point out the role played by the conservation law occurring in
this vertex in the chiral limit in the collinear singularity cancellation. If we set gv = ga = 1,
we require that in the limit of zero lepton mass, the Z 0 couples to a left-handed lepton.
We calculate the decay rate for the process in which the lepton is right-handed. At the
Born level this is given by:
0R =
GMZ3
48 2
o
n

1 4r gv2 + ga2 (1 r) + 3r gv2 ga2 2gv ga (1 4r) .
(40)
If in Eq. (40) we set gv = ga = 1, 0R vanishes in the chiral limit, since there isnt the term
related to the axial anomaly.
Let us now study the decay process with the lepton emitting a real photon (see Fig. 6)
and evaluate the probability that the outgoing lepton is right-handed. The electromagnetic
interaction doesnt couple states with different chirality, hence the decay rate is expected
to vanish for r 0.
The decay rate for the process described by the diagram of Fig. 6, differential with
respect to the lepton energy, is given by:

dR

GM 3 (y 2)(1 y)2 2
= Z
gv + ga2 A + 2gv ga (2r y)
2
dy
96 2 2 4(1 y + r)

(1 y) 2
gv + ga2 A(2r y) + 2gv ga y 2 2r
+
2(1 y + r)

2 2
2 gv 2ga2 Ar + gv2 + ga2 A + 2gv ga 4r y 2 + y 1
+
(y 1)
324
Fig. 6. Diagram corresponding to the decay rate (41).

y+A2
1
y +A
2
2
2
2
ln
+ gv + ga + 2gv ga 4r + r(y y + 1) y
.
ln
A
y A
yA2
(41)
Here
r=
m2l
MZ2
y is the usual dimensionless variable:

y=
2E1
MZ
where E1 is the lepton energy and

q
A = y 2 4r.
The physical region for y is
2 r 6 y 6 1.
(42)
The result obtained, as given by the emission of the photon by a single leg, is gauge
dependent. To have a gauge independent amplitude, the contribution of the right-hand
diagram of Fig. 7 must be added. For the purpose of the polarized amplitude, however,
the helicity flip contribution of the right-hand diagram of Fig. 7 gives zero in the massless
limit and is therefore negligible in our discussion.
From now on we consider the case gv = ga = 1, to have the condition of chirality
conservation in the Z 0 vertex for ml 0. Taking this limit in Eq. (41), we see that dR /dy
does not vanish:
dR
dy
(1 y)0 (Z 0 )
2
as r 0,
gv = ga = 1.
(43)
The result of this limit is the contribution related to the axial anomaly, which has the same
form of the one found in the pion case.
Let us now discuss the mass singularities cancellation mechanism for the Z 0 decay case.
If we keep the lepton mass different from zero, the chirality is not fixed by the interaction
325
Fig. 7. The other order diagrams.
occurring before the photon emission, even if we set gv = ga = 1. Thus, for ml 6= 0, the
soft and virtual photons contributions are different from zero and diverge in the infrared
limit. Indeed, if we let the lepton energy reach its kinematical limit, y MAX = 1, we see
that dR /dy diverges. We expect the infrared divergences to cancel, if we add all the
first order contributions, given by the diagrams of Figs. 6 and 7 and calculate the totally
inclusive decay rate.
Eq. (43) shows that the collinear limit gives a finite result. Thus, we conclude that, as in
the case of the radiative pion decay, evaluating dR /dy we have already summed over all
the collinear degeneration subspace. As a consequence, the collinear and infrared limit are
disconnected.
If we take the limit y 1 in Eq. (43), we obtain:
dR
0 as ml 0, y 1 and gv = ga = 1.
(44)
dy
In general, if a quantity is finite in the collinear limit, it is finite also in the infrared
limit, since the collinear subspace contains the infrared one. Eq. (44) indicates that in the
massless limit the soft photon contribution is zero. Indeed, it is factorized with respect to
the Born decay rate, which, for r 0 and gv = ga = 1, vanishes. As we have discussed in
Section 2, the presence of the anomalous term is directly connected to the emission of the
photon, hence it vanishes in the infrared limit.
The virtual photon contribution vanishes in the zero mass limit, as well. Indeed, it
is factorized respect to the Born decay rate, which goes to zero as r 0. The virtual
correction factor can produce only a logarithmic collinear singularity, not a power-like
one, needed for the cancellation of the chiral suppression.
We observe that taking the infrared limit x 0 in Eq. (22), gives a finite result (indeed
the cross section vanishes). This is a consequence of the fact that the cross section (22)
has been calculated to the leading order in the lepton mass. From Eq. (41), we see that the
infrared divergent term is given by:

R
d
2
4r
2r
y
(45)
dy IR (y 1) y
y
and it is proportional to the lepton mass. Performing the calculation, neglecting the mass
terms, as done in Ref. [11], means imposing the chirality conservation law in the Z 0 vertex;
326
thus the soft photon contribution is zero and the infrared divergences disappear. To the
leading order in the lepton mass, we have only the anomalous term, which vanishes in the
infrared limit.
As done for the pion case, we now examine the composition of the degeneration
ensembles for the radiative Z 0 decay with the right-handed lepton.
Eq. (44) indicates that, in the chiral limit, the soft photons dont contribute to the process
with the right-handed outgoing lepton, since, not carrying spin, they cannot contribute to
the helicity flip.
As we have already said, the virtual photon contribution vanishes in the zero mass limit.
In the limit ml 0, also the diagram with the photon emitted by l + doesnt contribute,
since, clearly, it violates the chirality conservation in the Z 0 vertex.
The collinear degeneration arises in the massless limit. The result (43) shows that, in this
limit, the collinear ensemble is constituted only by the states with the lepton accompanied
by hard collinear photons, just as in the case of the pion decay. Thus for ml 0 the infrared
ensemble is empty and also the states with the antilepton accompanied by a hard collinear
photon dont contribute.
5. Conclusions
We have shown that the Dolgov and Zakharov treatment of the axial anomaly can be
extended to processes characterized by a lepton which changes helicity by emitting a
photon, as it was already noticed in [10,11]. The corresponding decay rates dont vanish in
the chiral limit, thanks to a term independent of the lepton mass; we interpret the presence
of this term as related to the axial anomaly. This can be seen as a signal of the anomalous
symmetry breaking in processes different from the ones described by the usual anomalous
diagrams, as the 0 decay.
We have calculated the decay rates corresponding to the + and Z 0 radiative decays and
analysed their infrared and collinear limits. It is essential that the terms of all orders in the
lepton mass are kept, since the cancellation of the infrared and collinear divergences takes
place among these terms. We have examined the connection between the polarization of
the outgoing leptons and the application of the KLN theorem.
We have seen that for the helicity changing processes the cancellation of the collinear
singularities occurs through a mechanism different from the usual one of real and virtual
compensation. The coefficients in front of the collinear terms go to zero in the chiral
limit, producing the finitness of the distribution. However, we have found a difference
between the pion case and the more general case of the Z 0 decay. The former represents
a particular case because of the angular momentum conservation in the pion vertex: due
to this constraint, the virtual and real soft photon contributions are zero, even if we keep
the lepton mass different from zero. The Inner Bremsstrahlung contribution is finite both
in the infrared and in the collinear limits.
In the case of the radiative Z 0 decay, the decay rate diverges in the infrared limit, since,
for ml 6= 0, the soft photon contributions are not zero. However if we take the collinear
327
limit, we obtain a finite result. There are collinear degenerate states only when we take
the zero lepton mass limit. In this limit, we have seen that the virtual and real soft photon
contributions do vanish. In order to make the collinear limit finite, it is therefore sufficient
to sum over degenerate states constituted by the changing helicity lepton accompanied by
a hard collinear photon. The transition probability becomes finite after summing over the
photon final phase space.
We noticed that in the helicity changing processes the collinear limit results disconnected
from the infrared one. The contributions coming from the virtual photon emission and from
the emission of photons by particles different from the one changing helicity, are zero.
This situation is due to the fact that the Born part of the process fixes the fermion chirality
in the zero mass limit, while, after the photon emission, it is in a state of opposite chirality;
this reduces the collinear ensemble.
We conclude that the collinear singularity cancellation mechanism for helicity changing
processes is controlled by the anomalous breaking of the chiral symmetry. The axial
anomaly implies that the collinear limit gives a finite result, independent of the fermion
mass.
The extension the above remarks to other gauge theories like QCD, is possible. It could
allow a more sistematic and complete treatment of the infrared and collinear singularities.
Acknowledgements
We wish to thank V. Fadin and E. Kuraev for valuable comments and S. Forte, J. Kodaira
and L. Lipatov for useful discussions.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
S. Adler, Phys. Rev. 177 (1969) 2426.

J.S. Bell, R. Jackiw, Nuovo Cimento A 60 (1969) 47.
K. Fujikawa, Phys. Rev. Lett. 42 (1979) 1195.
A.D. Dolgov, V.I. Zakharov, Nucl. Phys. B 27 (1971) 525.
V.N. Gribov, Hungarian Academy of Sciences Report KFKI-1981-66, 1981, Lectures given at
the Institute for Theoretical Physics, Santa Barbara, 1988.
A.H. Mueller, Phys. Lett. B 234 (1990) 517.
R. Jackiw, in: Lectures on Current Algebra and Its Application, Princeton University Press,
Princeton, NJ, 1972.
H.B. Nielsen, M. Ninomiya, Int. J. Mod. Phys. A 6 (1991) 2913.
C. Adam, R.A. Bertlmann, Z. Phys. C 56 (1992) 123.
A.V. Smilga, Comments Nucl. Part. Phys. 20 (1991) 69.
B. Falk, L.M. Sehgal, Phys. Lett B 325 (1994) 509.
M.M. Tung, Phys. Rev. D 52 (1995) 1353.
S. Groote, J.G. Krner, J.A. Leyva, Phys. Lett. B 418 (1998) 192.
G. t Hooft, in: G. t Hooft et al. (Eds.), Recent Developments in Gauge Theories, Proceedings of
the Cargese Summer Institute, Cargese, France, 1979, NATO Advanced Study Institutes, Series
B: Physics, Vol. 59, Plenum, New York, 1980.
R.D. Carlitz, J.C. Collins, A.H. Mueller, Phys. Lett. B 214 (1988) 229.
328
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
S. Narison, G.M. Shore, G. Veneziano, Nucl. Phys. B 391 (1993) 69.

G.M. Shore, G. Veneziano, Mod. Phys. Lett. A 8 (1993) 373.
A. Freund, L.M. Sehgal, Phys. Lett. B 341 (1994) 90.
T.D. Lee, M. Nauenberg, Phys. Rev. 133 (1964) 1549.
T. Kinoshita, J. Math. Phys. 3 (1962) 650.
L. Trentadue, M. Verbeni, UPRF-99-21, December 1999; Phys. Lett. B 478 (2000) 137.
V.I. Zakharov, Phys. Rev. D 42 (1990) 1208.
K. Huang, Quarks, Leptons and Gauge Fields, World Scientific, Singapore, 1992.
D.A. Bryman, P. Depommier, C. Leroy, Phys. Rept. 88 (1982) 151, and references therein.
H.F. Contopanagos, M.B. Einhorn, Phys. Rev. D 45 (1992) 1322, and references therein.
F. Bloch, A. Nordsieck, Phys. Rev. 52 (1937) 54.
E.A. Kuraev, JETP Lett. 65 (1997) 127.
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 438; Sov. J. Nucl. Phys. 15 (1972)
675.
[29] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[30] Yu.L. Dokshitzer, Sov. Phys. JETP 46 (1977) 641.
[31] T. Kinoshita, Phys. Rev. Lett. 2 (1959) 477.

Meron-cluster algorithms and chiral-symmetry

breaking in a (2 + 1)D staggered fermion model
J. Cox a , K. Holland b,1
a Center for Theoretical Physics, Laboratory for Nuclear Science, Department of Physics, Massachusetts
Institute of Technology (MIT), Cambridge, MA 02139, USA

b Institute for Theoretical Physics, University of Bern, Sidlerstrasse 5, CH-3012 Bern, Switzerland
Abstract
The recently developed meron-cluster algorithm completely solves the exponentially difficult
sign problem for a number of models previously inaccessible to numerical simulation. We use
this algorithm in a high-precision study of a model of N = 1 flavor of staggered fermions in
(2 + 1) dimensions with a four-fermion interaction. This model cannot be explored using standard
algorithms. We find that the Z(2) chiral symmetry of this model is spontaneously broken at low
temperatures and that the finite-temperature chiral phase transition is in the universality class of the
2D Ising model, as expected. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ha; 11.30.Rd; 11.10.Wx; 02.70.Lq; 71.10.Fd
Keywords: Chiral symmetry; Symmetry breaking; Fermion algorithm; Cluster algorithm; Meron; Monte Carlo
1. Introduction
There are a number of models of interest which suffer from a very severe sign problem.
This includes QCD and other field theories with a nonzero chemical potential or a nonzero
vacuum angle or odd numbers of fermion flavors, frustrated quantum spin systems, like the
quantum antiferromagnet in an external magnetic field, and models for strongly-correlated
electrons, like the Hubbard model for high-temperature superconductivity. These models
have a Boltzmann weight which can be negative or even complex and so cannot be
interpreted as a probability. This difficulty can be overcome in numerical simulations
by including the sign or phase of the Boltzmann weight with observables. Unfortunately,
this leads to large cancellations and gives exponentially small observables. This requires
This work is supported in part by funds provided by the US Department of Energy (DOE) under cooperative
research agreement DE-FC02-94ER40818 and by the Schweizerischer Nationalfonds.
1 holland@itp.unibe.ch
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 9 9 - 6
332
J. Cox, K. Holland / Nuclear Physics B 583 (2000) 331346
exponentially large statistics, which makes it in practice impossible to simulate these

models numerically.
Recently, a new technique has been developed, called meron-cluster algorithms [1],
which completely solves the sign problem for some of these models [25]. It identifies the
origin of the sign problem with properties of the clusters, which enables it to be eliminated.
Cluster algorithms in general are extremely efficient at exploring configuration spaces and
very often do not suffer from critical slowing down as a phase transition is approached,
unlike many other algorithms. Combined with the ability to construct improved estimators,
we can perform a high-precision study of these models with only modest statistics.
In this paper, we explore a model of N = 1 flavor of staggered fermions in (2 + 1)
dimensions with a four-fermion interaction. This model has a very severe sign problem
and cannot be simulated with standard techniques. We build a meron-cluster algorithm,
which we use to perform a high-precision study. We find that the Z(2) chiral symmetry
of this model is spontaneously broken at low temperatures and, using finite-size scaling
analysis, we verify that the finite-temperature chiral phase transition is in the universality
class of the 2D Ising model. A recent study of the same model with Z(2) chiral symmetry
in (3 + 1) dimensions has shown, using a meron-cluster algorithm, that a finite temperature
chiral phase transition occurs which has the universal behavior of the 3D Ising model [2].
The work presented in this paper concerns a different universality class and also constructs
more observables than were previously considered.
The identification of the finite-temperature critical behavior is not entirely straightforward. A model of N fermion flavors with a four-fermion interaction shows mean-field
behavior in the N = limit [6]. On the other hand, at finite N one finds the nontrivial
critical behavior that one expects based on dimensional reduction and standard universality arguments. For example, in [7,8] it has been verified that the chiral phase transition in
a (2 + 1)D four-fermion interaction model with N = 4 flavors and Z(2) chiral symmetry
is in the universality class of the 2D Ising model. Due to the fermion sign problem, standard fermion simulation methods often do not work in models with too small a number of
flavors. The work presented in this paper shows that the same universal behavior holds for
N = 1 flavor.
The standard technique to deal with fermions in Monte Carlo simulations is to integrate
them out, resulting in a non-local bosonic theory. The meron-cluster algorithm does not
integrate out the fermions, instead we describe them with a local theory using a Fock
space basis of occupation number. The fermion sign arises as a nonlocal property due to
the permutation of fermion world lines. Using probabilistic rules, we connect neighboring
lattice sites, producing closed loops, which are the clusters. A cluster is flipped by making
all of its occupied sites empty and its empty ones occupied. Such a cluster flip can
change the fermion sign by changing the permutation of fermion world lines. A cluster
whose flip changes the sign we call a meron. We can tell if a cluster is a meron simply
from its structure. A typical configuration contains many merons, yet the observables of
interest are only nonzero for configurations with very few or no merons. The signals from
standard Monte Carlo algorithms are so exponentially small because the Markov chain
explores a vast configuration space, yet only an exponentially small subspace makes any
333
contribution to measurables. By restricting ourselves to only explore the relevant subspace,

we completely solve the sign problem.
This paper is organized as follows. In Section 2, we present the fermionic model which
we have studied, calculate its partition function using the Hamiltonian formulation and find
that there is a sign problem. In Section 3, we describe briefly the meron-cluster algorithm
which we have used to perform numerical simulations of this model. We present the results
of the simulations in Section 4 and give our conclusions in Section 5.
2. The staggered fermion model

We consider staggered fermions in the Hamiltonian formulation on a 2-dimensional
spatial lattice of extent L, which is even. The Hamiltonian operator is
X
X
hx,i + m
(1)x1+x2 x+ x ,
H=
x,i
hx,i = x,i x+ x+i

+ + x + G x+ x 12 + x+i 12 ,
x+i
x+i
(2.1)
where x,1 = 1 and x,2 = (1)x1 are the standard KawamotoSmit phases for staggered
fermions and G is a constant. The fermionic operators satisfy the usual anticommutation
relations {x , y } = {x+ , y+ } = 0, {x+ , y } = xy . The same model in (3 + 1)
dimensions was explored in [2]. We refer the reader to this paper, where various features of
the model and the meron-cluster algorithm are discussed in more detail than we give here.
In the hamiltonian formulation of the theory, fermion doubling on the lattice occurs
only in the spatial dimensions. Using staggered fermions, the four Dirac components of
a spinor are distributed spatially, reducing the number of fermion flavors by a factor of
four. Thus this (2 + 1)-dimensional model contains N = 1 fermion flavor. The continuum
fermion represented by this model has four spinor components. The model has a global
U (1) symmetry corresponding to conserved particle number, as the total particle number
operator commutes with the Hamiltonian
X
x+ x ,
[H, N] = 0.
(2.2)
N=
x
For the massless interacting Hamiltonian, i.e. m = 0 and G 6= 0, there is also a discrete
Z(2) symmetry corresponding to shifts by one lattice spacing. However the mass term
breaks that symmetry explicitly. For a single flavor of interacting massless fermions, the
symmetry of the model is U (1) Z(2) and we refer to the discrete symmetry as chiral
symmetry. A single noninteracting massless fermion flavor has a U (1) axial symmetry
(there is no gauge interaction, so this symmetry is not anomalously broken). The discrete
Z(2) symmetry is what remains of the continuous symmetry after the interaction has been
turned on. From now on, we set m = 0 and explore the behavior of the chiral symmetry
of this model. The symmetries of staggered fermions are discussed in detail in Ref. [9]. At
high temperatures, the Z(2) chiral symmetry is intact. As the temperature is lowered, if the
chiral symmetry is spontaneously broken at some finite temperature, from universality we
334
expect this to be a second-order phase transition. As the critical point is approached, the
correlation length diverges and the system becomes insensitive to the time extent. Due to
dimensional reduction, we expect a finite-temperature chiral phase transition in this model
to belong to the 2D Ising universality class.
The partition function of the model is

M

Z = Tr exp(H ) = lim Tr exp(H )
M

M
= lim Tr exp(H1 ) exp(H2 ) exp(H3 ) exp(H4 ) ,
(2.3)
M
where we use the SuzukiTrotter decomposition to divide the euclidean time extent into
4M time slices, the lattice spacing in the time direction being = /M. The hamiltonian
operator is decomposed into four parts:
X
X
X
X
hx,1 +
hx,2 +
hx,1 +
hx,2 = H1 + H2 + H3 + H4 . (2.4)
H=
x1 even
x2 even
x1 odd
x2 odd
All of the terms that contribute to a particular Hi commute with one another, as each term
is an interaction between nearest-neighbors and each lattice site appears in only one such
nearest-neighbor pair. However, the Hi do not commute with one another. We note that it
is not actually necessary to discretize the time direction, as it is possible to work directly
in the euclidean time continuum [10].
We can equivalently describe this model with bosonic operators, using a transformation
by Jordan and Wigner [11]. We order the lattice sites on each time slice arbitrarily into
a chain, which can be done in any number of spatial dimensions. For example, a possible
ordering of points in two spatial dimensions is by an index l = x1 + (x2 1)L. The
fermionic operators are now represented by a chain of Pauli matrices:

3
3
l+ ,
x = 13 23 . . . l1
l ,
x+ x = 12 l3 + 1 ,
x+ = 13 23 . . . l1

i j
(2.5)
l , m = 2ilm ij k lk ,
= 12 1 i 2 ,
where the spatial position x is denoted by the index l and the Pauli matrices satisfy the
usual commutation relations. To calculate the partition function of the theory, we use the
Fock space basis of occupation number nx = 0, 1, i.e., the eigenstates of 3 . The occupied
and empty states are respectively |1i and |0i, which satisfy 3 |1i = |1i and 3 |0i = |0i.
The time evolution operator exp(Hi ) acts on a time slice of occupation number
states, producing the next time slice. This is decomposed into the product of operators
exp(hx,i ) acting on nearest-neighbor occupation states. The transfer matrix is
exp G
0
0
0
2

0
0
cosh 2 sinh 2
G
,
(2.6)
exp(hx,i ) = exp

4
0
0
sinh 2 cosh 2

0
0
0
exp G
2
the basis being |00i, |01i, |10i and |11i, where e.g. |01i represents state |0i at x and |1i
If these nearest-neighbors are labelled l and m, the off-diagonal transfer matrix
at x + i.
335
3 3 . . . 3 . Note that this operator is diagonal in

elements have a factor = x,i l+1
l+2
m1
the occupation number basis.
The partition function of the theory is given as a path integral
X

Sign[n] exp S[n] ,
(2.7)
Zf =
n
where we sum over all possible configurations of occupation numbers n(x, t) = 0, 1 on

a (2 + 1)D spacetime lattice of points (x, t). The Boltzmann factor exp(S[n]) for
a configuration is the product of the Boltzmann factors for each spacetime plaquette
t), n(x, t + 1), n(x + i,
t + 1)]), which are:
exp(s[n(x, t), n(x + i,

G
,
exp s[0, 0, 0, 0] = exp s[1, 1, 1, 1] = exp
2

exp s[0, 1, 0, 1] = exp s[1, 0, 1, 0] = cosh ,
2

(2.8)
exp s[0, 1, 1, 0] = exp s[1, 0, 0, 1] = sinh .
2
All other plaquettes are illegal and have Boltzmann weight zero, as they represent
nonconservation of fermion number. Any configuration which contains illegal plaquettes
has itself Boltzmann weight zero and makes no contribution to the partition function. We
are only interested in legal configurations, which have to satisfy several constraints. Note
that here we have dropped the overall factor exp(G/4) that appeared in Eq. (2.6). The
sign of a configuration, Sign[n], is also a product of spacetime plaquette contributions
t), n(x, t + 1), n(x + i,
t + 1)] with
sign[n(x, t), n(x + i,
sign[0, 0, 0, 0] = sign[0, 1, 0, 1] = sign[1, 0, 1, 0] = sign[1, 1, 1, 1] = 1,
sign[0, 1, 1, 0] = sign[1, 0, 0, 1] = .
(2.9)
The occupied lattice sites define world-lines of fermions, which close due to the periodicity
of the euclidean time direction. The world-lines are free to permute during their time
evolution as the fermions interchange position and each configuration has a welldefined permutation of fermions. The Pauli exclusion principle tells us that the sign of
a configuration is the permutation sign of the fermions, hence Sign[n] = 1. This nonlocal
effect is contained in the factors of each spacetime plaquette.
The expectation value of a fermionic observable A[n] is given by
hA Signi
1 X
,
(2.10)
A[n] Sign[n] exp S[n] =
hAif =
Zf n
hSigni

1 X
Sign[n] exp S[n] ,
hSigni =
Zb n
where h. . .i means a measurement made in the bosonic ensemble, whose partition function
P
is Zb = n exp(S[n]). To measure one fermionic observable requires two bosonic
measurements. The quantities of physical interest which we measure are the chiral
condensate , the chiral susceptibility and a Binder cumulant U of the chiral
condensate, respectively:
336
[n] =
=

X
(1)x1+x2 n(x, t) 12 ,
4 x,t
( )2 f ,
V
U =1
h( )4 if
.
3[h( )2 if ]2
(2.11)
3. The meron-cluster algorithm

We now describe briefly the meron-cluster algorithm which we used to sample the
bosonic ensemble corresponding to the fermionic model without the sign factor. We set
G = 1, for which the bosonic model is the isotropic antiferromagnetic quantum Heisenberg
P
model, whose Hamiltonian is H = x,i (Sx1 S 1 + Sx2 S 2 + Sx3 S 3 ), where Sxi = 12 li
x+i
x+i
x+i
is a spin 1/2 operator at the lattice site x, labelled by l in the JordanWigner chain.
There already exist extremely efficient cluster algorithms to simulate bosonic quantum
spin systems [1214], and the first cluster algorithm for lattice fermions was constructed
in [15]. These algorithms can be implemented directly in the time continuum [10], i.e. the
SuzukiTrotter time discretization is not even necessary. In this study, we discretize the
time direction.
We use the same algorithm that was used in [2]. Each configuration is decomposed into
a set of clusters, which consist of connected lattice sites. A new configuration is generated
by flipping the clusters. When a cluster is flipped, all lattice sites contained in that cluster
change occupation number from n(x, t) to 1 n(x, t), i.e. the occupied sites become empty
and the empty ones occupied. To build the clusters, a probabilistic choice is made in each
t), n(x, t + 1), n(x + i,
t + 1)] as to
spacetime interaction plaquette [n(x, t), n(x + i,
which neighboring lattice sites are connected to one another. A cluster is a sequence of
connected sites. In this algorithm, the clusters are closed loops. The probabilistic choices
(called cluster break-ups) which build the clusters are designed to obey detailed balance
and we only allow break-ups which generate legal plaquettes under cluster flips. The cluster
rules are illustrated in Table 1. For plaquette configurations [0, 0, 0, 0] and [1, 1, 1, 1], i.e.
entirely empty or entirely occupied, we always connect sites with their time-like neighbors.
For configurations [1, 0, 0, 1] and [0, 1, 1, 0] where a fermion hops to a neighboring site,
we always connect sites with their space-like neighbors. For configurations [1, 0, 1, 0] and
[0, 1, 0, 1], i.e. a static fermion next to an empty site, we connect the sites with their timelike neighbors with probability p = 2/[1 + exp(/2)] and with their space-like neighbors
with probability 1 p. This algorithm was also used in [16]. It is extremely efficient, has
almost no detectable autocorrelations and its dynamical exponent for critical slowing down
is compatible with zero.
Each cluster has two orientations, with lattice site occupancies n(x, t) and 1 n(x, t).
When a cluster is flipped, the new configuration which is generated may have a different
sign from the previous one, depending on whether or not the permutation of fermion worldlines is changed. A cluster whose flip changes Sign[n] we call a meron, those which leave
Sign[n] unchanged we call nonmerons. Flipping a meron changes the topology of the
fermion world-lines. The term meron has been used before to denote half-instantons [17],
337
Table 1
Cluster break-ups of various plaquette configurations together with their probabilities, where p =
2/[1 + exp(/2)]. The dots represent occupied sites and the fat lines are the cluster connections
Weight
exp 2
Configuration
Break-ups

1

cosh 2
1p

sinh 2
such as in the 2D O(3) model at nonzero vacuum angle [18]. The number of merons
in a configuration is always even, as flipping all clusters leaves the sign unchanged. An
example of a meron-cluster is given in Fig. 1. When the meron-cluster is flipped the first
configuration with Sign[n] = 1 turns into the second configuration with Sign[n] = 1. For
cluster algorithms more general than the one described here, it is not always possible to
identify certain clusters as merons [19].
The meron concept alone gives us an exponential gain in statistics. Starting from
a configuration containing NC clusters, we consider the ensemble of 2NC configurations
where we allow all possible cluster orientations. If a configuration contains no merons, all
configurations in the ensemble have Sign[n] = 1. However, if it contains merons, half the
ensemble has Sign[n] = 1 and the other half Sign[n] = 1, which exactly cancel, giving
a contribution 0. The improved estimator gives hSigni = hN,0 i, i.e. the probability that
a configuration contains N = 0 merons, which is an exponential improvement on standard
algorithms, which measure a statistical average of 1. As explained in [2], this solves half
of the sign problem.
Sign[n] = 1
Sign[n] = 1
x
Fig. 1. Two configurations of fermion occupation numbers in (1 + 1) dimensions. The dots
represent occupied sites and the shaded plaquettes carry the interaction. With periodic spatial
boundary conditions, the two fermions interchange positions in the second configuration, giving
it Sign[n] = 1. Flipping the meron-cluster (represented by the fat line) changes one configuration
into the other, changing the fermion sign.
338
We also construct improved estimators for observables. The chiral susceptibility is

=

1 h( )2 Signi
1
( )2 f =
.
V
V
hSigni
(3.1)
P
The total chiral condensate for a given configuration, [n] = C C , is a sum of
cluster contributions. Averaging over the ensemble of 2NC configurations gives

P
2
C | C | N,0 + 2| C1 || C2 |N,2
.
(3.2)
=
V hN,0 i
This only gets contributions from configurations with N = 0 or N = 2 merons (C1 and C2
are the two merons). The vast majority of configurations contain many merons, but they
make no contribution to observables. The zero- and two-meron sectors of configuration
space are exponentially small, but they contain all of the contributions to . Restricting
ourselves to only explore this subspace, we exponentially enhance both the numerator and
denominator of Eq. (3.2), leaving the ratio invariant. This solves the remaining half of the
sign problem.
For the Binder cumulant U , we need to measure h( )4 if and hence

X

4
C i C j C k C l .
(3.3)
Sign( ) = Sign
Ci ,Cj ,Ck ,Cl
A clusters condensate contribution C changes sign when the cluster is flipped. When
a meron-cluster is flipped, Sign is changed. The nonzero terms in hSign( )4 i do not
change sign if any cluster in the configuration is flipped. These nonzero terms must contain
odd powers of C for all merons C in the configuration and even powers of C 0 for
all nonmerons C 0 . The average over the ensemble of 2NC configurations is
X

X

4
4
2
2
0
| C | + 6
| C | | C |
Sign( ) 2NC = N,0
C,C 0

+ N,2 4| C1 |3 | C2 | + 4| C2 |3 | C1 |

X
2
| C | | C1 || C2 |
+ 12
C

+ N,4 24| C1 || C2 || C3 || C4 | ,
(3.4)
where N is the number of merons in the configurations, C1 , C2 , C3 and C4 are the merons
and all sums in Eq. (3.4) are over nonmeron clusters. This average only gets contributions
from the zero-, two- and four-meron sectors and so we need only explore this subspace.
We average this quantity over the complete bosonic ensemble to measure hSign( )4 i
and hence U .
Consider the case of measuring . We expect that p(0)/p(2) (|C|/V )2 , where p(0)
and p(2) are the probabilities that a configuration has zero or two merons and |C| is the
average cluster size. In large volumes, the majority of configurations has two merons,
contributing 0 to hSigni. For even greater accuracy, we reweight the meron-sectors with
339
trial probabilities pt (0) and pt (2), so that they appear with roughly equal frequency. This
gives

P
2
C | C | N,0 pt (0) + 2| C1 || C2 |N,2 pt (2)
.
(3.5)
=
V hN,0 pt (0)i
The reweighting probabilities can be adjusted to minimize the statistical error. This
technique was previously used in [18]. To measure the Binder cumulant U , we use
reweighting probabilities pt (0), pt (2) and pt (4).
4. Numerical results
We have performed simulations of the staggered fermion model on lattices with
antiperiodic spatial boundary conditions from L = 4 up to L = 30 at inverse temperatures
in the range [1.0, 3.0], which includes the critical temperature where the chiral
symmetry is spontaneously broken. We have made separate runs with either a fixed number
of time slices (typically M = 10, i.e. 40 time slices) or with fixed lattice spacing in the
time direction ( = 0.1). In each simulation, we have made at least 1000 thermalization
sweeps followed by 10000 measurements, with these numbers increased by a factor of
10 for L 6 10. In one sweep of the lattice, a new cluster connection is proposed on
each interaction plaquette and each cluster is flipped with probability 1/2. To find the
optimal reweighting probabilities pt (N) which minimize the statistical error, we first make
a sample run without reweighting, only exploring the relevant meron-sectors. The observed
relative weights are then used in production runs, where the sectors appear with equal
probability. The major part of the sign problem is removed by the improved estimators, but
reweighting is necessary for accurate measurements in large volumes.
A sample of the data measured is given in Table 2. The table contains hSigni and
the susceptibility measured over all meron-sectors, and the reweighted hSignir and
r measured over the zero- and two-meron sectors only with the reweighting factor
pt (0)/pt (2). All of these data are produced with 1000 thermalization sweeps and 10000
measurements. As all of the contributions to come from the zero- and two-meron sectors,
and r should be identical. Note that hSignir , the fraction of zero-meron configurations
generated by sampling the zero- and two-meron sectors only, is typically a lot bigger than
hSigni, the fraction of zero-meron configurations generated over all meron sectors. In small
spacetime volumes, can be accurately measured even when sampling all meron sectors.
However, in large spacetime volumes, hSigni is too small to be measured and we can only
determine the susceptibility by restricting ourselves to the zero- and two-meron sectors.
The staggered fermion model suffers from a very severe sign problem which is solved by
the meron-cluster algorithm.
Fig. 2 shows the meron number probability distribution in an algorithm that samples all
meron sectors without reweighting. For small volumes the zero-meron sector and hence
hSigni are relatively large, while multimeron configurations are rare. On the other hand,
in larger volumes the vast majority of configurations has a large number of merons and
340
Table 2
Numerical results for the non-reweighted hSigni and susceptibility measured over all meron
sectors, and the reweighted hSignir and r measured over the zero- and two-meron sectors only
with a reweighting factor pt (0)/pt (2). For the larger volumes, hSigni and cannot be measured
L
hSigni
pt (0)/pt (2)
hSignir
8
8
8
8
10
12
14
16
20
24
28
1.0
1.5
2.0
2.4
2.4
2.4
2.4
2.4
2.4
2.4
2.4
0.804(5)
0.465(9)
0.214(6)
0.140(4)
0.057(3)
0.0203(8)
0.0052(6)
0.0005(2)
0.829(3)
2.84(3)
9.2(2)
16.6(3)
24.8(6)
33.0(1)
41.0(4)
80.0(40)
0.5/0.5
0.5/0.5
0.3/0.7
0.2/0.8
0.2/0.8
0.1/0.9
0.1/0.9
0.075/0.925
0.03/0.97
0.01/0.99
0.01/0.99
0.820(7)
0.52(1)
0.474(9)
0.501(9)
0.369(8)
0.443(7)
0.338(8)
0.314(4)
0.355(9)
0.46(1)
0.329(9)
0.826(6)
2.84(4)
9.0(1)
16.4(3)
24.2(5)
34.0(7)
44.0(1)
57.0(1)
82.0(3)
120.0(5)
156.0(8)
Fig. 2. The meron number probability distribution for various spatial sizes L = 8, 12, 20 and 28 at
= 2.4.
341
hence hSigni is exponentially small. For example, an extrapolation from smaller volumes
gives a rough estimate for the non-reweighted hSigni 109 on the L = 28 lattice at
= 2.4, while the reweighted hSignir = 0.329(9). Even if the configurations are entirely
uncorrelated, to achieve a similar accuracy without the meron-cluster algorithm one would
have to increase the statistics by a factor 1018 , which is obviously impossible. In fact, at
present there is no other method that can be used to simulate this model.
Fig. 3 shows the chiral susceptibility as a function of for various spatial sizes L. At
high temperatures (small ) is almost independent of the volume, indicating that chiral
symmetry is intact. On the other hand, at low temperatures (large ) increases with
the volume, which implies that chiral symmetry is spontaneously broken. To study the
critical behavior in detail, we have performed a finite-size scaling analysis for focusing
on the range [2.2, 2.6] around the critical point. Since a Z(2) chiral symmetry is
spontaneously broken at finite-temperature in this (2 + 1)D model, one expects to find
the critical behavior of the 2D Ising model. The corresponding finite-size scaling formula
valid close to c is [20]:
(L, ) = a(x) + b(y)L / ,
a(x) = a0 + a1 x + a2 x 2 + ,
x = c ,
b(y) = b0 + b1 y + b2 y + ,
y = ( c )L1/ .
(4.1)
For the 2D Ising model the critical exponents are given by = 1.0 and / = 1.75.
Fig. 3. The chiral susceptibility as a function of the inverse temperature for various spatial sizes
L = 8, 10, 12, 14 and 16 on lattices with 40 time slices. The chiral symmetry is intact at small and
spontaneously broken at large .
342
Fig. 4. The chiral susceptibility plotted against L for various values of computed with = 0.1. The
fit is to the finite-size scaling formula (4.1) with a(x) expanded to first order and b(y) expanded to
third order. The exponents are set to the 2D Ising model values. All curves are obtained from one fit.
The 2 per degree of freedom is 0.84 indicating a good agreement of our data with the finite-size
scaling ansatz and 2D Ising model critical exponents.
Assuming these values for the exponents, we obtain c = 2.43(1) for fixed = 0.1 from the
finite-size scaling fit, with a chi squared per degree of freedom of 0.84. The fit of the data
is plotted in Fig. 4. The same analysis of data generated in volumes with a coarser timediscretization 0.24 gives c = 2.36(2), which shows that the value of c is slightly
dependent on .
In the finite-size scaling equation (4.1), for large enough L one can neglect the term
a(x). Then /L / is a function of y = ( c )L1/ alone, i.e. the susceptibility data in
various volumes at various can be described by one universal function. We have varied
the value c to find if all the data can be collapsed onto one universal curve. In Fig. 5, we
plot the universal curve obtained by taking c = 2.43. The excellent agreement over a large
range of spatial volumes L and inverse temperatures is an indication of the quality of the
finite-size scaling fit.
We also measure UL , the Binder cumulant in volumes of extent L. In Fig. 6, we plot
the expected behavior of UL as L increases for different temperatures. For T > Tc , the
chiral symmetry is intact and UL flows into the T = fixed point U = 0. For T < Tc ,
the chiral symmetry is spontaneously broken and UL flows into the T = 0 fixed point
U = 2/3. If the universality class has a nontrivial fixed point U = U , then UL flows into
this value at T = Tc . By measuring UL in various volumes at many different temperatures,
we determine this flow numerically. We have measured the Binder cumulant values in
volumes up to L = 30 and we plot some of these values as a function of 1/L in Fig. 7.
These measurements are made with the number of time slices fixed at 40. Each curve in
343
Fig. 5. Finite-size scaling behavior of the chiral susceptibility . The data for various spatial sizes L
and inverse temperatures fall on one universal curve.
Fig. 6. The expected flow of UL as a function of 1/L. On each curve the temperature is constant.
the figure represents some fixed temperature. In Fig. 7, for small (i.e. high temperatures),
UL clearly flows into the infinite-temperature fixed point U = 0, while for large (low
temperatures), UL flows into the zero temperature fixed point U = 2/3. For close to c ,
we have to go to larger volumes to see this behavior. Near = 2.35, the cumulant values
appear to flow into a nontrivial fixed point U . Examining this region closely, we estimate
the critical inverse temperature as c = 2.36(2) and the fixed point value U = 0.60(1).
344
Fig. 7. Measured values of UL plotted versus 1/L for various in volumes with 40 time slices. Near
= 2.35, the values appear to flow into the nontrivial fixed point U = 0.60(1).
The finite-size scaling fit of measured at this 0.24 gives the same value of c . Note
that this deviates slightly from the critical temperature measured at = 0.1. The universal
fixed point value for the 2D Ising model is estimated as U 0.58 [21]. This is further
evidence that the chiral phase transition belongs to the 2D Ising universality class.
5. Conclusions
The meron-cluster algorithm has recently been developed to allow numerical simulations
in models which suffer from a very severe sign problem. In this paper, we have applied
this technique to investigate a model of staggered fermions. Unlike standard methods,
which integrate out the fermions, resulting in a nonlocal bosonic action, we use a Fock
space of occupation number to describe the fermions. We have a local bosonic action,
with an additional nonlocal sign factor which contains the Fermi statistics. Due to the
Pauli exclusion principle, configurations which have an odd permutation of fermion world
lines have a negative sign. This sign leads to very large cancellations in observables and
usually makes it impossible to make accurate measurements in numerical simulations. The
meron-cluster algorithm decomposes every configuration into closed loops of connected
sites, each loop being a cluster. Loops which change the fermion sign when flipped are
identified as meron-clusters. A meron-cluster identifies a pair of configurations with equal
weight and opposite sign. This results in an exact cancellation of two contributions 1 to
the path integral, such that only configurations without merons contribute to the partition
345
function. Observables only receive contributions from configurations which contain very
few or no merons, whereas the vast majority of configurations contain many merons. By
only exploring the sectors of configuration space with the relevant numbers of merons,
one makes an exponential gain in statistics. Combined with efficient reweighting of the
remaining meron sectors, this completely solves the sign problem. Cluster algorithms
are extremely efficient at exploring configuration spaces and generating uncorrelated
configurations and generally do not suffer from critical slowing down. Even in models
without a sign problem, the meron-cluster algorithm is more efficient than standard fermion
simulation methods.
In this paper, we examined a model of N = 1 flavor of staggered fermions in (2 + 1)
dimensions, which has a Z(2) chiral symmetry. The model has a very severe sign problem
and cannot be solved by standard fermion simulation algorithms. Using a meron-cluster
algorithm, we were able to make high-precision measurements of the chiral susceptibility
and Binder cumulant even in very large volumes and low temperatures. In order to perform
an accurate and reliable finite-size scaling analysis, it was necessary to go to volumes so
large, where the sign problem is so severe, that a standard algorithm would require statistics
on the order of 1018 to attain a similar accuracy. We were able to verify that the model
undergoes a finite-temperature chiral phase transition, which belongs to the universality
class of the 2D Ising model. This is the behavior expected from dimensional reduction
and universality. The same universal behavior was observed in the N = 4 flavor case [7,8].
However, the standard fermion algorithm that was used in that study does not work for
N < 4 due to the fermion sign problem.
It is quite natural to use cluster algorithms in models of discrete variables. A future
possible application of the meron-cluster algorithm is in exploring quantum link models
[22] which are used in the D-theory formulation of QCD [2325]. In D-theory, a model
of discrete quantum variables undergoes dimensional reduction, resulting in an effective
theory of continuous classical variables. In quantum link QCD the quarks arise as domain
wall fermions. The application of meron-cluster algorithms to domain wall fermions is in
progress. Also there are many applications to sign problems in condensed matter physics.
Investigations of antiferromagnets in a magnetic field and of systems in the Hubbard model
family are given in Ref. [35].
At present, the meron-cluster algorithm is the only method that allows us to solve the
fermion sign problem. A severe sign problem arises in lattice QCD calculations at nonzero
baryon number due to a complex action. It is therefore natural to ask if our algorithm
can be applied to this case. At nonzero chemical potential the 2D O(3) model, which is
a toy model for QCD, also suffers from a sign problem due to a complex action. When
applied to the D-theory formulation of this model, the meron-cluster algorithm solves the
sign problem completely [35]. It is an open question if such progress can be made in
investigations of QCD.
346
Acknowledgements
We would like to thank Shailesh Chandrasekharan and Uwe-Jens Wiese for helpful
discussions.
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]
S. Chandrasekharan, U.-J. Wiese, Phys. Rev. Lett. 83 (1999) 3116.

S. Chandrasekharan, J. Cox, K. Holland, U.-J. Wiese, hep-lat/9906021.
J. Cox, C. Gattringer, K. Holland, B. Scarlet, U.-J. Wiese, hep-lat/9909119.
S. Chandrasekharan, B. Scarlet, U.-J. Wiese, cond-mat/9909451.
S. Chandrasekharan, hep-lat/9909007, hep-lat/0001003.
B. Rosenstein, B.J. Warr, S.H. Park, Phys. Rev. D 39 (1989) 3088.
J.B. Kogut, M.A. Stephanov, C.G. Strouthos, Phys. Rev. D 58 (1998) 096001.
T. Reisz, hep-lat/9712017.
L. Susskind, Phys. Rev. D 16 (1977) 3031.
B.B. Beard, U.-J. Wiese, Phys. Rev. Lett. 77 (1996) 5130.
P. Jordan, E. Wigner, Z. Phys. 47 (1928) 631.
U.-J. Wiese, H.-P. Ying, Phys. Lett. A 168 (1992) 143.
H. G. Evertz, G. Lana, M. Marcu, Phys. Rev. Lett. 70 (1993) 875.
H.G. Evertz, The loop algorithm, in: D.J. Scalapino (Ed.), Numerical Methods for Lattice
Quantum Many-Body Problems, Frontiers in Physics, Addison-Wesley, Longman, New York,
1998.
U.-J. Wiese, Phys. Lett. B 311 (1993) 235.
U.-J. Wiese, H.-P. Ying, Z. Phys. B 93 (1994) 147.
C.G. Callan, R. Dashen, D.J. Gross, Phys. Lett. B 66 (1977) 375.
W. Bietenholz, A. Pochinsky, U.-J. Wiese, Phys. Rev. Lett. 75 (1995) 4524.
S. Chandrasekharan, in preparation.
H.W.J. Blte, E. Luitjen, J.R. Heringa, J. Phys. A 28 (1995) 6289.
K. Binder, Phys. Rev. Lett. 47 (1981) 693.
S. Chandrasekharan, U.-J. Wiese, Nucl. Phys. B 492 (1997) 455.
R.C. Brower, S. Chandrasekharan, U.-J. Wiese, Phys. Rev. D 60 (1999) 094502.
B.B. Beard, R.C. Brower, S. Chandrasekharan, D. Chen, A. Tsapalis, U.-J. Wiese, Nucl. Phys.
B (Proc. Suppl.) 63 (1998) 775.
U.-J. Wiese, Nucl. Phys. B (Proc. Suppl.) 73 (1999) 146.

Low-lying eigenvalues of the QCD Dirac operator

at finite temperature
P.H. Damgaard a , U.M. Heller b, , R. Niclasen a , K. Rummukainen c,d
a The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen, Denmark
b CSIT, Florida State University, Tallahassee, FL 32306-4120, USA
c Nordita, Blegdamsvej 17, DK-2100 Copenhagen, Denmark
d Helsinki Institute of Physics, P.O.Box 9, 00014 University of Helsinki, Helsinki, Finland
Abstract
We compute the low-lying spectrum of the staggered Dirac operator above and below the finite
temperature phase transition in both quenched QCD and in dynamical four flavor QCD. In both
cases we find, in the high temperature phase, a density with close to square root behavior, ()
( 0 )1/2 . In the quenched simulations we find, in addition, a volume independent tail of small
eigenvalues extending down to zero. In the dynamical simulations we also find a tail, decreasing
with decreasing mass, at the small end of the spectrum. However, the tail falls off quite quickly and
does not seem to extend to zero at these couplings. We find that the distribution of the smallest Dirac
operator eigenvalues provides an efficient observable for an accurate determination of the location of
the chiral phase transition, as first suggested by Jackson and Verbaarschot. 2000 Elsevier Science
1. Introduction
The correlations of Dirac operator eigenvalues in QCD and related theories have been
shown to have a fascinating relation to Random Matrix theory. There are two very different
domains of interest here. One is the so-called bulk of the eigenvalue spectrum of the
Dirac operator, far from both the infrared and the ultraviolet ends. The other is the
so-called hard edge at 0, i.e., the infrared end of the spectrum in theories with
spontaneous breaking of chiral symmetry. The relevance of Random Matrix theory in
describing eigenvalue correlations of the Dirac operator in the bulk of the spectrum was
first demonstrated by Halasz and Verbaarschot [1], and it has since been confirmed by
numerous lattice gauge theory studies [27]. From a theoretical point of view, these results
Corresponding author. E-mail:heller@csit.fsu.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 4 5 - X
348
P.H. Damgaard et al. / Nuclear Physics B 583 (2000) 347367
in the bulk of the spectrum remain to be better understood. In the other case, near 0, the
connection between universal Random Matrix theory results and the QCD Dirac operator
spectrum [815] is by now firmly established, and has also been extensively checked in
lattice gauge theory simulations [1621]. Already in the original work of Shuryak and
Verbaarschot [8] it was shown that the pertinent chiral Random Matrix theory partition
function coincides exactly with the effective field theory partition function in the relevant
microscopic limit. More recently an explicit relationship between the universal Random
Matrix theory eigenvalue distributions and those of the QCD Dirac operator has been
established [2227], the precise link being the partially quenched chiral condensate [25
29]. In this domain Random Matrix theory is an intriguing alternative description of exactly
the same phenomena that can be derived from the effective QCD partition function in the
large-volume limit where V 1/m4 [30].
There exists also an interesting physical situation that forces us to reconsider the two
different domains of the Dirac operator spectrum simultaneously. This is near the finitetemperature chiral phase transition, where the analytical connection between the effective
QCD partition function and Random Matrix theory breaks down even for the part of the
spectrum that is close to = 0. This situation is readily confronted in lattice gauge theory.
Indeed, it is for staggered fermions even rigorously proven [31], that chiral symmetry is
restored at high temperature. The low-lying spectrum of the Dirac operator must therefore
be quite different in the high temperature phase as compared with zero temperature. In
particular, the absence of chiral symmetry breaking implies, through the BanksCasher
relation,
h i = (0),
(1)
that the density of eigenvalues at zero vanishes. This could happen either just at that point
of = 0 (as, for example, in the free theory where () 3 ), or the spectrum could
develop a gap. The chiral phase transition occurs at the temperature Tc where the density
of Dirac operator eigenvalues just reaches zero at = 0. If the transition is continuous,
this will happen smoothly as the temperature T is increased towards Tc . In such a case,
an important question to settle is the precise power-law behavior of the spectral density
of the Dirac operator right at T = Tc , because this can be related to the critical exponents
of the phase transition [32]. The same chiral Random Matrix theory that yields universal
microscopic spectral correlators which exactly coincide with those of the Dirac operator at
T = 0 can be tuned in such a way as to just reach (multi-critical) points where [33]
() 2k
(2)
near = 0. Here k is an integer labeling the multi-criticality. At such points universality

of microscopic spectral correlators still holds in the Random Matrix theory context, but
there is no justification for assuming that these results are relevant for the Dirac operator
spectrum at T = Tc . 1 There is also a schematic Random Matrix model for the finite1 In particular, this behavior is only compatible with a continuous phase transition. But at a more fundamental
level, there is simply no longer any relation between the chiral Random Matrix theory ensemble and the effective
QCD partition function for temperatures T that do not satisfy T 3QCD .
349
temperature behavior of the Dirac operator spectrum [3436]. It gives a different behavior
at T = Tc : () 1/3 . Also here there is no physical justification for using it in
connection with the Dirac operator spectrum at finite temperature, but it is an interesting
model that depends on just one deterministic external parameter, and we shall therefore
return to it below.
Suppose, for a moment, that the Dirac spectrum actually develops a gap around = 0
above Tc . In Random Matrix theory the end of a spectrum around such a gap is referred
to as a soft edge. Generally, the (macroscopic) density of eigenvalues near a soft edge
behaves as (for > 0 ) [37,38]:
() ( 0 )2m+1/2
(3)
with 0 being the location of the edge. Here m is an integer that labels universality classes
classified by their Random Matrix theory potentials (m parameters in the potentials must
be tuned in order to reach each class). Thus the generic behavior, without any fine tuning,
corresponds to m = 0, which gives a square root approach to the soft edge:
() ( 0 )1/2 .
(4)
Random Matrix theory actually gives a more detailed, microscopic, description. This arises
from a blowing-up of the eigenvalue density function around the soft edge 0 with a
rescaling according to the macroscopic behavior (3). For example, for the generic m = 0
universality class, the corresponding microscopic eigenvalue density is [39]
() X[Ai(X)]2 + [Ai0 (X)]2 ,
(5)
where X ( 0 )N 2/3 and Ai(x) is the standard Airy function. Here N denotes the size
of the matrix, and the rescaling by N 2/3 is required in order to spread out the increasing
number of eigenvalues to obtain one well-defined limiting function. From the known
asymptotic behavior of the Airy functions one finds:

1
X
+O
,
for X ,
X
(6)
()
17
|X|1/2 exp(4|X|3/2/3), for X .

96
Thus, at the microscopic level the spectrum is not cut off sharply at 0 , but has an
exponentially suppressed tail beyond. Further, the square root behavior of the eigenvalue
density is modulated by wiggles corresponding to the distribution of particular eigenvalues
(smallest, second, third, etc.). (See, for example, Fig. 9.)
Until a few months ago, there existed only one low-statistics investigation of the
low-lying Dirac eigenvalue spectrum for staggered fermions in the high temperature
phase [40]. It did not unequivocally establish the existence of a gap. For example, some
low eigenvalues were found. However, they could possibly be attributed to would-be zero
modes from global gauge field topology, shifted away from zero by the explicit chiral
symmetry breaking of staggered fermions at finite lattice spacing [41]. This is a general
problem with staggered fermions that also we must face here: at finite lattice spacing, the
index theorem is not valid for staggered fermions, and gauge field topology (whichever way
350
one defines it on a discrete lattice) does not give rise to exact zero modes of the staggered
Dirac operator. At T = 0 no trace of non-trivial gauge field topology on the lowest-lying
spectrum of staggered eigenvalues has been found [42], except in the 2-d Schwinger model
at fairly small lattice spacing [43]. Finite temperature, which causes a depletion of genuine
non-zero eigenvalues near 0, further complicates this issue of a mix-up with would-be
zero modes.
Very recently a study of just the low-lying Dirac operator eigenvalues near Tc has
actually indicated the presence of a gap at large lattice volumes [44]. Again the statistics
was rather limited, and only a small number of the lowest-lying eigenvalues could be
included (varying between 8 and 10). The present study will in many ways follow the
same lines of thought as Ref. [44], but we shall have much larger statistics, and we shall
also probe some different aspects. In particular, we are also interested in seeing whether
quenching causes a different behavior of the smallest eigenvalues near Tc compared with
dynamical fermions.
Last year a study of low-lying eigenvalues of the overlap Dirac operator in quenched
finite temperature gauge theories appeared [45]. Overlap fermions are well suited for
such an investigation since they do not suffer from explicit chiral symmetry breaking
even at finite lattice spacing and since they have exact zero modes in topologically
non-trivial gauge fields. Ref. [45] found that effects of topology persisted in the high
temperature phase, although strongly suppressed compared to the low temperature phase.
More interestingly, an accumulation of low eigenvalues with an apparently finite density
in the infinite-volume limit was found very near 0 even above the (quenched) phase
transition temperature Tc . The statistical properties of the smallest group of eigenvalues
were consistent with them being due to a dilute gas of instantons and anti-instantons [45].
These results have led to the speculation that chiral symmetry might remain broken even
in the high temperature phase of quenched QCD with overlap fermions.
In this paper we describe high statistics investigations of the low-lying eigenvalue
spectrum of the staggered Dirac operator for SU(3) gauge group at finite temperature. We
do this for both quenched (Section 2) and dynamical QCD with four flavors of staggered
fermions (Section 3). The interest in the quenched case lies primarily in checking whether
also staggered fermions, although insensitive to topology at the gauge couplings we can
investigate here, show signs of unusual behavior of the smallest Dirac eigenvalues above
Tc (as was the case for overlap fermions [45]). Both quenched and unquenched simulations
give rise to Dirac operator spectra that can be compared with Random Matrix theory. In
particular, if the Dirac operator spectrum exhibits a gap, is it of the soft-edge kind of
Random Matrix theory? Are there indications that Random Matrix theory provides a more
accurate description of the Dirac operator spectrum as the three-volume is increased? We
shall try to answer these questions in what follows.
351
2. Quenched QCD at high temperature

For our quenched finite temperature Monte Carlo simulations we have used lattices with
temporal extent Nt = 4 and up to three spatial volumes: 83 , 123 and 163 . For Nt = 4 the
deconfinement phase transition for SU(3) pure gauge theory with Wilson action has been
very accurately determined, occurring at lattice gauge coupling c = 5.6925(2) [46]. It
has for long been assumed that the chiral phase transition of the quenched theory occurs
at exactly this deconfinement phase transition point of the pure gauge theory, but this has
recently been challenged [45].
We now give some technical details of our simulations. The gauge field configurations
were generated with a mixture of overrelaxation and heat bath updates, and were analyzed
after every 20th heat bath sweep. On each configuration we computed the 50 lowest-lying
eigenvalues, and in some cases even more, using the variational Ritz functional method [47,
48]. Many ensembles consisted of several thousand configurations. Gauge coupling, lattice
size and statistics of our ensembles is summarized in Table 1.
Table 1
Details of our quenched ensembles
V
#configurations
83 4
83 4
83 4
83 4
83 4
83 4
83 4
123 4
163 4
83 4
83 4
83 4
123 4
163 4
5.5
5.66
5.695
5.71
5.72
5.73
5.75
5.75
5.75
5.8
5.85
5.9
5.9
5.9
5931
5544
1769
1145
1426
2021
7029
1933
692
5335
6149
4263
1755
824
As is well-known, the staggered Dirac operator

1X
D
/ x,y =
(x) U (x)x+,y U (y)x,y+
2
/ e,o + D
/ o,e
D
(7)
is anti-hermitian, with purely imaginary eigenvalues i that come in pairs of opposite sign.
In Eq. (7)
(x) = (1)
< x
(8)
352
are the usual phase factors for staggered fermions. Denoting

(x) = (1)
(9)
/ connects even sites, i.e., those with (x) = +1,

we have also explicitly shown how D
/2
with odd sites, those with (x) = 1, and vice versa. This means that the operator D
is hermitian and positive semi-definite. The sign function (x) defined above plays the
/ : {D,
/ } = 0. As D
/ 2 does not mix
role of 5 in the continuum: it anticommutes with D
between even and odd lattice sites, we need only compute the eigenvalues, on, say, the even
/ 2 with eigenvalue
sublattice. One easily sees that if e is a normalized eigenvector of D
1
2
2
/ o,e e is a normalized eigenvector of D
/ with eigenvalue 2 , and non , then o D
zero only on odd sites. Moreover, as we will never encounter exact zero modes on genuine
quantum configurations, there is no difficulty with the above definition of o . We of course
/2
make use of these properties, and hence compute only the (positive) eigenvalues of D
restricted to the even sublattice, and then take the (positive) square root. All eigenvalues to
be shown in the following thus have an equal number of negative companions, of the exact
same magnitude.
The spectral density of the Dirac operator is given by
*
+
1 X
( n ) ,
(10)
()
V n
and it is readily computed numerically from our Monte Carlo simulations by a binning of
the measured eigenvalues per configuration at convenient small intervals.
In Fig. 1 we show the density of low lying eigenvalues on 83 4 lattices for several
values between 5.5 (in the confined phase) to 5.9 (in the deconfined phase). In each case we
computed the 50 lowest positive eigenvalues i of the staggered Dirac operator. The plots
R
are normalized by the following condition, 0 () d = #eigenvalues/V . In the confined
phase it is quite evident that the density at zero would be non-zero in the thermodynamic
limit. As the temperature is increased, the density at zero decreases. There is a qualitative
change of the eigenvalue density once the temperature is increased above Tc . Beyond the
first few eigenvalues the eigenvalue density assumes a shape compatible with a square
root behavior (4). However, a sizable tail, decreasing with increasing temperature, of small
eigenvalues persists, which seems to extend all the way down to zero.
In Fig. 2 we compare the eigenvalue density in the high temperature phase, at = 5.9,
for several spatial volumes. As can be seen the eigenvalue density is volume independent
to a surprising degree. In particular, the tail of small eigenvalues is seen to be volume
independent and appears to persist in the thermodynamic limit. We note that the tail is
much larger than would be compatible with the exponential tail from the Airy function
behavior of random matrix theory at a soft edge Eq. (6) (see Fig. 9 in Section 3 below for
an example).
It is tempting to speculate that the tail seen here in the quenched case with staggered
fermions is a reflection of the accumulation of small eigenvalues seen with overlap
fermions in [45] and attributed there to a dilute gas of instantons and anti-instantons. In the
353
Fig. 1. The measured distribution of the 50 lowest eigenvalues in quenched QCD at finite temperature
for several s on 83 4 lattices. One sees a clear suppression of the spectral density at the origin
(0) as the temperature (coupling) is increased, indicating a diminishing chiral condensate according
to the BanksCasher formula. For large temperatures the spectrum is heavily suppressed near the
origin, although in this quenched case a small tail always seems to extend towards = 0 (see below).
staggered case, the explicit chiral symmetry breaking at finite lattice spacing shifts the
small modes, presumably resulting in the observed tail.
We also looked at the unfolded level spacing between individual eigenvalues i and i + 1
si =
i+1 i
.
hi+1 i i
(11)
In Fig. 3 the distribution of the si between eigenvalue i and i + 1, i = 1, . . . , 6 is shown

and compared to the expected distribution, the Wigner surmise
P (s) =
32 2 4 s 2
s e .
2
(12)
Evidently, the level spacings between the first eigenvalues are not very accurately given by
Random Matrix theory correlations. But as we move into the bulk, say for i > 5 in the case
of the 83 4 lattice, the agreement with Random Matrix theory becomes almost perfect.
It should be stressed here that this is a volume-dependent statement. For larger volumes
one needs to go beyond more eigenvalues starting at the soft edge before one finds good
agreement. This is consistent with the fact that there is a definite scale around the soft edge,
where eigenvalue correlations are poorly described by Random Matrix theory. Going to
larger volumes simply forces more eigenvalues into that region.
354
Fig. 2. The density of the smallest eigenvalues in the high temperature phase of quenched QCD, at
= 5.9, for spatial volumes 83 , 123 and 163 . In the two large volumes we measured 100 eigenvalues
and only 50 in the small volume. Since the eigenvalue density increases with increasing volume, the
distributions are correspondingly cut off at smaller . There is no observable change in the eigenvalue
density as the three-volume is increased. This holds even at the very small tail extending towards
= 0, as shown in the magnified plot inserted.
3. QCD with four dynamical fermions at high temperature

For our dynamical simulations we chose to work with nf = 4 flavors. Four is the
natural number of flavors for staggered fermions in the continuum limit, and an efficient
exact simulation algorithm can be used, the Hybrid Monte Carlo algorithm. In addition,
the four flavor theory is known to have a rather strong first order finite temperature
phase transition [49]. Therefore, tunnelings into the low temperature phase are strongly
suppressed already on rather small systems and at temperatures quite close to Tc .
We made dynamical simulations with quark masses amq = 0.1, 0.05, 0.025, 0.01 and
0.002, and for couplings in the high temperature phase (here a is the lattice spacing).
Interestingly, the lowest eigenvalue provides a sensitive method for determining the critical
coupling c (amq ): in Fig. 4 the lowest eigenvalue on a 83 4 lattice with amq = 0.025 is
plotted against . ranges from 4.96 to 5.07 in intervals of 0.002. Each point is an average
over 5 configurations at that -value. A rise is observed between = 5.012 and = 5.022,
which can be clearly interpreted as the chiral symmetry restoring finite temperature phase
transition. (For published values of the critical couplings c (amq )) see Ref. [49].) The
possibility of using the magnitude of the smallest Dirac operator eigenvalues as probes for
chiral symmetry restoration was suggested by Jackson and Verbaarschot [3436] on the
basis of a similarly observed behavior in a Random Matrix theory context that we will also
discuss below.
355
Fig. 3. Distributions of the unfolded level spacings between the first few eigenvalues for = 5.75
on an 83 4 lattice, compared to the Wigner surmise. The agreement with Random Matrix theory
is clear once we go beyond the first few eigenvalues. However, there is not very good agreement for
the first 23 eigenvalues at this volume.
Fig. 4. The lowest eigenvalue, averaged over 5 configurations, on an 83 4 lattice with amq = 0.025
as a function of . One sees a clear change in behavior around 5.02, indicating the restoration
of chiral symmetry around this temperature. Just this first Dirac operator eigenvalue can thus serve
as an excellent indicator of the chiral phase transition point.
356
Most of our analysis with dynamical fermions was performed at = 5.2, which is above
c for all values of mq we used. Typically, we analyzed 3000 configurations for each
volume and mq , extracting the 50 smallest eigenvalues. Ensemble details are summarized
in Table 2. Note the large statistics gathered for = 5.2 and amq = 0.025.
Table 2
Details of our ensembles with dynamical fermions
V
amq
#configurations
83 4
83 4
83 4
83 4
83 4
83 4
123 4
83 4
123 4
83 4
83 4
83 4
83 4
83 4
5.1
5.1
5.1
5.2
5.2
5.2
5.2
5.2
5.2
5.2
5.2
5.2
5.3
5.4
0.1
0.025
0.01
0.1
0.05
0.025
0.025
0.01
0.01
0.008
0.005
0.002
0.025
0.025
8897
2909
4050
4348
2149
29632
13929
4050
2882
4200
7948
6950
4549
3746
As in the quenched case we find a small tail at the lower end of the eigenvalue
distribution, reaching towards = 0 from the main bulk of the distribution. This tail also
appears to be volume independent (see Fig. 6). However, in this case it is somewhat more
suppressed than in the quenched case, and for small values of the quark mass it does not
extend all the way to = 0, see Fig. 5.
In Fig. 7 we show eigenvalue distributions measured at = 5.2 and various amq . Note
that for larger amq , c is also larger, which causes a shift of the whole spectrum as amq
is varied. Therefore, a direct quantitative comparison of the distributions in Fig. 7 is not
straightforward. However, we note that the distributions at amq = 0.025 and amq = 0.002
are almost on top of each other, indicating that these distributions are very close to the
amq = 0 distribution.
Furthermore, as in the quenched case described in the previous section, we find the
macroscopic behavior of the eigenvalue density seemingly compatible with a square
root form. By fitting Eq. (3) to the bulk of the spectrum (leaving out the tail), we obtain
different powers, depending on how much of the tail we choose to cut off. This fitting has
been done for V = 83 4 at = 5.2 and amq = 0.025 and the resulting values can be seen
in Table 3. There is a clear tendency towards a slightly higher power (= 2m + 1/2) than the
square root which would be expected from Random Matrix theory. In Fig. 8 we show the
fit with the cut at 0.1. For comparison, we include the distribution of the first eigenvalue in
357
Fig. 5. Quenched spectrum of = 5.9 and a dynamical spectrum at = 5.4 with amq = 0.025, both
on an 83 4 lattice. A blowup of the y-axis near the end of the tail is also shown. There is a much
stronger suppression of small eigenvalues in the case of dynamical fermions, perhaps due to fewer
would be zero modes in topological non-trivial gauge field configurations.
Fig. 6. The density of the 50 smallest eigenvalues in the high temperature phase, = 5.2 and
amq = 0.025 for spatial volumes 83 and 123 . The normalization of the distributions is as in Fig. 2.
As in the quenched simulations, we observe no significant change in the eigenvalue spectrum at all
as the three-volume is increased.
358
Fig. 7. The density of the 50 smallest eigenvalues at = 5.2 for quark masses
amq = 0.1, 0.05, 0.025, 0.002 on 83 4 lattices. While there is some shift with decreasing quark
mass for the larger values of amq , the spectral density seems to approach a limiting function, which
we can identify as the chiral limit of the spectrum. Note that the large-mass behavior of the spectrum
is almost identical to the quenched spectrum in this region, in agreement with the notion that on
this scale of eigenvalues the large-mass fermion has completely decoupled, and effectively become
quenched.
the figure. Clearly, the tail at small is caused by the excessive width of the distribution of
the smallest eigenvalue.
In Fig. 9 we compare the spectral density for quark mass amq = 0.025 with the
prediction (5) of the Random Matrix theory for the density near a soft edge. In order to
make the comparison possible, we rescale the spectral density as

2/3
2/3
2
2
2
0
( 0 )
,
(13)
()
2 0 KV
0
0 KV
where we determine 0 and K from a fit to a square root: 0 /2K ( 0 ). Here V is

the lattice volume. We see that the tail predicted by the Random Matrix theory is much
more strongly suppressed than the measured distribution. Moreover, we can also observe
that the density of the eigenvalues themselves does not match: the measured distribution
includes 50 eigenvalues, whereas the function (5) has only 40 wiggles in the -range
of the distribution. However, if we attempt to perform the comparison by matching the
number of eigenvalues/wiggles, the overall fit becomes worse. In itself, this mismatch is
not surprising, when we remember that the overall shape is not well described by a square
root behavior in the first place (see Table 3).
In the quenched simulations we saw that correlations between the first 5 or 6 eigenvalues
were not very accurately given by Random Matrix theory. In Fig. 10 the distribution of si ,
359
Fig. 8. Power fit to the spectral density for = 5.2 with amq = 0.025 on an 83 4 lattice with cut
at 0.1 (see Table 3). Also included is the distribution of the first eigenvalue. The distribution of this
smallest eigenvalue is very different from that demanded by the Airy-function behavior (see below).
Fig. 9. Comparison of the spectral density for = 5.2, amq = 0.025 on a 83 4 lattice, with the
Random Matrix theory prediction for a soft edge, as in Eq. (5).
360
Table 3
The dependence of the fitted power on the where the small- tail is cut off. The data are from
83 4, = 5.2 and amq = 0.025 lattices, using the 50 smallest eigenvalues. The fits are almost
compatible with a square-root behavior of the spectrum at this soft edge, but there is a consistent
small upward shift in the exponent
Cut at
0.09
0.095
0.1
0.105
0.11
0.115
0.12
0.125
0.13
0.14
2m + 1/2
0.055 0.003
0.049 0.004
0.040 0.005
0.036 0.006
0.041 0.008
0.035 0.010
0.043 0.013
0.040 0.016
0.033 0.021
0.027 0.034
0.609 0.006
0.598 0.007
0.580 0.010
0.572 0.011
0.582 0.016
0.570 0.019
0.585 0.027
0.581 0.032
0.567 0.041
0.554 0.068
defined in Eq. (11), is shown for a dynamical simulation with = 5.2 , amq = 0.025 on a
83 4 lattice. Here we see, on the same lattice volume, a clear deviation from the Wigner
surmise only for the first two or three spacings, s1 , s2 and s3 . Again, comparisons can only
be made at equal volumes, as larger volumes imply more eigenvalues in the region around
the soft edge where correlations are poorly described by Random Matrix theory.
There is an interesting physical consequence of a genuine gap in the Dirac eigenvalue
spectrum. Consider the difference between the ( ) susceptibility P and the susceptibility
of its scalar partner (a0 ), which we denote by S [50,51]. In a manner similar to the Banks
Casher formula for the chiral condensate one finds that this difference can be written
Z
(; m)
2
d 2
,
(14)
= 4m
( + m2 )2
0
where the spectral density (; m) includes the zero modes due to topology as well. In
the infinite-volume limit this contribution from exact zero modes should vanish, and we
should be left with the integral over non-zero modes. If the spectral density has a gap
around the origin this means that vanishes in the chiral limit. Because axial U(1) rotates
P into S and vice versa, this would imply a restoration of axial U(1) at these high
temperatures [50,51]. Conversely, if one believes that this axial U(1) symmetry is not
restored at high temperature, one gets constraints on the behavior of the spectral density of
the Dirac operator near the origin. For a power-law behavior near 0 a non-vanishing
in the infinite-volume chiral limit can only be supported for (, 0) with 6 1, and
in fact would only remain finite in that limit if = 1.
Of course, the fact that we appear to see a gap in the eigenvalue spectrum with staggered
fermions at this particular coupling does not imply that this gap persists in physical units
as we take the continuum limit. If not, we are here studying a pure lattice artifact. The
only way to test this is to study the shift in the apparent cut-off eigenvalue 0 as we
361
Fig. 10. Distributions of the unfolded level spacings between the first few eigenvalues in the
dynamical case, compared to the Wigner surmise. As in the quenched case, we find a small
disagreement for the first eigenvalues. But in contrast to the quenched case the agreement sets in
after almost just the 2nd eigenvalue, at the same volume.
change the lattice spacing (this is, however, far beyond what we can do at present). A very
different uncertainty comes from the fact that we are working with fermions that are almost
insensitive to topology at the couplings and volumes available to us. This means that there
are some would be zero modes mixed up with our regular eigenvalues near the origin.
At zero temperature we found that these would-be zero modes somewhat surprisingly
behaved as the non-zero eigenvalues at realistic couplings and volumes [42]. But there
is no guarantee that this is the case at finite temperature. This means that the small tail
extending to zero in our quenched simulations, and some of the smallest eigenvalues in
the simulations with dynamical fermions may be due to these would-be zero modes. In
particular, the disagreement with Airy-function behavior very close to the soft edge may
be partly due to these would-be zero modes.
4. Random Matrix theory

We have just shown our lattice gauge theory data for the spectrum of the Dirac operator
below, around, and above Tc , and compared it with some of the analytical formulas of
Random Matrix theory in the limit where the size of these matrices N goes to infinity. It is
of interest to see how a simple model of a chiral phase transition in Random Matrix theory
behaves at finite N , as some external parameter (mimicking temperature T ) is changed.
The model we shall focus on here was proposed by Jackson and Verbaarschot in Ref. [34]
362
and has also been studied in Ref. [52]. It is based on a modified chiral ensemble of N N
complex matrices W , with partition function
Z
Z=
dW
Nf
Y
det(M imf ) exp[N Tr(W W )]
(15)
f =1
in a sector of topological charge zero. Here

0
W + T
M=
W +T
0
(16)
is a 2N 2N block hermitian matrix, and the external deterministic parameter T is playing

a rle reminiscent of temperature. In the normalization chosen here, a continuous phase
transition occurs at Tc = 1, where the spectral density () of the eigenvalues of the
random matrices just reaches zero [3436]. For T > Tc the spectrum becomes two-banded,
with a gap surrounding zero. The global shape of () in this model will be very different
from the actual (macroscopic) Dirac operator spectral density. But the interesting feature
lies in having here a simple model which qualitatively seems to describe some of the
observed behavior of the staggered Dirac operator with Nf = 4, in particular the apparent
gap in the density for T > Tc . 2
It is very simple to do quenched (i.e., Nf = 0) numerical simulations of the above
Random Matrix model, as it just corresponds to generating an ensemble of complex
matrices with Gaussian weight. Such matrices are maximally random in that their matrix
elements are independently of Gaussian distribution. This allows us to choose very large
random matrices numerically, and then finding the eigenvalues of the hermitian matrix M.
In this way we have studied the detailed behavior of the finite-N Random Matrix model
(15) just below Tc , at Tc , and just above Tc (where the two bands separate as the gap
develops). 3
We show some of our numerical results in Figs. 11 and 12. First, in Fig. 11 we display a
sequence of the macroscopic spectral density with increasing magnitude of the parameter
T : T = 0, 0.5, 0.75, 1.0, 1.2, 2.0. The plots were made by diagonalizing 10000 200
200 matrices. At T = 0 this macroscopic Random Matrix theory spectrum is just the
Wigner semi-circle law (we display only the > 0 part):
1 p
4 2 .
(17)
(, T = 0) =
2
As T increases, a dip in the spectral density around = 0 slowly develops, and it
subsequently turns into a gap. Of course, this macroscopic Random Matrix theory spectrum
is totally unlike the macroscopic Dirac operator spectrum. But it is interesting to zoom in
on the microscopic behavior of the Random Matrix spectral density at the soft edge of the
2 Many other details do not match at all. For instance the phase transition in the model (15) is continuous, in
contrast to the finite-temperature phase transition in the massless Nf = 4 theory, which is believed to be of first
order.
3 Similar numerical studies have been performed by K. Splittorff, M.Sc. thesis, The Niels Bohr Institute, 1999
(unpublished). Simulations of the macroscopic spectral density in this model can also be found in the original
paper by Jackson and Verbaarschot [34].
363
Fig. 11. Full spectra of 200 200 matrices at T = 0, 0.5, 0.75, 1.0, 1.2, 2.0. In this simple model a
gap indeed develops around = 0 at T = 1.
Fig. 12. A blowup of the graphs in Fig. 11. The microscopic spectrum changes smoothly from
Bessel-type behavior to Airy-type behavior, going through an intermediate critical distribution at
T = 1 corresponding to a macroscopic power-law behavior of type () 1/3 . The exact analytical
curve for the microscopic spectral density at T = 1 has been computed in Ref. [32].
gap. In this way, by blowing up the scale of the smallest eigenvalues, we obtain the plots
in Fig. 12 for the same parameter values of T as above. One sees clearly how the universal
Bessel-kernel behavior below Tc turns into the also universal Airy-kernel above Tc . We
364
Fig. 13. A blowup of a T = 3 spectrum compared with the Airy-prediction. There is nice agreement
near the edge.
note that it is has been shown in Ref. [3436] that the massless microscopic spectral density
of the above Random Matrix model has precisely the usual zero-temperature form,
s ( (T )) =
2

(T )
JNf (T ) JNf 1 (T ) JNf +1 (T ) ,
2
(18)
where (T ) is simply the eigenvalues rescaled by the (T -dependent) infinite-volume

spectral density at the origin (0, T ):
(T ) = 2N(0, T ).
In this model (0) approaches zero with a mean-field type of behavior [3436]:
p
(0, T ) = (0, 0) 1 T 2 .
(19)
(20)
For T bigger than Tc , but still close to it, we find, as expected, a deformation of the
Airy-kernel. In fact, the microscopic behavior there smoothly interpolates between the
Bessel-form and the Airy-form. The peaks corresponding to individual eigenvalues from
the Bessel-function behavior below Tc gradually smoothen out to become the inflection
points in the spectral density of the Airy-kind as the soft edge moves away from the origin.
To illustrate how accurately one reproduces the Airy-behavior in this kind of simulations,
we show in Fig. 13 the soft edge prediction appropriately scaled to fit simulation data at
T = 3. The Airy-behavior is perfectly reproduced close to the edge, but with deviations
after the first few eigenvalues (wiggles). The deviation is presumably caused by the limited
number of eigenvalues at N = 300, and when N is increased we expect the fit to improve.
365
5. Conclusions
The purpose of this study has been to find the extent to which Random Matrix theory
may be able to describe low-lying eigenvalue distributions and correlations between lowlying eigenvalues of the staggered Dirac operator at finite temperature. We have also been
interested in testing the quenched theory with staggered fermions in the light of recent
results with overlap fermions [45], which indicated that the chiral finite-temperature phase
transition in the quenched theory may be more subtle than previously expected. In the
quenched case we do not see the accumulation of small Dirac operator eigenvalues around
= 0 that was observed with overlap fermions. This is not entirely surprising in view of
the insensitivity of staggered fermions to gauge field topology at these lattice couplings
and lattice volumes [42,43]. With staggered fermions we do observe a strong depletion
of eigenvalues near the origin once the temperature T reaches the pure gauge theory
deconfinement phase transition temperature Tc . This is in agreement with the conventional
picture that chiral symmetry is restored in the quenched theory with staggered fermions at
precisely the deconfinement phase transition.
On the other hand, we find a clear difference in the behavior of the smallest Dirac
operator eigenvalues in the quenched theory and the theory with genuine, dynamical,
fermions. In the quenched case a small, volume independent tail of the eigenvalue
distribution extends to = 0 while in the full theory the tail does not reach the origin at
the couplings we have investigated. While the bulk of the eigenvalue distribution near the
(soft) edge is roughly compatible with a square root behavior, the tail of small eigenvalues
is considerably larger than the Airy function behavior that Random Matrix theory would
predict. Physically, a genuine gap in physical units in the eigenvalue spectrum above Tc
would imply the restoration of axial U(1) symmetry at these high temperatures. A very
likely scenario is therefore that the apparent gap found with staggered fermions at finite
bare coupling shrinks to zero in physical units as the continuum limit is reached.
However, an investigation of whether this is indeed the case is much beyond the scope
of the present paper.
Acknowledgements
We thank K. Splittorff and J. Verbaarschot for discussions. The work of P.H.D. and K.R.
has been partially supported by EU TMR grant no. ERBFMRXCT97-0122, and the work
of U.M.H. has been supported in part by DOE contracts DE-FG05-85ER250000 and DEFG05-96ER40979. In addition, P.H.D. and U.M.H. acknowledge the financial support of
NATO Science Collaborative Research Grant no. CRG 971487.
References
[1] M.A. Halasz, J.J.M. Verbaarschot, Phys. Rev. Lett. 74 (1995) 3920.
[2] R. Pullirsch, K. Rabitsch, T. Wettig, H. Markum, Phys. Lett. B 427 (1998) 119.
366
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
M.E. Berbenni-Bitsch et al., Phys. Lett. B 438 (1998) 14.

J. Ma, T. Guhr, T. Wettig, Eur. Phys. J. A 2 (1998) 87.
T. Guhr, J.Z. Ma, S. Meyer, T. Wilke, Phys. Rev. D 59 (1999) 054501.
B.A. Berg, H. Markum, R. Pullirsch, Phys. Rev. D 59 (1999) 097504.
R.G. Edwards, U.M. Heller, J. Kiskis, R. Narayanan, Phys. Rev. Lett. 82 (1999) 4188.
E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 (1993) 306.
J.J.M. Verbaarschot, I. Zahed, Phys. Rev. Lett. 70 (1993) 3852.
J.J.M. Verbaarschot, Phys. Lett. B 329 (1994) 351; Phys. Rev. Lett. 72 (1994) 2531; Nucl.
Phys. B 426 (1994) 559.
G. Akemann, P.H. Damgaard, U. Magnea, S. Nishigaki, Nucl. Phys. B 487 (1997) 721.
P.H. Damgaard, S.M. Nishigaki, Nucl. Phys. B 518 (1998) 495.
M.K. Sener,
J.J.M. Verbaarschot, Phys. Rev. Lett. 81 (1998) 248.

T. Nagao, S.M. Nishigaki, hep-th/0001137, hep-th/0003009.
G. Akemann, E. Kanzieper, hep-th/0001188.
M.E. Berbenni-Bitsch, S. Meyer, A. Schfer, J.J.M. Verbaarschot, T. Wettig, Phys. Rev. Lett. 80
(1998) 1146.
M.E. Berbenni-Bitsch, S. Meyer, T. Wettig, Phys. Rev. D 58 (1998) 071502.
P.H. Damgaard, U.M. Heller, A. Krasnitz, Phys. Lett. B 445 (1999) 366.
M. Gckeler, H. Hehl, P.E.L. Rakow, A. Schfer, T. Wettig, Phys. Rev. D 59 (1999) 094503.
R.G. Edwards, U.M. Heller, R. Narayanan, Phys. Rev. D 60 (1999) 077502.
F. Farchioni, I. Hip, C.B. Lang, M. Wohlgenannt, Nucl. Phys. B 549 (1999) 364.
P.H. Damgaard, Phys. Lett. B 424 (1998) 322.
G. Akemann, P.H. Damgaard, Nucl. Phys. B 519 (1998) 682; Phys. Lett. B 432 (1998) 390.
S.M. Nishigaki, P.H. Damgaard, T. Wettig, Phys. Rev. D 58 (1998) 087704.
J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 540 (1998) 317.
P.H. Damgaard, J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 547 (1999) 305.
D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 560 (1999) 259.
P.H. Damgaard, K. Splittorff, hep-th/9912146; hep-lat/0003017.
P.H. Damgaard, hep-lat/0001002.
H. Leutwyler, A. Smilga, Phys. Rev. D 46 (1992) 5607.
E.T. Tomboulis, L.G. Yaffe, Phys. Rev. Lett. 52 (1984) 2115.
R. Janik, M.A. Nowak, I. Zahed, Phys. Lett. B 392 (1997) 155; Phys. Lett. B 446 (1999) 9.
A.D. Jackson, J.J.M. Verbaarschot, Phys. Rev. D 53 (1996) 7223.
A.D. Jackson, M.K. Sener, J.J.M. Verbaarschot, Nucl. Phys. B 479 (1996) 707.
T. Guhr, T. Wettig, Nucl. Phys. B 506 (1997) 589.
M.J. Bowick, E. Brzin, Phys. Lett. B 268 (1991) 21.
E. Kanzieper, V. Freilikher, Phys. Rev. E 55 (1997) 3712.
P.J. Forrester, Nucl. Phys. B 402 (1993) 709.
J.B. Kogut, J.-F. Laga, D.K. Sinclair, Phys. Rev. D 58 (1998) 054504.
J. Smit, J.C. Vink, Nucl. Phys. B 286 (1987) 485.
P.H. Damgaard, U.M. Heller, R. Niclasen, R. Rummukainen, Phys. Rev. D 61 (2000) 014501.
F. Farchioni, I. Hip, C.B. Lang, Phys. Lett. B 471 (1999) 58.
F. Farchioni, Ph. de Forcrand, I. Hip, C.B. Lang, K. Splittorff, hep-lat/9912004.
R.G. Edwards, U.M. Heller, J. Kiskis, R. Narayanan, hep-lat/9910041.
J. Fingberg, U.M. Heller, F. Karsch, Nucl. Phys. B 392 (1993) 493.
B. Bunk, K. Jansen, M. Lscher, H. Simma, DESY report (September 1994).
T. Kalkreuter, H. Simma, Comp. Phys. Comm. 93 (1996) 33.
F.R. Brown et al., Phys. Lett. B 251 (1990) 181.
C. Bernard et al., Phys. Rev. Lett. 98 (1997) 253.
S. Chandrasekharan et al., Phys. Rev. Lett. 82 (1999) 2463.
367
[52] T. Wettig, A. Schfer, H.A. Weidenmller, Phys. Lett. B 367 (1996) 28; Phys. Lett. B (Erratum) 374 (1996) 362.

Percolation and magnetization in the continuous

spin Ising model
Piotr Bialas a,b , Philippe Blanchard a , Santo Fortunato a ,
Daniel Gandolfo a,c, , Helmut Satz a
a Fakultt fr Physik, Universitt Bielefeld, D-33615, Bielefeld, Germany
b Institute of Comp. Science, Jagellonian University, PL-30-072 Krakow, Poland
c Dpartement de Mathmatiques, Universit de Toulon et du Var, F-83957 La Garde Cedex, & CPT, CNRS,
Luminy, case 907, 13288 Marseille Cedex 09, France

Abstract
In the strong coupling limit the partition function of SU(2) gauge theory can be reduced to that
of the continuous spin Ising model with nearest neighbour pair-interactions. The random cluster
representation of the continuous spin Ising model in two dimensions is derived through a Fortuin
Kasteleyn transformation, and the properties of the corresponding cluster distribution are analyzed.
It is shown that for this model, the magnetic transition is equivalent to the percolation transition of
FortuinKasteleyn clusters, using local bond weights. These results are also illustrated by means of
numerical simulations. 2000 Elsevier Science B.V. All rights reserved.
PACS: 05.50.+q; 11.15.Ha; 75.10.H
Keywords: Phase transition; Gauge theories; Percolation; Wolff algorithm; FortuinKasteleyn transformation
1. Introduction
It has recently been proposed that the deconfinement transition in SU(2) gauge theory
can be characterized as percolation of Polyakov loop clusters [1]. This idea is based on
the fact that SU(2) gauge theory and the Ising model belong to the same universality
class [2], and that the magnetization transition in the Ising model can be specified as
percolation of clusters defined through local bond weights [3]. It thus seems natural that the
same holds for the corresponding Polyakov loop clusters in SU(2) gauge theory, provided
suitable bond weights can be defined. In [1], it was shown that in a lattice formulation
effectively corresponding to the strong coupling limit, SU(2) gauge theory indeed leads to
the predicted Ising critical exponents.
Corresponding author. E-mail: gandolfo@cpt.univ-mrs.fr
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 3 2 - 1
P. Bialas et al. / Nuclear Physics B 583 (2000) 368378
369
In the present paper, we want to consider the classical continuous spin model on Z2
introduced by Griffiths [4], an Ising model with spins taking values continuously between
1 and +1, and prove that in this case magnetization and percolation transitions coincide.
We also present a detailed numerical study of the model with simulations on the 2d square
lattice.
For SU(2) lattice gauge theory it was shown in the strong coupling limit [5] that the
partition function can be written in a form which, apart from a factor which depends
on the group measure, is the partition function of the classical spin model, with spins
varying continuously in some bounded real interval [4]. These continuous spins thus
are the natural counterparts of the Polyakov loops in SU(2) gauge theory. Following
the pioneering work [3], the relationship between the thermodynamical features of
spin models and the properties of the corresponding geometrical clusters have in the
past years received considerable attention (see [611]). We want to address here the
random percolation content of the continuous spin Ising model [4] obtained through
a FortuinKasteleyn transformation [3] of the partition function. A specific class of
geometrical clusters [12] shows a percolation transition whose critical behaviour matches
the conventional Ising counterparts. In particular, it can be shown that up to finite constants,
the spin magnetization equals the probability of long range cluster connectivity; moreover,
the spin susceptibility becomes equal (up to the same constants) to some mean cluster
size [13].
The numerical method to be used here is the Wolff algorithm [12]. It is based on the
formulation of a single cluster update algorithm for spin systems with unprecedented
performances regarding the problem of critical slowing down behaviour near phase
transition [12]. It turns out that the Wolff clusters exhibit critical percolation behaviour at
the point of the Ising phase transition. Very recently, this specific property of Wolff clusters
has been put into a rigorous framework [10], leading to a concept of Wolff measures which
provides the theoretical basis for the method to be used below. We shall extend this work to
the 2-D continuous spin model and corroborate the results through numerical simulations.
2. The model
Let Z2 be a finite lattice with edge set E and vertex set V = {1, . . . , N} and S
a random configuration of spin variables S = {Si } V , where the Si, iV take values
independently in = [1, +1]. Consider the Hamiltonian
X
Ji,j Si Sj ,
(2.1)
H (S) =
hi,j i
where the sum is over all nearest neighbour pairs (hi, j i) of spins and Ji,j > 0. Rewriting
the spin variables Si = li i , where i = sign(Si ) = 1 and li = |Si | [0, 1], one has
X
Ji,j li lj i j .
(2.2)
H (S) = H ( , l) =
i,j
370
2.1. Joint distribution over spin amplitudes and bond variables

Following Edwards and Sokal [8], the random cluster representation (RCR) [3,9] of this
model is derived in the following way. Let n {0, 1}E be a configuration of bond variables
on where,
nij = 1 means that there is a bond between sites i and j in V ,
nij = 0 is the opposite event.
Consider the following joint probability distribution of the random variables (, l, n),
Y
1
eJi,j li lj [pi,j nij,1 i j + (1 pij )nij,0 ]
P(S, n) = P(l, , n) = Z
hi,j i
1
= Z
w,J (l, , n),
(2.3)
hi,j i
where pi,j = 1 e2Ji,j li lj is a weighting factor to be interpreted as the probability for

a link nij to exist (nij = 1) between two nearest neighbour spins Si and Sj that share the
same sign (i j = 1), and (1 pi,j ) the probability of having no link (nij = 0) between
them. Z is the partition function of the model given by
"
#
XZ Y
X
dfi (li ) exp
Ji,j li lj i j ,
(2.4)
Z =
{ }
hi,j i
where dfi is the probability distribution of the spin amplitudes li and is the inverse
temperature.
Starting with the joint distribution (2.3), it is straightforward to express the weights
w,J (l, , n) in the following way
Y
Y
Y
eJi,j li lj
pi,j i j
(1 pij ).
w,J (l, , n) =
hi,j i
hi,j i,
nij =1
hi,j i,
nij =0
Now, we write n for a bond configuration fulfilling the compatibility condition that a bond
the number of clusters
n i,j exists between sites i and j (n i,j = 1) iff i = j and call c(n)
of bonds in the configuration n.
Then, the weights w,J take the form (FK-representation)
!
Y
Y
Y
Ji,j li lj
c(n)
=
e
pi,j
(1 pij ) .
(2.5)
2
w,J (l, n)
hi,j i
hi,j in
hi,j i
/ n
These weights are called the Wolff weights of the random cluster distribution. The joint
distribution on the configurations (l, n)
provide the right framework to relate the critical
behavior in the original spin system to its associated percolation representation.
differ from the (FK) weights of the Ising case
Remark 2.1. The Wolff weights w,J (l, n)
Q
J
l
l
i,j
i
j
that reflects the measure on the spin amplitudes in
(see [3,7]) by the term hi,j i e
the continuous case.
371
3. Wolff cluster algorithm and distributions

Non-local cluster Monte Carlo (MC) algorithms have brought significant improvements
in the simulation of Ising models near criticality. Starting from the ground-breaking
work of Fortuin and Kasteleyn [3], which relates the partition function of spin systems
with that of a correlated percolation model, Swendsen and Wang [6] derived a non-local
cluster algorithm which drastically reduces the critical slowing down phenomena near the
transition point. It seems well established [14] that, for SwendsenWang dynamics, the
dynamical critical exponent z defined by
= z
(3.1)
should be equal to zero for the 2-D Ising model. Here is the autocorrelation time in MC
simulations (measured in MC steps per site) and is the correlation length. In the case of
a local update algorithm like Metropolis or Heatbath, z is found to be close to 2.
Building on the SwendsenWang idea, Wolff [12] improved the method (see below)
and derived a non-local update MC scheme with a dynamical critical exponent z equal
to SwendsenWangs but with smaller autocorrelation time [15] and furthermore easier to
implement.
3.1. The Wolff algorithm
First let us recall briefly the main features of the Wolff method. Consider the 2-D
ferromagnetic Ising model on Z2 with coupling constant J and inverse temperature .
Starting from a randomly chosen spin 0 , visit all nearest neighbours and
with probability pb = 1 exp(2J ),
(III.1)
and only if they have the same orientation as 0 ,
include them in the same cluster as 0 ; spins not satisfying both conditions are excluded.
Repeat iteratively this procedure with newly added spins in the cluster until no more
neighbours fulfill the above compatibility condition (III.1). Now flip all the spins in that
cluster with probability 1. After that, erase all the bonds and start this procedure again. It
turns out that this dynamics verifies the detailed balance condition, i.e., it samples the Gibbs
distribution of the Ising model (see [12]). The distinguishing feature of the Wolff method
compared to SwendsenWangs is that, in the latter (following Fortuin and Kasteleyn [3])
one needs to build, with the same growing probability pb as before, all possible clusters
of like spins and then, with probability 1/2, flip all the spins in those clusters. Then all
the bonds are erased and one starts again from the newly created spin configuration. It can
also be shown that this method verifies the detailed balance condition [6]. In summary, the
following remarks can be made:
The building procedure of an individual cluster in both methods is clearly the same,
thus the Wolff cluster belongs to the set of SwendsenWang clusters.
When the Wolff cluster is built, the randomly chosen spin has obviously higher
probability to fall in a large SwendsenWang cluster than in a smaller one.
372
It results that the distribution of Wolff clusters is given by the distribution of Swendsen
Wang clusters (see below) modified by an additional weight that takes into account the size
of the clusters.
Remark 3.1. A (rather non economic) way of building the Wolff cluster would be to
construct the set of all N SwendsenWang bond clusters CiSW , i {1, . . . , N }, then to
pick at random a lattice spin x and flip with probability 1 all the spins belonging to the
cluster C SW (x ). Obviously, the randomly chosen spin x has higher probability to sit in a
large cluster.
Remark 3.2. In the case of our model, the bond probability pb (Si , Sj ) =
1 exp(2Ji,j li lj ) takes also into account the spin amplitudes. It turns out that, the
Wolff dynamics (as described above) is no longer ergodic (because spin amplitudes are
locally conserved) and must be supplemented by a local update method (like Metropolis or
Heatbath) in order to respect detailed balance. In practice it is sufficient to insert a Heatbath
sweep after several Wolff steps have been performed. This of course slightly alters the gain
brought by the non-local cluster update method concerning critical slowing down near the
phase transition.
The proof that the Wolff algorithm applied to our infinite-spin model fulfills both
ergodicity criteria and detailed balance conditions will be omitted here since it follows
closely the derivations that can be found in [1619].
3.2. Conditional distributions and properties of Wolff clusters
We state in this section our main results concerning the properties of the RC distribution.
First, it is straightforward to see that integrating P(l, , n) over n gives the marginal
distribution of the model with Hamiltonian (2.2).
Proposition 3.1. The marginal spin distribution in (2.3) gives the Boltzmann weight
associated with Hamiltonian (2.2).
Tracing over bond variables in (2.3), one gets
1
X
1
P(l, , n) = Z
nij =0
1
Y X

eJi,j li lj [pi,j nij,1 i j + (1 pij )nij,0 ]
hi,j i nij =0
1
= Z
Y
eJi,j li lj [pi,j i j + (1 pij )]
hi,j i
using the expression for pi,j and re-arranging terms, this can be rewritten as
Y

1
eJi,j li lj i j + eJi,j li lj (1 i j )
Z
hi,j i
1
= Z
Y
hi,j i
eJi,j li lj i j i j + eJi,j li lj i j (1 i j )
thus we get for this marginal distribution

1
Z
Ji,j li lj i j
1
= Z
exp
hi,j i
!
Ji,j li lj i j
hi,j i
1
exp
= Z
373
!
Ji,j Si Sj
hi,j i
eH (S)
,
Z
which is just the Boltzmann weight for the spin configuration S of our model.
The nice features that make Wolff clusters useful reside in the following propositions:
Proposition 3.2. The magnetization M of the spin system is equal to the probability of
long range connectivity in the percolation model.
Considering the joint bond-spin distribution (2.3) with weights (2.5), one can ask: given
a bond configuration, what is the conditional distribution on the spins? Let us assume (+)
boundary conditions for the spins (which amounts at fixing li i = 1 on the boundary). First,
due to the compatibility condition (III.1), if site k is connected to the boundary through
some path in the percolation model, then necessarily k = +1. The same is true with ()
boundary conditions due to the sign symmetry of the exponential factor in the statistical
weight. Second, a site that is not connected to the boundary is equally likely to be in a
1 state. From this it follows that M is exactly given by the probability of connectivity to
the boundary, thence long range order in the spin system is the same as percolation in the
random cluster model.
A more detailed argument will be found in [13]. It is actually easy to show that absence
of percolation is equivalent to zero magnetization. Consider again the random cluster
model (RCM) characterized by the joint distribution (2.3). First let us state the
Definition 3.1. Let P,i () be the probability for a site i to be connected to the boundary
of a finite domain Z2 through some path of active bonds in the percolation model. We
say that there is no percolation in the RCM iff
i ,
lim P,i () = P,i () = 0.
Z2
Proposition 3.3. Consider the joint bond-spin distribution (2.3) and suppose that for the
infinite-spin model at inverse temperature there corresponds no percolation in the joint
correlated geometrical bond representation; then the magnetization vanishes.
The argument is as follows. Let n = (i1 , i2 , . . . , in ) a finite set of spins and
(B)
M ( n ) the corresponding magnetization in the spin model with boundary conditions
(B). Let (B ), the boundary condition identical to (B) except that each spin value on
the boundary has been lowered by an amount . Then one has

(B)

(B)
( n ) ,
M ( n + ) M
374
where n + = (i1 + , i2 + , . . . , in + ). Clearly the geometrical clusters corresponding to these two boundary conditions are the same, thus for all bond configurations in which
sites i1 , i2 , . . . , in are not connected to the boundary, the corresponding contributions to the
magnetization are the same.
Now, using the uniqueness of the limiting state at given , it is easy to conclude that,
in the infinite volume limit of Definition 3.1, absence of percolation is the same as zero
magnetization.
From the proposition above follow also interesting results concerning high temperature
decay of correlations for functions of the spin variables when their relative distance goes
to infinity but we will not enter this question here and refer to [13].
The stronger statement of Proposition 3.2 whose details of proof will also be found in
the last reference (see also [10,11]) is the following: not only the onset of percolation is the
signature of the magnetic ordering but moreover, there exist a constant C depending only
smoothly on such that
P () > M() > C P ().
(3.2)
Explicitly, there is spontaneous magnetization if and only if there is percolation and the
critical behaviour is the same for these two quantities. Namely, considering on one hand
the order parameter exponent , defined for the spin system by M (Tc T ) , where Tc
is the critical temperature, and on the other hand the order parameter for the percolation
model, i.e., the fraction of sites P in the percolating cluster, which behaves like (p pc ) ,
where pc is the percolation threshold. It follows from (3.2) that = .

A similar result as (3.2) can be proven (see [13]) between the magnetic susceptibility of
the continuous spin system and its geometric counterpart.
In the following we shall carry out lattice simulations to determine the critical exponents
for cluster percolation, to illustrate that these indeed lead to the values of the Ising
universality class.
4. Numerical results
We have performed extensive simulations of our model, choosing six different lattice
sizes, namely 642 , 962 , 1282 , 1602 , 2402 and 3002 .
Our update step consisted of one heatbath for the spin amplitudes and three Wolff flippings for the signs, which turned out to be a good compromise to reduce sensibly the
correlation of the data without making the move be too much time-consuming. Every five
updates we measured the quantities of interest. The physical quantities are the energy and
the magnetization of the system. The percolation quantities have been evaluated in this
way: for a given configuration we consider the sign of the spins corresponding to the sign
of the global magnetization, say up. Then we form clusters of spins up using the local bond
weights 1 exp(2(J /kT )si sj ). Once we embody all spins up in clusters we assign to
each cluster a size s which is the number of spins belonging to it. We say that a cluster percolates if it spans the lattice in both directions, that is if it touches all four sides of the lattice.
375
This choice was made to avoid the possibility that, due to the finite lattice size, one could
find more than one percolating cluster, making ambiguous the evaluation of our variables.
Finally we assign to the percolation probability P the value of the cluster size of the
percolating cluster divided by the number of lattice sites (P = 0 if there is no percolation)
and to the average cluster size S the value of the expression
X ns s 2
P
,
S=
s ns s
s
where ns is the number of clusters of size s and the sums exclude the percolating cluster.
After some preliminary scans of our program for several values of the temperature
( = J /kT ), we focused on the beta range between 1.07 and 1.11, where the transition
seems to take place. The number of iterations for each run goes from 20000 (for
values close to the extremes of the range) to 50000 (around the center of the range). We
interpolated our data using the density of states method (DSM) [2023].
To locate the critical point of the physical transition we used the Binder cumulants
gr =
hM 4 i
hM 2 i
3.
Fig. 1 shows gr as a function of for the different lattice sizes we used. The lines cross
remarkably at the same point, which suggests that also in our case gr is a scaling function.
As a numerical proof we replot our lines as a function of ( crit )L1/ , choosing for the
exponent the exact value 1. The plot in Fig. 2 shows that indeed gr is a scaling function.
Fig. 1. Binder cumulants as a function of for our six lattice sizes.
376
Fig. 2. Rescaled Binder cumulants. We took crit = 1.0932 and for the exponent the exact value 1.
Fig. 3. Fraction of percolating samples as a function of for our six lattice sizes.
To get the critical point of the percolation transition we used the method suggested
in [24], which is based on the same principle. For each value we count the number of
configurations where we found a percolating cluster and we divide it by the total number
of configurations. If we plot the results as a function of for the different lattice sizes the
corresponding lines should cross at the threshold of the geometrical transition (Fig. 3).
377
Table 1
Critical point
Thermal results
1.093120+0.000120
0.000080
0.128+0.005
0.006
1.745+0.007
0.007
Percol. results
1.093200+0.000080
0.000080
0.130+0.028
0.029
1.753+0.006
0.006
As the figures show, the agreement between the two thresholds we found is excellent.
For the evaluation of the exponents we used the 2 method [25]. The results we got are
reported in Table 1.
The two thermal exponents ratios are in good agreement with the exact values.
Unfortunately we encountered some troubles in deriving the percolation / exponents
ratio because the best fit values change rapidly in the little range of beta values for which
the 2 corresponds to the 95% confidence level. Consequently, the relative error of /
turns out to be large. For a more precise evaluation of this ratio much higher statistics or
higher lattice sizes seem to be necessary.
However, the value of the other exponents ratio / is in good agreement with the
corresponding exact value 7/4 and its error seems to exclude the possibility for it to be
instead the random percolation ratio 43/24.
5. Conclusions
For the continuous spin Ising model, the percolation transition of FortuinKasteleyn
clusters is indeed equivalent to the magnetic transition due to spontaneous Z2 symmetry
breaking. We have also determined by means of lattice simulations the critical exponents
of the percolation variables and shown that they belong to the Ising universality class. The
simulation method used here can also be employed in the study of percolation in SU(N)
gauge theories, where analytic proofs do not exist. We expect that they will eventually
allow a general demonstration that the deconfinement transition in such theories can indeed
be defined as Polyakov cluster percolation.
Acknowledgements
D.G. gratefully acknowledges financial support and kind hospitality from the BiBoS
research center, Department of Physics, University of Bielefeld, Germany. S.F. was
supported by the EU-Network ERBFMRX-CT97-0122 and the DFG Forschergruppe Ka
1198/4-1; he also acknowledges the Centre de Physique Thorique, C.N.R.S, Luminy,
Marseille, France where part of this work was developed. P.B. was partially supported by
Alexander von Humboltd Foundation and KBN grant 2P03B 149 17. It is our pleasure
to thank L. Chayes for discussions and, for two of us, Ph.B. and D.G., for the joy of
collaboration.
378
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
S. Fortunato, H. Satz, hep-lat/9911020, Phys. Lett. B (in press).

B. Svetitsky, L.G. Yaffe, Nucl. Phys. B 210 [FS6] (1982) 423.
C.M. Fortuin, P.W. Kasteleyn, Physica 57 (1972) 536.
R.B. Griffiths, J. Math. Phys. 10 (9) (1969) 1559.
F. Green, F. Karsch, Nucl. Phys. B 238 (1984) 297.
R.H. Swendsen, J.S. Wang, Phys. Rev. Lett. 58 (1987) 86.
A. Coniglio, W. Klein, J. Phys. A 13 (1980) 2775.
R.G. Edwards, A.D. Sokal, Phys. Rev. D 38 (1988) 2009.
G. Grimmet, Ann. Prob. 23 (4) (1995) 1461.
L. Chayes, Commun. Math. Phys. 197 (3) (1998) 623.
L. Chayes, J. Phys. A 31 (13) (1998) L255.
U. Wolff, Phys. Rev. Lett. 62 (1989) 361.
Ph. Blanchard, L. Chayes, D. Gandolfo, submitted.
D.W. Heermann, A.N. Burkitt, Physica A 162 (1990) 210.
C.F. Baillie, D.A. Johnston, G.W. Kilcup, J. Supercomputing 4 (1990) 277.
U. Wolff, Nucl. Phys. B 322 (1989) 759.
T.E. Harris, Proc. Camb. Phil. Soc. 56 (1960) 13.
L. Chayes, J. Machta, Physica A 329 (1997) 542.
L. Chayes, J. Machta, Physica A 254 (1998) 477.
M. Falconi et al., Phys. Lett. B 108 (1982) 331.
E. Marinari, Nucl. Phys. B 235 (1984) 123.
G. Bhanot et al., Phys. Lett. B 183 (1986) 331.
A.M. Ferrenberg, R.H. Swendsen, Phys. Rev. Lett. 61 (1988) 2635; Phys. Rev. Lett. 63 (1989)
1195.
[24] K. Binder, D.W. Heermann, Monte Carlo Simulations in Statistical Physics, An Introduction,
Springer-Verlag, 1988, 4041.
[25] J. Engels et al., Phys. Lett. B 365 (1996) 219.

Fundamental strings in DpDq brane systems

Jrg Frhlich a,1 , Olivier Grandjean b,2 , Andreas Recknagel c,3 ,
Volker Schomerus d,4
a Institut fr Theoretische Physik, ETH-Hnggerberg, CH-8093 Zrich, Switzerland
b Department of Mathematics, Harvard University, Cambridge, MA 02138, USA
c Max-Planck-Institut fr Gravitationsphysik Albert-Einstein-Institut, Am Mhlenberg 1,
D-14476 Golm, Germany

d II. Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149,
D-22761 Hamburg, Germany

Received 28 January 2000; accepted 20 April 2000
Abstract
We study conformal field theory correlation functions relevant for string diagrams with open
strings that stretch between several parallel branes of different dimensions. In the framework of
conformal field theory, they involve boundary condition changing twist fields which intertwine
between Neumann and Dirichlet conditions. A KnizhnikZamolodchikov-like differential equation
for correlators of such boundary twist fields and ordinary string vertex operators is derived, and
explicit integral formulas for its solutions are provided. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.25.-w; 11.25.Hf; 02.60.Lj; 11.10.Kk
1. Introduction
D-branes [30] have become the most important ingredient of the new picture of string
theory that has emerged in recent years. They have shaped a new understanding of nonperturbative effects in string theory and of low-energy effective theories associated with
string theories. In the latter context, systems of many branes are of particular importance,
since they provide a natural way to include non-abelian gauge theories into string theory
[40]. Systems of several branes of different dimensions, most notably of D1- and D5branes, play a major role in proposals of how to derive the BekensteinHawking entropy of
1 juerg@itp.phys.ethz.ch
2 grandj@math.harvard.edu
3 anderl@aei-potsdam.mpg.de
4 vschomer@x4u.desy.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 3 7 - 6
382
J. Frhlich et al. / Nuclear Physics B 583 (2000) 381410
black holes from string theory [27,39]. More recently, stacks of branes and antibranes have
been reconsidered in connection with a K-theoretic classification of branes [41], based on
results concerning tachyon condensation in [37].
Some qualitative features of such systems can be uncovered within a target space
approach. But, e.g., the process of braneantibrane annihilation in the last-mentioned
application, or the properties of near-extremal black holes involve the analysis of nonBPS states for which the world-sheet approach is better suited, since it does not critically
depend on supersymmetry. Computation of CFT correlation functions is indispensable if
one wants to deal with problems like Hawking radiation off D1D5-systems, or for a clean
discussion of bound state formation [23].
The world-sheet formulation of string sectors with branes is well known in string theory,
although mainly in connection with flat targets. The general setup involves boundary
conformal field theory as introduced and developed by Cardy [810] and first exploited for
string theory by Sagnotti [34]. CFT on surfaces with boundaries exhibits a very interesting
internal structure and finds interesting applications beyond string theory: for many s-wave
dominated scattering processes, the universal behaviour is described by a boundary CFT
in two dimensions, irrespective of the dimensionality of the original system. The most
famous problem that could be tackled with boundary CFT methods is the Kondo effect in
condensed matter physics [1].
In string theory, methods of boundary CFT are not only valuable in the study of
situations without the BPS-property, but also to uncover non-classical features like
unexpected moduli [33,37] and non-commutative geometry; see, e.g., [3,15,16,35,36] and
references therein. Moreover, they allow one to analyze D-branes in non-geometric string
compactifications such as Gepner models [6,13,24,29,32,38].
In this paper, we ask how to compute CFT correlators describing string amplitudes of
arbitrary closed and open string vertex operators in the presence of multiple flat branes
in RD . The open strings involved stretch between two or more different branes, which
may have different dimensions. It appears that no systematic method for the computation
of those string diagrams, which contribute to scattering processes in higher orders of the
string coupling constant, is available in the literature. See, however, [25] for some sample
computations of scattering amplitudes in the presence of a pair of branes.
The world-sheet description requires surfaces with several boundary components. We
restrict our attention to diagrams without internal closed string loops so that we can map the
world-sheet to the disk or to the upper half-plane, but with different boundary conditions
assigned to consecutive intervals on the boundary; see Fig. 1. We focus on parallel branes
here; thus we can reduce our analysis to a one-dimensional target. Results for RD with
D > 1 follow by taking tensor products. The boundary state for a p-brane involves p + 1
Neumann and D p 1 Dirichlet boundary states of single free bosons.
The interesting transitions between boundary conditions in a one-dimensional target are
those from Neumann to Dirichlet or vice versa. They are mediated by boundary fields of a
special type, namely boundary condition changing twist fields.
Conformal boundary conditions which preserve the chiral algebra W of the theory are
parametrized by certain automorphisms of W, together with the amplitudes of one-
383
Fig. 1. The upper half-plane with a sequence of Neumann (solid intervals) and Dirichlet (dashed
intervals) boundary conditions along the real line. The dots between Neumann and Dirichlet intervals
mark insertions of boundary condition changing twist fields, while crosses on the boundary or in
the interior refer to insertions of ordinary open respectively closed string vertex operators. Such a
world-sheet diagram can be understood as Hawking radiation (closed string states) from a system of
branes (multiple changes of boundary conditions) with simultaneous inner excitations (open string
states).
point functions [32]. If W is the U(1) current algebra, can act as id on the currents,
and the 1-point functions determine the location of a brane. If the boundary condition is
constant along the boundary, arbitrary n-point functions can be expressed in terms of the
usual conformal blocks of W; see, e.g., [8,19,32].
If the gluing conditions described by an automorphism of W remain constant along
the boundary the computation of correlation functions is not, in principle, a difficult
problem. Otherwise, the simple Ward identities for the symmetry algebra W are broken,
and one has to find new methods to construct the twisted chiral blocks involving a
new type of boundary condition changing operators which correspond to twisted rather
than ordinary representations of W. It is the aim of this article to develop a convenient
formalism for computing such correlation functions in the case of a flat target space.
The plan of this paper is as follows: in the next section, we look at correlation functions
which contain just one insertion of a boundary twist field. We shall provide a complete
operator construction of the boundary CFT, from which one can derive correlators with
an arbitrary number of closed string vertex operators inserted in the bulk. When there are
more than two twist fields on the boundary, such techniques are no longer available. Our
strategy is then to derive Ward identities for the correlation functions. They will lead us to
KnizhnikZamolodchikov-like differential equations which describe the effect of moving
insertion points for bulk and boundary fields in terms of a flat connection. We explain this
idea in Section 3 and exploit it in the fourth section to give explicit integral formulas for the
correlators. While some of the technical steps in setting up the KnizhnikZamolodchikov
equations are rather involved, parts of our final results can be related to electrostatics.
Section 5 comments on possible generalizations and applications.
2. Operator formalism for a single twist field insertion

As a simple example, we consider open strings propagating freely in the target R, with
Dirichlet boundary conditions imposed at one end of the string and Neumann boundary
384
conditions at the other. We thus have to deal with a free bosonic field X(t, ) defined for
space variables [0, ] and subject to
t X(t, 0) = 0,
X(t, ) = 0
for all t R. Mapping the strip to the upper half-plane H by z = exp(t + i ), X(z, z )
satisfies Dirichlet boundary conditions for z R>0 and Neumann boundary conditions for
z R<0 :
X(z, z ) = x0 ,
for z = z > 0,
( )X(z,
z ) = 0,
for z = z < 0.
(1)
Our task is to compute correlation functions in this bosonic theory which involves two
insertions of twist fields on the boundary (see [12] for an early treatment of that problem).
Conformal symmetry allows us to place these boundary condition changing operators at
x1 = 0 and x2 = .
We propose to construct operators X(z, z ) satisfying (1), as well as open and closed
string vertex operators, and then to derive differential equations on the correlation functions
from the algebraic properties of these operators and from the symmetries of the theory.
First, we have to determine the space our fields are to act on.
2.1. The spectrum of boundary twist fields
The space in question is spanned by excited states of open strings stretching between
a Neumann and a Dirichlet boundary condition hence the name boundary condition
changing operators for the boundary fields uniquely associated to these states. There is
a relatively simple technique to determine the spectrum of boundary condition changing
operators that intertwine between two constant conformal boundary conditions B1 =
(1 , 1 ) and B2 = (2 , 2 ) in some boundary CFT. The state space of this boundary
theory is denoted by H12 . Because of the state-field correspondence, the spectrum of
boundary condition changing operators is described through the partition function of the
boundary theory,
c
(H )
Z12 (q) = TrH12 q H , where H (H ) = L0 .
24
By an interchange of space and time coordinates (world-sheet duality), the open string
1-loop diagram underlying Z12 may be viewed as a closed string tree diagram, i.e.,
1 (P )
c
(P )
(P )
Z12 (q) = hB1 |q 2 H |B2 i, where H (P ) = L0 + L 0 ,
12
where q = exp(2i/ ) is related to the variable q = exp(2i ) as usual. The closed
strings propagate between the boundary states |Bi i = |i ii associated with the boundary
conditions Bi . They allow to transfer boundary conditions from the upper half-plane
(where the Hamiltonian is H (H ) ) into a CFT on the full plane (with Hamiltonian H (P ) ),
see [9,26].
The boundary states implementing Dirichlet and Neumann conditions along the whole
boundary are of course well known, see, e.g., [7]. Let an(P ) , a n(P ) be two commuting sets
of oscillator modes (in the plane CFT) with standard commutation relations. The ground
385
states |ki of their Fock spaces are labeled by the momentum k R. Neumann and Dirichlet
boundary states are given by
X

1 (P ) (P )
1
an a n |0i,
|Ni = exp
n
2
n>1
X

Z
1 (P ) (P )
ikx0
a a
exp
|ki,
(2)
|D(x0 )i = dk e
n n n
n>1
where x0 R, as in (1), denotes the location of the D-brane, i.e., x|D(x

0 )i = x0 |D(x0 )i
for the center of mass coordinate x.
If we build the boundary state |pi for a p-brane in RD as a tensor product of |Ni
(p + 1 times) and |D(xi0 )i (i = p + 2, . . . , D) from Eq. (2), the partition function
Zpp (q) counts boundary fields that do not change the boundary condition. They describe
excitations of open strings attached to the p-brane. These open string vertices have the
form 1 n X1 Xn eikX with certain Lorentz tensors 1 n and with momentum
k parallel to the Neumann directions. The case n = 1 (where the polarization is
transversal) contains the massless modes: gauge fields living on the brane world-volume.
The partition function of the theory with Neumann boundary conditions on one side and
Dirichlet on the other follows from the boundary states as explained above,
c
ZND (q) = TrHND q HND = hN|q L0 24 |D(x0 )i

P 1
1
c P
1
= h0|e n=1 n an a n q L0 24 e m=1 m am a m |0i.
2
(3)
Orthonormality of the Fock ground states implies that only the contribution from the
vacuum sector survives in the second line. In particular, ZND is independent of the
parameter x0 . Computation of the vacuum expectation value above is straightforward. The
result can be written as
ZND (q) = q 1/48
Y
n=1
1 q n 2
1
1 X 1 (n 1 )2
2 .
q4
(q)
(4)
n=1
Our main conclusion concerns the conformal weights of the boundary fields that can induce
a transition between Dirichlet and Neumann type boundary conditions. The lowest weight
1
. Above this value, the spectrum of conformal weights has halfthat appears is h = 16
integer spacings. The boundary condition changing operator with conformal weight h =
1
16 corresponding to the lowest-energy state | i in the whole sector HND will be called
(x).
We will also refer to (x) as a twist field since the sum of irreducible Virasoro
characters in (4) can alternatively be regarded as the character of a twisted U(1)
representation. The absence of a vacuum state and the half-integer energy grading are
symptoms for the fact that the jump from Neumann to Dirichlet destroys the simple U(1)
Ward identities that are present in a boundary CFT with constant Neumann or Dirichlet
condition all along the boundary. See [21] for general results about twist fields and partition
functions in boundary conformal field theory.
386
It will be our main concern in the following to find substitutes for the broken Ward
identities, namely twisted KnizhnikZamolodchikov equations.
2.2. Construction of the basic fields
In order to construct a field X(z, z ) obeying the boundary conditions (1) on H = HND ,
we introduce a set of oscillator modes ar labelled by half-integers r Z + 12 . (We drop the
superscript (H ) for operators of the upper half-plane theory.) They are supposed to obey
the relations
[ar , as ] = rr,s ,
ar = ar .
Creation operators ar , r < 0, generate the Fock space H out of the ground state | i, which
is annihilated by modes ar with index r > 0. All the fields we shall consider act on this
z) of the
state space H. It is simple to verify that the decomposition X(z, z ) = X(z) X(
bosonic field yields the desired properties if
X ar
X ar
z) = i
zr ,
z r .
X(
X(z) = x0 + i
r
r
1
1
rZ+ 2
rZ+ 2
To make the square root well defined, we have to introduce a branch cut in the plane, which
extends from x = 0 to . Once the bosonic field is known, we obtain chiral currents as
X
ar zr1 ,
J (z) := iX(z, z ) =
rZ+ 12
J(z) := i X(z,
z ) =
ar z r1 .
rZ+ 21
Finally, the components T (z) and T (z) of the stress energy tensor are given by

1
1
J (w)J (z)
,
T (z) = lim
wz 2
(w z)2
and likewise for T (z). Since T and T are quadratic in J and J, they satisfy the usual
boundary condition T (z) = T (z) all along the real line Im z = 0. By the usual arguments
[8] this implies that the modes
Z
Z
dz n+1
d z n+1
z T (z) +
z T (z)
Ln :=
2i
2i
C+
obey commutation relations of the Virasoro algebra with central charge c = 1. Here C+
C is a closed oriented contour surrounding the origin, with C+ contained in the upper
half-plane and C contained in the lower half-plane. The commutation relation between
Ln and ar is easily checked to be of the form
[Ln , ar ] = ran+r .
It is convenient to introduce two generating fields T and J by the formal sums
X
X
T(w) =
Ln wn2 ,
J(w) =
ar wr1 .
nZ
rZ+ 21
387
One may think of T as being defined on the entire complex plane with T(w) = T (w) in the
upper half-plane, Im w > 0, and T(w) = T (w), for all w with Im w < 0. The generating
field J naturally lives on the two-fold branched cover of the complex plane defined by
2 = w. By introducing the branch cut from x = 0 to we have specified a coordinate
patch on this surface with the local coordinate being denoted by w. In this chart, J(w) =
J (w) for Im w > 0 and J(w) = J(w) for Im w < 0.
The commutation relations for the modes Ln , ar with the bosonic field X(z, z ) can be
expressed in terms of T and J as follows

T(w), X(z, z ) = X(z, z )(z w) + X(z,

z )(z w),

J(w), X(z, z ) = i(z w) + i(z w),
where
(z w) :=

r
z
1X z n 1 X
=
.
z
w
z
w
1
nZ
rZ+ 2
We state two simple consequences of these formulas that shall be important below. We split
J and T into two parts J(w) = J> (w) + J< (w) and T(w) = T> (w) + T< (w) such that
X
X
J> (w) :=
ar wr1 ,
T> (w) :=
Ln wn2 .
r>1/2
n>1
In the next subsection, we will use the commutation relations between the singular parts
T> , J> of the generating fields T, J and the bosonic field X(z, z ):
1/2

z
i
,
J> (w), X(z) =
w
wz

1
z X(z).
T> (w), X(z) =
(5)
wz
Moreover, we will need the following lemma, a proof of which is given in Appendix A.
Lemma 1. One may rewrite the generating field T(w) in terms of the objects J> (w) and
J< (w), namely
T(w) =

1
1 1
J< (w)J(w) + J(w)J> (w) +
.
2
16 w2
(6)
2.3. Bulk and boundary primary fields

Our next aim is to construct primary bulk and boundary fields. Here, the latter term refers
to open string vertex operators which can be inserted on R<0 or R>0 without changing the
boundary condition. They are in one-to-one correspondence to states in HDD or HNN , not
to states in HND . We will see that bulk fields can be regarded as products of such boundary
fields g (z). Therefore we discuss these chiral fields first admitting arbitrary complex
insertion points, not just z H. The fields g (z) are labeled by a real parameter g and
enjoy the properties
388
T(w), g (z) = z g (z)(x w) + hg (z)z (z w),
J(w), g (z) = gg (z)(z w).
We have used the definition of the formal -function specified above and h = 12 g2 . For the
commutators of the T> (w), J> (w) with the fields g (x), this implies
1
h
z g (z) +
g (z),
wz
(w z)2
1/2

z
g
g (z).
J> (w), g (z) =
w
wz
T> (w), g (z) =
(7)
(8)
Lemma 2. The unique solution (up to normalization) g (z) to the requirements (7), (8) is
given by
h
i
zh eigX< (z) eigX> (z) ,
g (z) =
2
where
X> (z) = i
X ar
r>0
zr
and X< (z) = X(z) X> (z). Note that g (z) is normal-ordered, i.e., the annihilators
ar , r > 0, appear to the right of the creation operators.
A proof can be found in Appendix A.
Our next aim is to describe the U(1)-primary bulk fields g (z, z ). By definition, they
obey the following commutation relations with respect to J> and T> ,
1
h
g (z, z ) +
g (z, z )
wz
(w z)2
h
1
g (z, z ),
g (z, z ) +
+
w z
(w z )2
1/2
1/2

z
z
g
g
g (z, z )
g (z, z ).
J> (w), g (z, z ) =
w
wz
w
w z
T> (w), g (z, z ) =
(9)
(10)
Note that each term from Eqs. (7), (8) appears a second time with z being replaced by
the variable z . One can easily work out commutation relations between the full generating
elements T(w), J(w) and the bulk primary fields g (z, z ). It is obvious from our discussion
of boundary fields that bulk fields g (z, z ) can be written as products of chiral vertex
operators,
g (z, z ) = g (z)g (z).
The formulas we have reviewed here would enable us to perform a direct computation
of arbitrary correlations functions G(Ez) for bulk-fields g (z, z ) with two twist fields
inserted at x = 0 and x = ,
G(Ez) := h |1 (z1 , z 1 ) n (zn , z n )| i,
389
where
(z , z ) = g (z , z ).
The calculations would proceed by moving all annihilation operators to the right until they
act on the ground state | i. The same techniques would apply if there are extra boundary
fields g (x) inserted in addition to the bulk fields g (z, z ). Since we will develop another,
more generally applicable approach to the computation of correlation functions below, we
do not enter details here.
Before we conclude this subsection, we would like to derive bulk-boundary operator
product expansions which allow to expand our bulk fields g (z, z ) in terms of boundary
operators [10]. The essential idea is that, in the presence of a boundary, bulk fields split into
products of chiral vertex operators inserted at points which are obtained from each other
by reflection at the real axis and with opposite charges (method of image charges). In
order to obtain concrete formulas, we first rewrite the bulk fields g (z, z ) = g (z)g (z)
in terms of X> (z, z ) = X> (z) X > (z) and X< (z, z ) = X< (z) X < (z),

2h

i
z + z 2h igX< (z,z) igX> (z,z)
h
(zz)
e
e
.
(11)
g (z, z ) =
2
z z
We have used the expression in Lemma 2 and then normal-ordered the right-hand side with
the help of the BCH formula, which leads to the additional z - and z -dependent factor.
Lemma 3 (bulk-boundary OPE). For arguments z = x + iy close to the boundary, i.e.,
y > 0 small, the operators g (z, z ) can be expanded in a series involving boundary primary
fields, with leading asymptotics
g (z, z )
eigx0
1,
y 2h
for x > 0,
g (z, z ) y 2h 2g (x),
for x < 0,
where 1 is the identity field.
Proof. Let us begin with the case x > 0 in which z z 0 as y becomes very small.
2h
2h

i
z
h z + z + 2 z
(zz)
eigX< (z,z) eigX> (z,z)
g (z, z ) =
2
z z
2h
2h
i
1
2x
x 2h
1 = 2h 1.
2
iy
y
We have also used that X< (x, x) = x0 and X> (x, x) = 0 for x > 0, which is a direct
consequence of the Dirichlet boundary condition.
If x < 0 and y tends to zero, the sum z + z vanishes and we can estimate the
behaviour of g (z, z ) according to
2h
2h

i
z z
h
(zz)
eigX< (z,z) eigX> (z,z)
g (z, z ) =
2
z + z 2 zz
390
2h
2h
i
2h iy
x
e2igX< (x) e2igX> (x) = y 2h 2g (x). 2
2
2x
Observe that boundary condition changing operators themselves do not arise from the
bulk-boundary OPE of bulk fields. Let us finally note that operator product expansions for
bulk fields or for boundary fields can be worked out with the same techniques. Those for
bulk fields g (z, z ) of course agree with the usual OPE of primary fields in the bulk.
2.4. Correlation functions and the KnizhnikZamolodchikov equation
This subsection contains the main result of this section, namely a derivation of
the KnizhnikZamolodchikov equation for correlation functions of bulk and boundary
primaries in the presence of a transition from Dirichlet to Neumann boundary conditions
at the origin. Let us look more closely at correlation functions containing n chiral fields
g (z),
F (Ez) := h |1 (z1 ) n (zn )| i,
where (z ) := g (z ).
Before we start, let us state two elementary formulas for the action of J> (w), T> (w) on
the ground state | i:

1
1
L
T> (w)| i =
h
+
(12)
1 | i,
w2
w
X
J> (w)| i =
ar | iwr1 = 0,
(13)
r>1/2
1
where h is the conformal weight of the state | i. We will recover the equation h = 16
in a moment. The object h | dual to | i obeys the relations h |J< (w) = 0 = h |T< (w).
A first differential equation is obtained by inserting the generating field T(w) into the
correlation function:
h |T(w)1 (z1 ) n (zn )| i = h |T> (w)1 (z1 ) n (zn )| i

#
" n

X
h
1
h
+
+ 2
=
w z
(w z )2
w
=1
1
h |1 (z1 ) n (zn )L1 | i.
w
Here, we have commuted T> (w) through the fields (z ) until it acts on the ground state
| i so that we can use formula (12).
Now we want to compute the same correlation function with the help of the affine
Sugawara construction, i.e., by exploiting Eq. (6):
F (Ez) +
h |T(w)1 (z1 ) n (zn )| i

1
= h |J> (w)J> (w)1 (z1 ) n (zn )| i
2
1
h |1 (z1 ) n (zn )| i
+
16w2
391

n
1 X z 1/2 g
1
=
h |J> (w)1 (z1 ) n (zn )| i +
F (Ez)
2
w
w z
16w2
=1
#
"
g g
1
1 X z z
+
F (Ez).
=
2 , w (w z )(w z ) 16w2
Comparison with our first formula for the insertion of T(w) yields h =
residue at w = 0 we get
1 X g g
F (Ez).
h |1 (z1 ) n (zn )L1 | i =
2 , z z
Finally, from the residue at w = z we obtain
"
#
r
h X z g g
F (Ez).
z F (Ez) = +
z
z z z
1
16 .
From the
(14)
6=
This is the KnizhnikZamolodchikov equation we were after. Note that the terms in
square brackets determine a flat connection, as in the ordinary KnizhnikZamolodchikov
equation.
We can solve (14) by a simple coordinate transformation. In fact, if we introduce
Q
coordinates u = z and the function Fu (u1 , . . . , un ) = uh F (u21 , . . . , u2n ), then the

system (14) of first order differential equations becomes
!
n
X
g2
g g
g g
+
u Fu (u1 , . . . , un ) =
2u
u u u + u
=1
6=
Fu (u1 , . . . , un ).
(15)
This equation is formally identical to the usual KnizhnikZamolodchikov equation with

2n fields of charges g inserted at the points u . The solution to (15) is given by
n
Y
Y u u g g
g2 /2
ui
.
(16)
Fu (u1 , . . . , un ) =
u + u
=1
16<6n
The free parameter can be determined from the boundary condition of the bosonic field
X on the positive real line, i.e., = (x0 ) depends on the position x0 of the D-brane.
3. Twisted KnizhnikZamolodchikov equation for multiple transitions

In the following, we study n-point functions of a free bosonic field theory on the
half-plane with several insertions of twist operators placed along the boundary. In the
corresponding string diagrams, open strings stretch between three or more branes of
various dimensions.
As long as only one DN-jump occurs, a simple Hilbert space formulation of the
boundary CFT is available, and we can solve, in principle, for correlation functions by
392
purely algebraic techniques, as indicated in the last section. In the presence of many
boundary condition changing twist fields, it may be simpler to resort to OPE methods and to
the theory of complex functions on higher genus Riemann surfaces, and this is the approach
we pursue in the present section. We begin with a very brief review of relevant input
from the theory of hyperelliptic surfaces. We then discuss Ward identities in the second
subsection. The latter allow us to derive a system of linear first order differential equations
for the correlation functions similar to the KnizhnikZamolodchikov equations. Note that
free bosons on higher genus surfaces without boundaries (i.e., higher loop diagrams of
closed strings propagating in a flat target) have been studied in great detail in [4,31].
3.1. Hyperelliptic surfaces
Our aim is to investigate a scenario in which a bosonic field X(z, z ) is defined on the
upper half-plane with boundary conditions switching between Dirichlet and Neumann
at 2g + 2 points xi , i = 1, . . . , 2g + 2, on the boundary. Without loss of generality, we
shall assume that x2g+2 = =: x0 . To be more precise, we impose Dirichlet boundary
conditions in the intervals ]xi , xi+1 [ for i odd and Neumann boundary conditions along
the rest of the boundary, i.e.,
for z = z Dk := ]x2k1, x2k [
X(z, z ) = xk0 ,
and
y X(z, z ) = 0,
for z = z Nk := ]x2k2, x2k1 [ .
The variable y is defined through z = x + iy, and k = 1, . . . , g + 1. In terms of the chiral
currents J (z) = iX(z, z ) and J(z) = i X(z,

z ), these conditions become
J (x) = J(x),
for x Dk ,
and
J (x) = J(x),
for x Nk .
As in the previous section, it is convenient to work with a single field J that contains all
information about the two chiral currents J and J. Such a field necessarily lives on a twofold branched cover of the complex w-plane, namely on the hyperelliptic surface, M, of
genus g which is described by the equation
2 = P (w) :=
2Y
g+1
(w xi ).
i=1
Introducing branch cuts along the intervals Nk = [x2k2, x2k1 ], we obtain a particular
coordinate patch of this surface with local coordinate w. In this chart, J(w) satisfies J(w) =
J (w) for Im w > 0 and J(w) = J(w) for Im w < 0. The Virasoro field T obeys the
gluing condition T (x) = T (x), all along the boundary, since it is quadratic in the currents.
Consequently, the generating field T(w) is defined on the complex w-plane and coincides
with T (respectively T ) on the upper (respectively lower) half-plane.
393
Fig. 2. The curves k run counterclockwise around the Neumann cuts on the real line. The curves e
k
run clockwise and close on the second sheet through Neumann intervals.
The coordinate w on the complex plane lifts to a meromorphic function of degree 2, also
denoted by w, on the hyperelliptic surface M. This function defines a two-fold covering of
the sphere branched over 2g + 2 points Q1 , . . . , Q2g+2 , where w(Qi ) = xi . A basis for the
space of holomorphic 1-forms on M is then given by
wk1 dw
,
k :=
P (w)
for k = 1, . . . , g.
It will be convenient to work with a canonical homology basis {k , k } on M chosen as in

Fig. 2. We denote by kl the period of the 1-form l along the cycle k , i.e.,
I
I l1
w dw
.
kl := l =
P (w)
k
The basis of holomorphic 1-forms, {k }, dual to the canonical homology basis {k , k } is

defined by the equation
I
l = kl .
k
In particular, we have the relation

k =
g
X
lk l ,
l=1
and the matrix is invertible. The period matrix is given by

I
kl := l ,
k
and it is known to be symmetric and to have positive definite imaginary part. The surface
M has an anti-holomorphic involution induced by complex conjugation on the complex
plane. In terms of the functions w and = P (w), it can be written as

(w, ) (w,
).
394
This involution will be used to extend the theory from the upper half-plane to the lower
half-plane while taking care of the boundary conditions on the real axis. There is a second
holomorphic involution that interchanges the two sheets of M and that can be written as
(w, ) (w, ).
This involution is used when passing from the sphere with cuts to its cover M.
3.2. The Ward identities
To begin with, we introduce the correlations that we plan to investigate below. Besides
the primary bulk fields g (z, z ) := exp(igX(z, z )), they involve 2g + 2 boundary twist
fields inserted at the points xi , which induce changes between Dirichlet and Neumann
boundary conditions. From our discussion in the previous section we know that there is
an infinite number of boundary condition changing operators that we could insert. But it
1
, since all others can be
is sufficient to study the fields (x) of conformal weight h = 16
obtained out of (x) by OPE with chiral fields. Thus, our discussion deals with correlators
of the form

(17)
G(Ez, xE) = 1 (z1 , z 1 ) n (zn , z n ) (x1 ) (x2g+1 ) ,
where we use the notation = g , and where the boundary field (x2g+2 ) = ()
is absorbed in the notation h i = h | |0i, with |0i denoting the vacuum state. We
could insert further boundary fields g (z). Their interpretation depends on whether they
are inserted in an interval with Dirichlet or Neumann boundary conditions. In the former
case, they could induce jumps in the Dirichlet parameters xk0 if they are associated with
open strings stretching between branes at different positions. A boundary operator g (z)
inserted in one of the Neumann intervals, on the other hand, creates an open string which
has both ends on the same brane (an euclidean 1-brane, in our case) and moves with some
definite momentum along its world-volume.
As far as Ward identities are concerned, correlation functions of such boundary fields are
actually more fundamental, since one may split the bulk fields g (z, z ) into a product of
g (z) and g (z). For this reason, most of our investigations below involve the correlation
functions

(18)
F (Ez, xE) = 1 (z1 ) n (zn ) (x1 ) (x2g+1 ) ,
from which the correlators G(Ez, xE ) can be reconstructed.
Our analysis will make essential use of the MittagLeffler theorem, and hence it is based
on the study of singularities in correlation functions. The latter are encoded in the operator
product expansions between chiral fields and the bulk and boundary fields appearing in
(17), (18). For the Virasoro field T one has the standard expansions:

hg
1
z g (z),
T(w)g (z)
+
(19)
(w z)2 w z

h
1
x (x).
T(w) (x)
+
(20)
2
(w x)
wx
395
Here and in the following, the symbol means equal up to terms which are regular as
w z. According to the rule g (z, z ) g (z)g (z), the operator product expansions
of T with g contain further terms in which z is replaced by z (note that hg = hg ).
These formulas may be compared with Eqs. (9) and (12) in Section 2. For our correlation
functions, Eqs. (19), (20) imply

T(w)1 (z1 ) n (zn ) (x1 ) (x2g+1 )

" n

X
1
h
+
=
(w z )2 w z z
=1
#
2g+1
X
1
h
+
F (Ez, xE).
(21)
+
(w xi )2 w xi xi
i=1
The situation is more subtle for the current J(w), which we recall is only well-defined on
a surface of genus g. More precisely, J(w) dw is a meromorphic 1-form on a hyperelliptic
surface. For the operator product expansion of J(w) with the field g (z), we shall use
g
P (z)
g (z) dw.
J(w)g (z) dw
(22)
P (w) w z
Indeed, the right hand side has a first order pole at w = z with residue g and is regular
otherwise. Eq. (22) generalizes formula (10) in Section 2.2. To determine the operator
product expansion between J and the twist field (x), we observe that the leading
contribution
J(w) (x) (w x)h 1h (x) +
must involve a field of conformal weight h = h + 1/2 + Z. Otherwise, the expansion

would not be consistent with the periodicity properties of J close to the branch point
at w = x. Among the boundary condition changing operators, there is one field with
1
which gives rise to the most singular contribution diverging
conformal weight h = 12 + 16
1/2
, cf. the spectrum (4) computed above.
with (w x)
After multiplication of the previous equation with dw, we are supposed to study the
right-hand side in the local coordinate = w x. The outcome is rather simple: the
form (w x)1/2 dw = 2 d on the right-hand side is regular at = 0 so that we conclude
J(w) (x) dw 0,
(23)
i.e., the singular part of the operator product expansion between J(w) dw and the twist
field (x) vanishes.
As we will see, the following important formula is a consequence of these operator
product expansions:

J(w)1 (z1 ) n (zn ) (x1 ) (x2g+1 )

#
" n
g
X g
P (z ) X k (Ez, xE)wk1
F (Ez, xE),
(24)
+
=
P (w) w z
P (w)
=1
k=1
396
where
k (Ez, xE ) =
g
X
1
kl
i x0
k
l=1
n
X
!
k
g B (z ) ,
=1
are obtained from the values of the bosonic field at

and the parameters k x0 := xk0 xk+1
0
the boundary.
To derive the formula (24) we exploit the fact that a meromorphic 1-form on a compact
Riemann surface is determined by its principal part up to some holomorphic 1-from.
Operator product expansions, on the other hand, contain all information about the principal
part. Hence, from Eqs. (22), (23) we conclude that

J(w)1 (z1 ) n (zn ) (x1 ) (x2g+1 )
g
n
X
X
g
P (z )
k (Ez, xE )wk1
.
(25)
F (Ez, xE ) +
=
P (w) w z
P (w)
=1
k=1
Note that the insertion points z and xi parametrize a whole family of meromorphic
1-forms, and the coefficients k may depend on them. Actually, we can determine this
dependence completely. To this end we integrate the above equation along a loop k
which surrounds the interval [x2k , x2k+1 ], as shown in Fig. 2. On the right-hand side of
our equation, this integral may be expressed in terms of the matrix and
I
1
P (z)
dw.
(26)
B k (z) :=
P (w) w z
k
Note that the matrix elements kl and the functions B k (z) depend on the insertion
points xi . With these conventions we find
I
g
n
X
X
dw(r.h.s. of (25)) =
g B k (z )F (Ez, xE ) +
kl l .
=1
l=1
Next, let us analyze the integral over the left-hand side of Eq. (25). By a deformation,
we can make the integration contour k symmetric under a reflection along the
real line. Now, we may split k into two parts k> , k< with the property Im k> > 0 and
Im k< 6 0, so that each piece lies entirely in one of the half-planes. With these conventions,
our contour can be written as a composition k = k> k< which obeys k< = k> . If we
recall, in addition, that the field J coincides with J on the upper and with J on the lower
half-plane we deduce
Z
Z
I
d w
J(w) dw =
J (w) dw + J(w)
k
k>
=i
k>

dX(w, w)
= i xk0 xk+1
.
0
k>
In the penultimate step, we have expressed the currents through the bosonic field X by
J (w) = iX(w, w)
and J(w)
= i X(w,
w).
The contour integral over the differential dX
397
is finally determined by the values of X at the boundary. For the integration over the lefthand side of (25), this result implies that
I

dw(l.h.s. of (25)) = i xk0 xk+1
F (Ez, xE ) =: ik x0 F (Ez, xE ).
0
k
Putting all this together, we arrive at the following formula for the function k (Ez, xE):
!
g
n
X
X
1
l
l
kl i x0
g B (z ) F (Ez, xE)
k (Ez, xE) =
l=1
=1
=: k (Ez, xE )F (Ez, xE ).
The functions k introduced here depend on the insertion points z , xi , the charges g and
the values xk0 of the bosonic field at the boundary. Additional information, e.g., on the
unknown function F (Ez, xE), is not needed. This concludes our derivation of Eq. (24).
3.3. The twisted KnizhnikZamolodchikov equations
We now start to exploit formula (24) together with the Sugawara construction of
the energy-momentum tensor T to compute the effect of inserting the field T into our
correlation functions. If we include one extra field g (u) into Eq. (24), differentiate with
respect to u and set g = 0, we obtain as a consequence of J(u) = 1g u g (u)|g=0

J(w)J(u)1 (z1 ) n (zn ) (x1 ) (x2g+1 )

" n
p
X
g g
P (z ) P (z )
=
P (u) P (w) w z u z
,=1
g X
g
n
X
X
k (Ez, xE )wk1 g P (z )
k l wk1 ul1
+
P (u) P (w) w z
P (w) P (u)
k=1 =1
k,l=1
!#
d
P (u)
1
1 l
k1
kl B (u)w
F (Ez, xE ).
+
P (w) du w u
+
At this stage we can subtract the term 1/(w u)2 , multiply by a factor 1/2 and perform
the limit u w. A short and elementary computation gives

T(w)1 (z1 ) n (zn ) (x1 ) (x2g+1 )

"
p
g n
n
1 X g g P (z ) P (z ) X X k (Ez, xE )wk1 g P (z )
+
=
2
P (w) w z w z
P (w)
w z
,=1
k=1 =1
#
g
g
00
1 l
B (w)0 wk1
1 X kl
1 X k l wk+l2
P (w)

F (Ez, xE).
+
2
P (w)
4 P (w) 2
P (w)
k,l=1
k,l=1
The same correlator has been computed in Eq. (21) directly with the help of operator
product expansions between the Virasoro field T and g , from (19), (20). Comparison
398
of the residues at w = z in the two different expressions gives the z-components of the
KnizhnikZamolodchikov equations:
" n p
#
g
X h
X P (z ) g g
X
g k zk1
F (Ez, xE ).
z F (Ez, xE ) =
z xi
P (z ) z z
P (z )
6=
k=1
If we restrict to the case of two twist field insertions, i.e., g = 0, the last term vanishes.
Putting x1 = 0 and hence P (w) = w, formula (14) is recovered.
Computation of the residues at w = xi yields a formula for the derivative of F with
respect to the position xi of twist fields. Using that
!
00
1X 1
P (w)
=
Resxi
8
xi xj
4 P (w)
j 6=i
and
Resxi
g
1 l
X
kl
B (w)0 wk1
P (w)
k,l=1
!
=
g
X
k,l=1
1
kl
Z
l
1
1 xik1
d
2 P ( ) xi
we obtain
"
!2
g
n
X
1
g P (z ) X
k1
Q
+
k (Ez, xE)xi
xi F (Ez, xE ) =
2 j 6=i (xi xj )
xi z
=1
k=1
#
I
g
xik1
1X 1
1 X 1
kl
d F (Ez, xE ).
8
xi xj
4
P ( )( xi )
j 6=i
k,l=1
(27)
To conclude this section, we come back to the original correlators G(Ez, xE) of bulk-fields
g (z, z ) and boundary twist fields (xi ). The differential equations they obey will be
formulated with the help of the following functions:
1
0 (z, w) =
P (z)
X
g

P (w)
P (w)
zk1
1
kl
,
B l (w) B l (w)
zw
z w
P (z)
k,l=1
2g+1
g
X

1 X 1
1
P (z)
zk1
1
kl
B l (z) B l (z) ,
0 (z) =
2
z xi
z z
P
(z)
P
(z)
i=1
k,l=1
g
X
zk1
1 l
x0 .
kl
(z) =
P
(z)
k,l=1
Taking into account the equation g (z, z ) = g (z)g (z), we conclude that the correlation
functions G(Ez, xE ) must satisfy the following set of first order linear differential equations:
G(Ez, xE) = A G(Ez, xE),
for all = z , z , xi
with the connection matrices Az , Az and Axi being defined by
(28)
Az =
399
g g 0 (z , z ) + g2 0 (z ) + ig (z ),
(29)
g g 0 (z , z ) + g2 0 (z ) + ig (z ),
(30)
6=
Az =
6=
!2
n
X
1
g 0 (x, z ) + i (x)
Axi = lim (x xi )
2 xxi
=1
I
g
xik1
i Hi (xi ) 1 X 1
kl
d.
8Hi (xi )
4
P ( )( xi )
k,l=1
(31)
Q
We have introduced the function Hi (xi ) := j 6=i (xi xj ). Note that the term in brackets
has a simple pole at x = xi which we cancel by the extra factor x xi before performing
the limit. Eqs. (28) through (31) constitute the main result of this section.
4. Construction of correlation functions

It remains to reconstruct the correlation functions G(Ez, xE ) from the system of linear first
order differential equations that we obtained in the previous section. Integration of the
equations is, in principle, straightforward, but it leaves one constant factor undetermined.
The latter is found explicitly in terms of the boundary conditions. Moreover, we shall
manage to express the correlators G(Ez, xE ) in terms of rather elementary building blocks.
4.1. Integration of the z-connection
To begin with, we simplify our problem by fixing the insertion points of the boundary
twist fields (xi ) and considering only the dependence of GxE (Ez) = G(Ez, xE ) on the positions
(z , z ) of bulk fields. This means that we have to integrate the z-connection Az dz +
Az d z defined in Eqs. (29), (30). The result will be written with the help of two functions
G0 (z, w) and S0 (w) which are given by
Zz
G0 (z, w) := 2 Re
0 (, w) d,
x
"
Zz
S0 (z) := lim 2 Re
(32)
#
0 ( ) d log(v v)
Re log P (v) .
vx
(33)
Here, the point x is chosen to lie in the Dirichlet-interval ]x1 , x2 [ . The integrand in Eq. (32)
is regular on the real axis and hence the integral is well defined. One can easily see that
its value neither depends on the starting point x ]x1 , x2 [ nor on the choice of the curve
from x to z. In contrast, the integrand 0 ( ) in our definition of S0 (z) has a simple
pole on the real axis. This is the reason why we subtract the divergent term log(v v)
before taking the limit v x. The contribution Re log P (v) is added to render the whole
400
function independent of the integration contour and, in particular, of the starting point x.
More details on the definition of S0 (z) and a discussion of its properties can be found in
Appendix B.
The two functions we have just defined possess a number of abstract properties that
characterize them uniquely. First of all, it is not difficult to see that G0 (z, w) is simply a
Greens function on the upper half-plane, i.e., it obeys
z G0 (z, w) = 4(z w),
for Im z > 0
and the boundary conditions
G0 (z, w) = 0, for z Ni .
Im z
A standard computation shows that G0 (z, w) is symmetric in its arguments, G0 (z, w) =
G0 (w, z).
The function S0 (z), on the other hand, is harmonic throughout the whole upper halfplane, i.e., z S0 (z) = 0. It diverges at the boundary with a leading singularity of the form

Di
.
(34)
S0 (z) log |z z | + , for Re z
Ni
G0 (z, w) = 0,
for z Di ,
We are now in a position to integrate the differential equations (29), (30) for the correlators
of bulk fields g (z, z ). The result is given by
ZzE
log GxE (Ez) =
(Az d + Az d ) + (w)
E
w
E
n
1 X
g g G0 (z , z )
2
,=1
6=
n
X
=1
g2 S0 (z ) +
g+1 Z
n
X
ig X
=1
i=1 D
xi0
G0 (, z) d.
Im
(35)
In the first line we have chosen some arbitrary curve in the configuration space of n
particles in the upper half-plane starting at points w with Im w > 0. The integration
constant (w)
E depends on the choice of the starting point and has to be fixed such that
the resulting function log GxE (Ez) satisfies the desired boundary conditions. In passing to the
second line, we have extended the integration to w = x on the boundary and inserted the
definitions (32), (33) of the functions G0 and S0 . Then we use the auxiliary formula
Z
g+1
g Zz
X
X
1
k1
1 l
i
G0 (z, ) d = 2 Re
x0
x0 d + x10 .
kl
4
Im
P
(
)
i=1
k=1
Di
To prove this formula one should notice that the function on the l.h.s. of the equation is
harmonic in the upper half-plane, that it satisfies Neumann boundary conditions along
the intervals Nk and that it approaches the constant values xk0 for z Dk . By explicit
401
computation, one can establish the same behaviour for the function on the r.h.s. Since these
properties are sufficient to determine the functions uniquely, the desired equation follows.
We employ it to bring the third term of Eq. (35) into a form that allows the most explicit
control of the boundary behaviour and hence is quite appropriate for fixing the remaining
(x).
4.2. Integration of the x-connection
Before we address the integration of the full connection (29)(31) below, we investigate
another simplified situation in which there are no bulk fields present. Consequently, the
charges g in Eq. (27) can be set to zero, and we are confronted with the problem of
x ) := G(E
x ):
solving the following equation for Z (E
!
2
g
X
1
x) =
kl
il x0 xik1
2Hi i log Z (E
k=1
I
g
x k1 Hi
1
1 X 1
i Hi
kl
i
d,
4
2
P ( )( xi )
k,l=1
l
where Hi denotes the function Hi = j 6=i (xi xj ) as before, and = {l x0 }. These

differential equations were solved by Zamolodchikov in [42]. Here we simply quote the
final result:
x) =
Z (E
2Y
g+1
(xi xj )
1/8
1/2
det
i>j
() e
i/(8)
g
X
k x0 l x0 kl .
k,l=1
4.3. The correlation functions G(Ez, xE )

The results of the previous two subsections can be combined into explicit expressions
for the correlators G(Ez, xE) of bulk fields g (z, z ) and boundary twist fields (x). To see
this we note that solutions of the differential equations (28) can be found by integrating
the connection 1-form Aw dw + Aw d w + Ai di along an arbitrary curve (t) =
(zE (t), xE (t)), t [0, 1], that ends at the point (Ez, xE) in the (2n + 2g + 1)-dimensional real
configuration space.
With some care (see the first subsection) we can start the integration with all insertion
points being on the real axis. Furthermore, we may choose such that all twist-fields are
moved to their final position at xE before we begin moving the bulk fields into their desired
locations.
In more mathematical terms this means that consists of two parts = (2) (1)
with t zE(1) (t [0, 12 ]) = 0 and t xE(2) (t [ 12 , 1]) = 0. As long as t 6 12 , the bulk fields are
located at points w = (t) belonging to the first Dirichlet interval D1 = ]x1 , x2 [ . This
implies 0 (x, (t)) = 0 for t [0, 12 ], so that one term in our expression (31) for Axi
drops out. Hence, we are precisely in the situation considered in the previous subsection,
x ).
and the integration of our connection 1-form over in the interval t [0, 12 ] gives Z (E
402
When we continue the integration to t = 1, we add the expression for log GxE (Ez) computed
in Eq. (35). After some rewriting, the following result for the correlations function G(Ez, xE )
is obtained:
!
!
n
n
X
X
2
(n1)
x ) exp
g S0 (z ) exp
ig x0 (z ) .
(36)
G(Ez, xE) = Z (E
=1
=1
(n1)
Here, x0 (z ) denotes the potential that is created by n 1 charges g
at points z 6= z
in the presence of the boundary with mixed boundary conditions,

Z
X
1
(z
)
=
d 2 wG0 (z , w)
g (w z )
x(n1)
0
2
6=
Im w>0
1
4
g+1 Zx2i
X
i=1x2i1
xi0
G0 (, z ) d.
Im
It is quite instructive to interpret each of the three factors in our final expression for the
correlation function G(Ez, xE) directly within conformal field theory. For the moment, let
us specify the number of bulk fields in G by some extra superscript, i.e., we shall write
G(Ez, xE) = G(n) (Ez, xE). Now consider the object

1
G(n) (Ez, xE )
,
log
(z ) :=
ig
exp(g2 S0 (z ))G(n1) (Ez0 , xE)
where zE0 denotes the set of n 1 bulk coordinates z , 6= . It is easy to determine the
behaviour of (z ) as a function of the bulk coordinate z from the bulk and bulk-boundary
operator product expansions of the fields g (z , z ), cf. Section 2.3:
X
g (z z ),
z (z ) = 2
6=
(z ) = 0, for z Ni .
Im z
These properties characterize the function (z ) uniquely, and by standard formulas from
(z ). An iteration of this construction along
electrostatics we obtain that (z ) = x(n1)
0
x ) = Z (E
x ) leads to our product formula (36) for the correlation function
with G(0) (E
G(Ez, xE).
x ) is a partition function
The three factors can be interpreted as follows: Z (E
corresponding to some line charge distribution provided by the twist fields alone; the term
(z ) = xi0 ,
n
X
for z Di ,
g x(n1)
(z )
0
=1
is the electrostatic potential corresponding to a configuration of n point particles with

charges g1 , . . . , gn located at the points z1 , . . . , zn and line charges distributions along the
Dirichlet intervals. The term
n
X
g2 S0 (z )
=1
403
can be interpreted as a renormalized electrostatic self-energy of the point particles located

at z1 , . . . , zn .
4.4. Path integral approach and extensions
Before we conclude, let us briefly sketch how the theory can be formulated in the
path integral approach. This is important for the following two reasons: first, as long as
one is only interested in free bosons, the path integral approach is a powerful alternative
to our analysis above involving KnizhnikZamolodchikov connections. The path integral
formulation allows for a more direct computation of the correlation functions (but it does
not easily extend to non-abelian group targets.) Secondly, using path integrals we will be
able to describe rather easily possible extensions of our analysis to compact targets and to
D-branes with B-fields.
To begin with, we consider once more the familiar situation of a non-compact
1-dimensional target and Dirichlet parameters xi0 . We denote by G the Greens function
of the Laplacian on the upper half-plane,
z G(z, w) = (z w),
subject to the boundary conditions
G(x, w) = 0,
for x D,
y G(x, w) = 0,
for x N.
The Gaussian measure with covariance G and mean 0 is denoted by G , and we use for
the corresponding random variable. With the help of the Dirichlet parameters xi0 , we define
a real-valued function on the upper half-plane by
= 0,
|Di = xi0 ,
y |Ni = 0,
i = 1, . . . , g + 1.
The effect of the Dirichlet parameters is incorporated through a shift of the random variable
by which gives us the bosonic field X = + . It appears when we construct the basic
fields of the theory, namely the vertex operators
g (z, z ) = :ei gX(z,z) : .
In this framework we could recover the correlation functions above by integrating products
of fields g (z, z ) using the Gaussian measure.
When the free boson X takes values in a circle S 1 , the boundary conditions depend
both on Dirichlet parameters xi0 and on Neumann parameters denoted by yi0 . The Neumann
parameters determine the Dirichlet parameters of the T-dual theory and can be thought of as
the strength of constant Wilson lines turned on along a Neumann direction. In Section 3, we
have worked in a local chart with the Neumann intervals cut out. Information on Neumann
parameters is, therefore, lost unless we take a second chart into account which has cuts
along the Dirichlet intervals. As above, we can derive KnizhnikZamolodchikov equations
404
for each chart; the full correlation functions of the compactified theory are to be built up
from the respective solutions in such a way that the boundary conditions are met.
The path integral computation of correlators is rather easy to adjust to the compactified
situation: if we set Neumann parameters to zero and restrict attention to the fields g (z, z ),
we can use precisely the same formulas as above.
In order to incorporate non-vanishing Neumann parameters yi0 and to compute more
(c)
general correlators for fields g,g (z, z ) with (g, g ) taken from an even, self-dual Lorentzian
lattice, we introduce a real-valued function on the upper half-plane defined by
= 0,
|Ni = yi0 ,
y |Di = 0,
i = 1, . . . , g + 1.
Then, we set
$g,z (w) := gG0 (w, z) + (w),
where G0 denotes the Greens function of the Laplacian on the upper half-plane with
interchanged Dirichlet and Neumann boundary conditions, i.e.,
G0 (w, z) = (w z),
y G0 (x, z) = 0,
0
G (x, z) = 0,
for x D,
for x N.
We now introduce the disorder operators Dg (z, z ) satisfying

Dg (z, z )F [dX] = F [dX + d$g,z ],
for any functional F . The left- respectively right-moving chiral vertex operators can then
be written as
g (z) = g/2 (z, z )Dg/2 (z, z ),
g (z) = g /2 (z, z )Dg/2 (z, z );
see also [17] for more details and for an application to soliton quantization in 2-dimensional
theories. The basic fields of the compactified theory are products
z),
g(c)
,g (z, z ) = g (z)g (
where (g, g ) lies in some even, self-dual Lorentzian lattice. For another approach to the
rational compactified boson, the reader is referred to [22].
Another extension would involve the appearance of B-fields on our D-branes. This has
attracted some interest recently, because of its relation with non-commutative geometry,
see, e.g., [3,15,16,35,36] and references therein. Non-trivial B-fields can only exist if one
of our branes is at least 2-dimensional. For simplicity, we shall focus on a pair of a Dp- and
a D0-brane. The field strength on the Dp-brane will be denoted by B. In terms of boundary
conditions for a multi-component free bosonic field, the situation is described as follows
t Xa (t, 0) = 0
and Xa (t, ) = Bba t Xb (t, ),
for a, b = 1, . . . , p.
The spectrum of the associated boundary condition changing operators and the Greens
functions in the presence of two twist fields have been discussed at various places (see,
405
e.g., [11,36]). Our techniques from Sections 3 and 4 allow to extend such investigations to
the case of multiple twist insertions. Instead of giving the details here, we simply state how
one has to adjust the path integral computation to the new scenario. This is rather easy:
all it requires is to replace the function G above by some matrix valued Greens function
GB = (Gab
B ). The latter is a Greens function for the Laplacian 1p on the upper half-plane
(1p denotes the p-dimensional identity matrix), subject to the boundary conditions
Gab
B (x, w) = 0,
for x D,
a
cb
y Gab
B (x, w) = iBc x GB (x, w),
for x N.
With the help of this function, the calculation of correlators proceeds as before.
5. Outlook
We have succeeded in decomposing the complete bulk and boundary correlators in the
presence of DN-transitions into functions with rather natural interpretations both from
the point of view of electrostatics and from the CFT perspective. This is useful for carrying
out the remaining step in the computation of string amplitudes, namely the integration over
insertion points of fields on the world-sheet. The calculation of such string amplitudes
gives effective actions involving a hyper-multiplet which comes with the twist fields. To
leading order, the bosonic part of these actions can be found in [14,25,27]. Multiple twist
insertions allow to compute higher order corrections.
When we turn on a B-field, the string amplitudes may be described through field
theories on some non-commutative space. It was suggested in [36] that these theories are
related to some model on an ordinary commutative space through a complicated non-linear
transformation. This statement can be checked order by order in the effective description.
After the appropriate (but straightforward) extension to non-vanishing B-fields, the
considerations presented above may be used to perform a similar analysis for theories
which contain a hyper-multiplet .
Keeping the bulk insertions fixed, the sequence of correlators with arbitrarily many twist
field insertions can be viewed as building blocks of the perturbation series of a relevant
perturbation by the twist field. This tachyon condensation is responsible, e.g., for the
formation of D0D2 bound states, as discussed in [23]. Upon integrating over twist field
x ) exp{g2 S0 (z)}, one would arrive at oneinsertion points in the one-point functions Z (E
point functions which characterize the boundary theory after tachyon condensation. Sens
approach [37] and the results of [33] allow one to circumvent the relevant boundary flow
and to replace it by a combination of marginal bulk and boundary deformations. However,
some questions as to the equivalence of both procedures remain open, and it might be useful
to have an independent check of these methods. The correlation functions constructed here
provide a starting point.
For applications to superstring theory, it is mandatory to extend our analysis to free
fermions. This does not pose serious problems, since systems of an even number of
fermions can be bosonized.
406
Problems of the type of our free boson problem are encountered in general boundary
CFT as soon as the parent CFT on the plane admits different boundary conditions. For
some general results on the rational case, see [2022]. The spectrum of boundary condition
changing operators can be derived as in Section 2, once boundary states for the constant
boundary conditions are known. Again, the computation of correlators becomes non-trivial
if boundary conditions with different gluing automorphisms are combined. In non-abelian
WZW models, which constitute and important generalization of the free boson case, the
Sugawara construction can be exploited in a similar fashion as for the free boson and
leads to twisted, non-abelian KnizhnikZamolodchikov equations. The partition functions
counting BCCOs in non-abelian boundary WZW models are linear combinations of the
twining characters investigated in [18] (see also [5] and references therein). Apart from the
models with affine Lie algebra symmetry, there is the rather large class of so-called quasirational CFTs [28] on the plane for which generalizations of KnizhnikZamolodchikov
equations exist even without a Sugawara form for the energy-momentum tensor [2]. It
might be interesting to see how such structures extend to boundary CFT.
Acknowledgements
We are indebted to P. Etingof for very helpful discussions. J.F. and O.G. thank
C. Schweigert for interesting discussions. The work of O.G. is supported in part by the
Department of Energy under Grant DE-FG02-94ER-25228 and by the National Science
Foundation under Grant DMS-94-24334. O.G. also acknowledges the Clay Mathematics
Institute for support.
O.G., A.R. and V.S. are grateful to the Research Group M at ETH Zrich for the
warm hospitality extended to them.
Appendix A. Proofs of Lemma 1 and Lemma 2

Proof of Lemma 1. We start from the usual expression for T in terms of J and rewrite it
until we can perform the limit w1 w2 .
J(w1 )J(w2 )
1
(w1 w2 )2

= J< (w1 )J(w2 ) + J(w2 )J> (w1 ) + J> (w1 ), J(w2 )
1
(w1 w2 )2
1
(w1 /w2 )1/2 + 12 (w2 /w1 )1/2 1
= J< (w1 )J(w2 ) + J(w2 )J> (w1 ) + 2
.
(w2 w1 )2
We can now perform the limit w1 w2 =: w to recover the generating field T(w) from
the last formula

1
1 1
T(w) = J< (w)J(w) + J(w)J> (w) +
,
2
16 w2
407
where we used
u1/2 + u1/2 2 1
= . 2
u1
2(1 u)2
8
lim
Proof of Lemma 2. The derivation of the commutation relation with J> (w) is straightforward. So, let us turn directly to the calculation of the commutator with T> (w). Recall that
h = g2 /2. Then,

T> (w), g (z)
h

i
T> (w), eigX< (z) eigX> (z) + eigX< (z) T> (w), eigX> (z)
=
2z
h

i
eigX< (z) ig T> (w), X< (z) h X< (x), [X< (x), T> (w)]
=
2z

+ ig T> (w), X> (z) + h X> (z), [X> (z), T> (w)] eigX> (z)
h
X

i
=
eigX< (z) ig T> (w), X(z) + h
wn2 zn r+s,n
2z
r,s<0,n>1
!!
X
n2 n
w
z r+s,n eigX> (z)
=
=
=
=
r,s>0,n>1
!!
X
1
ig
n2 n
z X(z) + h
eigX> (z)
e
w
z n
wz
wz
n>1

h

1
1
z
i
1
z eigX< (z) eigX> (z) + h
g (z)
w z 2z
w (w z)2 wz

1 h
1
1
1
z g (z) +
g (z) + h
g (z)
wz
wz z
(w z)2 z(w z)
h
1
z g (z) +
g (z).
wz
(w z)2
i
2z
h
igX< (z)
In the process of this computation we have inserted the commutation relation between
Ln , ar and Eq. (5). The rest involves only standard algebraic manipulations. 2
Appendix B. The function S0 (z)

In this appendix we want to explain a number of properties of the function S0 (z) that is
introduced in Section 4.1. To show that the limit limvx exists, we insert the definition of
0 ( ) into Eq. (33). After splitting off all non-singular terms in 0 we obtain:
"
Zz
S0 (z) = lim 2 Re
0 ( ) d log(v v)
Re log P (v)
vx
408
"
Zz
= lim 2 Re
vx
1
d log(v v)
+ regvx

log(z z ) log(v v)
+ reg0 vx .
= lim log(v v)
vx
Since the singularity from the integral cancels against the term log(v v),
the limit can be
taken.
Our second aim is to understand that S0 (z) S0x (z) does not depend on the choice of x.
Let us displace x by some small amount a R such that x + a is still in the Dirichlet
interval D1 . Comparison of S0x (z) and S0x+a (z) gives
S0x (z) S0x+a (z)
"
v+a
Z
0 ( ) d Re log P (v) + Re log P (v + a)
= lim 2 Re
vx
"
= lim 2 Re
v+a 2g1
Z
X
vx
"
= lim Re
vx
v
2X
g+1
i=1
i=1
P (v + a)
1/2
d + Re log
xi
P (v)
#
P (v + a)
= 0.
log(v + a xi ) log(v xi ) + Re log
P (v)
In passing to the second line we omitted all terms in the integrand which vanish when
comes close to the real axis.
Finally, we investigate the behaviour of S0 (z) at the boundary. Basically, one repeats
the analysis we have sketched above in our discussion of limvx . If the end-point z of
our integration approaches one of the Dirichlet intervals, this leads to the singularity
log |z z |. In the argument one needs that the quotient P (z)/P (z) in front of the
singular term 1/(z z ) satisfies limzx P (z)/P (z) = 1 for x Dk . This is no longer
true when z is sent to the real axis in one of the Neumann intervals Nk . In fact, upon
moving x from a Neumann into a Dirichlet interval, the polynomial P (x) changes sign,
causing the quotient P (z)/P (z ) to surround the origin of the complex plane once. After
taking the square root we conclude that limzx P (z)/P (z) = 1 for x Nk and hence
S0 (x) log |z z | near the Neumann intervals.
References
[1] I. Affleck, A.W.W. Ludwig, Critical theory of overscreened Kondo fixed points, Nucl. Phys.
B 360 (1991) 641.
[2] A.Yu. Alekseev, A. Recknagel, V. Schomerus, Generalization of the KnizhnikZamolodchikov
equations, Lett. Math. Phys. 41 (1997) 169, hep-th/9610066.
[3] A.Yu. Alekseev, A. Recknagel, V. Schomerus, Non-commutative world-volume geometries:
branes on SU(2) and fuzzy spheres, J. High Energy Phys. 9909 (1999) 023, hep-th/9908040.
[4] L. Alvarez-Gaume, J.B. Bost, G. Moore, P. Nelson, C. Vafa, Bosonization on higher genus
Riemann surfaces, Commun. Math. Phys. 112 (1987) 503.
409
[5] L. Birke, J. Fuchs, C. Schweigert, Symmetry breaking boundary conditions and WZW orbifolds,
hep-th/9905038.
[6] I. Brunner, M.R. Douglas, A. Lawrence, C. Rmelsberger, D-branes on the quintic, hepth/9906200.
[7] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Adding holes and crosscaps to the superstring,
Nucl. Phys. B 293 (1987) 83; Loop corrections to superstring equations of motion, Nucl. Phys.
B 308 (1988) 221.
[8] J.L. Cardy, Conformal invariance and surface critical behavior, Nucl. Phys. B 240 (1984) 514;
Effect of boundary conditions on the operator content of two-dimensional conformally invariant
theories, Nucl. Phys. B 275 (1986) 200.
[9] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324
(1989) 581.
[10] J.L. Cardy, D.C. Lewellen, Bulk and boundary operators in conformal field theory, Phys. Lett.
B 259 (1991) 274.
[11] B. Chen, H. Itoyama, T. Matsuo, K. Murakami, pp0 system with B-field, branes at angles and
noncommutative geometry, hep-th/9910263.
[12] E. Corrigan, D.B. Fairlie, Off-shell states in dual resonance theory, Nucl. Phys. B 91 (1975)
527.
[13] D.-E. Diaconescu, C. Rmelsberger, D-branes and bundles on elliptic fibrations, hepth/9910172.
[14] M.R. Douglas, Gauge fields and D-branes, J. Geom. Phys. 28 (1998) 255, hep-th/9604198.
[15] M.R. Douglas, C. Hull, D-branes and the noncommutative torus, J. High Energy Phys. 9802
(1998) 008, hep-th/9711165.
[16] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, The geometry of WZW branes, hep-th/9909030.
[17] J. Frhlich, P.A. Marchetti, Bosonization, topological solitons and fractional charges in twodimensional quantum field theory, Commun. Math. Phys. 116 (1988) 127.
[18] J. Fuchs, B. Schellekens, C. Schweigert, From Dynkin diagram symmetries to fixed point
structures, Commun. Math. Phys. 180 (1996) 39, hep-th/9506135; Twining characters, orbit
Lie algebras, and fixed point resolution, q-alg/9511026.
[19] J. Fuchs, C. Schweigert, Branes: from free fields to general conformal field theories, Nucl. Phys.
B 530 (1998) 99, hep-th/9712257.
[20] J. Fuchs, C. Schweigert, Orbifold analysis of broken bulk symmetries, hep-th/9811211.
[21] J. Fuchs, C. Schweigert, Symmetry breaking boundaries I. General theory, hep-th/9902132.
[22] J. Fuchs, C. Schweigert, Symmetry breaking boundaries II. More structures; examples, hepth/9908025.
[23] E. Gava, K.S. Narain, M.H. Sarmadi, On the bound states of p- and (p + 2)-branes, Nucl. Phys.
B 504 (1997) 214, hep-th/9704006.
[24] M. Gutperle, Y. Satoh, D-branes in Gepner models and supersymmetry, Nucl. Phys. B 543
(1999) 73, hep-th/9808080; D0-branes in Gepner models and N = 2 black holes, hepth/9902120.
[25] A. Hashimoto, Dynamics of DirichletNeumann open strings on D-branes, Nucl. Phys. B 496
(1997) 243, hep-th/9608127.
[26] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett. A 4
(1989) 251.
[27] J.M. Maldacena, Black holes in string theory, hep-th/9607235.
[28] W. Nahm, Quasi-rational fusion products, Int. J. Mod. Phys. B 8 (1994) 3693, hep-th/9402039.
[29] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477
(1996) 407, hep-th/9606112.
[30] J. Polchinski, Dirichlet branes and RamondRamond charges, Phys. Rev. Lett. 75 (1995) 4724,
hep-th/9510017; TASI lectures on D-branes, hep-th/9611050.
410
[31] A.K. Raina, An algebraic geometry study of the bc system with arbitrary twist fields and
arbitrary statistics, Commun. Math. Phys. 140 (1991) 373.
[32] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185, hepth/9712186.
[33] A. Recknagel, V. Schomerus, Boundary deformation theory and moduli spaces of D-branes,
Nucl. Phys. B 545 (1999) 233, hep-th/9811237.
[34] A. Sagnotti, Open strings and their symmetry groups, in: G. Mack et al. (Eds.), Non-Perturbative
Methods in Field Theory, Lecture Notes Cargse, 1987; Surprises in open-string perturbation
theory, Nucl. Phys. Proc. Suppl. 56B (1997) 332; hep-th/9702093.
[35] V. Schomerus, D-branes and deformation quantization, J. High Energy Phys. 9906 (1999) 030,
hep-th/9903205.
[36] N. Seiberg, E. Witten, String theory and noncommutative geometry, J. High Energy Phys. 9909
(1999) 032, hep-th/9908142.
[37] A. Sen, SO(32) spinors of type I and other solitons on brane-antibrane pair, J. High Energy
Phys. 9809 (1998) 023, hep-th/9808141; Descent relations among bosonic D-branes, hepth/9902105; Non-BPS states and branes in string theory, hep-th/9904207.
[38] S. Stanciu, D-branes in Kazama-Suzuki models, Nucl. Phys. B 526 (1998) 295, hep-th/9708166.
[39] A. Strominger, C. Vafa, Microscopic origin of the BekensteinHawking entropy, Phys. Lett.
B 379 (1996) 99, hep-th/9601029.
[40] E. Witten, Bound states of strings and D-branes, Nucl. Phys. B 460 (1996) 335, hep-th/9510135.
[41] E. Witten, D-branes and K-theory, J. High Energy Phys. 9812 (1998) 019, hep-th/9810188.
[42] Al.B. Zamolodchikov, Conformal scalar field on the hyperelliptic curve and critical Ashkin
Teller multipoint correlation functions, Nucl. Phys. B 285 (1987) 481.

New brane solutions from Killing spinor equations

Ali Kaya
Center for Theoretical Physics, Texas A & M University, College Station, TX 77843, USA
Abstract
In a recent paper, we have pointed out a relation between the Killing spinor and Einstein equations.
Using this relation, new brane solutions of D = 11 and D = 10 type IIB supergravity theories are
constructed. It is shown that in a brane solution, the flat world-volume directions, the smeared
transverse directions and the sphere located at a fixed radial distance can be replaced with any
Lorentzian Ricci flat, Euclidean Ricci flat and Einstein manifolds, respectively. The solution obtained
in this fashion is supersymmetric when the Ricci flat and Einstein manifolds have Killing spinors.
We generalize intersecting brane solutions, in which M5-, M2- and D3-branes also wrap over the
cycles determined by the Khler forms of Ricci flat Khler manifolds. New, singular, Ricci flat
manifolds as (generalized) cones over the U(1) bundles over Ricci flat Khler spaces are constructed.
These manifolds have covariantly constant spinors and give rise to new, supersymmetric, Ricci flat
compactifications of non-gauged supergravity theories. We find M2- and D3-brane solutions, which
asymptotically approach these singular vacua. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.27.+d
Keywords: p-branes; Supersymmetric solitons; Killing spinors
1. Introduction
Studying brane solutions of low energy supergravity equations is proved to be an
important way of searching non-perturbative properties of string/M theories. These
solutions play a crucial role both in the conjectured duality symmetries between seemingly
different string/M theories and in the AdS/CFT correspondence. Generically, a brane
solution has the interpretation of a p-dimensional extended black hole, which can be
characterized by few constants like the mass and the charges of the antisymmetric
tensor fields. The metric along the world-volume directions have Poincare, and along
the transverse directions have rotational invariances. The geometry is asymptotically flat
and the number of transverse directions dictates the radial coordinate dependence of the
ali@rainbow.tamu.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 8 3 - 2
412
A. Kaya / Nuclear Physics B 583 (2000) 411430
metric functions. By smearing out some transverse directions, solutions with slower falloff properties can also be constructed (for a review see, e.g., [1,2]).
Generalizations of the usual brane solutions have been studied from different points of
view. In [3], it has been shown that the transverse 7-sphere of the membrane solution can be
replaced with any Einstein manifold. The brane solutions having transverse hyper-Khler
spaces have been studied in [4]. In [5], a fivebrane solution wrapping on the manifold K3
has been constructed. Brane solutions with Ricci flat world-volumes have been obtained
in [6,7]. In [8,9], more general brane solutions with curved world-volume directions have
been studied. An intrinsic metric, representing non-trivially embedded D3-branes, has been
given in [10]. Brane solutions which are product of an AdS space with an Einstein space
have been constructed in [1113].
In this paper, we will systematically extend these observations by using a theorem proved
in [5] which states that, under certain conditions, the existence of a Killing spinor implies
the Einstein equations. This enables one to concentrate on the first order Killing spinor
equations without worrying about more complicated, second order Einstein equations. The
spaces having Killing spinors will play a crucial role in the constructions.
The organization of the paper is as follows. In Section 2, we review the theorem of [5].
In Sections 3 and 4, we specifically consider M2-, M5- and D3-branes. In Section 3, we
generalize the well known solutions and obtain branes having Ricci flat world-volumes
and smeared transverse directions, and non-spherical cross sections. In Section 4, using
Khler forms of Ricci flat Khler spaces, we construct intersecting brane solutions which
can also be interpreted as wrapping branes over the cycles determined by Khler forms. In
Section 5, we show that a generic brane background still obeys the field equations when
certain directions in the metric are replaced with more general spaces. In Section 6, we first
construct new, singular, Ricci flat manifolds, which have covariantly constant spinors, as
(generalized) cones over U(1) bundles over Ricci flat Khler manifolds. These manifolds
give rise to new supersymmetric compactifications of non-gauged supergravities. In the
same section, we find M2- and D3-brane solutions which asymptotically approach these
vacua. We conclude with some brief remarks in Section 7.
2. BPS solutions from Killing spinor equations: a theorem

It is well known that existence of a covariantly constant spinor on an Euclidean manifold
implies Ricci flatness. It turns out, this observation has a very useful generalization in the
supergravity theory context [5].
Let us consider a bosonic background (gMN ,FMNPQ ) of D = 11 supergravity theory [19]
which obeys:

1
1
FM PQR FNPQR gMN F PQRS FPQRS ,
(1)
RMN =
3
12
Q F QMNP =
1 MNPA1 ...A8
FA1 ...A4 FA5 ...A8 .

242
(2)
413
The linearized RaritaSchwinger equations on this background may be written as:

MNP DN P = 0,
where the supercovariant derivative DM is given by

1
1 P QRS
PQRSM M
FPQRS ,
DM = M +
144
8
(3)
(4)
and M is the usual covariant derivative acting on spinors. Let us further consider a
linearized spin 3/2 field M , which is obtained by the action of supercovariant derivative
on an arbitrary spinor :
M = DM .
(5)
Due to the invariance of D = 11 supergravity at the linearized fermionic level, this spin
3/2 field solves (3). To verify this claim, we insert (5) into (3) and, using only the 4-form
field equations, obtain

1
(6)
MNP DN DP = GMN TMN N = 0,
2
where GMN = RMN 12 gMN R is the Einstein tensor and

1
1
PQRS
PQR
FM
FNPQR gMN F
FPQRS
TMN =
3
8
(7)
is the energy momentum tensor. In setting (6) to zero, we have used the fact that the
background is chosen to obey Einstein equations (1).
Now consider a background which satisfies the 4-form field equations (2) but not
necessarily obeys the Einstein equations. Let us further assume the existence of at least
one Killing spinor obeying
DM 0 = 0.
Following (6), it is easy to see that

GMN TMN N 0 = 0.
(8)
(9)
Note that in deriving (6), we have only used the 4-form field equations which are also
assumed to be satisfied by the new background. (6) has been set to zero since that
background was chosen to obey Einstein equations. On the other hand, (9) is satisfied
since 0 is the Killing spinor. Therefore, (9) is still valid, even if the background is not
chosen to obey Einstein equations.
Although (9) is very close to the Einstein equations, one needs to impose further
conditions to proceed due to the Lorentzian signature of the metric. Since (GMN TMN ) is
symmetric, one can point-wise diagonalize it by elements of the group O(11). 1 However,
this does not give any new information, since O(11) transformation of O(1,10) gamma
matrices are not nice objects. Therefore, one should be able to split space and time
1 In claiming this, we assume that there is no topological obstruction to have a continuous map from space-time
to the group manifold O(11).
414
directions in (9). It turns out it is sufficient to assume existence of an orthonormal basis such
that (G0i T0i ) = 0, where 0 is the time and i is the spatial direction. When the indices of
(9) refer to this orthonormal basis, M = 0 component implies G00 T00 = 0. Along spatial
directions, when M = i, one can point-wise diagonalize the symmetric matrix (Gij Tij )
by O(10) transformations, under which the spatial gamma matrices still satisfy the same
Clifford algebra and thus invertible. Therefore, the diagonal entries of (Gij Tij ) should
also be zero, which then implies Einstein equations GMN TMN = 0. Summarizing above
considerations, we obtain the following result:
(i) if a background is known to solve the 4-form field equations (2),
(ii) if there exist an orthonormal basis such that G0i T0i = 0, 2
(iii) if it is known that there exist a Killing spinor on this background,
then this background, being preserve some fraction of supersymmetries, also solves the
Einstein equations of D = 11 supergravity.
The main advantage of the theorem for applications is that, to find a solution of the
second order Einstein equations, one can instead concentrate on the first order Killing
spinor equations which are of course easier to solve. One can also convince himself that it
is not hard to satisfy conditions (i) and (ii). Indeed, in constructing p-brane solutions, one
starts with ansatzs having the property (i) and (ii) (see, e.g., [1418]). As we will see for
a moment, condition (ii) can easily be satisfied by choosing the background to be static.
We finally note that a solution obtained in this fashion already preserves some fraction of
supersymmetries.
Although explicitly proven for D = 11 supergravity, it is possible to argue that such a
theorem can be established for all supergravities. The linearized supersymmetry invariance
of any supergravity theory requires
MNP DN DP = 0,
(10)
where DM = M + is the supercovariant derivative, is an arbitrary spinor and bosonic

fields refer to a background which obeys equations of motion. In (10), the derivatives of
the metric are expected to appear in the combination of the Einstein tensor. 3 Therefore,
after imposing all but Einstein equations, (10) should become

1
GMN TMN N = 0,
(11)
2
where TMN is the appropriate energy momentum tensor of the theory and in the last step
we have used the fact that the background obeys Einstein equations.
With this information, it is very easy to prove the theorem; dropping the condition that
the background satisfies Einstein equations and further assuming the existence of a Killing
spinor, one obtains (11) evaluated for the Killing spinor. With condition (iii), this implies
the Einstein equations. Therefore, if we replace (i) with
(i0 ) if the background satisfies all but Einstein equations,
MNP DN DP =
2 Here, we relax the corresponding condition imposed in [5]. I would like to thank C.N. Pope for the discussions
about this point.
3 This is due to the identity MNP = 1/2G
N
N P
MN .
415
then together with conditions (ii) and (iii) above, the result of the theorem should apply to
all supergravities.
To conclude this section, we finally comment on a simple way to satisfy condition (ii).
For a static background there exist an orthonormal basis such that G0i = 0. To see this,
using the fact that the background is static, we write the line element as

2
,
(12)
ds 2 = A2 dt 2 + dsM
2 is a line element on the Euclidean
where A is a time independent function and dsM
manifold M. One can easily show that, when expressed in the basis e0 = A dt and ei
(an orthonormal basis on M), the Ricci tensor of ds 2 obeys R0i = 0. This gives G0i = 0
since in any orthonormal basis g0i = 0. On the other hand, if the matter fields are chosen to
be static, then T0i = 0. 4 Therefore, one immediate way to satisfy condition (ii) is to take
background fields to be static. However, one should not rule out existence of interesting
non-static cases for which only the combination (G0i T0i ) = 0.
3. Generalized brane solutions

As mentioned in the introduction, a generic p-brane solution has Poincare invariance on
the world-volume and spherical symmetry along transverse directions. In this section, we
will focus on the M2-, M5- and D3-branes and try to obtain generalizations of the well
known solutions, by using the theorem of Section 2.
Definitions
Let Ld , Mm and Xn be Lorentzian Ricci flat, Euclidean Ricci flat and Einstein manifolds
of dimensions d, m and n, respectively. We will denote the basis one-forms on these spaces
as e , ea and e , where the indices , , . . . refer to Ld ; a, b, . . . refer to Mm and , , . . .
refer to Xn . We will assume that Ld and Mm have covariantly constant Killing spinors,
= 0,
(13)
a = 0,
(14)
and Xn has Killing spinors obeying

1
(15)
= r ,
2
where we set the inverse radius of Xn to 1. In these equations, we will not restrict
representations of 0-matrices to be irreducible. Indeed, we will view the Clifford algebras
on Ld , Mm and Xn as sub-algebras of a bigger Clifford algebra. For instance, in (15),
r can be any element of this bigger algebra which anti-commutes with and squares
2 and ds 2 ,
to identity. We will write the line elements on Ld , Mm and Xn as dsL2 d , dsM
Xn
m
respectively. The volume forms will be denoted by VL , VM and VX .
4 This can be regarded as the definition for matter fields to be static.
416
We will use 0 to denote differentiation with respect to the argument of the function, and
to mean equality up to a constant. The indices in tensor equations will refer to the tangent
space. Specifically, 0 and i will be used for time-like and spatial directions, respectively.
M2-brane
Let us start from the eleven dimensional membrane. Our aim is to write an ansatz, which
satisfies the conditions (i) and (ii) of the theorem of Section 2, and then work out the Killing
spinor equations. For the metric and 4-form field we assume
2
2
+ D 2 dsX
,
ds 2 = A2 dsL2 3 + B 2 dr 2 + C 2 dsM
m
n
F VM VX ,
(16)
(17)
where m + n = 7, is the Hodge dual corresponding to ds 2 and the metric functions

A, B, C and D are chosen to depend only on the coordinate r. We first note that the 4-form
field equations are identically satisfied without imposing any condition on the metric
functions. It is also easy to see that G0i = T0i = 0 in the orthonormal basis E = Ae ,
E r = B dr, E a = Cea and E = De . Therefore, the ansatz obeys the conditions (i) and
(ii) of Section 2. To find a supersymmetric solution of D = 11 supergravity, one needs to
impose conditions on metric functions which will ensure existence of at least one Killing
spinor obeying
A0
qe
r +
r = 0,
(18)
2AB
18C m D n
qe
= 0,
(19)
Dr = r +
18C m D n
C0
qe
a r +
r a = 0,
(20)
Da = a +
2CB
36C m D n
D0
qe
r +
r = 0,
(21)
D = +
2DB
36C m D n
where the indices and spinors refer to the basis one-forms defined above and the electrical
charge qe is defined to be the proportionality constant in (17) so that
qe
(22)
Fr = m n .
C D
The presence of the functions C and D in (22) is due to the fact that in (17) refers to the
line element (16). An easy way to solve these equations is to impose (13)(15) and
D = +
= 6,
(23)
which are of course consistent with each other. Then, the Killing spinor equations (18)
(20) imply
2qe
qe
qe
C0
D0
1
A0
=
=
=
(24)
,
,
+ .
m
n
m
n
AB 3C D
CB
3C D
DB
3C m D n D
Therefore, when (24) is satisfied, the background has at least one Killing spinor obeying
(13), (15) and (23). Now, by the theorem of Section 2, Einstein equations should be satisfied
417
identically. At this point it is instructive to verify this claim by a direct calculation. The
Einstein equations can be written as
0 2 0 0
A0 C 0 1
A
A0 D 0 1
4qe2
1
A
+m
+
+
n
=
,
(25)
2
AB
B AB
AC B 2
AD B 2 3C 2m D 2n
0 0
0 0
0 0
C
D
4qe2
1
1
1
A
+m
+n
=
,
(26)
3
B AB
B CB
B DB
3C 2m D 2n
0 2 0 0
A0 C 0 1
C
C 0 D0 1
2qe2
1
C
+3
+
+n
=
,
(27)
(m 1)
2
2
CB
B CB
AC B
CD B
3C 2m D 2n
0 2 0 0
A0 D 0 1
D
D
C 0 D0 1
2qe2
1
1
+
3
+
(n
1)
+
+
m
=
,
DB
B DB
AD B 2
CD B 2 3C 2m D 2n
D2
(28)
where we have used the fact that L3 and Mm are Ricci flat and Xn is Einstein. It is now
straightforward to check that, when the unknown functions A, B, C and D obey (24) they
also solve the Einstein equations, as required by the theorem.
There are four independent functions and three differential equations, which may be
thought to imply that functions are not constrained enough. However, this is simply
a manifestation of reparametrization invariance in coordinate r. We want to fix this
reparametrization freedom in a way, which will help us in solving the differential equations.
A convenient choice is
C m D n1 = r n1 + M,
(29)
where M = 2qe /(n 1). Then, the unknown functions obeying the first order coupled
equations (24) can be solved to give following solution

2
2
+ r 2 dsX
,
(30)
ds 2 = H 2/3 dsL2 3 + H 1/3 dr 2 + dsM
m
n
1
(31)
Fr = H 7/6(r H ) ,
2
where H = (1 + M/r n1 ).
M5-brane
For the generalized M5-brane we write the following ansatz,
2
2
+ D 2 dsX
,
ds 2 = A2 dsL2 6 + B 2 dr 2 + C 2 dsM
m
n
F VM VX ,
(32)
(33)
where m + n = 4 and the metric functions A, B, C and D are chosen to depend only on the
coordinate r. As for the M2-brane, the 4-from field equations are satisfied identically, and
G0i = T0i = 0 in the orthonormal basis E = Ae , E r = B dr, E a = Cea and E = De .
418
Therefore, the background satisfies conditions (i) and (ii) of Section 2. To solve the Killing
spinor equations, one can choose to obey (13)(15) and
a...b ... a...b...r = m!n!,
(34)
which are consistent with each other. One can then check that the Killing spinor equations
imply
qm
2qm
2qe
C0
D0
1
A0
=
=
=
(35)
,
,
+ ,
m
n
m
n
m
n
AB 3C D
CB
3C D
DB
3C D
D
where the magnetic charge qm is defined to be the proportionality constant in (33). Like in
the membrane case, a convenient way to fix the r-reparametrization invariance is to define
C m D n1 = r n1 + M, where M = 2qm /(n 1). Then, (35) can be solved to give the
following solution

2
2
+ r 2 dsX
,
(36)
ds 2 = H 1/3 dsL2 6 + H 2/3 dr 2 + dsM
m
n
1
(37)
Fa...b... = H 4/3(r H )a...b ... ,
2
where H = (1 + M/r n1 ).
D3-brane
The D3-brane solution of IIB supergravity in 10-dimensions has only non-vanishing
(anti)-self dual 5-form and the metric. The equations governing dynamics of these fields
can be written as [20]
1
(38)
RMN = F PQRS M FPQRSN ,
96
dF = 0, F = F.
(39)
The Killing spinors on such a background obey
i
NP ...Q M FNP ...Q = 0,
DM = M +
4 480
(40)
2 = I . For
= (11) and (11)
where is a Weyl spinor (11) = ; (11) = 0 . . . 9 , (11)
the generalized D3-brane we assume the form
2
2
+ D 2 dsX
,
ds 2 = A2 dsL2 4 + B 2 dr 2 + C 2 dsM
m
n
F (Vm Vn ) (Vm Vn ),
(41)
(42)
where m + n = 5, and the metric functions depend only on the coordinate r. The 5-form
field equations are identically satisfied and G0i = T0i = 0 in the basis E = Ae , E r =
B dr, E a = Cea and E = De . Therefore, the ansatz obeys the conditions (i0 ) and (ii) of
Section 2. On the other hand, if one choose to obey (13)(15) and impose further
= 24i,
(43)
the Killing spinor equations imply

q
A0
=
,
AB 4C m D n
q
C0
=
,
CB
4C m D n
q
D0
1
=
+ ,
DB
4C m D n D
(44)
419
where the dyonic charge q is defined to be the proportionality constant in (42). Therefore,
when A, B, C and D obey (44), all conditions of the theorem are satisfied and the
background should obey Einstein equations. Fixing r-reparametrization invariance by
C m D n1 = r n1 + M, we obtain the following solution,

2
2
+ r 2 dsX
,
(45)
ds 2 = H 1/2 dsL2 4 + H 1/2 dr 2 + dsM
m
n
1
(46)
F r = H 5/4 (r H ) ,
2
where M = q/(n 1) and H = (1 + M/r n1 ).
Interpretation and special cases
It is clear from the structure of the antisymmetric tensor fields and Killing spinor
projections that, in all solutions, Ld represents the curved world-volumes of the branes.
The dependence of the metric functions on the radial coordinate r implies that Mm and Xn
correspond to the smeared and actual transverse directions, respectively.
The solutions obtained so far have a very similar structure with the well known brane
solutions. Indeed, when Xn is chosen to be the n-sphere, (r, Xn ) space becomes flat. In
this case, H can be generalized to be any harmonic function on this flat space. Choosing,
furthermore, Ld and Mm to be flat gives the well known M2-, M5- and D3-brane solutions
which have m smeared transverse directions. Therefore, one can think of the new solutions
as the branes having curved world-volumes and smeared transverse directions, and nonspherical cross sections.
For m = 0, i.e., when Mm is empty, one obtains the solutions of [1113], which can
thus be viewed to be the members of a more general family found in this paper. For n = 1,
reparametrization fixing conditions should clearly be modified. A convinient choice is to
impose D = rC, which then gives the same solutions above with H = q log r + const.
The number of Killing spinors on Ld , Mm and Xn determines the number of unbroken
supersymmetries. Finally, it is also worth to mention that, field equations are still satisfied
even when Ld , Mm and Xn have no Killing spinors. As we will show in Section 5, this is
not a coincidence, and indeed there is another simple way of generating new solutions.
4. Generalized intersections
To generalize intersecting M2-, M5- and D3-brane solutions, we will make use of
Ricci flat Khler spaces. In this section, we will still use the definitions of Section 3, but
furthermore assume that Mm has a covariantly constant complex structure obeying
Ja b Jb c = a c ,
(47)
c Jab = 0.
(48)
This implies that Mm is a Ricci flat Khler space, m is an even integer and J is the
Khler two-form. Our basic strategy will still be the same; we will write an ansatz obeying
conditions (i) and (ii) of Section 2, and then work out the Killing spinor equations.
420
M2-brane intersections
We start with the following ansatz

2
2
+ D 2 dsX
,
ds 2 = A2 dt 2 + B 2 dr 2 + C 2 dsM
m
n
F (M J ) VX ,
(49)
(50)
where n + m = 9 with n = 1, 3, 5, 7, M is the Hodge dual on the manifold Mm and the

metric functions are assumed to depend only on r. For n = 7, the ansatz becomes a special
case of the generalized M2-brane ansatz of Section 3, and thus we will mainly consider
n = 1, 3, 5 cases. The 4-form field equations are identically satisfied since J is both closed
and co-closed on Mm . Furthermore, G0i = T0i = 0 in the basis E 0 = A dt, E r = B dr,
E a = Cea and E = De . Therefore, the conditions (i) and (ii) of Section 2 are satisfied.
To solve the Killing spinor equations, can consistently be chosen to obey (13), (14) and
t = 0,
0 = i,
Jab b = ia ,
(51)
which implies
(9 n)qe
(n 3)qe
(9 n)qe
C0
D0
1
A0
=
=
=
(52)
,
,
+ ,
7n
n
7n
n
7n
n
AB 3C D
CB
6C D
DB
6C D
D
where qe is the proportionality constant in (50). When n 6= 1, one can fix r-reparametrization invariance by imposing C m2 D n1 = r n1 + M and this gives the following solution

2
2
+ H (9n)/6 dr 2 + r 2 dsX
,
(53)
ds 2 = H (n9)/3 dt 2 + H (3n)/6 dsM
m
n
1
(54)
F0rab = H (n21)/12(r H )Jab ,
2
where M = 2qe /(n 1) and H = (1 + M/r n1 ). For n = 1, one should modify
reparametrization fixing condition. A convinient choice is to demand D = rC 4 , which then
gives the above solution with H = 2qe log r + const. As it will be made more clear when
we discuss the special cases, for n = 5, 3, 1 the solutions describe two, three and four M2brane intersections over a line, respectively. The structure of the background 4-form field
implies that membranes also wrap over the 2-cycle dual to Khler form of the space Mm .
M5-brane intersections
The discussion for intersecting M5-branes is not very different from M2-brane
intersections. Considering the fact that the 4-form field should give rise to magnetic type
of charges, we start with the following ansatz
2
2
+ D 2 dsX
,
ds 2 = A2 dsL2 d + B 2 dr 2 + C 2 dsM
m
2
F J VX ,
(55)
(56)
where m + d = 8 with m = 2, 4, 6 and the metric functions are assumed to depend only
on r. For m = 2, this reduces to a special case of the generalized M5-brane ansatz of
Section 3. The 4-form field equations are identically satisfied since J is both closed and
co-closed on Mm . On the other hand G0i = T0i = 0 in the basis E 0 = A dt, E r = B dr,
421
E a = Cea , E = De and therefore, the conditions (i) and (ii) of Section 2 are satisfied.
To solve the Killing spinor equations can consistently be chosen to obey (13)(15) and
r = 2i,
Jab b = ia ,
(57)
which implies
mqm
(n 6)qm
mqm
C0
D0
1
A0
=
=
=
(58)
,
,
+ ,
2
2
2
2
AB 6C D
CB
6C D
DB
3C 2 D 2 D
where qm is the proportionality constant in (50). To fix the r-reparametrization invariance
we impose C 2 D = r + M, where M = 2qm . Then, (58) can be solved to give

2
2
+ H m/3 dr 2 + r 2 dsX
,
(59)
ds 2 = H m/6 dsL2 8m + H (6m)/6 dsM
m
2
1
(60)
Fab = H (m+6)/6(r H )Jab ,
2
where H = (1 + M/r). As indicated above, for m = 2 the solution becomes one of the
M5-brane solution of Section 3. On the other hand, for m = 4 and m = 6, the solutions
describe two and three M5-branes intersecting over a three-brane and a string, respectively.
The structure of the background 4-form field implies that M5-branes also wrap over the
(m 2)-cycle dual to M J .
D3-brane intersections
Let us start by discussing possible ansatzs for the (anti)-self dual 5-form involving the
Khler two-form of Mm . The first obvious choice is to assume F J VX3 . Among the
possible cases, m = 2 corresponds to the smeared D3-brane of Section 3 and m = 6 is not
allowed since this does not leave any room for a time-like direction. One may also try to
write an ansatz involving M J such as F M J VXn . For m = 4, this becomes equal
to the choice F J VX3 and for m = 6, Xn space becomes one-dimensional. On the
other hand, m = 2 case turns out to be the same with the D3-brane ansatz of Section 3.
Summarizing, as a non-trivial ansatz representing intersecting D3-branes, one can write
2
2
+ D 2 dsX
,
ds 2 = A2 dsL2 2 + B 2 dr 2 + C 2 dsM
4
3
F (J VX ) (J VX ),
(61)
(62)
where the metric functions are assumed to depend only on r. The 5-form field equations
are identically satisfied and one can check that the background obeys condition (ii) of
the Section 2. To determine unknown functions, we solve the Killing spinor equations by
imposing (13)(15) and
= 2,
Jab b = ia ,
(63)
which implies
q
D0
q
C0
1
A0
=
=
0,
=
(64)
,
+ ,
AB 2C 2 D 3
CB
DB
2C 2 D 3 D
where q is the proportionality constant in (62). Reparametrization invariance can be fixed
by demanding C 2 D 2 = r 2 + M, where M = q/2. Then, (64) can be solved to obtain the
solution
422

2
2
ds 2 = H 1 dsL2 2 + dsM
+ H dr 2 + r 2 dsX
,
m
3
1
Fabr = H 3/2(r H )Jab ,
2
(65)
(66)
where H = (1 + M/r 2 ), which describes two D3-branes intersecting over a string and
wrapping over the cycle dual to J .
Interpretation and special cases
To be able to interpret these solutions properly, let us choose Xn to be the n-sphere.
In this case, one can introduce Cartesian coordinates to span (r, Xn ) space and H can
be generalized to be any harmonic function of these coordinates. When Ld and Mm are
also chosen to be flat, the solutions become the well known intersecting brane solutions in
which all harmonic functions are equal. 5
Comparing with this special case, it is easy to argue that Ld , Mm and Xn correspond
to common tangent, relative transverse and overall transverse directions. As mentioned
earlier, the branes also wrap over the cycles dual to the Khler two-form J or its Hodge
dual M J . For a given solution, this cycle can be written as a union of m/2 different
submanifolds. To see this we note that, in an orthonormal basis, J can be written as J =
e1 e2 + + em1 em . Therefore, the cycle dual to J or M J becomes the sum of
m/2 submanifolds dual to two-forms e1 e2 , . . . , em1 em or (m 2)-forms M (e1
e2 ), . . . , M (em1 em ), respectively. On the other hand, the Killing spinor projections
imply that there are m/2 intersecting branes each wrapping over one of these submanifolds.
We note that this interpretation is consistent with the special case where Mm is flat.
When Xn is different from the n-sphere, one obtains new solutions which have not been
encountered before. These solutions can be viewed to be the intersecting brane counterparts
of the brane solutions constructed in [1113], which have non-spherical transverse spaces.
In all cases the number of unbroken supersymmetries depend on the number of Killing
spinors on Ld , Mm and Xn . Like for the generalized brane solutions of the previous section,
one can check that the field equations are still satisfied, even when the spaces Ld , Mm and
Xn have no Killing spinors.
5. A general argument
In the last two sections, we have obtained supersymmetric, generalized brane solutions
by writing suitable ansatzs and working out Killing spinor equations. As pointed earlier,
the field equations are still satisfied, when the manifolds Ld , Mm and Xn have no Killing
spinors. This suggests the existence of a general rule which may hold at the level of field
equations.
5 In the next section, we will illustrate with an example how to generalize intersecting brane solutions when
harmonic functions are not equal.
423
Let us consider a brane solution which has a metric of the form,

2
2
+ dsN
,
ds 2 = A2 dsM
(67)
2 and ds 2 are line elements of m, n dimensional manifolds M and N ,

where dsM
N
respectively, and A is a function on N . We assume that the anti-symmetric tensor fields
of the solution are of the form F VM or F . . . , where VM is the volume form of
M and the dotted terms depend only on N . Furthermore, we consider the cases in which
scalar fields are independent of M and the anti-symmetric tensor field equations reduce to
dF = 0 and d F = 0. 6
Our claim is that, for such a brane solution, the field equations are still satisfied when M
e provided the Ricci tensors of both manifolds have
is replaced with a different manifold M,
e can be any manifold which is Ricci flat. Or, if
the same form. For instance, if M is flat, M
e
M is a sphere, M can be any Einstein manifold.
To prove this, let Aea and e be a basis for the tangent space, where ea and e are the
basis-one forms of M and N . We calculate the Ricci tensor of ds 2 as
1
A A b F b
b
a ,
R
(m
1)
a
(68)
(M)a
A
A2
A2
F
,
(69)
R = R(N) n
A
where the indices refer to the tangent space and R(M)ab , R(N) are the Ricci tensors of M
and N , respectively. The tensor quantities A and F are defined by the relations
Ra b =
dA = A e ,
(70)
dA = F e .
(71)
We note that the Ricci tensor of ds 2 depends only on the dimension m and Ricci tensor
R(M)ab of the manifold M.
e
Let us now analyze how the field equations may change, if one replaces M with M.
We first note that F is closed or co-closed irrespective of the choice of M, thus the form
equations are still satisfied. The possible terms in scalar and Einstein equations have at
most second order covariant derivatives of scalars. One can check that when M is replaced
e only a bc components of the spin connection may change. This ensures that the
with M,
second order covariant derivatives of scalars remain the same, since they are assumed to be
independent of M. Referring to the tangent space, it is easy to show that any tensor field
which is constructed from F , the terms containing up to second order covariant derivatives
of scalars and the metric have the same form. By (68) and (69), the last statement ensures
that both scalar and Einstein equations are still satisfied.
In some cases, M can be a Khler space and the anti-symmetric tensor fields may be of
e should also be
the form F J , where J is the Khler two-form. In such cases, M
e One can
chosen to be Khler and J should be replaced with the Khler two-form of M.
easily argue that the field equations are not affected from these replacements.
6 It seems one can relax the last assumption, but here, for simplicity, we do not consider more general cases.
424
Let us now present some examples which enables one to recover some solutions obtained
in the literature and in the previous sections. Consider, for instance, the well known
smeared M2-brane solution,
ds 2 = H 2/3 dt 2 + dx12 + dx22

2
2
2
+ dym+1
+ + dy10
,
+ H 1/3 dy12 + + dym
1
F = H 7/6 ( H ) ,
2
(72)
(73)
where H is harmonic on (ym+1 , . . . , y10 ). Referring the above discussion, it is easy to

see that the world-volume directions (t, x1, x2 ) and the smeared transverse directions
(y1 , . . . , ym ) can be replaced with more general Ricci flat manifolds. In this way, one can
obtain

2
2
2
+ dym+1
+ + dy10
,
ds 2 = H 2/3 dsL2 3 + H 1/3 dsM
m
1
F = H 7/6 ( H ) ,
2
(74)
(75)
which becomes one of the solutions constructed in Section 3.

In IIB theory, one can start from the single centered D3-brane solution

ds 2 = H 1/2 dt 2 + dx12 + + dx32 + H 1/2 dr 2 + r 2 d52 ,
1
F r = H 5/4 (r H ) ,
2
(76)
(77)
where d52 is the line element on the 5-sphere and H = (1 + M/r 4 ). The metric and the
5-form field are of the form discussed above with the flat world-volume (t, x1 , . . . , x3 ) and
the transverse 5-sphere play the role of manifold M in (67). Therefore, one can generalize
the well known solution as

2
,
ds 2 = H 1/2 dsL2 4 + H 1/2 dr 2 + r 2 dsX
5
1
F r = H 5/4 (r H ) ,
2
(78)
(79)
which corresponds to the solution obtained in [11] when L4 is flat.

Let us finally consider an example to illustrate how the intersecting solutions can be
generalized when the harmonic functions are not chosen to be equal. The background
representing two M5-branes intersecting over a three-brane is given by [17]

1/3 2/3
ds 2 = (H1 H2 )1/2 dt 2 + dx12 + dx22 + dx32 + H1 H2 dx42 + dx52

2/3 1/3
2
dx62 + dx72 + (H1 H2 )2/3 dx82 + + dx10
,
+ H1 H2
F dx4 dx5 dH2 + dx6 dx7 dH1 ,
(80)
(81)
where, H1 and H2 are harmonic and is the Hodge dual on (x8 , . . . , x10 ) space. Here, it is
easy to see that, (t, x1 , x2 , x3 ), (x4 , x5 ) and (x6 , x7 ) spaces play the role of manifold M in
425
(67). Since all these spaces are flat, one can replace them with arbitrary Ricci flat manifolds
L4 , M1 and M2 7 to obtain
1/3
ds 2 = (H1 H2 )1/3 dsL2 4 + H1

+ (H1H2 )2/3
2/3
2/3
1/3
2
dsM
+ H1 H2
1

2
2
dx5 + + dx10 ,
H2
F VM1 dH2 + VM2 dH1 .
2
dsM
2
(82)
(83)
When H1 = H2 = H , the 4-form field (81) can be written as

F J dH,
(84)
where J is the complex structure of the flat space (x4 , x5 , x6 , x7 ). In this case, these four
flat directions can be replaced with any 4-dimensional Ricci flat Khler space, which
corresponds to one class of intersections found in Section 4.
As can be inferred from these examples, in a brane solution the flat world-volume and
smeared transverse directions, and the transverse sphere at a fixed radial distance can
be replaced with more general Ricci flat and Einstein manifolds. In intersecting brane
solutions the common tangent and relative transverse directions have this property. We note
that, these replacements can be done at the level of field equations and supersymmetry of
the new solution obtained in this way is not manifest.
6. Solutions from U(1) bundles over Ricci flat Khler spaces

The brane solutions obtained so far have a number of common properties; for instance,
the warping factors are the same for the transverse and world-volume directions. In this
section, by using the theorem of Section 2, we will take a step in obtaining solutions which
fail to have this property, at least along the transverse directions.
We first construct singular, Ricci flat manifolds having a cone-like structure over U(1)
bundles over Ricci flat Khler spaces, which give rise to new supersymmetric vacua for
non-gauged supergravities. Since the conformal factors multiplying U(1) fibers and the
base spaces turn out to be neither equal to each other nor equal to the square of the
coordinate parametrizing the cone, we name these spaces as (generalized) cones.
As we will see in a moment, there are two singularities associated with the cone, since at
the origin the base space and at infinity the U(1) fibers shrink to zero size. Also, in viewing
the solutions as Ricci flat compactifications, unlike the conventional KaluzaKlein picture,
there is no natural way of assuming the internal spaces to be small compared to the spacetime.
At the end of this section, we will construct brane solutions which asymptotically approach these singular compactifications and preserve half of the available supersymmtries.
This is in fact not surprising, if one assumes stability of the vacua, since the fundamental
branes underlying the supergravity theories reveals themselves as supersymmetric soliton
solutions.
7 In two-dimensions, Ricci flatness implies flatness. Here, we would like to emphasize that for field equations
Ricci flatness of these manifolds is the key condition.
426
A singular Ricci flat space

It is well known that the cone over a manifold X
2
ds 2 = dr 2 + r 2 dsX
(85)
is Ricci flat if and only if X is Einstein. In this section, we would like to construct Ricci
flat manifolds having a cone-like structure over Ricci flat Khler spaces. It is clear that one
should modify (85) in a non-trivial way. Let us consider a metric of the form
2
+ D 2 (d A)2 ,
ds 2 = B 2 dr 2 + C 2 dsM
m
(86)
where Mm is an m-dimensional Ricci flat Khler space, A is the one-form potential for
the Khler two-form so that dA = J , is a periodic coordinate and the metric functions
are assumed to depend only on r. It is clear that we have a fiber bundle structure and is
the coordinate on U(1) fibers. We would like to determine B, C and D which give rise to
a Ricci flat space. Instead of calculating Ricci tensor and solving second order, coupled
differential equations we demand existence of a covariantly constant spinor which, by
considerations of Section 2, will imply Ricci flatness. In the tangent space basis E r = B dr,
E a = Cea and E D = D(d A), a covariantly constant spinor obeys

D
1 D
1
D
Jab E ab
Jab E b Da
ab +
D = d +
4
2C 2
4 C2
1 D0
1 C0
Ea ar +
ED Dr = 0,
(87)
2 CB
2 DB
where ab is the spin connection on Mm , and we have used the differential form notation.
Conistently imposing
+
= 0,
1
d + ab ab = 0,
4
Jab b = ia ,
Dr = i,
(88)
(89)
one obtains
D
mD
D0
C0
=
= 2.
,
(90)
CB
2C 2
DB
4C
Therefore, when the metric functions B, C and D obey (90), the Ricci tensor of (86)
vanishes. Fixing r-reparametrization invariance by B = 4/(m + 4), (90) can be solved to
give the following Ricci flat metric
ds 2 =
16
2
dr 2 + r 4/(m+4) dsM
+ r 2m/(4+m) (d A)2 ,
m
(m + 4)2
(91)
which we name as the space MC . Topologically, MC is a product of a line parametrized

by r and a U(1) bundle over a Ricci flat Khler space. The metric is singular, since
at the origin, when r = 0, the base space and at infinity, as r , the U(1) fibers
shrink to zero size. On the other hand, MC have covariantly constant spinors, and
thus gives rise to supersymmetric compactifications of non-gauged supergravities. In
427
these compactifications, one can view the coordinate r as a radial coordinate in the
uncompactified space-time. Therefore, the U(1) bundle over Mm plays the role of internal
space. As can be inferred from the structure of the metric, there is no natural way of
assuming the internal directions to be small. Thus these compactifications are very
different from the conventional KaluzaKlein ones. Although the metric is singular, the
existence of unbroken supersymmetries is a sign for stability. We now construct M2- and
D3-brane solutions which belong to these singular vacua.
M2- and D3-branes on MC compactifications
In searching brane solutions which asymptotically belong to above Ricci flat compactifications, we consider the cases in which MC plays the role of the total transverse space.
Since MC is an even dimensional manifold, this leaves the possibility of an even dimensional brane in D = 11, and an odd dimensional brane in D = 10. We think of the coordinate r as the radial coordinate in the transverse space.
In D = 11, the only even dimensional brane is the M2-brane and this suggests the
following ansatz
2
+ D 2 (d A)2 ,
ds 2 = A2 dsL2 3 + B 2 dr 2 + C 2 dsM
6
(92)
F VM (d A),
(93)
where all metric functions are assumed to depend on r. We remind the reader that we are
still using the definitions of Section 3. One can check that, the 4-form field equations are
identically satisfied, and G0i = T0i = 0 in the basis E = Ae , E r = B dr, E a = Cea and
E D = D(d A). Therefore, the conditions (i) and (ii) of Section 2 are satisfied. One can
check that, imposing
= 0,
= 0,
a = 0,
(94)
and
= 6,
Jab b = ia ,
Dr = i,
(95)

2qe
D
3D
C0
D0
qe
qe
A0
=
=
= 2
,
,
,
(96)
AB 3C 6 D
CB
DB
2C 2 3C 6 D
2C
3C 6 D
where qe is the proportionality constant in (93). Since the conditions imposed on the
Killing spinor are consistent with each other, when (96) is satisfied, the background should
obey the Einstein equations by the theorem of Section 2. We fix the r-reparametrization
invariance by imposing C 4 D 2 = r. Using this condition, (96) can be solved to give the
following solution

M 2/3 2
1
M 4/3 2
M 22/3 2 1
dsL3 + 7 1
dr +
dsM6
1
ds 2 = 1
r
r
r
r
r

M 8/3
3
+r 1
(d A)2 ,
(97)
r
428
Fr

M 3/2
M 8/3
= r
,
1
2
r
(98)
where M = 2qe .
It is clear that, in the solution, L3 represents the world-volume of the M2-brane and r is
a radial coordinate along transverse directions. Unlike all solutions obtained before, there
are three different warping factors multiplying the transverse directions, which have also
a non-trivial topological structure due to the presence of U(1) bundle. There is a horizon
located at r = M and asymptotically, as r , the solution becomes L3 MC , which
can be seen by a coordinate change dr r 7/2 dr, near infinity. The solution is a half
supersymmetry preserving state of the vacuum L3 MC since the presence of the M2brane brakes only half of the available supersymmetries.
As an example in D = 10, we consider the D3-brane of IIB theory and start with the
following ansatz
2
+ D 2 (d A)2 ,
ds 2 = A2 dsL2 4 + B 2 dr 2 + C 2 dsM
4
F VM (d A) VM (d A),
(99)
(100)
where, as usual, we assume that A, B, C and D depend only on r. One can check
that the ansatz obeys the condition (i0 ) and (ii) of Section 2, and therefore to obtain a
supersymmetric solution one needs to work out Killing spinor equations. Imposing
= 0,
= 0,
a = 0,
(101)
and
= 24i,
Jab b = ia ,
Dr = i,
(102)

q
C0
D
D0
D
q
q
A0
=
,
=
,
= 2
,
(103)
4
2
4
AB 4C D
CB
2C
4C D
DB
C
4C 4 D
where the dyonic charge q is defined to be the proportionality constant in (100). Fixing
r-reparametrization invariance by C 2 D 2 = r, one can solve the differential equations to
obtain the following D3-brane solution

q 3/2 2
q 1/2 2
1
q 13/2 2 1
2
1
dsL4 + 6 1
dr +
dsM4
ds = 1
r
r
r
r
r

q 3/2
+r 1
(d A)2 ,
r
2
F r

q 9/4
= qr 1
.
r
(104)
(105)
It is clear that, L4 corresponds to the world-volume of the D3-brane. Asymptotically, as

r , the solution becomes L4 MC which can be seen by a coordinate change dr
r 3 dr, near infinity. The solution is half supersymmetry preserving state of the vacuum
429
L4 MC , since the presence of D3-brane brakes half of the available supersymmetries.

There is also a horizon located at r = q. We finally note the non-orthogonal decomposition
of the transverse directions due to the U(1) bundle structure.
In this paper we do not attempt to determine singularity structures of these solutions.
The presence of unbroken supersymmetries make us to believe that the solutions are stable.
Another important open problem is to find realizations of unbroken supersymmetries on
the solutions, which enable one to determine how Bogomolny bounds are saturated.
7. Conclusions
The brane solutions of supergravity theories have played a crucial role in recent developments in the non-perturbative string/M theories. With this experience, it is reasonable
to claim that finding new solutions will also be important for future developments. In this
paper, we have constructed new classes of solutions in D = 11 and D = 10 dimensions.
The solutions obtained in Sections 3 and 4 have a very similar structure with the well
known solutions and can be viewed as the generalizations of them. Solutions obtained in
Section 6 belong to a different class in which the transverse directions do not have a single
warping factor. In Section 5, we have presented a general argument which allows one to
construct new solutions from the old ones by replacing certain directions with more general
manifolds.
In constructing new solutions, we have mainly used the theorem proved in [5]. The
results of the present paper show that the theorem reviewed in Section 2 is an important
way of obtaining new supersymmetric solutions. Here, we would like to mention two open
problems which can possible be attacked by using the theorem; the first one is to find nonstatic cases which may lead to time dependent or stationary supersymmetric solutions, and
the second one is to construct explicit examples of non-trivially embedded brane solutions,
as formally discussed in [10].
By the well known solution generating techniques, like applying S- or T-dualities, or
by dimensional reduction, one can obtain more new solutions in D = 10 or in lower
dimensions. As an interesting application of this, we note that it is possible to untwist a
U(1) bundle by a T-duality transformation along the coordinate parametrizing U(1) fiber.
As for the usual brane solutions, one may try to identify low energy theories defined on
the world-volumes. The number and the supermultiplet structure of collective coordinates
depend on the realizations of unbroken supersymmetries, the singularities of the solutions
and the isometries of the internal manifolds. A classification of spaces having Killing
spinors in diverse dimensions is the key ingredient for such an analysis. After identifying
the low energy field theories, the next important question is to learn how to take decoupling
or near horizon limits. We note that, brane solutions having transverse Einstein spaces
give rise to generalizations of the original AdS/CFT dualities which correspond to
compactifications on arbitrary Einstein manifolds. By defining sensible decoupling limits
for other type of solutions found in this paper, one may discover interesting realizations of
holographic principle.
430
References
[1] M.J. Duff, R.R. Khuri, J.X. Lu, String solitons, Phys. Rep. 259 (1995) 213, hep-th 9412184.
[2] J.P. Gauntlett, Intersecting branes, hep-th 9705011.
[3] M.J. Duff, H. Lu, C.N. Pope, E. Sezgin, Supermembranes with fewer supersymmetries, Phys.
Lett. B 371 (1996) 206, hep-th 9511162.
[4] J.P. Gauntlett, G.W. Gibbons, G. Papadopoulos, P.K. Townsend, Hyper-Khler manifolds and
multiply intersecting branes, Nucl. Phys. B 500 (1997) 133, hep-th 9702202.
[5] A. Kaya, A note on a relation between Killing spinor and Einstein equations, Phys. Lett. B 458
(1999) 267, hep-th 9902010.
[6] D. Brecher, M.J. Perry, Ricci flat branes, Nucl. Phys. B 566 (2000) 151, hep-th 9908018.
[7] J.M. Figueroa-OFarrill, More Ricci flat branes, Phys. Lett. B 471 (2000) 128, hep-th 9910086.
[8] B. Janssen, Curved branes and cosmological (a,b)-models, Jhep 0001 (2000) 044, hep-th
9910077.
[9] G. Papadopoulos, J. Russo, A.A. Tseytlin, Curved branes from string dualities, hep-th 9911253.
[10] R. Minasian, D. Tsimpis, On the geometry of non-trivially embedded branes, hep-th 9911042.
[11] A. Kehagias, New type IIB vacua and their F-theory interpretation, Phys. Lett. B 435 (1988)
337, hep-th 9805131.
[12] B.S. Acharya, J.M. Figueroa-OFarrill, C.M. Hull, B. Spence, Branes at conical singularities
and holography, Adv. Theor. Math. Phys. 2 (1999) 1249, hep-th 9808014.
[13] D.R. Morrison, M.R. Plesser, Non-sphereical horizons, I, Adv. Theor. Math. Phys. 3 (1999) 1,
hep-th 9810201.
[14] G. Horowitz, A. Strominger, Black strings and p-branes, Nucl. Phys. B 360 (1991) 197.
[15] R. Gven, Black p-brane solutions of D = 11 supergravity theory, Phys. Lett. B 276 (1992) 49.
[16] G. Papadopoulos, P.K. Townsend, Intersecting M-branes, Phys. Lett. 380 (1996) 273, hep-th
9603087.
[17] J.P. Gauntlett, D.A. Kastor, J. Traschen, Overlapping branes in M theory, Nucl. Phys. B 478
(1996) 544, hep-th 9604179.
[18] A. Tseytlin, Composite BPS configurations of p-branes in 10 and 11 dimensions, Class.
Quantum Gravity 14 (1997) 2085, hep-th 9702163.
[19] E. Cremmer, B. Julia, J. Sherk, Supergravity theory in 11-dimensions, Phys. Lett. B 76 (1978)
409.
[20] J.H. Schwarz, Covariant field equations of chiral N = 2, D = 10 supergravity, Nucl. Phys.
B 226 (1983) 269.

SO(d, d) transformations of RamondRamond

fields and spacetime spinors
S.F. Hassan 1
Centre de Physique Thorique, cole Polytechnique, 91128 Palaiseau, France
Abstract
We explicitly construct the SO(d, d) transformations of RamondRamond field strengths and
potentials, along with those of the spacetime supersymmetry parameters, the gravitinos and the
dilatinos in type-II theories. The results include the case when the SO(d, d) transformation involves
the time direction. The derivation is based on the compatibility of SO(d, d) transformations with
spacetime supersymmetry, which automatically guarantees compatibility with the equations of
motion. It involves constructing the spinor representation of a twist that an SO(d, d) action induces
between the local Lorentz frames associated with the left- and right-moving sectors of the worldsheet
theory. The relation to the transformation of RR potentials as SO(d, d) spinors is also clarified.
2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 04.65.+e
Keywords: T-duality; O(d, d) transformations; String theory; Supergravity
1. Introduction
The O(d, d) transformations in string theory, also referred to as the Narain group or
the generalized T-duality group, have played an important role in understanding toroidal
compactifications and dualities, as well as in constructing classical solutions to the lowenergy equations of motion [17]. The action of the O(d, d) group on the NSNS sector
fields (the graviton, the antisymmetric tensor field and the dilaton), which are all assumed
to be independent of d coordinates, has been known for a long time. In the worldsheet
formalism, the O(d, d) action on these fields can be obtained either by a canonical
transformation [3,4], or by gauging appropriate isometries in a non-linear -model [8].
In the framework of the low-energy effective theory, one can derive the transformation
of the fields by writing the effective action for the NSNS fields in a manifestly O(d, d)
invariant form [47].
1 fawad@cpht.polytechnique.fr
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 3 7 - 0
432
S.F. Hassan / Nuclear Physics B 583 (2000) 431453
Deriving the O(d, d) transformation of the RamondRamond fields has turned out
to be more complicated. Since the fundamental string does not carry RamondRamond
charges, the usual NSR formalism worldsheet methods cannot be used. In [9] the equations
of motion in IIA and IIB supergravities were used to derive the transformation of
the RamondRamond potentials under the single T-duality subgroup of O(d, d) which
interchanges the two theories. However, the generalization of this approach to the full
O(d, d) group (or to its SO(d, d) part, if we do not want to interchange IIA and
IIB theories) has not been straightforward. Before describing our derivation of the
transformation, which is based on supersymmetry, let us briefly review an interesting,
though as yet inconclusive, alternative approach that has so far been followed.
In [10] it was observed that when type-II theories are compactified on T 6 , the resulting
RR scalars fill up a MajoranaWeyl spinor representation of the associated T-duality
group SO(6, 6). This observation was further generalized in [11] where it was pointed
out that the components of RR potentials in type-II theories compactified on T d , when
arranged in representations of the Lorentz group SO(9 d, 1) of the uncompactified space,
fill up MajoranaWeyl spinor representations of the SO(d, d) group. This observation
is mainly based on the decomposition of representations of the U-duality group of the
compactified theory in terms of the representations of its S-duality and T-duality subgroups.
The U-duality group contains NSNS and RR charges in the same multiplet and, after
this decomposition, the RR charges are found to fall in the spinor representations of the
T-duality group SO(d, d). In [12,13], this problem was studied in more detail from the
group theory perspective. The explicit construction of this spinor representation in terms
of the supergravity fields was undertaken in [14] for a specific case of T 3 compactification
of type-IIA theory, and was generalized to T d compactifications in [15]. In this approach,
one constructs SO(d, d) MajoranaWeyl spinors out of combinations of RR potentials
and the NSNS 2-form field. The guiding principle in this construction is the requirement
that, in terms of these spinors, the kinetic energy terms for the RR fields in the lowenergy effective action should have a manifestly SO(d, d) invariant from. While, as
such, the construction ignores the electricmagnetic duality constraints between RR field
strengths (including the self-duality of the type-IIB 5-form), nevertheless, one expects it
to be consistent with these constraints and to be extendible to include the RR dependent
ChernSimons terms in the low-energy effective action. Although this formally proves the
SO(d, d) invariance of the RR kinetic energy terms in the low-energy effective action,
so far it has not been possible to extract from this the transformation of RR potentials
and field strengths under generic nontrivial SO(d, d) transformations (the reason for this
difficulty will be clarified in Section 6). It should be mentioned that in the case of IIA
on T 3 , the transformation of the RR scalars (in 7 dimensions) as an SO(3, 3) spinor has
also been studied in the framework noncommutative super YangMills compactifications
of matrix theory [14,16]. However, these results are not valid in string theory except in the
matrix theory limit of Gij 0, where many terms in the transformation drop out.
In this paper, we determine the SO(d, d) transformations of the RamondRamond field
strengths and potentials using a more straightforward approach. In the process we also
determine the SO(d, d) transformations of the spacetime supersymmetry parameters and
433
RNS fields (the gravitinos and the dilatinos). Our approach is based on the compatibility
of SO(d, d) transformations with spacetime supersymmetry. As a result, our derivation
automatically guarantees the SO(d, d) covariance of the equations of motion (and hence,
the invariance of the associated low-energy effective actions including the ChernSimons
terms), and is consistent with the electricmagnetic duality constraints on RR field
strengths, since in type-II theories all these are determined by supersymmetry. The method
we use is a generalization of the one used in [17] for a single T-duality transformation
and can be summarized as follows. In type-II supergravity theories, every spinor index
originates either in the left-moving or in the right-moving worldsheet sector of the
underlying string theory. On the other hand, it has been known that in flat-space nontrivial
O(d, d) transformations correspond to transforming the left-moving and right-moving
parts of the spacetime coordinates by independent Lorentz rotations: + X R + X,
X S X [5,6]. This feature also survives in curved backgrounds where, ignoring a
contribution from the worldsheet fermions, one gets X Q X, though now Q
are spacetime dependent through their dependences on the background fields [18,19].
Thus, one can regard a nontrivial O(d, d) transformation as twisting the left- and the rightmoving sectors of the worldsheet theory with respect to each other. By supersymmetry,
this translates to transforming the spinor indices originating in the two worldsheet sectors
by two different O(d, d) induced local Lorentz transformations. In particular, we may
choose a convention in which the spinor index associated with, say, the right-moving sector
remains unchanged, and only the one associated with the left-moving sector transforms.
This determines the basic O(d, d) action on the spacetime spinors as well as on the
RamondRamond fields which can be combined into bispinors. The exact form of the
transformation is fixed by the O(d, d) covariance of the supersymmetry transformations.
To make this structure behind the transformation of the RR fields manifest, it is important
that the supersymmetry transformations are written in terms of variables that appear more
natural in string theory. Note that in this approach we only have to construct the spinor
representation of the local Lorentz twist induced by O(d, d) which is an element of a
subgroup of the Lorentz group O(9, 1), and not that of the full O(d, d) group. The explicit
construction we give here is for the SO(d, d) part of the group. The single T-duality case
has been discusses in detail in [17].
This paper is organized as follows: in Section 2 we describe our notation and briefly
review some aspects of the O(d, d) transformations that will be needed in the rest of
the paper. In particular, we emphasize that after a nontrivial O(d, d) transformation, the
transformed theory contains two different vielbeins that are related by a local Lorentz
transformation. This is interpreted as a twist between the local Lorentz frames associated
with the left- and right-moving sectors of the worldsheet theory. In Section 3 we argue
that this local Lorentz twist should be absorbed by the spinors and hence construct its
spinor representation. We discuss the flat-space T-duality limit in some detail to highlight
some important features of the spinor representation. In Section 4, we obtain the O(d, d)
action on the supersymmetry variation parameters, the gravitinos and the dilatinos. In
Section 5, we use the transformation of the spacetime spinors along with the explicit
construction of the spinor representation of the local Lorentz twist to obtain the SO(d, d)
434
action on the RamondRamond field strengths. Then by a simple argument we show that
the transformations of the RR potentials have exactly the same form as those of the
field strengths. After restricting to type-IIA on T 3 , and taking the Gij 0 limit, we
reproduce the results obtained in the NCSYM formalism. We also write a simple SO(d, d)
covariant form for the RR kinetic energy terms in the low-energy effective action. In
Section 6, we discuss the relation between our approach and the alternative approach in
which components of RR potentials combine into SO(d, d) spinors. Our conclusions are
summarized in Section 7. Appendix A outlines the steps for the derivation of two sets of
equations needed in Section 4, and Appendix B contains some -matrix results.
2. Preliminaries
In this section, we will describe our notation and conventions and then briefly recall
the action of the SO(d, d) group on the massless NSNS sector fields for later reference.
Everything in this section applies equally well to the O(d, d) group but we restrict
ourselves to elements with positive determinant which do not interchange type-IIA and
type-IIB theories.
2.1. Notation
We denote the 10-dimensional spacetime coordinates by XM (M = 0, . . . , 9), and
assume that all fields are independent of the d coordinates Xi , for i = 0, . . . , d 1. For
the sake of generality, we have assumed that these coordinates also include time. The
remaining 10 d coordinates are denoted by X on which the fields may depend. a, b =
0, . . . , 9 are local Lorentz frame indices. We use the symbols G, B, Q, S, R to denote
10 10 matrices and G, B, Q, S, R to denote their d d blocks labeled by i, j .
2.2. Parameterization of SO(d, d)
It is well known that the low-energy effective action for the massless NSNS fields,
which are restricted to be independent of Xi , is invariant under an SO(d, d) group of
transformations [47]. Not all elements of this group have a nontrivial action on the
background fields [5]: in fact, out of its d(2d 1) elements, a d 2 -dimensional subgroup
corresponds to general linear coordinate transformations GL(d) and another d(d 1)/2dimensional subgroup corresponds to constant shifts in Bij , both of which are manifest
symmetries of the low-energy effective action even without restricting the fields to be
independent of Xi . To describe the action of these elements on the fields, we do not really
need the O(d, d) formalism. Only a d(d 1)/2-dimensional subgroup SO(d 1, 1)
SO(d 1, 1)/SO(d 1, 1) acts nontrivially on the background fields and therefore, in this
paper, we focus on the transformations generated by this subgroup. Let S, R SO(d
1, 1). We choose a basis in which elements of SO(d 1, 1) SO(d 1, 1) SO(d, d)
take the form
S
0
0
0 110d 0
S 0
O=
0
0 R
0
R
0
0
0

0
O=
.
OT
0
0

0
0
0
435
110d
(1)
Note that the transformations are embedded in 10 10 matrices and we use a hat to
distinguish between the SO(d 1, 1) invariant metric MN appearing above, and the
Minkowski metric ab of the local Lorentz frame. The case S = R corresponds to a
coordinate rotation which is already included in GL(d). The nontrivial transformations
are therefore parameterized by SO(d 1, 1) SO(d 1, 1)/SO(d 1, 1) which is
obtained by restricting O such that its only independent parameters are contained in
S 1 R (for example, by setting S = 1d ). In the following we may simply write SO(d, d),
understanding that we are only interested in its nontrivial elements.
2.3. Action on NSNS backgrounds
The action of the SO(d, d) group on the metric GMN and the antisymmetric tensor field
BMN is most conveniently written in terms of a matrix M which, in our conventions for
the parameterization of the group elements, is given by [47]:

1
( K + 110 ) G1 ( 1 K T + 110 ) ( 1 K + 110 ) G1 ( 1 K T 110)
.
M=
( 1 K 110 ) G1 ( 1 K T + 110 ) ( 1 K 110 ) G1 ( 1 K T 110 )
(2)
Here, K = G + B and the sign of BMN is chosen such that the worldsheet action has the
R
standard form d2 KMN + XM XN . The NSNS fields G, B and then transform
according to
q
T
2e
2
e det G.
e
e =e
det G/
(3)
M = OMO ,
Using (2) in (3), one can check that the transformation of the metric G can be written in
two equivalent forms [20]:
e1 = Q G1 QT = Q+ G1 QT+ ,
G
M
where the matrices QM
N and Q+N are given by

Q = 12 (S + R) + (S R) 1 (G + B) ,

Q+ = 12 (S + R) (S R) 1 (G B) .
(4)
(5)
These matrices will play an important role in our discussions later. Their relevance can be
understood better by noting that the canonical transformation that implements the SO(d, d)
eM = QM XN , where we have
transformation at the worldsheet level has the form X
N
ignored the worldsheet fermion contributions. On the worldsheet fermions, the action takes
436
M = QM N [19]. It is also useful to write the inverses of Q in terms of the

e
the form
N
variables of the transformed theory as,
1

1
e+B
e ,
+ R 1 + S 1 R 1 1 G
Q1
= 2 S
(6)
1

1
eB
e .
S + R 1 S 1 R 1 1 G
Q1
+ =
2
Then, from the form of S and R given in (1) it is clear that

(Q ) = Q1
(Q ) j = Q1
j = 0,
= .
(7)
In terms of the matrix Q the dilaton transformation takes the from

p
e
e = det Q .
(8)
2.4. Induced local Lorentz twist

In order to obtain the transformations of spinors and RR fields, we also need the
transformation of the vielbein eaM . From now on, for convenience, we will use the symbol
1 as the vielbein. From
e for the inverse vielbein eM
a and will refer to both e and e
equation (4) it is evident that a priori one can transform the vielbein in two different
ways [19],
M
N
= QM
e()a
N e a ,
M
N
e(+)a
= QM
+N e a ,
(9)
e1 = e1 e T . These two vielbeins are

both leading to the same transformed metric G
related by a local Lorentz transformation :
M
M
= e()a
ab ,
e(+)b
= e1 Q1
Q+ e.
(10)
The appearance of the two vielbeins e() can be easily understood in terms of the
worldsheet theory [17]: one may regard the vielbein eaM as originating in either the leftmoving or the right-moving sector of the worldsheet theory. It then transforms to either e(+)
or e() , depending on its worldsheet origin. This is consistent with the action of SO(d, d)
on the worldsheet variables as described above and also with the fact that the worldsheet
parity , which interchanges the two worldsheet sectors, also interchanges the
variables (S, R, B) and (R, S, B), and hence e(+) and e() .
It is then clear that an SO(d 1, 1) SO(d 1, 1)/SO(d 1, 1) transformation twists
the local Lorentz frame originating in the left-moving worldsheet sector with respect to
the one originating in the right-moving sector by an amount ab . Although all NSNS
sector fields and their supersymmetry variations are insensitive to , in general this
twist can be absorbed by the Ramond sector fields, leaving the two local Lorentz frames
untwisted. In fact, as we will see in Section 4, this is dictated by the requirement of
consistency with supersymmetry which determines the way the spinor representation of
the induced twist ab enters in the transformations of spacetime spinors and Ramond
Ramond fields under nontrivial SO(d, d) action. However, before that, we first construct
the spinor representation of the induced twist in the next section.
437
3. Spinor representation of the induced twist

Let us consider the -matrices M = eMa a that are associated with the local Lorentz
frames that originate in both the right- and the left-moving sectors of the worldsheet theory.
M = e M a or
After a nontrivial SO(d, d) transformation, these transform to either e()
()a
M
M
a
e(+) = e(+)a , depending on the worldsheet sector in which the Dirac algebra originates.
M
as the vielbein in terms of which the transformed theory is to be finally
Let us choose e()a
2
M
in terms of it, using (10). Then we get
written and express e(+)a
M
M
= 1 e()
,
e(+)
(11)
where is the spinor representation of the SO(d, d) induced twist defined by

1 a = ab b .
(12)
The factor in (11) can now be absorbed in the transformation of spacetime spinors.
Note that this only affects spinor indices that originate in the left-moving Ramond sector
of the worldsheet theory, leaving the ones associated with the right-moving Ramond sector
unchanged. We will see in the next section that this is in fact dictated by supersymmetry.
This situation is quite reminiscent of the construction of Ramond sector boundary states in
the presence of a background worldvolume gauge field studied in [21]. Below, we explicitly
construct the spinor representation associated with SO(d, d). For the simpler case of a
single T-duality, for which det O = 1, see [17].
In general, ab is an element of the local Lorentz group SO(9, 1) associated with the
left-moving sector of the worldsheet theory. Therefore, it can be parameterized by an
antisymmetric matrix Aab as
= ( + A)1 ( A),
A = (1 )(1 + )1 .
(13)
T =
. For the time being we assume that

Then, the antisymmetry of A implies that
1 + is nonsingular so that the above parameterization is well defined. We will later come
back to the singular case. The spinor representation can now be constructed in terms of
A as

1/2
12 Aab ab ,
= det( + A)
where the symbol stands for an exponential-like expansion with the products of matrices fully antisymmetrized:
12 Aab ab
=1+
n=5
X
(1)n
n=1
n!2n
Aa1 b1 . . . Aan bn a1 b1 ...an bn .
A priori, contains products of up to 10 -matrices. However, note that equaM N

tion (12) implies 1 M = (Q1
Q+ ) N . Furthermore, from (7) one can see that
2 Of course, this is a matter of choice and we could as well choose eM
(+)a or any other Lorentz equivalent
M
vielbein. However, with e()a
as the vielbein, it is natural to set S = 1 so that in the flat-space limit, e() e
while e(+) Re.
438
invariant and therefore should be con(Q1

Q+ ) = . This shows that keeps
structed in terms of the d matrices i alone, which anticommute with . To make this
feature explicit, let us define AMN = eMa eNb Aab . Then, using as given by (10) in the
expression for A in (13), we find that the only nonzero components of AMN are given by

1

1
(14)
1d + S 1 R + B ij .
Aij = d 1d S 1 R
Here A, S, R and d denote the d d blocks of A, S, R and labeled by i, j =

0, . . . , d 1, and the (ij ) indices on the right hand side are raised by the matrix inversion.
Note that the first term in the square bracket is the inverse of the antisymmetric matrix
that parameterizes the orthogonal transformation S 1 R through equations similar to (13).
Substituting for the determinant prefactor as well, the spinor representation takes the form
s

det(Q + Q+ )
1 ij
d2
A ij ,
(15)
=2
det Q
2
where now,
[d/2]
X (1)p

Ai1 i2 . . . Ai2p1 i2p i1 i2 ...i2p1 i2p .
12 Aij ij = 1 +
p!2p
(16)
p=1
Here, [d/2] stands for the integer part of d/2 so that, for example, the summation contains
only one term for d = 2, 3 (involving i1 i2 ) and two terms for d = 4, 5 (involving i1 i2
and i1 i2 i3 i4 ). In particular, it does not contain as should be the case.
The factor det Q
is the same quantity that appears in the transformation of e2 and det G. This will be
important in the derivation of the transformation rules of the RamondRamond fields as
well as in showing the invariance of the low-energy effective action. The remaining factor
has the explicit form

(17)
det(Q + Q+ ) = det 1d + S 1 R + 1d S 1 R d B ,
where we have used the fact that det S = 1. As will be shown below, this factor is essential
for getting the correct T-duality limit in cases with Bij = 0, which serves as an easy
check of the results. Note that both this factor and Aij depend only on B and have no
dependence on G and other components of B. Furthermore, note that, as expected,
depends on the combination S 1 R alone and equals identity for S = R, corresponding to
an SO(d) GL(d) transformation on the bosonic backgrounds which does not affect the
spinor index. One can easily restrict these formulae to the case when Xi do not contain the
time coordinate by replacing d by 1d .
3.1. Singular limits and T-duality with Bij = 0
When det(1 + ) = 0, the parameterization in (13) becomes singular, though in the
spinor representation , the singular terms cancel out and one is left with a well defined
expression. This case is important as it includes T-duality transformations in flat-space (as
well as in curved backgrounds as long as Bij = 0) and provides an easy nontrivial check
for the correctness of our results.
439
To see this more clearly, let us consider, as an example, the d = 4 case in the flat
background G = , B = 0, which remains unchanged by the transformation. We assume all
the four Xi to be spatial coordinates and set S = 1. Then, = R. T-duality transformations
along the four coordinates Xi correspond to taking R = 14 , where R is the 4 4
block of R. In this case the spinor representation is already known (see, for example,
[22,23]) and is simply given by = 1 2 3 4 . On the other hand, in the formalism here,
this transformation clearly corresponds to a singular case since det(1 + ) = 0. One can
easily see that the correct spinor representation is reproduced by first going away from the
T-duality point and parameterizing R as
cos
R = sin
O
sin
cos
O
cos
sin
sin
cos
(18)
The singular cases

p then correspond to either = or = or both. One can now
easily evaluate det(Q+ + Q ), which tends to zero in the singular limits, and Aij (some
components of which tend to in the singular limits) and get a well defined as
= cos
cos + sin cos 12 + sin cos 34 + sin sin 1234.

2
2
2
2
2
2
2
2
By setting = = we reproduce the correct for T-dualities along all the four
coordinates, while setting only one of the angles to corresponds to T-dualities along two
coordinates, combined with a nontrivial O(2, 2)ptransformation involving the other two.
This discussion highlights the importance of the det(Q+ + Q ) factor which is essential
in eliminating the terms in (16) particularly the identity that should not appear in the
flat-space T-duality
p limit.
Since neither det(Q+ + Q ) nor Aij depend on the metric, its inclusion does not
change the situation described above for the flat space, except that now also contains
the (det Q )1/2 factor. For example, if we consider two T-dualities along X1 and X2 with
B12 = 0, then = (det G)1/2 12 , as can be directly verified using the result for a single
T-duality in [17]. This case is relevant to the application of SO(d, d) transformations to the
simplest forms of Dp-brane solutions for which the NSNS two form vanishes.
In the more general case when B 6= 0, the singular limit det(Q+ + Q ) = 0 includes,
in particular, the interesting case Q+ = Q , or S 1 R = (1d + d1 B)(1d d1 B)1 ,
i.e., when B parameterizes the orthogonal matrix R1 S through equations similar to (13).
Clearly, this can arise only in even d with constant B and the corresponding will have
only one term which is proportional to i1 id . Thus we see that in this case, there exists
a special SO(d, d) transformation, the spinor representation of which has the same form
as that for d T-dualities in flat space. In general, even in non-flat backgrounds, one can
parameterize ab as R in (18) to show that a well defined spinor representation always
exists in the singular limits. Having constructed the spinor representation of the induced
twist, we set out to find the SO(d, d) action on the massless NSR and RR fields that
involve .
440
4. O(d, d) transformation of spacetime spinors

Let us consider the gravitinos 9M , the dilatinos and the spacetime supersymmetry
transformation parameters in type-II string theories. The subscripts denote the
worldsheet sector (+ for left-moving and for right-moving) in which the spinor
index of the spacetime spinor, or equivalently its Ramond component, originates. We
assign positive chirality to in both IIA and IIB theories which fixes the chiralities
of all other spinors. All spinors are assumed to be independent of d coordinates
Xi but may depend on the remaining 10 d coordinates X . The action of the
nontrivial elements of the O(d, d) group on these spinors can be obtained by demanding
consistency with supersymmetry transformations. These supersymmetry transformations
were constructed in [24] for type-IIA theory and in [25] for type-IIB theory. Here, we write
the supersymmetry transformations in the string metric and in conventions more natural
to string theory as given in [17]. Denoting the supersymmetry variations corresponding
to + and by + and respectively, it turns out that 9 and depend only
on the NSNS fields, on which the O(d, d) action is well-known, and are independent
of RR fields. This is expected from the fact that in string theory the supercharges act
independently on the left- and right-moving worldsheet sectors, interchanging R and NS
boundary conditions. Therefore, we can use these variations to determine the O(d, d)
action on the spinors. On the other hand, the variations 9 and depend on the RR
fields alone and will be used to obtain their transformations in the next section. In general,
these results are valid for O(d, d), and we restrict ourselves to the SO(d, d) subgroup
only when using the explicit construction of the spinor representation given in the previous
section. The derivation below closely follows the one in Section 3 of [17] and generalizes
it from the single T-duality case to the O(d, d) group actions.
= wMab 12 HMab , the gravitino

Defining the torsionful spin-connections as WMab
supersymmetry variations 9 are given by
ab + ,
(19)
9M = M + 14 WMab

ab
1 +
(20)
+ 9+M = M + 4 WMab + + ,
where, denotes 3-spinor terms the explicit forms of which are not needed here. To
. In the O(d, d)
obtain the O(d, d) action on 9M , we need the transformations of WMab
e
transformed theory, one can define two sets of torsionful spin-connections: W
()Mab
e
corresponding to the vielbein e() and W
corresponding
to
the
vielbein
e
,
where
(+)
(+)Mab
the two vielbeins are given by (9). One can then show that (see Appendix A)
1 N
e
(21)
W
()Mab = WNab Q+ M ,

+
1 N
e+
(22)
W
(+)Mab = WNab Q M .
Since we have chosen e() as the vielbein in terms of which the transformed theory is
to be written, the corresponding supersymmetry variations should be written in terms of
e
W
()Mab alone. Let us first consider 9M . Equation (7), along with the fact that is
1 M
M
independent of Xi , implies that M (Q1
+ ) N = N . Then multiplying (19) by (Q+ ) N
441
and using (21) leads to the corresponding variation in the transformed theory provided
M
eN = 9M (Q1
we identify = and 9
+ ) N , up to 3-spinor terms. Similar steps
applied to + 9+M , on the other hand, will lead to
M

ab
1 e+
+ + ,
+ 9+M Q1
N = M + 4 W(+)Mab

1
1 e+
M + 4 W()Mab ab + + .
=
e()M by a local
e(+)M is related to W
Here, we have used the fact that, due to (10), W
1
1
e
e
Lorentz transformation , W(+)M = W()M + M , which has a spinor
representation defined by (12). The above equation will lead to the correct form for
e+M in the transformed theory provided we identify + = + and
the variation + 9
M
e
+ 9+N = + 9+M (Q1
) N , again up to 3-spinor terms. Thus under an O(d, d) action,
the supersymmetry transformation parameters transform to
= ,
+ = + ,
(23)
while the transformation of the gravitinos can be read off from the O(d, d) action on the
supersymmetry variations 9M as
N
N
e+M = 9+N Q1
eM = 9N Q1
9
(24)
9
+ M,
M.
Note that though the transformations we obtained for 9M receive corrections cubic
in the spinors, our final results for 9M do not receive such corrections and are, in
this sense, exact. This can be understood on general grounds by noting that a 3-spinor
correction to (24) will induce 4- and higher spinor couplings with derivative interactions
in the supergravity Lagrangian, which should not exist. More precisely, such corrections
are ruled out by the compatibility of the O(d, d) transformations with the supersymmetry
variations of the NSNS fields. This was shown in [17] for the case of a single T-duality.
The same argument applies to the general case here simply because, in principle, any
nontrivial O(d, d) transformation can be constructed by intertwining T-dualities with
GL(d) transformations (an explicit construction, though not needed for this argument, will
be given later). Thus the O(d, d) transformations have the basic property that they do not
mix quantities with different spacetime fermion numbers.
Though the results we have so far are enough to derive the O(d, d) action on the
RamondRamond fields, for the sake of completeness, we also obtain the action on the
dilatinos. The dilatino supersymmetry variations are given by

1 MNK
HMNK + ,
(25)
= 12 M M 12
where again denotes 3-spinor terms. To relate these variations to the corresponding
ones in the O(d, d) transformed theory, we need the transformation of HMNK . It is most
useful to write this in the following two equivalent forms (see Appendix A):

M
N
K e
1 ij
e()b
e()c
Wj [ab ec]i . (26)
HMNK = eaM ebN ecK HMNK 3 Q1
e()a
(S R)
in the transformed theory, we get the O(d, d)
Using these and comparing with e
action on the dilatinos as
442

1 ij
e
9j ,
= 12 i Q1
(S R)
1
ij

1
1
e
9+j .
+ = + + 2 i Q+ (S R)
(27)
Note that so far we have not restricted ourselves to supersymmetric backgrounds for which
the spinors and their supersymmetry variations vanish. In such cases, the transformations
above lead to the O(d, d) action on the fermionic zero-modes on the background.
5. SO(d, d) transformation of RR fields

Having determined the O(d, d) action on and 9M , we are now in a position to write
down its action on the RamondRamond field strengths and potentials, again by demanding
consistency of the transformations with supersymmetry. For the explicit construction
of the transformations we restrict ourselves to SO(d, d). This will automatically insure
the SO(d, d) covariance of the supergravity equations of motion as well as that of the
electricmagnetic duality constraints on the RR fields since all these are determined by
supersymmetry. We will then compare our results with the ones obtained in the framework
of NCSYM approach to matrix theory compactifications for the SO(3, 3) action in type-IIA
on T 3 . We will end the section by writing down an SO(d, d) covariant form for the
RamondRamond kinetic terms in the low-energy effective action.
It is most convenient to start with the supersymmetry variation 9+M which involves
the RamondRamond field strengths. In both IIA and IIB theories this can be written
as [24,25] (see [17] for the required change of variables):
9+M =
1
e F M + .
2(8)
(28)
Here, denote 3-spinor terms and F is a bispinor that contains the RamondRamond
field strengths and has the expansion,
F=
X (1)n
n
n!
(n)
FM1 ...Mn M1 ...Mn .
(29)
The fact that and 9+M have the same chirality in IIB and opposite chiralities in
IIA implies that in IIA theory the summation contains only terms with even n (n =
0, 2, 4, 6, 8, 10), while in IIB theory it contains only terms with odd n (n = 1, 3, 5, 7, 9).
Note that, for the time being, we allow n = 0 and n = 10 in type IIA theory, so that our
formalism includes the massive type-IIA theory in [24]. The fact that has definite
chirality (in our case +1 in both IIA and IIB) implies that F satisfies the constraint
F = F 11 which, in terms of the components fields, translates to 1/(10 n)!
F (10n) (10n) = 1/n!F (n) (n) . However, in the following it is more convenient to retain
both F (n) and F (10n) in the summation. Furthermore, we assume that FM1 ...Mn are
independent of the coordinates Xi , which is in fact required by the Xi -independence of
and 9+M in (28). However, for the time being, we do not demand the RamondRamond
(n)
to be Xi independent as this would exclude the massive IIA theory
potentials CM
1 ...Mn
443
(we will briefly discuss this issue later). Let us now consider the supersymmetry variation
e+M in the theory obtained after an O(d, d) transformation
9
e+M =
9
1 e
e F e()M + .
2(8)
(30)
e has the same form as F above, with F (n) and N replaced by F

e(n) and eN ,
F
()
N
M
= QN
respectively. Since e()
M , this becomes
e=
F
X (1)n
n
n!
e(n) QN1 . . . QNn M1 ...Mn .

F
Mn
N1 ...Nn M1
(31)
Now, using (23) and (24) along with (8) in (30) and comparing the result with (28), we get
a compact expression for the O(d, d) action on the RamondRamond bispinor field as
p
e = det Q F .
(32)
F
In other words, e F , as well as (det G)1/4F transform as spinors under the induced Lorentz
twist. This expression is valid for O(d, d) and could interchange IIA and IIB theories for
elements with determinant 1. However, to obtain the transformation of the components
of F , we restrict ourselves to the SO(d, d) case and use the explicit expression for as
given by (15). The above equation then reduces to

dp
e = 2 2 det(Q + Q+ ) 1 Aij ij F .
(33)
F
2
The expression for the component fields are obtained by fusing together the antisymmetrized products of -matrices in the expansions of (16) and F (29), using a useful
-matrix identity in [2729] (that we quote in Appendix B), and then matching the result
e in (31). We will first consider a simple case and then write down
with the expansion of F
the general result.
5.1. The d = 2, 3 cases
Before writing down the general result, it is instructive to look at the simplest case of
[d/2] = 1 (corresponding to SO(2, 2) and SO(3, 3) transformations). In this case, one can
easily work out the transformation of RamondRamond field strengths, using the -matrix
identity (B.2) in Appendix B, to obtain
p

d2
e(n)
det(Q + Q+ ) FN(n)
12 Ai1 i2 n(n 1)Gi1 N1 Gi2 N2 FN(n2)
F
M1 ...Mn = 2
1 ...Nn
3 ...Nn
1 N1
(n)
(n+2)
1 Nn
+ 2nGi1 N1 Fi2 N2 ...Nn + Fi2 i1 N1 ...Nn Q [M . . . Q Mn ] .
(34)
1
This expression is valid in both IIA and IIB theories, depending on whether n is even
or odd, and includes the massive IIA theory if the field strengths F (0) or F (10) are
non-vanishing. Also, by construction, it does not mix odd-form and even-form field
strengths. However, it does transform the field strengths nontrivially. In particular, if
the original theory has only a nonvanishing q-form field strength F (q) , corresponding
to D(q 2)-branes, the theory obtained after the transformation will generically have
nonzero q, (q 2) and (q + 2)-from field strengths, depending on whether the
444
transformation involves directions parallel or transverse to the brane, or a mixture of

the two. On the other hand, by construction, the new configuration preserves the same
amount of supersymmetry as the original solution. Thus the new configuration can only be
a real (non-marginal) bound state of Dq, D(q 2) and D(q 4)-branes. A more detailed
discussion of this issue can be found in [26].
As a simple explicit example, let us consider the case of nontrivial SO(2, 2) transformations in flat-space. Setting S = 1, this corresponds to

sin 2 0 1
cos
sin
ij
,
R=
,
A =
sin cos
cos 2 1 0
p
with Q = 1 and det(Q + Q+ ) = 2 cos(/2). Taking care of the antisymmetrization
factors, one obtains
(n)
(n2)
e(n)
F
123 ...n = cos 2 F123 ...n + sin 2 F3 ...n ,
(n)
(n)
e(n)
F
12 ...n = cos 2 F12 ...n + sin 2 F22 ...n ,
(n)
(n)
e(n)
F
22 ...n = cos 2 F22 ...n sin 2 F12 ...n ,
e(n)... = cos F(n)... sin F (n+2) .
F
n
1
2 1 n
2 121 ...n
In particular, for = one reproduces the correct result for two T-dualities along X1 and
X2 , as can be verified directly by using the T-duality transformation of F (n) as given, for
example, in [17]. Furthermore, the first equation above implies that if F (0) is nonzero, then
e(2) . Therefore, in the case of massive IIA theory, one cannot restrict
one gets a nonzero F
12
the potential C (1) to be Xi independent. This is discussed in more detail in [30] for a single
T-duality and in [31] for two T-dualities.
5.2. The general SO(d, d) case
Following the same steps as described above, one can work out the transformation of
RamondRamond field strengths under the action of nontrivial elements of the SO(d, d)
group, for generic d . Using (B.2) in (33), and after some straightforward manipulations,
one obtains
"
[d/2]
X (1)p
p
(n)
d2
e
det(Q + Q+ )
A[i1 i2 . . . Ai2p1 i2p ]
FM1 ...Mn = 2
p!2p
p=0
!#
2p
X (2p)!
(n+2p2r)
n
Cr Gi1 N1 . . . Gir Nr Fi2p ...ir+1 Nr+1 ...Nn

(2p r)!
r=0
N1
1 Nn
(35)
Q1
[M . . . Q Mn ] ,
1
are the binomial expansion coefficients and Aij and det(Q + Q+ ) are given
where,
by (14) and (17), respectively. To evaluate Q1
, one can use either (5) or (6), according
nC
r
445
to convenience. Besides the nontrivial elements that we have considered so far, the full
SO(d, d) group also includes GL(d) transformations involving Xi and constant shifts in
Bij . F (n) simply transform as n-forms under the first set and are invariant under the second
set. This completes the construction of the full SO(d, d) action on the RamondRamond
field strengths.
Thought the transformation formula for F (n) looks rather bulky, in practice, at least some
components of it are simplified when restricted to specific solutions. This is especially the
case when the potentials C (n) can be chosen to be Xi independent, so that Fi1 ...in = 0.
5.3. Transformation of RamondRamond potentials
So far, we have only demanded the Xi -independence of the RR field strengths and not
that of the potentials since this would exclude the massive type-IIA theory. This is pointed
out above for the SO(2, 2) case in flat space and was considered in more detail in [30,31].
Let us now restrict ourselves to the standard type-IIA and IIB theories by setting the IIA
mass parameter F (0) to zero. The Xi -independence of the RamondRamond potentials is
now compatible with SO(d, d) transformations.
It is convenient to define the potentials C (n) such that
n!
(n3)
H[M1 M2 M3 CM4 ...Mn ] .
(36)
3!(n 3)!
In this convention, the potentials are invariant under the NSNS 2-form gauge transformations, but C (4) is not invariant under the SL(2, R) group of type-IIB theory. To obtain
the SO(d, d) action on the RamondRamond potentials, the straightforward (although tedious) procedure would be to substitute (36) in (35), and use the transformation of H as
given by (26), to work out the transformation of C (n) iteratively in n. The calculation may
get somewhat simplified if one chooses to work in the local Lorentz frame, since then one
will not have to bother about the factors of Q1
in (35). However, here we will obtain this
transformation by a very simple argument.
First, note that any nontrivial SO(d, d) transformation can, in principle, be constructed
as a combination of appropriately chosen discrete T-duality transformations and simple
coordinate rotations obtained by setting S = R = A. Since this statement is crucial for
our argument here and the one in the next section, we will spell it out in some detail:
any O(d) rotation, say R, can be decomposed as a product of reflections about planes
perpendicular to properly chosen axes, or equivalently, as a product of reflections Ti
about planes perpendicular to the coordinate axes x i , and properly chosen rotations Ak ;
R = Tin Akn . . . Ti1 Ak1 . If we choose S = Akn . . . Ak1 , then the O(d, d) transformation
implemented by R and S corresponds to a sequence of T-duality transformations Ti
intertwined with coordinate rotations Ak , which proves the statement above. Therefore,
knowing how F (n) transforms under a single T-duality, one can in principle construct its
transformation under any nontrivial SO(d, d) element by using this decomposition, and
reproduce equation (35). However, as emphasized in [17], F (n) and C (n) transform in
exactly the same way under a single T-duality (this is true up to a sign which is immaterial
since we need even number of T-dualities). Being n-forms, they also transform in the same
(n)
(n1)
FM1 ...Mn = n [M1 CM2 ...Mn ]
446
way under coordinate transformations. This implies that if we construct the action of an
SO(d, d) element on C (n) , using its decomposition in terms of T-dualities and rotations,
then we will end up with the same expression as for F (n) . Hence,
"
[d/2]
X (1)p
dp
(n)
e
2
det(Q + Q+ )
A[i1 i2 . . . Ai2p1 i2p ]
C
M1 ...Mn = 2
p!2p
p=0
!#
2p
X
(2p)! n
(n+2p2r)
Cr Gi1 N1 . . . Gir Nr Ci2p ...ir+1 Nr+1 ...Nn

(2p r)!
r=0
N1
1 Nn
(37)
Q1
[M . . . Q Mn ] .
1
As with the field strengths, C (n) are invariant under constant shifts of Bij and transform as
n-forms under GL(d). This completes the construction of the SO(d, d) action on Ramond
Ramond potentials. Defining a bispinor C in the same way as F in (29), the above
transformation takes the compact form
1
e 4 Ce= [det G] 4 C.
[det G]
1
(38)
We will not discuss the SO(d, d) action on the RR potentials in the massive IIA theory,
in which case some of the C (n) will necessarily have an Xi dependence and the above
transformations will get modified. However, we comment that, as shown in [17] for the
b(n) in
single T-duality case, it should be possible to define new Xi independent variables C
terms of C (n) and the mass parameter, such that the SO(d, d) action on the potentials has
b(n) .
exactly the same form as in (37), but now with the C (n) replaced by the new variables C
5.4. Comparison with results from NCSYM
In some cases, the transformation of the RR potentials has also been studied
in the framework of M-theory compactifications to super YangMills theories on
noncommutative tori [14,16], where the string theory SO(d, d) transformations are
implemented as a Morita equivalence. From the 10-dimensional string theory point of
view, the transformation obtained in this approach is valid only in the Gij 0 limit and
therefore, is not expected to coincide with the ones given above. To make a comparison
with these results, we restrict ourselves to the case considered in [14], which studies the
1-form and 3-form potentials in type IIA theory compactified on a 3-torus. The RR
potentials are assumed to have nonzero components only along the torus directions. In
our approach, the SO(3, 3) transformation of these potentials can be read off from (37) (or
more directly from (34) after replacing F (n) by C (n) ) as,
p
p
ei
23/2 C
det(Q+ + Q )
eij k
23/2 C
det(Q+ + Q )

l
= Cl 12 Apq (2Gpl Cq Cpql ) Q1
i,

l 1 m 1 n
= Clmn Apq (3Gpl Gqm Cn + 3Gpl Cqmn ) Q1
i Q j Q k .
447
The corresponding equations obtained in the NCSYM framework do not coincide with
these. However, in the limit Gij 0, we have
pq

Apq = 12 Q1 (S R)
(as Gij 0),
Q+ = Q Q,
and the above transformations go over to,
p

ei det(Q) Cl Q1 l + 1 Clmn Q1 l Q1 m Q1 n (S R)QT j k ,
C
4
i
i
j
k
p

eij k det(Q) Clmn Q1 l Q1 m Q1 n .
C
i
j
k
These coincide with the transformations obtained in the NCSYM approach [14,16] (also
see [32]). We can easily generalize the results obtained so far in the NCSYM approach by
taking the Gij 0 limit in (37).
5.5. SO(d, d) covariant form of the RR kinetic energy terms
Our analysis so far has been based on showing the SO(d, d) covariance of the
supersymmetry transformations in type-II theories. Since the equations of motion in these
theories are determined by supersymmetry, this approach also guarantees their SO(d, d)
covariance, and hence that of the corresponding low-energy effective actions. The SO(d, d)
covariant form of the effective action for the NSNS fields has been known for a long
time [47]. Here, we write down an SO(d, d) covariant form for the kinetic energy terms
of the RamondRamond fields in the type-II low-energy effective actions (ignoring the
subtlety associated with the self-dual 5-form in type-IIB theory).
In terms of the RamondRamond bispinor F given by (29), The RR kinetic energy
terms in both IIA and IIB theory can be written as
X 1

25 p
1p
Fa1 ...an F a1 ...an =
det(G)
det(G) Tr 0 F T 0 11 F 11 ,
8
n!
8
n
(39)
which can be checked using the trace formula (B.3) for the -matrices given in
Appendix B. The index on 0 is a local Lorentz frame index. The right-hand side is simply
11 for a spinor to the bispinor F . In type-IIA
a generalization of the expression
theory 11 F 11 = F while in IIB 11 F 11 = F , which restricts the summation on the
right-hand side of the equation to even n or odd n, respectively. The equation includes F (n)
and F (10n) separately and the duality between the two, including the self-duality of the
5-form, should be imposed by hand. The 11 factors in the expression on the right-hand
side have been inserted so that the kinetic energy terms for odd-forms and even-forms have
the same sign, as should be the case in our metric convention which is {, +, . . . , +}. Now,
using (31) and (4), one can easily check that the RR kinetic energy terms, as given by the
left hand side of (39), are manifestly SO(d, d) covariant since
1 = 0 T 0 ,
11 = 11 .
The RamondRamond fields also enter the action through ChernSimons terms that we do
not consider here, though it should be possible to write these too in a manifestly SO(d, d)
448
covariant form using the bispinor F and a similar quantity constructed in terms of the RR
potentials.
The SO(d, d) covariant form of the RR kinetic terms given above is, as such, not
restrictive enough to determine the SO(d, d) transformation of the RR field strengths if
we did not yet know the transformation. However, the addition of the ChernSimons terms
may change the situation. Another option is to use the kinetic terms alone, but express
the field strengths in terms of the potentials and the NSNS 3-form field strength H . This
form, along with the transformation of H given in (26) is restrictive enough to determine
transformation of the RR potentials. In particular, note that the antisymmetric part of the
1 appearing in (26) is related to Aij appearing in the transformation
factor Q1
(S R)
of C (n) in (37).
The above SO(d, d) invariant form of the action is not invariant under O(d, d)
transformations with det O = 1 (including single T-dualities) that interchange IIA and
IIB theories. This is because in such cases, 11 = 11 and after the transformation
one obtains the kinetic energy terms with the wrong sign.
6. RR potentials as SO(d, d) spinors

We have seen that O(d, d) transformations introduce a twist between the Lorentz
frames associated with the left- and the right-moving sectors of the worldsheet theory.
This twist could be regarded as an element of the spacetime Lorentz group O(9, 1),
though its action differs from that of the Lorentz group in that it affects only part of the
fields. It was shown that the spinor representation of this twist determines the O(d, d)
transformations of RR field strengths and potentials in a natural way. Though the approach
we have followed appears very natural from the perturbative string theory point of view,
there also exists, as described in the introduction, an alternative approach to the problem
in which the RR fields are arranged as components of an SO(d, d) spinor [1016]. While
this construction may appear ad hoc from the point of view of perturbative string theory, it
fits naturally in the U-duality group structure of non-perturbative string theory. However,
so far, it has not been possible to obtain the RR field transformations (35) and (37) in this
approach which, as we shall see, may not be an easy task. In the following, we will discuss
the relationship between these two approaches.
Pn=8 (n)
P
(n) and
Let F and C denote the sums of n-forms n=9
n=1 F
n=0 C , respectively.
Equation (36) can then be written as F = dC H C and the potentials C (n) defined
by it are invariant under the BMN gauge transformations. It is also common to use an
alternative set of potentials C 0(n) defined by
C0 = C eB ,
F = eB dC0 .
(40)
These are not invariant under the BMN gauge transformations. Comparing with the
construction in [14,15], it is clear that C0 is the field in terms of which the SO(d, d) spinor
is constructed. We will briefly review this construction here: Let I , for I = 1, . . . , 2d,
denote Gamma matrices of the Dirac algebra associated with SO(d, d) in a basis in which
449
the invariant metric is of the from diag ( d , d ). Defining a i = ( i d+i )/2, for
i = 1, . . . , d, one can easily verify that a i+ and aj = j k a k satisfy the Heisenberg
algebra {a i+ , aj } = ji . A basis for the spinor representation (Majorana in this case) is
then obtained by applying the raising operators a i+ on the vacuum state |0i, defined by
ai |0i = 0 for all i. Restricting to states with either even or odd numbers of a + s, one
obtains the two MajoranaWeyl basis states. Now consider the components C 0(n)
1 ...q i1 ...inq
of RR potentials, for all n and fixed q. The SO(d, d) MajoranaWeyl spinors constructed
out of the RR potentials are then given by
|C0 i1 ...q =
d
X
0(q+p)
C1 ...q i1 ...ip a i1 + . . . a ip + |0i.
(41)
p=0
Note that since n = q + p is either even (IIB) or odd (IIA), in the summation on
the right-hand side p runs over either even or odd values at a time, ensuring that the
spinor is MajoranaWeyl. The SO(d, d) action on C 0(n) can be obtained by constructing
the spinor representation of the group following the procedure described in Section 3.
However, to elucidate the relationship to the approach followed in this paper, we obtain the
transformation of C 0(n) under nontrivial SO(d, d) elements in a much simpler way, using
our results in Section 5.
The straightforward procedure of obtaining the transformations of C 0(n) from that of
(n)
C , using (40), runs into trouble since the field BMN transforms in a rather complicated
way [5]. However, we can circumvent this and obtain these transformations by a simple
argument: under single T-duality transformations both C (n) and BMN transform in a much
simpler way. Using these transformations (for example, as given in [17]) in (40) one can
easily work out the transformation of C 0(n) under a single T-duality, say along X9 , to obtain:
0(n1)
e0(n)
C
92 ...n = C2 ...n ,
e0(n) ... = C 0(n+1) .

C
91 ...n
n
1 2
(42)
Note that this is exactly how C (n) would transform in flat space. Now, using the
construction described above equation (37), we can intertwine single T-dualities with
coordinate rotations to obtain nontrivial SO(d, d) transformations of C 0(n) . On the other
hand, since C 0(n) and C (n) transform in the same way under coordinate rotations, this
SO(d, d) transformation of C 0(n) can also be obtained from that of C (n) (37) in the flatspace limit of GMN MN and BMN = 0. 3 After some rearrangements, and setting
S = 1d by a coordinate rotation, we obtain
"
[d/2]
X (1)p
dp
0(n)
e
2
det(1d + R)
2[i1 i2 . . . 2i2p1 i2p ]
C
M1 ...Mn = 2
p!2p
p=0
!#
2p
X (2p)!
i1
ir
n
i2p j2p
ir+1 jr+1 0(n+2p2r)
Cr
. . .
Cj2p ...jr+1 [Mr+1 ...Mn M1 . . . Mr ] , (43)
(2p r)!
r=0
3 The hat on reminds us that the transformation for C 0(n) obtained this way is still in curved space and that
spacetime indices are raised and lowered with GMN .
450
with

2ij = d (1d + R)1 (1d R) ij .
(44)
Clearly, 2 is the antisymmetric matrix that parameterizes nontrivial SO(d, d) elements

in (1) (with S = 1d ) and in terms of which the spinor representation of the group can
be constructed as described in Section 3. The dependence on S can be easily restored by
performing a coordinate rotation by an amount S 1 and at the same time replacing R with
S 1 R, as should be the case.
We have obtained equation (43) without having to assume that the RR potentials C 0(n)
transform as SO(d, d) spinors. However, the final result can also be written in terms of the
spinors (41) as

0
e
(45)
C ...q = NT C0 ...q ,
1
where NT stands for the spinor representation of the nontrivial elements of the
SO(d, d) parameterized by SO(d 1, 1) SO(d 1, 1)/SO(d 1, 1). Thus, the explicit
transformation of C (n) (37) leads us, by a very simple argument, to the transformation
of C 0(n) as SO(d, d) spinors (at least for the nontrivial part of the group). The converse,
however, is not true. To construct the transformation of C (n) from that of C 0(n) , we have
to use (40) but the SO(d, d) transformation of BMN [5] is expected to make things
complicated.
As we have seen in the previous section, C (n) and F (n) transform in exactly the same
way under SO(d, d) transformations. This implies that we can also construct SO(d, d)
spinors out of F (n) . In fact, in analogy with C0 in (40), we can define
F0 = F eB = dC0 .
(46)
Then |F0 i1 ...q constructed in analogy with (41) also transform as SO(d, d) spinors.
7. Conclusions
We have obtained the SO(d, d) transformations of the RamondRamond field strengths
and potentials and, in the process, have also determined the transformations of the space
time supersymmetry parameters, the gravitinos and the dilatinos in type-II theories. The
derivation is based on supersymmetry and is therefore guaranteed to be consistent with
the equations of motion provided the fields are independent of the d coordinates with
respect to which the transformation is performed. The transformations we obtain for the
RR field strengths also include the massive IIA theory, though for the RR potentials
we restrict ourselves to the usual massless IIA case. Besides the general cases, we also
discuss some special cases to highlight some features of the transformation and to check
the correctness of our results in these cases. Since the transformations could also include
the time direction, we have been careful to keep track of the indices by explicitly retaining
the flat metric .
Though we restrict ourselves to the SO(d, d) case, all formulas which do
not involve the explicit form of the spinor representation , constructed in Section 3, are
451
also valid for the O(d, d) case. In the SO(3, 3) case we reproduce the results obtained in
the NCSYM formalism of matrix theory compactifications by taking the limit Gij 0. It
is also shown that the RR kinetic energy terms in the low-energy effective action can be
easily written in an SO(d, d) invariant form in terms of the RR bispinor. We also clarify
the relation between our approach and an alternative one which is based on constructing
MajoranaWeyl spinor representations of the SO(d, d) group in terms of the Ramond
Ramond potentials.
The picture emerging is that O(d, d) transformations induce a rotation between the
local Lorentz frames originating in the left- and the right-moving sectors of the worldsheet
theory. The misalignment of the frames can be absorbed by the spinor index associated with
one of the frames. This is the basic mechanism by which the O(d, d) action is transfered
to the Ramond sector of the theory.
The transformations obtained in this paper can be used to construct nontrivial D-brane
configurations, starting from simpler ones. An important feature of the transformation
is that, applied to brane configurations for which the surviving supersymmetries are
independent of Xi , it can produce more complicated brane configurations without reducing
the amount of supersymmetry. This is an indication that the final configuration corresponds
to a nontrivial bound state of branes, since otherwise it would invariably break some of the
supersymmetries. Though we have not considered such applications here, as an example,
we mention [33]. Here, the authors construct a class of SO(4, 4) transformations by
explicitly intertwining T-dualities and spatial rotations to obtain a (D1, D5)-brane system
with a nonvanishing B-field. This solution corresponds to a genuine bound state of D1 and
D5 branes as opposed to the marginal bound state with zero B-field. Some other examples
were studied earlier in [27].
Acknowledgements
I would like to thank C. Angelantonj, I. Antoniadis and A. Armoni for useful discussions
and comments.
Appendix A. Some derivations

In this appendix we briefly describe the steps one could follow to derive equations (21),
(22) and (26). To derive (21), one starts from the covariant constancy of e() :
K
b
b
M
e b =
e M e()
W
a e() M + e() M N e() a ,
NK
()Na
M
M 1 GMP H
= NK
where NK
P NK are the torsionful connections. Using the O(d, d)
2
M
as given in [19] and rewriting the result in terms of WN ab , one
transformation of NK
will finally recover (21) after some manipulations. One will also have to use (7) as well
as Q+ = Q (S R) 1 G to simplify Q1
Q+ . To obtain (22) one follows the same
procedure, but now starting with the covariant constancy condition for e(+) .
452
e
To obtain (26) with the lower signs, we start from the definition of W
()Mab which can
be written as
M
M
N
K
d
e d
e
e()
e()
a e() b e() c HMNK = 2ad w
Mc W() Mc e() b .
e in terms of W and express the result in terms of H . Then, we expand
First, we write W
()
d
d
in
terms
of
the vielbein e() and express the result in terms of wMc
and Wia
w
e()
b . After
Mc
some further manipulations one gets (26) with the lower signs. To get the equation with
upper signs, replace e() by e(+) in the starting equation.
Appendix B. Some -matrix results

We use the metric signature {1, +1, . . ., +1} and define the -matrices such that
a b
(B.1)
, = 2ab .
In the MajoranaWeyl representation all a are real, with 0 antisymmetric and others
symmetric. To fuse products of -matrices into antisymmetrized ones, we use the identity
i+j
X
a1 ...ai b1 ...bj =
k=|ij |
s = 12 (i + j k),
i! j ! [b1
b ...b ]
s
. . . abt+1
a1 ...at ] s+1 j ,
s! t! u! [ai
t = 12 (i j + k),
with
(B.2)
u = 12 (i + j + k).
In the summation, only those values of k appear for which s, t and u are integers, i.e.,
k = |i j |, |i j | + 2, . . . , i + j 2, i + j . The trace of products of -matrices is given
by

a ]
[a1
. . . bkk ] .
(B.3)
Tr a1 ...al b1 ...bk = 25 kl (1)k(k1)/2k![b
1
All antisymmetrizations are with unit weight.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
K.S. Narain, Phys. Lett. B 169 (1986) 41.

K.S. Narain, M.H. Sarmadi, E. Witten, Nucl. Phys. B 279 (1987) 369.
A. Giveon, E. Rabinovici, G. Veneziano, Nucl. Phys B 322 (1989) 167.
K.A. Meissner, G. Veneziano, Phys. Lett. B 267 (1991) 33.
A. Sen, Phys. Lett. B 271 (1991) 295; Phys. Lett. B 274 (1992) 34, hep-th/9108011.
S.F. Hassan, A. Sen, Nucl. Phys. B 375 (1992) 103118, hep-th/9109038.
J. Maharana, J.H. Schwarz, Nucl. Phys. B 390 (1993) 3, hep-th/9207016.
A. Giveon, M. Rocek, Nucl. Phys. B 380 (1992) 128, hep-th/9112070.
E. Bergshoeff, C.M. Hull, T. Ortin, Nucl. Phys. B 451 (1995) 547, hep-th/9504081.
E. Witten, Nucl. Phys. B 443 (1995) 85, hep-th/9503124.
L. Andrianopoli, R. DAuria, S. Ferrara, P. Fr, M. Trigiante, Nucl. Phys. B 496 (1997) 617,
hep-th/9611014.
453
[13] L. Andrianopoli, R. DAuria, S. Ferrara, P. Fr, R. Minasian, M. Trigiante, Nucl. Phys. B 493
(1997) 249, hep-th/9612202.
[14] D. Brace, B. Morariu, B. Zumino, Nucl. Phys. B 549 (1999) 181, hep-th/9811213.
[15] M. Fukuma, T. Oota, H. Tanaka, Comments on T-dualities of RamondRamond potentials, hepth/9907132.
[16] A. Konechny, A. Schwarz, Phys. Lett. B 453 (1999) 23, hep-th/9901077.
[17] S.F. Hassan, Nucl. Phys. B 568 (2000) 145, hep-th/9907152.
[18] I. Bakas, K. Sfetsos, Phys. Lett. B 349 (1995) 448, hep-th/9502065.
[21] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 308 (1988) 221.
[22] J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
[23] J. Polchinski, S. Chaudhuri, C.V. Johnson, Notes on D-branes, hep-th/9602052.
[24] L.J. Romans, Phys. Lett. B 169 (1986) 374.
[25] J.H. Schwarz, Nucl. Phys. B 226 (1983) 269.
[26] J.C. Breckenridge, G. Michaud, R.C. Myers, Phys. Rev. D 55 (1997) 6438, hep-th/9611174.
[27] A.D. Kennedy, J. Math. Phys. 22 (1981) 1330.
[28] J. W. van Holten, A. Van Proeyen, J. Phys. A 15 (1982) 3763.
[29] A. Van Proeyen, Tools for supersymmetry, hep-th/9910030.
[30] E. Bergshoeff, M. de Roo, M.B. Green, G. Papadopoulos, P.K. Townsend, Nucl. Phys. B 470
(1996) 113, hep-th/9601150.
[31] M. Cvetic, H. L, C.N. Pope, K.S. Stelle, Nucl. Phys. B 573 (2000) 149, hep-th/9907202.
[32] B. Pioline, A. Schwarz, JHEP 9908 (1999) 021, hep-th/9908019.
[33] A. Dhar, G. Mandal, S.R. Wadia, K.P. Yogendran, Nucl. Phys. B 575 (2000) 177, hepth/9910194.

Further generalization of the Borel transform for

the non-perturbative regime
L.N. Epele, H. Fanchiotti, C.A. Garca Canal, M. Marucho
Laboratorio de Fsica Terica, Departamento de Fsica, Universidad Nacional de La Plata, C.C. 67,
1900 La Plata, Argentina
Abstract
A new generalization of the Borel transform improving the DuncanPernice proposal, and
designed for obtaining any non perturbative contributions is presented. This new transform leads
to a non-ambiguous reconstruction of the original theory. This generalized transform is applied to
the analysis of a one-dimensional spin chain and the two-dimensional non-linear sigma model on the
lattice. In both models the singularity structure related to renormalons is obtained. 2000 Elsevier
Science B.V. All rights reserved.
PACS: 11.15.Tk; 12.38.Lg
1. Introduction
In Quantum Field theory, different proposals for treating non-perturbative contributions
have been considered. Consequently, the problem of divergent perturbative series in the
coupling constant has been largely studied in the literature [17]. This kind of problems
are clearly present in the infrared regime (IR) regime of QCD, because there it has not been
possible to define an appropriate perturbative parameter.
In several theoretical models, for example 2D spin models, the Borel summation
technique [8] seems to provide a way out. However, concomitant with this difficulty, the
so-called renormalon problem [9] appears. This non-analytical structure, singularities in
the positive Borel axis, comes from particular Feynman diagrams, whose contributions
factorially grow. These non-analyticities are related, in a way that is not free of ambiguities,
to the non-perturbative contributions to the theory. Certainly, there can be present other
non-perturbative contributions of different origin.
Corresponding author. E-mail: afa@venus.fisica.unlp.edu.ar
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 4 2 - 4
L.N. Epele et al. / Nuclear Physics B 583 (2000) 454472
455
The conventional Borel transformation of the weak coupling expansion requires a

precise knowledge of the large order behavior of the perturbative series. In theories of
interest, like QCD in the IR, several difficulties weaken the program. In particular, it
is hard to hierarchize the diagrams that dominate the large order behavior [1013]. On
the other hand, the resummation of the asymptotic series becomes ambiguous due to the
singularities of IR renormalons on the positive real axes of the coupling constant. One
can mention also that several properties like the factorization of perturbative and nonperturbative contributions in the operator product expansion [9] and the presence of the
first singularity in this expansion [14] can be questioned.
An alternative formulation of the Borel transformation, that formally overcomes these
difficulties was proposed by Duncan and Pernice DP [15]. It is based on previous
studies on the subject [1618] and consists of replacing the Borel transformation by the
Laplace transformation with respect to the inverse of the coupling constant. This method
allows one the exact reconstruction of the theory. In other words, the singularities of the
DuncanPernice transformation are unambiguously related to those of the exact theory.
Unfortunately this technique cannot be implemented in terms of perturbative expansions.
In the particular case of asymptotically free theories, for example, one has to analyze them
defined on a finite lattice. It is then necessary to utilize a non-perturbative approach. In such
a case the IR renormalons appear only in the limit of infinite volume. But it is precisely
in this limiting process where the difficulties related to renormalons reappear. There is
no smooth (uniform) connection between the transformation of any amplitude in a finite
lattice and the corresponding one in the infinite volume limit.
The main purpose of this paper, based upon the DP [15] previous contribution, is to
present a new procedure, apt to the treatment of asymptotically free theories, to obtain
any non-perturbative contribution. It implies a new definition of the generalized Borel
transform introduced in DP. Our new definition allows an exact reconstruction of the
theory, necessary in the evaluation of physical amplitudes where the non-perturbative
contributions can be of the same order as the uncertainties of perturbation theory results.
The new proposal has been applied, for testing purposes, to the one-dimensional spin
chain and the 2D non-linear sigma model. In both cases, a complete analytical calculation
can be performed. Then, the smooth connection between the analysis done in a lattice of
finite volume and the continuous limit that can be implemented, via 1/N expansion, was
successfully tested. In fact, this advantage of the new transform, allows one to recover in
those examples, the expected non-perturbative structure in the continuous limit including
the first singularity. Consequently, the new transform should be particularly helpful in
connection with the Monte Carlo simulation of those models that need to be studied in
finite lattices.
In Section 2 we present a brief review on both the conventional Borel transform and the
DuncanPernice proposal. This section also includes the example of the one-dimensional
spin chain in order to show how the mentioned divergence appears. Section 3 is devoted
to the presentation of our new proposal, while Section 4 contains the corresponding
application of our definition to spin models in one and two dimensions. Finally Section 5
contains the final remarks, conclusions and a brief analysis of further applications.
456
2. The Borel transform

Let us first review the main steps towards the Borel regularization procedure. The
methods consists basically in defining a transformation T such that when applied to a given
quantum amplitude Z(f ), expressed in terms of a power series, generally asymptotic, of
the coupling f , produces another amplitude B(s). Finally, the original amplitude is reobtained by inversion of the Borel transform B(s). In other words
T [Z(f )] B(s) Z(f ) = T 1 [B(s)].
(1)
In the case of the conventional Borel transform, one starts by considering the amplitude
Z(f ) given by its asymptotic series expansion
X
cn f n ,
(2)
Z(f ) =
n
whose coefficients are known (for example via Feynman diagram calculation). Then one
defines
X cn
sn.
(3)
B(s) T [Z(f )] =
n!
n
Whenever the coefficients cn are not growing faster than n!, it can be expected that B(s)
will have a non-zero convergence radius. Consequently one can re-obtain Z(f ) as
Z(f ) = T 1 [B(s)] =
Z
exp(s/f ) B(s) ds.
(4)
This last expression clearly shows that this definition is not always applicable due to
the analytic continuation it is necessary to perform to cover all the integration range. This
continuation originates ambiguities. In fact, T 1 is not well defined when B(s) presents
singularities in the integration range. Of course, one can in principle distort the integration
path accordingly, but in this case ambiguities are generated because in general the result
will depend on the particular choice of the integration path.
A typical examples of a Borel summable theory is provided by the non-Gaussian integral
[15], while a non-summable Borel theory is the anharmonic double well [19].
There exist several generalizations to the conventional definition of Borel transform. For
example, Brown, Jaffe and Zhai [14], have proposed
X (1 + s)
cn s n ,
(s)
(5)
BBYZ
(n
+
1
+
s)
n
where b1 /b0 is a constant related to the parameters characterizing the problem under
consideration, in particular the renormalization group beta function. The inverse transform
provides the amplitude
(1 f )
Z(f ) =
f
Z
0
(s/f )s s/f
e
BBYZ (s) ds.
(1 + s)
(6)
457
Notice that in the limit 0, this definition goes into the conventional Borel transform.
Nevertheless, this generalization suffers of the same problems as the original transform.
For this reason, Duncan and Pernice recently introduced [15] a different proposal of
modification of the conventional definition that, in principle, is able to get rid of the
ambiguities. The proposal reads
Z
BDP (s) =
es/f Z(f ) d(1/f ),
(7)
which is an analytic function for Re(s) < 0 whenever the theory has, without ambiguities
in the integration path, an UV cutoff and asymptotic logarithmic behavior. Moreover, its
analytic continuation to the positive half plane is also an analytic function but this time
with a cut on the positive real axis.
The exact reconstruction of the theory, without ambiguities in the integration path, is
obtained through the inverse Laplace transform with the corresponding deformation of the
vertical integration to go around the cut, namely
Z
Z(f ) =
es/f Disc[BDP (s)] ds,
Re(s) > 0.
(8)
The calculation of the IR renormalon singularity structure starts from BDP (s) for
Re(s) < 0 where the integral is well defined. In general it is necessary to redefine the
L (s) in the negative s half
system in a finite lattice characterize by a size L to obtain BDP
plane. After that one has to analytically continue to the opposite half plane. In this way,
and after taking the limit L one should obtain picks of the function corresponding to
singularities of BDP (s) coming from the non-perturbative regime of the theory.
It is worth noticing that the amplitude under consideration has to be a connected one
such that the limit for infinite lattice volume is well defined. This is the case of the vacuum
amplitude and most of the two point Green functions. Moreover, as it has been stressed by
t Hooft [16], the location of singularities is universal, even if their properties are not. For
this reason, the singularities appearing in a given amplitude propagates to other ones.
Just to get an insight of the mechanism introduced, let us discuss the simple model
of a spin chain in one dimension as it was analyzed in Ref. [15]. This is a case of an
asymptotically free theory. The mass gap equation of this model is
L
1 X
1
1
=
SL (m)
f
L
m + n/L
(9)
n=1
that for L large and fixed

1+m
SL (m) ln
m + 1/L
(10)
that induces the obvious definition

S(m) = lim SL (m).
L
(11)
458
The corresponding propagator in the momentum space is

1
,
p2 = 1
(12)
1+m
showing that approximate calculations are not necessary because it could be obtained
directly in the infinite volume limit. In terms of the mass, one has
G(m) =
Z
BDP (s) =
S(m) s S(m)
e
G(m) dm
m
(13)
that provides immediately

1
1
BDP (s) =
s (1 s)
(14)
with evident singularities at s = 0 , 1. Clearly, for different elections of the momentum

there appear singularities for s being any natural number.
This structure of singularities should be obtained also by means of a calculation of the
transform in a finite lattice with the corresponding analytic continuation and the final limit
for L . Unfortunately, when one performs an integration by parts, for a finite volume,
the analytic continuation obtained is
L
(s) =
BDP
1 s SL (0)
1
e
G(0) +
s
s
G(m) s SL (m)
e
dm
m
(15)
showing that in the limit L this expression diverges as Ls due to the first term.
Consequently, this definition does not allow the interchange of analytic continuation with
the limit to an infinite volume.
It has been shown [15] that the origin of this divergent behaviour is in the integration
measure dm for m small and can be solved by adding a convenient multiplicative factor
L (s).
f (m) which is not present neither in the definition BDP (s) and consequently nor in BDP
The difficulty was ad-hoc surmounted in Ref. [15] by adding a term, with a different
integration measure for m small, in the original definition. This extra term cancels the
divergency coming from the analytic continuation, found above.
3. A new generalization of the Borel transform

In order to avoid the above mentioned problems, we have introduced a new generalization of the Borel transform, inspired in the proposal of Ref. [14], which includes a new
integration measure in terms of a real positive parameter .
If x 1/f , where f is the bare coupling constant, and Re(s) < 0, we define

Z
B (s)
ex s Z(x)
x
+1
s
showing the following characteristics
dx
(16)
459
1. The integration measure now acquires the form

s

x
+1
dx
instead of the usual dx.

This new integration measure gives rise to the correct behaviour of the transform when
the analytic continuation from a finite lattice of size L to large volume is performed. This
fact, a real improvement over the previous proposals, will become clear in the examples
presented below. In particular, one finds that the divergent term present in Eq. (15) is
cancelled out whenever the ansatz = L is used. In other words, the modification proposed
guarantee a soft an uniform connection between the theory defined on a lattice and the
continuous.
2. Notice that in Eq. (16), if |Z(x)| < M for x [0, ] and consequently the well known
Lebesgue bounded convergence theorem [21] can be applied to show that when computing
the 0 behaviour of B (s), the limit can be exchanged with the integral. In this way
the DuncanPernice definition is immediately recovered, namely
lim B (s) = BDP (s),
Re(s) < 0.
(17)
3. For any fixed and different from zero, all the properties of DuncanPernice
definition are preserved.
4. The term
s

x
+1
allows one to rewrite B (s) as a Laplace integral because, by means of the change of
variables
u(x) = x + ln() ln(x + )
one can write
Z
B (s) =
eu s H (u) du,
(18)
where evidently
H (u)
Z(x(u))
1
+x(u)
(19)
where x(u) is the inverse relation to u(x) defined above.

This way of expressing the transform guarantee the inversion. Following the steps of the
previous analysis of DuncanPernice, one has
Z
H (u) =
0

eu s Disc B (s) ds
(20)
460
or, in terms of the original amplitude

Z
s

1
1
s/f
+1
e
Disc B (s) ds.
Z(f ) =
f + 1
f + 1
(21)
Notice that the last expression reduces to

X

1
bk () k! [f ()]k+1 ,
Z(f ) =
f + 1
(22)
k=0
where
f () =
f
1 f ln
1
f
+1
(23)
whenever Disc[B (s)] admits an expansion of the form
X

bk () s k ,
Disc B (s) =
|s| < R.
(24)
k=0
Finally, one can check that, under the above conditions, the conventional Borel transform
is obtained in the limit 0.
5. In the particular case of asymptotically free theories one immediately concludes that
our B is an analytic function if an UV cutoff is present. Moreover, its continuation to
the positive half plane is also an analytic function with a cut on the positive real axis. For
further details see Appendix A.
4. Application to non Borel summable models

In this section we re-analyze the models considered by DuncanPernice in order to show
the potentialities of our proposal.
a) One-dimensional spin chain. One can compute B (s) for the propagator G(m), with
arbitrary momentum p2 , exactly in the limit of infinite volume, namely for L , to
obtain
s

Z
S(m)
S(m) s S(m)
e
+1
G(m)
dm, Re(s) < 0.
(25)
B (s) =
m
It is possible to verify that this equation, in the limit 0 and for p2 = 1, goes into the
DuncanPernice result because as mentioned above, the interchange of the limit and the
integration can be performed. For the details we refer to Appendix B.
Notice that in the opposite situation, corresponding to the limit, cannot be
interchanged with the integral in Eq. (25) since the bounded convergence theorem requires
that
Z
|Z(x)| dx < M
0
461
condition that in general is not fulfilled. For example, in the case of the spin change one
has Z(x) = 1 ex .
If one computes Eq. (25) for an arbitrary momentum p2 in the limit, the result
is (see Appendix C)
2

X

1
1
p 1 m1
1
1
(26)
B(s) = I p2 + 2 + 2
p
s (p + 1)
(s m) p2 + 1
m=1
that presents singularities for s = 0, 1, 2, . . . . Notice that, as should be expected from

t Hooft arguments [1618], the position of renormalon singularities in the Borel transform
is not changed, even if the corresponding residues could be different. In other words, the
singularity structure obtained by DuncanPernice, for 0 in our formulation, is not
modified when the limit is considered.
On the other hand and from the practical point of view, it is necessary to reformulate the
model in a finite lattice of size L by means of the mass gap equation (10). Consequently,
Eq. 25 changes into
s

Z
SL (m)
SL (m) s SL (m)
e
+1
G(m)
dm, Re(s) < 0.
(27)
BL (s) =
m
It is clear that for the determination of the renormalon singularity structure, both limits
L and are in order.
The example under consideration will clearly show that this double limit is in effect nonuniform. However, they exist when ordered, namely first and then L going to infinity.
Moreover the limits can be performed simultaneously whenever is of the same order
as L. This kind of ansatz provides the correct results in all the examples that have been
analyzed and it is potentially very useful for numerical calculations on a finite lattice.
Just to analyze the correct behaviour of the analytic continuation of the transform
provided by the integration by parts, we notice that the integration measure is modified
by the factor
s

SL (m)
+1
that in the limit m = 0 of integration behaves as

s

ln(L)
+1
and is finite and L independent on the rest of the integration interval. Then, whenever the
parameter is of the order of the lattice size L, this term, near the origin, contributes
with Ls . Consequently, the integral of the total differential involved in the analytic
continuation, does not diverge in the L limit, at variance with the DuncanPernice
proposal. More precisely
Ls

Z
SL (m)
SL (m) s SL (m)
e
+1
G(m)
dm, Re(s) < 0
(28)
BL (s) =
m
L
0
462
and through an integration by parts, similar to that performed to arrive in Eq. (15), the
potentially divergent term results in
Ls

SL (0)
1 s SL (0)
e
+1
G(0)
(29)
s
L
that is finite in the large volume limit because of the presence of the last parenthesis. This
success is clearly related to the integration measure naturally introduced in our modified
transform. In summary, via the proposed ansatz, the results are as follows
(s) = lim B (s),

lim BL
Re(s) > 0.
(30)
For details see Appendix C.

b) Non linear (N) model in 2D. In this model, that we include for completeness, the
mass gap equation is given by
1 X
1
1
= 2
(31)
+m
f
L
d(
k)
with

2 x
2 y
+ sin
,
d(k) 4 sin
L
L

2 x 2 y
,
, x, y = 1, 2, . . . , L.
k =
L
L
Then, one can use the approximate value, for L large and fixed, of the double sum in the
mass gap equation, to obtain

(1 + m4 )1/2 + 1 2 /2L2
1
m 1/2
1
ln
1+
.
f
4
4
(1 + m4 )1/2 1 + 2 /2L2
Just as in the previous one-dimensional model, the renormalon singularity structure
is dominated by the contributions to the transform coming from small m values. In this
regime, the last equation becomes

1
1 + m/16
1
sigma
ln
SL (m).
f
4
m/16 + 2 /4L2
Redefining m = m/16 and L = 4L2 / 2 , one can express the mass gap equation for
this model in terms of the corresponding one for the spin chain obtained in the previous
paragraph. One finally obtains

(m )
S chain
1 + m
1
sigma
ln 1
.
= L
SL (m) =
4
4
L + m
Notice that this model is also soluble in the continuum (L ) when the transform is
computed on the propagator in the momentum space for large N
Gsigma (m) =
1
.
d(q)
+m
If one takes d(q)

= 16, it clearly results that Gsigma (m) = (1/16) Gchain(m ). Then
1
,L
Bsigma
(s) =
64

463
SLchain
(m ) (s/4)S chain
(m )
L
e
Gchain (m )
m
0
chain
SL (m )
+1
s
dm .
Defining now 4 = , and (s/4) = s , the integral expression that results is exactly
the same as that for the one-dimensional spin chain, clearly in the modified variables,
namely
,L
(s) =
Bsigma
1 ,L
B
(s )
64 chain
that, as it has been previously shown it presents singularities for s = 0, 1, 2, . . . . This

singularities, in terms of the sigma model variables, appear at
s = 4s = 0, 4, 8, . . . .
The mentioned relationship between both models allows one to conclude that the
singularity structure is easily recovered if the transform is non-perturbatively computed
in the one-dimensional model.
5. Summary and conclusions

The DuncanPernice definition of the Borel transform allows a rigorous reconstruction
of the theory but it is not particularly apt for non-perturbative calculations that in those
theories with an UV cut off end a numerical simulation on finite lattices. The difficulty is
related to the absence of a soft and uniform correspondence between the lattice definition
and the continuous limit. It is understood that this is due to the behaviour, for small mass
values, of the integration measure in the transform definition. This is clearly seen in the
one-dimensional spin chain example.
We have presented a new generalization of the previous definition of Duncan and Pernice
that, when applied to asymptotic free theories allows one to obtain information about the
non perturbative-structure. To this end we need to include a different integration measure,
function of a parameter that allows beside the exact reconstruction of the theory, a soft
limit between the lattice definition and the continuous. In fact, the procedure is to fix
L and then consider L sufficiently large. Our generalized transform provides the correct
renormalon singularity structure in both models considered, namely the spin chain in one
dimension and the sigma model in two dimensions. Notice that the simplicity of the models
analyzed avoids the use of numerical simulations.
Due to the well-known analogy between spin models in D dimensions and 2D dimension
gauge models, one can plausibly think that the results obtained from the analysis of the
non-linear (N) model in two dimensions will apply to QCD in four dimensions, even if
in this last case and due to its inherent complexity, Monte Carlo method could be necessary
to use in order to compute the non-perturbative structure of singularities of the theory.
464
From a mathematical point of view, the new introduced tool to treat pathological cases
related to Borel transform, presents several important ventages, namely
It allows an exact reconstruction of the amplitude by means of the Laplace technique
that ends with an integral representation of the Borel type.
It has a generalized character through the arbitrary parameter .
In the limit 0 the DuncanPernice definition is recovered.
The use of asymptotic expansions is avoided and consequently their inherent errors
are not present. On the other hand, due to its generalized character, if Disc[B (s)]
is asymptotically expanded, one can analyze the corresponding behaviour of the
coefficients through
X

1
bk () k! [f ()]k+1
Z(f ) =
f + 1
k=0
in terms of an effective coupling. From this particular case of the generalized

definition, one re-obtain the conventional Borel transform in the limit 0.
This new tool could be applied to other Borel non-summable theories.
Acknowledgements
We warmly thanks A. Duncan and S. Pernice for very useful comments and encouragement. This work has been partially supported by CONICET and Agencia Nacional de
Promocin Cientfica y Tecnolgica, Argentina.
Appendix A
Analytic behaviour of the transform

Z
B (s)
exs Z(x)
x
+1
s
dx,
x = 1/f, Re(s) < 0
following the same procedure as DuncanPernice to study the analytic behaviour of their
modified definition of the Borel transform, we consider Z(x) as the partition function
defining the theory, namely
Z
Z(x) D exS() ,
where S() is the Euclidean action. In the examples of interest as the spin model or
lattice QCD, the action is a positive, bounded continuum function of the fields involved
0 6 S() 6 Smax . Then replacing
Z
B (s) =
Z
D
e
0
x(S()s)
x
+1
s
dx,
Re(s) < 0
465
and changing to the variable t = x(S() s)

Z
1
D
(S() s)
B (s) =
Z
e
t
+1
(S() s)
s
dt.
Notice that the last integral is no more than the hypergeometric function of the second type
[20]. Consequently
Z

B (s) = D G 1, 2 s, (S() s) , Re(s) < 0.
Then, the analytic properties of G(1, 2 s, (S() s)) allows one to conclude that
the transform B (s) is an analytic function on the negative half plane.
To the rest of the positive half axis, the discontinuity of B (s) through the cut is easily
computed in terms of the discontinuity of G(, , z) given by
Z

2i
e(S()s)
,
D
B (s) =
(s)
[(s S())]1s
Re(s) > S() > 0, (1 s) 6= n.
Appendix B
Let us now analyse the 0. In this limit, the DuncanPernice result is re-obtained.
We discuss explicitly the application to the spin model in one dimension
Starting with
Z
BDP (s) =
s

x
xs
+1
lim e G(m(x))
dx,
0
Re(s) < 0, x = 1/f
(B.1)
one has to show that the limit can be exchanged with the integration just to obtain the
generalized transform. In other words, one has to verify that
Z
s

x
+1
lim exs G(m(x))
dx
0
Z
exs G(m(x))
= lim
x
+1
s
dx
(B.2)
BDP (s) = lim B (s).

0
(B.3)
In the particular model under consideration, the propagator, when written in terms of the
inverse of the coupling constant via the mass gap equation in a finite lattice, takes the form
G(m(x)) =
1
= 1 ex .
m(x) + 1
(B.4)
466
Going back to B (s)

Z
B (s) =
xs
(1 e
x
+1
)
s
dx,
Re(s) < 0
(B.5)
and changing the scale by means of x = t one obtains

Z
B (s) =
t s
(t + 1)
Z
dt
et (s1)(t + 1)s dt,
Re(s) < 0,
(B.6)
where the integrals represent precisely confluent hypergeometric functions [20], namely

(B.7)
B (s) = G(1, 2 s, s) G 1, 2 s, (s 1) , Re(s) < 0.
The well known property
0
G(1, 2 s, z) =
X
r (1)zr
r=0
r (2)r!
(r + 2) + z1
(B.8)
that for small values of z is

0
G(1, 2 s, z) ' z1
(B.9)
allows us to conclude that

lim B (s) = lim G(1, 2 s, s) lim G 1, 2 s, (s 1)
1
1
= BDP (s)
=
s (1 s)
(B.10)
(B.11)
showing that the DuncanPernice result is reobtained. Clearly, in this example the validity
conditions for the bounded convergence theorem are satisfied.
Appendix C
Application of B (s) to the one-dimensional example. Determination of the singularity
structure in the infinite volume limit
From Eq. (11) and considering G(m) = 1/(m + p2 ) as the momentum space propagator
Z
B (s) = (m + 1)s1 (m)s1
changing variables to u =
Z
B (s) =
ln( m+1
m ) + 1, it results
e(u1) 1

s

m+1
1
1
ln
+1
dm
m
m + p2
+ p2
es(u1)us du.
467
To explicitly exhibit the singularity structure of the transform on the positive half plane,
one defines z = e(u1) with 0 6 z 6 1 and expresses the integrand denominator as
1
1
e(u1) 1
when
+ p2
1
p2
1
z
p 2 (1z)
+1
1
F (z),
p2

z
< 1 u > 1 + 1 ln 1 + 1

p2 (1 z)
p2
one can use the power series expansion

F (z) =
X
()n
n=0
p2n
(z)n
1
(1 z)n
and for |z| < 1, i.e., u > 1

F (z) =
X
()n
n=0
p2n
(z)n
n/2
Ck (1)zk ,
k=0
n/2
where Ck (1) are Guegenbauer polynomials.

Splitting now the integrand into two intervals just to safely allow the interchange of sum
and integration and using the above obtained expression for F (z), one has

1+ 1 ln 1+ 12
p
Z
1
es(u1)us du
B (s) =
1
2
+
p
e(u1) 1
1
X
X
n=0 k=0
()n n/2
C (1)
p2(n+1) k
1+ 1 ln 1+
1
p2
us es(u1)(snk) du.
The first integral is well defined for all s because the integrand is a continuous function
in the integration range. On the other hand, all integrals in the second term are finite for
Re(s) < n + k. If one defines now t = u(s k n) inside those integrals, one gets for
the second term T2
T2 =
n,k=0
()n n/2
1
C (1)e(snk)
p2(n+1) k
((s k n))s+1
Z
(skn) 1+ 1 ln 1+
1
p2
t s et dt,
where one can distinguish the presence of the incomplete gamma function, namely
Z
(a, x) =
x
t a1 et dt = ex x a G(1, 1 + a, x),
468
where G is the confluent hypergeometric function of the second type. Then the previous
expression goes into

1s
X
X
1
1 snk
()n n/2
1
C (1) 1 + 2
1 + ln 1 + 2
T2 =
p
p
p2(n+1) k
n=0 k=0

1
1
G 1, 2 s, (n + k s) 1 + ln 1 + 2
.
p
First term
1+ 1 ln 1+
I (p, s) =
1
1
p2

(e(u1) 1)
es(u1)us du
1 + p2 (e(u1) 1)
changing variables t = e(u1) the integral can be written as

1+
1
p2
I (p, s) =
1
s

(t 1)
1
s1
t
dt
1 + ln(t)
1 + p2 (t 1)
for large and fixed one can expand

s 2
1
ln (t) + O(1/3 )
exp s ln 1 + ln(t) = t s exp
2
and also

s 2
1
ln (t) + O(1/2 )
exp s ln 1 + ln(t) = t s exp
2

s 2
ln (t) + O(1/2 ) .
= t s 1 +
2
Going back to I
1+
1
p2
I (p, s) =
1
1+ 12
p
I (p) =
1
1+
1
(t 1)
t 1 dt +
1 + p2 (t 1)
1
p2
(t 1)
s
t 1 ln2 (t) dt,
2
1 + p2 (t 1)

1
(t 1)
1
t dt + O
2
1 + p (t 1)
and explicitly solving the integral

p2
1
1
1
ln
1
+
ln(2)
+
O
I (p) = 1 +
1 p2
1 p2
p2
and in the limit of

I (p) = I (p ) = 1 +

p2
1
1
ln 1 + 2 .
ln(2)
1 p2
1 p2
p
469
Second term
From the series expansion of G, it is easy to find the asymptotic behaviour

1
1
G 1, 2 s, (k + n s) 1 + ln 1 + 2
p
1

=
1
1
(k + n s) 1 + ln 1 + 2
p
that allows one to write
B(s) = B (s) = I (p) +
n,k=0

1 nk
1
()n n/2
1
+
C
(1)
(n + k s)
p2(n+1) k
p2
summing now the two terms, one gets
B(s) = B (s) = I (p2 ) +

X
1 nk
1
()n+1 n/2
1
+
C
(1)
(s k n)
p2(n+1) k
p2
n,k=0
after a change in the mute variables, m = n + k, the double sum can be re-expressed as
2

X
1
p 1 m1
1
1
1
2
B(s) = I (p ) + 2 + 2
s (p + 1)
(s m) p2 + 1
p
m=1
that presents singularities at s = 0, 1, 2, 3, . . . . In this way, we have recovered the Duncan

Pernice results in the infinity volume.
Application of B (s) to the one-dimensional example on a finite lattice. Solution of the
problem posed by the divergence
In the case of a finite lattice of size L, the transform of the propagator can be expressed
in terms of m and L as
Z
,L
B (s) = (1 1/L) (m + 1)s1 (m + 1/L)s1
0

s

m+1
1
1
ln
dm.
+1
m + p2
m + 1/L
Tomando L
Z
BL (s) = (1 1/L)
(m + 1)s1 (m + 1/L)s1

Ls

m+1
1
1
ln
+
1
dm.
m + 1/L
m + p2 L
We follow now the same steps as in the case of the continuum, namely we change variables

m+1
1
+1
u = ln
m + 1/L
470
so that
1
L
ln(L)+1
BL (s) = L
uLs eLs(u1)
1(1/L)eL(u1)
eL(u1) 1
+ p2
du.
In this case
1
p2
1
z(1/L)
p 2 (1z)
+1
1
FL (z),
p2
where z = e(u1) 1/L 6 z 6 1

n
1 X X ()2nr n/2
n k+r
1
F
(z)
=
C
(1)
z
k
2
2
2n
nr
r
p
p
p (L)
n,k=0 r=0
and one obtains

p2 +1
p2 +1/L
1+ L1 ln
BL (s) = L
1(1/L)eL(u1)
eL(u1) 1
+L
n
X
X
n,k=0 r=0
1
L
uLs eLs(u1)
+ p2
du

n
()2nr
n/2
Ck (1)
2(n+1)
nr
r
p
(L)
ln(L)+1
uLs eLs(u1)eLs(u1)(r+k) du.
1+ L1 ln
p2 +1
p2 +1/L
In principle, all the integrals here are well defined for all s because the corresponding
integrands are continuous and bonded functions in each integration range.
Limit to an infinite lattice
c) Determination of the first term. In this case one finds only two differences between the
case of a finite lattice and the case of an infinite one. The first one is related to the change in
the upper limit of the integral and does not introduce any difficulty. The second difference
is related to a change in the integrand denominator

2
1+ L1 ln p2 +1
p
+1/L
Z
(eL(u1) 1)
eLs(u1)uLs du
IL (p, s) = L
1 1/L + (p2 1/L)(eL(u1) 1)
1
this change does not affect its behaviour and consequently one can apply the same
procedure as before to estimate the value of the integral and afterwards one recover the
results for an infinite lattice taking the limit L . In this case one gets

p2 +1
p2 +1/L
IL (p) =
1
471

(t 1)
t 1 dt.
1 1/L + (p2 1/L)(t 1)
Solving the integral

1 1/L + (p2 1/L)
p2 1/L
ln
IL (p) = 1 +
1 1/L
1 p2

2
p +1
1
ln 2
1 p2
p + 1/L
and in the limit

IL (p) = I (p) = 1 +
L

p 2 +1
1
2
p +1/L

p2
1
1
ln 1 + 2 ,
ln(2) =
1 p2
1 p2
p
which are the results that were obtained directly in the infinity volume.
d) Determination of the second term. It can be seen that the limit L is only possible
for Re(s) < r + k where the integrals of the second term are finite

n
X
X
n
eL(srk)
()2nr
n/2
C
(1)
L
r [L(s r k)]Ls+1
p2(n+1) (L)nr k
n,k=0 r=0

L(srk) L1 ln(L)+1
Z
uLs eu du.
1

p2 +1
L(srk) 1+ L ln
p2 +1/L
This last integral, as it is easily recognized, represents in the limit L , the

incomplete gamma function. When it is written in term of G(, , z) the result is

n
X
X
n
1 kr
()2nr
n/2
C (1)
1+ 2
L
r
p
p2(n+1) (L)nr k
n,k=0 r=0
2

p +1
1
.
G 1, 2 Ls, L(s r k) 1 + ln 2
L
p + 1/L
Now, using the asymptotic expansion of G(, , z) we obtain

n
X
X
1 kr
n
1
()2nr
n/2
1
+
C
(1)
.
r (k + r s)
p2(n+1) (L)nr k
p2
n,k=0 r=0
On the other hand, the only term that survives in the sum over r is the one
corresponding to r = n. Note that we reobtain, for the second term, the same result of
the infinite lattice. Finally

= lim B (s)
.
B(s) = lim BL (s)
L
finite lattice, L
infinite lattice
472
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
M. Beneke, Nucl. Phys. B 405 (1993) 424.

A.H. Muller, Nucl. Phys. B 250 (1985) 327.
V.I. Zakharov, Nucl. Phys. B 385 (1992) 452.
C.N. Lovett-Truner, C.J. Maxwell, Nucl. Phys. B 432 (1994) 147; Nucl. Phys. B 452 (1995)
213.
J.C. Le Guillou, J. Zinn-Justin, Large-order Behaviour of Perturbation Theory, North-Holland,
Amsterdam, 1990.
G. Altarelli, CERN-TH/95-309.
J. Fisher, CERN-TH/95-325.
C.M. Bender, S.A. Orszag, Advanced Mathematical Methods for Scientist and Engeneers,
McGraw-Hill, 1978.
M. Beneke, Renormalons, hep-ph/9807443.
A.I. Vainshtein, V.I. Zakharov, Phys. Rev. Lett. 73 (1994) 1207.
D.J. Broadhurstand, A.G. Grozin, Phys. Rev. D 52 (1995) 4084.
N.V. Krasnikov, A.A. Pivovarov, Mod. Phys. Lett. A 11 (1996) 835.
R. Coquereaux, Phys. Rev. D 23 (1981) 2276.
L.S. Brown, L.G. Yaffe, C. Zhai, Phys. Rev. D 46 (1992) 4712.
A. Duncan, S. Pernice, Phys. Rev. D 51 (1995) 1956.
G. t Hooft, in: A. Zichichi (Ed.), The Whys of Subnuclear Physics, Plenum Press, New York,
1977.
W.Y. Crutchfield, Phys. Rev. D 19 (1979) 2370.
F. David, Nucl. Phys. B 209 (1979) 433.
E. Brezin, J.C. Le Guillou, J. Zinn-Justin, Phys. Rev. D 15 (1977) 1544.
M. Abramowitz, A. Stegun, Handbook of Mathematical Functions, Dover, 1970, Chapter 13.
H. Jeffreys, B.S. Jeffreys, Methods of Mathematical Physics, 3rd edn., Cambridge, 1996, p. 43.
Nuclear Physics B 583 [FS] (2000) 475512

gl(N|N) Super-current algebras for disordered

Dirac fermions in two dimensions
S. Guruswamy a , A. LeClair b , A.W.W. Ludwig c
a Institute for Theoretical Physics, Valckenierstraat 65, 1018 XE Amsterdam, The Netherlands
b Newman Laboratory, Cornell University, Ithaca, NY 14853, USA
c Department of Physics, University of California, Santa Barbara, CA 93106, USA
Received 28 January 2000; accepted 18 April 2000
Abstract
We consider the non-hermitian 2D Dirac Hamiltonian with (A) real random mass, imaginary
scalar potential and imaginary gauge field potentials, and (B) arbitrary complex random potentials
of all three kinds. In both cases this Hamiltonian gives rise to a delocalization transition at zero
energy with particlehole symmetry in every realization of disorder. Case (A) is in addition timereversal invariant, and can also be interpreted as the random-field XY statistical mechanics model
in two dimensions. The supersymmetric approach to disorder averaging results in currentcurrent
perturbations of gl(N|N) super-current algebras. Special properties of the gl(N|N) algebra allow
the exact computation of the -functions, and of the correlation functions of all currents. One of
them is the EdwardsAnderson order parameter. The theory is nearly conformal and possesses
a scale-invariant subsector which is not a current algebra. For N = 1, in addition, we obtain an
exact solution of all correlation functions. We also study the delocalization transition of case (B),
with broken time reversal symmetry, in the GadeWegner (random-flux) universality class, using a
sigma model whose target space is an analytic continuation of GL(N|N; C)/U (N|N), as well as its
PSL(N|N) variant, and a corresponding generalized random XY model. For N = 1 the sigma model
is solved exactly and shown to be identical to the currentcurrent perturbation. For the delocalization
transitions (case (A) and (B)) a density of states, diverging at zero energy, is found. 2000 Elsevier
1. Introduction
Quantum field theory in the presence of quenched disorder is an increasingly important
subject in condensed matter physics. Unfortunately, even in two dimensions, fully solvable
but non-trivial models are rather scarce. In this paper we consider, amongst others, what
is perhaps the simplest model of disordered fermions that can be exactly solved but also
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 4 5 - 5
476
S. Guruswamy et al. / Nuclear Physics B 583 [FS] (2000) 475512
exhibits some non-trivial behavior. We work in two spatial dimensions throughout. The
model can be defined by the Euclidean action 1
Z 2
d x
f
L (z + Az )L + R (z + Az )R
S0 =
2

+ m(x)R L + m (x)L R ,
(1.1)
where
d2 x = dx dy,
z = (x + iy)/2,
z = (x iy)/2,
and
Az = Ax iAy ,
Az = Ax + iAy .
The subscripts L, R on the fermion fields denote left versus right movers in the conformal
theory with m = m = 0. Letting m = V + iM, m = V iM, we take V and M to have
Gaussian probability distributions of equal 2 strength g and zero mean, and similarly for
Ax , Ay with strength gA :

Z 2
1
d x
m (x)m(x) ,
P [m, m ] = exp
g
2

Z 2
1
d x
Az (x)Az (x) .
(1.2)
P [Az , Az ] = exp
gA
2
(The couplings g, gA represent twice the variance.) Disorder averaging corresponds to a
Gaussian functional integral:
Z
hOi = Dm Dm DAz DAz P [m, m ]P [Az , Az ]hOi.
(1.3)
In the context of two-dimensional statistical mechanics, the above model may be thought
of as a random field XY model or equivalently the random field sine-Gordon model [1], at
the free-fermion point. The usual sine-Gordon model has m constant, which now acquires
a random phase, preventing the theory from becoming massive. Random field XY models
were first studied by Cardy and Ostlund [2] using replicas (for a recent discussion see,
e.g., [3]). The action in Eq. (1.1) can be related (see below) to a 2D Dirac Hamiltonian
H2 subject to a real random mass M, a random imaginary potential iV with strength
gM = gV = g, and also in a random imaginary gauge-field
H2 = (ix + iAx )x + (iy + iAy )y + M(x, y)3 iV (x, y),
(1.4)
where i are the standard Pauli matrices. This Hamiltonian is non-hermitian. This theory
should be compared with the models described in Refs. [4,5] which were argued to have a
transition in the universality class of the quantum Hall plateau transition; the latter models
have a real random mass and a real scalar potential with strength gM 6= gV , and a real
R
1 Partition function Z = D[ , ] exp(S).

2 The difference of the variances of Gaussian random variables V and M renormalizes to zero.
477
random gauge field, the corresponding 2D Dirac Hamiltonian being hermitian. 3 The same
model as in Eq. (1.4), also describes 4 the continuum limit of electrons hopping on a square
lattice with flux [7], exhibiting a delocalization transition at zero energy, first discussed
by Hatsugai, Wen and Kohmoto (HWK) in Ref. [7]. This involves a doubling of the number
of degrees of freedom (see Section 2 and Appendix B), leading to a random hermitian
4-component Hamiltonian. Applications of our results to this delocalization transition will
be described in section V .
We perform the disorder averaging using the supersymmetric method. This leads to a
leftright currentcurrent perturbation of the gl(N|N) conformal supercurrent algebra for
the N -species version, used to compute N th moment disorder averages. Conformal field
theory methods were previously used by Bernard for the case where gM (or gV ) = 0 [1],
which leads to the osp(2N|2N) algebra. Also, the case of gV = gM = 0, but gA 6= 0, where
the conformal symmetry is unbroken and the model is equivalent to free fields, was studied
in [4,8]. Though our model has a non-zero renormalization group -function, some special
features of the gl(N|N) algebra, in particular the nilpotency of the fermionic generators of
gl(N|N) and the existence of two quadratic casimirs, lead to a complete solution for all
correlation functions in the case N = 1, and to the solution of the correlation functions of
all Noether currents in the case of general N . The only non-vanishing -function is that
for gA which we compute exactly. It only depends on g. Integration of the renormalization
group equations allows us to compute for instance the density of states of the delocalization
problem of [7].
Our results reveal some interesting features of the currentcurrent perturbations of Lie
superalgebras in comparison with the analogous perturbation of the ordinary bosonic
Lie algebras. 5 Our models are integrable quantum field theories for the usual reasons
(Lax pair, etc.), and are not conformal. However, the structure of the theory allows a
solution which doesnt rely on exact S-matrices, form-factors, etc. Regarding the difficulty
of solution they lie somewhere between the conformal field theories and the massive
integrable theories.
Delocalization transitions of sublattice (or random flux) models, studied first by
Hikami et al. [10], and by Gade and Wegner [11,12] perturbatively using replica sigma
models, have been known for some time to exhibit R.G. -functions of the kind we
obtained for the currentcurrent perturbations of gl(N|N) supercurrent algebras. These
delocalization transitions possess a special particlehole symmetry, known to prevent
localization due to disorder even in one dimension [1317]. This is not an accident:
in the last section of this paper we exhibit a random Dirac Hamiltonian giving rise
to these theories (in 2D). From this Dirac Hamiltonian we derive a SUSY sigma
3 We point out that for this quantum Hall setting, based on the -functions computed in [1], the line g + g =
M
V
0 which is the subject of this paper, is a line of fixed points in the ultraviolet. One can check numerically that
starting from gM , gV , gA all positive, one can actually flow to the line gM + gV = 0 in the UV.
4 We were first made aware of the relationship between the model of Hatsugai et al. and the random XY model
by M.P.A. Fisher [6] in the context of the replica field theory.
5 The currentcurrent perturbation of su(2) at level 1 gives the GrossNeveu model, which is a massive theory
with factorizable S-matrix. See, for instance, [9].
478
model, which we find to live on a target manifold which is an analytic continuation

of GL(N|N; C)/U (N|N). We solve this sigma model exactly for the case N = 1, in
the presence of a topological WessZuminoWitten (WZW) term with coupling k. We
find that the sigma model on the analytically continued target GL(1|1; C)/U (1|1) with
WZW term is the same theory as the currentcurrent perturbation of gl(1|1) supercurrent
algebra.
An interesting and special feature of our currentcurrent perturbations is that (for any
value of N ) the non-vanishing -function is associated only with two of the generators
of the gl(N|N) algebra. These can be separated out and a scale invariant fixed point
theory results for any value of the disorder strength g, i.e., a line of fixed points (for
any value of N ). After we obtained the current algebra results in Section 4, the papers
[18,19] appeared on PSL(N|N) sigma models, exhibiting a line of fixed points as the
sigma model coupling constant is varied. We have added a paragraph to Section 4.3
explaining the connection of our results with these papers. Our bosonized formulation
may provide a useful way of studying the PSL(N|N) sigma models. Also, a proposal for a
relationship between the PSL sigma models and the integer quantum Hall plateau transition
has appeared recently [20].
The outline of the paper is as follows. In Section 2 we give a general symmetry-based
overview of the models we discuss in this paper. Sections 3 and 4 address the random
field XY model. In Section 3 we formulate the supersymmetric method and extend it to
the N -species version. In Section 4.1 we compute the non-perturbative -function for
the random field XY model, and the exact correlation functions using current algebra
techniques. In Section 4.2 some of these results are extended to the N -species case,
possessing gl(N|N) symmetry. The scale invariant subsector of this theory is discussed
in more detail in Section 4.3. A generalized random field XY model, whose associated
Dirac Hamiltonian lacks time-reversal symmetry is discussed in Section 4.4. (This Dirac
Hamiltonian underlies the GadeWegner sublattice models discussed later in Section 4.)
In Section 5 we turn to delocalization transition of Hatsugai et al. [7] and compute the
density of states as a function of energy. In Section 6.1 we discuss the corresponding
delocalization transition with broken time reversal symmetry, which is in the Gade
Wegner universality class [11,12]. We find that it is described by a sigma model on a
target space which is an analytic continuation of GL(1|1; C)/U (1|1), and we solve this
sigma model exactly. In Section 6.2 we relate the sigma model to the currentcurrent
perturbation discussed earlier (in Section 4.1). In Section 6.3 we comment on the N -species
generalization, the sigma model on analytically continued GL(N|N; C)/U (N|N), and its
scale invariant PSL(N|N) variant. A number of technical details are delegated to three
Appendices. In Appendix A we relate the N = 1 species theory discussed in Section 4.1,
to the random field XY model. Appendix B gives the steps leading from the general
non-hermitian Dirac Hamiltonian and its hermitianization to a corresponding conformal
field theory action. Finally, Appendix C contains details of the 1-loop RG equations of
Section 4.4.
479
2. The models and their symmetries: an overview

The quantum mechanics of non-interacting particles (such as electrons) subject to
random potentials and their possible universality classes of localization/delocalization
transitions are conventionally classified according to those symmetries which are present
for a fixed realization of disorder. Conventionally, only three such symmetry classes were
known. Recently many new symmetry classes for localization transitions were identified
by Zirnbauer [21], and Altland and Zirnbauer [22]. In this section we describe the class
of models discussed in this paper, all of which exhibit particlehole symmetry. These are
called chiral in Refs. [21,22].
2.1. Time reversal symmetric models
The fundamental model that we study in this paper can be formulated as the random field
XY model [2]. This is a 2D statistical mechanics model. At the free fermion point of the XY
model, however, the XY spin operator can be represented, using abelian bosonization, by
a bilinear L R of Dirac fermions governed by the action of Eq. (1.1). Alternatively, this
action can also be related to the 2-component random Dirac Hamiltonian H2 of Eq. (1.4)
in two spatial dimensions, containing a random real Dirac mass M, a random imaginary
scalar potential V and purely imaginary gauge potential AE term. (Details of this connection
are lain out in Appendix A.) The latter being a non-interacting system, this allows us to
use the supersymmetry (SUSY) method for disorder averaging [23,24]. This is the method
we use in this paper to study the random XY models. An apparent difficulty in applying
this method is the non-hermiticity [2527] of this quantum mechanical Hamiltonian H2 ,
which seems to prevent the functional integral over the bosonic fields, appearing in the
SUSY approach to disorder, from converging. This problem is however easily remedied by
considering an associated hermitian 4-component Hamiltonian (Appendix A)

0 H2
.
(2.1)
H4
H2 0
As discussed by HWK [7], the Hamiltonian H4 itself describes the quantum mechanics (in
the continuum limit) of fermions hopping, with real nearest neighbor hopping amplitudes,
on a 2D square lattice with -flux through each plaquette (see also [28]). This theory
possesses a delocalization transition at zero energy, the characteristic symmetries 6
of which are particlehole (since only nearest neighbor hopping amplitudes are nonvanishing) and time-reversal (reality of hopping amplitudes) symmetries. Hence there are
two apparently unrelated problems linked to this 4-component Hamiltonian, the random
field XY model, and the lattice fermion hopping model. In the latter model, a small
imaginary part has to be added to define the quantum mechanical Greens functions.
This is not necessary for the interpretation as a random XY statistical mechanics model.
As discussed in more detail below (and in Appendix A), the theory describing averaged
quantities of the random XY model has gl(1|1) global SUSY, and gl(N|N) SUSY
6 For the action of these symmetries in the continuum theory with Hamiltonian H , see Eqs. (B.2), (B.5).
4
480
when higher moment averages (e.g., of EdwardsAnderson order parameter type) are
considered.
2.2. Models with broken time-reversal symmetry
In the random field XY model discussed above, the complex XY spin, represented by
L R is coupled to a random ordering field, m(x) = m ei , with a random phase . The
corresponding random 2D Dirac Hamiltonian H4 (of Eqs. (1.4), (2.1)) is time reversal
invariant. Time-reversal symmetry breaking may for example be introduced by allowing,
in the language of the Dirac Hamiltonian H2 , in addition to the purely imaginary gauge
potential discussed above, also a real gauge potential. 7 The addition of a real random
gauge potential leads to the generation of both, real and imaginary parts of all three types
of potentials, and the theory flows off to strong coupling. 8 It is then more useful to
employ a description in terms of a non-linear sigma model (Section 6). The corresponding
quantum mechanical Hamiltonian H4 , possessing particlehole symmetry but not timereversal symmetry, is now, by universality, in the class of the GadeWegner sublattice
models [11,12,29] describing fermions hopping on the square lattice with complex nearest
neighbor hopping amplitudes. The corresponding generalized random XY model has
GL(1|1; C) global SUSY. In Section 4 we describe the corresponding GadeWegner type
delocalization transition in terms of a non-linear sigma model on a manifold which is
an analytic continuation of GL(1|1; C)/U (1|1) (we also allow for a topological Wess
Zumino term.)
It is noteworthy to re-emphasize that both models, with or without time reversal
symmetry, possess particlehole symmetry at zero energy, a feature which can circumvent
localization of all states due to disorder, at that energy, even in one dimension [1317].
3. Supersymmetric disorder averaging

In this section and the following Sections 4.1, 4.2, 4.3 we discuss the random field XY
model, described by the Hamiltonian of Eq. (1.4), which we can solve exactly.
To implement the supersymmetric method for disorder averaging [23,24], we augment
, L,R , coupled to the disorder in the same way

the theory with bosonic ghosts [30], L,R
f
as the fermions, with action S0b = S0 ( ). The partition function of the fermions
plus ghosts is now independent of the disordered potentials 9 so the order of functional
integration over the matter-ghosts and disorder can be interchanged. Integrating over
disorder, this yields an effective action for the fermions and ghosts:
7 Properties of the most general Dirac Hamiltonian where random Dirac mass, scalar and gauge potentials have
both real and imaginary parts, as well as the relationship with a corresponding SUSY field theory, are summarized
in Appendix B.
8 See Eq. (4.84) of Section 4.4, with the notations of Appendix B.
9 There is a subtlety arising from the convergence of the bosonic functional integral. A careful discussion of
this issue, done in Appendix A, results in the effective action described below.
Z
Seff =
481
i

d2 x h
L z L + R z R + L z L + R z R + gOm + gA OA , (3.1)
2
where

Om = R L + R L L R + L R ,

OA = L L + L L R R + R R .
(3.2)
In the sequel we will need to introduce a multi-species version of the above model.
Introducing a species index La , a = 1, . . . , N , we define the N -species model as
f
S0 =
N
d2 x X h
(z + Az )Ra
La (z + Az )La + Ra
2
i
a=1
+ m(x)Ra
La + m (x)La
Ra .
(3.3)
Introducing bosonic ghosts and integrating over disorder leads to the effective action:
Z 2 X

d x
(N)
z La + Ra
z Ra + La
z La + Ra
z Ra
La
Seff =
2
a=1

+ gOm + gA OA
(3.4)
where now
Om =
X
a,b
OA =

La + Ra
La Lb
Rb + Lb
Rb ,
Ra
(3.5)

La + La
La Rb
Rb + Rb
Rb .
La
(3.6)
a,b
4. Random XY models via gl(N |N ) current algebras

4.1. Random field XY model: N = 1 species
In this section we consider the N = 1 model in detail. The bosonic ghosts L (R ) have
Lorentz spin 1/2 (1/2). The conformal field theory for such a bosonic first order action
was treated in Ref. [30]. The Virasoro central charge for one copy of the , system is
c = 1, so that the total central charge of the matter plus ghosts is zero. This had to be the
case since the partition function is one.
In the effective action Eq. (3.1) for N = 1 species the operators Om and OA are bilinears
in the dimension 1 currents:
G+ = L L ,
G = L L ,
J = :L L :
J 0 = :L L :
(4.1)
G+ = R R ,
G = R R ,
J = :R R :
J0 = :R R :
(4.2)
(We denote all leftright currents as (JL , JR ) = (J, J).) Namely,
482
Om = J J J 0 J0 G G+ + G+ G ,
OA = (J0 J)(J 0 J ).
(4.3)
In the conformal limit g = gA = 0, the currents generate a gl(1|1) current algebra. The
currents J, J 0 are bosonic U (1) currents for the fermions and ghosts, respectively, and G
are their fermionic superpartners. From the operator product expansions (OPE)
1
,
zw
1
L (z)L (w) = L (z)L (w)
zw
L (z)L (w) = L (z)L (w)
(4.4)
one obtains the OPEs:

1
1
,
J 0 (z)J 0 (w)
,
(z w)2
(z w)2
1
1
G (w),
G (w),
J 0 (z)G (w)
J (z)G (w)
zw
zw

1
1
0
J
(w)
J
+
(w)
.
(4.5)
G (z)G+ (w)
(z w)2 z w
H dz
J (z), etc. These
Let j, j 0 , g denote the zero modes of the currents, defined as j = 2i
zero modes are generators for the global gl(1|1) algebra 10 :
J (z)J (w)
[j, g ] = [j 0 , g ] = g ,
{g+ , g } = j j 0 ,
2
g
= 0.
(4.6)
The algebra gl(1|1) has two quadratic Casimirs:

C = j 2 j 0 2 g g+ + g+ g ,
e = (j j 0 )2 .
C
(4.7)
The operators Om and OA are leftright currentcurrent perturbations with precisely the
e respectively. This implies that the gl(1|1)L gl(1|1)R
structure of the Casimirs C, C,
symmetry of the conformal field theory is broken to a diagonal gl(1|1) symmetry in Seff .
The model defined by Seff is a perturbation of a c = 0 conformal field theory, and can be
studied in the framework of Zamolodchikov [31]. The gl(1|1) conformal field theory was
studied in [32]. Currentcurrent perturbations are generically integrable in the case of the
bosonic Lie algebras, which means there are an infinite number of conserved currents. It is
likely that this feature also holds for the currentcurrent perturbations of the superalgebras.
However we do not pursue this here since, as we now show, the model can be solved by
straightforward manipulations of the functional integral.
The manner in which conformal symmetry is broken in Seff is contained in the functions. The OPEs of the perturbing operators are the following:
2
OA (0),
zz
OA (z, z )OA (0) 0.
Om (z, z )Om (0)
OA (z, z )Om (0) 0,
10 For the complete set of modes the current algebra (4.6) defines the affine Lie superalgebra gl(1|1).
b
(4.8)
483
Note that because of the supersymmetry, the 1/(zz)2 terms vanish. One obtains to lowest
order [3335]
dgA
= 2g 2 + ,
g = 0,
(4.9)
dl
where l is the log of the rescaling factor. The ultraviolet limit corresponds to l
where gA , whereas the infrared limit is l with gA +. This shows that if
gA disorder were not included from the start it would be generated under renormalization.
Since the operator Om is never generated in the OPE at higher orders, it follows that g = 0
to all orders.
It will be useful to bosonize the theory. The U (1) fermion current is bosonized with a
scalar field :
A (g) =
J = iz .
J = iz ,
(4.10)
In the conformal limit, (z, z ) = L (z) + R (z) and

L = exp(iL ),
L = exp(iL ),
R = exp(iR ),
R = exp(iR ).
(4.11)
Bosonization of the system was described in [30]. The J 0 currents are expressed
in terms of another scalar field 0 :
J 0 = iz 0 ,
J0 = iz 0 .
(4.12)
The scalar field 0 has the opposite sign kinetic term 11 compared to , which implies:
(z, z )(0) log zz,
0 (z, z ) 0 (0) + log zz.
(4.13)
Since the 0 field still contributes c = 1, in order to obtain the c = 1 of the system
one needs an additional fermionic system with c = 2. The result is
0
L = eiL z L ,
L = eiL L ,
R = eiR z R ,
R = eiR R ,
where as before 0 = L0 + R0 .
The system has the first order action
Z 2
d x
(L z L + R z R ).
S =
2
(4.14)
(4.15)
In order to have a more symmetric effective action it is convenient to formulate the

system in terms of a complex dimension zero fermion , with the action [36]
Z 2

d x
z z + z z .
(4.16)
S =
4
11 This is just a matter of notation. Since the kinetic term must be positive, the scalar field 0 is implicitly
considered to be analytically continued, so that 0 = i 0 is purely imaginary with 0 real.
484
The equation of motion z z = 0 implies that separates into left and right moving parts:
= L (z) + R (z). The fields are related to as follows:
z L = iz ,
z R = iz ,
L = iz ,
R = iz .
(4.17)
In order to absorb the J J + J 0 J0 terms in Om into the kinetic energy, we define

1
1
.
(4.18)
= ( 0 ),
1 + 2g
Putting all of this together one obtains the effective action 12
Z 2 h

d x
z + z 2 2gA z z + 1 + 2g ei z z
Seff =
4
i

+ 1 + 2g ei z z .
(4.19)
In order to compute correlation functions, we treat the effective theory in a number of

steps. First note that

(4.20)
Seff [+ , , gA ] = Seff + 2 2 gA , , gA = 0 .
Let (+ , , ) be a composite field or product of such fields at different spacetime
points. Making a change of variables + + + 2 2gA in the functional integral, one
relates correlation functions with gA 6= 0 to those in the gA = 0 theory:

(4.21)
(+ , , ) g = + + 2 2 gA , , g =0 .
A
Henceforth Seff denotes the effective action (4.20) with gA = 0.

The next step recognizes that + acts as a Lagrange multiplier. Now let ( , ) denote
a field or product of fields that is independent of + . Introducing a source for + then
functionally integrating over + gives a functional function: one has
Z

R d2 x +
4
( , ) g =0 = D[ , ](z z ) eSeff [+ =0] ( , )
e
A
Z

( = ,
), (4.22)
= D , eSeff [+ =0, =]
where is the solution to
z z = .
(4.23)
(We have set the partition function to 1.)

Insertions of the operator z z + in a correlation function can be obtained from
functional derivatives with respect to .
For instance,

z z + (z, z )( , ) g =0
A

Z
= 4
( = ,
)
.
(4.24)
D eSeff [+ =0, =]
(z,
z )
=0
12 Note that the free bosonic part of the action is positive due to the implicit analytic continuation of 0 , so that
z + z = z z z 0 z 0 = z z + z 0 z 0 , where 0 is real.
One needs

= 2ig z z z z
4 Seff + = 0, = =0

and

Seff + = = 0 = (1 + 2g)S .
485
(4.25)
(4.26)
We define a rescaled field,

=
(4.27)
such that
S = (1 + 2g)S =

d2 x
z z + z z .
4
(4.28)
The above formulas imply that all correlation functions are reduced to free-field
correlators in the rescaled system defined by S . For instance:

z z + (z, z )( , ) g =0 = 2 ig z z z z ( = 0, ) (4.29)
A

( , )
.
(4.30)
4
(z, z ) =0
All correlation functions h i on the RHS of the above equation are computed with the free
action (4.16). In particular

= (z, z ) (0) = log(zz).

(4.31)
(z, z )(0)
The second term in Eq. (4.29) generally involves -functions. More generally the insertion of z z + at different spacetime points is equivalent to the insertion of the operator
on the RHS of Eq. (4.25) at the same points in the free theory up to -functions:

(z z + )n ( , ) g =0
A
n

= (2 ig ) z z z z ( = 0, ) + -functions.
(4.32)
The above formulas are sufficient to compute all correlation functions. First, for
correlations not involving + , the field may be set to zero:

(4.33)
( , ) g =0 = ( = 0, ) g =0 .
A
Thus,

(z, z ) (0) = 0.
Equation (4.29) implies:

z z + (z, z ) g
(4.34)
A =0
Thus to all orders in g:

+ (z, z ) (0) g
A =0
= 4 2 (x).
= 2 log(zz).
(4.35)
(4.36)
486
Correlation functions involving multiple + insertions are more interesting. From (4.32)
one finds

1
= 8g 2 6
+ -functions.
(4.37)
z z + (z, z )w w + (w, w)
(z w)2 (z w)
2
Integrating this, one finds a log2 contribution:

+ (z, z )+ (0) g =0 = 4g 2 6 log2 zz/a 2 ,

A
(4.38)
where a is an ultraviolet cutoff scale.

Finally let us compute correlation functions of the matter U (1) field = (+ +
)/2 with gA 6= 0. Using Eq. (4.21) one obtains

(4.39)
(z, z )(0) = 2 1 + 2 2gA log zz/a 2 g 2 8 log2 zz/a 2 .
Similarly,

0
(z, z ) 0 (0) = 2 1 2 2gA log zz/a 2 g 2 8 log2 zz/a 2 .
(4.40)
The -function A (g) to all orders in g can be computed from (4.39) along with the
renormalization group equation

+ A (g)
(r)(0) = 0.
(4.41)
log a
gA
One finds
A (g) = 2g 2 4 =
2g 2
,
(1 + 2g)2
g = 0.
(4.42)
Note that this agrees with the lowest order result (4.9).
We consider now correlation functions of exponentials of and 0 . The free-field
property Eq. (4.32) implies that correlation functions of exponentials can be computed
using Wicks theorem in the free theory S . One has
Y
Y

eii (zi ,zi ) =

exp i j (zi , z i )(zj , z j )
(4.43)
i
i<j
|zi zj |2i j
2 (1+2 2 g
A)
i<j

exp i j g 2 8 log2 |zi zj |2 .
(4.44)
From this one deduces how the disorder affects the dimension of the exponential fields:

0
dim ei = 2 2 1 2 2gA .
(4.45)
dim ei = 2 2 1 + 2 2 gA ,
4.2. Random field XY model: N -species
We now extend some of the results of the last section to N -species. The effective action
Eq. (3.4) has a gl(N|N) Lie superalgebra symmetry. To see this, define the left-moving
currents
Jab = Lb La
,
La ,
Gab = Lb
0
Jab
= Lb La
,
Gab
487
(4.46)
= Lb La
,
, etc. In the conformal field

and similarly for the right-moving currents Jab = Rb Ra
theory with g = gA = 0, one has a gl(N|N) current algebra with the operator product
expansions

kad bc
1
ad Jbc (w) bc Jad (w) ,
+
2
(z w)
zw

k
1
ad bc
0
0
0
0
ad Jbc
(z)Jcd
(w)
(w) bc Jad
(w) ,
Jab
2
zw
(z w)
ac
bd
0
Gbd (w),
Gca (w),
Jab
(z)Gcd (w)
Jab (z)Gcd (w)
zw
zw
bc
ad
0
Gad (w),
G (w),
Jab
(z)Gcd (w)
Jab (z)Gcd (w)
zw
z w cb

kac bd
1
0
ac Jbd
+
(w) bd Jca (w) ,
Gab (z)Gcd (w)
2
zw
(z w)
where the level of the current algebra is k = 1.
The perturbing fields are currentcurrent perturbations which preserve the
diagonal (leftright) gl(N|N) symmetry:
X
0 0
Jab + Gab Gab Gba Gba ,
Jba Jab Jba
Om =
Jab (z)Jcd (w)
a,b
OA =
0

0
Jaa Jbb
Jbb .
Jaa
(4.47)
(4.48)
(4.49)
(4.50)
(4.51)
global
(4.52)
a,b
The structure of the perturbative expansion in g, gA about the conformal gl(N|N)

current algebra closely parallels the 1-species case. In particular, using the OPEs (4.51),
one can verify that the OPEs (4.8) still hold. In particular, the N dependence cancels. Since
the ultraviolet divergences arise from the singular terms in the OPE Eq. (4.8), this implies
that the -functions are independent of N and are the same as in Eq. (4.9). Thus the N species theory has the same -function as the 1-species theory.
Since the perturbative expansion can be derived from (4.8), we expect that correlation
functions in the gl(N|N) theory have the same structure as we found for gl(1|1). However,
since the functional integral manipulations we used to solve the bosonized gl(1|1) theory
are not straightforwardly extended to the gl(N|N) case, here we obtain expressions for the
correlation functions by using properties of the current algebra. This can be viewed as an
alternative method to the functional integral methods described in Section 4.1, and in fact
explain the results we obtained there in a more covariant fashion.
0 , G , G }. The perturbing
Let J denote an arbitrary left current, J {Jab , Jab
ab
ab
operators Om , OA are gl(N|N) invariant and can be expressed as
Om = C J J ,
e J J ,
OA = C
(4.53)
e define two independent Casimirs. In the abelian sub-vector space

where C and C
0 , . . . , J 0 ), one has
spanned by (J11 , J22 , . . . , JNN , J11
NN
488

C =
1N
0
0
1N

,

e
=
C

{1} {1}
,
{1} {1}
(4.54)
where 1N is the N N identity matrix, and {1} is the N N matrix with 1s in every
e is not), C defines a metric to be utilized to raise
entry. Since C is invertible (whereas C
e 0 0 , where C = C . In the subspace
e
and lower indices. We will need C = C 0 C 0 C
defined above,

e = {1} {1} .
(4.55)
C
{1} {1}
The OPEs (4.51) can be expressed as
C
1
f J (w),
+
J (z)J (w)
2
z w
(z w)
(4.56)
where f are structure constants.

The U (1) currents can be bosonized as in (4.10), (4.12):
Jaa = iz a ,
Jaa = iz a ,
0
Jaa
= iz a0 ,
0
Jaa
= iz a0 .
(4.57)
By bosonizing the fields as in (4.11), (4.14), one can verify that the shift property (4.20)
continues to hold:

(4.58)
Seff a , a0 , gA = Seff a 2 gA , a0 2 gA , gA = 0 ,
where

a a0 ,
(4.59)
and is defined in (4.18). This implies

O a , a0 g = O a + 2 gA , a0 + 2 gA g
A
A =0
(4.60)
where O is any operator.

The field has very simple correlation functions because it doesnt appear in any
exponential interactions. Collecting the kinetic terms for the scalar fields coming from
0 J0 )
the conformal field theory and from the terms g(Jaa Jaa Jaa
aa
Z 2 X
1
d x
z a+ z a ,
(4.61)
Skinetic = 2
4 a
where
a = a a0 .
(4.62)
Since doesnt couple to any of the remaining interaction terms, one has the exact 2-point
function

+
(4.63)
a (z, z )(0) = 2 2 log zz.
The above properties, together with current conservation provide strong constraints on
the 2-point functions of the currents. Euclidean rotational invariance, and the fact that
J , J are dimension 1 currents constrains the hJ Ji 2-point function to be of the form:

1
e F
e(g, zz2 ) ,
C F (g, zz2 ) + C
(4.64)
J (z, z )J (0) g =0 =
A
zz
489
e are scaling functions of g and

where we have used the unbroken gl(N|N) symmetry. F, F
2
the dimensionless combination zz , where is the energy scale that enters through the
renormalization group.
The field has the covariant description:
e J ,
iz = C
e J ,
iz = C
(4.65)
e0 C 6=
e0 hJ J i = 0. Since C
From (4.63) we see that away from zz = 0, one must have C
e
e
0, and C 0 C = 0, one deduces that F = 0.
The property (4.63), together with (4.60), leads to gA independence (up to -functions)
of the correlation function (4.64). Since the dependence on the scale must be compatible
e is only a function of the
with the -function for gA , the gA independence implies that F
dimensionless coupling g. Thus

J (z, z )J (0) g
A =0
1e e
C F (g).
zz
Current conservation, z J + z J = 0, requires

z J (z, z )J (0) = z J (z, z )J (0) .
(4.66)
(4.67)
Integrating this,

J (z, z )J (0) g
A =0

1 2
e(g) log zz .
e F
C + C
2
z
(4.68)
The coefficient 2 of the C term was derived from (4.63).

e(g), we use current algebra equations of motion. The
To fix the one unknown function F
equations of motion to first order in g can be obtained in conformal perturbation theory
I
dw
J (w, z )Om (z, z ).
(4.69)
z J (z, z ) = g
2i
z
Taking into account the -rescaling / that gives the kinetic term the standard
normalization, and using the OPEs (4.56), one obtains
z J = g 2 f J J = z J .
(4.70)
Using this in the correlation function hz J z J i, along with (4.64),

1 e e
0
0
J 0 J (z, z ) J J 0 (0) g =0 .
C F = g 2 4 f f
A
z2 z 2
(4.71)
e comes from
The non-zero contribution to F

J 0 J (z, z ) J J 0 (0) g
=0
4
C 0 C 0 ,
z2 z 2
(4.72)
where we have used Eq. (4.68). Now, in terms of the structure constants, the OPE
Om (z, z )Om (0) 2OA /zz implies
0
e = f f
2C
C 0 C 0 .
(4.73)
490
The last three equations thus give

e(g) = 2g 2 8 .
F
(4.74)
Let us summarize the above results by integrating the current correlations to find the
2-point functions of the scalar fields:

+
(4.75)
a (z, z )b+ (0) g =0 4g 2 8 log2 zz, a, b,
A

+
(4.76)
a (z, z )b (0) g =0 2 2ab log zz,
A

(4.77)
a (z, z )b (0) g =0 0, a, b.
A
Finally, using the shift property (4.60),

a (z, z )b (0) = 2 ab + 2 2gA log zz g 2 8 log2 zz,

0
a (z, z )b0 (0) = 2 ab 2 2 gA log zz g 2 8 log2 zz.
(4.78)
(4.79)
4.3. Nearly conformal structure

We have seen in the previous section that the N -species theory (describing N copies
of the random field XY model) is nearly conformal. This can be seen more clearly by
separating out the non-conformal pieces (coupling holomorphic and anti-holomorphic
coordinate dependences) by rewriting Eqs. (4.66), (4.68) as follows. Instead of the basis
of currents as in Eq. (4.47), we extract U (1) U (1) from the bosonic U (N) U (N)
subgroup. Explicitly, consider traceless generator matrices TA of su(N), (A = 1, . . . ,
N 2 1)
JA (TA )ab Jab ,
and
J
Jaa ,
0
JA0 (TA )ab Jab
,
J0
0
Jaa
and similar for the fermionic currents (which we denote by JA , A = 2N 2 + 1, . . . , 4N 2 ).

Forming the combinations
J J J 0 ,
J+ J + J 0 ,
and noting that all non-conformal terms in the currentcurrent correlators are multiplied by
eAB whose only non-vanishing matrix elements are A, B = in this basis,
the invariant C
we see that the two point correlation functions of all the remaining currents are
2 CAB

,
J
(z,
z
)J
(0)
0 (A, B 6= ).
JA (z, z )JB (0) =
A
z2
Note that we see from Eq. (4.72) that these correlators are invariant under independent
action of the symmetry group on the left and right chiral components of the currents.
The ability to separate off the J currents (and the corresponding fields, ) can be seen
from the Hamiltonian, expressed as
H = H0 + HI ,
491
where H0 is the non-interacting gl(N|N)-Sugawara Hamiltonian, and HI is the current

current interaction. Recalling that CAB = C AB there are no terms in the Hamiltonian H
which couple the generators J , J to the remaining ones, which allows to consistently set
J = J = 0, yielding a scale invariant theory.
Setting J = J = 0 effectively sets gA = 0. Since the non-zero -function is gA (g),
what is left is a conformal field theory since g = 0. In terms of groups, setting J =
0 is equivalent to dividing by the U (1) U (1) subgroup: GL(N|N)/U (1) U (1) =
PSL(N|N). Since the two U (1) bosons give c = 2, and the total central charge for
GL(N|N) is zero, what is left is a c = 2 conformal field theory. The conformal
PSL(N|N) sigma model was recently studied in [18,19] (see also Section 6 below).
In the N = 1-species case the emergence of conformal symmetry is easily understood:
Bosonizing the non-interacting theory, there are only two generators G+ = i , G =
i left once the currents J and thus the fields have been eliminated. The resulting
PSL(1|1) theory is free. (See Section 6 for a relationship with the PSL(1|1) sigma model.)
One proceeds similarly for the N > 1 cases. The N = 2 species case for example
describes the second moments of observables in the random XY model. An interesting
observable is the EdwardsAnderson order parameter, which is a bilinear of one of the
currents in the conformally invariant sector:

R1 L2 R2
= J12 J21 .
q12 L1
We see from Eq. (4.72) that the two-point of this field decays algebraically,

4
q12 (z, z )q21 (0) 2 2 .
z z
It was recently proposed by Zirnbauer that the PSL(N|N) sigma models describe the
integer quantum Hall transition [20]. The kind of disorder considered in the present paper,
which, as we have explained, leads to the PSL(N|N) sigma models, is not generally
believed to correspond to kind of disorder needed for the quantum Hall effect, at least not in
an obvious way. Nevertheless this is an interesting proposal that needs further investigation.
4.4. Generalized random XY model (broken time-reversal symmetry)

In this subsection we discuss a more general random XY model (at the free fermion
point). This model may be defined in terms of an associated 2-component Dirac
Hamiltonian of type H2 , where now all random potentials have both real (Ax , Ay , V , M)
and imaginary (A0x , A0y , V 0 , M 0 ) parts. As discussed in Appendix B, in this case the
corresponding 4-component Hamiltonian H4 lacks in general time reversal symmetry (but
still has particlehole symmetry), and the corresponding field theory action possesses
GL(1|1; C) global SUSY. 13 This subsection is devoted to the 1-loop RG equations of
this theory. To summarize the result, we find that the theory flows off to strong coupling.
The strong coupling physics should be captured by a non-linear sigma model of the kind
discussed in Section 6.
13 GL(N |N ; C) symmetry for the N th moment averages.
492
Performing the disorder average over independent random variables m = V + iM and

= i(V 0 + iM 0 ) with variance gm = mm and g = respectively 14 , we arrive at the
following Lagrangian for the averaged theory
X

g
0 0
Jaa + Gaa Gaa Gaa Gaa
Jaa Jaa Jaa
L susy = g1
a=1,2
+ g2
a6=b
X

0 0
Jab + Gab Gab Gba Gba ,
Jba Jab Jba
(4.80)
a,b=1,2
where
g1 gm g ,
g2 gm + g .
(4.81)
The additional contribution from the random vector potentials is

X
0

A
0
Jaa Jaa
Jaa
Jaa
L A
susy = g1 (1)
a=1,2
+ g2A (1)
a6=b
X
0

0
Jaa Jbb
Jbb ,
Jaa
(4.82)
a=1,2
where
g1A = gA gA0 ,
g2A = gA + gA0 .
(4.83)
[gA and gA0 are the variances of the imaginary and the real vector potential, respectively.]
We recover the time-reversal symmetric case, Eq. (4.52), by letting g = gA0 = 0. We
note that in the special case where only the potentials Ax , Ay , V 0 , M 0 are non-vanishing,
which corresponds to gm = 0, g > 0, we recover time reversal symmetry [Eq. (B.6) of
Appendix B]. The RG equations for this case are identical to those of the case gm >
0, g = 0, studied in the the previous section, upon letting gm g . This can be seen
by replacing in Eqs. (B.8), (B.9) the left moving fields L1 by iL1 , L2 by iL2 (and
by iL1
, and L2
by iL2
) and similarly for the -fields. All right moving fields
L1
remain unchanged. With these redefinitions the coupling constants in Eq. (4.80) become
g1 = g2 = g . On the other hand, as these redefinitions do not change the kinetic term
in Eq. (B.11) of Appendix B, the OPEs of these fields remain unchanged. This implies
that the OPEs of the gl(2|2) currents in Eq. (4.51) remain also unchanged, and the entire
analysis of Sections 4.1, 4.2 remains unchanged upon replacing gm by g .
Let us now return to the most general case. The 1-loop RG equations for the coupling
constants of the action above are derived in Appendix C (Eqs. (C.2), (C.3)), with the
result 15 :

dg A
dg1
= 0,
= 2 (gm )2 + (g )2 ,
dl
dl
0
dg A
dg2
= DgA0 g2 ,
= 4gm g .
(4.84)
dl
dl
14 Using the action of Eq. (B.11) of Appendix B.
15 In terms of the notations of the appendix, D (d 2 d 2 ).
22
21
493
The coupling g2 flows according to

dgA0
d2 log(g2 )
= D (g2 )2 (g1 )2 ,
=D
2
dl
dl
which means that g2 flows to large values. (Note from Eq. (4.81) that the expression
[(g2 )2 (g1 )2 ] is always non-negative.) The strong coupling physics should be described
via the sigma model technology of Section 6.
5. Density of states of the Hatsugai et al. Delocalization transition from current

algebras
In this section we apply some of the above results to the problem of localization of
electrons hopping with real amplitudes on a square lattice with flux per plaquette [7]. In
the continuum limit the Hamiltonian of this system is the 4-component (hermitian) Dirac
Hamiltonian H4 ,

0 H2
,
(5.1)
H4 =
H2
0
where H2 is defined in Eq. (1.4). The reality of lattice hopping amplitudes implies that the
Hamiltonian H4 must have time-reversal symmetry, which is manifest with this form of
H2 (see Eq. (B.2)). The corresponding SUSY field theory action (Eq. (B.11) of Appendix
B with 0) can be seen to possess GL(2|2; R) symmetry. (This model has also been
studied numerically in [37].) In this subsection we compute the density of states of the
Hamiltonian H4 . In particular, we will be interested in the eigenstates of the Schrdinger
single particle Green
equation H4 = E , where is a 4-component
R wave function. The
f
functions are defined by the functional integral D D exp(S0 ), where
Z 2
d x
f
(x)i(H4 E) (x)
(5.2)
S0 =
2
and E = E + i. For = 0+ , the functional integral defines the retarded Green function:

0

1
x , b = lim i a (x) (x 0 ) . (5.3)
GR (x, x 0 ; E)ab = lim x, a
b

H (E + i)
0+
0+
4
Making the identifications 16 (Appendix B)

t
, R2
, L1 , R1 ,
= R1
, L1
, R2 , L2 ,
= L2
the action takes the form of an E perturbation of the N = 2 species model:
Z 2
d x
(x) i(H4 E) (x)
S=
2
Z 2 X
2

d x
z La + Ra
z Ra + m(x)Ra
La + m (x)La
Ra
La
=
2
a=1
16 The superscript t denotes the transposed vector.
(5.4)
494

+ Az L1
L1 + L2
L2 + Az R1
R1 + R2
R2
Z
R2
L2
R1
+ h.c. .
iE
L1
(5.5)
For simplicity, we can set the physical A-couplings to zero from the beginning and then
let the gA coupling be generated dynamically under renormalization. Introducing ghosts
and integrating over disorder leads to the effective action:
Z 2
d x
(N=2)
OE ,
iE
(5.6)
S = Seff
2
where

R2
L2
R1
+ L1
R2 + L2
R1 + h.c. terms
OE = L1
(5.7)
(see Appendix B).

Of interest is the density of states (DOS)

1
1
1
lim Im Tr
,
(E) = Tr (H4 E) =
V
V 0+
H4 E i
(5.8)
where V is the two-dimensional volume. The averaged density of states can be expressed
in terms of the averaged retarded Green function
(E) =
1
lim Im tr GR (x, x; E),
0+
(5.9)
where tr denotes the trace over matrix indices of the Greens function of Eq. (5.3).
Thus the density of states is related to the one-point function of the, say, fermionic part
of the operator OE which couples to E:
(E) =

lim Re i L1
R2 L2
R1
E,
0+
(5.10)
where the one-point function on the right hand side is computed using Seff with the E term.
Generally, the E-term spoils the complete solvability of the theory. Our strategy will be
to use our solution of the E = 0 theory in conjunction with the renormalization group
R 2
to learn something about the density of states. Let us view SE = iE d2x OE as a
perturbation of the nearly conformal field theory. The anomalous scaling dimensions of
fields are controlled by Seff when E = 0.
We need the anomalous dimension of the field OE which contains

R2
: = : exp(iL1 + iR2 ): = exp i 1 + 2 /2 exp i(1 2 )/2 .
:L1
Here, a = La Ra denotes the dual field, whose 2-point function is found from
Eqs. (4.66), (4.74) to be

+
a (z, z ) b+ (0) g =0 4g 2 8 log2 (zz/a 2 ).
A
The requirement that for arbitrary gA this function is to satisfy Eq. (4.41) gives

+
a (z, z ) b+ (0) g 8 4 gA log zz/a 2 + 4g 2 8 log2 zz/a 2 .
A
495
This together with

+
a (z, z ) b (0) g 2 2 ab log zz/a 2
A
R2
:
yields the anomalous dimension 17 of the field :L1

1
(gA ) = 2 1 4 2gA + h(g)
2
where h(g) is independent of gA . Since the action is dimensionless,
dim(E) = 2 ,
(5.11)
ensuring that (E) dE has dimension 2, as it should in two dimensions.

The renormalization group equation for (E,
gA )

+ A (g)
(gA ) (E,
gA ) = 0
(2 )E
E
gA
(5.12)
implies that the renormalized density of states (l)

(E(l),
gA (l)) (where el > 1 is the

rescaling factor) is related to the unrenormalized DOS (0)
by
)
)
(Z l
( Zl
0

0
0
0
2 (l ) dl 2l .
(0)/
(l)
= exp (l ) dl = exp
0
Here
dE(l)
= [2 ]E,
dl
dgA (l)
= gA ,
dl
which permits to express the l-dependence in terms of the the renormalized coupling
constant ER E(l) and its bare value E E(0) 6 ER . This yields for the DOS
(E)/(ER ) =

ER
exp 2 gA (ER ) gA (E) /gA
E
(5.13)
(valid since gA (g, gA ) is independent of gA ). From this we find in the limit E/ER 0

ER
2(1 + 2g)2 p
exp
log(ER /E) .
(E)/
(E
R)
E
g
This diverging (but integrable) density of states was not seen in the simulations of [37].
Rather, a power law varying with the strength of disorder g was found. Such power law
behavior of the DOS is obtained from the RG analysis above for intermediate, but not
asymptotically small energies.
17 For N > 2 the free field property that leads to exact expressions such Eq. (4.43) is not expected to hold.
However if we assumed that for the purposes of computing the anomalous dimension we can parallel what we
did for N = 1, we would obtain h(g) = 2 /2.
496
6. GadeWegner delocalization transition via sigma model

6.1. Sigma model on analytically continued GL(1|1; C)/U (1|1)
Localization problems are usually described in terms of sigma models, using replicas
[38,39] or supersymmetry [23,24].
In this section we study the GadeWegner localization problem [11,12]. A version of
this model consists of electrons hopping on a 2-dimensional square lattice with arbitrary
complex hopping amplitudes (and flux per plaquette). The continuum limit of the
Hamiltonian of this model is again, as for the model discussed in Section 5, of the
form of a 4-component Dirac Hamiltonian. In contrast to the latter model, however, the
lattice Hamiltonian of the present model, and thus also its continuum limit H4 lacks time
reversal invariance due to the complex hopping amplitudes. As discussed in more detail
in Appendix B, the associated Dirac Hamiltonian H4 has now arbitrary complex random
scalar potential (with real and imaginary parts V 0 and V ) and complex Dirac mass (with
real and imaginary parts M and M 0 ) terms, which may be combined into two independent
complex random variables m and , with variance gm and g respectively. The explicit
form of the Hamiltonian is (from Eq. (B.7) of Appendix B):
0
m
m +
,
(6.1)
H4 = (i)
m +
m
0
where
m (V + iM),
i(V 0 + iM 0 ).
In general, an arbitrary complex vector potential will be generated upon RG flow in the
corresponding SUSY field theory. 18 We set the vector potentials to zero initially, and let
them be generated upon renormalization group transformations later on. In order to study
this system we use the conventional sigma-model technique.
The Lagrangian is
L = i(H4 E) + i(H4 E)
where E = E + i ( = 0+ ) and

1
1
2
2
=
=
3 .
3 ,
4
4
Let us define new fields

3
1

1
e
= + ,
e = 3 = + ,
=
2
4
e
4
2
18 See Eq. (4.84).
(6.2)
(6.3)
(6.4)

= 1 , 1 , 4 , 4 = + , ,

e .
e+ ,
e = 3 , 3 , 2 , 2 =
497
(6.5)
(6.6)
Using Eq. (B.10) of Appendix B the Lagrangian can be expressed as
e D + mi (x)
e
e e
e
e D
L=
i i mi (x)i i iE + ,
(6.7)
m (m ),
(6.8)
where a sum over the repeated index i = 1 is implied and
0 0 z 0
0 0 0 z
D = (+ 1)z + ( 1)z =
z 0 0 0 .
0 z 0 0
When E = 0, the Lagrangian has the following symmetry:

1
0
G
0
(G )
e
e,
e
e
,
G =
G =
0
G
0 (G1 )
(6.9)
(6.10)
where G is a 2 2 supermatrix with bosonic (fermionic) diagonal (off-diagonal) elements.
G ( 1)e
G = 1 and
That this is a symmetry follows from G ( 1)G = e
e
G G = 1. Since G is only required to be invertible, G is an element of GL(1|1; C), i.e.,
the complexification of GL(1|1). In the presence of the E term, the above transformation
G e
G = 1, which implies G G = 1. Thus the E
continues to be a symmetry only if G G = e
term breaks the symmetry to U (1|1).
Without loss of generality the random variables (m+ , m+ ) and (m , m ) may be taken
to be statistically independent 19 with the distribution (1.2) each, and the same variance g.
Upon disorder averaging one then obtains the term in the effective Lagrangian:
X
e i .
ei
i
(6.11)
Leff = g
i
i=
Introduce a grade [a] for the vector index a, where [1] = [3] = 1 (fermions), [2] = [4] =
0 (bosons), such that
ea .
ea b = (1)[a][b] b
(6.12)
Then using the fact that (1)[b] +2[a][b] = (1)[b] , one finds

eip
eiq = Str Mi M
ei ,
eiq iq = (1)[q] iq ip
eip
ip
2
(6.13)
ei are the 2 2 matrices of bilinears:

where Str denotes the supertrace and Mi , M
ei =
ei
e ;
M
i
ei )pq =
eip
eiq ,
= ip iq ,
(M
Mi = i i ,
(Mi )pq
p, q = 1, 2.
(6.14)
19 For general values of g and g , the variables m and m will be correlated and will have different
m
variances. Changing the ratio gm /g will however not affect the universality class of the sigma model. Therefore
we may chose this ratio to be unity.
498
e
Let us introduce two 4 4 block diagonal supermatrices of fields Q, Q

e
0
0
Q+
e Q+
,
Q
Q
e ,
0 Q
0 Q
and define
M
M+
0
0
M

,
e
M
e+
M
0
0
e
M

.
Consider the following Lagrangian:

e iE M
e (D + gQ + iE)M g QQ
e .
L = Str D g Q
(6.15)
e fields gives the effective interaction Leff =

A functional integral over the Q , Q
e
e
g Str(M+ M+ ) +g Str(M M ). If instead we first perform the Gaussian functional integral
e fields:
e fields, we obtain the effective action for the Q, Q
over the ,
Z 2

e (6.16)
e iE + STr log(D + gQ + iE) g d x Str QQ.
SQ,Q
e = STr log D g Q
2
In the above equation, STr incorporates the integral d2 x: For a diagonal functional operator
R 2
A(x, y) = A(x)(x y) we define STr A = a12 d2x Str A(x), where a is an ultraviolet
cutoff.
e follow from (6.10). The Lagrangian (6.15) then
The symmetry transformations on M, M
has the following symmetry when E = 0:
ee
ee
M
GM
G,
M GMG ,

1
1
e e
ee
Q
G Q
Q G QG 1 ,
G 1 .
(6.17)
When E 6= 0, the symmetry is as in (6.17) with G an element of U (1|1).

e develop vacuum expectation values
Due to the linear terms in SQ,Q
e , the fields Q, Q
e Let us take the vacuum expectation value (VEV) to be h(Q )ab i = q ab ,
hQi, hQi.
e one finds that for E 0, the
e )ab i = q ab . Minimizing S e with respect to Q, Q,
h(Q
Q,Q
VEVs q = q and q = q must be solutions to a set of self-consistent equations which
have the solution
1/2
1
exp(4/g) 1
(6.18)
q = q =
ag
which for small g goes to zero exponentially as follows:
q = q
1
exp(2/g).
ag
This follows from the saddle point equations

e = x (D + gQ)1 x ,
e 1 x ,
Q
Q = x D gQ
which, when written in momentum space, lead to

1
g q
log 1 +
=q
4
(ag q)
2
(and a similar equation with q q).
(6.19)
499
Since the VEV breaks the GL(1|1; C) symmetry to U (1|1), there are massless Goldstone
modes. These Goldstone modes as usual can be viewed as a symmetry transformation of
the VEV. Thus, let us define the fields

1
qGG
qU
0
0
=
,
U = G hQiG 1 =
0 qU 1
0
q(GG )1

1
qU 1 0
0
1
e1
e= e
e G = q(GG )
Q
=
.
(6.20)
U
G
0
qGG
0
qU
The latter are expressed in terms of U GG . An element G GL(1|1; C) can be
factorized into right U (1|1) cosets as G = Ga Gu where Gu Gu = 1 and Ga = Ga .
One has U = Ga Gu Gu Ga = G2a , thus the field U naively lives on the coset space
GL(1|1; C)/U (1|1).
A parametrization of the Goldstone modes U which manifests the required properties
that U be invertible and U = U is the following:

2
e 1 0
1

e 1 0
U=
,
2
1
0 e2
0 e2

e 1
1
2
e1
0
0
1
.
(6.21)
U =
0
e2
2 1
0
e2
The fields 1,2 are bosonic whereas , are fermionic. We remark, as discussed after Eq.
(6.30), that the field 1 is to be considered analytically continued and purely imaginary, in
order to ensure a positive kinetic term of the sigma model. Therefore the target space of the
sigma model is this analytical continuation of the coset space GL(1|1; C)/U (1|1). After
analytic continuation the bosonic submanifold consists of the set of matrices 20
2i

e 1
0
U=
U (1) GL(1; C)/U (1), 1 , 2 = real.
0
e22
The low energy -model for the Goldstone modes U is obtained by substituting
e and performing a derivative expansion using STr log(A + D) = STr log A +
e U, U
Q, Q
STr log(1 + A1 D). The zeroth order term in E is
2

1
1
1
e1 D U
e1 = q
DU
+
D
U
Str U 1 U.
(6.22)
Str
DU
2a 2g 2
a 2g2
Keeping only the linear term in E, and expressing everything in terms of U , one finally
obtains

Z

1
Str U 1 U
S = d2 x
2
8

2

1 A
1
1
iE
Str
U
+
U
Str
U
,
3
(6.23)
2 2
where for small g

4
1
=
4 exp(4/g) 1,
2 a 2g 2 q 2
20 Compare also Refs. [20,21].
3 =
2
2
exp(2/g).
a 2 gq
a
(6.24)
500
As explained below, the A /2 term is generated under renormalization.

In terms of the parametrization (6.21) we find the following expressions:

1
Str U 1 U = + + + ( )2 ,
4
2
1
Str U 1 U = ( )2 ,
4

Str U + U 1 = 2(1 ) cosh 22 2(1 ) cosh 21 ,
where (2 1 ).
(6.25)
(6.26)
(6.27)
(6.28)
(The effect of a WessZumino topological term is discussed below.)

We may study the sigma model as a perturbed conformal field theory. As an example we
will compute the density of states for the present model. Rescaling the fields 1,2 2 1,2 ,
2 , the sigma model action can be written as

Z 2 2
d x
Og + A O2 iEOE ,
(6.29)
S = SCFT +
2 4
where the conformal field theory has the same field content as our gl(1|1) model of
Section 4.1, with action
Z 2

d x
(6.30)
( 2 )2 ( 1 )2 +
SCFT =
8
and
1
(6.31)
Og = ( )2 ,
4
(6.32)
O2 = ( )2 ,

2
2
(6.33)
OE = 23 1 cosh(2 ) 1 cosh(1 ) .
4
4
We note that as in Section 4.1, the scalar field 1 is implicitly considered to be analytically
continued, i.e., 1 = i 1 is purely imaginary with 1 real. This is necessary in order for the
(bosonic) part of the action SCFT , from the scalar fields, to be positive. As mentioned above,
the target manifold of our sigma model is therefore not literally GL(1|1; C)/U (1|1), but
rather an analytical continuation of this.
Since is a logarithmic operator [40], it mixes under RG scale transformations with
the identity operator, thereby generating the A coupling.
This can also be seen, alternatively, by making the following change of variables
e ,
which yields

1
Str U 1 U = + + e2
4

( ) e2 .
(6.34)
The last term can be eliminated by shifting + ,

+ + + e2 .
(6.35)
501
Let us now also consider a topological WessZuminoWitten term with coupling k. To

lowest (cubic) order this term is easily computed with the result
Z

ik
d3 x ij k Str U 1 i U U 1 j U, U 1 k U
SWZW =
24
M3
Z

ik
d2 x ( ) ( )
=
2
+ quartic and higher terms,
(6.36)
where, as usual, M3 is a three-dimensional manifold whose boundary is the twodimensional space of interest and the two-dimensional antisymmetric tensor.
The sigma model including the WZW term can be solved using the Lagrange multiplier
method of Section 4. In order to see this, note that the fundamental matrix field is of the
form
U = e+ V
(6.37)
where V does not depend on + . This implies that the commutator in Eq. (6.36) does not
contain any + -dependence. Since the supertrace of a commutator vanishes, the WZW
term does not contain any + dependence either. 21 Therefore, as for the currentcurrent
perturbation of Section 4, + acts as a Lagrange multiplier, and we can use the same steps
to solve the present theory.
Making the same rescalings of the fields as above, the action for the sigma model with
WZW term is

Z 2
d x
1
Og k3 OWZW + A O2 iEOE + ,
(6.38)
S = SCFT +
2 2
2
where 22
+ 2 + 1 ,
2 1 ,
(6.39)
and only terms of linear order in have been written. 23 Here SCFT is given as before 24
by Eq. (6.30) upon replacing, , , and

1
(6.40)
Og = ,
2

1
(6.41)
OWZW = i ,
2
(6.42)
O2 = ( )2 ,

2
OE = 23 cosh(2 ) cosh(1 ) e1 + e2 .
(6.43)
4
We proceed with the 1-loop RG equations. One has the OPEs:
21 Note that this argument is equally valid for the N -component generalization, Section 6.3 below.
22 No confusion should arise between the fields ,
1,2 defined here and those of Eqs. (6.2), (6.3) and in
Appendix A, B.
23 As in Section 4, the term linear in will be sufficient in the following.
24 Again, is purely imaginary.

1
502
1
O2 (0) + ,
4zz
+1
O2 (0) + .
OWZW (z, z )OWZW(0)
4zz
Thus, as in Section 4, to lowest order the -functions are
Og (z, z )Og (0)
(6.44)
(6.45)

2
1 (k2 )2 + .
(6.46)
16
The sigma model coupling constant , and the WZW coupling k generate A terms
(denoted gA in Section 4) of opposite signs. The 1-loop RG equations have the same
structure as those found for the gl(1|1) currentcurrent perturbation model in Section 4,
where O2 is analogous to OA . In the presence of a WZW term, the sigma model becomes
the conformal gl(1|1) WZW model at a particular finite value of the sigma model coupling
1/2 = k.
We may solve the theory in Eq. (6.38) exactly using the Lagrange multiplier method of
Section 4.1. In order to compute correlators of + one needs

S

= + (k2 )i
4
(z, z ) =0 2

= 1 (k2 ) z z + 1 + (k2 ) z z .
= 0,
k = 0,
A =
As in Section 4.1 this gives the two point function of + fields:

z z + (z, z )w w + (w, w)
Furthermore, as in Eq. (4.35) we find

+ (z, z ) (0) = 2 log(zz),

2
1
1 (k2 )2
2
(z w)2 (z w)
2
+ -functions.
=
=0

(z, z ) (0) = 0.
As in Section 4.1, this yields the correlation functions 25

2

1 (k2 )2 log2 (zz/a 2 ),
1 (z, z )1 (0) = (1 4A ) log zz/a 2
16

2
1 (k2 )2 log2 (zz/a 2 ),
2 (z, z )2 (0) = (1 + 4A ) log zz/a 2
16
as well as the exact beta function (identical to the 1-loop result):
(6.47)
(6.48)

2
dA
= A =
1 (k2 )2 .
(6.49)
dl
16
We now proceed with the calculation of the density of states. The most relevant operator
in OE is cosh(2 ), whose anomalous dimension is

(6.50)
(A ) = dim cosh(2 ) = 2 (1 + 4A ).
25 The opposite sign of the non-interacting correlator of the bosonic field as compared to that of reflects
1
2
the above-mentioned analytic continuation of the former.
503
The analysis of Section 5 applies to the density of states with A playing the role of gA ,
yielding

ER
exp 2 A (ER ) A (E) /A
(E) (ER ) =
(6.51)
E
(valid since A is independent of A ). In the limit E 0 this gives a divergent (but
integrable) density of states
(
)
p
4 2
1
(E) exp p
log(ER /E)
E
2 1 (k2 )2
in agreement with the perturbative result obtained by Gade [11,12] using replicas when
k = 0, i.e., in the absence of the WZW term.
We end this section by commenting on the connection with the scale invariant PSL(1|1)
sigma model. Eliminating the two bosonic coordinates the fundamental PSL(1|1) sigma
model field becomes

2
(1
)
,
(6.52)
V=
2
(1 + )
whose dynamics is governed by the quadratic action obtained from SCFT of Eq. (6.30) by
eliminating 1 , 2 . In Section 6.3 we briefly discuss the generalization to the less trivial
PSL(N|N) case.
6.2. Sigma model on (analytically continued) GL(1|1)/U (1|1) as a currentcurrent
perturbation
Upon comparing the correlation functions of the sigma model (Eq. (6.47)) with those of
the N = 1-species XY model (Eq. (4.39), (4.40)), which is a currentcurrent perturbation,
one finds that the two models have identical correlation functions if one identifies the
coupling constants as follows:
2A = 2 gA ,

1 2
1 (k2 )2 = (2g)2 6 ,
4
(6.53)
where 2 = 1/(1 + 2g).
(6.54)
6.3. Sigma model on (analytically continued) GL(N|N)/U (N|N)

The generalization of the sigma model of the previous section to GL(N|N) is immediate.
We start with a theory of 2N bosons and 2N fermions (in the previous section N = 1). The
construction of the previous section provides us in the standard way with a field variable
on the coset manifold U GL(N|N; C)/U (N|N). Denoting the elements of the super-Lie
algebra gl(N|N) by J and JA , a parametrization of the fundamental matrix field, say of
the form 26

X

{X+ J+ }
U X , XA e
exp X J +
XA JA
A6=
26 This follows as in the N = 1 case from the decomposition G = G G into U (N |N ) cosets, see below
a u
Eq. (6.20).
504
yields an action generalizing Eq. (6.25). As in the N = 1 case, this manifold has to be
analytically continued, in order to ensure that the kinetic term of the bosonic variables is
positive 27 . Since the beta functions for the sigma model should be independent of N , they
should be the same as for the N = 1 model, Eqs. (6.46), (6.49), for all N . The PSL(N|N)
version is obtained by setting the coordinates X to zero. This sigma model in the presence
of a WZW term with coupling k becomes the gl(N|N) WZW conformal field theory at a
finite value 2 = 1/k of the sigma model coupling constant. The latter is a gl(N|N) current
algebra at level k. One might expect that the relationship noted in Section 6.2, is a special
case of a more general statement. We plan to address this issue in future work.
7. Conclusions
In this paper we have studied currentcurrent perturbations of gl(N|N) super-current
algebras. The existence of two quadratic Casimir invariants leads to a nearly conformal
theory for any value of the coupling. Current algebra techniques allowed us to compute
exactly the correlation functions of all current operators in the perturbed theory, as well as
the exact -functions (in a particular RG scheme). The current correlators are conformal
except for correlation functions involving the currents corresponding to the trace and the
supertrace of the super-Lie algebra, denoted here by J . In the case N = 1 we have
computed the correlation functions of all fields. We have applied this to two different,
but related disordered models, the 2D random field XY statistical mechanics model, and
the delocalization transition of electrons hopping on a 2D square lattice with -flux per
plaquette and real hopping amplitudes (HWK, [7]). Both theories can be formulated in
terms of the 2D Dirac Hamiltonian subject to random real mass, imaginary scalar and
imaginary vector potentials. This Hamiltonian is invariant under two discrete symmetries,
charge-conjugation (particlehole) and time-reversal, in every realization of disorder.
For the random XY model we have computed all correlation functions involving 1stmoment averages (described by the 1-species theory), as well as certain correlation
functions involving N th-moment averages (namely those involving the gl(N|N) currents).
In particular, we have computed the correlation function of the EdwardsAnderson
order parameter, which is scale invariant and has scaling dimension x = 2, in the 2nd
moment (N = 2-species) theory. For the random hopping model of Hatsugai et al. [7]
we have obtained the density of states which shows a divergence at zero energy. In
Section 4.4 we derived the 1-loop RG equations for a generalized random XY model:
When formulated in terms of a random 2D Dirac Hamiltonian, the latter still exhibits
particlehole symmetry, but lacks time-reversal symmetry, as opposed to that of HWK.
The delocalization transition exhibited by this more general random Dirac Hamiltonian is
in the GadeWegner universality class, describing hopping of electrons on a 2D square
lattice with general complex hopping amplitudes. In Section 6 we have derived a SUSY
sigma model from the underlying random Dirac Hamiltonian, found to be defined on
27 In parallel to the N = 1 case, the bosonic manifold becomes therefore U (N ) GL(N ; C)/U (N ) instead of
GL(N ; C)/U (N ) GL(N ; C)/U (N ) before analytic continuation (see also Refs. [20,21]).
505
a target manifold which is an analytic continuation of GL(N|N; C)/U (N|N). We have

solved this sigma model exactly for N = 1 (including also a WZW term), and shown that
it is identical to a currentcurrent perturbation of a gl(1|1) super-Lie current algebra. The
density of states is obtained from the sigma model.
Acknowledgements
We would like to thank M.P.A. Fisher for early discussions about the replica version
of the model by Hatsugai, Wen and Kohmoto. We also benefited from conversations with
D. Bernard. This work was supported in part by the foundation FOM of the Netherlands
(S.G.), by the National Science Foundation, in particular through the National Young
Investigator program (A.LeC.), and in part by the A.P. Sloan Foundation (A.W.W.L).
Appendix A. SUSY approach to the random field XY-model

In this Appendix we show that the SUSY approach to the random field XY model leads
to the N = 1 species action discussed in the bulk of the paper.
The non-hermitian Hamiltonian relevant for the 2D random field XY model is
H2 = (ix + iAx )1 + (iy + iAy )2 + M3 iV 12 .
In order to be able to integrate over bosonic variables, we need to consider the hermitian
Hamiltonian

0 H2
(A.1)
H4
H2 0
= ix 1 1 iy 2 1 + Ax 1 2 + Ay 2 2
+ M3 1 V 12 2 .
(A.2)
In the case of interest where all potentials Ax , Ay , M, V are real, H4 is time reversal
invariant, which means that H4 can be brought into real, symmetric form. Explicitly, this
can be achieved by conjugating H4 with the matrix
U = U U , where U (1 i1 )/ 2, U (1 i3 )/ 2,
yielding the real symmetric Hamiltonian
H4s U H4 U = ix (1 2 ) + iy (3 2 ) V (1 1 ) M(2 2 ).
(A.3)
For integration over fermionic variables an antisymmetric form is needed. This can be
obtained by conjugating by
U 0 U 1 2 ,
H4a U 0 H4 U 0 = ix (1 1 ) + iy (3 1 ) + V (1 2 ) M(2 1 ).
(A.4)
506
All correlation functions relevant for the XY model can be obtained from an action
constructed with a (real) 4-component fermion field , and a real 4-component boson field
, defined as follows:

1
1

+
2
,
=
,
=
3
4
4
with Lagrangian (density)
L Lf + Lb ,
Lf i t U 0 H4 U 0 ,
Lb i t U H4 U .
It will prove convenient to introduce 4-component complex fields and their adjoints by

+
+ , t U 0 ,
U 0 ,
Fermions:

+

,
t U ,
U ,
+
Bosons:
so that 28
Lf = i H4 = 2 i + H2 = 2 i H2 + ,

H2 = 2 i
H2 + .
Lb = i H4 = 2 i+
The so-defined fields are not independent from their adjoints. Rather, the above definitions
immediately give the following relations:
( )t = U 0 = U 0 U 0 ,
( )t = = U U .
Since U 0 U 0 = U U = i1 and U U = 1 3 this implies
1
2
1
2
2
1
1

2 =
3 = (i) 4 ,
4 .
3
4
3
4
3
Therefore, we may chose

3
L

+ = 1 , 2 R , L , and =
,
4
R
as well as

= 1 , 1 R , L ,
+

and =
4
4
L
R
(A.5)

,
(A.6)
as our independent 4 fermionic and 4 (real) bosonic integration variables. The resulting
supersymmetric action

L
L
+ i R , L H2
LSUSY = i R , L H2
R
R
is the N = 1-species theory discussed in the bulk of the paper.
28 Note that the latter equalities involving , arise because H s is symmetric, and because U 0 and U do
4
not mix the 2 2 blocks involving H2 .
507
Appendix B. The most general particlehole symmetric Dirac Hamiltonian

In this appendix we discuss in some detail the SUSY invariant action for the most
general random particlehole symmetric, but not necessarily time-reversal symmetric
Dirac Hamiltonian. We start from the hermitian quantum mechanical Dirac Hamiltonian
H4 defined by

i12
H2
(B.1)
H4
H2
i12
where
H2 (ix + iAx )1 + (iy + iAy )2 + M3 iV 12 .
(As usual, infinitesimals 0+ are needed to extract quantum mechanical Greens
functions.) This model is time reversal invariant if all the potentials (Ax , Ay , V , M) are
real. In this case the time reversal operation is implemented by
T H4 T = H4
with T = 1 3 .
(B.2)
We may consider a more general (non-hermitian) model, where we add to any of the
potentials an imaginary part, A0x , A0y , M 0 , V 0 . Note that A0x , A0y corresponds to a real vector
potential. The most general hermitian Hamiltonian H4 incorporating all these potentials is
H4 = ix 1 1 iy 2 1 + Ax 1 2
+ Ay 2 2 + M3 1 V 12 2
A0x 1
1 A0y 2
(B.3)
0
1 M 3 2 + V 12 1 .
(B.4)
In general, this Hamiltonian is no longer time reversal invariant. 29 However, it is always

invariant under particlehole transformations
CH4 C = H4
with C = 1 3 .
(B.5)
Time reversal symmetry is recovered for the special case where only Ax , Ay , V 0 , M 0 are
non-vanishing. In this case the time reversal operation is
T 0 H4 T 0 = H4
with T 0 = 2 12 .
(B.6)
These time reversal symmetry properties are summarized in the following table:
non-vanishing potentials
time reversal
Ax , Ay , V , M
YES (T )
Ax , Ay , V 0 , M 0
YES (T 0 )
others
NO
Let us now return to the most general situation, where time reversal symmetry is in
general absent. The 4 4 matrix corresponding to the Hamiltonian H4 has the following
block structure:
0
m
+ Az + iA0z
+ Az + iA0
m +
z
, (B.7)
iH4 =
0
m +
Az + iAz
0
m
0
Az + iAz
29 The imaginary parts A0 , A0 , M 0 , V 0 break invariance under time reversal, using T defined above.
x
y
508
where
m V + iM,
M 0 iV 0
and
Az Ax iAy ,
Az Az ,
(1)A0z A0y iA0x ,
A0z A0
z .
This matrix is manifestly anti-hermitian as it should. 30

Averaged Greens functions of the Hamiltonian H4 can be obtained from the following
action. We introduce 31 a 4-component complex bosonic field and its adjoint ,

L2
1

2
R2
,

3
L1
4
R1

, L1
, R2 , L2 ,
(B.8)
1 , 2 , 3 , 4 R1
as well as a 4-component (complex) fermionic field and :

L2
1

2
R2
3
L1
4
R1

, L1
, R2 , L2 .
1 , 2 , 3 , 4 R1
(B.9)
The SUSY Lagrangian (density) reads, in the absence of the vector potentials:
Lsusy = iH4 + iH4
h

i
1 + 1 4 + 4
1
= 1 4 + 4
h

i
3 + 3 (2 ) + (2 )
3
3 (2 ) + (2 )
h
i

+ m+ 1 3 + 1 3 + m 4 (2 ) + 4 (2 )
h

i
m+ 3 1 + 3 1 + m (2 )4 + (2 )4
h
+ 1 1 + 1 1 + (2 )(2 ) + (2 )(2 )
i
+ 3 3 + 3 3 + 4 4 + 4 4
(B.10)
where
m+ (m + ),
m (m ).
30 Note that [ i ], [ + i ], so that = [ i ] = .
x
y
x
y
x
y
31 The symbols , , , will be used to exhibit the chiral (L/R) nature of the corresponding fields
Ri Li Ri Li
(see below).
509
This may also be expressed in terms of the R- and L-moving fields, defined above:
X

La +
La
Ra + Ra +
Lsusy =
Ra
+m
+
X
Ra
La
Ra
L1
R1
La
+ Ra
La
+ R1
L1

+m
La
X
Ra
La
L2
R2
+ R2
L2
+ La
Ra
i
h

i
R1 + L1
R1 L2
R2 + L2
R2
L1
h
L2
R2
L1
+ L2 R1 L1 R2 + R1
L2
+ R1
i
+ R2
L1
R1 L2 + R2 L1 .
(B.11)
The vector potentials give rise to an additional contribution to the action of the form:

X
X

A
Ra Ra + Ra Ra + Az
La La + La La
Lsusy Az
a
R1 + R1
R1 R2
R2 R2
R2
+ iA0z R1
L1 + L1
L1 L2
L2 L2
L2
+ iA0z L1
h
h

i

i
0
0
0
0
J11 + J22
J22 + Az J11
J11 + J22
J22
= Az J11
h

i
0
0
J11 J22
J22
+ iA0 z J11
h

i
0
0
J11 J22
J22 .
+ iA0z J11
(B.12)
Appendix C. 1-loop RG equations for the generalized random XY model (broken

time reversal symmetry)
In this appendix we analyze the 1-loop RG equations for the random XY model without
time reversal symmetry.
Let us first consider the RG equations for the couplings g1 = gm g and g2 = gm + g ,
defined in Eqs. (4.80), (4.81). The most general form 32 of the 1-loop RG is:

dg1
1
1
1
1
= a1 g12 + b1 g1 g2 + c1 g22 + d11
g1 + d21
g2 g1A + d12
g1 + d22
g2 g2A ,
dl

dg2
2
2
2
2
= a2 g12 + b2 g1 g2 + c2 g22 + d11
g1 + d21
g2 g1A + d12
g1 + d22
g2 g2A .
dl
We proceed in the following steps, to simplify these equations.
32 Note that we have omitted terms on the r.h.s. which are quadratic in the vector potential couplings, i.e., of the
0 (see Eq. (4.82)). Since there are no poles
form giA gjA . The operators coupling to giA are of the form Jaa and Jbb
in the OPE of such operators with each other (see Eq. (4.51)) they do not generate 1-loop RG flows.
510
Step 1. We know that both RG equations vanish when g2 = g2A = 0 describing two
1 = a = d 2 = 0.
decoupled 1-species theories. This gives a1 = d11
2
11
Step 2. Adding and subtracting the two equations:

dgm
d(g1 + g2 )
1
2
=2
= (b1 + b2 )g1 g2 + (c1 + c2 )g22 + d21
+ d21
g2 g1A
dl
dl
1

A
2
1
2
+ d12
+ d12
+ d22
g1 + d22
g2 g2 ,

dg
d(g1 g2 )
1
2
= (2)
= (b1 b2 )g1 g2 + (c1 c2 )g22 + d21
d21
g2 g1A
dl
dl
1

A
2
1
2
+ d12
d12
d22
g1 + d22
g2 g2 .
Setting g = gA0 = 0, both equations must vanish identically. In this case we have g1 =
g2 = gm and g1A = g2A = gA , which yields, when inserted into the RG equations:
1
1
1
2
2
2
+ d12
+ d22
= b2 + c2 = d21
+ d12
+ d22
= 0.
b1 + c1 = d21
Step 3. Finally, setting gm = gA0 = 0, again both equations must vanish. In this case
we have g1 = g , g2 = g and g1A = g2A = gA , which yields, when inserted into the RG
equations:
1
1
1
2
2
2
d12
+ d22
= b2 + c2 = d21
d12
+ d22
= 0.
b1 + c1 = d21
Combining Steps 1, 2 and 3 we find:

1
2
= a2 = d11
= 0,
a1 = d11
b1 = b2 = c1 = c2 = 0,
1
2
d12
= d12
= 0,
1
1
2
2
+ d22
= d21
+ d22
= 0.
d21
(C.1)
The 1-loop RG equation thus simplifies to

dg1
1 A
1 A
= d21
g1 + d22
g2 g2 ,
dl

dg2
2 A
2 A
= d21
g1 + d22
g2 g2 .
dl
Using Eq. (4.83) this becomes

dg1
1
1
= d21
d22
gA0 g2 ,
dl

dg2
2
2
= d21
d22
gA0 g2 .
dl
Let us now consider explicitly the relevant OPEs: The coupling constant gA0 couples to a
leftright bilinear of a Cartan generator. Since the simple pole term in the OPE of a Cartan
generator with any operator can only give back the same operator, times a number (= the
eigenvalue of the Cartan generator when applied to this operator), we know that gA0 g2 can
511
1 d1 ) = 0
only generate again the operator coupling to g2 . Therefore we know that (d21
22
and the RG equations simplifies further to:

dg2
dg1
2
2
= 0,
= d21
d22
(C.2)
gA0 g2
dl
dl
(recall that g2 = gm + g , from Eq. (4.81)).
The corresponding RG equations for the vector potential couplings are of the form
dgA
2
= 1 gm
+ 1 gm g + 1 g2 ,
dl
dgA0
2
= 2 gm
+ 2 gm g + 2 g2 .
dl
Since gA0 is not generated when either g = 0 or when gm = 0, we have 2 = 2 = 0.
In addition, the flow for gA is the same in both cases, implying 1 = 1 . Adding and
subtracting the resulting equations

dg1A d(gA gA0 )
2
=
= 1 gm
+ g2 + (1 2 )gm g ,
dl
dl

dg2A d(gA + gA0 )
2
=
= 1 gm
+ g2 + (1 + 2 )gm g .
dl
dl
When g2 = 0 (i.e., g = gm , g1 = 2gm ) we know that
dg2A
dg1A
= 1 g12 ,
= 0.
dl
dl
This finally yields the following RG equations (1 = 0, 2 = 21 ):

dgA0
dgA
= 2 (gm )2 + (g )2 ,
= 4gm g ,
dl
dl
where we have used that 1 = 2 [see Eq. (4.9)].
(C.3)
References
[1] D. Bernard, (Perturbed) conformal field theory applied to 2D disordered systems: an introduction, hep-th/9509137.
[2] J.L. Cardy, S. Ostlund, Phys. Rev. B 25 (1982) 6899.
[3] C. Mudry, X.-G. Wen, Nucl. Phys. B 549 (1999) 613, and references therein.
[4] A.W.W. Ludwig, M.P.A. Fisher, R. Shankar, G. Grinstein, Phys. Rev. B 50 (1994) 7526.
[5] C.-M. Ho, J.T. Chalker, Phys. Rev. B 54 (1996) 8708.
[6] M.P.A. Fisher, private communication.
[7] Y. Hatsugai, X.G. Wen, M. Kohmoto, Phys. Rev. B 56 (1997) 1061.
[8] C. Mudry, C. Chamon, X.-G. Wen, Nucl. Phys. B 466 (1996) 383443.
[9] D. Bernard, A. LeClair, Commun. Math. Phys. 142 (1991) 99.
[10] S. Hikami, M. Shirai, F. Wegner, Nucl. Phys. B 408 (1993) 413.
[11] R. Gade, F. Wegner, Nucl. Phys. B 360 (1991) 213.
[12] R. Gade, Nucl. Phys. B 398 (1993) 499.
[13] T.P. Eggarter, R. Riedinger, Phys. Rev. 18 (1978) 569.
[14] B.M. McCoy, T.T. Wu, Phys. Rev. 176 (1968) 631.
512
[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]
R. Shankar, G. Murthy, Phys. Rev. B 36 (1987) 536.

L. Balents, M.P.A. Fisher, Phys. Rev. B 56 (1997) 12970.
D.S. Fisher, Phys. Rev. B 51 (1995) 6411.
M. Bershadsky, S. Zhukov, A. Vaintrob, Nucl. Phys. B 559 (1999) 205.
N. Berkovits, C. Vafa, E. Witten, JHEP 9903 (1999) 018.
M. Zirnbauer, Conformal field theory of the integer quantum Hall transition, hep-th/9905054.
M. Zirnbauer, J. Math. Phys. 37 (1996) 4986.
A. Altland, M. Zirnbauer, Phys. Rev. B 55 (1997) 1142.
K.B. Efetov, Adv. Phys. 32 (1983) 53.
K.B. Efetov, Supersymmetry in Disorder and Chaos, Cambridge Univ. Press, Cambridge, UK,
1997.
K.B. Efetov, Phys. Rev. Lett. 79 (1997) 491; Phys. Rev. B 56 (1997) 9630.
C. Mudry et al., Phys. Rev. B 58 (1998) 13539.
N. Hatano, D.R. Nelson, Phys. Rev. B 56 (1997) 8651.
T. Fukui, Nucl. Phys. B 562 (1999) 477.
A. Altland, B.D. Simons, J. Phys. A 32 (1999) L353.
D. Friedan, E. Martinec, S. Shenker, Nucl. Phys. B 271 (1986) 93.
A.B. Zamolodchikov, Int. J. Mod. Phys. A 4 (1989) 4235.
L. Rozansky, H. Saleur, Nucl. Phys. B 376 (1992) 461.
A.B. Zamolodchikov, Sov. J. Nucl. Phys. 46 (1987) 1090.
A.W.W. Ludwig, Nucl. Phys. B 285 (1987) 97.
J.L. Cardy, Scaling and Renormalization in Statistical Physics, Cambridge Univ. Press,
Cambridge, UK, 1997.
M.R. Gaberdiel, H.G. Kausch, Phys. Lett. B 386 (1996) 131137.
Y. Morita, Y. Hatsugai, Phys. Rev. Lett. 79 (1997) 3728.
F.J. Wegner, Z. Phys. B 35 (1979) 208.
A.J. McKane, M. Stone, Ann. Phys. 131 (1981) 36.
V. Gurarie, Nucl. Phys. B 410 (1993) 535.

Symplectic fermions
Horst G. Kausch
Department of Mathematical Sciences, University of Durham, South Road, Durham DH1 3LE, UK
Abstract
We study two-dimensional conformal field theories generated from a symplectic fermion
a free two-component fermion field of spin one and construct the maximal local supersymmetric
conformal field theory generated from it. This theory has central charge c = 2 and provides the
simplest example of a theory with logarithmic operators.
Twisted states with respect to the global SL(2, C)-symmetry of the symplectic fermions are
introduced and we describe in detail how one obtains a consistent set of twisted amplitudes. We
study orbifold models with respect to finite subgroups of SL(2, C) and obtain their modular invariant
partition functions. In the case of the cyclic orbifolds explicit expressions are given for all two-,
three- and four-point functions of the fundamental fields. The C2 -orbifold is shown to be isomorphic
to the bosonic local logarithmic conformal field theory of the triplet algebra encountered previously.
We discuss the relation of the C4 -orbifold to critical dense polymers. 2000 Elsevier Science B.V.
All rights reserved.
PACS: 11.25.Hf
Keywords: Conformal field theory
1. Introduction
Over the last few years it has become apparent that there is a class of conformal field
theories whose correlation functions have logarithmic branch cuts. Such conformal field
theories are believed to be important for the description of certain statistical models, in
particular in the theory of (multi)critical polymers and percolation [15], two-dimensional
turbulence [68], and the quantum Hall effect [9,10]. Models analysed so far include
the WZNW model on the supergroup GL(1, 1) [11], the coset model SL(2, C)/SU(2)
[10], the c = 2 model [12,13], gravitationally dressed conformal field theories [14] and
some critical disordered models [1519]. Singular vectors of some Virasoro models have
been constructed in [20], correlation functions have been calculated in [21,22], and more
structural properties of logarithmic conformal field theories have been analysed in [23].
h.g.kausch@dur.ac.uk
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 9 5 - 9
514
H.G. Kausch / Nuclear Physics B 583 [FS] (2000) 513541
The class of chiral logarithmic conformal field theories is also interesting from a
conceptual point of view. There exist logarithmic models which behave in many respects
like ordinary (non-logarithmic) chiral conformal field theories, and it is not yet clear in
which way these models differ structurally from conventional theories. They include the
models investigated in this paper, a series of quasirational logarithmic Virasoro models
[13] and a series of rational logarithmic models, the simplest of which is the triplet
algebra at c = 2 [24]. Here quasirational means that a countable set of representations
of the chiral algebra closes under fusion (with finite fusion rules), and rational that
the same holds for a certain finite set of representations, including all (finitely many)
irreducible representations. For rational models Zhus algebra [25] is finite-dimensional,
and it should be possible to read off all properties of the whole chiral theory from
the vacuum representation. In particular, one should be able to decide a priori whether
the chiral algebra leads to a logarithmic theory or not, without actually constructing all
the amplitudes. As yet little progress has been made in this direction, although it is
believed that unitary (rational) meromorphic conformal field theories always lead to nonlogarithmic theories.
The only rational logarithmic model which has been studied in detail so far, the
aforementioned triplet algebra at c = 2, possesses another oddity (apart from the
appearance of indecomposable reducible representations which lead to logarithmic
correlation functions), and it is quite possible that this is true in more generality [26]:
although the theory possesses a finite fusion algebra, the matrices corresponding to the
reducible representations cannot be diagonalised, and a straightforward application of
Verlindes formula does not make sense. This is mirrored by the fact that the modular
transformation properties of some of the characters cannot be described by constant
matrices as they depend on the modular parameter . This might suggest that these
logarithmic rational theories only make sense as chiral theories, and that they do not
correspond to modular invariant (non-chiral) conformal field theories. It was demonstrated
in [27] that, at least for the case of the triplet algebra at c = 2, this is not the case. The
resulting theory is in every aspect a standard (non-chiral) conformal field theory but for the
property that it does not factorise into standard chiral theories. Among other things, this
demonstrates that a non-chiral conformal field theory has significantly more structure than
the two chiral theories it is built from.
The general strategy for constructing a (non-chiral) conformal field theory is as follows.
Determine the two-, three- and four-point functions of the fundamental fields (the fields
that correspond to the fundamental states of the different representations). Given the twoand three-point functions of the fundamental fields, all other amplitudes can be derived
from these, and the consistency conditions of all amplitudes can be reduced to those being
obeyed by the four-point functions. This reduces the problem of constructing any amplitude
to a finite computation which can be done in principle. All data of a conformal field theory
can then be recovered from the complete set of amplitudes, see [28]. Due to the complex
structure of the non-chiral representations for the triplet model it was not feasible in [27] to
complete this programme. However, the two- and three-point functions agree with those of
a model built from a two-component free fermion field, the symplectic fermion model.
515
Since the two- and three-point amplitudes uniquely determine the theory, consistency of
the symplectic fermion model implies consistency of the triplet model.
Our strategy in this paper mirrors the one in [27]: firstly we construct the representations
of the chiral algebra generated by the symplectic fermions. In this case we obtain a unique
maximal indecomposable non-chiral representation. The non-chiral amplitudes are then
determined as co-invariants with respect to the comultiplication for the fermion field. After
imposing the consistency constraints on the hierarchy of amplitudes we arrive at a unique
local logarithmic conformal field theory SF generated from the symplectic fermions. This
theory admits a global SL(2, C) symmetry under which the fermions transform in the fundamental representation. We can thus introduce twisted states such that the symplectic
fermions acquire a phase when moved along a closed loop around a twisted state. Amplitudes involving twisted states are still co-invariants with respect to the (twisted) comultiplication for the symplectic fermions and we shall use this to explicitly calculate the local
two-, three- and four-point amplitudes for twist fields. The consistency constraints give a
system of quadratic equations in the three-point couplings which admits a simple solution.
Orbifold models SF [G] with respect to a finite subgroup of SL(2, C) are obtained by
restricting to (twisted and untwisted) states which are invariant under G. For the cyclic
subgroups this can be done explicitly with the local orbifold amplitudes given by the above
hierarchy of twisted amplitudes where all fields are G-invariant. We determine the different
sectors of the models and give their modular invariant partition functions. The C2 orbifold
model is the logarithmic conformal field theory for the triplet algebra discussed in [27]
while the C4 orbifold model is related to dense critical polymers [1]. Our method cannot be
applied to the non-abelian orbifolds, for these we can only determine the partition functions
by path-integral arguments.
This paper is organised as follows. In Section 2 we define the symplectic fermion model
and construct its non-chiral amplitudes. In Section 3 we introduce twisted sectors and find
their amplitudes. Finally, in Section 4 we investigate the various orbifold models with
respect to finite subgroups of the global SL(2, C) symmetry: the cyclic groups in Sections
4.1 and 4.2 and non-abelian orbifolds in Section 4.3. The equivalence of the C2 orbifold
with the triplet model is established in Section 4.4 and the relation of the C4 -orbifold to
critical dense polymers is discussed in Section 4.5. The more technical aspects of the paper
have been relegated to a number of appendices. The locality of the symplectic fermion
model is established in Appendix A. The twisted amplitudes are explicitly constructed
in Appendix B. The character of the orbifold chiral algebra is derived in Appendix C.
2. Symplectic fermions
We take as our starting point the (, ) ghost system with the first order action [29]
Z

1
+
d2 z
,
(1)
S=
where and are conjugate fermion fields of dimensions 1 and 0, respectively. The
operator product of the chiral fields has short-distance limit
516
1
1
,
(z)(w)
(2)
zw
zw
up to terms regular as z w. The stress tensor for this system has central charge c = 2
and reads
(z)(w)
T (z) = :(z)(z):
(3)
where : : indicates fermionic normal ordering. The system also possesses a natural U (1)
current given by
J (z) = :(z)(z):
(4)
which counts with charge +1 and with charge 1.

Viewed as a conformal field theory by itself this system has a number of problematic
features. If one wishes the vacuum state to be translation invariant the current is not a
primary field [29], rather one has
T (z)J (w)
J (w)
J (w)
1
.
+
+
3
2
(z w)
(z w)
zw
(5)
As a consequence one has charge asymmetry, J0 = 1 J0 . This implies anomalous charge

conservation in vacuum expectation values,
h|1 (z1 , z 1 ) n (zn , z n )|i = 0,
unless 6qi = 6 qi = 1,
(6)
where qi and qi are the charges of i with respect to the currents J and J. In terms of the
fermions this means that vacuum expectation values are non-vanishing only if they contain
exactly one unpaired and field. One has two ground states, the invariant vacuum
and 0 0 , with vanishing norms but h|0 0 |i = 1. Furthermore it is not possible to
construct an inner product on the space states compatible with the standard hermiticity
properties of the stress tensor, Ln = Ln . A further problem is the appearance of fields
which are neither primary fields nor descendants of a primary field, the simplest examples
of which are and J ; for further details see [3].
The ghost system can be bosonised by defining J (z) = (z) in which case the stress
tensor takes on the FeiginFuchs form. However, the aforementioned problems still persist.
In the Coulomb gas construction of the Virasoro minimal models they are resolved by
defining the physical states through a BRST resolution [30]. This identifies the two ground
states and removes the reducible representations of the Virasoro algebra leaving only the
field content of the minimal model. The central charge c = 2 falls into the pattern c =
1 6(p q)2/(pq) for the minimal series albeit with parameters p = 1, q = 2 outside the
allowed range. The screening charge for the BRST resolution corresponds to 0 and the
usual physical space ker 0 / im 0 is trivial, as observed in [1]. This is reflected in the
fact that the standard DotsenkoFateev correlators [31] vanish identically when evaluated
for c = 2 due to cancellations between the conformal blocks [1]. One can get finite
results by considering the limit c 2 however the the algebraic structure of the Virasoro
representations is discontinuous [13].
It was noted in [29] that the (, ) system contains an irreducible chiral algebra A
generated by and . Both of these are Virasoro primary fields of dimension one. We
517
can put them on an equal footing by grouping them together as a two-component fermion
field of dimension one. We shall now apply the algebraic methods introduced in [27] to
construct a local conformal field theory, the symplectic fermion model SF , containing
the chiral algebra A generated by . The resulting model is generated by a non-chiral
.
two-component free fermion field of dimension zero such that = , =
In contrast, the (, ) system can be obtained by adjoining to A A the two chiral fields
and . The two models thus correspond to different local slices of the same non-local
theory.
2.1. Chiral structure
The chiral algebra A of the symplectic fermion model is generated by a two-component
fermion field,
X
n zn1 ,
(7)
(z) =
nZ
of conformal weight one. The field has short-distance expansion

d
(z w)2
resulting in anti-commutators for the Fourier modes,
(z) (w)
(8)
{m , n } = md m+n ,
(9)
where d is an anti-symmetric tensor; we may choose a basis such that d + = 1. This

algebra has a unique irreducible highest weight representation. Its highest weight state
is annihilated by all non-negative fermion modes, m = 0 for m > 0. This representation
is isomorphic to A and provides its vacuum representation with as the vacuum state.
The chiral algebra contains a Virasoro algebra Vir of central charge c = 2 given by
X
1
Ln zn2 = d : (z) (z):
(10)
T (z) =
2
nZ
where d is the inverse of d such that d d = . In algebraic language the stress

tensor corresponds to the conformal state
1
1 .
(11)
L2 = d 1
2
The vacuum is Mbius invariant under this Virasoro algebra, Lm = 0 for m > 2.
While the irreducible representation is unique it can be extended to obtain reducible but
indecomposable representations. Denote by A] the maximal generalised highest weight
representation of A obtained as such an the extension of the vacuum representation. It is
freely generated by the negative modes, m for m < 0, from a four-dimensional space of
ground states. This space is spanned by two bosonic states and , and two fermionic
states, , with the action of the zero-modes 0 given by
0 = ,
0 = d ,
L0 = .
(12)
518
The states and span a two-dimensional Jordan block for L0 . A conformal field theory
based on this maximal representation will thus contain logarithmic operators.
2.2. Non-chiral theory
To construct the non-chiral theory we proceed as described in [27]. Details of the locality
constraints are presented in Appendix A. The result is that there is a unique non-chiral
symplectic fermion model SF . It has two bosonic ground states, , , and two fermionic
ground states , satisfying
0 = 0 = ,
0 = 0 = d ,
L0 = L 0 = .
(13)
The full non-chiral space of states W is generated by the free action of the negative modes,
m , m with m < 0, from the ground states. Using the comultiplication (A.3), amplitudes
of excited states can be expressed in terms of amplitudes involving only fundamental
fields, i.e., fields corresponding to the four ground states. The amplitudes of up to four
fundamental fields are
hi = 1,
hi = 2(),
hi = 2() (=),
hi = 2() 2(=) 2(5),
h i = d ,
h i = d

h i = d

(14)
(13 + 23 12 ),

13 + 14 + 23 + 24 212 34 34

(13 14 )(23 24 ) ,
h i = d d (12 + 34 ) d d (13 + 24 )
+ d d (14 + 23 ),
where
() =
() =
() =
X
ij
X
ij k
X
ij ,
ij j k ,
ij j k kl ,
ij kl
(=) =
(=) =
(5) =
X
ij
X
ij kl
X
2ij ,
2ij kl ,
ij j k ki
ij k
and
ij (zij ) = Z + ln |zij |2 .
(15)
The parameter Z could be set to zero by redefining 7 + Z, but it will be convenient

later on to retain this parameter.
519
The sums are over pairwise distinct labelled graphs (i.e., graphs with vertices); labelled
graphs that differ by a graph symmetry are only counted once. In amplitudes, acts as the
unit operator, except that its one-point function vanishes, hi = 0. The relevant operator
product expansions are, to lowest order in x,

(x) d + (x) ,
(x) (x) ,

(x) (x) 2 + (x) .
(16)
The operator product of with any field S is simply given by (x)S = S to all orders.
The structure of the theory is simple enough to allow us to determine not just the fourpoint amplitudes but all 2N -point amplitudes of the fields.
h 1 2N i
1 X
sign( )d 1 2 d 3 4 d 2N1 2N 3 4 2N1 2N ,
= 2N
2
(17)
S2N
where the sum is over all permutations of the fields and the prefactor ensures that each
grouping of the 2N fields into N pairs is only counted once. This can be proved by
induction: the amplitudes satisfy the correct differential equations in the zi and are thus
fixed up to addition of a constant; considering the OPE fixes that constant. By
contracting two fields one can obtain amplitudes involving . In particular, the N -point
amplitude for can be written a sum over graphs of N 1 links between the N vertices,
X
(1)N+Nc 2Nd (g),
(18)
h i =
N1
gGN
where the propagator (g) for the graph g is given as the product of a propagator ij for
each link between vertices i and j , Nc is the number of components including the isolated
vertex and the two-vertex loop = while Nd is the number of components excluding
and =. The combinatorial factor arises from contracting the d tensors in the 2N -point
amplitude h i with the d tensors in the OPE of and factors as the product of
the following factors for each component of the graph g,
1,
1,
2(1)n+1 ,
2(1)n+1 .
Here the last two subgraphs contain n vertices. For example, the five- and six-point
amplitudes of can be written as
hi = 2() 2() + (= =)
4( ) 2( =),
hi = 2() + 2() 2(= )
+4( ) + 4( ) 2( =)
+2( =).
520
These amplitudes define a fully consistent local conformal field theory on the Riemann
sphere. This does not imply that the theory can be defined consistently on higher genus
Riemann surfaces. Indeed, the partition function of this model is not modular invariant. To
define the theory on the torus one has to add additional sectors where the fermion fields
satisfy anti-periodic boundary conditions along the fundamental cycles of the torus the
different spin structures. This will be discussed in more detail in Section 4.
3. Twisted sectors
The symplectic fermion model admits a global SL(2, C) symmetry under which the
fermion fields transform in the fundamental representation. As in any theory where we
have a global group of automorphisms G we can introduce twist fields g for any g G,
such that, as a field from W is taken around the insertion of a twist field g we obtain
the original field up to the action of the automorphism g,

(19)
e2i z, e2i z g = (g)(z, z )g .
For the (chiral and anti-chiral) symplectic fermions this means,
(e2i z)g = (g )(z)g ,
(e2i z )g = (g 1 )(z)g .
(20)
We shall restrict to the case where G U (1) is an abelian subgroup of SL(2, C). In this
case all the twists commute and we can choose a basis for the two fermion fields such
that d + = 1 and the automorphisms are of the form
g: 7 e2i
(21)
for some and we can use that parameter to label the different twisted sectors. It also
follows that the symplectic fermions have a mode expansion
X
X
n
zn1 ,
(z) =
n
z n1 ,
(22)
(z) =
nZ
nZ
when acting on the -twisted sector. Denote by W the representation freely generated
from a ground state satisfying r = r = 0 for r > 0. This twisted representation
is irreducible since does not have zero-modes. The ground state is a Virasoro highest
weight state with conformal weight
(1 )
.
(23)
2
In this expression, and below, we always take 0 < < 1. We also write = 1 . We
will now construct the amplitudes for the twisted sectors. We need to determine only
the amplitudes of the cyclic states and since they determine all other amplitudes
through the twisted comultiplication (B.2). As usual, the system of differential equations
arising from the (twisted) comultiplication determines the amplitudes up to some structure
constants; the details can be found in Appendix B. Since the overall twist of an amplitude
has to vanish, all amplitudes with a single twist field vanish and the non-vanishing
amplitudes with two twist fields are
h =
521
h i = O |z12 |2 ,

z13 z23 2
2

,
Z + ln
h i = O |z12 |
z12

z13 z23 2
z14 z24 2
2

Z + ln
Z + ln
h i = O |z12 |
z12
z12

H (x)H (x) ,
where H (x) is given by

(1 x)
= 2<
H (x, x)
2 F1 (1, ; 1 + ; 1 x) + Y .
(24)
(25)
For each pair of twisted sectors we have two parameters, O and Z . The constant Y is
fixed by locality of h i and given in Appendix B.1. Amplitudes involving three
twist fields are

C , , z1 2 z1 3 z2 3 2 , for 1 + 2 + 3 = 1,
1 2 3 12
13
23
(26)
h1 2 3 i =
1 2 1 3 2 3 2
C

, for 1 + 2 + 3 = 2,
1 ,2 ,3 z12 z13 z23
from which we obtain the OPEs
C
1 2
1 ,O2,1
|x|212 1 +2 + ,
for 0 < 1 + 2 < 1,
1 +2
C1 ,2 ,21 2
21 2
1 (x)2 =
1 +2 1 + , for 1 < 1 + 2 < 2, (27)
O1 +2 1 |x|
O
for 1 + 2 = 1,
O1 |x|212 + ()(x) + ,
where
() (x) = ln |x|2 + 2Z Z .
(28)
Amplitudes involving four twist fields can be expressed in terms of hypergeometric

functions. As the expressions are quite complicated they are listed in Appendix B.2. By
contracting two fields in these four-point amplitudes we can express the four-point structure
constants in terms of the three-point couplings.
1 + 2 + 3 + 4 = 1:
F1 ,2 ,3 ,4 =
C1 ,2 ,11 2 C1 +2 ,3 ,4
,
O1 +2
1 + 2 + 3 + 4 = 3:
F1 ,2 ,3 ,4 =
C1 ,2 ,21 2 C1 +2 1,3 ,4
,
O1 +2 1
522
1 + 2 + 3 + 4 = 2:
C1 ,2 ,11 2 C1 +2 ,3 ,4
p
, if 1 + 2 < 1,
O1 +2 (1 , 2 )(3 , 4 )
C3 ,4 ,13 4 C3 +4 ,1 ,2
F1 ,2 ,3 ,4 =
p
, if 1 + 2 > 1,
O3 +4 (3 , 4 )(1 , 2 )
O1 O3 ,
if 1 + 2 = 1.
O
Here, (, 0 ) is a ratio of gamma functions,
(, 0 ) =
(1 0 ) () (0 )
,
( + 0 ) (1 ) (1 0 )
(29)
and F1 ,2 ,3 ,4 is the overall normalisation of the amplitude h1 2 3 3 i. Locality

also fixes the the parameter
Z = Z + , ,
,0 = 2(1) () (0 ),
(30)
where (x) = (x)0 / (x) is the digamma function. The structure constants F1 ,2 ,3 ,4
are completely symmetric in their arguments and considering all possible orderings yields
quadratic constraints on the three-point couplings. A solution to these locality constraints
is given by
s
(1 ) (2 ) (3 )
,
(31)
C1 ,2 ,3 = C1 ,2 ,3 =
O = 1,
(1 ) (2 ) (3 )
where 1 + 2 + 3 = 1. The four-point couplings are
s
(1 ) (2 ) (3 ) (4 )
,
F1 ,2 ,3 ,4 = F1 ,2 ,3 ,4 =
(1 ) (2 ) (3 ) (4 )
(32)
for 1 + 2 + 3 + 4 = 1 and F1 ,2 ,3 ,4 = 1 for 1 + 2 + 3 + 4 = 2.

This hierarchy of twisted amplitudes provides a consistent set of semi-local amplitudes:
by construction, the ground states of the twisted sectors are G-invariant and amplitudes
involving only those the ground states and the cyclic field are local and satisfy
all consistency constraints. However, excited states may not be G-invariant and their
amplitudes will acquire phase factors when moving fields along closed loops around such
an excited field. To obtain local amplitudes we have to restrict to G-invariant states; this
defines the orbifold model SF [G].
4. Orbifold models
An abelian subgroup G of SL(2, C is either U (1) or CN , the cyclic group of order N .
In all these cases irreducible representations of G are one-dimensional and can be labelled
by a weight (or twist) such that the generator g acts as exp(2i), as has been done
implicitly in the previous section. In the case of CN the twists are Z/N while for U (1)
523
the twists are continuous, R. In both cases the set of allowed twists can be identified
with G since shifting a twist by an integer results in the same representation.
Given the symplectic fermion model SF with space of states W and chiral algebra A
we construct the orbifold model SF[G] as follows [32]. The space of states is given by the
G-invariant subspaces H = W [G] of the (twisted) modules W for all G. They form
representations of the orbifold chiral algebra A[G]. The orbifold amplitudes, given by the
above twisted amplitudes where all fields are G-invariant, are fully local. The spaces H are
in general reducible representations of the orbifold chiral algebra A[G] and decompose as
M
H, .
(33)
H =
G
We first consider the Vir U (1) characters of the spaces W and W . The characters of
the G-invariant subspaces H0 and H, can then be obtained by specialising the U (1)
characters to the finite group G and using the orthogonality of G-characters.
Since W is irreducible with respect to the original chiral algebra A it is simply the
product of a chiral and an anti-chiral twisted sector,
W = V V ,
(34)
where V is generated from a ground state of conformal weight h = /2 by the modes

+
and n+1
with n > 0. These chiral spaces decompose under G as
n
M
V ,
(35)
V =
G
such that g acts as exp(2i) on V . The spaces V are modules of the orbifold chiral
algebra A[G] and we have chosen the ground state to be invariant under G. We can
introduce the chiral Vir U (1) character

1 Y

1 + gq n+ 1 + g 1 q n+
V (, g) = trV e2i (L0c/24) g = q 2 + 12
= ( )
n=0
m (2m+21)2/8
g q
(36)
mZ
where ( ) is the Dedekind -function, q = exp(2i ) and g is an element of U (1). In the

last line we used the Jacobi triple product identity. The untwisted sector W is not simply the
product of a chiral and an anti-chiral representation. However, since it is freely generated
we
from the ground states by the action of the chiral and anti-chiral algebras, A and A,
have the non-chiral Vir U (1) character
W (, g) = trV e2i (L0c/24) e2i (L0 c/24) g

!
Y
1
1
n
1
n
(1 + gq )(1 + g q )
= (q q)
12 (2 + g + g )
n=1
Y
n=1
1 + g q
1+g
1 n
524
= q
1
12
!
(1 + gq )(1 + g
n
1 n+1
n=0
= |( )|2
!
gm q
(2m1)2 /8
mZ
1
12
!
(1 + g q
n=0
g m q
(2m+1)2 /8
n+1
)(1 + g
1 n
q )
!
.
(37)
mZ
4.1. C2N orbifold

Specialising now to the case of G = C2N we have g 2N = 1 for any group element g and
the twists are of the form = l/2N . Thus the character of Vl/2N can be written in the form
V(l/2N) (, g) =
2N1
X
g k 2NkN+l,2N 2 ( ),
(38)
where n,m ( ) = n,m ( )/( ) with the classical theta function defined as
X
2
q mk .
n,m ( ) =
(39)
k=0
kZ+n/2m
The orbifold representations H = W [C2N ] with 6= 0 are reducible with respect to the
orbifold chiral algebra A[C2N ],
M
M
H, =
V V .
(40)
H =
C2N
C2N
The non-chiral character of H is just the modulus squared of the chiral character of V ,

2
2
(41)
Hl/2N,k/2N ( ) = V k/2N ( ) = 2NkN+l,2N 2 ( ) .
l/2N
For the untwisted sector we obtain

X
2
2 /8
q (2m1) /8 q (2m1)
H0 ( ) = |( )|2
mm0(2N)
= |( )|
2N1
X
2N1
X
l=0
kZ
!
q
(4Nk+2l1)2 /8
(2l1)N,2N 2 ( )
!
q
2 /8
(4N k+2l1)
kZ
2
.
(42)
l=0
The space of states of the orbifold theory SF[C2N ] consists of the indecomposable
(extended) vacuum module H0 , arising from the untwisted sector W, and 2N(2N 1)
sectors
H, = V V ,
with 6= 0, arising from the decomposition

M
H,
H = W [C2N ] =
C2N
of the twisted sectors W . The partition function of SF[C2N ] is thus
(43)
(44)
Z[C2N ] = H0 ( ) +
2N1
X 2N1
X
l=1
2N1
X 2N1
X
l=0
Hl/2N,k/2N ( )
k=0
2NkN+l,2N 2 ( )
k=0
4N 2 1
525
m,2N 2 ( )
2
.
(45)
m=0
This partition function is invariant under the modular group and is in fact identical to the
partition function of a free boson compactified on a circle of radius r = 2N using the
normalisation of [33]. Furthermore, defining as in [34] the Coulomb gas partition function
at radius as
Z(, ) =
X 1
1
2 1
2
q 4 (m+n/) q 4 (mn/)
2
|( )|
(46)
m,nZ
we have Z[C2N ]( ) = Z( 2N, ).

The orbifold chiral algebra A[C2N ] has character (see Appendix C)
#
"
N1
X
1
2
l
(1) (2N 2l 1)(2l+1)N,2N 2 ( ) .
( ) +
A[C2N ] ( ) =
2N
(47)
l=0
Because of (1/ )2 = ( )2 , this does not transform nicely under the modular group.
This is an indication that the chiral algebra A[C2N ] should not by itself form a sector of the
theory but rather be contained within the larger vacuum module H0 .
We can also decompose the chiral algebra A with respect to the Virasoro algebra of
central charge c = 2 contained within it. The character of A can then be expressed in
terms of the irreducible characters of the Virasoro algebra at c = 2 as
!
X
n
X
2ln
V ir
u
( ),
(48)
n+1,1
A (, u) =
n=0
l=0
and thus the character of the orbifold chiral algebra is given in terms of Virasoro characters
as

X
k
V ir
2k+1,1
( ),
(49)
1+2
A[C2N ] (, u) =
N
k=0
where bxc is the largest integer less or equal to x. Viewed as a W -algebra, the orbifold
chiral algebra A[C2N ] is generated by the Virasoro field L of conformal weight two, a
Virasoro primary field of weight three and a pair of Virasoro primary fields of weight
N(2N + 1). In the case of N = 1 the three Virasoro primary fields all have weight three
and we obtain the triplet algebra, see Section 4.4.
526
4.2. CN orbifold
We can repeat the same procedure for CN orbifolds with N odd. However, the CN invariant states still contain fermions. The partition function Z[CN ] corresponds to a single
spin-structure and is not modular invariant. The CN orbifold has an additional C2 symmetry
arising from the bosonfermion selection rule. Performing the sum over spin-structures in
the SF [CN ] model is the same as directly performing the orbifold construction with respect
to CN C2 = C2N .
4.3. Non-abelian orbifolds
The remaining finite subgroups of SL(2, C) are the binary tetrahedral, octahedral and
icosahedral groups T , O and I and one would like to construct SF [G] orbifolds for them
as well. The general structure of orbifold models was investigated in [32]: The twisted
sectors are labelled by conjugacy classes of G. They can be constructed as before and, in
fact, the structure of the g-twisted module depends only on the order of g. The n-point
amplitudes are labelled by n-tuples of group elements (g1 , . . . , gn ) such that g1 gn = 1.
Labels related by simultaneous conjugation yield the same amplitude. Our method for
constructing the amplitudes relies on being able to simultaneously diagonalise the twist for
all fields in the amplitude. This is not possible if some of the gi are not mutually commuting
as is necessarily the case for some amplitudes when G is non-abelian. A direct construction
of the T , O and I orbifolds is thus not possible.
However, we can determine the torus partition function of SF[G] as a sum over partition
functions of the symplectic fermion field with different boundary conditions: for g, h
G we denote the partition function of the h-twisted sector with an insertion of the operator
g as
c
c
(50)
g 2 = trWh gq L0 24 q L0 24 .
h
The G-orbifold partition function is then obtained as

Z[G] =
1 X
g2.
h
|G|
(51)
g,hG
gh=hg
For non-abelian G boundary conditions twisted by non-commuting group elements are not
consistent, hence the condition gh = hg. This allows us to interpret the twisted partition
functions as the Vir U (1) characters introduced previously,
g 2 = Wh (, g).
h
(52)
For C2N we recover the previous result which we denote by ZN = Z[C2N ]. To calculate the
partition function Z[G] for non-abelian G we follow [35] and add the contributions of the
mutually commuting subsets of G, which form cyclic groups, subtracting any overcounting.
As in [35] the result is
527
1
Z[DN ] = (ZN + 2Z2 Z1 ),
(53)
2
1
(54)
Z[T ] = (2Z3 + Z2 Z1 ),
2
1
(55)
Z[O] = (Z4 + Z3 + Z2 Z1 ),
2
1
(56)
Z[I] = (Z5 + Z3 + Z2 Z1 ).
2
Here, of course, one
has to keep in mind that ZN ( ) corresponds to a Coulomb gas partition
function at radius N 2 and not N as for c = 1.
4.4. The triplet model
We will now show that the C2 -twisted model is the logarithmic conformal field theory of
[27]. The chiral algebra of the theory in [27] is a W -algebra, the so-called triplet algebra
[3,36]; it is realised in the symplectic fermion model by
1
1 ,
L2 = d 1
(57)
2
a
a
= t
2
1 .
W3
Here the matrices (t a ) form the spin 1/2 representation of sl(2) in a CartanWeyl basis,
1
0
t
= ,
= 1,
(58)
t
2
with all other entries vanishing. The bosonic sector H0 = W0 [C2 ] of W0 contains the
representation R of the triplet algebra [27]. Explicitly, the ground states and are
identified in both representations and the higher level states of R can be expressed as
fermionic descendents as

,
= 1

= 1 ,
(59)
1
,
= 1
1
.
= 1
On the other hand, the bosonic sector has the non-chiral character

2
H ( ) = 21,2 ( ) = R ( ).
0
(60)
Since H0 and R have the same character but H0 contains R they are in fact identical,
H0 = R. By the same character argument we find that the orbifold algebra A[C2 ] is
identical to the triplet algebra.
The other two representations of the triplet model, the irreducible representations
V1/8,1/8 and V3/8,3/8, also have an interpretation in terms of the symplectic fermion
theory: they correspond to the bosonic sector H1/2 = W1/2 [C2 ] of the (unique) C2 -twisted
representation W1/2 . In this sector, the fermions are half-integrally moded, but all bosonic
operators (including the triplet algebra generators that are bilinear in the fermions) are still
integrally moded, and the twisted sector decomposes as
H1/2 = H1/2,0 H1/2,1/2
(61)
528
with respect to the triplet algebra. The ground state of the twisted sector is identified with
the ground state of V1/8,1/8 while the ground state of V3/8,3/8 is given as a fermionic
descendant in H1/2 ,
1/2
.
= 1/2
(62)
Thus, V1/8,1/8 is contained in H1/2,0 and V3/8,3/8 is contained in H1/2,1/2 and by

considering the characters,
H1/2,0 ( ) = |0,2 ( )|2 = V1/8,1/8 ( ),
H1/2,1/2 ( ) = |2,2 ( )|2 = V3/8,3/8 ( ),
(63)
we find that, in fact, H1/2,0 = V1/8,1/8 and H1/2,1/2 = V3/8,3/8 as representations of the
triplet algebra. The C2 orbifold and the triplet theory have the same partition function
Z[C2 ] =
3
X

k,2 ( )2 = Ztriplet.
(64)
k=0
The amplitudes determined in Section 3 agree with those in [27] on setting Z =

4 ln 2 and thus Z1/2 = 8 ln 2. Furthermore, all three-point functions of the fundamental
fields, , , and , of the triplet model agree with the corresponding excited
amplitudes in the C2 -twisted symplectic fermion model. Both models are therefore
isomorphic. This is the argument used in [27] to establish the consistency of the logarithmic
theory presented there.
4.5. Critical dense polymers
Many properties of polymers in solution can be modeled by considering simple
geometrical systems. In particular, dense polymers are obtained by considering a finite
number of self-avoiding and mutually-avoiding loops or chains on a lattice that cover a
finite fraction of the available volume. It was argued in [1] that their continuum limit should
correspond to a (, ) system. Specifically, the (, ) system on a torus with periodic or
anti-periodic boundary conditions describes the sector formed by an even number of noncontractible loops while the sector formed by an odd number of non-contractible loops
corresponds to Z4 -twisted boundary conditions. The total polymer partition function is
equal to the partition function Z[C4 ]. This result was obtained by realising dense polymers
as the n 0 limit of the low temperature phase of the O(n) model which in turn can be
mapped onto a Coulomb gas. This also reproduces the scaling dimensions
L2 4
(65)
16
for the geometric polymer L-leg operators L . However, the limit n 0 does not
commute with the thermodynamic limit or with changing the boundary conditions.
Furthermore, the physical quantities which have been determined for dense polymers,
the partition function Z[C4 ] and the scaling dimensions (65), are shared by the C4
orbifold models of both the (, ) system and the symplectic fermions. Both models
necessarily involve reducible but indecomposable representations of the Virasoro algebra.
xLD =
529
The structure of these representations is quite different in the two models resulting in
differences for the correlators. In particular, some amplitudes such as the four-point
amplitude of the Ramond ground states of dimensions 1/8 vanish when calculated
directly in the (, ) system or in its Coulomb gas formulation. To get a non-zero
result one has to take the generic Coulomb gas amplitude and take the limit c 2
resulting in an amplitude with logarithmic short-distance behaviour. This agrees with the
amplitude h1/2 1/2 1/2 1/2 i for the symplectic fermions. The same is also the case for
the other logarithmic amplitudes thus providing evidence that the correct description of the
continuum theory of dense polymers is in terms of the SF[C4 ] model. However, a more
detailed analysis is clearly needed: the 1-leg operator 1 is two-fold degenerate and is
represented by 1/4 and 3/4 . The reason for this degeneracy given in [1] is that sources
and sinks of polymers are distinguished. The four-point amplitude of the 1-leg operator 1
(Eq. (88) of [1]) agrees with the amplitude h1/4 3/4 1/43/4 i of the symplectic fermion
model. However, we also obtain a non-vanishing amplitude for four sources (or four sinks),
h1/4 1/4 1/4 1/4 i = h3/4 3/4 3/4 3/4 i =
(1/4)2
|z12 z13 z14 z23 z24 z34 |1/8 .
(3/4)2
An important problem is to understand the role played by the logarithmic fields and identify
their equivalents in the lattice realisation. In both the Z4 -twisted (, )-system and the
orbifold model SF[C4 ] there are two states of conformal dimension h = h = 0. According
to [1] these correspond to the identity and the density operator . In the (, )-system
the density operator is represented by = + and results in non-vanishing expectation
value hi and zero densitydensity correlation hi = 0. In the symplectic fermion model
the density would be some linear-combination of and resulting again in non-vanishing
expectation value but the the density-density correlation acquires a logarithmic behaviour
h(z1 )(z2 )i = A + B ln |z12 |2 .
For a detailed comparison with lattice correlators it would be useful to determine the
amplitudes of the SF[C4 ] in finite geometries, e.g., a rectangular region or strip with
various boundary conditions. The short-distance behaviour of these amplitudes will agree
with those on the Riemann sphere calculated here but the global behaviour will be modified
by finite-size corrections.
The symplectic fermion model provides definite predictions for polymer correlators
going beyond just the set of scaling dimensions. It is hoped that numerical lattice
simulations will be able to verify these predictions and distinguish them from the Z4 twisted (, ) system. On the lattice the logarithmic behaviour of the correlators may
however be masked by discretisation effects.
Similar considerations apply to dilute polymer or percolation models whose continuum
limit is described a c = 0 conformal field theory. The scaling dimensions for dilute
polymers can be reproduced by the Kac formula of conformal weights for c = 0 if one
also allows half-integer indices [1]. A conformal field theory with central charge c = 0
containing Virasoro primary of these conformal weights cannot be the Virasoro minimal
model at c = 0 since this has empty field content. By an analysis of fusion products
530
analogous to that described in [13] one can show that a c = 0 model which contains
fields from the Kac table has to include logarithmic fields. Preliminary results indicate
that the generalised highest weight representations occurring have a considerably more
complicated structure than in the c = 2 case. Further work in this direction will be
reported elsewhere.
Acknowledgements
The author would like to thank Matthias Gaberdiel for many useful discussions.
Appendix A. Locality of symplectic fermions

To construct non-chiral local representations we proceed as described in [27]. We start
off with the tensor product A] A ] of the left and right chiral representations. In a local
(n)
theory the operator S = L(n)

0 L0 , i.e., the nilpotent part of L0 L0 , has to vanish on all
states. The (maximal) non-chiral representation Wmax is thus given as the quotient space

(A.1)
Wmax = A] A ] /N ,
where N is the subrepresentation generated from S( ).
The space of ground states
then has the structure
0 = ,
0 = d ,
0 = ,
0 = ,
0 = d ,
(A.2)
Here, is the equivalence class of states in Wmax which contains as a representative.

Since S commutes with both chiral algebras and A] is freely generated from the ground
space representation by the negative modes m , m with m < 0, the same is true of Wmax .
The representation Wmax contains the states which are allowed a priori in a local theory.
However, the space of states W actually realised in the non-chiral local theory may be
smaller than Wmax , as we might be forced to set some of those states to zero when
requiring the locality of two- and three-point amplitudes. We construct these amplitudes
by the method described in [27]: the N -point amplitudes are co-invariants with respect to
the comultiplications i (n ), where is any field in the chiral algebra, n > h with
h the conformal weight of and i = 1, . . . , N . In our case it is sufficient to use the
comultiplications

X X
n nk (j )
(i)
) = n +
ij
(A.3)
zj i k ,
i (n
k
j 6=i
(j )
k=0
where k is the mode k acting on the j th field in the amplitude and ij is a sign factor
arising from interchanging of fermions when moving from ith to j th position. The
comultiplication allows us to express any amplitude in terms of amplitudes of the ground
531
states alone. These in turn satisfy systems of first order differential equations obtained by
identifying d/dzi with L1 acting on the ith field and noting that
L1 = 0,

L1 = d 1
,
L1 = 1
.
(A.4)
In this way the bosonic two-point amplitudes are found as

hi = 2C1 2C0 ln |z12 |2 ,
hi = C0 ,
h i = C0 ,
(A.5)
hi = 0,
h i = 0,
h i = 0,
while the amplitudes of two fermionic fields are
h i = d C0 ,
h i = d C0 ,
h i = C0 .
(A.6)
Here, C0 , C1 and are arbitrary constants. Amplitudes of one fermionic and one bosonic
field vanish. These amplitudes imply that and are linearly dependent and thus 1
= .
(A.7)

The fermionic states and are linearly dependent provided det() = 1. This is in fact
required by locality of the amplitude h i: Solving the system of differential equations
arising from the comultiplication we obtain

z13 z23
z 13 z 23
+ det()A0 ln
.
(A.8)
h i = d A1 + A0 ln
z12
z 12
This amplitude can be local only if det() = 1. Then the four fermionic states and
are related as
= d ,
= d .
(A.9)
The space of states W is then freely generated by the negative modes, m and m with
m < 0, from the four ground states; two bosonic states, and , and two fermionic states
.
We now determine the remaining three-point amplitudes:
hi = 0,
hi = A0 ,
hi = 2A1 2A0 ln |z12 |2 ,
1 We set states to zero whose amplitudes vanish identically.
532
hi = 3A2 + 2A1 ln |z12 z13 z23 |2

+ 2A0 ln |z12 |2 ln |z13 |2 + ln |z12 |2 ln |z23 |2 + ln |z13 |2 ln |z23 |2

A0 (ln |z12 |2 )2 + (ln |z13 |2 )2 + (ln |z23 |2 )2 ,
h i = d

(A.10)
A0 ,

z13 z23 2

.
A1 + A0 ln
h i = d
z12

From the two- and three-point amplitudes we can read off the OPEs. Locality requires that
the different ways of contracting two fields in a three-point amplitude are equivalent; this
implies
C1 A1 A2
=
=
= Z.
C0 A0 A1
(A.11)
We further introduce parameters and O by C0 = 4 O, A0 = 6 O. The OPEs are then

2 (x) = d + (Z + ln |x|2) ,
2 (x) = (Z + ln |x|2) ,
2

(x) = (Z + ln |x| ) 2 + Z + ln |x| .
2
(A.12)
The operator product of with any field S is simply given by (x)S = 2 S to all orders.
For 2 = 1, the fields can be thought as the unit operator, except that its one-point
function vanishes, hi = 0.
The two- and three-point amplitudes completely determine the theory and all higher
amplitudes can be constructed, at least in principle, from these by a gluing process.
Conversely, by taking limits of n-point amplitudes in which two fields are close together,
we can relate n-point amplitudes to (n 1)-point amplitudes and the OPEs. The different
ways of so contracting two fields all have to be compatible and this leads to consistency
relations. It is sufficient to check these for the four-point amplitudes. These are listed in the
main part of the paper and do indeed satisfy all consistency relations. This shows that the
symplectic fermions define a local conformal field theory.
In the main part of this paper we revert to regular greek letters instead of bold ones to
denote the non-chiral fields. We also fix the parameters as = = 1. They could easily
be restored by multiplying all n-point amplitudes by 2n O. Furthermore, by performing a
global chiral (or anti-chiral) SU(2) transformation, we can choose = d . The only
remaining parameter, Z, corresponds to the freedom of adding to an arbitrary multiple
of .
Appendix B. Twisted amplitudes

In abelian orbifolds the twist conditions satisfied by the twist fields in any amplitude
all commute. This implies that the twists can be simultaneously diagonalised and insertion
points of twist fields are rational branch points for the symplectic fermions. To calculate
amplitudes involving twisted fields we can then use a twisted comultiplication for the
533
chiral fermion fields. To derive the comultiplication formula consider (non-chiral) fields
j having twists j with respect to a chiral field S(w). If 1 + 2 + 3 + 4 Z then, as
a function of w,
(w z1 )1 (w z2 )2 (w z3 )3 (w z4 )4 hS(w)1 2 3 4 i,
(B.1)
is meromorphic with poles at w = zj . Taking a contour integral of w around the other fields
we obtain the comultiplication formula for S,
(S1 2 3 4 hS +1 )

X
2 3 4 2 r 3 s 4 t (1)
z13
z14
S1 +r+s+t hS +1
z12
=
r
s
t
r,s,t =0

X
1 3 4 1 r 3 s 4 t (2)
+
z23
z24
S2 +r+s+t hS +1 (B.2)
z21
r
s
t
r,s,t =0

X
1 2 4 1 r 2 s 4 t (3)
z32
z34
S3 +r+s+t hS +1
z31
+
r
s
t
r,s,t =0

X
1 2 3 1 r 2 s 3 t (4)
+
z42
z43
S4 +r+s+t hS +1 ,
z41
r
s
t
r,s,t =0
(j )
where Sm acts on the j th field. If some fields are fermionic there are additional minus
signs from the interchange of two fermion fields. As in the untwisted case any amplitude
functional satisfies
(B.3)
(Sm ) = 0, for m < hS ,
P
that is, j > 2(hS 1). Two- and three-point amplitudes satisfy analogous comultiplication properties.
We can use the comultiplication to derive differential equations for the amplitudes. Note
that
,
L1 = 1
+
+
2
L1 = (1 )2
1
1
.
(B.4)
To derive a differential equation for h1 i we typically use the comultiplication for

P
with 1 = 1 1 and all other j 6 0 such that j j = 0 to change the 1

mode
on the first field into 0 acting on the fields with j = 0. A second application of the
P
comultiplication for + with 1 = 1 and all other j 6 0 such that j j = 0 then
+
mode on the first field into + acting on the fields with j = 0. In cases
changes the
where we cannot satisfy these conditions on the j we can derive a second order differential
equation instead, as detailed below.
Appendix B.1. Two twist fields
Amplitudes with two twist fields, h1 2 i are non-zero only if 1 + 2 = 1. They
satisfy first order differential equations, for example,
534

1
h i = 0,
z12

z23
h i,
h i =
1
z12
z12 z13

z23
z13 z24
+
h i,
h i =
1
z12
z12 z13 z14 z23

z23
z24
h i +
h i
h i =
1
z12
z12 z13
z12 z14

z13 z24
z23
h + i
+
z12 z13 z14 z23

z24
z14 z23
h + i.
z12 z14 z13 z24
(B.5)
(B.6)
(B.7)
(B.8)
Imposing Mbius covariance and monodromy invariance fixes the functional form of the
amplitudes,
h i = D |z12 |2 ,
(B.9)
h i = D(1) |z12 |2 ,

z13 z23 2
(1)
,
h i = D |z12 |2 Z + ln
z12
(2)
h i = D |z12 |2 ,

z13 z23 2
,
h i = D(2) |z12 |2 Z(1) + ln
z12

z14 z24 2
(2)
(2)
2

,
Z + ln
h i = D |z12 |

z
(B.10)
(B.11)
(B.12)
(B.13)
(B.14)
12
h + i = D(2) |z12 |2 H (x, x),

+
(B.15)
(2)
i = D |z12 |2 H (x, x),
(B.16)

2
2

z13 z23
z14 z24
(2)
(1)
(2)

h i = D |z12 |2
Z + ln
Z + ln

z12
z12

(x, x)
+ X ,
(B.17)
H (x, x)H
is given by
where H (x, x)

(1 x)
= 2<
H (x, x)
2 F1 (1, ; 1 + ; 1 x) + Y
(B.18)

= ln |x|2 21, + Y + 2< (1 x) M(1, ; 1; x) .
Reading off the OPEs from the two- and three-point amplitudes and imposing locality on
the four-point amplitudes fixes the constants appearing in the amplitudes in terms of two
new free parameters, O and Z , for each pair of conjugate twisted sectors,
D = 2 O ,
(1)
D(1) = 4 O ,
(2)
Z = Z = Z ,
535
D(2) = 6 O ,
Y = Z Z + 21, ,
X = 0.
(B.19)
The OPEs are then given by

O 2(1)
|x|
+ (ln |x|2 + 2Z Z ) ,
O
2
(x) = ,

2 (x) = ln |x|2 + Z .
(x)1 =
(B.20)
(B.21)
(B.22)
Appendix B.2. Three twist fields

For the three-point amplitude h1 2 3 i we have to consider two cases. If 1 + 2 +
3 = 1 the amplitude satisfies

3
2
+
(B.23)
h1 2 3 i = 0
1 1
z12 z13
while for 1 + 2 + 3 = 2, we have

1 2 1 3
+
h1 2 3 i = 0.
1 (1 1 )
z12
z13
From this we obtain the amplitudes
1 2 1 3 2 3 2
C1 ,2 ,3 z12
z13 z23 ,
h1 2 3 i =
2
C1 ,2 ,3 z121 2 z131 3 z232 3 ,
for 1 + 2 + 3 = 1,
(B.24)
(B.25)
for 1 + 2 + 3 = 2.
Appendix B.3. Four twist fields

In the case of four twist fields we obtain

3
4
2
+
+
h1 2 3 4 i = 0
1 1
z12 z13 z14
for 1 + 2 + 3 + 4 = 1 and

1 2 1 3 1 4
+
+
h1 2 3 4 i = 0
1 (1 1 )
z12
z13
z14
(B.26)
(B.27)
for 1 + 2 + 3 + 4 = 3. The four-point amplitudes, manifestly symmetric in the fields,

are thus
Q
i j 2
F1 ,2 ,3 ,4 i<j zij
, for 1 + 2 + 3 + 4 = 1,
(B.28)
h1 2 3 4 i =

2
Q
F1 ,2 ,3 ,4 i<j ziji j , for 1 + 2 + 3 + 4 = 3.

The last case, 1 + 2 + 3 + 4 = 2, requires a second order differential equation. Using
1 = 2 1 , 2 = 2 , 3 = 3 , 4 = 4 and 1 = 1 + 1 , 2 = 2 1, 3 = 3
1, 4 = 4 1 we obtain
536

2
3
4
2
+
+
+ 3 + 4
12 + 1
+ 1
1
z12 z13 z14
z12 z13 z14

2 2 3 3 4 4 2 3 + 2 3 2 4 + 2 4 3 4 + 3 4
+ 1 1
+
+
2
2
2
z12 z13
z14 z12
z13 z14
z12
z13
z14
h1 2 3 4 i = 0.
Mbius covariance implies the amplitude can be written as
h1 2 3 4 i
h h +h +h 2h h +h +h h h h h h 2
= z12 1 2 3 4 z13 3 z14 1 2 3 4 z241 2 3 4

2
x 3 4 (1 x)2 3 f (x, x)
1 2 3 4 1 4 2 3 1 3 2 4
2
= z12 z34 z14 z23 z13 z24 (z13 z24 )1 f (x, x),
(B.29)
where f (x, x)
satisfies the hypergeometric equation of type (3 , 1 1 , 3 + 4 ),

x(1 x)f 00 + 3 + 4 (2 + 23 + 4 )x f 0
3 (2 + 3 + 4 1)f = 0.
If no two j add up to an integer, a system of two linear independent solutions in the
vicinity of x = 0 is given by
f1(0) = 2 F1 (3 , 1 1 ; 3 + 4 ; x),
(0)
f2 = x 13 4 (1 x)123 2 F1 (1 , 1 3 ; 1 + 2 ; x).
(B.30)
Another set of solutions, adapted to x 1, is

f1(1) = 2 F1 (3 , 1 1 ; 2 + 3 ; 1 x),
(1)
f2 = x 13 4 (1 x)123 2 F1 (1 , 1 3 ; 1 + 4 ; 1 x),
(B.31)
while a set of solutions in the vicinity of x = is given by

f1() = x 3 2 F1 (3 , 1 4 ; 1 + 3 ; 1/x),
f2() = x 4 (1 x)12 3 2 F1 (4 , 1 3 ; 2 + 4 ; 1/x).
(B.32)
Monodromy invariance requires that the non-chiral solution is of the form, up to an overall
constant,

2
2
f (x, x)
= (3 , 4 )f1(0) (1 , 2 )f2(0)
2
2

= (2 , 3 )f1(1) (1 , 4 )f2(1)
() 2
() 2
= (1 , 3 )f1 (2 , 4 )f2 ,
where
(, 0 ) =
(0 )
(1 0 ) ()
= (, 1 0 ).
( + 0 ) (1 ) (1 0 )
Note that, if 1 + 2 + 3 + 4 = 2 we have
537
(1 1 , 1 2 ) (1 3 , 1 4 )
=
(3 , 4 )
(1 , 2 )
(1 ) (2 ) (3 ) (4 )
.
=
(1 1 ) (1 2 ) (1 3 ) (1 4 )
Thus the full four-point amplitude, manifestly symmetric in the fields, can be written as
h1 2 3 4 i
F1
1 ,2 ,3 ,4
q
2 4
2
(z13 z24 )1
= (3 , 4 )(1 , 2 )z121 2 z343 4 z141 4 z232 3 z131 3 z24

2
2 F1 (3 , 1 1 ; 3 + 4 ; x)
q

2
(1 , 2 )(3 , 4 )z121 2 z343 4 z141 4 z232 3 z131 3 z242 4 (z13 z24 )3

2
2 F1 (1 , 1 3 ; 1 + 2 ; x) ,
q
2 4
2
(z13 z24 )1
= (2 , 3 )(1 , 4 )z121 2 z343 4 z141 4 z232 3 z131 3 z24

2
2 F1 (3 , 1 1 ; 2 + 3 ; 1 x)
q
1 2 3 4 1 4 2 3 1 3 2 4
2
(1 , 4 )(2 , 3 )z12
z34 z14 z23 z13 z24 (z13 z24 )3

2
2 F1 (1 , 1 3 ; 1 + 4 ; 1 x) ,
q
1 2 3 4 1 4 2 3 1 3 2 4
2
z34 z14 z23 z13 z24 (z12 z34 )4
= (1 , 3 )(1 , 4 )z12

2
2 F1 (3 , 1 4 ; 1 + 3 ; 1/x)
q
1 2 3 4 1 4 2 3 1 3 2 4
2
(1 , 4 )(1 , 3 )z12
z34 z14 z23 z13 z24 (z13 z24 )3

2
2 F1 (4 , 1 3 ; 2 + 4 ; 1/x)
(B.33)
(B.34)
(B.35)
in the vicinity of x 0, 1 and , respectively. The constant F1 ,2 ,3 ,4 is independent of

the ordering of the indices.
If two j add up to an integer, 1 + 4 = 1, 2 + 3 = 1, 3 + 4 6= 1 say, we encounter a
degenerate case of the hypergeometric equation. We still have the same system of solutions
near x = 0,
(0)
f1 = 2 F1 (3 , 4 ; 3 + 4 ; x),
f2 = x 134 2 F1 (1 3 , 1 4 ; 2 3 4 ; x).
(0)
Their analytic continuation to x 1 is given by

(3 + 4 ) (1)
f
ln(1 x) 3 ,4 + f(1) ,
(3 ) (4 )

(2 3 4 )
(0)
f (1) ln(1 x) 13 ,14 + f(1) ,
f2 =
(1 3 ) (1 4 )
(0)
f1 =
where
f1 = 2 F1 (3 , 4 ; 1; 1 x),
538
f1 = M(3 , 4 ; 1; 1 x),
a,b = 2(1) (a) (b),
X
(a)n (b)n
[ha,n + hb,n hc,n h1,n ]x n ,
M(a, b; c; x) =
(c)n n!
n=1
ha,n = (a + n) (a),
and (x) = (x)0 / (x) is the digamma function. The monodromy invariant solution is
2
2

f (x, x)
= (3 , 4 )f1(0) (1 , 2 )f2(0)

2

= 2 F1 (3 , 4 ; 1; 1 x) ln |1 x|2 3 ,13 4 ,14

+ c.c. .
+ 2 F1 (3 , 4 ; 1; 1 x)M(3 , 4 ; 1; 1 x)
If 3 6= 4 , that is we have two different pairs of conjugate fields, the solution near x
is still given by (B.35). If 3 = 4 = then
2

f (x, x)
= |x|2 2 F1 (, 1 ; 1; 1/x) ln |x|2 2,1

+ c.c. .
+ 2 F1 (, 1 ; 1; 1/x)M(, 1 ; 1; 1/x)
One can deal similarly with the other cases where we have one or more pairs of conjugate
fields. The generic solution for all cases is given by (B.33) (B.35), which are replaced by
a degenerate solution in the following cases, respectively:
1 + 2 = 3 + 4 = 1:
h1 2 3 4 i = F1 ,2 ,3 ,4 |z12 |
21 2
|z34 |

22 3

z z
13 24
23 4 z14 z23
2

2 F1 (2 , 3 ; 1; x) ln |x|2 1 ,2 3 ,4

+ c.c. ,
+ 2 F1 (2 , 3 ; 1; x)M(2, 3 ; 1; x)
1 + 4 = 2 + 3 = 1:
h1 2 3 4 i = F1 ,2 ,3 ,4 |z14 |
21 4
|z23 |

23 4

z z
13 24
22 3 z12 z34
2

2 F1 (3 , 4 ; 1; 1 x) ln |1 x|2 1 ,4 2 ,3

+ c.c. ,
+ 2 F1 (3 , 4 ; 1; 1 x)M(3 , 4 ; 1; 1 x)
1 + 3 = 2 + 4 = 1:
h1 2 3 4 i = F1 ,2 ,3 ,4 |z13 |
21 3
|z24 |

22 3

z z
12 34
22 4 z14 z23
2

2 F1 (2 , 3 ; 1; 1/x) ln |x|2 1 ,3 2 ,4

+ c.c. .
+ 2 F1 (2 , 3 ; 1; 1/x)M(2, 3 ; 1; 1/x)
539
Appendix C. Character formulae

We derive here the characters of A in the C2N orbifold model. Realising C2N as the
2N th roots of unity we can use the Jacobi triple product identity to obtain
A (, u) = q
1
12
(1 + uq n )(1 + u1 q n )
n=1
2N1
X
m=0
um
for u 6= 1
2 ( ),
1 + u1 (2m+1)N,2N
for u = 1
( )2 ,
2N1
X
uk Ak/2N ( ),
k=0
where u C with u2N = 1. The Virasoro characters of the spaces Ak/2N can then be
obtained by
Ak/2N ( )
2N1
1 X kl
A (, l )
2N
l=0
"
2N1
X
1
k
2
(1) ( ) +
=
2N
m=0
"
N1
X
1
(1)k ( )2 +
=
2N
#
!
(mk)l
(2m+1)N,2N 2 ( )
1 + l
l=N+1
#
!
N1
ml
(m+1)l
X
kl +
(2m+1)N,2N 2 ( ) ,
1 + l
N1
X
m=0 l=N+1
where = exp(i/N) is the primitive 2N th root of unity. The coefficients can be evaluated
further as
N1
X
kl
l=N+1
N1
X
ml + (m+1)l
1 + l
l=N+1
m
X
m
X
kl
r=m
(1)r+m
r=m
(
=
(1)r+m rl
N1
X
(rk)l =
l=N+1
m
X
(1)r+m 2Nr,k (1)rk
r=m
(2N 2m 1), for |k| 6 m,
(1)
mk
(1)
mk+1
(2m + 1),
for |k| > m,
where we assumed |k| 6 N . Thus the characters can be written as
540
"
|k|1
X
(1)k
( )2 +
Ak/2N ( ) =
(1)l+1 (2l + 1)(2l+1)N,2N 2 ( )
2N
l=0
#
N1
X
l
(1) (2N 2l 1)(2l+1)N,2N 2 ( ) .
+
l=|k|
References
[1] H. Saleur, Polymers and percolation in two dimensions and twisted N = 2 supersymmetry,
Nucl. Phys. B 382 (1992) 486, hep-th/9111007.
[2] M.A. Flohr, On modular invariant partition functions of conformal field theories with
logarithmic operators, Int. J. Mod. Phys. A 11 (1996) 4147, hep-th/9509166.
[3] H.G. Kausch, Curiosities at c = 2, hep-th/9510149.
[4] M.J. Martins, B. Nienhuis, R. Rietman, An intersecting loop model as a solvable super spin
chain, Phys. Rev. Lett. 81 (1998) 504, cond-mat/9709051.
[5] S. Guruswamy, A. LeClair, A.W.W. Ludwig, gl(N|N) super-current algebras for disordered
Dirac fermions in two dimensions, Nucl. Phys. B 583 (2000) 475, preceding article in this
issue, cond-mat/9909143.
[6] M.R. Rahimi Tabar, S. Rouhani, The Alfven effect and conformal fieldtheory, Nuovo Cimento
B 112 (1997) 1079, hep-th/9507166.
[7] M. Flohr, Two-dimensional turbulence: yet another conformal field theory solution, Nucl. Phys.
B 482 (1996) 567, hep-th/9606130.
[8] M.R. Rahimi Tabar, S. Rouhani, A logarithmic conformal field theory solution for twodimensional magnetohydrodynamics in presence of the Alfven effect, Europhys. Lett. 37
(1997) 447, hep-th/9606143.
[9] V. Gurarie, M. Flohr, C. Nayak, The Haldane-Rezayi quantum hall state and conformal field
theory, Nucl. Phys. B 498 (1997) 513, cond-mat/9701212.
[10] I.I. Kogan, A.M. Tsvelik, Logarithmic operators in the theory of plateau transitions, hepth/9912143.
[11] L. Rozansky, H. Saleur, Quantum field theory for the multi-variable AlexanderConway
polynomial, Nucl. Phys. B 376 (1992) 461, hep-th/9203069.
[12] V. Gurarie, Logarithmic operators in conformal field theory, Nucl. Phys. B 410 (1993) 535,
hep-th/9303160.
[13] M.R. Gaberdiel, H.G. Kausch, Indecomposable fusion products, Nucl. Phys. B 477 (1996) 293,
hep-th/9604026.
[14] A. Bilal, I.I. Kogan, On gravitational dressing of 2D field theories in chiral gauge, Nucl. Phys.
B 449 (1995) 569, hep-th/9503209.
[15] J.-S. Caux, I.I. Kogan, A.M. Tsvelik, Logarithmic operators and hidden continuous symmetry
in critical disordered models, Nucl. Phys. B 466 (1996) 444, hep-th/9511134.
[16] J.-S. Caux, I.I. Kogan, A. Lewis, A.M. Tsvelik, Logarithmic operators and dynamical extension
of the symmetry group in the bosonic SU(2)0 and SUSY SU(2)2 WZNW models, Nucl. Phys.
B 489 (1997) 469, hep-th/9606138.
[17] Z. Maassarani, D. Serban, Non-unitary conformal field theory and logarithmic operators for
disordered systems, Nucl. Phys. B 489 (1997) 603, hep-th/9605062.
[18] J.-S. Caux, N. Taniguchi, A.M. Tsvelik, Disordered Dirac fermions: Multifractality termination
and logarithmic conformal field theories, Nucl. Phys. B 525 (1998) 671, cond-mat/9801055.
[19] V. Gurarie, A.W.W. Ludwig, Conformal algebras of 2d disordered systems, cond-mat/
9911392.
541
[20] M. Flohr, Singular vectors in logarithmic conformal field theories, Nucl. Phys. B 514 (1998)
523, hep-th/9707090.
[21] A. Shafiekhani, M.R. Rahimi Tabar, Logarithmic operators in conformal field theory and the
W -algebra, Int. J. Mod. Phys. A 12 (1997) 3723, hep-th/9604007.
[22] M.R. Rahimi Tabar, A. Aghamohammadi, M. Khorrami, The logarithmic conformal field
theories, Nucl. Phys. B 497 (1997) 555, hep-th/9610168.
[23] F. Rohsiepe, On reducible but indecomposable representations of the Virasoro algebra, preprint
BONN-TH-96-17, hep-th/9611160.
[24] M.R. Gaberdiel, H.G. Kausch, A rational logarithmic conformal field theory, Phys. Lett. B 386
(1996) 131, hep-th/9606050.
[25] Y. Zhu, Vertex operator algebras, elliptic functions and modular forms, J. Am. Math. Soc. 9
(1996) 237.
[26] M. Flohr, On fusion rules in logarithmic conformal field theories, Int. J. Mod. Phys. A 12 (1997)
1943, hep-th/9605151.
[27] M.R. Gaberdiel, H.G. Kausch, A local logarithmic conformal field theory, Nucl. Phys. B 538
(1999) 631, hep-th/9807091.
[28] M.R. Gaberdiel, P. Goddard, Axiomatic conformal field theory, Commun. Math. Phys. 209
(2000) 549, hep-th/9810019.
[29] D. Friedan, E. Martinec, S. Shenker, Conformal invariance, supersymmetry and string theory,
Nucl. Phys. B 271 (1986) 93.
[30] G. Felder, BRST approach to minimal models, Nucl Phys. B 317 (1989) 215, Nucl. Phys. B 324
(1989) 548 (Erratum).
[31] V.S. Dotsenko, V.A. Fateev, Conformal algebra and multipoint correlation functions in 2D
statistical models, Nucl. Phys. B 240 [FS12] (1984) 312.
[32] R. Dijkgraaf, C. Vafa, E. Verlinde, H. Verlinde, The operator algebra of orbifold models,
Commun. Math. Phys. 123 (1989) 485.
[33] P. Ginsparg, Curiositites at c = 1, Nucl. Phys. B 295 (1988) 153.
[34] C. Itzykson, J.-M. Drouffe, Statistical field theory, Cambridge University Press, 1989.
[35] P. Ginsparg, Curiosities at c = 1, Nucl. Phys. B 295 (1988) 153.
[36] H.G. Kausch, Extended conformal algebras generated by a multiplet of primary fields, Phys.
Lett. B 259 (1991) 448.

Anderson localization in bipartite lattices

Michele Fabrizio a,b , Claudio Castellani c
a Istituto Nazionale di Fisica della Materia, and International School for Advanced Studies (SISSA),
via Beirut 2-4, I-34014 Trieste, Italy

b International Center for Theoretical Physics (ICTP), Trieste, Italy
c Istituto Nazionale di Fisica della Materia and Universit degli Studi di Roma La Sapienza,
Piazzale Aldo Moro 2, I-00185, Roma, Italy

Received 4 April 2000; revised 10 May 2000; accepted 18 May 2000
Abstract
We study the localization properties of a disordered tight-binding Hamiltonian on a generic
bipartite lattice close to the band center. By means of a fermionic replica trick method, we derive
the effective non-linear -model describing the diffusive modes, which we analyse by using the
WilsonPolyakov renormalization group. In addition to the standard parameters which define the
non-linear -model, namely, the conductance and the external frequency, a new parameter enters,
which may be related to the fluctuations of the staggered density of states. We find that, when both
the regular hopping and the disorder only couple one sublattice to the other, the quantum corrections
to the Kubo conductivity vanish at the band center, thus implying the existence of delocalized states.
In two dimensions, the RG equations predict that the conductance flows to a finite value, while both
the density of states and the staggered density of states fluctuations diverge. In three dimensions, we
find that, sufficiently close to the band center, all states are extended, independently of the disorder
strength. We also discuss the role of various symmetry breaking terms, as a regular hopping between
same sublattices, or an on-site disorder. 2000 Elsevier Science B.V. All rights reserved.
PACS: 73.20.Jc; 73.20.Fz; 71.30.+h
1. Introduction
An interesting and still debated issue in the physics of the Andersons localization
concerns the existence of delocalized states in dimensions d 6 2, the conditions under
which they appear, and their properties. This problem, which is, for instance, of relevance
in the theory of the integer quantum Hall effect [1], got recently a renewed interest after
evidences of a metalinsulator transition in two dimensions have been discovered [2].
One of the cases in which localization does not occur in any dimension is at the band
center energy of a tight binding model on a bipartite lattice, when both the regular hopping
and the disorder only couple one sublattice to the other, i.e., in the so-called two sublattice
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 1 1 - 4
M. Fabrizio, C. Castellani / Nuclear Physics B 583 [FS] (2000) 542583
543
model [46]. Although this is not a common physical situation, its consequences are
surprising, and seem to escape any quasi-classical interpretation, which, on the contrary,
provides simple physical explanations of other delocalization mechanisms [3]. Already
in 1976, Theodorou and Cohen [4] realized that a one dimensional tight binding model
with nearest neighbor random hopping has a single delocalized state at the band center
(see also Ref. [5]). Afterwards, Wegner [6] and Oppermann and Wegner [7] showed that a
delocalized state indeed exists under the above conditions in any dimension, within a largen expansion, being n the number of orbitals per site. Later on, Wegner and Gade [8] argued
that these models correspond to a particular class of non-linear -models for matrices
in the zero replica limit. They were able to show that the quantum corrections to the function which controls the scaling behavior of the conductance vanish at the band center
at all orders in the disorder strength. thus implying a metallic behavior at this value of
the chemical potential. Moreover, they showed that, contrary to the standard case, the function of the density of states is finite. These results were based upon the non-linear
-model derived by Gade [9] by means of a boson-replica trick method, in a particular two
sublattice Hamiltonian with broken time reversal invariance.
More recently, similar models have got a renewed interest for their implications in
different physical contexts, for instance random hopping models in a cubic lattice with
-flux per plaquette, which are equivalent to a particular random Dirac-fermion model,
or models with non-Hermitian stochastic operators, or, finally, random flux models in two
dimensions (see, e.g., Refs. [1013]).
In this paper, we present an analysis of a generic disordered tight-binding Hamiltonian
on a generic bipartite lattice. The starting model is therefore of quite general validity,
also describing systems with time reversal invariance, and reduces in particular cases
to the model discussed by Gade [9], or, in the honeycomb lattice, to models of Dirac
fermions [10,13], or, finally, to random flux models [11,12]. By means of a fermionicreplica trick method, we derive the generic non-linear -model describing the diffusive
modes, which we analyse by the Renormalization Group (RG). Needless to say, the
effective model belongs to the same class of non-linear -models identified by Wegner
and Gade, demonstrating once more the universality of this description in the theory of
Anderson localization [1416].
Since the work is quite technical, we prefer to give in the following section a short
summary of the main results.
1.1. Summary of the main results
In this section we shortly present the main results, with particular emphasis to the
connections and differences with the standard theory of the Andersons localization.
We consider a generic bipartite lattice and work with a unit cell which contains two
sites from opposite sublattices. The Pauli matrices s act on the two components of the
wavefunction, corresponding to the two sites within each unit cell. In this lattice, we study
a disordered tight-binding Hamiltonian which has the peculiar property of involving only
both Pauli matrices 1 and 2 . In other words, H satisfies the conditions
544
{H, 3 } = 0,
{H, 1 } 6= 0,
{H, 2 } 6= 0,
(1)
where { , } indicates the anticommutator. By means of a path-integral approach within

a fermionic replica trick method, we find that the low-energy diffusive modes at the band
center, E = 0, can be represented by the non-linear -model
Z

2xx
E (R)2
E (R)2 U
dR Tr U
S[U ] =
16
Z

2
E (R)2 2 ,
(2)
dR Tr U (R)2 U
32 2
where U (R) is unitary and belongs to the coset space U(4m)/Sp(2m), being m the
number of replicas. At finite energy E 6= 0, the symmetry of U (R) gets reduced
to Sp(2m)/ Sp(m) Sp(m), as in the standard case [20]. The enlarged symmetry is
accompanied by new diffusive modes appearing in the retardedretarded and advanced
advanced channels, which are instead massive in the standard case. In Eq. (2) xx is the
Kubo conductivity (in units of e2 /h ) in the Drude approximation. We find that the new
coupling constant, , is proportional to the fluctuations of the staggered density of states,
i.e., to the following correlation function

1 X iq(RR 0 )
e
(3)
s (E, R) s (E, R 0 ) ,
V
0
RR
where the bar indicates the impurity average, and s (E, R) is the staggered density of
states at energy E,
X
n (R) 3 n (R)(E n ),
s (E, R) =
n
being n (R) the two-component eigenfunction of energy n .

The structure of the above action was derived by Gade and Wegner [8] for a particular
Hamiltonian. Here, we derive it for a generic bipartite lattice and random hopping.
Moreover, we provide a simple physical interpretation of .
Going back to (2), Gade and Wegner [8] gave a beautiful proof, based just on symmetry
considerations, that the quantum corrections to the -function of xx vanish in the zero
replica limit. In Appendix E, we outline how their proof works in our case, which is
slightly, but not qualitatively, different from the U(N)/SO(N) case they have considered.
Essentially, one can show that the action (2) possesses an invariant coupling xx + m ,
which, in the m 0 limit, implies that xx is not renormalized, apart from its bare
dimensions.
On the contrary, both the density of states and have non vanishing -functions.
In d = 1, the system flows to strong coupling, hence we can not access the asymptotic
infrared behavior. Nevertheless, the starting flow of the running variables indicates that the
density of states diverges. In d = 2, 3, the system flows to weak coupling, hence we can
safely assume that the infrared behavior is captured by the RG equations. Indeed, in two
dimensions, the density of states diverges at E = 0, while, in d = 3, it saturates to a finite
value, although exponentially increased in 1/xx . Moreover, has an anomalous behavior
in d = 2, where it is predicted to diverge logarithmically. We explicitly estimate how these
545
quantities behave as E 0, by means of a two-cutoff scaling approach, as discussed by

Gade [9].
We have also analyzed various symmetry breaking terms. The simplest ones are those
which spoil the particular symmetry Eq. (1) of the model at E = 0, i.e., an on-site disorder
or a same-sublattice regular hopping. These perturbations bring the symmetry of U (R)
down to Sp(2m)/ Sp(m) Sp(m), as in the standard localization problem. However, by
evaluating the anomalous dimensions of these terms, we can estimate the cross-over lengths
above which the symmetry reduction is effective. While in d = 1, 2 these terms always lead
to a localized behavior also at the band center, in d = 3 the vicinity to the band center leads
to an increase of the window in which delocalized states exist.
Finally, if the impurity potential breaks time-reversal symmetry, the matrix field U (R)
is shown to belong to the coset space U(2m), which indeed agrees with the analysis of
Gade [9].
The paper is organized as follows. In Section 2, we introduce the Hamiltonian. In
Section 3, we derive the path-integral representation of the model, by using Grassmann
variables within the replica trick method, and, in Section 4, we study the symmetry
properties of the action. In Section 5, we evaluate the saddle point of the action, while in
Sections 6, 7, 8 we derive the effective non-linear -model describing the long-wavelength
fluctuations around the saddle point. The Renormalization Group analysis is presented in
Section 9, and the behavior in the presence of on-site disorder, of a same-sublattice regular
hopping or with broken time reversal invariance is studied in Sections 10, 11, and 12,
respectively. Finally, Sections 13 and 14 are devoted to a discussion of the results. We have
also included several appendices containing more technical parts.
2. The model
We consider a tight binding Hamiltonian on a bipartite lattice, of the form
X X
hRR 0 cR
cR 0 + cR
H=
0 cR ,
(4)
RA R 0 B
where A and B label the two sublattices and the hopping matrix elements hRR 0 are
randomly distributed. We take a unit cell which includes two sites from different
sublattices. In some cases, like the honeycomb lattice, this is indeed the primitive unit
cell. In other cases, like the square lattice, it is not.
In this representation, the Hamiltonian can be written as
X
h12
(5)
H=
RR 0 c1R c2R 0 + H.c. ,
RR 0
where 1 and 2 label now the two sites in the unit cell, while R and R 0 refer to the unit cells,
21
and h12
RR 0 = hR 0 R . By introducing the two component operators

c1R
cR =
,
c2R
we can also write the Hamiltonian as
546
H=
X
R,R 0
cR
HRR 0 cR 0
X1

i 12
21
h12
hRR 0 h21
RR 0 + hRR 0 cR 1 cR 0 +
RR 0 cR 2 cR 0 ,
2
2
0
(6)
R,R
where the i s (i = 1, 2, 3) are the Pauli matrices. We notice that, quite generally, the
Hamiltonian involves both 1 and 2 , but neither 3 nor the unit matrix 0 , so that it
satisfies the conditions in Eq. (1).
12
12
12
We can write h12
RR 0 = tRR 0 + RR 0 , where tRR 0 are the average values, which represent
12 are random
the regular (translationally invariant) hopping matrix elements, while RR
0
variables with zero average, which we assume to be Gaussian distributed with width
D
E

12 2
12 2
.
= u2 tRR
RR
0
0
The dimensionless parameter u is a measure of the disorder strength in units of the regular
hopping. In this way, the Hamiltonian is written as the sum of a regular part, H (0) , plus a
disordered part, Himp .
For the regular hopping, we define

1 12
1 12
21
21
wRR 0 = tRR
tRR 0 = tRR
0 + tRR 0 ,
0 tRR 0 ,
2
2
so that the non disorderd part, H (0) , of the Hamiltonian is
X (0)
X
cR HRR 0 cR 0 =
cR (tRR 0 1 + iwRR 0 2 )cR 0 .
H (0) =
R,R 0
(7)
R,R 0
Since, for any lattice vector R0 , tR R 0 = tR+R0 R 0 +R0 , as well as wR R 0 = wR+R0 R 0 +R0 , and,
moreover, tRR 0 = tR 0 R while wRR 0 = wR 0 R , the Fourier transforms satisfy tk = tk (tk
real) and wk = wk (wk imaginary). In momentum space, the Hamiltonian matrix, Hk(0) =
tk 1 + iwk 2 , is diagonalized by the unitary transformation ck = Uk dk , with
Uk = ei 4 2 ei 2 1 ,
(8)
where
tan k =
Im tk12
iwk
=
.
tk
Re tk12
(0)
Indeed, Uk Hk Uk = k 3 , where k2 = tk2 wk2 = |tk |2 + |wk |2 .

2.1. Current operator
cR with the non-disordered Hamiltonian (7) is

The commutator of the density cR
X
cR (tRR 0 1 + iwRR 0 2 )cR 0 cR

0 (tR 0 R 1 + iwRR 0 2 )cR ,
R0
or, in Fourier space,

X
(tk+q tk )ck 1 ck+q + i(wk+q wk )ck 2 ck+q .
k
(9)
547
Therefore, in the long wavelength limit, the current operator in the absence of disorder is
X
E k wk c 2 ck+q
E k tk c 1 ck+q + i
JEq(0) '
k
k
k
E k k c BEk E ck+q + k
E k c BE,k E ck+q
k
k
(10)
E k d 3 dk+q + k
E k d 2 dk+q ,

k
k
(11)
X
k
the last being the expression in the basis which diagonalizes the Hamiltonian H (0) . In (10),
E = (1 , 2 , 3 ), and the vectors
BEk = (cos k , sin k , 0),
BE,k = ( sin k , cos k , 0),
(12)
describe intra- and inter-band contributions to the current vertex. Notice that the regular
P
hopping Hamiltonian can be simply written as H (0) = k k ck BEk E ck .
Moreover, since also the impurity part of the whole Hamiltonian, (5), does not commute
with the density operator, in the disordered model the current operator acquires an
additional term proportional to the random hopping matrix elements, which we discuss
later.
3. Path integral
The starting point of our analysis is a path-integral representation of the generating
functional, in terms of Grassmann variables, following the work by Efetov, Larkin and
Khmelnitskii [20]. To this end, we introduce, for each unit cell R, the Grassmann variables
cR;a,p, and their complex conjugates cR;a,p, , where a = 1, 2 is the sublattice index, p =
is the index of the advanced (+) and retarded () components, and the index runs
over m identical copies of the model, as in the usual replica trick method. In what follows,
by convention, the Pauli matrices s act in the two sublattice space, the s in the space
of the Grassmann fields c and c, and the ss in the space.
In order to treat on equal footing both the particlehole and the particleparticle diffusive
modes (diffusons and cooperons, respectively), as implied by time-reversal invariance, it is
convenient to introduce the Nambu spinors R and R defined through

1
cR
,
R =
2 cR
where cR and cR are column vectors with components cR;a,p, and cR;a,p, , respectively,
and
R = [c ]t ,
where c = i2 is the charge conjugation operator. The action in terms of the spinors is
[see Eq. (6)]
548

R E + i s3 HRR 0 R 0
2
RR 0

X
X
(0)
=
R E + i s3 HRR 0 R 0 +
R Himp,RR 0 R 0
2
0
0
S=
RR
RR
= S0 + Simp ,
(13)
where E i/2 are the complex energies of the advanced/retarded components.

3.1. Disorder average
Before taking the disorder average, we notice that, in the spinor notation,
c2R 0 + H.c. 2 1R 2R 0 = 2 2R 0 1R .
c1R
Therefore, the impurity part of the action can be written as

X
12
2RR
Simp =
0 1R 2R 0 .
R,R 0
The generating functional, within the replica method, is

Z
Z = D D D P [ ] eS0 Simp ,
(14)
12 . The average
where P [ ] is the Gaussian probability distribution of the random bonds RR
0
over disorder changes the impurity action into
X

2
12 2
2u2 tRR
1R 2R 0
Simp =
0
R,R 0
12
2u2 tRR
0
2
1R 2R 0

2R 0 1R .
(15)
R,R 0
We define
12
WRR 0 = 2u2 tRR
0
2
Re,
so that Wq = Wq , and introduce
X1R = 1R 1R ,
where is a multilabel for Nambu, advanced/retarded and replica components, and
analogously X2R , as well as their Fourier transforms. By these definitions,

1 XX

Wq X1,q X2,q .
(16)
Simp =
V q
,
This form, as compared to (15), has the advantage to allow a simple HubbardStratonovich
transformation. Notice that the use of Nambu spinors has the great advantage to involve
just a single Fourier component of WRR 0 . If we write
Wq = q eiq ,
549
where q > 0 and q = q , and define

Y1q = ei
q
2
X1q ,
Y2q = ei
q
2
X2q .
Simp takes the simple form

Simp =
=

1 X
1 X q

Tr Y1,q Y2,q + Y2,q Y1,q
q Y1,q Y2,q =
V q
V q 2

1 X q
Tr Y0,q Y0,q Y3,q Y3,q ,
V q 4
(17)
where we have introduced Y0 = Y1 + Y2 as well as Y3 = Y1 Y2 . Moreover, our choice

of the impurity potential, which does not break, on average, the spatial symmetries of the
lattice, implies that q = q , see Eq. (9).
We notice that, if a term is written as

A2
A2 X
Tr X2 ,
X X =
4
4
where X = X and = 1, one can always decouple it, by introducing an Hermitian

matrix Q, by the following HubbardStratonovich transformation

Z

A2
2
Tr X
(18)
= N DQ exp A2 Tr Q2 + Tr QXt ,
exp
4
R
where the normalization factor N 1 = DQ exp[A2 Tr(Q2 )]. In the specific example,
(17) can be transformed into

1 X 1
i X
t
t
Tr Q0q Q0q + Q3q Q3q
Tr Q0q Y0q
+ iQ3q Y3q
Simp =
V q q
V q
=

1 X 1
Tr Q0q Q0q + Q3q Q3q
V q q

q t
q t
i X
X0q + i sin
X3q
Tr Q0q cos
V q
2
2

q t
q t
i X
X3q + i sin
X0q .
i Tr Q3q cos
V q
2
2
(19)
If we define Qq = Q0q 0 + iQ3q 3 , we obtain

Simp =
=

i
1 X 1
i X
Tr Qq Qq
p Qq e 2 q 3 p+q
V q 2q
V p,q
(20)
X

1 X 1
Tr Qq Qq i
R QR R
V q 2q
R

i
i X
+
p 1 e 2 q 3 Qq p+q ,
V p,q
(21)
550
where the last term vanishes at q = 0. Notice that the two-sublattice symmetry properties
of the Hamiltonian, see Eq. (1), are reflected in the particular form of the Q-matrix,
which only contains the Pauli matrices 0 and 3 . In particular, since the tensor Q(R)
(R) (R), 0 selects the uniform component while 3 the staggered component of the
product of the two Grassmann fields. Notice that the electron-Q coupling in the R basis
can be written as a local coupling but for the last term in (21). To simplify the notation of
this term we find it useful to define an operator L through

i
1 X iqR
e
(22)
1 e 2 q 3 Qq .
LQ(R)
=
V q
If we now transform the spinors in Eq. (20) to the diagonal basis, the coupling to the Qq
i
matrix transforms as Up+q

Qq e 2 q 3 Up . The simplest consequence is that, in the diagonal
basis, Qk1 ,k2 = Q0,k1 ,k2 0 + iQ1,k1 ,k2 1 depends on two wavevectors. However, in the case
of cubic lattices, see Eq. (A.1),
i
Qq e 2 q 3 Up+q = e 2 p+q 1 (Q0,q iQ3,q 1 ) e 2 p 1 e 2 q 1

Up+q
i
= e 2 (q +q )1 (Q0,q iQ3,q 1 ) = Q0,q iQ3,q 1 ,

since q = q . Therefore, for cubic lattices, we can also write
X

1 X 1
Tr Qq Qq i
R Q(R)R ,
Simp =
V q 2q
(23)
where now R is the Grassmann field of the dR operators, and the matrix Q(R) =
Q0 (R)0 + iQ1 (R)1 , with Q1 (R) = Q3 (R) defined above.
Finally, we notice that, in the case of the honeycomb lattice, at the wavevector q ,
connecting the two Dirac cones of the non-disordered dispersion band, q = 0. This
observation will turn useful when discussing the long-wavelength behavior of the model.
4. Symmetries
The action (13), at E = = 0, i.e., at the band center with zero complex frequency, is
invariant under a transformation R T R if
cT t ct HRR 0 T = HRR 0 .
Since the random matrix elements HRR 0 involve both Pauli matrices 1 and 2 , T has to
satisfy at the same time cT t ct 1 T = 1 and cT t ct 2 T = 2 . This implies that
cT t ct = 1 T 1 1
1 T
1 = 2 T
2 .
The condition (25) can be fulfilled only by a transformation T = T0 0 + T3 3 .

Under such a transformation
Q cT t ct QT = 1 T 1 1 QT 2 T 1 2 QT .
(24)
(25)
551
Since 1 Q1 = Q , then

T 1 1 QT 1 = T Q 1 T 1 1 = T 1 Q T 1 1 .
Hence, the transformation is also unitary, T = T 1 . Moreover, such a transformation

leaves the Q-manifold invariant, which implies that our HubbardStratonovich decoupling
scheme, which makes use of Q = Q0 0 + iQ3 3 , is exhaustive.
The unitary transformation, T , can be written as

W3
W0
0 +
3 ,
(26)
T = exp
2
2
where
W0 = W0 ,
W3 = W3 .
In addition, we must impose the charge conjugacy invariance, which, through Eq. (24),
implies that
cW0t ct = W0 = W0 ,
cW3t ct = W3 = W3 .
The number of independent parameters turns out to be 16m2 , which suggests that T is
related to a unitary group, specifically U(4m), as argued by Gade and Wegner [8]. In fact,
we can alternatively write
!
W0 +W3

U
0
0
e 2
,
(27)
T=
W0 W3
0 c U ct
0
e 2
where U is indeed a unitary transformation belonging to U(4m). The invariance of (13)
at finite frequency, 6= 0, implies the additional condition cT t ct s3 T = s3 , which reduces
the number of independent parameters to 8m2 + 2m, lowering the symmetry of U down to
Sp(2m).
If E 6= 0, T has to satisfy also
cT t ct T = 1 T 1 1 T = 1.
This implies that, at finite energy, i.e., away from the band center, T does not contain
anymore a 3 -component. Indeed, E lowers the symmetry of U down to Sp(2m), which is
further reduced to Sp(m) Sp(m) by a finite frequency, as in the standard situation [20].
5. Saddle point
The full action

X
i
i
(0)
k Eq0 + i s3 q0 Hk q0 + Qq e 2 q 3 k+q
S=
2
V
k,q

1 X 1
Tr Qq Qq ,
V q 2q
(28)
552
by integrating over the Nambu spinors, transforms into

1
1 X 1
(0)
Tr Qq Qq Tr ln E + i s3 H + iQ iLQ .
S[Q] =
V q 2q
2
2
The saddle point equation for homogeneous solutions at E = 0 and = 0+ is
Z 2
1
1
d k +
0
+ i0+ s3 + k + iQ ,
Q=i
i0 s3 k + iQ
2
4
4
P
12 )2 . The general solution is
where 0 = RR 0 2u2 (tRR
0
(29)
0 (0)s3 s3 ,
4
with (0) being the density of states at E = 0. In order to distinguish transverse from
longitudinal modes, it is convenient to parametrize the Q-matrix in the following way

Q(R)P = 1 T (R)1 1 Qsp + P (R) T (R)
Qsp =
Q(R) + 1 T (R)1 1 P (R)T (R).
(30)
Here P (R) describes the longitudinal modes, which we discuss more in detail in Section 7,
and T the transverse modes. Namely, T has the form given in Eq. (27),

W0 (R)
W3 (R)
W (R)
= exp
0 +
3 ,
(31)
T (R) = exp
2
2
2
with exp[(W0 + W3 )/2] belonging now to the coset space U(4m)/Sp(2m). This amounts
to impose that
{W0 , s3 } = 0,
[W3 , s3 ] = 0,
by which it derives that

Q(R) = 1 T (R)1 1 Qsp T (R) = Qsp eW0 (R)0+W3 (R)3 .
In the space, we can write

0
B
iA 0
,
W3 =
,
W0 =
0 iC
B 0
(32)
(33)
where A = A, C = C, and additionally, since cW t ct = 1 W 1 = 1 W 1 , then

cAt ct = A, cC t ct = C and cB t ct = B . By writing
A = A0 0 + i(A1 1 + A2 2 + A3 3 ),
(34)
B = B0 0 + i(B1 1 + B2 2 + B3 3 ),
(35)
C = C0 0 + i(C1 1 + C2 2 + C3 3 ),
(36)
we find that the above conditions imply that, for i = 0, . . . , 3,

Bi , Ai , Ci Re,
(37)
A0 = At0 ,
(38)
and
C0 = C0t ,
553
while, for j = 1, 2, 3,
Aj = Atj ,
Cj = Cjt .
(39)
6. Effective action
In this section, we derive the effective field theory describing the long wavelength
transverse fluctuations of Q(R) around the saddle point. In the case of honeycomb lattices,
we should worry about the momentum component of Q which couples the two Dirac cones.
However, one can see that the free action of Q diverges at this wavevector, so that we are
allowed to ignore the fluctuations around this momentum. For cubic lattices, we notice
that, since we use a doubled unit cell, the long-wavelength fluctuations of Q include both
uniform and staggered components.
6.1. Integration over the Grassmann fields
As we said, by integrating (28) over the Grassmann variables, we obtain the following
action of Q:

1
1 X 1
(0)
Tr Qq Qq + Tr ln E + i s3 H + iQ iLQ .
S[Q] =
V q 2q
2
2
We start by neglecting the longitudinal fluctuations. Then, since Q = T Qsp T , where we
define T = 1 T 1 2 T 2 , we can rewrite the second term of S[Q] as

1
Tr ln E T T + i T s3 T T H (0)T + iQsp V ,
(40)
2
2
where we define
.
V = iT LQT
(0)
Since HRR 0 involves either 1 and 2 , while T involves 0 and 3 , then

(0)
(0)
(0)
(0)
HRR 0 T (R 0 ) = T (R 0 ) HRR 0 = T (R) HRR 0 + T (R 0 ) T (R) HRR 0

(0)
E
E E 0 (0)
' T (R) HRR
0 T (R) R R HRR 0

(0)
1
+ ij T (R) Ri Ri0 Rj Rj0 HRR 0 .
2
Therefore the term T H (0)T which appears in (40) can be written at long wavelengths as
(0)
(0)
E T 1 (R) JE(0) (R 0 )RR 0
T (R)HRR 0 T (R 0 ) = HRR 0 iT (R)

(0)
1
+ T (R)ij T (R)1 Ri Ri0 Rj Rj0 HRR
0
2
(0)
HRR
0 + URR 0 ,
(41)
554
where we use the fact that the long-wavelength part of the current operator in real space is,
see (B.3),
X

HRR 0 cR 0 .
RE RE 0 cR
JE(R 0 ) = i
R
Notice that in (41) only the current vertex which derives from the regular hopping appears,
which is not the full current operator.
Moreover, a further current-like coupling will arise from the expansion in V (see below).
see (22), can be approximately
To this end, in the long-wavelength limit, the operator L,
written as

1
E 2 Q(R),
E
E
LQ(R)
' E Q(R)
3
2
E
where E = (q
= 0)/2.
Having defined U , (40) can be written as

1
(0)
Tr ln E T T + i T s3 T U V H + iQsp
2
2

1

1
= Tr ln G + Tr ln 1 + GE T T + G i T s3 T GU GV ,
2
2
2
(42)
(43)
where G = (H (0) + iQsp )1 is the Greens function in the absence of transverse

fluctuations.
The effective field theory is then derived by expanding S[Q] up to second order in U
and V , and first order in E and . In this way we get the following terms.
6.2. Expansion in the Q free action
The free part of the action
S0 [Q] =

1 X 1
Tr Qq Qq ,
V q 2q
can be expanded at small q. Since q = q , then

q ' 0 (1 q 2 ),
leading to
1
S0 [Q] '
20
1
=
20
Z
Z

dR Tr Q(R)Q(R) +

dR Tr Q2sp +
20

X 2
q Tr Qq Qq
2V 0 q

E
E
.
dR Tr Q(R)
Q(R)
(44)
The second term is a contribution to the currentcurrent correlation function of the part of
the current vertices proportional to the random hopping.
555
6.3. Expansion in E
Expansion of (43) in E gives

E
E
E
Tr GT T = i
Tr Qsp T T = i
Tr Q.
2
0
0
(45)
6.4. Expansion in
Expansion in gives
Tr(s3 Q).
i Tr GT s3 T =
4
20
(46)
6.5. Expansion in U
The second order expansion in U contains the following terms:
1
Tr(GU ),
2
(47)
and
1
Tr(GU GU ).
4
Taking in (47), the component of U containing second derivatives, we get

(0)
1
Tr T (R)ij T (R)1 Ri Ri0 Rj Rj0 HRR 0 G(R 0 , R) .
4
By means of the Ward identity (B.5), the above expression turns out to be
ij++

Tr T (R)ij T (R)1 ,
8
which, integrating by part, is also equal to
ij++
8
(48)
(49)

Tr T (R)i T (R)1 T (R)j T (R)1

1 ++
Tr Di Dj Di s3 1 Dj s3 1
16 ij

1
ij++ Tr Di Dj + Di s3 1 Dj s3 1 .
16
E
Here we have introduced a matrix D(R)
with the ith component
=
Di (R) = D0,i (R)0 + D3,i (R)3 T (R)i T (R)1 .

The part of (47) which contains first derivatives gives rise to a boundary term
Z
X

k2
1
E
E (R)3 ,
k 2
dR Tr W
2
V
k +
k
(50)
(51)
(52)
(53)
where has been defined by Eq. (29), which we discard by taking appropriate boundary
conditions.
556
Let us now analyse the term (48), where we have to keep of U only the part containing
first derivatives. By making use of (41), this term is, in momentum space,
o

1 X n E 1 E (0)
E T 1
E (0) G(k)
Tr T T
J
G(k
+
q)
T
J
k+q
k
q
q
4
kq
o

1 X n E 1 E (0)
E T 1
E (0)G(k)
Tr T T
J
G(k)
T
J
'
k
k
q
q
4
kq

1 XX E
(0)
(0)
E
Tr D(R) JEk G(k)D(R)
JEk G(k) ,
=
4
k
(54)
E k tk 1 + i
E k wk 2 . The non vanishing terms
valid for small q. Here the matrix JEk(0) =
E
E
E
in (54) have both Ds either D0 0 or D3 3 [see (52)].
In the diagonal basis, upon defining, as we did in Eq. (29), Qsp = s3 , with =
0 (0)/4, the Greens function is
k
1
= i 2
0 s3 2
3 s0
2
k 3 + is3
k +
k + 2
G0 (k)0 s3 + G3 (k)3 s0 .
G(k) =
(55)
Going back to the original basis,

G(k) = Uk G(k)Uk = G0 (k)0 s3 + G3 (k)BEk E s0 ,
(56)
where BEk has been defined in (12). Therefore,

3 G(k)3 = G0 (k)0 s3 G3 (k)BEk E s0 = s1 G(k)s1 .
(57)
By means of (57), we find that

3 JEk(0) G(k)+ 3 = JEk(0)3 G(k)+ 3 = JEk(0)G(k) ,
(58)
from which it derives that (54) can be written as the sum of two different terms

1 +
ij Tr Di Dj Di s3 1 Dj s3 1
(59)
16

1 ++
Tr Di Dj + Di s3 1 Dj s3 1 .
(60)
+
16 ij
By summing (59), (60), (50) and (51), we get

1 +
ij ij++ Tr Di Dj Di s3 1 Dj s3 1 ,
16
which is equal to
Z

2
(0)
Q(R)
Q(R)
,
(61)
dR
Tr
i
j
ij
32 2
where

1
(0)
+ ij++
ij =
2 ij
is the Kubo conductivity with the current vertices which involve only the regular hopping
[cf. Eq. (B.4)].
557
6.6. Expansion in V
The expansion in V up to second order, gives two terms
1
Tr(GV ),
2
(62)
and
1
Tr(GV GV ).
4
In addition, we must also consider the mixed term
1
Tr(GU GV ).
2
The first order term, (62), is
i
1
Tr(Qsp V )
Tr(GV ) =
2
0
Z

1
E
dR Tr Q(R) E Q(R)
=
3
0
Z

1
E E
E
Q(R) .
dR Tr E Q(R)
20
(63)
(64)
(65)
(66)
The first term, (65), is another boundary term, which we neglect. The second order term,
Eq. (63), gives
Z

1 X
E E
E
G0 (k)2 + G3 (k)2
dR Tr E Q(R)
Q(R) .
(67)
4V
k
Notice that the saddle point equation implies that

2
1 X
G3 (k)2 G0 (k)2 =
.
V
0
k
By using the above equation in (66), we find that (67) plus (66) give
Z

1 X
E E
E
Q(R)
G3 (k)2 dR Tr E Q(R)
2V
k
Z

1 X k2
E E
E
Q(R) .
dR Tr E Q(R)
=
2
2V
k + 2
k
(68)
The contribution of the mixing term (64), can be evaluated in a similar way, giving
Z

1 X k2
E E
E
E k Q(R)
Q(R) .
(69)
dR Tr
2
2
2V
k +
k
We notice that, since q = q , then (68) and (69) can be included in (61) if the following
redefinition of the current vertex is assumed:

E k BEk E + k
E 0 BE,k E .
E k
(70)
JEk(0) =
558
Therefore, the Kubo conductivity which appears in (61) has to be calculated with the above
expression of the regular current vertex. This has notable consequences. First of all, for cubic lattices, the interband contribution vanishes, hence the Kubo conductivity coincides
with the true one, since the enlargment of the unit cell was artificial. This is compatible
with (23), where we showed that the impurity action is local in the basis which diagonalizes
the regular hopping, implying that the regular current vertex contains only the intraband
operator.
To conclude, the effective action so far derived is therefore
Z

X 2
1
q Tr Qq Qq
dR Tr Q(R)Q(R) +
S[Q] =
20
2V 0 q
Z

2
(0)
ij
dR Tr i Q(R)j Q(R)
+
2
32
Z

E
Tr Q(R)
Tr s3 Q(R) .
(71)
+ dR i
0
20
7. Longitudinal fluctuations
The full expression of the Q-matrix we must indeed consider is the one given by (30),

Q(R)P = T (R)1 Qsp + P (R) T (R) Q(R) + T (R)1 P (R)T (R)
Q(R) + S(R),
(72)
where T (R) involves transverse fluctuations and

P (R) = (P00 s0 + P03 s3 )0 + i(P31 s1 + P32 s2 )3 ,
(73)
being all P s hermitean. Charge conjugation implies that cP t ct = P . The field P (R) takes
into account longitudinal fluctuations which are massive. Let us define (R R 0 ) the
Fourier transform of q1 . Then, the free action of the QP field is
Z

1
dR dR 0 (R R 0 ) Tr QP (R)QP (R 0 )
S[QP ] =
2
Z

1
(74)
dR Tr QP (R)QP (R)
=
20
Z
1
dR dR 0 (R R 0 )
4

Tr QP (R) QP (R 0 ) QP (R) QP (R 0 ) .
(75)
Since QQ = Q2sp , (74) gives
Z

1
dR Tr P (R)P (R) + 2Qsp P (R) + Q2sp .
20
The second term cancels with the first order expansion of Tr ln GP , since Qsp is the saddle
point solution. What is left, i.e.,
Z

1
(76)
dR Tr P (R)P (R) ,
20
559
is actually the mass term of the longitudinal modes, since the second order expansion in P
of Tr ln GP is zero. The other term, (75), can be analysed within a gradient expansion of
E P (R) (RE RE0 ) + . The details of the calculations are given in
QP (R) QP (R 0 ) = Q
Appendix C.1, so that, in this section, we just present the final results.
The free action of the longitudinal fields is found to be

1 X 1
Tr Pq Pq .
(77)
S0 [P ] =
V q 2q
Here, we neglect the contribution of the invariant measure, which, in the zero replica limit,
gives rise to fluctuations smaller by a factor u2 than Eq. (77). The integration over P with
the above action has several important consequences for the action of the transverse modes
(see Appendix C.1).
Firstly, all the terms of the Kubo conductivity with the random current vertices are
recovered. In addition, we find a new operator
Z

2
E
E
dR Tr Q (R)Q(R)
(78)
3 Tr Q (R)Q(R)
3 ,
4
8 32
which has contributions from two different terms. The first one is obtained by expanding
each Greens function in (48) at first order, GP = G0 iG0 P G0 , and the second is derived
by (75). They are analogous to the components of the Kubo conductivity with regular and
with random current vertices, respectively.
8. Effective non-linear -model
In conclusion, the final expression of the action of the transverse modes in the longwavelength limit is
Z

2
E
E
dR Tr Q(R)
Q(R)
S[Q] =
xx
2
32
Z

E
Tr Q(R)
Tr s3 Q(R)
+ dR i
0 Z
20

2
E
E
dR Tr Q (R)Q(R)
(79)
3 Tr Q (R)Q(R)
3 ,
8 32 4
where we make use of the fact that, in the models we consider, ij = ij xx . 1 Since
Q(R) = Qsp T (R)2 , see Eq. (32), expressing T (R) by means of U (R) as in Eq. (27), the
action at E = = 0 can also be written as
Z

2xx
E (R)2
E (R)2 U
dR Tr U
S[U ] =
16
Z
n
o2
2
E (R)2 ,
(80)
dR Tr U (R)2 U
32 2
1 In two dimensions, if the Hamiltonian breaks inversion symmetry, a topological term proportional to
i
i
h
dR Tr Q(R) Q(R) 3 ,
may appear, where is the Levi-Civita tensor.
560
as anticipated in the Section 1.1. As compared to the non-linear -model which is obtained
in the absence of the particlehole symmetry [20], the above action differs first of all
because of the symmetry properties of the matrix field U (R), which describes now the
Goldstone modes within the coset space U(4m)/Sp(2m). Moreover, it also differs for the
last term of (79), which, in the general case, even if present, is not related to massless
modes. An analogous term was originally obtained by Gade [9] in a two sublattice model
described by two on-site levels with a regular hopping of the form HRR 0 = tRR 0 1 , and a
local time-reversal symmetry breaking random potential Himp,R = w1R 1 + w2R 2 . We
discuss the broken time reversal symmetry case in Section 12.
Although the action may be parametrized by a simple unitary field U (R), as in (80), we
prefer to work with the matrix Q(R) which has the more transparent physical interpretation
Q(R) (R) (R).
Finally, it is important to notice that either xx and have contributions from both
the regular and the random current vertices. This implies that, even in the limit of strong
disorder, in which the average hopping is negligible with respect to its fluctuations, these
constants are finite and become of order unity [7,11].
8.1. Gaussian propagators
At second order in W , the dispersion term
Z
Z

2xx
2xx
E
E W
E
E
dR Tr W
dR Tr Q Q '
32 2
32
Z

2xx
E A
E + C
E C
E ,
E B
E + 2 Tr A
dR4 Tr B
=
32
where A, B and C are defined through Eqs. (34), (35) and (36). For the Bs we find the
quadratic action
4
xx X X X 2
k Bi,ab (k)Bi,ab (k),
2
i=0 k
so that
ab

Bi,ab (k)Bj,cd (k) = ij ac bd D(k),
(81)
where
D(k) =
1 1
.
xx k 2
For the As we have to take into account also the disconnected term:
Z

2
E
E
dR Tr Q (R)Q(R)
3 Tr Q (R)Q(R)
3
4
32 8
Z

2
E 3
E 3 Tr W
dR Tr W
'
64
Z

2
E + C
E
E + C
E
dR Tr A
Tr A
=
64
(82)
2
16
561

E 0 Tr A
E 0 .
E 0 + C
E 0 + C
dR Tr A
The non vanishing propagators are

A0,ab (k)A0,cd (k) = D(k) ac bd + ad bc D(k)
ab cd ,
xx + m

ab cd ,
C0,ab (k)C0,cd (k) = D(k) ac bd + ad bc D(k)
xx + m

ab cd ,
A0,ab (k)C0,cd (k) = D(k)
xx + m
where m is the number of replicas, while, for i = 1, 2, 3,

Ai,ab (k)Ai,cd (k) = D(k) ac bd ad bc ,

Ci,ab (k)Ci,cd (k) = D(k) ac bd ad bc .
(83)
(84)
(85)
(86)
(87)
Notice that the particular symmetry of the two-sublattice model leads to additional
diffusive modes in the retardedretarded and advancedadvanced channels, which are not
massless in the standard case [20].
8.2. Physical meaning of
Let us introduce an external source which couples to the staggered density of states,
which is accomplished by adding to the action a term
Z
dR R s3 3 (R)
R,
where, in the replica space, the source , = , . The fluctuations of the staggered
density of states is obtained by the derivative of the partition function with respect, for
instance, to and , with 6= . Inserting the source term in the action, and integrating
over the Grassmann fields after introducing the matrix Q, leads to the following expression
of the staggered density of states fluctuation in terms of Q:
F (R, R 0 ) =

1

Tr Q (R)s3 3 Tr Q (R 0 )s3 3 .
2 02
(88)
The Gaussian estimate of the above correlation function at momentum k is given by

F (k) = 16
=

2

A0, (k) + C0, (k) A0, (k) + C0, (k)
2
2
0
64 2
.
D(k)
2
2
xx + m
0
(89)
Therefore, is directly related to the singular behavior of the staggered density of states
fluctuations.
562
9. Renormalization group
In this section, we will apply the WilsonPolyakov Renormalization Group (RG)
procedure [17,18,20] to analyse the scaling behavior of the action. Indeed, some of the
calculations which we present are redumdant, given the proof by Gade and Wegner
that the -function is zero [8] (see Appendix E). Nevertheless, other results besides the
conductance -function are important, so that we describe the whole RG procedure.
9.1. RG equations
In the spirit of WilsonPolyakov RG approach [17,18,20], we assume that
T (R) = Tf (R)Ts (R),
where Tf involves fast modes with momentum q [/s, ], while Ts involves slow
modes with q [0, /s], being the high momentum cut-off, and the rescaling factor
s > 1. Within an -expansion, where = d 2, we define
Z
/s
1
dkE
D(k) L =
ln s + O().
(2)d
4 2 xx
It is straightforward to show the following result

E f + 2 Tr D
E s 1 Qf D
E
E Q
E s Q 1
E Q
= Tr Q
Tr Q
f
f

E s + 4 Tr D
E f ,
EsD
E s Q Q
2 2 Tr D
f
(90)
E s = Ts T
E s . Moreover,
where Qf = Tf Qsp Tf and D

1
E
E
Tr Q Q
3 Tr Q Q
3
4

E f 3 Tr W
E f 3 .
E s + W
E s + W
= Tr W
(91)
Since the fast and slow modes live in disconnected regions of momentum space, only the
stiffness term (90) generates corrections. By expanding the terms coupling slow and fast
modes up to second order in Wf , the stiffness generates an action term for the slow modes
which, after averaging over the fast ones, is
Z

2xx
E s + hF1 if 1 F22 ,
E s Q
(92)
dR Tr Q
f
2
2
32
where
Z

2xx
E f Qsp 1
E 1 Qsp Wf DW
dR 2 Tr D
F1 =
2
32

E sp W f2 1 ,
E 1 Qsp DQ
+ 2 Tr D
and
F2 = 4
2xx
32

E f W
E f .
dR Tr DW
(93)
(94)
563
The explicit evaluation of these terms is outlined in Appendix D. Here we just give the
final results. The Kubo conductivity is renormalized according to
xx (1 4Lm)xx ,
(95)
while the factor

+ 4Lxx .
(96)
For what it regards the renormalization of E and , we notice that

Q = Qsp Ts Tf2 Ts
1
Qsp Ts Wf2 Ts .
2
Since the slow and fast degrees of freedom are defined in different regions of momentum
space, only the last term is relevant. By means of Eq. (D.15), we find that

1
2 8m +
L .
(97)
Q Qs 1 +
2
xx + m
' Qsp Ts2 + Qsp Ts Wf Ts +
This leads to similar corrections to E and , which will have the same scaling behavior.
Finally, to describe the cross-over behavior in the presence of symmetry breaking terms,

we also need the scaling behavior of the operator Tr Q(R)2 . We find that

Tr Q2 f = Tr Qsp Ts Tf2 Ts Qsp Ts Tf2 Ts f

= 1 + 2 2 4m +
(98)
L Tr Q2s L(Tr Qs )2 .
xx + m
To implement the RG, we have to rescale the momenta in order to recover the original
range [0, ]. This is accomplished by the transformation k k/s, or, equivalently, R
Rs. Therefore, the stiffness as well as the fluctuation terms acquire a scaling factor s '
1 + ln s, while the E and terms a factor s d . Hence, after defining t = 1/(4 2 xx ),
c = 1/(4 2), and the coupling constant of the operator Tr Q2 , we get the following
-functions
t = t + 4mt 2 ,
c = c 4c ,

t
t
2+
,
E = dE + E
2
c + mt

t
.
= d + 2t 2 +
c + mt
2
(99)
(100)
(101)
(102)
At finite energy, E 6= 0, we may use a two-cutoff scaling approach [9]. Namely, we

can follow the previous RG equations up to a cross-over scale, scross = s(E), at which
the energy has flowed to a value E0 of order , which plays the role of the high-energy
cut-off in the theory. Above this scale, we must neglect all contributions coming from the
W3 modes, which acquire a mass. That is, we must abandon the RG equations (99)(102),
564
and let the coupling constants flow in accordance with the standard RG equations, which
amount only to a renormalization of t according to
t = t + t 2 .
(103)
If, by integrating (103), the inverse conductance t (s) flows to infinity, signalling an
insulating behavior, then we can define a localization length loc (E) as the scale at which
t has grown to a value of order unity.
9.2. RG in d = 2
In d = 2, the solution of the RG equations for t, c, and E is
t (s) = t (1) = t0 ,
1
1
1
=
+ 4 ln s =
+ 4 ln s,
c(s) c(1)
c0

t0
t0
E(s)
1
= 2+
2+
ln
ln s + t02 ln2 s.
E
2
c0
2
At finite energy, the crossover length, s(E), to the standard, non particle-hole symmetric,
model, which, as previously discussed, is defined through E[s(E)] = E0 , is given by
!)
(
r
E0
1
2
B ,
(104)
B + 4A ln
s(E) = exp
2A
E
being
A=
1 2
t ,
2 0
B =2+

t0
t0
2+
.
2
c0
Above s(E), t flows according to Eq. (103) with = 0, hence it grows to infinity, implying
that the wavefunctions are localized for any E 6= 0. The localization length loc (E) s(E),
apart from a multiplicative factor which is exp[(1 t0 )/t0 ] if t0 1. We see that, for

t0 2
1
t0
2+
,
E E0 exp 2 2 +
2
c0
2t0
the localization length has a power law behavior, namely
1/B
E0
,
loc (E)
E
otherwise, at very small energy, it diverges slower,
r
ln(E0 /E)
.
loc (E) exp
A
The density of states renormalizes like

t0
1
(s) t0
=
2+
ln s + t02 ln2 s.
ln
0
2
c0
2
565
At finite energy, the density of states flows until s < s(E), after which it stays constant.
This implies that the renormalized value is obtained by
ln
(s(E))
E0
(E)
2 ln s(E),
= ln
= ln
0
0
E
leading to

(E) = 0
!
r

4 ln(E0 /E)
1
E0
E0
,
' 0
exp
E s(E)2
E
A
(105)
the last equality valid at small energy.

9.3. RG in d = 3
In d = 3,
t (s) = t0 s 1 ,
c0 s 1
,
1 + 4c0 4c0 s 1

t0
t0
1
1
E(s)
2
= 3 ln s +
2 + + 4t0 1
t0 1 2 ,
ln
E
2
c0
s
s
c(s) =
where t0 is now the inverse of the conductance in units of e2 /(h l), l being a short distance
cut-off of the order of the mean free path. We find that both t and c running variables
vanish for s . The cross-over length diverges, as we approach E = 0, approximately
like
1/3 t
t
E0
0 (2+ c0 +t0 )
0
e 6
.
(106)
s(E) '
E
Hence, the density of states

t0
t
1
E0
(2+ c0 +t0 )
0
' 0 e 2
,
(E) = 0
3
E s(E)
(107)
saturates at E = 0 to a value exponentially increased in t0 with respect to the bare 0 .

At E 6= 0, the inverse conductance above scross = s(E) flows according to (103) with
= 1 and boundary condition t (scross ) = t0 /scross . We find that, if
t0 < scross ,
the system is metallic, otherwise it is insulating, with a localization length
loc (E)
t0 s(E)
.
t0 s(E)
This results implies that, for any amount of disorder, sufficiently close to E = 0, all
eigenfunctions are delocalized, in agreement with recent numerical results [19]. However,
if the disorder is gaussian, as we assumed, the random hopping model with zero regular
hopping seems to be characterized by an inverse Drude conductance, t0 , which is an
increasing function of |E|, being smaller than the critical value at E = 0 (see also
566
Ref. [7]). In this case, the presence of a finite mobility edge in d = 3, even for zero regular
hopping, would not depend crucially upon the intermediate RG flow in the vicinity to the
band center. Nevertheless, we expect that t0 at E = 0 varies for different kinds of disorder,
and eventually it may become greater than unity. In this case, it is just the vicinity to the
band center which makes it possible a finite mobility edge.
10. On site disorder

In this and in the following section, we analyse various symmetry breaking terms, which,
in the two sublattice representation, contain 0 and 3 , hence spoiling Eq. (1).
We start by adding an onsite disorder
X
u1,R c1,R
c1,R + u2,R c2,R
c2,R
Simp =
R
X
R
cR

u1,R u2,R
u1,R + u2,R
0 +
3 cR ,
2
2
where hui,R i = 0 and hui,R uj,R 0 i = ij RR 0 v 2 . Within the path integral, this term becomes,
once average over disorder is performed,
v 2 X
v 2 X

Y1,q Y1,q + Y2,q Y2,q =
Y Y
+ Y3,q Y3,q .
Simp =
2V q
4V q 0,q 0,q
By adding this term to (17), we get

1 X
q + v 2 Tr(Y0,q Y0,q ) q v 2 Tr(Y3,q Y3,q ). (108)
Simp + Simp =
4V q
If we assume that the onsite disorder is weak, i.e., q > v 2 at small q, the consequence is
that the Q free action becomes
1 X
1
1
0
=
Tr[Q0,q Q0,q ] +
Tr[Q3,q Q3,q ]
Simp
2
V q q + v
q v 2
=

2v 2
1
1 X
Tr
Q
Q
Tr[Q3,q Q3,q ].
+
q
q
V q 2(q + v 2 )
q2 v 4
(109)
Therefore, the on site disorder introduces a mass in the Q3 propagators. Specifically, since
2iQ3 3 = Q Q , the mass term can be written as

v2
Tr Qq Qq Qq Qq .
4(q2 v 4 )
Close to the saddle point, QQ = Q2sp , and, for small q, we get

Z

v2
dR Tr Q(R)Q(R) + Q(R) Q(R) ,
2
4
4(0 v )
which, at second order in W , reads
v2 2
2(02 v 4 )
=
2v 2 2
02 v 4
567

dR Tr W (R)W (R) + W (R)s3 W (R)s3
Z

dR Tr W3 (R)W3 (R) .
(110)
In the presence of this term, we could proceed, as before, in the framework of two cutoff
scaling theory. That is, we apply the previous RG equations until the above term becomes
of the order 0 , i.e., up to the scale which, by Eq. (102), is
p

1
B 2 + 4A ln B ,
scross = exp
(111)
2A
in d = 2, where A = 2t02 and B = 2 + t0 (2 + t0 /c0 ), while scross ' 1/d in d > 2, where
0
2,
v
in the limit of small v. Above this scale, the W3 propagator gets fully massive, and
the inverse conductivity flows with the RG equation (103). In d = 2, this implies that,
ultimately, the system gets localized, although the density of states has increased in the
first stage of the RG.
11. Same-sublattice regular hopping

We can also introduce a particlehole symmetry breaking term, by adding to the
Hamiltonian a regular term connecting same sublattices, e.g.,
(0)
(3)
(0)
(3)
HRR 0 = tRR 0 0 + tRR 0 3 Hk = tk 0 + tk 3 .

Expanding the action in H , after integrating over the Nambu spinors, we get an additional
term

1
(112)
S[Q] = Tr T H T G .
2
We define
1 X (0)
4
1 X (3)
4
0 Qsp i
tk G(k),
3 Qsp i
tk G(k),
0
V
0
V
k
so that (112) becomes

4
Tr Q 0 0 + 3 3 .
(113)
i
0
The 0 -term acts like an energy term. This implies that, if we just shift the chemical
potential, we do recover the same scenario as in the absence of this term and at E = 0.
On the contrary, the 3 -term is always a relevant perturbation, whose strength increases
under RG iteration as the energy E. We can define a crossover scale scross , which has the
same expression as s(E) in (104) and (106), for d = 2 and d = 3, respectively, provided
E 3 . Above this scale, the W3 modes get fully massive and their contribution to the
RG flow drops out.
568
(3)
Sometimes tRR
0 = 0, as, for instance, for next-nearest neighbor hopping in a square
lattice. In this case, 3 = 0 and we need to evaluate the second order term

1
S[Q] = Tr T H T GT H T G .
4
If we define F (R) = T (R)T (R) , which contains either 0 and 3 , this term can be written,
at long wavelengths, as

1 X (0) 2
Tr Fq G(k)Fq G(k) .
tk
S[Q] =
4
k
By introducing,

1 X (0) 2
pq
Tr i Gp i Gq ,
tk
i =
V
k
where p, q = , the following results hold

4C
4C
1
1
pq
pq
3 = (1 + pq) ,
0 = (1 pq) ,
2
0
2
0
where C is a constant of dimension energy square, with order of magnitude given by the
2
typical value of tk(0) close to the surface corresponding to E = 0. Therefore, we can
write,
Z

C
dR Tr F (R)F (R) F (R)s3 1 F (R)s3 1
S[Q] =
20
Z

C
(114)
dR Tr Q(R)2 ,
= const +
2
20
which is a mass term for the W3 propagators, similar to (110). Therefore, a same-sublattice
hopping introduces a cross-over length analogous to (111), with
2
.
C
Above this scale, the contribution of the W3 modes to the RG flow has to be dropped out.
12. Time-reversal symmetry breaking

If the random hopping breaks time-reversal symmetry, i.e.,
X
12
RR
Himp =
0 cR cR 0 + H.c.,
RR 0
12 gaussian distributed, after averaging, the impurity

with both real and imaginary part of RR
0
action can be written as

1 X
Wq Tr X1,0,q X2,0,q + X1,3,q X2,3,q ,
(115)
Simp =
V q
where
0 1R ,
X1,0,R = 1R
X1,3,R = 1R
3 1R ,
569
with the indices and running only over the replicas and the advanced/retarded
components. This implies that the manifold in which Q varies contains in this case only 0
and 3 components. Indeed, as in the time reversal invariant case we are able to parametrize
the 8m 8m matrix T in terms of a 4m 4m matrix U U(4m)/Sp(2m) [see Eq. (27)],
similarly, without time-reversal symmetry, T can be parametrized by means of a 2m 2m
matrix U U(2m), in agreement with Gade [9]. The effective non-linear -model is not
modified, but the expressions (D.1)(D.5) have to be substituted by

= 2D(k)ad Tr P0 ,
(116)
Bab Pbc Bcd
hBab Pbc Bcd i = 0,
Pad ,
hAab Pbc Acd i = 2D(K)ad Tr(P0 ) D(k)
xx + m
Pad ,
hCab Pbc Ccd i = 2D(K)ad Tr(P0 ) D(k)
xx + m
Pad ,
hAab Pbc Ccd i = D(k)
xx + m
where P = P0 + iP3 3 . Hence, the RG equations at m = 0 are, in this case,
t = t,
(117)
(118)
(119)
(120)
(121)
c = c 2c ,
Et 2
,
E = dE +
2c
which coincide with those obtained by Gade [9].
2
(122)
(123)
13. Discussion and comparison with the standard localization theory

In this section, we summarize the main differences between the model (2) and the
standard non-linear -model which is derived in the theory of Anderson localization [14
16,20], placing particular emphasis on the properties of the Q-matrix. The specific form
of the off-diagonal disorder we consider, which only couples one sublattice to the other,
leads, via the HubbardStratonovich decoupling, to the introduction of a space-varying
8m8m complex Q-matrix, Q = Q0 0 +iQ3 3 . Here, Q0 and Q3 are 4m4m hermitean
matrices, of which matricial structure refers to the retarded/advanced, spinor particle/hole
and m replica components. Contrary to the standard case, Q is not hermitean.
The evaluation of the saddle point, Qsp = 0 0 s3 (Section 5), as well as the derivation
of the effective action (Section 6) are analogous to the standard case [1416,20]. (We
recall that i , si and i indicate the Pauli matrices, including the unit matrix, acting on
sublattice, advanced/retarded and spinor components, respectively.) The non-linear model, Eq. (79), is obtained by integrating out the longitudinal massive Q-fluctuations
and only keeping the transverse soft modes. The real novelty with respect to localization
theory is not in the structure of the effective action. Indeed, the new term in (79), namely,
Z
2

E
,
dR Tr Q(R) Q(R)
3
570
even if present, would be irrelevant in the standard case. On the contrary, the essential
difference, as expected, lies in the ensamble spanned by the soft modes at the particle
hole symmetry point E = 0. We get QSoft = T 1 Qsp T , where the unitary matrix T only
contains 0 and 3 ,

W0 0 + W3 3
,
T = exp
2
and

W0 0 W3 3
.
T = 1 T 1 = 2 T 2 = exp
2
These expressions derive by the conditions (1), which fully specify the model, as shown in
Section 4. In that section, we also showed that the ensamble can be expressed in terms of
unitary 4m 4m matrices

W0 + W3
,
U = exp
2
as argued by Gade and Wegner [8]. Selecting the subset which leaves the saddle point
invariant gives Eq. (2) with U U(4m)/Sp(2m). In terms of T , the condition T 1 Qsp T 6=
Qsp , leads to the requirements [W0 , s3 ] 6= 0 and {W3 , s3 } 6= 0, which implies that W0 is
off-diagonal in the energy retarded/advanced space (as in the standard localization theory),
while W3 is diagonal. In other words, the omogeneous and staggered modes, W0 and W3 ,
respectively, have different structure in the energy space. The energy diagonal W3 -modes
betray the presence, at E = 0, of diffusive poles in the disorder averaged products of
retarded and advanced Greens functions, GR GR and GA GA , with GR,A = (H i0+ )1 .
In the localization theory [1416,20], only the mixed products GR GA have a singular
behavior. This explains why singular corrections to the density of states (which involves
connected diagrams with same energy Greens functions) are present in the two-sublattice
model, while they are absent in the localization theory.
In a square lattice, the energy diagonal modes have the transparent meaning of density
fluctations with wave-vector q nearby the nesting vector G = (, , . . .), see Appendix A.
Indeed
X

eiqR 1R 1R 2R 2R
Q3 (q)
RA
ei(q+G)R R R = Q(q + G),
RA,B
where A and B label the two sublattice, and, for R A, we have taken by definition 1R =
R and 2R = R+a x , being a the lattice spacing and x the unit vector in the x-direction.
As soon as E 6= 0, nesting is not more important and indeed Q3 becomes massive.
Finally, because of GR GR and GA GA , also the conductance acquires other corrections
with respect to standard localization theory. Indeed, these corrections add to give a
vanishing -function for xx , as first indicated by Gade and Wegner [8]. We have explained
in the Introduction section (see also Appendix E) that this is a consequence of a simple
571
abelian gauge symmetry generated by Eq. (1) at the particlehole symmetry point E = 0.
Similarly to the results for the density of states, this behavior of the -function is at odds
with the standard theory.
14. Conclusions
In this work, we have derived the effective non-linear -model of a disordered electronic
system on a generic bipartite lattice. This model, if the hopping matrix elements as well as
the disorder only couple one sublattice with the other, shows an interesting behavior close
to the band center, i.e., to the particlehole symmetry point. Namely, the wave-functions
are always delocalized at the band center, in any dimension. By a renormalization group
(RG) analysis, in the framework of an -expansion, = d 2, we have found that the
quantum corrections to the conductivity vanish if the chemical potential is exactly at the
band center, thus implying a metallic behavior. In two dimensions, in particular, the Kubo
conductivity flows to a fixed value by iterating the RG. On the contrary, we have found that
the staggered density of states fluctuations, which are controlled by a new parameter in the
non-linear -model, are singular. This result is reminiscent of what it is found in equivalent
one-dimensional models. In fact, models of disordered spinless fermions in one-dimension
can be mapped, by a JordanWigner transformation, onto disordered spin chains. In many
cases, it is known that, in spite of the presence of disorder, these spin chain models
display critical behavior, as shown in great detail by D. Fisher for random Heisenberg
antiferromagnets and random transverse-field Ising chains [21]. Indeed, as pointed out by
Fisher, the staggered spin fluctuations, in a random antiferromagnetic chain, also display
critical behavior in the form of a power law decay, (1)R hS(R)ihS(0)i R 2 , where
the bar indicates impurity average. Since the staggered spin-density corresponds to the
staggered density of the spinless fermions, this result is consistent with the outcome of our
analysis, which further suggests that a similar scenario generally holds in such models.
Moreover, as in one-dimension [4,5,21], we find that the density of states is strongly
modified by the disorder at the band center, and it actually diverges in d = 2. In reality, a
random Heisenberg chain, away from the XXZ limit, maps onto a spinless fermion random
hopping model in the presence of a random nearest-neighbor interaction. However, even
in the presence of this additional interaction, the Hamiltonian has still the abelian gaugelike symmetry described in Appendix E, which is at the origin of the delocalization of the
band center state. This observation is also compatible with Fishers result that the physical
behavior does not qualitatively change upon moving away from the XXZ limit towards the
isotropic XXX Heisenberg point.
Many of the results which we have derived were already known. The existence of
delocalized states at the band center of a two-sublattice model was argued already in 1979
by Wegner [6,7]. The effective non-linear -model when the disorder breaks time-reversal
invariance, as well as the RG equations, have earlier been derived by Gade [9], although in
a particular two-sublattice model. The extension to disordered systems with time-reversal
symmetry was later on argued by Gade and Wegner [8]. Finally, random flux models and
572
disordered Dirac fermion models have recently been the subject of an intensive theoretical
study [1013], for their implications to a variety of different physical problems.
In spite of that, our analysis has several novelties with respect to earlier studies.
First of all, the two-sublattice model which we study is quite general. Secondly, the
physical interpretation of the parameters which appear in the non-linear -model is quite
transparent. Thirdly, the explicit derivation of the non-linear -model and of the RG
equations with time-reversal invariance is presented.
Acknowledgements
We are grateful to A. Nersesyan, V. Kratsov, E. Tosatti, Yu Lu and G. Santoro for helpful
discussions and comments.
Appendix A. Specific examples

As an example, we consider a tight binding model with only nearest neighbor hopping
on a square and honeycomb lattice.
In the case of square lattice, the enlarged unit cell is the 2 2 one. The new
E 2 = 2(1, 1)/a, and the angle k of
E 1 = 2(1, 1)/a, G
reciprocal lattice vectors are G
Eq. (9) is
a
k = (k1 + k2 ) = kx a.
2
(A.1)
In the case of the honeycomb lattice, the unit cell contains already two lattice sites. The
energy is given by

2

2
3
3
3
3
2
kx a cos
ky a
kx a cos
ky a
+ 2 sin
,
k = 1 + 2 cos
2
2
2
2
and
k = tan1

3
3
k
a
cos
x
2
2 ky a

.
1 + 2 cos 32 kx a cos 23 ky a
2 sin
The Brillouin
zone is still honeycomb, with the y-axis one of its axes, and side equal to
4/(3 3 a).
Appendix B. Ward identity

Let us consider a generic Hamiltonian in the two sublattice representation
X
cR1 HR1 ,R2 cR2 ,
H=
R1 ,R2
(B.1)
573
where HR1 ,R2 is a 2 2 Hermitean matrix. The current operator

X
cR1 JER1 ,R2 (R)cR2 ,
JE(R) =
R1 ,R2
can be obtained by the continuity equation, leading to

X
E JE(R) = i
cR HR,R1 cR1 cR
HR1 ,R cR ,
(B.2)
R1
E the discrete version of the differential operator. The long-wavelength expression

being
E
for J (R) can be obtained by Fourier transformation, namely, through
X
X

E
E
qE JE(R) eiqER = i
HR1 ,R cR eiqER ,
cR HR,R1 cR1 cR
i
1
R
R,R1
and expanding both sides in q, we get, for the linear term,

X
X
qE JE(R) =
qE RE cR
HR,R1 cR1 cR
HR1 ,R cR
i
1
R
R,R1

qE RE1 RE cR
HR1 ,R cR ,
1
R,R1
hence
JE(R) = i

HR1 ,R cR .
RE1 RE cR
1
(B.3)
R1
Let us define the correlation functions

,i (R, R 0 ; t, t1 , t2 )

(t)JR1 ,R2 (R)cR2 (t)cR

(t1 )JRi 3 ,R4 (R 0 )cR4 (t2 ) ,
= T cR
1
3
(B.4)
where = 0, 1, 2, 3, JR01 ,R2 (R) = RR1 RR2 are the density matrix elements, and J i , for
i = 1, 2, 3, are the matrix element components of the current. By the continuity equation,
we find that
X

i
(t t1 ) Tr G(R1 , R; t2 t)JR,R
(R 0 )
t 0,i + j j,i = i
1
R1

+ (t t2 ) Tr G(R, R1 ; t t1 )JRi 1 ,R (R 0 ) .
If we integrate both sides by
Z
dt dt1 dt2 ei(E+)(t t1) eiE(t2 t ) ,
at = 0 we find
j j,i (R, R 0 ; E) = i

Tr G(R, R1 ; E)JRi 1 ,R (R 0 )
R1

i
(R 0 ) .
Tr G(R1 , R; E)JR,R
1
Using once more the continuity equation (B.2), we find
574

j i0 j,i (R, R 0 ; E) = Tr G(R, R 0 ; E)HR 0 ,R + G(R 0 , R; E)HR,R 0
X

Tr G(R, R1 ; E)HR1 ,R + Tr G(R1 , R; E)HR,R1 .
+ RR 0
R1
By Fourier transform,
X
X
E E0
E E0
j i0 j,i (R, R 0 ; E) eiqE(RR ) =
qi qj j,i (R, R 0 ; E) eiqE(RR )
RR 0
1e
E RE0 )
iqE(R
RR 0

Tr G(R, R 0 ; E)HR 0,R + G(R 0 , R; E)HR,R 0 .
RR 0
At small q, the above expression is

X
qi qj j,i (R, R 0 ; E)
RR 0

1X
qi qj Ri Ri0 Rj Rj0 Tr G(R, R 0 ; E)HR 0 ,R + G(R 0 , R; E)HR,R 0 ,
2 0
RR
leading to
X
X

j,i (R, R 0 ; E) =
Ri Ri0 Rj Rj0 Tr G(R, R 0 ; E)HR 0 ,R .
RR 0
(B.5)
RR 0
Appendix C. Longitudinal modes

As discussed in Section 7, the expression of the Q-matrix which includes also the
longitudinal modes is QP (R) = Q(R) + S(R), where Q(R) and S(R) have been defined
through (72). The free action for these fields contains a local term, Eq. (74), and a non local
one, Eq. (75). The latter is
Z
1
dR dR 0 (R R 0 )
4

Tr 1R Q(R 0 )1R Q(R 0 ) + 1R S(R 0 )1R S(R 0 ) + 21R Q(R 0 )1R S(R 0 ) , (C.1)
where we have defined the operator
1R f (R 0 ) = f (R) f (R 0 ) =
X
1 E E0 n E n
R R f (R 0 ).
n!
n=1
Let us apply this operator to Q(R) and S(R), keeping all terms which contains at most two
derivatives which act to the transverse matrices T . We obtain

0
E
) RE RE 0 ,
(C.2)
1R Q(R 0 ) ' Q(R
0
0
0
0
(C.3)
1R S(R ) ' T (R ) 1R P (R )T (R )

0
0
0
0
0
E (R ) RE RE (C.4)
E T (R ) P (R)T (R ) + T (R ) P (R) T
+

1 E2 0
E T (R 0 ) P (R) T
E (R 0 )
T (R ) P (R)T (R 0 ) + 2
+
2

E 2 T (R 0 ) RE RE0 2 .
(C.5)
+ T (R 0 ) P (R)
575
The term which is obtained by (C.3) times its hermitean conjugate, together with the local
piece (74) give the free action of the longitudinal modes

1 X 1
Tr Pq Pq .
(C.6)
S0 [P ] =
V q 2q
E = T T
E , give rise to the coupling between transverse
The mixed terms, after defining D
and longitudinal modes
Z

1
dR dR 0 (R R 0 ) RE RE0
S[Q, P ] =
2

E 0 ) P (R 0 )P (R) P (R) P (R 0 )
(C.7)
Tr D(R
Z

1
2
4
n
E 0 )1 P (R)D(R
E 0 )P (R 0 )
Tr 1 D(R

E 0 )P (R)
E 0 )1 P (R 0 )D(R
(C.8)
+ 1 D(R
o

E 0 ) P (R)P (R 0 ) + P (R 0 )P (R)
E 0 )D(R
(C.9)
Tr D(R
Z

1
4

0
E
)1R P (R 0 ) + H.c. .
(C.10)
Tr Q(R
The last term, (C.10), gives rise to higher gradient contributions, hence can be neglected.
C.1. Longitudinal propagators
Before averaging over the longitudinal modes, we have to evaluate the longitudinal
propagators. The matrix
P = (P0,0 s0 + P0,3 s3 )0 + i(P3,1 s1 + P3,2 s2 )3 ,
has to satisfy cP t ct = P , and, in addition, all P = P . For = (0, 0), (0, 3), (3, 1), by
writing
P = P(0) 0 + iPE E,
we find that
P(0) , PE Re,
t
P(0) = P(0) ,
t
PE = PE .
For = (3, 2), by writing

(0)
0 + PE3,2 E,
P3,2 = iP3,2
we must impose
(0) E
, P3,2 Re,
P3,2
(0)
(0) t
P3,2
= P3,2
,
If P(i) is a symmetric real matrix, its propagator is
(i) (i) G
P,ab P,cd = (ad bc + ac bd ),
2
t
PE3,2 = PE3,2 .
(C.11)
576
while if it is antisymmetric
(i) (i)
G
(C.12)
P,ab P,cd = (ad bc ac bd ),
2
where Gk = k /8V , and (a, b, c, d) are replica indices. By means of these propagators,
E i E , where M (i) are matrices in the replica space,
we readily find that, if M = Mi(0) + iM
then, for = (0, 0), (0, 3), (3, 1), the following results hold

(C.13)
hP MP i = GcM t ct + 2G Tr M (0) = GcM t ct + G Tr M ,
while, for = (3, 2),

hP3,2 MP3,2 i = GcM t ct + 2G Tr M (0) = GcM t ct + G Tr M .
(C.14)
More generally,

P Msi j P = GcM t ct si j + s3 si j s3 + 3 s1 si j s1 3 3 s2 si j s2 3

+ G Tr M si j + s3 si j s3 + 3 s1 si j s1 3 + 3 s2 si j s2 3 ,

P Msi j P = GcM t ct si j + s3 si j s3 3 s1 si j s1 3 + 3 s2 si j s2 3

+ G Tr M si j + s3 si j s3 3 s1 si j s1 3 3 s2 si j s2 3 .
For j = 0, 3 the above expression simplifies to

P Msi j P = 2Gc(Msi j )t ct + 2Gj Tr(Msi s0 ),

hP Msi j P i = 2Gs3 c(Msi j ) c s3 + 2Gj Tr(Msi s3 ),
t t
(C.15)
(C.16)
while for j = 1, 2
hP Msi j P i = 2Gs3 c(Msi j )t ct s3 + 2Gj Tr(Msi s3 ),
hP Msi j P i = 2Gc(Msi j ) c + 2Gj Tr(Msi s0 ).
t t
(C.17)
(C.18)
C.2. Averaging S[Q, P ]

We have now all what it is needed to proceed in the averaging over P . Here we just
E
sketch the calculation, which is quite involved and requires the matrix properties of D
which are determined in Appendix D. We just remark that (C.8) and (C.9) do not reproduce
the correct stiffness term. Indeed, it is (C.7), which contributes at second order, which
cancels the additional terms and allows to express everything in terms of the matrix Q.
By means of the previously calculated propagators of the longitudinal modes, we find
that
Z
n

Y
E
E
Q(R)
dR
Tr Q(R)
S[Q, P ] P =
2
4d
2 o
1
E
T
r
Q(R)
Q(R)
,
(C.19)
+
3
8 2
where
Z
(C.20)
Y = dR (R) (R)1 R 2 .
577
We notice that the first term is a contribution to the Kubo conductivity and the second to
the staggered density of states fluctuations of the diagrams where the current vertices are
those proportional to the random hopping.
C.3. Additional terms
The last class of corrections which generate new operators is obtained by expanding
each Greens function in (48) at first order, GP = G0 iG0 P G0 , leading to the term

1
(C.21)
Tr(GP GU GP GU ) P .
4
For the sake of clarity, we will analyse this term only in the case of a cubic lattice, where
the derivation is more straightforward. We will postpone a discussion about the general
case at the end of the section.
In the cubic lattice, according to Eq. (23), the electron Q-coupling can be brought to a
local one also in the diagonal basis. In this basis, Q = Q0 + iQ1 1 ,
P (R) = (P00 s0 + P03 s3 )0 + i(P11 s1 + P12 s2 )1 ,
(C.22)
the unitary matrix

W1 (R)
W0 (R)
0 +
1 ,
T (R) = exp
2
2
(C.23)
where W1 has the same form as W3 defined by Eq. (33), and T = 3 T 3 . Moreover, since
HRR 0 = RR 0 3 in the diagonal basis, being RR 0 the Fourier transfom of k , the current
operator which appears in the definition of URR 0 , Eq. (41), has only matrix elements JEk =
E k 3 .

Once averaged over P , (C.21) gives, among other terms which correct the Kubo
conductivity, a new term [see Eq. (Greenfunctiond)]
1 X XX
a,h b,h
a,h;b,h b,h a,h
pk Tr Uqb,h;a,h Gp+q
Gp Tr Uq
Gk Gk+q
2
a,b=1,2 h= pkq
a,h b,h
a,h
Gp Tr U a,h;b,h Gkb,h Gk+q
,
Tr U b,h;a,h Gp+q
where the first piece derives from P0 and the second from P1 . The structure in the
energy/sublattice indices can be shortly represented by
X
1
i i = (2 2 + s3 1 s3 1 + 3 3 + s3 s3 ),
4
i=1,...,4
so that the above term can be written, at small q, as

1 X X
pk Tr(Uq Gp i Gp ) Tr(Uq Gk i Gk ).
2
i=1,...,4 pkq
We remind that
E q JEp Gp ,
Gp Uq Gp = iGp D
(C.24)
578
E q is the Fourier transform of T (R)

E T (R)1 . We notice that only the diagonal
where D
E enters. Moreover, for the diagonal matrices sj = s0 , s3 , since
component in energy of D
E 1 1 , it derives that, for i = 0, 1, the following equality holds Tr Ds
E j i =
E =D
E 0 0 + D
D
E
Tr( W sj i /2). Therefore, just W1 1 contributes. By means of (55), for any s we

have

E 1,q s1 G(p)s1 2 i G(p)
E 1,q 1 JEp G(p)i G(p) = +
E p Tr D
i Tr D

E 1,q G23 3 2 i 3 G20 s3 2 i s3 + G3 G0 s3 [3 , 2 i ] .
E p Tr D
=
Only for 1 = 2 /4 the trace over the s is finite, leading to

E 1,q .
E p Tr D
2 G23 G20
(C.25)
In conclusion, going back to the original sublattice representation, and defining

(0) =
X

1
+
+
+
E p pk Tr 3 G+
E k

p 3 Gp Tr 3 Gk 3 Gk ,
2
4V d
(C.26)
pk
we obtain the following explicit expression of (C.21) [notice that Q = Q0 0 + iQ1 1

Q0 0 + iQ3 3 in the sublattice basis]
Z

2
(0)
E
E
(C.27)
dR Tr Q (R)Q(R)
3 Tr Q (R)Q(R)
3 .
4
8 32
This term, which derives from the regular current vertices, has the same form as the second
term in (C.19), which, on the contrary, is due to the random current vertices. Therefore, the
coupling constant which appears in the final expression (78) is the sum of both terms.
The symmetry of the operator (C.27), which, due to the trace, involves the Nambu,
energy and sublattice components 0 , s0 and 3 , respectively, suggests that the prefactor
represents the fluctuations of the staggered density of states, as discussed in Section 8.2. In
the case of a generic bipartite lattice, we do expect an analogous term to appear, because the
staggered-density of states fluctuations still acquire singular contributions, and the operator
is not forbidden by the symmetry properties of the Q-matrix. The reason why we decided
to show only the case of cubic lattices is that the distinction between the longitudinal
from the transverse modes is a bit ambiguous at large momenta, where the both are in a
sense massive. This is not a problem for cubic lattices, where one can show that the large
momentum components of the transverse modes do not contribute, hence (C.27) exhausts
the whole contribution. On the contrary, in other cases, we do have to keep into account
the contribution of the small-wavelength transverse modes to recover the full expression.
This makes the calculations more involved than in the case of cubic lattices.
Appendix D. Explicit derivation of the RG equations

In this Appendix we outline the derivation of the RG equations. Before that, it is
convenient to list some useful results.
579
D.4. Averages
Besides the propagators, we will also need the explicit expression of particular averages
which enter in the derivation of the RG equations. The following results hold:

(D.1)
= 4D(k)ad Tr(P0 ),
Bab Pbc Bcd
,
hBab Pbc Bcd i = 2D(k)Pad
(D.2)
where P = P0 + iPE E is a quaternion real matrix.

In addition,
+ 4D(K)ad Tr P0
hAab Pbc Acd i = 2D(k)Pad
Pad ,
D(k)
xx + m
+ 4D(K)ad Tr(P0 )
hCab Pbc Ccd i = 2D(k)Pad
Pad ,
D(k)
xx + m
Pad .
hAab Pbc Ccd i = D(k)
xx + m
For instance, if P = I, then

BB = 4D(k)mI,
hBBi = 2D(k)I,

I,
hAAi = 2D(k) + 4D(k)m D(k)
xx + m

I,
hCCi = 2D(k) + 4D(k)m D(k)
xx + m
I.
hACi = D(k)
xx + m
(D.3)
(D.4)
(D.5)
(D.6)
(D.7)
(D.8)
(D.9)
(D.10)
D.5. RG equations
First of all, we need to know the quaternion structure of the matrix D = T T . Since
D = D ,
cD t ct = 1 D1 ,
we can write D = D0 0 + D3 3 , where in the -space

!
A0 B0
,
D0 =
B0 C0
and
D3 = i
A3
B3
B3
C3
(D.11)
!
.
(D.12)
580
We can write each of the above matrices in quaternion form, P = P (0) 0 + iPE E , where
P (0) , PE Re, but, in addition, we must impose that
(0) t
(0)
(0) t
(0)
= A0 ,
AE t0 = AE0 ,
C0
= C0 ,
CE 0t = CE0 ,
A0

t
(0) t
(0)
(0)
(0)
t
A3
= A3 ,
AE 3 = AE3 ,
C3
= C3 ,
CE 3t = CE3 .
By making use of the above properties, and by means of the Eqs. (D.2), (D.1), (D.3), (D.4)
and (D.5), after defining
Z
L=
d2 k
D(k),
(2)2
/s
we get the following results

Wf D0 Wf f = LA D0
+ 2LB
B0
0
0
B0
where = /(xx + m), and

Wf D3 Wf f = LA D3

A3
+ 2LA i
0
0
C3

2LA
A0
0
0
C0
2LB i
0
B3
B3
0
(0)
4LI Tr iA(0)
3 + iC3 .

,
(D.13)
(D.14)
For further convenience, when useful, we have labelled LB the propagator of Bf , and LA
the ones of Af and Cf . The next useful result is

(D.15)
Wf Wf f = (4LB m 4LA m + 2LA + LA )I.
Through (D.13) and (D.14) we therefore get
Z

2xx
E f Qsp 1
E 1 Qsp Wf DW
dR 2 Tr D
hF1 if =
f
32 2

2
E sp W f 1
E 1 Qsp DQ
+ 2 Tr D
f
2xx
=
32
Z
dR4(LB LA + 2LA m + 2LB m) Tr

+ 4(2LA m + 2LB m)
2

2L Tr(D3 )
iA3
0
0
iC3

2
B0
0
!2
(D.16)
(D.17)
A0
0
0
C0
4(2LA m + 2LB m + LA + LB )
0
iB3
4(2LA m + 2LB m + 2LA )
0
B0
(D.18)
2
(D.19)
iB3
0
!2
.
(D.20)
581
The calculation of hF22 if is more involved, since one needs the average of four Wf s. For
sake of lengthy, we just quote the final result that such a term cancels (D.19) and (D.20).
We next notice that

1
Tr Q Q = 2 Tr Ds3 1 Ds3 1 DD
2
!2

2
0
B0
iA3 0
4 Tr
,
= 4 Tr
0 iC3
B0 0
and that

Tr Q Q3 = 2 2 Tr(D3 ).
Therefore, for LA = LB ,
Z

2xx
= 4Lm
Q
F1
dR
Tr
Q
32 2
2
1 2xx
(D.21)
Tr Q Q3 .
L
4
2 32
In the cases in which the A and C modes are gaped (LA = 0, and no D3 ), we obtain the
standard result
Z

2xx
1 2
(D.22)
dR Tr Q Q .
F1 2 F2 f = (2Lm + L)
2
32
1 2
2 F2 f
Appendix E. Gade and Wegners proof of the vanishing -function

The equations from (33) to (39) imply that W3 is not a traceless matrix. Indeed, we can
alternatively write W as
W = W0 +
1
Tr(W3 )3 W 0 + i3 ,
4m
(E.1)
with W 0 = W0 0 + W30 3 , now being W30 a traceless matrix. Since 3 commutes with W 0 ,
this means that
T (R)2 = eW (R) = ei(R)3 eW
0 (R)
ei(R)3 V (R),
(E.2)
which also defines the matrix field V (R). By means of this parametrization, the non-linear
-model (2) can also be written as
Z

2xx
E (R)
E (R)1 V
dR Tr V
S[T ] = S[V , ] =
32
Z

m
E
E
xx + m
dR (R)
(R).
(E.3)
+
2
Therefore, the action of V is distinct from that of , and the latter, being a phase, is
gaussian. This implies that the combination xx + m is not renormalized and scales
with its bare dimension , for any number of replicas. In turns, it means that, in the zero
replica limit, it is xx which is not renormalized! This is completely equivalent to the nice
582
proof given by Gade and Wegne Ref. [8] that the quantum corrections to the -function of
the conductance of a U(N)/SO(N) model vanish at all orders in the N 0 limit.
The other important result concerns the renormalization of an operator
T 2q = eiq3 V q .
Within RG,
eiq3

1 iq3
t 2
t
q ln s
.
e
2
c + mt
m
(E.4)
The second term, which is singular in the m 0 limit, has to be canceled by the one-loop
renormalization of V q . Gade and Wegner showed that this cancellation holds for any q.
Furthermore, they argued that, apart from the one-loop correction, the renormalization of
V q does not contain any other singular term in the m 0 limit. This, as they pointed out,
has very important consequences. In 2d, t does not scale, while c goes to zero. Therefore
the term which dominates the renormalization of T 2q for m = 0 is just the first term in the
right hand side of (E.4). This argument implies that the one-loop correction, which we have
derived for the density of states (q = 1 case), is sufficient to identify the correct asymptotic
behavior.
To conclude, let us discuss more in detail the origin of this gaussian field . In the
Grassmann variable path-integral representation, the action for the particlehole symmetric
model at E = = 0
X
R HRR 0 R 0 ,
S=
RR 0
posseses a simple abelian gauge-like symmetry

ei3 ,
(E.5)
because {3 , HRR 0 } = 0. It is just this symmetry which causes the appearance of the
gaussian part of the non-linear -model. Notice that, if time reversal invariance holds,
this symmetry implies a particlehole symmetric Hamiltonian, which is invariant under
but c2,R c2,R

. In fact, {3 , HRR 0 } = 0 also means that
the transformation c1,R c1,R
{1 3 , HRR 0 } = 0, being HRR 0 0 . The interesting fact is that (E.5) is not a symmetry
of the fermion operators. Indeed, under this transformation, cR ei3 cR , but cR
ei3 cR (chiral symmetry), and not cR ei3 cR as we would expect if cR and cR had
. Finally, we notice that if, besides 3 , the

to be identified with the operators cR and cR
Hamiltonian commutes with another Pauli matrix (as it can be the case for specifically built
particlehole symmetric models), the above gauge symmetry would be non abelian, hence
spoiling all peculiar properties which we have shown to occur. Indeed, in the last case,
the system can be mapped into a standard localization model with an additional sublattice
index.
References
[1] R.E. Prange, S.M. Girvin (Eds.), The Quantum Hall Effect, Springer-Verlag, New York, 1990.
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
583
Kravchenko et al., Phys. Rev. B 50 (1994) 8039; Phys. Rev. B 51 (1995) 7038.
G. Bergmann, Phys. Rev. B 28 (1983) 2914.
G. Theodorou, M.H. Cohen, Phys. Rev. B 13 (1976) 4597.
T.P. Eggarter, R. Riedinger, Phys. Rev. B 18 (1978) 569.
F. Wegner, Phys. Rev. B 19 (1979) 783.
R. Oppermann, F. Wegner, Z. Phys. B 34 (1979) 327.
R. Gade, F. Wegner, Nucl. Phys. B 360 (1991) 213.
R. Gade, Nucl. Phys. B 398 (1993) 499.
T. Fukui, Nucl. Phys. B 562 (1999) 477.
A. Altland, B.D. Simons, Nucl. Phys. B 562 (1999) 445.
A. Altland, B.D. Simons, J. Phys. A 32 (1999) L353, and references therein.
S. Guruswamy, A. LeClair, A.W.W. Ludwig, Nucl. Phys. B 583 (2000) 475, in this issue.
F. Wegner, Z. Phys. B 35 (1979) 207.
S. Hikami, Phys. Rev. B 24 (1981) 2671.
A.M.M. Pruisken, L. Schfer, Nucl. Phys. B 200 (1982) 20.
G. Wilson, J. Kogut, Phys. Rep. 12 (1974) 77.
A.M. Polyakov, Phys. Lett. B 59 (1975) 79.
P. Cain, R.A. Rmer, M. Schreiber, Ann. Phys. (Leipzig) 8 (1999) 507.
K.B. Efetov, A.I. Larkin, D.E. Khmelnitsky, Zh. Eksp. Teor. Fiz. 79 (1979) 1120 [Sov. Phys.
JETP 52 (1980) 568].
[21] D. Fisher, Phys. Rev. B 50 (1994) 3799; Phys. Rev. B 51 (1995) 6411.

Folding transitions of the square-diagonal

two-dimensional lattice
Emilio N.M. Cirillo a , Giuseppe Gonnella b , Alessandro Pelizzola c
a Dipartimento di Matematica, II Universit di Roma Tor Vergata, via della Ricerca Scientifica,
I00133 Roma, Italy

b Istituto Nazionale per la Fisica della Materia, Unit di Bari and Dipartimento di Fisica dellUniversit di
Bari and Istituto Nazionale di Fisica Nucleare, Sezione di Bari via Amendola 173, 70126 Bari, Italy
c Dipartimento di Fisica del Politecnico di Torino and Istituto Nazionale per la Fisica della Materia,
c. Duca degli Abruzzi 24, 10129 Torino, Italy

Abstract
The phase diagram of a vertex model introduced by P. Di Francesco (Nucl. Phys. B 525 (1998)
507) representing the configurations of a square lattice which can fold with different bending energies
along the main axes and the diagonals has been studied by Cluster Variation method. A very rich
structure with partially and completely folded phases, different disordered phases and a flat phase
is found. The crumpling transition between a disordered and the flat phase is first-order. The CVM
results are confimed by the analysis of the ground states and of the two limits where the model
reduces to an Ising model. 2000 Elsevier Science B.V. All rights reserved.
PACS: 05.50.+q; 64.60.-i; 82.65.Dp
1. Introduction
Polymerized membranes are two-dimensional networks of molecules with fixed connectivity [1]. Their configurational properties and the existence of a crumpling transition
between a folded and a flat phase are relevant for the behavior of some biological systems
[2]. Models of polymerized membranes can be defined on regular lattices. The constraint
of fixed connectivity gives rise to the definition of complicate vertex models which offer
the advantage that explicit analytical calculations can be performed [3]. In these models
the bonds between the vertices of the network have a fixed length and also, as a further
simplification, self-avoidance is not considered. In this way the only degrees of freedom
of the network are related to its possible states of folding and each state can be weighted
by a Boltzmann factor depending on the relative angle between adjacent plaquettes in the
network.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 8 9 - 3
E.N.M. Cirillo et al. / Nuclear Physics B 583 [FS] (2000) 584596
585
The model studied in this paper has been introduced by Di Francesco [4,5] and concerns
the folding properties of a square lattice which can be folded along the main axes and along
the diagonals. This model follows previous studies of the folding properties of a triangular
network embedded in d-dimensional lattices. In the case d = 2 [6,7] each triangle of the
network can be only in a up or down state. The entropy of this problem is the same as
in the three-colouring problem of the honeycomb lattice [8]. The introduction of a bending
rigidity induces a first-order crumpling transition which has been studied in [9,10]. The
model with the triangular network embedded in the face-centered-cubic lattice was defined
in [11] and a first-order transition has been found also in this case [12,13].
In this paper we consider the two-dimensional folding of the square-diagonal lattice
and study the phase diagram of this system in terms of the bending rigidities S and L
relative to short and long edges. Actually, the phase behavior of this model was already
considered in [5] by means of a transfer matrix analysis, but the ground state was not
properly investigated, and as a consequence some phases were missing. Here we consider
the complete set of the ground states of the model and study the phase diagram at finite
temperatures by the cluster variation method (CVM) [1417]. This method has already
proven to be useful for studying vertex models and folding problems [10,12,13].
The outline of the paper is the following. In the next section we define the model; the
ground states will be shown in Section 3. In Section 4 the CVM approximation used in this
paper will be described; our results will be presented in Section 5 and Section 6 will be
devoted to a discussion.
2. The square-diagonal folding model

We consider the folding configurations of the square-diagonal lattice where any couple
of adjacent triangles can be on the same plane or with the two triangles one on the other.
Triangles can be folded with respect to the main axes of the square lattice or with respect to
the diagonals so that in the definition of the bending energy different couples of triangles
with a short edge or a long edge in common have to be considered. We will start the
definition of the model by enumerating the folding states in terms of loop configurations.
2.1. Folding and loop configurations
First fix a reference orientation for all the edges of the lattice in such a way that for
each triangle the vectorial sum of the oriented edges is zero. There are two choices for
this global orientation and one is shown in Fig. 1. In any folding configuration each short
edge is mapped on one of the four vectors Ee1 , Ee2 ; this mapping defines the state of the
network. Then consider two triangles with a long edge in common as the pair shaded in
Fig. 1. This pair can be in two folding states that can be represented in the following way:
consider another square lattice as in Fig. 2. The lines joining the centers of the edges are
dashed or full depending on whether they are dual to the mapped short edge vectors Ee1
or Ee2 , respectively. The obvious observation that in each triangle there is a short edge
vector in the direction of Ee1 and an edge in the direction of Ee2 implies that for each
586
Fig. 1. The square-diagonal lattice.
Fig. 2. The mapping onto the vertex model.
center of the edges of the new square lattice there will be two dashed and two full lines. As
a consequence the dashed lines and the full lines will form two sets of orthogonal closed
loops [4].
Taking into account also the orientation of eE1 and eE2 , it comes out that each closed
loop has its own orientation independently from the others. Therefore the original
folding problem is equivalent to a dense two-loop problem where a sign representing the
orientation has to be attributed to each closed loop. Further details can be found in [4].
Consider now the square plaquettes of this new lattice. In each plaquette there are two
dashed and two full lines. On each line there can be two possible signs so that there
are 2 24 = 32 states for each plaquette. Of course, the sign conservation law along
the closed loops implies that only some of the 322 configurations of two neighboring
plaquettes are allowed. The logarithm of the number of all possible states of these plaquette
configurations gives the entropy of the folding problem.
587
2.2. Bending energy

Here we introduce a bending energy weighting the different folding configurations. We
attribute a Boltzmann weight eL or eL , respectively, to each unfolded or folded long
edge. Similar weights eS are associated to the two short edge folding states.
In the plaquette formulation long and short edge bending energies appear in an
asymmetric way. Long edge bending energy results in an interaction between each couple
of adjacent plaquettes in the loop formulation. An energy L has to be assigned to the
network any time a dashed line (or equivalently a full line) crosses the edge between two
plaquettes without bending. In the other case an energy L has to be assigned to any right
angle of dashed lines between neighboring plaquettes (see Fig. 2).
Differently, short edge bending energy results in different weights for the 32 basic
plaquette configurations. In the reference orientation state of Fig. 1, in each square the
two short edges on the same diagonal always have opposite orientations. These two short
edges acquire the same sign for each folding relative to their perpendicular diagonal. In this
case two triangles are folded and an energy 2S has to be assigned to the network. In the
loop formulation this is equivalent to have a couple of parallel lines in one plaquette with
the same sign. Similar considerations hold for an unfolded edge and for the other couple
of parallel lines in a plaquette. Therefore, in total, each plaquette can have an energy equal
to 4S , 0, or 4S [5].
2.3. The vertex representation
From the discussion above it is clear that our model can be thought as a vertex model
defined on a two-dimensional square lattice Z2 . A vertex variable x {1, . . . , 32} is
associated to each site x of the lattice; to each value of x corresponds one of the 32
vertices introduced above (a possible representation is shown in Fig. 3).
Constraints between neighboring vertices are implied by the sign conservation law
along the closed loops. By using the labelling of Fig. 3 these constraints can be written
in a simple form: we define the two 32 32 matrices h (, 0 ) = u2 ( ),u4 ( 0 ) and
v (, 0 ) = u3 ( ),u1 ( 0 ) where , 0 {1, . . . , 32} and u( ) is the vector u associated to
the vertex . A configuration = {x }Z2 is allowed if h (x , y ) = 1 and v (x , y ) = 1,
respectively, for any horizontal, hx, yih , and vertical, hx, yiv , pair of nearest neighboring
sites. Notice that on each bond there are only 256 allowed configurations out of the 322 =
1024 total ones. We remark that when we write hx, yih,v we suppose the sites x and y
lexicographically ordered.
The Hamiltonian of the model is in the form
X
X
X
H s (x ) +
H h (x , y ) +
H v (x , y ),
(2.1)
H ( ) =
xZ2
hx,yih
hx,yiv
where the two body potential is given by the two 32 32 matrices H v ( 0 , ) =

H h (, 0 ) = 2L v( ),1v( 0) L for any , 0 {1, . . . , 32} and the single body potential
588
Fig. 3. A possible representation of the vertex variables. Each vertex is labelled by a variable
{1, . . . , 32}. Following [5] a vector u = (u1 , u2 , u3 , u4 ) {0, 1, 2, 3}4 and a scalar v {0, 1}
are associated to any {1, . . . , 32} (see [4] for more details).
is given by the vector H s ( ) = 4S if = 1, . . . , 8, H s ( ) = S if = 25, . . . , 32 and

H s ( ) = 0 otherwise. Finally we can write the partition function of the model:

X Y
Y

h
v
(x , y )
(x , y ) exp H ( ) ,
(2.2)
Z(S , L ) =
hx,yih
hx,yiv
where the inverse temperature has been adsorbed in the Hamiltonian.
589
3. Ground states
The ground states of the network can be conveniently discussed in terms of loop
configurations. First consider what happens for S , L > 0 at very low temperatures.
A large positive L selects states with straight lines so that in a square lattice with periodic
boundary conditions there are the dashed and full closed loop shown in Fig. 4(a). A positive
value of S favors alternate sign values on these lines. This corresponds to a completely
flat state with an energy per plaquette given by E1 = 4S 2L . There are 8 possible
degenerate ground states of this kind and each of them can be realized with two basic
plaquette states.
Fig. 4. The ground states depicted in figures (a), (b), (c) and (d) correspond, respectively, to the
phases Fl, SF, F and LF in Fig. 6.
590
In the sector S < 0, L > 0 elementary triangles want to be folded along diagonals.
A positive value of L still select straight lines. All the dashed lines have the same sign
and this holds for the full lines, too. The network is in a state where two triangles sharing
a long edge are on the same plane and all the other triangles are above one of those two.
Also in this case the degeneracy is 8 while one plaquette state is enough for building the
global state which has an energy E2 = 4S 2L per plaquette and is reported in Fig. 4(b).
Loop-configuration ground states for the remaining sectors S < 0, L < 0 and S > 0,
L < 0 are, respectively, shown in Fig. 4(c) and Fig. 4(d). They have an energy equal to
E3 = 4S + 2L and E4 = 4S + 2L . They describe a completely folded state with
all triangle one above the other, and a state with four triangles with a common vertex on
the same plane and all other triangles folded above. There are needed 2 and 4 different
plaquettes for building up ground states in sectors 3 and 4, respectively.
4. The CVM approximation

In this section we describe the CVM approximation we used to study the phase diagram
of model (2.2). We do not enter into all the details related to the CVM technique, we just
refer to [1416].
The CVM is based on the minimization of the free-energy density functional obtained
by a truncation of the cluster (cumulant) expansion of the corresponding functional
appearing in the exact variational formulation of statistical mechanics. The existence of
different horizontal and vertical interactions in the Hamiltonian (2.1) and the structure
of the ground states of the model suggest that one should consider at least a square
(four-vertex) approximation, that is one should consider a square of four vertices as
the largest cluster in the expansion of the free-energy functional. Moreover, in order to
reproduce the structure of the ground states we are forced to partition our lattice into four
square lattices with spacing two (see Fig. 5). We denote by A, B, C and D these four
sublattices and we introduce four square density matrices: ABCD , BADC , CDAB and
DCBA . We also need four pair density matrices AB , BA , CD and DC for the four
horizontal different bonds, four pair density matrices AC , CA , BD and DB for the
four different vertical bonds and four single site matrices A , B , C and D . Finally,
we define the sets P = {ABCD, BADC, CDAB, DCBA}, Lh = {AB, BA, CD, DC},
Lv = {AC, CA, BD, DB} and S = {A, B, C, D}, and we write the free-energy functional
per plaquette as follows:
1 X

1 X

1X

TrX X H h +
TrX X H v +
TrX X H s
f {X : X P} =
4
4
4
v
h
XL
XL
1
4
TrX [X log X ]
XP
X
XLv
TrX [X log X ] +
X
XLh
X
XS
XS
TrX [X log X ]
!
TrX [X log X ] +
X
XP

X Tr[X ] 1 ,
(4.1)
591
where X , with X P, are four Lagrange multipliers ensuring that X with X P are
correctly normalized and for any X P Lh Lv S we have denoted by TrX the sum
over all the allowed configurations on the set X.
Following the recipe of the CVM one should find the densities {X : X P} that
minimize the functional (4.1). Hence the next step consists in taking derivatives of the
free energy with respect to the square densities. In order to do this we must take into
account that bond and single site densities can be obtained from square densities via a
partial tracing. Namely,
AB = TrCD ABCD ,
BA = TrDC BADC ,
CD = TrAB CDAB ,
DC = TrBA DCBA ,
AC = TrBD ABCD ,
BD = TrAC BADC ,
CA = TrDB CDAB ,
DB = TrCA DCBA ,
A = TrB AB ,
B = TrA BA ,
C = TrD CD ,
D = TrC DC .
(4.2)
Actually, each bond or single site density can be derived via partial tracing of different
higher order densities. This means that suitable Lagrange multiplier must be introduced
to ensure that different partial tracings lead to the same result. More precisely, each bond
belongs to two different squares (see Fig. 5): we have to associate a family of multiplier
to each bond to ensure that the same result is obtained by tracing over the two plaquettes
sharing the bond itself (for instance, we need TrCD ABCD = TrCD CDAB ). We get eight
different families {X : X Lh Lv } and each family contains 256 different multipliers,
one for each allowed bond configuration. Moreover, there exist four different bonds sharing
Fig. 5. Partition of the lattice used in the CVM calculation. The eight bonds and the four plaquettes
are explicitly depicted.
592
the same single site: to each site we associate three different multiplier families {X,i :
X S, i = 1, 2, 3}, each of them made of 32 different multipliers. Hence we have 20
different families of multipliers resulting in a total number of 8 256 + 12 32 = 2432
multipliers.
The functional that we have to minimize is no more the free energy (4.1), but the one in
which all the Lagrange multipliers are introduced:

g {X : X P} = 4f {X : X P} + TrA A,1 TrB [BA AB ] +

(4.3)
+ TrAB AB TrCD [CDAB ABCD ] + ,
where dots stand for other seven similar bond terms and eleven similar single site terms.
Now, let us label with {1, . . . , 4608} the allowed 4608 square states. We denote by
X () the vertex associated to the site X S corresponding to . To obtain the equilibrium
densities we have to set equal to zero the derivatives of the functional (4.3) taken with
respect to Y () with Y P and {1, . . . , 4608}. In the case Y = ABCD we obtain:
ABCD ()
n

= const exp H h A (), B () H v A (), C () H s A ()

+ AB A (), B () CD C (), D () + AC A (), C ()

BD B (), D () + A,1 A () + A,2 A () B,1 B ()

o
B,3 B () C,2 C () + C,3 C ()

1
AB A (), B () AC A (), C () A A () .
(4.4)
The equation above is actually a set of 4608 equations. Three similar sets can be found
by considering the cases X {BADC, CDAB, DCBA}. The complete set of 4 4608 =
18432 equations, together with the equations for the multipliers, can be solved by means
of the natural iteration method [18,19]. The iterative equations for the multipliers can be
obtained by taking derivatives of (4.3) with respect to the multipliers and are of the form
(X ) = (k)
(k+1)
X
X (X ) + const log
1 (X )
,
2 (X )
(4.5)
(k)
where X (X ) stands for the value of a multiplier at the k-th iteration, X Lh Lv S,

X is an allowed configuration on X and 1 and 2 are linear combinations of higher order
densities on clusters Y X traced over Y \ X.
5. Phase diagram
The phase diagram of the model, as predicted by our approximation, has been reported
in Fig. 6, where open symbols denote second order transitions, and full symbols denote
first order transitions. The transition lines have been drawn using a limited number of
points, since the precise determination of such points is a very computationally demanding
task. In particular, the transitions between the phase CO (color-ordered) and the phases F
593
Fig. 6. The phase diagram: open (full) symbols denote second (first) order transition points calculated
via CVM. Solid lines are eye-guides, dashed lines are suggested by the limiting behavior.
(folded) and LF (L-folded) have been simply sketched since it was possible to determine
their location only with a rough approximation.
The phases which appear in the diagram can be easily classified on the basis of
the symmetries of the model, which are manifest in the loop formulation. In the loop
formulation, an edge of the lattice is characterized by two variables: the color (dashed
or full) and the sign (+ or ). To each variable is associated a global Ising symmetry
of the model and the equilibrium phases can therefore be characterized by telling which
symmetries are broken in each phase. Order parameters can be easily defined for each
symmetry as the average values of the sign and color variables (the latter defined by
associating, e.g., +1 to a full edge and 1 to a dashed one).
First of all we have four long-range ordered phases, which break both sign and color
symmetries, corresponding to the four possible ground states described in Section 3. Phase
Fl (flat) is stable for large and positive S and L and represents a flat phase. Phase SF
(S-folded) is stable for large enough absolute values of S < 0 and L > 0. It represents
a partially folded phase in which folding occurs mainly along short edges. Phase F is
stable for large enough absolute values of S < 0 and L < 0. It represents a completely
folded phase. Phase LF is stable for large enough absolute values of S > 0 and L < 0. It
represents a partially folded phase in which folding occurs mainly at long edges. Like the
corresponding ground states, all these phases have degeneracy 8.
594
On the high temperature side of the ordered flat phase Fl we have a small slice of
the disordered phase D. In the central part of the phase diagram, we have the colorordered phase CO (only color symmetry is broken), which has larger entropy and larger
energy than D. It is noteworthy that the entropy of this phase at the point S = L = 0 is
S ' 0.9204 in our approximation, while the estimate in [4], obtained by transfer matrix
methods, corresponds, with our definitions, to S ' 0.9196.
Between phases SF and F we have the sign-ordered phase SO (only sign symmetry is
broken). For large negative S the transitions between this phase and phases SF and F tend
asymptotically to |L | = 12 ln(5 + 17)/4 ' 0.412, which is exactly the estimate for the
critical coupling of the square lattice Ising model in the present approximation [14]. This
is a consequence of a property of the model, which reduces to a square lattice Ising model
in the limit S . In the loop gas formulation, S implies that all loops in the
same set (that is, all dashed loops, or all full loops) must have the same sign. Looking at
Fig. 3 one can verify that to satisfy this condition the state of the system must be a mixture
of only 2 out of the 32 plaquette states. For this pair of allowed states there are four different
possibilities (and thus the phase SO has degeneracy 4): (1,5) (corresponding to + signs on
both loop sets), (2,6) (+ on full loops and on dashed loops), (3,7) ( on both loop
sets) and (4,8) ( on full loops and + on dashed loops). Hence we have the breaking of
the sign symmetry, which is restored only at the transition to the phase CO. Given a pair
of allowed states, the Hamiltonian reduces to an ordinary Ising Hamiltonian, with equal
states on adjacent plaquettes giving a contribution L to the total energy, and different
states giving +L . It is therefore an exact result that, in the limit S , there must
exist the three phases SF, F and SO (corresponding, respectively, to the ferromagnetic,
antiferromagnetic and disordered phases of the Ising model), separated by second order
phase transitions at |Lc | = 12 ln(1 + 2 ) ' 0.441.

A similar situation occurs in the limit L . Looking at Fig. 4 one sees that in this
limit both the full and the dashed lines form the smallest possible square loops. It follows
that the model reduces to two decoupled Ising models, where the Ising variables are the
loop signs and two loops (both full, or both dashed) having parallel edges in the same
plaquette interact with an energy 2S . Therefore the phases F and LF (corresponding,
respectively, to the antiferromagnetic and ferromagnetic phases of the limiting Ising model)
will undergo second order phase transitions towards the Color-ordered phase CO at |Sc | =
1
4 ln(1 + 2) ' 0.220. The estimate for this value in our approximation is of course |S | =
ln 5+4 17 ' 0.206.

Comparing our phase diagram with that by Di Francesco [5] one can see several striking
differences. First of all, among our ordered phases, only the completely flat and completely
folded ones were reported in [5], and no low temperature transition was found among them.
In addition, no intermediate phase (like our phases CO and SO) was found between the
ordered and the disordered phases. We note that the CVM results are fully confirmed by
the ground states analysis and the limits S and L .
1
4
595
6. Discussion
We have shown that the square-diagonal folding model has a rather complex phase
diagram exhibiting ordered, partially ordered and disordered phases. Here we shall discuss
a couple of issues regarding these phases, which can not be directly deduced from the
above analysis.
The first issue is that in the absence of interactions (S = L = 0) the stable phase
is not the disordered one. This might be ascribed to the presence of constraints in the
model, which make the S = L = 0 point differ from the true infinite temperature one (the
constraints can be viewed as additional interactions of infinite strengths, while at infinite
temperature all interactions should vanish).
Another interesting issue is why the partial ordered phases appear only in the S
and L limits and not in the opposite ones. Indeed, two different answers are
needed here.
In the limit L + the color symmetry is broken and one has no corners (except
possibly at the boundaries). Therefore, looking at the sign variables, we have Ising chains
in which all spins are constrained to have the same sign. When these chains interact, an
infinitely small coupling KS is sufficient to break also the sign symmetry and we get one
of the fully ordered phases Fl or SF.
On the other hand, for S + adjacent loops must have opposite signs. As a
consequence, when the loops are not arranged in a regular way we have frustration. The
disordered phase D might extend on the whole positive KS axis, but the CVM usually fails
to reproduce the absence of long-range order in frustrated two-dimensional models, e.g.,
in the triangular Ising antiferromagnet, and therefore such a feature could be real but not
reproduced by our approximation.
In conclusions, we have shown that the CVM techniques can be adapted to study
very complex vertex models as the one considered in this paper representing the folding
configurations of a square-diagonal lattice. As in the case of the folding of the triangular
lattice, a first-order crumpling transition between the flat phase and a disordered phase has
been found. The extension of this analysis to the d-dimensional folding problem appears
not easy from a numerical point of view.
Acknowledgements
We thank the referee for a clarifying comment. G.G. acknowledges support by PRA1999 INFM and MURST (PRIN 99). E.C. wishes to express his thanks to the European
network Stochastic Analysis and its Applications ERB-FMRX-CT96-0075 for financial
support.
References
[1] D.R. Nelson, T. Piran, S. Weinberg (Eds.), Statistical Mechanics of Membranes and Surfaces,
World Scientific, Singapore, 1989.
596
[2] D.R. Nelson, L. Peliti, J. Phys. (France) 48 (1987) 1085.

[3] M. Bowick, P. Di Francesco, O. Golinelli, E. Guitter, in: Proc. of the 4th meeting on Condensed
Matter and High Energy Physics, World Scientific, Singapore, 1997.
[4] P. Di Francesco, Nucl. Phys. B 525 (1998) 507.
[5] P. Di Francesco, Nucl. Phys. B 528 (1998) 453.
[6] Y. Kantor, M.V. Jaric, Europhys. Lett. 11 (1990) 157.
[7] P. Di Francesco, E. Guitter, Europhys. Lett. 26 (1994) 455.
[8] R.J. Baxter, J. Math. Phys. 11 (1970) 3.
[9] P. Di Francesco, E. Guitter, Phys. Rev. E 50 (1994) 4418.
[10] E.N.M. Cirillo, G. Gonnella, A. Pelizzola, Phys. Rev. E 53 (1996) 1479.
[11] M. Bowick, P. Di Francesco, O. Golinelli, E. Guitter, Nucl. Phys. B 450 (1995) 463.
[12] E.N.M. Cirillo, G. Gonnella, A. Pelizzola, Phys. Rev. E 53 (1996) 3253.
[13] M. Bowick, O. Golinelli, E. Guitter, S. Mori, Nucl. Phys. B 495 (1997) 583.
[14] R. Kikuchi, Phys. Rev. 81 (1951) 988.
[15] G. An, J. Stat. Phys. 52 (1988) 727.
[16] T. Morita, J. Stat. Phys. 59 (1990) 819.
[17] Prog. Theor. Phys. Suppl. 115 (1994).
[18] R. Kikuchi, J. Chem. Phys. 60 (1974) 1071.
[19] R. Kikuchi, J. Chem. Phys. 65 (1976) 4545.

Correlators in integrable quantum field theory:

the scaling RSOS models
G. Delfino
Laboratoire de Physique Thorique et Hautes Energies, Universit Pierre et Marie Curie, Tour 16 1er tage,
4 place Jussieu, 75252 Paris cedex 05, France
Received 7 December 1999; accepted 23 May 2000
Abstract
The study of the scaling limit of two-dimensional models of statistical mechanics within the
framework of integrable field theory is illustrated through the example of the RSOS models. Starting
from the exact description of regime III in terms of colliding particles, we compute the correlation
functions of the thermal, 1,2 and (for some cases) spin operators in the two-particle approximation.
The accuracy obtained for the moments of these correlators is analysed by computing the central
charge and the scaling dimensions and comparing with the exact results. We further consider the
(generally non-integrable) perturbation of the critical points with both the operators 1,3 and 1,2
and locate the branches solved on the lattice within the associated two-dimensional phase diagram.
Finally we discuss the fact that the RSOS models, the dilute q-state Potts model at and the O(n)
vector model are all described by the same perturbed conformal field theory. 2000 Elsevier Science
1. Introduction
Integrable field theory emerged in the last years as an elegant and effective tool for
the study of many two-dimensional statistical models directly in their scaling limit. The
approach relies on the fact that a large class of quantum field theories in 1 + 1 dimensions
admits an infinite number of integrals of motions (i.e., they are integrable) and can be
completely solved on-shell [1,2]. The matrix elements of local operators on the asymptotic
states are also exactly computable [3,4] and lead to spectral series for the correlation
functions whose quantitative effectiveness is remarkable. Among the statistical models
that have been studied in this framework we mention the Ising model in a magnetic field
[2,58], the q-state Potts model [912], the O(n) model [13,14], and the AshkinTeller
model [15].
aldo@lpthe.jussieu.fr
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 2 4 - 2
598
G. Delfino / Nuclear Physics B 583 [FS] (2000) 597613
It is the purpose of this paper to illustrate how the program outlined above applies to the
restricted solid-on-solid (RSOS) models introduced by Andrews, Baxter and Forrester
in Ref. [16]. For any integer p > 3 they are defined on the square lattice in terms of a spin
or height variable hi located at each site i and taking the integer values from 1 to p. The
heights of two nearest-neighbour sites i and j are constrained by the condition
|hi hj | = 1,
(1.1)
which, in particular, leads to a natural splitting of the lattice into two sublattices on which
the heights are all even or all odd. The Hamiltonian is further specified by a one-site and
a diagonal interaction terms. Their precise form does need to be reproduced here but it is
important that the energy of a configuration is invariant under the global transformation
hi p + 1 hi ,
(1.2)
which is the basic symmetry of the models.

In their general formulation the RSOS models contain a number of parameters which
grows linearly with p. In Ref. [16] the models were solved on two two-dimensional
manifolds of the parameter space which can be parameterised by a temperature-like
variable t together with a second coordinate v measuring the spatial anisotropy of the
lattice interaction. The scaling limit, however, is isotropic and v will be ignored in the
following. Hence, for each of the two solutions and for any p, the phase diagram reduces
to a line and exhibits a critical point separating two phases known as regimes I and II (for
the first solution) and regimes III and IV (for the second solution). Here, we will only be
interested in the second case, and more specifically in regime III.
By comparison of critical exponents, Huse showed [17] that the critical points separating
regimes III and IV for the different values of p correspond to the minimal unitary series of
conformal field theories characterised by the values of the central charge [18,19]
C =1
6
,
p(p + 1)
p = 3, 4, . . . .
(1.3)
These models contain a finite number of primary operators m,n (x) (m = 1, . . . , p 1;

n = 1, . . . , p) with scaling dimensions
[(p + 1)m pn]2 1
.
(1.4)
2p(p + 1)
It was shown in Ref. [20] that the unitary minimal models admit the LandauGinzburg
description
"
#
Z
p1
X
2
2
2k
gk ,
(1.5)
S = d x () +
Xm,n =
k=1
with g1 = g2 = = gp2 = 0. The scalar field (x) is then the continous version of the
shifted height variable hi (p + 1)/2, in such a way that the reflection symmetry (1.2) is
mapped into . The following identifications between normal ordered powers of
and conformal operators hold [20]
k k+1,k+1 ,
k = 0, . . . , p 2.
(1.6)
599
The effective action (1.5) makes transparent that the considered series of critical points
corresponds to the (p 1)-critical behaviour of a system with scalar order parameter and
Z2 symmetry, p = 3 being the ordinary Ising universality class, p = 4 the tricritical Ising,
and so on.
The RSOS models in regimes III and IV possess p 1 and p 2 degenerate ground
states, respectively [16], and can be described by the action (1.5) with suitable fine tunings
of the couplings gk leading to the appropriate number of degenerate minima in the effective
potential. For p = 3 the two regimes correspond to the low- and high-temperature phases
of the Ising model in zero magnetic field and are both massive in the scaling limit. For
p > 4, while regime III is still massive, regime IV become massless and corresponds to the
crossover between the critical points labelled by p and p 1 [21,22].
Solvability on the lattice naturally suggests integrability of the field theory describing
the scaling limit. In fact, the scaling dimension of the thermal operator (conjugated to t)
is known from the lattice solution and coincide with X1,3 , so that the scaling limit of the
RSOS models in regimes III and IV is described by the action
Z
(p)
(1.7)
A = ACFT + d 2 x 1,3 (x),
(p)
where ACFT is the action of the conformal theories with central charge (1.3), and is
a coupling with dimensions m2X1,3 . This 1,3 perturbation of conformal field theory is
known to be massive (regime III) or massless (regime IV) depending on the sign of
[21], and to be integrable in both directions [2]. The associated scattering theories are also
known [2325].
The paper is organised as follows. In the next section we briefly review the exact
scattering description for regime III and use it in Section 3 for the computation of form
factors of the operators 1,3 , 1,2 and (for p = 3, 4) 2,2 . In Section 4 we write down the
two-particle approximation for the correlation functions of these operators and analyse its
accuracy by computing the central charge and the scaling dimensions. Section 5 is devoted
to a discussion of the perturbation of the RSOS critical points with both the operators
1,3 and 1,2 . In the final section we briefly discuss the fact that the perturbed conformal
field theory (1.7) also describe the scaling limit of the dilute q-state Potts model along
its first order phase transition lines for 0 6 q 6 4, as well as the O(n) vector model for
2 6 n 6 2.
2. Scattering theory
In a (1 + 1)-dimensional theory with degenerate vacua the elementary excitations are
kinks interpolating among these vacua. It is known from the lattice solution that the j th
ground state in regime III (j = 1, . . . , p 1) is such that all the sites on one sublattice have
height j and all the sites on the other sublattice have height j + 1 (Fig. 1). The space
time trajectory of a kink is a domain wall separating two different ground states. Since the
pairing of two different ground states i and j can give an admissible configuration only
600
Fig. 1. The ground states 1, 2 and 3 of regime III for p = 4.
Fig. 2. Domain wall (segment) between the ground states 1 and 2. It corresponds to the spacetime
trajectory of the kink K12 .
if |i j | equals 1 (Fig. 2), the elementary excitations of the scattering theory are kinks 1
Kij ( ) interpolating between two vacua i and j = i 1. It follows from the precise form
of the lattice interaction that the interfacial tension between two ground states i and i + 1
does not depend on i, and this amounts to say that the kinks Ki,i1 all have the same mass.
Multikink excitations of the type
Ki1,i (1 )Ki,i1 (2 )
(2.1)
will connect ground states with arbitrary indices.

In an integrable field theory the scattering is completely elastic (no production
processes allowed) and multiparticle processes factorise into the product of the two-body
subprocesses, so that the problem of the determination of the S-matrix is reduced to the
computation of the two-kink amplitudes [1]. Taking into account the kink composition
rules together with invariance under time reversal and spatial inversion, the allowed
two-kink processes are those depicted in Fig. 3 and associated to the commutation
relations
Kj,j 1 (1 )Kj 1,j (2 ) = A
j (1 2 )Kj,j 1 (2 )Kj 1,j (1 )
+ Bj (1 2 )Kj,j 1 (2 )Kj 1,j (1 ),
Kj 1,j (1 )Kj,j 1 (2 ) = Cj (1 2 )Kj 1,j (2 )Kj,j 1 (1 ).
(2.2)
(2.3)
(2.4)
The scattering amplitudes are subject to a series of constraints. Invariance under the reflection j p j requires
1 The rapidity variable parameterises the on-shell momenta of the kink of mass m as (p 0 , p 1 ) =
(m cosh , m sinh ).
601
Fig. 3. The two-kink scattering amplitudes A

j , Bj and Cj .
A+
j ( ) = Apj ( ),
(2.5)
Bj ( ) = Bpj ( ),
(2.6)
Cj ( ) = Cpj ( ),
(2.7)
while crossing symmetry implies
A
j ( ) = Aj 1 (i ),
Bj ( ) = Cj (i ).
(2.8)
(2.9)
Commuting once again the r.h.s. of Eqs. (2.3) and (2.4) leads to the unitarity equations
A
j ( )Aj ( ) + Bj ( )Bj ( ) = 1,
(2.10)
A
j ( )Bj ( ) + Bj ( )Aj ( ) = 0,
(2.11)
Cj ( )Cj ( ) = 1.
(2.12)
A three-kink process can be factorised in two ways differing by the ordering of the twobody collisions. Equating the results leads to the factorisation equation

A
j Aj 1 Aj + Bj Cj Bj = Aj 1 Aj Aj 1 + Bj 1 Cj 1 Bj 1 ,
(2.13)
and similar others (the arguments of the three factors in each product are , + 0 and 0 ,
respectively).
The minimal solution to all these requirements is well known [16,23,24] and reads
A
j ( ) =
sj 1
sj
i/
1
s1 sinh p (ij )
S0 ( ),
sj sinh p1 (i )
(2.14)

sinh p
sj +1 sj 1 1+i/
S0 ( ),
Bj ( ) =
sj
sinh p1 (i )
(2.15)

1
sj +1 sj 1 i/ sinh p (i )
S0 ( ),
sj
sinh p1 (i )
(2.16)
Cj ( ) =
where
sj sin
j
,
p
(2.17)
602

0 1+
Y
2
2 + ip 0 1 + p n
S0 ( ) =

0 1 + n + 12 ip
0 1 + p2 n +
n=0
2

0 p n + 12 + ip
0 p2 (n + 1) ip
2

0 p2 (n + 1) + ip
0 p n + 12 ip
)
( Z
x
dx sinh(p 1) x2
.
= exp i
x sin
x sinh px
2 cosh 2
0
2
p
2
p
n+
ip
ip
(2.18)
It can be checked that the amplitudes do not possess poles in the physical strip Im
(0, ), what ensures that there are no bound states and that the amplitudes given above
entirely determine the scattering theory.
3. Form factors
Let us denote 8(x) a local scalar operator of the theory with zero topological charge,
namely such that its action on the vacuum |0j i only produces excitations beginning and
ending on this vacuum. All the operators we will consider in the following share this
property. We are interested in the two-particle form factors (Fig. 4)
8
(1 2 ) = h0j |8(0)|Kj,j 1 (1 )Kj 1,j (2 )i.
Fj,
(3.1)
Eq. (2.3) implies the relation

8
8
8
( ) = A
Fj,
j ( )Fj, ( ) + Bj ( )Fj, ( ),
(3.2)
while crossing leads to the equations [10]

8
( + 2i) = Fj81, ( ),
Fj,
(3.3)
8
( ) = i Res=i Fj81, ( ) = h0j |8|0j i h0j 1 |8|0j 1 i.
i Res=i Fj,
(3.4)
As a last necessary condition, the two-kink form factors are subject to the asymptotic bound
[5]
8
( ) 6 constant eX8 /2 ,
lim Fj,
where X8 denotes the scaling dimension of the operator 8(x).
8 of the operator 8.
Fig. 4. The two-kink form factor Fj,
(3.5)
It is easily checked that a class of solutions of Eqs. (3.2) and (3.3) is given by

2i sj 1 (1+i/)/2
F0 ( )
j ( ),
Fj, =
1
p
sj
sinh p ( i)
603
(3.6)
where {} {8},
(Z
)
x
dx sinh(1 p) x2 sin2 (i ) 2
F0 ( ) = i sinh exp
x
2
x sinh px
sinh x
2 cosh 2
(3.7)
is solution of the equations

F0 ( ) = S0 ( )F0 ( ),
(3.8)
F0 ( + 2i) = F0 ( ),
(3.9)
and behave as

F0 ( ) exp (1 + 1/p)/4 ,
+.
(3.10)
The functions j ( ) are free of poles and satisfy
j ( ) = j ( ),
(3.11)
j ( + 2i) = j 1 ( ).
(3.12)
These requirements, together with (3.5) imply that the j ( ) are polynomials in
cosh(/2). Let us consider those operators (x) which are relevant in the renormalisation
group sense (X < 2) for all values of p > 3. Then, the bound (3.5) implies that these
polynomials are at most of degree one, which means that the operator subspace {}
contains only two independent relevant operators. The trace of the stress-energy tensor
2 ( ) =
2(x) is a relevant operator and is even under the reflection symmetry, so that Fj,+
2
Fpj, ( ). Moreover h0j |2|0j i does not depend on j and, according to (3.4), the twokink matrix elements have no pole in = i . All these requirements are fulfilled if we
take
(3.13)
j2 ( ) = 2m2 cosh ,
2
with the normalisation constant fixed by
2
(i) = 2m2 .
Fj,
(3.14)
The other independent relevant operator in {} (let us denote it E(x)) corresponds to the
constant solution
jE ( ) = (1)j h01 |E|01 i.
(3.15)
The order parameter (x) is the most relevant operator which changes sign under
reflection, and this means in particular
h0j ||0j i = h0pj ||0pj i,
(3.16)
604
Fj,+ ( ) = Fpj, ( ).
(3.17)
For generic values of p, these properties are incompatible with those of the space of
solutions spanned by (3.6), (3.13), (3.15) and we conclude that 6 {}. We do not dispose
of the solution for Fj,+ ( ) for generic p and just quote the result for the two simplest cases
F1,+ ( ) = i h01 ||01 i tanh ,

2
Fj,+ ( ) =
p = 3,
(3.18)

sj +1 (1+i/)/2 F0 ( )
(1)j
h01 ||01i
j ( ),
2j (i)
sj
cosh 2
p = 4.
(3.19)
The functions
)
(Z
x
dx sin2 [2i|2 j | ] 2
j ( ) = exp
x
cosh x sinh 2x
(3.20)
satisfy the equations

1 ( ) = 3 ( ) =
sinh 14 (i + )
sinh 14 (i )
1 ( + 2i) = 2 ( ),
1 ( ),
(3.21)
(3.22)
and behave as exp(/8) when +. It can be checked that

p = 3.
Fj, ( )
E ( ) for
= Fj,
4. Correlation functions
The correlation functions are obtained using the resolution of the identity
Z
X
d1 dn
|nihn|
1=
(2)n
(4.1)
n=0 >>
n
1
to sum over all intermediate n-kink states |ni. A two-point function reads 2
X Z d1 d2 8
82
Fj,1 (1 2 )Fj,
h0j |81 (x)82 (0)|0j i =
(2 1 ) e|x|E2
2
2
=
(4.2)
1 >2

+ O e4m|x| ,
m|x| 1,
(4.3)
where E2 = m(cosh 1 + cosh 2 ) is the energy of the two-kink asymptotic state. The
two-kink approximation (4.3) is known to provide results of remarkable accuracy for
integrated correlators (see [10] and references therein), as can be checked through the use
of the sum rules [26,27]
2 Here and in the following we always refer to connected correlators.
605
Fig. 5. The two-kink approximation for the central charge sum rule (4.4) (dots) compared with the
exact formula (1.3) (continous line).
3
C=
4
d 2 x |x|2 h0j |2(x)2(0)|0j i,

Z
1
d 2 x h0j |2(x)8(0)|0j i,
X8 =
2h0j |8|0j i
(4.4)
(4.5)
allowing the determination of the ultraviolet conformal data (central charge and scaling
dimensions) in the form of moments of off-critical correlators.
In Fig. 5 we compare the exact formula for the central charge (1.3) with the result yielded
by the two-kink approximation (4.3) in the sum rule (4.4). Notice that p can be considered
as a continous parameter in the result of the latter computation, in agreement with the fact
that many observables in the theory (1.7) have a continous p-dependence. With this remark
in mind, in the remaining part of this section we will treat p as a real number 3 > 1.
For p = 1, in particular, the two-kink form factor (3.6), (3.13) simply reduces
to 2m2 (1)(1+i/)/2 . This residual rapidity dependence ensures the normalisation
condition (3.14) but is immaterial in the computation of the correlator (4.3). Hence, the
theory (1.7) at p = 1 is free and the two-kink computation for the central charge gives the
exact result C = 2. Due to the equivalence between the Ising model and a free neutral
fermion, the two-kink approximation is exact also for p = 3 and gives C = 1/2.
The computation of X2 through the sum rule (4.5) requires the knowledge of h2i.
2
( ) due to the vanishing of the residue
Although this quantity cannot be related to Fj,
(3.4), we can use the thermodynamic Bethe ansatz result [25]
3 In doing that, of course we loose unitarity unless p is an integer larger than 2. Here and in the following the
term unitarity refers to the absence of negative norm states in the Hilbert space.
606
Fig. 6. The two-kink approximations for the scaling dimension sum rule (4.5) for the operators 2(x)
(upper dotted line) and E (x) (lower dotted line). They are compared with the exact formulae (4.9)
and (4.10), respectively (continous lines).
p
.
(4.6)
2
Since 1,3 (x) is the operator which drives the theory away from criticality, we must have
h2i = m2 tan
2(x) 1,3 (x),
(4.7)
and X2 = X1,3 .
The two-kink computation of XE through (4.5) gives 1/4 at p = 1 and 1/8 at p = 3.
Since the theory is free for these two values of p, these results are expected to be exact.
Assuming that E(x) corresponds to a primary operator whose position in the Kacs table
does not depend on p, they can be substituted into the formula (1.4) to fix
E(x) 1,2 (x).
(4.8)
In Fig. 6 we compare the two-kink approximation for X2 and XE with the exact formulae
p1
,
p+1
p2
,
X1,2 =
2 (p + 1)
X1,3 = 2
(4.9)
(4.10)
respectively. Concerning the two values of p for which we determined the form factors of
the order parameter, the two-kink computation gives X = 1/8 at p = 3 and X = 0.0734
at p = 4. In view of the identification (x) 2,2 (x), these results must be compared with
the exact values 1/8 and 3/40 = 0.075, respectively.
607
Some comments are in order about the results yielded by the sum rules (4.4) and (4.5).
Consider the moment
Z
(4.11)
Ik = d 2 x |x|k h81 (x)82 (0)i,
and denote 83 (x) the leading operator determining the short distance behaviour
h81 (x)82 (0)i
h83 i
,
|x|8182
|x| 0
(4.12)
with
81 82 X81 + X82 X83 .
(4.13)
Then Ik is convergent if 2 + k 81 82 > 0. In the cases of interest for us we have 22 =

2X1,3 and 2E = X1,3 , so that the sum rules for C and X1,2 converge for all finite values
of p, while the sum rule for X1,3 converges only for p < 3. This failure of the sum rule
for the scaling dimension due to the divergence of the integral is originated by operator
mixing under renormalisation and was discussed in Ref. [27].
Let us now discuss the issue of the accuracy of the results obtained using the
approximated correlators (4.3) in the integrals (4.4) and (4.5). The spectral series for
the correlation functions is a large distance expansion and any partial sum including the
contributions up to n particles will appreciably depart from the exact result at sufficiently
small distances. In the moment (4.11), however, the factor |x|k causes a suppression of the
short distance contribution whose importance, for a fixed k, depends on the high energy
behaviour of the form factors. The two-kink contribution to Ik is given by
81
82
Z
(2 )Fj,
(2 )
Fj,
.
(4.14)
d
k+2
(cosh )
(k)
], where
The integrand here behaves asymptotically as exp[68
1 82
(k)
681 82 2 + k y81 y82 ,
(4.15)
with y8 defined by
8
( ) ey8/2 ,
Fj,
+.
(4.16)
In an unitary theory, this exponent is subject to the bound [5]

y8 6 X8 .
(k)
(4.17)
We see that 681 82 has to be positive to ensure the convergence of the integral and
that the suppression of the short distance contribution is proportional to this exponent.
Of course, this observation does not determine the absolute accuracy of the two-particle
approximation, but it helps understanding the accuracy pattern exhibited in Fig. 6. In fact,
the solutions of Section 3 determine
3(p 1)
,
(4.18)
y2 =
2p
p3
.
(4.19)
yE =
2p
608
Table 1
(k)
Accuracy 8 8 of the two particle approximation for the
1 2
kth moment of the correlator h81 (x)82 (0)i for three values
of the exponent (4.15). The numbers in parenthesis are the
corresponding values of p/(p + 1)
(k)
68 8
1 2
0.5
(0)
0.09
(0.66)
22
(2)
22
(0)
2 E
0.03
(0.86)
1.5
2.5
0.004
(0.55)
0.005
(0.86)
0.0008
(0.66)
0.007
(0.66)
0.004
(0.55)
(0)
Then, 622 goes to zero as p 3 and we observe that the deviation of the two-kink
approximation for X2 from the exact result becomes large as we approach this value.
Analogous considerations apply to the case of XE as p . In the central charge sum
(2)
rule (4.4), on the contrary, 622 tends to 1 as p , what means that the high energy
contribution is still strongly suppressed in this limit. The remarkable accuracy (1%) of the
two-kink approximation as p shows that also the contributions with a larger number
(k)
of kinks undergo a similar suppression. The relation between 681 82 and the accuracy of
the two-kink approximation (defined as the absolute deviation from the exact result divided
by the exact result) is illustrated in Table 1 through few examples.
5. Double perturbation and phase diagram

In this section we briefly consider what happens if we add to the action (1.7) the operator
E 1,2 , namely if we take 4
Z
Z
(p)
(p)
(p)
0
2
A = ACFT + d x 1,3 (x) + d 2 x 1,2 (x).
(5.1)
Within the usual conventions for the operator normalisations, the regime III we considered
in the previous sections corresponds to = 0 and < 0. If is very small we can use
form factor perturbation theory [8] around regime III. The correction to the energy density
j of the vacuum |0j i is proportional to the vacuum expectation value of the perturbing
operator 1,2 computed at = 0, and reads (remember (3.15) and (3.4))
j h0j |1,2|0j i = (1)j h01 |1,2 |01 i.
(5.2)
This means that only a subset of alternating vacua among the p 1 degenerate vacua of
regime III preserves the same energy when is negative and is small in (5.1). For p odd,
4 When useful in this section we explicitly label the operators by the superscript (p) identifying the critical
point they refer to.
609
in particular, we see that the Z2 symmetry characteristic of the case = 0 is broken by

the 1,2 perturbation. For p = 3 the action (5.1) describes the Ising model in a magnetic
field [7,8]. If p is odd the number of surviving degenerate vacua is (p 1)/2. If p is even,
instead, this number is p/2 or p/2 1 depending on the sign of .
It is clear that in presence of such a pattern of degeneracy breaking the kinks Kj,j 1 of
regime III are no longer asymptotic excitations when 6= 0. Rather, they will be confined
into pairs Kj,j +1 Kj +1,j +2 providing the new stable kinks in the perturbed theory. This
phenomenon appears in the formalism when we try to compute the correction to the mass
of the kinks Kj,j 1 , which is given by [8,28]
1,2
(i).
m Fj,
(5.3)
Since the form factor on the r.h.s. has a pole at = i (see (3.6) and (3.15)), it follows
that this correction is infinite, a fact that reveals the removal of the kinks Kj,j 1 from the
spectrum of the asymptotic excitations.
The conformal field theories with C < 1 perturbed with one of the operators 1,3 , 1,2
or 2,1 are integrable [2]. It is then natural to look for solvable lattice models whose scaling
limit corresponds to these quantum field theories. In Refs. [29,30] a dilute version of the
RSOS models was considered and found to be solvable on the lattice along four distinct
branches. It was found, in particular, that the scaling limit of branch 2 is described
by the action (5.1) with = 0, and that for p odd this branch possesses (p 1)/2
degenerate ground states. This result is consistent with our perturbative considerations and
suggests that their validity extends away from the perturbative region || 1. This is
not particularly surprising in view of the fact that a modification of the vacuum structure
requires a phase transition.
We already mentioned that for p > 3 the regime IV of the RSOS models ( > 0, = 0
(p1)
in (5.1)) corresponds to a massless flow to the critical point with action ACFT . It is known
(p)
(p1)
that the operator 1,2 renormalises into the operator 2,1 in the infrared limit of this flow
(p1)
[21]. Hence we conclude that the action (5.1) in the limit = + describes the 2,1
(p1)
perturbation of the critical point ACFT . This integrable perturbation was identified in [29,
30] as corresponding to the scaling limit of the lattice models along branch 1. 5 Once
again, such a branch possesses (p 1)/2 degenerate ground states for p odd [29,30], and
we are led to argue that the vacuum structure obtained above for small actually holds
true 6 for any value of 6= 0 in the action (5.1).
The phase diagram associated with the action (5.1) (for p > 4) is shown in Fig. 7.
Keeping the notation |0j i (j = 1, . . . , p 1) we used for the ground states of regime III, we
have that the ground states with j odd are the absolute minima of the free energy density
in the whole region < 0, while those with j even are the absolute minima for > 0.
The two subsets become simultaneously degenerate when = 0: regime III ( < 0) is a
5 The remaining two solvable branches of Refs. [29,30] are not related to perturbations of the critical points
considered in this paper.
6 See Ref. [31] for a discussion of the vacuum structure of the integrable perturbations of conformal field
theories from the point of view of quantum group reduction.
610
Fig. 7. Phase diagram associated to the action (5.1) and illustrating the different integrable branches
of the RSOS models.
line of first order phase transition while a phenomenon of vacuum coalescence makes the
transition continous across regime IV ( > 0).
6. Conclusion
In the central part of this paper we applied the S-matrix-form factor approach to the
regime III of the RSOS models. These models, however, are not the only lattice models
whose scaling limit is described by the 1,3 perturbation of C < 1 conformal field theories. It is well known [3234] that the same action (1.7) corresponds to the scaling dilute
q-state Potts model at the critical temperature and zero external field (with q =
4 cos2 /p [0, 4]), and to the scaling O(n) vector model in zero external field (with
n = 2 cos /p [2, 2]). The latter two models make sense for continous values of q and
n through mapping onto cluster and loop models, respectively. Excepting special values of
p (p = 3 in particular) these three models are characterised by different internal symmetries and then represent different universality classes of critical behaviour. In fact, the order
parameter has a different number of independent components in the three cases and corresponds to different operators (see Table 2). The fact that the three models are described
by the same action along the renormalisation group trajectories specified above means that
the theory (1.7) admits different microscopic descriptions distinguished by the choice of
local observables. 7 Each description is characterised by a specific set of mutually local
operators with well defined transformation properties under the group of internal symmetry. Of course, the perturbing operator 1,3 appears in all these local sets and is invariant
7 Famous examples of this kind of situation are the equivalence between the Ising model and free neutral
fermions, or that between the sine-Gordon and massive Thirring models.
611
Table 2
Some features of the lattice models whose continuum limit is described by the action (1.7). The
notation for the scaling dimensions refers to (1.4)
RSOS models
Dilute Potts model
O(n) model
q = 4 cos2 p
n = 2 cos p
Z2
Sq
O(n)
Order parameter:
number of components
scaling dimension
1
X2,2
q 1
Xp,p
n
X p1 , p+1
Number of vacua
p1
q +1
2(p 2)
2q
Symmetry
Number of elementary excitations
2 2
under the different symmetry groups. The observables associated to this operator, e.g., the
correlation length critical exponent = 1/(2 X1,3 ), are the same in the three cases.
At the conformal level, the possibility of different local descriptions appears through the
existence of different modular invariant partition functions for the same value of the central
charge C [3537]. In the S-matrix approach away from criticality the different nature of
the order parameter leads to the existence of different scattering descriptions for the action
(1.7). They all exhibit the same spectrum and very similar analytic form but differ from
each other by the nature and the number of the elementary excitations (see Table 2). In this
paper we used the scattering description based on the Z2 symmetry which characterises
the RSOS models. The massive dilute q-state Potts model at T = Tc [12,38] has q + 1
degenerate vacua located at the q vertices and at the center of a hypertetrahedron living
in the (q 1)-dimensional space of the independent order parameter components. The
elementary excitations are the 2q kinks interpolating from the center to the vertices, and
vice versa (Fig. 8). In the massive phase of the O(n) model there is a single vacuum and
the elementary excitations are n ordinary particles transforming according to the vector
representation of the group [13,14]. Of course, for non-integer values of q and n the number
of excitations is also non-integer, but this is not more surprising than the appearance of
operators with non-integer multiplicity in the modular invariant partition functions for the
two models at criticality (see [37]).
Fig. 8. Vacua and kinks for p = 6 in the regime III of the RSOS models (a), and in the massive dilute
Potts model at T = Tc (b).
612
The different number of excitations for a given p ensures that there is no one to
one correspondence between the three scattering descriptions, although some connections
certainly exists. 8 For each particle basis the asymptotic states have obvious transformation
properties under the relevant symmetry group and this fact allows a natural identification
within the form factor approach of the interesting operators (for example the order
parameter).
As was discussed above, no matter which particle basis is used, summation over the
intermediate asymptotic states must lead to the same result for the correlation functions
of some invariant operators, in particular the trace of the stressenergy tensor 2(x)
1,3 (x). Since each n-particle contribution to the spectral sum has a distinct large distance
behaviour exp(nmr), the identification is expected to occur term by term. It is easy to
check comparing the results of this paper with those of Refs. [12,14] that this is indeed the
case for the first (two-particle) contribution to h2(x)2(0)i.
Most of the considerations of this section can be extended to the case of the more general
action (5.1).
Acknowledgements
I thank John Cardy for interesting discussions.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.

A.B. Zamolodchikov, Adv. Stud. Pure Math. 19 (1989) 641; Int. J. Mod. Phys. A 3 (1988) 743.
M. Karowski, P. Weisz, Nucl. Phys. B 139 (1978) 445.
F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, World
Scientific, 1992.
G. Delfino, G. Mussardo, Nucl. Phys. B 455 (1995) 724.
G. Delfino, P. Simonetti, Phys. Lett. B 383 (1996) 327.
G. Delfino, Phys. Lett. B 419 (1998) 291.
G. Delfino, G. Mussardo, P. Simonetti, Nucl. Phys. B 473 (1996) 469.
L. Chim, A.B. Zamolodchikov, Int. J. Mod. Phys. A 7 (1992) 5317.
G. Delfino, J.L. Cardy, Nucl. Phys. B 519 (1998) 551.
G. Delfino, G.T. Barkema, J.L. Cardy, Nucl. Phys. B 565 (2000) 521.
G. Delfino, Nucl. Phys. B 554 (1999) 537.
A.B. Zamolodchikov, Mod. Phys. Lett. A 6 (1991) 1807.
J.L. Cardy, G. Mussardo, Nucl. Phys. B 410 (1993) 451.
G. Delfino, Phys. Lett. B 450 (1999) 196.
G. Andrews, R. Baxter, J. Forrester, J. Stat. Phys. 35 (1984) 193.
D.A. Huse, Phys. Rev. B 30 (1984) 3908.
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
8 A relation between the RSOS and dilute Potts scattering theory for p = 6 was pointed out in [38]. The issue
of the relation between the O(n) and RSOS scattering descriptions has been discussed in [39] from the point of
view of quantum group reduction of sine-Gordon model.
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
613
D. Friedan, Z. Qiu, S. Shenker, Phys. Rev. Lett. 52 (1984) 1575.

A.B. Zamolodchikov, Sov. J. Nucl. Phys. 44 (1986) 529.
A.B. Zamolodchikov, JETP Lett. 43 (1986) 730.
A.W.W. Ludwig, J.L. Cardy, Nucl. Phys. B 285 (1987) 687.
A.B. Zamolodchikov, Landau Institute preprint, 1989.
A. LeClair, Phys. Lett. B 230 (1989) 103.
P. Fendley, H. Saleur, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 8 (1993) 5751.
J.L. Cardy, Phys. Rev. Lett. 60 (1988) 2709.
G. Delfino, P. Simonetti, J.L. Cardy, Phys. Lett. B 387 (1996) 327.
S.O. Warnaar, B. Nienhuis, K.A. Seaton, Phys. Rev. Lett. 69 (1992) 710.
S.O. Warnaar, P.A. Pearce, K.A. Seaton, B. Nienhuis, J. Stat. Phys. 74 (1994) 469.
V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Nucl. Phys. B 516 (1998) 652.
Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312.
B. Nienhuis, J. Phys. A 15 (1982) 199.
B. Nienhuis, Phys. Rev. Lett. 49 (1982) 1062.
J.L. Cardy, Nucl. Phys. B 270 (1986) 186.
A. Cappelli, C. Itzykson, J.B. Zuber, Nucl. Phys. B 280 (1987) 445.
P. Di Francesco, H. Saleur, J.B. Zuber, Nucl. Phys. B 285 (1987) 454.
A.B. Zamolodchikov, in: H. Guo, Z. Qiu, H. Tye (Eds.), Fields, Strings and Quantum Gravity,
Gordon & Breach, 1989, p. 349.
[39] F.A. Smirnov, Phys. Lett. B 275 (1992) 109.

The intrinsic coupling in integrable quantum field

theories
J. Balog a , M. Niedermaier b , F. Niedermayer c , A. Patrascioiu d ,
E. Seiler e , P. Weisz e
a Research Institute for Particle and Nuclear Physics, 1525 Budapest 114, Hungary
b Department of Physics, University of Pittsburgh, Pittsburgh, PA 15260, USA
c Institute for Theoretical Physics, University of Bern, CH-3012 Bern, Switzerland
d Physics Department, University of Arizona, Tucson, AZ 85721, USA
e Max-Planck-Institut fr Physik, 80805 Munich, Germany
Received 24 January 2000; accepted 8 May 2000
Abstract
The intrinsic 4-point coupling, defined in terms of a truncated 4-point function at zero momentum,
provides a well-established measure for the interaction strength of a QFT. We show that this coupling
can be computed non-perturbatively and to high accuracy from the form factors of an (integrable)
QFT. The technique is illustrated and tested with the Ising model, the XY-model and the O(3)
nonlinear sigma-model. The results are compared to those from high precision lattice simulations.
2000 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Kk; 11.10.Lm
Keywords: Integrable systems; Bethe ansatz; Form factors; Ising model; XY-model; O(3) sigma-model
1. Introduction
The intrinsic coupling gR , also sometimes called physical or renormalized coupling,
is a quantity of great interest in a Quantum Field theory (QFT), especially for scalar fields.
In some cases, such as the 4 theories, its vanishing implies actually that the theory is
trivial in the sense that the higher correlation functions of the scalar field can be written as
sums of products of two point functions, as in a free theory [1]. On the other hand, a nonvanishing gR is not sufficient to assure the non-triviality of a theory; it only assures that a
certain four point vertex function does not vanish identically, but does not exclude that it
vanishes on shell.
Aside from that, gR is certainly a renormalization group invariant and a characteristic
physical quantity of a field theory. In particular it can be used to check the equivalence
or non-equivalence of different definitions of theories; this will be our main theme in this
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 7 7 - 7
J. Balog et al. / Nuclear Physics B 583 [FS] (2000) 614670
615
paper. gR is proportional to the connected sometimes called truncated four point

function at zero momentum, divided by the square of the zero momentum two point function and appropriate powers of the mass gap to make it dimensionless; details will be given
in the body of the paper.
We are dealing in this article with two main approaches to the construction of a QFT.
The first one starts from a suitably regularized functional integral and then removes the
regularization in a controlled way. This is a rather general procedure usable for a wide
variety of models; it has been successfully employed to construct QFTs in 2 and 3
dimensions obeying all the required axioms (see for instance [2]). Here we will make
use of a euclidean spacetime lattice as a regulator. Removal of the regularization, i.e.,
taking the continuum limit in a lattice theory requires the existence of a second order
phase transition point at which the characteristic length (correlation length) of the model
diverges. This approach raises the problem of universality, i.e., the question whether
different regularizations yield the same QFT after the regulator has been removed.
The other approach studied here is applicable to a large class of so-called integrable
models. It is not based on a Lagrangian, rather the dynamics is specified in terms of
a postulated exact bootstrap S-matrix, supposed to enjoy a factorization property that
allows to express all S-matrix elements in terms of the two-particle S-matrix [3]. In
physical terms this property is linked to the existence of an infinite number of conservation
laws and the absence of particle production. The postulated S-matrices are then used to set
up a system of recursive functional equations for the form factors; solving this system one
can in principle compute exactly all the form factors, in other words continue the S-matrix
off the mass shell [46]. Once the form factors are known, one can express the correlation
functions of the basic fields as well as other (composite) operators by inserting complete
sets of scattering states between them. This gives the correlation functions as hopefully
rapidly converging infinite series of convolution products of form factors. In particular
in this way one can express the intrinsic coupling in terms of the form factors.
In both approaches, in principle one has to verify in the end that the axioms of a QFT
hold. In the lattice approach with a reflection positive action, such as the standard nearest
neighbor action, essentially the only nontrivial question besides the existence of a critical
point concerns the restoration of euclidean (Poincar) invariance in the continuum limit. In
the form factor approach it is less obvious whether the axioms hold, in particular for the
form factor expansion of multi-point correlation functions. There exists, however, a formal
proof (disregarding convergence aspects) of locality [6] and it is hoped, of course, that the
other field theoretic axioms will hold as well, because the construction is to a large extent
inspired by them. It is also not clear from first principles though in practice there are
very natural guesses , which field in one construction should be identified with which
form factor sequence in the other. In any case, even assuming that all the axioms hold in
both constructions and that one has correctly identified the fields, it is a nontrivial question
whether the two approaches define the same theory, and in particular whether they give the
same value for gR .
In the lattice approach the intrinsic coupling of the two-dimensional O(n) models we are
discussing here has been widely studied, both by Monte Carlo simulations [79] and by
616
various expansions in a small parameter [1013]. For a more precise comparison with the
form factor approach, we also carried out our own high precision Monte Carlo simulations
which are reported in this paper.
In the form factor approach the series for gR , being a low energy quantity, is expected to
converge very rapidly. Our results give every indication that these hopes are fully justified,
though the actual computations turn out to be surprisingly intricate. In the present study
we want to develop this computational framework, outline the computations and compare
their results, where possible, to those obtained numerically from the lattice approach by
the different methods mentioned above.
Remarkably in all the examples considered the first non-trivial term in the series, which
contains only one and two particle intermediate states, appears to give about 98% of the
full answer (!). Moreover for this dominant contribution a general model-independent expression in terms of the 1- and 3-particle form factors and the derivative of the S-matrix
can be obtained.
In this paper we discuss three models which can be viewed as the O(n) nonlinear sigmamodels for n = 1, 2, 3. Though formally members of the O(n) series of nonlinear sigmamodels, the physics of these systems, their form factor description, and not the least our
motivation to study them is very different: the n = 1 case is just the massive continuum
limit of the Ising model. Here the spin form factors are very simple and we were able to
push the computation of the series up to all terms with a total particle number (summed
over the three intermediate states) of less or equal 8. The extremely rapid decay of the
terms is manifest and we use the observed pattern as a guideline for the other systems. The
final result amounts to a determination of gR with an estimated precision of better than
0.001%.
The n = 2 case is better known as the XY-model. Here we rely on a bootstrap description
of the model, to which we hope to return in more detail elsewhere [14]. Not all the form
factors are known explicitly, but the specific version of the 3-particle spin form factor
needed for the dominant contribution can be found by elementary techniques. We compare
this leading order result with that obtained by lattice techniques and find reasonable
agreement, which can be taken as support for the proposed bootstrap description.
Finally the n = 3 model is the first with a nonabelian symmetry group. The evaluation
of gR here is in part motivated by the controversy about the absence or presence of a
KosterlitzThouless type phase transition; see [9] for a more thorough discussion.
Let us remark that the form factor bootstrap has also been applied to the computation of
gR in the sinh-Gordon model; in this model the intrinsic coupling is especially interesting
because of its relevance to the issue of triviality versus weak-strong-duality. For details
see the accompanying paper [15].
The article is organized as follows. In the next section we describe the form factor
construction of Greens functions in terms of form factors generally and derive the formula
for the dominant contribution to the coupling. Further we prepare the ground for the
computation of the subleading terms in the specific models. We then give a few generalities
about the Monte Carlo simulations, and move on to discuss the three models as outlined
above one by one in more detail, comparing the results of the form factor construction to
617
those obtained by the the lattice definition of the models; for the latter the values of gR are
estimated by high temperature expansions as well as Monte Carlo simulations.
2. Construction of Green functions in terms of form factors
In this section we will consider a general massive QFT described in terms of its
generalized form factor sequences by which we mean matrix elements of local operators
between physical states. We will restrict our attention to the case of d = 2 dimensions
(although the extension to arbitrary d is often straightforward). The application of the
representation to the integrable models where the form factors are explicitly known will be
the subject of the next chapter.
2.1. Generalities
Our first goal is to construct the euclidean correlation functions (Schwinger functions)
from the generalized form factors. The Schwinger functions are convenient because
they have simpler properties than the Wightman functions and also because it facilitates
the comparison with lattice results later. For points xk R2 , k = 1, . . . , L, we denote
by (xk1 , xk2 ) their components and by xk = (ixk2 , xk1 ) a Wick rotated version. For
definiteness we will consider here correlation functions of n scalar fields a (x), a =
1, . . . , n (the generalization to other types of fields is straightforward). Then

S a1 aL (x1 , . . . , xL ) = a1 (x1 ) aL (xL ) ,
S a1 aL (x1 , . . . , xL ) = W a1 aL (x1 , . . . , xL ),
for x12 > > xL2 .
(2.1)
The first equation is the usual operator interpretation of the Schwinger functions. The second equation (2.1) then indicates the relation of the Schwinger function to the corresponding Wightman function for points (z1 , . . . , zL ) = (x1 , . . . , xL ) in the primitive tube of
analyticity. 1 Outside the primitive tube the Schwinger functions can in principle likewise
be obtained from the Wightman functions by analytic continuation and are then found to
be completely symmetric in all variables. In a form factor expansion however the primitive
domain is preferred in that only there the convergence of the momentum space integrals is
manifest through exponential damping factors (cf. below). We thus mimic the effect of the
analytic continuation by performing the symmetrization by hand
X a a
Ss1 sL (xs1 , . . . , xsL),
S a1 aL (x1 , . . . , xL ) =
sS L
a1 aL
(x1 , . . . , xL ) := (x1 , . . . , xL )W a1 aL (x1 , . . . , xL ),
S
(2.2)
where (x1 , . . . , xL ) is a generalized step function that vanishes unless x12 > >
xL2 holds and the sum is over all elements of the permutation group SL . The functions
a1 aL
(x1 , . . . , xL ) are expected to have a convergent expansion in terms of form factors
S
1 We use the signature (+, ) for (complexified) Minkowski space in which case (z , . . . , z ) is in the primitive
L
1
tube if Im(zk zk+1 ) V + , k = 1, . . . , L 1, where V + is the forward light cone.
618
in the interior of their support region (as well as for certain points on the boundary). The
cases of interest here are L = 2 and L = 4. Formally inserting a resolution of the identity
P
in terms of asymptotic multi-particle states 1 = m |mihm| one obtains
X
a1 a2
(x1 , x2 ) = (x1 , x2 )
e(x1 x2 )2 Em ei(x1 x2 )1 Pm
S
m
h0| (0)|mihm| a2 (0)|0i,

a1
(2.3)
and
a a2 a3 a4
S1
(x1 , x2 , x3 , x4 )
X
e(x1 x2 )2 Ek ei(x1 x2 )1 Pk
= (x1 , x2 , x3 , x4 )
k,l,m
(x2 x3 )2 El i(x2 x3 )1 Pl (x3 x4 )2 Em i(x3 x4 )1 Pm
h0| (0)|kihk| (0)|li hl| (0)|mihm| a4 (0)|0i.

a1
a2
a3
(2.4)
The states |mi are assumed to be improper eigenstates of the momentum operator P , and
Em , Pm denote the eigenvalues of P0 , P1 on |mi, respectively. To write down an explicit
parameterization of the complete set of states |mi requires of course the full knowledge of
the spectrum of stable particles. This is a basic input assumption for the integrable models
dealt with in the next section. Here for simplicity of notation we will consider the case
where there is only one multiplet of stable particle states of mass M. An explicit parameterization will then be given in Section 2.3.
We introduce their (dimensionless) Fourier transforms V by
(2)2 (2) (k1 + + kL )M 2(L1)V a1 aL (k1 , . . . , kL )
Z
a a
= d2 x1 d2 xL S1 L (x1 , . . . , xL )ei(k1 x1 ++kL xL ) ,
(2.5)
taking into account the translation invariance of S . The Fourier transform of the full
Schwinger function is then obtained by symmetrization
e
S a1 aL (k1 , . . . , kL ) = (2)2 (2)(k1 + + kL )M 2(L1)
X
V as1 asL (ks1 , . . . , ksL),
(2.6)
sSL
whereon rotational invariance gets restored. The desired representation of the two and four
point functions in terms of form factors is given by
X
Vma1 a2 (k1 , k2 ),
V a1 a2 (k1 , k2 ) =
m
V a1 a2 a3 a4 (k1 , k2 , k3 , k4 ) =
a a a3 a4
1 2
Vklm
(k1 , k2 , k3 , k4 ),
k,l,m
where
Vma1 a2 (k1 , k2 ) = 2M 2
(Pm + k11)
h0| a1 (0)|mihm| a2 (0)|0i,
Em ik12
(2.7)

a a a3 a4
1 2
Vklm
619
(k1 , k2 , k3 , k4 )
(Pk + k11 ) (Pl + k11 + k21 ) (Pm k41)

Ek ik12 El ik12 ik22 Em + ik42
h0| a1 (0)|kihk| a2 (0)|li hl| a3 (0)|mihm| a4 (0)|0i,
= (2M 2 )3
(2.8)
with the understanding that the sum of the momenta kj vanishes. Further we denote by
a1 a2 a3 a4
(k1 , k2 , k3 , k4 ) the quantities (2.8) with the integrations over
Vma1 a2 (k1 , k2 ) and Vklm
the rapidities performed, the measure being inherited from Eq. (2.25) below.
The key assumption of the form factor approach in this context is that the matrix
elements in (2.8) can be computed exactly via solutions of a recursive system of functional
equations, the so-called form factor equations or Smirnov axioms. Symbolically
a
(| ).
hl| a (0)|mi Fba1 bl a1 am (1 , . . . , l |1 , . . . , m ) =: FBA
(2.9)
The rhs, for which we shall often use the indicated shorthand notation, is called a
generalized form factor, the special case with either l = 0 or m = 0 are the form factors
proper. The form factors are meromorphic functions in the rapidities, while the generalized
form factors are distributions. The form factors can be computed, at least in principle, as
solutions of the before mentioned system of functional equations. The generalized form
factors can then be obtained from them by means of an explicit, though cumbersome,
combinatorial formula. We shall later just state the special cases of this formula required.
A discussion of the general formula can, e.g., be found in the appendix of [16].
Implicit in the products of matrix elements in (2.8) of course are appropriate index
contractions. For definiteness let us note them explicitly
h0| a (0)|mihm| b (0)|0i Imab ( ),
abcd
(| | ),
h0| a (0)|kihk| b (0)|li hl| c (0)|mihm| d (0)|0i Iklm
where
Imab ( ) =
(2.10)
FAa ( ) FAb T ( T ),
abcd
(| | ) =
Iklm
FAa () FAb T B (T | ) FBc T C ( T | ) FCd T ( T ).
(2.11)
A,B,C
Here AT = (ak , . . . , a1 ), T = (k , . . . , 1 ), etc. The construction is such that Imab ( ) is a

abcd
(| | ) is symmetric
completely symmetric function in = (1 , . . . , m ). Similarly Iklm
in each of the sets of variables = (1 , . . . , k ), = (1 , . . . , l ) and = (1 , . . . , m ),
individually.
2.2. The intrinsic coupling
As surveyed in the introduction the intrinsic coupling is defined in terms of the zero
momentum limit of a connected 4-point function. We may assume that the Schwinger
functions of the scalar fields with an odd number of arguments vanish; then the connected
L = 2, 4 Schwinger functions of interest here are
620
e
S a1 a2 (k1 , k2 ),
Sca1 a2 (k1 , k2 ) = e
e
S a1 a2 a3 a4 (k1 , k2 , k3 , k4 ) e
S a1 a2 (k1 , k2 )e
S a3 a4 (k3 , k4 )
Sca1 a2 a3 a4 (k1 , k2 , k3 , k4 ) = e
S a2 a4 (k2 , k4 )
e
S a1 a3 (k1 , k3 )e
S a2 a3 (k2 , k3 ).
e
S a1 a4 (k1 , k4 )e
(2.12)
Making explicit the overall delta-functions arising from translational invariance we

introduce the Green functions by
e
Sca1 aL (k1 , . . . , kL ) = (2)2 (2) (k1 + + kL ) Ga1 aL (k1 , . . . , kL ),
(2.13)
where the constraint k1 + + kL = 0 in the arguments of Ga1 aL will always be

understood.
In the following we will now assume that the theory is O(n) invariant and thus for the
2-point function we can write
Ga1 a2 (k, k) = a1 a2 G(k).
(2.14)
The intrinsic coupling is then defined by

M 2 1 X aabb
G
(0, 0, 0, 0),
G(0)2 n2
gR = N
(2.15)
a,b
where we leave the choice of positive constant N for later.

Performing the symmetrization (2.2) and the Fourier transform one recovers the familiar
expression for G(k) in terms of the spectral density
Z
G(k) =
d ()
0
where
() =
1
2 + k 2
(2.16)

1X
2 P 2 4E (P )
Em
h0| a (0)|mihm| a (0)|0i. (2.17)
m
m
m
n a
In order to compute e
S a1 a2 a3 a4 (k1 , k2 , k3 , k4 ) the symmetrized sum (2.6), (2.7) has to be
performed. For reasons that will become clear immediately we first single out the partial
sum with l = 0. Taking into account the S4 permutations one finds
XX aaaa
1 2 3 4
Vk0m
(ks1, ks2 , ks3 , ks4)
(2)2 (2) (k1 + k2 + k3 + k4 )M 6
k,m sS4
ea1 a2
=S
ea3 a4
(k1 , k2 )S
(k3 , k4 ) + e
S a1 a3 (k1 , k3 )e
S a2 a4 (k2 , k4 )
S a2 a3 (k2 , k3 ) + (2)2 (2)(k1 + k2 + k3 + k4 )M 6

+e
S a1 a4 (k1 , k4 )e
X
as1 as2 as3 as4 (ks1 , ks2, ks3 , ks4).
(2.18)
sS4
Here
a1 a2 a3 a4 (k1 , k2 , k3 , k4 ) =
(k11 + k21)M 6 a1 a2 a3 a4 H (k1 , k2 )G(k4 ),

2
(2.19)
621
with
Z
H (k1 , k2 ) =
d ()
2 +k k
2 + k11
12 22
(2
+ k12 )(2
+ k22 )
(2.20)
2
2 + k11
In obtaining (2.18) we defined the second denominator in (2.8) for l = 0 with the i
prescription as: i(k12 + k22 + i). Here and later the distributional identity
1
1
= P i(x),
x + i
x
will be heavily used, where P is the Principal Value prescription.
One observes that the first three terms in (2.18) are precisely the ones removed by the
definition of the connected 4-point function. Remarkably there is a remainder, the term,
which is present even in the free theory. Typically the spectral densities are decreasing or
bounded by a constant as . The functions G(k4 ) and H (k1 , k2 ) are then regular at
ki = 0. Inserting finally (2.18) into (2.6), (2.7) one obtains for the Green function (2.13)
M 6 Ga1 a2 a3 a4 (k1 , k2 , k3 , k4 )
X X a a a a
s1 s2 s3 s4
Vklm
(ks1 , ks2, ks3 , ks4)
=
k,l6=0,m sS4
as1 as2 as3 as4 (ks1, ks2 , ks3 , ks4).
(2.21)
sS4
On general grounds one expects the vertex function to be real analytic. In particular there
must also be terms involving the delta function in the Vklm above which cancel those of the
-term (we will demonstrate this explicitly in the computation of V121 in Appendix A).
Thus, provided the coupling is well-defined (finite) at all, the result will be independent
of the way the zero momentum limit is taken. It is therefore desirable to find a convenient
limiting procedure that simplifies the computation. To this end we first observe that in (2.8)
the kj 1 and kj 2 components enter asymmetrically. In particular as long as the intermediate
state is not the vacuum (i.e., l 6= 0 in the second formula, and recalling that we are assuming
that none of the operators involved has a non-zero vacuum expectation value) one can put
kj 2 = 0, j = 1, 2, 3, 4. We now compute the 4-point vertex function at zero momentum
through the limiting procedure:

,
(2.22)
Ga1 a2 a3 a4 (0, 0, 0, 0) = lim Ga1 a2 a3 a4 (k1 , k2 , k3 , k4 )
j 0
where
P4
j =1 sh j
|i | 6= |j |,
kj =(M sh j ,0)
= 0, and the limit is taken such that

for i 6= j, and |i j | 6= |k l | for distinct pairs.
(2.23)
In view of (2.19) it is clear that the limit prescription in (2.22), which we will use in the
following, has just been designed such that the term does not have to be considered in
computation of the coupling.
Before embarking on further computations let us comment on a few structural issues. On
physical grounds one expects the intrinsic coupling to be both finite (in a theory with a mass
622
gap) and positive (for N > 0) when the interaction is repulsive. Mathematically however it
is a quite challenging problem to actually prove this, whatever non-perturbative definition
of the theory one adopts. In the context of constructive (lattice) QFT such results seem to
be available only for a single phase 24 theory (see, e.g., [2] for a survey). In the present
context we wish to define the theory strictly terms of its form factors. Mathematically
speaking one should then try to prove in particular that the right-hand side of (2.21) defines
a real analytic function. For the dominant low particle contributions we demonstrate in
Appendix A explicitly that all non-analytic (e.g., distributional) terms indeed cancel out.
We have not attempted to prove this in general, nor can we estimate the rate of convergence
of the sums in (2.21) on general grounds. In all the examples considered later however the
series appears to be rapidly convergent; the terms are alternating in sign and decrease in
magnitude very quickly with increasing particle numbers.
2.3. State parameterization
Here we assume that the single particle spectrum consists only of an O(n) vector
multiplet of mass M. The one particle states |a, i are thus specified by an internal
isospin label a and the rapidity (i.e., the spatial momentum of the state is p = M sh ).
The states are normalized according to
ha, |b, i = 4ab ( ).
(2.24)
The condensed notation for the sum over states now becomes
X
|mihm| |0ih0|
m
Zm1
1
X
X Z d1 Z d2
dm
|a1 , 1 ; . . . ; am , m iin
+
4
4
4
a ,...,a
m=1
ha1 , 1 ; . . . ; am , m |.
in
(2.25)
It is often convenient (for a fixed m) to perform the change of variables

P j
1
je
= ln P ,
uj = j j +1 , j = 1, . . . , m 1,
j
2
je
(2.26)
since in terms of these variables the total energy and momentum of the states take on a
simpler form:
!
m
m
X
X
ch j , M
sh j
(Em , Pm ) M
j =1
j =1

M (m) (u) ch , M (m) (u) sh ,
q
2 P 2 of the mass operator are given by
where the eigenvalues Mm = Em
m
"
#1/2
X
(m)
ch(ui + + uj 1 )
.
M (u) = M m + 2
=
i<j
(2.27)
(2.28)
623
Correspondingly the integration measures in (2.25) above are replaced by

Z
d1
4
Z1
d2
Zm1
dm
Z
0
dm1 u
(4)m1
d
.
4
(2.29)
For later reference we also display the inverse transformation

j = uj + + um1 + um + , j = 1, . . . , m,
P

uj um1
1 + m1
1
j =1 e
where um := ln
.
Pm1 u ++u
m1
2
1 + j =1 e j
(2.30)
2.4. The two point function

The spectral function (2.17) appearing in the representation of the two point function
can be written as a sum of contributions of fixed particle number m
X
(m) (),
(2.31)
() =
0<m odd
where only odd numbers of particles contribute due to our assumption that the fields a
are parity odd. We normalize the fields a by
h0| a (0)|b, i = ba ,
(2.32)
rendering the 1-particle contribution to the spectral density simply

(1) () = ( M).
(2.33)
The m > 3-particle contribution to the spin spectral function (2.17) is given by
Z
(m)
() =

dm1 u
M (m) (u) Im (u),
m1
(4)
(2.34)
with
Im (u) :=
2
1 X X a
Fa1 am (1 , . . . , m ) ,
n a a ,...,a
1
(2.35)
which equals Im11 ( ) under the integral. The function F a featuring here corresponds to the
matrix element of a between vacuum and an m-particle in-state as in (2.9)
Faa1 am (1 , . . . , m ) = h0| a (0)|a1, 1 ; . . . ; am , m iin ,
m < < 1 .
The inverse 2-point function has a low momentum expansion of the form

G(k)1 = ZR1 MR2 + k 2 + O(k 4 ) ,
(2.36)
(2.37)
with
MR2 = M 2
2
,
2
ZR =
22
,
2
(2.38)
624
where 2 , 2 are spectral moments:

Z
2 = M
d
(),
2
Z
2 = M
d
().
4
(2.39)
2.5. The intrinsic coupling revisited

In (2.15) we left open the choice of the normalization constant N because for different
models different choices are convenient. In analytical and numerical lattice computations
(at fixed cutoff) it is often easier to compute the second moment mass MR instead of
the (exponential) spectral mass M (in lattice units). For ease of comparison with these
techniques we thus choose N = MR2 /M 2 , i.e., we define the intrinsic coupling by
gR =
MR2 1 X aabb
G
(0, 0, 0, 0).
G(0)2 n2
(2.40)
a,b
Using O(n) symmetry it follows

Ga1 a2 a3 a4 (0, 0, 0, 0) = M 6 4 a1 a2 a3 a4 + a1 a3 a2 a4 + a1 a4 a2 a3 ,
(2.41)
and hence we can write (2.40) as

gR =
n + 2 4
.
n 2 2
(2.42)
These spectral moments have, corresponding to the decomposition (2.31), an expansion

in contributions arising from states with a fixed (odd) number of particles
X
X
2;m ,
2 = 1 +
2;m .
(2.43)
2 = 1 +
3>m odd
3>m odd
Similarly, corresponding to the sum in (2.21) we have

X
4;klm,
4;klm = 4;mlk ,
4 =
(2.44)
k,l>0,m
where the sum goes over odd integers k, m and positive even integers l. To avoid writing
many O(n) indices we will use
4;klm =
X
1
Limj 0
vklm (s1 , s2 , s3 , s4 ),
3
(2.45)
sS4
where

1111
(k1 , k2 , k3 , k4 )k
vklm (1 , 2 , 3 , 4 ) = Vklm
j =(M
sh j ,0)
(2.46)
and the symbol Lim above means taking the limit j 0 with the j satisfying
P
j sh j = 0 and the constraints in Eq. (2.23).
625
3. The nonlinear O(n) sigma-models

As outlined in the introduction the form factor bootstrap (FFB) construction of an
integrable quantum field theory starts from postulates of the on shell properties of the
theory. By integrable here it is meant that the theory has an infinite set of conserved charges
which entail that there is no particle production. This property usually is characteristic
of non-relativistic quantum mechanics, remarkably here it holds for relativistic quantum
field theories (QFTs) (assuming that the FFB approach does indeed define a QFT). In
4 dimensions absence of particle production implies that the theory is free but in two
dimensions this is not so. In addition to the absence of particle production, one postulates
the spectrum of stable particle states and their 2-particle S-matrix which has to satisfy the
so-called YangBaxter (or factorization) equation (A.5).
In principle one could proceed without reference to a Lagrangian, but often contact
to a Lagrangian description is desirable. Thus typically postulates of specific S-matrices
are motivated by studies of associated Lagrangian QFTs. Unfortunately in most cases one
cannot solve the QFTs to the extent necessary to really derive the candidate S-matrix, rather
one has patches of partial information. This is in particular the case for the O(n) nonlinear
sigma-models formally described by a set of spin fields a , a = 1, . . . , n > 2, with the
constraint 2 = 1 and Lagrangian density ( )2 . There is a wealth of information on
these models which will be recalled when we study the various cases in the following
sections, and for an overview we refer the reader to our previous paper [9]. In particular
the spectrum of stable particles is thought to consist of an O(n) vector multiplet of mass M
without further bound states (i.e., of the form of the spectrum considered in Section 2.3).
The S-matrix element (for n > 2) has the decomposition
(3.1)
Sab;cd ( ) = 1 ( ) ab cd + 2 ( ) ac bd + 3 ( ) ad bc ,
where the center of mass energy is given by s = 2M ch /2.

Classically the theories have an infinite set of local and non-local conserved charges.
One can argue that there are no anomalies which obstruct the existence of such charges
in the quantum theory. In the case of the non-local charges for n > 3 the construction
of Lscher [17] is closely connected to the usual perturbative renormalizability and the
(perturbative) asymptotic freedom of the model. Knowledge of the action of the nonlocal charges on the asymptotic states then restricts the S-matrix to the form postulated
by Zamolodchikov and Zamolodchikov [3] for n > 3
s2 ( )
2i
,
(i ) (n 2) 2i
s2 ( )
,
2 ( ) = (n 2)
(n 2) 2i
s2 ( )
,
3 ( ) = 2i
(n 2) 2i
1 ( ) =
(3.2)
i.e., the invariant amplitudes are all given in terms of one amplitude which we have chosen
here to be the invariant amplitude s2 ( ) in the symmetric traceless (isospin 2) channel.
626
The amplitude s2 ( ) is off-hand determined only up to so-called CDD factors, which were
initially [3] fixed by selecting the solution with the minimal number of poles and zeros in
the physical strip. This solution for s2 ( ) is given by
( Z
)
d
en ()
sin( )K
(3.3)
s2 ( ) = exp 2i
with
+ e 2/(n2)
en () = e
.
K
1 + e
(3.4)
The proposed identification of (3.2)(3.4) with the S-matrix of the O(n) sigma-model
passes several non trivial tests. First, the leading terms of its large n-expansion coincide
with those obtained in leading orders of a field theoretical large n computation. Second, in
the determination of the exact M/ ratio a consistency condition arises when matching the
results of a perturbative computation against that obtained via the thermodynamic Bethe
ansatz [41]. This consistency condition is also sensitive to the CDD factor; the minimal
bootstrap solution (3.2)(3.4) passes the test.
We note that the above formulae have a smooth n 2 limit. A study of the possible
relation of the so defined FFB O(2) model to the continuum limit of the lattice XY-model
(from the massive phase) will be the topic of a future publication [14].
Further we remark that the S-matrix for the case n = 1 (Ising model) can also be written
in the form (3.1) by setting
1 ( ) = 2 ( ) = 0,
3 ( ) = 1,
n = 1.
(3.5)
The representation (3.1) is of course redundant in this case, but it does allow us in
the following to discuss all n > 1 simultaneously. For example in all cases we have an
expansion at low energies of the form
Sab;cd ( ) = ad bc + i Dab;cd + O( 2 ),
(3.6)
in sharp contrast to a weak perturbation of a free field theory.

Having all the on-shell information covers all the physical information on the theory one
observes from scattering of the stable particles, but off shell information is being explored
if the system is probed by external sources weakly coupled to local operators with given
quantum numbers.
3.1. Derivation of the leading term in the FF expansion of gR
For the leading 1-2-1 particle contribution to gR a general model-independent expression
can be given in terms of the derivative the S-matrix and the 3-particle form factor. For
notational reasons we restrict attention here to the O(n) models considered later in more
detail. The extension to a general integrable QFT without bound states is described in
Appendix A. For the O(n) models the formula reads

3
X
dj ( )
4;121 = 4i

d
j =1
=0
1
+
8

64
fc (u)fc (u) 2 .
du
u
ch2 u
1
627
(3.7)
a ( , , ) of the
Here fc ( ) is a particular version of the 3-particle form factor Fbcd
1 2 3
a
local field , supposed to correspond to the renormalized spin field Ra in a Lagrangian
construction. Explicitly
1
(i, , ).
fc ( ) := F1cc
(3.8)
In order to derive (3.7) consider first more generally the (1, l, 1) contribution in (2.21)
with l > 2. Using (2.8) and switching to the explicit notation introduced in Section 2.3 one
can perform the integrals over the rapidities of the 1 particles. Then one decomposes
the rapidity measure for the intermediate l particle contribution according to (2.26).
Using (2.29) the integration can be performed and by means of (2.30) one arrives at
v1l1 (1 , 2 , 3 , 4 )
Z
1
dl (1 , . . . , l , 1 , 2 )
1
I1l1 (1 | |4 )
=
Pl
2 ch2 1 ch2 4 l!
(4)l
j =1 ch j
Z
1
M2
dl1 u
1
I1l1 (1 | |4 ),
=
2
2
l1
2
(l)
(4)
8 ch 1 ch 4
ch M (u)2
(3.9)
1111 ( | | ) is a product of generalized form factors as

where I1l1 (1 | |4 ) := I1l1
1
4
in (2.10), (2.11). Explicitly the correspondence to the matrix elements is
X
h1, 1 | 1 (0)|b1, 1 ; . . . ; bl , l iin
I1l1 (1 | |4 ) =
b1 ,...,bl
in hb1 , 1 ; . . . ; bl , l | 1 (0)|1, 4i.

In the first expression we introduced the notation
!
l
X
sh j ,
(1 , . . . , l ) =
(3.10)
(3.11)
j =1
in the second one is defined by

M
(3.12)
sh = (l) (sh 1 + sh 2 ).
M (u)
As remarked before the generalized form factors can be expressed in terms of form factors
of the same operator and delta distributions by an explicit combinatorial formula. We shall
usually just display the specific version needed. A discussion of the general formula can be
found in the appendix of [16]. For the generalized form factor entering (3.10) the formula
reads

ha1 , 1 | a2 (0)|b1, 1 ; b2, 2 iin
1 >2
= Faa12b1 b2 (1
+ i i, 1, 2 ) + 4a1 b1 a2 b2 (1 + 1 )
+ 4Sb1 b2 ;a2 a1 (1 2 )(1 + 2 ).

Substituting this in Eq. (3.9) we obtain
(3.13)
628
v121 =
( Z
1
2
64 ch 1 ch 4
1
(i
F1xy
d1
1
ch 2 (ch 1 + ch 2 )
1
1 i, 1 , 2 )F1xy
(i 4 i, 1 , 2 )
8
F 1 (i 1 i, 4 , 3 )
ch 3 (ch 3 + ch 4 ) 111
)
8
1
F (i 4 i, 1 , 2 ) .
+
ch 2 (ch 1 + ch 2 ) 111
(3.14)
Here we used the simplifications discussed above and the real analyticity property (3.16e)
below. Moreover, we changed the integration variable from the difference of the two
rapidities to one of the rapidities (1 ). The other rapidity ( 2 ) is then the solution of the
transcendental equation
sh 1 + sh 2 + sh 1 + sh 2 = 0
(3.15)
and is an analytic function of 1 . There are no contributions from terms involving deltafunctions such as (1 + 4 ) appearing in I121 since we are taking the limit (2.22) where
these delta-functions vanish. (These terms are however crucial to cancel corresponding
singularities in the term; cf. Appendix A.)
The form factors appearing in (3.14) obey a system of functional equations which allow
one to further simplify the expression. Let us recall these equations in the form relevant to
d
(, , ) in the O(n) model.
the three-particle form factor Fabc
d
d
(, , ) = Sbc;yx ( )Faxy
(, , ),
Fabc
d
d
(, , ) = Fcab
( + 2i, , ),
Fabc
d
d
Fabc (, , ) = Fabc ( + , + , + ),
d
d
(, , ) = Fcba
( , , ),
Fabc
d
d
(, , )] = Fabc
( , , ).
[Fabc
(3.16a)
(3.16b)
(3.16c)
(3.16d)
(3.16e)
Here the S-matrix appearing in the exchange axiom (3.16a) is the O(n) S-matrix (3.1).
(3.16d) and (3.16e) express the parity invariance and real analyticity property of the
form factors, respectively. The homogeneous axioms (3.16) are supplemented by the
inhomogeneous residue equation

d
(, , ) = 2i ab cd Sbc;ad ( ) .
(3.17)
lim ( i)Fabc
+i
We now take advantage of the analytic properties of the form factors and change the
integration contour in (3.14) from the real axis to a curve C which is arbitrary except that
it has to stay within the physical strip 0 < Im 1 < /2. Along this contour we can put
= 0 and also the limit i 0 can safely be taken. The integrated part of (3.14) then
simplifies to
Z
1
d
(II)
fb ()fb (),
(3.18)
v121 =
128
ch2
C
629
where we introduced the shorthands

d
d
() := Fabc
(i, , ),
fabc
1
() =: bc fb ()
f1bc
(no sum).
(3.19)
Of course, one has to take into account the contribution of those singular points of the
integrand that get crossed when deforming the contour of integration. There are two such
singular points:
1 = 4 + i
and 2 = 1 i,
(3.20)
which never coincide if (2.23) holds.

Applying Cauchys theorem one can evaluate the contribution from the first singular
point using the residue axiom (3.17). This gives
1
16 ch2 1 ch2 4 ch 2 (ch 1 + ch 2 )

1
1
(i 1 i, 1 , 2 ) F111
(i 1 i, 2 , 1 ) ,
F111
where in the second term we also used the exchange axiom (3.16a). After taking the limit
0, which is possible if (2.23) holds, the contribution of the first singular point becomes
1
16 ch 1 ch 4 ch 3 (ch 3 + ch 4 )

1
1
(i 1 , 3 , 4 ) F111
(i 1 , 4 , 3 ) .
F111
2
The contribution of the second singularity is similar:

1
16 ch 1 ch 4 ch 2 (ch 1 + ch 2 )

1
1
(i 4 , 2 , 1 ) F111
(i 4 , 1 , 2 ) .
F111
2
Putting together the contribution of the singular points and the last two terms of (3.14) the
non-integrated contribution can be written as
(I ) .
v121 =
1
8 ch2 1 ch2 4 ch 3 (ch 3 + ch 4 )

1
1
(i 1 , 3 , 4 ) + F111
(i 1 , 4 , 3 ) ,
F111
.
where = indicates equality after the symmetrization over the elements of the permutation
group S4 has been carried out.
We now use the Smirnov axioms (3.16) and (3.17) to simplify the non-integrated part in
the (symmetrized) i 0 limit. It is convenient to first introduce the reduced form factor
d (, , ) by
Gabc
d
d
(, , ) = T3 (, , )Gabc
(, , ).
Fabc
Here and in the following we set

Y
T (i j ),
TN (1 , . . . , N ) :=
16i<j 6N
(3.21)
(3.22)
630
where T is basically the tanh-function T ( ) := tanh /2. Note T ( ) has a simple pole
at = i , T (i ) = T (i + ) = 2/ + O( ), and a simple zero at = 0. The
advantage of the representation (3.21) is that the singularities are carried by the tanh factors
d
is analytic everywhere in the physical strip. In particular,
and the reduced form factor Gabc
for small , and it can be expanded as
d
d
d
(i + , , ) = Jabc
+ ( )Kabc
+ ( )Ldabc + ,
Gabc
(3.23)
where the dots stand for terms higher order in , and . We can compute the constant
tensors appearing in the expansion (3.23) using the form factors equations. From the
residue axiom (3.17) we can immediately fix

d
d
= i ab cd + ac bd ,
+ Ldabc = Dbc;ad .
(3.24)
Kabc
Jabc
To determine the expansion coefficients individually we employ the exchange relation (3.16a) and find
d
= Dbc;ad Dbc;da ,
Kabc
Ldabc = Dbc;da .
(3.25)
Using the expansion (3.23), for small the non-integrated contribution becomes

3 4
(I ) . 1
2i + (3 4 )D + O( 2 ) ,
=
v121
4 (1 + 3 )(1 + 4 )
(3.26)
where

3
X
dj ( )
D = D11;11 = i

d
j =1
(3.27)
=0
This can be simplified by noting that upon averaging over the permutations
1
1
.
.
=
= 0,
1 + 4 1 + 3
(3.28)
and similarly
3
3
4
4
.
.
.
. 1
=
=
=
= .
1 + 4 1 + 3
1 + 4
1 + 3 2
(3.29)
After this simplification we have for the non-integrated contribution

1
(I ) .
(3.30)
v121 = D,
2
and hence the non-integrated part of the leading contribution to the four-point coupling
eventually becomes
(I )
= 4D.
4;121
(3.31)
(I )
For the Ising model the S-matrix is constant and therefore 4;121 = 0 for n = 1. For n > 2
we use (3.3) and find
(I )
4;121
4
= +8
Z
0
en ().
dK
(3.32)
631
The integrated part (3.18) (which is in this form rather useful for numerical evaluation) can
be written in an alternative form using the the residue axiom which implies
4 d
4i
d
( ) = Jabc
+ O(1) = (ab cd + ac bd ) + O(1).
(3.33)
fabc
Using this we can explicitly subtract the singular part in (3.18) and shift the contour back
to the real axis. Noting also that the integrand is an even function of we arrive at
(II)
4;121
1
=
8
Z
0

64
fb (u)fb (u) 2 .
du
u
ch2 u
1
(3.34)
The extension of the formula (3.7) to general integrable models without bound states is
described in Appendix A.
3.2. The three particle form factor
d
( ) in (3.19) is necessary to compute the
Only the special three-particle form factor fabc
leading contribution (3.7) to gR . It turns out to obey an autonomous system of functional
equations (in a single variable) that derives from the form factor equations satisfied by
d (, , ). Solving it allows one to compute f ( ) and hence to evaluate (3.7) in
Fabc
b
situations where the general form factors are not known.
d
d
( ) are real analytic, i.e., [fabc
( )] =
We begin by noting that the functions fabc
d
fabc ( ), in the physical strip 0 6 Im 6 , with simple poles at = 0 and =

i/2. Moreover, using (3.16b,d) one can easily deduce that it is symmetric in its last two
indices,
d
d
( ) = facb
( ).
fabc
(3.35)
Using (3.16a) one obtains

d
d
( ) = Sbc;yx (2 ) faxy
( ),
fabc
(3.36)
and finally combining (3.16ad) results in

d
d
(i ) = Sca;yx ( )Syb;zl (2 ) Slx;vw ( )fwvz
(i + ).
fabc
(3.37)
These are the consequences of the homogeneous form factor axioms; they are supplemented by the residue equations
d
(0) = 4i(abcd + ac bd ),
Res fabc

i
i
d
= i bc ad Sca;bd
.
Res fabc
2
2
(3.38)
(3.39)
In view of Eq. (3.35) we can parameterize f as

d
( ) = k( )ad bc + l( )[ac bd + ab cd ].
fabc
(II)
is given by
Then the contribution 4;121
(3.40)
632
(II)
4;121
=
1
8
Z
du

nk(u)k(u) + 2k(u)l(u) + 2k(u)l(u)
+ 4l(u)l(u)

ch2 u 64/u2 .
(3.41)
In terms of the two functions k( ) and l( ) Eq. (3.36) can be written as

k( ) = s2 (2 ) + n1 (2 ) k( ) + 21 (2 ) l( ),
l( ) = s2 (2 ) l( ),
(3.42)
while (3.37) becomes

k(i ) = A11 ( )k(i + ) + A12 ( )l(i + ) a( )s2 ( )2 s2 (2 ),

l(i ) = A21 ( )k(i + ) + A22 ( )l(i + ) a( )s2 ( )2 s2 (2 ).
(3.43)
Here
a( ) =
(n 2) + 2i
,
(i )(i 2 )[(n 2) i][(n 2) 2i]2
(3.44)
and

A11 ( ) = ( i) 2(n 2)2 3 + (n 2)(n 4) 2 i + (n + 2) 2 2i 3 ,
A12 ( ) = 4(n 2)i ( i)( + i),
A21 ( ) = 2(n 4)i 3 ,
A22 ( ) = A11 ( ).
(3.45)
The matrix A( ) satisfies

A( )1 = A( )a( )a( ),
detA( ) =
1
.
a( )a( )
(3.46)
The functional equations (3.42), (3.43) still contain the transcendental function s2 ( ). It
can be eliminated by the following standard procedure. We introduce the function u( ) as
the unique solution of
u( ) = s2 ( ) u( ),
(3.47)
u(i ) = u(i + ),
(3.48)
subject to the normalization condition

1
+ O( ).
Using the results of Appendix D one can immediately write down the solution
u(i ) =
(3.49)
1
u( ) = T ( ) e(),
2
(3.50)
en () defined in (3.4) has to be used. Introducing

where in (D.3) of course the kernel K
Y ( ) =
2i
u(i ) u(i + ) u(2 ),
u0
1
u0 = u0 (0) = e(0),
4
(3.51)
633
we parameterize k, l as
k( ) = Y ( )K( ) and l( ) = Y ( )L( ).
(3.52)
Rewriting then the functional equations (3.42), (3.43) in terms of K and L, the new system
involves only rational coefficient functions. Explicitly they read
K( ) =
and
1
[(n 2) i](i 2 )

[(n 2) + i](i + 2 )K( ) + 4i L( ) ,
L( ) = L( ),
(3.53)

K(i ) = A11 ( )K(i + ) + A12 ( )L(i + ) a( ),

L(i ) = A21 ( )K(i + ) + A22 ( )L(i + ) a( ),
(3.54)
respectively. The first equation of (3.53) can be used to eliminate L( ) in favor of K( )

via

1
i (n 2) (i 2 )K( )
L( ) =
4i

i + (n 2) (i + 2 )K( ) ,
(3.55)
and (3.55) also solves the second equation of (3.53). Inserting (3.55) into (3.54) results in
a single linear functional equation for K( ). The normalization of the solution is fixed by
the residue equations (3.39).
We expect that this procedure can be used to compute fb ( ) and hence the leading
contribution to the coupling for all O(n) models. For the O(2) model we demonstrate this
in Section 6.
3.3. Subleading contributions
In order to achieve higher accuracy and to obtain some clue on the rate of convergence of
the series (2.44) we will compute some of the subleading terms as well. It turns out that the
1-2-1 term indeed gives the numerically most important contribution to the coupling. But
based on the computation of the subleading terms the numerical result can also be endowed
with an intrinsic error estimate. Our results indicate that the next important contributions
to the coupling are (1, 2, 3) + (3, 2, 1) and (1, 4, 1). Its explicit evaluation is deferred to
Appendices C and B. The difficulty in the evaluation lies in the rapidly varying nature of the
integrands, which have in the multidimensional phase space many zeros and (integrable)
singularities. To deal with these we have either decomposed the integrand into appropriate
parts or avoided the singularities by shifting some contours of integration into the complex
plane.
The (1, 4, 1) contribution is the l = 4 case of Eq. (3.9) and will be evaluated in
Appendix B. Here we prepare the ground for the evaluation of the (1, 2, 3) + (3, 2, 1)
terms. More generally let us examine the (1, 2, m) + (m, 2, 1) contribution and to this end
return to (2.8). Performing the internal rapidity integrations one obtains
634
v12m (1 , 2 , 3 , 4 )
Z
d1 d2 (1 , 2 , 1 , 2 )
2 1
= 2
(4)2 ch 1 + ch 2
ch 1 m!
Z
dm (1 , . . . , m , 4 )
Pm
I12m (1 |2 , 1 | ).
(4)m
j =1 ch j
(3.56)
1111 by inserting the formula expressing the generalized form

Next one spells out I12m := I12m
factors in terms of ordinary form factors. Taking advantage of the S-matrix exchange
relations many of terms contribute equally upon integration and one ends up with four
terms
(I )
(II)
(III)
(IV)
v12m = v12m + v12m + v12m + v12m ,

with
(I )
(1 , 2 , 3 , 4 )
v12m
X X Z
1
(1 , 2 , 1 , 2 )
d2
16(4)m m!
ch 1 + ch 2
a ,...,a
b1 ,b2
(3.57)
dm
(1 , . . . , m , 4 ) 1
F1b1b2 (1 + i , 1 , 2 )
Pk
i=1 ch i
Fb12 b1 a1 a2 am (2 + i , 1 + i , 1 , . . . , m )Fa11 a2 am (1 , . . . , m ) , (3.58)

X X Z
1
(1 , 2 , 1 , 2 )
(II)
(1 , 2 , 3 , 4 )
d2
v12m
m1
ch 1 + ch 2
8(4)
(m 1)!
b1 ,b2 a2 ,...,am
Z
(1 , 2 , . . . , m , 4 ) 1
P
dm1
F1b1 b2 (1 + i , 1 , 2 )
ch 1 + m
i=2 ch i
(3.59)
Fb12 a2 am (2 + i , 2 , . . . , m )Fb11 a2 am (1 , 2 , . . . , m ) ,
Z
X
X
1
(1 , 2 , 1 , 2 )
(III)
d2
v12m (1 , 2 , 3 , 4 )
16(4)m2 (m 2)!
ch 1 + ch 2
b1 ,b2 a3 ,...,am
Z
(3 , . . . , m , 3 )
1
P
dm2
F1b
(1 + i , 1 , 2 )
1 b2
ch 1 + ch 2 + m
i=3 ch i
Fa13 am (3 , . . . , m )Fb11 b2 a3 am (1 , 2 , 3 , . . . , m ) ,
X Z
1
(1 , . . . , m , 4 )
(IV)
Pm
(1 , 2 , 3 , 4 )
v12m
dm
m1
16(4)
m! a ,...,a
i=1 ch i
1
(3.60)
+ i , 1 + i , 1 , . . . , m )Fa11 a2 am (1 , . . . , m ) ,
(3.61)
where stands for . All integrals range from to +.
Further details of the computation of the 1-2-3 contribution are given in Appendix C.
Note that for the numerical evaluation of the k + l + m = 6 contributions we need the
analytic expressions for the 5-particle form factor of the spin operator. Unfortunately these
are at present only known for n = 1 and n = 3. For the Ising model all the form factors are
explicitly known and in this case we have also computed the k + l + m = 8 contributions.
After a preparatory next section where we discuss the definition and measurement of the
intrinsic coupling in the lattice regularization, we will discuss the cases n = 1, 2, 3 in turn.
1
F11a
(2
1 a2 am
635
4. Lattice computations of g R
In the subsequent sections we will compare the results of the form factor bootstrap
coupling gR with those obtained from the lattice theory. As noted earlier, in the framework
of the lattice regularization there are two methods to compute gR in the O(n) models: high
temperature (= strong coupling) expansions and Monte Carlo simulations. Both approaches
usually take the standard lattice action on a square lattice
X
(x) (x + ),
(4.1)
S =
x,
P
as the starting point, where (x) (x) = a a (x) a (x) = 1.
The lattice definition of gR () is as in Eq. (2.40)
gR () =
1
22 G2 (0)2
1 X aabb
G4 ,
n2
(4.2)
a,b
where all quantities are defined analogously to the continuum theory

1 X X ikx
a
e (x) a (0) ,
n a x
X

a a a a
a1 (x1 ) a2 (x2 ) a3 (x3 ) a4 (0)
G4 1 2 3 4 =
G2 (k) =
x1 ,x2 ,x3

a
1

(x1 ) a2 (x2 ) a3 (x3 ) a4 (0) + 2 perms ,
(4.3)
(4.4)
and 2 is the second moment correlation length

22 =
2
,
4G2 (0)
2 =

1 X X 2
a
x (x) a (0) .
n a x
(4.5)
The coupling from the lattice regularization is defined as the continuum limit
gR = lim gR (),
c
(4.6)
where c is a critical point where the correlation length diverges (in lattice units).
Butera and Comi [12] have produced long high temperature series for G2 (0), 2 ,
and G4 in the O(n) model with standard action, and Pelissetto and Vicari [13] have
reanalyzed these series to compute estimates for the intrinsic coupling gR for n 6 4. Similar
computations have been performed previously by Campostrini et al. [11].
Our Monte Carlo simulations were of course done on a finite lattice, more precisely a
square lattice of size L (points) in each direction and periodic boundary conditions, both
with the standard action (4.1) and the fixed point action of Ref. [18]. The infinite volume
lattice coupling gR () is then obtained as the limit
gR () = lim gR (, L),
L
of a finite volume coupling gR (, L) which is proportional to Binders cumulant uL :
(4.7)
636
L
gR (, L) =
eff (, L)
uL = 1 +
2
uL ,
2 h( 2 )2 i
,
n
h 2 i2
(4.8)
P
where a = x a (x). In this definition eff (, L) is an effective correlation length
which converges to the second moment correlation length 2 in the limit L . In our
computations we used the particular definition (as, e.g., in Ref. [7]):
s
1
G2 (0)
1,
(4.9)
eff (, L) =
2 sin(/L) G2 (k0 )
where k0 = (2/L, 0).
In our analysis of the Monte Carlo data we shall make the working assumption that one
is allowed to replace the limiting procedure limc limL by
gR = lim gR (z),
z
gR (z) :=
lim
c , z fixed
z := L/ eff (, L),
gR (, L).
(4.10)
That is we attempt to first take the continuum limit at fixed physical volume and afterwards
take the physical volume to infinity. The z limit of g R (z) is expected to be reached
exponentially; for example in the leading order 1/n expansion [10]
(4.11)
gR (z) = g R () 1 c z exp(z) + .
The situation may however be slightly more complicated due to our particular definition of
eff . Indeed in the continuum limit at fixed physical volume we expect G2 (0)/G2(k0 )
G(0)/G(k) where k K0 = (2MR /z, 0) and the continuum expressions are in finite
physical volume. On the other hand for the continuum two point function defined in infinite
volume

2

1
2
1
(2 1) .
(4.12)
G(0)/G(K0 ) 1 2 1
z
K02
MR
In our simulations the values of 2/z are 1, i.e., not so small; nevertheless at such
values the correction factor on the rhs of (4.12) only deviates from 1 by the order 103 .
This deviation is much smaller than the statistical accuracy of our simulations, and hence
we ignore these additional effects in our analyses of the lattice data.
5. The Ising model

The particular field theory we are considering in this section is that obtained from
the Ising model in zero external field 2 for 0 < T Tc 0. The spinspin correlation
2 One can obtain an infinite number of field theories from the Ising model in the presence of an external field
H by taking the limit H 0, T Tc with h = H /|T Tc |15/8 fixed.
637
functions in the scaling limit are known exactly from the work of Wu et al. [19], and from
this knowledge Sato, Miwa and Jimbo [20] found that the S-matrix operator was given by
S = (1)N(N1)/2,
(5.1)
where N is the particle number operator. An energy independent phase is not observable
in a scattering experiment; the non-trivial S-matrix (5.1) reflects the fact that the off-shell
spinspin correlation functions are not that of a free field. The continuum limit of the Ising
model is also described by a free Majorana field, but this is non-local with respect to the
spin field; for a more detailed discussion we refer the reader to the lectures of McCoy [21].
5.1. Form factor determination
The generalized form factors are given by [22]
out
h1 , . . . , m | (0)|m+1, . . . , N iin
Y

= (2i)(N1)/2
T |i j |
16i<j 6m

T |k l | ,
Y
16r6m<s6N
P
T (r s )
(5.2)
m<k<l6N
with N an odd (positive) integer. We evaluate the dominant contribution to the coupling
using Eq. (3.7). The non-integrated part (3.31) vanishes. For the integral (3.34) we need
f1 ( ), which is readily obtained from (5.2),
f1 ( ) = 2i T (2 )/T 2 ( ).
(5.3)
Thus the dominant contribution to 4 is

1
4;121 =
2

16
5 47
.
2 =
du
2
4
u
2
6
T (u) ch u
T 2 (2u)
(5.4)
Numerically this gives 4;121 = 4.993427441(1) or gR 14.98 in the leading approximation.

The simplicity of the form factors (5.2) also makes the Ising model a good testing ground
for the computation of the subleading contributions, to which we turn now. The evaluation
of the spectral moments (2.39), (2.43) is straightforward. For m = 3, 5, 7 the results are
given in Table 1.
Table 1 suggests that the series (2.43) converge extremely rapidly and we would estimate
2 = 1 + 8.15259(1) 104 ,
2 = 1 + 1.094(1) 105 ,
(5.5)
where the estimated errors come both from the numerical integration and from estimating
the contributions of the higher particle terms. To get some check on this we may consider
the ratio 2 /2 for which from the leading terms (5.5) we get 2 /2 = 0.999 196 336(11).
This is in excellent agreement with the result 2 /2 = 0.999 196 33 of Campostrini et
638
Table 1
m-particle contributions to 2 , 2 in the Ising model
m
2;m
3
5
7
8.1446256566(1) 104
7.96(1) 107
7.8(1) 1010
2;m
1.094(1) 105
2.22(1) 1010
4.6(1) 1015
Table 2
k-l-m-particles contributions to 4 in the Ising model
k, l, m
4;klm
1, 2, 1
4.993427441(1)
1, 2, 3
1, 4, 1
0.046310(1)
0.002653(1)
3, 2, 3
1, 4, 3
1, 2, 5
1, 6, 1
0.0002884(3)
0.0000420(5)
0.00002562(2)
0.0000040(1)
al. [23], which they obtained by numerical evaluation of the exact formula for the 2-point
function 3 of Wu et al. [19].
The evaluation of 4 is more involved. In order to gain insight into the rate of decay of
the higher particle contributions as well as their sign pattern we pushed the computation up
to k + l + m 6 8. The k + l + m = 8 contributions in particular turned out to be a formidable
computation despite the deceptive simplicity of the form factors. The computation is based
on the formulae (3.9), (3.56) and similar ones for (m, 2, m), with m odd, and for (1, l, 3),
with l even. To give the reader a chance to follow the computations we have collected some
intermediate results in Appendices B, C. The final results for the contributions of the k-l-m
intermediate states with k + l + m 6 8 to 4 are summarized in Table 2.
The rapid decay of the terms is manifest. Increasing k + l + m by 2 gives a contribution
roughly two orders of magnitude smaller than the previous one. The sign pattern appears to
follow the rule: Sign(4;klm) = Sign(k + m l 1). Further terms with larger differences
|k l|, |l m| are suppressed as compared to those with smaller ones. In view of Table 2
we would thus (conservatively) estimate the k + l + m > 10 particle contributions to be
6 10% of the sum of the k + l + m = 8 contributions. This gives
4 = 4.90321(3).
(5.6)
Inserting into (2.42) with 2 , 2 taken from (5.5) then yields our final result
gR = 14.6975(1).
(5.7)
3 This famous Fredholm determinant (solving the Painlev III equation) is basically the summed up FF series;
see, e.g., [24].
639
Table 3
Previous determinations of gR in the Ising model
Method
High temperature
Borel summation
Monte Carlo
Value for gR
14.6943(17) [13], 14.67(5) [25]
15.5(8) [26]
14.3(1.0) [27]
Fig. 1. The data for gR at z 7.4 for different lattice spacings.
This amounts to a determination of gR to within < 0.001%. For comparison we collected

the results of some previous determinations in Table 3.
Finally we would like to mention that an analogous 4-point coupling hR can be defined
at criticality T = Tc by sending the magnetic field H to zero. Of course in this case the
definition of Binders cumulant has to be modified appropriately to take into account the
fact that the field has non-vanishing vacuum expectation value. Remarkably hR can be
computed exactly by taking advantage of the fact that the small H behavior of the partition
function is known exactly [28,29]. The final result is hR = 609/4 = 478.307.
5.2. Recent Monte Carlo simulation of the Ising model
Our Monte Carlo investigation of gR was performed on several IBM RISC 6000
workstations at the Werner-Heisenberg-Institut.
In this subsection eff is denoted simply by . We studied the dependence on the lattice
spacing by running at = 0.418 ( = 10.839936), = 0.4276 ( = 18.924790) and =
0.433345 ( = 33.873923) on lattices of size L = 80, L = 140 and L = 250, respectively.
These values were chosen in such a way that they have almost exactly the same value of
z = L/ 7.4. Fig. 1 shows that there is no significant dependence on the lattice spacing
(i.e., ). Therefore we decided to use all the data together to study the finite size effects.
640
Fig. 2. The data for gR vs
z exp(z) together with the fit.
Table 4
Ising data for = G2 (0) and gR (, L)
0.418
0.418
0.418
0.418
0.4276
0.4276
0.433345
# of runs
eff
40
60
80
140
140
250
250
200
200
321
300
100
225
202
10.839936
10.839936
10.839936
10.839936
18.924790
18.924790
33.873923
163.54(13)
172.81(11)
173.94(6)
174.08(4)
455.34(35)
455.90(12)
1254.48(72)
gR
11.941(11)
14.104(26)
14.587(30)
14.743(39)
14.610(54)
14.796(46)
14.567(37)
We studied the finite size dependence by measuring in addition gR on lattices of size

L = 40, 60, 80, 140 at = 0.418, ( = 10.839936). Finite size scaling works very well,
i.e., the results only depend on z = L/ . The dependence on z is still quite well described
by Eq. (4.11). This can be seen in Fig. 2. A least square fit produces
c = 3.91(3),
gR () = 14.69(2).
The fit quality is not fantastic ( 2
(5.8)
= 2.4 per d.o.f.) but acceptable. So our final Monte Carlo
estimate for gR is
gR = 14.69(2).
(5.9)
We report our numbers in Table 4. In this table we also indicate the number of
measurements. These were performed using the cluster algorithm as follows: one run
consisted of 100,000 clusters used for thermalization, followed by 20,000 sweeps of
641
the lattice used for measurements. Each run was repeated after changing the initial
configuration. One such run was considered as one independent measurement. The error
was computed out of this sample by using the jack-knife method.
Our estimated value for gR in Eq. (5.9) is in very good agreement with the values from
the analysis of the high temperature expansion given in Table 3; it is also consistent with
the value Eq. (5.7) obtained from the form factor construction.
6. The XY-model
In this section we compute the leading contribution to the four-point coupling in
the two-dimensional O(2) nonlinear -model better known as the XY-model. Starting
from the lattice formulation, after a chain of mappings consisting of several steps the
model is transformed to a system equivalent to the two-dimensional Coulomb gas. The
continuum limit of the Coulomb gas model (corresponding to the KosterlitzThouless
critical point [30]) is thought to have a dual description in terms of a sine-Gordon model at
the (extremal) sine-Gordon coupling 2 = 8 . For a review of the XY-model, see [31].
In the following we will start by discussing the XY-model S-matrix. The next step is to
solve the Smirnov equations for the three-particle form factors, which enter the formula
for the leading term. A general method for finding the sine-Gordon form factors is given
in [32]. This extends the results of Smirnov [6], where the form factors for an even number
of particles were found. The spin three-particle form factor we are interested in is probably
similar to the three-particle form factor of the fermion operator (corresponding to the
equivalent massive Thirring-model description), explicitly given in [32]. Here however
we need the three-particle form factor only for special rapidities and we found it simpler
to obtain this special version by going back to the functional equations. It is then used to
numerically evaluate the leading contribution to gR .
6.1. The XY-model S-matrix
We will regard the XY-model as the n = 2 member of the family of O(n) -models.
Recall that the formulae (3.2), (3.4) have a smooth n 2 limit; this has been noted and
commented on previously by Woo [33]. In this limit
1 ( ) =
and
s2 ( ),
(i )
(
s2 ( ) = exp 2i
2 ( ) = 0,
)
d
e2 () ,
sin( ) K
3 ( ) = s2 ( )
/2
e2 () = e
.
K
2 ch /2
(6.1)
(6.2)
In this paper we will assume that the spectrum of the XY-model in the (massive) continuum limit consists of an O(2) doublet of massive particles whose S-matrix is given by (6.1)
with (6.2). Of course, taking the formal n 2 limit of the bootstrap results valid for n > 3
would not be convincing in itself, but (6.1), (6.2) actually coincide with the 2 8 limit
642
of the sine-Gordon S-matrix, the prediction of the KosterlitzThouless theory! The S-matrix (6.1) and the corresponding scattering states as a consequence have a Uq=1 (su(2))
Hopf algebra symmetry, which as a Lie algebra is isomorphic to su(2). The latter is an
explicit symmetry in the alternative chiral GrossNeveu formulation of the model [31].
6.2. The three particle form factor
Next we calculate the three-particle form factor at the special rapidities necessary to
compute the leading contribution (3.34). For this purpose we note that the equations for
the functions k, l given in Section 3.2 can relatively easily be solved in this particular case
n = 2. We first note that Eq. (3.54) simplifies
K(i ) = K(i + ),
1
(i )(i 2 )L(i )
K(i + ) =
2i

(i + )(i + 2 )L(i + ) .
(6.3)
Inserting (3.55) yields

K(i ) =
3i 2 i 2
K(i + ).
3i + 2 i + 2
(6.4)
Luckily a term proportional to K(i ) drops out here and one is left with the simple
form (6.4). This can easily be converted into the form (D.1) and solved as
K(i ) = (2 5i)(2 7i) eD(/2) (i ch(/2)).
(6.5)
D( ) = 1/4 ( ) + 3/4 ( )
(6.6)
Here
in the notation of Appendix D and (z) is a polynomial function to be determined later.

Since Y ( ) already has the right singularity structure the functions K( ) and L( ) are
analytic in the physical strip. The residue axioms determine their value at = 0 and =
i/2 as
K(0) = 0,

u0
i
,
= 2
K
2
u (i/2)
L(0) = 1,

i
L
= s2 (i/2)K(i/2).
2
(6.7)
So far we have established that the solution can be expressed in terms of

Y ( ) = 2i
ch3 (/2) 2(i+)+(2)(0)

e
sh(/2) ch
(6.8)
and
K( ) = (2 + 3i)(2 + 5i) eD(
i
2 )
(sh(/2)).
(6.9)
The polynomial (z) can be determined using the residue constraints (6.7), which we can
rewrite as
K(0) = 0,
K(i/2) = e(0)2(i/2) ,
2
,
i
K(i/2) = e(0)(i/2)(i/2) .
643
K 0 (0) =
(6.10)
Using (6.8) and (D.4) one sees that for real

|Y ( )| e 3/4 .
(6.11)
This can be used to infer that the polynomial (z) can be at most second order, otherwise
the integral contribution to the leading term would diverge. Taking into account that
K(0) = 0 and the requirement of real analyticity one must have
(z) = i1 z + 2 z2 ,
(6.12)
for real constants 1 and 2 . Now it is easy to see that (6.7) determines 1 as
4
eD(i/2) .
15 3
In order to determine 2 we employ the following identities [34]
)
(Z
1
ch() 1
e
,
d
=
p() := exp
sh()
2 sin( /2)
1 =
(6.13)
(6.14)
)
(Z
ch() 1 2
1 2
e
,
d
=
q() := exp
sh()
cos( /2)
(6.15)
to obtain

16 2
q 2 (1)
=
.
exp (0) 2(i/2) + D(i/2) D(i/4) = p(1)
q(3/2)
5
(6.16)
This can be used to show that (6.10) is satisfied for the choice 2 = 0. Thus (sh 2 ) =
i1 sh 2 , and since 4 L( ) = (i 2 )K( ) (i + 2 )K( ), both k( ) = Y ( )K( )
and l( ) = Y ( )L( ) are known explicitly for the XY-model.
6.3. Calculation of the leading contribution
Having all the ingredients at our disposal we can compute the leading term (3.7) of the
intrinsic coupling. Firstly from (3.32) we have for n = 2
4
(ln 4 1).
(6.17)
Further substituting the explicit results for the functions k, l obtained above into Eq. (3.41)
and evaluating the resulting expression numerically we obtain
(I )
4;121 =
(II)
= 5.14902(1),
4;121
(6.18)
and hence
(I )
(II)
+ 4;121
= 4.65718.
4;121 = 4;121
(6.19)
644
Thus the leading contribution to the XY-model four-point coupling is

4
24;121 = 9.314.
(6.20)
gR = 2
2 2
Since 2 2 > 1 and since the next leading contributions to 4 are probably positive (as
they are in the Ising and O(3) models), we expect that the true value of gR will be less than
that given in (6.20) (probably by 24%).
6.4. Comparison with lattice results
For the XY-model with standard action Kim [8] gives the value
gR = 8.89(20)
(6.21)
for = 1/0.98. We are in the process of producing higher precision Monte Carlo data
for this model; so far we can only give a preliminary result, obtained on a lattice of size
L = 500 at = 1.0174:
gR = 9.14(12).
(6.22)
We will return to this issue in a separate publication, where we intend to analyze the finite
size corrections as well as the lattice artifacts.
We also wish to mention the results from the high temperature expansion: Butera and
Comi [12] obtain
gR = 9.15(10),
(6.23)
whereas Pelissetto and Vicari [13] give

gR = 9.10(5).
(6.24)
So there is an overall rough agreement between the lattice and the form factor results, but
the precision is not comparable to that obtained for the Ising model.
7. The O(3) nonlinear sigma-model

The O(3) nonlinear sigma-model is an important testing ground for quantum field
theoretical scenarios in nonabelian gauge theories. The form factor technique has been
particularly fruitful in studying its possible off-shell dynamics and can be confronted with
what can be achieved by perturbation theory or numerical simulations [35]. The intrinsic
coupling has been computed before by a number of different techniques; we compare the
results with ours at the end of this section. The present form factor determination takes
as usual the Zamolodchikov two-particle S-matrix [3] as its starting point; it is given by
Eqs. (3.1), (3.2) with n = 3 and
i
.
(7.1)
s2 ( ) =
+ i
The corresponding kernel (3.4) is simply given by
e3 () = e .
K
(7.2)
645
7.1. Form factor determination of gR

Following the by now routine procedure we first collect the ingredients for the evaluation
of the dominant (1, 2, 1) contribution to the intrinsic coupling. From (3.32) one readily
finds for n = 3
4
(I )
(7.3)
= .
4;121
The O(3) form factors have been computed in [6,35,36]. In particular the reduced 3-particle
form factor G in Eq. (3.21) is given by

Gaa1 a2 a3 (1 , 2 , 3 ) = 3 (1 , 2 , 3 ) aa1 a2 a3 (3 2 ) + aa2 a1 a3 (1 3 2i)

(7.4)
+ aa3 a1 a2 (2 1 ) ,
where
N (1 , . . . , N ) =
(i j ),
16i<j 6N
( i)
tanh .
(7.5)
(2i )
2
Correspondingly the functions k, l parametrizing fb via Eq. (3.40) are for n = 3 explicitly
given by
( ) =
3 T 2 (2 ) (2 i)
.
4T 4 ( )( 2 + 2 )2
Plugging this into the general formula Eq. (3.41) yields

Z 6 2 2
64
1
u (4u + 2 )(2u2 + 2 ) T 4 (2u)
(II)
du
.
4;121 =
8
4(u2 + 2 )5
T 8 (u) ch2 u u2
k( ) =
2
l( ),
i
l( ) =
(7.6)
(7.7)
Numerically we then obtain 4;121 = 4.16835492(1), so that as a first approximation

gR 53 4;121 = 6.9472. This is already in rough agreement with other determinations in
the continuum theory: the 1/n, the - and the g-expansions [7,13]. The leading order 1/n
computations have been performed in [37]. For the spectral integrals the result is

1
1
1
1
2 = 1 + 0.00026836 + O 2 . (7.8)
2 = 1 + 0.00671941 + O 2 ,
n
n
n
n
and for the coupling [10]

1
8
1
1 0.602033 + O 2 .
gR =
n
n
n
(7.9)
which gives the approximation gR 6.70 for the case n = 3. The results from the other
methods are given in Table 7. Considering the rather short series in each case it is amazing
how well the estimates by the various methods agree.
For a more precise determination we now return to the form factor approach and examine
the subleading contributions. Using the exact form factors [35] the results for the 3- and
5-particle contributions to 2 and 2 are readily evaluated and are listed in Table 5.
646
Table 5
m-particle contribution to 2 , 2 in the O(3) model
m
2;m
2;m
1.67995(1) 103
3.46494(1) 105
6.622(1) 106
7.114(1) 109
Table 6
k-l-m-particles contribution to 4 in the O(3) model
k, l, m
4;klm
4.16835492(1)
0.051748(1)
0.004065(1)
1, 2, 1
1, 2, 3
1, 4, 1
The size of the higher particle contributions to 2 and 2 can roughly be estimated by an
off hand extrapolation of Table 5; essentially they are negligible to the desired accuracy.
The latter could also be justified by referring to a more refined extrapolation scheme, based
on the scaling hypothesis of Ref. [35]. In upshot we obtain
2 = 1.001687(1),
2 = 1.000034657(1).
(7.10)
The computation of the subleading terms to 4 is much more involved. The starting point
is again the formulae (3.10) in Section 3.1. Due to the complexity of the form factors
however the computation is feasible only computer aided. The essential steps are given in
Appendices B, C. The computation has been performed independently by subsets of the
authors using slightly different techniques. The final results for the contributions of the
k-l-m intermediate states with k + l + m 6 6 to 4 are listed in Table 6.
The leading 1-2-1 contribution is a factor 42 greater in magnitude than the sum of
k-l-m contributions with k + l + m = 6. It is difficult to bound the rest of the contributions,
especially since the signs appear to be alternating. The computation of the states with
l + m + n = 8 would be quite an undertaking. But assuming that the pattern in Table 6
continues, as it seems to be the case in the Ising model (see Table 2), then we consider the
assumption that the sum of the remaining contributions k + l + m > 8 is 6 10% of the sum
of the k + l + m = 6 contributions to be reasonable and we then obtain
4 = 4.069(10),
(7.11)
and hence our final result

gR = 6.770(17).
(7.12)
This amounts to a determination of gR to within 0.3%. For comparison we give some

results of other already published determinations in Table 7. The first two are continuum
methods while the last one is based on the lattice regularization. We describe the two lattice
647
Table 7
Other determinations of gR in the O(3) model
Method
g-expansion
-expansion
High temperature
Value for gR
6.66(6) [7]
6.66(11) [13]
6.56(4) [10], 6.6(1) [11]
techniques in somewhat more detail in the next subsection, including in particular our own
recent Monte Carlo results.
7.2. Lattice computations of gR
High temperature expansion:
The analyses of the high temperature expansion for the spectral moments give 2 =
1.0013(2) [37] and 2 = 1.000029(5) [38]. The agreement with the FFB values Eq. (7.10)
is acceptable; note that these are smaller than that anticipated from the leading order of the
1/n approximation, Eqs. (7.8).
The various Pad approximations show the coupling falling rapidly as increases in the
region of small , then a region of rather flat behavior after which these approximations
show diverse behavior; some analyses indicate that in fact there is a shallow minimum
and that the continuum limit is actually approached from below (see, e.g., Refs. [13,23]).
In Ref. [11] Campostrini et al. quote for the case n = 3 the result gR = 6.6(1), and in a
more recent publication Pelissetto and Vicari cite 6.56(4) [13]. Butera and Comi on the
other hand are rather cautious, and did not quote a value for the case n = 3 in Ref. [12]; if
pressed they would at present cite gR = 6.6(2) [39].
Numerical simulations:
Monte Carlo computations of gR have a long history, see, e.g., Refs. [7,8]. In order
to attempt to match the apparent precision attained in the FFB approach, we recently
performed new high-precision measurements. These were performed on several IBM RISC
6000 workstations at the Werner-Heisenberg-Institut. In addition we made use of the
SGI 2000 machine of the University of Arizona, especially for the very time consuming
simulations on large lattices.
Based on the fixed point action [18] we have measured gR at three different values of
: 0.70, 0.85 and 1.00, corresponding to correlation length 3.2, 6.0 and 12.2, at the
values of z = L/ in the range 5.48.2. The data and their analysis can be found in [9], the
final result is gRFP = 6.77(2).
Monte Carlo measurements with the standard action were performed using a method
similar to the cluster estimator of [40]. We have reported the analysis of such simulations
already in our earlier paper [9]. But in the meantime we produced more data and we take
the opportunity to report them here.
The present status of the results of our simulations are given in Table 8. In this
table we also indicate the number of measurements. These were performed using the
648
Table 8
O(3) data for , = 3G2 (0) and gR
1.5
1.6
1.7
1.8
1.9
1.95
# of runs
gR (, L)
gR (, )
80
140
250
500
910
1230
344
370
367
382
127
68
11.030(7)
18.950(14)
34.500(15)
64.790(26)
122.330(74)
167.71(17)
175.95(11)
447.13(34)
1267.20(57)
3838.76(1.50)
11883.0(6.4)
20901.4(19.0)
6.553(16)
6.612(15)
6.665(14)
6.691(15)
6.737(21)
6.792(40)
6.616(16)
6.668(15)
6.730(14)
6.733(15)
6.792(21)
6.853(40)
Fig. 3. The extrapolated values of gR (, ).
cluster algorithm as follows: one run consisted of 100000 clusters used for thermalization,
followed by 20000 sweeps of the lattice used for measurements. Each run was repeated
after changing the initial configuration. One such run was considered as one independent
measurement. The error was computed out of this sample by using the jack-knife method.
Our measurements were taken at 6 different correlation lengths ranging from about 11
to about 168 on lattices satisfying L/ 7. To study the finite volume effects, we took in
addition data at 11 for lattices of sizes L with L/ 5.5, 9 and 13. As discussed
in [9], the finite size effects are well described by the formula
(4.11), even at finite (large)
correlation lengths. In the O(3) model the n = value c = 8 fits very well.
But unlike the Ising model, the lattice artifacts are by no means negligible. To study
them, we first use Eq. (4.11) to extrapolate our data to z = . In this extrapolation we use
the effective correlation length eff and neglect the fact that this is not exactly equal to the
exponential correlation length. In Fig. 3 we plot those extrapolated values of gR against
1/ which we identify with 1/ eff .
649
Unfortunately there is no rigorous result concerning the nature of the approach to

the continuum limit. At the time of our last analysis [9] the data point at the largest
value of 168 was not available. In that paper we fitted the data in the entire
range from 11 to 122 using a Symanzik type ansatz of the form gR ( ) =

gR () 1 + b1 2 log + b2 2 , and thereby obtained the result gR = 6.77(2). When
we now repeat the same fit for the new data, which in particular includes the new point
at 168, the result is only slightly changed to gR = 6.78(2) but the quality of the fit
becomes poorer. The fact that the two data points closest to the continuum limit lie above
6.78 is in this scenario interpreted as a statistical fluctuation.
On the other hand the present rather large central value at 168 could be interpreted
as an indication that the continuum limit is approached much slower than conventionally
assumed, perhaps as slow as 1/ ln (which may be expected in the O(2) model [42])!
If we adopt this viewpoint it is clear that, although qualitative fits can be made, without
further analytic information, our data are not sufficient to make a reliable quantitative
extrapolation to the continuum limit. However, independent of the assumed form of the
approach to the continuum limit, if the large value at 168 is confirmed by more
extensive studies it would practically establish a discrepancy between the form factor and
the lattice constructions of the O(3) sigma-model. This point, which needs complete control
over all systematic effects, albeit extremely difficult on such large lattices, is certainly
worthy of further investigations.
8. Conclusions
A new technique to compute the intrinsic 4-point coupling in a large class of twodimensional QFTs has been developed and tested. Starting from the form factor resolution
of the 4-point function the termwise zero momentum limit turned out to exist, providing a
decomposition of the coupling into terms with a definite number (k, l, m) of intermediate
particles. Based on the exactly known form factors these terms can be computed practically
exactly and in the models mainly considered (Ising and O(3)) were found to be rapidly decaying with increasing particle numbers. There is every reason to expect that this trend
continues, which allowed us to equip the results with an intrinsic error estimate. The final
results are
Ising model:
gR = 14.6975(1),
O(3) model:
gR = 6.770(17).
(8.1)
They amount to a determination of gR to within < 0.001% and 0.3%, respectively.

In addition we obtained the universal, model-independent formula (A.3) for the
dominant contribution to the coupling, which typically seems to account for about 98% of
the full answer. We illustrated its use in testing our proposed bootstrap description of the
XY-model. It would surely also be interesting to apply it, e.g., to supersymmetric theories,
where alternative techniques are hardly available.
650
The comparison with the lattice determinations of gR is quite impressive in the case of
the Ising model, where there is also very good agreement between the high temperature and
the Monte Carlo determinations. For O(2) we are so far lacking both precise Monte Carlo
and form factor data, but at this preliminary stage there is rough agreement. We intend to
return to this model in a separate publication.
The situation in O(3) is not completely clear: there is a less than perfect agreement
between the high temperature result and the new high precision Monte Carlo data, and
there is also room for doubt about the agreement between Monte Carlo and form factor.
We cannot resolve this question at the moment, mainly because even with our enormous
amount of Monte Carlo data it is at the moment not clear what the correct extrapolation to
the continuum is.
Note added in proof

In a recent paper by M. Caselle, M. Hasenbusch, A. Pelisetto and E. Vicari (hepth/003049), gR is computed in the Ising model using a completely different method to
that presented in this paper. Their result, gR = 14.69735(3), is in excellent agreement with
ours in Eq. (8.1).
Acknowledgements
This investigation was supported in part by the Hungarian National Science Fund OTKA
(under T030099), and also by the Schweizerische Nationalfonds. The work of M.N. was
supported by NSF grant 97-22097.
Appendix A. General formula for the dominant term

Here we describe the generalization of the formula (3.7) for the dominant 1-2-1 particle
contribution to gR to general integrable QFTs without bound states and operators other
than the fundamental field. The latter is particularly natural in the form factor approach
because fundamental and composite operators are treated on an equal footing. Thus let
Ol be possibly distinct, possibly non-scalar but parity odd operators Ol and write ol for
the quantum numbers labeling them. Parallel to (2.13) we define the Green functions by
e
Sco1 oL (k1 , . . . , kL ) = (2)2 (2) (k1 + + kL ) Go1 oL (k1 , . . . , kL ),
(A.1)
where e
Sco1 0L (k1 , . . . , kL ) is the Fourier transform of the connected part of the Euclidean
correlation function hO1 (x1 ) OL (xL )i. The obvious generalization of the intrinsic
coupling is
Go1 o2 o3 o4 (0, 0, 0, 0)
gR = N M 2 P
oj ok (0, 0)2 .
j <k G
(A.2)
651
Here M is again the mass gap and the constant N is conveniently adjusted to normalize
the 1-particle contribution to the denominator to unity. If Fao are the constant 1-particle
form factors of O, the leading 1-particle contribution to Go1 o2 (0, 0) is just Z o1 o2 :=
M 2 Fao1 C ab Fbo2 , where C ab is the charge conjugation matrix associated with the given
P
S-matrix (cf. below). Thus we take N = j <k (Z oj ok )2 . With these normalizations the
dominant 1-2-1 particle contribution to the coupling (A.2) is
1 X os1 os2 ;os3 os4
D
gR |121 =
2
sS4

1
du X
4
Faos1 Fbos4 C aa3 C bb3
+
2 Z os1 os2 Z os3 os4 +
2
4
u
16
ch
u
sS4
0

os2
a2 b2 a1 b1 os3
Fa3 a2 a1 (i, u, u)C

C
Fb3 b2 b1 (i, u, u) .
(A.3)
Here the symmetrization is over all elements of the permutation group S4 . D is defined in
cd ( ) by
terms of the given bootstrap S-matrix Sab

d ab
0
0
( )
Fao1 Fbo2 C cc C dd Fco03 Fdo04
(A.4)
D o1 o2 ;o3 o4 = i Scd
d
=0
o ( , , ) is the 3-particle form factor of O. Taking the results for the O(n)
and Fabc
1 2 3
models as a guideline one would expect that (A.3) typically yields about 98% of the full
answer for the coupling.
In the following we describe the derivation of (A.3). In contrast to that of (3.7) we
keep track here of the distributional terms like (2.19) and show explicitly that they cancel
out in the final answer. In particular this illustrates that the use of the simplifying limit
procedure (2.23) is justified.
To fix conventions we first recall the defining relations of a generic bootstrap S-matrix.
dc
( ), C, is called a two particle S-matrix if
A matrix-valued meromorphic function Sab
it satisfies the following set of equations. First the YangBaxter equation
kp
ji
pi
kj
nm
nm
(12 )Snc (13)Smp (23 ) = Sbc
(23 )Sam (13 )Spn (12 ),
Sab
(A.5)
where 12 = 1 2 , etc. Second unitarity (A.6a,b) and crossing invariance (A.6c)

mn
cd
( ) Snm
( ) = ad bc ,
Sab
(A.6a)
mc
nd
( ) Sbm
(2i ) = ad bc ,
San
(A.6b)
0 ca 0
dc
( ) = Caa 0 C dd Sbd
Sab
0 (i
(A.6c)
),
where (A.6c) together with one of the unitarity conditions (A.6a), (A.6b) implies the other.
Further real analyticity and Bose symmetry
dc
dc
dc
cd
( ),
Sab
( ) = Sba
( ).
(A.7)
Sab ( ) = Sab
Finally the normalization condition
dc
(0) = ac bd .
Sab
(A.8)
652
The indices a, b, . . . refer to a basis in a finite dimensional vector space V . Indices can
be raised and lowered by means of the (constant, symmetric, positive definite) charge
conjugation matrix Cab and its inverse C ab , satisfying Cad C db = ab . The S-matrix is
a meromorphic function of . Bound state poles, if any, are situated on the imaginary
axis in the so-called physical strip 0 6 Im < . From crossing invariance and the
dc (i) = C C dc is always regular, in contrast to
normalization (A.8) one infers that Sab
ab
dc (i) which may be singular.
Sab
Next we prepare the counterparts of Eqs (2.10), (2.11).
h0|O1 |mihm|O2 |0i Imo1 o2 ( ),
o o o3 o4
1 2
h0|O1 |kihk|O2 |lihl|O3 |mihm|O4 |0i Iklm
(| | ),
(A.9)
where
Imo1 o2 ( ) = FAo1 ( )C AB FBo2T ( T ),
o o o3 o4
1 2
Iklm
(A.10)
o
o4
T
(| | ) = FAo1 ()C AB FBo2T C (T | )C CD FD3T E ( T | )C EK FK
.
T
From the S-matrix exchange relations it follows that Imo1 o2 ( ) is a completely symmetric
o1 o2 o3 o4
(| | ) is symmetric in each of the sets
function in = (1 , . . . , m ). Similarly Iklm
of variables = (1 , . . . , k ), = (1 , . . . , l ) and = (1 , . . . , m ). As before we denote
by Vm (k1 , k2 ) and Vklm (k1 , k2 , k3 , k4 ) the quantities (2.8) with the integrations over the
rapidities performed, where the measure is inherited from (2.25). For simplicity we drop
the operator labels oj in the notation. When evaluated at kj = (M sh j , 0) we write
vklm (1 , 2 , 3 , 4 ), etc.
In the next step one inserts the expressions for the generalized form factors in terms of
the ordinary form factors; see [16] for an account in the present conventions. For m = 2
one obtains explicitly
v121 (1 , 2 , 3 , 4 ) =
Fao1 C ab C cd Fdo4
4Fmo2 C mn Fn 3 Cbc
(1 + 4 )
ch 2 (ch 1 + ch 2 )
8 ch2 1 ch2 4
o
4Fmo2 Fn 3
S mn (i 1 + 2 ) (2 + 4 )
+
ch 2 (ch 1 + ch 2 ) cb
1
o3
F o2 C mn Fcbn
(4 + i i, 1 , 2 )
+
ch 2 (ch 1 + ch 2 ) m
1
o2
F o3 C mn Fbcn
+
(1 + i i, 4 , 3 )
ch 3 (ch 3 + ch 4 ) m
Z
1
du
C mk C nl
+
2
4 4 ch (u/2) + (sh 1 + sh 2 )2
0
o2
Fbmn
(1 + i i, u/2, + u/2)

o3
Fckl (4 + i i, u/2, + u/2) .
(A.11)
653
The i 0 limit of this expression can be evaluated on general grounds. The key
observation is that a three particle form factor has the following universal small rapidity
expansion

1
1
2i Ca1 a2 Fao3 + Ca1 a3 Fao2

12 i 13 i

d cd
cd
= i Sab
( ) .
(A.12)
13Dacd2 a3 12 Dadc2 a3 Fdo , where Dab
d
=0

Fao1 a2 a3 (1 + i i, 2, 3 ) =
+ 2Ca1 c
This expression is uniquely determined by the following properties: (i) The numerator is
linear and boost invariant in the rapidities. (ii) It obeys the (linearized) S-matrix exchange
relations in 2 and 3 . (iii) It has simple poles at 21 + i and 31 + i with residues
dictated by the form factor residue equation (see, e.g., [16] for an account in the present
conventions). Using (A.12) in (A.11) one can compute the small i behavior of v121 . We
(I )
(II)
denote by v121 the contribution from the non-integrated part and by v121 that from the
integrated part. One finds
1
(I ) . o1 o3 o2 o4
Z
= Z
(1 + 3 ) (1 + 4 ) D o1 o2 ;o3 o4 ,
v121
4
2
(A.13)
.
where = again indicates that both sides give the same result for the symmetrized i 0
limit and D o1 o2 ;o3 o4 = D o2 o1 ;o4 o3 = D o4 o3 ;o2 o1 is given by (A.4).
Analyzing the small i behavior of the integral in (A.11) (by splitting it according to
R
R
R
+
0 du = 0 du + du, 0 ) one finds a distributional term and a regular one. The
result is
(II)
v121 =

o1 o4 o2 o3
Z
Z
+ Z o1 o3 Z o2 o4 (1 + 4 )
4
Z
1
du
F o1 F o4 C aa3 C bb3 Fao32a2 a1 (i, u, u)
+
4 16 ch2 u a b
0
a2 b2
a1 b1
o
Fb33b2 b1 (i, u, u)

2 o1 o4 o2 o3
o1 o3 o2 o4
Z
+Z
Z
Z
, (A.14)
u2
where the integrand is regular for u 0.

For the generalization of the term (2.19) one obtains in the 1-particle approximation
and in the i 0 limit
o1 o2 o3 o4 (k1 , k2 , k3 , k4 )|kj =(M sh j ,0) =
(1 + 2 ) Z o1 o2 Z o3 o4 .
2
(A.15)
Finally, combining (A.13), (A.14) and (A.15) according to (2.21) one sees that, as
promised in Section 2.2 all distributional terms drop out when computing the righthand side of (2.21). The final result thus does not depend on any prescription how to take
the i 0 limit and is given by (A.3), as asserted.
654
Appendix B. Computation of the 1-4-1 contribution

We start from the general formula Eq. (3.9) with l = 4. For 1 > 2 > 3 > 4 we have
for one of the factors occurring in (3.10)
ha, |S c (0)|b1 , 1 ; b2 , 2 ; b3, 3 ; b4, 4 iin
c
= Fab
(i + i, 1 , 2 , 3 , 4 )
1 b2 b3 b4

+ 4 ab1 ( 1 )Fbc2 b3 b4 (2 , 3 , 4 )
c
(1 , 3 , 4 )Sb1 b2 ,da (1 2 )
+ ( 2 )Fdb
3 b4
c
(1 , 2 , 4 )Sb2 b3 ,ef (2 3 )Sb1 f,da (1 3 )
+ ( 3 )Fdeb
4
c
(1 , 2 , 3 )Sb3 b4 ,fg (3 4 )
+ ( 4 )Fdef

Sb2 g,eh (2 4 )Sb1 h,da (1 4 ) .
(B.1)
For the 5-particle form factor we introduce the reduced form factor through
Faa1 a2 a3 a4 a5 (1 , 2 , 3 , 4 , 5 ) = T5 (1 , . . . , 5 )Gaa1 a2 a3 a4 a5 (1 , 2 , 3 , 4 , 5 ).
(B.2)
Multiplying out we obtain

I141 = K(I ) + K(II) + K(III) ,
(B.3)
where we momentarily omit the arguments (4 , 1 , 1 , 2 , 3 , 4 ). The shorthands are:

X
1
F1b
(i 1 i, 1 , 2 , 3 , 4 )
K(I ) =
1 b2 b3 b4
b1 ,b2 ,b3 ,b4
1
(i + 4 i, 1 , 2 , 3 , 4 ) ,
F1b
1 b2 b3 b4

K(II) = 4 (4 + 1 )K (II) (4 , 1 , 1 , 2 , 3 )

+ (4 4 )K (II) (1 , 4 , 1 , 2 , 3 )
(B.4)
+ (4 3 ) + (4 2 , 2 3 , 3 4 )
K
(III)
+ (4 1 , 1 2 , 2 3 , 3 4 ),

= (4)2 K (III) (4 , 1 , 3 , 4 ) (1 + 1 )(2 4 ) + (1 2 )
(B.5)
+ (2 3 ) + (2 4 , 4 3 , 3 2 )
+ (1 2 , 2 3 , 3 1 ) + (1 2 , 2 4 , 4 3 , 3 1 )
+ (1 3 , 3 2 , 2 4 , 4 1 ),
(B.6)
where
K (II) (, , 1 , 2 , 3 )
X
1
Fb11 b2 b3 (1 , 2 , 3 )F11b
(i + i, , 1, 2 , 3 ) ,
=
1 b2 b3
(B.7)
b1 ,b2 ,b3
K (III) (4 , 1 , 3 , 4 )
X
Fb11 b3 b4 (4 , 3 , 4 )Fb12 b3 b4 (1 , 3 , 4 ) S1b2 ,b1 1 (1 + 4 ).
=
b1 b2 ,b3 ,b4
Because of the symmetry in the i arguments of K(I ) , K (II) , K (III) one has
(B.8)
(I )
v141
(1 , 2 , 3 , 4 ) =
1
12288 3
Z
d1
d2
655
d3
d4
(1 , 2 , 3 , 4 , 1 , 2 ) (I )
K (4 , 1 , 1 , 2 , 3 , 4 ),
P4
k=1 ch k
(B.9)
Z
Z
Z
1
1
(II)
d1
d2
d3
v141 (1 , 2 , 3 , 4 ) =
P3
2
768
1 + k=1 ch k

(1 , 2 , 3 , 2 )K (II) (4 , 1 , 1 , 2 , 3 )

(B.10)
+ (1 , 2 , 3 , 3 )K (II) (1 , 4 , 1 , 2 , 3 ) ,
(III)
(1 , 2 , 3 , 4 ) =
v141
1
64
Z
d1
d2
(1 , 2 , 2 , 4 )
P
2 + 2k=1 ch k
K (III) (4 , 1 , 1 , 2 ).
(B.11)
The contribution (III) is very simple; we can set the i to zero to obtain
(III)
v141 (1 , 2 , 3 , 4 ) =
1
128
Z
d
1
T 2 (2)T 4 ()k (III) (),
ch (1 + ch )
(B.12)
where we decomposed
K (III) (0, 0, , ) = T 2 (2)T 4 ()k (III) ().
(B.13)
Writing similarly
K (II) (, , 1 , 2 , 3 ) = k (II) (, , 1 , 2 , 3 )
1
T ( )
T 2 (i j )
16i<j 63
3
Y
k=1
T (k )
,
T (k i)
(B.14)
one has
(II)
v141
2
X
(II,j )
v141 ,
(B.15)
j =1
with
(II,1)
v141 (1 , 2 , 3 , 4 ) =
1+
(
1
P3
k=1 ch k
1
1
2
T
(
768
1 + 4 )
Y
Z
d1
Z
d2
d3
T 2 (i j )
16i<j 63
(1 , 2 , 3 , 2 )k (II) (4 , 1 , 1 , 2 , 3 )
3
Y
k=1
T (k + 1 )
P
T (k 4 )
656
)
P
,
(1 , 2 , 3 , 3 )k (1 , 4 , 1 , 2 , 3 )
T (k 4 )
T (k + 1 )
k=1
(B.16)
Z
Z
i
(1 , 2 , 2 , 4 )
(I I,2)
d1
d2
v141 (1 , 2 , 3 , 4 ) =
P
128
2 + 2k=1 ch k
(II)
T 2 (1 2 )
2
Y
3
Y
T (k 4 )T (k + 1 )
k=1

k (II) (4 , 1 , 1 , 2 , 4 ) k (II) (1 , 4 , 1 , 2 , 1 ) .
(B.17)
In the latter term we can set the i to zero to obtain

(II,2)
v141 (0, 0, 0, 0)
Z

1
1
T 2 (2)T 4 () Im k (II) (0, 0, , , 0) .
d
=
128
ch (1 + ch )
(B.18)
Lastly we turn to the (I ) contribution. There are many ways to manipulate the integral
into a form more amenable to numerical evaluation. Here we proceed as follows: writing
Y
T 2 (i j )
K(I ) = k (I )(4 , 1 , 1 , 2 , 3 , 4 )
16i<j 64
4
Y
k=1
1
,
T (k + 1 + i)T (k 4 i)
(B.19)
we replace the 1/(x i) distributions by a sum of products of principal parts and delta
functions, thereby obtaining
(I )
v141
3
X
(I,j )
v141 .
(B.20)
j =1
The three terms are:

(I,1)
v141 (1 , 2 , 3 , 4 )
Z
Z
Z
Z
(1 , 2 , 3 , 4 , 1 , 2 )
1
d1
d2
d3
d4
=
P4
3
12288
k=1 ch k
Y
k (I )(4 , 1 , 1 , 2 , 3 , 4 )
T 2 (i j )
16i<j 64
4
Y
k=1
P
P
,
T (k + 1 ) T (k 4 )
(I,2)
v141
(1 , 2 , 3 , 4 ) =
i
1
2
1536 T (1 + 4 )
(B.21)
Z
d1
Z
d2
d3
1+
1
P3
k=1 ch k 16i<j 63
657
T 2 (i j )
3
Y
P
T (k 4 )
k=1
)
3
Y
P
(I )
,
T (k 4 )
+ (1 , 2 , 3 , 3 )k (4 , 1 , 1 , 2 , 3 , 4 )
T (k + 1 )
(1 , 2 , 3 , 2 )k (I ) (4 , 1 , 1 , 2 , 3 , 1 )
T (k + 1 )
k=1
(B.22)
(I,3)
(1 , 2 , 3 , 4 ) =
v141
(I )
1
256
Z
d1
(1 , 2 , 2 , 4 )
P
2 + 2k=1 ch k
d2
(4 , 1 , 1 , 2 , 1 , 4 )T (1 2 )
2
2
Y
T (k 4 )T (k + 1 ).
k=1
(B.23)
In the latter expression we can set the i to zero to obtain
(I,3)
v141 (0, 0, 0, 0)
Z
1
1
T 2 (2)T 4 ()k (I ) (0, 0, , , 0, 0).
d
=
512
ch (1 + ch )
(B.24)
(I,1)
For W (1) := v141 we now invoke the identity

Q
16i<j 64 sh(yi yj )
Q4
k=1 sh(yk + x)
1 X (1)k ch(yk )
ch(x)
sh(yk + x)
k=1
sh(yi yj ),
(B.25)
16i<j 64, i6=k6=j
to get

W (1) = lim W (1) [A](c ) + W (1) [B](c ) ,
c
(B.26)
with the notation

W
(1)
1
[X](c ) =
192 3
Zc
d1 GX (1 ),
X = A, B.
Here
P
GA (1 ) =
T (1 )
P
d2
T (2 )
Z
d3 FA (1 , 2 , 3 )
(B.27)
658
P
=
T (1 )
P
d2
T (2 )

d3 FA (1 , 2 , 3 )

FA (1 , 2 , 3 ) FA (1 , 2 , 3 ) + FA (1 , 2 , 3 ) ,
4G(0)
G(1 )
,
GB (1 ) = 2
12
sh (1 /2)
(B.28)
(B.29)
where
G(1 ) = ch2
1
2
Z
d3 FB (1 , 2 , 3 ).
d2
(B.30)
In these formulae
FX (1 , 2 , 3 )
Q4

1 k (I ) (0, 0, 1 , 2 , 3 , 4 )fX (1 , 2 , 3 , 4 ) k=1 ch2
P
Q
=
4
2 i j
16
ch 4
i<j ch
m=1 ch m
2
P3
where is given through sh = k=1 sh k and
k
2
1 3 1 4
sh
2
2
2 3 2 4 2 3 4
sh
sh
,
sh
2
2
2
2 3 2 2 4 2 3 4
sh
sh
.
fB (1 , 2 , 3 , 4 ) = sh2
2
2
2
For the [B] contribution it is numerically convenient to decompose
(B.31)
4 =
fA (1 , 2 , 3 , 4 ) = 3 sh
W (1) [B](c ) hG(0) + W (1) [B0] + W (1) [B1](c ),
(B.32)
(B.33)
(B.34)
where
W
(1)
1
[B0] =
96 3
W (1) [B1](c ) =
1
96 3
d1
0
Zcut
d1
1
and
Z1
G(1 ) G(0)
sh2 (1 /2)
G(1 )
sh2 (1 /2)

,
(

)
Z1
4
1
1
4 d1
= 0.0014539754.
h=
96 3
sh2 (1 /2) 12
(B.35)
(B.36)
Finally we recombine
v141 (1 , 2 , 3 , 4 ) =
3
X
j =1
with
W (j ) (1 , 2 , 3 , 4 ),
(B.37)
659
(I,1)
W (1) = v141
,
(I,2)
(II,1)
(I,3)
W (3) = v141
(II,2)
+ v141
W (2) = v141 + v141 ,
(B.38)
(III)
+ v141
.
Case n = 1:
Here we simply have (recall the 2-particle S-matrix = 1)
k (I ) = 16,
k (II) = 8i,
k (III) = 4,
(B.39)
from which one sees

(I,2)
(II,1)
v141 (0, 0, 0, 0) = v141 (0, 0, 0, 0),

1 (II,2)
(I,3)
(III)
v141 (0, 0, 0, 0) = v141 (0, 0, 0, 0) = v141 (0, 0, 0, 0).
2
(B.40)
W (2) = 0 = W (3) ,
(B.41)
Thus
with
given by Eq. (B.21) and
so that for the Ising case we simply get v141 =
k (I ) by Eq. (B.39). This is, as expected, the same expression as that obtained directly with
the form factor written as a product over principal parts as in Eq. (5.2).
W (1) ,
W (1)
Case n = 3:
Firstly for W (1) we obtain
W (1) = 0.0005420(1).
(B.42)
Next for W (2) one has

W (2) (1 , 2 , 3 , 4 )
1
1
=
2
768 T (1 + 4 )
(
Z
d1
Z
d3 J (1 , 2 , 3 )
d2
3
Y
P
T (k 4 )
k=1
)
3
Y
P
,
T (k + 4 )
+ (1 , 2 , 3 , 3 )w (2) (1 , 4 , 1 , 2 , 3 )
T (k 1 )
(1 , 2 , 3 , 2 )w
(2)
(1 , 4 , 1 , 2 , 3 )
T (k + 1 )
k=1
(B.43)
where
J (1 , 2 , 3 ) =
1+
1
P3
k=1 ch k
T 2 (i j ),
(B.44)
i (I )
k (4 , 1 , 1 , 2 , 3 , 1 ),
2
(B.45)
16i<j 63
and
w(2) (1 , 4 , 1 , 2 , 3 )
= k (II) (4 , 1 , 1 , 2 , 3 ) +
660
w (2) (1 , 4 , 1 , 2 , 3 )
= k (II) (1 , 4 , 1 , 2 , 3 )
i
+ k (I ) (4 , 1 , 1 , 2 , 3 , 4 ).
2
Explicit calculation reveals the fact
w (2) (1 , 4 , 1 , 2 , 3 ) = w(2) (4 , 1 , 1 , 2 , 3 ).
(B.46)
(B.47)
Now we expand w(2) for small i :

w(2) (1 , 4 , 1 , 2 , 3 ) = w0 (1 , 2 , 3 ) + (1 + 4 )w1 (1 , 2 , 3 )

+ (1 4 )w2 (1 , 2 , 3 ) + O i2 .
(B.48)
In fact we do not require w2 . Note that the functions wi are real, so that in particular
Im w(2) (0, 0, 1 , 2 , 3 ) = 0,
(B.49)
which is needed to avoid a singularity in W (2) for i 0. Hence

W (2) = W (2) [A] + W (2) [B],
(B.50)
with
W
(2)
1
[A] =
192 2
Z
d1
1
J (1 , 2 , 0 )w1 (1 , 2 , 0 ),
ch 0
d2
(B.51)
where 0 is determined through sh 0 = sh 1 sh 2 , and

W (2) [B](k1 , k2 , k3 , k4 )
1
1
=
768 2 T (1 + 4 )
Z
d1
Z
d2
d3 Z(1 , 2 , 3 )
)
P
+ (2 3 , 1 4 ) ,
T (k + 1 )
(1 , 2 , 3 , 2 )
T (k 4 )
k=1
(B.52)
(
3
Y
with
Z(1 , 2 , 3 ) = J (1 , 2 , 3 )w0 (1 , 2 , 3 ).
We then see that
W (2) [B]
(B.53)
is a sum of two parts
W (2) [B] = W (2) [B1] + W (2) [B2],
(B.54)
with
W
(2)
1
[B1] =
64 2
W (2) [B2] =
Z
d1
1
384 2
d2
d1
P
1
Z(1 , 2 , 0 )
,
ch 0
sh 1
d2

1 Z(1 , 2 , 3 )
.
ch 0 3
ch 3
3 =0
(B.55)
661
Numerically this gives

W (2) [A] = 4.41085(1) 104,
W (2) [B1] = 4.9600(1) 105 ,
W
(2)
[B2] = 1.1503(1) 10
(B.56)
and hence
W (2) = 0.00050219(1).
(B.57)
Finally we turn to the computation of W (3) . Due to Eq. (B.49) it follows that
(II,2)
(I,3)
(0, 0, 0, 0) = 2v141
(0, 0, 0, 0).
v141
(B.58)
Now explicit computation yields

2

k (I ) (0, 0, , , 0, 0) = 6 3 (0, , ) 40 2 + 32 2 ,
2

k (III) () = 12 63 (0, , ) 2 + 2 .
(B.59)
So for W (3) we arrive at

Z 6
5
sh (/2)T 6 () ( 2 + 2 )(4 2 + 2 )( 2 + 2 2 )
(3)
W =
64
6 ( 2 + 4 2 )2
ch5
0
= 0.0004682756.
(B.60)
Appendix C. Computation of the 1-2-3 contribution

We use the results (3.57)(3.61) with m = 3, and begin with contribution (I V ):
Z
1
(1 , 2 , 3 , 4 )
(IV)
d3
v123 (1 , 2 , 3 , 4 )
P3
2
1536
i=1 ch i
G (IV) (2 + i , 1 + i , 1 , 2 , 3 ).
Here
G (IV) (A) =
1
F11a
(A)Fa11 a2 a3 (A0 ) = T5 (A)T3 (A0 ) g (IV) (A),
1 a2 a3
(C.1)
(C.2)
a1 a2 a3
where A stands for 1 , 2 , 3 , 4 , 5 and A0 for 3 , 4 , 5 . Note that g (IV) (A) is totally
symmetric in the subset A0 .
We decompose the 1/(x i) factors to obtain terms involving products of principle
parts and delta-functions; only terms having less than three delta-functions contribute in
the Lim procedure, i.e.,
(IV)
=
v123
3
X
s=1
The terms are
(IV,s)
v123
.
(C.3)
662

(IV,1)
v123
(1 , 2 , 3 , 4 )
Z
(1 , 2 , 3 , 4 )
1
T
(
)
d3
P3
1
2
1536 2
i=1 ch i
Y
Y
g (IV) (i, i, 1, 2 , 3 )
T 2 (i j )
i<j
P
P
, (C.4)
T (k + 1 ) T (k + 2 )
(IV,2)
v123 (1 , 2 , 3 , 4 )
i
128
d2
(2 , 1 , 2 , 4 )
P
1 + 2i=1 ch i
g (IV) (i, i, 0, 1, 2 )T 2 (1 2 )
T (k + 2 )
P
,
T (k + 1 )
while for the term involving the product of two delta-functions one finds

(IV,3)
(1 , 2 , 3 , 4 ) = O 3 .
v123
(C.5)
(C.6)
The contribution (I V , 1) is antisymmetric in 1 2 and so it doesnt contribute in the

sum over permutations. In the contribution (IV, 2) we can take the 0 limit and obtain
i
(IV)
(0, 0, 0, 0)
v123
128
Z
d
T 2 (2)
g (IV) (i, i, 0, , ).
ch [1 + 2 ch ]
(C.7)
Now turn to the (I I I ) contribution:

(III)
v123 (1 , 2 , 3 , 4 )
Z
(1 , 2 , 1 , 2 )
1
P
d2 P
G (III) (1 + i , 1 , 2 , 3 ),
64
ch
[1
+
ch
]
i
j
i
j
where
G (III) (1 , 2 , 3 , 4 ) =
(C.8)
1
F1b
(1 , 2 , 3 )Fb11 b2 1 (2 , 3 , 4 )
1 b2
b1 b2
= T3 (1 , 2 , 3 )T3 (2 , 3 , 4 ) g (III) (1 , 2 , 3 , 4 ).
(C.9)
Here one can set the i to zero to obtain

(III)
(0, 0, 0, 0)
v123
1
128
Z
d
T 2 (2)
ch2 [1 + 2 ch ]
g (III) (i, , , 0).
(C.10)
For the (II) contribution:
Z
1
(1 , 2 , 1 , 2 )
d2
ch 1 + ch 2
256 2
Z
(1 , 1 , 2 , 4 ) (II)
d2
G (1 + i , 1 , 2 , 1 , 2 ),
P
ch 1 + 2i=1 ch i
(II)
(1 , 2 , 3 , 4 )
v123
where
(C.11)
G (II) (A) =
XX
663
1
1
1
F1bb
(A0 )Fba
(B)Fba
(B 0 )
1 a2
1 a2
b a1 ,a2
= T3 (A0 )T3 (B)T3 (B 0 ) g (II) (A),
(C.12)
A0
stands for 1 , 2 , 3 , B stands for 3 + i , 4 , 5 ,

where A stands for 1 , 2 , 3 , 4 , 5 ;
and B 0 stands for 2 , 4 , 5 . Making the familiar decomposition of the singular distributions
one obtains:
(II)
=
v123
3
X
(II,s)
v123
,
(C.13)
s=1
with
Z
1
(1 , 2 , 1 , 2 )
d2
256 2
ch 1 + ch 2
Z
(1 , 1 , 2 , 4 ) (II)
d2
g (1 + i, 1, 2 , 1 , 2 )T 2 (1 2 )
P
ch 1 + 2i=1 ch i
Y
Y
P
P
,
(C.14)
T (1 j )
T (1 2 )
T (1 + i )
T (2 j )
i
j
Z
i
(1 , 2 , 1 , 2 )
(I I,2)
v123 (1 , 2 , 3 , 4 )
d2
64
[ch 1 + ch 2 ][1 + ch 1 + ch 2 ]
Y
P
,
T (j + 3 )
g (II) (1 + i, 1 , 2 , 2 , 3 )T 2 (1 2 )
T (j + 1 )
(II,1)
v123 (1 , 2 , 3 , 4 )
Z
i
d2 T 2 (1 2 )
256

Y
(1 , 1 , 2 , 4 ) (II)
g (1 + i, 1 , 2 , 1 , 2 )
T (j + 1 )
P2
1 + i=1 ch i
j
(C.15)
(I I,3)
v123 (1 , 2 , 3 , 4 )
(2 , 1 , 2 , 4 ) (II)
P
+
g (1 + i, 2 , 1 , 1 , 2 )
P
T (j + 2 )
1 + 2i=1 ch i

P
,
(C.16)
T (j + 1 )
In the contributions s = 2, 3 we can set the i to zero to obtain

(II,2)
(0, 0, 0, 0)
v123
(I I,3)
(0, 0, 0, 0)
v123
i
128
i
128
Z
d
Z
d
Finally for the (I ) contribution

(I )
(1 , 2 , 3 , 4 )
v123
1
6144 3
T 2 (2)
ch2 [1 + 2 ch ]
g (II) (i, , , , 0), (C.17)
T 2 (2)
g (II) (i, 0, 0, , ),
ch [1 + 2 ch ]
Z
d2
(1 , 2 , 1 , 2 )
ch 1 + ch 2
(C.18)
664
d3
(1 , 2 , 3 , 4 ) (I )
G (1 + i , 1 , 2 , 1 , 2 , 3 ),
P3
i=1 ch i
(C.19)
with
G (I ) (A) =
X X
1
1
F1bb
(A0 )Fbba
(B)Fa11 a2 a3 (B 0 )
1 a2 a3
(C.20)
b a1 ,a2 ,a3
= T3 (A0 )T5 (B)T3 (B 0 ) g (I ) (A).
(C.21)
Here A stands for 1 , 2 , 3 , 4 , 5 , 6 ; A0 stands for 1 , 2 , 3 , B stands for 3 + i ,

2 + i, 4 , 5 , 6 , and B 0 stands for 4 , 5 , 6 .
Then we rearrange to terms where after doing the 2 integral the singularities in the 1
integral all have negative imaginary parts
(I )
=
v123
4
X
(I,s)
v123
.
(C.22)
s=1
The terms are:

1
(I,1)
v123 (1 , 2 , 3 , 4 )
6144 3
Z
Z
d2
(1 , 2 , 1 , 2 )
ch 1 + ch 2
(1 , 2 , 3 , 4 ) (I )
g (1 + i, 1 , 2 , 1 , 2 , 3 )
P3
i=1 ch i
Q
2
T 2 (1 2 )
i<j T (i j )
Q
,
T (1 + 1 + i)T (1 + 2 i) k T (1 + i k )T (2 i k )
(C.23)
Z
i
(
,
)
1
2
1
2
(I,2)
(1 , 2 , 3 , 4 )
v123
d2
ch 1 + ch 2
512 2
Z
(
,
)
1
1
2
4
d2
g (I ) (1 + i, 1 , 2 , 1 , 1 , 2 )
P
ch 1 + 2i=1 ch i
Q
T (1 2 )T 2 (1 2 ) i T (1 i )
Q
,
(C.24)
T (1 + 1 + i)T (1 + 2 i) j T (2 i j )
d3
i
(I,3)
T (1 2 )
v123 (1 , 2 , 3 , 4 )
3072 2
Z
(1 , 2 , 3 , 4 ) (I )
g (1 + i, 2, 1 , 1 , 2 , 3 )
d3
P3
i=1 ch i
Q
2
i<j T (i j )
,
Q
k T (1 i k )T (2 + i k )
1
(I,4)
T (1 2 )
v123 (1 , 2 , 3 , 4 )
256
Z
(2 , 1 , 2 , 4 ) (I )
g (1 + i, 2, 1 , 2 , 1 , 2 )
d2
P
1 + 2i=1 ch i
(C.25)
Q
T 2 (1 2 ) i T (2 + i )
Q
.
j T (1 + i + j )
665
(C.26)
As for the last contribution, it vanishes in the limit i 0

(I,4)
(0, 0, 0, 0) = 0.
v123
(C.27)
For the contribution (I, 1) we now perform the 2 integral and shift the 1 integral to
larger imaginary part, after which we can send all the i to zero to obtain
(I,1)
v123 (0, 0, 0, 0)
Z
du1
1
512 3
++i
Z
d
+i
ch4 (/2)
ch4
du2 T 2 (u1 )T 2 (u2 )T 2 (u1 + u2 )M (3)(u)2
g (I ) (i, , , 1, 2 , 3 )
Y ch k + ch
,
ch k ch
(C.28)
where the k are determined in terms of the us as in Eq. (2.30).

For the (I, 2) term we obtain
(I,2)
(I,5)
(I,6)
v123 v123 + v123 + O(),
(C.29)
where
Z
i
(1 , 2 , 1 , 2 )
(I,5)
d2
v123 (1 , 2 , 3 , 4 )
512 2
ch 1 + ch 2
Z
(
,
)
1 1 2
4
d2
g (I ) (1 + i, 1 , 2 , 1 , 1 , 2 )
P
ch 1 + 2i=1 ch i
Y
Y
P
P
,
T (1 j )
T (1 2 )T 2 (1 2 )
T (1 + i )
T (2 j )
i
j
Z
1
(1 , 2 , 1 , 2 )
(I,6)
P
d2 P
v123 (1 , 2 , 3 , 4 )
128
ch
i [1 + j ch j ]
i
g (I ) (1 + i, 1 , 2 , 1 , 2 , 3 )
Y
P
.
T (i + 3 )
T 2 (1 2 )
T (i + 1 )
(C.30)
(C.31)
In the latter we can do the 2 integral and set the i to zero to obtain
(I,6)
(0, 0, 0, 0)
v123
Z
T 2 (2)
1
g (I ) (i, , , , , 0).
d 2
256
ch [1 + 2 ch ]
Finally
(C.32)
666
1
(I,3)
v123
(0, 0, 0, 0) =
256
Z
d
T 2 (2)
g (I ) (i, 0, 0, , , 0).
ch [1 + 2 ch ]
(C.33)
Summarizing the results we have

v123 (0, 0, 0, 0) =
5
X
V (j ) ,
(C.34)
j =1
where the five terms are as follows:

(I,1)
V (1) = v123 (0, 0, 0, 0),
(C.35)
given in Eq. (C.28). Further

(III)
(IV)
(0, 0, 0, 0) + v123
(0, 0, 0, 0)
V (2) = v123
Z
T 2 (2) (2)
1
g (),
d
=
128
ch2
(C.36)
where
g (2) () =
g (III) (i, , , 0) ig (IV) (i, i, 0, , ) ch

.
1 + 2 ch
(C.37)
Next
(I,5)
(I I,1)
(0, 0, 0, 0) + v123
(0, 0, 0, 0)
V (3) = v123
Z
Z
ch2 (/2) P
(, 1 , 2 )
1
d
g (3) (, 1 , 2 )
d2
=
P
3
512 2
T
()
ch
ch + 2i=1 ch i
T 2 (1 2 )
T (j )
P
,
T (j + )
(C.38)
where
g (3) (, 1 , 2 ) = 2g (II) (i, , , 1, 2 ) + ig (I ) (i, , , , 1, 2 ).
(C.39)
Further
(I,3)
(II,3)
V (4) = v123 (0, 0, 0, 0) + v123 (0, 0, 0, 0)

Z
T 2 (2)
1
g (4) (),
d
=
256
ch [1 + 2 ch ]
(C.40)
where
g (4) () = ig (3) (0, , ).
Finally
(C.41)

(I,6)
(II,2)
V (5) = v123
(0, 0, 0, 0) + v123
(0, 0, 0, 0)
Z
T 2 (2)
1
g (5) (),
d 2
=
256
ch [1 + 2 ch ]
667
(C.42)
where
g (5) () = ig (3) (, , 0).
(C.43)
Case n = 1:
Here we have simply
g (I ) (A) = 16,
g (II) (B) = 8i,
g (III) (C) = 4,
g (IV) (D) = 8i,
(C.44)
and so
g (2) () = 4,
g (r) = 0 , r = 3, 4, 5.
(C.45)
Thus
V
(2)
(1)
1
=
32
sh2
ch
1
=
128 3
1
,
48
Z
d2 u T 2 (u1 )T 2 (u2 )T 2 (u1 + u2 )
S(u1 , u2 )
,
M (3) (u)2
(C.46)
with
++i
Z
S(u1 , u2 ) =
+i

3
[1 + ch ]2 Y ch k + ch
.
ch k ch
ch4
(C.47)
k=1
Numerically we find
V (1) = 0.000842721(1).
(C.48)
Case n = 3:
First we have
V
(1)
9
=
32768

2
d2 u(u1 )(u2 )(u1 + u2 ) M (3) (u)2 S(u1 , u2 ),
(C.49)
where

(u)2 =
and
u2 + 2
T 4 (u),
u2 (u2 + 4 2 )
(C.50)
668

++i
Z
S(u1 , u2 ) =
(1 + ch )4
ch6
+i
h(I ) (i, , , 1, 2 , 3 )

k2 2
ch k + ch 2
(4 2 + 2 ) Y
.
2
( + 2 )3
(k2 2 )2 + 2 2 (k2 + 2 ) + 4 ch k ch
k
(C.51)

V (1) = 0.000844527(1).
(C.52)
Doing the contractions yields

g (III) (i, , , 0) =
6 (2 2 + 2 ) T 2 (2)
,
4 ( 2 + 2 )2 2
g (IV) (i, i, 0, , ) = 2ig (III) (i, , , 0).
(C.53)
g (2) () = g (III) (i, , , 0),
(C.54)
Thus
and
V
(2)
5
=
512
Z
d
(2 2 + 2 )(4 2 + 2 )
= 0.0074380765.
2 ch6 ( 2 + 2 )2 ( 2 + 4 2 )
sh4
(C.55)
Next by explicit computation one verifies

g (4) () = 0,
g (5) () = g (5) (),
(C.56)
and thus 4
V (4) = V (5) = 0.
(C.57)
It remains to compute V (3) . Shifting the 1 integral we obtain the representation

V
(3)
1
=
256 2
Z
d
ch2 (/2) P
P
sh(/2) ch3 T ()
(3)
g (, 1 , 2 ) ch (1 /2) ch((2
+ )/2)
ch 2 ch ((1 2 )/2)[ch + ch 1 + ch 2 ]
1
T (1 2 )T (1 )T (2 ) ch ,
2
(C.58)
where
1 = ,
(C.59)
and 2 is determined by
sh 2 = sh sh 1 .
4 Eq. (C.57) is perhaps true for all n but we have not verified this conjecture.
(C.60)
669

V (3) = 0.000125112(1).
(C.61)
Appendix D. Building blocks of form factors

In form factor calculations one often encounters the problem of finding an analytic
function f ( ) satisfying
f ( ) = ( )f ( ),
f (i ) = f (i + ),
(D.1)
for given ( ). If ( ) has the Fourier representation

Z
( ) = e
i()
( ) = 2
sin( ) k(),
(D.2)
with some kernel function k()

then the minimal solution of (D.1) is given by [4]
Z
f ( ) = e() ,
( ) =
d ch ( + i ) 1
k().
sh
(D.3)
The function ( ) has the following properties. If k()

ez (z > 0) for then
( ) is analytic for z < Im < 2 + z and for real
ln 0

k (0) + const.
(D.4)
Re ( ) (i + ) k(0)
2
We encountered in Section 6 the following special case: for some (positive, real)
parameter
( ) = ei () =
(1 + )i +
,
(1 + )i
(D.5)
corresponding to the kernel

k () = e(1+).
(D.6)
We denote the corresponding solution by ( ).

References
[1] C. Newman, Comm. Math. Phys. 41 (1975) 1.
[2] J. Glimm, A. Jaffe, Quantum Physics, 2nd edn., Springer, 1987.
[3] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253; Nucl. Phys. B 133
(1978) 525.
[4] M. Karowski, P. Weisz, Nucl. Phys. B 139 (1978) 455.
[5] M. Karowski, in: W. Rhl (Ed.), Field Theoretical Methods in Particle Physics, 1980, p. 307.
670
[6] F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, World
Scientific, 1992.
[7] M. Falcioni, G. Martinelli, M.L. Paciello, G. Parisi, B. Taglienti, Nucl. Phys. B 225 (1983) 313.
[8] J. Kim, Phys. Lett. B 345 (1995) 469.
[9] J. Balog, M. Niedermaier, F. Niedermayer, A. Patrascioiu, E. Seiler, P. Weisz, Phys. Rev. D 60
(1999) 094508.
[10] M. Campostrini, A. Pelissetto, P. Rossi, E. Vicari, Nucl. Phys. B 459 (1996) 207; Nucl. Phys. B
(Proc. Suppl.) 47 (1996) 751.
[11] M. Campostrini, A. Pelissetto, P. Rossi, E. Vicari, Phys. Phys. D 54 (1996) 1782; Nucl. Phys. B
(Proc. Suppl.) 47 (1996) 755.
[12] P. Butera, M. Comi, Phys. Rev. B 54 (1996) 15828.
[13] A. Pelissetto, E. Vicari, Nucl. Phys. B 519 (1998) 626; Nucl. Phys. B 575 (2000) 579.
[14] J. Balog, M. Niedermaier, F. Niedermayer, A. Patrascioiu, E. Seiler, P. Weisz, in preparation.
[15] M. Niedermaier, Parametric holomorphy? Triviality versus duality in sinh-Gordon, hepth/9909170.
[16] M. Niedermaier, Nucl. Phys. B 519 (1998) 517.
[17] M. Lscher, Nucl. Phys. B 135 (1978) 1.
[18] P. Hasenfratz, F. Niedermayer, Nucl. Phys. B 414 (1994) 785.
[19] T.T. Wu, B. McCoy, C. Tracy, E. Baruch, Phys. Rev. B 13 (1976) 316.
[20] M. Sato, T. Miwa, M. Jimbo, Proc. Jpn. Acad. A 53 (1979) 6.
[21] B. McCoy, hep-th/9403084.
[22] B. Berg, M. Karowski, P. Weisz, Phys. Rev. D 19 (1979) 2477.
[23] M. Campostrini, A. Pelissetto, P. Rossi, E. Vicari, Phys. Rev. B 54 (1996) 7301.
[24] O. Babelon, D. Bernard, Phys. Lett. B 288 (1992) 113.
[25] G.A. Baker, Phys. Rev. B 15 (1977) 1552.
[26] J.C. Le Guillou, J. Zinn-Justin, Phys. Rev. Lett. 39 (1977) 95; Phys. Rev. B 21 (1980) 3976.
[27] J. Kim, A. Patrascioiu, Phys. Rev. D 47 (1993) 2588.
[28] A.B. Zamolodchikov, Int. J. Mod. Phys. A 3 (1988) 743.
[29] G. Delfino, Phys. Lett. B 419 (1998) 291.
[30] J.M. Kosterlitz, D.J. Thouless, J. Phys. C 6 (1973) 1181; J. Phys. C 7 (1974) 1046.
[31] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Oxford, 1989.
[32] H. Babujian, A. Fring, M. Karowski, A. Zapletal, Nucl. Phys. B 538 (1999) 535.
[33] C.H. Woo, Phys. Rev. D 20 (1979) 1880.
[34] A.P. Prudnikov, Yu.A. Brichkov, O.I. Marichev, Integrali i Ryadi, Nauka, Moskow, 1981,
formulae 2.4.22.27.
[35] J. Balog, M. Niedermaier, Nucl. Phys. B 500 (1997) 421; Phys. Rev. Lett. 78 (1997) 4151.
[36] J. Balog, M. Niedermaier, T. Hauer, Phys. Lett. B 386 (1996) 224.
[37] M. Campostrini, A. Pelissetto, P. Rossi, E. Vicari, Phys. Lett. B 402 (1997) 141.
[38] P. Rossi, E. Vicari, private communication.
[39] P. Butera, private communication.
[40] M. Lscher, U. Wolff, Nucl. Phys. B 339 (1990) 222.
[41] P. Hasenfratz, M. Maggiore, F. Niedermayer, Phys. Lett. B 245 (1990) 522.
[42] M. Hasenbusch, hep-lat/9408019.

Ising model description of the SU(2)1 quantum

critical point in a dimerized two-leg spin-1/2
ladder
Y.-J. Wang a,b , A.A. Nersesyan a,c
a The Abdus Salam International Centre for Theoretical Physics, Strada Costiera 11, 34014 Trieste, Italy
b Department of Physics, Nanjing University, 210093 Nanjing, China
c The Andronikashvili Institute of Physics, Tamarashvili 6, 380077 Tbilisi, Georgia
Abstract
A nonperturbative analytical description of the SU(2)1 quantum critical point in an explicitly
dimerized two-leg spin-1/2 Heisenberg ladder is presented. It is shown that this criticality essentially
coincides with that emerging in a weakly dimerized spin-1 chain with a small Haldane gap. The
approach is based on the mapping onto an SO(3)-symmetric model of three strongly coupled quantum
Ising chains. This mapping is used to establish the correspondence between all physical fields of the
spin ladder and those characterizing the SU(2)1 criticality at the infrared fixed point. 2000 Elsevier
PACS: 71.10.Pm; 71.10.Fd; 75.10.Jm
Keywords: Fermions in reduced dimensions; Lattice fermion models; Quantized spin models
1. Introduction
Quantum many-particle systems in one space dimension, such as 1D interacting
electrons, antiferromagnetic spin chains and ladders, exhibit universal properties in the
low-energy limit. The field-theoretical description of these properties is based on an
appropriately chosen conformally invariant (critical) theory which is defined in the highenergy, ultraviolet (UV) limit and then deformed by a number of perturbations consistent
with the structure and symmetry of the underlying microscopic model. While a single
relevant perturbation (with scaling dimension d < 2) drives the model away from criticality
towards a strong-coupling gapped phase in the infrared (IR) limit, the interplay between
several relevant perturbations can lead to a quantum critical point where conformal
invariance is restored, though with a central charge cIR less than that at the unstable UV
fixed point [1]. Understanding of such phase transitions in 1D models, where powerful
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 5 - 9
672
Y.-J. Wang, A.A. Nersesyan / Nuclear Physics B 583 [FS] (2000) 671690
nonperturbative techniques are available, is of great importance because asymptotically

exact results obtained in one space dimension can be relevant to the issue of quantum
critical points in higher-dimensional systems, most notably, in two-dimensional cuprite
superconductors (see, e.g., [2] and references therein).
An example of Abelian field theory displaying a quantum critical point was recently
discussed by Delfino and Mussardo [3]. They considered the double-frequency sineGordon (DSG) model, which is a two-dimensional Gaussian model perturbed by two
relevant vertex operators with the ratio of their scaling dimensions d1 /d2 = 4, and showed
that, upon fine tuning of its parameters, the model exhibits an Ising criticality with central
charge cIR = 1/2. In non-Abelian field theories, the existence of quantum critical points
belonging to the universality class of SU(2)k WessZuminoNovikovWitten (WZNW)
model and resulting from the interplay between several symmetry-preserving relevant
perturbations was anticipated some time ago by Affleck and Haldane [4,5]. They argued
that a massive phase of a translationally invariant spinchain Hamiltonian can be driven
to criticality by an external, parity-breaking perturbation, such as an explicit dimerization.
Apparently, the explicitly dimerized even-leg spin-1/2 ladders, which in the absence of
external perturbations are known to possess a fully gapped excitation spectrum (see for a
review Ref. [6]), are excellent candidates for such a scenario. A nice example of this kind
has been recently given by Martin-Delgado et al. [7,8] (see also [9]) who considered the
standard J -J two-leg spin ladder (J, J > 0)
XX
X
Si,n Si,n+1 + J
S1,n S2,n ,
(1)
Hstand = J
n i=1,2
and modified it by making the constituent chains dimerized with a relative phase : J
J1,2 (n) = J (1)n J 0 . At J = J 0 = 12 J the dimerized ladder transforms to a snake
looking, translationally invariant S = 1/2 Heisenberg chain which is critical and belongs
to the SU(2)1 WZNW universality class with central charge c = 1 [4,5]. Using the mapping
onto a nonlinear sigma-model with a topological ( ) term, in Ref. [7] (see also [10]) it was
argued that, at a given J > 0, there should exist a critical line J = J (J 0 ) along which
the model displays the IR properties of a single S = 1/2 Heisenberg chain.
An analytical description of the critical points resulting from the competition between
different relevant perturbations requires an entirely nonperturbative scheme able to
correctly identify those degrees of freedom that remain massive and the low-energy ones
that eventually become critical. Such a scheme has been recently proposed in Ref. [11] to
tackle the Ising transition in the DSG model. The approach was based on the mapping onto
a model of two quantum Ising (QI) chains coupled by an AshkinTeller (AT) interaction
and also by a magnetic-field type coupling h1z 2z . In the limit where the amplitude h
is considered as the largest energy scale in the problem, the relative Ising degrees of
freedom described by z = 1z 2z become locked, while the total degrees of freedom,
described, e.g., by 1z , asymptotically decouple and can be tuned to criticality. Universality
then implies that such separation between the fast and slow modes is valid close to the
transition at arbitrary h. This approach made it possible to study the Ising criticality in a
number of applications of the DSG model to physical systems [11,12].
673
In this paper, we extend this strategy to provide a consistent nonperturbative description

of the SU(2)1 criticality in the -dimerized two-leg spin ladder. Such a possibility naturally
follows from the correspondence between two weakly coupled Heisenberg chains and
an O(3) Z2 symmetric theory of four noncritical Ising models with order parameters
i , developed in Refs. [13,14]. The new ingredient introduced by the explicit relative
Q
dimerization is a perturbation h 4i=1 i with h J 0 . We stress, however, that the problem
of dimerized ladders is not only different from the DSG model in that instead of two
Ising models here we have four, but most importantly because in the adopted Ising
description and all subsequent nonlocal (change-of-basis) transformations one has to
maintain the unbroken spin rotational symmetry. For this reason, the SU(2)1 quantum
criticality reveals itself in the Ising-model description in a rather nontrivial way. The so
far existing conclusion on the existence of the c = 1 critical point in the S = 1/2 staggered
ladder [7] and in the related model of the explicitly dimerized S = 1 chain [15,16] was
reached either by mapping onto an O(3) nonlinear sigma model which, strictly speaking, is
valid for large S only, or using a perturbative (renormalization group) approach combined
with numerical methods. In other cases [9], the c = 1 criticality of the dimerized ladder (1)
was analyzed by the Abelian bosonization in the way suitable only for the XXZ-version
of the model but not for the SU(2) symmetric case where the explicit dimerization field is
the most relevant perturbation in the problem. Treating our QI-chain model in the strongcoupling limit and applying a nonlocal duality transformation [17,18], below we derive the
effective Hamiltonian describing the low-energy sector of the dimerized spin ladder which
represents a lattice spin-1/2 Heisenberg model with bond-alternating nearest-neighbor
interactions. The SU(2)1 criticality is reached when the external dimerization enforces
the exchange couplings on even and odd bonds to coincide. This mapping is then used
to establish the correspondence between the physical fields of the spin ladder and those
characterizing the SU(2)1 criticality at the IR fixed point.
The paper is organized as follows. In Section 2, we review the field-theoretical approach
to a weakly coupled two-leg Heisenberg ladder and show that the SU(2)1 criticality
emerging upon the explicit -phase dimerization of the constituent chains [7] is essentially
the quantum critical point appearing in a weakly dimerized effective spin-1 chain [15,16]
with a small Haldane gap [19]. In Section 3, we introduce an SO(3) symmetric quantum AT
model extended to include an interaction h1 2 3 . This model, which represents a lattice
version of the field theory discussed in Section 2, is treated in the large-h limit, and the
effective low-energy Hamiltonian displaying the SU(2)1 criticality is derived. In Section 4,
we establish the correspondence between the physical fields of the spin ladder and those
characterizing the SU(2)1 criticality at the IR fixed point. We conclude with a summary and
discussion. The paper contains two appendices where we provide some details concerning
the non-local duality transformation and the derivation of the effective Hamiltonian in the
large-h limit.
674
2. Spin ladder in the continuum limit

As shown in Refs. [13,14], at J J the original lattice model (1) can be mapped onto
an O(3) Z2 -symmetric theory of four massive Majorana fermions, or equivalently, four
noncritical 2D Ising models. In the continuum limit, the effective Hamiltonian density
ivt
(ER x ER EL x EL ) imt ER EL
2

ivs 4
R x R4 L4 x L4 ims R4 L4
2

1
+ g1 (ER EL )2 + g2 (ER EL ) R4 L4
(2)
2
describes a degenerate triplet of Majorana fields, E = ( 1 , 2 , 3 ), and a singlet Majorana
HM =
field, 4 . The mass terms in (2) originate from the relevant part of the interchain interaction,
J n1 n2 , where ni is the staggered magnetization of the ith chain: (1)n Si,n ni (x).
The triplet and singlet masses are given by mt = J , ms = 3J , where > 0 is a
nonuniversal constant. The marginal part of the transverse exchange, J J1 J2 , expressed
in terms of the smooth components of the local spin densities (vector currents Ji ), gives
rise to the four-fermion interaction in (2).
While the right and left components of the vector currents Ji = JiR + JiL admit a local
representation in terms of Majorana bilinears,
i
I = J1 + J2 = (E E ),
2
K = J1 J2 = iE 4 ; ( = R, L)
(3)
this is not so for the staggered fields ni and i , the latter being the dimerization operator
defined as (1)n Si,n Si,n+1 i (x). However, these fields with scaling dimension 1/2
can be expressed as products of the order and disorder parameters, a and a , of the related
Ising models (see Refs. [13,20] for more details):
n+ (1 2 3 4 , 1 2 3 4 , 1 2 3 4 ),
n (1 2 3 4 , 1 2 3 4 , 1 2 3 4 ),
+
1 2 3 4 ,
1 2 3 4 ,
(4)
(5)
(6)
where n = n1 n2 , = 1 2 . Since the Majorana mass is related to the temperature

of the associated Ising model via the relation m (T Tc )/Tc , the triplet of Ising models
is disordered (mt > 0) while the singlet Ising system is ordered (ms < 0). As follows from
Eqs. (4)(6), this fact plays a crucial role in the behaviour of the correlation functions. In
particular, the ground state of the system is parity symmetric (h i = 0), and the dynamical
spin susceptibility 00 (q, ), calculated by Fourier transforming the asymptotics of the
correlator hn (x, ) n (0, 0)i, exhibits the existence of a coherent S = 1 magnon peak
at 2 = ( q)2 vt2 + m2t . Thus, at low energies, the standard two-leg ladder represents a
Haldanes disordered spin liquids with n playing the role of the staggered magnetization
of the effective S = 1 chain. In fact, the model (2) can be thought as a continuum low-
675
energy theory of a spin-1 chain with a small Haldane gap, represented by a triplet of
Majorana fermions [21,22], coupled to a noncritical Ising model:
vt
b + g1 IR IL
(IR IR + IL IL ) + mt Tr
H=
2

ivs 4
R x R4 L4 x L4 ims R4 L4 + g2 KR KL + h4 Tr g.
(7)
2
The first term in the r.h. side of (7) is the Hamiltonian of the critical SU(2)2 WZNW model
describing universal properties of the S = 1 chain at the exactly integrable, multicritical
b is a 33 matrix
point [23,24], IR,L are the level-2 chiral vector currents defined in (3),
field, which is a primary field of the SU(2)2 WZNW model with scaling dimension d1 = 1,
b = iER EL ), and g is the WZNW field in the fundamental (22) representation,
(Tr
with dimension d2 = 3/8. The operator
t = Tr g = 1 2 3
(8)
is a parity-breaking, dimerization field of the triplet sector which couples to the order
parameter 4 of the singlet Ising system. Since the latter is in the ordered phase, in the
lowest order one can replace in Eq. (7) the operators 4 and R4 L4 by their expectation
values and only deal with the triplet sector. Thus, the original problem of the dimerized
spin ladder reduces to the problem of a spin-1 chain with a small Haldane gap and a weak
bond alternation:
vt
t,
b + h
(IR IR + IL IL ) + g1 IR IL + m
t Tr
(9)
Ht =
2
t are renormalized values of the coupling constant and triplet mass,
where g1 and m
respectively, and
h = hh4 i.
(10)
The neglected residual interaction between the singlet modes and those low-energy triplet
degrees of freedom that eventually become critical affects the precise location of the c = 1
critical point but does not alter the universal properties of the transition.
3. Generalized SO(3) quantum AshkinTeller model and the SU(2)1 criticality

Guided by the well-known correspondence between the theory of a massive Majorana
fermion and a QI model, i.e., the Ising chain in a transverse magnetic field [25,26], in this
section we consider the following model of three coupled QI chains
X X

(3)
z
z
x
a,n+1
+ 1a,n
J a,n
HQI =
a=1,2,3 n
XX
a<b n
X

z
z
z
z
z
z
z
x
x
a,n+1
b,n
b,n+1
+ a,n
b,n
1,n
2,n
3,n
.
a,n
+ h
(11)
The AT interaction between the chains is chosen to be self-dual and parametrized by a

single constant K. The last term in (11) couples all three chains together.
676
Let us first check that the lattice Hamiltonian (11) is indeed SO(3)-symmetric and then
show that, in the continuum limit, it reduces to the model (9). The hidden SO(3) symmetry
of the Hamiltonian (11) becomes manifest in the Majorana representation [27]. We introduce lattice Majorana fields, a,n and a,n (a = 1, 2, 3), satisfying the anticommutation
relations
{a,n , b,m } = {a,n , b,m } = 2ab nm ,
{a,n , b,m } = 0,
(12)
and then make use of the JordanWigner transformations:

!
n
Y
x
z
x
a,n = ia
a,m a,n .
a,n = ia,n a,n ,
(13)
m=1
Here a are Klein factors which anticommute among themselves,

{a , b } = 2ab ,
(14)
but commute with all the other operators. Quantum disorder operators
on the dual lattice {n + 1/2}, are given by
xa,n+1/2 = ia,n a,n+1 ,
za,n+1/2 =
n
Y
x,z
a,n+1/2 ,
x
a,m
.
defined
(15)
m=1
The order and disorder operators are related to each other by the duality transformations:
z
z
a,n+1
= xa,n+1/2,
a,n
x
za,n1/2 za,n+1/2 = a,n
.
(16)
With the definitions (13), (15), the lattice Majorana fields

z
z
za,n1/2 = a za,n1/2 a,n
,
a,n = a a,n
(17)
z
za,n+1/2
a,n = ia a,n
(18)
z
= ia za,n+1/2 a,n
,
satisfy the required anticommutation relations (12).

Using Eqs. (13), we fermionize the Hamiltonian (11). The first line in (11) immediately
transforms to an O(3)-symmetric sum of three lattice Majorana models:
X
(J En En+1 1En En ).
H0 = i
n
(3)
The AT part of HQI transforms to

XX

(a,n a,n+1 )(b,n b,n+1 ) + (a,n a,n )(b,n b,n ) ,
HAT = K
(19)
a<b n
and is also O(3)-symmetric because

X
3 1
(a,n a,m )(b,n b,m ) = + (En Em )2 .
2 2
a<b
z
z
z
2,n
3,n
involves products of three Majorana fields, 1m 2m 3m
The product of 1,n
and 1m 2m 3m . Notice that the product of the three components of a Majorana triplet,
677
1 2 3 = (1/3!) abc a b c , is an SO(3) invariant object. Therefore the h-perturbation

in (11) only breaks the discrete subgroup Z2 O(3), thus reducing the symmetry of the
model (11) to SO(3).
If the triplet of QI chains is close to the critical point and the interchain interaction is
weak,
| J | J
and K, h J ,
(20)
one can pass to the continuum limit in which the lattice operators a,n and a,n are replaced
by slowly varying Majorana fields
p
p
a,n 2a0 a (x).
a,n 2a0 a (x),
Here the factor 2 ensures the correct continuum anticommutation relations, {i (x), j (y)}
= {i (x), j (y)} = ij (x y). The Hamiltonian density of three decoupled QI chains then
becomes
H0 (x) [iv E (x) x E(x) imt E (x) E(x)]
(21)
with v = 2J a0 and mt = 2( J ). A global chiral rotation of the Majorana spinors,

a + a
a + a
,
aL = ,
(22)
2
2
transforms (21) to the non-interacting triplet part of the Hamiltonian (2). On the other hand,
up to irrelevant corrections, HAT in (19) transforms to the marginal interaction terms in (2)
in (11) and (9) is self-evident.

with g = 8Ka0 . The correspondence between the h-terms
Thus, we have shown that, in the weak-coupling limit (20), the 1D quantum model (11)
can be regarded as a symmetry preserving lattice counterpart of the continuum theory (9).
General universality considerations allow us to expect that, if the field-theoretical model (9)
displays a certain quantum critical behavior, this should also be a property of the quantum
lattice model (11) even when its parameters are not restricted by the condition (20). It is
limit where the
then legitimate to consider the model (11) in the strong-coupling, large-h,
description of the SU(2)1 criticality greatly simplifies.
z
z
, s2,n
, nz , where
Let us pass to a new set of Ising variables, s1,n
aR =
z
z
= 1,n
,
s1,n
z
z
s2,n
= 2,n
,
z
z
z
nz = 1,n
2,n
3,n
.
(23)
In the original (1 , 2 , 3 ) representation, the local (at a given site n) Hilbert space of the
three-chain model is spanned on the basis vectors |1 , 2 , 3 i which are eigenstates of 1z ,
2z and 3z :
az |1 , 2 , 3 i = a |1 , 2 , 3 i,
a = 1
(a = 1, 2, 3).
The new local basis |s1 , s2 , i is defined as

saz |s1 , s2 , i = sa |s1 , s2 , i (a = 1, 2),
z |s1 , s2 , i = |s1 , s2 , i,
(a = 1, 2, 3)
where sa = a , = 1 2 3 . Comparing matrix elements of the operators a,n
in the two bases, we find the following correspondence:
678
z
z
1,n
= s1,n
,
z
z
2,n
= s2,n
,
x
x
= s1,n
nx ,
1,n
z
z
z
3,n
= s1,n
s2,n
nz ,
x
x
2,n
= s2,n
nx ,
x
3,n
= nx .
(24)
, s , , respectively,
, 2,n+1/2
, n+1/2
be the disorder operators dual to s1,n
Let 1,n+1/2
2,n n
and obeying the duality relations similar to (16). Under the change of basis, the original
dual spins a,n+1/2 (a = 1, 2, 3) transform as follows:
z
z
n+1/2
,
z1,n+1/2 = 1,n+1/2
z
z
z2,n+1/2 = 2,n+1/2
n+1/2
,
z
,
z3,n+1/2 = n+1/2
x
,
x1,n+1/2 = 1,n+1/2
x
x2,n+1/2 = 2,n+1/2
,
x
x
x
2,n+1/2
n+1/2
.
x3,n+1/2 = 1,n+1/2
(25)
It is easy to check that the new pairs of mutually dual operators, (s1 , 1 ), (s2 , 2 ) and (, ),
satisfy the same algebra, Eqs. (17), (18), as the original operators (a , a ) (a = 1, 2, 3).
In terms of the new variables, the Hamiltonian (11) reads:
X

(3)
z
z
z
z
z
z
z
z
s1,n+1
+ s2,n
s2,n+1
+ s1,n
s1,n+1
s2,n
s2,n+1
s1,n
HQI = (J + K)
K
n
x
s1,n
X
X

x
x x
x
x
+ s2,n
+ s1,n
s2,n + h
nz
+ s2,n
+ 1 nx
s1,n
J X z z
z 2
+
s s
sz sz
z n+1
2 n 1,n 1,n+1 2,n 2,n+1 n
z
KX z z
z
z
z 2
+
s2,n+1
.
s1,n s1,n+1 + s2,n
n n+1
2 n
(26)
(3)
The advantage of this representation of HQI becomes transparent in the strong-coupling

limit:
h J , , K.
(27)
In the zeroth-order approximation, the Hamiltonian (26) describes a collection of noninteracting -spins in a strong magnetic field, with a large energy gap 2h separating the
fully polarized ground state from the excited states with one spin flip. Thus the degrees of
freedom are fast and can therefore be integrated out to produce an effective Hamiltonian
for the low-energy, (s1 , s2 ) part of the spectrum.
(3)
Using a unitary transformation in the form of a 1/h expansion, we project HQI onto
P
the lowest-energy state ( n = 1, n) of the zeroth-order Hamiltonian, H0 = h n nz ,
and thus obtain the effective model of two coupled QI chains, Heff [s1 , s2 ]. Apparently, the
(3)
x
symmetry of Heff [s1 , s2 ] might appear only as Z2 Z2 , for HQI in (26) contains only s1,2
z
and s1,2 . The hidden SU(2) symmetry of the effective model is revealed by the nonlocal
duality transformation first introduced by Kohmoto, den Nijs and Kadanoff (KNK) in
their study of the 2D AT model [17], and later on employed by Kohmoto and Tasaki in
their analysis of the bond-alternating S = 1/2 chain [18]. This transformation, which is
briefly reviewed in Appendix A, establishes an important correspondence between two
679
combinations of the variables s1 , s2 ( = z, x) and local SU(2) invariants written in terms

of new spin-1/2 operators Tn :
z
z
z
z
z
z
z
z
s1,n+1
+ s2,n
s2,n+1
+ s1,n
s1,n+1
s2,n
s2,n+1
4T2,n T2n+1 ,
s1,n
x
s1,n
x
+ s2,n
x x
+ s1,n
s2,n
4T2n1 T2,n ,
(28)
(29)
where {2n} and {2n + 1} denote even and odd sublattices of a new lattice with 2N sites.
Notice that the duality transformation that maps the l.h. sides of Eqs. (28) and (29) onto
each other is equivalent to a translation by one lattice spacing on the lattice where the
spin-1/2 operators Tn are defined.
To the accuracy O(1/h 2 ), the low-energy effective Hamiltonian is found to be (see
Appendix B)
Heff [s1 , s2 ] =
N
X
(J1 T2,n T2n+1 + J2 T2n1 T2,n )
n=1
+ J3
N
X
(T2n1 T2n+1 + T2,n T2n+2 ),
(30)
n=1
where

2

(J + 2K) ,
J1 = 4 J + K
h

2

2
K.
(31)
,
J3 = 4
J2 = 4 K +
h
h
Heff [s1 , s2 ] is manifestly SU(2) invariant and describes a spin-1/2 chain with bondalternating nearest-neighbor interactions J1 and J2 and the next-nearest-neighbor interaction J3 . The SU(2)1 critical regime occurs when the bond alternation vanishes: J1 = J2 .
Notice that the condition /h 1 ensures that the frustrating interaction J3 remains irrelevant at the transition. Eqs. (30), (31) constitute the central result of our work.
It is worth stressing here that the equivalence between two coupled critical QI chains
and a bond-alternating SU(2)-symmetric S = 1/2 Heisenberg model necessarily implies
that the AT interchain interaction is of the order of the cutoff and fine tuned in such a
way that all three terms in the l.h. sides of Eqs. (28), (29) have the same amplitude. This
is a manifestation of the well-known fact, implicitly present in earlier studies of the 2D
AT model [17], that there exists no free-fermion Majorana representation of the critical
SU(2)1 WZNW model. At the point where the effective N = 2 AT model Heff [s1 , s2 ] is
self-dual, the S = 1/2 Heisenberg chain becomes translationally invariant and critical. In
the 2D N = 2 AT model, this is the point where the c = 1 line of Gaussian critical points
with continuously varying exponents bifurcates into two Ising critical lines [17,28].
In the lowest order, the critical value of h is given by h c = 2 /J . With /h being
the small parameter in the expansions (31), the ratio /J h c / 1, implying that
the c = 1 criticality is reached if the three QI chains in (11) are strongly disordered. This
is a direct consequence of the imposed condition (27) which prevents us to establish a
quantitative correspondence between the parameters of the field-theoretical models (7),
680
(9) and the lattice model (11), and, in particular, determine the critical line. However,
universality arguments allow us to expect that the asymptotic decoupling between the
-degrees of freedom, which become frozen at the SU(2)1 transition, and the (s1 , s2 )
degrees of freedom that describe an effective critical S = 1/2 Heisenberg chain is a general
This means that, even
property of the quantum AT model (11) with an arbitrary value of h.
when the condition (27) is released, the low-energy sector of the model (11) will still be
described by an effective Hamiltonian (30), although the determination of its parameters
will remain as a complicated, yet unresolved, part of the problem.
Despite this shortcoming, it is possible to extract the scaling of the critical line in the
weak-coupling limit from general considerations and compare the results with the already
existing ones. Let us go back to the model (9). At small h and m
t , the scaling law for the
t ) simply follows from the comparison of the mass gaps, h 8/13
critical line h c = h c (m
and m
and m
t , generated by the h t -perturbations independently [16]:
13/8
t ]13/8 J .
h c [m
(32)
1/8
For the standard ladder where |ms | ' 3|mt | J , one has h4z i J , and the relation
(32) translates to
1 13/8
3/2
(33)
hc 4z J J ,
in agreement with the result of Ref. [8].
Returning to the adopted mean-field treatment of the singlet Ising component, it is worth
noticing here that the description of the lowest-energy part of the spectrum in terms of
the effective spin-1/2 Hamiltonian (30) is valid not only at the SU(2)1 transition but also
at small deviations from criticality, provided that the associated (dimerization) mass gap
mdimer |h h c |2/3 is much smaller than the singlet mass |ms |.
4. UVIR transmutation of physical fields
Let us now find out how the physical fields, characterizing the spin ladder in the UV
limit, are transformed at the IR fixed point where the system becomes SU(2)1 critical.
We start with the current operators J1,2 which, according to Eqs. (3), are expressed in
terms of the Majorana fields E and 4 . Let us first consider the total current,
i
I = J1 + J2 = (ER ER + EL EL ),
2
which is fully determined in the triplet sector of the model and, in the UV limit, actually
represents the level-2 vector current of the SU(2)2 WZNW model given by the first term
in Eq. (9). This current is nothing but the smooth part of the magnetization of the related
critical S = 1 Heisenberg chain. The x-component of this current is

(34)
I x (x) = i R2 (x)R3 (x) + L2 (x)L3 (x) .
Making the chiral rotation (22), we can define a local lattice operator
i
Inx = (2,n 3,n + 2,n 3,n ) a0 I x (x),
2
(35)
681
which transforms back to (34) in the continuum limit. Using definitions (17), (18), together
with the change-of-basis transformations (24), (25), we obtain:

i2 3 z z
2,n 3,n z2,n+1/2 z3,n+1/2 z2,n1/2 z3,n1/2
2

i2 3 z z
z
z
2,n1/2
.
(36)
n s1,n 2,n+1/2
=
2
Since nz is a noncritical field, it can be replaced by its non-zero expectation value. Using
formulas (A.3) of Appendix A, we express Inx in terms of the spin operators of the effective
critical S = 1/2 Heisenberg chain (30):

x
x
+ T2n+1
.
(37)
Inx = i2 3 1 h z i T2n
Inx =
In the continuum limit

Tm (a0 /2)T(x),
T(x) = J(x) + (1)m n(x),
(38)
where J = JR + JL and n are the smooth and staggered parts of the local magnetization
T(x). In (38) we took into account the fact that, by the KNK construction (see
Appendix A), the lattice spacing of the effective S = 1/2 Heisenberg chain is a0 /2.
Keeping only relevant operators, we find that I x (x), being the level-2 current in the UV
limit, transforms to

(39)
I x (x) i2 3 1 h z i J x (x) (a0 /2)x nx (x) + ,
where stand for irrelevant corrections. Quite similarly we find that

I y (x) i3 1 2 h z i J y (x) (a0 /2)x ny (x) + ,

I z (x) J z (x) (a0 /2)x nz (x) + .
(40)
(41)
h z i = 1,
Since (1 2 3 )2 = 1, we can replace 1 2 3 by i. In the leading order in 1/h,
and we arrive at the following correspondence
I(x) J(x) Ca0 x n(x)
(42)
(C being a nonuniversal constant), which shows that the level-2 vector current I, originally
defined at the UV fixed point, transforms in the IR limit not only to the level-1 vector
current J but also acquires a nonholomorphic piece represented by the operator x n with
scaling dimension 3/2 and conformal spin 1.
One can similarly show that the level-2 axial vector current (i.e., the spin current of the
critical S = 1 chain), defined at the UV fixed point as
i
I5 = (ER ER EL EL ),
2
transforms to
I5 (x) J5 (x) + Ca0 t n(x),
(43)
where J5 = JR JL is the axial vector current of the SU(2)1 WZNW model (i.e., the spin
current of the S = 1/2 Heisenberg chains). Notice that the constant C in (43) is the same
682
as in (42). As a result, both at the UV and IR fixed points, the currents I and I5 satisfy the
continuity equation:
t I + x I5 = 0.
Since the holomorphic property of both level-2 currents is lost in the IR limit, the second
equation x I + t I5 = 0 is no longer valid.
According to Eq. (3), the relative current K = J1 J2 involves the singlet Majorana
field 4 which remains massive across the transition. Therefore, the current K represents
a short-ranged field whose correlations decay exponentionally over the length scale s
vs /|ms |.
Let us now turn to the staggered fields n and which are defined at the UV fixed point
by Eqs. (4)(6). Since n+ and + are proportional to 4 and since the forth Ising model is
ordered (ms < 0, h4 i = 0), these two fields remain short-ranged at the transition. On the
other hand, with 4 replaced by its nonzero expectation value, n and transform to the
staggered magnetization and dimerization field of the effective S = 1 chain
n (1 2 3 , 1 2 3 , 1 2 3 ),
t = 1 2 3 .
(44)
(45)
Adopting a symmetric lattice definition, for (n )x we obtain:

1
(n )xn = 1,n z2,n+1/2 z3,n+1/2 + z2,n1/2z3,n1/2 .
2
(46)
Under transformations (24), (25) and formulas (A.3), we find that (n )xn becomes

1 z
z
z
2,n+1/2
2,n+1/2
(n )xn = s1,n
2

1
x
x
T2n1
.
= 1 T2,n
2
Passing then to the continuum limit, with (38) taken into account, we obtain:
(n )x (x) 1 nx ,
(47)
where n is the staggered magnetization of the critical S = 1/2 chain. Similarly

(n )y (x) 2 ny ,
(n )z (x) i1 2 h z inx .
(48)
The Klein factors can be identified with Pauli matrices:

1 = x ,
2 = y ,
i1 2 = z .
Thus, in the leading order in 1/h (h z i = 1), we arrive at the following correspondence:
(n ) (x) n
( = x, y, z).
(49)
The presence of the Klein factors in (49) should not be confusing. These factors drop out
of the Hamiltonians (11) and (30) and have no dynamics. Moreover, due to the unbroken
SU(2) symmetry, hn (x, )n (0, 0)i = 13 hn(x, ) n(0, 0)i, the Klein factors drop out
of the correlation functions as well. Thus, at the SU(2)1 critical point, the relative staggered
683
magnetization of the spin ladder transforms to the staggered magnetization of the effective
S = 1/2 chain:
n n.
(50)
It may seem from formulas (39), (40) and (48) that taking into account 1/h corrections to
would break SU(2) invariance of the UVIR transmutation rules (42), (50). The
point is that the adopted minimal choice of local lattice operators (see Eqs. (35), (46)),
although being consistent with the continuum representation of the corresponding fields,
is not unique. Other lattice regularizations of the field operators may be equally good in
this respect but differ in the h-dependence

of the prefactors and subleading corrections at
the critical point. With the manifestly SU(2) invariant correspondence (42), (50) firmly
established in the leading order, we can only state here that these corrections are not
universal and, as already explained in the preceding section, cannot be addressed within
this approach.
Finally, we consider transmutation of the dimerization field . From (45) and (24) it
follows that
h z i = 1
z I.
(51)
However, couples to the amplitude h of the external dimerization whose variation leads
to the deviation from criticality. The leading SU(2)-symmetric, parity-breaking, relevant
perturbation at the SU(2)1 criticality is the dimerization field of the effective S = 1/2
Heisenberg chain: (1)n Tn Tn+1 (x). Therefore, the improved version of the formula
(51), which includes a strongly fluctuating correction to the identity operator I , should read
(x) I + (x).
(52)
Since the scaling dimension of is 1/2, the behavior of hi close to the critical point is
given by
|h i|h = |h i|h c + const |h h c |1/3 sign(h h c ).
(53)
Thus, the average dimerization remains finite at the transition but has an infinite slope.
Let us now draw a physical picture emerging from the obtained results. In the limit of a
translationally invariant ladder, J 0 /J 0, the lowest-energy part of the spin fluctuation
spectrum displays a single coherent S = 1 massive magnon formed due to confinement of
the originally massless S = 1/2 spinons of individual chains. In the opposite limit of two
decoupled bond-alternating spin-1/2 chains, J /J 0 0, coherent massive magnons are
still present in the spectrum but are of a different nature: these represent two independent
sets of triplet states formed by a soliton, antisoliton and the first breather of the two
corresponding 2 = 2 sine-Gordon models (see, e.g., Ref. [20], Chapter 22). The
crossover between these two extreme cases involves a critical point where the spectrum
is entirely incoherent and consists of pairs of S = 1/2 spinons of the effective S = 1/2
Heisenberg chain. (The behavior of the current-current correlation function hI(x, )I(0, 0)i
at the transition will be modified due to the admixture of the nonholomorphic operator in
Eq. (42).) As follows from (30), (31), close to the transition, the low-energy properties
684
of the system are those of a single, weakly dimerized spin-1/2 Heisenberg chain. (The
fact that the average relative dimerization of the original ladder, h i, stays finite across
the transition implies that higher-energy degrees of freedom which remain massive at the
critical point are also dimerized.)
The two massive phases separated by the c = 1 critical point differ in the sign of the
explicit dimerization of the effective S = 1/2 chain (30) and are characterized by two
different string order parameters [29,30], each of them being nonzero only in one phase
and vanishing in the other. For the spin-1/2 alternating Heisenberg chain, the nonlocal
string order parameter associated with the breakdown of a hidden Z2 Z2 symmetry was
first considered by Kohmoto and Tasaki [18] and Hida [31]. Its representation in terms
of the Ising order and disorder parameters for two-leg spin-1/2 ladders was introduced in
Refs. [13,14]. The two different string order parameters of the alternating chain (30) are
defined as follows (due to the unbroken SU(2) symmetry, it is sufficient to consider only
the x-components of the corresponding operators):
!
2n1
X
x
x
Tj ,
O2k,2n1 = exp i
j =2k
x
O2k+1,2n
2n
X
= exp i
Tjx
(54)
j =2k+1
Using relations (A.4) and (A.5) given in Appendix A, we find that

z z
x
= s1,k
s1,n ,
O2k,2n1
x
O2k+1,2n
= z2,k+1/2z2,n+1/2 ,
(55)
z
z
, z1,k+1/2) and (s2,k
, z2,k+1/2) are pairs of the order and disorder operators
where (s1,k
of two QI chains representing the effective spin-1/2 Heisenberg Hamiltonian (30) (see
Section 3). As follows from (31), close to the transition, J1 J2 h h c . Since at h < h c
the two QI chains are disordered, one finds that

x
O2k,2n1 = 0.
(56)
O2k+1,2n 6= 0,
At h > h c these chains are ordered implying that in this case

x
O2k,2n1 6= 0.
O2k+1,2n = 0,
(57)
As shown by Hida [31], at h h c 0, the two string order parameters vanish as

|h h c |1/6 .
5. Summary and discussion

In this paper, we have proposed a nonperturbative approach to describe the SU(2)1
criticality in the dimerized, weakly-coupled two-leg spin-1/2 ladder. We have shown
that this criticality is in fact a quantum critical point of the effective spin-1 Haldane
spin liquid perturbed by the explicit dimerization. Using the mapping onto a generalized,
SO(3)-symmetric, quantum AshkinTeller model and employing a nonlocal duality
685
transformation, we have derived the low-energy effective Hamiltonian which represents

a lattice S = 1/2 Heisenberg chain with a small bond alternation. The SU(2)1 critical
point corresponds to the case when fine tuning of the parameters of the model restores
translational invariance. With the adopted approach we were able to find an asymptotically
exact correspondence between the physical fields of the original spin ladder and those
characterizing the SU(2)1 criticality at the IR fixed point.
To avoid confusion, let us emphasize that the effective spin-1/2 Heisenberg model
(30) should not be misleadingly associated with the one that corresponds to a strongly
dimerized, snake looking two-leg ladder. The Hamiltonian Heff [s1 , s2 ] was derived in
the large-h limit from the generalized, SO(3)-symmetric, quantum AT model (11) which
should be regarded as a regularized lattice version of a continuum field theory (9), the
latter describing universal properties of a weakly dimerized spin-1 chain. No such fieldtheoretical description is available if the in-chain staggering amplitude J 0 of the original
spin ladder is not small.
The Ising model description of quantum critical points in two-leg spin-1/2 ladders can
be extended to a more interesting situation where all four Ising models associated with the
triplet and singlet Majorana modes (see Eqs. (2) and (7)) become equally important. This
turns out to be the case for a generalized spin ladder [14]
X
(S1,n S1,n+1 )(S2,n S2,n+1 ),
(58)
Hgen = Hstand + V
n
which, apart from the standard on-rung exchange J present in H , Eq. (1), also includes
a biquadratic interchain interaction. At J = 0 the model (58) is known as the spin-orbital
chain [32] and has been recently studied by different groups (see, e.g., [9,3336]). If
J 6= 0 but |V | is large enough, the model (58) occurs in a non-Haldane, spontaneously
dimerized phase where the spectrum is entirely incoherent and consists of pairs of
topological kinks [14]. When an external longitudinal dimerization of the chains is
included, the generalized ladder is expected to exhibit a pattern of quantum criticalities
presumably richer than that in the standard-ladder case discussed in Ref. [7] and in the
present paper. The underlying SO(3) Z2 symmetry of the model (58) opens a room
for criticalities with central charge c = 1/2, 1 and 3/2, corresponding to the universality
classes of the Ising model and SU(2)k WZNW models with k = 1 and 2. Recent numerical
results [37] strongly suggest the appearance of critical points due to the interplay between
the biquadratic interaction V and the explicit dimerization. This and related problems are
presently under investigation.
Acknowledgements
We are grateful to M. Fabrizio, A.O. Gogolin, G. Mussardo, V. Rittenberg, A.M. Tsvelik
and Yu Lu for fruitful discussions. A.N. is partly supported by the INTAS-Georgia grant
No. 97-1340.
686
Appendix A. KNK duality transformation

Consider a model of two noncritical QI chains with an AT-like interchain interaction
N
X X
HQI =
a=1,2 j =1
N
X
j =1
z
z
x
sa,j
Aa sa,j
+1 + Ba sa,j

z
z
z
z
x x
s1,j
A3 s1,j
+1 s2,j s2,j +1 + B3 s1,j s2,j .
(A.1)
The KNK transformation [17,18], which is a special combination of spin rotation and
duality transformation, provides a mapping of the model (A.1) onto a quantum spin-1/2
chain. Here we outline basic steps of this transformation:
(i) duality transformation in the second QI copy:
z
z
x
= 2,j
s2,j
1/2 2,j +1/2 ,
z
z
x
s2,j
s2,j
+1 = 2,j +1/2 ;
(ii) reduction of the original ({j }) and dual ({j + 1/2}) chains to a single chain via the
following identification:
2j ,
s1,j
2,j
+1/2 2j +1 ;
(iii) relabeling the lattice sites, j j 1/4, and treating the sublattice {2j + 1/2} as
dual to {2j }.
(iv) duality transformation:
nx = zn1/2 zn+1/2 ,
z
nz n+1
= xn+1/2 ;
(v) a staggered -rotation around the y-axis:

nz (1)n nz ,
nx (1)n nx ;
(vi) global /2-rotation around the x-axis:

y
n nz ,
nz n .
This brings the Hamiltonian (A.1) to its final form a bond-alternating XYZ S = 1/2
chain:
HXYZ = 4
N
X
X
j =1 =x,y,z
T2j +1 + J2 T2j
J1 T2j
1 T2j ,
(A.2)
where Ti = (1/2)i are the spin-1/2 operators, and

y
J1x = A1 ,
J1 = A2 ,
J1z = A3 ;
J2x = B2 ,
y
J2
J2z = B3 .
= B1 ,
The KNK transformation establishes a correspondence between the spin operators on

and T
the even and odd lattice sites, T2j
2j +1 , and products of two Ising order and disorder
operators of the two-chain model (A.1), (s1 , 1 ) and (s2 , 2 ):
z
z
x
2T2j
= 1 s1,j
2,j
+1/2 ,
687
z
z
x
2T2j
+1 = 1 s1,j +1 2,j +1/2 ,
z
z
2T2j = 2 1,j
+1/2 s2,j ,
z
z
2T2j +1 = 2 1,j
+1/2 s2,j +1 ,

z
z
z z
z
= i(1 2 ) s1,j
s2,j
1,j +1/2 2,j
2T2j
+1/2 ,
z

z
z
z
z
2T2j
+1 = i(1 2 ) s1,j +1 s2,j +1 1,j +1/2 2,j +1/2 .
(A.3)
z
and 1(2),j +1/2 (see Eqs. (17), (18)), the
Given the standard algebra of the operators s1(2),j
Klein factors 1 , 2 in (A.3) ensure the correct algebra of the Pauli matrices 2Tn .
From (A.3) it follows that
z
z
x x
T2j +1 = s1,j
s1,j
4T2j
+1 ,
z
z
4T2j T2j +1 = s2,j
s2,j
+1 ,
z z
z
z
z
z
T2j +1 = s1,j
s1,j
4T2j
+1 s2,j s2,j +1 ;
x
x
4T2j
1 T2j
z
z
4T2j
1 T2j
x
= s2,j
,
x x
= s1,j s2,j .
y
y
4T2j 1 T2j
(A.4)
x
= s1,j
,
(A.5)
Eqs. (A.4), (A.5) lead to the correspondence (29). Therefore, the hidden SU(2) symmetry
of the following model of two coupled QI chains
H =
N
X
x
x
x x
+ s2,j
+ s1,j
s2,j
B s1,j
j =1
z
z
z
z
z
z
z
z
s1,j
+A s1,j
+1 + s2,j s2,j +1 + s1,j s1,j +1 s2,j s2,j +1
(A.6)
is encoded in the special Ising structure of the A and B terms in the r.h. side of (A.6).
Appendix B. Strong-coupling expansion

In this appendix, we provide some details concerning the unitary transformation of the
original model (26) which makes it possible to derive the effective Hamiltonian Heff [s1 , s2 ]
in the form of 1/h expansion.
It is suitable to rearrange the Hamiltonian in (26) in the following way:
HQI = H + V ,
H = H0 + W,
where
H0 = h
W = W1 + W2 ,
V = 1
nz ,
W1 = (J + K)
X
n
W2 =
Qn (s)nx ,
n,n+1 (s) K
2
1X
z
Pn,n+1 (s) nz n+1
.
2 n
In Eqs. (B.2)
(B.1)
Rn (s),
(B.2)
688
x
x
Qn (s) = s1,n
+ s2,n
+ 1,
x
x
x x
+ s2,n
+ s1,n
s2,n ,
Rn (s) = s1,n
z
z
z
z
z
z
z
z
s1,n+1
+ s2,n
s2,n+1
+ s1,n
s1,n+1
s2,n
s2,n+1
,
n,n+1 (s) = s1,n

z
z
z
z
z
z
z
z
s1,n+1
s2,n
s2,n+1
+ K s1,n
s1,n+1
+ s2,n
s2,n+1
.
Pn,n+1 (s) = J s1,n
(B.3)
Suppose that the eigenvalue problem for H is solved:

H |ai = Ea |ai.
Since [H0 , W ] = 0, any state |ai can be represented as a direct product |ai = | i |si,
where |si only involve quantum numbers characterizing the (s1 , s2 ) part of the spectrum,
while
| i =
N
Y
|i i
with |i i = | ii , | ii being eigenstates of the operator iz . Under the condition (27), the
spectrum of H coincides in the leading order with that of H0 . The ground state of H0 is a
fully polarized state
|0i = | . . . i ,
(B.4)
while the lowest excited states involve one -spin flip

| . . . i ,
| . . . i ,
| . . . i ,
(B.5)
With the s-degrees of freedom taken into account (H0 H0 + V ), the

and have a gap 2h.
ground state and 1-flip excited states transform to narrow bands. Our goal is to eliminate
V in the lowest order and project the resulting Hamiltonian onto the state |0i .
The procedure is standard. Consider a unitary transformation
eQI = eS HQI eS
H
1
(B.6)
= H + V + [S, H ] + [S, V ] + [S, [S, H ]] + ,
2
where S = S . Requiring that V + [S, H ] = 0, one finds that, in the lowest order in 1/h,
the projected Hamiltonian

2 ,
eQI P0 = 2N h + W1 + 1 P0 [S, V ] P0 + O (1/h)
(B.7)
P0 H
2
where
P0 = |0i h0| .
The matrix element of the commutator in (B.7) is given by

X
1
1
Vac Vcb
+
ha|[S, V ]|bi =
Ec Ea Ec Eb
c
Z
=
0
d e2h
X
c

ecb () ,
eac ()Vcb + Vac V
V
(B.8)
689
where
e() = eW V eW .
V
In obtaining (B.8), we assumed that the states |ai and |bi are of the form |0i |si. We
also took into account the fact that the energy differences in (B.8) are necessarily positive
because the off-diagonal operator V connects |0i with 1-flip states
(of the order of 2h)
(B.5). Using 1/h expansion, we obtain:

1
1
1 2
V + 2 [[W, V ], V ] P0 + O 3 .
(B.9)
P0 [S, V ]P0 = P0
4h
h
h
Calculating the commutators in (B.9) and using formulas (A.4) and (A.5), one arrives at
the results (30), (31) given in the main text.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
A.B. Zamolodchikov, JETP Lett. 43 (1986) 730 [Pisma Zh. Eksp. Teor. Fiz. 43 (1986) 565].
M. Vojta, Y. Zhang, S. Sachdev, cond-mat/0003163.
I. Affleck, Nucl. Phys. B 265 (1986) 409.
I. Affleck, F.D.M. Haldane, Phys. Rev. B 36 (1987) 5291.
E. Dagotto, T.M. Rice, Science 271 (1996) 618.
M.A. Martin-Delgado, R. Shankar, G. Sierra, Phys. Rev. Lett. 77 (1996) 3443.
M.A. Martin-Delgado, J. Dukelsky, G. Sierra, Phys. Lett. A 250 (1998) 430.
D.C. Cabra, M.D. Grynberg, Phys. Rev. Lett. 82 (1999) 1768.
K. Totsuka, M. Suzuki, J. Phys.: Condens. Matter 7 (1995) 6079.
M. Fabrizio, A.O. Gogolin, A.A. Nersesyan, Nucl. Phys. B 580 (2000) 647.
M. Fabrizio, A.O. Gogolin, A.A. Nersesyan, Phys. Rev. Lett. 83 (1999) 2014.
D. Shelton, A.A. Nersesyan, A.M. Tsvelik, Phys. Rev. B 53 (1996) 8521.
A.A. Nersesyan, A.M. Tsvelik, Phys. Rev. Lett. 78 (1997) 3939.
K. Totsuka, Y. Nishiyama, N. Hatano, M. Suzuki, J. Phys. C: Condens. Matter 7 (1995) 4895.
A. Kitazawa, K. Nomura, Phys. Rev. B 59 (11) (1999) 358.
M. Kohmoto, M. den Nijs, L.P. Kadanoff, Phys. Rev. B 24 (1981) 5229.
M. Kohmoto, H. Tasaki, Phys. Rev. B 46 (1992) 3486.
F.D.M. Haldane, Phys. Lett. A 93 (1983) 464; Phys. Rev. Lett. 50 (1983) 1153.
A.O. Gogolin, A.A. Nersesyan, A.M. Tsvelik, Bosonization and Strongly Correlated Systems,
Cambridge Univ. Press, Cambridge, 1999.
A.B. Zamolodchikov, V.A. Fateev, Sov. J. Nucl. Phys. 43 (1986) 657 [Yad. Fiz. 43 (1986) 1031].
A.M. Tsvelik, Phys. Rev. B 42 (10) (1990) 499.
L. Takhtajan, Phys. Lett. A 87 (1982) 479.
H. Babujan, Phys. Lett. A 90 (1982) 479.
P. Pfeuty, Ann. Phys. (NY) 57 (1970) 79.
E. Lieb, T. Schultz, D. Mattis, Ann. Phys. (NY) 16 (1961) 407.
R. Shankar, Phys. Rev. Lett. 55 (1985) 453.
R.V. Ditzian, J.R. Banavar, G.S. Grest, L.P. Kadanoff, Phys. Rev. B 22 (1980) 2542.
M. den Nijs, K. Rommelse, Phys. Rev. B 40 (1989) 4709.
S.M. Girvin, D.P. Arovas, Phys. Scripta T 27 (1989) 156.
K. Hida, Phys. Rev. B 45 (1992) 2207.
K.I. Kugel, D.I. Khomskii, Sov. Phys. Usp. 25 (1982) 231 [Usp. Fiz. Nauk 136 (1982) 621].
690
[33]
[34]
[35]
[36]
[37]
S.K. Pati, R.R.P. Singh, D.I. Khomskii, Phys. Rev. Lett. 81 (1998) 5406.
P. Azaria, A.O. Gogolin, P. Lecheminant, A.A. Nersesyan, Phys. Rev. Lett. 83 (1999) 624.
A.K. Kolezhuk, H.J. Mikeska, Phys. Rev. Lett. 80 (1998) 2709.
C. Itoi, S. Qin, I. Affleck, Phys. Rev. B 61 (2000) 6747.
M.J. Martins, B. Nienhuis, cond-mat/0004238.

Heisenberg spin chains based on s`(2|1) symmetry

S. Derkachov a,c, , D. Karakhanyan b,1 , R. Kirschner c,2
a Department of Mathematics, St. Petersburg Technology Institute, Sankt Petersburg, Russia
b Yerevan Physics Institute, Br. Alikhanian st. 2, 375036, Yerevan, Armenia
c Naturwissenschaftlich-Theoretisches Zentrum und Institut fr Theoretische Physik, Universitt Leipzig,
Augustusplatz 10, D-04109 Leipzig, Germany

Abstract
We find the general rational solution of the YangBaxter equation with symmetry s`(2|1). The
R-matrix obtained acts on the tensor product of arbitrary representations of the superalgebra s`(2|1).
Based on this R-matrix we construct the local Hamiltonians for integrable homogeneous periodic
chains and open chains. 2000 Elsevier Science B.V. All rights reserved.
PACS: 03.65.Fd; 05.50.+q; 04.20.Jb; 71.20.Ad; 12.38.Cy
Keywords: Integrable spin chains; YangBaxter equation; Graded algebras; Bjorken limit
1. Introduction
The superalgebra s`(2|1) [1,2] appears in various quantum systems as underlying
their symmetry and dynamics. Finite-dimensional representations describe spin states. For
example, a lattice site which is allowed to be empty or occupied by an electron with the two
spin states 12 , but not by two electrons, corresponds to the three-dimensional fundamental
representation. Chains consisting of sites carrying the fundamental representation with
integrable short-range interaction have been constructed. The Hamiltonian of tJ model is
obtained from the transfer matrix of the integrable model based on the three-dimensional
fundamental representation of s`(2|1) [8,9]. The superalgebra s`(2|1) has a series of
four-dimensional representations parametrized by a parameter b 6= 12 (for b = 12 the
representations are not fully reducible). Integrable models built from R-matrices defined
on tensor products of two different four-dimensional representations have been considered
in [6,7,10].
sergey.derkachov@itp.uni-leipzig.de
1 karakhan@lx2.yerphi.am
2 roland.kirschner@itp.uni-leipzig.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 4 6 - 1
692
S. Derkachov et al. / Nuclear Physics B 583 [FS] (2000) 691720
The simplest integrable chain structure is the homogeneous periodic spin chain; most
applications make use of this case. There exist some important modifications.
The construction of open spin chains is well known [11,12]. The treatment of
integrable inhomogeneous spin chains is more involved. An integrable model has been
constructed [13] in view of the relevance for systems with impurities and in particular for
the Kondo effect. The representations of s`(2|1) accommodate s`(2) representations of
spin s and s 12 and allow in this way to construct chains with mobile impurities [14]. For
the future development and generalisation see [1519].
The integrable chains turn out to describe approximately the effective interaction in
four-dimensional gauge theories in the Regge limit [20,21] and in the Bjorken limit [22,
23]. Unlike the above examples here the sites carry infinite-dimensional representations
of s`(2) accommodating all the momentum states of reggeons and partons. Besides of the
case of homogeneous periodic chains also the case of open chains is encountered both in
the Bjorken limit [24,25] and in the Regge limit [26,27]. A particular s`(2|1) representation
of interest are the infinitesimal conformal transformations in one dimension together with
their twofold supersymmetric extensions. This symmetry applies to the Bjorken limit of
four-dimensional supersymmetric YangMills theory at least up to one loop [28,29]. This
means, the one-loop renormalization of quasipartonic composite operators can be obtained
by s`(2|1) symmetric pairwise interactions of partons, the states (light-cone momenta,
helicity, fermion number) of which form an infinite-dimensional lowest weight module of
this algebra.
In the present paper we consider the algebra s`(2|1), its lowest weight modules
and construct on this basis the solutions of the YangBaxter equation, i.e., the Rmatrix acting on the tensor product of two those modules. R-matrices acting on
tensor products of two fundamental s`(2|1)-representations and of two four-dimensional
representations have been constructed in the above mentioned papers [610]. General
integrable models based on R-matrices acting on tensor product of two arbitrary finitedimensional atypical(chiral) representations of s`(2|1) have been constructed quite
recently [30] by generalizing the known approach [31,32] from s`(2) to the case
of s`(2|1).
We propose the alternative approach and generalize these results. Motivated by possible
applications to the Bjorken limit in QCD, we represent the lowest weight modules by
polynomials in one even (z) and two odd (, ) variables. The general R-matrices are in fact
operators acting on two-point functions, i.e., (polynomial) functions of two sets (z1 , 1 , 1 )
and (z2 , 2 , 2 ) representing the tensor product. We construct the R-operators acting on the
tensor product of two arbitrary (finite or infinite-dimensional) s`(2|1)-modules. This is
done by calculating the two-point eigenfunctions of the lowest weight and the eigenvalues
of the R-operator.
From the particular result for arbitrary but isomorphic representations we derive the
integrable nearest neighbour interaction Hamiltonian for homogeneous periodic chains
with sites carrying arbitrary isomorphic representations.
In the case of s`(2) there exist integrable nearest neighbour interactions in open chains
homogeneous inside but with arbitrary different representations corresponding to the end
693
points [24]. We extend this result to the case of s`(2|1) and construct the corresponding
Hamiltonians.
The presentation is organized as follows. In Section 2 we introduce definitions and
summarize the standard facts about the superalgebra s`(2|1) and its representations.
We represent the lowest weight modules by polynomials in one even (z) and two odd
variables (, ) and the s`(2|1)-generators as first order differential operators.
In Section 3 we derive the defining relation for the general R-matrix, i.e., the solution of
the YangBaxter equation acting on tensor products of two arbitrary representations, the
elements of which are polynomial functions of (z1 , 1 , 1 ) and (z2 , 2 , 2 ). We solve this
defining relation in the space of lowest weights.
In Section 4 we construct the local integrable Hamiltonians in the simplest case of
homogeneous periodic chain carrying arbitrary isomorphic representations on the sites.
In Section 5 we construct the local integrable Hamiltonians for the open chain with
arbitrary isomorphic representations inside and other arbitrary representations at the end
points.
Finally, in Section 6 we summarize. Appendix A contains some technical details of
calculations. In Appendix B we give the expression for the R-matrix acting in the tensor
product of chiral modules. In Appendix C we discuss shortly the case of finite-dimensional
representations and show that the obtained general R-matrix is reduced to the known
ones [10] for the tensor product of modules with minimal dimensions.
2. Algebra s`(2|1)
2.1. Commutators and Casimir operators
The superalgebra s`(2|1) has eight generators: four odd V , W and four even ones
S, S and B. The commutation relations have the following form [1]:
anticommutators

V , V = 0,
W , W = 0,
V , W = S ,
commutators

S, S = S ,
B,S =
0,
S , V = 0,

B, V = 12 V ,

1
,
S,V =
2V
S , V = V ,

V , V = 0,
W , W = 0,
V , W = S B;

S + , S = 2S,
[B,
S] =0,

S , W = 0,

B, W = 12 W ,

1
,
S,W =
2W
S , W = W .
These generators are linear combinations of the generators EAB of the superalgebra
g`(2|1) [2]. The commutation relations for the nine generators of g`(2|1) can be written
compactly in the form:
694
[EAB , ECD ] = CB EAD ()(A+B)(C+D) AD ECB ,

where the graded commutator is defined as:
[EAB , ECD ] EAB ECD ()(A+B)(C+D) ECD EAB .

The indices A, B, C, D = 1, 2, 3 and we choose the grading: 1 = 3 = 0; 2 = 1. The
connection between both sets of generators is the following:
S = E31 ,
W = E21 ,
S + = E13 ,
W + = E23 ,
S = 12 E11 12 E33 ,
V = E32 ,
V + = E12 ,
B = 12 E11 E22 12 E33 .
(2.1)
In the fundamental representation all generators eAB of g`(2|1) are (3 3)-matrices and
the basis in the space of these matrices can be chosen in the standard way:
(eAB )CD = AC BD ,
eAB eCD = CB eAD .
(2.2)
In the fundamental representation the generators W , V , S , S, B have the form:
0 00
0 0 0
0 0 0
W = 1 0 0 ,
V = 0 0 0 ,
S = 0 0 0 ,
1 00
0 0 0
0 1 0
0 01
0 00
01 0
W+ = 0 0 1 ,
V + = 0 0 0 ,
S+ = 0 0 0 ,
0 00
0 00
00 0
1
2 0 0
2 0 0
B = 0 1 0 .
(2.3)
S = 0 0 0 ,
0 0 12
0 0 12
There exists a simple construction for the central elements of the enveloping algebra of
g`(2|1) [2]. The first step is the construction of covariant operators: suppose we have two
(i)
covariant operators VCD , i.e., operators which have the following commutation relations
with generators

(i)
(i)
(i)
()(A+B)(C+D) AD VCB
, i = 1, 2.
= CB VAD
EAB , VCD
P
(1) (2)
It is easy to check that operator VAB = C ()C VAC VCB is also covariant. This simple
observation allows to construct covariant operators using generators ECD as simplest
building blocks. The second step is the construction of a central element from the covariant
P
operator: for any covariant operator VCD the operator V = C VCC belongs to the center
of the algebra
[EAB , V ] = 0.
Repeating this construction we obtain central elements Kn , n = 1, 2, 3, . . . , [4,5] for the
enveloping algebra of g`(2|1):
K1 =
K2 =
EAA ,
K3 =
695
()B EAB EBA ,
AB
()
C
B+
EAB EBC ECA ,
...
(2.4)
ABC
The eight generators of algebra s`(2|1) may be introduced by defining:

X
EAA ,
EAB EAB AB ()B
(2.5)
E31 = S ,
E21 = W ,
E13 = S + ,
E23 = W + ,
E11 = B S,
E32 = V ,
E12 = V + ,
E22 = 2B,
E33 = B + S.
It can be verified that generators EAB satisfy the same commutation relations as EAB :
[EAB , ECD ] = CB EAD ()(A+B)(C+D) AD ECB .

There exists only one restriction for the Cartan generators: E11 + E22 + E33 = 0 and the
independent eight generators can be chosen in the form (2.1).
The center of the enveloping algebra of the algebra s`(2|1) is generated by Casimir
operators Cn , n = 2, 3, . . . [4,5]. As shown in [2], in any finite-dimensional lowest weight
irreducible representation, the operators Cn with n > 4 can be expressed in terms of two
operators C2 and C3 :
1X
()B EAB EBA = S 2 B 2 + S + S + V + W + W + V ,

(2.6)
C2 =
2!
AB
1 X

()B+C EAB EBC ECA
C3 =
3!
ABC

3
= B S 2 B 2 + BS + S + B V + W + W + V
2

1
1
+ +
+ +
+ W V V W S + S+ V W W V
4
4

1
+
+
(2.7)
+ (S 1) V W W V .
2
It should be noted, however, that as operators, none of the higher Casimirs Cn can be
expressed in terms of C2 and C3 [4,5].
2.2. Global form of superconformal transformations
We represent the generators as first order differential operators, acting on the space of
polynomials (z, , ). Lowering (decreasing the polynomial degree) operators have the
form
1
1
(2.8)
V = + ,
W = +
S = ,
2
2
and generate the following global transformations
696
= (z ; , ),
eS (z; , )

= z + ; + , ,
eV (z; , )
2

= z + ; , + .
eW (z; , )
2
(2.9)
Rising (increasing the polynomial degree) operators

1
1
+
V = z + z + (` b),
2
2

1
1
+
W = z + z + (` + b),
2
2
+
2
S = z + z + z + 2`z b ,
(2.10)
generate the global transformations

b

z
1
+
= 1+
;
,
,
eS (z; , )
(1 z)
1 z 1 z 1 z
(1 z)2`

z
1
z
+
,
=
;
,
eV (z; , )
(1 + )`b
1 + /2 1 + /2

z
1
z
W +
; ,
.
(z; , ) =
e
(1 + )`+b
1 + /2
1 + /2
Two remaining elements of the Cartan subalgebra:
1
1
S = z + + + `,
2
2
generate the transformations:
1
1
B =
+ b
2 2

= e` e z; e/2, e/2 ,
eS (z; , )

= eb z; e/2, e/2 .
eB (z; , )
We use a natural convention here and assign scaling dimension one and U (1)-charge zero
to the even variable z and the scaling dimension 12 and U (1)-charge 12 to the odd variables
and .
2.3. s`(2|1)-lowest weight modules
The lowest weight s`(2|1)-module V`,b = VÈ, È = (`, b) is built on the lowest weight
vector obeying:
V = 0,
W = 0,
S = 0,
S = `,
B = b.
In generic situation ` 6= b the module is characterised uniquely by the action of the

Casimir operators on its elements:

C3 v = b `2 b2 v, v V`,b .
C2 v = `2 b2 v,
697
It is a vector space spanned by the following basis [1,3] with the even vectors
k
Bk = (S + )k1 W + V + , k Z+
Ak = S + ,
and the odd vectors
Vk = (S + )k V + ,
Wk = (S + )k W + ,
k Z+
We shall use the above realization of the s`(2|1)-generators as the differential operators
of first order acting on the infinite-dimensional (for generic `) space V`,b of polynomi of variables z, , .
als (z, , )
+
+
It is easy to obtain the expression for the coherent states eS , eS W + V + and
+
+
eS V + , eS W + for the lowest weight = 1 using the formulae of the previous
section. They are the generating functions, the power expansion in of which produces
the basis:

kb
Ak = (2`)k zk zk1 ,
2`

`b
k k1
(2`)k zk + b + ` +
z ,
Bk =
2`
2
k
Wk = (` + b)(2` + 1)k zk ,
Vk = (` b)(2` + 1)k z ,
0(2` + k)
.
(2.11)
(2`)k
0(2`)
It is useful to introduce the subspaces of functions with definite chirality. Let us define for
this purpose two operators called supercovariant derivatives:
1
D + = + ,
2
1
D = +
2
(2.12)
and two subspaces V`,b

ker D V`,b :
+
= (z+ , + ),
(z, , )
(z, , ) V`,b
z+ z + 12 ,
+ ,
= (z , ),
(z, , )
(z, , ) V`,b
z z 12 ,
are not s`(2|1)-invariant ones. Indeed, the

In the generic case the chiral subspaces V`,b
operators D have the following commutation relations with s`(2|1)-generators:
{D , V } = 0,
{D , W } = 0,
[D , S] = 12 D ,
{D + , V + } = D + ,
[D , S ] = 0,
[D , B] = 12 D ,
{D + , W + } = ` + b,
[D + , S + ] = (` + b) + z+ D + ,
{D , V + } = ` b,
{D , W + } = D ,
[D , S + ] = (` b) + z D ,
(2.13)
are s`(2|1)-invariant only under the condition

and it is easy to see that chiral subspaces V`,b
:
` = b. In this case the whole module V`,` has definite chirality V`,` = V`,`
D v = 0,
v V`,` ,
D + v = 0,
v V`,` .
698
The notions of chirality and chiral representations are used here as they are common in
supersymmetric field theory. In the mathematical literature about superalgebra representations generic representations are called typical and chiral representations are called atypical [1,3]. There exist some special values of `: ` = n; n 12 Z+ , for which the module
V`,b becomes a finite-dimensional vector space [1,3]. Indeed it is evident from (2.11) that
all basis vectors are equal zero for k > n + 1. There are three cases depending on the
relation between b and n. The first case is for generic b (typical representations):
dim Vn,b = 8n;
b 6= n,
k
,
k = z
k = 1, . . . , 2n 1,
k
,
k = z
2n

b
2n = z +
;
2n
0 = 1,
k = 0, . . . , 2n 1.
The second and third cases appear for b = n and here the representation spaces have
definite chirality (atypical representations):
,
Vn,n = Vn,n
b = n,
k
k
= z
,
dim Vn,n = 4n + 1;
k
k = 0, . . . , 2n;
k
= z
,
k = 0, . . . , 2n 1.
Let us introduce the special notation for the fundamental s`(2|1)-module:

V
1 1
2 , 2
VfE,
fE = ( 12 , 12 ).
In the basis

1
e1 = z 0 ,
0

0
e2 = 1 ,
0

0
e3 = 1 0
1
the s`(2|1)-generators take their fundamental form (2.3).

2.4. Tensor products of two s`(2|1)-modules
The tensor product of two s`(2|1)-modules has the following direct sum decomposition [3]:
V`1 ,b1 V`2 ,b2 = V`,b + 2
V`+n,b +
n=1
ì 6= bi ,
` = `1 + `2 ,
X
n=0
1
1
`+n+ 2 ,b 2
n=0
1,
1
`+n+ 2 ,b+ 2
b = b1 + b2 .
(2.14)
Note that this formula is applicable in the generic situation ì 6= bi . The direct sum
decomposition reduces for the tensor product involving chiral modules:
V`1 ,`1 V`2 ,b2 =
V`1 ,`1 V`2 ,`2 =
X
n=0
X
n=0
V`+n,b +
X
n=0
V`+n,b ,
1,
1
`+n+ 2 ,b 2
`2 6= b2 ,
V`1 ,`1 V`2 ,`2 =
X
n=0
1.
1
`+n+ 2 ,b 2
699
In Appendix C we discuss the modifications of (2.14) arising for finite-dimensional

representations [3].
For the proof of (2.14) one has to determine all possible lowest weight vectors appearing
in the tensor product V`1 ,b1 V`2 ,b2 . In the realization on functions of z, , the space
V`1 ,b1 V`2 ,b2 is isomorphic to the space of polynomials in two even variables z1 , z2
and four odd variables 1 , 1 , 2 , 2 called for the sake of brevity two-point functions.
The s`(2|1)-generators acting on the V`1 ,b1 V`2 ,b2 are just the sums of corresponding
generators acting in Vì ,bi . The lowest weight vectors of the irreducible representations in
the decomposition of V`1 ,b1 V`2 ,b2 are defined as the common solutions of the equations:
S = V = W = 0
(2.15)
which have the form

(z1 , z2 ; 1, 2 ; 1, 2 ) = Z12 , 12 , 12 ,
(2.16)
where
1
12 1 2 ,
12 1 2 .
Z12 z1 z2 + (1 2 + 1 2 ),
2
Indeed, from (2.15) follows immediately that the function has to be invariant with
respect to global transformations (2.9). This invariance predicts the general form of :
eW eV eaS (z1 , z2 ; 1, 2 ; 1 , 2 )

(1 + ) 1
(2 + ) 2
+
, z2 a +
+
;
= z1 a +
2
2
2
2

1 + , 2 + ; 1 + , 2 +
and choosing a = z2 + 12 2 2 , = 2 , = 2 we obtain (2.16).

There are additional restrictions of definite chirality for the lowest weights in the tensor
product of the chiral modules:
1 2 ,
, 12
D1 = 0 = Z12 12 12 12 , 12

D = 0 = Z12 1 12 12 , .
2
12
Now, the lowest weight vectors in the decomposition of the tensor product are constructed
from functions (2.16) being eigenfunctions of generators S and B. The eigenfunctions of
the operator S are the polynomials with scaling dimension n and the eigenfunctions of the
operator B are the polynomials with one of the possible U (1)-charges: 0, 12 .
Finally we obtain that all lowest weights in the space V`1 ,b1 V`2 ,b2 are divided on two
sets, the even lowest weights:
n
D1 n = 0,
n Z12 12 12 12 ,
Sn = (n + `)n ,
and the odd lowest weights:
Bn = bn
(2.17)
700

n
n 12 Z12
,
n
n+ 12 Z12
,
Sn = (n + ` + 12 )n ,
Bn = (b 12 )n
(2.18)
It is convenient to choose the chiral basis D1 n = 0 for the even lowest weights.
Thus we have obtained the full set of lowest weights appearing in the expansion of
V`1 ,b1 V`2 ,b2 and this proves the direct sum decomposition in generic situation (2.14).
The modifications in the case of the tensor product of chiral modules are evident: the
lowest weights in the tensor product of the chiral modules are obtained by imposing the
chirality restrictions. All lowest weights in the space V`1 ,`1 V`2 ,b2 , `2 6= b2 are n
and n and the lowest weights in the space V`1 ,`1 V`2 ,`2 are n . All lowest weights
in the space V`1 ,`1 V`2 ,`2 are 12(z1 z2 )n = n and all lowest weights in the space
V`1 ,`1 V`2 ,`2 are 12 (z1+ z2+ )n = n+ .
3. YangBaxter equation and general operator RÈ È (u)
1 2
Let Vì ,bi , i = 1, 2, 3 be three lowest weight s`(2|1)-modules. We shall use the shorthand notation:
È = (`, b),
VÈ = V`,b .
Let us consider the three operators RÈi Èj (u) which are acting in VÈi VÈj and obey the
YangBaxter equation in the space VÈ1 VÈ2 VÈ3 [33]:
RÈ1 È2 (u v)RÈ1 È3 (u)RÈ2 È3 (v) = RÈ2 È3 (v)RÈ1 È3 (u)RÈ1 È2 (u v).
(3.1)
We are going to find the general solution RÈi Èj (u) of YangBaxter equation by the
following three steps.
First one obtains the operator RfE,fE(u) in the simplest situation: Èi = fE for all i = 1, 2, 3
so that the space VfE = V 1 1 has the minimal possible dimension: dim VfE = 3. In the
2 , 2
E In the
second step we fix Èi = fE for i = 1, 2 and obtain the solution RfEÈ for arbitrary `.
third step we fix È3 = fE and using the result for the operator R E E we obtain and solve the
f `1
defining equation for the general R-matrix RÈi Èj (u). It should be noted that the analogous
approach was used for the derivation of the s`(2)-invariant R-matrix [31,32].
3.1. Fundamental solution RfE,fE
First one considers the simplest situation: ì = 12 , bi = 12 Èi = fEi . We shall prove
that the operator:
X
i
j
()B eAB
eBA ,
(3.2)
RfEi fEj (u) = u + Pij , Pij
AB
i
eAB
are generators acting in the space VfEi , is the solution of the YangBaxter
where
equation [33]:
RfE1 fE2 (u v)RfE1 fE3 (u)RfE2 fE3 (v) = RfE2 fE3 (v)RfE1 fE3 (u)RfE1 fE2 (u v).
701
Indeed, the proof is the following. The YangBaxter equation has the simple form in short
notations:
(u v + P12 )(u + P13 )(v + P23 ) = (v + P23 )(u + P13 )(u v + P12 ).
Comparing operator coefficients of uk on both sides of this equation yields:
u0 : P12 P13 P23 = P23 P13 P12 ,
(3.3)
u : P13 P23 + P12 P23 = P23 P12 + P23 P13 .
(3.4)
Using (2.2) one can prove that the operator Pij is the permutation:
j
i
= eAB Pij Pij Pj k = Pik Pij
Pij eAB
and this commutation relation for Pij allows to check that Eqs. (3.3), (3.4) hold and this
proves that RfEfE (3.2) obeys the YangBaxter equation.
3.2. The solution for the operator RfE,È(u)
E The operator:
We fix Èi = fEi for i = 1, 2 and obtain the solution RfEÈ for arbitrary `.
X
()B eAB EBA ,

(3.5)
RfEÈ(u) = u +
AB
E is the solution of the YangBaxter

where EAB are generators in arbitrary representation `,
equation:
RfE1 fE2 (u v)RfE1 È3 (u)RfE2 È3 (v) = RfE2 È3 (v)RfE1 È3 (u)RfE1 fE2 (u v).
The proof is the following [34,35] . The YangBaxter equation has the simple form in
short-hand notations:
(u v + P12 )(u + e 1 E)(v + 1 e E)
= (v + 1 e E)(u + e 1 E)(u v + P12 ).
Matching operator coefficients of uk on both sides of this equality yields:
u0 : P12 (e 1 E)(1 e E) = (1 e E)(e 1 E)P12 ,
u1 : (e 1 E)(1 e E) + P12 (1 e E)
= (1 e E)P12 + (1 e E)(e 1 E).
The first equation is a simple consequence of the properties of P12 . Using the fact that the
matrices eAB form a basis we obtain that the second equation is equivalent to the following
system of equations:
EAB ECD ()(A+B)(C+D) ECD EAB = CB EAD ()(A+B)(C+D) AD ECB .

This is nothing else but the commutation relations for generators EAB .
702
Let us represent the operator RfEÈ(u) (3.5) in the matrix form in the standard basis with
3 = 0; 2 = 1:
the grading 1,

1
0
0
e2 = 1 ,
e3 = 0 .
e1 = 0 ,
0
0
1
We shall use the following definition of the matrix of an operator:
F eA = eB FBA ,
(3.6)
which leads to the validity of the common rule for the matrix product, i.e., without
additional sign factors,
F GeA = eB (F G)BA ,
(F G)BA = FBC GCA .
Let us calculate the matrix of R-operator using the definitions (3.6) and (2.2):
(eAB EBA )eC = ()(B+A)C eAB eC EBA = ()(B+A)C eD (eAB )DC EBA .
Therefore the matrix element of operator eAB EBA has the form
X

()B (eAB EBA )CD = ()D C EDC .
AB
Finally one obtains
u + E11 E21
E31
()B eAB EBA = E12 u E22 E32 .

RfEÈ(u) = u +
AB
E13
E23 u + E33
X
This is the expression for the g`(2|1)-invariant R-matrix. The s`(2|1)-invariant R-matrix
can be derived from this result in a simple way: the operator K1 = E11 + E22 + E33 belongs
to the center of the algebra and therefore the R-operator RfEÈ(u K1 ) is also a solution
of the YangBaxter equation. Using the connection between EAB and s`(2|1)-generators
we obtain the s`(2|1)-invariant R-matrix [14]
S
u + (S + B) W
.
V +
u + 2B
V
RfEÈ(u K1 ) =
+
+
S
W
u + (B S)
3.3. General R-matrix RÈ1 È2 (u)
To obtain the defining relation for the general R-operator we consider the special case
È3 = fE in (3.1). Then one can choose the above matrix realization in VÈ3 and the operators
RÈ1 fE,RÈ2 fE are linear functions of spectral parameter u in this particular case
RÈi fE(u K1 ) = u + Ri ,
Si + Bi Wi Si
2Bi
Vi ,
Ri = Vi+
+
+
Si
Wi Bi Si
i = 1, 2.
703
Now the general R-matrix RÈ1 È2 (u) acting in the tensor product VÈ1 VÈ2 of arbitrary
modules, is fixed by the condition
RÈ1 È2 (u v)RÈ1 fE(u K1 )RÈ2 fE(v K1 )
= RÈ2 fE(v K1 )RÈ1 fE(u K1 )RÈ1 È2 (u v)
or equivalently:
(3.7)

uv u + v
uv
(R1 + R2 ) +
(R2 R1 ) + R1 R2
RÈ1 È2 (u v) 2 +
2
2

vu
uv v + u
(R2 + R1 ) +
(R1 R2 ) + R2 R1 RÈ1 È2 (u v).
+
=
2
2
2
After separation of u + v and u v dependence we obtain two equations(u v u):

RÈ1 È2 (u)(R1 + R2 ) = (R1 + R2 )RÈ1 È2 (u),

u
u
(R2 R1 ) + R1 R2 =
(R2 R1 ) + R2 R1 RÈ1 È2 (u).
RÈ1 È2 (u)
2
2
(3.8)
(3.9)
The first equation (3.8) expresses the fact that R(u) has to be invariant with respect to the
action of s`(2|1)-algebra and the second equation is the wanted defining relation for the
operator RÈ1 È2 (u).
The s`(2|1)-invariance of the operator RÈ1 È2 (u) allows to simplify the problem. Due to
s`(2|1)-invariance any eigenspace of the operator RÈ1 È2 is a lowest weight s`(2|1)-module
generated by some lowest weight eigenvector. Therefore without loss of generality we can
solve the defining relation (3.9) in the space of lowest weights. Let us consider in more
details the structure of eigenspace of the s`(2|1)-invariant operator acting on the tensor
product V`1 ,b1 V`2 ,b2 . As we have seen from direct sum decomposition:
V`1 ,b1 V`2 ,b2 = V`,b + 2
X
n=1
V`+n,b +
X
n=0
1
1
`+n+ 2 ,b 2
X
n=0
1
1
`+n+ 2 ,b+ 2
for every fixed n the space of lowest weight vectors with eigenvalue b is two-dimensional
and the ones with eigenvalues b 12 are one-dimensional. Therefore the operator RÈ1 È2
is diagonal on odd lowest weight vectors n+ and n but acts non-trivially on the twodimensional subspace of even lowest weight vectors spanned on n+ and n .
In matrix form we have:
+
+
n
An (u) Bn (u) 0
n
0
n Cn (u) Dn (u) 0
0 n

.
(3.10)
RÈ1 È2 (u)
+ = 0
0 Fn (u) 0 n+
n
n
n
0
0
0 En (u)
The matrix relation (3.9) leads to a set of recurrence relations for the coefficients
An , . . . , En . Some details of calculations can be found in appendix and here we present
the final expression for the general solution of these recurrence relations:
An (u) = (1)n+1
u + b1 b2
0 (u + `n + 1)
,
0 (u + `n ) (`1 b1 )(`2 + b2 )
704
0 (u + `n + 1) (`1 + b1 )(`2 b2 )
,
0 (u + `n + 1) (`1 b1 )(`2 + b2 )
0(u + `n )
,
Cn (u) = (1)n
0(u + `n )
0(u + `n )
Dn (u) = (1)n+1
0(u + `n + 1)
(`2 b2 )(`2 + b2 )(u b1 b2 ) (u + b1 + b2 )(u b2 `1 )(u b2 + `1 )
,
(`1 b1 )(`2 + b2 )
0(u + `n + 1) (u + b1 `2 )(u + b1 + `2 )
,
En (u) = (1)n
0(u + `n + 1)
(`1 b1 )(`2 + b2 )
0(u + `n + 1) (u b2 `1 )(u b2 + `1 )
,
(3.11)
Fn (u) = (1)n
0(u + `n + 1)
(`1 b1 )(`2 + b2 )
Bn (u) = (1)n
where we used the notations:

` n n + `1 + `2 ,
u
+ b1 b2 .
As usual the obtained general solution of the YangBaxter equation is fixed up to overall
normalization. We choose the normalization such that the R-matrix coincides with the
permutation operator for u = 0 and È1 = È2 . Note that the transformation from our basis
vectors n to the ones diagonalizing the R-matrix depends on n and this does not result
in a simpler formulae.
The obtained R-matrix (3.10) acts on the space of two-point functions which are
polynomials in zi , i , i , i = 1, 2. This holds also if the representation parameters
ì , bi , i = 1, 2 correspond to chiral or antichiral cases. If one or both modules in the
tensor product are chiral or antichiral then the tensor product representation space is a
proper subspace of the space of all two-point polynomials (compare (2.14)). It is important
to observe that in these cases the action of R-matrix can be consistently restricted to the
corresponding subspace. Indeed, after multiplying with the overall factor (`1 b1 )(`2 +
b2 ), the matrix becomes triangular in these cases in the way as expected. We list the reduced
R-matrices in all special cases involving chiral representations in Appendix B.
Note that we have calculated the matrix elements of R-matrix for the lowest weights
only. The action of the R-operator on an arbitrary vector can be calculated as follows. The
arbitrary vector V in the space V`1 ,b1 V`2 ,b2 can be represented in the form:
X

1 +
2
3 +
4
n + Vkn
n + Vkn
n + Vkn
n ,
Vkn
(3.12)
V=
k,n
i are some polynomials with respect to rising operators S + , V + , W + .
where operators Vkn
The R-matrix commutes with any s`(2|1)-generator and Eq. (3.10) shows the action of the
R-matrix on the lowest weights. Therefore we obtain for the arbitrary vector (3.12)
X

1
2
Vkn
An (u)n+ + Bn (u)n + Vkn
Cn (u)n+ + Dn (u)n
RÈ1 È2 (u)V =
k,n

3
4
Fn (u)n+ + Vkn
En (u)n .
+ Vkn
705
For calculating arbitrary matrix elements of the R-matrix it is sufficient to know the
s`(2|1)-invariance of R-matrix and the matrix elements for the lowest weights.
4. Homogeneous periodic chain

4.1. Commuting transfer matrices
Let us construct the set of commuting s`(2|1)-invariant operators the generating function
of which is the transfer-matrix TmE (u). We construct TmE (u) as the supertrace of a
monodromy matrix built of the elementary R-matrix blocks [34,35].
We introduce the N spaces VÈi and N operators Rm,
E Èi (u):
Rm,
E VÈi 7 Vm
E VÈi .
E Èi (u) : Vm
The periodicity convention N + 1 1 is implied. The monodromy matrix
RmE (u) Rm,
E È1 (u c1 ) Rm,
E ÈN (u cN )
(4.1)
acts then on the space VmE VÈ1 VÈN , and TmE (u) is obtained by taking the supertrace
in the auxiliary space VmE :
TmE (u) = strVmE RmE (u).
These monodromy matrices form the commutative family:
TmE 1 (u)TmE 2 (v) = TmE 2 (v)TmE 1 (u).
(4.2)
The relation (4.2) follows from the fact that there exists the operator RmE 1 ,mE 2 such that
RmE 1 ,mE 2 (u v)RmE 1 ,Èi (u)RmE 2 ,Èi (v) = RmE 2 ,Èi (v)RmE 1 ,Èi (u)RmE 1 ,mE 2 (u v).
The R-operator is even. Therefore by using standard arguments one derives an analogous
equation for the monodromy matrices:
RmE 1 ,mE 2 (u v)RmE 1 (u)RmE 2 (v) = RmE 2 (v)RmE 1 (u)RmE 1 ,mE 2 (u v).
(4.3)
From this one can derive easily that corresponding traces and supertraces are commuting
operators separately:
T+
E (u),
m
E (u) = trVmE Rm
T
E (u),
m
E (u) = strVmE Rm
+
+
+
T+
m
E 1 (u)Tm
E 2 (v) = Tm
E 2 (v)Tm
E 1 (u),
T
m
E (u)Tm
E (v) = Tm
E (v)Tm
E (u)
1
TmE (u) T
m
E (u)
is the generating function for the s`(2|1)-invariant operators.

but only
Instead of giving the general proof we demonstrate all this on the example of TfE(u),
where the auxiliary space corresponds to the fundamental representation. Let us represent
TfE(u) in the form:
TmE (u) = eAB TAB (u),
where operators TAB (u) act in tensor product VÈ1 VÈN and we assume the
summation over repeated indices. The general equation (4.3) has the form in this case
706
j i
j
TAB (u) eCD TCD (v)
u v + ()G eFi G eGF eAB
j
j
i
TAB (u) u v + ()G eFi G eGF .
= eCD TCD (v) eAB
The traces and supertraces of generators eCD are calculated as follows:

tr eAB (eAB )CC = AB ,
str eAB ()C (eAB )CC = ()A AB .
Using eAB eCD = CB eAD and taking (super-)traces in corresponding spaces one easily
obtains:
tr : TAA (u)TCC (v) = TCC (v)TAA (u),
str : ()A TAA (u)()C TCC (v) = ()C TCC (v)()A TAA (u).
The g`(2|1)-invariance can be demonstrated in the simplest example N = 2. The
generalization to arbitrary n is straightforward:

1
2
u + ()D eCD EDC
T+E (u) = tr u + ()B eAB EBA
f
1
2
1
2
+ EAA
EBA
,
+ 2 EAB
= 3u2 + u()A EAA

B
D
1
2
T E (u) = str u + () eAB EBA u + () eCD EDC
f

1
1
2
2
+ EAA
EBA
.
+ 2 ()B EAB
= u2 + u EAA
We have seen ((2.4) and discussion before) that only operators entering TE (u) are g`(2|1)f
invariant.
4.2. Local Hamiltonians
If one fixes the arbitrary representation m
E = È and the same representations È1 = =
E
E
`N = ` in remaining spaces VÈi we obtain the generating functions of Hamiltonians.
The local Hamiltonians can be obtained for the homogeneous (ci = 0) chain. The whole
construction is quite general and well known [36,37] and it needs the existence of regular
point for the R-matrix.
It is easy to see from the derived expression for the R-matrix that by condition Èi = Èj =
E
` the point u = 0 is regular:
RÈi ,Èj (0) = PÈi ,Èj ,
(4.4)
where P is permutation operator.

Indeed, the permutation operator P acts as follows on the lowest weight basis:
+
+
0 1 0 0
n
n
n
0 n
n1 0 0
P
+ = (1) 0 0 1 0 + .
n
n
n
n
0 0 0 1
The expression for matrix coefficients of R-operator takes the simple form in homogeneous
case (`1 = `2 , b1 = b2 ):
707
u
0(u + `n + 1)
,
0(u + `n ) (` + b)(` b)
0(u + `n + 1)
,
Bn (u) = (1)n
0(u + `n + 1)
0(u + `n )
,
Cn (u) = (1)n
0(u + `n )
u(u2 2`2 2b2 )
0(u + `n )
,
Dn (u) = (1)n+1
0(u + `n + 1)
(` + b)(` b)
0(u + `n + 1) (u + b `)(u + b + `)
,
En (u) = (1)n
0(u + `n + 1)
(` + b)(` b)
0(u + `n + 1) (u b `)(u b + `)
,
Fn (u) = (1)n
0(u + `n + 1)
(` + b)(` b)
An (u) = (1)n+1
(4.5)
where
`1 = `2 = `,
b1 = b2 = b,
u
,
`n n + 2`.
The equality (4.4) can be easily checked.

Performing the standard calculation [36,37] we obtain the following expression for
logarithmic derivative of the operator TÈ(u):
X
0
T1
E (0)TÈ(0) =
PÈi+1 ,Èi R0È
i+1 ,Èi
R0È ,È
i
i+1
(0)
(0) PÈi ,Èi+1 =
HÈi ,Èi+1 .
(4.6)
The resulting operator can be chosen as the Hamiltonian. It commutes with the integrals
of motions generated by TÈ(u) and is a sum of operators acting on two adjacent sites only.
The two-particle Hamiltonians in the sum are
HÈi ,Èi+1 = R0È ,È
i
i+1
(0) PÈi ,Èi+1
and have the following matrix elements:
2(`n + 1)
2 `2 + b2
`n (` b)(` + b)
HÈi ,Èi+1 =1
`n
(` b)(` + b)
2(`n )
0
0
2(`n + 1) +
2b
(` b)(` + b)
0
0
2(`n + 1)
2b
(` b)(` + b)
(4.7)
The eigenvalues of this matrix and corresponding eigenvalues can be easily calculated.
708
5. Inhomogeneous open chain

5.1. Integrals of motion
In this section we shall consider the open spin chain [12]. To start with we introduce the
operator RÈ1 ,È2 (u):
RÈ1 ,È2 (u) RÈ1 ,È2 (u) R1
E E (u).
` 1 ,` 2
It is possible to derive the following commutation relation for R(u):

RmE 1 ,mE 2 (u v) RmE 1 ,Èi (u) RmE 1 ,mE 2 (u + v) RmE 2 ,Èi (v)
= RmE 2 ,Èi (v) RmE 1 ,mE 2 (u + v) RmE 1 ,Èi (u) RmE 1 ,mE 2 (u v).
It is evident that the operator RmE (u) constructed from the monodromy matrix (4.1)
RmE (u) RmE (u) R1
m
E (u),
satisfies the analogous equation:
RmE 1 ,mE 2 (u v) RmE 1 (u) RmE 1 ,mE 2 (u + v) RmE 2 (v)
= RmE 2 (v) RmE 1 ,mE 2 (u + v) RmE 1 (u) RmE 1 ,mE 2 (u v).
Using this equation it is possible to show that corresponding supertraces are commuting
operators:
TmE 1 (u)TmE 2 (v) = TmE 2 (v)TmE 1 (u),
TmE (u) = strVmE RmE (u) = strVmE RmE (u) R1
m
E (u).
and therefore the operator TmE (u) is the generating function of integrals of motions for the
open chain.
5.2. Local Hamiltonian
Let us suppose that representations Èi and shifts ci in the product:
RÈ(u) R`,
E È1 (u + c1 )R`,
E È2 (u + c2 ) R`,
E ÈN (u + cN )
are fixed as follows:
E
È2 = È3 = = ÈN1 = `,
c2 = c3 = = cN1 = 0 RÈi ,È(0) = PÈi ,È,
i = 2, . . . , n 1,
where P is simply the permutation. In the case of s`(2) there exist integrable nearest
neighbour interactions for this slightly inhomogeneous chain [24]. We extend this result
to the case of s`(2|1) and construct the corresponding Hamiltonians.
The R-matrix (3.10) obeys the equation:
RÈ1 ,È2 (u) RÈ2 ,È1 (u) = P (u) 1,
(5.1)
709
where right side of equation is proportional of unit operator and P (u) is the function:
P (u)
(u + b1 `2 )(u + b1 + `2 )(u b2 + `1 )(u b2 `1 )

,
(`1 b1 )(`1 + b1 )(`2 b2 )(`2 + b2 )
which follows from the expression for matrix elements of the R-matrix (3.11). The explicit
form of the matrix RÈ2 ,È1 (u) is the following:

Dn (u) C n (u)
0
0
B n (u) A n (u)
0
0
,
RÈ2 ,È1 (u) =
0
0
0
Fn (u)
0
0
0
En (u)
where the matrix elements A n , B n , . . . are obtained from the elements An , Bn , . . . by formal
change of variables `1 `2 and b1 b2 . Due to Eq. (5.1) we have
R1
E E (u) RÈ2 ,È1 (u)
(5.2)
` 1 ,` 2
(u). This changes the

so that we can use the operator RÈ2 ,È1 (u) instead of operator R1
È1 ,È2
normalization of the operator TmE (u) only.
Let us calculate the first two coefficients in the Taylor expansion of the operator
TÈ(u) = strVÈ R`,
E È1 (u + c1 ) R`,
E ÈN (u + cN )RÈN ,È(u cN ) RÈ1 ,È(u c1 ).
The first coefficient is proportional to the unit operator for arbitrary representations Èi and
shifts ci due to the property (5.2),
TÈ(0) = strVÈ R`,
E È1 (c1 ) R`,
E ÈN (cN )RÈN ,È(cN ) RÈ1 ,È(c1 ) strVÈ 1.
Let us introduce the short-hand notation
0
0
(c)Rm,
(c)
H`,
Em
Em
Em
E È(c) + R`,
E (c) = R`,
E (c)Rm,
E
E È
(5.3)
and calculate the expression for T E0 (0) which contains several terms,
`
TÈ0 (0) = strVÈ H`,

E È1 (c1 ) + strVÈ R`,
E È1 (c1 )H`,
E È2 (0)RÈ1 ,È(c1 )
X
strVÈ R`,
+
E È1 (c1 ) P`,
E Èi1 H`,
E Èi (ci )PÈi1 ,È RÈ1 ,È(c1 ).
i=2,...,N
Note that this formula is true for c2 = = cN1 = 0 only. Let us consider each term
separately. The first term is constant
strVÈ H`,
E Èi (ci ) = const.
The ith term in the sum can be transformed easily to the simpler expression
trVÈ R`,
E È1 (c1 ) P`,
E Èi1 H`,
E Èi (ci )PÈi1 ,È RÈ1 ,È(c1 ) = HÈi1 ,Èi (ci ) strVÈ 1.
It turns out that the second term can be transformed to the analogous form (remember
c2 = 0) using the general formula
strVÈ R`,
E È1 (u)H`,
E È2 (0)RÈ1 ,È(u) = HÈ1 ,È2 (u) strVÈ 1 + const.
(5.4)
710
For the proof we start from the YangBaxter equation

RÈ2 ,È1 (u)RÈ2 ,È(u + v)RÈ1 ,È(v) = RÈ1 ,È(v)RÈ2 ,È(u + v)RÈ2 ,È1 (u),
differentiate this equation with respect to u, then put v = u and multiply both sides of
E
the obtained equation by the permutation PÈ2 ,È from the left (È2 = `):
0
R0`,
E È1 (u)R`,
E È (u)RÈ1 ,È(u) + R`,
E È (0)PÈ2 ,ÈRÈ1 ,È(u)
1
= RÈ1 ,È2 (u)PÈ2 ,ÈR0È ,È(0)RÈ2 ,È1 (u) + RÈ1 ,È2 (u)R0È ,È (u),
2
Calculation of the strVÈ using the equalities:

strVÈ R0`,
E È (u)RÈ1 ,È(u) = const,
1
strVÈ PÈ2 ,ÈR0È ,È(0) = const

2
leads to equality
0
0
strVÈ R`,
E È1 (u)R`,
E È (0)PÈ2 ,ÈRÈ1 ,È(u) = RÈ1 ,È2 (u)RÈ ,È (u) strVÈ 1 + const
2
and using the evident identity (consequence of Eq. (5.1)):

RÈ1 ,È2 (u)R0È ,È (u) = R0È ,È (u)RÈ2 ,È1 (u) + const
2
we arrive at the general formula (5.4). Finally we obtain the following representation for
T E0 (0):
`
T E1 (0)TÈ0 (0) = HÈ1 ,È2 (c1 ) +
N1
X
HÈi1 ,Èi (0) + HÈN1 ,ÈN (cN ) + const.
i=3
This operator, commuting with all integrals of motions, is a sum of two-particle operators
and can be considered as the Hamiltonian. The two-particle Hamiltonians entering the
sum can be easily calculated from the universal R-matrix by (5.3). The expression for the
HÈi1 ,Èi (0) coincides with the two-particle Hamiltonian for the periodic chain (4.7) up to
overall coefficient 2 but the expression for the HÈ1 ,È2 (c) is rather lengthy to be presented
here.
6. Conclusions
In this paper we have obtained the general rational solution of the YangBaxter equation
acting on the tensor product of arbitrary representations of the superalgebra s`(2|1).
We have represented the lowest weight module of s`(2|1) by polynomials in one even (z)
and two odd (, ) variables. Therefore the general R-matrices is an operator acting on twopoint functions being polynomials in the two sets of variables (z1 , 1 , 1 ) and (z2 , 2 , 2 ).
Instead of calculating this operator explicitly we have obtained its matrix elements on the
space of lowest weights. The eigenfunctions and eigenvalues can be easily calculated from
the obtained matrix.
From the general R-matrix for two isomorphic representations we have calculated the
nearest neighbour interaction Hamiltonians for an homogeneous closed chain. The result
applies both for the finite and infinite-dimensional representations on the sites.
711
Using the general R-matrix an integrable open chain can be constructed with arbitrary
isomorphic representations on the inner sites and other arbitrary representations at the end
points. The nearest neighbour interaction Hamiltonian has been calculated.
Acknowledgement
We thank L.N. Lipatov and A. Manashov for the stimulating discussion and critical
remarks. This work has been supported by Deutsche Forschungsgemeinschaft, grant No
KI 623/1-2 and by INTAS, grant No 96-524. One of us (D.K.) is grateful to Saxonian
Ministry of Science and Arts for supporting his visit at Leipzig University.
Appendix A
In this appendix we discuss briefly the derivation of the expression (3.11) for the general
R-matrix.
In matrix form the defining relation for the R-matrix reads as follows:
RÈ1 È2 (u)KAB = K AB RÈ1 È2 (u),
where:
A, B = 1, 2, 3,
(A.1)
K11 K12 K13

u
K = K21 K22 K23 (R2 R1 ) + R1 R2 ,
2
K31 K32 K33
u
K (R2 R1 ) + R2 R1 .
2
The matrix element K AB can be obtained from KAB by formal substitution 1 2 and
u u:
` 1 , b 1 , Z 1 `2 , b 2 , Z 2 ,
u u,
where Z (z, , ).
The operators KAB and lowest weights transform as follows:
KAB K AB ,
n (1)n n ,
n (1)n+1 n .
There are nine equations and we start from the simplest one.
A.1. Equation RK13 = K 13 R
The operator K13 commutes with the lowering generators V , W , S and the
covariant derivatives D1 . Therefore operator K13 maps lowest weight vectors with definite
chirality to lowest weight vector with the same chirality and decrease its power by one:
,
K13 n = n (u)n1
K13 n = n (u)n1
.
712
The explicit calculation gives:

n+ (u) = n(u + `n ),
n (u) = n(u + `n 1),
n (u) = n(u + `n ),
where
` n n + `1 + `2 ,
u
+ b1 b2 .
The action of operator K 13 on lowest weights vectors can be obtained from formulae for
K13 by formal substitution `1 , b1 , Z1 `2 , b2 , Z2 and u u:
,
K 13 n = n (u)n1
K13 n = n (u)n1
.
We project the operator equation RK13 = K 13 R onto the lowest weight vectors n :

+

+ Bn1 n1
RK13 n+ = K 13 Rn+ H n+ (u) An1 n1
RK13 n = K 13 Rn
Bn n+ (u)n1
,
= An n (u)n1

+
H n (u) Cn1 n1 + Dn1 n1

+
Dn n+ (u)n1
= Cn n (u)n1
which results in the recurrent relations:

n+ (u)An1 = n (u)An ,
n+ (u)Bn1 = n+ (u)Bn ,
n (u)Cn1 = n (u)Cn ,
n (u)Dn1 = n+ (u)Dn
with the following general solution:

0(u + `n + 1)
0(u + `n + 1)
,
Bn (u) = B(1)n
,
0(u + `n )
0(u + `n + 1)
0(u + `n )
0(u + `n )
,
Dn (u) = D(1)n+1
.
Cn (u) = C(1)n
0(u + `n )
0(u + `n + 1)
An (u) = A(1)n+1
The projection onto the odd lowest weight vectors n leads to analogous recurrent
relations
n (u)Fn1 = n (u)Fn ,
n (u)En1 = n (u)En
with general solution:

0(u + `n + 1)
0(u + `n + 1)
,
Fn (u) = F (1)n
.
0(u + `n + 1)
0(u + `n + 1)
We see that equation RK13 = K 13 R fixes the n-dependence of the matrix elements of REn (u) = E(1)n
matrix. The remaining equations fix the coefficients A, B, . . . up to overall normalization.

Due to commutativity of the operator K12 with lowering generators W , S we can
write down the general formulae for the action of operator K12 on even lowest weight
vectors:
+
+ b W + n1
+ c n1
.
K12 n = a W + n1
713
The coefficients a and b can be calculated using the following commutation relation:
{V , K12 } = K13 .
Remembering the known results for operator K13
{V , K12 } = K13 V K12 n = K13 n = n (u)n1
and the simple formula (b = b1 + b2 ):

V W + n = (`n + b)n
+
b (`n1 + b)n1
,
V K12 n = a (`n1 + b)n1
we obtain:
(`n1 + b)a + = n+ (u),
a = 0,
b+ = 0,
(`n1 + b)b = n (u).
The coefficient c+ can be calculated using the commutativity of K12 and D1+ :

+
+
+ c+ D1+ n1
=0
D1 , K12 = 0 D1+ K12 n+ = a + D1+ W + n1
and the simple formulae:
+
+
= (`1 + b1 )n1
,
D1+ W + n1
+
D1+ n1
= n1
.
We obtain
(`1 + b1 )a + = c+ .
The coefficient c can be calculated using the commutation relation with the covariant
derivative D2+ :
{D2+ , K12 } = D2+ W1 + (`2 + b2 )S1
+ c D2+ n1
= (`2 + b2 )S1 n
D2+ K12 n = b D2+ W + n1
and the formulae:
= (`2 + b2 )n1
,
D2+ W + n1
+
D2+ n1
= n1
,
S1 n = nn1
.
We obtain
(`2 + b2 )b + c = n(`2 + b2 ).
Finally we have (b = b1 + b2 ):
n+ (u)
+ (u)(`1 + b1 )
+
W + n1
n1 ,
n
`n1 + b
`n1 + b
(u)
n(u b1 b2 )(`2 + b2 )
W + n1
n1 .
+
K12 n = n
`n1 + b
`n1 + b
K12 n+ =
The analogous calculations for odd lowest weight vectors give:
714
n (u) + +
(u b)(`2 + b2 ) + n (u)(`1 + b1 )
W n1 +
n +
n ,
`n + b
`n + b
n(`n + b)
n (u)
W + n1
.
K12 n =
`n1 + b
K12 n+ =
The expression for the action of the operator K 12 is obtained by symmetry:

+ (u) +
+ (u)(`2 + b2 )
W n1 n
n1 ,
K 12 n = n
`n1 + b
`n1 + b
(u) + +
n(u + b)(`1 + b1 )
W n1
n1 ,
K 12 n+ = n
`n1 + b
`n1 + b
n (u) + +
(u + b)(`1 + b1 ) n (u)(`2 + b2 )
W n1 +
n
n ,
K 12 n+ =
`n + b
`n + b
n(`n + b)
n (u) +
W n1 .
K 12 n =
`n1 + b
The projection of the operator equation RK12 = K 12 R onto lowest weight vectors leads to
the following new recurrent relations (b = b1 + b2 ):
RK12 n+ = K 12 Rn+ H
(`1 + b1 )n+ (u)En1 = n(`1 + b1 )(u + b)An + (`2 + b2 )n+ (u)Bn ,
RK12 n = K 12 Rn H
n(`2 + b2 )(u b)En1
= n(`1 + b1 )(u + b1 + b2 )Cn + (`2 + b2 )n+ (u)Dn ,
RK12 n = RK 12 n H
n(`2 + b2 )(u b)An + (`1 + b1 )n (u)Cn = (`2 + b2 )n (u)Fn ,
n(`2 + b2 )(u b)Bn + (`1 + b1 )n (u)Dn = n(`1 + b1 )(u + b1 + b2 )Fn .
(A.2)

W V ,
B B
which has the following form in our representation:

,
b b.
Due to this automorphism the matrix KAB has definite symmetry properties with respect
to the transformation:
`1 , z1 `2 , z2 ,
1 , b1 2 , b2 .
The operators K12 ,K23 and lowest weights transform as follows:

K12 K23 ,
n (1)n n ,
n (1)n+1 n .
This symmetry allows to use the results of previous section and write down the formulae
for the action of the operator K23 on lowest weight vectors (b = b1 + b2 ):
715
n+ (u) + +
+ (u)(`2 b2 ) +
V n1 n
n1 ,
`n1 b
`n1 b
(u) +
n(u + b)(`1 b1 ) +
V n1 +
n1 ,
K23 n = n
`n1 b
`n1 b
n (u) +
(u + b)(`1 b1 ) + n (u)(`2 b2 )
V n1
n
n ,
K23 n =
`n b
`n b
n(`n b)
n (u)
+
V + n1
,
K23 n+ =
`n1 b
+ (u) +
+ (u)(`1 b1 ) +
V n1 n
n1 ,
K 23 n = n
`n1 b
`n1 b
(u) + +
n(u b)(`2 b2 ) +
V n1
n1 ,
K 23 n+ = n
`n1 b
`n1 b
n (u) +
(u b)(`2 b2 ) n (u)(`1 b1 ) +
V n1
n +
n ,
K 23 n =
`n b
`n b
n(`n b)
n (u) + +
V n1 .
K 23 n+ =
`n1 b
K23 n+ =
The projection of the operator equation RK23 = K 23 R onto lowest weight vectors leads to
the following new recurrent relations (b = b1 + b2 ):
RK23 n+ = K 23 Rn+ H
(`2 b2 )n+ (u)Fn1 = n(`2 b2 )(u b)An + (`1 b1 )n+ (u)Bn ,
RK23 n = K 23 Rn H
n(`1 b1 )(u + b)Fn1 = n(`2 b2 )(u b)Cn + (`1 b1 )n+ (u)Dn ,
RK23 n = RK 23 n H
n(`1 b1 )(u + b)An + (`2 b2 )n (u)Cn = (`1 b1 )n (u)En ,
n(`1 b1 )(u + b)Bn + (`2 b2 )n (u)Dn = n(`2 b2 )(u b)En .
(A.3)
The systems of equations (A.2), (A.3) fix the coefficients A, B, C, . . . up two arbitrary
constants. The next operator equation RK11 = K 11 R fixes the remaining ambiguity but we
avoid presenting the rather lengthy formulae here. Alternatively the missing equation can
be obtained as follows. The even lowest weight vectors n coincide for n = 0. Therefore
one obtains:
R0+ = R0 A(u + `0 ) + B =
C(u + `0 )
D
.
u + `0
u + `0
Finally the systems (A.2), (A.3) and this last equation fix the solution completely:
A = u + b1 b2 ,
B = (`1 + b1 )(`2 b2 ),
C = (`1 b1 )(`2 + b2 ),
D = (`2 b2 )(`2 + b2 )(u b1 b2 ) (u + b1 + b2 )(u b2 `1 )(u b2 + `1 ),

E = (u + b1 `2 )(u + b1 + `2 ),
F = (u b2 `1 )(u b2 + `1 ).
716
We have checked that the obtained R-matrix is really the solution also of the remaining
seven equations but the involved formulae become rather lengthy starting from the equation
RK11 = K 11 R.
Appendix B
The R-matrix acting in the tensor product of chiral modules can be obtained by simple
reduction from the general R-matrix (3.10). First of all we have to change the overall
normalization multiplying all matrix elements by the factor (`1 b1 )(`2 + b2 ). Let us
consider all possible special cases.
1. Chiral at 1, generic at 2:
V`1 ,`1 V`2 ,b2 =
`2 6= b2 ,
+
n
RÈ1 È2 (u)
n+
V`+n,b +
V
1,
1
`+n+ 2 ,b 2
n=0
n=0
n+ V`+n,b ,
n+ V
1,
1
`+n+ 2 ,b 2
+

n
An (u) 0
=R
,
n+
0 Fn (u)
0(u + `n + 1)
,
0(u + `n )
0(u + `n + 1)
(u + `1 b2 ).
Fn (u) = (1)n
0(u + `n + 1)
2. Atichiral at 1, generic at 2
An (u) = (1)n+1
V`1 ,`1 V`2 ,b2 =
V`+n,b +
n=0
n=0
n V`+n,b ,
1,
1
`+n+ 2 ,b+ 2
n V
`2 6= b2 ,
1,
1
`+n+ 2 ,b+ 2

n
n
Dn (u) 0
=
R
,
RÈ1 È2 (u)
n
n
0 En (u)
0(u + `n )
(u `1 b2 ),
Dn (u) = (1)n+1
0(u + `n + 1)
0(u + `n + 1)
.
En (u) = (1)n
0(u + `n + 1)
3. Chiral at 1, antichiral at 2:
V`1 ,`1 V`2 ,`2 =
V`+n,b ,
n+ V`+n,b ,
n=0
RÈ1 È2 (u)n+ = R An (u)n+ ,

4. Antichiral at 1, chiral at 2:
An (u) = (1)n
0 (u + `n + 1)
.
0 (u + `n )
V`+n,b ,
n V`+n,b ,
RÈ1 È2 (u)n = R Dn (u)n ;
Dn (u) = (1)n
V`1 ,`1 V`2 ,`2 =
717
n=0
0 (u + `n )
.
0 (u + `n + 1)
5. Antichiral at 1 and 2:
V`1 ,`1 V`2 ,`2 =
X
n=0
n V
1,
1
`+n+ 2 ,b+ 2
RÈ1 È2 (u)n = R En (u)n ,
1,
1
`+n+ 2 ,b+ 2
En (u) = (1)n
0 (u + `n + 1)
.
0 (u + `n + 1)
6. Chiral at 1 and 2:
V`1 ,`1 V`2 ,`2 =
X
n=0
n+ V
1,
1
`+n+ 2 ,b 2
RÈ1 È2 (u)n+ = R Fn (u)n+ ,
Fn (u) = (1)n
1,
1
`+n+ 2 ,b 2
0 (u + `n + 1)
.
0 (u + `n + 1)
Note that we do not fix the overall normalization of the obtained R-matrix.
Appendix C
In this appendix we discuss shortly the case of finite-dimensional representations and
show that the obtained general R-matrix reduces to the known ones [10] for the tensor
product of modules with minimal dimensions.
The tensor product of two finite-dimensional s`(2|1)-modules has the following direct
sum decomposition [3]:
V`1 ,b1 V`2 ,b2
= V`,b + V`+N,b + 2
N1
X
V`+n,b +
n=1
N1
X
n=0
1
1
`+n+ 2 ,b 2
N1
X
n=0
6 bi
ì =
n1
`1 = , n1 > 2,
2
1,
1
`+n+ 2 ,b+ 2
(C.1)
n2
`2 = , n2 > 2,
2
n1 + n2
,
b = b1 + b2 .
`=
N min(n1 , n2 ),
2
Note that this formula is applicable in the generic situation ì 6= bi , ì 6= 1/2. The direct
sum decomposition reduces for the module V 1 :
2 ,b
V`1 ,b1 V
`=
1
2 ,b2
= V`,b + V`+1,b + V
n1 + 1
,
2
b = b1 + b2 ,
1
1
`+ 2 ,b 2
n1 > 2,
+V
1,
1
`+ 2 ,b+ 2
`2 = 12 ,
718
V 1 ,b1 V 1 ,b2 = V1,b + V

2
b = b1 + b2 ,
1
1
2 ,b 2
+V
1,
1
2 ,b+ 2
`1 = `2 = 12 .
The origin of modifications for the tensor product involving the V 1 ,b -module is very
2
simple. The module V 1 ,b is the four-dimensional vector space with the following basis:
2
0 = 1,
0+ = ,
1 = z + b ;
0 = .
It is evident that we are able to construct the following two-point lowest weight vectors
only:

1
1
b2
0+ b2 +
0 ,
V`1 ,b1 V 1 0+ = 0 = 1,
2 ,b2
2
2
= 12.
+ = 12 ,
0
Note that

1
1
+
0 b2 +
0
b2
2
2

1
1
1 2 b2 +
1 2 ,
= z1 + b2 1 1 + z2 + b2 2 2 + b2
2
2
0+ = 12 ,
0 = 12 .
V 1 ,b1 V 1 ,b2 0+ = 0 = 1,
2
Let us consider the general R-matrix (3.10) acting in the tensor product V 1 ,b1 V 1 ,b2 .
2
2
In this case we have:

RÈ1 È2 (u) = A0 (u) + B0 (u) P1 + E0 (u) P2 + F0 (u) P3 ,
where Pi are projectors on the modules in the direct sum decomposition:
V 1 ,b1 V 1 ,b2 = V1,b + V
2
P1 V1,b ,
P2 V
1
1
2 ,b 2
1,
1
2 ,b 2
+V
1,
1
2 ,b+ 2
P3 V
b = b1 + b2 ,
`1 = `2 = 12 ,
1.
1
2 ,b+ 2
After a simple calculation one obtains:

RÈ1 È2 (u) P1 +
2u 1 + 2b1
2u 1 2b2
P2 +
P3 .
2u + 1 2b2
2u + 1 + 2b1
This result coincides with the expression for the R-matrix given in [10]:
R() = P1 +
4 1 + b1 + b2
4 1 b1 b2
P2 +
P3
4 + 1 b1 b2
4 + 1 + b1 + b2
up to the overall normalization and a redefinition of the spectral parameter:

b1 b2
.
2
The direct sum decomposition for the tensor product of the chiral modules (atypical
representations) is well known [3]. Now we need the simplest ones:
u = 2
1 1
2,2
1
2 ,b2
b = b1 + b2 ,
= V1,b + V
719
1,
1
2 ,b 2
`1 = `2 = 12 ,
b1 = 12 .
Let us consider the R-matrix (3.10) acting in the tensor product V

case we have (see Appendix B):
1 1
2,2
.
1
2 ,b2
In this
RÈ1 È2 (u) = A0 (u) P1 + F0 (u) P2 ,

where Pi are projectors on the modules in the direct sum decomposition:
P1 V1,b ,
P2 V
1.
1
2 ,b 2
After a simple calculation one obtains:

2u 1 2b2
P2 .
2u + 2
This result also coincides with the expression for such a R-matrix given in [10].
RÈ1 È2 (u) P1 +
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]
M. Scheunert, W. Nahm, V. Rittenberg, J. Math. Phys. 18 (1977) 155.

P.D. Jarvis, H.S. Green, J. Math. Phys. 20 (1979) 2115.
M. Marcu, J. Math. Phys. 21 (1980) 1277; J. Math. Phys. 21 (1980) 1284.
D. Arnaudon, C. Chryssomalakos, L. Frappat, J. Math. Phys. 36 (1995) 5262.
L. Frappat, P. Sorba, A. Sciarrino, Dictionary on Lie superalgebras, hep-th/9607161.
Z. Maassarani, J. Phys. A 28 (1995) 1305.
P.B. Ramos, M.J. Martins, Nucl. Phys. B 474 (1996) 678.
F. Eler, V. Korepin, Phys. Rev. B 46 (1992) 9147.
A. Foerster, M. Karowski, Nucl. Phys. B 396 (1993) 611.
M.P. Pfannmller, H. Frahm, Nucl. Phys. B 479 (1996) 575.
I. Cherednik, Theor. Math. Fiz. 61 (1984) 35.
E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
N. Andrei, H. Johannesson, Phys. Lett. A 100 (1984) 108.
H. Frahm, M.P. Pfannmller, A.M. Tsvelik, Phys. Rev. Lett. 81 (1998) 2116.
A. Foerster, M. Karowski, Nucl. Phys. B 408 (1993) 512.
A. Foerster, J. Phys. A 29 (1996) 7625.
D. Arnaudon, JHEP 12 (1997) 006.
J. Links, A. Foerster, J. Phys. A 32 (1999) 147.
A. Foerster, J. Links, A.P. Tonel, Nucl. Phys. B 552 (1999) 707.
L.N. Lipatov, High-energy asymptotics of multicolor QCD and exactly solvable lattice models,
Padova preprint DFPD-93-TH-70B, JETP Lett. 59 (1994) 596.
L.D. Faddeev, G.P. Korchemsky, Phys. Lett. B 342 (1995) 311.
V. Braun, S. Derkachov, A. Manashov, Phys. Rev. Lett. 81 (1998) 2020.
V. Braun, S. Derkachov, G. Korchemsky, A. Manashov, Nucl. Phys. B 553 (1999) 355.
S. Derkachov, G. Korchemsky, A. Manashov, Evolution equations for quark-gluon distributions
in multi-color QCD and open spin chains, hep-ph/9909539.
A. Belitsky, hep-ph/9907420, hep-ph/9903512, Phys. Lett. B 453 (1999) 59.
D. Karakhanian, R. Kirschner, Conserved currents of the three-reggeon interaction, hepth/9902147.
720
[27] D. Karakhanian, R. Kirschner, High-energy scattering in gauge theories and integrable spin
chains, hep-th/9902031, Fortschr. Phys. 48 (2000) 139.
[28] A.P. Bukhvostov, G.V. Frolov, L.N. Lipatov, E.A. Kuraev, Nucl. Phys. B 258 (1985) 601.
[29] A. Belitsky, D. Mller, A. Schfer, Implications of N = 1 supersymmetry for QCD conformal
operators, hep-ph/9811484, Phys. Lett. B 450 (1999) 126.
[30] H. Frahm, Doped Heisenberg chains: spin S generalizations of the supersymmetric t J model,
cond-mat/9904157, Nucl. Phys. B 559 (1999) 613.
[31] P.P. Kulish, N.Yu. Reshetikhin, E.K. Sklyanin, Lett. Math. Phys. 5 (1981) 393403.
[32] E.K. Sklyanin, Quantum inverse scattering method. Selected topics, in: Mo-Lin Ge (Ed.),
Quantum Group and Quantum Integrable Systems, Nankai Lectures in Mathematical Physics,
World Scientific, Singapore, 1992, pp. 6397; hep-th/9211111.
[33] P.P. Kulish, E.K. Sklyanin, Zap. Nauchn. Sem. LOMI 95 (1980) 129.
[34] P.P. Kulish, Zap. Nauchn. Sem. LOMI 145 (1985) 140; J. Sov. Math. 35 (1986) 1111.
[35] P.P. Kulish, YangBaxter equation and reflection equations in integrable models, hepth/9507070.
[36] V.O. Tarasov, L.A. Takhtajan, L.D. Faddeev, Theor. Math. Phys. 57 (1983) 163181.
[37] L.D. Faddeev, Les-Houches lectures 1995, hep-th/9605187.

Integrability of the Dn2 vertex models with open

boundary
M.J. Martins a,b , X.-W. Guan c
a School of Natural Sciences, Institute for Advanced Study, Olden Lane, Princeton, NJ 08540, USA
b Departamento de Fsica, Universidade Federal de So Carlos, Caixa Postal 676, 13565-905, So Carlos,
Brazil
c Institut fr Physik, Technische Universitt, D-09107 Chemnitz, Germany
Received 25 February 2000; accepted 27 April 2000
Abstract
We investigate various aspects of the integrability of the vertex models associated to the Dn2
affine Lie algebra with open boundaries. We first study the solutions of the corresponding reflection
equation compatible with the minimal symmetry of this system. We find three classes of general
solutions, one diagonal solution and two non-diagonal families with a free parameter. Next we
perform the Bethe ansatz analysis for some of the associated open D22 spin chains and we identify the
boundary having quantum group invariance. We also discuss a new D22 R-matrix. 2000 Elsevier
1. Introduction
Much work has been done in integrable lattice statistical mechanics models with open
boundary conditions, since Sklyanin [1] generalized the quantum inverse scattering method
to tackle the boundary problem. The bulk Boltzmann weights of an exactly solvable lattice
system are usually the non-null matrix elements of a R-matrix R() which satisfies the
YangBaxter equation. The integrability at boundary, for a given bulk theory, is governed
by the reflection equation, which reads
1
R12 ( ) K ()R21 ( + ) K ()
2
=K ()R12 ( + ) K ()R21 ( ),
(1)
where the matrix K () describes the reflection at one of the ends of an open chain. Similar
equation should also hold for the reflection K+ () at the opposite boundary. However, for
several relevant lattice models K+ () can be directly obtained from K (). For example,
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 5 9 - 5
722
M.J. Martins, X.-W. Guan / Nuclear Physics B 583 [FS] (2000) 721738
this is the case of models whose R() matrix satisfies extra properties such as unitarity,
P and T invariances and crossing symmetry [1,2].
Therefore, the first step toward constructing integrable models with open boundaries
is to search for solutions of the reflection equation. To date, solutions of this equation
have been found for a number of lattice models ranging from vertex systems based on Lie
algebras [38] to solid-on-solid models and their restriction [911]. Classification of such
solutions for particular systems [1216] as well as extensions to include supersymmetric
models [1721] can also be found in the literature.
In spite of all these works, there is an interesting vertex model based on the
non-exceptional Dn2 Lie algebra for which little is known about the solution of the
corresponding reflection equation. This is probably related to the fact that the Dn2 R-matrix
does not commute for different values of the rapidity [22], consequently the trivial diagonal
solution K () = I d does not hold for this system [3]. The purpose of this paper is to
bridge this gap, by presenting what we hope to be the minimal solution of the reflection
equation for Dn2 vertex models. This result offers us the possibility to understand a relevant
open problem which is the integrability of the Dn2 vertex model with quantum algebra
symmetry. In fact, this symmetry has been found for all vertex models based on nonexceptional Lie algebras [3] except for the Dn2 model. It turns out that, by carrying out
a Bethe ansatz analysis, we are able to identify this symmetry for the simplest D22 model
and conjecture it for arbitrary values of n.
We have organized this paper as follows. We start next section by considering the
reflection equation for the D22 vertex model. We find one diagonal solution without free
parameters and two non-diagonal families which depend on a free parameter. We also
derive the corresponding integrable one-dimensional open spin chains. In Section 3 we
present the Bethe ansatz solutions of the open D22 spin chain associated to the diagonal
K-matrix and to a special manifold of the first non-diagonal family. This allows us to
identify the quantum group symmetry for the D22 model. In Section 4 we generalize the
K-matrices results of Section 2 for arbitrary values of n > 2. Section 5 is reserved for our
conclusions as well as a discussion on possible new Dn2 R-matrices. In Appendix A we
collect some useful relations and Appendix B contains a new D22 R-matrix as well as its
boundary behaviour.
2. The D 22 K-matrices
The D22 vertex model has four independent degrees of freedom per bond and its
Boltzmann weights preserve only one U (1) symmetry out of two possible ones. Here we
are interested in looking at solutions of the reflection equation that commute with this
symmetry. We find that the most general K-matrix having this property is
0
0
0
Y1 ()
0
0
Y2 () Y5 ()
.
(2)
K () =
0
0
Y6 () Y3 ()
0
Y4 ()
723
Our next step is to substitute this ansatz in Eq. (1) and look for relations that constraint
the unknown elements Yj (), j = 1, . . . , 6. Although we have many functional equations,
a few of them are actually independent, and the most suitable ones have been collected in
Appendix A. The basic idea is to try to solve such equations algebraically, which hopefully
will produce a general ansatz for functions Yj () containing several arbitrary parameters.
The general strategy we use is to separate these equations in terms of ratio of functions
depending either on or on . From the relations (A.5)(A.7) one easily concludes that the
simplest possible solution is to take Y5 () = Y6 () = 0. This is the diagonal solution, and
by employing the separation variable method described above for the relations (A.5)
(A.7) we are able to fix the following ratios
e 1
Y2 ()
=
,
Y1 () e 1
e 2
Y3 ()
=
,
Y1 () e 2
e 3
Y4 ()
=
,
Y3 () e 3
(3)
where j , j = 1, 2, 3 are arbitrary constants.

These relations enable us to write an ansatz for three unknown functions in terms
of a normalizing factor, say Y1 (). Substituting the relations (3) back to the reflection
equation (1), we conclude that all the parameters j are fixed by
I
1 = 2 = 1/3 = ,
q
(4)
where q is the deformation parameter of the D22 R-matrix [22]. This leads us to our first
solutions with no free parameter,
e I / q
e + I / q
(1)
(1)
(1)
Y3 () =
Y2 () =
Y1 () = 1,
,
,
e I/ q
e + I/ q
e + I / q e I q
(5)
Y4(1) () =
.
e + I / q e I q
Next we turn our search for non-diagonal solutions now with both Y5 () and Y6 () non
null. From Eqs. (A.8)(A.11), we notice that it is possible to solve Y2 (), Y3 () and Y6 ()
in terms of Y5 (). At this point we should keep in mind that we are looking for regular Kmatrices, i.e., K (0) identity. After some simplifications, we find the following general
solutions

(6)
1 1 + e2 [Y5 () + Y6 ()] = 2 e [Y5 () Y6 ()],

2
(7)
e 1 [Y2 () + Y3 ()] = 3 e [Y6 () Y5 ()],
Y2 () Y3 () = 4 [Y5 () Y6 ()],
(8)
where j are four arbitrary parameters. These are linear equations which can be easily
solved for the ratios Y2 ()/Y5 (), Y3 ()/Y5 () and Y6 ()/Y5 (). Taking this into account
as well as Eqs. (A.5) and (A.7), we end up with the following ansatz for functions Yj ()

Y1 () = 5 e2 + 6 e + 7 /e , Y2 () = 1 + e2 4 e2 1 3 e ,
(9)

2
2
4 e 1 3 e , Y4 () = 8 e + 9 e + 10 e , (10)
Y3 () = 1 + e

Y5 () = e2 1 2 e + 1 1 + e2 ,
724

Y6 () = e2 1 2 e 1 1 + e2
(11)
having altogether ten free parameters. Substituting this ansatz back to the reflection
equation and after involving algebraic manipulations, we find that nine parameters are in
fact fixed, leading us to two classes of non-diagonal solution with a free parameter. The
first class is given by

Y1(2) (, ) = e2 + q 2 qe2 1 e ,

(2)
(12)
Y4 (, ) = e2 + q 2 q e2 e ,

(1 + e2) 2
2 e 1 q e (1 + q) 1 2 q ,
2

(1
+
e2)
(2)
2 e2 1 q e (1 + q) 1 2 q ,
Y3 (, ) =
2

(e2 1)
(1 q) 2 q + 1 e
Y5(2) (, ) = Y6(2) (, ) =
2
while the second family is

Y1(3) (, ) = e2 q e2 1 e ,

(3)
Y4 (, ) = e2 q e2 e ,
Y2(2) (, ) =
(13)
(14)
(15)
(16)
(1 + e2)
(1 q)( 1)e,
(17)
2
p

(e2 1) 2
2 e + 1 q + (1 + q)(1 + )e ,
(18)
Y5(3) (, ) =
2
p

(e2 1)
(3)
2 e2 + 1 q + (1 + q)(1 + )e ,
(19)
Y6 (, ) =
2
where is an arbitrary parameter.
Since the D22 R-matrix is PT invariant and crossing symmetric, the K+ () matrices at
the opposite boundary are easily derived from the above solutions [2,3]. More precisely,
we have
Y2(3) (, ) = Y3(3) (, ) =
t
(ln[q] , + )M,
K+ (, + ) = K
where M is a matrix related to the crossing matrix V by M

Appendix A, we have that for the D22 model M is given by

M = diag q, 1, 1, q 1 .
(20)
= V tV
[3]. From the results of

(21)
Having found the K () matrices, one can construct the corresponding commuting
transfer matrix (). Following Sklyanin [1], we have
i
ha
a (l)
(m)
(22)
t (l,m) () = Tra K + ()T () K ()T 1 () ,
where T () = RaL () Ra1 () is the monodromy matrix of the associated closed chain
with L sites. This means that the three families of K () matrices we found will produce
725
nine possible types of open boundary conditions. The corresponding Hamiltonian of the
spin chains with open boundaries are obtained by expanding the transfer matrix t (l,m) ()
(m)
in powers of . When Tr[K+ (0)] is non-null, the Hamiltonian H (l,m) is proportional to
the first-order expansion [1]
a (m)

a (l)
L1
X
Tra K (0)HLa
1 d K ()
(l,m)
|=0 +
,
(23)
=
Hk,k+1 +
H
(m)
2
d
Tr[K+
(0)]
k=1
d
Rk,k+1 ()|=0 is the two-body bulk Hamiltonian and is the
where Hk,k+1 = Pk,k+1 d
normalization R12 (0) = P12 . 1
For the first two solutions we indeed have Tr[K+ (0)] 6= 0 while for the third one
Tr[K+ (0)] = 0. In this last case one has to consider the second order expansion in
the spectral parameter [23,24]. We find convenient to write the expression for the
z
and ,i
with components =, acting
Hamiltonians in terms of Pauli matrices ,i
on the site i of a lattice of size L. In terms of these operators and up to irrelevant additive
constants, 2 we have
I (q 1/q) X
Hk,k+1
=
2
k=1
(
(q 1/q)2 X (l)
(l)
z
z
z
+
+I
( ),1
+ (l) ,1
,1
+ J ( ),1
,1
2
=,
X
(l)
+
z
z
(m) z
+ J ( ),1
,1
(m)
,L ,L
(+ ),L +
L1
(l,m)
=,
+
+ J(m) (+ ),L
,L
+ J(m) (+ ),L
,L
)
,
(24)
where the expression of the bulk part H k,k+1 is

(q 1/q) z
z +
+
,k + ,k
+ ,k+1
,k+1
,k+1 ,k+1
H k,k+1 =
2

+
+ z
z
,k + ,k
,k ,k+1 + ,k+1
,k

1 2 +
+
+
+
q
,k+1 + ,k
,k+1 ,k
,k+1
,k ,k+1 ,k
+
q
+ +

+ +
+ ,k
,k+1 ,k
,k+1 + ,k
,k+1 ,k
,k+1
+

+
z
z
,k+1 1 + ,k
,k+1
2 ,k ,k+1 + ,k

+
+
z
z
,k+1 + ,k
,k+1 1 + ,k
,k+1
+ ,k

1 +
+
z
z
,k ,k+1 + ,k
q+
,k+1 1 ,k
,k+1
+
q
1 The normalization we use for R() (see Appendix A) produces = (q 1/q)2 for D 2 model.
2
2 We also note that we have normalized the Hamiltonian by the pure imaginary number.
726

+
+
z
z
+ ,k
,k+1 + ,k
,k+1 1 ,k
,k+1

z

1 +
+
z
,k ,k+1 + ,k
q
,k+1 ,k
,k+1
q
z

+
+
z
,k+1 + ,k
,k+1 ,k
,k+1
+ ,k

(q + 1/q)
z
z
z
z
,k
,k
+ ,k+1
,k+1
+ 1
2
2

( q 1/ q)
z
z
z
z
,k
,k+1
+ ,k
,k+1
4

(q + 1/q) 3
z
z
z
z
+
,k
,k+1
+ ,k
,k+1
4
2

1
(q 1/q) X z
z
,k ,k+1 2 q +
Ik,k+1 .
+
2
q

=,
Turning to the boundary interactions we found that the chemical potentials are given by
q
q
(1)
(1)
,
(
)
=
1/2
I
(
)
=
1/2
+
I
1
+
q
1
+
q
(1 + q + 2 q)
(1 + q 2 q)
(2)
(2)
(l)
,
( ) =
,
( ) = ( ) =
2
2
(1 + q)( q 1)
(1 + q)( 2 q 2 1)
(3)( ) = (3) =
(25)
1
(l)
while the on-site parameters (l) and J ( ) are
(q 1)
(1) =
,
2(1 + q)
(q 1)
,
(l) = (2) =
2(1
+ q)
(1 + q)
(3) =
2(q 1)
(26)
and
(1)
J ( ) = J(1) = 0,
(q 1)(1 + 2 q)
(2)
(27)
,
J ( ) = J(2)( ) =
(l)
2 q 1)
J ( ) =
(1
+
q)(
(1 + q)(1 + ) + 4 q
(1 + q)(1 + ) 4 q
J(3)( ) =
, J(3) ( ) =
.
(q 1)( 1)
(q 1)( 1)
A natural question to be asked is which (if any) of these solutions would lead us to an
integrable D22 model with quantum algebra symmetry. One way to investigate that is by
applying the Bethe ansatz method to diagonalize the above open spin chains. This allows
us to extract information about the eigenspectrum, which in the case of quantum algebra
invariance, should be highly degenerated (see, e.g., [25]). In next section we will discuss
this problem in details.
727
3. Bethe ansatz analysis

The purpose of this section is to study the spectrum of some of the open spin chains
presented in Section 2 by the coordinate Bethe ansatz formalism. One of our motivations
is to identify the boundary that leads us to the quantum group symmetry. We begin by
P P
z
noticing that the total number of spins N s = L
=, ,i is a conserved quantity and
i=1
its eigenvalues ns labels the many possible disjoint sectors of the Hilbert space. Therefore,
the wave function solving the eigenvalue problem H |9ns i = E (l,m) (L)|9ns i can be written
as follows
XX
f (1 ,...,n ) (xQ1 , . . . , xQns )+1 ,xQ +ns ,xQns |0i,
(28)
|9ns i =
1
j xQj
where |0i denotes the ferromagnetic state (all spins up) and 1 6 xQ1 6 xQ2 6 6 xQns 6
L indicate the positions of the spins.
We will start our study by first considering the open spin chain H (1,1) corresponding
to the diagonal K-matrix solution. As it is customary we begin our discussion of the
eigenvalue problem in the sector of one down spin, ns = 1. In this sector, we find that
for 1 < x < L
I
E (1,1)(L)f ()(x)
2(1/q q)
= (L 2)f () (x) + f () (x + 1) + f ()(x 1) +
(1 q)2 ()
f (x),
4q
=, ,
(29)
where we have defined = q + 1/q. The matching condition at the left and right
boundaries gives us the following constraints
()
()
p1 d1
f (1)
f (0)
=
(30)
d1 p1
f () (0)
f () (1)
and
f () (L + 1)
f () (L + 1)

=
pL dL
dL pL

f () (L)
,
f () (L)
where the matrices parameters are given by
3I + q + q 2 (I + 3 q)
3I + q + q 2 (I + 3 q)
,
p1 =
,
p1 =
4q(I + q)
4q(I + q)
3 2q + 3q 2 2I q + 2I q q
,
pL =
4q
3 2q + 3q 2 + 2I q 2I q q
,
pL =
4q
d1 = d1 = dL = dL =
q 1/q
.
4
(31)
(32)
(33)
(34)
728
In order to go ahead it is crucial to notice that both boundary constraints (30) and
(31) can be diagonalized by the same unitary transformation U . After performing this
transformation the new components f (x) = Uf (x) satisfy
()
()
1 0
f (1)
f (0)
=
,
(35)
0 2
f() (0)
f() (1)
and
f() (L + 1)
f() (L + 1)

=
0
0 1

f() (L)
.
f() (L)
(36)
Clearly, Eq. (29) for 1 < x < L remains the same but now for the transformed amplitudes
f()(x). Now we reached a point in which one can try the usual Bethe ansatz (e.g., see
Refs. [2628]), namely
f() (x) = A (k)eikx A (k)eikx
(37)
and by substituting this ansatz in (29) we obtain the following eigenvalue

(1 q)2
I
E (1,1)(L) = (L 2) + 2 cos(k) +
.
2(1/q q)
4q
(38)
The fact that this ansatz should be also valid for the ends x = 1 and x = L provides us
constraints for the amplitudes A(k) and A(k), which reads
A(k) = eik A(k) and A(k) =
ik
) 2i(L+1)k
2e
e
A(k)
(1 2 eik )
(1
(39)
whose compatibility gives a restriction on the momentum k, namely

e2ikL
(eik
2)
( 2 eik 1)
= 1.
(40)
The next task is to generalize these results for arbitrary numbers of down spins. For a
general multiparticle state, we assume the Bethe ansatz wave function
f(1 ,...,n ) (xQ1 , . . . , xQns )
ns
Y
X
[ik x ]
sgn(P )
e pj Qj A(kP Q1 , . . . , kP QNe )Q1 ,...,Qns ,
=
P
(41)
j =1
where P is the sum over all the permutations of the momenta, including the negations
kj kj , and the symbol sgn accounts for the sign of the permutations and negations.
It turns out that for configurations such that |xQi xQj | > 2 the open spin chain H (1,1)
behaves as a free theory and the corresponding eigenvalues are
X
(1 q)2
I
E (1,1)(L) = (L 1) +
.
2 cos(kj ) +
2(1/q q)
4q
ns
(42)
j =1
The new ingredient for ns > 2 is that the nearest neighbor spin configurations enforce
constraints on the amplitude of the wave function. This condition enhances a relation
729
between the exchange of two states such as {(ki , i ); (kj , j )} and {(kj , j ); (ki , i )}
which ultimately is represented by the two-body scattering
A...j ,i ... (. . . , kj , ki , . . .) = Si,j (ki , kj )A...i ,j ... (. . . , ki , kj , . . .)
(43)
while the reflection at the left and right ends generalizes equation (29), which now reads
Ai ,... (kj , . . .) = eikj Ai ,... (kj , . . .),
A...,i (. . . , kj ) =
(1 2 eikj ) 2i(L+1)kj
e
A...i , (. . . , kj ).
(1 2 eikj )
(44)
(45)
Fortunately, the bulk two-body scattering amplitude Si,j (ki , kj ) has been recently
identified in Ref. [29] for the periodic chain. This result is of enormous help here since
it allows us to choose the suitable parametrization for the momenta kj in terms of the
S-matrix rapidities j , which is
eikj =
sinh(j i /2)
,
sinh(j + i /2)
(46)
where we have conveniently defined q = ei . For explicit expression of the non-null

S-matrix elements see Ref. [29].
In this general case, the compatibility between the bulk and boundary scattering
constraints (43)(45) leads us to the Bethe ansatz equation for the momenta kj
e2ikj L
(eikj
2)
( 2 eikj 1)
= j (k1 , . . . , kns ),
(47)
where j (k1 , . . . , kns ) are the eigenvalues of the auxiliary inhomogeneous transfer matrix
tj = Sj ns (kj , kns ) Sj 1 (kj , k1 )S1j (k1 , kj ) Snsj (kns , kj ). The integrability of this
latter inhomogenous problem follows from the fact that the 2 2 identity K-matrix is a
solution of the reflection equation associated to the two-body scattering Sij . As was shown
in Ref. [29] there is no need of a second Bethe ansatz to solve this auxiliary eigenvalue
problem. By adapting the results of Ref. [29] to our case and by relating the momenta kj
and the rapidities j by Eq. (46) we find that the Bethe ansatz equations are given by

sinh(j i /2) 2L cosh(j + i /2)
sinh(j + i /2)
cosh(j i /2)
ns
Y
sinh(j /2 k /2 i /2) sinh(j /2 + k /2 i /2)
,
=
sinh(j /2 k /2 + i /2) sinh(j /2 + k /2 + i /2)
k=1
j = 1, . . . , ns
(48)
and the eigenvalues (42) in terms of the rapidities j are

E (1,1)(L) = 8 sin3 ( )
ns
X
j =1
1
cos( ) cosh(2j )
4(L 1) sin(2 ) + 4 sin( ) sin2 ( /2).
(49)
730
Clearly, the Bethe ansatz equations for the open spin chain H (1,1) are not just the
doubling of the corresponding results of the closed chain with periodic boundary
conditions [29,30] due to an additional boundary left hand factor. We recall here that the
doubling property has been argued [31] to be one of the main features of a quantum
algebra invariant open spin chain at least for standard forms of comultiplication. Looking
at the spectrum of H (1,1), however, we notice a certain pattern of degeneracies which
suggests an underlying hidden symmetry. It could be that the diagonal boundary solution
corresponds to an asymmetric form of coproduct since this, in principle, is allowed too
[32,33].
Next we turn our attention to the first non-diagonal solution and its corresponding open
spin chain. In this case, at least for generic values of , the Bethe ansatz construction we
just explained above needs further generalizations. This can be seen even at the level of
one down spin state, since there is not a unique transformation that diagonalizes both left
and right boundary matrix problems. However, there is a particular manifold, + = q ,
in which our previous Bethe ansatz formulation is still valid. Fortunately, as we shall see
below, this special manifold will be sufficient to single out the boundary leading us to the
quantum algebra symmetry. Since for + = q , the Bethe ansatz analysis is very similar
to the one just described above, we restrict ourselves to present only the final results. We
found that the Bethe ansatz equations for the Hamiltonian H (2,2) at + = q are

sinh(j i /2) 2L
sinh(j + i /2)
ns
Y
sinh(j /2 k /2 i /2) sinh(j /2 + k /2 i /2)
,
=
sinh(j /2 k /2 + i /2) sinh(j /2 + k /2 + i /2)
k=1
j = 1, . . . , ns
(50)
while the corresponding eigenvalues are given by

ns
X
1
4(L 1) sin(2 )
cos( ) cosh(2j )
j =1
#
"
X

(2)
(2)
(2)
.
4 sin( )
( ) (+ ) + 2
E (2,2)(L) = 8 sin3 ( )
(51)
=,
Now the Bethe ansatz equations do have the doubling property at + = q and this
is an extra motivation to investigate the eigenspectrum of H (2,2) . It turns out that at the
value = 0 and therefore + = 0 we discover that the spectrum of the open chain H (2,2)
is specially highly degenerated. In fact, after some algebraic manipulations, we check that
for = 0 the Hamiltonian H (2,2) has the appropriate boundary coefficients to ensure
commutation with Uq (D22 ). Therefore, we finally managed to identify the quantum algebra
symmetry for the D22 vertex model.
Finally, it seems desirable to solve the open spins chains associated to the non-diagonal
solutions for arbitrary values of the parameters . The coordinate Bethe ansatz method,
however, leads us to cumbersome calculations even for the first excitation over the
731
reference state. In such general case it seems wise to tackle this problem by using a more
unifying technique such as the algebraic Bethe ansatz approach. Since the basics of this
method has been recently developed for the Dn2 vertex models [29] we hope to return to
this problem elsewhere.
4. The D 2n K-matrices
Here we shall consider the generalizations of the K-matrices solutions of Section 2 for
the general Dn2 model. This system has n 1 distinct U (1) conserved charges, and the
K-matrix ansatz compatible with these symmetries can be represented by the following
block diagonal matrix
Yn+2 (), . . . , Y2n () ,

(52)
K () = diag Y1 (), . . . , Yn1 (), A(),
where A()
is a 2 2 matrix

Y2n+1 ()
Yn ()
,
A()
=
Y2n+2 () Yn+1 ()
(53)
where Yj (), j = 1, . . . , 2n + 2 are functions we have determined by solving the reflection

equation. Notice that for n = 2 we recover our starting ansatz of Section 2.
Substituting this ansatz into the reflection equation, we realize that the simplest possible
solution is the symmetric one, namely
Y1 () = Y2 () = = Yn1 ()
and Yn+2 () = Yn+1 () = = Y2n ().
(54)
It turns out that the remaining functional equations for the functions Y1 (), Yn (),
Yn+1 (), Y2n (), Y2n+1 () and Y2n+2 () are very similar to those presented in Appendix A.
Therefore, they can be solved by the same procedure described in Section 2 and in what
follows we only quote our final results. As before we find three general families of
K-matrices, and the diagonal one is given by
Y1(1) () = 1,
(1)
() =
Yn+1
(1)
() =
Y2n
Yn(1) () =
e I q (n1)/2
,
e I q (n1)/2
(55)
e + I q (n1)/2
,
e + I q (n1)/2
e + I q (n1)/2 e I q (n1)/2
.
e + I q (n1)/2 e I q (n1)/2
The one-parameter families of non-diagonal K-matrices are given by

(2)
Y1 (, ) = e2 + q n1 2 q n1 e2 1 e ,

(2)
(, ) = e2 + q n1 2 q n1 e2 e ,
Y2n
Yn(2) (, ) =

(1 + e2) 2
2 e 1 q n1 e 1 + q n1 1 2 q n1 ,
2
(56)
(57)
(58)
732

(1 + e2)
2 e2 1 q n1 e 1 + q n1 1 2 q n1 , (59)
2

(e2 1)
(2)
(2)
1 q n1 2 q n1 + 1 e
(, ) = Y2n+2
(, ) =
(60)
Y2n+1
2
(2)
Yn+1
(, ) =
and

(3)
Y1 (, ) = e2 q n1 e2 1 e ,

(3)
(, ) = e2 q n1 e2 e ,
Y2n
(61)

(1 + e2)
(62)
1 q n1 1 e ,
2
p

(e2 1) 2
(3)
2 e + 1 q (n1)/2 + 1 + q n1 1 + e , (63)
(, ) =
Y2n+1
2
(3)
(, ) =
Yn(3) (, ) = Yn+1
(3)
Y2n+2 (, ) =
p
(e2 1)
2 e2 + 1 q (n1)/2
2

+ 1 + q n1 1 + e .
(64)
The next natural step is to search for asymmetric K-matrices for n > 3, i.e., those having
Y1 () 6= Y2 () 6= Yn1 () and Yn+2 () 6= Yn+3 () 6= Y2n (). In this case the number
of free parameters grows rapidly with n and the solution of the reflection equation becomes
more involving. To illustrate that, we consider the D32 model and for sake of simplicity
we look first for diagonal solutions. There are six functions Yj () to be determined and
their ratios are fixed by choosing some easy looking relations coming from the reflection
relation. More precisely, we have found the following equations:
e2 c1
Y2 ()
= 2
,
Y1 () e
c1
e c2
Y3 ()
=
,
Y1 () e c2
e c4
Y6 ()
=
,
Y4 () e c4
Y5 () e2 c5
= 2
,
Y6 ()
e c5
e c3
Y4 ()
=
,
Y1 () e c3
(65)
(66)
where cj are once again constants yet to be determined. Substituting these relations back
to the reflection equation we find only one possible manifold for the parameters cj , which
reads
c1 = c5 = 1 and c2 = c3 = c4 = 0.
(67)
After an appropriate normalization, this solution leads us to a new diagonal K-matrix

for the D32 model

D2
K 3 () = diag e2 , 1, 1, 1, 1, e2 .
(68)
It is plausible that this almost unity solution and its extensions generalizes for arbitrary
values of n > 4. Next we have looked at the possibility of asymmetric non-diagonal
solutions for the D32 model. It turns out that, within our algebraic approach, we did not
found any of such solutions. However, this possibility should not be completely rule out,
733
at least for general n, since we have so many free parameters that the chance to miss
a particular integrable manifold is high. In general, classification of the solutions of the
reflection equation seems to be an intricated problem even for simpler models [1216]. We
hope, however, that our K-matrices results prompt further investigation concerning this
problem for the Dn2 vertex models.
We would like to conclude this section with the following remarks. The K+ () matrices
can be obtained from K () by the isomorphism

t
(n 1) ln[q] M,
(69)
K+ () = K
where M is a diagonal matrix given by

M = diag q (2n3), q 2n5 , . . . , 1, 1, . . . , q (2n5) , q (2n3) .
(70)
Once we are equipped with K () matrices, the construction of the corresponding open
spin chains is possible along the lines of Section 2. Similarly, at least for the diagonal
solution, one can also repeat our Bethe ansatz construction without further technical
difficulties. In particular, we conjecture that the open spin chain associated to the first nondiagonal solution at = 0 is the one having the underlying quantum group symmetry.
5. Concluding remarks
In this paper we have made a great deal of progress towards the understanding of
the integrability of the Dn2 vertex model with open boundaries. We have investigated
the solutions of the associated reflection equation and found three general families of
K-matrices which respect the minimal U (1) symmetries of this system. We have carried
out a Bethe ansatz analysis for the simplest case, D22 model, revealing to us that the first
non-diagonal solution at = 0 possesses the special quantum algebra symmetry. In fact,
the structure of the K-matrices at this particular point leads us to conjecture that this will
be the case for arbitrary values of n.
We believe that our results open an avenue for further investigations. One clear
possibility is to use the Bethe ansatz results of Section 3 to compute the thermodynamic
behaviour, the bulk and the surface critical exponents. It would be also interesting to
generalize our results of Section 3 for all sort of open boundary conditions and for arbitrary
values of n. In this case, probably the most suitable tool would be instead the algebraic
Bethe ansatz approach. This method would allow us to show that indeed the Bethe ansatz
states are highest weight states of the underlying quantum algebra in the case of the first
non-diagonal family at = 0.
Other interesting issue is to apply the notion of the quantum group twisting [34] to find
out slightly different Dn2 R-matrices. As a result, this might lead us to integrable models
with very different behaviour, for an example see Ref. [35]. The practical implementation
of twisting, however, seems to be quite involving specially for an algebra such Dn2 . To shed
some light to this problem we proceed in a much more phenomenological way. Motived
by the structure of the non-diagonal solutions, we add extra Boltzmann weights to the
734
Jimbos R-matrix to account for such boundary terms at the level of the associated bulk
Hamiltonian. Next step is to try to solve the YangBaxter equation for this novel R-matrix
structure. It turns out that we succeed to find a new R-matrix solution for the D22 model.
Since this involves many technicalities, we have summarized it in Appendix B together
with the study of the corresponding solutions of the reflection equation. We hope that these
results will be useful to motivate further progress in this problem.
Acknowledgements
The work of M.J. Martins has been partially supported by the Lampadia Foundation and
by the Brazilian research Agencies CNPq and Fapesp. X-W. Guan thanks Fapesp and DFGSFB393 for financial support and the hospitality at the the Institut fr Physik, Technische
Universitt, Chemnitz.
Appendix A. R-matrix properties and reflection equations

In this appendix we briefly discuss some useful properties of the Dn2 R-matrix [22]. We
also present for n = 2 some relevant relations derived from the reflection equation. The
Dn2 R-matrix satisfies, besides the unitarity and regularity, extra relations denominated PT
invariance and crossing symmetry, PT-Symmetry:
t1 t2
().
P12 R12 ()P12 = R12
(A.1)
Crossing-symmetry:
R12 () =
1 t
1
[]
2
(n 1) ln[q] V 1 ,
V R12
[(n 1) ln[q] ]
(A.2)
where () is a normalization function and V is the following crossing matrix

1
V = antidiag q (2n3)/2, q (2n5)/2, . . . , , 1, 1, q, . . . , q (2n5)/2, q (2n3)/2 .

q
(A.3)
Here we find convenient to normalize the original Jimbos R-matrix by an overall factor
e2 q n and the function () is given by

e
q (n1)
.
(A.4)
() = e e
e
q (n1)
Next we present the simplest relations derived from the reflection equation we used in
Section 2. For sake of simplicity we shall use the following notation Yi (x) Yi , Yi (y)
Yi0 , wj (x y) wj , wj (x + y) = wj0 . Considering this notation, the relations we have
selected from the reflection equation are given by

w20 w3 Y10 Y2 + w4 Y1 Y20 w5 Y10 Y5 + w6 Y1 Y60

(A.5)
= w2 w40 Y10 Y1 + w50 Y20 Y5 + Y2 Y60 + w30 Y20 Y2 + Y5 Y60 ,
735

w20 w3 Y10 Y3 + w4 Y1 Y30 + w6 Y1 Y50 w5 Y10 Y6

= w2 w40 Y10 Y1 + w50 Y3 Y50 + Y30 Y6 + w30 Y30 Y3 + Y50 Y6 ,

w20 w3 Y30 Y4 w4 Y3 Y40 + w5 Y4 Y50 w6 Y40 Y6

= w2 w30 Y40 Y4 + w60 Y3 Y50 + Y30 Y6 + w40 Y30 Y3 + Y50 Y6 ,

w50 w5 Y60 Y6 Y5 Y50 + w3 Y20 (Y5 Y6 ) + Y2 Y60 Y50

= w30 w3 Y60 Y5 Y6 Y50 + w5 Y20 Y6 Y5 + Y3 Y60 Y50 ,

w30 w5 Y30 Y3 Y2 Y20 + w3 Y5 Y30 Y20 + Y50 Y2 Y3

= w50 w3 Y2 Y30 Y3 Y20 + w5 Y50 Y3 Y2 + Y6 Y30 Y20 ,

w60 w6 Y60 Y6 Y5 Y50 + w4 Y20 (Y5 Y6 ) + Y2 Y60 Y50

= w40 w4 Y60 Y5 Y6 Y50 + w6 Y20 (Y6 Y5 ) + Y3 Y60 Y50 ,

w40 w6 Y30 Y3 Y2 Y20 + w4 Y5 Y30 Y20 + Y50 (Y2 Y3 )

= w60 w4 Y30 Y2 Y3 Y20 + w6 Y50 (Y3 Y2 ) + Y6 Y30 Y20 .
(A.6)
(A.7)
(A.8)
(A.9)
(A.10)
(A.11)
The functions wj () are some of the Boltzmann weights of D22 model and are given
by [22]

e
q
,
w2 () = e e
q
e

1
e
q
1
e + 1 ,
(A.12)
w3 () = q
2
q
q
e

q
1
1
e
e + 1 ,
w4 () = q
2
q
q
e

q
1
1
e
e + 1 ,
(A.13)
q
w5 () =
2
q
q
e

q
1
1
e
e 1 .
(A.14)
w6 () = q
2
q
q
e
Appendix B. A new D 22 R-matrix
We begin by presenting the new D22 R-matrix

2 X

E E + q e2 1 e2 q 2
R() = e2 q 2
E E
6=, 0
or 6=2,3
6=2,3
(q 2 1)(e2 q 2 )
"
e + 1
X
+e e + 1
<2
=2,3
X
>3
=2,3
(E E + E 0 0 E 0 0 )
736
(q 2 1)(e2 q 2 )
"
1 e
+e e 1
<2
=2,3
X
(E E 0 + E 0 0 E 0 )
>3
=2,3
a ()E E 0 0
,6=2,3
b+ ()E E 0 0 + b+ ()E 0 0 E
6=2,3
=2,3
+ b ()E E 0 + b ()E 0 E
c+ ()E E 0 0 + c ()E E
=2,3
+ d + ()E 0 E 0 + d ()E 0 E 0

f () E 0 E + E E 0 E E 0 E 0 E ,
=2,3
(B.1)
where E are the elementary 4 4 matrices and we set 0 = 5 . The Boltzmann
weights are given by
2
a11 () = a44() = q 2 e2 1 ,

(B.2)
a14 () = a41()e2 = (q 1) q 2 1 e2 + q ,

q 3/2 2
q 1 e2 1 e 1 ,
2

q 1/2 2
q 1 e2 1 e q 2 ,
b1 =
2
1/2

q
e q 2 1 e2 1 e q 2 ,
b4 =
2
b1 =

q 1/2 2
e q 1 e2 1 e 1 ,
b4 =
2
2
2

2
e 2
e 2
f () =
q 1 e 1 ,
e 1 q2 1 ,
d =
4
4

e 2
q 1 e 1 e 3 + q 2 1 + 3q 2
4

+ q e2 1 e2 q 2 .
(B.3)
(B.4)
(B.5)
c =
(B.6)
This R-matrix has additional Boltzmann weights, the last term in Eq. (B.1), as compare
to the standard D22 R-matrix [22]. In addition, several other weights have also a different
737
functional dependence on the spectral parameter . For periodic boundary conditions,

such differences are not important since we verified, by using the algebraic Bethe ansatz
approach [29], that the corresponding Bethe ansatz equations and eigenvalues are the same
as those found for the standard D22 model [29,30]. This result is a strong indication that
indeed the R-matrix (B.1) can be obtained by twisting the usual D22 R-matrix. However,
the situation for open boundary conditions turns out to be a bit different. In fact, we did not
find any diagonal solution of the corresponding reflection equation. The basic K-matrices
are non-diagonal and we managed to find two classes of such solutions. The first family
depends only on a discrete parameter = and is given by

(1,)
(1,)
(B.7)
Y1 (, ) = Y4 (, ) = e2 + q ,

(1,)
(1,)
1
2
(B.8)
Y2 (, ) = Y3 (, ) = 2 (1 + q) 1 + e ,

(B.9)
Y5(1,)(, ) = Y6(1,)(, ) = 12 (1 q) 1 + e2
while the second family has an extra continuous parameter

(2,)
Y1 (, ) = e2 2 q + e2 e ,

Y4(2,)(, ) = 1 2 e2 q + e2 e ,

Y2(2,)(, ) = 12 1 e + e 1 + e2 ( + q),

(2,)
Y3 (, ) = 12 1 + e + e 1 + e2 ( + q),

Y5(2,)(, ) = 12 e2 1 ( q) e e 1 ,

(2,)
Y6 (, ) = 12 e2 1 ( q) + e e + 1 .
(B.10)
(B.11)
(B.12)
(B.13)
(B.14)
(B.15)
Finally, we remark that since this new R-matrix is only unitary, the associated K+ ()
matrices cannot be directly obtained by an isomorphism of the type described in (20).
However, as shown in Ref. [36], unitarity is a sufficient condition to allow one to construct
commutative transfer matrices leading to open spin chains. In this case one has to solve an
extra reflection equation to obtain the K+ () matrix [36].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
E.K. Sklyanin, J. Phys. A: Math. Gen. 21 (1988) 2375.

L. Mezincescu, R.I. Nepomechie, J. Phys. A: Math. Gen. 24 (1991) L17.
L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. A 7 (1991) 5231.
H.J. de Vega, A. Gonzalez-Ruiz, J. Phys. A: Math. Gen. 26 (1993) L519; J. Phys. A: Math.
Gen. 27 (1994) 6129.
B.Y. Hou, R.H. Yue, Phys. Lett. A 183 (1993) 169.
T. Inami, H. Konno, J. Phys. A: Math. Gen. 27 (1994) L913.
C.M. Yung, M.T. Batchelor, Nucl. Phys. B 435 (1995) 430.
M.T. Batchelor, V. Fridkin, A. Kuniba, Y.K. Zhou, Phys. Lett. B 376 (1996) 266.
R.E. Behrend, P.A. Pearce, D.L. OBrien, J. Stat. Phys. 84 (1996) 1.
C. Ahn, W.M. Koo, Nucl. Phys. B 468 (1996) 461.
Y.K. Zhou, Nucl. Phys. B 468 (1996) 504.
T. Inami, S. Odake, Y.-Z. Zhang, Nucl. Phys. B 47 (1996) 419.
738
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
C. Ahn, C.-K. You, Sol-int/9710024.

C.X. Liu, G.X. Ju, S.K. Wang, K. Wu, J. Phys. A 32 (1999) 3503.
J. Abad, M. Rios, Phys. Lett. B 352 (1995) 92.
A. Lima-Santos, Nucl. Phys. B 558 (1999) 637.
A. Gonzalez-Ruiz, Nucl. Phys. B 424 (1994) 468.
Z.-N. Hu, F.-C. Pu, Y. Wang, J. Phys. A: Math. Gen. 31 (1998) 5241.
A.J. Bracken, X.-Y. Ge, Y.-Z. Zhang, H.-Q. Zhou, Nucl. Phys. B 516 (1998) 588.
H.-Q. Zhou, X.-Y. Ge, J. Links, M.D. Gould, Nucl. Phys. B 546 (1999) 779.
H. Frahm, V. Slanov, J. Phys. A: Math. Gen. 32 (1999) 1547.
M. Jimbo, Commun. Math. Phys. 102 (1986) 537.
R. Cuerno, A. Gonzales-Ruiz, J. Phys. A: Math. Gen. 26 (1993) L605.
J.R. Links, M.D. Gould, Int. J. Mod. Phys. B 10 (1996) 3461.
V. Pasquier, H. Saleur, Nucl. Phys. B 330 (1990) 523.
F.C. Alcaraz, M.N. Barber, M.T. Batchelor, R.J. Baxter, G.R.W. Quispel, J. Phys. A: Math.
Gen. 20 (6397) 1987.
H. Asakawa, M. Suzuki, J. Phys. A: Math. Gen. 29 (1996) 225.
M. Shiroishi, M. Wadati, J. Phys. Soc. Jpn. 66 (1997) 1.
M.J. Martins, Phys. Rev. E 59 (1999) 7220.
N.Yu. Reshetikhin, Lett. Math. Phys. 14 (1987) 125.
S. Artz, L. Mezincescu, R. Nepomechie, J. Phys. A: Math. Gen. 28 (1995) 5131.
V. Chari, A. Pressley, A Guide to Quantum Groups, Cambridge University Press, 1994.
R.I. Nepomechie, J. Phys. A 33 (2000) L21.
N.Yu. Reshetikhin, Lett. Math. Phys. 20 (1990) 331.
A. Foerster, J. Links, I. Roditi, J. Phys. A: Math. Gen. 31 (1998) 687.
Y.-X. Chen, X.-D. Luo, K. Wu, J. Phys. A: Math. Gen. 32 (1999) 757.

Macroscopic and microscopic (non-)universality

of compact support random matrix theory
G. Akemann a , G. Vernizzi b
a Max-Planck-Institut fr Kernphysik, Saupfercheckweg 1, D-69117 Heidelberg, Germany
b The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
Abstract
A random matrix model with a -model like constraint, the restricted trace ensemble (RTE),
is solved in the large-n limit. In the macroscopic limit the smooth connected two-point resolvent
G(z, w) is found to be non-universal, extending previous results from monomial to arbitrary
polynomial potentials. Using loop equation techniques we give a closed though non-universal
expression for G(z, w), which extends recursively to all higher k-point resolvents. These findings are
in contrast to the usual unconstrained one-matrix model. However, in the microscopic large-n limit,
which probes only correlations at distance of the mean level spacing, we are able to show that the
constraint does not modify the universal sine-law. In the case of monomial potentials V (M) = M 2p ,
we provide a relation valid for finite-n between the k-point correlation function of the RTE and the
unconstrained model. In the microscopic large-n limit they coincide which proves the microscopic
universality of RTEs. 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.60.Kz; 05.40.-a; 02.10.Sp
Keywords: Large-N matrix model; Microscopic universality; Macroscopic non-universality; Restricted trace
ensemble
1. Introduction
Random matrix models enjoy a wide range of applications in physics due to their
property of being universal (for reviews see [1,2]). This property manifests itself in the
independence of correlation functions from the choice of the distribution function P(M)
exp[n Tr V (M)], where M is an n n matrix and V is a polynomial. Different classes
of universality are found depending on the way the large-n limit is taken [35] and which
part of the spectrum is investigated [58]. However, not in all applications a distribution
function of the above form is realistic. Here, all the eigenvalues (= energy levels) of the
matrix M are coupled through the Jacobian after diagonalization. There are situations as
for example in applications in nuclear physics, where in contrast to that the Hamiltonian
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 2 5 - 4
740
G. Akemann, G. Vernizzi / Nuclear Physics B 583 [FS] (2000) 739757
of the model couples only few energy levels and is still very well described by the
above random matrix model [1]. It is therefore very interesting to study deformations and
generalizations of the above distribution P(M) and to investigate first, if the correlations
remain unchanged and second, if the property of universality is maintained. In this work we
study a deformation which preserves the symmetry of the matrix model, which will be the
unitary transformations of the hermitian matrix M in our case. The symmetry of the model
is directly related to the properties of the Hamiltonian under rotations and time-reversal
[1]. There exists an interesting relation [9] between the restricted trace ensembles (RTEs)
which we will consider and the so-called Wigner ensembles [10,11], where different matrix
elements are weighted with different distribution functions, without being invariant under
unitary transformations.
It has been known only quite recently that there exist symmetry preserving deformations
of the distribution function that destroy the property of universality [9,12]. Two examples
are the trace squared ensembles which were originally introduced in the context of quantum
gravity [13,14] and the generalized RTEs which were introduced, in their simplest purequadratic form, by Bronk and Rosenzweig [15,16] in the context of nuclear physics. The
non-universality does not necessarily spoil the applicability to physical systems. In fact
the deformation may be introduced for physical reasons: indeed, trace-squared terms have
been recognized as corresponding to higher order intrinsic curvature terms in the string
action. Such terms have been added in order to cure an intrinsic instability of the theory
related to a crumpled surface phase of the string world sheet [13,14,17]. Moreover, in
the framework of random surfaces, these additional terms are interpreted as touching
interaction terms which make the random surfaces touch each other. The generalized RTEs
permit the same graphical interpretation as they have been shown to be a limiting case
of a special trace squared ensemble [18]. The RTEs are defined such that the exponential
weight function gets replaced by a constraint P (M) (A2 1/n Tr V (M)) (or the
Heaviside step function instead). By using the following representation of the -function,
(x) = liml / l exp[lx 2], the distribution P (M) can be easily seen to contain
trace squared terms. The relation between RTEs and the trace squared ensembles was
discussed in great detail for the spectral density in [18] using saddle point techniques,
and the canonical ensemble, the trace squared ensemble and the RTE were shown to agree
to leading order. In [12] also the two-point function was derived for the trace squared
ensembles and shown to be non-universal using saddle point equations. Here, we will
present a scheme to calculate general k-point functions of the RTEs in different large-n
regimes.
Namely, one has to distinguish two different types of large-n limits the macroscopic
and microscopic limit which may not all be affected by the deformation of the
distribution function P(M). In the example of generalized RTEs which we will study here
we find that this is precisely the case. While the macroscopic universality is destroyed by
the global constraint, the microscopic correlations at short distances remain unchanged.
In a sense, the canonical ensemble P(M) is replaced by its micro-canonical counterpart
P (M). Therefore we have an explicit model of statistical mechanics at hand, where
the correlation functions of both ensembles can be calculated analytically and then be
741
compared for discrepancies. Another remarkable property of RTEs is that they possess
a finite support already at finite n. Similar to the canonical Gaussian matrix model the
RTEs allow for an explicit calculation at finite n. This has been shown already in [18]
for the one-and two-point function and will be given here for general k-point functions.
Comparing the finite-n and n results the differences can be thought of as finite-size
corrections, when interpreting n as the continuum limit. In the RTE these corrections
appear in a different way than in the canonical model, due to the finite support at finite-n.
Let us explain now in more detail how the large-n limit can be taken.
In the macroscopic large-n limit no restrictions are made on the distance between
different eigenvalues. This leads to smooth, universal two- and higher k-point correlation
functions for the canonical, unconstrained models [3,4]. Applications can be found in
two-dimensional quantum gravity [19] as well as the theory of transport properties of
mesoscopic wires [2]. For the generalized RTEs we have shown in a previous publication
together with our collaborators [9], that for a certain class of potentials V (M) the 2-point
correlator is no longer universal. In this work we will extend these results to arbitrary
polynomial potentials (see also [20]) and to all higher k-point resolvents. Since in Ref. [9]
it was also shown that to leading order all k-point resolvents of the - and -measure are
equivalent we will restrict ourselves here to the former one.
In contrast to that, in the microscopic large-n limit correlations of eigenvalues ,
at a distance of the mean level spacing | | 1/n are calculated. It is this kind of
limit that finds a wide range of applications in nuclear physics, condensed matter physics
(see, e.g., [1]) and has initiated exact analytical solutions in the study of Dirac spectra
in QCD [21]. We will show that in generalized RTE with purely monomial potential
V (M) = M 2p (which includes of course the quadratic case for p = 1) the constraint does
not change the local properties at short distances, in the sense that the connected twopoint correlator behaves according to the well-known sine-law of the Gaussian canonical
ensemble [5]. Indeed, this result holds also for higher k-point correlation functions. This
does not come as a surprise since a global constraint should not change the local properties.
We believe that the same is true for more general potentials and that universality holds here
as well.
Consequently the present paper is split into two different parts. Section 2 is devoted
to the macroscopic large-n limit where we use loop equation techniques, closely following [22,23]. Although the loop equations are originally designed to calculate higher orders
in the 1/n-expansion we will restrict ourselves to the planar limit. Since we find nonuniversality to leading order in all connected correlation functions we do not calculate the
likewise non-universal higher orders in 1/n. The difficulty to deal with the constraint will
be treated in a similar way as the situation where the spectral density has a support consisting of several intervals [23]. In these multi-band phases additional constraints have to be
imposed to make the solution unique [24,25]. We will restrict ourselves to hermitian matrices M only, for non-hermitian matrices see [20]. Since only the planar solution is needed
for our results they can be easily extended to orthogonal and symplectic matrices using
[26]. Extensions to the complex matrix model are straightforward as well since the same
loop equation techniques exist in the literature [27,28]. Very recently finite-n results have
742
been obtained for the eigenvalue density of Gaussian ensembles with real symmetric and
complex matrices [29]. In the second part Section 3 the microscopic large-n limit is investigated for the RTE with a monomial potential V (M) = M 2p . Here, we generalize existing
results for finite-n of previous publication [18], by improving a technique already used in
Ref. [9]. Rescaling variables in the microscopic limit and using the inverse Laplace transform we are able to match the connected k-point correlator to the well known sine-law
behavior of the canonical ensemble, thereby proving the microscopic RTE universality. Let
us stress that for the RTEs no orthogonal polynomial techniques are applicable.
2. The macroscopic limit: non-universality

In order to calculate all correlation functions for the constrained matrix model with an
P
gj
j
arbitrary potential V (M) =
j =1 j M we introduce an auxiliary potential W (M) inside
the partition function 1
Z

Z DM exp[n Tr W (M)] n2 A2 n Tr V (M) ,
(2.1)
W (M)
X
tj
j =1
Mj ,
(2.2)
where the two sets of variables {tj } and ({gj }, A) are taken to be independent. All k-point
resolvent operators can then be obtained by taking functional derivatives of Z with respect
to W (p) as given below, where we then eventually set the auxiliary potential W to zero
at the end. A similar trick has been used in Ref. [12] in order to investigate a multi-trace
random matrix ensemble [13,14], showing their non-universality as well. Furthermore we
use the complex representation of the -function to obtain
Z
Z

d
DM exp n Tr W (M) + i V (M) A2 .
(2.3)
Z =
2
If we had used instead the following representation of the -function, (x) = liml / l
exp[lx 2 ], we would have obtained the trace squared ensemble: Pl (M) exp[2lnA2
Tr V (M) l Tr V (M)2 ], with the strength of the touching interaction being proportional
to l. However, it is not straightforward to derive and solve loop equations for such an
ensemble. In particular, we cannot directly employ the non-universality results of [12],
where it was crucial that the single- and multi-trace potentials were different. It is in the
form Eq. (2.3) that we can actually derive and solve the loop equations for the constrained
model. Throughout the paper the same notation as in [22] is used, which is redisplayed
here for completeness. The resolvent or 1-loop correlator is defined as

1 X hTr M k i
1
1
=
(2.4)
Tr
G(p)
n
pM n
pk+1
k=0
1 We added a trivial factor of n2 inside the delta function.
and higher k-point resolvents are given by

1
1
k2
Tr
Tr
G(p1 , . . . , pk ) n
p1 M
pk M ,conn
743
(2.5)
where conn stands for the connected part of the expectation value with respect to
Eq. (2.3). They are defined such that the leading part is of the order O(1). If we define
the free energy F as follows

(2.6)
Z exp n2 F
all resolvents can be obtained from it by successive applications of the loop insertion
operator
X j d
d
(p)
,
dW
pj +1 dtj
(2.7)
j =1
G(p1 , . . . , pk ) =
d
d
d
1
(pk )
(pk1 )
(p1 )F + k,1 .
dW
dW
dW
p1
(2.8)
In particular all higher resolvents can be derived from the 1-point resolvent alone.
2.1. The loop equation in the planar limit
The loop equation is derived in the usual way from the partition function Eq. (2.3) by
shifting variables M M + /(p M) and requiring it to be invariant under this shift,
i.e., dZ /d|=0 = 0,
I
0 ()
1 d
d Veff
G() = G(p)2 + 2
(p)G(p),
(2.9)
2i p
n dW
C
where we have defined the effective potential

(M).
Veff (M) = W (M) + iV
(2.10)
In Eq. (2.9) the integration contour C encircles the support [y, x] of the spectral density
() counterclockwise in the complex plane, not including the argument p
/ [y, x].
The parameter i inside the effective potential Veff is determined by the constraint
hTr V (M)i = nA2 as a function of all coupling constants, as we will see in more detail
below. This constraint can also obtained by requiring the invariance of Z under the shift
+ .
Let us stress again that the resolvents are given by differentiating with respect to W (p)
and not the effective potential Veff (p). Due to the -integral in the partition function Z
Eq. (2.3) we have hTr M k i 6= hi Tr M k i which is needed to determine G(p) Eq. (2.4).
Furthermore let us note that Eq. (2.9) looks almost identical to the loop equation of the
unconstrained hermitian matrix model [22] defined in the next section Eq. (3.1). However,
the role of the above mentioned auxiliary potential W as well as the constraint will modify
the results of [22]. The constraint leads to similar complications as in the situation where
the support of the spectral density consists of several intervals [23].
744
In order to solve the loop equation we introduce a 1/n2 expansion for all k-point
resolvent operators (2.5)
G(p1 , . . . , pk ) =
X
1
Gg (p1 , . . . , pk ),
n2g
(2.11)
g=0
where the leading part with genus g = 0 (planar) is of the order O(1). In Ref. [9] it was
shown that for monomial potentials the expectation values of the RTEs possess such an
expansion in 1/n2 as well. Here we assume that the same holds true for all polynomials
potentials. Inserting this expansion into the loop equation (2.9) and taking the large-n limit
we obtain to leading order
I
0 ()
d Veff
G0 () = G0 (p)2 .
(2.12)
2i p
C
If we make the ansatz that G0 (p) has just one cut in the complex plane or equivalently the
support of the eigenvalues consists of the single interval [y, x] we obtain
p

1 0
(p) M(p) (p x)(p y) ,
(2.13)
G0 (p) = Veff
2
where the analytic function M(p) is given by
I
V 0 ()
d
eff
.
(2.14)
M(p) =
2i ( p) ( x)( y)
C
For details of the derivation see for example Ref. [23]. The final result can be written as
follows, after deforming back the integration contour,
s
I
0 ()
1
d Veff
(p x)(p y)
.
(2.15)
G0 (p) =
2 2i p ( x)( y)
C
To make the solution complete we still have to determine the endpoints x and y as well as
the parameter i in terms of the coupling constants of W and V and the parameter A. The
first two equations can be obtained from the asymptotic behavior of G0 (p). According to
the definition (2.4) we have
1
(2.16)
lim G(p) = .
p
p
Since the leading term does not depend on n it comes from the planar part G0 (p) and we
obtain the conditions
I
0 ()
Veff
d
1
,
0=
2 2i ( x)( y)
C
1=
1
2
I
C
0 ()
Veff
d
.
2i ( x)( y)
(2.17)
The third equation needed we obtain from the constraint on Tr V (M) which we rewrite in
terms of the spectral density

1
lim G0 ( i) G0 ( + i)
2i 0
p
1
M() (x )( y).
=
2
The constraint then reads
Zx
Zx
p
d
2
M() (x )( y) V (),
A = d ()V () =
2
745
() =
(2.18)
(2.19)
which determines the parameter i contained in M(). This equation together with the
boundary conditions Eq. (2.17) determines the planar resolvent G0 (p) Eq. (2.4) completely
as a function of the coupling constants of W and V and of the parameter A.
2.2. Higher planar k-point resolvents
Starting from the the planar resolvent G0 (p) we can obtain all higher planar k-point
resolvents by successively applying the loop insertion operator d/dW to it, as it is
given in Eq. (2.8). Here we use the fact that all resolvents have the same expansion
in 1/n2 Eq. (2.11). For this purpose we introduce a set of new parameters Mk and
Jk , k N+ . These moments usually play the role of universal parameters of the higher
k-point resolvents encoding all the information of the potential in addition to the endpoints
x and y [22]. We will then rewrite the loop insertion operator in terms of these new
variables. This is done in order to make the successive application of d/dW to an algebraic
procedure. Finally we calculate explicitly the non-universal planar 2-point resolvent
G0 (p, q) and comment on the general situation.
Let us begin by defining
I
()
1
d 0
V ()
, ()
,
Mk
2i eff
( x)k
( x)( y)
C
I
()
d 0
V ()
, k = 1, 2, . . . .
(2.20)
Jk
2i eff
( y)k
C
Expanding the poles at x and y the moments can be explicitly written as functions of
dk1
1
M()|=x and similarly for y the
the coupling constants. Because of Mk = (k1)!
dk1
moments also characterize the multi-critical points of the model. We now rewrite the loop
insertion operator Eq. (2.7) in the following way
dx
dy
d i
d
(p) =
(p) +
(p)
+
(p)
+
(p)
dW
W
dW
x dW
y
dW
i

X
dJk
dMk
(p)
(p)
+
+
,
dW
Mk dW
Jk
(p)
W
k=1
X
j =1
j
.
pj +1 tj
(2.21)
(2.22)
746
dJk
k
While the dM
dV (p) and dV (p) can be obtained in a straightforward way from the definition
Eq. (2.20) (see, e.g., [22]) the remaining unknown quantities are derived by applying
d
dW (p) to the boundary conditions Eqs. (2.17) and (2.19) and solving a linear set of
equations. This is done in the Appendix A with the result reading

dx
(p)
4
G0 (p) (p) ,
(p) =
+
M1
dW
p x B(x y)

(p)
4
dy
(p) =
G0 (p) (p) ,
J1
dW
p y B(x y)

1
1 d i
(p) = G0 (p) (p) ,
(2.23)
i dW
B
where
Zx
1
lim
(2.24)
B i d V ()
2i 0
y

G0 ( i) ( i) G0 ( + i) ( + i) .
All quantities are expressed by elementary functions and the planar resolvent G0 (p)
Eq. (2.15) and we have set already the auxiliary potential W 0. The result for general W
can be derived from Eqs. (A.2) and (A.3).
We are now ready to apply the loop insertion operator d/dW in the form (2.21) to G0 (p)
Eq. (2.15), which does not explicitly depend on the moments:
d
(q)G0 (p)
dW

1
(q)
1
1
(q)
1
1 +
+
=
4(q p) (p) q x q y
2(p q)2 (p)
s
I
1
d V 0 () (p x)(p y)
d i
(q)
+
dW
2 2i p ( x)( y)
C

1
1
dx
dy
1
M1
(q) +
J1
(q) ,
(2.25)
+
4(p) p x
dW
p y dW
after performing some contour integrals. If we set W 0 we can use the results Eq. (2.23)
and finally obtain

1
G0 (p) (p) G0 (q) (q) ,
(2.26)
G0 (p, q) = Gcan
0 (p, q)
B
the planar connected 2-point resolvent of the constrained matrix model. The first part is the
well known universal 2-point resolvent of the corresponding unconstrained or canonical
matrix model (can) [3]

1
(p x)(q y) + (p y)(q x)
2
(2.27)
(p,
q)
=
Gcan
0
4(p q)2
(p x)(p y)(q x)(q y)
whereas the second part contains the non-universal terms G0 (p) and G0 (q). Still, the result
is given in closed form for an arbitrary polynomial potential V (M). Eq. (2.26) can be
compared with the earlier result [9]
G0 (p, q) =
G0 (p, q) = Gcan
0 (p, q)

1
p pG0 (p) q qG0 (q)
2p
747
(2.28)
for the special case of monomial potentials V (M) = M 2p , p N+ . For p = 1, 2 we have

checked explicitly that the two results Eq. (2.26) and Eq. (2.28) agree. The corresponding
resolvents can be found in [9] 2 and the parameter i was already determined in [18]. From
the procedure described above it is clear that also the higher k-point resolvents will remain
d
(pi )G0 (pj ) always contains terms proportional to
non-universal since the derivative dW
G0 (pi ) and G0 (pj ).
Let us also briefly comment on higher genus contributions. Expanding the loop equation
in 1/n2 together with Eq. (2.11) one obtains for genus one

(2.29)
K 2G0 (p) G1 (p) = G0 (p, p),
where
(p)
Kf
I
C
0 ()
d Veff
f ().
2i p
(2.30)
The right hand side of Eq. (2.29) is easily obtained from Eq. (2.26) by setting p = q.
However, it is no longer a rational function in contrast to Gcan
0 (p, p). Its non-universality
will then translate to G1 (p) after inverting the integral operator (K 2G0 (p)) and thus to
higher genera through
g1
X

d
(p)Gg1 (p), g > 1.
Gg 0 (p)Ggg 0 (p) +
K 2G0 (p) Gg (p) =
dW
0
(2.31)
g =1
For this reason we do not go through the tedious procedure of finding a basis for
(K 2G0 (p)) now including also square roots and inverting it.
3. The microscopic limit: universality

In this section we investigate correlations of eigenvalues at the distance of the mean
level spacing D 1/n, the so-called microscopic large-n limit. We will heavily exploit
the knowledge about correlations in the unconstrained canonical model for finite as well
as infinite n. Our main result is an explicit relation between the canonical and RTE
correlations for finite-n which then serves to determine the microscopic RTE correlations
and prove their universality.
Let us recall the known results about the canonical ensemble which also fixes our
notation. The partition function reads
Z

e(M) ,
Z DM exp n Tr V
(3.1)
2 In Eq. (2.18) of Ref. [9] the factor of 1/2p is missing.
748
e(M) is a polynomial. For the generalized RTE or micro-canonical ensemble

where V
Z

(3.2)
Z DM A2 n1 Tr V (M) ,
where we do not need to introduce an auxiliary potential in contrast to the previous section.
In order to obtain the same microscopic correlations for the two models we will eventually
e (as
have to relate the coupling constants of the respective polynomial potentials V and V
in the macroscopic limit, see [18]). The k-point density correlation function is defined as

(3.3)
(1 , . . . , k ) n1 Tr (1 M) n1 Tr (k M) ,
and similarly for the delta-measure. Its connected part conn is related in the following
way to the k-point resolvent defined in the previous section Eq. (2.5)

c (1 , . . . , k ) n1 Tr (1 M) n1 Tr (k M) conn ,

1
1 k
= 2k2
n
2i
X Y
lim
i G(1 + 1 i, . . . , k + k i),
(3.4)
0
i =
where we have not yet taken the large-n limit. In this section we deal with density
correlations instead of resolvents because they can be given more explicitly for finite-n.
Namely in the canonical model all density correlators can be expressed in terms of the
Kernel Kn (, ) of a set of orthonormal polynomials Pl () at finite-n [30]

(1 , . . . , k ) = det Kn (i , j ) ,
16i,j 6k
X
c
Kn (1 , 2 )Kn (2 , 3 ) Kn (k , 1 ),
(3.5)
(1 , . . . , k ) = (1)k+1
P
where the sum is taken over the (k 1)! distinct cyclic permutation P of the indices
(1, 2, . . . , k). The kernel and the polynomials are defined as
X
1 n e
e
Pk ()Pk (),
Kn (, ) = e 2 (V ()+V ())
n
k=0
Z
e
kl = d enV () Pk ()Pl ().
n1
(3.6)
The use of the orthogonal polynomial method is only possible for the canonical model
e(M)] inside the partition function Eq. (3.1) factorizes
because the measure exp[n Tr V
in terms of the eigenvalues i of the hermitian matrix M. For the RTE no such property
holds which forces us to seek for other methods. Here we will make use of homogeneity
properties for monomial potentials.
Let us finally give the universal results for the canonical ensemble in the microscopic
large-n limit, where we restrict ourselves to the origin scaling limit due to the local
translational invariance of the canonical ensemble. As mentioned in the beginning we
measure eigenvalues in units of the mean level spacing which is D = 1/(n(0)) at the
749
origin. Here (0) is the mean eigenvalue density at zero, taken in the macroscopic large-n
limit as given in Eq. (2.18). We then define new variables zi = i /D which are kept fixed
in the large-n limit. Since D 0 as n the variables i have to go to zero as well. In
this particular limit the microscopic correlators are defined as 3
S (z1 , . . . , zk ) lim (Dn)k (z1 D, . . . , zk D).
n
(3.7)
This limit, which is well behaved and finite, can be investigated by using the Darboux
Christoffel formula for the kernel Eq. (3.6) and the asymptotic large-n behavior of the
polynomials Pk () to obtain [5]
lim DnKn (z1 D, z2 D) =
sin((z1 z2 ))
.
(z1 z2 )
(3.8)
e(). Together
This is the universal sine-law which is valid for all polynomial potentials V
with Eqs. (3.5) and (3.7) it completely determines all k-point density correlators, connected
and not connected, where at coinciding arguments we have limn DnKn (zD, zD) = 1,

sin((zi zj ))
,
(3.9)
S (z1 , . . . , zk ) = det
16i,j 6k
(zi zj )
and similarly for Sc (z1 , . . . , zk ).
We conclude with the following remark. If we had taken the macroscopic large-n limit
instead, the connected correlators c (z1 , . . . , zk ) had been of the order O(1/n2k2) as one
can see from Eq. (3.4) together with the fact that the (connected) resolvents are of order
O(1). However, when taking the microscopic limit keeping the zi fixed, the asymptotic
kernel Eq. (3.8) is of order O(1) and hence are the connected and not connected microscopic k-point correlators from Eq. (3.5). Consequently, the knowledge of both connected
and not connected correlators, is equivalent here since they can be obtained from each
other by adding or subtracting l-point correlators (l < k) of order O(1). In that sense the
microscopic large-n limit modifies the usual large-n factorization of correlation functions.
3.1. Microscopic k-point correlation functions
As it has been mentioned already the correlation functions in the RTE cannot be
calculated using orthogonal polynomials because of the Dirac -function in the measure.
However, in the particular case of purely monomial potentials V (M) = M 2p , the
evaluation of the connected k-point correlator in the microscopic limit is straightforward.
We shall exploit some homogeneity properties of this case, which make it possible to relate
ensemble averages in the monomial RTE to ensemble averages of the same quantities in
e(M) = gM 2p . Let us define a function Fk (M; )
the canonical ensemble 4 with potential V
3 The factor nk appears as we had already defined the k-point correlator in Eq. (3.3) to be normalized to
unity. The appropriate unfolding procedure is usually defined for un-normalized correlators (see, e.g., [1]), which
provides us with the correct pre-factor 1/(0)k .
4 This technique is a slight generalization of Ref. [9], where only the macroscopic limit was investigated.
750
of the matrix M and of a set of parameters = (1 , . . . , k ), that satisfies the following

homogeneity property under a rescaling of the matrix

(3.10)
Fk (tM; ) = t a Fk M; t b for t, a, b R.
A simple example for such a function is the operator Tr (M)/n which has a = b = 1.
Also any such product is a homogeneous function, as inside the average of the k-point
correlation function Eq. (3.3), having a = k and b = 1. In Appendix B we derive the
following formula for such homogeneous functions Fk , which relates their canonical and
RTE average
a
"
#
b
n2
gn2 2p
(gn2 ) 2p ( 2p
)

1 hFk (M; ( t ) )i
L
(3.11)
A2 .
hFk (M; )i =
n2
n2 +a
2
Ap
t 2p
Here L1 [h(t)](x) is the inverse Laplace transform of a function h(t), evaluated at the
point x > 0 (for an integral representation see Eq. (B.5)). Eq. (3.11) holds for any finite n.
If we choose the 1- or 2-point correlator from Eq. (3.3) as an example we reproduce the
finite-n results of Ref. [18] for () and (, ) which were given in the case of a
Gaussian potential p = 1. If we choose for general k to take hFk (M; )i = (1 , . . . , k ),
which obviously fulfills the criterion (3.10), we obtain from Eq. (3.11) the following
expression for the k-point RTE correlator
k
"
#
n2
gn2 1
gn2 1
2p
2p
(gn2 ) 2p ( 2p
)

1 (( t ) 1 , . . . , ( t ) k )
2
L
. (3.12)
A
(1 , . . . , k ) =
2
2
n
n k
2
p
2p
A
t
Now we make use of the fact that at finite n the correlations (1 , . . . , k ) from Eq. (3.5)
can be written as a polynomial in all variables i times an exponential measure factor:
2n2
X
Pk
2p
i=1 i
(1 , . . . , k ) eng
(n)
l1 ,...,lk =0
c{l1 ,...,lk } 11 kk .
(3.13)
Due to this fact we can actually perform the inverse Laplace transformation
2
n
( 2p
)
(1 , . . . , k ) =
(gn2 ) 2p A
1
L
=
A2
h
t
n2
p 2
1 X 2p
A
i
n
k
n2
2n2
X
i=1
(n)
c{l1 ,...,lk } l11
1
(gn2 ) 2p
l1 ,...,lk =0
A2
n2 k6i li
2p
l11 kk
2p
i=1 i
(gn2 ) 2p A p

n2
2p
l1 ,...,lk =0
6l
(gn2 ) 2p i i
Pk
k
c{l1 ,...,lk }
n2 k6i li
2p
1
n
(n)
2n2
X
k
1X
kk
6i li
! n2 k6i li 1
2p
2p
(3.14)
i=1
where we have used the shift property L1 [h(t)e t ](x) = L1 [h(t)](x + ), the linearity
of the inverse Laplace transform and Eq. (B.7) of Appendix B. Eq. (3.14) is our first result,
751
the finite-n k-point correlation function for RTEs with monomial potential V (M) = M 2p in
e(M) = gM 2p .
terms of the corresponding canonical correlator at finite-n with potential V
In general, Eqs. (3.13) and (3.14) are different from each other at finite-n. This remains
true in the macroscopic large-n limit, as we have seen in the previous section. In the
remaining part we will show that in the microscopic large-n limit, however, they happen to
coincide.
In a first step, we determine the mean level spacing D = 1/(n (0)) in order to define
the appropriate microscopic scaling limit. From Eq. (3.14) at = 0 one can read off (0),
since then the sum over all li collapses. We obtain
(0) =
(n)
c{0}
n
( 2p
)
(gn2 A2 )
1
2p
2 1
( n2p
)
c{0}
(3.15)
(2pgA2 ) 2p
for its large-n value, with c{0} = (0) being the macroscopic large-n limit of the canonical
spectral density Eq. (2.18) at the origin (which exists and is finite). In order to have the same
mean level spacing as in the canonical ensemble we would have to set 2pgA2 = 1. This
identification of coupling constants occurs also in the macroscopic large-n limit in order
to match the corresponding macroscopic spectral densities (see Refs. [9,18]). However,
since we measure all correlations in units of D and D respectively we do not need to
identify D = D since they drop out in the microscopic correlators anyway, as we will see
below.
We will now take the microscopic limit analogue to Eq. (3.7) with rescaling i = zi D
of our finite-n relation (3.14). The large-n limit of the different factors can be obtained as
follows, starting with the -function term in Eq. (3.14)
!
k

2pgA2 X zi 2p
2
(3.16)
A2 = 1.
A 2p+1
c{0}
n
i=1
The remaining A-dependent terms yield
2
np +2
A
2
2pgA2
n2p+1
!n
k
X
zi 2p
c{0}
2 k6 l
i i
2p
i=1
!

k
2pg X zi 2p
n2 k 6i li
1 2p+1
=A
+
1
2p
c{0}
n
i=1

1
p1 (k+6i li )
1 + O 2p1
,
(3.17)
=A
n
which still has to be evaluated under the sum over li s. Here we have used in the second
step that the first factor inside the parenthesis is of O(n2 ) since
p1 (k+6i li )
k
X
li 6 k(2n 2),
(3.18)
i=1
is at most of O(n). Because of p > 1 the corrections are sub-leading. The factor containing
-functions we evaluate together with the explicit factors of n in Eq. (3.14)
752

2
n
( 2p
)
1
(gn2 ) 2p
(k+6i li )
= (2pg)
n
( 2p
(k+6i li )
)
2p
X
1
li
1
k+
2
4pn
k
1
2p
(k+6i li )
!
2p + k +
i=1
k
X
li + .
(3.19)
i=1
It remains to be shown that the second term and thus higher terms in the expansion are
sub-leading. A naive counting from Eq. (3.18) suggests that this might not be the case. In
order to use the microscopic results for the canonical correlators Eq. (3.9) we put together
our results obtained so far
lim (D n)k (z1 D , . . . , zk D )
(2pgA2 ) 2p
= lim
n
(c{0} )k
2pgA2
= lim
1
(c{0} )k
l1 ,...,lk =0
k
X
1
4pn2
(n)
c{l1 ,...,lk }
1
4pn2
!
1
2p (k+6i li )
l1 ,...,lk =0
2p +
zi zi
i=1
2n2
X
1
1

zk (2pgA2 ) 2p lk
z1 (2pgA2 ) 2p l1
nc{0}
nc{0}
!
!
!
k
k
X
X
li 2p + k +
li +
k+
2n2
X
(n)
c{l1 ,...,lk }
z1
nc{0}
l1
i=1
k
X
i=1
i=1
zi zi +

zk
nc{0}
lk
.
(3.20)
Here, the pre-factors from Eqs. (3.17) and (3.19) have canceled with the factors of
1
(2pgA2 ) 2p from the unfolding. Now we know from the canonical ensemble Eqs. (3.13),
(3.7) and (3.9) that the limit of the sum over the li exists and is finite:
S (z1 , . . . , zk ) = lim
1
(c{0} )k
2n2
X
l1 ,...,lk =0

(n)
c{l1 ,...,lk }
z1
nc{0}
l1
zk
nc{0}
lk
.
(3.21)
Hence the term in Eq. (3.20) proportional to 1/(4pn2 ) is indeed sub-leading and we have
as a final result
,S (z1 , . . . , zk ) = S (z1 , . . . , zk ),
(3.22)
or more explicitly in terms of Eq. (3.9). Since our derivation holds for the RTE with
an arbitrary monomial potential V (M) = M 2p , we have not only derived all k-point
correlation functions but also proved their universality for the given class of potentials.
Let us finally point out that the equivalence Eq. (3.22) also holds for the corresponding
connected k-point correlation functions. As we have mentioned already at the end of the
previous subsection, they are of the same order in the microscopic limit and they can be
obtained from each other by adding or subtracting lower l-point correlators.
753
4. Conclusions
We have shown for RTEs as an example of constrained random matrix models that in
one and the same model correlation functions may exhibit universal and non-universal
behavior in different large-n regimes. In particular, in the macroscopic large-n limit all
planar connected k-point resolvents are non-universal and a closed expression was given
for the resolvent G0 (z, w) for an arbitrary potential. Hence when switching from the
canonical to the RTE or micro-canonical ensemble, the delta-function constraint destroys
the macroscopic universality.
A different behavior appears in the study of correlations of eigenvalues at the scale
of the mean level spacing 1/n. Here we recover the sine-law of the canonical ensemble
and prove its universality for the class of RTEs with monomial potential. This leads us to
conjecture that microscopic universality holds also for more general RTEs. The result in
the microscopic limit is not unexpected since a global constraint should not influence the
local statistics of eigenvalues.
Acknowledgements
We wish to thank G. Cicuta and L. Molinari for their very enjoyable collaboration and
many discussions on work done prior to this publication. Furthermore one of us (G.A.)
wishes to thank the Physics Department of Parma for its hospitality extended to him on
several occasions. The work of G.V. is supported in part by MURST within the project of
Theoretical Physics of fundamental Interactions.
Appendix A. The functions
dy
dx
dW (p), dW (p)
and
d i
dW (p)
d
(p) Eq. (2.21) to the boundary conditions
In this appendix we apply the operator dW
dy
dx
(p), dW
(p)
Eqs. (2.17) and (2.19) and solve the linear set of equations for the quantities dW
d i
and dW (p). Using the identity
(p)W 0 () =
,
W
(p )2
(A.1)
we obtain
I

1
d i
dx
dy
d 0
(p) + J1
(p) +
V ()()
(p),
M1
0 = p (p) +
2
dW
dW
2i
dW
C

1
dx
dy
xM1
(p) + yJ1
(p)
0 = p p(p) +
2
dW
dW
I
d
i
d
V 0 ()()
(p),
+
2i
dW
C
from Eq. (2.17) and
(A.2)
754
0=

(p)
(p)
dx
dx
C
D
(p)
(p)
M1
+
J1
2
dW
px
2
dW
py
I
Zx
p
d
d V 0 ()
d i
(p)
V () (x )( y)
()
+
dW
2
2i
y
1
2
I
C
d V 0 ()
2i p
( x)( y)
,
(p x)(p y)
(A.3)
from Eq. (2.19) after some calculation. Here we have introduced

Zx
C
y
Zx
D
y
d V () p
(x )( y),
2 x
d V () p
(x )( y).
2 y
(A.4)
d
(p) to M() which is then no
We note that special care has to be taken when applying dW
longer an analytic function in . Therefore the contour C has to be deformed in Eq. (2.14)
d
(p)
to contain only the pole and cut in the integrand, and not the new pole introduced by dW
(see also Appendix A in [23]). The linear set of equations Eq. (A.2) and Eq. (A.3) simplifies
() in that case. It is this limit
considerably when setting W 0 because of Veff () = iV
which we will need in order give a closed final expression for the planar 2-point resolvent
of the pure delta-measure without the auxiliary potential in Eq. (2.1). Using Eq. (2.17) it
follows
(p)
(p)
dx
dx
(p)
+ J1
(p)
,
dW
px
dW
py

(p)
(p)
x
dx
dx
y
2 d i
(p)
(p)
(p),
0=
M1
+
J1
+
2
dW
px
2
dW
py
i dW

(p)
(p)
D
dx
dx
C
(p)
(p)
M1
+
J1
0=
2
dW
px
2
dW
py
0 = M1

1
1 2 d i
A
(p) +
G0 (p) (p) .
i
dW
i
(A.5)
This can be easily solved for the desired quantities. We obtain

M1

(p)
4
dx
(p) =
+
G0 (p) (p) ,
dW
p x B(x y)
J1

(p)
4
dy
(p) =
G0 (p) (p) ,
dW
p y B(x y)

1
1 d i
(p) = G0 (p) (p) ,
i dW
B
(A.6)
as given in Eq. (2.23) where the following abbreviation has been introduced

C+D
2
.
B = i A + 2
x y
755
(A.7)
One can easily convince oneself that it equals the form given in Eq. (2.24).
Appendix B. RTE via inverse Laplace transform

In this appendix we derive Eq. (3.11) which expresses expectation values with respect
to the delta-measure in terms of averages with respect to the canonical measure Eq. (3.1)
using the inverse Laplace transform. Let Fk (M; ) be a function of the hermitian n n
matrix M and of the set of parameters = (1 , . . . , k ), such that it satisfies the
homogeneity property

(B.1)
Fk (tM; ) = t a Fk M; t b
for some real t, a and b. In other words Fk is a homogeneous function of degree a with
respect to matrix elements Mij and of degree (b) with respect to each parameter i . An
example for such a function is the operator inside the average of Eq. (3.3). The matrix
integral considered in this appendix is:
Z

(B.2)
I [Fk ] = DM A2 n1 Tr M 2p Fk (M; ),
with p an integer number. It is proportional to the
R average hFk i . Introducing the complex
representation of the delta function 2(x) = dy exp[ixy] and then scaling all matrix
elements by a factor (gn2 /(iy + 0+ ))1/2p , Eq. (B.2) reads:
Z+
I [Fk ] =

n2 Z

1
2p
2p
dy iyA2
gn2
gn2
ng Tr[M 2p ]
e
Fk
M; ,
DM e
2
iy + 0+
iy + 0+
(B.3)
where we have interchanged integrals. By using the homogeneity property (B.1) we can
rewrite the matrix integral as a canonical ensemble average, up to the normalization factor
Z Eq. (3.1):
Z+
I [Fk ] = Z
= Z gn

n2 +a
b
2p
2p
dy iyA2
gn2
gn2
e
F
M;
k
+
+
2
iy + 0
iy + 0
n2 +a
2p
hFk (M; ( gnt ) 2p )i
n2 +a
2p

A2 .
(B.4)
t
In the last step we used the complex representation of the inverse Laplace transform, i.e.,
1
L1 [h(t)](x) =
2i
+i+0
Z +
dt et x h(t),
i+0+
x > 0.
(B.5)
756
In order to obtain the correct normalization of the delta-average we evaluate Eq. (B.4)
in the case F = 1 (with a = b = 0):
n2
n2 A p 2
Z,
I [1] = gn2 2p
n2
( 2p
)
where we used the formula

1
x
1
(x),
(x)
=
L
( + 1)
t +1
(B.6)
Re( ) > 1.
(B.7)
Finally, by putting together Eqs. (B.4) and (B.6), the ensemble average hFk i I [Fk ]/I [1]
with respect to the measure (A2 Tr[M 2p ]/n) reads:
b
a
2

gn2 2p

(gn2 ) 2p
n
1 hFk (M; ( t ) )i
A2 ,
(B.8)
L
hFk i = n2
2 +a
n
2p
2
Ap
t 2p
which is just Eq. (3.11).
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]
T. Guhr, A. Mller-Groeling, H.A. Weidenmller, Phys. Rep. 299 (1998) 190.

C.W. Beenakker, Rev. Mod. Phys. 69 (1997) 731.
J. Ambjrn, J. Jurkiewicz, Y. Makeenko, Phys. Lett. B 251 (1990) 517.
G. Akemann, J. Ambjrn, J. Phys. A 29 (1996) L555.
E. Brzin, A. Zee, Nucl. Phys. B 402 (1993) 613.
M.J. Bowick, E. Brzin, Phys. Lett. B 268 (1991) 21.
E. Kanzieper, V. Freilikher, Phys. Rev. E 55 (1997) 3712.
G. Akemann, G.M. Cicuta, L. Molinari, G. Vernizzi, Phys. Rev. E 60 (1999) 5287.
A. Khorunzhy, B. Khoruzhenko, L. Pastur, J. Phys. A 28 (1995) L31; J. Math. Phys. 37 (1996)
5033.
J. DAnna, A. Zee, Phys. Rev. E 53 (1996) 1399.
S. Iso, A. Kavalov, Nucl. Phys. B 501 (1997) 670.
S.R. Das, A. Dhar, A.M. Sengupta, S.R. Wadia, Mod. Phys. Lett. A 5 (1990) 1041.
G.M. Cicuta, E. Montaldi, Mod. Phys. Lett. A 5 (1990) 1927.
N. Rosenzweig, Statistical Physics, Brandeis Summer Institute, 1962, G. Uhlenbeck et al.,
Benjamin, 1963.
B.V. Bronk, Ph.D. thesis, Princeton University, 1964, as cited in [30], Chapter 19.
G.P. Korchemsky, Mod. Phys. Lett. A 7 (1992) 3081.
G. Akemann, G.M. Cicuta, L. Molinari, G. Vernizzi, Phys. Rev. E 59 (1999) 1489.
J. Ambjrn, B. Durhuus, T. Jonsson, Quantum Geometry, Cambridge University Press, 1997.
G. Vernizzi, Classes of universality in random matrix theory, Ph.D. thesis, Parma, 1999.
E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 (1993) 306.
J. Ambjrn, L. Chekhov, C.F. Kristjansen, Yu. Makeenko, Nucl. Phys. B 404 (1993) 127.
G. Akemann, Nucl. Phys. B 482 (1996) 403.
F. David, Nucl. Phys. B 348 (1991) 507.
J. Jurkiewicz, Phys. Lett. B 245 (1990) 178.
C. Itoi, Nucl. Phys. B 493 (1997) 651.
J. Ambjrn, C.F. Kristjansen, Yu. Makeenko, Mod. Phys. Lett. A 7 (1992) 3187.
[28] G. Akemann, Nucl. Phys. B 507 (1997) 475.

[29] R. Delannay, G. Le Car, J. Phys. A 33 (2000) 2611.
[30] M.L. Mehta, Random Matrices, Academic Press, 1991.
757

CUMULATIVE AUTHOR INDEX B581B583
Akemann, G.
Alvarez, O.
Andreev, O.
Antoniadis, I.
Arhrib, A.
B583 (2000) 739

B582 (2000) 139
B583 (2000) 145
B583 (2000) 35
B581 (2000) 34
Balog, J.
Br, O.
Baulieu, L.
Blanger, G.
Benakli, K.
Bertola, M.
Bertolini, M.
Bialas, P.
Bialas, P.
Bilal, A.
Blanchard, P.
Blumenhagen, R.
Blmlein, J.
Bohm, A.R.
Boudjema, F.
Brignole, A.
Bros, J.
B583 (2000) 614

B581 (2000) 499
B581 (2000) 604
B581 (2000) 3
B583 (2000) 35
B581 (2000) 575
B582 (2000) 393
B581 (2000) 477
B583 (2000) 368
B582 (2000) 65
B583 (2000) 368
B582 (2000) 44
B581 (2000) 449
B581 (2000) 91
B581 (2000) 3
B582 (2000) 759
B581 (2000) 575
Caffo, M.
Campbell, B.A.
Campos, I.
Capdequi Peyranre, M.
Capitani, S.
Carter, G.W.
Casas, J.A.
Castellani, C.
Castro Alvaredo, O.A.
Chakrabarti, S.K.
Chan, Ch.S.
Chetyrkin, K.G.
Chu, C.-S.
Cirillo, E.N.M.
Coste, A.
Cox, J.
Cski, C.
Czyz, H.
B581 (2000) 274

B581 (2000) 240
B581 (2000) 499
B581 (2000) 34
B582 (2000) 762
B582 (2000) 571
B581 (2000) 61
B583 (2000) 542
B581 (2000) 643
B582 (2000) 627
B581 (2000) 156
B583 (2000) 3
B582 (2000) 65
B583 (2000) 584
B581 (2000) 679
B583 (2000) 331
B581 (2000) 309
B581 (2000) 274
Damgaard, P.H.
de Azcrraga, J.A.
Delfino, G.
Denef, F.
Derendinger, J.-P.
Derkachov, S.
Diakonov, D.
Daz, M.A.
Di Clemente, V.
Donato, F.
Dorn, H.
Dubin, A.Yu.
B583 (2000) 347

B581 (2000) 743
B583 (2000) 597
B581 (2000) 135
B582 (2000) 231
B583 (2000) 691
B582 (2000) 571
B583 (2000) 182
B581 (2000) 61
B581 (2000) 3
B583 (2000) 145
B582 (2000) 677
Eden, B.
Einhorn, M.B.
Enqvist, K.
Epele, L.N.
Erickson, J.K.
Erlich, J.
Evans, N.
B581 (2000) 523

B582 (2000) 216
B582 (2000) 763
B583 (2000) 454
B582 (2000) 155
B581 (2000) 309
B581 (2000) 391
Fabrizio, M.
Fanchiotti, H.
Ferreira, C.N.
Feruglio, F.
Fortunato, S.
Fosco, C.D.
Fries, R.J.
Frhlich, J.
B583 (2000) 542

B583 (2000) 454
B581 (2000) 165
B582 (2000) 759
B583 (2000) 368
B582 (2000) 716
B582 (2000) 537
B583 (2000) 381
Gandolfo, D.
Gannon, T.
Garca Canal, C.A.
Geyer, B.
Godbole, R.
Gonnella, G.
Gonzlez-Sprinberg, G.A.
Gorini, V.
Grlich, L.
Govindarajan, T.R.
Grandjean, O.
Grzadkowski, B.
Guan, X.-W.
B583 (2000) 368

B581 (2000) 679
B583 (2000) 454
B581 (2000) 341
B581 (2000) 3
B583 (2000) 584
B582 (2000) 3
B581 (2000) 575
B582 (2000) 44
B583 (2000) 291
B583 (2000) 381
B583 (2000) 49
B583 (2000) 721
760
Guimares, M.E.X.
Gunion, J.F.
Guruswamy, S.
B581 (2000) 165

B583 (2000) 49
B583 (2000) 475
Harshman, N.L.
Hassan, S.F.
Helayl-Neto, J.A.
Heller, U.M.
Herdeiro, C.A.R.
Hikami, K.
Holland, K.
Hollik, W.
Hollowood, T.J.
Hong, D.K.
Hormuzdiar, J.
Howe, P.S.
Hsu, S.D.H.
Hull, C.M.
B581 (2000) 91
B583 (2000) 431
B581 (2000) 165
B583 (2000) 347
B582 (2000) 363
B581 (2000) 761
B583 (2000) 331
B581 (2000) 34
B581 (2000) 309
B582 (2000) 451
B581 (2000) 391
B581 (2000) 523
B581 (2000) 391
B583 (2000) 237
Inoue, R.
Intriligator, K.
Ishibashi, N.
Iso, S.
B581 (2000) 761

B581 (2000) 257
B583 (2000) 159
B583 (2000) 159
Kaminsky, K.
Karakhanyan, D.
Kaste, P.
Kausch, H.G.
Kawai, H.
Kaya, A.
Ketov, S.V.
Ketov, S.V.
Kim, J.E.
Kimura, Y.
Kinar, Y.
Kirschner, R.
Kiselev, V.V.
Kitazawa, Y.
Kitazawa, Y.
Kniehl, B.A.
Kogut, J.B.
Krs, B.
Kotikov, A.V.
Kramer, G.
Kyae, B.
B581 (2000) 240

B583 (2000) 691
B582 (2000) 203
B583 (2000) 513
B583 (2000) 159
B583 (2000) 411
B582 (2000) 95
B582 (2000) 119
B582 (2000) 296
B581 (2000) 295
B583 (2000) 76
B583 (2000) 691
B581 (2000) 432
B581 (2000) 295
B583 (2000) 159
B582 (2000) 514
B582 (2000) 477
B582 (2000) 44
B582 (2000) 19
B582 (2000) 514
B582 (2000) 296
Laine, M.
Lazar, M.
LeClair, A.
Lee, H.M.
Lerche, W.
Lesgourgues, J.
Li, T.
Lipatov, L.N.
Lpez, A.
Losada, M.
B582 (2000) 277

B581 (2000) 341
B583 (2000) 475
B582 (2000) 296
B582 (2000) 203
B582 (2000) 593
B582 (2000) 176
B582 (2000) 19
B582 (2000) 716
B582 (2000) 277
Ludwig, A.W.W.
Lunardini, C.
Lscher, M.
Lst, D.
Ltken, C.A.
B583 (2000) 475

B583 (2000) 260
B582 (2000) 762
B582 (2000) 44
B582 (2000) 203
Macfarlane, A.J.
Mangano, M.L.
Martins, M.J.
Marucho, M.
McDonald, J.
Miramontes, J.L.
Moeller, N.
Montes, X.
Morel, A.
Moschella, U.
Moultaka, G.
Mukhopadhyay, B.
Mller, B.
B581 (2000) 743

B582 (2000) 759
B583 (2000) 721
B583 (2000) 454
B582 (2000) 763
B581 (2000) 643
B583 (2000) 105
B582 (2000) 259
B581 (2000) 477
B581 (2000) 575
B581 (2000) 34
B582 (2000) 627
B582 (2000) 537
Nastase, H.
Nastase, H.
Nersesyan, A.A.
Niclasen, R.
Niedermaier, M.
Niedermayer, F.
B581 (2000) 179

B583 (2000) 211
B583 (2000) 671
B583 (2000) 347
B583 (2000) 614
B583 (2000) 614
Oeckl, R.
Ohsawa, A.
Onishchenko, A.I.
B581 (2000) 559

B581 (2000) 73
B581 (2000) 432
Patrascioiu, A.
Paul, P.L.
Pelizzola, A.
Pernici, M.
Petersson, B.
Petrov, K.
Pickering, A.
Pliszka, J.
Polyakov, A.
Ptter, B.
B583 (2000) 614

B581 (2000) 156
B583 (2000) 584
B582 (2000) 733
B581 (2000) 477
B581 (2000) 477
B581 (2000) 523
B583 (2000) 49
B581 (2000) 116
B582 (2000) 514
Quirs, M.
Quirs, M.
B581 (2000) 61
B583 (2000) 35
Rasmussen, J.
Recknagel, A.
Reisz, T.
Remiddi, E.
Restrepo, D.A.
Rtey, A.
Robaschik, D.
Rosier-Lees, S.
Ruelle, P.
Rummukainen, K.
B582 (2000) 649

B583 (2000) 381
B581 (2000) 477
B581 (2000) 274
B583 (2000) 182
B583 (2000) 3
B581 (2000) 449
B581 (2000) 3
B581 (2000) 679
B583 (2000) 347
761
Russo, R.
Rychkov, V.
B582 (2000) 65
B581 (2000) 116
Trigiante, M.
Troost, J.
B582 (2000) 393

B581 (2000) 135
Santamaria, A.
Satz, H.
Sauser, R.
Schaeffer, R.
Schfer, A.
Schaposnik, F.A.
Scharnhorst, K.
Schomerus, V.
Schreiber, E.
Schwetz, M.
Seiler, E.
Semenoff, G.W.
Sevrin, A.
Shirman, Y.
Shore, G.M.
Smirnov, A.Yu.
Sokatchev, E.
Sommer, R.
Sonnenschein, J.
Stein, E.
Stephanov, M.A.
Suneeta, V.
B582 (2000) 3
B583 (2000) 368
B582 (2000) 231
B581 (2000) 575
B582 (2000) 537
B582 (2000) 716
B581 (2000) 718
B583 (2000) 381
B583 (2000) 76
B581 (2000) 391
B583 (2000) 614
B582 (2000) 155
B581 (2000) 135
B581 (2000) 309
B581 (2000) 409
B583 (2000) 260
B581 (2000) 523
B582 (2000) 762
B583 (2000) 76
B582 (2000) 537
B582 (2000) 477
B583 (2000) 291
Vaidya, S.
Valle, J.W.F.
Vaman, D.
Vaman, D.
van Nieuwenhuizen, P.
Verbaarschot, J.J.M.
Verbeni, M.
Verlinde, H.
Vernizzi, G.
Vidal, J.
Volkov, M.S.
B583 (2000) 291

B583 (2000) 182
B581 (2000) 179
B583 (2000) 211
B581 (2000) 179
B582 (2000) 477
B583 (2000) 307
B581 (2000) 156
B583 (2000) 739
B582 (2000) 3
B582 (2000) 313
Walcher, J.
Wang, Y.-J.
Weiss, N.
Weisz, P.
West, P.C.
White, B.E.
Wiedemann, U.A.
Wipf, A.
Wittig, H.
B582 (2000) 203

B583 (2000) 671
B583 (2000) 76
B583 (2000) 614
B581 (2000) 523
B581 (2000) 409
B582 (2000) 409
B582 (2000) 313
B582 (2000) 762
Tamada, M.
Tanaka, T.
Taylor, W.
Toublan, D.
Trentadue, L.
B581 (2000) 73
B582 (2000) 259
B583 (2000) 105
B582 (2000) 477
B583 (2000) 307
Zarembo, K.
Zayas, L.A.P.
Zhitnitsky, A.
Zwanziger, D.
Zwirner, F.
B582 (2000) 155

B582 (2000) 216
B582 (2000) 477
B581 (2000) 604
B582 (2000) 759

Nucl - Phys.B v.583

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.583

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 583 (2000) 334

Renormalization and running of quark mass and

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

The article is organized as follows. In Section 2 we discuss the scheme dependence of

2. Scheme dependence of quark mass and field

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

where the renormalized quark field is

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

with the conversion functions being themselves finite series in

The running of m0 and 0 is governed by the corresponding anomalous dimensions

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

properly chosen Green functions with predefined dependent configurations of external

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

3. Three-loop quark propagator in MS-scheme

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

4. RI scheme versus MS and MOM schemes

At a scale of = 2 GeV and nf = 4, the numerical contributions of the leading order to

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

C2RI = 1.0 + 0.0 0.0101 0.0095,

= 1.0 0.1333 0.0701 0.0458.

C2RI = 1.0 + 0.0 0.00333 0.00296

and we have evaluated the coefficients in the series in z with

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

gauge are given in Appendix C.

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

The corresponding equations for the RI0 scheme are:

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

where (x stands for either s ()/ or s (0 )/ )

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

Appendix A. Quark propagators

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

256 L q 256 L q  128 q 256

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

Appendix B. Conversion functions

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

Appendix C. Anomalous dimensions

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

1261 50047 2 66577 4 460151 6

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

1261 198679 2 778421 4 202580155 6

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

1027 7673 2 174565 4 3993865 6

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

1027 8069 2 240973 4 5232091 6

G. t Hooft, Nucl. Phys. B 61 (1973) 455.

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

K.G. Chetyrkin, A. Rtey / Nuclear Physics B 583 (2000) 334

Nuclear Physics B 583 (2000) 3548

Radiative symmetry breaking in brane models

Received 12 April 2000; accepted 25 May 2000

I. Antoniadis et al. / Nuclear Physics B 583 (2000) 3548

2. One-loop effective potential

256 L q 256 L q 128 q 256